]>
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 CM |
1205 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
1206 | 1, 0, NULL, GFP_NOFS); | |
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 CM |
1953 | * extent_io.c submission hook. This does the right thing for csum calculation |
1954 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1955 | */ |
8c27cb35 | 1956 | static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio, |
c6100a4b JB |
1957 | int mirror_num, unsigned long bio_flags, |
1958 | u64 bio_offset) | |
44b8bd7e | 1959 | { |
c6100a4b | 1960 | struct inode *inode = private_data; |
0b246afa | 1961 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
44b8bd7e | 1962 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0d51e28a | 1963 | enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA; |
4e4cbee9 | 1964 | blk_status_t ret = 0; |
19b9bdb0 | 1965 | int skip_sum; |
b812ce28 | 1966 | int async = !atomic_read(&BTRFS_I(inode)->sync_writers); |
44b8bd7e | 1967 | |
6cbff00f | 1968 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1969 | |
70ddc553 | 1970 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) |
0d51e28a | 1971 | metadata = BTRFS_WQ_ENDIO_FREE_SPACE; |
0417341e | 1972 | |
37226b21 | 1973 | if (bio_op(bio) != REQ_OP_WRITE) { |
0b246afa | 1974 | ret = btrfs_bio_wq_end_io(fs_info, bio, metadata); |
5fd02043 | 1975 | if (ret) |
61891923 | 1976 | goto out; |
5fd02043 | 1977 | |
d20f7043 | 1978 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
61891923 SB |
1979 | ret = btrfs_submit_compressed_read(inode, bio, |
1980 | mirror_num, | |
1981 | bio_flags); | |
1982 | goto out; | |
c2db1073 | 1983 | } else if (!skip_sum) { |
2ff7e61e | 1984 | ret = btrfs_lookup_bio_sums(inode, bio, NULL); |
c2db1073 | 1985 | if (ret) |
61891923 | 1986 | goto out; |
c2db1073 | 1987 | } |
4d1b5fb4 | 1988 | goto mapit; |
b812ce28 | 1989 | } else if (async && !skip_sum) { |
17d217fe YZ |
1990 | /* csum items have already been cloned */ |
1991 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1992 | goto mapit; | |
19b9bdb0 | 1993 | /* we're doing a write, do the async checksumming */ |
c6100a4b JB |
1994 | ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags, |
1995 | bio_offset, inode, | |
0b246afa JM |
1996 | __btrfs_submit_bio_start, |
1997 | __btrfs_submit_bio_done); | |
61891923 | 1998 | goto out; |
b812ce28 | 1999 | } else if (!skip_sum) { |
2ff7e61e | 2000 | ret = btrfs_csum_one_bio(inode, bio, 0, 0); |
b812ce28 JB |
2001 | if (ret) |
2002 | goto out; | |
19b9bdb0 CM |
2003 | } |
2004 | ||
0b86a832 | 2005 | mapit: |
2ff7e61e | 2006 | ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); |
61891923 SB |
2007 | |
2008 | out: | |
4e4cbee9 CH |
2009 | if (ret) { |
2010 | bio->bi_status = ret; | |
4246a0b6 CH |
2011 | bio_endio(bio); |
2012 | } | |
61891923 | 2013 | return ret; |
065631f6 | 2014 | } |
6885f308 | 2015 | |
d352ac68 CM |
2016 | /* |
2017 | * given a list of ordered sums record them in the inode. This happens | |
2018 | * at IO completion time based on sums calculated at bio submission time. | |
2019 | */ | |
ba1da2f4 | 2020 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
df9f628e | 2021 | struct inode *inode, struct list_head *list) |
e6dcd2dc | 2022 | { |
e6dcd2dc CM |
2023 | struct btrfs_ordered_sum *sum; |
2024 | ||
c6e30871 | 2025 | list_for_each_entry(sum, list, list) { |
39847c4d | 2026 | trans->adding_csums = 1; |
d20f7043 CM |
2027 | btrfs_csum_file_blocks(trans, |
2028 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
39847c4d | 2029 | trans->adding_csums = 0; |
e6dcd2dc CM |
2030 | } |
2031 | return 0; | |
2032 | } | |
2033 | ||
2ac55d41 | 2034 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
e3b8a485 | 2035 | unsigned int extra_bits, |
ba8b04c1 | 2036 | struct extent_state **cached_state, int dedupe) |
ea8c2819 | 2037 | { |
09cbfeaf | 2038 | WARN_ON((end & (PAGE_SIZE - 1)) == 0); |
ea8c2819 | 2039 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
e3b8a485 | 2040 | extra_bits, cached_state); |
ea8c2819 CM |
2041 | } |
2042 | ||
d352ac68 | 2043 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
2044 | struct btrfs_writepage_fixup { |
2045 | struct page *page; | |
2046 | struct btrfs_work work; | |
2047 | }; | |
2048 | ||
b2950863 | 2049 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
2050 | { |
2051 | struct btrfs_writepage_fixup *fixup; | |
2052 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 2053 | struct extent_state *cached_state = NULL; |
364ecf36 | 2054 | struct extent_changeset *data_reserved = NULL; |
247e743c CM |
2055 | struct page *page; |
2056 | struct inode *inode; | |
2057 | u64 page_start; | |
2058 | u64 page_end; | |
87826df0 | 2059 | int ret; |
247e743c CM |
2060 | |
2061 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
2062 | page = fixup->page; | |
4a096752 | 2063 | again: |
247e743c CM |
2064 | lock_page(page); |
2065 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
2066 | ClearPageChecked(page); | |
2067 | goto out_page; | |
2068 | } | |
2069 | ||
2070 | inode = page->mapping->host; | |
2071 | page_start = page_offset(page); | |
09cbfeaf | 2072 | page_end = page_offset(page) + PAGE_SIZE - 1; |
247e743c | 2073 | |
ff13db41 | 2074 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, |
d0082371 | 2075 | &cached_state); |
4a096752 CM |
2076 | |
2077 | /* already ordered? We're done */ | |
8b62b72b | 2078 | if (PagePrivate2(page)) |
247e743c | 2079 | goto out; |
4a096752 | 2080 | |
a776c6fa | 2081 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start, |
09cbfeaf | 2082 | PAGE_SIZE); |
4a096752 | 2083 | if (ordered) { |
2ac55d41 JB |
2084 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
2085 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
2086 | unlock_page(page); |
2087 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 2088 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
2089 | goto again; |
2090 | } | |
247e743c | 2091 | |
364ecf36 | 2092 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start, |
09cbfeaf | 2093 | PAGE_SIZE); |
87826df0 JM |
2094 | if (ret) { |
2095 | mapping_set_error(page->mapping, ret); | |
2096 | end_extent_writepage(page, ret, page_start, page_end); | |
2097 | ClearPageChecked(page); | |
2098 | goto out; | |
2099 | } | |
2100 | ||
e3b8a485 | 2101 | btrfs_set_extent_delalloc(inode, page_start, page_end, 0, &cached_state, |
ba8b04c1 | 2102 | 0); |
247e743c | 2103 | ClearPageChecked(page); |
87826df0 | 2104 | set_page_dirty(page); |
8b62f87b | 2105 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); |
247e743c | 2106 | out: |
2ac55d41 JB |
2107 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
2108 | &cached_state, GFP_NOFS); | |
247e743c CM |
2109 | out_page: |
2110 | unlock_page(page); | |
09cbfeaf | 2111 | put_page(page); |
b897abec | 2112 | kfree(fixup); |
364ecf36 | 2113 | extent_changeset_free(data_reserved); |
247e743c CM |
2114 | } |
2115 | ||
2116 | /* | |
2117 | * There are a few paths in the higher layers of the kernel that directly | |
2118 | * set the page dirty bit without asking the filesystem if it is a | |
2119 | * good idea. This causes problems because we want to make sure COW | |
2120 | * properly happens and the data=ordered rules are followed. | |
2121 | * | |
c8b97818 | 2122 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
2123 | * hasn't been properly setup for IO. We kick off an async process |
2124 | * to fix it up. The async helper will wait for ordered extents, set | |
2125 | * the delalloc bit and make it safe to write the page. | |
2126 | */ | |
b2950863 | 2127 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
2128 | { |
2129 | struct inode *inode = page->mapping->host; | |
0b246afa | 2130 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
247e743c | 2131 | struct btrfs_writepage_fixup *fixup; |
247e743c | 2132 | |
8b62b72b CM |
2133 | /* this page is properly in the ordered list */ |
2134 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
2135 | return 0; |
2136 | ||
2137 | if (PageChecked(page)) | |
2138 | return -EAGAIN; | |
2139 | ||
2140 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
2141 | if (!fixup) | |
2142 | return -EAGAIN; | |
f421950f | 2143 | |
247e743c | 2144 | SetPageChecked(page); |
09cbfeaf | 2145 | get_page(page); |
9e0af237 LB |
2146 | btrfs_init_work(&fixup->work, btrfs_fixup_helper, |
2147 | btrfs_writepage_fixup_worker, NULL, NULL); | |
247e743c | 2148 | fixup->page = page; |
0b246afa | 2149 | btrfs_queue_work(fs_info->fixup_workers, &fixup->work); |
87826df0 | 2150 | return -EBUSY; |
247e743c CM |
2151 | } |
2152 | ||
d899e052 YZ |
2153 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
2154 | struct inode *inode, u64 file_pos, | |
2155 | u64 disk_bytenr, u64 disk_num_bytes, | |
2156 | u64 num_bytes, u64 ram_bytes, | |
2157 | u8 compression, u8 encryption, | |
2158 | u16 other_encoding, int extent_type) | |
2159 | { | |
2160 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2161 | struct btrfs_file_extent_item *fi; | |
2162 | struct btrfs_path *path; | |
2163 | struct extent_buffer *leaf; | |
2164 | struct btrfs_key ins; | |
a12b877b | 2165 | u64 qg_released; |
1acae57b | 2166 | int extent_inserted = 0; |
d899e052 YZ |
2167 | int ret; |
2168 | ||
2169 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
2170 | if (!path) |
2171 | return -ENOMEM; | |
d899e052 | 2172 | |
a1ed835e CM |
2173 | /* |
2174 | * we may be replacing one extent in the tree with another. | |
2175 | * The new extent is pinned in the extent map, and we don't want | |
2176 | * to drop it from the cache until it is completely in the btree. | |
2177 | * | |
2178 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
2179 | * the caller is expected to unpin it and allow it to be merged | |
2180 | * with the others. | |
2181 | */ | |
1acae57b FDBM |
2182 | ret = __btrfs_drop_extents(trans, root, inode, path, file_pos, |
2183 | file_pos + num_bytes, NULL, 0, | |
2184 | 1, sizeof(*fi), &extent_inserted); | |
79787eaa JM |
2185 | if (ret) |
2186 | goto out; | |
d899e052 | 2187 | |
1acae57b | 2188 | if (!extent_inserted) { |
4a0cc7ca | 2189 | ins.objectid = btrfs_ino(BTRFS_I(inode)); |
1acae57b FDBM |
2190 | ins.offset = file_pos; |
2191 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
2192 | ||
2193 | path->leave_spinning = 1; | |
2194 | ret = btrfs_insert_empty_item(trans, root, path, &ins, | |
2195 | sizeof(*fi)); | |
2196 | if (ret) | |
2197 | goto out; | |
2198 | } | |
d899e052 YZ |
2199 | leaf = path->nodes[0]; |
2200 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2201 | struct btrfs_file_extent_item); | |
2202 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
2203 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
2204 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
2205 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
2206 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
2207 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2208 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
2209 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
2210 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
2211 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 | 2212 | |
d899e052 | 2213 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 2214 | btrfs_release_path(path); |
d899e052 YZ |
2215 | |
2216 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
2217 | |
2218 | ins.objectid = disk_bytenr; | |
2219 | ins.offset = disk_num_bytes; | |
2220 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
a12b877b | 2221 | |
297d750b | 2222 | /* |
5846a3c2 QW |
2223 | * Release the reserved range from inode dirty range map, as it is |
2224 | * already moved into delayed_ref_head | |
297d750b | 2225 | */ |
a12b877b QW |
2226 | ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes); |
2227 | if (ret < 0) | |
2228 | goto out; | |
2229 | qg_released = ret; | |
84f7d8e6 JB |
2230 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
2231 | btrfs_ino(BTRFS_I(inode)), | |
2232 | file_pos, qg_released, &ins); | |
79787eaa | 2233 | out: |
d899e052 | 2234 | btrfs_free_path(path); |
b9473439 | 2235 | |
79787eaa | 2236 | return ret; |
d899e052 YZ |
2237 | } |
2238 | ||
38c227d8 LB |
2239 | /* snapshot-aware defrag */ |
2240 | struct sa_defrag_extent_backref { | |
2241 | struct rb_node node; | |
2242 | struct old_sa_defrag_extent *old; | |
2243 | u64 root_id; | |
2244 | u64 inum; | |
2245 | u64 file_pos; | |
2246 | u64 extent_offset; | |
2247 | u64 num_bytes; | |
2248 | u64 generation; | |
2249 | }; | |
2250 | ||
2251 | struct old_sa_defrag_extent { | |
2252 | struct list_head list; | |
2253 | struct new_sa_defrag_extent *new; | |
2254 | ||
2255 | u64 extent_offset; | |
2256 | u64 bytenr; | |
2257 | u64 offset; | |
2258 | u64 len; | |
2259 | int count; | |
2260 | }; | |
2261 | ||
2262 | struct new_sa_defrag_extent { | |
2263 | struct rb_root root; | |
2264 | struct list_head head; | |
2265 | struct btrfs_path *path; | |
2266 | struct inode *inode; | |
2267 | u64 file_pos; | |
2268 | u64 len; | |
2269 | u64 bytenr; | |
2270 | u64 disk_len; | |
2271 | u8 compress_type; | |
2272 | }; | |
2273 | ||
2274 | static int backref_comp(struct sa_defrag_extent_backref *b1, | |
2275 | struct sa_defrag_extent_backref *b2) | |
2276 | { | |
2277 | if (b1->root_id < b2->root_id) | |
2278 | return -1; | |
2279 | else if (b1->root_id > b2->root_id) | |
2280 | return 1; | |
2281 | ||
2282 | if (b1->inum < b2->inum) | |
2283 | return -1; | |
2284 | else if (b1->inum > b2->inum) | |
2285 | return 1; | |
2286 | ||
2287 | if (b1->file_pos < b2->file_pos) | |
2288 | return -1; | |
2289 | else if (b1->file_pos > b2->file_pos) | |
2290 | return 1; | |
2291 | ||
2292 | /* | |
2293 | * [------------------------------] ===> (a range of space) | |
2294 | * |<--->| |<---->| =============> (fs/file tree A) | |
2295 | * |<---------------------------->| ===> (fs/file tree B) | |
2296 | * | |
2297 | * A range of space can refer to two file extents in one tree while | |
2298 | * refer to only one file extent in another tree. | |
2299 | * | |
2300 | * So we may process a disk offset more than one time(two extents in A) | |
2301 | * and locate at the same extent(one extent in B), then insert two same | |
2302 | * backrefs(both refer to the extent in B). | |
2303 | */ | |
2304 | return 0; | |
2305 | } | |
2306 | ||
2307 | static void backref_insert(struct rb_root *root, | |
2308 | struct sa_defrag_extent_backref *backref) | |
2309 | { | |
2310 | struct rb_node **p = &root->rb_node; | |
2311 | struct rb_node *parent = NULL; | |
2312 | struct sa_defrag_extent_backref *entry; | |
2313 | int ret; | |
2314 | ||
2315 | while (*p) { | |
2316 | parent = *p; | |
2317 | entry = rb_entry(parent, struct sa_defrag_extent_backref, node); | |
2318 | ||
2319 | ret = backref_comp(backref, entry); | |
2320 | if (ret < 0) | |
2321 | p = &(*p)->rb_left; | |
2322 | else | |
2323 | p = &(*p)->rb_right; | |
2324 | } | |
2325 | ||
2326 | rb_link_node(&backref->node, parent, p); | |
2327 | rb_insert_color(&backref->node, root); | |
2328 | } | |
2329 | ||
2330 | /* | |
2331 | * Note the backref might has changed, and in this case we just return 0. | |
2332 | */ | |
2333 | static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id, | |
2334 | void *ctx) | |
2335 | { | |
2336 | struct btrfs_file_extent_item *extent; | |
38c227d8 LB |
2337 | struct old_sa_defrag_extent *old = ctx; |
2338 | struct new_sa_defrag_extent *new = old->new; | |
2339 | struct btrfs_path *path = new->path; | |
2340 | struct btrfs_key key; | |
2341 | struct btrfs_root *root; | |
2342 | struct sa_defrag_extent_backref *backref; | |
2343 | struct extent_buffer *leaf; | |
2344 | struct inode *inode = new->inode; | |
0b246afa | 2345 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
38c227d8 LB |
2346 | int slot; |
2347 | int ret; | |
2348 | u64 extent_offset; | |
2349 | u64 num_bytes; | |
2350 | ||
2351 | if (BTRFS_I(inode)->root->root_key.objectid == root_id && | |
4a0cc7ca | 2352 | inum == btrfs_ino(BTRFS_I(inode))) |
38c227d8 LB |
2353 | return 0; |
2354 | ||
2355 | key.objectid = root_id; | |
2356 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2357 | key.offset = (u64)-1; | |
2358 | ||
38c227d8 LB |
2359 | root = btrfs_read_fs_root_no_name(fs_info, &key); |
2360 | if (IS_ERR(root)) { | |
2361 | if (PTR_ERR(root) == -ENOENT) | |
2362 | return 0; | |
2363 | WARN_ON(1); | |
ab8d0fc4 | 2364 | btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu", |
38c227d8 LB |
2365 | inum, offset, root_id); |
2366 | return PTR_ERR(root); | |
2367 | } | |
2368 | ||
2369 | key.objectid = inum; | |
2370 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2371 | if (offset > (u64)-1 << 32) | |
2372 | key.offset = 0; | |
2373 | else | |
2374 | key.offset = offset; | |
2375 | ||
2376 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
fae7f21c | 2377 | if (WARN_ON(ret < 0)) |
38c227d8 | 2378 | return ret; |
50f1319c | 2379 | ret = 0; |
38c227d8 LB |
2380 | |
2381 | while (1) { | |
2382 | cond_resched(); | |
2383 | ||
2384 | leaf = path->nodes[0]; | |
2385 | slot = path->slots[0]; | |
2386 | ||
2387 | if (slot >= btrfs_header_nritems(leaf)) { | |
2388 | ret = btrfs_next_leaf(root, path); | |
2389 | if (ret < 0) { | |
2390 | goto out; | |
2391 | } else if (ret > 0) { | |
2392 | ret = 0; | |
2393 | goto out; | |
2394 | } | |
2395 | continue; | |
2396 | } | |
2397 | ||
2398 | path->slots[0]++; | |
2399 | ||
2400 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
2401 | ||
2402 | if (key.objectid > inum) | |
2403 | goto out; | |
2404 | ||
2405 | if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY) | |
2406 | continue; | |
2407 | ||
2408 | extent = btrfs_item_ptr(leaf, slot, | |
2409 | struct btrfs_file_extent_item); | |
2410 | ||
2411 | if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr) | |
2412 | continue; | |
2413 | ||
e68afa49 LB |
2414 | /* |
2415 | * 'offset' refers to the exact key.offset, | |
2416 | * NOT the 'offset' field in btrfs_extent_data_ref, ie. | |
2417 | * (key.offset - extent_offset). | |
2418 | */ | |
2419 | if (key.offset != offset) | |
38c227d8 LB |
2420 | continue; |
2421 | ||
e68afa49 | 2422 | extent_offset = btrfs_file_extent_offset(leaf, extent); |
38c227d8 | 2423 | num_bytes = btrfs_file_extent_num_bytes(leaf, extent); |
e68afa49 | 2424 | |
38c227d8 LB |
2425 | if (extent_offset >= old->extent_offset + old->offset + |
2426 | old->len || extent_offset + num_bytes <= | |
2427 | old->extent_offset + old->offset) | |
2428 | continue; | |
38c227d8 LB |
2429 | break; |
2430 | } | |
2431 | ||
2432 | backref = kmalloc(sizeof(*backref), GFP_NOFS); | |
2433 | if (!backref) { | |
2434 | ret = -ENOENT; | |
2435 | goto out; | |
2436 | } | |
2437 | ||
2438 | backref->root_id = root_id; | |
2439 | backref->inum = inum; | |
e68afa49 | 2440 | backref->file_pos = offset; |
38c227d8 LB |
2441 | backref->num_bytes = num_bytes; |
2442 | backref->extent_offset = extent_offset; | |
2443 | backref->generation = btrfs_file_extent_generation(leaf, extent); | |
2444 | backref->old = old; | |
2445 | backref_insert(&new->root, backref); | |
2446 | old->count++; | |
2447 | out: | |
2448 | btrfs_release_path(path); | |
2449 | WARN_ON(ret); | |
2450 | return ret; | |
2451 | } | |
2452 | ||
2453 | static noinline bool record_extent_backrefs(struct btrfs_path *path, | |
2454 | struct new_sa_defrag_extent *new) | |
2455 | { | |
0b246afa | 2456 | struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb); |
38c227d8 LB |
2457 | struct old_sa_defrag_extent *old, *tmp; |
2458 | int ret; | |
2459 | ||
2460 | new->path = path; | |
2461 | ||
2462 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
e68afa49 LB |
2463 | ret = iterate_inodes_from_logical(old->bytenr + |
2464 | old->extent_offset, fs_info, | |
38c227d8 | 2465 | path, record_one_backref, |
c995ab3c | 2466 | old, false); |
4724b106 JB |
2467 | if (ret < 0 && ret != -ENOENT) |
2468 | return false; | |
38c227d8 LB |
2469 | |
2470 | /* no backref to be processed for this extent */ | |
2471 | if (!old->count) { | |
2472 | list_del(&old->list); | |
2473 | kfree(old); | |
2474 | } | |
2475 | } | |
2476 | ||
2477 | if (list_empty(&new->head)) | |
2478 | return false; | |
2479 | ||
2480 | return true; | |
2481 | } | |
2482 | ||
2483 | static int relink_is_mergable(struct extent_buffer *leaf, | |
2484 | struct btrfs_file_extent_item *fi, | |
116e0024 | 2485 | struct new_sa_defrag_extent *new) |
38c227d8 | 2486 | { |
116e0024 | 2487 | if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr) |
38c227d8 LB |
2488 | return 0; |
2489 | ||
2490 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2491 | return 0; | |
2492 | ||
116e0024 LB |
2493 | if (btrfs_file_extent_compression(leaf, fi) != new->compress_type) |
2494 | return 0; | |
2495 | ||
2496 | if (btrfs_file_extent_encryption(leaf, fi) || | |
38c227d8 LB |
2497 | btrfs_file_extent_other_encoding(leaf, fi)) |
2498 | return 0; | |
2499 | ||
2500 | return 1; | |
2501 | } | |
2502 | ||
2503 | /* | |
2504 | * Note the backref might has changed, and in this case we just return 0. | |
2505 | */ | |
2506 | static noinline int relink_extent_backref(struct btrfs_path *path, | |
2507 | struct sa_defrag_extent_backref *prev, | |
2508 | struct sa_defrag_extent_backref *backref) | |
2509 | { | |
2510 | struct btrfs_file_extent_item *extent; | |
2511 | struct btrfs_file_extent_item *item; | |
2512 | struct btrfs_ordered_extent *ordered; | |
2513 | struct btrfs_trans_handle *trans; | |
38c227d8 LB |
2514 | struct btrfs_root *root; |
2515 | struct btrfs_key key; | |
2516 | struct extent_buffer *leaf; | |
2517 | struct old_sa_defrag_extent *old = backref->old; | |
2518 | struct new_sa_defrag_extent *new = old->new; | |
0b246afa | 2519 | struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb); |
38c227d8 LB |
2520 | struct inode *inode; |
2521 | struct extent_state *cached = NULL; | |
2522 | int ret = 0; | |
2523 | u64 start; | |
2524 | u64 len; | |
2525 | u64 lock_start; | |
2526 | u64 lock_end; | |
2527 | bool merge = false; | |
2528 | int index; | |
2529 | ||
2530 | if (prev && prev->root_id == backref->root_id && | |
2531 | prev->inum == backref->inum && | |
2532 | prev->file_pos + prev->num_bytes == backref->file_pos) | |
2533 | merge = true; | |
2534 | ||
2535 | /* step 1: get root */ | |
2536 | key.objectid = backref->root_id; | |
2537 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2538 | key.offset = (u64)-1; | |
2539 | ||
38c227d8 LB |
2540 | index = srcu_read_lock(&fs_info->subvol_srcu); |
2541 | ||
2542 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2543 | if (IS_ERR(root)) { | |
2544 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2545 | if (PTR_ERR(root) == -ENOENT) | |
2546 | return 0; | |
2547 | return PTR_ERR(root); | |
2548 | } | |
38c227d8 | 2549 | |
bcbba5e6 WS |
2550 | if (btrfs_root_readonly(root)) { |
2551 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2552 | return 0; | |
2553 | } | |
2554 | ||
38c227d8 LB |
2555 | /* step 2: get inode */ |
2556 | key.objectid = backref->inum; | |
2557 | key.type = BTRFS_INODE_ITEM_KEY; | |
2558 | key.offset = 0; | |
2559 | ||
2560 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
2561 | if (IS_ERR(inode)) { | |
2562 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2563 | return 0; | |
2564 | } | |
2565 | ||
2566 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2567 | ||
2568 | /* step 3: relink backref */ | |
2569 | lock_start = backref->file_pos; | |
2570 | lock_end = backref->file_pos + backref->num_bytes - 1; | |
2571 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
ff13db41 | 2572 | &cached); |
38c227d8 LB |
2573 | |
2574 | ordered = btrfs_lookup_first_ordered_extent(inode, lock_end); | |
2575 | if (ordered) { | |
2576 | btrfs_put_ordered_extent(ordered); | |
2577 | goto out_unlock; | |
2578 | } | |
2579 | ||
2580 | trans = btrfs_join_transaction(root); | |
2581 | if (IS_ERR(trans)) { | |
2582 | ret = PTR_ERR(trans); | |
2583 | goto out_unlock; | |
2584 | } | |
2585 | ||
2586 | key.objectid = backref->inum; | |
2587 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2588 | key.offset = backref->file_pos; | |
2589 | ||
2590 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2591 | if (ret < 0) { | |
2592 | goto out_free_path; | |
2593 | } else if (ret > 0) { | |
2594 | ret = 0; | |
2595 | goto out_free_path; | |
2596 | } | |
2597 | ||
2598 | extent = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2599 | struct btrfs_file_extent_item); | |
2600 | ||
2601 | if (btrfs_file_extent_generation(path->nodes[0], extent) != | |
2602 | backref->generation) | |
2603 | goto out_free_path; | |
2604 | ||
2605 | btrfs_release_path(path); | |
2606 | ||
2607 | start = backref->file_pos; | |
2608 | if (backref->extent_offset < old->extent_offset + old->offset) | |
2609 | start += old->extent_offset + old->offset - | |
2610 | backref->extent_offset; | |
2611 | ||
2612 | len = min(backref->extent_offset + backref->num_bytes, | |
2613 | old->extent_offset + old->offset + old->len); | |
2614 | len -= max(backref->extent_offset, old->extent_offset + old->offset); | |
2615 | ||
2616 | ret = btrfs_drop_extents(trans, root, inode, start, | |
2617 | start + len, 1); | |
2618 | if (ret) | |
2619 | goto out_free_path; | |
2620 | again: | |
4a0cc7ca | 2621 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
38c227d8 LB |
2622 | key.type = BTRFS_EXTENT_DATA_KEY; |
2623 | key.offset = start; | |
2624 | ||
a09a0a70 | 2625 | path->leave_spinning = 1; |
38c227d8 LB |
2626 | if (merge) { |
2627 | struct btrfs_file_extent_item *fi; | |
2628 | u64 extent_len; | |
2629 | struct btrfs_key found_key; | |
2630 | ||
3c9665df | 2631 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
38c227d8 LB |
2632 | if (ret < 0) |
2633 | goto out_free_path; | |
2634 | ||
2635 | path->slots[0]--; | |
2636 | leaf = path->nodes[0]; | |
2637 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2638 | ||
2639 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2640 | struct btrfs_file_extent_item); | |
2641 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
2642 | ||
116e0024 LB |
2643 | if (extent_len + found_key.offset == start && |
2644 | relink_is_mergable(leaf, fi, new)) { | |
38c227d8 LB |
2645 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2646 | extent_len + len); | |
2647 | btrfs_mark_buffer_dirty(leaf); | |
2648 | inode_add_bytes(inode, len); | |
2649 | ||
2650 | ret = 1; | |
2651 | goto out_free_path; | |
2652 | } else { | |
2653 | merge = false; | |
2654 | btrfs_release_path(path); | |
2655 | goto again; | |
2656 | } | |
2657 | } | |
2658 | ||
2659 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2660 | sizeof(*extent)); | |
2661 | if (ret) { | |
66642832 | 2662 | btrfs_abort_transaction(trans, ret); |
38c227d8 LB |
2663 | goto out_free_path; |
2664 | } | |
2665 | ||
2666 | leaf = path->nodes[0]; | |
2667 | item = btrfs_item_ptr(leaf, path->slots[0], | |
2668 | struct btrfs_file_extent_item); | |
2669 | btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr); | |
2670 | btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len); | |
2671 | btrfs_set_file_extent_offset(leaf, item, start - new->file_pos); | |
2672 | btrfs_set_file_extent_num_bytes(leaf, item, len); | |
2673 | btrfs_set_file_extent_ram_bytes(leaf, item, new->len); | |
2674 | btrfs_set_file_extent_generation(leaf, item, trans->transid); | |
2675 | btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG); | |
2676 | btrfs_set_file_extent_compression(leaf, item, new->compress_type); | |
2677 | btrfs_set_file_extent_encryption(leaf, item, 0); | |
2678 | btrfs_set_file_extent_other_encoding(leaf, item, 0); | |
2679 | ||
2680 | btrfs_mark_buffer_dirty(leaf); | |
2681 | inode_add_bytes(inode, len); | |
a09a0a70 | 2682 | btrfs_release_path(path); |
38c227d8 | 2683 | |
84f7d8e6 | 2684 | ret = btrfs_inc_extent_ref(trans, root, new->bytenr, |
38c227d8 LB |
2685 | new->disk_len, 0, |
2686 | backref->root_id, backref->inum, | |
b06c4bf5 | 2687 | new->file_pos); /* start - extent_offset */ |
38c227d8 | 2688 | if (ret) { |
66642832 | 2689 | btrfs_abort_transaction(trans, ret); |
38c227d8 LB |
2690 | goto out_free_path; |
2691 | } | |
2692 | ||
2693 | ret = 1; | |
2694 | out_free_path: | |
2695 | btrfs_release_path(path); | |
a09a0a70 | 2696 | path->leave_spinning = 0; |
3a45bb20 | 2697 | btrfs_end_transaction(trans); |
38c227d8 LB |
2698 | out_unlock: |
2699 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
2700 | &cached, GFP_NOFS); | |
2701 | iput(inode); | |
2702 | return ret; | |
2703 | } | |
2704 | ||
6f519564 LB |
2705 | static void free_sa_defrag_extent(struct new_sa_defrag_extent *new) |
2706 | { | |
2707 | struct old_sa_defrag_extent *old, *tmp; | |
2708 | ||
2709 | if (!new) | |
2710 | return; | |
2711 | ||
2712 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
6f519564 LB |
2713 | kfree(old); |
2714 | } | |
2715 | kfree(new); | |
2716 | } | |
2717 | ||
38c227d8 LB |
2718 | static void relink_file_extents(struct new_sa_defrag_extent *new) |
2719 | { | |
0b246afa | 2720 | struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb); |
38c227d8 | 2721 | struct btrfs_path *path; |
38c227d8 LB |
2722 | struct sa_defrag_extent_backref *backref; |
2723 | struct sa_defrag_extent_backref *prev = NULL; | |
2724 | struct inode *inode; | |
2725 | struct btrfs_root *root; | |
2726 | struct rb_node *node; | |
2727 | int ret; | |
2728 | ||
2729 | inode = new->inode; | |
2730 | root = BTRFS_I(inode)->root; | |
2731 | ||
2732 | path = btrfs_alloc_path(); | |
2733 | if (!path) | |
2734 | return; | |
2735 | ||
2736 | if (!record_extent_backrefs(path, new)) { | |
2737 | btrfs_free_path(path); | |
2738 | goto out; | |
2739 | } | |
2740 | btrfs_release_path(path); | |
2741 | ||
2742 | while (1) { | |
2743 | node = rb_first(&new->root); | |
2744 | if (!node) | |
2745 | break; | |
2746 | rb_erase(node, &new->root); | |
2747 | ||
2748 | backref = rb_entry(node, struct sa_defrag_extent_backref, node); | |
2749 | ||
2750 | ret = relink_extent_backref(path, prev, backref); | |
2751 | WARN_ON(ret < 0); | |
2752 | ||
2753 | kfree(prev); | |
2754 | ||
2755 | if (ret == 1) | |
2756 | prev = backref; | |
2757 | else | |
2758 | prev = NULL; | |
2759 | cond_resched(); | |
2760 | } | |
2761 | kfree(prev); | |
2762 | ||
2763 | btrfs_free_path(path); | |
38c227d8 | 2764 | out: |
6f519564 LB |
2765 | free_sa_defrag_extent(new); |
2766 | ||
0b246afa JM |
2767 | atomic_dec(&fs_info->defrag_running); |
2768 | wake_up(&fs_info->transaction_wait); | |
38c227d8 LB |
2769 | } |
2770 | ||
2771 | static struct new_sa_defrag_extent * | |
2772 | record_old_file_extents(struct inode *inode, | |
2773 | struct btrfs_ordered_extent *ordered) | |
2774 | { | |
0b246afa | 2775 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
38c227d8 LB |
2776 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2777 | struct btrfs_path *path; | |
2778 | struct btrfs_key key; | |
6f519564 | 2779 | struct old_sa_defrag_extent *old; |
38c227d8 LB |
2780 | struct new_sa_defrag_extent *new; |
2781 | int ret; | |
2782 | ||
2783 | new = kmalloc(sizeof(*new), GFP_NOFS); | |
2784 | if (!new) | |
2785 | return NULL; | |
2786 | ||
2787 | new->inode = inode; | |
2788 | new->file_pos = ordered->file_offset; | |
2789 | new->len = ordered->len; | |
2790 | new->bytenr = ordered->start; | |
2791 | new->disk_len = ordered->disk_len; | |
2792 | new->compress_type = ordered->compress_type; | |
2793 | new->root = RB_ROOT; | |
2794 | INIT_LIST_HEAD(&new->head); | |
2795 | ||
2796 | path = btrfs_alloc_path(); | |
2797 | if (!path) | |
2798 | goto out_kfree; | |
2799 | ||
4a0cc7ca | 2800 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
38c227d8 LB |
2801 | key.type = BTRFS_EXTENT_DATA_KEY; |
2802 | key.offset = new->file_pos; | |
2803 | ||
2804 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2805 | if (ret < 0) | |
2806 | goto out_free_path; | |
2807 | if (ret > 0 && path->slots[0] > 0) | |
2808 | path->slots[0]--; | |
2809 | ||
2810 | /* find out all the old extents for the file range */ | |
2811 | while (1) { | |
2812 | struct btrfs_file_extent_item *extent; | |
2813 | struct extent_buffer *l; | |
2814 | int slot; | |
2815 | u64 num_bytes; | |
2816 | u64 offset; | |
2817 | u64 end; | |
2818 | u64 disk_bytenr; | |
2819 | u64 extent_offset; | |
2820 | ||
2821 | l = path->nodes[0]; | |
2822 | slot = path->slots[0]; | |
2823 | ||
2824 | if (slot >= btrfs_header_nritems(l)) { | |
2825 | ret = btrfs_next_leaf(root, path); | |
2826 | if (ret < 0) | |
6f519564 | 2827 | goto out_free_path; |
38c227d8 LB |
2828 | else if (ret > 0) |
2829 | break; | |
2830 | continue; | |
2831 | } | |
2832 | ||
2833 | btrfs_item_key_to_cpu(l, &key, slot); | |
2834 | ||
4a0cc7ca | 2835 | if (key.objectid != btrfs_ino(BTRFS_I(inode))) |
38c227d8 LB |
2836 | break; |
2837 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2838 | break; | |
2839 | if (key.offset >= new->file_pos + new->len) | |
2840 | break; | |
2841 | ||
2842 | extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item); | |
2843 | ||
2844 | num_bytes = btrfs_file_extent_num_bytes(l, extent); | |
2845 | if (key.offset + num_bytes < new->file_pos) | |
2846 | goto next; | |
2847 | ||
2848 | disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent); | |
2849 | if (!disk_bytenr) | |
2850 | goto next; | |
2851 | ||
2852 | extent_offset = btrfs_file_extent_offset(l, extent); | |
2853 | ||
2854 | old = kmalloc(sizeof(*old), GFP_NOFS); | |
2855 | if (!old) | |
6f519564 | 2856 | goto out_free_path; |
38c227d8 LB |
2857 | |
2858 | offset = max(new->file_pos, key.offset); | |
2859 | end = min(new->file_pos + new->len, key.offset + num_bytes); | |
2860 | ||
2861 | old->bytenr = disk_bytenr; | |
2862 | old->extent_offset = extent_offset; | |
2863 | old->offset = offset - key.offset; | |
2864 | old->len = end - offset; | |
2865 | old->new = new; | |
2866 | old->count = 0; | |
2867 | list_add_tail(&old->list, &new->head); | |
2868 | next: | |
2869 | path->slots[0]++; | |
2870 | cond_resched(); | |
2871 | } | |
2872 | ||
2873 | btrfs_free_path(path); | |
0b246afa | 2874 | atomic_inc(&fs_info->defrag_running); |
38c227d8 LB |
2875 | |
2876 | return new; | |
2877 | ||
38c227d8 LB |
2878 | out_free_path: |
2879 | btrfs_free_path(path); | |
2880 | out_kfree: | |
6f519564 | 2881 | free_sa_defrag_extent(new); |
38c227d8 LB |
2882 | return NULL; |
2883 | } | |
2884 | ||
2ff7e61e | 2885 | static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
2886 | u64 start, u64 len) |
2887 | { | |
2888 | struct btrfs_block_group_cache *cache; | |
2889 | ||
0b246afa | 2890 | cache = btrfs_lookup_block_group(fs_info, start); |
e570fd27 MX |
2891 | ASSERT(cache); |
2892 | ||
2893 | spin_lock(&cache->lock); | |
2894 | cache->delalloc_bytes -= len; | |
2895 | spin_unlock(&cache->lock); | |
2896 | ||
2897 | btrfs_put_block_group(cache); | |
2898 | } | |
2899 | ||
d352ac68 CM |
2900 | /* as ordered data IO finishes, this gets called so we can finish |
2901 | * an ordered extent if the range of bytes in the file it covers are | |
2902 | * fully written. | |
2903 | */ | |
5fd02043 | 2904 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 2905 | { |
5fd02043 | 2906 | struct inode *inode = ordered_extent->inode; |
0b246afa | 2907 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc | 2908 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 2909 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 2910 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 2911 | struct extent_state *cached_state = NULL; |
38c227d8 | 2912 | struct new_sa_defrag_extent *new = NULL; |
261507a0 | 2913 | int compress_type = 0; |
77cef2ec JB |
2914 | int ret = 0; |
2915 | u64 logical_len = ordered_extent->len; | |
82d5902d | 2916 | bool nolock; |
77cef2ec | 2917 | bool truncated = false; |
a7e3b975 FM |
2918 | bool range_locked = false; |
2919 | bool clear_new_delalloc_bytes = false; | |
2920 | ||
2921 | if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && | |
2922 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) && | |
2923 | !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags)) | |
2924 | clear_new_delalloc_bytes = true; | |
e6dcd2dc | 2925 | |
70ddc553 | 2926 | nolock = btrfs_is_free_space_inode(BTRFS_I(inode)); |
0cb59c99 | 2927 | |
5fd02043 JB |
2928 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
2929 | ret = -EIO; | |
2930 | goto out; | |
2931 | } | |
2932 | ||
7ab7956e NB |
2933 | btrfs_free_io_failure_record(BTRFS_I(inode), |
2934 | ordered_extent->file_offset, | |
2935 | ordered_extent->file_offset + | |
2936 | ordered_extent->len - 1); | |
f612496b | 2937 | |
77cef2ec JB |
2938 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) { |
2939 | truncated = true; | |
2940 | logical_len = ordered_extent->truncated_len; | |
2941 | /* Truncated the entire extent, don't bother adding */ | |
2942 | if (!logical_len) | |
2943 | goto out; | |
2944 | } | |
2945 | ||
c2167754 | 2946 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 2947 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
94ed938a QW |
2948 | |
2949 | /* | |
2950 | * For mwrite(mmap + memset to write) case, we still reserve | |
2951 | * space for NOCOW range. | |
2952 | * As NOCOW won't cause a new delayed ref, just free the space | |
2953 | */ | |
bc42bda2 | 2954 | btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset, |
94ed938a | 2955 | ordered_extent->len); |
6c760c07 JB |
2956 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2957 | if (nolock) | |
2958 | trans = btrfs_join_transaction_nolock(root); | |
2959 | else | |
2960 | trans = btrfs_join_transaction(root); | |
2961 | if (IS_ERR(trans)) { | |
2962 | ret = PTR_ERR(trans); | |
2963 | trans = NULL; | |
2964 | goto out; | |
c2167754 | 2965 | } |
69fe2d75 | 2966 | trans->block_rsv = &BTRFS_I(inode)->block_rsv; |
6c760c07 JB |
2967 | ret = btrfs_update_inode_fallback(trans, root, inode); |
2968 | if (ret) /* -ENOMEM or corruption */ | |
66642832 | 2969 | btrfs_abort_transaction(trans, ret); |
c2167754 YZ |
2970 | goto out; |
2971 | } | |
e6dcd2dc | 2972 | |
a7e3b975 | 2973 | range_locked = true; |
2ac55d41 JB |
2974 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
2975 | ordered_extent->file_offset + ordered_extent->len - 1, | |
ff13db41 | 2976 | &cached_state); |
e6dcd2dc | 2977 | |
38c227d8 LB |
2978 | ret = test_range_bit(io_tree, ordered_extent->file_offset, |
2979 | ordered_extent->file_offset + ordered_extent->len - 1, | |
452e62b7 | 2980 | EXTENT_DEFRAG, 0, cached_state); |
38c227d8 LB |
2981 | if (ret) { |
2982 | u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item); | |
8101c8db | 2983 | if (0 && last_snapshot >= BTRFS_I(inode)->generation) |
38c227d8 LB |
2984 | /* the inode is shared */ |
2985 | new = record_old_file_extents(inode, ordered_extent); | |
2986 | ||
2987 | clear_extent_bit(io_tree, ordered_extent->file_offset, | |
2988 | ordered_extent->file_offset + ordered_extent->len - 1, | |
2989 | EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS); | |
2990 | } | |
2991 | ||
0cb59c99 | 2992 | if (nolock) |
7a7eaa40 | 2993 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 2994 | else |
7a7eaa40 | 2995 | trans = btrfs_join_transaction(root); |
79787eaa JM |
2996 | if (IS_ERR(trans)) { |
2997 | ret = PTR_ERR(trans); | |
2998 | trans = NULL; | |
a7e3b975 | 2999 | goto out; |
79787eaa | 3000 | } |
a79b7d4b | 3001 | |
69fe2d75 | 3002 | trans->block_rsv = &BTRFS_I(inode)->block_rsv; |
c2167754 | 3003 | |
c8b97818 | 3004 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 3005 | compress_type = ordered_extent->compress_type; |
d899e052 | 3006 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 3007 | BUG_ON(compress_type); |
b430b775 JM |
3008 | btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset, |
3009 | ordered_extent->len); | |
7a6d7067 | 3010 | ret = btrfs_mark_extent_written(trans, BTRFS_I(inode), |
d899e052 YZ |
3011 | ordered_extent->file_offset, |
3012 | ordered_extent->file_offset + | |
77cef2ec | 3013 | logical_len); |
d899e052 | 3014 | } else { |
0b246afa | 3015 | BUG_ON(root == fs_info->tree_root); |
d899e052 YZ |
3016 | ret = insert_reserved_file_extent(trans, inode, |
3017 | ordered_extent->file_offset, | |
3018 | ordered_extent->start, | |
3019 | ordered_extent->disk_len, | |
77cef2ec | 3020 | logical_len, logical_len, |
261507a0 | 3021 | compress_type, 0, 0, |
d899e052 | 3022 | BTRFS_FILE_EXTENT_REG); |
e570fd27 | 3023 | if (!ret) |
2ff7e61e | 3024 | btrfs_release_delalloc_bytes(fs_info, |
e570fd27 MX |
3025 | ordered_extent->start, |
3026 | ordered_extent->disk_len); | |
d899e052 | 3027 | } |
5dc562c5 JB |
3028 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
3029 | ordered_extent->file_offset, ordered_extent->len, | |
3030 | trans->transid); | |
79787eaa | 3031 | if (ret < 0) { |
66642832 | 3032 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 3033 | goto out; |
79787eaa | 3034 | } |
2ac55d41 | 3035 | |
df9f628e | 3036 | add_pending_csums(trans, inode, &ordered_extent->list); |
e6dcd2dc | 3037 | |
6c760c07 JB |
3038 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
3039 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
3040 | if (ret) { /* -ENOMEM or corruption */ | |
66642832 | 3041 | btrfs_abort_transaction(trans, ret); |
a7e3b975 | 3042 | goto out; |
1ef30be1 JB |
3043 | } |
3044 | ret = 0; | |
c2167754 | 3045 | out: |
a7e3b975 FM |
3046 | if (range_locked || clear_new_delalloc_bytes) { |
3047 | unsigned int clear_bits = 0; | |
3048 | ||
3049 | if (range_locked) | |
3050 | clear_bits |= EXTENT_LOCKED; | |
3051 | if (clear_new_delalloc_bytes) | |
3052 | clear_bits |= EXTENT_DELALLOC_NEW; | |
3053 | clear_extent_bit(&BTRFS_I(inode)->io_tree, | |
3054 | ordered_extent->file_offset, | |
3055 | ordered_extent->file_offset + | |
3056 | ordered_extent->len - 1, | |
3057 | clear_bits, | |
3058 | (clear_bits & EXTENT_LOCKED) ? 1 : 0, | |
3059 | 0, &cached_state, GFP_NOFS); | |
3060 | } | |
3061 | ||
a698d075 | 3062 | if (trans) |
3a45bb20 | 3063 | btrfs_end_transaction(trans); |
0cb59c99 | 3064 | |
77cef2ec JB |
3065 | if (ret || truncated) { |
3066 | u64 start, end; | |
3067 | ||
3068 | if (truncated) | |
3069 | start = ordered_extent->file_offset + logical_len; | |
3070 | else | |
3071 | start = ordered_extent->file_offset; | |
3072 | end = ordered_extent->file_offset + ordered_extent->len - 1; | |
3073 | clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS); | |
3074 | ||
3075 | /* Drop the cache for the part of the extent we didn't write. */ | |
dcdbc059 | 3076 | btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0); |
5fd02043 | 3077 | |
0bec9ef5 JB |
3078 | /* |
3079 | * If the ordered extent had an IOERR or something else went | |
3080 | * wrong we need to return the space for this ordered extent | |
77cef2ec JB |
3081 | * back to the allocator. We only free the extent in the |
3082 | * truncated case if we didn't write out the extent at all. | |
0bec9ef5 | 3083 | */ |
77cef2ec JB |
3084 | if ((ret || !logical_len) && |
3085 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && | |
0bec9ef5 | 3086 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) |
2ff7e61e JM |
3087 | btrfs_free_reserved_extent(fs_info, |
3088 | ordered_extent->start, | |
e570fd27 | 3089 | ordered_extent->disk_len, 1); |
0bec9ef5 JB |
3090 | } |
3091 | ||
3092 | ||
5fd02043 | 3093 | /* |
8bad3c02 LB |
3094 | * This needs to be done to make sure anybody waiting knows we are done |
3095 | * updating everything for this ordered extent. | |
5fd02043 JB |
3096 | */ |
3097 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
3098 | ||
38c227d8 | 3099 | /* for snapshot-aware defrag */ |
6f519564 LB |
3100 | if (new) { |
3101 | if (ret) { | |
3102 | free_sa_defrag_extent(new); | |
0b246afa | 3103 | atomic_dec(&fs_info->defrag_running); |
6f519564 LB |
3104 | } else { |
3105 | relink_file_extents(new); | |
3106 | } | |
3107 | } | |
38c227d8 | 3108 | |
e6dcd2dc CM |
3109 | /* once for us */ |
3110 | btrfs_put_ordered_extent(ordered_extent); | |
3111 | /* once for the tree */ | |
3112 | btrfs_put_ordered_extent(ordered_extent); | |
3113 | ||
5fd02043 JB |
3114 | return ret; |
3115 | } | |
3116 | ||
3117 | static void finish_ordered_fn(struct btrfs_work *work) | |
3118 | { | |
3119 | struct btrfs_ordered_extent *ordered_extent; | |
3120 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
3121 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
3122 | } |
3123 | ||
c3988d63 | 3124 | static void btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
3125 | struct extent_state *state, int uptodate) |
3126 | { | |
5fd02043 | 3127 | struct inode *inode = page->mapping->host; |
0b246afa | 3128 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5fd02043 | 3129 | struct btrfs_ordered_extent *ordered_extent = NULL; |
9e0af237 LB |
3130 | struct btrfs_workqueue *wq; |
3131 | btrfs_work_func_t func; | |
5fd02043 | 3132 | |
1abe9b8a | 3133 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
3134 | ||
8b62b72b | 3135 | ClearPagePrivate2(page); |
5fd02043 JB |
3136 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
3137 | end - start + 1, uptodate)) | |
c3988d63 | 3138 | return; |
5fd02043 | 3139 | |
70ddc553 | 3140 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) { |
0b246afa | 3141 | wq = fs_info->endio_freespace_worker; |
9e0af237 LB |
3142 | func = btrfs_freespace_write_helper; |
3143 | } else { | |
0b246afa | 3144 | wq = fs_info->endio_write_workers; |
9e0af237 LB |
3145 | func = btrfs_endio_write_helper; |
3146 | } | |
5fd02043 | 3147 | |
9e0af237 LB |
3148 | btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL, |
3149 | NULL); | |
3150 | btrfs_queue_work(wq, &ordered_extent->work); | |
211f90e6 CM |
3151 | } |
3152 | ||
dc380aea MX |
3153 | static int __readpage_endio_check(struct inode *inode, |
3154 | struct btrfs_io_bio *io_bio, | |
3155 | int icsum, struct page *page, | |
3156 | int pgoff, u64 start, size_t len) | |
3157 | { | |
3158 | char *kaddr; | |
3159 | u32 csum_expected; | |
3160 | u32 csum = ~(u32)0; | |
dc380aea MX |
3161 | |
3162 | csum_expected = *(((u32 *)io_bio->csum) + icsum); | |
3163 | ||
3164 | kaddr = kmap_atomic(page); | |
3165 | csum = btrfs_csum_data(kaddr + pgoff, csum, len); | |
0b5e3daf | 3166 | btrfs_csum_final(csum, (u8 *)&csum); |
dc380aea MX |
3167 | if (csum != csum_expected) |
3168 | goto zeroit; | |
3169 | ||
3170 | kunmap_atomic(kaddr); | |
3171 | return 0; | |
3172 | zeroit: | |
0970a22e | 3173 | btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected, |
6f6b643e | 3174 | io_bio->mirror_num); |
dc380aea MX |
3175 | memset(kaddr + pgoff, 1, len); |
3176 | flush_dcache_page(page); | |
3177 | kunmap_atomic(kaddr); | |
dc380aea MX |
3178 | return -EIO; |
3179 | } | |
3180 | ||
d352ac68 CM |
3181 | /* |
3182 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
3183 | * if there's a match, we allow the bio to finish. If not, the code in |
3184 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 3185 | */ |
facc8a22 MX |
3186 | static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio, |
3187 | u64 phy_offset, struct page *page, | |
3188 | u64 start, u64 end, int mirror) | |
07157aac | 3189 | { |
4eee4fa4 | 3190 | size_t offset = start - page_offset(page); |
07157aac | 3191 | struct inode *inode = page->mapping->host; |
d1310b2e | 3192 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
ff79f819 | 3193 | struct btrfs_root *root = BTRFS_I(inode)->root; |
d1310b2e | 3194 | |
d20f7043 CM |
3195 | if (PageChecked(page)) { |
3196 | ClearPageChecked(page); | |
dc380aea | 3197 | return 0; |
d20f7043 | 3198 | } |
6cbff00f CH |
3199 | |
3200 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
dc380aea | 3201 | return 0; |
17d217fe YZ |
3202 | |
3203 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 3204 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
91166212 | 3205 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM); |
b6cda9bc | 3206 | return 0; |
17d217fe | 3207 | } |
d20f7043 | 3208 | |
facc8a22 | 3209 | phy_offset >>= inode->i_sb->s_blocksize_bits; |
dc380aea MX |
3210 | return __readpage_endio_check(inode, io_bio, phy_offset, page, offset, |
3211 | start, (size_t)(end - start + 1)); | |
07157aac | 3212 | } |
b888db2b | 3213 | |
24bbcf04 YZ |
3214 | void btrfs_add_delayed_iput(struct inode *inode) |
3215 | { | |
0b246afa | 3216 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
8089fe62 | 3217 | struct btrfs_inode *binode = BTRFS_I(inode); |
24bbcf04 YZ |
3218 | |
3219 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
3220 | return; | |
3221 | ||
24bbcf04 | 3222 | spin_lock(&fs_info->delayed_iput_lock); |
8089fe62 DS |
3223 | if (binode->delayed_iput_count == 0) { |
3224 | ASSERT(list_empty(&binode->delayed_iput)); | |
3225 | list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs); | |
3226 | } else { | |
3227 | binode->delayed_iput_count++; | |
3228 | } | |
24bbcf04 YZ |
3229 | spin_unlock(&fs_info->delayed_iput_lock); |
3230 | } | |
3231 | ||
2ff7e61e | 3232 | void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info) |
24bbcf04 | 3233 | { |
24bbcf04 | 3234 | |
24bbcf04 | 3235 | spin_lock(&fs_info->delayed_iput_lock); |
8089fe62 DS |
3236 | while (!list_empty(&fs_info->delayed_iputs)) { |
3237 | struct btrfs_inode *inode; | |
3238 | ||
3239 | inode = list_first_entry(&fs_info->delayed_iputs, | |
3240 | struct btrfs_inode, delayed_iput); | |
3241 | if (inode->delayed_iput_count) { | |
3242 | inode->delayed_iput_count--; | |
3243 | list_move_tail(&inode->delayed_iput, | |
3244 | &fs_info->delayed_iputs); | |
3245 | } else { | |
3246 | list_del_init(&inode->delayed_iput); | |
3247 | } | |
3248 | spin_unlock(&fs_info->delayed_iput_lock); | |
3249 | iput(&inode->vfs_inode); | |
3250 | spin_lock(&fs_info->delayed_iput_lock); | |
24bbcf04 | 3251 | } |
8089fe62 | 3252 | spin_unlock(&fs_info->delayed_iput_lock); |
24bbcf04 YZ |
3253 | } |
3254 | ||
d68fc57b | 3255 | /* |
42b2aa86 | 3256 | * This is called in transaction commit time. If there are no orphan |
d68fc57b YZ |
3257 | * files in the subvolume, it removes orphan item and frees block_rsv |
3258 | * structure. | |
3259 | */ | |
3260 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
3261 | struct btrfs_root *root) | |
3262 | { | |
0b246afa | 3263 | struct btrfs_fs_info *fs_info = root->fs_info; |
90290e19 | 3264 | struct btrfs_block_rsv *block_rsv; |
d68fc57b YZ |
3265 | int ret; |
3266 | ||
8a35d95f | 3267 | if (atomic_read(&root->orphan_inodes) || |
d68fc57b YZ |
3268 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) |
3269 | return; | |
3270 | ||
90290e19 | 3271 | spin_lock(&root->orphan_lock); |
8a35d95f | 3272 | if (atomic_read(&root->orphan_inodes)) { |
90290e19 JB |
3273 | spin_unlock(&root->orphan_lock); |
3274 | return; | |
3275 | } | |
3276 | ||
3277 | if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) { | |
3278 | spin_unlock(&root->orphan_lock); | |
3279 | return; | |
3280 | } | |
3281 | ||
3282 | block_rsv = root->orphan_block_rsv; | |
3283 | root->orphan_block_rsv = NULL; | |
3284 | spin_unlock(&root->orphan_lock); | |
3285 | ||
27cdeb70 | 3286 | if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) && |
d68fc57b | 3287 | btrfs_root_refs(&root->root_item) > 0) { |
0b246afa | 3288 | ret = btrfs_del_orphan_item(trans, fs_info->tree_root, |
d68fc57b | 3289 | root->root_key.objectid); |
4ef31a45 | 3290 | if (ret) |
66642832 | 3291 | btrfs_abort_transaction(trans, ret); |
4ef31a45 | 3292 | else |
27cdeb70 MX |
3293 | clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, |
3294 | &root->state); | |
d68fc57b YZ |
3295 | } |
3296 | ||
90290e19 JB |
3297 | if (block_rsv) { |
3298 | WARN_ON(block_rsv->size > 0); | |
2ff7e61e | 3299 | btrfs_free_block_rsv(fs_info, block_rsv); |
d68fc57b YZ |
3300 | } |
3301 | } | |
3302 | ||
7b128766 JB |
3303 | /* |
3304 | * This creates an orphan entry for the given inode in case something goes | |
3305 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
3306 | * |
3307 | * NOTE: caller of this function should reserve 5 units of metadata for | |
3308 | * this function. | |
7b128766 | 3309 | */ |
73f2e545 NB |
3310 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, |
3311 | struct btrfs_inode *inode) | |
7b128766 | 3312 | { |
73f2e545 NB |
3313 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
3314 | struct btrfs_root *root = inode->root; | |
d68fc57b YZ |
3315 | struct btrfs_block_rsv *block_rsv = NULL; |
3316 | int reserve = 0; | |
3317 | int insert = 0; | |
3318 | int ret; | |
7b128766 | 3319 | |
d68fc57b | 3320 | if (!root->orphan_block_rsv) { |
2ff7e61e JM |
3321 | block_rsv = btrfs_alloc_block_rsv(fs_info, |
3322 | BTRFS_BLOCK_RSV_TEMP); | |
b532402e TI |
3323 | if (!block_rsv) |
3324 | return -ENOMEM; | |
d68fc57b | 3325 | } |
7b128766 | 3326 | |
d68fc57b YZ |
3327 | spin_lock(&root->orphan_lock); |
3328 | if (!root->orphan_block_rsv) { | |
3329 | root->orphan_block_rsv = block_rsv; | |
3330 | } else if (block_rsv) { | |
2ff7e61e | 3331 | btrfs_free_block_rsv(fs_info, block_rsv); |
d68fc57b | 3332 | block_rsv = NULL; |
7b128766 | 3333 | } |
7b128766 | 3334 | |
8a35d95f | 3335 | if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
73f2e545 | 3336 | &inode->runtime_flags)) { |
d68fc57b YZ |
3337 | #if 0 |
3338 | /* | |
3339 | * For proper ENOSPC handling, we should do orphan | |
3340 | * cleanup when mounting. But this introduces backward | |
3341 | * compatibility issue. | |
3342 | */ | |
3343 | if (!xchg(&root->orphan_item_inserted, 1)) | |
3344 | insert = 2; | |
3345 | else | |
3346 | insert = 1; | |
3347 | #endif | |
3348 | insert = 1; | |
321f0e70 | 3349 | atomic_inc(&root->orphan_inodes); |
7b128766 JB |
3350 | } |
3351 | ||
72ac3c0d | 3352 | if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
73f2e545 | 3353 | &inode->runtime_flags)) |
d68fc57b | 3354 | reserve = 1; |
d68fc57b | 3355 | spin_unlock(&root->orphan_lock); |
7b128766 | 3356 | |
d68fc57b YZ |
3357 | /* grab metadata reservation from transaction handle */ |
3358 | if (reserve) { | |
3359 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
3b6571c1 JB |
3360 | ASSERT(!ret); |
3361 | if (ret) { | |
3362 | atomic_dec(&root->orphan_inodes); | |
3363 | clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, | |
73f2e545 | 3364 | &inode->runtime_flags); |
3b6571c1 JB |
3365 | if (insert) |
3366 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, | |
73f2e545 | 3367 | &inode->runtime_flags); |
3b6571c1 JB |
3368 | return ret; |
3369 | } | |
d68fc57b | 3370 | } |
7b128766 | 3371 | |
d68fc57b YZ |
3372 | /* insert an orphan item to track this unlinked/truncated file */ |
3373 | if (insert >= 1) { | |
73f2e545 | 3374 | ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode)); |
4ef31a45 | 3375 | if (ret) { |
703c88e0 | 3376 | atomic_dec(&root->orphan_inodes); |
4ef31a45 JB |
3377 | if (reserve) { |
3378 | clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, | |
73f2e545 | 3379 | &inode->runtime_flags); |
4ef31a45 JB |
3380 | btrfs_orphan_release_metadata(inode); |
3381 | } | |
3382 | if (ret != -EEXIST) { | |
e8e7cff6 | 3383 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
73f2e545 | 3384 | &inode->runtime_flags); |
66642832 | 3385 | btrfs_abort_transaction(trans, ret); |
4ef31a45 JB |
3386 | return ret; |
3387 | } | |
79787eaa JM |
3388 | } |
3389 | ret = 0; | |
d68fc57b YZ |
3390 | } |
3391 | ||
3392 | /* insert an orphan item to track subvolume contains orphan files */ | |
3393 | if (insert >= 2) { | |
0b246afa | 3394 | ret = btrfs_insert_orphan_item(trans, fs_info->tree_root, |
d68fc57b | 3395 | root->root_key.objectid); |
79787eaa | 3396 | if (ret && ret != -EEXIST) { |
66642832 | 3397 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3398 | return ret; |
3399 | } | |
d68fc57b YZ |
3400 | } |
3401 | return 0; | |
7b128766 JB |
3402 | } |
3403 | ||
3404 | /* | |
3405 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
3406 | * item for this particular inode. | |
3407 | */ | |
48a3b636 | 3408 | static int btrfs_orphan_del(struct btrfs_trans_handle *trans, |
3d6ae7bb | 3409 | struct btrfs_inode *inode) |
7b128766 | 3410 | { |
3d6ae7bb | 3411 | struct btrfs_root *root = inode->root; |
d68fc57b YZ |
3412 | int delete_item = 0; |
3413 | int release_rsv = 0; | |
7b128766 JB |
3414 | int ret = 0; |
3415 | ||
d68fc57b | 3416 | spin_lock(&root->orphan_lock); |
8a35d95f | 3417 | if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3d6ae7bb | 3418 | &inode->runtime_flags)) |
d68fc57b | 3419 | delete_item = 1; |
7b128766 | 3420 | |
72ac3c0d | 3421 | if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
3d6ae7bb | 3422 | &inode->runtime_flags)) |
d68fc57b | 3423 | release_rsv = 1; |
d68fc57b | 3424 | spin_unlock(&root->orphan_lock); |
7b128766 | 3425 | |
703c88e0 | 3426 | if (delete_item) { |
8a35d95f | 3427 | atomic_dec(&root->orphan_inodes); |
703c88e0 FDBM |
3428 | if (trans) |
3429 | ret = btrfs_del_orphan_item(trans, root, | |
3d6ae7bb | 3430 | btrfs_ino(inode)); |
8a35d95f | 3431 | } |
7b128766 | 3432 | |
703c88e0 FDBM |
3433 | if (release_rsv) |
3434 | btrfs_orphan_release_metadata(inode); | |
3435 | ||
4ef31a45 | 3436 | return ret; |
7b128766 JB |
3437 | } |
3438 | ||
3439 | /* | |
3440 | * this cleans up any orphans that may be left on the list from the last use | |
3441 | * of this root. | |
3442 | */ | |
66b4ffd1 | 3443 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 | 3444 | { |
0b246afa | 3445 | struct btrfs_fs_info *fs_info = root->fs_info; |
7b128766 JB |
3446 | struct btrfs_path *path; |
3447 | struct extent_buffer *leaf; | |
7b128766 JB |
3448 | struct btrfs_key key, found_key; |
3449 | struct btrfs_trans_handle *trans; | |
3450 | struct inode *inode; | |
8f6d7f4f | 3451 | u64 last_objectid = 0; |
7b128766 JB |
3452 | int ret = 0, nr_unlink = 0, nr_truncate = 0; |
3453 | ||
d68fc57b | 3454 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 3455 | return 0; |
c71bf099 YZ |
3456 | |
3457 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
3458 | if (!path) { |
3459 | ret = -ENOMEM; | |
3460 | goto out; | |
3461 | } | |
e4058b54 | 3462 | path->reada = READA_BACK; |
7b128766 JB |
3463 | |
3464 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
962a298f | 3465 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
7b128766 JB |
3466 | key.offset = (u64)-1; |
3467 | ||
7b128766 JB |
3468 | while (1) { |
3469 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
3470 | if (ret < 0) |
3471 | goto out; | |
7b128766 JB |
3472 | |
3473 | /* | |
3474 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 3475 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
3476 | * find the key and see if we have stuff that matches |
3477 | */ | |
3478 | if (ret > 0) { | |
66b4ffd1 | 3479 | ret = 0; |
7b128766 JB |
3480 | if (path->slots[0] == 0) |
3481 | break; | |
3482 | path->slots[0]--; | |
3483 | } | |
3484 | ||
3485 | /* pull out the item */ | |
3486 | leaf = path->nodes[0]; | |
7b128766 JB |
3487 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3488 | ||
3489 | /* make sure the item matches what we want */ | |
3490 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
3491 | break; | |
962a298f | 3492 | if (found_key.type != BTRFS_ORPHAN_ITEM_KEY) |
7b128766 JB |
3493 | break; |
3494 | ||
3495 | /* release the path since we're done with it */ | |
b3b4aa74 | 3496 | btrfs_release_path(path); |
7b128766 JB |
3497 | |
3498 | /* | |
3499 | * this is where we are basically btrfs_lookup, without the | |
3500 | * crossing root thing. we store the inode number in the | |
3501 | * offset of the orphan item. | |
3502 | */ | |
8f6d7f4f JB |
3503 | |
3504 | if (found_key.offset == last_objectid) { | |
0b246afa JM |
3505 | btrfs_err(fs_info, |
3506 | "Error removing orphan entry, stopping orphan cleanup"); | |
8f6d7f4f JB |
3507 | ret = -EINVAL; |
3508 | goto out; | |
3509 | } | |
3510 | ||
3511 | last_objectid = found_key.offset; | |
3512 | ||
5d4f98a2 YZ |
3513 | found_key.objectid = found_key.offset; |
3514 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
3515 | found_key.offset = 0; | |
0b246afa | 3516 | inode = btrfs_iget(fs_info->sb, &found_key, root, NULL); |
8c6ffba0 | 3517 | ret = PTR_ERR_OR_ZERO(inode); |
67710892 | 3518 | if (ret && ret != -ENOENT) |
66b4ffd1 | 3519 | goto out; |
7b128766 | 3520 | |
0b246afa | 3521 | if (ret == -ENOENT && root == fs_info->tree_root) { |
f8e9e0b0 AJ |
3522 | struct btrfs_root *dead_root; |
3523 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3524 | int is_dead_root = 0; | |
3525 | ||
3526 | /* | |
3527 | * this is an orphan in the tree root. Currently these | |
3528 | * could come from 2 sources: | |
3529 | * a) a snapshot deletion in progress | |
3530 | * b) a free space cache inode | |
3531 | * We need to distinguish those two, as the snapshot | |
3532 | * orphan must not get deleted. | |
3533 | * find_dead_roots already ran before us, so if this | |
3534 | * is a snapshot deletion, we should find the root | |
3535 | * in the dead_roots list | |
3536 | */ | |
3537 | spin_lock(&fs_info->trans_lock); | |
3538 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
3539 | root_list) { | |
3540 | if (dead_root->root_key.objectid == | |
3541 | found_key.objectid) { | |
3542 | is_dead_root = 1; | |
3543 | break; | |
3544 | } | |
3545 | } | |
3546 | spin_unlock(&fs_info->trans_lock); | |
3547 | if (is_dead_root) { | |
3548 | /* prevent this orphan from being found again */ | |
3549 | key.offset = found_key.objectid - 1; | |
3550 | continue; | |
3551 | } | |
3552 | } | |
7b128766 | 3553 | /* |
a8c9e576 JB |
3554 | * Inode is already gone but the orphan item is still there, |
3555 | * kill the orphan item. | |
7b128766 | 3556 | */ |
67710892 | 3557 | if (ret == -ENOENT) { |
a8c9e576 | 3558 | trans = btrfs_start_transaction(root, 1); |
66b4ffd1 JB |
3559 | if (IS_ERR(trans)) { |
3560 | ret = PTR_ERR(trans); | |
3561 | goto out; | |
3562 | } | |
0b246afa JM |
3563 | btrfs_debug(fs_info, "auto deleting %Lu", |
3564 | found_key.objectid); | |
a8c9e576 JB |
3565 | ret = btrfs_del_orphan_item(trans, root, |
3566 | found_key.objectid); | |
3a45bb20 | 3567 | btrfs_end_transaction(trans); |
4ef31a45 JB |
3568 | if (ret) |
3569 | goto out; | |
7b128766 JB |
3570 | continue; |
3571 | } | |
3572 | ||
a8c9e576 JB |
3573 | /* |
3574 | * add this inode to the orphan list so btrfs_orphan_del does | |
3575 | * the proper thing when we hit it | |
3576 | */ | |
8a35d95f JB |
3577 | set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3578 | &BTRFS_I(inode)->runtime_flags); | |
925396ec | 3579 | atomic_inc(&root->orphan_inodes); |
a8c9e576 | 3580 | |
7b128766 JB |
3581 | /* if we have links, this was a truncate, lets do that */ |
3582 | if (inode->i_nlink) { | |
fae7f21c | 3583 | if (WARN_ON(!S_ISREG(inode->i_mode))) { |
a41ad394 JB |
3584 | iput(inode); |
3585 | continue; | |
3586 | } | |
7b128766 | 3587 | nr_truncate++; |
f3fe820c JB |
3588 | |
3589 | /* 1 for the orphan item deletion. */ | |
3590 | trans = btrfs_start_transaction(root, 1); | |
3591 | if (IS_ERR(trans)) { | |
c69b26b0 | 3592 | iput(inode); |
f3fe820c JB |
3593 | ret = PTR_ERR(trans); |
3594 | goto out; | |
3595 | } | |
73f2e545 | 3596 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
3a45bb20 | 3597 | btrfs_end_transaction(trans); |
c69b26b0 JB |
3598 | if (ret) { |
3599 | iput(inode); | |
f3fe820c | 3600 | goto out; |
c69b26b0 | 3601 | } |
f3fe820c | 3602 | |
66b4ffd1 | 3603 | ret = btrfs_truncate(inode); |
4a7d0f68 | 3604 | if (ret) |
3d6ae7bb | 3605 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
7b128766 JB |
3606 | } else { |
3607 | nr_unlink++; | |
3608 | } | |
3609 | ||
3610 | /* this will do delete_inode and everything for us */ | |
3611 | iput(inode); | |
66b4ffd1 JB |
3612 | if (ret) |
3613 | goto out; | |
7b128766 | 3614 | } |
3254c876 MX |
3615 | /* release the path since we're done with it */ |
3616 | btrfs_release_path(path); | |
3617 | ||
d68fc57b YZ |
3618 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
3619 | ||
3620 | if (root->orphan_block_rsv) | |
2ff7e61e | 3621 | btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, |
d68fc57b YZ |
3622 | (u64)-1); |
3623 | ||
27cdeb70 MX |
3624 | if (root->orphan_block_rsv || |
3625 | test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) { | |
7a7eaa40 | 3626 | trans = btrfs_join_transaction(root); |
66b4ffd1 | 3627 | if (!IS_ERR(trans)) |
3a45bb20 | 3628 | btrfs_end_transaction(trans); |
d68fc57b | 3629 | } |
7b128766 JB |
3630 | |
3631 | if (nr_unlink) | |
0b246afa | 3632 | btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink); |
7b128766 | 3633 | if (nr_truncate) |
0b246afa | 3634 | btrfs_debug(fs_info, "truncated %d orphans", nr_truncate); |
66b4ffd1 JB |
3635 | |
3636 | out: | |
3637 | if (ret) | |
0b246afa | 3638 | btrfs_err(fs_info, "could not do orphan cleanup %d", ret); |
66b4ffd1 JB |
3639 | btrfs_free_path(path); |
3640 | return ret; | |
7b128766 JB |
3641 | } |
3642 | ||
46a53cca CM |
3643 | /* |
3644 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
3645 | * don't find any xattrs, we know there can't be any acls. | |
3646 | * | |
3647 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
3648 | */ | |
3649 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
63541927 FDBM |
3650 | int slot, u64 objectid, |
3651 | int *first_xattr_slot) | |
46a53cca CM |
3652 | { |
3653 | u32 nritems = btrfs_header_nritems(leaf); | |
3654 | struct btrfs_key found_key; | |
f23b5a59 JB |
3655 | static u64 xattr_access = 0; |
3656 | static u64 xattr_default = 0; | |
46a53cca CM |
3657 | int scanned = 0; |
3658 | ||
f23b5a59 | 3659 | if (!xattr_access) { |
97d79299 AG |
3660 | xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS, |
3661 | strlen(XATTR_NAME_POSIX_ACL_ACCESS)); | |
3662 | xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT, | |
3663 | strlen(XATTR_NAME_POSIX_ACL_DEFAULT)); | |
f23b5a59 JB |
3664 | } |
3665 | ||
46a53cca | 3666 | slot++; |
63541927 | 3667 | *first_xattr_slot = -1; |
46a53cca CM |
3668 | while (slot < nritems) { |
3669 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3670 | ||
3671 | /* we found a different objectid, there must not be acls */ | |
3672 | if (found_key.objectid != objectid) | |
3673 | return 0; | |
3674 | ||
3675 | /* we found an xattr, assume we've got an acl */ | |
f23b5a59 | 3676 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) { |
63541927 FDBM |
3677 | if (*first_xattr_slot == -1) |
3678 | *first_xattr_slot = slot; | |
f23b5a59 JB |
3679 | if (found_key.offset == xattr_access || |
3680 | found_key.offset == xattr_default) | |
3681 | return 1; | |
3682 | } | |
46a53cca CM |
3683 | |
3684 | /* | |
3685 | * we found a key greater than an xattr key, there can't | |
3686 | * be any acls later on | |
3687 | */ | |
3688 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
3689 | return 0; | |
3690 | ||
3691 | slot++; | |
3692 | scanned++; | |
3693 | ||
3694 | /* | |
3695 | * it goes inode, inode backrefs, xattrs, extents, | |
3696 | * so if there are a ton of hard links to an inode there can | |
3697 | * be a lot of backrefs. Don't waste time searching too hard, | |
3698 | * this is just an optimization | |
3699 | */ | |
3700 | if (scanned >= 8) | |
3701 | break; | |
3702 | } | |
3703 | /* we hit the end of the leaf before we found an xattr or | |
3704 | * something larger than an xattr. We have to assume the inode | |
3705 | * has acls | |
3706 | */ | |
63541927 FDBM |
3707 | if (*first_xattr_slot == -1) |
3708 | *first_xattr_slot = slot; | |
46a53cca CM |
3709 | return 1; |
3710 | } | |
3711 | ||
d352ac68 CM |
3712 | /* |
3713 | * read an inode from the btree into the in-memory inode | |
3714 | */ | |
67710892 | 3715 | static int btrfs_read_locked_inode(struct inode *inode) |
39279cc3 | 3716 | { |
0b246afa | 3717 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 3718 | struct btrfs_path *path; |
5f39d397 | 3719 | struct extent_buffer *leaf; |
39279cc3 CM |
3720 | struct btrfs_inode_item *inode_item; |
3721 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3722 | struct btrfs_key location; | |
67de1176 | 3723 | unsigned long ptr; |
46a53cca | 3724 | int maybe_acls; |
618e21d5 | 3725 | u32 rdev; |
39279cc3 | 3726 | int ret; |
2f7e33d4 | 3727 | bool filled = false; |
63541927 | 3728 | int first_xattr_slot; |
2f7e33d4 MX |
3729 | |
3730 | ret = btrfs_fill_inode(inode, &rdev); | |
3731 | if (!ret) | |
3732 | filled = true; | |
39279cc3 CM |
3733 | |
3734 | path = btrfs_alloc_path(); | |
67710892 FM |
3735 | if (!path) { |
3736 | ret = -ENOMEM; | |
1748f843 | 3737 | goto make_bad; |
67710892 | 3738 | } |
1748f843 | 3739 | |
39279cc3 | 3740 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 3741 | |
39279cc3 | 3742 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
67710892 FM |
3743 | if (ret) { |
3744 | if (ret > 0) | |
3745 | ret = -ENOENT; | |
39279cc3 | 3746 | goto make_bad; |
67710892 | 3747 | } |
39279cc3 | 3748 | |
5f39d397 | 3749 | leaf = path->nodes[0]; |
2f7e33d4 MX |
3750 | |
3751 | if (filled) | |
67de1176 | 3752 | goto cache_index; |
2f7e33d4 | 3753 | |
5f39d397 CM |
3754 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
3755 | struct btrfs_inode_item); | |
5f39d397 | 3756 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 3757 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
3758 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
3759 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
6ef06d27 | 3760 | btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item)); |
5f39d397 | 3761 | |
a937b979 DS |
3762 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime); |
3763 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime); | |
5f39d397 | 3764 | |
a937b979 DS |
3765 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime); |
3766 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime); | |
5f39d397 | 3767 | |
a937b979 DS |
3768 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime); |
3769 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime); | |
5f39d397 | 3770 | |
9cc97d64 | 3771 | BTRFS_I(inode)->i_otime.tv_sec = |
3772 | btrfs_timespec_sec(leaf, &inode_item->otime); | |
3773 | BTRFS_I(inode)->i_otime.tv_nsec = | |
3774 | btrfs_timespec_nsec(leaf, &inode_item->otime); | |
5f39d397 | 3775 | |
a76a3cd4 | 3776 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 3777 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
3778 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
3779 | ||
6e17d30b YD |
3780 | inode->i_version = btrfs_inode_sequence(leaf, inode_item); |
3781 | inode->i_generation = BTRFS_I(inode)->generation; | |
3782 | inode->i_rdev = 0; | |
3783 | rdev = btrfs_inode_rdev(leaf, inode_item); | |
3784 | ||
3785 | BTRFS_I(inode)->index_cnt = (u64)-1; | |
3786 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); | |
3787 | ||
3788 | cache_index: | |
5dc562c5 JB |
3789 | /* |
3790 | * If we were modified in the current generation and evicted from memory | |
3791 | * and then re-read we need to do a full sync since we don't have any | |
3792 | * idea about which extents were modified before we were evicted from | |
3793 | * cache. | |
6e17d30b YD |
3794 | * |
3795 | * This is required for both inode re-read from disk and delayed inode | |
3796 | * in delayed_nodes_tree. | |
5dc562c5 | 3797 | */ |
0b246afa | 3798 | if (BTRFS_I(inode)->last_trans == fs_info->generation) |
5dc562c5 JB |
3799 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
3800 | &BTRFS_I(inode)->runtime_flags); | |
3801 | ||
bde6c242 FM |
3802 | /* |
3803 | * We don't persist the id of the transaction where an unlink operation | |
3804 | * against the inode was last made. So here we assume the inode might | |
3805 | * have been evicted, and therefore the exact value of last_unlink_trans | |
3806 | * lost, and set it to last_trans to avoid metadata inconsistencies | |
3807 | * between the inode and its parent if the inode is fsync'ed and the log | |
3808 | * replayed. For example, in the scenario: | |
3809 | * | |
3810 | * touch mydir/foo | |
3811 | * ln mydir/foo mydir/bar | |
3812 | * sync | |
3813 | * unlink mydir/bar | |
3814 | * echo 2 > /proc/sys/vm/drop_caches # evicts inode | |
3815 | * xfs_io -c fsync mydir/foo | |
3816 | * <power failure> | |
3817 | * mount fs, triggers fsync log replay | |
3818 | * | |
3819 | * We must make sure that when we fsync our inode foo we also log its | |
3820 | * parent inode, otherwise after log replay the parent still has the | |
3821 | * dentry with the "bar" name but our inode foo has a link count of 1 | |
3822 | * and doesn't have an inode ref with the name "bar" anymore. | |
3823 | * | |
3824 | * Setting last_unlink_trans to last_trans is a pessimistic approach, | |
01327610 | 3825 | * but it guarantees correctness at the expense of occasional full |
bde6c242 FM |
3826 | * transaction commits on fsync if our inode is a directory, or if our |
3827 | * inode is not a directory, logging its parent unnecessarily. | |
3828 | */ | |
3829 | BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans; | |
3830 | ||
67de1176 MX |
3831 | path->slots[0]++; |
3832 | if (inode->i_nlink != 1 || | |
3833 | path->slots[0] >= btrfs_header_nritems(leaf)) | |
3834 | goto cache_acl; | |
3835 | ||
3836 | btrfs_item_key_to_cpu(leaf, &location, path->slots[0]); | |
4a0cc7ca | 3837 | if (location.objectid != btrfs_ino(BTRFS_I(inode))) |
67de1176 MX |
3838 | goto cache_acl; |
3839 | ||
3840 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3841 | if (location.type == BTRFS_INODE_REF_KEY) { | |
3842 | struct btrfs_inode_ref *ref; | |
3843 | ||
3844 | ref = (struct btrfs_inode_ref *)ptr; | |
3845 | BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref); | |
3846 | } else if (location.type == BTRFS_INODE_EXTREF_KEY) { | |
3847 | struct btrfs_inode_extref *extref; | |
3848 | ||
3849 | extref = (struct btrfs_inode_extref *)ptr; | |
3850 | BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf, | |
3851 | extref); | |
3852 | } | |
2f7e33d4 | 3853 | cache_acl: |
46a53cca CM |
3854 | /* |
3855 | * try to precache a NULL acl entry for files that don't have | |
3856 | * any xattrs or acls | |
3857 | */ | |
33345d01 | 3858 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
f85b7379 | 3859 | btrfs_ino(BTRFS_I(inode)), &first_xattr_slot); |
63541927 FDBM |
3860 | if (first_xattr_slot != -1) { |
3861 | path->slots[0] = first_xattr_slot; | |
3862 | ret = btrfs_load_inode_props(inode, path); | |
3863 | if (ret) | |
0b246afa | 3864 | btrfs_err(fs_info, |
351fd353 | 3865 | "error loading props for ino %llu (root %llu): %d", |
4a0cc7ca | 3866 | btrfs_ino(BTRFS_I(inode)), |
63541927 FDBM |
3867 | root->root_key.objectid, ret); |
3868 | } | |
3869 | btrfs_free_path(path); | |
3870 | ||
72c04902 AV |
3871 | if (!maybe_acls) |
3872 | cache_no_acl(inode); | |
46a53cca | 3873 | |
39279cc3 | 3874 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
3875 | case S_IFREG: |
3876 | inode->i_mapping->a_ops = &btrfs_aops; | |
d1310b2e | 3877 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
3878 | inode->i_fop = &btrfs_file_operations; |
3879 | inode->i_op = &btrfs_file_inode_operations; | |
3880 | break; | |
3881 | case S_IFDIR: | |
3882 | inode->i_fop = &btrfs_dir_file_operations; | |
67ade058 | 3883 | inode->i_op = &btrfs_dir_inode_operations; |
39279cc3 CM |
3884 | break; |
3885 | case S_IFLNK: | |
3886 | inode->i_op = &btrfs_symlink_inode_operations; | |
21fc61c7 | 3887 | inode_nohighmem(inode); |
39279cc3 CM |
3888 | inode->i_mapping->a_ops = &btrfs_symlink_aops; |
3889 | break; | |
618e21d5 | 3890 | default: |
0279b4cd | 3891 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
3892 | init_special_inode(inode, inode->i_mode, rdev); |
3893 | break; | |
39279cc3 | 3894 | } |
6cbff00f CH |
3895 | |
3896 | btrfs_update_iflags(inode); | |
67710892 | 3897 | return 0; |
39279cc3 CM |
3898 | |
3899 | make_bad: | |
39279cc3 | 3900 | btrfs_free_path(path); |
39279cc3 | 3901 | make_bad_inode(inode); |
67710892 | 3902 | return ret; |
39279cc3 CM |
3903 | } |
3904 | ||
d352ac68 CM |
3905 | /* |
3906 | * given a leaf and an inode, copy the inode fields into the leaf | |
3907 | */ | |
e02119d5 CM |
3908 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3909 | struct extent_buffer *leaf, | |
5f39d397 | 3910 | struct btrfs_inode_item *item, |
39279cc3 CM |
3911 | struct inode *inode) |
3912 | { | |
51fab693 LB |
3913 | struct btrfs_map_token token; |
3914 | ||
3915 | btrfs_init_map_token(&token); | |
5f39d397 | 3916 | |
51fab693 LB |
3917 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); |
3918 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3919 | btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size, | |
3920 | &token); | |
3921 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3922 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
5f39d397 | 3923 | |
a937b979 | 3924 | btrfs_set_token_timespec_sec(leaf, &item->atime, |
51fab693 | 3925 | inode->i_atime.tv_sec, &token); |
a937b979 | 3926 | btrfs_set_token_timespec_nsec(leaf, &item->atime, |
51fab693 | 3927 | inode->i_atime.tv_nsec, &token); |
5f39d397 | 3928 | |
a937b979 | 3929 | btrfs_set_token_timespec_sec(leaf, &item->mtime, |
51fab693 | 3930 | inode->i_mtime.tv_sec, &token); |
a937b979 | 3931 | btrfs_set_token_timespec_nsec(leaf, &item->mtime, |
51fab693 | 3932 | inode->i_mtime.tv_nsec, &token); |
5f39d397 | 3933 | |
a937b979 | 3934 | btrfs_set_token_timespec_sec(leaf, &item->ctime, |
51fab693 | 3935 | inode->i_ctime.tv_sec, &token); |
a937b979 | 3936 | btrfs_set_token_timespec_nsec(leaf, &item->ctime, |
51fab693 | 3937 | inode->i_ctime.tv_nsec, &token); |
5f39d397 | 3938 | |
9cc97d64 | 3939 | btrfs_set_token_timespec_sec(leaf, &item->otime, |
3940 | BTRFS_I(inode)->i_otime.tv_sec, &token); | |
3941 | btrfs_set_token_timespec_nsec(leaf, &item->otime, | |
3942 | BTRFS_I(inode)->i_otime.tv_nsec, &token); | |
3943 | ||
51fab693 LB |
3944 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), |
3945 | &token); | |
3946 | btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation, | |
3947 | &token); | |
3948 | btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token); | |
3949 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); | |
3950 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3951 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3952 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
39279cc3 CM |
3953 | } |
3954 | ||
d352ac68 CM |
3955 | /* |
3956 | * copy everything in the in-memory inode into the btree. | |
3957 | */ | |
2115133f | 3958 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 3959 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
3960 | { |
3961 | struct btrfs_inode_item *inode_item; | |
3962 | struct btrfs_path *path; | |
5f39d397 | 3963 | struct extent_buffer *leaf; |
39279cc3 CM |
3964 | int ret; |
3965 | ||
3966 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
3967 | if (!path) |
3968 | return -ENOMEM; | |
3969 | ||
b9473439 | 3970 | path->leave_spinning = 1; |
16cdcec7 MX |
3971 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
3972 | 1); | |
39279cc3 CM |
3973 | if (ret) { |
3974 | if (ret > 0) | |
3975 | ret = -ENOENT; | |
3976 | goto failed; | |
3977 | } | |
3978 | ||
5f39d397 CM |
3979 | leaf = path->nodes[0]; |
3980 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 3981 | struct btrfs_inode_item); |
39279cc3 | 3982 | |
e02119d5 | 3983 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 3984 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 3985 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
3986 | ret = 0; |
3987 | failed: | |
39279cc3 CM |
3988 | btrfs_free_path(path); |
3989 | return ret; | |
3990 | } | |
3991 | ||
2115133f CM |
3992 | /* |
3993 | * copy everything in the in-memory inode into the btree. | |
3994 | */ | |
3995 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
3996 | struct btrfs_root *root, struct inode *inode) | |
3997 | { | |
0b246afa | 3998 | struct btrfs_fs_info *fs_info = root->fs_info; |
2115133f CM |
3999 | int ret; |
4000 | ||
4001 | /* | |
4002 | * If the inode is a free space inode, we can deadlock during commit | |
4003 | * if we put it into the delayed code. | |
4004 | * | |
4005 | * The data relocation inode should also be directly updated | |
4006 | * without delay | |
4007 | */ | |
70ddc553 | 4008 | if (!btrfs_is_free_space_inode(BTRFS_I(inode)) |
1d52c78a | 4009 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
0b246afa | 4010 | && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) { |
8ea05e3a AB |
4011 | btrfs_update_root_times(trans, root); |
4012 | ||
2115133f CM |
4013 | ret = btrfs_delayed_update_inode(trans, root, inode); |
4014 | if (!ret) | |
4015 | btrfs_set_inode_last_trans(trans, inode); | |
4016 | return ret; | |
4017 | } | |
4018 | ||
4019 | return btrfs_update_inode_item(trans, root, inode); | |
4020 | } | |
4021 | ||
be6aef60 JB |
4022 | noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
4023 | struct btrfs_root *root, | |
4024 | struct inode *inode) | |
2115133f CM |
4025 | { |
4026 | int ret; | |
4027 | ||
4028 | ret = btrfs_update_inode(trans, root, inode); | |
4029 | if (ret == -ENOSPC) | |
4030 | return btrfs_update_inode_item(trans, root, inode); | |
4031 | return ret; | |
4032 | } | |
4033 | ||
d352ac68 CM |
4034 | /* |
4035 | * unlink helper that gets used here in inode.c and in the tree logging | |
4036 | * recovery code. It remove a link in a directory with a given name, and | |
4037 | * also drops the back refs in the inode to the directory | |
4038 | */ | |
92986796 AV |
4039 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
4040 | struct btrfs_root *root, | |
4ec5934e NB |
4041 | struct btrfs_inode *dir, |
4042 | struct btrfs_inode *inode, | |
92986796 | 4043 | const char *name, int name_len) |
39279cc3 | 4044 | { |
0b246afa | 4045 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 4046 | struct btrfs_path *path; |
39279cc3 | 4047 | int ret = 0; |
5f39d397 | 4048 | struct extent_buffer *leaf; |
39279cc3 | 4049 | struct btrfs_dir_item *di; |
5f39d397 | 4050 | struct btrfs_key key; |
aec7477b | 4051 | u64 index; |
33345d01 LZ |
4052 | u64 ino = btrfs_ino(inode); |
4053 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
4054 | |
4055 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
4056 | if (!path) { |
4057 | ret = -ENOMEM; | |
554233a6 | 4058 | goto out; |
54aa1f4d CM |
4059 | } |
4060 | ||
b9473439 | 4061 | path->leave_spinning = 1; |
33345d01 | 4062 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 CM |
4063 | name, name_len, -1); |
4064 | if (IS_ERR(di)) { | |
4065 | ret = PTR_ERR(di); | |
4066 | goto err; | |
4067 | } | |
4068 | if (!di) { | |
4069 | ret = -ENOENT; | |
4070 | goto err; | |
4071 | } | |
5f39d397 CM |
4072 | leaf = path->nodes[0]; |
4073 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 4074 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
4075 | if (ret) |
4076 | goto err; | |
b3b4aa74 | 4077 | btrfs_release_path(path); |
39279cc3 | 4078 | |
67de1176 MX |
4079 | /* |
4080 | * If we don't have dir index, we have to get it by looking up | |
4081 | * the inode ref, since we get the inode ref, remove it directly, | |
4082 | * it is unnecessary to do delayed deletion. | |
4083 | * | |
4084 | * But if we have dir index, needn't search inode ref to get it. | |
4085 | * Since the inode ref is close to the inode item, it is better | |
4086 | * that we delay to delete it, and just do this deletion when | |
4087 | * we update the inode item. | |
4088 | */ | |
4ec5934e | 4089 | if (inode->dir_index) { |
67de1176 MX |
4090 | ret = btrfs_delayed_delete_inode_ref(inode); |
4091 | if (!ret) { | |
4ec5934e | 4092 | index = inode->dir_index; |
67de1176 MX |
4093 | goto skip_backref; |
4094 | } | |
4095 | } | |
4096 | ||
33345d01 LZ |
4097 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
4098 | dir_ino, &index); | |
aec7477b | 4099 | if (ret) { |
0b246afa | 4100 | btrfs_info(fs_info, |
c2cf52eb | 4101 | "failed to delete reference to %.*s, inode %llu parent %llu", |
c1c9ff7c | 4102 | name_len, name, ino, dir_ino); |
66642832 | 4103 | btrfs_abort_transaction(trans, ret); |
aec7477b JB |
4104 | goto err; |
4105 | } | |
67de1176 | 4106 | skip_backref: |
2ff7e61e | 4107 | ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index); |
79787eaa | 4108 | if (ret) { |
66642832 | 4109 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 4110 | goto err; |
79787eaa | 4111 | } |
39279cc3 | 4112 | |
4ec5934e NB |
4113 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode, |
4114 | dir_ino); | |
79787eaa | 4115 | if (ret != 0 && ret != -ENOENT) { |
66642832 | 4116 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4117 | goto err; |
4118 | } | |
e02119d5 | 4119 | |
4ec5934e NB |
4120 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir, |
4121 | index); | |
6418c961 CM |
4122 | if (ret == -ENOENT) |
4123 | ret = 0; | |
d4e3991b | 4124 | else if (ret) |
66642832 | 4125 | btrfs_abort_transaction(trans, ret); |
39279cc3 CM |
4126 | err: |
4127 | btrfs_free_path(path); | |
e02119d5 CM |
4128 | if (ret) |
4129 | goto out; | |
4130 | ||
6ef06d27 | 4131 | btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2); |
4ec5934e NB |
4132 | inode_inc_iversion(&inode->vfs_inode); |
4133 | inode_inc_iversion(&dir->vfs_inode); | |
4134 | inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime = | |
4135 | dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode); | |
4136 | ret = btrfs_update_inode(trans, root, &dir->vfs_inode); | |
e02119d5 | 4137 | out: |
39279cc3 CM |
4138 | return ret; |
4139 | } | |
4140 | ||
92986796 AV |
4141 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
4142 | struct btrfs_root *root, | |
4ec5934e | 4143 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
92986796 AV |
4144 | const char *name, int name_len) |
4145 | { | |
4146 | int ret; | |
4147 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
4148 | if (!ret) { | |
4ec5934e NB |
4149 | drop_nlink(&inode->vfs_inode); |
4150 | ret = btrfs_update_inode(trans, root, &inode->vfs_inode); | |
92986796 AV |
4151 | } |
4152 | return ret; | |
4153 | } | |
39279cc3 | 4154 | |
a22285a6 YZ |
4155 | /* |
4156 | * helper to start transaction for unlink and rmdir. | |
4157 | * | |
d52be818 JB |
4158 | * unlink and rmdir are special in btrfs, they do not always free space, so |
4159 | * if we cannot make our reservations the normal way try and see if there is | |
4160 | * plenty of slack room in the global reserve to migrate, otherwise we cannot | |
4161 | * allow the unlink to occur. | |
a22285a6 | 4162 | */ |
d52be818 | 4163 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir) |
4df27c4d | 4164 | { |
a22285a6 | 4165 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 4166 | |
e70bea5f JB |
4167 | /* |
4168 | * 1 for the possible orphan item | |
4169 | * 1 for the dir item | |
4170 | * 1 for the dir index | |
4171 | * 1 for the inode ref | |
e70bea5f JB |
4172 | * 1 for the inode |
4173 | */ | |
8eab77ff | 4174 | return btrfs_start_transaction_fallback_global_rsv(root, 5, 5); |
a22285a6 YZ |
4175 | } |
4176 | ||
4177 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
4178 | { | |
4179 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4180 | struct btrfs_trans_handle *trans; | |
2b0143b5 | 4181 | struct inode *inode = d_inode(dentry); |
a22285a6 | 4182 | int ret; |
a22285a6 | 4183 | |
d52be818 | 4184 | trans = __unlink_start_trans(dir); |
a22285a6 YZ |
4185 | if (IS_ERR(trans)) |
4186 | return PTR_ERR(trans); | |
5f39d397 | 4187 | |
4ec5934e NB |
4188 | btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)), |
4189 | 0); | |
12fcfd22 | 4190 | |
4ec5934e NB |
4191 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
4192 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, | |
4193 | dentry->d_name.len); | |
b532402e TI |
4194 | if (ret) |
4195 | goto out; | |
7b128766 | 4196 | |
a22285a6 | 4197 | if (inode->i_nlink == 0) { |
73f2e545 | 4198 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
b532402e TI |
4199 | if (ret) |
4200 | goto out; | |
a22285a6 | 4201 | } |
7b128766 | 4202 | |
b532402e | 4203 | out: |
3a45bb20 | 4204 | btrfs_end_transaction(trans); |
2ff7e61e | 4205 | btrfs_btree_balance_dirty(root->fs_info); |
39279cc3 CM |
4206 | return ret; |
4207 | } | |
4208 | ||
4df27c4d YZ |
4209 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
4210 | struct btrfs_root *root, | |
4211 | struct inode *dir, u64 objectid, | |
4212 | const char *name, int name_len) | |
4213 | { | |
0b246afa | 4214 | struct btrfs_fs_info *fs_info = root->fs_info; |
4df27c4d YZ |
4215 | struct btrfs_path *path; |
4216 | struct extent_buffer *leaf; | |
4217 | struct btrfs_dir_item *di; | |
4218 | struct btrfs_key key; | |
4219 | u64 index; | |
4220 | int ret; | |
4a0cc7ca | 4221 | u64 dir_ino = btrfs_ino(BTRFS_I(dir)); |
4df27c4d YZ |
4222 | |
4223 | path = btrfs_alloc_path(); | |
4224 | if (!path) | |
4225 | return -ENOMEM; | |
4226 | ||
33345d01 | 4227 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 4228 | name, name_len, -1); |
79787eaa JM |
4229 | if (IS_ERR_OR_NULL(di)) { |
4230 | if (!di) | |
4231 | ret = -ENOENT; | |
4232 | else | |
4233 | ret = PTR_ERR(di); | |
4234 | goto out; | |
4235 | } | |
4df27c4d YZ |
4236 | |
4237 | leaf = path->nodes[0]; | |
4238 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
4239 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
4240 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa | 4241 | if (ret) { |
66642832 | 4242 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4243 | goto out; |
4244 | } | |
b3b4aa74 | 4245 | btrfs_release_path(path); |
4df27c4d | 4246 | |
0b246afa JM |
4247 | ret = btrfs_del_root_ref(trans, fs_info, objectid, |
4248 | root->root_key.objectid, dir_ino, | |
4249 | &index, name, name_len); | |
4df27c4d | 4250 | if (ret < 0) { |
79787eaa | 4251 | if (ret != -ENOENT) { |
66642832 | 4252 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4253 | goto out; |
4254 | } | |
33345d01 | 4255 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 4256 | name, name_len); |
79787eaa JM |
4257 | if (IS_ERR_OR_NULL(di)) { |
4258 | if (!di) | |
4259 | ret = -ENOENT; | |
4260 | else | |
4261 | ret = PTR_ERR(di); | |
66642832 | 4262 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4263 | goto out; |
4264 | } | |
4df27c4d YZ |
4265 | |
4266 | leaf = path->nodes[0]; | |
4267 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 4268 | btrfs_release_path(path); |
4df27c4d YZ |
4269 | index = key.offset; |
4270 | } | |
945d8962 | 4271 | btrfs_release_path(path); |
4df27c4d | 4272 | |
e67bbbb9 | 4273 | ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index); |
79787eaa | 4274 | if (ret) { |
66642832 | 4275 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4276 | goto out; |
4277 | } | |
4df27c4d | 4278 | |
6ef06d27 | 4279 | btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2); |
0c4d2d95 | 4280 | inode_inc_iversion(dir); |
c2050a45 | 4281 | dir->i_mtime = dir->i_ctime = current_time(dir); |
5a24e84c | 4282 | ret = btrfs_update_inode_fallback(trans, root, dir); |
79787eaa | 4283 | if (ret) |
66642832 | 4284 | btrfs_abort_transaction(trans, ret); |
79787eaa | 4285 | out: |
71d7aed0 | 4286 | btrfs_free_path(path); |
79787eaa | 4287 | return ret; |
4df27c4d YZ |
4288 | } |
4289 | ||
39279cc3 CM |
4290 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
4291 | { | |
2b0143b5 | 4292 | struct inode *inode = d_inode(dentry); |
1832a6d5 | 4293 | int err = 0; |
39279cc3 | 4294 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 4295 | struct btrfs_trans_handle *trans; |
44f714da | 4296 | u64 last_unlink_trans; |
39279cc3 | 4297 | |
b3ae244e | 4298 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 4299 | return -ENOTEMPTY; |
4a0cc7ca | 4300 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) |
b3ae244e | 4301 | return -EPERM; |
134d4512 | 4302 | |
d52be818 | 4303 | trans = __unlink_start_trans(dir); |
a22285a6 | 4304 | if (IS_ERR(trans)) |
5df6a9f6 | 4305 | return PTR_ERR(trans); |
5df6a9f6 | 4306 | |
4a0cc7ca | 4307 | if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
4df27c4d YZ |
4308 | err = btrfs_unlink_subvol(trans, root, dir, |
4309 | BTRFS_I(inode)->location.objectid, | |
4310 | dentry->d_name.name, | |
4311 | dentry->d_name.len); | |
4312 | goto out; | |
4313 | } | |
4314 | ||
73f2e545 | 4315 | err = btrfs_orphan_add(trans, BTRFS_I(inode)); |
7b128766 | 4316 | if (err) |
4df27c4d | 4317 | goto out; |
7b128766 | 4318 | |
44f714da FM |
4319 | last_unlink_trans = BTRFS_I(inode)->last_unlink_trans; |
4320 | ||
39279cc3 | 4321 | /* now the directory is empty */ |
4ec5934e NB |
4322 | err = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
4323 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, | |
4324 | dentry->d_name.len); | |
44f714da | 4325 | if (!err) { |
6ef06d27 | 4326 | btrfs_i_size_write(BTRFS_I(inode), 0); |
44f714da FM |
4327 | /* |
4328 | * Propagate the last_unlink_trans value of the deleted dir to | |
4329 | * its parent directory. This is to prevent an unrecoverable | |
4330 | * log tree in the case we do something like this: | |
4331 | * 1) create dir foo | |
4332 | * 2) create snapshot under dir foo | |
4333 | * 3) delete the snapshot | |
4334 | * 4) rmdir foo | |
4335 | * 5) mkdir foo | |
4336 | * 6) fsync foo or some file inside foo | |
4337 | */ | |
4338 | if (last_unlink_trans >= trans->transid) | |
4339 | BTRFS_I(dir)->last_unlink_trans = last_unlink_trans; | |
4340 | } | |
4df27c4d | 4341 | out: |
3a45bb20 | 4342 | btrfs_end_transaction(trans); |
2ff7e61e | 4343 | btrfs_btree_balance_dirty(root->fs_info); |
3954401f | 4344 | |
39279cc3 CM |
4345 | return err; |
4346 | } | |
4347 | ||
28f75a0e CM |
4348 | static int truncate_space_check(struct btrfs_trans_handle *trans, |
4349 | struct btrfs_root *root, | |
4350 | u64 bytes_deleted) | |
4351 | { | |
0b246afa | 4352 | struct btrfs_fs_info *fs_info = root->fs_info; |
28f75a0e CM |
4353 | int ret; |
4354 | ||
dc95f7bf JB |
4355 | /* |
4356 | * This is only used to apply pressure to the enospc system, we don't | |
4357 | * intend to use this reservation at all. | |
4358 | */ | |
2ff7e61e | 4359 | bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted); |
0b246afa JM |
4360 | bytes_deleted *= fs_info->nodesize; |
4361 | ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv, | |
28f75a0e | 4362 | bytes_deleted, BTRFS_RESERVE_NO_FLUSH); |
dc95f7bf | 4363 | if (!ret) { |
0b246afa | 4364 | trace_btrfs_space_reservation(fs_info, "transaction", |
dc95f7bf JB |
4365 | trans->transid, |
4366 | bytes_deleted, 1); | |
28f75a0e | 4367 | trans->bytes_reserved += bytes_deleted; |
dc95f7bf | 4368 | } |
28f75a0e CM |
4369 | return ret; |
4370 | ||
4371 | } | |
4372 | ||
ddfae63c JB |
4373 | /* |
4374 | * Return this if we need to call truncate_block for the last bit of the | |
4375 | * truncate. | |
4376 | */ | |
4377 | #define NEED_TRUNCATE_BLOCK 1 | |
0305cd5f | 4378 | |
39279cc3 CM |
4379 | /* |
4380 | * this can truncate away extent items, csum items and directory items. | |
4381 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 4382 | * any higher than new_size |
39279cc3 CM |
4383 | * |
4384 | * csum items that cross the new i_size are truncated to the new size | |
4385 | * as well. | |
7b128766 JB |
4386 | * |
4387 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
4388 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 4389 | */ |
8082510e YZ |
4390 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
4391 | struct btrfs_root *root, | |
4392 | struct inode *inode, | |
4393 | u64 new_size, u32 min_type) | |
39279cc3 | 4394 | { |
0b246afa | 4395 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 4396 | struct btrfs_path *path; |
5f39d397 | 4397 | struct extent_buffer *leaf; |
39279cc3 | 4398 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
4399 | struct btrfs_key key; |
4400 | struct btrfs_key found_key; | |
39279cc3 | 4401 | u64 extent_start = 0; |
db94535d | 4402 | u64 extent_num_bytes = 0; |
5d4f98a2 | 4403 | u64 extent_offset = 0; |
39279cc3 | 4404 | u64 item_end = 0; |
c1aa4575 | 4405 | u64 last_size = new_size; |
8082510e | 4406 | u32 found_type = (u8)-1; |
39279cc3 CM |
4407 | int found_extent; |
4408 | int del_item; | |
85e21bac CM |
4409 | int pending_del_nr = 0; |
4410 | int pending_del_slot = 0; | |
179e29e4 | 4411 | int extent_type = -1; |
8082510e YZ |
4412 | int ret; |
4413 | int err = 0; | |
4a0cc7ca | 4414 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
28ed1345 | 4415 | u64 bytes_deleted = 0; |
897ca819 TM |
4416 | bool be_nice = false; |
4417 | bool should_throttle = false; | |
4418 | bool should_end = false; | |
8082510e YZ |
4419 | |
4420 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 4421 | |
28ed1345 CM |
4422 | /* |
4423 | * for non-free space inodes and ref cows, we want to back off from | |
4424 | * time to time | |
4425 | */ | |
70ddc553 | 4426 | if (!btrfs_is_free_space_inode(BTRFS_I(inode)) && |
28ed1345 | 4427 | test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
897ca819 | 4428 | be_nice = true; |
28ed1345 | 4429 | |
0eb0e19c MF |
4430 | path = btrfs_alloc_path(); |
4431 | if (!path) | |
4432 | return -ENOMEM; | |
e4058b54 | 4433 | path->reada = READA_BACK; |
0eb0e19c | 4434 | |
5dc562c5 JB |
4435 | /* |
4436 | * We want to drop from the next block forward in case this new size is | |
4437 | * not block aligned since we will be keeping the last block of the | |
4438 | * extent just the way it is. | |
4439 | */ | |
27cdeb70 | 4440 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
0b246afa | 4441 | root == fs_info->tree_root) |
dcdbc059 | 4442 | btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size, |
0b246afa | 4443 | fs_info->sectorsize), |
da17066c | 4444 | (u64)-1, 0); |
8082510e | 4445 | |
16cdcec7 MX |
4446 | /* |
4447 | * This function is also used to drop the items in the log tree before | |
4448 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
4449 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
4450 | * items. | |
4451 | */ | |
4452 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
4ccb5c72 | 4453 | btrfs_kill_delayed_inode_items(BTRFS_I(inode)); |
16cdcec7 | 4454 | |
33345d01 | 4455 | key.objectid = ino; |
39279cc3 | 4456 | key.offset = (u64)-1; |
5f39d397 CM |
4457 | key.type = (u8)-1; |
4458 | ||
85e21bac | 4459 | search_again: |
28ed1345 CM |
4460 | /* |
4461 | * with a 16K leaf size and 128MB extents, you can actually queue | |
4462 | * up a huge file in a single leaf. Most of the time that | |
4463 | * bytes_deleted is > 0, it will be huge by the time we get here | |
4464 | */ | |
ee22184b | 4465 | if (be_nice && bytes_deleted > SZ_32M) { |
3a45bb20 | 4466 | if (btrfs_should_end_transaction(trans)) { |
28ed1345 CM |
4467 | err = -EAGAIN; |
4468 | goto error; | |
4469 | } | |
4470 | } | |
4471 | ||
4472 | ||
b9473439 | 4473 | path->leave_spinning = 1; |
85e21bac | 4474 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
4475 | if (ret < 0) { |
4476 | err = ret; | |
4477 | goto out; | |
4478 | } | |
d397712b | 4479 | |
85e21bac | 4480 | if (ret > 0) { |
e02119d5 CM |
4481 | /* there are no items in the tree for us to truncate, we're |
4482 | * done | |
4483 | */ | |
8082510e YZ |
4484 | if (path->slots[0] == 0) |
4485 | goto out; | |
85e21bac CM |
4486 | path->slots[0]--; |
4487 | } | |
4488 | ||
d397712b | 4489 | while (1) { |
39279cc3 | 4490 | fi = NULL; |
5f39d397 CM |
4491 | leaf = path->nodes[0]; |
4492 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
962a298f | 4493 | found_type = found_key.type; |
39279cc3 | 4494 | |
33345d01 | 4495 | if (found_key.objectid != ino) |
39279cc3 | 4496 | break; |
5f39d397 | 4497 | |
85e21bac | 4498 | if (found_type < min_type) |
39279cc3 CM |
4499 | break; |
4500 | ||
5f39d397 | 4501 | item_end = found_key.offset; |
39279cc3 | 4502 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 4503 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 4504 | struct btrfs_file_extent_item); |
179e29e4 CM |
4505 | extent_type = btrfs_file_extent_type(leaf, fi); |
4506 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4507 | item_end += |
db94535d | 4508 | btrfs_file_extent_num_bytes(leaf, fi); |
09ed2f16 LB |
4509 | |
4510 | trace_btrfs_truncate_show_fi_regular( | |
4511 | BTRFS_I(inode), leaf, fi, | |
4512 | found_key.offset); | |
179e29e4 | 4513 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 4514 | item_end += btrfs_file_extent_inline_len(leaf, |
514ac8ad | 4515 | path->slots[0], fi); |
09ed2f16 LB |
4516 | |
4517 | trace_btrfs_truncate_show_fi_inline( | |
4518 | BTRFS_I(inode), leaf, fi, path->slots[0], | |
4519 | found_key.offset); | |
39279cc3 | 4520 | } |
008630c1 | 4521 | item_end--; |
39279cc3 | 4522 | } |
8082510e YZ |
4523 | if (found_type > min_type) { |
4524 | del_item = 1; | |
4525 | } else { | |
76b42abb | 4526 | if (item_end < new_size) |
b888db2b | 4527 | break; |
8082510e YZ |
4528 | if (found_key.offset >= new_size) |
4529 | del_item = 1; | |
4530 | else | |
4531 | del_item = 0; | |
39279cc3 | 4532 | } |
39279cc3 | 4533 | found_extent = 0; |
39279cc3 | 4534 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
4535 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
4536 | goto delete; | |
4537 | ||
4538 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 4539 | u64 num_dec; |
db94535d | 4540 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 4541 | if (!del_item) { |
db94535d CM |
4542 | u64 orig_num_bytes = |
4543 | btrfs_file_extent_num_bytes(leaf, fi); | |
fda2832f QW |
4544 | extent_num_bytes = ALIGN(new_size - |
4545 | found_key.offset, | |
0b246afa | 4546 | fs_info->sectorsize); |
db94535d CM |
4547 | btrfs_set_file_extent_num_bytes(leaf, fi, |
4548 | extent_num_bytes); | |
4549 | num_dec = (orig_num_bytes - | |
9069218d | 4550 | extent_num_bytes); |
27cdeb70 MX |
4551 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4552 | &root->state) && | |
4553 | extent_start != 0) | |
a76a3cd4 | 4554 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 4555 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 4556 | } else { |
db94535d CM |
4557 | extent_num_bytes = |
4558 | btrfs_file_extent_disk_num_bytes(leaf, | |
4559 | fi); | |
5d4f98a2 YZ |
4560 | extent_offset = found_key.offset - |
4561 | btrfs_file_extent_offset(leaf, fi); | |
4562 | ||
39279cc3 | 4563 | /* FIXME blocksize != 4096 */ |
9069218d | 4564 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
4565 | if (extent_start != 0) { |
4566 | found_extent = 1; | |
27cdeb70 MX |
4567 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4568 | &root->state)) | |
a76a3cd4 | 4569 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 4570 | } |
39279cc3 | 4571 | } |
9069218d | 4572 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
4573 | /* |
4574 | * we can't truncate inline items that have had | |
4575 | * special encodings | |
4576 | */ | |
4577 | if (!del_item && | |
c8b97818 | 4578 | btrfs_file_extent_encryption(leaf, fi) == 0 && |
ddfae63c JB |
4579 | btrfs_file_extent_other_encoding(leaf, fi) == 0 && |
4580 | btrfs_file_extent_compression(leaf, fi) == 0) { | |
4581 | u32 size = (u32)(new_size - found_key.offset); | |
4582 | ||
4583 | btrfs_set_file_extent_ram_bytes(leaf, fi, size); | |
4584 | size = btrfs_file_extent_calc_inline_size(size); | |
4585 | btrfs_truncate_item(root->fs_info, path, size, 1); | |
4586 | } else if (!del_item) { | |
514ac8ad | 4587 | /* |
ddfae63c JB |
4588 | * We have to bail so the last_size is set to |
4589 | * just before this extent. | |
514ac8ad | 4590 | */ |
ddfae63c JB |
4591 | err = NEED_TRUNCATE_BLOCK; |
4592 | break; | |
4593 | } | |
0305cd5f | 4594 | |
ddfae63c | 4595 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
0305cd5f | 4596 | inode_sub_bytes(inode, item_end + 1 - new_size); |
39279cc3 | 4597 | } |
179e29e4 | 4598 | delete: |
ddfae63c JB |
4599 | if (del_item) |
4600 | last_size = found_key.offset; | |
4601 | else | |
4602 | last_size = new_size; | |
39279cc3 | 4603 | if (del_item) { |
85e21bac CM |
4604 | if (!pending_del_nr) { |
4605 | /* no pending yet, add ourselves */ | |
4606 | pending_del_slot = path->slots[0]; | |
4607 | pending_del_nr = 1; | |
4608 | } else if (pending_del_nr && | |
4609 | path->slots[0] + 1 == pending_del_slot) { | |
4610 | /* hop on the pending chunk */ | |
4611 | pending_del_nr++; | |
4612 | pending_del_slot = path->slots[0]; | |
4613 | } else { | |
d397712b | 4614 | BUG(); |
85e21bac | 4615 | } |
39279cc3 CM |
4616 | } else { |
4617 | break; | |
4618 | } | |
897ca819 | 4619 | should_throttle = false; |
28f75a0e | 4620 | |
27cdeb70 MX |
4621 | if (found_extent && |
4622 | (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || | |
0b246afa | 4623 | root == fs_info->tree_root)) { |
b9473439 | 4624 | btrfs_set_path_blocking(path); |
28ed1345 | 4625 | bytes_deleted += extent_num_bytes; |
84f7d8e6 | 4626 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
4627 | extent_num_bytes, 0, |
4628 | btrfs_header_owner(leaf), | |
b06c4bf5 | 4629 | ino, extent_offset); |
39279cc3 | 4630 | BUG_ON(ret); |
2ff7e61e JM |
4631 | if (btrfs_should_throttle_delayed_refs(trans, fs_info)) |
4632 | btrfs_async_run_delayed_refs(fs_info, | |
dd4b857a WX |
4633 | trans->delayed_ref_updates * 2, |
4634 | trans->transid, 0); | |
28f75a0e CM |
4635 | if (be_nice) { |
4636 | if (truncate_space_check(trans, root, | |
4637 | extent_num_bytes)) { | |
897ca819 | 4638 | should_end = true; |
28f75a0e CM |
4639 | } |
4640 | if (btrfs_should_throttle_delayed_refs(trans, | |
2ff7e61e | 4641 | fs_info)) |
897ca819 | 4642 | should_throttle = true; |
28f75a0e | 4643 | } |
39279cc3 | 4644 | } |
85e21bac | 4645 | |
8082510e YZ |
4646 | if (found_type == BTRFS_INODE_ITEM_KEY) |
4647 | break; | |
4648 | ||
4649 | if (path->slots[0] == 0 || | |
1262133b | 4650 | path->slots[0] != pending_del_slot || |
28f75a0e | 4651 | should_throttle || should_end) { |
8082510e YZ |
4652 | if (pending_del_nr) { |
4653 | ret = btrfs_del_items(trans, root, path, | |
4654 | pending_del_slot, | |
4655 | pending_del_nr); | |
79787eaa | 4656 | if (ret) { |
66642832 | 4657 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4658 | goto error; |
4659 | } | |
8082510e YZ |
4660 | pending_del_nr = 0; |
4661 | } | |
b3b4aa74 | 4662 | btrfs_release_path(path); |
28f75a0e | 4663 | if (should_throttle) { |
1262133b JB |
4664 | unsigned long updates = trans->delayed_ref_updates; |
4665 | if (updates) { | |
4666 | trans->delayed_ref_updates = 0; | |
2ff7e61e JM |
4667 | ret = btrfs_run_delayed_refs(trans, |
4668 | fs_info, | |
4669 | updates * 2); | |
1262133b JB |
4670 | if (ret && !err) |
4671 | err = ret; | |
4672 | } | |
4673 | } | |
28f75a0e CM |
4674 | /* |
4675 | * if we failed to refill our space rsv, bail out | |
4676 | * and let the transaction restart | |
4677 | */ | |
4678 | if (should_end) { | |
4679 | err = -EAGAIN; | |
4680 | goto error; | |
4681 | } | |
85e21bac | 4682 | goto search_again; |
8082510e YZ |
4683 | } else { |
4684 | path->slots[0]--; | |
85e21bac | 4685 | } |
39279cc3 | 4686 | } |
8082510e | 4687 | out: |
85e21bac CM |
4688 | if (pending_del_nr) { |
4689 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
4690 | pending_del_nr); | |
79787eaa | 4691 | if (ret) |
66642832 | 4692 | btrfs_abort_transaction(trans, ret); |
85e21bac | 4693 | } |
79787eaa | 4694 | error: |
76b42abb FM |
4695 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
4696 | ASSERT(last_size >= new_size); | |
4697 | if (!err && last_size > new_size) | |
4698 | last_size = new_size; | |
7f4f6e0a | 4699 | btrfs_ordered_update_i_size(inode, last_size, NULL); |
76b42abb | 4700 | } |
28ed1345 | 4701 | |
39279cc3 | 4702 | btrfs_free_path(path); |
28ed1345 | 4703 | |
ee22184b | 4704 | if (be_nice && bytes_deleted > SZ_32M) { |
28ed1345 CM |
4705 | unsigned long updates = trans->delayed_ref_updates; |
4706 | if (updates) { | |
4707 | trans->delayed_ref_updates = 0; | |
2ff7e61e JM |
4708 | ret = btrfs_run_delayed_refs(trans, fs_info, |
4709 | updates * 2); | |
28ed1345 CM |
4710 | if (ret && !err) |
4711 | err = ret; | |
4712 | } | |
4713 | } | |
8082510e | 4714 | return err; |
39279cc3 CM |
4715 | } |
4716 | ||
4717 | /* | |
9703fefe | 4718 | * btrfs_truncate_block - read, zero a chunk and write a block |
2aaa6655 JB |
4719 | * @inode - inode that we're zeroing |
4720 | * @from - the offset to start zeroing | |
4721 | * @len - the length to zero, 0 to zero the entire range respective to the | |
4722 | * offset | |
4723 | * @front - zero up to the offset instead of from the offset on | |
4724 | * | |
9703fefe | 4725 | * This will find the block for the "from" offset and cow the block and zero the |
2aaa6655 | 4726 | * part we want to zero. This is used with truncate and hole punching. |
39279cc3 | 4727 | */ |
9703fefe | 4728 | int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len, |
2aaa6655 | 4729 | int front) |
39279cc3 | 4730 | { |
0b246afa | 4731 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2aaa6655 | 4732 | struct address_space *mapping = inode->i_mapping; |
e6dcd2dc CM |
4733 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4734 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 4735 | struct extent_state *cached_state = NULL; |
364ecf36 | 4736 | struct extent_changeset *data_reserved = NULL; |
e6dcd2dc | 4737 | char *kaddr; |
0b246afa | 4738 | u32 blocksize = fs_info->sectorsize; |
09cbfeaf | 4739 | pgoff_t index = from >> PAGE_SHIFT; |
9703fefe | 4740 | unsigned offset = from & (blocksize - 1); |
39279cc3 | 4741 | struct page *page; |
3b16a4e3 | 4742 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 4743 | int ret = 0; |
9703fefe CR |
4744 | u64 block_start; |
4745 | u64 block_end; | |
39279cc3 | 4746 | |
2aaa6655 JB |
4747 | if ((offset & (blocksize - 1)) == 0 && |
4748 | (!len || ((len & (blocksize - 1)) == 0))) | |
39279cc3 | 4749 | goto out; |
9703fefe | 4750 | |
8b62f87b JB |
4751 | block_start = round_down(from, blocksize); |
4752 | block_end = block_start + blocksize - 1; | |
4753 | ||
364ecf36 | 4754 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, |
8b62f87b | 4755 | block_start, blocksize); |
5d5e103a JB |
4756 | if (ret) |
4757 | goto out; | |
39279cc3 | 4758 | |
211c17f5 | 4759 | again: |
3b16a4e3 | 4760 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 4761 | if (!page) { |
bc42bda2 | 4762 | btrfs_delalloc_release_space(inode, data_reserved, |
8b62f87b JB |
4763 | block_start, blocksize); |
4764 | btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize); | |
ac6a2b36 | 4765 | ret = -ENOMEM; |
39279cc3 | 4766 | goto out; |
5d5e103a | 4767 | } |
e6dcd2dc | 4768 | |
39279cc3 | 4769 | if (!PageUptodate(page)) { |
9ebefb18 | 4770 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 4771 | lock_page(page); |
211c17f5 CM |
4772 | if (page->mapping != mapping) { |
4773 | unlock_page(page); | |
09cbfeaf | 4774 | put_page(page); |
211c17f5 CM |
4775 | goto again; |
4776 | } | |
39279cc3 CM |
4777 | if (!PageUptodate(page)) { |
4778 | ret = -EIO; | |
89642229 | 4779 | goto out_unlock; |
39279cc3 CM |
4780 | } |
4781 | } | |
211c17f5 | 4782 | wait_on_page_writeback(page); |
e6dcd2dc | 4783 | |
9703fefe | 4784 | lock_extent_bits(io_tree, block_start, block_end, &cached_state); |
e6dcd2dc CM |
4785 | set_page_extent_mapped(page); |
4786 | ||
9703fefe | 4787 | ordered = btrfs_lookup_ordered_extent(inode, block_start); |
e6dcd2dc | 4788 | if (ordered) { |
9703fefe | 4789 | unlock_extent_cached(io_tree, block_start, block_end, |
2ac55d41 | 4790 | &cached_state, GFP_NOFS); |
e6dcd2dc | 4791 | unlock_page(page); |
09cbfeaf | 4792 | put_page(page); |
eb84ae03 | 4793 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
4794 | btrfs_put_ordered_extent(ordered); |
4795 | goto again; | |
4796 | } | |
4797 | ||
9703fefe | 4798 | clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end, |
9e8a4a8b LB |
4799 | EXTENT_DIRTY | EXTENT_DELALLOC | |
4800 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 4801 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 4802 | |
e3b8a485 | 4803 | ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0, |
ba8b04c1 | 4804 | &cached_state, 0); |
9ed74f2d | 4805 | if (ret) { |
9703fefe | 4806 | unlock_extent_cached(io_tree, block_start, block_end, |
2ac55d41 | 4807 | &cached_state, GFP_NOFS); |
9ed74f2d JB |
4808 | goto out_unlock; |
4809 | } | |
4810 | ||
9703fefe | 4811 | if (offset != blocksize) { |
2aaa6655 | 4812 | if (!len) |
9703fefe | 4813 | len = blocksize - offset; |
e6dcd2dc | 4814 | kaddr = kmap(page); |
2aaa6655 | 4815 | if (front) |
9703fefe CR |
4816 | memset(kaddr + (block_start - page_offset(page)), |
4817 | 0, offset); | |
2aaa6655 | 4818 | else |
9703fefe CR |
4819 | memset(kaddr + (block_start - page_offset(page)) + offset, |
4820 | 0, len); | |
e6dcd2dc CM |
4821 | flush_dcache_page(page); |
4822 | kunmap(page); | |
4823 | } | |
247e743c | 4824 | ClearPageChecked(page); |
e6dcd2dc | 4825 | set_page_dirty(page); |
9703fefe | 4826 | unlock_extent_cached(io_tree, block_start, block_end, &cached_state, |
2ac55d41 | 4827 | GFP_NOFS); |
39279cc3 | 4828 | |
89642229 | 4829 | out_unlock: |
5d5e103a | 4830 | if (ret) |
bc42bda2 | 4831 | btrfs_delalloc_release_space(inode, data_reserved, block_start, |
9703fefe | 4832 | blocksize); |
8b62f87b | 4833 | btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize); |
39279cc3 | 4834 | unlock_page(page); |
09cbfeaf | 4835 | put_page(page); |
39279cc3 | 4836 | out: |
364ecf36 | 4837 | extent_changeset_free(data_reserved); |
39279cc3 CM |
4838 | return ret; |
4839 | } | |
4840 | ||
16e7549f JB |
4841 | static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode, |
4842 | u64 offset, u64 len) | |
4843 | { | |
0b246afa | 4844 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
16e7549f JB |
4845 | struct btrfs_trans_handle *trans; |
4846 | int ret; | |
4847 | ||
4848 | /* | |
4849 | * Still need to make sure the inode looks like it's been updated so | |
4850 | * that any holes get logged if we fsync. | |
4851 | */ | |
0b246afa JM |
4852 | if (btrfs_fs_incompat(fs_info, NO_HOLES)) { |
4853 | BTRFS_I(inode)->last_trans = fs_info->generation; | |
16e7549f JB |
4854 | BTRFS_I(inode)->last_sub_trans = root->log_transid; |
4855 | BTRFS_I(inode)->last_log_commit = root->last_log_commit; | |
4856 | return 0; | |
4857 | } | |
4858 | ||
4859 | /* | |
4860 | * 1 - for the one we're dropping | |
4861 | * 1 - for the one we're adding | |
4862 | * 1 - for updating the inode. | |
4863 | */ | |
4864 | trans = btrfs_start_transaction(root, 3); | |
4865 | if (IS_ERR(trans)) | |
4866 | return PTR_ERR(trans); | |
4867 | ||
4868 | ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1); | |
4869 | if (ret) { | |
66642832 | 4870 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4871 | btrfs_end_transaction(trans); |
16e7549f JB |
4872 | return ret; |
4873 | } | |
4874 | ||
f85b7379 DS |
4875 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)), |
4876 | offset, 0, 0, len, 0, len, 0, 0, 0); | |
16e7549f | 4877 | if (ret) |
66642832 | 4878 | btrfs_abort_transaction(trans, ret); |
16e7549f JB |
4879 | else |
4880 | btrfs_update_inode(trans, root, inode); | |
3a45bb20 | 4881 | btrfs_end_transaction(trans); |
16e7549f JB |
4882 | return ret; |
4883 | } | |
4884 | ||
695a0d0d JB |
4885 | /* |
4886 | * This function puts in dummy file extents for the area we're creating a hole | |
4887 | * for. So if we are truncating this file to a larger size we need to insert | |
4888 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
4889 | * the range between oldsize and size | |
4890 | */ | |
a41ad394 | 4891 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 4892 | { |
0b246afa | 4893 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
9036c102 YZ |
4894 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4895 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 4896 | struct extent_map *em = NULL; |
2ac55d41 | 4897 | struct extent_state *cached_state = NULL; |
5dc562c5 | 4898 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
0b246afa JM |
4899 | u64 hole_start = ALIGN(oldsize, fs_info->sectorsize); |
4900 | u64 block_end = ALIGN(size, fs_info->sectorsize); | |
9036c102 YZ |
4901 | u64 last_byte; |
4902 | u64 cur_offset; | |
4903 | u64 hole_size; | |
9ed74f2d | 4904 | int err = 0; |
39279cc3 | 4905 | |
a71754fc | 4906 | /* |
9703fefe CR |
4907 | * If our size started in the middle of a block we need to zero out the |
4908 | * rest of the block before we expand the i_size, otherwise we could | |
a71754fc JB |
4909 | * expose stale data. |
4910 | */ | |
9703fefe | 4911 | err = btrfs_truncate_block(inode, oldsize, 0, 0); |
a71754fc JB |
4912 | if (err) |
4913 | return err; | |
4914 | ||
9036c102 YZ |
4915 | if (size <= hole_start) |
4916 | return 0; | |
4917 | ||
9036c102 YZ |
4918 | while (1) { |
4919 | struct btrfs_ordered_extent *ordered; | |
fa7c1494 | 4920 | |
ff13db41 | 4921 | lock_extent_bits(io_tree, hole_start, block_end - 1, |
d0082371 | 4922 | &cached_state); |
a776c6fa | 4923 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start, |
fa7c1494 | 4924 | block_end - hole_start); |
9036c102 YZ |
4925 | if (!ordered) |
4926 | break; | |
2ac55d41 JB |
4927 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
4928 | &cached_state, GFP_NOFS); | |
fa7c1494 | 4929 | btrfs_start_ordered_extent(inode, ordered, 1); |
9036c102 YZ |
4930 | btrfs_put_ordered_extent(ordered); |
4931 | } | |
39279cc3 | 4932 | |
9036c102 YZ |
4933 | cur_offset = hole_start; |
4934 | while (1) { | |
fc4f21b1 | 4935 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset, |
9036c102 | 4936 | block_end - cur_offset, 0); |
79787eaa JM |
4937 | if (IS_ERR(em)) { |
4938 | err = PTR_ERR(em); | |
f2767956 | 4939 | em = NULL; |
79787eaa JM |
4940 | break; |
4941 | } | |
9036c102 | 4942 | last_byte = min(extent_map_end(em), block_end); |
0b246afa | 4943 | last_byte = ALIGN(last_byte, fs_info->sectorsize); |
8082510e | 4944 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 4945 | struct extent_map *hole_em; |
9036c102 | 4946 | hole_size = last_byte - cur_offset; |
9ed74f2d | 4947 | |
16e7549f JB |
4948 | err = maybe_insert_hole(root, inode, cur_offset, |
4949 | hole_size); | |
4950 | if (err) | |
3893e33b | 4951 | break; |
dcdbc059 | 4952 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
5dc562c5 JB |
4953 | cur_offset + hole_size - 1, 0); |
4954 | hole_em = alloc_extent_map(); | |
4955 | if (!hole_em) { | |
4956 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
4957 | &BTRFS_I(inode)->runtime_flags); | |
4958 | goto next; | |
4959 | } | |
4960 | hole_em->start = cur_offset; | |
4961 | hole_em->len = hole_size; | |
4962 | hole_em->orig_start = cur_offset; | |
8082510e | 4963 | |
5dc562c5 JB |
4964 | hole_em->block_start = EXTENT_MAP_HOLE; |
4965 | hole_em->block_len = 0; | |
b4939680 | 4966 | hole_em->orig_block_len = 0; |
cc95bef6 | 4967 | hole_em->ram_bytes = hole_size; |
0b246afa | 4968 | hole_em->bdev = fs_info->fs_devices->latest_bdev; |
5dc562c5 | 4969 | hole_em->compress_type = BTRFS_COMPRESS_NONE; |
0b246afa | 4970 | hole_em->generation = fs_info->generation; |
8082510e | 4971 | |
5dc562c5 JB |
4972 | while (1) { |
4973 | write_lock(&em_tree->lock); | |
09a2a8f9 | 4974 | err = add_extent_mapping(em_tree, hole_em, 1); |
5dc562c5 JB |
4975 | write_unlock(&em_tree->lock); |
4976 | if (err != -EEXIST) | |
4977 | break; | |
dcdbc059 NB |
4978 | btrfs_drop_extent_cache(BTRFS_I(inode), |
4979 | cur_offset, | |
5dc562c5 JB |
4980 | cur_offset + |
4981 | hole_size - 1, 0); | |
4982 | } | |
4983 | free_extent_map(hole_em); | |
9036c102 | 4984 | } |
16e7549f | 4985 | next: |
9036c102 | 4986 | free_extent_map(em); |
a22285a6 | 4987 | em = NULL; |
9036c102 | 4988 | cur_offset = last_byte; |
8082510e | 4989 | if (cur_offset >= block_end) |
9036c102 YZ |
4990 | break; |
4991 | } | |
a22285a6 | 4992 | free_extent_map(em); |
2ac55d41 JB |
4993 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
4994 | GFP_NOFS); | |
9036c102 YZ |
4995 | return err; |
4996 | } | |
39279cc3 | 4997 | |
3972f260 | 4998 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 4999 | { |
f4a2f4c5 MX |
5000 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5001 | struct btrfs_trans_handle *trans; | |
a41ad394 | 5002 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
5003 | loff_t newsize = attr->ia_size; |
5004 | int mask = attr->ia_valid; | |
8082510e YZ |
5005 | int ret; |
5006 | ||
3972f260 ES |
5007 | /* |
5008 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
5009 | * special case where we need to update the times despite not having | |
5010 | * these flags set. For all other operations the VFS set these flags | |
5011 | * explicitly if it wants a timestamp update. | |
5012 | */ | |
dff6efc3 CH |
5013 | if (newsize != oldsize) { |
5014 | inode_inc_iversion(inode); | |
5015 | if (!(mask & (ATTR_CTIME | ATTR_MTIME))) | |
5016 | inode->i_ctime = inode->i_mtime = | |
c2050a45 | 5017 | current_time(inode); |
dff6efc3 | 5018 | } |
3972f260 | 5019 | |
a41ad394 | 5020 | if (newsize > oldsize) { |
9ea24bbe | 5021 | /* |
ea14b57f | 5022 | * Don't do an expanding truncate while snapshotting is ongoing. |
9ea24bbe FM |
5023 | * This is to ensure the snapshot captures a fully consistent |
5024 | * state of this file - if the snapshot captures this expanding | |
5025 | * truncation, it must capture all writes that happened before | |
5026 | * this truncation. | |
5027 | */ | |
0bc19f90 | 5028 | btrfs_wait_for_snapshot_creation(root); |
a41ad394 | 5029 | ret = btrfs_cont_expand(inode, oldsize, newsize); |
9ea24bbe | 5030 | if (ret) { |
ea14b57f | 5031 | btrfs_end_write_no_snapshotting(root); |
8082510e | 5032 | return ret; |
9ea24bbe | 5033 | } |
8082510e | 5034 | |
f4a2f4c5 | 5035 | trans = btrfs_start_transaction(root, 1); |
9ea24bbe | 5036 | if (IS_ERR(trans)) { |
ea14b57f | 5037 | btrfs_end_write_no_snapshotting(root); |
f4a2f4c5 | 5038 | return PTR_ERR(trans); |
9ea24bbe | 5039 | } |
f4a2f4c5 MX |
5040 | |
5041 | i_size_write(inode, newsize); | |
5042 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
27772b68 | 5043 | pagecache_isize_extended(inode, oldsize, newsize); |
f4a2f4c5 | 5044 | ret = btrfs_update_inode(trans, root, inode); |
ea14b57f | 5045 | btrfs_end_write_no_snapshotting(root); |
3a45bb20 | 5046 | btrfs_end_transaction(trans); |
a41ad394 | 5047 | } else { |
8082510e | 5048 | |
a41ad394 JB |
5049 | /* |
5050 | * We're truncating a file that used to have good data down to | |
5051 | * zero. Make sure it gets into the ordered flush list so that | |
5052 | * any new writes get down to disk quickly. | |
5053 | */ | |
5054 | if (newsize == 0) | |
72ac3c0d JB |
5055 | set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
5056 | &BTRFS_I(inode)->runtime_flags); | |
8082510e | 5057 | |
f3fe820c JB |
5058 | /* |
5059 | * 1 for the orphan item we're going to add | |
5060 | * 1 for the orphan item deletion. | |
5061 | */ | |
5062 | trans = btrfs_start_transaction(root, 2); | |
5063 | if (IS_ERR(trans)) | |
5064 | return PTR_ERR(trans); | |
5065 | ||
5066 | /* | |
5067 | * We need to do this in case we fail at _any_ point during the | |
5068 | * actual truncate. Once we do the truncate_setsize we could | |
5069 | * invalidate pages which forces any outstanding ordered io to | |
5070 | * be instantly completed which will give us extents that need | |
5071 | * to be truncated. If we fail to get an orphan inode down we | |
5072 | * could have left over extents that were never meant to live, | |
01327610 | 5073 | * so we need to guarantee from this point on that everything |
f3fe820c JB |
5074 | * will be consistent. |
5075 | */ | |
73f2e545 | 5076 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
3a45bb20 | 5077 | btrfs_end_transaction(trans); |
f3fe820c JB |
5078 | if (ret) |
5079 | return ret; | |
5080 | ||
a41ad394 JB |
5081 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
5082 | truncate_setsize(inode, newsize); | |
2e60a51e MX |
5083 | |
5084 | /* Disable nonlocked read DIO to avoid the end less truncate */ | |
abcefb1e | 5085 | btrfs_inode_block_unlocked_dio(BTRFS_I(inode)); |
2e60a51e | 5086 | inode_dio_wait(inode); |
0b581701 | 5087 | btrfs_inode_resume_unlocked_dio(BTRFS_I(inode)); |
2e60a51e | 5088 | |
a41ad394 | 5089 | ret = btrfs_truncate(inode); |
7f4f6e0a JB |
5090 | if (ret && inode->i_nlink) { |
5091 | int err; | |
5092 | ||
19fd2df5 LB |
5093 | /* To get a stable disk_i_size */ |
5094 | err = btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
5095 | if (err) { | |
3d6ae7bb | 5096 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
19fd2df5 LB |
5097 | return err; |
5098 | } | |
5099 | ||
7f4f6e0a JB |
5100 | /* |
5101 | * failed to truncate, disk_i_size is only adjusted down | |
5102 | * as we remove extents, so it should represent the true | |
5103 | * size of the inode, so reset the in memory size and | |
5104 | * delete our orphan entry. | |
5105 | */ | |
5106 | trans = btrfs_join_transaction(root); | |
5107 | if (IS_ERR(trans)) { | |
3d6ae7bb | 5108 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
7f4f6e0a JB |
5109 | return ret; |
5110 | } | |
5111 | i_size_write(inode, BTRFS_I(inode)->disk_i_size); | |
3d6ae7bb | 5112 | err = btrfs_orphan_del(trans, BTRFS_I(inode)); |
7f4f6e0a | 5113 | if (err) |
66642832 | 5114 | btrfs_abort_transaction(trans, err); |
3a45bb20 | 5115 | btrfs_end_transaction(trans); |
7f4f6e0a | 5116 | } |
8082510e YZ |
5117 | } |
5118 | ||
a41ad394 | 5119 | return ret; |
8082510e YZ |
5120 | } |
5121 | ||
9036c102 YZ |
5122 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
5123 | { | |
2b0143b5 | 5124 | struct inode *inode = d_inode(dentry); |
b83cc969 | 5125 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 5126 | int err; |
39279cc3 | 5127 | |
b83cc969 LZ |
5128 | if (btrfs_root_readonly(root)) |
5129 | return -EROFS; | |
5130 | ||
31051c85 | 5131 | err = setattr_prepare(dentry, attr); |
9036c102 YZ |
5132 | if (err) |
5133 | return err; | |
2bf5a725 | 5134 | |
5a3f23d5 | 5135 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 5136 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
5137 | if (err) |
5138 | return err; | |
39279cc3 | 5139 | } |
9036c102 | 5140 | |
1025774c CH |
5141 | if (attr->ia_valid) { |
5142 | setattr_copy(inode, attr); | |
0c4d2d95 | 5143 | inode_inc_iversion(inode); |
22c44fe6 | 5144 | err = btrfs_dirty_inode(inode); |
1025774c | 5145 | |
22c44fe6 | 5146 | if (!err && attr->ia_valid & ATTR_MODE) |
996a710d | 5147 | err = posix_acl_chmod(inode, inode->i_mode); |
1025774c | 5148 | } |
33268eaf | 5149 | |
39279cc3 CM |
5150 | return err; |
5151 | } | |
61295eb8 | 5152 | |
131e404a FDBM |
5153 | /* |
5154 | * While truncating the inode pages during eviction, we get the VFS calling | |
5155 | * btrfs_invalidatepage() against each page of the inode. This is slow because | |
5156 | * the calls to btrfs_invalidatepage() result in a huge amount of calls to | |
5157 | * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting | |
5158 | * extent_state structures over and over, wasting lots of time. | |
5159 | * | |
5160 | * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all | |
5161 | * those expensive operations on a per page basis and do only the ordered io | |
5162 | * finishing, while we release here the extent_map and extent_state structures, | |
5163 | * without the excessive merging and splitting. | |
5164 | */ | |
5165 | static void evict_inode_truncate_pages(struct inode *inode) | |
5166 | { | |
5167 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
5168 | struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree; | |
5169 | struct rb_node *node; | |
5170 | ||
5171 | ASSERT(inode->i_state & I_FREEING); | |
91b0abe3 | 5172 | truncate_inode_pages_final(&inode->i_data); |
131e404a FDBM |
5173 | |
5174 | write_lock(&map_tree->lock); | |
5175 | while (!RB_EMPTY_ROOT(&map_tree->map)) { | |
5176 | struct extent_map *em; | |
5177 | ||
5178 | node = rb_first(&map_tree->map); | |
5179 | em = rb_entry(node, struct extent_map, rb_node); | |
180589ef WS |
5180 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
5181 | clear_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
131e404a FDBM |
5182 | remove_extent_mapping(map_tree, em); |
5183 | free_extent_map(em); | |
7064dd5c FM |
5184 | if (need_resched()) { |
5185 | write_unlock(&map_tree->lock); | |
5186 | cond_resched(); | |
5187 | write_lock(&map_tree->lock); | |
5188 | } | |
131e404a FDBM |
5189 | } |
5190 | write_unlock(&map_tree->lock); | |
5191 | ||
6ca07097 FM |
5192 | /* |
5193 | * Keep looping until we have no more ranges in the io tree. | |
5194 | * We can have ongoing bios started by readpages (called from readahead) | |
9c6429d9 FM |
5195 | * that have their endio callback (extent_io.c:end_bio_extent_readpage) |
5196 | * still in progress (unlocked the pages in the bio but did not yet | |
5197 | * unlocked the ranges in the io tree). Therefore this means some | |
6ca07097 FM |
5198 | * ranges can still be locked and eviction started because before |
5199 | * submitting those bios, which are executed by a separate task (work | |
5200 | * queue kthread), inode references (inode->i_count) were not taken | |
5201 | * (which would be dropped in the end io callback of each bio). | |
5202 | * Therefore here we effectively end up waiting for those bios and | |
5203 | * anyone else holding locked ranges without having bumped the inode's | |
5204 | * reference count - if we don't do it, when they access the inode's | |
5205 | * io_tree to unlock a range it may be too late, leading to an | |
5206 | * use-after-free issue. | |
5207 | */ | |
131e404a FDBM |
5208 | spin_lock(&io_tree->lock); |
5209 | while (!RB_EMPTY_ROOT(&io_tree->state)) { | |
5210 | struct extent_state *state; | |
5211 | struct extent_state *cached_state = NULL; | |
6ca07097 FM |
5212 | u64 start; |
5213 | u64 end; | |
131e404a FDBM |
5214 | |
5215 | node = rb_first(&io_tree->state); | |
5216 | state = rb_entry(node, struct extent_state, rb_node); | |
6ca07097 FM |
5217 | start = state->start; |
5218 | end = state->end; | |
131e404a FDBM |
5219 | spin_unlock(&io_tree->lock); |
5220 | ||
ff13db41 | 5221 | lock_extent_bits(io_tree, start, end, &cached_state); |
b9d0b389 QW |
5222 | |
5223 | /* | |
5224 | * If still has DELALLOC flag, the extent didn't reach disk, | |
5225 | * and its reserved space won't be freed by delayed_ref. | |
5226 | * So we need to free its reserved space here. | |
5227 | * (Refer to comment in btrfs_invalidatepage, case 2) | |
5228 | * | |
5229 | * Note, end is the bytenr of last byte, so we need + 1 here. | |
5230 | */ | |
5231 | if (state->state & EXTENT_DELALLOC) | |
bc42bda2 | 5232 | btrfs_qgroup_free_data(inode, NULL, start, end - start + 1); |
b9d0b389 | 5233 | |
6ca07097 | 5234 | clear_extent_bit(io_tree, start, end, |
131e404a FDBM |
5235 | EXTENT_LOCKED | EXTENT_DIRTY | |
5236 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
5237 | EXTENT_DEFRAG, 1, 1, | |
5238 | &cached_state, GFP_NOFS); | |
131e404a | 5239 | |
7064dd5c | 5240 | cond_resched(); |
131e404a FDBM |
5241 | spin_lock(&io_tree->lock); |
5242 | } | |
5243 | spin_unlock(&io_tree->lock); | |
5244 | } | |
5245 | ||
bd555975 | 5246 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 | 5247 | { |
0b246afa | 5248 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
5249 | struct btrfs_trans_handle *trans; |
5250 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
726c35fa | 5251 | struct btrfs_block_rsv *rsv, *global_rsv; |
3bce876f | 5252 | int steal_from_global = 0; |
3d48d981 | 5253 | u64 min_size; |
39279cc3 CM |
5254 | int ret; |
5255 | ||
1abe9b8a | 5256 | trace_btrfs_inode_evict(inode); |
5257 | ||
3d48d981 NB |
5258 | if (!root) { |
5259 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); | |
5260 | return; | |
5261 | } | |
5262 | ||
0b246afa | 5263 | min_size = btrfs_calc_trunc_metadata_size(fs_info, 1); |
3d48d981 | 5264 | |
131e404a FDBM |
5265 | evict_inode_truncate_pages(inode); |
5266 | ||
69e9c6c6 SB |
5267 | if (inode->i_nlink && |
5268 | ((btrfs_root_refs(&root->root_item) != 0 && | |
5269 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) || | |
70ddc553 | 5270 | btrfs_is_free_space_inode(BTRFS_I(inode)))) |
bd555975 AV |
5271 | goto no_delete; |
5272 | ||
39279cc3 | 5273 | if (is_bad_inode(inode)) { |
3d6ae7bb | 5274 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
39279cc3 CM |
5275 | goto no_delete; |
5276 | } | |
bd555975 | 5277 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
a30e577c JM |
5278 | if (!special_file(inode->i_mode)) |
5279 | btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
5f39d397 | 5280 | |
7ab7956e | 5281 | btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1); |
f612496b | 5282 | |
0b246afa | 5283 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) { |
6bf02314 | 5284 | BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
8a35d95f | 5285 | &BTRFS_I(inode)->runtime_flags)); |
c71bf099 YZ |
5286 | goto no_delete; |
5287 | } | |
5288 | ||
76dda93c | 5289 | if (inode->i_nlink > 0) { |
69e9c6c6 SB |
5290 | BUG_ON(btrfs_root_refs(&root->root_item) != 0 && |
5291 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID); | |
76dda93c YZ |
5292 | goto no_delete; |
5293 | } | |
5294 | ||
aa79021f | 5295 | ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode)); |
0e8c36a9 | 5296 | if (ret) { |
3d6ae7bb | 5297 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
0e8c36a9 MX |
5298 | goto no_delete; |
5299 | } | |
5300 | ||
2ff7e61e | 5301 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
4289a667 | 5302 | if (!rsv) { |
3d6ae7bb | 5303 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
4289a667 JB |
5304 | goto no_delete; |
5305 | } | |
4a338542 | 5306 | rsv->size = min_size; |
ca7e70f5 | 5307 | rsv->failfast = 1; |
0b246afa | 5308 | global_rsv = &fs_info->global_block_rsv; |
4289a667 | 5309 | |
6ef06d27 | 5310 | btrfs_i_size_write(BTRFS_I(inode), 0); |
5f39d397 | 5311 | |
4289a667 | 5312 | /* |
8407aa46 MX |
5313 | * This is a bit simpler than btrfs_truncate since we've already |
5314 | * reserved our space for our orphan item in the unlink, so we just | |
5315 | * need to reserve some slack space in case we add bytes and update | |
5316 | * inode item when doing the truncate. | |
4289a667 | 5317 | */ |
8082510e | 5318 | while (1) { |
08e007d2 MX |
5319 | ret = btrfs_block_rsv_refill(root, rsv, min_size, |
5320 | BTRFS_RESERVE_FLUSH_LIMIT); | |
726c35fa JB |
5321 | |
5322 | /* | |
5323 | * Try and steal from the global reserve since we will | |
5324 | * likely not use this space anyway, we want to try as | |
5325 | * hard as possible to get this to work. | |
5326 | */ | |
5327 | if (ret) | |
3bce876f JB |
5328 | steal_from_global++; |
5329 | else | |
5330 | steal_from_global = 0; | |
5331 | ret = 0; | |
d68fc57b | 5332 | |
3bce876f JB |
5333 | /* |
5334 | * steal_from_global == 0: we reserved stuff, hooray! | |
5335 | * steal_from_global == 1: we didn't reserve stuff, boo! | |
5336 | * steal_from_global == 2: we've committed, still not a lot of | |
5337 | * room but maybe we'll have room in the global reserve this | |
5338 | * time. | |
5339 | * steal_from_global == 3: abandon all hope! | |
5340 | */ | |
5341 | if (steal_from_global > 2) { | |
0b246afa JM |
5342 | btrfs_warn(fs_info, |
5343 | "Could not get space for a delete, will truncate on mount %d", | |
5344 | ret); | |
3d6ae7bb | 5345 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
2ff7e61e | 5346 | btrfs_free_block_rsv(fs_info, rsv); |
4289a667 | 5347 | goto no_delete; |
d68fc57b | 5348 | } |
7b128766 | 5349 | |
0e8c36a9 | 5350 | trans = btrfs_join_transaction(root); |
4289a667 | 5351 | if (IS_ERR(trans)) { |
3d6ae7bb | 5352 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
2ff7e61e | 5353 | btrfs_free_block_rsv(fs_info, rsv); |
4289a667 | 5354 | goto no_delete; |
d68fc57b | 5355 | } |
7b128766 | 5356 | |
3bce876f | 5357 | /* |
01327610 | 5358 | * We can't just steal from the global reserve, we need to make |
3bce876f JB |
5359 | * sure there is room to do it, if not we need to commit and try |
5360 | * again. | |
5361 | */ | |
5362 | if (steal_from_global) { | |
2ff7e61e | 5363 | if (!btrfs_check_space_for_delayed_refs(trans, fs_info)) |
3bce876f | 5364 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, |
25d609f8 | 5365 | min_size, 0); |
3bce876f JB |
5366 | else |
5367 | ret = -ENOSPC; | |
5368 | } | |
5369 | ||
5370 | /* | |
5371 | * Couldn't steal from the global reserve, we have too much | |
5372 | * pending stuff built up, commit the transaction and try it | |
5373 | * again. | |
5374 | */ | |
5375 | if (ret) { | |
3a45bb20 | 5376 | ret = btrfs_commit_transaction(trans); |
3bce876f | 5377 | if (ret) { |
3d6ae7bb | 5378 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
2ff7e61e | 5379 | btrfs_free_block_rsv(fs_info, rsv); |
3bce876f JB |
5380 | goto no_delete; |
5381 | } | |
5382 | continue; | |
5383 | } else { | |
5384 | steal_from_global = 0; | |
5385 | } | |
5386 | ||
4289a667 JB |
5387 | trans->block_rsv = rsv; |
5388 | ||
d68fc57b | 5389 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
28ed1345 | 5390 | if (ret != -ENOSPC && ret != -EAGAIN) |
8082510e | 5391 | break; |
85e21bac | 5392 | |
0b246afa | 5393 | trans->block_rsv = &fs_info->trans_block_rsv; |
3a45bb20 | 5394 | btrfs_end_transaction(trans); |
8082510e | 5395 | trans = NULL; |
2ff7e61e | 5396 | btrfs_btree_balance_dirty(fs_info); |
8082510e | 5397 | } |
5f39d397 | 5398 | |
2ff7e61e | 5399 | btrfs_free_block_rsv(fs_info, rsv); |
4289a667 | 5400 | |
4ef31a45 JB |
5401 | /* |
5402 | * Errors here aren't a big deal, it just means we leave orphan items | |
5403 | * in the tree. They will be cleaned up on the next mount. | |
5404 | */ | |
8082510e | 5405 | if (ret == 0) { |
4289a667 | 5406 | trans->block_rsv = root->orphan_block_rsv; |
3d6ae7bb | 5407 | btrfs_orphan_del(trans, BTRFS_I(inode)); |
4ef31a45 | 5408 | } else { |
3d6ae7bb | 5409 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
8082510e | 5410 | } |
54aa1f4d | 5411 | |
0b246afa JM |
5412 | trans->block_rsv = &fs_info->trans_block_rsv; |
5413 | if (!(root == fs_info->tree_root || | |
581bb050 | 5414 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) |
4a0cc7ca | 5415 | btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode))); |
581bb050 | 5416 | |
3a45bb20 | 5417 | btrfs_end_transaction(trans); |
2ff7e61e | 5418 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 | 5419 | no_delete: |
f48d1cf5 | 5420 | btrfs_remove_delayed_node(BTRFS_I(inode)); |
dbd5768f | 5421 | clear_inode(inode); |
39279cc3 CM |
5422 | } |
5423 | ||
5424 | /* | |
5425 | * this returns the key found in the dir entry in the location pointer. | |
5426 | * If no dir entries were found, location->objectid is 0. | |
5427 | */ | |
5428 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
5429 | struct btrfs_key *location) | |
5430 | { | |
5431 | const char *name = dentry->d_name.name; | |
5432 | int namelen = dentry->d_name.len; | |
5433 | struct btrfs_dir_item *di; | |
5434 | struct btrfs_path *path; | |
5435 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 5436 | int ret = 0; |
39279cc3 CM |
5437 | |
5438 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
5439 | if (!path) |
5440 | return -ENOMEM; | |
3954401f | 5441 | |
f85b7379 DS |
5442 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)), |
5443 | name, namelen, 0); | |
0d9f7f3e Y |
5444 | if (IS_ERR(di)) |
5445 | ret = PTR_ERR(di); | |
d397712b | 5446 | |
c704005d | 5447 | if (IS_ERR_OR_NULL(di)) |
3954401f | 5448 | goto out_err; |
d397712b | 5449 | |
5f39d397 | 5450 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
56a0e706 LB |
5451 | if (location->type != BTRFS_INODE_ITEM_KEY && |
5452 | location->type != BTRFS_ROOT_ITEM_KEY) { | |
5453 | btrfs_warn(root->fs_info, | |
5454 | "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))", | |
5455 | __func__, name, btrfs_ino(BTRFS_I(dir)), | |
5456 | location->objectid, location->type, location->offset); | |
5457 | goto out_err; | |
5458 | } | |
39279cc3 | 5459 | out: |
39279cc3 CM |
5460 | btrfs_free_path(path); |
5461 | return ret; | |
3954401f CM |
5462 | out_err: |
5463 | location->objectid = 0; | |
5464 | goto out; | |
39279cc3 CM |
5465 | } |
5466 | ||
5467 | /* | |
5468 | * when we hit a tree root in a directory, the btrfs part of the inode | |
5469 | * needs to be changed to reflect the root directory of the tree root. This | |
5470 | * is kind of like crossing a mount point. | |
5471 | */ | |
2ff7e61e | 5472 | static int fixup_tree_root_location(struct btrfs_fs_info *fs_info, |
4df27c4d YZ |
5473 | struct inode *dir, |
5474 | struct dentry *dentry, | |
5475 | struct btrfs_key *location, | |
5476 | struct btrfs_root **sub_root) | |
39279cc3 | 5477 | { |
4df27c4d YZ |
5478 | struct btrfs_path *path; |
5479 | struct btrfs_root *new_root; | |
5480 | struct btrfs_root_ref *ref; | |
5481 | struct extent_buffer *leaf; | |
1d4c08e0 | 5482 | struct btrfs_key key; |
4df27c4d YZ |
5483 | int ret; |
5484 | int err = 0; | |
39279cc3 | 5485 | |
4df27c4d YZ |
5486 | path = btrfs_alloc_path(); |
5487 | if (!path) { | |
5488 | err = -ENOMEM; | |
5489 | goto out; | |
5490 | } | |
39279cc3 | 5491 | |
4df27c4d | 5492 | err = -ENOENT; |
1d4c08e0 DS |
5493 | key.objectid = BTRFS_I(dir)->root->root_key.objectid; |
5494 | key.type = BTRFS_ROOT_REF_KEY; | |
5495 | key.offset = location->objectid; | |
5496 | ||
0b246afa | 5497 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
4df27c4d YZ |
5498 | if (ret) { |
5499 | if (ret < 0) | |
5500 | err = ret; | |
5501 | goto out; | |
5502 | } | |
39279cc3 | 5503 | |
4df27c4d YZ |
5504 | leaf = path->nodes[0]; |
5505 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
4a0cc7ca | 5506 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) || |
4df27c4d YZ |
5507 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
5508 | goto out; | |
39279cc3 | 5509 | |
4df27c4d YZ |
5510 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
5511 | (unsigned long)(ref + 1), | |
5512 | dentry->d_name.len); | |
5513 | if (ret) | |
5514 | goto out; | |
5515 | ||
b3b4aa74 | 5516 | btrfs_release_path(path); |
4df27c4d | 5517 | |
0b246afa | 5518 | new_root = btrfs_read_fs_root_no_name(fs_info, location); |
4df27c4d YZ |
5519 | if (IS_ERR(new_root)) { |
5520 | err = PTR_ERR(new_root); | |
5521 | goto out; | |
5522 | } | |
5523 | ||
4df27c4d YZ |
5524 | *sub_root = new_root; |
5525 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
5526 | location->type = BTRFS_INODE_ITEM_KEY; | |
5527 | location->offset = 0; | |
5528 | err = 0; | |
5529 | out: | |
5530 | btrfs_free_path(path); | |
5531 | return err; | |
39279cc3 CM |
5532 | } |
5533 | ||
5d4f98a2 YZ |
5534 | static void inode_tree_add(struct inode *inode) |
5535 | { | |
5536 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5537 | struct btrfs_inode *entry; | |
03e860bd NP |
5538 | struct rb_node **p; |
5539 | struct rb_node *parent; | |
cef21937 | 5540 | struct rb_node *new = &BTRFS_I(inode)->rb_node; |
4a0cc7ca | 5541 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
5d4f98a2 | 5542 | |
1d3382cb | 5543 | if (inode_unhashed(inode)) |
76dda93c | 5544 | return; |
e1409cef | 5545 | parent = NULL; |
5d4f98a2 | 5546 | spin_lock(&root->inode_lock); |
e1409cef | 5547 | p = &root->inode_tree.rb_node; |
5d4f98a2 YZ |
5548 | while (*p) { |
5549 | parent = *p; | |
5550 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
5551 | ||
4a0cc7ca | 5552 | if (ino < btrfs_ino(BTRFS_I(&entry->vfs_inode))) |
03e860bd | 5553 | p = &parent->rb_left; |
4a0cc7ca | 5554 | else if (ino > btrfs_ino(BTRFS_I(&entry->vfs_inode))) |
03e860bd | 5555 | p = &parent->rb_right; |
5d4f98a2 YZ |
5556 | else { |
5557 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 5558 | (I_WILL_FREE | I_FREEING))); |
cef21937 | 5559 | rb_replace_node(parent, new, &root->inode_tree); |
03e860bd NP |
5560 | RB_CLEAR_NODE(parent); |
5561 | spin_unlock(&root->inode_lock); | |
cef21937 | 5562 | return; |
5d4f98a2 YZ |
5563 | } |
5564 | } | |
cef21937 FDBM |
5565 | rb_link_node(new, parent, p); |
5566 | rb_insert_color(new, &root->inode_tree); | |
5d4f98a2 YZ |
5567 | spin_unlock(&root->inode_lock); |
5568 | } | |
5569 | ||
5570 | static void inode_tree_del(struct inode *inode) | |
5571 | { | |
0b246afa | 5572 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5d4f98a2 | 5573 | struct btrfs_root *root = BTRFS_I(inode)->root; |
76dda93c | 5574 | int empty = 0; |
5d4f98a2 | 5575 | |
03e860bd | 5576 | spin_lock(&root->inode_lock); |
5d4f98a2 | 5577 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 5578 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 5579 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 5580 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 5581 | } |
03e860bd | 5582 | spin_unlock(&root->inode_lock); |
76dda93c | 5583 | |
69e9c6c6 | 5584 | if (empty && btrfs_root_refs(&root->root_item) == 0) { |
0b246afa | 5585 | synchronize_srcu(&fs_info->subvol_srcu); |
76dda93c YZ |
5586 | spin_lock(&root->inode_lock); |
5587 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
5588 | spin_unlock(&root->inode_lock); | |
5589 | if (empty) | |
5590 | btrfs_add_dead_root(root); | |
5591 | } | |
5592 | } | |
5593 | ||
143bede5 | 5594 | void btrfs_invalidate_inodes(struct btrfs_root *root) |
76dda93c | 5595 | { |
0b246afa | 5596 | struct btrfs_fs_info *fs_info = root->fs_info; |
76dda93c YZ |
5597 | struct rb_node *node; |
5598 | struct rb_node *prev; | |
5599 | struct btrfs_inode *entry; | |
5600 | struct inode *inode; | |
5601 | u64 objectid = 0; | |
5602 | ||
0b246afa | 5603 | if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
7813b3db | 5604 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); |
76dda93c YZ |
5605 | |
5606 | spin_lock(&root->inode_lock); | |
5607 | again: | |
5608 | node = root->inode_tree.rb_node; | |
5609 | prev = NULL; | |
5610 | while (node) { | |
5611 | prev = node; | |
5612 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
5613 | ||
4a0cc7ca | 5614 | if (objectid < btrfs_ino(BTRFS_I(&entry->vfs_inode))) |
76dda93c | 5615 | node = node->rb_left; |
4a0cc7ca | 5616 | else if (objectid > btrfs_ino(BTRFS_I(&entry->vfs_inode))) |
76dda93c YZ |
5617 | node = node->rb_right; |
5618 | else | |
5619 | break; | |
5620 | } | |
5621 | if (!node) { | |
5622 | while (prev) { | |
5623 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
4a0cc7ca | 5624 | if (objectid <= btrfs_ino(BTRFS_I(&entry->vfs_inode))) { |
76dda93c YZ |
5625 | node = prev; |
5626 | break; | |
5627 | } | |
5628 | prev = rb_next(prev); | |
5629 | } | |
5630 | } | |
5631 | while (node) { | |
5632 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4a0cc7ca | 5633 | objectid = btrfs_ino(BTRFS_I(&entry->vfs_inode)) + 1; |
76dda93c YZ |
5634 | inode = igrab(&entry->vfs_inode); |
5635 | if (inode) { | |
5636 | spin_unlock(&root->inode_lock); | |
5637 | if (atomic_read(&inode->i_count) > 1) | |
5638 | d_prune_aliases(inode); | |
5639 | /* | |
45321ac5 | 5640 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
5641 | * the inode cache when its usage count |
5642 | * hits zero. | |
5643 | */ | |
5644 | iput(inode); | |
5645 | cond_resched(); | |
5646 | spin_lock(&root->inode_lock); | |
5647 | goto again; | |
5648 | } | |
5649 | ||
5650 | if (cond_resched_lock(&root->inode_lock)) | |
5651 | goto again; | |
5652 | ||
5653 | node = rb_next(node); | |
5654 | } | |
5655 | spin_unlock(&root->inode_lock); | |
5d4f98a2 YZ |
5656 | } |
5657 | ||
e02119d5 CM |
5658 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
5659 | { | |
5660 | struct btrfs_iget_args *args = p; | |
90d3e592 CM |
5661 | inode->i_ino = args->location->objectid; |
5662 | memcpy(&BTRFS_I(inode)->location, args->location, | |
5663 | sizeof(*args->location)); | |
e02119d5 | 5664 | BTRFS_I(inode)->root = args->root; |
39279cc3 CM |
5665 | return 0; |
5666 | } | |
5667 | ||
5668 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
5669 | { | |
5670 | struct btrfs_iget_args *args = opaque; | |
90d3e592 | 5671 | return args->location->objectid == BTRFS_I(inode)->location.objectid && |
d397712b | 5672 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
5673 | } |
5674 | ||
5d4f98a2 | 5675 | static struct inode *btrfs_iget_locked(struct super_block *s, |
90d3e592 | 5676 | struct btrfs_key *location, |
5d4f98a2 | 5677 | struct btrfs_root *root) |
39279cc3 CM |
5678 | { |
5679 | struct inode *inode; | |
5680 | struct btrfs_iget_args args; | |
90d3e592 | 5681 | unsigned long hashval = btrfs_inode_hash(location->objectid, root); |
778ba82b | 5682 | |
90d3e592 | 5683 | args.location = location; |
39279cc3 CM |
5684 | args.root = root; |
5685 | ||
778ba82b | 5686 | inode = iget5_locked(s, hashval, btrfs_find_actor, |
39279cc3 CM |
5687 | btrfs_init_locked_inode, |
5688 | (void *)&args); | |
5689 | return inode; | |
5690 | } | |
5691 | ||
1a54ef8c BR |
5692 | /* Get an inode object given its location and corresponding root. |
5693 | * Returns in *is_new if the inode was read from disk | |
5694 | */ | |
5695 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 5696 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
5697 | { |
5698 | struct inode *inode; | |
5699 | ||
90d3e592 | 5700 | inode = btrfs_iget_locked(s, location, root); |
1a54ef8c | 5701 | if (!inode) |
5d4f98a2 | 5702 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
5703 | |
5704 | if (inode->i_state & I_NEW) { | |
67710892 FM |
5705 | int ret; |
5706 | ||
5707 | ret = btrfs_read_locked_inode(inode); | |
1748f843 MF |
5708 | if (!is_bad_inode(inode)) { |
5709 | inode_tree_add(inode); | |
5710 | unlock_new_inode(inode); | |
5711 | if (new) | |
5712 | *new = 1; | |
5713 | } else { | |
e0b6d65b ST |
5714 | unlock_new_inode(inode); |
5715 | iput(inode); | |
67710892 FM |
5716 | ASSERT(ret < 0); |
5717 | inode = ERR_PTR(ret < 0 ? ret : -ESTALE); | |
1748f843 MF |
5718 | } |
5719 | } | |
5720 | ||
1a54ef8c BR |
5721 | return inode; |
5722 | } | |
5723 | ||
4df27c4d YZ |
5724 | static struct inode *new_simple_dir(struct super_block *s, |
5725 | struct btrfs_key *key, | |
5726 | struct btrfs_root *root) | |
5727 | { | |
5728 | struct inode *inode = new_inode(s); | |
5729 | ||
5730 | if (!inode) | |
5731 | return ERR_PTR(-ENOMEM); | |
5732 | ||
4df27c4d YZ |
5733 | BTRFS_I(inode)->root = root; |
5734 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
72ac3c0d | 5735 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
5736 | |
5737 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
848cce0d | 5738 | inode->i_op = &btrfs_dir_ro_inode_operations; |
1fdf4194 | 5739 | inode->i_opflags &= ~IOP_XATTR; |
4df27c4d YZ |
5740 | inode->i_fop = &simple_dir_operations; |
5741 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
c2050a45 | 5742 | inode->i_mtime = current_time(inode); |
9cc97d64 | 5743 | inode->i_atime = inode->i_mtime; |
5744 | inode->i_ctime = inode->i_mtime; | |
5745 | BTRFS_I(inode)->i_otime = inode->i_mtime; | |
4df27c4d YZ |
5746 | |
5747 | return inode; | |
5748 | } | |
5749 | ||
3de4586c | 5750 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 5751 | { |
0b246afa | 5752 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
d397712b | 5753 | struct inode *inode; |
4df27c4d | 5754 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
5755 | struct btrfs_root *sub_root = root; |
5756 | struct btrfs_key location; | |
76dda93c | 5757 | int index; |
b4aff1f8 | 5758 | int ret = 0; |
39279cc3 CM |
5759 | |
5760 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
5761 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 5762 | |
39e3c955 | 5763 | ret = btrfs_inode_by_name(dir, dentry, &location); |
39279cc3 CM |
5764 | if (ret < 0) |
5765 | return ERR_PTR(ret); | |
5f39d397 | 5766 | |
4df27c4d | 5767 | if (location.objectid == 0) |
5662344b | 5768 | return ERR_PTR(-ENOENT); |
4df27c4d YZ |
5769 | |
5770 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 5771 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
5772 | return inode; |
5773 | } | |
5774 | ||
0b246afa | 5775 | index = srcu_read_lock(&fs_info->subvol_srcu); |
2ff7e61e | 5776 | ret = fixup_tree_root_location(fs_info, dir, dentry, |
4df27c4d YZ |
5777 | &location, &sub_root); |
5778 | if (ret < 0) { | |
5779 | if (ret != -ENOENT) | |
5780 | inode = ERR_PTR(ret); | |
5781 | else | |
5782 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
5783 | } else { | |
73f73415 | 5784 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 5785 | } |
0b246afa | 5786 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
76dda93c | 5787 | |
34d19bad | 5788 | if (!IS_ERR(inode) && root != sub_root) { |
0b246afa | 5789 | down_read(&fs_info->cleanup_work_sem); |
bc98a42c | 5790 | if (!sb_rdonly(inode->i_sb)) |
66b4ffd1 | 5791 | ret = btrfs_orphan_cleanup(sub_root); |
0b246afa | 5792 | up_read(&fs_info->cleanup_work_sem); |
01cd3367 JB |
5793 | if (ret) { |
5794 | iput(inode); | |
66b4ffd1 | 5795 | inode = ERR_PTR(ret); |
01cd3367 | 5796 | } |
c71bf099 YZ |
5797 | } |
5798 | ||
3de4586c CM |
5799 | return inode; |
5800 | } | |
5801 | ||
fe15ce44 | 5802 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
5803 | { |
5804 | struct btrfs_root *root; | |
2b0143b5 | 5805 | struct inode *inode = d_inode(dentry); |
76dda93c | 5806 | |
848cce0d | 5807 | if (!inode && !IS_ROOT(dentry)) |
2b0143b5 | 5808 | inode = d_inode(dentry->d_parent); |
76dda93c | 5809 | |
848cce0d LZ |
5810 | if (inode) { |
5811 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
5812 | if (btrfs_root_refs(&root->root_item) == 0) |
5813 | return 1; | |
848cce0d | 5814 | |
4a0cc7ca | 5815 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
848cce0d | 5816 | return 1; |
efefb143 | 5817 | } |
76dda93c YZ |
5818 | return 0; |
5819 | } | |
5820 | ||
b4aff1f8 JB |
5821 | static void btrfs_dentry_release(struct dentry *dentry) |
5822 | { | |
944a4515 | 5823 | kfree(dentry->d_fsdata); |
b4aff1f8 JB |
5824 | } |
5825 | ||
3de4586c | 5826 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 5827 | unsigned int flags) |
3de4586c | 5828 | { |
5662344b | 5829 | struct inode *inode; |
a66e7cc6 | 5830 | |
5662344b TI |
5831 | inode = btrfs_lookup_dentry(dir, dentry); |
5832 | if (IS_ERR(inode)) { | |
5833 | if (PTR_ERR(inode) == -ENOENT) | |
5834 | inode = NULL; | |
5835 | else | |
5836 | return ERR_CAST(inode); | |
5837 | } | |
5838 | ||
41d28bca | 5839 | return d_splice_alias(inode, dentry); |
39279cc3 CM |
5840 | } |
5841 | ||
16cdcec7 | 5842 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
5843 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
5844 | }; | |
5845 | ||
23b5ec74 JB |
5846 | /* |
5847 | * All this infrastructure exists because dir_emit can fault, and we are holding | |
5848 | * the tree lock when doing readdir. For now just allocate a buffer and copy | |
5849 | * our information into that, and then dir_emit from the buffer. This is | |
5850 | * similar to what NFS does, only we don't keep the buffer around in pagecache | |
5851 | * because I'm afraid I'll mess that up. Long term we need to make filldir do | |
5852 | * copy_to_user_inatomic so we don't have to worry about page faulting under the | |
5853 | * tree lock. | |
5854 | */ | |
5855 | static int btrfs_opendir(struct inode *inode, struct file *file) | |
5856 | { | |
5857 | struct btrfs_file_private *private; | |
5858 | ||
5859 | private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL); | |
5860 | if (!private) | |
5861 | return -ENOMEM; | |
5862 | private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL); | |
5863 | if (!private->filldir_buf) { | |
5864 | kfree(private); | |
5865 | return -ENOMEM; | |
5866 | } | |
5867 | file->private_data = private; | |
5868 | return 0; | |
5869 | } | |
5870 | ||
5871 | struct dir_entry { | |
5872 | u64 ino; | |
5873 | u64 offset; | |
5874 | unsigned type; | |
5875 | int name_len; | |
5876 | }; | |
5877 | ||
5878 | static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx) | |
5879 | { | |
5880 | while (entries--) { | |
5881 | struct dir_entry *entry = addr; | |
5882 | char *name = (char *)(entry + 1); | |
5883 | ||
5884 | ctx->pos = entry->offset; | |
5885 | if (!dir_emit(ctx, name, entry->name_len, entry->ino, | |
5886 | entry->type)) | |
5887 | return 1; | |
5888 | addr += sizeof(struct dir_entry) + entry->name_len; | |
5889 | ctx->pos++; | |
5890 | } | |
5891 | return 0; | |
5892 | } | |
5893 | ||
9cdda8d3 | 5894 | static int btrfs_real_readdir(struct file *file, struct dir_context *ctx) |
39279cc3 | 5895 | { |
9cdda8d3 | 5896 | struct inode *inode = file_inode(file); |
2ff7e61e | 5897 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 5898 | struct btrfs_root *root = BTRFS_I(inode)->root; |
23b5ec74 | 5899 | struct btrfs_file_private *private = file->private_data; |
39279cc3 CM |
5900 | struct btrfs_dir_item *di; |
5901 | struct btrfs_key key; | |
5f39d397 | 5902 | struct btrfs_key found_key; |
39279cc3 | 5903 | struct btrfs_path *path; |
23b5ec74 | 5904 | void *addr; |
16cdcec7 MX |
5905 | struct list_head ins_list; |
5906 | struct list_head del_list; | |
39279cc3 | 5907 | int ret; |
5f39d397 | 5908 | struct extent_buffer *leaf; |
39279cc3 | 5909 | int slot; |
5f39d397 CM |
5910 | char *name_ptr; |
5911 | int name_len; | |
23b5ec74 JB |
5912 | int entries = 0; |
5913 | int total_len = 0; | |
02dbfc99 | 5914 | bool put = false; |
c2951f32 | 5915 | struct btrfs_key location; |
5f39d397 | 5916 | |
9cdda8d3 AV |
5917 | if (!dir_emit_dots(file, ctx)) |
5918 | return 0; | |
5919 | ||
49593bfa | 5920 | path = btrfs_alloc_path(); |
16cdcec7 MX |
5921 | if (!path) |
5922 | return -ENOMEM; | |
ff5714cc | 5923 | |
23b5ec74 | 5924 | addr = private->filldir_buf; |
e4058b54 | 5925 | path->reada = READA_FORWARD; |
49593bfa | 5926 | |
c2951f32 JM |
5927 | INIT_LIST_HEAD(&ins_list); |
5928 | INIT_LIST_HEAD(&del_list); | |
5929 | put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list); | |
16cdcec7 | 5930 | |
23b5ec74 | 5931 | again: |
c2951f32 | 5932 | key.type = BTRFS_DIR_INDEX_KEY; |
9cdda8d3 | 5933 | key.offset = ctx->pos; |
4a0cc7ca | 5934 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
5f39d397 | 5935 | |
39279cc3 CM |
5936 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5937 | if (ret < 0) | |
5938 | goto err; | |
49593bfa DW |
5939 | |
5940 | while (1) { | |
23b5ec74 JB |
5941 | struct dir_entry *entry; |
5942 | ||
5f39d397 | 5943 | leaf = path->nodes[0]; |
39279cc3 | 5944 | slot = path->slots[0]; |
b9e03af0 LZ |
5945 | if (slot >= btrfs_header_nritems(leaf)) { |
5946 | ret = btrfs_next_leaf(root, path); | |
5947 | if (ret < 0) | |
5948 | goto err; | |
5949 | else if (ret > 0) | |
5950 | break; | |
5951 | continue; | |
39279cc3 | 5952 | } |
3de4586c | 5953 | |
5f39d397 CM |
5954 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
5955 | ||
5956 | if (found_key.objectid != key.objectid) | |
39279cc3 | 5957 | break; |
c2951f32 | 5958 | if (found_key.type != BTRFS_DIR_INDEX_KEY) |
39279cc3 | 5959 | break; |
9cdda8d3 | 5960 | if (found_key.offset < ctx->pos) |
b9e03af0 | 5961 | goto next; |
c2951f32 | 5962 | if (btrfs_should_delete_dir_index(&del_list, found_key.offset)) |
16cdcec7 | 5963 | goto next; |
39279cc3 | 5964 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
e79a3327 | 5965 | if (verify_dir_item(fs_info, leaf, slot, di)) |
c2951f32 | 5966 | goto next; |
22a94d44 | 5967 | |
c2951f32 | 5968 | name_len = btrfs_dir_name_len(leaf, di); |
23b5ec74 JB |
5969 | if ((total_len + sizeof(struct dir_entry) + name_len) >= |
5970 | PAGE_SIZE) { | |
5971 | btrfs_release_path(path); | |
5972 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
5973 | if (ret) | |
5974 | goto nopos; | |
5975 | addr = private->filldir_buf; | |
5976 | entries = 0; | |
5977 | total_len = 0; | |
5978 | goto again; | |
c2951f32 | 5979 | } |
23b5ec74 JB |
5980 | |
5981 | entry = addr; | |
5982 | entry->name_len = name_len; | |
5983 | name_ptr = (char *)(entry + 1); | |
c2951f32 JM |
5984 | read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1), |
5985 | name_len); | |
23b5ec74 | 5986 | entry->type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; |
c2951f32 | 5987 | btrfs_dir_item_key_to_cpu(leaf, di, &location); |
23b5ec74 JB |
5988 | entry->ino = location.objectid; |
5989 | entry->offset = found_key.offset; | |
5990 | entries++; | |
5991 | addr += sizeof(struct dir_entry) + name_len; | |
5992 | total_len += sizeof(struct dir_entry) + name_len; | |
b9e03af0 LZ |
5993 | next: |
5994 | path->slots[0]++; | |
39279cc3 | 5995 | } |
23b5ec74 JB |
5996 | btrfs_release_path(path); |
5997 | ||
5998 | ret = btrfs_filldir(private->filldir_buf, entries, ctx); | |
5999 | if (ret) | |
6000 | goto nopos; | |
49593bfa | 6001 | |
d2fbb2b5 | 6002 | ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list); |
c2951f32 | 6003 | if (ret) |
bc4ef759 DS |
6004 | goto nopos; |
6005 | ||
db62efbb ZB |
6006 | /* |
6007 | * Stop new entries from being returned after we return the last | |
6008 | * entry. | |
6009 | * | |
6010 | * New directory entries are assigned a strictly increasing | |
6011 | * offset. This means that new entries created during readdir | |
6012 | * are *guaranteed* to be seen in the future by that readdir. | |
6013 | * This has broken buggy programs which operate on names as | |
6014 | * they're returned by readdir. Until we re-use freed offsets | |
6015 | * we have this hack to stop new entries from being returned | |
6016 | * under the assumption that they'll never reach this huge | |
6017 | * offset. | |
6018 | * | |
6019 | * This is being careful not to overflow 32bit loff_t unless the | |
6020 | * last entry requires it because doing so has broken 32bit apps | |
6021 | * in the past. | |
6022 | */ | |
c2951f32 JM |
6023 | if (ctx->pos >= INT_MAX) |
6024 | ctx->pos = LLONG_MAX; | |
6025 | else | |
6026 | ctx->pos = INT_MAX; | |
39279cc3 CM |
6027 | nopos: |
6028 | ret = 0; | |
6029 | err: | |
02dbfc99 OS |
6030 | if (put) |
6031 | btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list); | |
39279cc3 | 6032 | btrfs_free_path(path); |
39279cc3 CM |
6033 | return ret; |
6034 | } | |
6035 | ||
a9185b41 | 6036 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
6037 | { |
6038 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6039 | struct btrfs_trans_handle *trans; | |
6040 | int ret = 0; | |
0af3d00b | 6041 | bool nolock = false; |
39279cc3 | 6042 | |
72ac3c0d | 6043 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
4ca8b41e CM |
6044 | return 0; |
6045 | ||
70ddc553 NB |
6046 | if (btrfs_fs_closing(root->fs_info) && |
6047 | btrfs_is_free_space_inode(BTRFS_I(inode))) | |
82d5902d | 6048 | nolock = true; |
0af3d00b | 6049 | |
a9185b41 | 6050 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b | 6051 | if (nolock) |
7a7eaa40 | 6052 | trans = btrfs_join_transaction_nolock(root); |
0af3d00b | 6053 | else |
7a7eaa40 | 6054 | trans = btrfs_join_transaction(root); |
3612b495 TI |
6055 | if (IS_ERR(trans)) |
6056 | return PTR_ERR(trans); | |
3a45bb20 | 6057 | ret = btrfs_commit_transaction(trans); |
39279cc3 CM |
6058 | } |
6059 | return ret; | |
6060 | } | |
6061 | ||
6062 | /* | |
54aa1f4d | 6063 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
6064 | * inode changes. But, it is most likely to find the inode in cache. |
6065 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
6066 | * to keep or drop this code. | |
6067 | */ | |
48a3b636 | 6068 | static int btrfs_dirty_inode(struct inode *inode) |
39279cc3 | 6069 | { |
2ff7e61e | 6070 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
6071 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6072 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
6073 | int ret; |
6074 | ||
72ac3c0d | 6075 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 6076 | return 0; |
39279cc3 | 6077 | |
7a7eaa40 | 6078 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
6079 | if (IS_ERR(trans)) |
6080 | return PTR_ERR(trans); | |
8929ecfa YZ |
6081 | |
6082 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
6083 | if (ret && ret == -ENOSPC) { |
6084 | /* whoops, lets try again with the full transaction */ | |
3a45bb20 | 6085 | btrfs_end_transaction(trans); |
94b60442 | 6086 | trans = btrfs_start_transaction(root, 1); |
22c44fe6 JB |
6087 | if (IS_ERR(trans)) |
6088 | return PTR_ERR(trans); | |
8929ecfa | 6089 | |
94b60442 | 6090 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 6091 | } |
3a45bb20 | 6092 | btrfs_end_transaction(trans); |
16cdcec7 | 6093 | if (BTRFS_I(inode)->delayed_node) |
2ff7e61e | 6094 | btrfs_balance_delayed_items(fs_info); |
22c44fe6 JB |
6095 | |
6096 | return ret; | |
6097 | } | |
6098 | ||
6099 | /* | |
6100 | * This is a copy of file_update_time. We need this so we can return error on | |
6101 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
6102 | */ | |
e41f941a JB |
6103 | static int btrfs_update_time(struct inode *inode, struct timespec *now, |
6104 | int flags) | |
22c44fe6 | 6105 | { |
2bc55652 AB |
6106 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6107 | ||
6108 | if (btrfs_root_readonly(root)) | |
6109 | return -EROFS; | |
6110 | ||
e41f941a | 6111 | if (flags & S_VERSION) |
22c44fe6 | 6112 | inode_inc_iversion(inode); |
e41f941a JB |
6113 | if (flags & S_CTIME) |
6114 | inode->i_ctime = *now; | |
6115 | if (flags & S_MTIME) | |
6116 | inode->i_mtime = *now; | |
6117 | if (flags & S_ATIME) | |
6118 | inode->i_atime = *now; | |
6119 | return btrfs_dirty_inode(inode); | |
39279cc3 CM |
6120 | } |
6121 | ||
d352ac68 CM |
6122 | /* |
6123 | * find the highest existing sequence number in a directory | |
6124 | * and then set the in-memory index_cnt variable to reflect | |
6125 | * free sequence numbers | |
6126 | */ | |
4c570655 | 6127 | static int btrfs_set_inode_index_count(struct btrfs_inode *inode) |
aec7477b | 6128 | { |
4c570655 | 6129 | struct btrfs_root *root = inode->root; |
aec7477b JB |
6130 | struct btrfs_key key, found_key; |
6131 | struct btrfs_path *path; | |
6132 | struct extent_buffer *leaf; | |
6133 | int ret; | |
6134 | ||
4c570655 | 6135 | key.objectid = btrfs_ino(inode); |
962a298f | 6136 | key.type = BTRFS_DIR_INDEX_KEY; |
aec7477b JB |
6137 | key.offset = (u64)-1; |
6138 | ||
6139 | path = btrfs_alloc_path(); | |
6140 | if (!path) | |
6141 | return -ENOMEM; | |
6142 | ||
6143 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
6144 | if (ret < 0) | |
6145 | goto out; | |
6146 | /* FIXME: we should be able to handle this */ | |
6147 | if (ret == 0) | |
6148 | goto out; | |
6149 | ret = 0; | |
6150 | ||
6151 | /* | |
6152 | * MAGIC NUMBER EXPLANATION: | |
6153 | * since we search a directory based on f_pos we have to start at 2 | |
6154 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
6155 | * else has to start at 2 | |
6156 | */ | |
6157 | if (path->slots[0] == 0) { | |
4c570655 | 6158 | inode->index_cnt = 2; |
aec7477b JB |
6159 | goto out; |
6160 | } | |
6161 | ||
6162 | path->slots[0]--; | |
6163 | ||
6164 | leaf = path->nodes[0]; | |
6165 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
6166 | ||
4c570655 | 6167 | if (found_key.objectid != btrfs_ino(inode) || |
962a298f | 6168 | found_key.type != BTRFS_DIR_INDEX_KEY) { |
4c570655 | 6169 | inode->index_cnt = 2; |
aec7477b JB |
6170 | goto out; |
6171 | } | |
6172 | ||
4c570655 | 6173 | inode->index_cnt = found_key.offset + 1; |
aec7477b JB |
6174 | out: |
6175 | btrfs_free_path(path); | |
6176 | return ret; | |
6177 | } | |
6178 | ||
d352ac68 CM |
6179 | /* |
6180 | * helper to find a free sequence number in a given directory. This current | |
6181 | * code is very simple, later versions will do smarter things in the btree | |
6182 | */ | |
877574e2 | 6183 | int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index) |
aec7477b JB |
6184 | { |
6185 | int ret = 0; | |
6186 | ||
877574e2 NB |
6187 | if (dir->index_cnt == (u64)-1) { |
6188 | ret = btrfs_inode_delayed_dir_index_count(dir); | |
16cdcec7 MX |
6189 | if (ret) { |
6190 | ret = btrfs_set_inode_index_count(dir); | |
6191 | if (ret) | |
6192 | return ret; | |
6193 | } | |
aec7477b JB |
6194 | } |
6195 | ||
877574e2 NB |
6196 | *index = dir->index_cnt; |
6197 | dir->index_cnt++; | |
aec7477b JB |
6198 | |
6199 | return ret; | |
6200 | } | |
6201 | ||
b0d5d10f CM |
6202 | static int btrfs_insert_inode_locked(struct inode *inode) |
6203 | { | |
6204 | struct btrfs_iget_args args; | |
6205 | args.location = &BTRFS_I(inode)->location; | |
6206 | args.root = BTRFS_I(inode)->root; | |
6207 | ||
6208 | return insert_inode_locked4(inode, | |
6209 | btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root), | |
6210 | btrfs_find_actor, &args); | |
6211 | } | |
6212 | ||
19aee8de AJ |
6213 | /* |
6214 | * Inherit flags from the parent inode. | |
6215 | * | |
6216 | * Currently only the compression flags and the cow flags are inherited. | |
6217 | */ | |
6218 | static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir) | |
6219 | { | |
6220 | unsigned int flags; | |
6221 | ||
6222 | if (!dir) | |
6223 | return; | |
6224 | ||
6225 | flags = BTRFS_I(dir)->flags; | |
6226 | ||
6227 | if (flags & BTRFS_INODE_NOCOMPRESS) { | |
6228 | BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS; | |
6229 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; | |
6230 | } else if (flags & BTRFS_INODE_COMPRESS) { | |
6231 | BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS; | |
6232 | BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS; | |
6233 | } | |
6234 | ||
6235 | if (flags & BTRFS_INODE_NODATACOW) { | |
6236 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; | |
6237 | if (S_ISREG(inode->i_mode)) | |
6238 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
6239 | } | |
6240 | ||
6241 | btrfs_update_iflags(inode); | |
6242 | } | |
6243 | ||
39279cc3 CM |
6244 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
6245 | struct btrfs_root *root, | |
aec7477b | 6246 | struct inode *dir, |
9c58309d | 6247 | const char *name, int name_len, |
175a4eb7 AV |
6248 | u64 ref_objectid, u64 objectid, |
6249 | umode_t mode, u64 *index) | |
39279cc3 | 6250 | { |
0b246afa | 6251 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 6252 | struct inode *inode; |
5f39d397 | 6253 | struct btrfs_inode_item *inode_item; |
39279cc3 | 6254 | struct btrfs_key *location; |
5f39d397 | 6255 | struct btrfs_path *path; |
9c58309d CM |
6256 | struct btrfs_inode_ref *ref; |
6257 | struct btrfs_key key[2]; | |
6258 | u32 sizes[2]; | |
ef3b9af5 | 6259 | int nitems = name ? 2 : 1; |
9c58309d | 6260 | unsigned long ptr; |
39279cc3 | 6261 | int ret; |
39279cc3 | 6262 | |
5f39d397 | 6263 | path = btrfs_alloc_path(); |
d8926bb3 MF |
6264 | if (!path) |
6265 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 6266 | |
0b246afa | 6267 | inode = new_inode(fs_info->sb); |
8fb27640 YS |
6268 | if (!inode) { |
6269 | btrfs_free_path(path); | |
39279cc3 | 6270 | return ERR_PTR(-ENOMEM); |
8fb27640 | 6271 | } |
39279cc3 | 6272 | |
5762b5c9 FM |
6273 | /* |
6274 | * O_TMPFILE, set link count to 0, so that after this point, | |
6275 | * we fill in an inode item with the correct link count. | |
6276 | */ | |
6277 | if (!name) | |
6278 | set_nlink(inode, 0); | |
6279 | ||
581bb050 LZ |
6280 | /* |
6281 | * we have to initialize this early, so we can reclaim the inode | |
6282 | * number if we fail afterwards in this function. | |
6283 | */ | |
6284 | inode->i_ino = objectid; | |
6285 | ||
ef3b9af5 | 6286 | if (dir && name) { |
1abe9b8a | 6287 | trace_btrfs_inode_request(dir); |
6288 | ||
877574e2 | 6289 | ret = btrfs_set_inode_index(BTRFS_I(dir), index); |
09771430 | 6290 | if (ret) { |
8fb27640 | 6291 | btrfs_free_path(path); |
09771430 | 6292 | iput(inode); |
aec7477b | 6293 | return ERR_PTR(ret); |
09771430 | 6294 | } |
ef3b9af5 FM |
6295 | } else if (dir) { |
6296 | *index = 0; | |
aec7477b JB |
6297 | } |
6298 | /* | |
6299 | * index_cnt is ignored for everything but a dir, | |
6300 | * btrfs_get_inode_index_count has an explanation for the magic | |
6301 | * number | |
6302 | */ | |
6303 | BTRFS_I(inode)->index_cnt = 2; | |
67de1176 | 6304 | BTRFS_I(inode)->dir_index = *index; |
39279cc3 | 6305 | BTRFS_I(inode)->root = root; |
e02119d5 | 6306 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 6307 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 6308 | |
5dc562c5 JB |
6309 | /* |
6310 | * We could have gotten an inode number from somebody who was fsynced | |
6311 | * and then removed in this same transaction, so let's just set full | |
6312 | * sync since it will be a full sync anyway and this will blow away the | |
6313 | * old info in the log. | |
6314 | */ | |
6315 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
6316 | ||
9c58309d | 6317 | key[0].objectid = objectid; |
962a298f | 6318 | key[0].type = BTRFS_INODE_ITEM_KEY; |
9c58309d CM |
6319 | key[0].offset = 0; |
6320 | ||
9c58309d | 6321 | sizes[0] = sizeof(struct btrfs_inode_item); |
ef3b9af5 FM |
6322 | |
6323 | if (name) { | |
6324 | /* | |
6325 | * Start new inodes with an inode_ref. This is slightly more | |
6326 | * efficient for small numbers of hard links since they will | |
6327 | * be packed into one item. Extended refs will kick in if we | |
6328 | * add more hard links than can fit in the ref item. | |
6329 | */ | |
6330 | key[1].objectid = objectid; | |
962a298f | 6331 | key[1].type = BTRFS_INODE_REF_KEY; |
ef3b9af5 FM |
6332 | key[1].offset = ref_objectid; |
6333 | ||
6334 | sizes[1] = name_len + sizeof(*ref); | |
6335 | } | |
9c58309d | 6336 | |
b0d5d10f CM |
6337 | location = &BTRFS_I(inode)->location; |
6338 | location->objectid = objectid; | |
6339 | location->offset = 0; | |
962a298f | 6340 | location->type = BTRFS_INODE_ITEM_KEY; |
b0d5d10f CM |
6341 | |
6342 | ret = btrfs_insert_inode_locked(inode); | |
6343 | if (ret < 0) | |
6344 | goto fail; | |
6345 | ||
b9473439 | 6346 | path->leave_spinning = 1; |
ef3b9af5 | 6347 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems); |
9c58309d | 6348 | if (ret != 0) |
b0d5d10f | 6349 | goto fail_unlock; |
5f39d397 | 6350 | |
ecc11fab | 6351 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 6352 | inode_set_bytes(inode, 0); |
9cc97d64 | 6353 | |
c2050a45 | 6354 | inode->i_mtime = current_time(inode); |
9cc97d64 | 6355 | inode->i_atime = inode->i_mtime; |
6356 | inode->i_ctime = inode->i_mtime; | |
6357 | BTRFS_I(inode)->i_otime = inode->i_mtime; | |
6358 | ||
5f39d397 CM |
6359 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
6360 | struct btrfs_inode_item); | |
b159fa28 | 6361 | memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item, |
293f7e07 | 6362 | sizeof(*inode_item)); |
e02119d5 | 6363 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d | 6364 | |
ef3b9af5 FM |
6365 | if (name) { |
6366 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
6367 | struct btrfs_inode_ref); | |
6368 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
6369 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); | |
6370 | ptr = (unsigned long)(ref + 1); | |
6371 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
6372 | } | |
9c58309d | 6373 | |
5f39d397 CM |
6374 | btrfs_mark_buffer_dirty(path->nodes[0]); |
6375 | btrfs_free_path(path); | |
6376 | ||
6cbff00f CH |
6377 | btrfs_inherit_iflags(inode, dir); |
6378 | ||
569254b0 | 6379 | if (S_ISREG(mode)) { |
0b246afa | 6380 | if (btrfs_test_opt(fs_info, NODATASUM)) |
94272164 | 6381 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; |
0b246afa | 6382 | if (btrfs_test_opt(fs_info, NODATACOW)) |
f2bdf9a8 JB |
6383 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW | |
6384 | BTRFS_INODE_NODATASUM; | |
94272164 CM |
6385 | } |
6386 | ||
5d4f98a2 | 6387 | inode_tree_add(inode); |
1abe9b8a | 6388 | |
6389 | trace_btrfs_inode_new(inode); | |
1973f0fa | 6390 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 6391 | |
8ea05e3a AB |
6392 | btrfs_update_root_times(trans, root); |
6393 | ||
63541927 FDBM |
6394 | ret = btrfs_inode_inherit_props(trans, inode, dir); |
6395 | if (ret) | |
0b246afa | 6396 | btrfs_err(fs_info, |
63541927 | 6397 | "error inheriting props for ino %llu (root %llu): %d", |
f85b7379 | 6398 | btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret); |
63541927 | 6399 | |
39279cc3 | 6400 | return inode; |
b0d5d10f CM |
6401 | |
6402 | fail_unlock: | |
6403 | unlock_new_inode(inode); | |
5f39d397 | 6404 | fail: |
ef3b9af5 | 6405 | if (dir && name) |
aec7477b | 6406 | BTRFS_I(dir)->index_cnt--; |
5f39d397 | 6407 | btrfs_free_path(path); |
09771430 | 6408 | iput(inode); |
5f39d397 | 6409 | return ERR_PTR(ret); |
39279cc3 CM |
6410 | } |
6411 | ||
6412 | static inline u8 btrfs_inode_type(struct inode *inode) | |
6413 | { | |
6414 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
6415 | } | |
6416 | ||
d352ac68 CM |
6417 | /* |
6418 | * utility function to add 'inode' into 'parent_inode' with | |
6419 | * a give name and a given sequence number. | |
6420 | * if 'add_backref' is true, also insert a backref from the | |
6421 | * inode to the parent directory. | |
6422 | */ | |
e02119d5 | 6423 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
db0a669f | 6424 | struct btrfs_inode *parent_inode, struct btrfs_inode *inode, |
e02119d5 | 6425 | const char *name, int name_len, int add_backref, u64 index) |
39279cc3 | 6426 | { |
db0a669f | 6427 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
4df27c4d | 6428 | int ret = 0; |
39279cc3 | 6429 | struct btrfs_key key; |
db0a669f NB |
6430 | struct btrfs_root *root = parent_inode->root; |
6431 | u64 ino = btrfs_ino(inode); | |
6432 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 6433 | |
33345d01 | 6434 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
db0a669f | 6435 | memcpy(&key, &inode->root->root_key, sizeof(key)); |
4df27c4d | 6436 | } else { |
33345d01 | 6437 | key.objectid = ino; |
962a298f | 6438 | key.type = BTRFS_INODE_ITEM_KEY; |
4df27c4d YZ |
6439 | key.offset = 0; |
6440 | } | |
6441 | ||
33345d01 | 6442 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
0b246afa JM |
6443 | ret = btrfs_add_root_ref(trans, fs_info, key.objectid, |
6444 | root->root_key.objectid, parent_ino, | |
6445 | index, name, name_len); | |
4df27c4d | 6446 | } else if (add_backref) { |
33345d01 LZ |
6447 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
6448 | parent_ino, index); | |
4df27c4d | 6449 | } |
39279cc3 | 6450 | |
79787eaa JM |
6451 | /* Nothing to clean up yet */ |
6452 | if (ret) | |
6453 | return ret; | |
4df27c4d | 6454 | |
79787eaa JM |
6455 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
6456 | parent_inode, &key, | |
db0a669f | 6457 | btrfs_inode_type(&inode->vfs_inode), index); |
9c52057c | 6458 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
6459 | goto fail_dir_item; |
6460 | else if (ret) { | |
66642832 | 6461 | btrfs_abort_transaction(trans, ret); |
79787eaa | 6462 | return ret; |
39279cc3 | 6463 | } |
79787eaa | 6464 | |
db0a669f | 6465 | btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size + |
79787eaa | 6466 | name_len * 2); |
db0a669f NB |
6467 | inode_inc_iversion(&parent_inode->vfs_inode); |
6468 | parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime = | |
6469 | current_time(&parent_inode->vfs_inode); | |
6470 | ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode); | |
79787eaa | 6471 | if (ret) |
66642832 | 6472 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 6473 | return ret; |
fe66a05a CM |
6474 | |
6475 | fail_dir_item: | |
6476 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
6477 | u64 local_index; | |
6478 | int err; | |
0b246afa JM |
6479 | err = btrfs_del_root_ref(trans, fs_info, key.objectid, |
6480 | root->root_key.objectid, parent_ino, | |
6481 | &local_index, name, name_len); | |
fe66a05a CM |
6482 | |
6483 | } else if (add_backref) { | |
6484 | u64 local_index; | |
6485 | int err; | |
6486 | ||
6487 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
6488 | ino, parent_ino, &local_index); | |
6489 | } | |
6490 | return ret; | |
39279cc3 CM |
6491 | } |
6492 | ||
6493 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
cef415af NB |
6494 | struct btrfs_inode *dir, struct dentry *dentry, |
6495 | struct btrfs_inode *inode, int backref, u64 index) | |
39279cc3 | 6496 | { |
a1b075d2 JB |
6497 | int err = btrfs_add_link(trans, dir, inode, |
6498 | dentry->d_name.name, dentry->d_name.len, | |
6499 | backref, index); | |
39279cc3 CM |
6500 | if (err > 0) |
6501 | err = -EEXIST; | |
6502 | return err; | |
6503 | } | |
6504 | ||
618e21d5 | 6505 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 6506 | umode_t mode, dev_t rdev) |
618e21d5 | 6507 | { |
2ff7e61e | 6508 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
618e21d5 JB |
6509 | struct btrfs_trans_handle *trans; |
6510 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6511 | struct inode *inode = NULL; |
618e21d5 JB |
6512 | int err; |
6513 | int drop_inode = 0; | |
6514 | u64 objectid; | |
00e4e6b3 | 6515 | u64 index = 0; |
618e21d5 | 6516 | |
9ed74f2d JB |
6517 | /* |
6518 | * 2 for inode item and ref | |
6519 | * 2 for dir items | |
6520 | * 1 for xattr if selinux is on | |
6521 | */ | |
a22285a6 YZ |
6522 | trans = btrfs_start_transaction(root, 5); |
6523 | if (IS_ERR(trans)) | |
6524 | return PTR_ERR(trans); | |
1832a6d5 | 6525 | |
581bb050 LZ |
6526 | err = btrfs_find_free_ino(root, &objectid); |
6527 | if (err) | |
6528 | goto out_unlock; | |
6529 | ||
aec7477b | 6530 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6531 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6532 | mode, &index); | |
7cf96da3 TI |
6533 | if (IS_ERR(inode)) { |
6534 | err = PTR_ERR(inode); | |
618e21d5 | 6535 | goto out_unlock; |
7cf96da3 | 6536 | } |
618e21d5 | 6537 | |
ad19db71 CS |
6538 | /* |
6539 | * If the active LSM wants to access the inode during | |
6540 | * d_instantiate it needs these. Smack checks to see | |
6541 | * if the filesystem supports xattrs by looking at the | |
6542 | * ops vector. | |
6543 | */ | |
ad19db71 | 6544 | inode->i_op = &btrfs_special_inode_operations; |
b0d5d10f CM |
6545 | init_special_inode(inode, inode->i_mode, rdev); |
6546 | ||
6547 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
618e21d5 | 6548 | if (err) |
b0d5d10f CM |
6549 | goto out_unlock_inode; |
6550 | ||
cef415af NB |
6551 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6552 | 0, index); | |
b0d5d10f CM |
6553 | if (err) { |
6554 | goto out_unlock_inode; | |
6555 | } else { | |
1b4ab1bb | 6556 | btrfs_update_inode(trans, root, inode); |
b0d5d10f | 6557 | unlock_new_inode(inode); |
08c422c2 | 6558 | d_instantiate(dentry, inode); |
618e21d5 | 6559 | } |
b0d5d10f | 6560 | |
618e21d5 | 6561 | out_unlock: |
3a45bb20 | 6562 | btrfs_end_transaction(trans); |
2ff7e61e | 6563 | btrfs_btree_balance_dirty(fs_info); |
618e21d5 JB |
6564 | if (drop_inode) { |
6565 | inode_dec_link_count(inode); | |
6566 | iput(inode); | |
6567 | } | |
618e21d5 | 6568 | return err; |
b0d5d10f CM |
6569 | |
6570 | out_unlock_inode: | |
6571 | drop_inode = 1; | |
6572 | unlock_new_inode(inode); | |
6573 | goto out_unlock; | |
6574 | ||
618e21d5 JB |
6575 | } |
6576 | ||
39279cc3 | 6577 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 6578 | umode_t mode, bool excl) |
39279cc3 | 6579 | { |
2ff7e61e | 6580 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
6581 | struct btrfs_trans_handle *trans; |
6582 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6583 | struct inode *inode = NULL; |
43baa579 | 6584 | int drop_inode_on_err = 0; |
a22285a6 | 6585 | int err; |
39279cc3 | 6586 | u64 objectid; |
00e4e6b3 | 6587 | u64 index = 0; |
39279cc3 | 6588 | |
9ed74f2d JB |
6589 | /* |
6590 | * 2 for inode item and ref | |
6591 | * 2 for dir items | |
6592 | * 1 for xattr if selinux is on | |
6593 | */ | |
a22285a6 YZ |
6594 | trans = btrfs_start_transaction(root, 5); |
6595 | if (IS_ERR(trans)) | |
6596 | return PTR_ERR(trans); | |
9ed74f2d | 6597 | |
581bb050 LZ |
6598 | err = btrfs_find_free_ino(root, &objectid); |
6599 | if (err) | |
6600 | goto out_unlock; | |
6601 | ||
aec7477b | 6602 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6603 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6604 | mode, &index); | |
7cf96da3 TI |
6605 | if (IS_ERR(inode)) { |
6606 | err = PTR_ERR(inode); | |
39279cc3 | 6607 | goto out_unlock; |
7cf96da3 | 6608 | } |
43baa579 | 6609 | drop_inode_on_err = 1; |
ad19db71 CS |
6610 | /* |
6611 | * If the active LSM wants to access the inode during | |
6612 | * d_instantiate it needs these. Smack checks to see | |
6613 | * if the filesystem supports xattrs by looking at the | |
6614 | * ops vector. | |
6615 | */ | |
6616 | inode->i_fop = &btrfs_file_operations; | |
6617 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 6618 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
6619 | |
6620 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
6621 | if (err) | |
6622 | goto out_unlock_inode; | |
6623 | ||
6624 | err = btrfs_update_inode(trans, root, inode); | |
6625 | if (err) | |
6626 | goto out_unlock_inode; | |
ad19db71 | 6627 | |
cef415af NB |
6628 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6629 | 0, index); | |
39279cc3 | 6630 | if (err) |
b0d5d10f | 6631 | goto out_unlock_inode; |
43baa579 | 6632 | |
43baa579 | 6633 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
b0d5d10f | 6634 | unlock_new_inode(inode); |
43baa579 FB |
6635 | d_instantiate(dentry, inode); |
6636 | ||
39279cc3 | 6637 | out_unlock: |
3a45bb20 | 6638 | btrfs_end_transaction(trans); |
43baa579 | 6639 | if (err && drop_inode_on_err) { |
39279cc3 CM |
6640 | inode_dec_link_count(inode); |
6641 | iput(inode); | |
6642 | } | |
2ff7e61e | 6643 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 | 6644 | return err; |
b0d5d10f CM |
6645 | |
6646 | out_unlock_inode: | |
6647 | unlock_new_inode(inode); | |
6648 | goto out_unlock; | |
6649 | ||
39279cc3 CM |
6650 | } |
6651 | ||
6652 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
6653 | struct dentry *dentry) | |
6654 | { | |
271dba45 | 6655 | struct btrfs_trans_handle *trans = NULL; |
39279cc3 | 6656 | struct btrfs_root *root = BTRFS_I(dir)->root; |
2b0143b5 | 6657 | struct inode *inode = d_inode(old_dentry); |
2ff7e61e | 6658 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
00e4e6b3 | 6659 | u64 index; |
39279cc3 CM |
6660 | int err; |
6661 | int drop_inode = 0; | |
6662 | ||
4a8be425 TH |
6663 | /* do not allow sys_link's with other subvols of the same device */ |
6664 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 6665 | return -EXDEV; |
4a8be425 | 6666 | |
f186373f | 6667 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 6668 | return -EMLINK; |
4a8be425 | 6669 | |
877574e2 | 6670 | err = btrfs_set_inode_index(BTRFS_I(dir), &index); |
aec7477b JB |
6671 | if (err) |
6672 | goto fail; | |
6673 | ||
a22285a6 | 6674 | /* |
7e6b6465 | 6675 | * 2 items for inode and inode ref |
a22285a6 | 6676 | * 2 items for dir items |
7e6b6465 | 6677 | * 1 item for parent inode |
a22285a6 | 6678 | */ |
7e6b6465 | 6679 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
6680 | if (IS_ERR(trans)) { |
6681 | err = PTR_ERR(trans); | |
271dba45 | 6682 | trans = NULL; |
a22285a6 YZ |
6683 | goto fail; |
6684 | } | |
5f39d397 | 6685 | |
67de1176 MX |
6686 | /* There are several dir indexes for this inode, clear the cache. */ |
6687 | BTRFS_I(inode)->dir_index = 0ULL; | |
8b558c5f | 6688 | inc_nlink(inode); |
0c4d2d95 | 6689 | inode_inc_iversion(inode); |
c2050a45 | 6690 | inode->i_ctime = current_time(inode); |
7de9c6ee | 6691 | ihold(inode); |
e9976151 | 6692 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 6693 | |
cef415af NB |
6694 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6695 | 1, index); | |
5f39d397 | 6696 | |
a5719521 | 6697 | if (err) { |
54aa1f4d | 6698 | drop_inode = 1; |
a5719521 | 6699 | } else { |
10d9f309 | 6700 | struct dentry *parent = dentry->d_parent; |
a5719521 | 6701 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
6702 | if (err) |
6703 | goto fail; | |
ef3b9af5 FM |
6704 | if (inode->i_nlink == 1) { |
6705 | /* | |
6706 | * If new hard link count is 1, it's a file created | |
6707 | * with open(2) O_TMPFILE flag. | |
6708 | */ | |
3d6ae7bb | 6709 | err = btrfs_orphan_del(trans, BTRFS_I(inode)); |
ef3b9af5 FM |
6710 | if (err) |
6711 | goto fail; | |
6712 | } | |
08c422c2 | 6713 | d_instantiate(dentry, inode); |
9ca5fbfb | 6714 | btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent); |
a5719521 | 6715 | } |
39279cc3 | 6716 | |
1832a6d5 | 6717 | fail: |
271dba45 | 6718 | if (trans) |
3a45bb20 | 6719 | btrfs_end_transaction(trans); |
39279cc3 CM |
6720 | if (drop_inode) { |
6721 | inode_dec_link_count(inode); | |
6722 | iput(inode); | |
6723 | } | |
2ff7e61e | 6724 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6725 | return err; |
6726 | } | |
6727 | ||
18bb1db3 | 6728 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 6729 | { |
2ff7e61e | 6730 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
b9d86667 | 6731 | struct inode *inode = NULL; |
39279cc3 CM |
6732 | struct btrfs_trans_handle *trans; |
6733 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
6734 | int err = 0; | |
6735 | int drop_on_err = 0; | |
b9d86667 | 6736 | u64 objectid = 0; |
00e4e6b3 | 6737 | u64 index = 0; |
39279cc3 | 6738 | |
9ed74f2d JB |
6739 | /* |
6740 | * 2 items for inode and ref | |
6741 | * 2 items for dir items | |
6742 | * 1 for xattr if selinux is on | |
6743 | */ | |
a22285a6 YZ |
6744 | trans = btrfs_start_transaction(root, 5); |
6745 | if (IS_ERR(trans)) | |
6746 | return PTR_ERR(trans); | |
39279cc3 | 6747 | |
581bb050 LZ |
6748 | err = btrfs_find_free_ino(root, &objectid); |
6749 | if (err) | |
6750 | goto out_fail; | |
6751 | ||
aec7477b | 6752 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6753 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6754 | S_IFDIR | mode, &index); | |
39279cc3 CM |
6755 | if (IS_ERR(inode)) { |
6756 | err = PTR_ERR(inode); | |
6757 | goto out_fail; | |
6758 | } | |
5f39d397 | 6759 | |
39279cc3 | 6760 | drop_on_err = 1; |
b0d5d10f CM |
6761 | /* these must be set before we unlock the inode */ |
6762 | inode->i_op = &btrfs_dir_inode_operations; | |
6763 | inode->i_fop = &btrfs_dir_file_operations; | |
33268eaf | 6764 | |
2a7dba39 | 6765 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf | 6766 | if (err) |
b0d5d10f | 6767 | goto out_fail_inode; |
39279cc3 | 6768 | |
6ef06d27 | 6769 | btrfs_i_size_write(BTRFS_I(inode), 0); |
39279cc3 CM |
6770 | err = btrfs_update_inode(trans, root, inode); |
6771 | if (err) | |
b0d5d10f | 6772 | goto out_fail_inode; |
5f39d397 | 6773 | |
db0a669f NB |
6774 | err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), |
6775 | dentry->d_name.name, | |
6776 | dentry->d_name.len, 0, index); | |
39279cc3 | 6777 | if (err) |
b0d5d10f | 6778 | goto out_fail_inode; |
5f39d397 | 6779 | |
39279cc3 | 6780 | d_instantiate(dentry, inode); |
b0d5d10f CM |
6781 | /* |
6782 | * mkdir is special. We're unlocking after we call d_instantiate | |
6783 | * to avoid a race with nfsd calling d_instantiate. | |
6784 | */ | |
6785 | unlock_new_inode(inode); | |
39279cc3 | 6786 | drop_on_err = 0; |
39279cc3 CM |
6787 | |
6788 | out_fail: | |
3a45bb20 | 6789 | btrfs_end_transaction(trans); |
c7cfb8a5 WS |
6790 | if (drop_on_err) { |
6791 | inode_dec_link_count(inode); | |
39279cc3 | 6792 | iput(inode); |
c7cfb8a5 | 6793 | } |
2ff7e61e | 6794 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 | 6795 | return err; |
b0d5d10f CM |
6796 | |
6797 | out_fail_inode: | |
6798 | unlock_new_inode(inode); | |
6799 | goto out_fail; | |
39279cc3 CM |
6800 | } |
6801 | ||
e6c4efd8 QW |
6802 | /* Find next extent map of a given extent map, caller needs to ensure locks */ |
6803 | static struct extent_map *next_extent_map(struct extent_map *em) | |
6804 | { | |
6805 | struct rb_node *next; | |
6806 | ||
6807 | next = rb_next(&em->rb_node); | |
6808 | if (!next) | |
6809 | return NULL; | |
6810 | return container_of(next, struct extent_map, rb_node); | |
6811 | } | |
6812 | ||
6813 | static struct extent_map *prev_extent_map(struct extent_map *em) | |
6814 | { | |
6815 | struct rb_node *prev; | |
6816 | ||
6817 | prev = rb_prev(&em->rb_node); | |
6818 | if (!prev) | |
6819 | return NULL; | |
6820 | return container_of(prev, struct extent_map, rb_node); | |
6821 | } | |
6822 | ||
d352ac68 | 6823 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
e6c4efd8 | 6824 | * the existing extent is the nearest extent to map_start, |
d352ac68 | 6825 | * and an extent that you want to insert, deal with overlap and insert |
e6c4efd8 | 6826 | * the best fitted new extent into the tree. |
d352ac68 | 6827 | */ |
3b951516 CM |
6828 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
6829 | struct extent_map *existing, | |
e6dcd2dc | 6830 | struct extent_map *em, |
51f395ad | 6831 | u64 map_start) |
3b951516 | 6832 | { |
e6c4efd8 QW |
6833 | struct extent_map *prev; |
6834 | struct extent_map *next; | |
6835 | u64 start; | |
6836 | u64 end; | |
3b951516 | 6837 | u64 start_diff; |
3b951516 | 6838 | |
e6dcd2dc | 6839 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
e6c4efd8 QW |
6840 | |
6841 | if (existing->start > map_start) { | |
6842 | next = existing; | |
6843 | prev = prev_extent_map(next); | |
6844 | } else { | |
6845 | prev = existing; | |
6846 | next = next_extent_map(prev); | |
6847 | } | |
6848 | ||
6849 | start = prev ? extent_map_end(prev) : em->start; | |
6850 | start = max_t(u64, start, em->start); | |
6851 | end = next ? next->start : extent_map_end(em); | |
6852 | end = min_t(u64, end, extent_map_end(em)); | |
6853 | start_diff = start - em->start; | |
6854 | em->start = start; | |
6855 | em->len = end - start; | |
c8b97818 CM |
6856 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
6857 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 6858 | em->block_start += start_diff; |
c8b97818 CM |
6859 | em->block_len -= start_diff; |
6860 | } | |
09a2a8f9 | 6861 | return add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
6862 | } |
6863 | ||
c8b97818 | 6864 | static noinline int uncompress_inline(struct btrfs_path *path, |
e40da0e5 | 6865 | struct page *page, |
c8b97818 CM |
6866 | size_t pg_offset, u64 extent_offset, |
6867 | struct btrfs_file_extent_item *item) | |
6868 | { | |
6869 | int ret; | |
6870 | struct extent_buffer *leaf = path->nodes[0]; | |
6871 | char *tmp; | |
6872 | size_t max_size; | |
6873 | unsigned long inline_size; | |
6874 | unsigned long ptr; | |
261507a0 | 6875 | int compress_type; |
c8b97818 CM |
6876 | |
6877 | WARN_ON(pg_offset != 0); | |
261507a0 | 6878 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
6879 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
6880 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
dd3cc16b | 6881 | btrfs_item_nr(path->slots[0])); |
c8b97818 | 6882 | tmp = kmalloc(inline_size, GFP_NOFS); |
8d413713 TI |
6883 | if (!tmp) |
6884 | return -ENOMEM; | |
c8b97818 CM |
6885 | ptr = btrfs_file_extent_inline_start(item); |
6886 | ||
6887 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
6888 | ||
09cbfeaf | 6889 | max_size = min_t(unsigned long, PAGE_SIZE, max_size); |
261507a0 LZ |
6890 | ret = btrfs_decompress(compress_type, tmp, page, |
6891 | extent_offset, inline_size, max_size); | |
e1699d2d ZB |
6892 | |
6893 | /* | |
6894 | * decompression code contains a memset to fill in any space between the end | |
6895 | * of the uncompressed data and the end of max_size in case the decompressed | |
6896 | * data ends up shorter than ram_bytes. That doesn't cover the hole between | |
6897 | * the end of an inline extent and the beginning of the next block, so we | |
6898 | * cover that region here. | |
6899 | */ | |
6900 | ||
6901 | if (max_size + pg_offset < PAGE_SIZE) { | |
6902 | char *map = kmap(page); | |
6903 | memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset); | |
6904 | kunmap(page); | |
6905 | } | |
c8b97818 | 6906 | kfree(tmp); |
166ae5a4 | 6907 | return ret; |
c8b97818 CM |
6908 | } |
6909 | ||
d352ac68 CM |
6910 | /* |
6911 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
6912 | * the ugly parts come from merging extents from the disk with the in-ram |
6913 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
6914 | * where the in-ram extents might be locked pending data=ordered completion. |
6915 | * | |
6916 | * This also copies inline extents directly into the page. | |
6917 | */ | |
fc4f21b1 NB |
6918 | struct extent_map *btrfs_get_extent(struct btrfs_inode *inode, |
6919 | struct page *page, | |
6920 | size_t pg_offset, u64 start, u64 len, | |
6921 | int create) | |
a52d9a80 | 6922 | { |
fc4f21b1 | 6923 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
a52d9a80 CM |
6924 | int ret; |
6925 | int err = 0; | |
a52d9a80 CM |
6926 | u64 extent_start = 0; |
6927 | u64 extent_end = 0; | |
fc4f21b1 | 6928 | u64 objectid = btrfs_ino(inode); |
a52d9a80 | 6929 | u32 found_type; |
f421950f | 6930 | struct btrfs_path *path = NULL; |
fc4f21b1 | 6931 | struct btrfs_root *root = inode->root; |
a52d9a80 | 6932 | struct btrfs_file_extent_item *item; |
5f39d397 CM |
6933 | struct extent_buffer *leaf; |
6934 | struct btrfs_key found_key; | |
a52d9a80 | 6935 | struct extent_map *em = NULL; |
fc4f21b1 NB |
6936 | struct extent_map_tree *em_tree = &inode->extent_tree; |
6937 | struct extent_io_tree *io_tree = &inode->io_tree; | |
a52d9a80 | 6938 | struct btrfs_trans_handle *trans = NULL; |
7ffbb598 | 6939 | const bool new_inline = !page || create; |
a52d9a80 | 6940 | |
a52d9a80 | 6941 | again: |
890871be | 6942 | read_lock(&em_tree->lock); |
d1310b2e | 6943 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d | 6944 | if (em) |
0b246afa | 6945 | em->bdev = fs_info->fs_devices->latest_bdev; |
890871be | 6946 | read_unlock(&em_tree->lock); |
d1310b2e | 6947 | |
a52d9a80 | 6948 | if (em) { |
e1c4b745 CM |
6949 | if (em->start > start || em->start + em->len <= start) |
6950 | free_extent_map(em); | |
6951 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
6952 | free_extent_map(em); |
6953 | else | |
6954 | goto out; | |
a52d9a80 | 6955 | } |
172ddd60 | 6956 | em = alloc_extent_map(); |
a52d9a80 | 6957 | if (!em) { |
d1310b2e CM |
6958 | err = -ENOMEM; |
6959 | goto out; | |
a52d9a80 | 6960 | } |
0b246afa | 6961 | em->bdev = fs_info->fs_devices->latest_bdev; |
d1310b2e | 6962 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 6963 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 6964 | em->len = (u64)-1; |
c8b97818 | 6965 | em->block_len = (u64)-1; |
f421950f CM |
6966 | |
6967 | if (!path) { | |
6968 | path = btrfs_alloc_path(); | |
026fd317 JB |
6969 | if (!path) { |
6970 | err = -ENOMEM; | |
6971 | goto out; | |
6972 | } | |
6973 | /* | |
6974 | * Chances are we'll be called again, so go ahead and do | |
6975 | * readahead | |
6976 | */ | |
e4058b54 | 6977 | path->reada = READA_FORWARD; |
f421950f CM |
6978 | } |
6979 | ||
179e29e4 CM |
6980 | ret = btrfs_lookup_file_extent(trans, root, path, |
6981 | objectid, start, trans != NULL); | |
a52d9a80 CM |
6982 | if (ret < 0) { |
6983 | err = ret; | |
6984 | goto out; | |
6985 | } | |
6986 | ||
6987 | if (ret != 0) { | |
6988 | if (path->slots[0] == 0) | |
6989 | goto not_found; | |
6990 | path->slots[0]--; | |
6991 | } | |
6992 | ||
5f39d397 CM |
6993 | leaf = path->nodes[0]; |
6994 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 6995 | struct btrfs_file_extent_item); |
a52d9a80 | 6996 | /* are we inside the extent that was found? */ |
5f39d397 | 6997 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
962a298f | 6998 | found_type = found_key.type; |
5f39d397 | 6999 | if (found_key.objectid != objectid || |
a52d9a80 | 7000 | found_type != BTRFS_EXTENT_DATA_KEY) { |
25a50341 JB |
7001 | /* |
7002 | * If we backup past the first extent we want to move forward | |
7003 | * and see if there is an extent in front of us, otherwise we'll | |
7004 | * say there is a hole for our whole search range which can | |
7005 | * cause problems. | |
7006 | */ | |
7007 | extent_end = start; | |
7008 | goto next; | |
a52d9a80 CM |
7009 | } |
7010 | ||
5f39d397 CM |
7011 | found_type = btrfs_file_extent_type(leaf, item); |
7012 | extent_start = found_key.offset; | |
d899e052 YZ |
7013 | if (found_type == BTRFS_FILE_EXTENT_REG || |
7014 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 7015 | extent_end = extent_start + |
db94535d | 7016 | btrfs_file_extent_num_bytes(leaf, item); |
09ed2f16 LB |
7017 | |
7018 | trace_btrfs_get_extent_show_fi_regular(inode, leaf, item, | |
7019 | extent_start); | |
9036c102 YZ |
7020 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
7021 | size_t size; | |
514ac8ad | 7022 | size = btrfs_file_extent_inline_len(leaf, path->slots[0], item); |
da17066c | 7023 | extent_end = ALIGN(extent_start + size, |
0b246afa | 7024 | fs_info->sectorsize); |
09ed2f16 LB |
7025 | |
7026 | trace_btrfs_get_extent_show_fi_inline(inode, leaf, item, | |
7027 | path->slots[0], | |
7028 | extent_start); | |
9036c102 | 7029 | } |
25a50341 | 7030 | next: |
9036c102 YZ |
7031 | if (start >= extent_end) { |
7032 | path->slots[0]++; | |
7033 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
7034 | ret = btrfs_next_leaf(root, path); | |
7035 | if (ret < 0) { | |
7036 | err = ret; | |
7037 | goto out; | |
a52d9a80 | 7038 | } |
9036c102 YZ |
7039 | if (ret > 0) |
7040 | goto not_found; | |
7041 | leaf = path->nodes[0]; | |
a52d9a80 | 7042 | } |
9036c102 YZ |
7043 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
7044 | if (found_key.objectid != objectid || | |
7045 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
7046 | goto not_found; | |
7047 | if (start + len <= found_key.offset) | |
7048 | goto not_found; | |
e2eca69d WS |
7049 | if (start > found_key.offset) |
7050 | goto next; | |
9036c102 | 7051 | em->start = start; |
70c8a91c | 7052 | em->orig_start = start; |
9036c102 YZ |
7053 | em->len = found_key.offset - start; |
7054 | goto not_found_em; | |
7055 | } | |
7056 | ||
fc4f21b1 | 7057 | btrfs_extent_item_to_extent_map(inode, path, item, |
9cdc5124 | 7058 | new_inline, em); |
7ffbb598 | 7059 | |
d899e052 YZ |
7060 | if (found_type == BTRFS_FILE_EXTENT_REG || |
7061 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 CM |
7062 | goto insert; |
7063 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 7064 | unsigned long ptr; |
a52d9a80 | 7065 | char *map; |
3326d1b0 CM |
7066 | size_t size; |
7067 | size_t extent_offset; | |
7068 | size_t copy_size; | |
a52d9a80 | 7069 | |
7ffbb598 | 7070 | if (new_inline) |
689f9346 | 7071 | goto out; |
5f39d397 | 7072 | |
514ac8ad | 7073 | size = btrfs_file_extent_inline_len(leaf, path->slots[0], item); |
9036c102 | 7074 | extent_offset = page_offset(page) + pg_offset - extent_start; |
09cbfeaf KS |
7075 | copy_size = min_t(u64, PAGE_SIZE - pg_offset, |
7076 | size - extent_offset); | |
3326d1b0 | 7077 | em->start = extent_start + extent_offset; |
0b246afa | 7078 | em->len = ALIGN(copy_size, fs_info->sectorsize); |
b4939680 | 7079 | em->orig_block_len = em->len; |
70c8a91c | 7080 | em->orig_start = em->start; |
689f9346 | 7081 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 7082 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
7083 | if (btrfs_file_extent_compression(leaf, item) != |
7084 | BTRFS_COMPRESS_NONE) { | |
e40da0e5 | 7085 | ret = uncompress_inline(path, page, pg_offset, |
c8b97818 | 7086 | extent_offset, item); |
166ae5a4 ZB |
7087 | if (ret) { |
7088 | err = ret; | |
7089 | goto out; | |
7090 | } | |
c8b97818 CM |
7091 | } else { |
7092 | map = kmap(page); | |
7093 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
7094 | copy_size); | |
09cbfeaf | 7095 | if (pg_offset + copy_size < PAGE_SIZE) { |
93c82d57 | 7096 | memset(map + pg_offset + copy_size, 0, |
09cbfeaf | 7097 | PAGE_SIZE - pg_offset - |
93c82d57 CM |
7098 | copy_size); |
7099 | } | |
c8b97818 CM |
7100 | kunmap(page); |
7101 | } | |
179e29e4 CM |
7102 | flush_dcache_page(page); |
7103 | } else if (create && PageUptodate(page)) { | |
6bf7e080 | 7104 | BUG(); |
179e29e4 CM |
7105 | if (!trans) { |
7106 | kunmap(page); | |
7107 | free_extent_map(em); | |
7108 | em = NULL; | |
ff5714cc | 7109 | |
b3b4aa74 | 7110 | btrfs_release_path(path); |
7a7eaa40 | 7111 | trans = btrfs_join_transaction(root); |
ff5714cc | 7112 | |
3612b495 TI |
7113 | if (IS_ERR(trans)) |
7114 | return ERR_CAST(trans); | |
179e29e4 CM |
7115 | goto again; |
7116 | } | |
c8b97818 | 7117 | map = kmap(page); |
70dec807 | 7118 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 7119 | copy_size); |
c8b97818 | 7120 | kunmap(page); |
179e29e4 | 7121 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 7122 | } |
d1310b2e | 7123 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 7124 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 | 7125 | goto insert; |
a52d9a80 CM |
7126 | } |
7127 | not_found: | |
7128 | em->start = start; | |
70c8a91c | 7129 | em->orig_start = start; |
d1310b2e | 7130 | em->len = len; |
a52d9a80 | 7131 | not_found_em: |
5f39d397 | 7132 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 7133 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 | 7134 | insert: |
b3b4aa74 | 7135 | btrfs_release_path(path); |
d1310b2e | 7136 | if (em->start > start || extent_map_end(em) <= start) { |
0b246afa | 7137 | btrfs_err(fs_info, |
5d163e0e JM |
7138 | "bad extent! em: [%llu %llu] passed [%llu %llu]", |
7139 | em->start, em->len, start, len); | |
a52d9a80 CM |
7140 | err = -EIO; |
7141 | goto out; | |
7142 | } | |
d1310b2e CM |
7143 | |
7144 | err = 0; | |
890871be | 7145 | write_lock(&em_tree->lock); |
09a2a8f9 | 7146 | ret = add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
7147 | /* it is possible that someone inserted the extent into the tree |
7148 | * while we had the lock dropped. It is also possible that | |
7149 | * an overlapping map exists in the tree | |
7150 | */ | |
a52d9a80 | 7151 | if (ret == -EEXIST) { |
3b951516 | 7152 | struct extent_map *existing; |
e6dcd2dc CM |
7153 | |
7154 | ret = 0; | |
7155 | ||
e6c4efd8 QW |
7156 | existing = search_extent_mapping(em_tree, start, len); |
7157 | /* | |
7158 | * existing will always be non-NULL, since there must be | |
7159 | * extent causing the -EEXIST. | |
7160 | */ | |
8dff9c85 | 7161 | if (existing->start == em->start && |
8e2bd3b7 | 7162 | extent_map_end(existing) >= extent_map_end(em) && |
8dff9c85 CM |
7163 | em->block_start == existing->block_start) { |
7164 | /* | |
8e2bd3b7 OS |
7165 | * The existing extent map already encompasses the |
7166 | * entire extent map we tried to add. | |
8dff9c85 CM |
7167 | */ |
7168 | free_extent_map(em); | |
7169 | em = existing; | |
7170 | err = 0; | |
7171 | ||
7172 | } else if (start >= extent_map_end(existing) || | |
32be3a1a | 7173 | start <= existing->start) { |
e6c4efd8 QW |
7174 | /* |
7175 | * The existing extent map is the one nearest to | |
7176 | * the [start, start + len) range which overlaps | |
7177 | */ | |
7178 | err = merge_extent_mapping(em_tree, existing, | |
7179 | em, start); | |
e1c4b745 | 7180 | free_extent_map(existing); |
e6c4efd8 | 7181 | if (err) { |
3b951516 CM |
7182 | free_extent_map(em); |
7183 | em = NULL; | |
7184 | } | |
7185 | } else { | |
7186 | free_extent_map(em); | |
7187 | em = existing; | |
e6dcd2dc | 7188 | err = 0; |
a52d9a80 | 7189 | } |
a52d9a80 | 7190 | } |
890871be | 7191 | write_unlock(&em_tree->lock); |
a52d9a80 | 7192 | out: |
1abe9b8a | 7193 | |
fc4f21b1 | 7194 | trace_btrfs_get_extent(root, inode, em); |
1abe9b8a | 7195 | |
527afb44 | 7196 | btrfs_free_path(path); |
a52d9a80 | 7197 | if (trans) { |
3a45bb20 | 7198 | ret = btrfs_end_transaction(trans); |
d397712b | 7199 | if (!err) |
a52d9a80 CM |
7200 | err = ret; |
7201 | } | |
a52d9a80 CM |
7202 | if (err) { |
7203 | free_extent_map(em); | |
a52d9a80 CM |
7204 | return ERR_PTR(err); |
7205 | } | |
79787eaa | 7206 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
7207 | return em; |
7208 | } | |
7209 | ||
fc4f21b1 NB |
7210 | struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode, |
7211 | struct page *page, | |
7212 | size_t pg_offset, u64 start, u64 len, | |
7213 | int create) | |
ec29ed5b CM |
7214 | { |
7215 | struct extent_map *em; | |
7216 | struct extent_map *hole_em = NULL; | |
7217 | u64 range_start = start; | |
7218 | u64 end; | |
7219 | u64 found; | |
7220 | u64 found_end; | |
7221 | int err = 0; | |
7222 | ||
7223 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
7224 | if (IS_ERR(em)) | |
7225 | return em; | |
9986277e DC |
7226 | /* |
7227 | * If our em maps to: | |
7228 | * - a hole or | |
7229 | * - a pre-alloc extent, | |
7230 | * there might actually be delalloc bytes behind it. | |
7231 | */ | |
7232 | if (em->block_start != EXTENT_MAP_HOLE && | |
7233 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7234 | return em; | |
7235 | else | |
7236 | hole_em = em; | |
ec29ed5b CM |
7237 | |
7238 | /* check to see if we've wrapped (len == -1 or similar) */ | |
7239 | end = start + len; | |
7240 | if (end < start) | |
7241 | end = (u64)-1; | |
7242 | else | |
7243 | end -= 1; | |
7244 | ||
7245 | em = NULL; | |
7246 | ||
7247 | /* ok, we didn't find anything, lets look for delalloc */ | |
fc4f21b1 | 7248 | found = count_range_bits(&inode->io_tree, &range_start, |
ec29ed5b CM |
7249 | end, len, EXTENT_DELALLOC, 1); |
7250 | found_end = range_start + found; | |
7251 | if (found_end < range_start) | |
7252 | found_end = (u64)-1; | |
7253 | ||
7254 | /* | |
7255 | * we didn't find anything useful, return | |
7256 | * the original results from get_extent() | |
7257 | */ | |
7258 | if (range_start > end || found_end <= start) { | |
7259 | em = hole_em; | |
7260 | hole_em = NULL; | |
7261 | goto out; | |
7262 | } | |
7263 | ||
7264 | /* adjust the range_start to make sure it doesn't | |
7265 | * go backwards from the start they passed in | |
7266 | */ | |
67871254 | 7267 | range_start = max(start, range_start); |
ec29ed5b CM |
7268 | found = found_end - range_start; |
7269 | ||
7270 | if (found > 0) { | |
7271 | u64 hole_start = start; | |
7272 | u64 hole_len = len; | |
7273 | ||
172ddd60 | 7274 | em = alloc_extent_map(); |
ec29ed5b CM |
7275 | if (!em) { |
7276 | err = -ENOMEM; | |
7277 | goto out; | |
7278 | } | |
7279 | /* | |
7280 | * when btrfs_get_extent can't find anything it | |
7281 | * returns one huge hole | |
7282 | * | |
7283 | * make sure what it found really fits our range, and | |
7284 | * adjust to make sure it is based on the start from | |
7285 | * the caller | |
7286 | */ | |
7287 | if (hole_em) { | |
7288 | u64 calc_end = extent_map_end(hole_em); | |
7289 | ||
7290 | if (calc_end <= start || (hole_em->start > end)) { | |
7291 | free_extent_map(hole_em); | |
7292 | hole_em = NULL; | |
7293 | } else { | |
7294 | hole_start = max(hole_em->start, start); | |
7295 | hole_len = calc_end - hole_start; | |
7296 | } | |
7297 | } | |
7298 | em->bdev = NULL; | |
7299 | if (hole_em && range_start > hole_start) { | |
7300 | /* our hole starts before our delalloc, so we | |
7301 | * have to return just the parts of the hole | |
7302 | * that go until the delalloc starts | |
7303 | */ | |
7304 | em->len = min(hole_len, | |
7305 | range_start - hole_start); | |
7306 | em->start = hole_start; | |
7307 | em->orig_start = hole_start; | |
7308 | /* | |
7309 | * don't adjust block start at all, | |
7310 | * it is fixed at EXTENT_MAP_HOLE | |
7311 | */ | |
7312 | em->block_start = hole_em->block_start; | |
7313 | em->block_len = hole_len; | |
f9e4fb53 LB |
7314 | if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags)) |
7315 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
ec29ed5b CM |
7316 | } else { |
7317 | em->start = range_start; | |
7318 | em->len = found; | |
7319 | em->orig_start = range_start; | |
7320 | em->block_start = EXTENT_MAP_DELALLOC; | |
7321 | em->block_len = found; | |
7322 | } | |
7323 | } else if (hole_em) { | |
7324 | return hole_em; | |
7325 | } | |
7326 | out: | |
7327 | ||
7328 | free_extent_map(hole_em); | |
7329 | if (err) { | |
7330 | free_extent_map(em); | |
7331 | return ERR_PTR(err); | |
7332 | } | |
7333 | return em; | |
7334 | } | |
7335 | ||
5f9a8a51 FM |
7336 | static struct extent_map *btrfs_create_dio_extent(struct inode *inode, |
7337 | const u64 start, | |
7338 | const u64 len, | |
7339 | const u64 orig_start, | |
7340 | const u64 block_start, | |
7341 | const u64 block_len, | |
7342 | const u64 orig_block_len, | |
7343 | const u64 ram_bytes, | |
7344 | const int type) | |
7345 | { | |
7346 | struct extent_map *em = NULL; | |
7347 | int ret; | |
7348 | ||
5f9a8a51 | 7349 | if (type != BTRFS_ORDERED_NOCOW) { |
6f9994db LB |
7350 | em = create_io_em(inode, start, len, orig_start, |
7351 | block_start, block_len, orig_block_len, | |
7352 | ram_bytes, | |
7353 | BTRFS_COMPRESS_NONE, /* compress_type */ | |
7354 | type); | |
5f9a8a51 FM |
7355 | if (IS_ERR(em)) |
7356 | goto out; | |
7357 | } | |
7358 | ret = btrfs_add_ordered_extent_dio(inode, start, block_start, | |
7359 | len, block_len, type); | |
7360 | if (ret) { | |
7361 | if (em) { | |
7362 | free_extent_map(em); | |
dcdbc059 | 7363 | btrfs_drop_extent_cache(BTRFS_I(inode), start, |
5f9a8a51 FM |
7364 | start + len - 1, 0); |
7365 | } | |
7366 | em = ERR_PTR(ret); | |
7367 | } | |
7368 | out: | |
5f9a8a51 FM |
7369 | |
7370 | return em; | |
7371 | } | |
7372 | ||
4b46fce2 JB |
7373 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
7374 | u64 start, u64 len) | |
7375 | { | |
0b246afa | 7376 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 7377 | struct btrfs_root *root = BTRFS_I(inode)->root; |
70c8a91c | 7378 | struct extent_map *em; |
4b46fce2 JB |
7379 | struct btrfs_key ins; |
7380 | u64 alloc_hint; | |
7381 | int ret; | |
4b46fce2 | 7382 | |
4b46fce2 | 7383 | alloc_hint = get_extent_allocation_hint(inode, start, len); |
0b246afa | 7384 | ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize, |
da17066c | 7385 | 0, alloc_hint, &ins, 1, 1); |
00361589 JB |
7386 | if (ret) |
7387 | return ERR_PTR(ret); | |
4b46fce2 | 7388 | |
5f9a8a51 FM |
7389 | em = btrfs_create_dio_extent(inode, start, ins.offset, start, |
7390 | ins.objectid, ins.offset, ins.offset, | |
6288d6ea | 7391 | ins.offset, BTRFS_ORDERED_REGULAR); |
0b246afa | 7392 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
5f9a8a51 | 7393 | if (IS_ERR(em)) |
2ff7e61e JM |
7394 | btrfs_free_reserved_extent(fs_info, ins.objectid, |
7395 | ins.offset, 1); | |
de0ee0ed | 7396 | |
4b46fce2 JB |
7397 | return em; |
7398 | } | |
7399 | ||
46bfbb5c CM |
7400 | /* |
7401 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
7402 | * block must be cow'd | |
7403 | */ | |
00361589 | 7404 | noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len, |
7ee9e440 JB |
7405 | u64 *orig_start, u64 *orig_block_len, |
7406 | u64 *ram_bytes) | |
46bfbb5c | 7407 | { |
2ff7e61e | 7408 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
46bfbb5c CM |
7409 | struct btrfs_path *path; |
7410 | int ret; | |
7411 | struct extent_buffer *leaf; | |
7412 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7b2b7085 | 7413 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
46bfbb5c CM |
7414 | struct btrfs_file_extent_item *fi; |
7415 | struct btrfs_key key; | |
7416 | u64 disk_bytenr; | |
7417 | u64 backref_offset; | |
7418 | u64 extent_end; | |
7419 | u64 num_bytes; | |
7420 | int slot; | |
7421 | int found_type; | |
7ee9e440 | 7422 | bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW); |
e77751aa | 7423 | |
46bfbb5c CM |
7424 | path = btrfs_alloc_path(); |
7425 | if (!path) | |
7426 | return -ENOMEM; | |
7427 | ||
f85b7379 DS |
7428 | ret = btrfs_lookup_file_extent(NULL, root, path, |
7429 | btrfs_ino(BTRFS_I(inode)), offset, 0); | |
46bfbb5c CM |
7430 | if (ret < 0) |
7431 | goto out; | |
7432 | ||
7433 | slot = path->slots[0]; | |
7434 | if (ret == 1) { | |
7435 | if (slot == 0) { | |
7436 | /* can't find the item, must cow */ | |
7437 | ret = 0; | |
7438 | goto out; | |
7439 | } | |
7440 | slot--; | |
7441 | } | |
7442 | ret = 0; | |
7443 | leaf = path->nodes[0]; | |
7444 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4a0cc7ca | 7445 | if (key.objectid != btrfs_ino(BTRFS_I(inode)) || |
46bfbb5c CM |
7446 | key.type != BTRFS_EXTENT_DATA_KEY) { |
7447 | /* not our file or wrong item type, must cow */ | |
7448 | goto out; | |
7449 | } | |
7450 | ||
7451 | if (key.offset > offset) { | |
7452 | /* Wrong offset, must cow */ | |
7453 | goto out; | |
7454 | } | |
7455 | ||
7456 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
7457 | found_type = btrfs_file_extent_type(leaf, fi); | |
7458 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
7459 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
7460 | /* not a regular extent, must cow */ | |
7461 | goto out; | |
7462 | } | |
7ee9e440 JB |
7463 | |
7464 | if (!nocow && found_type == BTRFS_FILE_EXTENT_REG) | |
7465 | goto out; | |
7466 | ||
e77751aa MX |
7467 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
7468 | if (extent_end <= offset) | |
7469 | goto out; | |
7470 | ||
46bfbb5c | 7471 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
7ee9e440 JB |
7472 | if (disk_bytenr == 0) |
7473 | goto out; | |
7474 | ||
7475 | if (btrfs_file_extent_compression(leaf, fi) || | |
7476 | btrfs_file_extent_encryption(leaf, fi) || | |
7477 | btrfs_file_extent_other_encoding(leaf, fi)) | |
7478 | goto out; | |
7479 | ||
46bfbb5c CM |
7480 | backref_offset = btrfs_file_extent_offset(leaf, fi); |
7481 | ||
7ee9e440 JB |
7482 | if (orig_start) { |
7483 | *orig_start = key.offset - backref_offset; | |
7484 | *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
7485 | *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); | |
7486 | } | |
eb384b55 | 7487 | |
2ff7e61e | 7488 | if (btrfs_extent_readonly(fs_info, disk_bytenr)) |
46bfbb5c | 7489 | goto out; |
7b2b7085 MX |
7490 | |
7491 | num_bytes = min(offset + *len, extent_end) - offset; | |
7492 | if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
7493 | u64 range_end; | |
7494 | ||
da17066c JM |
7495 | range_end = round_up(offset + num_bytes, |
7496 | root->fs_info->sectorsize) - 1; | |
7b2b7085 MX |
7497 | ret = test_range_bit(io_tree, offset, range_end, |
7498 | EXTENT_DELALLOC, 0, NULL); | |
7499 | if (ret) { | |
7500 | ret = -EAGAIN; | |
7501 | goto out; | |
7502 | } | |
7503 | } | |
7504 | ||
1bda19eb | 7505 | btrfs_release_path(path); |
46bfbb5c CM |
7506 | |
7507 | /* | |
7508 | * look for other files referencing this extent, if we | |
7509 | * find any we must cow | |
7510 | */ | |
00361589 | 7511 | |
e4c3b2dc | 7512 | ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)), |
00361589 | 7513 | key.offset - backref_offset, disk_bytenr); |
00361589 JB |
7514 | if (ret) { |
7515 | ret = 0; | |
7516 | goto out; | |
7517 | } | |
46bfbb5c CM |
7518 | |
7519 | /* | |
7520 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
7521 | * in this extent we are about to write. If there | |
7522 | * are any csums in that range we have to cow in order | |
7523 | * to keep the csums correct | |
7524 | */ | |
7525 | disk_bytenr += backref_offset; | |
7526 | disk_bytenr += offset - key.offset; | |
2ff7e61e JM |
7527 | if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes)) |
7528 | goto out; | |
46bfbb5c CM |
7529 | /* |
7530 | * all of the above have passed, it is safe to overwrite this extent | |
7531 | * without cow | |
7532 | */ | |
eb384b55 | 7533 | *len = num_bytes; |
46bfbb5c CM |
7534 | ret = 1; |
7535 | out: | |
7536 | btrfs_free_path(path); | |
7537 | return ret; | |
7538 | } | |
7539 | ||
fc4adbff AG |
7540 | bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end) |
7541 | { | |
7542 | struct radix_tree_root *root = &inode->i_mapping->page_tree; | |
e03733da | 7543 | bool found = false; |
fc4adbff AG |
7544 | void **pagep = NULL; |
7545 | struct page *page = NULL; | |
cc2b702c DS |
7546 | unsigned long start_idx; |
7547 | unsigned long end_idx; | |
fc4adbff | 7548 | |
09cbfeaf | 7549 | start_idx = start >> PAGE_SHIFT; |
fc4adbff AG |
7550 | |
7551 | /* | |
7552 | * end is the last byte in the last page. end == start is legal | |
7553 | */ | |
09cbfeaf | 7554 | end_idx = end >> PAGE_SHIFT; |
fc4adbff AG |
7555 | |
7556 | rcu_read_lock(); | |
7557 | ||
7558 | /* Most of the code in this while loop is lifted from | |
7559 | * find_get_page. It's been modified to begin searching from a | |
7560 | * page and return just the first page found in that range. If the | |
7561 | * found idx is less than or equal to the end idx then we know that | |
7562 | * a page exists. If no pages are found or if those pages are | |
7563 | * outside of the range then we're fine (yay!) */ | |
7564 | while (page == NULL && | |
7565 | radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) { | |
7566 | page = radix_tree_deref_slot(pagep); | |
7567 | if (unlikely(!page)) | |
7568 | break; | |
7569 | ||
7570 | if (radix_tree_exception(page)) { | |
809f9016 FM |
7571 | if (radix_tree_deref_retry(page)) { |
7572 | page = NULL; | |
fc4adbff | 7573 | continue; |
809f9016 | 7574 | } |
fc4adbff AG |
7575 | /* |
7576 | * Otherwise, shmem/tmpfs must be storing a swap entry | |
7577 | * here as an exceptional entry: so return it without | |
7578 | * attempting to raise page count. | |
7579 | */ | |
6fdef6d4 | 7580 | page = NULL; |
fc4adbff AG |
7581 | break; /* TODO: Is this relevant for this use case? */ |
7582 | } | |
7583 | ||
91405151 FM |
7584 | if (!page_cache_get_speculative(page)) { |
7585 | page = NULL; | |
fc4adbff | 7586 | continue; |
91405151 | 7587 | } |
fc4adbff AG |
7588 | |
7589 | /* | |
7590 | * Has the page moved? | |
7591 | * This is part of the lockless pagecache protocol. See | |
7592 | * include/linux/pagemap.h for details. | |
7593 | */ | |
7594 | if (unlikely(page != *pagep)) { | |
09cbfeaf | 7595 | put_page(page); |
fc4adbff AG |
7596 | page = NULL; |
7597 | } | |
7598 | } | |
7599 | ||
7600 | if (page) { | |
7601 | if (page->index <= end_idx) | |
7602 | found = true; | |
09cbfeaf | 7603 | put_page(page); |
fc4adbff AG |
7604 | } |
7605 | ||
7606 | rcu_read_unlock(); | |
7607 | return found; | |
7608 | } | |
7609 | ||
eb838e73 JB |
7610 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
7611 | struct extent_state **cached_state, int writing) | |
7612 | { | |
7613 | struct btrfs_ordered_extent *ordered; | |
7614 | int ret = 0; | |
7615 | ||
7616 | while (1) { | |
7617 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
ff13db41 | 7618 | cached_state); |
eb838e73 JB |
7619 | /* |
7620 | * We're concerned with the entire range that we're going to be | |
01327610 | 7621 | * doing DIO to, so we need to make sure there's no ordered |
eb838e73 JB |
7622 | * extents in this range. |
7623 | */ | |
a776c6fa | 7624 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart, |
eb838e73 JB |
7625 | lockend - lockstart + 1); |
7626 | ||
7627 | /* | |
7628 | * We need to make sure there are no buffered pages in this | |
7629 | * range either, we could have raced between the invalidate in | |
7630 | * generic_file_direct_write and locking the extent. The | |
7631 | * invalidate needs to happen so that reads after a write do not | |
7632 | * get stale data. | |
7633 | */ | |
fc4adbff AG |
7634 | if (!ordered && |
7635 | (!writing || | |
7636 | !btrfs_page_exists_in_range(inode, lockstart, lockend))) | |
eb838e73 JB |
7637 | break; |
7638 | ||
7639 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
7640 | cached_state, GFP_NOFS); | |
7641 | ||
7642 | if (ordered) { | |
ade77029 FM |
7643 | /* |
7644 | * If we are doing a DIO read and the ordered extent we | |
7645 | * found is for a buffered write, we can not wait for it | |
7646 | * to complete and retry, because if we do so we can | |
7647 | * deadlock with concurrent buffered writes on page | |
7648 | * locks. This happens only if our DIO read covers more | |
7649 | * than one extent map, if at this point has already | |
7650 | * created an ordered extent for a previous extent map | |
7651 | * and locked its range in the inode's io tree, and a | |
7652 | * concurrent write against that previous extent map's | |
7653 | * range and this range started (we unlock the ranges | |
7654 | * in the io tree only when the bios complete and | |
7655 | * buffered writes always lock pages before attempting | |
7656 | * to lock range in the io tree). | |
7657 | */ | |
7658 | if (writing || | |
7659 | test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) | |
7660 | btrfs_start_ordered_extent(inode, ordered, 1); | |
7661 | else | |
7662 | ret = -ENOTBLK; | |
eb838e73 JB |
7663 | btrfs_put_ordered_extent(ordered); |
7664 | } else { | |
eb838e73 | 7665 | /* |
b850ae14 FM |
7666 | * We could trigger writeback for this range (and wait |
7667 | * for it to complete) and then invalidate the pages for | |
7668 | * this range (through invalidate_inode_pages2_range()), | |
7669 | * but that can lead us to a deadlock with a concurrent | |
7670 | * call to readpages() (a buffered read or a defrag call | |
7671 | * triggered a readahead) on a page lock due to an | |
7672 | * ordered dio extent we created before but did not have | |
7673 | * yet a corresponding bio submitted (whence it can not | |
7674 | * complete), which makes readpages() wait for that | |
7675 | * ordered extent to complete while holding a lock on | |
7676 | * that page. | |
eb838e73 | 7677 | */ |
b850ae14 | 7678 | ret = -ENOTBLK; |
eb838e73 JB |
7679 | } |
7680 | ||
ade77029 FM |
7681 | if (ret) |
7682 | break; | |
7683 | ||
eb838e73 JB |
7684 | cond_resched(); |
7685 | } | |
7686 | ||
7687 | return ret; | |
7688 | } | |
7689 | ||
6f9994db LB |
7690 | /* The callers of this must take lock_extent() */ |
7691 | static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len, | |
7692 | u64 orig_start, u64 block_start, | |
7693 | u64 block_len, u64 orig_block_len, | |
7694 | u64 ram_bytes, int compress_type, | |
7695 | int type) | |
69ffb543 JB |
7696 | { |
7697 | struct extent_map_tree *em_tree; | |
7698 | struct extent_map *em; | |
7699 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7700 | int ret; | |
7701 | ||
6f9994db LB |
7702 | ASSERT(type == BTRFS_ORDERED_PREALLOC || |
7703 | type == BTRFS_ORDERED_COMPRESSED || | |
7704 | type == BTRFS_ORDERED_NOCOW || | |
1af4a0aa | 7705 | type == BTRFS_ORDERED_REGULAR); |
6f9994db | 7706 | |
69ffb543 JB |
7707 | em_tree = &BTRFS_I(inode)->extent_tree; |
7708 | em = alloc_extent_map(); | |
7709 | if (!em) | |
7710 | return ERR_PTR(-ENOMEM); | |
7711 | ||
7712 | em->start = start; | |
7713 | em->orig_start = orig_start; | |
7714 | em->len = len; | |
7715 | em->block_len = block_len; | |
7716 | em->block_start = block_start; | |
7717 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
b4939680 | 7718 | em->orig_block_len = orig_block_len; |
cc95bef6 | 7719 | em->ram_bytes = ram_bytes; |
70c8a91c | 7720 | em->generation = -1; |
69ffb543 | 7721 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
1af4a0aa | 7722 | if (type == BTRFS_ORDERED_PREALLOC) { |
b11e234d | 7723 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
1af4a0aa | 7724 | } else if (type == BTRFS_ORDERED_COMPRESSED) { |
6f9994db LB |
7725 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
7726 | em->compress_type = compress_type; | |
7727 | } | |
69ffb543 JB |
7728 | |
7729 | do { | |
dcdbc059 | 7730 | btrfs_drop_extent_cache(BTRFS_I(inode), em->start, |
69ffb543 JB |
7731 | em->start + em->len - 1, 0); |
7732 | write_lock(&em_tree->lock); | |
09a2a8f9 | 7733 | ret = add_extent_mapping(em_tree, em, 1); |
69ffb543 | 7734 | write_unlock(&em_tree->lock); |
6f9994db LB |
7735 | /* |
7736 | * The caller has taken lock_extent(), who could race with us | |
7737 | * to add em? | |
7738 | */ | |
69ffb543 JB |
7739 | } while (ret == -EEXIST); |
7740 | ||
7741 | if (ret) { | |
7742 | free_extent_map(em); | |
7743 | return ERR_PTR(ret); | |
7744 | } | |
7745 | ||
6f9994db | 7746 | /* em got 2 refs now, callers needs to do free_extent_map once. */ |
69ffb543 JB |
7747 | return em; |
7748 | } | |
7749 | ||
4b46fce2 JB |
7750 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
7751 | struct buffer_head *bh_result, int create) | |
7752 | { | |
0b246afa | 7753 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 7754 | struct extent_map *em; |
eb838e73 | 7755 | struct extent_state *cached_state = NULL; |
50745b0a | 7756 | struct btrfs_dio_data *dio_data = NULL; |
4b46fce2 | 7757 | u64 start = iblock << inode->i_blkbits; |
eb838e73 | 7758 | u64 lockstart, lockend; |
4b46fce2 | 7759 | u64 len = bh_result->b_size; |
eb838e73 | 7760 | int unlock_bits = EXTENT_LOCKED; |
0934856d | 7761 | int ret = 0; |
eb838e73 | 7762 | |
172a5049 | 7763 | if (create) |
3266789f | 7764 | unlock_bits |= EXTENT_DIRTY; |
172a5049 | 7765 | else |
0b246afa | 7766 | len = min_t(u64, len, fs_info->sectorsize); |
eb838e73 | 7767 | |
c329861d JB |
7768 | lockstart = start; |
7769 | lockend = start + len - 1; | |
7770 | ||
e1cbbfa5 JB |
7771 | if (current->journal_info) { |
7772 | /* | |
7773 | * Need to pull our outstanding extents and set journal_info to NULL so | |
01327610 | 7774 | * that anything that needs to check if there's a transaction doesn't get |
e1cbbfa5 JB |
7775 | * confused. |
7776 | */ | |
50745b0a | 7777 | dio_data = current->journal_info; |
e1cbbfa5 JB |
7778 | current->journal_info = NULL; |
7779 | } | |
7780 | ||
eb838e73 JB |
7781 | /* |
7782 | * If this errors out it's because we couldn't invalidate pagecache for | |
7783 | * this range and we need to fallback to buffered. | |
7784 | */ | |
9c9464cc FM |
7785 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, |
7786 | create)) { | |
7787 | ret = -ENOTBLK; | |
7788 | goto err; | |
7789 | } | |
eb838e73 | 7790 | |
fc4f21b1 | 7791 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0); |
eb838e73 JB |
7792 | if (IS_ERR(em)) { |
7793 | ret = PTR_ERR(em); | |
7794 | goto unlock_err; | |
7795 | } | |
4b46fce2 JB |
7796 | |
7797 | /* | |
7798 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
7799 | * io. INLINE is special, and we could probably kludge it in here, but | |
7800 | * it's still buffered so for safety lets just fall back to the generic | |
7801 | * buffered path. | |
7802 | * | |
7803 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
7804 | * decompress it, so there will be buffering required no matter what we | |
7805 | * do, so go ahead and fallback to buffered. | |
7806 | * | |
01327610 | 7807 | * We return -ENOTBLK because that's what makes DIO go ahead and go back |
4b46fce2 JB |
7808 | * to buffered IO. Don't blame me, this is the price we pay for using |
7809 | * the generic code. | |
7810 | */ | |
7811 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
7812 | em->block_start == EXTENT_MAP_INLINE) { | |
7813 | free_extent_map(em); | |
eb838e73 JB |
7814 | ret = -ENOTBLK; |
7815 | goto unlock_err; | |
4b46fce2 JB |
7816 | } |
7817 | ||
7818 | /* Just a good old fashioned hole, return */ | |
7819 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
7820 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
7821 | free_extent_map(em); | |
eb838e73 | 7822 | goto unlock_err; |
4b46fce2 JB |
7823 | } |
7824 | ||
7825 | /* | |
7826 | * We don't allocate a new extent in the following cases | |
7827 | * | |
7828 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
7829 | * existing extent. | |
7830 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
7831 | * just use the extent. | |
7832 | * | |
7833 | */ | |
46bfbb5c | 7834 | if (!create) { |
eb838e73 JB |
7835 | len = min(len, em->len - (start - em->start)); |
7836 | lockstart = start + len; | |
7837 | goto unlock; | |
46bfbb5c | 7838 | } |
4b46fce2 JB |
7839 | |
7840 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
7841 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
7842 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 | 7843 | int type; |
eb384b55 | 7844 | u64 block_start, orig_start, orig_block_len, ram_bytes; |
4b46fce2 JB |
7845 | |
7846 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7847 | type = BTRFS_ORDERED_PREALLOC; | |
7848 | else | |
7849 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 7850 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 7851 | block_start = em->block_start + (start - em->start); |
46bfbb5c | 7852 | |
00361589 | 7853 | if (can_nocow_extent(inode, start, &len, &orig_start, |
f78c436c | 7854 | &orig_block_len, &ram_bytes) == 1 && |
0b246afa | 7855 | btrfs_inc_nocow_writers(fs_info, block_start)) { |
5f9a8a51 | 7856 | struct extent_map *em2; |
0b901916 | 7857 | |
5f9a8a51 FM |
7858 | em2 = btrfs_create_dio_extent(inode, start, len, |
7859 | orig_start, block_start, | |
7860 | len, orig_block_len, | |
7861 | ram_bytes, type); | |
0b246afa | 7862 | btrfs_dec_nocow_writers(fs_info, block_start); |
69ffb543 JB |
7863 | if (type == BTRFS_ORDERED_PREALLOC) { |
7864 | free_extent_map(em); | |
5f9a8a51 | 7865 | em = em2; |
69ffb543 | 7866 | } |
5f9a8a51 FM |
7867 | if (em2 && IS_ERR(em2)) { |
7868 | ret = PTR_ERR(em2); | |
eb838e73 | 7869 | goto unlock_err; |
46bfbb5c | 7870 | } |
18513091 WX |
7871 | /* |
7872 | * For inode marked NODATACOW or extent marked PREALLOC, | |
7873 | * use the existing or preallocated extent, so does not | |
7874 | * need to adjust btrfs_space_info's bytes_may_use. | |
7875 | */ | |
7876 | btrfs_free_reserved_data_space_noquota(inode, | |
7877 | start, len); | |
46bfbb5c | 7878 | goto unlock; |
4b46fce2 | 7879 | } |
4b46fce2 | 7880 | } |
00361589 | 7881 | |
46bfbb5c CM |
7882 | /* |
7883 | * this will cow the extent, reset the len in case we changed | |
7884 | * it above | |
7885 | */ | |
7886 | len = bh_result->b_size; | |
70c8a91c JB |
7887 | free_extent_map(em); |
7888 | em = btrfs_new_extent_direct(inode, start, len); | |
eb838e73 JB |
7889 | if (IS_ERR(em)) { |
7890 | ret = PTR_ERR(em); | |
7891 | goto unlock_err; | |
7892 | } | |
46bfbb5c CM |
7893 | len = min(len, em->len - (start - em->start)); |
7894 | unlock: | |
4b46fce2 JB |
7895 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> |
7896 | inode->i_blkbits; | |
46bfbb5c | 7897 | bh_result->b_size = len; |
4b46fce2 JB |
7898 | bh_result->b_bdev = em->bdev; |
7899 | set_buffer_mapped(bh_result); | |
c3473e83 JB |
7900 | if (create) { |
7901 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7902 | set_buffer_new(bh_result); | |
7903 | ||
7904 | /* | |
7905 | * Need to update the i_size under the extent lock so buffered | |
7906 | * readers will get the updated i_size when we unlock. | |
7907 | */ | |
4aaedfb0 | 7908 | if (!dio_data->overwrite && start + len > i_size_read(inode)) |
c3473e83 | 7909 | i_size_write(inode, start + len); |
0934856d | 7910 | |
50745b0a | 7911 | WARN_ON(dio_data->reserve < len); |
7912 | dio_data->reserve -= len; | |
f28a4928 | 7913 | dio_data->unsubmitted_oe_range_end = start + len; |
50745b0a | 7914 | current->journal_info = dio_data; |
c3473e83 | 7915 | } |
4b46fce2 | 7916 | |
eb838e73 JB |
7917 | /* |
7918 | * In the case of write we need to clear and unlock the entire range, | |
7919 | * in the case of read we need to unlock only the end area that we | |
7920 | * aren't using if there is any left over space. | |
7921 | */ | |
24c03fa5 | 7922 | if (lockstart < lockend) { |
0934856d MX |
7923 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, |
7924 | lockend, unlock_bits, 1, 0, | |
7925 | &cached_state, GFP_NOFS); | |
24c03fa5 | 7926 | } else { |
eb838e73 | 7927 | free_extent_state(cached_state); |
24c03fa5 | 7928 | } |
eb838e73 | 7929 | |
4b46fce2 JB |
7930 | free_extent_map(em); |
7931 | ||
7932 | return 0; | |
eb838e73 JB |
7933 | |
7934 | unlock_err: | |
eb838e73 JB |
7935 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
7936 | unlock_bits, 1, 0, &cached_state, GFP_NOFS); | |
9c9464cc | 7937 | err: |
50745b0a | 7938 | if (dio_data) |
7939 | current->journal_info = dio_data; | |
eb838e73 | 7940 | return ret; |
4b46fce2 JB |
7941 | } |
7942 | ||
58efbc9f OS |
7943 | static inline blk_status_t submit_dio_repair_bio(struct inode *inode, |
7944 | struct bio *bio, | |
7945 | int mirror_num) | |
8b110e39 | 7946 | { |
2ff7e61e | 7947 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
58efbc9f | 7948 | blk_status_t ret; |
8b110e39 | 7949 | |
37226b21 | 7950 | BUG_ON(bio_op(bio) == REQ_OP_WRITE); |
8b110e39 MX |
7951 | |
7952 | bio_get(bio); | |
7953 | ||
2ff7e61e | 7954 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR); |
8b110e39 MX |
7955 | if (ret) |
7956 | goto err; | |
7957 | ||
2ff7e61e | 7958 | ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); |
8b110e39 MX |
7959 | err: |
7960 | bio_put(bio); | |
7961 | return ret; | |
7962 | } | |
7963 | ||
7964 | static int btrfs_check_dio_repairable(struct inode *inode, | |
7965 | struct bio *failed_bio, | |
7966 | struct io_failure_record *failrec, | |
7967 | int failed_mirror) | |
7968 | { | |
ab8d0fc4 | 7969 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
8b110e39 MX |
7970 | int num_copies; |
7971 | ||
ab8d0fc4 | 7972 | num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len); |
8b110e39 MX |
7973 | if (num_copies == 1) { |
7974 | /* | |
7975 | * we only have a single copy of the data, so don't bother with | |
7976 | * all the retry and error correction code that follows. no | |
7977 | * matter what the error is, it is very likely to persist. | |
7978 | */ | |
ab8d0fc4 JM |
7979 | btrfs_debug(fs_info, |
7980 | "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d", | |
7981 | num_copies, failrec->this_mirror, failed_mirror); | |
8b110e39 MX |
7982 | return 0; |
7983 | } | |
7984 | ||
7985 | failrec->failed_mirror = failed_mirror; | |
7986 | failrec->this_mirror++; | |
7987 | if (failrec->this_mirror == failed_mirror) | |
7988 | failrec->this_mirror++; | |
7989 | ||
7990 | if (failrec->this_mirror > num_copies) { | |
ab8d0fc4 JM |
7991 | btrfs_debug(fs_info, |
7992 | "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d", | |
7993 | num_copies, failrec->this_mirror, failed_mirror); | |
8b110e39 MX |
7994 | return 0; |
7995 | } | |
7996 | ||
7997 | return 1; | |
7998 | } | |
7999 | ||
58efbc9f OS |
8000 | static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio, |
8001 | struct page *page, unsigned int pgoff, | |
8002 | u64 start, u64 end, int failed_mirror, | |
8003 | bio_end_io_t *repair_endio, void *repair_arg) | |
8b110e39 MX |
8004 | { |
8005 | struct io_failure_record *failrec; | |
7870d082 JB |
8006 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
8007 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
8b110e39 MX |
8008 | struct bio *bio; |
8009 | int isector; | |
f1c77c55 | 8010 | unsigned int read_mode = 0; |
17347cec | 8011 | int segs; |
8b110e39 | 8012 | int ret; |
58efbc9f | 8013 | blk_status_t status; |
8b110e39 | 8014 | |
37226b21 | 8015 | BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE); |
8b110e39 MX |
8016 | |
8017 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
8018 | if (ret) | |
58efbc9f | 8019 | return errno_to_blk_status(ret); |
8b110e39 MX |
8020 | |
8021 | ret = btrfs_check_dio_repairable(inode, failed_bio, failrec, | |
8022 | failed_mirror); | |
8023 | if (!ret) { | |
7870d082 | 8024 | free_io_failure(failure_tree, io_tree, failrec); |
58efbc9f | 8025 | return BLK_STS_IOERR; |
8b110e39 MX |
8026 | } |
8027 | ||
17347cec LB |
8028 | segs = bio_segments(failed_bio); |
8029 | if (segs > 1 || | |
8030 | (failed_bio->bi_io_vec->bv_len > btrfs_inode_sectorsize(inode))) | |
70fd7614 | 8031 | read_mode |= REQ_FAILFAST_DEV; |
8b110e39 MX |
8032 | |
8033 | isector = start - btrfs_io_bio(failed_bio)->logical; | |
8034 | isector >>= inode->i_sb->s_blocksize_bits; | |
8035 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
2dabb324 | 8036 | pgoff, isector, repair_endio, repair_arg); |
37226b21 | 8037 | bio_set_op_attrs(bio, REQ_OP_READ, read_mode); |
8b110e39 MX |
8038 | |
8039 | btrfs_debug(BTRFS_I(inode)->root->fs_info, | |
913e1535 | 8040 | "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d", |
8b110e39 MX |
8041 | read_mode, failrec->this_mirror, failrec->in_validation); |
8042 | ||
58efbc9f OS |
8043 | status = submit_dio_repair_bio(inode, bio, failrec->this_mirror); |
8044 | if (status) { | |
7870d082 | 8045 | free_io_failure(failure_tree, io_tree, failrec); |
8b110e39 MX |
8046 | bio_put(bio); |
8047 | } | |
8048 | ||
58efbc9f | 8049 | return status; |
8b110e39 MX |
8050 | } |
8051 | ||
8052 | struct btrfs_retry_complete { | |
8053 | struct completion done; | |
8054 | struct inode *inode; | |
8055 | u64 start; | |
8056 | int uptodate; | |
8057 | }; | |
8058 | ||
4246a0b6 | 8059 | static void btrfs_retry_endio_nocsum(struct bio *bio) |
8b110e39 MX |
8060 | { |
8061 | struct btrfs_retry_complete *done = bio->bi_private; | |
7870d082 | 8062 | struct inode *inode = done->inode; |
8b110e39 | 8063 | struct bio_vec *bvec; |
7870d082 | 8064 | struct extent_io_tree *io_tree, *failure_tree; |
8b110e39 MX |
8065 | int i; |
8066 | ||
4e4cbee9 | 8067 | if (bio->bi_status) |
8b110e39 MX |
8068 | goto end; |
8069 | ||
2dabb324 | 8070 | ASSERT(bio->bi_vcnt == 1); |
7870d082 JB |
8071 | io_tree = &BTRFS_I(inode)->io_tree; |
8072 | failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
8073 | ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(inode)); | |
2dabb324 | 8074 | |
8b110e39 | 8075 | done->uptodate = 1; |
c09abff8 | 8076 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
8b110e39 | 8077 | bio_for_each_segment_all(bvec, bio, i) |
7870d082 JB |
8078 | clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree, |
8079 | io_tree, done->start, bvec->bv_page, | |
8080 | btrfs_ino(BTRFS_I(inode)), 0); | |
8b110e39 MX |
8081 | end: |
8082 | complete(&done->done); | |
8083 | bio_put(bio); | |
8084 | } | |
8085 | ||
58efbc9f OS |
8086 | static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode, |
8087 | struct btrfs_io_bio *io_bio) | |
4b46fce2 | 8088 | { |
2dabb324 | 8089 | struct btrfs_fs_info *fs_info; |
17347cec LB |
8090 | struct bio_vec bvec; |
8091 | struct bvec_iter iter; | |
8b110e39 | 8092 | struct btrfs_retry_complete done; |
4b46fce2 | 8093 | u64 start; |
2dabb324 CR |
8094 | unsigned int pgoff; |
8095 | u32 sectorsize; | |
8096 | int nr_sectors; | |
58efbc9f OS |
8097 | blk_status_t ret; |
8098 | blk_status_t err = BLK_STS_OK; | |
4b46fce2 | 8099 | |
2dabb324 | 8100 | fs_info = BTRFS_I(inode)->root->fs_info; |
da17066c | 8101 | sectorsize = fs_info->sectorsize; |
2dabb324 | 8102 | |
8b110e39 MX |
8103 | start = io_bio->logical; |
8104 | done.inode = inode; | |
17347cec | 8105 | io_bio->bio.bi_iter = io_bio->iter; |
8b110e39 | 8106 | |
17347cec LB |
8107 | bio_for_each_segment(bvec, &io_bio->bio, iter) { |
8108 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len); | |
8109 | pgoff = bvec.bv_offset; | |
2dabb324 CR |
8110 | |
8111 | next_block_or_try_again: | |
8b110e39 MX |
8112 | done.uptodate = 0; |
8113 | done.start = start; | |
8114 | init_completion(&done.done); | |
8115 | ||
17347cec | 8116 | ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page, |
2dabb324 CR |
8117 | pgoff, start, start + sectorsize - 1, |
8118 | io_bio->mirror_num, | |
8119 | btrfs_retry_endio_nocsum, &done); | |
629ebf4f LB |
8120 | if (ret) { |
8121 | err = ret; | |
8122 | goto next; | |
8123 | } | |
8b110e39 | 8124 | |
9c17f6cd | 8125 | wait_for_completion_io(&done.done); |
8b110e39 MX |
8126 | |
8127 | if (!done.uptodate) { | |
8128 | /* We might have another mirror, so try again */ | |
2dabb324 | 8129 | goto next_block_or_try_again; |
8b110e39 MX |
8130 | } |
8131 | ||
629ebf4f | 8132 | next: |
2dabb324 CR |
8133 | start += sectorsize; |
8134 | ||
97bf5a55 LB |
8135 | nr_sectors--; |
8136 | if (nr_sectors) { | |
2dabb324 | 8137 | pgoff += sectorsize; |
97bf5a55 | 8138 | ASSERT(pgoff < PAGE_SIZE); |
2dabb324 CR |
8139 | goto next_block_or_try_again; |
8140 | } | |
8b110e39 MX |
8141 | } |
8142 | ||
629ebf4f | 8143 | return err; |
8b110e39 MX |
8144 | } |
8145 | ||
4246a0b6 | 8146 | static void btrfs_retry_endio(struct bio *bio) |
8b110e39 MX |
8147 | { |
8148 | struct btrfs_retry_complete *done = bio->bi_private; | |
8149 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
7870d082 JB |
8150 | struct extent_io_tree *io_tree, *failure_tree; |
8151 | struct inode *inode = done->inode; | |
8b110e39 MX |
8152 | struct bio_vec *bvec; |
8153 | int uptodate; | |
8154 | int ret; | |
8155 | int i; | |
8156 | ||
4e4cbee9 | 8157 | if (bio->bi_status) |
8b110e39 MX |
8158 | goto end; |
8159 | ||
8160 | uptodate = 1; | |
2dabb324 | 8161 | |
2dabb324 | 8162 | ASSERT(bio->bi_vcnt == 1); |
2e949b0a | 8163 | ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(done->inode)); |
2dabb324 | 8164 | |
7870d082 JB |
8165 | io_tree = &BTRFS_I(inode)->io_tree; |
8166 | failure_tree = &BTRFS_I(inode)->io_failure_tree; | |
8167 | ||
c09abff8 | 8168 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
8b110e39 | 8169 | bio_for_each_segment_all(bvec, bio, i) { |
7870d082 JB |
8170 | ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page, |
8171 | bvec->bv_offset, done->start, | |
8172 | bvec->bv_len); | |
8b110e39 | 8173 | if (!ret) |
7870d082 JB |
8174 | clean_io_failure(BTRFS_I(inode)->root->fs_info, |
8175 | failure_tree, io_tree, done->start, | |
8176 | bvec->bv_page, | |
8177 | btrfs_ino(BTRFS_I(inode)), | |
8178 | bvec->bv_offset); | |
8b110e39 MX |
8179 | else |
8180 | uptodate = 0; | |
8181 | } | |
8182 | ||
8183 | done->uptodate = uptodate; | |
8184 | end: | |
8185 | complete(&done->done); | |
8186 | bio_put(bio); | |
8187 | } | |
8188 | ||
4e4cbee9 CH |
8189 | static blk_status_t __btrfs_subio_endio_read(struct inode *inode, |
8190 | struct btrfs_io_bio *io_bio, blk_status_t err) | |
8b110e39 | 8191 | { |
2dabb324 | 8192 | struct btrfs_fs_info *fs_info; |
17347cec LB |
8193 | struct bio_vec bvec; |
8194 | struct bvec_iter iter; | |
8b110e39 MX |
8195 | struct btrfs_retry_complete done; |
8196 | u64 start; | |
8197 | u64 offset = 0; | |
2dabb324 CR |
8198 | u32 sectorsize; |
8199 | int nr_sectors; | |
8200 | unsigned int pgoff; | |
8201 | int csum_pos; | |
ef7cdac1 | 8202 | bool uptodate = (err == 0); |
8b110e39 | 8203 | int ret; |
58efbc9f | 8204 | blk_status_t status; |
dc380aea | 8205 | |
2dabb324 | 8206 | fs_info = BTRFS_I(inode)->root->fs_info; |
da17066c | 8207 | sectorsize = fs_info->sectorsize; |
2dabb324 | 8208 | |
58efbc9f | 8209 | err = BLK_STS_OK; |
c1dc0896 | 8210 | start = io_bio->logical; |
8b110e39 | 8211 | done.inode = inode; |
17347cec | 8212 | io_bio->bio.bi_iter = io_bio->iter; |
8b110e39 | 8213 | |
17347cec LB |
8214 | bio_for_each_segment(bvec, &io_bio->bio, iter) { |
8215 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len); | |
2dabb324 | 8216 | |
17347cec | 8217 | pgoff = bvec.bv_offset; |
2dabb324 | 8218 | next_block: |
ef7cdac1 LB |
8219 | if (uptodate) { |
8220 | csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset); | |
8221 | ret = __readpage_endio_check(inode, io_bio, csum_pos, | |
8222 | bvec.bv_page, pgoff, start, sectorsize); | |
8223 | if (likely(!ret)) | |
8224 | goto next; | |
8225 | } | |
8b110e39 MX |
8226 | try_again: |
8227 | done.uptodate = 0; | |
8228 | done.start = start; | |
8229 | init_completion(&done.done); | |
8230 | ||
58efbc9f OS |
8231 | status = dio_read_error(inode, &io_bio->bio, bvec.bv_page, |
8232 | pgoff, start, start + sectorsize - 1, | |
8233 | io_bio->mirror_num, btrfs_retry_endio, | |
8234 | &done); | |
8235 | if (status) { | |
8236 | err = status; | |
8b110e39 MX |
8237 | goto next; |
8238 | } | |
8239 | ||
9c17f6cd | 8240 | wait_for_completion_io(&done.done); |
8b110e39 MX |
8241 | |
8242 | if (!done.uptodate) { | |
8243 | /* We might have another mirror, so try again */ | |
8244 | goto try_again; | |
8245 | } | |
8246 | next: | |
2dabb324 CR |
8247 | offset += sectorsize; |
8248 | start += sectorsize; | |
8249 | ||
8250 | ASSERT(nr_sectors); | |
8251 | ||
97bf5a55 LB |
8252 | nr_sectors--; |
8253 | if (nr_sectors) { | |
2dabb324 | 8254 | pgoff += sectorsize; |
97bf5a55 | 8255 | ASSERT(pgoff < PAGE_SIZE); |
2dabb324 CR |
8256 | goto next_block; |
8257 | } | |
2c30c71b | 8258 | } |
c1dc0896 MX |
8259 | |
8260 | return err; | |
8261 | } | |
8262 | ||
4e4cbee9 CH |
8263 | static blk_status_t btrfs_subio_endio_read(struct inode *inode, |
8264 | struct btrfs_io_bio *io_bio, blk_status_t err) | |
8b110e39 MX |
8265 | { |
8266 | bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
8267 | ||
8268 | if (skip_csum) { | |
8269 | if (unlikely(err)) | |
8270 | return __btrfs_correct_data_nocsum(inode, io_bio); | |
8271 | else | |
58efbc9f | 8272 | return BLK_STS_OK; |
8b110e39 MX |
8273 | } else { |
8274 | return __btrfs_subio_endio_read(inode, io_bio, err); | |
8275 | } | |
8276 | } | |
8277 | ||
4246a0b6 | 8278 | static void btrfs_endio_direct_read(struct bio *bio) |
c1dc0896 MX |
8279 | { |
8280 | struct btrfs_dio_private *dip = bio->bi_private; | |
8281 | struct inode *inode = dip->inode; | |
8282 | struct bio *dio_bio; | |
8283 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
4e4cbee9 | 8284 | blk_status_t err = bio->bi_status; |
c1dc0896 | 8285 | |
99c4e3b9 | 8286 | if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED) |
8b110e39 | 8287 | err = btrfs_subio_endio_read(inode, io_bio, err); |
c1dc0896 | 8288 | |
4b46fce2 | 8289 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, |
d0082371 | 8290 | dip->logical_offset + dip->bytes - 1); |
9be3395b | 8291 | dio_bio = dip->dio_bio; |
4b46fce2 | 8292 | |
4b46fce2 | 8293 | kfree(dip); |
c0da7aa1 | 8294 | |
99c4e3b9 | 8295 | dio_bio->bi_status = err; |
4055351c | 8296 | dio_end_io(dio_bio); |
23ea8e5a MX |
8297 | |
8298 | if (io_bio->end_io) | |
4e4cbee9 | 8299 | io_bio->end_io(io_bio, blk_status_to_errno(err)); |
9be3395b | 8300 | bio_put(bio); |
4b46fce2 JB |
8301 | } |
8302 | ||
52427260 QW |
8303 | static void __endio_write_update_ordered(struct inode *inode, |
8304 | const u64 offset, const u64 bytes, | |
8305 | const bool uptodate) | |
4b46fce2 | 8306 | { |
0b246afa | 8307 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 8308 | struct btrfs_ordered_extent *ordered = NULL; |
52427260 QW |
8309 | struct btrfs_workqueue *wq; |
8310 | btrfs_work_func_t func; | |
14543774 FM |
8311 | u64 ordered_offset = offset; |
8312 | u64 ordered_bytes = bytes; | |
67c003f9 | 8313 | u64 last_offset; |
4b46fce2 JB |
8314 | int ret; |
8315 | ||
52427260 QW |
8316 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) { |
8317 | wq = fs_info->endio_freespace_worker; | |
8318 | func = btrfs_freespace_write_helper; | |
8319 | } else { | |
8320 | wq = fs_info->endio_write_workers; | |
8321 | func = btrfs_endio_write_helper; | |
8322 | } | |
8323 | ||
163cf09c | 8324 | again: |
67c003f9 | 8325 | last_offset = ordered_offset; |
163cf09c CM |
8326 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, |
8327 | &ordered_offset, | |
4246a0b6 | 8328 | ordered_bytes, |
14543774 | 8329 | uptodate); |
4b46fce2 | 8330 | if (!ret) |
163cf09c | 8331 | goto out_test; |
4b46fce2 | 8332 | |
52427260 QW |
8333 | btrfs_init_work(&ordered->work, func, finish_ordered_fn, NULL, NULL); |
8334 | btrfs_queue_work(wq, &ordered->work); | |
163cf09c | 8335 | out_test: |
67c003f9 NA |
8336 | /* |
8337 | * If btrfs_dec_test_ordered_pending does not find any ordered extent | |
8338 | * in the range, we can exit. | |
8339 | */ | |
8340 | if (ordered_offset == last_offset) | |
8341 | return; | |
163cf09c CM |
8342 | /* |
8343 | * our bio might span multiple ordered extents. If we haven't | |
8344 | * completed the accounting for the whole dio, go back and try again | |
8345 | */ | |
14543774 FM |
8346 | if (ordered_offset < offset + bytes) { |
8347 | ordered_bytes = offset + bytes - ordered_offset; | |
5fd02043 | 8348 | ordered = NULL; |
163cf09c CM |
8349 | goto again; |
8350 | } | |
14543774 FM |
8351 | } |
8352 | ||
8353 | static void btrfs_endio_direct_write(struct bio *bio) | |
8354 | { | |
8355 | struct btrfs_dio_private *dip = bio->bi_private; | |
8356 | struct bio *dio_bio = dip->dio_bio; | |
8357 | ||
52427260 | 8358 | __endio_write_update_ordered(dip->inode, dip->logical_offset, |
4e4cbee9 | 8359 | dip->bytes, !bio->bi_status); |
4b46fce2 | 8360 | |
4b46fce2 | 8361 | kfree(dip); |
c0da7aa1 | 8362 | |
4e4cbee9 | 8363 | dio_bio->bi_status = bio->bi_status; |
4055351c | 8364 | dio_end_io(dio_bio); |
9be3395b | 8365 | bio_put(bio); |
4b46fce2 JB |
8366 | } |
8367 | ||
8c27cb35 | 8368 | static blk_status_t __btrfs_submit_bio_start_direct_io(void *private_data, |
eaf25d93 CM |
8369 | struct bio *bio, int mirror_num, |
8370 | unsigned long bio_flags, u64 offset) | |
8371 | { | |
c6100a4b | 8372 | struct inode *inode = private_data; |
4e4cbee9 | 8373 | blk_status_t ret; |
2ff7e61e | 8374 | ret = btrfs_csum_one_bio(inode, bio, offset, 1); |
79787eaa | 8375 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
8376 | return 0; |
8377 | } | |
8378 | ||
4246a0b6 | 8379 | static void btrfs_end_dio_bio(struct bio *bio) |
e65e1535 MX |
8380 | { |
8381 | struct btrfs_dio_private *dip = bio->bi_private; | |
4e4cbee9 | 8382 | blk_status_t err = bio->bi_status; |
e65e1535 | 8383 | |
8b110e39 MX |
8384 | if (err) |
8385 | btrfs_warn(BTRFS_I(dip->inode)->root->fs_info, | |
6296b960 | 8386 | "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d", |
f85b7379 DS |
8387 | btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio), |
8388 | bio->bi_opf, | |
8b110e39 MX |
8389 | (unsigned long long)bio->bi_iter.bi_sector, |
8390 | bio->bi_iter.bi_size, err); | |
8391 | ||
8392 | if (dip->subio_endio) | |
8393 | err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err); | |
c1dc0896 MX |
8394 | |
8395 | if (err) { | |
e65e1535 MX |
8396 | dip->errors = 1; |
8397 | ||
8398 | /* | |
8399 | * before atomic variable goto zero, we must make sure | |
8400 | * dip->errors is perceived to be set. | |
8401 | */ | |
4e857c58 | 8402 | smp_mb__before_atomic(); |
e65e1535 MX |
8403 | } |
8404 | ||
8405 | /* if there are more bios still pending for this dio, just exit */ | |
8406 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
8407 | goto out; | |
8408 | ||
9be3395b | 8409 | if (dip->errors) { |
e65e1535 | 8410 | bio_io_error(dip->orig_bio); |
9be3395b | 8411 | } else { |
2dbe0c77 | 8412 | dip->dio_bio->bi_status = BLK_STS_OK; |
4246a0b6 | 8413 | bio_endio(dip->orig_bio); |
e65e1535 MX |
8414 | } |
8415 | out: | |
8416 | bio_put(bio); | |
8417 | } | |
8418 | ||
4e4cbee9 | 8419 | static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode, |
c1dc0896 MX |
8420 | struct btrfs_dio_private *dip, |
8421 | struct bio *bio, | |
8422 | u64 file_offset) | |
8423 | { | |
8424 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
8425 | struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio); | |
4e4cbee9 | 8426 | blk_status_t ret; |
c1dc0896 MX |
8427 | |
8428 | /* | |
8429 | * We load all the csum data we need when we submit | |
8430 | * the first bio to reduce the csum tree search and | |
8431 | * contention. | |
8432 | */ | |
8433 | if (dip->logical_offset == file_offset) { | |
2ff7e61e | 8434 | ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio, |
c1dc0896 MX |
8435 | file_offset); |
8436 | if (ret) | |
8437 | return ret; | |
8438 | } | |
8439 | ||
8440 | if (bio == dip->orig_bio) | |
8441 | return 0; | |
8442 | ||
8443 | file_offset -= dip->logical_offset; | |
8444 | file_offset >>= inode->i_sb->s_blocksize_bits; | |
8445 | io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset); | |
8446 | ||
8447 | return 0; | |
8448 | } | |
8449 | ||
58efbc9f OS |
8450 | static inline blk_status_t |
8451 | __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, u64 file_offset, | |
66ba772e | 8452 | int async_submit) |
e65e1535 | 8453 | { |
0b246afa | 8454 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
facc8a22 | 8455 | struct btrfs_dio_private *dip = bio->bi_private; |
37226b21 | 8456 | bool write = bio_op(bio) == REQ_OP_WRITE; |
4e4cbee9 | 8457 | blk_status_t ret; |
e65e1535 | 8458 | |
b812ce28 JB |
8459 | if (async_submit) |
8460 | async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers); | |
8461 | ||
e65e1535 | 8462 | bio_get(bio); |
5fd02043 JB |
8463 | |
8464 | if (!write) { | |
0b246afa | 8465 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA); |
5fd02043 JB |
8466 | if (ret) |
8467 | goto err; | |
8468 | } | |
e65e1535 | 8469 | |
e6961cac | 8470 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) |
1ae39938 JB |
8471 | goto map; |
8472 | ||
8473 | if (write && async_submit) { | |
c6100a4b JB |
8474 | ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0, |
8475 | file_offset, inode, | |
0b246afa JM |
8476 | __btrfs_submit_bio_start_direct_io, |
8477 | __btrfs_submit_bio_done); | |
e65e1535 | 8478 | goto err; |
1ae39938 JB |
8479 | } else if (write) { |
8480 | /* | |
8481 | * If we aren't doing async submit, calculate the csum of the | |
8482 | * bio now. | |
8483 | */ | |
2ff7e61e | 8484 | ret = btrfs_csum_one_bio(inode, bio, file_offset, 1); |
1ae39938 JB |
8485 | if (ret) |
8486 | goto err; | |
23ea8e5a | 8487 | } else { |
2ff7e61e | 8488 | ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio, |
c1dc0896 | 8489 | file_offset); |
c2db1073 TI |
8490 | if (ret) |
8491 | goto err; | |
8492 | } | |
1ae39938 | 8493 | map: |
9b4a9b28 | 8494 | ret = btrfs_map_bio(fs_info, bio, 0, 0); |
e65e1535 MX |
8495 | err: |
8496 | bio_put(bio); | |
8497 | return ret; | |
8498 | } | |
8499 | ||
e6961cac | 8500 | static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip) |
e65e1535 MX |
8501 | { |
8502 | struct inode *inode = dip->inode; | |
0b246afa | 8503 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e65e1535 MX |
8504 | struct bio *bio; |
8505 | struct bio *orig_bio = dip->orig_bio; | |
4f024f37 | 8506 | u64 start_sector = orig_bio->bi_iter.bi_sector; |
e65e1535 | 8507 | u64 file_offset = dip->logical_offset; |
e65e1535 | 8508 | u64 map_length; |
1ae39938 | 8509 | int async_submit = 0; |
725130ba LB |
8510 | u64 submit_len; |
8511 | int clone_offset = 0; | |
8512 | int clone_len; | |
5f4dc8fc | 8513 | int ret; |
58efbc9f | 8514 | blk_status_t status; |
e65e1535 | 8515 | |
4f024f37 | 8516 | map_length = orig_bio->bi_iter.bi_size; |
725130ba | 8517 | submit_len = map_length; |
0b246afa JM |
8518 | ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9, |
8519 | &map_length, NULL, 0); | |
7a5c3c9b | 8520 | if (ret) |
e65e1535 | 8521 | return -EIO; |
facc8a22 | 8522 | |
725130ba | 8523 | if (map_length >= submit_len) { |
02f57c7a | 8524 | bio = orig_bio; |
c1dc0896 | 8525 | dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED; |
02f57c7a JB |
8526 | goto submit; |
8527 | } | |
8528 | ||
53b381b3 | 8529 | /* async crcs make it difficult to collect full stripe writes. */ |
1b86826d | 8530 | if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 DW |
8531 | async_submit = 0; |
8532 | else | |
8533 | async_submit = 1; | |
8534 | ||
725130ba LB |
8535 | /* bio split */ |
8536 | ASSERT(map_length <= INT_MAX); | |
02f57c7a | 8537 | atomic_inc(&dip->pending_bios); |
3c91ee69 | 8538 | do { |
725130ba | 8539 | clone_len = min_t(int, submit_len, map_length); |
02f57c7a | 8540 | |
725130ba LB |
8541 | /* |
8542 | * This will never fail as it's passing GPF_NOFS and | |
8543 | * the allocation is backed by btrfs_bioset. | |
8544 | */ | |
e477094f | 8545 | bio = btrfs_bio_clone_partial(orig_bio, clone_offset, |
725130ba LB |
8546 | clone_len); |
8547 | bio->bi_private = dip; | |
8548 | bio->bi_end_io = btrfs_end_dio_bio; | |
8549 | btrfs_io_bio(bio)->logical = file_offset; | |
8550 | ||
8551 | ASSERT(submit_len >= clone_len); | |
8552 | submit_len -= clone_len; | |
8553 | if (submit_len == 0) | |
8554 | break; | |
e65e1535 | 8555 | |
725130ba LB |
8556 | /* |
8557 | * Increase the count before we submit the bio so we know | |
8558 | * the end IO handler won't happen before we increase the | |
8559 | * count. Otherwise, the dip might get freed before we're | |
8560 | * done setting it up. | |
8561 | */ | |
8562 | atomic_inc(&dip->pending_bios); | |
e65e1535 | 8563 | |
66ba772e | 8564 | status = __btrfs_submit_dio_bio(bio, inode, file_offset, |
58efbc9f OS |
8565 | async_submit); |
8566 | if (status) { | |
725130ba LB |
8567 | bio_put(bio); |
8568 | atomic_dec(&dip->pending_bios); | |
8569 | goto out_err; | |
8570 | } | |
e65e1535 | 8571 | |
725130ba LB |
8572 | clone_offset += clone_len; |
8573 | start_sector += clone_len >> 9; | |
8574 | file_offset += clone_len; | |
5f4dc8fc | 8575 | |
725130ba LB |
8576 | map_length = submit_len; |
8577 | ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), | |
8578 | start_sector << 9, &map_length, NULL, 0); | |
8579 | if (ret) | |
8580 | goto out_err; | |
3c91ee69 | 8581 | } while (submit_len > 0); |
e65e1535 | 8582 | |
02f57c7a | 8583 | submit: |
66ba772e | 8584 | status = __btrfs_submit_dio_bio(bio, inode, file_offset, async_submit); |
58efbc9f | 8585 | if (!status) |
e65e1535 MX |
8586 | return 0; |
8587 | ||
8588 | bio_put(bio); | |
8589 | out_err: | |
8590 | dip->errors = 1; | |
8591 | /* | |
8592 | * before atomic variable goto zero, we must | |
8593 | * make sure dip->errors is perceived to be set. | |
8594 | */ | |
4e857c58 | 8595 | smp_mb__before_atomic(); |
e65e1535 MX |
8596 | if (atomic_dec_and_test(&dip->pending_bios)) |
8597 | bio_io_error(dip->orig_bio); | |
8598 | ||
8599 | /* bio_end_io() will handle error, so we needn't return it */ | |
8600 | return 0; | |
8601 | } | |
8602 | ||
8a4c1e42 MC |
8603 | static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode, |
8604 | loff_t file_offset) | |
4b46fce2 | 8605 | { |
61de718f | 8606 | struct btrfs_dio_private *dip = NULL; |
3892ac90 LB |
8607 | struct bio *bio = NULL; |
8608 | struct btrfs_io_bio *io_bio; | |
8a4c1e42 | 8609 | bool write = (bio_op(dio_bio) == REQ_OP_WRITE); |
4b46fce2 JB |
8610 | int ret = 0; |
8611 | ||
8b6c1d56 | 8612 | bio = btrfs_bio_clone(dio_bio); |
9be3395b | 8613 | |
c1dc0896 | 8614 | dip = kzalloc(sizeof(*dip), GFP_NOFS); |
4b46fce2 JB |
8615 | if (!dip) { |
8616 | ret = -ENOMEM; | |
61de718f | 8617 | goto free_ordered; |
4b46fce2 | 8618 | } |
4b46fce2 | 8619 | |
9be3395b | 8620 | dip->private = dio_bio->bi_private; |
4b46fce2 JB |
8621 | dip->inode = inode; |
8622 | dip->logical_offset = file_offset; | |
4f024f37 KO |
8623 | dip->bytes = dio_bio->bi_iter.bi_size; |
8624 | dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9; | |
3892ac90 LB |
8625 | bio->bi_private = dip; |
8626 | dip->orig_bio = bio; | |
9be3395b | 8627 | dip->dio_bio = dio_bio; |
e65e1535 | 8628 | atomic_set(&dip->pending_bios, 0); |
3892ac90 LB |
8629 | io_bio = btrfs_io_bio(bio); |
8630 | io_bio->logical = file_offset; | |
4b46fce2 | 8631 | |
c1dc0896 | 8632 | if (write) { |
3892ac90 | 8633 | bio->bi_end_io = btrfs_endio_direct_write; |
c1dc0896 | 8634 | } else { |
3892ac90 | 8635 | bio->bi_end_io = btrfs_endio_direct_read; |
c1dc0896 MX |
8636 | dip->subio_endio = btrfs_subio_endio_read; |
8637 | } | |
4b46fce2 | 8638 | |
f28a4928 FM |
8639 | /* |
8640 | * Reset the range for unsubmitted ordered extents (to a 0 length range) | |
8641 | * even if we fail to submit a bio, because in such case we do the | |
8642 | * corresponding error handling below and it must not be done a second | |
8643 | * time by btrfs_direct_IO(). | |
8644 | */ | |
8645 | if (write) { | |
8646 | struct btrfs_dio_data *dio_data = current->journal_info; | |
8647 | ||
8648 | dio_data->unsubmitted_oe_range_end = dip->logical_offset + | |
8649 | dip->bytes; | |
8650 | dio_data->unsubmitted_oe_range_start = | |
8651 | dio_data->unsubmitted_oe_range_end; | |
8652 | } | |
8653 | ||
e6961cac | 8654 | ret = btrfs_submit_direct_hook(dip); |
e65e1535 | 8655 | if (!ret) |
eaf25d93 | 8656 | return; |
9be3395b | 8657 | |
3892ac90 LB |
8658 | if (io_bio->end_io) |
8659 | io_bio->end_io(io_bio, ret); | |
9be3395b | 8660 | |
4b46fce2 JB |
8661 | free_ordered: |
8662 | /* | |
61de718f FM |
8663 | * If we arrived here it means either we failed to submit the dip |
8664 | * or we either failed to clone the dio_bio or failed to allocate the | |
8665 | * dip. If we cloned the dio_bio and allocated the dip, we can just | |
8666 | * call bio_endio against our io_bio so that we get proper resource | |
8667 | * cleanup if we fail to submit the dip, otherwise, we must do the | |
8668 | * same as btrfs_endio_direct_[write|read] because we can't call these | |
8669 | * callbacks - they require an allocated dip and a clone of dio_bio. | |
4b46fce2 | 8670 | */ |
3892ac90 | 8671 | if (bio && dip) { |
054ec2f6 | 8672 | bio_io_error(bio); |
61de718f | 8673 | /* |
3892ac90 | 8674 | * The end io callbacks free our dip, do the final put on bio |
61de718f FM |
8675 | * and all the cleanup and final put for dio_bio (through |
8676 | * dio_end_io()). | |
8677 | */ | |
8678 | dip = NULL; | |
3892ac90 | 8679 | bio = NULL; |
61de718f | 8680 | } else { |
14543774 | 8681 | if (write) |
52427260 | 8682 | __endio_write_update_ordered(inode, |
14543774 FM |
8683 | file_offset, |
8684 | dio_bio->bi_iter.bi_size, | |
52427260 | 8685 | false); |
14543774 | 8686 | else |
61de718f FM |
8687 | unlock_extent(&BTRFS_I(inode)->io_tree, file_offset, |
8688 | file_offset + dio_bio->bi_iter.bi_size - 1); | |
14543774 | 8689 | |
4e4cbee9 | 8690 | dio_bio->bi_status = BLK_STS_IOERR; |
61de718f FM |
8691 | /* |
8692 | * Releases and cleans up our dio_bio, no need to bio_put() | |
8693 | * nor bio_endio()/bio_io_error() against dio_bio. | |
8694 | */ | |
4055351c | 8695 | dio_end_io(dio_bio); |
4b46fce2 | 8696 | } |
3892ac90 LB |
8697 | if (bio) |
8698 | bio_put(bio); | |
61de718f | 8699 | kfree(dip); |
4b46fce2 JB |
8700 | } |
8701 | ||
2ff7e61e | 8702 | static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info, |
2ff7e61e | 8703 | const struct iov_iter *iter, loff_t offset) |
5a5f79b5 CM |
8704 | { |
8705 | int seg; | |
a1b75f7d | 8706 | int i; |
0b246afa | 8707 | unsigned int blocksize_mask = fs_info->sectorsize - 1; |
5a5f79b5 | 8708 | ssize_t retval = -EINVAL; |
5a5f79b5 CM |
8709 | |
8710 | if (offset & blocksize_mask) | |
8711 | goto out; | |
8712 | ||
28060d5d AV |
8713 | if (iov_iter_alignment(iter) & blocksize_mask) |
8714 | goto out; | |
a1b75f7d | 8715 | |
28060d5d | 8716 | /* If this is a write we don't need to check anymore */ |
cd27e455 | 8717 | if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter)) |
28060d5d AV |
8718 | return 0; |
8719 | /* | |
8720 | * Check to make sure we don't have duplicate iov_base's in this | |
8721 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
8722 | * when reading back. | |
8723 | */ | |
8724 | for (seg = 0; seg < iter->nr_segs; seg++) { | |
8725 | for (i = seg + 1; i < iter->nr_segs; i++) { | |
8726 | if (iter->iov[seg].iov_base == iter->iov[i].iov_base) | |
a1b75f7d JB |
8727 | goto out; |
8728 | } | |
5a5f79b5 CM |
8729 | } |
8730 | retval = 0; | |
8731 | out: | |
8732 | return retval; | |
8733 | } | |
eb838e73 | 8734 | |
c8b8e32d | 8735 | static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
16432985 | 8736 | { |
4b46fce2 JB |
8737 | struct file *file = iocb->ki_filp; |
8738 | struct inode *inode = file->f_mapping->host; | |
0b246afa | 8739 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
50745b0a | 8740 | struct btrfs_dio_data dio_data = { 0 }; |
364ecf36 | 8741 | struct extent_changeset *data_reserved = NULL; |
c8b8e32d | 8742 | loff_t offset = iocb->ki_pos; |
0934856d | 8743 | size_t count = 0; |
2e60a51e | 8744 | int flags = 0; |
38851cc1 MX |
8745 | bool wakeup = true; |
8746 | bool relock = false; | |
0934856d | 8747 | ssize_t ret; |
4b46fce2 | 8748 | |
8c70c9f8 | 8749 | if (check_direct_IO(fs_info, iter, offset)) |
5a5f79b5 | 8750 | return 0; |
3f7c579c | 8751 | |
fe0f07d0 | 8752 | inode_dio_begin(inode); |
38851cc1 | 8753 | |
0e267c44 | 8754 | /* |
41bd9ca4 MX |
8755 | * The generic stuff only does filemap_write_and_wait_range, which |
8756 | * isn't enough if we've written compressed pages to this area, so | |
8757 | * we need to flush the dirty pages again to make absolutely sure | |
8758 | * that any outstanding dirty pages are on disk. | |
0e267c44 | 8759 | */ |
a6cbcd4a | 8760 | count = iov_iter_count(iter); |
41bd9ca4 MX |
8761 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, |
8762 | &BTRFS_I(inode)->runtime_flags)) | |
9a025a08 WS |
8763 | filemap_fdatawrite_range(inode->i_mapping, offset, |
8764 | offset + count - 1); | |
0e267c44 | 8765 | |
6f673763 | 8766 | if (iov_iter_rw(iter) == WRITE) { |
38851cc1 MX |
8767 | /* |
8768 | * If the write DIO is beyond the EOF, we need update | |
8769 | * the isize, but it is protected by i_mutex. So we can | |
8770 | * not unlock the i_mutex at this case. | |
8771 | */ | |
8772 | if (offset + count <= inode->i_size) { | |
4aaedfb0 | 8773 | dio_data.overwrite = 1; |
5955102c | 8774 | inode_unlock(inode); |
38851cc1 | 8775 | relock = true; |
edf064e7 GR |
8776 | } else if (iocb->ki_flags & IOCB_NOWAIT) { |
8777 | ret = -EAGAIN; | |
8778 | goto out; | |
38851cc1 | 8779 | } |
364ecf36 QW |
8780 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, |
8781 | offset, count); | |
0934856d | 8782 | if (ret) |
38851cc1 | 8783 | goto out; |
e1cbbfa5 JB |
8784 | |
8785 | /* | |
8786 | * We need to know how many extents we reserved so that we can | |
8787 | * do the accounting properly if we go over the number we | |
8788 | * originally calculated. Abuse current->journal_info for this. | |
8789 | */ | |
da17066c | 8790 | dio_data.reserve = round_up(count, |
0b246afa | 8791 | fs_info->sectorsize); |
f28a4928 FM |
8792 | dio_data.unsubmitted_oe_range_start = (u64)offset; |
8793 | dio_data.unsubmitted_oe_range_end = (u64)offset; | |
50745b0a | 8794 | current->journal_info = &dio_data; |
97dcdea0 | 8795 | down_read(&BTRFS_I(inode)->dio_sem); |
ee39b432 DS |
8796 | } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK, |
8797 | &BTRFS_I(inode)->runtime_flags)) { | |
fe0f07d0 | 8798 | inode_dio_end(inode); |
38851cc1 MX |
8799 | flags = DIO_LOCKING | DIO_SKIP_HOLES; |
8800 | wakeup = false; | |
0934856d MX |
8801 | } |
8802 | ||
17f8c842 | 8803 | ret = __blockdev_direct_IO(iocb, inode, |
0b246afa | 8804 | fs_info->fs_devices->latest_bdev, |
c8b8e32d | 8805 | iter, btrfs_get_blocks_direct, NULL, |
17f8c842 | 8806 | btrfs_submit_direct, flags); |
6f673763 | 8807 | if (iov_iter_rw(iter) == WRITE) { |
97dcdea0 | 8808 | up_read(&BTRFS_I(inode)->dio_sem); |
e1cbbfa5 | 8809 | current->journal_info = NULL; |
ddba1bfc | 8810 | if (ret < 0 && ret != -EIOCBQUEUED) { |
50745b0a | 8811 | if (dio_data.reserve) |
bc42bda2 QW |
8812 | btrfs_delalloc_release_space(inode, data_reserved, |
8813 | offset, dio_data.reserve); | |
f28a4928 FM |
8814 | /* |
8815 | * On error we might have left some ordered extents | |
8816 | * without submitting corresponding bios for them, so | |
8817 | * cleanup them up to avoid other tasks getting them | |
8818 | * and waiting for them to complete forever. | |
8819 | */ | |
8820 | if (dio_data.unsubmitted_oe_range_start < | |
8821 | dio_data.unsubmitted_oe_range_end) | |
52427260 | 8822 | __endio_write_update_ordered(inode, |
f28a4928 FM |
8823 | dio_data.unsubmitted_oe_range_start, |
8824 | dio_data.unsubmitted_oe_range_end - | |
8825 | dio_data.unsubmitted_oe_range_start, | |
52427260 | 8826 | false); |
ddba1bfc | 8827 | } else if (ret >= 0 && (size_t)ret < count) |
bc42bda2 QW |
8828 | btrfs_delalloc_release_space(inode, data_reserved, |
8829 | offset, count - (size_t)ret); | |
69fe2d75 | 8830 | btrfs_delalloc_release_extents(BTRFS_I(inode), count); |
0934856d | 8831 | } |
38851cc1 | 8832 | out: |
2e60a51e | 8833 | if (wakeup) |
fe0f07d0 | 8834 | inode_dio_end(inode); |
38851cc1 | 8835 | if (relock) |
5955102c | 8836 | inode_lock(inode); |
0934856d | 8837 | |
364ecf36 | 8838 | extent_changeset_free(data_reserved); |
0934856d | 8839 | return ret; |
16432985 CM |
8840 | } |
8841 | ||
05dadc09 TI |
8842 | #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC) |
8843 | ||
1506fcc8 YS |
8844 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
8845 | __u64 start, __u64 len) | |
8846 | { | |
05dadc09 TI |
8847 | int ret; |
8848 | ||
8849 | ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS); | |
8850 | if (ret) | |
8851 | return ret; | |
8852 | ||
ec29ed5b | 8853 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
8854 | } |
8855 | ||
a52d9a80 | 8856 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 8857 | { |
d1310b2e CM |
8858 | struct extent_io_tree *tree; |
8859 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 8860 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 8861 | } |
1832a6d5 | 8862 | |
a52d9a80 | 8863 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 8864 | { |
d1310b2e | 8865 | struct extent_io_tree *tree; |
be7bd730 JB |
8866 | struct inode *inode = page->mapping->host; |
8867 | int ret; | |
b888db2b CM |
8868 | |
8869 | if (current->flags & PF_MEMALLOC) { | |
8870 | redirty_page_for_writepage(wbc, page); | |
8871 | unlock_page(page); | |
8872 | return 0; | |
8873 | } | |
be7bd730 JB |
8874 | |
8875 | /* | |
8876 | * If we are under memory pressure we will call this directly from the | |
8877 | * VM, we need to make sure we have the inode referenced for the ordered | |
8878 | * extent. If not just return like we didn't do anything. | |
8879 | */ | |
8880 | if (!igrab(inode)) { | |
8881 | redirty_page_for_writepage(wbc, page); | |
8882 | return AOP_WRITEPAGE_ACTIVATE; | |
8883 | } | |
d1310b2e | 8884 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
be7bd730 JB |
8885 | ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
8886 | btrfs_add_delayed_iput(inode); | |
8887 | return ret; | |
9ebefb18 CM |
8888 | } |
8889 | ||
48a3b636 ES |
8890 | static int btrfs_writepages(struct address_space *mapping, |
8891 | struct writeback_control *wbc) | |
b293f02e | 8892 | { |
d1310b2e | 8893 | struct extent_io_tree *tree; |
771ed689 | 8894 | |
d1310b2e | 8895 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
8896 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
8897 | } | |
8898 | ||
3ab2fb5a CM |
8899 | static int |
8900 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
8901 | struct list_head *pages, unsigned nr_pages) | |
8902 | { | |
d1310b2e CM |
8903 | struct extent_io_tree *tree; |
8904 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
8905 | return extent_readpages(tree, mapping, pages, nr_pages, |
8906 | btrfs_get_extent); | |
8907 | } | |
e6dcd2dc | 8908 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 8909 | { |
d1310b2e CM |
8910 | struct extent_io_tree *tree; |
8911 | struct extent_map_tree *map; | |
a52d9a80 | 8912 | int ret; |
8c2383c3 | 8913 | |
d1310b2e CM |
8914 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
8915 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 8916 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
8917 | if (ret == 1) { |
8918 | ClearPagePrivate(page); | |
8919 | set_page_private(page, 0); | |
09cbfeaf | 8920 | put_page(page); |
39279cc3 | 8921 | } |
a52d9a80 | 8922 | return ret; |
39279cc3 CM |
8923 | } |
8924 | ||
e6dcd2dc CM |
8925 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
8926 | { | |
98509cfc CM |
8927 | if (PageWriteback(page) || PageDirty(page)) |
8928 | return 0; | |
3ba7ab22 | 8929 | return __btrfs_releasepage(page, gfp_flags); |
e6dcd2dc CM |
8930 | } |
8931 | ||
d47992f8 LC |
8932 | static void btrfs_invalidatepage(struct page *page, unsigned int offset, |
8933 | unsigned int length) | |
39279cc3 | 8934 | { |
5fd02043 | 8935 | struct inode *inode = page->mapping->host; |
d1310b2e | 8936 | struct extent_io_tree *tree; |
e6dcd2dc | 8937 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 8938 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 8939 | u64 page_start = page_offset(page); |
09cbfeaf | 8940 | u64 page_end = page_start + PAGE_SIZE - 1; |
dbfdb6d1 CR |
8941 | u64 start; |
8942 | u64 end; | |
131e404a | 8943 | int inode_evicting = inode->i_state & I_FREEING; |
39279cc3 | 8944 | |
8b62b72b CM |
8945 | /* |
8946 | * we have the page locked, so new writeback can't start, | |
8947 | * and the dirty bit won't be cleared while we are here. | |
8948 | * | |
8949 | * Wait for IO on this page so that we can safely clear | |
8950 | * the PagePrivate2 bit and do ordered accounting | |
8951 | */ | |
e6dcd2dc | 8952 | wait_on_page_writeback(page); |
8b62b72b | 8953 | |
5fd02043 | 8954 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
8955 | if (offset) { |
8956 | btrfs_releasepage(page, GFP_NOFS); | |
8957 | return; | |
8958 | } | |
131e404a FDBM |
8959 | |
8960 | if (!inode_evicting) | |
ff13db41 | 8961 | lock_extent_bits(tree, page_start, page_end, &cached_state); |
dbfdb6d1 CR |
8962 | again: |
8963 | start = page_start; | |
a776c6fa | 8964 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start, |
dbfdb6d1 | 8965 | page_end - start + 1); |
e6dcd2dc | 8966 | if (ordered) { |
dbfdb6d1 | 8967 | end = min(page_end, ordered->file_offset + ordered->len - 1); |
eb84ae03 CM |
8968 | /* |
8969 | * IO on this page will never be started, so we need | |
8970 | * to account for any ordered extents now | |
8971 | */ | |
131e404a | 8972 | if (!inode_evicting) |
dbfdb6d1 | 8973 | clear_extent_bit(tree, start, end, |
131e404a | 8974 | EXTENT_DIRTY | EXTENT_DELALLOC | |
a7e3b975 | 8975 | EXTENT_DELALLOC_NEW | |
131e404a FDBM |
8976 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
8977 | EXTENT_DEFRAG, 1, 0, &cached_state, | |
8978 | GFP_NOFS); | |
8b62b72b CM |
8979 | /* |
8980 | * whoever cleared the private bit is responsible | |
8981 | * for the finish_ordered_io | |
8982 | */ | |
77cef2ec JB |
8983 | if (TestClearPagePrivate2(page)) { |
8984 | struct btrfs_ordered_inode_tree *tree; | |
8985 | u64 new_len; | |
8986 | ||
8987 | tree = &BTRFS_I(inode)->ordered_tree; | |
8988 | ||
8989 | spin_lock_irq(&tree->lock); | |
8990 | set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags); | |
dbfdb6d1 | 8991 | new_len = start - ordered->file_offset; |
77cef2ec JB |
8992 | if (new_len < ordered->truncated_len) |
8993 | ordered->truncated_len = new_len; | |
8994 | spin_unlock_irq(&tree->lock); | |
8995 | ||
8996 | if (btrfs_dec_test_ordered_pending(inode, &ordered, | |
dbfdb6d1 CR |
8997 | start, |
8998 | end - start + 1, 1)) | |
77cef2ec | 8999 | btrfs_finish_ordered_io(ordered); |
8b62b72b | 9000 | } |
e6dcd2dc | 9001 | btrfs_put_ordered_extent(ordered); |
131e404a FDBM |
9002 | if (!inode_evicting) { |
9003 | cached_state = NULL; | |
dbfdb6d1 | 9004 | lock_extent_bits(tree, start, end, |
131e404a FDBM |
9005 | &cached_state); |
9006 | } | |
dbfdb6d1 CR |
9007 | |
9008 | start = end + 1; | |
9009 | if (start < page_end) | |
9010 | goto again; | |
131e404a FDBM |
9011 | } |
9012 | ||
b9d0b389 QW |
9013 | /* |
9014 | * Qgroup reserved space handler | |
9015 | * Page here will be either | |
9016 | * 1) Already written to disk | |
9017 | * In this case, its reserved space is released from data rsv map | |
9018 | * and will be freed by delayed_ref handler finally. | |
9019 | * So even we call qgroup_free_data(), it won't decrease reserved | |
9020 | * space. | |
9021 | * 2) Not written to disk | |
0b34c261 GR |
9022 | * This means the reserved space should be freed here. However, |
9023 | * if a truncate invalidates the page (by clearing PageDirty) | |
9024 | * and the page is accounted for while allocating extent | |
9025 | * in btrfs_check_data_free_space() we let delayed_ref to | |
9026 | * free the entire extent. | |
b9d0b389 | 9027 | */ |
0b34c261 | 9028 | if (PageDirty(page)) |
bc42bda2 | 9029 | btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE); |
131e404a FDBM |
9030 | if (!inode_evicting) { |
9031 | clear_extent_bit(tree, page_start, page_end, | |
9032 | EXTENT_LOCKED | EXTENT_DIRTY | | |
a7e3b975 FM |
9033 | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW | |
9034 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1, | |
131e404a FDBM |
9035 | &cached_state, GFP_NOFS); |
9036 | ||
9037 | __btrfs_releasepage(page, GFP_NOFS); | |
e6dcd2dc | 9038 | } |
e6dcd2dc | 9039 | |
4a096752 | 9040 | ClearPageChecked(page); |
9ad6b7bc | 9041 | if (PagePrivate(page)) { |
9ad6b7bc CM |
9042 | ClearPagePrivate(page); |
9043 | set_page_private(page, 0); | |
09cbfeaf | 9044 | put_page(page); |
9ad6b7bc | 9045 | } |
39279cc3 CM |
9046 | } |
9047 | ||
9ebefb18 CM |
9048 | /* |
9049 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
9050 | * called from a page fault handler when a page is first dirtied. Hence we must | |
9051 | * be careful to check for EOF conditions here. We set the page up correctly | |
9052 | * for a written page which means we get ENOSPC checking when writing into | |
9053 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
9054 | * support these features. | |
9055 | * | |
9056 | * We are not allowed to take the i_mutex here so we have to play games to | |
9057 | * protect against truncate races as the page could now be beyond EOF. Because | |
9058 | * vmtruncate() writes the inode size before removing pages, once we have the | |
9059 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
9060 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
9061 | * unlock the page. | |
9062 | */ | |
11bac800 | 9063 | int btrfs_page_mkwrite(struct vm_fault *vmf) |
9ebefb18 | 9064 | { |
c2ec175c | 9065 | struct page *page = vmf->page; |
11bac800 | 9066 | struct inode *inode = file_inode(vmf->vma->vm_file); |
0b246afa | 9067 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc CM |
9068 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
9069 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 9070 | struct extent_state *cached_state = NULL; |
364ecf36 | 9071 | struct extent_changeset *data_reserved = NULL; |
e6dcd2dc CM |
9072 | char *kaddr; |
9073 | unsigned long zero_start; | |
9ebefb18 | 9074 | loff_t size; |
1832a6d5 | 9075 | int ret; |
9998eb70 | 9076 | int reserved = 0; |
d0b7da88 | 9077 | u64 reserved_space; |
a52d9a80 | 9078 | u64 page_start; |
e6dcd2dc | 9079 | u64 page_end; |
d0b7da88 CR |
9080 | u64 end; |
9081 | ||
09cbfeaf | 9082 | reserved_space = PAGE_SIZE; |
9ebefb18 | 9083 | |
b2b5ef5c | 9084 | sb_start_pagefault(inode->i_sb); |
df480633 | 9085 | page_start = page_offset(page); |
09cbfeaf | 9086 | page_end = page_start + PAGE_SIZE - 1; |
d0b7da88 | 9087 | end = page_end; |
df480633 | 9088 | |
d0b7da88 CR |
9089 | /* |
9090 | * Reserving delalloc space after obtaining the page lock can lead to | |
9091 | * deadlock. For example, if a dirty page is locked by this function | |
9092 | * and the call to btrfs_delalloc_reserve_space() ends up triggering | |
9093 | * dirty page write out, then the btrfs_writepage() function could | |
9094 | * end up waiting indefinitely to get a lock on the page currently | |
9095 | * being processed by btrfs_page_mkwrite() function. | |
9096 | */ | |
364ecf36 | 9097 | ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start, |
d0b7da88 | 9098 | reserved_space); |
9998eb70 | 9099 | if (!ret) { |
11bac800 | 9100 | ret = file_update_time(vmf->vma->vm_file); |
9998eb70 CM |
9101 | reserved = 1; |
9102 | } | |
56a76f82 NP |
9103 | if (ret) { |
9104 | if (ret == -ENOMEM) | |
9105 | ret = VM_FAULT_OOM; | |
9106 | else /* -ENOSPC, -EIO, etc */ | |
9107 | ret = VM_FAULT_SIGBUS; | |
9998eb70 CM |
9108 | if (reserved) |
9109 | goto out; | |
9110 | goto out_noreserve; | |
56a76f82 | 9111 | } |
1832a6d5 | 9112 | |
56a76f82 | 9113 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 9114 | again: |
9ebefb18 | 9115 | lock_page(page); |
9ebefb18 | 9116 | size = i_size_read(inode); |
a52d9a80 | 9117 | |
9ebefb18 | 9118 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 9119 | (page_start >= size)) { |
9ebefb18 CM |
9120 | /* page got truncated out from underneath us */ |
9121 | goto out_unlock; | |
9122 | } | |
e6dcd2dc CM |
9123 | wait_on_page_writeback(page); |
9124 | ||
ff13db41 | 9125 | lock_extent_bits(io_tree, page_start, page_end, &cached_state); |
e6dcd2dc CM |
9126 | set_page_extent_mapped(page); |
9127 | ||
eb84ae03 CM |
9128 | /* |
9129 | * we can't set the delalloc bits if there are pending ordered | |
9130 | * extents. Drop our locks and wait for them to finish | |
9131 | */ | |
a776c6fa NB |
9132 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start, |
9133 | PAGE_SIZE); | |
e6dcd2dc | 9134 | if (ordered) { |
2ac55d41 JB |
9135 | unlock_extent_cached(io_tree, page_start, page_end, |
9136 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 9137 | unlock_page(page); |
eb84ae03 | 9138 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
9139 | btrfs_put_ordered_extent(ordered); |
9140 | goto again; | |
9141 | } | |
9142 | ||
09cbfeaf | 9143 | if (page->index == ((size - 1) >> PAGE_SHIFT)) { |
da17066c | 9144 | reserved_space = round_up(size - page_start, |
0b246afa | 9145 | fs_info->sectorsize); |
09cbfeaf | 9146 | if (reserved_space < PAGE_SIZE) { |
d0b7da88 | 9147 | end = page_start + reserved_space - 1; |
bc42bda2 QW |
9148 | btrfs_delalloc_release_space(inode, data_reserved, |
9149 | page_start, PAGE_SIZE - reserved_space); | |
d0b7da88 CR |
9150 | } |
9151 | } | |
9152 | ||
fbf19087 | 9153 | /* |
5416034f LB |
9154 | * page_mkwrite gets called when the page is firstly dirtied after it's |
9155 | * faulted in, but write(2) could also dirty a page and set delalloc | |
9156 | * bits, thus in this case for space account reason, we still need to | |
9157 | * clear any delalloc bits within this page range since we have to | |
9158 | * reserve data&meta space before lock_page() (see above comments). | |
fbf19087 | 9159 | */ |
d0b7da88 | 9160 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end, |
9e8a4a8b LB |
9161 | EXTENT_DIRTY | EXTENT_DELALLOC | |
9162 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 9163 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 9164 | |
e3b8a485 | 9165 | ret = btrfs_set_extent_delalloc(inode, page_start, end, 0, |
ba8b04c1 | 9166 | &cached_state, 0); |
9ed74f2d | 9167 | if (ret) { |
2ac55d41 JB |
9168 | unlock_extent_cached(io_tree, page_start, page_end, |
9169 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
9170 | ret = VM_FAULT_SIGBUS; |
9171 | goto out_unlock; | |
9172 | } | |
e6dcd2dc | 9173 | ret = 0; |
9ebefb18 CM |
9174 | |
9175 | /* page is wholly or partially inside EOF */ | |
09cbfeaf KS |
9176 | if (page_start + PAGE_SIZE > size) |
9177 | zero_start = size & ~PAGE_MASK; | |
9ebefb18 | 9178 | else |
09cbfeaf | 9179 | zero_start = PAGE_SIZE; |
9ebefb18 | 9180 | |
09cbfeaf | 9181 | if (zero_start != PAGE_SIZE) { |
e6dcd2dc | 9182 | kaddr = kmap(page); |
09cbfeaf | 9183 | memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start); |
e6dcd2dc CM |
9184 | flush_dcache_page(page); |
9185 | kunmap(page); | |
9186 | } | |
247e743c | 9187 | ClearPageChecked(page); |
e6dcd2dc | 9188 | set_page_dirty(page); |
50a9b214 | 9189 | SetPageUptodate(page); |
5a3f23d5 | 9190 | |
0b246afa | 9191 | BTRFS_I(inode)->last_trans = fs_info->generation; |
257c62e1 | 9192 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; |
46d8bc34 | 9193 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit; |
257c62e1 | 9194 | |
2ac55d41 | 9195 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
9196 | |
9197 | out_unlock: | |
b2b5ef5c | 9198 | if (!ret) { |
8b62f87b | 9199 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); |
b2b5ef5c | 9200 | sb_end_pagefault(inode->i_sb); |
364ecf36 | 9201 | extent_changeset_free(data_reserved); |
50a9b214 | 9202 | return VM_FAULT_LOCKED; |
b2b5ef5c | 9203 | } |
9ebefb18 | 9204 | unlock_page(page); |
1832a6d5 | 9205 | out: |
8b62f87b | 9206 | btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); |
bc42bda2 QW |
9207 | btrfs_delalloc_release_space(inode, data_reserved, page_start, |
9208 | reserved_space); | |
9998eb70 | 9209 | out_noreserve: |
b2b5ef5c | 9210 | sb_end_pagefault(inode->i_sb); |
364ecf36 | 9211 | extent_changeset_free(data_reserved); |
9ebefb18 CM |
9212 | return ret; |
9213 | } | |
9214 | ||
a41ad394 | 9215 | static int btrfs_truncate(struct inode *inode) |
39279cc3 | 9216 | { |
0b246afa | 9217 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 9218 | struct btrfs_root *root = BTRFS_I(inode)->root; |
fcb80c2a | 9219 | struct btrfs_block_rsv *rsv; |
a71754fc | 9220 | int ret = 0; |
3893e33b | 9221 | int err = 0; |
39279cc3 | 9222 | struct btrfs_trans_handle *trans; |
0b246afa JM |
9223 | u64 mask = fs_info->sectorsize - 1; |
9224 | u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1); | |
39279cc3 | 9225 | |
0ef8b726 JB |
9226 | ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask), |
9227 | (u64)-1); | |
9228 | if (ret) | |
9229 | return ret; | |
39279cc3 | 9230 | |
fcb80c2a | 9231 | /* |
01327610 | 9232 | * Yes ladies and gentlemen, this is indeed ugly. The fact is we have |
fcb80c2a JB |
9233 | * 3 things going on here |
9234 | * | |
9235 | * 1) We need to reserve space for our orphan item and the space to | |
9236 | * delete our orphan item. Lord knows we don't want to have a dangling | |
9237 | * orphan item because we didn't reserve space to remove it. | |
9238 | * | |
9239 | * 2) We need to reserve space to update our inode. | |
9240 | * | |
9241 | * 3) We need to have something to cache all the space that is going to | |
9242 | * be free'd up by the truncate operation, but also have some slack | |
9243 | * space reserved in case it uses space during the truncate (thank you | |
9244 | * very much snapshotting). | |
9245 | * | |
01327610 | 9246 | * And we need these to all be separate. The fact is we can use a lot of |
fcb80c2a | 9247 | * space doing the truncate, and we have no earthly idea how much space |
01327610 | 9248 | * we will use, so we need the truncate reservation to be separate so it |
fcb80c2a JB |
9249 | * doesn't end up using space reserved for updating the inode or |
9250 | * removing the orphan item. We also need to be able to stop the | |
9251 | * transaction and start a new one, which means we need to be able to | |
9252 | * update the inode several times, and we have no idea of knowing how | |
9253 | * many times that will be, so we can't just reserve 1 item for the | |
01327610 | 9254 | * entirety of the operation, so that has to be done separately as well. |
fcb80c2a JB |
9255 | * Then there is the orphan item, which does indeed need to be held on |
9256 | * to for the whole operation, and we need nobody to touch this reserved | |
9257 | * space except the orphan code. | |
9258 | * | |
9259 | * So that leaves us with | |
9260 | * | |
9261 | * 1) root->orphan_block_rsv - for the orphan deletion. | |
9262 | * 2) rsv - for the truncate reservation, which we will steal from the | |
9263 | * transaction reservation. | |
9264 | * 3) fs_info->trans_block_rsv - this will have 1 items worth left for | |
9265 | * updating the inode. | |
9266 | */ | |
2ff7e61e | 9267 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
9268 | if (!rsv) |
9269 | return -ENOMEM; | |
4a338542 | 9270 | rsv->size = min_size; |
ca7e70f5 | 9271 | rsv->failfast = 1; |
f0cd846e | 9272 | |
907cbceb | 9273 | /* |
07127184 | 9274 | * 1 for the truncate slack space |
907cbceb JB |
9275 | * 1 for updating the inode. |
9276 | */ | |
f3fe820c | 9277 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a JB |
9278 | if (IS_ERR(trans)) { |
9279 | err = PTR_ERR(trans); | |
9280 | goto out; | |
9281 | } | |
f0cd846e | 9282 | |
907cbceb | 9283 | /* Migrate the slack space for the truncate to our reserve */ |
0b246afa | 9284 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv, |
25d609f8 | 9285 | min_size, 0); |
fcb80c2a | 9286 | BUG_ON(ret); |
f0cd846e | 9287 | |
5dc562c5 JB |
9288 | /* |
9289 | * So if we truncate and then write and fsync we normally would just | |
9290 | * write the extents that changed, which is a problem if we need to | |
9291 | * first truncate that entire inode. So set this flag so we write out | |
9292 | * all of the extents in the inode to the sync log so we're completely | |
9293 | * safe. | |
9294 | */ | |
9295 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
ca7e70f5 | 9296 | trans->block_rsv = rsv; |
907cbceb | 9297 | |
8082510e YZ |
9298 | while (1) { |
9299 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
9300 | inode->i_size, | |
9301 | BTRFS_EXTENT_DATA_KEY); | |
ddfae63c | 9302 | trans->block_rsv = &fs_info->trans_block_rsv; |
28ed1345 | 9303 | if (ret != -ENOSPC && ret != -EAGAIN) { |
3893e33b | 9304 | err = ret; |
8082510e | 9305 | break; |
3893e33b | 9306 | } |
39279cc3 | 9307 | |
8082510e | 9308 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
9309 | if (ret) { |
9310 | err = ret; | |
9311 | break; | |
9312 | } | |
ca7e70f5 | 9313 | |
3a45bb20 | 9314 | btrfs_end_transaction(trans); |
2ff7e61e | 9315 | btrfs_btree_balance_dirty(fs_info); |
ca7e70f5 JB |
9316 | |
9317 | trans = btrfs_start_transaction(root, 2); | |
9318 | if (IS_ERR(trans)) { | |
9319 | ret = err = PTR_ERR(trans); | |
9320 | trans = NULL; | |
9321 | break; | |
9322 | } | |
9323 | ||
47b5d646 | 9324 | btrfs_block_rsv_release(fs_info, rsv, -1); |
0b246afa | 9325 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, |
25d609f8 | 9326 | rsv, min_size, 0); |
ca7e70f5 JB |
9327 | BUG_ON(ret); /* shouldn't happen */ |
9328 | trans->block_rsv = rsv; | |
8082510e YZ |
9329 | } |
9330 | ||
ddfae63c JB |
9331 | /* |
9332 | * We can't call btrfs_truncate_block inside a trans handle as we could | |
9333 | * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know | |
9334 | * we've truncated everything except the last little bit, and can do | |
9335 | * btrfs_truncate_block and then update the disk_i_size. | |
9336 | */ | |
9337 | if (ret == NEED_TRUNCATE_BLOCK) { | |
9338 | btrfs_end_transaction(trans); | |
9339 | btrfs_btree_balance_dirty(fs_info); | |
9340 | ||
9341 | ret = btrfs_truncate_block(inode, inode->i_size, 0, 0); | |
9342 | if (ret) | |
9343 | goto out; | |
9344 | trans = btrfs_start_transaction(root, 1); | |
9345 | if (IS_ERR(trans)) { | |
9346 | ret = PTR_ERR(trans); | |
9347 | goto out; | |
9348 | } | |
9349 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); | |
9350 | } | |
9351 | ||
8082510e | 9352 | if (ret == 0 && inode->i_nlink > 0) { |
fcb80c2a | 9353 | trans->block_rsv = root->orphan_block_rsv; |
3d6ae7bb | 9354 | ret = btrfs_orphan_del(trans, BTRFS_I(inode)); |
3893e33b JB |
9355 | if (ret) |
9356 | err = ret; | |
8082510e YZ |
9357 | } |
9358 | ||
917c16b2 | 9359 | if (trans) { |
0b246afa | 9360 | trans->block_rsv = &fs_info->trans_block_rsv; |
917c16b2 CM |
9361 | ret = btrfs_update_inode(trans, root, inode); |
9362 | if (ret && !err) | |
9363 | err = ret; | |
7b128766 | 9364 | |
3a45bb20 | 9365 | ret = btrfs_end_transaction(trans); |
2ff7e61e | 9366 | btrfs_btree_balance_dirty(fs_info); |
917c16b2 | 9367 | } |
fcb80c2a | 9368 | out: |
2ff7e61e | 9369 | btrfs_free_block_rsv(fs_info, rsv); |
fcb80c2a | 9370 | |
3893e33b JB |
9371 | if (ret && !err) |
9372 | err = ret; | |
a41ad394 | 9373 | |
3893e33b | 9374 | return err; |
39279cc3 CM |
9375 | } |
9376 | ||
d352ac68 CM |
9377 | /* |
9378 | * create a new subvolume directory/inode (helper for the ioctl). | |
9379 | */ | |
d2fb3437 | 9380 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
63541927 FDBM |
9381 | struct btrfs_root *new_root, |
9382 | struct btrfs_root *parent_root, | |
9383 | u64 new_dirid) | |
39279cc3 | 9384 | { |
39279cc3 | 9385 | struct inode *inode; |
76dda93c | 9386 | int err; |
00e4e6b3 | 9387 | u64 index = 0; |
39279cc3 | 9388 | |
12fc9d09 FA |
9389 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
9390 | new_dirid, new_dirid, | |
9391 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
9392 | &index); | |
54aa1f4d | 9393 | if (IS_ERR(inode)) |
f46b5a66 | 9394 | return PTR_ERR(inode); |
39279cc3 CM |
9395 | inode->i_op = &btrfs_dir_inode_operations; |
9396 | inode->i_fop = &btrfs_dir_file_operations; | |
9397 | ||
bfe86848 | 9398 | set_nlink(inode, 1); |
6ef06d27 | 9399 | btrfs_i_size_write(BTRFS_I(inode), 0); |
b0d5d10f | 9400 | unlock_new_inode(inode); |
3b96362c | 9401 | |
63541927 FDBM |
9402 | err = btrfs_subvol_inherit_props(trans, new_root, parent_root); |
9403 | if (err) | |
9404 | btrfs_err(new_root->fs_info, | |
351fd353 | 9405 | "error inheriting subvolume %llu properties: %d", |
63541927 FDBM |
9406 | new_root->root_key.objectid, err); |
9407 | ||
76dda93c | 9408 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 9409 | |
76dda93c | 9410 | iput(inode); |
ce598979 | 9411 | return err; |
39279cc3 CM |
9412 | } |
9413 | ||
39279cc3 CM |
9414 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
9415 | { | |
69fe2d75 | 9416 | struct btrfs_fs_info *fs_info = btrfs_sb(sb); |
39279cc3 | 9417 | struct btrfs_inode *ei; |
2ead6ae7 | 9418 | struct inode *inode; |
39279cc3 CM |
9419 | |
9420 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
9421 | if (!ei) | |
9422 | return NULL; | |
2ead6ae7 YZ |
9423 | |
9424 | ei->root = NULL; | |
2ead6ae7 | 9425 | ei->generation = 0; |
15ee9bc7 | 9426 | ei->last_trans = 0; |
257c62e1 | 9427 | ei->last_sub_trans = 0; |
e02119d5 | 9428 | ei->logged_trans = 0; |
2ead6ae7 | 9429 | ei->delalloc_bytes = 0; |
a7e3b975 | 9430 | ei->new_delalloc_bytes = 0; |
47059d93 | 9431 | ei->defrag_bytes = 0; |
2ead6ae7 YZ |
9432 | ei->disk_i_size = 0; |
9433 | ei->flags = 0; | |
7709cde3 | 9434 | ei->csum_bytes = 0; |
2ead6ae7 | 9435 | ei->index_cnt = (u64)-1; |
67de1176 | 9436 | ei->dir_index = 0; |
2ead6ae7 | 9437 | ei->last_unlink_trans = 0; |
46d8bc34 | 9438 | ei->last_log_commit = 0; |
8089fe62 | 9439 | ei->delayed_iput_count = 0; |
2ead6ae7 | 9440 | |
9e0baf60 JB |
9441 | spin_lock_init(&ei->lock); |
9442 | ei->outstanding_extents = 0; | |
69fe2d75 JB |
9443 | if (sb->s_magic != BTRFS_TEST_MAGIC) |
9444 | btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv, | |
9445 | BTRFS_BLOCK_RSV_DELALLOC); | |
72ac3c0d | 9446 | ei->runtime_flags = 0; |
b52aa8c9 | 9447 | ei->prop_compress = BTRFS_COMPRESS_NONE; |
eec63c65 | 9448 | ei->defrag_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 9449 | |
16cdcec7 MX |
9450 | ei->delayed_node = NULL; |
9451 | ||
9cc97d64 | 9452 | ei->i_otime.tv_sec = 0; |
9453 | ei->i_otime.tv_nsec = 0; | |
9454 | ||
2ead6ae7 | 9455 | inode = &ei->vfs_inode; |
a8067e02 | 9456 | extent_map_tree_init(&ei->extent_tree); |
c6100a4b JB |
9457 | extent_io_tree_init(&ei->io_tree, inode); |
9458 | extent_io_tree_init(&ei->io_failure_tree, inode); | |
0b32f4bb JB |
9459 | ei->io_tree.track_uptodate = 1; |
9460 | ei->io_failure_tree.track_uptodate = 1; | |
b812ce28 | 9461 | atomic_set(&ei->sync_writers, 0); |
2ead6ae7 | 9462 | mutex_init(&ei->log_mutex); |
f248679e | 9463 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 9464 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 9465 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
8089fe62 | 9466 | INIT_LIST_HEAD(&ei->delayed_iput); |
2ead6ae7 | 9467 | RB_CLEAR_NODE(&ei->rb_node); |
5f9a8a51 | 9468 | init_rwsem(&ei->dio_sem); |
2ead6ae7 YZ |
9469 | |
9470 | return inode; | |
39279cc3 CM |
9471 | } |
9472 | ||
aaedb55b JB |
9473 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
9474 | void btrfs_test_destroy_inode(struct inode *inode) | |
9475 | { | |
dcdbc059 | 9476 | btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0); |
aaedb55b JB |
9477 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
9478 | } | |
9479 | #endif | |
9480 | ||
fa0d7e3d NP |
9481 | static void btrfs_i_callback(struct rcu_head *head) |
9482 | { | |
9483 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
9484 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
9485 | } | |
9486 | ||
39279cc3 CM |
9487 | void btrfs_destroy_inode(struct inode *inode) |
9488 | { | |
0b246afa | 9489 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc | 9490 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
9491 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9492 | ||
b3d9b7a3 | 9493 | WARN_ON(!hlist_empty(&inode->i_dentry)); |
39279cc3 | 9494 | WARN_ON(inode->i_data.nrpages); |
69fe2d75 JB |
9495 | WARN_ON(BTRFS_I(inode)->block_rsv.reserved); |
9496 | WARN_ON(BTRFS_I(inode)->block_rsv.size); | |
9e0baf60 | 9497 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
7709cde3 | 9498 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
a7e3b975 | 9499 | WARN_ON(BTRFS_I(inode)->new_delalloc_bytes); |
7709cde3 | 9500 | WARN_ON(BTRFS_I(inode)->csum_bytes); |
47059d93 | 9501 | WARN_ON(BTRFS_I(inode)->defrag_bytes); |
39279cc3 | 9502 | |
a6dbd429 JB |
9503 | /* |
9504 | * This can happen where we create an inode, but somebody else also | |
9505 | * created the same inode and we need to destroy the one we already | |
9506 | * created. | |
9507 | */ | |
9508 | if (!root) | |
9509 | goto free; | |
9510 | ||
8a35d95f JB |
9511 | if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
9512 | &BTRFS_I(inode)->runtime_flags)) { | |
0b246afa | 9513 | btrfs_info(fs_info, "inode %llu still on the orphan list", |
4a0cc7ca | 9514 | btrfs_ino(BTRFS_I(inode))); |
8a35d95f | 9515 | atomic_dec(&root->orphan_inodes); |
7b128766 | 9516 | } |
7b128766 | 9517 | |
d397712b | 9518 | while (1) { |
e6dcd2dc CM |
9519 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
9520 | if (!ordered) | |
9521 | break; | |
9522 | else { | |
0b246afa | 9523 | btrfs_err(fs_info, |
5d163e0e JM |
9524 | "found ordered extent %llu %llu on inode cleanup", |
9525 | ordered->file_offset, ordered->len); | |
e6dcd2dc CM |
9526 | btrfs_remove_ordered_extent(inode, ordered); |
9527 | btrfs_put_ordered_extent(ordered); | |
9528 | btrfs_put_ordered_extent(ordered); | |
9529 | } | |
9530 | } | |
56fa9d07 | 9531 | btrfs_qgroup_check_reserved_leak(inode); |
5d4f98a2 | 9532 | inode_tree_del(inode); |
dcdbc059 | 9533 | btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0); |
a6dbd429 | 9534 | free: |
fa0d7e3d | 9535 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
9536 | } |
9537 | ||
45321ac5 | 9538 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
9539 | { |
9540 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 9541 | |
6379ef9f NA |
9542 | if (root == NULL) |
9543 | return 1; | |
9544 | ||
fa6ac876 | 9545 | /* the snap/subvol tree is on deleting */ |
69e9c6c6 | 9546 | if (btrfs_root_refs(&root->root_item) == 0) |
45321ac5 | 9547 | return 1; |
76dda93c | 9548 | else |
45321ac5 | 9549 | return generic_drop_inode(inode); |
76dda93c YZ |
9550 | } |
9551 | ||
0ee0fda0 | 9552 | static void init_once(void *foo) |
39279cc3 CM |
9553 | { |
9554 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
9555 | ||
9556 | inode_init_once(&ei->vfs_inode); | |
9557 | } | |
9558 | ||
9559 | void btrfs_destroy_cachep(void) | |
9560 | { | |
8c0a8537 KS |
9561 | /* |
9562 | * Make sure all delayed rcu free inodes are flushed before we | |
9563 | * destroy cache. | |
9564 | */ | |
9565 | rcu_barrier(); | |
5598e900 KM |
9566 | kmem_cache_destroy(btrfs_inode_cachep); |
9567 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
5598e900 KM |
9568 | kmem_cache_destroy(btrfs_path_cachep); |
9569 | kmem_cache_destroy(btrfs_free_space_cachep); | |
39279cc3 CM |
9570 | } |
9571 | ||
f5c29bd9 | 9572 | int __init btrfs_init_cachep(void) |
39279cc3 | 9573 | { |
837e1972 | 9574 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 | 9575 | sizeof(struct btrfs_inode), 0, |
5d097056 VD |
9576 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT, |
9577 | init_once); | |
39279cc3 CM |
9578 | if (!btrfs_inode_cachep) |
9579 | goto fail; | |
9601e3f6 | 9580 | |
837e1972 | 9581 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 | 9582 | sizeof(struct btrfs_trans_handle), 0, |
fba4b697 | 9583 | SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
9584 | if (!btrfs_trans_handle_cachep) |
9585 | goto fail; | |
9601e3f6 | 9586 | |
837e1972 | 9587 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 | 9588 | sizeof(struct btrfs_path), 0, |
fba4b697 | 9589 | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
9590 | if (!btrfs_path_cachep) |
9591 | goto fail; | |
9601e3f6 | 9592 | |
837e1972 | 9593 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 | 9594 | sizeof(struct btrfs_free_space), 0, |
fba4b697 | 9595 | SLAB_MEM_SPREAD, NULL); |
dc89e982 JB |
9596 | if (!btrfs_free_space_cachep) |
9597 | goto fail; | |
9598 | ||
39279cc3 CM |
9599 | return 0; |
9600 | fail: | |
9601 | btrfs_destroy_cachep(); | |
9602 | return -ENOMEM; | |
9603 | } | |
9604 | ||
a528d35e DH |
9605 | static int btrfs_getattr(const struct path *path, struct kstat *stat, |
9606 | u32 request_mask, unsigned int flags) | |
39279cc3 | 9607 | { |
df0af1a5 | 9608 | u64 delalloc_bytes; |
a528d35e | 9609 | struct inode *inode = d_inode(path->dentry); |
fadc0d8b | 9610 | u32 blocksize = inode->i_sb->s_blocksize; |
04a87e34 YS |
9611 | u32 bi_flags = BTRFS_I(inode)->flags; |
9612 | ||
9613 | stat->result_mask |= STATX_BTIME; | |
9614 | stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec; | |
9615 | stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec; | |
9616 | if (bi_flags & BTRFS_INODE_APPEND) | |
9617 | stat->attributes |= STATX_ATTR_APPEND; | |
9618 | if (bi_flags & BTRFS_INODE_COMPRESS) | |
9619 | stat->attributes |= STATX_ATTR_COMPRESSED; | |
9620 | if (bi_flags & BTRFS_INODE_IMMUTABLE) | |
9621 | stat->attributes |= STATX_ATTR_IMMUTABLE; | |
9622 | if (bi_flags & BTRFS_INODE_NODUMP) | |
9623 | stat->attributes |= STATX_ATTR_NODUMP; | |
9624 | ||
9625 | stat->attributes_mask |= (STATX_ATTR_APPEND | | |
9626 | STATX_ATTR_COMPRESSED | | |
9627 | STATX_ATTR_IMMUTABLE | | |
9628 | STATX_ATTR_NODUMP); | |
fadc0d8b | 9629 | |
39279cc3 | 9630 | generic_fillattr(inode, stat); |
0ee5dc67 | 9631 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
df0af1a5 MX |
9632 | |
9633 | spin_lock(&BTRFS_I(inode)->lock); | |
a7e3b975 | 9634 | delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes; |
df0af1a5 | 9635 | spin_unlock(&BTRFS_I(inode)->lock); |
fadc0d8b | 9636 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
df0af1a5 | 9637 | ALIGN(delalloc_bytes, blocksize)) >> 9; |
39279cc3 CM |
9638 | return 0; |
9639 | } | |
9640 | ||
cdd1fedf DF |
9641 | static int btrfs_rename_exchange(struct inode *old_dir, |
9642 | struct dentry *old_dentry, | |
9643 | struct inode *new_dir, | |
9644 | struct dentry *new_dentry) | |
9645 | { | |
0b246afa | 9646 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
cdd1fedf DF |
9647 | struct btrfs_trans_handle *trans; |
9648 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
9649 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; | |
9650 | struct inode *new_inode = new_dentry->d_inode; | |
9651 | struct inode *old_inode = old_dentry->d_inode; | |
c2050a45 | 9652 | struct timespec ctime = current_time(old_inode); |
cdd1fedf | 9653 | struct dentry *parent; |
4a0cc7ca NB |
9654 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
9655 | u64 new_ino = btrfs_ino(BTRFS_I(new_inode)); | |
cdd1fedf DF |
9656 | u64 old_idx = 0; |
9657 | u64 new_idx = 0; | |
9658 | u64 root_objectid; | |
9659 | int ret; | |
86e8aa0e FM |
9660 | bool root_log_pinned = false; |
9661 | bool dest_log_pinned = false; | |
cdd1fedf DF |
9662 | |
9663 | /* we only allow rename subvolume link between subvolumes */ | |
9664 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) | |
9665 | return -EXDEV; | |
9666 | ||
9667 | /* close the race window with snapshot create/destroy ioctl */ | |
9668 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 9669 | down_read(&fs_info->subvol_sem); |
cdd1fedf | 9670 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9671 | down_read(&fs_info->subvol_sem); |
cdd1fedf DF |
9672 | |
9673 | /* | |
9674 | * We want to reserve the absolute worst case amount of items. So if | |
9675 | * both inodes are subvols and we need to unlink them then that would | |
9676 | * require 4 item modifications, but if they are both normal inodes it | |
9677 | * would require 5 item modifications, so we'll assume their normal | |
9678 | * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items | |
9679 | * should cover the worst case number of items we'll modify. | |
9680 | */ | |
9681 | trans = btrfs_start_transaction(root, 12); | |
9682 | if (IS_ERR(trans)) { | |
9683 | ret = PTR_ERR(trans); | |
9684 | goto out_notrans; | |
9685 | } | |
9686 | ||
9687 | /* | |
9688 | * We need to find a free sequence number both in the source and | |
9689 | * in the destination directory for the exchange. | |
9690 | */ | |
877574e2 | 9691 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx); |
cdd1fedf DF |
9692 | if (ret) |
9693 | goto out_fail; | |
877574e2 | 9694 | ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx); |
cdd1fedf DF |
9695 | if (ret) |
9696 | goto out_fail; | |
9697 | ||
9698 | BTRFS_I(old_inode)->dir_index = 0ULL; | |
9699 | BTRFS_I(new_inode)->dir_index = 0ULL; | |
9700 | ||
9701 | /* Reference for the source. */ | |
9702 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9703 | /* force full log commit if subvolume involved. */ | |
0b246afa | 9704 | btrfs_set_log_full_commit(fs_info, trans); |
cdd1fedf | 9705 | } else { |
376e5a57 FM |
9706 | btrfs_pin_log_trans(root); |
9707 | root_log_pinned = true; | |
cdd1fedf DF |
9708 | ret = btrfs_insert_inode_ref(trans, dest, |
9709 | new_dentry->d_name.name, | |
9710 | new_dentry->d_name.len, | |
9711 | old_ino, | |
f85b7379 DS |
9712 | btrfs_ino(BTRFS_I(new_dir)), |
9713 | old_idx); | |
cdd1fedf DF |
9714 | if (ret) |
9715 | goto out_fail; | |
cdd1fedf DF |
9716 | } |
9717 | ||
9718 | /* And now for the dest. */ | |
9719 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9720 | /* force full log commit if subvolume involved. */ | |
0b246afa | 9721 | btrfs_set_log_full_commit(fs_info, trans); |
cdd1fedf | 9722 | } else { |
376e5a57 FM |
9723 | btrfs_pin_log_trans(dest); |
9724 | dest_log_pinned = true; | |
cdd1fedf DF |
9725 | ret = btrfs_insert_inode_ref(trans, root, |
9726 | old_dentry->d_name.name, | |
9727 | old_dentry->d_name.len, | |
9728 | new_ino, | |
f85b7379 DS |
9729 | btrfs_ino(BTRFS_I(old_dir)), |
9730 | new_idx); | |
cdd1fedf DF |
9731 | if (ret) |
9732 | goto out_fail; | |
cdd1fedf DF |
9733 | } |
9734 | ||
9735 | /* Update inode version and ctime/mtime. */ | |
9736 | inode_inc_iversion(old_dir); | |
9737 | inode_inc_iversion(new_dir); | |
9738 | inode_inc_iversion(old_inode); | |
9739 | inode_inc_iversion(new_inode); | |
9740 | old_dir->i_ctime = old_dir->i_mtime = ctime; | |
9741 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
9742 | old_inode->i_ctime = ctime; | |
9743 | new_inode->i_ctime = ctime; | |
9744 | ||
9745 | if (old_dentry->d_parent != new_dentry->d_parent) { | |
f85b7379 DS |
9746 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
9747 | BTRFS_I(old_inode), 1); | |
9748 | btrfs_record_unlink_dir(trans, BTRFS_I(new_dir), | |
9749 | BTRFS_I(new_inode), 1); | |
cdd1fedf DF |
9750 | } |
9751 | ||
9752 | /* src is a subvolume */ | |
9753 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9754 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; | |
9755 | ret = btrfs_unlink_subvol(trans, root, old_dir, | |
9756 | root_objectid, | |
9757 | old_dentry->d_name.name, | |
9758 | old_dentry->d_name.len); | |
9759 | } else { /* src is an inode */ | |
4ec5934e NB |
9760 | ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir), |
9761 | BTRFS_I(old_dentry->d_inode), | |
cdd1fedf DF |
9762 | old_dentry->d_name.name, |
9763 | old_dentry->d_name.len); | |
9764 | if (!ret) | |
9765 | ret = btrfs_update_inode(trans, root, old_inode); | |
9766 | } | |
9767 | if (ret) { | |
66642832 | 9768 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9769 | goto out_fail; |
9770 | } | |
9771 | ||
9772 | /* dest is a subvolume */ | |
9773 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9774 | root_objectid = BTRFS_I(new_inode)->root->root_key.objectid; | |
9775 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
9776 | root_objectid, | |
9777 | new_dentry->d_name.name, | |
9778 | new_dentry->d_name.len); | |
9779 | } else { /* dest is an inode */ | |
4ec5934e NB |
9780 | ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir), |
9781 | BTRFS_I(new_dentry->d_inode), | |
cdd1fedf DF |
9782 | new_dentry->d_name.name, |
9783 | new_dentry->d_name.len); | |
9784 | if (!ret) | |
9785 | ret = btrfs_update_inode(trans, dest, new_inode); | |
9786 | } | |
9787 | if (ret) { | |
66642832 | 9788 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9789 | goto out_fail; |
9790 | } | |
9791 | ||
db0a669f | 9792 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
cdd1fedf DF |
9793 | new_dentry->d_name.name, |
9794 | new_dentry->d_name.len, 0, old_idx); | |
9795 | if (ret) { | |
66642832 | 9796 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9797 | goto out_fail; |
9798 | } | |
9799 | ||
db0a669f | 9800 | ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode), |
cdd1fedf DF |
9801 | old_dentry->d_name.name, |
9802 | old_dentry->d_name.len, 0, new_idx); | |
9803 | if (ret) { | |
66642832 | 9804 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9805 | goto out_fail; |
9806 | } | |
9807 | ||
9808 | if (old_inode->i_nlink == 1) | |
9809 | BTRFS_I(old_inode)->dir_index = old_idx; | |
9810 | if (new_inode->i_nlink == 1) | |
9811 | BTRFS_I(new_inode)->dir_index = new_idx; | |
9812 | ||
86e8aa0e | 9813 | if (root_log_pinned) { |
cdd1fedf | 9814 | parent = new_dentry->d_parent; |
f85b7379 DS |
9815 | btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir), |
9816 | parent); | |
cdd1fedf | 9817 | btrfs_end_log_trans(root); |
86e8aa0e | 9818 | root_log_pinned = false; |
cdd1fedf | 9819 | } |
86e8aa0e | 9820 | if (dest_log_pinned) { |
cdd1fedf | 9821 | parent = old_dentry->d_parent; |
f85b7379 DS |
9822 | btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir), |
9823 | parent); | |
cdd1fedf | 9824 | btrfs_end_log_trans(dest); |
86e8aa0e | 9825 | dest_log_pinned = false; |
cdd1fedf DF |
9826 | } |
9827 | out_fail: | |
86e8aa0e FM |
9828 | /* |
9829 | * If we have pinned a log and an error happened, we unpin tasks | |
9830 | * trying to sync the log and force them to fallback to a transaction | |
9831 | * commit if the log currently contains any of the inodes involved in | |
9832 | * this rename operation (to ensure we do not persist a log with an | |
9833 | * inconsistent state for any of these inodes or leading to any | |
9834 | * inconsistencies when replayed). If the transaction was aborted, the | |
9835 | * abortion reason is propagated to userspace when attempting to commit | |
9836 | * the transaction. If the log does not contain any of these inodes, we | |
9837 | * allow the tasks to sync it. | |
9838 | */ | |
9839 | if (ret && (root_log_pinned || dest_log_pinned)) { | |
0f8939b8 NB |
9840 | if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) || |
9841 | btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) || | |
9842 | btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) || | |
86e8aa0e | 9843 | (new_inode && |
0f8939b8 | 9844 | btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation))) |
0b246afa | 9845 | btrfs_set_log_full_commit(fs_info, trans); |
86e8aa0e FM |
9846 | |
9847 | if (root_log_pinned) { | |
9848 | btrfs_end_log_trans(root); | |
9849 | root_log_pinned = false; | |
9850 | } | |
9851 | if (dest_log_pinned) { | |
9852 | btrfs_end_log_trans(dest); | |
9853 | dest_log_pinned = false; | |
9854 | } | |
9855 | } | |
3a45bb20 | 9856 | ret = btrfs_end_transaction(trans); |
cdd1fedf DF |
9857 | out_notrans: |
9858 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 9859 | up_read(&fs_info->subvol_sem); |
cdd1fedf | 9860 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9861 | up_read(&fs_info->subvol_sem); |
cdd1fedf DF |
9862 | |
9863 | return ret; | |
9864 | } | |
9865 | ||
9866 | static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans, | |
9867 | struct btrfs_root *root, | |
9868 | struct inode *dir, | |
9869 | struct dentry *dentry) | |
9870 | { | |
9871 | int ret; | |
9872 | struct inode *inode; | |
9873 | u64 objectid; | |
9874 | u64 index; | |
9875 | ||
9876 | ret = btrfs_find_free_ino(root, &objectid); | |
9877 | if (ret) | |
9878 | return ret; | |
9879 | ||
9880 | inode = btrfs_new_inode(trans, root, dir, | |
9881 | dentry->d_name.name, | |
9882 | dentry->d_name.len, | |
4a0cc7ca | 9883 | btrfs_ino(BTRFS_I(dir)), |
cdd1fedf DF |
9884 | objectid, |
9885 | S_IFCHR | WHITEOUT_MODE, | |
9886 | &index); | |
9887 | ||
9888 | if (IS_ERR(inode)) { | |
9889 | ret = PTR_ERR(inode); | |
9890 | return ret; | |
9891 | } | |
9892 | ||
9893 | inode->i_op = &btrfs_special_inode_operations; | |
9894 | init_special_inode(inode, inode->i_mode, | |
9895 | WHITEOUT_DEV); | |
9896 | ||
9897 | ret = btrfs_init_inode_security(trans, inode, dir, | |
9898 | &dentry->d_name); | |
9899 | if (ret) | |
c9901618 | 9900 | goto out; |
cdd1fedf | 9901 | |
cef415af NB |
9902 | ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, |
9903 | BTRFS_I(inode), 0, index); | |
cdd1fedf | 9904 | if (ret) |
c9901618 | 9905 | goto out; |
cdd1fedf DF |
9906 | |
9907 | ret = btrfs_update_inode(trans, root, inode); | |
c9901618 | 9908 | out: |
cdd1fedf | 9909 | unlock_new_inode(inode); |
c9901618 FM |
9910 | if (ret) |
9911 | inode_dec_link_count(inode); | |
cdd1fedf DF |
9912 | iput(inode); |
9913 | ||
c9901618 | 9914 | return ret; |
cdd1fedf DF |
9915 | } |
9916 | ||
d397712b | 9917 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
cdd1fedf DF |
9918 | struct inode *new_dir, struct dentry *new_dentry, |
9919 | unsigned int flags) | |
39279cc3 | 9920 | { |
0b246afa | 9921 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
39279cc3 | 9922 | struct btrfs_trans_handle *trans; |
5062af35 | 9923 | unsigned int trans_num_items; |
39279cc3 | 9924 | struct btrfs_root *root = BTRFS_I(old_dir)->root; |
4df27c4d | 9925 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
2b0143b5 DH |
9926 | struct inode *new_inode = d_inode(new_dentry); |
9927 | struct inode *old_inode = d_inode(old_dentry); | |
00e4e6b3 | 9928 | u64 index = 0; |
4df27c4d | 9929 | u64 root_objectid; |
39279cc3 | 9930 | int ret; |
4a0cc7ca | 9931 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
3dc9e8f7 | 9932 | bool log_pinned = false; |
39279cc3 | 9933 | |
4a0cc7ca | 9934 | if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
9935 | return -EPERM; |
9936 | ||
4df27c4d | 9937 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 9938 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
9939 | return -EXDEV; |
9940 | ||
33345d01 | 9941 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
4a0cc7ca | 9942 | (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID)) |
39279cc3 | 9943 | return -ENOTEMPTY; |
5f39d397 | 9944 | |
4df27c4d YZ |
9945 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
9946 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
9947 | return -ENOTEMPTY; | |
9c52057c CM |
9948 | |
9949 | ||
9950 | /* check for collisions, even if the name isn't there */ | |
4871c158 | 9951 | ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, |
9c52057c CM |
9952 | new_dentry->d_name.name, |
9953 | new_dentry->d_name.len); | |
9954 | ||
9955 | if (ret) { | |
9956 | if (ret == -EEXIST) { | |
9957 | /* we shouldn't get | |
9958 | * eexist without a new_inode */ | |
fae7f21c | 9959 | if (WARN_ON(!new_inode)) { |
9c52057c CM |
9960 | return ret; |
9961 | } | |
9962 | } else { | |
9963 | /* maybe -EOVERFLOW */ | |
9964 | return ret; | |
9965 | } | |
9966 | } | |
9967 | ret = 0; | |
9968 | ||
5a3f23d5 | 9969 | /* |
8d875f95 CM |
9970 | * we're using rename to replace one file with another. Start IO on it |
9971 | * now so we don't add too much work to the end of the transaction | |
5a3f23d5 | 9972 | */ |
8d875f95 | 9973 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size) |
5a3f23d5 CM |
9974 | filemap_flush(old_inode->i_mapping); |
9975 | ||
76dda93c | 9976 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 9977 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9978 | down_read(&fs_info->subvol_sem); |
a22285a6 YZ |
9979 | /* |
9980 | * We want to reserve the absolute worst case amount of items. So if | |
9981 | * both inodes are subvols and we need to unlink them then that would | |
9982 | * require 4 item modifications, but if they are both normal inodes it | |
cdd1fedf | 9983 | * would require 5 item modifications, so we'll assume they are normal |
a22285a6 YZ |
9984 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items |
9985 | * should cover the worst case number of items we'll modify. | |
5062af35 FM |
9986 | * If our rename has the whiteout flag, we need more 5 units for the |
9987 | * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item | |
9988 | * when selinux is enabled). | |
a22285a6 | 9989 | */ |
5062af35 FM |
9990 | trans_num_items = 11; |
9991 | if (flags & RENAME_WHITEOUT) | |
9992 | trans_num_items += 5; | |
9993 | trans = btrfs_start_transaction(root, trans_num_items); | |
b44c59a8 | 9994 | if (IS_ERR(trans)) { |
cdd1fedf DF |
9995 | ret = PTR_ERR(trans); |
9996 | goto out_notrans; | |
9997 | } | |
76dda93c | 9998 | |
4df27c4d YZ |
9999 | if (dest != root) |
10000 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 10001 | |
877574e2 | 10002 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index); |
a5719521 YZ |
10003 | if (ret) |
10004 | goto out_fail; | |
5a3f23d5 | 10005 | |
67de1176 | 10006 | BTRFS_I(old_inode)->dir_index = 0ULL; |
33345d01 | 10007 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d | 10008 | /* force full log commit if subvolume involved. */ |
0b246afa | 10009 | btrfs_set_log_full_commit(fs_info, trans); |
4df27c4d | 10010 | } else { |
c4aba954 FM |
10011 | btrfs_pin_log_trans(root); |
10012 | log_pinned = true; | |
a5719521 YZ |
10013 | ret = btrfs_insert_inode_ref(trans, dest, |
10014 | new_dentry->d_name.name, | |
10015 | new_dentry->d_name.len, | |
33345d01 | 10016 | old_ino, |
4a0cc7ca | 10017 | btrfs_ino(BTRFS_I(new_dir)), index); |
a5719521 YZ |
10018 | if (ret) |
10019 | goto out_fail; | |
4df27c4d | 10020 | } |
5a3f23d5 | 10021 | |
0c4d2d95 JB |
10022 | inode_inc_iversion(old_dir); |
10023 | inode_inc_iversion(new_dir); | |
10024 | inode_inc_iversion(old_inode); | |
04b285f3 DD |
10025 | old_dir->i_ctime = old_dir->i_mtime = |
10026 | new_dir->i_ctime = new_dir->i_mtime = | |
c2050a45 | 10027 | old_inode->i_ctime = current_time(old_dir); |
5f39d397 | 10028 | |
12fcfd22 | 10029 | if (old_dentry->d_parent != new_dentry->d_parent) |
f85b7379 DS |
10030 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
10031 | BTRFS_I(old_inode), 1); | |
12fcfd22 | 10032 | |
33345d01 | 10033 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
10034 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
10035 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
10036 | old_dentry->d_name.name, | |
10037 | old_dentry->d_name.len); | |
10038 | } else { | |
4ec5934e NB |
10039 | ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir), |
10040 | BTRFS_I(d_inode(old_dentry)), | |
92986796 AV |
10041 | old_dentry->d_name.name, |
10042 | old_dentry->d_name.len); | |
10043 | if (!ret) | |
10044 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 10045 | } |
79787eaa | 10046 | if (ret) { |
66642832 | 10047 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
10048 | goto out_fail; |
10049 | } | |
39279cc3 CM |
10050 | |
10051 | if (new_inode) { | |
0c4d2d95 | 10052 | inode_inc_iversion(new_inode); |
c2050a45 | 10053 | new_inode->i_ctime = current_time(new_inode); |
4a0cc7ca | 10054 | if (unlikely(btrfs_ino(BTRFS_I(new_inode)) == |
4df27c4d YZ |
10055 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
10056 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
10057 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
10058 | root_objectid, | |
10059 | new_dentry->d_name.name, | |
10060 | new_dentry->d_name.len); | |
10061 | BUG_ON(new_inode->i_nlink == 0); | |
10062 | } else { | |
4ec5934e NB |
10063 | ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir), |
10064 | BTRFS_I(d_inode(new_dentry)), | |
4df27c4d YZ |
10065 | new_dentry->d_name.name, |
10066 | new_dentry->d_name.len); | |
10067 | } | |
4ef31a45 | 10068 | if (!ret && new_inode->i_nlink == 0) |
73f2e545 NB |
10069 | ret = btrfs_orphan_add(trans, |
10070 | BTRFS_I(d_inode(new_dentry))); | |
79787eaa | 10071 | if (ret) { |
66642832 | 10072 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
10073 | goto out_fail; |
10074 | } | |
39279cc3 | 10075 | } |
aec7477b | 10076 | |
db0a669f | 10077 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
4df27c4d | 10078 | new_dentry->d_name.name, |
a5719521 | 10079 | new_dentry->d_name.len, 0, index); |
79787eaa | 10080 | if (ret) { |
66642832 | 10081 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
10082 | goto out_fail; |
10083 | } | |
39279cc3 | 10084 | |
67de1176 MX |
10085 | if (old_inode->i_nlink == 1) |
10086 | BTRFS_I(old_inode)->dir_index = index; | |
10087 | ||
3dc9e8f7 | 10088 | if (log_pinned) { |
10d9f309 | 10089 | struct dentry *parent = new_dentry->d_parent; |
3dc9e8f7 | 10090 | |
f85b7379 DS |
10091 | btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir), |
10092 | parent); | |
4df27c4d | 10093 | btrfs_end_log_trans(root); |
3dc9e8f7 | 10094 | log_pinned = false; |
4df27c4d | 10095 | } |
cdd1fedf DF |
10096 | |
10097 | if (flags & RENAME_WHITEOUT) { | |
10098 | ret = btrfs_whiteout_for_rename(trans, root, old_dir, | |
10099 | old_dentry); | |
10100 | ||
10101 | if (ret) { | |
66642832 | 10102 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
10103 | goto out_fail; |
10104 | } | |
4df27c4d | 10105 | } |
39279cc3 | 10106 | out_fail: |
3dc9e8f7 FM |
10107 | /* |
10108 | * If we have pinned the log and an error happened, we unpin tasks | |
10109 | * trying to sync the log and force them to fallback to a transaction | |
10110 | * commit if the log currently contains any of the inodes involved in | |
10111 | * this rename operation (to ensure we do not persist a log with an | |
10112 | * inconsistent state for any of these inodes or leading to any | |
10113 | * inconsistencies when replayed). If the transaction was aborted, the | |
10114 | * abortion reason is propagated to userspace when attempting to commit | |
10115 | * the transaction. If the log does not contain any of these inodes, we | |
10116 | * allow the tasks to sync it. | |
10117 | */ | |
10118 | if (ret && log_pinned) { | |
0f8939b8 NB |
10119 | if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) || |
10120 | btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) || | |
10121 | btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) || | |
3dc9e8f7 | 10122 | (new_inode && |
0f8939b8 | 10123 | btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation))) |
0b246afa | 10124 | btrfs_set_log_full_commit(fs_info, trans); |
3dc9e8f7 FM |
10125 | |
10126 | btrfs_end_log_trans(root); | |
10127 | log_pinned = false; | |
10128 | } | |
3a45bb20 | 10129 | btrfs_end_transaction(trans); |
b44c59a8 | 10130 | out_notrans: |
33345d01 | 10131 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 10132 | up_read(&fs_info->subvol_sem); |
9ed74f2d | 10133 | |
39279cc3 CM |
10134 | return ret; |
10135 | } | |
10136 | ||
80ace85c MS |
10137 | static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry, |
10138 | struct inode *new_dir, struct dentry *new_dentry, | |
10139 | unsigned int flags) | |
10140 | { | |
cdd1fedf | 10141 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
80ace85c MS |
10142 | return -EINVAL; |
10143 | ||
cdd1fedf DF |
10144 | if (flags & RENAME_EXCHANGE) |
10145 | return btrfs_rename_exchange(old_dir, old_dentry, new_dir, | |
10146 | new_dentry); | |
10147 | ||
10148 | return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags); | |
80ace85c MS |
10149 | } |
10150 | ||
8ccf6f19 MX |
10151 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
10152 | { | |
10153 | struct btrfs_delalloc_work *delalloc_work; | |
9f23e289 | 10154 | struct inode *inode; |
8ccf6f19 MX |
10155 | |
10156 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
10157 | work); | |
9f23e289 | 10158 | inode = delalloc_work->inode; |
30424601 DS |
10159 | filemap_flush(inode->i_mapping); |
10160 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
10161 | &BTRFS_I(inode)->runtime_flags)) | |
9f23e289 | 10162 | filemap_flush(inode->i_mapping); |
8ccf6f19 MX |
10163 | |
10164 | if (delalloc_work->delay_iput) | |
9f23e289 | 10165 | btrfs_add_delayed_iput(inode); |
8ccf6f19 | 10166 | else |
9f23e289 | 10167 | iput(inode); |
8ccf6f19 MX |
10168 | complete(&delalloc_work->completion); |
10169 | } | |
10170 | ||
10171 | struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode, | |
651d494a | 10172 | int delay_iput) |
8ccf6f19 MX |
10173 | { |
10174 | struct btrfs_delalloc_work *work; | |
10175 | ||
100d5702 | 10176 | work = kmalloc(sizeof(*work), GFP_NOFS); |
8ccf6f19 MX |
10177 | if (!work) |
10178 | return NULL; | |
10179 | ||
10180 | init_completion(&work->completion); | |
10181 | INIT_LIST_HEAD(&work->list); | |
10182 | work->inode = inode; | |
8ccf6f19 | 10183 | work->delay_iput = delay_iput; |
9e0af237 LB |
10184 | WARN_ON_ONCE(!inode); |
10185 | btrfs_init_work(&work->work, btrfs_flush_delalloc_helper, | |
10186 | btrfs_run_delalloc_work, NULL, NULL); | |
8ccf6f19 MX |
10187 | |
10188 | return work; | |
10189 | } | |
10190 | ||
10191 | void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work) | |
10192 | { | |
10193 | wait_for_completion(&work->completion); | |
100d5702 | 10194 | kfree(work); |
8ccf6f19 MX |
10195 | } |
10196 | ||
d352ac68 CM |
10197 | /* |
10198 | * some fairly slow code that needs optimization. This walks the list | |
10199 | * of all the inodes with pending delalloc and forces them to disk. | |
10200 | */ | |
6c255e67 MX |
10201 | static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput, |
10202 | int nr) | |
ea8c2819 | 10203 | { |
ea8c2819 | 10204 | struct btrfs_inode *binode; |
5b21f2ed | 10205 | struct inode *inode; |
8ccf6f19 MX |
10206 | struct btrfs_delalloc_work *work, *next; |
10207 | struct list_head works; | |
1eafa6c7 | 10208 | struct list_head splice; |
8ccf6f19 | 10209 | int ret = 0; |
ea8c2819 | 10210 | |
8ccf6f19 | 10211 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 10212 | INIT_LIST_HEAD(&splice); |
63607cc8 | 10213 | |
573bfb72 | 10214 | mutex_lock(&root->delalloc_mutex); |
eb73c1b7 MX |
10215 | spin_lock(&root->delalloc_lock); |
10216 | list_splice_init(&root->delalloc_inodes, &splice); | |
1eafa6c7 MX |
10217 | while (!list_empty(&splice)) { |
10218 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 10219 | delalloc_inodes); |
1eafa6c7 | 10220 | |
eb73c1b7 MX |
10221 | list_move_tail(&binode->delalloc_inodes, |
10222 | &root->delalloc_inodes); | |
5b21f2ed | 10223 | inode = igrab(&binode->vfs_inode); |
df0af1a5 | 10224 | if (!inode) { |
eb73c1b7 | 10225 | cond_resched_lock(&root->delalloc_lock); |
1eafa6c7 | 10226 | continue; |
df0af1a5 | 10227 | } |
eb73c1b7 | 10228 | spin_unlock(&root->delalloc_lock); |
1eafa6c7 | 10229 | |
651d494a | 10230 | work = btrfs_alloc_delalloc_work(inode, delay_iput); |
5d99a998 | 10231 | if (!work) { |
f4ab9ea7 JB |
10232 | if (delay_iput) |
10233 | btrfs_add_delayed_iput(inode); | |
10234 | else | |
10235 | iput(inode); | |
1eafa6c7 | 10236 | ret = -ENOMEM; |
a1ecaabb | 10237 | goto out; |
5b21f2ed | 10238 | } |
1eafa6c7 | 10239 | list_add_tail(&work->list, &works); |
a44903ab QW |
10240 | btrfs_queue_work(root->fs_info->flush_workers, |
10241 | &work->work); | |
6c255e67 MX |
10242 | ret++; |
10243 | if (nr != -1 && ret >= nr) | |
a1ecaabb | 10244 | goto out; |
5b21f2ed | 10245 | cond_resched(); |
eb73c1b7 | 10246 | spin_lock(&root->delalloc_lock); |
ea8c2819 | 10247 | } |
eb73c1b7 | 10248 | spin_unlock(&root->delalloc_lock); |
8c8bee1d | 10249 | |
a1ecaabb | 10250 | out: |
eb73c1b7 MX |
10251 | list_for_each_entry_safe(work, next, &works, list) { |
10252 | list_del_init(&work->list); | |
10253 | btrfs_wait_and_free_delalloc_work(work); | |
10254 | } | |
10255 | ||
10256 | if (!list_empty_careful(&splice)) { | |
10257 | spin_lock(&root->delalloc_lock); | |
10258 | list_splice_tail(&splice, &root->delalloc_inodes); | |
10259 | spin_unlock(&root->delalloc_lock); | |
10260 | } | |
573bfb72 | 10261 | mutex_unlock(&root->delalloc_mutex); |
eb73c1b7 MX |
10262 | return ret; |
10263 | } | |
1eafa6c7 | 10264 | |
eb73c1b7 MX |
10265 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
10266 | { | |
0b246afa | 10267 | struct btrfs_fs_info *fs_info = root->fs_info; |
eb73c1b7 | 10268 | int ret; |
1eafa6c7 | 10269 | |
0b246afa | 10270 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
10271 | return -EROFS; |
10272 | ||
6c255e67 MX |
10273 | ret = __start_delalloc_inodes(root, delay_iput, -1); |
10274 | if (ret > 0) | |
10275 | ret = 0; | |
eb73c1b7 MX |
10276 | return ret; |
10277 | } | |
10278 | ||
6c255e67 MX |
10279 | int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput, |
10280 | int nr) | |
eb73c1b7 MX |
10281 | { |
10282 | struct btrfs_root *root; | |
10283 | struct list_head splice; | |
10284 | int ret; | |
10285 | ||
2c21b4d7 | 10286 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
10287 | return -EROFS; |
10288 | ||
10289 | INIT_LIST_HEAD(&splice); | |
10290 | ||
573bfb72 | 10291 | mutex_lock(&fs_info->delalloc_root_mutex); |
eb73c1b7 MX |
10292 | spin_lock(&fs_info->delalloc_root_lock); |
10293 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
6c255e67 | 10294 | while (!list_empty(&splice) && nr) { |
eb73c1b7 MX |
10295 | root = list_first_entry(&splice, struct btrfs_root, |
10296 | delalloc_root); | |
10297 | root = btrfs_grab_fs_root(root); | |
10298 | BUG_ON(!root); | |
10299 | list_move_tail(&root->delalloc_root, | |
10300 | &fs_info->delalloc_roots); | |
10301 | spin_unlock(&fs_info->delalloc_root_lock); | |
10302 | ||
6c255e67 | 10303 | ret = __start_delalloc_inodes(root, delay_iput, nr); |
eb73c1b7 | 10304 | btrfs_put_fs_root(root); |
6c255e67 | 10305 | if (ret < 0) |
eb73c1b7 MX |
10306 | goto out; |
10307 | ||
6c255e67 MX |
10308 | if (nr != -1) { |
10309 | nr -= ret; | |
10310 | WARN_ON(nr < 0); | |
10311 | } | |
eb73c1b7 | 10312 | spin_lock(&fs_info->delalloc_root_lock); |
8ccf6f19 | 10313 | } |
eb73c1b7 | 10314 | spin_unlock(&fs_info->delalloc_root_lock); |
1eafa6c7 | 10315 | |
6c255e67 | 10316 | ret = 0; |
eb73c1b7 | 10317 | out: |
1eafa6c7 | 10318 | if (!list_empty_careful(&splice)) { |
eb73c1b7 MX |
10319 | spin_lock(&fs_info->delalloc_root_lock); |
10320 | list_splice_tail(&splice, &fs_info->delalloc_roots); | |
10321 | spin_unlock(&fs_info->delalloc_root_lock); | |
1eafa6c7 | 10322 | } |
573bfb72 | 10323 | mutex_unlock(&fs_info->delalloc_root_mutex); |
8ccf6f19 | 10324 | return ret; |
ea8c2819 CM |
10325 | } |
10326 | ||
39279cc3 CM |
10327 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
10328 | const char *symname) | |
10329 | { | |
0b246afa | 10330 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
10331 | struct btrfs_trans_handle *trans; |
10332 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
10333 | struct btrfs_path *path; | |
10334 | struct btrfs_key key; | |
1832a6d5 | 10335 | struct inode *inode = NULL; |
39279cc3 CM |
10336 | int err; |
10337 | int drop_inode = 0; | |
10338 | u64 objectid; | |
67871254 | 10339 | u64 index = 0; |
39279cc3 CM |
10340 | int name_len; |
10341 | int datasize; | |
5f39d397 | 10342 | unsigned long ptr; |
39279cc3 | 10343 | struct btrfs_file_extent_item *ei; |
5f39d397 | 10344 | struct extent_buffer *leaf; |
39279cc3 | 10345 | |
f06becc4 | 10346 | name_len = strlen(symname); |
0b246afa | 10347 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info)) |
39279cc3 | 10348 | return -ENAMETOOLONG; |
1832a6d5 | 10349 | |
9ed74f2d JB |
10350 | /* |
10351 | * 2 items for inode item and ref | |
10352 | * 2 items for dir items | |
9269d12b FM |
10353 | * 1 item for updating parent inode item |
10354 | * 1 item for the inline extent item | |
9ed74f2d JB |
10355 | * 1 item for xattr if selinux is on |
10356 | */ | |
9269d12b | 10357 | trans = btrfs_start_transaction(root, 7); |
a22285a6 YZ |
10358 | if (IS_ERR(trans)) |
10359 | return PTR_ERR(trans); | |
1832a6d5 | 10360 | |
581bb050 LZ |
10361 | err = btrfs_find_free_ino(root, &objectid); |
10362 | if (err) | |
10363 | goto out_unlock; | |
10364 | ||
aec7477b | 10365 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
10366 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), |
10367 | objectid, S_IFLNK|S_IRWXUGO, &index); | |
7cf96da3 TI |
10368 | if (IS_ERR(inode)) { |
10369 | err = PTR_ERR(inode); | |
39279cc3 | 10370 | goto out_unlock; |
7cf96da3 | 10371 | } |
39279cc3 | 10372 | |
ad19db71 CS |
10373 | /* |
10374 | * If the active LSM wants to access the inode during | |
10375 | * d_instantiate it needs these. Smack checks to see | |
10376 | * if the filesystem supports xattrs by looking at the | |
10377 | * ops vector. | |
10378 | */ | |
10379 | inode->i_fop = &btrfs_file_operations; | |
10380 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 10381 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
10382 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
10383 | ||
10384 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
10385 | if (err) | |
10386 | goto out_unlock_inode; | |
ad19db71 | 10387 | |
39279cc3 | 10388 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10389 | if (!path) { |
10390 | err = -ENOMEM; | |
b0d5d10f | 10391 | goto out_unlock_inode; |
d8926bb3 | 10392 | } |
4a0cc7ca | 10393 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
39279cc3 | 10394 | key.offset = 0; |
962a298f | 10395 | key.type = BTRFS_EXTENT_DATA_KEY; |
39279cc3 CM |
10396 | datasize = btrfs_file_extent_calc_inline_size(name_len); |
10397 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
10398 | datasize); | |
54aa1f4d | 10399 | if (err) { |
b0839166 | 10400 | btrfs_free_path(path); |
b0d5d10f | 10401 | goto out_unlock_inode; |
54aa1f4d | 10402 | } |
5f39d397 CM |
10403 | leaf = path->nodes[0]; |
10404 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
10405 | struct btrfs_file_extent_item); | |
10406 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
10407 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 10408 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
10409 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
10410 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
10411 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
10412 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
10413 | ||
39279cc3 | 10414 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
10415 | write_extent_buffer(leaf, symname, ptr, name_len); |
10416 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 10417 | btrfs_free_path(path); |
5f39d397 | 10418 | |
39279cc3 | 10419 | inode->i_op = &btrfs_symlink_inode_operations; |
21fc61c7 | 10420 | inode_nohighmem(inode); |
39279cc3 | 10421 | inode->i_mapping->a_ops = &btrfs_symlink_aops; |
d899e052 | 10422 | inode_set_bytes(inode, name_len); |
6ef06d27 | 10423 | btrfs_i_size_write(BTRFS_I(inode), name_len); |
54aa1f4d | 10424 | err = btrfs_update_inode(trans, root, inode); |
d50866d0 FM |
10425 | /* |
10426 | * Last step, add directory indexes for our symlink inode. This is the | |
10427 | * last step to avoid extra cleanup of these indexes if an error happens | |
10428 | * elsewhere above. | |
10429 | */ | |
10430 | if (!err) | |
cef415af NB |
10431 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, |
10432 | BTRFS_I(inode), 0, index); | |
b0d5d10f | 10433 | if (err) { |
54aa1f4d | 10434 | drop_inode = 1; |
b0d5d10f CM |
10435 | goto out_unlock_inode; |
10436 | } | |
10437 | ||
10438 | unlock_new_inode(inode); | |
10439 | d_instantiate(dentry, inode); | |
39279cc3 CM |
10440 | |
10441 | out_unlock: | |
3a45bb20 | 10442 | btrfs_end_transaction(trans); |
39279cc3 CM |
10443 | if (drop_inode) { |
10444 | inode_dec_link_count(inode); | |
10445 | iput(inode); | |
10446 | } | |
2ff7e61e | 10447 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 | 10448 | return err; |
b0d5d10f CM |
10449 | |
10450 | out_unlock_inode: | |
10451 | drop_inode = 1; | |
10452 | unlock_new_inode(inode); | |
10453 | goto out_unlock; | |
39279cc3 | 10454 | } |
16432985 | 10455 | |
0af3d00b JB |
10456 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
10457 | u64 start, u64 num_bytes, u64 min_size, | |
10458 | loff_t actual_len, u64 *alloc_hint, | |
10459 | struct btrfs_trans_handle *trans) | |
d899e052 | 10460 | { |
0b246afa | 10461 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5dc562c5 JB |
10462 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
10463 | struct extent_map *em; | |
d899e052 YZ |
10464 | struct btrfs_root *root = BTRFS_I(inode)->root; |
10465 | struct btrfs_key ins; | |
d899e052 | 10466 | u64 cur_offset = start; |
55a61d1d | 10467 | u64 i_size; |
154ea289 | 10468 | u64 cur_bytes; |
0b670dc4 | 10469 | u64 last_alloc = (u64)-1; |
d899e052 | 10470 | int ret = 0; |
0af3d00b | 10471 | bool own_trans = true; |
18513091 | 10472 | u64 end = start + num_bytes - 1; |
d899e052 | 10473 | |
0af3d00b JB |
10474 | if (trans) |
10475 | own_trans = false; | |
d899e052 | 10476 | while (num_bytes > 0) { |
0af3d00b JB |
10477 | if (own_trans) { |
10478 | trans = btrfs_start_transaction(root, 3); | |
10479 | if (IS_ERR(trans)) { | |
10480 | ret = PTR_ERR(trans); | |
10481 | break; | |
10482 | } | |
5a303d5d YZ |
10483 | } |
10484 | ||
ee22184b | 10485 | cur_bytes = min_t(u64, num_bytes, SZ_256M); |
154ea289 | 10486 | cur_bytes = max(cur_bytes, min_size); |
0b670dc4 JB |
10487 | /* |
10488 | * If we are severely fragmented we could end up with really | |
10489 | * small allocations, so if the allocator is returning small | |
10490 | * chunks lets make its job easier by only searching for those | |
10491 | * sized chunks. | |
10492 | */ | |
10493 | cur_bytes = min(cur_bytes, last_alloc); | |
18513091 WX |
10494 | ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes, |
10495 | min_size, 0, *alloc_hint, &ins, 1, 0); | |
5a303d5d | 10496 | if (ret) { |
0af3d00b | 10497 | if (own_trans) |
3a45bb20 | 10498 | btrfs_end_transaction(trans); |
a22285a6 | 10499 | break; |
d899e052 | 10500 | } |
0b246afa | 10501 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
5a303d5d | 10502 | |
0b670dc4 | 10503 | last_alloc = ins.offset; |
d899e052 YZ |
10504 | ret = insert_reserved_file_extent(trans, inode, |
10505 | cur_offset, ins.objectid, | |
10506 | ins.offset, ins.offset, | |
920bbbfb | 10507 | ins.offset, 0, 0, 0, |
d899e052 | 10508 | BTRFS_FILE_EXTENT_PREALLOC); |
79787eaa | 10509 | if (ret) { |
2ff7e61e | 10510 | btrfs_free_reserved_extent(fs_info, ins.objectid, |
e570fd27 | 10511 | ins.offset, 0); |
66642832 | 10512 | btrfs_abort_transaction(trans, ret); |
79787eaa | 10513 | if (own_trans) |
3a45bb20 | 10514 | btrfs_end_transaction(trans); |
79787eaa JM |
10515 | break; |
10516 | } | |
31193213 | 10517 | |
dcdbc059 | 10518 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
a1ed835e | 10519 | cur_offset + ins.offset -1, 0); |
5a303d5d | 10520 | |
5dc562c5 JB |
10521 | em = alloc_extent_map(); |
10522 | if (!em) { | |
10523 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
10524 | &BTRFS_I(inode)->runtime_flags); | |
10525 | goto next; | |
10526 | } | |
10527 | ||
10528 | em->start = cur_offset; | |
10529 | em->orig_start = cur_offset; | |
10530 | em->len = ins.offset; | |
10531 | em->block_start = ins.objectid; | |
10532 | em->block_len = ins.offset; | |
b4939680 | 10533 | em->orig_block_len = ins.offset; |
cc95bef6 | 10534 | em->ram_bytes = ins.offset; |
0b246afa | 10535 | em->bdev = fs_info->fs_devices->latest_bdev; |
5dc562c5 JB |
10536 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); |
10537 | em->generation = trans->transid; | |
10538 | ||
10539 | while (1) { | |
10540 | write_lock(&em_tree->lock); | |
09a2a8f9 | 10541 | ret = add_extent_mapping(em_tree, em, 1); |
5dc562c5 JB |
10542 | write_unlock(&em_tree->lock); |
10543 | if (ret != -EEXIST) | |
10544 | break; | |
dcdbc059 | 10545 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
5dc562c5 JB |
10546 | cur_offset + ins.offset - 1, |
10547 | 0); | |
10548 | } | |
10549 | free_extent_map(em); | |
10550 | next: | |
d899e052 YZ |
10551 | num_bytes -= ins.offset; |
10552 | cur_offset += ins.offset; | |
efa56464 | 10553 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 10554 | |
0c4d2d95 | 10555 | inode_inc_iversion(inode); |
c2050a45 | 10556 | inode->i_ctime = current_time(inode); |
6cbff00f | 10557 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 10558 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
10559 | (actual_len > inode->i_size) && |
10560 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 10561 | if (cur_offset > actual_len) |
55a61d1d | 10562 | i_size = actual_len; |
d1ea6a61 | 10563 | else |
55a61d1d JB |
10564 | i_size = cur_offset; |
10565 | i_size_write(inode, i_size); | |
10566 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
10567 | } |
10568 | ||
d899e052 | 10569 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
10570 | |
10571 | if (ret) { | |
66642832 | 10572 | btrfs_abort_transaction(trans, ret); |
79787eaa | 10573 | if (own_trans) |
3a45bb20 | 10574 | btrfs_end_transaction(trans); |
79787eaa JM |
10575 | break; |
10576 | } | |
d899e052 | 10577 | |
0af3d00b | 10578 | if (own_trans) |
3a45bb20 | 10579 | btrfs_end_transaction(trans); |
5a303d5d | 10580 | } |
18513091 | 10581 | if (cur_offset < end) |
bc42bda2 | 10582 | btrfs_free_reserved_data_space(inode, NULL, cur_offset, |
18513091 | 10583 | end - cur_offset + 1); |
d899e052 YZ |
10584 | return ret; |
10585 | } | |
10586 | ||
0af3d00b JB |
10587 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
10588 | u64 start, u64 num_bytes, u64 min_size, | |
10589 | loff_t actual_len, u64 *alloc_hint) | |
10590 | { | |
10591 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10592 | min_size, actual_len, alloc_hint, | |
10593 | NULL); | |
10594 | } | |
10595 | ||
10596 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
10597 | struct btrfs_trans_handle *trans, int mode, | |
10598 | u64 start, u64 num_bytes, u64 min_size, | |
10599 | loff_t actual_len, u64 *alloc_hint) | |
10600 | { | |
10601 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10602 | min_size, actual_len, alloc_hint, trans); | |
10603 | } | |
10604 | ||
e6dcd2dc CM |
10605 | static int btrfs_set_page_dirty(struct page *page) |
10606 | { | |
e6dcd2dc CM |
10607 | return __set_page_dirty_nobuffers(page); |
10608 | } | |
10609 | ||
10556cb2 | 10610 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 10611 | { |
b83cc969 | 10612 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 10613 | umode_t mode = inode->i_mode; |
b83cc969 | 10614 | |
cb6db4e5 JM |
10615 | if (mask & MAY_WRITE && |
10616 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
10617 | if (btrfs_root_readonly(root)) | |
10618 | return -EROFS; | |
10619 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
10620 | return -EACCES; | |
10621 | } | |
2830ba7f | 10622 | return generic_permission(inode, mask); |
fdebe2bd | 10623 | } |
39279cc3 | 10624 | |
ef3b9af5 FM |
10625 | static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) |
10626 | { | |
2ff7e61e | 10627 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
ef3b9af5 FM |
10628 | struct btrfs_trans_handle *trans; |
10629 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
10630 | struct inode *inode = NULL; | |
10631 | u64 objectid; | |
10632 | u64 index; | |
10633 | int ret = 0; | |
10634 | ||
10635 | /* | |
10636 | * 5 units required for adding orphan entry | |
10637 | */ | |
10638 | trans = btrfs_start_transaction(root, 5); | |
10639 | if (IS_ERR(trans)) | |
10640 | return PTR_ERR(trans); | |
10641 | ||
10642 | ret = btrfs_find_free_ino(root, &objectid); | |
10643 | if (ret) | |
10644 | goto out; | |
10645 | ||
10646 | inode = btrfs_new_inode(trans, root, dir, NULL, 0, | |
f85b7379 | 10647 | btrfs_ino(BTRFS_I(dir)), objectid, mode, &index); |
ef3b9af5 FM |
10648 | if (IS_ERR(inode)) { |
10649 | ret = PTR_ERR(inode); | |
10650 | inode = NULL; | |
10651 | goto out; | |
10652 | } | |
10653 | ||
ef3b9af5 FM |
10654 | inode->i_fop = &btrfs_file_operations; |
10655 | inode->i_op = &btrfs_file_inode_operations; | |
10656 | ||
10657 | inode->i_mapping->a_ops = &btrfs_aops; | |
ef3b9af5 FM |
10658 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
10659 | ||
b0d5d10f CM |
10660 | ret = btrfs_init_inode_security(trans, inode, dir, NULL); |
10661 | if (ret) | |
10662 | goto out_inode; | |
10663 | ||
10664 | ret = btrfs_update_inode(trans, root, inode); | |
10665 | if (ret) | |
10666 | goto out_inode; | |
73f2e545 | 10667 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
ef3b9af5 | 10668 | if (ret) |
b0d5d10f | 10669 | goto out_inode; |
ef3b9af5 | 10670 | |
5762b5c9 FM |
10671 | /* |
10672 | * We set number of links to 0 in btrfs_new_inode(), and here we set | |
10673 | * it to 1 because d_tmpfile() will issue a warning if the count is 0, | |
10674 | * through: | |
10675 | * | |
10676 | * d_tmpfile() -> inode_dec_link_count() -> drop_nlink() | |
10677 | */ | |
10678 | set_nlink(inode, 1); | |
b0d5d10f | 10679 | unlock_new_inode(inode); |
ef3b9af5 FM |
10680 | d_tmpfile(dentry, inode); |
10681 | mark_inode_dirty(inode); | |
10682 | ||
10683 | out: | |
3a45bb20 | 10684 | btrfs_end_transaction(trans); |
ef3b9af5 FM |
10685 | if (ret) |
10686 | iput(inode); | |
2ff7e61e | 10687 | btrfs_btree_balance_dirty(fs_info); |
ef3b9af5 | 10688 | return ret; |
b0d5d10f CM |
10689 | |
10690 | out_inode: | |
10691 | unlock_new_inode(inode); | |
10692 | goto out; | |
10693 | ||
ef3b9af5 FM |
10694 | } |
10695 | ||
20a7db8a | 10696 | __attribute__((const)) |
9d0d1c8b | 10697 | static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror) |
20a7db8a | 10698 | { |
9d0d1c8b | 10699 | return -EAGAIN; |
20a7db8a DS |
10700 | } |
10701 | ||
c6100a4b JB |
10702 | static struct btrfs_fs_info *iotree_fs_info(void *private_data) |
10703 | { | |
10704 | struct inode *inode = private_data; | |
10705 | return btrfs_sb(inode->i_sb); | |
10706 | } | |
10707 | ||
10708 | static void btrfs_check_extent_io_range(void *private_data, const char *caller, | |
10709 | u64 start, u64 end) | |
10710 | { | |
10711 | struct inode *inode = private_data; | |
10712 | u64 isize; | |
10713 | ||
10714 | isize = i_size_read(inode); | |
10715 | if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) { | |
10716 | btrfs_debug_rl(BTRFS_I(inode)->root->fs_info, | |
10717 | "%s: ino %llu isize %llu odd range [%llu,%llu]", | |
10718 | caller, btrfs_ino(BTRFS_I(inode)), isize, start, end); | |
10719 | } | |
10720 | } | |
10721 | ||
10722 | void btrfs_set_range_writeback(void *private_data, u64 start, u64 end) | |
10723 | { | |
10724 | struct inode *inode = private_data; | |
10725 | unsigned long index = start >> PAGE_SHIFT; | |
10726 | unsigned long end_index = end >> PAGE_SHIFT; | |
10727 | struct page *page; | |
10728 | ||
10729 | while (index <= end_index) { | |
10730 | page = find_get_page(inode->i_mapping, index); | |
10731 | ASSERT(page); /* Pages should be in the extent_io_tree */ | |
10732 | set_page_writeback(page); | |
10733 | put_page(page); | |
10734 | index++; | |
10735 | } | |
10736 | } | |
10737 | ||
6e1d5dcc | 10738 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 10739 | .getattr = btrfs_getattr, |
39279cc3 CM |
10740 | .lookup = btrfs_lookup, |
10741 | .create = btrfs_create, | |
10742 | .unlink = btrfs_unlink, | |
10743 | .link = btrfs_link, | |
10744 | .mkdir = btrfs_mkdir, | |
10745 | .rmdir = btrfs_rmdir, | |
2773bf00 | 10746 | .rename = btrfs_rename2, |
39279cc3 CM |
10747 | .symlink = btrfs_symlink, |
10748 | .setattr = btrfs_setattr, | |
618e21d5 | 10749 | .mknod = btrfs_mknod, |
5103e947 | 10750 | .listxattr = btrfs_listxattr, |
fdebe2bd | 10751 | .permission = btrfs_permission, |
4e34e719 | 10752 | .get_acl = btrfs_get_acl, |
996a710d | 10753 | .set_acl = btrfs_set_acl, |
93fd63c2 | 10754 | .update_time = btrfs_update_time, |
ef3b9af5 | 10755 | .tmpfile = btrfs_tmpfile, |
39279cc3 | 10756 | }; |
6e1d5dcc | 10757 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 10758 | .lookup = btrfs_lookup, |
fdebe2bd | 10759 | .permission = btrfs_permission, |
93fd63c2 | 10760 | .update_time = btrfs_update_time, |
39279cc3 | 10761 | }; |
76dda93c | 10762 | |
828c0950 | 10763 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
10764 | .llseek = generic_file_llseek, |
10765 | .read = generic_read_dir, | |
02dbfc99 | 10766 | .iterate_shared = btrfs_real_readdir, |
23b5ec74 | 10767 | .open = btrfs_opendir, |
34287aa3 | 10768 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 10769 | #ifdef CONFIG_COMPAT |
4c63c245 | 10770 | .compat_ioctl = btrfs_compat_ioctl, |
39279cc3 | 10771 | #endif |
6bf13c0c | 10772 | .release = btrfs_release_file, |
e02119d5 | 10773 | .fsync = btrfs_sync_file, |
39279cc3 CM |
10774 | }; |
10775 | ||
20e5506b | 10776 | static const struct extent_io_ops btrfs_extent_io_ops = { |
4d53dddb | 10777 | /* mandatory callbacks */ |
065631f6 | 10778 | .submit_bio_hook = btrfs_submit_bio_hook, |
07157aac | 10779 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
4d53dddb | 10780 | .merge_bio_hook = btrfs_merge_bio_hook, |
9d0d1c8b | 10781 | .readpage_io_failed_hook = btrfs_readpage_io_failed_hook, |
c6100a4b JB |
10782 | .tree_fs_info = iotree_fs_info, |
10783 | .set_range_writeback = btrfs_set_range_writeback, | |
4d53dddb DS |
10784 | |
10785 | /* optional callbacks */ | |
10786 | .fill_delalloc = run_delalloc_range, | |
e6dcd2dc | 10787 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 10788 | .writepage_start_hook = btrfs_writepage_start_hook, |
b0c68f8b CM |
10789 | .set_bit_hook = btrfs_set_bit_hook, |
10790 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
10791 | .merge_extent_hook = btrfs_merge_extent_hook, |
10792 | .split_extent_hook = btrfs_split_extent_hook, | |
c6100a4b | 10793 | .check_extent_io_range = btrfs_check_extent_io_range, |
07157aac CM |
10794 | }; |
10795 | ||
35054394 CM |
10796 | /* |
10797 | * btrfs doesn't support the bmap operation because swapfiles | |
10798 | * use bmap to make a mapping of extents in the file. They assume | |
10799 | * these extents won't change over the life of the file and they | |
10800 | * use the bmap result to do IO directly to the drive. | |
10801 | * | |
10802 | * the btrfs bmap call would return logical addresses that aren't | |
10803 | * suitable for IO and they also will change frequently as COW | |
10804 | * operations happen. So, swapfile + btrfs == corruption. | |
10805 | * | |
10806 | * For now we're avoiding this by dropping bmap. | |
10807 | */ | |
7f09410b | 10808 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
10809 | .readpage = btrfs_readpage, |
10810 | .writepage = btrfs_writepage, | |
b293f02e | 10811 | .writepages = btrfs_writepages, |
3ab2fb5a | 10812 | .readpages = btrfs_readpages, |
16432985 | 10813 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
10814 | .invalidatepage = btrfs_invalidatepage, |
10815 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 10816 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 10817 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
10818 | }; |
10819 | ||
7f09410b | 10820 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
10821 | .readpage = btrfs_readpage, |
10822 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
10823 | .invalidatepage = btrfs_invalidatepage, |
10824 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
10825 | }; |
10826 | ||
6e1d5dcc | 10827 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
10828 | .getattr = btrfs_getattr, |
10829 | .setattr = btrfs_setattr, | |
5103e947 | 10830 | .listxattr = btrfs_listxattr, |
fdebe2bd | 10831 | .permission = btrfs_permission, |
1506fcc8 | 10832 | .fiemap = btrfs_fiemap, |
4e34e719 | 10833 | .get_acl = btrfs_get_acl, |
996a710d | 10834 | .set_acl = btrfs_set_acl, |
e41f941a | 10835 | .update_time = btrfs_update_time, |
39279cc3 | 10836 | }; |
6e1d5dcc | 10837 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
10838 | .getattr = btrfs_getattr, |
10839 | .setattr = btrfs_setattr, | |
fdebe2bd | 10840 | .permission = btrfs_permission, |
33268eaf | 10841 | .listxattr = btrfs_listxattr, |
4e34e719 | 10842 | .get_acl = btrfs_get_acl, |
996a710d | 10843 | .set_acl = btrfs_set_acl, |
e41f941a | 10844 | .update_time = btrfs_update_time, |
618e21d5 | 10845 | }; |
6e1d5dcc | 10846 | static const struct inode_operations btrfs_symlink_inode_operations = { |
6b255391 | 10847 | .get_link = page_get_link, |
f209561a | 10848 | .getattr = btrfs_getattr, |
22c44fe6 | 10849 | .setattr = btrfs_setattr, |
fdebe2bd | 10850 | .permission = btrfs_permission, |
0279b4cd | 10851 | .listxattr = btrfs_listxattr, |
e41f941a | 10852 | .update_time = btrfs_update_time, |
39279cc3 | 10853 | }; |
76dda93c | 10854 | |
82d339d9 | 10855 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c | 10856 | .d_delete = btrfs_dentry_delete, |
b4aff1f8 | 10857 | .d_release = btrfs_dentry_release, |
76dda93c | 10858 | }; |