]>
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; | |
74 | }; | |
75 | ||
6e1d5dcc AD |
76 | static const struct inode_operations btrfs_dir_inode_operations; |
77 | static const struct inode_operations btrfs_symlink_inode_operations; | |
78 | static const struct inode_operations btrfs_dir_ro_inode_operations; | |
79 | static const struct inode_operations btrfs_special_inode_operations; | |
80 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b AD |
81 | static const struct address_space_operations btrfs_aops; |
82 | static const struct address_space_operations btrfs_symlink_aops; | |
828c0950 | 83 | static const struct file_operations btrfs_dir_file_operations; |
20e5506b | 84 | static const struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
85 | |
86 | static struct kmem_cache *btrfs_inode_cachep; | |
87 | struct kmem_cache *btrfs_trans_handle_cachep; | |
88 | struct kmem_cache *btrfs_transaction_cachep; | |
39279cc3 | 89 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 90 | struct kmem_cache *btrfs_free_space_cachep; |
39279cc3 CM |
91 | |
92 | #define S_SHIFT 12 | |
4d4ab6d6 | 93 | static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { |
39279cc3 CM |
94 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, |
95 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
96 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
97 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
98 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
99 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
100 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
101 | }; | |
102 | ||
3972f260 | 103 | static int btrfs_setsize(struct inode *inode, struct iattr *attr); |
a41ad394 | 104 | static int btrfs_truncate(struct inode *inode); |
5fd02043 | 105 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent); |
771ed689 CM |
106 | static noinline int cow_file_range(struct inode *inode, |
107 | struct page *locked_page, | |
dda3245e WX |
108 | u64 start, u64 end, u64 delalloc_end, |
109 | int *page_started, unsigned long *nr_written, | |
110 | int unlock, struct btrfs_dedupe_hash *hash); | |
70c8a91c JB |
111 | static struct extent_map *create_pinned_em(struct inode *inode, u64 start, |
112 | u64 len, u64 orig_start, | |
113 | u64 block_start, u64 block_len, | |
cc95bef6 JB |
114 | u64 orig_block_len, u64 ram_bytes, |
115 | int type); | |
7b128766 | 116 | |
48a3b636 | 117 | static int btrfs_dirty_inode(struct inode *inode); |
7b128766 | 118 | |
6a3891c5 JB |
119 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
120 | void btrfs_test_inode_set_ops(struct inode *inode) | |
121 | { | |
122 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; | |
123 | } | |
124 | #endif | |
125 | ||
f34f57a3 | 126 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
2a7dba39 EP |
127 | struct inode *inode, struct inode *dir, |
128 | const struct qstr *qstr) | |
0279b4cd JO |
129 | { |
130 | int err; | |
131 | ||
f34f57a3 | 132 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 133 | if (!err) |
2a7dba39 | 134 | err = btrfs_xattr_security_init(trans, inode, dir, qstr); |
0279b4cd JO |
135 | return err; |
136 | } | |
137 | ||
c8b97818 CM |
138 | /* |
139 | * this does all the hard work for inserting an inline extent into | |
140 | * the btree. The caller should have done a btrfs_drop_extents so that | |
141 | * no overlapping inline items exist in the btree | |
142 | */ | |
40f76580 | 143 | static int insert_inline_extent(struct btrfs_trans_handle *trans, |
1acae57b | 144 | struct btrfs_path *path, int extent_inserted, |
c8b97818 CM |
145 | struct btrfs_root *root, struct inode *inode, |
146 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 147 | int compress_type, |
c8b97818 CM |
148 | struct page **compressed_pages) |
149 | { | |
c8b97818 CM |
150 | struct extent_buffer *leaf; |
151 | struct page *page = NULL; | |
152 | char *kaddr; | |
153 | unsigned long ptr; | |
154 | struct btrfs_file_extent_item *ei; | |
155 | int err = 0; | |
156 | int ret; | |
157 | size_t cur_size = size; | |
c8b97818 | 158 | unsigned long offset; |
c8b97818 | 159 | |
fe3f566c | 160 | if (compressed_size && compressed_pages) |
c8b97818 | 161 | cur_size = compressed_size; |
c8b97818 | 162 | |
1acae57b | 163 | inode_add_bytes(inode, size); |
c8b97818 | 164 | |
1acae57b FDBM |
165 | if (!extent_inserted) { |
166 | struct btrfs_key key; | |
167 | size_t datasize; | |
c8b97818 | 168 | |
4a0cc7ca | 169 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
1acae57b | 170 | key.offset = start; |
962a298f | 171 | key.type = BTRFS_EXTENT_DATA_KEY; |
c8b97818 | 172 | |
1acae57b FDBM |
173 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
174 | path->leave_spinning = 1; | |
175 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
176 | datasize); | |
177 | if (ret) { | |
178 | err = ret; | |
179 | goto fail; | |
180 | } | |
c8b97818 CM |
181 | } |
182 | leaf = path->nodes[0]; | |
183 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
184 | struct btrfs_file_extent_item); | |
185 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
186 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
187 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
188 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
189 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
190 | ptr = btrfs_file_extent_inline_start(ei); | |
191 | ||
261507a0 | 192 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
193 | struct page *cpage; |
194 | int i = 0; | |
d397712b | 195 | while (compressed_size > 0) { |
c8b97818 | 196 | cpage = compressed_pages[i]; |
5b050f04 | 197 | cur_size = min_t(unsigned long, compressed_size, |
09cbfeaf | 198 | PAGE_SIZE); |
c8b97818 | 199 | |
7ac687d9 | 200 | kaddr = kmap_atomic(cpage); |
c8b97818 | 201 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
7ac687d9 | 202 | kunmap_atomic(kaddr); |
c8b97818 CM |
203 | |
204 | i++; | |
205 | ptr += cur_size; | |
206 | compressed_size -= cur_size; | |
207 | } | |
208 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 209 | compress_type); |
c8b97818 CM |
210 | } else { |
211 | page = find_get_page(inode->i_mapping, | |
09cbfeaf | 212 | start >> PAGE_SHIFT); |
c8b97818 | 213 | btrfs_set_file_extent_compression(leaf, ei, 0); |
7ac687d9 | 214 | kaddr = kmap_atomic(page); |
09cbfeaf | 215 | offset = start & (PAGE_SIZE - 1); |
c8b97818 | 216 | write_extent_buffer(leaf, kaddr + offset, ptr, size); |
7ac687d9 | 217 | kunmap_atomic(kaddr); |
09cbfeaf | 218 | put_page(page); |
c8b97818 CM |
219 | } |
220 | btrfs_mark_buffer_dirty(leaf); | |
1acae57b | 221 | btrfs_release_path(path); |
c8b97818 | 222 | |
c2167754 YZ |
223 | /* |
224 | * we're an inline extent, so nobody can | |
225 | * extend the file past i_size without locking | |
226 | * a page we already have locked. | |
227 | * | |
228 | * We must do any isize and inode updates | |
229 | * before we unlock the pages. Otherwise we | |
230 | * could end up racing with unlink. | |
231 | */ | |
c8b97818 | 232 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
79787eaa | 233 | ret = btrfs_update_inode(trans, root, inode); |
c2167754 | 234 | |
79787eaa | 235 | return ret; |
c8b97818 | 236 | fail: |
c8b97818 CM |
237 | return err; |
238 | } | |
239 | ||
240 | ||
241 | /* | |
242 | * conditionally insert an inline extent into the file. This | |
243 | * does the checks required to make sure the data is small enough | |
244 | * to fit as an inline extent. | |
245 | */ | |
00361589 JB |
246 | static noinline int cow_file_range_inline(struct btrfs_root *root, |
247 | struct inode *inode, u64 start, | |
248 | u64 end, size_t compressed_size, | |
249 | int compress_type, | |
250 | struct page **compressed_pages) | |
c8b97818 | 251 | { |
0b246afa | 252 | struct btrfs_fs_info *fs_info = root->fs_info; |
00361589 | 253 | struct btrfs_trans_handle *trans; |
c8b97818 CM |
254 | u64 isize = i_size_read(inode); |
255 | u64 actual_end = min(end + 1, isize); | |
256 | u64 inline_len = actual_end - start; | |
0b246afa | 257 | u64 aligned_end = ALIGN(end, fs_info->sectorsize); |
c8b97818 CM |
258 | u64 data_len = inline_len; |
259 | int ret; | |
1acae57b FDBM |
260 | struct btrfs_path *path; |
261 | int extent_inserted = 0; | |
262 | u32 extent_item_size; | |
c8b97818 CM |
263 | |
264 | if (compressed_size) | |
265 | data_len = compressed_size; | |
266 | ||
267 | if (start > 0 || | |
0b246afa JM |
268 | actual_end > fs_info->sectorsize || |
269 | data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) || | |
c8b97818 | 270 | (!compressed_size && |
0b246afa | 271 | (actual_end & (fs_info->sectorsize - 1)) == 0) || |
c8b97818 | 272 | end + 1 < isize || |
0b246afa | 273 | data_len > fs_info->max_inline) { |
c8b97818 CM |
274 | return 1; |
275 | } | |
276 | ||
1acae57b FDBM |
277 | path = btrfs_alloc_path(); |
278 | if (!path) | |
279 | return -ENOMEM; | |
280 | ||
00361589 | 281 | trans = btrfs_join_transaction(root); |
1acae57b FDBM |
282 | if (IS_ERR(trans)) { |
283 | btrfs_free_path(path); | |
00361589 | 284 | return PTR_ERR(trans); |
1acae57b | 285 | } |
0b246afa | 286 | trans->block_rsv = &fs_info->delalloc_block_rsv; |
00361589 | 287 | |
1acae57b FDBM |
288 | if (compressed_size && compressed_pages) |
289 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
290 | compressed_size); | |
291 | else | |
292 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
293 | inline_len); | |
294 | ||
295 | ret = __btrfs_drop_extents(trans, root, inode, path, | |
296 | start, aligned_end, NULL, | |
297 | 1, 1, extent_item_size, &extent_inserted); | |
00361589 | 298 | if (ret) { |
66642832 | 299 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
300 | goto out; |
301 | } | |
c8b97818 CM |
302 | |
303 | if (isize > actual_end) | |
304 | inline_len = min_t(u64, isize, actual_end); | |
1acae57b FDBM |
305 | ret = insert_inline_extent(trans, path, extent_inserted, |
306 | root, inode, start, | |
c8b97818 | 307 | inline_len, compressed_size, |
fe3f566c | 308 | compress_type, compressed_pages); |
2adcac1a | 309 | if (ret && ret != -ENOSPC) { |
66642832 | 310 | btrfs_abort_transaction(trans, ret); |
00361589 | 311 | goto out; |
2adcac1a | 312 | } else if (ret == -ENOSPC) { |
00361589 JB |
313 | ret = 1; |
314 | goto out; | |
79787eaa | 315 | } |
2adcac1a | 316 | |
bdc20e67 | 317 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); |
0ca1f7ce | 318 | btrfs_delalloc_release_metadata(inode, end + 1 - start); |
a1ed835e | 319 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
00361589 | 320 | out: |
94ed938a QW |
321 | /* |
322 | * Don't forget to free the reserved space, as for inlined extent | |
323 | * it won't count as data extent, free them directly here. | |
324 | * And at reserve time, it's always aligned to page size, so | |
325 | * just free one page here. | |
326 | */ | |
09cbfeaf | 327 | btrfs_qgroup_free_data(inode, 0, PAGE_SIZE); |
1acae57b | 328 | btrfs_free_path(path); |
3a45bb20 | 329 | btrfs_end_transaction(trans); |
00361589 | 330 | return ret; |
c8b97818 CM |
331 | } |
332 | ||
771ed689 CM |
333 | struct async_extent { |
334 | u64 start; | |
335 | u64 ram_size; | |
336 | u64 compressed_size; | |
337 | struct page **pages; | |
338 | unsigned long nr_pages; | |
261507a0 | 339 | int compress_type; |
771ed689 CM |
340 | struct list_head list; |
341 | }; | |
342 | ||
343 | struct async_cow { | |
344 | struct inode *inode; | |
345 | struct btrfs_root *root; | |
346 | struct page *locked_page; | |
347 | u64 start; | |
348 | u64 end; | |
349 | struct list_head extents; | |
350 | struct btrfs_work work; | |
351 | }; | |
352 | ||
353 | static noinline int add_async_extent(struct async_cow *cow, | |
354 | u64 start, u64 ram_size, | |
355 | u64 compressed_size, | |
356 | struct page **pages, | |
261507a0 LZ |
357 | unsigned long nr_pages, |
358 | int compress_type) | |
771ed689 CM |
359 | { |
360 | struct async_extent *async_extent; | |
361 | ||
362 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 363 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
364 | async_extent->start = start; |
365 | async_extent->ram_size = ram_size; | |
366 | async_extent->compressed_size = compressed_size; | |
367 | async_extent->pages = pages; | |
368 | async_extent->nr_pages = nr_pages; | |
261507a0 | 369 | async_extent->compress_type = compress_type; |
771ed689 CM |
370 | list_add_tail(&async_extent->list, &cow->extents); |
371 | return 0; | |
372 | } | |
373 | ||
f79707b0 WS |
374 | static inline int inode_need_compress(struct inode *inode) |
375 | { | |
0b246afa | 376 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
f79707b0 WS |
377 | |
378 | /* force compress */ | |
0b246afa | 379 | if (btrfs_test_opt(fs_info, FORCE_COMPRESS)) |
f79707b0 WS |
380 | return 1; |
381 | /* bad compression ratios */ | |
382 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) | |
383 | return 0; | |
0b246afa | 384 | if (btrfs_test_opt(fs_info, COMPRESS) || |
f79707b0 WS |
385 | BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS || |
386 | BTRFS_I(inode)->force_compress) | |
387 | return 1; | |
388 | return 0; | |
389 | } | |
390 | ||
26d30f85 AJ |
391 | static inline void inode_should_defrag(struct inode *inode, |
392 | u64 start, u64 end, u64 num_bytes, u64 small_write) | |
393 | { | |
394 | /* If this is a small write inside eof, kick off a defrag */ | |
395 | if (num_bytes < small_write && | |
396 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
397 | btrfs_add_inode_defrag(NULL, inode); | |
398 | } | |
399 | ||
d352ac68 | 400 | /* |
771ed689 CM |
401 | * we create compressed extents in two phases. The first |
402 | * phase compresses a range of pages that have already been | |
403 | * locked (both pages and state bits are locked). | |
c8b97818 | 404 | * |
771ed689 CM |
405 | * This is done inside an ordered work queue, and the compression |
406 | * is spread across many cpus. The actual IO submission is step | |
407 | * two, and the ordered work queue takes care of making sure that | |
408 | * happens in the same order things were put onto the queue by | |
409 | * writepages and friends. | |
c8b97818 | 410 | * |
771ed689 CM |
411 | * If this code finds it can't get good compression, it puts an |
412 | * entry onto the work queue to write the uncompressed bytes. This | |
413 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
414 | * are written in the same order that the flusher thread sent them |
415 | * down. | |
d352ac68 | 416 | */ |
c44f649e | 417 | static noinline void compress_file_range(struct inode *inode, |
771ed689 CM |
418 | struct page *locked_page, |
419 | u64 start, u64 end, | |
420 | struct async_cow *async_cow, | |
421 | int *num_added) | |
b888db2b | 422 | { |
0b246afa | 423 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
b888db2b | 424 | struct btrfs_root *root = BTRFS_I(inode)->root; |
db94535d | 425 | u64 num_bytes; |
0b246afa | 426 | u64 blocksize = fs_info->sectorsize; |
c8b97818 | 427 | u64 actual_end; |
42dc7bab | 428 | u64 isize = i_size_read(inode); |
e6dcd2dc | 429 | int ret = 0; |
c8b97818 CM |
430 | struct page **pages = NULL; |
431 | unsigned long nr_pages; | |
432 | unsigned long nr_pages_ret = 0; | |
433 | unsigned long total_compressed = 0; | |
434 | unsigned long total_in = 0; | |
ee22184b BL |
435 | unsigned long max_compressed = SZ_128K; |
436 | unsigned long max_uncompressed = SZ_128K; | |
c8b97818 CM |
437 | int i; |
438 | int will_compress; | |
0b246afa | 439 | int compress_type = fs_info->compress_type; |
4adaa611 | 440 | int redirty = 0; |
b888db2b | 441 | |
26d30f85 | 442 | inode_should_defrag(inode, start, end, end - start + 1, SZ_16K); |
4cb5300b | 443 | |
42dc7bab | 444 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
445 | again: |
446 | will_compress = 0; | |
09cbfeaf KS |
447 | nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1; |
448 | nr_pages = min_t(unsigned long, nr_pages, SZ_128K / PAGE_SIZE); | |
be20aa9d | 449 | |
f03d9301 CM |
450 | /* |
451 | * we don't want to send crud past the end of i_size through | |
452 | * compression, that's just a waste of CPU time. So, if the | |
453 | * end of the file is before the start of our current | |
454 | * requested range of bytes, we bail out to the uncompressed | |
455 | * cleanup code that can deal with all of this. | |
456 | * | |
457 | * It isn't really the fastest way to fix things, but this is a | |
458 | * very uncommon corner. | |
459 | */ | |
460 | if (actual_end <= start) | |
461 | goto cleanup_and_bail_uncompressed; | |
462 | ||
c8b97818 CM |
463 | total_compressed = actual_end - start; |
464 | ||
4bcbb332 SW |
465 | /* |
466 | * skip compression for a small file range(<=blocksize) that | |
01327610 | 467 | * isn't an inline extent, since it doesn't save disk space at all. |
4bcbb332 SW |
468 | */ |
469 | if (total_compressed <= blocksize && | |
470 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
471 | goto cleanup_and_bail_uncompressed; | |
472 | ||
c8b97818 CM |
473 | /* we want to make sure that amount of ram required to uncompress |
474 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
475 | * of a compressed extent to 128k. This is a crucial number |
476 | * because it also controls how easily we can spread reads across | |
477 | * cpus for decompression. | |
478 | * | |
479 | * We also want to make sure the amount of IO required to do | |
480 | * a random read is reasonably small, so we limit the size of | |
481 | * a compressed extent to 128k. | |
c8b97818 CM |
482 | */ |
483 | total_compressed = min(total_compressed, max_uncompressed); | |
fda2832f | 484 | num_bytes = ALIGN(end - start + 1, blocksize); |
be20aa9d | 485 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
486 | total_in = 0; |
487 | ret = 0; | |
db94535d | 488 | |
771ed689 CM |
489 | /* |
490 | * we do compression for mount -o compress and when the | |
491 | * inode has not been flagged as nocompress. This flag can | |
492 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 493 | */ |
f79707b0 | 494 | if (inode_need_compress(inode)) { |
c8b97818 | 495 | WARN_ON(pages); |
31e818fe | 496 | pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS); |
560f7d75 LZ |
497 | if (!pages) { |
498 | /* just bail out to the uncompressed code */ | |
499 | goto cont; | |
500 | } | |
c8b97818 | 501 | |
261507a0 LZ |
502 | if (BTRFS_I(inode)->force_compress) |
503 | compress_type = BTRFS_I(inode)->force_compress; | |
504 | ||
4adaa611 CM |
505 | /* |
506 | * we need to call clear_page_dirty_for_io on each | |
507 | * page in the range. Otherwise applications with the file | |
508 | * mmap'd can wander in and change the page contents while | |
509 | * we are compressing them. | |
510 | * | |
511 | * If the compression fails for any reason, we set the pages | |
512 | * dirty again later on. | |
513 | */ | |
514 | extent_range_clear_dirty_for_io(inode, start, end); | |
515 | redirty = 1; | |
261507a0 LZ |
516 | ret = btrfs_compress_pages(compress_type, |
517 | inode->i_mapping, start, | |
518 | total_compressed, pages, | |
519 | nr_pages, &nr_pages_ret, | |
520 | &total_in, | |
521 | &total_compressed, | |
522 | max_compressed); | |
c8b97818 CM |
523 | |
524 | if (!ret) { | |
525 | unsigned long offset = total_compressed & | |
09cbfeaf | 526 | (PAGE_SIZE - 1); |
c8b97818 CM |
527 | struct page *page = pages[nr_pages_ret - 1]; |
528 | char *kaddr; | |
529 | ||
530 | /* zero the tail end of the last page, we might be | |
531 | * sending it down to disk | |
532 | */ | |
533 | if (offset) { | |
7ac687d9 | 534 | kaddr = kmap_atomic(page); |
c8b97818 | 535 | memset(kaddr + offset, 0, |
09cbfeaf | 536 | PAGE_SIZE - offset); |
7ac687d9 | 537 | kunmap_atomic(kaddr); |
c8b97818 CM |
538 | } |
539 | will_compress = 1; | |
540 | } | |
541 | } | |
560f7d75 | 542 | cont: |
c8b97818 CM |
543 | if (start == 0) { |
544 | /* lets try to make an inline extent */ | |
771ed689 | 545 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 546 | /* we didn't compress the entire range, try |
771ed689 | 547 | * to make an uncompressed inline extent. |
c8b97818 | 548 | */ |
00361589 | 549 | ret = cow_file_range_inline(root, inode, start, end, |
f74670f7 | 550 | 0, BTRFS_COMPRESS_NONE, NULL); |
c8b97818 | 551 | } else { |
771ed689 | 552 | /* try making a compressed inline extent */ |
00361589 | 553 | ret = cow_file_range_inline(root, inode, start, end, |
fe3f566c LZ |
554 | total_compressed, |
555 | compress_type, pages); | |
c8b97818 | 556 | } |
79787eaa | 557 | if (ret <= 0) { |
151a41bc JB |
558 | unsigned long clear_flags = EXTENT_DELALLOC | |
559 | EXTENT_DEFRAG; | |
e6eb4314 FM |
560 | unsigned long page_error_op; |
561 | ||
151a41bc | 562 | clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0; |
e6eb4314 | 563 | page_error_op = ret < 0 ? PAGE_SET_ERROR : 0; |
151a41bc | 564 | |
771ed689 | 565 | /* |
79787eaa JM |
566 | * inline extent creation worked or returned error, |
567 | * we don't need to create any more async work items. | |
568 | * Unlock and free up our temp pages. | |
771ed689 | 569 | */ |
ba8b04c1 QW |
570 | extent_clear_unlock_delalloc(inode, start, end, end, |
571 | NULL, clear_flags, | |
572 | PAGE_UNLOCK | | |
c2790a2e JB |
573 | PAGE_CLEAR_DIRTY | |
574 | PAGE_SET_WRITEBACK | | |
e6eb4314 | 575 | page_error_op | |
c2790a2e | 576 | PAGE_END_WRITEBACK); |
18513091 WX |
577 | btrfs_free_reserved_data_space_noquota(inode, start, |
578 | end - start + 1); | |
c8b97818 CM |
579 | goto free_pages_out; |
580 | } | |
581 | } | |
582 | ||
583 | if (will_compress) { | |
584 | /* | |
585 | * we aren't doing an inline extent round the compressed size | |
586 | * up to a block size boundary so the allocator does sane | |
587 | * things | |
588 | */ | |
fda2832f | 589 | total_compressed = ALIGN(total_compressed, blocksize); |
c8b97818 CM |
590 | |
591 | /* | |
592 | * one last check to make sure the compression is really a | |
593 | * win, compare the page count read with the blocks on disk | |
594 | */ | |
09cbfeaf | 595 | total_in = ALIGN(total_in, PAGE_SIZE); |
c8b97818 CM |
596 | if (total_compressed >= total_in) { |
597 | will_compress = 0; | |
598 | } else { | |
c8b97818 | 599 | num_bytes = total_in; |
c8bb0c8b AS |
600 | *num_added += 1; |
601 | ||
602 | /* | |
603 | * The async work queues will take care of doing actual | |
604 | * allocation on disk for these compressed pages, and | |
605 | * will submit them to the elevator. | |
606 | */ | |
607 | add_async_extent(async_cow, start, num_bytes, | |
608 | total_compressed, pages, nr_pages_ret, | |
609 | compress_type); | |
610 | ||
611 | if (start + num_bytes < end) { | |
612 | start += num_bytes; | |
613 | pages = NULL; | |
614 | cond_resched(); | |
615 | goto again; | |
616 | } | |
617 | return; | |
c8b97818 CM |
618 | } |
619 | } | |
c8bb0c8b | 620 | if (pages) { |
c8b97818 CM |
621 | /* |
622 | * the compression code ran but failed to make things smaller, | |
623 | * free any pages it allocated and our page pointer array | |
624 | */ | |
625 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 626 | WARN_ON(pages[i]->mapping); |
09cbfeaf | 627 | put_page(pages[i]); |
c8b97818 CM |
628 | } |
629 | kfree(pages); | |
630 | pages = NULL; | |
631 | total_compressed = 0; | |
632 | nr_pages_ret = 0; | |
633 | ||
634 | /* flag the file so we don't compress in the future */ | |
0b246afa | 635 | if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) && |
1e701a32 | 636 | !(BTRFS_I(inode)->force_compress)) { |
a555f810 | 637 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 638 | } |
c8b97818 | 639 | } |
f03d9301 | 640 | cleanup_and_bail_uncompressed: |
c8bb0c8b AS |
641 | /* |
642 | * No compression, but we still need to write the pages in the file | |
643 | * we've been given so far. redirty the locked page if it corresponds | |
644 | * to our extent and set things up for the async work queue to run | |
645 | * cow_file_range to do the normal delalloc dance. | |
646 | */ | |
647 | if (page_offset(locked_page) >= start && | |
648 | page_offset(locked_page) <= end) | |
649 | __set_page_dirty_nobuffers(locked_page); | |
650 | /* unlocked later on in the async handlers */ | |
651 | ||
652 | if (redirty) | |
653 | extent_range_redirty_for_io(inode, start, end); | |
654 | add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0, | |
655 | BTRFS_COMPRESS_NONE); | |
656 | *num_added += 1; | |
3b951516 | 657 | |
c44f649e | 658 | return; |
771ed689 CM |
659 | |
660 | free_pages_out: | |
661 | for (i = 0; i < nr_pages_ret; i++) { | |
662 | WARN_ON(pages[i]->mapping); | |
09cbfeaf | 663 | put_page(pages[i]); |
771ed689 | 664 | } |
d397712b | 665 | kfree(pages); |
771ed689 | 666 | } |
771ed689 | 667 | |
40ae837b FM |
668 | static void free_async_extent_pages(struct async_extent *async_extent) |
669 | { | |
670 | int i; | |
671 | ||
672 | if (!async_extent->pages) | |
673 | return; | |
674 | ||
675 | for (i = 0; i < async_extent->nr_pages; i++) { | |
676 | WARN_ON(async_extent->pages[i]->mapping); | |
09cbfeaf | 677 | put_page(async_extent->pages[i]); |
40ae837b FM |
678 | } |
679 | kfree(async_extent->pages); | |
680 | async_extent->nr_pages = 0; | |
681 | async_extent->pages = NULL; | |
771ed689 CM |
682 | } |
683 | ||
684 | /* | |
685 | * phase two of compressed writeback. This is the ordered portion | |
686 | * of the code, which only gets called in the order the work was | |
687 | * queued. We walk all the async extents created by compress_file_range | |
688 | * and send them down to the disk. | |
689 | */ | |
dec8f175 | 690 | static noinline void submit_compressed_extents(struct inode *inode, |
771ed689 CM |
691 | struct async_cow *async_cow) |
692 | { | |
0b246afa | 693 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
771ed689 CM |
694 | struct async_extent *async_extent; |
695 | u64 alloc_hint = 0; | |
771ed689 CM |
696 | struct btrfs_key ins; |
697 | struct extent_map *em; | |
698 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
699 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
700 | struct extent_io_tree *io_tree; | |
f5a84ee3 | 701 | int ret = 0; |
771ed689 | 702 | |
3e04e7f1 | 703 | again: |
d397712b | 704 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
705 | async_extent = list_entry(async_cow->extents.next, |
706 | struct async_extent, list); | |
707 | list_del(&async_extent->list); | |
c8b97818 | 708 | |
771ed689 CM |
709 | io_tree = &BTRFS_I(inode)->io_tree; |
710 | ||
f5a84ee3 | 711 | retry: |
771ed689 CM |
712 | /* did the compression code fall back to uncompressed IO? */ |
713 | if (!async_extent->pages) { | |
714 | int page_started = 0; | |
715 | unsigned long nr_written = 0; | |
716 | ||
717 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 | 718 | async_extent->start + |
d0082371 | 719 | async_extent->ram_size - 1); |
771ed689 CM |
720 | |
721 | /* allocate blocks */ | |
f5a84ee3 JB |
722 | ret = cow_file_range(inode, async_cow->locked_page, |
723 | async_extent->start, | |
724 | async_extent->start + | |
725 | async_extent->ram_size - 1, | |
dda3245e WX |
726 | async_extent->start + |
727 | async_extent->ram_size - 1, | |
728 | &page_started, &nr_written, 0, | |
729 | NULL); | |
771ed689 | 730 | |
79787eaa JM |
731 | /* JDM XXX */ |
732 | ||
771ed689 CM |
733 | /* |
734 | * if page_started, cow_file_range inserted an | |
735 | * inline extent and took care of all the unlocking | |
736 | * and IO for us. Otherwise, we need to submit | |
737 | * all those pages down to the drive. | |
738 | */ | |
f5a84ee3 | 739 | if (!page_started && !ret) |
771ed689 CM |
740 | extent_write_locked_range(io_tree, |
741 | inode, async_extent->start, | |
d397712b | 742 | async_extent->start + |
771ed689 CM |
743 | async_extent->ram_size - 1, |
744 | btrfs_get_extent, | |
745 | WB_SYNC_ALL); | |
3e04e7f1 JB |
746 | else if (ret) |
747 | unlock_page(async_cow->locked_page); | |
771ed689 CM |
748 | kfree(async_extent); |
749 | cond_resched(); | |
750 | continue; | |
751 | } | |
752 | ||
753 | lock_extent(io_tree, async_extent->start, | |
d0082371 | 754 | async_extent->start + async_extent->ram_size - 1); |
771ed689 | 755 | |
18513091 | 756 | ret = btrfs_reserve_extent(root, async_extent->ram_size, |
771ed689 CM |
757 | async_extent->compressed_size, |
758 | async_extent->compressed_size, | |
e570fd27 | 759 | 0, alloc_hint, &ins, 1, 1); |
f5a84ee3 | 760 | if (ret) { |
40ae837b | 761 | free_async_extent_pages(async_extent); |
3e04e7f1 | 762 | |
fdf8e2ea JB |
763 | if (ret == -ENOSPC) { |
764 | unlock_extent(io_tree, async_extent->start, | |
765 | async_extent->start + | |
766 | async_extent->ram_size - 1); | |
ce62003f LB |
767 | |
768 | /* | |
769 | * we need to redirty the pages if we decide to | |
770 | * fallback to uncompressed IO, otherwise we | |
771 | * will not submit these pages down to lower | |
772 | * layers. | |
773 | */ | |
774 | extent_range_redirty_for_io(inode, | |
775 | async_extent->start, | |
776 | async_extent->start + | |
777 | async_extent->ram_size - 1); | |
778 | ||
79787eaa | 779 | goto retry; |
fdf8e2ea | 780 | } |
3e04e7f1 | 781 | goto out_free; |
f5a84ee3 | 782 | } |
c2167754 YZ |
783 | /* |
784 | * here we're doing allocation and writeback of the | |
785 | * compressed pages | |
786 | */ | |
787 | btrfs_drop_extent_cache(inode, async_extent->start, | |
788 | async_extent->start + | |
789 | async_extent->ram_size - 1, 0); | |
790 | ||
172ddd60 | 791 | em = alloc_extent_map(); |
b9aa55be LB |
792 | if (!em) { |
793 | ret = -ENOMEM; | |
3e04e7f1 | 794 | goto out_free_reserve; |
b9aa55be | 795 | } |
771ed689 CM |
796 | em->start = async_extent->start; |
797 | em->len = async_extent->ram_size; | |
445a6944 | 798 | em->orig_start = em->start; |
2ab28f32 JB |
799 | em->mod_start = em->start; |
800 | em->mod_len = em->len; | |
c8b97818 | 801 | |
771ed689 CM |
802 | em->block_start = ins.objectid; |
803 | em->block_len = ins.offset; | |
b4939680 | 804 | em->orig_block_len = ins.offset; |
cc95bef6 | 805 | em->ram_bytes = async_extent->ram_size; |
0b246afa | 806 | em->bdev = fs_info->fs_devices->latest_bdev; |
261507a0 | 807 | em->compress_type = async_extent->compress_type; |
771ed689 CM |
808 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
809 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
70c8a91c | 810 | em->generation = -1; |
771ed689 | 811 | |
d397712b | 812 | while (1) { |
890871be | 813 | write_lock(&em_tree->lock); |
09a2a8f9 | 814 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 815 | write_unlock(&em_tree->lock); |
771ed689 CM |
816 | if (ret != -EEXIST) { |
817 | free_extent_map(em); | |
818 | break; | |
819 | } | |
820 | btrfs_drop_extent_cache(inode, async_extent->start, | |
821 | async_extent->start + | |
822 | async_extent->ram_size - 1, 0); | |
823 | } | |
824 | ||
3e04e7f1 JB |
825 | if (ret) |
826 | goto out_free_reserve; | |
827 | ||
261507a0 LZ |
828 | ret = btrfs_add_ordered_extent_compress(inode, |
829 | async_extent->start, | |
830 | ins.objectid, | |
831 | async_extent->ram_size, | |
832 | ins.offset, | |
833 | BTRFS_ORDERED_COMPRESSED, | |
834 | async_extent->compress_type); | |
d9f85963 FM |
835 | if (ret) { |
836 | btrfs_drop_extent_cache(inode, async_extent->start, | |
837 | async_extent->start + | |
838 | async_extent->ram_size - 1, 0); | |
3e04e7f1 | 839 | goto out_free_reserve; |
d9f85963 | 840 | } |
0b246afa | 841 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
771ed689 | 842 | |
771ed689 CM |
843 | /* |
844 | * clear dirty, set writeback and unlock the pages. | |
845 | */ | |
c2790a2e | 846 | extent_clear_unlock_delalloc(inode, async_extent->start, |
ba8b04c1 QW |
847 | async_extent->start + |
848 | async_extent->ram_size - 1, | |
a791e35e CM |
849 | async_extent->start + |
850 | async_extent->ram_size - 1, | |
151a41bc JB |
851 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC, |
852 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
c2790a2e | 853 | PAGE_SET_WRITEBACK); |
771ed689 | 854 | ret = btrfs_submit_compressed_write(inode, |
d397712b CM |
855 | async_extent->start, |
856 | async_extent->ram_size, | |
857 | ins.objectid, | |
858 | ins.offset, async_extent->pages, | |
859 | async_extent->nr_pages); | |
fce2a4e6 FM |
860 | if (ret) { |
861 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
862 | struct page *p = async_extent->pages[0]; | |
863 | const u64 start = async_extent->start; | |
864 | const u64 end = start + async_extent->ram_size - 1; | |
865 | ||
866 | p->mapping = inode->i_mapping; | |
867 | tree->ops->writepage_end_io_hook(p, start, end, | |
868 | NULL, 0); | |
869 | p->mapping = NULL; | |
ba8b04c1 QW |
870 | extent_clear_unlock_delalloc(inode, start, end, end, |
871 | NULL, 0, | |
fce2a4e6 FM |
872 | PAGE_END_WRITEBACK | |
873 | PAGE_SET_ERROR); | |
40ae837b | 874 | free_async_extent_pages(async_extent); |
fce2a4e6 | 875 | } |
771ed689 CM |
876 | alloc_hint = ins.objectid + ins.offset; |
877 | kfree(async_extent); | |
878 | cond_resched(); | |
879 | } | |
dec8f175 | 880 | return; |
3e04e7f1 | 881 | out_free_reserve: |
0b246afa | 882 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 883 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 884 | out_free: |
c2790a2e | 885 | extent_clear_unlock_delalloc(inode, async_extent->start, |
ba8b04c1 QW |
886 | async_extent->start + |
887 | async_extent->ram_size - 1, | |
3e04e7f1 JB |
888 | async_extent->start + |
889 | async_extent->ram_size - 1, | |
c2790a2e | 890 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc JB |
891 | EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING, |
892 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
704de49d FM |
893 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK | |
894 | PAGE_SET_ERROR); | |
40ae837b | 895 | free_async_extent_pages(async_extent); |
79787eaa | 896 | kfree(async_extent); |
3e04e7f1 | 897 | goto again; |
771ed689 CM |
898 | } |
899 | ||
4b46fce2 JB |
900 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
901 | u64 num_bytes) | |
902 | { | |
903 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
904 | struct extent_map *em; | |
905 | u64 alloc_hint = 0; | |
906 | ||
907 | read_lock(&em_tree->lock); | |
908 | em = search_extent_mapping(em_tree, start, num_bytes); | |
909 | if (em) { | |
910 | /* | |
911 | * if block start isn't an actual block number then find the | |
912 | * first block in this inode and use that as a hint. If that | |
913 | * block is also bogus then just don't worry about it. | |
914 | */ | |
915 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
916 | free_extent_map(em); | |
917 | em = search_extent_mapping(em_tree, 0, 0); | |
918 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
919 | alloc_hint = em->block_start; | |
920 | if (em) | |
921 | free_extent_map(em); | |
922 | } else { | |
923 | alloc_hint = em->block_start; | |
924 | free_extent_map(em); | |
925 | } | |
926 | } | |
927 | read_unlock(&em_tree->lock); | |
928 | ||
929 | return alloc_hint; | |
930 | } | |
931 | ||
771ed689 CM |
932 | /* |
933 | * when extent_io.c finds a delayed allocation range in the file, | |
934 | * the call backs end up in this code. The basic idea is to | |
935 | * allocate extents on disk for the range, and create ordered data structs | |
936 | * in ram to track those extents. | |
937 | * | |
938 | * locked_page is the page that writepage had locked already. We use | |
939 | * it to make sure we don't do extra locks or unlocks. | |
940 | * | |
941 | * *page_started is set to one if we unlock locked_page and do everything | |
942 | * required to start IO on it. It may be clean and already done with | |
943 | * IO when we return. | |
944 | */ | |
00361589 JB |
945 | static noinline int cow_file_range(struct inode *inode, |
946 | struct page *locked_page, | |
dda3245e WX |
947 | u64 start, u64 end, u64 delalloc_end, |
948 | int *page_started, unsigned long *nr_written, | |
949 | int unlock, struct btrfs_dedupe_hash *hash) | |
771ed689 | 950 | { |
0b246afa | 951 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
00361589 | 952 | struct btrfs_root *root = BTRFS_I(inode)->root; |
771ed689 CM |
953 | u64 alloc_hint = 0; |
954 | u64 num_bytes; | |
955 | unsigned long ram_size; | |
956 | u64 disk_num_bytes; | |
957 | u64 cur_alloc_size; | |
0b246afa | 958 | u64 blocksize = fs_info->sectorsize; |
771ed689 CM |
959 | struct btrfs_key ins; |
960 | struct extent_map *em; | |
961 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
962 | int ret = 0; | |
963 | ||
02ecd2c2 JB |
964 | if (btrfs_is_free_space_inode(inode)) { |
965 | WARN_ON_ONCE(1); | |
29bce2f3 JB |
966 | ret = -EINVAL; |
967 | goto out_unlock; | |
02ecd2c2 | 968 | } |
771ed689 | 969 | |
fda2832f | 970 | num_bytes = ALIGN(end - start + 1, blocksize); |
771ed689 CM |
971 | num_bytes = max(blocksize, num_bytes); |
972 | disk_num_bytes = num_bytes; | |
771ed689 | 973 | |
26d30f85 | 974 | inode_should_defrag(inode, start, end, num_bytes, SZ_64K); |
4cb5300b | 975 | |
771ed689 CM |
976 | if (start == 0) { |
977 | /* lets try to make an inline extent */ | |
f74670f7 AJ |
978 | ret = cow_file_range_inline(root, inode, start, end, 0, |
979 | BTRFS_COMPRESS_NONE, NULL); | |
771ed689 | 980 | if (ret == 0) { |
ba8b04c1 QW |
981 | extent_clear_unlock_delalloc(inode, start, end, |
982 | delalloc_end, NULL, | |
c2790a2e | 983 | EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc | 984 | EXTENT_DEFRAG, PAGE_UNLOCK | |
c2790a2e JB |
985 | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
986 | PAGE_END_WRITEBACK); | |
18513091 WX |
987 | btrfs_free_reserved_data_space_noquota(inode, start, |
988 | end - start + 1); | |
771ed689 | 989 | *nr_written = *nr_written + |
09cbfeaf | 990 | (end - start + PAGE_SIZE) / PAGE_SIZE; |
771ed689 | 991 | *page_started = 1; |
771ed689 | 992 | goto out; |
79787eaa | 993 | } else if (ret < 0) { |
79787eaa | 994 | goto out_unlock; |
771ed689 CM |
995 | } |
996 | } | |
997 | ||
998 | BUG_ON(disk_num_bytes > | |
0b246afa | 999 | btrfs_super_total_bytes(fs_info->super_copy)); |
771ed689 | 1000 | |
4b46fce2 | 1001 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
771ed689 CM |
1002 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
1003 | ||
d397712b | 1004 | while (disk_num_bytes > 0) { |
a791e35e CM |
1005 | unsigned long op; |
1006 | ||
287a0ab9 | 1007 | cur_alloc_size = disk_num_bytes; |
18513091 | 1008 | ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size, |
0b246afa | 1009 | fs_info->sectorsize, 0, alloc_hint, |
e570fd27 | 1010 | &ins, 1, 1); |
00361589 | 1011 | if (ret < 0) |
79787eaa | 1012 | goto out_unlock; |
d397712b | 1013 | |
172ddd60 | 1014 | em = alloc_extent_map(); |
b9aa55be LB |
1015 | if (!em) { |
1016 | ret = -ENOMEM; | |
ace68bac | 1017 | goto out_reserve; |
b9aa55be | 1018 | } |
e6dcd2dc | 1019 | em->start = start; |
445a6944 | 1020 | em->orig_start = em->start; |
771ed689 CM |
1021 | ram_size = ins.offset; |
1022 | em->len = ins.offset; | |
2ab28f32 JB |
1023 | em->mod_start = em->start; |
1024 | em->mod_len = em->len; | |
c8b97818 | 1025 | |
e6dcd2dc | 1026 | em->block_start = ins.objectid; |
c8b97818 | 1027 | em->block_len = ins.offset; |
b4939680 | 1028 | em->orig_block_len = ins.offset; |
cc95bef6 | 1029 | em->ram_bytes = ram_size; |
0b246afa | 1030 | em->bdev = fs_info->fs_devices->latest_bdev; |
7f3c74fb | 1031 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
70c8a91c | 1032 | em->generation = -1; |
c8b97818 | 1033 | |
d397712b | 1034 | while (1) { |
890871be | 1035 | write_lock(&em_tree->lock); |
09a2a8f9 | 1036 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 1037 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
1038 | if (ret != -EEXIST) { |
1039 | free_extent_map(em); | |
1040 | break; | |
1041 | } | |
1042 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 1043 | start + ram_size - 1, 0); |
e6dcd2dc | 1044 | } |
ace68bac LB |
1045 | if (ret) |
1046 | goto out_reserve; | |
e6dcd2dc | 1047 | |
98d20f67 | 1048 | cur_alloc_size = ins.offset; |
e6dcd2dc | 1049 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 1050 | ram_size, cur_alloc_size, 0); |
ace68bac | 1051 | if (ret) |
d9f85963 | 1052 | goto out_drop_extent_cache; |
c8b97818 | 1053 | |
17d217fe YZ |
1054 | if (root->root_key.objectid == |
1055 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1056 | ret = btrfs_reloc_clone_csums(inode, start, | |
1057 | cur_alloc_size); | |
00361589 | 1058 | if (ret) |
d9f85963 | 1059 | goto out_drop_extent_cache; |
17d217fe YZ |
1060 | } |
1061 | ||
0b246afa | 1062 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
9cfa3e34 | 1063 | |
d397712b | 1064 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 1065 | break; |
d397712b | 1066 | |
c8b97818 CM |
1067 | /* we're not doing compressed IO, don't unlock the first |
1068 | * page (which the caller expects to stay locked), don't | |
1069 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
1070 | * |
1071 | * Do set the Private2 bit so we know this page was properly | |
1072 | * setup for writepage | |
c8b97818 | 1073 | */ |
c2790a2e JB |
1074 | op = unlock ? PAGE_UNLOCK : 0; |
1075 | op |= PAGE_SET_PRIVATE2; | |
a791e35e | 1076 | |
c2790a2e | 1077 | extent_clear_unlock_delalloc(inode, start, |
ba8b04c1 QW |
1078 | start + ram_size - 1, |
1079 | delalloc_end, locked_page, | |
c2790a2e JB |
1080 | EXTENT_LOCKED | EXTENT_DELALLOC, |
1081 | op); | |
c8b97818 | 1082 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
1083 | num_bytes -= cur_alloc_size; |
1084 | alloc_hint = ins.objectid + ins.offset; | |
1085 | start += cur_alloc_size; | |
b888db2b | 1086 | } |
79787eaa | 1087 | out: |
be20aa9d | 1088 | return ret; |
b7d5b0a8 | 1089 | |
d9f85963 FM |
1090 | out_drop_extent_cache: |
1091 | btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0); | |
ace68bac | 1092 | out_reserve: |
0b246afa | 1093 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
2ff7e61e | 1094 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1); |
79787eaa | 1095 | out_unlock: |
ba8b04c1 QW |
1096 | extent_clear_unlock_delalloc(inode, start, end, delalloc_end, |
1097 | locked_page, | |
151a41bc JB |
1098 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
1099 | EXTENT_DELALLOC | EXTENT_DEFRAG, | |
1100 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
1101 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK); | |
79787eaa | 1102 | goto out; |
771ed689 | 1103 | } |
c8b97818 | 1104 | |
771ed689 CM |
1105 | /* |
1106 | * work queue call back to started compression on a file and pages | |
1107 | */ | |
1108 | static noinline void async_cow_start(struct btrfs_work *work) | |
1109 | { | |
1110 | struct async_cow *async_cow; | |
1111 | int num_added = 0; | |
1112 | async_cow = container_of(work, struct async_cow, work); | |
1113 | ||
1114 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
1115 | async_cow->start, async_cow->end, async_cow, | |
1116 | &num_added); | |
8180ef88 | 1117 | if (num_added == 0) { |
cb77fcd8 | 1118 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 | 1119 | async_cow->inode = NULL; |
8180ef88 | 1120 | } |
771ed689 CM |
1121 | } |
1122 | ||
1123 | /* | |
1124 | * work queue call back to submit previously compressed pages | |
1125 | */ | |
1126 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1127 | { | |
0b246afa | 1128 | struct btrfs_fs_info *fs_info; |
771ed689 CM |
1129 | struct async_cow *async_cow; |
1130 | struct btrfs_root *root; | |
1131 | unsigned long nr_pages; | |
1132 | ||
1133 | async_cow = container_of(work, struct async_cow, work); | |
1134 | ||
1135 | root = async_cow->root; | |
0b246afa | 1136 | fs_info = root->fs_info; |
09cbfeaf KS |
1137 | nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >> |
1138 | PAGE_SHIFT; | |
771ed689 | 1139 | |
ee863954 DS |
1140 | /* |
1141 | * atomic_sub_return implies a barrier for waitqueue_active | |
1142 | */ | |
0b246afa | 1143 | if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) < |
ee22184b | 1144 | 5 * SZ_1M && |
0b246afa JM |
1145 | waitqueue_active(&fs_info->async_submit_wait)) |
1146 | wake_up(&fs_info->async_submit_wait); | |
771ed689 | 1147 | |
d397712b | 1148 | if (async_cow->inode) |
771ed689 | 1149 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 1150 | } |
c8b97818 | 1151 | |
771ed689 CM |
1152 | static noinline void async_cow_free(struct btrfs_work *work) |
1153 | { | |
1154 | struct async_cow *async_cow; | |
1155 | async_cow = container_of(work, struct async_cow, work); | |
8180ef88 | 1156 | if (async_cow->inode) |
cb77fcd8 | 1157 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 CM |
1158 | kfree(async_cow); |
1159 | } | |
1160 | ||
1161 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
1162 | u64 start, u64 end, int *page_started, | |
1163 | unsigned long *nr_written) | |
1164 | { | |
0b246afa | 1165 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
771ed689 CM |
1166 | struct async_cow *async_cow; |
1167 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1168 | unsigned long nr_pages; | |
1169 | u64 cur_end; | |
771ed689 | 1170 | |
a3429ab7 CM |
1171 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
1172 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 1173 | while (start < end) { |
771ed689 | 1174 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
79787eaa | 1175 | BUG_ON(!async_cow); /* -ENOMEM */ |
8180ef88 | 1176 | async_cow->inode = igrab(inode); |
771ed689 CM |
1177 | async_cow->root = root; |
1178 | async_cow->locked_page = locked_page; | |
1179 | async_cow->start = start; | |
1180 | ||
f79707b0 | 1181 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS && |
0b246afa | 1182 | !btrfs_test_opt(fs_info, FORCE_COMPRESS)) |
771ed689 CM |
1183 | cur_end = end; |
1184 | else | |
ee22184b | 1185 | cur_end = min(end, start + SZ_512K - 1); |
771ed689 CM |
1186 | |
1187 | async_cow->end = cur_end; | |
1188 | INIT_LIST_HEAD(&async_cow->extents); | |
1189 | ||
9e0af237 LB |
1190 | btrfs_init_work(&async_cow->work, |
1191 | btrfs_delalloc_helper, | |
1192 | async_cow_start, async_cow_submit, | |
1193 | async_cow_free); | |
771ed689 | 1194 | |
09cbfeaf KS |
1195 | nr_pages = (cur_end - start + PAGE_SIZE) >> |
1196 | PAGE_SHIFT; | |
0b246afa | 1197 | atomic_add(nr_pages, &fs_info->async_delalloc_pages); |
771ed689 | 1198 | |
0b246afa | 1199 | btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work); |
771ed689 | 1200 | |
0b246afa JM |
1201 | while (atomic_read(&fs_info->async_submit_draining) && |
1202 | atomic_read(&fs_info->async_delalloc_pages)) { | |
1203 | wait_event(fs_info->async_submit_wait, | |
1204 | (atomic_read(&fs_info->async_delalloc_pages) == | |
1205 | 0)); | |
771ed689 CM |
1206 | } |
1207 | ||
1208 | *nr_written += nr_pages; | |
1209 | start = cur_end + 1; | |
1210 | } | |
1211 | *page_started = 1; | |
1212 | return 0; | |
be20aa9d CM |
1213 | } |
1214 | ||
2ff7e61e | 1215 | static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info, |
17d217fe YZ |
1216 | u64 bytenr, u64 num_bytes) |
1217 | { | |
1218 | int ret; | |
1219 | struct btrfs_ordered_sum *sums; | |
1220 | LIST_HEAD(list); | |
1221 | ||
0b246afa | 1222 | ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr, |
a2de733c | 1223 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1224 | if (ret == 0 && list_empty(&list)) |
1225 | return 0; | |
1226 | ||
1227 | while (!list_empty(&list)) { | |
1228 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1229 | list_del(&sums->list); | |
1230 | kfree(sums); | |
1231 | } | |
1232 | return 1; | |
1233 | } | |
1234 | ||
d352ac68 CM |
1235 | /* |
1236 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1237 | * of the extents that exist in the file, and COWs the file as required. | |
1238 | * | |
1239 | * If no cow copies or snapshots exist, we write directly to the existing | |
1240 | * blocks on disk | |
1241 | */ | |
7f366cfe CM |
1242 | static noinline int run_delalloc_nocow(struct inode *inode, |
1243 | struct page *locked_page, | |
771ed689 CM |
1244 | u64 start, u64 end, int *page_started, int force, |
1245 | unsigned long *nr_written) | |
be20aa9d | 1246 | { |
0b246afa | 1247 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
be20aa9d | 1248 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 1249 | struct btrfs_trans_handle *trans; |
be20aa9d | 1250 | struct extent_buffer *leaf; |
be20aa9d | 1251 | struct btrfs_path *path; |
80ff3856 | 1252 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1253 | struct btrfs_key found_key; |
80ff3856 YZ |
1254 | u64 cow_start; |
1255 | u64 cur_offset; | |
1256 | u64 extent_end; | |
5d4f98a2 | 1257 | u64 extent_offset; |
80ff3856 YZ |
1258 | u64 disk_bytenr; |
1259 | u64 num_bytes; | |
b4939680 | 1260 | u64 disk_num_bytes; |
cc95bef6 | 1261 | u64 ram_bytes; |
80ff3856 | 1262 | int extent_type; |
79787eaa | 1263 | int ret, err; |
d899e052 | 1264 | int type; |
80ff3856 YZ |
1265 | int nocow; |
1266 | int check_prev = 1; | |
82d5902d | 1267 | bool nolock; |
4a0cc7ca | 1268 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
be20aa9d CM |
1269 | |
1270 | path = btrfs_alloc_path(); | |
17ca04af | 1271 | if (!path) { |
ba8b04c1 QW |
1272 | extent_clear_unlock_delalloc(inode, start, end, end, |
1273 | locked_page, | |
c2790a2e | 1274 | EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc JB |
1275 | EXTENT_DO_ACCOUNTING | |
1276 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1277 | PAGE_CLEAR_DIRTY | |
1278 | PAGE_SET_WRITEBACK | | |
1279 | PAGE_END_WRITEBACK); | |
d8926bb3 | 1280 | return -ENOMEM; |
17ca04af | 1281 | } |
82d5902d | 1282 | |
83eea1f1 | 1283 | nolock = btrfs_is_free_space_inode(inode); |
82d5902d LZ |
1284 | |
1285 | if (nolock) | |
7a7eaa40 | 1286 | trans = btrfs_join_transaction_nolock(root); |
82d5902d | 1287 | else |
7a7eaa40 | 1288 | trans = btrfs_join_transaction(root); |
ff5714cc | 1289 | |
79787eaa | 1290 | if (IS_ERR(trans)) { |
ba8b04c1 QW |
1291 | extent_clear_unlock_delalloc(inode, start, end, end, |
1292 | locked_page, | |
c2790a2e | 1293 | EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc JB |
1294 | EXTENT_DO_ACCOUNTING | |
1295 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1296 | PAGE_CLEAR_DIRTY | |
1297 | PAGE_SET_WRITEBACK | | |
1298 | PAGE_END_WRITEBACK); | |
79787eaa JM |
1299 | btrfs_free_path(path); |
1300 | return PTR_ERR(trans); | |
1301 | } | |
1302 | ||
0b246afa | 1303 | trans->block_rsv = &fs_info->delalloc_block_rsv; |
be20aa9d | 1304 | |
80ff3856 YZ |
1305 | cow_start = (u64)-1; |
1306 | cur_offset = start; | |
1307 | while (1) { | |
33345d01 | 1308 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
80ff3856 | 1309 | cur_offset, 0); |
d788a349 | 1310 | if (ret < 0) |
79787eaa | 1311 | goto error; |
80ff3856 YZ |
1312 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1313 | leaf = path->nodes[0]; | |
1314 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1315 | path->slots[0] - 1); | |
33345d01 | 1316 | if (found_key.objectid == ino && |
80ff3856 YZ |
1317 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1318 | path->slots[0]--; | |
1319 | } | |
1320 | check_prev = 0; | |
1321 | next_slot: | |
1322 | leaf = path->nodes[0]; | |
1323 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1324 | ret = btrfs_next_leaf(root, path); | |
d788a349 | 1325 | if (ret < 0) |
79787eaa | 1326 | goto error; |
80ff3856 YZ |
1327 | if (ret > 0) |
1328 | break; | |
1329 | leaf = path->nodes[0]; | |
1330 | } | |
be20aa9d | 1331 | |
80ff3856 YZ |
1332 | nocow = 0; |
1333 | disk_bytenr = 0; | |
17d217fe | 1334 | num_bytes = 0; |
80ff3856 YZ |
1335 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1336 | ||
1d512cb7 FM |
1337 | if (found_key.objectid > ino) |
1338 | break; | |
1339 | if (WARN_ON_ONCE(found_key.objectid < ino) || | |
1340 | found_key.type < BTRFS_EXTENT_DATA_KEY) { | |
1341 | path->slots[0]++; | |
1342 | goto next_slot; | |
1343 | } | |
1344 | if (found_key.type > BTRFS_EXTENT_DATA_KEY || | |
80ff3856 YZ |
1345 | found_key.offset > end) |
1346 | break; | |
1347 | ||
1348 | if (found_key.offset > cur_offset) { | |
1349 | extent_end = found_key.offset; | |
e9061e21 | 1350 | extent_type = 0; |
80ff3856 YZ |
1351 | goto out_check; |
1352 | } | |
1353 | ||
1354 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1355 | struct btrfs_file_extent_item); | |
1356 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1357 | ||
cc95bef6 | 1358 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
d899e052 YZ |
1359 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1360 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1361 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1362 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1363 | extent_end = found_key.offset + |
1364 | btrfs_file_extent_num_bytes(leaf, fi); | |
b4939680 JB |
1365 | disk_num_bytes = |
1366 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
80ff3856 YZ |
1367 | if (extent_end <= start) { |
1368 | path->slots[0]++; | |
1369 | goto next_slot; | |
1370 | } | |
17d217fe YZ |
1371 | if (disk_bytenr == 0) |
1372 | goto out_check; | |
80ff3856 YZ |
1373 | if (btrfs_file_extent_compression(leaf, fi) || |
1374 | btrfs_file_extent_encryption(leaf, fi) || | |
1375 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1376 | goto out_check; | |
d899e052 YZ |
1377 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1378 | goto out_check; | |
2ff7e61e | 1379 | if (btrfs_extent_readonly(fs_info, disk_bytenr)) |
80ff3856 | 1380 | goto out_check; |
33345d01 | 1381 | if (btrfs_cross_ref_exist(trans, root, ino, |
5d4f98a2 YZ |
1382 | found_key.offset - |
1383 | extent_offset, disk_bytenr)) | |
17d217fe | 1384 | goto out_check; |
5d4f98a2 | 1385 | disk_bytenr += extent_offset; |
17d217fe YZ |
1386 | disk_bytenr += cur_offset - found_key.offset; |
1387 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
e9894fd3 WS |
1388 | /* |
1389 | * if there are pending snapshots for this root, | |
1390 | * we fall into common COW way. | |
1391 | */ | |
1392 | if (!nolock) { | |
9ea24bbe | 1393 | err = btrfs_start_write_no_snapshoting(root); |
e9894fd3 WS |
1394 | if (!err) |
1395 | goto out_check; | |
1396 | } | |
17d217fe YZ |
1397 | /* |
1398 | * force cow if csum exists in the range. | |
1399 | * this ensure that csum for a given extent are | |
1400 | * either valid or do not exist. | |
1401 | */ | |
2ff7e61e JM |
1402 | if (csum_exist_in_range(fs_info, disk_bytenr, |
1403 | num_bytes)) | |
17d217fe | 1404 | goto out_check; |
0b246afa | 1405 | if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) |
f78c436c | 1406 | goto out_check; |
80ff3856 YZ |
1407 | nocow = 1; |
1408 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1409 | extent_end = found_key.offset + | |
514ac8ad CM |
1410 | btrfs_file_extent_inline_len(leaf, |
1411 | path->slots[0], fi); | |
da17066c | 1412 | extent_end = ALIGN(extent_end, |
0b246afa | 1413 | fs_info->sectorsize); |
80ff3856 YZ |
1414 | } else { |
1415 | BUG_ON(1); | |
1416 | } | |
1417 | out_check: | |
1418 | if (extent_end <= start) { | |
1419 | path->slots[0]++; | |
e9894fd3 | 1420 | if (!nolock && nocow) |
9ea24bbe | 1421 | btrfs_end_write_no_snapshoting(root); |
f78c436c | 1422 | if (nocow) |
0b246afa | 1423 | btrfs_dec_nocow_writers(fs_info, disk_bytenr); |
80ff3856 YZ |
1424 | goto next_slot; |
1425 | } | |
1426 | if (!nocow) { | |
1427 | if (cow_start == (u64)-1) | |
1428 | cow_start = cur_offset; | |
1429 | cur_offset = extent_end; | |
1430 | if (cur_offset > end) | |
1431 | break; | |
1432 | path->slots[0]++; | |
1433 | goto next_slot; | |
7ea394f1 YZ |
1434 | } |
1435 | ||
b3b4aa74 | 1436 | btrfs_release_path(path); |
80ff3856 | 1437 | if (cow_start != (u64)-1) { |
00361589 JB |
1438 | ret = cow_file_range(inode, locked_page, |
1439 | cow_start, found_key.offset - 1, | |
dda3245e WX |
1440 | end, page_started, nr_written, 1, |
1441 | NULL); | |
e9894fd3 WS |
1442 | if (ret) { |
1443 | if (!nolock && nocow) | |
9ea24bbe | 1444 | btrfs_end_write_no_snapshoting(root); |
f78c436c | 1445 | if (nocow) |
0b246afa | 1446 | btrfs_dec_nocow_writers(fs_info, |
f78c436c | 1447 | disk_bytenr); |
79787eaa | 1448 | goto error; |
e9894fd3 | 1449 | } |
80ff3856 | 1450 | cow_start = (u64)-1; |
7ea394f1 | 1451 | } |
80ff3856 | 1452 | |
d899e052 YZ |
1453 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1454 | struct extent_map *em; | |
1455 | struct extent_map_tree *em_tree; | |
1456 | em_tree = &BTRFS_I(inode)->extent_tree; | |
172ddd60 | 1457 | em = alloc_extent_map(); |
79787eaa | 1458 | BUG_ON(!em); /* -ENOMEM */ |
d899e052 | 1459 | em->start = cur_offset; |
70c8a91c | 1460 | em->orig_start = found_key.offset - extent_offset; |
d899e052 YZ |
1461 | em->len = num_bytes; |
1462 | em->block_len = num_bytes; | |
1463 | em->block_start = disk_bytenr; | |
b4939680 | 1464 | em->orig_block_len = disk_num_bytes; |
cc95bef6 | 1465 | em->ram_bytes = ram_bytes; |
0b246afa | 1466 | em->bdev = fs_info->fs_devices->latest_bdev; |
2ab28f32 JB |
1467 | em->mod_start = em->start; |
1468 | em->mod_len = em->len; | |
d899e052 | 1469 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
b11e234d | 1470 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
70c8a91c | 1471 | em->generation = -1; |
d899e052 | 1472 | while (1) { |
890871be | 1473 | write_lock(&em_tree->lock); |
09a2a8f9 | 1474 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 1475 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1476 | if (ret != -EEXIST) { |
1477 | free_extent_map(em); | |
1478 | break; | |
1479 | } | |
1480 | btrfs_drop_extent_cache(inode, em->start, | |
1481 | em->start + em->len - 1, 0); | |
1482 | } | |
1483 | type = BTRFS_ORDERED_PREALLOC; | |
1484 | } else { | |
1485 | type = BTRFS_ORDERED_NOCOW; | |
1486 | } | |
80ff3856 YZ |
1487 | |
1488 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 | 1489 | num_bytes, num_bytes, type); |
f78c436c | 1490 | if (nocow) |
0b246afa | 1491 | btrfs_dec_nocow_writers(fs_info, disk_bytenr); |
79787eaa | 1492 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 1493 | |
efa56464 YZ |
1494 | if (root->root_key.objectid == |
1495 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1496 | ret = btrfs_reloc_clone_csums(inode, cur_offset, | |
1497 | num_bytes); | |
e9894fd3 WS |
1498 | if (ret) { |
1499 | if (!nolock && nocow) | |
9ea24bbe | 1500 | btrfs_end_write_no_snapshoting(root); |
79787eaa | 1501 | goto error; |
e9894fd3 | 1502 | } |
efa56464 YZ |
1503 | } |
1504 | ||
c2790a2e | 1505 | extent_clear_unlock_delalloc(inode, cur_offset, |
ba8b04c1 | 1506 | cur_offset + num_bytes - 1, end, |
c2790a2e | 1507 | locked_page, EXTENT_LOCKED | |
18513091 WX |
1508 | EXTENT_DELALLOC | |
1509 | EXTENT_CLEAR_DATA_RESV, | |
1510 | PAGE_UNLOCK | PAGE_SET_PRIVATE2); | |
1511 | ||
e9894fd3 | 1512 | if (!nolock && nocow) |
9ea24bbe | 1513 | btrfs_end_write_no_snapshoting(root); |
80ff3856 YZ |
1514 | cur_offset = extent_end; |
1515 | if (cur_offset > end) | |
1516 | break; | |
be20aa9d | 1517 | } |
b3b4aa74 | 1518 | btrfs_release_path(path); |
80ff3856 | 1519 | |
17ca04af | 1520 | if (cur_offset <= end && cow_start == (u64)-1) { |
80ff3856 | 1521 | cow_start = cur_offset; |
17ca04af JB |
1522 | cur_offset = end; |
1523 | } | |
1524 | ||
80ff3856 | 1525 | if (cow_start != (u64)-1) { |
dda3245e WX |
1526 | ret = cow_file_range(inode, locked_page, cow_start, end, end, |
1527 | page_started, nr_written, 1, NULL); | |
d788a349 | 1528 | if (ret) |
79787eaa | 1529 | goto error; |
80ff3856 YZ |
1530 | } |
1531 | ||
79787eaa | 1532 | error: |
3a45bb20 | 1533 | err = btrfs_end_transaction(trans); |
79787eaa JM |
1534 | if (!ret) |
1535 | ret = err; | |
1536 | ||
17ca04af | 1537 | if (ret && cur_offset < end) |
ba8b04c1 | 1538 | extent_clear_unlock_delalloc(inode, cur_offset, end, end, |
c2790a2e | 1539 | locked_page, EXTENT_LOCKED | |
151a41bc JB |
1540 | EXTENT_DELALLOC | EXTENT_DEFRAG | |
1541 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
1542 | PAGE_CLEAR_DIRTY | | |
c2790a2e JB |
1543 | PAGE_SET_WRITEBACK | |
1544 | PAGE_END_WRITEBACK); | |
7ea394f1 | 1545 | btrfs_free_path(path); |
79787eaa | 1546 | return ret; |
be20aa9d CM |
1547 | } |
1548 | ||
47059d93 WS |
1549 | static inline int need_force_cow(struct inode *inode, u64 start, u64 end) |
1550 | { | |
1551 | ||
1552 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
1553 | !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)) | |
1554 | return 0; | |
1555 | ||
1556 | /* | |
1557 | * @defrag_bytes is a hint value, no spinlock held here, | |
1558 | * if is not zero, it means the file is defragging. | |
1559 | * Force cow if given extent needs to be defragged. | |
1560 | */ | |
1561 | if (BTRFS_I(inode)->defrag_bytes && | |
1562 | test_range_bit(&BTRFS_I(inode)->io_tree, start, end, | |
1563 | EXTENT_DEFRAG, 0, NULL)) | |
1564 | return 1; | |
1565 | ||
1566 | return 0; | |
1567 | } | |
1568 | ||
d352ac68 CM |
1569 | /* |
1570 | * extent_io.c call back to do delayed allocation processing | |
1571 | */ | |
c8b97818 | 1572 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1573 | u64 start, u64 end, int *page_started, |
1574 | unsigned long *nr_written) | |
be20aa9d | 1575 | { |
be20aa9d | 1576 | int ret; |
47059d93 | 1577 | int force_cow = need_force_cow(inode, start, end); |
a2135011 | 1578 | |
47059d93 | 1579 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) { |
c8b97818 | 1580 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1581 | page_started, 1, nr_written); |
47059d93 | 1582 | } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) { |
d899e052 | 1583 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1584 | page_started, 0, nr_written); |
7816030e | 1585 | } else if (!inode_need_compress(inode)) { |
dda3245e WX |
1586 | ret = cow_file_range(inode, locked_page, start, end, end, |
1587 | page_started, nr_written, 1, NULL); | |
7ddf5a42 JB |
1588 | } else { |
1589 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
1590 | &BTRFS_I(inode)->runtime_flags); | |
771ed689 | 1591 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1592 | page_started, nr_written); |
7ddf5a42 | 1593 | } |
b888db2b CM |
1594 | return ret; |
1595 | } | |
1596 | ||
1bf85046 JM |
1597 | static void btrfs_split_extent_hook(struct inode *inode, |
1598 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1599 | { |
dcab6a3b JB |
1600 | u64 size; |
1601 | ||
0ca1f7ce | 1602 | /* not delalloc, ignore it */ |
9ed74f2d | 1603 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1604 | return; |
9ed74f2d | 1605 | |
dcab6a3b JB |
1606 | size = orig->end - orig->start + 1; |
1607 | if (size > BTRFS_MAX_EXTENT_SIZE) { | |
823bb20a | 1608 | u32 num_extents; |
dcab6a3b JB |
1609 | u64 new_size; |
1610 | ||
1611 | /* | |
ba117213 JB |
1612 | * See the explanation in btrfs_merge_extent_hook, the same |
1613 | * applies here, just in reverse. | |
dcab6a3b JB |
1614 | */ |
1615 | new_size = orig->end - split + 1; | |
823bb20a | 1616 | num_extents = count_max_extents(new_size); |
ba117213 | 1617 | new_size = split - orig->start; |
823bb20a DS |
1618 | num_extents += count_max_extents(new_size); |
1619 | if (count_max_extents(size) >= num_extents) | |
dcab6a3b JB |
1620 | return; |
1621 | } | |
1622 | ||
9e0baf60 JB |
1623 | spin_lock(&BTRFS_I(inode)->lock); |
1624 | BTRFS_I(inode)->outstanding_extents++; | |
1625 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1626 | } |
1627 | ||
1628 | /* | |
1629 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1630 | * extents so we can keep track of new extents that are just merged onto old | |
1631 | * extents, such as when we are doing sequential writes, so we can properly | |
1632 | * account for the metadata space we'll need. | |
1633 | */ | |
1bf85046 JM |
1634 | static void btrfs_merge_extent_hook(struct inode *inode, |
1635 | struct extent_state *new, | |
1636 | struct extent_state *other) | |
9ed74f2d | 1637 | { |
dcab6a3b | 1638 | u64 new_size, old_size; |
823bb20a | 1639 | u32 num_extents; |
dcab6a3b | 1640 | |
9ed74f2d JB |
1641 | /* not delalloc, ignore it */ |
1642 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1643 | return; |
9ed74f2d | 1644 | |
8461a3de JB |
1645 | if (new->start > other->start) |
1646 | new_size = new->end - other->start + 1; | |
1647 | else | |
1648 | new_size = other->end - new->start + 1; | |
dcab6a3b JB |
1649 | |
1650 | /* we're not bigger than the max, unreserve the space and go */ | |
1651 | if (new_size <= BTRFS_MAX_EXTENT_SIZE) { | |
1652 | spin_lock(&BTRFS_I(inode)->lock); | |
1653 | BTRFS_I(inode)->outstanding_extents--; | |
1654 | spin_unlock(&BTRFS_I(inode)->lock); | |
1655 | return; | |
1656 | } | |
1657 | ||
1658 | /* | |
ba117213 JB |
1659 | * We have to add up either side to figure out how many extents were |
1660 | * accounted for before we merged into one big extent. If the number of | |
1661 | * extents we accounted for is <= the amount we need for the new range | |
1662 | * then we can return, otherwise drop. Think of it like this | |
1663 | * | |
1664 | * [ 4k][MAX_SIZE] | |
1665 | * | |
1666 | * So we've grown the extent by a MAX_SIZE extent, this would mean we | |
1667 | * need 2 outstanding extents, on one side we have 1 and the other side | |
1668 | * we have 1 so they are == and we can return. But in this case | |
1669 | * | |
1670 | * [MAX_SIZE+4k][MAX_SIZE+4k] | |
1671 | * | |
1672 | * Each range on their own accounts for 2 extents, but merged together | |
1673 | * they are only 3 extents worth of accounting, so we need to drop in | |
1674 | * this case. | |
dcab6a3b | 1675 | */ |
ba117213 | 1676 | old_size = other->end - other->start + 1; |
823bb20a | 1677 | num_extents = count_max_extents(old_size); |
ba117213 | 1678 | old_size = new->end - new->start + 1; |
823bb20a DS |
1679 | num_extents += count_max_extents(old_size); |
1680 | if (count_max_extents(new_size) >= num_extents) | |
dcab6a3b JB |
1681 | return; |
1682 | ||
9e0baf60 JB |
1683 | spin_lock(&BTRFS_I(inode)->lock); |
1684 | BTRFS_I(inode)->outstanding_extents--; | |
1685 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1686 | } |
1687 | ||
eb73c1b7 MX |
1688 | static void btrfs_add_delalloc_inodes(struct btrfs_root *root, |
1689 | struct inode *inode) | |
1690 | { | |
0b246afa JM |
1691 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
1692 | ||
eb73c1b7 MX |
1693 | spin_lock(&root->delalloc_lock); |
1694 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1695 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1696 | &root->delalloc_inodes); | |
1697 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1698 | &BTRFS_I(inode)->runtime_flags); | |
1699 | root->nr_delalloc_inodes++; | |
1700 | if (root->nr_delalloc_inodes == 1) { | |
0b246afa | 1701 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
1702 | BUG_ON(!list_empty(&root->delalloc_root)); |
1703 | list_add_tail(&root->delalloc_root, | |
0b246afa JM |
1704 | &fs_info->delalloc_roots); |
1705 | spin_unlock(&fs_info->delalloc_root_lock); | |
eb73c1b7 MX |
1706 | } |
1707 | } | |
1708 | spin_unlock(&root->delalloc_lock); | |
1709 | } | |
1710 | ||
1711 | static void btrfs_del_delalloc_inode(struct btrfs_root *root, | |
1712 | struct inode *inode) | |
1713 | { | |
0b246afa JM |
1714 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
1715 | ||
eb73c1b7 MX |
1716 | spin_lock(&root->delalloc_lock); |
1717 | if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1718 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1719 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1720 | &BTRFS_I(inode)->runtime_flags); | |
1721 | root->nr_delalloc_inodes--; | |
1722 | if (!root->nr_delalloc_inodes) { | |
0b246afa | 1723 | spin_lock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
1724 | BUG_ON(list_empty(&root->delalloc_root)); |
1725 | list_del_init(&root->delalloc_root); | |
0b246afa | 1726 | spin_unlock(&fs_info->delalloc_root_lock); |
eb73c1b7 MX |
1727 | } |
1728 | } | |
1729 | spin_unlock(&root->delalloc_lock); | |
1730 | } | |
1731 | ||
d352ac68 CM |
1732 | /* |
1733 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1734 | * bytes in this file, and to maintain the list of inodes that | |
1735 | * have pending delalloc work to be done. | |
1736 | */ | |
1bf85046 | 1737 | static void btrfs_set_bit_hook(struct inode *inode, |
9ee49a04 | 1738 | struct extent_state *state, unsigned *bits) |
291d673e | 1739 | { |
9ed74f2d | 1740 | |
0b246afa JM |
1741 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
1742 | ||
47059d93 WS |
1743 | if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC)) |
1744 | WARN_ON(1); | |
75eff68e CM |
1745 | /* |
1746 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1747 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1748 | * bit, which is only set or cleared with irqs on |
1749 | */ | |
0ca1f7ce | 1750 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1751 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1752 | u64 len = state->end + 1 - state->start; |
83eea1f1 | 1753 | bool do_list = !btrfs_is_free_space_inode(inode); |
9ed74f2d | 1754 | |
9e0baf60 | 1755 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1756 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1757 | } else { |
1758 | spin_lock(&BTRFS_I(inode)->lock); | |
1759 | BTRFS_I(inode)->outstanding_extents++; | |
1760 | spin_unlock(&BTRFS_I(inode)->lock); | |
1761 | } | |
287a0ab9 | 1762 | |
6a3891c5 | 1763 | /* For sanity tests */ |
0b246afa | 1764 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
1765 | return; |
1766 | ||
0b246afa JM |
1767 | __percpu_counter_add(&fs_info->delalloc_bytes, len, |
1768 | fs_info->delalloc_batch); | |
df0af1a5 | 1769 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1770 | BTRFS_I(inode)->delalloc_bytes += len; |
47059d93 WS |
1771 | if (*bits & EXTENT_DEFRAG) |
1772 | BTRFS_I(inode)->defrag_bytes += len; | |
df0af1a5 | 1773 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1774 | &BTRFS_I(inode)->runtime_flags)) |
1775 | btrfs_add_delalloc_inodes(root, inode); | |
df0af1a5 | 1776 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1777 | } |
291d673e CM |
1778 | } |
1779 | ||
d352ac68 CM |
1780 | /* |
1781 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1782 | */ | |
1bf85046 | 1783 | static void btrfs_clear_bit_hook(struct inode *inode, |
41074888 | 1784 | struct extent_state *state, |
9ee49a04 | 1785 | unsigned *bits) |
291d673e | 1786 | { |
0b246afa | 1787 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
47059d93 | 1788 | u64 len = state->end + 1 - state->start; |
823bb20a | 1789 | u32 num_extents = count_max_extents(len); |
47059d93 WS |
1790 | |
1791 | spin_lock(&BTRFS_I(inode)->lock); | |
1792 | if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) | |
1793 | BTRFS_I(inode)->defrag_bytes -= len; | |
1794 | spin_unlock(&BTRFS_I(inode)->lock); | |
1795 | ||
75eff68e CM |
1796 | /* |
1797 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1798 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1799 | * bit, which is only set or cleared with irqs on |
1800 | */ | |
0ca1f7ce | 1801 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1802 | struct btrfs_root *root = BTRFS_I(inode)->root; |
83eea1f1 | 1803 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 1804 | |
9e0baf60 | 1805 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1806 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1807 | } else if (!(*bits & EXTENT_DO_ACCOUNTING)) { |
1808 | spin_lock(&BTRFS_I(inode)->lock); | |
dcab6a3b | 1809 | BTRFS_I(inode)->outstanding_extents -= num_extents; |
9e0baf60 JB |
1810 | spin_unlock(&BTRFS_I(inode)->lock); |
1811 | } | |
0ca1f7ce | 1812 | |
b6d08f06 JB |
1813 | /* |
1814 | * We don't reserve metadata space for space cache inodes so we | |
1815 | * don't need to call dellalloc_release_metadata if there is an | |
1816 | * error. | |
1817 | */ | |
1818 | if (*bits & EXTENT_DO_ACCOUNTING && | |
0b246afa | 1819 | root != fs_info->tree_root) |
0ca1f7ce YZ |
1820 | btrfs_delalloc_release_metadata(inode, len); |
1821 | ||
6a3891c5 | 1822 | /* For sanity tests. */ |
0b246afa | 1823 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
1824 | return; |
1825 | ||
0cb59c99 | 1826 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
18513091 WX |
1827 | && do_list && !(state->state & EXTENT_NORESERVE) |
1828 | && (*bits & (EXTENT_DO_ACCOUNTING | | |
1829 | EXTENT_CLEAR_DATA_RESV))) | |
51773bec QW |
1830 | btrfs_free_reserved_data_space_noquota(inode, |
1831 | state->start, len); | |
9ed74f2d | 1832 | |
0b246afa JM |
1833 | __percpu_counter_add(&fs_info->delalloc_bytes, -len, |
1834 | fs_info->delalloc_batch); | |
df0af1a5 | 1835 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1836 | BTRFS_I(inode)->delalloc_bytes -= len; |
0cb59c99 | 1837 | if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 && |
df0af1a5 | 1838 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1839 | &BTRFS_I(inode)->runtime_flags)) |
1840 | btrfs_del_delalloc_inode(root, inode); | |
df0af1a5 | 1841 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1842 | } |
291d673e CM |
1843 | } |
1844 | ||
d352ac68 CM |
1845 | /* |
1846 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1847 | * we don't create bios that span stripes or chunks | |
6f034ece LB |
1848 | * |
1849 | * return 1 if page cannot be merged to bio | |
1850 | * return 0 if page can be merged to bio | |
1851 | * return error otherwise | |
d352ac68 | 1852 | */ |
81a75f67 | 1853 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1854 | size_t size, struct bio *bio, |
1855 | unsigned long bio_flags) | |
239b14b3 | 1856 | { |
0b246afa JM |
1857 | struct inode *inode = page->mapping->host; |
1858 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
4f024f37 | 1859 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
239b14b3 CM |
1860 | u64 length = 0; |
1861 | u64 map_length; | |
239b14b3 CM |
1862 | int ret; |
1863 | ||
771ed689 CM |
1864 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1865 | return 0; | |
1866 | ||
4f024f37 | 1867 | length = bio->bi_iter.bi_size; |
239b14b3 | 1868 | map_length = length; |
0b246afa JM |
1869 | ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length, |
1870 | NULL, 0); | |
6f034ece LB |
1871 | if (ret < 0) |
1872 | return ret; | |
d397712b | 1873 | if (map_length < length + size) |
239b14b3 | 1874 | return 1; |
3444a972 | 1875 | return 0; |
239b14b3 CM |
1876 | } |
1877 | ||
d352ac68 CM |
1878 | /* |
1879 | * in order to insert checksums into the metadata in large chunks, | |
1880 | * we wait until bio submission time. All the pages in the bio are | |
1881 | * checksummed and sums are attached onto the ordered extent record. | |
1882 | * | |
1883 | * At IO completion time the cums attached on the ordered extent record | |
1884 | * are inserted into the btree | |
1885 | */ | |
81a75f67 MC |
1886 | static int __btrfs_submit_bio_start(struct inode *inode, struct bio *bio, |
1887 | int mirror_num, unsigned long bio_flags, | |
eaf25d93 | 1888 | u64 bio_offset) |
065631f6 | 1889 | { |
065631f6 | 1890 | int ret = 0; |
e015640f | 1891 | |
2ff7e61e | 1892 | ret = btrfs_csum_one_bio(inode, bio, 0, 0); |
79787eaa | 1893 | BUG_ON(ret); /* -ENOMEM */ |
4a69a410 CM |
1894 | return 0; |
1895 | } | |
e015640f | 1896 | |
4a69a410 CM |
1897 | /* |
1898 | * in order to insert checksums into the metadata in large chunks, | |
1899 | * we wait until bio submission time. All the pages in the bio are | |
1900 | * checksummed and sums are attached onto the ordered extent record. | |
1901 | * | |
1902 | * At IO completion time the cums attached on the ordered extent record | |
1903 | * are inserted into the btree | |
1904 | */ | |
81a75f67 | 1905 | static int __btrfs_submit_bio_done(struct inode *inode, struct bio *bio, |
eaf25d93 CM |
1906 | int mirror_num, unsigned long bio_flags, |
1907 | u64 bio_offset) | |
4a69a410 | 1908 | { |
2ff7e61e | 1909 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
61891923 SB |
1910 | int ret; |
1911 | ||
2ff7e61e | 1912 | ret = btrfs_map_bio(fs_info, bio, mirror_num, 1); |
4246a0b6 CH |
1913 | if (ret) { |
1914 | bio->bi_error = ret; | |
1915 | bio_endio(bio); | |
1916 | } | |
61891923 | 1917 | return ret; |
44b8bd7e CM |
1918 | } |
1919 | ||
d352ac68 | 1920 | /* |
cad321ad CM |
1921 | * extent_io.c submission hook. This does the right thing for csum calculation |
1922 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1923 | */ |
81a75f67 | 1924 | static int btrfs_submit_bio_hook(struct inode *inode, struct bio *bio, |
eaf25d93 CM |
1925 | int mirror_num, unsigned long bio_flags, |
1926 | u64 bio_offset) | |
44b8bd7e | 1927 | { |
0b246afa | 1928 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
44b8bd7e | 1929 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0d51e28a | 1930 | enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA; |
44b8bd7e | 1931 | int ret = 0; |
19b9bdb0 | 1932 | int skip_sum; |
b812ce28 | 1933 | int async = !atomic_read(&BTRFS_I(inode)->sync_writers); |
44b8bd7e | 1934 | |
6cbff00f | 1935 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1936 | |
83eea1f1 | 1937 | if (btrfs_is_free_space_inode(inode)) |
0d51e28a | 1938 | metadata = BTRFS_WQ_ENDIO_FREE_SPACE; |
0417341e | 1939 | |
37226b21 | 1940 | if (bio_op(bio) != REQ_OP_WRITE) { |
0b246afa | 1941 | ret = btrfs_bio_wq_end_io(fs_info, bio, metadata); |
5fd02043 | 1942 | if (ret) |
61891923 | 1943 | goto out; |
5fd02043 | 1944 | |
d20f7043 | 1945 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
61891923 SB |
1946 | ret = btrfs_submit_compressed_read(inode, bio, |
1947 | mirror_num, | |
1948 | bio_flags); | |
1949 | goto out; | |
c2db1073 | 1950 | } else if (!skip_sum) { |
2ff7e61e | 1951 | ret = btrfs_lookup_bio_sums(inode, bio, NULL); |
c2db1073 | 1952 | if (ret) |
61891923 | 1953 | goto out; |
c2db1073 | 1954 | } |
4d1b5fb4 | 1955 | goto mapit; |
b812ce28 | 1956 | } else if (async && !skip_sum) { |
17d217fe YZ |
1957 | /* csum items have already been cloned */ |
1958 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1959 | goto mapit; | |
19b9bdb0 | 1960 | /* we're doing a write, do the async checksumming */ |
0b246afa JM |
1961 | ret = btrfs_wq_submit_bio(fs_info, inode, bio, mirror_num, |
1962 | bio_flags, bio_offset, | |
1963 | __btrfs_submit_bio_start, | |
1964 | __btrfs_submit_bio_done); | |
61891923 | 1965 | goto out; |
b812ce28 | 1966 | } else if (!skip_sum) { |
2ff7e61e | 1967 | ret = btrfs_csum_one_bio(inode, bio, 0, 0); |
b812ce28 JB |
1968 | if (ret) |
1969 | goto out; | |
19b9bdb0 CM |
1970 | } |
1971 | ||
0b86a832 | 1972 | mapit: |
2ff7e61e | 1973 | ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); |
61891923 SB |
1974 | |
1975 | out: | |
4246a0b6 CH |
1976 | if (ret < 0) { |
1977 | bio->bi_error = ret; | |
1978 | bio_endio(bio); | |
1979 | } | |
61891923 | 1980 | return ret; |
065631f6 | 1981 | } |
6885f308 | 1982 | |
d352ac68 CM |
1983 | /* |
1984 | * given a list of ordered sums record them in the inode. This happens | |
1985 | * at IO completion time based on sums calculated at bio submission time. | |
1986 | */ | |
ba1da2f4 | 1987 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1988 | struct inode *inode, u64 file_offset, |
1989 | struct list_head *list) | |
1990 | { | |
e6dcd2dc CM |
1991 | struct btrfs_ordered_sum *sum; |
1992 | ||
c6e30871 | 1993 | list_for_each_entry(sum, list, list) { |
39847c4d | 1994 | trans->adding_csums = 1; |
d20f7043 CM |
1995 | btrfs_csum_file_blocks(trans, |
1996 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
39847c4d | 1997 | trans->adding_csums = 0; |
e6dcd2dc CM |
1998 | } |
1999 | return 0; | |
2000 | } | |
2001 | ||
2ac55d41 | 2002 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
ba8b04c1 | 2003 | struct extent_state **cached_state, int dedupe) |
ea8c2819 | 2004 | { |
09cbfeaf | 2005 | WARN_ON((end & (PAGE_SIZE - 1)) == 0); |
ea8c2819 | 2006 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
7cd8c752 | 2007 | cached_state); |
ea8c2819 CM |
2008 | } |
2009 | ||
d352ac68 | 2010 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
2011 | struct btrfs_writepage_fixup { |
2012 | struct page *page; | |
2013 | struct btrfs_work work; | |
2014 | }; | |
2015 | ||
b2950863 | 2016 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
2017 | { |
2018 | struct btrfs_writepage_fixup *fixup; | |
2019 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 2020 | struct extent_state *cached_state = NULL; |
247e743c CM |
2021 | struct page *page; |
2022 | struct inode *inode; | |
2023 | u64 page_start; | |
2024 | u64 page_end; | |
87826df0 | 2025 | int ret; |
247e743c CM |
2026 | |
2027 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
2028 | page = fixup->page; | |
4a096752 | 2029 | again: |
247e743c CM |
2030 | lock_page(page); |
2031 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
2032 | ClearPageChecked(page); | |
2033 | goto out_page; | |
2034 | } | |
2035 | ||
2036 | inode = page->mapping->host; | |
2037 | page_start = page_offset(page); | |
09cbfeaf | 2038 | page_end = page_offset(page) + PAGE_SIZE - 1; |
247e743c | 2039 | |
ff13db41 | 2040 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, |
d0082371 | 2041 | &cached_state); |
4a096752 CM |
2042 | |
2043 | /* already ordered? We're done */ | |
8b62b72b | 2044 | if (PagePrivate2(page)) |
247e743c | 2045 | goto out; |
4a096752 | 2046 | |
dbfdb6d1 | 2047 | ordered = btrfs_lookup_ordered_range(inode, page_start, |
09cbfeaf | 2048 | PAGE_SIZE); |
4a096752 | 2049 | if (ordered) { |
2ac55d41 JB |
2050 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
2051 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
2052 | unlock_page(page); |
2053 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 2054 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
2055 | goto again; |
2056 | } | |
247e743c | 2057 | |
7cf5b976 | 2058 | ret = btrfs_delalloc_reserve_space(inode, page_start, |
09cbfeaf | 2059 | PAGE_SIZE); |
87826df0 JM |
2060 | if (ret) { |
2061 | mapping_set_error(page->mapping, ret); | |
2062 | end_extent_writepage(page, ret, page_start, page_end); | |
2063 | ClearPageChecked(page); | |
2064 | goto out; | |
2065 | } | |
2066 | ||
ba8b04c1 QW |
2067 | btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state, |
2068 | 0); | |
247e743c | 2069 | ClearPageChecked(page); |
87826df0 | 2070 | set_page_dirty(page); |
247e743c | 2071 | out: |
2ac55d41 JB |
2072 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
2073 | &cached_state, GFP_NOFS); | |
247e743c CM |
2074 | out_page: |
2075 | unlock_page(page); | |
09cbfeaf | 2076 | put_page(page); |
b897abec | 2077 | kfree(fixup); |
247e743c CM |
2078 | } |
2079 | ||
2080 | /* | |
2081 | * There are a few paths in the higher layers of the kernel that directly | |
2082 | * set the page dirty bit without asking the filesystem if it is a | |
2083 | * good idea. This causes problems because we want to make sure COW | |
2084 | * properly happens and the data=ordered rules are followed. | |
2085 | * | |
c8b97818 | 2086 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
2087 | * hasn't been properly setup for IO. We kick off an async process |
2088 | * to fix it up. The async helper will wait for ordered extents, set | |
2089 | * the delalloc bit and make it safe to write the page. | |
2090 | */ | |
b2950863 | 2091 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
2092 | { |
2093 | struct inode *inode = page->mapping->host; | |
0b246afa | 2094 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
247e743c | 2095 | struct btrfs_writepage_fixup *fixup; |
247e743c | 2096 | |
8b62b72b CM |
2097 | /* this page is properly in the ordered list */ |
2098 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
2099 | return 0; |
2100 | ||
2101 | if (PageChecked(page)) | |
2102 | return -EAGAIN; | |
2103 | ||
2104 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
2105 | if (!fixup) | |
2106 | return -EAGAIN; | |
f421950f | 2107 | |
247e743c | 2108 | SetPageChecked(page); |
09cbfeaf | 2109 | get_page(page); |
9e0af237 LB |
2110 | btrfs_init_work(&fixup->work, btrfs_fixup_helper, |
2111 | btrfs_writepage_fixup_worker, NULL, NULL); | |
247e743c | 2112 | fixup->page = page; |
0b246afa | 2113 | btrfs_queue_work(fs_info->fixup_workers, &fixup->work); |
87826df0 | 2114 | return -EBUSY; |
247e743c CM |
2115 | } |
2116 | ||
d899e052 YZ |
2117 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
2118 | struct inode *inode, u64 file_pos, | |
2119 | u64 disk_bytenr, u64 disk_num_bytes, | |
2120 | u64 num_bytes, u64 ram_bytes, | |
2121 | u8 compression, u8 encryption, | |
2122 | u16 other_encoding, int extent_type) | |
2123 | { | |
2124 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2125 | struct btrfs_file_extent_item *fi; | |
2126 | struct btrfs_path *path; | |
2127 | struct extent_buffer *leaf; | |
2128 | struct btrfs_key ins; | |
1acae57b | 2129 | int extent_inserted = 0; |
d899e052 YZ |
2130 | int ret; |
2131 | ||
2132 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
2133 | if (!path) |
2134 | return -ENOMEM; | |
d899e052 | 2135 | |
a1ed835e CM |
2136 | /* |
2137 | * we may be replacing one extent in the tree with another. | |
2138 | * The new extent is pinned in the extent map, and we don't want | |
2139 | * to drop it from the cache until it is completely in the btree. | |
2140 | * | |
2141 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
2142 | * the caller is expected to unpin it and allow it to be merged | |
2143 | * with the others. | |
2144 | */ | |
1acae57b FDBM |
2145 | ret = __btrfs_drop_extents(trans, root, inode, path, file_pos, |
2146 | file_pos + num_bytes, NULL, 0, | |
2147 | 1, sizeof(*fi), &extent_inserted); | |
79787eaa JM |
2148 | if (ret) |
2149 | goto out; | |
d899e052 | 2150 | |
1acae57b | 2151 | if (!extent_inserted) { |
4a0cc7ca | 2152 | ins.objectid = btrfs_ino(BTRFS_I(inode)); |
1acae57b FDBM |
2153 | ins.offset = file_pos; |
2154 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
2155 | ||
2156 | path->leave_spinning = 1; | |
2157 | ret = btrfs_insert_empty_item(trans, root, path, &ins, | |
2158 | sizeof(*fi)); | |
2159 | if (ret) | |
2160 | goto out; | |
2161 | } | |
d899e052 YZ |
2162 | leaf = path->nodes[0]; |
2163 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2164 | struct btrfs_file_extent_item); | |
2165 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
2166 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
2167 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
2168 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
2169 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
2170 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2171 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
2172 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
2173 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
2174 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 | 2175 | |
d899e052 | 2176 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 2177 | btrfs_release_path(path); |
d899e052 YZ |
2178 | |
2179 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
2180 | |
2181 | ins.objectid = disk_bytenr; | |
2182 | ins.offset = disk_num_bytes; | |
2183 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2ff7e61e | 2184 | ret = btrfs_alloc_reserved_file_extent(trans, root->root_key.objectid, |
f85b7379 | 2185 | btrfs_ino(BTRFS_I(inode)), file_pos, ram_bytes, &ins); |
297d750b | 2186 | /* |
5846a3c2 QW |
2187 | * Release the reserved range from inode dirty range map, as it is |
2188 | * already moved into delayed_ref_head | |
297d750b QW |
2189 | */ |
2190 | btrfs_qgroup_release_data(inode, file_pos, ram_bytes); | |
79787eaa | 2191 | out: |
d899e052 | 2192 | btrfs_free_path(path); |
b9473439 | 2193 | |
79787eaa | 2194 | return ret; |
d899e052 YZ |
2195 | } |
2196 | ||
38c227d8 LB |
2197 | /* snapshot-aware defrag */ |
2198 | struct sa_defrag_extent_backref { | |
2199 | struct rb_node node; | |
2200 | struct old_sa_defrag_extent *old; | |
2201 | u64 root_id; | |
2202 | u64 inum; | |
2203 | u64 file_pos; | |
2204 | u64 extent_offset; | |
2205 | u64 num_bytes; | |
2206 | u64 generation; | |
2207 | }; | |
2208 | ||
2209 | struct old_sa_defrag_extent { | |
2210 | struct list_head list; | |
2211 | struct new_sa_defrag_extent *new; | |
2212 | ||
2213 | u64 extent_offset; | |
2214 | u64 bytenr; | |
2215 | u64 offset; | |
2216 | u64 len; | |
2217 | int count; | |
2218 | }; | |
2219 | ||
2220 | struct new_sa_defrag_extent { | |
2221 | struct rb_root root; | |
2222 | struct list_head head; | |
2223 | struct btrfs_path *path; | |
2224 | struct inode *inode; | |
2225 | u64 file_pos; | |
2226 | u64 len; | |
2227 | u64 bytenr; | |
2228 | u64 disk_len; | |
2229 | u8 compress_type; | |
2230 | }; | |
2231 | ||
2232 | static int backref_comp(struct sa_defrag_extent_backref *b1, | |
2233 | struct sa_defrag_extent_backref *b2) | |
2234 | { | |
2235 | if (b1->root_id < b2->root_id) | |
2236 | return -1; | |
2237 | else if (b1->root_id > b2->root_id) | |
2238 | return 1; | |
2239 | ||
2240 | if (b1->inum < b2->inum) | |
2241 | return -1; | |
2242 | else if (b1->inum > b2->inum) | |
2243 | return 1; | |
2244 | ||
2245 | if (b1->file_pos < b2->file_pos) | |
2246 | return -1; | |
2247 | else if (b1->file_pos > b2->file_pos) | |
2248 | return 1; | |
2249 | ||
2250 | /* | |
2251 | * [------------------------------] ===> (a range of space) | |
2252 | * |<--->| |<---->| =============> (fs/file tree A) | |
2253 | * |<---------------------------->| ===> (fs/file tree B) | |
2254 | * | |
2255 | * A range of space can refer to two file extents in one tree while | |
2256 | * refer to only one file extent in another tree. | |
2257 | * | |
2258 | * So we may process a disk offset more than one time(two extents in A) | |
2259 | * and locate at the same extent(one extent in B), then insert two same | |
2260 | * backrefs(both refer to the extent in B). | |
2261 | */ | |
2262 | return 0; | |
2263 | } | |
2264 | ||
2265 | static void backref_insert(struct rb_root *root, | |
2266 | struct sa_defrag_extent_backref *backref) | |
2267 | { | |
2268 | struct rb_node **p = &root->rb_node; | |
2269 | struct rb_node *parent = NULL; | |
2270 | struct sa_defrag_extent_backref *entry; | |
2271 | int ret; | |
2272 | ||
2273 | while (*p) { | |
2274 | parent = *p; | |
2275 | entry = rb_entry(parent, struct sa_defrag_extent_backref, node); | |
2276 | ||
2277 | ret = backref_comp(backref, entry); | |
2278 | if (ret < 0) | |
2279 | p = &(*p)->rb_left; | |
2280 | else | |
2281 | p = &(*p)->rb_right; | |
2282 | } | |
2283 | ||
2284 | rb_link_node(&backref->node, parent, p); | |
2285 | rb_insert_color(&backref->node, root); | |
2286 | } | |
2287 | ||
2288 | /* | |
2289 | * Note the backref might has changed, and in this case we just return 0. | |
2290 | */ | |
2291 | static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id, | |
2292 | void *ctx) | |
2293 | { | |
2294 | struct btrfs_file_extent_item *extent; | |
38c227d8 LB |
2295 | struct old_sa_defrag_extent *old = ctx; |
2296 | struct new_sa_defrag_extent *new = old->new; | |
2297 | struct btrfs_path *path = new->path; | |
2298 | struct btrfs_key key; | |
2299 | struct btrfs_root *root; | |
2300 | struct sa_defrag_extent_backref *backref; | |
2301 | struct extent_buffer *leaf; | |
2302 | struct inode *inode = new->inode; | |
0b246afa | 2303 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
38c227d8 LB |
2304 | int slot; |
2305 | int ret; | |
2306 | u64 extent_offset; | |
2307 | u64 num_bytes; | |
2308 | ||
2309 | if (BTRFS_I(inode)->root->root_key.objectid == root_id && | |
4a0cc7ca | 2310 | inum == btrfs_ino(BTRFS_I(inode))) |
38c227d8 LB |
2311 | return 0; |
2312 | ||
2313 | key.objectid = root_id; | |
2314 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2315 | key.offset = (u64)-1; | |
2316 | ||
38c227d8 LB |
2317 | root = btrfs_read_fs_root_no_name(fs_info, &key); |
2318 | if (IS_ERR(root)) { | |
2319 | if (PTR_ERR(root) == -ENOENT) | |
2320 | return 0; | |
2321 | WARN_ON(1); | |
ab8d0fc4 | 2322 | btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu", |
38c227d8 LB |
2323 | inum, offset, root_id); |
2324 | return PTR_ERR(root); | |
2325 | } | |
2326 | ||
2327 | key.objectid = inum; | |
2328 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2329 | if (offset > (u64)-1 << 32) | |
2330 | key.offset = 0; | |
2331 | else | |
2332 | key.offset = offset; | |
2333 | ||
2334 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
fae7f21c | 2335 | if (WARN_ON(ret < 0)) |
38c227d8 | 2336 | return ret; |
50f1319c | 2337 | ret = 0; |
38c227d8 LB |
2338 | |
2339 | while (1) { | |
2340 | cond_resched(); | |
2341 | ||
2342 | leaf = path->nodes[0]; | |
2343 | slot = path->slots[0]; | |
2344 | ||
2345 | if (slot >= btrfs_header_nritems(leaf)) { | |
2346 | ret = btrfs_next_leaf(root, path); | |
2347 | if (ret < 0) { | |
2348 | goto out; | |
2349 | } else if (ret > 0) { | |
2350 | ret = 0; | |
2351 | goto out; | |
2352 | } | |
2353 | continue; | |
2354 | } | |
2355 | ||
2356 | path->slots[0]++; | |
2357 | ||
2358 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
2359 | ||
2360 | if (key.objectid > inum) | |
2361 | goto out; | |
2362 | ||
2363 | if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY) | |
2364 | continue; | |
2365 | ||
2366 | extent = btrfs_item_ptr(leaf, slot, | |
2367 | struct btrfs_file_extent_item); | |
2368 | ||
2369 | if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr) | |
2370 | continue; | |
2371 | ||
e68afa49 LB |
2372 | /* |
2373 | * 'offset' refers to the exact key.offset, | |
2374 | * NOT the 'offset' field in btrfs_extent_data_ref, ie. | |
2375 | * (key.offset - extent_offset). | |
2376 | */ | |
2377 | if (key.offset != offset) | |
38c227d8 LB |
2378 | continue; |
2379 | ||
e68afa49 | 2380 | extent_offset = btrfs_file_extent_offset(leaf, extent); |
38c227d8 | 2381 | num_bytes = btrfs_file_extent_num_bytes(leaf, extent); |
e68afa49 | 2382 | |
38c227d8 LB |
2383 | if (extent_offset >= old->extent_offset + old->offset + |
2384 | old->len || extent_offset + num_bytes <= | |
2385 | old->extent_offset + old->offset) | |
2386 | continue; | |
38c227d8 LB |
2387 | break; |
2388 | } | |
2389 | ||
2390 | backref = kmalloc(sizeof(*backref), GFP_NOFS); | |
2391 | if (!backref) { | |
2392 | ret = -ENOENT; | |
2393 | goto out; | |
2394 | } | |
2395 | ||
2396 | backref->root_id = root_id; | |
2397 | backref->inum = inum; | |
e68afa49 | 2398 | backref->file_pos = offset; |
38c227d8 LB |
2399 | backref->num_bytes = num_bytes; |
2400 | backref->extent_offset = extent_offset; | |
2401 | backref->generation = btrfs_file_extent_generation(leaf, extent); | |
2402 | backref->old = old; | |
2403 | backref_insert(&new->root, backref); | |
2404 | old->count++; | |
2405 | out: | |
2406 | btrfs_release_path(path); | |
2407 | WARN_ON(ret); | |
2408 | return ret; | |
2409 | } | |
2410 | ||
2411 | static noinline bool record_extent_backrefs(struct btrfs_path *path, | |
2412 | struct new_sa_defrag_extent *new) | |
2413 | { | |
0b246afa | 2414 | struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb); |
38c227d8 LB |
2415 | struct old_sa_defrag_extent *old, *tmp; |
2416 | int ret; | |
2417 | ||
2418 | new->path = path; | |
2419 | ||
2420 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
e68afa49 LB |
2421 | ret = iterate_inodes_from_logical(old->bytenr + |
2422 | old->extent_offset, fs_info, | |
38c227d8 LB |
2423 | path, record_one_backref, |
2424 | old); | |
4724b106 JB |
2425 | if (ret < 0 && ret != -ENOENT) |
2426 | return false; | |
38c227d8 LB |
2427 | |
2428 | /* no backref to be processed for this extent */ | |
2429 | if (!old->count) { | |
2430 | list_del(&old->list); | |
2431 | kfree(old); | |
2432 | } | |
2433 | } | |
2434 | ||
2435 | if (list_empty(&new->head)) | |
2436 | return false; | |
2437 | ||
2438 | return true; | |
2439 | } | |
2440 | ||
2441 | static int relink_is_mergable(struct extent_buffer *leaf, | |
2442 | struct btrfs_file_extent_item *fi, | |
116e0024 | 2443 | struct new_sa_defrag_extent *new) |
38c227d8 | 2444 | { |
116e0024 | 2445 | if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr) |
38c227d8 LB |
2446 | return 0; |
2447 | ||
2448 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2449 | return 0; | |
2450 | ||
116e0024 LB |
2451 | if (btrfs_file_extent_compression(leaf, fi) != new->compress_type) |
2452 | return 0; | |
2453 | ||
2454 | if (btrfs_file_extent_encryption(leaf, fi) || | |
38c227d8 LB |
2455 | btrfs_file_extent_other_encoding(leaf, fi)) |
2456 | return 0; | |
2457 | ||
2458 | return 1; | |
2459 | } | |
2460 | ||
2461 | /* | |
2462 | * Note the backref might has changed, and in this case we just return 0. | |
2463 | */ | |
2464 | static noinline int relink_extent_backref(struct btrfs_path *path, | |
2465 | struct sa_defrag_extent_backref *prev, | |
2466 | struct sa_defrag_extent_backref *backref) | |
2467 | { | |
2468 | struct btrfs_file_extent_item *extent; | |
2469 | struct btrfs_file_extent_item *item; | |
2470 | struct btrfs_ordered_extent *ordered; | |
2471 | struct btrfs_trans_handle *trans; | |
38c227d8 LB |
2472 | struct btrfs_root *root; |
2473 | struct btrfs_key key; | |
2474 | struct extent_buffer *leaf; | |
2475 | struct old_sa_defrag_extent *old = backref->old; | |
2476 | struct new_sa_defrag_extent *new = old->new; | |
0b246afa | 2477 | struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb); |
38c227d8 LB |
2478 | struct inode *inode; |
2479 | struct extent_state *cached = NULL; | |
2480 | int ret = 0; | |
2481 | u64 start; | |
2482 | u64 len; | |
2483 | u64 lock_start; | |
2484 | u64 lock_end; | |
2485 | bool merge = false; | |
2486 | int index; | |
2487 | ||
2488 | if (prev && prev->root_id == backref->root_id && | |
2489 | prev->inum == backref->inum && | |
2490 | prev->file_pos + prev->num_bytes == backref->file_pos) | |
2491 | merge = true; | |
2492 | ||
2493 | /* step 1: get root */ | |
2494 | key.objectid = backref->root_id; | |
2495 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2496 | key.offset = (u64)-1; | |
2497 | ||
38c227d8 LB |
2498 | index = srcu_read_lock(&fs_info->subvol_srcu); |
2499 | ||
2500 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2501 | if (IS_ERR(root)) { | |
2502 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2503 | if (PTR_ERR(root) == -ENOENT) | |
2504 | return 0; | |
2505 | return PTR_ERR(root); | |
2506 | } | |
38c227d8 | 2507 | |
bcbba5e6 WS |
2508 | if (btrfs_root_readonly(root)) { |
2509 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2510 | return 0; | |
2511 | } | |
2512 | ||
38c227d8 LB |
2513 | /* step 2: get inode */ |
2514 | key.objectid = backref->inum; | |
2515 | key.type = BTRFS_INODE_ITEM_KEY; | |
2516 | key.offset = 0; | |
2517 | ||
2518 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
2519 | if (IS_ERR(inode)) { | |
2520 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2521 | return 0; | |
2522 | } | |
2523 | ||
2524 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2525 | ||
2526 | /* step 3: relink backref */ | |
2527 | lock_start = backref->file_pos; | |
2528 | lock_end = backref->file_pos + backref->num_bytes - 1; | |
2529 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
ff13db41 | 2530 | &cached); |
38c227d8 LB |
2531 | |
2532 | ordered = btrfs_lookup_first_ordered_extent(inode, lock_end); | |
2533 | if (ordered) { | |
2534 | btrfs_put_ordered_extent(ordered); | |
2535 | goto out_unlock; | |
2536 | } | |
2537 | ||
2538 | trans = btrfs_join_transaction(root); | |
2539 | if (IS_ERR(trans)) { | |
2540 | ret = PTR_ERR(trans); | |
2541 | goto out_unlock; | |
2542 | } | |
2543 | ||
2544 | key.objectid = backref->inum; | |
2545 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2546 | key.offset = backref->file_pos; | |
2547 | ||
2548 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2549 | if (ret < 0) { | |
2550 | goto out_free_path; | |
2551 | } else if (ret > 0) { | |
2552 | ret = 0; | |
2553 | goto out_free_path; | |
2554 | } | |
2555 | ||
2556 | extent = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2557 | struct btrfs_file_extent_item); | |
2558 | ||
2559 | if (btrfs_file_extent_generation(path->nodes[0], extent) != | |
2560 | backref->generation) | |
2561 | goto out_free_path; | |
2562 | ||
2563 | btrfs_release_path(path); | |
2564 | ||
2565 | start = backref->file_pos; | |
2566 | if (backref->extent_offset < old->extent_offset + old->offset) | |
2567 | start += old->extent_offset + old->offset - | |
2568 | backref->extent_offset; | |
2569 | ||
2570 | len = min(backref->extent_offset + backref->num_bytes, | |
2571 | old->extent_offset + old->offset + old->len); | |
2572 | len -= max(backref->extent_offset, old->extent_offset + old->offset); | |
2573 | ||
2574 | ret = btrfs_drop_extents(trans, root, inode, start, | |
2575 | start + len, 1); | |
2576 | if (ret) | |
2577 | goto out_free_path; | |
2578 | again: | |
4a0cc7ca | 2579 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
38c227d8 LB |
2580 | key.type = BTRFS_EXTENT_DATA_KEY; |
2581 | key.offset = start; | |
2582 | ||
a09a0a70 | 2583 | path->leave_spinning = 1; |
38c227d8 LB |
2584 | if (merge) { |
2585 | struct btrfs_file_extent_item *fi; | |
2586 | u64 extent_len; | |
2587 | struct btrfs_key found_key; | |
2588 | ||
3c9665df | 2589 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
38c227d8 LB |
2590 | if (ret < 0) |
2591 | goto out_free_path; | |
2592 | ||
2593 | path->slots[0]--; | |
2594 | leaf = path->nodes[0]; | |
2595 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2596 | ||
2597 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2598 | struct btrfs_file_extent_item); | |
2599 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
2600 | ||
116e0024 LB |
2601 | if (extent_len + found_key.offset == start && |
2602 | relink_is_mergable(leaf, fi, new)) { | |
38c227d8 LB |
2603 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2604 | extent_len + len); | |
2605 | btrfs_mark_buffer_dirty(leaf); | |
2606 | inode_add_bytes(inode, len); | |
2607 | ||
2608 | ret = 1; | |
2609 | goto out_free_path; | |
2610 | } else { | |
2611 | merge = false; | |
2612 | btrfs_release_path(path); | |
2613 | goto again; | |
2614 | } | |
2615 | } | |
2616 | ||
2617 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2618 | sizeof(*extent)); | |
2619 | if (ret) { | |
66642832 | 2620 | btrfs_abort_transaction(trans, ret); |
38c227d8 LB |
2621 | goto out_free_path; |
2622 | } | |
2623 | ||
2624 | leaf = path->nodes[0]; | |
2625 | item = btrfs_item_ptr(leaf, path->slots[0], | |
2626 | struct btrfs_file_extent_item); | |
2627 | btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr); | |
2628 | btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len); | |
2629 | btrfs_set_file_extent_offset(leaf, item, start - new->file_pos); | |
2630 | btrfs_set_file_extent_num_bytes(leaf, item, len); | |
2631 | btrfs_set_file_extent_ram_bytes(leaf, item, new->len); | |
2632 | btrfs_set_file_extent_generation(leaf, item, trans->transid); | |
2633 | btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG); | |
2634 | btrfs_set_file_extent_compression(leaf, item, new->compress_type); | |
2635 | btrfs_set_file_extent_encryption(leaf, item, 0); | |
2636 | btrfs_set_file_extent_other_encoding(leaf, item, 0); | |
2637 | ||
2638 | btrfs_mark_buffer_dirty(leaf); | |
2639 | inode_add_bytes(inode, len); | |
a09a0a70 | 2640 | btrfs_release_path(path); |
38c227d8 | 2641 | |
2ff7e61e | 2642 | ret = btrfs_inc_extent_ref(trans, fs_info, new->bytenr, |
38c227d8 LB |
2643 | new->disk_len, 0, |
2644 | backref->root_id, backref->inum, | |
b06c4bf5 | 2645 | new->file_pos); /* start - extent_offset */ |
38c227d8 | 2646 | if (ret) { |
66642832 | 2647 | btrfs_abort_transaction(trans, ret); |
38c227d8 LB |
2648 | goto out_free_path; |
2649 | } | |
2650 | ||
2651 | ret = 1; | |
2652 | out_free_path: | |
2653 | btrfs_release_path(path); | |
a09a0a70 | 2654 | path->leave_spinning = 0; |
3a45bb20 | 2655 | btrfs_end_transaction(trans); |
38c227d8 LB |
2656 | out_unlock: |
2657 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
2658 | &cached, GFP_NOFS); | |
2659 | iput(inode); | |
2660 | return ret; | |
2661 | } | |
2662 | ||
6f519564 LB |
2663 | static void free_sa_defrag_extent(struct new_sa_defrag_extent *new) |
2664 | { | |
2665 | struct old_sa_defrag_extent *old, *tmp; | |
2666 | ||
2667 | if (!new) | |
2668 | return; | |
2669 | ||
2670 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
6f519564 LB |
2671 | kfree(old); |
2672 | } | |
2673 | kfree(new); | |
2674 | } | |
2675 | ||
38c227d8 LB |
2676 | static void relink_file_extents(struct new_sa_defrag_extent *new) |
2677 | { | |
0b246afa | 2678 | struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb); |
38c227d8 | 2679 | struct btrfs_path *path; |
38c227d8 LB |
2680 | struct sa_defrag_extent_backref *backref; |
2681 | struct sa_defrag_extent_backref *prev = NULL; | |
2682 | struct inode *inode; | |
2683 | struct btrfs_root *root; | |
2684 | struct rb_node *node; | |
2685 | int ret; | |
2686 | ||
2687 | inode = new->inode; | |
2688 | root = BTRFS_I(inode)->root; | |
2689 | ||
2690 | path = btrfs_alloc_path(); | |
2691 | if (!path) | |
2692 | return; | |
2693 | ||
2694 | if (!record_extent_backrefs(path, new)) { | |
2695 | btrfs_free_path(path); | |
2696 | goto out; | |
2697 | } | |
2698 | btrfs_release_path(path); | |
2699 | ||
2700 | while (1) { | |
2701 | node = rb_first(&new->root); | |
2702 | if (!node) | |
2703 | break; | |
2704 | rb_erase(node, &new->root); | |
2705 | ||
2706 | backref = rb_entry(node, struct sa_defrag_extent_backref, node); | |
2707 | ||
2708 | ret = relink_extent_backref(path, prev, backref); | |
2709 | WARN_ON(ret < 0); | |
2710 | ||
2711 | kfree(prev); | |
2712 | ||
2713 | if (ret == 1) | |
2714 | prev = backref; | |
2715 | else | |
2716 | prev = NULL; | |
2717 | cond_resched(); | |
2718 | } | |
2719 | kfree(prev); | |
2720 | ||
2721 | btrfs_free_path(path); | |
38c227d8 | 2722 | out: |
6f519564 LB |
2723 | free_sa_defrag_extent(new); |
2724 | ||
0b246afa JM |
2725 | atomic_dec(&fs_info->defrag_running); |
2726 | wake_up(&fs_info->transaction_wait); | |
38c227d8 LB |
2727 | } |
2728 | ||
2729 | static struct new_sa_defrag_extent * | |
2730 | record_old_file_extents(struct inode *inode, | |
2731 | struct btrfs_ordered_extent *ordered) | |
2732 | { | |
0b246afa | 2733 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
38c227d8 LB |
2734 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2735 | struct btrfs_path *path; | |
2736 | struct btrfs_key key; | |
6f519564 | 2737 | struct old_sa_defrag_extent *old; |
38c227d8 LB |
2738 | struct new_sa_defrag_extent *new; |
2739 | int ret; | |
2740 | ||
2741 | new = kmalloc(sizeof(*new), GFP_NOFS); | |
2742 | if (!new) | |
2743 | return NULL; | |
2744 | ||
2745 | new->inode = inode; | |
2746 | new->file_pos = ordered->file_offset; | |
2747 | new->len = ordered->len; | |
2748 | new->bytenr = ordered->start; | |
2749 | new->disk_len = ordered->disk_len; | |
2750 | new->compress_type = ordered->compress_type; | |
2751 | new->root = RB_ROOT; | |
2752 | INIT_LIST_HEAD(&new->head); | |
2753 | ||
2754 | path = btrfs_alloc_path(); | |
2755 | if (!path) | |
2756 | goto out_kfree; | |
2757 | ||
4a0cc7ca | 2758 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
38c227d8 LB |
2759 | key.type = BTRFS_EXTENT_DATA_KEY; |
2760 | key.offset = new->file_pos; | |
2761 | ||
2762 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2763 | if (ret < 0) | |
2764 | goto out_free_path; | |
2765 | if (ret > 0 && path->slots[0] > 0) | |
2766 | path->slots[0]--; | |
2767 | ||
2768 | /* find out all the old extents for the file range */ | |
2769 | while (1) { | |
2770 | struct btrfs_file_extent_item *extent; | |
2771 | struct extent_buffer *l; | |
2772 | int slot; | |
2773 | u64 num_bytes; | |
2774 | u64 offset; | |
2775 | u64 end; | |
2776 | u64 disk_bytenr; | |
2777 | u64 extent_offset; | |
2778 | ||
2779 | l = path->nodes[0]; | |
2780 | slot = path->slots[0]; | |
2781 | ||
2782 | if (slot >= btrfs_header_nritems(l)) { | |
2783 | ret = btrfs_next_leaf(root, path); | |
2784 | if (ret < 0) | |
6f519564 | 2785 | goto out_free_path; |
38c227d8 LB |
2786 | else if (ret > 0) |
2787 | break; | |
2788 | continue; | |
2789 | } | |
2790 | ||
2791 | btrfs_item_key_to_cpu(l, &key, slot); | |
2792 | ||
4a0cc7ca | 2793 | if (key.objectid != btrfs_ino(BTRFS_I(inode))) |
38c227d8 LB |
2794 | break; |
2795 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2796 | break; | |
2797 | if (key.offset >= new->file_pos + new->len) | |
2798 | break; | |
2799 | ||
2800 | extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item); | |
2801 | ||
2802 | num_bytes = btrfs_file_extent_num_bytes(l, extent); | |
2803 | if (key.offset + num_bytes < new->file_pos) | |
2804 | goto next; | |
2805 | ||
2806 | disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent); | |
2807 | if (!disk_bytenr) | |
2808 | goto next; | |
2809 | ||
2810 | extent_offset = btrfs_file_extent_offset(l, extent); | |
2811 | ||
2812 | old = kmalloc(sizeof(*old), GFP_NOFS); | |
2813 | if (!old) | |
6f519564 | 2814 | goto out_free_path; |
38c227d8 LB |
2815 | |
2816 | offset = max(new->file_pos, key.offset); | |
2817 | end = min(new->file_pos + new->len, key.offset + num_bytes); | |
2818 | ||
2819 | old->bytenr = disk_bytenr; | |
2820 | old->extent_offset = extent_offset; | |
2821 | old->offset = offset - key.offset; | |
2822 | old->len = end - offset; | |
2823 | old->new = new; | |
2824 | old->count = 0; | |
2825 | list_add_tail(&old->list, &new->head); | |
2826 | next: | |
2827 | path->slots[0]++; | |
2828 | cond_resched(); | |
2829 | } | |
2830 | ||
2831 | btrfs_free_path(path); | |
0b246afa | 2832 | atomic_inc(&fs_info->defrag_running); |
38c227d8 LB |
2833 | |
2834 | return new; | |
2835 | ||
38c227d8 LB |
2836 | out_free_path: |
2837 | btrfs_free_path(path); | |
2838 | out_kfree: | |
6f519564 | 2839 | free_sa_defrag_extent(new); |
38c227d8 LB |
2840 | return NULL; |
2841 | } | |
2842 | ||
2ff7e61e | 2843 | static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
2844 | u64 start, u64 len) |
2845 | { | |
2846 | struct btrfs_block_group_cache *cache; | |
2847 | ||
0b246afa | 2848 | cache = btrfs_lookup_block_group(fs_info, start); |
e570fd27 MX |
2849 | ASSERT(cache); |
2850 | ||
2851 | spin_lock(&cache->lock); | |
2852 | cache->delalloc_bytes -= len; | |
2853 | spin_unlock(&cache->lock); | |
2854 | ||
2855 | btrfs_put_block_group(cache); | |
2856 | } | |
2857 | ||
d352ac68 CM |
2858 | /* as ordered data IO finishes, this gets called so we can finish |
2859 | * an ordered extent if the range of bytes in the file it covers are | |
2860 | * fully written. | |
2861 | */ | |
5fd02043 | 2862 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 2863 | { |
5fd02043 | 2864 | struct inode *inode = ordered_extent->inode; |
0b246afa | 2865 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc | 2866 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 2867 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 2868 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 2869 | struct extent_state *cached_state = NULL; |
38c227d8 | 2870 | struct new_sa_defrag_extent *new = NULL; |
261507a0 | 2871 | int compress_type = 0; |
77cef2ec JB |
2872 | int ret = 0; |
2873 | u64 logical_len = ordered_extent->len; | |
82d5902d | 2874 | bool nolock; |
77cef2ec | 2875 | bool truncated = false; |
e6dcd2dc | 2876 | |
83eea1f1 | 2877 | nolock = btrfs_is_free_space_inode(inode); |
0cb59c99 | 2878 | |
5fd02043 JB |
2879 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
2880 | ret = -EIO; | |
2881 | goto out; | |
2882 | } | |
2883 | ||
f612496b MX |
2884 | btrfs_free_io_failure_record(inode, ordered_extent->file_offset, |
2885 | ordered_extent->file_offset + | |
2886 | ordered_extent->len - 1); | |
2887 | ||
77cef2ec JB |
2888 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) { |
2889 | truncated = true; | |
2890 | logical_len = ordered_extent->truncated_len; | |
2891 | /* Truncated the entire extent, don't bother adding */ | |
2892 | if (!logical_len) | |
2893 | goto out; | |
2894 | } | |
2895 | ||
c2167754 | 2896 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 2897 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
94ed938a QW |
2898 | |
2899 | /* | |
2900 | * For mwrite(mmap + memset to write) case, we still reserve | |
2901 | * space for NOCOW range. | |
2902 | * As NOCOW won't cause a new delayed ref, just free the space | |
2903 | */ | |
2904 | btrfs_qgroup_free_data(inode, ordered_extent->file_offset, | |
2905 | ordered_extent->len); | |
6c760c07 JB |
2906 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2907 | if (nolock) | |
2908 | trans = btrfs_join_transaction_nolock(root); | |
2909 | else | |
2910 | trans = btrfs_join_transaction(root); | |
2911 | if (IS_ERR(trans)) { | |
2912 | ret = PTR_ERR(trans); | |
2913 | trans = NULL; | |
2914 | goto out; | |
c2167754 | 2915 | } |
0b246afa | 2916 | trans->block_rsv = &fs_info->delalloc_block_rsv; |
6c760c07 JB |
2917 | ret = btrfs_update_inode_fallback(trans, root, inode); |
2918 | if (ret) /* -ENOMEM or corruption */ | |
66642832 | 2919 | btrfs_abort_transaction(trans, ret); |
c2167754 YZ |
2920 | goto out; |
2921 | } | |
e6dcd2dc | 2922 | |
2ac55d41 JB |
2923 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
2924 | ordered_extent->file_offset + ordered_extent->len - 1, | |
ff13db41 | 2925 | &cached_state); |
e6dcd2dc | 2926 | |
38c227d8 LB |
2927 | ret = test_range_bit(io_tree, ordered_extent->file_offset, |
2928 | ordered_extent->file_offset + ordered_extent->len - 1, | |
2929 | EXTENT_DEFRAG, 1, cached_state); | |
2930 | if (ret) { | |
2931 | u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item); | |
8101c8db | 2932 | if (0 && last_snapshot >= BTRFS_I(inode)->generation) |
38c227d8 LB |
2933 | /* the inode is shared */ |
2934 | new = record_old_file_extents(inode, ordered_extent); | |
2935 | ||
2936 | clear_extent_bit(io_tree, ordered_extent->file_offset, | |
2937 | ordered_extent->file_offset + ordered_extent->len - 1, | |
2938 | EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS); | |
2939 | } | |
2940 | ||
0cb59c99 | 2941 | if (nolock) |
7a7eaa40 | 2942 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 2943 | else |
7a7eaa40 | 2944 | trans = btrfs_join_transaction(root); |
79787eaa JM |
2945 | if (IS_ERR(trans)) { |
2946 | ret = PTR_ERR(trans); | |
2947 | trans = NULL; | |
2948 | goto out_unlock; | |
2949 | } | |
a79b7d4b | 2950 | |
0b246afa | 2951 | trans->block_rsv = &fs_info->delalloc_block_rsv; |
c2167754 | 2952 | |
c8b97818 | 2953 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 2954 | compress_type = ordered_extent->compress_type; |
d899e052 | 2955 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 2956 | BUG_ON(compress_type); |
920bbbfb | 2957 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
2958 | ordered_extent->file_offset, |
2959 | ordered_extent->file_offset + | |
77cef2ec | 2960 | logical_len); |
d899e052 | 2961 | } else { |
0b246afa | 2962 | BUG_ON(root == fs_info->tree_root); |
d899e052 YZ |
2963 | ret = insert_reserved_file_extent(trans, inode, |
2964 | ordered_extent->file_offset, | |
2965 | ordered_extent->start, | |
2966 | ordered_extent->disk_len, | |
77cef2ec | 2967 | logical_len, logical_len, |
261507a0 | 2968 | compress_type, 0, 0, |
d899e052 | 2969 | BTRFS_FILE_EXTENT_REG); |
e570fd27 | 2970 | if (!ret) |
2ff7e61e | 2971 | btrfs_release_delalloc_bytes(fs_info, |
e570fd27 MX |
2972 | ordered_extent->start, |
2973 | ordered_extent->disk_len); | |
d899e052 | 2974 | } |
5dc562c5 JB |
2975 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
2976 | ordered_extent->file_offset, ordered_extent->len, | |
2977 | trans->transid); | |
79787eaa | 2978 | if (ret < 0) { |
66642832 | 2979 | btrfs_abort_transaction(trans, ret); |
5fd02043 | 2980 | goto out_unlock; |
79787eaa | 2981 | } |
2ac55d41 | 2982 | |
e6dcd2dc CM |
2983 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
2984 | &ordered_extent->list); | |
2985 | ||
6c760c07 JB |
2986 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2987 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
2988 | if (ret) { /* -ENOMEM or corruption */ | |
66642832 | 2989 | btrfs_abort_transaction(trans, ret); |
6c760c07 | 2990 | goto out_unlock; |
1ef30be1 JB |
2991 | } |
2992 | ret = 0; | |
5fd02043 JB |
2993 | out_unlock: |
2994 | unlock_extent_cached(io_tree, ordered_extent->file_offset, | |
2995 | ordered_extent->file_offset + | |
2996 | ordered_extent->len - 1, &cached_state, GFP_NOFS); | |
c2167754 | 2997 | out: |
0b246afa | 2998 | if (root != fs_info->tree_root) |
0cb59c99 | 2999 | btrfs_delalloc_release_metadata(inode, ordered_extent->len); |
a698d075 | 3000 | if (trans) |
3a45bb20 | 3001 | btrfs_end_transaction(trans); |
0cb59c99 | 3002 | |
77cef2ec JB |
3003 | if (ret || truncated) { |
3004 | u64 start, end; | |
3005 | ||
3006 | if (truncated) | |
3007 | start = ordered_extent->file_offset + logical_len; | |
3008 | else | |
3009 | start = ordered_extent->file_offset; | |
3010 | end = ordered_extent->file_offset + ordered_extent->len - 1; | |
3011 | clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS); | |
3012 | ||
3013 | /* Drop the cache for the part of the extent we didn't write. */ | |
3014 | btrfs_drop_extent_cache(inode, start, end, 0); | |
5fd02043 | 3015 | |
0bec9ef5 JB |
3016 | /* |
3017 | * If the ordered extent had an IOERR or something else went | |
3018 | * wrong we need to return the space for this ordered extent | |
77cef2ec JB |
3019 | * back to the allocator. We only free the extent in the |
3020 | * truncated case if we didn't write out the extent at all. | |
0bec9ef5 | 3021 | */ |
77cef2ec JB |
3022 | if ((ret || !logical_len) && |
3023 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && | |
0bec9ef5 | 3024 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) |
2ff7e61e JM |
3025 | btrfs_free_reserved_extent(fs_info, |
3026 | ordered_extent->start, | |
e570fd27 | 3027 | ordered_extent->disk_len, 1); |
0bec9ef5 JB |
3028 | } |
3029 | ||
3030 | ||
5fd02043 | 3031 | /* |
8bad3c02 LB |
3032 | * This needs to be done to make sure anybody waiting knows we are done |
3033 | * updating everything for this ordered extent. | |
5fd02043 JB |
3034 | */ |
3035 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
3036 | ||
38c227d8 | 3037 | /* for snapshot-aware defrag */ |
6f519564 LB |
3038 | if (new) { |
3039 | if (ret) { | |
3040 | free_sa_defrag_extent(new); | |
0b246afa | 3041 | atomic_dec(&fs_info->defrag_running); |
6f519564 LB |
3042 | } else { |
3043 | relink_file_extents(new); | |
3044 | } | |
3045 | } | |
38c227d8 | 3046 | |
e6dcd2dc CM |
3047 | /* once for us */ |
3048 | btrfs_put_ordered_extent(ordered_extent); | |
3049 | /* once for the tree */ | |
3050 | btrfs_put_ordered_extent(ordered_extent); | |
3051 | ||
5fd02043 JB |
3052 | return ret; |
3053 | } | |
3054 | ||
3055 | static void finish_ordered_fn(struct btrfs_work *work) | |
3056 | { | |
3057 | struct btrfs_ordered_extent *ordered_extent; | |
3058 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
3059 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
3060 | } |
3061 | ||
b2950863 | 3062 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
3063 | struct extent_state *state, int uptodate) |
3064 | { | |
5fd02043 | 3065 | struct inode *inode = page->mapping->host; |
0b246afa | 3066 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5fd02043 | 3067 | struct btrfs_ordered_extent *ordered_extent = NULL; |
9e0af237 LB |
3068 | struct btrfs_workqueue *wq; |
3069 | btrfs_work_func_t func; | |
5fd02043 | 3070 | |
1abe9b8a | 3071 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
3072 | ||
8b62b72b | 3073 | ClearPagePrivate2(page); |
5fd02043 JB |
3074 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
3075 | end - start + 1, uptodate)) | |
3076 | return 0; | |
3077 | ||
9e0af237 | 3078 | if (btrfs_is_free_space_inode(inode)) { |
0b246afa | 3079 | wq = fs_info->endio_freespace_worker; |
9e0af237 LB |
3080 | func = btrfs_freespace_write_helper; |
3081 | } else { | |
0b246afa | 3082 | wq = fs_info->endio_write_workers; |
9e0af237 LB |
3083 | func = btrfs_endio_write_helper; |
3084 | } | |
5fd02043 | 3085 | |
9e0af237 LB |
3086 | btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL, |
3087 | NULL); | |
3088 | btrfs_queue_work(wq, &ordered_extent->work); | |
5fd02043 JB |
3089 | |
3090 | return 0; | |
211f90e6 CM |
3091 | } |
3092 | ||
dc380aea MX |
3093 | static int __readpage_endio_check(struct inode *inode, |
3094 | struct btrfs_io_bio *io_bio, | |
3095 | int icsum, struct page *page, | |
3096 | int pgoff, u64 start, size_t len) | |
3097 | { | |
3098 | char *kaddr; | |
3099 | u32 csum_expected; | |
3100 | u32 csum = ~(u32)0; | |
dc380aea MX |
3101 | |
3102 | csum_expected = *(((u32 *)io_bio->csum) + icsum); | |
3103 | ||
3104 | kaddr = kmap_atomic(page); | |
3105 | csum = btrfs_csum_data(kaddr + pgoff, csum, len); | |
0b5e3daf | 3106 | btrfs_csum_final(csum, (u8 *)&csum); |
dc380aea MX |
3107 | if (csum != csum_expected) |
3108 | goto zeroit; | |
3109 | ||
3110 | kunmap_atomic(kaddr); | |
3111 | return 0; | |
3112 | zeroit: | |
94647322 DS |
3113 | btrfs_warn_rl(BTRFS_I(inode)->root->fs_info, |
3114 | "csum failed ino %llu off %llu csum %u expected csum %u", | |
f85b7379 | 3115 | btrfs_ino(BTRFS_I(inode)), start, csum, csum_expected); |
dc380aea MX |
3116 | memset(kaddr + pgoff, 1, len); |
3117 | flush_dcache_page(page); | |
3118 | kunmap_atomic(kaddr); | |
3119 | if (csum_expected == 0) | |
3120 | return 0; | |
3121 | return -EIO; | |
3122 | } | |
3123 | ||
d352ac68 CM |
3124 | /* |
3125 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
3126 | * if there's a match, we allow the bio to finish. If not, the code in |
3127 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 3128 | */ |
facc8a22 MX |
3129 | static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio, |
3130 | u64 phy_offset, struct page *page, | |
3131 | u64 start, u64 end, int mirror) | |
07157aac | 3132 | { |
4eee4fa4 | 3133 | size_t offset = start - page_offset(page); |
07157aac | 3134 | struct inode *inode = page->mapping->host; |
d1310b2e | 3135 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
ff79f819 | 3136 | struct btrfs_root *root = BTRFS_I(inode)->root; |
d1310b2e | 3137 | |
d20f7043 CM |
3138 | if (PageChecked(page)) { |
3139 | ClearPageChecked(page); | |
dc380aea | 3140 | return 0; |
d20f7043 | 3141 | } |
6cbff00f CH |
3142 | |
3143 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
dc380aea | 3144 | return 0; |
17d217fe YZ |
3145 | |
3146 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 3147 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
91166212 | 3148 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM); |
b6cda9bc | 3149 | return 0; |
17d217fe | 3150 | } |
d20f7043 | 3151 | |
facc8a22 | 3152 | phy_offset >>= inode->i_sb->s_blocksize_bits; |
dc380aea MX |
3153 | return __readpage_endio_check(inode, io_bio, phy_offset, page, offset, |
3154 | start, (size_t)(end - start + 1)); | |
07157aac | 3155 | } |
b888db2b | 3156 | |
24bbcf04 YZ |
3157 | void btrfs_add_delayed_iput(struct inode *inode) |
3158 | { | |
0b246afa | 3159 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
8089fe62 | 3160 | struct btrfs_inode *binode = BTRFS_I(inode); |
24bbcf04 YZ |
3161 | |
3162 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
3163 | return; | |
3164 | ||
24bbcf04 | 3165 | spin_lock(&fs_info->delayed_iput_lock); |
8089fe62 DS |
3166 | if (binode->delayed_iput_count == 0) { |
3167 | ASSERT(list_empty(&binode->delayed_iput)); | |
3168 | list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs); | |
3169 | } else { | |
3170 | binode->delayed_iput_count++; | |
3171 | } | |
24bbcf04 YZ |
3172 | spin_unlock(&fs_info->delayed_iput_lock); |
3173 | } | |
3174 | ||
2ff7e61e | 3175 | void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info) |
24bbcf04 | 3176 | { |
24bbcf04 | 3177 | |
24bbcf04 | 3178 | spin_lock(&fs_info->delayed_iput_lock); |
8089fe62 DS |
3179 | while (!list_empty(&fs_info->delayed_iputs)) { |
3180 | struct btrfs_inode *inode; | |
3181 | ||
3182 | inode = list_first_entry(&fs_info->delayed_iputs, | |
3183 | struct btrfs_inode, delayed_iput); | |
3184 | if (inode->delayed_iput_count) { | |
3185 | inode->delayed_iput_count--; | |
3186 | list_move_tail(&inode->delayed_iput, | |
3187 | &fs_info->delayed_iputs); | |
3188 | } else { | |
3189 | list_del_init(&inode->delayed_iput); | |
3190 | } | |
3191 | spin_unlock(&fs_info->delayed_iput_lock); | |
3192 | iput(&inode->vfs_inode); | |
3193 | spin_lock(&fs_info->delayed_iput_lock); | |
24bbcf04 | 3194 | } |
8089fe62 | 3195 | spin_unlock(&fs_info->delayed_iput_lock); |
24bbcf04 YZ |
3196 | } |
3197 | ||
d68fc57b | 3198 | /* |
42b2aa86 | 3199 | * This is called in transaction commit time. If there are no orphan |
d68fc57b YZ |
3200 | * files in the subvolume, it removes orphan item and frees block_rsv |
3201 | * structure. | |
3202 | */ | |
3203 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
3204 | struct btrfs_root *root) | |
3205 | { | |
0b246afa | 3206 | struct btrfs_fs_info *fs_info = root->fs_info; |
90290e19 | 3207 | struct btrfs_block_rsv *block_rsv; |
d68fc57b YZ |
3208 | int ret; |
3209 | ||
8a35d95f | 3210 | if (atomic_read(&root->orphan_inodes) || |
d68fc57b YZ |
3211 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) |
3212 | return; | |
3213 | ||
90290e19 | 3214 | spin_lock(&root->orphan_lock); |
8a35d95f | 3215 | if (atomic_read(&root->orphan_inodes)) { |
90290e19 JB |
3216 | spin_unlock(&root->orphan_lock); |
3217 | return; | |
3218 | } | |
3219 | ||
3220 | if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) { | |
3221 | spin_unlock(&root->orphan_lock); | |
3222 | return; | |
3223 | } | |
3224 | ||
3225 | block_rsv = root->orphan_block_rsv; | |
3226 | root->orphan_block_rsv = NULL; | |
3227 | spin_unlock(&root->orphan_lock); | |
3228 | ||
27cdeb70 | 3229 | if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) && |
d68fc57b | 3230 | btrfs_root_refs(&root->root_item) > 0) { |
0b246afa | 3231 | ret = btrfs_del_orphan_item(trans, fs_info->tree_root, |
d68fc57b | 3232 | root->root_key.objectid); |
4ef31a45 | 3233 | if (ret) |
66642832 | 3234 | btrfs_abort_transaction(trans, ret); |
4ef31a45 | 3235 | else |
27cdeb70 MX |
3236 | clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, |
3237 | &root->state); | |
d68fc57b YZ |
3238 | } |
3239 | ||
90290e19 JB |
3240 | if (block_rsv) { |
3241 | WARN_ON(block_rsv->size > 0); | |
2ff7e61e | 3242 | btrfs_free_block_rsv(fs_info, block_rsv); |
d68fc57b YZ |
3243 | } |
3244 | } | |
3245 | ||
7b128766 JB |
3246 | /* |
3247 | * This creates an orphan entry for the given inode in case something goes | |
3248 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
3249 | * |
3250 | * NOTE: caller of this function should reserve 5 units of metadata for | |
3251 | * this function. | |
7b128766 JB |
3252 | */ |
3253 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
3254 | { | |
0b246afa | 3255 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
7b128766 | 3256 | struct btrfs_root *root = BTRFS_I(inode)->root; |
d68fc57b YZ |
3257 | struct btrfs_block_rsv *block_rsv = NULL; |
3258 | int reserve = 0; | |
3259 | int insert = 0; | |
3260 | int ret; | |
7b128766 | 3261 | |
d68fc57b | 3262 | if (!root->orphan_block_rsv) { |
2ff7e61e JM |
3263 | block_rsv = btrfs_alloc_block_rsv(fs_info, |
3264 | BTRFS_BLOCK_RSV_TEMP); | |
b532402e TI |
3265 | if (!block_rsv) |
3266 | return -ENOMEM; | |
d68fc57b | 3267 | } |
7b128766 | 3268 | |
d68fc57b YZ |
3269 | spin_lock(&root->orphan_lock); |
3270 | if (!root->orphan_block_rsv) { | |
3271 | root->orphan_block_rsv = block_rsv; | |
3272 | } else if (block_rsv) { | |
2ff7e61e | 3273 | btrfs_free_block_rsv(fs_info, block_rsv); |
d68fc57b | 3274 | block_rsv = NULL; |
7b128766 | 3275 | } |
7b128766 | 3276 | |
8a35d95f JB |
3277 | if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3278 | &BTRFS_I(inode)->runtime_flags)) { | |
d68fc57b YZ |
3279 | #if 0 |
3280 | /* | |
3281 | * For proper ENOSPC handling, we should do orphan | |
3282 | * cleanup when mounting. But this introduces backward | |
3283 | * compatibility issue. | |
3284 | */ | |
3285 | if (!xchg(&root->orphan_item_inserted, 1)) | |
3286 | insert = 2; | |
3287 | else | |
3288 | insert = 1; | |
3289 | #endif | |
3290 | insert = 1; | |
321f0e70 | 3291 | atomic_inc(&root->orphan_inodes); |
7b128766 JB |
3292 | } |
3293 | ||
72ac3c0d JB |
3294 | if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
3295 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3296 | reserve = 1; |
d68fc57b | 3297 | spin_unlock(&root->orphan_lock); |
7b128766 | 3298 | |
d68fc57b YZ |
3299 | /* grab metadata reservation from transaction handle */ |
3300 | if (reserve) { | |
3301 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
3b6571c1 JB |
3302 | ASSERT(!ret); |
3303 | if (ret) { | |
3304 | atomic_dec(&root->orphan_inodes); | |
3305 | clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, | |
3306 | &BTRFS_I(inode)->runtime_flags); | |
3307 | if (insert) | |
3308 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, | |
3309 | &BTRFS_I(inode)->runtime_flags); | |
3310 | return ret; | |
3311 | } | |
d68fc57b | 3312 | } |
7b128766 | 3313 | |
d68fc57b YZ |
3314 | /* insert an orphan item to track this unlinked/truncated file */ |
3315 | if (insert >= 1) { | |
f85b7379 DS |
3316 | ret = btrfs_insert_orphan_item(trans, root, |
3317 | btrfs_ino(BTRFS_I(inode))); | |
4ef31a45 | 3318 | if (ret) { |
703c88e0 | 3319 | atomic_dec(&root->orphan_inodes); |
4ef31a45 JB |
3320 | if (reserve) { |
3321 | clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, | |
3322 | &BTRFS_I(inode)->runtime_flags); | |
3323 | btrfs_orphan_release_metadata(inode); | |
3324 | } | |
3325 | if (ret != -EEXIST) { | |
e8e7cff6 JB |
3326 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3327 | &BTRFS_I(inode)->runtime_flags); | |
66642832 | 3328 | btrfs_abort_transaction(trans, ret); |
4ef31a45 JB |
3329 | return ret; |
3330 | } | |
79787eaa JM |
3331 | } |
3332 | ret = 0; | |
d68fc57b YZ |
3333 | } |
3334 | ||
3335 | /* insert an orphan item to track subvolume contains orphan files */ | |
3336 | if (insert >= 2) { | |
0b246afa | 3337 | ret = btrfs_insert_orphan_item(trans, fs_info->tree_root, |
d68fc57b | 3338 | root->root_key.objectid); |
79787eaa | 3339 | if (ret && ret != -EEXIST) { |
66642832 | 3340 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
3341 | return ret; |
3342 | } | |
d68fc57b YZ |
3343 | } |
3344 | return 0; | |
7b128766 JB |
3345 | } |
3346 | ||
3347 | /* | |
3348 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
3349 | * item for this particular inode. | |
3350 | */ | |
48a3b636 ES |
3351 | static int btrfs_orphan_del(struct btrfs_trans_handle *trans, |
3352 | struct inode *inode) | |
7b128766 JB |
3353 | { |
3354 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
3355 | int delete_item = 0; |
3356 | int release_rsv = 0; | |
7b128766 JB |
3357 | int ret = 0; |
3358 | ||
d68fc57b | 3359 | spin_lock(&root->orphan_lock); |
8a35d95f JB |
3360 | if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3361 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3362 | delete_item = 1; |
7b128766 | 3363 | |
72ac3c0d JB |
3364 | if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
3365 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3366 | release_rsv = 1; |
d68fc57b | 3367 | spin_unlock(&root->orphan_lock); |
7b128766 | 3368 | |
703c88e0 | 3369 | if (delete_item) { |
8a35d95f | 3370 | atomic_dec(&root->orphan_inodes); |
703c88e0 FDBM |
3371 | if (trans) |
3372 | ret = btrfs_del_orphan_item(trans, root, | |
4a0cc7ca | 3373 | btrfs_ino(BTRFS_I(inode))); |
8a35d95f | 3374 | } |
7b128766 | 3375 | |
703c88e0 FDBM |
3376 | if (release_rsv) |
3377 | btrfs_orphan_release_metadata(inode); | |
3378 | ||
4ef31a45 | 3379 | return ret; |
7b128766 JB |
3380 | } |
3381 | ||
3382 | /* | |
3383 | * this cleans up any orphans that may be left on the list from the last use | |
3384 | * of this root. | |
3385 | */ | |
66b4ffd1 | 3386 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 | 3387 | { |
0b246afa | 3388 | struct btrfs_fs_info *fs_info = root->fs_info; |
7b128766 JB |
3389 | struct btrfs_path *path; |
3390 | struct extent_buffer *leaf; | |
7b128766 JB |
3391 | struct btrfs_key key, found_key; |
3392 | struct btrfs_trans_handle *trans; | |
3393 | struct inode *inode; | |
8f6d7f4f | 3394 | u64 last_objectid = 0; |
7b128766 JB |
3395 | int ret = 0, nr_unlink = 0, nr_truncate = 0; |
3396 | ||
d68fc57b | 3397 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 3398 | return 0; |
c71bf099 YZ |
3399 | |
3400 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
3401 | if (!path) { |
3402 | ret = -ENOMEM; | |
3403 | goto out; | |
3404 | } | |
e4058b54 | 3405 | path->reada = READA_BACK; |
7b128766 JB |
3406 | |
3407 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
962a298f | 3408 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
7b128766 JB |
3409 | key.offset = (u64)-1; |
3410 | ||
7b128766 JB |
3411 | while (1) { |
3412 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
3413 | if (ret < 0) |
3414 | goto out; | |
7b128766 JB |
3415 | |
3416 | /* | |
3417 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 3418 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
3419 | * find the key and see if we have stuff that matches |
3420 | */ | |
3421 | if (ret > 0) { | |
66b4ffd1 | 3422 | ret = 0; |
7b128766 JB |
3423 | if (path->slots[0] == 0) |
3424 | break; | |
3425 | path->slots[0]--; | |
3426 | } | |
3427 | ||
3428 | /* pull out the item */ | |
3429 | leaf = path->nodes[0]; | |
7b128766 JB |
3430 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3431 | ||
3432 | /* make sure the item matches what we want */ | |
3433 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
3434 | break; | |
962a298f | 3435 | if (found_key.type != BTRFS_ORPHAN_ITEM_KEY) |
7b128766 JB |
3436 | break; |
3437 | ||
3438 | /* release the path since we're done with it */ | |
b3b4aa74 | 3439 | btrfs_release_path(path); |
7b128766 JB |
3440 | |
3441 | /* | |
3442 | * this is where we are basically btrfs_lookup, without the | |
3443 | * crossing root thing. we store the inode number in the | |
3444 | * offset of the orphan item. | |
3445 | */ | |
8f6d7f4f JB |
3446 | |
3447 | if (found_key.offset == last_objectid) { | |
0b246afa JM |
3448 | btrfs_err(fs_info, |
3449 | "Error removing orphan entry, stopping orphan cleanup"); | |
8f6d7f4f JB |
3450 | ret = -EINVAL; |
3451 | goto out; | |
3452 | } | |
3453 | ||
3454 | last_objectid = found_key.offset; | |
3455 | ||
5d4f98a2 YZ |
3456 | found_key.objectid = found_key.offset; |
3457 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
3458 | found_key.offset = 0; | |
0b246afa | 3459 | inode = btrfs_iget(fs_info->sb, &found_key, root, NULL); |
8c6ffba0 | 3460 | ret = PTR_ERR_OR_ZERO(inode); |
67710892 | 3461 | if (ret && ret != -ENOENT) |
66b4ffd1 | 3462 | goto out; |
7b128766 | 3463 | |
0b246afa | 3464 | if (ret == -ENOENT && root == fs_info->tree_root) { |
f8e9e0b0 AJ |
3465 | struct btrfs_root *dead_root; |
3466 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3467 | int is_dead_root = 0; | |
3468 | ||
3469 | /* | |
3470 | * this is an orphan in the tree root. Currently these | |
3471 | * could come from 2 sources: | |
3472 | * a) a snapshot deletion in progress | |
3473 | * b) a free space cache inode | |
3474 | * We need to distinguish those two, as the snapshot | |
3475 | * orphan must not get deleted. | |
3476 | * find_dead_roots already ran before us, so if this | |
3477 | * is a snapshot deletion, we should find the root | |
3478 | * in the dead_roots list | |
3479 | */ | |
3480 | spin_lock(&fs_info->trans_lock); | |
3481 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
3482 | root_list) { | |
3483 | if (dead_root->root_key.objectid == | |
3484 | found_key.objectid) { | |
3485 | is_dead_root = 1; | |
3486 | break; | |
3487 | } | |
3488 | } | |
3489 | spin_unlock(&fs_info->trans_lock); | |
3490 | if (is_dead_root) { | |
3491 | /* prevent this orphan from being found again */ | |
3492 | key.offset = found_key.objectid - 1; | |
3493 | continue; | |
3494 | } | |
3495 | } | |
7b128766 | 3496 | /* |
a8c9e576 JB |
3497 | * Inode is already gone but the orphan item is still there, |
3498 | * kill the orphan item. | |
7b128766 | 3499 | */ |
67710892 | 3500 | if (ret == -ENOENT) { |
a8c9e576 | 3501 | trans = btrfs_start_transaction(root, 1); |
66b4ffd1 JB |
3502 | if (IS_ERR(trans)) { |
3503 | ret = PTR_ERR(trans); | |
3504 | goto out; | |
3505 | } | |
0b246afa JM |
3506 | btrfs_debug(fs_info, "auto deleting %Lu", |
3507 | found_key.objectid); | |
a8c9e576 JB |
3508 | ret = btrfs_del_orphan_item(trans, root, |
3509 | found_key.objectid); | |
3a45bb20 | 3510 | btrfs_end_transaction(trans); |
4ef31a45 JB |
3511 | if (ret) |
3512 | goto out; | |
7b128766 JB |
3513 | continue; |
3514 | } | |
3515 | ||
a8c9e576 JB |
3516 | /* |
3517 | * add this inode to the orphan list so btrfs_orphan_del does | |
3518 | * the proper thing when we hit it | |
3519 | */ | |
8a35d95f JB |
3520 | set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3521 | &BTRFS_I(inode)->runtime_flags); | |
925396ec | 3522 | atomic_inc(&root->orphan_inodes); |
a8c9e576 | 3523 | |
7b128766 JB |
3524 | /* if we have links, this was a truncate, lets do that */ |
3525 | if (inode->i_nlink) { | |
fae7f21c | 3526 | if (WARN_ON(!S_ISREG(inode->i_mode))) { |
a41ad394 JB |
3527 | iput(inode); |
3528 | continue; | |
3529 | } | |
7b128766 | 3530 | nr_truncate++; |
f3fe820c JB |
3531 | |
3532 | /* 1 for the orphan item deletion. */ | |
3533 | trans = btrfs_start_transaction(root, 1); | |
3534 | if (IS_ERR(trans)) { | |
c69b26b0 | 3535 | iput(inode); |
f3fe820c JB |
3536 | ret = PTR_ERR(trans); |
3537 | goto out; | |
3538 | } | |
3539 | ret = btrfs_orphan_add(trans, inode); | |
3a45bb20 | 3540 | btrfs_end_transaction(trans); |
c69b26b0 JB |
3541 | if (ret) { |
3542 | iput(inode); | |
f3fe820c | 3543 | goto out; |
c69b26b0 | 3544 | } |
f3fe820c | 3545 | |
66b4ffd1 | 3546 | ret = btrfs_truncate(inode); |
4a7d0f68 JB |
3547 | if (ret) |
3548 | btrfs_orphan_del(NULL, inode); | |
7b128766 JB |
3549 | } else { |
3550 | nr_unlink++; | |
3551 | } | |
3552 | ||
3553 | /* this will do delete_inode and everything for us */ | |
3554 | iput(inode); | |
66b4ffd1 JB |
3555 | if (ret) |
3556 | goto out; | |
7b128766 | 3557 | } |
3254c876 MX |
3558 | /* release the path since we're done with it */ |
3559 | btrfs_release_path(path); | |
3560 | ||
d68fc57b YZ |
3561 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
3562 | ||
3563 | if (root->orphan_block_rsv) | |
2ff7e61e | 3564 | btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, |
d68fc57b YZ |
3565 | (u64)-1); |
3566 | ||
27cdeb70 MX |
3567 | if (root->orphan_block_rsv || |
3568 | test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) { | |
7a7eaa40 | 3569 | trans = btrfs_join_transaction(root); |
66b4ffd1 | 3570 | if (!IS_ERR(trans)) |
3a45bb20 | 3571 | btrfs_end_transaction(trans); |
d68fc57b | 3572 | } |
7b128766 JB |
3573 | |
3574 | if (nr_unlink) | |
0b246afa | 3575 | btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink); |
7b128766 | 3576 | if (nr_truncate) |
0b246afa | 3577 | btrfs_debug(fs_info, "truncated %d orphans", nr_truncate); |
66b4ffd1 JB |
3578 | |
3579 | out: | |
3580 | if (ret) | |
0b246afa | 3581 | btrfs_err(fs_info, "could not do orphan cleanup %d", ret); |
66b4ffd1 JB |
3582 | btrfs_free_path(path); |
3583 | return ret; | |
7b128766 JB |
3584 | } |
3585 | ||
46a53cca CM |
3586 | /* |
3587 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
3588 | * don't find any xattrs, we know there can't be any acls. | |
3589 | * | |
3590 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
3591 | */ | |
3592 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
63541927 FDBM |
3593 | int slot, u64 objectid, |
3594 | int *first_xattr_slot) | |
46a53cca CM |
3595 | { |
3596 | u32 nritems = btrfs_header_nritems(leaf); | |
3597 | struct btrfs_key found_key; | |
f23b5a59 JB |
3598 | static u64 xattr_access = 0; |
3599 | static u64 xattr_default = 0; | |
46a53cca CM |
3600 | int scanned = 0; |
3601 | ||
f23b5a59 | 3602 | if (!xattr_access) { |
97d79299 AG |
3603 | xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS, |
3604 | strlen(XATTR_NAME_POSIX_ACL_ACCESS)); | |
3605 | xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT, | |
3606 | strlen(XATTR_NAME_POSIX_ACL_DEFAULT)); | |
f23b5a59 JB |
3607 | } |
3608 | ||
46a53cca | 3609 | slot++; |
63541927 | 3610 | *first_xattr_slot = -1; |
46a53cca CM |
3611 | while (slot < nritems) { |
3612 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3613 | ||
3614 | /* we found a different objectid, there must not be acls */ | |
3615 | if (found_key.objectid != objectid) | |
3616 | return 0; | |
3617 | ||
3618 | /* we found an xattr, assume we've got an acl */ | |
f23b5a59 | 3619 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) { |
63541927 FDBM |
3620 | if (*first_xattr_slot == -1) |
3621 | *first_xattr_slot = slot; | |
f23b5a59 JB |
3622 | if (found_key.offset == xattr_access || |
3623 | found_key.offset == xattr_default) | |
3624 | return 1; | |
3625 | } | |
46a53cca CM |
3626 | |
3627 | /* | |
3628 | * we found a key greater than an xattr key, there can't | |
3629 | * be any acls later on | |
3630 | */ | |
3631 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
3632 | return 0; | |
3633 | ||
3634 | slot++; | |
3635 | scanned++; | |
3636 | ||
3637 | /* | |
3638 | * it goes inode, inode backrefs, xattrs, extents, | |
3639 | * so if there are a ton of hard links to an inode there can | |
3640 | * be a lot of backrefs. Don't waste time searching too hard, | |
3641 | * this is just an optimization | |
3642 | */ | |
3643 | if (scanned >= 8) | |
3644 | break; | |
3645 | } | |
3646 | /* we hit the end of the leaf before we found an xattr or | |
3647 | * something larger than an xattr. We have to assume the inode | |
3648 | * has acls | |
3649 | */ | |
63541927 FDBM |
3650 | if (*first_xattr_slot == -1) |
3651 | *first_xattr_slot = slot; | |
46a53cca CM |
3652 | return 1; |
3653 | } | |
3654 | ||
d352ac68 CM |
3655 | /* |
3656 | * read an inode from the btree into the in-memory inode | |
3657 | */ | |
67710892 | 3658 | static int btrfs_read_locked_inode(struct inode *inode) |
39279cc3 | 3659 | { |
0b246afa | 3660 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 3661 | struct btrfs_path *path; |
5f39d397 | 3662 | struct extent_buffer *leaf; |
39279cc3 CM |
3663 | struct btrfs_inode_item *inode_item; |
3664 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3665 | struct btrfs_key location; | |
67de1176 | 3666 | unsigned long ptr; |
46a53cca | 3667 | int maybe_acls; |
618e21d5 | 3668 | u32 rdev; |
39279cc3 | 3669 | int ret; |
2f7e33d4 | 3670 | bool filled = false; |
63541927 | 3671 | int first_xattr_slot; |
2f7e33d4 MX |
3672 | |
3673 | ret = btrfs_fill_inode(inode, &rdev); | |
3674 | if (!ret) | |
3675 | filled = true; | |
39279cc3 CM |
3676 | |
3677 | path = btrfs_alloc_path(); | |
67710892 FM |
3678 | if (!path) { |
3679 | ret = -ENOMEM; | |
1748f843 | 3680 | goto make_bad; |
67710892 | 3681 | } |
1748f843 | 3682 | |
39279cc3 | 3683 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 3684 | |
39279cc3 | 3685 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
67710892 FM |
3686 | if (ret) { |
3687 | if (ret > 0) | |
3688 | ret = -ENOENT; | |
39279cc3 | 3689 | goto make_bad; |
67710892 | 3690 | } |
39279cc3 | 3691 | |
5f39d397 | 3692 | leaf = path->nodes[0]; |
2f7e33d4 MX |
3693 | |
3694 | if (filled) | |
67de1176 | 3695 | goto cache_index; |
2f7e33d4 | 3696 | |
5f39d397 CM |
3697 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
3698 | struct btrfs_inode_item); | |
5f39d397 | 3699 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 3700 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
3701 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
3702 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
dbe674a9 | 3703 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 | 3704 | |
a937b979 DS |
3705 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime); |
3706 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime); | |
5f39d397 | 3707 | |
a937b979 DS |
3708 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime); |
3709 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime); | |
5f39d397 | 3710 | |
a937b979 DS |
3711 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime); |
3712 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime); | |
5f39d397 | 3713 | |
9cc97d64 | 3714 | BTRFS_I(inode)->i_otime.tv_sec = |
3715 | btrfs_timespec_sec(leaf, &inode_item->otime); | |
3716 | BTRFS_I(inode)->i_otime.tv_nsec = | |
3717 | btrfs_timespec_nsec(leaf, &inode_item->otime); | |
5f39d397 | 3718 | |
a76a3cd4 | 3719 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 3720 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
3721 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
3722 | ||
6e17d30b YD |
3723 | inode->i_version = btrfs_inode_sequence(leaf, inode_item); |
3724 | inode->i_generation = BTRFS_I(inode)->generation; | |
3725 | inode->i_rdev = 0; | |
3726 | rdev = btrfs_inode_rdev(leaf, inode_item); | |
3727 | ||
3728 | BTRFS_I(inode)->index_cnt = (u64)-1; | |
3729 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); | |
3730 | ||
3731 | cache_index: | |
5dc562c5 JB |
3732 | /* |
3733 | * If we were modified in the current generation and evicted from memory | |
3734 | * and then re-read we need to do a full sync since we don't have any | |
3735 | * idea about which extents were modified before we were evicted from | |
3736 | * cache. | |
6e17d30b YD |
3737 | * |
3738 | * This is required for both inode re-read from disk and delayed inode | |
3739 | * in delayed_nodes_tree. | |
5dc562c5 | 3740 | */ |
0b246afa | 3741 | if (BTRFS_I(inode)->last_trans == fs_info->generation) |
5dc562c5 JB |
3742 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
3743 | &BTRFS_I(inode)->runtime_flags); | |
3744 | ||
bde6c242 FM |
3745 | /* |
3746 | * We don't persist the id of the transaction where an unlink operation | |
3747 | * against the inode was last made. So here we assume the inode might | |
3748 | * have been evicted, and therefore the exact value of last_unlink_trans | |
3749 | * lost, and set it to last_trans to avoid metadata inconsistencies | |
3750 | * between the inode and its parent if the inode is fsync'ed and the log | |
3751 | * replayed. For example, in the scenario: | |
3752 | * | |
3753 | * touch mydir/foo | |
3754 | * ln mydir/foo mydir/bar | |
3755 | * sync | |
3756 | * unlink mydir/bar | |
3757 | * echo 2 > /proc/sys/vm/drop_caches # evicts inode | |
3758 | * xfs_io -c fsync mydir/foo | |
3759 | * <power failure> | |
3760 | * mount fs, triggers fsync log replay | |
3761 | * | |
3762 | * We must make sure that when we fsync our inode foo we also log its | |
3763 | * parent inode, otherwise after log replay the parent still has the | |
3764 | * dentry with the "bar" name but our inode foo has a link count of 1 | |
3765 | * and doesn't have an inode ref with the name "bar" anymore. | |
3766 | * | |
3767 | * Setting last_unlink_trans to last_trans is a pessimistic approach, | |
01327610 | 3768 | * but it guarantees correctness at the expense of occasional full |
bde6c242 FM |
3769 | * transaction commits on fsync if our inode is a directory, or if our |
3770 | * inode is not a directory, logging its parent unnecessarily. | |
3771 | */ | |
3772 | BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans; | |
3773 | ||
67de1176 MX |
3774 | path->slots[0]++; |
3775 | if (inode->i_nlink != 1 || | |
3776 | path->slots[0] >= btrfs_header_nritems(leaf)) | |
3777 | goto cache_acl; | |
3778 | ||
3779 | btrfs_item_key_to_cpu(leaf, &location, path->slots[0]); | |
4a0cc7ca | 3780 | if (location.objectid != btrfs_ino(BTRFS_I(inode))) |
67de1176 MX |
3781 | goto cache_acl; |
3782 | ||
3783 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3784 | if (location.type == BTRFS_INODE_REF_KEY) { | |
3785 | struct btrfs_inode_ref *ref; | |
3786 | ||
3787 | ref = (struct btrfs_inode_ref *)ptr; | |
3788 | BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref); | |
3789 | } else if (location.type == BTRFS_INODE_EXTREF_KEY) { | |
3790 | struct btrfs_inode_extref *extref; | |
3791 | ||
3792 | extref = (struct btrfs_inode_extref *)ptr; | |
3793 | BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf, | |
3794 | extref); | |
3795 | } | |
2f7e33d4 | 3796 | cache_acl: |
46a53cca CM |
3797 | /* |
3798 | * try to precache a NULL acl entry for files that don't have | |
3799 | * any xattrs or acls | |
3800 | */ | |
33345d01 | 3801 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
f85b7379 | 3802 | btrfs_ino(BTRFS_I(inode)), &first_xattr_slot); |
63541927 FDBM |
3803 | if (first_xattr_slot != -1) { |
3804 | path->slots[0] = first_xattr_slot; | |
3805 | ret = btrfs_load_inode_props(inode, path); | |
3806 | if (ret) | |
0b246afa | 3807 | btrfs_err(fs_info, |
351fd353 | 3808 | "error loading props for ino %llu (root %llu): %d", |
4a0cc7ca | 3809 | btrfs_ino(BTRFS_I(inode)), |
63541927 FDBM |
3810 | root->root_key.objectid, ret); |
3811 | } | |
3812 | btrfs_free_path(path); | |
3813 | ||
72c04902 AV |
3814 | if (!maybe_acls) |
3815 | cache_no_acl(inode); | |
46a53cca | 3816 | |
39279cc3 | 3817 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
3818 | case S_IFREG: |
3819 | inode->i_mapping->a_ops = &btrfs_aops; | |
d1310b2e | 3820 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
3821 | inode->i_fop = &btrfs_file_operations; |
3822 | inode->i_op = &btrfs_file_inode_operations; | |
3823 | break; | |
3824 | case S_IFDIR: | |
3825 | inode->i_fop = &btrfs_dir_file_operations; | |
67ade058 | 3826 | inode->i_op = &btrfs_dir_inode_operations; |
39279cc3 CM |
3827 | break; |
3828 | case S_IFLNK: | |
3829 | inode->i_op = &btrfs_symlink_inode_operations; | |
21fc61c7 | 3830 | inode_nohighmem(inode); |
39279cc3 CM |
3831 | inode->i_mapping->a_ops = &btrfs_symlink_aops; |
3832 | break; | |
618e21d5 | 3833 | default: |
0279b4cd | 3834 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
3835 | init_special_inode(inode, inode->i_mode, rdev); |
3836 | break; | |
39279cc3 | 3837 | } |
6cbff00f CH |
3838 | |
3839 | btrfs_update_iflags(inode); | |
67710892 | 3840 | return 0; |
39279cc3 CM |
3841 | |
3842 | make_bad: | |
39279cc3 | 3843 | btrfs_free_path(path); |
39279cc3 | 3844 | make_bad_inode(inode); |
67710892 | 3845 | return ret; |
39279cc3 CM |
3846 | } |
3847 | ||
d352ac68 CM |
3848 | /* |
3849 | * given a leaf and an inode, copy the inode fields into the leaf | |
3850 | */ | |
e02119d5 CM |
3851 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3852 | struct extent_buffer *leaf, | |
5f39d397 | 3853 | struct btrfs_inode_item *item, |
39279cc3 CM |
3854 | struct inode *inode) |
3855 | { | |
51fab693 LB |
3856 | struct btrfs_map_token token; |
3857 | ||
3858 | btrfs_init_map_token(&token); | |
5f39d397 | 3859 | |
51fab693 LB |
3860 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); |
3861 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3862 | btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size, | |
3863 | &token); | |
3864 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3865 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
5f39d397 | 3866 | |
a937b979 | 3867 | btrfs_set_token_timespec_sec(leaf, &item->atime, |
51fab693 | 3868 | inode->i_atime.tv_sec, &token); |
a937b979 | 3869 | btrfs_set_token_timespec_nsec(leaf, &item->atime, |
51fab693 | 3870 | inode->i_atime.tv_nsec, &token); |
5f39d397 | 3871 | |
a937b979 | 3872 | btrfs_set_token_timespec_sec(leaf, &item->mtime, |
51fab693 | 3873 | inode->i_mtime.tv_sec, &token); |
a937b979 | 3874 | btrfs_set_token_timespec_nsec(leaf, &item->mtime, |
51fab693 | 3875 | inode->i_mtime.tv_nsec, &token); |
5f39d397 | 3876 | |
a937b979 | 3877 | btrfs_set_token_timespec_sec(leaf, &item->ctime, |
51fab693 | 3878 | inode->i_ctime.tv_sec, &token); |
a937b979 | 3879 | btrfs_set_token_timespec_nsec(leaf, &item->ctime, |
51fab693 | 3880 | inode->i_ctime.tv_nsec, &token); |
5f39d397 | 3881 | |
9cc97d64 | 3882 | btrfs_set_token_timespec_sec(leaf, &item->otime, |
3883 | BTRFS_I(inode)->i_otime.tv_sec, &token); | |
3884 | btrfs_set_token_timespec_nsec(leaf, &item->otime, | |
3885 | BTRFS_I(inode)->i_otime.tv_nsec, &token); | |
3886 | ||
51fab693 LB |
3887 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), |
3888 | &token); | |
3889 | btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation, | |
3890 | &token); | |
3891 | btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token); | |
3892 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); | |
3893 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3894 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3895 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
39279cc3 CM |
3896 | } |
3897 | ||
d352ac68 CM |
3898 | /* |
3899 | * copy everything in the in-memory inode into the btree. | |
3900 | */ | |
2115133f | 3901 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 3902 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
3903 | { |
3904 | struct btrfs_inode_item *inode_item; | |
3905 | struct btrfs_path *path; | |
5f39d397 | 3906 | struct extent_buffer *leaf; |
39279cc3 CM |
3907 | int ret; |
3908 | ||
3909 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
3910 | if (!path) |
3911 | return -ENOMEM; | |
3912 | ||
b9473439 | 3913 | path->leave_spinning = 1; |
16cdcec7 MX |
3914 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
3915 | 1); | |
39279cc3 CM |
3916 | if (ret) { |
3917 | if (ret > 0) | |
3918 | ret = -ENOENT; | |
3919 | goto failed; | |
3920 | } | |
3921 | ||
5f39d397 CM |
3922 | leaf = path->nodes[0]; |
3923 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 3924 | struct btrfs_inode_item); |
39279cc3 | 3925 | |
e02119d5 | 3926 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 3927 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 3928 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
3929 | ret = 0; |
3930 | failed: | |
39279cc3 CM |
3931 | btrfs_free_path(path); |
3932 | return ret; | |
3933 | } | |
3934 | ||
2115133f CM |
3935 | /* |
3936 | * copy everything in the in-memory inode into the btree. | |
3937 | */ | |
3938 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
3939 | struct btrfs_root *root, struct inode *inode) | |
3940 | { | |
0b246afa | 3941 | struct btrfs_fs_info *fs_info = root->fs_info; |
2115133f CM |
3942 | int ret; |
3943 | ||
3944 | /* | |
3945 | * If the inode is a free space inode, we can deadlock during commit | |
3946 | * if we put it into the delayed code. | |
3947 | * | |
3948 | * The data relocation inode should also be directly updated | |
3949 | * without delay | |
3950 | */ | |
83eea1f1 | 3951 | if (!btrfs_is_free_space_inode(inode) |
1d52c78a | 3952 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
0b246afa | 3953 | && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) { |
8ea05e3a AB |
3954 | btrfs_update_root_times(trans, root); |
3955 | ||
2115133f CM |
3956 | ret = btrfs_delayed_update_inode(trans, root, inode); |
3957 | if (!ret) | |
3958 | btrfs_set_inode_last_trans(trans, inode); | |
3959 | return ret; | |
3960 | } | |
3961 | ||
3962 | return btrfs_update_inode_item(trans, root, inode); | |
3963 | } | |
3964 | ||
be6aef60 JB |
3965 | noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
3966 | struct btrfs_root *root, | |
3967 | struct inode *inode) | |
2115133f CM |
3968 | { |
3969 | int ret; | |
3970 | ||
3971 | ret = btrfs_update_inode(trans, root, inode); | |
3972 | if (ret == -ENOSPC) | |
3973 | return btrfs_update_inode_item(trans, root, inode); | |
3974 | return ret; | |
3975 | } | |
3976 | ||
d352ac68 CM |
3977 | /* |
3978 | * unlink helper that gets used here in inode.c and in the tree logging | |
3979 | * recovery code. It remove a link in a directory with a given name, and | |
3980 | * also drops the back refs in the inode to the directory | |
3981 | */ | |
92986796 AV |
3982 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3983 | struct btrfs_root *root, | |
4ec5934e NB |
3984 | struct btrfs_inode *dir, |
3985 | struct btrfs_inode *inode, | |
92986796 | 3986 | const char *name, int name_len) |
39279cc3 | 3987 | { |
0b246afa | 3988 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 3989 | struct btrfs_path *path; |
39279cc3 | 3990 | int ret = 0; |
5f39d397 | 3991 | struct extent_buffer *leaf; |
39279cc3 | 3992 | struct btrfs_dir_item *di; |
5f39d397 | 3993 | struct btrfs_key key; |
aec7477b | 3994 | u64 index; |
4ec5934e NB |
3995 | u64 ino = btrfs_ino(inode); |
3996 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
3997 | |
3998 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
3999 | if (!path) { |
4000 | ret = -ENOMEM; | |
554233a6 | 4001 | goto out; |
54aa1f4d CM |
4002 | } |
4003 | ||
b9473439 | 4004 | path->leave_spinning = 1; |
33345d01 | 4005 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 CM |
4006 | name, name_len, -1); |
4007 | if (IS_ERR(di)) { | |
4008 | ret = PTR_ERR(di); | |
4009 | goto err; | |
4010 | } | |
4011 | if (!di) { | |
4012 | ret = -ENOENT; | |
4013 | goto err; | |
4014 | } | |
5f39d397 CM |
4015 | leaf = path->nodes[0]; |
4016 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 4017 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
4018 | if (ret) |
4019 | goto err; | |
b3b4aa74 | 4020 | btrfs_release_path(path); |
39279cc3 | 4021 | |
67de1176 MX |
4022 | /* |
4023 | * If we don't have dir index, we have to get it by looking up | |
4024 | * the inode ref, since we get the inode ref, remove it directly, | |
4025 | * it is unnecessary to do delayed deletion. | |
4026 | * | |
4027 | * But if we have dir index, needn't search inode ref to get it. | |
4028 | * Since the inode ref is close to the inode item, it is better | |
4029 | * that we delay to delete it, and just do this deletion when | |
4030 | * we update the inode item. | |
4031 | */ | |
4ec5934e NB |
4032 | if (inode->dir_index) { |
4033 | ret = btrfs_delayed_delete_inode_ref(inode); | |
67de1176 | 4034 | if (!ret) { |
4ec5934e | 4035 | index = inode->dir_index; |
67de1176 MX |
4036 | goto skip_backref; |
4037 | } | |
4038 | } | |
4039 | ||
33345d01 LZ |
4040 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
4041 | dir_ino, &index); | |
aec7477b | 4042 | if (ret) { |
0b246afa | 4043 | btrfs_info(fs_info, |
c2cf52eb | 4044 | "failed to delete reference to %.*s, inode %llu parent %llu", |
c1c9ff7c | 4045 | name_len, name, ino, dir_ino); |
66642832 | 4046 | btrfs_abort_transaction(trans, ret); |
aec7477b JB |
4047 | goto err; |
4048 | } | |
67de1176 | 4049 | skip_backref: |
4ec5934e | 4050 | ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index); |
79787eaa | 4051 | if (ret) { |
66642832 | 4052 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 4053 | goto err; |
79787eaa | 4054 | } |
39279cc3 | 4055 | |
4ec5934e NB |
4056 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode, |
4057 | dir_ino); | |
79787eaa | 4058 | if (ret != 0 && ret != -ENOENT) { |
66642832 | 4059 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4060 | goto err; |
4061 | } | |
e02119d5 | 4062 | |
4ec5934e NB |
4063 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir, |
4064 | index); | |
6418c961 CM |
4065 | if (ret == -ENOENT) |
4066 | ret = 0; | |
d4e3991b | 4067 | else if (ret) |
66642832 | 4068 | btrfs_abort_transaction(trans, ret); |
39279cc3 CM |
4069 | err: |
4070 | btrfs_free_path(path); | |
e02119d5 CM |
4071 | if (ret) |
4072 | goto out; | |
4073 | ||
4ec5934e NB |
4074 | btrfs_i_size_write(&dir->vfs_inode, |
4075 | dir->vfs_inode.i_size - name_len * 2); | |
4076 | inode_inc_iversion(&inode->vfs_inode); | |
4077 | inode_inc_iversion(&dir->vfs_inode); | |
4078 | inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime = | |
4079 | dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode); | |
4080 | ret = btrfs_update_inode(trans, root, &dir->vfs_inode); | |
e02119d5 | 4081 | out: |
39279cc3 CM |
4082 | return ret; |
4083 | } | |
4084 | ||
92986796 AV |
4085 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
4086 | struct btrfs_root *root, | |
4ec5934e | 4087 | struct btrfs_inode *dir, struct btrfs_inode *inode, |
92986796 AV |
4088 | const char *name, int name_len) |
4089 | { | |
4090 | int ret; | |
4091 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
4092 | if (!ret) { | |
4ec5934e NB |
4093 | drop_nlink(&inode->vfs_inode); |
4094 | ret = btrfs_update_inode(trans, root, &inode->vfs_inode); | |
92986796 AV |
4095 | } |
4096 | return ret; | |
4097 | } | |
39279cc3 | 4098 | |
a22285a6 YZ |
4099 | /* |
4100 | * helper to start transaction for unlink and rmdir. | |
4101 | * | |
d52be818 JB |
4102 | * unlink and rmdir are special in btrfs, they do not always free space, so |
4103 | * if we cannot make our reservations the normal way try and see if there is | |
4104 | * plenty of slack room in the global reserve to migrate, otherwise we cannot | |
4105 | * allow the unlink to occur. | |
a22285a6 | 4106 | */ |
d52be818 | 4107 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir) |
4df27c4d | 4108 | { |
a22285a6 | 4109 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 4110 | |
e70bea5f JB |
4111 | /* |
4112 | * 1 for the possible orphan item | |
4113 | * 1 for the dir item | |
4114 | * 1 for the dir index | |
4115 | * 1 for the inode ref | |
e70bea5f JB |
4116 | * 1 for the inode |
4117 | */ | |
8eab77ff | 4118 | return btrfs_start_transaction_fallback_global_rsv(root, 5, 5); |
a22285a6 YZ |
4119 | } |
4120 | ||
4121 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
4122 | { | |
4123 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4124 | struct btrfs_trans_handle *trans; | |
2b0143b5 | 4125 | struct inode *inode = d_inode(dentry); |
a22285a6 | 4126 | int ret; |
a22285a6 | 4127 | |
d52be818 | 4128 | trans = __unlink_start_trans(dir); |
a22285a6 YZ |
4129 | if (IS_ERR(trans)) |
4130 | return PTR_ERR(trans); | |
5f39d397 | 4131 | |
4ec5934e NB |
4132 | btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)), |
4133 | 0); | |
12fcfd22 | 4134 | |
4ec5934e NB |
4135 | ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
4136 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, | |
4137 | dentry->d_name.len); | |
b532402e TI |
4138 | if (ret) |
4139 | goto out; | |
7b128766 | 4140 | |
a22285a6 | 4141 | if (inode->i_nlink == 0) { |
7b128766 | 4142 | ret = btrfs_orphan_add(trans, inode); |
b532402e TI |
4143 | if (ret) |
4144 | goto out; | |
a22285a6 | 4145 | } |
7b128766 | 4146 | |
b532402e | 4147 | out: |
3a45bb20 | 4148 | btrfs_end_transaction(trans); |
2ff7e61e | 4149 | btrfs_btree_balance_dirty(root->fs_info); |
39279cc3 CM |
4150 | return ret; |
4151 | } | |
4152 | ||
4df27c4d YZ |
4153 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
4154 | struct btrfs_root *root, | |
4155 | struct inode *dir, u64 objectid, | |
4156 | const char *name, int name_len) | |
4157 | { | |
0b246afa | 4158 | struct btrfs_fs_info *fs_info = root->fs_info; |
4df27c4d YZ |
4159 | struct btrfs_path *path; |
4160 | struct extent_buffer *leaf; | |
4161 | struct btrfs_dir_item *di; | |
4162 | struct btrfs_key key; | |
4163 | u64 index; | |
4164 | int ret; | |
4a0cc7ca | 4165 | u64 dir_ino = btrfs_ino(BTRFS_I(dir)); |
4df27c4d YZ |
4166 | |
4167 | path = btrfs_alloc_path(); | |
4168 | if (!path) | |
4169 | return -ENOMEM; | |
4170 | ||
33345d01 | 4171 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 4172 | name, name_len, -1); |
79787eaa JM |
4173 | if (IS_ERR_OR_NULL(di)) { |
4174 | if (!di) | |
4175 | ret = -ENOENT; | |
4176 | else | |
4177 | ret = PTR_ERR(di); | |
4178 | goto out; | |
4179 | } | |
4df27c4d YZ |
4180 | |
4181 | leaf = path->nodes[0]; | |
4182 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
4183 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
4184 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa | 4185 | if (ret) { |
66642832 | 4186 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4187 | goto out; |
4188 | } | |
b3b4aa74 | 4189 | btrfs_release_path(path); |
4df27c4d | 4190 | |
0b246afa JM |
4191 | ret = btrfs_del_root_ref(trans, fs_info, objectid, |
4192 | root->root_key.objectid, dir_ino, | |
4193 | &index, name, name_len); | |
4df27c4d | 4194 | if (ret < 0) { |
79787eaa | 4195 | if (ret != -ENOENT) { |
66642832 | 4196 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4197 | goto out; |
4198 | } | |
33345d01 | 4199 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 4200 | name, name_len); |
79787eaa JM |
4201 | if (IS_ERR_OR_NULL(di)) { |
4202 | if (!di) | |
4203 | ret = -ENOENT; | |
4204 | else | |
4205 | ret = PTR_ERR(di); | |
66642832 | 4206 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4207 | goto out; |
4208 | } | |
4df27c4d YZ |
4209 | |
4210 | leaf = path->nodes[0]; | |
4211 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 4212 | btrfs_release_path(path); |
4df27c4d YZ |
4213 | index = key.offset; |
4214 | } | |
945d8962 | 4215 | btrfs_release_path(path); |
4df27c4d | 4216 | |
e67bbbb9 | 4217 | ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index); |
79787eaa | 4218 | if (ret) { |
66642832 | 4219 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4220 | goto out; |
4221 | } | |
4df27c4d YZ |
4222 | |
4223 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 | 4224 | inode_inc_iversion(dir); |
c2050a45 | 4225 | dir->i_mtime = dir->i_ctime = current_time(dir); |
5a24e84c | 4226 | ret = btrfs_update_inode_fallback(trans, root, dir); |
79787eaa | 4227 | if (ret) |
66642832 | 4228 | btrfs_abort_transaction(trans, ret); |
79787eaa | 4229 | out: |
71d7aed0 | 4230 | btrfs_free_path(path); |
79787eaa | 4231 | return ret; |
4df27c4d YZ |
4232 | } |
4233 | ||
39279cc3 CM |
4234 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
4235 | { | |
2b0143b5 | 4236 | struct inode *inode = d_inode(dentry); |
1832a6d5 | 4237 | int err = 0; |
39279cc3 | 4238 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 4239 | struct btrfs_trans_handle *trans; |
44f714da | 4240 | u64 last_unlink_trans; |
39279cc3 | 4241 | |
b3ae244e | 4242 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 4243 | return -ENOTEMPTY; |
4a0cc7ca | 4244 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) |
b3ae244e | 4245 | return -EPERM; |
134d4512 | 4246 | |
d52be818 | 4247 | trans = __unlink_start_trans(dir); |
a22285a6 | 4248 | if (IS_ERR(trans)) |
5df6a9f6 | 4249 | return PTR_ERR(trans); |
5df6a9f6 | 4250 | |
4a0cc7ca | 4251 | if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
4df27c4d YZ |
4252 | err = btrfs_unlink_subvol(trans, root, dir, |
4253 | BTRFS_I(inode)->location.objectid, | |
4254 | dentry->d_name.name, | |
4255 | dentry->d_name.len); | |
4256 | goto out; | |
4257 | } | |
4258 | ||
7b128766 JB |
4259 | err = btrfs_orphan_add(trans, inode); |
4260 | if (err) | |
4df27c4d | 4261 | goto out; |
7b128766 | 4262 | |
44f714da FM |
4263 | last_unlink_trans = BTRFS_I(inode)->last_unlink_trans; |
4264 | ||
39279cc3 | 4265 | /* now the directory is empty */ |
4ec5934e NB |
4266 | err = btrfs_unlink_inode(trans, root, BTRFS_I(dir), |
4267 | BTRFS_I(d_inode(dentry)), dentry->d_name.name, | |
4268 | dentry->d_name.len); | |
44f714da | 4269 | if (!err) { |
dbe674a9 | 4270 | btrfs_i_size_write(inode, 0); |
44f714da FM |
4271 | /* |
4272 | * Propagate the last_unlink_trans value of the deleted dir to | |
4273 | * its parent directory. This is to prevent an unrecoverable | |
4274 | * log tree in the case we do something like this: | |
4275 | * 1) create dir foo | |
4276 | * 2) create snapshot under dir foo | |
4277 | * 3) delete the snapshot | |
4278 | * 4) rmdir foo | |
4279 | * 5) mkdir foo | |
4280 | * 6) fsync foo or some file inside foo | |
4281 | */ | |
4282 | if (last_unlink_trans >= trans->transid) | |
4283 | BTRFS_I(dir)->last_unlink_trans = last_unlink_trans; | |
4284 | } | |
4df27c4d | 4285 | out: |
3a45bb20 | 4286 | btrfs_end_transaction(trans); |
2ff7e61e | 4287 | btrfs_btree_balance_dirty(root->fs_info); |
3954401f | 4288 | |
39279cc3 CM |
4289 | return err; |
4290 | } | |
4291 | ||
28f75a0e CM |
4292 | static int truncate_space_check(struct btrfs_trans_handle *trans, |
4293 | struct btrfs_root *root, | |
4294 | u64 bytes_deleted) | |
4295 | { | |
0b246afa | 4296 | struct btrfs_fs_info *fs_info = root->fs_info; |
28f75a0e CM |
4297 | int ret; |
4298 | ||
dc95f7bf JB |
4299 | /* |
4300 | * This is only used to apply pressure to the enospc system, we don't | |
4301 | * intend to use this reservation at all. | |
4302 | */ | |
2ff7e61e | 4303 | bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted); |
0b246afa JM |
4304 | bytes_deleted *= fs_info->nodesize; |
4305 | ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv, | |
28f75a0e | 4306 | bytes_deleted, BTRFS_RESERVE_NO_FLUSH); |
dc95f7bf | 4307 | if (!ret) { |
0b246afa | 4308 | trace_btrfs_space_reservation(fs_info, "transaction", |
dc95f7bf JB |
4309 | trans->transid, |
4310 | bytes_deleted, 1); | |
28f75a0e | 4311 | trans->bytes_reserved += bytes_deleted; |
dc95f7bf | 4312 | } |
28f75a0e CM |
4313 | return ret; |
4314 | ||
4315 | } | |
4316 | ||
0305cd5f FM |
4317 | static int truncate_inline_extent(struct inode *inode, |
4318 | struct btrfs_path *path, | |
4319 | struct btrfs_key *found_key, | |
4320 | const u64 item_end, | |
4321 | const u64 new_size) | |
4322 | { | |
4323 | struct extent_buffer *leaf = path->nodes[0]; | |
4324 | int slot = path->slots[0]; | |
4325 | struct btrfs_file_extent_item *fi; | |
4326 | u32 size = (u32)(new_size - found_key->offset); | |
4327 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4328 | ||
4329 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
4330 | ||
4331 | if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) { | |
4332 | loff_t offset = new_size; | |
09cbfeaf | 4333 | loff_t page_end = ALIGN(offset, PAGE_SIZE); |
0305cd5f FM |
4334 | |
4335 | /* | |
4336 | * Zero out the remaining of the last page of our inline extent, | |
4337 | * instead of directly truncating our inline extent here - that | |
4338 | * would be much more complex (decompressing all the data, then | |
4339 | * compressing the truncated data, which might be bigger than | |
4340 | * the size of the inline extent, resize the extent, etc). | |
4341 | * We release the path because to get the page we might need to | |
4342 | * read the extent item from disk (data not in the page cache). | |
4343 | */ | |
4344 | btrfs_release_path(path); | |
9703fefe CR |
4345 | return btrfs_truncate_block(inode, offset, page_end - offset, |
4346 | 0); | |
0305cd5f FM |
4347 | } |
4348 | ||
4349 | btrfs_set_file_extent_ram_bytes(leaf, fi, size); | |
4350 | size = btrfs_file_extent_calc_inline_size(size); | |
2ff7e61e | 4351 | btrfs_truncate_item(root->fs_info, path, size, 1); |
0305cd5f FM |
4352 | |
4353 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) | |
4354 | inode_sub_bytes(inode, item_end + 1 - new_size); | |
4355 | ||
4356 | return 0; | |
4357 | } | |
4358 | ||
39279cc3 CM |
4359 | /* |
4360 | * this can truncate away extent items, csum items and directory items. | |
4361 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 4362 | * any higher than new_size |
39279cc3 CM |
4363 | * |
4364 | * csum items that cross the new i_size are truncated to the new size | |
4365 | * as well. | |
7b128766 JB |
4366 | * |
4367 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
4368 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 4369 | */ |
8082510e YZ |
4370 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
4371 | struct btrfs_root *root, | |
4372 | struct inode *inode, | |
4373 | u64 new_size, u32 min_type) | |
39279cc3 | 4374 | { |
0b246afa | 4375 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 4376 | struct btrfs_path *path; |
5f39d397 | 4377 | struct extent_buffer *leaf; |
39279cc3 | 4378 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
4379 | struct btrfs_key key; |
4380 | struct btrfs_key found_key; | |
39279cc3 | 4381 | u64 extent_start = 0; |
db94535d | 4382 | u64 extent_num_bytes = 0; |
5d4f98a2 | 4383 | u64 extent_offset = 0; |
39279cc3 | 4384 | u64 item_end = 0; |
c1aa4575 | 4385 | u64 last_size = new_size; |
8082510e | 4386 | u32 found_type = (u8)-1; |
39279cc3 CM |
4387 | int found_extent; |
4388 | int del_item; | |
85e21bac CM |
4389 | int pending_del_nr = 0; |
4390 | int pending_del_slot = 0; | |
179e29e4 | 4391 | int extent_type = -1; |
8082510e YZ |
4392 | int ret; |
4393 | int err = 0; | |
4a0cc7ca | 4394 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
28ed1345 | 4395 | u64 bytes_deleted = 0; |
1262133b JB |
4396 | bool be_nice = 0; |
4397 | bool should_throttle = 0; | |
28f75a0e | 4398 | bool should_end = 0; |
8082510e YZ |
4399 | |
4400 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 4401 | |
28ed1345 CM |
4402 | /* |
4403 | * for non-free space inodes and ref cows, we want to back off from | |
4404 | * time to time | |
4405 | */ | |
4406 | if (!btrfs_is_free_space_inode(inode) && | |
4407 | test_bit(BTRFS_ROOT_REF_COWS, &root->state)) | |
4408 | be_nice = 1; | |
4409 | ||
0eb0e19c MF |
4410 | path = btrfs_alloc_path(); |
4411 | if (!path) | |
4412 | return -ENOMEM; | |
e4058b54 | 4413 | path->reada = READA_BACK; |
0eb0e19c | 4414 | |
5dc562c5 JB |
4415 | /* |
4416 | * We want to drop from the next block forward in case this new size is | |
4417 | * not block aligned since we will be keeping the last block of the | |
4418 | * extent just the way it is. | |
4419 | */ | |
27cdeb70 | 4420 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
0b246afa | 4421 | root == fs_info->tree_root) |
fda2832f | 4422 | btrfs_drop_extent_cache(inode, ALIGN(new_size, |
0b246afa | 4423 | fs_info->sectorsize), |
da17066c | 4424 | (u64)-1, 0); |
8082510e | 4425 | |
16cdcec7 MX |
4426 | /* |
4427 | * This function is also used to drop the items in the log tree before | |
4428 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
4429 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
4430 | * items. | |
4431 | */ | |
4432 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
4ccb5c72 | 4433 | btrfs_kill_delayed_inode_items(BTRFS_I(inode)); |
16cdcec7 | 4434 | |
33345d01 | 4435 | key.objectid = ino; |
39279cc3 | 4436 | key.offset = (u64)-1; |
5f39d397 CM |
4437 | key.type = (u8)-1; |
4438 | ||
85e21bac | 4439 | search_again: |
28ed1345 CM |
4440 | /* |
4441 | * with a 16K leaf size and 128MB extents, you can actually queue | |
4442 | * up a huge file in a single leaf. Most of the time that | |
4443 | * bytes_deleted is > 0, it will be huge by the time we get here | |
4444 | */ | |
ee22184b | 4445 | if (be_nice && bytes_deleted > SZ_32M) { |
3a45bb20 | 4446 | if (btrfs_should_end_transaction(trans)) { |
28ed1345 CM |
4447 | err = -EAGAIN; |
4448 | goto error; | |
4449 | } | |
4450 | } | |
4451 | ||
4452 | ||
b9473439 | 4453 | path->leave_spinning = 1; |
85e21bac | 4454 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
4455 | if (ret < 0) { |
4456 | err = ret; | |
4457 | goto out; | |
4458 | } | |
d397712b | 4459 | |
85e21bac | 4460 | if (ret > 0) { |
e02119d5 CM |
4461 | /* there are no items in the tree for us to truncate, we're |
4462 | * done | |
4463 | */ | |
8082510e YZ |
4464 | if (path->slots[0] == 0) |
4465 | goto out; | |
85e21bac CM |
4466 | path->slots[0]--; |
4467 | } | |
4468 | ||
d397712b | 4469 | while (1) { |
39279cc3 | 4470 | fi = NULL; |
5f39d397 CM |
4471 | leaf = path->nodes[0]; |
4472 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
962a298f | 4473 | found_type = found_key.type; |
39279cc3 | 4474 | |
33345d01 | 4475 | if (found_key.objectid != ino) |
39279cc3 | 4476 | break; |
5f39d397 | 4477 | |
85e21bac | 4478 | if (found_type < min_type) |
39279cc3 CM |
4479 | break; |
4480 | ||
5f39d397 | 4481 | item_end = found_key.offset; |
39279cc3 | 4482 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 4483 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 4484 | struct btrfs_file_extent_item); |
179e29e4 CM |
4485 | extent_type = btrfs_file_extent_type(leaf, fi); |
4486 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4487 | item_end += |
db94535d | 4488 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 4489 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 4490 | item_end += btrfs_file_extent_inline_len(leaf, |
514ac8ad | 4491 | path->slots[0], fi); |
39279cc3 | 4492 | } |
008630c1 | 4493 | item_end--; |
39279cc3 | 4494 | } |
8082510e YZ |
4495 | if (found_type > min_type) { |
4496 | del_item = 1; | |
4497 | } else { | |
91298eec LB |
4498 | if (item_end < new_size) { |
4499 | /* | |
4500 | * With NO_HOLES mode, for the following mapping | |
4501 | * | |
4502 | * [0-4k][hole][8k-12k] | |
4503 | * | |
4504 | * if truncating isize down to 6k, it ends up | |
4505 | * isize being 8k. | |
4506 | */ | |
4507 | if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) | |
4508 | last_size = new_size; | |
b888db2b | 4509 | break; |
91298eec | 4510 | } |
8082510e YZ |
4511 | if (found_key.offset >= new_size) |
4512 | del_item = 1; | |
4513 | else | |
4514 | del_item = 0; | |
39279cc3 | 4515 | } |
39279cc3 | 4516 | found_extent = 0; |
39279cc3 | 4517 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
4518 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
4519 | goto delete; | |
4520 | ||
7f4f6e0a JB |
4521 | if (del_item) |
4522 | last_size = found_key.offset; | |
4523 | else | |
4524 | last_size = new_size; | |
4525 | ||
179e29e4 | 4526 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
39279cc3 | 4527 | u64 num_dec; |
db94535d | 4528 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 4529 | if (!del_item) { |
db94535d CM |
4530 | u64 orig_num_bytes = |
4531 | btrfs_file_extent_num_bytes(leaf, fi); | |
fda2832f QW |
4532 | extent_num_bytes = ALIGN(new_size - |
4533 | found_key.offset, | |
0b246afa | 4534 | fs_info->sectorsize); |
db94535d CM |
4535 | btrfs_set_file_extent_num_bytes(leaf, fi, |
4536 | extent_num_bytes); | |
4537 | num_dec = (orig_num_bytes - | |
9069218d | 4538 | extent_num_bytes); |
27cdeb70 MX |
4539 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4540 | &root->state) && | |
4541 | extent_start != 0) | |
a76a3cd4 | 4542 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 4543 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 4544 | } else { |
db94535d CM |
4545 | extent_num_bytes = |
4546 | btrfs_file_extent_disk_num_bytes(leaf, | |
4547 | fi); | |
5d4f98a2 YZ |
4548 | extent_offset = found_key.offset - |
4549 | btrfs_file_extent_offset(leaf, fi); | |
4550 | ||
39279cc3 | 4551 | /* FIXME blocksize != 4096 */ |
9069218d | 4552 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
4553 | if (extent_start != 0) { |
4554 | found_extent = 1; | |
27cdeb70 MX |
4555 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4556 | &root->state)) | |
a76a3cd4 | 4557 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 4558 | } |
39279cc3 | 4559 | } |
9069218d | 4560 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
4561 | /* |
4562 | * we can't truncate inline items that have had | |
4563 | * special encodings | |
4564 | */ | |
4565 | if (!del_item && | |
c8b97818 CM |
4566 | btrfs_file_extent_encryption(leaf, fi) == 0 && |
4567 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
514ac8ad CM |
4568 | |
4569 | /* | |
0305cd5f FM |
4570 | * Need to release path in order to truncate a |
4571 | * compressed extent. So delete any accumulated | |
4572 | * extent items so far. | |
514ac8ad | 4573 | */ |
0305cd5f FM |
4574 | if (btrfs_file_extent_compression(leaf, fi) != |
4575 | BTRFS_COMPRESS_NONE && pending_del_nr) { | |
4576 | err = btrfs_del_items(trans, root, path, | |
4577 | pending_del_slot, | |
4578 | pending_del_nr); | |
4579 | if (err) { | |
4580 | btrfs_abort_transaction(trans, | |
0305cd5f FM |
4581 | err); |
4582 | goto error; | |
4583 | } | |
4584 | pending_del_nr = 0; | |
4585 | } | |
4586 | ||
4587 | err = truncate_inline_extent(inode, path, | |
4588 | &found_key, | |
4589 | item_end, | |
4590 | new_size); | |
4591 | if (err) { | |
66642832 | 4592 | btrfs_abort_transaction(trans, err); |
0305cd5f FM |
4593 | goto error; |
4594 | } | |
27cdeb70 MX |
4595 | } else if (test_bit(BTRFS_ROOT_REF_COWS, |
4596 | &root->state)) { | |
0305cd5f | 4597 | inode_sub_bytes(inode, item_end + 1 - new_size); |
9069218d | 4598 | } |
39279cc3 | 4599 | } |
179e29e4 | 4600 | delete: |
39279cc3 | 4601 | if (del_item) { |
85e21bac CM |
4602 | if (!pending_del_nr) { |
4603 | /* no pending yet, add ourselves */ | |
4604 | pending_del_slot = path->slots[0]; | |
4605 | pending_del_nr = 1; | |
4606 | } else if (pending_del_nr && | |
4607 | path->slots[0] + 1 == pending_del_slot) { | |
4608 | /* hop on the pending chunk */ | |
4609 | pending_del_nr++; | |
4610 | pending_del_slot = path->slots[0]; | |
4611 | } else { | |
d397712b | 4612 | BUG(); |
85e21bac | 4613 | } |
39279cc3 CM |
4614 | } else { |
4615 | break; | |
4616 | } | |
28f75a0e CM |
4617 | should_throttle = 0; |
4618 | ||
27cdeb70 MX |
4619 | if (found_extent && |
4620 | (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || | |
0b246afa | 4621 | root == fs_info->tree_root)) { |
b9473439 | 4622 | btrfs_set_path_blocking(path); |
28ed1345 | 4623 | bytes_deleted += extent_num_bytes; |
2ff7e61e | 4624 | ret = btrfs_free_extent(trans, fs_info, extent_start, |
5d4f98a2 YZ |
4625 | extent_num_bytes, 0, |
4626 | btrfs_header_owner(leaf), | |
b06c4bf5 | 4627 | ino, extent_offset); |
39279cc3 | 4628 | BUG_ON(ret); |
2ff7e61e JM |
4629 | if (btrfs_should_throttle_delayed_refs(trans, fs_info)) |
4630 | btrfs_async_run_delayed_refs(fs_info, | |
dd4b857a WX |
4631 | trans->delayed_ref_updates * 2, |
4632 | trans->transid, 0); | |
28f75a0e CM |
4633 | if (be_nice) { |
4634 | if (truncate_space_check(trans, root, | |
4635 | extent_num_bytes)) { | |
4636 | should_end = 1; | |
4637 | } | |
4638 | if (btrfs_should_throttle_delayed_refs(trans, | |
2ff7e61e | 4639 | fs_info)) |
28f75a0e | 4640 | should_throttle = 1; |
28f75a0e | 4641 | } |
39279cc3 | 4642 | } |
85e21bac | 4643 | |
8082510e YZ |
4644 | if (found_type == BTRFS_INODE_ITEM_KEY) |
4645 | break; | |
4646 | ||
4647 | if (path->slots[0] == 0 || | |
1262133b | 4648 | path->slots[0] != pending_del_slot || |
28f75a0e | 4649 | should_throttle || should_end) { |
8082510e YZ |
4650 | if (pending_del_nr) { |
4651 | ret = btrfs_del_items(trans, root, path, | |
4652 | pending_del_slot, | |
4653 | pending_del_nr); | |
79787eaa | 4654 | if (ret) { |
66642832 | 4655 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
4656 | goto error; |
4657 | } | |
8082510e YZ |
4658 | pending_del_nr = 0; |
4659 | } | |
b3b4aa74 | 4660 | btrfs_release_path(path); |
28f75a0e | 4661 | if (should_throttle) { |
1262133b JB |
4662 | unsigned long updates = trans->delayed_ref_updates; |
4663 | if (updates) { | |
4664 | trans->delayed_ref_updates = 0; | |
2ff7e61e JM |
4665 | ret = btrfs_run_delayed_refs(trans, |
4666 | fs_info, | |
4667 | updates * 2); | |
1262133b JB |
4668 | if (ret && !err) |
4669 | err = ret; | |
4670 | } | |
4671 | } | |
28f75a0e CM |
4672 | /* |
4673 | * if we failed to refill our space rsv, bail out | |
4674 | * and let the transaction restart | |
4675 | */ | |
4676 | if (should_end) { | |
4677 | err = -EAGAIN; | |
4678 | goto error; | |
4679 | } | |
85e21bac | 4680 | goto search_again; |
8082510e YZ |
4681 | } else { |
4682 | path->slots[0]--; | |
85e21bac | 4683 | } |
39279cc3 | 4684 | } |
8082510e | 4685 | out: |
85e21bac CM |
4686 | if (pending_del_nr) { |
4687 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
4688 | pending_del_nr); | |
79787eaa | 4689 | if (ret) |
66642832 | 4690 | btrfs_abort_transaction(trans, ret); |
85e21bac | 4691 | } |
79787eaa | 4692 | error: |
c1aa4575 | 4693 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) |
7f4f6e0a | 4694 | btrfs_ordered_update_i_size(inode, last_size, NULL); |
28ed1345 | 4695 | |
39279cc3 | 4696 | btrfs_free_path(path); |
28ed1345 | 4697 | |
ee22184b | 4698 | if (be_nice && bytes_deleted > SZ_32M) { |
28ed1345 CM |
4699 | unsigned long updates = trans->delayed_ref_updates; |
4700 | if (updates) { | |
4701 | trans->delayed_ref_updates = 0; | |
2ff7e61e JM |
4702 | ret = btrfs_run_delayed_refs(trans, fs_info, |
4703 | updates * 2); | |
28ed1345 CM |
4704 | if (ret && !err) |
4705 | err = ret; | |
4706 | } | |
4707 | } | |
8082510e | 4708 | return err; |
39279cc3 CM |
4709 | } |
4710 | ||
4711 | /* | |
9703fefe | 4712 | * btrfs_truncate_block - read, zero a chunk and write a block |
2aaa6655 JB |
4713 | * @inode - inode that we're zeroing |
4714 | * @from - the offset to start zeroing | |
4715 | * @len - the length to zero, 0 to zero the entire range respective to the | |
4716 | * offset | |
4717 | * @front - zero up to the offset instead of from the offset on | |
4718 | * | |
9703fefe | 4719 | * This will find the block for the "from" offset and cow the block and zero the |
2aaa6655 | 4720 | * part we want to zero. This is used with truncate and hole punching. |
39279cc3 | 4721 | */ |
9703fefe | 4722 | int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len, |
2aaa6655 | 4723 | int front) |
39279cc3 | 4724 | { |
0b246afa | 4725 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2aaa6655 | 4726 | struct address_space *mapping = inode->i_mapping; |
e6dcd2dc CM |
4727 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4728 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 4729 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 4730 | char *kaddr; |
0b246afa | 4731 | u32 blocksize = fs_info->sectorsize; |
09cbfeaf | 4732 | pgoff_t index = from >> PAGE_SHIFT; |
9703fefe | 4733 | unsigned offset = from & (blocksize - 1); |
39279cc3 | 4734 | struct page *page; |
3b16a4e3 | 4735 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 4736 | int ret = 0; |
9703fefe CR |
4737 | u64 block_start; |
4738 | u64 block_end; | |
39279cc3 | 4739 | |
2aaa6655 JB |
4740 | if ((offset & (blocksize - 1)) == 0 && |
4741 | (!len || ((len & (blocksize - 1)) == 0))) | |
39279cc3 | 4742 | goto out; |
9703fefe | 4743 | |
7cf5b976 | 4744 | ret = btrfs_delalloc_reserve_space(inode, |
9703fefe | 4745 | round_down(from, blocksize), blocksize); |
5d5e103a JB |
4746 | if (ret) |
4747 | goto out; | |
39279cc3 | 4748 | |
211c17f5 | 4749 | again: |
3b16a4e3 | 4750 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 4751 | if (!page) { |
7cf5b976 | 4752 | btrfs_delalloc_release_space(inode, |
9703fefe CR |
4753 | round_down(from, blocksize), |
4754 | blocksize); | |
ac6a2b36 | 4755 | ret = -ENOMEM; |
39279cc3 | 4756 | goto out; |
5d5e103a | 4757 | } |
e6dcd2dc | 4758 | |
9703fefe CR |
4759 | block_start = round_down(from, blocksize); |
4760 | block_end = block_start + blocksize - 1; | |
e6dcd2dc | 4761 | |
39279cc3 | 4762 | if (!PageUptodate(page)) { |
9ebefb18 | 4763 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 4764 | lock_page(page); |
211c17f5 CM |
4765 | if (page->mapping != mapping) { |
4766 | unlock_page(page); | |
09cbfeaf | 4767 | put_page(page); |
211c17f5 CM |
4768 | goto again; |
4769 | } | |
39279cc3 CM |
4770 | if (!PageUptodate(page)) { |
4771 | ret = -EIO; | |
89642229 | 4772 | goto out_unlock; |
39279cc3 CM |
4773 | } |
4774 | } | |
211c17f5 | 4775 | wait_on_page_writeback(page); |
e6dcd2dc | 4776 | |
9703fefe | 4777 | lock_extent_bits(io_tree, block_start, block_end, &cached_state); |
e6dcd2dc CM |
4778 | set_page_extent_mapped(page); |
4779 | ||
9703fefe | 4780 | ordered = btrfs_lookup_ordered_extent(inode, block_start); |
e6dcd2dc | 4781 | if (ordered) { |
9703fefe | 4782 | unlock_extent_cached(io_tree, block_start, block_end, |
2ac55d41 | 4783 | &cached_state, GFP_NOFS); |
e6dcd2dc | 4784 | unlock_page(page); |
09cbfeaf | 4785 | put_page(page); |
eb84ae03 | 4786 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
4787 | btrfs_put_ordered_extent(ordered); |
4788 | goto again; | |
4789 | } | |
4790 | ||
9703fefe | 4791 | clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end, |
9e8a4a8b LB |
4792 | EXTENT_DIRTY | EXTENT_DELALLOC | |
4793 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 4794 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 4795 | |
9703fefe | 4796 | ret = btrfs_set_extent_delalloc(inode, block_start, block_end, |
ba8b04c1 | 4797 | &cached_state, 0); |
9ed74f2d | 4798 | if (ret) { |
9703fefe | 4799 | unlock_extent_cached(io_tree, block_start, block_end, |
2ac55d41 | 4800 | &cached_state, GFP_NOFS); |
9ed74f2d JB |
4801 | goto out_unlock; |
4802 | } | |
4803 | ||
9703fefe | 4804 | if (offset != blocksize) { |
2aaa6655 | 4805 | if (!len) |
9703fefe | 4806 | len = blocksize - offset; |
e6dcd2dc | 4807 | kaddr = kmap(page); |
2aaa6655 | 4808 | if (front) |
9703fefe CR |
4809 | memset(kaddr + (block_start - page_offset(page)), |
4810 | 0, offset); | |
2aaa6655 | 4811 | else |
9703fefe CR |
4812 | memset(kaddr + (block_start - page_offset(page)) + offset, |
4813 | 0, len); | |
e6dcd2dc CM |
4814 | flush_dcache_page(page); |
4815 | kunmap(page); | |
4816 | } | |
247e743c | 4817 | ClearPageChecked(page); |
e6dcd2dc | 4818 | set_page_dirty(page); |
9703fefe | 4819 | unlock_extent_cached(io_tree, block_start, block_end, &cached_state, |
2ac55d41 | 4820 | GFP_NOFS); |
39279cc3 | 4821 | |
89642229 | 4822 | out_unlock: |
5d5e103a | 4823 | if (ret) |
9703fefe CR |
4824 | btrfs_delalloc_release_space(inode, block_start, |
4825 | blocksize); | |
39279cc3 | 4826 | unlock_page(page); |
09cbfeaf | 4827 | put_page(page); |
39279cc3 CM |
4828 | out: |
4829 | return ret; | |
4830 | } | |
4831 | ||
16e7549f JB |
4832 | static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode, |
4833 | u64 offset, u64 len) | |
4834 | { | |
0b246afa | 4835 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
16e7549f JB |
4836 | struct btrfs_trans_handle *trans; |
4837 | int ret; | |
4838 | ||
4839 | /* | |
4840 | * Still need to make sure the inode looks like it's been updated so | |
4841 | * that any holes get logged if we fsync. | |
4842 | */ | |
0b246afa JM |
4843 | if (btrfs_fs_incompat(fs_info, NO_HOLES)) { |
4844 | BTRFS_I(inode)->last_trans = fs_info->generation; | |
16e7549f JB |
4845 | BTRFS_I(inode)->last_sub_trans = root->log_transid; |
4846 | BTRFS_I(inode)->last_log_commit = root->last_log_commit; | |
4847 | return 0; | |
4848 | } | |
4849 | ||
4850 | /* | |
4851 | * 1 - for the one we're dropping | |
4852 | * 1 - for the one we're adding | |
4853 | * 1 - for updating the inode. | |
4854 | */ | |
4855 | trans = btrfs_start_transaction(root, 3); | |
4856 | if (IS_ERR(trans)) | |
4857 | return PTR_ERR(trans); | |
4858 | ||
4859 | ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1); | |
4860 | if (ret) { | |
66642832 | 4861 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4862 | btrfs_end_transaction(trans); |
16e7549f JB |
4863 | return ret; |
4864 | } | |
4865 | ||
f85b7379 DS |
4866 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)), |
4867 | offset, 0, 0, len, 0, len, 0, 0, 0); | |
16e7549f | 4868 | if (ret) |
66642832 | 4869 | btrfs_abort_transaction(trans, ret); |
16e7549f JB |
4870 | else |
4871 | btrfs_update_inode(trans, root, inode); | |
3a45bb20 | 4872 | btrfs_end_transaction(trans); |
16e7549f JB |
4873 | return ret; |
4874 | } | |
4875 | ||
695a0d0d JB |
4876 | /* |
4877 | * This function puts in dummy file extents for the area we're creating a hole | |
4878 | * for. So if we are truncating this file to a larger size we need to insert | |
4879 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
4880 | * the range between oldsize and size | |
4881 | */ | |
a41ad394 | 4882 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 4883 | { |
0b246afa | 4884 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
9036c102 YZ |
4885 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4886 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 4887 | struct extent_map *em = NULL; |
2ac55d41 | 4888 | struct extent_state *cached_state = NULL; |
5dc562c5 | 4889 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
0b246afa JM |
4890 | u64 hole_start = ALIGN(oldsize, fs_info->sectorsize); |
4891 | u64 block_end = ALIGN(size, fs_info->sectorsize); | |
9036c102 YZ |
4892 | u64 last_byte; |
4893 | u64 cur_offset; | |
4894 | u64 hole_size; | |
9ed74f2d | 4895 | int err = 0; |
39279cc3 | 4896 | |
a71754fc | 4897 | /* |
9703fefe CR |
4898 | * If our size started in the middle of a block we need to zero out the |
4899 | * rest of the block before we expand the i_size, otherwise we could | |
a71754fc JB |
4900 | * expose stale data. |
4901 | */ | |
9703fefe | 4902 | err = btrfs_truncate_block(inode, oldsize, 0, 0); |
a71754fc JB |
4903 | if (err) |
4904 | return err; | |
4905 | ||
9036c102 YZ |
4906 | if (size <= hole_start) |
4907 | return 0; | |
4908 | ||
9036c102 YZ |
4909 | while (1) { |
4910 | struct btrfs_ordered_extent *ordered; | |
fa7c1494 | 4911 | |
ff13db41 | 4912 | lock_extent_bits(io_tree, hole_start, block_end - 1, |
d0082371 | 4913 | &cached_state); |
fa7c1494 MX |
4914 | ordered = btrfs_lookup_ordered_range(inode, hole_start, |
4915 | block_end - hole_start); | |
9036c102 YZ |
4916 | if (!ordered) |
4917 | break; | |
2ac55d41 JB |
4918 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
4919 | &cached_state, GFP_NOFS); | |
fa7c1494 | 4920 | btrfs_start_ordered_extent(inode, ordered, 1); |
9036c102 YZ |
4921 | btrfs_put_ordered_extent(ordered); |
4922 | } | |
39279cc3 | 4923 | |
9036c102 YZ |
4924 | cur_offset = hole_start; |
4925 | while (1) { | |
4926 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
4927 | block_end - cur_offset, 0); | |
79787eaa JM |
4928 | if (IS_ERR(em)) { |
4929 | err = PTR_ERR(em); | |
f2767956 | 4930 | em = NULL; |
79787eaa JM |
4931 | break; |
4932 | } | |
9036c102 | 4933 | last_byte = min(extent_map_end(em), block_end); |
0b246afa | 4934 | last_byte = ALIGN(last_byte, fs_info->sectorsize); |
8082510e | 4935 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 4936 | struct extent_map *hole_em; |
9036c102 | 4937 | hole_size = last_byte - cur_offset; |
9ed74f2d | 4938 | |
16e7549f JB |
4939 | err = maybe_insert_hole(root, inode, cur_offset, |
4940 | hole_size); | |
4941 | if (err) | |
3893e33b | 4942 | break; |
5dc562c5 JB |
4943 | btrfs_drop_extent_cache(inode, cur_offset, |
4944 | cur_offset + hole_size - 1, 0); | |
4945 | hole_em = alloc_extent_map(); | |
4946 | if (!hole_em) { | |
4947 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
4948 | &BTRFS_I(inode)->runtime_flags); | |
4949 | goto next; | |
4950 | } | |
4951 | hole_em->start = cur_offset; | |
4952 | hole_em->len = hole_size; | |
4953 | hole_em->orig_start = cur_offset; | |
8082510e | 4954 | |
5dc562c5 JB |
4955 | hole_em->block_start = EXTENT_MAP_HOLE; |
4956 | hole_em->block_len = 0; | |
b4939680 | 4957 | hole_em->orig_block_len = 0; |
cc95bef6 | 4958 | hole_em->ram_bytes = hole_size; |
0b246afa | 4959 | hole_em->bdev = fs_info->fs_devices->latest_bdev; |
5dc562c5 | 4960 | hole_em->compress_type = BTRFS_COMPRESS_NONE; |
0b246afa | 4961 | hole_em->generation = fs_info->generation; |
8082510e | 4962 | |
5dc562c5 JB |
4963 | while (1) { |
4964 | write_lock(&em_tree->lock); | |
09a2a8f9 | 4965 | err = add_extent_mapping(em_tree, hole_em, 1); |
5dc562c5 JB |
4966 | write_unlock(&em_tree->lock); |
4967 | if (err != -EEXIST) | |
4968 | break; | |
4969 | btrfs_drop_extent_cache(inode, cur_offset, | |
4970 | cur_offset + | |
4971 | hole_size - 1, 0); | |
4972 | } | |
4973 | free_extent_map(hole_em); | |
9036c102 | 4974 | } |
16e7549f | 4975 | next: |
9036c102 | 4976 | free_extent_map(em); |
a22285a6 | 4977 | em = NULL; |
9036c102 | 4978 | cur_offset = last_byte; |
8082510e | 4979 | if (cur_offset >= block_end) |
9036c102 YZ |
4980 | break; |
4981 | } | |
a22285a6 | 4982 | free_extent_map(em); |
2ac55d41 JB |
4983 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
4984 | GFP_NOFS); | |
9036c102 YZ |
4985 | return err; |
4986 | } | |
39279cc3 | 4987 | |
3972f260 | 4988 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 4989 | { |
f4a2f4c5 MX |
4990 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4991 | struct btrfs_trans_handle *trans; | |
a41ad394 | 4992 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
4993 | loff_t newsize = attr->ia_size; |
4994 | int mask = attr->ia_valid; | |
8082510e YZ |
4995 | int ret; |
4996 | ||
3972f260 ES |
4997 | /* |
4998 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
4999 | * special case where we need to update the times despite not having | |
5000 | * these flags set. For all other operations the VFS set these flags | |
5001 | * explicitly if it wants a timestamp update. | |
5002 | */ | |
dff6efc3 CH |
5003 | if (newsize != oldsize) { |
5004 | inode_inc_iversion(inode); | |
5005 | if (!(mask & (ATTR_CTIME | ATTR_MTIME))) | |
5006 | inode->i_ctime = inode->i_mtime = | |
c2050a45 | 5007 | current_time(inode); |
dff6efc3 | 5008 | } |
3972f260 | 5009 | |
a41ad394 | 5010 | if (newsize > oldsize) { |
9ea24bbe FM |
5011 | /* |
5012 | * Don't do an expanding truncate while snapshoting is ongoing. | |
5013 | * This is to ensure the snapshot captures a fully consistent | |
5014 | * state of this file - if the snapshot captures this expanding | |
5015 | * truncation, it must capture all writes that happened before | |
5016 | * this truncation. | |
5017 | */ | |
0bc19f90 | 5018 | btrfs_wait_for_snapshot_creation(root); |
a41ad394 | 5019 | ret = btrfs_cont_expand(inode, oldsize, newsize); |
9ea24bbe FM |
5020 | if (ret) { |
5021 | btrfs_end_write_no_snapshoting(root); | |
8082510e | 5022 | return ret; |
9ea24bbe | 5023 | } |
8082510e | 5024 | |
f4a2f4c5 | 5025 | trans = btrfs_start_transaction(root, 1); |
9ea24bbe FM |
5026 | if (IS_ERR(trans)) { |
5027 | btrfs_end_write_no_snapshoting(root); | |
f4a2f4c5 | 5028 | return PTR_ERR(trans); |
9ea24bbe | 5029 | } |
f4a2f4c5 MX |
5030 | |
5031 | i_size_write(inode, newsize); | |
5032 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
27772b68 | 5033 | pagecache_isize_extended(inode, oldsize, newsize); |
f4a2f4c5 | 5034 | ret = btrfs_update_inode(trans, root, inode); |
9ea24bbe | 5035 | btrfs_end_write_no_snapshoting(root); |
3a45bb20 | 5036 | btrfs_end_transaction(trans); |
a41ad394 | 5037 | } else { |
8082510e | 5038 | |
a41ad394 JB |
5039 | /* |
5040 | * We're truncating a file that used to have good data down to | |
5041 | * zero. Make sure it gets into the ordered flush list so that | |
5042 | * any new writes get down to disk quickly. | |
5043 | */ | |
5044 | if (newsize == 0) | |
72ac3c0d JB |
5045 | set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
5046 | &BTRFS_I(inode)->runtime_flags); | |
8082510e | 5047 | |
f3fe820c JB |
5048 | /* |
5049 | * 1 for the orphan item we're going to add | |
5050 | * 1 for the orphan item deletion. | |
5051 | */ | |
5052 | trans = btrfs_start_transaction(root, 2); | |
5053 | if (IS_ERR(trans)) | |
5054 | return PTR_ERR(trans); | |
5055 | ||
5056 | /* | |
5057 | * We need to do this in case we fail at _any_ point during the | |
5058 | * actual truncate. Once we do the truncate_setsize we could | |
5059 | * invalidate pages which forces any outstanding ordered io to | |
5060 | * be instantly completed which will give us extents that need | |
5061 | * to be truncated. If we fail to get an orphan inode down we | |
5062 | * could have left over extents that were never meant to live, | |
01327610 | 5063 | * so we need to guarantee from this point on that everything |
f3fe820c JB |
5064 | * will be consistent. |
5065 | */ | |
5066 | ret = btrfs_orphan_add(trans, inode); | |
3a45bb20 | 5067 | btrfs_end_transaction(trans); |
f3fe820c JB |
5068 | if (ret) |
5069 | return ret; | |
5070 | ||
a41ad394 JB |
5071 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
5072 | truncate_setsize(inode, newsize); | |
2e60a51e MX |
5073 | |
5074 | /* Disable nonlocked read DIO to avoid the end less truncate */ | |
5075 | btrfs_inode_block_unlocked_dio(inode); | |
5076 | inode_dio_wait(inode); | |
5077 | btrfs_inode_resume_unlocked_dio(inode); | |
5078 | ||
a41ad394 | 5079 | ret = btrfs_truncate(inode); |
7f4f6e0a JB |
5080 | if (ret && inode->i_nlink) { |
5081 | int err; | |
5082 | ||
5083 | /* | |
5084 | * failed to truncate, disk_i_size is only adjusted down | |
5085 | * as we remove extents, so it should represent the true | |
5086 | * size of the inode, so reset the in memory size and | |
5087 | * delete our orphan entry. | |
5088 | */ | |
5089 | trans = btrfs_join_transaction(root); | |
5090 | if (IS_ERR(trans)) { | |
5091 | btrfs_orphan_del(NULL, inode); | |
5092 | return ret; | |
5093 | } | |
5094 | i_size_write(inode, BTRFS_I(inode)->disk_i_size); | |
5095 | err = btrfs_orphan_del(trans, inode); | |
5096 | if (err) | |
66642832 | 5097 | btrfs_abort_transaction(trans, err); |
3a45bb20 | 5098 | btrfs_end_transaction(trans); |
7f4f6e0a | 5099 | } |
8082510e YZ |
5100 | } |
5101 | ||
a41ad394 | 5102 | return ret; |
8082510e YZ |
5103 | } |
5104 | ||
9036c102 YZ |
5105 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
5106 | { | |
2b0143b5 | 5107 | struct inode *inode = d_inode(dentry); |
b83cc969 | 5108 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 5109 | int err; |
39279cc3 | 5110 | |
b83cc969 LZ |
5111 | if (btrfs_root_readonly(root)) |
5112 | return -EROFS; | |
5113 | ||
31051c85 | 5114 | err = setattr_prepare(dentry, attr); |
9036c102 YZ |
5115 | if (err) |
5116 | return err; | |
2bf5a725 | 5117 | |
5a3f23d5 | 5118 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 5119 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
5120 | if (err) |
5121 | return err; | |
39279cc3 | 5122 | } |
9036c102 | 5123 | |
1025774c CH |
5124 | if (attr->ia_valid) { |
5125 | setattr_copy(inode, attr); | |
0c4d2d95 | 5126 | inode_inc_iversion(inode); |
22c44fe6 | 5127 | err = btrfs_dirty_inode(inode); |
1025774c | 5128 | |
22c44fe6 | 5129 | if (!err && attr->ia_valid & ATTR_MODE) |
996a710d | 5130 | err = posix_acl_chmod(inode, inode->i_mode); |
1025774c | 5131 | } |
33268eaf | 5132 | |
39279cc3 CM |
5133 | return err; |
5134 | } | |
61295eb8 | 5135 | |
131e404a FDBM |
5136 | /* |
5137 | * While truncating the inode pages during eviction, we get the VFS calling | |
5138 | * btrfs_invalidatepage() against each page of the inode. This is slow because | |
5139 | * the calls to btrfs_invalidatepage() result in a huge amount of calls to | |
5140 | * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting | |
5141 | * extent_state structures over and over, wasting lots of time. | |
5142 | * | |
5143 | * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all | |
5144 | * those expensive operations on a per page basis and do only the ordered io | |
5145 | * finishing, while we release here the extent_map and extent_state structures, | |
5146 | * without the excessive merging and splitting. | |
5147 | */ | |
5148 | static void evict_inode_truncate_pages(struct inode *inode) | |
5149 | { | |
5150 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
5151 | struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree; | |
5152 | struct rb_node *node; | |
5153 | ||
5154 | ASSERT(inode->i_state & I_FREEING); | |
91b0abe3 | 5155 | truncate_inode_pages_final(&inode->i_data); |
131e404a FDBM |
5156 | |
5157 | write_lock(&map_tree->lock); | |
5158 | while (!RB_EMPTY_ROOT(&map_tree->map)) { | |
5159 | struct extent_map *em; | |
5160 | ||
5161 | node = rb_first(&map_tree->map); | |
5162 | em = rb_entry(node, struct extent_map, rb_node); | |
180589ef WS |
5163 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
5164 | clear_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
131e404a FDBM |
5165 | remove_extent_mapping(map_tree, em); |
5166 | free_extent_map(em); | |
7064dd5c FM |
5167 | if (need_resched()) { |
5168 | write_unlock(&map_tree->lock); | |
5169 | cond_resched(); | |
5170 | write_lock(&map_tree->lock); | |
5171 | } | |
131e404a FDBM |
5172 | } |
5173 | write_unlock(&map_tree->lock); | |
5174 | ||
6ca07097 FM |
5175 | /* |
5176 | * Keep looping until we have no more ranges in the io tree. | |
5177 | * We can have ongoing bios started by readpages (called from readahead) | |
9c6429d9 FM |
5178 | * that have their endio callback (extent_io.c:end_bio_extent_readpage) |
5179 | * still in progress (unlocked the pages in the bio but did not yet | |
5180 | * unlocked the ranges in the io tree). Therefore this means some | |
6ca07097 FM |
5181 | * ranges can still be locked and eviction started because before |
5182 | * submitting those bios, which are executed by a separate task (work | |
5183 | * queue kthread), inode references (inode->i_count) were not taken | |
5184 | * (which would be dropped in the end io callback of each bio). | |
5185 | * Therefore here we effectively end up waiting for those bios and | |
5186 | * anyone else holding locked ranges without having bumped the inode's | |
5187 | * reference count - if we don't do it, when they access the inode's | |
5188 | * io_tree to unlock a range it may be too late, leading to an | |
5189 | * use-after-free issue. | |
5190 | */ | |
131e404a FDBM |
5191 | spin_lock(&io_tree->lock); |
5192 | while (!RB_EMPTY_ROOT(&io_tree->state)) { | |
5193 | struct extent_state *state; | |
5194 | struct extent_state *cached_state = NULL; | |
6ca07097 FM |
5195 | u64 start; |
5196 | u64 end; | |
131e404a FDBM |
5197 | |
5198 | node = rb_first(&io_tree->state); | |
5199 | state = rb_entry(node, struct extent_state, rb_node); | |
6ca07097 FM |
5200 | start = state->start; |
5201 | end = state->end; | |
131e404a FDBM |
5202 | spin_unlock(&io_tree->lock); |
5203 | ||
ff13db41 | 5204 | lock_extent_bits(io_tree, start, end, &cached_state); |
b9d0b389 QW |
5205 | |
5206 | /* | |
5207 | * If still has DELALLOC flag, the extent didn't reach disk, | |
5208 | * and its reserved space won't be freed by delayed_ref. | |
5209 | * So we need to free its reserved space here. | |
5210 | * (Refer to comment in btrfs_invalidatepage, case 2) | |
5211 | * | |
5212 | * Note, end is the bytenr of last byte, so we need + 1 here. | |
5213 | */ | |
5214 | if (state->state & EXTENT_DELALLOC) | |
5215 | btrfs_qgroup_free_data(inode, start, end - start + 1); | |
5216 | ||
6ca07097 | 5217 | clear_extent_bit(io_tree, start, end, |
131e404a FDBM |
5218 | EXTENT_LOCKED | EXTENT_DIRTY | |
5219 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
5220 | EXTENT_DEFRAG, 1, 1, | |
5221 | &cached_state, GFP_NOFS); | |
131e404a | 5222 | |
7064dd5c | 5223 | cond_resched(); |
131e404a FDBM |
5224 | spin_lock(&io_tree->lock); |
5225 | } | |
5226 | spin_unlock(&io_tree->lock); | |
5227 | } | |
5228 | ||
bd555975 | 5229 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 | 5230 | { |
0b246afa | 5231 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
5232 | struct btrfs_trans_handle *trans; |
5233 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
726c35fa | 5234 | struct btrfs_block_rsv *rsv, *global_rsv; |
3bce876f | 5235 | int steal_from_global = 0; |
3d48d981 | 5236 | u64 min_size; |
39279cc3 CM |
5237 | int ret; |
5238 | ||
1abe9b8a | 5239 | trace_btrfs_inode_evict(inode); |
5240 | ||
3d48d981 NB |
5241 | if (!root) { |
5242 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); | |
5243 | return; | |
5244 | } | |
5245 | ||
0b246afa | 5246 | min_size = btrfs_calc_trunc_metadata_size(fs_info, 1); |
3d48d981 | 5247 | |
131e404a FDBM |
5248 | evict_inode_truncate_pages(inode); |
5249 | ||
69e9c6c6 SB |
5250 | if (inode->i_nlink && |
5251 | ((btrfs_root_refs(&root->root_item) != 0 && | |
5252 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) || | |
5253 | btrfs_is_free_space_inode(inode))) | |
bd555975 AV |
5254 | goto no_delete; |
5255 | ||
39279cc3 | 5256 | if (is_bad_inode(inode)) { |
7b128766 | 5257 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
5258 | goto no_delete; |
5259 | } | |
bd555975 | 5260 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
a30e577c JM |
5261 | if (!special_file(inode->i_mode)) |
5262 | btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
5f39d397 | 5263 | |
f612496b MX |
5264 | btrfs_free_io_failure_record(inode, 0, (u64)-1); |
5265 | ||
0b246afa | 5266 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) { |
6bf02314 | 5267 | BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
8a35d95f | 5268 | &BTRFS_I(inode)->runtime_flags)); |
c71bf099 YZ |
5269 | goto no_delete; |
5270 | } | |
5271 | ||
76dda93c | 5272 | if (inode->i_nlink > 0) { |
69e9c6c6 SB |
5273 | BUG_ON(btrfs_root_refs(&root->root_item) != 0 && |
5274 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID); | |
76dda93c YZ |
5275 | goto no_delete; |
5276 | } | |
5277 | ||
aa79021f | 5278 | ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode)); |
0e8c36a9 MX |
5279 | if (ret) { |
5280 | btrfs_orphan_del(NULL, inode); | |
5281 | goto no_delete; | |
5282 | } | |
5283 | ||
2ff7e61e | 5284 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
4289a667 JB |
5285 | if (!rsv) { |
5286 | btrfs_orphan_del(NULL, inode); | |
5287 | goto no_delete; | |
5288 | } | |
4a338542 | 5289 | rsv->size = min_size; |
ca7e70f5 | 5290 | rsv->failfast = 1; |
0b246afa | 5291 | global_rsv = &fs_info->global_block_rsv; |
4289a667 | 5292 | |
dbe674a9 | 5293 | btrfs_i_size_write(inode, 0); |
5f39d397 | 5294 | |
4289a667 | 5295 | /* |
8407aa46 MX |
5296 | * This is a bit simpler than btrfs_truncate since we've already |
5297 | * reserved our space for our orphan item in the unlink, so we just | |
5298 | * need to reserve some slack space in case we add bytes and update | |
5299 | * inode item when doing the truncate. | |
4289a667 | 5300 | */ |
8082510e | 5301 | while (1) { |
08e007d2 MX |
5302 | ret = btrfs_block_rsv_refill(root, rsv, min_size, |
5303 | BTRFS_RESERVE_FLUSH_LIMIT); | |
726c35fa JB |
5304 | |
5305 | /* | |
5306 | * Try and steal from the global reserve since we will | |
5307 | * likely not use this space anyway, we want to try as | |
5308 | * hard as possible to get this to work. | |
5309 | */ | |
5310 | if (ret) | |
3bce876f JB |
5311 | steal_from_global++; |
5312 | else | |
5313 | steal_from_global = 0; | |
5314 | ret = 0; | |
d68fc57b | 5315 | |
3bce876f JB |
5316 | /* |
5317 | * steal_from_global == 0: we reserved stuff, hooray! | |
5318 | * steal_from_global == 1: we didn't reserve stuff, boo! | |
5319 | * steal_from_global == 2: we've committed, still not a lot of | |
5320 | * room but maybe we'll have room in the global reserve this | |
5321 | * time. | |
5322 | * steal_from_global == 3: abandon all hope! | |
5323 | */ | |
5324 | if (steal_from_global > 2) { | |
0b246afa JM |
5325 | btrfs_warn(fs_info, |
5326 | "Could not get space for a delete, will truncate on mount %d", | |
5327 | ret); | |
4289a667 | 5328 | btrfs_orphan_del(NULL, inode); |
2ff7e61e | 5329 | btrfs_free_block_rsv(fs_info, rsv); |
4289a667 | 5330 | goto no_delete; |
d68fc57b | 5331 | } |
7b128766 | 5332 | |
0e8c36a9 | 5333 | trans = btrfs_join_transaction(root); |
4289a667 JB |
5334 | if (IS_ERR(trans)) { |
5335 | btrfs_orphan_del(NULL, inode); | |
2ff7e61e | 5336 | btrfs_free_block_rsv(fs_info, rsv); |
4289a667 | 5337 | goto no_delete; |
d68fc57b | 5338 | } |
7b128766 | 5339 | |
3bce876f | 5340 | /* |
01327610 | 5341 | * We can't just steal from the global reserve, we need to make |
3bce876f JB |
5342 | * sure there is room to do it, if not we need to commit and try |
5343 | * again. | |
5344 | */ | |
5345 | if (steal_from_global) { | |
2ff7e61e | 5346 | if (!btrfs_check_space_for_delayed_refs(trans, fs_info)) |
3bce876f | 5347 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, |
25d609f8 | 5348 | min_size, 0); |
3bce876f JB |
5349 | else |
5350 | ret = -ENOSPC; | |
5351 | } | |
5352 | ||
5353 | /* | |
5354 | * Couldn't steal from the global reserve, we have too much | |
5355 | * pending stuff built up, commit the transaction and try it | |
5356 | * again. | |
5357 | */ | |
5358 | if (ret) { | |
3a45bb20 | 5359 | ret = btrfs_commit_transaction(trans); |
3bce876f JB |
5360 | if (ret) { |
5361 | btrfs_orphan_del(NULL, inode); | |
2ff7e61e | 5362 | btrfs_free_block_rsv(fs_info, rsv); |
3bce876f JB |
5363 | goto no_delete; |
5364 | } | |
5365 | continue; | |
5366 | } else { | |
5367 | steal_from_global = 0; | |
5368 | } | |
5369 | ||
4289a667 JB |
5370 | trans->block_rsv = rsv; |
5371 | ||
d68fc57b | 5372 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
28ed1345 | 5373 | if (ret != -ENOSPC && ret != -EAGAIN) |
8082510e | 5374 | break; |
85e21bac | 5375 | |
0b246afa | 5376 | trans->block_rsv = &fs_info->trans_block_rsv; |
3a45bb20 | 5377 | btrfs_end_transaction(trans); |
8082510e | 5378 | trans = NULL; |
2ff7e61e | 5379 | btrfs_btree_balance_dirty(fs_info); |
8082510e | 5380 | } |
5f39d397 | 5381 | |
2ff7e61e | 5382 | btrfs_free_block_rsv(fs_info, rsv); |
4289a667 | 5383 | |
4ef31a45 JB |
5384 | /* |
5385 | * Errors here aren't a big deal, it just means we leave orphan items | |
5386 | * in the tree. They will be cleaned up on the next mount. | |
5387 | */ | |
8082510e | 5388 | if (ret == 0) { |
4289a667 | 5389 | trans->block_rsv = root->orphan_block_rsv; |
4ef31a45 JB |
5390 | btrfs_orphan_del(trans, inode); |
5391 | } else { | |
5392 | btrfs_orphan_del(NULL, inode); | |
8082510e | 5393 | } |
54aa1f4d | 5394 | |
0b246afa JM |
5395 | trans->block_rsv = &fs_info->trans_block_rsv; |
5396 | if (!(root == fs_info->tree_root || | |
581bb050 | 5397 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) |
4a0cc7ca | 5398 | btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode))); |
581bb050 | 5399 | |
3a45bb20 | 5400 | btrfs_end_transaction(trans); |
2ff7e61e | 5401 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 | 5402 | no_delete: |
f48d1cf5 | 5403 | btrfs_remove_delayed_node(BTRFS_I(inode)); |
dbd5768f | 5404 | clear_inode(inode); |
39279cc3 CM |
5405 | } |
5406 | ||
5407 | /* | |
5408 | * this returns the key found in the dir entry in the location pointer. | |
5409 | * If no dir entries were found, location->objectid is 0. | |
5410 | */ | |
5411 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
5412 | struct btrfs_key *location) | |
5413 | { | |
5414 | const char *name = dentry->d_name.name; | |
5415 | int namelen = dentry->d_name.len; | |
5416 | struct btrfs_dir_item *di; | |
5417 | struct btrfs_path *path; | |
5418 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 5419 | int ret = 0; |
39279cc3 CM |
5420 | |
5421 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
5422 | if (!path) |
5423 | return -ENOMEM; | |
3954401f | 5424 | |
f85b7379 DS |
5425 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)), |
5426 | name, namelen, 0); | |
0d9f7f3e Y |
5427 | if (IS_ERR(di)) |
5428 | ret = PTR_ERR(di); | |
d397712b | 5429 | |
c704005d | 5430 | if (IS_ERR_OR_NULL(di)) |
3954401f | 5431 | goto out_err; |
d397712b | 5432 | |
5f39d397 | 5433 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 5434 | out: |
39279cc3 CM |
5435 | btrfs_free_path(path); |
5436 | return ret; | |
3954401f CM |
5437 | out_err: |
5438 | location->objectid = 0; | |
5439 | goto out; | |
39279cc3 CM |
5440 | } |
5441 | ||
5442 | /* | |
5443 | * when we hit a tree root in a directory, the btrfs part of the inode | |
5444 | * needs to be changed to reflect the root directory of the tree root. This | |
5445 | * is kind of like crossing a mount point. | |
5446 | */ | |
2ff7e61e | 5447 | static int fixup_tree_root_location(struct btrfs_fs_info *fs_info, |
4df27c4d YZ |
5448 | struct inode *dir, |
5449 | struct dentry *dentry, | |
5450 | struct btrfs_key *location, | |
5451 | struct btrfs_root **sub_root) | |
39279cc3 | 5452 | { |
4df27c4d YZ |
5453 | struct btrfs_path *path; |
5454 | struct btrfs_root *new_root; | |
5455 | struct btrfs_root_ref *ref; | |
5456 | struct extent_buffer *leaf; | |
1d4c08e0 | 5457 | struct btrfs_key key; |
4df27c4d YZ |
5458 | int ret; |
5459 | int err = 0; | |
39279cc3 | 5460 | |
4df27c4d YZ |
5461 | path = btrfs_alloc_path(); |
5462 | if (!path) { | |
5463 | err = -ENOMEM; | |
5464 | goto out; | |
5465 | } | |
39279cc3 | 5466 | |
4df27c4d | 5467 | err = -ENOENT; |
1d4c08e0 DS |
5468 | key.objectid = BTRFS_I(dir)->root->root_key.objectid; |
5469 | key.type = BTRFS_ROOT_REF_KEY; | |
5470 | key.offset = location->objectid; | |
5471 | ||
0b246afa | 5472 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
4df27c4d YZ |
5473 | if (ret) { |
5474 | if (ret < 0) | |
5475 | err = ret; | |
5476 | goto out; | |
5477 | } | |
39279cc3 | 5478 | |
4df27c4d YZ |
5479 | leaf = path->nodes[0]; |
5480 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
4a0cc7ca | 5481 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) || |
4df27c4d YZ |
5482 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
5483 | goto out; | |
39279cc3 | 5484 | |
4df27c4d YZ |
5485 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
5486 | (unsigned long)(ref + 1), | |
5487 | dentry->d_name.len); | |
5488 | if (ret) | |
5489 | goto out; | |
5490 | ||
b3b4aa74 | 5491 | btrfs_release_path(path); |
4df27c4d | 5492 | |
0b246afa | 5493 | new_root = btrfs_read_fs_root_no_name(fs_info, location); |
4df27c4d YZ |
5494 | if (IS_ERR(new_root)) { |
5495 | err = PTR_ERR(new_root); | |
5496 | goto out; | |
5497 | } | |
5498 | ||
4df27c4d YZ |
5499 | *sub_root = new_root; |
5500 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
5501 | location->type = BTRFS_INODE_ITEM_KEY; | |
5502 | location->offset = 0; | |
5503 | err = 0; | |
5504 | out: | |
5505 | btrfs_free_path(path); | |
5506 | return err; | |
39279cc3 CM |
5507 | } |
5508 | ||
5d4f98a2 YZ |
5509 | static void inode_tree_add(struct inode *inode) |
5510 | { | |
5511 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5512 | struct btrfs_inode *entry; | |
03e860bd FNP |
5513 | struct rb_node **p; |
5514 | struct rb_node *parent; | |
cef21937 | 5515 | struct rb_node *new = &BTRFS_I(inode)->rb_node; |
4a0cc7ca | 5516 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
5d4f98a2 | 5517 | |
1d3382cb | 5518 | if (inode_unhashed(inode)) |
76dda93c | 5519 | return; |
e1409cef | 5520 | parent = NULL; |
5d4f98a2 | 5521 | spin_lock(&root->inode_lock); |
e1409cef | 5522 | p = &root->inode_tree.rb_node; |
5d4f98a2 YZ |
5523 | while (*p) { |
5524 | parent = *p; | |
5525 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
5526 | ||
4a0cc7ca | 5527 | if (ino < btrfs_ino(BTRFS_I(&entry->vfs_inode))) |
03e860bd | 5528 | p = &parent->rb_left; |
4a0cc7ca | 5529 | else if (ino > btrfs_ino(BTRFS_I(&entry->vfs_inode))) |
03e860bd | 5530 | p = &parent->rb_right; |
5d4f98a2 YZ |
5531 | else { |
5532 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 5533 | (I_WILL_FREE | I_FREEING))); |
cef21937 | 5534 | rb_replace_node(parent, new, &root->inode_tree); |
03e860bd FNP |
5535 | RB_CLEAR_NODE(parent); |
5536 | spin_unlock(&root->inode_lock); | |
cef21937 | 5537 | return; |
5d4f98a2 YZ |
5538 | } |
5539 | } | |
cef21937 FDBM |
5540 | rb_link_node(new, parent, p); |
5541 | rb_insert_color(new, &root->inode_tree); | |
5d4f98a2 YZ |
5542 | spin_unlock(&root->inode_lock); |
5543 | } | |
5544 | ||
5545 | static void inode_tree_del(struct inode *inode) | |
5546 | { | |
0b246afa | 5547 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5d4f98a2 | 5548 | struct btrfs_root *root = BTRFS_I(inode)->root; |
76dda93c | 5549 | int empty = 0; |
5d4f98a2 | 5550 | |
03e860bd | 5551 | spin_lock(&root->inode_lock); |
5d4f98a2 | 5552 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 5553 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 5554 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 5555 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 5556 | } |
03e860bd | 5557 | spin_unlock(&root->inode_lock); |
76dda93c | 5558 | |
69e9c6c6 | 5559 | if (empty && btrfs_root_refs(&root->root_item) == 0) { |
0b246afa | 5560 | synchronize_srcu(&fs_info->subvol_srcu); |
76dda93c YZ |
5561 | spin_lock(&root->inode_lock); |
5562 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
5563 | spin_unlock(&root->inode_lock); | |
5564 | if (empty) | |
5565 | btrfs_add_dead_root(root); | |
5566 | } | |
5567 | } | |
5568 | ||
143bede5 | 5569 | void btrfs_invalidate_inodes(struct btrfs_root *root) |
76dda93c | 5570 | { |
0b246afa | 5571 | struct btrfs_fs_info *fs_info = root->fs_info; |
76dda93c YZ |
5572 | struct rb_node *node; |
5573 | struct rb_node *prev; | |
5574 | struct btrfs_inode *entry; | |
5575 | struct inode *inode; | |
5576 | u64 objectid = 0; | |
5577 | ||
0b246afa | 5578 | if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
7813b3db | 5579 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); |
76dda93c YZ |
5580 | |
5581 | spin_lock(&root->inode_lock); | |
5582 | again: | |
5583 | node = root->inode_tree.rb_node; | |
5584 | prev = NULL; | |
5585 | while (node) { | |
5586 | prev = node; | |
5587 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
5588 | ||
4a0cc7ca | 5589 | if (objectid < btrfs_ino(BTRFS_I(&entry->vfs_inode))) |
76dda93c | 5590 | node = node->rb_left; |
4a0cc7ca | 5591 | else if (objectid > btrfs_ino(BTRFS_I(&entry->vfs_inode))) |
76dda93c YZ |
5592 | node = node->rb_right; |
5593 | else | |
5594 | break; | |
5595 | } | |
5596 | if (!node) { | |
5597 | while (prev) { | |
5598 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
4a0cc7ca | 5599 | if (objectid <= btrfs_ino(BTRFS_I(&entry->vfs_inode))) { |
76dda93c YZ |
5600 | node = prev; |
5601 | break; | |
5602 | } | |
5603 | prev = rb_next(prev); | |
5604 | } | |
5605 | } | |
5606 | while (node) { | |
5607 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
4a0cc7ca | 5608 | objectid = btrfs_ino(BTRFS_I(&entry->vfs_inode)) + 1; |
76dda93c YZ |
5609 | inode = igrab(&entry->vfs_inode); |
5610 | if (inode) { | |
5611 | spin_unlock(&root->inode_lock); | |
5612 | if (atomic_read(&inode->i_count) > 1) | |
5613 | d_prune_aliases(inode); | |
5614 | /* | |
45321ac5 | 5615 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
5616 | * the inode cache when its usage count |
5617 | * hits zero. | |
5618 | */ | |
5619 | iput(inode); | |
5620 | cond_resched(); | |
5621 | spin_lock(&root->inode_lock); | |
5622 | goto again; | |
5623 | } | |
5624 | ||
5625 | if (cond_resched_lock(&root->inode_lock)) | |
5626 | goto again; | |
5627 | ||
5628 | node = rb_next(node); | |
5629 | } | |
5630 | spin_unlock(&root->inode_lock); | |
5d4f98a2 YZ |
5631 | } |
5632 | ||
e02119d5 CM |
5633 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
5634 | { | |
5635 | struct btrfs_iget_args *args = p; | |
90d3e592 CM |
5636 | inode->i_ino = args->location->objectid; |
5637 | memcpy(&BTRFS_I(inode)->location, args->location, | |
5638 | sizeof(*args->location)); | |
e02119d5 | 5639 | BTRFS_I(inode)->root = args->root; |
39279cc3 CM |
5640 | return 0; |
5641 | } | |
5642 | ||
5643 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
5644 | { | |
5645 | struct btrfs_iget_args *args = opaque; | |
90d3e592 | 5646 | return args->location->objectid == BTRFS_I(inode)->location.objectid && |
d397712b | 5647 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
5648 | } |
5649 | ||
5d4f98a2 | 5650 | static struct inode *btrfs_iget_locked(struct super_block *s, |
90d3e592 | 5651 | struct btrfs_key *location, |
5d4f98a2 | 5652 | struct btrfs_root *root) |
39279cc3 CM |
5653 | { |
5654 | struct inode *inode; | |
5655 | struct btrfs_iget_args args; | |
90d3e592 | 5656 | unsigned long hashval = btrfs_inode_hash(location->objectid, root); |
778ba82b | 5657 | |
90d3e592 | 5658 | args.location = location; |
39279cc3 CM |
5659 | args.root = root; |
5660 | ||
778ba82b | 5661 | inode = iget5_locked(s, hashval, btrfs_find_actor, |
39279cc3 CM |
5662 | btrfs_init_locked_inode, |
5663 | (void *)&args); | |
5664 | return inode; | |
5665 | } | |
5666 | ||
1a54ef8c BR |
5667 | /* Get an inode object given its location and corresponding root. |
5668 | * Returns in *is_new if the inode was read from disk | |
5669 | */ | |
5670 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 5671 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
5672 | { |
5673 | struct inode *inode; | |
5674 | ||
90d3e592 | 5675 | inode = btrfs_iget_locked(s, location, root); |
1a54ef8c | 5676 | if (!inode) |
5d4f98a2 | 5677 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
5678 | |
5679 | if (inode->i_state & I_NEW) { | |
67710892 FM |
5680 | int ret; |
5681 | ||
5682 | ret = btrfs_read_locked_inode(inode); | |
1748f843 MF |
5683 | if (!is_bad_inode(inode)) { |
5684 | inode_tree_add(inode); | |
5685 | unlock_new_inode(inode); | |
5686 | if (new) | |
5687 | *new = 1; | |
5688 | } else { | |
e0b6d65b ST |
5689 | unlock_new_inode(inode); |
5690 | iput(inode); | |
67710892 FM |
5691 | ASSERT(ret < 0); |
5692 | inode = ERR_PTR(ret < 0 ? ret : -ESTALE); | |
1748f843 MF |
5693 | } |
5694 | } | |
5695 | ||
1a54ef8c BR |
5696 | return inode; |
5697 | } | |
5698 | ||
4df27c4d YZ |
5699 | static struct inode *new_simple_dir(struct super_block *s, |
5700 | struct btrfs_key *key, | |
5701 | struct btrfs_root *root) | |
5702 | { | |
5703 | struct inode *inode = new_inode(s); | |
5704 | ||
5705 | if (!inode) | |
5706 | return ERR_PTR(-ENOMEM); | |
5707 | ||
4df27c4d YZ |
5708 | BTRFS_I(inode)->root = root; |
5709 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
72ac3c0d | 5710 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
5711 | |
5712 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
848cce0d | 5713 | inode->i_op = &btrfs_dir_ro_inode_operations; |
1fdf4194 | 5714 | inode->i_opflags &= ~IOP_XATTR; |
4df27c4d YZ |
5715 | inode->i_fop = &simple_dir_operations; |
5716 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
c2050a45 | 5717 | inode->i_mtime = current_time(inode); |
9cc97d64 | 5718 | inode->i_atime = inode->i_mtime; |
5719 | inode->i_ctime = inode->i_mtime; | |
5720 | BTRFS_I(inode)->i_otime = inode->i_mtime; | |
4df27c4d YZ |
5721 | |
5722 | return inode; | |
5723 | } | |
5724 | ||
3de4586c | 5725 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 5726 | { |
0b246afa | 5727 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
d397712b | 5728 | struct inode *inode; |
4df27c4d | 5729 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
5730 | struct btrfs_root *sub_root = root; |
5731 | struct btrfs_key location; | |
76dda93c | 5732 | int index; |
b4aff1f8 | 5733 | int ret = 0; |
39279cc3 CM |
5734 | |
5735 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
5736 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 5737 | |
39e3c955 | 5738 | ret = btrfs_inode_by_name(dir, dentry, &location); |
39279cc3 CM |
5739 | if (ret < 0) |
5740 | return ERR_PTR(ret); | |
5f39d397 | 5741 | |
4df27c4d | 5742 | if (location.objectid == 0) |
5662344b | 5743 | return ERR_PTR(-ENOENT); |
4df27c4d YZ |
5744 | |
5745 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 5746 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
5747 | return inode; |
5748 | } | |
5749 | ||
5750 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
5751 | ||
0b246afa | 5752 | index = srcu_read_lock(&fs_info->subvol_srcu); |
2ff7e61e | 5753 | ret = fixup_tree_root_location(fs_info, dir, dentry, |
4df27c4d YZ |
5754 | &location, &sub_root); |
5755 | if (ret < 0) { | |
5756 | if (ret != -ENOENT) | |
5757 | inode = ERR_PTR(ret); | |
5758 | else | |
5759 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
5760 | } else { | |
73f73415 | 5761 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 5762 | } |
0b246afa | 5763 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
76dda93c | 5764 | |
34d19bad | 5765 | if (!IS_ERR(inode) && root != sub_root) { |
0b246afa | 5766 | down_read(&fs_info->cleanup_work_sem); |
c71bf099 | 5767 | if (!(inode->i_sb->s_flags & MS_RDONLY)) |
66b4ffd1 | 5768 | ret = btrfs_orphan_cleanup(sub_root); |
0b246afa | 5769 | up_read(&fs_info->cleanup_work_sem); |
01cd3367 JB |
5770 | if (ret) { |
5771 | iput(inode); | |
66b4ffd1 | 5772 | inode = ERR_PTR(ret); |
01cd3367 | 5773 | } |
c71bf099 YZ |
5774 | } |
5775 | ||
3de4586c CM |
5776 | return inode; |
5777 | } | |
5778 | ||
fe15ce44 | 5779 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
5780 | { |
5781 | struct btrfs_root *root; | |
2b0143b5 | 5782 | struct inode *inode = d_inode(dentry); |
76dda93c | 5783 | |
848cce0d | 5784 | if (!inode && !IS_ROOT(dentry)) |
2b0143b5 | 5785 | inode = d_inode(dentry->d_parent); |
76dda93c | 5786 | |
848cce0d LZ |
5787 | if (inode) { |
5788 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
5789 | if (btrfs_root_refs(&root->root_item) == 0) |
5790 | return 1; | |
848cce0d | 5791 | |
4a0cc7ca | 5792 | if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
848cce0d | 5793 | return 1; |
efefb143 | 5794 | } |
76dda93c YZ |
5795 | return 0; |
5796 | } | |
5797 | ||
b4aff1f8 JB |
5798 | static void btrfs_dentry_release(struct dentry *dentry) |
5799 | { | |
944a4515 | 5800 | kfree(dentry->d_fsdata); |
b4aff1f8 JB |
5801 | } |
5802 | ||
3de4586c | 5803 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 5804 | unsigned int flags) |
3de4586c | 5805 | { |
5662344b | 5806 | struct inode *inode; |
a66e7cc6 | 5807 | |
5662344b TI |
5808 | inode = btrfs_lookup_dentry(dir, dentry); |
5809 | if (IS_ERR(inode)) { | |
5810 | if (PTR_ERR(inode) == -ENOENT) | |
5811 | inode = NULL; | |
5812 | else | |
5813 | return ERR_CAST(inode); | |
5814 | } | |
5815 | ||
41d28bca | 5816 | return d_splice_alias(inode, dentry); |
39279cc3 CM |
5817 | } |
5818 | ||
16cdcec7 | 5819 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
5820 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
5821 | }; | |
5822 | ||
9cdda8d3 | 5823 | static int btrfs_real_readdir(struct file *file, struct dir_context *ctx) |
39279cc3 | 5824 | { |
9cdda8d3 | 5825 | struct inode *inode = file_inode(file); |
2ff7e61e | 5826 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
5827 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5828 | struct btrfs_item *item; | |
5829 | struct btrfs_dir_item *di; | |
5830 | struct btrfs_key key; | |
5f39d397 | 5831 | struct btrfs_key found_key; |
39279cc3 | 5832 | struct btrfs_path *path; |
16cdcec7 MX |
5833 | struct list_head ins_list; |
5834 | struct list_head del_list; | |
39279cc3 | 5835 | int ret; |
5f39d397 | 5836 | struct extent_buffer *leaf; |
39279cc3 | 5837 | int slot; |
39279cc3 CM |
5838 | unsigned char d_type; |
5839 | int over = 0; | |
5f39d397 CM |
5840 | char tmp_name[32]; |
5841 | char *name_ptr; | |
5842 | int name_len; | |
02dbfc99 | 5843 | bool put = false; |
c2951f32 | 5844 | struct btrfs_key location; |
5f39d397 | 5845 | |
9cdda8d3 AV |
5846 | if (!dir_emit_dots(file, ctx)) |
5847 | return 0; | |
5848 | ||
49593bfa | 5849 | path = btrfs_alloc_path(); |
16cdcec7 MX |
5850 | if (!path) |
5851 | return -ENOMEM; | |
ff5714cc | 5852 | |
e4058b54 | 5853 | path->reada = READA_FORWARD; |
49593bfa | 5854 | |
c2951f32 JM |
5855 | INIT_LIST_HEAD(&ins_list); |
5856 | INIT_LIST_HEAD(&del_list); | |
5857 | put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list); | |
16cdcec7 | 5858 | |
c2951f32 | 5859 | key.type = BTRFS_DIR_INDEX_KEY; |
9cdda8d3 | 5860 | key.offset = ctx->pos; |
4a0cc7ca | 5861 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
5f39d397 | 5862 | |
39279cc3 CM |
5863 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5864 | if (ret < 0) | |
5865 | goto err; | |
49593bfa DW |
5866 | |
5867 | while (1) { | |
5f39d397 | 5868 | leaf = path->nodes[0]; |
39279cc3 | 5869 | slot = path->slots[0]; |
b9e03af0 LZ |
5870 | if (slot >= btrfs_header_nritems(leaf)) { |
5871 | ret = btrfs_next_leaf(root, path); | |
5872 | if (ret < 0) | |
5873 | goto err; | |
5874 | else if (ret > 0) | |
5875 | break; | |
5876 | continue; | |
39279cc3 | 5877 | } |
3de4586c | 5878 | |
dd3cc16b | 5879 | item = btrfs_item_nr(slot); |
5f39d397 CM |
5880 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
5881 | ||
5882 | if (found_key.objectid != key.objectid) | |
39279cc3 | 5883 | break; |
c2951f32 | 5884 | if (found_key.type != BTRFS_DIR_INDEX_KEY) |
39279cc3 | 5885 | break; |
9cdda8d3 | 5886 | if (found_key.offset < ctx->pos) |
b9e03af0 | 5887 | goto next; |
c2951f32 | 5888 | if (btrfs_should_delete_dir_index(&del_list, found_key.offset)) |
16cdcec7 | 5889 | goto next; |
5f39d397 | 5890 | |
9cdda8d3 | 5891 | ctx->pos = found_key.offset; |
49593bfa | 5892 | |
39279cc3 | 5893 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
2ff7e61e | 5894 | if (verify_dir_item(fs_info, leaf, di)) |
c2951f32 | 5895 | goto next; |
22a94d44 | 5896 | |
c2951f32 JM |
5897 | name_len = btrfs_dir_name_len(leaf, di); |
5898 | if (name_len <= sizeof(tmp_name)) { | |
5899 | name_ptr = tmp_name; | |
5900 | } else { | |
5901 | name_ptr = kmalloc(name_len, GFP_KERNEL); | |
5902 | if (!name_ptr) { | |
5903 | ret = -ENOMEM; | |
5904 | goto err; | |
5f39d397 | 5905 | } |
c2951f32 JM |
5906 | } |
5907 | read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1), | |
5908 | name_len); | |
3de4586c | 5909 | |
c2951f32 JM |
5910 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; |
5911 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
fede766f | 5912 | |
c2951f32 JM |
5913 | over = !dir_emit(ctx, name_ptr, name_len, location.objectid, |
5914 | d_type); | |
5f39d397 | 5915 | |
c2951f32 JM |
5916 | if (name_ptr != tmp_name) |
5917 | kfree(name_ptr); | |
5f39d397 | 5918 | |
c2951f32 JM |
5919 | if (over) |
5920 | goto nopos; | |
d2fbb2b5 | 5921 | ctx->pos++; |
b9e03af0 LZ |
5922 | next: |
5923 | path->slots[0]++; | |
39279cc3 | 5924 | } |
49593bfa | 5925 | |
d2fbb2b5 | 5926 | ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list); |
c2951f32 | 5927 | if (ret) |
bc4ef759 DS |
5928 | goto nopos; |
5929 | ||
db62efbb ZB |
5930 | /* |
5931 | * Stop new entries from being returned after we return the last | |
5932 | * entry. | |
5933 | * | |
5934 | * New directory entries are assigned a strictly increasing | |
5935 | * offset. This means that new entries created during readdir | |
5936 | * are *guaranteed* to be seen in the future by that readdir. | |
5937 | * This has broken buggy programs which operate on names as | |
5938 | * they're returned by readdir. Until we re-use freed offsets | |
5939 | * we have this hack to stop new entries from being returned | |
5940 | * under the assumption that they'll never reach this huge | |
5941 | * offset. | |
5942 | * | |
5943 | * This is being careful not to overflow 32bit loff_t unless the | |
5944 | * last entry requires it because doing so has broken 32bit apps | |
5945 | * in the past. | |
5946 | */ | |
c2951f32 JM |
5947 | if (ctx->pos >= INT_MAX) |
5948 | ctx->pos = LLONG_MAX; | |
5949 | else | |
5950 | ctx->pos = INT_MAX; | |
39279cc3 CM |
5951 | nopos: |
5952 | ret = 0; | |
5953 | err: | |
02dbfc99 OS |
5954 | if (put) |
5955 | btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list); | |
39279cc3 | 5956 | btrfs_free_path(path); |
39279cc3 CM |
5957 | return ret; |
5958 | } | |
5959 | ||
a9185b41 | 5960 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
5961 | { |
5962 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5963 | struct btrfs_trans_handle *trans; | |
5964 | int ret = 0; | |
0af3d00b | 5965 | bool nolock = false; |
39279cc3 | 5966 | |
72ac3c0d | 5967 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
4ca8b41e CM |
5968 | return 0; |
5969 | ||
83eea1f1 | 5970 | if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode)) |
82d5902d | 5971 | nolock = true; |
0af3d00b | 5972 | |
a9185b41 | 5973 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b | 5974 | if (nolock) |
7a7eaa40 | 5975 | trans = btrfs_join_transaction_nolock(root); |
0af3d00b | 5976 | else |
7a7eaa40 | 5977 | trans = btrfs_join_transaction(root); |
3612b495 TI |
5978 | if (IS_ERR(trans)) |
5979 | return PTR_ERR(trans); | |
3a45bb20 | 5980 | ret = btrfs_commit_transaction(trans); |
39279cc3 CM |
5981 | } |
5982 | return ret; | |
5983 | } | |
5984 | ||
5985 | /* | |
54aa1f4d | 5986 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
5987 | * inode changes. But, it is most likely to find the inode in cache. |
5988 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
5989 | * to keep or drop this code. | |
5990 | */ | |
48a3b636 | 5991 | static int btrfs_dirty_inode(struct inode *inode) |
39279cc3 | 5992 | { |
2ff7e61e | 5993 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 CM |
5994 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5995 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
5996 | int ret; |
5997 | ||
72ac3c0d | 5998 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 5999 | return 0; |
39279cc3 | 6000 | |
7a7eaa40 | 6001 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
6002 | if (IS_ERR(trans)) |
6003 | return PTR_ERR(trans); | |
8929ecfa YZ |
6004 | |
6005 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
6006 | if (ret && ret == -ENOSPC) { |
6007 | /* whoops, lets try again with the full transaction */ | |
3a45bb20 | 6008 | btrfs_end_transaction(trans); |
94b60442 | 6009 | trans = btrfs_start_transaction(root, 1); |
22c44fe6 JB |
6010 | if (IS_ERR(trans)) |
6011 | return PTR_ERR(trans); | |
8929ecfa | 6012 | |
94b60442 | 6013 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 6014 | } |
3a45bb20 | 6015 | btrfs_end_transaction(trans); |
16cdcec7 | 6016 | if (BTRFS_I(inode)->delayed_node) |
2ff7e61e | 6017 | btrfs_balance_delayed_items(fs_info); |
22c44fe6 JB |
6018 | |
6019 | return ret; | |
6020 | } | |
6021 | ||
6022 | /* | |
6023 | * This is a copy of file_update_time. We need this so we can return error on | |
6024 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
6025 | */ | |
e41f941a JB |
6026 | static int btrfs_update_time(struct inode *inode, struct timespec *now, |
6027 | int flags) | |
22c44fe6 | 6028 | { |
2bc55652 AB |
6029 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6030 | ||
6031 | if (btrfs_root_readonly(root)) | |
6032 | return -EROFS; | |
6033 | ||
e41f941a | 6034 | if (flags & S_VERSION) |
22c44fe6 | 6035 | inode_inc_iversion(inode); |
e41f941a JB |
6036 | if (flags & S_CTIME) |
6037 | inode->i_ctime = *now; | |
6038 | if (flags & S_MTIME) | |
6039 | inode->i_mtime = *now; | |
6040 | if (flags & S_ATIME) | |
6041 | inode->i_atime = *now; | |
6042 | return btrfs_dirty_inode(inode); | |
39279cc3 CM |
6043 | } |
6044 | ||
d352ac68 CM |
6045 | /* |
6046 | * find the highest existing sequence number in a directory | |
6047 | * and then set the in-memory index_cnt variable to reflect | |
6048 | * free sequence numbers | |
6049 | */ | |
aec7477b JB |
6050 | static int btrfs_set_inode_index_count(struct inode *inode) |
6051 | { | |
6052 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6053 | struct btrfs_key key, found_key; | |
6054 | struct btrfs_path *path; | |
6055 | struct extent_buffer *leaf; | |
6056 | int ret; | |
6057 | ||
4a0cc7ca | 6058 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
962a298f | 6059 | key.type = BTRFS_DIR_INDEX_KEY; |
aec7477b JB |
6060 | key.offset = (u64)-1; |
6061 | ||
6062 | path = btrfs_alloc_path(); | |
6063 | if (!path) | |
6064 | return -ENOMEM; | |
6065 | ||
6066 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
6067 | if (ret < 0) | |
6068 | goto out; | |
6069 | /* FIXME: we should be able to handle this */ | |
6070 | if (ret == 0) | |
6071 | goto out; | |
6072 | ret = 0; | |
6073 | ||
6074 | /* | |
6075 | * MAGIC NUMBER EXPLANATION: | |
6076 | * since we search a directory based on f_pos we have to start at 2 | |
6077 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
6078 | * else has to start at 2 | |
6079 | */ | |
6080 | if (path->slots[0] == 0) { | |
6081 | BTRFS_I(inode)->index_cnt = 2; | |
6082 | goto out; | |
6083 | } | |
6084 | ||
6085 | path->slots[0]--; | |
6086 | ||
6087 | leaf = path->nodes[0]; | |
6088 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
6089 | ||
4a0cc7ca | 6090 | if (found_key.objectid != btrfs_ino(BTRFS_I(inode)) || |
962a298f | 6091 | found_key.type != BTRFS_DIR_INDEX_KEY) { |
aec7477b JB |
6092 | BTRFS_I(inode)->index_cnt = 2; |
6093 | goto out; | |
6094 | } | |
6095 | ||
6096 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
6097 | out: | |
6098 | btrfs_free_path(path); | |
6099 | return ret; | |
6100 | } | |
6101 | ||
d352ac68 CM |
6102 | /* |
6103 | * helper to find a free sequence number in a given directory. This current | |
6104 | * code is very simple, later versions will do smarter things in the btree | |
6105 | */ | |
3de4586c | 6106 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
6107 | { |
6108 | int ret = 0; | |
6109 | ||
6110 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
f5cc7b80 | 6111 | ret = btrfs_inode_delayed_dir_index_count(BTRFS_I(dir)); |
16cdcec7 MX |
6112 | if (ret) { |
6113 | ret = btrfs_set_inode_index_count(dir); | |
6114 | if (ret) | |
6115 | return ret; | |
6116 | } | |
aec7477b JB |
6117 | } |
6118 | ||
00e4e6b3 | 6119 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
6120 | BTRFS_I(dir)->index_cnt++; |
6121 | ||
6122 | return ret; | |
6123 | } | |
6124 | ||
b0d5d10f CM |
6125 | static int btrfs_insert_inode_locked(struct inode *inode) |
6126 | { | |
6127 | struct btrfs_iget_args args; | |
6128 | args.location = &BTRFS_I(inode)->location; | |
6129 | args.root = BTRFS_I(inode)->root; | |
6130 | ||
6131 | return insert_inode_locked4(inode, | |
6132 | btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root), | |
6133 | btrfs_find_actor, &args); | |
6134 | } | |
6135 | ||
39279cc3 CM |
6136 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
6137 | struct btrfs_root *root, | |
aec7477b | 6138 | struct inode *dir, |
9c58309d | 6139 | const char *name, int name_len, |
175a4eb7 AV |
6140 | u64 ref_objectid, u64 objectid, |
6141 | umode_t mode, u64 *index) | |
39279cc3 | 6142 | { |
0b246afa | 6143 | struct btrfs_fs_info *fs_info = root->fs_info; |
39279cc3 | 6144 | struct inode *inode; |
5f39d397 | 6145 | struct btrfs_inode_item *inode_item; |
39279cc3 | 6146 | struct btrfs_key *location; |
5f39d397 | 6147 | struct btrfs_path *path; |
9c58309d CM |
6148 | struct btrfs_inode_ref *ref; |
6149 | struct btrfs_key key[2]; | |
6150 | u32 sizes[2]; | |
ef3b9af5 | 6151 | int nitems = name ? 2 : 1; |
9c58309d | 6152 | unsigned long ptr; |
39279cc3 | 6153 | int ret; |
39279cc3 | 6154 | |
5f39d397 | 6155 | path = btrfs_alloc_path(); |
d8926bb3 MF |
6156 | if (!path) |
6157 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 6158 | |
0b246afa | 6159 | inode = new_inode(fs_info->sb); |
8fb27640 YS |
6160 | if (!inode) { |
6161 | btrfs_free_path(path); | |
39279cc3 | 6162 | return ERR_PTR(-ENOMEM); |
8fb27640 | 6163 | } |
39279cc3 | 6164 | |
5762b5c9 FM |
6165 | /* |
6166 | * O_TMPFILE, set link count to 0, so that after this point, | |
6167 | * we fill in an inode item with the correct link count. | |
6168 | */ | |
6169 | if (!name) | |
6170 | set_nlink(inode, 0); | |
6171 | ||
581bb050 LZ |
6172 | /* |
6173 | * we have to initialize this early, so we can reclaim the inode | |
6174 | * number if we fail afterwards in this function. | |
6175 | */ | |
6176 | inode->i_ino = objectid; | |
6177 | ||
ef3b9af5 | 6178 | if (dir && name) { |
1abe9b8a | 6179 | trace_btrfs_inode_request(dir); |
6180 | ||
3de4586c | 6181 | ret = btrfs_set_inode_index(dir, index); |
09771430 | 6182 | if (ret) { |
8fb27640 | 6183 | btrfs_free_path(path); |
09771430 | 6184 | iput(inode); |
aec7477b | 6185 | return ERR_PTR(ret); |
09771430 | 6186 | } |
ef3b9af5 FM |
6187 | } else if (dir) { |
6188 | *index = 0; | |
aec7477b JB |
6189 | } |
6190 | /* | |
6191 | * index_cnt is ignored for everything but a dir, | |
6192 | * btrfs_get_inode_index_count has an explanation for the magic | |
6193 | * number | |
6194 | */ | |
6195 | BTRFS_I(inode)->index_cnt = 2; | |
67de1176 | 6196 | BTRFS_I(inode)->dir_index = *index; |
39279cc3 | 6197 | BTRFS_I(inode)->root = root; |
e02119d5 | 6198 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 6199 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 6200 | |
5dc562c5 JB |
6201 | /* |
6202 | * We could have gotten an inode number from somebody who was fsynced | |
6203 | * and then removed in this same transaction, so let's just set full | |
6204 | * sync since it will be a full sync anyway and this will blow away the | |
6205 | * old info in the log. | |
6206 | */ | |
6207 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
6208 | ||
9c58309d | 6209 | key[0].objectid = objectid; |
962a298f | 6210 | key[0].type = BTRFS_INODE_ITEM_KEY; |
9c58309d CM |
6211 | key[0].offset = 0; |
6212 | ||
9c58309d | 6213 | sizes[0] = sizeof(struct btrfs_inode_item); |
ef3b9af5 FM |
6214 | |
6215 | if (name) { | |
6216 | /* | |
6217 | * Start new inodes with an inode_ref. This is slightly more | |
6218 | * efficient for small numbers of hard links since they will | |
6219 | * be packed into one item. Extended refs will kick in if we | |
6220 | * add more hard links than can fit in the ref item. | |
6221 | */ | |
6222 | key[1].objectid = objectid; | |
962a298f | 6223 | key[1].type = BTRFS_INODE_REF_KEY; |
ef3b9af5 FM |
6224 | key[1].offset = ref_objectid; |
6225 | ||
6226 | sizes[1] = name_len + sizeof(*ref); | |
6227 | } | |
9c58309d | 6228 | |
b0d5d10f CM |
6229 | location = &BTRFS_I(inode)->location; |
6230 | location->objectid = objectid; | |
6231 | location->offset = 0; | |
962a298f | 6232 | location->type = BTRFS_INODE_ITEM_KEY; |
b0d5d10f CM |
6233 | |
6234 | ret = btrfs_insert_inode_locked(inode); | |
6235 | if (ret < 0) | |
6236 | goto fail; | |
6237 | ||
b9473439 | 6238 | path->leave_spinning = 1; |
ef3b9af5 | 6239 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems); |
9c58309d | 6240 | if (ret != 0) |
b0d5d10f | 6241 | goto fail_unlock; |
5f39d397 | 6242 | |
ecc11fab | 6243 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 6244 | inode_set_bytes(inode, 0); |
9cc97d64 | 6245 | |
c2050a45 | 6246 | inode->i_mtime = current_time(inode); |
9cc97d64 | 6247 | inode->i_atime = inode->i_mtime; |
6248 | inode->i_ctime = inode->i_mtime; | |
6249 | BTRFS_I(inode)->i_otime = inode->i_mtime; | |
6250 | ||
5f39d397 CM |
6251 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
6252 | struct btrfs_inode_item); | |
b159fa28 | 6253 | memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item, |
293f7e07 | 6254 | sizeof(*inode_item)); |
e02119d5 | 6255 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d | 6256 | |
ef3b9af5 FM |
6257 | if (name) { |
6258 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
6259 | struct btrfs_inode_ref); | |
6260 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
6261 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); | |
6262 | ptr = (unsigned long)(ref + 1); | |
6263 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
6264 | } | |
9c58309d | 6265 | |
5f39d397 CM |
6266 | btrfs_mark_buffer_dirty(path->nodes[0]); |
6267 | btrfs_free_path(path); | |
6268 | ||
6cbff00f CH |
6269 | btrfs_inherit_iflags(inode, dir); |
6270 | ||
569254b0 | 6271 | if (S_ISREG(mode)) { |
0b246afa | 6272 | if (btrfs_test_opt(fs_info, NODATASUM)) |
94272164 | 6273 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; |
0b246afa | 6274 | if (btrfs_test_opt(fs_info, NODATACOW)) |
f2bdf9a8 JB |
6275 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW | |
6276 | BTRFS_INODE_NODATASUM; | |
94272164 CM |
6277 | } |
6278 | ||
5d4f98a2 | 6279 | inode_tree_add(inode); |
1abe9b8a | 6280 | |
6281 | trace_btrfs_inode_new(inode); | |
1973f0fa | 6282 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 6283 | |
8ea05e3a AB |
6284 | btrfs_update_root_times(trans, root); |
6285 | ||
63541927 FDBM |
6286 | ret = btrfs_inode_inherit_props(trans, inode, dir); |
6287 | if (ret) | |
0b246afa | 6288 | btrfs_err(fs_info, |
63541927 | 6289 | "error inheriting props for ino %llu (root %llu): %d", |
f85b7379 | 6290 | btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret); |
63541927 | 6291 | |
39279cc3 | 6292 | return inode; |
b0d5d10f CM |
6293 | |
6294 | fail_unlock: | |
6295 | unlock_new_inode(inode); | |
5f39d397 | 6296 | fail: |
ef3b9af5 | 6297 | if (dir && name) |
aec7477b | 6298 | BTRFS_I(dir)->index_cnt--; |
5f39d397 | 6299 | btrfs_free_path(path); |
09771430 | 6300 | iput(inode); |
5f39d397 | 6301 | return ERR_PTR(ret); |
39279cc3 CM |
6302 | } |
6303 | ||
6304 | static inline u8 btrfs_inode_type(struct inode *inode) | |
6305 | { | |
6306 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
6307 | } | |
6308 | ||
d352ac68 CM |
6309 | /* |
6310 | * utility function to add 'inode' into 'parent_inode' with | |
6311 | * a give name and a given sequence number. | |
6312 | * if 'add_backref' is true, also insert a backref from the | |
6313 | * inode to the parent directory. | |
6314 | */ | |
e02119d5 CM |
6315 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
6316 | struct inode *parent_inode, struct inode *inode, | |
6317 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 6318 | { |
0b246afa | 6319 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4df27c4d | 6320 | int ret = 0; |
39279cc3 | 6321 | struct btrfs_key key; |
e02119d5 | 6322 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
4a0cc7ca NB |
6323 | u64 ino = btrfs_ino(BTRFS_I(inode)); |
6324 | u64 parent_ino = btrfs_ino(BTRFS_I(parent_inode)); | |
5f39d397 | 6325 | |
33345d01 | 6326 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
6327 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); |
6328 | } else { | |
33345d01 | 6329 | key.objectid = ino; |
962a298f | 6330 | key.type = BTRFS_INODE_ITEM_KEY; |
4df27c4d YZ |
6331 | key.offset = 0; |
6332 | } | |
6333 | ||
33345d01 | 6334 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
0b246afa JM |
6335 | ret = btrfs_add_root_ref(trans, fs_info, key.objectid, |
6336 | root->root_key.objectid, parent_ino, | |
6337 | index, name, name_len); | |
4df27c4d | 6338 | } else if (add_backref) { |
33345d01 LZ |
6339 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
6340 | parent_ino, index); | |
4df27c4d | 6341 | } |
39279cc3 | 6342 | |
79787eaa JM |
6343 | /* Nothing to clean up yet */ |
6344 | if (ret) | |
6345 | return ret; | |
4df27c4d | 6346 | |
79787eaa JM |
6347 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
6348 | parent_inode, &key, | |
6349 | btrfs_inode_type(inode), index); | |
9c52057c | 6350 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
6351 | goto fail_dir_item; |
6352 | else if (ret) { | |
66642832 | 6353 | btrfs_abort_transaction(trans, ret); |
79787eaa | 6354 | return ret; |
39279cc3 | 6355 | } |
79787eaa JM |
6356 | |
6357 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
6358 | name_len * 2); | |
0c4d2d95 | 6359 | inode_inc_iversion(parent_inode); |
04b285f3 | 6360 | parent_inode->i_mtime = parent_inode->i_ctime = |
c2050a45 | 6361 | current_time(parent_inode); |
79787eaa JM |
6362 | ret = btrfs_update_inode(trans, root, parent_inode); |
6363 | if (ret) | |
66642832 | 6364 | btrfs_abort_transaction(trans, ret); |
39279cc3 | 6365 | return ret; |
fe66a05a CM |
6366 | |
6367 | fail_dir_item: | |
6368 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
6369 | u64 local_index; | |
6370 | int err; | |
0b246afa JM |
6371 | err = btrfs_del_root_ref(trans, fs_info, key.objectid, |
6372 | root->root_key.objectid, parent_ino, | |
6373 | &local_index, name, name_len); | |
fe66a05a CM |
6374 | |
6375 | } else if (add_backref) { | |
6376 | u64 local_index; | |
6377 | int err; | |
6378 | ||
6379 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
6380 | ino, parent_ino, &local_index); | |
6381 | } | |
6382 | return ret; | |
39279cc3 CM |
6383 | } |
6384 | ||
6385 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
a1b075d2 JB |
6386 | struct inode *dir, struct dentry *dentry, |
6387 | struct inode *inode, int backref, u64 index) | |
39279cc3 | 6388 | { |
a1b075d2 JB |
6389 | int err = btrfs_add_link(trans, dir, inode, |
6390 | dentry->d_name.name, dentry->d_name.len, | |
6391 | backref, index); | |
39279cc3 CM |
6392 | if (err > 0) |
6393 | err = -EEXIST; | |
6394 | return err; | |
6395 | } | |
6396 | ||
618e21d5 | 6397 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 6398 | umode_t mode, dev_t rdev) |
618e21d5 | 6399 | { |
2ff7e61e | 6400 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
618e21d5 JB |
6401 | struct btrfs_trans_handle *trans; |
6402 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6403 | struct inode *inode = NULL; |
618e21d5 JB |
6404 | int err; |
6405 | int drop_inode = 0; | |
6406 | u64 objectid; | |
00e4e6b3 | 6407 | u64 index = 0; |
618e21d5 | 6408 | |
9ed74f2d JB |
6409 | /* |
6410 | * 2 for inode item and ref | |
6411 | * 2 for dir items | |
6412 | * 1 for xattr if selinux is on | |
6413 | */ | |
a22285a6 YZ |
6414 | trans = btrfs_start_transaction(root, 5); |
6415 | if (IS_ERR(trans)) | |
6416 | return PTR_ERR(trans); | |
1832a6d5 | 6417 | |
581bb050 LZ |
6418 | err = btrfs_find_free_ino(root, &objectid); |
6419 | if (err) | |
6420 | goto out_unlock; | |
6421 | ||
aec7477b | 6422 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6423 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6424 | mode, &index); | |
7cf96da3 TI |
6425 | if (IS_ERR(inode)) { |
6426 | err = PTR_ERR(inode); | |
618e21d5 | 6427 | goto out_unlock; |
7cf96da3 | 6428 | } |
618e21d5 | 6429 | |
ad19db71 CS |
6430 | /* |
6431 | * If the active LSM wants to access the inode during | |
6432 | * d_instantiate it needs these. Smack checks to see | |
6433 | * if the filesystem supports xattrs by looking at the | |
6434 | * ops vector. | |
6435 | */ | |
ad19db71 | 6436 | inode->i_op = &btrfs_special_inode_operations; |
b0d5d10f CM |
6437 | init_special_inode(inode, inode->i_mode, rdev); |
6438 | ||
6439 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
618e21d5 | 6440 | if (err) |
b0d5d10f CM |
6441 | goto out_unlock_inode; |
6442 | ||
6443 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); | |
6444 | if (err) { | |
6445 | goto out_unlock_inode; | |
6446 | } else { | |
1b4ab1bb | 6447 | btrfs_update_inode(trans, root, inode); |
b0d5d10f | 6448 | unlock_new_inode(inode); |
08c422c2 | 6449 | d_instantiate(dentry, inode); |
618e21d5 | 6450 | } |
b0d5d10f | 6451 | |
618e21d5 | 6452 | out_unlock: |
3a45bb20 | 6453 | btrfs_end_transaction(trans); |
2ff7e61e JM |
6454 | btrfs_balance_delayed_items(fs_info); |
6455 | btrfs_btree_balance_dirty(fs_info); | |
618e21d5 JB |
6456 | if (drop_inode) { |
6457 | inode_dec_link_count(inode); | |
6458 | iput(inode); | |
6459 | } | |
618e21d5 | 6460 | return err; |
b0d5d10f CM |
6461 | |
6462 | out_unlock_inode: | |
6463 | drop_inode = 1; | |
6464 | unlock_new_inode(inode); | |
6465 | goto out_unlock; | |
6466 | ||
618e21d5 JB |
6467 | } |
6468 | ||
39279cc3 | 6469 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 6470 | umode_t mode, bool excl) |
39279cc3 | 6471 | { |
2ff7e61e | 6472 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
6473 | struct btrfs_trans_handle *trans; |
6474 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6475 | struct inode *inode = NULL; |
43baa579 | 6476 | int drop_inode_on_err = 0; |
a22285a6 | 6477 | int err; |
39279cc3 | 6478 | u64 objectid; |
00e4e6b3 | 6479 | u64 index = 0; |
39279cc3 | 6480 | |
9ed74f2d JB |
6481 | /* |
6482 | * 2 for inode item and ref | |
6483 | * 2 for dir items | |
6484 | * 1 for xattr if selinux is on | |
6485 | */ | |
a22285a6 YZ |
6486 | trans = btrfs_start_transaction(root, 5); |
6487 | if (IS_ERR(trans)) | |
6488 | return PTR_ERR(trans); | |
9ed74f2d | 6489 | |
581bb050 LZ |
6490 | err = btrfs_find_free_ino(root, &objectid); |
6491 | if (err) | |
6492 | goto out_unlock; | |
6493 | ||
aec7477b | 6494 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6495 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6496 | mode, &index); | |
7cf96da3 TI |
6497 | if (IS_ERR(inode)) { |
6498 | err = PTR_ERR(inode); | |
39279cc3 | 6499 | goto out_unlock; |
7cf96da3 | 6500 | } |
43baa579 | 6501 | drop_inode_on_err = 1; |
ad19db71 CS |
6502 | /* |
6503 | * If the active LSM wants to access the inode during | |
6504 | * d_instantiate it needs these. Smack checks to see | |
6505 | * if the filesystem supports xattrs by looking at the | |
6506 | * ops vector. | |
6507 | */ | |
6508 | inode->i_fop = &btrfs_file_operations; | |
6509 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 6510 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
6511 | |
6512 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
6513 | if (err) | |
6514 | goto out_unlock_inode; | |
6515 | ||
6516 | err = btrfs_update_inode(trans, root, inode); | |
6517 | if (err) | |
6518 | goto out_unlock_inode; | |
ad19db71 | 6519 | |
a1b075d2 | 6520 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 | 6521 | if (err) |
b0d5d10f | 6522 | goto out_unlock_inode; |
43baa579 | 6523 | |
43baa579 | 6524 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
b0d5d10f | 6525 | unlock_new_inode(inode); |
43baa579 FB |
6526 | d_instantiate(dentry, inode); |
6527 | ||
39279cc3 | 6528 | out_unlock: |
3a45bb20 | 6529 | btrfs_end_transaction(trans); |
43baa579 | 6530 | if (err && drop_inode_on_err) { |
39279cc3 CM |
6531 | inode_dec_link_count(inode); |
6532 | iput(inode); | |
6533 | } | |
2ff7e61e JM |
6534 | btrfs_balance_delayed_items(fs_info); |
6535 | btrfs_btree_balance_dirty(fs_info); | |
39279cc3 | 6536 | return err; |
b0d5d10f CM |
6537 | |
6538 | out_unlock_inode: | |
6539 | unlock_new_inode(inode); | |
6540 | goto out_unlock; | |
6541 | ||
39279cc3 CM |
6542 | } |
6543 | ||
6544 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
6545 | struct dentry *dentry) | |
6546 | { | |
271dba45 | 6547 | struct btrfs_trans_handle *trans = NULL; |
39279cc3 | 6548 | struct btrfs_root *root = BTRFS_I(dir)->root; |
2b0143b5 | 6549 | struct inode *inode = d_inode(old_dentry); |
2ff7e61e | 6550 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
00e4e6b3 | 6551 | u64 index; |
39279cc3 CM |
6552 | int err; |
6553 | int drop_inode = 0; | |
6554 | ||
4a8be425 TH |
6555 | /* do not allow sys_link's with other subvols of the same device */ |
6556 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 6557 | return -EXDEV; |
4a8be425 | 6558 | |
f186373f | 6559 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 6560 | return -EMLINK; |
4a8be425 | 6561 | |
3de4586c | 6562 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
6563 | if (err) |
6564 | goto fail; | |
6565 | ||
a22285a6 | 6566 | /* |
7e6b6465 | 6567 | * 2 items for inode and inode ref |
a22285a6 | 6568 | * 2 items for dir items |
7e6b6465 | 6569 | * 1 item for parent inode |
a22285a6 | 6570 | */ |
7e6b6465 | 6571 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
6572 | if (IS_ERR(trans)) { |
6573 | err = PTR_ERR(trans); | |
271dba45 | 6574 | trans = NULL; |
a22285a6 YZ |
6575 | goto fail; |
6576 | } | |
5f39d397 | 6577 | |
67de1176 MX |
6578 | /* There are several dir indexes for this inode, clear the cache. */ |
6579 | BTRFS_I(inode)->dir_index = 0ULL; | |
8b558c5f | 6580 | inc_nlink(inode); |
0c4d2d95 | 6581 | inode_inc_iversion(inode); |
c2050a45 | 6582 | inode->i_ctime = current_time(inode); |
7de9c6ee | 6583 | ihold(inode); |
e9976151 | 6584 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 6585 | |
a1b075d2 | 6586 | err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index); |
5f39d397 | 6587 | |
a5719521 | 6588 | if (err) { |
54aa1f4d | 6589 | drop_inode = 1; |
a5719521 | 6590 | } else { |
10d9f309 | 6591 | struct dentry *parent = dentry->d_parent; |
a5719521 | 6592 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
6593 | if (err) |
6594 | goto fail; | |
ef3b9af5 FM |
6595 | if (inode->i_nlink == 1) { |
6596 | /* | |
6597 | * If new hard link count is 1, it's a file created | |
6598 | * with open(2) O_TMPFILE flag. | |
6599 | */ | |
6600 | err = btrfs_orphan_del(trans, inode); | |
6601 | if (err) | |
6602 | goto fail; | |
6603 | } | |
08c422c2 | 6604 | d_instantiate(dentry, inode); |
9ca5fbfb | 6605 | btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent); |
a5719521 | 6606 | } |
39279cc3 | 6607 | |
2ff7e61e | 6608 | btrfs_balance_delayed_items(fs_info); |
1832a6d5 | 6609 | fail: |
271dba45 | 6610 | if (trans) |
3a45bb20 | 6611 | btrfs_end_transaction(trans); |
39279cc3 CM |
6612 | if (drop_inode) { |
6613 | inode_dec_link_count(inode); | |
6614 | iput(inode); | |
6615 | } | |
2ff7e61e | 6616 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 CM |
6617 | return err; |
6618 | } | |
6619 | ||
18bb1db3 | 6620 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 6621 | { |
2ff7e61e | 6622 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
b9d86667 | 6623 | struct inode *inode = NULL; |
39279cc3 CM |
6624 | struct btrfs_trans_handle *trans; |
6625 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
6626 | int err = 0; | |
6627 | int drop_on_err = 0; | |
b9d86667 | 6628 | u64 objectid = 0; |
00e4e6b3 | 6629 | u64 index = 0; |
39279cc3 | 6630 | |
9ed74f2d JB |
6631 | /* |
6632 | * 2 items for inode and ref | |
6633 | * 2 items for dir items | |
6634 | * 1 for xattr if selinux is on | |
6635 | */ | |
a22285a6 YZ |
6636 | trans = btrfs_start_transaction(root, 5); |
6637 | if (IS_ERR(trans)) | |
6638 | return PTR_ERR(trans); | |
39279cc3 | 6639 | |
581bb050 LZ |
6640 | err = btrfs_find_free_ino(root, &objectid); |
6641 | if (err) | |
6642 | goto out_fail; | |
6643 | ||
aec7477b | 6644 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
6645 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid, |
6646 | S_IFDIR | mode, &index); | |
39279cc3 CM |
6647 | if (IS_ERR(inode)) { |
6648 | err = PTR_ERR(inode); | |
6649 | goto out_fail; | |
6650 | } | |
5f39d397 | 6651 | |
39279cc3 | 6652 | drop_on_err = 1; |
b0d5d10f CM |
6653 | /* these must be set before we unlock the inode */ |
6654 | inode->i_op = &btrfs_dir_inode_operations; | |
6655 | inode->i_fop = &btrfs_dir_file_operations; | |
33268eaf | 6656 | |
2a7dba39 | 6657 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf | 6658 | if (err) |
b0d5d10f | 6659 | goto out_fail_inode; |
39279cc3 | 6660 | |
dbe674a9 | 6661 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
6662 | err = btrfs_update_inode(trans, root, inode); |
6663 | if (err) | |
b0d5d10f | 6664 | goto out_fail_inode; |
5f39d397 | 6665 | |
a1b075d2 JB |
6666 | err = btrfs_add_link(trans, dir, inode, dentry->d_name.name, |
6667 | dentry->d_name.len, 0, index); | |
39279cc3 | 6668 | if (err) |
b0d5d10f | 6669 | goto out_fail_inode; |
5f39d397 | 6670 | |
39279cc3 | 6671 | d_instantiate(dentry, inode); |
b0d5d10f CM |
6672 | /* |
6673 | * mkdir is special. We're unlocking after we call d_instantiate | |
6674 | * to avoid a race with nfsd calling d_instantiate. | |
6675 | */ | |
6676 | unlock_new_inode(inode); | |
39279cc3 | 6677 | drop_on_err = 0; |
39279cc3 CM |
6678 | |
6679 | out_fail: | |
3a45bb20 | 6680 | btrfs_end_transaction(trans); |
c7cfb8a5 WS |
6681 | if (drop_on_err) { |
6682 | inode_dec_link_count(inode); | |
39279cc3 | 6683 | iput(inode); |
c7cfb8a5 | 6684 | } |
2ff7e61e JM |
6685 | btrfs_balance_delayed_items(fs_info); |
6686 | btrfs_btree_balance_dirty(fs_info); | |
39279cc3 | 6687 | return err; |
b0d5d10f CM |
6688 | |
6689 | out_fail_inode: | |
6690 | unlock_new_inode(inode); | |
6691 | goto out_fail; | |
39279cc3 CM |
6692 | } |
6693 | ||
e6c4efd8 QW |
6694 | /* Find next extent map of a given extent map, caller needs to ensure locks */ |
6695 | static struct extent_map *next_extent_map(struct extent_map *em) | |
6696 | { | |
6697 | struct rb_node *next; | |
6698 | ||
6699 | next = rb_next(&em->rb_node); | |
6700 | if (!next) | |
6701 | return NULL; | |
6702 | return container_of(next, struct extent_map, rb_node); | |
6703 | } | |
6704 | ||
6705 | static struct extent_map *prev_extent_map(struct extent_map *em) | |
6706 | { | |
6707 | struct rb_node *prev; | |
6708 | ||
6709 | prev = rb_prev(&em->rb_node); | |
6710 | if (!prev) | |
6711 | return NULL; | |
6712 | return container_of(prev, struct extent_map, rb_node); | |
6713 | } | |
6714 | ||
d352ac68 | 6715 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
e6c4efd8 | 6716 | * the existing extent is the nearest extent to map_start, |
d352ac68 | 6717 | * and an extent that you want to insert, deal with overlap and insert |
e6c4efd8 | 6718 | * the best fitted new extent into the tree. |
d352ac68 | 6719 | */ |
3b951516 CM |
6720 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
6721 | struct extent_map *existing, | |
e6dcd2dc | 6722 | struct extent_map *em, |
51f395ad | 6723 | u64 map_start) |
3b951516 | 6724 | { |
e6c4efd8 QW |
6725 | struct extent_map *prev; |
6726 | struct extent_map *next; | |
6727 | u64 start; | |
6728 | u64 end; | |
3b951516 | 6729 | u64 start_diff; |
3b951516 | 6730 | |
e6dcd2dc | 6731 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
e6c4efd8 QW |
6732 | |
6733 | if (existing->start > map_start) { | |
6734 | next = existing; | |
6735 | prev = prev_extent_map(next); | |
6736 | } else { | |
6737 | prev = existing; | |
6738 | next = next_extent_map(prev); | |
6739 | } | |
6740 | ||
6741 | start = prev ? extent_map_end(prev) : em->start; | |
6742 | start = max_t(u64, start, em->start); | |
6743 | end = next ? next->start : extent_map_end(em); | |
6744 | end = min_t(u64, end, extent_map_end(em)); | |
6745 | start_diff = start - em->start; | |
6746 | em->start = start; | |
6747 | em->len = end - start; | |
c8b97818 CM |
6748 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
6749 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 6750 | em->block_start += start_diff; |
c8b97818 CM |
6751 | em->block_len -= start_diff; |
6752 | } | |
09a2a8f9 | 6753 | return add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
6754 | } |
6755 | ||
c8b97818 | 6756 | static noinline int uncompress_inline(struct btrfs_path *path, |
e40da0e5 | 6757 | struct page *page, |
c8b97818 CM |
6758 | size_t pg_offset, u64 extent_offset, |
6759 | struct btrfs_file_extent_item *item) | |
6760 | { | |
6761 | int ret; | |
6762 | struct extent_buffer *leaf = path->nodes[0]; | |
6763 | char *tmp; | |
6764 | size_t max_size; | |
6765 | unsigned long inline_size; | |
6766 | unsigned long ptr; | |
261507a0 | 6767 | int compress_type; |
c8b97818 CM |
6768 | |
6769 | WARN_ON(pg_offset != 0); | |
261507a0 | 6770 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
6771 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
6772 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
dd3cc16b | 6773 | btrfs_item_nr(path->slots[0])); |
c8b97818 | 6774 | tmp = kmalloc(inline_size, GFP_NOFS); |
8d413713 TI |
6775 | if (!tmp) |
6776 | return -ENOMEM; | |
c8b97818 CM |
6777 | ptr = btrfs_file_extent_inline_start(item); |
6778 | ||
6779 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
6780 | ||
09cbfeaf | 6781 | max_size = min_t(unsigned long, PAGE_SIZE, max_size); |
261507a0 LZ |
6782 | ret = btrfs_decompress(compress_type, tmp, page, |
6783 | extent_offset, inline_size, max_size); | |
c8b97818 | 6784 | kfree(tmp); |
166ae5a4 | 6785 | return ret; |
c8b97818 CM |
6786 | } |
6787 | ||
d352ac68 CM |
6788 | /* |
6789 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
6790 | * the ugly parts come from merging extents from the disk with the in-ram |
6791 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
6792 | * where the in-ram extents might be locked pending data=ordered completion. |
6793 | * | |
6794 | * This also copies inline extents directly into the page. | |
6795 | */ | |
d397712b | 6796 | |
a52d9a80 | 6797 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 6798 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
6799 | int create) |
6800 | { | |
0b246afa | 6801 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
a52d9a80 CM |
6802 | int ret; |
6803 | int err = 0; | |
a52d9a80 CM |
6804 | u64 extent_start = 0; |
6805 | u64 extent_end = 0; | |
4a0cc7ca | 6806 | u64 objectid = btrfs_ino(BTRFS_I(inode)); |
a52d9a80 | 6807 | u32 found_type; |
f421950f | 6808 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
6809 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6810 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
6811 | struct extent_buffer *leaf; |
6812 | struct btrfs_key found_key; | |
a52d9a80 CM |
6813 | struct extent_map *em = NULL; |
6814 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 6815 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 6816 | struct btrfs_trans_handle *trans = NULL; |
7ffbb598 | 6817 | const bool new_inline = !page || create; |
a52d9a80 | 6818 | |
a52d9a80 | 6819 | again: |
890871be | 6820 | read_lock(&em_tree->lock); |
d1310b2e | 6821 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d | 6822 | if (em) |
0b246afa | 6823 | em->bdev = fs_info->fs_devices->latest_bdev; |
890871be | 6824 | read_unlock(&em_tree->lock); |
d1310b2e | 6825 | |
a52d9a80 | 6826 | if (em) { |
e1c4b745 CM |
6827 | if (em->start > start || em->start + em->len <= start) |
6828 | free_extent_map(em); | |
6829 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
6830 | free_extent_map(em); |
6831 | else | |
6832 | goto out; | |
a52d9a80 | 6833 | } |
172ddd60 | 6834 | em = alloc_extent_map(); |
a52d9a80 | 6835 | if (!em) { |
d1310b2e CM |
6836 | err = -ENOMEM; |
6837 | goto out; | |
a52d9a80 | 6838 | } |
0b246afa | 6839 | em->bdev = fs_info->fs_devices->latest_bdev; |
d1310b2e | 6840 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 6841 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 6842 | em->len = (u64)-1; |
c8b97818 | 6843 | em->block_len = (u64)-1; |
f421950f CM |
6844 | |
6845 | if (!path) { | |
6846 | path = btrfs_alloc_path(); | |
026fd317 JB |
6847 | if (!path) { |
6848 | err = -ENOMEM; | |
6849 | goto out; | |
6850 | } | |
6851 | /* | |
6852 | * Chances are we'll be called again, so go ahead and do | |
6853 | * readahead | |
6854 | */ | |
e4058b54 | 6855 | path->reada = READA_FORWARD; |
f421950f CM |
6856 | } |
6857 | ||
179e29e4 CM |
6858 | ret = btrfs_lookup_file_extent(trans, root, path, |
6859 | objectid, start, trans != NULL); | |
a52d9a80 CM |
6860 | if (ret < 0) { |
6861 | err = ret; | |
6862 | goto out; | |
6863 | } | |
6864 | ||
6865 | if (ret != 0) { | |
6866 | if (path->slots[0] == 0) | |
6867 | goto not_found; | |
6868 | path->slots[0]--; | |
6869 | } | |
6870 | ||
5f39d397 CM |
6871 | leaf = path->nodes[0]; |
6872 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 6873 | struct btrfs_file_extent_item); |
a52d9a80 | 6874 | /* are we inside the extent that was found? */ |
5f39d397 | 6875 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
962a298f | 6876 | found_type = found_key.type; |
5f39d397 | 6877 | if (found_key.objectid != objectid || |
a52d9a80 | 6878 | found_type != BTRFS_EXTENT_DATA_KEY) { |
25a50341 JB |
6879 | /* |
6880 | * If we backup past the first extent we want to move forward | |
6881 | * and see if there is an extent in front of us, otherwise we'll | |
6882 | * say there is a hole for our whole search range which can | |
6883 | * cause problems. | |
6884 | */ | |
6885 | extent_end = start; | |
6886 | goto next; | |
a52d9a80 CM |
6887 | } |
6888 | ||
5f39d397 CM |
6889 | found_type = btrfs_file_extent_type(leaf, item); |
6890 | extent_start = found_key.offset; | |
d899e052 YZ |
6891 | if (found_type == BTRFS_FILE_EXTENT_REG || |
6892 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 6893 | extent_end = extent_start + |
db94535d | 6894 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
6895 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
6896 | size_t size; | |
514ac8ad | 6897 | size = btrfs_file_extent_inline_len(leaf, path->slots[0], item); |
da17066c | 6898 | extent_end = ALIGN(extent_start + size, |
0b246afa | 6899 | fs_info->sectorsize); |
9036c102 | 6900 | } |
25a50341 | 6901 | next: |
9036c102 YZ |
6902 | if (start >= extent_end) { |
6903 | path->slots[0]++; | |
6904 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
6905 | ret = btrfs_next_leaf(root, path); | |
6906 | if (ret < 0) { | |
6907 | err = ret; | |
6908 | goto out; | |
a52d9a80 | 6909 | } |
9036c102 YZ |
6910 | if (ret > 0) |
6911 | goto not_found; | |
6912 | leaf = path->nodes[0]; | |
a52d9a80 | 6913 | } |
9036c102 YZ |
6914 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
6915 | if (found_key.objectid != objectid || | |
6916 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
6917 | goto not_found; | |
6918 | if (start + len <= found_key.offset) | |
6919 | goto not_found; | |
e2eca69d WS |
6920 | if (start > found_key.offset) |
6921 | goto next; | |
9036c102 | 6922 | em->start = start; |
70c8a91c | 6923 | em->orig_start = start; |
9036c102 YZ |
6924 | em->len = found_key.offset - start; |
6925 | goto not_found_em; | |
6926 | } | |
6927 | ||
7ffbb598 FM |
6928 | btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em); |
6929 | ||
d899e052 YZ |
6930 | if (found_type == BTRFS_FILE_EXTENT_REG || |
6931 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 CM |
6932 | goto insert; |
6933 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 6934 | unsigned long ptr; |
a52d9a80 | 6935 | char *map; |
3326d1b0 CM |
6936 | size_t size; |
6937 | size_t extent_offset; | |
6938 | size_t copy_size; | |
a52d9a80 | 6939 | |
7ffbb598 | 6940 | if (new_inline) |
689f9346 | 6941 | goto out; |
5f39d397 | 6942 | |
514ac8ad | 6943 | size = btrfs_file_extent_inline_len(leaf, path->slots[0], item); |
9036c102 | 6944 | extent_offset = page_offset(page) + pg_offset - extent_start; |
09cbfeaf KS |
6945 | copy_size = min_t(u64, PAGE_SIZE - pg_offset, |
6946 | size - extent_offset); | |
3326d1b0 | 6947 | em->start = extent_start + extent_offset; |
0b246afa | 6948 | em->len = ALIGN(copy_size, fs_info->sectorsize); |
b4939680 | 6949 | em->orig_block_len = em->len; |
70c8a91c | 6950 | em->orig_start = em->start; |
689f9346 | 6951 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 6952 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
6953 | if (btrfs_file_extent_compression(leaf, item) != |
6954 | BTRFS_COMPRESS_NONE) { | |
e40da0e5 | 6955 | ret = uncompress_inline(path, page, pg_offset, |
c8b97818 | 6956 | extent_offset, item); |
166ae5a4 ZB |
6957 | if (ret) { |
6958 | err = ret; | |
6959 | goto out; | |
6960 | } | |
c8b97818 CM |
6961 | } else { |
6962 | map = kmap(page); | |
6963 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
6964 | copy_size); | |
09cbfeaf | 6965 | if (pg_offset + copy_size < PAGE_SIZE) { |
93c82d57 | 6966 | memset(map + pg_offset + copy_size, 0, |
09cbfeaf | 6967 | PAGE_SIZE - pg_offset - |
93c82d57 CM |
6968 | copy_size); |
6969 | } | |
c8b97818 CM |
6970 | kunmap(page); |
6971 | } | |
179e29e4 CM |
6972 | flush_dcache_page(page); |
6973 | } else if (create && PageUptodate(page)) { | |
6bf7e080 | 6974 | BUG(); |
179e29e4 CM |
6975 | if (!trans) { |
6976 | kunmap(page); | |
6977 | free_extent_map(em); | |
6978 | em = NULL; | |
ff5714cc | 6979 | |
b3b4aa74 | 6980 | btrfs_release_path(path); |
7a7eaa40 | 6981 | trans = btrfs_join_transaction(root); |
ff5714cc | 6982 | |
3612b495 TI |
6983 | if (IS_ERR(trans)) |
6984 | return ERR_CAST(trans); | |
179e29e4 CM |
6985 | goto again; |
6986 | } | |
c8b97818 | 6987 | map = kmap(page); |
70dec807 | 6988 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 6989 | copy_size); |
c8b97818 | 6990 | kunmap(page); |
179e29e4 | 6991 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 6992 | } |
d1310b2e | 6993 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 6994 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 | 6995 | goto insert; |
a52d9a80 CM |
6996 | } |
6997 | not_found: | |
6998 | em->start = start; | |
70c8a91c | 6999 | em->orig_start = start; |
d1310b2e | 7000 | em->len = len; |
a52d9a80 | 7001 | not_found_em: |
5f39d397 | 7002 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 7003 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 | 7004 | insert: |
b3b4aa74 | 7005 | btrfs_release_path(path); |
d1310b2e | 7006 | if (em->start > start || extent_map_end(em) <= start) { |
0b246afa | 7007 | btrfs_err(fs_info, |
5d163e0e JM |
7008 | "bad extent! em: [%llu %llu] passed [%llu %llu]", |
7009 | em->start, em->len, start, len); | |
a52d9a80 CM |
7010 | err = -EIO; |
7011 | goto out; | |
7012 | } | |
d1310b2e CM |
7013 | |
7014 | err = 0; | |
890871be | 7015 | write_lock(&em_tree->lock); |
09a2a8f9 | 7016 | ret = add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
7017 | /* it is possible that someone inserted the extent into the tree |
7018 | * while we had the lock dropped. It is also possible that | |
7019 | * an overlapping map exists in the tree | |
7020 | */ | |
a52d9a80 | 7021 | if (ret == -EEXIST) { |
3b951516 | 7022 | struct extent_map *existing; |
e6dcd2dc CM |
7023 | |
7024 | ret = 0; | |
7025 | ||
e6c4efd8 QW |
7026 | existing = search_extent_mapping(em_tree, start, len); |
7027 | /* | |
7028 | * existing will always be non-NULL, since there must be | |
7029 | * extent causing the -EEXIST. | |
7030 | */ | |
8dff9c85 | 7031 | if (existing->start == em->start && |
8e2bd3b7 | 7032 | extent_map_end(existing) >= extent_map_end(em) && |
8dff9c85 CM |
7033 | em->block_start == existing->block_start) { |
7034 | /* | |
8e2bd3b7 OS |
7035 | * The existing extent map already encompasses the |
7036 | * entire extent map we tried to add. | |
8dff9c85 CM |
7037 | */ |
7038 | free_extent_map(em); | |
7039 | em = existing; | |
7040 | err = 0; | |
7041 | ||
7042 | } else if (start >= extent_map_end(existing) || | |
32be3a1a | 7043 | start <= existing->start) { |
e6c4efd8 QW |
7044 | /* |
7045 | * The existing extent map is the one nearest to | |
7046 | * the [start, start + len) range which overlaps | |
7047 | */ | |
7048 | err = merge_extent_mapping(em_tree, existing, | |
7049 | em, start); | |
e1c4b745 | 7050 | free_extent_map(existing); |
e6c4efd8 | 7051 | if (err) { |
3b951516 CM |
7052 | free_extent_map(em); |
7053 | em = NULL; | |
7054 | } | |
7055 | } else { | |
7056 | free_extent_map(em); | |
7057 | em = existing; | |
e6dcd2dc | 7058 | err = 0; |
a52d9a80 | 7059 | } |
a52d9a80 | 7060 | } |
890871be | 7061 | write_unlock(&em_tree->lock); |
a52d9a80 | 7062 | out: |
1abe9b8a | 7063 | |
4a0cc7ca | 7064 | trace_btrfs_get_extent(root, BTRFS_I(inode), em); |
1abe9b8a | 7065 | |
527afb44 | 7066 | btrfs_free_path(path); |
a52d9a80 | 7067 | if (trans) { |
3a45bb20 | 7068 | ret = btrfs_end_transaction(trans); |
d397712b | 7069 | if (!err) |
a52d9a80 CM |
7070 | err = ret; |
7071 | } | |
a52d9a80 CM |
7072 | if (err) { |
7073 | free_extent_map(em); | |
a52d9a80 CM |
7074 | return ERR_PTR(err); |
7075 | } | |
79787eaa | 7076 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
7077 | return em; |
7078 | } | |
7079 | ||
ec29ed5b CM |
7080 | struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page, |
7081 | size_t pg_offset, u64 start, u64 len, | |
7082 | int create) | |
7083 | { | |
7084 | struct extent_map *em; | |
7085 | struct extent_map *hole_em = NULL; | |
7086 | u64 range_start = start; | |
7087 | u64 end; | |
7088 | u64 found; | |
7089 | u64 found_end; | |
7090 | int err = 0; | |
7091 | ||
7092 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
7093 | if (IS_ERR(em)) | |
7094 | return em; | |
7095 | if (em) { | |
7096 | /* | |
f9e4fb53 LB |
7097 | * if our em maps to |
7098 | * - a hole or | |
7099 | * - a pre-alloc extent, | |
7100 | * there might actually be delalloc bytes behind it. | |
ec29ed5b | 7101 | */ |
f9e4fb53 LB |
7102 | if (em->block_start != EXTENT_MAP_HOLE && |
7103 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
ec29ed5b CM |
7104 | return em; |
7105 | else | |
7106 | hole_em = em; | |
7107 | } | |
7108 | ||
7109 | /* check to see if we've wrapped (len == -1 or similar) */ | |
7110 | end = start + len; | |
7111 | if (end < start) | |
7112 | end = (u64)-1; | |
7113 | else | |
7114 | end -= 1; | |
7115 | ||
7116 | em = NULL; | |
7117 | ||
7118 | /* ok, we didn't find anything, lets look for delalloc */ | |
7119 | found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start, | |
7120 | end, len, EXTENT_DELALLOC, 1); | |
7121 | found_end = range_start + found; | |
7122 | if (found_end < range_start) | |
7123 | found_end = (u64)-1; | |
7124 | ||
7125 | /* | |
7126 | * we didn't find anything useful, return | |
7127 | * the original results from get_extent() | |
7128 | */ | |
7129 | if (range_start > end || found_end <= start) { | |
7130 | em = hole_em; | |
7131 | hole_em = NULL; | |
7132 | goto out; | |
7133 | } | |
7134 | ||
7135 | /* adjust the range_start to make sure it doesn't | |
7136 | * go backwards from the start they passed in | |
7137 | */ | |
67871254 | 7138 | range_start = max(start, range_start); |
ec29ed5b CM |
7139 | found = found_end - range_start; |
7140 | ||
7141 | if (found > 0) { | |
7142 | u64 hole_start = start; | |
7143 | u64 hole_len = len; | |
7144 | ||
172ddd60 | 7145 | em = alloc_extent_map(); |
ec29ed5b CM |
7146 | if (!em) { |
7147 | err = -ENOMEM; | |
7148 | goto out; | |
7149 | } | |
7150 | /* | |
7151 | * when btrfs_get_extent can't find anything it | |
7152 | * returns one huge hole | |
7153 | * | |
7154 | * make sure what it found really fits our range, and | |
7155 | * adjust to make sure it is based on the start from | |
7156 | * the caller | |
7157 | */ | |
7158 | if (hole_em) { | |
7159 | u64 calc_end = extent_map_end(hole_em); | |
7160 | ||
7161 | if (calc_end <= start || (hole_em->start > end)) { | |
7162 | free_extent_map(hole_em); | |
7163 | hole_em = NULL; | |
7164 | } else { | |
7165 | hole_start = max(hole_em->start, start); | |
7166 | hole_len = calc_end - hole_start; | |
7167 | } | |
7168 | } | |
7169 | em->bdev = NULL; | |
7170 | if (hole_em && range_start > hole_start) { | |
7171 | /* our hole starts before our delalloc, so we | |
7172 | * have to return just the parts of the hole | |
7173 | * that go until the delalloc starts | |
7174 | */ | |
7175 | em->len = min(hole_len, | |
7176 | range_start - hole_start); | |
7177 | em->start = hole_start; | |
7178 | em->orig_start = hole_start; | |
7179 | /* | |
7180 | * don't adjust block start at all, | |
7181 | * it is fixed at EXTENT_MAP_HOLE | |
7182 | */ | |
7183 | em->block_start = hole_em->block_start; | |
7184 | em->block_len = hole_len; | |
f9e4fb53 LB |
7185 | if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags)) |
7186 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
ec29ed5b CM |
7187 | } else { |
7188 | em->start = range_start; | |
7189 | em->len = found; | |
7190 | em->orig_start = range_start; | |
7191 | em->block_start = EXTENT_MAP_DELALLOC; | |
7192 | em->block_len = found; | |
7193 | } | |
7194 | } else if (hole_em) { | |
7195 | return hole_em; | |
7196 | } | |
7197 | out: | |
7198 | ||
7199 | free_extent_map(hole_em); | |
7200 | if (err) { | |
7201 | free_extent_map(em); | |
7202 | return ERR_PTR(err); | |
7203 | } | |
7204 | return em; | |
7205 | } | |
7206 | ||
5f9a8a51 FM |
7207 | static struct extent_map *btrfs_create_dio_extent(struct inode *inode, |
7208 | const u64 start, | |
7209 | const u64 len, | |
7210 | const u64 orig_start, | |
7211 | const u64 block_start, | |
7212 | const u64 block_len, | |
7213 | const u64 orig_block_len, | |
7214 | const u64 ram_bytes, | |
7215 | const int type) | |
7216 | { | |
7217 | struct extent_map *em = NULL; | |
7218 | int ret; | |
7219 | ||
5f9a8a51 FM |
7220 | if (type != BTRFS_ORDERED_NOCOW) { |
7221 | em = create_pinned_em(inode, start, len, orig_start, | |
7222 | block_start, block_len, orig_block_len, | |
7223 | ram_bytes, type); | |
7224 | if (IS_ERR(em)) | |
7225 | goto out; | |
7226 | } | |
7227 | ret = btrfs_add_ordered_extent_dio(inode, start, block_start, | |
7228 | len, block_len, type); | |
7229 | if (ret) { | |
7230 | if (em) { | |
7231 | free_extent_map(em); | |
7232 | btrfs_drop_extent_cache(inode, start, | |
7233 | start + len - 1, 0); | |
7234 | } | |
7235 | em = ERR_PTR(ret); | |
7236 | } | |
7237 | out: | |
5f9a8a51 FM |
7238 | |
7239 | return em; | |
7240 | } | |
7241 | ||
4b46fce2 JB |
7242 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
7243 | u64 start, u64 len) | |
7244 | { | |
0b246afa | 7245 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 7246 | struct btrfs_root *root = BTRFS_I(inode)->root; |
70c8a91c | 7247 | struct extent_map *em; |
4b46fce2 JB |
7248 | struct btrfs_key ins; |
7249 | u64 alloc_hint; | |
7250 | int ret; | |
4b46fce2 | 7251 | |
4b46fce2 | 7252 | alloc_hint = get_extent_allocation_hint(inode, start, len); |
0b246afa | 7253 | ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize, |
da17066c | 7254 | 0, alloc_hint, &ins, 1, 1); |
00361589 JB |
7255 | if (ret) |
7256 | return ERR_PTR(ret); | |
4b46fce2 | 7257 | |
5f9a8a51 FM |
7258 | em = btrfs_create_dio_extent(inode, start, ins.offset, start, |
7259 | ins.objectid, ins.offset, ins.offset, | |
7260 | ins.offset, 0); | |
0b246afa | 7261 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
5f9a8a51 | 7262 | if (IS_ERR(em)) |
2ff7e61e JM |
7263 | btrfs_free_reserved_extent(fs_info, ins.objectid, |
7264 | ins.offset, 1); | |
de0ee0ed | 7265 | |
4b46fce2 JB |
7266 | return em; |
7267 | } | |
7268 | ||
46bfbb5c CM |
7269 | /* |
7270 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
7271 | * block must be cow'd | |
7272 | */ | |
00361589 | 7273 | noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len, |
7ee9e440 JB |
7274 | u64 *orig_start, u64 *orig_block_len, |
7275 | u64 *ram_bytes) | |
46bfbb5c | 7276 | { |
2ff7e61e | 7277 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
00361589 | 7278 | struct btrfs_trans_handle *trans; |
46bfbb5c CM |
7279 | struct btrfs_path *path; |
7280 | int ret; | |
7281 | struct extent_buffer *leaf; | |
7282 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7b2b7085 | 7283 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
46bfbb5c CM |
7284 | struct btrfs_file_extent_item *fi; |
7285 | struct btrfs_key key; | |
7286 | u64 disk_bytenr; | |
7287 | u64 backref_offset; | |
7288 | u64 extent_end; | |
7289 | u64 num_bytes; | |
7290 | int slot; | |
7291 | int found_type; | |
7ee9e440 | 7292 | bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW); |
e77751aa | 7293 | |
46bfbb5c CM |
7294 | path = btrfs_alloc_path(); |
7295 | if (!path) | |
7296 | return -ENOMEM; | |
7297 | ||
f85b7379 DS |
7298 | ret = btrfs_lookup_file_extent(NULL, root, path, |
7299 | btrfs_ino(BTRFS_I(inode)), offset, 0); | |
46bfbb5c CM |
7300 | if (ret < 0) |
7301 | goto out; | |
7302 | ||
7303 | slot = path->slots[0]; | |
7304 | if (ret == 1) { | |
7305 | if (slot == 0) { | |
7306 | /* can't find the item, must cow */ | |
7307 | ret = 0; | |
7308 | goto out; | |
7309 | } | |
7310 | slot--; | |
7311 | } | |
7312 | ret = 0; | |
7313 | leaf = path->nodes[0]; | |
7314 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
4a0cc7ca | 7315 | if (key.objectid != btrfs_ino(BTRFS_I(inode)) || |
46bfbb5c CM |
7316 | key.type != BTRFS_EXTENT_DATA_KEY) { |
7317 | /* not our file or wrong item type, must cow */ | |
7318 | goto out; | |
7319 | } | |
7320 | ||
7321 | if (key.offset > offset) { | |
7322 | /* Wrong offset, must cow */ | |
7323 | goto out; | |
7324 | } | |
7325 | ||
7326 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
7327 | found_type = btrfs_file_extent_type(leaf, fi); | |
7328 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
7329 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
7330 | /* not a regular extent, must cow */ | |
7331 | goto out; | |
7332 | } | |
7ee9e440 JB |
7333 | |
7334 | if (!nocow && found_type == BTRFS_FILE_EXTENT_REG) | |
7335 | goto out; | |
7336 | ||
e77751aa MX |
7337 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
7338 | if (extent_end <= offset) | |
7339 | goto out; | |
7340 | ||
46bfbb5c | 7341 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
7ee9e440 JB |
7342 | if (disk_bytenr == 0) |
7343 | goto out; | |
7344 | ||
7345 | if (btrfs_file_extent_compression(leaf, fi) || | |
7346 | btrfs_file_extent_encryption(leaf, fi) || | |
7347 | btrfs_file_extent_other_encoding(leaf, fi)) | |
7348 | goto out; | |
7349 | ||
46bfbb5c CM |
7350 | backref_offset = btrfs_file_extent_offset(leaf, fi); |
7351 | ||
7ee9e440 JB |
7352 | if (orig_start) { |
7353 | *orig_start = key.offset - backref_offset; | |
7354 | *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
7355 | *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); | |
7356 | } | |
eb384b55 | 7357 | |
2ff7e61e | 7358 | if (btrfs_extent_readonly(fs_info, disk_bytenr)) |
46bfbb5c | 7359 | goto out; |
7b2b7085 MX |
7360 | |
7361 | num_bytes = min(offset + *len, extent_end) - offset; | |
7362 | if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
7363 | u64 range_end; | |
7364 | ||
da17066c JM |
7365 | range_end = round_up(offset + num_bytes, |
7366 | root->fs_info->sectorsize) - 1; | |
7b2b7085 MX |
7367 | ret = test_range_bit(io_tree, offset, range_end, |
7368 | EXTENT_DELALLOC, 0, NULL); | |
7369 | if (ret) { | |
7370 | ret = -EAGAIN; | |
7371 | goto out; | |
7372 | } | |
7373 | } | |
7374 | ||
1bda19eb | 7375 | btrfs_release_path(path); |
46bfbb5c CM |
7376 | |
7377 | /* | |
7378 | * look for other files referencing this extent, if we | |
7379 | * find any we must cow | |
7380 | */ | |
00361589 JB |
7381 | trans = btrfs_join_transaction(root); |
7382 | if (IS_ERR(trans)) { | |
7383 | ret = 0; | |
46bfbb5c | 7384 | goto out; |
00361589 JB |
7385 | } |
7386 | ||
4a0cc7ca | 7387 | ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(BTRFS_I(inode)), |
00361589 | 7388 | key.offset - backref_offset, disk_bytenr); |
3a45bb20 | 7389 | btrfs_end_transaction(trans); |
00361589 JB |
7390 | if (ret) { |
7391 | ret = 0; | |
7392 | goto out; | |
7393 | } | |
46bfbb5c CM |
7394 | |
7395 | /* | |
7396 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
7397 | * in this extent we are about to write. If there | |
7398 | * are any csums in that range we have to cow in order | |
7399 | * to keep the csums correct | |
7400 | */ | |
7401 | disk_bytenr += backref_offset; | |
7402 | disk_bytenr += offset - key.offset; | |
2ff7e61e JM |
7403 | if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes)) |
7404 | goto out; | |
46bfbb5c CM |
7405 | /* |
7406 | * all of the above have passed, it is safe to overwrite this extent | |
7407 | * without cow | |
7408 | */ | |
eb384b55 | 7409 | *len = num_bytes; |
46bfbb5c CM |
7410 | ret = 1; |
7411 | out: | |
7412 | btrfs_free_path(path); | |
7413 | return ret; | |
7414 | } | |
7415 | ||
fc4adbff AG |
7416 | bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end) |
7417 | { | |
7418 | struct radix_tree_root *root = &inode->i_mapping->page_tree; | |
7419 | int found = false; | |
7420 | void **pagep = NULL; | |
7421 | struct page *page = NULL; | |
7422 | int start_idx; | |
7423 | int end_idx; | |
7424 | ||
09cbfeaf | 7425 | start_idx = start >> PAGE_SHIFT; |
fc4adbff AG |
7426 | |
7427 | /* | |
7428 | * end is the last byte in the last page. end == start is legal | |
7429 | */ | |
09cbfeaf | 7430 | end_idx = end >> PAGE_SHIFT; |
fc4adbff AG |
7431 | |
7432 | rcu_read_lock(); | |
7433 | ||
7434 | /* Most of the code in this while loop is lifted from | |
7435 | * find_get_page. It's been modified to begin searching from a | |
7436 | * page and return just the first page found in that range. If the | |
7437 | * found idx is less than or equal to the end idx then we know that | |
7438 | * a page exists. If no pages are found or if those pages are | |
7439 | * outside of the range then we're fine (yay!) */ | |
7440 | while (page == NULL && | |
7441 | radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) { | |
7442 | page = radix_tree_deref_slot(pagep); | |
7443 | if (unlikely(!page)) | |
7444 | break; | |
7445 | ||
7446 | if (radix_tree_exception(page)) { | |
809f9016 FM |
7447 | if (radix_tree_deref_retry(page)) { |
7448 | page = NULL; | |
fc4adbff | 7449 | continue; |
809f9016 | 7450 | } |
fc4adbff AG |
7451 | /* |
7452 | * Otherwise, shmem/tmpfs must be storing a swap entry | |
7453 | * here as an exceptional entry: so return it without | |
7454 | * attempting to raise page count. | |
7455 | */ | |
6fdef6d4 | 7456 | page = NULL; |
fc4adbff AG |
7457 | break; /* TODO: Is this relevant for this use case? */ |
7458 | } | |
7459 | ||
91405151 FM |
7460 | if (!page_cache_get_speculative(page)) { |
7461 | page = NULL; | |
fc4adbff | 7462 | continue; |
91405151 | 7463 | } |
fc4adbff AG |
7464 | |
7465 | /* | |
7466 | * Has the page moved? | |
7467 | * This is part of the lockless pagecache protocol. See | |
7468 | * include/linux/pagemap.h for details. | |
7469 | */ | |
7470 | if (unlikely(page != *pagep)) { | |
09cbfeaf | 7471 | put_page(page); |
fc4adbff AG |
7472 | page = NULL; |
7473 | } | |
7474 | } | |
7475 | ||
7476 | if (page) { | |
7477 | if (page->index <= end_idx) | |
7478 | found = true; | |
09cbfeaf | 7479 | put_page(page); |
fc4adbff AG |
7480 | } |
7481 | ||
7482 | rcu_read_unlock(); | |
7483 | return found; | |
7484 | } | |
7485 | ||
eb838e73 JB |
7486 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
7487 | struct extent_state **cached_state, int writing) | |
7488 | { | |
7489 | struct btrfs_ordered_extent *ordered; | |
7490 | int ret = 0; | |
7491 | ||
7492 | while (1) { | |
7493 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
ff13db41 | 7494 | cached_state); |
eb838e73 JB |
7495 | /* |
7496 | * We're concerned with the entire range that we're going to be | |
01327610 | 7497 | * doing DIO to, so we need to make sure there's no ordered |
eb838e73 JB |
7498 | * extents in this range. |
7499 | */ | |
7500 | ordered = btrfs_lookup_ordered_range(inode, lockstart, | |
7501 | lockend - lockstart + 1); | |
7502 | ||
7503 | /* | |
7504 | * We need to make sure there are no buffered pages in this | |
7505 | * range either, we could have raced between the invalidate in | |
7506 | * generic_file_direct_write and locking the extent. The | |
7507 | * invalidate needs to happen so that reads after a write do not | |
7508 | * get stale data. | |
7509 | */ | |
fc4adbff AG |
7510 | if (!ordered && |
7511 | (!writing || | |
7512 | !btrfs_page_exists_in_range(inode, lockstart, lockend))) | |
eb838e73 JB |
7513 | break; |
7514 | ||
7515 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
7516 | cached_state, GFP_NOFS); | |
7517 | ||
7518 | if (ordered) { | |
ade77029 FM |
7519 | /* |
7520 | * If we are doing a DIO read and the ordered extent we | |
7521 | * found is for a buffered write, we can not wait for it | |
7522 | * to complete and retry, because if we do so we can | |
7523 | * deadlock with concurrent buffered writes on page | |
7524 | * locks. This happens only if our DIO read covers more | |
7525 | * than one extent map, if at this point has already | |
7526 | * created an ordered extent for a previous extent map | |
7527 | * and locked its range in the inode's io tree, and a | |
7528 | * concurrent write against that previous extent map's | |
7529 | * range and this range started (we unlock the ranges | |
7530 | * in the io tree only when the bios complete and | |
7531 | * buffered writes always lock pages before attempting | |
7532 | * to lock range in the io tree). | |
7533 | */ | |
7534 | if (writing || | |
7535 | test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) | |
7536 | btrfs_start_ordered_extent(inode, ordered, 1); | |
7537 | else | |
7538 | ret = -ENOTBLK; | |
eb838e73 JB |
7539 | btrfs_put_ordered_extent(ordered); |
7540 | } else { | |
eb838e73 | 7541 | /* |
b850ae14 FM |
7542 | * We could trigger writeback for this range (and wait |
7543 | * for it to complete) and then invalidate the pages for | |
7544 | * this range (through invalidate_inode_pages2_range()), | |
7545 | * but that can lead us to a deadlock with a concurrent | |
7546 | * call to readpages() (a buffered read or a defrag call | |
7547 | * triggered a readahead) on a page lock due to an | |
7548 | * ordered dio extent we created before but did not have | |
7549 | * yet a corresponding bio submitted (whence it can not | |
7550 | * complete), which makes readpages() wait for that | |
7551 | * ordered extent to complete while holding a lock on | |
7552 | * that page. | |
eb838e73 | 7553 | */ |
b850ae14 | 7554 | ret = -ENOTBLK; |
eb838e73 JB |
7555 | } |
7556 | ||
ade77029 FM |
7557 | if (ret) |
7558 | break; | |
7559 | ||
eb838e73 JB |
7560 | cond_resched(); |
7561 | } | |
7562 | ||
7563 | return ret; | |
7564 | } | |
7565 | ||
69ffb543 JB |
7566 | static struct extent_map *create_pinned_em(struct inode *inode, u64 start, |
7567 | u64 len, u64 orig_start, | |
7568 | u64 block_start, u64 block_len, | |
cc95bef6 JB |
7569 | u64 orig_block_len, u64 ram_bytes, |
7570 | int type) | |
69ffb543 JB |
7571 | { |
7572 | struct extent_map_tree *em_tree; | |
7573 | struct extent_map *em; | |
7574 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7575 | int ret; | |
7576 | ||
7577 | em_tree = &BTRFS_I(inode)->extent_tree; | |
7578 | em = alloc_extent_map(); | |
7579 | if (!em) | |
7580 | return ERR_PTR(-ENOMEM); | |
7581 | ||
7582 | em->start = start; | |
7583 | em->orig_start = orig_start; | |
2ab28f32 JB |
7584 | em->mod_start = start; |
7585 | em->mod_len = len; | |
69ffb543 JB |
7586 | em->len = len; |
7587 | em->block_len = block_len; | |
7588 | em->block_start = block_start; | |
7589 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
b4939680 | 7590 | em->orig_block_len = orig_block_len; |
cc95bef6 | 7591 | em->ram_bytes = ram_bytes; |
70c8a91c | 7592 | em->generation = -1; |
69ffb543 JB |
7593 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
7594 | if (type == BTRFS_ORDERED_PREALLOC) | |
b11e234d | 7595 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
69ffb543 JB |
7596 | |
7597 | do { | |
7598 | btrfs_drop_extent_cache(inode, em->start, | |
7599 | em->start + em->len - 1, 0); | |
7600 | write_lock(&em_tree->lock); | |
09a2a8f9 | 7601 | ret = add_extent_mapping(em_tree, em, 1); |
69ffb543 JB |
7602 | write_unlock(&em_tree->lock); |
7603 | } while (ret == -EEXIST); | |
7604 | ||
7605 | if (ret) { | |
7606 | free_extent_map(em); | |
7607 | return ERR_PTR(ret); | |
7608 | } | |
7609 | ||
7610 | return em; | |
7611 | } | |
7612 | ||
9c9464cc FM |
7613 | static void adjust_dio_outstanding_extents(struct inode *inode, |
7614 | struct btrfs_dio_data *dio_data, | |
7615 | const u64 len) | |
7616 | { | |
823bb20a | 7617 | unsigned num_extents = count_max_extents(len); |
9c9464cc | 7618 | |
9c9464cc FM |
7619 | /* |
7620 | * If we have an outstanding_extents count still set then we're | |
7621 | * within our reservation, otherwise we need to adjust our inode | |
7622 | * counter appropriately. | |
7623 | */ | |
c2931667 | 7624 | if (dio_data->outstanding_extents >= num_extents) { |
9c9464cc FM |
7625 | dio_data->outstanding_extents -= num_extents; |
7626 | } else { | |
c2931667 LB |
7627 | /* |
7628 | * If dio write length has been split due to no large enough | |
7629 | * contiguous space, we need to compensate our inode counter | |
7630 | * appropriately. | |
7631 | */ | |
7632 | u64 num_needed = num_extents - dio_data->outstanding_extents; | |
7633 | ||
9c9464cc | 7634 | spin_lock(&BTRFS_I(inode)->lock); |
c2931667 | 7635 | BTRFS_I(inode)->outstanding_extents += num_needed; |
9c9464cc FM |
7636 | spin_unlock(&BTRFS_I(inode)->lock); |
7637 | } | |
7638 | } | |
7639 | ||
4b46fce2 JB |
7640 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
7641 | struct buffer_head *bh_result, int create) | |
7642 | { | |
0b246afa | 7643 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 7644 | struct extent_map *em; |
eb838e73 | 7645 | struct extent_state *cached_state = NULL; |
50745b0a | 7646 | struct btrfs_dio_data *dio_data = NULL; |
4b46fce2 | 7647 | u64 start = iblock << inode->i_blkbits; |
eb838e73 | 7648 | u64 lockstart, lockend; |
4b46fce2 | 7649 | u64 len = bh_result->b_size; |
eb838e73 | 7650 | int unlock_bits = EXTENT_LOCKED; |
0934856d | 7651 | int ret = 0; |
eb838e73 | 7652 | |
172a5049 | 7653 | if (create) |
3266789f | 7654 | unlock_bits |= EXTENT_DIRTY; |
172a5049 | 7655 | else |
0b246afa | 7656 | len = min_t(u64, len, fs_info->sectorsize); |
eb838e73 | 7657 | |
c329861d JB |
7658 | lockstart = start; |
7659 | lockend = start + len - 1; | |
7660 | ||
e1cbbfa5 JB |
7661 | if (current->journal_info) { |
7662 | /* | |
7663 | * Need to pull our outstanding extents and set journal_info to NULL so | |
01327610 | 7664 | * that anything that needs to check if there's a transaction doesn't get |
e1cbbfa5 JB |
7665 | * confused. |
7666 | */ | |
50745b0a | 7667 | dio_data = current->journal_info; |
e1cbbfa5 JB |
7668 | current->journal_info = NULL; |
7669 | } | |
7670 | ||
eb838e73 JB |
7671 | /* |
7672 | * If this errors out it's because we couldn't invalidate pagecache for | |
7673 | * this range and we need to fallback to buffered. | |
7674 | */ | |
9c9464cc FM |
7675 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, |
7676 | create)) { | |
7677 | ret = -ENOTBLK; | |
7678 | goto err; | |
7679 | } | |
eb838e73 | 7680 | |
4b46fce2 | 7681 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); |
eb838e73 JB |
7682 | if (IS_ERR(em)) { |
7683 | ret = PTR_ERR(em); | |
7684 | goto unlock_err; | |
7685 | } | |
4b46fce2 JB |
7686 | |
7687 | /* | |
7688 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
7689 | * io. INLINE is special, and we could probably kludge it in here, but | |
7690 | * it's still buffered so for safety lets just fall back to the generic | |
7691 | * buffered path. | |
7692 | * | |
7693 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
7694 | * decompress it, so there will be buffering required no matter what we | |
7695 | * do, so go ahead and fallback to buffered. | |
7696 | * | |
01327610 | 7697 | * We return -ENOTBLK because that's what makes DIO go ahead and go back |
4b46fce2 JB |
7698 | * to buffered IO. Don't blame me, this is the price we pay for using |
7699 | * the generic code. | |
7700 | */ | |
7701 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
7702 | em->block_start == EXTENT_MAP_INLINE) { | |
7703 | free_extent_map(em); | |
eb838e73 JB |
7704 | ret = -ENOTBLK; |
7705 | goto unlock_err; | |
4b46fce2 JB |
7706 | } |
7707 | ||
7708 | /* Just a good old fashioned hole, return */ | |
7709 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
7710 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
7711 | free_extent_map(em); | |
eb838e73 | 7712 | goto unlock_err; |
4b46fce2 JB |
7713 | } |
7714 | ||
7715 | /* | |
7716 | * We don't allocate a new extent in the following cases | |
7717 | * | |
7718 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
7719 | * existing extent. | |
7720 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
7721 | * just use the extent. | |
7722 | * | |
7723 | */ | |
46bfbb5c | 7724 | if (!create) { |
eb838e73 JB |
7725 | len = min(len, em->len - (start - em->start)); |
7726 | lockstart = start + len; | |
7727 | goto unlock; | |
46bfbb5c | 7728 | } |
4b46fce2 JB |
7729 | |
7730 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
7731 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
7732 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 | 7733 | int type; |
eb384b55 | 7734 | u64 block_start, orig_start, orig_block_len, ram_bytes; |
4b46fce2 JB |
7735 | |
7736 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7737 | type = BTRFS_ORDERED_PREALLOC; | |
7738 | else | |
7739 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 7740 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 7741 | block_start = em->block_start + (start - em->start); |
46bfbb5c | 7742 | |
00361589 | 7743 | if (can_nocow_extent(inode, start, &len, &orig_start, |
f78c436c | 7744 | &orig_block_len, &ram_bytes) == 1 && |
0b246afa | 7745 | btrfs_inc_nocow_writers(fs_info, block_start)) { |
5f9a8a51 | 7746 | struct extent_map *em2; |
0b901916 | 7747 | |
5f9a8a51 FM |
7748 | em2 = btrfs_create_dio_extent(inode, start, len, |
7749 | orig_start, block_start, | |
7750 | len, orig_block_len, | |
7751 | ram_bytes, type); | |
0b246afa | 7752 | btrfs_dec_nocow_writers(fs_info, block_start); |
69ffb543 JB |
7753 | if (type == BTRFS_ORDERED_PREALLOC) { |
7754 | free_extent_map(em); | |
5f9a8a51 | 7755 | em = em2; |
69ffb543 | 7756 | } |
5f9a8a51 FM |
7757 | if (em2 && IS_ERR(em2)) { |
7758 | ret = PTR_ERR(em2); | |
eb838e73 | 7759 | goto unlock_err; |
46bfbb5c | 7760 | } |
18513091 WX |
7761 | /* |
7762 | * For inode marked NODATACOW or extent marked PREALLOC, | |
7763 | * use the existing or preallocated extent, so does not | |
7764 | * need to adjust btrfs_space_info's bytes_may_use. | |
7765 | */ | |
7766 | btrfs_free_reserved_data_space_noquota(inode, | |
7767 | start, len); | |
46bfbb5c | 7768 | goto unlock; |
4b46fce2 | 7769 | } |
4b46fce2 | 7770 | } |
00361589 | 7771 | |
46bfbb5c CM |
7772 | /* |
7773 | * this will cow the extent, reset the len in case we changed | |
7774 | * it above | |
7775 | */ | |
7776 | len = bh_result->b_size; | |
70c8a91c JB |
7777 | free_extent_map(em); |
7778 | em = btrfs_new_extent_direct(inode, start, len); | |
eb838e73 JB |
7779 | if (IS_ERR(em)) { |
7780 | ret = PTR_ERR(em); | |
7781 | goto unlock_err; | |
7782 | } | |
46bfbb5c CM |
7783 | len = min(len, em->len - (start - em->start)); |
7784 | unlock: | |
4b46fce2 JB |
7785 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> |
7786 | inode->i_blkbits; | |
46bfbb5c | 7787 | bh_result->b_size = len; |
4b46fce2 JB |
7788 | bh_result->b_bdev = em->bdev; |
7789 | set_buffer_mapped(bh_result); | |
c3473e83 JB |
7790 | if (create) { |
7791 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7792 | set_buffer_new(bh_result); | |
7793 | ||
7794 | /* | |
7795 | * Need to update the i_size under the extent lock so buffered | |
7796 | * readers will get the updated i_size when we unlock. | |
7797 | */ | |
7798 | if (start + len > i_size_read(inode)) | |
7799 | i_size_write(inode, start + len); | |
0934856d | 7800 | |
9c9464cc | 7801 | adjust_dio_outstanding_extents(inode, dio_data, len); |
50745b0a | 7802 | WARN_ON(dio_data->reserve < len); |
7803 | dio_data->reserve -= len; | |
f28a4928 | 7804 | dio_data->unsubmitted_oe_range_end = start + len; |
50745b0a | 7805 | current->journal_info = dio_data; |
c3473e83 | 7806 | } |
4b46fce2 | 7807 | |
eb838e73 JB |
7808 | /* |
7809 | * In the case of write we need to clear and unlock the entire range, | |
7810 | * in the case of read we need to unlock only the end area that we | |
7811 | * aren't using if there is any left over space. | |
7812 | */ | |
24c03fa5 | 7813 | if (lockstart < lockend) { |
0934856d MX |
7814 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, |
7815 | lockend, unlock_bits, 1, 0, | |
7816 | &cached_state, GFP_NOFS); | |
24c03fa5 | 7817 | } else { |
eb838e73 | 7818 | free_extent_state(cached_state); |
24c03fa5 | 7819 | } |
eb838e73 | 7820 | |
4b46fce2 JB |
7821 | free_extent_map(em); |
7822 | ||
7823 | return 0; | |
eb838e73 JB |
7824 | |
7825 | unlock_err: | |
eb838e73 JB |
7826 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
7827 | unlock_bits, 1, 0, &cached_state, GFP_NOFS); | |
9c9464cc | 7828 | err: |
50745b0a | 7829 | if (dio_data) |
7830 | current->journal_info = dio_data; | |
9c9464cc FM |
7831 | /* |
7832 | * Compensate the delalloc release we do in btrfs_direct_IO() when we | |
7833 | * write less data then expected, so that we don't underflow our inode's | |
7834 | * outstanding extents counter. | |
7835 | */ | |
7836 | if (create && dio_data) | |
7837 | adjust_dio_outstanding_extents(inode, dio_data, len); | |
7838 | ||
eb838e73 | 7839 | return ret; |
4b46fce2 JB |
7840 | } |
7841 | ||
8b110e39 | 7842 | static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio, |
81a75f67 | 7843 | int mirror_num) |
8b110e39 | 7844 | { |
2ff7e61e | 7845 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
8b110e39 MX |
7846 | int ret; |
7847 | ||
37226b21 | 7848 | BUG_ON(bio_op(bio) == REQ_OP_WRITE); |
8b110e39 MX |
7849 | |
7850 | bio_get(bio); | |
7851 | ||
2ff7e61e | 7852 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR); |
8b110e39 MX |
7853 | if (ret) |
7854 | goto err; | |
7855 | ||
2ff7e61e | 7856 | ret = btrfs_map_bio(fs_info, bio, mirror_num, 0); |
8b110e39 MX |
7857 | err: |
7858 | bio_put(bio); | |
7859 | return ret; | |
7860 | } | |
7861 | ||
7862 | static int btrfs_check_dio_repairable(struct inode *inode, | |
7863 | struct bio *failed_bio, | |
7864 | struct io_failure_record *failrec, | |
7865 | int failed_mirror) | |
7866 | { | |
ab8d0fc4 | 7867 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
8b110e39 MX |
7868 | int num_copies; |
7869 | ||
ab8d0fc4 | 7870 | num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len); |
8b110e39 MX |
7871 | if (num_copies == 1) { |
7872 | /* | |
7873 | * we only have a single copy of the data, so don't bother with | |
7874 | * all the retry and error correction code that follows. no | |
7875 | * matter what the error is, it is very likely to persist. | |
7876 | */ | |
ab8d0fc4 JM |
7877 | btrfs_debug(fs_info, |
7878 | "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d", | |
7879 | num_copies, failrec->this_mirror, failed_mirror); | |
8b110e39 MX |
7880 | return 0; |
7881 | } | |
7882 | ||
7883 | failrec->failed_mirror = failed_mirror; | |
7884 | failrec->this_mirror++; | |
7885 | if (failrec->this_mirror == failed_mirror) | |
7886 | failrec->this_mirror++; | |
7887 | ||
7888 | if (failrec->this_mirror > num_copies) { | |
ab8d0fc4 JM |
7889 | btrfs_debug(fs_info, |
7890 | "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d", | |
7891 | num_copies, failrec->this_mirror, failed_mirror); | |
8b110e39 MX |
7892 | return 0; |
7893 | } | |
7894 | ||
7895 | return 1; | |
7896 | } | |
7897 | ||
7898 | static int dio_read_error(struct inode *inode, struct bio *failed_bio, | |
2dabb324 CR |
7899 | struct page *page, unsigned int pgoff, |
7900 | u64 start, u64 end, int failed_mirror, | |
7901 | bio_end_io_t *repair_endio, void *repair_arg) | |
8b110e39 MX |
7902 | { |
7903 | struct io_failure_record *failrec; | |
7904 | struct bio *bio; | |
7905 | int isector; | |
70fd7614 | 7906 | int read_mode = 0; |
8b110e39 MX |
7907 | int ret; |
7908 | ||
37226b21 | 7909 | BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE); |
8b110e39 MX |
7910 | |
7911 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
7912 | if (ret) | |
7913 | return ret; | |
7914 | ||
7915 | ret = btrfs_check_dio_repairable(inode, failed_bio, failrec, | |
7916 | failed_mirror); | |
7917 | if (!ret) { | |
7918 | free_io_failure(inode, failrec); | |
7919 | return -EIO; | |
7920 | } | |
7921 | ||
2dabb324 CR |
7922 | if ((failed_bio->bi_vcnt > 1) |
7923 | || (failed_bio->bi_io_vec->bv_len | |
da17066c | 7924 | > btrfs_inode_sectorsize(inode))) |
70fd7614 | 7925 | read_mode |= REQ_FAILFAST_DEV; |
8b110e39 MX |
7926 | |
7927 | isector = start - btrfs_io_bio(failed_bio)->logical; | |
7928 | isector >>= inode->i_sb->s_blocksize_bits; | |
7929 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
2dabb324 | 7930 | pgoff, isector, repair_endio, repair_arg); |
8b110e39 MX |
7931 | if (!bio) { |
7932 | free_io_failure(inode, failrec); | |
7933 | return -EIO; | |
7934 | } | |
37226b21 | 7935 | bio_set_op_attrs(bio, REQ_OP_READ, read_mode); |
8b110e39 MX |
7936 | |
7937 | btrfs_debug(BTRFS_I(inode)->root->fs_info, | |
7938 | "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n", | |
7939 | read_mode, failrec->this_mirror, failrec->in_validation); | |
7940 | ||
81a75f67 | 7941 | ret = submit_dio_repair_bio(inode, bio, failrec->this_mirror); |
8b110e39 MX |
7942 | if (ret) { |
7943 | free_io_failure(inode, failrec); | |
7944 | bio_put(bio); | |
7945 | } | |
7946 | ||
7947 | return ret; | |
7948 | } | |
7949 | ||
7950 | struct btrfs_retry_complete { | |
7951 | struct completion done; | |
7952 | struct inode *inode; | |
7953 | u64 start; | |
7954 | int uptodate; | |
7955 | }; | |
7956 | ||
4246a0b6 | 7957 | static void btrfs_retry_endio_nocsum(struct bio *bio) |
8b110e39 MX |
7958 | { |
7959 | struct btrfs_retry_complete *done = bio->bi_private; | |
2dabb324 | 7960 | struct inode *inode; |
8b110e39 MX |
7961 | struct bio_vec *bvec; |
7962 | int i; | |
7963 | ||
4246a0b6 | 7964 | if (bio->bi_error) |
8b110e39 MX |
7965 | goto end; |
7966 | ||
2dabb324 CR |
7967 | ASSERT(bio->bi_vcnt == 1); |
7968 | inode = bio->bi_io_vec->bv_page->mapping->host; | |
da17066c | 7969 | ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(inode)); |
2dabb324 | 7970 | |
8b110e39 MX |
7971 | done->uptodate = 1; |
7972 | bio_for_each_segment_all(bvec, bio, i) | |
7973 | clean_io_failure(done->inode, done->start, bvec->bv_page, 0); | |
7974 | end: | |
7975 | complete(&done->done); | |
7976 | bio_put(bio); | |
7977 | } | |
7978 | ||
7979 | static int __btrfs_correct_data_nocsum(struct inode *inode, | |
7980 | struct btrfs_io_bio *io_bio) | |
4b46fce2 | 7981 | { |
2dabb324 | 7982 | struct btrfs_fs_info *fs_info; |
2c30c71b | 7983 | struct bio_vec *bvec; |
8b110e39 | 7984 | struct btrfs_retry_complete done; |
4b46fce2 | 7985 | u64 start; |
2dabb324 CR |
7986 | unsigned int pgoff; |
7987 | u32 sectorsize; | |
7988 | int nr_sectors; | |
2c30c71b | 7989 | int i; |
c1dc0896 | 7990 | int ret; |
4b46fce2 | 7991 | |
2dabb324 | 7992 | fs_info = BTRFS_I(inode)->root->fs_info; |
da17066c | 7993 | sectorsize = fs_info->sectorsize; |
2dabb324 | 7994 | |
8b110e39 MX |
7995 | start = io_bio->logical; |
7996 | done.inode = inode; | |
7997 | ||
7998 | bio_for_each_segment_all(bvec, &io_bio->bio, i) { | |
2dabb324 CR |
7999 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len); |
8000 | pgoff = bvec->bv_offset; | |
8001 | ||
8002 | next_block_or_try_again: | |
8b110e39 MX |
8003 | done.uptodate = 0; |
8004 | done.start = start; | |
8005 | init_completion(&done.done); | |
8006 | ||
2dabb324 CR |
8007 | ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, |
8008 | pgoff, start, start + sectorsize - 1, | |
8009 | io_bio->mirror_num, | |
8010 | btrfs_retry_endio_nocsum, &done); | |
8b110e39 MX |
8011 | if (ret) |
8012 | return ret; | |
8013 | ||
8014 | wait_for_completion(&done.done); | |
8015 | ||
8016 | if (!done.uptodate) { | |
8017 | /* We might have another mirror, so try again */ | |
2dabb324 | 8018 | goto next_block_or_try_again; |
8b110e39 MX |
8019 | } |
8020 | ||
2dabb324 CR |
8021 | start += sectorsize; |
8022 | ||
8023 | if (nr_sectors--) { | |
8024 | pgoff += sectorsize; | |
8025 | goto next_block_or_try_again; | |
8026 | } | |
8b110e39 MX |
8027 | } |
8028 | ||
8029 | return 0; | |
8030 | } | |
8031 | ||
4246a0b6 | 8032 | static void btrfs_retry_endio(struct bio *bio) |
8b110e39 MX |
8033 | { |
8034 | struct btrfs_retry_complete *done = bio->bi_private; | |
8035 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
2dabb324 | 8036 | struct inode *inode; |
8b110e39 | 8037 | struct bio_vec *bvec; |
2dabb324 | 8038 | u64 start; |
8b110e39 MX |
8039 | int uptodate; |
8040 | int ret; | |
8041 | int i; | |
8042 | ||
4246a0b6 | 8043 | if (bio->bi_error) |
8b110e39 MX |
8044 | goto end; |
8045 | ||
8046 | uptodate = 1; | |
2dabb324 CR |
8047 | |
8048 | start = done->start; | |
8049 | ||
8050 | ASSERT(bio->bi_vcnt == 1); | |
8051 | inode = bio->bi_io_vec->bv_page->mapping->host; | |
da17066c | 8052 | ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(inode)); |
2dabb324 | 8053 | |
8b110e39 MX |
8054 | bio_for_each_segment_all(bvec, bio, i) { |
8055 | ret = __readpage_endio_check(done->inode, io_bio, i, | |
2dabb324 CR |
8056 | bvec->bv_page, bvec->bv_offset, |
8057 | done->start, bvec->bv_len); | |
8b110e39 MX |
8058 | if (!ret) |
8059 | clean_io_failure(done->inode, done->start, | |
2dabb324 | 8060 | bvec->bv_page, bvec->bv_offset); |
8b110e39 MX |
8061 | else |
8062 | uptodate = 0; | |
8063 | } | |
8064 | ||
8065 | done->uptodate = uptodate; | |
8066 | end: | |
8067 | complete(&done->done); | |
8068 | bio_put(bio); | |
8069 | } | |
8070 | ||
8071 | static int __btrfs_subio_endio_read(struct inode *inode, | |
8072 | struct btrfs_io_bio *io_bio, int err) | |
8073 | { | |
2dabb324 | 8074 | struct btrfs_fs_info *fs_info; |
8b110e39 MX |
8075 | struct bio_vec *bvec; |
8076 | struct btrfs_retry_complete done; | |
8077 | u64 start; | |
8078 | u64 offset = 0; | |
2dabb324 CR |
8079 | u32 sectorsize; |
8080 | int nr_sectors; | |
8081 | unsigned int pgoff; | |
8082 | int csum_pos; | |
8b110e39 MX |
8083 | int i; |
8084 | int ret; | |
dc380aea | 8085 | |
2dabb324 | 8086 | fs_info = BTRFS_I(inode)->root->fs_info; |
da17066c | 8087 | sectorsize = fs_info->sectorsize; |
2dabb324 | 8088 | |
8b110e39 | 8089 | err = 0; |
c1dc0896 | 8090 | start = io_bio->logical; |
8b110e39 MX |
8091 | done.inode = inode; |
8092 | ||
c1dc0896 | 8093 | bio_for_each_segment_all(bvec, &io_bio->bio, i) { |
2dabb324 CR |
8094 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len); |
8095 | ||
8096 | pgoff = bvec->bv_offset; | |
8097 | next_block: | |
8098 | csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset); | |
8099 | ret = __readpage_endio_check(inode, io_bio, csum_pos, | |
8100 | bvec->bv_page, pgoff, start, | |
8101 | sectorsize); | |
8b110e39 MX |
8102 | if (likely(!ret)) |
8103 | goto next; | |
8104 | try_again: | |
8105 | done.uptodate = 0; | |
8106 | done.start = start; | |
8107 | init_completion(&done.done); | |
8108 | ||
2dabb324 CR |
8109 | ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, |
8110 | pgoff, start, start + sectorsize - 1, | |
8111 | io_bio->mirror_num, | |
8112 | btrfs_retry_endio, &done); | |
8b110e39 MX |
8113 | if (ret) { |
8114 | err = ret; | |
8115 | goto next; | |
8116 | } | |
8117 | ||
8118 | wait_for_completion(&done.done); | |
8119 | ||
8120 | if (!done.uptodate) { | |
8121 | /* We might have another mirror, so try again */ | |
8122 | goto try_again; | |
8123 | } | |
8124 | next: | |
2dabb324 CR |
8125 | offset += sectorsize; |
8126 | start += sectorsize; | |
8127 | ||
8128 | ASSERT(nr_sectors); | |
8129 | ||
8130 | if (--nr_sectors) { | |
8131 | pgoff += sectorsize; | |
8132 | goto next_block; | |
8133 | } | |
2c30c71b | 8134 | } |
c1dc0896 MX |
8135 | |
8136 | return err; | |
8137 | } | |
8138 | ||
8b110e39 MX |
8139 | static int btrfs_subio_endio_read(struct inode *inode, |
8140 | struct btrfs_io_bio *io_bio, int err) | |
8141 | { | |
8142 | bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
8143 | ||
8144 | if (skip_csum) { | |
8145 | if (unlikely(err)) | |
8146 | return __btrfs_correct_data_nocsum(inode, io_bio); | |
8147 | else | |
8148 | return 0; | |
8149 | } else { | |
8150 | return __btrfs_subio_endio_read(inode, io_bio, err); | |
8151 | } | |
8152 | } | |
8153 | ||
4246a0b6 | 8154 | static void btrfs_endio_direct_read(struct bio *bio) |
c1dc0896 MX |
8155 | { |
8156 | struct btrfs_dio_private *dip = bio->bi_private; | |
8157 | struct inode *inode = dip->inode; | |
8158 | struct bio *dio_bio; | |
8159 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
4246a0b6 | 8160 | int err = bio->bi_error; |
c1dc0896 | 8161 | |
8b110e39 MX |
8162 | if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED) |
8163 | err = btrfs_subio_endio_read(inode, io_bio, err); | |
c1dc0896 | 8164 | |
4b46fce2 | 8165 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, |
d0082371 | 8166 | dip->logical_offset + dip->bytes - 1); |
9be3395b | 8167 | dio_bio = dip->dio_bio; |
4b46fce2 | 8168 | |
4b46fce2 | 8169 | kfree(dip); |
c0da7aa1 | 8170 | |
1636d1d7 | 8171 | dio_bio->bi_error = bio->bi_error; |
4246a0b6 | 8172 | dio_end_io(dio_bio, bio->bi_error); |
23ea8e5a MX |
8173 | |
8174 | if (io_bio->end_io) | |
8175 | io_bio->end_io(io_bio, err); | |
9be3395b | 8176 | bio_put(bio); |
4b46fce2 JB |
8177 | } |
8178 | ||
14543774 FM |
8179 | static void btrfs_endio_direct_write_update_ordered(struct inode *inode, |
8180 | const u64 offset, | |
8181 | const u64 bytes, | |
8182 | const int uptodate) | |
4b46fce2 | 8183 | { |
0b246afa | 8184 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4b46fce2 | 8185 | struct btrfs_ordered_extent *ordered = NULL; |
14543774 FM |
8186 | u64 ordered_offset = offset; |
8187 | u64 ordered_bytes = bytes; | |
4b46fce2 JB |
8188 | int ret; |
8189 | ||
163cf09c CM |
8190 | again: |
8191 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
8192 | &ordered_offset, | |
4246a0b6 | 8193 | ordered_bytes, |
14543774 | 8194 | uptodate); |
4b46fce2 | 8195 | if (!ret) |
163cf09c | 8196 | goto out_test; |
4b46fce2 | 8197 | |
9e0af237 LB |
8198 | btrfs_init_work(&ordered->work, btrfs_endio_write_helper, |
8199 | finish_ordered_fn, NULL, NULL); | |
0b246afa | 8200 | btrfs_queue_work(fs_info->endio_write_workers, &ordered->work); |
163cf09c CM |
8201 | out_test: |
8202 | /* | |
8203 | * our bio might span multiple ordered extents. If we haven't | |
8204 | * completed the accounting for the whole dio, go back and try again | |
8205 | */ | |
14543774 FM |
8206 | if (ordered_offset < offset + bytes) { |
8207 | ordered_bytes = offset + bytes - ordered_offset; | |
5fd02043 | 8208 | ordered = NULL; |
163cf09c CM |
8209 | goto again; |
8210 | } | |
14543774 FM |
8211 | } |
8212 | ||
8213 | static void btrfs_endio_direct_write(struct bio *bio) | |
8214 | { | |
8215 | struct btrfs_dio_private *dip = bio->bi_private; | |
8216 | struct bio *dio_bio = dip->dio_bio; | |
8217 | ||
8218 | btrfs_endio_direct_write_update_ordered(dip->inode, | |
8219 | dip->logical_offset, | |
8220 | dip->bytes, | |
8221 | !bio->bi_error); | |
4b46fce2 | 8222 | |
4b46fce2 | 8223 | kfree(dip); |
c0da7aa1 | 8224 | |
1636d1d7 | 8225 | dio_bio->bi_error = bio->bi_error; |
4246a0b6 | 8226 | dio_end_io(dio_bio, bio->bi_error); |
9be3395b | 8227 | bio_put(bio); |
4b46fce2 JB |
8228 | } |
8229 | ||
81a75f67 | 8230 | static int __btrfs_submit_bio_start_direct_io(struct inode *inode, |
eaf25d93 CM |
8231 | struct bio *bio, int mirror_num, |
8232 | unsigned long bio_flags, u64 offset) | |
8233 | { | |
8234 | int ret; | |
2ff7e61e | 8235 | ret = btrfs_csum_one_bio(inode, bio, offset, 1); |
79787eaa | 8236 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
8237 | return 0; |
8238 | } | |
8239 | ||
4246a0b6 | 8240 | static void btrfs_end_dio_bio(struct bio *bio) |
e65e1535 MX |
8241 | { |
8242 | struct btrfs_dio_private *dip = bio->bi_private; | |
4246a0b6 | 8243 | int err = bio->bi_error; |
e65e1535 | 8244 | |
8b110e39 MX |
8245 | if (err) |
8246 | btrfs_warn(BTRFS_I(dip->inode)->root->fs_info, | |
6296b960 | 8247 | "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d", |
f85b7379 DS |
8248 | btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio), |
8249 | bio->bi_opf, | |
8b110e39 MX |
8250 | (unsigned long long)bio->bi_iter.bi_sector, |
8251 | bio->bi_iter.bi_size, err); | |
8252 | ||
8253 | if (dip->subio_endio) | |
8254 | err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err); | |
c1dc0896 MX |
8255 | |
8256 | if (err) { | |
e65e1535 MX |
8257 | dip->errors = 1; |
8258 | ||
8259 | /* | |
8260 | * before atomic variable goto zero, we must make sure | |
8261 | * dip->errors is perceived to be set. | |
8262 | */ | |
4e857c58 | 8263 | smp_mb__before_atomic(); |
e65e1535 MX |
8264 | } |
8265 | ||
8266 | /* if there are more bios still pending for this dio, just exit */ | |
8267 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
8268 | goto out; | |
8269 | ||
9be3395b | 8270 | if (dip->errors) { |
e65e1535 | 8271 | bio_io_error(dip->orig_bio); |
9be3395b | 8272 | } else { |
4246a0b6 CH |
8273 | dip->dio_bio->bi_error = 0; |
8274 | bio_endio(dip->orig_bio); | |
e65e1535 MX |
8275 | } |
8276 | out: | |
8277 | bio_put(bio); | |
8278 | } | |
8279 | ||
8280 | static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev, | |
8281 | u64 first_sector, gfp_t gfp_flags) | |
8282 | { | |
da2f0f74 | 8283 | struct bio *bio; |
22365979 | 8284 | bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags); |
da2f0f74 CM |
8285 | if (bio) |
8286 | bio_associate_current(bio); | |
8287 | return bio; | |
e65e1535 MX |
8288 | } |
8289 | ||
2ff7e61e | 8290 | static inline int btrfs_lookup_and_bind_dio_csum(struct inode *inode, |
c1dc0896 MX |
8291 | struct btrfs_dio_private *dip, |
8292 | struct bio *bio, | |
8293 | u64 file_offset) | |
8294 | { | |
8295 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
8296 | struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio); | |
8297 | int ret; | |
8298 | ||
8299 | /* | |
8300 | * We load all the csum data we need when we submit | |
8301 | * the first bio to reduce the csum tree search and | |
8302 | * contention. | |
8303 | */ | |
8304 | if (dip->logical_offset == file_offset) { | |
2ff7e61e | 8305 | ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio, |
c1dc0896 MX |
8306 | file_offset); |
8307 | if (ret) | |
8308 | return ret; | |
8309 | } | |
8310 | ||
8311 | if (bio == dip->orig_bio) | |
8312 | return 0; | |
8313 | ||
8314 | file_offset -= dip->logical_offset; | |
8315 | file_offset >>= inode->i_sb->s_blocksize_bits; | |
8316 | io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset); | |
8317 | ||
8318 | return 0; | |
8319 | } | |
8320 | ||
e65e1535 | 8321 | static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, |
81a75f67 | 8322 | u64 file_offset, int skip_sum, |
c329861d | 8323 | int async_submit) |
e65e1535 | 8324 | { |
0b246afa | 8325 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
facc8a22 | 8326 | struct btrfs_dio_private *dip = bio->bi_private; |
37226b21 | 8327 | bool write = bio_op(bio) == REQ_OP_WRITE; |
e65e1535 MX |
8328 | int ret; |
8329 | ||
b812ce28 JB |
8330 | if (async_submit) |
8331 | async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers); | |
8332 | ||
e65e1535 | 8333 | bio_get(bio); |
5fd02043 JB |
8334 | |
8335 | if (!write) { | |
0b246afa | 8336 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA); |
5fd02043 JB |
8337 | if (ret) |
8338 | goto err; | |
8339 | } | |
e65e1535 | 8340 | |
1ae39938 JB |
8341 | if (skip_sum) |
8342 | goto map; | |
8343 | ||
8344 | if (write && async_submit) { | |
0b246afa JM |
8345 | ret = btrfs_wq_submit_bio(fs_info, inode, bio, 0, 0, |
8346 | file_offset, | |
8347 | __btrfs_submit_bio_start_direct_io, | |
8348 | __btrfs_submit_bio_done); | |
e65e1535 | 8349 | goto err; |
1ae39938 JB |
8350 | } else if (write) { |
8351 | /* | |
8352 | * If we aren't doing async submit, calculate the csum of the | |
8353 | * bio now. | |
8354 | */ | |
2ff7e61e | 8355 | ret = btrfs_csum_one_bio(inode, bio, file_offset, 1); |
1ae39938 JB |
8356 | if (ret) |
8357 | goto err; | |
23ea8e5a | 8358 | } else { |
2ff7e61e | 8359 | ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio, |
c1dc0896 | 8360 | file_offset); |
c2db1073 TI |
8361 | if (ret) |
8362 | goto err; | |
8363 | } | |
1ae39938 | 8364 | map: |
2ff7e61e | 8365 | ret = btrfs_map_bio(fs_info, bio, 0, async_submit); |
e65e1535 MX |
8366 | err: |
8367 | bio_put(bio); | |
8368 | return ret; | |
8369 | } | |
8370 | ||
81a75f67 | 8371 | static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip, |
e65e1535 MX |
8372 | int skip_sum) |
8373 | { | |
8374 | struct inode *inode = dip->inode; | |
0b246afa | 8375 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e65e1535 | 8376 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e65e1535 MX |
8377 | struct bio *bio; |
8378 | struct bio *orig_bio = dip->orig_bio; | |
6a2de22f | 8379 | struct bio_vec *bvec; |
4f024f37 | 8380 | u64 start_sector = orig_bio->bi_iter.bi_sector; |
e65e1535 MX |
8381 | u64 file_offset = dip->logical_offset; |
8382 | u64 submit_len = 0; | |
8383 | u64 map_length; | |
0b246afa | 8384 | u32 blocksize = fs_info->sectorsize; |
1ae39938 | 8385 | int async_submit = 0; |
5f4dc8fc CR |
8386 | int nr_sectors; |
8387 | int ret; | |
6a2de22f | 8388 | int i, j; |
e65e1535 | 8389 | |
4f024f37 | 8390 | map_length = orig_bio->bi_iter.bi_size; |
0b246afa JM |
8391 | ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9, |
8392 | &map_length, NULL, 0); | |
7a5c3c9b | 8393 | if (ret) |
e65e1535 | 8394 | return -EIO; |
facc8a22 | 8395 | |
4f024f37 | 8396 | if (map_length >= orig_bio->bi_iter.bi_size) { |
02f57c7a | 8397 | bio = orig_bio; |
c1dc0896 | 8398 | dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED; |
02f57c7a JB |
8399 | goto submit; |
8400 | } | |
8401 | ||
53b381b3 | 8402 | /* async crcs make it difficult to collect full stripe writes. */ |
ffe2d203 | 8403 | if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 DW |
8404 | async_submit = 0; |
8405 | else | |
8406 | async_submit = 1; | |
8407 | ||
02f57c7a JB |
8408 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS); |
8409 | if (!bio) | |
8410 | return -ENOMEM; | |
7a5c3c9b | 8411 | |
ef295ecf | 8412 | bio->bi_opf = orig_bio->bi_opf; |
02f57c7a JB |
8413 | bio->bi_private = dip; |
8414 | bio->bi_end_io = btrfs_end_dio_bio; | |
c1dc0896 | 8415 | btrfs_io_bio(bio)->logical = file_offset; |
02f57c7a JB |
8416 | atomic_inc(&dip->pending_bios); |
8417 | ||
6a2de22f | 8418 | bio_for_each_segment_all(bvec, orig_bio, j) { |
0b246afa | 8419 | nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len); |
5f4dc8fc CR |
8420 | i = 0; |
8421 | next_block: | |
8422 | if (unlikely(map_length < submit_len + blocksize || | |
8423 | bio_add_page(bio, bvec->bv_page, blocksize, | |
8424 | bvec->bv_offset + (i * blocksize)) < blocksize)) { | |
e65e1535 MX |
8425 | /* |
8426 | * inc the count before we submit the bio so | |
8427 | * we know the end IO handler won't happen before | |
8428 | * we inc the count. Otherwise, the dip might get freed | |
8429 | * before we're done setting it up | |
8430 | */ | |
8431 | atomic_inc(&dip->pending_bios); | |
81a75f67 | 8432 | ret = __btrfs_submit_dio_bio(bio, inode, |
e65e1535 | 8433 | file_offset, skip_sum, |
c329861d | 8434 | async_submit); |
e65e1535 MX |
8435 | if (ret) { |
8436 | bio_put(bio); | |
8437 | atomic_dec(&dip->pending_bios); | |
8438 | goto out_err; | |
8439 | } | |
8440 | ||
e65e1535 MX |
8441 | start_sector += submit_len >> 9; |
8442 | file_offset += submit_len; | |
8443 | ||
8444 | submit_len = 0; | |
e65e1535 MX |
8445 | |
8446 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, | |
8447 | start_sector, GFP_NOFS); | |
8448 | if (!bio) | |
8449 | goto out_err; | |
ef295ecf | 8450 | bio->bi_opf = orig_bio->bi_opf; |
e65e1535 MX |
8451 | bio->bi_private = dip; |
8452 | bio->bi_end_io = btrfs_end_dio_bio; | |
c1dc0896 | 8453 | btrfs_io_bio(bio)->logical = file_offset; |
e65e1535 | 8454 | |
4f024f37 | 8455 | map_length = orig_bio->bi_iter.bi_size; |
0b246afa | 8456 | ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), |
3ec706c8 | 8457 | start_sector << 9, |
e65e1535 MX |
8458 | &map_length, NULL, 0); |
8459 | if (ret) { | |
8460 | bio_put(bio); | |
8461 | goto out_err; | |
8462 | } | |
5f4dc8fc CR |
8463 | |
8464 | goto next_block; | |
e65e1535 | 8465 | } else { |
5f4dc8fc CR |
8466 | submit_len += blocksize; |
8467 | if (--nr_sectors) { | |
8468 | i++; | |
8469 | goto next_block; | |
8470 | } | |
e65e1535 MX |
8471 | } |
8472 | } | |
8473 | ||
02f57c7a | 8474 | submit: |
81a75f67 | 8475 | ret = __btrfs_submit_dio_bio(bio, inode, file_offset, skip_sum, |
c329861d | 8476 | async_submit); |
e65e1535 MX |
8477 | if (!ret) |
8478 | return 0; | |
8479 | ||
8480 | bio_put(bio); | |
8481 | out_err: | |
8482 | dip->errors = 1; | |
8483 | /* | |
8484 | * before atomic variable goto zero, we must | |
8485 | * make sure dip->errors is perceived to be set. | |
8486 | */ | |
4e857c58 | 8487 | smp_mb__before_atomic(); |
e65e1535 MX |
8488 | if (atomic_dec_and_test(&dip->pending_bios)) |
8489 | bio_io_error(dip->orig_bio); | |
8490 | ||
8491 | /* bio_end_io() will handle error, so we needn't return it */ | |
8492 | return 0; | |
8493 | } | |
8494 | ||
8a4c1e42 MC |
8495 | static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode, |
8496 | loff_t file_offset) | |
4b46fce2 | 8497 | { |
61de718f FM |
8498 | struct btrfs_dio_private *dip = NULL; |
8499 | struct bio *io_bio = NULL; | |
23ea8e5a | 8500 | struct btrfs_io_bio *btrfs_bio; |
4b46fce2 | 8501 | int skip_sum; |
8a4c1e42 | 8502 | bool write = (bio_op(dio_bio) == REQ_OP_WRITE); |
4b46fce2 JB |
8503 | int ret = 0; |
8504 | ||
8505 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
8506 | ||
9be3395b | 8507 | io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS); |
9be3395b CM |
8508 | if (!io_bio) { |
8509 | ret = -ENOMEM; | |
8510 | goto free_ordered; | |
8511 | } | |
8512 | ||
c1dc0896 | 8513 | dip = kzalloc(sizeof(*dip), GFP_NOFS); |
4b46fce2 JB |
8514 | if (!dip) { |
8515 | ret = -ENOMEM; | |
61de718f | 8516 | goto free_ordered; |
4b46fce2 | 8517 | } |
4b46fce2 | 8518 | |
9be3395b | 8519 | dip->private = dio_bio->bi_private; |
4b46fce2 JB |
8520 | dip->inode = inode; |
8521 | dip->logical_offset = file_offset; | |
4f024f37 KO |
8522 | dip->bytes = dio_bio->bi_iter.bi_size; |
8523 | dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9; | |
9be3395b | 8524 | io_bio->bi_private = dip; |
9be3395b CM |
8525 | dip->orig_bio = io_bio; |
8526 | dip->dio_bio = dio_bio; | |
e65e1535 | 8527 | atomic_set(&dip->pending_bios, 0); |
c1dc0896 MX |
8528 | btrfs_bio = btrfs_io_bio(io_bio); |
8529 | btrfs_bio->logical = file_offset; | |
4b46fce2 | 8530 | |
c1dc0896 | 8531 | if (write) { |
9be3395b | 8532 | io_bio->bi_end_io = btrfs_endio_direct_write; |
c1dc0896 | 8533 | } else { |
9be3395b | 8534 | io_bio->bi_end_io = btrfs_endio_direct_read; |
c1dc0896 MX |
8535 | dip->subio_endio = btrfs_subio_endio_read; |
8536 | } | |
4b46fce2 | 8537 | |
f28a4928 FM |
8538 | /* |
8539 | * Reset the range for unsubmitted ordered extents (to a 0 length range) | |
8540 | * even if we fail to submit a bio, because in such case we do the | |
8541 | * corresponding error handling below and it must not be done a second | |
8542 | * time by btrfs_direct_IO(). | |
8543 | */ | |
8544 | if (write) { | |
8545 | struct btrfs_dio_data *dio_data = current->journal_info; | |
8546 | ||
8547 | dio_data->unsubmitted_oe_range_end = dip->logical_offset + | |
8548 | dip->bytes; | |
8549 | dio_data->unsubmitted_oe_range_start = | |
8550 | dio_data->unsubmitted_oe_range_end; | |
8551 | } | |
8552 | ||
81a75f67 | 8553 | ret = btrfs_submit_direct_hook(dip, skip_sum); |
e65e1535 | 8554 | if (!ret) |
eaf25d93 | 8555 | return; |
9be3395b | 8556 | |
23ea8e5a MX |
8557 | if (btrfs_bio->end_io) |
8558 | btrfs_bio->end_io(btrfs_bio, ret); | |
9be3395b | 8559 | |
4b46fce2 JB |
8560 | free_ordered: |
8561 | /* | |
61de718f FM |
8562 | * If we arrived here it means either we failed to submit the dip |
8563 | * or we either failed to clone the dio_bio or failed to allocate the | |
8564 | * dip. If we cloned the dio_bio and allocated the dip, we can just | |
8565 | * call bio_endio against our io_bio so that we get proper resource | |
8566 | * cleanup if we fail to submit the dip, otherwise, we must do the | |
8567 | * same as btrfs_endio_direct_[write|read] because we can't call these | |
8568 | * callbacks - they require an allocated dip and a clone of dio_bio. | |
4b46fce2 | 8569 | */ |
61de718f | 8570 | if (io_bio && dip) { |
4246a0b6 CH |
8571 | io_bio->bi_error = -EIO; |
8572 | bio_endio(io_bio); | |
61de718f FM |
8573 | /* |
8574 | * The end io callbacks free our dip, do the final put on io_bio | |
8575 | * and all the cleanup and final put for dio_bio (through | |
8576 | * dio_end_io()). | |
8577 | */ | |
8578 | dip = NULL; | |
8579 | io_bio = NULL; | |
8580 | } else { | |
14543774 FM |
8581 | if (write) |
8582 | btrfs_endio_direct_write_update_ordered(inode, | |
8583 | file_offset, | |
8584 | dio_bio->bi_iter.bi_size, | |
8585 | 0); | |
8586 | else | |
61de718f FM |
8587 | unlock_extent(&BTRFS_I(inode)->io_tree, file_offset, |
8588 | file_offset + dio_bio->bi_iter.bi_size - 1); | |
14543774 | 8589 | |
4246a0b6 | 8590 | dio_bio->bi_error = -EIO; |
61de718f FM |
8591 | /* |
8592 | * Releases and cleans up our dio_bio, no need to bio_put() | |
8593 | * nor bio_endio()/bio_io_error() against dio_bio. | |
8594 | */ | |
8595 | dio_end_io(dio_bio, ret); | |
4b46fce2 | 8596 | } |
61de718f FM |
8597 | if (io_bio) |
8598 | bio_put(io_bio); | |
8599 | kfree(dip); | |
4b46fce2 JB |
8600 | } |
8601 | ||
2ff7e61e JM |
8602 | static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info, |
8603 | struct kiocb *iocb, | |
8604 | const struct iov_iter *iter, loff_t offset) | |
5a5f79b5 CM |
8605 | { |
8606 | int seg; | |
a1b75f7d | 8607 | int i; |
0b246afa | 8608 | unsigned int blocksize_mask = fs_info->sectorsize - 1; |
5a5f79b5 | 8609 | ssize_t retval = -EINVAL; |
5a5f79b5 CM |
8610 | |
8611 | if (offset & blocksize_mask) | |
8612 | goto out; | |
8613 | ||
28060d5d AV |
8614 | if (iov_iter_alignment(iter) & blocksize_mask) |
8615 | goto out; | |
a1b75f7d | 8616 | |
28060d5d | 8617 | /* If this is a write we don't need to check anymore */ |
cd27e455 | 8618 | if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter)) |
28060d5d AV |
8619 | return 0; |
8620 | /* | |
8621 | * Check to make sure we don't have duplicate iov_base's in this | |
8622 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
8623 | * when reading back. | |
8624 | */ | |
8625 | for (seg = 0; seg < iter->nr_segs; seg++) { | |
8626 | for (i = seg + 1; i < iter->nr_segs; i++) { | |
8627 | if (iter->iov[seg].iov_base == iter->iov[i].iov_base) | |
a1b75f7d JB |
8628 | goto out; |
8629 | } | |
5a5f79b5 CM |
8630 | } |
8631 | retval = 0; | |
8632 | out: | |
8633 | return retval; | |
8634 | } | |
eb838e73 | 8635 | |
c8b8e32d | 8636 | static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
16432985 | 8637 | { |
4b46fce2 JB |
8638 | struct file *file = iocb->ki_filp; |
8639 | struct inode *inode = file->f_mapping->host; | |
0b246afa | 8640 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
50745b0a | 8641 | struct btrfs_dio_data dio_data = { 0 }; |
c8b8e32d | 8642 | loff_t offset = iocb->ki_pos; |
0934856d | 8643 | size_t count = 0; |
2e60a51e | 8644 | int flags = 0; |
38851cc1 MX |
8645 | bool wakeup = true; |
8646 | bool relock = false; | |
0934856d | 8647 | ssize_t ret; |
4b46fce2 | 8648 | |
2ff7e61e | 8649 | if (check_direct_IO(fs_info, iocb, iter, offset)) |
5a5f79b5 | 8650 | return 0; |
3f7c579c | 8651 | |
fe0f07d0 | 8652 | inode_dio_begin(inode); |
4e857c58 | 8653 | smp_mb__after_atomic(); |
38851cc1 | 8654 | |
0e267c44 | 8655 | /* |
41bd9ca4 MX |
8656 | * The generic stuff only does filemap_write_and_wait_range, which |
8657 | * isn't enough if we've written compressed pages to this area, so | |
8658 | * we need to flush the dirty pages again to make absolutely sure | |
8659 | * that any outstanding dirty pages are on disk. | |
0e267c44 | 8660 | */ |
a6cbcd4a | 8661 | count = iov_iter_count(iter); |
41bd9ca4 MX |
8662 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, |
8663 | &BTRFS_I(inode)->runtime_flags)) | |
9a025a08 WS |
8664 | filemap_fdatawrite_range(inode->i_mapping, offset, |
8665 | offset + count - 1); | |
0e267c44 | 8666 | |
6f673763 | 8667 | if (iov_iter_rw(iter) == WRITE) { |
38851cc1 MX |
8668 | /* |
8669 | * If the write DIO is beyond the EOF, we need update | |
8670 | * the isize, but it is protected by i_mutex. So we can | |
8671 | * not unlock the i_mutex at this case. | |
8672 | */ | |
8673 | if (offset + count <= inode->i_size) { | |
5955102c | 8674 | inode_unlock(inode); |
38851cc1 MX |
8675 | relock = true; |
8676 | } | |
7cf5b976 | 8677 | ret = btrfs_delalloc_reserve_space(inode, offset, count); |
0934856d | 8678 | if (ret) |
38851cc1 | 8679 | goto out; |
823bb20a | 8680 | dio_data.outstanding_extents = count_max_extents(count); |
e1cbbfa5 JB |
8681 | |
8682 | /* | |
8683 | * We need to know how many extents we reserved so that we can | |
8684 | * do the accounting properly if we go over the number we | |
8685 | * originally calculated. Abuse current->journal_info for this. | |
8686 | */ | |
da17066c | 8687 | dio_data.reserve = round_up(count, |
0b246afa | 8688 | fs_info->sectorsize); |
f28a4928 FM |
8689 | dio_data.unsubmitted_oe_range_start = (u64)offset; |
8690 | dio_data.unsubmitted_oe_range_end = (u64)offset; | |
50745b0a | 8691 | current->journal_info = &dio_data; |
97dcdea0 | 8692 | down_read(&BTRFS_I(inode)->dio_sem); |
ee39b432 DS |
8693 | } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK, |
8694 | &BTRFS_I(inode)->runtime_flags)) { | |
fe0f07d0 | 8695 | inode_dio_end(inode); |
38851cc1 MX |
8696 | flags = DIO_LOCKING | DIO_SKIP_HOLES; |
8697 | wakeup = false; | |
0934856d MX |
8698 | } |
8699 | ||
17f8c842 | 8700 | ret = __blockdev_direct_IO(iocb, inode, |
0b246afa | 8701 | fs_info->fs_devices->latest_bdev, |
c8b8e32d | 8702 | iter, btrfs_get_blocks_direct, NULL, |
17f8c842 | 8703 | btrfs_submit_direct, flags); |
6f673763 | 8704 | if (iov_iter_rw(iter) == WRITE) { |
97dcdea0 | 8705 | up_read(&BTRFS_I(inode)->dio_sem); |
e1cbbfa5 | 8706 | current->journal_info = NULL; |
ddba1bfc | 8707 | if (ret < 0 && ret != -EIOCBQUEUED) { |
50745b0a | 8708 | if (dio_data.reserve) |
7cf5b976 QW |
8709 | btrfs_delalloc_release_space(inode, offset, |
8710 | dio_data.reserve); | |
f28a4928 FM |
8711 | /* |
8712 | * On error we might have left some ordered extents | |
8713 | * without submitting corresponding bios for them, so | |
8714 | * cleanup them up to avoid other tasks getting them | |
8715 | * and waiting for them to complete forever. | |
8716 | */ | |
8717 | if (dio_data.unsubmitted_oe_range_start < | |
8718 | dio_data.unsubmitted_oe_range_end) | |
8719 | btrfs_endio_direct_write_update_ordered(inode, | |
8720 | dio_data.unsubmitted_oe_range_start, | |
8721 | dio_data.unsubmitted_oe_range_end - | |
8722 | dio_data.unsubmitted_oe_range_start, | |
8723 | 0); | |
ddba1bfc | 8724 | } else if (ret >= 0 && (size_t)ret < count) |
7cf5b976 QW |
8725 | btrfs_delalloc_release_space(inode, offset, |
8726 | count - (size_t)ret); | |
0934856d | 8727 | } |
38851cc1 | 8728 | out: |
2e60a51e | 8729 | if (wakeup) |
fe0f07d0 | 8730 | inode_dio_end(inode); |
38851cc1 | 8731 | if (relock) |
5955102c | 8732 | inode_lock(inode); |
0934856d MX |
8733 | |
8734 | return ret; | |
16432985 CM |
8735 | } |
8736 | ||
05dadc09 TI |
8737 | #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC) |
8738 | ||
1506fcc8 YS |
8739 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
8740 | __u64 start, __u64 len) | |
8741 | { | |
05dadc09 TI |
8742 | int ret; |
8743 | ||
8744 | ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS); | |
8745 | if (ret) | |
8746 | return ret; | |
8747 | ||
ec29ed5b | 8748 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
8749 | } |
8750 | ||
a52d9a80 | 8751 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 8752 | { |
d1310b2e CM |
8753 | struct extent_io_tree *tree; |
8754 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 8755 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 8756 | } |
1832a6d5 | 8757 | |
a52d9a80 | 8758 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 8759 | { |
d1310b2e | 8760 | struct extent_io_tree *tree; |
be7bd730 JB |
8761 | struct inode *inode = page->mapping->host; |
8762 | int ret; | |
b888db2b CM |
8763 | |
8764 | if (current->flags & PF_MEMALLOC) { | |
8765 | redirty_page_for_writepage(wbc, page); | |
8766 | unlock_page(page); | |
8767 | return 0; | |
8768 | } | |
be7bd730 JB |
8769 | |
8770 | /* | |
8771 | * If we are under memory pressure we will call this directly from the | |
8772 | * VM, we need to make sure we have the inode referenced for the ordered | |
8773 | * extent. If not just return like we didn't do anything. | |
8774 | */ | |
8775 | if (!igrab(inode)) { | |
8776 | redirty_page_for_writepage(wbc, page); | |
8777 | return AOP_WRITEPAGE_ACTIVATE; | |
8778 | } | |
d1310b2e | 8779 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
be7bd730 JB |
8780 | ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
8781 | btrfs_add_delayed_iput(inode); | |
8782 | return ret; | |
9ebefb18 CM |
8783 | } |
8784 | ||
48a3b636 ES |
8785 | static int btrfs_writepages(struct address_space *mapping, |
8786 | struct writeback_control *wbc) | |
b293f02e | 8787 | { |
d1310b2e | 8788 | struct extent_io_tree *tree; |
771ed689 | 8789 | |
d1310b2e | 8790 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
8791 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
8792 | } | |
8793 | ||
3ab2fb5a CM |
8794 | static int |
8795 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
8796 | struct list_head *pages, unsigned nr_pages) | |
8797 | { | |
d1310b2e CM |
8798 | struct extent_io_tree *tree; |
8799 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
8800 | return extent_readpages(tree, mapping, pages, nr_pages, |
8801 | btrfs_get_extent); | |
8802 | } | |
e6dcd2dc | 8803 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 8804 | { |
d1310b2e CM |
8805 | struct extent_io_tree *tree; |
8806 | struct extent_map_tree *map; | |
a52d9a80 | 8807 | int ret; |
8c2383c3 | 8808 | |
d1310b2e CM |
8809 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
8810 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 8811 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
8812 | if (ret == 1) { |
8813 | ClearPagePrivate(page); | |
8814 | set_page_private(page, 0); | |
09cbfeaf | 8815 | put_page(page); |
39279cc3 | 8816 | } |
a52d9a80 | 8817 | return ret; |
39279cc3 CM |
8818 | } |
8819 | ||
e6dcd2dc CM |
8820 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
8821 | { | |
98509cfc CM |
8822 | if (PageWriteback(page) || PageDirty(page)) |
8823 | return 0; | |
3ba7ab22 | 8824 | return __btrfs_releasepage(page, gfp_flags); |
e6dcd2dc CM |
8825 | } |
8826 | ||
d47992f8 LC |
8827 | static void btrfs_invalidatepage(struct page *page, unsigned int offset, |
8828 | unsigned int length) | |
39279cc3 | 8829 | { |
5fd02043 | 8830 | struct inode *inode = page->mapping->host; |
d1310b2e | 8831 | struct extent_io_tree *tree; |
e6dcd2dc | 8832 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 8833 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 8834 | u64 page_start = page_offset(page); |
09cbfeaf | 8835 | u64 page_end = page_start + PAGE_SIZE - 1; |
dbfdb6d1 CR |
8836 | u64 start; |
8837 | u64 end; | |
131e404a | 8838 | int inode_evicting = inode->i_state & I_FREEING; |
39279cc3 | 8839 | |
8b62b72b CM |
8840 | /* |
8841 | * we have the page locked, so new writeback can't start, | |
8842 | * and the dirty bit won't be cleared while we are here. | |
8843 | * | |
8844 | * Wait for IO on this page so that we can safely clear | |
8845 | * the PagePrivate2 bit and do ordered accounting | |
8846 | */ | |
e6dcd2dc | 8847 | wait_on_page_writeback(page); |
8b62b72b | 8848 | |
5fd02043 | 8849 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
8850 | if (offset) { |
8851 | btrfs_releasepage(page, GFP_NOFS); | |
8852 | return; | |
8853 | } | |
131e404a FDBM |
8854 | |
8855 | if (!inode_evicting) | |
ff13db41 | 8856 | lock_extent_bits(tree, page_start, page_end, &cached_state); |
dbfdb6d1 CR |
8857 | again: |
8858 | start = page_start; | |
8859 | ordered = btrfs_lookup_ordered_range(inode, start, | |
8860 | page_end - start + 1); | |
e6dcd2dc | 8861 | if (ordered) { |
dbfdb6d1 | 8862 | end = min(page_end, ordered->file_offset + ordered->len - 1); |
eb84ae03 CM |
8863 | /* |
8864 | * IO on this page will never be started, so we need | |
8865 | * to account for any ordered extents now | |
8866 | */ | |
131e404a | 8867 | if (!inode_evicting) |
dbfdb6d1 | 8868 | clear_extent_bit(tree, start, end, |
131e404a FDBM |
8869 | EXTENT_DIRTY | EXTENT_DELALLOC | |
8870 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | | |
8871 | EXTENT_DEFRAG, 1, 0, &cached_state, | |
8872 | GFP_NOFS); | |
8b62b72b CM |
8873 | /* |
8874 | * whoever cleared the private bit is responsible | |
8875 | * for the finish_ordered_io | |
8876 | */ | |
77cef2ec JB |
8877 | if (TestClearPagePrivate2(page)) { |
8878 | struct btrfs_ordered_inode_tree *tree; | |
8879 | u64 new_len; | |
8880 | ||
8881 | tree = &BTRFS_I(inode)->ordered_tree; | |
8882 | ||
8883 | spin_lock_irq(&tree->lock); | |
8884 | set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags); | |
dbfdb6d1 | 8885 | new_len = start - ordered->file_offset; |
77cef2ec JB |
8886 | if (new_len < ordered->truncated_len) |
8887 | ordered->truncated_len = new_len; | |
8888 | spin_unlock_irq(&tree->lock); | |
8889 | ||
8890 | if (btrfs_dec_test_ordered_pending(inode, &ordered, | |
dbfdb6d1 CR |
8891 | start, |
8892 | end - start + 1, 1)) | |
77cef2ec | 8893 | btrfs_finish_ordered_io(ordered); |
8b62b72b | 8894 | } |
e6dcd2dc | 8895 | btrfs_put_ordered_extent(ordered); |
131e404a FDBM |
8896 | if (!inode_evicting) { |
8897 | cached_state = NULL; | |
dbfdb6d1 | 8898 | lock_extent_bits(tree, start, end, |
131e404a FDBM |
8899 | &cached_state); |
8900 | } | |
dbfdb6d1 CR |
8901 | |
8902 | start = end + 1; | |
8903 | if (start < page_end) | |
8904 | goto again; | |
131e404a FDBM |
8905 | } |
8906 | ||
b9d0b389 QW |
8907 | /* |
8908 | * Qgroup reserved space handler | |
8909 | * Page here will be either | |
8910 | * 1) Already written to disk | |
8911 | * In this case, its reserved space is released from data rsv map | |
8912 | * and will be freed by delayed_ref handler finally. | |
8913 | * So even we call qgroup_free_data(), it won't decrease reserved | |
8914 | * space. | |
8915 | * 2) Not written to disk | |
0b34c261 GR |
8916 | * This means the reserved space should be freed here. However, |
8917 | * if a truncate invalidates the page (by clearing PageDirty) | |
8918 | * and the page is accounted for while allocating extent | |
8919 | * in btrfs_check_data_free_space() we let delayed_ref to | |
8920 | * free the entire extent. | |
b9d0b389 | 8921 | */ |
0b34c261 GR |
8922 | if (PageDirty(page)) |
8923 | btrfs_qgroup_free_data(inode, page_start, PAGE_SIZE); | |
131e404a FDBM |
8924 | if (!inode_evicting) { |
8925 | clear_extent_bit(tree, page_start, page_end, | |
8926 | EXTENT_LOCKED | EXTENT_DIRTY | | |
8927 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
8928 | EXTENT_DEFRAG, 1, 1, | |
8929 | &cached_state, GFP_NOFS); | |
8930 | ||
8931 | __btrfs_releasepage(page, GFP_NOFS); | |
e6dcd2dc | 8932 | } |
e6dcd2dc | 8933 | |
4a096752 | 8934 | ClearPageChecked(page); |
9ad6b7bc | 8935 | if (PagePrivate(page)) { |
9ad6b7bc CM |
8936 | ClearPagePrivate(page); |
8937 | set_page_private(page, 0); | |
09cbfeaf | 8938 | put_page(page); |
9ad6b7bc | 8939 | } |
39279cc3 CM |
8940 | } |
8941 | ||
9ebefb18 CM |
8942 | /* |
8943 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
8944 | * called from a page fault handler when a page is first dirtied. Hence we must | |
8945 | * be careful to check for EOF conditions here. We set the page up correctly | |
8946 | * for a written page which means we get ENOSPC checking when writing into | |
8947 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
8948 | * support these features. | |
8949 | * | |
8950 | * We are not allowed to take the i_mutex here so we have to play games to | |
8951 | * protect against truncate races as the page could now be beyond EOF. Because | |
8952 | * vmtruncate() writes the inode size before removing pages, once we have the | |
8953 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
8954 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
8955 | * unlock the page. | |
8956 | */ | |
c2ec175c | 8957 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 8958 | { |
c2ec175c | 8959 | struct page *page = vmf->page; |
496ad9aa | 8960 | struct inode *inode = file_inode(vma->vm_file); |
0b246afa | 8961 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc CM |
8962 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
8963 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 8964 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
8965 | char *kaddr; |
8966 | unsigned long zero_start; | |
9ebefb18 | 8967 | loff_t size; |
1832a6d5 | 8968 | int ret; |
9998eb70 | 8969 | int reserved = 0; |
d0b7da88 | 8970 | u64 reserved_space; |
a52d9a80 | 8971 | u64 page_start; |
e6dcd2dc | 8972 | u64 page_end; |
d0b7da88 CR |
8973 | u64 end; |
8974 | ||
09cbfeaf | 8975 | reserved_space = PAGE_SIZE; |
9ebefb18 | 8976 | |
b2b5ef5c | 8977 | sb_start_pagefault(inode->i_sb); |
df480633 | 8978 | page_start = page_offset(page); |
09cbfeaf | 8979 | page_end = page_start + PAGE_SIZE - 1; |
d0b7da88 | 8980 | end = page_end; |
df480633 | 8981 | |
d0b7da88 CR |
8982 | /* |
8983 | * Reserving delalloc space after obtaining the page lock can lead to | |
8984 | * deadlock. For example, if a dirty page is locked by this function | |
8985 | * and the call to btrfs_delalloc_reserve_space() ends up triggering | |
8986 | * dirty page write out, then the btrfs_writepage() function could | |
8987 | * end up waiting indefinitely to get a lock on the page currently | |
8988 | * being processed by btrfs_page_mkwrite() function. | |
8989 | */ | |
7cf5b976 | 8990 | ret = btrfs_delalloc_reserve_space(inode, page_start, |
d0b7da88 | 8991 | reserved_space); |
9998eb70 | 8992 | if (!ret) { |
e41f941a | 8993 | ret = file_update_time(vma->vm_file); |
9998eb70 CM |
8994 | reserved = 1; |
8995 | } | |
56a76f82 NP |
8996 | if (ret) { |
8997 | if (ret == -ENOMEM) | |
8998 | ret = VM_FAULT_OOM; | |
8999 | else /* -ENOSPC, -EIO, etc */ | |
9000 | ret = VM_FAULT_SIGBUS; | |
9998eb70 CM |
9001 | if (reserved) |
9002 | goto out; | |
9003 | goto out_noreserve; | |
56a76f82 | 9004 | } |
1832a6d5 | 9005 | |
56a76f82 | 9006 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 9007 | again: |
9ebefb18 | 9008 | lock_page(page); |
9ebefb18 | 9009 | size = i_size_read(inode); |
a52d9a80 | 9010 | |
9ebefb18 | 9011 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 9012 | (page_start >= size)) { |
9ebefb18 CM |
9013 | /* page got truncated out from underneath us */ |
9014 | goto out_unlock; | |
9015 | } | |
e6dcd2dc CM |
9016 | wait_on_page_writeback(page); |
9017 | ||
ff13db41 | 9018 | lock_extent_bits(io_tree, page_start, page_end, &cached_state); |
e6dcd2dc CM |
9019 | set_page_extent_mapped(page); |
9020 | ||
eb84ae03 CM |
9021 | /* |
9022 | * we can't set the delalloc bits if there are pending ordered | |
9023 | * extents. Drop our locks and wait for them to finish | |
9024 | */ | |
d0b7da88 | 9025 | ordered = btrfs_lookup_ordered_range(inode, page_start, page_end); |
e6dcd2dc | 9026 | if (ordered) { |
2ac55d41 JB |
9027 | unlock_extent_cached(io_tree, page_start, page_end, |
9028 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 9029 | unlock_page(page); |
eb84ae03 | 9030 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
9031 | btrfs_put_ordered_extent(ordered); |
9032 | goto again; | |
9033 | } | |
9034 | ||
09cbfeaf | 9035 | if (page->index == ((size - 1) >> PAGE_SHIFT)) { |
da17066c | 9036 | reserved_space = round_up(size - page_start, |
0b246afa | 9037 | fs_info->sectorsize); |
09cbfeaf | 9038 | if (reserved_space < PAGE_SIZE) { |
d0b7da88 CR |
9039 | end = page_start + reserved_space - 1; |
9040 | spin_lock(&BTRFS_I(inode)->lock); | |
9041 | BTRFS_I(inode)->outstanding_extents++; | |
9042 | spin_unlock(&BTRFS_I(inode)->lock); | |
9043 | btrfs_delalloc_release_space(inode, page_start, | |
09cbfeaf | 9044 | PAGE_SIZE - reserved_space); |
d0b7da88 CR |
9045 | } |
9046 | } | |
9047 | ||
fbf19087 JB |
9048 | /* |
9049 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
9050 | * if it was already dirty, so for space accounting reasons we need to | |
9051 | * clear any delalloc bits for the range we are fixing to save. There | |
9052 | * is probably a better way to do this, but for now keep consistent with | |
9053 | * prepare_pages in the normal write path. | |
9054 | */ | |
d0b7da88 | 9055 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end, |
9e8a4a8b LB |
9056 | EXTENT_DIRTY | EXTENT_DELALLOC | |
9057 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 9058 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 9059 | |
d0b7da88 | 9060 | ret = btrfs_set_extent_delalloc(inode, page_start, end, |
ba8b04c1 | 9061 | &cached_state, 0); |
9ed74f2d | 9062 | if (ret) { |
2ac55d41 JB |
9063 | unlock_extent_cached(io_tree, page_start, page_end, |
9064 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
9065 | ret = VM_FAULT_SIGBUS; |
9066 | goto out_unlock; | |
9067 | } | |
e6dcd2dc | 9068 | ret = 0; |
9ebefb18 CM |
9069 | |
9070 | /* page is wholly or partially inside EOF */ | |
09cbfeaf KS |
9071 | if (page_start + PAGE_SIZE > size) |
9072 | zero_start = size & ~PAGE_MASK; | |
9ebefb18 | 9073 | else |
09cbfeaf | 9074 | zero_start = PAGE_SIZE; |
9ebefb18 | 9075 | |
09cbfeaf | 9076 | if (zero_start != PAGE_SIZE) { |
e6dcd2dc | 9077 | kaddr = kmap(page); |
09cbfeaf | 9078 | memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start); |
e6dcd2dc CM |
9079 | flush_dcache_page(page); |
9080 | kunmap(page); | |
9081 | } | |
247e743c | 9082 | ClearPageChecked(page); |
e6dcd2dc | 9083 | set_page_dirty(page); |
50a9b214 | 9084 | SetPageUptodate(page); |
5a3f23d5 | 9085 | |
0b246afa | 9086 | BTRFS_I(inode)->last_trans = fs_info->generation; |
257c62e1 | 9087 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; |
46d8bc34 | 9088 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit; |
257c62e1 | 9089 | |
2ac55d41 | 9090 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
9091 | |
9092 | out_unlock: | |
b2b5ef5c JK |
9093 | if (!ret) { |
9094 | sb_end_pagefault(inode->i_sb); | |
50a9b214 | 9095 | return VM_FAULT_LOCKED; |
b2b5ef5c | 9096 | } |
9ebefb18 | 9097 | unlock_page(page); |
1832a6d5 | 9098 | out: |
d0b7da88 | 9099 | btrfs_delalloc_release_space(inode, page_start, reserved_space); |
9998eb70 | 9100 | out_noreserve: |
b2b5ef5c | 9101 | sb_end_pagefault(inode->i_sb); |
9ebefb18 CM |
9102 | return ret; |
9103 | } | |
9104 | ||
a41ad394 | 9105 | static int btrfs_truncate(struct inode *inode) |
39279cc3 | 9106 | { |
0b246afa | 9107 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
39279cc3 | 9108 | struct btrfs_root *root = BTRFS_I(inode)->root; |
fcb80c2a | 9109 | struct btrfs_block_rsv *rsv; |
a71754fc | 9110 | int ret = 0; |
3893e33b | 9111 | int err = 0; |
39279cc3 | 9112 | struct btrfs_trans_handle *trans; |
0b246afa JM |
9113 | u64 mask = fs_info->sectorsize - 1; |
9114 | u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1); | |
39279cc3 | 9115 | |
0ef8b726 JB |
9116 | ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask), |
9117 | (u64)-1); | |
9118 | if (ret) | |
9119 | return ret; | |
39279cc3 | 9120 | |
fcb80c2a | 9121 | /* |
01327610 | 9122 | * Yes ladies and gentlemen, this is indeed ugly. The fact is we have |
fcb80c2a JB |
9123 | * 3 things going on here |
9124 | * | |
9125 | * 1) We need to reserve space for our orphan item and the space to | |
9126 | * delete our orphan item. Lord knows we don't want to have a dangling | |
9127 | * orphan item because we didn't reserve space to remove it. | |
9128 | * | |
9129 | * 2) We need to reserve space to update our inode. | |
9130 | * | |
9131 | * 3) We need to have something to cache all the space that is going to | |
9132 | * be free'd up by the truncate operation, but also have some slack | |
9133 | * space reserved in case it uses space during the truncate (thank you | |
9134 | * very much snapshotting). | |
9135 | * | |
01327610 | 9136 | * And we need these to all be separate. The fact is we can use a lot of |
fcb80c2a | 9137 | * space doing the truncate, and we have no earthly idea how much space |
01327610 | 9138 | * we will use, so we need the truncate reservation to be separate so it |
fcb80c2a JB |
9139 | * doesn't end up using space reserved for updating the inode or |
9140 | * removing the orphan item. We also need to be able to stop the | |
9141 | * transaction and start a new one, which means we need to be able to | |
9142 | * update the inode several times, and we have no idea of knowing how | |
9143 | * many times that will be, so we can't just reserve 1 item for the | |
01327610 | 9144 | * entirety of the operation, so that has to be done separately as well. |
fcb80c2a JB |
9145 | * Then there is the orphan item, which does indeed need to be held on |
9146 | * to for the whole operation, and we need nobody to touch this reserved | |
9147 | * space except the orphan code. | |
9148 | * | |
9149 | * So that leaves us with | |
9150 | * | |
9151 | * 1) root->orphan_block_rsv - for the orphan deletion. | |
9152 | * 2) rsv - for the truncate reservation, which we will steal from the | |
9153 | * transaction reservation. | |
9154 | * 3) fs_info->trans_block_rsv - this will have 1 items worth left for | |
9155 | * updating the inode. | |
9156 | */ | |
2ff7e61e | 9157 | rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
9158 | if (!rsv) |
9159 | return -ENOMEM; | |
4a338542 | 9160 | rsv->size = min_size; |
ca7e70f5 | 9161 | rsv->failfast = 1; |
f0cd846e | 9162 | |
907cbceb | 9163 | /* |
07127184 | 9164 | * 1 for the truncate slack space |
907cbceb JB |
9165 | * 1 for updating the inode. |
9166 | */ | |
f3fe820c | 9167 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a JB |
9168 | if (IS_ERR(trans)) { |
9169 | err = PTR_ERR(trans); | |
9170 | goto out; | |
9171 | } | |
f0cd846e | 9172 | |
907cbceb | 9173 | /* Migrate the slack space for the truncate to our reserve */ |
0b246afa | 9174 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv, |
25d609f8 | 9175 | min_size, 0); |
fcb80c2a | 9176 | BUG_ON(ret); |
f0cd846e | 9177 | |
5dc562c5 JB |
9178 | /* |
9179 | * So if we truncate and then write and fsync we normally would just | |
9180 | * write the extents that changed, which is a problem if we need to | |
9181 | * first truncate that entire inode. So set this flag so we write out | |
9182 | * all of the extents in the inode to the sync log so we're completely | |
9183 | * safe. | |
9184 | */ | |
9185 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
ca7e70f5 | 9186 | trans->block_rsv = rsv; |
907cbceb | 9187 | |
8082510e YZ |
9188 | while (1) { |
9189 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
9190 | inode->i_size, | |
9191 | BTRFS_EXTENT_DATA_KEY); | |
28ed1345 | 9192 | if (ret != -ENOSPC && ret != -EAGAIN) { |
3893e33b | 9193 | err = ret; |
8082510e | 9194 | break; |
3893e33b | 9195 | } |
39279cc3 | 9196 | |
0b246afa | 9197 | trans->block_rsv = &fs_info->trans_block_rsv; |
8082510e | 9198 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
9199 | if (ret) { |
9200 | err = ret; | |
9201 | break; | |
9202 | } | |
ca7e70f5 | 9203 | |
3a45bb20 | 9204 | btrfs_end_transaction(trans); |
2ff7e61e | 9205 | btrfs_btree_balance_dirty(fs_info); |
ca7e70f5 JB |
9206 | |
9207 | trans = btrfs_start_transaction(root, 2); | |
9208 | if (IS_ERR(trans)) { | |
9209 | ret = err = PTR_ERR(trans); | |
9210 | trans = NULL; | |
9211 | break; | |
9212 | } | |
9213 | ||
47b5d646 | 9214 | btrfs_block_rsv_release(fs_info, rsv, -1); |
0b246afa | 9215 | ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, |
25d609f8 | 9216 | rsv, min_size, 0); |
ca7e70f5 JB |
9217 | BUG_ON(ret); /* shouldn't happen */ |
9218 | trans->block_rsv = rsv; | |
8082510e YZ |
9219 | } |
9220 | ||
9221 | if (ret == 0 && inode->i_nlink > 0) { | |
fcb80c2a | 9222 | trans->block_rsv = root->orphan_block_rsv; |
8082510e | 9223 | ret = btrfs_orphan_del(trans, inode); |
3893e33b JB |
9224 | if (ret) |
9225 | err = ret; | |
8082510e YZ |
9226 | } |
9227 | ||
917c16b2 | 9228 | if (trans) { |
0b246afa | 9229 | trans->block_rsv = &fs_info->trans_block_rsv; |
917c16b2 CM |
9230 | ret = btrfs_update_inode(trans, root, inode); |
9231 | if (ret && !err) | |
9232 | err = ret; | |
7b128766 | 9233 | |
3a45bb20 | 9234 | ret = btrfs_end_transaction(trans); |
2ff7e61e | 9235 | btrfs_btree_balance_dirty(fs_info); |
917c16b2 | 9236 | } |
fcb80c2a | 9237 | out: |
2ff7e61e | 9238 | btrfs_free_block_rsv(fs_info, rsv); |
fcb80c2a | 9239 | |
3893e33b JB |
9240 | if (ret && !err) |
9241 | err = ret; | |
a41ad394 | 9242 | |
3893e33b | 9243 | return err; |
39279cc3 CM |
9244 | } |
9245 | ||
d352ac68 CM |
9246 | /* |
9247 | * create a new subvolume directory/inode (helper for the ioctl). | |
9248 | */ | |
d2fb3437 | 9249 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
63541927 FDBM |
9250 | struct btrfs_root *new_root, |
9251 | struct btrfs_root *parent_root, | |
9252 | u64 new_dirid) | |
39279cc3 | 9253 | { |
39279cc3 | 9254 | struct inode *inode; |
76dda93c | 9255 | int err; |
00e4e6b3 | 9256 | u64 index = 0; |
39279cc3 | 9257 | |
12fc9d09 FA |
9258 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
9259 | new_dirid, new_dirid, | |
9260 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
9261 | &index); | |
54aa1f4d | 9262 | if (IS_ERR(inode)) |
f46b5a66 | 9263 | return PTR_ERR(inode); |
39279cc3 CM |
9264 | inode->i_op = &btrfs_dir_inode_operations; |
9265 | inode->i_fop = &btrfs_dir_file_operations; | |
9266 | ||
bfe86848 | 9267 | set_nlink(inode, 1); |
dbe674a9 | 9268 | btrfs_i_size_write(inode, 0); |
b0d5d10f | 9269 | unlock_new_inode(inode); |
3b96362c | 9270 | |
63541927 FDBM |
9271 | err = btrfs_subvol_inherit_props(trans, new_root, parent_root); |
9272 | if (err) | |
9273 | btrfs_err(new_root->fs_info, | |
351fd353 | 9274 | "error inheriting subvolume %llu properties: %d", |
63541927 FDBM |
9275 | new_root->root_key.objectid, err); |
9276 | ||
76dda93c | 9277 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 9278 | |
76dda93c | 9279 | iput(inode); |
ce598979 | 9280 | return err; |
39279cc3 CM |
9281 | } |
9282 | ||
39279cc3 CM |
9283 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
9284 | { | |
9285 | struct btrfs_inode *ei; | |
2ead6ae7 | 9286 | struct inode *inode; |
39279cc3 CM |
9287 | |
9288 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
9289 | if (!ei) | |
9290 | return NULL; | |
2ead6ae7 YZ |
9291 | |
9292 | ei->root = NULL; | |
2ead6ae7 | 9293 | ei->generation = 0; |
15ee9bc7 | 9294 | ei->last_trans = 0; |
257c62e1 | 9295 | ei->last_sub_trans = 0; |
e02119d5 | 9296 | ei->logged_trans = 0; |
2ead6ae7 | 9297 | ei->delalloc_bytes = 0; |
47059d93 | 9298 | ei->defrag_bytes = 0; |
2ead6ae7 YZ |
9299 | ei->disk_i_size = 0; |
9300 | ei->flags = 0; | |
7709cde3 | 9301 | ei->csum_bytes = 0; |
2ead6ae7 | 9302 | ei->index_cnt = (u64)-1; |
67de1176 | 9303 | ei->dir_index = 0; |
2ead6ae7 | 9304 | ei->last_unlink_trans = 0; |
46d8bc34 | 9305 | ei->last_log_commit = 0; |
8089fe62 | 9306 | ei->delayed_iput_count = 0; |
2ead6ae7 | 9307 | |
9e0baf60 JB |
9308 | spin_lock_init(&ei->lock); |
9309 | ei->outstanding_extents = 0; | |
9310 | ei->reserved_extents = 0; | |
2ead6ae7 | 9311 | |
72ac3c0d | 9312 | ei->runtime_flags = 0; |
261507a0 | 9313 | ei->force_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 9314 | |
16cdcec7 MX |
9315 | ei->delayed_node = NULL; |
9316 | ||
9cc97d64 | 9317 | ei->i_otime.tv_sec = 0; |
9318 | ei->i_otime.tv_nsec = 0; | |
9319 | ||
2ead6ae7 | 9320 | inode = &ei->vfs_inode; |
a8067e02 | 9321 | extent_map_tree_init(&ei->extent_tree); |
f993c883 DS |
9322 | extent_io_tree_init(&ei->io_tree, &inode->i_data); |
9323 | extent_io_tree_init(&ei->io_failure_tree, &inode->i_data); | |
0b32f4bb JB |
9324 | ei->io_tree.track_uptodate = 1; |
9325 | ei->io_failure_tree.track_uptodate = 1; | |
b812ce28 | 9326 | atomic_set(&ei->sync_writers, 0); |
2ead6ae7 | 9327 | mutex_init(&ei->log_mutex); |
f248679e | 9328 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 9329 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 9330 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
8089fe62 | 9331 | INIT_LIST_HEAD(&ei->delayed_iput); |
2ead6ae7 | 9332 | RB_CLEAR_NODE(&ei->rb_node); |
5f9a8a51 | 9333 | init_rwsem(&ei->dio_sem); |
2ead6ae7 YZ |
9334 | |
9335 | return inode; | |
39279cc3 CM |
9336 | } |
9337 | ||
aaedb55b JB |
9338 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
9339 | void btrfs_test_destroy_inode(struct inode *inode) | |
9340 | { | |
9341 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); | |
9342 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); | |
9343 | } | |
9344 | #endif | |
9345 | ||
fa0d7e3d NP |
9346 | static void btrfs_i_callback(struct rcu_head *head) |
9347 | { | |
9348 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
9349 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
9350 | } | |
9351 | ||
39279cc3 CM |
9352 | void btrfs_destroy_inode(struct inode *inode) |
9353 | { | |
0b246afa | 9354 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
e6dcd2dc | 9355 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
9356 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9357 | ||
b3d9b7a3 | 9358 | WARN_ON(!hlist_empty(&inode->i_dentry)); |
39279cc3 | 9359 | WARN_ON(inode->i_data.nrpages); |
9e0baf60 JB |
9360 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
9361 | WARN_ON(BTRFS_I(inode)->reserved_extents); | |
7709cde3 JB |
9362 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
9363 | WARN_ON(BTRFS_I(inode)->csum_bytes); | |
47059d93 | 9364 | WARN_ON(BTRFS_I(inode)->defrag_bytes); |
39279cc3 | 9365 | |
a6dbd429 JB |
9366 | /* |
9367 | * This can happen where we create an inode, but somebody else also | |
9368 | * created the same inode and we need to destroy the one we already | |
9369 | * created. | |
9370 | */ | |
9371 | if (!root) | |
9372 | goto free; | |
9373 | ||
8a35d95f JB |
9374 | if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
9375 | &BTRFS_I(inode)->runtime_flags)) { | |
0b246afa | 9376 | btrfs_info(fs_info, "inode %llu still on the orphan list", |
4a0cc7ca | 9377 | btrfs_ino(BTRFS_I(inode))); |
8a35d95f | 9378 | atomic_dec(&root->orphan_inodes); |
7b128766 | 9379 | } |
7b128766 | 9380 | |
d397712b | 9381 | while (1) { |
e6dcd2dc CM |
9382 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
9383 | if (!ordered) | |
9384 | break; | |
9385 | else { | |
0b246afa | 9386 | btrfs_err(fs_info, |
5d163e0e JM |
9387 | "found ordered extent %llu %llu on inode cleanup", |
9388 | ordered->file_offset, ordered->len); | |
e6dcd2dc CM |
9389 | btrfs_remove_ordered_extent(inode, ordered); |
9390 | btrfs_put_ordered_extent(ordered); | |
9391 | btrfs_put_ordered_extent(ordered); | |
9392 | } | |
9393 | } | |
56fa9d07 | 9394 | btrfs_qgroup_check_reserved_leak(inode); |
5d4f98a2 | 9395 | inode_tree_del(inode); |
5b21f2ed | 9396 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 9397 | free: |
fa0d7e3d | 9398 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
9399 | } |
9400 | ||
45321ac5 | 9401 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
9402 | { |
9403 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 9404 | |
6379ef9f NA |
9405 | if (root == NULL) |
9406 | return 1; | |
9407 | ||
fa6ac876 | 9408 | /* the snap/subvol tree is on deleting */ |
69e9c6c6 | 9409 | if (btrfs_root_refs(&root->root_item) == 0) |
45321ac5 | 9410 | return 1; |
76dda93c | 9411 | else |
45321ac5 | 9412 | return generic_drop_inode(inode); |
76dda93c YZ |
9413 | } |
9414 | ||
0ee0fda0 | 9415 | static void init_once(void *foo) |
39279cc3 CM |
9416 | { |
9417 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
9418 | ||
9419 | inode_init_once(&ei->vfs_inode); | |
9420 | } | |
9421 | ||
9422 | void btrfs_destroy_cachep(void) | |
9423 | { | |
8c0a8537 KS |
9424 | /* |
9425 | * Make sure all delayed rcu free inodes are flushed before we | |
9426 | * destroy cache. | |
9427 | */ | |
9428 | rcu_barrier(); | |
5598e900 KM |
9429 | kmem_cache_destroy(btrfs_inode_cachep); |
9430 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
9431 | kmem_cache_destroy(btrfs_transaction_cachep); | |
9432 | kmem_cache_destroy(btrfs_path_cachep); | |
9433 | kmem_cache_destroy(btrfs_free_space_cachep); | |
39279cc3 CM |
9434 | } |
9435 | ||
9436 | int btrfs_init_cachep(void) | |
9437 | { | |
837e1972 | 9438 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 | 9439 | sizeof(struct btrfs_inode), 0, |
5d097056 VD |
9440 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT, |
9441 | init_once); | |
39279cc3 CM |
9442 | if (!btrfs_inode_cachep) |
9443 | goto fail; | |
9601e3f6 | 9444 | |
837e1972 | 9445 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 | 9446 | sizeof(struct btrfs_trans_handle), 0, |
fba4b697 | 9447 | SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
9448 | if (!btrfs_trans_handle_cachep) |
9449 | goto fail; | |
9601e3f6 | 9450 | |
837e1972 | 9451 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction", |
9601e3f6 | 9452 | sizeof(struct btrfs_transaction), 0, |
fba4b697 | 9453 | SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
9454 | if (!btrfs_transaction_cachep) |
9455 | goto fail; | |
9601e3f6 | 9456 | |
837e1972 | 9457 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 | 9458 | sizeof(struct btrfs_path), 0, |
fba4b697 | 9459 | SLAB_MEM_SPREAD, NULL); |
39279cc3 CM |
9460 | if (!btrfs_path_cachep) |
9461 | goto fail; | |
9601e3f6 | 9462 | |
837e1972 | 9463 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 | 9464 | sizeof(struct btrfs_free_space), 0, |
fba4b697 | 9465 | SLAB_MEM_SPREAD, NULL); |
dc89e982 JB |
9466 | if (!btrfs_free_space_cachep) |
9467 | goto fail; | |
9468 | ||
39279cc3 CM |
9469 | return 0; |
9470 | fail: | |
9471 | btrfs_destroy_cachep(); | |
9472 | return -ENOMEM; | |
9473 | } | |
9474 | ||
9475 | static int btrfs_getattr(struct vfsmount *mnt, | |
9476 | struct dentry *dentry, struct kstat *stat) | |
9477 | { | |
df0af1a5 | 9478 | u64 delalloc_bytes; |
2b0143b5 | 9479 | struct inode *inode = d_inode(dentry); |
fadc0d8b DS |
9480 | u32 blocksize = inode->i_sb->s_blocksize; |
9481 | ||
39279cc3 | 9482 | generic_fillattr(inode, stat); |
0ee5dc67 | 9483 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
df0af1a5 MX |
9484 | |
9485 | spin_lock(&BTRFS_I(inode)->lock); | |
9486 | delalloc_bytes = BTRFS_I(inode)->delalloc_bytes; | |
9487 | spin_unlock(&BTRFS_I(inode)->lock); | |
fadc0d8b | 9488 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
df0af1a5 | 9489 | ALIGN(delalloc_bytes, blocksize)) >> 9; |
39279cc3 CM |
9490 | return 0; |
9491 | } | |
9492 | ||
cdd1fedf DF |
9493 | static int btrfs_rename_exchange(struct inode *old_dir, |
9494 | struct dentry *old_dentry, | |
9495 | struct inode *new_dir, | |
9496 | struct dentry *new_dentry) | |
9497 | { | |
0b246afa | 9498 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
cdd1fedf DF |
9499 | struct btrfs_trans_handle *trans; |
9500 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
9501 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; | |
9502 | struct inode *new_inode = new_dentry->d_inode; | |
9503 | struct inode *old_inode = old_dentry->d_inode; | |
c2050a45 | 9504 | struct timespec ctime = current_time(old_inode); |
cdd1fedf | 9505 | struct dentry *parent; |
4a0cc7ca NB |
9506 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
9507 | u64 new_ino = btrfs_ino(BTRFS_I(new_inode)); | |
cdd1fedf DF |
9508 | u64 old_idx = 0; |
9509 | u64 new_idx = 0; | |
9510 | u64 root_objectid; | |
9511 | int ret; | |
86e8aa0e FM |
9512 | bool root_log_pinned = false; |
9513 | bool dest_log_pinned = false; | |
cdd1fedf DF |
9514 | |
9515 | /* we only allow rename subvolume link between subvolumes */ | |
9516 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) | |
9517 | return -EXDEV; | |
9518 | ||
9519 | /* close the race window with snapshot create/destroy ioctl */ | |
9520 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 9521 | down_read(&fs_info->subvol_sem); |
cdd1fedf | 9522 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9523 | down_read(&fs_info->subvol_sem); |
cdd1fedf DF |
9524 | |
9525 | /* | |
9526 | * We want to reserve the absolute worst case amount of items. So if | |
9527 | * both inodes are subvols and we need to unlink them then that would | |
9528 | * require 4 item modifications, but if they are both normal inodes it | |
9529 | * would require 5 item modifications, so we'll assume their normal | |
9530 | * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items | |
9531 | * should cover the worst case number of items we'll modify. | |
9532 | */ | |
9533 | trans = btrfs_start_transaction(root, 12); | |
9534 | if (IS_ERR(trans)) { | |
9535 | ret = PTR_ERR(trans); | |
9536 | goto out_notrans; | |
9537 | } | |
9538 | ||
9539 | /* | |
9540 | * We need to find a free sequence number both in the source and | |
9541 | * in the destination directory for the exchange. | |
9542 | */ | |
9543 | ret = btrfs_set_inode_index(new_dir, &old_idx); | |
9544 | if (ret) | |
9545 | goto out_fail; | |
9546 | ret = btrfs_set_inode_index(old_dir, &new_idx); | |
9547 | if (ret) | |
9548 | goto out_fail; | |
9549 | ||
9550 | BTRFS_I(old_inode)->dir_index = 0ULL; | |
9551 | BTRFS_I(new_inode)->dir_index = 0ULL; | |
9552 | ||
9553 | /* Reference for the source. */ | |
9554 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9555 | /* force full log commit if subvolume involved. */ | |
0b246afa | 9556 | btrfs_set_log_full_commit(fs_info, trans); |
cdd1fedf | 9557 | } else { |
376e5a57 FM |
9558 | btrfs_pin_log_trans(root); |
9559 | root_log_pinned = true; | |
cdd1fedf DF |
9560 | ret = btrfs_insert_inode_ref(trans, dest, |
9561 | new_dentry->d_name.name, | |
9562 | new_dentry->d_name.len, | |
9563 | old_ino, | |
f85b7379 DS |
9564 | btrfs_ino(BTRFS_I(new_dir)), |
9565 | old_idx); | |
cdd1fedf DF |
9566 | if (ret) |
9567 | goto out_fail; | |
cdd1fedf DF |
9568 | } |
9569 | ||
9570 | /* And now for the dest. */ | |
9571 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9572 | /* force full log commit if subvolume involved. */ | |
0b246afa | 9573 | btrfs_set_log_full_commit(fs_info, trans); |
cdd1fedf | 9574 | } else { |
376e5a57 FM |
9575 | btrfs_pin_log_trans(dest); |
9576 | dest_log_pinned = true; | |
cdd1fedf DF |
9577 | ret = btrfs_insert_inode_ref(trans, root, |
9578 | old_dentry->d_name.name, | |
9579 | old_dentry->d_name.len, | |
9580 | new_ino, | |
f85b7379 DS |
9581 | btrfs_ino(BTRFS_I(old_dir)), |
9582 | new_idx); | |
cdd1fedf DF |
9583 | if (ret) |
9584 | goto out_fail; | |
cdd1fedf DF |
9585 | } |
9586 | ||
9587 | /* Update inode version and ctime/mtime. */ | |
9588 | inode_inc_iversion(old_dir); | |
9589 | inode_inc_iversion(new_dir); | |
9590 | inode_inc_iversion(old_inode); | |
9591 | inode_inc_iversion(new_inode); | |
9592 | old_dir->i_ctime = old_dir->i_mtime = ctime; | |
9593 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
9594 | old_inode->i_ctime = ctime; | |
9595 | new_inode->i_ctime = ctime; | |
9596 | ||
9597 | if (old_dentry->d_parent != new_dentry->d_parent) { | |
f85b7379 DS |
9598 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
9599 | BTRFS_I(old_inode), 1); | |
9600 | btrfs_record_unlink_dir(trans, BTRFS_I(new_dir), | |
9601 | BTRFS_I(new_inode), 1); | |
cdd1fedf DF |
9602 | } |
9603 | ||
9604 | /* src is a subvolume */ | |
9605 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9606 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; | |
9607 | ret = btrfs_unlink_subvol(trans, root, old_dir, | |
9608 | root_objectid, | |
9609 | old_dentry->d_name.name, | |
9610 | old_dentry->d_name.len); | |
9611 | } else { /* src is an inode */ | |
4ec5934e NB |
9612 | ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir), |
9613 | BTRFS_I(old_dentry->d_inode), | |
cdd1fedf DF |
9614 | old_dentry->d_name.name, |
9615 | old_dentry->d_name.len); | |
9616 | if (!ret) | |
9617 | ret = btrfs_update_inode(trans, root, old_inode); | |
9618 | } | |
9619 | if (ret) { | |
66642832 | 9620 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9621 | goto out_fail; |
9622 | } | |
9623 | ||
9624 | /* dest is a subvolume */ | |
9625 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) { | |
9626 | root_objectid = BTRFS_I(new_inode)->root->root_key.objectid; | |
9627 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
9628 | root_objectid, | |
9629 | new_dentry->d_name.name, | |
9630 | new_dentry->d_name.len); | |
9631 | } else { /* dest is an inode */ | |
4ec5934e NB |
9632 | ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir), |
9633 | BTRFS_I(new_dentry->d_inode), | |
cdd1fedf DF |
9634 | new_dentry->d_name.name, |
9635 | new_dentry->d_name.len); | |
9636 | if (!ret) | |
9637 | ret = btrfs_update_inode(trans, dest, new_inode); | |
9638 | } | |
9639 | if (ret) { | |
66642832 | 9640 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9641 | goto out_fail; |
9642 | } | |
9643 | ||
9644 | ret = btrfs_add_link(trans, new_dir, old_inode, | |
9645 | new_dentry->d_name.name, | |
9646 | new_dentry->d_name.len, 0, old_idx); | |
9647 | if (ret) { | |
66642832 | 9648 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9649 | goto out_fail; |
9650 | } | |
9651 | ||
9652 | ret = btrfs_add_link(trans, old_dir, new_inode, | |
9653 | old_dentry->d_name.name, | |
9654 | old_dentry->d_name.len, 0, new_idx); | |
9655 | if (ret) { | |
66642832 | 9656 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9657 | goto out_fail; |
9658 | } | |
9659 | ||
9660 | if (old_inode->i_nlink == 1) | |
9661 | BTRFS_I(old_inode)->dir_index = old_idx; | |
9662 | if (new_inode->i_nlink == 1) | |
9663 | BTRFS_I(new_inode)->dir_index = new_idx; | |
9664 | ||
86e8aa0e | 9665 | if (root_log_pinned) { |
cdd1fedf | 9666 | parent = new_dentry->d_parent; |
f85b7379 DS |
9667 | btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir), |
9668 | parent); | |
cdd1fedf | 9669 | btrfs_end_log_trans(root); |
86e8aa0e | 9670 | root_log_pinned = false; |
cdd1fedf | 9671 | } |
86e8aa0e | 9672 | if (dest_log_pinned) { |
cdd1fedf | 9673 | parent = old_dentry->d_parent; |
f85b7379 DS |
9674 | btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir), |
9675 | parent); | |
cdd1fedf | 9676 | btrfs_end_log_trans(dest); |
86e8aa0e | 9677 | dest_log_pinned = false; |
cdd1fedf DF |
9678 | } |
9679 | out_fail: | |
86e8aa0e FM |
9680 | /* |
9681 | * If we have pinned a log and an error happened, we unpin tasks | |
9682 | * trying to sync the log and force them to fallback to a transaction | |
9683 | * commit if the log currently contains any of the inodes involved in | |
9684 | * this rename operation (to ensure we do not persist a log with an | |
9685 | * inconsistent state for any of these inodes or leading to any | |
9686 | * inconsistencies when replayed). If the transaction was aborted, the | |
9687 | * abortion reason is propagated to userspace when attempting to commit | |
9688 | * the transaction. If the log does not contain any of these inodes, we | |
9689 | * allow the tasks to sync it. | |
9690 | */ | |
9691 | if (ret && (root_log_pinned || dest_log_pinned)) { | |
0f8939b8 NB |
9692 | if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) || |
9693 | btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) || | |
9694 | btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) || | |
86e8aa0e | 9695 | (new_inode && |
0f8939b8 | 9696 | btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation))) |
0b246afa | 9697 | btrfs_set_log_full_commit(fs_info, trans); |
86e8aa0e FM |
9698 | |
9699 | if (root_log_pinned) { | |
9700 | btrfs_end_log_trans(root); | |
9701 | root_log_pinned = false; | |
9702 | } | |
9703 | if (dest_log_pinned) { | |
9704 | btrfs_end_log_trans(dest); | |
9705 | dest_log_pinned = false; | |
9706 | } | |
9707 | } | |
3a45bb20 | 9708 | ret = btrfs_end_transaction(trans); |
cdd1fedf DF |
9709 | out_notrans: |
9710 | if (new_ino == BTRFS_FIRST_FREE_OBJECTID) | |
0b246afa | 9711 | up_read(&fs_info->subvol_sem); |
cdd1fedf | 9712 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9713 | up_read(&fs_info->subvol_sem); |
cdd1fedf DF |
9714 | |
9715 | return ret; | |
9716 | } | |
9717 | ||
9718 | static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans, | |
9719 | struct btrfs_root *root, | |
9720 | struct inode *dir, | |
9721 | struct dentry *dentry) | |
9722 | { | |
9723 | int ret; | |
9724 | struct inode *inode; | |
9725 | u64 objectid; | |
9726 | u64 index; | |
9727 | ||
9728 | ret = btrfs_find_free_ino(root, &objectid); | |
9729 | if (ret) | |
9730 | return ret; | |
9731 | ||
9732 | inode = btrfs_new_inode(trans, root, dir, | |
9733 | dentry->d_name.name, | |
9734 | dentry->d_name.len, | |
4a0cc7ca | 9735 | btrfs_ino(BTRFS_I(dir)), |
cdd1fedf DF |
9736 | objectid, |
9737 | S_IFCHR | WHITEOUT_MODE, | |
9738 | &index); | |
9739 | ||
9740 | if (IS_ERR(inode)) { | |
9741 | ret = PTR_ERR(inode); | |
9742 | return ret; | |
9743 | } | |
9744 | ||
9745 | inode->i_op = &btrfs_special_inode_operations; | |
9746 | init_special_inode(inode, inode->i_mode, | |
9747 | WHITEOUT_DEV); | |
9748 | ||
9749 | ret = btrfs_init_inode_security(trans, inode, dir, | |
9750 | &dentry->d_name); | |
9751 | if (ret) | |
c9901618 | 9752 | goto out; |
cdd1fedf DF |
9753 | |
9754 | ret = btrfs_add_nondir(trans, dir, dentry, | |
9755 | inode, 0, index); | |
9756 | if (ret) | |
c9901618 | 9757 | goto out; |
cdd1fedf DF |
9758 | |
9759 | ret = btrfs_update_inode(trans, root, inode); | |
c9901618 | 9760 | out: |
cdd1fedf | 9761 | unlock_new_inode(inode); |
c9901618 FM |
9762 | if (ret) |
9763 | inode_dec_link_count(inode); | |
cdd1fedf DF |
9764 | iput(inode); |
9765 | ||
c9901618 | 9766 | return ret; |
cdd1fedf DF |
9767 | } |
9768 | ||
d397712b | 9769 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
cdd1fedf DF |
9770 | struct inode *new_dir, struct dentry *new_dentry, |
9771 | unsigned int flags) | |
39279cc3 | 9772 | { |
0b246afa | 9773 | struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb); |
39279cc3 | 9774 | struct btrfs_trans_handle *trans; |
5062af35 | 9775 | unsigned int trans_num_items; |
39279cc3 | 9776 | struct btrfs_root *root = BTRFS_I(old_dir)->root; |
4df27c4d | 9777 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
2b0143b5 DH |
9778 | struct inode *new_inode = d_inode(new_dentry); |
9779 | struct inode *old_inode = d_inode(old_dentry); | |
00e4e6b3 | 9780 | u64 index = 0; |
4df27c4d | 9781 | u64 root_objectid; |
39279cc3 | 9782 | int ret; |
4a0cc7ca | 9783 | u64 old_ino = btrfs_ino(BTRFS_I(old_inode)); |
3dc9e8f7 | 9784 | bool log_pinned = false; |
39279cc3 | 9785 | |
4a0cc7ca | 9786 | if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
9787 | return -EPERM; |
9788 | ||
4df27c4d | 9789 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 9790 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
9791 | return -EXDEV; |
9792 | ||
33345d01 | 9793 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
4a0cc7ca | 9794 | (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID)) |
39279cc3 | 9795 | return -ENOTEMPTY; |
5f39d397 | 9796 | |
4df27c4d YZ |
9797 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
9798 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
9799 | return -ENOTEMPTY; | |
9c52057c CM |
9800 | |
9801 | ||
9802 | /* check for collisions, even if the name isn't there */ | |
4871c158 | 9803 | ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, |
9c52057c CM |
9804 | new_dentry->d_name.name, |
9805 | new_dentry->d_name.len); | |
9806 | ||
9807 | if (ret) { | |
9808 | if (ret == -EEXIST) { | |
9809 | /* we shouldn't get | |
9810 | * eexist without a new_inode */ | |
fae7f21c | 9811 | if (WARN_ON(!new_inode)) { |
9c52057c CM |
9812 | return ret; |
9813 | } | |
9814 | } else { | |
9815 | /* maybe -EOVERFLOW */ | |
9816 | return ret; | |
9817 | } | |
9818 | } | |
9819 | ret = 0; | |
9820 | ||
5a3f23d5 | 9821 | /* |
8d875f95 CM |
9822 | * we're using rename to replace one file with another. Start IO on it |
9823 | * now so we don't add too much work to the end of the transaction | |
5a3f23d5 | 9824 | */ |
8d875f95 | 9825 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size) |
5a3f23d5 CM |
9826 | filemap_flush(old_inode->i_mapping); |
9827 | ||
76dda93c | 9828 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 9829 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9830 | down_read(&fs_info->subvol_sem); |
a22285a6 YZ |
9831 | /* |
9832 | * We want to reserve the absolute worst case amount of items. So if | |
9833 | * both inodes are subvols and we need to unlink them then that would | |
9834 | * require 4 item modifications, but if they are both normal inodes it | |
cdd1fedf | 9835 | * would require 5 item modifications, so we'll assume they are normal |
a22285a6 YZ |
9836 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items |
9837 | * should cover the worst case number of items we'll modify. | |
5062af35 FM |
9838 | * If our rename has the whiteout flag, we need more 5 units for the |
9839 | * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item | |
9840 | * when selinux is enabled). | |
a22285a6 | 9841 | */ |
5062af35 FM |
9842 | trans_num_items = 11; |
9843 | if (flags & RENAME_WHITEOUT) | |
9844 | trans_num_items += 5; | |
9845 | trans = btrfs_start_transaction(root, trans_num_items); | |
b44c59a8 | 9846 | if (IS_ERR(trans)) { |
cdd1fedf DF |
9847 | ret = PTR_ERR(trans); |
9848 | goto out_notrans; | |
9849 | } | |
76dda93c | 9850 | |
4df27c4d YZ |
9851 | if (dest != root) |
9852 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 9853 | |
a5719521 YZ |
9854 | ret = btrfs_set_inode_index(new_dir, &index); |
9855 | if (ret) | |
9856 | goto out_fail; | |
5a3f23d5 | 9857 | |
67de1176 | 9858 | BTRFS_I(old_inode)->dir_index = 0ULL; |
33345d01 | 9859 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d | 9860 | /* force full log commit if subvolume involved. */ |
0b246afa | 9861 | btrfs_set_log_full_commit(fs_info, trans); |
4df27c4d | 9862 | } else { |
c4aba954 FM |
9863 | btrfs_pin_log_trans(root); |
9864 | log_pinned = true; | |
a5719521 YZ |
9865 | ret = btrfs_insert_inode_ref(trans, dest, |
9866 | new_dentry->d_name.name, | |
9867 | new_dentry->d_name.len, | |
33345d01 | 9868 | old_ino, |
4a0cc7ca | 9869 | btrfs_ino(BTRFS_I(new_dir)), index); |
a5719521 YZ |
9870 | if (ret) |
9871 | goto out_fail; | |
4df27c4d | 9872 | } |
5a3f23d5 | 9873 | |
0c4d2d95 JB |
9874 | inode_inc_iversion(old_dir); |
9875 | inode_inc_iversion(new_dir); | |
9876 | inode_inc_iversion(old_inode); | |
04b285f3 DD |
9877 | old_dir->i_ctime = old_dir->i_mtime = |
9878 | new_dir->i_ctime = new_dir->i_mtime = | |
c2050a45 | 9879 | old_inode->i_ctime = current_time(old_dir); |
5f39d397 | 9880 | |
12fcfd22 | 9881 | if (old_dentry->d_parent != new_dentry->d_parent) |
f85b7379 DS |
9882 | btrfs_record_unlink_dir(trans, BTRFS_I(old_dir), |
9883 | BTRFS_I(old_inode), 1); | |
12fcfd22 | 9884 | |
33345d01 | 9885 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
9886 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
9887 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
9888 | old_dentry->d_name.name, | |
9889 | old_dentry->d_name.len); | |
9890 | } else { | |
4ec5934e NB |
9891 | ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir), |
9892 | BTRFS_I(d_inode(old_dentry)), | |
92986796 AV |
9893 | old_dentry->d_name.name, |
9894 | old_dentry->d_name.len); | |
9895 | if (!ret) | |
9896 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 9897 | } |
79787eaa | 9898 | if (ret) { |
66642832 | 9899 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9900 | goto out_fail; |
9901 | } | |
39279cc3 CM |
9902 | |
9903 | if (new_inode) { | |
0c4d2d95 | 9904 | inode_inc_iversion(new_inode); |
c2050a45 | 9905 | new_inode->i_ctime = current_time(new_inode); |
4a0cc7ca | 9906 | if (unlikely(btrfs_ino(BTRFS_I(new_inode)) == |
4df27c4d YZ |
9907 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
9908 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
9909 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
9910 | root_objectid, | |
9911 | new_dentry->d_name.name, | |
9912 | new_dentry->d_name.len); | |
9913 | BUG_ON(new_inode->i_nlink == 0); | |
9914 | } else { | |
4ec5934e NB |
9915 | ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir), |
9916 | BTRFS_I(d_inode(new_dentry)), | |
4df27c4d YZ |
9917 | new_dentry->d_name.name, |
9918 | new_dentry->d_name.len); | |
9919 | } | |
4ef31a45 | 9920 | if (!ret && new_inode->i_nlink == 0) |
2b0143b5 | 9921 | ret = btrfs_orphan_add(trans, d_inode(new_dentry)); |
79787eaa | 9922 | if (ret) { |
66642832 | 9923 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9924 | goto out_fail; |
9925 | } | |
39279cc3 | 9926 | } |
aec7477b | 9927 | |
4df27c4d YZ |
9928 | ret = btrfs_add_link(trans, new_dir, old_inode, |
9929 | new_dentry->d_name.name, | |
a5719521 | 9930 | new_dentry->d_name.len, 0, index); |
79787eaa | 9931 | if (ret) { |
66642832 | 9932 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9933 | goto out_fail; |
9934 | } | |
39279cc3 | 9935 | |
67de1176 MX |
9936 | if (old_inode->i_nlink == 1) |
9937 | BTRFS_I(old_inode)->dir_index = index; | |
9938 | ||
3dc9e8f7 | 9939 | if (log_pinned) { |
10d9f309 | 9940 | struct dentry *parent = new_dentry->d_parent; |
3dc9e8f7 | 9941 | |
f85b7379 DS |
9942 | btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir), |
9943 | parent); | |
4df27c4d | 9944 | btrfs_end_log_trans(root); |
3dc9e8f7 | 9945 | log_pinned = false; |
4df27c4d | 9946 | } |
cdd1fedf DF |
9947 | |
9948 | if (flags & RENAME_WHITEOUT) { | |
9949 | ret = btrfs_whiteout_for_rename(trans, root, old_dir, | |
9950 | old_dentry); | |
9951 | ||
9952 | if (ret) { | |
66642832 | 9953 | btrfs_abort_transaction(trans, ret); |
cdd1fedf DF |
9954 | goto out_fail; |
9955 | } | |
4df27c4d | 9956 | } |
39279cc3 | 9957 | out_fail: |
3dc9e8f7 FM |
9958 | /* |
9959 | * If we have pinned the log and an error happened, we unpin tasks | |
9960 | * trying to sync the log and force them to fallback to a transaction | |
9961 | * commit if the log currently contains any of the inodes involved in | |
9962 | * this rename operation (to ensure we do not persist a log with an | |
9963 | * inconsistent state for any of these inodes or leading to any | |
9964 | * inconsistencies when replayed). If the transaction was aborted, the | |
9965 | * abortion reason is propagated to userspace when attempting to commit | |
9966 | * the transaction. If the log does not contain any of these inodes, we | |
9967 | * allow the tasks to sync it. | |
9968 | */ | |
9969 | if (ret && log_pinned) { | |
0f8939b8 NB |
9970 | if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) || |
9971 | btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) || | |
9972 | btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) || | |
3dc9e8f7 | 9973 | (new_inode && |
0f8939b8 | 9974 | btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation))) |
0b246afa | 9975 | btrfs_set_log_full_commit(fs_info, trans); |
3dc9e8f7 FM |
9976 | |
9977 | btrfs_end_log_trans(root); | |
9978 | log_pinned = false; | |
9979 | } | |
3a45bb20 | 9980 | btrfs_end_transaction(trans); |
b44c59a8 | 9981 | out_notrans: |
33345d01 | 9982 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
0b246afa | 9983 | up_read(&fs_info->subvol_sem); |
9ed74f2d | 9984 | |
39279cc3 CM |
9985 | return ret; |
9986 | } | |
9987 | ||
80ace85c MS |
9988 | static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry, |
9989 | struct inode *new_dir, struct dentry *new_dentry, | |
9990 | unsigned int flags) | |
9991 | { | |
cdd1fedf | 9992 | if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) |
80ace85c MS |
9993 | return -EINVAL; |
9994 | ||
cdd1fedf DF |
9995 | if (flags & RENAME_EXCHANGE) |
9996 | return btrfs_rename_exchange(old_dir, old_dentry, new_dir, | |
9997 | new_dentry); | |
9998 | ||
9999 | return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags); | |
80ace85c MS |
10000 | } |
10001 | ||
8ccf6f19 MX |
10002 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
10003 | { | |
10004 | struct btrfs_delalloc_work *delalloc_work; | |
9f23e289 | 10005 | struct inode *inode; |
8ccf6f19 MX |
10006 | |
10007 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
10008 | work); | |
9f23e289 | 10009 | inode = delalloc_work->inode; |
30424601 DS |
10010 | filemap_flush(inode->i_mapping); |
10011 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
10012 | &BTRFS_I(inode)->runtime_flags)) | |
9f23e289 | 10013 | filemap_flush(inode->i_mapping); |
8ccf6f19 MX |
10014 | |
10015 | if (delalloc_work->delay_iput) | |
9f23e289 | 10016 | btrfs_add_delayed_iput(inode); |
8ccf6f19 | 10017 | else |
9f23e289 | 10018 | iput(inode); |
8ccf6f19 MX |
10019 | complete(&delalloc_work->completion); |
10020 | } | |
10021 | ||
10022 | struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode, | |
651d494a | 10023 | int delay_iput) |
8ccf6f19 MX |
10024 | { |
10025 | struct btrfs_delalloc_work *work; | |
10026 | ||
100d5702 | 10027 | work = kmalloc(sizeof(*work), GFP_NOFS); |
8ccf6f19 MX |
10028 | if (!work) |
10029 | return NULL; | |
10030 | ||
10031 | init_completion(&work->completion); | |
10032 | INIT_LIST_HEAD(&work->list); | |
10033 | work->inode = inode; | |
8ccf6f19 | 10034 | work->delay_iput = delay_iput; |
9e0af237 LB |
10035 | WARN_ON_ONCE(!inode); |
10036 | btrfs_init_work(&work->work, btrfs_flush_delalloc_helper, | |
10037 | btrfs_run_delalloc_work, NULL, NULL); | |
8ccf6f19 MX |
10038 | |
10039 | return work; | |
10040 | } | |
10041 | ||
10042 | void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work) | |
10043 | { | |
10044 | wait_for_completion(&work->completion); | |
100d5702 | 10045 | kfree(work); |
8ccf6f19 MX |
10046 | } |
10047 | ||
d352ac68 CM |
10048 | /* |
10049 | * some fairly slow code that needs optimization. This walks the list | |
10050 | * of all the inodes with pending delalloc and forces them to disk. | |
10051 | */ | |
6c255e67 MX |
10052 | static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput, |
10053 | int nr) | |
ea8c2819 | 10054 | { |
ea8c2819 | 10055 | struct btrfs_inode *binode; |
5b21f2ed | 10056 | struct inode *inode; |
8ccf6f19 MX |
10057 | struct btrfs_delalloc_work *work, *next; |
10058 | struct list_head works; | |
1eafa6c7 | 10059 | struct list_head splice; |
8ccf6f19 | 10060 | int ret = 0; |
ea8c2819 | 10061 | |
8ccf6f19 | 10062 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 10063 | INIT_LIST_HEAD(&splice); |
63607cc8 | 10064 | |
573bfb72 | 10065 | mutex_lock(&root->delalloc_mutex); |
eb73c1b7 MX |
10066 | spin_lock(&root->delalloc_lock); |
10067 | list_splice_init(&root->delalloc_inodes, &splice); | |
1eafa6c7 MX |
10068 | while (!list_empty(&splice)) { |
10069 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 10070 | delalloc_inodes); |
1eafa6c7 | 10071 | |
eb73c1b7 MX |
10072 | list_move_tail(&binode->delalloc_inodes, |
10073 | &root->delalloc_inodes); | |
5b21f2ed | 10074 | inode = igrab(&binode->vfs_inode); |
df0af1a5 | 10075 | if (!inode) { |
eb73c1b7 | 10076 | cond_resched_lock(&root->delalloc_lock); |
1eafa6c7 | 10077 | continue; |
df0af1a5 | 10078 | } |
eb73c1b7 | 10079 | spin_unlock(&root->delalloc_lock); |
1eafa6c7 | 10080 | |
651d494a | 10081 | work = btrfs_alloc_delalloc_work(inode, delay_iput); |
5d99a998 | 10082 | if (!work) { |
f4ab9ea7 JB |
10083 | if (delay_iput) |
10084 | btrfs_add_delayed_iput(inode); | |
10085 | else | |
10086 | iput(inode); | |
1eafa6c7 | 10087 | ret = -ENOMEM; |
a1ecaabb | 10088 | goto out; |
5b21f2ed | 10089 | } |
1eafa6c7 | 10090 | list_add_tail(&work->list, &works); |
a44903ab QW |
10091 | btrfs_queue_work(root->fs_info->flush_workers, |
10092 | &work->work); | |
6c255e67 MX |
10093 | ret++; |
10094 | if (nr != -1 && ret >= nr) | |
a1ecaabb | 10095 | goto out; |
5b21f2ed | 10096 | cond_resched(); |
eb73c1b7 | 10097 | spin_lock(&root->delalloc_lock); |
ea8c2819 | 10098 | } |
eb73c1b7 | 10099 | spin_unlock(&root->delalloc_lock); |
8c8bee1d | 10100 | |
a1ecaabb | 10101 | out: |
eb73c1b7 MX |
10102 | list_for_each_entry_safe(work, next, &works, list) { |
10103 | list_del_init(&work->list); | |
10104 | btrfs_wait_and_free_delalloc_work(work); | |
10105 | } | |
10106 | ||
10107 | if (!list_empty_careful(&splice)) { | |
10108 | spin_lock(&root->delalloc_lock); | |
10109 | list_splice_tail(&splice, &root->delalloc_inodes); | |
10110 | spin_unlock(&root->delalloc_lock); | |
10111 | } | |
573bfb72 | 10112 | mutex_unlock(&root->delalloc_mutex); |
eb73c1b7 MX |
10113 | return ret; |
10114 | } | |
1eafa6c7 | 10115 | |
eb73c1b7 MX |
10116 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
10117 | { | |
0b246afa | 10118 | struct btrfs_fs_info *fs_info = root->fs_info; |
eb73c1b7 | 10119 | int ret; |
1eafa6c7 | 10120 | |
0b246afa | 10121 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
10122 | return -EROFS; |
10123 | ||
6c255e67 MX |
10124 | ret = __start_delalloc_inodes(root, delay_iput, -1); |
10125 | if (ret > 0) | |
10126 | ret = 0; | |
eb73c1b7 MX |
10127 | /* |
10128 | * the filemap_flush will queue IO into the worker threads, but | |
8c8bee1d CM |
10129 | * we have to make sure the IO is actually started and that |
10130 | * ordered extents get created before we return | |
10131 | */ | |
0b246afa JM |
10132 | atomic_inc(&fs_info->async_submit_draining); |
10133 | while (atomic_read(&fs_info->nr_async_submits) || | |
10134 | atomic_read(&fs_info->async_delalloc_pages)) { | |
10135 | wait_event(fs_info->async_submit_wait, | |
10136 | (atomic_read(&fs_info->nr_async_submits) == 0 && | |
10137 | atomic_read(&fs_info->async_delalloc_pages) == 0)); | |
10138 | } | |
10139 | atomic_dec(&fs_info->async_submit_draining); | |
eb73c1b7 MX |
10140 | return ret; |
10141 | } | |
10142 | ||
6c255e67 MX |
10143 | int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput, |
10144 | int nr) | |
eb73c1b7 MX |
10145 | { |
10146 | struct btrfs_root *root; | |
10147 | struct list_head splice; | |
10148 | int ret; | |
10149 | ||
2c21b4d7 | 10150 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
10151 | return -EROFS; |
10152 | ||
10153 | INIT_LIST_HEAD(&splice); | |
10154 | ||
573bfb72 | 10155 | mutex_lock(&fs_info->delalloc_root_mutex); |
eb73c1b7 MX |
10156 | spin_lock(&fs_info->delalloc_root_lock); |
10157 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
6c255e67 | 10158 | while (!list_empty(&splice) && nr) { |
eb73c1b7 MX |
10159 | root = list_first_entry(&splice, struct btrfs_root, |
10160 | delalloc_root); | |
10161 | root = btrfs_grab_fs_root(root); | |
10162 | BUG_ON(!root); | |
10163 | list_move_tail(&root->delalloc_root, | |
10164 | &fs_info->delalloc_roots); | |
10165 | spin_unlock(&fs_info->delalloc_root_lock); | |
10166 | ||
6c255e67 | 10167 | ret = __start_delalloc_inodes(root, delay_iput, nr); |
eb73c1b7 | 10168 | btrfs_put_fs_root(root); |
6c255e67 | 10169 | if (ret < 0) |
eb73c1b7 MX |
10170 | goto out; |
10171 | ||
6c255e67 MX |
10172 | if (nr != -1) { |
10173 | nr -= ret; | |
10174 | WARN_ON(nr < 0); | |
10175 | } | |
eb73c1b7 | 10176 | spin_lock(&fs_info->delalloc_root_lock); |
8ccf6f19 | 10177 | } |
eb73c1b7 | 10178 | spin_unlock(&fs_info->delalloc_root_lock); |
1eafa6c7 | 10179 | |
6c255e67 | 10180 | ret = 0; |
eb73c1b7 MX |
10181 | atomic_inc(&fs_info->async_submit_draining); |
10182 | while (atomic_read(&fs_info->nr_async_submits) || | |
10183 | atomic_read(&fs_info->async_delalloc_pages)) { | |
10184 | wait_event(fs_info->async_submit_wait, | |
10185 | (atomic_read(&fs_info->nr_async_submits) == 0 && | |
10186 | atomic_read(&fs_info->async_delalloc_pages) == 0)); | |
10187 | } | |
10188 | atomic_dec(&fs_info->async_submit_draining); | |
eb73c1b7 | 10189 | out: |
1eafa6c7 | 10190 | if (!list_empty_careful(&splice)) { |
eb73c1b7 MX |
10191 | spin_lock(&fs_info->delalloc_root_lock); |
10192 | list_splice_tail(&splice, &fs_info->delalloc_roots); | |
10193 | spin_unlock(&fs_info->delalloc_root_lock); | |
1eafa6c7 | 10194 | } |
573bfb72 | 10195 | mutex_unlock(&fs_info->delalloc_root_mutex); |
8ccf6f19 | 10196 | return ret; |
ea8c2819 CM |
10197 | } |
10198 | ||
39279cc3 CM |
10199 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
10200 | const char *symname) | |
10201 | { | |
0b246afa | 10202 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
39279cc3 CM |
10203 | struct btrfs_trans_handle *trans; |
10204 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
10205 | struct btrfs_path *path; | |
10206 | struct btrfs_key key; | |
1832a6d5 | 10207 | struct inode *inode = NULL; |
39279cc3 CM |
10208 | int err; |
10209 | int drop_inode = 0; | |
10210 | u64 objectid; | |
67871254 | 10211 | u64 index = 0; |
39279cc3 CM |
10212 | int name_len; |
10213 | int datasize; | |
5f39d397 | 10214 | unsigned long ptr; |
39279cc3 | 10215 | struct btrfs_file_extent_item *ei; |
5f39d397 | 10216 | struct extent_buffer *leaf; |
39279cc3 | 10217 | |
f06becc4 | 10218 | name_len = strlen(symname); |
0b246afa | 10219 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info)) |
39279cc3 | 10220 | return -ENAMETOOLONG; |
1832a6d5 | 10221 | |
9ed74f2d JB |
10222 | /* |
10223 | * 2 items for inode item and ref | |
10224 | * 2 items for dir items | |
9269d12b FM |
10225 | * 1 item for updating parent inode item |
10226 | * 1 item for the inline extent item | |
9ed74f2d JB |
10227 | * 1 item for xattr if selinux is on |
10228 | */ | |
9269d12b | 10229 | trans = btrfs_start_transaction(root, 7); |
a22285a6 YZ |
10230 | if (IS_ERR(trans)) |
10231 | return PTR_ERR(trans); | |
1832a6d5 | 10232 | |
581bb050 LZ |
10233 | err = btrfs_find_free_ino(root, &objectid); |
10234 | if (err) | |
10235 | goto out_unlock; | |
10236 | ||
aec7477b | 10237 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
f85b7379 DS |
10238 | dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), |
10239 | objectid, S_IFLNK|S_IRWXUGO, &index); | |
7cf96da3 TI |
10240 | if (IS_ERR(inode)) { |
10241 | err = PTR_ERR(inode); | |
39279cc3 | 10242 | goto out_unlock; |
7cf96da3 | 10243 | } |
39279cc3 | 10244 | |
ad19db71 CS |
10245 | /* |
10246 | * If the active LSM wants to access the inode during | |
10247 | * d_instantiate it needs these. Smack checks to see | |
10248 | * if the filesystem supports xattrs by looking at the | |
10249 | * ops vector. | |
10250 | */ | |
10251 | inode->i_fop = &btrfs_file_operations; | |
10252 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 10253 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
10254 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
10255 | ||
10256 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
10257 | if (err) | |
10258 | goto out_unlock_inode; | |
ad19db71 | 10259 | |
39279cc3 | 10260 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10261 | if (!path) { |
10262 | err = -ENOMEM; | |
b0d5d10f | 10263 | goto out_unlock_inode; |
d8926bb3 | 10264 | } |
4a0cc7ca | 10265 | key.objectid = btrfs_ino(BTRFS_I(inode)); |
39279cc3 | 10266 | key.offset = 0; |
962a298f | 10267 | key.type = BTRFS_EXTENT_DATA_KEY; |
39279cc3 CM |
10268 | datasize = btrfs_file_extent_calc_inline_size(name_len); |
10269 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
10270 | datasize); | |
54aa1f4d | 10271 | if (err) { |
b0839166 | 10272 | btrfs_free_path(path); |
b0d5d10f | 10273 | goto out_unlock_inode; |
54aa1f4d | 10274 | } |
5f39d397 CM |
10275 | leaf = path->nodes[0]; |
10276 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
10277 | struct btrfs_file_extent_item); | |
10278 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
10279 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 10280 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
10281 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
10282 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
10283 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
10284 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
10285 | ||
39279cc3 | 10286 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
10287 | write_extent_buffer(leaf, symname, ptr, name_len); |
10288 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 10289 | btrfs_free_path(path); |
5f39d397 | 10290 | |
39279cc3 | 10291 | inode->i_op = &btrfs_symlink_inode_operations; |
21fc61c7 | 10292 | inode_nohighmem(inode); |
39279cc3 | 10293 | inode->i_mapping->a_ops = &btrfs_symlink_aops; |
d899e052 | 10294 | inode_set_bytes(inode, name_len); |
f06becc4 | 10295 | btrfs_i_size_write(inode, name_len); |
54aa1f4d | 10296 | err = btrfs_update_inode(trans, root, inode); |
d50866d0 FM |
10297 | /* |
10298 | * Last step, add directory indexes for our symlink inode. This is the | |
10299 | * last step to avoid extra cleanup of these indexes if an error happens | |
10300 | * elsewhere above. | |
10301 | */ | |
10302 | if (!err) | |
10303 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); | |
b0d5d10f | 10304 | if (err) { |
54aa1f4d | 10305 | drop_inode = 1; |
b0d5d10f CM |
10306 | goto out_unlock_inode; |
10307 | } | |
10308 | ||
10309 | unlock_new_inode(inode); | |
10310 | d_instantiate(dentry, inode); | |
39279cc3 CM |
10311 | |
10312 | out_unlock: | |
3a45bb20 | 10313 | btrfs_end_transaction(trans); |
39279cc3 CM |
10314 | if (drop_inode) { |
10315 | inode_dec_link_count(inode); | |
10316 | iput(inode); | |
10317 | } | |
2ff7e61e | 10318 | btrfs_btree_balance_dirty(fs_info); |
39279cc3 | 10319 | return err; |
b0d5d10f CM |
10320 | |
10321 | out_unlock_inode: | |
10322 | drop_inode = 1; | |
10323 | unlock_new_inode(inode); | |
10324 | goto out_unlock; | |
39279cc3 | 10325 | } |
16432985 | 10326 | |
0af3d00b JB |
10327 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
10328 | u64 start, u64 num_bytes, u64 min_size, | |
10329 | loff_t actual_len, u64 *alloc_hint, | |
10330 | struct btrfs_trans_handle *trans) | |
d899e052 | 10331 | { |
0b246afa | 10332 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
5dc562c5 JB |
10333 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
10334 | struct extent_map *em; | |
d899e052 YZ |
10335 | struct btrfs_root *root = BTRFS_I(inode)->root; |
10336 | struct btrfs_key ins; | |
d899e052 | 10337 | u64 cur_offset = start; |
55a61d1d | 10338 | u64 i_size; |
154ea289 | 10339 | u64 cur_bytes; |
0b670dc4 | 10340 | u64 last_alloc = (u64)-1; |
d899e052 | 10341 | int ret = 0; |
0af3d00b | 10342 | bool own_trans = true; |
18513091 | 10343 | u64 end = start + num_bytes - 1; |
d899e052 | 10344 | |
0af3d00b JB |
10345 | if (trans) |
10346 | own_trans = false; | |
d899e052 | 10347 | while (num_bytes > 0) { |
0af3d00b JB |
10348 | if (own_trans) { |
10349 | trans = btrfs_start_transaction(root, 3); | |
10350 | if (IS_ERR(trans)) { | |
10351 | ret = PTR_ERR(trans); | |
10352 | break; | |
10353 | } | |
5a303d5d YZ |
10354 | } |
10355 | ||
ee22184b | 10356 | cur_bytes = min_t(u64, num_bytes, SZ_256M); |
154ea289 | 10357 | cur_bytes = max(cur_bytes, min_size); |
0b670dc4 JB |
10358 | /* |
10359 | * If we are severely fragmented we could end up with really | |
10360 | * small allocations, so if the allocator is returning small | |
10361 | * chunks lets make its job easier by only searching for those | |
10362 | * sized chunks. | |
10363 | */ | |
10364 | cur_bytes = min(cur_bytes, last_alloc); | |
18513091 WX |
10365 | ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes, |
10366 | min_size, 0, *alloc_hint, &ins, 1, 0); | |
5a303d5d | 10367 | if (ret) { |
0af3d00b | 10368 | if (own_trans) |
3a45bb20 | 10369 | btrfs_end_transaction(trans); |
a22285a6 | 10370 | break; |
d899e052 | 10371 | } |
0b246afa | 10372 | btrfs_dec_block_group_reservations(fs_info, ins.objectid); |
5a303d5d | 10373 | |
0b670dc4 | 10374 | last_alloc = ins.offset; |
d899e052 YZ |
10375 | ret = insert_reserved_file_extent(trans, inode, |
10376 | cur_offset, ins.objectid, | |
10377 | ins.offset, ins.offset, | |
920bbbfb | 10378 | ins.offset, 0, 0, 0, |
d899e052 | 10379 | BTRFS_FILE_EXTENT_PREALLOC); |
79787eaa | 10380 | if (ret) { |
2ff7e61e | 10381 | btrfs_free_reserved_extent(fs_info, ins.objectid, |
e570fd27 | 10382 | ins.offset, 0); |
66642832 | 10383 | btrfs_abort_transaction(trans, ret); |
79787eaa | 10384 | if (own_trans) |
3a45bb20 | 10385 | btrfs_end_transaction(trans); |
79787eaa JM |
10386 | break; |
10387 | } | |
31193213 | 10388 | |
a1ed835e CM |
10389 | btrfs_drop_extent_cache(inode, cur_offset, |
10390 | cur_offset + ins.offset -1, 0); | |
5a303d5d | 10391 | |
5dc562c5 JB |
10392 | em = alloc_extent_map(); |
10393 | if (!em) { | |
10394 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
10395 | &BTRFS_I(inode)->runtime_flags); | |
10396 | goto next; | |
10397 | } | |
10398 | ||
10399 | em->start = cur_offset; | |
10400 | em->orig_start = cur_offset; | |
10401 | em->len = ins.offset; | |
10402 | em->block_start = ins.objectid; | |
10403 | em->block_len = ins.offset; | |
b4939680 | 10404 | em->orig_block_len = ins.offset; |
cc95bef6 | 10405 | em->ram_bytes = ins.offset; |
0b246afa | 10406 | em->bdev = fs_info->fs_devices->latest_bdev; |
5dc562c5 JB |
10407 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); |
10408 | em->generation = trans->transid; | |
10409 | ||
10410 | while (1) { | |
10411 | write_lock(&em_tree->lock); | |
09a2a8f9 | 10412 | ret = add_extent_mapping(em_tree, em, 1); |
5dc562c5 JB |
10413 | write_unlock(&em_tree->lock); |
10414 | if (ret != -EEXIST) | |
10415 | break; | |
10416 | btrfs_drop_extent_cache(inode, cur_offset, | |
10417 | cur_offset + ins.offset - 1, | |
10418 | 0); | |
10419 | } | |
10420 | free_extent_map(em); | |
10421 | next: | |
d899e052 YZ |
10422 | num_bytes -= ins.offset; |
10423 | cur_offset += ins.offset; | |
efa56464 | 10424 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 10425 | |
0c4d2d95 | 10426 | inode_inc_iversion(inode); |
c2050a45 | 10427 | inode->i_ctime = current_time(inode); |
6cbff00f | 10428 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 10429 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
10430 | (actual_len > inode->i_size) && |
10431 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 10432 | if (cur_offset > actual_len) |
55a61d1d | 10433 | i_size = actual_len; |
d1ea6a61 | 10434 | else |
55a61d1d JB |
10435 | i_size = cur_offset; |
10436 | i_size_write(inode, i_size); | |
10437 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
10438 | } |
10439 | ||
d899e052 | 10440 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
10441 | |
10442 | if (ret) { | |
66642832 | 10443 | btrfs_abort_transaction(trans, ret); |
79787eaa | 10444 | if (own_trans) |
3a45bb20 | 10445 | btrfs_end_transaction(trans); |
79787eaa JM |
10446 | break; |
10447 | } | |
d899e052 | 10448 | |
0af3d00b | 10449 | if (own_trans) |
3a45bb20 | 10450 | btrfs_end_transaction(trans); |
5a303d5d | 10451 | } |
18513091 WX |
10452 | if (cur_offset < end) |
10453 | btrfs_free_reserved_data_space(inode, cur_offset, | |
10454 | end - cur_offset + 1); | |
d899e052 YZ |
10455 | return ret; |
10456 | } | |
10457 | ||
0af3d00b JB |
10458 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
10459 | u64 start, u64 num_bytes, u64 min_size, | |
10460 | loff_t actual_len, u64 *alloc_hint) | |
10461 | { | |
10462 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10463 | min_size, actual_len, alloc_hint, | |
10464 | NULL); | |
10465 | } | |
10466 | ||
10467 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
10468 | struct btrfs_trans_handle *trans, int mode, | |
10469 | u64 start, u64 num_bytes, u64 min_size, | |
10470 | loff_t actual_len, u64 *alloc_hint) | |
10471 | { | |
10472 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
10473 | min_size, actual_len, alloc_hint, trans); | |
10474 | } | |
10475 | ||
e6dcd2dc CM |
10476 | static int btrfs_set_page_dirty(struct page *page) |
10477 | { | |
e6dcd2dc CM |
10478 | return __set_page_dirty_nobuffers(page); |
10479 | } | |
10480 | ||
10556cb2 | 10481 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 10482 | { |
b83cc969 | 10483 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 10484 | umode_t mode = inode->i_mode; |
b83cc969 | 10485 | |
cb6db4e5 JM |
10486 | if (mask & MAY_WRITE && |
10487 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
10488 | if (btrfs_root_readonly(root)) | |
10489 | return -EROFS; | |
10490 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
10491 | return -EACCES; | |
10492 | } | |
2830ba7f | 10493 | return generic_permission(inode, mask); |
fdebe2bd | 10494 | } |
39279cc3 | 10495 | |
ef3b9af5 FM |
10496 | static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) |
10497 | { | |
2ff7e61e | 10498 | struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); |
ef3b9af5 FM |
10499 | struct btrfs_trans_handle *trans; |
10500 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
10501 | struct inode *inode = NULL; | |
10502 | u64 objectid; | |
10503 | u64 index; | |
10504 | int ret = 0; | |
10505 | ||
10506 | /* | |
10507 | * 5 units required for adding orphan entry | |
10508 | */ | |
10509 | trans = btrfs_start_transaction(root, 5); | |
10510 | if (IS_ERR(trans)) | |
10511 | return PTR_ERR(trans); | |
10512 | ||
10513 | ret = btrfs_find_free_ino(root, &objectid); | |
10514 | if (ret) | |
10515 | goto out; | |
10516 | ||
10517 | inode = btrfs_new_inode(trans, root, dir, NULL, 0, | |
f85b7379 | 10518 | btrfs_ino(BTRFS_I(dir)), objectid, mode, &index); |
ef3b9af5 FM |
10519 | if (IS_ERR(inode)) { |
10520 | ret = PTR_ERR(inode); | |
10521 | inode = NULL; | |
10522 | goto out; | |
10523 | } | |
10524 | ||
ef3b9af5 FM |
10525 | inode->i_fop = &btrfs_file_operations; |
10526 | inode->i_op = &btrfs_file_inode_operations; | |
10527 | ||
10528 | inode->i_mapping->a_ops = &btrfs_aops; | |
ef3b9af5 FM |
10529 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
10530 | ||
b0d5d10f CM |
10531 | ret = btrfs_init_inode_security(trans, inode, dir, NULL); |
10532 | if (ret) | |
10533 | goto out_inode; | |
10534 | ||
10535 | ret = btrfs_update_inode(trans, root, inode); | |
10536 | if (ret) | |
10537 | goto out_inode; | |
ef3b9af5 FM |
10538 | ret = btrfs_orphan_add(trans, inode); |
10539 | if (ret) | |
b0d5d10f | 10540 | goto out_inode; |
ef3b9af5 | 10541 | |
5762b5c9 FM |
10542 | /* |
10543 | * We set number of links to 0 in btrfs_new_inode(), and here we set | |
10544 | * it to 1 because d_tmpfile() will issue a warning if the count is 0, | |
10545 | * through: | |
10546 | * | |
10547 | * d_tmpfile() -> inode_dec_link_count() -> drop_nlink() | |
10548 | */ | |
10549 | set_nlink(inode, 1); | |
b0d5d10f | 10550 | unlock_new_inode(inode); |
ef3b9af5 FM |
10551 | d_tmpfile(dentry, inode); |
10552 | mark_inode_dirty(inode); | |
10553 | ||
10554 | out: | |
3a45bb20 | 10555 | btrfs_end_transaction(trans); |
ef3b9af5 FM |
10556 | if (ret) |
10557 | iput(inode); | |
2ff7e61e JM |
10558 | btrfs_balance_delayed_items(fs_info); |
10559 | btrfs_btree_balance_dirty(fs_info); | |
ef3b9af5 | 10560 | return ret; |
b0d5d10f CM |
10561 | |
10562 | out_inode: | |
10563 | unlock_new_inode(inode); | |
10564 | goto out; | |
10565 | ||
ef3b9af5 FM |
10566 | } |
10567 | ||
6e1d5dcc | 10568 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 10569 | .getattr = btrfs_getattr, |
39279cc3 CM |
10570 | .lookup = btrfs_lookup, |
10571 | .create = btrfs_create, | |
10572 | .unlink = btrfs_unlink, | |
10573 | .link = btrfs_link, | |
10574 | .mkdir = btrfs_mkdir, | |
10575 | .rmdir = btrfs_rmdir, | |
2773bf00 | 10576 | .rename = btrfs_rename2, |
39279cc3 CM |
10577 | .symlink = btrfs_symlink, |
10578 | .setattr = btrfs_setattr, | |
618e21d5 | 10579 | .mknod = btrfs_mknod, |
5103e947 | 10580 | .listxattr = btrfs_listxattr, |
fdebe2bd | 10581 | .permission = btrfs_permission, |
4e34e719 | 10582 | .get_acl = btrfs_get_acl, |
996a710d | 10583 | .set_acl = btrfs_set_acl, |
93fd63c2 | 10584 | .update_time = btrfs_update_time, |
ef3b9af5 | 10585 | .tmpfile = btrfs_tmpfile, |
39279cc3 | 10586 | }; |
6e1d5dcc | 10587 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 10588 | .lookup = btrfs_lookup, |
fdebe2bd | 10589 | .permission = btrfs_permission, |
93fd63c2 | 10590 | .update_time = btrfs_update_time, |
39279cc3 | 10591 | }; |
76dda93c | 10592 | |
828c0950 | 10593 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
10594 | .llseek = generic_file_llseek, |
10595 | .read = generic_read_dir, | |
02dbfc99 | 10596 | .iterate_shared = btrfs_real_readdir, |
34287aa3 | 10597 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 10598 | #ifdef CONFIG_COMPAT |
4c63c245 | 10599 | .compat_ioctl = btrfs_compat_ioctl, |
39279cc3 | 10600 | #endif |
6bf13c0c | 10601 | .release = btrfs_release_file, |
e02119d5 | 10602 | .fsync = btrfs_sync_file, |
39279cc3 CM |
10603 | }; |
10604 | ||
20e5506b | 10605 | static const struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 10606 | .fill_delalloc = run_delalloc_range, |
065631f6 | 10607 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 10608 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 10609 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 10610 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 10611 | .writepage_start_hook = btrfs_writepage_start_hook, |
b0c68f8b CM |
10612 | .set_bit_hook = btrfs_set_bit_hook, |
10613 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
10614 | .merge_extent_hook = btrfs_merge_extent_hook, |
10615 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
10616 | }; |
10617 | ||
35054394 CM |
10618 | /* |
10619 | * btrfs doesn't support the bmap operation because swapfiles | |
10620 | * use bmap to make a mapping of extents in the file. They assume | |
10621 | * these extents won't change over the life of the file and they | |
10622 | * use the bmap result to do IO directly to the drive. | |
10623 | * | |
10624 | * the btrfs bmap call would return logical addresses that aren't | |
10625 | * suitable for IO and they also will change frequently as COW | |
10626 | * operations happen. So, swapfile + btrfs == corruption. | |
10627 | * | |
10628 | * For now we're avoiding this by dropping bmap. | |
10629 | */ | |
7f09410b | 10630 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
10631 | .readpage = btrfs_readpage, |
10632 | .writepage = btrfs_writepage, | |
b293f02e | 10633 | .writepages = btrfs_writepages, |
3ab2fb5a | 10634 | .readpages = btrfs_readpages, |
16432985 | 10635 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
10636 | .invalidatepage = btrfs_invalidatepage, |
10637 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 10638 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 10639 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
10640 | }; |
10641 | ||
7f09410b | 10642 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
10643 | .readpage = btrfs_readpage, |
10644 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
10645 | .invalidatepage = btrfs_invalidatepage, |
10646 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
10647 | }; |
10648 | ||
6e1d5dcc | 10649 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
10650 | .getattr = btrfs_getattr, |
10651 | .setattr = btrfs_setattr, | |
5103e947 | 10652 | .listxattr = btrfs_listxattr, |
fdebe2bd | 10653 | .permission = btrfs_permission, |
1506fcc8 | 10654 | .fiemap = btrfs_fiemap, |
4e34e719 | 10655 | .get_acl = btrfs_get_acl, |
996a710d | 10656 | .set_acl = btrfs_set_acl, |
e41f941a | 10657 | .update_time = btrfs_update_time, |
39279cc3 | 10658 | }; |
6e1d5dcc | 10659 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
10660 | .getattr = btrfs_getattr, |
10661 | .setattr = btrfs_setattr, | |
fdebe2bd | 10662 | .permission = btrfs_permission, |
33268eaf | 10663 | .listxattr = btrfs_listxattr, |
4e34e719 | 10664 | .get_acl = btrfs_get_acl, |
996a710d | 10665 | .set_acl = btrfs_set_acl, |
e41f941a | 10666 | .update_time = btrfs_update_time, |
618e21d5 | 10667 | }; |
6e1d5dcc | 10668 | static const struct inode_operations btrfs_symlink_inode_operations = { |
6b255391 | 10669 | .get_link = page_get_link, |
f209561a | 10670 | .getattr = btrfs_getattr, |
22c44fe6 | 10671 | .setattr = btrfs_setattr, |
fdebe2bd | 10672 | .permission = btrfs_permission, |
0279b4cd | 10673 | .listxattr = btrfs_listxattr, |
e41f941a | 10674 | .update_time = btrfs_update_time, |
39279cc3 | 10675 | }; |
76dda93c | 10676 | |
82d339d9 | 10677 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c | 10678 | .d_delete = btrfs_dentry_delete, |
b4aff1f8 | 10679 | .d_release = btrfs_dentry_release, |
76dda93c | 10680 | }; |