]>
Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
8f18cf13 | 19 | #include <linux/kernel.h> |
065631f6 | 20 | #include <linux/bio.h> |
39279cc3 | 21 | #include <linux/buffer_head.h> |
f2eb0a24 | 22 | #include <linux/file.h> |
39279cc3 CM |
23 | #include <linux/fs.h> |
24 | #include <linux/pagemap.h> | |
25 | #include <linux/highmem.h> | |
26 | #include <linux/time.h> | |
27 | #include <linux/init.h> | |
28 | #include <linux/string.h> | |
39279cc3 CM |
29 | #include <linux/backing-dev.h> |
30 | #include <linux/mpage.h> | |
31 | #include <linux/swap.h> | |
32 | #include <linux/writeback.h> | |
39279cc3 | 33 | #include <linux/compat.h> |
9ebefb18 | 34 | #include <linux/bit_spinlock.h> |
5103e947 | 35 | #include <linux/xattr.h> |
33268eaf | 36 | #include <linux/posix_acl.h> |
d899e052 | 37 | #include <linux/falloc.h> |
5a0e3ad6 | 38 | #include <linux/slab.h> |
7a36ddec | 39 | #include <linux/ratelimit.h> |
22c44fe6 | 40 | #include <linux/mount.h> |
55e301fd | 41 | #include <linux/btrfs.h> |
53b381b3 | 42 | #include <linux/blkdev.h> |
f23b5a59 | 43 | #include <linux/posix_acl_xattr.h> |
e2e40f2c | 44 | #include <linux/uio.h> |
69fe2d75 | 45 | #include <linux/magic.h> |
39279cc3 CM |
46 | #include "ctree.h" |
47 | #include "disk-io.h" | |
48 | #include "transaction.h" | |
49 | #include "btrfs_inode.h" | |
39279cc3 | 50 | #include "print-tree.h" |
e6dcd2dc | 51 | #include "ordered-data.h" |
95819c05 | 52 | #include "xattr.h" |
e02119d5 | 53 | #include "tree-log.h" |
4a54c8c1 | 54 | #include "volumes.h" |
c8b97818 | 55 | #include "compression.h" |
b4ce94de | 56 | #include "locking.h" |
dc89e982 | 57 | #include "free-space-cache.h" |
581bb050 | 58 | #include "inode-map.h" |
38c227d8 | 59 | #include "backref.h" |
f23b5a59 | 60 | #include "hash.h" |
63541927 | 61 | #include "props.h" |
31193213 | 62 | #include "qgroup.h" |
dda3245e | 63 | #include "dedupe.h" |
39279cc3 CM |
64 | |
65 | struct btrfs_iget_args { | |
90d3e592 | 66 | struct btrfs_key *location; |
39279cc3 CM |
67 | struct btrfs_root *root; |
68 | }; | |
69 | ||
f28a4928 | 70 | struct btrfs_dio_data { |
f28a4928 FM |
71 | u64 reserve; |
72 | u64 unsubmitted_oe_range_start; | |
73 | u64 unsubmitted_oe_range_end; | |
4aaedfb0 | 74 | int overwrite; |
f28a4928 FM |
75 | }; |
76 | ||
6e1d5dcc AD |
77 | static const struct inode_operations btrfs_dir_inode_operations; |
78 | static const struct inode_operations btrfs_symlink_inode_operations; | |
79 | static const struct inode_operations btrfs_dir_ro_inode_operations; | |
80 | static const struct inode_operations btrfs_special_inode_operations; | |
81 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b AD |
82 | static const struct address_space_operations btrfs_aops; |
83 | static const struct address_space_operations btrfs_symlink_aops; | |
828c0950 | 84 | static const struct file_operations btrfs_dir_file_operations; |
20e5506b | 85 | static const struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
86 | |
87 | static struct kmem_cache *btrfs_inode_cachep; | |
88 | struct kmem_cache *btrfs_trans_handle_cachep; | |
39279cc3 | 89 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 90 | struct kmem_cache *btrfs_free_space_cachep; |
39279cc3 CM |
91 | |
92 | #define S_SHIFT 12 | |
4d4ab6d6 | 93 | static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { |
39279cc3 CM |
94 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, |
95 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
96 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
97 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
98 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
99 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
100 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
101 | }; | |
102 | ||
3972f260 | 103 | static int btrfs_setsize(struct inode *inode, struct iattr *attr); |
a41ad394 | 104 | static int btrfs_truncate(struct inode *inode); |
5fd02043 | 105 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent); |
771ed689 CM |
106 | static noinline int cow_file_range(struct inode *inode, |
107 | struct page *locked_page, | |
dda3245e WX |
108 | u64 start, u64 end, u64 delalloc_end, |
109 | int *page_started, unsigned long *nr_written, | |
110 | int unlock, struct btrfs_dedupe_hash *hash); | |
6f9994db LB |
111 | static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len, |
112 | u64 orig_start, u64 block_start, | |
113 | u64 block_len, u64 orig_block_len, | |
114 | u64 ram_bytes, int compress_type, | |
115 | int type); | |
7b128766 | 116 | |
52427260 QW |
117 | static void __endio_write_update_ordered(struct inode *inode, |
118 | const u64 offset, const u64 bytes, | |
119 | const bool uptodate); | |
120 | ||
121 | /* | |
122 | * Cleanup all submitted ordered extents in specified range to handle errors | |
123 | * from the fill_dellaloc() callback. | |
124 | * | |
125 | * NOTE: caller must ensure that when an error happens, it can not call | |
126 | * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING | |
127 | * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata | |
128 | * to be released, which we want to happen only when finishing the ordered | |
129 | * extent (btrfs_finish_ordered_io()). Also note that the caller of the | |
130 | * fill_delalloc() callback already does proper cleanup for the first page of | |
131 | * the range, that is, it invokes the callback writepage_end_io_hook() for the | |
132 | * range of the first page. | |
133 | */ | |
134 | static inline void btrfs_cleanup_ordered_extents(struct inode *inode, | |
135 | const u64 offset, | |
136 | const u64 bytes) | |
137 | { | |
63d71450 NA |
138 | unsigned long index = offset >> PAGE_SHIFT; |
139 | unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT; | |
140 | struct page *page; | |
141 | ||
142 | while (index <= end_index) { | |
143 | page = find_get_page(inode->i_mapping, index); | |
144 | index++; | |
145 | if (!page) | |
146 | continue; | |
147 | ClearPagePrivate2(page); | |
148 | put_page(page); | |
149 | } | |
52427260 QW |
150 | return __endio_write_update_ordered(inode, offset + PAGE_SIZE, |
151 | bytes - PAGE_SIZE, false); | |
152 | } | |
153 | ||
48a3b636 | 154 | static int btrfs_dirty_inode(struct inode *inode); |
7b128766 | 155 | |
6a3891c5 JB |
156 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
157 | void btrfs_test_inode_set_ops(struct inode *inode) | |
158 | { | |
159 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; | |
160 | } | |
161 | #endif | |
162 | ||
f34f57a3 | 163 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
2a7dba39 EP |
164 | struct inode *inode, struct inode *dir, |
165 | const struct qstr *qstr) | |
0279b4cd JO |
166 | { |
167 | int err; | |
168 | ||
f34f57a3 | 169 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 170 | if (!err) |
2a7dba39 | 171 | err = btrfs_xattr_security_init(trans, inode, dir, qstr); |
0279b4cd JO |
172 | return err; |
173 | } | |
174 | ||
c8b97818 CM |
175 | /* |
176 | * this does all the hard work for inserting an inline extent into | |
177 | * the btree. The caller should have done a btrfs_drop_extents so that | |
178 | * no overlapping inline items exist in the btree | |
179 | */ | |
40f76580 | 180 | static int insert_inline_extent(struct btrfs_trans_handle *trans, |
1acae57b | 181 | struct btrfs_path *path, int extent_inserted, |
c8b97818 CM |
182 | struct btrfs_root *root, struct inode *inode, |
183 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 184 | int compress_type, |
c8b97818 CM |
185 | struct page **compressed_pages) |
186 | { | |
c8b97818 CM |
187 | struct extent_buffer *leaf; |
188 | struct page *page = NULL; | |
189 | char *kaddr; | |
190 | unsigned long ptr; | |
191 | struct btrfs_file_extent_item *ei; | |
c8b97818 CM |
192 | int ret; |
193 | size_t cur_size = size; | |
c8b97818 | 194 | unsigned long offset; |
c8b97818 | 195 | |
fe3f566c | 196 | if (compressed_size && compressed_pages) |
c8b97818 | 197 | cur_size = compressed_size; |
c8b97818 | 198 | |
1acae57b | 199 | inode_add_bytes(inode, size); |
c8b97818 | 200 | |
1acae57b FDBM |
201 | if (!extent_inserted) { |
202 | struct btrfs_key key; | |
203 | size_t datasize; | |
c8b97818 | 204 | |
4a0cc7ca | 205 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
1acae57b | 206 | key.offset = start; |
962a298f | 207 | key.type = BTRFS_EXTENT_DATA_KEY; |
c8b97818 | 208 | |
1acae57b FDBM |
209 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
210 | path->leave_spinning = 1; | |
211 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
212 | datasize); | |
79b4f4c6 | 213 | if (ret) |
1acae57b | 214 | goto fail; |
c8b97818 CM |
215 | } |
216 | leaf = path->nodes[0]; | |
217 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
218 | struct btrfs_file_extent_item); | |
219 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
220 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
221 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
222 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
223 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
224 | ptr = btrfs_file_extent_inline_start(ei); | |
225 | ||
261507a0 | 226 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
227 | struct page *cpage; |
228 | int i = 0; | |
d397712b | 229 | while (compressed_size > 0) { |
c8b97818 | 230 | cpage = compressed_pages[i]; |
5b050f04 | 231 | cur_size = min_t(unsigned long, compressed_size, |
09cbfeaf | 232 | PAGE_SIZE); |
c8b97818 | 233 | |
7ac687d9 | 234 | kaddr = kmap_atomic(cpage); |
c8b97818 | 235 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
7ac687d9 | 236 | kunmap_atomic(kaddr); |
c8b97818 CM |
237 | |
238 | i++; | |
239 | ptr += cur_size; | |
240 | compressed_size -= cur_size; | |
241 | } | |
242 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 243 | compress_type); |
c8b97818 CM |
244 | } else { |
245 | page = find_get_page(inode->i_mapping, | |
09cbfeaf | 246 | start >> PAGE_SHIFT); |
c8b97818 | 247 | btrfs_set_file_extent_compression(leaf, ei, 0); |
7ac687d9 | 248 | kaddr = kmap_atomic(page); |
09cbfeaf | 249 | offset = start & (PAGE_SIZE - 1); |
c8b97818 | 250 | write_extent_buffer(leaf, kaddr + offset, ptr, size); |
7ac687d9 | 251 | kunmap_atomic(kaddr); |
09cbfeaf | 252 | put_page(page); |
c8b97818 CM |
253 | } |
254 | btrfs_mark_buffer_dirty(leaf); | |
1acae57b | 255 | btrfs_release_path(path); |
c8b97818 | 256 | |
c2167754 YZ |
257 | /* |
258 | * we're an inline extent, so nobody can | |
259 | * extend the file past i_size without locking | |
260 | * a page we already have locked. | |
261 | * | |
262 | * We must do any isize and inode updates | |
263 | * before we unlock the pages. Otherwise we | |
264 | * could end up racing with unlink. | |
265 | */ | |
c8b97818 | 266 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
79787eaa | 267 | ret = btrfs_update_inode(trans, root, inode); |
c2167754 | 268 | |
c8b97818 | 269 | fail: |
79b4f4c6 | 270 | return ret; |
c8b97818 CM |
271 | } |
272 | ||
273 | ||
274 | /* | |
275 | * conditionally insert an inline extent into the file. This | |
276 | * does the checks required to make sure the data is small enough | |
277 | * to fit as an inline extent. | |
278 | */ | |
00361589 JB |
279 | static noinline int cow_file_range_inline(struct btrfs_root *root, |
280 | struct inode *inode, u64 start, | |
281 | u64 end, size_t compressed_size, | |
282 | int compress_type, | |
283 | struct page **compressed_pages) | |
c8b97818 | 284 | { |
0b246afa | 285 | struct btrfs_fs_info *fs_info = root->fs_info; |
00361589 | 286 | struct btrfs_trans_handle *trans; |
c8b97818 CM |
287 | u64 isize = i_size_read(inode); |
288 | u64 actual_end = min(end + 1, isize); | |
289 | u64 inline_len = actual_end - start; | |
0b246afa | 290 | u64 aligned_end = ALIGN(end, fs_info->sectorsize); |
c8b97818 CM |
291 | u64 data_len = inline_len; |
292 | int ret; | |
1acae57b FDBM |
293 | struct btrfs_path *path; |
294 | int extent_inserted = 0; | |
295 | u32 extent_item_size; | |
c8b97818 CM |
296 | |
297 | if (compressed_size) | |
298 | data_len = compressed_size; | |
299 | ||
300 | if (start > 0 || | |
0b246afa JM |
301 | actual_end > fs_info->sectorsize || |
302 | data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) || | |
c8b97818 | 303 | (!compressed_size && |
0b246afa | 304 | (actual_end & (fs_info->sectorsize - 1)) == 0) || |
c8b97818 | 305 | end + 1 < isize || |
0b246afa | 306 | data_len > fs_info->max_inline) { |
c8b97818 CM |
307 | return 1; |
308 | } | |
309 | ||
1acae57b FDBM |
310 | path = btrfs_alloc_path(); |
311 | if (!path) | |
312 | return -ENOMEM; | |
313 | ||
00361589 | 314 | trans = btrfs_join_transaction(root); |
1acae57b FDBM |
315 | if (IS_ERR(trans)) { |
316 | btrfs_free_path(path); | |
00361589 | 317 | return PTR_ERR(trans); |
1acae57b | 318 | } |
69fe2d75 | 319 | trans->block_rsv = &BTRFS_I(inode)->block_rsv; |
00361589 | 320 | |
1acae57b FDBM |
321 | if (compressed_size && compressed_pages) |
322 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
323 | compressed_size); | |
324 | else | |
325 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
326 | inline_len); | |
327 | ||
328 | ret = __btrfs_drop_extents(trans, root, inode, path, | |
329 | start, aligned_end, NULL, | |
330 | 1, 1, extent_item_size, &extent_inserted); | |
00361589 | 331 | if (ret) { |
66642832 | 332 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
333 | goto out; |
334 | } | |
c8b97818 CM |
335 | |
336 | if (isize > actual_end) | |
337 | inline_len = min_t(u64, isize, actual_end); | |
1acae57b FDBM |
338 | ret = insert_inline_extent(trans, path, extent_inserted, |
339 | root, inode, start, | |
c8b97818 | 340 | inline_len, compressed_size, |
fe3f566c | 341 | compress_type, compressed_pages); |
2adcac1a | 342 | if (ret && ret != -ENOSPC) { |
66642832 | 343 | btrfs_abort_transaction(trans, ret); |
00361589 | 344 | goto out; |
2adcac1a | 345 | } else if (ret == -ENOSPC) { |
00361589 JB |
346 | ret = 1; |
347 | goto out; | |
79787eaa | 348 | } |
2adcac1a | 349 | |
bdc20e67 | 350 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); |
dcdbc059 | 351 | btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0); |
00361589 | 352 | out: |
94ed938a QW |
353 | /* |
354 | * Don't forget to free the reserved space, as for inlined extent | |
355 | * it won't count as data extent, free them directly here. | |
356 | * And at reserve time, it's always aligned to page size, so | |
357 | * just free one page here. | |
358 | */ | |
bc42bda2 | 359 | btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE); |
1acae57b | 360 | btrfs_free_path(path); |
3a45bb20 | 361 | btrfs_end_transaction(trans); |
00361589 | 362 | return ret; |
c8b97818 CM |
363 | } |
364 | ||
771ed689 CM |
365 | struct async_extent { |
366 | u64 start; | |
367 | u64 ram_size; | |
368 | u64 compressed_size; | |
369 | struct page **pages; | |
370 | unsigned long nr_pages; | |
261507a0 | 371 | int compress_type; |
771ed689 CM |
372 | struct list_head list; |
373 | }; | |
374 | ||
375 | struct async_cow { | |
376 | struct inode *inode; | |
377 | struct btrfs_root *root; | |
378 | struct page *locked_page; | |
379 | u64 start; | |
380 | u64 end; | |
f82b7359 | 381 | unsigned int write_flags; |
771ed689 CM |
382 | struct list_head extents; |
383 | struct btrfs_work work; | |
384 | }; | |
385 | ||
386 | static noinline int add_async_extent(struct async_cow *cow, | |
387 | u64 start, u64 ram_size, | |
388 | u64 compressed_size, | |
389 | struct page **pages, | |
261507a0 LZ |
390 | unsigned long nr_pages, |
391 | int compress_type) | |
771ed689 CM |
392 | { |
393 | struct async_extent *async_extent; | |
394 | ||
395 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 396 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
397 | async_extent->start = start; |
398 | async_extent->ram_size = ram_size; | |
399 | async_extent->compressed_size = compressed_size; | |
400 | async_extent->pages = pages; | |
401 | async_extent->nr_pages = nr_pages; | |
261507a0 | 402 | async_extent->compress_type = compress_type; |
771ed689 CM |
403 | list_add_tail(&async_extent->list, &cow->extents); |
404 | return 0; | |
405 | } | |
406 | ||
c2fcdcdf | 407 | static inline int inode_need_compress(struct inode *inode, u64 start, u64 end) |
f79707b0 | 408 | { |
0b246afa | 409 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
f79707b0 WS |
410 | |
411 | /* force compress */ | |
0b246afa | 412 | if (btrfs_test_opt(fs_info, FORCE_COMPRESS)) |
f79707b0 | 413 | return 1; |
eec63c65 DS |
414 | /* defrag ioctl */ |
415 | if (BTRFS_I(inode)->defrag_compress) | |
416 | return 1; | |
f79707b0 WS |
417 | /* bad compression ratios */ |
418 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) | |
419 | return 0; | |
0b246afa | 420 | if (btrfs_test_opt(fs_info, COMPRESS) || |
f79707b0 | 421 | BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS || |
b52aa8c9 | 422 | BTRFS_I(inode)->prop_compress) |
c2fcdcdf | 423 | return btrfs_compress_heuristic(inode, start, end); |
f79707b0 WS |
424 | return 0; |
425 | } | |
426 | ||
6158e1ce | 427 | static inline void inode_should_defrag(struct btrfs_inode *inode, |
26d30f85 AJ |
428 | u64 start, u64 end, u64 num_bytes, u64 small_write) |
429 | { | |
430 | /* If this is a small write inside eof, kick off a defrag */ | |
431 | if (num_bytes < small_write && | |
6158e1ce | 432 | (start > 0 || end + 1 < inode->disk_i_size)) |
26d30f85 AJ |
433 | btrfs_add_inode_defrag(NULL, inode); |
434 | } | |
435 | ||
d352ac68 | 436 | /* |
771ed689 CM |
437 | * we create compressed extents in two phases. The first |
438 | * phase compresses a range of pages that have already been | |
439 | * locked (both pages and state bits are locked). | |
c8b97818 | 440 | * |
771ed689 CM |
441 | * This is done inside an ordered work queue, and the compression |
442 | * is spread across many cpus. The actual IO submission is step | |
443 | * two, and the ordered work queue takes care of making sure that | |
444 | * happens in the same order things were put onto the queue by | |
445 | * writepages and friends. | |
c8b97818 | 446 | * |
771ed689 CM |
447 | * If this code finds it can't get good compression, it puts an |
448 | * entry onto the work queue to write the uncompressed bytes. This | |
449 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
450 | * are written in the same order that the flusher thread sent them |
451 | * down. | |
d352ac68 | 452 | */ |
c44f649e | 453 | static noinline void compress_file_range(struct inode *inode, |
771ed689 CM |
454 | struct page *locked_page, |
455 | u64 start, u64 end, | |
456 | struct async_cow *async_cow, | |
457 | int *num_added) | |
b888db2b | 458 | { |
0b246afa | 459 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
b888db2b | 460 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0b246afa | 461 | u64 blocksize = fs_info->sectorsize; |
c8b97818 | 462 | u64 actual_end; |
42dc7bab | 463 | u64 isize = i_size_read(inode); |
e6dcd2dc | 464 | int ret = 0; |
c8b97818 CM |
465 | struct page **pages = NULL; |
466 | unsigned long nr_pages; | |
c8b97818 CM |
467 | unsigned long total_compressed = 0; |
468 | unsigned long total_in = 0; | |
c8b97818 CM |
469 | int i; |
470 | int will_compress; | |
0b246afa | 471 | int compress_type = fs_info->compress_type; |
4adaa611 | 472 | int redirty = 0; |
b888db2b | 473 | |
6158e1ce NB |
474 | inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1, |
475 | SZ_16K); | |
4cb5300b | 476 | |
42dc7bab | 477 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
478 | again: |
479 | will_compress = 0; | |
09cbfeaf | 480 | nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1; |
069eac78 DS |
481 | BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0); |
482 | nr_pages = min_t(unsigned long, nr_pages, | |
483 | BTRFS_MAX_COMPRESSED / PAGE_SIZE); | |
be20aa9d | 484 | |
f03d9301 CM |
485 | /* |
486 | * we don't want to send crud past the end of i_size through | |
487 | * compression, that's just a waste of CPU time. So, if the | |
488 | * end of the file is before the start of our current | |
489 | * requested range of bytes, we bail out to the uncompressed | |
490 | * cleanup code that can deal with all of this. | |
491 | * | |
492 | * It isn't really the fastest way to fix things, but this is a | |
493 | * very uncommon corner. | |
494 | */ | |
495 | if (actual_end <= start) | |
496 | goto cleanup_and_bail_uncompressed; | |
497 | ||
c8b97818 CM |
498 | total_compressed = actual_end - start; |
499 | ||
4bcbb332 SW |
500 | /* |
501 | * skip compression for a small file range(<=blocksize) that | |
01327610 | 502 | * isn't an inline extent, since it doesn't save disk space at all. |
4bcbb332 SW |
503 | */ |
504 | if (total_compressed <= blocksize && | |
505 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
506 | goto cleanup_and_bail_uncompressed; | |
507 | ||
069eac78 DS |
508 | total_compressed = min_t(unsigned long, total_compressed, |
509 | BTRFS_MAX_UNCOMPRESSED); | |
c8b97818 CM |
510 | total_in = 0; |
511 | ret = 0; | |
db94535d | 512 | |
771ed689 CM |
513 | /* |
514 | * we do compression for mount -o compress and when the | |
515 | * inode has not been flagged as nocompress. This flag can | |
516 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 517 | */ |
c2fcdcdf | 518 | if (inode_need_compress(inode, start, end)) { |
c8b97818 | 519 | WARN_ON(pages); |
31e818fe | 520 | pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS); |
560f7d75 LZ |
521 | if (!pages) { |
522 | /* just bail out to the uncompressed code */ | |
523 | goto cont; | |
524 | } | |
c8b97818 | 525 | |
eec63c65 DS |
526 | if (BTRFS_I(inode)->defrag_compress) |
527 | compress_type = BTRFS_I(inode)->defrag_compress; | |
528 | else if (BTRFS_I(inode)->prop_compress) | |
b52aa8c9 | 529 | compress_type = BTRFS_I(inode)->prop_compress; |
261507a0 | 530 | |
4adaa611 CM |
531 | /* |
532 | * we need to call clear_page_dirty_for_io on each | |
533 | * page in the range. Otherwise applications with the file | |
534 | * mmap'd can wander in and change the page contents while | |
535 | * we are compressing them. | |
536 | * | |
537 | * If the compression fails for any reason, we set the pages | |
538 | * dirty again later on. | |
539 | */ | |
540 | extent_range_clear_dirty_for_io(inode, start, end); | |
541 | redirty = 1; | |
f51d2b59 DS |
542 | |
543 | /* Compression level is applied here and only here */ | |
544 | ret = btrfs_compress_pages( | |
545 | compress_type | (fs_info->compress_level << 4), | |
261507a0 | 546 | inode->i_mapping, start, |
38c31464 | 547 | pages, |
4d3a800e | 548 | &nr_pages, |
261507a0 | 549 | &total_in, |
e5d74902 | 550 | &total_compressed); |
c8b97818 CM |
551 | |
552 | if (!ret) { | |
553 | unsigned long offset = total_compressed & | |
09cbfeaf | 554 | (PAGE_SIZE - 1); |
4d3a800e | 555 | struct page *page = pages[nr_pages - 1]; |
c8b97818 CM |
556 | char *kaddr; |
557 | ||
558 | /* zero the tail end of the last page, we might be | |
559 | * sending it down to disk | |
560 | */ | |
561 | if (offset) { | |
7ac687d9 | 562 | kaddr = kmap_atomic(page); |
c8b97818 | 563 | memset(kaddr + offset, 0, |
09cbfeaf | 564 | PAGE_SIZE - offset); |
7ac687d9 | 565 | kunmap_atomic(kaddr); |
c8b97818 CM |
566 | } |
567 | will_compress = 1; | |
568 | } | |
569 | } | |
560f7d75 | 570 | cont: |
c8b97818 CM |
571 | if (start == 0) { |
572 | /* lets try to make an inline extent */ | |
6018ba0a | 573 | if (ret || total_in < actual_end) { |
c8b97818 | 574 | /* we didn't compress the entire range, try |
771ed689 | 575 | * to make an uncompressed inline extent. |
c8b97818 | 576 | */ |
00361589 | 577 | ret = cow_file_range_inline(root, inode, start, end, |
f74670f7 | 578 | 0, BTRFS_COMPRESS_NONE, NULL); |
c8b97818 | 579 | } else { |
771ed689 | 580 | /* try making a compressed inline extent */ |
00361589 | 581 | ret = cow_file_range_inline(root, inode, start, end, |
fe3f566c LZ |
582 | total_compressed, |
583 | compress_type, pages); | |
c8b97818 | 584 | } |
79787eaa | 585 | if (ret <= 0) { |
151a41bc | 586 | unsigned long clear_flags = EXTENT_DELALLOC | |
8b62f87b JB |
587 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | |
588 | EXTENT_DO_ACCOUNTING; | |
e6eb4314 FM |
589 | unsigned long page_error_op; |
590 | ||
e6eb4314 | 591 | page_error_op = ret < 0 ? PAGE_SET_ERROR : 0; |
151a41bc | 592 | |
771ed689 | 593 | /* |
79787eaa JM |
594 | * inline extent creation worked or returned error, |
595 | * we don't need to create any more async work items. | |
596 | * Unlock and free up our temp pages. | |
8b62f87b JB |
597 | * |
598 | * We use DO_ACCOUNTING here because we need the | |
599 | * delalloc_release_metadata to be done _after_ we drop | |
600 | * our outstanding extent for clearing delalloc for this | |
601 | * range. | |
771ed689 | 602 | */ |
ba8b04c1 QW |
603 | extent_clear_unlock_delalloc(inode, start, end, end, |
604 | NULL, clear_flags, | |
605 | PAGE_UNLOCK | | |
c2790a2e JB |
606 | PAGE_CLEAR_DIRTY | |
607 | PAGE_SET_WRITEBACK | | |
e6eb4314 | 608 | page_error_op | |
c2790a2e | 609 | PAGE_END_WRITEBACK); |
c8b97818 CM |
610 | goto free_pages_out; |
611 | } | |
612 | } | |
613 | ||
614 | if (will_compress) { | |
615 | /* | |
616 | * we aren't doing an inline extent round the compressed size | |
617 | * up to a block size boundary so the allocator does sane | |
618 | * things | |
619 | */ | |
fda2832f | 620 | total_compressed = ALIGN(total_compressed, blocksize); |
c8b97818 CM |
621 | |
622 | /* | |
623 | * one last check to make sure the compression is really a | |
170607eb TT |
624 | * win, compare the page count read with the blocks on disk, |
625 | * compression must free at least one sector size | |
c8b97818 | 626 | */ |
09cbfeaf | 627 | total_in = ALIGN(total_in, PAGE_SIZE); |
170607eb | 628 | if (total_compressed + blocksize <= total_in) { |
c8bb0c8b AS |
629 | *num_added += 1; |
630 | ||
631 | /* | |
632 | * The async work queues will take care of doing actual | |
633 | * allocation on disk for these compressed pages, and | |
634 | * will submit them to the elevator. | |
635 | */ | |
1170862d | 636 | add_async_extent(async_cow, start, total_in, |
4d3a800e | 637 | total_compressed, pages, nr_pages, |
c8bb0c8b AS |
638 | compress_type); |
639 | ||
1170862d TT |
640 | if (start + total_in < end) { |
641 | start += total_in; | |
c8bb0c8b AS |
642 | pages = NULL; |
643 | cond_resched(); | |
644 | goto again; | |
645 | } | |
646 | return; | |
c8b97818 CM |
647 | } |
648 | } | |
c8bb0c8b | 649 | if (pages) { |
c8b97818 CM |
650 | /* |
651 | * the compression code ran but failed to make things smaller, | |
652 | * free any pages it allocated and our page pointer array | |
653 | */ | |
4d3a800e | 654 | for (i = 0; i < nr_pages; i++) { |
70b99e69 | 655 | WARN_ON(pages[i]->mapping); |
09cbfeaf | 656 | put_page(pages[i]); |
c8b97818 CM |
657 | } |
658 | kfree(pages); | |
659 | pages = NULL; | |
660 | total_compressed = 0; | |
4d3a800e | 661 | nr_pages = 0; |
c8b97818 CM |
662 | |
663 | /* flag the file so we don't compress in the future */ | |
0b246afa | 664 | if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) && |
b52aa8c9 | 665 | !(BTRFS_I(inode)->prop_compress)) { |
a555f810 | 666 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 667 | } |
c8b97818 | 668 | } |
f03d9301 | 669 | cleanup_and_bail_uncompressed: |
c8bb0c8b AS |
670 | /* |
671 | * No compression, but we still need to write the pages in the file | |
672 | * we've been given so far. redirty the locked page if it corresponds | |
673 | * to our extent and set things up for the async work queue to run | |
674 | * cow_file_range to do the normal delalloc dance. | |
675 | */ | |
676 | if (page_offset(locked_page) >= start && | |
677 | page_offset(locked_page) <= end) | |
678 | __set_page_dirty_nobuffers(locked_page); | |
679 | /* unlocked later on in the async handlers */ | |
680 | ||
681 | if (redirty) | |
682 | extent_range_redirty_for_io(inode, start, end); | |
683 | add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0, | |
684 | BTRFS_COMPRESS_NONE); | |
685 | *num_added += 1; | |
3b951516 | 686 | |
c44f649e | 687 | return; |
771ed689 CM |
688 | |
689 | free_pages_out: | |
4d3a800e | 690 | for (i = 0; i < nr_pages; i++) { |
771ed689 | 691 | WARN_ON(pages[i]->mapping); |
09cbfeaf | 692 | put_page(pages[i]); |
771ed689 | 693 | } |
d397712b | 694 | kfree(pages); |
771ed689 | 695 | } |
771ed689 | 696 | |
40ae837b FM |
697 | static void free_async_extent_pages(struct async_extent *async_extent) |
698 | { | |
699 | int i; | |
700 | ||
701 | if (!async_extent->pages) | |
702 | return; | |
703 | ||
704 | for (i = 0; i < async_extent->nr_pages; i++) { | |
705 | WARN_ON(async_extent->pages[i]->mapping); | |
09cbfeaf | 706 | put_page(async_extent->pages[i]); |
40ae837b FM |
707 | } |
708 | kfree(async_extent->pages); | |
709 | async_extent->nr_pages = 0; | |
710 | async_extent->pages = NULL; | |
771ed689 CM |
711 | } |
712 | ||
713 | /* | |
714 | * phase two of compressed writeback. This is the ordered portion | |
715 | * of the code, which only gets called in the order the work was | |
716 | * queued. We walk all the async extents created by compress_file_range | |
717 | * and send them down to the disk. | |
718 | */ | |
dec8f175 | 719 | static noinline void submit_compressed_extents(struct inode *inode, |
771ed689 CM |
720 | struct async_cow *async_cow) |
721 | { | |
0b246afa | 722 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
771ed689 CM |
723 | struct async_extent *async_extent; |
724 | u64 alloc_hint = 0; | |
771ed689 CM |
725 | struct btrfs_key ins; |
726 | struct extent_map *em; | |
727 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
771ed689 | 728 | struct extent_io_tree *io_tree; |
f5a84ee3 | 729 | int ret = 0; |
771ed689 | 730 | |
3e04e7f1 | 731 | again: |
d397712b | 732 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
733 | async_extent = list_entry(async_cow->extents.next, |
734 | struct async_extent, list); | |
735 | list_del(&async_extent->list); | |
c8b97818 | 736 | |
771ed689 CM |
737 | io_tree = &BTRFS_I(inode)->io_tree; |
738 | ||
f5a84ee3 | 739 | retry: |
771ed689 CM |
740 | /* did the compression code fall back to uncompressed IO? */ |
741 | if (!async_extent->pages) { | |
742 | int page_started = 0; | |
743 | unsigned long nr_written = 0; | |
744 | ||
745 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 | 746 | async_extent->start + |
d0082371 | 747 | async_extent->ram_size - 1); |
771ed689 CM |
748 | |
749 | /* allocate blocks */ | |
f5a84ee3 JB |
750 | ret = cow_file_range(inode, async_cow->locked_page, |
751 | async_extent->start, | |
752 | async_extent->start + | |
753 | async_extent->ram_size - 1, | |
dda3245e WX |
754 | async_extent->start + |
755 | async_extent->ram_size - 1, | |
756 | &page_started, &nr_written, 0, | |
757 | NULL); | |
771ed689 | 758 | |
79787eaa JM |
759 | /* JDM XXX */ |
760 | ||
771ed689 CM |
761 | /* |
762 | * if page_started, cow_file_range inserted an | |
763 | * inline extent and took care of all the unlocking | |
764 | * and IO for us. Otherwise, we need to submit | |
765 | * all those pages down to the drive. | |
766 | */ | |
f5a84ee3 | 767 | if (!page_started && !ret) |
771ed689 CM |
768 | extent_write_locked_range(io_tree, |
769 | inode, async_extent->start, | |
d397712b | 770 | async_extent->start + |
771ed689 CM |
771 | async_extent->ram_size - 1, |
772 | btrfs_get_extent, | |
773 | WB_SYNC_ALL); | |
3e04e7f1 JB |
774 | else if (ret) |
775 | unlock_page(async_cow->locked_page); | |
771ed689 CM |
776 | kfree(async_extent); |
777 | cond_resched(); | |
778 | continue; | |
779 | } | |
780 | ||
781 | lock_extent(io_tree, async_extent->start, | |
d0082371 | 782 | async_extent->start + async_extent->ram_size - 1); |
771ed689 | 783 | |
18513091 | 784 | ret = btrfs_reserve_extent(root, async_extent->ram_size, |
771ed689 CM |
785 | async_extent->compressed_size, |
786 | async_extent->compressed_size, | |
e570fd27 | 787 | 0, alloc_hint, &ins, 1, 1); |
f5a84ee3 | 788 | if (ret) { |
40ae837b | 789 | free_async_extent_pages(async_extent); |
3e04e7f1 | 790 | |
fdf8e2ea JB |
791 | if (ret == -ENOSPC) { |
792 | unlock_extent(io_tree, async_extent->start, | |
793 | async_extent->start + | |
794 | async_extent->ram_size - 1); | |
ce62003f LB |
795 | |
796 | /* | |
797 | * we need to redirty the pages if we decide to | |
798 | * fallback to uncompressed IO, otherwise we | |
799 | * will not submit these pages down to lower | |
800 | * layers. | |
801 | */ | |
802 | extent_range_redirty_for_io(inode, | |
803 | async_extent->start, | |
804 | async_extent->start + | |
805 | async_extent->ram_size - 1); | |
806 | ||
79787eaa | 807 | goto retry; |
fdf8e2ea | 808 | } |
3e04e7f1 | 809 | goto out_free; |
f5a84ee3 | 810 | } |
c2167754 YZ |
811 | /* |
812 | * here we're doing allocation and writeback of the | |
813 | * compressed pages | |
814 | */ | |
6f9994db LB |
815 | em = create_io_em(inode, async_extent->start, |
816 | async_extent->ram_size, /* len */ | |
817 | async_extent->start, /* orig_start */ | |
818 | ins.objectid, /* block_start */ | |
819 | ins.offset, /* block_len */ | |
820 | ins.offset, /* orig_block_len */ | |
821 | async_extent->ram_size, /* ram_bytes */ | |
822 | async_extent->compress_type, | |
823 | BTRFS_ORDERED_COMPRESSED); | |
824 | if (IS_ERR(em)) | |
825 | /* ret value is not necessary due to void function */ | |
3e04e7f1 | 826 | goto out_free_reserve; |
6f9994db | 827 | free_extent_map(em); |
3e04e7f1 | 828 | |
261507a0 LZ |
829 | ret = btrfs_add_ordered_extent_compress(inode, |
830 | async_extent->start, | |
831 | ins.objectid, | |
832 | async_extent->ram_size, | |
833 | ins.offset, | |
834 | BTRFS_ORDERED_COMPRESSED, | |
835 | async_extent->compress_type); | |
d9f85963 | 836 | if (ret) { |
dcdbc059 NB |
837 | btrfs_drop_extent_cache(BTRFS_I(inode), |
838 | async_extent->start, | |
d9f85963 FM |
839 | async_extent->start + |
840 | async_extent->ram_size - 1, 0); | |
3e04e7f1 | 841 | goto out_free_reserve; |
d9f85963 | 842 | } |
0b246afa | 843 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
771ed689 | 844 | |
771ed689 CM |
845 | /* |
846 | * clear dirty, set writeback and unlock the pages. | |
847 | */ | |
c2790a2e | 848 | extent_clear_unlock_delalloc(inode, async_extent->start, |
ba8b04c1 QW |
849 | async_extent->start + |
850 | async_extent->ram_size - 1, | |
a791e35e CM |
851 | async_extent->start + |
852 | async_extent->ram_size - 1, | |
151a41bc JB |
853 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC, |
854 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
c2790a2e | 855 | PAGE_SET_WRITEBACK); |
4e4cbee9 | 856 | if (btrfs_submit_compressed_write(inode, |
d397712b CM |
857 | async_extent->start, |
858 | async_extent->ram_size, | |
859 | ins.objectid, | |
860 | ins.offset, async_extent->pages, | |
f82b7359 LB |
861 | async_extent->nr_pages, |
862 | async_cow->write_flags)) { | |
fce2a4e6 FM |
863 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
864 | struct page *p = async_extent->pages[0]; | |
865 | const u64 start = async_extent->start; | |
866 | const u64 end = start + async_extent->ram_size - 1; | |
867 | ||
868 | p->mapping = inode->i_mapping; | |
869 | tree->ops->writepage_end_io_hook(p, start, end, | |
870 | NULL, 0); | |
871 | p->mapping = NULL; | |
ba8b04c1 QW |
872 | extent_clear_unlock_delalloc(inode, start, end, end, |
873 | NULL, 0, | |
fce2a4e6 FM |
874 | PAGE_END_WRITEBACK | |
875 | PAGE_SET_ERROR); | |
40ae837b | 876 | free_async_extent_pages(async_extent); |
fce2a4e6 | 877 | } |
771ed689 CM |
878 | alloc_hint = ins.objectid + ins.offset; |
879 | kfree(async_extent); | |
880 | cond_resched(); | |
881 | } | |
dec8f175 | 882 | return; |
3e04e7f1 | 883 | out_free_reserve: |
0b246afa | 884 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 885 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 886 | out_free: |
c2790a2e | 887 | extent_clear_unlock_delalloc(inode, async_extent->start, |
ba8b04c1 QW |
888 | async_extent->start + |
889 | async_extent->ram_size - 1, | |
3e04e7f1 JB |
890 | async_extent->start + |
891 | async_extent->ram_size - 1, | |
c2790a2e | 892 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC | |
a7e3b975 | 893 | EXTENT_DELALLOC_NEW | |
151a41bc JB |
894 | EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING, |
895 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
704de49d FM |
896 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK | |
897 | PAGE_SET_ERROR); | |
40ae837b | 898 | free_async_extent_pages(async_extent); |
79787eaa | 899 | kfree(async_extent); |
3e04e7f1 | 900 | goto again; |
771ed689 CM |
901 | } |
902 | ||
4b46fce2 JB |
903 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
904 | u64 num_bytes) | |
905 | { | |
906 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
907 | struct extent_map *em; | |
908 | u64 alloc_hint = 0; | |
909 | ||
910 | read_lock(&em_tree->lock); | |
911 | em = search_extent_mapping(em_tree, start, num_bytes); | |
912 | if (em) { | |
913 | /* | |
914 | * if block start isn't an actual block number then find the | |
915 | * first block in this inode and use that as a hint. If that | |
916 | * block is also bogus then just don't worry about it. | |
917 | */ | |
918 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
919 | free_extent_map(em); | |
920 | em = search_extent_mapping(em_tree, 0, 0); | |
921 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
922 | alloc_hint = em->block_start; | |
923 | if (em) | |
924 | free_extent_map(em); | |
925 | } else { | |
926 | alloc_hint = em->block_start; | |
927 | free_extent_map(em); | |
928 | } | |
929 | } | |
930 | read_unlock(&em_tree->lock); | |
931 | ||
932 | return alloc_hint; | |
933 | } | |
934 | ||
771ed689 CM |
935 | /* |
936 | * when extent_io.c finds a delayed allocation range in the file, | |
937 | * the call backs end up in this code. The basic idea is to | |
938 | * allocate extents on disk for the range, and create ordered data structs | |
939 | * in ram to track those extents. | |
940 | * | |
941 | * locked_page is the page that writepage had locked already. We use | |
942 | * it to make sure we don't do extra locks or unlocks. | |
943 | * | |
944 | * *page_started is set to one if we unlock locked_page and do everything | |
945 | * required to start IO on it. It may be clean and already done with | |
946 | * IO when we return. | |
947 | */ | |
00361589 JB |
948 | static noinline int cow_file_range(struct inode *inode, |
949 | struct page *locked_page, | |
dda3245e WX |
950 | u64 start, u64 end, u64 delalloc_end, |
951 | int *page_started, unsigned long *nr_written, | |
952 | int unlock, struct btrfs_dedupe_hash *hash) | |
771ed689 | 953 | { |
0b246afa | 954 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
00361589 | 955 | struct btrfs_root *root = BTRFS_I(inode)->root; |
771ed689 CM |
956 | u64 alloc_hint = 0; |
957 | u64 num_bytes; | |
958 | unsigned long ram_size; | |
959 | u64 disk_num_bytes; | |
a315e68f | 960 | u64 cur_alloc_size = 0; |
0b246afa | 961 | u64 blocksize = fs_info->sectorsize; |
771ed689 CM |
962 | struct btrfs_key ins; |
963 | struct extent_map *em; | |
a315e68f FM |
964 | unsigned clear_bits; |
965 | unsigned long page_ops; | |
966 | bool extent_reserved = false; | |
771ed689 CM |
967 | int ret = 0; |
968 | ||
70ddc553 | 969 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) { |
02ecd2c2 | 970 | WARN_ON_ONCE(1); |
29bce2f3 JB |
971 | ret = -EINVAL; |
972 | goto out_unlock; | |
02ecd2c2 | 973 | } |
771ed689 | 974 | |
fda2832f | 975 | num_bytes = ALIGN(end - start + 1, blocksize); |
771ed689 CM |
976 | num_bytes = max(blocksize, num_bytes); |
977 | disk_num_bytes = num_bytes; | |
771ed689 | 978 | |
6158e1ce | 979 | inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K); |
4cb5300b | 980 | |
771ed689 CM |
981 | if (start == 0) { |
982 | /* lets try to make an inline extent */ | |
f74670f7 AJ |
983 | ret = cow_file_range_inline(root, inode, start, end, 0, |
984 | BTRFS_COMPRESS_NONE, NULL); | |
771ed689 | 985 | if (ret == 0) { |
8b62f87b JB |
986 | /* |
987 | * We use DO_ACCOUNTING here because we need the | |
988 | * delalloc_release_metadata to be run _after_ we drop | |
989 | * our outstanding extent for clearing delalloc for this | |
990 | * range. | |
991 | */ | |
ba8b04c1 QW |
992 | extent_clear_unlock_delalloc(inode, start, end, |
993 | delalloc_end, NULL, | |
c2790a2e | 994 | EXTENT_LOCKED | EXTENT_DELALLOC | |
8b62f87b JB |
995 | EXTENT_DELALLOC_NEW | EXTENT_DEFRAG | |
996 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
c2790a2e JB |
997 | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
998 | PAGE_END_WRITEBACK); | |
771ed689 | 999 | *nr_written = *nr_written + |
09cbfeaf | 1000 | (end - start + PAGE_SIZE) / PAGE_SIZE; |
771ed689 | 1001 | *page_started = 1; |
771ed689 | 1002 | goto out; |
79787eaa | 1003 | } else if (ret < 0) { |
79787eaa | 1004 | goto out_unlock; |
771ed689 CM |
1005 | } |
1006 | } | |
1007 | ||
1008 | BUG_ON(disk_num_bytes > | |
0b246afa | 1009 | btrfs_super_total_bytes(fs_info->super_copy)); |
771ed689 | 1010 | |
4b46fce2 | 1011 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
dcdbc059 NB |
1012 | btrfs_drop_extent_cache(BTRFS_I(inode), start, |
1013 | start + num_bytes - 1, 0); | |
771ed689 | 1014 | |
d397712b | 1015 | while (disk_num_bytes > 0) { |
287a0ab9 | 1016 | cur_alloc_size = disk_num_bytes; |
18513091 | 1017 | ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size, |
0b246afa | 1018 | fs_info->sectorsize, 0, alloc_hint, |
e570fd27 | 1019 | &ins, 1, 1); |
00361589 | 1020 | if (ret < 0) |
79787eaa | 1021 | goto out_unlock; |
a315e68f FM |
1022 | cur_alloc_size = ins.offset; |
1023 | extent_reserved = true; | |
d397712b | 1024 | |
771ed689 | 1025 | ram_size = ins.offset; |
6f9994db LB |
1026 | em = create_io_em(inode, start, ins.offset, /* len */ |
1027 | start, /* orig_start */ | |
1028 | ins.objectid, /* block_start */ | |
1029 | ins.offset, /* block_len */ | |
1030 | ins.offset, /* orig_block_len */ | |
1031 | ram_size, /* ram_bytes */ | |
1032 | BTRFS_COMPRESS_NONE, /* compress_type */ | |
1af4a0aa | 1033 | BTRFS_ORDERED_REGULAR /* type */); |
6f9994db | 1034 | if (IS_ERR(em)) |
ace68bac | 1035 | goto out_reserve; |
6f9994db | 1036 | free_extent_map(em); |
e6dcd2dc | 1037 | |
e6dcd2dc | 1038 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 1039 | ram_size, cur_alloc_size, 0); |
ace68bac | 1040 | if (ret) |
d9f85963 | 1041 | goto out_drop_extent_cache; |
c8b97818 | 1042 | |
17d217fe YZ |
1043 | if (root->root_key.objectid == |
1044 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1045 | ret = btrfs_reloc_clone_csums(inode, start, | |
1046 | cur_alloc_size); | |
4dbd80fb QW |
1047 | /* |
1048 | * Only drop cache here, and process as normal. | |
1049 | * | |
1050 | * We must not allow extent_clear_unlock_delalloc() | |
1051 | * at out_unlock label to free meta of this ordered | |
1052 | * extent, as its meta should be freed by | |
1053 | * btrfs_finish_ordered_io(). | |
1054 | * | |
1055 | * So we must continue until @start is increased to | |
1056 | * skip current ordered extent. | |
1057 | */ | |
00361589 | 1058 | if (ret) |
4dbd80fb QW |
1059 | btrfs_drop_extent_cache(BTRFS_I(inode), start, |
1060 | start + ram_size - 1, 0); | |
17d217fe YZ |
1061 | } |
1062 | ||
0b246afa | 1063 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
9cfa3e34 | 1064 | |
c8b97818 CM |
1065 | /* we're not doing compressed IO, don't unlock the first |
1066 | * page (which the caller expects to stay locked), don't | |
1067 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
1068 | * |
1069 | * Do set the Private2 bit so we know this page was properly | |
1070 | * setup for writepage | |
c8b97818 | 1071 | */ |
a315e68f FM |
1072 | page_ops = unlock ? PAGE_UNLOCK : 0; |
1073 | page_ops |= PAGE_SET_PRIVATE2; | |
a791e35e | 1074 | |
c2790a2e | 1075 | extent_clear_unlock_delalloc(inode, start, |
ba8b04c1 QW |
1076 | start + ram_size - 1, |
1077 | delalloc_end, locked_page, | |
c2790a2e | 1078 | EXTENT_LOCKED | EXTENT_DELALLOC, |
a315e68f | 1079 | page_ops); |
4dbd80fb QW |
1080 | if (disk_num_bytes < cur_alloc_size) |
1081 | disk_num_bytes = 0; | |
1082 | else | |
1083 | disk_num_bytes -= cur_alloc_size; | |
c59f8951 CM |
1084 | num_bytes -= cur_alloc_size; |
1085 | alloc_hint = ins.objectid + ins.offset; | |
1086 | start += cur_alloc_size; | |
a315e68f | 1087 | extent_reserved = false; |
4dbd80fb QW |
1088 | |
1089 | /* | |
1090 | * btrfs_reloc_clone_csums() error, since start is increased | |
1091 | * extent_clear_unlock_delalloc() at out_unlock label won't | |
1092 | * free metadata of current ordered extent, we're OK to exit. | |
1093 | */ | |
1094 | if (ret) | |
1095 | goto out_unlock; | |
b888db2b | 1096 | } |
79787eaa | 1097 | out: |
be20aa9d | 1098 | return ret; |
b7d5b0a8 | 1099 | |
d9f85963 | 1100 | out_drop_extent_cache: |
dcdbc059 | 1101 | btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0); |
ace68bac | 1102 | out_reserve: |
0b246afa | 1103 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 1104 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 1105 | out_unlock: |
a7e3b975 FM |
1106 | clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW | |
1107 | EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV; | |
a315e68f FM |
1108 | page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
1109 | PAGE_END_WRITEBACK; | |
1110 | /* | |
1111 | * If we reserved an extent for our delalloc range (or a subrange) and | |
1112 | * failed to create the respective ordered extent, then it means that | |
1113 | * when we reserved the extent we decremented the extent's size from | |
1114 | * the data space_info's bytes_may_use counter and incremented the | |
1115 | * space_info's bytes_reserved counter by the same amount. We must make | |
1116 | * sure extent_clear_unlock_delalloc() does not try to decrement again | |
1117 | * the data space_info's bytes_may_use counter, therefore we do not pass | |
1118 | * it the flag EXTENT_CLEAR_DATA_RESV. | |
1119 | */ | |
1120 | if (extent_reserved) { | |
1121 | extent_clear_unlock_delalloc(inode, start, | |
1122 | start + cur_alloc_size, | |
1123 | start + cur_alloc_size, | |
1124 | locked_page, | |
1125 | clear_bits, | |
1126 | page_ops); | |
1127 | start += cur_alloc_size; | |
1128 | if (start >= end) | |
1129 | goto out; | |
1130 | } | |
ba8b04c1 QW |
1131 | extent_clear_unlock_delalloc(inode, start, end, delalloc_end, |
1132 | locked_page, | |
a315e68f FM |
1133 | clear_bits | EXTENT_CLEAR_DATA_RESV, |
1134 | page_ops); | |
79787eaa | 1135 | goto out; |
771ed689 | 1136 | } |
c8b97818 | 1137 | |
771ed689 CM |
1138 | /* |
1139 | * work queue call back to started compression on a file and pages | |
1140 | */ | |
1141 | static noinline void async_cow_start(struct btrfs_work *work) | |
1142 | { | |
1143 | struct async_cow *async_cow; | |
1144 | int num_added = 0; | |
1145 | async_cow = container_of(work, struct async_cow, work); | |
1146 | ||
1147 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
1148 | async_cow->start, async_cow->end, async_cow, | |
1149 | &num_added); | |
8180ef88 | 1150 | if (num_added == 0) { |
cb77fcd8 | 1151 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 | 1152 | async_cow->inode = NULL; |
8180ef88 | 1153 | } |
771ed689 CM |
1154 | } |
1155 | ||
1156 | /* | |
1157 | * work queue call back to submit previously compressed pages | |
1158 | */ | |
1159 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1160 | { | |
0b246afa | 1161 | struct btrfs_fs_info *fs_info; |
771ed689 CM |
1162 | struct async_cow *async_cow; |
1163 | struct btrfs_root *root; | |
1164 | unsigned long nr_pages; | |
1165 | ||
1166 | async_cow = container_of(work, struct async_cow, work); | |
1167 | ||
1168 | root = async_cow->root; | |
0b246afa | 1169 | fs_info = root->fs_info; |
09cbfeaf KS |
1170 | nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >> |
1171 | PAGE_SHIFT; | |
771ed689 | 1172 | |
ee863954 DS |
1173 | /* |
1174 | * atomic_sub_return implies a barrier for waitqueue_active | |
1175 | */ | |
0b246afa | 1176 | if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) < |
ee22184b | 1177 | 5 * SZ_1M && |
0b246afa JM |
1178 | waitqueue_active(&fs_info->async_submit_wait)) |
1179 | wake_up(&fs_info->async_submit_wait); | |
771ed689 | 1180 | |
d397712b | 1181 | if (async_cow->inode) |
771ed689 | 1182 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 1183 | } |
c8b97818 | 1184 | |
771ed689 CM |
1185 | static noinline void async_cow_free(struct btrfs_work *work) |
1186 | { | |
1187 | struct async_cow *async_cow; | |
1188 | async_cow = container_of(work, struct async_cow, work); | |
8180ef88 | 1189 | if (async_cow->inode) |
cb77fcd8 | 1190 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 CM |
1191 | kfree(async_cow); |
1192 | } | |
1193 | ||
1194 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
1195 | u64 start, u64 end, int *page_started, | |
f82b7359 LB |
1196 | unsigned long *nr_written, |
1197 | unsigned int write_flags) | |
771ed689 | 1198 | { |
0b246afa | 1199 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
771ed689 CM |
1200 | struct async_cow *async_cow; |
1201 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1202 | unsigned long nr_pages; | |
1203 | u64 cur_end; | |
771ed689 | 1204 | |
a3429ab7 | 1205 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
ae0f1625 | 1206 | 1, 0, NULL); |
d397712b | 1207 | while (start < end) { |
771ed689 | 1208 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
79787eaa | 1209 | BUG_ON(!async_cow); /* -ENOMEM */ |
8180ef88 | 1210 | async_cow->inode = igrab(inode); |
771ed689 CM |
1211 | async_cow->root = root; |
1212 | async_cow->locked_page = locked_page; | |
1213 | async_cow->start = start; | |
f82b7359 | 1214 | async_cow->write_flags = write_flags; |
771ed689 | 1215 | |
f79707b0 | 1216 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS && |
0b246afa | 1217 | !btrfs_test_opt(fs_info, FORCE_COMPRESS)) |
771ed689 CM |
1218 | cur_end = end; |
1219 | else | |
ee22184b | 1220 | cur_end = min(end, start + SZ_512K - 1); |
771ed689 CM |
1221 | |
1222 | async_cow->end = cur_end; | |
1223 | INIT_LIST_HEAD(&async_cow->extents); | |
1224 | ||
9e0af237 LB |
1225 | btrfs_init_work(&async_cow->work, |
1226 | btrfs_delalloc_helper, | |
1227 | async_cow_start, async_cow_submit, | |
1228 | async_cow_free); | |
771ed689 | 1229 | |
09cbfeaf KS |
1230 | nr_pages = (cur_end - start + PAGE_SIZE) >> |
1231 | PAGE_SHIFT; | |
0b246afa | 1232 | atomic_add(nr_pages, &fs_info->async_delalloc_pages); |
771ed689 | 1233 | |
0b246afa | 1234 | btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work); |
771ed689 | 1235 | |
771ed689 CM |
1236 | *nr_written += nr_pages; |
1237 | start = cur_end + 1; | |
1238 | } | |
1239 | *page_started = 1; | |
1240 | return 0; | |
be20aa9d CM |
1241 | } |
1242 | ||
2ff7e61e | 1243 | static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info, |
17d217fe YZ |
1244 | u64 bytenr, u64 num_bytes) |
1245 | { | |
1246 | int ret; | |
1247 | struct btrfs_ordered_sum *sums; | |
1248 | LIST_HEAD(list); | |
1249 | ||
0b246afa | 1250 | ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr, |
a2de733c | 1251 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1252 | if (ret == 0 && list_empty(&list)) |
1253 | return 0; | |
1254 | ||
1255 | while (!list_empty(&list)) { | |
1256 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1257 | list_del(&sums->list); | |
1258 | kfree(sums); | |
1259 | } | |
1260 | return 1; | |
1261 | } | |
1262 | ||
d352ac68 CM |
1263 | /* |
1264 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1265 | * of the extents that exist in the file, and COWs the file as required. | |
1266 | * | |
1267 | * If no cow copies or snapshots exist, we write directly to the existing | |
1268 | * blocks on disk | |
1269 | */ | |
7f366cfe CM |
1270 | static noinline int run_delalloc_nocow(struct inode *inode, |
1271 | struct page *locked_page, | |
771ed689 CM |
1272 | u64 start, u64 end, int *page_started, int force, |
1273 | unsigned long *nr_written) | |
be20aa9d | 1274 | { |
0b246afa | 1275 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
be20aa9d CM |
1276 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1277 | struct extent_buffer *leaf; | |
be20aa9d | 1278 | struct btrfs_path *path; |
80ff3856 | 1279 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1280 | struct btrfs_key found_key; |
6f9994db | 1281 | struct extent_map *em; |
80ff3856 YZ |
1282 | u64 cow_start; |
1283 | u64 cur_offset; | |
1284 | u64 extent_end; | |
5d4f98a2 | 1285 | u64 extent_offset; |
80ff3856 YZ |
1286 | u64 disk_bytenr; |
1287 | u64 num_bytes; | |
b4939680 | 1288 | u64 disk_num_bytes; |
cc95bef6 | 1289 | u64 ram_bytes; |
80ff3856 | 1290 | int extent_type; |
79787eaa | 1291 | int ret, err; |
d899e052 | 1292 | int type; |
80ff3856 YZ |
1293 | int nocow; |
1294 | int check_prev = 1; | |
82d5902d | 1295 | bool nolock; |
4a0cc7ca | 1296 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
be20aa9d CM |
1297 | |
1298 | path = btrfs_alloc_path(); | |
17ca04af | 1299 | if (!path) { |
ba8b04c1 QW |
1300 | extent_clear_unlock_delalloc(inode, start, end, end, |
1301 | locked_page, | |
c2790a2e | 1302 | EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc JB |
1303 | EXTENT_DO_ACCOUNTING | |
1304 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1305 | PAGE_CLEAR_DIRTY | |
1306 | PAGE_SET_WRITEBACK | | |
1307 | PAGE_END_WRITEBACK); | |
d8926bb3 | 1308 | return -ENOMEM; |
17ca04af | 1309 | } |
82d5902d | 1310 | |
70ddc553 | 1311 | nolock = btrfs_is_free_space_inode(BTRFS_I(inode)); |
82d5902d | 1312 | |
80ff3856 YZ |
1313 | cow_start = (u64)-1; |
1314 | cur_offset = start; | |
1315 | while (1) { | |
e4c3b2dc | 1316 | ret = btrfs_lookup_file_extent(NULL, root, path, ino, |
80ff3856 | 1317 | cur_offset, 0); |
d788a349 | 1318 | if (ret < 0) |
79787eaa | 1319 | goto error; |
80ff3856 YZ |
1320 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1321 | leaf = path->nodes[0]; | |
1322 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1323 | path->slots[0] - 1); | |
33345d01 | 1324 | if (found_key.objectid == ino && |
80ff3856 YZ |
1325 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1326 | path->slots[0]--; | |
1327 | } | |
1328 | check_prev = 0; | |
1329 | next_slot: | |
1330 | leaf = path->nodes[0]; | |
1331 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1332 | ret = btrfs_next_leaf(root, path); | |
d788a349 | 1333 | if (ret < 0) |
79787eaa | 1334 | goto error; |
80ff3856 YZ |
1335 | if (ret > 0) |
1336 | break; | |
1337 | leaf = path->nodes[0]; | |
1338 | } | |
be20aa9d | 1339 | |
80ff3856 YZ |
1340 | nocow = 0; |
1341 | disk_bytenr = 0; | |
17d217fe | 1342 | num_bytes = 0; |
80ff3856 YZ |
1343 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1344 | ||
1d512cb7 FM |
1345 | if (found_key.objectid > ino) |
1346 | break; | |
1347 | if (WARN_ON_ONCE(found_key.objectid < ino) || | |
1348 | found_key.type < BTRFS_EXTENT_DATA_KEY) { | |
1349 | path->slots[0]++; | |
1350 | goto next_slot; | |
1351 | } | |
1352 | if (found_key.type > BTRFS_EXTENT_DATA_KEY || | |
80ff3856 YZ |
1353 | found_key.offset > end) |
1354 | break; | |
1355 | ||
1356 | if (found_key.offset > cur_offset) { | |
1357 | extent_end = found_key.offset; | |
e9061e21 | 1358 | extent_type = 0; |
80ff3856 YZ |
1359 | goto out_check; |
1360 | } | |
1361 | ||
1362 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1363 | struct btrfs_file_extent_item); | |
1364 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1365 | ||
cc95bef6 | 1366 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
d899e052 YZ |
1367 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1368 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1369 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1370 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1371 | extent_end = found_key.offset + |
1372 | btrfs_file_extent_num_bytes(leaf, fi); | |
b4939680 JB |
1373 | disk_num_bytes = |
1374 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
80ff3856 YZ |
1375 | if (extent_end <= start) { |
1376 | path->slots[0]++; | |
1377 | goto next_slot; | |
1378 | } | |
17d217fe YZ |
1379 | if (disk_bytenr == 0) |
1380 | goto out_check; | |
80ff3856 YZ |
1381 | if (btrfs_file_extent_compression(leaf, fi) || |
1382 | btrfs_file_extent_encryption(leaf, fi) || | |
1383 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1384 | goto out_check; | |
d899e052 YZ |
1385 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1386 | goto out_check; | |
2ff7e61e | 1387 | if (btrfs_extent_readonly(fs_info, disk_bytenr)) |
80ff3856 | 1388 | goto out_check; |
e4c3b2dc | 1389 | if (btrfs_cross_ref_exist(root, ino, |
5d4f98a2 YZ |
1390 | found_key.offset - |
1391 | extent_offset, disk_bytenr)) | |
17d217fe | 1392 | goto out_check; |
5d4f98a2 | 1393 | disk_bytenr += extent_offset; |
17d217fe YZ |
1394 | disk_bytenr += cur_offset - found_key.offset; |
1395 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
e9894fd3 WS |
1396 | /* |
1397 | * if there are pending snapshots for this root, | |
1398 | * we fall into common COW way. | |
1399 | */ | |
1400 | if (!nolock) { | |
ea14b57f | 1401 | err = btrfs_start_write_no_snapshotting(root); |
e9894fd3 WS |
1402 | if (!err) |
1403 | goto out_check; | |
1404 | } | |
17d217fe YZ |
1405 | /* |
1406 | * force cow if csum exists in the range. | |
1407 | * this ensure that csum for a given extent are | |
1408 | * either valid or do not exist. | |
1409 | */ | |
2ff7e61e | 1410 | if (csum_exist_in_range(fs_info, disk_bytenr, |
91e1f56a RK |
1411 | num_bytes)) { |
1412 | if (!nolock) | |
ea14b57f | 1413 | btrfs_end_write_no_snapshotting(root); |
17d217fe | 1414 | goto out_check; |
91e1f56a RK |
1415 | } |
1416 | if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) { | |
1417 | if (!nolock) | |
ea14b57f | 1418 | btrfs_end_write_no_snapshotting(root); |
f78c436c | 1419 | goto out_check; |
91e1f56a | 1420 | } |
80ff3856 YZ |
1421 | nocow = 1; |
1422 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1423 | extent_end = found_key.offset + | |
514ac8ad CM |
1424 | btrfs_file_extent_inline_len(leaf, |
1425 | path->slots[0], fi); | |
da17066c | 1426 | extent_end = ALIGN(extent_end, |
0b246afa | 1427 | fs_info->sectorsize); |
80ff3856 YZ |
1428 | } else { |
1429 | BUG_ON(1); | |
1430 | } | |
1431 | out_check: | |
1432 | if (extent_end <= start) { | |
1433 | path->slots[0]++; | |
e9894fd3 | 1434 | if (!nolock && nocow) |
ea14b57f | 1435 | btrfs_end_write_no_snapshotting(root); |
f78c436c | 1436 | if (nocow) |
0b246afa | 1437 | btrfs_dec_nocow_writers(fs_info, disk_bytenr); |
80ff3856 YZ |
1438 | goto next_slot; |
1439 | } | |
1440 | if (!nocow) { | |
1441 | if (cow_start == (u64)-1) | |
1442 | cow_start = cur_offset; | |
1443 | cur_offset = extent_end; | |
1444 | if (cur_offset > end) | |
1445 | break; | |
1446 | path->slots[0]++; | |
1447 | goto next_slot; | |
7ea394f1 YZ |
1448 | } |
1449 | ||
b3b4aa74 | 1450 | btrfs_release_path(path); |
80ff3856 | 1451 | if (cow_start != (u64)-1) { |
00361589 JB |
1452 | ret = cow_file_range(inode, locked_page, |
1453 | cow_start, found_key.offset - 1, | |
dda3245e WX |
1454 | end, page_started, nr_written, 1, |
1455 | NULL); | |
e9894fd3 WS |
1456 | if (ret) { |
1457 | if (!nolock && nocow) | |
ea14b57f | 1458 | btrfs_end_write_no_snapshotting(root); |
f78c436c | 1459 | if (nocow) |
0b246afa | 1460 | btrfs_dec_nocow_writers(fs_info, |
f78c436c | 1461 | disk_bytenr); |
79787eaa | 1462 | goto error; |
e9894fd3 | 1463 | } |
80ff3856 | 1464 | cow_start = (u64)-1; |
7ea394f1 | 1465 | } |
80ff3856 | 1466 | |
d899e052 | 1467 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
6f9994db LB |
1468 | u64 orig_start = found_key.offset - extent_offset; |
1469 | ||
1470 | em = create_io_em(inode, cur_offset, num_bytes, | |
1471 | orig_start, | |
1472 | disk_bytenr, /* block_start */ | |
1473 | num_bytes, /* block_len */ | |
1474 | disk_num_bytes, /* orig_block_len */ | |
1475 | ram_bytes, BTRFS_COMPRESS_NONE, | |
1476 | BTRFS_ORDERED_PREALLOC); | |
1477 | if (IS_ERR(em)) { | |
1478 | if (!nolock && nocow) | |
ea14b57f | 1479 | btrfs_end_write_no_snapshotting(root); |
6f9994db LB |
1480 | if (nocow) |
1481 | btrfs_dec_nocow_writers(fs_info, | |
1482 | disk_bytenr); | |
1483 | ret = PTR_ERR(em); | |
1484 | goto error; | |
d899e052 | 1485 | } |
6f9994db LB |
1486 | free_extent_map(em); |
1487 | } | |
1488 | ||
1489 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
d899e052 YZ |
1490 | type = BTRFS_ORDERED_PREALLOC; |
1491 | } else { | |
1492 | type = BTRFS_ORDERED_NOCOW; | |
1493 | } | |
80ff3856 YZ |
1494 | |
1495 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 | 1496 | num_bytes, num_bytes, type); |
f78c436c | 1497 | if (nocow) |
0b246afa | 1498 | btrfs_dec_nocow_writers(fs_info, disk_bytenr); |
79787eaa | 1499 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 1500 | |
efa56464 | 1501 | if (root->root_key.objectid == |
4dbd80fb QW |
1502 | BTRFS_DATA_RELOC_TREE_OBJECTID) |
1503 | /* | |
1504 | * Error handled later, as we must prevent | |
1505 | * extent_clear_unlock_delalloc() in error handler | |
1506 | * from freeing metadata of created ordered extent. | |
1507 | */ | |
efa56464 YZ |
1508 | ret = btrfs_reloc_clone_csums(inode, cur_offset, |
1509 | num_bytes); | |
efa56464 | 1510 | |
c2790a2e | 1511 | extent_clear_unlock_delalloc(inode, cur_offset, |
ba8b04c1 | 1512 | cur_offset + num_bytes - 1, end, |
c2790a2e | 1513 | locked_page, EXTENT_LOCKED | |
18513091 WX |
1514 | EXTENT_DELALLOC | |
1515 | EXTENT_CLEAR_DATA_RESV, | |
1516 | PAGE_UNLOCK | PAGE_SET_PRIVATE2); | |
1517 | ||
e9894fd3 | 1518 | if (!nolock && nocow) |
ea14b57f | 1519 | btrfs_end_write_no_snapshotting(root); |
80ff3856 | 1520 | cur_offset = extent_end; |
4dbd80fb QW |
1521 | |
1522 | /* | |
1523 | * btrfs_reloc_clone_csums() error, now we're OK to call error | |
1524 | * handler, as metadata for created ordered extent will only | |
1525 | * be freed by btrfs_finish_ordered_io(). | |
1526 | */ | |
1527 | if (ret) | |
1528 | goto error; | |
80ff3856 YZ |
1529 | if (cur_offset > end) |
1530 | break; | |
be20aa9d | 1531 | } |
b3b4aa74 | 1532 | btrfs_release_path(path); |
80ff3856 | 1533 | |
17ca04af | 1534 | if (cur_offset <= end && cow_start == (u64)-1) { |
80ff3856 | 1535 | cow_start = cur_offset; |
17ca04af JB |
1536 | cur_offset = end; |
1537 | } | |
1538 | ||
80ff3856 | 1539 | if (cow_start != (u64)-1) { |
dda3245e WX |
1540 | ret = cow_file_range(inode, locked_page, cow_start, end, end, |
1541 | page_started, nr_written, 1, NULL); | |
d788a349 | 1542 | if (ret) |
79787eaa | 1543 | goto error; |
80ff3856 YZ |
1544 | } |
1545 | ||
79787eaa | 1546 | error: |
17ca04af | 1547 | if (ret && cur_offset < end) |
ba8b04c1 | 1548 | extent_clear_unlock_delalloc(inode, cur_offset, end, end, |
c2790a2e | 1549 | locked_page, EXTENT_LOCKED | |
151a41bc JB |
1550 | EXTENT_DELALLOC | EXTENT_DEFRAG | |
1551 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
1552 | PAGE_CLEAR_DIRTY | | |
c2790a2e JB |
1553 | PAGE_SET_WRITEBACK | |
1554 | PAGE_END_WRITEBACK); | |
7ea394f1 | 1555 | btrfs_free_path(path); |
79787eaa | 1556 | return ret; |
be20aa9d CM |
1557 | } |
1558 | ||
47059d93 WS |
1559 | static inline int need_force_cow(struct inode *inode, u64 start, u64 end) |
1560 | { | |
1561 | ||
1562 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
1563 | !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)) | |
1564 | return 0; | |
1565 | ||
1566 | /* | |
1567 | * @defrag_bytes is a hint value, no spinlock held here, | |
1568 | * if is not zero, it means the file is defragging. | |
1569 | * Force cow if given extent needs to be defragged. | |
1570 | */ | |
1571 | if (BTRFS_I(inode)->defrag_bytes && | |
1572 | test_range_bit(&BTRFS_I(inode)->io_tree, start, end, | |
1573 | EXTENT_DEFRAG, 0, NULL)) | |
1574 | return 1; | |
1575 | ||
1576 | return 0; | |
1577 | } | |
1578 | ||
d352ac68 CM |
1579 | /* |
1580 | * extent_io.c call back to do delayed allocation processing | |
1581 | */ | |
c6100a4b | 1582 | static int run_delalloc_range(void *private_data, struct page *locked_page, |
771ed689 | 1583 | u64 start, u64 end, int *page_started, |
f82b7359 LB |
1584 | unsigned long *nr_written, |
1585 | struct writeback_control *wbc) | |
be20aa9d | 1586 | { |
c6100a4b | 1587 | struct inode *inode = private_data; |
be20aa9d | 1588 | int ret; |
47059d93 | 1589 | int force_cow = need_force_cow(inode, start, end); |
f82b7359 | 1590 | unsigned int write_flags = wbc_to_write_flags(wbc); |
a2135011 | 1591 | |
47059d93 | 1592 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) { |
c8b97818 | 1593 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1594 | page_started, 1, nr_written); |
47059d93 | 1595 | } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) { |
d899e052 | 1596 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1597 | page_started, 0, nr_written); |
c2fcdcdf | 1598 | } else if (!inode_need_compress(inode, start, end)) { |
dda3245e WX |
1599 | ret = cow_file_range(inode, locked_page, start, end, end, |
1600 | page_started, nr_written, 1, NULL); | |
7ddf5a42 JB |
1601 | } else { |
1602 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
1603 | &BTRFS_I(inode)->runtime_flags); | |
771ed689 | 1604 | ret = cow_file_range_async(inode, locked_page, start, end, |
f82b7359 LB |
1605 | page_started, nr_written, |
1606 | write_flags); | |
7ddf5a42 | 1607 | } |
52427260 QW |
1608 | if (ret) |
1609 | btrfs_cleanup_ordered_extents(inode, start, end - start + 1); | |
b888db2b CM |
1610 | return ret; |
1611 | } | |
1612 | ||
c6100a4b | 1613 | static void btrfs_split_extent_hook(void *private_data, |
1bf85046 | 1614 | struct extent_state *orig, u64 split) |
9ed74f2d | 1615 | { |
c6100a4b | 1616 | struct inode *inode = private_data; |
dcab6a3b JB |
1617 | u64 size; |
1618 | ||
0ca1f7ce | 1619 | /* not delalloc, ignore it */ |
9ed74f2d | 1620 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1621 | return; |
9ed74f2d | 1622 | |
dcab6a3b JB |
1623 | size = orig->end - orig->start + 1; |
1624 | if (size > BTRFS_MAX_EXTENT_SIZE) { | |
823bb20a | 1625 | u32 num_extents; |
dcab6a3b JB |
1626 | u64 new_size; |
1627 | ||
1628 | /* | |
ba117213 JB |
1629 | * See the explanation in btrfs_merge_extent_hook, the same |
1630 | * applies here, just in reverse. | |
dcab6a3b JB |
1631 | */ |
1632 | new_size = orig->end - split + 1; | |
823bb20a | 1633 | num_extents = count_max_extents(new_size); |
ba117213 | 1634 | new_size = split - orig->start; |
823bb20a DS |
1635 | num_extents += count_max_extents(new_size); |
1636 | if (count_max_extents(size) >= num_extents) | |
dcab6a3b JB |
1637 | return; |
1638 | } | |
1639 | ||
9e0baf60 | 1640 | spin_lock(&BTRFS_I(inode)->lock); |
8b62f87b | 1641 | btrfs_mod_outstanding_extents(BTRFS_I(inode), 1); |
9e0baf60 | 1642 | spin_unlock(&BTRFS_I(inode)->lock); |
9ed74f2d JB |
1643 | } |
1644 | ||
1645 | /* | |
1646 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1647 | * extents so we can keep track of new extents that are just merged onto old | |
1648 | * extents, such as when we are doing sequential writes, so we can properly | |
1649 | * account for the metadata space we'll need. | |
1650 | */ | |
c6100a4b | 1651 | static void btrfs_merge_extent_hook(void *private_data, |
1bf85046 JM |
1652 | struct extent_state *new, |
1653 | struct extent_state *other) | |
9ed74f2d | 1654 | { |
c6100a4b | 1655 | struct inode *inode = private_data; |
dcab6a3b | 1656 | u64 new_size, old_size; |
823bb20a | 1657 | u32 num_extents; |
dcab6a3b | 1658 | |
9ed74f2d JB |
1659 | /* not delalloc, ignore it */ |
1660 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1661 | return; |
9ed74f2d | 1662 | |
8461a3de JB |
1663 | if (new->start > other->start) |
1664 | new_size = new->end - other->start + 1; | |
1665 | else | |
1666 | new_size = other->end - new->start + 1; | |
dcab6a3b JB |
1667 | |
1668 | /* we're not bigger than the max, unreserve the space and go */ | |
1669 | if (new_size <= BTRFS_MAX_EXTENT_SIZE) { | |
1670 | spin_lock(&BTRFS_I(inode)->lock); | |
8b62f87b | 1671 | btrfs_mod_outstanding_extents(BTRFS_I(inode), -1); |
dcab6a3b JB |
1672 | spin_unlock(&BTRFS_I(inode)->lock); |
1673 | return; | |
1674 | } | |
1675 | ||
1676 | /* | |
ba117213 JB |
1677 | * We have to add up either side to figure out how many extents were |
1678 | * accounted for before we merged into one big extent. If the number of | |
1679 | * extents we accounted for is <= the amount we need for the new range | |
1680 | * then we can return, otherwise drop. Think of it like this | |
1681 | * | |
1682 | * [ 4k][MAX_SIZE] | |
1683 | * | |
1684 | * So we've grown the extent by a MAX_SIZE extent, this would mean we | |
1685 | * need 2 outstanding extents, on one side we have 1 and the other side | |
1686 | * we have 1 so they are == and we can return. But in this case | |
1687 | * | |
1688 | * [MAX_SIZE+4k][MAX_SIZE+4k] | |
1689 | * | |
1690 | * Each range on their own accounts for 2 extents, but merged together | |
1691 | * they are only 3 extents worth of accounting, so we need to drop in | |
1692 | * this case. | |
dcab6a3b | 1693 | */ |
ba117213 | 1694 | old_size = other->end - other->start + 1; |
823bb20a | 1695 | num_extents = count_max_extents(old_size); |
ba117213 | 1696 | old_size = new->end - new->start + 1; |
823bb20a DS |
1697 | num_extents += count_max_extents(old_size); |
1698 | if (count_max_extents(new_size) >= num_extents) | |
dcab6a3b JB |
1699 | return; |
1700 | ||
9e0baf60 | 1701 | spin_lock(&BTRFS_I(inode)->lock); |
8b62f87b | 1702 | btrfs_mod_outstanding_extents(BTRFS_I(inode), -1); |
9e0baf60 | 1703 | spin_unlock(&BTRFS_I(inode)->lock); |
9ed74f2d JB |
1704 | } |
1705 | ||
eb73c1b7 MX |
1706 | static void btrfs_add_delalloc_inodes(struct btrfs_root *root, |
1707 | struct inode *inode) | |
1708 | { | |
0b246afa JM |
1709 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
1710 | ||
eb73c1b7 MX |
1711 | spin_lock(&root->delalloc_lock); |
1712 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1713 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1714 | &root->delalloc_inodes); | |
1715 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1716 | &BTRFS_I(inode)->runtime_flags); | |
1717 | root->nr_delalloc_inodes++; | |
1718 | if (root->nr_delalloc_inodes == 1) { | |
0b246afa | 1719 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
1720 | BUG_ON(!list_empty(&root->delalloc_root)); |
1721 | list_add_tail(&root->delalloc_root, | |
0b246afa JM |
1722 | &fs_info->delalloc_roots); |
1723 | spin_unlock(&fs_info->delalloc_root_lock); | |
eb73c1b7 MX |
1724 | } |
1725 | } | |
1726 | spin_unlock(&root->delalloc_lock); | |
1727 | } | |
1728 | ||
1729 | static void btrfs_del_delalloc_inode(struct btrfs_root *root, | |
9e3e97f4 | 1730 | struct btrfs_inode *inode) |
eb73c1b7 | 1731 | { |
9e3e97f4 | 1732 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
0b246afa | 1733 | |
eb73c1b7 | 1734 | spin_lock(&root->delalloc_lock); |
9e3e97f4 NB |
1735 | if (!list_empty(&inode->delalloc_inodes)) { |
1736 | list_del_init(&inode->delalloc_inodes); | |
eb73c1b7 | 1737 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 1738 | &inode->runtime_flags); |
eb73c1b7 MX |
1739 | root->nr_delalloc_inodes--; |
1740 | if (!root->nr_delalloc_inodes) { | |
0b246afa | 1741 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
1742 | BUG_ON(list_empty(&root->delalloc_root)); |
1743 | list_del_init(&root->delalloc_root); | |
0b246afa | 1744 | spin_unlock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
1745 | } |
1746 | } | |
1747 | spin_unlock(&root->delalloc_lock); | |
1748 | } | |
1749 | ||
d352ac68 CM |
1750 | /* |
1751 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1752 | * bytes in this file, and to maintain the list of inodes that | |
1753 | * have pending delalloc work to be done. | |
1754 | */ | |
c6100a4b | 1755 | static void btrfs_set_bit_hook(void *private_data, |
9ee49a04 | 1756 | struct extent_state *state, unsigned *bits) |
291d673e | 1757 | { |
c6100a4b | 1758 | struct inode *inode = private_data; |
9ed74f2d | 1759 | |
0b246afa JM |
1760 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
1761 | ||
47059d93 WS |
1762 | if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC)) |
1763 | WARN_ON(1); | |
75eff68e CM |
1764 | /* |
1765 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1766 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1767 | * bit, which is only set or cleared with irqs on |
1768 | */ | |
0ca1f7ce | 1769 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1770 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1771 | u64 len = state->end + 1 - state->start; |
8b62f87b | 1772 | u32 num_extents = count_max_extents(len); |
70ddc553 | 1773 | bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode)); |
9ed74f2d | 1774 | |
8b62f87b JB |
1775 | spin_lock(&BTRFS_I(inode)->lock); |
1776 | btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents); | |
1777 | spin_unlock(&BTRFS_I(inode)->lock); | |
287a0ab9 | 1778 | |
6a3891c5 | 1779 | /* For sanity tests */ |
0b246afa | 1780 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
1781 | return; |
1782 | ||
104b4e51 NB |
1783 | percpu_counter_add_batch(&fs_info->delalloc_bytes, len, |
1784 | fs_info->delalloc_batch); | |
df0af1a5 | 1785 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1786 | BTRFS_I(inode)->delalloc_bytes += len; |
47059d93 WS |
1787 | if (*bits & EXTENT_DEFRAG) |
1788 | BTRFS_I(inode)->defrag_bytes += len; | |
df0af1a5 | 1789 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1790 | &BTRFS_I(inode)->runtime_flags)) |
1791 | btrfs_add_delalloc_inodes(root, inode); | |
df0af1a5 | 1792 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1793 | } |
a7e3b975 FM |
1794 | |
1795 | if (!(state->state & EXTENT_DELALLOC_NEW) && | |
1796 | (*bits & EXTENT_DELALLOC_NEW)) { | |
1797 | spin_lock(&BTRFS_I(inode)->lock); | |
1798 | BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 - | |
1799 | state->start; | |
1800 | spin_unlock(&BTRFS_I(inode)->lock); | |
1801 | } | |
291d673e CM |
1802 | } |
1803 | ||
d352ac68 CM |
1804 | /* |
1805 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1806 | */ | |
c6100a4b | 1807 | static void btrfs_clear_bit_hook(void *private_data, |
41074888 | 1808 | struct extent_state *state, |
9ee49a04 | 1809 | unsigned *bits) |
291d673e | 1810 | { |
c6100a4b | 1811 | struct btrfs_inode *inode = BTRFS_I((struct inode *)private_data); |
6fc0ef68 | 1812 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
47059d93 | 1813 | u64 len = state->end + 1 - state->start; |
823bb20a | 1814 | u32 num_extents = count_max_extents(len); |
47059d93 | 1815 | |
4a4b964f FM |
1816 | if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) { |
1817 | spin_lock(&inode->lock); | |
6fc0ef68 | 1818 | inode->defrag_bytes -= len; |
4a4b964f FM |
1819 | spin_unlock(&inode->lock); |
1820 | } | |
47059d93 | 1821 | |
75eff68e CM |
1822 | /* |
1823 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1824 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1825 | * bit, which is only set or cleared with irqs on |
1826 | */ | |
0ca1f7ce | 1827 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
6fc0ef68 | 1828 | struct btrfs_root *root = inode->root; |
83eea1f1 | 1829 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 1830 | |
8b62f87b JB |
1831 | spin_lock(&inode->lock); |
1832 | btrfs_mod_outstanding_extents(inode, -num_extents); | |
1833 | spin_unlock(&inode->lock); | |
0ca1f7ce | 1834 | |
b6d08f06 JB |
1835 | /* |
1836 | * We don't reserve metadata space for space cache inodes so we | |
1837 | * don't need to call dellalloc_release_metadata if there is an | |
1838 | * error. | |
1839 | */ | |
a315e68f | 1840 | if (*bits & EXTENT_CLEAR_META_RESV && |
0b246afa | 1841 | root != fs_info->tree_root) |
0ca1f7ce YZ |
1842 | btrfs_delalloc_release_metadata(inode, len); |
1843 | ||
6a3891c5 | 1844 | /* For sanity tests. */ |
0b246afa | 1845 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
1846 | return; |
1847 | ||
a315e68f FM |
1848 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID && |
1849 | do_list && !(state->state & EXTENT_NORESERVE) && | |
1850 | (*bits & EXTENT_CLEAR_DATA_RESV)) | |
6fc0ef68 NB |
1851 | btrfs_free_reserved_data_space_noquota( |
1852 | &inode->vfs_inode, | |
51773bec | 1853 | state->start, len); |
9ed74f2d | 1854 | |
104b4e51 NB |
1855 | percpu_counter_add_batch(&fs_info->delalloc_bytes, -len, |
1856 | fs_info->delalloc_batch); | |
6fc0ef68 NB |
1857 | spin_lock(&inode->lock); |
1858 | inode->delalloc_bytes -= len; | |
1859 | if (do_list && inode->delalloc_bytes == 0 && | |
df0af1a5 | 1860 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 1861 | &inode->runtime_flags)) |
eb73c1b7 | 1862 | btrfs_del_delalloc_inode(root, inode); |
6fc0ef68 | 1863 | spin_unlock(&inode->lock); |
291d673e | 1864 | } |
a7e3b975 FM |
1865 | |
1866 | if ((state->state & EXTENT_DELALLOC_NEW) && | |
1867 | (*bits & EXTENT_DELALLOC_NEW)) { | |
1868 | spin_lock(&inode->lock); | |
1869 | ASSERT(inode->new_delalloc_bytes >= len); | |
1870 | inode->new_delalloc_bytes -= len; | |
1871 | spin_unlock(&inode->lock); | |
1872 | } | |
291d673e CM |
1873 | } |
1874 | ||
d352ac68 CM |
1875 | /* |
1876 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1877 | * we don't create bios that span stripes or chunks | |
6f034ece LB |
1878 | * |
1879 | * return 1 if page cannot be merged to bio | |
1880 | * return 0 if page can be merged to bio | |
1881 | * return error otherwise | |
d352ac68 | 1882 | */ |
81a75f67 | 1883 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1884 | size_t size, struct bio *bio, |
1885 | unsigned long bio_flags) | |
239b14b3 | 1886 | { |
0b246afa JM |
1887 | struct inode *inode = page->mapping->host; |
1888 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
4f024f37 | 1889 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
239b14b3 CM |
1890 | u64 length = 0; |
1891 | u64 map_length; | |
239b14b3 CM |
1892 | int ret; |
1893 | ||
771ed689 CM |
1894 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1895 | return 0; | |
1896 | ||
4f024f37 | 1897 | length = bio->bi_iter.bi_size; |
239b14b3 | 1898 | map_length = length; |
0b246afa JM |
1899 | ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length, |
1900 | NULL, 0); | |
6f034ece LB |
1901 | if (ret < 0) |
1902 | return ret; | |
d397712b | 1903 | if (map_length < length + size) |
239b14b3 | 1904 | return 1; |
3444a972 | 1905 | return 0; |
239b14b3 CM |
1906 | } |
1907 | ||
d352ac68 CM |
1908 | /* |
1909 | * in order to insert checksums into the metadata in large chunks, | |
1910 | * we wait until bio submission time. All the pages in the bio are | |
1911 | * checksummed and sums are attached onto the ordered extent record. | |
1912 | * | |
1913 | * At IO completion time the cums attached on the ordered extent record | |
1914 | * are inserted into the btree | |
1915 | */ | |
8c27cb35 | 1916 | static blk_status_t __btrfs_submit_bio_start(void *private_data, struct bio *bio, |
81a75f67 | 1917 | int mirror_num, unsigned long bio_flags, |
eaf25d93 | 1918 | u64 bio_offset) |
065631f6 | 1919 | { |
c6100a4b | 1920 | struct inode *inode = private_data; |
4e4cbee9 | 1921 | blk_status_t ret = 0; |
e015640f | 1922 | |
2ff7e61e | 1923 | ret = btrfs_csum_one_bio(inode, bio, 0, 0); |
79787eaa | 1924 | BUG_ON(ret); /* -ENOMEM */ |
4a69a410 CM |
1925 | return 0; |
1926 | } | |
e015640f | 1927 | |
4a69a410 CM |
1928 | /* |
1929 | * in order to insert checksums into the metadata in large chunks, | |
1930 | * we wait until bio submission time. All the pages in the bio are | |
1931 | * checksummed and sums are attached onto the ordered extent record. | |
1932 | * | |
1933 | * At IO completion time the cums attached on the ordered extent record | |
1934 | * are inserted into the btree | |
1935 | */ | |
8c27cb35 | 1936 | static blk_status_t __btrfs_submit_bio_done(void *private_data, struct bio *bio, |
eaf25d93 CM |
1937 | int mirror_num, unsigned long bio_flags, |
1938 | u64 bio_offset) | |
4a69a410 | 1939 | { |
c6100a4b | 1940 | struct inode *inode = private_data; |
2ff7e61e | 1941 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4e4cbee9 | 1942 | blk_status_t ret; |
61891923 | 1943 | |
2ff7e61e | 1944 | ret = btrfs_map_bio(fs_info, bio, mirror_num, 1); |
4246a0b6 | 1945 | if (ret) { |
4e4cbee9 | 1946 | bio->bi_status = ret; |
4246a0b6 CH |
1947 | bio_endio(bio); |
1948 | } | |
61891923 | 1949 | return ret; |
44b8bd7e CM |
1950 | } |
1951 | ||
d352ac68 | 1952 | /* |
cad321ad | 1953 | * extent_io.c submission hook. This does the right thing for csum calculation |
4c274bc6 LB |
1954 | * on write, or reading the csums from the tree before a read. |
1955 | * | |
1956 | * Rules about async/sync submit, | |
1957 | * a) read: sync submit | |
1958 | * | |
1959 | * b) write without checksum: sync submit | |
1960 | * | |
1961 | * c) write with checksum: | |
1962 | * c-1) if bio is issued by fsync: sync submit | |
1963 | * (sync_writers != 0) | |
1964 | * | |
1965 | * c-2) if root is reloc root: sync submit | |
1966 | * (only in case of buffered IO) | |
1967 | * | |
1968 | * c-3) otherwise: async submit | |
d352ac68 | 1969 | */ |
8c27cb35 | 1970 | static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio, |
c6100a4b JB |
1971 | int mirror_num, unsigned long bio_flags, |
1972 | u64 bio_offset) | |
44b8bd7e | 1973 | { |
c6100a4b | 1974 | struct inode *inode = private_data; |
0b246afa | 1975 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
44b8bd7e | 1976 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0d51e28a | 1977 | enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA; |
4e4cbee9 | 1978 | blk_status_t ret = 0; |
19b9bdb0 | 1979 | int skip_sum; |
b812ce28 | 1980 | int async = !atomic_read(&BTRFS_I(inode)->sync_writers); |
44b8bd7e | 1981 | |
6cbff00f | 1982 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1983 | |
70ddc553 | 1984 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) |
0d51e28a | 1985 | metadata = BTRFS_WQ_ENDIO_FREE_SPACE; |
0417341e | 1986 | |
37226b21 | 1987 | if (bio_op(bio) != REQ_OP_WRITE) { |
0b246afa | 1988 | ret = btrfs_bio_wq_end_io(fs_info, bio, metadata); |
5fd02043 | 1989 | if (ret) |
61891923 | 1990 | goto out; |
5fd02043 | 1991 | |
d20f7043 | 1992 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
61891923 SB |
1993 | ret = btrfs_submit_compressed_read(inode, bio, |
1994 | mirror_num, | |
1995 | bio_flags); | |
1996 | goto out; | |
c2db1073 | 1997 | } else if (!skip_sum) { |
2ff7e61e | 1998 | ret = btrfs_lookup_bio_sums(inode, bio, NULL); |
c2db1073 | 1999 | if (ret) |
61891923 | 2000 | goto out; |
c2db1073 | 2001 | } |
4d1b5fb4 | 2002 | goto mapit; |
b812ce28 | 2003 | } else if (async && !skip_sum) { |
17d217fe YZ |
2004 | /* csum items have already been cloned */ |
2005 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
2006 | goto mapit; | |
19b9bdb0 | 2007 | /* we're doing a write, do the async checksumming */ |
c6100a4b JB |
2008 | ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags, |
2009 | bio_offset, inode, | |
0b246afa JM |
2010 | __btrfs_submit_bio_start, |
2011 | __btrfs_submit_bio_done); | |
61891923 | 2012 | goto out; |
b812ce28 | 2013 | } else if (!skip_sum) { |
2ff7e61e | 2014 | ret = btrfs_csum_one_bio(inode, bio, 0, 0); |
b812ce28 JB |
2015 | if (ret) |
2016 | goto out; | |
19b9bdb0 CM |
2017 | } |
2018 | ||
0b86a832 | 2019 | mapit: |
2ff7e61e | 2020 | ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); |
61891923 SB |
2021 | |
2022 | out: | |
4e4cbee9 CH |
2023 | if (ret) { |
2024 | bio->bi_status = ret; | |
4246a0b6 CH |
2025 | bio_endio(bio); |
2026 | } | |
61891923 | 2027 | return ret; |
065631f6 | 2028 | } |
6885f308 | 2029 | |
d352ac68 CM |
2030 | /* |
2031 | * given a list of ordered sums record them in the inode. This happens | |
2032 | * at IO completion time based on sums calculated at bio submission time. | |
2033 | */ | |
ba1da2f4 | 2034 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
df9f628e | 2035 | struct inode *inode, struct list_head *list) |
e6dcd2dc | 2036 | { |
e6dcd2dc CM |
2037 | struct btrfs_ordered_sum *sum; |
2038 | ||
c6e30871 | 2039 | list_for_each_entry(sum, list, list) { |
39847c4d | 2040 | trans->adding_csums = 1; |
d20f7043 CM |
2041 | btrfs_csum_file_blocks(trans, |
2042 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
39847c4d | 2043 | trans->adding_csums = 0; |
e6dcd2dc CM |
2044 | } |
2045 | return 0; | |
2046 | } | |
2047 | ||
2ac55d41 | 2048 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
e3b8a485 | 2049 | unsigned int extra_bits, |
ba8b04c1 | 2050 | struct extent_state **cached_state, int dedupe) |
ea8c2819 | 2051 | { |
09cbfeaf | 2052 | WARN_ON((end & (PAGE_SIZE - 1)) == 0); |
ea8c2819 | 2053 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
e3b8a485 | 2054 | extra_bits, cached_state); |
ea8c2819 CM |
2055 | } |
2056 | ||
d352ac68 | 2057 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
2058 | struct btrfs_writepage_fixup { |
2059 | struct page *page; | |
2060 | struct btrfs_work work; | |
2061 | }; | |
2062 | ||
b2950863 | 2063 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
2064 | { |
2065 | struct btrfs_writepage_fixup *fixup; | |
2066 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 2067 | struct extent_state *cached_state = NULL; |
364ecf36 | 2068 | struct extent_changeset *data_reserved = NULL; |
247e743c CM |
2069 | struct page *page; |
2070 | struct inode *inode; | |
2071 | u64 page_start; | |
2072 | u64 page_end; | |
87826df0 | 2073 | int ret; |
247e743c CM |
2074 | |
2075 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
2076 | page = fixup->page; | |
4a096752 | 2077 | again: |
247e743c CM |
2078 | lock_page(page); |
2079 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
2080 | ClearPageChecked(page); | |
2081 | goto out_page; | |
2082 | } | |
2083 | ||
2084 | inode = page->mapping->host; | |
2085 | page_start = page_offset(page); | |
09cbfeaf | 2086 | page_end = page_offset(page) + PAGE_SIZE - 1; |
247e743c | 2087 | |
ff13db41 | 2088 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, |
d0082371 | 2089 | &cached_state); |
4a096752 CM |
2090 | |
2091 | /* already ordered? We're done */ | |
8b62b72b | 2092 | if (PagePrivate2(page)) |
247e743c | 2093 | goto out; |
4a096752 | 2094 | |
a776c6fa | 2095 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start, |
09cbfeaf | 2096 | PAGE_SIZE); |
4a096752 | 2097 | if (ordered) { |
2ac55d41 JB |
2098 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
2099 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
2100 | unlock_page(page); |
2101 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 2102 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
2103 | goto again; |
2104 | } | |
247e743c | 2105 | |
364ecf36 | 2106 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start, |
09cbfeaf | 2107 | PAGE_SIZE); |
87826df0 JM |
2108 | if (ret) { |
2109 | mapping_set_error(page->mapping, ret); | |
2110 | end_extent_writepage(page, ret, page_start, page_end); | |
2111 | ClearPageChecked(page); | |
2112 | goto out; | |
2113 | } | |
2114 | ||
e3b8a485 | 2115 | btrfs_set_extent_delalloc(inode, page_start, page_end, 0, &cached_state, |
ba8b04c1 | 2116 | 0); |
247e743c | 2117 | ClearPageChecked(page); |
87826df0 | 2118 | set_page_dirty(page); |
8b62f87b | 2119 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); |
247e743c | 2120 | out: |
2ac55d41 JB |
2121 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
2122 | &cached_state, GFP_NOFS); | |
247e743c CM |
2123 | out_page: |
2124 | unlock_page(page); | |
09cbfeaf | 2125 | put_page(page); |
b897abec | 2126 | kfree(fixup); |
364ecf36 | 2127 | extent_changeset_free(data_reserved); |
247e743c CM |
2128 | } |
2129 | ||
2130 | /* | |
2131 | * There are a few paths in the higher layers of the kernel that directly | |
2132 | * set the page dirty bit without asking the filesystem if it is a | |
2133 | * good idea. This causes problems because we want to make sure COW | |
2134 | * properly happens and the data=ordered rules are followed. | |
2135 | * | |
c8b97818 | 2136 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
2137 | * hasn't been properly setup for IO. We kick off an async process |
2138 | * to fix it up. The async helper will wait for ordered extents, set | |
2139 | * the delalloc bit and make it safe to write the page. | |
2140 | */ | |
b2950863 | 2141 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
2142 | { |
2143 | struct inode *inode = page->mapping->host; | |
0b246afa | 2144 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
247e743c | 2145 | struct btrfs_writepage_fixup *fixup; |
247e743c | 2146 | |
8b62b72b CM |
2147 | /* this page is properly in the ordered list */ |
2148 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
2149 | return 0; |
2150 | ||
2151 | if (PageChecked(page)) | |
2152 | return -EAGAIN; | |
2153 | ||
2154 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
2155 | if (!fixup) | |
2156 | return -EAGAIN; | |
f421950f | 2157 | |
247e743c | 2158 | SetPageChecked(page); |
09cbfeaf | 2159 | get_page(page); |
9e0af237 LB |
2160 | btrfs_init_work(&fixup->work, btrfs_fixup_helper, |
2161 | btrfs_writepage_fixup_worker, NULL, NULL); | |
247e743c | 2162 | fixup->page = page; |
0b246afa | 2163 | btrfs_queue_work(fs_info->fixup_workers, &fixup->work); |
87826df0 | 2164 | return -EBUSY; |
247e743c CM |
2165 | } |
2166 | ||
d899e052 YZ |
2167 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
2168 | struct inode *inode, u64 file_pos, | |
2169 | u64 disk_bytenr, u64 disk_num_bytes, | |
2170 | u64 num_bytes, u64 ram_bytes, | |
2171 | u8 compression, u8 encryption, | |
2172 | u16 other_encoding, int extent_type) | |
2173 | { | |
2174 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2175 | struct btrfs_file_extent_item *fi; | |
2176 | struct btrfs_path *path; | |
2177 | struct extent_buffer *leaf; | |
2178 | struct btrfs_key ins; | |
a12b877b | 2179 | u64 qg_released; |
1acae57b | 2180 | int extent_inserted = 0; |
d899e052 YZ |
2181 | int ret; |
2182 | ||
2183 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
2184 | if (!path) |
2185 | return -ENOMEM; | |
d899e052 | 2186 | |
a1ed835e CM |
2187 | /* |
2188 | * we may be replacing one extent in the tree with another. | |
2189 | * The new extent is pinned in the extent map, and we don't want | |
2190 | * to drop it from the cache until it is completely in the btree. | |
2191 | * | |
2192 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
2193 | * the caller is expected to unpin it and allow it to be merged | |
2194 | * with the others. | |
2195 | */ | |
1acae57b FDBM |
2196 | ret = __btrfs_drop_extents(trans, root, inode, path, file_pos, |
2197 | file_pos + num_bytes, NULL, 0, | |
2198 | 1, sizeof(*fi), &extent_inserted); | |
79787eaa JM |
2199 | if (ret) |
2200 | goto out; | |
d899e052 | 2201 | |
1acae57b | 2202 | if (!extent_inserted) { |
4a0cc7ca | 2203 | ins.objectid = btrfs_ino(BTRFS_I(inode)); |
1acae57b FDBM |
2204 | ins.offset = file_pos; |
2205 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
2206 | ||
2207 | path->leave_spinning = 1; | |
2208 | ret = btrfs_insert_empty_item(trans, root, path, &ins, | |
2209 | sizeof(*fi)); | |
2210 | if (ret) | |
2211 | goto out; | |
2212 | } | |
d899e052 YZ |
2213 | leaf = path->nodes[0]; |
2214 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2215 | struct btrfs_file_extent_item); | |
2216 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
2217 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
2218 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
2219 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
2220 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
2221 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2222 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
2223 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
2224 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
2225 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 | 2226 | |
d899e052 | 2227 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 2228 | btrfs_release_path(path); |
d899e052 YZ |
2229 | |
2230 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
2231 | |
2232 | ins.objectid = disk_bytenr; | |
2233 | ins.offset = disk_num_bytes; | |
2234 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
a12b877b | 2235 | |
297d750b | 2236 | /* |
5846a3c2 QW |
2237 | * Release the reserved range from inode dirty range map, as it is |
2238 | * already moved into delayed_ref_head | |
297d750b | 2239 | */ |
a12b877b QW |
2240 | ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes); |
2241 | if (ret < 0) | |
2242 | goto out; | |
2243 | qg_released = ret; | |
84f7d8e6 JB |
2244 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
2245 | btrfs_ino(BTRFS_I(inode)), | |
2246 | file_pos, qg_released, &ins); | |
79787eaa | 2247 | out: |
d899e052 | 2248 | btrfs_free_path(path); |
b9473439 | 2249 | |
79787eaa | 2250 | return ret; |
d899e052 YZ |
2251 | } |
2252 | ||
38c227d8 LB |
2253 | /* snapshot-aware defrag */ |
2254 | struct sa_defrag_extent_backref { | |
2255 | struct rb_node node; | |
2256 | struct old_sa_defrag_extent *old; | |
2257 | u64 root_id; | |
2258 | u64 inum; | |
2259 | u64 file_pos; | |
2260 | u64 extent_offset; | |
2261 | u64 num_bytes; | |
2262 | u64 generation; | |
2263 | }; | |
2264 | ||
2265 | struct old_sa_defrag_extent { | |
2266 | struct list_head list; | |
2267 | struct new_sa_defrag_extent *new; | |
2268 | ||
2269 | u64 extent_offset; | |
2270 | u64 bytenr; | |
2271 | u64 offset; | |
2272 | u64 len; | |
2273 | int count; | |
2274 | }; | |
2275 | ||
2276 | struct new_sa_defrag_extent { | |
2277 | struct rb_root root; | |
2278 | struct list_head head; | |
2279 | struct btrfs_path *path; | |
2280 | struct inode *inode; | |
2281 | u64 file_pos; | |
2282 | u64 len; | |
2283 | u64 bytenr; | |
2284 | u64 disk_len; | |
2285 | u8 compress_type; | |
2286 | }; | |
2287 | ||
2288 | static int backref_comp(struct sa_defrag_extent_backref *b1, | |
2289 | struct sa_defrag_extent_backref *b2) | |
2290 | { | |
2291 | if (b1->root_id < b2->root_id) | |
2292 | return -1; | |
2293 | else if (b1->root_id > b2->root_id) | |
2294 | return 1; | |
2295 | ||
2296 | if (b1->inum < b2->inum) | |
2297 | return -1; | |
2298 | else if (b1->inum > b2->inum) | |
2299 | return 1; | |
2300 | ||
2301 | if (b1->file_pos < b2->file_pos) | |
2302 | return -1; | |
2303 | else if (b1->file_pos > b2->file_pos) | |
2304 | return 1; | |
2305 | ||
2306 | /* | |
2307 | * [------------------------------] ===> (a range of space) | |
2308 | * |<--->| |<---->| =============> (fs/file tree A) | |
2309 | * |<---------------------------->| ===> (fs/file tree B) | |
2310 | * | |
2311 | * A range of space can refer to two file extents in one tree while | |
2312 | * refer to only one file extent in another tree. | |
2313 | * | |
2314 | * So we may process a disk offset more than one time(two extents in A) | |
2315 | * and locate at the same extent(one extent in B), then insert two same | |
2316 | * backrefs(both refer to the extent in B). | |
2317 | */ | |
2318 | return 0; | |
2319 | } | |
2320 | ||
2321 | static void backref_insert(struct rb_root *root, | |
2322 | struct sa_defrag_extent_backref *backref) | |
2323 | { | |
2324 | struct rb_node **p = &root->rb_node; | |
2325 | struct rb_node *parent = NULL; | |
2326 | struct sa_defrag_extent_backref *entry; | |
2327 | int ret; | |
2328 | ||
2329 | while (*p) { | |
2330 | parent = *p; | |
2331 | entry = rb_entry(parent, struct sa_defrag_extent_backref, node); | |
2332 | ||
2333 | ret = backref_comp(backref, entry); | |
2334 | if (ret < 0) | |
2335 | p = &(*p)->rb_left; | |
2336 | else | |
2337 | p = &(*p)->rb_right; | |
2338 | } | |
2339 | ||
2340 | rb_link_node(&backref->node, parent, p); | |
2341 | rb_insert_color(&backref->node, root); | |
2342 | } | |
2343 | ||
2344 | /* | |
2345 | * Note the backref might has changed, and in this case we just return 0. | |
2346 | */ | |
2347 | static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id, | |
2348 | void *ctx) | |
2349 | { | |
2350 | struct btrfs_file_extent_item *extent; | |
38c227d8 LB |
2351 | struct old_sa_defrag_extent *old = ctx; |
2352 | struct new_sa_defrag_extent *new = old->new; | |
2353 | struct btrfs_path *path = new->path; | |
2354 | struct btrfs_key key; | |
2355 | struct btrfs_root *root; | |
2356 | struct sa_defrag_extent_backref *backref; | |
2357 | struct extent_buffer *leaf; | |
2358 | struct inode *inode = new->inode; | |
0b246afa | 2359 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
38c227d8 LB |
2360 | int slot; |
2361 | int ret; | |
2362 | u64 extent_offset; | |
2363 | u64 num_bytes; | |
2364 | ||
2365 | if (BTRFS_I(inode)->root->root_key.objectid == root_id && | |
4a0cc7ca | 2366 | inum == btrfs_ino(BTRFS_I(inode))) |
38c227d8 LB |
2367 | return 0; |
2368 | ||
2369 | key.objectid = root_id; | |
2370 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2371 | key.offset = (u64)-1; | |
2372 | ||
38c227d8 LB |
2373 | root = btrfs_read_fs_root_no_name(fs_info, &key); |
2374 | if (IS_ERR(root)) { | |
2375 | if (PTR_ERR(root) == -ENOENT) | |
2376 | return 0; | |
2377 | WARN_ON(1); | |
ab8d0fc4 | 2378 | btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu", |
38c227d8 LB |
2379 | inum, offset, root_id); |
2380 | return PTR_ERR(root); | |
2381 | } | |
2382 | ||
2383 | key.objectid = inum; | |
2384 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2385 | if (offset > (u64)-1 << 32) | |
2386 | key.offset = 0; | |
2387 | else | |
2388 | key.offset = offset; | |
2389 | ||
2390 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
fae7f21c | 2391 | if (WARN_ON(ret < 0)) |
38c227d8 | 2392 | return ret; |
50f1319c | 2393 | ret = 0; |
38c227d8 LB |
2394 | |
2395 | while (1) { | |
2396 | cond_resched(); | |
2397 | ||
2398 | leaf = path->nodes[0]; | |
2399 | slot = path->slots[0]; | |
2400 | ||
2401 | if (slot >= btrfs_header_nritems(leaf)) { | |
2402 | ret = btrfs_next_leaf(root, path); | |
2403 | if (ret < 0) { | |
2404 | goto out; | |
2405 | } else if (ret > 0) { | |
2406 | ret = 0; | |
2407 | goto out; | |
2408 | } | |
2409 | continue; | |
2410 | } | |
2411 | ||
2412 | path->slots[0]++; | |
2413 | ||
2414 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
2415 | ||
2416 | if (key.objectid > inum) | |
2417 | goto out; | |
2418 | ||
2419 | if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY) | |
2420 | continue; | |
2421 | ||
2422 | extent = btrfs_item_ptr(leaf, slot, | |
2423 | struct btrfs_file_extent_item); | |
2424 | ||
2425 | if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr) | |
2426 | continue; | |
2427 | ||
e68afa49 LB |
2428 | /* |
2429 | * 'offset' refers to the exact key.offset, | |
2430 | * NOT the 'offset' field in btrfs_extent_data_ref, ie. | |
2431 | * (key.offset - extent_offset). | |
2432 | */ | |
2433 | if (key.offset != offset) | |
38c227d8 LB |
2434 | continue; |
2435 | ||
e68afa49 | 2436 | extent_offset = btrfs_file_extent_offset(leaf, extent); |
38c227d8 | 2437 | num_bytes = btrfs_file_extent_num_bytes(leaf, extent); |
e68afa49 | 2438 | |
38c227d8 LB |
2439 | if (extent_offset >= old->extent_offset + old->offset + |
2440 | old->len || extent_offset + num_bytes <= | |
2441 | old->extent_offset + old->offset) | |
2442 | continue; | |
38c227d8 LB |
2443 | break; |
2444 | } | |
2445 | ||
2446 | backref = kmalloc(sizeof(*backref), GFP_NOFS); | |
2447 | if (!backref) { | |
2448 | ret = -ENOENT; | |
2449 | goto out; | |
2450 | } | |
2451 | ||
2452 | backref->root_id = root_id; | |
2453 | backref->inum = inum; | |
e68afa49 | 2454 | backref->file_pos = offset; |
38c227d8 LB |
2455 | backref->num_bytes = num_bytes; |
2456 | backref->extent_offset = extent_offset; | |
2457 | backref->generation = btrfs_file_extent_generation(leaf, extent); | |
2458 | backref->old = old; | |
2459 | backref_insert(&new->root, backref); | |
2460 | old->count++; | |
2461 | out: | |
2462 | btrfs_release_path(path); | |
2463 | WARN_ON(ret); | |
2464 | return ret; | |
2465 | } | |
2466 | ||
2467 | static noinline bool record_extent_backrefs(struct btrfs_path *path, | |
2468 | struct new_sa_defrag_extent *new) | |
2469 | { | |
0b246afa | 2470 | struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb); |
38c227d8 LB |
2471 | struct old_sa_defrag_extent *old, *tmp; |
2472 | int ret; | |
2473 | ||
2474 | new->path = path; | |
2475 | ||
2476 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
e68afa49 LB |
2477 | ret = iterate_inodes_from_logical(old->bytenr + |
2478 | old->extent_offset, fs_info, | |
38c227d8 | 2479 | path, record_one_backref, |
c995ab3c | 2480 | old, false); |
4724b106 JB |
2481 | if (ret < 0 && ret != -ENOENT) |
2482 | return false; | |
38c227d8 LB |
2483 | |
2484 | /* no backref to be processed for this extent */ | |
2485 | if (!old->count) { | |
2486 | list_del(&old->list); | |
2487 | kfree(old); | |
2488 | } | |
2489 | } | |
2490 | ||
2491 | if (list_empty(&new->head)) | |
2492 | return false; | |
2493 | ||
2494 | return true; | |
2495 | } | |
2496 | ||
2497 | static int relink_is_mergable(struct extent_buffer *leaf, | |
2498 | struct btrfs_file_extent_item *fi, | |
116e0024 | 2499 | struct new_sa_defrag_extent *new) |
38c227d8 | 2500 | { |
116e0024 | 2501 | if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr) |
38c227d8 LB |
2502 | return 0; |
2503 | ||
2504 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2505 | return 0; | |
2506 | ||
116e0024 LB |
2507 | if (btrfs_file_extent_compression(leaf, fi) != new->compress_type) |
2508 | return 0; | |
2509 | ||
2510 | if (btrfs_file_extent_encryption(leaf, fi) || | |
38c227d8 LB |
2511 | btrfs_file_extent_other_encoding(leaf, fi)) |
2512 | return 0; | |
2513 | ||
2514 | return 1; | |
2515 | } | |
2516 | ||
2517 | /* | |
2518 | * Note the backref might has changed, and in this case we just return 0. | |
2519 | */ | |
2520 | static noinline int relink_extent_backref(struct btrfs_path *path, | |
2521 | struct sa_defrag_extent_backref *prev, | |
2522 | struct sa_defrag_extent_backref *backref) | |
2523 | { | |
2524 | struct btrfs_file_extent_item *extent; | |
2525 | struct btrfs_file_extent_item *item; | |
2526 | struct btrfs_ordered_extent *ordered; | |
2527 | struct btrfs_trans_handle *trans; | |
38c227d8 LB |
2528 | struct btrfs_root *root; |
2529 | struct btrfs_key key; | |
2530 | struct extent_buffer *leaf; | |
2531 | struct old_sa_defrag_extent *old = backref->old; | |
2532 | struct new_sa_defrag_extent *new = old->new; | |
0b246afa | 2533 | struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb); |
38c227d8 LB |
2534 | struct inode *inode; |
2535 | struct extent_state *cached = NULL; | |
2536 | int ret = 0; | |
2537 | u64 start; | |
2538 | u64 len; | |
2539 | u64 lock_start; | |
2540 | u64 lock_end; | |
2541 | bool merge = false; | |
2542 | int index; | |
2543 | ||
2544 | if (prev && prev->root_id == backref->root_id && | |
2545 | prev->inum == backref->inum && | |
2546 | prev->file_pos + prev->num_bytes == backref->file_pos) | |
2547 | merge = true; | |
2548 | ||
2549 | /* step 1: get root */ | |
2550 | key.objectid = backref->root_id; | |
2551 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2552 | key.offset = (u64)-1; | |
2553 | ||
38c227d8 LB |
2554 | index = srcu_read_lock(&fs_info->subvol_srcu); |
2555 | ||
2556 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2557 | if (IS_ERR(root)) { | |
2558 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2559 | if (PTR_ERR(root) == -ENOENT) | |
2560 | return 0; | |
2561 | return PTR_ERR(root); | |
2562 | } | |
38c227d8 | 2563 | |
bcbba5e6 WS |
2564 | if (btrfs_root_readonly(root)) { |
2565 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2566 | return 0; | |
2567 | } | |
2568 | ||
38c227d8 LB |
2569 | /* step 2: get inode */ |
2570 | key.objectid = backref->inum; | |
2571 | key.type = BTRFS_INODE_ITEM_KEY; | |
2572 | key.offset = 0; | |
2573 | ||
2574 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
2575 | if (IS_ERR(inode)) { | |
2576 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2577 | return 0; | |
2578 | } | |
2579 | ||
2580 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2581 | ||
2582 | /* step 3: relink backref */ | |
2583 | lock_start = backref->file_pos; | |
2584 | lock_end = backref->file_pos + backref->num_bytes - 1; | |
2585 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
ff13db41 | 2586 | &cached); |
38c227d8 LB |
2587 | |
2588 | ordered = btrfs_lookup_first_ordered_extent(inode, lock_end); | |
2589 | if (ordered) { | |
2590 | btrfs_put_ordered_extent(ordered); | |
2591 | goto out_unlock; | |
2592 | } | |
2593 | ||
2594 | trans = btrfs_join_transaction(root); | |
2595 | if (IS_ERR(trans)) { | |
2596 | ret = PTR_ERR(trans); | |
2597 | goto out_unlock; | |
2598 | } | |
2599 | ||
2600 | key.objectid = backref->inum; | |
2601 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2602 | key.offset = backref->file_pos; | |
2603 | ||
2604 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2605 | if (ret < 0) { | |
2606 | goto out_free_path; | |
2607 | } else if (ret > 0) { | |
2608 | ret = 0; | |
2609 | goto out_free_path; | |
2610 | } | |
2611 | ||
2612 | extent = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2613 | struct btrfs_file_extent_item); | |
2614 | ||
2615 | if (btrfs_file_extent_generation(path->nodes[0], extent) != | |
2616 | backref->generation) | |
2617 | goto out_free_path; | |
2618 | ||
2619 | btrfs_release_path(path); | |
2620 | ||
2621 | start = backref->file_pos; | |
2622 | if (backref->extent_offset < old->extent_offset + old->offset) | |
2623 | start += old->extent_offset + old->offset - | |
2624 | backref->extent_offset; | |
2625 | ||
2626 | len = min(backref->extent_offset + backref->num_bytes, | |
2627 | old->extent_offset + old->offset + old->len); | |
2628 | len -= max(backref->extent_offset, old->extent_offset + old->offset); | |
2629 | ||
2630 | ret = btrfs_drop_extents(trans, root, inode, start, | |
2631 | start + len, 1); | |
2632 | if (ret) | |
2633 | goto out_free_path; | |
2634 | again: | |
4a0cc7ca | 2635 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
38c227d8 LB |
2636 | key.type = BTRFS_EXTENT_DATA_KEY; |
2637 | key.offset = start; | |
2638 | ||
a09a0a70 | 2639 | path->leave_spinning = 1; |
38c227d8 LB |
2640 | if (merge) { |
2641 | struct btrfs_file_extent_item *fi; | |
2642 | u64 extent_len; | |
2643 | struct btrfs_key found_key; | |
2644 | ||
3c9665df | 2645 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
38c227d8 LB |
2646 | if (ret < 0) |
2647 | goto out_free_path; | |
2648 | ||
2649 | path->slots[0]--; | |
2650 | leaf = path->nodes[0]; | |
2651 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2652 | ||
2653 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2654 | struct btrfs_file_extent_item); | |
2655 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
2656 | ||
116e0024 LB |
2657 | if (extent_len + found_key.offset == start && |
2658 | relink_is_mergable(leaf, fi, new)) { | |
38c227d8 LB |
2659 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2660 | extent_len + len); | |
2661 | btrfs_mark_buffer_dirty(leaf); | |
2662 | inode_add_bytes(inode, len); | |
2663 | ||
2664 | ret = 1; | |
2665 | goto out_free_path; | |
2666 | } else { | |
2667 | merge = false; | |
2668 | btrfs_release_path(path); | |
2669 | goto again; | |
2670 | } | |
2671 | } | |
2672 | ||
2673 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2674 | sizeof(*extent)); | |
2675 | if (ret) { | |
66642832 | 2676 | btrfs_abort_transaction(trans, ret); |
38c227d8 LB |
2677 | goto out_free_path; |
2678 | } | |
2679 | ||
2680 | leaf = path->nodes[0]; | |
2681 | item = btrfs_item_ptr(leaf, path->slots[0], | |
2682 | struct btrfs_file_extent_item); | |
2683 | btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr); | |
2684 | btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len); | |
2685 | btrfs_set_file_extent_offset(leaf, item, start - new->file_pos); | |
2686 | btrfs_set_file_extent_num_bytes(leaf, item, len); | |
2687 | btrfs_set_file_extent_ram_bytes(leaf, item, new->len); | |
2688 | btrfs_set_file_extent_generation(leaf, item, trans->transid); | |
2689 | btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG); | |
2690 | btrfs_set_file_extent_compression(leaf, item, new->compress_type); | |
2691 | btrfs_set_file_extent_encryption(leaf, item, 0); | |
2692 | btrfs_set_file_extent_other_encoding(leaf, item, 0); | |
2693 | ||
2694 | btrfs_mark_buffer_dirty(leaf); | |
2695 | inode_add_bytes(inode, len); | |
a09a0a70 | 2696 | btrfs_release_path(path); |
38c227d8 | 2697 | |
84f7d8e6 | 2698 | ret = btrfs_inc_extent_ref(trans, root, new->bytenr, |
38c227d8 LB |
2699 | new->disk_len, 0, |
2700 | backref->root_id, backref->inum, | |
b06c4bf5 | 2701 | new->file_pos); /* start - extent_offset */ |
38c227d8 | 2702 | if (ret) { |
66642832 | 2703 | btrfs_abort_transaction(trans, ret); |
38c227d8 LB |
2704 | goto out_free_path; |
2705 | } | |
2706 | ||
2707 | ret = 1; | |
2708 | out_free_path: | |
2709 | btrfs_release_path(path); | |
a09a0a70 | 2710 | path->leave_spinning = 0; |
3a45bb20 | 2711 | btrfs_end_transaction(trans); |
38c227d8 LB |
2712 | out_unlock: |
2713 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
2714 | &cached, GFP_NOFS); | |
2715 | iput(inode); | |
2716 | return ret; | |
2717 | } | |
2718 | ||
6f519564 LB |
2719 | static void free_sa_defrag_extent(struct new_sa_defrag_extent *new) |
2720 | { | |
2721 | struct old_sa_defrag_extent *old, *tmp; | |
2722 | ||
2723 | if (!new) | |
2724 | return; | |
2725 | ||
2726 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
6f519564 LB |
2727 | kfree(old); |
2728 | } | |
2729 | kfree(new); | |
2730 | } | |
2731 | ||
38c227d8 LB |
2732 | static void relink_file_extents(struct new_sa_defrag_extent *new) |
2733 | { | |
0b246afa | 2734 | struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb); |
38c227d8 | 2735 | struct btrfs_path *path; |
38c227d8 LB |
2736 | struct sa_defrag_extent_backref *backref; |
2737 | struct sa_defrag_extent_backref *prev = NULL; | |
2738 | struct inode *inode; | |
2739 | struct btrfs_root *root; | |
2740 | struct rb_node *node; | |
2741 | int ret; | |
2742 | ||
2743 | inode = new->inode; | |
2744 | root = BTRFS_I(inode)->root; | |
2745 | ||
2746 | path = btrfs_alloc_path(); | |
2747 | if (!path) | |
2748 | return; | |
2749 | ||
2750 | if (!record_extent_backrefs(path, new)) { | |
2751 | btrfs_free_path(path); | |
2752 | goto out; | |
2753 | } | |
2754 | btrfs_release_path(path); | |
2755 | ||
2756 | while (1) { | |
2757 | node = rb_first(&new->root); | |
2758 | if (!node) | |
2759 | break; | |
2760 | rb_erase(node, &new->root); | |
2761 | ||
2762 | backref = rb_entry(node, struct sa_defrag_extent_backref, node); | |
2763 | ||
2764 | ret = relink_extent_backref(path, prev, backref); | |
2765 | WARN_ON(ret < 0); | |
2766 | ||
2767 | kfree(prev); | |
2768 | ||
2769 | if (ret == 1) | |
2770 | prev = backref; | |
2771 | else | |
2772 | prev = NULL; | |
2773 | cond_resched(); | |
2774 | } | |
2775 | kfree(prev); | |
2776 | ||
2777 | btrfs_free_path(path); | |
38c227d8 | 2778 | out: |
6f519564 LB |
2779 | free_sa_defrag_extent(new); |
2780 | ||
0b246afa JM |
2781 | atomic_dec(&fs_info->defrag_running); |
2782 | wake_up(&fs_info->transaction_wait); | |
38c227d8 LB |
2783 | } |
2784 | ||
2785 | static struct new_sa_defrag_extent * | |
2786 | record_old_file_extents(struct inode *inode, | |
2787 | struct btrfs_ordered_extent *ordered) | |
2788 | { | |
0b246afa | 2789 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
38c227d8 LB |
2790 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2791 | struct btrfs_path *path; | |
2792 | struct btrfs_key key; | |
6f519564 | 2793 | struct old_sa_defrag_extent *old; |
38c227d8 LB |
2794 | struct new_sa_defrag_extent *new; |
2795 | int ret; | |
2796 | ||
2797 | new = kmalloc(sizeof(*new), GFP_NOFS); | |
2798 | if (!new) | |
2799 | return NULL; | |
2800 | ||
2801 | new->inode = inode; | |
2802 | new->file_pos = ordered->file_offset; | |
2803 | new->len = ordered->len; | |
2804 | new->bytenr = ordered->start; | |
2805 | new->disk_len = ordered->disk_len; | |
2806 | new->compress_type = ordered->compress_type; | |
2807 | new->root = RB_ROOT; | |
2808 | INIT_LIST_HEAD(&new->head); | |
2809 | ||
2810 | path = btrfs_alloc_path(); | |
2811 | if (!path) | |
2812 | goto out_kfree; | |
2813 | ||
4a0cc7ca | 2814 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
38c227d8 LB |
2815 | key.type = BTRFS_EXTENT_DATA_KEY; |
2816 | key.offset = new->file_pos; | |
2817 | ||
2818 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2819 | if (ret < 0) | |
2820 | goto out_free_path; | |
2821 | if (ret > 0 && path->slots[0] > 0) | |
2822 | path->slots[0]--; | |
2823 | ||
2824 | /* find out all the old extents for the file range */ | |
2825 | while (1) { | |
2826 | struct btrfs_file_extent_item *extent; | |
2827 | struct extent_buffer *l; | |
2828 | int slot; | |
2829 | u64 num_bytes; | |
2830 | u64 offset; | |
2831 | u64 end; | |
2832 | u64 disk_bytenr; | |
2833 | u64 extent_offset; | |
2834 | ||
2835 | l = path->nodes[0]; | |
2836 | slot = path->slots[0]; | |
2837 | ||
2838 | if (slot >= btrfs_header_nritems(l)) { | |
2839 | ret = btrfs_next_leaf(root, path); | |
2840 | if (ret < 0) | |
6f519564 | 2841 | goto out_free_path; |
38c227d8 LB |
2842 | else if (ret > 0) |
2843 | break; | |
2844 | continue; | |
2845 | } | |
2846 | ||
2847 | btrfs_item_key_to_cpu(l, &key, slot); | |
2848 | ||
4a0cc7ca | 2849 | if (key.objectid != btrfs_ino(BTRFS_I(inode))) |
38c227d8 LB |
2850 | break; |
2851 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2852 | break; | |
2853 | if (key.offset >= new->file_pos + new->len) | |
2854 | break; | |
2855 | ||
2856 | extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item); | |
2857 | ||
2858 | num_bytes = btrfs_file_extent_num_bytes(l, extent); | |
2859 | if (key.offset + num_bytes < new->file_pos) | |
2860 | goto next; | |
2861 | ||
2862 | disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent); | |
2863 | if (!disk_bytenr) | |
2864 | goto next; | |
2865 | ||
2866 | extent_offset = btrfs_file_extent_offset(l, extent); | |
2867 | ||
2868 | old = kmalloc(sizeof(*old), GFP_NOFS); | |
2869 | if (!old) | |
6f519564 | 2870 | goto out_free_path; |
38c227d8 LB |
2871 | |
2872 | offset = max(new->file_pos, key.offset); | |
2873 | end = min(new->file_pos + new->len, key.offset + num_bytes); | |
2874 | ||
2875 | old->bytenr = disk_bytenr; | |
2876 | old->extent_offset = extent_offset; | |
2877 | old->offset = offset - key.offset; | |
2878 | old->len = end - offset; | |
2879 | old->new = new; | |
2880 | old->count = 0; | |
2881 | list_add_tail(&old->list, &new->head); | |
2882 | next: | |
2883 | path->slots[0]++; | |
2884 | cond_resched(); | |
2885 | } | |
2886 | ||
2887 | btrfs_free_path(path); | |
0b246afa | 2888 | atomic_inc(&fs_info->defrag_running); |
38c227d8 LB |
2889 | |
2890 | return new; | |
2891 | ||
38c227d8 LB |
2892 | out_free_path: |
2893 | btrfs_free_path(path); | |
2894 | out_kfree: | |
6f519564 | 2895 | free_sa_defrag_extent(new); |
38c227d8 LB |
2896 | return NULL; |
2897 | } | |
2898 | ||
2ff7e61e | 2899 | static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
2900 | u64 start, u64 len) |
2901 | { | |
2902 | struct btrfs_block_group_cache *cache; | |
2903 | ||
0b246afa | 2904 | cache = btrfs_lookup_block_group(fs_info, start); |
e570fd27 MX |
2905 | ASSERT(cache); |
2906 | ||
2907 | spin_lock(&cache->lock); | |
2908 | cache->delalloc_bytes -= len; | |
2909 | spin_unlock(&cache->lock); | |
2910 | ||
2911 | btrfs_put_block_group(cache); | |
2912 | } | |
2913 | ||
d352ac68 CM |
2914 | /* as ordered data IO finishes, this gets called so we can finish |
2915 | * an ordered extent if the range of bytes in the file it covers are | |
2916 | * fully written. | |
2917 | */ | |
5fd02043 | 2918 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 2919 | { |
5fd02043 | 2920 | struct inode *inode = ordered_extent->inode; |
0b246afa | 2921 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc | 2922 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 2923 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 2924 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 2925 | struct extent_state *cached_state = NULL; |
38c227d8 | 2926 | struct new_sa_defrag_extent *new = NULL; |
261507a0 | 2927 | int compress_type = 0; |
77cef2ec JB |
2928 | int ret = 0; |
2929 | u64 logical_len = ordered_extent->len; | |
82d5902d | 2930 | bool nolock; |
77cef2ec | 2931 | bool truncated = false; |
a7e3b975 FM |
2932 | bool range_locked = false; |
2933 | bool clear_new_delalloc_bytes = false; | |
2934 | ||
2935 | if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && | |
2936 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) && | |
2937 | !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags)) | |
2938 | clear_new_delalloc_bytes = true; | |
e6dcd2dc | 2939 | |
70ddc553 | 2940 | nolock = btrfs_is_free_space_inode(BTRFS_I(inode)); |
0cb59c99 | 2941 | |
5fd02043 JB |
2942 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
2943 | ret = -EIO; | |
2944 | goto out; | |
2945 | } | |
2946 | ||
7ab7956e NB |
2947 | btrfs_free_io_failure_record(BTRFS_I(inode), |
2948 | ordered_extent->file_offset, | |
2949 | ordered_extent->file_offset + | |
2950 | ordered_extent->len - 1); | |
f612496b | 2951 | |
77cef2ec JB |
2952 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) { |
2953 | truncated = true; | |
2954 | logical_len = ordered_extent->truncated_len; | |
2955 | /* Truncated the entire extent, don't bother adding */ | |
2956 | if (!logical_len) | |
2957 | goto out; | |
2958 | } | |
2959 | ||
c2167754 | 2960 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 2961 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
94ed938a QW |
2962 | |
2963 | /* | |
2964 | * For mwrite(mmap + memset to write) case, we still reserve | |
2965 | * space for NOCOW range. | |
2966 | * As NOCOW won't cause a new delayed ref, just free the space | |
2967 | */ | |
bc42bda2 | 2968 | btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset, |
94ed938a | 2969 | ordered_extent->len); |
6c760c07 JB |
2970 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2971 | if (nolock) | |
2972 | trans = btrfs_join_transaction_nolock(root); | |
2973 | else | |
2974 | trans = btrfs_join_transaction(root); | |
2975 | if (IS_ERR(trans)) { | |
2976 | ret = PTR_ERR(trans); | |
2977 | trans = NULL; | |
2978 | goto out; | |
c2167754 | 2979 | } |
69fe2d75 | 2980 | trans->block_rsv = &BTRFS_I(inode)->block_rsv; |
6c760c07 JB |
2981 | ret = btrfs_update_inode_fallback(trans, root, inode); |
2982 | if (ret) /* -ENOMEM or corruption */ | |
66642832 | 2983 | btrfs_abort_transaction(trans, ret); |
c2167754 YZ |
2984 | goto out; |
2985 | } | |
e6dcd2dc | 2986 | |
a7e3b975 | 2987 | range_locked = true; |
2ac55d41 JB |
2988 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
2989 | ordered_extent->file_offset + ordered_extent->len - 1, | |
ff13db41 | 2990 | &cached_state); |
e6dcd2dc | 2991 | |
38c227d8 LB |
2992 | ret = test_range_bit(io_tree, ordered_extent->file_offset, |
2993 | ordered_extent->file_offset + ordered_extent->len - 1, | |
452e62b7 | 2994 | EXTENT_DEFRAG, 0, cached_state); |
38c227d8 LB |
2995 | if (ret) { |
2996 | u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item); | |
8101c8db | 2997 | if (0 && last_snapshot >= BTRFS_I(inode)->generation) |
38c227d8 LB |
2998 | /* the inode is shared */ |
2999 | new = record_old_file_extents(inode, ordered_extent); | |
3000 | ||
3001 | clear_extent_bit(io_tree, ordered_extent->file_offset, | |
3002 | ordered_extent->file_offset + ordered_extent->len - 1, | |
ae0f1625 | 3003 | EXTENT_DEFRAG, 0, 0, &cached_state); |
38c227d8 LB |
3004 | } |
3005 | ||
0cb59c99 | 3006 | if (nolock) |
7a7eaa40 | 3007 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 3008 | else |
7a7eaa40 | 3009 | trans = btrfs_join_transaction(root); |
79787eaa JM |
3010 | if (IS_ERR(trans)) { |
3011 | ret = PTR_ERR(trans); | |
3012 | trans = NULL; | |
a7e3b975 | 3013 | goto out; |
79787eaa | 3014 | } |
a79b7d4b | 3015 | |
69fe2d75 | 3016 | trans->block_rsv = &BTRFS_I(inode)->block_rsv; |
c2167754 | 3017 | |
c8b97818 | 3018 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 3019 | compress_type = ordered_extent->compress_type; |
d899e052 | 3020 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 3021 | BUG_ON(compress_type); |
b430b775 JM |
3022 | btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset, |
3023 | ordered_extent->len); | |
7a6d7067 | 3024 | ret = btrfs_mark_extent_written(trans, BTRFS_I(inode), |
d899e052 YZ |
3025 | ordered_extent->file_offset, |
3026 | ordered_extent->file_offset + | |
77cef2ec | 3027 | logical_len); |
d899e052 | 3028 | } else { |
0b246afa | 3029 | BUG_ON(root == fs_info->tree_root); |
d899e052 YZ |
3030 | ret = insert_reserved_file_extent(trans, inode, |
3031 | ordered_extent->file_offset, | |
3032 | ordered_extent->start, | |
3033 | ordered_extent->disk_len, | |
77cef2ec | 3034 | logical_len, logical_len, |
261507a0 | 3035 | compress_type, 0, 0, |
d899e052 | 3036 | BTRFS_FILE_EXTENT_REG); |
e570fd27 | 3037 | if (!ret) |
2ff7e61e | 3038 | btrfs_release_delalloc_bytes(fs_info, |
e570fd27 MX |
3039 | ordered_extent->start, |
3040 | ordered_extent->disk_len); | |
d899e052 | 3041 | } |
5dc562c5 JB |
3042 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
3043 | ordered_extent->file_offset, ordered_extent->len, | |
3044 | trans->transid); | |
79787eaa | 3045 | if (ret < 0) { |
66642832 | 3046 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 3047 | goto out; |
79787eaa | 3048 | } |
2ac55d41 | 3049 | |
df9f628e | 3050 | add_pending_csums(trans, inode, &ordered_extent->list); |
e6dcd2dc | 3051 | |
6c760c07 JB |
3052 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
3053 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
3054 | if (ret) { /* -ENOMEM or corruption */ | |
66642832 | 3055 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 3056 | goto out; |
1ef30be1 JB |
3057 | } |
3058 | ret = 0; | |
c2167754 | 3059 | out: |
a7e3b975 FM |
3060 | if (range_locked || clear_new_delalloc_bytes) { |
3061 | unsigned int clear_bits = 0; | |
3062 | ||
3063 | if (range_locked) | |
3064 | clear_bits |= EXTENT_LOCKED; | |
3065 | if (clear_new_delalloc_bytes) | |
3066 | clear_bits |= EXTENT_DELALLOC_NEW; | |
3067 | clear_extent_bit(&BTRFS_I(inode)->io_tree, | |
3068 | ordered_extent->file_offset, | |
3069 | ordered_extent->file_offset + | |
3070 | ordered_extent->len - 1, | |
3071 | clear_bits, | |
3072 | (clear_bits & EXTENT_LOCKED) ? 1 : 0, | |
ae0f1625 | 3073 | 0, &cached_state); |
a7e3b975 FM |
3074 | } |
3075 | ||
a698d075 | 3076 | if (trans) |
3a45bb20 | 3077 | btrfs_end_transaction(trans); |
0cb59c99 | 3078 | |
77cef2ec JB |
3079 | if (ret || truncated) { |
3080 | u64 start, end; | |
3081 | ||
3082 | if (truncated) | |
3083 | start = ordered_extent->file_offset + logical_len; | |
3084 | else | |
3085 | start = ordered_extent->file_offset; | |
3086 | end = ordered_extent->file_offset + ordered_extent->len - 1; | |
3087 | clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS); | |
3088 | ||
3089 | /* Drop the cache for the part of the extent we didn't write. */ | |
dcdbc059 | 3090 | btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0); |
5fd02043 | 3091 | |
0bec9ef5 JB |
3092 | /* |
3093 | * If the ordered extent had an IOERR or something else went | |
3094 | * wrong we need to return the space for this ordered extent | |
77cef2ec JB |
3095 | * back to the allocator. We only free the extent in the |
3096 | * truncated case if we didn't write out the extent at all. | |
0bec9ef5 | 3097 | */ |
77cef2ec JB |
3098 | if ((ret || !logical_len) && |
3099 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && | |
0bec9ef5 | 3100 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) |
2ff7e61e JM |
3101 | btrfs_free_reserved_extent(fs_info, |
3102 | ordered_extent->start, | |
e570fd27 | 3103 | ordered_extent->disk_len, 1); |
0bec9ef5 JB |
3104 | } |
3105 | ||
3106 | ||
5fd02043 | 3107 | /* |
8bad3c02 LB |
3108 | * This needs to be done to make sure anybody waiting knows we are done |
3109 | * updating everything for this ordered extent. | |
5fd02043 JB |
3110 | */ |
3111 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
3112 | ||
38c227d8 | 3113 | /* for snapshot-aware defrag */ |
6f519564 LB |
3114 | if (new) { |
3115 | if (ret) { | |
3116 | free_sa_defrag_extent(new); | |
0b246afa | 3117 | atomic_dec(&fs_info->defrag_running); |
6f519564 LB |
3118 | } else { |
3119 | relink_file_extents(new); | |
3120 | } | |
3121 | } | |
38c227d8 | 3122 | |
e6dcd2dc CM |
3123 | /* once for us */ |
3124 | btrfs_put_ordered_extent(ordered_extent); | |
3125 | /* once for the tree */ | |
3126 | btrfs_put_ordered_extent(ordered_extent); | |
3127 | ||
5fd02043 JB |
3128 | return ret; |
3129 | } | |
3130 | ||
3131 | static void finish_ordered_fn(struct btrfs_work *work) | |
3132 | { | |
3133 | struct btrfs_ordered_extent *ordered_extent; | |
3134 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
3135 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
3136 | } |
3137 | ||
c3988d63 | 3138 | static void btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
3139 | struct extent_state *state, int uptodate) |
3140 | { | |
5fd02043 | 3141 | struct inode *inode = page->mapping->host; |
0b246afa | 3142 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5fd02043 | 3143 | struct btrfs_ordered_extent *ordered_extent = NULL; |
9e0af237 LB |
3144 | struct btrfs_workqueue *wq; |
3145 | btrfs_work_func_t func; | |
5fd02043 | 3146 | |
1abe9b8a | 3147 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
3148 | ||
8b62b72b | 3149 | ClearPagePrivate2(page); |
5fd02043 JB |
3150 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
3151 | end - start + 1, uptodate)) | |
c3988d63 | 3152 | return; |
5fd02043 | 3153 | |
70ddc553 | 3154 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) { |
0b246afa | 3155 | wq = fs_info->endio_freespace_worker; |
9e0af237 LB |
3156 | func = btrfs_freespace_write_helper; |
3157 | } else { | |
0b246afa | 3158 | wq = fs_info->endio_write_workers; |
9e0af237 LB |
3159 | func = btrfs_endio_write_helper; |
3160 | } | |
5fd02043 | 3161 | |
9e0af237 LB |
3162 | btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL, |
3163 | NULL); | |
3164 | btrfs_queue_work(wq, &ordered_extent->work); | |
211f90e6 CM |
3165 | } |
3166 | ||
dc380aea MX |
3167 | static int __readpage_endio_check(struct inode *inode, |
3168 | struct btrfs_io_bio *io_bio, | |
3169 | int icsum, struct page *page, | |
3170 | int pgoff, u64 start, size_t len) | |
3171 | { | |
3172 | char *kaddr; | |
3173 | u32 csum_expected; | |
3174 | u32 csum = ~(u32)0; | |
dc380aea MX |
3175 | |
3176 | csum_expected = *(((u32 *)io_bio->csum) + icsum); | |
3177 | ||
3178 | kaddr = kmap_atomic(page); | |
3179 | csum = btrfs_csum_data(kaddr + pgoff, csum, len); | |
0b5e3daf | 3180 | btrfs_csum_final(csum, (u8 *)&csum); |
dc380aea MX |
3181 | if (csum != csum_expected) |
3182 | goto zeroit; | |
3183 | ||
3184 | kunmap_atomic(kaddr); | |
3185 | return 0; | |
3186 | zeroit: | |
0970a22e | 3187 | btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected, |
6f6b643e | 3188 | io_bio->mirror_num); |
dc380aea MX |
3189 | memset(kaddr + pgoff, 1, len); |
3190 | flush_dcache_page(page); | |
3191 | kunmap_atomic(kaddr); | |
dc380aea MX |
3192 | return -EIO; |
3193 | } | |
3194 | ||
d352ac68 CM |
3195 | /* |
3196 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
3197 | * if there's a match, we allow the bio to finish. If not, the code in |
3198 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 3199 | */ |
facc8a22 MX |
3200 | static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio, |
3201 | u64 phy_offset, struct page *page, | |
3202 | u64 start, u64 end, int mirror) | |
07157aac | 3203 | { |
4eee4fa4 | 3204 | size_t offset = start - page_offset(page); |
07157aac | 3205 | struct inode *inode = page->mapping->host; |
d1310b2e | 3206 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
ff79f819 | 3207 | struct btrfs_root *root = BTRFS_I(inode)->root; |
d1310b2e | 3208 | |
d20f7043 CM |
3209 | if (PageChecked(page)) { |
3210 | ClearPageChecked(page); | |
dc380aea | 3211 | return 0; |
d20f7043 | 3212 | } |
6cbff00f CH |
3213 | |
3214 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
dc380aea | 3215 | return 0; |
17d217fe YZ |
3216 | |
3217 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 3218 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
91166212 | 3219 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM); |
b6cda9bc | 3220 | return 0; |
17d217fe | 3221 | } |
d20f7043 | 3222 | |
facc8a22 | 3223 | phy_offset >>= inode->i_sb->s_blocksize_bits; |
dc380aea MX |
3224 | return __readpage_endio_check(inode, io_bio, phy_offset, page, offset, |
3225 | start, (size_t)(end - start + 1)); | |
07157aac | 3226 | } |
b888db2b | 3227 | |
24bbcf04 YZ |
3228 | void btrfs_add_delayed_iput(struct inode *inode) |
3229 | { | |
0b246afa | 3230 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
8089fe62 | 3231 | struct btrfs_inode *binode = BTRFS_I(inode); |
24bbcf04 YZ |
3232 | |
3233 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
3234 | return; | |
3235 | ||
24bbcf04 | 3236 | spin_lock(&fs_info->delayed_iput_lock); |
8089fe62 DS |
3237 | if (binode->delayed_iput_count == 0) { |
3238 | ASSERT(list_empty(&binode->delayed_iput)); | |
3239 | list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs); | |
3240 | } else { | |
3241 | binode->delayed_iput_count++; | |
3242 | } | |
24bbcf04 YZ |
3243 | spin_unlock(&fs_info->delayed_iput_lock); |
3244 | } | |
3245 | ||
2ff7e61e | 3246 | void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info) |
24bbcf04 | 3247 | { |
24bbcf04 | 3248 | |
24bbcf04 | 3249 | spin_lock(&fs_info->delayed_iput_lock); |
8089fe62 DS |
3250 | while (!list_empty(&fs_info->delayed_iputs)) { |
3251 | struct btrfs_inode *inode; | |
3252 | ||
3253 | inode = list_first_entry(&fs_info->delayed_iputs, | |
3254 | struct btrfs_inode, delayed_iput); | |
3255 | if (inode->delayed_iput_count) { | |
3256 | inode->delayed_iput_count--; | |
3257 | list_move_tail(&inode->delayed_iput, | |
3258 | &fs_info->delayed_iputs); | |
3259 | } else { | |
3260 | list_del_init(&inode->delayed_iput); | |
3261 | } | |
3262 | spin_unlock(&fs_info->delayed_iput_lock); | |
3263 | iput(&inode->vfs_inode); | |
3264 | spin_lock(&fs_info->delayed_iput_lock); | |
24bbcf04 | 3265 | } |
8089fe62 | 3266 | spin_unlock(&fs_info->delayed_iput_lock); |
24bbcf04 YZ |
3267 | } |
3268 | ||
d68fc57b | 3269 | /* |
42b2aa86 | 3270 | * This is called in transaction commit time. If there are no orphan |
d68fc57b YZ |
3271 | * files in the subvolume, it removes orphan item and frees block_rsv |
3272 | * structure. | |
3273 | */ | |
3274 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
3275 | struct btrfs_root *root) | |
3276 | { | |
0b246afa | 3277 | struct btrfs_fs_info *fs_info = root->fs_info; |
90290e19 | 3278 | struct btrfs_block_rsv *block_rsv; |
d68fc57b YZ |
3279 | int ret; |
3280 | ||
8a35d95f | 3281 | if (atomic_read(&root->orphan_inodes) || |
d68fc57b YZ |
3282 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) |
3283 | return; | |
3284 | ||
90290e19 | 3285 | spin_lock(&root->orphan_lock); |
8a35d95f | 3286 | if (atomic_read(&root->orphan_inodes)) { |
90290e19 JB |
3287 | spin_unlock(&root->orphan_lock); |
3288 | return; | |
3289 | } | |
3290 | ||
3291 | if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) { | |
3292 | spin_unlock(&root->orphan_lock); | |
3293 | return; | |
3294 | } | |
3295 | ||
3296 | block_rsv = root->orphan_block_rsv; | |
3297 | root->orphan_block_rsv = NULL; | |
3298 | spin_unlock(&root->orphan_lock); | |
3299 | ||
27cdeb70 | 3300 | if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) && |
d68fc57b | 3301 | btrfs_root_refs(&root->root_item) > 0) { |
0b246afa | 3302 | ret = btrfs_del_orphan_item(trans, fs_info->tree_root, |
d68fc57b | 3303 | root->root_key.objectid); |
4ef31a45 | 3304 | if (ret) |
66642832 | 3305 | btrfs_abort_transaction(trans, ret); |
4ef31a45 | 3306 | else |
27cdeb70 MX |
3307 | clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, |
3308 | &root->state); | |
d68fc57b YZ |
3309 | } |
3310 | ||
90290e19 JB |
3311 | if (block_rsv) { |
3312 | WARN_ON(block_rsv->size > 0); | |
2ff7e61e | 3313 | btrfs_free_block_rsv(fs_info, block_rsv); |
d68fc57b YZ |
3314 | } |
3315 | } | |
3316 | ||
7b128766 JB |
3317 | /* |
3318 | * This creates an orphan entry for the given inode in case something goes | |
3319 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
3320 | * |
3321 | * NOTE: caller of this function should reserve 5 units of metadata for | |
3322 | * this function. | |
7b128766 | 3323 | */ |
73f2e545 NB |
3324 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, |
3325 | struct btrfs_inode *inode) | |
7b128766 | 3326 | { |
73f2e545 NB |
3327 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
3328 | struct btrfs_root *root = inode->root; | |
d68fc57b YZ |
3329 | struct btrfs_block_rsv *block_rsv = NULL; |
3330 | int reserve = 0; | |
3331 | int insert = 0; | |
3332 | int ret; | |
7b128766 | 3333 | |
d68fc57b | 3334 | if (!root->orphan_block_rsv) { |
2ff7e61e JM |
3335 | block_rsv = btrfs_alloc_block_rsv(fs_info, |
3336 | BTRFS_BLOCK_RSV_TEMP); | |
b532402e TI |
3337 | if (!block_rsv) |
3338 | return -ENOMEM; | |
d68fc57b | 3339 | } |
7b128766 | 3340 | |
d68fc57b YZ |
3341 | spin_lock(&root->orphan_lock); |
3342 | if (!root->orphan_block_rsv) { | |
3343 | root->orphan_block_rsv = block_rsv; | |
3344 | } else if (block_rsv) { | |
2ff7e61e | 3345 | btrfs_free_block_rsv(fs_info, block_rsv); |
d68fc57b | 3346 | block_rsv = NULL; |
7b128766 | 3347 | } |
7b128766 | 3348 | |
8a35d95f | 3349 | if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
73f2e545 | 3350 | &inode->runtime_flags)) { |
d68fc57b YZ |
3351 | #if 0 |
3352 | /* | |
3353 | * For proper ENOSPC handling, we should do orphan | |
3354 | * cleanup when mounting. But this introduces backward | |
3355 | * compatibility issue. | |
3356 | */ | |
3357 | if (!xchg(&root->orphan_item_inserted, 1)) | |
3358 | insert = 2; | |
3359 | else | |
3360 | insert = 1; | |
3361 | #endif | |
3362 | insert = 1; | |
321f0e70 | 3363 | atomic_inc(&root->orphan_inodes); |
7b128766 JB |
3364 | } |
3365 | ||
72ac3c0d | 3366 | if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
73f2e545 | 3367 | &inode->runtime_flags)) |
d68fc57b | 3368 | reserve = 1; |
d68fc57b | 3369 | spin_unlock(&root->orphan_lock); |
7b128766 | 3370 | |
d68fc57b YZ |
3371 | /* grab metadata reservation from transaction handle */ |
3372 | if (reserve) { | |
3373 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
3b6571c1 JB |
3374 | ASSERT(!ret); |
3375 | if (ret) { | |
3376 | atomic_dec(&root->orphan_inodes); | |
3377 | clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, | |
73f2e545 | 3378 | &inode->runtime_flags); |
3b6571c1 JB |
3379 | if (insert) |
3380 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, | |
73f2e545 | 3381 | &inode->runtime_flags); |
3b6571c1 JB |
3382 | return ret; |
3383 | } | |
d68fc57b | 3384 | } |
7b128766 | 3385 | |
d68fc57b YZ |
3386 | /* insert an orphan item to track this unlinked/truncated file */ |
3387 | if (insert >= 1) { | |
73f2e545 | 3388 | ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode)); |
4ef31a45 | 3389 | if (ret) { |
703c88e0 | 3390 | atomic_dec(&root->orphan_inodes); |
4ef31a45 JB |
3391 | if (reserve) { |
3392 | clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, | |
73f2e545 | 3393 | &inode->runtime_flags); |
4ef31a45 JB |
3394 | btrfs_orphan_release_metadata(inode); |
3395 | } | |
3396 | if (ret != -EEXIST) { | |
e8e7cff6 | 3397 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
73f2e545 | 3398 | &inode->runtime_flags); |
66642832 | 3399 | btrfs_abort_transaction(trans, ret); |
4ef31a45 JB |
3400 | return ret; |
3401 | } | |
79787eaa JM |
3402 | } |
3403 | ret = 0; | |
d68fc57b YZ |
3404 | } |
3405 | ||
3406 | /* insert an orphan item to track subvolume contains orphan files */ | |
3407 | if (insert >= 2) { | |
0b246afa | 3408 | ret = btrfs_insert_orphan_item(trans, fs_info->tree_root, |
d68fc57b | 3409 | root->root_key.objectid); |
79787eaa | 3410 | if (ret && ret != -EEXIST) { |
66642832 | 3411 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3412 | return ret; |
3413 | } | |
d68fc57b YZ |
3414 | } |
3415 | return 0; | |
7b128766 JB |
3416 | } |
3417 | ||
3418 | /* | |
3419 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
3420 | * item for this particular inode. | |
3421 | */ | |
48a3b636 | 3422 | static int btrfs_orphan_del(struct btrfs_trans_handle *trans, |
3d6ae7bb | 3423 | struct btrfs_inode *inode) |
7b128766 | 3424 | { |
3d6ae7bb | 3425 | struct btrfs_root *root = inode->root; |
d68fc57b YZ |
3426 | int delete_item = 0; |
3427 | int release_rsv = 0; | |
7b128766 JB |
3428 | int ret = 0; |
3429 | ||
d68fc57b | 3430 | spin_lock(&root->orphan_lock); |
8a35d95f | 3431 | if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3d6ae7bb | 3432 | &inode->runtime_flags)) |
d68fc57b | 3433 | delete_item = 1; |
7b128766 | 3434 | |
72ac3c0d | 3435 | if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
3d6ae7bb | 3436 | &inode->runtime_flags)) |
d68fc57b | 3437 | release_rsv = 1; |
d68fc57b | 3438 | spin_unlock(&root->orphan_lock); |
7b128766 | 3439 | |
703c88e0 | 3440 | if (delete_item) { |
8a35d95f | 3441 | atomic_dec(&root->orphan_inodes); |
703c88e0 FDBM |
3442 | if (trans) |
3443 | ret = btrfs_del_orphan_item(trans, root, | |
3d6ae7bb | 3444 | btrfs_ino(inode)); |
8a35d95f | 3445 | } |
7b128766 | 3446 | |
703c88e0 FDBM |
3447 | if (release_rsv) |
3448 | btrfs_orphan_release_metadata(inode); | |
3449 | ||
4ef31a45 | 3450 | return ret; |
7b128766 JB |
3451 | } |
3452 | ||
3453 | /* | |
3454 | * this cleans up any orphans that may be left on the list from the last use | |
3455 | * of this root. | |
3456 | */ | |
66b4ffd1 | 3457 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 | 3458 | { |
0b246afa | 3459 | struct btrfs_fs_info *fs_info = root->fs_info; |
7b128766 JB |
3460 | struct btrfs_path *path; |
3461 | struct extent_buffer *leaf; | |
7b128766 JB |
3462 | struct btrfs_key key, found_key; |
3463 | struct btrfs_trans_handle *trans; | |
3464 | struct inode *inode; | |
8f6d7f4f | 3465 | u64 last_objectid = 0; |
7b128766 JB |
3466 | int ret = 0, nr_unlink = 0, nr_truncate = 0; |
3467 | ||
d68fc57b | 3468 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 3469 | return 0; |
c71bf099 YZ |
3470 | |
3471 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
3472 | if (!path) { |
3473 | ret = -ENOMEM; | |
3474 | goto out; | |
3475 | } | |
e4058b54 | 3476 | path->reada = READA_BACK; |
7b128766 JB |
3477 | |
3478 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
962a298f | 3479 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
7b128766 JB |
3480 | key.offset = (u64)-1; |
3481 | ||
7b128766 JB |
3482 | while (1) { |
3483 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
3484 | if (ret < 0) |
3485 | goto out; | |
7b128766 JB |
3486 | |
3487 | /* | |
3488 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 3489 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
3490 | * find the key and see if we have stuff that matches |
3491 | */ | |
3492 | if (ret > 0) { | |
66b4ffd1 | 3493 | ret = 0; |
7b128766 JB |
3494 | if (path->slots[0] == 0) |
3495 | break; | |
3496 | path->slots[0]--; | |
3497 | } | |
3498 | ||
3499 | /* pull out the item */ | |
3500 | leaf = path->nodes[0]; | |
7b128766 JB |
3501 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3502 | ||
3503 | /* make sure the item matches what we want */ | |
3504 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
3505 | break; | |
962a298f | 3506 | if (found_key.type != BTRFS_ORPHAN_ITEM_KEY) |
7b128766 JB |
3507 | break; |
3508 | ||
3509 | /* release the path since we're done with it */ | |
b3b4aa74 | 3510 | btrfs_release_path(path); |
7b128766 JB |
3511 | |
3512 | /* | |
3513 | * this is where we are basically btrfs_lookup, without the | |
3514 | * crossing root thing. we store the inode number in the | |
3515 | * offset of the orphan item. | |
3516 | */ | |
8f6d7f4f JB |
3517 | |
3518 | if (found_key.offset == last_objectid) { | |
0b246afa JM |
3519 | btrfs_err(fs_info, |
3520 | "Error removing orphan entry, stopping orphan cleanup"); | |
8f6d7f4f JB |
3521 | ret = -EINVAL; |
3522 | goto out; | |
3523 | } | |
3524 | ||
3525 | last_objectid = found_key.offset; | |
3526 | ||
5d4f98a2 YZ |
3527 | found_key.objectid = found_key.offset; |
3528 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
3529 | found_key.offset = 0; | |
0b246afa | 3530 | inode = btrfs_iget(fs_info->sb, &found_key, root, NULL); |
8c6ffba0 | 3531 | ret = PTR_ERR_OR_ZERO(inode); |
67710892 | 3532 | if (ret && ret != -ENOENT) |
66b4ffd1 | 3533 | goto out; |
7b128766 | 3534 | |
0b246afa | 3535 | if (ret == -ENOENT && root == fs_info->tree_root) { |
f8e9e0b0 AJ |
3536 | struct btrfs_root *dead_root; |
3537 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3538 | int is_dead_root = 0; | |
3539 | ||
3540 | /* | |
3541 | * this is an orphan in the tree root. Currently these | |
3542 | * could come from 2 sources: | |
3543 | * a) a snapshot deletion in progress | |
3544 | * b) a free space cache inode | |
3545 | * We need to distinguish those two, as the snapshot | |
3546 | * orphan must not get deleted. | |
3547 | * find_dead_roots already ran before us, so if this | |
3548 | * is a snapshot deletion, we should find the root | |
3549 | * in the dead_roots list | |
3550 | */ | |
3551 | spin_lock(&fs_info->trans_lock); | |
3552 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
3553 | root_list) { | |
3554 | if (dead_root->root_key.objectid == | |
3555 | found_key.objectid) { | |
3556 | is_dead_root = 1; | |
3557 | break; | |
3558 | } | |
3559 | } | |
3560 | spin_unlock(&fs_info->trans_lock); | |
3561 | if (is_dead_root) { | |
3562 | /* prevent this orphan from being found again */ | |
3563 | key.offset = found_key.objectid - 1; | |
3564 | continue; | |
3565 | } | |
3566 | } | |
7b128766 | 3567 | /* |
a8c9e576 JB |
3568 | * Inode is already gone but the orphan item is still there, |
3569 | * kill the orphan item. | |
7b128766 | 3570 | */ |
67710892 | 3571 | if (ret == -ENOENT) { |
a8c9e576 | 3572 | trans = btrfs_start_transaction(root, 1); |
66b4ffd1 JB |
3573 | if (IS_ERR(trans)) { |
3574 | ret = PTR_ERR(trans); | |
3575 | goto out; | |
3576 | } | |
0b246afa JM |
3577 | btrfs_debug(fs_info, "auto deleting %Lu", |
3578 | found_key.objectid); | |
a8c9e576 JB |
3579 | ret = btrfs_del_orphan_item(trans, root, |
3580 | found_key.objectid); | |
3a45bb20 | 3581 | btrfs_end_transaction(trans); |
4ef31a45 JB |
3582 | if (ret) |
3583 | goto out; | |
7b128766 JB |
3584 | continue; |
3585 | } | |
3586 | ||
a8c9e576 JB |
3587 | /* |
3588 | * add this inode to the orphan list so btrfs_orphan_del does | |
3589 | * the proper thing when we hit it | |
3590 | */ | |
8a35d95f JB |
3591 | set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3592 | &BTRFS_I(inode)->runtime_flags); | |
925396ec | 3593 | atomic_inc(&root->orphan_inodes); |
a8c9e576 | 3594 | |
7b128766 JB |
3595 | /* if we have links, this was a truncate, lets do that */ |
3596 | if (inode->i_nlink) { | |
fae7f21c | 3597 | if (WARN_ON(!S_ISREG(inode->i_mode))) { |
a41ad394 JB |
3598 | iput(inode); |
3599 | continue; | |
3600 | } | |
7b128766 | 3601 | nr_truncate++; |
f3fe820c JB |
3602 | |
3603 | /* 1 for the orphan item deletion. */ | |
3604 | trans = btrfs_start_transaction(root, 1); | |
3605 | if (IS_ERR(trans)) { | |
c69b26b0 | 3606 | iput(inode); |
f3fe820c JB |
3607 | ret = PTR_ERR(trans); |
3608 | goto out; | |
3609 | } | |
73f2e545 | 3610 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
3a45bb20 | 3611 | btrfs_end_transaction(trans); |
c69b26b0 JB |
3612 | if (ret) { |
3613 | iput(inode); | |
f3fe820c | 3614 | goto out; |
c69b26b0 | 3615 | } |
f3fe820c | 3616 | |
66b4ffd1 | 3617 | ret = btrfs_truncate(inode); |
4a7d0f68 | 3618 | if (ret) |
3d6ae7bb | 3619 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
7b128766 JB |
3620 | } else { |
3621 | nr_unlink++; | |
3622 | } | |
3623 | ||
3624 | /* this will do delete_inode and everything for us */ | |
3625 | iput(inode); | |
66b4ffd1 JB |
3626 | if (ret) |
3627 | goto out; | |
7b128766 | 3628 | } |
3254c876 MX |
3629 | /* release the path since we're done with it */ |
3630 | btrfs_release_path(path); | |
3631 | ||
d68fc57b YZ |
3632 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
3633 | ||
3634 | if (root->orphan_block_rsv) | |
2ff7e61e | 3635 | btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, |
d68fc57b YZ |
3636 | (u64)-1); |
3637 | ||
27cdeb70 MX |
3638 | if (root->orphan_block_rsv || |
3639 | test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) { | |
7a7eaa40 | 3640 | trans = btrfs_join_transaction(root); |
66b4ffd1 | 3641 | if (!IS_ERR(trans)) |
3a45bb20 | 3642 | btrfs_end_transaction(trans); |
d68fc57b | 3643 | } |
7b128766 JB |
3644 | |
3645 | if (nr_unlink) | |
0b246afa | 3646 | btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink); |
7b128766 | 3647 | if (nr_truncate) |
0b246afa | 3648 | btrfs_debug(fs_info, "truncated %d orphans", nr_truncate); |
66b4ffd1 JB |
3649 | |
3650 | out: | |
3651 | if (ret) | |
0b246afa | 3652 | btrfs_err(fs_info, "could not do orphan cleanup %d", ret); |
66b4ffd1 JB |
3653 | btrfs_free_path(path); |
3654 | return ret; | |
7b128766 JB |
3655 | } |
3656 | ||
46a53cca CM |
3657 | /* |
3658 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
3659 | * don't find any xattrs, we know there can't be any acls. | |
3660 | * | |
3661 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
3662 | */ | |
3663 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
63541927 FDBM |
3664 | int slot, u64 objectid, |
3665 | int *first_xattr_slot) | |
46a53cca CM |
3666 | { |
3667 | u32 nritems = btrfs_header_nritems(leaf); | |
3668 | struct btrfs_key found_key; | |
f23b5a59 JB |
3669 | static u64 xattr_access = 0; |
3670 | static u64 xattr_default = 0; | |
46a53cca CM |
3671 | int scanned = 0; |
3672 | ||
f23b5a59 | 3673 | if (!xattr_access) { |
97d79299 AG |
3674 | xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS, |
3675 | strlen(XATTR_NAME_POSIX_ACL_ACCESS)); | |
3676 | xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT, | |
3677 | strlen(XATTR_NAME_POSIX_ACL_DEFAULT)); | |
f23b5a59 JB |
3678 | } |
3679 | ||
46a53cca | 3680 | slot++; |
63541927 | 3681 | *first_xattr_slot = -1; |
46a53cca CM |
3682 | while (slot < nritems) { |
3683 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3684 | ||
3685 | /* we found a different objectid, there must not be acls */ | |
3686 | if (found_key.objectid != objectid) | |
3687 | return 0; | |
3688 | ||
3689 | /* we found an xattr, assume we've got an acl */ | |
f23b5a59 | 3690 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) { |
63541927 FDBM |
3691 | if (*first_xattr_slot == -1) |
3692 | *first_xattr_slot = slot; | |
f23b5a59 JB |
3693 | if (found_key.offset == xattr_access || |
3694 | found_key.offset == xattr_default) | |
3695 | return 1; | |
3696 | } | |
46a53cca CM |
3697 | |
3698 | /* | |
3699 | * we found a key greater than an xattr key, there can't | |
3700 | * be any acls later on | |
3701 | */ | |
3702 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
3703 | return 0; | |
3704 | ||
3705 | slot++; | |
3706 | scanned++; | |
3707 | ||
3708 | /* | |
3709 | * it goes inode, inode backrefs, xattrs, extents, | |
3710 | * so if there are a ton of hard links to an inode there can | |
3711 | * be a lot of backrefs. Don't waste time searching too hard, | |
3712 | * this is just an optimization | |
3713 | */ | |
3714 | if (scanned >= 8) | |
3715 | break; | |
3716 | } | |
3717 | /* we hit the end of the leaf before we found an xattr or | |
3718 | * something larger than an xattr. We have to assume the inode | |
3719 | * has acls | |
3720 | */ | |
63541927 FDBM |
3721 | if (*first_xattr_slot == -1) |
3722 | *first_xattr_slot = slot; | |
46a53cca CM |
3723 | return 1; |
3724 | } | |
3725 | ||
d352ac68 CM |
3726 | /* |
3727 | * read an inode from the btree into the in-memory inode | |
3728 | */ | |
67710892 | 3729 | static int btrfs_read_locked_inode(struct inode *inode) |
39279cc3 | 3730 | { |
0b246afa | 3731 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 3732 | struct btrfs_path *path; |
5f39d397 | 3733 | struct extent_buffer *leaf; |
39279cc3 CM |
3734 | struct btrfs_inode_item *inode_item; |
3735 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3736 | struct btrfs_key location; | |
67de1176 | 3737 | unsigned long ptr; |
46a53cca | 3738 | int maybe_acls; |
618e21d5 | 3739 | u32 rdev; |
39279cc3 | 3740 | int ret; |
2f7e33d4 | 3741 | bool filled = false; |
63541927 | 3742 | int first_xattr_slot; |
2f7e33d4 MX |
3743 | |
3744 | ret = btrfs_fill_inode(inode, &rdev); | |
3745 | if (!ret) | |
3746 | filled = true; | |
39279cc3 CM |
3747 | |
3748 | path = btrfs_alloc_path(); | |
67710892 FM |
3749 | if (!path) { |
3750 | ret = -ENOMEM; | |
1748f843 | 3751 | goto make_bad; |
67710892 | 3752 | } |
1748f843 | 3753 | |
39279cc3 | 3754 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 3755 | |
39279cc3 | 3756 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
67710892 FM |
3757 | if (ret) { |
3758 | if (ret > 0) | |
3759 | ret = -ENOENT; | |
39279cc3 | 3760 | goto make_bad; |
67710892 | 3761 | } |
39279cc3 | 3762 | |
5f39d397 | 3763 | leaf = path->nodes[0]; |
2f7e33d4 MX |
3764 | |
3765 | if (filled) | |
67de1176 | 3766 | goto cache_index; |
2f7e33d4 | 3767 | |
5f39d397 CM |
3768 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
3769 | struct btrfs_inode_item); | |
5f39d397 | 3770 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 3771 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
3772 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
3773 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
6ef06d27 | 3774 | btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item)); |
5f39d397 | 3775 | |
a937b979 DS |
3776 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime); |
3777 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime); | |
5f39d397 | 3778 | |
a937b979 DS |
3779 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime); |
3780 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime); | |
5f39d397 | 3781 | |
a937b979 DS |
3782 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime); |
3783 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime); | |
5f39d397 | 3784 | |
9cc97d64 | 3785 | BTRFS_I(inode)->i_otime.tv_sec = |
3786 | btrfs_timespec_sec(leaf, &inode_item->otime); | |
3787 | BTRFS_I(inode)->i_otime.tv_nsec = | |
3788 | btrfs_timespec_nsec(leaf, &inode_item->otime); | |
5f39d397 | 3789 | |
a76a3cd4 | 3790 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 3791 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
3792 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
3793 | ||
6e17d30b YD |
3794 | inode->i_version = btrfs_inode_sequence(leaf, inode_item); |
3795 | inode->i_generation = BTRFS_I(inode)->generation; | |
3796 | inode->i_rdev = 0; | |
3797 | rdev = btrfs_inode_rdev(leaf, inode_item); | |
3798 | ||
3799 | BTRFS_I(inode)->index_cnt = (u64)-1; | |
3800 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); | |
3801 | ||
3802 | cache_index: | |
5dc562c5 JB |
3803 | /* |
3804 | * If we were modified in the current generation and evicted from memory | |
3805 | * and then re-read we need to do a full sync since we don't have any | |
3806 | * idea about which extents were modified before we were evicted from | |
3807 | * cache. | |
6e17d30b YD |
3808 | * |
3809 | * This is required for both inode re-read from disk and delayed inode | |
3810 | * in delayed_nodes_tree. | |
5dc562c5 | 3811 | */ |
0b246afa | 3812 | if (BTRFS_I(inode)->last_trans == fs_info->generation) |
5dc562c5 JB |
3813 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
3814 | &BTRFS_I(inode)->runtime_flags); | |
3815 | ||
bde6c242 FM |
3816 | /* |
3817 | * We don't persist the id of the transaction where an unlink operation | |
3818 | * against the inode was last made. So here we assume the inode might | |
3819 | * have been evicted, and therefore the exact value of last_unlink_trans | |
3820 | * lost, and set it to last_trans to avoid metadata inconsistencies | |
3821 | * between the inode and its parent if the inode is fsync'ed and the log | |
3822 | * replayed. For example, in the scenario: | |
3823 | * | |
3824 | * touch mydir/foo | |
3825 | * ln mydir/foo mydir/bar | |
3826 | * sync | |
3827 | * unlink mydir/bar | |
3828 | * echo 2 > /proc/sys/vm/drop_caches # evicts inode | |
3829 | * xfs_io -c fsync mydir/foo | |
3830 | * <power failure> | |
3831 | * mount fs, triggers fsync log replay | |
3832 | * | |
3833 | * We must make sure that when we fsync our inode foo we also log its | |
3834 | * parent inode, otherwise after log replay the parent still has the | |
3835 | * dentry with the "bar" name but our inode foo has a link count of 1 | |
3836 | * and doesn't have an inode ref with the name "bar" anymore. | |
3837 | * | |
3838 | * Setting last_unlink_trans to last_trans is a pessimistic approach, | |
01327610 | 3839 | * but it guarantees correctness at the expense of occasional full |
bde6c242 FM |
3840 | * transaction commits on fsync if our inode is a directory, or if our |
3841 | * inode is not a directory, logging its parent unnecessarily. | |
3842 | */ | |
3843 | BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans; | |
3844 | ||
67de1176 MX |
3845 | path->slots[0]++; |
3846 | if (inode->i_nlink != 1 || | |
3847 | path->slots[0] >= btrfs_header_nritems(leaf)) | |
3848 | goto cache_acl; | |
3849 | ||
3850 | btrfs_item_key_to_cpu(leaf, &location, path->slots[0]); | |
4a0cc7ca | 3851 | if (location.objectid != btrfs_ino(BTRFS_I(inode))) |
67de1176 MX |
3852 | goto cache_acl; |
3853 | ||
3854 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3855 | if (location.type == BTRFS_INODE_REF_KEY) { | |
3856 | struct btrfs_inode_ref *ref; | |
3857 | ||
3858 | ref = (struct btrfs_inode_ref *)ptr; | |
3859 | BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref); | |
3860 | } else if (location.type == BTRFS_INODE_EXTREF_KEY) { | |
3861 | struct btrfs_inode_extref *extref; | |
3862 | ||
3863 | extref = (struct btrfs_inode_extref *)ptr; | |
3864 | BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf, | |
3865 | extref); | |
3866 | } | |
2f7e33d4 | 3867 | cache_acl: |
46a53cca CM |
3868 | /* |
3869 | * try to precache a NULL acl entry for files that don't have | |
3870 | * any xattrs or acls | |
3871 | */ | |
33345d01 | 3872 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
f85b7379 | 3873 | btrfs_ino(BTRFS_I(inode)), &first_xattr_slot); |
63541927 FDBM |
3874 | if (first_xattr_slot != -1) { |
3875 | path->slots[0] = first_xattr_slot; | |
3876 | ret = btrfs_load_inode_props(inode, path); | |
3877 | if (ret) | |
0b246afa | 3878 | btrfs_err(fs_info, |
351fd353 | 3879 | "error loading props for ino %llu (root %llu): %d", |
4a0cc7ca | 3880 | btrfs_ino(BTRFS_I(inode)), |
63541927 FDBM |
3881 | root->root_key.objectid, ret); |
3882 | } | |
3883 | btrfs_free_path(path); | |
3884 | ||
72c04902 AV |
3885 | if (!maybe_acls) |
3886 | cache_no_acl(inode); | |
46a53cca | 3887 | |
39279cc3 | 3888 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
3889 | case S_IFREG: |
3890 | inode->i_mapping->a_ops = &btrfs_aops; | |
d1310b2e | 3891 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
3892 | inode->i_fop = &btrfs_file_operations; |
3893 | inode->i_op = &btrfs_file_inode_operations; | |
3894 | break; | |
3895 | case S_IFDIR: | |
3896 | inode->i_fop = &btrfs_dir_file_operations; | |
67ade058 | 3897 | inode->i_op = &btrfs_dir_inode_operations; |
39279cc3 CM |
3898 | break; |
3899 | case S_IFLNK: | |
3900 | inode->i_op = &btrfs_symlink_inode_operations; | |
21fc61c7 | 3901 | inode_nohighmem(inode); |
39279cc3 CM |
3902 | inode->i_mapping->a_ops = &btrfs_symlink_aops; |
3903 | break; | |
618e21d5 | 3904 | default: |
0279b4cd | 3905 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
3906 | init_special_inode(inode, inode->i_mode, rdev); |
3907 | break; | |
39279cc3 | 3908 | } |
6cbff00f CH |
3909 | |
3910 | btrfs_update_iflags(inode); | |
67710892 | 3911 | return 0; |
39279cc3 CM |
3912 | |
3913 | make_bad: | |
39279cc3 | 3914 | btrfs_free_path(path); |
39279cc3 | 3915 | make_bad_inode(inode); |
67710892 | 3916 | return ret; |
39279cc3 CM |
3917 | } |
3918 | ||
d352ac68 CM |
3919 | /* |
3920 | * given a leaf and an inode, copy the inode fields into the leaf | |
3921 | */ | |
e02119d5 CM |
3922 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3923 | struct extent_buffer *leaf, | |
5f39d397 | 3924 | struct btrfs_inode_item *item, |
39279cc3 CM |
3925 | struct inode *inode) |
3926 | { | |
51fab693 LB |
3927 | struct btrfs_map_token token; |
3928 | ||
3929 | btrfs_init_map_token(&token); | |
5f39d397 | 3930 | |
51fab693 LB |
3931 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); |
3932 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3933 | btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size, | |
3934 | &token); | |
3935 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3936 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
5f39d397 | 3937 | |
a937b979 | 3938 | btrfs_set_token_timespec_sec(leaf, &item->atime, |
51fab693 | 3939 | inode->i_atime.tv_sec, &token); |
a937b979 | 3940 | btrfs_set_token_timespec_nsec(leaf, &item->atime, |
51fab693 | 3941 | inode->i_atime.tv_nsec, &token); |
5f39d397 | 3942 | |
a937b979 | 3943 | btrfs_set_token_timespec_sec(leaf, &item->mtime, |
51fab693 | 3944 | inode->i_mtime.tv_sec, &token); |
a937b979 | 3945 | btrfs_set_token_timespec_nsec(leaf, &item->mtime, |
51fab693 | 3946 | inode->i_mtime.tv_nsec, &token); |
5f39d397 | 3947 | |
a937b979 | 3948 | btrfs_set_token_timespec_sec(leaf, &item->ctime, |
51fab693 | 3949 | inode->i_ctime.tv_sec, &token); |
a937b979 | 3950 | btrfs_set_token_timespec_nsec(leaf, &item->ctime, |
51fab693 | 3951 | inode->i_ctime.tv_nsec, &token); |
5f39d397 | 3952 | |
9cc97d64 | 3953 | btrfs_set_token_timespec_sec(leaf, &item->otime, |
3954 | BTRFS_I(inode)->i_otime.tv_sec, &token); | |
3955 | btrfs_set_token_timespec_nsec(leaf, &item->otime, | |
3956 | BTRFS_I(inode)->i_otime.tv_nsec, &token); | |
3957 | ||
51fab693 LB |
3958 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), |
3959 | &token); | |
3960 | btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation, | |
3961 | &token); | |
3962 | btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token); | |
3963 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); | |
3964 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3965 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3966 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
39279cc3 CM |
3967 | } |
3968 | ||
d352ac68 CM |
3969 | /* |
3970 | * copy everything in the in-memory inode into the btree. | |
3971 | */ | |
2115133f | 3972 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 3973 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
3974 | { |
3975 | struct btrfs_inode_item *inode_item; | |
3976 | struct btrfs_path *path; | |
5f39d397 | 3977 | struct extent_buffer *leaf; |
39279cc3 CM |
3978 | int ret; |
3979 | ||
3980 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
3981 | if (!path) |
3982 | return -ENOMEM; | |
3983 | ||
b9473439 | 3984 | path->leave_spinning = 1; |
16cdcec7 MX |
3985 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
3986 | 1); | |
39279cc3 CM |
3987 | if (ret) { |
3988 | if (ret > 0) | |
3989 | ret = -ENOENT; | |
3990 | goto failed; | |
3991 | } | |
3992 | ||
5f39d397 CM |
3993 | leaf = path->nodes[0]; |
3994 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 3995 | struct btrfs_inode_item); |
39279cc3 | 3996 | |
e02119d5 | 3997 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 3998 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 3999 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
4000 | ret = 0; |
4001 | failed: | |
39279cc3 CM |
4002 | btrfs_free_path(path); |
4003 | return ret; | |
4004 | } | |
4005 | ||
2115133f CM |
4006 | /* |
4007 | * copy everything in the in-memory inode into the btree. | |
4008 | */ | |
4009 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
4010 | struct btrfs_root *root, struct inode *inode) | |
4011 | { | |
0b246afa | 4012 | struct btrfs_fs_info *fs_info = root->fs_info; |
2115133f CM |
4013 | int ret; |
4014 | ||
4015 | /* | |
4016 | * If the inode is a free space inode, we can deadlock during commit | |
4017 | * if we put it into the delayed code. | |
4018 | * | |
4019 | * The data relocation inode should also be directly updated | |
4020 | * without delay | |
4021 | */ | |
70ddc553 | 4022 | if (!btrfs_is_free_space_inode(BTRFS_I(inode)) |
1d52c78a | 4023 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
0b246afa | 4024 | && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) { |
8ea05e3a AB |
4025 | btrfs_update_root_times(trans, root); |
4026 | ||
2115133f CM |
4027 | ret = btrfs_delayed_update_inode(trans, root, inode); |
4028 | if (!ret) | |
4029 | btrfs_set_inode_last_trans(trans, inode); | |
4030 | return ret; | |
4031 | } | |
4032 | ||
4033 | return btrfs_update_inode_item(trans, root, inode); | |
4034 | } | |
4035 | ||
be6aef60 JB |
4036 | noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
4037 | struct btrfs_root *root, | |
4038 | struct inode *inode) | |
2115133f CM |
4039 | { |
4040 | int ret; | |
4041 | ||
4042 | ret = btrfs_update_inode(trans, root, inode); | |
4043 | if (ret == -ENOSPC) | |
4044 | return btrfs_update_inode_item(trans, root, inode); | |
4045 | return ret; | |
4046 | } | |
4047 | ||
d352ac68 CM |
4048 | /* |
4049 | * unlink helper that gets used here in inode.c and in the tree logging | |
4050 | * recovery code. It remove a link in a directory with a given name, and | |
4051 | * also drops the back refs in the inode to the directory | |
4052 | */ | |
92986796 AV |
4053 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
4054 | struct btrfs_root *root, | |
4ec5934e NB |
4055 | struct btrfs_inode *dir, |
4056 | struct btrfs_inode *inode, | |
92986796 | 4057 | const char *name, int name_len) |
39279cc3 | 4058 | { |
0b246afa | 4059 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 4060 | struct btrfs_path *path; |
39279cc3 | 4061 | int ret = 0; |
5f39d397 | 4062 | struct extent_buffer *leaf; |
39279cc3 | 4063 | struct btrfs_dir_item *di; |
5f39d397 | 4064 | struct btrfs_key key; |
aec7477b | 4065 | u64 index; |
33345d01 LZ |
4066 | u64 ino = btrfs_ino(inode); |
4067 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
4068 | |
4069 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
4070 | if (!path) { |
4071 | ret = -ENOMEM; | |
554233a6 | 4072 | goto out; |
54aa1f4d CM |
4073 | } |
4074 | ||
b9473439 | 4075 | path->leave_spinning = 1; |
33345d01 | 4076 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 CM |
4077 | name, name_len, -1); |
4078 | if (IS_ERR(di)) { | |
4079 | ret = PTR_ERR(di); | |
4080 | goto err; | |
4081 | } | |
4082 | if (!di) { | |
4083 | ret = -ENOENT; | |
4084 | goto err; | |
4085 | } | |
5f39d397 CM |
4086 | leaf = path->nodes[0]; |
4087 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 4088 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
4089 | if (ret) |
4090 | goto err; | |
b3b4aa74 | 4091 | btrfs_release_path(path); |
39279cc3 | 4092 | |
67de1176 MX |
4093 | /* |
4094 | * If we don't have dir index, we have to get it by looking up | |
4095 | * the inode ref, since we get the inode ref, remove it directly, | |
4096 | * it is unnecessary to do delayed deletion. | |
4097 | * | |
4098 | * But if we have dir index, needn't search inode ref to get it. | |
4099 | * Since the inode ref is close to the inode item, it is better | |
4100 | * that we delay to delete it, and just do this deletion when | |
4101 | * we update the inode item. | |
4102 | */ | |
4ec5934e | 4103 | if (inode->dir_index) { |
67de1176 MX |
4104 | ret = btrfs_delayed_delete_inode_ref(inode); |
4105 | if (!ret) { | |
4ec5934e | 4106 | index = inode->dir_index; |
67de1176 MX |
4107 | goto skip_backref; |
4108 | } | |
4109 | } | |
4110 | ||
33345d01 LZ |
4111 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
4112 | dir_ino, &index); | |
aec7477b | 4113 | if (ret) { |
0b246afa | 4114 | btrfs_info(fs_info, |
c2cf52eb | 4115 | "failed to delete reference to %.*s, inode %llu parent %llu", |
c1c9ff7c | 4116 | name_len, name, ino, dir_ino); |
66642832 | 4117 | btrfs_abort_transaction(trans, ret); |
aec7477b JB |
4118 | goto err; |
4119 | } | |
67de1176 | 4120 | skip_backref: |
2ff7e61e | 4121 | ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index); |
79787eaa | 4122 | if (ret) { |
66642832 | 4123 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 4124 | goto err; |
79787eaa | 4125 | } |
39279cc3 | 4126 | |
4ec5934e NB |
4127 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode, |
4128 | dir_ino); | |
79787eaa | 4129 | if (ret != 0 && ret != -ENOENT) { |
66642832 | 4130 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4131 | goto err; |
4132 | } | |
e02119d5 | 4133 | |
4ec5934e NB |
4134 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir, |
4135 | index); | |
6418c961 CM |
4136 | if (ret == -ENOENT) |
4137 | ret = 0; | |
d4e3991b | 4138 | else if (ret) |
66642832 | 4139 | btrfs_abort_transaction(trans, ret); |
39279cc3 CM |
4140 | err: |
4141 | btrfs_free_path(path); | |
e02119d5 CM |
4142 | if (ret) |
4143 | goto out; | |
4144 | ||
6ef06d27 | 4145 | btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2); |
4ec5934e NB |
4146 | inode_inc_iversion(&inode->vfs_inode); |
4147 | inode_inc_iversion(&dir->vfs_inode); | |
4148 | inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime = | |
4149 | dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode); | |
4150 | ret = btrfs_update_inode(trans, root, &dir->vfs_inode); | |
e02119d5 | 4151 | out: |
39279cc3 CM |
4152 | return ret; |
4153 | } | |
4154 | ||
92986796 AV |
4155 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
4156 | struct btrfs_root *root, | |
4ec5934e | 4157 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
92986796 AV |
4158 | const char *name, int name_len) |
4159 | { | |
4160 | int ret; | |
4161 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
4162 | if (!ret) { | |
4ec5934e NB |
4163 | drop_nlink(&inode->vfs_inode); |
4164 | ret = btrfs_update_inode(trans, root, &inode->vfs_inode); | |
92986796 AV |
4165 | } |
4166 | return ret; | |
4167 | } | |
39279cc3 | 4168 | |
a22285a6 YZ |
4169 | /* |
4170 | * helper to start transaction for unlink and rmdir. | |
4171 | * | |
d52be818 JB |
4172 | * unlink and rmdir are special in btrfs, they do not always free space, so |
4173 | * if we cannot make our reservations the normal way try and see if there is | |
4174 | * plenty of slack room in the global reserve to migrate, otherwise we cannot | |
4175 | * allow the unlink to occur. | |
a22285a6 | 4176 | */ |
d52be818 | 4177 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir) |
4df27c4d | 4178 | { |
a22285a6 | 4179 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 4180 | |
e70bea5f JB |
4181 | /* |
4182 | * 1 for the possible orphan item | |
4183 | * 1 for the dir item | |
4184 | * 1 for the dir index | |
4185 | * 1 for the inode ref | |
e70bea5f JB |
4186 | * 1 for the inode |
4187 | */ | |
8eab77ff | 4188 | return btrfs_start_transaction_fallback_global_rsv(root, 5, 5); |
a22285a6 YZ |
4189 | } |
4190 | ||
4191 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
4192 | { | |
4193 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4194 | struct btrfs_trans_handle *trans; | |
2b0143b5 | 4195 | struct inode *inode = d_inode(dentry); |
a22285a6 | 4196 | int ret; |
a22285a6 | 4197 | |
d52be818 | 4198 | trans = __unlink_start_trans(dir); |
a22285a6 YZ |
4199 | if (IS_ERR(trans)) |
4200 | return PTR_ERR(trans); | |
5f39d397 | 4201 | |
4ec5934e NB |
4202 | btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)), |
4203 | 0); | |
12fcfd22 | 4204 | |
4ec5934e NB |
4205 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
4206 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, | |
4207 | dentry->d_name.len); | |
b532402e TI |
4208 | if (ret) |
4209 | goto out; | |
7b128766 | 4210 | |
a22285a6 | 4211 | if (inode->i_nlink == 0) { |
73f2e545 | 4212 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
b532402e TI |
4213 | if (ret) |
4214 | goto out; | |
a22285a6 | 4215 | } |
7b128766 | 4216 | |
b532402e | 4217 | out: |
3a45bb20 | 4218 | btrfs_end_transaction(trans); |
2ff7e61e | 4219 | btrfs_btree_balance_dirty(root->fs_info); |
39279cc3 CM |
4220 | return ret; |
4221 | } | |
4222 | ||
4df27c4d YZ |
4223 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
4224 | struct btrfs_root *root, | |
4225 | struct inode *dir, u64 objectid, | |
4226 | const char *name, int name_len) | |
4227 | { | |
0b246afa | 4228 | struct btrfs_fs_info *fs_info = root->fs_info; |
4df27c4d YZ |
4229 | struct btrfs_path *path; |
4230 | struct extent_buffer *leaf; | |
4231 | struct btrfs_dir_item *di; | |
4232 | struct btrfs_key key; | |
4233 | u64 index; | |
4234 | int ret; | |
4a0cc7ca | 4235 | u64 dir_ino = btrfs_ino(BTRFS_I(dir)); |
4df27c4d YZ |
4236 | |
4237 | path = btrfs_alloc_path(); | |
4238 | if (!path) | |
4239 | return -ENOMEM; | |
4240 | ||
33345d01 | 4241 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 4242 | name, name_len, -1); |
79787eaa JM |
4243 | if (IS_ERR_OR_NULL(di)) { |
4244 | if (!di) | |
4245 | ret = -ENOENT; | |
4246 | else | |
4247 | ret = PTR_ERR(di); | |
4248 | goto out; | |
4249 | } | |
4df27c4d YZ |
4250 | |
4251 | leaf = path->nodes[0]; | |
4252 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
4253 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
4254 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa | 4255 | if (ret) { |
66642832 | 4256 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4257 | goto out; |
4258 | } | |
b3b4aa74 | 4259 | btrfs_release_path(path); |
4df27c4d | 4260 | |
0b246afa JM |
4261 | ret = btrfs_del_root_ref(trans, fs_info, objectid, |
4262 | root->root_key.objectid, dir_ino, | |
4263 | &index, name, name_len); | |
4df27c4d | 4264 | if (ret < 0) { |
79787eaa | 4265 | if (ret != -ENOENT) { |
66642832 | 4266 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4267 | goto out; |
4268 | } | |
33345d01 | 4269 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 4270 | name, name_len); |
79787eaa JM |
4271 | if (IS_ERR_OR_NULL(di)) { |
4272 | if (!di) | |
4273 | ret = -ENOENT; | |
4274 | else | |
4275 | ret = PTR_ERR(di); | |
66642832 | 4276 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4277 | goto out; |
4278 | } | |
4df27c4d YZ |
4279 | |
4280 | leaf = path->nodes[0]; | |
4281 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 4282 | btrfs_release_path(path); |
4df27c4d YZ |
4283 | index = key.offset; |
4284 | } | |
945d8962 | 4285 | btrfs_release_path(path); |
4df27c4d | 4286 | |
e67bbbb9 | 4287 | ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index); |
79787eaa | 4288 | if (ret) { |
66642832 | 4289 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4290 | goto out; |
4291 | } | |
4df27c4d | 4292 | |
6ef06d27 | 4293 | btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2); |
0c4d2d95 | 4294 | inode_inc_iversion(dir); |
c2050a45 | 4295 | dir->i_mtime = dir->i_ctime = current_time(dir); |
5a24e84c | 4296 | ret = btrfs_update_inode_fallback(trans, root, dir); |
79787eaa | 4297 | if (ret) |
66642832 | 4298 | btrfs_abort_transaction(trans, ret); |
79787eaa | 4299 | out: |
71d7aed0 | 4300 | btrfs_free_path(path); |
79787eaa | 4301 | return ret; |
4df27c4d YZ |
4302 | } |
4303 | ||
39279cc3 CM |
4304 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
4305 | { | |
2b0143b5 | 4306 | struct inode *inode = d_inode(dentry); |
1832a6d5 | 4307 | int err = 0; |
39279cc3 | 4308 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 4309 | struct btrfs_trans_handle *trans; |
44f714da | 4310 | u64 last_unlink_trans; |
39279cc3 | 4311 | |
b3ae244e | 4312 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 4313 | return -ENOTEMPTY; |
4a0cc7ca | 4314 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) |
b3ae244e | 4315 | return -EPERM; |
134d4512 | 4316 | |
d52be818 | 4317 | trans = __unlink_start_trans(dir); |
a22285a6 | 4318 | if (IS_ERR(trans)) |
5df6a9f6 | 4319 | return PTR_ERR(trans); |
5df6a9f6 | 4320 | |
4a0cc7ca | 4321 | if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
4df27c4d YZ |
4322 | err = btrfs_unlink_subvol(trans, root, dir, |
4323 | BTRFS_I(inode)->location.objectid, | |
4324 | dentry->d_name.name, | |
4325 | dentry->d_name.len); | |
4326 | goto out; | |
4327 | } | |
4328 | ||
73f2e545 | 4329 | err = btrfs_orphan_add(trans, BTRFS_I(inode)); |
7b128766 | 4330 | if (err) |
4df27c4d | 4331 | goto out; |
7b128766 | 4332 | |
44f714da FM |
4333 | last_unlink_trans = BTRFS_I(inode)->last_unlink_trans; |
4334 | ||
39279cc3 | 4335 | /* now the directory is empty */ |
4ec5934e NB |
4336 | err = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
4337 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, | |
4338 | dentry->d_name.len); | |
44f714da | 4339 | if (!err) { |
6ef06d27 | 4340 | btrfs_i_size_write(BTRFS_I(inode), 0); |
44f714da FM |
4341 | /* |
4342 | * Propagate the last_unlink_trans value of the deleted dir to | |
4343 | * its parent directory. This is to prevent an unrecoverable | |
4344 | * log tree in the case we do something like this: | |
4345 | * 1) create dir foo | |
4346 | * 2) create snapshot under dir foo | |
4347 | * 3) delete the snapshot | |
4348 | * 4) rmdir foo | |
4349 | * 5) mkdir foo | |
4350 | * 6) fsync foo or some file inside foo | |
4351 | */ | |
4352 | if (last_unlink_trans >= trans->transid) | |
4353 | BTRFS_I(dir)->last_unlink_trans = last_unlink_trans; | |
4354 | } | |
4df27c4d | 4355 | out: |
3a45bb20 | 4356 | btrfs_end_transaction(trans); |
2ff7e61e | 4357 | btrfs_btree_balance_dirty(root->fs_info); |
3954401f | 4358 | |
39279cc3 CM |
4359 | return err; |
4360 | } | |
4361 | ||
28f75a0e CM |
4362 | static int truncate_space_check(struct btrfs_trans_handle *trans, |
4363 | struct btrfs_root *root, | |
4364 | u64 bytes_deleted) | |
4365 | { | |
0b246afa | 4366 | struct btrfs_fs_info *fs_info = root->fs_info; |
28f75a0e CM |
4367 | int ret; |
4368 | ||
dc95f7bf JB |
4369 | /* |
4370 | * This is only used to apply pressure to the enospc system, we don't | |
4371 | * intend to use this reservation at all. | |
4372 | */ | |
2ff7e61e | 4373 | bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted); |
0b246afa JM |
4374 | bytes_deleted *= fs_info->nodesize; |
4375 | ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv, | |
28f75a0e | 4376 | bytes_deleted, BTRFS_RESERVE_NO_FLUSH); |
dc95f7bf | 4377 | if (!ret) { |
0b246afa | 4378 | trace_btrfs_space_reservation(fs_info, "transaction", |
dc95f7bf JB |
4379 | trans->transid, |
4380 | bytes_deleted, 1); | |
28f75a0e | 4381 | trans->bytes_reserved += bytes_deleted; |
dc95f7bf | 4382 | } |
28f75a0e CM |
4383 | return ret; |
4384 | ||
4385 | } | |
4386 | ||
ddfae63c JB |
4387 | /* |
4388 | * Return this if we need to call truncate_block for the last bit of the | |
4389 | * truncate. | |
4390 | */ | |
4391 | #define NEED_TRUNCATE_BLOCK 1 | |
0305cd5f | 4392 | |
39279cc3 CM |
4393 | /* |
4394 | * this can truncate away extent items, csum items and directory items. | |
4395 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 4396 | * any higher than new_size |
39279cc3 CM |
4397 | * |
4398 | * csum items that cross the new i_size are truncated to the new size | |
4399 | * as well. | |
7b128766 JB |
4400 | * |
4401 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
4402 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 4403 | */ |
8082510e YZ |
4404 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
4405 | struct btrfs_root *root, | |
4406 | struct inode *inode, | |
4407 | u64 new_size, u32 min_type) | |
39279cc3 | 4408 | { |
0b246afa | 4409 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 4410 | struct btrfs_path *path; |
5f39d397 | 4411 | struct extent_buffer *leaf; |
39279cc3 | 4412 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
4413 | struct btrfs_key key; |
4414 | struct btrfs_key found_key; | |
39279cc3 | 4415 | u64 extent_start = 0; |
db94535d | 4416 | u64 extent_num_bytes = 0; |
5d4f98a2 | 4417 | u64 extent_offset = 0; |
39279cc3 | 4418 | u64 item_end = 0; |
c1aa4575 | 4419 | u64 last_size = new_size; |
8082510e | 4420 | u32 found_type = (u8)-1; |
39279cc3 CM |
4421 | int found_extent; |
4422 | int del_item; | |
85e21bac CM |
4423 | int pending_del_nr = 0; |
4424 | int pending_del_slot = 0; | |
179e29e4 | 4425 | int extent_type = -1; |
8082510e YZ |
4426 | int ret; |
4427 | int err = 0; | |
4a0cc7ca | 4428 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
28ed1345 | 4429 | u64 bytes_deleted = 0; |
897ca819 TM |
4430 | bool be_nice = false; |
4431 | bool should_throttle = false; | |
4432 | bool should_end = false; | |
8082510e YZ |
4433 | |
4434 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 4435 | |
28ed1345 CM |
4436 | /* |
4437 | * for non-free space inodes and ref cows, we want to back off from | |
4438 | * time to time | |
4439 | */ | |
70ddc553 | 4440 | if (!btrfs_is_free_space_inode(BTRFS_I(inode)) && |
28ed1345 | 4441 | test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
897ca819 | 4442 | be_nice = true; |
28ed1345 | 4443 | |
0eb0e19c MF |
4444 | path = btrfs_alloc_path(); |
4445 | if (!path) | |
4446 | return -ENOMEM; | |
e4058b54 | 4447 | path->reada = READA_BACK; |
0eb0e19c | 4448 | |
5dc562c5 JB |
4449 | /* |
4450 | * We want to drop from the next block forward in case this new size is | |
4451 | * not block aligned since we will be keeping the last block of the | |
4452 | * extent just the way it is. | |
4453 | */ | |
27cdeb70 | 4454 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
0b246afa | 4455 | root == fs_info->tree_root) |
dcdbc059 | 4456 | btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size, |
0b246afa | 4457 | fs_info->sectorsize), |
da17066c | 4458 | (u64)-1, 0); |
8082510e | 4459 | |
16cdcec7 MX |
4460 | /* |
4461 | * This function is also used to drop the items in the log tree before | |
4462 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
4463 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
4464 | * items. | |
4465 | */ | |
4466 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
4ccb5c72 | 4467 | btrfs_kill_delayed_inode_items(BTRFS_I(inode)); |
16cdcec7 | 4468 | |
33345d01 | 4469 | key.objectid = ino; |
39279cc3 | 4470 | key.offset = (u64)-1; |
5f39d397 CM |
4471 | key.type = (u8)-1; |
4472 | ||
85e21bac | 4473 | search_again: |
28ed1345 CM |
4474 | /* |
4475 | * with a 16K leaf size and 128MB extents, you can actually queue | |
4476 | * up a huge file in a single leaf. Most of the time that | |
4477 | * bytes_deleted is > 0, it will be huge by the time we get here | |
4478 | */ | |
ee22184b | 4479 | if (be_nice && bytes_deleted > SZ_32M) { |
3a45bb20 | 4480 | if (btrfs_should_end_transaction(trans)) { |
28ed1345 CM |
4481 | err = -EAGAIN; |
4482 | goto error; | |
4483 | } | |
4484 | } | |
4485 | ||
4486 | ||
b9473439 | 4487 | path->leave_spinning = 1; |
85e21bac | 4488 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
4489 | if (ret < 0) { |
4490 | err = ret; | |
4491 | goto out; | |
4492 | } | |
d397712b | 4493 | |
85e21bac | 4494 | if (ret > 0) { |
e02119d5 CM |
4495 | /* there are no items in the tree for us to truncate, we're |
4496 | * done | |
4497 | */ | |
8082510e YZ |
4498 | if (path->slots[0] == 0) |
4499 | goto out; | |
85e21bac CM |
4500 | path->slots[0]--; |
4501 | } | |
4502 | ||
d397712b | 4503 | while (1) { |
39279cc3 | 4504 | fi = NULL; |
5f39d397 CM |
4505 | leaf = path->nodes[0]; |
4506 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
962a298f | 4507 | found_type = found_key.type; |
39279cc3 | 4508 | |
33345d01 | 4509 | if (found_key.objectid != ino) |
39279cc3 | 4510 | break; |
5f39d397 | 4511 | |
85e21bac | 4512 | if (found_type < min_type) |
39279cc3 CM |
4513 | break; |
4514 | ||
5f39d397 | 4515 | item_end = found_key.offset; |
39279cc3 | 4516 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 4517 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 4518 | struct btrfs_file_extent_item); |
179e29e4 CM |
4519 | extent_type = btrfs_file_extent_type(leaf, fi); |
4520 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4521 | item_end += |
db94535d | 4522 | btrfs_file_extent_num_bytes(leaf, fi); |
09ed2f16 LB |
4523 | |
4524 | trace_btrfs_truncate_show_fi_regular( | |
4525 | BTRFS_I(inode), leaf, fi, | |
4526 | found_key.offset); | |
179e29e4 | 4527 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 4528 | item_end += btrfs_file_extent_inline_len(leaf, |
514ac8ad | 4529 | path->slots[0], fi); |
09ed2f16 LB |
4530 | |
4531 | trace_btrfs_truncate_show_fi_inline( | |
4532 | BTRFS_I(inode), leaf, fi, path->slots[0], | |
4533 | found_key.offset); | |
39279cc3 | 4534 | } |
008630c1 | 4535 | item_end--; |
39279cc3 | 4536 | } |
8082510e YZ |
4537 | if (found_type > min_type) { |
4538 | del_item = 1; | |
4539 | } else { | |
76b42abb | 4540 | if (item_end < new_size) |
b888db2b | 4541 | break; |
8082510e YZ |
4542 | if (found_key.offset >= new_size) |
4543 | del_item = 1; | |
4544 | else | |
4545 | del_item = 0; | |
39279cc3 | 4546 | } |
39279cc3 | 4547 | found_extent = 0; |
39279cc3 | 4548 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
4549 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
4550 | goto delete; | |
4551 | ||
4552 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 4553 | u64 num_dec; |
db94535d | 4554 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 4555 | if (!del_item) { |
db94535d CM |
4556 | u64 orig_num_bytes = |
4557 | btrfs_file_extent_num_bytes(leaf, fi); | |
fda2832f QW |
4558 | extent_num_bytes = ALIGN(new_size - |
4559 | found_key.offset, | |
0b246afa | 4560 | fs_info->sectorsize); |
db94535d CM |
4561 | btrfs_set_file_extent_num_bytes(leaf, fi, |
4562 | extent_num_bytes); | |
4563 | num_dec = (orig_num_bytes - | |
9069218d | 4564 | extent_num_bytes); |
27cdeb70 MX |
4565 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4566 | &root->state) && | |
4567 | extent_start != 0) | |
a76a3cd4 | 4568 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 4569 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 4570 | } else { |
db94535d CM |
4571 | extent_num_bytes = |
4572 | btrfs_file_extent_disk_num_bytes(leaf, | |
4573 | fi); | |
5d4f98a2 YZ |
4574 | extent_offset = found_key.offset - |
4575 | btrfs_file_extent_offset(leaf, fi); | |
4576 | ||
39279cc3 | 4577 | /* FIXME blocksize != 4096 */ |
9069218d | 4578 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
4579 | if (extent_start != 0) { |
4580 | found_extent = 1; | |
27cdeb70 MX |
4581 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4582 | &root->state)) | |
a76a3cd4 | 4583 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 4584 | } |
39279cc3 | 4585 | } |
9069218d | 4586 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
4587 | /* |
4588 | * we can't truncate inline items that have had | |
4589 | * special encodings | |
4590 | */ | |
4591 | if (!del_item && | |
c8b97818 | 4592 | btrfs_file_extent_encryption(leaf, fi) == 0 && |
ddfae63c JB |
4593 | btrfs_file_extent_other_encoding(leaf, fi) == 0 && |
4594 | btrfs_file_extent_compression(leaf, fi) == 0) { | |
4595 | u32 size = (u32)(new_size - found_key.offset); | |
4596 | ||
4597 | btrfs_set_file_extent_ram_bytes(leaf, fi, size); | |
4598 | size = btrfs_file_extent_calc_inline_size(size); | |
4599 | btrfs_truncate_item(root->fs_info, path, size, 1); | |
4600 | } else if (!del_item) { | |
514ac8ad | 4601 | /* |
ddfae63c JB |
4602 | * We have to bail so the last_size is set to |
4603 | * just before this extent. | |
514ac8ad | 4604 | */ |
ddfae63c JB |
4605 | err = NEED_TRUNCATE_BLOCK; |
4606 | break; | |
4607 | } | |
0305cd5f | 4608 | |
ddfae63c | 4609 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
0305cd5f | 4610 | inode_sub_bytes(inode, item_end + 1 - new_size); |
39279cc3 | 4611 | } |
179e29e4 | 4612 | delete: |
ddfae63c JB |
4613 | if (del_item) |
4614 | last_size = found_key.offset; | |
4615 | else | |
4616 | last_size = new_size; | |
39279cc3 | 4617 | if (del_item) { |
85e21bac CM |
4618 | if (!pending_del_nr) { |
4619 | /* no pending yet, add ourselves */ | |
4620 | pending_del_slot = path->slots[0]; | |
4621 | pending_del_nr = 1; | |
4622 | } else if (pending_del_nr && | |
4623 | path->slots[0] + 1 == pending_del_slot) { | |
4624 | /* hop on the pending chunk */ | |
4625 | pending_del_nr++; | |
4626 | pending_del_slot = path->slots[0]; | |
4627 | } else { | |
d397712b | 4628 | BUG(); |
85e21bac | 4629 | } |
39279cc3 CM |
4630 | } else { |
4631 | break; | |
4632 | } | |
897ca819 | 4633 | should_throttle = false; |
28f75a0e | 4634 | |
27cdeb70 MX |
4635 | if (found_extent && |
4636 | (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || | |
0b246afa | 4637 | root == fs_info->tree_root)) { |
b9473439 | 4638 | btrfs_set_path_blocking(path); |
28ed1345 | 4639 | bytes_deleted += extent_num_bytes; |
84f7d8e6 | 4640 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
4641 | extent_num_bytes, 0, |
4642 | btrfs_header_owner(leaf), | |
b06c4bf5 | 4643 | ino, extent_offset); |
39279cc3 | 4644 | BUG_ON(ret); |
2ff7e61e JM |
4645 | if (btrfs_should_throttle_delayed_refs(trans, fs_info)) |
4646 | btrfs_async_run_delayed_refs(fs_info, | |
dd4b857a WX |
4647 | trans->delayed_ref_updates * 2, |
4648 | trans->transid, 0); | |
28f75a0e CM |
4649 | if (be_nice) { |
4650 | if (truncate_space_check(trans, root, | |
4651 | extent_num_bytes)) { | |
897ca819 | 4652 | should_end = true; |
28f75a0e CM |
4653 | } |
4654 | if (btrfs_should_throttle_delayed_refs(trans, | |
2ff7e61e | 4655 | fs_info)) |
897ca819 | 4656 | should_throttle = true; |
28f75a0e | 4657 | } |
39279cc3 | 4658 | } |
85e21bac | 4659 | |
8082510e YZ |
4660 | if (found_type == BTRFS_INODE_ITEM_KEY) |
4661 | break; | |
4662 | ||
4663 | if (path->slots[0] == 0 || | |
1262133b | 4664 | path->slots[0] != pending_del_slot || |
28f75a0e | 4665 | should_throttle || should_end) { |
8082510e YZ |
4666 | if (pending_del_nr) { |
4667 | ret = btrfs_del_items(trans, root, path, | |
4668 | pending_del_slot, | |
4669 | pending_del_nr); | |
79787eaa | 4670 | if (ret) { |
66642832 | 4671 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4672 | goto error; |
4673 | } | |
8082510e YZ |
4674 | pending_del_nr = 0; |
4675 | } | |
b3b4aa74 | 4676 | btrfs_release_path(path); |
28f75a0e | 4677 | if (should_throttle) { |
1262133b JB |
4678 | unsigned long updates = trans->delayed_ref_updates; |
4679 | if (updates) { | |
4680 | trans->delayed_ref_updates = 0; | |
2ff7e61e JM |
4681 | ret = btrfs_run_delayed_refs(trans, |
4682 | fs_info, | |
4683 | updates * 2); | |
1262133b JB |
4684 | if (ret && !err) |
4685 | err = ret; | |
4686 | } | |
4687 | } | |
28f75a0e CM |
4688 | /* |
4689 | * if we failed to refill our space rsv, bail out | |
4690 | * and let the transaction restart | |
4691 | */ | |
4692 | if (should_end) { | |
4693 | err = -EAGAIN; | |
4694 | goto error; | |
4695 | } | |
85e21bac | 4696 | goto search_again; |
8082510e YZ |
4697 | } else { |
4698 | path->slots[0]--; | |
85e21bac | 4699 | } |
39279cc3 | 4700 | } |
8082510e | 4701 | out: |
85e21bac CM |
4702 | if (pending_del_nr) { |
4703 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
4704 | pending_del_nr); | |
79787eaa | 4705 | if (ret) |
66642832 | 4706 | btrfs_abort_transaction(trans, ret); |
85e21bac | 4707 | } |
79787eaa | 4708 | error: |
76b42abb FM |
4709 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
4710 | ASSERT(last_size >= new_size); | |
4711 | if (!err && last_size > new_size) | |
4712 | last_size = new_size; | |
7f4f6e0a | 4713 | btrfs_ordered_update_i_size(inode, last_size, NULL); |
76b42abb | 4714 | } |
28ed1345 | 4715 | |
39279cc3 | 4716 | btrfs_free_path(path); |
28ed1345 | 4717 | |
ee22184b | 4718 | if (be_nice && bytes_deleted > SZ_32M) { |
28ed1345 CM |
4719 | unsigned long updates = trans->delayed_ref_updates; |
4720 | if (updates) { | |
4721 | trans->delayed_ref_updates = 0; | |
2ff7e61e JM |
4722 | ret = btrfs_run_delayed_refs(trans, fs_info, |
4723 | updates * 2); | |
28ed1345 CM |
4724 | if (ret && !err) |
4725 | err = ret; | |
4726 | } | |
4727 | } | |
8082510e | 4728 | return err; |
39279cc3 CM |
4729 | } |
4730 | ||
4731 | /* | |
9703fefe | 4732 | * btrfs_truncate_block - read, zero a chunk and write a block |
2aaa6655 JB |
4733 | * @inode - inode that we're zeroing |
4734 | * @from - the offset to start zeroing | |
4735 | * @len - the length to zero, 0 to zero the entire range respective to the | |
4736 | * offset | |
4737 | * @front - zero up to the offset instead of from the offset on | |
4738 | * | |
9703fefe | 4739 | * This will find the block for the "from" offset and cow the block and zero the |
2aaa6655 | 4740 | * part we want to zero. This is used with truncate and hole punching. |
39279cc3 | 4741 | */ |
9703fefe | 4742 | int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len, |
2aaa6655 | 4743 | int front) |
39279cc3 | 4744 | { |
0b246afa | 4745 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2aaa6655 | 4746 | struct address_space *mapping = inode->i_mapping; |
e6dcd2dc CM |
4747 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4748 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 4749 | struct extent_state *cached_state = NULL; |
364ecf36 | 4750 | struct extent_changeset *data_reserved = NULL; |
e6dcd2dc | 4751 | char *kaddr; |
0b246afa | 4752 | u32 blocksize = fs_info->sectorsize; |
09cbfeaf | 4753 | pgoff_t index = from >> PAGE_SHIFT; |
9703fefe | 4754 | unsigned offset = from & (blocksize - 1); |
39279cc3 | 4755 | struct page *page; |
3b16a4e3 | 4756 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 4757 | int ret = 0; |
9703fefe CR |
4758 | u64 block_start; |
4759 | u64 block_end; | |
39279cc3 | 4760 | |
2aaa6655 JB |
4761 | if ((offset & (blocksize - 1)) == 0 && |
4762 | (!len || ((len & (blocksize - 1)) == 0))) | |
39279cc3 | 4763 | goto out; |
9703fefe | 4764 | |
8b62f87b JB |
4765 | block_start = round_down(from, blocksize); |
4766 | block_end = block_start + blocksize - 1; | |
4767 | ||
364ecf36 | 4768 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, |
8b62f87b | 4769 | block_start, blocksize); |
5d5e103a JB |
4770 | if (ret) |
4771 | goto out; | |
39279cc3 | 4772 | |
211c17f5 | 4773 | again: |
3b16a4e3 | 4774 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 4775 | if (!page) { |
bc42bda2 | 4776 | btrfs_delalloc_release_space(inode, data_reserved, |
8b62f87b JB |
4777 | block_start, blocksize); |
4778 | btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize); | |
ac6a2b36 | 4779 | ret = -ENOMEM; |
39279cc3 | 4780 | goto out; |
5d5e103a | 4781 | } |
e6dcd2dc | 4782 | |
39279cc3 | 4783 | if (!PageUptodate(page)) { |
9ebefb18 | 4784 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 4785 | lock_page(page); |
211c17f5 CM |
4786 | if (page->mapping != mapping) { |
4787 | unlock_page(page); | |
09cbfeaf | 4788 | put_page(page); |
211c17f5 CM |
4789 | goto again; |
4790 | } | |
39279cc3 CM |
4791 | if (!PageUptodate(page)) { |
4792 | ret = -EIO; | |
89642229 | 4793 | goto out_unlock; |
39279cc3 CM |
4794 | } |
4795 | } | |
211c17f5 | 4796 | wait_on_page_writeback(page); |
e6dcd2dc | 4797 | |
9703fefe | 4798 | lock_extent_bits(io_tree, block_start, block_end, &cached_state); |
e6dcd2dc CM |
4799 | set_page_extent_mapped(page); |
4800 | ||
9703fefe | 4801 | ordered = btrfs_lookup_ordered_extent(inode, block_start); |
e6dcd2dc | 4802 | if (ordered) { |
9703fefe | 4803 | unlock_extent_cached(io_tree, block_start, block_end, |
2ac55d41 | 4804 | &cached_state, GFP_NOFS); |
e6dcd2dc | 4805 | unlock_page(page); |
09cbfeaf | 4806 | put_page(page); |
eb84ae03 | 4807 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
4808 | btrfs_put_ordered_extent(ordered); |
4809 | goto again; | |
4810 | } | |
4811 | ||
9703fefe | 4812 | clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end, |
9e8a4a8b LB |
4813 | EXTENT_DIRTY | EXTENT_DELALLOC | |
4814 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
ae0f1625 | 4815 | 0, 0, &cached_state); |
5d5e103a | 4816 | |
e3b8a485 | 4817 | ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0, |
ba8b04c1 | 4818 | &cached_state, 0); |
9ed74f2d | 4819 | if (ret) { |
9703fefe | 4820 | unlock_extent_cached(io_tree, block_start, block_end, |
2ac55d41 | 4821 | &cached_state, GFP_NOFS); |
9ed74f2d JB |
4822 | goto out_unlock; |
4823 | } | |
4824 | ||
9703fefe | 4825 | if (offset != blocksize) { |
2aaa6655 | 4826 | if (!len) |
9703fefe | 4827 | len = blocksize - offset; |
e6dcd2dc | 4828 | kaddr = kmap(page); |
2aaa6655 | 4829 | if (front) |
9703fefe CR |
4830 | memset(kaddr + (block_start - page_offset(page)), |
4831 | 0, offset); | |
2aaa6655 | 4832 | else |
9703fefe CR |
4833 | memset(kaddr + (block_start - page_offset(page)) + offset, |
4834 | 0, len); | |
e6dcd2dc CM |
4835 | flush_dcache_page(page); |
4836 | kunmap(page); | |
4837 | } | |
247e743c | 4838 | ClearPageChecked(page); |
e6dcd2dc | 4839 | set_page_dirty(page); |
9703fefe | 4840 | unlock_extent_cached(io_tree, block_start, block_end, &cached_state, |
2ac55d41 | 4841 | GFP_NOFS); |
39279cc3 | 4842 | |
89642229 | 4843 | out_unlock: |
5d5e103a | 4844 | if (ret) |
bc42bda2 | 4845 | btrfs_delalloc_release_space(inode, data_reserved, block_start, |
9703fefe | 4846 | blocksize); |
8b62f87b | 4847 | btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize); |
39279cc3 | 4848 | unlock_page(page); |
09cbfeaf | 4849 | put_page(page); |
39279cc3 | 4850 | out: |
364ecf36 | 4851 | extent_changeset_free(data_reserved); |
39279cc3 CM |
4852 | return ret; |
4853 | } | |
4854 | ||
16e7549f JB |
4855 | static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode, |
4856 | u64 offset, u64 len) | |
4857 | { | |
0b246afa | 4858 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
16e7549f JB |
4859 | struct btrfs_trans_handle *trans; |
4860 | int ret; | |
4861 | ||
4862 | /* | |
4863 | * Still need to make sure the inode looks like it's been updated so | |
4864 | * that any holes get logged if we fsync. | |
4865 | */ | |
0b246afa JM |
4866 | if (btrfs_fs_incompat(fs_info, NO_HOLES)) { |
4867 | BTRFS_I(inode)->last_trans = fs_info->generation; | |
16e7549f JB |
4868 | BTRFS_I(inode)->last_sub_trans = root->log_transid; |
4869 | BTRFS_I(inode)->last_log_commit = root->last_log_commit; | |
4870 | return 0; | |
4871 | } | |
4872 | ||
4873 | /* | |
4874 | * 1 - for the one we're dropping | |
4875 | * 1 - for the one we're adding | |
4876 | * 1 - for updating the inode. | |
4877 | */ | |
4878 | trans = btrfs_start_transaction(root, 3); | |
4879 | if (IS_ERR(trans)) | |
4880 | return PTR_ERR(trans); | |
4881 | ||
4882 | ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1); | |
4883 | if (ret) { | |
66642832 | 4884 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4885 | btrfs_end_transaction(trans); |
16e7549f JB |
4886 | return ret; |
4887 | } | |
4888 | ||
f85b7379 DS |
4889 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)), |
4890 | offset, 0, 0, len, 0, len, 0, 0, 0); | |
16e7549f | 4891 | if (ret) |
66642832 | 4892 | btrfs_abort_transaction(trans, ret); |
16e7549f JB |
4893 | else |
4894 | btrfs_update_inode(trans, root, inode); | |
3a45bb20 | 4895 | btrfs_end_transaction(trans); |
16e7549f JB |
4896 | return ret; |
4897 | } | |
4898 | ||
695a0d0d JB |
4899 | /* |
4900 | * This function puts in dummy file extents for the area we're creating a hole | |
4901 | * for. So if we are truncating this file to a larger size we need to insert | |
4902 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
4903 | * the range between oldsize and size | |
4904 | */ | |
a41ad394 | 4905 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 4906 | { |
0b246afa | 4907 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
9036c102 YZ |
4908 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4909 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 4910 | struct extent_map *em = NULL; |
2ac55d41 | 4911 | struct extent_state *cached_state = NULL; |
5dc562c5 | 4912 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
0b246afa JM |
4913 | u64 hole_start = ALIGN(oldsize, fs_info->sectorsize); |
4914 | u64 block_end = ALIGN(size, fs_info->sectorsize); | |
9036c102 YZ |
4915 | u64 last_byte; |
4916 | u64 cur_offset; | |
4917 | u64 hole_size; | |
9ed74f2d | 4918 | int err = 0; |
39279cc3 | 4919 | |
a71754fc | 4920 | /* |
9703fefe CR |
4921 | * If our size started in the middle of a block we need to zero out the |
4922 | * rest of the block before we expand the i_size, otherwise we could | |
a71754fc JB |
4923 | * expose stale data. |
4924 | */ | |
9703fefe | 4925 | err = btrfs_truncate_block(inode, oldsize, 0, 0); |
a71754fc JB |
4926 | if (err) |
4927 | return err; | |
4928 | ||
9036c102 YZ |
4929 | if (size <= hole_start) |
4930 | return 0; | |
4931 | ||
9036c102 YZ |
4932 | while (1) { |
4933 | struct btrfs_ordered_extent *ordered; | |
fa7c1494 | 4934 | |
ff13db41 | 4935 | lock_extent_bits(io_tree, hole_start, block_end - 1, |
d0082371 | 4936 | &cached_state); |
a776c6fa | 4937 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start, |
fa7c1494 | 4938 | block_end - hole_start); |
9036c102 YZ |
4939 | if (!ordered) |
4940 | break; | |
2ac55d41 JB |
4941 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
4942 | &cached_state, GFP_NOFS); | |
fa7c1494 | 4943 | btrfs_start_ordered_extent(inode, ordered, 1); |
9036c102 YZ |
4944 | btrfs_put_ordered_extent(ordered); |
4945 | } | |
39279cc3 | 4946 | |
9036c102 YZ |
4947 | cur_offset = hole_start; |
4948 | while (1) { | |
fc4f21b1 | 4949 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset, |
9036c102 | 4950 | block_end - cur_offset, 0); |
79787eaa JM |
4951 | if (IS_ERR(em)) { |
4952 | err = PTR_ERR(em); | |
f2767956 | 4953 | em = NULL; |
79787eaa JM |
4954 | break; |
4955 | } | |
9036c102 | 4956 | last_byte = min(extent_map_end(em), block_end); |
0b246afa | 4957 | last_byte = ALIGN(last_byte, fs_info->sectorsize); |
8082510e | 4958 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 4959 | struct extent_map *hole_em; |
9036c102 | 4960 | hole_size = last_byte - cur_offset; |
9ed74f2d | 4961 | |
16e7549f JB |
4962 | err = maybe_insert_hole(root, inode, cur_offset, |
4963 | hole_size); | |
4964 | if (err) | |
3893e33b | 4965 | break; |
dcdbc059 | 4966 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
5dc562c5 JB |
4967 | cur_offset + hole_size - 1, 0); |
4968 | hole_em = alloc_extent_map(); | |
4969 | if (!hole_em) { | |
4970 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
4971 | &BTRFS_I(inode)->runtime_flags); | |
4972 | goto next; | |
4973 | } | |
4974 | hole_em->start = cur_offset; | |
4975 | hole_em->len = hole_size; | |
4976 | hole_em->orig_start = cur_offset; | |
8082510e | 4977 | |
5dc562c5 JB |
4978 | hole_em->block_start = EXTENT_MAP_HOLE; |
4979 | hole_em->block_len = 0; | |
b4939680 | 4980 | hole_em->orig_block_len = 0; |
cc95bef6 | 4981 | hole_em->ram_bytes = hole_size; |
0b246afa | 4982 | hole_em->bdev = fs_info->fs_devices->latest_bdev; |
5dc562c5 | 4983 | hole_em->compress_type = BTRFS_COMPRESS_NONE; |
0b246afa | 4984 | hole_em->generation = fs_info->generation; |
8082510e | 4985 | |
5dc562c5 JB |
4986 | while (1) { |
4987 | write_lock(&em_tree->lock); | |
09a2a8f9 | 4988 | err = add_extent_mapping(em_tree, hole_em, 1); |
5dc562c5 JB |
4989 | write_unlock(&em_tree->lock); |
4990 | if (err != -EEXIST) | |
4991 | break; | |
dcdbc059 NB |
4992 | btrfs_drop_extent_cache(BTRFS_I(inode), |
4993 | cur_offset, | |
5dc562c5 JB |
4994 | cur_offset + |
4995 | hole_size - 1, 0); | |
4996 | } | |
4997 | free_extent_map(hole_em); | |
9036c102 | 4998 | } |
16e7549f | 4999 | next: |
9036c102 | 5000 | free_extent_map(em); |
a22285a6 | 5001 | em = NULL; |
9036c102 | 5002 | cur_offset = last_byte; |
8082510e | 5003 | if (cur_offset >= block_end) |
9036c102 YZ |
5004 | break; |
5005 | } | |
a22285a6 | 5006 | free_extent_map(em); |
2ac55d41 JB |
5007 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
5008 | GFP_NOFS); | |
9036c102 YZ |
5009 | return err; |
5010 | } | |
39279cc3 | 5011 | |
3972f260 | 5012 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 5013 | { |
f4a2f4c5 MX |
5014 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5015 | struct btrfs_trans_handle *trans; | |
a41ad394 | 5016 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
5017 | loff_t newsize = attr->ia_size; |
5018 | int mask = attr->ia_valid; | |
8082510e YZ |
5019 | int ret; |
5020 | ||
3972f260 ES |
5021 | /* |
5022 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
5023 | * special case where we need to update the times despite not having | |
5024 | * these flags set. For all other operations the VFS set these flags | |
5025 | * explicitly if it wants a timestamp update. | |
5026 | */ | |
dff6efc3 CH |
5027 | if (newsize != oldsize) { |
5028 | inode_inc_iversion(inode); | |
5029 | if (!(mask & (ATTR_CTIME | ATTR_MTIME))) | |
5030 | inode->i_ctime = inode->i_mtime = | |
c2050a45 | 5031 | current_time(inode); |
dff6efc3 | 5032 | } |
3972f260 | 5033 | |
a41ad394 | 5034 | if (newsize > oldsize) { |
9ea24bbe | 5035 | /* |
ea14b57f | 5036 | * Don't do an expanding truncate while snapshotting is ongoing. |
9ea24bbe FM |
5037 | * This is to ensure the snapshot captures a fully consistent |
5038 | * state of this file - if the snapshot captures this expanding | |
5039 | * truncation, it must capture all writes that happened before | |
5040 | * this truncation. | |
5041 | */ | |
0bc19f90 | 5042 | btrfs_wait_for_snapshot_creation(root); |
a41ad394 | 5043 | ret = btrfs_cont_expand(inode, oldsize, newsize); |
9ea24bbe | 5044 | if (ret) { |
ea14b57f | 5045 | btrfs_end_write_no_snapshotting(root); |
8082510e | 5046 | return ret; |
9ea24bbe | 5047 | } |
8082510e | 5048 | |
f4a2f4c5 | 5049 | trans = btrfs_start_transaction(root, 1); |
9ea24bbe | 5050 | if (IS_ERR(trans)) { |
ea14b57f | 5051 | btrfs_end_write_no_snapshotting(root); |
f4a2f4c5 | 5052 | return PTR_ERR(trans); |
9ea24bbe | 5053 | } |
f4a2f4c5 MX |
5054 | |
5055 | i_size_write(inode, newsize); | |
5056 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
27772b68 | 5057 | pagecache_isize_extended(inode, oldsize, newsize); |
f4a2f4c5 | 5058 | ret = btrfs_update_inode(trans, root, inode); |
ea14b57f | 5059 | btrfs_end_write_no_snapshotting(root); |
3a45bb20 | 5060 | btrfs_end_transaction(trans); |
a41ad394 | 5061 | } else { |
8082510e | 5062 | |
a41ad394 JB |
5063 | /* |
5064 | * We're truncating a file that used to have good data down to | |
5065 | * zero. Make sure it gets into the ordered flush list so that | |
5066 | * any new writes get down to disk quickly. | |
5067 | */ | |
5068 | if (newsize == 0) | |
72ac3c0d JB |
5069 | set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
5070 | &BTRFS_I(inode)->runtime_flags); | |
8082510e | 5071 | |
f3fe820c JB |
5072 | /* |
5073 | * 1 for the orphan item we're going to add | |
5074 | * 1 for the orphan item deletion. | |
5075 | */ | |
5076 | trans = btrfs_start_transaction(root, 2); | |
5077 | if (IS_ERR(trans)) | |
5078 | return PTR_ERR(trans); | |
5079 | ||
5080 | /* | |
5081 | * We need to do this in case we fail at _any_ point during the | |
5082 | * actual truncate. Once we do the truncate_setsize we could | |
5083 | * invalidate pages which forces any outstanding ordered io to | |
5084 | * be instantly completed which will give us extents that need | |
5085 | * to be truncated. If we fail to get an orphan inode down we | |
5086 | * could have left over extents that were never meant to live, | |
01327610 | 5087 | * so we need to guarantee from this point on that everything |
f3fe820c JB |
5088 | * will be consistent. |
5089 | */ | |
73f2e545 | 5090 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
3a45bb20 | 5091 | btrfs_end_transaction(trans); |
f3fe820c JB |
5092 | if (ret) |
5093 | return ret; | |
5094 | ||
a41ad394 JB |
5095 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
5096 | truncate_setsize(inode, newsize); | |
2e60a51e MX |
5097 | |
5098 | /* Disable nonlocked read DIO to avoid the end less truncate */ | |
abcefb1e | 5099 | btrfs_inode_block_unlocked_dio(BTRFS_I(inode)); |
2e60a51e | 5100 | inode_dio_wait(inode); |
0b581701 | 5101 | btrfs_inode_resume_unlocked_dio(BTRFS_I(inode)); |
2e60a51e | 5102 | |
a41ad394 | 5103 | ret = btrfs_truncate(inode); |
7f4f6e0a JB |
5104 | if (ret && inode->i_nlink) { |
5105 | int err; | |
5106 | ||
19fd2df5 LB |
5107 | /* To get a stable disk_i_size */ |
5108 | err = btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
5109 | if (err) { | |
3d6ae7bb | 5110 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
19fd2df5 LB |
5111 | return err; |
5112 | } | |
5113 | ||
7f4f6e0a JB |
5114 | /* |
5115 | * failed to truncate, disk_i_size is only adjusted down | |
5116 | * as we remove extents, so it should represent the true | |
5117 | * size of the inode, so reset the in memory size and | |
5118 | * delete our orphan entry. | |
5119 | */ | |
5120 | trans = btrfs_join_transaction(root); | |
5121 | if (IS_ERR(trans)) { | |
3d6ae7bb | 5122 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
7f4f6e0a JB |
5123 | return ret; |
5124 | } | |
5125 | i_size_write(inode, BTRFS_I(inode)->disk_i_size); | |
3d6ae7bb | 5126 | err = btrfs_orphan_del(trans, BTRFS_I(inode)); |
7f4f6e0a | 5127 | if (err) |
66642832 | 5128 | btrfs_abort_transaction(trans, err); |
3a45bb20 | 5129 | btrfs_end_transaction(trans); |
7f4f6e0a | 5130 | } |
8082510e YZ |
5131 | } |
5132 | ||
a41ad394 | 5133 | return ret; |
8082510e YZ |
5134 | } |
5135 | ||
9036c102 YZ |
5136 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
5137 | { | |
2b0143b5 | 5138 | struct inode *inode = d_inode(dentry); |
b83cc969 | 5139 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 5140 | int err; |
39279cc3 | 5141 | |
b83cc969 LZ |
5142 | if (btrfs_root_readonly(root)) |
5143 | return -EROFS; | |
5144 | ||
31051c85 | 5145 | err = setattr_prepare(dentry, attr); |
9036c102 YZ |
5146 | if (err) |
5147 | return err; | |
2bf5a725 | 5148 | |
5a3f23d5 | 5149 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 5150 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
5151 | if (err) |
5152 | return err; | |
39279cc3 | 5153 | } |
9036c102 | 5154 | |
1025774c CH |
5155 | if (attr->ia_valid) { |
5156 | setattr_copy(inode, attr); | |
0c4d2d95 | 5157 | inode_inc_iversion(inode); |
22c44fe6 | 5158 | err = btrfs_dirty_inode(inode); |
1025774c | 5159 | |
22c44fe6 | 5160 | if (!err && attr->ia_valid & ATTR_MODE) |
996a710d | 5161 | err = posix_acl_chmod(inode, inode->i_mode); |
1025774c | 5162 | } |
33268eaf | 5163 | |
39279cc3 CM |
5164 | return err; |
5165 | } | |
61295eb8 | 5166 | |
131e404a FDBM |
5167 | /* |
5168 | * While truncating the inode pages during eviction, we get the VFS calling | |
5169 | * btrfs_invalidatepage() against each page of the inode. This is slow because | |
5170 | * the calls to btrfs_invalidatepage() result in a huge amount of calls to | |
5171 | * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting | |
5172 | * extent_state structures over and over, wasting lots of time. | |
5173 | * | |
5174 | * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all | |
5175 | * those expensive operations on a per page basis and do only the ordered io | |
5176 | * finishing, while we release here the extent_map and extent_state structures, | |
5177 | * without the excessive merging and splitting. | |
5178 | */ | |
5179 | static void evict_inode_truncate_pages(struct inode *inode) | |
5180 | { | |
5181 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
5182 | struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree; | |
5183 | struct rb_node *node; | |
5184 | ||
5185 | ASSERT(inode->i_state & I_FREEING); | |
91b0abe3 | 5186 | truncate_inode_pages_final(&inode->i_data); |
131e404a FDBM |
5187 | |
5188 | write_lock(&map_tree->lock); | |
5189 | while (!RB_EMPTY_ROOT(&map_tree->map)) { | |
5190 | struct extent_map *em; | |
5191 | ||
5192 | node = rb_first(&map_tree->map); | |
5193 | em = rb_entry(node, struct extent_map, rb_node); | |
180589ef WS |
5194 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
5195 | clear_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
131e404a FDBM |
5196 | remove_extent_mapping(map_tree, em); |
5197 | free_extent_map(em); | |
7064dd5c FM |
5198 | if (need_resched()) { |
5199 | write_unlock(&map_tree->lock); | |
5200 | cond_resched(); | |
5201 | write_lock(&map_tree->lock); | |
5202 | } | |
131e404a FDBM |
5203 | } |
5204 | write_unlock(&map_tree->lock); | |
5205 | ||
6ca07097 FM |
5206 | /* |
5207 | * Keep looping until we have no more ranges in the io tree. | |
5208 | * We can have ongoing bios started by readpages (called from readahead) | |
9c6429d9 FM |
5209 | * that have their endio callback (extent_io.c:end_bio_extent_readpage) |
5210 | * still in progress (unlocked the pages in the bio but did not yet | |
5211 | * unlocked the ranges in the io tree). Therefore this means some | |
6ca07097 FM |
5212 | * ranges can still be locked and eviction started because before |
5213 | * submitting those bios, which are executed by a separate task (work | |
5214 | * queue kthread), inode references (inode->i_count) were not taken | |
5215 | * (which would be dropped in the end io callback of each bio). | |
5216 | * Therefore here we effectively end up waiting for those bios and | |
5217 | * anyone else holding locked ranges without having bumped the inode's | |
5218 | * reference count - if we don't do it, when they access the inode's | |
5219 | * io_tree to unlock a range it may be too late, leading to an | |
5220 | * use-after-free issue. | |
5221 | */ | |
131e404a FDBM |
5222 | spin_lock(&io_tree->lock); |
5223 | while (!RB_EMPTY_ROOT(&io_tree->state)) { | |
5224 | struct extent_state *state; | |
5225 | struct extent_state *cached_state = NULL; | |
6ca07097 FM |
5226 | u64 start; |
5227 | u64 end; | |
131e404a FDBM |
5228 | |
5229 | node = rb_first(&io_tree->state); | |
5230 | state = rb_entry(node, struct extent_state, rb_node); | |
6ca07097 FM |
5231 | start = state->start; |
5232 | end = state->end; | |
131e404a FDBM |
5233 | spin_unlock(&io_tree->lock); |
5234 | ||
ff13db41 | 5235 | lock_extent_bits(io_tree, start, end, &cached_state); |
b9d0b389 QW |
5236 | |
5237 | /* | |
5238 | * If still has DELALLOC flag, the extent didn't reach disk, | |
5239 | * and its reserved space won't be freed by delayed_ref. | |
5240 | * So we need to free its reserved space here. | |
5241 | * (Refer to comment in btrfs_invalidatepage, case 2) | |
5242 | * | |
5243 | * Note, end is the bytenr of last byte, so we need + 1 here. | |
5244 | */ | |
5245 | if (state->state & EXTENT_DELALLOC) | |
bc42bda2 | 5246 | btrfs_qgroup_free_data(inode, NULL, start, end - start + 1); |
b9d0b389 | 5247 | |
6ca07097 | 5248 | clear_extent_bit(io_tree, start, end, |
131e404a FDBM |
5249 | EXTENT_LOCKED | EXTENT_DIRTY | |
5250 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
ae0f1625 | 5251 | EXTENT_DEFRAG, 1, 1, &cached_state); |
131e404a | 5252 | |
7064dd5c | 5253 | cond_resched(); |
131e404a FDBM |
5254 | spin_lock(&io_tree->lock); |
5255 | } | |
5256 | spin_unlock(&io_tree->lock); | |
5257 | } | |
5258 | ||
bd555975 | 5259 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 | 5260 | { |
0b246afa | 5261 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
5262 | struct btrfs_trans_handle *trans; |
5263 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
726c35fa | 5264 | struct btrfs_block_rsv *rsv, *global_rsv; |
3bce876f | 5265 | int steal_from_global = 0; |
3d48d981 | 5266 | u64 min_size; |
39279cc3 CM |
5267 | int ret; |
5268 | ||
1abe9b8a | 5269 | trace_btrfs_inode_evict(inode); |
5270 | ||
3d48d981 NB |
5271 | if (!root) { |
5272 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); | |
5273 | return; | |
5274 | } | |
5275 | ||
0b246afa | 5276 | min_size = btrfs_calc_trunc_metadata_size(fs_info, 1); |
3d48d981 | 5277 | |
131e404a FDBM |
5278 | evict_inode_truncate_pages(inode); |
5279 | ||
69e9c6c6 SB |
5280 | if (inode->i_nlink && |
5281 | ((btrfs_root_refs(&root->root_item) != 0 && | |
5282 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) || | |
70ddc553 | 5283 | btrfs_is_free_space_inode(BTRFS_I(inode)))) |
bd555975 AV |
5284 | goto no_delete; |
5285 | ||
39279cc3 | 5286 | if (is_bad_inode(inode)) { |
3d6ae7bb | 5287 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
39279cc3 CM |
5288 | goto no_delete; |
5289 | } | |
bd555975 | 5290 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
a30e577c JM |
5291 | if (!special_file(inode->i_mode)) |
5292 | btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
5f39d397 | 5293 | |
7ab7956e | 5294 | btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1); |
f612496b | 5295 | |
0b246afa | 5296 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) { |
6bf02314 | 5297 | BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
8a35d95f | 5298 | &BTRFS_I(inode)->runtime_flags)); |
c71bf099 YZ |
5299 | goto no_delete; |
5300 | } | |
5301 | ||
76dda93c | 5302 | if (inode->i_nlink > 0) { |
69e9c6c6 SB |
5303 | BUG_ON(btrfs_root_refs(&root->root_item) != 0 && |
5304 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID); | |
76dda93c YZ |
5305 | goto no_delete; |
5306 | } | |
5307 | ||
aa79021f | 5308 | ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode)); |
0e8c36a9 | 5309 | if (ret) { |
3d6ae7bb | 5310 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
0e8c36a9 MX |
5311 | goto no_delete; |
5312 | } | |
5313 | ||
2ff7e61e | 5314 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
4289a667 | 5315 | if (!rsv) { |
3d6ae7bb | 5316 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
4289a667 JB |
5317 | goto no_delete; |
5318 | } | |
4a338542 | 5319 | rsv->size = min_size; |
ca7e70f5 | 5320 | rsv->failfast = 1; |
0b246afa | 5321 | global_rsv = &fs_info->global_block_rsv; |
4289a667 | 5322 | |
6ef06d27 | 5323 | btrfs_i_size_write(BTRFS_I(inode), 0); |
5f39d397 | 5324 | |
4289a667 | 5325 | /* |
8407aa46 MX |
5326 | * This is a bit simpler than btrfs_truncate since we've already |
5327 | * reserved our space for our orphan item in the unlink, so we just | |
5328 | * need to reserve some slack space in case we add bytes and update | |
5329 | * inode item when doing the truncate. | |
4289a667 | 5330 | */ |
8082510e | 5331 | while (1) { |
08e007d2 MX |
5332 | ret = btrfs_block_rsv_refill(root, rsv, min_size, |
5333 | BTRFS_RESERVE_FLUSH_LIMIT); | |
726c35fa JB |
5334 | |
5335 | /* | |
5336 | * Try and steal from the global reserve since we will | |
5337 | * likely not use this space anyway, we want to try as | |
5338 | * hard as possible to get this to work. | |
5339 | */ | |
5340 | if (ret) | |
3bce876f JB |
5341 | steal_from_global++; |
5342 | else | |
5343 | steal_from_global = 0; | |
5344 | ret = 0; | |
d68fc57b | 5345 | |
3bce876f JB |
5346 | /* |
5347 | * steal_from_global == 0: we reserved stuff, hooray! | |
5348 | * steal_from_global == 1: we didn't reserve stuff, boo! | |
5349 | * steal_from_global == 2: we've committed, still not a lot of | |
5350 | * room but maybe we'll have room in the global reserve this | |
5351 | * time. | |
5352 | * steal_from_global == 3: abandon all hope! | |
5353 | */ | |
5354 | if (steal_from_global > 2) { | |
0b246afa JM |
5355 | btrfs_warn(fs_info, |
5356 | "Could not get space for a delete, will truncate on mount %d", | |
5357 | ret); | |
3d6ae7bb | 5358 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
2ff7e61e | 5359 | btrfs_free_block_rsv(fs_info, rsv); |
4289a667 | 5360 | goto no_delete; |
d68fc57b | 5361 | } |
7b128766 | 5362 | |
0e8c36a9 | 5363 | trans = btrfs_join_transaction(root); |
4289a667 | 5364 | if (IS_ERR(trans)) { |
3d6ae7bb | 5365 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
2ff7e61e | 5366 | btrfs_free_block_rsv(fs_info, rsv); |
4289a667 | 5367 | goto no_delete; |
d68fc57b | 5368 | } |
7b128766 | 5369 | |
3bce876f | 5370 | /* |
01327610 | 5371 | * We can't just steal from the global reserve, we need to make |
3bce876f JB |
5372 | * sure there is room to do it, if not we need to commit and try |
5373 | * again. | |
5374 | */ | |
5375 | if (steal_from_global) { | |
2ff7e61e | 5376 | if (!btrfs_check_space_for_delayed_refs(trans, fs_info)) |
3bce876f | 5377 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, |
25d609f8 | 5378 | min_size, 0); |
3bce876f JB |
5379 | else |
5380 | ret = -ENOSPC; | |
5381 | } | |
5382 | ||
5383 | /* | |
5384 | * Couldn't steal from the global reserve, we have too much | |
5385 | * pending stuff built up, commit the transaction and try it | |
5386 | * again. | |
5387 | */ | |
5388 | if (ret) { | |
3a45bb20 | 5389 | ret = btrfs_commit_transaction(trans); |
3bce876f | 5390 | if (ret) { |
3d6ae7bb | 5391 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
2ff7e61e | 5392 | btrfs_free_block_rsv(fs_info, rsv); |
3bce876f JB |
5393 | goto no_delete; |
5394 | } | |
5395 | continue; | |
5396 | } else { | |
5397 | steal_from_global = 0; | |
5398 | } | |
5399 | ||
4289a667 JB |
5400 | trans->block_rsv = rsv; |
5401 | ||
d68fc57b | 5402 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
28ed1345 | 5403 | if (ret != -ENOSPC && ret != -EAGAIN) |
8082510e | 5404 | break; |
85e21bac | 5405 | |
0b246afa | 5406 | trans->block_rsv = &fs_info->trans_block_rsv; |
3a45bb20 | 5407 | btrfs_end_transaction(trans); |
8082510e | 5408 | trans = NULL; |
2ff7e61e | 5409 | btrfs_btree_balance_dirty(fs_info); |
8082510e | 5410 | } |
5f39d397 | 5411 | |
2ff7e61e | 5412 | btrfs_free_block_rsv(fs_info, rsv); |
4289a667 | 5413 | |
4ef31a45 JB |
5414 | /* |
5415 | * Errors here aren't a big deal, it just means we leave orphan items | |
5416 | * in the tree. They will be cleaned up on the next mount. | |
5417 | */ | |
8082510e | 5418 | if (ret == 0) { |
4289a667 | 5419 | trans->block_rsv = root->orphan_block_rsv; |
3d6ae7bb | 5420 | btrfs_orphan_del(trans, BTRFS_I(inode)); |
4ef31a45 | 5421 | } else { |
3d6ae7bb | 5422 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
8082510e | 5423 | } |
54aa1f4d | 5424 | |
0b246afa JM |
5425 | trans->block_rsv = &fs_info->trans_block_rsv; |
5426 | if (!(root == fs_info->tree_root || | |
581bb050 | 5427 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) |
4a0cc7ca | 5428 | btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode))); |
581bb050 | 5429 | |
3a45bb20 | 5430 | btrfs_end_transaction(trans); |
2ff7e61e | 5431 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 | 5432 | no_delete: |
f48d1cf5 | 5433 | btrfs_remove_delayed_node(BTRFS_I(inode)); |
dbd5768f | 5434 | clear_inode(inode); |
39279cc3 CM |
5435 | } |
5436 | ||
5437 | /* | |
5438 | * this returns the key found in the dir entry in the location pointer. | |
5439 | * If no dir entries were found, location->objectid is 0. | |
5440 | */ | |
5441 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
5442 | struct btrfs_key *location) | |
5443 | { | |
5444 | const char *name = dentry->d_name.name; | |
5445 | int namelen = dentry->d_name.len; | |
5446 | struct btrfs_dir_item *di; | |
5447 | struct btrfs_path *path; | |
5448 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 5449 | int ret = 0; |
39279cc3 CM |
5450 | |
5451 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
5452 | if (!path) |
5453 | return -ENOMEM; | |
3954401f | 5454 | |
f85b7379 DS |
5455 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)), |
5456 | name, namelen, 0); | |
0d9f7f3e Y |
5457 | if (IS_ERR(di)) |
5458 | ret = PTR_ERR(di); | |
d397712b | 5459 | |
c704005d | 5460 | if (IS_ERR_OR_NULL(di)) |
3954401f | 5461 | goto out_err; |
d397712b | 5462 | |
5f39d397 | 5463 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
56a0e706 LB |
5464 | if (location->type != BTRFS_INODE_ITEM_KEY && |
5465 | location->type != BTRFS_ROOT_ITEM_KEY) { | |
5466 | btrfs_warn(root->fs_info, | |
5467 | "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))", | |
5468 | __func__, name, btrfs_ino(BTRFS_I(dir)), | |
5469 | location->objectid, location->type, location->offset); | |
5470 | goto out_err; | |
5471 | } | |
39279cc3 | 5472 | out: |
39279cc3 CM |
5473 | btrfs_free_path(path); |
5474 | return ret; | |
3954401f CM |
5475 | out_err: |
5476 | location->objectid = 0; | |
5477 | goto out; | |
39279cc3 CM |
5478 | } |
5479 | ||
5480 | /* | |
5481 | * when we hit a tree root in a directory, the btrfs part of the inode | |
5482 | * needs to be changed to reflect the root directory of the tree root. This | |
5483 | * is kind of like crossing a mount point. | |
5484 | */ | |
2ff7e61e | 5485 | static int fixup_tree_root_location(struct btrfs_fs_info *fs_info, |
4df27c4d YZ |
5486 | struct inode *dir, |
5487 | struct dentry *dentry, | |
5488 | struct btrfs_key *location, | |
5489 | struct btrfs_root **sub_root) | |
39279cc3 | 5490 | { |
4df27c4d YZ |
5491 | struct btrfs_path *path; |
5492 | struct btrfs_root *new_root; | |
5493 | struct btrfs_root_ref *ref; | |
5494 | struct extent_buffer *leaf; | |
1d4c08e0 | 5495 | struct btrfs_key key; |
4df27c4d YZ |
5496 | int ret; |
5497 | int err = 0; | |
39279cc3 | 5498 | |
4df27c4d YZ |
5499 | path = btrfs_alloc_path(); |
5500 | if (!path) { | |
5501 | err = -ENOMEM; | |
5502 | goto out; | |
5503 | } | |
39279cc3 | 5504 | |
4df27c4d | 5505 | err = -ENOENT; |
1d4c08e0 DS |
5506 | key.objectid = BTRFS_I(dir)->root->root_key.objectid; |
5507 | key.type = BTRFS_ROOT_REF_KEY; | |
5508 | key.offset = location->objectid; | |
5509 | ||
0b246afa | 5510 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
4df27c4d YZ |
5511 | if (ret) { |
5512 | if (ret < 0) | |
5513 | err = ret; | |
5514 | goto out; | |
5515 | } | |
39279cc3 | 5516 | |
4df27c4d YZ |
5517 | leaf = path->nodes[0]; |
5518 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
4a0cc7ca | 5519 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) || |
4df27c4d YZ |
5520 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
5521 | goto out; | |
39279cc3 | 5522 | |
4df27c4d YZ |
5523 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
5524 | (unsigned long)(ref + 1), | |
5525 | dentry->d_name.len); | |
5526 | if (ret) | |
5527 | goto out; | |
5528 | ||
b3b4aa74 | 5529 | btrfs_release_path(path); |
4df27c4d | 5530 | |
0b246afa | 5531 | new_root = btrfs_read_fs_root_no_name(fs_info, location); |
4df27c4d YZ |
5532 | if (IS_ERR(new_root)) { |
5533 | err = PTR_ERR(new_root); | |
5534 | goto out; | |
5535 | } | |
5536 | ||
4df27c4d YZ |
5537 | *sub_root = new_root; |
5538 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
5539 | location->type = BTRFS_INODE_ITEM_KEY; | |
5540 | location->offset = 0; | |
5541 | err = 0; | |
5542 | out: | |
5543 | btrfs_free_path(path); | |
5544 | return err; | |
39279cc3 CM |
5545 | } |
5546 | ||
5d4f98a2 YZ |
5547 | static void inode_tree_add(struct inode *inode) |
5548 | { | |
5549 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5550 | struct btrfs_inode *entry; | |
03e860bd NP |
5551 | struct rb_node **p; |
5552 | struct rb_node *parent; | |
cef21937 | 5553 | struct rb_node *new = &BTRFS_I(inode)->rb_node; |
4a0cc7ca | 5554 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
5d4f98a2 | 5555 | |
1d3382cb | 5556 | if (inode_unhashed(inode)) |
76dda93c | 5557 | return; |
e1409cef | 5558 | parent = NULL; |
5d4f98a2 | 5559 | spin_lock(&root->inode_lock); |
e1409cef | 5560 | p = &root->inode_tree.rb_node; |
5d4f98a2 YZ |
5561 | while (*p) { |
5562 | parent = *p; | |
5563 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
5564 | ||
4a0cc7ca | 5565 | if (ino < btrfs_ino(BTRFS_I(&entry->vfs_inode))) |
03e860bd | 5566 | p = &parent->rb_left; |
4a0cc7ca | 5567 | else if (ino > btrfs_ino(BTRFS_I(&entry->vfs_inode))) |
03e860bd | 5568 | p = &parent->rb_right; |
5d4f98a2 YZ |
5569 | else { |
5570 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 5571 | (I_WILL_FREE | I_FREEING))); |
cef21937 | 5572 | rb_replace_node(parent, new, &root->inode_tree); |
03e860bd NP |
5573 | RB_CLEAR_NODE(parent); |
5574 | spin_unlock(&root->inode_lock); | |
cef21937 | 5575 | return; |
5d4f98a2 YZ |
5576 | } |
5577 | } | |
cef21937 FDBM |
5578 | rb_link_node(new, parent, p); |
5579 | rb_insert_color(new, &root->inode_tree); | |
5d4f98a2 YZ |
5580 | spin_unlock(&root->inode_lock); |
5581 | } | |
5582 | ||
5583 | static void inode_tree_del(struct inode *inode) | |
5584 | { | |
0b246afa | 5585 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5d4f98a2 | 5586 | struct btrfs_root *root = BTRFS_I(inode)->root; |
76dda93c | 5587 | int empty = 0; |
5d4f98a2 | 5588 | |
03e860bd | 5589 | spin_lock(&root->inode_lock); |
5d4f98a2 | 5590 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 5591 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 5592 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 5593 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 5594 | } |
03e860bd | 5595 | spin_unlock(&root->inode_lock); |
76dda93c | 5596 | |
69e9c6c6 | 5597 | if (empty && btrfs_root_refs(&root->root_item) == 0) { |
0b246afa | 5598 | synchronize_srcu(&fs_info->subvol_srcu); |
76dda93c YZ |
5599 | spin_lock(&root->inode_lock); |
5600 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
5601 | spin_unlock(&root->inode_lock); | |
5602 | if (empty) | |
5603 | btrfs_add_dead_root(root); | |
5604 | } | |
5605 | } | |
5606 | ||
143bede5 | 5607 | void btrfs_invalidate_inodes(struct btrfs_root *root) |
76dda93c | 5608 | { |
0b246afa | 5609 | struct btrfs_fs_info *fs_info = root->fs_info; |
76dda93c YZ |
5610 | struct rb_node *node; |
5611 | struct rb_node *prev; | |
5612 | struct btrfs_inode *entry; | |
5613 | struct inode *inode; | |
5614 | u64 objectid = 0; | |
5615 | ||
0b246afa | 5616 | if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
7813b3db | 5617 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); |
76dda93c YZ |
5618 | |
5619 | spin_lock(&root->inode_lock); | |
5620 | again: | |
5621 | node = root->inode_tree.rb_node; | |
5622 | prev = NULL; | |
5623 | while (node) { | |
5624 | prev = node; | |
5625 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
5626 | ||
4a0cc7ca | 5627 | if (objectid < btrfs_ino(BTRFS_I(&entry->vfs_inode))) |
76dda93c | 5628 | node = node->rb_left; |
4a0cc7ca | 5629 | else if (objectid > btrfs_ino(BTRFS_I(&entry->vfs_inode))) |
76dda93c YZ |
5630 | node = node->rb_right; |
5631 | else | |
5632 | break; | |
5633 | } | |
5634 | if (!node) { | |
5635 | while (prev) { | |
5636 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
4a0cc7ca | 5637 | if (objectid <= btrfs_ino(BTRFS_I(&entry->vfs_inode))) { |
76dda93c YZ |
5638 | node = prev; |
5639 | break; | |
5640 | } | |
5641 | prev = rb_next(prev); | |
5642 | } | |
5643 | } | |
5644 | while (node) { | |
5645 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4a0cc7ca | 5646 | objectid = btrfs_ino(BTRFS_I(&entry->vfs_inode)) + 1; |
76dda93c YZ |
5647 | inode = igrab(&entry->vfs_inode); |
5648 | if (inode) { | |
5649 | spin_unlock(&root->inode_lock); | |
5650 | if (atomic_read(&inode->i_count) > 1) | |
5651 | d_prune_aliases(inode); | |
5652 | /* | |
45321ac5 | 5653 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
5654 | * the inode cache when its usage count |
5655 | * hits zero. | |
5656 | */ | |
5657 | iput(inode); | |
5658 | cond_resched(); | |
5659 | spin_lock(&root->inode_lock); | |
5660 | goto again; | |
5661 | } | |
5662 | ||
5663 | if (cond_resched_lock(&root->inode_lock)) | |
5664 | goto again; | |
5665 | ||
5666 | node = rb_next(node); | |
5667 | } | |
5668 | spin_unlock(&root->inode_lock); | |
5d4f98a2 YZ |
5669 | } |
5670 | ||
e02119d5 CM |
5671 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
5672 | { | |
5673 | struct btrfs_iget_args *args = p; | |
90d3e592 CM |
5674 | inode->i_ino = args->location->objectid; |
5675 | memcpy(&BTRFS_I(inode)->location, args->location, | |
5676 | sizeof(*args->location)); | |
e02119d5 | 5677 | BTRFS_I(inode)->root = args->root; |
39279cc3 CM |
5678 | return 0; |
5679 | } | |
5680 | ||
5681 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
5682 | { | |
5683 | struct btrfs_iget_args *args = opaque; | |
90d3e592 | 5684 | return args->location->objectid == BTRFS_I(inode)->location.objectid && |
d397712b | 5685 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
5686 | } |
5687 | ||
5d4f98a2 | 5688 | static struct inode *btrfs_iget_locked(struct super_block *s, |
90d3e592 | 5689 | struct btrfs_key *location, |
5d4f98a2 | 5690 | struct btrfs_root *root) |
39279cc3 CM |
5691 | { |
5692 | struct inode *inode; | |
5693 | struct btrfs_iget_args args; | |
90d3e592 | 5694 | unsigned long hashval = btrfs_inode_hash(location->objectid, root); |
778ba82b | 5695 | |
90d3e592 | 5696 | args.location = location; |
39279cc3 CM |
5697 | args.root = root; |
5698 | ||
778ba82b | 5699 | inode = iget5_locked(s, hashval, btrfs_find_actor, |
39279cc3 CM |
5700 | btrfs_init_locked_inode, |
5701 | (void *)&args); | |
5702 | return inode; | |
5703 | } | |
5704 | ||
1a54ef8c BR |
5705 | /* Get an inode object given its location and corresponding root. |
5706 | * Returns in *is_new if the inode was read from disk | |
5707 | */ | |
5708 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 5709 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
5710 | { |
5711 | struct inode *inode; | |
5712 | ||
90d3e592 | 5713 | inode = btrfs_iget_locked(s, location, root); |
1a54ef8c | 5714 | if (!inode) |
5d4f98a2 | 5715 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
5716 | |
5717 | if (inode->i_state & I_NEW) { | |
67710892 FM |
5718 | int ret; |
5719 | ||
5720 | ret = btrfs_read_locked_inode(inode); | |
1748f843 MF |
5721 | if (!is_bad_inode(inode)) { |
5722 | inode_tree_add(inode); | |
5723 | unlock_new_inode(inode); | |
5724 | if (new) | |
5725 | *new = 1; | |
5726 | } else { | |
e0b6d65b ST |
5727 | unlock_new_inode(inode); |
5728 | iput(inode); | |
67710892 FM |
5729 | ASSERT(ret < 0); |
5730 | inode = ERR_PTR(ret < 0 ? ret : -ESTALE); | |
1748f843 MF |
5731 | } |
5732 | } | |
5733 | ||
1a54ef8c BR |
5734 | return inode; |
5735 | } | |
5736 | ||
4df27c4d YZ |
5737 | static struct inode *new_simple_dir(struct super_block *s, |
5738 | struct btrfs_key *key, | |
5739 | struct btrfs_root *root) | |
5740 | { | |
5741 | struct inode *inode = new_inode(s); | |
5742 | ||
5743 | if (!inode) | |
5744 | return ERR_PTR(-ENOMEM); | |
5745 | ||
4df27c4d YZ |
5746 | BTRFS_I(inode)->root = root; |
5747 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
72ac3c0d | 5748 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
5749 | |
5750 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
848cce0d | 5751 | inode->i_op = &btrfs_dir_ro_inode_operations; |
1fdf4194 | 5752 | inode->i_opflags &= ~IOP_XATTR; |
4df27c4d YZ |
5753 | inode->i_fop = &simple_dir_operations; |
5754 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
c2050a45 | 5755 | inode->i_mtime = current_time(inode); |
9cc97d64 | 5756 | inode->i_atime = inode->i_mtime; |
5757 | inode->i_ctime = inode->i_mtime; | |
5758 | BTRFS_I(inode)->i_otime = inode->i_mtime; | |
4df27c4d YZ |
5759 | |
5760 | return inode; | |
5761 | } | |
5762 | ||
3de4586c | 5763 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 5764 | { |
0b246afa | 5765 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
d397712b | 5766 | struct inode *inode; |
4df27c4d | 5767 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
5768 | struct btrfs_root *sub_root = root; |
5769 | struct btrfs_key location; | |
76dda93c | 5770 | int index; |
b4aff1f8 | 5771 | int ret = 0; |
39279cc3 CM |
5772 | |
5773 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
5774 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 5775 | |
39e3c955 | 5776 | ret = btrfs_inode_by_name(dir, dentry, &location); |
39279cc3 CM |
5777 | if (ret < 0) |
5778 | return ERR_PTR(ret); | |
5f39d397 | 5779 | |
4df27c4d | 5780 | if (location.objectid == 0) |
5662344b | 5781 | return ERR_PTR(-ENOENT); |
4df27c4d YZ |
5782 | |
5783 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 5784 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
5785 | return inode; |
5786 | } | |
5787 | ||
0b246afa | 5788 | index = srcu_read_lock(&fs_info->subvol_srcu); |
2ff7e61e | 5789 | ret = fixup_tree_root_location(fs_info, dir, dentry, |
4df27c4d YZ |
5790 | &location, &sub_root); |
5791 | if (ret < 0) { | |
5792 | if (ret != -ENOENT) | |
5793 | inode = ERR_PTR(ret); | |
5794 | else | |
5795 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
5796 | } else { | |
73f73415 | 5797 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 5798 | } |
0b246afa | 5799 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
76dda93c | 5800 | |
34d19bad | 5801 | if (!IS_ERR(inode) && root != sub_root) { |
0b246afa | 5802 | down_read(&fs_info->cleanup_work_sem); |
bc98a42c | 5803 | if (!sb_rdonly(inode->i_sb)) |
66b4ffd1 | 5804 | ret = btrfs_orphan_cleanup(sub_root); |
0b246afa | 5805 | up_read(&fs_info->cleanup_work_sem); |
01cd3367 JB |
5806 | if (ret) { |
5807 | iput(inode); | |
66b4ffd1 | 5808 | inode = ERR_PTR(ret); |
01cd3367 | 5809 | } |
c71bf099 YZ |
5810 | } |
5811 | ||
3de4586c CM |
5812 | return inode; |
5813 | } | |
5814 | ||
fe15ce44 | 5815 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
5816 | { |
5817 | struct btrfs_root *root; | |
2b0143b5 | 5818 | struct inode *inode = d_inode(dentry); |
76dda93c | 5819 | |
848cce0d | 5820 | if (!inode && !IS_ROOT(dentry)) |
2b0143b5 | 5821 | inode = d_inode(dentry->d_parent); |
76dda93c | 5822 | |
848cce0d LZ |
5823 | if (inode) { |
5824 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
5825 | if (btrfs_root_refs(&root->root_item) == 0) |
5826 | return 1; | |
848cce0d | 5827 | |
4a0cc7ca | 5828 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
848cce0d | 5829 | return 1; |
efefb143 | 5830 | } |
76dda93c YZ |
5831 | return 0; |
5832 | } | |
5833 | ||
b4aff1f8 JB |
5834 | static void btrfs_dentry_release(struct dentry *dentry) |
5835 | { | |
944a4515 | 5836 | kfree(dentry->d_fsdata); |
b4aff1f8 JB |
5837 | } |
5838 | ||
3de4586c | 5839 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 5840 | unsigned int flags) |
3de4586c | 5841 | { |
5662344b | 5842 | struct inode *inode; |
a66e7cc6 | 5843 | |
5662344b TI |
5844 | inode = btrfs_lookup_dentry(dir, dentry); |
5845 | if (IS_ERR(inode)) { | |
5846 | if (PTR_ERR(inode) == -ENOENT) | |
5847 | inode = NULL; | |
5848 | else | |
5849 | return ERR_CAST(inode); | |
5850 | } | |
5851 | ||
41d28bca | 5852 | return d_splice_alias(inode, dentry); |
39279cc3 CM |
5853 | } |
5854 | ||
16cdcec7 | 5855 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
5856 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
5857 | }; | |
5858 | ||
23b5ec74 JB |
5859 | /* |
5860 | * All this infrastructure exists because dir_emit can fault, and we are holding | |
5861 | * the tree lock when doing readdir. For now just allocate a buffer and copy | |
5862 | * our information into that, and then dir_emit from the buffer. This is | |
5863 | * similar to what NFS does, only we don't keep the buffer around in pagecache | |
5864 | * because I'm afraid I'll mess that up. Long term we need to make filldir do | |
5865 | * copy_to_user_inatomic so we don't have to worry about page faulting under the | |
5866 | * tree lock. | |
5867 | */ | |
5868 | static int btrfs_opendir(struct inode *inode, struct file *file) | |
5869 | { | |
5870 | struct btrfs_file_private *private; | |
5871 | ||
5872 | private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL); | |
5873 | if (!private) | |
5874 | return -ENOMEM; | |
5875 | private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
5876 | if (!private->filldir_buf) { | |
5877 | kfree(private); | |
5878 | return -ENOMEM; | |
5879 | } | |
5880 | file->private_data = private; | |
5881 | return 0; | |
5882 | } | |
5883 | ||
5884 | struct dir_entry { | |
5885 | u64 ino; | |
5886 | u64 offset; | |
5887 | unsigned type; | |
5888 | int name_len; | |
5889 | }; | |
5890 | ||
5891 | static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx) | |
5892 | { | |
5893 | while (entries--) { | |
5894 | struct dir_entry *entry = addr; | |
5895 | char *name = (char *)(entry + 1); | |
5896 | ||
5897 | ctx->pos = entry->offset; | |
5898 | if (!dir_emit(ctx, name, entry->name_len, entry->ino, | |
5899 | entry->type)) | |
5900 | return 1; | |
5901 | addr += sizeof(struct dir_entry) + entry->name_len; | |
5902 | ctx->pos++; | |
5903 | } | |
5904 | return 0; | |
5905 | } | |
5906 | ||
9cdda8d3 | 5907 | static int btrfs_real_readdir(struct file *file, struct dir_context *ctx) |
39279cc3 | 5908 | { |
9cdda8d3 | 5909 | struct inode *inode = file_inode(file); |
2ff7e61e | 5910 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 5911 | struct btrfs_root *root = BTRFS_I(inode)->root; |
23b5ec74 | 5912 | struct btrfs_file_private *private = file->private_data; |
39279cc3 CM |
5913 | struct btrfs_dir_item *di; |
5914 | struct btrfs_key key; | |
5f39d397 | 5915 | struct btrfs_key found_key; |
39279cc3 | 5916 | struct btrfs_path *path; |
23b5ec74 | 5917 | void *addr; |
16cdcec7 MX |
5918 | struct list_head ins_list; |
5919 | struct list_head del_list; | |
39279cc3 | 5920 | int ret; |
5f39d397 | 5921 | struct extent_buffer *leaf; |
39279cc3 | 5922 | int slot; |
5f39d397 CM |
5923 | char *name_ptr; |
5924 | int name_len; | |
23b5ec74 JB |
5925 | int entries = 0; |
5926 | int total_len = 0; | |
02dbfc99 | 5927 | bool put = false; |
c2951f32 | 5928 | struct btrfs_key location; |
5f39d397 | 5929 | |
9cdda8d3 AV |
5930 | if (!dir_emit_dots(file, ctx)) |
5931 | return 0; | |
5932 | ||
49593bfa | 5933 | path = btrfs_alloc_path(); |
16cdcec7 MX |
5934 | if (!path) |
5935 | return -ENOMEM; | |
ff5714cc | 5936 | |
23b5ec74 | 5937 | addr = private->filldir_buf; |
e4058b54 | 5938 | path->reada = READA_FORWARD; |
49593bfa | 5939 | |
c2951f32 JM |
5940 | INIT_LIST_HEAD(&ins_list); |
5941 | INIT_LIST_HEAD(&del_list); | |
5942 | put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list); | |
16cdcec7 | 5943 | |
23b5ec74 | 5944 | again: |
c2951f32 | 5945 | key.type = BTRFS_DIR_INDEX_KEY; |
9cdda8d3 | 5946 | key.offset = ctx->pos; |
4a0cc7ca | 5947 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
5f39d397 | 5948 | |
39279cc3 CM |
5949 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5950 | if (ret < 0) | |
5951 | goto err; | |
49593bfa DW |
5952 | |
5953 | while (1) { | |
23b5ec74 JB |
5954 | struct dir_entry *entry; |
5955 | ||
5f39d397 | 5956 | leaf = path->nodes[0]; |
39279cc3 | 5957 | slot = path->slots[0]; |
b9e03af0 LZ |
5958 | if (slot >= btrfs_header_nritems(leaf)) { |
5959 | ret = btrfs_next_leaf(root, path); | |
5960 | if (ret < 0) | |
5961 | goto err; | |
5962 | else if (ret > 0) | |
5963 | break; | |
5964 | continue; | |
39279cc3 | 5965 | } |
3de4586c | 5966 | |
5f39d397 CM |
5967 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
5968 | ||
5969 | if (found_key.objectid != key.objectid) | |
39279cc3 | 5970 | break; |
c2951f32 | 5971 | if (found_key.type != BTRFS_DIR_INDEX_KEY) |
39279cc3 | 5972 | break; |
9cdda8d3 | 5973 | if (found_key.offset < ctx->pos) |
b9e03af0 | 5974 | goto next; |
c2951f32 | 5975 | if (btrfs_should_delete_dir_index(&del_list, found_key.offset)) |
16cdcec7 | 5976 | goto next; |
39279cc3 | 5977 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
e79a3327 | 5978 | if (verify_dir_item(fs_info, leaf, slot, di)) |
c2951f32 | 5979 | goto next; |
22a94d44 | 5980 | |
c2951f32 | 5981 | name_len = btrfs_dir_name_len(leaf, di); |
23b5ec74 JB |
5982 | if ((total_len + sizeof(struct dir_entry) + name_len) >= |
5983 | PAGE_SIZE) { | |
5984 | btrfs_release_path(path); | |
5985 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
5986 | if (ret) | |
5987 | goto nopos; | |
5988 | addr = private->filldir_buf; | |
5989 | entries = 0; | |
5990 | total_len = 0; | |
5991 | goto again; | |
c2951f32 | 5992 | } |
23b5ec74 JB |
5993 | |
5994 | entry = addr; | |
5995 | entry->name_len = name_len; | |
5996 | name_ptr = (char *)(entry + 1); | |
c2951f32 JM |
5997 | read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1), |
5998 | name_len); | |
23b5ec74 | 5999 | entry->type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; |
c2951f32 | 6000 | btrfs_dir_item_key_to_cpu(leaf, di, &location); |
23b5ec74 JB |
6001 | entry->ino = location.objectid; |
6002 | entry->offset = found_key.offset; | |
6003 | entries++; | |
6004 | addr += sizeof(struct dir_entry) + name_len; | |
6005 | total_len += sizeof(struct dir_entry) + name_len; | |
b9e03af0 LZ |
6006 | next: |
6007 | path->slots[0]++; | |
39279cc3 | 6008 | } |
23b5ec74 JB |
6009 | btrfs_release_path(path); |
6010 | ||
6011 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
6012 | if (ret) | |
6013 | goto nopos; | |
49593bfa | 6014 | |
d2fbb2b5 | 6015 | ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list); |
c2951f32 | 6016 | if (ret) |
bc4ef759 DS |
6017 | goto nopos; |
6018 | ||
db62efbb ZB |
6019 | /* |
6020 | * Stop new entries from being returned after we return the last | |
6021 | * entry. | |
6022 | * | |
6023 | * New directory entries are assigned a strictly increasing | |
6024 | * offset. This means that new entries created during readdir | |
6025 | * are *guaranteed* to be seen in the future by that readdir. | |
6026 | * This has broken buggy programs which operate on names as | |
6027 | * they're returned by readdir. Until we re-use freed offsets | |
6028 | * we have this hack to stop new entries from being returned | |
6029 | * under the assumption that they'll never reach this huge | |
6030 | * offset. | |
6031 | * | |
6032 | * This is being careful not to overflow 32bit loff_t unless the | |
6033 | * last entry requires it because doing so has broken 32bit apps | |
6034 | * in the past. | |
6035 | */ | |
c2951f32 JM |
6036 | if (ctx->pos >= INT_MAX) |
6037 | ctx->pos = LLONG_MAX; | |
6038 | else | |
6039 | ctx->pos = INT_MAX; | |
39279cc3 CM |
6040 | nopos: |
6041 | ret = 0; | |
6042 | err: | |
02dbfc99 OS |
6043 | if (put) |
6044 | btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list); | |
39279cc3 | 6045 | btrfs_free_path(path); |
39279cc3 CM |
6046 | return ret; |
6047 | } | |
6048 | ||
a9185b41 | 6049 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
6050 | { |
6051 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6052 | struct btrfs_trans_handle *trans; | |
6053 | int ret = 0; | |
0af3d00b | 6054 | bool nolock = false; |
39279cc3 | 6055 | |
72ac3c0d | 6056 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
4ca8b41e CM |
6057 | return 0; |
6058 | ||
70ddc553 NB |
6059 | if (btrfs_fs_closing(root->fs_info) && |
6060 | btrfs_is_free_space_inode(BTRFS_I(inode))) | |
82d5902d | 6061 | nolock = true; |
0af3d00b | 6062 | |
a9185b41 | 6063 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b | 6064 | if (nolock) |
7a7eaa40 | 6065 | trans = btrfs_join_transaction_nolock(root); |
0af3d00b | 6066 | else |
7a7eaa40 | 6067 | trans = btrfs_join_transaction(root); |
3612b495 TI |
6068 | if (IS_ERR(trans)) |
6069 | return PTR_ERR(trans); | |
3a45bb20 | 6070 | ret = btrfs_commit_transaction(trans); |
39279cc3 CM |
6071 | } |
6072 | return ret; | |
6073 | } | |
6074 | ||
6075 | /* | |
54aa1f4d | 6076 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
6077 | * inode changes. But, it is most likely to find the inode in cache. |
6078 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
6079 | * to keep or drop this code. | |
6080 | */ | |
48a3b636 | 6081 | static int btrfs_dirty_inode(struct inode *inode) |
39279cc3 | 6082 | { |
2ff7e61e | 6083 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
6084 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6085 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
6086 | int ret; |
6087 | ||
72ac3c0d | 6088 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 6089 | return 0; |
39279cc3 | 6090 | |
7a7eaa40 | 6091 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
6092 | if (IS_ERR(trans)) |
6093 | return PTR_ERR(trans); | |
8929ecfa YZ |
6094 | |
6095 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
6096 | if (ret && ret == -ENOSPC) { |
6097 | /* whoops, lets try again with the full transaction */ | |
3a45bb20 | 6098 | btrfs_end_transaction(trans); |
94b60442 | 6099 | trans = btrfs_start_transaction(root, 1); |
22c44fe6 JB |
6100 | if (IS_ERR(trans)) |
6101 | return PTR_ERR(trans); | |
8929ecfa | 6102 | |
94b60442 | 6103 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 6104 | } |
3a45bb20 | 6105 | btrfs_end_transaction(trans); |
16cdcec7 | 6106 | if (BTRFS_I(inode)->delayed_node) |
2ff7e61e | 6107 | btrfs_balance_delayed_items(fs_info); |
22c44fe6 JB |
6108 | |
6109 | return ret; | |
6110 | } | |
6111 | ||
6112 | /* | |
6113 | * This is a copy of file_update_time. We need this so we can return error on | |
6114 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
6115 | */ | |
e41f941a JB |
6116 | static int btrfs_update_time(struct inode *inode, struct timespec *now, |
6117 | int flags) | |
22c44fe6 | 6118 | { |
2bc55652 AB |
6119 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6120 | ||
6121 | if (btrfs_root_readonly(root)) | |
6122 | return -EROFS; | |
6123 | ||
e41f941a | 6124 | if (flags & S_VERSION) |
22c44fe6 | 6125 | inode_inc_iversion(inode); |
e41f941a JB |
6126 | if (flags & S_CTIME) |
6127 | inode->i_ctime = *now; | |
6128 | if (flags & S_MTIME) | |
6129 | inode->i_mtime = *now; | |
6130 | if (flags & S_ATIME) | |
6131 | inode->i_atime = *now; | |
6132 | return btrfs_dirty_inode(inode); | |
39279cc3 CM |
6133 | } |
6134 | ||
d352ac68 CM |
6135 | /* |
6136 | * find the highest existing sequence number in a directory | |
6137 | * and then set the in-memory index_cnt variable to reflect | |
6138 | * free sequence numbers | |
6139 | */ | |
4c570655 | 6140 | static int btrfs_set_inode_index_count(struct btrfs_inode *inode) |
aec7477b | 6141 | { |
4c570655 | 6142 | struct btrfs_root *root = inode->root; |
aec7477b JB |
6143 | struct btrfs_key key, found_key; |
6144 | struct btrfs_path *path; | |
6145 | struct extent_buffer *leaf; | |
6146 | int ret; | |
6147 | ||
4c570655 | 6148 | key.objectid = btrfs_ino(inode); |
962a298f | 6149 | key.type = BTRFS_DIR_INDEX_KEY; |
aec7477b JB |
6150 | key.offset = (u64)-1; |
6151 | ||
6152 | path = btrfs_alloc_path(); | |
6153 | if (!path) | |
6154 | return -ENOMEM; | |
6155 | ||
6156 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
6157 | if (ret < 0) | |
6158 | goto out; | |
6159 | /* FIXME: we should be able to handle this */ | |
6160 | if (ret == 0) | |
6161 | goto out; | |
6162 | ret = 0; | |
6163 | ||
6164 | /* | |
6165 | * MAGIC NUMBER EXPLANATION: | |
6166 | * since we search a directory based on f_pos we have to start at 2 | |
6167 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
6168 | * else has to start at 2 | |
6169 | */ | |
6170 | if (path->slots[0] == 0) { | |
4c570655 | 6171 | inode->index_cnt = 2; |
aec7477b JB |
6172 | goto out; |
6173 | } | |
6174 | ||
6175 | path->slots[0]--; | |
6176 | ||
6177 | leaf = path->nodes[0]; | |
6178 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
6179 | ||
4c570655 | 6180 | if (found_key.objectid != btrfs_ino(inode) || |
962a298f | 6181 | found_key.type != BTRFS_DIR_INDEX_KEY) { |
4c570655 | 6182 | inode->index_cnt = 2; |
aec7477b JB |
6183 | goto out; |
6184 | } | |
6185 | ||
4c570655 | 6186 | inode->index_cnt = found_key.offset + 1; |
aec7477b JB |
6187 | out: |
6188 | btrfs_free_path(path); | |
6189 | return ret; | |
6190 | } | |
6191 | ||
d352ac68 CM |
6192 | /* |
6193 | * helper to find a free sequence number in a given directory. This current | |
6194 | * code is very simple, later versions will do smarter things in the btree | |
6195 | */ | |
877574e2 | 6196 | int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index) |
aec7477b JB |
6197 | { |
6198 | int ret = 0; | |
6199 | ||
877574e2 NB |
6200 | if (dir->index_cnt == (u64)-1) { |
6201 | ret = btrfs_inode_delayed_dir_index_count(dir); | |
16cdcec7 MX |
6202 | if (ret) { |
6203 | ret = btrfs_set_inode_index_count(dir); | |
6204 | if (ret) | |
6205 | return ret; | |
6206 | } | |
aec7477b JB |
6207 | } |
6208 | ||
877574e2 NB |
6209 | *index = dir->index_cnt; |
6210 | dir->index_cnt++; | |
aec7477b JB |
6211 | |
6212 | return ret; | |
6213 | } | |
6214 | ||
b0d5d10f CM |
6215 | static int btrfs_insert_inode_locked(struct inode *inode) |
6216 | { | |
6217 | struct btrfs_iget_args args; | |
6218 | args.location = &BTRFS_I(inode)->location; | |
6219 | args.root = BTRFS_I(inode)->root; | |
6220 | ||
6221 | return insert_inode_locked4(inode, | |
6222 | btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root), | |
6223 | btrfs_find_actor, &args); | |
6224 | } | |
6225 | ||
19aee8de AJ |
6226 | /* |
6227 | * Inherit flags from the parent inode. | |
6228 | * | |
6229 | * Currently only the compression flags and the cow flags are inherited. | |
6230 | */ | |
6231 | static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir) | |
6232 | { | |
6233 | unsigned int flags; | |
6234 | ||
6235 | if (!dir) | |
6236 | return; | |
6237 | ||
6238 | flags = BTRFS_I(dir)->flags; | |
6239 | ||
6240 | if (flags & BTRFS_INODE_NOCOMPRESS) { | |
6241 | BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS; | |
6242 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; | |
6243 | } else if (flags & BTRFS_INODE_COMPRESS) { | |
6244 | BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS; | |
6245 | BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS; | |
6246 | } | |
6247 | ||
6248 | if (flags & BTRFS_INODE_NODATACOW) { | |
6249 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; | |
6250 | if (S_ISREG(inode->i_mode)) | |
6251 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
6252 | } | |
6253 | ||
6254 | btrfs_update_iflags(inode); | |
6255 | } | |
6256 | ||
39279cc3 CM |
6257 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
6258 | struct btrfs_root *root, | |
aec7477b | 6259 | struct inode *dir, |
9c58309d | 6260 | const char *name, int name_len, |
175a4eb7 AV |
6261 | u64 ref_objectid, u64 objectid, |
6262 | umode_t mode, u64 *index) | |
39279cc3 | 6263 | { |
0b246afa | 6264 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 6265 | struct inode *inode; |
5f39d397 | 6266 | struct btrfs_inode_item *inode_item; |
39279cc3 | 6267 | struct btrfs_key *location; |
5f39d397 | 6268 | struct btrfs_path *path; |
9c58309d CM |
6269 | struct btrfs_inode_ref *ref; |
6270 | struct btrfs_key key[2]; | |
6271 | u32 sizes[2]; | |
ef3b9af5 | 6272 | int nitems = name ? 2 : 1; |
9c58309d | 6273 | unsigned long ptr; |
39279cc3 | 6274 | int ret; |
39279cc3 | 6275 | |
5f39d397 | 6276 | path = btrfs_alloc_path(); |
d8926bb3 MF |
6277 | if (!path) |
6278 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 6279 | |
0b246afa | 6280 | inode = new_inode(fs_info->sb); |
8fb27640 YS |
6281 | if (!inode) { |
6282 | btrfs_free_path(path); | |
39279cc3 | 6283 | return ERR_PTR(-ENOMEM); |
8fb27640 | 6284 | } |
39279cc3 | 6285 | |
5762b5c9 FM |
6286 | /* |
6287 | * O_TMPFILE, set link count to 0, so that after this point, | |
6288 | * we fill in an inode item with the correct link count. | |
6289 | */ | |
6290 | if (!name) | |
6291 | set_nlink(inode, 0); | |
6292 | ||
581bb050 LZ |
6293 | /* |
6294 | * we have to initialize this early, so we can reclaim the inode | |
6295 | * number if we fail afterwards in this function. | |
6296 | */ | |
6297 | inode->i_ino = objectid; | |
6298 | ||
ef3b9af5 | 6299 | if (dir && name) { |
1abe9b8a | 6300 | trace_btrfs_inode_request(dir); |
6301 | ||
877574e2 | 6302 | ret = btrfs_set_inode_index(BTRFS_I(dir), index); |
09771430 | 6303 | if (ret) { |
8fb27640 | 6304 | btrfs_free_path(path); |
09771430 | 6305 | iput(inode); |
aec7477b | 6306 | return ERR_PTR(ret); |
09771430 | 6307 | } |
ef3b9af5 FM |
6308 | } else if (dir) { |
6309 | *index = 0; | |
aec7477b JB |
6310 | } |
6311 | /* | |
6312 | * index_cnt is ignored for everything but a dir, | |
6313 | * btrfs_get_inode_index_count has an explanation for the magic | |
6314 | * number | |
6315 | */ | |
6316 | BTRFS_I(inode)->index_cnt = 2; | |
67de1176 | 6317 | BTRFS_I(inode)->dir_index = *index; |
39279cc3 | 6318 | BTRFS_I(inode)->root = root; |
e02119d5 | 6319 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 6320 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 6321 | |
5dc562c5 JB |
6322 | /* |
6323 | * We could have gotten an inode number from somebody who was fsynced | |
6324 | * and then removed in this same transaction, so let's just set full | |
6325 | * sync since it will be a full sync anyway and this will blow away the | |
6326 | * old info in the log. | |
6327 | */ | |
6328 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
6329 | ||
9c58309d | 6330 | key[0].objectid = objectid; |
962a298f | 6331 | key[0].type = BTRFS_INODE_ITEM_KEY; |
9c58309d CM |
6332 | key[0].offset = 0; |
6333 | ||
9c58309d | 6334 | sizes[0] = sizeof(struct btrfs_inode_item); |
ef3b9af5 FM |
6335 | |
6336 | if (name) { | |
6337 | /* | |
6338 | * Start new inodes with an inode_ref. This is slightly more | |
6339 | * efficient for small numbers of hard links since they will | |
6340 | * be packed into one item. Extended refs will kick in if we | |
6341 | * add more hard links than can fit in the ref item. | |
6342 | */ | |
6343 | key[1].objectid = objectid; | |
962a298f | 6344 | key[1].type = BTRFS_INODE_REF_KEY; |
ef3b9af5 FM |
6345 | key[1].offset = ref_objectid; |
6346 | ||
6347 | sizes[1] = name_len + sizeof(*ref); | |
6348 | } | |
9c58309d | 6349 | |
b0d5d10f CM |
6350 | location = &BTRFS_I(inode)->location; |
6351 | location->objectid = objectid; | |
6352 | location->offset = 0; | |
962a298f | 6353 | location->type = BTRFS_INODE_ITEM_KEY; |
b0d5d10f CM |
6354 | |
6355 | ret = btrfs_insert_inode_locked(inode); | |
6356 | if (ret < 0) | |
6357 | goto fail; | |
6358 | ||
b9473439 | 6359 | path->leave_spinning = 1; |
ef3b9af5 | 6360 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems); |
9c58309d | 6361 | if (ret != 0) |
b0d5d10f | 6362 | goto fail_unlock; |
5f39d397 | 6363 | |
ecc11fab | 6364 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 6365 | inode_set_bytes(inode, 0); |
9cc97d64 | 6366 | |
c2050a45 | 6367 | inode->i_mtime = current_time(inode); |
9cc97d64 | 6368 | inode->i_atime = inode->i_mtime; |
6369 | inode->i_ctime = inode->i_mtime; | |
6370 | BTRFS_I(inode)->i_otime = inode->i_mtime; | |
6371 | ||
5f39d397 CM |
6372 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
6373 | struct btrfs_inode_item); | |
b159fa28 | 6374 | memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item, |
293f7e07 | 6375 | sizeof(*inode_item)); |
e02119d5 | 6376 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d | 6377 | |
ef3b9af5 FM |
6378 | if (name) { |
6379 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
6380 | struct btrfs_inode_ref); | |
6381 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
6382 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); | |
6383 | ptr = (unsigned long)(ref + 1); | |
6384 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
6385 | } | |
9c58309d | 6386 | |
5f39d397 CM |
6387 | btrfs_mark_buffer_dirty(path->nodes[0]); |
6388 | btrfs_free_path(path); | |
6389 | ||
6cbff00f CH |
6390 | btrfs_inherit_iflags(inode, dir); |
6391 | ||
569254b0 | 6392 | if (S_ISREG(mode)) { |
0b246afa | 6393 | if (btrfs_test_opt(fs_info, NODATASUM)) |
94272164 | 6394 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; |
0b246afa | 6395 | if (btrfs_test_opt(fs_info, NODATACOW)) |
f2bdf9a8 JB |
6396 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW | |
6397 | BTRFS_INODE_NODATASUM; | |
94272164 CM |
6398 | } |
6399 | ||
5d4f98a2 | 6400 | inode_tree_add(inode); |
1abe9b8a | 6401 | |
6402 | trace_btrfs_inode_new(inode); | |
1973f0fa | 6403 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 6404 | |
8ea05e3a AB |
6405 | btrfs_update_root_times(trans, root); |
6406 | ||
63541927 FDBM |
6407 | ret = btrfs_inode_inherit_props(trans, inode, dir); |
6408 | if (ret) | |
0b246afa | 6409 | btrfs_err(fs_info, |
63541927 | 6410 | "error inheriting props for ino %llu (root %llu): %d", |
f85b7379 | 6411 | btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret); |
63541927 | 6412 | |
39279cc3 | 6413 | return inode; |
b0d5d10f CM |
6414 | |
6415 | fail_unlock: | |
6416 | unlock_new_inode(inode); | |
5f39d397 | 6417 | fail: |
ef3b9af5 | 6418 | if (dir && name) |
aec7477b | 6419 | BTRFS_I(dir)->index_cnt--; |
5f39d397 | 6420 | btrfs_free_path(path); |
09771430 | 6421 | iput(inode); |
5f39d397 | 6422 | return ERR_PTR(ret); |
39279cc3 CM |
6423 | } |
6424 | ||
6425 | static inline u8 btrfs_inode_type(struct inode *inode) | |
6426 | { | |
6427 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
6428 | } | |
6429 | ||
d352ac68 CM |
6430 | /* |
6431 | * utility function to add 'inode' into 'parent_inode' with | |
6432 | * a give name and a given sequence number. | |
6433 | * if 'add_backref' is true, also insert a backref from the | |
6434 | * inode to the parent directory. | |
6435 | */ | |
e02119d5 | 6436 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
db0a669f | 6437 | struct btrfs_inode *parent_inode, struct btrfs_inode *inode, |
e02119d5 | 6438 | const char *name, int name_len, int add_backref, u64 index) |
39279cc3 | 6439 | { |
db0a669f | 6440 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
4df27c4d | 6441 | int ret = 0; |
39279cc3 | 6442 | struct btrfs_key key; |
db0a669f NB |
6443 | struct btrfs_root *root = parent_inode->root; |
6444 | u64 ino = btrfs_ino(inode); | |
6445 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 6446 | |
33345d01 | 6447 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
db0a669f | 6448 | memcpy(&key, &inode->root->root_key, sizeof(key)); |
4df27c4d | 6449 | } else { |
33345d01 | 6450 | key.objectid = ino; |
962a298f | 6451 | key.type = BTRFS_INODE_ITEM_KEY; |
4df27c4d YZ |
6452 | key.offset = 0; |
6453 | } | |
6454 | ||
33345d01 | 6455 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
0b246afa JM |
6456 | ret = btrfs_add_root_ref(trans, fs_info, key.objectid, |
6457 | root->root_key.objectid, parent_ino, | |
6458 | index, name, name_len); | |
4df27c4d | 6459 | } else if (add_backref) { |
33345d01 LZ |
6460 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
6461 | parent_ino, index); | |
4df27c4d | 6462 | } |
39279cc3 | 6463 | |
79787eaa JM |
6464 | /* Nothing to clean up yet */ |
6465 | if (ret) | |
6466 | return ret; | |
4df27c4d | 6467 | |
79787eaa JM |
6468 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
6469 | parent_inode, &key, | |
db0a669f | 6470 | btrfs_inode_type(&inode->vfs_inode), index); |
9c52057c | 6471 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
6472 | goto fail_dir_item; |
6473 | else if (ret) { | |
66642832 | 6474 | btrfs_abort_transaction(trans, ret); |
79787eaa | 6475 | return ret; |
39279cc3 | 6476 | } |
79787eaa | 6477 | |
db0a669f | 6478 | btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size + |
79787eaa | 6479 | name_len * 2); |
db0a669f NB |
6480 | inode_inc_iversion(&parent_inode->vfs_inode); |
6481 | parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime = | |
6482 | current_time(&parent_inode->vfs_inode); | |
6483 | ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode); | |
79787eaa | 6484 | if (ret) |
66642832 | 6485 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 6486 | return ret; |
fe66a05a CM |
6487 | |
6488 | fail_dir_item: | |
6489 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
6490 | u64 local_index; | |
6491 | int err; | |
0b246afa JM |
6492 | err = btrfs_del_root_ref(trans, fs_info, key.objectid, |
6493 | root->root_key.objectid, parent_ino, | |
6494 | &local_index, name, name_len); | |
fe66a05a CM |
6495 | |
6496 | } else if (add_backref) { | |
6497 | u64 local_index; | |
6498 | int err; | |
6499 | ||
6500 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
6501 | ino, parent_ino, &local_index); | |
6502 | } | |
6503 | return ret; | |
39279cc3 CM |
6504 | } |
6505 | ||
6506 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
cef415af NB |
6507 | struct btrfs_inode *dir, struct dentry *dentry, |
6508 | struct btrfs_inode *inode, int backref, u64 index) | |
39279cc3 | 6509 | { |
a1b075d2 JB |
6510 | int err = btrfs_add_link(trans, dir, inode, |
6511 | dentry->d_name.name, dentry->d_name.len, | |
6512 | backref, index); | |
39279cc3 CM |
6513 | if (err > 0) |
6514 | err = -EEXIST; | |
6515 | return err; | |
6516 | } | |
6517 | ||
618e21d5 | 6518 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 6519 | umode_t mode, dev_t rdev) |
618e21d5 | 6520 | { |
2ff7e61e | 6521 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
618e21d5 JB |
6522 | struct btrfs_trans_handle *trans; |
6523 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6524 | struct inode *inode = NULL; |
618e21d5 JB |
6525 | int err; |
6526 | int drop_inode = 0; | |
6527 | u64 objectid; | |
00e4e6b3 | 6528 | u64 index = 0; |
618e21d5 | 6529 | |
9ed74f2d JB |
6530 | /* |
6531 | * 2 for inode item and ref | |
6532 | * 2 for dir items | |
6533 | * 1 for xattr if selinux is on | |
6534 | */ | |
a22285a6 YZ |
6535 | trans = btrfs_start_transaction(root, 5); |
6536 | if (IS_ERR(trans)) | |
6537 | return PTR_ERR(trans); | |
1832a6d5 | 6538 | |
581bb050 LZ |
6539 | err = btrfs_find_free_ino(root, &objectid); |
6540 | if (err) | |
6541 | goto out_unlock; | |
6542 | ||
aec7477b | 6543 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6544 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6545 | mode, &index); | |
7cf96da3 TI |
6546 | if (IS_ERR(inode)) { |
6547 | err = PTR_ERR(inode); | |
618e21d5 | 6548 | goto out_unlock; |
7cf96da3 | 6549 | } |
618e21d5 | 6550 | |
ad19db71 CS |
6551 | /* |
6552 | * If the active LSM wants to access the inode during | |
6553 | * d_instantiate it needs these. Smack checks to see | |
6554 | * if the filesystem supports xattrs by looking at the | |
6555 | * ops vector. | |
6556 | */ | |
ad19db71 | 6557 | inode->i_op = &btrfs_special_inode_operations; |
b0d5d10f CM |
6558 | init_special_inode(inode, inode->i_mode, rdev); |
6559 | ||
6560 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
618e21d5 | 6561 | if (err) |
b0d5d10f CM |
6562 | goto out_unlock_inode; |
6563 | ||
cef415af NB |
6564 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6565 | 0, index); | |
b0d5d10f CM |
6566 | if (err) { |
6567 | goto out_unlock_inode; | |
6568 | } else { | |
1b4ab1bb | 6569 | btrfs_update_inode(trans, root, inode); |
b0d5d10f | 6570 | unlock_new_inode(inode); |
08c422c2 | 6571 | d_instantiate(dentry, inode); |
618e21d5 | 6572 | } |
b0d5d10f | 6573 | |
618e21d5 | 6574 | out_unlock: |
3a45bb20 | 6575 | btrfs_end_transaction(trans); |
2ff7e61e | 6576 | btrfs_btree_balance_dirty(fs_info); |
618e21d5 JB |
6577 | if (drop_inode) { |
6578 | inode_dec_link_count(inode); | |
6579 | iput(inode); | |
6580 | } | |
618e21d5 | 6581 | return err; |
b0d5d10f CM |
6582 | |
6583 | out_unlock_inode: | |
6584 | drop_inode = 1; | |
6585 | unlock_new_inode(inode); | |
6586 | goto out_unlock; | |
6587 | ||
618e21d5 JB |
6588 | } |
6589 | ||
39279cc3 | 6590 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 6591 | umode_t mode, bool excl) |
39279cc3 | 6592 | { |
2ff7e61e | 6593 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
6594 | struct btrfs_trans_handle *trans; |
6595 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6596 | struct inode *inode = NULL; |
43baa579 | 6597 | int drop_inode_on_err = 0; |
a22285a6 | 6598 | int err; |
39279cc3 | 6599 | u64 objectid; |
00e4e6b3 | 6600 | u64 index = 0; |
39279cc3 | 6601 | |
9ed74f2d JB |
6602 | /* |
6603 | * 2 for inode item and ref | |
6604 | * 2 for dir items | |
6605 | * 1 for xattr if selinux is on | |
6606 | */ | |
a22285a6 YZ |
6607 | trans = btrfs_start_transaction(root, 5); |
6608 | if (IS_ERR(trans)) | |
6609 | return PTR_ERR(trans); | |
9ed74f2d | 6610 | |
581bb050 LZ |
6611 | err = btrfs_find_free_ino(root, &objectid); |
6612 | if (err) | |
6613 | goto out_unlock; | |
6614 | ||
aec7477b | 6615 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6616 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6617 | mode, &index); | |
7cf96da3 TI |
6618 | if (IS_ERR(inode)) { |
6619 | err = PTR_ERR(inode); | |
39279cc3 | 6620 | goto out_unlock; |
7cf96da3 | 6621 | } |
43baa579 | 6622 | drop_inode_on_err = 1; |
ad19db71 CS |
6623 | /* |
6624 | * If the active LSM wants to access the inode during | |
6625 | * d_instantiate it needs these. Smack checks to see | |
6626 | * if the filesystem supports xattrs by looking at the | |
6627 | * ops vector. | |
6628 | */ | |
6629 | inode->i_fop = &btrfs_file_operations; | |
6630 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 6631 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
6632 | |
6633 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
6634 | if (err) | |
6635 | goto out_unlock_inode; | |
6636 | ||
6637 | err = btrfs_update_inode(trans, root, inode); | |
6638 | if (err) | |
6639 | goto out_unlock_inode; | |
ad19db71 | 6640 | |
cef415af NB |
6641 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6642 | 0, index); | |
39279cc3 | 6643 | if (err) |
b0d5d10f | 6644 | goto out_unlock_inode; |
43baa579 | 6645 | |
43baa579 | 6646 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
b0d5d10f | 6647 | unlock_new_inode(inode); |
43baa579 FB |
6648 | d_instantiate(dentry, inode); |
6649 | ||
39279cc3 | 6650 | out_unlock: |
3a45bb20 | 6651 | btrfs_end_transaction(trans); |
43baa579 | 6652 | if (err && drop_inode_on_err) { |
39279cc3 CM |
6653 | inode_dec_link_count(inode); |
6654 | iput(inode); | |
6655 | } | |
2ff7e61e | 6656 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 | 6657 | return err; |
b0d5d10f CM |
6658 | |
6659 | out_unlock_inode: | |
6660 | unlock_new_inode(inode); | |
6661 | goto out_unlock; | |
6662 | ||
39279cc3 CM |
6663 | } |
6664 | ||
6665 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
6666 | struct dentry *dentry) | |
6667 | { | |
271dba45 | 6668 | struct btrfs_trans_handle *trans = NULL; |
39279cc3 | 6669 | struct btrfs_root *root = BTRFS_I(dir)->root; |
2b0143b5 | 6670 | struct inode *inode = d_inode(old_dentry); |
2ff7e61e | 6671 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
00e4e6b3 | 6672 | u64 index; |
39279cc3 CM |
6673 | int err; |
6674 | int drop_inode = 0; | |
6675 | ||
4a8be425 TH |
6676 | /* do not allow sys_link's with other subvols of the same device */ |
6677 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 6678 | return -EXDEV; |
4a8be425 | 6679 | |
f186373f | 6680 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 6681 | return -EMLINK; |
4a8be425 | 6682 | |
877574e2 | 6683 | err = btrfs_set_inode_index(BTRFS_I(dir), &index); |
aec7477b JB |
6684 | if (err) |
6685 | goto fail; | |
6686 | ||
a22285a6 | 6687 | /* |
7e6b6465 | 6688 | * 2 items for inode and inode ref |
a22285a6 | 6689 | * 2 items for dir items |
7e6b6465 | 6690 | * 1 item for parent inode |
a22285a6 | 6691 | */ |
7e6b6465 | 6692 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
6693 | if (IS_ERR(trans)) { |
6694 | err = PTR_ERR(trans); | |
271dba45 | 6695 | trans = NULL; |
a22285a6 YZ |
6696 | goto fail; |
6697 | } | |
5f39d397 | 6698 | |
67de1176 MX |
6699 | /* There are several dir indexes for this inode, clear the cache. */ |
6700 | BTRFS_I(inode)->dir_index = 0ULL; | |
8b558c5f | 6701 | inc_nlink(inode); |
0c4d2d95 | 6702 | inode_inc_iversion(inode); |
c2050a45 | 6703 | inode->i_ctime = current_time(inode); |
7de9c6ee | 6704 | ihold(inode); |
e9976151 | 6705 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 6706 | |
cef415af NB |
6707 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6708 | 1, index); | |
5f39d397 | 6709 | |
a5719521 | 6710 | if (err) { |
54aa1f4d | 6711 | drop_inode = 1; |
a5719521 | 6712 | } else { |
10d9f309 | 6713 | struct dentry *parent = dentry->d_parent; |
a5719521 | 6714 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
6715 | if (err) |
6716 | goto fail; | |
ef3b9af5 FM |
6717 | if (inode->i_nlink == 1) { |
6718 | /* | |
6719 | * If new hard link count is 1, it's a file created | |
6720 | * with open(2) O_TMPFILE flag. | |
6721 | */ | |
3d6ae7bb | 6722 | err = btrfs_orphan_del(trans, BTRFS_I(inode)); |
ef3b9af5 FM |
6723 | if (err) |
6724 | goto fail; | |
6725 | } | |
08c422c2 | 6726 | d_instantiate(dentry, inode); |
9ca5fbfb | 6727 | btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent); |
a5719521 | 6728 | } |
39279cc3 | 6729 | |
1832a6d5 | 6730 | fail: |
271dba45 | 6731 | if (trans) |
3a45bb20 | 6732 | btrfs_end_transaction(trans); |
39279cc3 CM |
6733 | if (drop_inode) { |
6734 | inode_dec_link_count(inode); | |
6735 | iput(inode); | |
6736 | } | |
2ff7e61e | 6737 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6738 | return err; |
6739 | } | |
6740 | ||
18bb1db3 | 6741 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 6742 | { |
2ff7e61e | 6743 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
b9d86667 | 6744 | struct inode *inode = NULL; |
39279cc3 CM |
6745 | struct btrfs_trans_handle *trans; |
6746 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
6747 | int err = 0; | |
6748 | int drop_on_err = 0; | |
b9d86667 | 6749 | u64 objectid = 0; |
00e4e6b3 | 6750 | u64 index = 0; |
39279cc3 | 6751 | |
9ed74f2d JB |
6752 | /* |
6753 | * 2 items for inode and ref | |
6754 | * 2 items for dir items | |
6755 | * 1 for xattr if selinux is on | |
6756 | */ | |
a22285a6 YZ |
6757 | trans = btrfs_start_transaction(root, 5); |
6758 | if (IS_ERR(trans)) | |
6759 | return PTR_ERR(trans); | |
39279cc3 | 6760 | |
581bb050 LZ |
6761 | err = btrfs_find_free_ino(root, &objectid); |
6762 | if (err) | |
6763 | goto out_fail; | |
6764 | ||
aec7477b | 6765 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6766 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6767 | S_IFDIR | mode, &index); | |
39279cc3 CM |
6768 | if (IS_ERR(inode)) { |
6769 | err = PTR_ERR(inode); | |
6770 | goto out_fail; | |
6771 | } | |
5f39d397 | 6772 | |
39279cc3 | 6773 | drop_on_err = 1; |
b0d5d10f CM |
6774 | /* these must be set before we unlock the inode */ |
6775 | inode->i_op = &btrfs_dir_inode_operations; | |
6776 | inode->i_fop = &btrfs_dir_file_operations; | |
33268eaf | 6777 | |
2a7dba39 | 6778 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf | 6779 | if (err) |
b0d5d10f | 6780 | goto out_fail_inode; |
39279cc3 | 6781 | |
6ef06d27 | 6782 | btrfs_i_size_write(BTRFS_I(inode), 0); |
39279cc3 CM |
6783 | err = btrfs_update_inode(trans, root, inode); |
6784 | if (err) | |
b0d5d10f | 6785 | goto out_fail_inode; |
5f39d397 | 6786 | |
db0a669f NB |
6787 | err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), |
6788 | dentry->d_name.name, | |
6789 | dentry->d_name.len, 0, index); | |
39279cc3 | 6790 | if (err) |
b0d5d10f | 6791 | goto out_fail_inode; |
5f39d397 | 6792 | |
39279cc3 | 6793 | d_instantiate(dentry, inode); |
b0d5d10f CM |
6794 | /* |
6795 | * mkdir is special. We're unlocking after we call d_instantiate | |
6796 | * to avoid a race with nfsd calling d_instantiate. | |
6797 | */ | |
6798 | unlock_new_inode(inode); | |
39279cc3 | 6799 | drop_on_err = 0; |
39279cc3 CM |
6800 | |
6801 | out_fail: | |
3a45bb20 | 6802 | btrfs_end_transaction(trans); |
c7cfb8a5 WS |
6803 | if (drop_on_err) { |
6804 | inode_dec_link_count(inode); | |
39279cc3 | 6805 | iput(inode); |
c7cfb8a5 | 6806 | } |
2ff7e61e | 6807 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 | 6808 | return err; |
b0d5d10f CM |
6809 | |
6810 | out_fail_inode: | |
6811 | unlock_new_inode(inode); | |
6812 | goto out_fail; | |
39279cc3 CM |
6813 | } |
6814 | ||
e6c4efd8 QW |
6815 | /* Find next extent map of a given extent map, caller needs to ensure locks */ |
6816 | static struct extent_map *next_extent_map(struct extent_map *em) | |
6817 | { | |
6818 | struct rb_node *next; | |
6819 | ||
6820 | next = rb_next(&em->rb_node); | |
6821 | if (!next) | |
6822 | return NULL; | |
6823 | return container_of(next, struct extent_map, rb_node); | |
6824 | } | |
6825 | ||
6826 | static struct extent_map *prev_extent_map(struct extent_map *em) | |
6827 | { | |
6828 | struct rb_node *prev; | |
6829 | ||
6830 | prev = rb_prev(&em->rb_node); | |
6831 | if (!prev) | |
6832 | return NULL; | |
6833 | return container_of(prev, struct extent_map, rb_node); | |
6834 | } | |
6835 | ||
d352ac68 | 6836 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
e6c4efd8 | 6837 | * the existing extent is the nearest extent to map_start, |
d352ac68 | 6838 | * and an extent that you want to insert, deal with overlap and insert |
e6c4efd8 | 6839 | * the best fitted new extent into the tree. |
d352ac68 | 6840 | */ |
3b951516 CM |
6841 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
6842 | struct extent_map *existing, | |
e6dcd2dc | 6843 | struct extent_map *em, |
51f395ad | 6844 | u64 map_start) |
3b951516 | 6845 | { |
e6c4efd8 QW |
6846 | struct extent_map *prev; |
6847 | struct extent_map *next; | |
6848 | u64 start; | |
6849 | u64 end; | |
3b951516 | 6850 | u64 start_diff; |
3b951516 | 6851 | |
e6dcd2dc | 6852 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
e6c4efd8 QW |
6853 | |
6854 | if (existing->start > map_start) { | |
6855 | next = existing; | |
6856 | prev = prev_extent_map(next); | |
6857 | } else { | |
6858 | prev = existing; | |
6859 | next = next_extent_map(prev); | |
6860 | } | |
6861 | ||
6862 | start = prev ? extent_map_end(prev) : em->start; | |
6863 | start = max_t(u64, start, em->start); | |
6864 | end = next ? next->start : extent_map_end(em); | |
6865 | end = min_t(u64, end, extent_map_end(em)); | |
6866 | start_diff = start - em->start; | |
6867 | em->start = start; | |
6868 | em->len = end - start; | |
c8b97818 CM |
6869 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
6870 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 6871 | em->block_start += start_diff; |
c8b97818 CM |
6872 | em->block_len -= start_diff; |
6873 | } | |
09a2a8f9 | 6874 | return add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
6875 | } |
6876 | ||
c8b97818 | 6877 | static noinline int uncompress_inline(struct btrfs_path *path, |
e40da0e5 | 6878 | struct page *page, |
c8b97818 CM |
6879 | size_t pg_offset, u64 extent_offset, |
6880 | struct btrfs_file_extent_item *item) | |
6881 | { | |
6882 | int ret; | |
6883 | struct extent_buffer *leaf = path->nodes[0]; | |
6884 | char *tmp; | |
6885 | size_t max_size; | |
6886 | unsigned long inline_size; | |
6887 | unsigned long ptr; | |
261507a0 | 6888 | int compress_type; |
c8b97818 CM |
6889 | |
6890 | WARN_ON(pg_offset != 0); | |
261507a0 | 6891 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
6892 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
6893 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
dd3cc16b | 6894 | btrfs_item_nr(path->slots[0])); |
c8b97818 | 6895 | tmp = kmalloc(inline_size, GFP_NOFS); |
8d413713 TI |
6896 | if (!tmp) |
6897 | return -ENOMEM; | |
c8b97818 CM |
6898 | ptr = btrfs_file_extent_inline_start(item); |
6899 | ||
6900 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
6901 | ||
09cbfeaf | 6902 | max_size = min_t(unsigned long, PAGE_SIZE, max_size); |
261507a0 LZ |
6903 | ret = btrfs_decompress(compress_type, tmp, page, |
6904 | extent_offset, inline_size, max_size); | |
e1699d2d ZB |
6905 | |
6906 | /* | |
6907 | * decompression code contains a memset to fill in any space between the end | |
6908 | * of the uncompressed data and the end of max_size in case the decompressed | |
6909 | * data ends up shorter than ram_bytes. That doesn't cover the hole between | |
6910 | * the end of an inline extent and the beginning of the next block, so we | |
6911 | * cover that region here. | |
6912 | */ | |
6913 | ||
6914 | if (max_size + pg_offset < PAGE_SIZE) { | |
6915 | char *map = kmap(page); | |
6916 | memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset); | |
6917 | kunmap(page); | |
6918 | } | |
c8b97818 | 6919 | kfree(tmp); |
166ae5a4 | 6920 | return ret; |
c8b97818 CM |
6921 | } |
6922 | ||
d352ac68 CM |
6923 | /* |
6924 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
6925 | * the ugly parts come from merging extents from the disk with the in-ram |
6926 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
6927 | * where the in-ram extents might be locked pending data=ordered completion. |
6928 | * | |
6929 | * This also copies inline extents directly into the page. | |
6930 | */ | |
fc4f21b1 NB |
6931 | struct extent_map *btrfs_get_extent(struct btrfs_inode *inode, |
6932 | struct page *page, | |
6933 | size_t pg_offset, u64 start, u64 len, | |
6934 | int create) | |
a52d9a80 | 6935 | { |
fc4f21b1 | 6936 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
a52d9a80 CM |
6937 | int ret; |
6938 | int err = 0; | |
a52d9a80 CM |
6939 | u64 extent_start = 0; |
6940 | u64 extent_end = 0; | |
fc4f21b1 | 6941 | u64 objectid = btrfs_ino(inode); |
a52d9a80 | 6942 | u32 found_type; |
f421950f | 6943 | struct btrfs_path *path = NULL; |
fc4f21b1 | 6944 | struct btrfs_root *root = inode->root; |
a52d9a80 | 6945 | struct btrfs_file_extent_item *item; |
5f39d397 CM |
6946 | struct extent_buffer *leaf; |
6947 | struct btrfs_key found_key; | |
a52d9a80 | 6948 | struct extent_map *em = NULL; |
fc4f21b1 NB |
6949 | struct extent_map_tree *em_tree = &inode->extent_tree; |
6950 | struct extent_io_tree *io_tree = &inode->io_tree; | |
a52d9a80 | 6951 | struct btrfs_trans_handle *trans = NULL; |
7ffbb598 | 6952 | const bool new_inline = !page || create; |
a52d9a80 | 6953 | |
a52d9a80 | 6954 | again: |
890871be | 6955 | read_lock(&em_tree->lock); |
d1310b2e | 6956 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d | 6957 | if (em) |
0b246afa | 6958 | em->bdev = fs_info->fs_devices->latest_bdev; |
890871be | 6959 | read_unlock(&em_tree->lock); |
d1310b2e | 6960 | |
a52d9a80 | 6961 | if (em) { |
e1c4b745 CM |
6962 | if (em->start > start || em->start + em->len <= start) |
6963 | free_extent_map(em); | |
6964 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
6965 | free_extent_map(em); |
6966 | else | |
6967 | goto out; | |
a52d9a80 | 6968 | } |
172ddd60 | 6969 | em = alloc_extent_map(); |
a52d9a80 | 6970 | if (!em) { |
d1310b2e CM |
6971 | err = -ENOMEM; |
6972 | goto out; | |
a52d9a80 | 6973 | } |
0b246afa | 6974 | em->bdev = fs_info->fs_devices->latest_bdev; |
d1310b2e | 6975 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 6976 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 6977 | em->len = (u64)-1; |
c8b97818 | 6978 | em->block_len = (u64)-1; |
f421950f CM |
6979 | |
6980 | if (!path) { | |
6981 | path = btrfs_alloc_path(); | |
026fd317 JB |
6982 | if (!path) { |
6983 | err = -ENOMEM; | |
6984 | goto out; | |
6985 | } | |
6986 | /* | |
6987 | * Chances are we'll be called again, so go ahead and do | |
6988 | * readahead | |
6989 | */ | |
e4058b54 | 6990 | path->reada = READA_FORWARD; |
f421950f CM |
6991 | } |
6992 | ||
179e29e4 CM |
6993 | ret = btrfs_lookup_file_extent(trans, root, path, |
6994 | objectid, start, trans != NULL); | |
a52d9a80 CM |
6995 | if (ret < 0) { |
6996 | err = ret; | |
6997 | goto out; | |
6998 | } | |
6999 | ||
7000 | if (ret != 0) { | |
7001 | if (path->slots[0] == 0) | |
7002 | goto not_found; | |
7003 | path->slots[0]--; | |
7004 | } | |
7005 | ||
5f39d397 CM |
7006 | leaf = path->nodes[0]; |
7007 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 7008 | struct btrfs_file_extent_item); |
a52d9a80 | 7009 | /* are we inside the extent that was found? */ |
5f39d397 | 7010 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
962a298f | 7011 | found_type = found_key.type; |
5f39d397 | 7012 | if (found_key.objectid != objectid || |
a52d9a80 | 7013 | found_type != BTRFS_EXTENT_DATA_KEY) { |
25a50341 JB |
7014 | /* |
7015 | * If we backup past the first extent we want to move forward | |
7016 | * and see if there is an extent in front of us, otherwise we'll | |
7017 | * say there is a hole for our whole search range which can | |
7018 | * cause problems. | |
7019 | */ | |
7020 | extent_end = start; | |
7021 | goto next; | |
a52d9a80 CM |
7022 | } |
7023 | ||
5f39d397 CM |
7024 | found_type = btrfs_file_extent_type(leaf, item); |
7025 | extent_start = found_key.offset; | |
d899e052 YZ |
7026 | if (found_type == BTRFS_FILE_EXTENT_REG || |
7027 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 7028 | extent_end = extent_start + |
db94535d | 7029 | btrfs_file_extent_num_bytes(leaf, item); |
09ed2f16 LB |
7030 | |
7031 | trace_btrfs_get_extent_show_fi_regular(inode, leaf, item, | |
7032 | extent_start); | |
9036c102 YZ |
7033 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
7034 | size_t size; | |
514ac8ad | 7035 | size = btrfs_file_extent_inline_len(leaf, path->slots[0], item); |
da17066c | 7036 | extent_end = ALIGN(extent_start + size, |
0b246afa | 7037 | fs_info->sectorsize); |
09ed2f16 LB |
7038 | |
7039 | trace_btrfs_get_extent_show_fi_inline(inode, leaf, item, | |
7040 | path->slots[0], | |
7041 | extent_start); | |
9036c102 | 7042 | } |
25a50341 | 7043 | next: |
9036c102 YZ |
7044 | if (start >= extent_end) { |
7045 | path->slots[0]++; | |
7046 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
7047 | ret = btrfs_next_leaf(root, path); | |
7048 | if (ret < 0) { | |
7049 | err = ret; | |
7050 | goto out; | |
a52d9a80 | 7051 | } |
9036c102 YZ |
7052 | if (ret > 0) |
7053 | goto not_found; | |
7054 | leaf = path->nodes[0]; | |
a52d9a80 | 7055 | } |
9036c102 YZ |
7056 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
7057 | if (found_key.objectid != objectid || | |
7058 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
7059 | goto not_found; | |
7060 | if (start + len <= found_key.offset) | |
7061 | goto not_found; | |
e2eca69d WS |
7062 | if (start > found_key.offset) |
7063 | goto next; | |
9036c102 | 7064 | em->start = start; |
70c8a91c | 7065 | em->orig_start = start; |
9036c102 YZ |
7066 | em->len = found_key.offset - start; |
7067 | goto not_found_em; | |
7068 | } | |
7069 | ||
fc4f21b1 | 7070 | btrfs_extent_item_to_extent_map(inode, path, item, |
9cdc5124 | 7071 | new_inline, em); |
7ffbb598 | 7072 | |
d899e052 YZ |
7073 | if (found_type == BTRFS_FILE_EXTENT_REG || |
7074 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 CM |
7075 | goto insert; |
7076 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 7077 | unsigned long ptr; |
a52d9a80 | 7078 | char *map; |
3326d1b0 CM |
7079 | size_t size; |
7080 | size_t extent_offset; | |
7081 | size_t copy_size; | |
a52d9a80 | 7082 | |
7ffbb598 | 7083 | if (new_inline) |
689f9346 | 7084 | goto out; |
5f39d397 | 7085 | |
514ac8ad | 7086 | size = btrfs_file_extent_inline_len(leaf, path->slots[0], item); |
9036c102 | 7087 | extent_offset = page_offset(page) + pg_offset - extent_start; |
09cbfeaf KS |
7088 | copy_size = min_t(u64, PAGE_SIZE - pg_offset, |
7089 | size - extent_offset); | |
3326d1b0 | 7090 | em->start = extent_start + extent_offset; |
0b246afa | 7091 | em->len = ALIGN(copy_size, fs_info->sectorsize); |
b4939680 | 7092 | em->orig_block_len = em->len; |
70c8a91c | 7093 | em->orig_start = em->start; |
689f9346 | 7094 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 7095 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
7096 | if (btrfs_file_extent_compression(leaf, item) != |
7097 | BTRFS_COMPRESS_NONE) { | |
e40da0e5 | 7098 | ret = uncompress_inline(path, page, pg_offset, |
c8b97818 | 7099 | extent_offset, item); |
166ae5a4 ZB |
7100 | if (ret) { |
7101 | err = ret; | |
7102 | goto out; | |
7103 | } | |
c8b97818 CM |
7104 | } else { |
7105 | map = kmap(page); | |
7106 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
7107 | copy_size); | |
09cbfeaf | 7108 | if (pg_offset + copy_size < PAGE_SIZE) { |
93c82d57 | 7109 | memset(map + pg_offset + copy_size, 0, |
09cbfeaf | 7110 | PAGE_SIZE - pg_offset - |
93c82d57 CM |
7111 | copy_size); |
7112 | } | |
c8b97818 CM |
7113 | kunmap(page); |
7114 | } | |
179e29e4 CM |
7115 | flush_dcache_page(page); |
7116 | } else if (create && PageUptodate(page)) { | |
6bf7e080 | 7117 | BUG(); |
179e29e4 CM |
7118 | if (!trans) { |
7119 | kunmap(page); | |
7120 | free_extent_map(em); | |
7121 | em = NULL; | |
ff5714cc | 7122 | |
b3b4aa74 | 7123 | btrfs_release_path(path); |
7a7eaa40 | 7124 | trans = btrfs_join_transaction(root); |
ff5714cc | 7125 | |
3612b495 TI |
7126 | if (IS_ERR(trans)) |
7127 | return ERR_CAST(trans); | |
179e29e4 CM |
7128 | goto again; |
7129 | } | |
c8b97818 | 7130 | map = kmap(page); |
70dec807 | 7131 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 7132 | copy_size); |
c8b97818 | 7133 | kunmap(page); |
179e29e4 | 7134 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 7135 | } |
d1310b2e | 7136 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 7137 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 | 7138 | goto insert; |
a52d9a80 CM |
7139 | } |
7140 | not_found: | |
7141 | em->start = start; | |
70c8a91c | 7142 | em->orig_start = start; |
d1310b2e | 7143 | em->len = len; |
a52d9a80 | 7144 | not_found_em: |
5f39d397 | 7145 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 7146 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 | 7147 | insert: |
b3b4aa74 | 7148 | btrfs_release_path(path); |
d1310b2e | 7149 | if (em->start > start || extent_map_end(em) <= start) { |
0b246afa | 7150 | btrfs_err(fs_info, |
5d163e0e JM |
7151 | "bad extent! em: [%llu %llu] passed [%llu %llu]", |
7152 | em->start, em->len, start, len); | |
a52d9a80 CM |
7153 | err = -EIO; |
7154 | goto out; | |
7155 | } | |
d1310b2e CM |
7156 | |
7157 | err = 0; | |
890871be | 7158 | write_lock(&em_tree->lock); |
09a2a8f9 | 7159 | ret = add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
7160 | /* it is possible that someone inserted the extent into the tree |
7161 | * while we had the lock dropped. It is also possible that | |
7162 | * an overlapping map exists in the tree | |
7163 | */ | |
a52d9a80 | 7164 | if (ret == -EEXIST) { |
3b951516 | 7165 | struct extent_map *existing; |
e6dcd2dc CM |
7166 | |
7167 | ret = 0; | |
7168 | ||
e6c4efd8 QW |
7169 | existing = search_extent_mapping(em_tree, start, len); |
7170 | /* | |
7171 | * existing will always be non-NULL, since there must be | |
7172 | * extent causing the -EEXIST. | |
7173 | */ | |
8dff9c85 | 7174 | if (existing->start == em->start && |
8e2bd3b7 | 7175 | extent_map_end(existing) >= extent_map_end(em) && |
8dff9c85 CM |
7176 | em->block_start == existing->block_start) { |
7177 | /* | |
8e2bd3b7 OS |
7178 | * The existing extent map already encompasses the |
7179 | * entire extent map we tried to add. | |
8dff9c85 CM |
7180 | */ |
7181 | free_extent_map(em); | |
7182 | em = existing; | |
7183 | err = 0; | |
7184 | ||
7185 | } else if (start >= extent_map_end(existing) || | |
32be3a1a | 7186 | start <= existing->start) { |
e6c4efd8 QW |
7187 | /* |
7188 | * The existing extent map is the one nearest to | |
7189 | * the [start, start + len) range which overlaps | |
7190 | */ | |
7191 | err = merge_extent_mapping(em_tree, existing, | |
7192 | em, start); | |
e1c4b745 | 7193 | free_extent_map(existing); |
e6c4efd8 | 7194 | if (err) { |
3b951516 CM |
7195 | free_extent_map(em); |
7196 | em = NULL; | |
7197 | } | |
7198 | } else { | |
7199 | free_extent_map(em); | |
7200 | em = existing; | |
e6dcd2dc | 7201 | err = 0; |
a52d9a80 | 7202 | } |
a52d9a80 | 7203 | } |
890871be | 7204 | write_unlock(&em_tree->lock); |
a52d9a80 | 7205 | out: |
1abe9b8a | 7206 | |
fc4f21b1 | 7207 | trace_btrfs_get_extent(root, inode, em); |
1abe9b8a | 7208 | |
527afb44 | 7209 | btrfs_free_path(path); |
a52d9a80 | 7210 | if (trans) { |
3a45bb20 | 7211 | ret = btrfs_end_transaction(trans); |
d397712b | 7212 | if (!err) |
a52d9a80 CM |
7213 | err = ret; |
7214 | } | |
a52d9a80 CM |
7215 | if (err) { |
7216 | free_extent_map(em); | |
a52d9a80 CM |
7217 | return ERR_PTR(err); |
7218 | } | |
79787eaa | 7219 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
7220 | return em; |
7221 | } | |
7222 | ||
fc4f21b1 NB |
7223 | struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode, |
7224 | struct page *page, | |
7225 | size_t pg_offset, u64 start, u64 len, | |
7226 | int create) | |
ec29ed5b CM |
7227 | { |
7228 | struct extent_map *em; | |
7229 | struct extent_map *hole_em = NULL; | |
7230 | u64 range_start = start; | |
7231 | u64 end; | |
7232 | u64 found; | |
7233 | u64 found_end; | |
7234 | int err = 0; | |
7235 | ||
7236 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
7237 | if (IS_ERR(em)) | |
7238 | return em; | |
9986277e DC |
7239 | /* |
7240 | * If our em maps to: | |
7241 | * - a hole or | |
7242 | * - a pre-alloc extent, | |
7243 | * there might actually be delalloc bytes behind it. | |
7244 | */ | |
7245 | if (em->block_start != EXTENT_MAP_HOLE && | |
7246 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7247 | return em; | |
7248 | else | |
7249 | hole_em = em; | |
ec29ed5b CM |
7250 | |
7251 | /* check to see if we've wrapped (len == -1 or similar) */ | |
7252 | end = start + len; | |
7253 | if (end < start) | |
7254 | end = (u64)-1; | |
7255 | else | |
7256 | end -= 1; | |
7257 | ||
7258 | em = NULL; | |
7259 | ||
7260 | /* ok, we didn't find anything, lets look for delalloc */ | |
fc4f21b1 | 7261 | found = count_range_bits(&inode->io_tree, &range_start, |
ec29ed5b CM |
7262 | end, len, EXTENT_DELALLOC, 1); |
7263 | found_end = range_start + found; | |
7264 | if (found_end < range_start) | |
7265 | found_end = (u64)-1; | |
7266 | ||
7267 | /* | |
7268 | * we didn't find anything useful, return | |
7269 | * the original results from get_extent() | |
7270 | */ | |
7271 | if (range_start > end || found_end <= start) { | |
7272 | em = hole_em; | |
7273 | hole_em = NULL; | |
7274 | goto out; | |
7275 | } | |
7276 | ||
7277 | /* adjust the range_start to make sure it doesn't | |
7278 | * go backwards from the start they passed in | |
7279 | */ | |
67871254 | 7280 | range_start = max(start, range_start); |
ec29ed5b CM |
7281 | found = found_end - range_start; |
7282 | ||
7283 | if (found > 0) { | |
7284 | u64 hole_start = start; | |
7285 | u64 hole_len = len; | |
7286 | ||
172ddd60 | 7287 | em = alloc_extent_map(); |
ec29ed5b CM |
7288 | if (!em) { |
7289 | err = -ENOMEM; | |
7290 | goto out; | |
7291 | } | |
7292 | /* | |
7293 | * when btrfs_get_extent can't find anything it | |
7294 | * returns one huge hole | |
7295 | * | |
7296 | * make sure what it found really fits our range, and | |
7297 | * adjust to make sure it is based on the start from | |
7298 | * the caller | |
7299 | */ | |
7300 | if (hole_em) { | |
7301 | u64 calc_end = extent_map_end(hole_em); | |
7302 | ||
7303 | if (calc_end <= start || (hole_em->start > end)) { | |
7304 | free_extent_map(hole_em); | |
7305 | hole_em = NULL; | |
7306 | } else { | |
7307 | hole_start = max(hole_em->start, start); | |
7308 | hole_len = calc_end - hole_start; | |
7309 | } | |
7310 | } | |
7311 | em->bdev = NULL; | |
7312 | if (hole_em && range_start > hole_start) { | |
7313 | /* our hole starts before our delalloc, so we | |
7314 | * have to return just the parts of the hole | |
7315 | * that go until the delalloc starts | |
7316 | */ | |
7317 | em->len = min(hole_len, | |
7318 | range_start - hole_start); | |
7319 | em->start = hole_start; | |
7320 | em->orig_start = hole_start; | |
7321 | /* | |
7322 | * don't adjust block start at all, | |
7323 | * it is fixed at EXTENT_MAP_HOLE | |
7324 | */ | |
7325 | em->block_start = hole_em->block_start; | |
7326 | em->block_len = hole_len; | |
f9e4fb53 LB |
7327 | if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags)) |
7328 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
ec29ed5b CM |
7329 | } else { |
7330 | em->start = range_start; | |
7331 | em->len = found; | |
7332 | em->orig_start = range_start; | |
7333 | em->block_start = EXTENT_MAP_DELALLOC; | |
7334 | em->block_len = found; | |
7335 | } | |
7336 | } else if (hole_em) { | |
7337 | return hole_em; | |
7338 | } | |
7339 | out: | |
7340 | ||
7341 | free_extent_map(hole_em); | |
7342 | if (err) { | |
7343 | free_extent_map(em); | |
7344 | return ERR_PTR(err); | |
7345 | } | |
7346 | return em; | |
7347 | } | |
7348 | ||
5f9a8a51 FM |
7349 | static struct extent_map *btrfs_create_dio_extent(struct inode *inode, |
7350 | const u64 start, | |
7351 | const u64 len, | |
7352 | const u64 orig_start, | |
7353 | const u64 block_start, | |
7354 | const u64 block_len, | |
7355 | const u64 orig_block_len, | |
7356 | const u64 ram_bytes, | |
7357 | const int type) | |
7358 | { | |
7359 | struct extent_map *em = NULL; | |
7360 | int ret; | |
7361 | ||
5f9a8a51 | 7362 | if (type != BTRFS_ORDERED_NOCOW) { |
6f9994db LB |
7363 | em = create_io_em(inode, start, len, orig_start, |
7364 | block_start, block_len, orig_block_len, | |
7365 | ram_bytes, | |
7366 | BTRFS_COMPRESS_NONE, /* compress_type */ | |
7367 | type); | |
5f9a8a51 FM |
7368 | if (IS_ERR(em)) |
7369 | goto out; | |
7370 | } | |
7371 | ret = btrfs_add_ordered_extent_dio(inode, start, block_start, | |
7372 | len, block_len, type); | |
7373 | if (ret) { | |
7374 | if (em) { | |
7375 | free_extent_map(em); | |
dcdbc059 | 7376 | btrfs_drop_extent_cache(BTRFS_I(inode), start, |
5f9a8a51 FM |
7377 | start + len - 1, 0); |
7378 | } | |
7379 | em = ERR_PTR(ret); | |
7380 | } | |
7381 | out: | |
5f9a8a51 FM |
7382 | |
7383 | return em; | |
7384 | } | |
7385 | ||
4b46fce2 JB |
7386 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
7387 | u64 start, u64 len) | |
7388 | { | |
0b246afa | 7389 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 7390 | struct btrfs_root *root = BTRFS_I(inode)->root; |
70c8a91c | 7391 | struct extent_map *em; |
4b46fce2 JB |
7392 | struct btrfs_key ins; |
7393 | u64 alloc_hint; | |
7394 | int ret; | |
4b46fce2 | 7395 | |
4b46fce2 | 7396 | alloc_hint = get_extent_allocation_hint(inode, start, len); |
0b246afa | 7397 | ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize, |
da17066c | 7398 | 0, alloc_hint, &ins, 1, 1); |
00361589 JB |
7399 | if (ret) |
7400 | return ERR_PTR(ret); | |
4b46fce2 | 7401 | |
5f9a8a51 FM |
7402 | em = btrfs_create_dio_extent(inode, start, ins.offset, start, |
7403 | ins.objectid, ins.offset, ins.offset, | |
6288d6ea | 7404 | ins.offset, BTRFS_ORDERED_REGULAR); |
0b246afa | 7405 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
5f9a8a51 | 7406 | if (IS_ERR(em)) |
2ff7e61e JM |
7407 | btrfs_free_reserved_extent(fs_info, ins.objectid, |
7408 | ins.offset, 1); | |
de0ee0ed | 7409 | |
4b46fce2 JB |
7410 | return em; |
7411 | } | |
7412 | ||
46bfbb5c CM |
7413 | /* |
7414 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
7415 | * block must be cow'd | |
7416 | */ | |
00361589 | 7417 | noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len, |
7ee9e440 JB |
7418 | u64 *orig_start, u64 *orig_block_len, |
7419 | u64 *ram_bytes) | |
46bfbb5c | 7420 | { |
2ff7e61e | 7421 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
46bfbb5c CM |
7422 | struct btrfs_path *path; |
7423 | int ret; | |
7424 | struct extent_buffer *leaf; | |
7425 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7b2b7085 | 7426 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
46bfbb5c CM |
7427 | struct btrfs_file_extent_item *fi; |
7428 | struct btrfs_key key; | |
7429 | u64 disk_bytenr; | |
7430 | u64 backref_offset; | |
7431 | u64 extent_end; | |
7432 | u64 num_bytes; | |
7433 | int slot; | |
7434 | int found_type; | |
7ee9e440 | 7435 | bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW); |
e77751aa | 7436 | |
46bfbb5c CM |
7437 | path = btrfs_alloc_path(); |
7438 | if (!path) | |
7439 | return -ENOMEM; | |
7440 | ||
f85b7379 DS |
7441 | ret = btrfs_lookup_file_extent(NULL, root, path, |
7442 | btrfs_ino(BTRFS_I(inode)), offset, 0); | |
46bfbb5c CM |
7443 | if (ret < 0) |
7444 | goto out; | |
7445 | ||
7446 | slot = path->slots[0]; | |
7447 | if (ret == 1) { | |
7448 | if (slot == 0) { | |
7449 | /* can't find the item, must cow */ | |
7450 | ret = 0; | |
7451 | goto out; | |
7452 | } | |
7453 | slot--; | |
7454 | } | |
7455 | ret = 0; | |
7456 | leaf = path->nodes[0]; | |
7457 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4a0cc7ca | 7458 | if (key.objectid != btrfs_ino(BTRFS_I(inode)) || |
46bfbb5c CM |
7459 | key.type != BTRFS_EXTENT_DATA_KEY) { |
7460 | /* not our file or wrong item type, must cow */ | |
7461 | goto out; | |
7462 | } | |
7463 | ||
7464 | if (key.offset > offset) { | |
7465 | /* Wrong offset, must cow */ | |
7466 | goto out; | |
7467 | } | |
7468 | ||
7469 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
7470 | found_type = btrfs_file_extent_type(leaf, fi); | |
7471 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
7472 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
7473 | /* not a regular extent, must cow */ | |
7474 | goto out; | |
7475 | } | |
7ee9e440 JB |
7476 | |
7477 | if (!nocow && found_type == BTRFS_FILE_EXTENT_REG) | |
7478 | goto out; | |
7479 | ||
e77751aa MX |
7480 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
7481 | if (extent_end <= offset) | |
7482 | goto out; | |
7483 | ||
46bfbb5c | 7484 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
7ee9e440 JB |
7485 | if (disk_bytenr == 0) |
7486 | goto out; | |
7487 | ||
7488 | if (btrfs_file_extent_compression(leaf, fi) || | |
7489 | btrfs_file_extent_encryption(leaf, fi) || | |
7490 | btrfs_file_extent_other_encoding(leaf, fi)) | |
7491 | goto out; | |
7492 | ||
46bfbb5c CM |
7493 | backref_offset = btrfs_file_extent_offset(leaf, fi); |
7494 | ||
7ee9e440 JB |
7495 | if (orig_start) { |
7496 | *orig_start = key.offset - backref_offset; | |
7497 | *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
7498 | *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); | |
7499 | } | |
eb384b55 | 7500 | |
2ff7e61e | 7501 | if (btrfs_extent_readonly(fs_info, disk_bytenr)) |
46bfbb5c | 7502 | goto out; |
7b2b7085 MX |
7503 | |
7504 | num_bytes = min(offset + *len, extent_end) - offset; | |
7505 | if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
7506 | u64 range_end; | |
7507 | ||
da17066c JM |
7508 | range_end = round_up(offset + num_bytes, |
7509 | root->fs_info->sectorsize) - 1; | |
7b2b7085 MX |
7510 | ret = test_range_bit(io_tree, offset, range_end, |
7511 | EXTENT_DELALLOC, 0, NULL); | |
7512 | if (ret) { | |
7513 | ret = -EAGAIN; | |
7514 | goto out; | |
7515 | } | |
7516 | } | |
7517 | ||
1bda19eb | 7518 | btrfs_release_path(path); |
46bfbb5c CM |
7519 | |
7520 | /* | |
7521 | * look for other files referencing this extent, if we | |
7522 | * find any we must cow | |
7523 | */ | |
00361589 | 7524 | |
e4c3b2dc | 7525 | ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)), |
00361589 | 7526 | key.offset - backref_offset, disk_bytenr); |
00361589 JB |
7527 | if (ret) { |
7528 | ret = 0; | |
7529 | goto out; | |
7530 | } | |
46bfbb5c CM |
7531 | |
7532 | /* | |
7533 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
7534 | * in this extent we are about to write. If there | |
7535 | * are any csums in that range we have to cow in order | |
7536 | * to keep the csums correct | |
7537 | */ | |
7538 | disk_bytenr += backref_offset; | |
7539 | disk_bytenr += offset - key.offset; | |
2ff7e61e JM |
7540 | if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes)) |
7541 | goto out; | |
46bfbb5c CM |
7542 | /* |
7543 | * all of the above have passed, it is safe to overwrite this extent | |
7544 | * without cow | |
7545 | */ | |
eb384b55 | 7546 | *len = num_bytes; |
46bfbb5c CM |
7547 | ret = 1; |
7548 | out: | |
7549 | btrfs_free_path(path); | |
7550 | return ret; | |
7551 | } | |
7552 | ||
fc4adbff AG |
7553 | bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end) |
7554 | { | |
7555 | struct radix_tree_root *root = &inode->i_mapping->page_tree; | |
e03733da | 7556 | bool found = false; |
fc4adbff AG |
7557 | void **pagep = NULL; |
7558 | struct page *page = NULL; | |
cc2b702c DS |
7559 | unsigned long start_idx; |
7560 | unsigned long end_idx; | |
fc4adbff | 7561 | |
09cbfeaf | 7562 | start_idx = start >> PAGE_SHIFT; |
fc4adbff AG |
7563 | |
7564 | /* | |
7565 | * end is the last byte in the last page. end == start is legal | |
7566 | */ | |
09cbfeaf | 7567 | end_idx = end >> PAGE_SHIFT; |
fc4adbff AG |
7568 | |
7569 | rcu_read_lock(); | |
7570 | ||
7571 | /* Most of the code in this while loop is lifted from | |
7572 | * find_get_page. It's been modified to begin searching from a | |
7573 | * page and return just the first page found in that range. If the | |
7574 | * found idx is less than or equal to the end idx then we know that | |
7575 | * a page exists. If no pages are found or if those pages are | |
7576 | * outside of the range then we're fine (yay!) */ | |
7577 | while (page == NULL && | |
7578 | radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) { | |
7579 | page = radix_tree_deref_slot(pagep); | |
7580 | if (unlikely(!page)) | |
7581 | break; | |
7582 | ||
7583 | if (radix_tree_exception(page)) { | |
809f9016 FM |
7584 | if (radix_tree_deref_retry(page)) { |
7585 | page = NULL; | |
fc4adbff | 7586 | continue; |
809f9016 | 7587 | } |
fc4adbff AG |
7588 | /* |
7589 | * Otherwise, shmem/tmpfs must be storing a swap entry | |
7590 | * here as an exceptional entry: so return it without | |
7591 | * attempting to raise page count. | |
7592 | */ | |
6fdef6d4 | 7593 | page = NULL; |
fc4adbff AG |
7594 | break; /* TODO: Is this relevant for this use case? */ |
7595 | } | |
7596 | ||
91405151 FM |
7597 | if (!page_cache_get_speculative(page)) { |
7598 | page = NULL; | |
fc4adbff | 7599 | continue; |
91405151 | 7600 | } |
fc4adbff AG |
7601 | |
7602 | /* | |
7603 | * Has the page moved? | |
7604 | * This is part of the lockless pagecache protocol. See | |
7605 | * include/linux/pagemap.h for details. | |
7606 | */ | |
7607 | if (unlikely(page != *pagep)) { | |
09cbfeaf | 7608 | put_page(page); |
fc4adbff AG |
7609 | page = NULL; |
7610 | } | |
7611 | } | |
7612 | ||
7613 | if (page) { | |
7614 | if (page->index <= end_idx) | |
7615 | found = true; | |
09cbfeaf | 7616 | put_page(page); |
fc4adbff AG |
7617 | } |
7618 | ||
7619 | rcu_read_unlock(); | |
7620 | return found; | |
7621 | } | |
7622 | ||
eb838e73 JB |
7623 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
7624 | struct extent_state **cached_state, int writing) | |
7625 | { | |
7626 | struct btrfs_ordered_extent *ordered; | |
7627 | int ret = 0; | |
7628 | ||
7629 | while (1) { | |
7630 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
ff13db41 | 7631 | cached_state); |
eb838e73 JB |
7632 | /* |
7633 | * We're concerned with the entire range that we're going to be | |
01327610 | 7634 | * doing DIO to, so we need to make sure there's no ordered |
eb838e73 JB |
7635 | * extents in this range. |
7636 | */ | |
a776c6fa | 7637 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart, |
eb838e73 JB |
7638 | lockend - lockstart + 1); |
7639 | ||
7640 | /* | |
7641 | * We need to make sure there are no buffered pages in this | |
7642 | * range either, we could have raced between the invalidate in | |
7643 | * generic_file_direct_write and locking the extent. The | |
7644 | * invalidate needs to happen so that reads after a write do not | |
7645 | * get stale data. | |
7646 | */ | |
fc4adbff AG |
7647 | if (!ordered && |
7648 | (!writing || | |
7649 | !btrfs_page_exists_in_range(inode, lockstart, lockend))) | |
eb838e73 JB |
7650 | break; |
7651 | ||
7652 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
7653 | cached_state, GFP_NOFS); | |
7654 | ||
7655 | if (ordered) { | |
ade77029 FM |
7656 | /* |
7657 | * If we are doing a DIO read and the ordered extent we | |
7658 | * found is for a buffered write, we can not wait for it | |
7659 | * to complete and retry, because if we do so we can | |
7660 | * deadlock with concurrent buffered writes on page | |
7661 | * locks. This happens only if our DIO read covers more | |
7662 | * than one extent map, if at this point has already | |
7663 | * created an ordered extent for a previous extent map | |
7664 | * and locked its range in the inode's io tree, and a | |
7665 | * concurrent write against that previous extent map's | |
7666 | * range and this range started (we unlock the ranges | |
7667 | * in the io tree only when the bios complete and | |
7668 | * buffered writes always lock pages before attempting | |
7669 | * to lock range in the io tree). | |
7670 | */ | |
7671 | if (writing || | |
7672 | test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) | |
7673 | btrfs_start_ordered_extent(inode, ordered, 1); | |
7674 | else | |
7675 | ret = -ENOTBLK; | |
eb838e73 JB |
7676 | btrfs_put_ordered_extent(ordered); |
7677 | } else { | |
eb838e73 | 7678 | /* |
b850ae14 FM |
7679 | * We could trigger writeback for this range (and wait |
7680 | * for it to complete) and then invalidate the pages for | |
7681 | * this range (through invalidate_inode_pages2_range()), | |
7682 | * but that can lead us to a deadlock with a concurrent | |
7683 | * call to readpages() (a buffered read or a defrag call | |
7684 | * triggered a readahead) on a page lock due to an | |
7685 | * ordered dio extent we created before but did not have | |
7686 | * yet a corresponding bio submitted (whence it can not | |
7687 | * complete), which makes readpages() wait for that | |
7688 | * ordered extent to complete while holding a lock on | |
7689 | * that page. | |
eb838e73 | 7690 | */ |
b850ae14 | 7691 | ret = -ENOTBLK; |
eb838e73 JB |
7692 | } |
7693 | ||
ade77029 FM |
7694 | if (ret) |
7695 | break; | |
7696 | ||
eb838e73 JB |
7697 | cond_resched(); |
7698 | } | |
7699 | ||
7700 | return ret; | |
7701 | } | |
7702 | ||
6f9994db LB |
7703 | /* The callers of this must take lock_extent() */ |
7704 | static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len, | |
7705 | u64 orig_start, u64 block_start, | |
7706 | u64 block_len, u64 orig_block_len, | |
7707 | u64 ram_bytes, int compress_type, | |
7708 | int type) | |
69ffb543 JB |
7709 | { |
7710 | struct extent_map_tree *em_tree; | |
7711 | struct extent_map *em; | |
7712 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7713 | int ret; | |
7714 | ||
6f9994db LB |
7715 | ASSERT(type == BTRFS_ORDERED_PREALLOC || |
7716 | type == BTRFS_ORDERED_COMPRESSED || | |
7717 | type == BTRFS_ORDERED_NOCOW || | |
1af4a0aa | 7718 | type == BTRFS_ORDERED_REGULAR); |
6f9994db | 7719 | |
69ffb543 JB |
7720 | em_tree = &BTRFS_I(inode)->extent_tree; |
7721 | em = alloc_extent_map(); | |
7722 | if (!em) | |
7723 | return ERR_PTR(-ENOMEM); | |
7724 | ||
7725 | em->start = start; | |
7726 | em->orig_start = orig_start; | |
7727 | em->len = len; | |
7728 | em->block_len = block_len; | |
7729 | em->block_start = block_start; | |
7730 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
b4939680 | 7731 | em->orig_block_len = orig_block_len; |
cc95bef6 | 7732 | em->ram_bytes = ram_bytes; |
70c8a91c | 7733 | em->generation = -1; |
69ffb543 | 7734 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
1af4a0aa | 7735 | if (type == BTRFS_ORDERED_PREALLOC) { |
b11e234d | 7736 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
1af4a0aa | 7737 | } else if (type == BTRFS_ORDERED_COMPRESSED) { |
6f9994db LB |
7738 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
7739 | em->compress_type = compress_type; | |
7740 | } | |
69ffb543 JB |
7741 | |
7742 | do { | |
dcdbc059 | 7743 | btrfs_drop_extent_cache(BTRFS_I(inode), em->start, |
69ffb543 JB |
7744 | em->start + em->len - 1, 0); |
7745 | write_lock(&em_tree->lock); | |
09a2a8f9 | 7746 | ret = add_extent_mapping(em_tree, em, 1); |
69ffb543 | 7747 | write_unlock(&em_tree->lock); |
6f9994db LB |
7748 | /* |
7749 | * The caller has taken lock_extent(), who could race with us | |
7750 | * to add em? | |
7751 | */ | |
69ffb543 JB |
7752 | } while (ret == -EEXIST); |
7753 | ||
7754 | if (ret) { | |
7755 | free_extent_map(em); | |
7756 | return ERR_PTR(ret); | |
7757 | } | |
7758 | ||
6f9994db | 7759 | /* em got 2 refs now, callers needs to do free_extent_map once. */ |
69ffb543 JB |
7760 | return em; |
7761 | } | |
7762 | ||
4b46fce2 JB |
7763 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
7764 | struct buffer_head *bh_result, int create) | |
7765 | { | |
0b246afa | 7766 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 7767 | struct extent_map *em; |
eb838e73 | 7768 | struct extent_state *cached_state = NULL; |
50745b0a | 7769 | struct btrfs_dio_data *dio_data = NULL; |
4b46fce2 | 7770 | u64 start = iblock << inode->i_blkbits; |
eb838e73 | 7771 | u64 lockstart, lockend; |
4b46fce2 | 7772 | u64 len = bh_result->b_size; |
eb838e73 | 7773 | int unlock_bits = EXTENT_LOCKED; |
0934856d | 7774 | int ret = 0; |
eb838e73 | 7775 | |
172a5049 | 7776 | if (create) |
3266789f | 7777 | unlock_bits |= EXTENT_DIRTY; |
172a5049 | 7778 | else |
0b246afa | 7779 | len = min_t(u64, len, fs_info->sectorsize); |
eb838e73 | 7780 | |
c329861d JB |
7781 | lockstart = start; |
7782 | lockend = start + len - 1; | |
7783 | ||
e1cbbfa5 JB |
7784 | if (current->journal_info) { |
7785 | /* | |
7786 | * Need to pull our outstanding extents and set journal_info to NULL so | |
01327610 | 7787 | * that anything that needs to check if there's a transaction doesn't get |
e1cbbfa5 JB |
7788 | * confused. |
7789 | */ | |
50745b0a | 7790 | dio_data = current->journal_info; |
e1cbbfa5 JB |
7791 | current->journal_info = NULL; |
7792 | } | |
7793 | ||
eb838e73 JB |
7794 | /* |
7795 | * If this errors out it's because we couldn't invalidate pagecache for | |
7796 | * this range and we need to fallback to buffered. | |
7797 | */ | |
9c9464cc FM |
7798 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, |
7799 | create)) { | |
7800 | ret = -ENOTBLK; | |
7801 | goto err; | |
7802 | } | |
eb838e73 | 7803 | |
fc4f21b1 | 7804 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0); |
eb838e73 JB |
7805 | if (IS_ERR(em)) { |
7806 | ret = PTR_ERR(em); | |
7807 | goto unlock_err; | |
7808 | } | |
4b46fce2 JB |
7809 | |
7810 | /* | |
7811 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
7812 | * io. INLINE is special, and we could probably kludge it in here, but | |
7813 | * it's still buffered so for safety lets just fall back to the generic | |
7814 | * buffered path. | |
7815 | * | |
7816 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
7817 | * decompress it, so there will be buffering required no matter what we | |
7818 | * do, so go ahead and fallback to buffered. | |
7819 | * | |
01327610 | 7820 | * We return -ENOTBLK because that's what makes DIO go ahead and go back |
4b46fce2 JB |
7821 | * to buffered IO. Don't blame me, this is the price we pay for using |
7822 | * the generic code. | |
7823 | */ | |
7824 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
7825 | em->block_start == EXTENT_MAP_INLINE) { | |
7826 | free_extent_map(em); | |
eb838e73 JB |
7827 | ret = -ENOTBLK; |
7828 | goto unlock_err; | |
4b46fce2 JB |
7829 | } |
7830 | ||
7831 | /* Just a good old fashioned hole, return */ | |
7832 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
7833 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
7834 | free_extent_map(em); | |
eb838e73 | 7835 | goto unlock_err; |
4b46fce2 JB |
7836 | } |
7837 | ||
7838 | /* | |
7839 | * We don't allocate a new extent in the following cases | |
7840 | * | |
7841 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
7842 | * existing extent. | |
7843 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
7844 | * just use the extent. | |
7845 | * | |
7846 | */ | |
46bfbb5c | 7847 | if (!create) { |
eb838e73 JB |
7848 | len = min(len, em->len - (start - em->start)); |
7849 | lockstart = start + len; | |
7850 | goto unlock; | |
46bfbb5c | 7851 | } |
4b46fce2 JB |
7852 | |
7853 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
7854 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
7855 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 | 7856 | int type; |
eb384b55 | 7857 | u64 block_start, orig_start, orig_block_len, ram_bytes; |
4b46fce2 JB |
7858 | |
7859 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7860 | type = BTRFS_ORDERED_PREALLOC; | |
7861 | else | |
7862 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 7863 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 7864 | block_start = em->block_start + (start - em->start); |
46bfbb5c | 7865 | |
00361589 | 7866 | if (can_nocow_extent(inode, start, &len, &orig_start, |
f78c436c | 7867 | &orig_block_len, &ram_bytes) == 1 && |
0b246afa | 7868 | btrfs_inc_nocow_writers(fs_info, block_start)) { |
5f9a8a51 | 7869 | struct extent_map *em2; |
0b901916 | 7870 | |
5f9a8a51 FM |
7871 | em2 = btrfs_create_dio_extent(inode, start, len, |
7872 | orig_start, block_start, | |
7873 | len, orig_block_len, | |
7874 | ram_bytes, type); | |
0b246afa | 7875 | btrfs_dec_nocow_writers(fs_info, block_start); |
69ffb543 JB |
7876 | if (type == BTRFS_ORDERED_PREALLOC) { |
7877 | free_extent_map(em); | |
5f9a8a51 | 7878 | em = em2; |
69ffb543 | 7879 | } |
5f9a8a51 FM |
7880 | if (em2 && IS_ERR(em2)) { |
7881 | ret = PTR_ERR(em2); | |
eb838e73 | 7882 | goto unlock_err; |
46bfbb5c | 7883 | } |
18513091 WX |
7884 | /* |
7885 | * For inode marked NODATACOW or extent marked PREALLOC, | |
7886 | * use the existing or preallocated extent, so does not | |
7887 | * need to adjust btrfs_space_info's bytes_may_use. | |
7888 | */ | |
7889 | btrfs_free_reserved_data_space_noquota(inode, | |
7890 | start, len); | |
46bfbb5c | 7891 | goto unlock; |
4b46fce2 | 7892 | } |
4b46fce2 | 7893 | } |
00361589 | 7894 | |
46bfbb5c CM |
7895 | /* |
7896 | * this will cow the extent, reset the len in case we changed | |
7897 | * it above | |
7898 | */ | |
7899 | len = bh_result->b_size; | |
70c8a91c JB |
7900 | free_extent_map(em); |
7901 | em = btrfs_new_extent_direct(inode, start, len); | |
eb838e73 JB |
7902 | if (IS_ERR(em)) { |
7903 | ret = PTR_ERR(em); | |
7904 | goto unlock_err; | |
7905 | } | |
46bfbb5c CM |
7906 | len = min(len, em->len - (start - em->start)); |
7907 | unlock: | |
4b46fce2 JB |
7908 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> |
7909 | inode->i_blkbits; | |
46bfbb5c | 7910 | bh_result->b_size = len; |
4b46fce2 JB |
7911 | bh_result->b_bdev = em->bdev; |
7912 | set_buffer_mapped(bh_result); | |
c3473e83 JB |
7913 | if (create) { |
7914 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7915 | set_buffer_new(bh_result); | |
7916 | ||
7917 | /* | |
7918 | * Need to update the i_size under the extent lock so buffered | |
7919 | * readers will get the updated i_size when we unlock. | |
7920 | */ | |
4aaedfb0 | 7921 | if (!dio_data->overwrite && start + len > i_size_read(inode)) |
c3473e83 | 7922 | i_size_write(inode, start + len); |
0934856d | 7923 | |
50745b0a | 7924 | WARN_ON(dio_data->reserve < len); |
7925 | dio_data->reserve -= len; | |
f28a4928 | 7926 | dio_data->unsubmitted_oe_range_end = start + len; |
50745b0a | 7927 | current->journal_info = dio_data; |
c3473e83 | 7928 | } |
4b46fce2 | 7929 | |
eb838e73 JB |
7930 | /* |
7931 | * In the case of write we need to clear and unlock the entire range, | |
7932 | * in the case of read we need to unlock only the end area that we | |
7933 | * aren't using if there is any left over space. | |
7934 | */ | |
24c03fa5 | 7935 | if (lockstart < lockend) { |
0934856d MX |
7936 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, |
7937 | lockend, unlock_bits, 1, 0, | |
ae0f1625 | 7938 | &cached_state); |
24c03fa5 | 7939 | } else { |
eb838e73 | 7940 | free_extent_state(cached_state); |
24c03fa5 | 7941 | } |
eb838e73 | 7942 | |
4b46fce2 JB |
7943 | free_extent_map(em); |
7944 | ||
7945 | return 0; | |
eb838e73 JB |
7946 | |
7947 | unlock_err: | |
eb838e73 | 7948 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
ae0f1625 | 7949 | unlock_bits, 1, 0, &cached_state); |
9c9464cc | 7950 | err: |
50745b0a | 7951 | if (dio_data) |
7952 | current->journal_info = dio_data; | |
eb838e73 | 7953 | return ret; |
4b46fce2 JB |
7954 | } |
7955 | ||
58efbc9f OS |
7956 | static inline blk_status_t submit_dio_repair_bio(struct inode *inode, |
7957 | struct bio *bio, | |
7958 | int mirror_num) | |
8b110e39 | 7959 | { |
2ff7e61e | 7960 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
58efbc9f | 7961 | blk_status_t ret; |
8b110e39 | 7962 | |
37226b21 | 7963 | BUG_ON(bio_op(bio) == REQ_OP_WRITE); |
8b110e39 MX |
7964 | |
7965 | bio_get(bio); | |
7966 | ||
2ff7e61e | 7967 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR); |
8b110e39 MX |
7968 | if (ret) |
7969 | goto err; | |
7970 | ||
2ff7e61e | 7971 | ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); |
8b110e39 MX |
7972 | err: |
7973 | bio_put(bio); | |
7974 | return ret; | |
7975 | } | |
7976 | ||
7977 | static int btrfs_check_dio_repairable(struct inode *inode, | |
7978 | struct bio *failed_bio, | |
7979 | struct io_failure_record *failrec, | |
7980 | int failed_mirror) | |
7981 | { | |
ab8d0fc4 | 7982 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
8b110e39 MX |
7983 | int num_copies; |
7984 | ||
ab8d0fc4 | 7985 | num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len); |
8b110e39 MX |
7986 | if (num_copies == 1) { |
7987 | /* | |
7988 | * we only have a single copy of the data, so don't bother with | |
7989 | * all the retry and error correction code that follows. no | |
7990 | * matter what the error is, it is very likely to persist. | |
7991 | */ | |
ab8d0fc4 JM |
7992 | btrfs_debug(fs_info, |
7993 | "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d", | |
7994 | num_copies, failrec->this_mirror, failed_mirror); | |
8b110e39 MX |
7995 | return 0; |
7996 | } | |
7997 | ||
7998 | failrec->failed_mirror = failed_mirror; | |
7999 | failrec->this_mirror++; | |
8000 | if (failrec->this_mirror == failed_mirror) | |
8001 | failrec->this_mirror++; | |
8002 | ||
8003 | if (failrec->this_mirror > num_copies) { | |
ab8d0fc4 JM |
8004 | btrfs_debug(fs_info, |
8005 | "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d", | |
8006 | num_copies, failrec->this_mirror, failed_mirror); | |
8b110e39 MX |
8007 | return 0; |
8008 | } | |
8009 | ||
8010 | return 1; | |
8011 | } | |
8012 | ||
58efbc9f OS |
8013 | static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio, |
8014 | struct page *page, unsigned int pgoff, | |
8015 | u64 start, u64 end, int failed_mirror, | |
8016 | bio_end_io_t *repair_endio, void *repair_arg) | |
8b110e39 MX |
8017 | { |
8018 | struct io_failure_record *failrec; | |
7870d082 JB |
8019 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
8020 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
8b110e39 MX |
8021 | struct bio *bio; |
8022 | int isector; | |
f1c77c55 | 8023 | unsigned int read_mode = 0; |
17347cec | 8024 | int segs; |
8b110e39 | 8025 | int ret; |
58efbc9f | 8026 | blk_status_t status; |
8b110e39 | 8027 | |
37226b21 | 8028 | BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE); |
8b110e39 MX |
8029 | |
8030 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
8031 | if (ret) | |
58efbc9f | 8032 | return errno_to_blk_status(ret); |
8b110e39 MX |
8033 | |
8034 | ret = btrfs_check_dio_repairable(inode, failed_bio, failrec, | |
8035 | failed_mirror); | |
8036 | if (!ret) { | |
7870d082 | 8037 | free_io_failure(failure_tree, io_tree, failrec); |
58efbc9f | 8038 | return BLK_STS_IOERR; |
8b110e39 MX |
8039 | } |
8040 | ||
17347cec LB |
8041 | segs = bio_segments(failed_bio); |
8042 | if (segs > 1 || | |
8043 | (failed_bio->bi_io_vec->bv_len > btrfs_inode_sectorsize(inode))) | |
70fd7614 | 8044 | read_mode |= REQ_FAILFAST_DEV; |
8b110e39 MX |
8045 | |
8046 | isector = start - btrfs_io_bio(failed_bio)->logical; | |
8047 | isector >>= inode->i_sb->s_blocksize_bits; | |
8048 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
2dabb324 | 8049 | pgoff, isector, repair_endio, repair_arg); |
37226b21 | 8050 | bio_set_op_attrs(bio, REQ_OP_READ, read_mode); |
8b110e39 MX |
8051 | |
8052 | btrfs_debug(BTRFS_I(inode)->root->fs_info, | |
913e1535 | 8053 | "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d", |
8b110e39 MX |
8054 | read_mode, failrec->this_mirror, failrec->in_validation); |
8055 | ||
58efbc9f OS |
8056 | status = submit_dio_repair_bio(inode, bio, failrec->this_mirror); |
8057 | if (status) { | |
7870d082 | 8058 | free_io_failure(failure_tree, io_tree, failrec); |
8b110e39 MX |
8059 | bio_put(bio); |
8060 | } | |
8061 | ||
58efbc9f | 8062 | return status; |
8b110e39 MX |
8063 | } |
8064 | ||
8065 | struct btrfs_retry_complete { | |
8066 | struct completion done; | |
8067 | struct inode *inode; | |
8068 | u64 start; | |
8069 | int uptodate; | |
8070 | }; | |
8071 | ||
4246a0b6 | 8072 | static void btrfs_retry_endio_nocsum(struct bio *bio) |
8b110e39 MX |
8073 | { |
8074 | struct btrfs_retry_complete *done = bio->bi_private; | |
7870d082 | 8075 | struct inode *inode = done->inode; |
8b110e39 | 8076 | struct bio_vec *bvec; |
7870d082 | 8077 | struct extent_io_tree *io_tree, *failure_tree; |
8b110e39 MX |
8078 | int i; |
8079 | ||
4e4cbee9 | 8080 | if (bio->bi_status) |
8b110e39 MX |
8081 | goto end; |
8082 | ||
2dabb324 | 8083 | ASSERT(bio->bi_vcnt == 1); |
7870d082 JB |
8084 | io_tree = &BTRFS_I(inode)->io_tree; |
8085 | failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
8086 | ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(inode)); | |
2dabb324 | 8087 | |
8b110e39 | 8088 | done->uptodate = 1; |
c09abff8 | 8089 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
8b110e39 | 8090 | bio_for_each_segment_all(bvec, bio, i) |
7870d082 JB |
8091 | clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree, |
8092 | io_tree, done->start, bvec->bv_page, | |
8093 | btrfs_ino(BTRFS_I(inode)), 0); | |
8b110e39 MX |
8094 | end: |
8095 | complete(&done->done); | |
8096 | bio_put(bio); | |
8097 | } | |
8098 | ||
58efbc9f OS |
8099 | static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode, |
8100 | struct btrfs_io_bio *io_bio) | |
4b46fce2 | 8101 | { |
2dabb324 | 8102 | struct btrfs_fs_info *fs_info; |
17347cec LB |
8103 | struct bio_vec bvec; |
8104 | struct bvec_iter iter; | |
8b110e39 | 8105 | struct btrfs_retry_complete done; |
4b46fce2 | 8106 | u64 start; |
2dabb324 CR |
8107 | unsigned int pgoff; |
8108 | u32 sectorsize; | |
8109 | int nr_sectors; | |
58efbc9f OS |
8110 | blk_status_t ret; |
8111 | blk_status_t err = BLK_STS_OK; | |
4b46fce2 | 8112 | |
2dabb324 | 8113 | fs_info = BTRFS_I(inode)->root->fs_info; |
da17066c | 8114 | sectorsize = fs_info->sectorsize; |
2dabb324 | 8115 | |
8b110e39 MX |
8116 | start = io_bio->logical; |
8117 | done.inode = inode; | |
17347cec | 8118 | io_bio->bio.bi_iter = io_bio->iter; |
8b110e39 | 8119 | |
17347cec LB |
8120 | bio_for_each_segment(bvec, &io_bio->bio, iter) { |
8121 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len); | |
8122 | pgoff = bvec.bv_offset; | |
2dabb324 CR |
8123 | |
8124 | next_block_or_try_again: | |
8b110e39 MX |
8125 | done.uptodate = 0; |
8126 | done.start = start; | |
8127 | init_completion(&done.done); | |
8128 | ||
17347cec | 8129 | ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page, |
2dabb324 CR |
8130 | pgoff, start, start + sectorsize - 1, |
8131 | io_bio->mirror_num, | |
8132 | btrfs_retry_endio_nocsum, &done); | |
629ebf4f LB |
8133 | if (ret) { |
8134 | err = ret; | |
8135 | goto next; | |
8136 | } | |
8b110e39 | 8137 | |
9c17f6cd | 8138 | wait_for_completion_io(&done.done); |
8b110e39 MX |
8139 | |
8140 | if (!done.uptodate) { | |
8141 | /* We might have another mirror, so try again */ | |
2dabb324 | 8142 | goto next_block_or_try_again; |
8b110e39 MX |
8143 | } |
8144 | ||
629ebf4f | 8145 | next: |
2dabb324 CR |
8146 | start += sectorsize; |
8147 | ||
97bf5a55 LB |
8148 | nr_sectors--; |
8149 | if (nr_sectors) { | |
2dabb324 | 8150 | pgoff += sectorsize; |
97bf5a55 | 8151 | ASSERT(pgoff < PAGE_SIZE); |
2dabb324 CR |
8152 | goto next_block_or_try_again; |
8153 | } | |
8b110e39 MX |
8154 | } |
8155 | ||
629ebf4f | 8156 | return err; |
8b110e39 MX |
8157 | } |
8158 | ||
4246a0b6 | 8159 | static void btrfs_retry_endio(struct bio *bio) |
8b110e39 MX |
8160 | { |
8161 | struct btrfs_retry_complete *done = bio->bi_private; | |
8162 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
7870d082 JB |
8163 | struct extent_io_tree *io_tree, *failure_tree; |
8164 | struct inode *inode = done->inode; | |
8b110e39 MX |
8165 | struct bio_vec *bvec; |
8166 | int uptodate; | |
8167 | int ret; | |
8168 | int i; | |
8169 | ||
4e4cbee9 | 8170 | if (bio->bi_status) |
8b110e39 MX |
8171 | goto end; |
8172 | ||
8173 | uptodate = 1; | |
2dabb324 | 8174 | |
2dabb324 | 8175 | ASSERT(bio->bi_vcnt == 1); |
2e949b0a | 8176 | ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(done->inode)); |
2dabb324 | 8177 | |
7870d082 JB |
8178 | io_tree = &BTRFS_I(inode)->io_tree; |
8179 | failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
8180 | ||
c09abff8 | 8181 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
8b110e39 | 8182 | bio_for_each_segment_all(bvec, bio, i) { |
7870d082 JB |
8183 | ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page, |
8184 | bvec->bv_offset, done->start, | |
8185 | bvec->bv_len); | |
8b110e39 | 8186 | if (!ret) |
7870d082 JB |
8187 | clean_io_failure(BTRFS_I(inode)->root->fs_info, |
8188 | failure_tree, io_tree, done->start, | |
8189 | bvec->bv_page, | |
8190 | btrfs_ino(BTRFS_I(inode)), | |
8191 | bvec->bv_offset); | |
8b110e39 MX |
8192 | else |
8193 | uptodate = 0; | |
8194 | } | |
8195 | ||
8196 | done->uptodate = uptodate; | |
8197 | end: | |
8198 | complete(&done->done); | |
8199 | bio_put(bio); | |
8200 | } | |
8201 | ||
4e4cbee9 CH |
8202 | static blk_status_t __btrfs_subio_endio_read(struct inode *inode, |
8203 | struct btrfs_io_bio *io_bio, blk_status_t err) | |
8b110e39 | 8204 | { |
2dabb324 | 8205 | struct btrfs_fs_info *fs_info; |
17347cec LB |
8206 | struct bio_vec bvec; |
8207 | struct bvec_iter iter; | |
8b110e39 MX |
8208 | struct btrfs_retry_complete done; |
8209 | u64 start; | |
8210 | u64 offset = 0; | |
2dabb324 CR |
8211 | u32 sectorsize; |
8212 | int nr_sectors; | |
8213 | unsigned int pgoff; | |
8214 | int csum_pos; | |
ef7cdac1 | 8215 | bool uptodate = (err == 0); |
8b110e39 | 8216 | int ret; |
58efbc9f | 8217 | blk_status_t status; |
dc380aea | 8218 | |
2dabb324 | 8219 | fs_info = BTRFS_I(inode)->root->fs_info; |
da17066c | 8220 | sectorsize = fs_info->sectorsize; |
2dabb324 | 8221 | |
58efbc9f | 8222 | err = BLK_STS_OK; |
c1dc0896 | 8223 | start = io_bio->logical; |
8b110e39 | 8224 | done.inode = inode; |
17347cec | 8225 | io_bio->bio.bi_iter = io_bio->iter; |
8b110e39 | 8226 | |
17347cec LB |
8227 | bio_for_each_segment(bvec, &io_bio->bio, iter) { |
8228 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len); | |
2dabb324 | 8229 | |
17347cec | 8230 | pgoff = bvec.bv_offset; |
2dabb324 | 8231 | next_block: |
ef7cdac1 LB |
8232 | if (uptodate) { |
8233 | csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset); | |
8234 | ret = __readpage_endio_check(inode, io_bio, csum_pos, | |
8235 | bvec.bv_page, pgoff, start, sectorsize); | |
8236 | if (likely(!ret)) | |
8237 | goto next; | |
8238 | } | |
8b110e39 MX |
8239 | try_again: |
8240 | done.uptodate = 0; | |
8241 | done.start = start; | |
8242 | init_completion(&done.done); | |
8243 | ||
58efbc9f OS |
8244 | status = dio_read_error(inode, &io_bio->bio, bvec.bv_page, |
8245 | pgoff, start, start + sectorsize - 1, | |
8246 | io_bio->mirror_num, btrfs_retry_endio, | |
8247 | &done); | |
8248 | if (status) { | |
8249 | err = status; | |
8b110e39 MX |
8250 | goto next; |
8251 | } | |
8252 | ||
9c17f6cd | 8253 | wait_for_completion_io(&done.done); |
8b110e39 MX |
8254 | |
8255 | if (!done.uptodate) { | |
8256 | /* We might have another mirror, so try again */ | |
8257 | goto try_again; | |
8258 | } | |
8259 | next: | |
2dabb324 CR |
8260 | offset += sectorsize; |
8261 | start += sectorsize; | |
8262 | ||
8263 | ASSERT(nr_sectors); | |
8264 | ||
97bf5a55 LB |
8265 | nr_sectors--; |
8266 | if (nr_sectors) { | |
2dabb324 | 8267 | pgoff += sectorsize; |
97bf5a55 | 8268 | ASSERT(pgoff < PAGE_SIZE); |
2dabb324 CR |
8269 | goto next_block; |
8270 | } | |
2c30c71b | 8271 | } |
c1dc0896 MX |
8272 | |
8273 | return err; | |
8274 | } | |
8275 | ||
4e4cbee9 CH |
8276 | static blk_status_t btrfs_subio_endio_read(struct inode *inode, |
8277 | struct btrfs_io_bio *io_bio, blk_status_t err) | |
8b110e39 MX |
8278 | { |
8279 | bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
8280 | ||
8281 | if (skip_csum) { | |
8282 | if (unlikely(err)) | |
8283 | return __btrfs_correct_data_nocsum(inode, io_bio); | |
8284 | else | |
58efbc9f | 8285 | return BLK_STS_OK; |
8b110e39 MX |
8286 | } else { |
8287 | return __btrfs_subio_endio_read(inode, io_bio, err); | |
8288 | } | |
8289 | } | |
8290 | ||
4246a0b6 | 8291 | static void btrfs_endio_direct_read(struct bio *bio) |
c1dc0896 MX |
8292 | { |
8293 | struct btrfs_dio_private *dip = bio->bi_private; | |
8294 | struct inode *inode = dip->inode; | |
8295 | struct bio *dio_bio; | |
8296 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
4e4cbee9 | 8297 | blk_status_t err = bio->bi_status; |
c1dc0896 | 8298 | |
99c4e3b9 | 8299 | if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED) |
8b110e39 | 8300 | err = btrfs_subio_endio_read(inode, io_bio, err); |
c1dc0896 | 8301 | |
4b46fce2 | 8302 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, |
d0082371 | 8303 | dip->logical_offset + dip->bytes - 1); |
9be3395b | 8304 | dio_bio = dip->dio_bio; |
4b46fce2 | 8305 | |
4b46fce2 | 8306 | kfree(dip); |
c0da7aa1 | 8307 | |
99c4e3b9 | 8308 | dio_bio->bi_status = err; |
4055351c | 8309 | dio_end_io(dio_bio); |
23ea8e5a MX |
8310 | |
8311 | if (io_bio->end_io) | |
4e4cbee9 | 8312 | io_bio->end_io(io_bio, blk_status_to_errno(err)); |
9be3395b | 8313 | bio_put(bio); |
4b46fce2 JB |
8314 | } |
8315 | ||
52427260 QW |
8316 | static void __endio_write_update_ordered(struct inode *inode, |
8317 | const u64 offset, const u64 bytes, | |
8318 | const bool uptodate) | |
4b46fce2 | 8319 | { |
0b246afa | 8320 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 8321 | struct btrfs_ordered_extent *ordered = NULL; |
52427260 QW |
8322 | struct btrfs_workqueue *wq; |
8323 | btrfs_work_func_t func; | |
14543774 FM |
8324 | u64 ordered_offset = offset; |
8325 | u64 ordered_bytes = bytes; | |
67c003f9 | 8326 | u64 last_offset; |
4b46fce2 JB |
8327 | int ret; |
8328 | ||
52427260 QW |
8329 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) { |
8330 | wq = fs_info->endio_freespace_worker; | |
8331 | func = btrfs_freespace_write_helper; | |
8332 | } else { | |
8333 | wq = fs_info->endio_write_workers; | |
8334 | func = btrfs_endio_write_helper; | |
8335 | } | |
8336 | ||
163cf09c | 8337 | again: |
67c003f9 | 8338 | last_offset = ordered_offset; |
163cf09c CM |
8339 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, |
8340 | &ordered_offset, | |
4246a0b6 | 8341 | ordered_bytes, |
14543774 | 8342 | uptodate); |
4b46fce2 | 8343 | if (!ret) |
163cf09c | 8344 | goto out_test; |
4b46fce2 | 8345 | |
52427260 QW |
8346 | btrfs_init_work(&ordered->work, func, finish_ordered_fn, NULL, NULL); |
8347 | btrfs_queue_work(wq, &ordered->work); | |
163cf09c | 8348 | out_test: |
67c003f9 NA |
8349 | /* |
8350 | * If btrfs_dec_test_ordered_pending does not find any ordered extent | |
8351 | * in the range, we can exit. | |
8352 | */ | |
8353 | if (ordered_offset == last_offset) | |
8354 | return; | |
163cf09c CM |
8355 | /* |
8356 | * our bio might span multiple ordered extents. If we haven't | |
8357 | * completed the accounting for the whole dio, go back and try again | |
8358 | */ | |
14543774 FM |
8359 | if (ordered_offset < offset + bytes) { |
8360 | ordered_bytes = offset + bytes - ordered_offset; | |
5fd02043 | 8361 | ordered = NULL; |
163cf09c CM |
8362 | goto again; |
8363 | } | |
14543774 FM |
8364 | } |
8365 | ||
8366 | static void btrfs_endio_direct_write(struct bio *bio) | |
8367 | { | |
8368 | struct btrfs_dio_private *dip = bio->bi_private; | |
8369 | struct bio *dio_bio = dip->dio_bio; | |
8370 | ||
52427260 | 8371 | __endio_write_update_ordered(dip->inode, dip->logical_offset, |
4e4cbee9 | 8372 | dip->bytes, !bio->bi_status); |
4b46fce2 | 8373 | |
4b46fce2 | 8374 | kfree(dip); |
c0da7aa1 | 8375 | |
4e4cbee9 | 8376 | dio_bio->bi_status = bio->bi_status; |
4055351c | 8377 | dio_end_io(dio_bio); |
9be3395b | 8378 | bio_put(bio); |
4b46fce2 JB |
8379 | } |
8380 | ||
8c27cb35 | 8381 | static blk_status_t __btrfs_submit_bio_start_direct_io(void *private_data, |
eaf25d93 CM |
8382 | struct bio *bio, int mirror_num, |
8383 | unsigned long bio_flags, u64 offset) | |
8384 | { | |
c6100a4b | 8385 | struct inode *inode = private_data; |
4e4cbee9 | 8386 | blk_status_t ret; |
2ff7e61e | 8387 | ret = btrfs_csum_one_bio(inode, bio, offset, 1); |
79787eaa | 8388 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
8389 | return 0; |
8390 | } | |
8391 | ||
4246a0b6 | 8392 | static void btrfs_end_dio_bio(struct bio *bio) |
e65e1535 MX |
8393 | { |
8394 | struct btrfs_dio_private *dip = bio->bi_private; | |
4e4cbee9 | 8395 | blk_status_t err = bio->bi_status; |
e65e1535 | 8396 | |
8b110e39 MX |
8397 | if (err) |
8398 | btrfs_warn(BTRFS_I(dip->inode)->root->fs_info, | |
6296b960 | 8399 | "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d", |
f85b7379 DS |
8400 | btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio), |
8401 | bio->bi_opf, | |
8b110e39 MX |
8402 | (unsigned long long)bio->bi_iter.bi_sector, |
8403 | bio->bi_iter.bi_size, err); | |
8404 | ||
8405 | if (dip->subio_endio) | |
8406 | err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err); | |
c1dc0896 MX |
8407 | |
8408 | if (err) { | |
e65e1535 MX |
8409 | dip->errors = 1; |
8410 | ||
8411 | /* | |
8412 | * before atomic variable goto zero, we must make sure | |
8413 | * dip->errors is perceived to be set. | |
8414 | */ | |
4e857c58 | 8415 | smp_mb__before_atomic(); |
e65e1535 MX |
8416 | } |
8417 | ||
8418 | /* if there are more bios still pending for this dio, just exit */ | |
8419 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
8420 | goto out; | |
8421 | ||
9be3395b | 8422 | if (dip->errors) { |
e65e1535 | 8423 | bio_io_error(dip->orig_bio); |
9be3395b | 8424 | } else { |
2dbe0c77 | 8425 | dip->dio_bio->bi_status = BLK_STS_OK; |
4246a0b6 | 8426 | bio_endio(dip->orig_bio); |
e65e1535 MX |
8427 | } |
8428 | out: | |
8429 | bio_put(bio); | |
8430 | } | |
8431 | ||
4e4cbee9 | 8432 | static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode, |
c1dc0896 MX |
8433 | struct btrfs_dio_private *dip, |
8434 | struct bio *bio, | |
8435 | u64 file_offset) | |
8436 | { | |
8437 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
8438 | struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio); | |
4e4cbee9 | 8439 | blk_status_t ret; |
c1dc0896 MX |
8440 | |
8441 | /* | |
8442 | * We load all the csum data we need when we submit | |
8443 | * the first bio to reduce the csum tree search and | |
8444 | * contention. | |
8445 | */ | |
8446 | if (dip->logical_offset == file_offset) { | |
2ff7e61e | 8447 | ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio, |
c1dc0896 MX |
8448 | file_offset); |
8449 | if (ret) | |
8450 | return ret; | |
8451 | } | |
8452 | ||
8453 | if (bio == dip->orig_bio) | |
8454 | return 0; | |
8455 | ||
8456 | file_offset -= dip->logical_offset; | |
8457 | file_offset >>= inode->i_sb->s_blocksize_bits; | |
8458 | io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset); | |
8459 | ||
8460 | return 0; | |
8461 | } | |
8462 | ||
58efbc9f OS |
8463 | static inline blk_status_t |
8464 | __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, u64 file_offset, | |
66ba772e | 8465 | int async_submit) |
e65e1535 | 8466 | { |
0b246afa | 8467 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
facc8a22 | 8468 | struct btrfs_dio_private *dip = bio->bi_private; |
37226b21 | 8469 | bool write = bio_op(bio) == REQ_OP_WRITE; |
4e4cbee9 | 8470 | blk_status_t ret; |
e65e1535 | 8471 | |
4c274bc6 | 8472 | /* Check btrfs_submit_bio_hook() for rules about async submit. */ |
b812ce28 JB |
8473 | if (async_submit) |
8474 | async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers); | |
8475 | ||
e65e1535 | 8476 | bio_get(bio); |
5fd02043 JB |
8477 | |
8478 | if (!write) { | |
0b246afa | 8479 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA); |
5fd02043 JB |
8480 | if (ret) |
8481 | goto err; | |
8482 | } | |
e65e1535 | 8483 | |
e6961cac | 8484 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) |
1ae39938 JB |
8485 | goto map; |
8486 | ||
8487 | if (write && async_submit) { | |
c6100a4b JB |
8488 | ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0, |
8489 | file_offset, inode, | |
0b246afa JM |
8490 | __btrfs_submit_bio_start_direct_io, |
8491 | __btrfs_submit_bio_done); | |
e65e1535 | 8492 | goto err; |
1ae39938 JB |
8493 | } else if (write) { |
8494 | /* | |
8495 | * If we aren't doing async submit, calculate the csum of the | |
8496 | * bio now. | |
8497 | */ | |
2ff7e61e | 8498 | ret = btrfs_csum_one_bio(inode, bio, file_offset, 1); |
1ae39938 JB |
8499 | if (ret) |
8500 | goto err; | |
23ea8e5a | 8501 | } else { |
2ff7e61e | 8502 | ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio, |
c1dc0896 | 8503 | file_offset); |
c2db1073 TI |
8504 | if (ret) |
8505 | goto err; | |
8506 | } | |
1ae39938 | 8507 | map: |
9b4a9b28 | 8508 | ret = btrfs_map_bio(fs_info, bio, 0, 0); |
e65e1535 MX |
8509 | err: |
8510 | bio_put(bio); | |
8511 | return ret; | |
8512 | } | |
8513 | ||
e6961cac | 8514 | static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip) |
e65e1535 MX |
8515 | { |
8516 | struct inode *inode = dip->inode; | |
0b246afa | 8517 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e65e1535 MX |
8518 | struct bio *bio; |
8519 | struct bio *orig_bio = dip->orig_bio; | |
4f024f37 | 8520 | u64 start_sector = orig_bio->bi_iter.bi_sector; |
e65e1535 | 8521 | u64 file_offset = dip->logical_offset; |
e65e1535 | 8522 | u64 map_length; |
1ae39938 | 8523 | int async_submit = 0; |
725130ba LB |
8524 | u64 submit_len; |
8525 | int clone_offset = 0; | |
8526 | int clone_len; | |
5f4dc8fc | 8527 | int ret; |
58efbc9f | 8528 | blk_status_t status; |
e65e1535 | 8529 | |
4f024f37 | 8530 | map_length = orig_bio->bi_iter.bi_size; |
725130ba | 8531 | submit_len = map_length; |
0b246afa JM |
8532 | ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9, |
8533 | &map_length, NULL, 0); | |
7a5c3c9b | 8534 | if (ret) |
e65e1535 | 8535 | return -EIO; |
facc8a22 | 8536 | |
725130ba | 8537 | if (map_length >= submit_len) { |
02f57c7a | 8538 | bio = orig_bio; |
c1dc0896 | 8539 | dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED; |
02f57c7a JB |
8540 | goto submit; |
8541 | } | |
8542 | ||
53b381b3 | 8543 | /* async crcs make it difficult to collect full stripe writes. */ |
1b86826d | 8544 | if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 DW |
8545 | async_submit = 0; |
8546 | else | |
8547 | async_submit = 1; | |
8548 | ||
725130ba LB |
8549 | /* bio split */ |
8550 | ASSERT(map_length <= INT_MAX); | |
02f57c7a | 8551 | atomic_inc(&dip->pending_bios); |
3c91ee69 | 8552 | do { |
725130ba | 8553 | clone_len = min_t(int, submit_len, map_length); |
02f57c7a | 8554 | |
725130ba LB |
8555 | /* |
8556 | * This will never fail as it's passing GPF_NOFS and | |
8557 | * the allocation is backed by btrfs_bioset. | |
8558 | */ | |
e477094f | 8559 | bio = btrfs_bio_clone_partial(orig_bio, clone_offset, |
725130ba LB |
8560 | clone_len); |
8561 | bio->bi_private = dip; | |
8562 | bio->bi_end_io = btrfs_end_dio_bio; | |
8563 | btrfs_io_bio(bio)->logical = file_offset; | |
8564 | ||
8565 | ASSERT(submit_len >= clone_len); | |
8566 | submit_len -= clone_len; | |
8567 | if (submit_len == 0) | |
8568 | break; | |
e65e1535 | 8569 | |
725130ba LB |
8570 | /* |
8571 | * Increase the count before we submit the bio so we know | |
8572 | * the end IO handler won't happen before we increase the | |
8573 | * count. Otherwise, the dip might get freed before we're | |
8574 | * done setting it up. | |
8575 | */ | |
8576 | atomic_inc(&dip->pending_bios); | |
e65e1535 | 8577 | |
66ba772e | 8578 | status = __btrfs_submit_dio_bio(bio, inode, file_offset, |
58efbc9f OS |
8579 | async_submit); |
8580 | if (status) { | |
725130ba LB |
8581 | bio_put(bio); |
8582 | atomic_dec(&dip->pending_bios); | |
8583 | goto out_err; | |
8584 | } | |
e65e1535 | 8585 | |
725130ba LB |
8586 | clone_offset += clone_len; |
8587 | start_sector += clone_len >> 9; | |
8588 | file_offset += clone_len; | |
5f4dc8fc | 8589 | |
725130ba LB |
8590 | map_length = submit_len; |
8591 | ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), | |
8592 | start_sector << 9, &map_length, NULL, 0); | |
8593 | if (ret) | |
8594 | goto out_err; | |
3c91ee69 | 8595 | } while (submit_len > 0); |
e65e1535 | 8596 | |
02f57c7a | 8597 | submit: |
66ba772e | 8598 | status = __btrfs_submit_dio_bio(bio, inode, file_offset, async_submit); |
58efbc9f | 8599 | if (!status) |
e65e1535 MX |
8600 | return 0; |
8601 | ||
8602 | bio_put(bio); | |
8603 | out_err: | |
8604 | dip->errors = 1; | |
8605 | /* | |
8606 | * before atomic variable goto zero, we must | |
8607 | * make sure dip->errors is perceived to be set. | |
8608 | */ | |
4e857c58 | 8609 | smp_mb__before_atomic(); |
e65e1535 MX |
8610 | if (atomic_dec_and_test(&dip->pending_bios)) |
8611 | bio_io_error(dip->orig_bio); | |
8612 | ||
8613 | /* bio_end_io() will handle error, so we needn't return it */ | |
8614 | return 0; | |
8615 | } | |
8616 | ||
8a4c1e42 MC |
8617 | static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode, |
8618 | loff_t file_offset) | |
4b46fce2 | 8619 | { |
61de718f | 8620 | struct btrfs_dio_private *dip = NULL; |
3892ac90 LB |
8621 | struct bio *bio = NULL; |
8622 | struct btrfs_io_bio *io_bio; | |
8a4c1e42 | 8623 | bool write = (bio_op(dio_bio) == REQ_OP_WRITE); |
4b46fce2 JB |
8624 | int ret = 0; |
8625 | ||
8b6c1d56 | 8626 | bio = btrfs_bio_clone(dio_bio); |
9be3395b | 8627 | |
c1dc0896 | 8628 | dip = kzalloc(sizeof(*dip), GFP_NOFS); |
4b46fce2 JB |
8629 | if (!dip) { |
8630 | ret = -ENOMEM; | |
61de718f | 8631 | goto free_ordered; |
4b46fce2 | 8632 | } |
4b46fce2 | 8633 | |
9be3395b | 8634 | dip->private = dio_bio->bi_private; |
4b46fce2 JB |
8635 | dip->inode = inode; |
8636 | dip->logical_offset = file_offset; | |
4f024f37 KO |
8637 | dip->bytes = dio_bio->bi_iter.bi_size; |
8638 | dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9; | |
3892ac90 LB |
8639 | bio->bi_private = dip; |
8640 | dip->orig_bio = bio; | |
9be3395b | 8641 | dip->dio_bio = dio_bio; |
e65e1535 | 8642 | atomic_set(&dip->pending_bios, 0); |
3892ac90 LB |
8643 | io_bio = btrfs_io_bio(bio); |
8644 | io_bio->logical = file_offset; | |
4b46fce2 | 8645 | |
c1dc0896 | 8646 | if (write) { |
3892ac90 | 8647 | bio->bi_end_io = btrfs_endio_direct_write; |
c1dc0896 | 8648 | } else { |
3892ac90 | 8649 | bio->bi_end_io = btrfs_endio_direct_read; |
c1dc0896 MX |
8650 | dip->subio_endio = btrfs_subio_endio_read; |
8651 | } | |
4b46fce2 | 8652 | |
f28a4928 FM |
8653 | /* |
8654 | * Reset the range for unsubmitted ordered extents (to a 0 length range) | |
8655 | * even if we fail to submit a bio, because in such case we do the | |
8656 | * corresponding error handling below and it must not be done a second | |
8657 | * time by btrfs_direct_IO(). | |
8658 | */ | |
8659 | if (write) { | |
8660 | struct btrfs_dio_data *dio_data = current->journal_info; | |
8661 | ||
8662 | dio_data->unsubmitted_oe_range_end = dip->logical_offset + | |
8663 | dip->bytes; | |
8664 | dio_data->unsubmitted_oe_range_start = | |
8665 | dio_data->unsubmitted_oe_range_end; | |
8666 | } | |
8667 | ||
e6961cac | 8668 | ret = btrfs_submit_direct_hook(dip); |
e65e1535 | 8669 | if (!ret) |
eaf25d93 | 8670 | return; |
9be3395b | 8671 | |
3892ac90 LB |
8672 | if (io_bio->end_io) |
8673 | io_bio->end_io(io_bio, ret); | |
9be3395b | 8674 | |
4b46fce2 JB |
8675 | free_ordered: |
8676 | /* | |
61de718f FM |
8677 | * If we arrived here it means either we failed to submit the dip |
8678 | * or we either failed to clone the dio_bio or failed to allocate the | |
8679 | * dip. If we cloned the dio_bio and allocated the dip, we can just | |
8680 | * call bio_endio against our io_bio so that we get proper resource | |
8681 | * cleanup if we fail to submit the dip, otherwise, we must do the | |
8682 | * same as btrfs_endio_direct_[write|read] because we can't call these | |
8683 | * callbacks - they require an allocated dip and a clone of dio_bio. | |
4b46fce2 | 8684 | */ |
3892ac90 | 8685 | if (bio && dip) { |
054ec2f6 | 8686 | bio_io_error(bio); |
61de718f | 8687 | /* |
3892ac90 | 8688 | * The end io callbacks free our dip, do the final put on bio |
61de718f FM |
8689 | * and all the cleanup and final put for dio_bio (through |
8690 | * dio_end_io()). | |
8691 | */ | |
8692 | dip = NULL; | |
3892ac90 | 8693 | bio = NULL; |
61de718f | 8694 | } else { |
14543774 | 8695 | if (write) |
52427260 | 8696 | __endio_write_update_ordered(inode, |
14543774 FM |
8697 | file_offset, |
8698 | dio_bio->bi_iter.bi_size, | |
52427260 | 8699 | false); |
14543774 | 8700 | else |
61de718f FM |
8701 | unlock_extent(&BTRFS_I(inode)->io_tree, file_offset, |
8702 | file_offset + dio_bio->bi_iter.bi_size - 1); | |
14543774 | 8703 | |
4e4cbee9 | 8704 | dio_bio->bi_status = BLK_STS_IOERR; |
61de718f FM |
8705 | /* |
8706 | * Releases and cleans up our dio_bio, no need to bio_put() | |
8707 | * nor bio_endio()/bio_io_error() against dio_bio. | |
8708 | */ | |
4055351c | 8709 | dio_end_io(dio_bio); |
4b46fce2 | 8710 | } |
3892ac90 LB |
8711 | if (bio) |
8712 | bio_put(bio); | |
61de718f | 8713 | kfree(dip); |
4b46fce2 JB |
8714 | } |
8715 | ||
2ff7e61e | 8716 | static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info, |
2ff7e61e | 8717 | const struct iov_iter *iter, loff_t offset) |
5a5f79b5 CM |
8718 | { |
8719 | int seg; | |
a1b75f7d | 8720 | int i; |
0b246afa | 8721 | unsigned int blocksize_mask = fs_info->sectorsize - 1; |
5a5f79b5 | 8722 | ssize_t retval = -EINVAL; |
5a5f79b5 CM |
8723 | |
8724 | if (offset & blocksize_mask) | |
8725 | goto out; | |
8726 | ||
28060d5d AV |
8727 | if (iov_iter_alignment(iter) & blocksize_mask) |
8728 | goto out; | |
a1b75f7d | 8729 | |
28060d5d | 8730 | /* If this is a write we don't need to check anymore */ |
cd27e455 | 8731 | if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter)) |
28060d5d AV |
8732 | return 0; |
8733 | /* | |
8734 | * Check to make sure we don't have duplicate iov_base's in this | |
8735 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
8736 | * when reading back. | |
8737 | */ | |
8738 | for (seg = 0; seg < iter->nr_segs; seg++) { | |
8739 | for (i = seg + 1; i < iter->nr_segs; i++) { | |
8740 | if (iter->iov[seg].iov_base == iter->iov[i].iov_base) | |
a1b75f7d JB |
8741 | goto out; |
8742 | } | |
5a5f79b5 CM |
8743 | } |
8744 | retval = 0; | |
8745 | out: | |
8746 | return retval; | |
8747 | } | |
eb838e73 | 8748 | |
c8b8e32d | 8749 | static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
16432985 | 8750 | { |
4b46fce2 JB |
8751 | struct file *file = iocb->ki_filp; |
8752 | struct inode *inode = file->f_mapping->host; | |
0b246afa | 8753 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
50745b0a | 8754 | struct btrfs_dio_data dio_data = { 0 }; |
364ecf36 | 8755 | struct extent_changeset *data_reserved = NULL; |
c8b8e32d | 8756 | loff_t offset = iocb->ki_pos; |
0934856d | 8757 | size_t count = 0; |
2e60a51e | 8758 | int flags = 0; |
38851cc1 MX |
8759 | bool wakeup = true; |
8760 | bool relock = false; | |
0934856d | 8761 | ssize_t ret; |
4b46fce2 | 8762 | |
8c70c9f8 | 8763 | if (check_direct_IO(fs_info, iter, offset)) |
5a5f79b5 | 8764 | return 0; |
3f7c579c | 8765 | |
fe0f07d0 | 8766 | inode_dio_begin(inode); |
38851cc1 | 8767 | |
0e267c44 | 8768 | /* |
41bd9ca4 MX |
8769 | * The generic stuff only does filemap_write_and_wait_range, which |
8770 | * isn't enough if we've written compressed pages to this area, so | |
8771 | * we need to flush the dirty pages again to make absolutely sure | |
8772 | * that any outstanding dirty pages are on disk. | |
0e267c44 | 8773 | */ |
a6cbcd4a | 8774 | count = iov_iter_count(iter); |
41bd9ca4 MX |
8775 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, |
8776 | &BTRFS_I(inode)->runtime_flags)) | |
9a025a08 WS |
8777 | filemap_fdatawrite_range(inode->i_mapping, offset, |
8778 | offset + count - 1); | |
0e267c44 | 8779 | |
6f673763 | 8780 | if (iov_iter_rw(iter) == WRITE) { |
38851cc1 MX |
8781 | /* |
8782 | * If the write DIO is beyond the EOF, we need update | |
8783 | * the isize, but it is protected by i_mutex. So we can | |
8784 | * not unlock the i_mutex at this case. | |
8785 | */ | |
8786 | if (offset + count <= inode->i_size) { | |
4aaedfb0 | 8787 | dio_data.overwrite = 1; |
5955102c | 8788 | inode_unlock(inode); |
38851cc1 | 8789 | relock = true; |
edf064e7 GR |
8790 | } else if (iocb->ki_flags & IOCB_NOWAIT) { |
8791 | ret = -EAGAIN; | |
8792 | goto out; | |
38851cc1 | 8793 | } |
364ecf36 QW |
8794 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, |
8795 | offset, count); | |
0934856d | 8796 | if (ret) |
38851cc1 | 8797 | goto out; |
e1cbbfa5 JB |
8798 | |
8799 | /* | |
8800 | * We need to know how many extents we reserved so that we can | |
8801 | * do the accounting properly if we go over the number we | |
8802 | * originally calculated. Abuse current->journal_info for this. | |
8803 | */ | |
da17066c | 8804 | dio_data.reserve = round_up(count, |
0b246afa | 8805 | fs_info->sectorsize); |
f28a4928 FM |
8806 | dio_data.unsubmitted_oe_range_start = (u64)offset; |
8807 | dio_data.unsubmitted_oe_range_end = (u64)offset; | |
50745b0a | 8808 | current->journal_info = &dio_data; |
97dcdea0 | 8809 | down_read(&BTRFS_I(inode)->dio_sem); |
ee39b432 DS |
8810 | } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK, |
8811 | &BTRFS_I(inode)->runtime_flags)) { | |
fe0f07d0 | 8812 | inode_dio_end(inode); |
38851cc1 MX |
8813 | flags = DIO_LOCKING | DIO_SKIP_HOLES; |
8814 | wakeup = false; | |
0934856d MX |
8815 | } |
8816 | ||
17f8c842 | 8817 | ret = __blockdev_direct_IO(iocb, inode, |
0b246afa | 8818 | fs_info->fs_devices->latest_bdev, |
c8b8e32d | 8819 | iter, btrfs_get_blocks_direct, NULL, |
17f8c842 | 8820 | btrfs_submit_direct, flags); |
6f673763 | 8821 | if (iov_iter_rw(iter) == WRITE) { |
97dcdea0 | 8822 | up_read(&BTRFS_I(inode)->dio_sem); |
e1cbbfa5 | 8823 | current->journal_info = NULL; |
ddba1bfc | 8824 | if (ret < 0 && ret != -EIOCBQUEUED) { |
50745b0a | 8825 | if (dio_data.reserve) |
bc42bda2 QW |
8826 | btrfs_delalloc_release_space(inode, data_reserved, |
8827 | offset, dio_data.reserve); | |
f28a4928 FM |
8828 | /* |
8829 | * On error we might have left some ordered extents | |
8830 | * without submitting corresponding bios for them, so | |
8831 | * cleanup them up to avoid other tasks getting them | |
8832 | * and waiting for them to complete forever. | |
8833 | */ | |
8834 | if (dio_data.unsubmitted_oe_range_start < | |
8835 | dio_data.unsubmitted_oe_range_end) | |
52427260 | 8836 | __endio_write_update_ordered(inode, |
f28a4928 FM |
8837 | dio_data.unsubmitted_oe_range_start, |
8838 | dio_data.unsubmitted_oe_range_end - | |
8839 | dio_data.unsubmitted_oe_range_start, | |
52427260 | 8840 | false); |
ddba1bfc | 8841 | } else if (ret >= 0 && (size_t)ret < count) |
bc42bda2 QW |
8842 | btrfs_delalloc_release_space(inode, data_reserved, |
8843 | offset, count - (size_t)ret); | |
69fe2d75 | 8844 | btrfs_delalloc_release_extents(BTRFS_I(inode), count); |
0934856d | 8845 | } |
38851cc1 | 8846 | out: |
2e60a51e | 8847 | if (wakeup) |
fe0f07d0 | 8848 | inode_dio_end(inode); |
38851cc1 | 8849 | if (relock) |
5955102c | 8850 | inode_lock(inode); |
0934856d | 8851 | |
364ecf36 | 8852 | extent_changeset_free(data_reserved); |
0934856d | 8853 | return ret; |
16432985 CM |
8854 | } |
8855 | ||
05dadc09 TI |
8856 | #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC) |
8857 | ||
1506fcc8 YS |
8858 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
8859 | __u64 start, __u64 len) | |
8860 | { | |
05dadc09 TI |
8861 | int ret; |
8862 | ||
8863 | ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS); | |
8864 | if (ret) | |
8865 | return ret; | |
8866 | ||
ec29ed5b | 8867 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
8868 | } |
8869 | ||
a52d9a80 | 8870 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 8871 | { |
d1310b2e CM |
8872 | struct extent_io_tree *tree; |
8873 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 8874 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 8875 | } |
1832a6d5 | 8876 | |
a52d9a80 | 8877 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 8878 | { |
d1310b2e | 8879 | struct extent_io_tree *tree; |
be7bd730 JB |
8880 | struct inode *inode = page->mapping->host; |
8881 | int ret; | |
b888db2b CM |
8882 | |
8883 | if (current->flags & PF_MEMALLOC) { | |
8884 | redirty_page_for_writepage(wbc, page); | |
8885 | unlock_page(page); | |
8886 | return 0; | |
8887 | } | |
be7bd730 JB |
8888 | |
8889 | /* | |
8890 | * If we are under memory pressure we will call this directly from the | |
8891 | * VM, we need to make sure we have the inode referenced for the ordered | |
8892 | * extent. If not just return like we didn't do anything. | |
8893 | */ | |
8894 | if (!igrab(inode)) { | |
8895 | redirty_page_for_writepage(wbc, page); | |
8896 | return AOP_WRITEPAGE_ACTIVATE; | |
8897 | } | |
d1310b2e | 8898 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
be7bd730 JB |
8899 | ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
8900 | btrfs_add_delayed_iput(inode); | |
8901 | return ret; | |
9ebefb18 CM |
8902 | } |
8903 | ||
48a3b636 ES |
8904 | static int btrfs_writepages(struct address_space *mapping, |
8905 | struct writeback_control *wbc) | |
b293f02e | 8906 | { |
d1310b2e | 8907 | struct extent_io_tree *tree; |
771ed689 | 8908 | |
d1310b2e | 8909 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
8910 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
8911 | } | |
8912 | ||
3ab2fb5a CM |
8913 | static int |
8914 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
8915 | struct list_head *pages, unsigned nr_pages) | |
8916 | { | |
d1310b2e CM |
8917 | struct extent_io_tree *tree; |
8918 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
8919 | return extent_readpages(tree, mapping, pages, nr_pages, |
8920 | btrfs_get_extent); | |
8921 | } | |
e6dcd2dc | 8922 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 8923 | { |
d1310b2e CM |
8924 | struct extent_io_tree *tree; |
8925 | struct extent_map_tree *map; | |
a52d9a80 | 8926 | int ret; |
8c2383c3 | 8927 | |
d1310b2e CM |
8928 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
8929 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 8930 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
8931 | if (ret == 1) { |
8932 | ClearPagePrivate(page); | |
8933 | set_page_private(page, 0); | |
09cbfeaf | 8934 | put_page(page); |
39279cc3 | 8935 | } |
a52d9a80 | 8936 | return ret; |
39279cc3 CM |
8937 | } |
8938 | ||
e6dcd2dc CM |
8939 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
8940 | { | |
98509cfc CM |
8941 | if (PageWriteback(page) || PageDirty(page)) |
8942 | return 0; | |
3ba7ab22 | 8943 | return __btrfs_releasepage(page, gfp_flags); |
e6dcd2dc CM |
8944 | } |
8945 | ||
d47992f8 LC |
8946 | static void btrfs_invalidatepage(struct page *page, unsigned int offset, |
8947 | unsigned int length) | |
39279cc3 | 8948 | { |
5fd02043 | 8949 | struct inode *inode = page->mapping->host; |
d1310b2e | 8950 | struct extent_io_tree *tree; |
e6dcd2dc | 8951 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 8952 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 8953 | u64 page_start = page_offset(page); |
09cbfeaf | 8954 | u64 page_end = page_start + PAGE_SIZE - 1; |
dbfdb6d1 CR |
8955 | u64 start; |
8956 | u64 end; | |
131e404a | 8957 | int inode_evicting = inode->i_state & I_FREEING; |
39279cc3 | 8958 | |
8b62b72b CM |
8959 | /* |
8960 | * we have the page locked, so new writeback can't start, | |
8961 | * and the dirty bit won't be cleared while we are here. | |
8962 | * | |
8963 | * Wait for IO on this page so that we can safely clear | |
8964 | * the PagePrivate2 bit and do ordered accounting | |
8965 | */ | |
e6dcd2dc | 8966 | wait_on_page_writeback(page); |
8b62b72b | 8967 | |
5fd02043 | 8968 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
8969 | if (offset) { |
8970 | btrfs_releasepage(page, GFP_NOFS); | |
8971 | return; | |
8972 | } | |
131e404a FDBM |
8973 | |
8974 | if (!inode_evicting) | |
ff13db41 | 8975 | lock_extent_bits(tree, page_start, page_end, &cached_state); |
dbfdb6d1 CR |
8976 | again: |
8977 | start = page_start; | |
a776c6fa | 8978 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start, |
dbfdb6d1 | 8979 | page_end - start + 1); |
e6dcd2dc | 8980 | if (ordered) { |
dbfdb6d1 | 8981 | end = min(page_end, ordered->file_offset + ordered->len - 1); |
eb84ae03 CM |
8982 | /* |
8983 | * IO on this page will never be started, so we need | |
8984 | * to account for any ordered extents now | |
8985 | */ | |
131e404a | 8986 | if (!inode_evicting) |
dbfdb6d1 | 8987 | clear_extent_bit(tree, start, end, |
131e404a | 8988 | EXTENT_DIRTY | EXTENT_DELALLOC | |
a7e3b975 | 8989 | EXTENT_DELALLOC_NEW | |
131e404a | 8990 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
ae0f1625 | 8991 | EXTENT_DEFRAG, 1, 0, &cached_state); |
8b62b72b CM |
8992 | /* |
8993 | * whoever cleared the private bit is responsible | |
8994 | * for the finish_ordered_io | |
8995 | */ | |
77cef2ec JB |
8996 | if (TestClearPagePrivate2(page)) { |
8997 | struct btrfs_ordered_inode_tree *tree; | |
8998 | u64 new_len; | |
8999 | ||
9000 | tree = &BTRFS_I(inode)->ordered_tree; | |
9001 | ||
9002 | spin_lock_irq(&tree->lock); | |
9003 | set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags); | |
dbfdb6d1 | 9004 | new_len = start - ordered->file_offset; |
77cef2ec JB |
9005 | if (new_len < ordered->truncated_len) |
9006 | ordered->truncated_len = new_len; | |
9007 | spin_unlock_irq(&tree->lock); | |
9008 | ||
9009 | if (btrfs_dec_test_ordered_pending(inode, &ordered, | |
dbfdb6d1 CR |
9010 | start, |
9011 | end - start + 1, 1)) | |
77cef2ec | 9012 | btrfs_finish_ordered_io(ordered); |
8b62b72b | 9013 | } |
e6dcd2dc | 9014 | btrfs_put_ordered_extent(ordered); |
131e404a FDBM |
9015 | if (!inode_evicting) { |
9016 | cached_state = NULL; | |
dbfdb6d1 | 9017 | lock_extent_bits(tree, start, end, |
131e404a FDBM |
9018 | &cached_state); |
9019 | } | |
dbfdb6d1 CR |
9020 | |
9021 | start = end + 1; | |
9022 | if (start < page_end) | |
9023 | goto again; | |
131e404a FDBM |
9024 | } |
9025 | ||
b9d0b389 QW |
9026 | /* |
9027 | * Qgroup reserved space handler | |
9028 | * Page here will be either | |
9029 | * 1) Already written to disk | |
9030 | * In this case, its reserved space is released from data rsv map | |
9031 | * and will be freed by delayed_ref handler finally. | |
9032 | * So even we call qgroup_free_data(), it won't decrease reserved | |
9033 | * space. | |
9034 | * 2) Not written to disk | |
0b34c261 GR |
9035 | * This means the reserved space should be freed here. However, |
9036 | * if a truncate invalidates the page (by clearing PageDirty) | |
9037 | * and the page is accounted for while allocating extent | |
9038 | * in btrfs_check_data_free_space() we let delayed_ref to | |
9039 | * free the entire extent. | |
b9d0b389 | 9040 | */ |
0b34c261 | 9041 | if (PageDirty(page)) |
bc42bda2 | 9042 | btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE); |
131e404a FDBM |
9043 | if (!inode_evicting) { |
9044 | clear_extent_bit(tree, page_start, page_end, | |
9045 | EXTENT_LOCKED | EXTENT_DIRTY | | |
a7e3b975 FM |
9046 | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW | |
9047 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1, | |
ae0f1625 | 9048 | &cached_state); |
131e404a FDBM |
9049 | |
9050 | __btrfs_releasepage(page, GFP_NOFS); | |
e6dcd2dc | 9051 | } |
e6dcd2dc | 9052 | |
4a096752 | 9053 | ClearPageChecked(page); |
9ad6b7bc | 9054 | if (PagePrivate(page)) { |
9ad6b7bc CM |
9055 | ClearPagePrivate(page); |
9056 | set_page_private(page, 0); | |
09cbfeaf | 9057 | put_page(page); |
9ad6b7bc | 9058 | } |
39279cc3 CM |
9059 | } |
9060 | ||
9ebefb18 CM |
9061 | /* |
9062 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
9063 | * called from a page fault handler when a page is first dirtied. Hence we must | |
9064 | * be careful to check for EOF conditions here. We set the page up correctly | |
9065 | * for a written page which means we get ENOSPC checking when writing into | |
9066 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
9067 | * support these features. | |
9068 | * | |
9069 | * We are not allowed to take the i_mutex here so we have to play games to | |
9070 | * protect against truncate races as the page could now be beyond EOF. Because | |
9071 | * vmtruncate() writes the inode size before removing pages, once we have the | |
9072 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
9073 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
9074 | * unlock the page. | |
9075 | */ | |
11bac800 | 9076 | int btrfs_page_mkwrite(struct vm_fault *vmf) |
9ebefb18 | 9077 | { |
c2ec175c | 9078 | struct page *page = vmf->page; |
11bac800 | 9079 | struct inode *inode = file_inode(vmf->vma->vm_file); |
0b246afa | 9080 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc CM |
9081 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
9082 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 9083 | struct extent_state *cached_state = NULL; |
364ecf36 | 9084 | struct extent_changeset *data_reserved = NULL; |
e6dcd2dc CM |
9085 | char *kaddr; |
9086 | unsigned long zero_start; | |
9ebefb18 | 9087 | loff_t size; |
1832a6d5 | 9088 | int ret; |
9998eb70 | 9089 | int reserved = 0; |
d0b7da88 | 9090 | u64 reserved_space; |
a52d9a80 | 9091 | u64 page_start; |
e6dcd2dc | 9092 | u64 page_end; |
d0b7da88 CR |
9093 | u64 end; |
9094 | ||
09cbfeaf | 9095 | reserved_space = PAGE_SIZE; |
9ebefb18 | 9096 | |
b2b5ef5c | 9097 | sb_start_pagefault(inode->i_sb); |
df480633 | 9098 | page_start = page_offset(page); |
09cbfeaf | 9099 | page_end = page_start + PAGE_SIZE - 1; |
d0b7da88 | 9100 | end = page_end; |
df480633 | 9101 | |
d0b7da88 CR |
9102 | /* |
9103 | * Reserving delalloc space after obtaining the page lock can lead to | |
9104 | * deadlock. For example, if a dirty page is locked by this function | |
9105 | * and the call to btrfs_delalloc_reserve_space() ends up triggering | |
9106 | * dirty page write out, then the btrfs_writepage() function could | |
9107 | * end up waiting indefinitely to get a lock on the page currently | |
9108 | * being processed by btrfs_page_mkwrite() function. | |
9109 | */ | |
364ecf36 | 9110 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start, |
d0b7da88 | 9111 | reserved_space); |
9998eb70 | 9112 | if (!ret) { |
11bac800 | 9113 | ret = file_update_time(vmf->vma->vm_file); |
9998eb70 CM |
9114 | reserved = 1; |
9115 | } | |
56a76f82 NP |
9116 | if (ret) { |
9117 | if (ret == -ENOMEM) | |
9118 | ret = VM_FAULT_OOM; | |
9119 | else /* -ENOSPC, -EIO, etc */ | |
9120 | ret = VM_FAULT_SIGBUS; | |
9998eb70 CM |
9121 | if (reserved) |
9122 | goto out; | |
9123 | goto out_noreserve; | |
56a76f82 | 9124 | } |
1832a6d5 | 9125 | |
56a76f82 | 9126 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 9127 | again: |
9ebefb18 | 9128 | lock_page(page); |
9ebefb18 | 9129 | size = i_size_read(inode); |
a52d9a80 | 9130 | |
9ebefb18 | 9131 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 9132 | (page_start >= size)) { |
9ebefb18 CM |
9133 | /* page got truncated out from underneath us */ |
9134 | goto out_unlock; | |
9135 | } | |
e6dcd2dc CM |
9136 | wait_on_page_writeback(page); |
9137 | ||
ff13db41 | 9138 | lock_extent_bits(io_tree, page_start, page_end, &cached_state); |
e6dcd2dc CM |
9139 | set_page_extent_mapped(page); |
9140 | ||
eb84ae03 CM |
9141 | /* |
9142 | * we can't set the delalloc bits if there are pending ordered | |
9143 | * extents. Drop our locks and wait for them to finish | |
9144 | */ | |
a776c6fa NB |
9145 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start, |
9146 | PAGE_SIZE); | |
e6dcd2dc | 9147 | if (ordered) { |
2ac55d41 JB |
9148 | unlock_extent_cached(io_tree, page_start, page_end, |
9149 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 9150 | unlock_page(page); |
eb84ae03 | 9151 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
9152 | btrfs_put_ordered_extent(ordered); |
9153 | goto again; | |
9154 | } | |
9155 | ||
09cbfeaf | 9156 | if (page->index == ((size - 1) >> PAGE_SHIFT)) { |
da17066c | 9157 | reserved_space = round_up(size - page_start, |
0b246afa | 9158 | fs_info->sectorsize); |
09cbfeaf | 9159 | if (reserved_space < PAGE_SIZE) { |
d0b7da88 | 9160 | end = page_start + reserved_space - 1; |
bc42bda2 QW |
9161 | btrfs_delalloc_release_space(inode, data_reserved, |
9162 | page_start, PAGE_SIZE - reserved_space); | |
d0b7da88 CR |
9163 | } |
9164 | } | |
9165 | ||
fbf19087 | 9166 | /* |
5416034f LB |
9167 | * page_mkwrite gets called when the page is firstly dirtied after it's |
9168 | * faulted in, but write(2) could also dirty a page and set delalloc | |
9169 | * bits, thus in this case for space account reason, we still need to | |
9170 | * clear any delalloc bits within this page range since we have to | |
9171 | * reserve data&meta space before lock_page() (see above comments). | |
fbf19087 | 9172 | */ |
d0b7da88 | 9173 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end, |
9e8a4a8b LB |
9174 | EXTENT_DIRTY | EXTENT_DELALLOC | |
9175 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
ae0f1625 | 9176 | 0, 0, &cached_state); |
fbf19087 | 9177 | |
e3b8a485 | 9178 | ret = btrfs_set_extent_delalloc(inode, page_start, end, 0, |
ba8b04c1 | 9179 | &cached_state, 0); |
9ed74f2d | 9180 | if (ret) { |
2ac55d41 JB |
9181 | unlock_extent_cached(io_tree, page_start, page_end, |
9182 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
9183 | ret = VM_FAULT_SIGBUS; |
9184 | goto out_unlock; | |
9185 | } | |
e6dcd2dc | 9186 | ret = 0; |
9ebefb18 CM |
9187 | |
9188 | /* page is wholly or partially inside EOF */ | |
09cbfeaf KS |
9189 | if (page_start + PAGE_SIZE > size) |
9190 | zero_start = size & ~PAGE_MASK; | |
9ebefb18 | 9191 | else |
09cbfeaf | 9192 | zero_start = PAGE_SIZE; |
9ebefb18 | 9193 | |
09cbfeaf | 9194 | if (zero_start != PAGE_SIZE) { |
e6dcd2dc | 9195 | kaddr = kmap(page); |
09cbfeaf | 9196 | memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start); |
e6dcd2dc CM |
9197 | flush_dcache_page(page); |
9198 | kunmap(page); | |
9199 | } | |
247e743c | 9200 | ClearPageChecked(page); |
e6dcd2dc | 9201 | set_page_dirty(page); |
50a9b214 | 9202 | SetPageUptodate(page); |
5a3f23d5 | 9203 | |
0b246afa | 9204 | BTRFS_I(inode)->last_trans = fs_info->generation; |
257c62e1 | 9205 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; |
46d8bc34 | 9206 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit; |
257c62e1 | 9207 | |
2ac55d41 | 9208 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
9209 | |
9210 | out_unlock: | |
b2b5ef5c | 9211 | if (!ret) { |
8b62f87b | 9212 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); |
b2b5ef5c | 9213 | sb_end_pagefault(inode->i_sb); |
364ecf36 | 9214 | extent_changeset_free(data_reserved); |
50a9b214 | 9215 | return VM_FAULT_LOCKED; |
b2b5ef5c | 9216 | } |
9ebefb18 | 9217 | unlock_page(page); |
1832a6d5 | 9218 | out: |
8b62f87b | 9219 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); |
bc42bda2 QW |
9220 | btrfs_delalloc_release_space(inode, data_reserved, page_start, |
9221 | reserved_space); | |
9998eb70 | 9222 | out_noreserve: |
b2b5ef5c | 9223 | sb_end_pagefault(inode->i_sb); |
364ecf36 | 9224 | extent_changeset_free(data_reserved); |
9ebefb18 CM |
9225 | return ret; |
9226 | } | |
9227 | ||
a41ad394 | 9228 | static int btrfs_truncate(struct inode *inode) |
39279cc3 | 9229 | { |
0b246afa | 9230 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 9231 | struct btrfs_root *root = BTRFS_I(inode)->root; |
fcb80c2a | 9232 | struct btrfs_block_rsv *rsv; |
a71754fc | 9233 | int ret = 0; |
3893e33b | 9234 | int err = 0; |
39279cc3 | 9235 | struct btrfs_trans_handle *trans; |
0b246afa JM |
9236 | u64 mask = fs_info->sectorsize - 1; |
9237 | u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1); | |
39279cc3 | 9238 | |
0ef8b726 JB |
9239 | ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask), |
9240 | (u64)-1); | |
9241 | if (ret) | |
9242 | return ret; | |
39279cc3 | 9243 | |
fcb80c2a | 9244 | /* |
01327610 | 9245 | * Yes ladies and gentlemen, this is indeed ugly. The fact is we have |
fcb80c2a JB |
9246 | * 3 things going on here |
9247 | * | |
9248 | * 1) We need to reserve space for our orphan item and the space to | |
9249 | * delete our orphan item. Lord knows we don't want to have a dangling | |
9250 | * orphan item because we didn't reserve space to remove it. | |
9251 | * | |
9252 | * 2) We need to reserve space to update our inode. | |
9253 | * | |
9254 | * 3) We need to have something to cache all the space that is going to | |
9255 | * be free'd up by the truncate operation, but also have some slack | |
9256 | * space reserved in case it uses space during the truncate (thank you | |
9257 | * very much snapshotting). | |
9258 | * | |
01327610 | 9259 | * And we need these to all be separate. The fact is we can use a lot of |
fcb80c2a | 9260 | * space doing the truncate, and we have no earthly idea how much space |
01327610 | 9261 | * we will use, so we need the truncate reservation to be separate so it |
fcb80c2a JB |
9262 | * doesn't end up using space reserved for updating the inode or |
9263 | * removing the orphan item. We also need to be able to stop the | |
9264 | * transaction and start a new one, which means we need to be able to | |
9265 | * update the inode several times, and we have no idea of knowing how | |
9266 | * many times that will be, so we can't just reserve 1 item for the | |
01327610 | 9267 | * entirety of the operation, so that has to be done separately as well. |
fcb80c2a JB |
9268 | * Then there is the orphan item, which does indeed need to be held on |
9269 | * to for the whole operation, and we need nobody to touch this reserved | |
9270 | * space except the orphan code. | |
9271 | * | |
9272 | * So that leaves us with | |
9273 | * | |
9274 | * 1) root->orphan_block_rsv - for the orphan deletion. | |
9275 | * 2) rsv - for the truncate reservation, which we will steal from the | |
9276 | * transaction reservation. | |
9277 | * 3) fs_info->trans_block_rsv - this will have 1 items worth left for | |
9278 | * updating the inode. | |
9279 | */ | |
2ff7e61e | 9280 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
9281 | if (!rsv) |
9282 | return -ENOMEM; | |
4a338542 | 9283 | rsv->size = min_size; |
ca7e70f5 | 9284 | rsv->failfast = 1; |
f0cd846e | 9285 | |
907cbceb | 9286 | /* |
07127184 | 9287 | * 1 for the truncate slack space |
907cbceb JB |
9288 | * 1 for updating the inode. |
9289 | */ | |
f3fe820c | 9290 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a JB |
9291 | if (IS_ERR(trans)) { |
9292 | err = PTR_ERR(trans); | |
9293 | goto out; | |
9294 | } | |
f0cd846e | 9295 | |
907cbceb | 9296 | /* Migrate the slack space for the truncate to our reserve */ |
0b246afa | 9297 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv, |
25d609f8 | 9298 | min_size, 0); |
fcb80c2a | 9299 | BUG_ON(ret); |
f0cd846e | 9300 | |
5dc562c5 JB |
9301 | /* |
9302 | * So if we truncate and then write and fsync we normally would just | |
9303 | * write the extents that changed, which is a problem if we need to | |
9304 | * first truncate that entire inode. So set this flag so we write out | |
9305 | * all of the extents in the inode to the sync log so we're completely | |
9306 | * safe. | |
9307 | */ | |
9308 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
ca7e70f5 | 9309 | trans->block_rsv = rsv; |
907cbceb | 9310 | |
8082510e YZ |
9311 | while (1) { |
9312 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
9313 | inode->i_size, | |
9314 | BTRFS_EXTENT_DATA_KEY); | |
ddfae63c | 9315 | trans->block_rsv = &fs_info->trans_block_rsv; |
28ed1345 | 9316 | if (ret != -ENOSPC && ret != -EAGAIN) { |
3893e33b | 9317 | err = ret; |
8082510e | 9318 | break; |
3893e33b | 9319 | } |
39279cc3 | 9320 | |
8082510e | 9321 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
9322 | if (ret) { |
9323 | err = ret; | |
9324 | break; | |
9325 | } | |
ca7e70f5 | 9326 | |
3a45bb20 | 9327 | btrfs_end_transaction(trans); |
2ff7e61e | 9328 | btrfs_btree_balance_dirty(fs_info); |
ca7e70f5 JB |
9329 | |
9330 | trans = btrfs_start_transaction(root, 2); | |
9331 | if (IS_ERR(trans)) { | |
9332 | ret = err = PTR_ERR(trans); | |
9333 | trans = NULL; | |
9334 | break; | |
9335 | } | |
9336 | ||
47b5d646 | 9337 | btrfs_block_rsv_release(fs_info, rsv, -1); |
0b246afa | 9338 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, |
25d609f8 | 9339 | rsv, min_size, 0); |
ca7e70f5 JB |
9340 | BUG_ON(ret); /* shouldn't happen */ |
9341 | trans->block_rsv = rsv; | |
8082510e YZ |
9342 | } |
9343 | ||
ddfae63c JB |
9344 | /* |
9345 | * We can't call btrfs_truncate_block inside a trans handle as we could | |
9346 | * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know | |
9347 | * we've truncated everything except the last little bit, and can do | |
9348 | * btrfs_truncate_block and then update the disk_i_size. | |
9349 | */ | |
9350 | if (ret == NEED_TRUNCATE_BLOCK) { | |
9351 | btrfs_end_transaction(trans); | |
9352 | btrfs_btree_balance_dirty(fs_info); | |
9353 | ||
9354 | ret = btrfs_truncate_block(inode, inode->i_size, 0, 0); | |
9355 | if (ret) | |
9356 | goto out; | |
9357 | trans = btrfs_start_transaction(root, 1); | |
9358 | if (IS_ERR(trans)) { | |
9359 | ret = PTR_ERR(trans); | |
9360 | goto out; | |
9361 | } | |
9362 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); | |
9363 | } | |
9364 | ||
8082510e | 9365 | if (ret == 0 && inode->i_nlink > 0) { |
fcb80c2a | 9366 | trans->block_rsv = root->orphan_block_rsv; |
3d6ae7bb | 9367 | ret = btrfs_orphan_del(trans, BTRFS_I(inode)); |
3893e33b JB |
9368 | if (ret) |
9369 | err = ret; | |
8082510e YZ |
9370 | } |
9371 | ||
917c16b2 | 9372 | if (trans) { |
0b246afa | 9373 | trans->block_rsv = &fs_info->trans_block_rsv; |
917c16b2 CM |
9374 | ret = btrfs_update_inode(trans, root, inode); |
9375 | if (ret && !err) | |
9376 | err = ret; | |
7b128766 | 9377 | |
3a45bb20 | 9378 | ret = btrfs_end_transaction(trans); |
2ff7e61e | 9379 | btrfs_btree_balance_dirty(fs_info); |
917c16b2 | 9380 | } |
fcb80c2a | 9381 | out: |
2ff7e61e | 9382 | btrfs_free_block_rsv(fs_info, rsv); |
fcb80c2a | 9383 | |
3893e33b JB |
9384 | if (ret && !err) |
9385 | err = ret; | |
a41ad394 | 9386 | |
3893e33b | 9387 | return err; |
39279cc3 CM |
9388 | } |
9389 | ||
d352ac68 CM |
9390 | /* |
9391 | * create a new subvolume directory/inode (helper for the ioctl). | |
9392 | */ | |
d2fb3437 | 9393 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
63541927 FDBM |
9394 | struct btrfs_root *new_root, |
9395 | struct btrfs_root *parent_root, | |
9396 | u64 new_dirid) | |
39279cc3 | 9397 | { |
39279cc3 | 9398 | struct inode *inode; |
76dda93c | 9399 | int err; |
00e4e6b3 | 9400 | u64 index = 0; |
39279cc3 | 9401 | |
12fc9d09 FA |
9402 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
9403 | new_dirid, new_dirid, | |
9404 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
9405 | &index); | |
54aa1f4d | 9406 | if (IS_ERR(inode)) |
f46b5a66 | 9407 | return PTR_ERR(inode); |
39279cc3 CM |
9408 | inode->i_op = &btrfs_dir_inode_operations; |
9409 | inode->i_fop = &btrfs_dir_file_operations; | |
9410 | ||
bfe86848 | 9411 | set_nlink(inode, 1); |
6ef06d27 | 9412 | btrfs_i_size_write(BTRFS_I(inode), 0); |
b0d5d10f | 9413 | unlock_new_inode(inode); |
3b96362c | 9414 | |
63541927 FDBM |
9415 | err = btrfs_subvol_inherit_props(trans, new_root, parent_root); |
9416 | if (err) | |
9417 | btrfs_err(new_root->fs_info, | |
351fd353 | 9418 | "error inheriting subvolume %llu properties: %d", |
63541927 FDBM |
9419 | new_root->root_key.objectid, err); |
9420 | ||
76dda93c | 9421 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 9422 | |
76dda93c | 9423 | iput(inode); |
ce598979 | 9424 | return err; |
39279cc3 CM |
9425 | } |
9426 | ||
39279cc3 CM |
9427 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
9428 | { | |
69fe2d75 | 9429 | struct btrfs_fs_info *fs_info = btrfs_sb(sb); |
39279cc3 | 9430 | struct btrfs_inode *ei; |
2ead6ae7 | 9431 | struct inode *inode; |
39279cc3 CM |
9432 | |
9433 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
9434 | if (!ei) | |
9435 | return NULL; | |
2ead6ae7 YZ |
9436 | |
9437 | ei->root = NULL; | |
2ead6ae7 | 9438 | ei->generation = 0; |
15ee9bc7 | 9439 | ei->last_trans = 0; |
257c62e1 | 9440 | ei->last_sub_trans = 0; |
e02119d5 | 9441 | ei->logged_trans = 0; |
2ead6ae7 | 9442 | ei->delalloc_bytes = 0; |
a7e3b975 | 9443 | ei->new_delalloc_bytes = 0; |
47059d93 | 9444 | ei->defrag_bytes = 0; |
2ead6ae7 YZ |
9445 | ei->disk_i_size = 0; |
9446 | ei->flags = 0; | |
7709cde3 | 9447 | ei->csum_bytes = 0; |
2ead6ae7 | 9448 | ei->index_cnt = (u64)-1; |
67de1176 | 9449 | ei->dir_index = 0; |
2ead6ae7 | 9450 | ei->last_unlink_trans = 0; |
46d8bc34 | 9451 | ei->last_log_commit = 0; |
8089fe62 | 9452 | ei->delayed_iput_count = 0; |
2ead6ae7 | 9453 | |
9e0baf60 JB |
9454 | spin_lock_init(&ei->lock); |
9455 | ei->outstanding_extents = 0; | |
69fe2d75 JB |
9456 | if (sb->s_magic != BTRFS_TEST_MAGIC) |
9457 | btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv, | |
9458 | BTRFS_BLOCK_RSV_DELALLOC); | |
72ac3c0d | 9459 | ei->runtime_flags = 0; |
b52aa8c9 | 9460 | ei->prop_compress = BTRFS_COMPRESS_NONE; |
eec63c65 | 9461 | ei->defrag_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 9462 | |
16cdcec7 MX |
9463 | ei->delayed_node = NULL; |
9464 | ||
9cc97d64 | 9465 | ei->i_otime.tv_sec = 0; |
9466 | ei->i_otime.tv_nsec = 0; | |
9467 | ||
2ead6ae7 | 9468 | inode = &ei->vfs_inode; |
a8067e02 | 9469 | extent_map_tree_init(&ei->extent_tree); |
c6100a4b JB |
9470 | extent_io_tree_init(&ei->io_tree, inode); |
9471 | extent_io_tree_init(&ei->io_failure_tree, inode); | |
0b32f4bb JB |
9472 | ei->io_tree.track_uptodate = 1; |
9473 | ei->io_failure_tree.track_uptodate = 1; | |
b812ce28 | 9474 | atomic_set(&ei->sync_writers, 0); |
2ead6ae7 | 9475 | mutex_init(&ei->log_mutex); |
f248679e | 9476 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 9477 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 9478 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
8089fe62 | 9479 | INIT_LIST_HEAD(&ei->delayed_iput); |
2ead6ae7 | 9480 | RB_CLEAR_NODE(&ei->rb_node); |
5f9a8a51 | 9481 | init_rwsem(&ei->dio_sem); |
2ead6ae7 YZ |
9482 | |
9483 | return inode; | |
39279cc3 CM |
9484 | } |
9485 | ||
aaedb55b JB |
9486 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
9487 | void btrfs_test_destroy_inode(struct inode *inode) | |
9488 | { | |
dcdbc059 | 9489 | btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0); |
aaedb55b JB |
9490 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
9491 | } | |
9492 | #endif | |
9493 | ||
fa0d7e3d NP |
9494 | static void btrfs_i_callback(struct rcu_head *head) |
9495 | { | |
9496 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
9497 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
9498 | } | |
9499 | ||
39279cc3 CM |
9500 | void btrfs_destroy_inode(struct inode *inode) |
9501 | { | |
0b246afa | 9502 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc | 9503 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
9504 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9505 | ||
b3d9b7a3 | 9506 | WARN_ON(!hlist_empty(&inode->i_dentry)); |
39279cc3 | 9507 | WARN_ON(inode->i_data.nrpages); |
69fe2d75 JB |
9508 | WARN_ON(BTRFS_I(inode)->block_rsv.reserved); |
9509 | WARN_ON(BTRFS_I(inode)->block_rsv.size); | |
9e0baf60 | 9510 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
7709cde3 | 9511 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
a7e3b975 | 9512 | WARN_ON(BTRFS_I(inode)->new_delalloc_bytes); |
7709cde3 | 9513 | WARN_ON(BTRFS_I(inode)->csum_bytes); |
47059d93 | 9514 | WARN_ON(BTRFS_I(inode)->defrag_bytes); |
39279cc3 | 9515 | |
a6dbd429 JB |
9516 | /* |
9517 | * This can happen where we create an inode, but somebody else also | |
9518 | * created the same inode and we need to destroy the one we already | |
9519 | * created. | |
9520 | */ | |
9521 | if (!root) | |
9522 | goto free; | |
9523 | ||
8a35d95f JB |
9524 | if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
9525 | &BTRFS_I(inode)->runtime_flags)) { | |
0b246afa | 9526 | btrfs_info(fs_info, "inode %llu still on the orphan list", |
4a0cc7ca | 9527 | btrfs_ino(BTRFS_I(inode))); |
8a35d95f | 9528 | atomic_dec(&root->orphan_inodes); |
7b128766 | 9529 | } |
7b128766 | 9530 | |
d397712b | 9531 | while (1) { |
e6dcd2dc CM |
9532 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
9533 | if (!ordered) | |
9534 | break; | |
9535 | else { | |
0b246afa | 9536 | btrfs_err(fs_info, |
5d163e0e JM |
9537 | "found ordered extent %llu %llu on inode cleanup", |
9538 | ordered->file_offset, ordered->len); | |
e6dcd2dc CM |
9539 | btrfs_remove_ordered_extent(inode, ordered); |
9540 | btrfs_put_ordered_extent(ordered); | |
9541 | btrfs_put_ordered_extent(ordered); | |
9542 | } | |
9543 | } | |
56fa9d07 | 9544 | btrfs_qgroup_check_reserved_leak(inode); |
5d4f98a2 | 9545 | inode_tree_del(inode); |
dcdbc059 | 9546 | btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0); |
a6dbd429 | 9547 | free: |
fa0d7e3d | 9548 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
9549 | } |
9550 | ||
45321ac5 | 9551 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
9552 | { |
9553 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 9554 | |
6379ef9f NA |
9555 | if (root == NULL) |
9556 | return 1; | |
9557 | ||
fa6ac876 | 9558 | /* the snap/subvol tree is on deleting */ |
69e9c6c6 | 9559 | if (btrfs_root_refs(&root->root_item) == 0) |
45321ac5 | 9560 | return 1; |
76dda93c | 9561 | else |
45321ac5 | 9562 | return generic_drop_inode(inode); |
76dda93c YZ |
9563 | } |
9564 | ||
0ee0fda0 | 9565 | static void init_once(void *foo) |
39279cc3 CM |
9566 | { |
9567 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
9568 | ||
9569 | inode_init_once(&ei->vfs_inode); | |
9570 | } | |
9571 | ||
9572 | void btrfs_destroy_cachep(void) | |
9573 | { | |
8c0a8537 KS |
9574 | /* |
9575 | * Make sure all delayed rcu free inodes are flushed before we | |
9576 | * destroy cache. | |
9577 | */ | |
9578 | rcu_barrier(); | |
5598e900 KM |
9579 | kmem_cache_destroy(btrfs_inode_cachep); |
9580 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
5598e900 KM |
9581 | kmem_cache_destroy(btrfs_path_cachep); |
9582 | kmem_cache_destroy(btrfs_free_space_cachep); | |
39279cc3 CM |
9583 | } |
9584 | ||
f5c29bd9 | 9585 | int __init btrfs_init_cachep(void) |
39279cc3 | 9586 | { |
837e1972 | 9587 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 | 9588 | sizeof(struct btrfs_inode), 0, |
5d097056 VD |
9589 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT, |
9590 | init_once); | |
39279cc3 CM |
9591 | if (!btrfs_inode_cachep) |
9592 | goto fail; | |
9601e3f6 | 9593 | |
837e1972 | 9594 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 | 9595 | sizeof(struct btrfs_trans_handle), 0, |
fba4b697 | 9596 | SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
9597 | if (!btrfs_trans_handle_cachep) |
9598 | goto fail; | |
9601e3f6 | 9599 | |
837e1972 | 9600 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 | 9601 | sizeof(struct btrfs_path), 0, |
fba4b697 | 9602 | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
9603 | if (!btrfs_path_cachep) |
9604 | goto fail; | |
9601e3f6 | 9605 | |
837e1972 | 9606 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 | 9607 | sizeof(struct btrfs_free_space), 0, |
fba4b697 | 9608 | SLAB_MEM_SPREAD, NULL); |
dc89e982 JB |
9609 | if (!btrfs_free_space_cachep) |
9610 | goto fail; | |
9611 | ||
39279cc3 CM |
9612 | return 0; |
9613 | fail: | |
9614 | btrfs_destroy_cachep(); | |
9615 | return -ENOMEM; | |
9616 | } | |
9617 | ||
a528d35e DH |
9618 | static int btrfs_getattr(const struct path *path, struct kstat *stat, |
9619 | u32 request_mask, unsigned int flags) | |
39279cc3 | 9620 | { |
df0af1a5 | 9621 | u64 delalloc_bytes; |
a528d35e | 9622 | struct inode *inode = d_inode(path->dentry); |
fadc0d8b | 9623 | u32 blocksize = inode->i_sb->s_blocksize; |
04a87e34 YS |
9624 | u32 bi_flags = BTRFS_I(inode)->flags; |
9625 | ||
9626 | stat->result_mask |= STATX_BTIME; | |
9627 | stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec; | |
9628 | stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec; | |
9629 | if (bi_flags & BTRFS_INODE_APPEND) | |
9630 | stat->attributes |= STATX_ATTR_APPEND; | |
9631 | if (bi_flags & BTRFS_INODE_COMPRESS) | |
9632 | stat->attributes |= STATX_ATTR_COMPRESSED; | |
9633 | if (bi_flags & BTRFS_INODE_IMMUTABLE) | |
9634 | stat->attributes |= STATX_ATTR_IMMUTABLE; | |
9635 | if (bi_flags & BTRFS_INODE_NODUMP) | |
9636 | stat->attributes |= STATX_ATTR_NODUMP; | |
9637 | ||
9638 | stat->attributes_mask |= (STATX_ATTR_APPEND | | |
9639 | STATX_ATTR_COMPRESSED | | |
9640 | STATX_ATTR_IMMUTABLE | | |
9641 | STATX_ATTR_NODUMP); | |
fadc0d8b | 9642 | |
39279cc3 | 9643 | generic_fillattr(inode, stat); |
0ee5dc67 | 9644 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
df0af1a5 MX |
9645 | |
9646 | spin_lock(&BTRFS_I(inode)->lock); | |
a7e3b975 | 9647 | delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes; |
df0af1a5 | 9648 | spin_unlock(&BTRFS_I(inode)->lock); |
fadc0d8b | 9649 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
df0af1a5 | 9650 | ALIGN(delalloc_bytes, blocksize)) >> 9; |
39279cc3 CM |
9651 | return 0; |
9652 | } | |
9653 | ||
cdd1fedf DF |
9654 | static int btrfs_rename_exchange(struct inode *old_dir, |
9655 | struct dentry *old_dentry, | |
9656 | struct inode *new_dir, | |
9657 | struct dentry *new_dentry) | |
9658 | { | |
0b246afa | 9659 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
cdd1fedf DF |
9660 | struct btrfs_trans_handle *trans; |
9661 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
9662 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; | |
9663 | struct inode *new_inode = new_dentry->d_inode; | |
9664 | struct inode *old_inode = old_dentry->d_inode; | |
c2050a45 | 9665 | struct timespec ctime = current_time(old_inode); |
cdd1fedf | 9666 | struct dentry *parent; |
4a0cc7ca NB |
9667 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
9668 | u64 new_ino = btrfs_ino(BTRFS_I(new_inode)); | |
cdd1fedf DF |
9669 | u64 old_idx = 0; |
9670 | u64 new_idx = 0; | |
9671 | u64 root_objectid; | |
9672 | int ret; | |
86e8aa0e FM |
9673 | bool root_log_pinned = false; |
9674 | bool dest_log_pinned = false; | |
cdd1fedf DF |
9675 | |
9676 | /* we only allow rename subvolume link between subvolumes */ | |
9677 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) | |
9678 | return -EXDEV; | |
9679 | ||
9680 | /* close the race window with snapshot create/destroy ioctl */ | |
9681 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 9682 | down_read(&fs_info->subvol_sem); |
cdd1fedf | 9683 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9684 | down_read(&fs_info->subvol_sem); |
cdd1fedf DF |
9685 | |
9686 | /* | |
9687 | * We want to reserve the absolute worst case amount of items. So if | |
9688 | * both inodes are subvols and we need to unlink them then that would | |
9689 | * require 4 item modifications, but if they are both normal inodes it | |
9690 | * would require 5 item modifications, so we'll assume their normal | |
9691 | * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items | |
9692 | * should cover the worst case number of items we'll modify. | |
9693 | */ | |
9694 | trans = btrfs_start_transaction(root, 12); | |
9695 | if (IS_ERR(trans)) { | |
9696 | ret = PTR_ERR(trans); | |
9697 | goto out_notrans; | |
9698 | } | |
9699 | ||
9700 | /* | |
9701 | * We need to find a free sequence number both in the source and | |
9702 | * in the destination directory for the exchange. | |
9703 | */ | |
877574e2 | 9704 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx); |
cdd1fedf DF |
9705 | if (ret) |
9706 | goto out_fail; | |
877574e2 | 9707 | ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx); |
cdd1fedf DF |
9708 | if (ret) |
9709 | goto out_fail; | |
9710 | ||
9711 | BTRFS_I(old_inode)->dir_index = 0ULL; | |
9712 | BTRFS_I(new_inode)->dir_index = 0ULL; | |
9713 | ||
9714 | /* Reference for the source. */ | |
9715 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9716 | /* force full log commit if subvolume involved. */ | |
0b246afa | 9717 | btrfs_set_log_full_commit(fs_info, trans); |
cdd1fedf | 9718 | } else { |
376e5a57 FM |
9719 | btrfs_pin_log_trans(root); |
9720 | root_log_pinned = true; | |
cdd1fedf DF |
9721 | ret = btrfs_insert_inode_ref(trans, dest, |
9722 | new_dentry->d_name.name, | |
9723 | new_dentry->d_name.len, | |
9724 | old_ino, | |
f85b7379 DS |
9725 | btrfs_ino(BTRFS_I(new_dir)), |
9726 | old_idx); | |
cdd1fedf DF |
9727 | if (ret) |
9728 | goto out_fail; | |
cdd1fedf DF |
9729 | } |
9730 | ||
9731 | /* And now for the dest. */ | |
9732 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9733 | /* force full log commit if subvolume involved. */ | |
0b246afa | 9734 | btrfs_set_log_full_commit(fs_info, trans); |
cdd1fedf | 9735 | } else { |
376e5a57 FM |
9736 | btrfs_pin_log_trans(dest); |
9737 | dest_log_pinned = true; | |
cdd1fedf DF |
9738 | ret = btrfs_insert_inode_ref(trans, root, |
9739 | old_dentry->d_name.name, | |
9740 | old_dentry->d_name.len, | |
9741 | new_ino, | |
f85b7379 DS |
9742 | btrfs_ino(BTRFS_I(old_dir)), |
9743 | new_idx); | |
cdd1fedf DF |
9744 | if (ret) |
9745 | goto out_fail; | |
cdd1fedf DF |
9746 | } |
9747 | ||
9748 | /* Update inode version and ctime/mtime. */ | |
9749 | inode_inc_iversion(old_dir); | |
9750 | inode_inc_iversion(new_dir); | |
9751 | inode_inc_iversion(old_inode); | |
9752 | inode_inc_iversion(new_inode); | |
9753 | old_dir->i_ctime = old_dir->i_mtime = ctime; | |
9754 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
9755 | old_inode->i_ctime = ctime; | |
9756 | new_inode->i_ctime = ctime; | |
9757 | ||
9758 | if (old_dentry->d_parent != new_dentry->d_parent) { | |
f85b7379 DS |
9759 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
9760 | BTRFS_I(old_inode), 1); | |
9761 | btrfs_record_unlink_dir(trans, BTRFS_I(new_dir), | |
9762 | BTRFS_I(new_inode), 1); | |
cdd1fedf DF |
9763 | } |
9764 | ||
9765 | /* src is a subvolume */ | |
9766 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9767 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; | |
9768 | ret = btrfs_unlink_subvol(trans, root, old_dir, | |
9769 | root_objectid, | |
9770 | old_dentry->d_name.name, | |
9771 | old_dentry->d_name.len); | |
9772 | } else { /* src is an inode */ | |
4ec5934e NB |
9773 | ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir), |
9774 | BTRFS_I(old_dentry->d_inode), | |
cdd1fedf DF |
9775 | old_dentry->d_name.name, |
9776 | old_dentry->d_name.len); | |
9777 | if (!ret) | |
9778 | ret = btrfs_update_inode(trans, root, old_inode); | |
9779 | } | |
9780 | if (ret) { | |
66642832 | 9781 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9782 | goto out_fail; |
9783 | } | |
9784 | ||
9785 | /* dest is a subvolume */ | |
9786 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9787 | root_objectid = BTRFS_I(new_inode)->root->root_key.objectid; | |
9788 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
9789 | root_objectid, | |
9790 | new_dentry->d_name.name, | |
9791 | new_dentry->d_name.len); | |
9792 | } else { /* dest is an inode */ | |
4ec5934e NB |
9793 | ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir), |
9794 | BTRFS_I(new_dentry->d_inode), | |
cdd1fedf DF |
9795 | new_dentry->d_name.name, |
9796 | new_dentry->d_name.len); | |
9797 | if (!ret) | |
9798 | ret = btrfs_update_inode(trans, dest, new_inode); | |
9799 | } | |
9800 | if (ret) { | |
66642832 | 9801 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9802 | goto out_fail; |
9803 | } | |
9804 | ||
db0a669f | 9805 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
cdd1fedf DF |
9806 | new_dentry->d_name.name, |
9807 | new_dentry->d_name.len, 0, old_idx); | |
9808 | if (ret) { | |
66642832 | 9809 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9810 | goto out_fail; |
9811 | } | |
9812 | ||
db0a669f | 9813 | ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode), |
cdd1fedf DF |
9814 | old_dentry->d_name.name, |
9815 | old_dentry->d_name.len, 0, new_idx); | |
9816 | if (ret) { | |
66642832 | 9817 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9818 | goto out_fail; |
9819 | } | |
9820 | ||
9821 | if (old_inode->i_nlink == 1) | |
9822 | BTRFS_I(old_inode)->dir_index = old_idx; | |
9823 | if (new_inode->i_nlink == 1) | |
9824 | BTRFS_I(new_inode)->dir_index = new_idx; | |
9825 | ||
86e8aa0e | 9826 | if (root_log_pinned) { |
cdd1fedf | 9827 | parent = new_dentry->d_parent; |
f85b7379 DS |
9828 | btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir), |
9829 | parent); | |
cdd1fedf | 9830 | btrfs_end_log_trans(root); |
86e8aa0e | 9831 | root_log_pinned = false; |
cdd1fedf | 9832 | } |
86e8aa0e | 9833 | if (dest_log_pinned) { |
cdd1fedf | 9834 | parent = old_dentry->d_parent; |
f85b7379 DS |
9835 | btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir), |
9836 | parent); | |
cdd1fedf | 9837 | btrfs_end_log_trans(dest); |
86e8aa0e | 9838 | dest_log_pinned = false; |
cdd1fedf DF |
9839 | } |
9840 | out_fail: | |
86e8aa0e FM |
9841 | /* |
9842 | * If we have pinned a log and an error happened, we unpin tasks | |
9843 | * trying to sync the log and force them to fallback to a transaction | |
9844 | * commit if the log currently contains any of the inodes involved in | |
9845 | * this rename operation (to ensure we do not persist a log with an | |
9846 | * inconsistent state for any of these inodes or leading to any | |
9847 | * inconsistencies when replayed). If the transaction was aborted, the | |
9848 | * abortion reason is propagated to userspace when attempting to commit | |
9849 | * the transaction. If the log does not contain any of these inodes, we | |
9850 | * allow the tasks to sync it. | |
9851 | */ | |
9852 | if (ret && (root_log_pinned || dest_log_pinned)) { | |
0f8939b8 NB |
9853 | if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) || |
9854 | btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) || | |
9855 | btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) || | |
86e8aa0e | 9856 | (new_inode && |
0f8939b8 | 9857 | btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation))) |
0b246afa | 9858 | btrfs_set_log_full_commit(fs_info, trans); |
86e8aa0e FM |
9859 | |
9860 | if (root_log_pinned) { | |
9861 | btrfs_end_log_trans(root); | |
9862 | root_log_pinned = false; | |
9863 | } | |
9864 | if (dest_log_pinned) { | |
9865 | btrfs_end_log_trans(dest); | |
9866 | dest_log_pinned = false; | |
9867 | } | |
9868 | } | |
3a45bb20 | 9869 | ret = btrfs_end_transaction(trans); |
cdd1fedf DF |
9870 | out_notrans: |
9871 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 9872 | up_read(&fs_info->subvol_sem); |
cdd1fedf | 9873 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9874 | up_read(&fs_info->subvol_sem); |
cdd1fedf DF |
9875 | |
9876 | return ret; | |
9877 | } | |
9878 | ||
9879 | static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans, | |
9880 | struct btrfs_root *root, | |
9881 | struct inode *dir, | |
9882 | struct dentry *dentry) | |
9883 | { | |
9884 | int ret; | |
9885 | struct inode *inode; | |
9886 | u64 objectid; | |
9887 | u64 index; | |
9888 | ||
9889 | ret = btrfs_find_free_ino(root, &objectid); | |
9890 | if (ret) | |
9891 | return ret; | |
9892 | ||
9893 | inode = btrfs_new_inode(trans, root, dir, | |
9894 | dentry->d_name.name, | |
9895 | dentry->d_name.len, | |
4a0cc7ca | 9896 | btrfs_ino(BTRFS_I(dir)), |
cdd1fedf DF |
9897 | objectid, |
9898 | S_IFCHR | WHITEOUT_MODE, | |
9899 | &index); | |
9900 | ||
9901 | if (IS_ERR(inode)) { | |
9902 | ret = PTR_ERR(inode); | |
9903 | return ret; | |
9904 | } | |
9905 | ||
9906 | inode->i_op = &btrfs_special_inode_operations; | |
9907 | init_special_inode(inode, inode->i_mode, | |
9908 | WHITEOUT_DEV); | |
9909 | ||
9910 | ret = btrfs_init_inode_security(trans, inode, dir, | |
9911 | &dentry->d_name); | |
9912 | if (ret) | |
c9901618 | 9913 | goto out; |
cdd1fedf | 9914 | |
cef415af NB |
9915 | ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, |
9916 | BTRFS_I(inode), 0, index); | |
cdd1fedf | 9917 | if (ret) |
c9901618 | 9918 | goto out; |
cdd1fedf DF |
9919 | |
9920 | ret = btrfs_update_inode(trans, root, inode); | |
c9901618 | 9921 | out: |
cdd1fedf | 9922 | unlock_new_inode(inode); |
c9901618 FM |
9923 | if (ret) |
9924 | inode_dec_link_count(inode); | |
cdd1fedf DF |
9925 | iput(inode); |
9926 | ||
c9901618 | 9927 | return ret; |
cdd1fedf DF |
9928 | } |
9929 | ||
d397712b | 9930 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
cdd1fedf DF |
9931 | struct inode *new_dir, struct dentry *new_dentry, |
9932 | unsigned int flags) | |
39279cc3 | 9933 | { |
0b246afa | 9934 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
39279cc3 | 9935 | struct btrfs_trans_handle *trans; |
5062af35 | 9936 | unsigned int trans_num_items; |
39279cc3 | 9937 | struct btrfs_root *root = BTRFS_I(old_dir)->root; |
4df27c4d | 9938 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
2b0143b5 DH |
9939 | struct inode *new_inode = d_inode(new_dentry); |
9940 | struct inode *old_inode = d_inode(old_dentry); | |
00e4e6b3 | 9941 | u64 index = 0; |
4df27c4d | 9942 | u64 root_objectid; |
39279cc3 | 9943 | int ret; |
4a0cc7ca | 9944 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
3dc9e8f7 | 9945 | bool log_pinned = false; |
39279cc3 | 9946 | |
4a0cc7ca | 9947 | if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
9948 | return -EPERM; |
9949 | ||
4df27c4d | 9950 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 9951 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
9952 | return -EXDEV; |
9953 | ||
33345d01 | 9954 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
4a0cc7ca | 9955 | (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID)) |
39279cc3 | 9956 | return -ENOTEMPTY; |
5f39d397 | 9957 | |
4df27c4d YZ |
9958 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
9959 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
9960 | return -ENOTEMPTY; | |
9c52057c CM |
9961 | |
9962 | ||
9963 | /* check for collisions, even if the name isn't there */ | |
4871c158 | 9964 | ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, |
9c52057c CM |
9965 | new_dentry->d_name.name, |
9966 | new_dentry->d_name.len); | |
9967 | ||
9968 | if (ret) { | |
9969 | if (ret == -EEXIST) { | |
9970 | /* we shouldn't get | |
9971 | * eexist without a new_inode */ | |
fae7f21c | 9972 | if (WARN_ON(!new_inode)) { |
9c52057c CM |
9973 | return ret; |
9974 | } | |
9975 | } else { | |
9976 | /* maybe -EOVERFLOW */ | |
9977 | return ret; | |
9978 | } | |
9979 | } | |
9980 | ret = 0; | |
9981 | ||
5a3f23d5 | 9982 | /* |
8d875f95 CM |
9983 | * we're using rename to replace one file with another. Start IO on it |
9984 | * now so we don't add too much work to the end of the transaction | |
5a3f23d5 | 9985 | */ |
8d875f95 | 9986 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size) |
5a3f23d5 CM |
9987 | filemap_flush(old_inode->i_mapping); |
9988 | ||
76dda93c | 9989 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 9990 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9991 | down_read(&fs_info->subvol_sem); |
a22285a6 YZ |
9992 | /* |
9993 | * We want to reserve the absolute worst case amount of items. So if | |
9994 | * both inodes are subvols and we need to unlink them then that would | |
9995 | * require 4 item modifications, but if they are both normal inodes it | |
cdd1fedf | 9996 | * would require 5 item modifications, so we'll assume they are normal |
a22285a6 YZ |
9997 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items |
9998 | * should cover the worst case number of items we'll modify. | |
5062af35 FM |
9999 | * If our rename has the whiteout flag, we need more 5 units for the |
10000 | * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item | |
10001 | * when selinux is enabled). | |
a22285a6 | 10002 | */ |
5062af35 FM |
10003 | trans_num_items = 11; |
10004 | if (flags & RENAME_WHITEOUT) | |
10005 | trans_num_items += 5; | |
10006 | trans = btrfs_start_transaction(root, trans_num_items); | |
b44c59a8 | 10007 | if (IS_ERR(trans)) { |
cdd1fedf DF |
10008 | ret = PTR_ERR(trans); |
10009 | goto out_notrans; | |
10010 | } | |
76dda93c | 10011 | |
4df27c4d YZ |
10012 | if (dest != root) |
10013 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 10014 | |
877574e2 | 10015 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index); |
a5719521 YZ |
10016 | if (ret) |
10017 | goto out_fail; | |
5a3f23d5 | 10018 | |
67de1176 | 10019 | BTRFS_I(old_inode)->dir_index = 0ULL; |
33345d01 | 10020 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d | 10021 | /* force full log commit if subvolume involved. */ |
0b246afa | 10022 | btrfs_set_log_full_commit(fs_info, trans); |
4df27c4d | 10023 | } else { |
c4aba954 FM |
10024 | btrfs_pin_log_trans(root); |
10025 | log_pinned = true; | |
a5719521 YZ |
10026 | ret = btrfs_insert_inode_ref(trans, dest, |
10027 | new_dentry->d_name.name, | |
10028 | new_dentry->d_name.len, | |
33345d01 | 10029 | old_ino, |
4a0cc7ca | 10030 | btrfs_ino(BTRFS_I(new_dir)), index); |
a5719521 YZ |
10031 | if (ret) |
10032 | goto out_fail; | |
4df27c4d | 10033 | } |
5a3f23d5 | 10034 | |
0c4d2d95 JB |
10035 | inode_inc_iversion(old_dir); |
10036 | inode_inc_iversion(new_dir); | |
10037 | inode_inc_iversion(old_inode); | |
04b285f3 DD |
10038 | old_dir->i_ctime = old_dir->i_mtime = |
10039 | new_dir->i_ctime = new_dir->i_mtime = | |
c2050a45 | 10040 | old_inode->i_ctime = current_time(old_dir); |
5f39d397 | 10041 | |
12fcfd22 | 10042 | if (old_dentry->d_parent != new_dentry->d_parent) |
f85b7379 DS |
10043 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
10044 | BTRFS_I(old_inode), 1); | |
12fcfd22 | 10045 | |
33345d01 | 10046 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
10047 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
10048 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
10049 | old_dentry->d_name.name, | |
10050 | old_dentry->d_name.len); | |
10051 | } else { | |
4ec5934e NB |
10052 | ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir), |
10053 | BTRFS_I(d_inode(old_dentry)), | |
92986796 AV |
10054 | old_dentry->d_name.name, |
10055 | old_dentry->d_name.len); | |
10056 | if (!ret) | |
10057 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 10058 | } |
79787eaa | 10059 | if (ret) { |
66642832 | 10060 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
10061 | goto out_fail; |
10062 | } | |
39279cc3 CM |
10063 | |
10064 | if (new_inode) { | |
0c4d2d95 | 10065 | inode_inc_iversion(new_inode); |
c2050a45 | 10066 | new_inode->i_ctime = current_time(new_inode); |
4a0cc7ca | 10067 | if (unlikely(btrfs_ino(BTRFS_I(new_inode)) == |
4df27c4d YZ |
10068 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
10069 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
10070 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
10071 | root_objectid, | |
10072 | new_dentry->d_name.name, | |
10073 | new_dentry->d_name.len); | |
10074 | BUG_ON(new_inode->i_nlink == 0); | |
10075 | } else { | |
4ec5934e NB |
10076 | ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir), |
10077 | BTRFS_I(d_inode(new_dentry)), | |
4df27c4d YZ |
10078 | new_dentry->d_name.name, |
10079 | new_dentry->d_name.len); | |
10080 | } | |
4ef31a45 | 10081 | if (!ret && new_inode->i_nlink == 0) |
73f2e545 NB |
10082 | ret = btrfs_orphan_add(trans, |
10083 | BTRFS_I(d_inode(new_dentry))); | |
79787eaa | 10084 | if (ret) { |
66642832 | 10085 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
10086 | goto out_fail; |
10087 | } | |
39279cc3 | 10088 | } |
aec7477b | 10089 | |
db0a669f | 10090 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
4df27c4d | 10091 | new_dentry->d_name.name, |
a5719521 | 10092 | new_dentry->d_name.len, 0, index); |
79787eaa | 10093 | if (ret) { |
66642832 | 10094 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
10095 | goto out_fail; |
10096 | } | |
39279cc3 | 10097 | |
67de1176 MX |
10098 | if (old_inode->i_nlink == 1) |
10099 | BTRFS_I(old_inode)->dir_index = index; | |
10100 | ||
3dc9e8f7 | 10101 | if (log_pinned) { |
10d9f309 | 10102 | struct dentry *parent = new_dentry->d_parent; |
3dc9e8f7 | 10103 | |
f85b7379 DS |
10104 | btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir), |
10105 | parent); | |
4df27c4d | 10106 | btrfs_end_log_trans(root); |
3dc9e8f7 | 10107 | log_pinned = false; |
4df27c4d | 10108 | } |
cdd1fedf DF |
10109 | |
10110 | if (flags & RENAME_WHITEOUT) { | |
10111 | ret = btrfs_whiteout_for_rename(trans, root, old_dir, | |
10112 | old_dentry); | |
10113 | ||
10114 | if (ret) { | |
66642832 | 10115 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
10116 | goto out_fail; |
10117 | } | |
4df27c4d | 10118 | } |
39279cc3 | 10119 | out_fail: |
3dc9e8f7 FM |
10120 | /* |
10121 | * If we have pinned the log and an error happened, we unpin tasks | |
10122 | * trying to sync the log and force them to fallback to a transaction | |
10123 | * commit if the log currently contains any of the inodes involved in | |
10124 | * this rename operation (to ensure we do not persist a log with an | |
10125 | * inconsistent state for any of these inodes or leading to any | |
10126 | * inconsistencies when replayed). If the transaction was aborted, the | |
10127 | * abortion reason is propagated to userspace when attempting to commit | |
10128 | * the transaction. If the log does not contain any of these inodes, we | |
10129 | * allow the tasks to sync it. | |
10130 | */ | |
10131 | if (ret && log_pinned) { | |
0f8939b8 NB |
10132 | if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) || |
10133 | btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) || | |
10134 | btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) || | |
3dc9e8f7 | 10135 | (new_inode && |
0f8939b8 | 10136 | btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation))) |
0b246afa | 10137 | btrfs_set_log_full_commit(fs_info, trans); |
3dc9e8f7 FM |
10138 | |
10139 | btrfs_end_log_trans(root); | |
10140 | log_pinned = false; | |
10141 | } | |
3a45bb20 | 10142 | btrfs_end_transaction(trans); |
b44c59a8 | 10143 | out_notrans: |
33345d01 | 10144 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 10145 | up_read(&fs_info->subvol_sem); |
9ed74f2d | 10146 | |
39279cc3 CM |
10147 | return ret; |
10148 | } | |
10149 | ||
80ace85c MS |
10150 | static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry, |
10151 | struct inode *new_dir, struct dentry *new_dentry, | |
10152 | unsigned int flags) | |
10153 | { | |
cdd1fedf | 10154 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
80ace85c MS |
10155 | return -EINVAL; |
10156 | ||
cdd1fedf DF |
10157 | if (flags & RENAME_EXCHANGE) |
10158 | return btrfs_rename_exchange(old_dir, old_dentry, new_dir, | |
10159 | new_dentry); | |
10160 | ||
10161 | return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags); | |
80ace85c MS |
10162 | } |
10163 | ||
8ccf6f19 MX |
10164 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
10165 | { | |
10166 | struct btrfs_delalloc_work *delalloc_work; | |
9f23e289 | 10167 | struct inode *inode; |
8ccf6f19 MX |
10168 | |
10169 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
10170 | work); | |
9f23e289 | 10171 | inode = delalloc_work->inode; |
30424601 DS |
10172 | filemap_flush(inode->i_mapping); |
10173 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
10174 | &BTRFS_I(inode)->runtime_flags)) | |
9f23e289 | 10175 | filemap_flush(inode->i_mapping); |
8ccf6f19 MX |
10176 | |
10177 | if (delalloc_work->delay_iput) | |
9f23e289 | 10178 | btrfs_add_delayed_iput(inode); |
8ccf6f19 | 10179 | else |
9f23e289 | 10180 | iput(inode); |
8ccf6f19 MX |
10181 | complete(&delalloc_work->completion); |
10182 | } | |
10183 | ||
10184 | struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode, | |
651d494a | 10185 | int delay_iput) |
8ccf6f19 MX |
10186 | { |
10187 | struct btrfs_delalloc_work *work; | |
10188 | ||
100d5702 | 10189 | work = kmalloc(sizeof(*work), GFP_NOFS); |
8ccf6f19 MX |
10190 | if (!work) |
10191 | return NULL; | |
10192 | ||
10193 | init_completion(&work->completion); | |
10194 | INIT_LIST_HEAD(&work->list); | |
10195 | work->inode = inode; | |
8ccf6f19 | 10196 | work->delay_iput = delay_iput; |
9e0af237 LB |
10197 | WARN_ON_ONCE(!inode); |
10198 | btrfs_init_work(&work->work, btrfs_flush_delalloc_helper, | |
10199 | btrfs_run_delalloc_work, NULL, NULL); | |
8ccf6f19 MX |
10200 | |
10201 | return work; | |
10202 | } | |
10203 | ||
10204 | void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work) | |
10205 | { | |
10206 | wait_for_completion(&work->completion); | |
100d5702 | 10207 | kfree(work); |
8ccf6f19 MX |
10208 | } |
10209 | ||
d352ac68 CM |
10210 | /* |
10211 | * some fairly slow code that needs optimization. This walks the list | |
10212 | * of all the inodes with pending delalloc and forces them to disk. | |
10213 | */ | |
6c255e67 MX |
10214 | static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput, |
10215 | int nr) | |
ea8c2819 | 10216 | { |
ea8c2819 | 10217 | struct btrfs_inode *binode; |
5b21f2ed | 10218 | struct inode *inode; |
8ccf6f19 MX |
10219 | struct btrfs_delalloc_work *work, *next; |
10220 | struct list_head works; | |
1eafa6c7 | 10221 | struct list_head splice; |
8ccf6f19 | 10222 | int ret = 0; |
ea8c2819 | 10223 | |
8ccf6f19 | 10224 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 10225 | INIT_LIST_HEAD(&splice); |
63607cc8 | 10226 | |
573bfb72 | 10227 | mutex_lock(&root->delalloc_mutex); |
eb73c1b7 MX |
10228 | spin_lock(&root->delalloc_lock); |
10229 | list_splice_init(&root->delalloc_inodes, &splice); | |
1eafa6c7 MX |
10230 | while (!list_empty(&splice)) { |
10231 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 10232 | delalloc_inodes); |
1eafa6c7 | 10233 | |
eb73c1b7 MX |
10234 | list_move_tail(&binode->delalloc_inodes, |
10235 | &root->delalloc_inodes); | |
5b21f2ed | 10236 | inode = igrab(&binode->vfs_inode); |
df0af1a5 | 10237 | if (!inode) { |
eb73c1b7 | 10238 | cond_resched_lock(&root->delalloc_lock); |
1eafa6c7 | 10239 | continue; |
df0af1a5 | 10240 | } |
eb73c1b7 | 10241 | spin_unlock(&root->delalloc_lock); |
1eafa6c7 | 10242 | |
651d494a | 10243 | work = btrfs_alloc_delalloc_work(inode, delay_iput); |
5d99a998 | 10244 | if (!work) { |
f4ab9ea7 JB |
10245 | if (delay_iput) |
10246 | btrfs_add_delayed_iput(inode); | |
10247 | else | |
10248 | iput(inode); | |
1eafa6c7 | 10249 | ret = -ENOMEM; |
a1ecaabb | 10250 | goto out; |
5b21f2ed | 10251 | } |
1eafa6c7 | 10252 | list_add_tail(&work->list, &works); |
a44903ab QW |
10253 | btrfs_queue_work(root->fs_info->flush_workers, |
10254 | &work->work); | |
6c255e67 MX |
10255 | ret++; |
10256 | if (nr != -1 && ret >= nr) | |
a1ecaabb | 10257 | goto out; |
5b21f2ed | 10258 | cond_resched(); |
eb73c1b7 | 10259 | spin_lock(&root->delalloc_lock); |
ea8c2819 | 10260 | } |
eb73c1b7 | 10261 | spin_unlock(&root->delalloc_lock); |
8c8bee1d | 10262 | |
a1ecaabb | 10263 | out: |
eb73c1b7 MX |
10264 | list_for_each_entry_safe(work, next, &works, list) { |
10265 | list_del_init(&work->list); | |
10266 | btrfs_wait_and_free_delalloc_work(work); | |
10267 | } | |
10268 | ||
10269 | if (!list_empty_careful(&splice)) { | |
10270 | spin_lock(&root->delalloc_lock); | |
10271 | list_splice_tail(&splice, &root->delalloc_inodes); | |
10272 | spin_unlock(&root->delalloc_lock); | |
10273 | } | |
573bfb72 | 10274 | mutex_unlock(&root->delalloc_mutex); |
eb73c1b7 MX |
10275 | return ret; |
10276 | } | |
1eafa6c7 | 10277 | |
eb73c1b7 MX |
10278 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
10279 | { | |
0b246afa | 10280 | struct btrfs_fs_info *fs_info = root->fs_info; |
eb73c1b7 | 10281 | int ret; |
1eafa6c7 | 10282 | |
0b246afa | 10283 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
10284 | return -EROFS; |
10285 | ||
6c255e67 MX |
10286 | ret = __start_delalloc_inodes(root, delay_iput, -1); |
10287 | if (ret > 0) | |
10288 | ret = 0; | |
eb73c1b7 MX |
10289 | return ret; |
10290 | } | |
10291 | ||
6c255e67 MX |
10292 | int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput, |
10293 | int nr) | |
eb73c1b7 MX |
10294 | { |
10295 | struct btrfs_root *root; | |
10296 | struct list_head splice; | |
10297 | int ret; | |
10298 | ||
2c21b4d7 | 10299 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
10300 | return -EROFS; |
10301 | ||
10302 | INIT_LIST_HEAD(&splice); | |
10303 | ||
573bfb72 | 10304 | mutex_lock(&fs_info->delalloc_root_mutex); |
eb73c1b7 MX |
10305 | spin_lock(&fs_info->delalloc_root_lock); |
10306 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
6c255e67 | 10307 | while (!list_empty(&splice) && nr) { |
eb73c1b7 MX |
10308 | root = list_first_entry(&splice, struct btrfs_root, |
10309 | delalloc_root); | |
10310 | root = btrfs_grab_fs_root(root); | |
10311 | BUG_ON(!root); | |
10312 | list_move_tail(&root->delalloc_root, | |
10313 | &fs_info->delalloc_roots); | |
10314 | spin_unlock(&fs_info->delalloc_root_lock); | |
10315 | ||
6c255e67 | 10316 | ret = __start_delalloc_inodes(root, delay_iput, nr); |
eb73c1b7 | 10317 | btrfs_put_fs_root(root); |
6c255e67 | 10318 | if (ret < 0) |
eb73c1b7 MX |
10319 | goto out; |
10320 | ||
6c255e67 MX |
10321 | if (nr != -1) { |
10322 | nr -= ret; | |
10323 | WARN_ON(nr < 0); | |
10324 | } | |
eb73c1b7 | 10325 | spin_lock(&fs_info->delalloc_root_lock); |
8ccf6f19 | 10326 | } |
eb73c1b7 | 10327 | spin_unlock(&fs_info->delalloc_root_lock); |
1eafa6c7 | 10328 | |
6c255e67 | 10329 | ret = 0; |
eb73c1b7 | 10330 | out: |
1eafa6c7 | 10331 | if (!list_empty_careful(&splice)) { |
eb73c1b7 MX |
10332 | spin_lock(&fs_info->delalloc_root_lock); |
10333 | list_splice_tail(&splice, &fs_info->delalloc_roots); | |
10334 | spin_unlock(&fs_info->delalloc_root_lock); | |
1eafa6c7 | 10335 | } |
573bfb72 | 10336 | mutex_unlock(&fs_info->delalloc_root_mutex); |
8ccf6f19 | 10337 | return ret; |
ea8c2819 CM |
10338 | } |
10339 | ||
39279cc3 CM |
10340 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
10341 | const char *symname) | |
10342 | { | |
0b246afa | 10343 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
10344 | struct btrfs_trans_handle *trans; |
10345 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
10346 | struct btrfs_path *path; | |
10347 | struct btrfs_key key; | |
1832a6d5 | 10348 | struct inode *inode = NULL; |
39279cc3 CM |
10349 | int err; |
10350 | int drop_inode = 0; | |
10351 | u64 objectid; | |
67871254 | 10352 | u64 index = 0; |
39279cc3 CM |
10353 | int name_len; |
10354 | int datasize; | |
5f39d397 | 10355 | unsigned long ptr; |
39279cc3 | 10356 | struct btrfs_file_extent_item *ei; |
5f39d397 | 10357 | struct extent_buffer *leaf; |
39279cc3 | 10358 | |
f06becc4 | 10359 | name_len = strlen(symname); |
0b246afa | 10360 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info)) |
39279cc3 | 10361 | return -ENAMETOOLONG; |
1832a6d5 | 10362 | |
9ed74f2d JB |
10363 | /* |
10364 | * 2 items for inode item and ref | |
10365 | * 2 items for dir items | |
9269d12b FM |
10366 | * 1 item for updating parent inode item |
10367 | * 1 item for the inline extent item | |
9ed74f2d JB |
10368 | * 1 item for xattr if selinux is on |
10369 | */ | |
9269d12b | 10370 | trans = btrfs_start_transaction(root, 7); |
a22285a6 YZ |
10371 | if (IS_ERR(trans)) |
10372 | return PTR_ERR(trans); | |
1832a6d5 | 10373 | |
581bb050 LZ |
10374 | err = btrfs_find_free_ino(root, &objectid); |
10375 | if (err) | |
10376 | goto out_unlock; | |
10377 | ||
aec7477b | 10378 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
10379 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), |
10380 | objectid, S_IFLNK|S_IRWXUGO, &index); | |
7cf96da3 TI |
10381 | if (IS_ERR(inode)) { |
10382 | err = PTR_ERR(inode); | |
39279cc3 | 10383 | goto out_unlock; |
7cf96da3 | 10384 | } |
39279cc3 | 10385 | |
ad19db71 CS |
10386 | /* |
10387 | * If the active LSM wants to access the inode during | |
10388 | * d_instantiate it needs these. Smack checks to see | |
10389 | * if the filesystem supports xattrs by looking at the | |
10390 | * ops vector. | |
10391 | */ | |
10392 | inode->i_fop = &btrfs_file_operations; | |
10393 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 10394 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
10395 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
10396 | ||
10397 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
10398 | if (err) | |
10399 | goto out_unlock_inode; | |
ad19db71 | 10400 | |
39279cc3 | 10401 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10402 | if (!path) { |
10403 | err = -ENOMEM; | |
b0d5d10f | 10404 | goto out_unlock_inode; |
d8926bb3 | 10405 | } |
4a0cc7ca | 10406 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
39279cc3 | 10407 | key.offset = 0; |
962a298f | 10408 | key.type = BTRFS_EXTENT_DATA_KEY; |
39279cc3 CM |
10409 | datasize = btrfs_file_extent_calc_inline_size(name_len); |
10410 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
10411 | datasize); | |
54aa1f4d | 10412 | if (err) { |
b0839166 | 10413 | btrfs_free_path(path); |
b0d5d10f | 10414 | goto out_unlock_inode; |
54aa1f4d | 10415 | } |
5f39d397 CM |
10416 | leaf = path->nodes[0]; |
10417 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
10418 | struct btrfs_file_extent_item); | |
10419 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
10420 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 10421 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
10422 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
10423 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
10424 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
10425 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
10426 | ||
39279cc3 | 10427 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
10428 | write_extent_buffer(leaf, symname, ptr, name_len); |
10429 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 10430 | btrfs_free_path(path); |
5f39d397 | 10431 | |
39279cc3 | 10432 | inode->i_op = &btrfs_symlink_inode_operations; |
21fc61c7 | 10433 | inode_nohighmem(inode); |
39279cc3 | 10434 | inode->i_mapping->a_ops = &btrfs_symlink_aops; |
d899e052 | 10435 | inode_set_bytes(inode, name_len); |
6ef06d27 | 10436 | btrfs_i_size_write(BTRFS_I(inode), name_len); |
54aa1f4d | 10437 | err = btrfs_update_inode(trans, root, inode); |
d50866d0 FM |
10438 | /* |
10439 | * Last step, add directory indexes for our symlink inode. This is the | |
10440 | * last step to avoid extra cleanup of these indexes if an error happens | |
10441 | * elsewhere above. | |
10442 | */ | |
10443 | if (!err) | |
cef415af NB |
10444 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, |
10445 | BTRFS_I(inode), 0, index); | |
b0d5d10f | 10446 | if (err) { |
54aa1f4d | 10447 | drop_inode = 1; |
b0d5d10f CM |
10448 | goto out_unlock_inode; |
10449 | } | |
10450 | ||
10451 | unlock_new_inode(inode); | |
10452 | d_instantiate(dentry, inode); | |
39279cc3 CM |
10453 | |
10454 | out_unlock: | |
3a45bb20 | 10455 | btrfs_end_transaction(trans); |
39279cc3 CM |
10456 | if (drop_inode) { |
10457 | inode_dec_link_count(inode); | |
10458 | iput(inode); | |
10459 | } | |
2ff7e61e | 10460 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 | 10461 | return err; |
b0d5d10f CM |
10462 | |
10463 | out_unlock_inode: | |
10464 | drop_inode = 1; | |
10465 | unlock_new_inode(inode); | |
10466 | goto out_unlock; | |
39279cc3 | 10467 | } |
16432985 | 10468 | |
0af3d00b JB |
10469 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
10470 | u64 start, u64 num_bytes, u64 min_size, | |
10471 | loff_t actual_len, u64 *alloc_hint, | |
10472 | struct btrfs_trans_handle *trans) | |
d899e052 | 10473 | { |
0b246afa | 10474 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5dc562c5 JB |
10475 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
10476 | struct extent_map *em; | |
d899e052 YZ |
10477 | struct btrfs_root *root = BTRFS_I(inode)->root; |
10478 | struct btrfs_key ins; | |
d899e052 | 10479 | u64 cur_offset = start; |
55a61d1d | 10480 | u64 i_size; |
154ea289 | 10481 | u64 cur_bytes; |
0b670dc4 | 10482 | u64 last_alloc = (u64)-1; |
d899e052 | 10483 | int ret = 0; |
0af3d00b | 10484 | bool own_trans = true; |
18513091 | 10485 | u64 end = start + num_bytes - 1; |
d899e052 | 10486 | |
0af3d00b JB |
10487 | if (trans) |
10488 | own_trans = false; | |
d899e052 | 10489 | while (num_bytes > 0) { |
0af3d00b JB |
10490 | if (own_trans) { |
10491 | trans = btrfs_start_transaction(root, 3); | |
10492 | if (IS_ERR(trans)) { | |
10493 | ret = PTR_ERR(trans); | |
10494 | break; | |
10495 | } | |
5a303d5d YZ |
10496 | } |
10497 | ||
ee22184b | 10498 | cur_bytes = min_t(u64, num_bytes, SZ_256M); |
154ea289 | 10499 | cur_bytes = max(cur_bytes, min_size); |
0b670dc4 JB |
10500 | /* |
10501 | * If we are severely fragmented we could end up with really | |
10502 | * small allocations, so if the allocator is returning small | |
10503 | * chunks lets make its job easier by only searching for those | |
10504 | * sized chunks. | |
10505 | */ | |
10506 | cur_bytes = min(cur_bytes, last_alloc); | |
18513091 WX |
10507 | ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes, |
10508 | min_size, 0, *alloc_hint, &ins, 1, 0); | |
5a303d5d | 10509 | if (ret) { |
0af3d00b | 10510 | if (own_trans) |
3a45bb20 | 10511 | btrfs_end_transaction(trans); |
a22285a6 | 10512 | break; |
d899e052 | 10513 | } |
0b246afa | 10514 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
5a303d5d | 10515 | |
0b670dc4 | 10516 | last_alloc = ins.offset; |
d899e052 YZ |
10517 | ret = insert_reserved_file_extent(trans, inode, |
10518 | cur_offset, ins.objectid, | |
10519 | ins.offset, ins.offset, | |
920bbbfb | 10520 | ins.offset, 0, 0, 0, |
d899e052 | 10521 | BTRFS_FILE_EXTENT_PREALLOC); |
79787eaa | 10522 | if (ret) { |
2ff7e61e | 10523 | btrfs_free_reserved_extent(fs_info, ins.objectid, |
e570fd27 | 10524 | ins.offset, 0); |
66642832 | 10525 | btrfs_abort_transaction(trans, ret); |
79787eaa | 10526 | if (own_trans) |
3a45bb20 | 10527 | btrfs_end_transaction(trans); |
79787eaa JM |
10528 | break; |
10529 | } | |
31193213 | 10530 | |
dcdbc059 | 10531 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
a1ed835e | 10532 | cur_offset + ins.offset -1, 0); |
5a303d5d | 10533 | |
5dc562c5 JB |
10534 | em = alloc_extent_map(); |
10535 | if (!em) { | |
10536 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
10537 | &BTRFS_I(inode)->runtime_flags); | |
10538 | goto next; | |
10539 | } | |
10540 | ||
10541 | em->start = cur_offset; | |
10542 | em->orig_start = cur_offset; | |
10543 | em->len = ins.offset; | |
10544 | em->block_start = ins.objectid; | |
10545 | em->block_len = ins.offset; | |
b4939680 | 10546 | em->orig_block_len = ins.offset; |
cc95bef6 | 10547 | em->ram_bytes = ins.offset; |
0b246afa | 10548 | em->bdev = fs_info->fs_devices->latest_bdev; |
5dc562c5 JB |
10549 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); |
10550 | em->generation = trans->transid; | |
10551 | ||
10552 | while (1) { | |
10553 | write_lock(&em_tree->lock); | |
09a2a8f9 | 10554 | ret = add_extent_mapping(em_tree, em, 1); |
5dc562c5 JB |
10555 | write_unlock(&em_tree->lock); |
10556 | if (ret != -EEXIST) | |
10557 | break; | |
dcdbc059 | 10558 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
5dc562c5 JB |
10559 | cur_offset + ins.offset - 1, |
10560 | 0); | |
10561 | } | |
10562 | free_extent_map(em); | |
10563 | next: | |
d899e052 YZ |
10564 | num_bytes -= ins.offset; |
10565 | cur_offset += ins.offset; | |
efa56464 | 10566 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 10567 | |
0c4d2d95 | 10568 | inode_inc_iversion(inode); |
c2050a45 | 10569 | inode->i_ctime = current_time(inode); |
6cbff00f | 10570 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 10571 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
10572 | (actual_len > inode->i_size) && |
10573 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 10574 | if (cur_offset > actual_len) |
55a61d1d | 10575 | i_size = actual_len; |
d1ea6a61 | 10576 | else |
55a61d1d JB |
10577 | i_size = cur_offset; |
10578 | i_size_write(inode, i_size); | |
10579 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
10580 | } |
10581 | ||
d899e052 | 10582 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
10583 | |
10584 | if (ret) { | |
66642832 | 10585 | btrfs_abort_transaction(trans, ret); |
79787eaa | 10586 | if (own_trans) |
3a45bb20 | 10587 | btrfs_end_transaction(trans); |
79787eaa JM |
10588 | break; |
10589 | } | |
d899e052 | 10590 | |
0af3d00b | 10591 | if (own_trans) |
3a45bb20 | 10592 | btrfs_end_transaction(trans); |
5a303d5d | 10593 | } |
18513091 | 10594 | if (cur_offset < end) |
bc42bda2 | 10595 | btrfs_free_reserved_data_space(inode, NULL, cur_offset, |
18513091 | 10596 | end - cur_offset + 1); |
d899e052 YZ |
10597 | return ret; |
10598 | } | |
10599 | ||
0af3d00b JB |
10600 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
10601 | u64 start, u64 num_bytes, u64 min_size, | |
10602 | loff_t actual_len, u64 *alloc_hint) | |
10603 | { | |
10604 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10605 | min_size, actual_len, alloc_hint, | |
10606 | NULL); | |
10607 | } | |
10608 | ||
10609 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
10610 | struct btrfs_trans_handle *trans, int mode, | |
10611 | u64 start, u64 num_bytes, u64 min_size, | |
10612 | loff_t actual_len, u64 *alloc_hint) | |
10613 | { | |
10614 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10615 | min_size, actual_len, alloc_hint, trans); | |
10616 | } | |
10617 | ||
e6dcd2dc CM |
10618 | static int btrfs_set_page_dirty(struct page *page) |
10619 | { | |
e6dcd2dc CM |
10620 | return __set_page_dirty_nobuffers(page); |
10621 | } | |
10622 | ||
10556cb2 | 10623 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 10624 | { |
b83cc969 | 10625 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 10626 | umode_t mode = inode->i_mode; |
b83cc969 | 10627 | |
cb6db4e5 JM |
10628 | if (mask & MAY_WRITE && |
10629 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
10630 | if (btrfs_root_readonly(root)) | |
10631 | return -EROFS; | |
10632 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
10633 | return -EACCES; | |
10634 | } | |
2830ba7f | 10635 | return generic_permission(inode, mask); |
fdebe2bd | 10636 | } |
39279cc3 | 10637 | |
ef3b9af5 FM |
10638 | static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) |
10639 | { | |
2ff7e61e | 10640 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
ef3b9af5 FM |
10641 | struct btrfs_trans_handle *trans; |
10642 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
10643 | struct inode *inode = NULL; | |
10644 | u64 objectid; | |
10645 | u64 index; | |
10646 | int ret = 0; | |
10647 | ||
10648 | /* | |
10649 | * 5 units required for adding orphan entry | |
10650 | */ | |
10651 | trans = btrfs_start_transaction(root, 5); | |
10652 | if (IS_ERR(trans)) | |
10653 | return PTR_ERR(trans); | |
10654 | ||
10655 | ret = btrfs_find_free_ino(root, &objectid); | |
10656 | if (ret) | |
10657 | goto out; | |
10658 | ||
10659 | inode = btrfs_new_inode(trans, root, dir, NULL, 0, | |
f85b7379 | 10660 | btrfs_ino(BTRFS_I(dir)), objectid, mode, &index); |
ef3b9af5 FM |
10661 | if (IS_ERR(inode)) { |
10662 | ret = PTR_ERR(inode); | |
10663 | inode = NULL; | |
10664 | goto out; | |
10665 | } | |
10666 | ||
ef3b9af5 FM |
10667 | inode->i_fop = &btrfs_file_operations; |
10668 | inode->i_op = &btrfs_file_inode_operations; | |
10669 | ||
10670 | inode->i_mapping->a_ops = &btrfs_aops; | |
ef3b9af5 FM |
10671 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
10672 | ||
b0d5d10f CM |
10673 | ret = btrfs_init_inode_security(trans, inode, dir, NULL); |
10674 | if (ret) | |
10675 | goto out_inode; | |
10676 | ||
10677 | ret = btrfs_update_inode(trans, root, inode); | |
10678 | if (ret) | |
10679 | goto out_inode; | |
73f2e545 | 10680 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
ef3b9af5 | 10681 | if (ret) |
b0d5d10f | 10682 | goto out_inode; |
ef3b9af5 | 10683 | |
5762b5c9 FM |
10684 | /* |
10685 | * We set number of links to 0 in btrfs_new_inode(), and here we set | |
10686 | * it to 1 because d_tmpfile() will issue a warning if the count is 0, | |
10687 | * through: | |
10688 | * | |
10689 | * d_tmpfile() -> inode_dec_link_count() -> drop_nlink() | |
10690 | */ | |
10691 | set_nlink(inode, 1); | |
b0d5d10f | 10692 | unlock_new_inode(inode); |
ef3b9af5 FM |
10693 | d_tmpfile(dentry, inode); |
10694 | mark_inode_dirty(inode); | |
10695 | ||
10696 | out: | |
3a45bb20 | 10697 | btrfs_end_transaction(trans); |
ef3b9af5 FM |
10698 | if (ret) |
10699 | iput(inode); | |
2ff7e61e | 10700 | btrfs_btree_balance_dirty(fs_info); |
ef3b9af5 | 10701 | return ret; |
b0d5d10f CM |
10702 | |
10703 | out_inode: | |
10704 | unlock_new_inode(inode); | |
10705 | goto out; | |
10706 | ||
ef3b9af5 FM |
10707 | } |
10708 | ||
20a7db8a | 10709 | __attribute__((const)) |
9d0d1c8b | 10710 | static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror) |
20a7db8a | 10711 | { |
9d0d1c8b | 10712 | return -EAGAIN; |
20a7db8a DS |
10713 | } |
10714 | ||
c6100a4b JB |
10715 | static struct btrfs_fs_info *iotree_fs_info(void *private_data) |
10716 | { | |
10717 | struct inode *inode = private_data; | |
10718 | return btrfs_sb(inode->i_sb); | |
10719 | } | |
10720 | ||
10721 | static void btrfs_check_extent_io_range(void *private_data, const char *caller, | |
10722 | u64 start, u64 end) | |
10723 | { | |
10724 | struct inode *inode = private_data; | |
10725 | u64 isize; | |
10726 | ||
10727 | isize = i_size_read(inode); | |
10728 | if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) { | |
10729 | btrfs_debug_rl(BTRFS_I(inode)->root->fs_info, | |
10730 | "%s: ino %llu isize %llu odd range [%llu,%llu]", | |
10731 | caller, btrfs_ino(BTRFS_I(inode)), isize, start, end); | |
10732 | } | |
10733 | } | |
10734 | ||
10735 | void btrfs_set_range_writeback(void *private_data, u64 start, u64 end) | |
10736 | { | |
10737 | struct inode *inode = private_data; | |
10738 | unsigned long index = start >> PAGE_SHIFT; | |
10739 | unsigned long end_index = end >> PAGE_SHIFT; | |
10740 | struct page *page; | |
10741 | ||
10742 | while (index <= end_index) { | |
10743 | page = find_get_page(inode->i_mapping, index); | |
10744 | ASSERT(page); /* Pages should be in the extent_io_tree */ | |
10745 | set_page_writeback(page); | |
10746 | put_page(page); | |
10747 | index++; | |
10748 | } | |
10749 | } | |
10750 | ||
6e1d5dcc | 10751 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 10752 | .getattr = btrfs_getattr, |
39279cc3 CM |
10753 | .lookup = btrfs_lookup, |
10754 | .create = btrfs_create, | |
10755 | .unlink = btrfs_unlink, | |
10756 | .link = btrfs_link, | |
10757 | .mkdir = btrfs_mkdir, | |
10758 | .rmdir = btrfs_rmdir, | |
2773bf00 | 10759 | .rename = btrfs_rename2, |
39279cc3 CM |
10760 | .symlink = btrfs_symlink, |
10761 | .setattr = btrfs_setattr, | |
618e21d5 | 10762 | .mknod = btrfs_mknod, |
5103e947 | 10763 | .listxattr = btrfs_listxattr, |
fdebe2bd | 10764 | .permission = btrfs_permission, |
4e34e719 | 10765 | .get_acl = btrfs_get_acl, |
996a710d | 10766 | .set_acl = btrfs_set_acl, |
93fd63c2 | 10767 | .update_time = btrfs_update_time, |
ef3b9af5 | 10768 | .tmpfile = btrfs_tmpfile, |
39279cc3 | 10769 | }; |
6e1d5dcc | 10770 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 10771 | .lookup = btrfs_lookup, |
fdebe2bd | 10772 | .permission = btrfs_permission, |
93fd63c2 | 10773 | .update_time = btrfs_update_time, |
39279cc3 | 10774 | }; |
76dda93c | 10775 | |
828c0950 | 10776 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
10777 | .llseek = generic_file_llseek, |
10778 | .read = generic_read_dir, | |
02dbfc99 | 10779 | .iterate_shared = btrfs_real_readdir, |
23b5ec74 | 10780 | .open = btrfs_opendir, |
34287aa3 | 10781 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 10782 | #ifdef CONFIG_COMPAT |
4c63c245 | 10783 | .compat_ioctl = btrfs_compat_ioctl, |
39279cc3 | 10784 | #endif |
6bf13c0c | 10785 | .release = btrfs_release_file, |
e02119d5 | 10786 | .fsync = btrfs_sync_file, |
39279cc3 CM |
10787 | }; |
10788 | ||
20e5506b | 10789 | static const struct extent_io_ops btrfs_extent_io_ops = { |
4d53dddb | 10790 | /* mandatory callbacks */ |
065631f6 | 10791 | .submit_bio_hook = btrfs_submit_bio_hook, |
07157aac | 10792 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
4d53dddb | 10793 | .merge_bio_hook = btrfs_merge_bio_hook, |
9d0d1c8b | 10794 | .readpage_io_failed_hook = btrfs_readpage_io_failed_hook, |
c6100a4b JB |
10795 | .tree_fs_info = iotree_fs_info, |
10796 | .set_range_writeback = btrfs_set_range_writeback, | |
4d53dddb DS |
10797 | |
10798 | /* optional callbacks */ | |
10799 | .fill_delalloc = run_delalloc_range, | |
e6dcd2dc | 10800 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 10801 | .writepage_start_hook = btrfs_writepage_start_hook, |
b0c68f8b CM |
10802 | .set_bit_hook = btrfs_set_bit_hook, |
10803 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
10804 | .merge_extent_hook = btrfs_merge_extent_hook, |
10805 | .split_extent_hook = btrfs_split_extent_hook, | |
c6100a4b | 10806 | .check_extent_io_range = btrfs_check_extent_io_range, |
07157aac CM |
10807 | }; |
10808 | ||
35054394 CM |
10809 | /* |
10810 | * btrfs doesn't support the bmap operation because swapfiles | |
10811 | * use bmap to make a mapping of extents in the file. They assume | |
10812 | * these extents won't change over the life of the file and they | |
10813 | * use the bmap result to do IO directly to the drive. | |
10814 | * | |
10815 | * the btrfs bmap call would return logical addresses that aren't | |
10816 | * suitable for IO and they also will change frequently as COW | |
10817 | * operations happen. So, swapfile + btrfs == corruption. | |
10818 | * | |
10819 | * For now we're avoiding this by dropping bmap. | |
10820 | */ | |
7f09410b | 10821 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
10822 | .readpage = btrfs_readpage, |
10823 | .writepage = btrfs_writepage, | |
b293f02e | 10824 | .writepages = btrfs_writepages, |
3ab2fb5a | 10825 | .readpages = btrfs_readpages, |
16432985 | 10826 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
10827 | .invalidatepage = btrfs_invalidatepage, |
10828 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 10829 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 10830 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
10831 | }; |
10832 | ||
7f09410b | 10833 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
10834 | .readpage = btrfs_readpage, |
10835 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
10836 | .invalidatepage = btrfs_invalidatepage, |
10837 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
10838 | }; |
10839 | ||
6e1d5dcc | 10840 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
10841 | .getattr = btrfs_getattr, |
10842 | .setattr = btrfs_setattr, | |
5103e947 | 10843 | .listxattr = btrfs_listxattr, |
fdebe2bd | 10844 | .permission = btrfs_permission, |
1506fcc8 | 10845 | .fiemap = btrfs_fiemap, |
4e34e719 | 10846 | .get_acl = btrfs_get_acl, |
996a710d | 10847 | .set_acl = btrfs_set_acl, |
e41f941a | 10848 | .update_time = btrfs_update_time, |
39279cc3 | 10849 | }; |
6e1d5dcc | 10850 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
10851 | .getattr = btrfs_getattr, |
10852 | .setattr = btrfs_setattr, | |
fdebe2bd | 10853 | .permission = btrfs_permission, |
33268eaf | 10854 | .listxattr = btrfs_listxattr, |
4e34e719 | 10855 | .get_acl = btrfs_get_acl, |
996a710d | 10856 | .set_acl = btrfs_set_acl, |
e41f941a | 10857 | .update_time = btrfs_update_time, |
618e21d5 | 10858 | }; |
6e1d5dcc | 10859 | static const struct inode_operations btrfs_symlink_inode_operations = { |
6b255391 | 10860 | .get_link = page_get_link, |
f209561a | 10861 | .getattr = btrfs_getattr, |
22c44fe6 | 10862 | .setattr = btrfs_setattr, |
fdebe2bd | 10863 | .permission = btrfs_permission, |
0279b4cd | 10864 | .listxattr = btrfs_listxattr, |
e41f941a | 10865 | .update_time = btrfs_update_time, |
39279cc3 | 10866 | }; |
76dda93c | 10867 | |
82d339d9 | 10868 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c | 10869 | .d_delete = btrfs_dentry_delete, |
b4aff1f8 | 10870 | .d_release = btrfs_dentry_release, |
76dda93c | 10871 | }; |