]>
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> |
39279cc3 CM |
45 | #include "ctree.h" |
46 | #include "disk-io.h" | |
47 | #include "transaction.h" | |
48 | #include "btrfs_inode.h" | |
39279cc3 | 49 | #include "print-tree.h" |
e6dcd2dc | 50 | #include "ordered-data.h" |
95819c05 | 51 | #include "xattr.h" |
e02119d5 | 52 | #include "tree-log.h" |
4a54c8c1 | 53 | #include "volumes.h" |
c8b97818 | 54 | #include "compression.h" |
b4ce94de | 55 | #include "locking.h" |
dc89e982 | 56 | #include "free-space-cache.h" |
581bb050 | 57 | #include "inode-map.h" |
38c227d8 | 58 | #include "backref.h" |
f23b5a59 | 59 | #include "hash.h" |
63541927 | 60 | #include "props.h" |
31193213 | 61 | #include "qgroup.h" |
dda3245e | 62 | #include "dedupe.h" |
39279cc3 CM |
63 | |
64 | struct btrfs_iget_args { | |
90d3e592 | 65 | struct btrfs_key *location; |
39279cc3 CM |
66 | struct btrfs_root *root; |
67 | }; | |
68 | ||
f28a4928 FM |
69 | struct btrfs_dio_data { |
70 | u64 outstanding_extents; | |
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; | |
89 | struct kmem_cache *btrfs_transaction_cachep; | |
39279cc3 | 90 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 91 | struct kmem_cache *btrfs_free_space_cachep; |
39279cc3 CM |
92 | |
93 | #define S_SHIFT 12 | |
4d4ab6d6 | 94 | static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { |
39279cc3 CM |
95 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, |
96 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
97 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
98 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
99 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
100 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
101 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
102 | }; | |
103 | ||
3972f260 | 104 | static int btrfs_setsize(struct inode *inode, struct iattr *attr); |
a41ad394 | 105 | static int btrfs_truncate(struct inode *inode); |
5fd02043 | 106 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent); |
771ed689 CM |
107 | static noinline int cow_file_range(struct inode *inode, |
108 | struct page *locked_page, | |
dda3245e WX |
109 | u64 start, u64 end, u64 delalloc_end, |
110 | int *page_started, unsigned long *nr_written, | |
111 | int unlock, struct btrfs_dedupe_hash *hash); | |
6f9994db LB |
112 | static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len, |
113 | u64 orig_start, u64 block_start, | |
114 | u64 block_len, u64 orig_block_len, | |
115 | u64 ram_bytes, int compress_type, | |
116 | int type); | |
7b128766 | 117 | |
48a3b636 | 118 | static int btrfs_dirty_inode(struct inode *inode); |
7b128766 | 119 | |
6a3891c5 JB |
120 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
121 | void btrfs_test_inode_set_ops(struct inode *inode) | |
122 | { | |
123 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; | |
124 | } | |
125 | #endif | |
126 | ||
f34f57a3 | 127 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
2a7dba39 EP |
128 | struct inode *inode, struct inode *dir, |
129 | const struct qstr *qstr) | |
0279b4cd JO |
130 | { |
131 | int err; | |
132 | ||
f34f57a3 | 133 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 134 | if (!err) |
2a7dba39 | 135 | err = btrfs_xattr_security_init(trans, inode, dir, qstr); |
0279b4cd JO |
136 | return err; |
137 | } | |
138 | ||
c8b97818 CM |
139 | /* |
140 | * this does all the hard work for inserting an inline extent into | |
141 | * the btree. The caller should have done a btrfs_drop_extents so that | |
142 | * no overlapping inline items exist in the btree | |
143 | */ | |
40f76580 | 144 | static int insert_inline_extent(struct btrfs_trans_handle *trans, |
1acae57b | 145 | struct btrfs_path *path, int extent_inserted, |
c8b97818 CM |
146 | struct btrfs_root *root, struct inode *inode, |
147 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 148 | int compress_type, |
c8b97818 CM |
149 | struct page **compressed_pages) |
150 | { | |
c8b97818 CM |
151 | struct extent_buffer *leaf; |
152 | struct page *page = NULL; | |
153 | char *kaddr; | |
154 | unsigned long ptr; | |
155 | struct btrfs_file_extent_item *ei; | |
156 | int err = 0; | |
157 | int ret; | |
158 | size_t cur_size = size; | |
c8b97818 | 159 | unsigned long offset; |
c8b97818 | 160 | |
fe3f566c | 161 | if (compressed_size && compressed_pages) |
c8b97818 | 162 | cur_size = compressed_size; |
c8b97818 | 163 | |
1acae57b | 164 | inode_add_bytes(inode, size); |
c8b97818 | 165 | |
1acae57b FDBM |
166 | if (!extent_inserted) { |
167 | struct btrfs_key key; | |
168 | size_t datasize; | |
c8b97818 | 169 | |
4a0cc7ca | 170 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
1acae57b | 171 | key.offset = start; |
962a298f | 172 | key.type = BTRFS_EXTENT_DATA_KEY; |
c8b97818 | 173 | |
1acae57b FDBM |
174 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
175 | path->leave_spinning = 1; | |
176 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
177 | datasize); | |
178 | if (ret) { | |
179 | err = ret; | |
180 | goto fail; | |
181 | } | |
c8b97818 CM |
182 | } |
183 | leaf = path->nodes[0]; | |
184 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
185 | struct btrfs_file_extent_item); | |
186 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
187 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
188 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
189 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
190 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
191 | ptr = btrfs_file_extent_inline_start(ei); | |
192 | ||
261507a0 | 193 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
194 | struct page *cpage; |
195 | int i = 0; | |
d397712b | 196 | while (compressed_size > 0) { |
c8b97818 | 197 | cpage = compressed_pages[i]; |
5b050f04 | 198 | cur_size = min_t(unsigned long, compressed_size, |
09cbfeaf | 199 | PAGE_SIZE); |
c8b97818 | 200 | |
7ac687d9 | 201 | kaddr = kmap_atomic(cpage); |
c8b97818 | 202 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
7ac687d9 | 203 | kunmap_atomic(kaddr); |
c8b97818 CM |
204 | |
205 | i++; | |
206 | ptr += cur_size; | |
207 | compressed_size -= cur_size; | |
208 | } | |
209 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 210 | compress_type); |
c8b97818 CM |
211 | } else { |
212 | page = find_get_page(inode->i_mapping, | |
09cbfeaf | 213 | start >> PAGE_SHIFT); |
c8b97818 | 214 | btrfs_set_file_extent_compression(leaf, ei, 0); |
7ac687d9 | 215 | kaddr = kmap_atomic(page); |
09cbfeaf | 216 | offset = start & (PAGE_SIZE - 1); |
c8b97818 | 217 | write_extent_buffer(leaf, kaddr + offset, ptr, size); |
7ac687d9 | 218 | kunmap_atomic(kaddr); |
09cbfeaf | 219 | put_page(page); |
c8b97818 CM |
220 | } |
221 | btrfs_mark_buffer_dirty(leaf); | |
1acae57b | 222 | btrfs_release_path(path); |
c8b97818 | 223 | |
c2167754 YZ |
224 | /* |
225 | * we're an inline extent, so nobody can | |
226 | * extend the file past i_size without locking | |
227 | * a page we already have locked. | |
228 | * | |
229 | * We must do any isize and inode updates | |
230 | * before we unlock the pages. Otherwise we | |
231 | * could end up racing with unlink. | |
232 | */ | |
c8b97818 | 233 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
79787eaa | 234 | ret = btrfs_update_inode(trans, root, inode); |
c2167754 | 235 | |
79787eaa | 236 | return ret; |
c8b97818 | 237 | fail: |
c8b97818 CM |
238 | return err; |
239 | } | |
240 | ||
241 | ||
242 | /* | |
243 | * conditionally insert an inline extent into the file. This | |
244 | * does the checks required to make sure the data is small enough | |
245 | * to fit as an inline extent. | |
246 | */ | |
00361589 JB |
247 | static noinline int cow_file_range_inline(struct btrfs_root *root, |
248 | struct inode *inode, u64 start, | |
249 | u64 end, size_t compressed_size, | |
250 | int compress_type, | |
251 | struct page **compressed_pages) | |
c8b97818 | 252 | { |
0b246afa | 253 | struct btrfs_fs_info *fs_info = root->fs_info; |
00361589 | 254 | struct btrfs_trans_handle *trans; |
c8b97818 CM |
255 | u64 isize = i_size_read(inode); |
256 | u64 actual_end = min(end + 1, isize); | |
257 | u64 inline_len = actual_end - start; | |
0b246afa | 258 | u64 aligned_end = ALIGN(end, fs_info->sectorsize); |
c8b97818 CM |
259 | u64 data_len = inline_len; |
260 | int ret; | |
1acae57b FDBM |
261 | struct btrfs_path *path; |
262 | int extent_inserted = 0; | |
263 | u32 extent_item_size; | |
c8b97818 CM |
264 | |
265 | if (compressed_size) | |
266 | data_len = compressed_size; | |
267 | ||
268 | if (start > 0 || | |
0b246afa JM |
269 | actual_end > fs_info->sectorsize || |
270 | data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) || | |
c8b97818 | 271 | (!compressed_size && |
0b246afa | 272 | (actual_end & (fs_info->sectorsize - 1)) == 0) || |
c8b97818 | 273 | end + 1 < isize || |
0b246afa | 274 | data_len > fs_info->max_inline) { |
c8b97818 CM |
275 | return 1; |
276 | } | |
277 | ||
1acae57b FDBM |
278 | path = btrfs_alloc_path(); |
279 | if (!path) | |
280 | return -ENOMEM; | |
281 | ||
00361589 | 282 | trans = btrfs_join_transaction(root); |
1acae57b FDBM |
283 | if (IS_ERR(trans)) { |
284 | btrfs_free_path(path); | |
00361589 | 285 | return PTR_ERR(trans); |
1acae57b | 286 | } |
0b246afa | 287 | trans->block_rsv = &fs_info->delalloc_block_rsv; |
00361589 | 288 | |
1acae57b FDBM |
289 | if (compressed_size && compressed_pages) |
290 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
291 | compressed_size); | |
292 | else | |
293 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
294 | inline_len); | |
295 | ||
296 | ret = __btrfs_drop_extents(trans, root, inode, path, | |
297 | start, aligned_end, NULL, | |
298 | 1, 1, extent_item_size, &extent_inserted); | |
00361589 | 299 | if (ret) { |
66642832 | 300 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
301 | goto out; |
302 | } | |
c8b97818 CM |
303 | |
304 | if (isize > actual_end) | |
305 | inline_len = min_t(u64, isize, actual_end); | |
1acae57b FDBM |
306 | ret = insert_inline_extent(trans, path, extent_inserted, |
307 | root, inode, start, | |
c8b97818 | 308 | inline_len, compressed_size, |
fe3f566c | 309 | compress_type, compressed_pages); |
2adcac1a | 310 | if (ret && ret != -ENOSPC) { |
66642832 | 311 | btrfs_abort_transaction(trans, ret); |
00361589 | 312 | goto out; |
2adcac1a | 313 | } else if (ret == -ENOSPC) { |
00361589 JB |
314 | ret = 1; |
315 | goto out; | |
79787eaa | 316 | } |
2adcac1a | 317 | |
bdc20e67 | 318 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); |
691fa059 | 319 | btrfs_delalloc_release_metadata(BTRFS_I(inode), end + 1 - start); |
dcdbc059 | 320 | btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0); |
00361589 | 321 | out: |
94ed938a QW |
322 | /* |
323 | * Don't forget to free the reserved space, as for inlined extent | |
324 | * it won't count as data extent, free them directly here. | |
325 | * And at reserve time, it's always aligned to page size, so | |
326 | * just free one page here. | |
327 | */ | |
09cbfeaf | 328 | btrfs_qgroup_free_data(inode, 0, PAGE_SIZE); |
1acae57b | 329 | btrfs_free_path(path); |
3a45bb20 | 330 | btrfs_end_transaction(trans); |
00361589 | 331 | return ret; |
c8b97818 CM |
332 | } |
333 | ||
771ed689 CM |
334 | struct async_extent { |
335 | u64 start; | |
336 | u64 ram_size; | |
337 | u64 compressed_size; | |
338 | struct page **pages; | |
339 | unsigned long nr_pages; | |
261507a0 | 340 | int compress_type; |
771ed689 CM |
341 | struct list_head list; |
342 | }; | |
343 | ||
344 | struct async_cow { | |
345 | struct inode *inode; | |
346 | struct btrfs_root *root; | |
347 | struct page *locked_page; | |
348 | u64 start; | |
349 | u64 end; | |
350 | struct list_head extents; | |
351 | struct btrfs_work work; | |
352 | }; | |
353 | ||
354 | static noinline int add_async_extent(struct async_cow *cow, | |
355 | u64 start, u64 ram_size, | |
356 | u64 compressed_size, | |
357 | struct page **pages, | |
261507a0 LZ |
358 | unsigned long nr_pages, |
359 | int compress_type) | |
771ed689 CM |
360 | { |
361 | struct async_extent *async_extent; | |
362 | ||
363 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 364 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
365 | async_extent->start = start; |
366 | async_extent->ram_size = ram_size; | |
367 | async_extent->compressed_size = compressed_size; | |
368 | async_extent->pages = pages; | |
369 | async_extent->nr_pages = nr_pages; | |
261507a0 | 370 | async_extent->compress_type = compress_type; |
771ed689 CM |
371 | list_add_tail(&async_extent->list, &cow->extents); |
372 | return 0; | |
373 | } | |
374 | ||
f79707b0 WS |
375 | static inline int inode_need_compress(struct inode *inode) |
376 | { | |
0b246afa | 377 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
f79707b0 WS |
378 | |
379 | /* force compress */ | |
0b246afa | 380 | if (btrfs_test_opt(fs_info, FORCE_COMPRESS)) |
f79707b0 WS |
381 | return 1; |
382 | /* bad compression ratios */ | |
383 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) | |
384 | return 0; | |
0b246afa | 385 | if (btrfs_test_opt(fs_info, COMPRESS) || |
f79707b0 WS |
386 | BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS || |
387 | BTRFS_I(inode)->force_compress) | |
388 | return 1; | |
389 | return 0; | |
390 | } | |
391 | ||
6158e1ce | 392 | static inline void inode_should_defrag(struct btrfs_inode *inode, |
26d30f85 AJ |
393 | u64 start, u64 end, u64 num_bytes, u64 small_write) |
394 | { | |
395 | /* If this is a small write inside eof, kick off a defrag */ | |
396 | if (num_bytes < small_write && | |
6158e1ce | 397 | (start > 0 || end + 1 < inode->disk_i_size)) |
26d30f85 AJ |
398 | btrfs_add_inode_defrag(NULL, inode); |
399 | } | |
400 | ||
d352ac68 | 401 | /* |
771ed689 CM |
402 | * we create compressed extents in two phases. The first |
403 | * phase compresses a range of pages that have already been | |
404 | * locked (both pages and state bits are locked). | |
c8b97818 | 405 | * |
771ed689 CM |
406 | * This is done inside an ordered work queue, and the compression |
407 | * is spread across many cpus. The actual IO submission is step | |
408 | * two, and the ordered work queue takes care of making sure that | |
409 | * happens in the same order things were put onto the queue by | |
410 | * writepages and friends. | |
c8b97818 | 411 | * |
771ed689 CM |
412 | * If this code finds it can't get good compression, it puts an |
413 | * entry onto the work queue to write the uncompressed bytes. This | |
414 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
415 | * are written in the same order that the flusher thread sent them |
416 | * down. | |
d352ac68 | 417 | */ |
c44f649e | 418 | static noinline void compress_file_range(struct inode *inode, |
771ed689 CM |
419 | struct page *locked_page, |
420 | u64 start, u64 end, | |
421 | struct async_cow *async_cow, | |
422 | int *num_added) | |
b888db2b | 423 | { |
0b246afa | 424 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
b888db2b | 425 | struct btrfs_root *root = BTRFS_I(inode)->root; |
db94535d | 426 | u64 num_bytes; |
0b246afa | 427 | u64 blocksize = fs_info->sectorsize; |
c8b97818 | 428 | u64 actual_end; |
42dc7bab | 429 | u64 isize = i_size_read(inode); |
e6dcd2dc | 430 | int ret = 0; |
c8b97818 CM |
431 | struct page **pages = NULL; |
432 | unsigned long nr_pages; | |
c8b97818 CM |
433 | unsigned long total_compressed = 0; |
434 | unsigned long total_in = 0; | |
c8b97818 CM |
435 | int i; |
436 | int will_compress; | |
0b246afa | 437 | int compress_type = fs_info->compress_type; |
4adaa611 | 438 | int redirty = 0; |
b888db2b | 439 | |
6158e1ce NB |
440 | inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1, |
441 | SZ_16K); | |
4cb5300b | 442 | |
42dc7bab | 443 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
444 | again: |
445 | will_compress = 0; | |
09cbfeaf | 446 | nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1; |
069eac78 DS |
447 | BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0); |
448 | nr_pages = min_t(unsigned long, nr_pages, | |
449 | BTRFS_MAX_COMPRESSED / PAGE_SIZE); | |
be20aa9d | 450 | |
f03d9301 CM |
451 | /* |
452 | * we don't want to send crud past the end of i_size through | |
453 | * compression, that's just a waste of CPU time. So, if the | |
454 | * end of the file is before the start of our current | |
455 | * requested range of bytes, we bail out to the uncompressed | |
456 | * cleanup code that can deal with all of this. | |
457 | * | |
458 | * It isn't really the fastest way to fix things, but this is a | |
459 | * very uncommon corner. | |
460 | */ | |
461 | if (actual_end <= start) | |
462 | goto cleanup_and_bail_uncompressed; | |
463 | ||
c8b97818 CM |
464 | total_compressed = actual_end - start; |
465 | ||
4bcbb332 SW |
466 | /* |
467 | * skip compression for a small file range(<=blocksize) that | |
01327610 | 468 | * isn't an inline extent, since it doesn't save disk space at all. |
4bcbb332 SW |
469 | */ |
470 | if (total_compressed <= blocksize && | |
471 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
472 | goto cleanup_and_bail_uncompressed; | |
473 | ||
069eac78 DS |
474 | total_compressed = min_t(unsigned long, total_compressed, |
475 | BTRFS_MAX_UNCOMPRESSED); | |
fda2832f | 476 | num_bytes = ALIGN(end - start + 1, blocksize); |
be20aa9d | 477 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
478 | total_in = 0; |
479 | ret = 0; | |
db94535d | 480 | |
771ed689 CM |
481 | /* |
482 | * we do compression for mount -o compress and when the | |
483 | * inode has not been flagged as nocompress. This flag can | |
484 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 485 | */ |
f79707b0 | 486 | if (inode_need_compress(inode)) { |
c8b97818 | 487 | WARN_ON(pages); |
31e818fe | 488 | pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS); |
560f7d75 LZ |
489 | if (!pages) { |
490 | /* just bail out to the uncompressed code */ | |
491 | goto cont; | |
492 | } | |
c8b97818 | 493 | |
261507a0 LZ |
494 | if (BTRFS_I(inode)->force_compress) |
495 | compress_type = BTRFS_I(inode)->force_compress; | |
496 | ||
4adaa611 CM |
497 | /* |
498 | * we need to call clear_page_dirty_for_io on each | |
499 | * page in the range. Otherwise applications with the file | |
500 | * mmap'd can wander in and change the page contents while | |
501 | * we are compressing them. | |
502 | * | |
503 | * If the compression fails for any reason, we set the pages | |
504 | * dirty again later on. | |
505 | */ | |
506 | extent_range_clear_dirty_for_io(inode, start, end); | |
507 | redirty = 1; | |
261507a0 LZ |
508 | ret = btrfs_compress_pages(compress_type, |
509 | inode->i_mapping, start, | |
38c31464 | 510 | pages, |
4d3a800e | 511 | &nr_pages, |
261507a0 | 512 | &total_in, |
e5d74902 | 513 | &total_compressed); |
c8b97818 CM |
514 | |
515 | if (!ret) { | |
516 | unsigned long offset = total_compressed & | |
09cbfeaf | 517 | (PAGE_SIZE - 1); |
4d3a800e | 518 | struct page *page = pages[nr_pages - 1]; |
c8b97818 CM |
519 | char *kaddr; |
520 | ||
521 | /* zero the tail end of the last page, we might be | |
522 | * sending it down to disk | |
523 | */ | |
524 | if (offset) { | |
7ac687d9 | 525 | kaddr = kmap_atomic(page); |
c8b97818 | 526 | memset(kaddr + offset, 0, |
09cbfeaf | 527 | PAGE_SIZE - offset); |
7ac687d9 | 528 | kunmap_atomic(kaddr); |
c8b97818 CM |
529 | } |
530 | will_compress = 1; | |
531 | } | |
532 | } | |
560f7d75 | 533 | cont: |
c8b97818 CM |
534 | if (start == 0) { |
535 | /* lets try to make an inline extent */ | |
771ed689 | 536 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 537 | /* we didn't compress the entire range, try |
771ed689 | 538 | * to make an uncompressed inline extent. |
c8b97818 | 539 | */ |
00361589 | 540 | ret = cow_file_range_inline(root, inode, start, end, |
f74670f7 | 541 | 0, BTRFS_COMPRESS_NONE, NULL); |
c8b97818 | 542 | } else { |
771ed689 | 543 | /* try making a compressed inline extent */ |
00361589 | 544 | ret = cow_file_range_inline(root, inode, start, end, |
fe3f566c LZ |
545 | total_compressed, |
546 | compress_type, pages); | |
c8b97818 | 547 | } |
79787eaa | 548 | if (ret <= 0) { |
151a41bc JB |
549 | unsigned long clear_flags = EXTENT_DELALLOC | |
550 | EXTENT_DEFRAG; | |
e6eb4314 FM |
551 | unsigned long page_error_op; |
552 | ||
151a41bc | 553 | clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0; |
e6eb4314 | 554 | page_error_op = ret < 0 ? PAGE_SET_ERROR : 0; |
151a41bc | 555 | |
771ed689 | 556 | /* |
79787eaa JM |
557 | * inline extent creation worked or returned error, |
558 | * we don't need to create any more async work items. | |
559 | * Unlock and free up our temp pages. | |
771ed689 | 560 | */ |
ba8b04c1 QW |
561 | extent_clear_unlock_delalloc(inode, start, end, end, |
562 | NULL, clear_flags, | |
563 | PAGE_UNLOCK | | |
c2790a2e JB |
564 | PAGE_CLEAR_DIRTY | |
565 | PAGE_SET_WRITEBACK | | |
e6eb4314 | 566 | page_error_op | |
c2790a2e | 567 | PAGE_END_WRITEBACK); |
18513091 WX |
568 | btrfs_free_reserved_data_space_noquota(inode, start, |
569 | end - start + 1); | |
c8b97818 CM |
570 | goto free_pages_out; |
571 | } | |
572 | } | |
573 | ||
574 | if (will_compress) { | |
575 | /* | |
576 | * we aren't doing an inline extent round the compressed size | |
577 | * up to a block size boundary so the allocator does sane | |
578 | * things | |
579 | */ | |
fda2832f | 580 | total_compressed = ALIGN(total_compressed, blocksize); |
c8b97818 CM |
581 | |
582 | /* | |
583 | * one last check to make sure the compression is really a | |
584 | * win, compare the page count read with the blocks on disk | |
585 | */ | |
09cbfeaf | 586 | total_in = ALIGN(total_in, PAGE_SIZE); |
c8b97818 CM |
587 | if (total_compressed >= total_in) { |
588 | will_compress = 0; | |
589 | } else { | |
c8b97818 | 590 | num_bytes = total_in; |
c8bb0c8b AS |
591 | *num_added += 1; |
592 | ||
593 | /* | |
594 | * The async work queues will take care of doing actual | |
595 | * allocation on disk for these compressed pages, and | |
596 | * will submit them to the elevator. | |
597 | */ | |
598 | add_async_extent(async_cow, start, num_bytes, | |
4d3a800e | 599 | total_compressed, pages, nr_pages, |
c8bb0c8b AS |
600 | compress_type); |
601 | ||
602 | if (start + num_bytes < end) { | |
603 | start += num_bytes; | |
604 | pages = NULL; | |
605 | cond_resched(); | |
606 | goto again; | |
607 | } | |
608 | return; | |
c8b97818 CM |
609 | } |
610 | } | |
c8bb0c8b | 611 | if (pages) { |
c8b97818 CM |
612 | /* |
613 | * the compression code ran but failed to make things smaller, | |
614 | * free any pages it allocated and our page pointer array | |
615 | */ | |
4d3a800e | 616 | for (i = 0; i < nr_pages; i++) { |
70b99e69 | 617 | WARN_ON(pages[i]->mapping); |
09cbfeaf | 618 | put_page(pages[i]); |
c8b97818 CM |
619 | } |
620 | kfree(pages); | |
621 | pages = NULL; | |
622 | total_compressed = 0; | |
4d3a800e | 623 | nr_pages = 0; |
c8b97818 CM |
624 | |
625 | /* flag the file so we don't compress in the future */ | |
0b246afa | 626 | if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) && |
1e701a32 | 627 | !(BTRFS_I(inode)->force_compress)) { |
a555f810 | 628 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 629 | } |
c8b97818 | 630 | } |
f03d9301 | 631 | cleanup_and_bail_uncompressed: |
c8bb0c8b AS |
632 | /* |
633 | * No compression, but we still need to write the pages in the file | |
634 | * we've been given so far. redirty the locked page if it corresponds | |
635 | * to our extent and set things up for the async work queue to run | |
636 | * cow_file_range to do the normal delalloc dance. | |
637 | */ | |
638 | if (page_offset(locked_page) >= start && | |
639 | page_offset(locked_page) <= end) | |
640 | __set_page_dirty_nobuffers(locked_page); | |
641 | /* unlocked later on in the async handlers */ | |
642 | ||
643 | if (redirty) | |
644 | extent_range_redirty_for_io(inode, start, end); | |
645 | add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0, | |
646 | BTRFS_COMPRESS_NONE); | |
647 | *num_added += 1; | |
3b951516 | 648 | |
c44f649e | 649 | return; |
771ed689 CM |
650 | |
651 | free_pages_out: | |
4d3a800e | 652 | for (i = 0; i < nr_pages; i++) { |
771ed689 | 653 | WARN_ON(pages[i]->mapping); |
09cbfeaf | 654 | put_page(pages[i]); |
771ed689 | 655 | } |
d397712b | 656 | kfree(pages); |
771ed689 | 657 | } |
771ed689 | 658 | |
40ae837b FM |
659 | static void free_async_extent_pages(struct async_extent *async_extent) |
660 | { | |
661 | int i; | |
662 | ||
663 | if (!async_extent->pages) | |
664 | return; | |
665 | ||
666 | for (i = 0; i < async_extent->nr_pages; i++) { | |
667 | WARN_ON(async_extent->pages[i]->mapping); | |
09cbfeaf | 668 | put_page(async_extent->pages[i]); |
40ae837b FM |
669 | } |
670 | kfree(async_extent->pages); | |
671 | async_extent->nr_pages = 0; | |
672 | async_extent->pages = NULL; | |
771ed689 CM |
673 | } |
674 | ||
675 | /* | |
676 | * phase two of compressed writeback. This is the ordered portion | |
677 | * of the code, which only gets called in the order the work was | |
678 | * queued. We walk all the async extents created by compress_file_range | |
679 | * and send them down to the disk. | |
680 | */ | |
dec8f175 | 681 | static noinline void submit_compressed_extents(struct inode *inode, |
771ed689 CM |
682 | struct async_cow *async_cow) |
683 | { | |
0b246afa | 684 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
771ed689 CM |
685 | struct async_extent *async_extent; |
686 | u64 alloc_hint = 0; | |
771ed689 CM |
687 | struct btrfs_key ins; |
688 | struct extent_map *em; | |
689 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
771ed689 | 690 | struct extent_io_tree *io_tree; |
f5a84ee3 | 691 | int ret = 0; |
771ed689 | 692 | |
3e04e7f1 | 693 | again: |
d397712b | 694 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
695 | async_extent = list_entry(async_cow->extents.next, |
696 | struct async_extent, list); | |
697 | list_del(&async_extent->list); | |
c8b97818 | 698 | |
771ed689 CM |
699 | io_tree = &BTRFS_I(inode)->io_tree; |
700 | ||
f5a84ee3 | 701 | retry: |
771ed689 CM |
702 | /* did the compression code fall back to uncompressed IO? */ |
703 | if (!async_extent->pages) { | |
704 | int page_started = 0; | |
705 | unsigned long nr_written = 0; | |
706 | ||
707 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 | 708 | async_extent->start + |
d0082371 | 709 | async_extent->ram_size - 1); |
771ed689 CM |
710 | |
711 | /* allocate blocks */ | |
f5a84ee3 JB |
712 | ret = cow_file_range(inode, async_cow->locked_page, |
713 | async_extent->start, | |
714 | async_extent->start + | |
715 | async_extent->ram_size - 1, | |
dda3245e WX |
716 | async_extent->start + |
717 | async_extent->ram_size - 1, | |
718 | &page_started, &nr_written, 0, | |
719 | NULL); | |
771ed689 | 720 | |
79787eaa JM |
721 | /* JDM XXX */ |
722 | ||
771ed689 CM |
723 | /* |
724 | * if page_started, cow_file_range inserted an | |
725 | * inline extent and took care of all the unlocking | |
726 | * and IO for us. Otherwise, we need to submit | |
727 | * all those pages down to the drive. | |
728 | */ | |
f5a84ee3 | 729 | if (!page_started && !ret) |
771ed689 CM |
730 | extent_write_locked_range(io_tree, |
731 | inode, async_extent->start, | |
d397712b | 732 | async_extent->start + |
771ed689 CM |
733 | async_extent->ram_size - 1, |
734 | btrfs_get_extent, | |
735 | WB_SYNC_ALL); | |
3e04e7f1 JB |
736 | else if (ret) |
737 | unlock_page(async_cow->locked_page); | |
771ed689 CM |
738 | kfree(async_extent); |
739 | cond_resched(); | |
740 | continue; | |
741 | } | |
742 | ||
743 | lock_extent(io_tree, async_extent->start, | |
d0082371 | 744 | async_extent->start + async_extent->ram_size - 1); |
771ed689 | 745 | |
18513091 | 746 | ret = btrfs_reserve_extent(root, async_extent->ram_size, |
771ed689 CM |
747 | async_extent->compressed_size, |
748 | async_extent->compressed_size, | |
e570fd27 | 749 | 0, alloc_hint, &ins, 1, 1); |
f5a84ee3 | 750 | if (ret) { |
40ae837b | 751 | free_async_extent_pages(async_extent); |
3e04e7f1 | 752 | |
fdf8e2ea JB |
753 | if (ret == -ENOSPC) { |
754 | unlock_extent(io_tree, async_extent->start, | |
755 | async_extent->start + | |
756 | async_extent->ram_size - 1); | |
ce62003f LB |
757 | |
758 | /* | |
759 | * we need to redirty the pages if we decide to | |
760 | * fallback to uncompressed IO, otherwise we | |
761 | * will not submit these pages down to lower | |
762 | * layers. | |
763 | */ | |
764 | extent_range_redirty_for_io(inode, | |
765 | async_extent->start, | |
766 | async_extent->start + | |
767 | async_extent->ram_size - 1); | |
768 | ||
79787eaa | 769 | goto retry; |
fdf8e2ea | 770 | } |
3e04e7f1 | 771 | goto out_free; |
f5a84ee3 | 772 | } |
c2167754 YZ |
773 | /* |
774 | * here we're doing allocation and writeback of the | |
775 | * compressed pages | |
776 | */ | |
6f9994db LB |
777 | em = create_io_em(inode, async_extent->start, |
778 | async_extent->ram_size, /* len */ | |
779 | async_extent->start, /* orig_start */ | |
780 | ins.objectid, /* block_start */ | |
781 | ins.offset, /* block_len */ | |
782 | ins.offset, /* orig_block_len */ | |
783 | async_extent->ram_size, /* ram_bytes */ | |
784 | async_extent->compress_type, | |
785 | BTRFS_ORDERED_COMPRESSED); | |
786 | if (IS_ERR(em)) | |
787 | /* ret value is not necessary due to void function */ | |
3e04e7f1 | 788 | goto out_free_reserve; |
6f9994db | 789 | free_extent_map(em); |
3e04e7f1 | 790 | |
261507a0 LZ |
791 | ret = btrfs_add_ordered_extent_compress(inode, |
792 | async_extent->start, | |
793 | ins.objectid, | |
794 | async_extent->ram_size, | |
795 | ins.offset, | |
796 | BTRFS_ORDERED_COMPRESSED, | |
797 | async_extent->compress_type); | |
d9f85963 | 798 | if (ret) { |
dcdbc059 NB |
799 | btrfs_drop_extent_cache(BTRFS_I(inode), |
800 | async_extent->start, | |
d9f85963 FM |
801 | async_extent->start + |
802 | async_extent->ram_size - 1, 0); | |
3e04e7f1 | 803 | goto out_free_reserve; |
d9f85963 | 804 | } |
0b246afa | 805 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
771ed689 | 806 | |
771ed689 CM |
807 | /* |
808 | * clear dirty, set writeback and unlock the pages. | |
809 | */ | |
c2790a2e | 810 | extent_clear_unlock_delalloc(inode, async_extent->start, |
ba8b04c1 QW |
811 | async_extent->start + |
812 | async_extent->ram_size - 1, | |
a791e35e CM |
813 | async_extent->start + |
814 | async_extent->ram_size - 1, | |
151a41bc JB |
815 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC, |
816 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
c2790a2e | 817 | PAGE_SET_WRITEBACK); |
771ed689 | 818 | ret = btrfs_submit_compressed_write(inode, |
d397712b CM |
819 | async_extent->start, |
820 | async_extent->ram_size, | |
821 | ins.objectid, | |
822 | ins.offset, async_extent->pages, | |
823 | async_extent->nr_pages); | |
fce2a4e6 FM |
824 | if (ret) { |
825 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
826 | struct page *p = async_extent->pages[0]; | |
827 | const u64 start = async_extent->start; | |
828 | const u64 end = start + async_extent->ram_size - 1; | |
829 | ||
830 | p->mapping = inode->i_mapping; | |
831 | tree->ops->writepage_end_io_hook(p, start, end, | |
832 | NULL, 0); | |
833 | p->mapping = NULL; | |
ba8b04c1 QW |
834 | extent_clear_unlock_delalloc(inode, start, end, end, |
835 | NULL, 0, | |
fce2a4e6 FM |
836 | PAGE_END_WRITEBACK | |
837 | PAGE_SET_ERROR); | |
40ae837b | 838 | free_async_extent_pages(async_extent); |
fce2a4e6 | 839 | } |
771ed689 CM |
840 | alloc_hint = ins.objectid + ins.offset; |
841 | kfree(async_extent); | |
842 | cond_resched(); | |
843 | } | |
dec8f175 | 844 | return; |
3e04e7f1 | 845 | out_free_reserve: |
0b246afa | 846 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 847 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 848 | out_free: |
c2790a2e | 849 | extent_clear_unlock_delalloc(inode, async_extent->start, |
ba8b04c1 QW |
850 | async_extent->start + |
851 | async_extent->ram_size - 1, | |
3e04e7f1 JB |
852 | async_extent->start + |
853 | async_extent->ram_size - 1, | |
c2790a2e | 854 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc JB |
855 | EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING, |
856 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
704de49d FM |
857 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK | |
858 | PAGE_SET_ERROR); | |
40ae837b | 859 | free_async_extent_pages(async_extent); |
79787eaa | 860 | kfree(async_extent); |
3e04e7f1 | 861 | goto again; |
771ed689 CM |
862 | } |
863 | ||
4b46fce2 JB |
864 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
865 | u64 num_bytes) | |
866 | { | |
867 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
868 | struct extent_map *em; | |
869 | u64 alloc_hint = 0; | |
870 | ||
871 | read_lock(&em_tree->lock); | |
872 | em = search_extent_mapping(em_tree, start, num_bytes); | |
873 | if (em) { | |
874 | /* | |
875 | * if block start isn't an actual block number then find the | |
876 | * first block in this inode and use that as a hint. If that | |
877 | * block is also bogus then just don't worry about it. | |
878 | */ | |
879 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
880 | free_extent_map(em); | |
881 | em = search_extent_mapping(em_tree, 0, 0); | |
882 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
883 | alloc_hint = em->block_start; | |
884 | if (em) | |
885 | free_extent_map(em); | |
886 | } else { | |
887 | alloc_hint = em->block_start; | |
888 | free_extent_map(em); | |
889 | } | |
890 | } | |
891 | read_unlock(&em_tree->lock); | |
892 | ||
893 | return alloc_hint; | |
894 | } | |
895 | ||
771ed689 CM |
896 | /* |
897 | * when extent_io.c finds a delayed allocation range in the file, | |
898 | * the call backs end up in this code. The basic idea is to | |
899 | * allocate extents on disk for the range, and create ordered data structs | |
900 | * in ram to track those extents. | |
901 | * | |
902 | * locked_page is the page that writepage had locked already. We use | |
903 | * it to make sure we don't do extra locks or unlocks. | |
904 | * | |
905 | * *page_started is set to one if we unlock locked_page and do everything | |
906 | * required to start IO on it. It may be clean and already done with | |
907 | * IO when we return. | |
908 | */ | |
00361589 JB |
909 | static noinline int cow_file_range(struct inode *inode, |
910 | struct page *locked_page, | |
dda3245e WX |
911 | u64 start, u64 end, u64 delalloc_end, |
912 | int *page_started, unsigned long *nr_written, | |
913 | int unlock, struct btrfs_dedupe_hash *hash) | |
771ed689 | 914 | { |
0b246afa | 915 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
00361589 | 916 | struct btrfs_root *root = BTRFS_I(inode)->root; |
771ed689 CM |
917 | u64 alloc_hint = 0; |
918 | u64 num_bytes; | |
919 | unsigned long ram_size; | |
920 | u64 disk_num_bytes; | |
921 | u64 cur_alloc_size; | |
0b246afa | 922 | u64 blocksize = fs_info->sectorsize; |
771ed689 CM |
923 | struct btrfs_key ins; |
924 | struct extent_map *em; | |
771ed689 CM |
925 | int ret = 0; |
926 | ||
70ddc553 | 927 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) { |
02ecd2c2 | 928 | WARN_ON_ONCE(1); |
29bce2f3 JB |
929 | ret = -EINVAL; |
930 | goto out_unlock; | |
02ecd2c2 | 931 | } |
771ed689 | 932 | |
fda2832f | 933 | num_bytes = ALIGN(end - start + 1, blocksize); |
771ed689 CM |
934 | num_bytes = max(blocksize, num_bytes); |
935 | disk_num_bytes = num_bytes; | |
771ed689 | 936 | |
6158e1ce | 937 | inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K); |
4cb5300b | 938 | |
771ed689 CM |
939 | if (start == 0) { |
940 | /* lets try to make an inline extent */ | |
f74670f7 AJ |
941 | ret = cow_file_range_inline(root, inode, start, end, 0, |
942 | BTRFS_COMPRESS_NONE, NULL); | |
771ed689 | 943 | if (ret == 0) { |
ba8b04c1 QW |
944 | extent_clear_unlock_delalloc(inode, start, end, |
945 | delalloc_end, NULL, | |
c2790a2e | 946 | EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc | 947 | EXTENT_DEFRAG, PAGE_UNLOCK | |
c2790a2e JB |
948 | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
949 | PAGE_END_WRITEBACK); | |
18513091 WX |
950 | btrfs_free_reserved_data_space_noquota(inode, start, |
951 | end - start + 1); | |
771ed689 | 952 | *nr_written = *nr_written + |
09cbfeaf | 953 | (end - start + PAGE_SIZE) / PAGE_SIZE; |
771ed689 | 954 | *page_started = 1; |
771ed689 | 955 | goto out; |
79787eaa | 956 | } else if (ret < 0) { |
79787eaa | 957 | goto out_unlock; |
771ed689 CM |
958 | } |
959 | } | |
960 | ||
961 | BUG_ON(disk_num_bytes > | |
0b246afa | 962 | btrfs_super_total_bytes(fs_info->super_copy)); |
771ed689 | 963 | |
4b46fce2 | 964 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
dcdbc059 NB |
965 | btrfs_drop_extent_cache(BTRFS_I(inode), start, |
966 | start + num_bytes - 1, 0); | |
771ed689 | 967 | |
d397712b | 968 | while (disk_num_bytes > 0) { |
a791e35e CM |
969 | unsigned long op; |
970 | ||
287a0ab9 | 971 | cur_alloc_size = disk_num_bytes; |
18513091 | 972 | ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size, |
0b246afa | 973 | fs_info->sectorsize, 0, alloc_hint, |
e570fd27 | 974 | &ins, 1, 1); |
00361589 | 975 | if (ret < 0) |
79787eaa | 976 | goto out_unlock; |
d397712b | 977 | |
771ed689 | 978 | ram_size = ins.offset; |
6f9994db LB |
979 | em = create_io_em(inode, start, ins.offset, /* len */ |
980 | start, /* orig_start */ | |
981 | ins.objectid, /* block_start */ | |
982 | ins.offset, /* block_len */ | |
983 | ins.offset, /* orig_block_len */ | |
984 | ram_size, /* ram_bytes */ | |
985 | BTRFS_COMPRESS_NONE, /* compress_type */ | |
1af4a0aa | 986 | BTRFS_ORDERED_REGULAR /* type */); |
6f9994db | 987 | if (IS_ERR(em)) |
ace68bac | 988 | goto out_reserve; |
6f9994db | 989 | free_extent_map(em); |
e6dcd2dc | 990 | |
98d20f67 | 991 | cur_alloc_size = ins.offset; |
e6dcd2dc | 992 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 993 | ram_size, cur_alloc_size, 0); |
ace68bac | 994 | if (ret) |
d9f85963 | 995 | goto out_drop_extent_cache; |
c8b97818 | 996 | |
17d217fe YZ |
997 | if (root->root_key.objectid == |
998 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
999 | ret = btrfs_reloc_clone_csums(inode, start, | |
1000 | cur_alloc_size); | |
00361589 | 1001 | if (ret) |
d9f85963 | 1002 | goto out_drop_extent_cache; |
17d217fe YZ |
1003 | } |
1004 | ||
0b246afa | 1005 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
9cfa3e34 | 1006 | |
d397712b | 1007 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 1008 | break; |
d397712b | 1009 | |
c8b97818 CM |
1010 | /* we're not doing compressed IO, don't unlock the first |
1011 | * page (which the caller expects to stay locked), don't | |
1012 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
1013 | * |
1014 | * Do set the Private2 bit so we know this page was properly | |
1015 | * setup for writepage | |
c8b97818 | 1016 | */ |
c2790a2e JB |
1017 | op = unlock ? PAGE_UNLOCK : 0; |
1018 | op |= PAGE_SET_PRIVATE2; | |
a791e35e | 1019 | |
c2790a2e | 1020 | extent_clear_unlock_delalloc(inode, start, |
ba8b04c1 QW |
1021 | start + ram_size - 1, |
1022 | delalloc_end, locked_page, | |
c2790a2e JB |
1023 | EXTENT_LOCKED | EXTENT_DELALLOC, |
1024 | op); | |
c8b97818 | 1025 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
1026 | num_bytes -= cur_alloc_size; |
1027 | alloc_hint = ins.objectid + ins.offset; | |
1028 | start += cur_alloc_size; | |
b888db2b | 1029 | } |
79787eaa | 1030 | out: |
be20aa9d | 1031 | return ret; |
b7d5b0a8 | 1032 | |
d9f85963 | 1033 | out_drop_extent_cache: |
dcdbc059 | 1034 | btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0); |
ace68bac | 1035 | out_reserve: |
0b246afa | 1036 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 1037 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 1038 | out_unlock: |
ba8b04c1 QW |
1039 | extent_clear_unlock_delalloc(inode, start, end, delalloc_end, |
1040 | locked_page, | |
151a41bc JB |
1041 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
1042 | EXTENT_DELALLOC | EXTENT_DEFRAG, | |
1043 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
1044 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK); | |
79787eaa | 1045 | goto out; |
771ed689 | 1046 | } |
c8b97818 | 1047 | |
771ed689 CM |
1048 | /* |
1049 | * work queue call back to started compression on a file and pages | |
1050 | */ | |
1051 | static noinline void async_cow_start(struct btrfs_work *work) | |
1052 | { | |
1053 | struct async_cow *async_cow; | |
1054 | int num_added = 0; | |
1055 | async_cow = container_of(work, struct async_cow, work); | |
1056 | ||
1057 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
1058 | async_cow->start, async_cow->end, async_cow, | |
1059 | &num_added); | |
8180ef88 | 1060 | if (num_added == 0) { |
cb77fcd8 | 1061 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 | 1062 | async_cow->inode = NULL; |
8180ef88 | 1063 | } |
771ed689 CM |
1064 | } |
1065 | ||
1066 | /* | |
1067 | * work queue call back to submit previously compressed pages | |
1068 | */ | |
1069 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1070 | { | |
0b246afa | 1071 | struct btrfs_fs_info *fs_info; |
771ed689 CM |
1072 | struct async_cow *async_cow; |
1073 | struct btrfs_root *root; | |
1074 | unsigned long nr_pages; | |
1075 | ||
1076 | async_cow = container_of(work, struct async_cow, work); | |
1077 | ||
1078 | root = async_cow->root; | |
0b246afa | 1079 | fs_info = root->fs_info; |
09cbfeaf KS |
1080 | nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >> |
1081 | PAGE_SHIFT; | |
771ed689 | 1082 | |
ee863954 DS |
1083 | /* |
1084 | * atomic_sub_return implies a barrier for waitqueue_active | |
1085 | */ | |
0b246afa | 1086 | if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) < |
ee22184b | 1087 | 5 * SZ_1M && |
0b246afa JM |
1088 | waitqueue_active(&fs_info->async_submit_wait)) |
1089 | wake_up(&fs_info->async_submit_wait); | |
771ed689 | 1090 | |
d397712b | 1091 | if (async_cow->inode) |
771ed689 | 1092 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 1093 | } |
c8b97818 | 1094 | |
771ed689 CM |
1095 | static noinline void async_cow_free(struct btrfs_work *work) |
1096 | { | |
1097 | struct async_cow *async_cow; | |
1098 | async_cow = container_of(work, struct async_cow, work); | |
8180ef88 | 1099 | if (async_cow->inode) |
cb77fcd8 | 1100 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 CM |
1101 | kfree(async_cow); |
1102 | } | |
1103 | ||
1104 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
1105 | u64 start, u64 end, int *page_started, | |
1106 | unsigned long *nr_written) | |
1107 | { | |
0b246afa | 1108 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
771ed689 CM |
1109 | struct async_cow *async_cow; |
1110 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1111 | unsigned long nr_pages; | |
1112 | u64 cur_end; | |
771ed689 | 1113 | |
a3429ab7 CM |
1114 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
1115 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 1116 | while (start < end) { |
771ed689 | 1117 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
79787eaa | 1118 | BUG_ON(!async_cow); /* -ENOMEM */ |
8180ef88 | 1119 | async_cow->inode = igrab(inode); |
771ed689 CM |
1120 | async_cow->root = root; |
1121 | async_cow->locked_page = locked_page; | |
1122 | async_cow->start = start; | |
1123 | ||
f79707b0 | 1124 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS && |
0b246afa | 1125 | !btrfs_test_opt(fs_info, FORCE_COMPRESS)) |
771ed689 CM |
1126 | cur_end = end; |
1127 | else | |
ee22184b | 1128 | cur_end = min(end, start + SZ_512K - 1); |
771ed689 CM |
1129 | |
1130 | async_cow->end = cur_end; | |
1131 | INIT_LIST_HEAD(&async_cow->extents); | |
1132 | ||
9e0af237 LB |
1133 | btrfs_init_work(&async_cow->work, |
1134 | btrfs_delalloc_helper, | |
1135 | async_cow_start, async_cow_submit, | |
1136 | async_cow_free); | |
771ed689 | 1137 | |
09cbfeaf KS |
1138 | nr_pages = (cur_end - start + PAGE_SIZE) >> |
1139 | PAGE_SHIFT; | |
0b246afa | 1140 | atomic_add(nr_pages, &fs_info->async_delalloc_pages); |
771ed689 | 1141 | |
0b246afa | 1142 | btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work); |
771ed689 | 1143 | |
0b246afa JM |
1144 | while (atomic_read(&fs_info->async_submit_draining) && |
1145 | atomic_read(&fs_info->async_delalloc_pages)) { | |
1146 | wait_event(fs_info->async_submit_wait, | |
1147 | (atomic_read(&fs_info->async_delalloc_pages) == | |
1148 | 0)); | |
771ed689 CM |
1149 | } |
1150 | ||
1151 | *nr_written += nr_pages; | |
1152 | start = cur_end + 1; | |
1153 | } | |
1154 | *page_started = 1; | |
1155 | return 0; | |
be20aa9d CM |
1156 | } |
1157 | ||
2ff7e61e | 1158 | static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info, |
17d217fe YZ |
1159 | u64 bytenr, u64 num_bytes) |
1160 | { | |
1161 | int ret; | |
1162 | struct btrfs_ordered_sum *sums; | |
1163 | LIST_HEAD(list); | |
1164 | ||
0b246afa | 1165 | ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr, |
a2de733c | 1166 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1167 | if (ret == 0 && list_empty(&list)) |
1168 | return 0; | |
1169 | ||
1170 | while (!list_empty(&list)) { | |
1171 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1172 | list_del(&sums->list); | |
1173 | kfree(sums); | |
1174 | } | |
1175 | return 1; | |
1176 | } | |
1177 | ||
d352ac68 CM |
1178 | /* |
1179 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1180 | * of the extents that exist in the file, and COWs the file as required. | |
1181 | * | |
1182 | * If no cow copies or snapshots exist, we write directly to the existing | |
1183 | * blocks on disk | |
1184 | */ | |
7f366cfe CM |
1185 | static noinline int run_delalloc_nocow(struct inode *inode, |
1186 | struct page *locked_page, | |
771ed689 CM |
1187 | u64 start, u64 end, int *page_started, int force, |
1188 | unsigned long *nr_written) | |
be20aa9d | 1189 | { |
0b246afa | 1190 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
be20aa9d CM |
1191 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1192 | struct extent_buffer *leaf; | |
be20aa9d | 1193 | struct btrfs_path *path; |
80ff3856 | 1194 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1195 | struct btrfs_key found_key; |
6f9994db | 1196 | struct extent_map *em; |
80ff3856 YZ |
1197 | u64 cow_start; |
1198 | u64 cur_offset; | |
1199 | u64 extent_end; | |
5d4f98a2 | 1200 | u64 extent_offset; |
80ff3856 YZ |
1201 | u64 disk_bytenr; |
1202 | u64 num_bytes; | |
b4939680 | 1203 | u64 disk_num_bytes; |
cc95bef6 | 1204 | u64 ram_bytes; |
80ff3856 | 1205 | int extent_type; |
79787eaa | 1206 | int ret, err; |
d899e052 | 1207 | int type; |
80ff3856 YZ |
1208 | int nocow; |
1209 | int check_prev = 1; | |
82d5902d | 1210 | bool nolock; |
4a0cc7ca | 1211 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
be20aa9d CM |
1212 | |
1213 | path = btrfs_alloc_path(); | |
17ca04af | 1214 | if (!path) { |
ba8b04c1 QW |
1215 | extent_clear_unlock_delalloc(inode, start, end, end, |
1216 | locked_page, | |
c2790a2e | 1217 | EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc JB |
1218 | EXTENT_DO_ACCOUNTING | |
1219 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1220 | PAGE_CLEAR_DIRTY | |
1221 | PAGE_SET_WRITEBACK | | |
1222 | PAGE_END_WRITEBACK); | |
d8926bb3 | 1223 | return -ENOMEM; |
17ca04af | 1224 | } |
82d5902d | 1225 | |
70ddc553 | 1226 | nolock = btrfs_is_free_space_inode(BTRFS_I(inode)); |
82d5902d | 1227 | |
80ff3856 YZ |
1228 | cow_start = (u64)-1; |
1229 | cur_offset = start; | |
1230 | while (1) { | |
e4c3b2dc | 1231 | ret = btrfs_lookup_file_extent(NULL, root, path, ino, |
80ff3856 | 1232 | cur_offset, 0); |
d788a349 | 1233 | if (ret < 0) |
79787eaa | 1234 | goto error; |
80ff3856 YZ |
1235 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1236 | leaf = path->nodes[0]; | |
1237 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1238 | path->slots[0] - 1); | |
33345d01 | 1239 | if (found_key.objectid == ino && |
80ff3856 YZ |
1240 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1241 | path->slots[0]--; | |
1242 | } | |
1243 | check_prev = 0; | |
1244 | next_slot: | |
1245 | leaf = path->nodes[0]; | |
1246 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1247 | ret = btrfs_next_leaf(root, path); | |
d788a349 | 1248 | if (ret < 0) |
79787eaa | 1249 | goto error; |
80ff3856 YZ |
1250 | if (ret > 0) |
1251 | break; | |
1252 | leaf = path->nodes[0]; | |
1253 | } | |
be20aa9d | 1254 | |
80ff3856 YZ |
1255 | nocow = 0; |
1256 | disk_bytenr = 0; | |
17d217fe | 1257 | num_bytes = 0; |
80ff3856 YZ |
1258 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1259 | ||
1d512cb7 FM |
1260 | if (found_key.objectid > ino) |
1261 | break; | |
1262 | if (WARN_ON_ONCE(found_key.objectid < ino) || | |
1263 | found_key.type < BTRFS_EXTENT_DATA_KEY) { | |
1264 | path->slots[0]++; | |
1265 | goto next_slot; | |
1266 | } | |
1267 | if (found_key.type > BTRFS_EXTENT_DATA_KEY || | |
80ff3856 YZ |
1268 | found_key.offset > end) |
1269 | break; | |
1270 | ||
1271 | if (found_key.offset > cur_offset) { | |
1272 | extent_end = found_key.offset; | |
e9061e21 | 1273 | extent_type = 0; |
80ff3856 YZ |
1274 | goto out_check; |
1275 | } | |
1276 | ||
1277 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1278 | struct btrfs_file_extent_item); | |
1279 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1280 | ||
cc95bef6 | 1281 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
d899e052 YZ |
1282 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1283 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1284 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1285 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1286 | extent_end = found_key.offset + |
1287 | btrfs_file_extent_num_bytes(leaf, fi); | |
b4939680 JB |
1288 | disk_num_bytes = |
1289 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
80ff3856 YZ |
1290 | if (extent_end <= start) { |
1291 | path->slots[0]++; | |
1292 | goto next_slot; | |
1293 | } | |
17d217fe YZ |
1294 | if (disk_bytenr == 0) |
1295 | goto out_check; | |
80ff3856 YZ |
1296 | if (btrfs_file_extent_compression(leaf, fi) || |
1297 | btrfs_file_extent_encryption(leaf, fi) || | |
1298 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1299 | goto out_check; | |
d899e052 YZ |
1300 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1301 | goto out_check; | |
2ff7e61e | 1302 | if (btrfs_extent_readonly(fs_info, disk_bytenr)) |
80ff3856 | 1303 | goto out_check; |
e4c3b2dc | 1304 | if (btrfs_cross_ref_exist(root, ino, |
5d4f98a2 YZ |
1305 | found_key.offset - |
1306 | extent_offset, disk_bytenr)) | |
17d217fe | 1307 | goto out_check; |
5d4f98a2 | 1308 | disk_bytenr += extent_offset; |
17d217fe YZ |
1309 | disk_bytenr += cur_offset - found_key.offset; |
1310 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
e9894fd3 WS |
1311 | /* |
1312 | * if there are pending snapshots for this root, | |
1313 | * we fall into common COW way. | |
1314 | */ | |
1315 | if (!nolock) { | |
9ea24bbe | 1316 | err = btrfs_start_write_no_snapshoting(root); |
e9894fd3 WS |
1317 | if (!err) |
1318 | goto out_check; | |
1319 | } | |
17d217fe YZ |
1320 | /* |
1321 | * force cow if csum exists in the range. | |
1322 | * this ensure that csum for a given extent are | |
1323 | * either valid or do not exist. | |
1324 | */ | |
2ff7e61e | 1325 | if (csum_exist_in_range(fs_info, disk_bytenr, |
91e1f56a RK |
1326 | num_bytes)) { |
1327 | if (!nolock) | |
1328 | btrfs_end_write_no_snapshoting(root); | |
17d217fe | 1329 | goto out_check; |
91e1f56a RK |
1330 | } |
1331 | if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) { | |
1332 | if (!nolock) | |
1333 | btrfs_end_write_no_snapshoting(root); | |
f78c436c | 1334 | goto out_check; |
91e1f56a | 1335 | } |
80ff3856 YZ |
1336 | nocow = 1; |
1337 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1338 | extent_end = found_key.offset + | |
514ac8ad CM |
1339 | btrfs_file_extent_inline_len(leaf, |
1340 | path->slots[0], fi); | |
da17066c | 1341 | extent_end = ALIGN(extent_end, |
0b246afa | 1342 | fs_info->sectorsize); |
80ff3856 YZ |
1343 | } else { |
1344 | BUG_ON(1); | |
1345 | } | |
1346 | out_check: | |
1347 | if (extent_end <= start) { | |
1348 | path->slots[0]++; | |
e9894fd3 | 1349 | if (!nolock && nocow) |
9ea24bbe | 1350 | btrfs_end_write_no_snapshoting(root); |
f78c436c | 1351 | if (nocow) |
0b246afa | 1352 | btrfs_dec_nocow_writers(fs_info, disk_bytenr); |
80ff3856 YZ |
1353 | goto next_slot; |
1354 | } | |
1355 | if (!nocow) { | |
1356 | if (cow_start == (u64)-1) | |
1357 | cow_start = cur_offset; | |
1358 | cur_offset = extent_end; | |
1359 | if (cur_offset > end) | |
1360 | break; | |
1361 | path->slots[0]++; | |
1362 | goto next_slot; | |
7ea394f1 YZ |
1363 | } |
1364 | ||
b3b4aa74 | 1365 | btrfs_release_path(path); |
80ff3856 | 1366 | if (cow_start != (u64)-1) { |
00361589 JB |
1367 | ret = cow_file_range(inode, locked_page, |
1368 | cow_start, found_key.offset - 1, | |
dda3245e WX |
1369 | end, page_started, nr_written, 1, |
1370 | NULL); | |
e9894fd3 WS |
1371 | if (ret) { |
1372 | if (!nolock && nocow) | |
9ea24bbe | 1373 | btrfs_end_write_no_snapshoting(root); |
f78c436c | 1374 | if (nocow) |
0b246afa | 1375 | btrfs_dec_nocow_writers(fs_info, |
f78c436c | 1376 | disk_bytenr); |
79787eaa | 1377 | goto error; |
e9894fd3 | 1378 | } |
80ff3856 | 1379 | cow_start = (u64)-1; |
7ea394f1 | 1380 | } |
80ff3856 | 1381 | |
d899e052 | 1382 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
6f9994db LB |
1383 | u64 orig_start = found_key.offset - extent_offset; |
1384 | ||
1385 | em = create_io_em(inode, cur_offset, num_bytes, | |
1386 | orig_start, | |
1387 | disk_bytenr, /* block_start */ | |
1388 | num_bytes, /* block_len */ | |
1389 | disk_num_bytes, /* orig_block_len */ | |
1390 | ram_bytes, BTRFS_COMPRESS_NONE, | |
1391 | BTRFS_ORDERED_PREALLOC); | |
1392 | if (IS_ERR(em)) { | |
1393 | if (!nolock && nocow) | |
1394 | btrfs_end_write_no_snapshoting(root); | |
1395 | if (nocow) | |
1396 | btrfs_dec_nocow_writers(fs_info, | |
1397 | disk_bytenr); | |
1398 | ret = PTR_ERR(em); | |
1399 | goto error; | |
d899e052 | 1400 | } |
6f9994db LB |
1401 | free_extent_map(em); |
1402 | } | |
1403 | ||
1404 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
d899e052 YZ |
1405 | type = BTRFS_ORDERED_PREALLOC; |
1406 | } else { | |
1407 | type = BTRFS_ORDERED_NOCOW; | |
1408 | } | |
80ff3856 YZ |
1409 | |
1410 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 | 1411 | num_bytes, num_bytes, type); |
f78c436c | 1412 | if (nocow) |
0b246afa | 1413 | btrfs_dec_nocow_writers(fs_info, disk_bytenr); |
79787eaa | 1414 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 1415 | |
efa56464 YZ |
1416 | if (root->root_key.objectid == |
1417 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1418 | ret = btrfs_reloc_clone_csums(inode, cur_offset, | |
1419 | num_bytes); | |
e9894fd3 WS |
1420 | if (ret) { |
1421 | if (!nolock && nocow) | |
9ea24bbe | 1422 | btrfs_end_write_no_snapshoting(root); |
79787eaa | 1423 | goto error; |
e9894fd3 | 1424 | } |
efa56464 YZ |
1425 | } |
1426 | ||
c2790a2e | 1427 | extent_clear_unlock_delalloc(inode, cur_offset, |
ba8b04c1 | 1428 | cur_offset + num_bytes - 1, end, |
c2790a2e | 1429 | locked_page, EXTENT_LOCKED | |
18513091 WX |
1430 | EXTENT_DELALLOC | |
1431 | EXTENT_CLEAR_DATA_RESV, | |
1432 | PAGE_UNLOCK | PAGE_SET_PRIVATE2); | |
1433 | ||
e9894fd3 | 1434 | if (!nolock && nocow) |
9ea24bbe | 1435 | btrfs_end_write_no_snapshoting(root); |
80ff3856 YZ |
1436 | cur_offset = extent_end; |
1437 | if (cur_offset > end) | |
1438 | break; | |
be20aa9d | 1439 | } |
b3b4aa74 | 1440 | btrfs_release_path(path); |
80ff3856 | 1441 | |
17ca04af | 1442 | if (cur_offset <= end && cow_start == (u64)-1) { |
80ff3856 | 1443 | cow_start = cur_offset; |
17ca04af JB |
1444 | cur_offset = end; |
1445 | } | |
1446 | ||
80ff3856 | 1447 | if (cow_start != (u64)-1) { |
dda3245e WX |
1448 | ret = cow_file_range(inode, locked_page, cow_start, end, end, |
1449 | page_started, nr_written, 1, NULL); | |
d788a349 | 1450 | if (ret) |
79787eaa | 1451 | goto error; |
80ff3856 YZ |
1452 | } |
1453 | ||
79787eaa | 1454 | error: |
17ca04af | 1455 | if (ret && cur_offset < end) |
ba8b04c1 | 1456 | extent_clear_unlock_delalloc(inode, cur_offset, end, end, |
c2790a2e | 1457 | locked_page, EXTENT_LOCKED | |
151a41bc JB |
1458 | EXTENT_DELALLOC | EXTENT_DEFRAG | |
1459 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
1460 | PAGE_CLEAR_DIRTY | | |
c2790a2e JB |
1461 | PAGE_SET_WRITEBACK | |
1462 | PAGE_END_WRITEBACK); | |
7ea394f1 | 1463 | btrfs_free_path(path); |
79787eaa | 1464 | return ret; |
be20aa9d CM |
1465 | } |
1466 | ||
47059d93 WS |
1467 | static inline int need_force_cow(struct inode *inode, u64 start, u64 end) |
1468 | { | |
1469 | ||
1470 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
1471 | !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)) | |
1472 | return 0; | |
1473 | ||
1474 | /* | |
1475 | * @defrag_bytes is a hint value, no spinlock held here, | |
1476 | * if is not zero, it means the file is defragging. | |
1477 | * Force cow if given extent needs to be defragged. | |
1478 | */ | |
1479 | if (BTRFS_I(inode)->defrag_bytes && | |
1480 | test_range_bit(&BTRFS_I(inode)->io_tree, start, end, | |
1481 | EXTENT_DEFRAG, 0, NULL)) | |
1482 | return 1; | |
1483 | ||
1484 | return 0; | |
1485 | } | |
1486 | ||
d352ac68 CM |
1487 | /* |
1488 | * extent_io.c call back to do delayed allocation processing | |
1489 | */ | |
c8b97818 | 1490 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1491 | u64 start, u64 end, int *page_started, |
1492 | unsigned long *nr_written) | |
be20aa9d | 1493 | { |
be20aa9d | 1494 | int ret; |
47059d93 | 1495 | int force_cow = need_force_cow(inode, start, end); |
a2135011 | 1496 | |
47059d93 | 1497 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) { |
c8b97818 | 1498 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1499 | page_started, 1, nr_written); |
47059d93 | 1500 | } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) { |
d899e052 | 1501 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1502 | page_started, 0, nr_written); |
7816030e | 1503 | } else if (!inode_need_compress(inode)) { |
dda3245e WX |
1504 | ret = cow_file_range(inode, locked_page, start, end, end, |
1505 | page_started, nr_written, 1, NULL); | |
7ddf5a42 JB |
1506 | } else { |
1507 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
1508 | &BTRFS_I(inode)->runtime_flags); | |
771ed689 | 1509 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1510 | page_started, nr_written); |
7ddf5a42 | 1511 | } |
b888db2b CM |
1512 | return ret; |
1513 | } | |
1514 | ||
1bf85046 JM |
1515 | static void btrfs_split_extent_hook(struct inode *inode, |
1516 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1517 | { |
dcab6a3b JB |
1518 | u64 size; |
1519 | ||
0ca1f7ce | 1520 | /* not delalloc, ignore it */ |
9ed74f2d | 1521 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1522 | return; |
9ed74f2d | 1523 | |
dcab6a3b JB |
1524 | size = orig->end - orig->start + 1; |
1525 | if (size > BTRFS_MAX_EXTENT_SIZE) { | |
823bb20a | 1526 | u32 num_extents; |
dcab6a3b JB |
1527 | u64 new_size; |
1528 | ||
1529 | /* | |
ba117213 JB |
1530 | * See the explanation in btrfs_merge_extent_hook, the same |
1531 | * applies here, just in reverse. | |
dcab6a3b JB |
1532 | */ |
1533 | new_size = orig->end - split + 1; | |
823bb20a | 1534 | num_extents = count_max_extents(new_size); |
ba117213 | 1535 | new_size = split - orig->start; |
823bb20a DS |
1536 | num_extents += count_max_extents(new_size); |
1537 | if (count_max_extents(size) >= num_extents) | |
dcab6a3b JB |
1538 | return; |
1539 | } | |
1540 | ||
9e0baf60 JB |
1541 | spin_lock(&BTRFS_I(inode)->lock); |
1542 | BTRFS_I(inode)->outstanding_extents++; | |
1543 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1544 | } |
1545 | ||
1546 | /* | |
1547 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1548 | * extents so we can keep track of new extents that are just merged onto old | |
1549 | * extents, such as when we are doing sequential writes, so we can properly | |
1550 | * account for the metadata space we'll need. | |
1551 | */ | |
1bf85046 JM |
1552 | static void btrfs_merge_extent_hook(struct inode *inode, |
1553 | struct extent_state *new, | |
1554 | struct extent_state *other) | |
9ed74f2d | 1555 | { |
dcab6a3b | 1556 | u64 new_size, old_size; |
823bb20a | 1557 | u32 num_extents; |
dcab6a3b | 1558 | |
9ed74f2d JB |
1559 | /* not delalloc, ignore it */ |
1560 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1561 | return; |
9ed74f2d | 1562 | |
8461a3de JB |
1563 | if (new->start > other->start) |
1564 | new_size = new->end - other->start + 1; | |
1565 | else | |
1566 | new_size = other->end - new->start + 1; | |
dcab6a3b JB |
1567 | |
1568 | /* we're not bigger than the max, unreserve the space and go */ | |
1569 | if (new_size <= BTRFS_MAX_EXTENT_SIZE) { | |
1570 | spin_lock(&BTRFS_I(inode)->lock); | |
1571 | BTRFS_I(inode)->outstanding_extents--; | |
1572 | spin_unlock(&BTRFS_I(inode)->lock); | |
1573 | return; | |
1574 | } | |
1575 | ||
1576 | /* | |
ba117213 JB |
1577 | * We have to add up either side to figure out how many extents were |
1578 | * accounted for before we merged into one big extent. If the number of | |
1579 | * extents we accounted for is <= the amount we need for the new range | |
1580 | * then we can return, otherwise drop. Think of it like this | |
1581 | * | |
1582 | * [ 4k][MAX_SIZE] | |
1583 | * | |
1584 | * So we've grown the extent by a MAX_SIZE extent, this would mean we | |
1585 | * need 2 outstanding extents, on one side we have 1 and the other side | |
1586 | * we have 1 so they are == and we can return. But in this case | |
1587 | * | |
1588 | * [MAX_SIZE+4k][MAX_SIZE+4k] | |
1589 | * | |
1590 | * Each range on their own accounts for 2 extents, but merged together | |
1591 | * they are only 3 extents worth of accounting, so we need to drop in | |
1592 | * this case. | |
dcab6a3b | 1593 | */ |
ba117213 | 1594 | old_size = other->end - other->start + 1; |
823bb20a | 1595 | num_extents = count_max_extents(old_size); |
ba117213 | 1596 | old_size = new->end - new->start + 1; |
823bb20a DS |
1597 | num_extents += count_max_extents(old_size); |
1598 | if (count_max_extents(new_size) >= num_extents) | |
dcab6a3b JB |
1599 | return; |
1600 | ||
9e0baf60 JB |
1601 | spin_lock(&BTRFS_I(inode)->lock); |
1602 | BTRFS_I(inode)->outstanding_extents--; | |
1603 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1604 | } |
1605 | ||
eb73c1b7 MX |
1606 | static void btrfs_add_delalloc_inodes(struct btrfs_root *root, |
1607 | struct inode *inode) | |
1608 | { | |
0b246afa JM |
1609 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
1610 | ||
eb73c1b7 MX |
1611 | spin_lock(&root->delalloc_lock); |
1612 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1613 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1614 | &root->delalloc_inodes); | |
1615 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1616 | &BTRFS_I(inode)->runtime_flags); | |
1617 | root->nr_delalloc_inodes++; | |
1618 | if (root->nr_delalloc_inodes == 1) { | |
0b246afa | 1619 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
1620 | BUG_ON(!list_empty(&root->delalloc_root)); |
1621 | list_add_tail(&root->delalloc_root, | |
0b246afa JM |
1622 | &fs_info->delalloc_roots); |
1623 | spin_unlock(&fs_info->delalloc_root_lock); | |
eb73c1b7 MX |
1624 | } |
1625 | } | |
1626 | spin_unlock(&root->delalloc_lock); | |
1627 | } | |
1628 | ||
1629 | static void btrfs_del_delalloc_inode(struct btrfs_root *root, | |
9e3e97f4 | 1630 | struct btrfs_inode *inode) |
eb73c1b7 | 1631 | { |
9e3e97f4 | 1632 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
0b246afa | 1633 | |
eb73c1b7 | 1634 | spin_lock(&root->delalloc_lock); |
9e3e97f4 NB |
1635 | if (!list_empty(&inode->delalloc_inodes)) { |
1636 | list_del_init(&inode->delalloc_inodes); | |
eb73c1b7 | 1637 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 1638 | &inode->runtime_flags); |
eb73c1b7 MX |
1639 | root->nr_delalloc_inodes--; |
1640 | if (!root->nr_delalloc_inodes) { | |
0b246afa | 1641 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
1642 | BUG_ON(list_empty(&root->delalloc_root)); |
1643 | list_del_init(&root->delalloc_root); | |
0b246afa | 1644 | spin_unlock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
1645 | } |
1646 | } | |
1647 | spin_unlock(&root->delalloc_lock); | |
1648 | } | |
1649 | ||
d352ac68 CM |
1650 | /* |
1651 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1652 | * bytes in this file, and to maintain the list of inodes that | |
1653 | * have pending delalloc work to be done. | |
1654 | */ | |
1bf85046 | 1655 | static void btrfs_set_bit_hook(struct inode *inode, |
9ee49a04 | 1656 | struct extent_state *state, unsigned *bits) |
291d673e | 1657 | { |
9ed74f2d | 1658 | |
0b246afa JM |
1659 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
1660 | ||
47059d93 WS |
1661 | if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC)) |
1662 | WARN_ON(1); | |
75eff68e CM |
1663 | /* |
1664 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1665 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1666 | * bit, which is only set or cleared with irqs on |
1667 | */ | |
0ca1f7ce | 1668 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1669 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1670 | u64 len = state->end + 1 - state->start; |
70ddc553 | 1671 | bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode)); |
9ed74f2d | 1672 | |
9e0baf60 | 1673 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1674 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1675 | } else { |
1676 | spin_lock(&BTRFS_I(inode)->lock); | |
1677 | BTRFS_I(inode)->outstanding_extents++; | |
1678 | spin_unlock(&BTRFS_I(inode)->lock); | |
1679 | } | |
287a0ab9 | 1680 | |
6a3891c5 | 1681 | /* For sanity tests */ |
0b246afa | 1682 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
1683 | return; |
1684 | ||
0b246afa JM |
1685 | __percpu_counter_add(&fs_info->delalloc_bytes, len, |
1686 | fs_info->delalloc_batch); | |
df0af1a5 | 1687 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1688 | BTRFS_I(inode)->delalloc_bytes += len; |
47059d93 WS |
1689 | if (*bits & EXTENT_DEFRAG) |
1690 | BTRFS_I(inode)->defrag_bytes += len; | |
df0af1a5 | 1691 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1692 | &BTRFS_I(inode)->runtime_flags)) |
1693 | btrfs_add_delalloc_inodes(root, inode); | |
df0af1a5 | 1694 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1695 | } |
291d673e CM |
1696 | } |
1697 | ||
d352ac68 CM |
1698 | /* |
1699 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1700 | */ | |
6fc0ef68 | 1701 | static void btrfs_clear_bit_hook(struct btrfs_inode *inode, |
41074888 | 1702 | struct extent_state *state, |
9ee49a04 | 1703 | unsigned *bits) |
291d673e | 1704 | { |
6fc0ef68 | 1705 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
47059d93 | 1706 | u64 len = state->end + 1 - state->start; |
823bb20a | 1707 | u32 num_extents = count_max_extents(len); |
47059d93 | 1708 | |
6fc0ef68 | 1709 | spin_lock(&inode->lock); |
47059d93 | 1710 | if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) |
6fc0ef68 NB |
1711 | inode->defrag_bytes -= len; |
1712 | spin_unlock(&inode->lock); | |
47059d93 | 1713 | |
75eff68e CM |
1714 | /* |
1715 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1716 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1717 | * bit, which is only set or cleared with irqs on |
1718 | */ | |
0ca1f7ce | 1719 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
6fc0ef68 | 1720 | struct btrfs_root *root = inode->root; |
83eea1f1 | 1721 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 1722 | |
9e0baf60 | 1723 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1724 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 | 1725 | } else if (!(*bits & EXTENT_DO_ACCOUNTING)) { |
6fc0ef68 NB |
1726 | spin_lock(&inode->lock); |
1727 | inode->outstanding_extents -= num_extents; | |
1728 | spin_unlock(&inode->lock); | |
9e0baf60 | 1729 | } |
0ca1f7ce | 1730 | |
b6d08f06 JB |
1731 | /* |
1732 | * We don't reserve metadata space for space cache inodes so we | |
1733 | * don't need to call dellalloc_release_metadata if there is an | |
1734 | * error. | |
1735 | */ | |
1736 | if (*bits & EXTENT_DO_ACCOUNTING && | |
0b246afa | 1737 | root != fs_info->tree_root) |
0ca1f7ce YZ |
1738 | btrfs_delalloc_release_metadata(inode, len); |
1739 | ||
6a3891c5 | 1740 | /* For sanity tests. */ |
0b246afa | 1741 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
1742 | return; |
1743 | ||
0cb59c99 | 1744 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
18513091 WX |
1745 | && do_list && !(state->state & EXTENT_NORESERVE) |
1746 | && (*bits & (EXTENT_DO_ACCOUNTING | | |
1747 | EXTENT_CLEAR_DATA_RESV))) | |
6fc0ef68 NB |
1748 | btrfs_free_reserved_data_space_noquota( |
1749 | &inode->vfs_inode, | |
51773bec | 1750 | state->start, len); |
9ed74f2d | 1751 | |
0b246afa JM |
1752 | __percpu_counter_add(&fs_info->delalloc_bytes, -len, |
1753 | fs_info->delalloc_batch); | |
6fc0ef68 NB |
1754 | spin_lock(&inode->lock); |
1755 | inode->delalloc_bytes -= len; | |
1756 | if (do_list && inode->delalloc_bytes == 0 && | |
df0af1a5 | 1757 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
9e3e97f4 | 1758 | &inode->runtime_flags)) |
eb73c1b7 | 1759 | btrfs_del_delalloc_inode(root, inode); |
6fc0ef68 | 1760 | spin_unlock(&inode->lock); |
291d673e | 1761 | } |
291d673e CM |
1762 | } |
1763 | ||
d352ac68 CM |
1764 | /* |
1765 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1766 | * we don't create bios that span stripes or chunks | |
6f034ece LB |
1767 | * |
1768 | * return 1 if page cannot be merged to bio | |
1769 | * return 0 if page can be merged to bio | |
1770 | * return error otherwise | |
d352ac68 | 1771 | */ |
81a75f67 | 1772 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1773 | size_t size, struct bio *bio, |
1774 | unsigned long bio_flags) | |
239b14b3 | 1775 | { |
0b246afa JM |
1776 | struct inode *inode = page->mapping->host; |
1777 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
4f024f37 | 1778 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
239b14b3 CM |
1779 | u64 length = 0; |
1780 | u64 map_length; | |
239b14b3 CM |
1781 | int ret; |
1782 | ||
771ed689 CM |
1783 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1784 | return 0; | |
1785 | ||
4f024f37 | 1786 | length = bio->bi_iter.bi_size; |
239b14b3 | 1787 | map_length = length; |
0b246afa JM |
1788 | ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length, |
1789 | NULL, 0); | |
6f034ece LB |
1790 | if (ret < 0) |
1791 | return ret; | |
d397712b | 1792 | if (map_length < length + size) |
239b14b3 | 1793 | return 1; |
3444a972 | 1794 | return 0; |
239b14b3 CM |
1795 | } |
1796 | ||
d352ac68 CM |
1797 | /* |
1798 | * in order to insert checksums into the metadata in large chunks, | |
1799 | * we wait until bio submission time. All the pages in the bio are | |
1800 | * checksummed and sums are attached onto the ordered extent record. | |
1801 | * | |
1802 | * At IO completion time the cums attached on the ordered extent record | |
1803 | * are inserted into the btree | |
1804 | */ | |
81a75f67 MC |
1805 | static int __btrfs_submit_bio_start(struct inode *inode, struct bio *bio, |
1806 | int mirror_num, unsigned long bio_flags, | |
eaf25d93 | 1807 | u64 bio_offset) |
065631f6 | 1808 | { |
065631f6 | 1809 | int ret = 0; |
e015640f | 1810 | |
2ff7e61e | 1811 | ret = btrfs_csum_one_bio(inode, bio, 0, 0); |
79787eaa | 1812 | BUG_ON(ret); /* -ENOMEM */ |
4a69a410 CM |
1813 | return 0; |
1814 | } | |
e015640f | 1815 | |
4a69a410 CM |
1816 | /* |
1817 | * in order to insert checksums into the metadata in large chunks, | |
1818 | * we wait until bio submission time. All the pages in the bio are | |
1819 | * checksummed and sums are attached onto the ordered extent record. | |
1820 | * | |
1821 | * At IO completion time the cums attached on the ordered extent record | |
1822 | * are inserted into the btree | |
1823 | */ | |
81a75f67 | 1824 | static int __btrfs_submit_bio_done(struct inode *inode, struct bio *bio, |
eaf25d93 CM |
1825 | int mirror_num, unsigned long bio_flags, |
1826 | u64 bio_offset) | |
4a69a410 | 1827 | { |
2ff7e61e | 1828 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
61891923 SB |
1829 | int ret; |
1830 | ||
2ff7e61e | 1831 | ret = btrfs_map_bio(fs_info, bio, mirror_num, 1); |
4246a0b6 CH |
1832 | if (ret) { |
1833 | bio->bi_error = ret; | |
1834 | bio_endio(bio); | |
1835 | } | |
61891923 | 1836 | return ret; |
44b8bd7e CM |
1837 | } |
1838 | ||
d352ac68 | 1839 | /* |
cad321ad CM |
1840 | * extent_io.c submission hook. This does the right thing for csum calculation |
1841 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1842 | */ |
81a75f67 | 1843 | static int btrfs_submit_bio_hook(struct inode *inode, struct bio *bio, |
eaf25d93 CM |
1844 | int mirror_num, unsigned long bio_flags, |
1845 | u64 bio_offset) | |
44b8bd7e | 1846 | { |
0b246afa | 1847 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
44b8bd7e | 1848 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0d51e28a | 1849 | enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA; |
44b8bd7e | 1850 | int ret = 0; |
19b9bdb0 | 1851 | int skip_sum; |
b812ce28 | 1852 | int async = !atomic_read(&BTRFS_I(inode)->sync_writers); |
44b8bd7e | 1853 | |
6cbff00f | 1854 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1855 | |
70ddc553 | 1856 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) |
0d51e28a | 1857 | metadata = BTRFS_WQ_ENDIO_FREE_SPACE; |
0417341e | 1858 | |
37226b21 | 1859 | if (bio_op(bio) != REQ_OP_WRITE) { |
0b246afa | 1860 | ret = btrfs_bio_wq_end_io(fs_info, bio, metadata); |
5fd02043 | 1861 | if (ret) |
61891923 | 1862 | goto out; |
5fd02043 | 1863 | |
d20f7043 | 1864 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
61891923 SB |
1865 | ret = btrfs_submit_compressed_read(inode, bio, |
1866 | mirror_num, | |
1867 | bio_flags); | |
1868 | goto out; | |
c2db1073 | 1869 | } else if (!skip_sum) { |
2ff7e61e | 1870 | ret = btrfs_lookup_bio_sums(inode, bio, NULL); |
c2db1073 | 1871 | if (ret) |
61891923 | 1872 | goto out; |
c2db1073 | 1873 | } |
4d1b5fb4 | 1874 | goto mapit; |
b812ce28 | 1875 | } else if (async && !skip_sum) { |
17d217fe YZ |
1876 | /* csum items have already been cloned */ |
1877 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1878 | goto mapit; | |
19b9bdb0 | 1879 | /* we're doing a write, do the async checksumming */ |
0b246afa JM |
1880 | ret = btrfs_wq_submit_bio(fs_info, inode, bio, mirror_num, |
1881 | bio_flags, bio_offset, | |
1882 | __btrfs_submit_bio_start, | |
1883 | __btrfs_submit_bio_done); | |
61891923 | 1884 | goto out; |
b812ce28 | 1885 | } else if (!skip_sum) { |
2ff7e61e | 1886 | ret = btrfs_csum_one_bio(inode, bio, 0, 0); |
b812ce28 JB |
1887 | if (ret) |
1888 | goto out; | |
19b9bdb0 CM |
1889 | } |
1890 | ||
0b86a832 | 1891 | mapit: |
2ff7e61e | 1892 | ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); |
61891923 SB |
1893 | |
1894 | out: | |
4246a0b6 CH |
1895 | if (ret < 0) { |
1896 | bio->bi_error = ret; | |
1897 | bio_endio(bio); | |
1898 | } | |
61891923 | 1899 | return ret; |
065631f6 | 1900 | } |
6885f308 | 1901 | |
d352ac68 CM |
1902 | /* |
1903 | * given a list of ordered sums record them in the inode. This happens | |
1904 | * at IO completion time based on sums calculated at bio submission time. | |
1905 | */ | |
ba1da2f4 | 1906 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
df9f628e | 1907 | struct inode *inode, struct list_head *list) |
e6dcd2dc | 1908 | { |
e6dcd2dc CM |
1909 | struct btrfs_ordered_sum *sum; |
1910 | ||
c6e30871 | 1911 | list_for_each_entry(sum, list, list) { |
39847c4d | 1912 | trans->adding_csums = 1; |
d20f7043 CM |
1913 | btrfs_csum_file_blocks(trans, |
1914 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
39847c4d | 1915 | trans->adding_csums = 0; |
e6dcd2dc CM |
1916 | } |
1917 | return 0; | |
1918 | } | |
1919 | ||
2ac55d41 | 1920 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
ba8b04c1 | 1921 | struct extent_state **cached_state, int dedupe) |
ea8c2819 | 1922 | { |
09cbfeaf | 1923 | WARN_ON((end & (PAGE_SIZE - 1)) == 0); |
ea8c2819 | 1924 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
7cd8c752 | 1925 | cached_state); |
ea8c2819 CM |
1926 | } |
1927 | ||
d352ac68 | 1928 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1929 | struct btrfs_writepage_fixup { |
1930 | struct page *page; | |
1931 | struct btrfs_work work; | |
1932 | }; | |
1933 | ||
b2950863 | 1934 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1935 | { |
1936 | struct btrfs_writepage_fixup *fixup; | |
1937 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 1938 | struct extent_state *cached_state = NULL; |
247e743c CM |
1939 | struct page *page; |
1940 | struct inode *inode; | |
1941 | u64 page_start; | |
1942 | u64 page_end; | |
87826df0 | 1943 | int ret; |
247e743c CM |
1944 | |
1945 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1946 | page = fixup->page; | |
4a096752 | 1947 | again: |
247e743c CM |
1948 | lock_page(page); |
1949 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1950 | ClearPageChecked(page); | |
1951 | goto out_page; | |
1952 | } | |
1953 | ||
1954 | inode = page->mapping->host; | |
1955 | page_start = page_offset(page); | |
09cbfeaf | 1956 | page_end = page_offset(page) + PAGE_SIZE - 1; |
247e743c | 1957 | |
ff13db41 | 1958 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, |
d0082371 | 1959 | &cached_state); |
4a096752 CM |
1960 | |
1961 | /* already ordered? We're done */ | |
8b62b72b | 1962 | if (PagePrivate2(page)) |
247e743c | 1963 | goto out; |
4a096752 | 1964 | |
a776c6fa | 1965 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start, |
09cbfeaf | 1966 | PAGE_SIZE); |
4a096752 | 1967 | if (ordered) { |
2ac55d41 JB |
1968 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
1969 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
1970 | unlock_page(page); |
1971 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 1972 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
1973 | goto again; |
1974 | } | |
247e743c | 1975 | |
7cf5b976 | 1976 | ret = btrfs_delalloc_reserve_space(inode, page_start, |
09cbfeaf | 1977 | PAGE_SIZE); |
87826df0 JM |
1978 | if (ret) { |
1979 | mapping_set_error(page->mapping, ret); | |
1980 | end_extent_writepage(page, ret, page_start, page_end); | |
1981 | ClearPageChecked(page); | |
1982 | goto out; | |
1983 | } | |
1984 | ||
ba8b04c1 QW |
1985 | btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state, |
1986 | 0); | |
247e743c | 1987 | ClearPageChecked(page); |
87826df0 | 1988 | set_page_dirty(page); |
247e743c | 1989 | out: |
2ac55d41 JB |
1990 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
1991 | &cached_state, GFP_NOFS); | |
247e743c CM |
1992 | out_page: |
1993 | unlock_page(page); | |
09cbfeaf | 1994 | put_page(page); |
b897abec | 1995 | kfree(fixup); |
247e743c CM |
1996 | } |
1997 | ||
1998 | /* | |
1999 | * There are a few paths in the higher layers of the kernel that directly | |
2000 | * set the page dirty bit without asking the filesystem if it is a | |
2001 | * good idea. This causes problems because we want to make sure COW | |
2002 | * properly happens and the data=ordered rules are followed. | |
2003 | * | |
c8b97818 | 2004 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
2005 | * hasn't been properly setup for IO. We kick off an async process |
2006 | * to fix it up. The async helper will wait for ordered extents, set | |
2007 | * the delalloc bit and make it safe to write the page. | |
2008 | */ | |
b2950863 | 2009 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
2010 | { |
2011 | struct inode *inode = page->mapping->host; | |
0b246afa | 2012 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
247e743c | 2013 | struct btrfs_writepage_fixup *fixup; |
247e743c | 2014 | |
8b62b72b CM |
2015 | /* this page is properly in the ordered list */ |
2016 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
2017 | return 0; |
2018 | ||
2019 | if (PageChecked(page)) | |
2020 | return -EAGAIN; | |
2021 | ||
2022 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
2023 | if (!fixup) | |
2024 | return -EAGAIN; | |
f421950f | 2025 | |
247e743c | 2026 | SetPageChecked(page); |
09cbfeaf | 2027 | get_page(page); |
9e0af237 LB |
2028 | btrfs_init_work(&fixup->work, btrfs_fixup_helper, |
2029 | btrfs_writepage_fixup_worker, NULL, NULL); | |
247e743c | 2030 | fixup->page = page; |
0b246afa | 2031 | btrfs_queue_work(fs_info->fixup_workers, &fixup->work); |
87826df0 | 2032 | return -EBUSY; |
247e743c CM |
2033 | } |
2034 | ||
d899e052 YZ |
2035 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
2036 | struct inode *inode, u64 file_pos, | |
2037 | u64 disk_bytenr, u64 disk_num_bytes, | |
2038 | u64 num_bytes, u64 ram_bytes, | |
2039 | u8 compression, u8 encryption, | |
2040 | u16 other_encoding, int extent_type) | |
2041 | { | |
2042 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2043 | struct btrfs_file_extent_item *fi; | |
2044 | struct btrfs_path *path; | |
2045 | struct extent_buffer *leaf; | |
2046 | struct btrfs_key ins; | |
1acae57b | 2047 | int extent_inserted = 0; |
d899e052 YZ |
2048 | int ret; |
2049 | ||
2050 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
2051 | if (!path) |
2052 | return -ENOMEM; | |
d899e052 | 2053 | |
a1ed835e CM |
2054 | /* |
2055 | * we may be replacing one extent in the tree with another. | |
2056 | * The new extent is pinned in the extent map, and we don't want | |
2057 | * to drop it from the cache until it is completely in the btree. | |
2058 | * | |
2059 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
2060 | * the caller is expected to unpin it and allow it to be merged | |
2061 | * with the others. | |
2062 | */ | |
1acae57b FDBM |
2063 | ret = __btrfs_drop_extents(trans, root, inode, path, file_pos, |
2064 | file_pos + num_bytes, NULL, 0, | |
2065 | 1, sizeof(*fi), &extent_inserted); | |
79787eaa JM |
2066 | if (ret) |
2067 | goto out; | |
d899e052 | 2068 | |
1acae57b | 2069 | if (!extent_inserted) { |
4a0cc7ca | 2070 | ins.objectid = btrfs_ino(BTRFS_I(inode)); |
1acae57b FDBM |
2071 | ins.offset = file_pos; |
2072 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
2073 | ||
2074 | path->leave_spinning = 1; | |
2075 | ret = btrfs_insert_empty_item(trans, root, path, &ins, | |
2076 | sizeof(*fi)); | |
2077 | if (ret) | |
2078 | goto out; | |
2079 | } | |
d899e052 YZ |
2080 | leaf = path->nodes[0]; |
2081 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2082 | struct btrfs_file_extent_item); | |
2083 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
2084 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
2085 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
2086 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
2087 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
2088 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2089 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
2090 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
2091 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
2092 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 | 2093 | |
d899e052 | 2094 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 2095 | btrfs_release_path(path); |
d899e052 YZ |
2096 | |
2097 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
2098 | |
2099 | ins.objectid = disk_bytenr; | |
2100 | ins.offset = disk_num_bytes; | |
2101 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2ff7e61e | 2102 | ret = btrfs_alloc_reserved_file_extent(trans, root->root_key.objectid, |
f85b7379 | 2103 | btrfs_ino(BTRFS_I(inode)), file_pos, ram_bytes, &ins); |
297d750b | 2104 | /* |
5846a3c2 QW |
2105 | * Release the reserved range from inode dirty range map, as it is |
2106 | * already moved into delayed_ref_head | |
297d750b QW |
2107 | */ |
2108 | btrfs_qgroup_release_data(inode, file_pos, ram_bytes); | |
79787eaa | 2109 | out: |
d899e052 | 2110 | btrfs_free_path(path); |
b9473439 | 2111 | |
79787eaa | 2112 | return ret; |
d899e052 YZ |
2113 | } |
2114 | ||
38c227d8 LB |
2115 | /* snapshot-aware defrag */ |
2116 | struct sa_defrag_extent_backref { | |
2117 | struct rb_node node; | |
2118 | struct old_sa_defrag_extent *old; | |
2119 | u64 root_id; | |
2120 | u64 inum; | |
2121 | u64 file_pos; | |
2122 | u64 extent_offset; | |
2123 | u64 num_bytes; | |
2124 | u64 generation; | |
2125 | }; | |
2126 | ||
2127 | struct old_sa_defrag_extent { | |
2128 | struct list_head list; | |
2129 | struct new_sa_defrag_extent *new; | |
2130 | ||
2131 | u64 extent_offset; | |
2132 | u64 bytenr; | |
2133 | u64 offset; | |
2134 | u64 len; | |
2135 | int count; | |
2136 | }; | |
2137 | ||
2138 | struct new_sa_defrag_extent { | |
2139 | struct rb_root root; | |
2140 | struct list_head head; | |
2141 | struct btrfs_path *path; | |
2142 | struct inode *inode; | |
2143 | u64 file_pos; | |
2144 | u64 len; | |
2145 | u64 bytenr; | |
2146 | u64 disk_len; | |
2147 | u8 compress_type; | |
2148 | }; | |
2149 | ||
2150 | static int backref_comp(struct sa_defrag_extent_backref *b1, | |
2151 | struct sa_defrag_extent_backref *b2) | |
2152 | { | |
2153 | if (b1->root_id < b2->root_id) | |
2154 | return -1; | |
2155 | else if (b1->root_id > b2->root_id) | |
2156 | return 1; | |
2157 | ||
2158 | if (b1->inum < b2->inum) | |
2159 | return -1; | |
2160 | else if (b1->inum > b2->inum) | |
2161 | return 1; | |
2162 | ||
2163 | if (b1->file_pos < b2->file_pos) | |
2164 | return -1; | |
2165 | else if (b1->file_pos > b2->file_pos) | |
2166 | return 1; | |
2167 | ||
2168 | /* | |
2169 | * [------------------------------] ===> (a range of space) | |
2170 | * |<--->| |<---->| =============> (fs/file tree A) | |
2171 | * |<---------------------------->| ===> (fs/file tree B) | |
2172 | * | |
2173 | * A range of space can refer to two file extents in one tree while | |
2174 | * refer to only one file extent in another tree. | |
2175 | * | |
2176 | * So we may process a disk offset more than one time(two extents in A) | |
2177 | * and locate at the same extent(one extent in B), then insert two same | |
2178 | * backrefs(both refer to the extent in B). | |
2179 | */ | |
2180 | return 0; | |
2181 | } | |
2182 | ||
2183 | static void backref_insert(struct rb_root *root, | |
2184 | struct sa_defrag_extent_backref *backref) | |
2185 | { | |
2186 | struct rb_node **p = &root->rb_node; | |
2187 | struct rb_node *parent = NULL; | |
2188 | struct sa_defrag_extent_backref *entry; | |
2189 | int ret; | |
2190 | ||
2191 | while (*p) { | |
2192 | parent = *p; | |
2193 | entry = rb_entry(parent, struct sa_defrag_extent_backref, node); | |
2194 | ||
2195 | ret = backref_comp(backref, entry); | |
2196 | if (ret < 0) | |
2197 | p = &(*p)->rb_left; | |
2198 | else | |
2199 | p = &(*p)->rb_right; | |
2200 | } | |
2201 | ||
2202 | rb_link_node(&backref->node, parent, p); | |
2203 | rb_insert_color(&backref->node, root); | |
2204 | } | |
2205 | ||
2206 | /* | |
2207 | * Note the backref might has changed, and in this case we just return 0. | |
2208 | */ | |
2209 | static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id, | |
2210 | void *ctx) | |
2211 | { | |
2212 | struct btrfs_file_extent_item *extent; | |
38c227d8 LB |
2213 | struct old_sa_defrag_extent *old = ctx; |
2214 | struct new_sa_defrag_extent *new = old->new; | |
2215 | struct btrfs_path *path = new->path; | |
2216 | struct btrfs_key key; | |
2217 | struct btrfs_root *root; | |
2218 | struct sa_defrag_extent_backref *backref; | |
2219 | struct extent_buffer *leaf; | |
2220 | struct inode *inode = new->inode; | |
0b246afa | 2221 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
38c227d8 LB |
2222 | int slot; |
2223 | int ret; | |
2224 | u64 extent_offset; | |
2225 | u64 num_bytes; | |
2226 | ||
2227 | if (BTRFS_I(inode)->root->root_key.objectid == root_id && | |
4a0cc7ca | 2228 | inum == btrfs_ino(BTRFS_I(inode))) |
38c227d8 LB |
2229 | return 0; |
2230 | ||
2231 | key.objectid = root_id; | |
2232 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2233 | key.offset = (u64)-1; | |
2234 | ||
38c227d8 LB |
2235 | root = btrfs_read_fs_root_no_name(fs_info, &key); |
2236 | if (IS_ERR(root)) { | |
2237 | if (PTR_ERR(root) == -ENOENT) | |
2238 | return 0; | |
2239 | WARN_ON(1); | |
ab8d0fc4 | 2240 | btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu", |
38c227d8 LB |
2241 | inum, offset, root_id); |
2242 | return PTR_ERR(root); | |
2243 | } | |
2244 | ||
2245 | key.objectid = inum; | |
2246 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2247 | if (offset > (u64)-1 << 32) | |
2248 | key.offset = 0; | |
2249 | else | |
2250 | key.offset = offset; | |
2251 | ||
2252 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
fae7f21c | 2253 | if (WARN_ON(ret < 0)) |
38c227d8 | 2254 | return ret; |
50f1319c | 2255 | ret = 0; |
38c227d8 LB |
2256 | |
2257 | while (1) { | |
2258 | cond_resched(); | |
2259 | ||
2260 | leaf = path->nodes[0]; | |
2261 | slot = path->slots[0]; | |
2262 | ||
2263 | if (slot >= btrfs_header_nritems(leaf)) { | |
2264 | ret = btrfs_next_leaf(root, path); | |
2265 | if (ret < 0) { | |
2266 | goto out; | |
2267 | } else if (ret > 0) { | |
2268 | ret = 0; | |
2269 | goto out; | |
2270 | } | |
2271 | continue; | |
2272 | } | |
2273 | ||
2274 | path->slots[0]++; | |
2275 | ||
2276 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
2277 | ||
2278 | if (key.objectid > inum) | |
2279 | goto out; | |
2280 | ||
2281 | if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY) | |
2282 | continue; | |
2283 | ||
2284 | extent = btrfs_item_ptr(leaf, slot, | |
2285 | struct btrfs_file_extent_item); | |
2286 | ||
2287 | if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr) | |
2288 | continue; | |
2289 | ||
e68afa49 LB |
2290 | /* |
2291 | * 'offset' refers to the exact key.offset, | |
2292 | * NOT the 'offset' field in btrfs_extent_data_ref, ie. | |
2293 | * (key.offset - extent_offset). | |
2294 | */ | |
2295 | if (key.offset != offset) | |
38c227d8 LB |
2296 | continue; |
2297 | ||
e68afa49 | 2298 | extent_offset = btrfs_file_extent_offset(leaf, extent); |
38c227d8 | 2299 | num_bytes = btrfs_file_extent_num_bytes(leaf, extent); |
e68afa49 | 2300 | |
38c227d8 LB |
2301 | if (extent_offset >= old->extent_offset + old->offset + |
2302 | old->len || extent_offset + num_bytes <= | |
2303 | old->extent_offset + old->offset) | |
2304 | continue; | |
38c227d8 LB |
2305 | break; |
2306 | } | |
2307 | ||
2308 | backref = kmalloc(sizeof(*backref), GFP_NOFS); | |
2309 | if (!backref) { | |
2310 | ret = -ENOENT; | |
2311 | goto out; | |
2312 | } | |
2313 | ||
2314 | backref->root_id = root_id; | |
2315 | backref->inum = inum; | |
e68afa49 | 2316 | backref->file_pos = offset; |
38c227d8 LB |
2317 | backref->num_bytes = num_bytes; |
2318 | backref->extent_offset = extent_offset; | |
2319 | backref->generation = btrfs_file_extent_generation(leaf, extent); | |
2320 | backref->old = old; | |
2321 | backref_insert(&new->root, backref); | |
2322 | old->count++; | |
2323 | out: | |
2324 | btrfs_release_path(path); | |
2325 | WARN_ON(ret); | |
2326 | return ret; | |
2327 | } | |
2328 | ||
2329 | static noinline bool record_extent_backrefs(struct btrfs_path *path, | |
2330 | struct new_sa_defrag_extent *new) | |
2331 | { | |
0b246afa | 2332 | struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb); |
38c227d8 LB |
2333 | struct old_sa_defrag_extent *old, *tmp; |
2334 | int ret; | |
2335 | ||
2336 | new->path = path; | |
2337 | ||
2338 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
e68afa49 LB |
2339 | ret = iterate_inodes_from_logical(old->bytenr + |
2340 | old->extent_offset, fs_info, | |
38c227d8 LB |
2341 | path, record_one_backref, |
2342 | old); | |
4724b106 JB |
2343 | if (ret < 0 && ret != -ENOENT) |
2344 | return false; | |
38c227d8 LB |
2345 | |
2346 | /* no backref to be processed for this extent */ | |
2347 | if (!old->count) { | |
2348 | list_del(&old->list); | |
2349 | kfree(old); | |
2350 | } | |
2351 | } | |
2352 | ||
2353 | if (list_empty(&new->head)) | |
2354 | return false; | |
2355 | ||
2356 | return true; | |
2357 | } | |
2358 | ||
2359 | static int relink_is_mergable(struct extent_buffer *leaf, | |
2360 | struct btrfs_file_extent_item *fi, | |
116e0024 | 2361 | struct new_sa_defrag_extent *new) |
38c227d8 | 2362 | { |
116e0024 | 2363 | if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr) |
38c227d8 LB |
2364 | return 0; |
2365 | ||
2366 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2367 | return 0; | |
2368 | ||
116e0024 LB |
2369 | if (btrfs_file_extent_compression(leaf, fi) != new->compress_type) |
2370 | return 0; | |
2371 | ||
2372 | if (btrfs_file_extent_encryption(leaf, fi) || | |
38c227d8 LB |
2373 | btrfs_file_extent_other_encoding(leaf, fi)) |
2374 | return 0; | |
2375 | ||
2376 | return 1; | |
2377 | } | |
2378 | ||
2379 | /* | |
2380 | * Note the backref might has changed, and in this case we just return 0. | |
2381 | */ | |
2382 | static noinline int relink_extent_backref(struct btrfs_path *path, | |
2383 | struct sa_defrag_extent_backref *prev, | |
2384 | struct sa_defrag_extent_backref *backref) | |
2385 | { | |
2386 | struct btrfs_file_extent_item *extent; | |
2387 | struct btrfs_file_extent_item *item; | |
2388 | struct btrfs_ordered_extent *ordered; | |
2389 | struct btrfs_trans_handle *trans; | |
38c227d8 LB |
2390 | struct btrfs_root *root; |
2391 | struct btrfs_key key; | |
2392 | struct extent_buffer *leaf; | |
2393 | struct old_sa_defrag_extent *old = backref->old; | |
2394 | struct new_sa_defrag_extent *new = old->new; | |
0b246afa | 2395 | struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb); |
38c227d8 LB |
2396 | struct inode *inode; |
2397 | struct extent_state *cached = NULL; | |
2398 | int ret = 0; | |
2399 | u64 start; | |
2400 | u64 len; | |
2401 | u64 lock_start; | |
2402 | u64 lock_end; | |
2403 | bool merge = false; | |
2404 | int index; | |
2405 | ||
2406 | if (prev && prev->root_id == backref->root_id && | |
2407 | prev->inum == backref->inum && | |
2408 | prev->file_pos + prev->num_bytes == backref->file_pos) | |
2409 | merge = true; | |
2410 | ||
2411 | /* step 1: get root */ | |
2412 | key.objectid = backref->root_id; | |
2413 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2414 | key.offset = (u64)-1; | |
2415 | ||
38c227d8 LB |
2416 | index = srcu_read_lock(&fs_info->subvol_srcu); |
2417 | ||
2418 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2419 | if (IS_ERR(root)) { | |
2420 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2421 | if (PTR_ERR(root) == -ENOENT) | |
2422 | return 0; | |
2423 | return PTR_ERR(root); | |
2424 | } | |
38c227d8 | 2425 | |
bcbba5e6 WS |
2426 | if (btrfs_root_readonly(root)) { |
2427 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2428 | return 0; | |
2429 | } | |
2430 | ||
38c227d8 LB |
2431 | /* step 2: get inode */ |
2432 | key.objectid = backref->inum; | |
2433 | key.type = BTRFS_INODE_ITEM_KEY; | |
2434 | key.offset = 0; | |
2435 | ||
2436 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
2437 | if (IS_ERR(inode)) { | |
2438 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2439 | return 0; | |
2440 | } | |
2441 | ||
2442 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2443 | ||
2444 | /* step 3: relink backref */ | |
2445 | lock_start = backref->file_pos; | |
2446 | lock_end = backref->file_pos + backref->num_bytes - 1; | |
2447 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
ff13db41 | 2448 | &cached); |
38c227d8 LB |
2449 | |
2450 | ordered = btrfs_lookup_first_ordered_extent(inode, lock_end); | |
2451 | if (ordered) { | |
2452 | btrfs_put_ordered_extent(ordered); | |
2453 | goto out_unlock; | |
2454 | } | |
2455 | ||
2456 | trans = btrfs_join_transaction(root); | |
2457 | if (IS_ERR(trans)) { | |
2458 | ret = PTR_ERR(trans); | |
2459 | goto out_unlock; | |
2460 | } | |
2461 | ||
2462 | key.objectid = backref->inum; | |
2463 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2464 | key.offset = backref->file_pos; | |
2465 | ||
2466 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2467 | if (ret < 0) { | |
2468 | goto out_free_path; | |
2469 | } else if (ret > 0) { | |
2470 | ret = 0; | |
2471 | goto out_free_path; | |
2472 | } | |
2473 | ||
2474 | extent = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2475 | struct btrfs_file_extent_item); | |
2476 | ||
2477 | if (btrfs_file_extent_generation(path->nodes[0], extent) != | |
2478 | backref->generation) | |
2479 | goto out_free_path; | |
2480 | ||
2481 | btrfs_release_path(path); | |
2482 | ||
2483 | start = backref->file_pos; | |
2484 | if (backref->extent_offset < old->extent_offset + old->offset) | |
2485 | start += old->extent_offset + old->offset - | |
2486 | backref->extent_offset; | |
2487 | ||
2488 | len = min(backref->extent_offset + backref->num_bytes, | |
2489 | old->extent_offset + old->offset + old->len); | |
2490 | len -= max(backref->extent_offset, old->extent_offset + old->offset); | |
2491 | ||
2492 | ret = btrfs_drop_extents(trans, root, inode, start, | |
2493 | start + len, 1); | |
2494 | if (ret) | |
2495 | goto out_free_path; | |
2496 | again: | |
4a0cc7ca | 2497 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
38c227d8 LB |
2498 | key.type = BTRFS_EXTENT_DATA_KEY; |
2499 | key.offset = start; | |
2500 | ||
a09a0a70 | 2501 | path->leave_spinning = 1; |
38c227d8 LB |
2502 | if (merge) { |
2503 | struct btrfs_file_extent_item *fi; | |
2504 | u64 extent_len; | |
2505 | struct btrfs_key found_key; | |
2506 | ||
3c9665df | 2507 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
38c227d8 LB |
2508 | if (ret < 0) |
2509 | goto out_free_path; | |
2510 | ||
2511 | path->slots[0]--; | |
2512 | leaf = path->nodes[0]; | |
2513 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2514 | ||
2515 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2516 | struct btrfs_file_extent_item); | |
2517 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
2518 | ||
116e0024 LB |
2519 | if (extent_len + found_key.offset == start && |
2520 | relink_is_mergable(leaf, fi, new)) { | |
38c227d8 LB |
2521 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2522 | extent_len + len); | |
2523 | btrfs_mark_buffer_dirty(leaf); | |
2524 | inode_add_bytes(inode, len); | |
2525 | ||
2526 | ret = 1; | |
2527 | goto out_free_path; | |
2528 | } else { | |
2529 | merge = false; | |
2530 | btrfs_release_path(path); | |
2531 | goto again; | |
2532 | } | |
2533 | } | |
2534 | ||
2535 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2536 | sizeof(*extent)); | |
2537 | if (ret) { | |
66642832 | 2538 | btrfs_abort_transaction(trans, ret); |
38c227d8 LB |
2539 | goto out_free_path; |
2540 | } | |
2541 | ||
2542 | leaf = path->nodes[0]; | |
2543 | item = btrfs_item_ptr(leaf, path->slots[0], | |
2544 | struct btrfs_file_extent_item); | |
2545 | btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr); | |
2546 | btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len); | |
2547 | btrfs_set_file_extent_offset(leaf, item, start - new->file_pos); | |
2548 | btrfs_set_file_extent_num_bytes(leaf, item, len); | |
2549 | btrfs_set_file_extent_ram_bytes(leaf, item, new->len); | |
2550 | btrfs_set_file_extent_generation(leaf, item, trans->transid); | |
2551 | btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG); | |
2552 | btrfs_set_file_extent_compression(leaf, item, new->compress_type); | |
2553 | btrfs_set_file_extent_encryption(leaf, item, 0); | |
2554 | btrfs_set_file_extent_other_encoding(leaf, item, 0); | |
2555 | ||
2556 | btrfs_mark_buffer_dirty(leaf); | |
2557 | inode_add_bytes(inode, len); | |
a09a0a70 | 2558 | btrfs_release_path(path); |
38c227d8 | 2559 | |
2ff7e61e | 2560 | ret = btrfs_inc_extent_ref(trans, fs_info, new->bytenr, |
38c227d8 LB |
2561 | new->disk_len, 0, |
2562 | backref->root_id, backref->inum, | |
b06c4bf5 | 2563 | new->file_pos); /* start - extent_offset */ |
38c227d8 | 2564 | if (ret) { |
66642832 | 2565 | btrfs_abort_transaction(trans, ret); |
38c227d8 LB |
2566 | goto out_free_path; |
2567 | } | |
2568 | ||
2569 | ret = 1; | |
2570 | out_free_path: | |
2571 | btrfs_release_path(path); | |
a09a0a70 | 2572 | path->leave_spinning = 0; |
3a45bb20 | 2573 | btrfs_end_transaction(trans); |
38c227d8 LB |
2574 | out_unlock: |
2575 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
2576 | &cached, GFP_NOFS); | |
2577 | iput(inode); | |
2578 | return ret; | |
2579 | } | |
2580 | ||
6f519564 LB |
2581 | static void free_sa_defrag_extent(struct new_sa_defrag_extent *new) |
2582 | { | |
2583 | struct old_sa_defrag_extent *old, *tmp; | |
2584 | ||
2585 | if (!new) | |
2586 | return; | |
2587 | ||
2588 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
6f519564 LB |
2589 | kfree(old); |
2590 | } | |
2591 | kfree(new); | |
2592 | } | |
2593 | ||
38c227d8 LB |
2594 | static void relink_file_extents(struct new_sa_defrag_extent *new) |
2595 | { | |
0b246afa | 2596 | struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb); |
38c227d8 | 2597 | struct btrfs_path *path; |
38c227d8 LB |
2598 | struct sa_defrag_extent_backref *backref; |
2599 | struct sa_defrag_extent_backref *prev = NULL; | |
2600 | struct inode *inode; | |
2601 | struct btrfs_root *root; | |
2602 | struct rb_node *node; | |
2603 | int ret; | |
2604 | ||
2605 | inode = new->inode; | |
2606 | root = BTRFS_I(inode)->root; | |
2607 | ||
2608 | path = btrfs_alloc_path(); | |
2609 | if (!path) | |
2610 | return; | |
2611 | ||
2612 | if (!record_extent_backrefs(path, new)) { | |
2613 | btrfs_free_path(path); | |
2614 | goto out; | |
2615 | } | |
2616 | btrfs_release_path(path); | |
2617 | ||
2618 | while (1) { | |
2619 | node = rb_first(&new->root); | |
2620 | if (!node) | |
2621 | break; | |
2622 | rb_erase(node, &new->root); | |
2623 | ||
2624 | backref = rb_entry(node, struct sa_defrag_extent_backref, node); | |
2625 | ||
2626 | ret = relink_extent_backref(path, prev, backref); | |
2627 | WARN_ON(ret < 0); | |
2628 | ||
2629 | kfree(prev); | |
2630 | ||
2631 | if (ret == 1) | |
2632 | prev = backref; | |
2633 | else | |
2634 | prev = NULL; | |
2635 | cond_resched(); | |
2636 | } | |
2637 | kfree(prev); | |
2638 | ||
2639 | btrfs_free_path(path); | |
38c227d8 | 2640 | out: |
6f519564 LB |
2641 | free_sa_defrag_extent(new); |
2642 | ||
0b246afa JM |
2643 | atomic_dec(&fs_info->defrag_running); |
2644 | wake_up(&fs_info->transaction_wait); | |
38c227d8 LB |
2645 | } |
2646 | ||
2647 | static struct new_sa_defrag_extent * | |
2648 | record_old_file_extents(struct inode *inode, | |
2649 | struct btrfs_ordered_extent *ordered) | |
2650 | { | |
0b246afa | 2651 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
38c227d8 LB |
2652 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2653 | struct btrfs_path *path; | |
2654 | struct btrfs_key key; | |
6f519564 | 2655 | struct old_sa_defrag_extent *old; |
38c227d8 LB |
2656 | struct new_sa_defrag_extent *new; |
2657 | int ret; | |
2658 | ||
2659 | new = kmalloc(sizeof(*new), GFP_NOFS); | |
2660 | if (!new) | |
2661 | return NULL; | |
2662 | ||
2663 | new->inode = inode; | |
2664 | new->file_pos = ordered->file_offset; | |
2665 | new->len = ordered->len; | |
2666 | new->bytenr = ordered->start; | |
2667 | new->disk_len = ordered->disk_len; | |
2668 | new->compress_type = ordered->compress_type; | |
2669 | new->root = RB_ROOT; | |
2670 | INIT_LIST_HEAD(&new->head); | |
2671 | ||
2672 | path = btrfs_alloc_path(); | |
2673 | if (!path) | |
2674 | goto out_kfree; | |
2675 | ||
4a0cc7ca | 2676 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
38c227d8 LB |
2677 | key.type = BTRFS_EXTENT_DATA_KEY; |
2678 | key.offset = new->file_pos; | |
2679 | ||
2680 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2681 | if (ret < 0) | |
2682 | goto out_free_path; | |
2683 | if (ret > 0 && path->slots[0] > 0) | |
2684 | path->slots[0]--; | |
2685 | ||
2686 | /* find out all the old extents for the file range */ | |
2687 | while (1) { | |
2688 | struct btrfs_file_extent_item *extent; | |
2689 | struct extent_buffer *l; | |
2690 | int slot; | |
2691 | u64 num_bytes; | |
2692 | u64 offset; | |
2693 | u64 end; | |
2694 | u64 disk_bytenr; | |
2695 | u64 extent_offset; | |
2696 | ||
2697 | l = path->nodes[0]; | |
2698 | slot = path->slots[0]; | |
2699 | ||
2700 | if (slot >= btrfs_header_nritems(l)) { | |
2701 | ret = btrfs_next_leaf(root, path); | |
2702 | if (ret < 0) | |
6f519564 | 2703 | goto out_free_path; |
38c227d8 LB |
2704 | else if (ret > 0) |
2705 | break; | |
2706 | continue; | |
2707 | } | |
2708 | ||
2709 | btrfs_item_key_to_cpu(l, &key, slot); | |
2710 | ||
4a0cc7ca | 2711 | if (key.objectid != btrfs_ino(BTRFS_I(inode))) |
38c227d8 LB |
2712 | break; |
2713 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2714 | break; | |
2715 | if (key.offset >= new->file_pos + new->len) | |
2716 | break; | |
2717 | ||
2718 | extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item); | |
2719 | ||
2720 | num_bytes = btrfs_file_extent_num_bytes(l, extent); | |
2721 | if (key.offset + num_bytes < new->file_pos) | |
2722 | goto next; | |
2723 | ||
2724 | disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent); | |
2725 | if (!disk_bytenr) | |
2726 | goto next; | |
2727 | ||
2728 | extent_offset = btrfs_file_extent_offset(l, extent); | |
2729 | ||
2730 | old = kmalloc(sizeof(*old), GFP_NOFS); | |
2731 | if (!old) | |
6f519564 | 2732 | goto out_free_path; |
38c227d8 LB |
2733 | |
2734 | offset = max(new->file_pos, key.offset); | |
2735 | end = min(new->file_pos + new->len, key.offset + num_bytes); | |
2736 | ||
2737 | old->bytenr = disk_bytenr; | |
2738 | old->extent_offset = extent_offset; | |
2739 | old->offset = offset - key.offset; | |
2740 | old->len = end - offset; | |
2741 | old->new = new; | |
2742 | old->count = 0; | |
2743 | list_add_tail(&old->list, &new->head); | |
2744 | next: | |
2745 | path->slots[0]++; | |
2746 | cond_resched(); | |
2747 | } | |
2748 | ||
2749 | btrfs_free_path(path); | |
0b246afa | 2750 | atomic_inc(&fs_info->defrag_running); |
38c227d8 LB |
2751 | |
2752 | return new; | |
2753 | ||
38c227d8 LB |
2754 | out_free_path: |
2755 | btrfs_free_path(path); | |
2756 | out_kfree: | |
6f519564 | 2757 | free_sa_defrag_extent(new); |
38c227d8 LB |
2758 | return NULL; |
2759 | } | |
2760 | ||
2ff7e61e | 2761 | static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
2762 | u64 start, u64 len) |
2763 | { | |
2764 | struct btrfs_block_group_cache *cache; | |
2765 | ||
0b246afa | 2766 | cache = btrfs_lookup_block_group(fs_info, start); |
e570fd27 MX |
2767 | ASSERT(cache); |
2768 | ||
2769 | spin_lock(&cache->lock); | |
2770 | cache->delalloc_bytes -= len; | |
2771 | spin_unlock(&cache->lock); | |
2772 | ||
2773 | btrfs_put_block_group(cache); | |
2774 | } | |
2775 | ||
d352ac68 CM |
2776 | /* as ordered data IO finishes, this gets called so we can finish |
2777 | * an ordered extent if the range of bytes in the file it covers are | |
2778 | * fully written. | |
2779 | */ | |
5fd02043 | 2780 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 2781 | { |
5fd02043 | 2782 | struct inode *inode = ordered_extent->inode; |
0b246afa | 2783 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc | 2784 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 2785 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 2786 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 2787 | struct extent_state *cached_state = NULL; |
38c227d8 | 2788 | struct new_sa_defrag_extent *new = NULL; |
261507a0 | 2789 | int compress_type = 0; |
77cef2ec JB |
2790 | int ret = 0; |
2791 | u64 logical_len = ordered_extent->len; | |
82d5902d | 2792 | bool nolock; |
77cef2ec | 2793 | bool truncated = false; |
e6dcd2dc | 2794 | |
70ddc553 | 2795 | nolock = btrfs_is_free_space_inode(BTRFS_I(inode)); |
0cb59c99 | 2796 | |
5fd02043 JB |
2797 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
2798 | ret = -EIO; | |
2799 | goto out; | |
2800 | } | |
2801 | ||
7ab7956e NB |
2802 | btrfs_free_io_failure_record(BTRFS_I(inode), |
2803 | ordered_extent->file_offset, | |
2804 | ordered_extent->file_offset + | |
2805 | ordered_extent->len - 1); | |
f612496b | 2806 | |
77cef2ec JB |
2807 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) { |
2808 | truncated = true; | |
2809 | logical_len = ordered_extent->truncated_len; | |
2810 | /* Truncated the entire extent, don't bother adding */ | |
2811 | if (!logical_len) | |
2812 | goto out; | |
2813 | } | |
2814 | ||
c2167754 | 2815 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 2816 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
94ed938a QW |
2817 | |
2818 | /* | |
2819 | * For mwrite(mmap + memset to write) case, we still reserve | |
2820 | * space for NOCOW range. | |
2821 | * As NOCOW won't cause a new delayed ref, just free the space | |
2822 | */ | |
2823 | btrfs_qgroup_free_data(inode, ordered_extent->file_offset, | |
2824 | ordered_extent->len); | |
6c760c07 JB |
2825 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2826 | if (nolock) | |
2827 | trans = btrfs_join_transaction_nolock(root); | |
2828 | else | |
2829 | trans = btrfs_join_transaction(root); | |
2830 | if (IS_ERR(trans)) { | |
2831 | ret = PTR_ERR(trans); | |
2832 | trans = NULL; | |
2833 | goto out; | |
c2167754 | 2834 | } |
0b246afa | 2835 | trans->block_rsv = &fs_info->delalloc_block_rsv; |
6c760c07 JB |
2836 | ret = btrfs_update_inode_fallback(trans, root, inode); |
2837 | if (ret) /* -ENOMEM or corruption */ | |
66642832 | 2838 | btrfs_abort_transaction(trans, ret); |
c2167754 YZ |
2839 | goto out; |
2840 | } | |
e6dcd2dc | 2841 | |
2ac55d41 JB |
2842 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
2843 | ordered_extent->file_offset + ordered_extent->len - 1, | |
ff13db41 | 2844 | &cached_state); |
e6dcd2dc | 2845 | |
38c227d8 LB |
2846 | ret = test_range_bit(io_tree, ordered_extent->file_offset, |
2847 | ordered_extent->file_offset + ordered_extent->len - 1, | |
2848 | EXTENT_DEFRAG, 1, cached_state); | |
2849 | if (ret) { | |
2850 | u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item); | |
8101c8db | 2851 | if (0 && last_snapshot >= BTRFS_I(inode)->generation) |
38c227d8 LB |
2852 | /* the inode is shared */ |
2853 | new = record_old_file_extents(inode, ordered_extent); | |
2854 | ||
2855 | clear_extent_bit(io_tree, ordered_extent->file_offset, | |
2856 | ordered_extent->file_offset + ordered_extent->len - 1, | |
2857 | EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS); | |
2858 | } | |
2859 | ||
0cb59c99 | 2860 | if (nolock) |
7a7eaa40 | 2861 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 2862 | else |
7a7eaa40 | 2863 | trans = btrfs_join_transaction(root); |
79787eaa JM |
2864 | if (IS_ERR(trans)) { |
2865 | ret = PTR_ERR(trans); | |
2866 | trans = NULL; | |
2867 | goto out_unlock; | |
2868 | } | |
a79b7d4b | 2869 | |
0b246afa | 2870 | trans->block_rsv = &fs_info->delalloc_block_rsv; |
c2167754 | 2871 | |
c8b97818 | 2872 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 2873 | compress_type = ordered_extent->compress_type; |
d899e052 | 2874 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 2875 | BUG_ON(compress_type); |
7a6d7067 | 2876 | ret = btrfs_mark_extent_written(trans, BTRFS_I(inode), |
d899e052 YZ |
2877 | ordered_extent->file_offset, |
2878 | ordered_extent->file_offset + | |
77cef2ec | 2879 | logical_len); |
d899e052 | 2880 | } else { |
0b246afa | 2881 | BUG_ON(root == fs_info->tree_root); |
d899e052 YZ |
2882 | ret = insert_reserved_file_extent(trans, inode, |
2883 | ordered_extent->file_offset, | |
2884 | ordered_extent->start, | |
2885 | ordered_extent->disk_len, | |
77cef2ec | 2886 | logical_len, logical_len, |
261507a0 | 2887 | compress_type, 0, 0, |
d899e052 | 2888 | BTRFS_FILE_EXTENT_REG); |
e570fd27 | 2889 | if (!ret) |
2ff7e61e | 2890 | btrfs_release_delalloc_bytes(fs_info, |
e570fd27 MX |
2891 | ordered_extent->start, |
2892 | ordered_extent->disk_len); | |
d899e052 | 2893 | } |
5dc562c5 JB |
2894 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
2895 | ordered_extent->file_offset, ordered_extent->len, | |
2896 | trans->transid); | |
79787eaa | 2897 | if (ret < 0) { |
66642832 | 2898 | btrfs_abort_transaction(trans, ret); |
5fd02043 | 2899 | goto out_unlock; |
79787eaa | 2900 | } |
2ac55d41 | 2901 | |
df9f628e | 2902 | add_pending_csums(trans, inode, &ordered_extent->list); |
e6dcd2dc | 2903 | |
6c760c07 JB |
2904 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2905 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
2906 | if (ret) { /* -ENOMEM or corruption */ | |
66642832 | 2907 | btrfs_abort_transaction(trans, ret); |
6c760c07 | 2908 | goto out_unlock; |
1ef30be1 JB |
2909 | } |
2910 | ret = 0; | |
5fd02043 JB |
2911 | out_unlock: |
2912 | unlock_extent_cached(io_tree, ordered_extent->file_offset, | |
2913 | ordered_extent->file_offset + | |
2914 | ordered_extent->len - 1, &cached_state, GFP_NOFS); | |
c2167754 | 2915 | out: |
0b246afa | 2916 | if (root != fs_info->tree_root) |
691fa059 NB |
2917 | btrfs_delalloc_release_metadata(BTRFS_I(inode), |
2918 | ordered_extent->len); | |
a698d075 | 2919 | if (trans) |
3a45bb20 | 2920 | btrfs_end_transaction(trans); |
0cb59c99 | 2921 | |
77cef2ec JB |
2922 | if (ret || truncated) { |
2923 | u64 start, end; | |
2924 | ||
2925 | if (truncated) | |
2926 | start = ordered_extent->file_offset + logical_len; | |
2927 | else | |
2928 | start = ordered_extent->file_offset; | |
2929 | end = ordered_extent->file_offset + ordered_extent->len - 1; | |
2930 | clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS); | |
2931 | ||
2932 | /* Drop the cache for the part of the extent we didn't write. */ | |
dcdbc059 | 2933 | btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0); |
5fd02043 | 2934 | |
0bec9ef5 JB |
2935 | /* |
2936 | * If the ordered extent had an IOERR or something else went | |
2937 | * wrong we need to return the space for this ordered extent | |
77cef2ec JB |
2938 | * back to the allocator. We only free the extent in the |
2939 | * truncated case if we didn't write out the extent at all. | |
0bec9ef5 | 2940 | */ |
77cef2ec JB |
2941 | if ((ret || !logical_len) && |
2942 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && | |
0bec9ef5 | 2943 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) |
2ff7e61e JM |
2944 | btrfs_free_reserved_extent(fs_info, |
2945 | ordered_extent->start, | |
e570fd27 | 2946 | ordered_extent->disk_len, 1); |
0bec9ef5 JB |
2947 | } |
2948 | ||
2949 | ||
5fd02043 | 2950 | /* |
8bad3c02 LB |
2951 | * This needs to be done to make sure anybody waiting knows we are done |
2952 | * updating everything for this ordered extent. | |
5fd02043 JB |
2953 | */ |
2954 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
2955 | ||
38c227d8 | 2956 | /* for snapshot-aware defrag */ |
6f519564 LB |
2957 | if (new) { |
2958 | if (ret) { | |
2959 | free_sa_defrag_extent(new); | |
0b246afa | 2960 | atomic_dec(&fs_info->defrag_running); |
6f519564 LB |
2961 | } else { |
2962 | relink_file_extents(new); | |
2963 | } | |
2964 | } | |
38c227d8 | 2965 | |
e6dcd2dc CM |
2966 | /* once for us */ |
2967 | btrfs_put_ordered_extent(ordered_extent); | |
2968 | /* once for the tree */ | |
2969 | btrfs_put_ordered_extent(ordered_extent); | |
2970 | ||
5fd02043 JB |
2971 | return ret; |
2972 | } | |
2973 | ||
2974 | static void finish_ordered_fn(struct btrfs_work *work) | |
2975 | { | |
2976 | struct btrfs_ordered_extent *ordered_extent; | |
2977 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
2978 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
2979 | } |
2980 | ||
c3988d63 | 2981 | static void btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
2982 | struct extent_state *state, int uptodate) |
2983 | { | |
5fd02043 | 2984 | struct inode *inode = page->mapping->host; |
0b246afa | 2985 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5fd02043 | 2986 | struct btrfs_ordered_extent *ordered_extent = NULL; |
9e0af237 LB |
2987 | struct btrfs_workqueue *wq; |
2988 | btrfs_work_func_t func; | |
5fd02043 | 2989 | |
1abe9b8a | 2990 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
2991 | ||
8b62b72b | 2992 | ClearPagePrivate2(page); |
5fd02043 JB |
2993 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
2994 | end - start + 1, uptodate)) | |
c3988d63 | 2995 | return; |
5fd02043 | 2996 | |
70ddc553 | 2997 | if (btrfs_is_free_space_inode(BTRFS_I(inode))) { |
0b246afa | 2998 | wq = fs_info->endio_freespace_worker; |
9e0af237 LB |
2999 | func = btrfs_freespace_write_helper; |
3000 | } else { | |
0b246afa | 3001 | wq = fs_info->endio_write_workers; |
9e0af237 LB |
3002 | func = btrfs_endio_write_helper; |
3003 | } | |
5fd02043 | 3004 | |
9e0af237 LB |
3005 | btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL, |
3006 | NULL); | |
3007 | btrfs_queue_work(wq, &ordered_extent->work); | |
211f90e6 CM |
3008 | } |
3009 | ||
dc380aea MX |
3010 | static int __readpage_endio_check(struct inode *inode, |
3011 | struct btrfs_io_bio *io_bio, | |
3012 | int icsum, struct page *page, | |
3013 | int pgoff, u64 start, size_t len) | |
3014 | { | |
3015 | char *kaddr; | |
3016 | u32 csum_expected; | |
3017 | u32 csum = ~(u32)0; | |
dc380aea MX |
3018 | |
3019 | csum_expected = *(((u32 *)io_bio->csum) + icsum); | |
3020 | ||
3021 | kaddr = kmap_atomic(page); | |
3022 | csum = btrfs_csum_data(kaddr + pgoff, csum, len); | |
0b5e3daf | 3023 | btrfs_csum_final(csum, (u8 *)&csum); |
dc380aea MX |
3024 | if (csum != csum_expected) |
3025 | goto zeroit; | |
3026 | ||
3027 | kunmap_atomic(kaddr); | |
3028 | return 0; | |
3029 | zeroit: | |
0970a22e | 3030 | btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected, |
6f6b643e | 3031 | io_bio->mirror_num); |
dc380aea MX |
3032 | memset(kaddr + pgoff, 1, len); |
3033 | flush_dcache_page(page); | |
3034 | kunmap_atomic(kaddr); | |
3035 | if (csum_expected == 0) | |
3036 | return 0; | |
3037 | return -EIO; | |
3038 | } | |
3039 | ||
d352ac68 CM |
3040 | /* |
3041 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
3042 | * if there's a match, we allow the bio to finish. If not, the code in |
3043 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 3044 | */ |
facc8a22 MX |
3045 | static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio, |
3046 | u64 phy_offset, struct page *page, | |
3047 | u64 start, u64 end, int mirror) | |
07157aac | 3048 | { |
4eee4fa4 | 3049 | size_t offset = start - page_offset(page); |
07157aac | 3050 | struct inode *inode = page->mapping->host; |
d1310b2e | 3051 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
ff79f819 | 3052 | struct btrfs_root *root = BTRFS_I(inode)->root; |
d1310b2e | 3053 | |
d20f7043 CM |
3054 | if (PageChecked(page)) { |
3055 | ClearPageChecked(page); | |
dc380aea | 3056 | return 0; |
d20f7043 | 3057 | } |
6cbff00f CH |
3058 | |
3059 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
dc380aea | 3060 | return 0; |
17d217fe YZ |
3061 | |
3062 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 3063 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
91166212 | 3064 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM); |
b6cda9bc | 3065 | return 0; |
17d217fe | 3066 | } |
d20f7043 | 3067 | |
facc8a22 | 3068 | phy_offset >>= inode->i_sb->s_blocksize_bits; |
dc380aea MX |
3069 | return __readpage_endio_check(inode, io_bio, phy_offset, page, offset, |
3070 | start, (size_t)(end - start + 1)); | |
07157aac | 3071 | } |
b888db2b | 3072 | |
24bbcf04 YZ |
3073 | void btrfs_add_delayed_iput(struct inode *inode) |
3074 | { | |
0b246afa | 3075 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
8089fe62 | 3076 | struct btrfs_inode *binode = BTRFS_I(inode); |
24bbcf04 YZ |
3077 | |
3078 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
3079 | return; | |
3080 | ||
24bbcf04 | 3081 | spin_lock(&fs_info->delayed_iput_lock); |
8089fe62 DS |
3082 | if (binode->delayed_iput_count == 0) { |
3083 | ASSERT(list_empty(&binode->delayed_iput)); | |
3084 | list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs); | |
3085 | } else { | |
3086 | binode->delayed_iput_count++; | |
3087 | } | |
24bbcf04 YZ |
3088 | spin_unlock(&fs_info->delayed_iput_lock); |
3089 | } | |
3090 | ||
2ff7e61e | 3091 | void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info) |
24bbcf04 | 3092 | { |
24bbcf04 | 3093 | |
24bbcf04 | 3094 | spin_lock(&fs_info->delayed_iput_lock); |
8089fe62 DS |
3095 | while (!list_empty(&fs_info->delayed_iputs)) { |
3096 | struct btrfs_inode *inode; | |
3097 | ||
3098 | inode = list_first_entry(&fs_info->delayed_iputs, | |
3099 | struct btrfs_inode, delayed_iput); | |
3100 | if (inode->delayed_iput_count) { | |
3101 | inode->delayed_iput_count--; | |
3102 | list_move_tail(&inode->delayed_iput, | |
3103 | &fs_info->delayed_iputs); | |
3104 | } else { | |
3105 | list_del_init(&inode->delayed_iput); | |
3106 | } | |
3107 | spin_unlock(&fs_info->delayed_iput_lock); | |
3108 | iput(&inode->vfs_inode); | |
3109 | spin_lock(&fs_info->delayed_iput_lock); | |
24bbcf04 | 3110 | } |
8089fe62 | 3111 | spin_unlock(&fs_info->delayed_iput_lock); |
24bbcf04 YZ |
3112 | } |
3113 | ||
d68fc57b | 3114 | /* |
42b2aa86 | 3115 | * This is called in transaction commit time. If there are no orphan |
d68fc57b YZ |
3116 | * files in the subvolume, it removes orphan item and frees block_rsv |
3117 | * structure. | |
3118 | */ | |
3119 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
3120 | struct btrfs_root *root) | |
3121 | { | |
0b246afa | 3122 | struct btrfs_fs_info *fs_info = root->fs_info; |
90290e19 | 3123 | struct btrfs_block_rsv *block_rsv; |
d68fc57b YZ |
3124 | int ret; |
3125 | ||
8a35d95f | 3126 | if (atomic_read(&root->orphan_inodes) || |
d68fc57b YZ |
3127 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) |
3128 | return; | |
3129 | ||
90290e19 | 3130 | spin_lock(&root->orphan_lock); |
8a35d95f | 3131 | if (atomic_read(&root->orphan_inodes)) { |
90290e19 JB |
3132 | spin_unlock(&root->orphan_lock); |
3133 | return; | |
3134 | } | |
3135 | ||
3136 | if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) { | |
3137 | spin_unlock(&root->orphan_lock); | |
3138 | return; | |
3139 | } | |
3140 | ||
3141 | block_rsv = root->orphan_block_rsv; | |
3142 | root->orphan_block_rsv = NULL; | |
3143 | spin_unlock(&root->orphan_lock); | |
3144 | ||
27cdeb70 | 3145 | if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) && |
d68fc57b | 3146 | btrfs_root_refs(&root->root_item) > 0) { |
0b246afa | 3147 | ret = btrfs_del_orphan_item(trans, fs_info->tree_root, |
d68fc57b | 3148 | root->root_key.objectid); |
4ef31a45 | 3149 | if (ret) |
66642832 | 3150 | btrfs_abort_transaction(trans, ret); |
4ef31a45 | 3151 | else |
27cdeb70 MX |
3152 | clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, |
3153 | &root->state); | |
d68fc57b YZ |
3154 | } |
3155 | ||
90290e19 JB |
3156 | if (block_rsv) { |
3157 | WARN_ON(block_rsv->size > 0); | |
2ff7e61e | 3158 | btrfs_free_block_rsv(fs_info, block_rsv); |
d68fc57b YZ |
3159 | } |
3160 | } | |
3161 | ||
7b128766 JB |
3162 | /* |
3163 | * This creates an orphan entry for the given inode in case something goes | |
3164 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
3165 | * |
3166 | * NOTE: caller of this function should reserve 5 units of metadata for | |
3167 | * this function. | |
7b128766 | 3168 | */ |
73f2e545 NB |
3169 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, |
3170 | struct btrfs_inode *inode) | |
7b128766 | 3171 | { |
73f2e545 NB |
3172 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
3173 | struct btrfs_root *root = inode->root; | |
d68fc57b YZ |
3174 | struct btrfs_block_rsv *block_rsv = NULL; |
3175 | int reserve = 0; | |
3176 | int insert = 0; | |
3177 | int ret; | |
7b128766 | 3178 | |
d68fc57b | 3179 | if (!root->orphan_block_rsv) { |
2ff7e61e JM |
3180 | block_rsv = btrfs_alloc_block_rsv(fs_info, |
3181 | BTRFS_BLOCK_RSV_TEMP); | |
b532402e TI |
3182 | if (!block_rsv) |
3183 | return -ENOMEM; | |
d68fc57b | 3184 | } |
7b128766 | 3185 | |
d68fc57b YZ |
3186 | spin_lock(&root->orphan_lock); |
3187 | if (!root->orphan_block_rsv) { | |
3188 | root->orphan_block_rsv = block_rsv; | |
3189 | } else if (block_rsv) { | |
2ff7e61e | 3190 | btrfs_free_block_rsv(fs_info, block_rsv); |
d68fc57b | 3191 | block_rsv = NULL; |
7b128766 | 3192 | } |
7b128766 | 3193 | |
8a35d95f | 3194 | if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
73f2e545 | 3195 | &inode->runtime_flags)) { |
d68fc57b YZ |
3196 | #if 0 |
3197 | /* | |
3198 | * For proper ENOSPC handling, we should do orphan | |
3199 | * cleanup when mounting. But this introduces backward | |
3200 | * compatibility issue. | |
3201 | */ | |
3202 | if (!xchg(&root->orphan_item_inserted, 1)) | |
3203 | insert = 2; | |
3204 | else | |
3205 | insert = 1; | |
3206 | #endif | |
3207 | insert = 1; | |
321f0e70 | 3208 | atomic_inc(&root->orphan_inodes); |
7b128766 JB |
3209 | } |
3210 | ||
72ac3c0d | 3211 | if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
73f2e545 | 3212 | &inode->runtime_flags)) |
d68fc57b | 3213 | reserve = 1; |
d68fc57b | 3214 | spin_unlock(&root->orphan_lock); |
7b128766 | 3215 | |
d68fc57b YZ |
3216 | /* grab metadata reservation from transaction handle */ |
3217 | if (reserve) { | |
3218 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
3b6571c1 JB |
3219 | ASSERT(!ret); |
3220 | if (ret) { | |
3221 | atomic_dec(&root->orphan_inodes); | |
3222 | clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, | |
73f2e545 | 3223 | &inode->runtime_flags); |
3b6571c1 JB |
3224 | if (insert) |
3225 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, | |
73f2e545 | 3226 | &inode->runtime_flags); |
3b6571c1 JB |
3227 | return ret; |
3228 | } | |
d68fc57b | 3229 | } |
7b128766 | 3230 | |
d68fc57b YZ |
3231 | /* insert an orphan item to track this unlinked/truncated file */ |
3232 | if (insert >= 1) { | |
73f2e545 | 3233 | ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode)); |
4ef31a45 | 3234 | if (ret) { |
703c88e0 | 3235 | atomic_dec(&root->orphan_inodes); |
4ef31a45 JB |
3236 | if (reserve) { |
3237 | clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, | |
73f2e545 | 3238 | &inode->runtime_flags); |
4ef31a45 JB |
3239 | btrfs_orphan_release_metadata(inode); |
3240 | } | |
3241 | if (ret != -EEXIST) { | |
e8e7cff6 | 3242 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
73f2e545 | 3243 | &inode->runtime_flags); |
66642832 | 3244 | btrfs_abort_transaction(trans, ret); |
4ef31a45 JB |
3245 | return ret; |
3246 | } | |
79787eaa JM |
3247 | } |
3248 | ret = 0; | |
d68fc57b YZ |
3249 | } |
3250 | ||
3251 | /* insert an orphan item to track subvolume contains orphan files */ | |
3252 | if (insert >= 2) { | |
0b246afa | 3253 | ret = btrfs_insert_orphan_item(trans, fs_info->tree_root, |
d68fc57b | 3254 | root->root_key.objectid); |
79787eaa | 3255 | if (ret && ret != -EEXIST) { |
66642832 | 3256 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3257 | return ret; |
3258 | } | |
d68fc57b YZ |
3259 | } |
3260 | return 0; | |
7b128766 JB |
3261 | } |
3262 | ||
3263 | /* | |
3264 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
3265 | * item for this particular inode. | |
3266 | */ | |
48a3b636 | 3267 | static int btrfs_orphan_del(struct btrfs_trans_handle *trans, |
3d6ae7bb | 3268 | struct btrfs_inode *inode) |
7b128766 | 3269 | { |
3d6ae7bb | 3270 | struct btrfs_root *root = inode->root; |
d68fc57b YZ |
3271 | int delete_item = 0; |
3272 | int release_rsv = 0; | |
7b128766 JB |
3273 | int ret = 0; |
3274 | ||
d68fc57b | 3275 | spin_lock(&root->orphan_lock); |
8a35d95f | 3276 | if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3d6ae7bb | 3277 | &inode->runtime_flags)) |
d68fc57b | 3278 | delete_item = 1; |
7b128766 | 3279 | |
72ac3c0d | 3280 | if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
3d6ae7bb | 3281 | &inode->runtime_flags)) |
d68fc57b | 3282 | release_rsv = 1; |
d68fc57b | 3283 | spin_unlock(&root->orphan_lock); |
7b128766 | 3284 | |
703c88e0 | 3285 | if (delete_item) { |
8a35d95f | 3286 | atomic_dec(&root->orphan_inodes); |
703c88e0 FDBM |
3287 | if (trans) |
3288 | ret = btrfs_del_orphan_item(trans, root, | |
3d6ae7bb | 3289 | btrfs_ino(inode)); |
8a35d95f | 3290 | } |
7b128766 | 3291 | |
703c88e0 FDBM |
3292 | if (release_rsv) |
3293 | btrfs_orphan_release_metadata(inode); | |
3294 | ||
4ef31a45 | 3295 | return ret; |
7b128766 JB |
3296 | } |
3297 | ||
3298 | /* | |
3299 | * this cleans up any orphans that may be left on the list from the last use | |
3300 | * of this root. | |
3301 | */ | |
66b4ffd1 | 3302 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 | 3303 | { |
0b246afa | 3304 | struct btrfs_fs_info *fs_info = root->fs_info; |
7b128766 JB |
3305 | struct btrfs_path *path; |
3306 | struct extent_buffer *leaf; | |
7b128766 JB |
3307 | struct btrfs_key key, found_key; |
3308 | struct btrfs_trans_handle *trans; | |
3309 | struct inode *inode; | |
8f6d7f4f | 3310 | u64 last_objectid = 0; |
7b128766 JB |
3311 | int ret = 0, nr_unlink = 0, nr_truncate = 0; |
3312 | ||
d68fc57b | 3313 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 3314 | return 0; |
c71bf099 YZ |
3315 | |
3316 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
3317 | if (!path) { |
3318 | ret = -ENOMEM; | |
3319 | goto out; | |
3320 | } | |
e4058b54 | 3321 | path->reada = READA_BACK; |
7b128766 JB |
3322 | |
3323 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
962a298f | 3324 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
7b128766 JB |
3325 | key.offset = (u64)-1; |
3326 | ||
7b128766 JB |
3327 | while (1) { |
3328 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
3329 | if (ret < 0) |
3330 | goto out; | |
7b128766 JB |
3331 | |
3332 | /* | |
3333 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 3334 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
3335 | * find the key and see if we have stuff that matches |
3336 | */ | |
3337 | if (ret > 0) { | |
66b4ffd1 | 3338 | ret = 0; |
7b128766 JB |
3339 | if (path->slots[0] == 0) |
3340 | break; | |
3341 | path->slots[0]--; | |
3342 | } | |
3343 | ||
3344 | /* pull out the item */ | |
3345 | leaf = path->nodes[0]; | |
7b128766 JB |
3346 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3347 | ||
3348 | /* make sure the item matches what we want */ | |
3349 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
3350 | break; | |
962a298f | 3351 | if (found_key.type != BTRFS_ORPHAN_ITEM_KEY) |
7b128766 JB |
3352 | break; |
3353 | ||
3354 | /* release the path since we're done with it */ | |
b3b4aa74 | 3355 | btrfs_release_path(path); |
7b128766 JB |
3356 | |
3357 | /* | |
3358 | * this is where we are basically btrfs_lookup, without the | |
3359 | * crossing root thing. we store the inode number in the | |
3360 | * offset of the orphan item. | |
3361 | */ | |
8f6d7f4f JB |
3362 | |
3363 | if (found_key.offset == last_objectid) { | |
0b246afa JM |
3364 | btrfs_err(fs_info, |
3365 | "Error removing orphan entry, stopping orphan cleanup"); | |
8f6d7f4f JB |
3366 | ret = -EINVAL; |
3367 | goto out; | |
3368 | } | |
3369 | ||
3370 | last_objectid = found_key.offset; | |
3371 | ||
5d4f98a2 YZ |
3372 | found_key.objectid = found_key.offset; |
3373 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
3374 | found_key.offset = 0; | |
0b246afa | 3375 | inode = btrfs_iget(fs_info->sb, &found_key, root, NULL); |
8c6ffba0 | 3376 | ret = PTR_ERR_OR_ZERO(inode); |
67710892 | 3377 | if (ret && ret != -ENOENT) |
66b4ffd1 | 3378 | goto out; |
7b128766 | 3379 | |
0b246afa | 3380 | if (ret == -ENOENT && root == fs_info->tree_root) { |
f8e9e0b0 AJ |
3381 | struct btrfs_root *dead_root; |
3382 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3383 | int is_dead_root = 0; | |
3384 | ||
3385 | /* | |
3386 | * this is an orphan in the tree root. Currently these | |
3387 | * could come from 2 sources: | |
3388 | * a) a snapshot deletion in progress | |
3389 | * b) a free space cache inode | |
3390 | * We need to distinguish those two, as the snapshot | |
3391 | * orphan must not get deleted. | |
3392 | * find_dead_roots already ran before us, so if this | |
3393 | * is a snapshot deletion, we should find the root | |
3394 | * in the dead_roots list | |
3395 | */ | |
3396 | spin_lock(&fs_info->trans_lock); | |
3397 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
3398 | root_list) { | |
3399 | if (dead_root->root_key.objectid == | |
3400 | found_key.objectid) { | |
3401 | is_dead_root = 1; | |
3402 | break; | |
3403 | } | |
3404 | } | |
3405 | spin_unlock(&fs_info->trans_lock); | |
3406 | if (is_dead_root) { | |
3407 | /* prevent this orphan from being found again */ | |
3408 | key.offset = found_key.objectid - 1; | |
3409 | continue; | |
3410 | } | |
3411 | } | |
7b128766 | 3412 | /* |
a8c9e576 JB |
3413 | * Inode is already gone but the orphan item is still there, |
3414 | * kill the orphan item. | |
7b128766 | 3415 | */ |
67710892 | 3416 | if (ret == -ENOENT) { |
a8c9e576 | 3417 | trans = btrfs_start_transaction(root, 1); |
66b4ffd1 JB |
3418 | if (IS_ERR(trans)) { |
3419 | ret = PTR_ERR(trans); | |
3420 | goto out; | |
3421 | } | |
0b246afa JM |
3422 | btrfs_debug(fs_info, "auto deleting %Lu", |
3423 | found_key.objectid); | |
a8c9e576 JB |
3424 | ret = btrfs_del_orphan_item(trans, root, |
3425 | found_key.objectid); | |
3a45bb20 | 3426 | btrfs_end_transaction(trans); |
4ef31a45 JB |
3427 | if (ret) |
3428 | goto out; | |
7b128766 JB |
3429 | continue; |
3430 | } | |
3431 | ||
a8c9e576 JB |
3432 | /* |
3433 | * add this inode to the orphan list so btrfs_orphan_del does | |
3434 | * the proper thing when we hit it | |
3435 | */ | |
8a35d95f JB |
3436 | set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3437 | &BTRFS_I(inode)->runtime_flags); | |
925396ec | 3438 | atomic_inc(&root->orphan_inodes); |
a8c9e576 | 3439 | |
7b128766 JB |
3440 | /* if we have links, this was a truncate, lets do that */ |
3441 | if (inode->i_nlink) { | |
fae7f21c | 3442 | if (WARN_ON(!S_ISREG(inode->i_mode))) { |
a41ad394 JB |
3443 | iput(inode); |
3444 | continue; | |
3445 | } | |
7b128766 | 3446 | nr_truncate++; |
f3fe820c JB |
3447 | |
3448 | /* 1 for the orphan item deletion. */ | |
3449 | trans = btrfs_start_transaction(root, 1); | |
3450 | if (IS_ERR(trans)) { | |
c69b26b0 | 3451 | iput(inode); |
f3fe820c JB |
3452 | ret = PTR_ERR(trans); |
3453 | goto out; | |
3454 | } | |
73f2e545 | 3455 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
3a45bb20 | 3456 | btrfs_end_transaction(trans); |
c69b26b0 JB |
3457 | if (ret) { |
3458 | iput(inode); | |
f3fe820c | 3459 | goto out; |
c69b26b0 | 3460 | } |
f3fe820c | 3461 | |
66b4ffd1 | 3462 | ret = btrfs_truncate(inode); |
4a7d0f68 | 3463 | if (ret) |
3d6ae7bb | 3464 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
7b128766 JB |
3465 | } else { |
3466 | nr_unlink++; | |
3467 | } | |
3468 | ||
3469 | /* this will do delete_inode and everything for us */ | |
3470 | iput(inode); | |
66b4ffd1 JB |
3471 | if (ret) |
3472 | goto out; | |
7b128766 | 3473 | } |
3254c876 MX |
3474 | /* release the path since we're done with it */ |
3475 | btrfs_release_path(path); | |
3476 | ||
d68fc57b YZ |
3477 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
3478 | ||
3479 | if (root->orphan_block_rsv) | |
2ff7e61e | 3480 | btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, |
d68fc57b YZ |
3481 | (u64)-1); |
3482 | ||
27cdeb70 MX |
3483 | if (root->orphan_block_rsv || |
3484 | test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) { | |
7a7eaa40 | 3485 | trans = btrfs_join_transaction(root); |
66b4ffd1 | 3486 | if (!IS_ERR(trans)) |
3a45bb20 | 3487 | btrfs_end_transaction(trans); |
d68fc57b | 3488 | } |
7b128766 JB |
3489 | |
3490 | if (nr_unlink) | |
0b246afa | 3491 | btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink); |
7b128766 | 3492 | if (nr_truncate) |
0b246afa | 3493 | btrfs_debug(fs_info, "truncated %d orphans", nr_truncate); |
66b4ffd1 JB |
3494 | |
3495 | out: | |
3496 | if (ret) | |
0b246afa | 3497 | btrfs_err(fs_info, "could not do orphan cleanup %d", ret); |
66b4ffd1 JB |
3498 | btrfs_free_path(path); |
3499 | return ret; | |
7b128766 JB |
3500 | } |
3501 | ||
46a53cca CM |
3502 | /* |
3503 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
3504 | * don't find any xattrs, we know there can't be any acls. | |
3505 | * | |
3506 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
3507 | */ | |
3508 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
63541927 FDBM |
3509 | int slot, u64 objectid, |
3510 | int *first_xattr_slot) | |
46a53cca CM |
3511 | { |
3512 | u32 nritems = btrfs_header_nritems(leaf); | |
3513 | struct btrfs_key found_key; | |
f23b5a59 JB |
3514 | static u64 xattr_access = 0; |
3515 | static u64 xattr_default = 0; | |
46a53cca CM |
3516 | int scanned = 0; |
3517 | ||
f23b5a59 | 3518 | if (!xattr_access) { |
97d79299 AG |
3519 | xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS, |
3520 | strlen(XATTR_NAME_POSIX_ACL_ACCESS)); | |
3521 | xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT, | |
3522 | strlen(XATTR_NAME_POSIX_ACL_DEFAULT)); | |
f23b5a59 JB |
3523 | } |
3524 | ||
46a53cca | 3525 | slot++; |
63541927 | 3526 | *first_xattr_slot = -1; |
46a53cca CM |
3527 | while (slot < nritems) { |
3528 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3529 | ||
3530 | /* we found a different objectid, there must not be acls */ | |
3531 | if (found_key.objectid != objectid) | |
3532 | return 0; | |
3533 | ||
3534 | /* we found an xattr, assume we've got an acl */ | |
f23b5a59 | 3535 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) { |
63541927 FDBM |
3536 | if (*first_xattr_slot == -1) |
3537 | *first_xattr_slot = slot; | |
f23b5a59 JB |
3538 | if (found_key.offset == xattr_access || |
3539 | found_key.offset == xattr_default) | |
3540 | return 1; | |
3541 | } | |
46a53cca CM |
3542 | |
3543 | /* | |
3544 | * we found a key greater than an xattr key, there can't | |
3545 | * be any acls later on | |
3546 | */ | |
3547 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
3548 | return 0; | |
3549 | ||
3550 | slot++; | |
3551 | scanned++; | |
3552 | ||
3553 | /* | |
3554 | * it goes inode, inode backrefs, xattrs, extents, | |
3555 | * so if there are a ton of hard links to an inode there can | |
3556 | * be a lot of backrefs. Don't waste time searching too hard, | |
3557 | * this is just an optimization | |
3558 | */ | |
3559 | if (scanned >= 8) | |
3560 | break; | |
3561 | } | |
3562 | /* we hit the end of the leaf before we found an xattr or | |
3563 | * something larger than an xattr. We have to assume the inode | |
3564 | * has acls | |
3565 | */ | |
63541927 FDBM |
3566 | if (*first_xattr_slot == -1) |
3567 | *first_xattr_slot = slot; | |
46a53cca CM |
3568 | return 1; |
3569 | } | |
3570 | ||
d352ac68 CM |
3571 | /* |
3572 | * read an inode from the btree into the in-memory inode | |
3573 | */ | |
67710892 | 3574 | static int btrfs_read_locked_inode(struct inode *inode) |
39279cc3 | 3575 | { |
0b246afa | 3576 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 3577 | struct btrfs_path *path; |
5f39d397 | 3578 | struct extent_buffer *leaf; |
39279cc3 CM |
3579 | struct btrfs_inode_item *inode_item; |
3580 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3581 | struct btrfs_key location; | |
67de1176 | 3582 | unsigned long ptr; |
46a53cca | 3583 | int maybe_acls; |
618e21d5 | 3584 | u32 rdev; |
39279cc3 | 3585 | int ret; |
2f7e33d4 | 3586 | bool filled = false; |
63541927 | 3587 | int first_xattr_slot; |
2f7e33d4 MX |
3588 | |
3589 | ret = btrfs_fill_inode(inode, &rdev); | |
3590 | if (!ret) | |
3591 | filled = true; | |
39279cc3 CM |
3592 | |
3593 | path = btrfs_alloc_path(); | |
67710892 FM |
3594 | if (!path) { |
3595 | ret = -ENOMEM; | |
1748f843 | 3596 | goto make_bad; |
67710892 | 3597 | } |
1748f843 | 3598 | |
39279cc3 | 3599 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 3600 | |
39279cc3 | 3601 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
67710892 FM |
3602 | if (ret) { |
3603 | if (ret > 0) | |
3604 | ret = -ENOENT; | |
39279cc3 | 3605 | goto make_bad; |
67710892 | 3606 | } |
39279cc3 | 3607 | |
5f39d397 | 3608 | leaf = path->nodes[0]; |
2f7e33d4 MX |
3609 | |
3610 | if (filled) | |
67de1176 | 3611 | goto cache_index; |
2f7e33d4 | 3612 | |
5f39d397 CM |
3613 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
3614 | struct btrfs_inode_item); | |
5f39d397 | 3615 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 3616 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
3617 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
3618 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
6ef06d27 | 3619 | btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item)); |
5f39d397 | 3620 | |
a937b979 DS |
3621 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime); |
3622 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime); | |
5f39d397 | 3623 | |
a937b979 DS |
3624 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime); |
3625 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime); | |
5f39d397 | 3626 | |
a937b979 DS |
3627 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime); |
3628 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime); | |
5f39d397 | 3629 | |
9cc97d64 | 3630 | BTRFS_I(inode)->i_otime.tv_sec = |
3631 | btrfs_timespec_sec(leaf, &inode_item->otime); | |
3632 | BTRFS_I(inode)->i_otime.tv_nsec = | |
3633 | btrfs_timespec_nsec(leaf, &inode_item->otime); | |
5f39d397 | 3634 | |
a76a3cd4 | 3635 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 3636 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
3637 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
3638 | ||
6e17d30b YD |
3639 | inode->i_version = btrfs_inode_sequence(leaf, inode_item); |
3640 | inode->i_generation = BTRFS_I(inode)->generation; | |
3641 | inode->i_rdev = 0; | |
3642 | rdev = btrfs_inode_rdev(leaf, inode_item); | |
3643 | ||
3644 | BTRFS_I(inode)->index_cnt = (u64)-1; | |
3645 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); | |
3646 | ||
3647 | cache_index: | |
5dc562c5 JB |
3648 | /* |
3649 | * If we were modified in the current generation and evicted from memory | |
3650 | * and then re-read we need to do a full sync since we don't have any | |
3651 | * idea about which extents were modified before we were evicted from | |
3652 | * cache. | |
6e17d30b YD |
3653 | * |
3654 | * This is required for both inode re-read from disk and delayed inode | |
3655 | * in delayed_nodes_tree. | |
5dc562c5 | 3656 | */ |
0b246afa | 3657 | if (BTRFS_I(inode)->last_trans == fs_info->generation) |
5dc562c5 JB |
3658 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
3659 | &BTRFS_I(inode)->runtime_flags); | |
3660 | ||
bde6c242 FM |
3661 | /* |
3662 | * We don't persist the id of the transaction where an unlink operation | |
3663 | * against the inode was last made. So here we assume the inode might | |
3664 | * have been evicted, and therefore the exact value of last_unlink_trans | |
3665 | * lost, and set it to last_trans to avoid metadata inconsistencies | |
3666 | * between the inode and its parent if the inode is fsync'ed and the log | |
3667 | * replayed. For example, in the scenario: | |
3668 | * | |
3669 | * touch mydir/foo | |
3670 | * ln mydir/foo mydir/bar | |
3671 | * sync | |
3672 | * unlink mydir/bar | |
3673 | * echo 2 > /proc/sys/vm/drop_caches # evicts inode | |
3674 | * xfs_io -c fsync mydir/foo | |
3675 | * <power failure> | |
3676 | * mount fs, triggers fsync log replay | |
3677 | * | |
3678 | * We must make sure that when we fsync our inode foo we also log its | |
3679 | * parent inode, otherwise after log replay the parent still has the | |
3680 | * dentry with the "bar" name but our inode foo has a link count of 1 | |
3681 | * and doesn't have an inode ref with the name "bar" anymore. | |
3682 | * | |
3683 | * Setting last_unlink_trans to last_trans is a pessimistic approach, | |
01327610 | 3684 | * but it guarantees correctness at the expense of occasional full |
bde6c242 FM |
3685 | * transaction commits on fsync if our inode is a directory, or if our |
3686 | * inode is not a directory, logging its parent unnecessarily. | |
3687 | */ | |
3688 | BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans; | |
3689 | ||
67de1176 MX |
3690 | path->slots[0]++; |
3691 | if (inode->i_nlink != 1 || | |
3692 | path->slots[0] >= btrfs_header_nritems(leaf)) | |
3693 | goto cache_acl; | |
3694 | ||
3695 | btrfs_item_key_to_cpu(leaf, &location, path->slots[0]); | |
4a0cc7ca | 3696 | if (location.objectid != btrfs_ino(BTRFS_I(inode))) |
67de1176 MX |
3697 | goto cache_acl; |
3698 | ||
3699 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3700 | if (location.type == BTRFS_INODE_REF_KEY) { | |
3701 | struct btrfs_inode_ref *ref; | |
3702 | ||
3703 | ref = (struct btrfs_inode_ref *)ptr; | |
3704 | BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref); | |
3705 | } else if (location.type == BTRFS_INODE_EXTREF_KEY) { | |
3706 | struct btrfs_inode_extref *extref; | |
3707 | ||
3708 | extref = (struct btrfs_inode_extref *)ptr; | |
3709 | BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf, | |
3710 | extref); | |
3711 | } | |
2f7e33d4 | 3712 | cache_acl: |
46a53cca CM |
3713 | /* |
3714 | * try to precache a NULL acl entry for files that don't have | |
3715 | * any xattrs or acls | |
3716 | */ | |
33345d01 | 3717 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
f85b7379 | 3718 | btrfs_ino(BTRFS_I(inode)), &first_xattr_slot); |
63541927 FDBM |
3719 | if (first_xattr_slot != -1) { |
3720 | path->slots[0] = first_xattr_slot; | |
3721 | ret = btrfs_load_inode_props(inode, path); | |
3722 | if (ret) | |
0b246afa | 3723 | btrfs_err(fs_info, |
351fd353 | 3724 | "error loading props for ino %llu (root %llu): %d", |
4a0cc7ca | 3725 | btrfs_ino(BTRFS_I(inode)), |
63541927 FDBM |
3726 | root->root_key.objectid, ret); |
3727 | } | |
3728 | btrfs_free_path(path); | |
3729 | ||
72c04902 AV |
3730 | if (!maybe_acls) |
3731 | cache_no_acl(inode); | |
46a53cca | 3732 | |
39279cc3 | 3733 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
3734 | case S_IFREG: |
3735 | inode->i_mapping->a_ops = &btrfs_aops; | |
d1310b2e | 3736 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
3737 | inode->i_fop = &btrfs_file_operations; |
3738 | inode->i_op = &btrfs_file_inode_operations; | |
3739 | break; | |
3740 | case S_IFDIR: | |
3741 | inode->i_fop = &btrfs_dir_file_operations; | |
67ade058 | 3742 | inode->i_op = &btrfs_dir_inode_operations; |
39279cc3 CM |
3743 | break; |
3744 | case S_IFLNK: | |
3745 | inode->i_op = &btrfs_symlink_inode_operations; | |
21fc61c7 | 3746 | inode_nohighmem(inode); |
39279cc3 CM |
3747 | inode->i_mapping->a_ops = &btrfs_symlink_aops; |
3748 | break; | |
618e21d5 | 3749 | default: |
0279b4cd | 3750 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
3751 | init_special_inode(inode, inode->i_mode, rdev); |
3752 | break; | |
39279cc3 | 3753 | } |
6cbff00f CH |
3754 | |
3755 | btrfs_update_iflags(inode); | |
67710892 | 3756 | return 0; |
39279cc3 CM |
3757 | |
3758 | make_bad: | |
39279cc3 | 3759 | btrfs_free_path(path); |
39279cc3 | 3760 | make_bad_inode(inode); |
67710892 | 3761 | return ret; |
39279cc3 CM |
3762 | } |
3763 | ||
d352ac68 CM |
3764 | /* |
3765 | * given a leaf and an inode, copy the inode fields into the leaf | |
3766 | */ | |
e02119d5 CM |
3767 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3768 | struct extent_buffer *leaf, | |
5f39d397 | 3769 | struct btrfs_inode_item *item, |
39279cc3 CM |
3770 | struct inode *inode) |
3771 | { | |
51fab693 LB |
3772 | struct btrfs_map_token token; |
3773 | ||
3774 | btrfs_init_map_token(&token); | |
5f39d397 | 3775 | |
51fab693 LB |
3776 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); |
3777 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3778 | btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size, | |
3779 | &token); | |
3780 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3781 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
5f39d397 | 3782 | |
a937b979 | 3783 | btrfs_set_token_timespec_sec(leaf, &item->atime, |
51fab693 | 3784 | inode->i_atime.tv_sec, &token); |
a937b979 | 3785 | btrfs_set_token_timespec_nsec(leaf, &item->atime, |
51fab693 | 3786 | inode->i_atime.tv_nsec, &token); |
5f39d397 | 3787 | |
a937b979 | 3788 | btrfs_set_token_timespec_sec(leaf, &item->mtime, |
51fab693 | 3789 | inode->i_mtime.tv_sec, &token); |
a937b979 | 3790 | btrfs_set_token_timespec_nsec(leaf, &item->mtime, |
51fab693 | 3791 | inode->i_mtime.tv_nsec, &token); |
5f39d397 | 3792 | |
a937b979 | 3793 | btrfs_set_token_timespec_sec(leaf, &item->ctime, |
51fab693 | 3794 | inode->i_ctime.tv_sec, &token); |
a937b979 | 3795 | btrfs_set_token_timespec_nsec(leaf, &item->ctime, |
51fab693 | 3796 | inode->i_ctime.tv_nsec, &token); |
5f39d397 | 3797 | |
9cc97d64 | 3798 | btrfs_set_token_timespec_sec(leaf, &item->otime, |
3799 | BTRFS_I(inode)->i_otime.tv_sec, &token); | |
3800 | btrfs_set_token_timespec_nsec(leaf, &item->otime, | |
3801 | BTRFS_I(inode)->i_otime.tv_nsec, &token); | |
3802 | ||
51fab693 LB |
3803 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), |
3804 | &token); | |
3805 | btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation, | |
3806 | &token); | |
3807 | btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token); | |
3808 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); | |
3809 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3810 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3811 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
39279cc3 CM |
3812 | } |
3813 | ||
d352ac68 CM |
3814 | /* |
3815 | * copy everything in the in-memory inode into the btree. | |
3816 | */ | |
2115133f | 3817 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 3818 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
3819 | { |
3820 | struct btrfs_inode_item *inode_item; | |
3821 | struct btrfs_path *path; | |
5f39d397 | 3822 | struct extent_buffer *leaf; |
39279cc3 CM |
3823 | int ret; |
3824 | ||
3825 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
3826 | if (!path) |
3827 | return -ENOMEM; | |
3828 | ||
b9473439 | 3829 | path->leave_spinning = 1; |
16cdcec7 MX |
3830 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
3831 | 1); | |
39279cc3 CM |
3832 | if (ret) { |
3833 | if (ret > 0) | |
3834 | ret = -ENOENT; | |
3835 | goto failed; | |
3836 | } | |
3837 | ||
5f39d397 CM |
3838 | leaf = path->nodes[0]; |
3839 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 3840 | struct btrfs_inode_item); |
39279cc3 | 3841 | |
e02119d5 | 3842 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 3843 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 3844 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
3845 | ret = 0; |
3846 | failed: | |
39279cc3 CM |
3847 | btrfs_free_path(path); |
3848 | return ret; | |
3849 | } | |
3850 | ||
2115133f CM |
3851 | /* |
3852 | * copy everything in the in-memory inode into the btree. | |
3853 | */ | |
3854 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
3855 | struct btrfs_root *root, struct inode *inode) | |
3856 | { | |
0b246afa | 3857 | struct btrfs_fs_info *fs_info = root->fs_info; |
2115133f CM |
3858 | int ret; |
3859 | ||
3860 | /* | |
3861 | * If the inode is a free space inode, we can deadlock during commit | |
3862 | * if we put it into the delayed code. | |
3863 | * | |
3864 | * The data relocation inode should also be directly updated | |
3865 | * without delay | |
3866 | */ | |
70ddc553 | 3867 | if (!btrfs_is_free_space_inode(BTRFS_I(inode)) |
1d52c78a | 3868 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
0b246afa | 3869 | && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) { |
8ea05e3a AB |
3870 | btrfs_update_root_times(trans, root); |
3871 | ||
2115133f CM |
3872 | ret = btrfs_delayed_update_inode(trans, root, inode); |
3873 | if (!ret) | |
3874 | btrfs_set_inode_last_trans(trans, inode); | |
3875 | return ret; | |
3876 | } | |
3877 | ||
3878 | return btrfs_update_inode_item(trans, root, inode); | |
3879 | } | |
3880 | ||
be6aef60 JB |
3881 | noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
3882 | struct btrfs_root *root, | |
3883 | struct inode *inode) | |
2115133f CM |
3884 | { |
3885 | int ret; | |
3886 | ||
3887 | ret = btrfs_update_inode(trans, root, inode); | |
3888 | if (ret == -ENOSPC) | |
3889 | return btrfs_update_inode_item(trans, root, inode); | |
3890 | return ret; | |
3891 | } | |
3892 | ||
d352ac68 CM |
3893 | /* |
3894 | * unlink helper that gets used here in inode.c and in the tree logging | |
3895 | * recovery code. It remove a link in a directory with a given name, and | |
3896 | * also drops the back refs in the inode to the directory | |
3897 | */ | |
92986796 AV |
3898 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3899 | struct btrfs_root *root, | |
4ec5934e NB |
3900 | struct btrfs_inode *dir, |
3901 | struct btrfs_inode *inode, | |
92986796 | 3902 | const char *name, int name_len) |
39279cc3 | 3903 | { |
0b246afa | 3904 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 3905 | struct btrfs_path *path; |
39279cc3 | 3906 | int ret = 0; |
5f39d397 | 3907 | struct extent_buffer *leaf; |
39279cc3 | 3908 | struct btrfs_dir_item *di; |
5f39d397 | 3909 | struct btrfs_key key; |
aec7477b | 3910 | u64 index; |
33345d01 LZ |
3911 | u64 ino = btrfs_ino(inode); |
3912 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
3913 | |
3914 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
3915 | if (!path) { |
3916 | ret = -ENOMEM; | |
554233a6 | 3917 | goto out; |
54aa1f4d CM |
3918 | } |
3919 | ||
b9473439 | 3920 | path->leave_spinning = 1; |
33345d01 | 3921 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 CM |
3922 | name, name_len, -1); |
3923 | if (IS_ERR(di)) { | |
3924 | ret = PTR_ERR(di); | |
3925 | goto err; | |
3926 | } | |
3927 | if (!di) { | |
3928 | ret = -ENOENT; | |
3929 | goto err; | |
3930 | } | |
5f39d397 CM |
3931 | leaf = path->nodes[0]; |
3932 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 3933 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
3934 | if (ret) |
3935 | goto err; | |
b3b4aa74 | 3936 | btrfs_release_path(path); |
39279cc3 | 3937 | |
67de1176 MX |
3938 | /* |
3939 | * If we don't have dir index, we have to get it by looking up | |
3940 | * the inode ref, since we get the inode ref, remove it directly, | |
3941 | * it is unnecessary to do delayed deletion. | |
3942 | * | |
3943 | * But if we have dir index, needn't search inode ref to get it. | |
3944 | * Since the inode ref is close to the inode item, it is better | |
3945 | * that we delay to delete it, and just do this deletion when | |
3946 | * we update the inode item. | |
3947 | */ | |
4ec5934e | 3948 | if (inode->dir_index) { |
67de1176 MX |
3949 | ret = btrfs_delayed_delete_inode_ref(inode); |
3950 | if (!ret) { | |
4ec5934e | 3951 | index = inode->dir_index; |
67de1176 MX |
3952 | goto skip_backref; |
3953 | } | |
3954 | } | |
3955 | ||
33345d01 LZ |
3956 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
3957 | dir_ino, &index); | |
aec7477b | 3958 | if (ret) { |
0b246afa | 3959 | btrfs_info(fs_info, |
c2cf52eb | 3960 | "failed to delete reference to %.*s, inode %llu parent %llu", |
c1c9ff7c | 3961 | name_len, name, ino, dir_ino); |
66642832 | 3962 | btrfs_abort_transaction(trans, ret); |
aec7477b JB |
3963 | goto err; |
3964 | } | |
67de1176 | 3965 | skip_backref: |
2ff7e61e | 3966 | ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index); |
79787eaa | 3967 | if (ret) { |
66642832 | 3968 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 3969 | goto err; |
79787eaa | 3970 | } |
39279cc3 | 3971 | |
4ec5934e NB |
3972 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode, |
3973 | dir_ino); | |
79787eaa | 3974 | if (ret != 0 && ret != -ENOENT) { |
66642832 | 3975 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3976 | goto err; |
3977 | } | |
e02119d5 | 3978 | |
4ec5934e NB |
3979 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir, |
3980 | index); | |
6418c961 CM |
3981 | if (ret == -ENOENT) |
3982 | ret = 0; | |
d4e3991b | 3983 | else if (ret) |
66642832 | 3984 | btrfs_abort_transaction(trans, ret); |
39279cc3 CM |
3985 | err: |
3986 | btrfs_free_path(path); | |
e02119d5 CM |
3987 | if (ret) |
3988 | goto out; | |
3989 | ||
6ef06d27 | 3990 | btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2); |
4ec5934e NB |
3991 | inode_inc_iversion(&inode->vfs_inode); |
3992 | inode_inc_iversion(&dir->vfs_inode); | |
3993 | inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime = | |
3994 | dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode); | |
3995 | ret = btrfs_update_inode(trans, root, &dir->vfs_inode); | |
e02119d5 | 3996 | out: |
39279cc3 CM |
3997 | return ret; |
3998 | } | |
3999 | ||
92986796 AV |
4000 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
4001 | struct btrfs_root *root, | |
4ec5934e | 4002 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
92986796 AV |
4003 | const char *name, int name_len) |
4004 | { | |
4005 | int ret; | |
4006 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
4007 | if (!ret) { | |
4ec5934e NB |
4008 | drop_nlink(&inode->vfs_inode); |
4009 | ret = btrfs_update_inode(trans, root, &inode->vfs_inode); | |
92986796 AV |
4010 | } |
4011 | return ret; | |
4012 | } | |
39279cc3 | 4013 | |
a22285a6 YZ |
4014 | /* |
4015 | * helper to start transaction for unlink and rmdir. | |
4016 | * | |
d52be818 JB |
4017 | * unlink and rmdir are special in btrfs, they do not always free space, so |
4018 | * if we cannot make our reservations the normal way try and see if there is | |
4019 | * plenty of slack room in the global reserve to migrate, otherwise we cannot | |
4020 | * allow the unlink to occur. | |
a22285a6 | 4021 | */ |
d52be818 | 4022 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir) |
4df27c4d | 4023 | { |
a22285a6 | 4024 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 4025 | |
e70bea5f JB |
4026 | /* |
4027 | * 1 for the possible orphan item | |
4028 | * 1 for the dir item | |
4029 | * 1 for the dir index | |
4030 | * 1 for the inode ref | |
e70bea5f JB |
4031 | * 1 for the inode |
4032 | */ | |
8eab77ff | 4033 | return btrfs_start_transaction_fallback_global_rsv(root, 5, 5); |
a22285a6 YZ |
4034 | } |
4035 | ||
4036 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
4037 | { | |
4038 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4039 | struct btrfs_trans_handle *trans; | |
2b0143b5 | 4040 | struct inode *inode = d_inode(dentry); |
a22285a6 | 4041 | int ret; |
a22285a6 | 4042 | |
d52be818 | 4043 | trans = __unlink_start_trans(dir); |
a22285a6 YZ |
4044 | if (IS_ERR(trans)) |
4045 | return PTR_ERR(trans); | |
5f39d397 | 4046 | |
4ec5934e NB |
4047 | btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)), |
4048 | 0); | |
12fcfd22 | 4049 | |
4ec5934e NB |
4050 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
4051 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, | |
4052 | dentry->d_name.len); | |
b532402e TI |
4053 | if (ret) |
4054 | goto out; | |
7b128766 | 4055 | |
a22285a6 | 4056 | if (inode->i_nlink == 0) { |
73f2e545 | 4057 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
b532402e TI |
4058 | if (ret) |
4059 | goto out; | |
a22285a6 | 4060 | } |
7b128766 | 4061 | |
b532402e | 4062 | out: |
3a45bb20 | 4063 | btrfs_end_transaction(trans); |
2ff7e61e | 4064 | btrfs_btree_balance_dirty(root->fs_info); |
39279cc3 CM |
4065 | return ret; |
4066 | } | |
4067 | ||
4df27c4d YZ |
4068 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
4069 | struct btrfs_root *root, | |
4070 | struct inode *dir, u64 objectid, | |
4071 | const char *name, int name_len) | |
4072 | { | |
0b246afa | 4073 | struct btrfs_fs_info *fs_info = root->fs_info; |
4df27c4d YZ |
4074 | struct btrfs_path *path; |
4075 | struct extent_buffer *leaf; | |
4076 | struct btrfs_dir_item *di; | |
4077 | struct btrfs_key key; | |
4078 | u64 index; | |
4079 | int ret; | |
4a0cc7ca | 4080 | u64 dir_ino = btrfs_ino(BTRFS_I(dir)); |
4df27c4d YZ |
4081 | |
4082 | path = btrfs_alloc_path(); | |
4083 | if (!path) | |
4084 | return -ENOMEM; | |
4085 | ||
33345d01 | 4086 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 4087 | name, name_len, -1); |
79787eaa JM |
4088 | if (IS_ERR_OR_NULL(di)) { |
4089 | if (!di) | |
4090 | ret = -ENOENT; | |
4091 | else | |
4092 | ret = PTR_ERR(di); | |
4093 | goto out; | |
4094 | } | |
4df27c4d YZ |
4095 | |
4096 | leaf = path->nodes[0]; | |
4097 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
4098 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
4099 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa | 4100 | if (ret) { |
66642832 | 4101 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4102 | goto out; |
4103 | } | |
b3b4aa74 | 4104 | btrfs_release_path(path); |
4df27c4d | 4105 | |
0b246afa JM |
4106 | ret = btrfs_del_root_ref(trans, fs_info, objectid, |
4107 | root->root_key.objectid, dir_ino, | |
4108 | &index, name, name_len); | |
4df27c4d | 4109 | if (ret < 0) { |
79787eaa | 4110 | if (ret != -ENOENT) { |
66642832 | 4111 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4112 | goto out; |
4113 | } | |
33345d01 | 4114 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 4115 | name, name_len); |
79787eaa JM |
4116 | if (IS_ERR_OR_NULL(di)) { |
4117 | if (!di) | |
4118 | ret = -ENOENT; | |
4119 | else | |
4120 | ret = PTR_ERR(di); | |
66642832 | 4121 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4122 | goto out; |
4123 | } | |
4df27c4d YZ |
4124 | |
4125 | leaf = path->nodes[0]; | |
4126 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 4127 | btrfs_release_path(path); |
4df27c4d YZ |
4128 | index = key.offset; |
4129 | } | |
945d8962 | 4130 | btrfs_release_path(path); |
4df27c4d | 4131 | |
e67bbbb9 | 4132 | ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index); |
79787eaa | 4133 | if (ret) { |
66642832 | 4134 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4135 | goto out; |
4136 | } | |
4df27c4d | 4137 | |
6ef06d27 | 4138 | btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2); |
0c4d2d95 | 4139 | inode_inc_iversion(dir); |
c2050a45 | 4140 | dir->i_mtime = dir->i_ctime = current_time(dir); |
5a24e84c | 4141 | ret = btrfs_update_inode_fallback(trans, root, dir); |
79787eaa | 4142 | if (ret) |
66642832 | 4143 | btrfs_abort_transaction(trans, ret); |
79787eaa | 4144 | out: |
71d7aed0 | 4145 | btrfs_free_path(path); |
79787eaa | 4146 | return ret; |
4df27c4d YZ |
4147 | } |
4148 | ||
39279cc3 CM |
4149 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
4150 | { | |
2b0143b5 | 4151 | struct inode *inode = d_inode(dentry); |
1832a6d5 | 4152 | int err = 0; |
39279cc3 | 4153 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 4154 | struct btrfs_trans_handle *trans; |
44f714da | 4155 | u64 last_unlink_trans; |
39279cc3 | 4156 | |
b3ae244e | 4157 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 4158 | return -ENOTEMPTY; |
4a0cc7ca | 4159 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) |
b3ae244e | 4160 | return -EPERM; |
134d4512 | 4161 | |
d52be818 | 4162 | trans = __unlink_start_trans(dir); |
a22285a6 | 4163 | if (IS_ERR(trans)) |
5df6a9f6 | 4164 | return PTR_ERR(trans); |
5df6a9f6 | 4165 | |
4a0cc7ca | 4166 | if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
4df27c4d YZ |
4167 | err = btrfs_unlink_subvol(trans, root, dir, |
4168 | BTRFS_I(inode)->location.objectid, | |
4169 | dentry->d_name.name, | |
4170 | dentry->d_name.len); | |
4171 | goto out; | |
4172 | } | |
4173 | ||
73f2e545 | 4174 | err = btrfs_orphan_add(trans, BTRFS_I(inode)); |
7b128766 | 4175 | if (err) |
4df27c4d | 4176 | goto out; |
7b128766 | 4177 | |
44f714da FM |
4178 | last_unlink_trans = BTRFS_I(inode)->last_unlink_trans; |
4179 | ||
39279cc3 | 4180 | /* now the directory is empty */ |
4ec5934e NB |
4181 | err = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
4182 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, | |
4183 | dentry->d_name.len); | |
44f714da | 4184 | if (!err) { |
6ef06d27 | 4185 | btrfs_i_size_write(BTRFS_I(inode), 0); |
44f714da FM |
4186 | /* |
4187 | * Propagate the last_unlink_trans value of the deleted dir to | |
4188 | * its parent directory. This is to prevent an unrecoverable | |
4189 | * log tree in the case we do something like this: | |
4190 | * 1) create dir foo | |
4191 | * 2) create snapshot under dir foo | |
4192 | * 3) delete the snapshot | |
4193 | * 4) rmdir foo | |
4194 | * 5) mkdir foo | |
4195 | * 6) fsync foo or some file inside foo | |
4196 | */ | |
4197 | if (last_unlink_trans >= trans->transid) | |
4198 | BTRFS_I(dir)->last_unlink_trans = last_unlink_trans; | |
4199 | } | |
4df27c4d | 4200 | out: |
3a45bb20 | 4201 | btrfs_end_transaction(trans); |
2ff7e61e | 4202 | btrfs_btree_balance_dirty(root->fs_info); |
3954401f | 4203 | |
39279cc3 CM |
4204 | return err; |
4205 | } | |
4206 | ||
28f75a0e CM |
4207 | static int truncate_space_check(struct btrfs_trans_handle *trans, |
4208 | struct btrfs_root *root, | |
4209 | u64 bytes_deleted) | |
4210 | { | |
0b246afa | 4211 | struct btrfs_fs_info *fs_info = root->fs_info; |
28f75a0e CM |
4212 | int ret; |
4213 | ||
dc95f7bf JB |
4214 | /* |
4215 | * This is only used to apply pressure to the enospc system, we don't | |
4216 | * intend to use this reservation at all. | |
4217 | */ | |
2ff7e61e | 4218 | bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted); |
0b246afa JM |
4219 | bytes_deleted *= fs_info->nodesize; |
4220 | ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv, | |
28f75a0e | 4221 | bytes_deleted, BTRFS_RESERVE_NO_FLUSH); |
dc95f7bf | 4222 | if (!ret) { |
0b246afa | 4223 | trace_btrfs_space_reservation(fs_info, "transaction", |
dc95f7bf JB |
4224 | trans->transid, |
4225 | bytes_deleted, 1); | |
28f75a0e | 4226 | trans->bytes_reserved += bytes_deleted; |
dc95f7bf | 4227 | } |
28f75a0e CM |
4228 | return ret; |
4229 | ||
4230 | } | |
4231 | ||
0305cd5f FM |
4232 | static int truncate_inline_extent(struct inode *inode, |
4233 | struct btrfs_path *path, | |
4234 | struct btrfs_key *found_key, | |
4235 | const u64 item_end, | |
4236 | const u64 new_size) | |
4237 | { | |
4238 | struct extent_buffer *leaf = path->nodes[0]; | |
4239 | int slot = path->slots[0]; | |
4240 | struct btrfs_file_extent_item *fi; | |
4241 | u32 size = (u32)(new_size - found_key->offset); | |
4242 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4243 | ||
4244 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
4245 | ||
4246 | if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) { | |
4247 | loff_t offset = new_size; | |
09cbfeaf | 4248 | loff_t page_end = ALIGN(offset, PAGE_SIZE); |
0305cd5f FM |
4249 | |
4250 | /* | |
4251 | * Zero out the remaining of the last page of our inline extent, | |
4252 | * instead of directly truncating our inline extent here - that | |
4253 | * would be much more complex (decompressing all the data, then | |
4254 | * compressing the truncated data, which might be bigger than | |
4255 | * the size of the inline extent, resize the extent, etc). | |
4256 | * We release the path because to get the page we might need to | |
4257 | * read the extent item from disk (data not in the page cache). | |
4258 | */ | |
4259 | btrfs_release_path(path); | |
9703fefe CR |
4260 | return btrfs_truncate_block(inode, offset, page_end - offset, |
4261 | 0); | |
0305cd5f FM |
4262 | } |
4263 | ||
4264 | btrfs_set_file_extent_ram_bytes(leaf, fi, size); | |
4265 | size = btrfs_file_extent_calc_inline_size(size); | |
2ff7e61e | 4266 | btrfs_truncate_item(root->fs_info, path, size, 1); |
0305cd5f FM |
4267 | |
4268 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) | |
4269 | inode_sub_bytes(inode, item_end + 1 - new_size); | |
4270 | ||
4271 | return 0; | |
4272 | } | |
4273 | ||
39279cc3 CM |
4274 | /* |
4275 | * this can truncate away extent items, csum items and directory items. | |
4276 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 4277 | * any higher than new_size |
39279cc3 CM |
4278 | * |
4279 | * csum items that cross the new i_size are truncated to the new size | |
4280 | * as well. | |
7b128766 JB |
4281 | * |
4282 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
4283 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 4284 | */ |
8082510e YZ |
4285 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
4286 | struct btrfs_root *root, | |
4287 | struct inode *inode, | |
4288 | u64 new_size, u32 min_type) | |
39279cc3 | 4289 | { |
0b246afa | 4290 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 4291 | struct btrfs_path *path; |
5f39d397 | 4292 | struct extent_buffer *leaf; |
39279cc3 | 4293 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
4294 | struct btrfs_key key; |
4295 | struct btrfs_key found_key; | |
39279cc3 | 4296 | u64 extent_start = 0; |
db94535d | 4297 | u64 extent_num_bytes = 0; |
5d4f98a2 | 4298 | u64 extent_offset = 0; |
39279cc3 | 4299 | u64 item_end = 0; |
c1aa4575 | 4300 | u64 last_size = new_size; |
8082510e | 4301 | u32 found_type = (u8)-1; |
39279cc3 CM |
4302 | int found_extent; |
4303 | int del_item; | |
85e21bac CM |
4304 | int pending_del_nr = 0; |
4305 | int pending_del_slot = 0; | |
179e29e4 | 4306 | int extent_type = -1; |
8082510e YZ |
4307 | int ret; |
4308 | int err = 0; | |
4a0cc7ca | 4309 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
28ed1345 | 4310 | u64 bytes_deleted = 0; |
1262133b JB |
4311 | bool be_nice = 0; |
4312 | bool should_throttle = 0; | |
28f75a0e | 4313 | bool should_end = 0; |
8082510e YZ |
4314 | |
4315 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 4316 | |
28ed1345 CM |
4317 | /* |
4318 | * for non-free space inodes and ref cows, we want to back off from | |
4319 | * time to time | |
4320 | */ | |
70ddc553 | 4321 | if (!btrfs_is_free_space_inode(BTRFS_I(inode)) && |
28ed1345 CM |
4322 | test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
4323 | be_nice = 1; | |
4324 | ||
0eb0e19c MF |
4325 | path = btrfs_alloc_path(); |
4326 | if (!path) | |
4327 | return -ENOMEM; | |
e4058b54 | 4328 | path->reada = READA_BACK; |
0eb0e19c | 4329 | |
5dc562c5 JB |
4330 | /* |
4331 | * We want to drop from the next block forward in case this new size is | |
4332 | * not block aligned since we will be keeping the last block of the | |
4333 | * extent just the way it is. | |
4334 | */ | |
27cdeb70 | 4335 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
0b246afa | 4336 | root == fs_info->tree_root) |
dcdbc059 | 4337 | btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size, |
0b246afa | 4338 | fs_info->sectorsize), |
da17066c | 4339 | (u64)-1, 0); |
8082510e | 4340 | |
16cdcec7 MX |
4341 | /* |
4342 | * This function is also used to drop the items in the log tree before | |
4343 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
4344 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
4345 | * items. | |
4346 | */ | |
4347 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
4ccb5c72 | 4348 | btrfs_kill_delayed_inode_items(BTRFS_I(inode)); |
16cdcec7 | 4349 | |
33345d01 | 4350 | key.objectid = ino; |
39279cc3 | 4351 | key.offset = (u64)-1; |
5f39d397 CM |
4352 | key.type = (u8)-1; |
4353 | ||
85e21bac | 4354 | search_again: |
28ed1345 CM |
4355 | /* |
4356 | * with a 16K leaf size and 128MB extents, you can actually queue | |
4357 | * up a huge file in a single leaf. Most of the time that | |
4358 | * bytes_deleted is > 0, it will be huge by the time we get here | |
4359 | */ | |
ee22184b | 4360 | if (be_nice && bytes_deleted > SZ_32M) { |
3a45bb20 | 4361 | if (btrfs_should_end_transaction(trans)) { |
28ed1345 CM |
4362 | err = -EAGAIN; |
4363 | goto error; | |
4364 | } | |
4365 | } | |
4366 | ||
4367 | ||
b9473439 | 4368 | path->leave_spinning = 1; |
85e21bac | 4369 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
4370 | if (ret < 0) { |
4371 | err = ret; | |
4372 | goto out; | |
4373 | } | |
d397712b | 4374 | |
85e21bac | 4375 | if (ret > 0) { |
e02119d5 CM |
4376 | /* there are no items in the tree for us to truncate, we're |
4377 | * done | |
4378 | */ | |
8082510e YZ |
4379 | if (path->slots[0] == 0) |
4380 | goto out; | |
85e21bac CM |
4381 | path->slots[0]--; |
4382 | } | |
4383 | ||
d397712b | 4384 | while (1) { |
39279cc3 | 4385 | fi = NULL; |
5f39d397 CM |
4386 | leaf = path->nodes[0]; |
4387 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
962a298f | 4388 | found_type = found_key.type; |
39279cc3 | 4389 | |
33345d01 | 4390 | if (found_key.objectid != ino) |
39279cc3 | 4391 | break; |
5f39d397 | 4392 | |
85e21bac | 4393 | if (found_type < min_type) |
39279cc3 CM |
4394 | break; |
4395 | ||
5f39d397 | 4396 | item_end = found_key.offset; |
39279cc3 | 4397 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 4398 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 4399 | struct btrfs_file_extent_item); |
179e29e4 CM |
4400 | extent_type = btrfs_file_extent_type(leaf, fi); |
4401 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4402 | item_end += |
db94535d | 4403 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 4404 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 4405 | item_end += btrfs_file_extent_inline_len(leaf, |
514ac8ad | 4406 | path->slots[0], fi); |
39279cc3 | 4407 | } |
008630c1 | 4408 | item_end--; |
39279cc3 | 4409 | } |
8082510e YZ |
4410 | if (found_type > min_type) { |
4411 | del_item = 1; | |
4412 | } else { | |
76b42abb | 4413 | if (item_end < new_size) |
b888db2b | 4414 | break; |
8082510e YZ |
4415 | if (found_key.offset >= new_size) |
4416 | del_item = 1; | |
4417 | else | |
4418 | del_item = 0; | |
39279cc3 | 4419 | } |
39279cc3 | 4420 | found_extent = 0; |
39279cc3 | 4421 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
4422 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
4423 | goto delete; | |
4424 | ||
7f4f6e0a JB |
4425 | if (del_item) |
4426 | last_size = found_key.offset; | |
4427 | else | |
4428 | last_size = new_size; | |
4429 | ||
179e29e4 | 4430 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
39279cc3 | 4431 | u64 num_dec; |
db94535d | 4432 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 4433 | if (!del_item) { |
db94535d CM |
4434 | u64 orig_num_bytes = |
4435 | btrfs_file_extent_num_bytes(leaf, fi); | |
fda2832f QW |
4436 | extent_num_bytes = ALIGN(new_size - |
4437 | found_key.offset, | |
0b246afa | 4438 | fs_info->sectorsize); |
db94535d CM |
4439 | btrfs_set_file_extent_num_bytes(leaf, fi, |
4440 | extent_num_bytes); | |
4441 | num_dec = (orig_num_bytes - | |
9069218d | 4442 | extent_num_bytes); |
27cdeb70 MX |
4443 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4444 | &root->state) && | |
4445 | extent_start != 0) | |
a76a3cd4 | 4446 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 4447 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 4448 | } else { |
db94535d CM |
4449 | extent_num_bytes = |
4450 | btrfs_file_extent_disk_num_bytes(leaf, | |
4451 | fi); | |
5d4f98a2 YZ |
4452 | extent_offset = found_key.offset - |
4453 | btrfs_file_extent_offset(leaf, fi); | |
4454 | ||
39279cc3 | 4455 | /* FIXME blocksize != 4096 */ |
9069218d | 4456 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
4457 | if (extent_start != 0) { |
4458 | found_extent = 1; | |
27cdeb70 MX |
4459 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4460 | &root->state)) | |
a76a3cd4 | 4461 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 4462 | } |
39279cc3 | 4463 | } |
9069218d | 4464 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
4465 | /* |
4466 | * we can't truncate inline items that have had | |
4467 | * special encodings | |
4468 | */ | |
4469 | if (!del_item && | |
c8b97818 CM |
4470 | btrfs_file_extent_encryption(leaf, fi) == 0 && |
4471 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
514ac8ad CM |
4472 | |
4473 | /* | |
0305cd5f FM |
4474 | * Need to release path in order to truncate a |
4475 | * compressed extent. So delete any accumulated | |
4476 | * extent items so far. | |
514ac8ad | 4477 | */ |
0305cd5f FM |
4478 | if (btrfs_file_extent_compression(leaf, fi) != |
4479 | BTRFS_COMPRESS_NONE && pending_del_nr) { | |
4480 | err = btrfs_del_items(trans, root, path, | |
4481 | pending_del_slot, | |
4482 | pending_del_nr); | |
4483 | if (err) { | |
4484 | btrfs_abort_transaction(trans, | |
0305cd5f FM |
4485 | err); |
4486 | goto error; | |
4487 | } | |
4488 | pending_del_nr = 0; | |
4489 | } | |
4490 | ||
4491 | err = truncate_inline_extent(inode, path, | |
4492 | &found_key, | |
4493 | item_end, | |
4494 | new_size); | |
4495 | if (err) { | |
66642832 | 4496 | btrfs_abort_transaction(trans, err); |
0305cd5f FM |
4497 | goto error; |
4498 | } | |
27cdeb70 MX |
4499 | } else if (test_bit(BTRFS_ROOT_REF_COWS, |
4500 | &root->state)) { | |
0305cd5f | 4501 | inode_sub_bytes(inode, item_end + 1 - new_size); |
9069218d | 4502 | } |
39279cc3 | 4503 | } |
179e29e4 | 4504 | delete: |
39279cc3 | 4505 | if (del_item) { |
85e21bac CM |
4506 | if (!pending_del_nr) { |
4507 | /* no pending yet, add ourselves */ | |
4508 | pending_del_slot = path->slots[0]; | |
4509 | pending_del_nr = 1; | |
4510 | } else if (pending_del_nr && | |
4511 | path->slots[0] + 1 == pending_del_slot) { | |
4512 | /* hop on the pending chunk */ | |
4513 | pending_del_nr++; | |
4514 | pending_del_slot = path->slots[0]; | |
4515 | } else { | |
d397712b | 4516 | BUG(); |
85e21bac | 4517 | } |
39279cc3 CM |
4518 | } else { |
4519 | break; | |
4520 | } | |
28f75a0e CM |
4521 | should_throttle = 0; |
4522 | ||
27cdeb70 MX |
4523 | if (found_extent && |
4524 | (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || | |
0b246afa | 4525 | root == fs_info->tree_root)) { |
b9473439 | 4526 | btrfs_set_path_blocking(path); |
28ed1345 | 4527 | bytes_deleted += extent_num_bytes; |
2ff7e61e | 4528 | ret = btrfs_free_extent(trans, fs_info, extent_start, |
5d4f98a2 YZ |
4529 | extent_num_bytes, 0, |
4530 | btrfs_header_owner(leaf), | |
b06c4bf5 | 4531 | ino, extent_offset); |
39279cc3 | 4532 | BUG_ON(ret); |
2ff7e61e JM |
4533 | if (btrfs_should_throttle_delayed_refs(trans, fs_info)) |
4534 | btrfs_async_run_delayed_refs(fs_info, | |
dd4b857a WX |
4535 | trans->delayed_ref_updates * 2, |
4536 | trans->transid, 0); | |
28f75a0e CM |
4537 | if (be_nice) { |
4538 | if (truncate_space_check(trans, root, | |
4539 | extent_num_bytes)) { | |
4540 | should_end = 1; | |
4541 | } | |
4542 | if (btrfs_should_throttle_delayed_refs(trans, | |
2ff7e61e | 4543 | fs_info)) |
28f75a0e | 4544 | should_throttle = 1; |
28f75a0e | 4545 | } |
39279cc3 | 4546 | } |
85e21bac | 4547 | |
8082510e YZ |
4548 | if (found_type == BTRFS_INODE_ITEM_KEY) |
4549 | break; | |
4550 | ||
4551 | if (path->slots[0] == 0 || | |
1262133b | 4552 | path->slots[0] != pending_del_slot || |
28f75a0e | 4553 | should_throttle || should_end) { |
8082510e YZ |
4554 | if (pending_del_nr) { |
4555 | ret = btrfs_del_items(trans, root, path, | |
4556 | pending_del_slot, | |
4557 | pending_del_nr); | |
79787eaa | 4558 | if (ret) { |
66642832 | 4559 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4560 | goto error; |
4561 | } | |
8082510e YZ |
4562 | pending_del_nr = 0; |
4563 | } | |
b3b4aa74 | 4564 | btrfs_release_path(path); |
28f75a0e | 4565 | if (should_throttle) { |
1262133b JB |
4566 | unsigned long updates = trans->delayed_ref_updates; |
4567 | if (updates) { | |
4568 | trans->delayed_ref_updates = 0; | |
2ff7e61e JM |
4569 | ret = btrfs_run_delayed_refs(trans, |
4570 | fs_info, | |
4571 | updates * 2); | |
1262133b JB |
4572 | if (ret && !err) |
4573 | err = ret; | |
4574 | } | |
4575 | } | |
28f75a0e CM |
4576 | /* |
4577 | * if we failed to refill our space rsv, bail out | |
4578 | * and let the transaction restart | |
4579 | */ | |
4580 | if (should_end) { | |
4581 | err = -EAGAIN; | |
4582 | goto error; | |
4583 | } | |
85e21bac | 4584 | goto search_again; |
8082510e YZ |
4585 | } else { |
4586 | path->slots[0]--; | |
85e21bac | 4587 | } |
39279cc3 | 4588 | } |
8082510e | 4589 | out: |
85e21bac CM |
4590 | if (pending_del_nr) { |
4591 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
4592 | pending_del_nr); | |
79787eaa | 4593 | if (ret) |
66642832 | 4594 | btrfs_abort_transaction(trans, ret); |
85e21bac | 4595 | } |
79787eaa | 4596 | error: |
76b42abb FM |
4597 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
4598 | ASSERT(last_size >= new_size); | |
4599 | if (!err && last_size > new_size) | |
4600 | last_size = new_size; | |
7f4f6e0a | 4601 | btrfs_ordered_update_i_size(inode, last_size, NULL); |
76b42abb | 4602 | } |
28ed1345 | 4603 | |
39279cc3 | 4604 | btrfs_free_path(path); |
28ed1345 | 4605 | |
19fd2df5 LB |
4606 | if (err == 0) { |
4607 | /* only inline file may have last_size != new_size */ | |
4608 | if (new_size >= fs_info->sectorsize || | |
4609 | new_size > fs_info->max_inline) | |
4610 | ASSERT(last_size == new_size); | |
4611 | } | |
4612 | ||
ee22184b | 4613 | if (be_nice && bytes_deleted > SZ_32M) { |
28ed1345 CM |
4614 | unsigned long updates = trans->delayed_ref_updates; |
4615 | if (updates) { | |
4616 | trans->delayed_ref_updates = 0; | |
2ff7e61e JM |
4617 | ret = btrfs_run_delayed_refs(trans, fs_info, |
4618 | updates * 2); | |
28ed1345 CM |
4619 | if (ret && !err) |
4620 | err = ret; | |
4621 | } | |
4622 | } | |
8082510e | 4623 | return err; |
39279cc3 CM |
4624 | } |
4625 | ||
4626 | /* | |
9703fefe | 4627 | * btrfs_truncate_block - read, zero a chunk and write a block |
2aaa6655 JB |
4628 | * @inode - inode that we're zeroing |
4629 | * @from - the offset to start zeroing | |
4630 | * @len - the length to zero, 0 to zero the entire range respective to the | |
4631 | * offset | |
4632 | * @front - zero up to the offset instead of from the offset on | |
4633 | * | |
9703fefe | 4634 | * This will find the block for the "from" offset and cow the block and zero the |
2aaa6655 | 4635 | * part we want to zero. This is used with truncate and hole punching. |
39279cc3 | 4636 | */ |
9703fefe | 4637 | int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len, |
2aaa6655 | 4638 | int front) |
39279cc3 | 4639 | { |
0b246afa | 4640 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2aaa6655 | 4641 | struct address_space *mapping = inode->i_mapping; |
e6dcd2dc CM |
4642 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4643 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 4644 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 4645 | char *kaddr; |
0b246afa | 4646 | u32 blocksize = fs_info->sectorsize; |
09cbfeaf | 4647 | pgoff_t index = from >> PAGE_SHIFT; |
9703fefe | 4648 | unsigned offset = from & (blocksize - 1); |
39279cc3 | 4649 | struct page *page; |
3b16a4e3 | 4650 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 4651 | int ret = 0; |
9703fefe CR |
4652 | u64 block_start; |
4653 | u64 block_end; | |
39279cc3 | 4654 | |
2aaa6655 JB |
4655 | if ((offset & (blocksize - 1)) == 0 && |
4656 | (!len || ((len & (blocksize - 1)) == 0))) | |
39279cc3 | 4657 | goto out; |
9703fefe | 4658 | |
7cf5b976 | 4659 | ret = btrfs_delalloc_reserve_space(inode, |
9703fefe | 4660 | round_down(from, blocksize), blocksize); |
5d5e103a JB |
4661 | if (ret) |
4662 | goto out; | |
39279cc3 | 4663 | |
211c17f5 | 4664 | again: |
3b16a4e3 | 4665 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 4666 | if (!page) { |
7cf5b976 | 4667 | btrfs_delalloc_release_space(inode, |
9703fefe CR |
4668 | round_down(from, blocksize), |
4669 | blocksize); | |
ac6a2b36 | 4670 | ret = -ENOMEM; |
39279cc3 | 4671 | goto out; |
5d5e103a | 4672 | } |
e6dcd2dc | 4673 | |
9703fefe CR |
4674 | block_start = round_down(from, blocksize); |
4675 | block_end = block_start + blocksize - 1; | |
e6dcd2dc | 4676 | |
39279cc3 | 4677 | if (!PageUptodate(page)) { |
9ebefb18 | 4678 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 4679 | lock_page(page); |
211c17f5 CM |
4680 | if (page->mapping != mapping) { |
4681 | unlock_page(page); | |
09cbfeaf | 4682 | put_page(page); |
211c17f5 CM |
4683 | goto again; |
4684 | } | |
39279cc3 CM |
4685 | if (!PageUptodate(page)) { |
4686 | ret = -EIO; | |
89642229 | 4687 | goto out_unlock; |
39279cc3 CM |
4688 | } |
4689 | } | |
211c17f5 | 4690 | wait_on_page_writeback(page); |
e6dcd2dc | 4691 | |
9703fefe | 4692 | lock_extent_bits(io_tree, block_start, block_end, &cached_state); |
e6dcd2dc CM |
4693 | set_page_extent_mapped(page); |
4694 | ||
9703fefe | 4695 | ordered = btrfs_lookup_ordered_extent(inode, block_start); |
e6dcd2dc | 4696 | if (ordered) { |
9703fefe | 4697 | unlock_extent_cached(io_tree, block_start, block_end, |
2ac55d41 | 4698 | &cached_state, GFP_NOFS); |
e6dcd2dc | 4699 | unlock_page(page); |
09cbfeaf | 4700 | put_page(page); |
eb84ae03 | 4701 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
4702 | btrfs_put_ordered_extent(ordered); |
4703 | goto again; | |
4704 | } | |
4705 | ||
9703fefe | 4706 | clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end, |
9e8a4a8b LB |
4707 | EXTENT_DIRTY | EXTENT_DELALLOC | |
4708 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 4709 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 4710 | |
9703fefe | 4711 | ret = btrfs_set_extent_delalloc(inode, block_start, block_end, |
ba8b04c1 | 4712 | &cached_state, 0); |
9ed74f2d | 4713 | if (ret) { |
9703fefe | 4714 | unlock_extent_cached(io_tree, block_start, block_end, |
2ac55d41 | 4715 | &cached_state, GFP_NOFS); |
9ed74f2d JB |
4716 | goto out_unlock; |
4717 | } | |
4718 | ||
9703fefe | 4719 | if (offset != blocksize) { |
2aaa6655 | 4720 | if (!len) |
9703fefe | 4721 | len = blocksize - offset; |
e6dcd2dc | 4722 | kaddr = kmap(page); |
2aaa6655 | 4723 | if (front) |
9703fefe CR |
4724 | memset(kaddr + (block_start - page_offset(page)), |
4725 | 0, offset); | |
2aaa6655 | 4726 | else |
9703fefe CR |
4727 | memset(kaddr + (block_start - page_offset(page)) + offset, |
4728 | 0, len); | |
e6dcd2dc CM |
4729 | flush_dcache_page(page); |
4730 | kunmap(page); | |
4731 | } | |
247e743c | 4732 | ClearPageChecked(page); |
e6dcd2dc | 4733 | set_page_dirty(page); |
9703fefe | 4734 | unlock_extent_cached(io_tree, block_start, block_end, &cached_state, |
2ac55d41 | 4735 | GFP_NOFS); |
39279cc3 | 4736 | |
89642229 | 4737 | out_unlock: |
5d5e103a | 4738 | if (ret) |
9703fefe CR |
4739 | btrfs_delalloc_release_space(inode, block_start, |
4740 | blocksize); | |
39279cc3 | 4741 | unlock_page(page); |
09cbfeaf | 4742 | put_page(page); |
39279cc3 CM |
4743 | out: |
4744 | return ret; | |
4745 | } | |
4746 | ||
16e7549f JB |
4747 | static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode, |
4748 | u64 offset, u64 len) | |
4749 | { | |
0b246afa | 4750 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
16e7549f JB |
4751 | struct btrfs_trans_handle *trans; |
4752 | int ret; | |
4753 | ||
4754 | /* | |
4755 | * Still need to make sure the inode looks like it's been updated so | |
4756 | * that any holes get logged if we fsync. | |
4757 | */ | |
0b246afa JM |
4758 | if (btrfs_fs_incompat(fs_info, NO_HOLES)) { |
4759 | BTRFS_I(inode)->last_trans = fs_info->generation; | |
16e7549f JB |
4760 | BTRFS_I(inode)->last_sub_trans = root->log_transid; |
4761 | BTRFS_I(inode)->last_log_commit = root->last_log_commit; | |
4762 | return 0; | |
4763 | } | |
4764 | ||
4765 | /* | |
4766 | * 1 - for the one we're dropping | |
4767 | * 1 - for the one we're adding | |
4768 | * 1 - for updating the inode. | |
4769 | */ | |
4770 | trans = btrfs_start_transaction(root, 3); | |
4771 | if (IS_ERR(trans)) | |
4772 | return PTR_ERR(trans); | |
4773 | ||
4774 | ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1); | |
4775 | if (ret) { | |
66642832 | 4776 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4777 | btrfs_end_transaction(trans); |
16e7549f JB |
4778 | return ret; |
4779 | } | |
4780 | ||
f85b7379 DS |
4781 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)), |
4782 | offset, 0, 0, len, 0, len, 0, 0, 0); | |
16e7549f | 4783 | if (ret) |
66642832 | 4784 | btrfs_abort_transaction(trans, ret); |
16e7549f JB |
4785 | else |
4786 | btrfs_update_inode(trans, root, inode); | |
3a45bb20 | 4787 | btrfs_end_transaction(trans); |
16e7549f JB |
4788 | return ret; |
4789 | } | |
4790 | ||
695a0d0d JB |
4791 | /* |
4792 | * This function puts in dummy file extents for the area we're creating a hole | |
4793 | * for. So if we are truncating this file to a larger size we need to insert | |
4794 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
4795 | * the range between oldsize and size | |
4796 | */ | |
a41ad394 | 4797 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 4798 | { |
0b246afa | 4799 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
9036c102 YZ |
4800 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4801 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 4802 | struct extent_map *em = NULL; |
2ac55d41 | 4803 | struct extent_state *cached_state = NULL; |
5dc562c5 | 4804 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
0b246afa JM |
4805 | u64 hole_start = ALIGN(oldsize, fs_info->sectorsize); |
4806 | u64 block_end = ALIGN(size, fs_info->sectorsize); | |
9036c102 YZ |
4807 | u64 last_byte; |
4808 | u64 cur_offset; | |
4809 | u64 hole_size; | |
9ed74f2d | 4810 | int err = 0; |
39279cc3 | 4811 | |
a71754fc | 4812 | /* |
9703fefe CR |
4813 | * If our size started in the middle of a block we need to zero out the |
4814 | * rest of the block before we expand the i_size, otherwise we could | |
a71754fc JB |
4815 | * expose stale data. |
4816 | */ | |
9703fefe | 4817 | err = btrfs_truncate_block(inode, oldsize, 0, 0); |
a71754fc JB |
4818 | if (err) |
4819 | return err; | |
4820 | ||
9036c102 YZ |
4821 | if (size <= hole_start) |
4822 | return 0; | |
4823 | ||
9036c102 YZ |
4824 | while (1) { |
4825 | struct btrfs_ordered_extent *ordered; | |
fa7c1494 | 4826 | |
ff13db41 | 4827 | lock_extent_bits(io_tree, hole_start, block_end - 1, |
d0082371 | 4828 | &cached_state); |
a776c6fa | 4829 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start, |
fa7c1494 | 4830 | block_end - hole_start); |
9036c102 YZ |
4831 | if (!ordered) |
4832 | break; | |
2ac55d41 JB |
4833 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
4834 | &cached_state, GFP_NOFS); | |
fa7c1494 | 4835 | btrfs_start_ordered_extent(inode, ordered, 1); |
9036c102 YZ |
4836 | btrfs_put_ordered_extent(ordered); |
4837 | } | |
39279cc3 | 4838 | |
9036c102 YZ |
4839 | cur_offset = hole_start; |
4840 | while (1) { | |
fc4f21b1 | 4841 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset, |
9036c102 | 4842 | block_end - cur_offset, 0); |
79787eaa JM |
4843 | if (IS_ERR(em)) { |
4844 | err = PTR_ERR(em); | |
f2767956 | 4845 | em = NULL; |
79787eaa JM |
4846 | break; |
4847 | } | |
9036c102 | 4848 | last_byte = min(extent_map_end(em), block_end); |
0b246afa | 4849 | last_byte = ALIGN(last_byte, fs_info->sectorsize); |
8082510e | 4850 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 4851 | struct extent_map *hole_em; |
9036c102 | 4852 | hole_size = last_byte - cur_offset; |
9ed74f2d | 4853 | |
16e7549f JB |
4854 | err = maybe_insert_hole(root, inode, cur_offset, |
4855 | hole_size); | |
4856 | if (err) | |
3893e33b | 4857 | break; |
dcdbc059 | 4858 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
5dc562c5 JB |
4859 | cur_offset + hole_size - 1, 0); |
4860 | hole_em = alloc_extent_map(); | |
4861 | if (!hole_em) { | |
4862 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
4863 | &BTRFS_I(inode)->runtime_flags); | |
4864 | goto next; | |
4865 | } | |
4866 | hole_em->start = cur_offset; | |
4867 | hole_em->len = hole_size; | |
4868 | hole_em->orig_start = cur_offset; | |
8082510e | 4869 | |
5dc562c5 JB |
4870 | hole_em->block_start = EXTENT_MAP_HOLE; |
4871 | hole_em->block_len = 0; | |
b4939680 | 4872 | hole_em->orig_block_len = 0; |
cc95bef6 | 4873 | hole_em->ram_bytes = hole_size; |
0b246afa | 4874 | hole_em->bdev = fs_info->fs_devices->latest_bdev; |
5dc562c5 | 4875 | hole_em->compress_type = BTRFS_COMPRESS_NONE; |
0b246afa | 4876 | hole_em->generation = fs_info->generation; |
8082510e | 4877 | |
5dc562c5 JB |
4878 | while (1) { |
4879 | write_lock(&em_tree->lock); | |
09a2a8f9 | 4880 | err = add_extent_mapping(em_tree, hole_em, 1); |
5dc562c5 JB |
4881 | write_unlock(&em_tree->lock); |
4882 | if (err != -EEXIST) | |
4883 | break; | |
dcdbc059 NB |
4884 | btrfs_drop_extent_cache(BTRFS_I(inode), |
4885 | cur_offset, | |
5dc562c5 JB |
4886 | cur_offset + |
4887 | hole_size - 1, 0); | |
4888 | } | |
4889 | free_extent_map(hole_em); | |
9036c102 | 4890 | } |
16e7549f | 4891 | next: |
9036c102 | 4892 | free_extent_map(em); |
a22285a6 | 4893 | em = NULL; |
9036c102 | 4894 | cur_offset = last_byte; |
8082510e | 4895 | if (cur_offset >= block_end) |
9036c102 YZ |
4896 | break; |
4897 | } | |
a22285a6 | 4898 | free_extent_map(em); |
2ac55d41 JB |
4899 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
4900 | GFP_NOFS); | |
9036c102 YZ |
4901 | return err; |
4902 | } | |
39279cc3 | 4903 | |
3972f260 | 4904 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 4905 | { |
f4a2f4c5 MX |
4906 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4907 | struct btrfs_trans_handle *trans; | |
a41ad394 | 4908 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
4909 | loff_t newsize = attr->ia_size; |
4910 | int mask = attr->ia_valid; | |
8082510e YZ |
4911 | int ret; |
4912 | ||
3972f260 ES |
4913 | /* |
4914 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
4915 | * special case where we need to update the times despite not having | |
4916 | * these flags set. For all other operations the VFS set these flags | |
4917 | * explicitly if it wants a timestamp update. | |
4918 | */ | |
dff6efc3 CH |
4919 | if (newsize != oldsize) { |
4920 | inode_inc_iversion(inode); | |
4921 | if (!(mask & (ATTR_CTIME | ATTR_MTIME))) | |
4922 | inode->i_ctime = inode->i_mtime = | |
c2050a45 | 4923 | current_time(inode); |
dff6efc3 | 4924 | } |
3972f260 | 4925 | |
a41ad394 | 4926 | if (newsize > oldsize) { |
9ea24bbe FM |
4927 | /* |
4928 | * Don't do an expanding truncate while snapshoting is ongoing. | |
4929 | * This is to ensure the snapshot captures a fully consistent | |
4930 | * state of this file - if the snapshot captures this expanding | |
4931 | * truncation, it must capture all writes that happened before | |
4932 | * this truncation. | |
4933 | */ | |
0bc19f90 | 4934 | btrfs_wait_for_snapshot_creation(root); |
a41ad394 | 4935 | ret = btrfs_cont_expand(inode, oldsize, newsize); |
9ea24bbe FM |
4936 | if (ret) { |
4937 | btrfs_end_write_no_snapshoting(root); | |
8082510e | 4938 | return ret; |
9ea24bbe | 4939 | } |
8082510e | 4940 | |
f4a2f4c5 | 4941 | trans = btrfs_start_transaction(root, 1); |
9ea24bbe FM |
4942 | if (IS_ERR(trans)) { |
4943 | btrfs_end_write_no_snapshoting(root); | |
f4a2f4c5 | 4944 | return PTR_ERR(trans); |
9ea24bbe | 4945 | } |
f4a2f4c5 MX |
4946 | |
4947 | i_size_write(inode, newsize); | |
4948 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
27772b68 | 4949 | pagecache_isize_extended(inode, oldsize, newsize); |
f4a2f4c5 | 4950 | ret = btrfs_update_inode(trans, root, inode); |
9ea24bbe | 4951 | btrfs_end_write_no_snapshoting(root); |
3a45bb20 | 4952 | btrfs_end_transaction(trans); |
a41ad394 | 4953 | } else { |
8082510e | 4954 | |
a41ad394 JB |
4955 | /* |
4956 | * We're truncating a file that used to have good data down to | |
4957 | * zero. Make sure it gets into the ordered flush list so that | |
4958 | * any new writes get down to disk quickly. | |
4959 | */ | |
4960 | if (newsize == 0) | |
72ac3c0d JB |
4961 | set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
4962 | &BTRFS_I(inode)->runtime_flags); | |
8082510e | 4963 | |
f3fe820c JB |
4964 | /* |
4965 | * 1 for the orphan item we're going to add | |
4966 | * 1 for the orphan item deletion. | |
4967 | */ | |
4968 | trans = btrfs_start_transaction(root, 2); | |
4969 | if (IS_ERR(trans)) | |
4970 | return PTR_ERR(trans); | |
4971 | ||
4972 | /* | |
4973 | * We need to do this in case we fail at _any_ point during the | |
4974 | * actual truncate. Once we do the truncate_setsize we could | |
4975 | * invalidate pages which forces any outstanding ordered io to | |
4976 | * be instantly completed which will give us extents that need | |
4977 | * to be truncated. If we fail to get an orphan inode down we | |
4978 | * could have left over extents that were never meant to live, | |
01327610 | 4979 | * so we need to guarantee from this point on that everything |
f3fe820c JB |
4980 | * will be consistent. |
4981 | */ | |
73f2e545 | 4982 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
3a45bb20 | 4983 | btrfs_end_transaction(trans); |
f3fe820c JB |
4984 | if (ret) |
4985 | return ret; | |
4986 | ||
a41ad394 JB |
4987 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
4988 | truncate_setsize(inode, newsize); | |
2e60a51e MX |
4989 | |
4990 | /* Disable nonlocked read DIO to avoid the end less truncate */ | |
abcefb1e | 4991 | btrfs_inode_block_unlocked_dio(BTRFS_I(inode)); |
2e60a51e | 4992 | inode_dio_wait(inode); |
0b581701 | 4993 | btrfs_inode_resume_unlocked_dio(BTRFS_I(inode)); |
2e60a51e | 4994 | |
a41ad394 | 4995 | ret = btrfs_truncate(inode); |
7f4f6e0a JB |
4996 | if (ret && inode->i_nlink) { |
4997 | int err; | |
4998 | ||
19fd2df5 LB |
4999 | /* To get a stable disk_i_size */ |
5000 | err = btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
5001 | if (err) { | |
3d6ae7bb | 5002 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
19fd2df5 LB |
5003 | return err; |
5004 | } | |
5005 | ||
7f4f6e0a JB |
5006 | /* |
5007 | * failed to truncate, disk_i_size is only adjusted down | |
5008 | * as we remove extents, so it should represent the true | |
5009 | * size of the inode, so reset the in memory size and | |
5010 | * delete our orphan entry. | |
5011 | */ | |
5012 | trans = btrfs_join_transaction(root); | |
5013 | if (IS_ERR(trans)) { | |
3d6ae7bb | 5014 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
7f4f6e0a JB |
5015 | return ret; |
5016 | } | |
5017 | i_size_write(inode, BTRFS_I(inode)->disk_i_size); | |
3d6ae7bb | 5018 | err = btrfs_orphan_del(trans, BTRFS_I(inode)); |
7f4f6e0a | 5019 | if (err) |
66642832 | 5020 | btrfs_abort_transaction(trans, err); |
3a45bb20 | 5021 | btrfs_end_transaction(trans); |
7f4f6e0a | 5022 | } |
8082510e YZ |
5023 | } |
5024 | ||
a41ad394 | 5025 | return ret; |
8082510e YZ |
5026 | } |
5027 | ||
9036c102 YZ |
5028 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
5029 | { | |
2b0143b5 | 5030 | struct inode *inode = d_inode(dentry); |
b83cc969 | 5031 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 5032 | int err; |
39279cc3 | 5033 | |
b83cc969 LZ |
5034 | if (btrfs_root_readonly(root)) |
5035 | return -EROFS; | |
5036 | ||
31051c85 | 5037 | err = setattr_prepare(dentry, attr); |
9036c102 YZ |
5038 | if (err) |
5039 | return err; | |
2bf5a725 | 5040 | |
5a3f23d5 | 5041 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 5042 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
5043 | if (err) |
5044 | return err; | |
39279cc3 | 5045 | } |
9036c102 | 5046 | |
1025774c CH |
5047 | if (attr->ia_valid) { |
5048 | setattr_copy(inode, attr); | |
0c4d2d95 | 5049 | inode_inc_iversion(inode); |
22c44fe6 | 5050 | err = btrfs_dirty_inode(inode); |
1025774c | 5051 | |
22c44fe6 | 5052 | if (!err && attr->ia_valid & ATTR_MODE) |
996a710d | 5053 | err = posix_acl_chmod(inode, inode->i_mode); |
1025774c | 5054 | } |
33268eaf | 5055 | |
39279cc3 CM |
5056 | return err; |
5057 | } | |
61295eb8 | 5058 | |
131e404a FDBM |
5059 | /* |
5060 | * While truncating the inode pages during eviction, we get the VFS calling | |
5061 | * btrfs_invalidatepage() against each page of the inode. This is slow because | |
5062 | * the calls to btrfs_invalidatepage() result in a huge amount of calls to | |
5063 | * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting | |
5064 | * extent_state structures over and over, wasting lots of time. | |
5065 | * | |
5066 | * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all | |
5067 | * those expensive operations on a per page basis and do only the ordered io | |
5068 | * finishing, while we release here the extent_map and extent_state structures, | |
5069 | * without the excessive merging and splitting. | |
5070 | */ | |
5071 | static void evict_inode_truncate_pages(struct inode *inode) | |
5072 | { | |
5073 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
5074 | struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree; | |
5075 | struct rb_node *node; | |
5076 | ||
5077 | ASSERT(inode->i_state & I_FREEING); | |
91b0abe3 | 5078 | truncate_inode_pages_final(&inode->i_data); |
131e404a FDBM |
5079 | |
5080 | write_lock(&map_tree->lock); | |
5081 | while (!RB_EMPTY_ROOT(&map_tree->map)) { | |
5082 | struct extent_map *em; | |
5083 | ||
5084 | node = rb_first(&map_tree->map); | |
5085 | em = rb_entry(node, struct extent_map, rb_node); | |
180589ef WS |
5086 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
5087 | clear_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
131e404a FDBM |
5088 | remove_extent_mapping(map_tree, em); |
5089 | free_extent_map(em); | |
7064dd5c FM |
5090 | if (need_resched()) { |
5091 | write_unlock(&map_tree->lock); | |
5092 | cond_resched(); | |
5093 | write_lock(&map_tree->lock); | |
5094 | } | |
131e404a FDBM |
5095 | } |
5096 | write_unlock(&map_tree->lock); | |
5097 | ||
6ca07097 FM |
5098 | /* |
5099 | * Keep looping until we have no more ranges in the io tree. | |
5100 | * We can have ongoing bios started by readpages (called from readahead) | |
9c6429d9 FM |
5101 | * that have their endio callback (extent_io.c:end_bio_extent_readpage) |
5102 | * still in progress (unlocked the pages in the bio but did not yet | |
5103 | * unlocked the ranges in the io tree). Therefore this means some | |
6ca07097 FM |
5104 | * ranges can still be locked and eviction started because before |
5105 | * submitting those bios, which are executed by a separate task (work | |
5106 | * queue kthread), inode references (inode->i_count) were not taken | |
5107 | * (which would be dropped in the end io callback of each bio). | |
5108 | * Therefore here we effectively end up waiting for those bios and | |
5109 | * anyone else holding locked ranges without having bumped the inode's | |
5110 | * reference count - if we don't do it, when they access the inode's | |
5111 | * io_tree to unlock a range it may be too late, leading to an | |
5112 | * use-after-free issue. | |
5113 | */ | |
131e404a FDBM |
5114 | spin_lock(&io_tree->lock); |
5115 | while (!RB_EMPTY_ROOT(&io_tree->state)) { | |
5116 | struct extent_state *state; | |
5117 | struct extent_state *cached_state = NULL; | |
6ca07097 FM |
5118 | u64 start; |
5119 | u64 end; | |
131e404a FDBM |
5120 | |
5121 | node = rb_first(&io_tree->state); | |
5122 | state = rb_entry(node, struct extent_state, rb_node); | |
6ca07097 FM |
5123 | start = state->start; |
5124 | end = state->end; | |
131e404a FDBM |
5125 | spin_unlock(&io_tree->lock); |
5126 | ||
ff13db41 | 5127 | lock_extent_bits(io_tree, start, end, &cached_state); |
b9d0b389 QW |
5128 | |
5129 | /* | |
5130 | * If still has DELALLOC flag, the extent didn't reach disk, | |
5131 | * and its reserved space won't be freed by delayed_ref. | |
5132 | * So we need to free its reserved space here. | |
5133 | * (Refer to comment in btrfs_invalidatepage, case 2) | |
5134 | * | |
5135 | * Note, end is the bytenr of last byte, so we need + 1 here. | |
5136 | */ | |
5137 | if (state->state & EXTENT_DELALLOC) | |
5138 | btrfs_qgroup_free_data(inode, start, end - start + 1); | |
5139 | ||
6ca07097 | 5140 | clear_extent_bit(io_tree, start, end, |
131e404a FDBM |
5141 | EXTENT_LOCKED | EXTENT_DIRTY | |
5142 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
5143 | EXTENT_DEFRAG, 1, 1, | |
5144 | &cached_state, GFP_NOFS); | |
131e404a | 5145 | |
7064dd5c | 5146 | cond_resched(); |
131e404a FDBM |
5147 | spin_lock(&io_tree->lock); |
5148 | } | |
5149 | spin_unlock(&io_tree->lock); | |
5150 | } | |
5151 | ||
bd555975 | 5152 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 | 5153 | { |
0b246afa | 5154 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
5155 | struct btrfs_trans_handle *trans; |
5156 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
726c35fa | 5157 | struct btrfs_block_rsv *rsv, *global_rsv; |
3bce876f | 5158 | int steal_from_global = 0; |
3d48d981 | 5159 | u64 min_size; |
39279cc3 CM |
5160 | int ret; |
5161 | ||
1abe9b8a | 5162 | trace_btrfs_inode_evict(inode); |
5163 | ||
3d48d981 NB |
5164 | if (!root) { |
5165 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); | |
5166 | return; | |
5167 | } | |
5168 | ||
0b246afa | 5169 | min_size = btrfs_calc_trunc_metadata_size(fs_info, 1); |
3d48d981 | 5170 | |
131e404a FDBM |
5171 | evict_inode_truncate_pages(inode); |
5172 | ||
69e9c6c6 SB |
5173 | if (inode->i_nlink && |
5174 | ((btrfs_root_refs(&root->root_item) != 0 && | |
5175 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) || | |
70ddc553 | 5176 | btrfs_is_free_space_inode(BTRFS_I(inode)))) |
bd555975 AV |
5177 | goto no_delete; |
5178 | ||
39279cc3 | 5179 | if (is_bad_inode(inode)) { |
3d6ae7bb | 5180 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
39279cc3 CM |
5181 | goto no_delete; |
5182 | } | |
bd555975 | 5183 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
a30e577c JM |
5184 | if (!special_file(inode->i_mode)) |
5185 | btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
5f39d397 | 5186 | |
7ab7956e | 5187 | btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1); |
f612496b | 5188 | |
0b246afa | 5189 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) { |
6bf02314 | 5190 | BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
8a35d95f | 5191 | &BTRFS_I(inode)->runtime_flags)); |
c71bf099 YZ |
5192 | goto no_delete; |
5193 | } | |
5194 | ||
76dda93c | 5195 | if (inode->i_nlink > 0) { |
69e9c6c6 SB |
5196 | BUG_ON(btrfs_root_refs(&root->root_item) != 0 && |
5197 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID); | |
76dda93c YZ |
5198 | goto no_delete; |
5199 | } | |
5200 | ||
aa79021f | 5201 | ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode)); |
0e8c36a9 | 5202 | if (ret) { |
3d6ae7bb | 5203 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
0e8c36a9 MX |
5204 | goto no_delete; |
5205 | } | |
5206 | ||
2ff7e61e | 5207 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
4289a667 | 5208 | if (!rsv) { |
3d6ae7bb | 5209 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
4289a667 JB |
5210 | goto no_delete; |
5211 | } | |
4a338542 | 5212 | rsv->size = min_size; |
ca7e70f5 | 5213 | rsv->failfast = 1; |
0b246afa | 5214 | global_rsv = &fs_info->global_block_rsv; |
4289a667 | 5215 | |
6ef06d27 | 5216 | btrfs_i_size_write(BTRFS_I(inode), 0); |
5f39d397 | 5217 | |
4289a667 | 5218 | /* |
8407aa46 MX |
5219 | * This is a bit simpler than btrfs_truncate since we've already |
5220 | * reserved our space for our orphan item in the unlink, so we just | |
5221 | * need to reserve some slack space in case we add bytes and update | |
5222 | * inode item when doing the truncate. | |
4289a667 | 5223 | */ |
8082510e | 5224 | while (1) { |
08e007d2 MX |
5225 | ret = btrfs_block_rsv_refill(root, rsv, min_size, |
5226 | BTRFS_RESERVE_FLUSH_LIMIT); | |
726c35fa JB |
5227 | |
5228 | /* | |
5229 | * Try and steal from the global reserve since we will | |
5230 | * likely not use this space anyway, we want to try as | |
5231 | * hard as possible to get this to work. | |
5232 | */ | |
5233 | if (ret) | |
3bce876f JB |
5234 | steal_from_global++; |
5235 | else | |
5236 | steal_from_global = 0; | |
5237 | ret = 0; | |
d68fc57b | 5238 | |
3bce876f JB |
5239 | /* |
5240 | * steal_from_global == 0: we reserved stuff, hooray! | |
5241 | * steal_from_global == 1: we didn't reserve stuff, boo! | |
5242 | * steal_from_global == 2: we've committed, still not a lot of | |
5243 | * room but maybe we'll have room in the global reserve this | |
5244 | * time. | |
5245 | * steal_from_global == 3: abandon all hope! | |
5246 | */ | |
5247 | if (steal_from_global > 2) { | |
0b246afa JM |
5248 | btrfs_warn(fs_info, |
5249 | "Could not get space for a delete, will truncate on mount %d", | |
5250 | ret); | |
3d6ae7bb | 5251 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
2ff7e61e | 5252 | btrfs_free_block_rsv(fs_info, rsv); |
4289a667 | 5253 | goto no_delete; |
d68fc57b | 5254 | } |
7b128766 | 5255 | |
0e8c36a9 | 5256 | trans = btrfs_join_transaction(root); |
4289a667 | 5257 | if (IS_ERR(trans)) { |
3d6ae7bb | 5258 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
2ff7e61e | 5259 | btrfs_free_block_rsv(fs_info, rsv); |
4289a667 | 5260 | goto no_delete; |
d68fc57b | 5261 | } |
7b128766 | 5262 | |
3bce876f | 5263 | /* |
01327610 | 5264 | * We can't just steal from the global reserve, we need to make |
3bce876f JB |
5265 | * sure there is room to do it, if not we need to commit and try |
5266 | * again. | |
5267 | */ | |
5268 | if (steal_from_global) { | |
2ff7e61e | 5269 | if (!btrfs_check_space_for_delayed_refs(trans, fs_info)) |
3bce876f | 5270 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, |
25d609f8 | 5271 | min_size, 0); |
3bce876f JB |
5272 | else |
5273 | ret = -ENOSPC; | |
5274 | } | |
5275 | ||
5276 | /* | |
5277 | * Couldn't steal from the global reserve, we have too much | |
5278 | * pending stuff built up, commit the transaction and try it | |
5279 | * again. | |
5280 | */ | |
5281 | if (ret) { | |
3a45bb20 | 5282 | ret = btrfs_commit_transaction(trans); |
3bce876f | 5283 | if (ret) { |
3d6ae7bb | 5284 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
2ff7e61e | 5285 | btrfs_free_block_rsv(fs_info, rsv); |
3bce876f JB |
5286 | goto no_delete; |
5287 | } | |
5288 | continue; | |
5289 | } else { | |
5290 | steal_from_global = 0; | |
5291 | } | |
5292 | ||
4289a667 JB |
5293 | trans->block_rsv = rsv; |
5294 | ||
d68fc57b | 5295 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
28ed1345 | 5296 | if (ret != -ENOSPC && ret != -EAGAIN) |
8082510e | 5297 | break; |
85e21bac | 5298 | |
0b246afa | 5299 | trans->block_rsv = &fs_info->trans_block_rsv; |
3a45bb20 | 5300 | btrfs_end_transaction(trans); |
8082510e | 5301 | trans = NULL; |
2ff7e61e | 5302 | btrfs_btree_balance_dirty(fs_info); |
8082510e | 5303 | } |
5f39d397 | 5304 | |
2ff7e61e | 5305 | btrfs_free_block_rsv(fs_info, rsv); |
4289a667 | 5306 | |
4ef31a45 JB |
5307 | /* |
5308 | * Errors here aren't a big deal, it just means we leave orphan items | |
5309 | * in the tree. They will be cleaned up on the next mount. | |
5310 | */ | |
8082510e | 5311 | if (ret == 0) { |
4289a667 | 5312 | trans->block_rsv = root->orphan_block_rsv; |
3d6ae7bb | 5313 | btrfs_orphan_del(trans, BTRFS_I(inode)); |
4ef31a45 | 5314 | } else { |
3d6ae7bb | 5315 | btrfs_orphan_del(NULL, BTRFS_I(inode)); |
8082510e | 5316 | } |
54aa1f4d | 5317 | |
0b246afa JM |
5318 | trans->block_rsv = &fs_info->trans_block_rsv; |
5319 | if (!(root == fs_info->tree_root || | |
581bb050 | 5320 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) |
4a0cc7ca | 5321 | btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode))); |
581bb050 | 5322 | |
3a45bb20 | 5323 | btrfs_end_transaction(trans); |
2ff7e61e | 5324 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 | 5325 | no_delete: |
f48d1cf5 | 5326 | btrfs_remove_delayed_node(BTRFS_I(inode)); |
dbd5768f | 5327 | clear_inode(inode); |
39279cc3 CM |
5328 | } |
5329 | ||
5330 | /* | |
5331 | * this returns the key found in the dir entry in the location pointer. | |
5332 | * If no dir entries were found, location->objectid is 0. | |
5333 | */ | |
5334 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
5335 | struct btrfs_key *location) | |
5336 | { | |
5337 | const char *name = dentry->d_name.name; | |
5338 | int namelen = dentry->d_name.len; | |
5339 | struct btrfs_dir_item *di; | |
5340 | struct btrfs_path *path; | |
5341 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 5342 | int ret = 0; |
39279cc3 CM |
5343 | |
5344 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
5345 | if (!path) |
5346 | return -ENOMEM; | |
3954401f | 5347 | |
f85b7379 DS |
5348 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)), |
5349 | name, namelen, 0); | |
0d9f7f3e Y |
5350 | if (IS_ERR(di)) |
5351 | ret = PTR_ERR(di); | |
d397712b | 5352 | |
c704005d | 5353 | if (IS_ERR_OR_NULL(di)) |
3954401f | 5354 | goto out_err; |
d397712b | 5355 | |
5f39d397 | 5356 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 5357 | out: |
39279cc3 CM |
5358 | btrfs_free_path(path); |
5359 | return ret; | |
3954401f CM |
5360 | out_err: |
5361 | location->objectid = 0; | |
5362 | goto out; | |
39279cc3 CM |
5363 | } |
5364 | ||
5365 | /* | |
5366 | * when we hit a tree root in a directory, the btrfs part of the inode | |
5367 | * needs to be changed to reflect the root directory of the tree root. This | |
5368 | * is kind of like crossing a mount point. | |
5369 | */ | |
2ff7e61e | 5370 | static int fixup_tree_root_location(struct btrfs_fs_info *fs_info, |
4df27c4d YZ |
5371 | struct inode *dir, |
5372 | struct dentry *dentry, | |
5373 | struct btrfs_key *location, | |
5374 | struct btrfs_root **sub_root) | |
39279cc3 | 5375 | { |
4df27c4d YZ |
5376 | struct btrfs_path *path; |
5377 | struct btrfs_root *new_root; | |
5378 | struct btrfs_root_ref *ref; | |
5379 | struct extent_buffer *leaf; | |
1d4c08e0 | 5380 | struct btrfs_key key; |
4df27c4d YZ |
5381 | int ret; |
5382 | int err = 0; | |
39279cc3 | 5383 | |
4df27c4d YZ |
5384 | path = btrfs_alloc_path(); |
5385 | if (!path) { | |
5386 | err = -ENOMEM; | |
5387 | goto out; | |
5388 | } | |
39279cc3 | 5389 | |
4df27c4d | 5390 | err = -ENOENT; |
1d4c08e0 DS |
5391 | key.objectid = BTRFS_I(dir)->root->root_key.objectid; |
5392 | key.type = BTRFS_ROOT_REF_KEY; | |
5393 | key.offset = location->objectid; | |
5394 | ||
0b246afa | 5395 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
4df27c4d YZ |
5396 | if (ret) { |
5397 | if (ret < 0) | |
5398 | err = ret; | |
5399 | goto out; | |
5400 | } | |
39279cc3 | 5401 | |
4df27c4d YZ |
5402 | leaf = path->nodes[0]; |
5403 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
4a0cc7ca | 5404 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) || |
4df27c4d YZ |
5405 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
5406 | goto out; | |
39279cc3 | 5407 | |
4df27c4d YZ |
5408 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
5409 | (unsigned long)(ref + 1), | |
5410 | dentry->d_name.len); | |
5411 | if (ret) | |
5412 | goto out; | |
5413 | ||
b3b4aa74 | 5414 | btrfs_release_path(path); |
4df27c4d | 5415 | |
0b246afa | 5416 | new_root = btrfs_read_fs_root_no_name(fs_info, location); |
4df27c4d YZ |
5417 | if (IS_ERR(new_root)) { |
5418 | err = PTR_ERR(new_root); | |
5419 | goto out; | |
5420 | } | |
5421 | ||
4df27c4d YZ |
5422 | *sub_root = new_root; |
5423 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
5424 | location->type = BTRFS_INODE_ITEM_KEY; | |
5425 | location->offset = 0; | |
5426 | err = 0; | |
5427 | out: | |
5428 | btrfs_free_path(path); | |
5429 | return err; | |
39279cc3 CM |
5430 | } |
5431 | ||
5d4f98a2 YZ |
5432 | static void inode_tree_add(struct inode *inode) |
5433 | { | |
5434 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5435 | struct btrfs_inode *entry; | |
03e860bd FNP |
5436 | struct rb_node **p; |
5437 | struct rb_node *parent; | |
cef21937 | 5438 | struct rb_node *new = &BTRFS_I(inode)->rb_node; |
4a0cc7ca | 5439 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
5d4f98a2 | 5440 | |
1d3382cb | 5441 | if (inode_unhashed(inode)) |
76dda93c | 5442 | return; |
e1409cef | 5443 | parent = NULL; |
5d4f98a2 | 5444 | spin_lock(&root->inode_lock); |
e1409cef | 5445 | p = &root->inode_tree.rb_node; |
5d4f98a2 YZ |
5446 | while (*p) { |
5447 | parent = *p; | |
5448 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
5449 | ||
4a0cc7ca | 5450 | if (ino < btrfs_ino(BTRFS_I(&entry->vfs_inode))) |
03e860bd | 5451 | p = &parent->rb_left; |
4a0cc7ca | 5452 | else if (ino > btrfs_ino(BTRFS_I(&entry->vfs_inode))) |
03e860bd | 5453 | p = &parent->rb_right; |
5d4f98a2 YZ |
5454 | else { |
5455 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 5456 | (I_WILL_FREE | I_FREEING))); |
cef21937 | 5457 | rb_replace_node(parent, new, &root->inode_tree); |
03e860bd FNP |
5458 | RB_CLEAR_NODE(parent); |
5459 | spin_unlock(&root->inode_lock); | |
cef21937 | 5460 | return; |
5d4f98a2 YZ |
5461 | } |
5462 | } | |
cef21937 FDBM |
5463 | rb_link_node(new, parent, p); |
5464 | rb_insert_color(new, &root->inode_tree); | |
5d4f98a2 YZ |
5465 | spin_unlock(&root->inode_lock); |
5466 | } | |
5467 | ||
5468 | static void inode_tree_del(struct inode *inode) | |
5469 | { | |
0b246afa | 5470 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5d4f98a2 | 5471 | struct btrfs_root *root = BTRFS_I(inode)->root; |
76dda93c | 5472 | int empty = 0; |
5d4f98a2 | 5473 | |
03e860bd | 5474 | spin_lock(&root->inode_lock); |
5d4f98a2 | 5475 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 5476 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 5477 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 5478 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 5479 | } |
03e860bd | 5480 | spin_unlock(&root->inode_lock); |
76dda93c | 5481 | |
69e9c6c6 | 5482 | if (empty && btrfs_root_refs(&root->root_item) == 0) { |
0b246afa | 5483 | synchronize_srcu(&fs_info->subvol_srcu); |
76dda93c YZ |
5484 | spin_lock(&root->inode_lock); |
5485 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
5486 | spin_unlock(&root->inode_lock); | |
5487 | if (empty) | |
5488 | btrfs_add_dead_root(root); | |
5489 | } | |
5490 | } | |
5491 | ||
143bede5 | 5492 | void btrfs_invalidate_inodes(struct btrfs_root *root) |
76dda93c | 5493 | { |
0b246afa | 5494 | struct btrfs_fs_info *fs_info = root->fs_info; |
76dda93c YZ |
5495 | struct rb_node *node; |
5496 | struct rb_node *prev; | |
5497 | struct btrfs_inode *entry; | |
5498 | struct inode *inode; | |
5499 | u64 objectid = 0; | |
5500 | ||
0b246afa | 5501 | if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
7813b3db | 5502 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); |
76dda93c YZ |
5503 | |
5504 | spin_lock(&root->inode_lock); | |
5505 | again: | |
5506 | node = root->inode_tree.rb_node; | |
5507 | prev = NULL; | |
5508 | while (node) { | |
5509 | prev = node; | |
5510 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
5511 | ||
4a0cc7ca | 5512 | if (objectid < btrfs_ino(BTRFS_I(&entry->vfs_inode))) |
76dda93c | 5513 | node = node->rb_left; |
4a0cc7ca | 5514 | else if (objectid > btrfs_ino(BTRFS_I(&entry->vfs_inode))) |
76dda93c YZ |
5515 | node = node->rb_right; |
5516 | else | |
5517 | break; | |
5518 | } | |
5519 | if (!node) { | |
5520 | while (prev) { | |
5521 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
4a0cc7ca | 5522 | if (objectid <= btrfs_ino(BTRFS_I(&entry->vfs_inode))) { |
76dda93c YZ |
5523 | node = prev; |
5524 | break; | |
5525 | } | |
5526 | prev = rb_next(prev); | |
5527 | } | |
5528 | } | |
5529 | while (node) { | |
5530 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4a0cc7ca | 5531 | objectid = btrfs_ino(BTRFS_I(&entry->vfs_inode)) + 1; |
76dda93c YZ |
5532 | inode = igrab(&entry->vfs_inode); |
5533 | if (inode) { | |
5534 | spin_unlock(&root->inode_lock); | |
5535 | if (atomic_read(&inode->i_count) > 1) | |
5536 | d_prune_aliases(inode); | |
5537 | /* | |
45321ac5 | 5538 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
5539 | * the inode cache when its usage count |
5540 | * hits zero. | |
5541 | */ | |
5542 | iput(inode); | |
5543 | cond_resched(); | |
5544 | spin_lock(&root->inode_lock); | |
5545 | goto again; | |
5546 | } | |
5547 | ||
5548 | if (cond_resched_lock(&root->inode_lock)) | |
5549 | goto again; | |
5550 | ||
5551 | node = rb_next(node); | |
5552 | } | |
5553 | spin_unlock(&root->inode_lock); | |
5d4f98a2 YZ |
5554 | } |
5555 | ||
e02119d5 CM |
5556 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
5557 | { | |
5558 | struct btrfs_iget_args *args = p; | |
90d3e592 CM |
5559 | inode->i_ino = args->location->objectid; |
5560 | memcpy(&BTRFS_I(inode)->location, args->location, | |
5561 | sizeof(*args->location)); | |
e02119d5 | 5562 | BTRFS_I(inode)->root = args->root; |
39279cc3 CM |
5563 | return 0; |
5564 | } | |
5565 | ||
5566 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
5567 | { | |
5568 | struct btrfs_iget_args *args = opaque; | |
90d3e592 | 5569 | return args->location->objectid == BTRFS_I(inode)->location.objectid && |
d397712b | 5570 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
5571 | } |
5572 | ||
5d4f98a2 | 5573 | static struct inode *btrfs_iget_locked(struct super_block *s, |
90d3e592 | 5574 | struct btrfs_key *location, |
5d4f98a2 | 5575 | struct btrfs_root *root) |
39279cc3 CM |
5576 | { |
5577 | struct inode *inode; | |
5578 | struct btrfs_iget_args args; | |
90d3e592 | 5579 | unsigned long hashval = btrfs_inode_hash(location->objectid, root); |
778ba82b | 5580 | |
90d3e592 | 5581 | args.location = location; |
39279cc3 CM |
5582 | args.root = root; |
5583 | ||
778ba82b | 5584 | inode = iget5_locked(s, hashval, btrfs_find_actor, |
39279cc3 CM |
5585 | btrfs_init_locked_inode, |
5586 | (void *)&args); | |
5587 | return inode; | |
5588 | } | |
5589 | ||
1a54ef8c BR |
5590 | /* Get an inode object given its location and corresponding root. |
5591 | * Returns in *is_new if the inode was read from disk | |
5592 | */ | |
5593 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 5594 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
5595 | { |
5596 | struct inode *inode; | |
5597 | ||
90d3e592 | 5598 | inode = btrfs_iget_locked(s, location, root); |
1a54ef8c | 5599 | if (!inode) |
5d4f98a2 | 5600 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
5601 | |
5602 | if (inode->i_state & I_NEW) { | |
67710892 FM |
5603 | int ret; |
5604 | ||
5605 | ret = btrfs_read_locked_inode(inode); | |
1748f843 MF |
5606 | if (!is_bad_inode(inode)) { |
5607 | inode_tree_add(inode); | |
5608 | unlock_new_inode(inode); | |
5609 | if (new) | |
5610 | *new = 1; | |
5611 | } else { | |
e0b6d65b ST |
5612 | unlock_new_inode(inode); |
5613 | iput(inode); | |
67710892 FM |
5614 | ASSERT(ret < 0); |
5615 | inode = ERR_PTR(ret < 0 ? ret : -ESTALE); | |
1748f843 MF |
5616 | } |
5617 | } | |
5618 | ||
1a54ef8c BR |
5619 | return inode; |
5620 | } | |
5621 | ||
4df27c4d YZ |
5622 | static struct inode *new_simple_dir(struct super_block *s, |
5623 | struct btrfs_key *key, | |
5624 | struct btrfs_root *root) | |
5625 | { | |
5626 | struct inode *inode = new_inode(s); | |
5627 | ||
5628 | if (!inode) | |
5629 | return ERR_PTR(-ENOMEM); | |
5630 | ||
4df27c4d YZ |
5631 | BTRFS_I(inode)->root = root; |
5632 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
72ac3c0d | 5633 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
5634 | |
5635 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
848cce0d | 5636 | inode->i_op = &btrfs_dir_ro_inode_operations; |
1fdf4194 | 5637 | inode->i_opflags &= ~IOP_XATTR; |
4df27c4d YZ |
5638 | inode->i_fop = &simple_dir_operations; |
5639 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
c2050a45 | 5640 | inode->i_mtime = current_time(inode); |
9cc97d64 | 5641 | inode->i_atime = inode->i_mtime; |
5642 | inode->i_ctime = inode->i_mtime; | |
5643 | BTRFS_I(inode)->i_otime = inode->i_mtime; | |
4df27c4d YZ |
5644 | |
5645 | return inode; | |
5646 | } | |
5647 | ||
3de4586c | 5648 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 5649 | { |
0b246afa | 5650 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
d397712b | 5651 | struct inode *inode; |
4df27c4d | 5652 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
5653 | struct btrfs_root *sub_root = root; |
5654 | struct btrfs_key location; | |
76dda93c | 5655 | int index; |
b4aff1f8 | 5656 | int ret = 0; |
39279cc3 CM |
5657 | |
5658 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
5659 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 5660 | |
39e3c955 | 5661 | ret = btrfs_inode_by_name(dir, dentry, &location); |
39279cc3 CM |
5662 | if (ret < 0) |
5663 | return ERR_PTR(ret); | |
5f39d397 | 5664 | |
4df27c4d | 5665 | if (location.objectid == 0) |
5662344b | 5666 | return ERR_PTR(-ENOENT); |
4df27c4d YZ |
5667 | |
5668 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 5669 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
5670 | return inode; |
5671 | } | |
5672 | ||
5673 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
5674 | ||
0b246afa | 5675 | index = srcu_read_lock(&fs_info->subvol_srcu); |
2ff7e61e | 5676 | ret = fixup_tree_root_location(fs_info, dir, dentry, |
4df27c4d YZ |
5677 | &location, &sub_root); |
5678 | if (ret < 0) { | |
5679 | if (ret != -ENOENT) | |
5680 | inode = ERR_PTR(ret); | |
5681 | else | |
5682 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
5683 | } else { | |
73f73415 | 5684 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 5685 | } |
0b246afa | 5686 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
76dda93c | 5687 | |
34d19bad | 5688 | if (!IS_ERR(inode) && root != sub_root) { |
0b246afa | 5689 | down_read(&fs_info->cleanup_work_sem); |
c71bf099 | 5690 | if (!(inode->i_sb->s_flags & MS_RDONLY)) |
66b4ffd1 | 5691 | ret = btrfs_orphan_cleanup(sub_root); |
0b246afa | 5692 | up_read(&fs_info->cleanup_work_sem); |
01cd3367 JB |
5693 | if (ret) { |
5694 | iput(inode); | |
66b4ffd1 | 5695 | inode = ERR_PTR(ret); |
01cd3367 | 5696 | } |
c71bf099 YZ |
5697 | } |
5698 | ||
3de4586c CM |
5699 | return inode; |
5700 | } | |
5701 | ||
fe15ce44 | 5702 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
5703 | { |
5704 | struct btrfs_root *root; | |
2b0143b5 | 5705 | struct inode *inode = d_inode(dentry); |
76dda93c | 5706 | |
848cce0d | 5707 | if (!inode && !IS_ROOT(dentry)) |
2b0143b5 | 5708 | inode = d_inode(dentry->d_parent); |
76dda93c | 5709 | |
848cce0d LZ |
5710 | if (inode) { |
5711 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
5712 | if (btrfs_root_refs(&root->root_item) == 0) |
5713 | return 1; | |
848cce0d | 5714 | |
4a0cc7ca | 5715 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
848cce0d | 5716 | return 1; |
efefb143 | 5717 | } |
76dda93c YZ |
5718 | return 0; |
5719 | } | |
5720 | ||
b4aff1f8 JB |
5721 | static void btrfs_dentry_release(struct dentry *dentry) |
5722 | { | |
944a4515 | 5723 | kfree(dentry->d_fsdata); |
b4aff1f8 JB |
5724 | } |
5725 | ||
3de4586c | 5726 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 5727 | unsigned int flags) |
3de4586c | 5728 | { |
5662344b | 5729 | struct inode *inode; |
a66e7cc6 | 5730 | |
5662344b TI |
5731 | inode = btrfs_lookup_dentry(dir, dentry); |
5732 | if (IS_ERR(inode)) { | |
5733 | if (PTR_ERR(inode) == -ENOENT) | |
5734 | inode = NULL; | |
5735 | else | |
5736 | return ERR_CAST(inode); | |
5737 | } | |
5738 | ||
41d28bca | 5739 | return d_splice_alias(inode, dentry); |
39279cc3 CM |
5740 | } |
5741 | ||
16cdcec7 | 5742 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
5743 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
5744 | }; | |
5745 | ||
9cdda8d3 | 5746 | static int btrfs_real_readdir(struct file *file, struct dir_context *ctx) |
39279cc3 | 5747 | { |
9cdda8d3 | 5748 | struct inode *inode = file_inode(file); |
2ff7e61e | 5749 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
5750 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5751 | struct btrfs_item *item; | |
5752 | struct btrfs_dir_item *di; | |
5753 | struct btrfs_key key; | |
5f39d397 | 5754 | struct btrfs_key found_key; |
39279cc3 | 5755 | struct btrfs_path *path; |
16cdcec7 MX |
5756 | struct list_head ins_list; |
5757 | struct list_head del_list; | |
39279cc3 | 5758 | int ret; |
5f39d397 | 5759 | struct extent_buffer *leaf; |
39279cc3 | 5760 | int slot; |
39279cc3 CM |
5761 | unsigned char d_type; |
5762 | int over = 0; | |
5f39d397 CM |
5763 | char tmp_name[32]; |
5764 | char *name_ptr; | |
5765 | int name_len; | |
02dbfc99 | 5766 | bool put = false; |
c2951f32 | 5767 | struct btrfs_key location; |
5f39d397 | 5768 | |
9cdda8d3 AV |
5769 | if (!dir_emit_dots(file, ctx)) |
5770 | return 0; | |
5771 | ||
49593bfa | 5772 | path = btrfs_alloc_path(); |
16cdcec7 MX |
5773 | if (!path) |
5774 | return -ENOMEM; | |
ff5714cc | 5775 | |
e4058b54 | 5776 | path->reada = READA_FORWARD; |
49593bfa | 5777 | |
c2951f32 JM |
5778 | INIT_LIST_HEAD(&ins_list); |
5779 | INIT_LIST_HEAD(&del_list); | |
5780 | put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list); | |
16cdcec7 | 5781 | |
c2951f32 | 5782 | key.type = BTRFS_DIR_INDEX_KEY; |
9cdda8d3 | 5783 | key.offset = ctx->pos; |
4a0cc7ca | 5784 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
5f39d397 | 5785 | |
39279cc3 CM |
5786 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5787 | if (ret < 0) | |
5788 | goto err; | |
49593bfa DW |
5789 | |
5790 | while (1) { | |
5f39d397 | 5791 | leaf = path->nodes[0]; |
39279cc3 | 5792 | slot = path->slots[0]; |
b9e03af0 LZ |
5793 | if (slot >= btrfs_header_nritems(leaf)) { |
5794 | ret = btrfs_next_leaf(root, path); | |
5795 | if (ret < 0) | |
5796 | goto err; | |
5797 | else if (ret > 0) | |
5798 | break; | |
5799 | continue; | |
39279cc3 | 5800 | } |
3de4586c | 5801 | |
dd3cc16b | 5802 | item = btrfs_item_nr(slot); |
5f39d397 CM |
5803 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
5804 | ||
5805 | if (found_key.objectid != key.objectid) | |
39279cc3 | 5806 | break; |
c2951f32 | 5807 | if (found_key.type != BTRFS_DIR_INDEX_KEY) |
39279cc3 | 5808 | break; |
9cdda8d3 | 5809 | if (found_key.offset < ctx->pos) |
b9e03af0 | 5810 | goto next; |
c2951f32 | 5811 | if (btrfs_should_delete_dir_index(&del_list, found_key.offset)) |
16cdcec7 | 5812 | goto next; |
5f39d397 | 5813 | |
9cdda8d3 | 5814 | ctx->pos = found_key.offset; |
49593bfa | 5815 | |
39279cc3 | 5816 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
2ff7e61e | 5817 | if (verify_dir_item(fs_info, leaf, di)) |
c2951f32 | 5818 | goto next; |
22a94d44 | 5819 | |
c2951f32 JM |
5820 | name_len = btrfs_dir_name_len(leaf, di); |
5821 | if (name_len <= sizeof(tmp_name)) { | |
5822 | name_ptr = tmp_name; | |
5823 | } else { | |
5824 | name_ptr = kmalloc(name_len, GFP_KERNEL); | |
5825 | if (!name_ptr) { | |
5826 | ret = -ENOMEM; | |
5827 | goto err; | |
5f39d397 | 5828 | } |
c2951f32 JM |
5829 | } |
5830 | read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1), | |
5831 | name_len); | |
3de4586c | 5832 | |
c2951f32 JM |
5833 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; |
5834 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
fede766f | 5835 | |
c2951f32 JM |
5836 | over = !dir_emit(ctx, name_ptr, name_len, location.objectid, |
5837 | d_type); | |
5f39d397 | 5838 | |
c2951f32 JM |
5839 | if (name_ptr != tmp_name) |
5840 | kfree(name_ptr); | |
5f39d397 | 5841 | |
c2951f32 JM |
5842 | if (over) |
5843 | goto nopos; | |
d2fbb2b5 | 5844 | ctx->pos++; |
b9e03af0 LZ |
5845 | next: |
5846 | path->slots[0]++; | |
39279cc3 | 5847 | } |
49593bfa | 5848 | |
d2fbb2b5 | 5849 | ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list); |
c2951f32 | 5850 | if (ret) |
bc4ef759 DS |
5851 | goto nopos; |
5852 | ||
db62efbb ZB |
5853 | /* |
5854 | * Stop new entries from being returned after we return the last | |
5855 | * entry. | |
5856 | * | |
5857 | * New directory entries are assigned a strictly increasing | |
5858 | * offset. This means that new entries created during readdir | |
5859 | * are *guaranteed* to be seen in the future by that readdir. | |
5860 | * This has broken buggy programs which operate on names as | |
5861 | * they're returned by readdir. Until we re-use freed offsets | |
5862 | * we have this hack to stop new entries from being returned | |
5863 | * under the assumption that they'll never reach this huge | |
5864 | * offset. | |
5865 | * | |
5866 | * This is being careful not to overflow 32bit loff_t unless the | |
5867 | * last entry requires it because doing so has broken 32bit apps | |
5868 | * in the past. | |
5869 | */ | |
c2951f32 JM |
5870 | if (ctx->pos >= INT_MAX) |
5871 | ctx->pos = LLONG_MAX; | |
5872 | else | |
5873 | ctx->pos = INT_MAX; | |
39279cc3 CM |
5874 | nopos: |
5875 | ret = 0; | |
5876 | err: | |
02dbfc99 OS |
5877 | if (put) |
5878 | btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list); | |
39279cc3 | 5879 | btrfs_free_path(path); |
39279cc3 CM |
5880 | return ret; |
5881 | } | |
5882 | ||
a9185b41 | 5883 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
5884 | { |
5885 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5886 | struct btrfs_trans_handle *trans; | |
5887 | int ret = 0; | |
0af3d00b | 5888 | bool nolock = false; |
39279cc3 | 5889 | |
72ac3c0d | 5890 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
4ca8b41e CM |
5891 | return 0; |
5892 | ||
70ddc553 NB |
5893 | if (btrfs_fs_closing(root->fs_info) && |
5894 | btrfs_is_free_space_inode(BTRFS_I(inode))) | |
82d5902d | 5895 | nolock = true; |
0af3d00b | 5896 | |
a9185b41 | 5897 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b | 5898 | if (nolock) |
7a7eaa40 | 5899 | trans = btrfs_join_transaction_nolock(root); |
0af3d00b | 5900 | else |
7a7eaa40 | 5901 | trans = btrfs_join_transaction(root); |
3612b495 TI |
5902 | if (IS_ERR(trans)) |
5903 | return PTR_ERR(trans); | |
3a45bb20 | 5904 | ret = btrfs_commit_transaction(trans); |
39279cc3 CM |
5905 | } |
5906 | return ret; | |
5907 | } | |
5908 | ||
5909 | /* | |
54aa1f4d | 5910 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
5911 | * inode changes. But, it is most likely to find the inode in cache. |
5912 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
5913 | * to keep or drop this code. | |
5914 | */ | |
48a3b636 | 5915 | static int btrfs_dirty_inode(struct inode *inode) |
39279cc3 | 5916 | { |
2ff7e61e | 5917 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
5918 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5919 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
5920 | int ret; |
5921 | ||
72ac3c0d | 5922 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 5923 | return 0; |
39279cc3 | 5924 | |
7a7eaa40 | 5925 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
5926 | if (IS_ERR(trans)) |
5927 | return PTR_ERR(trans); | |
8929ecfa YZ |
5928 | |
5929 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
5930 | if (ret && ret == -ENOSPC) { |
5931 | /* whoops, lets try again with the full transaction */ | |
3a45bb20 | 5932 | btrfs_end_transaction(trans); |
94b60442 | 5933 | trans = btrfs_start_transaction(root, 1); |
22c44fe6 JB |
5934 | if (IS_ERR(trans)) |
5935 | return PTR_ERR(trans); | |
8929ecfa | 5936 | |
94b60442 | 5937 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 5938 | } |
3a45bb20 | 5939 | btrfs_end_transaction(trans); |
16cdcec7 | 5940 | if (BTRFS_I(inode)->delayed_node) |
2ff7e61e | 5941 | btrfs_balance_delayed_items(fs_info); |
22c44fe6 JB |
5942 | |
5943 | return ret; | |
5944 | } | |
5945 | ||
5946 | /* | |
5947 | * This is a copy of file_update_time. We need this so we can return error on | |
5948 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
5949 | */ | |
e41f941a JB |
5950 | static int btrfs_update_time(struct inode *inode, struct timespec *now, |
5951 | int flags) | |
22c44fe6 | 5952 | { |
2bc55652 AB |
5953 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5954 | ||
5955 | if (btrfs_root_readonly(root)) | |
5956 | return -EROFS; | |
5957 | ||
e41f941a | 5958 | if (flags & S_VERSION) |
22c44fe6 | 5959 | inode_inc_iversion(inode); |
e41f941a JB |
5960 | if (flags & S_CTIME) |
5961 | inode->i_ctime = *now; | |
5962 | if (flags & S_MTIME) | |
5963 | inode->i_mtime = *now; | |
5964 | if (flags & S_ATIME) | |
5965 | inode->i_atime = *now; | |
5966 | return btrfs_dirty_inode(inode); | |
39279cc3 CM |
5967 | } |
5968 | ||
d352ac68 CM |
5969 | /* |
5970 | * find the highest existing sequence number in a directory | |
5971 | * and then set the in-memory index_cnt variable to reflect | |
5972 | * free sequence numbers | |
5973 | */ | |
4c570655 | 5974 | static int btrfs_set_inode_index_count(struct btrfs_inode *inode) |
aec7477b | 5975 | { |
4c570655 | 5976 | struct btrfs_root *root = inode->root; |
aec7477b JB |
5977 | struct btrfs_key key, found_key; |
5978 | struct btrfs_path *path; | |
5979 | struct extent_buffer *leaf; | |
5980 | int ret; | |
5981 | ||
4c570655 | 5982 | key.objectid = btrfs_ino(inode); |
962a298f | 5983 | key.type = BTRFS_DIR_INDEX_KEY; |
aec7477b JB |
5984 | key.offset = (u64)-1; |
5985 | ||
5986 | path = btrfs_alloc_path(); | |
5987 | if (!path) | |
5988 | return -ENOMEM; | |
5989 | ||
5990 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5991 | if (ret < 0) | |
5992 | goto out; | |
5993 | /* FIXME: we should be able to handle this */ | |
5994 | if (ret == 0) | |
5995 | goto out; | |
5996 | ret = 0; | |
5997 | ||
5998 | /* | |
5999 | * MAGIC NUMBER EXPLANATION: | |
6000 | * since we search a directory based on f_pos we have to start at 2 | |
6001 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
6002 | * else has to start at 2 | |
6003 | */ | |
6004 | if (path->slots[0] == 0) { | |
4c570655 | 6005 | inode->index_cnt = 2; |
aec7477b JB |
6006 | goto out; |
6007 | } | |
6008 | ||
6009 | path->slots[0]--; | |
6010 | ||
6011 | leaf = path->nodes[0]; | |
6012 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
6013 | ||
4c570655 | 6014 | if (found_key.objectid != btrfs_ino(inode) || |
962a298f | 6015 | found_key.type != BTRFS_DIR_INDEX_KEY) { |
4c570655 | 6016 | inode->index_cnt = 2; |
aec7477b JB |
6017 | goto out; |
6018 | } | |
6019 | ||
4c570655 | 6020 | inode->index_cnt = found_key.offset + 1; |
aec7477b JB |
6021 | out: |
6022 | btrfs_free_path(path); | |
6023 | return ret; | |
6024 | } | |
6025 | ||
d352ac68 CM |
6026 | /* |
6027 | * helper to find a free sequence number in a given directory. This current | |
6028 | * code is very simple, later versions will do smarter things in the btree | |
6029 | */ | |
877574e2 | 6030 | int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index) |
aec7477b JB |
6031 | { |
6032 | int ret = 0; | |
6033 | ||
877574e2 NB |
6034 | if (dir->index_cnt == (u64)-1) { |
6035 | ret = btrfs_inode_delayed_dir_index_count(dir); | |
16cdcec7 MX |
6036 | if (ret) { |
6037 | ret = btrfs_set_inode_index_count(dir); | |
6038 | if (ret) | |
6039 | return ret; | |
6040 | } | |
aec7477b JB |
6041 | } |
6042 | ||
877574e2 NB |
6043 | *index = dir->index_cnt; |
6044 | dir->index_cnt++; | |
aec7477b JB |
6045 | |
6046 | return ret; | |
6047 | } | |
6048 | ||
b0d5d10f CM |
6049 | static int btrfs_insert_inode_locked(struct inode *inode) |
6050 | { | |
6051 | struct btrfs_iget_args args; | |
6052 | args.location = &BTRFS_I(inode)->location; | |
6053 | args.root = BTRFS_I(inode)->root; | |
6054 | ||
6055 | return insert_inode_locked4(inode, | |
6056 | btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root), | |
6057 | btrfs_find_actor, &args); | |
6058 | } | |
6059 | ||
39279cc3 CM |
6060 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
6061 | struct btrfs_root *root, | |
aec7477b | 6062 | struct inode *dir, |
9c58309d | 6063 | const char *name, int name_len, |
175a4eb7 AV |
6064 | u64 ref_objectid, u64 objectid, |
6065 | umode_t mode, u64 *index) | |
39279cc3 | 6066 | { |
0b246afa | 6067 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 6068 | struct inode *inode; |
5f39d397 | 6069 | struct btrfs_inode_item *inode_item; |
39279cc3 | 6070 | struct btrfs_key *location; |
5f39d397 | 6071 | struct btrfs_path *path; |
9c58309d CM |
6072 | struct btrfs_inode_ref *ref; |
6073 | struct btrfs_key key[2]; | |
6074 | u32 sizes[2]; | |
ef3b9af5 | 6075 | int nitems = name ? 2 : 1; |
9c58309d | 6076 | unsigned long ptr; |
39279cc3 | 6077 | int ret; |
39279cc3 | 6078 | |
5f39d397 | 6079 | path = btrfs_alloc_path(); |
d8926bb3 MF |
6080 | if (!path) |
6081 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 6082 | |
0b246afa | 6083 | inode = new_inode(fs_info->sb); |
8fb27640 YS |
6084 | if (!inode) { |
6085 | btrfs_free_path(path); | |
39279cc3 | 6086 | return ERR_PTR(-ENOMEM); |
8fb27640 | 6087 | } |
39279cc3 | 6088 | |
5762b5c9 FM |
6089 | /* |
6090 | * O_TMPFILE, set link count to 0, so that after this point, | |
6091 | * we fill in an inode item with the correct link count. | |
6092 | */ | |
6093 | if (!name) | |
6094 | set_nlink(inode, 0); | |
6095 | ||
581bb050 LZ |
6096 | /* |
6097 | * we have to initialize this early, so we can reclaim the inode | |
6098 | * number if we fail afterwards in this function. | |
6099 | */ | |
6100 | inode->i_ino = objectid; | |
6101 | ||
ef3b9af5 | 6102 | if (dir && name) { |
1abe9b8a | 6103 | trace_btrfs_inode_request(dir); |
6104 | ||
877574e2 | 6105 | ret = btrfs_set_inode_index(BTRFS_I(dir), index); |
09771430 | 6106 | if (ret) { |
8fb27640 | 6107 | btrfs_free_path(path); |
09771430 | 6108 | iput(inode); |
aec7477b | 6109 | return ERR_PTR(ret); |
09771430 | 6110 | } |
ef3b9af5 FM |
6111 | } else if (dir) { |
6112 | *index = 0; | |
aec7477b JB |
6113 | } |
6114 | /* | |
6115 | * index_cnt is ignored for everything but a dir, | |
6116 | * btrfs_get_inode_index_count has an explanation for the magic | |
6117 | * number | |
6118 | */ | |
6119 | BTRFS_I(inode)->index_cnt = 2; | |
67de1176 | 6120 | BTRFS_I(inode)->dir_index = *index; |
39279cc3 | 6121 | BTRFS_I(inode)->root = root; |
e02119d5 | 6122 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 6123 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 6124 | |
5dc562c5 JB |
6125 | /* |
6126 | * We could have gotten an inode number from somebody who was fsynced | |
6127 | * and then removed in this same transaction, so let's just set full | |
6128 | * sync since it will be a full sync anyway and this will blow away the | |
6129 | * old info in the log. | |
6130 | */ | |
6131 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
6132 | ||
9c58309d | 6133 | key[0].objectid = objectid; |
962a298f | 6134 | key[0].type = BTRFS_INODE_ITEM_KEY; |
9c58309d CM |
6135 | key[0].offset = 0; |
6136 | ||
9c58309d | 6137 | sizes[0] = sizeof(struct btrfs_inode_item); |
ef3b9af5 FM |
6138 | |
6139 | if (name) { | |
6140 | /* | |
6141 | * Start new inodes with an inode_ref. This is slightly more | |
6142 | * efficient for small numbers of hard links since they will | |
6143 | * be packed into one item. Extended refs will kick in if we | |
6144 | * add more hard links than can fit in the ref item. | |
6145 | */ | |
6146 | key[1].objectid = objectid; | |
962a298f | 6147 | key[1].type = BTRFS_INODE_REF_KEY; |
ef3b9af5 FM |
6148 | key[1].offset = ref_objectid; |
6149 | ||
6150 | sizes[1] = name_len + sizeof(*ref); | |
6151 | } | |
9c58309d | 6152 | |
b0d5d10f CM |
6153 | location = &BTRFS_I(inode)->location; |
6154 | location->objectid = objectid; | |
6155 | location->offset = 0; | |
962a298f | 6156 | location->type = BTRFS_INODE_ITEM_KEY; |
b0d5d10f CM |
6157 | |
6158 | ret = btrfs_insert_inode_locked(inode); | |
6159 | if (ret < 0) | |
6160 | goto fail; | |
6161 | ||
b9473439 | 6162 | path->leave_spinning = 1; |
ef3b9af5 | 6163 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems); |
9c58309d | 6164 | if (ret != 0) |
b0d5d10f | 6165 | goto fail_unlock; |
5f39d397 | 6166 | |
ecc11fab | 6167 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 6168 | inode_set_bytes(inode, 0); |
9cc97d64 | 6169 | |
c2050a45 | 6170 | inode->i_mtime = current_time(inode); |
9cc97d64 | 6171 | inode->i_atime = inode->i_mtime; |
6172 | inode->i_ctime = inode->i_mtime; | |
6173 | BTRFS_I(inode)->i_otime = inode->i_mtime; | |
6174 | ||
5f39d397 CM |
6175 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
6176 | struct btrfs_inode_item); | |
b159fa28 | 6177 | memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item, |
293f7e07 | 6178 | sizeof(*inode_item)); |
e02119d5 | 6179 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d | 6180 | |
ef3b9af5 FM |
6181 | if (name) { |
6182 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
6183 | struct btrfs_inode_ref); | |
6184 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
6185 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); | |
6186 | ptr = (unsigned long)(ref + 1); | |
6187 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
6188 | } | |
9c58309d | 6189 | |
5f39d397 CM |
6190 | btrfs_mark_buffer_dirty(path->nodes[0]); |
6191 | btrfs_free_path(path); | |
6192 | ||
6cbff00f CH |
6193 | btrfs_inherit_iflags(inode, dir); |
6194 | ||
569254b0 | 6195 | if (S_ISREG(mode)) { |
0b246afa | 6196 | if (btrfs_test_opt(fs_info, NODATASUM)) |
94272164 | 6197 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; |
0b246afa | 6198 | if (btrfs_test_opt(fs_info, NODATACOW)) |
f2bdf9a8 JB |
6199 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW | |
6200 | BTRFS_INODE_NODATASUM; | |
94272164 CM |
6201 | } |
6202 | ||
5d4f98a2 | 6203 | inode_tree_add(inode); |
1abe9b8a | 6204 | |
6205 | trace_btrfs_inode_new(inode); | |
1973f0fa | 6206 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 6207 | |
8ea05e3a AB |
6208 | btrfs_update_root_times(trans, root); |
6209 | ||
63541927 FDBM |
6210 | ret = btrfs_inode_inherit_props(trans, inode, dir); |
6211 | if (ret) | |
0b246afa | 6212 | btrfs_err(fs_info, |
63541927 | 6213 | "error inheriting props for ino %llu (root %llu): %d", |
f85b7379 | 6214 | btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret); |
63541927 | 6215 | |
39279cc3 | 6216 | return inode; |
b0d5d10f CM |
6217 | |
6218 | fail_unlock: | |
6219 | unlock_new_inode(inode); | |
5f39d397 | 6220 | fail: |
ef3b9af5 | 6221 | if (dir && name) |
aec7477b | 6222 | BTRFS_I(dir)->index_cnt--; |
5f39d397 | 6223 | btrfs_free_path(path); |
09771430 | 6224 | iput(inode); |
5f39d397 | 6225 | return ERR_PTR(ret); |
39279cc3 CM |
6226 | } |
6227 | ||
6228 | static inline u8 btrfs_inode_type(struct inode *inode) | |
6229 | { | |
6230 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
6231 | } | |
6232 | ||
d352ac68 CM |
6233 | /* |
6234 | * utility function to add 'inode' into 'parent_inode' with | |
6235 | * a give name and a given sequence number. | |
6236 | * if 'add_backref' is true, also insert a backref from the | |
6237 | * inode to the parent directory. | |
6238 | */ | |
e02119d5 | 6239 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
db0a669f | 6240 | struct btrfs_inode *parent_inode, struct btrfs_inode *inode, |
e02119d5 | 6241 | const char *name, int name_len, int add_backref, u64 index) |
39279cc3 | 6242 | { |
db0a669f | 6243 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
4df27c4d | 6244 | int ret = 0; |
39279cc3 | 6245 | struct btrfs_key key; |
db0a669f NB |
6246 | struct btrfs_root *root = parent_inode->root; |
6247 | u64 ino = btrfs_ino(inode); | |
6248 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 6249 | |
33345d01 | 6250 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
db0a669f | 6251 | memcpy(&key, &inode->root->root_key, sizeof(key)); |
4df27c4d | 6252 | } else { |
33345d01 | 6253 | key.objectid = ino; |
962a298f | 6254 | key.type = BTRFS_INODE_ITEM_KEY; |
4df27c4d YZ |
6255 | key.offset = 0; |
6256 | } | |
6257 | ||
33345d01 | 6258 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
0b246afa JM |
6259 | ret = btrfs_add_root_ref(trans, fs_info, key.objectid, |
6260 | root->root_key.objectid, parent_ino, | |
6261 | index, name, name_len); | |
4df27c4d | 6262 | } else if (add_backref) { |
33345d01 LZ |
6263 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
6264 | parent_ino, index); | |
4df27c4d | 6265 | } |
39279cc3 | 6266 | |
79787eaa JM |
6267 | /* Nothing to clean up yet */ |
6268 | if (ret) | |
6269 | return ret; | |
4df27c4d | 6270 | |
79787eaa JM |
6271 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
6272 | parent_inode, &key, | |
db0a669f | 6273 | btrfs_inode_type(&inode->vfs_inode), index); |
9c52057c | 6274 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
6275 | goto fail_dir_item; |
6276 | else if (ret) { | |
66642832 | 6277 | btrfs_abort_transaction(trans, ret); |
79787eaa | 6278 | return ret; |
39279cc3 | 6279 | } |
79787eaa | 6280 | |
db0a669f | 6281 | btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size + |
79787eaa | 6282 | name_len * 2); |
db0a669f NB |
6283 | inode_inc_iversion(&parent_inode->vfs_inode); |
6284 | parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime = | |
6285 | current_time(&parent_inode->vfs_inode); | |
6286 | ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode); | |
79787eaa | 6287 | if (ret) |
66642832 | 6288 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 6289 | return ret; |
fe66a05a CM |
6290 | |
6291 | fail_dir_item: | |
6292 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
6293 | u64 local_index; | |
6294 | int err; | |
0b246afa JM |
6295 | err = btrfs_del_root_ref(trans, fs_info, key.objectid, |
6296 | root->root_key.objectid, parent_ino, | |
6297 | &local_index, name, name_len); | |
fe66a05a CM |
6298 | |
6299 | } else if (add_backref) { | |
6300 | u64 local_index; | |
6301 | int err; | |
6302 | ||
6303 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
6304 | ino, parent_ino, &local_index); | |
6305 | } | |
6306 | return ret; | |
39279cc3 CM |
6307 | } |
6308 | ||
6309 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
cef415af NB |
6310 | struct btrfs_inode *dir, struct dentry *dentry, |
6311 | struct btrfs_inode *inode, int backref, u64 index) | |
39279cc3 | 6312 | { |
a1b075d2 JB |
6313 | int err = btrfs_add_link(trans, dir, inode, |
6314 | dentry->d_name.name, dentry->d_name.len, | |
6315 | backref, index); | |
39279cc3 CM |
6316 | if (err > 0) |
6317 | err = -EEXIST; | |
6318 | return err; | |
6319 | } | |
6320 | ||
618e21d5 | 6321 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 6322 | umode_t mode, dev_t rdev) |
618e21d5 | 6323 | { |
2ff7e61e | 6324 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
618e21d5 JB |
6325 | struct btrfs_trans_handle *trans; |
6326 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6327 | struct inode *inode = NULL; |
618e21d5 JB |
6328 | int err; |
6329 | int drop_inode = 0; | |
6330 | u64 objectid; | |
00e4e6b3 | 6331 | u64 index = 0; |
618e21d5 | 6332 | |
9ed74f2d JB |
6333 | /* |
6334 | * 2 for inode item and ref | |
6335 | * 2 for dir items | |
6336 | * 1 for xattr if selinux is on | |
6337 | */ | |
a22285a6 YZ |
6338 | trans = btrfs_start_transaction(root, 5); |
6339 | if (IS_ERR(trans)) | |
6340 | return PTR_ERR(trans); | |
1832a6d5 | 6341 | |
581bb050 LZ |
6342 | err = btrfs_find_free_ino(root, &objectid); |
6343 | if (err) | |
6344 | goto out_unlock; | |
6345 | ||
aec7477b | 6346 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6347 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6348 | mode, &index); | |
7cf96da3 TI |
6349 | if (IS_ERR(inode)) { |
6350 | err = PTR_ERR(inode); | |
618e21d5 | 6351 | goto out_unlock; |
7cf96da3 | 6352 | } |
618e21d5 | 6353 | |
ad19db71 CS |
6354 | /* |
6355 | * If the active LSM wants to access the inode during | |
6356 | * d_instantiate it needs these. Smack checks to see | |
6357 | * if the filesystem supports xattrs by looking at the | |
6358 | * ops vector. | |
6359 | */ | |
ad19db71 | 6360 | inode->i_op = &btrfs_special_inode_operations; |
b0d5d10f CM |
6361 | init_special_inode(inode, inode->i_mode, rdev); |
6362 | ||
6363 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
618e21d5 | 6364 | if (err) |
b0d5d10f CM |
6365 | goto out_unlock_inode; |
6366 | ||
cef415af NB |
6367 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6368 | 0, index); | |
b0d5d10f CM |
6369 | if (err) { |
6370 | goto out_unlock_inode; | |
6371 | } else { | |
1b4ab1bb | 6372 | btrfs_update_inode(trans, root, inode); |
b0d5d10f | 6373 | unlock_new_inode(inode); |
08c422c2 | 6374 | d_instantiate(dentry, inode); |
618e21d5 | 6375 | } |
b0d5d10f | 6376 | |
618e21d5 | 6377 | out_unlock: |
3a45bb20 | 6378 | btrfs_end_transaction(trans); |
2ff7e61e JM |
6379 | btrfs_balance_delayed_items(fs_info); |
6380 | btrfs_btree_balance_dirty(fs_info); | |
618e21d5 JB |
6381 | if (drop_inode) { |
6382 | inode_dec_link_count(inode); | |
6383 | iput(inode); | |
6384 | } | |
618e21d5 | 6385 | return err; |
b0d5d10f CM |
6386 | |
6387 | out_unlock_inode: | |
6388 | drop_inode = 1; | |
6389 | unlock_new_inode(inode); | |
6390 | goto out_unlock; | |
6391 | ||
618e21d5 JB |
6392 | } |
6393 | ||
39279cc3 | 6394 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 6395 | umode_t mode, bool excl) |
39279cc3 | 6396 | { |
2ff7e61e | 6397 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
6398 | struct btrfs_trans_handle *trans; |
6399 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6400 | struct inode *inode = NULL; |
43baa579 | 6401 | int drop_inode_on_err = 0; |
a22285a6 | 6402 | int err; |
39279cc3 | 6403 | u64 objectid; |
00e4e6b3 | 6404 | u64 index = 0; |
39279cc3 | 6405 | |
9ed74f2d JB |
6406 | /* |
6407 | * 2 for inode item and ref | |
6408 | * 2 for dir items | |
6409 | * 1 for xattr if selinux is on | |
6410 | */ | |
a22285a6 YZ |
6411 | trans = btrfs_start_transaction(root, 5); |
6412 | if (IS_ERR(trans)) | |
6413 | return PTR_ERR(trans); | |
9ed74f2d | 6414 | |
581bb050 LZ |
6415 | err = btrfs_find_free_ino(root, &objectid); |
6416 | if (err) | |
6417 | goto out_unlock; | |
6418 | ||
aec7477b | 6419 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6420 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6421 | mode, &index); | |
7cf96da3 TI |
6422 | if (IS_ERR(inode)) { |
6423 | err = PTR_ERR(inode); | |
39279cc3 | 6424 | goto out_unlock; |
7cf96da3 | 6425 | } |
43baa579 | 6426 | drop_inode_on_err = 1; |
ad19db71 CS |
6427 | /* |
6428 | * If the active LSM wants to access the inode during | |
6429 | * d_instantiate it needs these. Smack checks to see | |
6430 | * if the filesystem supports xattrs by looking at the | |
6431 | * ops vector. | |
6432 | */ | |
6433 | inode->i_fop = &btrfs_file_operations; | |
6434 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 6435 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
6436 | |
6437 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
6438 | if (err) | |
6439 | goto out_unlock_inode; | |
6440 | ||
6441 | err = btrfs_update_inode(trans, root, inode); | |
6442 | if (err) | |
6443 | goto out_unlock_inode; | |
ad19db71 | 6444 | |
cef415af NB |
6445 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6446 | 0, index); | |
39279cc3 | 6447 | if (err) |
b0d5d10f | 6448 | goto out_unlock_inode; |
43baa579 | 6449 | |
43baa579 | 6450 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
b0d5d10f | 6451 | unlock_new_inode(inode); |
43baa579 FB |
6452 | d_instantiate(dentry, inode); |
6453 | ||
39279cc3 | 6454 | out_unlock: |
3a45bb20 | 6455 | btrfs_end_transaction(trans); |
43baa579 | 6456 | if (err && drop_inode_on_err) { |
39279cc3 CM |
6457 | inode_dec_link_count(inode); |
6458 | iput(inode); | |
6459 | } | |
2ff7e61e JM |
6460 | btrfs_balance_delayed_items(fs_info); |
6461 | btrfs_btree_balance_dirty(fs_info); | |
39279cc3 | 6462 | return err; |
b0d5d10f CM |
6463 | |
6464 | out_unlock_inode: | |
6465 | unlock_new_inode(inode); | |
6466 | goto out_unlock; | |
6467 | ||
39279cc3 CM |
6468 | } |
6469 | ||
6470 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
6471 | struct dentry *dentry) | |
6472 | { | |
271dba45 | 6473 | struct btrfs_trans_handle *trans = NULL; |
39279cc3 | 6474 | struct btrfs_root *root = BTRFS_I(dir)->root; |
2b0143b5 | 6475 | struct inode *inode = d_inode(old_dentry); |
2ff7e61e | 6476 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
00e4e6b3 | 6477 | u64 index; |
39279cc3 CM |
6478 | int err; |
6479 | int drop_inode = 0; | |
6480 | ||
4a8be425 TH |
6481 | /* do not allow sys_link's with other subvols of the same device */ |
6482 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 6483 | return -EXDEV; |
4a8be425 | 6484 | |
f186373f | 6485 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 6486 | return -EMLINK; |
4a8be425 | 6487 | |
877574e2 | 6488 | err = btrfs_set_inode_index(BTRFS_I(dir), &index); |
aec7477b JB |
6489 | if (err) |
6490 | goto fail; | |
6491 | ||
a22285a6 | 6492 | /* |
7e6b6465 | 6493 | * 2 items for inode and inode ref |
a22285a6 | 6494 | * 2 items for dir items |
7e6b6465 | 6495 | * 1 item for parent inode |
a22285a6 | 6496 | */ |
7e6b6465 | 6497 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
6498 | if (IS_ERR(trans)) { |
6499 | err = PTR_ERR(trans); | |
271dba45 | 6500 | trans = NULL; |
a22285a6 YZ |
6501 | goto fail; |
6502 | } | |
5f39d397 | 6503 | |
67de1176 MX |
6504 | /* There are several dir indexes for this inode, clear the cache. */ |
6505 | BTRFS_I(inode)->dir_index = 0ULL; | |
8b558c5f | 6506 | inc_nlink(inode); |
0c4d2d95 | 6507 | inode_inc_iversion(inode); |
c2050a45 | 6508 | inode->i_ctime = current_time(inode); |
7de9c6ee | 6509 | ihold(inode); |
e9976151 | 6510 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 6511 | |
cef415af NB |
6512 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode), |
6513 | 1, index); | |
5f39d397 | 6514 | |
a5719521 | 6515 | if (err) { |
54aa1f4d | 6516 | drop_inode = 1; |
a5719521 | 6517 | } else { |
10d9f309 | 6518 | struct dentry *parent = dentry->d_parent; |
a5719521 | 6519 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
6520 | if (err) |
6521 | goto fail; | |
ef3b9af5 FM |
6522 | if (inode->i_nlink == 1) { |
6523 | /* | |
6524 | * If new hard link count is 1, it's a file created | |
6525 | * with open(2) O_TMPFILE flag. | |
6526 | */ | |
3d6ae7bb | 6527 | err = btrfs_orphan_del(trans, BTRFS_I(inode)); |
ef3b9af5 FM |
6528 | if (err) |
6529 | goto fail; | |
6530 | } | |
08c422c2 | 6531 | d_instantiate(dentry, inode); |
9ca5fbfb | 6532 | btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent); |
a5719521 | 6533 | } |
39279cc3 | 6534 | |
2ff7e61e | 6535 | btrfs_balance_delayed_items(fs_info); |
1832a6d5 | 6536 | fail: |
271dba45 | 6537 | if (trans) |
3a45bb20 | 6538 | btrfs_end_transaction(trans); |
39279cc3 CM |
6539 | if (drop_inode) { |
6540 | inode_dec_link_count(inode); | |
6541 | iput(inode); | |
6542 | } | |
2ff7e61e | 6543 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6544 | return err; |
6545 | } | |
6546 | ||
18bb1db3 | 6547 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 6548 | { |
2ff7e61e | 6549 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
b9d86667 | 6550 | struct inode *inode = NULL; |
39279cc3 CM |
6551 | struct btrfs_trans_handle *trans; |
6552 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
6553 | int err = 0; | |
6554 | int drop_on_err = 0; | |
b9d86667 | 6555 | u64 objectid = 0; |
00e4e6b3 | 6556 | u64 index = 0; |
39279cc3 | 6557 | |
9ed74f2d JB |
6558 | /* |
6559 | * 2 items for inode and ref | |
6560 | * 2 items for dir items | |
6561 | * 1 for xattr if selinux is on | |
6562 | */ | |
a22285a6 YZ |
6563 | trans = btrfs_start_transaction(root, 5); |
6564 | if (IS_ERR(trans)) | |
6565 | return PTR_ERR(trans); | |
39279cc3 | 6566 | |
581bb050 LZ |
6567 | err = btrfs_find_free_ino(root, &objectid); |
6568 | if (err) | |
6569 | goto out_fail; | |
6570 | ||
aec7477b | 6571 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6572 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6573 | S_IFDIR | mode, &index); | |
39279cc3 CM |
6574 | if (IS_ERR(inode)) { |
6575 | err = PTR_ERR(inode); | |
6576 | goto out_fail; | |
6577 | } | |
5f39d397 | 6578 | |
39279cc3 | 6579 | drop_on_err = 1; |
b0d5d10f CM |
6580 | /* these must be set before we unlock the inode */ |
6581 | inode->i_op = &btrfs_dir_inode_operations; | |
6582 | inode->i_fop = &btrfs_dir_file_operations; | |
33268eaf | 6583 | |
2a7dba39 | 6584 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf | 6585 | if (err) |
b0d5d10f | 6586 | goto out_fail_inode; |
39279cc3 | 6587 | |
6ef06d27 | 6588 | btrfs_i_size_write(BTRFS_I(inode), 0); |
39279cc3 CM |
6589 | err = btrfs_update_inode(trans, root, inode); |
6590 | if (err) | |
b0d5d10f | 6591 | goto out_fail_inode; |
5f39d397 | 6592 | |
db0a669f NB |
6593 | err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), |
6594 | dentry->d_name.name, | |
6595 | dentry->d_name.len, 0, index); | |
39279cc3 | 6596 | if (err) |
b0d5d10f | 6597 | goto out_fail_inode; |
5f39d397 | 6598 | |
39279cc3 | 6599 | d_instantiate(dentry, inode); |
b0d5d10f CM |
6600 | /* |
6601 | * mkdir is special. We're unlocking after we call d_instantiate | |
6602 | * to avoid a race with nfsd calling d_instantiate. | |
6603 | */ | |
6604 | unlock_new_inode(inode); | |
39279cc3 | 6605 | drop_on_err = 0; |
39279cc3 CM |
6606 | |
6607 | out_fail: | |
3a45bb20 | 6608 | btrfs_end_transaction(trans); |
c7cfb8a5 WS |
6609 | if (drop_on_err) { |
6610 | inode_dec_link_count(inode); | |
39279cc3 | 6611 | iput(inode); |
c7cfb8a5 | 6612 | } |
2ff7e61e JM |
6613 | btrfs_balance_delayed_items(fs_info); |
6614 | btrfs_btree_balance_dirty(fs_info); | |
39279cc3 | 6615 | return err; |
b0d5d10f CM |
6616 | |
6617 | out_fail_inode: | |
6618 | unlock_new_inode(inode); | |
6619 | goto out_fail; | |
39279cc3 CM |
6620 | } |
6621 | ||
e6c4efd8 QW |
6622 | /* Find next extent map of a given extent map, caller needs to ensure locks */ |
6623 | static struct extent_map *next_extent_map(struct extent_map *em) | |
6624 | { | |
6625 | struct rb_node *next; | |
6626 | ||
6627 | next = rb_next(&em->rb_node); | |
6628 | if (!next) | |
6629 | return NULL; | |
6630 | return container_of(next, struct extent_map, rb_node); | |
6631 | } | |
6632 | ||
6633 | static struct extent_map *prev_extent_map(struct extent_map *em) | |
6634 | { | |
6635 | struct rb_node *prev; | |
6636 | ||
6637 | prev = rb_prev(&em->rb_node); | |
6638 | if (!prev) | |
6639 | return NULL; | |
6640 | return container_of(prev, struct extent_map, rb_node); | |
6641 | } | |
6642 | ||
d352ac68 | 6643 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
e6c4efd8 | 6644 | * the existing extent is the nearest extent to map_start, |
d352ac68 | 6645 | * and an extent that you want to insert, deal with overlap and insert |
e6c4efd8 | 6646 | * the best fitted new extent into the tree. |
d352ac68 | 6647 | */ |
3b951516 CM |
6648 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
6649 | struct extent_map *existing, | |
e6dcd2dc | 6650 | struct extent_map *em, |
51f395ad | 6651 | u64 map_start) |
3b951516 | 6652 | { |
e6c4efd8 QW |
6653 | struct extent_map *prev; |
6654 | struct extent_map *next; | |
6655 | u64 start; | |
6656 | u64 end; | |
3b951516 | 6657 | u64 start_diff; |
3b951516 | 6658 | |
e6dcd2dc | 6659 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
e6c4efd8 QW |
6660 | |
6661 | if (existing->start > map_start) { | |
6662 | next = existing; | |
6663 | prev = prev_extent_map(next); | |
6664 | } else { | |
6665 | prev = existing; | |
6666 | next = next_extent_map(prev); | |
6667 | } | |
6668 | ||
6669 | start = prev ? extent_map_end(prev) : em->start; | |
6670 | start = max_t(u64, start, em->start); | |
6671 | end = next ? next->start : extent_map_end(em); | |
6672 | end = min_t(u64, end, extent_map_end(em)); | |
6673 | start_diff = start - em->start; | |
6674 | em->start = start; | |
6675 | em->len = end - start; | |
c8b97818 CM |
6676 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
6677 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 6678 | em->block_start += start_diff; |
c8b97818 CM |
6679 | em->block_len -= start_diff; |
6680 | } | |
09a2a8f9 | 6681 | return add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
6682 | } |
6683 | ||
c8b97818 | 6684 | static noinline int uncompress_inline(struct btrfs_path *path, |
e40da0e5 | 6685 | struct page *page, |
c8b97818 CM |
6686 | size_t pg_offset, u64 extent_offset, |
6687 | struct btrfs_file_extent_item *item) | |
6688 | { | |
6689 | int ret; | |
6690 | struct extent_buffer *leaf = path->nodes[0]; | |
6691 | char *tmp; | |
6692 | size_t max_size; | |
6693 | unsigned long inline_size; | |
6694 | unsigned long ptr; | |
261507a0 | 6695 | int compress_type; |
c8b97818 CM |
6696 | |
6697 | WARN_ON(pg_offset != 0); | |
261507a0 | 6698 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
6699 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
6700 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
dd3cc16b | 6701 | btrfs_item_nr(path->slots[0])); |
c8b97818 | 6702 | tmp = kmalloc(inline_size, GFP_NOFS); |
8d413713 TI |
6703 | if (!tmp) |
6704 | return -ENOMEM; | |
c8b97818 CM |
6705 | ptr = btrfs_file_extent_inline_start(item); |
6706 | ||
6707 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
6708 | ||
09cbfeaf | 6709 | max_size = min_t(unsigned long, PAGE_SIZE, max_size); |
261507a0 LZ |
6710 | ret = btrfs_decompress(compress_type, tmp, page, |
6711 | extent_offset, inline_size, max_size); | |
e1699d2d ZB |
6712 | |
6713 | /* | |
6714 | * decompression code contains a memset to fill in any space between the end | |
6715 | * of the uncompressed data and the end of max_size in case the decompressed | |
6716 | * data ends up shorter than ram_bytes. That doesn't cover the hole between | |
6717 | * the end of an inline extent and the beginning of the next block, so we | |
6718 | * cover that region here. | |
6719 | */ | |
6720 | ||
6721 | if (max_size + pg_offset < PAGE_SIZE) { | |
6722 | char *map = kmap(page); | |
6723 | memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset); | |
6724 | kunmap(page); | |
6725 | } | |
c8b97818 | 6726 | kfree(tmp); |
166ae5a4 | 6727 | return ret; |
c8b97818 CM |
6728 | } |
6729 | ||
d352ac68 CM |
6730 | /* |
6731 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
6732 | * the ugly parts come from merging extents from the disk with the in-ram |
6733 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
6734 | * where the in-ram extents might be locked pending data=ordered completion. |
6735 | * | |
6736 | * This also copies inline extents directly into the page. | |
6737 | */ | |
d397712b | 6738 | |
fc4f21b1 NB |
6739 | struct extent_map *btrfs_get_extent(struct btrfs_inode *inode, |
6740 | struct page *page, | |
6741 | size_t pg_offset, u64 start, u64 len, | |
6742 | int create) | |
a52d9a80 | 6743 | { |
fc4f21b1 | 6744 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
a52d9a80 CM |
6745 | int ret; |
6746 | int err = 0; | |
a52d9a80 CM |
6747 | u64 extent_start = 0; |
6748 | u64 extent_end = 0; | |
fc4f21b1 | 6749 | u64 objectid = btrfs_ino(inode); |
a52d9a80 | 6750 | u32 found_type; |
f421950f | 6751 | struct btrfs_path *path = NULL; |
fc4f21b1 | 6752 | struct btrfs_root *root = inode->root; |
a52d9a80 | 6753 | struct btrfs_file_extent_item *item; |
5f39d397 CM |
6754 | struct extent_buffer *leaf; |
6755 | struct btrfs_key found_key; | |
a52d9a80 | 6756 | struct extent_map *em = NULL; |
fc4f21b1 NB |
6757 | struct extent_map_tree *em_tree = &inode->extent_tree; |
6758 | struct extent_io_tree *io_tree = &inode->io_tree; | |
a52d9a80 | 6759 | struct btrfs_trans_handle *trans = NULL; |
7ffbb598 | 6760 | const bool new_inline = !page || create; |
a52d9a80 | 6761 | |
a52d9a80 | 6762 | again: |
890871be | 6763 | read_lock(&em_tree->lock); |
d1310b2e | 6764 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d | 6765 | if (em) |
0b246afa | 6766 | em->bdev = fs_info->fs_devices->latest_bdev; |
890871be | 6767 | read_unlock(&em_tree->lock); |
d1310b2e | 6768 | |
a52d9a80 | 6769 | if (em) { |
e1c4b745 CM |
6770 | if (em->start > start || em->start + em->len <= start) |
6771 | free_extent_map(em); | |
6772 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
6773 | free_extent_map(em); |
6774 | else | |
6775 | goto out; | |
a52d9a80 | 6776 | } |
172ddd60 | 6777 | em = alloc_extent_map(); |
a52d9a80 | 6778 | if (!em) { |
d1310b2e CM |
6779 | err = -ENOMEM; |
6780 | goto out; | |
a52d9a80 | 6781 | } |
0b246afa | 6782 | em->bdev = fs_info->fs_devices->latest_bdev; |
d1310b2e | 6783 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 6784 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 6785 | em->len = (u64)-1; |
c8b97818 | 6786 | em->block_len = (u64)-1; |
f421950f CM |
6787 | |
6788 | if (!path) { | |
6789 | path = btrfs_alloc_path(); | |
026fd317 JB |
6790 | if (!path) { |
6791 | err = -ENOMEM; | |
6792 | goto out; | |
6793 | } | |
6794 | /* | |
6795 | * Chances are we'll be called again, so go ahead and do | |
6796 | * readahead | |
6797 | */ | |
e4058b54 | 6798 | path->reada = READA_FORWARD; |
f421950f CM |
6799 | } |
6800 | ||
179e29e4 CM |
6801 | ret = btrfs_lookup_file_extent(trans, root, path, |
6802 | objectid, start, trans != NULL); | |
a52d9a80 CM |
6803 | if (ret < 0) { |
6804 | err = ret; | |
6805 | goto out; | |
6806 | } | |
6807 | ||
6808 | if (ret != 0) { | |
6809 | if (path->slots[0] == 0) | |
6810 | goto not_found; | |
6811 | path->slots[0]--; | |
6812 | } | |
6813 | ||
5f39d397 CM |
6814 | leaf = path->nodes[0]; |
6815 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 6816 | struct btrfs_file_extent_item); |
a52d9a80 | 6817 | /* are we inside the extent that was found? */ |
5f39d397 | 6818 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
962a298f | 6819 | found_type = found_key.type; |
5f39d397 | 6820 | if (found_key.objectid != objectid || |
a52d9a80 | 6821 | found_type != BTRFS_EXTENT_DATA_KEY) { |
25a50341 JB |
6822 | /* |
6823 | * If we backup past the first extent we want to move forward | |
6824 | * and see if there is an extent in front of us, otherwise we'll | |
6825 | * say there is a hole for our whole search range which can | |
6826 | * cause problems. | |
6827 | */ | |
6828 | extent_end = start; | |
6829 | goto next; | |
a52d9a80 CM |
6830 | } |
6831 | ||
5f39d397 CM |
6832 | found_type = btrfs_file_extent_type(leaf, item); |
6833 | extent_start = found_key.offset; | |
d899e052 YZ |
6834 | if (found_type == BTRFS_FILE_EXTENT_REG || |
6835 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 6836 | extent_end = extent_start + |
db94535d | 6837 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
6838 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
6839 | size_t size; | |
514ac8ad | 6840 | size = btrfs_file_extent_inline_len(leaf, path->slots[0], item); |
da17066c | 6841 | extent_end = ALIGN(extent_start + size, |
0b246afa | 6842 | fs_info->sectorsize); |
9036c102 | 6843 | } |
25a50341 | 6844 | next: |
9036c102 YZ |
6845 | if (start >= extent_end) { |
6846 | path->slots[0]++; | |
6847 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
6848 | ret = btrfs_next_leaf(root, path); | |
6849 | if (ret < 0) { | |
6850 | err = ret; | |
6851 | goto out; | |
a52d9a80 | 6852 | } |
9036c102 YZ |
6853 | if (ret > 0) |
6854 | goto not_found; | |
6855 | leaf = path->nodes[0]; | |
a52d9a80 | 6856 | } |
9036c102 YZ |
6857 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
6858 | if (found_key.objectid != objectid || | |
6859 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
6860 | goto not_found; | |
6861 | if (start + len <= found_key.offset) | |
6862 | goto not_found; | |
e2eca69d WS |
6863 | if (start > found_key.offset) |
6864 | goto next; | |
9036c102 | 6865 | em->start = start; |
70c8a91c | 6866 | em->orig_start = start; |
9036c102 YZ |
6867 | em->len = found_key.offset - start; |
6868 | goto not_found_em; | |
6869 | } | |
6870 | ||
fc4f21b1 | 6871 | btrfs_extent_item_to_extent_map(inode, path, item, |
9cdc5124 | 6872 | new_inline, em); |
7ffbb598 | 6873 | |
d899e052 YZ |
6874 | if (found_type == BTRFS_FILE_EXTENT_REG || |
6875 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 CM |
6876 | goto insert; |
6877 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 6878 | unsigned long ptr; |
a52d9a80 | 6879 | char *map; |
3326d1b0 CM |
6880 | size_t size; |
6881 | size_t extent_offset; | |
6882 | size_t copy_size; | |
a52d9a80 | 6883 | |
7ffbb598 | 6884 | if (new_inline) |
689f9346 | 6885 | goto out; |
5f39d397 | 6886 | |
514ac8ad | 6887 | size = btrfs_file_extent_inline_len(leaf, path->slots[0], item); |
9036c102 | 6888 | extent_offset = page_offset(page) + pg_offset - extent_start; |
09cbfeaf KS |
6889 | copy_size = min_t(u64, PAGE_SIZE - pg_offset, |
6890 | size - extent_offset); | |
3326d1b0 | 6891 | em->start = extent_start + extent_offset; |
0b246afa | 6892 | em->len = ALIGN(copy_size, fs_info->sectorsize); |
b4939680 | 6893 | em->orig_block_len = em->len; |
70c8a91c | 6894 | em->orig_start = em->start; |
689f9346 | 6895 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 6896 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
6897 | if (btrfs_file_extent_compression(leaf, item) != |
6898 | BTRFS_COMPRESS_NONE) { | |
e40da0e5 | 6899 | ret = uncompress_inline(path, page, pg_offset, |
c8b97818 | 6900 | extent_offset, item); |
166ae5a4 ZB |
6901 | if (ret) { |
6902 | err = ret; | |
6903 | goto out; | |
6904 | } | |
c8b97818 CM |
6905 | } else { |
6906 | map = kmap(page); | |
6907 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
6908 | copy_size); | |
09cbfeaf | 6909 | if (pg_offset + copy_size < PAGE_SIZE) { |
93c82d57 | 6910 | memset(map + pg_offset + copy_size, 0, |
09cbfeaf | 6911 | PAGE_SIZE - pg_offset - |
93c82d57 CM |
6912 | copy_size); |
6913 | } | |
c8b97818 CM |
6914 | kunmap(page); |
6915 | } | |
179e29e4 CM |
6916 | flush_dcache_page(page); |
6917 | } else if (create && PageUptodate(page)) { | |
6bf7e080 | 6918 | BUG(); |
179e29e4 CM |
6919 | if (!trans) { |
6920 | kunmap(page); | |
6921 | free_extent_map(em); | |
6922 | em = NULL; | |
ff5714cc | 6923 | |
b3b4aa74 | 6924 | btrfs_release_path(path); |
7a7eaa40 | 6925 | trans = btrfs_join_transaction(root); |
ff5714cc | 6926 | |
3612b495 TI |
6927 | if (IS_ERR(trans)) |
6928 | return ERR_CAST(trans); | |
179e29e4 CM |
6929 | goto again; |
6930 | } | |
c8b97818 | 6931 | map = kmap(page); |
70dec807 | 6932 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 6933 | copy_size); |
c8b97818 | 6934 | kunmap(page); |
179e29e4 | 6935 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 6936 | } |
d1310b2e | 6937 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 6938 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 | 6939 | goto insert; |
a52d9a80 CM |
6940 | } |
6941 | not_found: | |
6942 | em->start = start; | |
70c8a91c | 6943 | em->orig_start = start; |
d1310b2e | 6944 | em->len = len; |
a52d9a80 | 6945 | not_found_em: |
5f39d397 | 6946 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 6947 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 | 6948 | insert: |
b3b4aa74 | 6949 | btrfs_release_path(path); |
d1310b2e | 6950 | if (em->start > start || extent_map_end(em) <= start) { |
0b246afa | 6951 | btrfs_err(fs_info, |
5d163e0e JM |
6952 | "bad extent! em: [%llu %llu] passed [%llu %llu]", |
6953 | em->start, em->len, start, len); | |
a52d9a80 CM |
6954 | err = -EIO; |
6955 | goto out; | |
6956 | } | |
d1310b2e CM |
6957 | |
6958 | err = 0; | |
890871be | 6959 | write_lock(&em_tree->lock); |
09a2a8f9 | 6960 | ret = add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
6961 | /* it is possible that someone inserted the extent into the tree |
6962 | * while we had the lock dropped. It is also possible that | |
6963 | * an overlapping map exists in the tree | |
6964 | */ | |
a52d9a80 | 6965 | if (ret == -EEXIST) { |
3b951516 | 6966 | struct extent_map *existing; |
e6dcd2dc CM |
6967 | |
6968 | ret = 0; | |
6969 | ||
e6c4efd8 QW |
6970 | existing = search_extent_mapping(em_tree, start, len); |
6971 | /* | |
6972 | * existing will always be non-NULL, since there must be | |
6973 | * extent causing the -EEXIST. | |
6974 | */ | |
8dff9c85 | 6975 | if (existing->start == em->start && |
8e2bd3b7 | 6976 | extent_map_end(existing) >= extent_map_end(em) && |
8dff9c85 CM |
6977 | em->block_start == existing->block_start) { |
6978 | /* | |
8e2bd3b7 OS |
6979 | * The existing extent map already encompasses the |
6980 | * entire extent map we tried to add. | |
8dff9c85 CM |
6981 | */ |
6982 | free_extent_map(em); | |
6983 | em = existing; | |
6984 | err = 0; | |
6985 | ||
6986 | } else if (start >= extent_map_end(existing) || | |
32be3a1a | 6987 | start <= existing->start) { |
e6c4efd8 QW |
6988 | /* |
6989 | * The existing extent map is the one nearest to | |
6990 | * the [start, start + len) range which overlaps | |
6991 | */ | |
6992 | err = merge_extent_mapping(em_tree, existing, | |
6993 | em, start); | |
e1c4b745 | 6994 | free_extent_map(existing); |
e6c4efd8 | 6995 | if (err) { |
3b951516 CM |
6996 | free_extent_map(em); |
6997 | em = NULL; | |
6998 | } | |
6999 | } else { | |
7000 | free_extent_map(em); | |
7001 | em = existing; | |
e6dcd2dc | 7002 | err = 0; |
a52d9a80 | 7003 | } |
a52d9a80 | 7004 | } |
890871be | 7005 | write_unlock(&em_tree->lock); |
a52d9a80 | 7006 | out: |
1abe9b8a | 7007 | |
fc4f21b1 | 7008 | trace_btrfs_get_extent(root, inode, em); |
1abe9b8a | 7009 | |
527afb44 | 7010 | btrfs_free_path(path); |
a52d9a80 | 7011 | if (trans) { |
3a45bb20 | 7012 | ret = btrfs_end_transaction(trans); |
d397712b | 7013 | if (!err) |
a52d9a80 CM |
7014 | err = ret; |
7015 | } | |
a52d9a80 CM |
7016 | if (err) { |
7017 | free_extent_map(em); | |
a52d9a80 CM |
7018 | return ERR_PTR(err); |
7019 | } | |
79787eaa | 7020 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
7021 | return em; |
7022 | } | |
7023 | ||
fc4f21b1 NB |
7024 | struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode, |
7025 | struct page *page, | |
7026 | size_t pg_offset, u64 start, u64 len, | |
7027 | int create) | |
ec29ed5b CM |
7028 | { |
7029 | struct extent_map *em; | |
7030 | struct extent_map *hole_em = NULL; | |
7031 | u64 range_start = start; | |
7032 | u64 end; | |
7033 | u64 found; | |
7034 | u64 found_end; | |
7035 | int err = 0; | |
7036 | ||
7037 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
7038 | if (IS_ERR(em)) | |
7039 | return em; | |
7040 | if (em) { | |
7041 | /* | |
f9e4fb53 LB |
7042 | * if our em maps to |
7043 | * - a hole or | |
7044 | * - a pre-alloc extent, | |
7045 | * there might actually be delalloc bytes behind it. | |
ec29ed5b | 7046 | */ |
f9e4fb53 LB |
7047 | if (em->block_start != EXTENT_MAP_HOLE && |
7048 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
ec29ed5b CM |
7049 | return em; |
7050 | else | |
7051 | hole_em = em; | |
7052 | } | |
7053 | ||
7054 | /* check to see if we've wrapped (len == -1 or similar) */ | |
7055 | end = start + len; | |
7056 | if (end < start) | |
7057 | end = (u64)-1; | |
7058 | else | |
7059 | end -= 1; | |
7060 | ||
7061 | em = NULL; | |
7062 | ||
7063 | /* ok, we didn't find anything, lets look for delalloc */ | |
fc4f21b1 | 7064 | found = count_range_bits(&inode->io_tree, &range_start, |
ec29ed5b CM |
7065 | end, len, EXTENT_DELALLOC, 1); |
7066 | found_end = range_start + found; | |
7067 | if (found_end < range_start) | |
7068 | found_end = (u64)-1; | |
7069 | ||
7070 | /* | |
7071 | * we didn't find anything useful, return | |
7072 | * the original results from get_extent() | |
7073 | */ | |
7074 | if (range_start > end || found_end <= start) { | |
7075 | em = hole_em; | |
7076 | hole_em = NULL; | |
7077 | goto out; | |
7078 | } | |
7079 | ||
7080 | /* adjust the range_start to make sure it doesn't | |
7081 | * go backwards from the start they passed in | |
7082 | */ | |
67871254 | 7083 | range_start = max(start, range_start); |
ec29ed5b CM |
7084 | found = found_end - range_start; |
7085 | ||
7086 | if (found > 0) { | |
7087 | u64 hole_start = start; | |
7088 | u64 hole_len = len; | |
7089 | ||
172ddd60 | 7090 | em = alloc_extent_map(); |
ec29ed5b CM |
7091 | if (!em) { |
7092 | err = -ENOMEM; | |
7093 | goto out; | |
7094 | } | |
7095 | /* | |
7096 | * when btrfs_get_extent can't find anything it | |
7097 | * returns one huge hole | |
7098 | * | |
7099 | * make sure what it found really fits our range, and | |
7100 | * adjust to make sure it is based on the start from | |
7101 | * the caller | |
7102 | */ | |
7103 | if (hole_em) { | |
7104 | u64 calc_end = extent_map_end(hole_em); | |
7105 | ||
7106 | if (calc_end <= start || (hole_em->start > end)) { | |
7107 | free_extent_map(hole_em); | |
7108 | hole_em = NULL; | |
7109 | } else { | |
7110 | hole_start = max(hole_em->start, start); | |
7111 | hole_len = calc_end - hole_start; | |
7112 | } | |
7113 | } | |
7114 | em->bdev = NULL; | |
7115 | if (hole_em && range_start > hole_start) { | |
7116 | /* our hole starts before our delalloc, so we | |
7117 | * have to return just the parts of the hole | |
7118 | * that go until the delalloc starts | |
7119 | */ | |
7120 | em->len = min(hole_len, | |
7121 | range_start - hole_start); | |
7122 | em->start = hole_start; | |
7123 | em->orig_start = hole_start; | |
7124 | /* | |
7125 | * don't adjust block start at all, | |
7126 | * it is fixed at EXTENT_MAP_HOLE | |
7127 | */ | |
7128 | em->block_start = hole_em->block_start; | |
7129 | em->block_len = hole_len; | |
f9e4fb53 LB |
7130 | if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags)) |
7131 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
ec29ed5b CM |
7132 | } else { |
7133 | em->start = range_start; | |
7134 | em->len = found; | |
7135 | em->orig_start = range_start; | |
7136 | em->block_start = EXTENT_MAP_DELALLOC; | |
7137 | em->block_len = found; | |
7138 | } | |
7139 | } else if (hole_em) { | |
7140 | return hole_em; | |
7141 | } | |
7142 | out: | |
7143 | ||
7144 | free_extent_map(hole_em); | |
7145 | if (err) { | |
7146 | free_extent_map(em); | |
7147 | return ERR_PTR(err); | |
7148 | } | |
7149 | return em; | |
7150 | } | |
7151 | ||
5f9a8a51 FM |
7152 | static struct extent_map *btrfs_create_dio_extent(struct inode *inode, |
7153 | const u64 start, | |
7154 | const u64 len, | |
7155 | const u64 orig_start, | |
7156 | const u64 block_start, | |
7157 | const u64 block_len, | |
7158 | const u64 orig_block_len, | |
7159 | const u64 ram_bytes, | |
7160 | const int type) | |
7161 | { | |
7162 | struct extent_map *em = NULL; | |
7163 | int ret; | |
7164 | ||
5f9a8a51 | 7165 | if (type != BTRFS_ORDERED_NOCOW) { |
6f9994db LB |
7166 | em = create_io_em(inode, start, len, orig_start, |
7167 | block_start, block_len, orig_block_len, | |
7168 | ram_bytes, | |
7169 | BTRFS_COMPRESS_NONE, /* compress_type */ | |
7170 | type); | |
5f9a8a51 FM |
7171 | if (IS_ERR(em)) |
7172 | goto out; | |
7173 | } | |
7174 | ret = btrfs_add_ordered_extent_dio(inode, start, block_start, | |
7175 | len, block_len, type); | |
7176 | if (ret) { | |
7177 | if (em) { | |
7178 | free_extent_map(em); | |
dcdbc059 | 7179 | btrfs_drop_extent_cache(BTRFS_I(inode), start, |
5f9a8a51 FM |
7180 | start + len - 1, 0); |
7181 | } | |
7182 | em = ERR_PTR(ret); | |
7183 | } | |
7184 | out: | |
5f9a8a51 FM |
7185 | |
7186 | return em; | |
7187 | } | |
7188 | ||
4b46fce2 JB |
7189 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
7190 | u64 start, u64 len) | |
7191 | { | |
0b246afa | 7192 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 7193 | struct btrfs_root *root = BTRFS_I(inode)->root; |
70c8a91c | 7194 | struct extent_map *em; |
4b46fce2 JB |
7195 | struct btrfs_key ins; |
7196 | u64 alloc_hint; | |
7197 | int ret; | |
4b46fce2 | 7198 | |
4b46fce2 | 7199 | alloc_hint = get_extent_allocation_hint(inode, start, len); |
0b246afa | 7200 | ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize, |
da17066c | 7201 | 0, alloc_hint, &ins, 1, 1); |
00361589 JB |
7202 | if (ret) |
7203 | return ERR_PTR(ret); | |
4b46fce2 | 7204 | |
5f9a8a51 FM |
7205 | em = btrfs_create_dio_extent(inode, start, ins.offset, start, |
7206 | ins.objectid, ins.offset, ins.offset, | |
6288d6ea | 7207 | ins.offset, BTRFS_ORDERED_REGULAR); |
0b246afa | 7208 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
5f9a8a51 | 7209 | if (IS_ERR(em)) |
2ff7e61e JM |
7210 | btrfs_free_reserved_extent(fs_info, ins.objectid, |
7211 | ins.offset, 1); | |
de0ee0ed | 7212 | |
4b46fce2 JB |
7213 | return em; |
7214 | } | |
7215 | ||
46bfbb5c CM |
7216 | /* |
7217 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
7218 | * block must be cow'd | |
7219 | */ | |
00361589 | 7220 | noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len, |
7ee9e440 JB |
7221 | u64 *orig_start, u64 *orig_block_len, |
7222 | u64 *ram_bytes) | |
46bfbb5c | 7223 | { |
2ff7e61e | 7224 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
46bfbb5c CM |
7225 | struct btrfs_path *path; |
7226 | int ret; | |
7227 | struct extent_buffer *leaf; | |
7228 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7b2b7085 | 7229 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
46bfbb5c CM |
7230 | struct btrfs_file_extent_item *fi; |
7231 | struct btrfs_key key; | |
7232 | u64 disk_bytenr; | |
7233 | u64 backref_offset; | |
7234 | u64 extent_end; | |
7235 | u64 num_bytes; | |
7236 | int slot; | |
7237 | int found_type; | |
7ee9e440 | 7238 | bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW); |
e77751aa | 7239 | |
46bfbb5c CM |
7240 | path = btrfs_alloc_path(); |
7241 | if (!path) | |
7242 | return -ENOMEM; | |
7243 | ||
f85b7379 DS |
7244 | ret = btrfs_lookup_file_extent(NULL, root, path, |
7245 | btrfs_ino(BTRFS_I(inode)), offset, 0); | |
46bfbb5c CM |
7246 | if (ret < 0) |
7247 | goto out; | |
7248 | ||
7249 | slot = path->slots[0]; | |
7250 | if (ret == 1) { | |
7251 | if (slot == 0) { | |
7252 | /* can't find the item, must cow */ | |
7253 | ret = 0; | |
7254 | goto out; | |
7255 | } | |
7256 | slot--; | |
7257 | } | |
7258 | ret = 0; | |
7259 | leaf = path->nodes[0]; | |
7260 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4a0cc7ca | 7261 | if (key.objectid != btrfs_ino(BTRFS_I(inode)) || |
46bfbb5c CM |
7262 | key.type != BTRFS_EXTENT_DATA_KEY) { |
7263 | /* not our file or wrong item type, must cow */ | |
7264 | goto out; | |
7265 | } | |
7266 | ||
7267 | if (key.offset > offset) { | |
7268 | /* Wrong offset, must cow */ | |
7269 | goto out; | |
7270 | } | |
7271 | ||
7272 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
7273 | found_type = btrfs_file_extent_type(leaf, fi); | |
7274 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
7275 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
7276 | /* not a regular extent, must cow */ | |
7277 | goto out; | |
7278 | } | |
7ee9e440 JB |
7279 | |
7280 | if (!nocow && found_type == BTRFS_FILE_EXTENT_REG) | |
7281 | goto out; | |
7282 | ||
e77751aa MX |
7283 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
7284 | if (extent_end <= offset) | |
7285 | goto out; | |
7286 | ||
46bfbb5c | 7287 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
7ee9e440 JB |
7288 | if (disk_bytenr == 0) |
7289 | goto out; | |
7290 | ||
7291 | if (btrfs_file_extent_compression(leaf, fi) || | |
7292 | btrfs_file_extent_encryption(leaf, fi) || | |
7293 | btrfs_file_extent_other_encoding(leaf, fi)) | |
7294 | goto out; | |
7295 | ||
46bfbb5c CM |
7296 | backref_offset = btrfs_file_extent_offset(leaf, fi); |
7297 | ||
7ee9e440 JB |
7298 | if (orig_start) { |
7299 | *orig_start = key.offset - backref_offset; | |
7300 | *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
7301 | *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); | |
7302 | } | |
eb384b55 | 7303 | |
2ff7e61e | 7304 | if (btrfs_extent_readonly(fs_info, disk_bytenr)) |
46bfbb5c | 7305 | goto out; |
7b2b7085 MX |
7306 | |
7307 | num_bytes = min(offset + *len, extent_end) - offset; | |
7308 | if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
7309 | u64 range_end; | |
7310 | ||
da17066c JM |
7311 | range_end = round_up(offset + num_bytes, |
7312 | root->fs_info->sectorsize) - 1; | |
7b2b7085 MX |
7313 | ret = test_range_bit(io_tree, offset, range_end, |
7314 | EXTENT_DELALLOC, 0, NULL); | |
7315 | if (ret) { | |
7316 | ret = -EAGAIN; | |
7317 | goto out; | |
7318 | } | |
7319 | } | |
7320 | ||
1bda19eb | 7321 | btrfs_release_path(path); |
46bfbb5c CM |
7322 | |
7323 | /* | |
7324 | * look for other files referencing this extent, if we | |
7325 | * find any we must cow | |
7326 | */ | |
00361589 | 7327 | |
e4c3b2dc | 7328 | ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)), |
00361589 | 7329 | key.offset - backref_offset, disk_bytenr); |
00361589 JB |
7330 | if (ret) { |
7331 | ret = 0; | |
7332 | goto out; | |
7333 | } | |
46bfbb5c CM |
7334 | |
7335 | /* | |
7336 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
7337 | * in this extent we are about to write. If there | |
7338 | * are any csums in that range we have to cow in order | |
7339 | * to keep the csums correct | |
7340 | */ | |
7341 | disk_bytenr += backref_offset; | |
7342 | disk_bytenr += offset - key.offset; | |
2ff7e61e JM |
7343 | if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes)) |
7344 | goto out; | |
46bfbb5c CM |
7345 | /* |
7346 | * all of the above have passed, it is safe to overwrite this extent | |
7347 | * without cow | |
7348 | */ | |
eb384b55 | 7349 | *len = num_bytes; |
46bfbb5c CM |
7350 | ret = 1; |
7351 | out: | |
7352 | btrfs_free_path(path); | |
7353 | return ret; | |
7354 | } | |
7355 | ||
fc4adbff AG |
7356 | bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end) |
7357 | { | |
7358 | struct radix_tree_root *root = &inode->i_mapping->page_tree; | |
7359 | int found = false; | |
7360 | void **pagep = NULL; | |
7361 | struct page *page = NULL; | |
7362 | int start_idx; | |
7363 | int end_idx; | |
7364 | ||
09cbfeaf | 7365 | start_idx = start >> PAGE_SHIFT; |
fc4adbff AG |
7366 | |
7367 | /* | |
7368 | * end is the last byte in the last page. end == start is legal | |
7369 | */ | |
09cbfeaf | 7370 | end_idx = end >> PAGE_SHIFT; |
fc4adbff AG |
7371 | |
7372 | rcu_read_lock(); | |
7373 | ||
7374 | /* Most of the code in this while loop is lifted from | |
7375 | * find_get_page. It's been modified to begin searching from a | |
7376 | * page and return just the first page found in that range. If the | |
7377 | * found idx is less than or equal to the end idx then we know that | |
7378 | * a page exists. If no pages are found or if those pages are | |
7379 | * outside of the range then we're fine (yay!) */ | |
7380 | while (page == NULL && | |
7381 | radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) { | |
7382 | page = radix_tree_deref_slot(pagep); | |
7383 | if (unlikely(!page)) | |
7384 | break; | |
7385 | ||
7386 | if (radix_tree_exception(page)) { | |
809f9016 FM |
7387 | if (radix_tree_deref_retry(page)) { |
7388 | page = NULL; | |
fc4adbff | 7389 | continue; |
809f9016 | 7390 | } |
fc4adbff AG |
7391 | /* |
7392 | * Otherwise, shmem/tmpfs must be storing a swap entry | |
7393 | * here as an exceptional entry: so return it without | |
7394 | * attempting to raise page count. | |
7395 | */ | |
6fdef6d4 | 7396 | page = NULL; |
fc4adbff AG |
7397 | break; /* TODO: Is this relevant for this use case? */ |
7398 | } | |
7399 | ||
91405151 FM |
7400 | if (!page_cache_get_speculative(page)) { |
7401 | page = NULL; | |
fc4adbff | 7402 | continue; |
91405151 | 7403 | } |
fc4adbff AG |
7404 | |
7405 | /* | |
7406 | * Has the page moved? | |
7407 | * This is part of the lockless pagecache protocol. See | |
7408 | * include/linux/pagemap.h for details. | |
7409 | */ | |
7410 | if (unlikely(page != *pagep)) { | |
09cbfeaf | 7411 | put_page(page); |
fc4adbff AG |
7412 | page = NULL; |
7413 | } | |
7414 | } | |
7415 | ||
7416 | if (page) { | |
7417 | if (page->index <= end_idx) | |
7418 | found = true; | |
09cbfeaf | 7419 | put_page(page); |
fc4adbff AG |
7420 | } |
7421 | ||
7422 | rcu_read_unlock(); | |
7423 | return found; | |
7424 | } | |
7425 | ||
eb838e73 JB |
7426 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
7427 | struct extent_state **cached_state, int writing) | |
7428 | { | |
7429 | struct btrfs_ordered_extent *ordered; | |
7430 | int ret = 0; | |
7431 | ||
7432 | while (1) { | |
7433 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
ff13db41 | 7434 | cached_state); |
eb838e73 JB |
7435 | /* |
7436 | * We're concerned with the entire range that we're going to be | |
01327610 | 7437 | * doing DIO to, so we need to make sure there's no ordered |
eb838e73 JB |
7438 | * extents in this range. |
7439 | */ | |
a776c6fa | 7440 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart, |
eb838e73 JB |
7441 | lockend - lockstart + 1); |
7442 | ||
7443 | /* | |
7444 | * We need to make sure there are no buffered pages in this | |
7445 | * range either, we could have raced between the invalidate in | |
7446 | * generic_file_direct_write and locking the extent. The | |
7447 | * invalidate needs to happen so that reads after a write do not | |
7448 | * get stale data. | |
7449 | */ | |
fc4adbff AG |
7450 | if (!ordered && |
7451 | (!writing || | |
7452 | !btrfs_page_exists_in_range(inode, lockstart, lockend))) | |
eb838e73 JB |
7453 | break; |
7454 | ||
7455 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
7456 | cached_state, GFP_NOFS); | |
7457 | ||
7458 | if (ordered) { | |
ade77029 FM |
7459 | /* |
7460 | * If we are doing a DIO read and the ordered extent we | |
7461 | * found is for a buffered write, we can not wait for it | |
7462 | * to complete and retry, because if we do so we can | |
7463 | * deadlock with concurrent buffered writes on page | |
7464 | * locks. This happens only if our DIO read covers more | |
7465 | * than one extent map, if at this point has already | |
7466 | * created an ordered extent for a previous extent map | |
7467 | * and locked its range in the inode's io tree, and a | |
7468 | * concurrent write against that previous extent map's | |
7469 | * range and this range started (we unlock the ranges | |
7470 | * in the io tree only when the bios complete and | |
7471 | * buffered writes always lock pages before attempting | |
7472 | * to lock range in the io tree). | |
7473 | */ | |
7474 | if (writing || | |
7475 | test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) | |
7476 | btrfs_start_ordered_extent(inode, ordered, 1); | |
7477 | else | |
7478 | ret = -ENOTBLK; | |
eb838e73 JB |
7479 | btrfs_put_ordered_extent(ordered); |
7480 | } else { | |
eb838e73 | 7481 | /* |
b850ae14 FM |
7482 | * We could trigger writeback for this range (and wait |
7483 | * for it to complete) and then invalidate the pages for | |
7484 | * this range (through invalidate_inode_pages2_range()), | |
7485 | * but that can lead us to a deadlock with a concurrent | |
7486 | * call to readpages() (a buffered read or a defrag call | |
7487 | * triggered a readahead) on a page lock due to an | |
7488 | * ordered dio extent we created before but did not have | |
7489 | * yet a corresponding bio submitted (whence it can not | |
7490 | * complete), which makes readpages() wait for that | |
7491 | * ordered extent to complete while holding a lock on | |
7492 | * that page. | |
eb838e73 | 7493 | */ |
b850ae14 | 7494 | ret = -ENOTBLK; |
eb838e73 JB |
7495 | } |
7496 | ||
ade77029 FM |
7497 | if (ret) |
7498 | break; | |
7499 | ||
eb838e73 JB |
7500 | cond_resched(); |
7501 | } | |
7502 | ||
7503 | return ret; | |
7504 | } | |
7505 | ||
6f9994db LB |
7506 | /* The callers of this must take lock_extent() */ |
7507 | static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len, | |
7508 | u64 orig_start, u64 block_start, | |
7509 | u64 block_len, u64 orig_block_len, | |
7510 | u64 ram_bytes, int compress_type, | |
7511 | int type) | |
69ffb543 JB |
7512 | { |
7513 | struct extent_map_tree *em_tree; | |
7514 | struct extent_map *em; | |
7515 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7516 | int ret; | |
7517 | ||
6f9994db LB |
7518 | ASSERT(type == BTRFS_ORDERED_PREALLOC || |
7519 | type == BTRFS_ORDERED_COMPRESSED || | |
7520 | type == BTRFS_ORDERED_NOCOW || | |
1af4a0aa | 7521 | type == BTRFS_ORDERED_REGULAR); |
6f9994db | 7522 | |
69ffb543 JB |
7523 | em_tree = &BTRFS_I(inode)->extent_tree; |
7524 | em = alloc_extent_map(); | |
7525 | if (!em) | |
7526 | return ERR_PTR(-ENOMEM); | |
7527 | ||
7528 | em->start = start; | |
7529 | em->orig_start = orig_start; | |
7530 | em->len = len; | |
7531 | em->block_len = block_len; | |
7532 | em->block_start = block_start; | |
7533 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
b4939680 | 7534 | em->orig_block_len = orig_block_len; |
cc95bef6 | 7535 | em->ram_bytes = ram_bytes; |
70c8a91c | 7536 | em->generation = -1; |
69ffb543 | 7537 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
1af4a0aa | 7538 | if (type == BTRFS_ORDERED_PREALLOC) { |
b11e234d | 7539 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
1af4a0aa | 7540 | } else if (type == BTRFS_ORDERED_COMPRESSED) { |
6f9994db LB |
7541 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
7542 | em->compress_type = compress_type; | |
7543 | } | |
69ffb543 JB |
7544 | |
7545 | do { | |
dcdbc059 | 7546 | btrfs_drop_extent_cache(BTRFS_I(inode), em->start, |
69ffb543 JB |
7547 | em->start + em->len - 1, 0); |
7548 | write_lock(&em_tree->lock); | |
09a2a8f9 | 7549 | ret = add_extent_mapping(em_tree, em, 1); |
69ffb543 | 7550 | write_unlock(&em_tree->lock); |
6f9994db LB |
7551 | /* |
7552 | * The caller has taken lock_extent(), who could race with us | |
7553 | * to add em? | |
7554 | */ | |
69ffb543 JB |
7555 | } while (ret == -EEXIST); |
7556 | ||
7557 | if (ret) { | |
7558 | free_extent_map(em); | |
7559 | return ERR_PTR(ret); | |
7560 | } | |
7561 | ||
6f9994db | 7562 | /* em got 2 refs now, callers needs to do free_extent_map once. */ |
69ffb543 JB |
7563 | return em; |
7564 | } | |
7565 | ||
9c9464cc FM |
7566 | static void adjust_dio_outstanding_extents(struct inode *inode, |
7567 | struct btrfs_dio_data *dio_data, | |
7568 | const u64 len) | |
7569 | { | |
823bb20a | 7570 | unsigned num_extents = count_max_extents(len); |
9c9464cc | 7571 | |
9c9464cc FM |
7572 | /* |
7573 | * If we have an outstanding_extents count still set then we're | |
7574 | * within our reservation, otherwise we need to adjust our inode | |
7575 | * counter appropriately. | |
7576 | */ | |
c2931667 | 7577 | if (dio_data->outstanding_extents >= num_extents) { |
9c9464cc FM |
7578 | dio_data->outstanding_extents -= num_extents; |
7579 | } else { | |
c2931667 LB |
7580 | /* |
7581 | * If dio write length has been split due to no large enough | |
7582 | * contiguous space, we need to compensate our inode counter | |
7583 | * appropriately. | |
7584 | */ | |
7585 | u64 num_needed = num_extents - dio_data->outstanding_extents; | |
7586 | ||
9c9464cc | 7587 | spin_lock(&BTRFS_I(inode)->lock); |
c2931667 | 7588 | BTRFS_I(inode)->outstanding_extents += num_needed; |
9c9464cc FM |
7589 | spin_unlock(&BTRFS_I(inode)->lock); |
7590 | } | |
7591 | } | |
7592 | ||
4b46fce2 JB |
7593 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
7594 | struct buffer_head *bh_result, int create) | |
7595 | { | |
0b246afa | 7596 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 7597 | struct extent_map *em; |
eb838e73 | 7598 | struct extent_state *cached_state = NULL; |
50745b0a | 7599 | struct btrfs_dio_data *dio_data = NULL; |
4b46fce2 | 7600 | u64 start = iblock << inode->i_blkbits; |
eb838e73 | 7601 | u64 lockstart, lockend; |
4b46fce2 | 7602 | u64 len = bh_result->b_size; |
eb838e73 | 7603 | int unlock_bits = EXTENT_LOCKED; |
0934856d | 7604 | int ret = 0; |
eb838e73 | 7605 | |
172a5049 | 7606 | if (create) |
3266789f | 7607 | unlock_bits |= EXTENT_DIRTY; |
172a5049 | 7608 | else |
0b246afa | 7609 | len = min_t(u64, len, fs_info->sectorsize); |
eb838e73 | 7610 | |
c329861d JB |
7611 | lockstart = start; |
7612 | lockend = start + len - 1; | |
7613 | ||
e1cbbfa5 JB |
7614 | if (current->journal_info) { |
7615 | /* | |
7616 | * Need to pull our outstanding extents and set journal_info to NULL so | |
01327610 | 7617 | * that anything that needs to check if there's a transaction doesn't get |
e1cbbfa5 JB |
7618 | * confused. |
7619 | */ | |
50745b0a | 7620 | dio_data = current->journal_info; |
e1cbbfa5 JB |
7621 | current->journal_info = NULL; |
7622 | } | |
7623 | ||
eb838e73 JB |
7624 | /* |
7625 | * If this errors out it's because we couldn't invalidate pagecache for | |
7626 | * this range and we need to fallback to buffered. | |
7627 | */ | |
9c9464cc FM |
7628 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, |
7629 | create)) { | |
7630 | ret = -ENOTBLK; | |
7631 | goto err; | |
7632 | } | |
eb838e73 | 7633 | |
fc4f21b1 | 7634 | em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0); |
eb838e73 JB |
7635 | if (IS_ERR(em)) { |
7636 | ret = PTR_ERR(em); | |
7637 | goto unlock_err; | |
7638 | } | |
4b46fce2 JB |
7639 | |
7640 | /* | |
7641 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
7642 | * io. INLINE is special, and we could probably kludge it in here, but | |
7643 | * it's still buffered so for safety lets just fall back to the generic | |
7644 | * buffered path. | |
7645 | * | |
7646 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
7647 | * decompress it, so there will be buffering required no matter what we | |
7648 | * do, so go ahead and fallback to buffered. | |
7649 | * | |
01327610 | 7650 | * We return -ENOTBLK because that's what makes DIO go ahead and go back |
4b46fce2 JB |
7651 | * to buffered IO. Don't blame me, this is the price we pay for using |
7652 | * the generic code. | |
7653 | */ | |
7654 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
7655 | em->block_start == EXTENT_MAP_INLINE) { | |
7656 | free_extent_map(em); | |
eb838e73 JB |
7657 | ret = -ENOTBLK; |
7658 | goto unlock_err; | |
4b46fce2 JB |
7659 | } |
7660 | ||
7661 | /* Just a good old fashioned hole, return */ | |
7662 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
7663 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
7664 | free_extent_map(em); | |
eb838e73 | 7665 | goto unlock_err; |
4b46fce2 JB |
7666 | } |
7667 | ||
7668 | /* | |
7669 | * We don't allocate a new extent in the following cases | |
7670 | * | |
7671 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
7672 | * existing extent. | |
7673 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
7674 | * just use the extent. | |
7675 | * | |
7676 | */ | |
46bfbb5c | 7677 | if (!create) { |
eb838e73 JB |
7678 | len = min(len, em->len - (start - em->start)); |
7679 | lockstart = start + len; | |
7680 | goto unlock; | |
46bfbb5c | 7681 | } |
4b46fce2 JB |
7682 | |
7683 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
7684 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
7685 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 | 7686 | int type; |
eb384b55 | 7687 | u64 block_start, orig_start, orig_block_len, ram_bytes; |
4b46fce2 JB |
7688 | |
7689 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7690 | type = BTRFS_ORDERED_PREALLOC; | |
7691 | else | |
7692 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 7693 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 7694 | block_start = em->block_start + (start - em->start); |
46bfbb5c | 7695 | |
00361589 | 7696 | if (can_nocow_extent(inode, start, &len, &orig_start, |
f78c436c | 7697 | &orig_block_len, &ram_bytes) == 1 && |
0b246afa | 7698 | btrfs_inc_nocow_writers(fs_info, block_start)) { |
5f9a8a51 | 7699 | struct extent_map *em2; |
0b901916 | 7700 | |
5f9a8a51 FM |
7701 | em2 = btrfs_create_dio_extent(inode, start, len, |
7702 | orig_start, block_start, | |
7703 | len, orig_block_len, | |
7704 | ram_bytes, type); | |
0b246afa | 7705 | btrfs_dec_nocow_writers(fs_info, block_start); |
69ffb543 JB |
7706 | if (type == BTRFS_ORDERED_PREALLOC) { |
7707 | free_extent_map(em); | |
5f9a8a51 | 7708 | em = em2; |
69ffb543 | 7709 | } |
5f9a8a51 FM |
7710 | if (em2 && IS_ERR(em2)) { |
7711 | ret = PTR_ERR(em2); | |
eb838e73 | 7712 | goto unlock_err; |
46bfbb5c | 7713 | } |
18513091 WX |
7714 | /* |
7715 | * For inode marked NODATACOW or extent marked PREALLOC, | |
7716 | * use the existing or preallocated extent, so does not | |
7717 | * need to adjust btrfs_space_info's bytes_may_use. | |
7718 | */ | |
7719 | btrfs_free_reserved_data_space_noquota(inode, | |
7720 | start, len); | |
46bfbb5c | 7721 | goto unlock; |
4b46fce2 | 7722 | } |
4b46fce2 | 7723 | } |
00361589 | 7724 | |
46bfbb5c CM |
7725 | /* |
7726 | * this will cow the extent, reset the len in case we changed | |
7727 | * it above | |
7728 | */ | |
7729 | len = bh_result->b_size; | |
70c8a91c JB |
7730 | free_extent_map(em); |
7731 | em = btrfs_new_extent_direct(inode, start, len); | |
eb838e73 JB |
7732 | if (IS_ERR(em)) { |
7733 | ret = PTR_ERR(em); | |
7734 | goto unlock_err; | |
7735 | } | |
46bfbb5c CM |
7736 | len = min(len, em->len - (start - em->start)); |
7737 | unlock: | |
4b46fce2 JB |
7738 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> |
7739 | inode->i_blkbits; | |
46bfbb5c | 7740 | bh_result->b_size = len; |
4b46fce2 JB |
7741 | bh_result->b_bdev = em->bdev; |
7742 | set_buffer_mapped(bh_result); | |
c3473e83 JB |
7743 | if (create) { |
7744 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7745 | set_buffer_new(bh_result); | |
7746 | ||
7747 | /* | |
7748 | * Need to update the i_size under the extent lock so buffered | |
7749 | * readers will get the updated i_size when we unlock. | |
7750 | */ | |
4aaedfb0 | 7751 | if (!dio_data->overwrite && start + len > i_size_read(inode)) |
c3473e83 | 7752 | i_size_write(inode, start + len); |
0934856d | 7753 | |
9c9464cc | 7754 | adjust_dio_outstanding_extents(inode, dio_data, len); |
50745b0a | 7755 | WARN_ON(dio_data->reserve < len); |
7756 | dio_data->reserve -= len; | |
f28a4928 | 7757 | dio_data->unsubmitted_oe_range_end = start + len; |
50745b0a | 7758 | current->journal_info = dio_data; |
c3473e83 | 7759 | } |
4b46fce2 | 7760 | |
eb838e73 JB |
7761 | /* |
7762 | * In the case of write we need to clear and unlock the entire range, | |
7763 | * in the case of read we need to unlock only the end area that we | |
7764 | * aren't using if there is any left over space. | |
7765 | */ | |
24c03fa5 | 7766 | if (lockstart < lockend) { |
0934856d MX |
7767 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, |
7768 | lockend, unlock_bits, 1, 0, | |
7769 | &cached_state, GFP_NOFS); | |
24c03fa5 | 7770 | } else { |
eb838e73 | 7771 | free_extent_state(cached_state); |
24c03fa5 | 7772 | } |
eb838e73 | 7773 | |
4b46fce2 JB |
7774 | free_extent_map(em); |
7775 | ||
7776 | return 0; | |
eb838e73 JB |
7777 | |
7778 | unlock_err: | |
eb838e73 JB |
7779 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
7780 | unlock_bits, 1, 0, &cached_state, GFP_NOFS); | |
9c9464cc | 7781 | err: |
50745b0a | 7782 | if (dio_data) |
7783 | current->journal_info = dio_data; | |
9c9464cc FM |
7784 | /* |
7785 | * Compensate the delalloc release we do in btrfs_direct_IO() when we | |
7786 | * write less data then expected, so that we don't underflow our inode's | |
7787 | * outstanding extents counter. | |
7788 | */ | |
7789 | if (create && dio_data) | |
7790 | adjust_dio_outstanding_extents(inode, dio_data, len); | |
7791 | ||
eb838e73 | 7792 | return ret; |
4b46fce2 JB |
7793 | } |
7794 | ||
8b110e39 | 7795 | static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio, |
81a75f67 | 7796 | int mirror_num) |
8b110e39 | 7797 | { |
2ff7e61e | 7798 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
8b110e39 MX |
7799 | int ret; |
7800 | ||
37226b21 | 7801 | BUG_ON(bio_op(bio) == REQ_OP_WRITE); |
8b110e39 MX |
7802 | |
7803 | bio_get(bio); | |
7804 | ||
2ff7e61e | 7805 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR); |
8b110e39 MX |
7806 | if (ret) |
7807 | goto err; | |
7808 | ||
2ff7e61e | 7809 | ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); |
8b110e39 MX |
7810 | err: |
7811 | bio_put(bio); | |
7812 | return ret; | |
7813 | } | |
7814 | ||
7815 | static int btrfs_check_dio_repairable(struct inode *inode, | |
7816 | struct bio *failed_bio, | |
7817 | struct io_failure_record *failrec, | |
7818 | int failed_mirror) | |
7819 | { | |
ab8d0fc4 | 7820 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
8b110e39 MX |
7821 | int num_copies; |
7822 | ||
ab8d0fc4 | 7823 | num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len); |
8b110e39 MX |
7824 | if (num_copies == 1) { |
7825 | /* | |
7826 | * we only have a single copy of the data, so don't bother with | |
7827 | * all the retry and error correction code that follows. no | |
7828 | * matter what the error is, it is very likely to persist. | |
7829 | */ | |
ab8d0fc4 JM |
7830 | btrfs_debug(fs_info, |
7831 | "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d", | |
7832 | num_copies, failrec->this_mirror, failed_mirror); | |
8b110e39 MX |
7833 | return 0; |
7834 | } | |
7835 | ||
7836 | failrec->failed_mirror = failed_mirror; | |
7837 | failrec->this_mirror++; | |
7838 | if (failrec->this_mirror == failed_mirror) | |
7839 | failrec->this_mirror++; | |
7840 | ||
7841 | if (failrec->this_mirror > num_copies) { | |
ab8d0fc4 JM |
7842 | btrfs_debug(fs_info, |
7843 | "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d", | |
7844 | num_copies, failrec->this_mirror, failed_mirror); | |
8b110e39 MX |
7845 | return 0; |
7846 | } | |
7847 | ||
7848 | return 1; | |
7849 | } | |
7850 | ||
7851 | static int dio_read_error(struct inode *inode, struct bio *failed_bio, | |
2dabb324 CR |
7852 | struct page *page, unsigned int pgoff, |
7853 | u64 start, u64 end, int failed_mirror, | |
7854 | bio_end_io_t *repair_endio, void *repair_arg) | |
8b110e39 MX |
7855 | { |
7856 | struct io_failure_record *failrec; | |
7857 | struct bio *bio; | |
7858 | int isector; | |
70fd7614 | 7859 | int read_mode = 0; |
8b110e39 MX |
7860 | int ret; |
7861 | ||
37226b21 | 7862 | BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE); |
8b110e39 MX |
7863 | |
7864 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
7865 | if (ret) | |
7866 | return ret; | |
7867 | ||
7868 | ret = btrfs_check_dio_repairable(inode, failed_bio, failrec, | |
7869 | failed_mirror); | |
7870 | if (!ret) { | |
4ac1f4ac | 7871 | free_io_failure(BTRFS_I(inode), failrec); |
8b110e39 MX |
7872 | return -EIO; |
7873 | } | |
7874 | ||
2dabb324 CR |
7875 | if ((failed_bio->bi_vcnt > 1) |
7876 | || (failed_bio->bi_io_vec->bv_len | |
da17066c | 7877 | > btrfs_inode_sectorsize(inode))) |
70fd7614 | 7878 | read_mode |= REQ_FAILFAST_DEV; |
8b110e39 MX |
7879 | |
7880 | isector = start - btrfs_io_bio(failed_bio)->logical; | |
7881 | isector >>= inode->i_sb->s_blocksize_bits; | |
7882 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
2dabb324 | 7883 | pgoff, isector, repair_endio, repair_arg); |
8b110e39 | 7884 | if (!bio) { |
4ac1f4ac | 7885 | free_io_failure(BTRFS_I(inode), failrec); |
8b110e39 MX |
7886 | return -EIO; |
7887 | } | |
37226b21 | 7888 | bio_set_op_attrs(bio, REQ_OP_READ, read_mode); |
8b110e39 MX |
7889 | |
7890 | btrfs_debug(BTRFS_I(inode)->root->fs_info, | |
7891 | "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n", | |
7892 | read_mode, failrec->this_mirror, failrec->in_validation); | |
7893 | ||
81a75f67 | 7894 | ret = submit_dio_repair_bio(inode, bio, failrec->this_mirror); |
8b110e39 | 7895 | if (ret) { |
4ac1f4ac | 7896 | free_io_failure(BTRFS_I(inode), failrec); |
8b110e39 MX |
7897 | bio_put(bio); |
7898 | } | |
7899 | ||
7900 | return ret; | |
7901 | } | |
7902 | ||
7903 | struct btrfs_retry_complete { | |
7904 | struct completion done; | |
7905 | struct inode *inode; | |
7906 | u64 start; | |
7907 | int uptodate; | |
7908 | }; | |
7909 | ||
4246a0b6 | 7910 | static void btrfs_retry_endio_nocsum(struct bio *bio) |
8b110e39 MX |
7911 | { |
7912 | struct btrfs_retry_complete *done = bio->bi_private; | |
2dabb324 | 7913 | struct inode *inode; |
8b110e39 MX |
7914 | struct bio_vec *bvec; |
7915 | int i; | |
7916 | ||
4246a0b6 | 7917 | if (bio->bi_error) |
8b110e39 MX |
7918 | goto end; |
7919 | ||
2dabb324 CR |
7920 | ASSERT(bio->bi_vcnt == 1); |
7921 | inode = bio->bi_io_vec->bv_page->mapping->host; | |
da17066c | 7922 | ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(inode)); |
2dabb324 | 7923 | |
8b110e39 MX |
7924 | done->uptodate = 1; |
7925 | bio_for_each_segment_all(bvec, bio, i) | |
b30cb441 | 7926 | clean_io_failure(BTRFS_I(done->inode), done->start, bvec->bv_page, 0); |
8b110e39 MX |
7927 | end: |
7928 | complete(&done->done); | |
7929 | bio_put(bio); | |
7930 | } | |
7931 | ||
7932 | static int __btrfs_correct_data_nocsum(struct inode *inode, | |
7933 | struct btrfs_io_bio *io_bio) | |
4b46fce2 | 7934 | { |
2dabb324 | 7935 | struct btrfs_fs_info *fs_info; |
2c30c71b | 7936 | struct bio_vec *bvec; |
8b110e39 | 7937 | struct btrfs_retry_complete done; |
4b46fce2 | 7938 | u64 start; |
2dabb324 CR |
7939 | unsigned int pgoff; |
7940 | u32 sectorsize; | |
7941 | int nr_sectors; | |
2c30c71b | 7942 | int i; |
c1dc0896 | 7943 | int ret; |
4b46fce2 | 7944 | |
2dabb324 | 7945 | fs_info = BTRFS_I(inode)->root->fs_info; |
da17066c | 7946 | sectorsize = fs_info->sectorsize; |
2dabb324 | 7947 | |
8b110e39 MX |
7948 | start = io_bio->logical; |
7949 | done.inode = inode; | |
7950 | ||
7951 | bio_for_each_segment_all(bvec, &io_bio->bio, i) { | |
2dabb324 CR |
7952 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len); |
7953 | pgoff = bvec->bv_offset; | |
7954 | ||
7955 | next_block_or_try_again: | |
8b110e39 MX |
7956 | done.uptodate = 0; |
7957 | done.start = start; | |
7958 | init_completion(&done.done); | |
7959 | ||
2dabb324 CR |
7960 | ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, |
7961 | pgoff, start, start + sectorsize - 1, | |
7962 | io_bio->mirror_num, | |
7963 | btrfs_retry_endio_nocsum, &done); | |
8b110e39 MX |
7964 | if (ret) |
7965 | return ret; | |
7966 | ||
7967 | wait_for_completion(&done.done); | |
7968 | ||
7969 | if (!done.uptodate) { | |
7970 | /* We might have another mirror, so try again */ | |
2dabb324 | 7971 | goto next_block_or_try_again; |
8b110e39 MX |
7972 | } |
7973 | ||
2dabb324 CR |
7974 | start += sectorsize; |
7975 | ||
7976 | if (nr_sectors--) { | |
7977 | pgoff += sectorsize; | |
7978 | goto next_block_or_try_again; | |
7979 | } | |
8b110e39 MX |
7980 | } |
7981 | ||
7982 | return 0; | |
7983 | } | |
7984 | ||
4246a0b6 | 7985 | static void btrfs_retry_endio(struct bio *bio) |
8b110e39 MX |
7986 | { |
7987 | struct btrfs_retry_complete *done = bio->bi_private; | |
7988 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
2dabb324 | 7989 | struct inode *inode; |
8b110e39 | 7990 | struct bio_vec *bvec; |
2dabb324 | 7991 | u64 start; |
8b110e39 MX |
7992 | int uptodate; |
7993 | int ret; | |
7994 | int i; | |
7995 | ||
4246a0b6 | 7996 | if (bio->bi_error) |
8b110e39 MX |
7997 | goto end; |
7998 | ||
7999 | uptodate = 1; | |
2dabb324 CR |
8000 | |
8001 | start = done->start; | |
8002 | ||
8003 | ASSERT(bio->bi_vcnt == 1); | |
8004 | inode = bio->bi_io_vec->bv_page->mapping->host; | |
da17066c | 8005 | ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(inode)); |
2dabb324 | 8006 | |
8b110e39 MX |
8007 | bio_for_each_segment_all(bvec, bio, i) { |
8008 | ret = __readpage_endio_check(done->inode, io_bio, i, | |
2dabb324 CR |
8009 | bvec->bv_page, bvec->bv_offset, |
8010 | done->start, bvec->bv_len); | |
8b110e39 | 8011 | if (!ret) |
b30cb441 | 8012 | clean_io_failure(BTRFS_I(done->inode), done->start, |
2dabb324 | 8013 | bvec->bv_page, bvec->bv_offset); |
8b110e39 MX |
8014 | else |
8015 | uptodate = 0; | |
8016 | } | |
8017 | ||
8018 | done->uptodate = uptodate; | |
8019 | end: | |
8020 | complete(&done->done); | |
8021 | bio_put(bio); | |
8022 | } | |
8023 | ||
8024 | static int __btrfs_subio_endio_read(struct inode *inode, | |
8025 | struct btrfs_io_bio *io_bio, int err) | |
8026 | { | |
2dabb324 | 8027 | struct btrfs_fs_info *fs_info; |
8b110e39 MX |
8028 | struct bio_vec *bvec; |
8029 | struct btrfs_retry_complete done; | |
8030 | u64 start; | |
8031 | u64 offset = 0; | |
2dabb324 CR |
8032 | u32 sectorsize; |
8033 | int nr_sectors; | |
8034 | unsigned int pgoff; | |
8035 | int csum_pos; | |
8b110e39 MX |
8036 | int i; |
8037 | int ret; | |
dc380aea | 8038 | |
2dabb324 | 8039 | fs_info = BTRFS_I(inode)->root->fs_info; |
da17066c | 8040 | sectorsize = fs_info->sectorsize; |
2dabb324 | 8041 | |
8b110e39 | 8042 | err = 0; |
c1dc0896 | 8043 | start = io_bio->logical; |
8b110e39 MX |
8044 | done.inode = inode; |
8045 | ||
c1dc0896 | 8046 | bio_for_each_segment_all(bvec, &io_bio->bio, i) { |
2dabb324 CR |
8047 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len); |
8048 | ||
8049 | pgoff = bvec->bv_offset; | |
8050 | next_block: | |
8051 | csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset); | |
8052 | ret = __readpage_endio_check(inode, io_bio, csum_pos, | |
8053 | bvec->bv_page, pgoff, start, | |
8054 | sectorsize); | |
8b110e39 MX |
8055 | if (likely(!ret)) |
8056 | goto next; | |
8057 | try_again: | |
8058 | done.uptodate = 0; | |
8059 | done.start = start; | |
8060 | init_completion(&done.done); | |
8061 | ||
2dabb324 CR |
8062 | ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, |
8063 | pgoff, start, start + sectorsize - 1, | |
8064 | io_bio->mirror_num, | |
8065 | btrfs_retry_endio, &done); | |
8b110e39 MX |
8066 | if (ret) { |
8067 | err = ret; | |
8068 | goto next; | |
8069 | } | |
8070 | ||
8071 | wait_for_completion(&done.done); | |
8072 | ||
8073 | if (!done.uptodate) { | |
8074 | /* We might have another mirror, so try again */ | |
8075 | goto try_again; | |
8076 | } | |
8077 | next: | |
2dabb324 CR |
8078 | offset += sectorsize; |
8079 | start += sectorsize; | |
8080 | ||
8081 | ASSERT(nr_sectors); | |
8082 | ||
8083 | if (--nr_sectors) { | |
8084 | pgoff += sectorsize; | |
8085 | goto next_block; | |
8086 | } | |
2c30c71b | 8087 | } |
c1dc0896 MX |
8088 | |
8089 | return err; | |
8090 | } | |
8091 | ||
8b110e39 MX |
8092 | static int btrfs_subio_endio_read(struct inode *inode, |
8093 | struct btrfs_io_bio *io_bio, int err) | |
8094 | { | |
8095 | bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
8096 | ||
8097 | if (skip_csum) { | |
8098 | if (unlikely(err)) | |
8099 | return __btrfs_correct_data_nocsum(inode, io_bio); | |
8100 | else | |
8101 | return 0; | |
8102 | } else { | |
8103 | return __btrfs_subio_endio_read(inode, io_bio, err); | |
8104 | } | |
8105 | } | |
8106 | ||
4246a0b6 | 8107 | static void btrfs_endio_direct_read(struct bio *bio) |
c1dc0896 MX |
8108 | { |
8109 | struct btrfs_dio_private *dip = bio->bi_private; | |
8110 | struct inode *inode = dip->inode; | |
8111 | struct bio *dio_bio; | |
8112 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
4246a0b6 | 8113 | int err = bio->bi_error; |
c1dc0896 | 8114 | |
8b110e39 MX |
8115 | if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED) |
8116 | err = btrfs_subio_endio_read(inode, io_bio, err); | |
c1dc0896 | 8117 | |
4b46fce2 | 8118 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, |
d0082371 | 8119 | dip->logical_offset + dip->bytes - 1); |
9be3395b | 8120 | dio_bio = dip->dio_bio; |
4b46fce2 | 8121 | |
4b46fce2 | 8122 | kfree(dip); |
c0da7aa1 | 8123 | |
1636d1d7 | 8124 | dio_bio->bi_error = bio->bi_error; |
4246a0b6 | 8125 | dio_end_io(dio_bio, bio->bi_error); |
23ea8e5a MX |
8126 | |
8127 | if (io_bio->end_io) | |
8128 | io_bio->end_io(io_bio, err); | |
9be3395b | 8129 | bio_put(bio); |
4b46fce2 JB |
8130 | } |
8131 | ||
14543774 FM |
8132 | static void btrfs_endio_direct_write_update_ordered(struct inode *inode, |
8133 | const u64 offset, | |
8134 | const u64 bytes, | |
8135 | const int uptodate) | |
4b46fce2 | 8136 | { |
0b246afa | 8137 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 8138 | struct btrfs_ordered_extent *ordered = NULL; |
14543774 FM |
8139 | u64 ordered_offset = offset; |
8140 | u64 ordered_bytes = bytes; | |
4b46fce2 JB |
8141 | int ret; |
8142 | ||
163cf09c CM |
8143 | again: |
8144 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
8145 | &ordered_offset, | |
4246a0b6 | 8146 | ordered_bytes, |
14543774 | 8147 | uptodate); |
4b46fce2 | 8148 | if (!ret) |
163cf09c | 8149 | goto out_test; |
4b46fce2 | 8150 | |
9e0af237 LB |
8151 | btrfs_init_work(&ordered->work, btrfs_endio_write_helper, |
8152 | finish_ordered_fn, NULL, NULL); | |
0b246afa | 8153 | btrfs_queue_work(fs_info->endio_write_workers, &ordered->work); |
163cf09c CM |
8154 | out_test: |
8155 | /* | |
8156 | * our bio might span multiple ordered extents. If we haven't | |
8157 | * completed the accounting for the whole dio, go back and try again | |
8158 | */ | |
14543774 FM |
8159 | if (ordered_offset < offset + bytes) { |
8160 | ordered_bytes = offset + bytes - ordered_offset; | |
5fd02043 | 8161 | ordered = NULL; |
163cf09c CM |
8162 | goto again; |
8163 | } | |
14543774 FM |
8164 | } |
8165 | ||
8166 | static void btrfs_endio_direct_write(struct bio *bio) | |
8167 | { | |
8168 | struct btrfs_dio_private *dip = bio->bi_private; | |
8169 | struct bio *dio_bio = dip->dio_bio; | |
8170 | ||
8171 | btrfs_endio_direct_write_update_ordered(dip->inode, | |
8172 | dip->logical_offset, | |
8173 | dip->bytes, | |
8174 | !bio->bi_error); | |
4b46fce2 | 8175 | |
4b46fce2 | 8176 | kfree(dip); |
c0da7aa1 | 8177 | |
1636d1d7 | 8178 | dio_bio->bi_error = bio->bi_error; |
4246a0b6 | 8179 | dio_end_io(dio_bio, bio->bi_error); |
9be3395b | 8180 | bio_put(bio); |
4b46fce2 JB |
8181 | } |
8182 | ||
81a75f67 | 8183 | static int __btrfs_submit_bio_start_direct_io(struct inode *inode, |
eaf25d93 CM |
8184 | struct bio *bio, int mirror_num, |
8185 | unsigned long bio_flags, u64 offset) | |
8186 | { | |
8187 | int ret; | |
2ff7e61e | 8188 | ret = btrfs_csum_one_bio(inode, bio, offset, 1); |
79787eaa | 8189 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
8190 | return 0; |
8191 | } | |
8192 | ||
4246a0b6 | 8193 | static void btrfs_end_dio_bio(struct bio *bio) |
e65e1535 MX |
8194 | { |
8195 | struct btrfs_dio_private *dip = bio->bi_private; | |
4246a0b6 | 8196 | int err = bio->bi_error; |
e65e1535 | 8197 | |
8b110e39 MX |
8198 | if (err) |
8199 | btrfs_warn(BTRFS_I(dip->inode)->root->fs_info, | |
6296b960 | 8200 | "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d", |
f85b7379 DS |
8201 | btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio), |
8202 | bio->bi_opf, | |
8b110e39 MX |
8203 | (unsigned long long)bio->bi_iter.bi_sector, |
8204 | bio->bi_iter.bi_size, err); | |
8205 | ||
8206 | if (dip->subio_endio) | |
8207 | err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err); | |
c1dc0896 MX |
8208 | |
8209 | if (err) { | |
e65e1535 MX |
8210 | dip->errors = 1; |
8211 | ||
8212 | /* | |
8213 | * before atomic variable goto zero, we must make sure | |
8214 | * dip->errors is perceived to be set. | |
8215 | */ | |
4e857c58 | 8216 | smp_mb__before_atomic(); |
e65e1535 MX |
8217 | } |
8218 | ||
8219 | /* if there are more bios still pending for this dio, just exit */ | |
8220 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
8221 | goto out; | |
8222 | ||
9be3395b | 8223 | if (dip->errors) { |
e65e1535 | 8224 | bio_io_error(dip->orig_bio); |
9be3395b | 8225 | } else { |
4246a0b6 CH |
8226 | dip->dio_bio->bi_error = 0; |
8227 | bio_endio(dip->orig_bio); | |
e65e1535 MX |
8228 | } |
8229 | out: | |
8230 | bio_put(bio); | |
8231 | } | |
8232 | ||
8233 | static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev, | |
8234 | u64 first_sector, gfp_t gfp_flags) | |
8235 | { | |
da2f0f74 | 8236 | struct bio *bio; |
22365979 | 8237 | bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags); |
da2f0f74 CM |
8238 | if (bio) |
8239 | bio_associate_current(bio); | |
8240 | return bio; | |
e65e1535 MX |
8241 | } |
8242 | ||
2ff7e61e | 8243 | static inline int btrfs_lookup_and_bind_dio_csum(struct inode *inode, |
c1dc0896 MX |
8244 | struct btrfs_dio_private *dip, |
8245 | struct bio *bio, | |
8246 | u64 file_offset) | |
8247 | { | |
8248 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
8249 | struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio); | |
8250 | int ret; | |
8251 | ||
8252 | /* | |
8253 | * We load all the csum data we need when we submit | |
8254 | * the first bio to reduce the csum tree search and | |
8255 | * contention. | |
8256 | */ | |
8257 | if (dip->logical_offset == file_offset) { | |
2ff7e61e | 8258 | ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio, |
c1dc0896 MX |
8259 | file_offset); |
8260 | if (ret) | |
8261 | return ret; | |
8262 | } | |
8263 | ||
8264 | if (bio == dip->orig_bio) | |
8265 | return 0; | |
8266 | ||
8267 | file_offset -= dip->logical_offset; | |
8268 | file_offset >>= inode->i_sb->s_blocksize_bits; | |
8269 | io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset); | |
8270 | ||
8271 | return 0; | |
8272 | } | |
8273 | ||
e65e1535 | 8274 | static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, |
81a75f67 | 8275 | u64 file_offset, int skip_sum, |
c329861d | 8276 | int async_submit) |
e65e1535 | 8277 | { |
0b246afa | 8278 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
facc8a22 | 8279 | struct btrfs_dio_private *dip = bio->bi_private; |
37226b21 | 8280 | bool write = bio_op(bio) == REQ_OP_WRITE; |
e65e1535 MX |
8281 | int ret; |
8282 | ||
b812ce28 JB |
8283 | if (async_submit) |
8284 | async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers); | |
8285 | ||
e65e1535 | 8286 | bio_get(bio); |
5fd02043 JB |
8287 | |
8288 | if (!write) { | |
0b246afa | 8289 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA); |
5fd02043 JB |
8290 | if (ret) |
8291 | goto err; | |
8292 | } | |
e65e1535 | 8293 | |
1ae39938 JB |
8294 | if (skip_sum) |
8295 | goto map; | |
8296 | ||
8297 | if (write && async_submit) { | |
0b246afa JM |
8298 | ret = btrfs_wq_submit_bio(fs_info, inode, bio, 0, 0, |
8299 | file_offset, | |
8300 | __btrfs_submit_bio_start_direct_io, | |
8301 | __btrfs_submit_bio_done); | |
e65e1535 | 8302 | goto err; |
1ae39938 JB |
8303 | } else if (write) { |
8304 | /* | |
8305 | * If we aren't doing async submit, calculate the csum of the | |
8306 | * bio now. | |
8307 | */ | |
2ff7e61e | 8308 | ret = btrfs_csum_one_bio(inode, bio, file_offset, 1); |
1ae39938 JB |
8309 | if (ret) |
8310 | goto err; | |
23ea8e5a | 8311 | } else { |
2ff7e61e | 8312 | ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio, |
c1dc0896 | 8313 | file_offset); |
c2db1073 TI |
8314 | if (ret) |
8315 | goto err; | |
8316 | } | |
1ae39938 | 8317 | map: |
2ff7e61e | 8318 | ret = btrfs_map_bio(fs_info, bio, 0, async_submit); |
e65e1535 MX |
8319 | err: |
8320 | bio_put(bio); | |
8321 | return ret; | |
8322 | } | |
8323 | ||
81a75f67 | 8324 | static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip, |
e65e1535 MX |
8325 | int skip_sum) |
8326 | { | |
8327 | struct inode *inode = dip->inode; | |
0b246afa | 8328 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e65e1535 | 8329 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e65e1535 MX |
8330 | struct bio *bio; |
8331 | struct bio *orig_bio = dip->orig_bio; | |
6a2de22f | 8332 | struct bio_vec *bvec; |
4f024f37 | 8333 | u64 start_sector = orig_bio->bi_iter.bi_sector; |
e65e1535 MX |
8334 | u64 file_offset = dip->logical_offset; |
8335 | u64 submit_len = 0; | |
8336 | u64 map_length; | |
0b246afa | 8337 | u32 blocksize = fs_info->sectorsize; |
1ae39938 | 8338 | int async_submit = 0; |
5f4dc8fc CR |
8339 | int nr_sectors; |
8340 | int ret; | |
6a2de22f | 8341 | int i, j; |
e65e1535 | 8342 | |
4f024f37 | 8343 | map_length = orig_bio->bi_iter.bi_size; |
0b246afa JM |
8344 | ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9, |
8345 | &map_length, NULL, 0); | |
7a5c3c9b | 8346 | if (ret) |
e65e1535 | 8347 | return -EIO; |
facc8a22 | 8348 | |
4f024f37 | 8349 | if (map_length >= orig_bio->bi_iter.bi_size) { |
02f57c7a | 8350 | bio = orig_bio; |
c1dc0896 | 8351 | dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED; |
02f57c7a JB |
8352 | goto submit; |
8353 | } | |
8354 | ||
53b381b3 | 8355 | /* async crcs make it difficult to collect full stripe writes. */ |
ffe2d203 | 8356 | if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 DW |
8357 | async_submit = 0; |
8358 | else | |
8359 | async_submit = 1; | |
8360 | ||
02f57c7a JB |
8361 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS); |
8362 | if (!bio) | |
8363 | return -ENOMEM; | |
7a5c3c9b | 8364 | |
ef295ecf | 8365 | bio->bi_opf = orig_bio->bi_opf; |
02f57c7a JB |
8366 | bio->bi_private = dip; |
8367 | bio->bi_end_io = btrfs_end_dio_bio; | |
c1dc0896 | 8368 | btrfs_io_bio(bio)->logical = file_offset; |
02f57c7a JB |
8369 | atomic_inc(&dip->pending_bios); |
8370 | ||
6a2de22f | 8371 | bio_for_each_segment_all(bvec, orig_bio, j) { |
0b246afa | 8372 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len); |
5f4dc8fc CR |
8373 | i = 0; |
8374 | next_block: | |
8375 | if (unlikely(map_length < submit_len + blocksize || | |
8376 | bio_add_page(bio, bvec->bv_page, blocksize, | |
8377 | bvec->bv_offset + (i * blocksize)) < blocksize)) { | |
e65e1535 MX |
8378 | /* |
8379 | * inc the count before we submit the bio so | |
8380 | * we know the end IO handler won't happen before | |
8381 | * we inc the count. Otherwise, the dip might get freed | |
8382 | * before we're done setting it up | |
8383 | */ | |
8384 | atomic_inc(&dip->pending_bios); | |
81a75f67 | 8385 | ret = __btrfs_submit_dio_bio(bio, inode, |
e65e1535 | 8386 | file_offset, skip_sum, |
c329861d | 8387 | async_submit); |
e65e1535 MX |
8388 | if (ret) { |
8389 | bio_put(bio); | |
8390 | atomic_dec(&dip->pending_bios); | |
8391 | goto out_err; | |
8392 | } | |
8393 | ||
e65e1535 MX |
8394 | start_sector += submit_len >> 9; |
8395 | file_offset += submit_len; | |
8396 | ||
8397 | submit_len = 0; | |
e65e1535 MX |
8398 | |
8399 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, | |
8400 | start_sector, GFP_NOFS); | |
8401 | if (!bio) | |
8402 | goto out_err; | |
ef295ecf | 8403 | bio->bi_opf = orig_bio->bi_opf; |
e65e1535 MX |
8404 | bio->bi_private = dip; |
8405 | bio->bi_end_io = btrfs_end_dio_bio; | |
c1dc0896 | 8406 | btrfs_io_bio(bio)->logical = file_offset; |
e65e1535 | 8407 | |
4f024f37 | 8408 | map_length = orig_bio->bi_iter.bi_size; |
0b246afa | 8409 | ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), |
3ec706c8 | 8410 | start_sector << 9, |
e65e1535 MX |
8411 | &map_length, NULL, 0); |
8412 | if (ret) { | |
8413 | bio_put(bio); | |
8414 | goto out_err; | |
8415 | } | |
5f4dc8fc CR |
8416 | |
8417 | goto next_block; | |
e65e1535 | 8418 | } else { |
5f4dc8fc CR |
8419 | submit_len += blocksize; |
8420 | if (--nr_sectors) { | |
8421 | i++; | |
8422 | goto next_block; | |
8423 | } | |
e65e1535 MX |
8424 | } |
8425 | } | |
8426 | ||
02f57c7a | 8427 | submit: |
81a75f67 | 8428 | ret = __btrfs_submit_dio_bio(bio, inode, file_offset, skip_sum, |
c329861d | 8429 | async_submit); |
e65e1535 MX |
8430 | if (!ret) |
8431 | return 0; | |
8432 | ||
8433 | bio_put(bio); | |
8434 | out_err: | |
8435 | dip->errors = 1; | |
8436 | /* | |
8437 | * before atomic variable goto zero, we must | |
8438 | * make sure dip->errors is perceived to be set. | |
8439 | */ | |
4e857c58 | 8440 | smp_mb__before_atomic(); |
e65e1535 MX |
8441 | if (atomic_dec_and_test(&dip->pending_bios)) |
8442 | bio_io_error(dip->orig_bio); | |
8443 | ||
8444 | /* bio_end_io() will handle error, so we needn't return it */ | |
8445 | return 0; | |
8446 | } | |
8447 | ||
8a4c1e42 MC |
8448 | static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode, |
8449 | loff_t file_offset) | |
4b46fce2 | 8450 | { |
61de718f FM |
8451 | struct btrfs_dio_private *dip = NULL; |
8452 | struct bio *io_bio = NULL; | |
23ea8e5a | 8453 | struct btrfs_io_bio *btrfs_bio; |
4b46fce2 | 8454 | int skip_sum; |
8a4c1e42 | 8455 | bool write = (bio_op(dio_bio) == REQ_OP_WRITE); |
4b46fce2 JB |
8456 | int ret = 0; |
8457 | ||
8458 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
8459 | ||
9be3395b | 8460 | io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS); |
9be3395b CM |
8461 | if (!io_bio) { |
8462 | ret = -ENOMEM; | |
8463 | goto free_ordered; | |
8464 | } | |
8465 | ||
c1dc0896 | 8466 | dip = kzalloc(sizeof(*dip), GFP_NOFS); |
4b46fce2 JB |
8467 | if (!dip) { |
8468 | ret = -ENOMEM; | |
61de718f | 8469 | goto free_ordered; |
4b46fce2 | 8470 | } |
4b46fce2 | 8471 | |
9be3395b | 8472 | dip->private = dio_bio->bi_private; |
4b46fce2 JB |
8473 | dip->inode = inode; |
8474 | dip->logical_offset = file_offset; | |
4f024f37 KO |
8475 | dip->bytes = dio_bio->bi_iter.bi_size; |
8476 | dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9; | |
9be3395b | 8477 | io_bio->bi_private = dip; |
9be3395b CM |
8478 | dip->orig_bio = io_bio; |
8479 | dip->dio_bio = dio_bio; | |
e65e1535 | 8480 | atomic_set(&dip->pending_bios, 0); |
c1dc0896 MX |
8481 | btrfs_bio = btrfs_io_bio(io_bio); |
8482 | btrfs_bio->logical = file_offset; | |
4b46fce2 | 8483 | |
c1dc0896 | 8484 | if (write) { |
9be3395b | 8485 | io_bio->bi_end_io = btrfs_endio_direct_write; |
c1dc0896 | 8486 | } else { |
9be3395b | 8487 | io_bio->bi_end_io = btrfs_endio_direct_read; |
c1dc0896 MX |
8488 | dip->subio_endio = btrfs_subio_endio_read; |
8489 | } | |
4b46fce2 | 8490 | |
f28a4928 FM |
8491 | /* |
8492 | * Reset the range for unsubmitted ordered extents (to a 0 length range) | |
8493 | * even if we fail to submit a bio, because in such case we do the | |
8494 | * corresponding error handling below and it must not be done a second | |
8495 | * time by btrfs_direct_IO(). | |
8496 | */ | |
8497 | if (write) { | |
8498 | struct btrfs_dio_data *dio_data = current->journal_info; | |
8499 | ||
8500 | dio_data->unsubmitted_oe_range_end = dip->logical_offset + | |
8501 | dip->bytes; | |
8502 | dio_data->unsubmitted_oe_range_start = | |
8503 | dio_data->unsubmitted_oe_range_end; | |
8504 | } | |
8505 | ||
81a75f67 | 8506 | ret = btrfs_submit_direct_hook(dip, skip_sum); |
e65e1535 | 8507 | if (!ret) |
eaf25d93 | 8508 | return; |
9be3395b | 8509 | |
23ea8e5a MX |
8510 | if (btrfs_bio->end_io) |
8511 | btrfs_bio->end_io(btrfs_bio, ret); | |
9be3395b | 8512 | |
4b46fce2 JB |
8513 | free_ordered: |
8514 | /* | |
61de718f FM |
8515 | * If we arrived here it means either we failed to submit the dip |
8516 | * or we either failed to clone the dio_bio or failed to allocate the | |
8517 | * dip. If we cloned the dio_bio and allocated the dip, we can just | |
8518 | * call bio_endio against our io_bio so that we get proper resource | |
8519 | * cleanup if we fail to submit the dip, otherwise, we must do the | |
8520 | * same as btrfs_endio_direct_[write|read] because we can't call these | |
8521 | * callbacks - they require an allocated dip and a clone of dio_bio. | |
4b46fce2 | 8522 | */ |
61de718f | 8523 | if (io_bio && dip) { |
4246a0b6 CH |
8524 | io_bio->bi_error = -EIO; |
8525 | bio_endio(io_bio); | |
61de718f FM |
8526 | /* |
8527 | * The end io callbacks free our dip, do the final put on io_bio | |
8528 | * and all the cleanup and final put for dio_bio (through | |
8529 | * dio_end_io()). | |
8530 | */ | |
8531 | dip = NULL; | |
8532 | io_bio = NULL; | |
8533 | } else { | |
14543774 FM |
8534 | if (write) |
8535 | btrfs_endio_direct_write_update_ordered(inode, | |
8536 | file_offset, | |
8537 | dio_bio->bi_iter.bi_size, | |
8538 | 0); | |
8539 | else | |
61de718f FM |
8540 | unlock_extent(&BTRFS_I(inode)->io_tree, file_offset, |
8541 | file_offset + dio_bio->bi_iter.bi_size - 1); | |
14543774 | 8542 | |
4246a0b6 | 8543 | dio_bio->bi_error = -EIO; |
61de718f FM |
8544 | /* |
8545 | * Releases and cleans up our dio_bio, no need to bio_put() | |
8546 | * nor bio_endio()/bio_io_error() against dio_bio. | |
8547 | */ | |
8548 | dio_end_io(dio_bio, ret); | |
4b46fce2 | 8549 | } |
61de718f FM |
8550 | if (io_bio) |
8551 | bio_put(io_bio); | |
8552 | kfree(dip); | |
4b46fce2 JB |
8553 | } |
8554 | ||
2ff7e61e JM |
8555 | static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info, |
8556 | struct kiocb *iocb, | |
8557 | const struct iov_iter *iter, loff_t offset) | |
5a5f79b5 CM |
8558 | { |
8559 | int seg; | |
a1b75f7d | 8560 | int i; |
0b246afa | 8561 | unsigned int blocksize_mask = fs_info->sectorsize - 1; |
5a5f79b5 | 8562 | ssize_t retval = -EINVAL; |
5a5f79b5 CM |
8563 | |
8564 | if (offset & blocksize_mask) | |
8565 | goto out; | |
8566 | ||
28060d5d AV |
8567 | if (iov_iter_alignment(iter) & blocksize_mask) |
8568 | goto out; | |
a1b75f7d | 8569 | |
28060d5d | 8570 | /* If this is a write we don't need to check anymore */ |
cd27e455 | 8571 | if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter)) |
28060d5d AV |
8572 | return 0; |
8573 | /* | |
8574 | * Check to make sure we don't have duplicate iov_base's in this | |
8575 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
8576 | * when reading back. | |
8577 | */ | |
8578 | for (seg = 0; seg < iter->nr_segs; seg++) { | |
8579 | for (i = seg + 1; i < iter->nr_segs; i++) { | |
8580 | if (iter->iov[seg].iov_base == iter->iov[i].iov_base) | |
a1b75f7d JB |
8581 | goto out; |
8582 | } | |
5a5f79b5 CM |
8583 | } |
8584 | retval = 0; | |
8585 | out: | |
8586 | return retval; | |
8587 | } | |
eb838e73 | 8588 | |
c8b8e32d | 8589 | static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
16432985 | 8590 | { |
4b46fce2 JB |
8591 | struct file *file = iocb->ki_filp; |
8592 | struct inode *inode = file->f_mapping->host; | |
0b246afa | 8593 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
50745b0a | 8594 | struct btrfs_dio_data dio_data = { 0 }; |
c8b8e32d | 8595 | loff_t offset = iocb->ki_pos; |
0934856d | 8596 | size_t count = 0; |
2e60a51e | 8597 | int flags = 0; |
38851cc1 MX |
8598 | bool wakeup = true; |
8599 | bool relock = false; | |
0934856d | 8600 | ssize_t ret; |
4b46fce2 | 8601 | |
2ff7e61e | 8602 | if (check_direct_IO(fs_info, iocb, iter, offset)) |
5a5f79b5 | 8603 | return 0; |
3f7c579c | 8604 | |
fe0f07d0 | 8605 | inode_dio_begin(inode); |
4e857c58 | 8606 | smp_mb__after_atomic(); |
38851cc1 | 8607 | |
0e267c44 | 8608 | /* |
41bd9ca4 MX |
8609 | * The generic stuff only does filemap_write_and_wait_range, which |
8610 | * isn't enough if we've written compressed pages to this area, so | |
8611 | * we need to flush the dirty pages again to make absolutely sure | |
8612 | * that any outstanding dirty pages are on disk. | |
0e267c44 | 8613 | */ |
a6cbcd4a | 8614 | count = iov_iter_count(iter); |
41bd9ca4 MX |
8615 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, |
8616 | &BTRFS_I(inode)->runtime_flags)) | |
9a025a08 WS |
8617 | filemap_fdatawrite_range(inode->i_mapping, offset, |
8618 | offset + count - 1); | |
0e267c44 | 8619 | |
6f673763 | 8620 | if (iov_iter_rw(iter) == WRITE) { |
38851cc1 MX |
8621 | /* |
8622 | * If the write DIO is beyond the EOF, we need update | |
8623 | * the isize, but it is protected by i_mutex. So we can | |
8624 | * not unlock the i_mutex at this case. | |
8625 | */ | |
8626 | if (offset + count <= inode->i_size) { | |
4aaedfb0 | 8627 | dio_data.overwrite = 1; |
5955102c | 8628 | inode_unlock(inode); |
38851cc1 MX |
8629 | relock = true; |
8630 | } | |
7cf5b976 | 8631 | ret = btrfs_delalloc_reserve_space(inode, offset, count); |
0934856d | 8632 | if (ret) |
38851cc1 | 8633 | goto out; |
823bb20a | 8634 | dio_data.outstanding_extents = count_max_extents(count); |
e1cbbfa5 JB |
8635 | |
8636 | /* | |
8637 | * We need to know how many extents we reserved so that we can | |
8638 | * do the accounting properly if we go over the number we | |
8639 | * originally calculated. Abuse current->journal_info for this. | |
8640 | */ | |
da17066c | 8641 | dio_data.reserve = round_up(count, |
0b246afa | 8642 | fs_info->sectorsize); |
f28a4928 FM |
8643 | dio_data.unsubmitted_oe_range_start = (u64)offset; |
8644 | dio_data.unsubmitted_oe_range_end = (u64)offset; | |
50745b0a | 8645 | current->journal_info = &dio_data; |
97dcdea0 | 8646 | down_read(&BTRFS_I(inode)->dio_sem); |
ee39b432 DS |
8647 | } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK, |
8648 | &BTRFS_I(inode)->runtime_flags)) { | |
fe0f07d0 | 8649 | inode_dio_end(inode); |
38851cc1 MX |
8650 | flags = DIO_LOCKING | DIO_SKIP_HOLES; |
8651 | wakeup = false; | |
0934856d MX |
8652 | } |
8653 | ||
17f8c842 | 8654 | ret = __blockdev_direct_IO(iocb, inode, |
0b246afa | 8655 | fs_info->fs_devices->latest_bdev, |
c8b8e32d | 8656 | iter, btrfs_get_blocks_direct, NULL, |
17f8c842 | 8657 | btrfs_submit_direct, flags); |
6f673763 | 8658 | if (iov_iter_rw(iter) == WRITE) { |
97dcdea0 | 8659 | up_read(&BTRFS_I(inode)->dio_sem); |
e1cbbfa5 | 8660 | current->journal_info = NULL; |
ddba1bfc | 8661 | if (ret < 0 && ret != -EIOCBQUEUED) { |
50745b0a | 8662 | if (dio_data.reserve) |
7cf5b976 QW |
8663 | btrfs_delalloc_release_space(inode, offset, |
8664 | dio_data.reserve); | |
f28a4928 FM |
8665 | /* |
8666 | * On error we might have left some ordered extents | |
8667 | * without submitting corresponding bios for them, so | |
8668 | * cleanup them up to avoid other tasks getting them | |
8669 | * and waiting for them to complete forever. | |
8670 | */ | |
8671 | if (dio_data.unsubmitted_oe_range_start < | |
8672 | dio_data.unsubmitted_oe_range_end) | |
8673 | btrfs_endio_direct_write_update_ordered(inode, | |
8674 | dio_data.unsubmitted_oe_range_start, | |
8675 | dio_data.unsubmitted_oe_range_end - | |
8676 | dio_data.unsubmitted_oe_range_start, | |
8677 | 0); | |
ddba1bfc | 8678 | } else if (ret >= 0 && (size_t)ret < count) |
7cf5b976 QW |
8679 | btrfs_delalloc_release_space(inode, offset, |
8680 | count - (size_t)ret); | |
0934856d | 8681 | } |
38851cc1 | 8682 | out: |
2e60a51e | 8683 | if (wakeup) |
fe0f07d0 | 8684 | inode_dio_end(inode); |
38851cc1 | 8685 | if (relock) |
5955102c | 8686 | inode_lock(inode); |
0934856d MX |
8687 | |
8688 | return ret; | |
16432985 CM |
8689 | } |
8690 | ||
05dadc09 TI |
8691 | #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC) |
8692 | ||
1506fcc8 YS |
8693 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
8694 | __u64 start, __u64 len) | |
8695 | { | |
05dadc09 TI |
8696 | int ret; |
8697 | ||
8698 | ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS); | |
8699 | if (ret) | |
8700 | return ret; | |
8701 | ||
ec29ed5b | 8702 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
8703 | } |
8704 | ||
a52d9a80 | 8705 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 8706 | { |
d1310b2e CM |
8707 | struct extent_io_tree *tree; |
8708 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 8709 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 8710 | } |
1832a6d5 | 8711 | |
a52d9a80 | 8712 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 8713 | { |
d1310b2e | 8714 | struct extent_io_tree *tree; |
be7bd730 JB |
8715 | struct inode *inode = page->mapping->host; |
8716 | int ret; | |
b888db2b CM |
8717 | |
8718 | if (current->flags & PF_MEMALLOC) { | |
8719 | redirty_page_for_writepage(wbc, page); | |
8720 | unlock_page(page); | |
8721 | return 0; | |
8722 | } | |
be7bd730 JB |
8723 | |
8724 | /* | |
8725 | * If we are under memory pressure we will call this directly from the | |
8726 | * VM, we need to make sure we have the inode referenced for the ordered | |
8727 | * extent. If not just return like we didn't do anything. | |
8728 | */ | |
8729 | if (!igrab(inode)) { | |
8730 | redirty_page_for_writepage(wbc, page); | |
8731 | return AOP_WRITEPAGE_ACTIVATE; | |
8732 | } | |
d1310b2e | 8733 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
be7bd730 JB |
8734 | ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
8735 | btrfs_add_delayed_iput(inode); | |
8736 | return ret; | |
9ebefb18 CM |
8737 | } |
8738 | ||
48a3b636 ES |
8739 | static int btrfs_writepages(struct address_space *mapping, |
8740 | struct writeback_control *wbc) | |
b293f02e | 8741 | { |
d1310b2e | 8742 | struct extent_io_tree *tree; |
771ed689 | 8743 | |
d1310b2e | 8744 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
8745 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
8746 | } | |
8747 | ||
3ab2fb5a CM |
8748 | static int |
8749 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
8750 | struct list_head *pages, unsigned nr_pages) | |
8751 | { | |
d1310b2e CM |
8752 | struct extent_io_tree *tree; |
8753 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
8754 | return extent_readpages(tree, mapping, pages, nr_pages, |
8755 | btrfs_get_extent); | |
8756 | } | |
e6dcd2dc | 8757 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 8758 | { |
d1310b2e CM |
8759 | struct extent_io_tree *tree; |
8760 | struct extent_map_tree *map; | |
a52d9a80 | 8761 | int ret; |
8c2383c3 | 8762 | |
d1310b2e CM |
8763 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
8764 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 8765 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
8766 | if (ret == 1) { |
8767 | ClearPagePrivate(page); | |
8768 | set_page_private(page, 0); | |
09cbfeaf | 8769 | put_page(page); |
39279cc3 | 8770 | } |
a52d9a80 | 8771 | return ret; |
39279cc3 CM |
8772 | } |
8773 | ||
e6dcd2dc CM |
8774 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
8775 | { | |
98509cfc CM |
8776 | if (PageWriteback(page) || PageDirty(page)) |
8777 | return 0; | |
3ba7ab22 | 8778 | return __btrfs_releasepage(page, gfp_flags); |
e6dcd2dc CM |
8779 | } |
8780 | ||
d47992f8 LC |
8781 | static void btrfs_invalidatepage(struct page *page, unsigned int offset, |
8782 | unsigned int length) | |
39279cc3 | 8783 | { |
5fd02043 | 8784 | struct inode *inode = page->mapping->host; |
d1310b2e | 8785 | struct extent_io_tree *tree; |
e6dcd2dc | 8786 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 8787 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 8788 | u64 page_start = page_offset(page); |
09cbfeaf | 8789 | u64 page_end = page_start + PAGE_SIZE - 1; |
dbfdb6d1 CR |
8790 | u64 start; |
8791 | u64 end; | |
131e404a | 8792 | int inode_evicting = inode->i_state & I_FREEING; |
39279cc3 | 8793 | |
8b62b72b CM |
8794 | /* |
8795 | * we have the page locked, so new writeback can't start, | |
8796 | * and the dirty bit won't be cleared while we are here. | |
8797 | * | |
8798 | * Wait for IO on this page so that we can safely clear | |
8799 | * the PagePrivate2 bit and do ordered accounting | |
8800 | */ | |
e6dcd2dc | 8801 | wait_on_page_writeback(page); |
8b62b72b | 8802 | |
5fd02043 | 8803 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
8804 | if (offset) { |
8805 | btrfs_releasepage(page, GFP_NOFS); | |
8806 | return; | |
8807 | } | |
131e404a FDBM |
8808 | |
8809 | if (!inode_evicting) | |
ff13db41 | 8810 | lock_extent_bits(tree, page_start, page_end, &cached_state); |
dbfdb6d1 CR |
8811 | again: |
8812 | start = page_start; | |
a776c6fa | 8813 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start, |
dbfdb6d1 | 8814 | page_end - start + 1); |
e6dcd2dc | 8815 | if (ordered) { |
dbfdb6d1 | 8816 | end = min(page_end, ordered->file_offset + ordered->len - 1); |
eb84ae03 CM |
8817 | /* |
8818 | * IO on this page will never be started, so we need | |
8819 | * to account for any ordered extents now | |
8820 | */ | |
131e404a | 8821 | if (!inode_evicting) |
dbfdb6d1 | 8822 | clear_extent_bit(tree, start, end, |
131e404a FDBM |
8823 | EXTENT_DIRTY | EXTENT_DELALLOC | |
8824 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | | |
8825 | EXTENT_DEFRAG, 1, 0, &cached_state, | |
8826 | GFP_NOFS); | |
8b62b72b CM |
8827 | /* |
8828 | * whoever cleared the private bit is responsible | |
8829 | * for the finish_ordered_io | |
8830 | */ | |
77cef2ec JB |
8831 | if (TestClearPagePrivate2(page)) { |
8832 | struct btrfs_ordered_inode_tree *tree; | |
8833 | u64 new_len; | |
8834 | ||
8835 | tree = &BTRFS_I(inode)->ordered_tree; | |
8836 | ||
8837 | spin_lock_irq(&tree->lock); | |
8838 | set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags); | |
dbfdb6d1 | 8839 | new_len = start - ordered->file_offset; |
77cef2ec JB |
8840 | if (new_len < ordered->truncated_len) |
8841 | ordered->truncated_len = new_len; | |
8842 | spin_unlock_irq(&tree->lock); | |
8843 | ||
8844 | if (btrfs_dec_test_ordered_pending(inode, &ordered, | |
dbfdb6d1 CR |
8845 | start, |
8846 | end - start + 1, 1)) | |
77cef2ec | 8847 | btrfs_finish_ordered_io(ordered); |
8b62b72b | 8848 | } |
e6dcd2dc | 8849 | btrfs_put_ordered_extent(ordered); |
131e404a FDBM |
8850 | if (!inode_evicting) { |
8851 | cached_state = NULL; | |
dbfdb6d1 | 8852 | lock_extent_bits(tree, start, end, |
131e404a FDBM |
8853 | &cached_state); |
8854 | } | |
dbfdb6d1 CR |
8855 | |
8856 | start = end + 1; | |
8857 | if (start < page_end) | |
8858 | goto again; | |
131e404a FDBM |
8859 | } |
8860 | ||
b9d0b389 QW |
8861 | /* |
8862 | * Qgroup reserved space handler | |
8863 | * Page here will be either | |
8864 | * 1) Already written to disk | |
8865 | * In this case, its reserved space is released from data rsv map | |
8866 | * and will be freed by delayed_ref handler finally. | |
8867 | * So even we call qgroup_free_data(), it won't decrease reserved | |
8868 | * space. | |
8869 | * 2) Not written to disk | |
0b34c261 GR |
8870 | * This means the reserved space should be freed here. However, |
8871 | * if a truncate invalidates the page (by clearing PageDirty) | |
8872 | * and the page is accounted for while allocating extent | |
8873 | * in btrfs_check_data_free_space() we let delayed_ref to | |
8874 | * free the entire extent. | |
b9d0b389 | 8875 | */ |
0b34c261 GR |
8876 | if (PageDirty(page)) |
8877 | btrfs_qgroup_free_data(inode, page_start, PAGE_SIZE); | |
131e404a FDBM |
8878 | if (!inode_evicting) { |
8879 | clear_extent_bit(tree, page_start, page_end, | |
8880 | EXTENT_LOCKED | EXTENT_DIRTY | | |
8881 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
8882 | EXTENT_DEFRAG, 1, 1, | |
8883 | &cached_state, GFP_NOFS); | |
8884 | ||
8885 | __btrfs_releasepage(page, GFP_NOFS); | |
e6dcd2dc | 8886 | } |
e6dcd2dc | 8887 | |
4a096752 | 8888 | ClearPageChecked(page); |
9ad6b7bc | 8889 | if (PagePrivate(page)) { |
9ad6b7bc CM |
8890 | ClearPagePrivate(page); |
8891 | set_page_private(page, 0); | |
09cbfeaf | 8892 | put_page(page); |
9ad6b7bc | 8893 | } |
39279cc3 CM |
8894 | } |
8895 | ||
9ebefb18 CM |
8896 | /* |
8897 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
8898 | * called from a page fault handler when a page is first dirtied. Hence we must | |
8899 | * be careful to check for EOF conditions here. We set the page up correctly | |
8900 | * for a written page which means we get ENOSPC checking when writing into | |
8901 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
8902 | * support these features. | |
8903 | * | |
8904 | * We are not allowed to take the i_mutex here so we have to play games to | |
8905 | * protect against truncate races as the page could now be beyond EOF. Because | |
8906 | * vmtruncate() writes the inode size before removing pages, once we have the | |
8907 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
8908 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
8909 | * unlock the page. | |
8910 | */ | |
11bac800 | 8911 | int btrfs_page_mkwrite(struct vm_fault *vmf) |
9ebefb18 | 8912 | { |
c2ec175c | 8913 | struct page *page = vmf->page; |
11bac800 | 8914 | struct inode *inode = file_inode(vmf->vma->vm_file); |
0b246afa | 8915 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc CM |
8916 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
8917 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 8918 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
8919 | char *kaddr; |
8920 | unsigned long zero_start; | |
9ebefb18 | 8921 | loff_t size; |
1832a6d5 | 8922 | int ret; |
9998eb70 | 8923 | int reserved = 0; |
d0b7da88 | 8924 | u64 reserved_space; |
a52d9a80 | 8925 | u64 page_start; |
e6dcd2dc | 8926 | u64 page_end; |
d0b7da88 CR |
8927 | u64 end; |
8928 | ||
09cbfeaf | 8929 | reserved_space = PAGE_SIZE; |
9ebefb18 | 8930 | |
b2b5ef5c | 8931 | sb_start_pagefault(inode->i_sb); |
df480633 | 8932 | page_start = page_offset(page); |
09cbfeaf | 8933 | page_end = page_start + PAGE_SIZE - 1; |
d0b7da88 | 8934 | end = page_end; |
df480633 | 8935 | |
d0b7da88 CR |
8936 | /* |
8937 | * Reserving delalloc space after obtaining the page lock can lead to | |
8938 | * deadlock. For example, if a dirty page is locked by this function | |
8939 | * and the call to btrfs_delalloc_reserve_space() ends up triggering | |
8940 | * dirty page write out, then the btrfs_writepage() function could | |
8941 | * end up waiting indefinitely to get a lock on the page currently | |
8942 | * being processed by btrfs_page_mkwrite() function. | |
8943 | */ | |
7cf5b976 | 8944 | ret = btrfs_delalloc_reserve_space(inode, page_start, |
d0b7da88 | 8945 | reserved_space); |
9998eb70 | 8946 | if (!ret) { |
11bac800 | 8947 | ret = file_update_time(vmf->vma->vm_file); |
9998eb70 CM |
8948 | reserved = 1; |
8949 | } | |
56a76f82 NP |
8950 | if (ret) { |
8951 | if (ret == -ENOMEM) | |
8952 | ret = VM_FAULT_OOM; | |
8953 | else /* -ENOSPC, -EIO, etc */ | |
8954 | ret = VM_FAULT_SIGBUS; | |
9998eb70 CM |
8955 | if (reserved) |
8956 | goto out; | |
8957 | goto out_noreserve; | |
56a76f82 | 8958 | } |
1832a6d5 | 8959 | |
56a76f82 | 8960 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 8961 | again: |
9ebefb18 | 8962 | lock_page(page); |
9ebefb18 | 8963 | size = i_size_read(inode); |
a52d9a80 | 8964 | |
9ebefb18 | 8965 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 8966 | (page_start >= size)) { |
9ebefb18 CM |
8967 | /* page got truncated out from underneath us */ |
8968 | goto out_unlock; | |
8969 | } | |
e6dcd2dc CM |
8970 | wait_on_page_writeback(page); |
8971 | ||
ff13db41 | 8972 | lock_extent_bits(io_tree, page_start, page_end, &cached_state); |
e6dcd2dc CM |
8973 | set_page_extent_mapped(page); |
8974 | ||
eb84ae03 CM |
8975 | /* |
8976 | * we can't set the delalloc bits if there are pending ordered | |
8977 | * extents. Drop our locks and wait for them to finish | |
8978 | */ | |
a776c6fa NB |
8979 | ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start, |
8980 | PAGE_SIZE); | |
e6dcd2dc | 8981 | if (ordered) { |
2ac55d41 JB |
8982 | unlock_extent_cached(io_tree, page_start, page_end, |
8983 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 8984 | unlock_page(page); |
eb84ae03 | 8985 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
8986 | btrfs_put_ordered_extent(ordered); |
8987 | goto again; | |
8988 | } | |
8989 | ||
09cbfeaf | 8990 | if (page->index == ((size - 1) >> PAGE_SHIFT)) { |
da17066c | 8991 | reserved_space = round_up(size - page_start, |
0b246afa | 8992 | fs_info->sectorsize); |
09cbfeaf | 8993 | if (reserved_space < PAGE_SIZE) { |
d0b7da88 CR |
8994 | end = page_start + reserved_space - 1; |
8995 | spin_lock(&BTRFS_I(inode)->lock); | |
8996 | BTRFS_I(inode)->outstanding_extents++; | |
8997 | spin_unlock(&BTRFS_I(inode)->lock); | |
8998 | btrfs_delalloc_release_space(inode, page_start, | |
09cbfeaf | 8999 | PAGE_SIZE - reserved_space); |
d0b7da88 CR |
9000 | } |
9001 | } | |
9002 | ||
fbf19087 | 9003 | /* |
5416034f LB |
9004 | * page_mkwrite gets called when the page is firstly dirtied after it's |
9005 | * faulted in, but write(2) could also dirty a page and set delalloc | |
9006 | * bits, thus in this case for space account reason, we still need to | |
9007 | * clear any delalloc bits within this page range since we have to | |
9008 | * reserve data&meta space before lock_page() (see above comments). | |
fbf19087 | 9009 | */ |
d0b7da88 | 9010 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end, |
9e8a4a8b LB |
9011 | EXTENT_DIRTY | EXTENT_DELALLOC | |
9012 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 9013 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 9014 | |
d0b7da88 | 9015 | ret = btrfs_set_extent_delalloc(inode, page_start, end, |
ba8b04c1 | 9016 | &cached_state, 0); |
9ed74f2d | 9017 | if (ret) { |
2ac55d41 JB |
9018 | unlock_extent_cached(io_tree, page_start, page_end, |
9019 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
9020 | ret = VM_FAULT_SIGBUS; |
9021 | goto out_unlock; | |
9022 | } | |
e6dcd2dc | 9023 | ret = 0; |
9ebefb18 CM |
9024 | |
9025 | /* page is wholly or partially inside EOF */ | |
09cbfeaf KS |
9026 | if (page_start + PAGE_SIZE > size) |
9027 | zero_start = size & ~PAGE_MASK; | |
9ebefb18 | 9028 | else |
09cbfeaf | 9029 | zero_start = PAGE_SIZE; |
9ebefb18 | 9030 | |
09cbfeaf | 9031 | if (zero_start != PAGE_SIZE) { |
e6dcd2dc | 9032 | kaddr = kmap(page); |
09cbfeaf | 9033 | memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start); |
e6dcd2dc CM |
9034 | flush_dcache_page(page); |
9035 | kunmap(page); | |
9036 | } | |
247e743c | 9037 | ClearPageChecked(page); |
e6dcd2dc | 9038 | set_page_dirty(page); |
50a9b214 | 9039 | SetPageUptodate(page); |
5a3f23d5 | 9040 | |
0b246afa | 9041 | BTRFS_I(inode)->last_trans = fs_info->generation; |
257c62e1 | 9042 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; |
46d8bc34 | 9043 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit; |
257c62e1 | 9044 | |
2ac55d41 | 9045 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
9046 | |
9047 | out_unlock: | |
b2b5ef5c JK |
9048 | if (!ret) { |
9049 | sb_end_pagefault(inode->i_sb); | |
50a9b214 | 9050 | return VM_FAULT_LOCKED; |
b2b5ef5c | 9051 | } |
9ebefb18 | 9052 | unlock_page(page); |
1832a6d5 | 9053 | out: |
d0b7da88 | 9054 | btrfs_delalloc_release_space(inode, page_start, reserved_space); |
9998eb70 | 9055 | out_noreserve: |
b2b5ef5c | 9056 | sb_end_pagefault(inode->i_sb); |
9ebefb18 CM |
9057 | return ret; |
9058 | } | |
9059 | ||
a41ad394 | 9060 | static int btrfs_truncate(struct inode *inode) |
39279cc3 | 9061 | { |
0b246afa | 9062 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 9063 | struct btrfs_root *root = BTRFS_I(inode)->root; |
fcb80c2a | 9064 | struct btrfs_block_rsv *rsv; |
a71754fc | 9065 | int ret = 0; |
3893e33b | 9066 | int err = 0; |
39279cc3 | 9067 | struct btrfs_trans_handle *trans; |
0b246afa JM |
9068 | u64 mask = fs_info->sectorsize - 1; |
9069 | u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1); | |
39279cc3 | 9070 | |
0ef8b726 JB |
9071 | ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask), |
9072 | (u64)-1); | |
9073 | if (ret) | |
9074 | return ret; | |
39279cc3 | 9075 | |
fcb80c2a | 9076 | /* |
01327610 | 9077 | * Yes ladies and gentlemen, this is indeed ugly. The fact is we have |
fcb80c2a JB |
9078 | * 3 things going on here |
9079 | * | |
9080 | * 1) We need to reserve space for our orphan item and the space to | |
9081 | * delete our orphan item. Lord knows we don't want to have a dangling | |
9082 | * orphan item because we didn't reserve space to remove it. | |
9083 | * | |
9084 | * 2) We need to reserve space to update our inode. | |
9085 | * | |
9086 | * 3) We need to have something to cache all the space that is going to | |
9087 | * be free'd up by the truncate operation, but also have some slack | |
9088 | * space reserved in case it uses space during the truncate (thank you | |
9089 | * very much snapshotting). | |
9090 | * | |
01327610 | 9091 | * And we need these to all be separate. The fact is we can use a lot of |
fcb80c2a | 9092 | * space doing the truncate, and we have no earthly idea how much space |
01327610 | 9093 | * we will use, so we need the truncate reservation to be separate so it |
fcb80c2a JB |
9094 | * doesn't end up using space reserved for updating the inode or |
9095 | * removing the orphan item. We also need to be able to stop the | |
9096 | * transaction and start a new one, which means we need to be able to | |
9097 | * update the inode several times, and we have no idea of knowing how | |
9098 | * many times that will be, so we can't just reserve 1 item for the | |
01327610 | 9099 | * entirety of the operation, so that has to be done separately as well. |
fcb80c2a JB |
9100 | * Then there is the orphan item, which does indeed need to be held on |
9101 | * to for the whole operation, and we need nobody to touch this reserved | |
9102 | * space except the orphan code. | |
9103 | * | |
9104 | * So that leaves us with | |
9105 | * | |
9106 | * 1) root->orphan_block_rsv - for the orphan deletion. | |
9107 | * 2) rsv - for the truncate reservation, which we will steal from the | |
9108 | * transaction reservation. | |
9109 | * 3) fs_info->trans_block_rsv - this will have 1 items worth left for | |
9110 | * updating the inode. | |
9111 | */ | |
2ff7e61e | 9112 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
9113 | if (!rsv) |
9114 | return -ENOMEM; | |
4a338542 | 9115 | rsv->size = min_size; |
ca7e70f5 | 9116 | rsv->failfast = 1; |
f0cd846e | 9117 | |
907cbceb | 9118 | /* |
07127184 | 9119 | * 1 for the truncate slack space |
907cbceb JB |
9120 | * 1 for updating the inode. |
9121 | */ | |
f3fe820c | 9122 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a JB |
9123 | if (IS_ERR(trans)) { |
9124 | err = PTR_ERR(trans); | |
9125 | goto out; | |
9126 | } | |
f0cd846e | 9127 | |
907cbceb | 9128 | /* Migrate the slack space for the truncate to our reserve */ |
0b246afa | 9129 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv, |
25d609f8 | 9130 | min_size, 0); |
fcb80c2a | 9131 | BUG_ON(ret); |
f0cd846e | 9132 | |
5dc562c5 JB |
9133 | /* |
9134 | * So if we truncate and then write and fsync we normally would just | |
9135 | * write the extents that changed, which is a problem if we need to | |
9136 | * first truncate that entire inode. So set this flag so we write out | |
9137 | * all of the extents in the inode to the sync log so we're completely | |
9138 | * safe. | |
9139 | */ | |
9140 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
ca7e70f5 | 9141 | trans->block_rsv = rsv; |
907cbceb | 9142 | |
8082510e YZ |
9143 | while (1) { |
9144 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
9145 | inode->i_size, | |
9146 | BTRFS_EXTENT_DATA_KEY); | |
28ed1345 | 9147 | if (ret != -ENOSPC && ret != -EAGAIN) { |
3893e33b | 9148 | err = ret; |
8082510e | 9149 | break; |
3893e33b | 9150 | } |
39279cc3 | 9151 | |
0b246afa | 9152 | trans->block_rsv = &fs_info->trans_block_rsv; |
8082510e | 9153 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
9154 | if (ret) { |
9155 | err = ret; | |
9156 | break; | |
9157 | } | |
ca7e70f5 | 9158 | |
3a45bb20 | 9159 | btrfs_end_transaction(trans); |
2ff7e61e | 9160 | btrfs_btree_balance_dirty(fs_info); |
ca7e70f5 JB |
9161 | |
9162 | trans = btrfs_start_transaction(root, 2); | |
9163 | if (IS_ERR(trans)) { | |
9164 | ret = err = PTR_ERR(trans); | |
9165 | trans = NULL; | |
9166 | break; | |
9167 | } | |
9168 | ||
47b5d646 | 9169 | btrfs_block_rsv_release(fs_info, rsv, -1); |
0b246afa | 9170 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, |
25d609f8 | 9171 | rsv, min_size, 0); |
ca7e70f5 JB |
9172 | BUG_ON(ret); /* shouldn't happen */ |
9173 | trans->block_rsv = rsv; | |
8082510e YZ |
9174 | } |
9175 | ||
9176 | if (ret == 0 && inode->i_nlink > 0) { | |
fcb80c2a | 9177 | trans->block_rsv = root->orphan_block_rsv; |
3d6ae7bb | 9178 | ret = btrfs_orphan_del(trans, BTRFS_I(inode)); |
3893e33b JB |
9179 | if (ret) |
9180 | err = ret; | |
8082510e YZ |
9181 | } |
9182 | ||
917c16b2 | 9183 | if (trans) { |
0b246afa | 9184 | trans->block_rsv = &fs_info->trans_block_rsv; |
917c16b2 CM |
9185 | ret = btrfs_update_inode(trans, root, inode); |
9186 | if (ret && !err) | |
9187 | err = ret; | |
7b128766 | 9188 | |
3a45bb20 | 9189 | ret = btrfs_end_transaction(trans); |
2ff7e61e | 9190 | btrfs_btree_balance_dirty(fs_info); |
917c16b2 | 9191 | } |
fcb80c2a | 9192 | out: |
2ff7e61e | 9193 | btrfs_free_block_rsv(fs_info, rsv); |
fcb80c2a | 9194 | |
3893e33b JB |
9195 | if (ret && !err) |
9196 | err = ret; | |
a41ad394 | 9197 | |
3893e33b | 9198 | return err; |
39279cc3 CM |
9199 | } |
9200 | ||
d352ac68 CM |
9201 | /* |
9202 | * create a new subvolume directory/inode (helper for the ioctl). | |
9203 | */ | |
d2fb3437 | 9204 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
63541927 FDBM |
9205 | struct btrfs_root *new_root, |
9206 | struct btrfs_root *parent_root, | |
9207 | u64 new_dirid) | |
39279cc3 | 9208 | { |
39279cc3 | 9209 | struct inode *inode; |
76dda93c | 9210 | int err; |
00e4e6b3 | 9211 | u64 index = 0; |
39279cc3 | 9212 | |
12fc9d09 FA |
9213 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
9214 | new_dirid, new_dirid, | |
9215 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
9216 | &index); | |
54aa1f4d | 9217 | if (IS_ERR(inode)) |
f46b5a66 | 9218 | return PTR_ERR(inode); |
39279cc3 CM |
9219 | inode->i_op = &btrfs_dir_inode_operations; |
9220 | inode->i_fop = &btrfs_dir_file_operations; | |
9221 | ||
bfe86848 | 9222 | set_nlink(inode, 1); |
6ef06d27 | 9223 | btrfs_i_size_write(BTRFS_I(inode), 0); |
b0d5d10f | 9224 | unlock_new_inode(inode); |
3b96362c | 9225 | |
63541927 FDBM |
9226 | err = btrfs_subvol_inherit_props(trans, new_root, parent_root); |
9227 | if (err) | |
9228 | btrfs_err(new_root->fs_info, | |
351fd353 | 9229 | "error inheriting subvolume %llu properties: %d", |
63541927 FDBM |
9230 | new_root->root_key.objectid, err); |
9231 | ||
76dda93c | 9232 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 9233 | |
76dda93c | 9234 | iput(inode); |
ce598979 | 9235 | return err; |
39279cc3 CM |
9236 | } |
9237 | ||
39279cc3 CM |
9238 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
9239 | { | |
9240 | struct btrfs_inode *ei; | |
2ead6ae7 | 9241 | struct inode *inode; |
39279cc3 CM |
9242 | |
9243 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
9244 | if (!ei) | |
9245 | return NULL; | |
2ead6ae7 YZ |
9246 | |
9247 | ei->root = NULL; | |
2ead6ae7 | 9248 | ei->generation = 0; |
15ee9bc7 | 9249 | ei->last_trans = 0; |
257c62e1 | 9250 | ei->last_sub_trans = 0; |
e02119d5 | 9251 | ei->logged_trans = 0; |
2ead6ae7 | 9252 | ei->delalloc_bytes = 0; |
47059d93 | 9253 | ei->defrag_bytes = 0; |
2ead6ae7 YZ |
9254 | ei->disk_i_size = 0; |
9255 | ei->flags = 0; | |
7709cde3 | 9256 | ei->csum_bytes = 0; |
2ead6ae7 | 9257 | ei->index_cnt = (u64)-1; |
67de1176 | 9258 | ei->dir_index = 0; |
2ead6ae7 | 9259 | ei->last_unlink_trans = 0; |
46d8bc34 | 9260 | ei->last_log_commit = 0; |
8089fe62 | 9261 | ei->delayed_iput_count = 0; |
2ead6ae7 | 9262 | |
9e0baf60 JB |
9263 | spin_lock_init(&ei->lock); |
9264 | ei->outstanding_extents = 0; | |
9265 | ei->reserved_extents = 0; | |
2ead6ae7 | 9266 | |
72ac3c0d | 9267 | ei->runtime_flags = 0; |
261507a0 | 9268 | ei->force_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 9269 | |
16cdcec7 MX |
9270 | ei->delayed_node = NULL; |
9271 | ||
9cc97d64 | 9272 | ei->i_otime.tv_sec = 0; |
9273 | ei->i_otime.tv_nsec = 0; | |
9274 | ||
2ead6ae7 | 9275 | inode = &ei->vfs_inode; |
a8067e02 | 9276 | extent_map_tree_init(&ei->extent_tree); |
f993c883 DS |
9277 | extent_io_tree_init(&ei->io_tree, &inode->i_data); |
9278 | extent_io_tree_init(&ei->io_failure_tree, &inode->i_data); | |
0b32f4bb JB |
9279 | ei->io_tree.track_uptodate = 1; |
9280 | ei->io_failure_tree.track_uptodate = 1; | |
b812ce28 | 9281 | atomic_set(&ei->sync_writers, 0); |
2ead6ae7 | 9282 | mutex_init(&ei->log_mutex); |
f248679e | 9283 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 9284 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 9285 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
8089fe62 | 9286 | INIT_LIST_HEAD(&ei->delayed_iput); |
2ead6ae7 | 9287 | RB_CLEAR_NODE(&ei->rb_node); |
5f9a8a51 | 9288 | init_rwsem(&ei->dio_sem); |
2ead6ae7 YZ |
9289 | |
9290 | return inode; | |
39279cc3 CM |
9291 | } |
9292 | ||
aaedb55b JB |
9293 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
9294 | void btrfs_test_destroy_inode(struct inode *inode) | |
9295 | { | |
dcdbc059 | 9296 | btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0); |
aaedb55b JB |
9297 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
9298 | } | |
9299 | #endif | |
9300 | ||
fa0d7e3d NP |
9301 | static void btrfs_i_callback(struct rcu_head *head) |
9302 | { | |
9303 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
9304 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
9305 | } | |
9306 | ||
39279cc3 CM |
9307 | void btrfs_destroy_inode(struct inode *inode) |
9308 | { | |
0b246afa | 9309 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc | 9310 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
9311 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9312 | ||
b3d9b7a3 | 9313 | WARN_ON(!hlist_empty(&inode->i_dentry)); |
39279cc3 | 9314 | WARN_ON(inode->i_data.nrpages); |
9e0baf60 JB |
9315 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
9316 | WARN_ON(BTRFS_I(inode)->reserved_extents); | |
7709cde3 JB |
9317 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
9318 | WARN_ON(BTRFS_I(inode)->csum_bytes); | |
47059d93 | 9319 | WARN_ON(BTRFS_I(inode)->defrag_bytes); |
39279cc3 | 9320 | |
a6dbd429 JB |
9321 | /* |
9322 | * This can happen where we create an inode, but somebody else also | |
9323 | * created the same inode and we need to destroy the one we already | |
9324 | * created. | |
9325 | */ | |
9326 | if (!root) | |
9327 | goto free; | |
9328 | ||
8a35d95f JB |
9329 | if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
9330 | &BTRFS_I(inode)->runtime_flags)) { | |
0b246afa | 9331 | btrfs_info(fs_info, "inode %llu still on the orphan list", |
4a0cc7ca | 9332 | btrfs_ino(BTRFS_I(inode))); |
8a35d95f | 9333 | atomic_dec(&root->orphan_inodes); |
7b128766 | 9334 | } |
7b128766 | 9335 | |
d397712b | 9336 | while (1) { |
e6dcd2dc CM |
9337 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
9338 | if (!ordered) | |
9339 | break; | |
9340 | else { | |
0b246afa | 9341 | btrfs_err(fs_info, |
5d163e0e JM |
9342 | "found ordered extent %llu %llu on inode cleanup", |
9343 | ordered->file_offset, ordered->len); | |
e6dcd2dc CM |
9344 | btrfs_remove_ordered_extent(inode, ordered); |
9345 | btrfs_put_ordered_extent(ordered); | |
9346 | btrfs_put_ordered_extent(ordered); | |
9347 | } | |
9348 | } | |
56fa9d07 | 9349 | btrfs_qgroup_check_reserved_leak(inode); |
5d4f98a2 | 9350 | inode_tree_del(inode); |
dcdbc059 | 9351 | btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0); |
a6dbd429 | 9352 | free: |
fa0d7e3d | 9353 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
9354 | } |
9355 | ||
45321ac5 | 9356 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
9357 | { |
9358 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 9359 | |
6379ef9f NA |
9360 | if (root == NULL) |
9361 | return 1; | |
9362 | ||
fa6ac876 | 9363 | /* the snap/subvol tree is on deleting */ |
69e9c6c6 | 9364 | if (btrfs_root_refs(&root->root_item) == 0) |
45321ac5 | 9365 | return 1; |
76dda93c | 9366 | else |
45321ac5 | 9367 | return generic_drop_inode(inode); |
76dda93c YZ |
9368 | } |
9369 | ||
0ee0fda0 | 9370 | static void init_once(void *foo) |
39279cc3 CM |
9371 | { |
9372 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
9373 | ||
9374 | inode_init_once(&ei->vfs_inode); | |
9375 | } | |
9376 | ||
9377 | void btrfs_destroy_cachep(void) | |
9378 | { | |
8c0a8537 KS |
9379 | /* |
9380 | * Make sure all delayed rcu free inodes are flushed before we | |
9381 | * destroy cache. | |
9382 | */ | |
9383 | rcu_barrier(); | |
5598e900 KM |
9384 | kmem_cache_destroy(btrfs_inode_cachep); |
9385 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
9386 | kmem_cache_destroy(btrfs_transaction_cachep); | |
9387 | kmem_cache_destroy(btrfs_path_cachep); | |
9388 | kmem_cache_destroy(btrfs_free_space_cachep); | |
39279cc3 CM |
9389 | } |
9390 | ||
9391 | int btrfs_init_cachep(void) | |
9392 | { | |
837e1972 | 9393 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 | 9394 | sizeof(struct btrfs_inode), 0, |
5d097056 VD |
9395 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT, |
9396 | init_once); | |
39279cc3 CM |
9397 | if (!btrfs_inode_cachep) |
9398 | goto fail; | |
9601e3f6 | 9399 | |
837e1972 | 9400 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 | 9401 | sizeof(struct btrfs_trans_handle), 0, |
fba4b697 | 9402 | SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
9403 | if (!btrfs_trans_handle_cachep) |
9404 | goto fail; | |
9601e3f6 | 9405 | |
837e1972 | 9406 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction", |
9601e3f6 | 9407 | sizeof(struct btrfs_transaction), 0, |
fba4b697 | 9408 | SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
9409 | if (!btrfs_transaction_cachep) |
9410 | goto fail; | |
9601e3f6 | 9411 | |
837e1972 | 9412 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 | 9413 | sizeof(struct btrfs_path), 0, |
fba4b697 | 9414 | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
9415 | if (!btrfs_path_cachep) |
9416 | goto fail; | |
9601e3f6 | 9417 | |
837e1972 | 9418 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 | 9419 | sizeof(struct btrfs_free_space), 0, |
fba4b697 | 9420 | SLAB_MEM_SPREAD, NULL); |
dc89e982 JB |
9421 | if (!btrfs_free_space_cachep) |
9422 | goto fail; | |
9423 | ||
39279cc3 CM |
9424 | return 0; |
9425 | fail: | |
9426 | btrfs_destroy_cachep(); | |
9427 | return -ENOMEM; | |
9428 | } | |
9429 | ||
a528d35e DH |
9430 | static int btrfs_getattr(const struct path *path, struct kstat *stat, |
9431 | u32 request_mask, unsigned int flags) | |
39279cc3 | 9432 | { |
df0af1a5 | 9433 | u64 delalloc_bytes; |
a528d35e | 9434 | struct inode *inode = d_inode(path->dentry); |
fadc0d8b DS |
9435 | u32 blocksize = inode->i_sb->s_blocksize; |
9436 | ||
39279cc3 | 9437 | generic_fillattr(inode, stat); |
0ee5dc67 | 9438 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
df0af1a5 MX |
9439 | |
9440 | spin_lock(&BTRFS_I(inode)->lock); | |
9441 | delalloc_bytes = BTRFS_I(inode)->delalloc_bytes; | |
9442 | spin_unlock(&BTRFS_I(inode)->lock); | |
fadc0d8b | 9443 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
df0af1a5 | 9444 | ALIGN(delalloc_bytes, blocksize)) >> 9; |
39279cc3 CM |
9445 | return 0; |
9446 | } | |
9447 | ||
cdd1fedf DF |
9448 | static int btrfs_rename_exchange(struct inode *old_dir, |
9449 | struct dentry *old_dentry, | |
9450 | struct inode *new_dir, | |
9451 | struct dentry *new_dentry) | |
9452 | { | |
0b246afa | 9453 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
cdd1fedf DF |
9454 | struct btrfs_trans_handle *trans; |
9455 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
9456 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; | |
9457 | struct inode *new_inode = new_dentry->d_inode; | |
9458 | struct inode *old_inode = old_dentry->d_inode; | |
c2050a45 | 9459 | struct timespec ctime = current_time(old_inode); |
cdd1fedf | 9460 | struct dentry *parent; |
4a0cc7ca NB |
9461 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
9462 | u64 new_ino = btrfs_ino(BTRFS_I(new_inode)); | |
cdd1fedf DF |
9463 | u64 old_idx = 0; |
9464 | u64 new_idx = 0; | |
9465 | u64 root_objectid; | |
9466 | int ret; | |
86e8aa0e FM |
9467 | bool root_log_pinned = false; |
9468 | bool dest_log_pinned = false; | |
cdd1fedf DF |
9469 | |
9470 | /* we only allow rename subvolume link between subvolumes */ | |
9471 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) | |
9472 | return -EXDEV; | |
9473 | ||
9474 | /* close the race window with snapshot create/destroy ioctl */ | |
9475 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 9476 | down_read(&fs_info->subvol_sem); |
cdd1fedf | 9477 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9478 | down_read(&fs_info->subvol_sem); |
cdd1fedf DF |
9479 | |
9480 | /* | |
9481 | * We want to reserve the absolute worst case amount of items. So if | |
9482 | * both inodes are subvols and we need to unlink them then that would | |
9483 | * require 4 item modifications, but if they are both normal inodes it | |
9484 | * would require 5 item modifications, so we'll assume their normal | |
9485 | * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items | |
9486 | * should cover the worst case number of items we'll modify. | |
9487 | */ | |
9488 | trans = btrfs_start_transaction(root, 12); | |
9489 | if (IS_ERR(trans)) { | |
9490 | ret = PTR_ERR(trans); | |
9491 | goto out_notrans; | |
9492 | } | |
9493 | ||
9494 | /* | |
9495 | * We need to find a free sequence number both in the source and | |
9496 | * in the destination directory for the exchange. | |
9497 | */ | |
877574e2 | 9498 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx); |
cdd1fedf DF |
9499 | if (ret) |
9500 | goto out_fail; | |
877574e2 | 9501 | ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx); |
cdd1fedf DF |
9502 | if (ret) |
9503 | goto out_fail; | |
9504 | ||
9505 | BTRFS_I(old_inode)->dir_index = 0ULL; | |
9506 | BTRFS_I(new_inode)->dir_index = 0ULL; | |
9507 | ||
9508 | /* Reference for the source. */ | |
9509 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9510 | /* force full log commit if subvolume involved. */ | |
0b246afa | 9511 | btrfs_set_log_full_commit(fs_info, trans); |
cdd1fedf | 9512 | } else { |
376e5a57 FM |
9513 | btrfs_pin_log_trans(root); |
9514 | root_log_pinned = true; | |
cdd1fedf DF |
9515 | ret = btrfs_insert_inode_ref(trans, dest, |
9516 | new_dentry->d_name.name, | |
9517 | new_dentry->d_name.len, | |
9518 | old_ino, | |
f85b7379 DS |
9519 | btrfs_ino(BTRFS_I(new_dir)), |
9520 | old_idx); | |
cdd1fedf DF |
9521 | if (ret) |
9522 | goto out_fail; | |
cdd1fedf DF |
9523 | } |
9524 | ||
9525 | /* And now for the dest. */ | |
9526 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9527 | /* force full log commit if subvolume involved. */ | |
0b246afa | 9528 | btrfs_set_log_full_commit(fs_info, trans); |
cdd1fedf | 9529 | } else { |
376e5a57 FM |
9530 | btrfs_pin_log_trans(dest); |
9531 | dest_log_pinned = true; | |
cdd1fedf DF |
9532 | ret = btrfs_insert_inode_ref(trans, root, |
9533 | old_dentry->d_name.name, | |
9534 | old_dentry->d_name.len, | |
9535 | new_ino, | |
f85b7379 DS |
9536 | btrfs_ino(BTRFS_I(old_dir)), |
9537 | new_idx); | |
cdd1fedf DF |
9538 | if (ret) |
9539 | goto out_fail; | |
cdd1fedf DF |
9540 | } |
9541 | ||
9542 | /* Update inode version and ctime/mtime. */ | |
9543 | inode_inc_iversion(old_dir); | |
9544 | inode_inc_iversion(new_dir); | |
9545 | inode_inc_iversion(old_inode); | |
9546 | inode_inc_iversion(new_inode); | |
9547 | old_dir->i_ctime = old_dir->i_mtime = ctime; | |
9548 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
9549 | old_inode->i_ctime = ctime; | |
9550 | new_inode->i_ctime = ctime; | |
9551 | ||
9552 | if (old_dentry->d_parent != new_dentry->d_parent) { | |
f85b7379 DS |
9553 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
9554 | BTRFS_I(old_inode), 1); | |
9555 | btrfs_record_unlink_dir(trans, BTRFS_I(new_dir), | |
9556 | BTRFS_I(new_inode), 1); | |
cdd1fedf DF |
9557 | } |
9558 | ||
9559 | /* src is a subvolume */ | |
9560 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9561 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; | |
9562 | ret = btrfs_unlink_subvol(trans, root, old_dir, | |
9563 | root_objectid, | |
9564 | old_dentry->d_name.name, | |
9565 | old_dentry->d_name.len); | |
9566 | } else { /* src is an inode */ | |
4ec5934e NB |
9567 | ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir), |
9568 | BTRFS_I(old_dentry->d_inode), | |
cdd1fedf DF |
9569 | old_dentry->d_name.name, |
9570 | old_dentry->d_name.len); | |
9571 | if (!ret) | |
9572 | ret = btrfs_update_inode(trans, root, old_inode); | |
9573 | } | |
9574 | if (ret) { | |
66642832 | 9575 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9576 | goto out_fail; |
9577 | } | |
9578 | ||
9579 | /* dest is a subvolume */ | |
9580 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9581 | root_objectid = BTRFS_I(new_inode)->root->root_key.objectid; | |
9582 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
9583 | root_objectid, | |
9584 | new_dentry->d_name.name, | |
9585 | new_dentry->d_name.len); | |
9586 | } else { /* dest is an inode */ | |
4ec5934e NB |
9587 | ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir), |
9588 | BTRFS_I(new_dentry->d_inode), | |
cdd1fedf DF |
9589 | new_dentry->d_name.name, |
9590 | new_dentry->d_name.len); | |
9591 | if (!ret) | |
9592 | ret = btrfs_update_inode(trans, dest, new_inode); | |
9593 | } | |
9594 | if (ret) { | |
66642832 | 9595 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9596 | goto out_fail; |
9597 | } | |
9598 | ||
db0a669f | 9599 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
cdd1fedf DF |
9600 | new_dentry->d_name.name, |
9601 | new_dentry->d_name.len, 0, old_idx); | |
9602 | if (ret) { | |
66642832 | 9603 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9604 | goto out_fail; |
9605 | } | |
9606 | ||
db0a669f | 9607 | ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode), |
cdd1fedf DF |
9608 | old_dentry->d_name.name, |
9609 | old_dentry->d_name.len, 0, new_idx); | |
9610 | if (ret) { | |
66642832 | 9611 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9612 | goto out_fail; |
9613 | } | |
9614 | ||
9615 | if (old_inode->i_nlink == 1) | |
9616 | BTRFS_I(old_inode)->dir_index = old_idx; | |
9617 | if (new_inode->i_nlink == 1) | |
9618 | BTRFS_I(new_inode)->dir_index = new_idx; | |
9619 | ||
86e8aa0e | 9620 | if (root_log_pinned) { |
cdd1fedf | 9621 | parent = new_dentry->d_parent; |
f85b7379 DS |
9622 | btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir), |
9623 | parent); | |
cdd1fedf | 9624 | btrfs_end_log_trans(root); |
86e8aa0e | 9625 | root_log_pinned = false; |
cdd1fedf | 9626 | } |
86e8aa0e | 9627 | if (dest_log_pinned) { |
cdd1fedf | 9628 | parent = old_dentry->d_parent; |
f85b7379 DS |
9629 | btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir), |
9630 | parent); | |
cdd1fedf | 9631 | btrfs_end_log_trans(dest); |
86e8aa0e | 9632 | dest_log_pinned = false; |
cdd1fedf DF |
9633 | } |
9634 | out_fail: | |
86e8aa0e FM |
9635 | /* |
9636 | * If we have pinned a log and an error happened, we unpin tasks | |
9637 | * trying to sync the log and force them to fallback to a transaction | |
9638 | * commit if the log currently contains any of the inodes involved in | |
9639 | * this rename operation (to ensure we do not persist a log with an | |
9640 | * inconsistent state for any of these inodes or leading to any | |
9641 | * inconsistencies when replayed). If the transaction was aborted, the | |
9642 | * abortion reason is propagated to userspace when attempting to commit | |
9643 | * the transaction. If the log does not contain any of these inodes, we | |
9644 | * allow the tasks to sync it. | |
9645 | */ | |
9646 | if (ret && (root_log_pinned || dest_log_pinned)) { | |
0f8939b8 NB |
9647 | if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) || |
9648 | btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) || | |
9649 | btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) || | |
86e8aa0e | 9650 | (new_inode && |
0f8939b8 | 9651 | btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation))) |
0b246afa | 9652 | btrfs_set_log_full_commit(fs_info, trans); |
86e8aa0e FM |
9653 | |
9654 | if (root_log_pinned) { | |
9655 | btrfs_end_log_trans(root); | |
9656 | root_log_pinned = false; | |
9657 | } | |
9658 | if (dest_log_pinned) { | |
9659 | btrfs_end_log_trans(dest); | |
9660 | dest_log_pinned = false; | |
9661 | } | |
9662 | } | |
3a45bb20 | 9663 | ret = btrfs_end_transaction(trans); |
cdd1fedf DF |
9664 | out_notrans: |
9665 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 9666 | up_read(&fs_info->subvol_sem); |
cdd1fedf | 9667 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9668 | up_read(&fs_info->subvol_sem); |
cdd1fedf DF |
9669 | |
9670 | return ret; | |
9671 | } | |
9672 | ||
9673 | static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans, | |
9674 | struct btrfs_root *root, | |
9675 | struct inode *dir, | |
9676 | struct dentry *dentry) | |
9677 | { | |
9678 | int ret; | |
9679 | struct inode *inode; | |
9680 | u64 objectid; | |
9681 | u64 index; | |
9682 | ||
9683 | ret = btrfs_find_free_ino(root, &objectid); | |
9684 | if (ret) | |
9685 | return ret; | |
9686 | ||
9687 | inode = btrfs_new_inode(trans, root, dir, | |
9688 | dentry->d_name.name, | |
9689 | dentry->d_name.len, | |
4a0cc7ca | 9690 | btrfs_ino(BTRFS_I(dir)), |
cdd1fedf DF |
9691 | objectid, |
9692 | S_IFCHR | WHITEOUT_MODE, | |
9693 | &index); | |
9694 | ||
9695 | if (IS_ERR(inode)) { | |
9696 | ret = PTR_ERR(inode); | |
9697 | return ret; | |
9698 | } | |
9699 | ||
9700 | inode->i_op = &btrfs_special_inode_operations; | |
9701 | init_special_inode(inode, inode->i_mode, | |
9702 | WHITEOUT_DEV); | |
9703 | ||
9704 | ret = btrfs_init_inode_security(trans, inode, dir, | |
9705 | &dentry->d_name); | |
9706 | if (ret) | |
c9901618 | 9707 | goto out; |
cdd1fedf | 9708 | |
cef415af NB |
9709 | ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, |
9710 | BTRFS_I(inode), 0, index); | |
cdd1fedf | 9711 | if (ret) |
c9901618 | 9712 | goto out; |
cdd1fedf DF |
9713 | |
9714 | ret = btrfs_update_inode(trans, root, inode); | |
c9901618 | 9715 | out: |
cdd1fedf | 9716 | unlock_new_inode(inode); |
c9901618 FM |
9717 | if (ret) |
9718 | inode_dec_link_count(inode); | |
cdd1fedf DF |
9719 | iput(inode); |
9720 | ||
c9901618 | 9721 | return ret; |
cdd1fedf DF |
9722 | } |
9723 | ||
d397712b | 9724 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
cdd1fedf DF |
9725 | struct inode *new_dir, struct dentry *new_dentry, |
9726 | unsigned int flags) | |
39279cc3 | 9727 | { |
0b246afa | 9728 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
39279cc3 | 9729 | struct btrfs_trans_handle *trans; |
5062af35 | 9730 | unsigned int trans_num_items; |
39279cc3 | 9731 | struct btrfs_root *root = BTRFS_I(old_dir)->root; |
4df27c4d | 9732 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
2b0143b5 DH |
9733 | struct inode *new_inode = d_inode(new_dentry); |
9734 | struct inode *old_inode = d_inode(old_dentry); | |
00e4e6b3 | 9735 | u64 index = 0; |
4df27c4d | 9736 | u64 root_objectid; |
39279cc3 | 9737 | int ret; |
4a0cc7ca | 9738 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
3dc9e8f7 | 9739 | bool log_pinned = false; |
39279cc3 | 9740 | |
4a0cc7ca | 9741 | if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
9742 | return -EPERM; |
9743 | ||
4df27c4d | 9744 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 9745 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
9746 | return -EXDEV; |
9747 | ||
33345d01 | 9748 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
4a0cc7ca | 9749 | (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID)) |
39279cc3 | 9750 | return -ENOTEMPTY; |
5f39d397 | 9751 | |
4df27c4d YZ |
9752 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
9753 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
9754 | return -ENOTEMPTY; | |
9c52057c CM |
9755 | |
9756 | ||
9757 | /* check for collisions, even if the name isn't there */ | |
4871c158 | 9758 | ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, |
9c52057c CM |
9759 | new_dentry->d_name.name, |
9760 | new_dentry->d_name.len); | |
9761 | ||
9762 | if (ret) { | |
9763 | if (ret == -EEXIST) { | |
9764 | /* we shouldn't get | |
9765 | * eexist without a new_inode */ | |
fae7f21c | 9766 | if (WARN_ON(!new_inode)) { |
9c52057c CM |
9767 | return ret; |
9768 | } | |
9769 | } else { | |
9770 | /* maybe -EOVERFLOW */ | |
9771 | return ret; | |
9772 | } | |
9773 | } | |
9774 | ret = 0; | |
9775 | ||
5a3f23d5 | 9776 | /* |
8d875f95 CM |
9777 | * we're using rename to replace one file with another. Start IO on it |
9778 | * now so we don't add too much work to the end of the transaction | |
5a3f23d5 | 9779 | */ |
8d875f95 | 9780 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size) |
5a3f23d5 CM |
9781 | filemap_flush(old_inode->i_mapping); |
9782 | ||
76dda93c | 9783 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 9784 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9785 | down_read(&fs_info->subvol_sem); |
a22285a6 YZ |
9786 | /* |
9787 | * We want to reserve the absolute worst case amount of items. So if | |
9788 | * both inodes are subvols and we need to unlink them then that would | |
9789 | * require 4 item modifications, but if they are both normal inodes it | |
cdd1fedf | 9790 | * would require 5 item modifications, so we'll assume they are normal |
a22285a6 YZ |
9791 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items |
9792 | * should cover the worst case number of items we'll modify. | |
5062af35 FM |
9793 | * If our rename has the whiteout flag, we need more 5 units for the |
9794 | * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item | |
9795 | * when selinux is enabled). | |
a22285a6 | 9796 | */ |
5062af35 FM |
9797 | trans_num_items = 11; |
9798 | if (flags & RENAME_WHITEOUT) | |
9799 | trans_num_items += 5; | |
9800 | trans = btrfs_start_transaction(root, trans_num_items); | |
b44c59a8 | 9801 | if (IS_ERR(trans)) { |
cdd1fedf DF |
9802 | ret = PTR_ERR(trans); |
9803 | goto out_notrans; | |
9804 | } | |
76dda93c | 9805 | |
4df27c4d YZ |
9806 | if (dest != root) |
9807 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 9808 | |
877574e2 | 9809 | ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index); |
a5719521 YZ |
9810 | if (ret) |
9811 | goto out_fail; | |
5a3f23d5 | 9812 | |
67de1176 | 9813 | BTRFS_I(old_inode)->dir_index = 0ULL; |
33345d01 | 9814 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d | 9815 | /* force full log commit if subvolume involved. */ |
0b246afa | 9816 | btrfs_set_log_full_commit(fs_info, trans); |
4df27c4d | 9817 | } else { |
c4aba954 FM |
9818 | btrfs_pin_log_trans(root); |
9819 | log_pinned = true; | |
a5719521 YZ |
9820 | ret = btrfs_insert_inode_ref(trans, dest, |
9821 | new_dentry->d_name.name, | |
9822 | new_dentry->d_name.len, | |
33345d01 | 9823 | old_ino, |
4a0cc7ca | 9824 | btrfs_ino(BTRFS_I(new_dir)), index); |
a5719521 YZ |
9825 | if (ret) |
9826 | goto out_fail; | |
4df27c4d | 9827 | } |
5a3f23d5 | 9828 | |
0c4d2d95 JB |
9829 | inode_inc_iversion(old_dir); |
9830 | inode_inc_iversion(new_dir); | |
9831 | inode_inc_iversion(old_inode); | |
04b285f3 DD |
9832 | old_dir->i_ctime = old_dir->i_mtime = |
9833 | new_dir->i_ctime = new_dir->i_mtime = | |
c2050a45 | 9834 | old_inode->i_ctime = current_time(old_dir); |
5f39d397 | 9835 | |
12fcfd22 | 9836 | if (old_dentry->d_parent != new_dentry->d_parent) |
f85b7379 DS |
9837 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
9838 | BTRFS_I(old_inode), 1); | |
12fcfd22 | 9839 | |
33345d01 | 9840 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
9841 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
9842 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
9843 | old_dentry->d_name.name, | |
9844 | old_dentry->d_name.len); | |
9845 | } else { | |
4ec5934e NB |
9846 | ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir), |
9847 | BTRFS_I(d_inode(old_dentry)), | |
92986796 AV |
9848 | old_dentry->d_name.name, |
9849 | old_dentry->d_name.len); | |
9850 | if (!ret) | |
9851 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 9852 | } |
79787eaa | 9853 | if (ret) { |
66642832 | 9854 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9855 | goto out_fail; |
9856 | } | |
39279cc3 CM |
9857 | |
9858 | if (new_inode) { | |
0c4d2d95 | 9859 | inode_inc_iversion(new_inode); |
c2050a45 | 9860 | new_inode->i_ctime = current_time(new_inode); |
4a0cc7ca | 9861 | if (unlikely(btrfs_ino(BTRFS_I(new_inode)) == |
4df27c4d YZ |
9862 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
9863 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
9864 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
9865 | root_objectid, | |
9866 | new_dentry->d_name.name, | |
9867 | new_dentry->d_name.len); | |
9868 | BUG_ON(new_inode->i_nlink == 0); | |
9869 | } else { | |
4ec5934e NB |
9870 | ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir), |
9871 | BTRFS_I(d_inode(new_dentry)), | |
4df27c4d YZ |
9872 | new_dentry->d_name.name, |
9873 | new_dentry->d_name.len); | |
9874 | } | |
4ef31a45 | 9875 | if (!ret && new_inode->i_nlink == 0) |
73f2e545 NB |
9876 | ret = btrfs_orphan_add(trans, |
9877 | BTRFS_I(d_inode(new_dentry))); | |
79787eaa | 9878 | if (ret) { |
66642832 | 9879 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9880 | goto out_fail; |
9881 | } | |
39279cc3 | 9882 | } |
aec7477b | 9883 | |
db0a669f | 9884 | ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode), |
4df27c4d | 9885 | new_dentry->d_name.name, |
a5719521 | 9886 | new_dentry->d_name.len, 0, index); |
79787eaa | 9887 | if (ret) { |
66642832 | 9888 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9889 | goto out_fail; |
9890 | } | |
39279cc3 | 9891 | |
67de1176 MX |
9892 | if (old_inode->i_nlink == 1) |
9893 | BTRFS_I(old_inode)->dir_index = index; | |
9894 | ||
3dc9e8f7 | 9895 | if (log_pinned) { |
10d9f309 | 9896 | struct dentry *parent = new_dentry->d_parent; |
3dc9e8f7 | 9897 | |
f85b7379 DS |
9898 | btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir), |
9899 | parent); | |
4df27c4d | 9900 | btrfs_end_log_trans(root); |
3dc9e8f7 | 9901 | log_pinned = false; |
4df27c4d | 9902 | } |
cdd1fedf DF |
9903 | |
9904 | if (flags & RENAME_WHITEOUT) { | |
9905 | ret = btrfs_whiteout_for_rename(trans, root, old_dir, | |
9906 | old_dentry); | |
9907 | ||
9908 | if (ret) { | |
66642832 | 9909 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9910 | goto out_fail; |
9911 | } | |
4df27c4d | 9912 | } |
39279cc3 | 9913 | out_fail: |
3dc9e8f7 FM |
9914 | /* |
9915 | * If we have pinned the log and an error happened, we unpin tasks | |
9916 | * trying to sync the log and force them to fallback to a transaction | |
9917 | * commit if the log currently contains any of the inodes involved in | |
9918 | * this rename operation (to ensure we do not persist a log with an | |
9919 | * inconsistent state for any of these inodes or leading to any | |
9920 | * inconsistencies when replayed). If the transaction was aborted, the | |
9921 | * abortion reason is propagated to userspace when attempting to commit | |
9922 | * the transaction. If the log does not contain any of these inodes, we | |
9923 | * allow the tasks to sync it. | |
9924 | */ | |
9925 | if (ret && log_pinned) { | |
0f8939b8 NB |
9926 | if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) || |
9927 | btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) || | |
9928 | btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) || | |
3dc9e8f7 | 9929 | (new_inode && |
0f8939b8 | 9930 | btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation))) |
0b246afa | 9931 | btrfs_set_log_full_commit(fs_info, trans); |
3dc9e8f7 FM |
9932 | |
9933 | btrfs_end_log_trans(root); | |
9934 | log_pinned = false; | |
9935 | } | |
3a45bb20 | 9936 | btrfs_end_transaction(trans); |
b44c59a8 | 9937 | out_notrans: |
33345d01 | 9938 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9939 | up_read(&fs_info->subvol_sem); |
9ed74f2d | 9940 | |
39279cc3 CM |
9941 | return ret; |
9942 | } | |
9943 | ||
80ace85c MS |
9944 | static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry, |
9945 | struct inode *new_dir, struct dentry *new_dentry, | |
9946 | unsigned int flags) | |
9947 | { | |
cdd1fedf | 9948 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
80ace85c MS |
9949 | return -EINVAL; |
9950 | ||
cdd1fedf DF |
9951 | if (flags & RENAME_EXCHANGE) |
9952 | return btrfs_rename_exchange(old_dir, old_dentry, new_dir, | |
9953 | new_dentry); | |
9954 | ||
9955 | return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags); | |
80ace85c MS |
9956 | } |
9957 | ||
8ccf6f19 MX |
9958 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
9959 | { | |
9960 | struct btrfs_delalloc_work *delalloc_work; | |
9f23e289 | 9961 | struct inode *inode; |
8ccf6f19 MX |
9962 | |
9963 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
9964 | work); | |
9f23e289 | 9965 | inode = delalloc_work->inode; |
30424601 DS |
9966 | filemap_flush(inode->i_mapping); |
9967 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
9968 | &BTRFS_I(inode)->runtime_flags)) | |
9f23e289 | 9969 | filemap_flush(inode->i_mapping); |
8ccf6f19 MX |
9970 | |
9971 | if (delalloc_work->delay_iput) | |
9f23e289 | 9972 | btrfs_add_delayed_iput(inode); |
8ccf6f19 | 9973 | else |
9f23e289 | 9974 | iput(inode); |
8ccf6f19 MX |
9975 | complete(&delalloc_work->completion); |
9976 | } | |
9977 | ||
9978 | struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode, | |
651d494a | 9979 | int delay_iput) |
8ccf6f19 MX |
9980 | { |
9981 | struct btrfs_delalloc_work *work; | |
9982 | ||
100d5702 | 9983 | work = kmalloc(sizeof(*work), GFP_NOFS); |
8ccf6f19 MX |
9984 | if (!work) |
9985 | return NULL; | |
9986 | ||
9987 | init_completion(&work->completion); | |
9988 | INIT_LIST_HEAD(&work->list); | |
9989 | work->inode = inode; | |
8ccf6f19 | 9990 | work->delay_iput = delay_iput; |
9e0af237 LB |
9991 | WARN_ON_ONCE(!inode); |
9992 | btrfs_init_work(&work->work, btrfs_flush_delalloc_helper, | |
9993 | btrfs_run_delalloc_work, NULL, NULL); | |
8ccf6f19 MX |
9994 | |
9995 | return work; | |
9996 | } | |
9997 | ||
9998 | void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work) | |
9999 | { | |
10000 | wait_for_completion(&work->completion); | |
100d5702 | 10001 | kfree(work); |
8ccf6f19 MX |
10002 | } |
10003 | ||
d352ac68 CM |
10004 | /* |
10005 | * some fairly slow code that needs optimization. This walks the list | |
10006 | * of all the inodes with pending delalloc and forces them to disk. | |
10007 | */ | |
6c255e67 MX |
10008 | static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput, |
10009 | int nr) | |
ea8c2819 | 10010 | { |
ea8c2819 | 10011 | struct btrfs_inode *binode; |
5b21f2ed | 10012 | struct inode *inode; |
8ccf6f19 MX |
10013 | struct btrfs_delalloc_work *work, *next; |
10014 | struct list_head works; | |
1eafa6c7 | 10015 | struct list_head splice; |
8ccf6f19 | 10016 | int ret = 0; |
ea8c2819 | 10017 | |
8ccf6f19 | 10018 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 10019 | INIT_LIST_HEAD(&splice); |
63607cc8 | 10020 | |
573bfb72 | 10021 | mutex_lock(&root->delalloc_mutex); |
eb73c1b7 MX |
10022 | spin_lock(&root->delalloc_lock); |
10023 | list_splice_init(&root->delalloc_inodes, &splice); | |
1eafa6c7 MX |
10024 | while (!list_empty(&splice)) { |
10025 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 10026 | delalloc_inodes); |
1eafa6c7 | 10027 | |
eb73c1b7 MX |
10028 | list_move_tail(&binode->delalloc_inodes, |
10029 | &root->delalloc_inodes); | |
5b21f2ed | 10030 | inode = igrab(&binode->vfs_inode); |
df0af1a5 | 10031 | if (!inode) { |
eb73c1b7 | 10032 | cond_resched_lock(&root->delalloc_lock); |
1eafa6c7 | 10033 | continue; |
df0af1a5 | 10034 | } |
eb73c1b7 | 10035 | spin_unlock(&root->delalloc_lock); |
1eafa6c7 | 10036 | |
651d494a | 10037 | work = btrfs_alloc_delalloc_work(inode, delay_iput); |
5d99a998 | 10038 | if (!work) { |
f4ab9ea7 JB |
10039 | if (delay_iput) |
10040 | btrfs_add_delayed_iput(inode); | |
10041 | else | |
10042 | iput(inode); | |
1eafa6c7 | 10043 | ret = -ENOMEM; |
a1ecaabb | 10044 | goto out; |
5b21f2ed | 10045 | } |
1eafa6c7 | 10046 | list_add_tail(&work->list, &works); |
a44903ab QW |
10047 | btrfs_queue_work(root->fs_info->flush_workers, |
10048 | &work->work); | |
6c255e67 MX |
10049 | ret++; |
10050 | if (nr != -1 && ret >= nr) | |
a1ecaabb | 10051 | goto out; |
5b21f2ed | 10052 | cond_resched(); |
eb73c1b7 | 10053 | spin_lock(&root->delalloc_lock); |
ea8c2819 | 10054 | } |
eb73c1b7 | 10055 | spin_unlock(&root->delalloc_lock); |
8c8bee1d | 10056 | |
a1ecaabb | 10057 | out: |
eb73c1b7 MX |
10058 | list_for_each_entry_safe(work, next, &works, list) { |
10059 | list_del_init(&work->list); | |
10060 | btrfs_wait_and_free_delalloc_work(work); | |
10061 | } | |
10062 | ||
10063 | if (!list_empty_careful(&splice)) { | |
10064 | spin_lock(&root->delalloc_lock); | |
10065 | list_splice_tail(&splice, &root->delalloc_inodes); | |
10066 | spin_unlock(&root->delalloc_lock); | |
10067 | } | |
573bfb72 | 10068 | mutex_unlock(&root->delalloc_mutex); |
eb73c1b7 MX |
10069 | return ret; |
10070 | } | |
1eafa6c7 | 10071 | |
eb73c1b7 MX |
10072 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
10073 | { | |
0b246afa | 10074 | struct btrfs_fs_info *fs_info = root->fs_info; |
eb73c1b7 | 10075 | int ret; |
1eafa6c7 | 10076 | |
0b246afa | 10077 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
10078 | return -EROFS; |
10079 | ||
6c255e67 MX |
10080 | ret = __start_delalloc_inodes(root, delay_iput, -1); |
10081 | if (ret > 0) | |
10082 | ret = 0; | |
eb73c1b7 MX |
10083 | /* |
10084 | * the filemap_flush will queue IO into the worker threads, but | |
8c8bee1d CM |
10085 | * we have to make sure the IO is actually started and that |
10086 | * ordered extents get created before we return | |
10087 | */ | |
0b246afa JM |
10088 | atomic_inc(&fs_info->async_submit_draining); |
10089 | while (atomic_read(&fs_info->nr_async_submits) || | |
10090 | atomic_read(&fs_info->async_delalloc_pages)) { | |
10091 | wait_event(fs_info->async_submit_wait, | |
10092 | (atomic_read(&fs_info->nr_async_submits) == 0 && | |
10093 | atomic_read(&fs_info->async_delalloc_pages) == 0)); | |
10094 | } | |
10095 | atomic_dec(&fs_info->async_submit_draining); | |
eb73c1b7 MX |
10096 | return ret; |
10097 | } | |
10098 | ||
6c255e67 MX |
10099 | int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput, |
10100 | int nr) | |
eb73c1b7 MX |
10101 | { |
10102 | struct btrfs_root *root; | |
10103 | struct list_head splice; | |
10104 | int ret; | |
10105 | ||
2c21b4d7 | 10106 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
10107 | return -EROFS; |
10108 | ||
10109 | INIT_LIST_HEAD(&splice); | |
10110 | ||
573bfb72 | 10111 | mutex_lock(&fs_info->delalloc_root_mutex); |
eb73c1b7 MX |
10112 | spin_lock(&fs_info->delalloc_root_lock); |
10113 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
6c255e67 | 10114 | while (!list_empty(&splice) && nr) { |
eb73c1b7 MX |
10115 | root = list_first_entry(&splice, struct btrfs_root, |
10116 | delalloc_root); | |
10117 | root = btrfs_grab_fs_root(root); | |
10118 | BUG_ON(!root); | |
10119 | list_move_tail(&root->delalloc_root, | |
10120 | &fs_info->delalloc_roots); | |
10121 | spin_unlock(&fs_info->delalloc_root_lock); | |
10122 | ||
6c255e67 | 10123 | ret = __start_delalloc_inodes(root, delay_iput, nr); |
eb73c1b7 | 10124 | btrfs_put_fs_root(root); |
6c255e67 | 10125 | if (ret < 0) |
eb73c1b7 MX |
10126 | goto out; |
10127 | ||
6c255e67 MX |
10128 | if (nr != -1) { |
10129 | nr -= ret; | |
10130 | WARN_ON(nr < 0); | |
10131 | } | |
eb73c1b7 | 10132 | spin_lock(&fs_info->delalloc_root_lock); |
8ccf6f19 | 10133 | } |
eb73c1b7 | 10134 | spin_unlock(&fs_info->delalloc_root_lock); |
1eafa6c7 | 10135 | |
6c255e67 | 10136 | ret = 0; |
eb73c1b7 MX |
10137 | atomic_inc(&fs_info->async_submit_draining); |
10138 | while (atomic_read(&fs_info->nr_async_submits) || | |
10139 | atomic_read(&fs_info->async_delalloc_pages)) { | |
10140 | wait_event(fs_info->async_submit_wait, | |
10141 | (atomic_read(&fs_info->nr_async_submits) == 0 && | |
10142 | atomic_read(&fs_info->async_delalloc_pages) == 0)); | |
10143 | } | |
10144 | atomic_dec(&fs_info->async_submit_draining); | |
eb73c1b7 | 10145 | out: |
1eafa6c7 | 10146 | if (!list_empty_careful(&splice)) { |
eb73c1b7 MX |
10147 | spin_lock(&fs_info->delalloc_root_lock); |
10148 | list_splice_tail(&splice, &fs_info->delalloc_roots); | |
10149 | spin_unlock(&fs_info->delalloc_root_lock); | |
1eafa6c7 | 10150 | } |
573bfb72 | 10151 | mutex_unlock(&fs_info->delalloc_root_mutex); |
8ccf6f19 | 10152 | return ret; |
ea8c2819 CM |
10153 | } |
10154 | ||
39279cc3 CM |
10155 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
10156 | const char *symname) | |
10157 | { | |
0b246afa | 10158 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
10159 | struct btrfs_trans_handle *trans; |
10160 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
10161 | struct btrfs_path *path; | |
10162 | struct btrfs_key key; | |
1832a6d5 | 10163 | struct inode *inode = NULL; |
39279cc3 CM |
10164 | int err; |
10165 | int drop_inode = 0; | |
10166 | u64 objectid; | |
67871254 | 10167 | u64 index = 0; |
39279cc3 CM |
10168 | int name_len; |
10169 | int datasize; | |
5f39d397 | 10170 | unsigned long ptr; |
39279cc3 | 10171 | struct btrfs_file_extent_item *ei; |
5f39d397 | 10172 | struct extent_buffer *leaf; |
39279cc3 | 10173 | |
f06becc4 | 10174 | name_len = strlen(symname); |
0b246afa | 10175 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info)) |
39279cc3 | 10176 | return -ENAMETOOLONG; |
1832a6d5 | 10177 | |
9ed74f2d JB |
10178 | /* |
10179 | * 2 items for inode item and ref | |
10180 | * 2 items for dir items | |
9269d12b FM |
10181 | * 1 item for updating parent inode item |
10182 | * 1 item for the inline extent item | |
9ed74f2d JB |
10183 | * 1 item for xattr if selinux is on |
10184 | */ | |
9269d12b | 10185 | trans = btrfs_start_transaction(root, 7); |
a22285a6 YZ |
10186 | if (IS_ERR(trans)) |
10187 | return PTR_ERR(trans); | |
1832a6d5 | 10188 | |
581bb050 LZ |
10189 | err = btrfs_find_free_ino(root, &objectid); |
10190 | if (err) | |
10191 | goto out_unlock; | |
10192 | ||
aec7477b | 10193 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
10194 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), |
10195 | objectid, S_IFLNK|S_IRWXUGO, &index); | |
7cf96da3 TI |
10196 | if (IS_ERR(inode)) { |
10197 | err = PTR_ERR(inode); | |
39279cc3 | 10198 | goto out_unlock; |
7cf96da3 | 10199 | } |
39279cc3 | 10200 | |
ad19db71 CS |
10201 | /* |
10202 | * If the active LSM wants to access the inode during | |
10203 | * d_instantiate it needs these. Smack checks to see | |
10204 | * if the filesystem supports xattrs by looking at the | |
10205 | * ops vector. | |
10206 | */ | |
10207 | inode->i_fop = &btrfs_file_operations; | |
10208 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 10209 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
10210 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
10211 | ||
10212 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
10213 | if (err) | |
10214 | goto out_unlock_inode; | |
ad19db71 | 10215 | |
39279cc3 | 10216 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10217 | if (!path) { |
10218 | err = -ENOMEM; | |
b0d5d10f | 10219 | goto out_unlock_inode; |
d8926bb3 | 10220 | } |
4a0cc7ca | 10221 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
39279cc3 | 10222 | key.offset = 0; |
962a298f | 10223 | key.type = BTRFS_EXTENT_DATA_KEY; |
39279cc3 CM |
10224 | datasize = btrfs_file_extent_calc_inline_size(name_len); |
10225 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
10226 | datasize); | |
54aa1f4d | 10227 | if (err) { |
b0839166 | 10228 | btrfs_free_path(path); |
b0d5d10f | 10229 | goto out_unlock_inode; |
54aa1f4d | 10230 | } |
5f39d397 CM |
10231 | leaf = path->nodes[0]; |
10232 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
10233 | struct btrfs_file_extent_item); | |
10234 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
10235 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 10236 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
10237 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
10238 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
10239 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
10240 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
10241 | ||
39279cc3 | 10242 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
10243 | write_extent_buffer(leaf, symname, ptr, name_len); |
10244 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 10245 | btrfs_free_path(path); |
5f39d397 | 10246 | |
39279cc3 | 10247 | inode->i_op = &btrfs_symlink_inode_operations; |
21fc61c7 | 10248 | inode_nohighmem(inode); |
39279cc3 | 10249 | inode->i_mapping->a_ops = &btrfs_symlink_aops; |
d899e052 | 10250 | inode_set_bytes(inode, name_len); |
6ef06d27 | 10251 | btrfs_i_size_write(BTRFS_I(inode), name_len); |
54aa1f4d | 10252 | err = btrfs_update_inode(trans, root, inode); |
d50866d0 FM |
10253 | /* |
10254 | * Last step, add directory indexes for our symlink inode. This is the | |
10255 | * last step to avoid extra cleanup of these indexes if an error happens | |
10256 | * elsewhere above. | |
10257 | */ | |
10258 | if (!err) | |
cef415af NB |
10259 | err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, |
10260 | BTRFS_I(inode), 0, index); | |
b0d5d10f | 10261 | if (err) { |
54aa1f4d | 10262 | drop_inode = 1; |
b0d5d10f CM |
10263 | goto out_unlock_inode; |
10264 | } | |
10265 | ||
10266 | unlock_new_inode(inode); | |
10267 | d_instantiate(dentry, inode); | |
39279cc3 CM |
10268 | |
10269 | out_unlock: | |
3a45bb20 | 10270 | btrfs_end_transaction(trans); |
39279cc3 CM |
10271 | if (drop_inode) { |
10272 | inode_dec_link_count(inode); | |
10273 | iput(inode); | |
10274 | } | |
2ff7e61e | 10275 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 | 10276 | return err; |
b0d5d10f CM |
10277 | |
10278 | out_unlock_inode: | |
10279 | drop_inode = 1; | |
10280 | unlock_new_inode(inode); | |
10281 | goto out_unlock; | |
39279cc3 | 10282 | } |
16432985 | 10283 | |
0af3d00b JB |
10284 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
10285 | u64 start, u64 num_bytes, u64 min_size, | |
10286 | loff_t actual_len, u64 *alloc_hint, | |
10287 | struct btrfs_trans_handle *trans) | |
d899e052 | 10288 | { |
0b246afa | 10289 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5dc562c5 JB |
10290 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
10291 | struct extent_map *em; | |
d899e052 YZ |
10292 | struct btrfs_root *root = BTRFS_I(inode)->root; |
10293 | struct btrfs_key ins; | |
d899e052 | 10294 | u64 cur_offset = start; |
55a61d1d | 10295 | u64 i_size; |
154ea289 | 10296 | u64 cur_bytes; |
0b670dc4 | 10297 | u64 last_alloc = (u64)-1; |
d899e052 | 10298 | int ret = 0; |
0af3d00b | 10299 | bool own_trans = true; |
18513091 | 10300 | u64 end = start + num_bytes - 1; |
d899e052 | 10301 | |
0af3d00b JB |
10302 | if (trans) |
10303 | own_trans = false; | |
d899e052 | 10304 | while (num_bytes > 0) { |
0af3d00b JB |
10305 | if (own_trans) { |
10306 | trans = btrfs_start_transaction(root, 3); | |
10307 | if (IS_ERR(trans)) { | |
10308 | ret = PTR_ERR(trans); | |
10309 | break; | |
10310 | } | |
5a303d5d YZ |
10311 | } |
10312 | ||
ee22184b | 10313 | cur_bytes = min_t(u64, num_bytes, SZ_256M); |
154ea289 | 10314 | cur_bytes = max(cur_bytes, min_size); |
0b670dc4 JB |
10315 | /* |
10316 | * If we are severely fragmented we could end up with really | |
10317 | * small allocations, so if the allocator is returning small | |
10318 | * chunks lets make its job easier by only searching for those | |
10319 | * sized chunks. | |
10320 | */ | |
10321 | cur_bytes = min(cur_bytes, last_alloc); | |
18513091 WX |
10322 | ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes, |
10323 | min_size, 0, *alloc_hint, &ins, 1, 0); | |
5a303d5d | 10324 | if (ret) { |
0af3d00b | 10325 | if (own_trans) |
3a45bb20 | 10326 | btrfs_end_transaction(trans); |
a22285a6 | 10327 | break; |
d899e052 | 10328 | } |
0b246afa | 10329 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
5a303d5d | 10330 | |
0b670dc4 | 10331 | last_alloc = ins.offset; |
d899e052 YZ |
10332 | ret = insert_reserved_file_extent(trans, inode, |
10333 | cur_offset, ins.objectid, | |
10334 | ins.offset, ins.offset, | |
920bbbfb | 10335 | ins.offset, 0, 0, 0, |
d899e052 | 10336 | BTRFS_FILE_EXTENT_PREALLOC); |
79787eaa | 10337 | if (ret) { |
2ff7e61e | 10338 | btrfs_free_reserved_extent(fs_info, ins.objectid, |
e570fd27 | 10339 | ins.offset, 0); |
66642832 | 10340 | btrfs_abort_transaction(trans, ret); |
79787eaa | 10341 | if (own_trans) |
3a45bb20 | 10342 | btrfs_end_transaction(trans); |
79787eaa JM |
10343 | break; |
10344 | } | |
31193213 | 10345 | |
dcdbc059 | 10346 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
a1ed835e | 10347 | cur_offset + ins.offset -1, 0); |
5a303d5d | 10348 | |
5dc562c5 JB |
10349 | em = alloc_extent_map(); |
10350 | if (!em) { | |
10351 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
10352 | &BTRFS_I(inode)->runtime_flags); | |
10353 | goto next; | |
10354 | } | |
10355 | ||
10356 | em->start = cur_offset; | |
10357 | em->orig_start = cur_offset; | |
10358 | em->len = ins.offset; | |
10359 | em->block_start = ins.objectid; | |
10360 | em->block_len = ins.offset; | |
b4939680 | 10361 | em->orig_block_len = ins.offset; |
cc95bef6 | 10362 | em->ram_bytes = ins.offset; |
0b246afa | 10363 | em->bdev = fs_info->fs_devices->latest_bdev; |
5dc562c5 JB |
10364 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); |
10365 | em->generation = trans->transid; | |
10366 | ||
10367 | while (1) { | |
10368 | write_lock(&em_tree->lock); | |
09a2a8f9 | 10369 | ret = add_extent_mapping(em_tree, em, 1); |
5dc562c5 JB |
10370 | write_unlock(&em_tree->lock); |
10371 | if (ret != -EEXIST) | |
10372 | break; | |
dcdbc059 | 10373 | btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset, |
5dc562c5 JB |
10374 | cur_offset + ins.offset - 1, |
10375 | 0); | |
10376 | } | |
10377 | free_extent_map(em); | |
10378 | next: | |
d899e052 YZ |
10379 | num_bytes -= ins.offset; |
10380 | cur_offset += ins.offset; | |
efa56464 | 10381 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 10382 | |
0c4d2d95 | 10383 | inode_inc_iversion(inode); |
c2050a45 | 10384 | inode->i_ctime = current_time(inode); |
6cbff00f | 10385 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 10386 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
10387 | (actual_len > inode->i_size) && |
10388 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 10389 | if (cur_offset > actual_len) |
55a61d1d | 10390 | i_size = actual_len; |
d1ea6a61 | 10391 | else |
55a61d1d JB |
10392 | i_size = cur_offset; |
10393 | i_size_write(inode, i_size); | |
10394 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
10395 | } |
10396 | ||
d899e052 | 10397 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
10398 | |
10399 | if (ret) { | |
66642832 | 10400 | btrfs_abort_transaction(trans, ret); |
79787eaa | 10401 | if (own_trans) |
3a45bb20 | 10402 | btrfs_end_transaction(trans); |
79787eaa JM |
10403 | break; |
10404 | } | |
d899e052 | 10405 | |
0af3d00b | 10406 | if (own_trans) |
3a45bb20 | 10407 | btrfs_end_transaction(trans); |
5a303d5d | 10408 | } |
18513091 WX |
10409 | if (cur_offset < end) |
10410 | btrfs_free_reserved_data_space(inode, cur_offset, | |
10411 | end - cur_offset + 1); | |
d899e052 YZ |
10412 | return ret; |
10413 | } | |
10414 | ||
0af3d00b JB |
10415 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
10416 | u64 start, u64 num_bytes, u64 min_size, | |
10417 | loff_t actual_len, u64 *alloc_hint) | |
10418 | { | |
10419 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10420 | min_size, actual_len, alloc_hint, | |
10421 | NULL); | |
10422 | } | |
10423 | ||
10424 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
10425 | struct btrfs_trans_handle *trans, int mode, | |
10426 | u64 start, u64 num_bytes, u64 min_size, | |
10427 | loff_t actual_len, u64 *alloc_hint) | |
10428 | { | |
10429 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10430 | min_size, actual_len, alloc_hint, trans); | |
10431 | } | |
10432 | ||
e6dcd2dc CM |
10433 | static int btrfs_set_page_dirty(struct page *page) |
10434 | { | |
e6dcd2dc CM |
10435 | return __set_page_dirty_nobuffers(page); |
10436 | } | |
10437 | ||
10556cb2 | 10438 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 10439 | { |
b83cc969 | 10440 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 10441 | umode_t mode = inode->i_mode; |
b83cc969 | 10442 | |
cb6db4e5 JM |
10443 | if (mask & MAY_WRITE && |
10444 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
10445 | if (btrfs_root_readonly(root)) | |
10446 | return -EROFS; | |
10447 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
10448 | return -EACCES; | |
10449 | } | |
2830ba7f | 10450 | return generic_permission(inode, mask); |
fdebe2bd | 10451 | } |
39279cc3 | 10452 | |
ef3b9af5 FM |
10453 | static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) |
10454 | { | |
2ff7e61e | 10455 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
ef3b9af5 FM |
10456 | struct btrfs_trans_handle *trans; |
10457 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
10458 | struct inode *inode = NULL; | |
10459 | u64 objectid; | |
10460 | u64 index; | |
10461 | int ret = 0; | |
10462 | ||
10463 | /* | |
10464 | * 5 units required for adding orphan entry | |
10465 | */ | |
10466 | trans = btrfs_start_transaction(root, 5); | |
10467 | if (IS_ERR(trans)) | |
10468 | return PTR_ERR(trans); | |
10469 | ||
10470 | ret = btrfs_find_free_ino(root, &objectid); | |
10471 | if (ret) | |
10472 | goto out; | |
10473 | ||
10474 | inode = btrfs_new_inode(trans, root, dir, NULL, 0, | |
f85b7379 | 10475 | btrfs_ino(BTRFS_I(dir)), objectid, mode, &index); |
ef3b9af5 FM |
10476 | if (IS_ERR(inode)) { |
10477 | ret = PTR_ERR(inode); | |
10478 | inode = NULL; | |
10479 | goto out; | |
10480 | } | |
10481 | ||
ef3b9af5 FM |
10482 | inode->i_fop = &btrfs_file_operations; |
10483 | inode->i_op = &btrfs_file_inode_operations; | |
10484 | ||
10485 | inode->i_mapping->a_ops = &btrfs_aops; | |
ef3b9af5 FM |
10486 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
10487 | ||
b0d5d10f CM |
10488 | ret = btrfs_init_inode_security(trans, inode, dir, NULL); |
10489 | if (ret) | |
10490 | goto out_inode; | |
10491 | ||
10492 | ret = btrfs_update_inode(trans, root, inode); | |
10493 | if (ret) | |
10494 | goto out_inode; | |
73f2e545 | 10495 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
ef3b9af5 | 10496 | if (ret) |
b0d5d10f | 10497 | goto out_inode; |
ef3b9af5 | 10498 | |
5762b5c9 FM |
10499 | /* |
10500 | * We set number of links to 0 in btrfs_new_inode(), and here we set | |
10501 | * it to 1 because d_tmpfile() will issue a warning if the count is 0, | |
10502 | * through: | |
10503 | * | |
10504 | * d_tmpfile() -> inode_dec_link_count() -> drop_nlink() | |
10505 | */ | |
10506 | set_nlink(inode, 1); | |
b0d5d10f | 10507 | unlock_new_inode(inode); |
ef3b9af5 FM |
10508 | d_tmpfile(dentry, inode); |
10509 | mark_inode_dirty(inode); | |
10510 | ||
10511 | out: | |
3a45bb20 | 10512 | btrfs_end_transaction(trans); |
ef3b9af5 FM |
10513 | if (ret) |
10514 | iput(inode); | |
2ff7e61e JM |
10515 | btrfs_balance_delayed_items(fs_info); |
10516 | btrfs_btree_balance_dirty(fs_info); | |
ef3b9af5 | 10517 | return ret; |
b0d5d10f CM |
10518 | |
10519 | out_inode: | |
10520 | unlock_new_inode(inode); | |
10521 | goto out; | |
10522 | ||
ef3b9af5 FM |
10523 | } |
10524 | ||
20a7db8a | 10525 | __attribute__((const)) |
9d0d1c8b | 10526 | static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror) |
20a7db8a | 10527 | { |
9d0d1c8b | 10528 | return -EAGAIN; |
20a7db8a DS |
10529 | } |
10530 | ||
6e1d5dcc | 10531 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 10532 | .getattr = btrfs_getattr, |
39279cc3 CM |
10533 | .lookup = btrfs_lookup, |
10534 | .create = btrfs_create, | |
10535 | .unlink = btrfs_unlink, | |
10536 | .link = btrfs_link, | |
10537 | .mkdir = btrfs_mkdir, | |
10538 | .rmdir = btrfs_rmdir, | |
2773bf00 | 10539 | .rename = btrfs_rename2, |
39279cc3 CM |
10540 | .symlink = btrfs_symlink, |
10541 | .setattr = btrfs_setattr, | |
618e21d5 | 10542 | .mknod = btrfs_mknod, |
5103e947 | 10543 | .listxattr = btrfs_listxattr, |
fdebe2bd | 10544 | .permission = btrfs_permission, |
4e34e719 | 10545 | .get_acl = btrfs_get_acl, |
996a710d | 10546 | .set_acl = btrfs_set_acl, |
93fd63c2 | 10547 | .update_time = btrfs_update_time, |
ef3b9af5 | 10548 | .tmpfile = btrfs_tmpfile, |
39279cc3 | 10549 | }; |
6e1d5dcc | 10550 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 10551 | .lookup = btrfs_lookup, |
fdebe2bd | 10552 | .permission = btrfs_permission, |
93fd63c2 | 10553 | .update_time = btrfs_update_time, |
39279cc3 | 10554 | }; |
76dda93c | 10555 | |
828c0950 | 10556 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
10557 | .llseek = generic_file_llseek, |
10558 | .read = generic_read_dir, | |
02dbfc99 | 10559 | .iterate_shared = btrfs_real_readdir, |
34287aa3 | 10560 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 10561 | #ifdef CONFIG_COMPAT |
4c63c245 | 10562 | .compat_ioctl = btrfs_compat_ioctl, |
39279cc3 | 10563 | #endif |
6bf13c0c | 10564 | .release = btrfs_release_file, |
e02119d5 | 10565 | .fsync = btrfs_sync_file, |
39279cc3 CM |
10566 | }; |
10567 | ||
20e5506b | 10568 | static const struct extent_io_ops btrfs_extent_io_ops = { |
4d53dddb | 10569 | /* mandatory callbacks */ |
065631f6 | 10570 | .submit_bio_hook = btrfs_submit_bio_hook, |
07157aac | 10571 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
4d53dddb | 10572 | .merge_bio_hook = btrfs_merge_bio_hook, |
9d0d1c8b | 10573 | .readpage_io_failed_hook = btrfs_readpage_io_failed_hook, |
4d53dddb DS |
10574 | |
10575 | /* optional callbacks */ | |
10576 | .fill_delalloc = run_delalloc_range, | |
e6dcd2dc | 10577 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 10578 | .writepage_start_hook = btrfs_writepage_start_hook, |
b0c68f8b CM |
10579 | .set_bit_hook = btrfs_set_bit_hook, |
10580 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
10581 | .merge_extent_hook = btrfs_merge_extent_hook, |
10582 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
10583 | }; |
10584 | ||
35054394 CM |
10585 | /* |
10586 | * btrfs doesn't support the bmap operation because swapfiles | |
10587 | * use bmap to make a mapping of extents in the file. They assume | |
10588 | * these extents won't change over the life of the file and they | |
10589 | * use the bmap result to do IO directly to the drive. | |
10590 | * | |
10591 | * the btrfs bmap call would return logical addresses that aren't | |
10592 | * suitable for IO and they also will change frequently as COW | |
10593 | * operations happen. So, swapfile + btrfs == corruption. | |
10594 | * | |
10595 | * For now we're avoiding this by dropping bmap. | |
10596 | */ | |
7f09410b | 10597 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
10598 | .readpage = btrfs_readpage, |
10599 | .writepage = btrfs_writepage, | |
b293f02e | 10600 | .writepages = btrfs_writepages, |
3ab2fb5a | 10601 | .readpages = btrfs_readpages, |
16432985 | 10602 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
10603 | .invalidatepage = btrfs_invalidatepage, |
10604 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 10605 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 10606 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
10607 | }; |
10608 | ||
7f09410b | 10609 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
10610 | .readpage = btrfs_readpage, |
10611 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
10612 | .invalidatepage = btrfs_invalidatepage, |
10613 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
10614 | }; |
10615 | ||
6e1d5dcc | 10616 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
10617 | .getattr = btrfs_getattr, |
10618 | .setattr = btrfs_setattr, | |
5103e947 | 10619 | .listxattr = btrfs_listxattr, |
fdebe2bd | 10620 | .permission = btrfs_permission, |
1506fcc8 | 10621 | .fiemap = btrfs_fiemap, |
4e34e719 | 10622 | .get_acl = btrfs_get_acl, |
996a710d | 10623 | .set_acl = btrfs_set_acl, |
e41f941a | 10624 | .update_time = btrfs_update_time, |
39279cc3 | 10625 | }; |
6e1d5dcc | 10626 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
10627 | .getattr = btrfs_getattr, |
10628 | .setattr = btrfs_setattr, | |
fdebe2bd | 10629 | .permission = btrfs_permission, |
33268eaf | 10630 | .listxattr = btrfs_listxattr, |
4e34e719 | 10631 | .get_acl = btrfs_get_acl, |
996a710d | 10632 | .set_acl = btrfs_set_acl, |
e41f941a | 10633 | .update_time = btrfs_update_time, |
618e21d5 | 10634 | }; |
6e1d5dcc | 10635 | static const struct inode_operations btrfs_symlink_inode_operations = { |
6b255391 | 10636 | .get_link = page_get_link, |
f209561a | 10637 | .getattr = btrfs_getattr, |
22c44fe6 | 10638 | .setattr = btrfs_setattr, |
fdebe2bd | 10639 | .permission = btrfs_permission, |
0279b4cd | 10640 | .listxattr = btrfs_listxattr, |
e41f941a | 10641 | .update_time = btrfs_update_time, |
39279cc3 | 10642 | }; |
76dda93c | 10643 | |
82d339d9 | 10644 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c | 10645 | .d_delete = btrfs_dentry_delete, |
b4aff1f8 | 10646 | .d_release = btrfs_dentry_release, |
76dda93c | 10647 | }; |