]>
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> | |
33 | #include <linux/statfs.h> | |
34 | #include <linux/compat.h> | |
9ebefb18 | 35 | #include <linux/bit_spinlock.h> |
5103e947 | 36 | #include <linux/xattr.h> |
33268eaf | 37 | #include <linux/posix_acl.h> |
d899e052 | 38 | #include <linux/falloc.h> |
5a0e3ad6 | 39 | #include <linux/slab.h> |
7a36ddec | 40 | #include <linux/ratelimit.h> |
22c44fe6 | 41 | #include <linux/mount.h> |
55e301fd | 42 | #include <linux/btrfs.h> |
53b381b3 | 43 | #include <linux/blkdev.h> |
f23b5a59 | 44 | #include <linux/posix_acl_xattr.h> |
e2e40f2c | 45 | #include <linux/uio.h> |
39279cc3 CM |
46 | #include "ctree.h" |
47 | #include "disk-io.h" | |
48 | #include "transaction.h" | |
49 | #include "btrfs_inode.h" | |
39279cc3 | 50 | #include "print-tree.h" |
e6dcd2dc | 51 | #include "ordered-data.h" |
95819c05 | 52 | #include "xattr.h" |
e02119d5 | 53 | #include "tree-log.h" |
4a54c8c1 | 54 | #include "volumes.h" |
c8b97818 | 55 | #include "compression.h" |
b4ce94de | 56 | #include "locking.h" |
dc89e982 | 57 | #include "free-space-cache.h" |
581bb050 | 58 | #include "inode-map.h" |
38c227d8 | 59 | #include "backref.h" |
f23b5a59 | 60 | #include "hash.h" |
63541927 | 61 | #include "props.h" |
31193213 | 62 | #include "qgroup.h" |
39279cc3 CM |
63 | |
64 | struct btrfs_iget_args { | |
90d3e592 | 65 | struct btrfs_key *location; |
39279cc3 CM |
66 | struct btrfs_root *root; |
67 | }; | |
68 | ||
6e1d5dcc AD |
69 | static const struct inode_operations btrfs_dir_inode_operations; |
70 | static const struct inode_operations btrfs_symlink_inode_operations; | |
71 | static const struct inode_operations btrfs_dir_ro_inode_operations; | |
72 | static const struct inode_operations btrfs_special_inode_operations; | |
73 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b AD |
74 | static const struct address_space_operations btrfs_aops; |
75 | static const struct address_space_operations btrfs_symlink_aops; | |
828c0950 | 76 | static const struct file_operations btrfs_dir_file_operations; |
d1310b2e | 77 | static struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
78 | |
79 | static struct kmem_cache *btrfs_inode_cachep; | |
8ccf6f19 | 80 | static struct kmem_cache *btrfs_delalloc_work_cachep; |
39279cc3 CM |
81 | struct kmem_cache *btrfs_trans_handle_cachep; |
82 | struct kmem_cache *btrfs_transaction_cachep; | |
39279cc3 | 83 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 84 | struct kmem_cache *btrfs_free_space_cachep; |
39279cc3 CM |
85 | |
86 | #define S_SHIFT 12 | |
87 | static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { | |
88 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, | |
89 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
90 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
91 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
92 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
93 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
94 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
95 | }; | |
96 | ||
3972f260 | 97 | static int btrfs_setsize(struct inode *inode, struct iattr *attr); |
a41ad394 | 98 | static int btrfs_truncate(struct inode *inode); |
5fd02043 | 99 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent); |
771ed689 CM |
100 | static noinline int cow_file_range(struct inode *inode, |
101 | struct page *locked_page, | |
102 | u64 start, u64 end, int *page_started, | |
103 | unsigned long *nr_written, int unlock); | |
70c8a91c JB |
104 | static struct extent_map *create_pinned_em(struct inode *inode, u64 start, |
105 | u64 len, u64 orig_start, | |
106 | u64 block_start, u64 block_len, | |
cc95bef6 JB |
107 | u64 orig_block_len, u64 ram_bytes, |
108 | int type); | |
7b128766 | 109 | |
48a3b636 | 110 | static int btrfs_dirty_inode(struct inode *inode); |
7b128766 | 111 | |
6a3891c5 JB |
112 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
113 | void btrfs_test_inode_set_ops(struct inode *inode) | |
114 | { | |
115 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; | |
116 | } | |
117 | #endif | |
118 | ||
f34f57a3 | 119 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
2a7dba39 EP |
120 | struct inode *inode, struct inode *dir, |
121 | const struct qstr *qstr) | |
0279b4cd JO |
122 | { |
123 | int err; | |
124 | ||
f34f57a3 | 125 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 126 | if (!err) |
2a7dba39 | 127 | err = btrfs_xattr_security_init(trans, inode, dir, qstr); |
0279b4cd JO |
128 | return err; |
129 | } | |
130 | ||
c8b97818 CM |
131 | /* |
132 | * this does all the hard work for inserting an inline extent into | |
133 | * the btree. The caller should have done a btrfs_drop_extents so that | |
134 | * no overlapping inline items exist in the btree | |
135 | */ | |
40f76580 | 136 | static int insert_inline_extent(struct btrfs_trans_handle *trans, |
1acae57b | 137 | struct btrfs_path *path, int extent_inserted, |
c8b97818 CM |
138 | struct btrfs_root *root, struct inode *inode, |
139 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 140 | int compress_type, |
c8b97818 CM |
141 | struct page **compressed_pages) |
142 | { | |
c8b97818 CM |
143 | struct extent_buffer *leaf; |
144 | struct page *page = NULL; | |
145 | char *kaddr; | |
146 | unsigned long ptr; | |
147 | struct btrfs_file_extent_item *ei; | |
148 | int err = 0; | |
149 | int ret; | |
150 | size_t cur_size = size; | |
c8b97818 | 151 | unsigned long offset; |
c8b97818 | 152 | |
fe3f566c | 153 | if (compressed_size && compressed_pages) |
c8b97818 | 154 | cur_size = compressed_size; |
c8b97818 | 155 | |
1acae57b | 156 | inode_add_bytes(inode, size); |
c8b97818 | 157 | |
1acae57b FDBM |
158 | if (!extent_inserted) { |
159 | struct btrfs_key key; | |
160 | size_t datasize; | |
c8b97818 | 161 | |
1acae57b FDBM |
162 | key.objectid = btrfs_ino(inode); |
163 | key.offset = start; | |
962a298f | 164 | key.type = BTRFS_EXTENT_DATA_KEY; |
c8b97818 | 165 | |
1acae57b FDBM |
166 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
167 | path->leave_spinning = 1; | |
168 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
169 | datasize); | |
170 | if (ret) { | |
171 | err = ret; | |
172 | goto fail; | |
173 | } | |
c8b97818 CM |
174 | } |
175 | leaf = path->nodes[0]; | |
176 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
177 | struct btrfs_file_extent_item); | |
178 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
179 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
180 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
181 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
182 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
183 | ptr = btrfs_file_extent_inline_start(ei); | |
184 | ||
261507a0 | 185 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
186 | struct page *cpage; |
187 | int i = 0; | |
d397712b | 188 | while (compressed_size > 0) { |
c8b97818 | 189 | cpage = compressed_pages[i]; |
5b050f04 | 190 | cur_size = min_t(unsigned long, compressed_size, |
c8b97818 CM |
191 | PAGE_CACHE_SIZE); |
192 | ||
7ac687d9 | 193 | kaddr = kmap_atomic(cpage); |
c8b97818 | 194 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
7ac687d9 | 195 | kunmap_atomic(kaddr); |
c8b97818 CM |
196 | |
197 | i++; | |
198 | ptr += cur_size; | |
199 | compressed_size -= cur_size; | |
200 | } | |
201 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 202 | compress_type); |
c8b97818 CM |
203 | } else { |
204 | page = find_get_page(inode->i_mapping, | |
205 | start >> PAGE_CACHE_SHIFT); | |
206 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
7ac687d9 | 207 | kaddr = kmap_atomic(page); |
c8b97818 CM |
208 | offset = start & (PAGE_CACHE_SIZE - 1); |
209 | write_extent_buffer(leaf, kaddr + offset, ptr, size); | |
7ac687d9 | 210 | kunmap_atomic(kaddr); |
c8b97818 CM |
211 | page_cache_release(page); |
212 | } | |
213 | btrfs_mark_buffer_dirty(leaf); | |
1acae57b | 214 | btrfs_release_path(path); |
c8b97818 | 215 | |
c2167754 YZ |
216 | /* |
217 | * we're an inline extent, so nobody can | |
218 | * extend the file past i_size without locking | |
219 | * a page we already have locked. | |
220 | * | |
221 | * We must do any isize and inode updates | |
222 | * before we unlock the pages. Otherwise we | |
223 | * could end up racing with unlink. | |
224 | */ | |
c8b97818 | 225 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
79787eaa | 226 | ret = btrfs_update_inode(trans, root, inode); |
c2167754 | 227 | |
79787eaa | 228 | return ret; |
c8b97818 | 229 | fail: |
c8b97818 CM |
230 | return err; |
231 | } | |
232 | ||
233 | ||
234 | /* | |
235 | * conditionally insert an inline extent into the file. This | |
236 | * does the checks required to make sure the data is small enough | |
237 | * to fit as an inline extent. | |
238 | */ | |
00361589 JB |
239 | static noinline int cow_file_range_inline(struct btrfs_root *root, |
240 | struct inode *inode, u64 start, | |
241 | u64 end, size_t compressed_size, | |
242 | int compress_type, | |
243 | struct page **compressed_pages) | |
c8b97818 | 244 | { |
00361589 | 245 | struct btrfs_trans_handle *trans; |
c8b97818 CM |
246 | u64 isize = i_size_read(inode); |
247 | u64 actual_end = min(end + 1, isize); | |
248 | u64 inline_len = actual_end - start; | |
fda2832f | 249 | u64 aligned_end = ALIGN(end, root->sectorsize); |
c8b97818 CM |
250 | u64 data_len = inline_len; |
251 | int ret; | |
1acae57b FDBM |
252 | struct btrfs_path *path; |
253 | int extent_inserted = 0; | |
254 | u32 extent_item_size; | |
c8b97818 CM |
255 | |
256 | if (compressed_size) | |
257 | data_len = compressed_size; | |
258 | ||
259 | if (start > 0 || | |
354877be WS |
260 | actual_end > PAGE_CACHE_SIZE || |
261 | data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) || | |
c8b97818 CM |
262 | (!compressed_size && |
263 | (actual_end & (root->sectorsize - 1)) == 0) || | |
264 | end + 1 < isize || | |
265 | data_len > root->fs_info->max_inline) { | |
266 | return 1; | |
267 | } | |
268 | ||
1acae57b FDBM |
269 | path = btrfs_alloc_path(); |
270 | if (!path) | |
271 | return -ENOMEM; | |
272 | ||
00361589 | 273 | trans = btrfs_join_transaction(root); |
1acae57b FDBM |
274 | if (IS_ERR(trans)) { |
275 | btrfs_free_path(path); | |
00361589 | 276 | return PTR_ERR(trans); |
1acae57b | 277 | } |
00361589 JB |
278 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
279 | ||
1acae57b FDBM |
280 | if (compressed_size && compressed_pages) |
281 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
282 | compressed_size); | |
283 | else | |
284 | extent_item_size = btrfs_file_extent_calc_inline_size( | |
285 | inline_len); | |
286 | ||
287 | ret = __btrfs_drop_extents(trans, root, inode, path, | |
288 | start, aligned_end, NULL, | |
289 | 1, 1, extent_item_size, &extent_inserted); | |
00361589 JB |
290 | if (ret) { |
291 | btrfs_abort_transaction(trans, root, ret); | |
292 | goto out; | |
293 | } | |
c8b97818 CM |
294 | |
295 | if (isize > actual_end) | |
296 | inline_len = min_t(u64, isize, actual_end); | |
1acae57b FDBM |
297 | ret = insert_inline_extent(trans, path, extent_inserted, |
298 | root, inode, start, | |
c8b97818 | 299 | inline_len, compressed_size, |
fe3f566c | 300 | compress_type, compressed_pages); |
2adcac1a | 301 | if (ret && ret != -ENOSPC) { |
79787eaa | 302 | btrfs_abort_transaction(trans, root, ret); |
00361589 | 303 | goto out; |
2adcac1a | 304 | } else if (ret == -ENOSPC) { |
00361589 JB |
305 | ret = 1; |
306 | goto out; | |
79787eaa | 307 | } |
2adcac1a | 308 | |
bdc20e67 | 309 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); |
0ca1f7ce | 310 | btrfs_delalloc_release_metadata(inode, end + 1 - start); |
a1ed835e | 311 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
00361589 | 312 | out: |
1acae57b | 313 | btrfs_free_path(path); |
00361589 JB |
314 | btrfs_end_transaction(trans, root); |
315 | return ret; | |
c8b97818 CM |
316 | } |
317 | ||
771ed689 CM |
318 | struct async_extent { |
319 | u64 start; | |
320 | u64 ram_size; | |
321 | u64 compressed_size; | |
322 | struct page **pages; | |
323 | unsigned long nr_pages; | |
261507a0 | 324 | int compress_type; |
771ed689 CM |
325 | struct list_head list; |
326 | }; | |
327 | ||
328 | struct async_cow { | |
329 | struct inode *inode; | |
330 | struct btrfs_root *root; | |
331 | struct page *locked_page; | |
332 | u64 start; | |
333 | u64 end; | |
334 | struct list_head extents; | |
335 | struct btrfs_work work; | |
336 | }; | |
337 | ||
338 | static noinline int add_async_extent(struct async_cow *cow, | |
339 | u64 start, u64 ram_size, | |
340 | u64 compressed_size, | |
341 | struct page **pages, | |
261507a0 LZ |
342 | unsigned long nr_pages, |
343 | int compress_type) | |
771ed689 CM |
344 | { |
345 | struct async_extent *async_extent; | |
346 | ||
347 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
79787eaa | 348 | BUG_ON(!async_extent); /* -ENOMEM */ |
771ed689 CM |
349 | async_extent->start = start; |
350 | async_extent->ram_size = ram_size; | |
351 | async_extent->compressed_size = compressed_size; | |
352 | async_extent->pages = pages; | |
353 | async_extent->nr_pages = nr_pages; | |
261507a0 | 354 | async_extent->compress_type = compress_type; |
771ed689 CM |
355 | list_add_tail(&async_extent->list, &cow->extents); |
356 | return 0; | |
357 | } | |
358 | ||
f79707b0 WS |
359 | static inline int inode_need_compress(struct inode *inode) |
360 | { | |
361 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
362 | ||
363 | /* force compress */ | |
364 | if (btrfs_test_opt(root, FORCE_COMPRESS)) | |
365 | return 1; | |
366 | /* bad compression ratios */ | |
367 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) | |
368 | return 0; | |
369 | if (btrfs_test_opt(root, COMPRESS) || | |
370 | BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS || | |
371 | BTRFS_I(inode)->force_compress) | |
372 | return 1; | |
373 | return 0; | |
374 | } | |
375 | ||
d352ac68 | 376 | /* |
771ed689 CM |
377 | * we create compressed extents in two phases. The first |
378 | * phase compresses a range of pages that have already been | |
379 | * locked (both pages and state bits are locked). | |
c8b97818 | 380 | * |
771ed689 CM |
381 | * This is done inside an ordered work queue, and the compression |
382 | * is spread across many cpus. The actual IO submission is step | |
383 | * two, and the ordered work queue takes care of making sure that | |
384 | * happens in the same order things were put onto the queue by | |
385 | * writepages and friends. | |
c8b97818 | 386 | * |
771ed689 CM |
387 | * If this code finds it can't get good compression, it puts an |
388 | * entry onto the work queue to write the uncompressed bytes. This | |
389 | * makes sure that both compressed inodes and uncompressed inodes | |
b2570314 AB |
390 | * are written in the same order that the flusher thread sent them |
391 | * down. | |
d352ac68 | 392 | */ |
c44f649e | 393 | static noinline void compress_file_range(struct inode *inode, |
771ed689 CM |
394 | struct page *locked_page, |
395 | u64 start, u64 end, | |
396 | struct async_cow *async_cow, | |
397 | int *num_added) | |
b888db2b CM |
398 | { |
399 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
db94535d | 400 | u64 num_bytes; |
db94535d | 401 | u64 blocksize = root->sectorsize; |
c8b97818 | 402 | u64 actual_end; |
42dc7bab | 403 | u64 isize = i_size_read(inode); |
e6dcd2dc | 404 | int ret = 0; |
c8b97818 CM |
405 | struct page **pages = NULL; |
406 | unsigned long nr_pages; | |
407 | unsigned long nr_pages_ret = 0; | |
408 | unsigned long total_compressed = 0; | |
409 | unsigned long total_in = 0; | |
410 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 411 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
412 | int i; |
413 | int will_compress; | |
261507a0 | 414 | int compress_type = root->fs_info->compress_type; |
4adaa611 | 415 | int redirty = 0; |
b888db2b | 416 | |
4cb13e5d LB |
417 | /* if this is a small write inside eof, kick off a defrag */ |
418 | if ((end - start + 1) < 16 * 1024 && | |
419 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
4cb5300b CM |
420 | btrfs_add_inode_defrag(NULL, inode); |
421 | ||
42dc7bab | 422 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
423 | again: |
424 | will_compress = 0; | |
425 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
426 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 427 | |
f03d9301 CM |
428 | /* |
429 | * we don't want to send crud past the end of i_size through | |
430 | * compression, that's just a waste of CPU time. So, if the | |
431 | * end of the file is before the start of our current | |
432 | * requested range of bytes, we bail out to the uncompressed | |
433 | * cleanup code that can deal with all of this. | |
434 | * | |
435 | * It isn't really the fastest way to fix things, but this is a | |
436 | * very uncommon corner. | |
437 | */ | |
438 | if (actual_end <= start) | |
439 | goto cleanup_and_bail_uncompressed; | |
440 | ||
c8b97818 CM |
441 | total_compressed = actual_end - start; |
442 | ||
4bcbb332 SW |
443 | /* |
444 | * skip compression for a small file range(<=blocksize) that | |
445 | * isn't an inline extent, since it dosen't save disk space at all. | |
446 | */ | |
447 | if (total_compressed <= blocksize && | |
448 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
449 | goto cleanup_and_bail_uncompressed; | |
450 | ||
c8b97818 CM |
451 | /* we want to make sure that amount of ram required to uncompress |
452 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
453 | * of a compressed extent to 128k. This is a crucial number |
454 | * because it also controls how easily we can spread reads across | |
455 | * cpus for decompression. | |
456 | * | |
457 | * We also want to make sure the amount of IO required to do | |
458 | * a random read is reasonably small, so we limit the size of | |
459 | * a compressed extent to 128k. | |
c8b97818 CM |
460 | */ |
461 | total_compressed = min(total_compressed, max_uncompressed); | |
fda2832f | 462 | num_bytes = ALIGN(end - start + 1, blocksize); |
be20aa9d | 463 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
464 | total_in = 0; |
465 | ret = 0; | |
db94535d | 466 | |
771ed689 CM |
467 | /* |
468 | * we do compression for mount -o compress and when the | |
469 | * inode has not been flagged as nocompress. This flag can | |
470 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 471 | */ |
f79707b0 | 472 | if (inode_need_compress(inode)) { |
c8b97818 | 473 | WARN_ON(pages); |
31e818fe | 474 | pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS); |
560f7d75 LZ |
475 | if (!pages) { |
476 | /* just bail out to the uncompressed code */ | |
477 | goto cont; | |
478 | } | |
c8b97818 | 479 | |
261507a0 LZ |
480 | if (BTRFS_I(inode)->force_compress) |
481 | compress_type = BTRFS_I(inode)->force_compress; | |
482 | ||
4adaa611 CM |
483 | /* |
484 | * we need to call clear_page_dirty_for_io on each | |
485 | * page in the range. Otherwise applications with the file | |
486 | * mmap'd can wander in and change the page contents while | |
487 | * we are compressing them. | |
488 | * | |
489 | * If the compression fails for any reason, we set the pages | |
490 | * dirty again later on. | |
491 | */ | |
492 | extent_range_clear_dirty_for_io(inode, start, end); | |
493 | redirty = 1; | |
261507a0 LZ |
494 | ret = btrfs_compress_pages(compress_type, |
495 | inode->i_mapping, start, | |
496 | total_compressed, pages, | |
497 | nr_pages, &nr_pages_ret, | |
498 | &total_in, | |
499 | &total_compressed, | |
500 | max_compressed); | |
c8b97818 CM |
501 | |
502 | if (!ret) { | |
503 | unsigned long offset = total_compressed & | |
504 | (PAGE_CACHE_SIZE - 1); | |
505 | struct page *page = pages[nr_pages_ret - 1]; | |
506 | char *kaddr; | |
507 | ||
508 | /* zero the tail end of the last page, we might be | |
509 | * sending it down to disk | |
510 | */ | |
511 | if (offset) { | |
7ac687d9 | 512 | kaddr = kmap_atomic(page); |
c8b97818 CM |
513 | memset(kaddr + offset, 0, |
514 | PAGE_CACHE_SIZE - offset); | |
7ac687d9 | 515 | kunmap_atomic(kaddr); |
c8b97818 CM |
516 | } |
517 | will_compress = 1; | |
518 | } | |
519 | } | |
560f7d75 | 520 | cont: |
c8b97818 CM |
521 | if (start == 0) { |
522 | /* lets try to make an inline extent */ | |
771ed689 | 523 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 524 | /* we didn't compress the entire range, try |
771ed689 | 525 | * to make an uncompressed inline extent. |
c8b97818 | 526 | */ |
00361589 JB |
527 | ret = cow_file_range_inline(root, inode, start, end, |
528 | 0, 0, NULL); | |
c8b97818 | 529 | } else { |
771ed689 | 530 | /* try making a compressed inline extent */ |
00361589 | 531 | ret = cow_file_range_inline(root, inode, start, end, |
fe3f566c LZ |
532 | total_compressed, |
533 | compress_type, pages); | |
c8b97818 | 534 | } |
79787eaa | 535 | if (ret <= 0) { |
151a41bc JB |
536 | unsigned long clear_flags = EXTENT_DELALLOC | |
537 | EXTENT_DEFRAG; | |
e6eb4314 FM |
538 | unsigned long page_error_op; |
539 | ||
151a41bc | 540 | clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0; |
e6eb4314 | 541 | page_error_op = ret < 0 ? PAGE_SET_ERROR : 0; |
151a41bc | 542 | |
771ed689 | 543 | /* |
79787eaa JM |
544 | * inline extent creation worked or returned error, |
545 | * we don't need to create any more async work items. | |
546 | * Unlock and free up our temp pages. | |
771ed689 | 547 | */ |
c2790a2e | 548 | extent_clear_unlock_delalloc(inode, start, end, NULL, |
151a41bc | 549 | clear_flags, PAGE_UNLOCK | |
c2790a2e JB |
550 | PAGE_CLEAR_DIRTY | |
551 | PAGE_SET_WRITEBACK | | |
e6eb4314 | 552 | page_error_op | |
c2790a2e | 553 | PAGE_END_WRITEBACK); |
c8b97818 CM |
554 | goto free_pages_out; |
555 | } | |
556 | } | |
557 | ||
558 | if (will_compress) { | |
559 | /* | |
560 | * we aren't doing an inline extent round the compressed size | |
561 | * up to a block size boundary so the allocator does sane | |
562 | * things | |
563 | */ | |
fda2832f | 564 | total_compressed = ALIGN(total_compressed, blocksize); |
c8b97818 CM |
565 | |
566 | /* | |
567 | * one last check to make sure the compression is really a | |
568 | * win, compare the page count read with the blocks on disk | |
569 | */ | |
fda2832f | 570 | total_in = ALIGN(total_in, PAGE_CACHE_SIZE); |
c8b97818 CM |
571 | if (total_compressed >= total_in) { |
572 | will_compress = 0; | |
573 | } else { | |
c8b97818 CM |
574 | num_bytes = total_in; |
575 | } | |
576 | } | |
577 | if (!will_compress && pages) { | |
578 | /* | |
579 | * the compression code ran but failed to make things smaller, | |
580 | * free any pages it allocated and our page pointer array | |
581 | */ | |
582 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 583 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
584 | page_cache_release(pages[i]); |
585 | } | |
586 | kfree(pages); | |
587 | pages = NULL; | |
588 | total_compressed = 0; | |
589 | nr_pages_ret = 0; | |
590 | ||
591 | /* flag the file so we don't compress in the future */ | |
1e701a32 CM |
592 | if (!btrfs_test_opt(root, FORCE_COMPRESS) && |
593 | !(BTRFS_I(inode)->force_compress)) { | |
a555f810 | 594 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 595 | } |
c8b97818 | 596 | } |
771ed689 CM |
597 | if (will_compress) { |
598 | *num_added += 1; | |
c8b97818 | 599 | |
771ed689 CM |
600 | /* the async work queues will take care of doing actual |
601 | * allocation on disk for these compressed pages, | |
602 | * and will submit them to the elevator. | |
603 | */ | |
604 | add_async_extent(async_cow, start, num_bytes, | |
261507a0 LZ |
605 | total_compressed, pages, nr_pages_ret, |
606 | compress_type); | |
179e29e4 | 607 | |
24ae6365 | 608 | if (start + num_bytes < end) { |
771ed689 CM |
609 | start += num_bytes; |
610 | pages = NULL; | |
611 | cond_resched(); | |
612 | goto again; | |
613 | } | |
614 | } else { | |
f03d9301 | 615 | cleanup_and_bail_uncompressed: |
771ed689 CM |
616 | /* |
617 | * No compression, but we still need to write the pages in | |
618 | * the file we've been given so far. redirty the locked | |
619 | * page if it corresponds to our extent and set things up | |
620 | * for the async work queue to run cow_file_range to do | |
621 | * the normal delalloc dance | |
622 | */ | |
623 | if (page_offset(locked_page) >= start && | |
624 | page_offset(locked_page) <= end) { | |
625 | __set_page_dirty_nobuffers(locked_page); | |
626 | /* unlocked later on in the async handlers */ | |
627 | } | |
4adaa611 CM |
628 | if (redirty) |
629 | extent_range_redirty_for_io(inode, start, end); | |
261507a0 LZ |
630 | add_async_extent(async_cow, start, end - start + 1, |
631 | 0, NULL, 0, BTRFS_COMPRESS_NONE); | |
771ed689 CM |
632 | *num_added += 1; |
633 | } | |
3b951516 | 634 | |
c44f649e | 635 | return; |
771ed689 CM |
636 | |
637 | free_pages_out: | |
638 | for (i = 0; i < nr_pages_ret; i++) { | |
639 | WARN_ON(pages[i]->mapping); | |
640 | page_cache_release(pages[i]); | |
641 | } | |
d397712b | 642 | kfree(pages); |
771ed689 | 643 | } |
771ed689 | 644 | |
40ae837b FM |
645 | static void free_async_extent_pages(struct async_extent *async_extent) |
646 | { | |
647 | int i; | |
648 | ||
649 | if (!async_extent->pages) | |
650 | return; | |
651 | ||
652 | for (i = 0; i < async_extent->nr_pages; i++) { | |
653 | WARN_ON(async_extent->pages[i]->mapping); | |
654 | page_cache_release(async_extent->pages[i]); | |
655 | } | |
656 | kfree(async_extent->pages); | |
657 | async_extent->nr_pages = 0; | |
658 | async_extent->pages = NULL; | |
771ed689 CM |
659 | } |
660 | ||
661 | /* | |
662 | * phase two of compressed writeback. This is the ordered portion | |
663 | * of the code, which only gets called in the order the work was | |
664 | * queued. We walk all the async extents created by compress_file_range | |
665 | * and send them down to the disk. | |
666 | */ | |
dec8f175 | 667 | static noinline void submit_compressed_extents(struct inode *inode, |
771ed689 CM |
668 | struct async_cow *async_cow) |
669 | { | |
670 | struct async_extent *async_extent; | |
671 | u64 alloc_hint = 0; | |
771ed689 CM |
672 | struct btrfs_key ins; |
673 | struct extent_map *em; | |
674 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
675 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
676 | struct extent_io_tree *io_tree; | |
f5a84ee3 | 677 | int ret = 0; |
771ed689 | 678 | |
3e04e7f1 | 679 | again: |
d397712b | 680 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
681 | async_extent = list_entry(async_cow->extents.next, |
682 | struct async_extent, list); | |
683 | list_del(&async_extent->list); | |
c8b97818 | 684 | |
771ed689 CM |
685 | io_tree = &BTRFS_I(inode)->io_tree; |
686 | ||
f5a84ee3 | 687 | retry: |
771ed689 CM |
688 | /* did the compression code fall back to uncompressed IO? */ |
689 | if (!async_extent->pages) { | |
690 | int page_started = 0; | |
691 | unsigned long nr_written = 0; | |
692 | ||
693 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 | 694 | async_extent->start + |
d0082371 | 695 | async_extent->ram_size - 1); |
771ed689 CM |
696 | |
697 | /* allocate blocks */ | |
f5a84ee3 JB |
698 | ret = cow_file_range(inode, async_cow->locked_page, |
699 | async_extent->start, | |
700 | async_extent->start + | |
701 | async_extent->ram_size - 1, | |
702 | &page_started, &nr_written, 0); | |
771ed689 | 703 | |
79787eaa JM |
704 | /* JDM XXX */ |
705 | ||
771ed689 CM |
706 | /* |
707 | * if page_started, cow_file_range inserted an | |
708 | * inline extent and took care of all the unlocking | |
709 | * and IO for us. Otherwise, we need to submit | |
710 | * all those pages down to the drive. | |
711 | */ | |
f5a84ee3 | 712 | if (!page_started && !ret) |
771ed689 CM |
713 | extent_write_locked_range(io_tree, |
714 | inode, async_extent->start, | |
d397712b | 715 | async_extent->start + |
771ed689 CM |
716 | async_extent->ram_size - 1, |
717 | btrfs_get_extent, | |
718 | WB_SYNC_ALL); | |
3e04e7f1 JB |
719 | else if (ret) |
720 | unlock_page(async_cow->locked_page); | |
771ed689 CM |
721 | kfree(async_extent); |
722 | cond_resched(); | |
723 | continue; | |
724 | } | |
725 | ||
726 | lock_extent(io_tree, async_extent->start, | |
d0082371 | 727 | async_extent->start + async_extent->ram_size - 1); |
771ed689 | 728 | |
00361589 | 729 | ret = btrfs_reserve_extent(root, |
771ed689 CM |
730 | async_extent->compressed_size, |
731 | async_extent->compressed_size, | |
e570fd27 | 732 | 0, alloc_hint, &ins, 1, 1); |
f5a84ee3 | 733 | if (ret) { |
40ae837b | 734 | free_async_extent_pages(async_extent); |
3e04e7f1 | 735 | |
fdf8e2ea JB |
736 | if (ret == -ENOSPC) { |
737 | unlock_extent(io_tree, async_extent->start, | |
738 | async_extent->start + | |
739 | async_extent->ram_size - 1); | |
ce62003f LB |
740 | |
741 | /* | |
742 | * we need to redirty the pages if we decide to | |
743 | * fallback to uncompressed IO, otherwise we | |
744 | * will not submit these pages down to lower | |
745 | * layers. | |
746 | */ | |
747 | extent_range_redirty_for_io(inode, | |
748 | async_extent->start, | |
749 | async_extent->start + | |
750 | async_extent->ram_size - 1); | |
751 | ||
79787eaa | 752 | goto retry; |
fdf8e2ea | 753 | } |
3e04e7f1 | 754 | goto out_free; |
f5a84ee3 | 755 | } |
c2167754 YZ |
756 | /* |
757 | * here we're doing allocation and writeback of the | |
758 | * compressed pages | |
759 | */ | |
760 | btrfs_drop_extent_cache(inode, async_extent->start, | |
761 | async_extent->start + | |
762 | async_extent->ram_size - 1, 0); | |
763 | ||
172ddd60 | 764 | em = alloc_extent_map(); |
b9aa55be LB |
765 | if (!em) { |
766 | ret = -ENOMEM; | |
3e04e7f1 | 767 | goto out_free_reserve; |
b9aa55be | 768 | } |
771ed689 CM |
769 | em->start = async_extent->start; |
770 | em->len = async_extent->ram_size; | |
445a6944 | 771 | em->orig_start = em->start; |
2ab28f32 JB |
772 | em->mod_start = em->start; |
773 | em->mod_len = em->len; | |
c8b97818 | 774 | |
771ed689 CM |
775 | em->block_start = ins.objectid; |
776 | em->block_len = ins.offset; | |
b4939680 | 777 | em->orig_block_len = ins.offset; |
cc95bef6 | 778 | em->ram_bytes = async_extent->ram_size; |
771ed689 | 779 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
261507a0 | 780 | em->compress_type = async_extent->compress_type; |
771ed689 CM |
781 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
782 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
70c8a91c | 783 | em->generation = -1; |
771ed689 | 784 | |
d397712b | 785 | while (1) { |
890871be | 786 | write_lock(&em_tree->lock); |
09a2a8f9 | 787 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 788 | write_unlock(&em_tree->lock); |
771ed689 CM |
789 | if (ret != -EEXIST) { |
790 | free_extent_map(em); | |
791 | break; | |
792 | } | |
793 | btrfs_drop_extent_cache(inode, async_extent->start, | |
794 | async_extent->start + | |
795 | async_extent->ram_size - 1, 0); | |
796 | } | |
797 | ||
3e04e7f1 JB |
798 | if (ret) |
799 | goto out_free_reserve; | |
800 | ||
261507a0 LZ |
801 | ret = btrfs_add_ordered_extent_compress(inode, |
802 | async_extent->start, | |
803 | ins.objectid, | |
804 | async_extent->ram_size, | |
805 | ins.offset, | |
806 | BTRFS_ORDERED_COMPRESSED, | |
807 | async_extent->compress_type); | |
d9f85963 FM |
808 | if (ret) { |
809 | btrfs_drop_extent_cache(inode, async_extent->start, | |
810 | async_extent->start + | |
811 | async_extent->ram_size - 1, 0); | |
3e04e7f1 | 812 | goto out_free_reserve; |
d9f85963 | 813 | } |
771ed689 | 814 | |
771ed689 CM |
815 | /* |
816 | * clear dirty, set writeback and unlock the pages. | |
817 | */ | |
c2790a2e | 818 | extent_clear_unlock_delalloc(inode, async_extent->start, |
a791e35e CM |
819 | async_extent->start + |
820 | async_extent->ram_size - 1, | |
151a41bc JB |
821 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC, |
822 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
c2790a2e | 823 | PAGE_SET_WRITEBACK); |
771ed689 | 824 | ret = btrfs_submit_compressed_write(inode, |
d397712b CM |
825 | async_extent->start, |
826 | async_extent->ram_size, | |
827 | ins.objectid, | |
828 | ins.offset, async_extent->pages, | |
829 | async_extent->nr_pages); | |
fce2a4e6 FM |
830 | if (ret) { |
831 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
832 | struct page *p = async_extent->pages[0]; | |
833 | const u64 start = async_extent->start; | |
834 | const u64 end = start + async_extent->ram_size - 1; | |
835 | ||
836 | p->mapping = inode->i_mapping; | |
837 | tree->ops->writepage_end_io_hook(p, start, end, | |
838 | NULL, 0); | |
839 | p->mapping = NULL; | |
840 | extent_clear_unlock_delalloc(inode, start, end, NULL, 0, | |
841 | PAGE_END_WRITEBACK | | |
842 | PAGE_SET_ERROR); | |
40ae837b | 843 | free_async_extent_pages(async_extent); |
fce2a4e6 | 844 | } |
771ed689 CM |
845 | alloc_hint = ins.objectid + ins.offset; |
846 | kfree(async_extent); | |
847 | cond_resched(); | |
848 | } | |
dec8f175 | 849 | return; |
3e04e7f1 | 850 | out_free_reserve: |
e570fd27 | 851 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1); |
79787eaa | 852 | out_free: |
c2790a2e | 853 | extent_clear_unlock_delalloc(inode, async_extent->start, |
3e04e7f1 JB |
854 | async_extent->start + |
855 | async_extent->ram_size - 1, | |
c2790a2e | 856 | NULL, EXTENT_LOCKED | EXTENT_DELALLOC | |
151a41bc JB |
857 | EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING, |
858 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
704de49d FM |
859 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK | |
860 | PAGE_SET_ERROR); | |
40ae837b | 861 | free_async_extent_pages(async_extent); |
79787eaa | 862 | kfree(async_extent); |
3e04e7f1 | 863 | goto again; |
771ed689 CM |
864 | } |
865 | ||
4b46fce2 JB |
866 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
867 | u64 num_bytes) | |
868 | { | |
869 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
870 | struct extent_map *em; | |
871 | u64 alloc_hint = 0; | |
872 | ||
873 | read_lock(&em_tree->lock); | |
874 | em = search_extent_mapping(em_tree, start, num_bytes); | |
875 | if (em) { | |
876 | /* | |
877 | * if block start isn't an actual block number then find the | |
878 | * first block in this inode and use that as a hint. If that | |
879 | * block is also bogus then just don't worry about it. | |
880 | */ | |
881 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
882 | free_extent_map(em); | |
883 | em = search_extent_mapping(em_tree, 0, 0); | |
884 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
885 | alloc_hint = em->block_start; | |
886 | if (em) | |
887 | free_extent_map(em); | |
888 | } else { | |
889 | alloc_hint = em->block_start; | |
890 | free_extent_map(em); | |
891 | } | |
892 | } | |
893 | read_unlock(&em_tree->lock); | |
894 | ||
895 | return alloc_hint; | |
896 | } | |
897 | ||
771ed689 CM |
898 | /* |
899 | * when extent_io.c finds a delayed allocation range in the file, | |
900 | * the call backs end up in this code. The basic idea is to | |
901 | * allocate extents on disk for the range, and create ordered data structs | |
902 | * in ram to track those extents. | |
903 | * | |
904 | * locked_page is the page that writepage had locked already. We use | |
905 | * it to make sure we don't do extra locks or unlocks. | |
906 | * | |
907 | * *page_started is set to one if we unlock locked_page and do everything | |
908 | * required to start IO on it. It may be clean and already done with | |
909 | * IO when we return. | |
910 | */ | |
00361589 JB |
911 | static noinline int cow_file_range(struct inode *inode, |
912 | struct page *locked_page, | |
913 | u64 start, u64 end, int *page_started, | |
914 | unsigned long *nr_written, | |
915 | int unlock) | |
771ed689 | 916 | { |
00361589 | 917 | struct btrfs_root *root = BTRFS_I(inode)->root; |
771ed689 CM |
918 | u64 alloc_hint = 0; |
919 | u64 num_bytes; | |
920 | unsigned long ram_size; | |
921 | u64 disk_num_bytes; | |
922 | u64 cur_alloc_size; | |
923 | u64 blocksize = root->sectorsize; | |
771ed689 CM |
924 | struct btrfs_key ins; |
925 | struct extent_map *em; | |
926 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
927 | int ret = 0; | |
928 | ||
02ecd2c2 JB |
929 | if (btrfs_is_free_space_inode(inode)) { |
930 | WARN_ON_ONCE(1); | |
29bce2f3 JB |
931 | ret = -EINVAL; |
932 | goto out_unlock; | |
02ecd2c2 | 933 | } |
771ed689 | 934 | |
fda2832f | 935 | num_bytes = ALIGN(end - start + 1, blocksize); |
771ed689 CM |
936 | num_bytes = max(blocksize, num_bytes); |
937 | disk_num_bytes = num_bytes; | |
771ed689 | 938 | |
4cb5300b | 939 | /* if this is a small write inside eof, kick off defrag */ |
4cb13e5d LB |
940 | if (num_bytes < 64 * 1024 && |
941 | (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size)) | |
00361589 | 942 | btrfs_add_inode_defrag(NULL, inode); |
4cb5300b | 943 | |
771ed689 CM |
944 | if (start == 0) { |
945 | /* lets try to make an inline extent */ | |
00361589 JB |
946 | ret = cow_file_range_inline(root, inode, start, end, 0, 0, |
947 | NULL); | |
771ed689 | 948 | if (ret == 0) { |
c2790a2e JB |
949 | extent_clear_unlock_delalloc(inode, start, end, NULL, |
950 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc | 951 | EXTENT_DEFRAG, PAGE_UNLOCK | |
c2790a2e JB |
952 | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK | |
953 | PAGE_END_WRITEBACK); | |
c2167754 | 954 | |
771ed689 CM |
955 | *nr_written = *nr_written + |
956 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
957 | *page_started = 1; | |
771ed689 | 958 | goto out; |
79787eaa | 959 | } else if (ret < 0) { |
79787eaa | 960 | goto out_unlock; |
771ed689 CM |
961 | } |
962 | } | |
963 | ||
964 | BUG_ON(disk_num_bytes > | |
6c41761f | 965 | btrfs_super_total_bytes(root->fs_info->super_copy)); |
771ed689 | 966 | |
4b46fce2 | 967 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
771ed689 CM |
968 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
969 | ||
d397712b | 970 | while (disk_num_bytes > 0) { |
a791e35e CM |
971 | unsigned long op; |
972 | ||
287a0ab9 | 973 | cur_alloc_size = disk_num_bytes; |
00361589 | 974 | ret = btrfs_reserve_extent(root, cur_alloc_size, |
771ed689 | 975 | root->sectorsize, 0, alloc_hint, |
e570fd27 | 976 | &ins, 1, 1); |
00361589 | 977 | if (ret < 0) |
79787eaa | 978 | goto out_unlock; |
d397712b | 979 | |
172ddd60 | 980 | em = alloc_extent_map(); |
b9aa55be LB |
981 | if (!em) { |
982 | ret = -ENOMEM; | |
ace68bac | 983 | goto out_reserve; |
b9aa55be | 984 | } |
e6dcd2dc | 985 | em->start = start; |
445a6944 | 986 | em->orig_start = em->start; |
771ed689 CM |
987 | ram_size = ins.offset; |
988 | em->len = ins.offset; | |
2ab28f32 JB |
989 | em->mod_start = em->start; |
990 | em->mod_len = em->len; | |
c8b97818 | 991 | |
e6dcd2dc | 992 | em->block_start = ins.objectid; |
c8b97818 | 993 | em->block_len = ins.offset; |
b4939680 | 994 | em->orig_block_len = ins.offset; |
cc95bef6 | 995 | em->ram_bytes = ram_size; |
e6dcd2dc | 996 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 997 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
70c8a91c | 998 | em->generation = -1; |
c8b97818 | 999 | |
d397712b | 1000 | while (1) { |
890871be | 1001 | write_lock(&em_tree->lock); |
09a2a8f9 | 1002 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 1003 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
1004 | if (ret != -EEXIST) { |
1005 | free_extent_map(em); | |
1006 | break; | |
1007 | } | |
1008 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 1009 | start + ram_size - 1, 0); |
e6dcd2dc | 1010 | } |
ace68bac LB |
1011 | if (ret) |
1012 | goto out_reserve; | |
e6dcd2dc | 1013 | |
98d20f67 | 1014 | cur_alloc_size = ins.offset; |
e6dcd2dc | 1015 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 1016 | ram_size, cur_alloc_size, 0); |
ace68bac | 1017 | if (ret) |
d9f85963 | 1018 | goto out_drop_extent_cache; |
c8b97818 | 1019 | |
17d217fe YZ |
1020 | if (root->root_key.objectid == |
1021 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1022 | ret = btrfs_reloc_clone_csums(inode, start, | |
1023 | cur_alloc_size); | |
00361589 | 1024 | if (ret) |
d9f85963 | 1025 | goto out_drop_extent_cache; |
17d217fe YZ |
1026 | } |
1027 | ||
d397712b | 1028 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 1029 | break; |
d397712b | 1030 | |
c8b97818 CM |
1031 | /* we're not doing compressed IO, don't unlock the first |
1032 | * page (which the caller expects to stay locked), don't | |
1033 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
1034 | * |
1035 | * Do set the Private2 bit so we know this page was properly | |
1036 | * setup for writepage | |
c8b97818 | 1037 | */ |
c2790a2e JB |
1038 | op = unlock ? PAGE_UNLOCK : 0; |
1039 | op |= PAGE_SET_PRIVATE2; | |
a791e35e | 1040 | |
c2790a2e JB |
1041 | extent_clear_unlock_delalloc(inode, start, |
1042 | start + ram_size - 1, locked_page, | |
1043 | EXTENT_LOCKED | EXTENT_DELALLOC, | |
1044 | op); | |
c8b97818 | 1045 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
1046 | num_bytes -= cur_alloc_size; |
1047 | alloc_hint = ins.objectid + ins.offset; | |
1048 | start += cur_alloc_size; | |
b888db2b | 1049 | } |
79787eaa | 1050 | out: |
be20aa9d | 1051 | return ret; |
b7d5b0a8 | 1052 | |
d9f85963 FM |
1053 | out_drop_extent_cache: |
1054 | btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0); | |
ace68bac | 1055 | out_reserve: |
e570fd27 | 1056 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1); |
79787eaa | 1057 | out_unlock: |
c2790a2e | 1058 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
151a41bc JB |
1059 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | |
1060 | EXTENT_DELALLOC | EXTENT_DEFRAG, | |
1061 | PAGE_UNLOCK | PAGE_CLEAR_DIRTY | | |
1062 | PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK); | |
79787eaa | 1063 | goto out; |
771ed689 | 1064 | } |
c8b97818 | 1065 | |
771ed689 CM |
1066 | /* |
1067 | * work queue call back to started compression on a file and pages | |
1068 | */ | |
1069 | static noinline void async_cow_start(struct btrfs_work *work) | |
1070 | { | |
1071 | struct async_cow *async_cow; | |
1072 | int num_added = 0; | |
1073 | async_cow = container_of(work, struct async_cow, work); | |
1074 | ||
1075 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
1076 | async_cow->start, async_cow->end, async_cow, | |
1077 | &num_added); | |
8180ef88 | 1078 | if (num_added == 0) { |
cb77fcd8 | 1079 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 | 1080 | async_cow->inode = NULL; |
8180ef88 | 1081 | } |
771ed689 CM |
1082 | } |
1083 | ||
1084 | /* | |
1085 | * work queue call back to submit previously compressed pages | |
1086 | */ | |
1087 | static noinline void async_cow_submit(struct btrfs_work *work) | |
1088 | { | |
1089 | struct async_cow *async_cow; | |
1090 | struct btrfs_root *root; | |
1091 | unsigned long nr_pages; | |
1092 | ||
1093 | async_cow = container_of(work, struct async_cow, work); | |
1094 | ||
1095 | root = async_cow->root; | |
1096 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
1097 | PAGE_CACHE_SHIFT; | |
1098 | ||
ee863954 DS |
1099 | /* |
1100 | * atomic_sub_return implies a barrier for waitqueue_active | |
1101 | */ | |
66657b31 | 1102 | if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) < |
287082b0 | 1103 | 5 * 1024 * 1024 && |
771ed689 CM |
1104 | waitqueue_active(&root->fs_info->async_submit_wait)) |
1105 | wake_up(&root->fs_info->async_submit_wait); | |
1106 | ||
d397712b | 1107 | if (async_cow->inode) |
771ed689 | 1108 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 1109 | } |
c8b97818 | 1110 | |
771ed689 CM |
1111 | static noinline void async_cow_free(struct btrfs_work *work) |
1112 | { | |
1113 | struct async_cow *async_cow; | |
1114 | async_cow = container_of(work, struct async_cow, work); | |
8180ef88 | 1115 | if (async_cow->inode) |
cb77fcd8 | 1116 | btrfs_add_delayed_iput(async_cow->inode); |
771ed689 CM |
1117 | kfree(async_cow); |
1118 | } | |
1119 | ||
1120 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
1121 | u64 start, u64 end, int *page_started, | |
1122 | unsigned long *nr_written) | |
1123 | { | |
1124 | struct async_cow *async_cow; | |
1125 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1126 | unsigned long nr_pages; | |
1127 | u64 cur_end; | |
287082b0 | 1128 | int limit = 10 * 1024 * 1024; |
771ed689 | 1129 | |
a3429ab7 CM |
1130 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
1131 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 1132 | while (start < end) { |
771ed689 | 1133 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
79787eaa | 1134 | BUG_ON(!async_cow); /* -ENOMEM */ |
8180ef88 | 1135 | async_cow->inode = igrab(inode); |
771ed689 CM |
1136 | async_cow->root = root; |
1137 | async_cow->locked_page = locked_page; | |
1138 | async_cow->start = start; | |
1139 | ||
f79707b0 WS |
1140 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS && |
1141 | !btrfs_test_opt(root, FORCE_COMPRESS)) | |
771ed689 CM |
1142 | cur_end = end; |
1143 | else | |
1144 | cur_end = min(end, start + 512 * 1024 - 1); | |
1145 | ||
1146 | async_cow->end = cur_end; | |
1147 | INIT_LIST_HEAD(&async_cow->extents); | |
1148 | ||
9e0af237 LB |
1149 | btrfs_init_work(&async_cow->work, |
1150 | btrfs_delalloc_helper, | |
1151 | async_cow_start, async_cow_submit, | |
1152 | async_cow_free); | |
771ed689 | 1153 | |
771ed689 CM |
1154 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
1155 | PAGE_CACHE_SHIFT; | |
1156 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
1157 | ||
afe3d242 QW |
1158 | btrfs_queue_work(root->fs_info->delalloc_workers, |
1159 | &async_cow->work); | |
771ed689 CM |
1160 | |
1161 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
1162 | wait_event(root->fs_info->async_submit_wait, | |
1163 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
1164 | limit)); | |
1165 | } | |
1166 | ||
d397712b | 1167 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
1168 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
1169 | wait_event(root->fs_info->async_submit_wait, | |
1170 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
1171 | 0)); | |
1172 | } | |
1173 | ||
1174 | *nr_written += nr_pages; | |
1175 | start = cur_end + 1; | |
1176 | } | |
1177 | *page_started = 1; | |
1178 | return 0; | |
be20aa9d CM |
1179 | } |
1180 | ||
d397712b | 1181 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
1182 | u64 bytenr, u64 num_bytes) |
1183 | { | |
1184 | int ret; | |
1185 | struct btrfs_ordered_sum *sums; | |
1186 | LIST_HEAD(list); | |
1187 | ||
07d400a6 | 1188 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
a2de733c | 1189 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1190 | if (ret == 0 && list_empty(&list)) |
1191 | return 0; | |
1192 | ||
1193 | while (!list_empty(&list)) { | |
1194 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1195 | list_del(&sums->list); | |
1196 | kfree(sums); | |
1197 | } | |
1198 | return 1; | |
1199 | } | |
1200 | ||
d352ac68 CM |
1201 | /* |
1202 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1203 | * of the extents that exist in the file, and COWs the file as required. | |
1204 | * | |
1205 | * If no cow copies or snapshots exist, we write directly to the existing | |
1206 | * blocks on disk | |
1207 | */ | |
7f366cfe CM |
1208 | static noinline int run_delalloc_nocow(struct inode *inode, |
1209 | struct page *locked_page, | |
771ed689 CM |
1210 | u64 start, u64 end, int *page_started, int force, |
1211 | unsigned long *nr_written) | |
be20aa9d | 1212 | { |
be20aa9d | 1213 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 1214 | struct btrfs_trans_handle *trans; |
be20aa9d | 1215 | struct extent_buffer *leaf; |
be20aa9d | 1216 | struct btrfs_path *path; |
80ff3856 | 1217 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1218 | struct btrfs_key found_key; |
80ff3856 YZ |
1219 | u64 cow_start; |
1220 | u64 cur_offset; | |
1221 | u64 extent_end; | |
5d4f98a2 | 1222 | u64 extent_offset; |
80ff3856 YZ |
1223 | u64 disk_bytenr; |
1224 | u64 num_bytes; | |
b4939680 | 1225 | u64 disk_num_bytes; |
cc95bef6 | 1226 | u64 ram_bytes; |
80ff3856 | 1227 | int extent_type; |
79787eaa | 1228 | int ret, err; |
d899e052 | 1229 | int type; |
80ff3856 YZ |
1230 | int nocow; |
1231 | int check_prev = 1; | |
82d5902d | 1232 | bool nolock; |
33345d01 | 1233 | u64 ino = btrfs_ino(inode); |
be20aa9d CM |
1234 | |
1235 | path = btrfs_alloc_path(); | |
17ca04af | 1236 | if (!path) { |
c2790a2e JB |
1237 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
1238 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc JB |
1239 | EXTENT_DO_ACCOUNTING | |
1240 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1241 | PAGE_CLEAR_DIRTY | |
1242 | PAGE_SET_WRITEBACK | | |
1243 | PAGE_END_WRITEBACK); | |
d8926bb3 | 1244 | return -ENOMEM; |
17ca04af | 1245 | } |
82d5902d | 1246 | |
83eea1f1 | 1247 | nolock = btrfs_is_free_space_inode(inode); |
82d5902d LZ |
1248 | |
1249 | if (nolock) | |
7a7eaa40 | 1250 | trans = btrfs_join_transaction_nolock(root); |
82d5902d | 1251 | else |
7a7eaa40 | 1252 | trans = btrfs_join_transaction(root); |
ff5714cc | 1253 | |
79787eaa | 1254 | if (IS_ERR(trans)) { |
c2790a2e JB |
1255 | extent_clear_unlock_delalloc(inode, start, end, locked_page, |
1256 | EXTENT_LOCKED | EXTENT_DELALLOC | | |
151a41bc JB |
1257 | EXTENT_DO_ACCOUNTING | |
1258 | EXTENT_DEFRAG, PAGE_UNLOCK | | |
c2790a2e JB |
1259 | PAGE_CLEAR_DIRTY | |
1260 | PAGE_SET_WRITEBACK | | |
1261 | PAGE_END_WRITEBACK); | |
79787eaa JM |
1262 | btrfs_free_path(path); |
1263 | return PTR_ERR(trans); | |
1264 | } | |
1265 | ||
74b21075 | 1266 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
be20aa9d | 1267 | |
80ff3856 YZ |
1268 | cow_start = (u64)-1; |
1269 | cur_offset = start; | |
1270 | while (1) { | |
33345d01 | 1271 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
80ff3856 | 1272 | cur_offset, 0); |
d788a349 | 1273 | if (ret < 0) |
79787eaa | 1274 | goto error; |
80ff3856 YZ |
1275 | if (ret > 0 && path->slots[0] > 0 && check_prev) { |
1276 | leaf = path->nodes[0]; | |
1277 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1278 | path->slots[0] - 1); | |
33345d01 | 1279 | if (found_key.objectid == ino && |
80ff3856 YZ |
1280 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1281 | path->slots[0]--; | |
1282 | } | |
1283 | check_prev = 0; | |
1284 | next_slot: | |
1285 | leaf = path->nodes[0]; | |
1286 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1287 | ret = btrfs_next_leaf(root, path); | |
d788a349 | 1288 | if (ret < 0) |
79787eaa | 1289 | goto error; |
80ff3856 YZ |
1290 | if (ret > 0) |
1291 | break; | |
1292 | leaf = path->nodes[0]; | |
1293 | } | |
be20aa9d | 1294 | |
80ff3856 YZ |
1295 | nocow = 0; |
1296 | disk_bytenr = 0; | |
17d217fe | 1297 | num_bytes = 0; |
80ff3856 YZ |
1298 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1299 | ||
33345d01 | 1300 | if (found_key.objectid > ino || |
80ff3856 YZ |
1301 | found_key.type > BTRFS_EXTENT_DATA_KEY || |
1302 | found_key.offset > end) | |
1303 | break; | |
1304 | ||
1305 | if (found_key.offset > cur_offset) { | |
1306 | extent_end = found_key.offset; | |
e9061e21 | 1307 | extent_type = 0; |
80ff3856 YZ |
1308 | goto out_check; |
1309 | } | |
1310 | ||
1311 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1312 | struct btrfs_file_extent_item); | |
1313 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1314 | ||
cc95bef6 | 1315 | ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); |
d899e052 YZ |
1316 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1317 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1318 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1319 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1320 | extent_end = found_key.offset + |
1321 | btrfs_file_extent_num_bytes(leaf, fi); | |
b4939680 JB |
1322 | disk_num_bytes = |
1323 | btrfs_file_extent_disk_num_bytes(leaf, fi); | |
80ff3856 YZ |
1324 | if (extent_end <= start) { |
1325 | path->slots[0]++; | |
1326 | goto next_slot; | |
1327 | } | |
17d217fe YZ |
1328 | if (disk_bytenr == 0) |
1329 | goto out_check; | |
80ff3856 YZ |
1330 | if (btrfs_file_extent_compression(leaf, fi) || |
1331 | btrfs_file_extent_encryption(leaf, fi) || | |
1332 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1333 | goto out_check; | |
d899e052 YZ |
1334 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1335 | goto out_check; | |
d2fb3437 | 1336 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1337 | goto out_check; |
33345d01 | 1338 | if (btrfs_cross_ref_exist(trans, root, ino, |
5d4f98a2 YZ |
1339 | found_key.offset - |
1340 | extent_offset, disk_bytenr)) | |
17d217fe | 1341 | goto out_check; |
5d4f98a2 | 1342 | disk_bytenr += extent_offset; |
17d217fe YZ |
1343 | disk_bytenr += cur_offset - found_key.offset; |
1344 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
e9894fd3 WS |
1345 | /* |
1346 | * if there are pending snapshots for this root, | |
1347 | * we fall into common COW way. | |
1348 | */ | |
1349 | if (!nolock) { | |
9ea24bbe | 1350 | err = btrfs_start_write_no_snapshoting(root); |
e9894fd3 WS |
1351 | if (!err) |
1352 | goto out_check; | |
1353 | } | |
17d217fe YZ |
1354 | /* |
1355 | * force cow if csum exists in the range. | |
1356 | * this ensure that csum for a given extent are | |
1357 | * either valid or do not exist. | |
1358 | */ | |
1359 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1360 | goto out_check; | |
80ff3856 YZ |
1361 | nocow = 1; |
1362 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1363 | extent_end = found_key.offset + | |
514ac8ad CM |
1364 | btrfs_file_extent_inline_len(leaf, |
1365 | path->slots[0], fi); | |
80ff3856 YZ |
1366 | extent_end = ALIGN(extent_end, root->sectorsize); |
1367 | } else { | |
1368 | BUG_ON(1); | |
1369 | } | |
1370 | out_check: | |
1371 | if (extent_end <= start) { | |
1372 | path->slots[0]++; | |
e9894fd3 | 1373 | if (!nolock && nocow) |
9ea24bbe | 1374 | btrfs_end_write_no_snapshoting(root); |
80ff3856 YZ |
1375 | goto next_slot; |
1376 | } | |
1377 | if (!nocow) { | |
1378 | if (cow_start == (u64)-1) | |
1379 | cow_start = cur_offset; | |
1380 | cur_offset = extent_end; | |
1381 | if (cur_offset > end) | |
1382 | break; | |
1383 | path->slots[0]++; | |
1384 | goto next_slot; | |
7ea394f1 YZ |
1385 | } |
1386 | ||
b3b4aa74 | 1387 | btrfs_release_path(path); |
80ff3856 | 1388 | if (cow_start != (u64)-1) { |
00361589 JB |
1389 | ret = cow_file_range(inode, locked_page, |
1390 | cow_start, found_key.offset - 1, | |
1391 | page_started, nr_written, 1); | |
e9894fd3 WS |
1392 | if (ret) { |
1393 | if (!nolock && nocow) | |
9ea24bbe | 1394 | btrfs_end_write_no_snapshoting(root); |
79787eaa | 1395 | goto error; |
e9894fd3 | 1396 | } |
80ff3856 | 1397 | cow_start = (u64)-1; |
7ea394f1 | 1398 | } |
80ff3856 | 1399 | |
d899e052 YZ |
1400 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1401 | struct extent_map *em; | |
1402 | struct extent_map_tree *em_tree; | |
1403 | em_tree = &BTRFS_I(inode)->extent_tree; | |
172ddd60 | 1404 | em = alloc_extent_map(); |
79787eaa | 1405 | BUG_ON(!em); /* -ENOMEM */ |
d899e052 | 1406 | em->start = cur_offset; |
70c8a91c | 1407 | em->orig_start = found_key.offset - extent_offset; |
d899e052 YZ |
1408 | em->len = num_bytes; |
1409 | em->block_len = num_bytes; | |
1410 | em->block_start = disk_bytenr; | |
b4939680 | 1411 | em->orig_block_len = disk_num_bytes; |
cc95bef6 | 1412 | em->ram_bytes = ram_bytes; |
d899e052 | 1413 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
2ab28f32 JB |
1414 | em->mod_start = em->start; |
1415 | em->mod_len = em->len; | |
d899e052 | 1416 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
b11e234d | 1417 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
70c8a91c | 1418 | em->generation = -1; |
d899e052 | 1419 | while (1) { |
890871be | 1420 | write_lock(&em_tree->lock); |
09a2a8f9 | 1421 | ret = add_extent_mapping(em_tree, em, 1); |
890871be | 1422 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1423 | if (ret != -EEXIST) { |
1424 | free_extent_map(em); | |
1425 | break; | |
1426 | } | |
1427 | btrfs_drop_extent_cache(inode, em->start, | |
1428 | em->start + em->len - 1, 0); | |
1429 | } | |
1430 | type = BTRFS_ORDERED_PREALLOC; | |
1431 | } else { | |
1432 | type = BTRFS_ORDERED_NOCOW; | |
1433 | } | |
80ff3856 YZ |
1434 | |
1435 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 | 1436 | num_bytes, num_bytes, type); |
79787eaa | 1437 | BUG_ON(ret); /* -ENOMEM */ |
771ed689 | 1438 | |
efa56464 YZ |
1439 | if (root->root_key.objectid == |
1440 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1441 | ret = btrfs_reloc_clone_csums(inode, cur_offset, | |
1442 | num_bytes); | |
e9894fd3 WS |
1443 | if (ret) { |
1444 | if (!nolock && nocow) | |
9ea24bbe | 1445 | btrfs_end_write_no_snapshoting(root); |
79787eaa | 1446 | goto error; |
e9894fd3 | 1447 | } |
efa56464 YZ |
1448 | } |
1449 | ||
c2790a2e JB |
1450 | extent_clear_unlock_delalloc(inode, cur_offset, |
1451 | cur_offset + num_bytes - 1, | |
1452 | locked_page, EXTENT_LOCKED | | |
1453 | EXTENT_DELALLOC, PAGE_UNLOCK | | |
1454 | PAGE_SET_PRIVATE2); | |
e9894fd3 | 1455 | if (!nolock && nocow) |
9ea24bbe | 1456 | btrfs_end_write_no_snapshoting(root); |
80ff3856 YZ |
1457 | cur_offset = extent_end; |
1458 | if (cur_offset > end) | |
1459 | break; | |
be20aa9d | 1460 | } |
b3b4aa74 | 1461 | btrfs_release_path(path); |
80ff3856 | 1462 | |
17ca04af | 1463 | if (cur_offset <= end && cow_start == (u64)-1) { |
80ff3856 | 1464 | cow_start = cur_offset; |
17ca04af JB |
1465 | cur_offset = end; |
1466 | } | |
1467 | ||
80ff3856 | 1468 | if (cow_start != (u64)-1) { |
00361589 JB |
1469 | ret = cow_file_range(inode, locked_page, cow_start, end, |
1470 | page_started, nr_written, 1); | |
d788a349 | 1471 | if (ret) |
79787eaa | 1472 | goto error; |
80ff3856 YZ |
1473 | } |
1474 | ||
79787eaa | 1475 | error: |
a698d075 | 1476 | err = btrfs_end_transaction(trans, root); |
79787eaa JM |
1477 | if (!ret) |
1478 | ret = err; | |
1479 | ||
17ca04af | 1480 | if (ret && cur_offset < end) |
c2790a2e JB |
1481 | extent_clear_unlock_delalloc(inode, cur_offset, end, |
1482 | locked_page, EXTENT_LOCKED | | |
151a41bc JB |
1483 | EXTENT_DELALLOC | EXTENT_DEFRAG | |
1484 | EXTENT_DO_ACCOUNTING, PAGE_UNLOCK | | |
1485 | PAGE_CLEAR_DIRTY | | |
c2790a2e JB |
1486 | PAGE_SET_WRITEBACK | |
1487 | PAGE_END_WRITEBACK); | |
7ea394f1 | 1488 | btrfs_free_path(path); |
79787eaa | 1489 | return ret; |
be20aa9d CM |
1490 | } |
1491 | ||
47059d93 WS |
1492 | static inline int need_force_cow(struct inode *inode, u64 start, u64 end) |
1493 | { | |
1494 | ||
1495 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
1496 | !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)) | |
1497 | return 0; | |
1498 | ||
1499 | /* | |
1500 | * @defrag_bytes is a hint value, no spinlock held here, | |
1501 | * if is not zero, it means the file is defragging. | |
1502 | * Force cow if given extent needs to be defragged. | |
1503 | */ | |
1504 | if (BTRFS_I(inode)->defrag_bytes && | |
1505 | test_range_bit(&BTRFS_I(inode)->io_tree, start, end, | |
1506 | EXTENT_DEFRAG, 0, NULL)) | |
1507 | return 1; | |
1508 | ||
1509 | return 0; | |
1510 | } | |
1511 | ||
d352ac68 CM |
1512 | /* |
1513 | * extent_io.c call back to do delayed allocation processing | |
1514 | */ | |
c8b97818 | 1515 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1516 | u64 start, u64 end, int *page_started, |
1517 | unsigned long *nr_written) | |
be20aa9d | 1518 | { |
be20aa9d | 1519 | int ret; |
47059d93 | 1520 | int force_cow = need_force_cow(inode, start, end); |
a2135011 | 1521 | |
47059d93 | 1522 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) { |
c8b97818 | 1523 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1524 | page_started, 1, nr_written); |
47059d93 | 1525 | } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) { |
d899e052 | 1526 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1527 | page_started, 0, nr_written); |
7816030e | 1528 | } else if (!inode_need_compress(inode)) { |
7f366cfe CM |
1529 | ret = cow_file_range(inode, locked_page, start, end, |
1530 | page_started, nr_written, 1); | |
7ddf5a42 JB |
1531 | } else { |
1532 | set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
1533 | &BTRFS_I(inode)->runtime_flags); | |
771ed689 | 1534 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1535 | page_started, nr_written); |
7ddf5a42 | 1536 | } |
b888db2b CM |
1537 | return ret; |
1538 | } | |
1539 | ||
1bf85046 JM |
1540 | static void btrfs_split_extent_hook(struct inode *inode, |
1541 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1542 | { |
dcab6a3b JB |
1543 | u64 size; |
1544 | ||
0ca1f7ce | 1545 | /* not delalloc, ignore it */ |
9ed74f2d | 1546 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1547 | return; |
9ed74f2d | 1548 | |
dcab6a3b JB |
1549 | size = orig->end - orig->start + 1; |
1550 | if (size > BTRFS_MAX_EXTENT_SIZE) { | |
1551 | u64 num_extents; | |
1552 | u64 new_size; | |
1553 | ||
1554 | /* | |
ba117213 JB |
1555 | * See the explanation in btrfs_merge_extent_hook, the same |
1556 | * applies here, just in reverse. | |
dcab6a3b JB |
1557 | */ |
1558 | new_size = orig->end - split + 1; | |
ba117213 | 1559 | num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1, |
dcab6a3b | 1560 | BTRFS_MAX_EXTENT_SIZE); |
ba117213 JB |
1561 | new_size = split - orig->start; |
1562 | num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1, | |
1563 | BTRFS_MAX_EXTENT_SIZE); | |
1564 | if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1, | |
1565 | BTRFS_MAX_EXTENT_SIZE) >= num_extents) | |
dcab6a3b JB |
1566 | return; |
1567 | } | |
1568 | ||
9e0baf60 JB |
1569 | spin_lock(&BTRFS_I(inode)->lock); |
1570 | BTRFS_I(inode)->outstanding_extents++; | |
1571 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1572 | } |
1573 | ||
1574 | /* | |
1575 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1576 | * extents so we can keep track of new extents that are just merged onto old | |
1577 | * extents, such as when we are doing sequential writes, so we can properly | |
1578 | * account for the metadata space we'll need. | |
1579 | */ | |
1bf85046 JM |
1580 | static void btrfs_merge_extent_hook(struct inode *inode, |
1581 | struct extent_state *new, | |
1582 | struct extent_state *other) | |
9ed74f2d | 1583 | { |
dcab6a3b JB |
1584 | u64 new_size, old_size; |
1585 | u64 num_extents; | |
1586 | ||
9ed74f2d JB |
1587 | /* not delalloc, ignore it */ |
1588 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1589 | return; |
9ed74f2d | 1590 | |
8461a3de JB |
1591 | if (new->start > other->start) |
1592 | new_size = new->end - other->start + 1; | |
1593 | else | |
1594 | new_size = other->end - new->start + 1; | |
dcab6a3b JB |
1595 | |
1596 | /* we're not bigger than the max, unreserve the space and go */ | |
1597 | if (new_size <= BTRFS_MAX_EXTENT_SIZE) { | |
1598 | spin_lock(&BTRFS_I(inode)->lock); | |
1599 | BTRFS_I(inode)->outstanding_extents--; | |
1600 | spin_unlock(&BTRFS_I(inode)->lock); | |
1601 | return; | |
1602 | } | |
1603 | ||
1604 | /* | |
ba117213 JB |
1605 | * We have to add up either side to figure out how many extents were |
1606 | * accounted for before we merged into one big extent. If the number of | |
1607 | * extents we accounted for is <= the amount we need for the new range | |
1608 | * then we can return, otherwise drop. Think of it like this | |
1609 | * | |
1610 | * [ 4k][MAX_SIZE] | |
1611 | * | |
1612 | * So we've grown the extent by a MAX_SIZE extent, this would mean we | |
1613 | * need 2 outstanding extents, on one side we have 1 and the other side | |
1614 | * we have 1 so they are == and we can return. But in this case | |
1615 | * | |
1616 | * [MAX_SIZE+4k][MAX_SIZE+4k] | |
1617 | * | |
1618 | * Each range on their own accounts for 2 extents, but merged together | |
1619 | * they are only 3 extents worth of accounting, so we need to drop in | |
1620 | * this case. | |
dcab6a3b | 1621 | */ |
ba117213 | 1622 | old_size = other->end - other->start + 1; |
dcab6a3b JB |
1623 | num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1, |
1624 | BTRFS_MAX_EXTENT_SIZE); | |
ba117213 JB |
1625 | old_size = new->end - new->start + 1; |
1626 | num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1, | |
1627 | BTRFS_MAX_EXTENT_SIZE); | |
1628 | ||
dcab6a3b | 1629 | if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1, |
ba117213 | 1630 | BTRFS_MAX_EXTENT_SIZE) >= num_extents) |
dcab6a3b JB |
1631 | return; |
1632 | ||
9e0baf60 JB |
1633 | spin_lock(&BTRFS_I(inode)->lock); |
1634 | BTRFS_I(inode)->outstanding_extents--; | |
1635 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1636 | } |
1637 | ||
eb73c1b7 MX |
1638 | static void btrfs_add_delalloc_inodes(struct btrfs_root *root, |
1639 | struct inode *inode) | |
1640 | { | |
1641 | spin_lock(&root->delalloc_lock); | |
1642 | if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1643 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, | |
1644 | &root->delalloc_inodes); | |
1645 | set_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1646 | &BTRFS_I(inode)->runtime_flags); | |
1647 | root->nr_delalloc_inodes++; | |
1648 | if (root->nr_delalloc_inodes == 1) { | |
1649 | spin_lock(&root->fs_info->delalloc_root_lock); | |
1650 | BUG_ON(!list_empty(&root->delalloc_root)); | |
1651 | list_add_tail(&root->delalloc_root, | |
1652 | &root->fs_info->delalloc_roots); | |
1653 | spin_unlock(&root->fs_info->delalloc_root_lock); | |
1654 | } | |
1655 | } | |
1656 | spin_unlock(&root->delalloc_lock); | |
1657 | } | |
1658 | ||
1659 | static void btrfs_del_delalloc_inode(struct btrfs_root *root, | |
1660 | struct inode *inode) | |
1661 | { | |
1662 | spin_lock(&root->delalloc_lock); | |
1663 | if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) { | |
1664 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1665 | clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, | |
1666 | &BTRFS_I(inode)->runtime_flags); | |
1667 | root->nr_delalloc_inodes--; | |
1668 | if (!root->nr_delalloc_inodes) { | |
1669 | spin_lock(&root->fs_info->delalloc_root_lock); | |
1670 | BUG_ON(list_empty(&root->delalloc_root)); | |
1671 | list_del_init(&root->delalloc_root); | |
1672 | spin_unlock(&root->fs_info->delalloc_root_lock); | |
1673 | } | |
1674 | } | |
1675 | spin_unlock(&root->delalloc_lock); | |
1676 | } | |
1677 | ||
d352ac68 CM |
1678 | /* |
1679 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1680 | * bytes in this file, and to maintain the list of inodes that | |
1681 | * have pending delalloc work to be done. | |
1682 | */ | |
1bf85046 | 1683 | static void btrfs_set_bit_hook(struct inode *inode, |
9ee49a04 | 1684 | struct extent_state *state, unsigned *bits) |
291d673e | 1685 | { |
9ed74f2d | 1686 | |
47059d93 WS |
1687 | if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC)) |
1688 | WARN_ON(1); | |
75eff68e CM |
1689 | /* |
1690 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1691 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1692 | * bit, which is only set or cleared with irqs on |
1693 | */ | |
0ca1f7ce | 1694 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1695 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1696 | u64 len = state->end + 1 - state->start; |
83eea1f1 | 1697 | bool do_list = !btrfs_is_free_space_inode(inode); |
9ed74f2d | 1698 | |
9e0baf60 | 1699 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1700 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1701 | } else { |
1702 | spin_lock(&BTRFS_I(inode)->lock); | |
1703 | BTRFS_I(inode)->outstanding_extents++; | |
1704 | spin_unlock(&BTRFS_I(inode)->lock); | |
1705 | } | |
287a0ab9 | 1706 | |
6a3891c5 JB |
1707 | /* For sanity tests */ |
1708 | if (btrfs_test_is_dummy_root(root)) | |
1709 | return; | |
1710 | ||
963d678b MX |
1711 | __percpu_counter_add(&root->fs_info->delalloc_bytes, len, |
1712 | root->fs_info->delalloc_batch); | |
df0af1a5 | 1713 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1714 | BTRFS_I(inode)->delalloc_bytes += len; |
47059d93 WS |
1715 | if (*bits & EXTENT_DEFRAG) |
1716 | BTRFS_I(inode)->defrag_bytes += len; | |
df0af1a5 | 1717 | if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1718 | &BTRFS_I(inode)->runtime_flags)) |
1719 | btrfs_add_delalloc_inodes(root, inode); | |
df0af1a5 | 1720 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1721 | } |
291d673e CM |
1722 | } |
1723 | ||
d352ac68 CM |
1724 | /* |
1725 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1726 | */ | |
1bf85046 | 1727 | static void btrfs_clear_bit_hook(struct inode *inode, |
41074888 | 1728 | struct extent_state *state, |
9ee49a04 | 1729 | unsigned *bits) |
291d673e | 1730 | { |
47059d93 | 1731 | u64 len = state->end + 1 - state->start; |
dcab6a3b JB |
1732 | u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1, |
1733 | BTRFS_MAX_EXTENT_SIZE); | |
47059d93 WS |
1734 | |
1735 | spin_lock(&BTRFS_I(inode)->lock); | |
1736 | if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) | |
1737 | BTRFS_I(inode)->defrag_bytes -= len; | |
1738 | spin_unlock(&BTRFS_I(inode)->lock); | |
1739 | ||
75eff68e CM |
1740 | /* |
1741 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1742 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1743 | * bit, which is only set or cleared with irqs on |
1744 | */ | |
0ca1f7ce | 1745 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1746 | struct btrfs_root *root = BTRFS_I(inode)->root; |
83eea1f1 | 1747 | bool do_list = !btrfs_is_free_space_inode(inode); |
bcbfce8a | 1748 | |
9e0baf60 | 1749 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1750 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1751 | } else if (!(*bits & EXTENT_DO_ACCOUNTING)) { |
1752 | spin_lock(&BTRFS_I(inode)->lock); | |
dcab6a3b | 1753 | BTRFS_I(inode)->outstanding_extents -= num_extents; |
9e0baf60 JB |
1754 | spin_unlock(&BTRFS_I(inode)->lock); |
1755 | } | |
0ca1f7ce | 1756 | |
b6d08f06 JB |
1757 | /* |
1758 | * We don't reserve metadata space for space cache inodes so we | |
1759 | * don't need to call dellalloc_release_metadata if there is an | |
1760 | * error. | |
1761 | */ | |
1762 | if (*bits & EXTENT_DO_ACCOUNTING && | |
1763 | root != root->fs_info->tree_root) | |
0ca1f7ce YZ |
1764 | btrfs_delalloc_release_metadata(inode, len); |
1765 | ||
6a3891c5 JB |
1766 | /* For sanity tests. */ |
1767 | if (btrfs_test_is_dummy_root(root)) | |
1768 | return; | |
1769 | ||
0cb59c99 | 1770 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
7ee9e440 | 1771 | && do_list && !(state->state & EXTENT_NORESERVE)) |
0ca1f7ce | 1772 | btrfs_free_reserved_data_space(inode, len); |
9ed74f2d | 1773 | |
963d678b MX |
1774 | __percpu_counter_add(&root->fs_info->delalloc_bytes, -len, |
1775 | root->fs_info->delalloc_batch); | |
df0af1a5 | 1776 | spin_lock(&BTRFS_I(inode)->lock); |
0ca1f7ce | 1777 | BTRFS_I(inode)->delalloc_bytes -= len; |
0cb59c99 | 1778 | if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 && |
df0af1a5 | 1779 | test_bit(BTRFS_INODE_IN_DELALLOC_LIST, |
eb73c1b7 MX |
1780 | &BTRFS_I(inode)->runtime_flags)) |
1781 | btrfs_del_delalloc_inode(root, inode); | |
df0af1a5 | 1782 | spin_unlock(&BTRFS_I(inode)->lock); |
291d673e | 1783 | } |
291d673e CM |
1784 | } |
1785 | ||
d352ac68 CM |
1786 | /* |
1787 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1788 | * we don't create bios that span stripes or chunks | |
1789 | */ | |
64a16701 | 1790 | int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset, |
c8b97818 CM |
1791 | size_t size, struct bio *bio, |
1792 | unsigned long bio_flags) | |
239b14b3 CM |
1793 | { |
1794 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
4f024f37 | 1795 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
239b14b3 CM |
1796 | u64 length = 0; |
1797 | u64 map_length; | |
239b14b3 CM |
1798 | int ret; |
1799 | ||
771ed689 CM |
1800 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1801 | return 0; | |
1802 | ||
4f024f37 | 1803 | length = bio->bi_iter.bi_size; |
239b14b3 | 1804 | map_length = length; |
64a16701 | 1805 | ret = btrfs_map_block(root->fs_info, rw, logical, |
f188591e | 1806 | &map_length, NULL, 0); |
3ec706c8 | 1807 | /* Will always return 0 with map_multi == NULL */ |
3444a972 | 1808 | BUG_ON(ret < 0); |
d397712b | 1809 | if (map_length < length + size) |
239b14b3 | 1810 | return 1; |
3444a972 | 1811 | return 0; |
239b14b3 CM |
1812 | } |
1813 | ||
d352ac68 CM |
1814 | /* |
1815 | * in order to insert checksums into the metadata in large chunks, | |
1816 | * we wait until bio submission time. All the pages in the bio are | |
1817 | * checksummed and sums are attached onto the ordered extent record. | |
1818 | * | |
1819 | * At IO completion time the cums attached on the ordered extent record | |
1820 | * are inserted into the btree | |
1821 | */ | |
d397712b CM |
1822 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1823 | struct bio *bio, int mirror_num, | |
eaf25d93 CM |
1824 | unsigned long bio_flags, |
1825 | u64 bio_offset) | |
065631f6 | 1826 | { |
065631f6 | 1827 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1828 | int ret = 0; |
e015640f | 1829 | |
d20f7043 | 1830 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
79787eaa | 1831 | BUG_ON(ret); /* -ENOMEM */ |
4a69a410 CM |
1832 | return 0; |
1833 | } | |
e015640f | 1834 | |
4a69a410 CM |
1835 | /* |
1836 | * in order to insert checksums into the metadata in large chunks, | |
1837 | * we wait until bio submission time. All the pages in the bio are | |
1838 | * checksummed and sums are attached onto the ordered extent record. | |
1839 | * | |
1840 | * At IO completion time the cums attached on the ordered extent record | |
1841 | * are inserted into the btree | |
1842 | */ | |
b2950863 | 1843 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1844 | int mirror_num, unsigned long bio_flags, |
1845 | u64 bio_offset) | |
4a69a410 CM |
1846 | { |
1847 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
61891923 SB |
1848 | int ret; |
1849 | ||
1850 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 1); | |
4246a0b6 CH |
1851 | if (ret) { |
1852 | bio->bi_error = ret; | |
1853 | bio_endio(bio); | |
1854 | } | |
61891923 | 1855 | return ret; |
44b8bd7e CM |
1856 | } |
1857 | ||
d352ac68 | 1858 | /* |
cad321ad CM |
1859 | * extent_io.c submission hook. This does the right thing for csum calculation |
1860 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1861 | */ |
b2950863 | 1862 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1863 | int mirror_num, unsigned long bio_flags, |
1864 | u64 bio_offset) | |
44b8bd7e CM |
1865 | { |
1866 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
0d51e28a | 1867 | enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA; |
44b8bd7e | 1868 | int ret = 0; |
19b9bdb0 | 1869 | int skip_sum; |
b812ce28 | 1870 | int async = !atomic_read(&BTRFS_I(inode)->sync_writers); |
44b8bd7e | 1871 | |
6cbff00f | 1872 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1873 | |
83eea1f1 | 1874 | if (btrfs_is_free_space_inode(inode)) |
0d51e28a | 1875 | metadata = BTRFS_WQ_ENDIO_FREE_SPACE; |
0417341e | 1876 | |
7b6d91da | 1877 | if (!(rw & REQ_WRITE)) { |
5fd02043 JB |
1878 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata); |
1879 | if (ret) | |
61891923 | 1880 | goto out; |
5fd02043 | 1881 | |
d20f7043 | 1882 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
61891923 SB |
1883 | ret = btrfs_submit_compressed_read(inode, bio, |
1884 | mirror_num, | |
1885 | bio_flags); | |
1886 | goto out; | |
c2db1073 TI |
1887 | } else if (!skip_sum) { |
1888 | ret = btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
1889 | if (ret) | |
61891923 | 1890 | goto out; |
c2db1073 | 1891 | } |
4d1b5fb4 | 1892 | goto mapit; |
b812ce28 | 1893 | } else if (async && !skip_sum) { |
17d217fe YZ |
1894 | /* csum items have already been cloned */ |
1895 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1896 | goto mapit; | |
19b9bdb0 | 1897 | /* we're doing a write, do the async checksumming */ |
61891923 | 1898 | ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, |
44b8bd7e | 1899 | inode, rw, bio, mirror_num, |
eaf25d93 CM |
1900 | bio_flags, bio_offset, |
1901 | __btrfs_submit_bio_start, | |
4a69a410 | 1902 | __btrfs_submit_bio_done); |
61891923 | 1903 | goto out; |
b812ce28 JB |
1904 | } else if (!skip_sum) { |
1905 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); | |
1906 | if (ret) | |
1907 | goto out; | |
19b9bdb0 CM |
1908 | } |
1909 | ||
0b86a832 | 1910 | mapit: |
61891923 SB |
1911 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 0); |
1912 | ||
1913 | out: | |
4246a0b6 CH |
1914 | if (ret < 0) { |
1915 | bio->bi_error = ret; | |
1916 | bio_endio(bio); | |
1917 | } | |
61891923 | 1918 | return ret; |
065631f6 | 1919 | } |
6885f308 | 1920 | |
d352ac68 CM |
1921 | /* |
1922 | * given a list of ordered sums record them in the inode. This happens | |
1923 | * at IO completion time based on sums calculated at bio submission time. | |
1924 | */ | |
ba1da2f4 | 1925 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1926 | struct inode *inode, u64 file_offset, |
1927 | struct list_head *list) | |
1928 | { | |
e6dcd2dc CM |
1929 | struct btrfs_ordered_sum *sum; |
1930 | ||
c6e30871 | 1931 | list_for_each_entry(sum, list, list) { |
39847c4d | 1932 | trans->adding_csums = 1; |
d20f7043 CM |
1933 | btrfs_csum_file_blocks(trans, |
1934 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
39847c4d | 1935 | trans->adding_csums = 0; |
e6dcd2dc CM |
1936 | } |
1937 | return 0; | |
1938 | } | |
1939 | ||
2ac55d41 JB |
1940 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
1941 | struct extent_state **cached_state) | |
ea8c2819 | 1942 | { |
6c1500f2 | 1943 | WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0); |
ea8c2819 | 1944 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
2ac55d41 | 1945 | cached_state, GFP_NOFS); |
ea8c2819 CM |
1946 | } |
1947 | ||
d352ac68 | 1948 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1949 | struct btrfs_writepage_fixup { |
1950 | struct page *page; | |
1951 | struct btrfs_work work; | |
1952 | }; | |
1953 | ||
b2950863 | 1954 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1955 | { |
1956 | struct btrfs_writepage_fixup *fixup; | |
1957 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 1958 | struct extent_state *cached_state = NULL; |
247e743c CM |
1959 | struct page *page; |
1960 | struct inode *inode; | |
1961 | u64 page_start; | |
1962 | u64 page_end; | |
87826df0 | 1963 | int ret; |
247e743c CM |
1964 | |
1965 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1966 | page = fixup->page; | |
4a096752 | 1967 | again: |
247e743c CM |
1968 | lock_page(page); |
1969 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1970 | ClearPageChecked(page); | |
1971 | goto out_page; | |
1972 | } | |
1973 | ||
1974 | inode = page->mapping->host; | |
1975 | page_start = page_offset(page); | |
1976 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1977 | ||
2ac55d41 | 1978 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0, |
d0082371 | 1979 | &cached_state); |
4a096752 CM |
1980 | |
1981 | /* already ordered? We're done */ | |
8b62b72b | 1982 | if (PagePrivate2(page)) |
247e743c | 1983 | goto out; |
4a096752 CM |
1984 | |
1985 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1986 | if (ordered) { | |
2ac55d41 JB |
1987 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
1988 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
1989 | unlock_page(page); |
1990 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 1991 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
1992 | goto again; |
1993 | } | |
247e743c | 1994 | |
87826df0 JM |
1995 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
1996 | if (ret) { | |
1997 | mapping_set_error(page->mapping, ret); | |
1998 | end_extent_writepage(page, ret, page_start, page_end); | |
1999 | ClearPageChecked(page); | |
2000 | goto out; | |
2001 | } | |
2002 | ||
2ac55d41 | 2003 | btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state); |
247e743c | 2004 | ClearPageChecked(page); |
87826df0 | 2005 | set_page_dirty(page); |
247e743c | 2006 | out: |
2ac55d41 JB |
2007 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
2008 | &cached_state, GFP_NOFS); | |
247e743c CM |
2009 | out_page: |
2010 | unlock_page(page); | |
2011 | page_cache_release(page); | |
b897abec | 2012 | kfree(fixup); |
247e743c CM |
2013 | } |
2014 | ||
2015 | /* | |
2016 | * There are a few paths in the higher layers of the kernel that directly | |
2017 | * set the page dirty bit without asking the filesystem if it is a | |
2018 | * good idea. This causes problems because we want to make sure COW | |
2019 | * properly happens and the data=ordered rules are followed. | |
2020 | * | |
c8b97818 | 2021 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
2022 | * hasn't been properly setup for IO. We kick off an async process |
2023 | * to fix it up. The async helper will wait for ordered extents, set | |
2024 | * the delalloc bit and make it safe to write the page. | |
2025 | */ | |
b2950863 | 2026 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
2027 | { |
2028 | struct inode *inode = page->mapping->host; | |
2029 | struct btrfs_writepage_fixup *fixup; | |
2030 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 2031 | |
8b62b72b CM |
2032 | /* this page is properly in the ordered list */ |
2033 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
2034 | return 0; |
2035 | ||
2036 | if (PageChecked(page)) | |
2037 | return -EAGAIN; | |
2038 | ||
2039 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
2040 | if (!fixup) | |
2041 | return -EAGAIN; | |
f421950f | 2042 | |
247e743c CM |
2043 | SetPageChecked(page); |
2044 | page_cache_get(page); | |
9e0af237 LB |
2045 | btrfs_init_work(&fixup->work, btrfs_fixup_helper, |
2046 | btrfs_writepage_fixup_worker, NULL, NULL); | |
247e743c | 2047 | fixup->page = page; |
dc6e3209 | 2048 | btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work); |
87826df0 | 2049 | return -EBUSY; |
247e743c CM |
2050 | } |
2051 | ||
d899e052 YZ |
2052 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
2053 | struct inode *inode, u64 file_pos, | |
2054 | u64 disk_bytenr, u64 disk_num_bytes, | |
2055 | u64 num_bytes, u64 ram_bytes, | |
2056 | u8 compression, u8 encryption, | |
2057 | u16 other_encoding, int extent_type) | |
2058 | { | |
2059 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2060 | struct btrfs_file_extent_item *fi; | |
2061 | struct btrfs_path *path; | |
2062 | struct extent_buffer *leaf; | |
2063 | struct btrfs_key ins; | |
1acae57b | 2064 | int extent_inserted = 0; |
d899e052 YZ |
2065 | int ret; |
2066 | ||
2067 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
2068 | if (!path) |
2069 | return -ENOMEM; | |
d899e052 | 2070 | |
a1ed835e CM |
2071 | /* |
2072 | * we may be replacing one extent in the tree with another. | |
2073 | * The new extent is pinned in the extent map, and we don't want | |
2074 | * to drop it from the cache until it is completely in the btree. | |
2075 | * | |
2076 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
2077 | * the caller is expected to unpin it and allow it to be merged | |
2078 | * with the others. | |
2079 | */ | |
1acae57b FDBM |
2080 | ret = __btrfs_drop_extents(trans, root, inode, path, file_pos, |
2081 | file_pos + num_bytes, NULL, 0, | |
2082 | 1, sizeof(*fi), &extent_inserted); | |
79787eaa JM |
2083 | if (ret) |
2084 | goto out; | |
d899e052 | 2085 | |
1acae57b FDBM |
2086 | if (!extent_inserted) { |
2087 | ins.objectid = btrfs_ino(inode); | |
2088 | ins.offset = file_pos; | |
2089 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
2090 | ||
2091 | path->leave_spinning = 1; | |
2092 | ret = btrfs_insert_empty_item(trans, root, path, &ins, | |
2093 | sizeof(*fi)); | |
2094 | if (ret) | |
2095 | goto out; | |
2096 | } | |
d899e052 YZ |
2097 | leaf = path->nodes[0]; |
2098 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2099 | struct btrfs_file_extent_item); | |
2100 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
2101 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
2102 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
2103 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
2104 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
2105 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
2106 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
2107 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
2108 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
2109 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 | 2110 | |
d899e052 | 2111 | btrfs_mark_buffer_dirty(leaf); |
ce195332 | 2112 | btrfs_release_path(path); |
d899e052 YZ |
2113 | |
2114 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
2115 | |
2116 | ins.objectid = disk_bytenr; | |
2117 | ins.offset = disk_num_bytes; | |
2118 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
2119 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
2120 | root->root_key.objectid, | |
33345d01 | 2121 | btrfs_ino(inode), file_pos, &ins); |
79787eaa | 2122 | out: |
d899e052 | 2123 | btrfs_free_path(path); |
b9473439 | 2124 | |
79787eaa | 2125 | return ret; |
d899e052 YZ |
2126 | } |
2127 | ||
38c227d8 LB |
2128 | /* snapshot-aware defrag */ |
2129 | struct sa_defrag_extent_backref { | |
2130 | struct rb_node node; | |
2131 | struct old_sa_defrag_extent *old; | |
2132 | u64 root_id; | |
2133 | u64 inum; | |
2134 | u64 file_pos; | |
2135 | u64 extent_offset; | |
2136 | u64 num_bytes; | |
2137 | u64 generation; | |
2138 | }; | |
2139 | ||
2140 | struct old_sa_defrag_extent { | |
2141 | struct list_head list; | |
2142 | struct new_sa_defrag_extent *new; | |
2143 | ||
2144 | u64 extent_offset; | |
2145 | u64 bytenr; | |
2146 | u64 offset; | |
2147 | u64 len; | |
2148 | int count; | |
2149 | }; | |
2150 | ||
2151 | struct new_sa_defrag_extent { | |
2152 | struct rb_root root; | |
2153 | struct list_head head; | |
2154 | struct btrfs_path *path; | |
2155 | struct inode *inode; | |
2156 | u64 file_pos; | |
2157 | u64 len; | |
2158 | u64 bytenr; | |
2159 | u64 disk_len; | |
2160 | u8 compress_type; | |
2161 | }; | |
2162 | ||
2163 | static int backref_comp(struct sa_defrag_extent_backref *b1, | |
2164 | struct sa_defrag_extent_backref *b2) | |
2165 | { | |
2166 | if (b1->root_id < b2->root_id) | |
2167 | return -1; | |
2168 | else if (b1->root_id > b2->root_id) | |
2169 | return 1; | |
2170 | ||
2171 | if (b1->inum < b2->inum) | |
2172 | return -1; | |
2173 | else if (b1->inum > b2->inum) | |
2174 | return 1; | |
2175 | ||
2176 | if (b1->file_pos < b2->file_pos) | |
2177 | return -1; | |
2178 | else if (b1->file_pos > b2->file_pos) | |
2179 | return 1; | |
2180 | ||
2181 | /* | |
2182 | * [------------------------------] ===> (a range of space) | |
2183 | * |<--->| |<---->| =============> (fs/file tree A) | |
2184 | * |<---------------------------->| ===> (fs/file tree B) | |
2185 | * | |
2186 | * A range of space can refer to two file extents in one tree while | |
2187 | * refer to only one file extent in another tree. | |
2188 | * | |
2189 | * So we may process a disk offset more than one time(two extents in A) | |
2190 | * and locate at the same extent(one extent in B), then insert two same | |
2191 | * backrefs(both refer to the extent in B). | |
2192 | */ | |
2193 | return 0; | |
2194 | } | |
2195 | ||
2196 | static void backref_insert(struct rb_root *root, | |
2197 | struct sa_defrag_extent_backref *backref) | |
2198 | { | |
2199 | struct rb_node **p = &root->rb_node; | |
2200 | struct rb_node *parent = NULL; | |
2201 | struct sa_defrag_extent_backref *entry; | |
2202 | int ret; | |
2203 | ||
2204 | while (*p) { | |
2205 | parent = *p; | |
2206 | entry = rb_entry(parent, struct sa_defrag_extent_backref, node); | |
2207 | ||
2208 | ret = backref_comp(backref, entry); | |
2209 | if (ret < 0) | |
2210 | p = &(*p)->rb_left; | |
2211 | else | |
2212 | p = &(*p)->rb_right; | |
2213 | } | |
2214 | ||
2215 | rb_link_node(&backref->node, parent, p); | |
2216 | rb_insert_color(&backref->node, root); | |
2217 | } | |
2218 | ||
2219 | /* | |
2220 | * Note the backref might has changed, and in this case we just return 0. | |
2221 | */ | |
2222 | static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id, | |
2223 | void *ctx) | |
2224 | { | |
2225 | struct btrfs_file_extent_item *extent; | |
2226 | struct btrfs_fs_info *fs_info; | |
2227 | struct old_sa_defrag_extent *old = ctx; | |
2228 | struct new_sa_defrag_extent *new = old->new; | |
2229 | struct btrfs_path *path = new->path; | |
2230 | struct btrfs_key key; | |
2231 | struct btrfs_root *root; | |
2232 | struct sa_defrag_extent_backref *backref; | |
2233 | struct extent_buffer *leaf; | |
2234 | struct inode *inode = new->inode; | |
2235 | int slot; | |
2236 | int ret; | |
2237 | u64 extent_offset; | |
2238 | u64 num_bytes; | |
2239 | ||
2240 | if (BTRFS_I(inode)->root->root_key.objectid == root_id && | |
2241 | inum == btrfs_ino(inode)) | |
2242 | return 0; | |
2243 | ||
2244 | key.objectid = root_id; | |
2245 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2246 | key.offset = (u64)-1; | |
2247 | ||
2248 | fs_info = BTRFS_I(inode)->root->fs_info; | |
2249 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2250 | if (IS_ERR(root)) { | |
2251 | if (PTR_ERR(root) == -ENOENT) | |
2252 | return 0; | |
2253 | WARN_ON(1); | |
2254 | pr_debug("inum=%llu, offset=%llu, root_id=%llu\n", | |
2255 | inum, offset, root_id); | |
2256 | return PTR_ERR(root); | |
2257 | } | |
2258 | ||
2259 | key.objectid = inum; | |
2260 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2261 | if (offset > (u64)-1 << 32) | |
2262 | key.offset = 0; | |
2263 | else | |
2264 | key.offset = offset; | |
2265 | ||
2266 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
fae7f21c | 2267 | if (WARN_ON(ret < 0)) |
38c227d8 | 2268 | return ret; |
50f1319c | 2269 | ret = 0; |
38c227d8 LB |
2270 | |
2271 | while (1) { | |
2272 | cond_resched(); | |
2273 | ||
2274 | leaf = path->nodes[0]; | |
2275 | slot = path->slots[0]; | |
2276 | ||
2277 | if (slot >= btrfs_header_nritems(leaf)) { | |
2278 | ret = btrfs_next_leaf(root, path); | |
2279 | if (ret < 0) { | |
2280 | goto out; | |
2281 | } else if (ret > 0) { | |
2282 | ret = 0; | |
2283 | goto out; | |
2284 | } | |
2285 | continue; | |
2286 | } | |
2287 | ||
2288 | path->slots[0]++; | |
2289 | ||
2290 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
2291 | ||
2292 | if (key.objectid > inum) | |
2293 | goto out; | |
2294 | ||
2295 | if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY) | |
2296 | continue; | |
2297 | ||
2298 | extent = btrfs_item_ptr(leaf, slot, | |
2299 | struct btrfs_file_extent_item); | |
2300 | ||
2301 | if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr) | |
2302 | continue; | |
2303 | ||
e68afa49 LB |
2304 | /* |
2305 | * 'offset' refers to the exact key.offset, | |
2306 | * NOT the 'offset' field in btrfs_extent_data_ref, ie. | |
2307 | * (key.offset - extent_offset). | |
2308 | */ | |
2309 | if (key.offset != offset) | |
38c227d8 LB |
2310 | continue; |
2311 | ||
e68afa49 | 2312 | extent_offset = btrfs_file_extent_offset(leaf, extent); |
38c227d8 | 2313 | num_bytes = btrfs_file_extent_num_bytes(leaf, extent); |
e68afa49 | 2314 | |
38c227d8 LB |
2315 | if (extent_offset >= old->extent_offset + old->offset + |
2316 | old->len || extent_offset + num_bytes <= | |
2317 | old->extent_offset + old->offset) | |
2318 | continue; | |
38c227d8 LB |
2319 | break; |
2320 | } | |
2321 | ||
2322 | backref = kmalloc(sizeof(*backref), GFP_NOFS); | |
2323 | if (!backref) { | |
2324 | ret = -ENOENT; | |
2325 | goto out; | |
2326 | } | |
2327 | ||
2328 | backref->root_id = root_id; | |
2329 | backref->inum = inum; | |
e68afa49 | 2330 | backref->file_pos = offset; |
38c227d8 LB |
2331 | backref->num_bytes = num_bytes; |
2332 | backref->extent_offset = extent_offset; | |
2333 | backref->generation = btrfs_file_extent_generation(leaf, extent); | |
2334 | backref->old = old; | |
2335 | backref_insert(&new->root, backref); | |
2336 | old->count++; | |
2337 | out: | |
2338 | btrfs_release_path(path); | |
2339 | WARN_ON(ret); | |
2340 | return ret; | |
2341 | } | |
2342 | ||
2343 | static noinline bool record_extent_backrefs(struct btrfs_path *path, | |
2344 | struct new_sa_defrag_extent *new) | |
2345 | { | |
2346 | struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info; | |
2347 | struct old_sa_defrag_extent *old, *tmp; | |
2348 | int ret; | |
2349 | ||
2350 | new->path = path; | |
2351 | ||
2352 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
e68afa49 LB |
2353 | ret = iterate_inodes_from_logical(old->bytenr + |
2354 | old->extent_offset, fs_info, | |
38c227d8 LB |
2355 | path, record_one_backref, |
2356 | old); | |
4724b106 JB |
2357 | if (ret < 0 && ret != -ENOENT) |
2358 | return false; | |
38c227d8 LB |
2359 | |
2360 | /* no backref to be processed for this extent */ | |
2361 | if (!old->count) { | |
2362 | list_del(&old->list); | |
2363 | kfree(old); | |
2364 | } | |
2365 | } | |
2366 | ||
2367 | if (list_empty(&new->head)) | |
2368 | return false; | |
2369 | ||
2370 | return true; | |
2371 | } | |
2372 | ||
2373 | static int relink_is_mergable(struct extent_buffer *leaf, | |
2374 | struct btrfs_file_extent_item *fi, | |
116e0024 | 2375 | struct new_sa_defrag_extent *new) |
38c227d8 | 2376 | { |
116e0024 | 2377 | if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr) |
38c227d8 LB |
2378 | return 0; |
2379 | ||
2380 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) | |
2381 | return 0; | |
2382 | ||
116e0024 LB |
2383 | if (btrfs_file_extent_compression(leaf, fi) != new->compress_type) |
2384 | return 0; | |
2385 | ||
2386 | if (btrfs_file_extent_encryption(leaf, fi) || | |
38c227d8 LB |
2387 | btrfs_file_extent_other_encoding(leaf, fi)) |
2388 | return 0; | |
2389 | ||
2390 | return 1; | |
2391 | } | |
2392 | ||
2393 | /* | |
2394 | * Note the backref might has changed, and in this case we just return 0. | |
2395 | */ | |
2396 | static noinline int relink_extent_backref(struct btrfs_path *path, | |
2397 | struct sa_defrag_extent_backref *prev, | |
2398 | struct sa_defrag_extent_backref *backref) | |
2399 | { | |
2400 | struct btrfs_file_extent_item *extent; | |
2401 | struct btrfs_file_extent_item *item; | |
2402 | struct btrfs_ordered_extent *ordered; | |
2403 | struct btrfs_trans_handle *trans; | |
2404 | struct btrfs_fs_info *fs_info; | |
2405 | struct btrfs_root *root; | |
2406 | struct btrfs_key key; | |
2407 | struct extent_buffer *leaf; | |
2408 | struct old_sa_defrag_extent *old = backref->old; | |
2409 | struct new_sa_defrag_extent *new = old->new; | |
2410 | struct inode *src_inode = new->inode; | |
2411 | struct inode *inode; | |
2412 | struct extent_state *cached = NULL; | |
2413 | int ret = 0; | |
2414 | u64 start; | |
2415 | u64 len; | |
2416 | u64 lock_start; | |
2417 | u64 lock_end; | |
2418 | bool merge = false; | |
2419 | int index; | |
2420 | ||
2421 | if (prev && prev->root_id == backref->root_id && | |
2422 | prev->inum == backref->inum && | |
2423 | prev->file_pos + prev->num_bytes == backref->file_pos) | |
2424 | merge = true; | |
2425 | ||
2426 | /* step 1: get root */ | |
2427 | key.objectid = backref->root_id; | |
2428 | key.type = BTRFS_ROOT_ITEM_KEY; | |
2429 | key.offset = (u64)-1; | |
2430 | ||
2431 | fs_info = BTRFS_I(src_inode)->root->fs_info; | |
2432 | index = srcu_read_lock(&fs_info->subvol_srcu); | |
2433 | ||
2434 | root = btrfs_read_fs_root_no_name(fs_info, &key); | |
2435 | if (IS_ERR(root)) { | |
2436 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2437 | if (PTR_ERR(root) == -ENOENT) | |
2438 | return 0; | |
2439 | return PTR_ERR(root); | |
2440 | } | |
38c227d8 | 2441 | |
bcbba5e6 WS |
2442 | if (btrfs_root_readonly(root)) { |
2443 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2444 | return 0; | |
2445 | } | |
2446 | ||
38c227d8 LB |
2447 | /* step 2: get inode */ |
2448 | key.objectid = backref->inum; | |
2449 | key.type = BTRFS_INODE_ITEM_KEY; | |
2450 | key.offset = 0; | |
2451 | ||
2452 | inode = btrfs_iget(fs_info->sb, &key, root, NULL); | |
2453 | if (IS_ERR(inode)) { | |
2454 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2455 | return 0; | |
2456 | } | |
2457 | ||
2458 | srcu_read_unlock(&fs_info->subvol_srcu, index); | |
2459 | ||
2460 | /* step 3: relink backref */ | |
2461 | lock_start = backref->file_pos; | |
2462 | lock_end = backref->file_pos + backref->num_bytes - 1; | |
2463 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
2464 | 0, &cached); | |
2465 | ||
2466 | ordered = btrfs_lookup_first_ordered_extent(inode, lock_end); | |
2467 | if (ordered) { | |
2468 | btrfs_put_ordered_extent(ordered); | |
2469 | goto out_unlock; | |
2470 | } | |
2471 | ||
2472 | trans = btrfs_join_transaction(root); | |
2473 | if (IS_ERR(trans)) { | |
2474 | ret = PTR_ERR(trans); | |
2475 | goto out_unlock; | |
2476 | } | |
2477 | ||
2478 | key.objectid = backref->inum; | |
2479 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2480 | key.offset = backref->file_pos; | |
2481 | ||
2482 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2483 | if (ret < 0) { | |
2484 | goto out_free_path; | |
2485 | } else if (ret > 0) { | |
2486 | ret = 0; | |
2487 | goto out_free_path; | |
2488 | } | |
2489 | ||
2490 | extent = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
2491 | struct btrfs_file_extent_item); | |
2492 | ||
2493 | if (btrfs_file_extent_generation(path->nodes[0], extent) != | |
2494 | backref->generation) | |
2495 | goto out_free_path; | |
2496 | ||
2497 | btrfs_release_path(path); | |
2498 | ||
2499 | start = backref->file_pos; | |
2500 | if (backref->extent_offset < old->extent_offset + old->offset) | |
2501 | start += old->extent_offset + old->offset - | |
2502 | backref->extent_offset; | |
2503 | ||
2504 | len = min(backref->extent_offset + backref->num_bytes, | |
2505 | old->extent_offset + old->offset + old->len); | |
2506 | len -= max(backref->extent_offset, old->extent_offset + old->offset); | |
2507 | ||
2508 | ret = btrfs_drop_extents(trans, root, inode, start, | |
2509 | start + len, 1); | |
2510 | if (ret) | |
2511 | goto out_free_path; | |
2512 | again: | |
2513 | key.objectid = btrfs_ino(inode); | |
2514 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2515 | key.offset = start; | |
2516 | ||
a09a0a70 | 2517 | path->leave_spinning = 1; |
38c227d8 LB |
2518 | if (merge) { |
2519 | struct btrfs_file_extent_item *fi; | |
2520 | u64 extent_len; | |
2521 | struct btrfs_key found_key; | |
2522 | ||
3c9665df | 2523 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
38c227d8 LB |
2524 | if (ret < 0) |
2525 | goto out_free_path; | |
2526 | ||
2527 | path->slots[0]--; | |
2528 | leaf = path->nodes[0]; | |
2529 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2530 | ||
2531 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
2532 | struct btrfs_file_extent_item); | |
2533 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
2534 | ||
116e0024 LB |
2535 | if (extent_len + found_key.offset == start && |
2536 | relink_is_mergable(leaf, fi, new)) { | |
38c227d8 LB |
2537 | btrfs_set_file_extent_num_bytes(leaf, fi, |
2538 | extent_len + len); | |
2539 | btrfs_mark_buffer_dirty(leaf); | |
2540 | inode_add_bytes(inode, len); | |
2541 | ||
2542 | ret = 1; | |
2543 | goto out_free_path; | |
2544 | } else { | |
2545 | merge = false; | |
2546 | btrfs_release_path(path); | |
2547 | goto again; | |
2548 | } | |
2549 | } | |
2550 | ||
2551 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2552 | sizeof(*extent)); | |
2553 | if (ret) { | |
2554 | btrfs_abort_transaction(trans, root, ret); | |
2555 | goto out_free_path; | |
2556 | } | |
2557 | ||
2558 | leaf = path->nodes[0]; | |
2559 | item = btrfs_item_ptr(leaf, path->slots[0], | |
2560 | struct btrfs_file_extent_item); | |
2561 | btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr); | |
2562 | btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len); | |
2563 | btrfs_set_file_extent_offset(leaf, item, start - new->file_pos); | |
2564 | btrfs_set_file_extent_num_bytes(leaf, item, len); | |
2565 | btrfs_set_file_extent_ram_bytes(leaf, item, new->len); | |
2566 | btrfs_set_file_extent_generation(leaf, item, trans->transid); | |
2567 | btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG); | |
2568 | btrfs_set_file_extent_compression(leaf, item, new->compress_type); | |
2569 | btrfs_set_file_extent_encryption(leaf, item, 0); | |
2570 | btrfs_set_file_extent_other_encoding(leaf, item, 0); | |
2571 | ||
2572 | btrfs_mark_buffer_dirty(leaf); | |
2573 | inode_add_bytes(inode, len); | |
a09a0a70 | 2574 | btrfs_release_path(path); |
38c227d8 LB |
2575 | |
2576 | ret = btrfs_inc_extent_ref(trans, root, new->bytenr, | |
2577 | new->disk_len, 0, | |
2578 | backref->root_id, backref->inum, | |
2579 | new->file_pos, 0); /* start - extent_offset */ | |
2580 | if (ret) { | |
2581 | btrfs_abort_transaction(trans, root, ret); | |
2582 | goto out_free_path; | |
2583 | } | |
2584 | ||
2585 | ret = 1; | |
2586 | out_free_path: | |
2587 | btrfs_release_path(path); | |
a09a0a70 | 2588 | path->leave_spinning = 0; |
38c227d8 LB |
2589 | btrfs_end_transaction(trans, root); |
2590 | out_unlock: | |
2591 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end, | |
2592 | &cached, GFP_NOFS); | |
2593 | iput(inode); | |
2594 | return ret; | |
2595 | } | |
2596 | ||
6f519564 LB |
2597 | static void free_sa_defrag_extent(struct new_sa_defrag_extent *new) |
2598 | { | |
2599 | struct old_sa_defrag_extent *old, *tmp; | |
2600 | ||
2601 | if (!new) | |
2602 | return; | |
2603 | ||
2604 | list_for_each_entry_safe(old, tmp, &new->head, list) { | |
6f519564 LB |
2605 | kfree(old); |
2606 | } | |
2607 | kfree(new); | |
2608 | } | |
2609 | ||
38c227d8 LB |
2610 | static void relink_file_extents(struct new_sa_defrag_extent *new) |
2611 | { | |
2612 | struct btrfs_path *path; | |
38c227d8 LB |
2613 | struct sa_defrag_extent_backref *backref; |
2614 | struct sa_defrag_extent_backref *prev = NULL; | |
2615 | struct inode *inode; | |
2616 | struct btrfs_root *root; | |
2617 | struct rb_node *node; | |
2618 | int ret; | |
2619 | ||
2620 | inode = new->inode; | |
2621 | root = BTRFS_I(inode)->root; | |
2622 | ||
2623 | path = btrfs_alloc_path(); | |
2624 | if (!path) | |
2625 | return; | |
2626 | ||
2627 | if (!record_extent_backrefs(path, new)) { | |
2628 | btrfs_free_path(path); | |
2629 | goto out; | |
2630 | } | |
2631 | btrfs_release_path(path); | |
2632 | ||
2633 | while (1) { | |
2634 | node = rb_first(&new->root); | |
2635 | if (!node) | |
2636 | break; | |
2637 | rb_erase(node, &new->root); | |
2638 | ||
2639 | backref = rb_entry(node, struct sa_defrag_extent_backref, node); | |
2640 | ||
2641 | ret = relink_extent_backref(path, prev, backref); | |
2642 | WARN_ON(ret < 0); | |
2643 | ||
2644 | kfree(prev); | |
2645 | ||
2646 | if (ret == 1) | |
2647 | prev = backref; | |
2648 | else | |
2649 | prev = NULL; | |
2650 | cond_resched(); | |
2651 | } | |
2652 | kfree(prev); | |
2653 | ||
2654 | btrfs_free_path(path); | |
38c227d8 | 2655 | out: |
6f519564 LB |
2656 | free_sa_defrag_extent(new); |
2657 | ||
38c227d8 LB |
2658 | atomic_dec(&root->fs_info->defrag_running); |
2659 | wake_up(&root->fs_info->transaction_wait); | |
38c227d8 LB |
2660 | } |
2661 | ||
2662 | static struct new_sa_defrag_extent * | |
2663 | record_old_file_extents(struct inode *inode, | |
2664 | struct btrfs_ordered_extent *ordered) | |
2665 | { | |
2666 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2667 | struct btrfs_path *path; | |
2668 | struct btrfs_key key; | |
6f519564 | 2669 | struct old_sa_defrag_extent *old; |
38c227d8 LB |
2670 | struct new_sa_defrag_extent *new; |
2671 | int ret; | |
2672 | ||
2673 | new = kmalloc(sizeof(*new), GFP_NOFS); | |
2674 | if (!new) | |
2675 | return NULL; | |
2676 | ||
2677 | new->inode = inode; | |
2678 | new->file_pos = ordered->file_offset; | |
2679 | new->len = ordered->len; | |
2680 | new->bytenr = ordered->start; | |
2681 | new->disk_len = ordered->disk_len; | |
2682 | new->compress_type = ordered->compress_type; | |
2683 | new->root = RB_ROOT; | |
2684 | INIT_LIST_HEAD(&new->head); | |
2685 | ||
2686 | path = btrfs_alloc_path(); | |
2687 | if (!path) | |
2688 | goto out_kfree; | |
2689 | ||
2690 | key.objectid = btrfs_ino(inode); | |
2691 | key.type = BTRFS_EXTENT_DATA_KEY; | |
2692 | key.offset = new->file_pos; | |
2693 | ||
2694 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2695 | if (ret < 0) | |
2696 | goto out_free_path; | |
2697 | if (ret > 0 && path->slots[0] > 0) | |
2698 | path->slots[0]--; | |
2699 | ||
2700 | /* find out all the old extents for the file range */ | |
2701 | while (1) { | |
2702 | struct btrfs_file_extent_item *extent; | |
2703 | struct extent_buffer *l; | |
2704 | int slot; | |
2705 | u64 num_bytes; | |
2706 | u64 offset; | |
2707 | u64 end; | |
2708 | u64 disk_bytenr; | |
2709 | u64 extent_offset; | |
2710 | ||
2711 | l = path->nodes[0]; | |
2712 | slot = path->slots[0]; | |
2713 | ||
2714 | if (slot >= btrfs_header_nritems(l)) { | |
2715 | ret = btrfs_next_leaf(root, path); | |
2716 | if (ret < 0) | |
6f519564 | 2717 | goto out_free_path; |
38c227d8 LB |
2718 | else if (ret > 0) |
2719 | break; | |
2720 | continue; | |
2721 | } | |
2722 | ||
2723 | btrfs_item_key_to_cpu(l, &key, slot); | |
2724 | ||
2725 | if (key.objectid != btrfs_ino(inode)) | |
2726 | break; | |
2727 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2728 | break; | |
2729 | if (key.offset >= new->file_pos + new->len) | |
2730 | break; | |
2731 | ||
2732 | extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item); | |
2733 | ||
2734 | num_bytes = btrfs_file_extent_num_bytes(l, extent); | |
2735 | if (key.offset + num_bytes < new->file_pos) | |
2736 | goto next; | |
2737 | ||
2738 | disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent); | |
2739 | if (!disk_bytenr) | |
2740 | goto next; | |
2741 | ||
2742 | extent_offset = btrfs_file_extent_offset(l, extent); | |
2743 | ||
2744 | old = kmalloc(sizeof(*old), GFP_NOFS); | |
2745 | if (!old) | |
6f519564 | 2746 | goto out_free_path; |
38c227d8 LB |
2747 | |
2748 | offset = max(new->file_pos, key.offset); | |
2749 | end = min(new->file_pos + new->len, key.offset + num_bytes); | |
2750 | ||
2751 | old->bytenr = disk_bytenr; | |
2752 | old->extent_offset = extent_offset; | |
2753 | old->offset = offset - key.offset; | |
2754 | old->len = end - offset; | |
2755 | old->new = new; | |
2756 | old->count = 0; | |
2757 | list_add_tail(&old->list, &new->head); | |
2758 | next: | |
2759 | path->slots[0]++; | |
2760 | cond_resched(); | |
2761 | } | |
2762 | ||
2763 | btrfs_free_path(path); | |
2764 | atomic_inc(&root->fs_info->defrag_running); | |
2765 | ||
2766 | return new; | |
2767 | ||
38c227d8 LB |
2768 | out_free_path: |
2769 | btrfs_free_path(path); | |
2770 | out_kfree: | |
6f519564 | 2771 | free_sa_defrag_extent(new); |
38c227d8 LB |
2772 | return NULL; |
2773 | } | |
2774 | ||
e570fd27 MX |
2775 | static void btrfs_release_delalloc_bytes(struct btrfs_root *root, |
2776 | u64 start, u64 len) | |
2777 | { | |
2778 | struct btrfs_block_group_cache *cache; | |
2779 | ||
2780 | cache = btrfs_lookup_block_group(root->fs_info, start); | |
2781 | ASSERT(cache); | |
2782 | ||
2783 | spin_lock(&cache->lock); | |
2784 | cache->delalloc_bytes -= len; | |
2785 | spin_unlock(&cache->lock); | |
2786 | ||
2787 | btrfs_put_block_group(cache); | |
2788 | } | |
2789 | ||
d352ac68 CM |
2790 | /* as ordered data IO finishes, this gets called so we can finish |
2791 | * an ordered extent if the range of bytes in the file it covers are | |
2792 | * fully written. | |
2793 | */ | |
5fd02043 | 2794 | static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent) |
e6dcd2dc | 2795 | { |
5fd02043 | 2796 | struct inode *inode = ordered_extent->inode; |
e6dcd2dc | 2797 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 2798 | struct btrfs_trans_handle *trans = NULL; |
e6dcd2dc | 2799 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 2800 | struct extent_state *cached_state = NULL; |
38c227d8 | 2801 | struct new_sa_defrag_extent *new = NULL; |
261507a0 | 2802 | int compress_type = 0; |
77cef2ec JB |
2803 | int ret = 0; |
2804 | u64 logical_len = ordered_extent->len; | |
82d5902d | 2805 | bool nolock; |
77cef2ec | 2806 | bool truncated = false; |
e6dcd2dc | 2807 | |
83eea1f1 | 2808 | nolock = btrfs_is_free_space_inode(inode); |
0cb59c99 | 2809 | |
5fd02043 JB |
2810 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) { |
2811 | ret = -EIO; | |
2812 | goto out; | |
2813 | } | |
2814 | ||
f612496b MX |
2815 | btrfs_free_io_failure_record(inode, ordered_extent->file_offset, |
2816 | ordered_extent->file_offset + | |
2817 | ordered_extent->len - 1); | |
2818 | ||
77cef2ec JB |
2819 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) { |
2820 | truncated = true; | |
2821 | logical_len = ordered_extent->truncated_len; | |
2822 | /* Truncated the entire extent, don't bother adding */ | |
2823 | if (!logical_len) | |
2824 | goto out; | |
2825 | } | |
2826 | ||
c2167754 | 2827 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
79787eaa | 2828 | BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */ |
6c760c07 JB |
2829 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2830 | if (nolock) | |
2831 | trans = btrfs_join_transaction_nolock(root); | |
2832 | else | |
2833 | trans = btrfs_join_transaction(root); | |
2834 | if (IS_ERR(trans)) { | |
2835 | ret = PTR_ERR(trans); | |
2836 | trans = NULL; | |
2837 | goto out; | |
c2167754 | 2838 | } |
6c760c07 JB |
2839 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
2840 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
2841 | if (ret) /* -ENOMEM or corruption */ | |
2842 | btrfs_abort_transaction(trans, root, ret); | |
c2167754 YZ |
2843 | goto out; |
2844 | } | |
e6dcd2dc | 2845 | |
2ac55d41 JB |
2846 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
2847 | ordered_extent->file_offset + ordered_extent->len - 1, | |
d0082371 | 2848 | 0, &cached_state); |
e6dcd2dc | 2849 | |
38c227d8 LB |
2850 | ret = test_range_bit(io_tree, ordered_extent->file_offset, |
2851 | ordered_extent->file_offset + ordered_extent->len - 1, | |
2852 | EXTENT_DEFRAG, 1, cached_state); | |
2853 | if (ret) { | |
2854 | u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item); | |
8101c8db | 2855 | if (0 && last_snapshot >= BTRFS_I(inode)->generation) |
38c227d8 LB |
2856 | /* the inode is shared */ |
2857 | new = record_old_file_extents(inode, ordered_extent); | |
2858 | ||
2859 | clear_extent_bit(io_tree, ordered_extent->file_offset, | |
2860 | ordered_extent->file_offset + ordered_extent->len - 1, | |
2861 | EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS); | |
2862 | } | |
2863 | ||
0cb59c99 | 2864 | if (nolock) |
7a7eaa40 | 2865 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 2866 | else |
7a7eaa40 | 2867 | trans = btrfs_join_transaction(root); |
79787eaa JM |
2868 | if (IS_ERR(trans)) { |
2869 | ret = PTR_ERR(trans); | |
2870 | trans = NULL; | |
2871 | goto out_unlock; | |
2872 | } | |
a79b7d4b | 2873 | |
0ca1f7ce | 2874 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
c2167754 | 2875 | |
c8b97818 | 2876 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 2877 | compress_type = ordered_extent->compress_type; |
d899e052 | 2878 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 2879 | BUG_ON(compress_type); |
920bbbfb | 2880 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
2881 | ordered_extent->file_offset, |
2882 | ordered_extent->file_offset + | |
77cef2ec | 2883 | logical_len); |
d899e052 | 2884 | } else { |
0af3d00b | 2885 | BUG_ON(root == root->fs_info->tree_root); |
d899e052 YZ |
2886 | ret = insert_reserved_file_extent(trans, inode, |
2887 | ordered_extent->file_offset, | |
2888 | ordered_extent->start, | |
2889 | ordered_extent->disk_len, | |
77cef2ec | 2890 | logical_len, logical_len, |
261507a0 | 2891 | compress_type, 0, 0, |
d899e052 | 2892 | BTRFS_FILE_EXTENT_REG); |
e570fd27 MX |
2893 | if (!ret) |
2894 | btrfs_release_delalloc_bytes(root, | |
2895 | ordered_extent->start, | |
2896 | ordered_extent->disk_len); | |
d899e052 | 2897 | } |
5dc562c5 JB |
2898 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
2899 | ordered_extent->file_offset, ordered_extent->len, | |
2900 | trans->transid); | |
79787eaa JM |
2901 | if (ret < 0) { |
2902 | btrfs_abort_transaction(trans, root, ret); | |
5fd02043 | 2903 | goto out_unlock; |
79787eaa | 2904 | } |
2ac55d41 | 2905 | |
e6dcd2dc CM |
2906 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
2907 | &ordered_extent->list); | |
2908 | ||
6c760c07 JB |
2909 | btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
2910 | ret = btrfs_update_inode_fallback(trans, root, inode); | |
2911 | if (ret) { /* -ENOMEM or corruption */ | |
2912 | btrfs_abort_transaction(trans, root, ret); | |
2913 | goto out_unlock; | |
1ef30be1 JB |
2914 | } |
2915 | ret = 0; | |
5fd02043 JB |
2916 | out_unlock: |
2917 | unlock_extent_cached(io_tree, ordered_extent->file_offset, | |
2918 | ordered_extent->file_offset + | |
2919 | ordered_extent->len - 1, &cached_state, GFP_NOFS); | |
c2167754 | 2920 | out: |
5b0e95bf | 2921 | if (root != root->fs_info->tree_root) |
0cb59c99 | 2922 | btrfs_delalloc_release_metadata(inode, ordered_extent->len); |
a698d075 MX |
2923 | if (trans) |
2924 | btrfs_end_transaction(trans, root); | |
0cb59c99 | 2925 | |
77cef2ec JB |
2926 | if (ret || truncated) { |
2927 | u64 start, end; | |
2928 | ||
2929 | if (truncated) | |
2930 | start = ordered_extent->file_offset + logical_len; | |
2931 | else | |
2932 | start = ordered_extent->file_offset; | |
2933 | end = ordered_extent->file_offset + ordered_extent->len - 1; | |
2934 | clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS); | |
2935 | ||
2936 | /* Drop the cache for the part of the extent we didn't write. */ | |
2937 | btrfs_drop_extent_cache(inode, start, end, 0); | |
5fd02043 | 2938 | |
0bec9ef5 JB |
2939 | /* |
2940 | * If the ordered extent had an IOERR or something else went | |
2941 | * wrong we need to return the space for this ordered extent | |
77cef2ec JB |
2942 | * back to the allocator. We only free the extent in the |
2943 | * truncated case if we didn't write out the extent at all. | |
0bec9ef5 | 2944 | */ |
77cef2ec JB |
2945 | if ((ret || !logical_len) && |
2946 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) && | |
0bec9ef5 JB |
2947 | !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) |
2948 | btrfs_free_reserved_extent(root, ordered_extent->start, | |
e570fd27 | 2949 | ordered_extent->disk_len, 1); |
0bec9ef5 JB |
2950 | } |
2951 | ||
2952 | ||
5fd02043 | 2953 | /* |
8bad3c02 LB |
2954 | * This needs to be done to make sure anybody waiting knows we are done |
2955 | * updating everything for this ordered extent. | |
5fd02043 JB |
2956 | */ |
2957 | btrfs_remove_ordered_extent(inode, ordered_extent); | |
2958 | ||
38c227d8 | 2959 | /* for snapshot-aware defrag */ |
6f519564 LB |
2960 | if (new) { |
2961 | if (ret) { | |
2962 | free_sa_defrag_extent(new); | |
2963 | atomic_dec(&root->fs_info->defrag_running); | |
2964 | } else { | |
2965 | relink_file_extents(new); | |
2966 | } | |
2967 | } | |
38c227d8 | 2968 | |
e6dcd2dc CM |
2969 | /* once for us */ |
2970 | btrfs_put_ordered_extent(ordered_extent); | |
2971 | /* once for the tree */ | |
2972 | btrfs_put_ordered_extent(ordered_extent); | |
2973 | ||
5fd02043 JB |
2974 | return ret; |
2975 | } | |
2976 | ||
2977 | static void finish_ordered_fn(struct btrfs_work *work) | |
2978 | { | |
2979 | struct btrfs_ordered_extent *ordered_extent; | |
2980 | ordered_extent = container_of(work, struct btrfs_ordered_extent, work); | |
2981 | btrfs_finish_ordered_io(ordered_extent); | |
e6dcd2dc CM |
2982 | } |
2983 | ||
b2950863 | 2984 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
2985 | struct extent_state *state, int uptodate) |
2986 | { | |
5fd02043 JB |
2987 | struct inode *inode = page->mapping->host; |
2988 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
2989 | struct btrfs_ordered_extent *ordered_extent = NULL; | |
9e0af237 LB |
2990 | struct btrfs_workqueue *wq; |
2991 | btrfs_work_func_t func; | |
5fd02043 | 2992 | |
1abe9b8a | 2993 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
2994 | ||
8b62b72b | 2995 | ClearPagePrivate2(page); |
5fd02043 JB |
2996 | if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
2997 | end - start + 1, uptodate)) | |
2998 | return 0; | |
2999 | ||
9e0af237 LB |
3000 | if (btrfs_is_free_space_inode(inode)) { |
3001 | wq = root->fs_info->endio_freespace_worker; | |
3002 | func = btrfs_freespace_write_helper; | |
3003 | } else { | |
3004 | wq = root->fs_info->endio_write_workers; | |
3005 | func = btrfs_endio_write_helper; | |
3006 | } | |
5fd02043 | 3007 | |
9e0af237 LB |
3008 | btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL, |
3009 | NULL); | |
3010 | btrfs_queue_work(wq, &ordered_extent->work); | |
5fd02043 JB |
3011 | |
3012 | return 0; | |
211f90e6 CM |
3013 | } |
3014 | ||
dc380aea MX |
3015 | static int __readpage_endio_check(struct inode *inode, |
3016 | struct btrfs_io_bio *io_bio, | |
3017 | int icsum, struct page *page, | |
3018 | int pgoff, u64 start, size_t len) | |
3019 | { | |
3020 | char *kaddr; | |
3021 | u32 csum_expected; | |
3022 | u32 csum = ~(u32)0; | |
dc380aea MX |
3023 | |
3024 | csum_expected = *(((u32 *)io_bio->csum) + icsum); | |
3025 | ||
3026 | kaddr = kmap_atomic(page); | |
3027 | csum = btrfs_csum_data(kaddr + pgoff, csum, len); | |
3028 | btrfs_csum_final(csum, (char *)&csum); | |
3029 | if (csum != csum_expected) | |
3030 | goto zeroit; | |
3031 | ||
3032 | kunmap_atomic(kaddr); | |
3033 | return 0; | |
3034 | zeroit: | |
94647322 DS |
3035 | btrfs_warn_rl(BTRFS_I(inode)->root->fs_info, |
3036 | "csum failed ino %llu off %llu csum %u expected csum %u", | |
dc380aea MX |
3037 | btrfs_ino(inode), start, csum, csum_expected); |
3038 | memset(kaddr + pgoff, 1, len); | |
3039 | flush_dcache_page(page); | |
3040 | kunmap_atomic(kaddr); | |
3041 | if (csum_expected == 0) | |
3042 | return 0; | |
3043 | return -EIO; | |
3044 | } | |
3045 | ||
d352ac68 CM |
3046 | /* |
3047 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
3048 | * if there's a match, we allow the bio to finish. If not, the code in |
3049 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 3050 | */ |
facc8a22 MX |
3051 | static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio, |
3052 | u64 phy_offset, struct page *page, | |
3053 | u64 start, u64 end, int mirror) | |
07157aac | 3054 | { |
4eee4fa4 | 3055 | size_t offset = start - page_offset(page); |
07157aac | 3056 | struct inode *inode = page->mapping->host; |
d1310b2e | 3057 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
ff79f819 | 3058 | struct btrfs_root *root = BTRFS_I(inode)->root; |
d1310b2e | 3059 | |
d20f7043 CM |
3060 | if (PageChecked(page)) { |
3061 | ClearPageChecked(page); | |
dc380aea | 3062 | return 0; |
d20f7043 | 3063 | } |
6cbff00f CH |
3064 | |
3065 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
dc380aea | 3066 | return 0; |
17d217fe YZ |
3067 | |
3068 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 3069 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
3070 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
3071 | GFP_NOFS); | |
b6cda9bc | 3072 | return 0; |
17d217fe | 3073 | } |
d20f7043 | 3074 | |
facc8a22 | 3075 | phy_offset >>= inode->i_sb->s_blocksize_bits; |
dc380aea MX |
3076 | return __readpage_endio_check(inode, io_bio, phy_offset, page, offset, |
3077 | start, (size_t)(end - start + 1)); | |
07157aac | 3078 | } |
b888db2b | 3079 | |
24bbcf04 YZ |
3080 | struct delayed_iput { |
3081 | struct list_head list; | |
3082 | struct inode *inode; | |
3083 | }; | |
3084 | ||
79787eaa JM |
3085 | /* JDM: If this is fs-wide, why can't we add a pointer to |
3086 | * btrfs_inode instead and avoid the allocation? */ | |
24bbcf04 YZ |
3087 | void btrfs_add_delayed_iput(struct inode *inode) |
3088 | { | |
3089 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
3090 | struct delayed_iput *delayed; | |
3091 | ||
3092 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
3093 | return; | |
3094 | ||
3095 | delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL); | |
3096 | delayed->inode = inode; | |
3097 | ||
3098 | spin_lock(&fs_info->delayed_iput_lock); | |
3099 | list_add_tail(&delayed->list, &fs_info->delayed_iputs); | |
3100 | spin_unlock(&fs_info->delayed_iput_lock); | |
3101 | } | |
3102 | ||
3103 | void btrfs_run_delayed_iputs(struct btrfs_root *root) | |
3104 | { | |
3105 | LIST_HEAD(list); | |
3106 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3107 | struct delayed_iput *delayed; | |
3108 | int empty; | |
3109 | ||
3110 | spin_lock(&fs_info->delayed_iput_lock); | |
3111 | empty = list_empty(&fs_info->delayed_iputs); | |
3112 | spin_unlock(&fs_info->delayed_iput_lock); | |
3113 | if (empty) | |
3114 | return; | |
3115 | ||
d7c15171 ZL |
3116 | down_read(&fs_info->delayed_iput_sem); |
3117 | ||
24bbcf04 YZ |
3118 | spin_lock(&fs_info->delayed_iput_lock); |
3119 | list_splice_init(&fs_info->delayed_iputs, &list); | |
3120 | spin_unlock(&fs_info->delayed_iput_lock); | |
3121 | ||
3122 | while (!list_empty(&list)) { | |
3123 | delayed = list_entry(list.next, struct delayed_iput, list); | |
3124 | list_del(&delayed->list); | |
3125 | iput(delayed->inode); | |
3126 | kfree(delayed); | |
3127 | } | |
d7c15171 ZL |
3128 | |
3129 | up_read(&root->fs_info->delayed_iput_sem); | |
24bbcf04 YZ |
3130 | } |
3131 | ||
d68fc57b | 3132 | /* |
42b2aa86 | 3133 | * This is called in transaction commit time. If there are no orphan |
d68fc57b YZ |
3134 | * files in the subvolume, it removes orphan item and frees block_rsv |
3135 | * structure. | |
3136 | */ | |
3137 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
3138 | struct btrfs_root *root) | |
3139 | { | |
90290e19 | 3140 | struct btrfs_block_rsv *block_rsv; |
d68fc57b YZ |
3141 | int ret; |
3142 | ||
8a35d95f | 3143 | if (atomic_read(&root->orphan_inodes) || |
d68fc57b YZ |
3144 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) |
3145 | return; | |
3146 | ||
90290e19 | 3147 | spin_lock(&root->orphan_lock); |
8a35d95f | 3148 | if (atomic_read(&root->orphan_inodes)) { |
90290e19 JB |
3149 | spin_unlock(&root->orphan_lock); |
3150 | return; | |
3151 | } | |
3152 | ||
3153 | if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) { | |
3154 | spin_unlock(&root->orphan_lock); | |
3155 | return; | |
3156 | } | |
3157 | ||
3158 | block_rsv = root->orphan_block_rsv; | |
3159 | root->orphan_block_rsv = NULL; | |
3160 | spin_unlock(&root->orphan_lock); | |
3161 | ||
27cdeb70 | 3162 | if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) && |
d68fc57b YZ |
3163 | btrfs_root_refs(&root->root_item) > 0) { |
3164 | ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root, | |
3165 | root->root_key.objectid); | |
4ef31a45 JB |
3166 | if (ret) |
3167 | btrfs_abort_transaction(trans, root, ret); | |
3168 | else | |
27cdeb70 MX |
3169 | clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, |
3170 | &root->state); | |
d68fc57b YZ |
3171 | } |
3172 | ||
90290e19 JB |
3173 | if (block_rsv) { |
3174 | WARN_ON(block_rsv->size > 0); | |
3175 | btrfs_free_block_rsv(root, block_rsv); | |
d68fc57b YZ |
3176 | } |
3177 | } | |
3178 | ||
7b128766 JB |
3179 | /* |
3180 | * This creates an orphan entry for the given inode in case something goes | |
3181 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
3182 | * |
3183 | * NOTE: caller of this function should reserve 5 units of metadata for | |
3184 | * this function. | |
7b128766 JB |
3185 | */ |
3186 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
3187 | { | |
3188 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
3189 | struct btrfs_block_rsv *block_rsv = NULL; |
3190 | int reserve = 0; | |
3191 | int insert = 0; | |
3192 | int ret; | |
7b128766 | 3193 | |
d68fc57b | 3194 | if (!root->orphan_block_rsv) { |
66d8f3dd | 3195 | block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
b532402e TI |
3196 | if (!block_rsv) |
3197 | return -ENOMEM; | |
d68fc57b | 3198 | } |
7b128766 | 3199 | |
d68fc57b YZ |
3200 | spin_lock(&root->orphan_lock); |
3201 | if (!root->orphan_block_rsv) { | |
3202 | root->orphan_block_rsv = block_rsv; | |
3203 | } else if (block_rsv) { | |
3204 | btrfs_free_block_rsv(root, block_rsv); | |
3205 | block_rsv = NULL; | |
7b128766 | 3206 | } |
7b128766 | 3207 | |
8a35d95f JB |
3208 | if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3209 | &BTRFS_I(inode)->runtime_flags)) { | |
d68fc57b YZ |
3210 | #if 0 |
3211 | /* | |
3212 | * For proper ENOSPC handling, we should do orphan | |
3213 | * cleanup when mounting. But this introduces backward | |
3214 | * compatibility issue. | |
3215 | */ | |
3216 | if (!xchg(&root->orphan_item_inserted, 1)) | |
3217 | insert = 2; | |
3218 | else | |
3219 | insert = 1; | |
3220 | #endif | |
3221 | insert = 1; | |
321f0e70 | 3222 | atomic_inc(&root->orphan_inodes); |
7b128766 JB |
3223 | } |
3224 | ||
72ac3c0d JB |
3225 | if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
3226 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3227 | reserve = 1; |
d68fc57b | 3228 | spin_unlock(&root->orphan_lock); |
7b128766 | 3229 | |
d68fc57b YZ |
3230 | /* grab metadata reservation from transaction handle */ |
3231 | if (reserve) { | |
3232 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
79787eaa | 3233 | BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */ |
d68fc57b | 3234 | } |
7b128766 | 3235 | |
d68fc57b YZ |
3236 | /* insert an orphan item to track this unlinked/truncated file */ |
3237 | if (insert >= 1) { | |
33345d01 | 3238 | ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode)); |
4ef31a45 | 3239 | if (ret) { |
703c88e0 | 3240 | atomic_dec(&root->orphan_inodes); |
4ef31a45 JB |
3241 | if (reserve) { |
3242 | clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, | |
3243 | &BTRFS_I(inode)->runtime_flags); | |
3244 | btrfs_orphan_release_metadata(inode); | |
3245 | } | |
3246 | if (ret != -EEXIST) { | |
e8e7cff6 JB |
3247 | clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3248 | &BTRFS_I(inode)->runtime_flags); | |
4ef31a45 JB |
3249 | btrfs_abort_transaction(trans, root, ret); |
3250 | return ret; | |
3251 | } | |
79787eaa JM |
3252 | } |
3253 | ret = 0; | |
d68fc57b YZ |
3254 | } |
3255 | ||
3256 | /* insert an orphan item to track subvolume contains orphan files */ | |
3257 | if (insert >= 2) { | |
3258 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
3259 | root->root_key.objectid); | |
79787eaa JM |
3260 | if (ret && ret != -EEXIST) { |
3261 | btrfs_abort_transaction(trans, root, ret); | |
3262 | return ret; | |
3263 | } | |
d68fc57b YZ |
3264 | } |
3265 | return 0; | |
7b128766 JB |
3266 | } |
3267 | ||
3268 | /* | |
3269 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
3270 | * item for this particular inode. | |
3271 | */ | |
48a3b636 ES |
3272 | static int btrfs_orphan_del(struct btrfs_trans_handle *trans, |
3273 | struct inode *inode) | |
7b128766 JB |
3274 | { |
3275 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
3276 | int delete_item = 0; |
3277 | int release_rsv = 0; | |
7b128766 JB |
3278 | int ret = 0; |
3279 | ||
d68fc57b | 3280 | spin_lock(&root->orphan_lock); |
8a35d95f JB |
3281 | if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3282 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3283 | delete_item = 1; |
7b128766 | 3284 | |
72ac3c0d JB |
3285 | if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED, |
3286 | &BTRFS_I(inode)->runtime_flags)) | |
d68fc57b | 3287 | release_rsv = 1; |
d68fc57b | 3288 | spin_unlock(&root->orphan_lock); |
7b128766 | 3289 | |
703c88e0 | 3290 | if (delete_item) { |
8a35d95f | 3291 | atomic_dec(&root->orphan_inodes); |
703c88e0 FDBM |
3292 | if (trans) |
3293 | ret = btrfs_del_orphan_item(trans, root, | |
3294 | btrfs_ino(inode)); | |
8a35d95f | 3295 | } |
7b128766 | 3296 | |
703c88e0 FDBM |
3297 | if (release_rsv) |
3298 | btrfs_orphan_release_metadata(inode); | |
3299 | ||
4ef31a45 | 3300 | return ret; |
7b128766 JB |
3301 | } |
3302 | ||
3303 | /* | |
3304 | * this cleans up any orphans that may be left on the list from the last use | |
3305 | * of this root. | |
3306 | */ | |
66b4ffd1 | 3307 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 JB |
3308 | { |
3309 | struct btrfs_path *path; | |
3310 | struct extent_buffer *leaf; | |
7b128766 JB |
3311 | struct btrfs_key key, found_key; |
3312 | struct btrfs_trans_handle *trans; | |
3313 | struct inode *inode; | |
8f6d7f4f | 3314 | u64 last_objectid = 0; |
7b128766 JB |
3315 | int ret = 0, nr_unlink = 0, nr_truncate = 0; |
3316 | ||
d68fc57b | 3317 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 3318 | return 0; |
c71bf099 YZ |
3319 | |
3320 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
3321 | if (!path) { |
3322 | ret = -ENOMEM; | |
3323 | goto out; | |
3324 | } | |
7b128766 JB |
3325 | path->reada = -1; |
3326 | ||
3327 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
962a298f | 3328 | key.type = BTRFS_ORPHAN_ITEM_KEY; |
7b128766 JB |
3329 | key.offset = (u64)-1; |
3330 | ||
7b128766 JB |
3331 | while (1) { |
3332 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
3333 | if (ret < 0) |
3334 | goto out; | |
7b128766 JB |
3335 | |
3336 | /* | |
3337 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 3338 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
3339 | * find the key and see if we have stuff that matches |
3340 | */ | |
3341 | if (ret > 0) { | |
66b4ffd1 | 3342 | ret = 0; |
7b128766 JB |
3343 | if (path->slots[0] == 0) |
3344 | break; | |
3345 | path->slots[0]--; | |
3346 | } | |
3347 | ||
3348 | /* pull out the item */ | |
3349 | leaf = path->nodes[0]; | |
7b128766 JB |
3350 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
3351 | ||
3352 | /* make sure the item matches what we want */ | |
3353 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
3354 | break; | |
962a298f | 3355 | if (found_key.type != BTRFS_ORPHAN_ITEM_KEY) |
7b128766 JB |
3356 | break; |
3357 | ||
3358 | /* release the path since we're done with it */ | |
b3b4aa74 | 3359 | btrfs_release_path(path); |
7b128766 JB |
3360 | |
3361 | /* | |
3362 | * this is where we are basically btrfs_lookup, without the | |
3363 | * crossing root thing. we store the inode number in the | |
3364 | * offset of the orphan item. | |
3365 | */ | |
8f6d7f4f JB |
3366 | |
3367 | if (found_key.offset == last_objectid) { | |
c2cf52eb SK |
3368 | btrfs_err(root->fs_info, |
3369 | "Error removing orphan entry, stopping orphan cleanup"); | |
8f6d7f4f JB |
3370 | ret = -EINVAL; |
3371 | goto out; | |
3372 | } | |
3373 | ||
3374 | last_objectid = found_key.offset; | |
3375 | ||
5d4f98a2 YZ |
3376 | found_key.objectid = found_key.offset; |
3377 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
3378 | found_key.offset = 0; | |
73f73415 | 3379 | inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL); |
8c6ffba0 | 3380 | ret = PTR_ERR_OR_ZERO(inode); |
a8c9e576 | 3381 | if (ret && ret != -ESTALE) |
66b4ffd1 | 3382 | goto out; |
7b128766 | 3383 | |
f8e9e0b0 AJ |
3384 | if (ret == -ESTALE && root == root->fs_info->tree_root) { |
3385 | struct btrfs_root *dead_root; | |
3386 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3387 | int is_dead_root = 0; | |
3388 | ||
3389 | /* | |
3390 | * this is an orphan in the tree root. Currently these | |
3391 | * could come from 2 sources: | |
3392 | * a) a snapshot deletion in progress | |
3393 | * b) a free space cache inode | |
3394 | * We need to distinguish those two, as the snapshot | |
3395 | * orphan must not get deleted. | |
3396 | * find_dead_roots already ran before us, so if this | |
3397 | * is a snapshot deletion, we should find the root | |
3398 | * in the dead_roots list | |
3399 | */ | |
3400 | spin_lock(&fs_info->trans_lock); | |
3401 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
3402 | root_list) { | |
3403 | if (dead_root->root_key.objectid == | |
3404 | found_key.objectid) { | |
3405 | is_dead_root = 1; | |
3406 | break; | |
3407 | } | |
3408 | } | |
3409 | spin_unlock(&fs_info->trans_lock); | |
3410 | if (is_dead_root) { | |
3411 | /* prevent this orphan from being found again */ | |
3412 | key.offset = found_key.objectid - 1; | |
3413 | continue; | |
3414 | } | |
3415 | } | |
7b128766 | 3416 | /* |
a8c9e576 JB |
3417 | * Inode is already gone but the orphan item is still there, |
3418 | * kill the orphan item. | |
7b128766 | 3419 | */ |
a8c9e576 JB |
3420 | if (ret == -ESTALE) { |
3421 | trans = btrfs_start_transaction(root, 1); | |
66b4ffd1 JB |
3422 | if (IS_ERR(trans)) { |
3423 | ret = PTR_ERR(trans); | |
3424 | goto out; | |
3425 | } | |
c2cf52eb SK |
3426 | btrfs_debug(root->fs_info, "auto deleting %Lu", |
3427 | found_key.objectid); | |
a8c9e576 JB |
3428 | ret = btrfs_del_orphan_item(trans, root, |
3429 | found_key.objectid); | |
5b21f2ed | 3430 | btrfs_end_transaction(trans, root); |
4ef31a45 JB |
3431 | if (ret) |
3432 | goto out; | |
7b128766 JB |
3433 | continue; |
3434 | } | |
3435 | ||
a8c9e576 JB |
3436 | /* |
3437 | * add this inode to the orphan list so btrfs_orphan_del does | |
3438 | * the proper thing when we hit it | |
3439 | */ | |
8a35d95f JB |
3440 | set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
3441 | &BTRFS_I(inode)->runtime_flags); | |
925396ec | 3442 | atomic_inc(&root->orphan_inodes); |
a8c9e576 | 3443 | |
7b128766 JB |
3444 | /* if we have links, this was a truncate, lets do that */ |
3445 | if (inode->i_nlink) { | |
fae7f21c | 3446 | if (WARN_ON(!S_ISREG(inode->i_mode))) { |
a41ad394 JB |
3447 | iput(inode); |
3448 | continue; | |
3449 | } | |
7b128766 | 3450 | nr_truncate++; |
f3fe820c JB |
3451 | |
3452 | /* 1 for the orphan item deletion. */ | |
3453 | trans = btrfs_start_transaction(root, 1); | |
3454 | if (IS_ERR(trans)) { | |
c69b26b0 | 3455 | iput(inode); |
f3fe820c JB |
3456 | ret = PTR_ERR(trans); |
3457 | goto out; | |
3458 | } | |
3459 | ret = btrfs_orphan_add(trans, inode); | |
3460 | btrfs_end_transaction(trans, root); | |
c69b26b0 JB |
3461 | if (ret) { |
3462 | iput(inode); | |
f3fe820c | 3463 | goto out; |
c69b26b0 | 3464 | } |
f3fe820c | 3465 | |
66b4ffd1 | 3466 | ret = btrfs_truncate(inode); |
4a7d0f68 JB |
3467 | if (ret) |
3468 | btrfs_orphan_del(NULL, inode); | |
7b128766 JB |
3469 | } else { |
3470 | nr_unlink++; | |
3471 | } | |
3472 | ||
3473 | /* this will do delete_inode and everything for us */ | |
3474 | iput(inode); | |
66b4ffd1 JB |
3475 | if (ret) |
3476 | goto out; | |
7b128766 | 3477 | } |
3254c876 MX |
3478 | /* release the path since we're done with it */ |
3479 | btrfs_release_path(path); | |
3480 | ||
d68fc57b YZ |
3481 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
3482 | ||
3483 | if (root->orphan_block_rsv) | |
3484 | btrfs_block_rsv_release(root, root->orphan_block_rsv, | |
3485 | (u64)-1); | |
3486 | ||
27cdeb70 MX |
3487 | if (root->orphan_block_rsv || |
3488 | test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) { | |
7a7eaa40 | 3489 | trans = btrfs_join_transaction(root); |
66b4ffd1 JB |
3490 | if (!IS_ERR(trans)) |
3491 | btrfs_end_transaction(trans, root); | |
d68fc57b | 3492 | } |
7b128766 JB |
3493 | |
3494 | if (nr_unlink) | |
4884b476 | 3495 | btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink); |
7b128766 | 3496 | if (nr_truncate) |
4884b476 | 3497 | btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate); |
66b4ffd1 JB |
3498 | |
3499 | out: | |
3500 | if (ret) | |
68b663d1 | 3501 | btrfs_err(root->fs_info, |
c2cf52eb | 3502 | "could not do orphan cleanup %d", ret); |
66b4ffd1 JB |
3503 | btrfs_free_path(path); |
3504 | return ret; | |
7b128766 JB |
3505 | } |
3506 | ||
46a53cca CM |
3507 | /* |
3508 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
3509 | * don't find any xattrs, we know there can't be any acls. | |
3510 | * | |
3511 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
3512 | */ | |
3513 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
63541927 FDBM |
3514 | int slot, u64 objectid, |
3515 | int *first_xattr_slot) | |
46a53cca CM |
3516 | { |
3517 | u32 nritems = btrfs_header_nritems(leaf); | |
3518 | struct btrfs_key found_key; | |
f23b5a59 JB |
3519 | static u64 xattr_access = 0; |
3520 | static u64 xattr_default = 0; | |
46a53cca CM |
3521 | int scanned = 0; |
3522 | ||
f23b5a59 JB |
3523 | if (!xattr_access) { |
3524 | xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS, | |
3525 | strlen(POSIX_ACL_XATTR_ACCESS)); | |
3526 | xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT, | |
3527 | strlen(POSIX_ACL_XATTR_DEFAULT)); | |
3528 | } | |
3529 | ||
46a53cca | 3530 | slot++; |
63541927 | 3531 | *first_xattr_slot = -1; |
46a53cca CM |
3532 | while (slot < nritems) { |
3533 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3534 | ||
3535 | /* we found a different objectid, there must not be acls */ | |
3536 | if (found_key.objectid != objectid) | |
3537 | return 0; | |
3538 | ||
3539 | /* we found an xattr, assume we've got an acl */ | |
f23b5a59 | 3540 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) { |
63541927 FDBM |
3541 | if (*first_xattr_slot == -1) |
3542 | *first_xattr_slot = slot; | |
f23b5a59 JB |
3543 | if (found_key.offset == xattr_access || |
3544 | found_key.offset == xattr_default) | |
3545 | return 1; | |
3546 | } | |
46a53cca CM |
3547 | |
3548 | /* | |
3549 | * we found a key greater than an xattr key, there can't | |
3550 | * be any acls later on | |
3551 | */ | |
3552 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
3553 | return 0; | |
3554 | ||
3555 | slot++; | |
3556 | scanned++; | |
3557 | ||
3558 | /* | |
3559 | * it goes inode, inode backrefs, xattrs, extents, | |
3560 | * so if there are a ton of hard links to an inode there can | |
3561 | * be a lot of backrefs. Don't waste time searching too hard, | |
3562 | * this is just an optimization | |
3563 | */ | |
3564 | if (scanned >= 8) | |
3565 | break; | |
3566 | } | |
3567 | /* we hit the end of the leaf before we found an xattr or | |
3568 | * something larger than an xattr. We have to assume the inode | |
3569 | * has acls | |
3570 | */ | |
63541927 FDBM |
3571 | if (*first_xattr_slot == -1) |
3572 | *first_xattr_slot = slot; | |
46a53cca CM |
3573 | return 1; |
3574 | } | |
3575 | ||
d352ac68 CM |
3576 | /* |
3577 | * read an inode from the btree into the in-memory inode | |
3578 | */ | |
5d4f98a2 | 3579 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
3580 | { |
3581 | struct btrfs_path *path; | |
5f39d397 | 3582 | struct extent_buffer *leaf; |
39279cc3 CM |
3583 | struct btrfs_inode_item *inode_item; |
3584 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3585 | struct btrfs_key location; | |
67de1176 | 3586 | unsigned long ptr; |
46a53cca | 3587 | int maybe_acls; |
618e21d5 | 3588 | u32 rdev; |
39279cc3 | 3589 | int ret; |
2f7e33d4 | 3590 | bool filled = false; |
63541927 | 3591 | int first_xattr_slot; |
2f7e33d4 MX |
3592 | |
3593 | ret = btrfs_fill_inode(inode, &rdev); | |
3594 | if (!ret) | |
3595 | filled = true; | |
39279cc3 CM |
3596 | |
3597 | path = btrfs_alloc_path(); | |
1748f843 MF |
3598 | if (!path) |
3599 | goto make_bad; | |
3600 | ||
39279cc3 | 3601 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 3602 | |
39279cc3 | 3603 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 3604 | if (ret) |
39279cc3 | 3605 | goto make_bad; |
39279cc3 | 3606 | |
5f39d397 | 3607 | leaf = path->nodes[0]; |
2f7e33d4 MX |
3608 | |
3609 | if (filled) | |
67de1176 | 3610 | goto cache_index; |
2f7e33d4 | 3611 | |
5f39d397 CM |
3612 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
3613 | struct btrfs_inode_item); | |
5f39d397 | 3614 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 3615 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
2f2f43d3 EB |
3616 | i_uid_write(inode, btrfs_inode_uid(leaf, inode_item)); |
3617 | i_gid_write(inode, btrfs_inode_gid(leaf, inode_item)); | |
dbe674a9 | 3618 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 | 3619 | |
a937b979 DS |
3620 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime); |
3621 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime); | |
5f39d397 | 3622 | |
a937b979 DS |
3623 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime); |
3624 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime); | |
5f39d397 | 3625 | |
a937b979 DS |
3626 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime); |
3627 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime); | |
5f39d397 | 3628 | |
9cc97d64 | 3629 | BTRFS_I(inode)->i_otime.tv_sec = |
3630 | btrfs_timespec_sec(leaf, &inode_item->otime); | |
3631 | BTRFS_I(inode)->i_otime.tv_nsec = | |
3632 | btrfs_timespec_nsec(leaf, &inode_item->otime); | |
5f39d397 | 3633 | |
a76a3cd4 | 3634 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 3635 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
5dc562c5 JB |
3636 | BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item); |
3637 | ||
6e17d30b YD |
3638 | inode->i_version = btrfs_inode_sequence(leaf, inode_item); |
3639 | inode->i_generation = BTRFS_I(inode)->generation; | |
3640 | inode->i_rdev = 0; | |
3641 | rdev = btrfs_inode_rdev(leaf, inode_item); | |
3642 | ||
3643 | BTRFS_I(inode)->index_cnt = (u64)-1; | |
3644 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); | |
3645 | ||
3646 | cache_index: | |
5dc562c5 JB |
3647 | /* |
3648 | * If we were modified in the current generation and evicted from memory | |
3649 | * and then re-read we need to do a full sync since we don't have any | |
3650 | * idea about which extents were modified before we were evicted from | |
3651 | * cache. | |
6e17d30b YD |
3652 | * |
3653 | * This is required for both inode re-read from disk and delayed inode | |
3654 | * in delayed_nodes_tree. | |
5dc562c5 JB |
3655 | */ |
3656 | if (BTRFS_I(inode)->last_trans == root->fs_info->generation) | |
3657 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
3658 | &BTRFS_I(inode)->runtime_flags); | |
3659 | ||
bde6c242 FM |
3660 | /* |
3661 | * We don't persist the id of the transaction where an unlink operation | |
3662 | * against the inode was last made. So here we assume the inode might | |
3663 | * have been evicted, and therefore the exact value of last_unlink_trans | |
3664 | * lost, and set it to last_trans to avoid metadata inconsistencies | |
3665 | * between the inode and its parent if the inode is fsync'ed and the log | |
3666 | * replayed. For example, in the scenario: | |
3667 | * | |
3668 | * touch mydir/foo | |
3669 | * ln mydir/foo mydir/bar | |
3670 | * sync | |
3671 | * unlink mydir/bar | |
3672 | * echo 2 > /proc/sys/vm/drop_caches # evicts inode | |
3673 | * xfs_io -c fsync mydir/foo | |
3674 | * <power failure> | |
3675 | * mount fs, triggers fsync log replay | |
3676 | * | |
3677 | * We must make sure that when we fsync our inode foo we also log its | |
3678 | * parent inode, otherwise after log replay the parent still has the | |
3679 | * dentry with the "bar" name but our inode foo has a link count of 1 | |
3680 | * and doesn't have an inode ref with the name "bar" anymore. | |
3681 | * | |
3682 | * Setting last_unlink_trans to last_trans is a pessimistic approach, | |
3683 | * but it guarantees correctness at the expense of ocassional full | |
3684 | * transaction commits on fsync if our inode is a directory, or if our | |
3685 | * inode is not a directory, logging its parent unnecessarily. | |
3686 | */ | |
3687 | BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans; | |
3688 | ||
67de1176 MX |
3689 | path->slots[0]++; |
3690 | if (inode->i_nlink != 1 || | |
3691 | path->slots[0] >= btrfs_header_nritems(leaf)) | |
3692 | goto cache_acl; | |
3693 | ||
3694 | btrfs_item_key_to_cpu(leaf, &location, path->slots[0]); | |
3695 | if (location.objectid != btrfs_ino(inode)) | |
3696 | goto cache_acl; | |
3697 | ||
3698 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3699 | if (location.type == BTRFS_INODE_REF_KEY) { | |
3700 | struct btrfs_inode_ref *ref; | |
3701 | ||
3702 | ref = (struct btrfs_inode_ref *)ptr; | |
3703 | BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref); | |
3704 | } else if (location.type == BTRFS_INODE_EXTREF_KEY) { | |
3705 | struct btrfs_inode_extref *extref; | |
3706 | ||
3707 | extref = (struct btrfs_inode_extref *)ptr; | |
3708 | BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf, | |
3709 | extref); | |
3710 | } | |
2f7e33d4 | 3711 | cache_acl: |
46a53cca CM |
3712 | /* |
3713 | * try to precache a NULL acl entry for files that don't have | |
3714 | * any xattrs or acls | |
3715 | */ | |
33345d01 | 3716 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
63541927 FDBM |
3717 | btrfs_ino(inode), &first_xattr_slot); |
3718 | if (first_xattr_slot != -1) { | |
3719 | path->slots[0] = first_xattr_slot; | |
3720 | ret = btrfs_load_inode_props(inode, path); | |
3721 | if (ret) | |
3722 | btrfs_err(root->fs_info, | |
351fd353 | 3723 | "error loading props for ino %llu (root %llu): %d", |
63541927 FDBM |
3724 | btrfs_ino(inode), |
3725 | root->root_key.objectid, ret); | |
3726 | } | |
3727 | btrfs_free_path(path); | |
3728 | ||
72c04902 AV |
3729 | if (!maybe_acls) |
3730 | cache_no_acl(inode); | |
46a53cca | 3731 | |
39279cc3 | 3732 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
3733 | case S_IFREG: |
3734 | inode->i_mapping->a_ops = &btrfs_aops; | |
d1310b2e | 3735 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
3736 | inode->i_fop = &btrfs_file_operations; |
3737 | inode->i_op = &btrfs_file_inode_operations; | |
3738 | break; | |
3739 | case S_IFDIR: | |
3740 | inode->i_fop = &btrfs_dir_file_operations; | |
3741 | if (root == root->fs_info->tree_root) | |
3742 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
3743 | else | |
3744 | inode->i_op = &btrfs_dir_inode_operations; | |
3745 | break; | |
3746 | case S_IFLNK: | |
3747 | inode->i_op = &btrfs_symlink_inode_operations; | |
3748 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
3749 | break; | |
618e21d5 | 3750 | default: |
0279b4cd | 3751 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
3752 | init_special_inode(inode, inode->i_mode, rdev); |
3753 | break; | |
39279cc3 | 3754 | } |
6cbff00f CH |
3755 | |
3756 | btrfs_update_iflags(inode); | |
39279cc3 CM |
3757 | return; |
3758 | ||
3759 | make_bad: | |
39279cc3 | 3760 | btrfs_free_path(path); |
39279cc3 CM |
3761 | make_bad_inode(inode); |
3762 | } | |
3763 | ||
d352ac68 CM |
3764 | /* |
3765 | * given a leaf and an inode, copy the inode fields into the leaf | |
3766 | */ | |
e02119d5 CM |
3767 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
3768 | struct extent_buffer *leaf, | |
5f39d397 | 3769 | struct btrfs_inode_item *item, |
39279cc3 CM |
3770 | struct inode *inode) |
3771 | { | |
51fab693 LB |
3772 | struct btrfs_map_token token; |
3773 | ||
3774 | btrfs_init_map_token(&token); | |
5f39d397 | 3775 | |
51fab693 LB |
3776 | btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token); |
3777 | btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token); | |
3778 | btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size, | |
3779 | &token); | |
3780 | btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token); | |
3781 | btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token); | |
5f39d397 | 3782 | |
a937b979 | 3783 | btrfs_set_token_timespec_sec(leaf, &item->atime, |
51fab693 | 3784 | inode->i_atime.tv_sec, &token); |
a937b979 | 3785 | btrfs_set_token_timespec_nsec(leaf, &item->atime, |
51fab693 | 3786 | inode->i_atime.tv_nsec, &token); |
5f39d397 | 3787 | |
a937b979 | 3788 | btrfs_set_token_timespec_sec(leaf, &item->mtime, |
51fab693 | 3789 | inode->i_mtime.tv_sec, &token); |
a937b979 | 3790 | btrfs_set_token_timespec_nsec(leaf, &item->mtime, |
51fab693 | 3791 | inode->i_mtime.tv_nsec, &token); |
5f39d397 | 3792 | |
a937b979 | 3793 | btrfs_set_token_timespec_sec(leaf, &item->ctime, |
51fab693 | 3794 | inode->i_ctime.tv_sec, &token); |
a937b979 | 3795 | btrfs_set_token_timespec_nsec(leaf, &item->ctime, |
51fab693 | 3796 | inode->i_ctime.tv_nsec, &token); |
5f39d397 | 3797 | |
9cc97d64 | 3798 | btrfs_set_token_timespec_sec(leaf, &item->otime, |
3799 | BTRFS_I(inode)->i_otime.tv_sec, &token); | |
3800 | btrfs_set_token_timespec_nsec(leaf, &item->otime, | |
3801 | BTRFS_I(inode)->i_otime.tv_nsec, &token); | |
3802 | ||
51fab693 LB |
3803 | btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode), |
3804 | &token); | |
3805 | btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation, | |
3806 | &token); | |
3807 | btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token); | |
3808 | btrfs_set_token_inode_transid(leaf, item, trans->transid, &token); | |
3809 | btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token); | |
3810 | btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token); | |
3811 | btrfs_set_token_inode_block_group(leaf, item, 0, &token); | |
39279cc3 CM |
3812 | } |
3813 | ||
d352ac68 CM |
3814 | /* |
3815 | * copy everything in the in-memory inode into the btree. | |
3816 | */ | |
2115133f | 3817 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 3818 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
3819 | { |
3820 | struct btrfs_inode_item *inode_item; | |
3821 | struct btrfs_path *path; | |
5f39d397 | 3822 | struct extent_buffer *leaf; |
39279cc3 CM |
3823 | int ret; |
3824 | ||
3825 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
3826 | if (!path) |
3827 | return -ENOMEM; | |
3828 | ||
b9473439 | 3829 | path->leave_spinning = 1; |
16cdcec7 MX |
3830 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
3831 | 1); | |
39279cc3 CM |
3832 | if (ret) { |
3833 | if (ret > 0) | |
3834 | ret = -ENOENT; | |
3835 | goto failed; | |
3836 | } | |
3837 | ||
5f39d397 CM |
3838 | leaf = path->nodes[0]; |
3839 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 3840 | struct btrfs_inode_item); |
39279cc3 | 3841 | |
e02119d5 | 3842 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 3843 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 3844 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
3845 | ret = 0; |
3846 | failed: | |
39279cc3 CM |
3847 | btrfs_free_path(path); |
3848 | return ret; | |
3849 | } | |
3850 | ||
2115133f CM |
3851 | /* |
3852 | * copy everything in the in-memory inode into the btree. | |
3853 | */ | |
3854 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
3855 | struct btrfs_root *root, struct inode *inode) | |
3856 | { | |
3857 | int ret; | |
3858 | ||
3859 | /* | |
3860 | * If the inode is a free space inode, we can deadlock during commit | |
3861 | * if we put it into the delayed code. | |
3862 | * | |
3863 | * The data relocation inode should also be directly updated | |
3864 | * without delay | |
3865 | */ | |
83eea1f1 | 3866 | if (!btrfs_is_free_space_inode(inode) |
1d52c78a JB |
3867 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
3868 | && !root->fs_info->log_root_recovering) { | |
8ea05e3a AB |
3869 | btrfs_update_root_times(trans, root); |
3870 | ||
2115133f CM |
3871 | ret = btrfs_delayed_update_inode(trans, root, inode); |
3872 | if (!ret) | |
3873 | btrfs_set_inode_last_trans(trans, inode); | |
3874 | return ret; | |
3875 | } | |
3876 | ||
3877 | return btrfs_update_inode_item(trans, root, inode); | |
3878 | } | |
3879 | ||
be6aef60 JB |
3880 | noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
3881 | struct btrfs_root *root, | |
3882 | struct inode *inode) | |
2115133f CM |
3883 | { |
3884 | int ret; | |
3885 | ||
3886 | ret = btrfs_update_inode(trans, root, inode); | |
3887 | if (ret == -ENOSPC) | |
3888 | return btrfs_update_inode_item(trans, root, inode); | |
3889 | return ret; | |
3890 | } | |
3891 | ||
d352ac68 CM |
3892 | /* |
3893 | * unlink helper that gets used here in inode.c and in the tree logging | |
3894 | * recovery code. It remove a link in a directory with a given name, and | |
3895 | * also drops the back refs in the inode to the directory | |
3896 | */ | |
92986796 AV |
3897 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3898 | struct btrfs_root *root, | |
3899 | struct inode *dir, struct inode *inode, | |
3900 | const char *name, int name_len) | |
39279cc3 CM |
3901 | { |
3902 | struct btrfs_path *path; | |
39279cc3 | 3903 | int ret = 0; |
5f39d397 | 3904 | struct extent_buffer *leaf; |
39279cc3 | 3905 | struct btrfs_dir_item *di; |
5f39d397 | 3906 | struct btrfs_key key; |
aec7477b | 3907 | u64 index; |
33345d01 LZ |
3908 | u64 ino = btrfs_ino(inode); |
3909 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
3910 | |
3911 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
3912 | if (!path) { |
3913 | ret = -ENOMEM; | |
554233a6 | 3914 | goto out; |
54aa1f4d CM |
3915 | } |
3916 | ||
b9473439 | 3917 | path->leave_spinning = 1; |
33345d01 | 3918 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 CM |
3919 | name, name_len, -1); |
3920 | if (IS_ERR(di)) { | |
3921 | ret = PTR_ERR(di); | |
3922 | goto err; | |
3923 | } | |
3924 | if (!di) { | |
3925 | ret = -ENOENT; | |
3926 | goto err; | |
3927 | } | |
5f39d397 CM |
3928 | leaf = path->nodes[0]; |
3929 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 3930 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
3931 | if (ret) |
3932 | goto err; | |
b3b4aa74 | 3933 | btrfs_release_path(path); |
39279cc3 | 3934 | |
67de1176 MX |
3935 | /* |
3936 | * If we don't have dir index, we have to get it by looking up | |
3937 | * the inode ref, since we get the inode ref, remove it directly, | |
3938 | * it is unnecessary to do delayed deletion. | |
3939 | * | |
3940 | * But if we have dir index, needn't search inode ref to get it. | |
3941 | * Since the inode ref is close to the inode item, it is better | |
3942 | * that we delay to delete it, and just do this deletion when | |
3943 | * we update the inode item. | |
3944 | */ | |
3945 | if (BTRFS_I(inode)->dir_index) { | |
3946 | ret = btrfs_delayed_delete_inode_ref(inode); | |
3947 | if (!ret) { | |
3948 | index = BTRFS_I(inode)->dir_index; | |
3949 | goto skip_backref; | |
3950 | } | |
3951 | } | |
3952 | ||
33345d01 LZ |
3953 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
3954 | dir_ino, &index); | |
aec7477b | 3955 | if (ret) { |
c2cf52eb SK |
3956 | btrfs_info(root->fs_info, |
3957 | "failed to delete reference to %.*s, inode %llu parent %llu", | |
c1c9ff7c | 3958 | name_len, name, ino, dir_ino); |
79787eaa | 3959 | btrfs_abort_transaction(trans, root, ret); |
aec7477b JB |
3960 | goto err; |
3961 | } | |
67de1176 | 3962 | skip_backref: |
16cdcec7 | 3963 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
3964 | if (ret) { |
3965 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 3966 | goto err; |
79787eaa | 3967 | } |
39279cc3 | 3968 | |
e02119d5 | 3969 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
33345d01 | 3970 | inode, dir_ino); |
79787eaa JM |
3971 | if (ret != 0 && ret != -ENOENT) { |
3972 | btrfs_abort_transaction(trans, root, ret); | |
3973 | goto err; | |
3974 | } | |
e02119d5 CM |
3975 | |
3976 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
3977 | dir, index); | |
6418c961 CM |
3978 | if (ret == -ENOENT) |
3979 | ret = 0; | |
d4e3991b ZB |
3980 | else if (ret) |
3981 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 CM |
3982 | err: |
3983 | btrfs_free_path(path); | |
e02119d5 CM |
3984 | if (ret) |
3985 | goto out; | |
3986 | ||
3987 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 JB |
3988 | inode_inc_iversion(inode); |
3989 | inode_inc_iversion(dir); | |
e02119d5 | 3990 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
b9959295 | 3991 | ret = btrfs_update_inode(trans, root, dir); |
e02119d5 | 3992 | out: |
39279cc3 CM |
3993 | return ret; |
3994 | } | |
3995 | ||
92986796 AV |
3996 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
3997 | struct btrfs_root *root, | |
3998 | struct inode *dir, struct inode *inode, | |
3999 | const char *name, int name_len) | |
4000 | { | |
4001 | int ret; | |
4002 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
4003 | if (!ret) { | |
8b558c5f | 4004 | drop_nlink(inode); |
92986796 AV |
4005 | ret = btrfs_update_inode(trans, root, inode); |
4006 | } | |
4007 | return ret; | |
4008 | } | |
39279cc3 | 4009 | |
a22285a6 YZ |
4010 | /* |
4011 | * helper to start transaction for unlink and rmdir. | |
4012 | * | |
d52be818 JB |
4013 | * unlink and rmdir are special in btrfs, they do not always free space, so |
4014 | * if we cannot make our reservations the normal way try and see if there is | |
4015 | * plenty of slack room in the global reserve to migrate, otherwise we cannot | |
4016 | * allow the unlink to occur. | |
a22285a6 | 4017 | */ |
d52be818 | 4018 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir) |
4df27c4d | 4019 | { |
39279cc3 | 4020 | struct btrfs_trans_handle *trans; |
a22285a6 | 4021 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d YZ |
4022 | int ret; |
4023 | ||
e70bea5f JB |
4024 | /* |
4025 | * 1 for the possible orphan item | |
4026 | * 1 for the dir item | |
4027 | * 1 for the dir index | |
4028 | * 1 for the inode ref | |
e70bea5f JB |
4029 | * 1 for the inode |
4030 | */ | |
6e137ed3 | 4031 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
4032 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) |
4033 | return trans; | |
4df27c4d | 4034 | |
d52be818 JB |
4035 | if (PTR_ERR(trans) == -ENOSPC) { |
4036 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5); | |
4df27c4d | 4037 | |
d52be818 JB |
4038 | trans = btrfs_start_transaction(root, 0); |
4039 | if (IS_ERR(trans)) | |
4040 | return trans; | |
4041 | ret = btrfs_cond_migrate_bytes(root->fs_info, | |
4042 | &root->fs_info->trans_block_rsv, | |
4043 | num_bytes, 5); | |
4044 | if (ret) { | |
4045 | btrfs_end_transaction(trans, root); | |
4046 | return ERR_PTR(ret); | |
a22285a6 | 4047 | } |
5a77d76c | 4048 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
d52be818 | 4049 | trans->bytes_reserved = num_bytes; |
a22285a6 | 4050 | } |
d52be818 | 4051 | return trans; |
a22285a6 YZ |
4052 | } |
4053 | ||
4054 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
4055 | { | |
4056 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4057 | struct btrfs_trans_handle *trans; | |
2b0143b5 | 4058 | struct inode *inode = d_inode(dentry); |
a22285a6 | 4059 | int ret; |
a22285a6 | 4060 | |
d52be818 | 4061 | trans = __unlink_start_trans(dir); |
a22285a6 YZ |
4062 | if (IS_ERR(trans)) |
4063 | return PTR_ERR(trans); | |
5f39d397 | 4064 | |
2b0143b5 | 4065 | btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0); |
12fcfd22 | 4066 | |
2b0143b5 | 4067 | ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry), |
e02119d5 | 4068 | dentry->d_name.name, dentry->d_name.len); |
b532402e TI |
4069 | if (ret) |
4070 | goto out; | |
7b128766 | 4071 | |
a22285a6 | 4072 | if (inode->i_nlink == 0) { |
7b128766 | 4073 | ret = btrfs_orphan_add(trans, inode); |
b532402e TI |
4074 | if (ret) |
4075 | goto out; | |
a22285a6 | 4076 | } |
7b128766 | 4077 | |
b532402e | 4078 | out: |
d52be818 | 4079 | btrfs_end_transaction(trans, root); |
b53d3f5d | 4080 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
4081 | return ret; |
4082 | } | |
4083 | ||
4df27c4d YZ |
4084 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
4085 | struct btrfs_root *root, | |
4086 | struct inode *dir, u64 objectid, | |
4087 | const char *name, int name_len) | |
4088 | { | |
4089 | struct btrfs_path *path; | |
4090 | struct extent_buffer *leaf; | |
4091 | struct btrfs_dir_item *di; | |
4092 | struct btrfs_key key; | |
4093 | u64 index; | |
4094 | int ret; | |
33345d01 | 4095 | u64 dir_ino = btrfs_ino(dir); |
4df27c4d YZ |
4096 | |
4097 | path = btrfs_alloc_path(); | |
4098 | if (!path) | |
4099 | return -ENOMEM; | |
4100 | ||
33345d01 | 4101 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 4102 | name, name_len, -1); |
79787eaa JM |
4103 | if (IS_ERR_OR_NULL(di)) { |
4104 | if (!di) | |
4105 | ret = -ENOENT; | |
4106 | else | |
4107 | ret = PTR_ERR(di); | |
4108 | goto out; | |
4109 | } | |
4df27c4d YZ |
4110 | |
4111 | leaf = path->nodes[0]; | |
4112 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
4113 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
4114 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
79787eaa JM |
4115 | if (ret) { |
4116 | btrfs_abort_transaction(trans, root, ret); | |
4117 | goto out; | |
4118 | } | |
b3b4aa74 | 4119 | btrfs_release_path(path); |
4df27c4d YZ |
4120 | |
4121 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
4122 | objectid, root->root_key.objectid, | |
33345d01 | 4123 | dir_ino, &index, name, name_len); |
4df27c4d | 4124 | if (ret < 0) { |
79787eaa JM |
4125 | if (ret != -ENOENT) { |
4126 | btrfs_abort_transaction(trans, root, ret); | |
4127 | goto out; | |
4128 | } | |
33345d01 | 4129 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 4130 | name, name_len); |
79787eaa JM |
4131 | if (IS_ERR_OR_NULL(di)) { |
4132 | if (!di) | |
4133 | ret = -ENOENT; | |
4134 | else | |
4135 | ret = PTR_ERR(di); | |
4136 | btrfs_abort_transaction(trans, root, ret); | |
4137 | goto out; | |
4138 | } | |
4df27c4d YZ |
4139 | |
4140 | leaf = path->nodes[0]; | |
4141 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 4142 | btrfs_release_path(path); |
4df27c4d YZ |
4143 | index = key.offset; |
4144 | } | |
945d8962 | 4145 | btrfs_release_path(path); |
4df27c4d | 4146 | |
16cdcec7 | 4147 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
79787eaa JM |
4148 | if (ret) { |
4149 | btrfs_abort_transaction(trans, root, ret); | |
4150 | goto out; | |
4151 | } | |
4df27c4d YZ |
4152 | |
4153 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
0c4d2d95 | 4154 | inode_inc_iversion(dir); |
4df27c4d | 4155 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; |
5a24e84c | 4156 | ret = btrfs_update_inode_fallback(trans, root, dir); |
79787eaa JM |
4157 | if (ret) |
4158 | btrfs_abort_transaction(trans, root, ret); | |
4159 | out: | |
71d7aed0 | 4160 | btrfs_free_path(path); |
79787eaa | 4161 | return ret; |
4df27c4d YZ |
4162 | } |
4163 | ||
39279cc3 CM |
4164 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
4165 | { | |
2b0143b5 | 4166 | struct inode *inode = d_inode(dentry); |
1832a6d5 | 4167 | int err = 0; |
39279cc3 | 4168 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 4169 | struct btrfs_trans_handle *trans; |
39279cc3 | 4170 | |
b3ae244e | 4171 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) |
134d4512 | 4172 | return -ENOTEMPTY; |
b3ae244e DS |
4173 | if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) |
4174 | return -EPERM; | |
134d4512 | 4175 | |
d52be818 | 4176 | trans = __unlink_start_trans(dir); |
a22285a6 | 4177 | if (IS_ERR(trans)) |
5df6a9f6 | 4178 | return PTR_ERR(trans); |
5df6a9f6 | 4179 | |
33345d01 | 4180 | if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
4df27c4d YZ |
4181 | err = btrfs_unlink_subvol(trans, root, dir, |
4182 | BTRFS_I(inode)->location.objectid, | |
4183 | dentry->d_name.name, | |
4184 | dentry->d_name.len); | |
4185 | goto out; | |
4186 | } | |
4187 | ||
7b128766 JB |
4188 | err = btrfs_orphan_add(trans, inode); |
4189 | if (err) | |
4df27c4d | 4190 | goto out; |
7b128766 | 4191 | |
39279cc3 | 4192 | /* now the directory is empty */ |
2b0143b5 | 4193 | err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry), |
e02119d5 | 4194 | dentry->d_name.name, dentry->d_name.len); |
d397712b | 4195 | if (!err) |
dbe674a9 | 4196 | btrfs_i_size_write(inode, 0); |
4df27c4d | 4197 | out: |
d52be818 | 4198 | btrfs_end_transaction(trans, root); |
b53d3f5d | 4199 | btrfs_btree_balance_dirty(root); |
3954401f | 4200 | |
39279cc3 CM |
4201 | return err; |
4202 | } | |
4203 | ||
28f75a0e CM |
4204 | static int truncate_space_check(struct btrfs_trans_handle *trans, |
4205 | struct btrfs_root *root, | |
4206 | u64 bytes_deleted) | |
4207 | { | |
4208 | int ret; | |
4209 | ||
4210 | bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted); | |
4211 | ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv, | |
4212 | bytes_deleted, BTRFS_RESERVE_NO_FLUSH); | |
4213 | if (!ret) | |
4214 | trans->bytes_reserved += bytes_deleted; | |
4215 | return ret; | |
4216 | ||
4217 | } | |
4218 | ||
0305cd5f FM |
4219 | static int truncate_inline_extent(struct inode *inode, |
4220 | struct btrfs_path *path, | |
4221 | struct btrfs_key *found_key, | |
4222 | const u64 item_end, | |
4223 | const u64 new_size) | |
4224 | { | |
4225 | struct extent_buffer *leaf = path->nodes[0]; | |
4226 | int slot = path->slots[0]; | |
4227 | struct btrfs_file_extent_item *fi; | |
4228 | u32 size = (u32)(new_size - found_key->offset); | |
4229 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4230 | ||
4231 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
4232 | ||
4233 | if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) { | |
4234 | loff_t offset = new_size; | |
4235 | loff_t page_end = ALIGN(offset, PAGE_CACHE_SIZE); | |
4236 | ||
4237 | /* | |
4238 | * Zero out the remaining of the last page of our inline extent, | |
4239 | * instead of directly truncating our inline extent here - that | |
4240 | * would be much more complex (decompressing all the data, then | |
4241 | * compressing the truncated data, which might be bigger than | |
4242 | * the size of the inline extent, resize the extent, etc). | |
4243 | * We release the path because to get the page we might need to | |
4244 | * read the extent item from disk (data not in the page cache). | |
4245 | */ | |
4246 | btrfs_release_path(path); | |
4247 | return btrfs_truncate_page(inode, offset, page_end - offset, 0); | |
4248 | } | |
4249 | ||
4250 | btrfs_set_file_extent_ram_bytes(leaf, fi, size); | |
4251 | size = btrfs_file_extent_calc_inline_size(size); | |
4252 | btrfs_truncate_item(root, path, size, 1); | |
4253 | ||
4254 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) | |
4255 | inode_sub_bytes(inode, item_end + 1 - new_size); | |
4256 | ||
4257 | return 0; | |
4258 | } | |
4259 | ||
39279cc3 CM |
4260 | /* |
4261 | * this can truncate away extent items, csum items and directory items. | |
4262 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 4263 | * any higher than new_size |
39279cc3 CM |
4264 | * |
4265 | * csum items that cross the new i_size are truncated to the new size | |
4266 | * as well. | |
7b128766 JB |
4267 | * |
4268 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
4269 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 4270 | */ |
8082510e YZ |
4271 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
4272 | struct btrfs_root *root, | |
4273 | struct inode *inode, | |
4274 | u64 new_size, u32 min_type) | |
39279cc3 | 4275 | { |
39279cc3 | 4276 | struct btrfs_path *path; |
5f39d397 | 4277 | struct extent_buffer *leaf; |
39279cc3 | 4278 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
4279 | struct btrfs_key key; |
4280 | struct btrfs_key found_key; | |
39279cc3 | 4281 | u64 extent_start = 0; |
db94535d | 4282 | u64 extent_num_bytes = 0; |
5d4f98a2 | 4283 | u64 extent_offset = 0; |
39279cc3 | 4284 | u64 item_end = 0; |
c1aa4575 | 4285 | u64 last_size = new_size; |
8082510e | 4286 | u32 found_type = (u8)-1; |
39279cc3 CM |
4287 | int found_extent; |
4288 | int del_item; | |
85e21bac CM |
4289 | int pending_del_nr = 0; |
4290 | int pending_del_slot = 0; | |
179e29e4 | 4291 | int extent_type = -1; |
8082510e YZ |
4292 | int ret; |
4293 | int err = 0; | |
33345d01 | 4294 | u64 ino = btrfs_ino(inode); |
28ed1345 | 4295 | u64 bytes_deleted = 0; |
1262133b JB |
4296 | bool be_nice = 0; |
4297 | bool should_throttle = 0; | |
28f75a0e | 4298 | bool should_end = 0; |
8082510e YZ |
4299 | |
4300 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 4301 | |
28ed1345 CM |
4302 | /* |
4303 | * for non-free space inodes and ref cows, we want to back off from | |
4304 | * time to time | |
4305 | */ | |
4306 | if (!btrfs_is_free_space_inode(inode) && | |
4307 | test_bit(BTRFS_ROOT_REF_COWS, &root->state)) | |
4308 | be_nice = 1; | |
4309 | ||
0eb0e19c MF |
4310 | path = btrfs_alloc_path(); |
4311 | if (!path) | |
4312 | return -ENOMEM; | |
4313 | path->reada = -1; | |
4314 | ||
5dc562c5 JB |
4315 | /* |
4316 | * We want to drop from the next block forward in case this new size is | |
4317 | * not block aligned since we will be keeping the last block of the | |
4318 | * extent just the way it is. | |
4319 | */ | |
27cdeb70 MX |
4320 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
4321 | root == root->fs_info->tree_root) | |
fda2832f QW |
4322 | btrfs_drop_extent_cache(inode, ALIGN(new_size, |
4323 | root->sectorsize), (u64)-1, 0); | |
8082510e | 4324 | |
16cdcec7 MX |
4325 | /* |
4326 | * This function is also used to drop the items in the log tree before | |
4327 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
4328 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
4329 | * items. | |
4330 | */ | |
4331 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
4332 | btrfs_kill_delayed_inode_items(inode); | |
4333 | ||
33345d01 | 4334 | key.objectid = ino; |
39279cc3 | 4335 | key.offset = (u64)-1; |
5f39d397 CM |
4336 | key.type = (u8)-1; |
4337 | ||
85e21bac | 4338 | search_again: |
28ed1345 CM |
4339 | /* |
4340 | * with a 16K leaf size and 128MB extents, you can actually queue | |
4341 | * up a huge file in a single leaf. Most of the time that | |
4342 | * bytes_deleted is > 0, it will be huge by the time we get here | |
4343 | */ | |
4344 | if (be_nice && bytes_deleted > 32 * 1024 * 1024) { | |
4345 | if (btrfs_should_end_transaction(trans, root)) { | |
4346 | err = -EAGAIN; | |
4347 | goto error; | |
4348 | } | |
4349 | } | |
4350 | ||
4351 | ||
b9473439 | 4352 | path->leave_spinning = 1; |
85e21bac | 4353 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
4354 | if (ret < 0) { |
4355 | err = ret; | |
4356 | goto out; | |
4357 | } | |
d397712b | 4358 | |
85e21bac | 4359 | if (ret > 0) { |
e02119d5 CM |
4360 | /* there are no items in the tree for us to truncate, we're |
4361 | * done | |
4362 | */ | |
8082510e YZ |
4363 | if (path->slots[0] == 0) |
4364 | goto out; | |
85e21bac CM |
4365 | path->slots[0]--; |
4366 | } | |
4367 | ||
d397712b | 4368 | while (1) { |
39279cc3 | 4369 | fi = NULL; |
5f39d397 CM |
4370 | leaf = path->nodes[0]; |
4371 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
962a298f | 4372 | found_type = found_key.type; |
39279cc3 | 4373 | |
33345d01 | 4374 | if (found_key.objectid != ino) |
39279cc3 | 4375 | break; |
5f39d397 | 4376 | |
85e21bac | 4377 | if (found_type < min_type) |
39279cc3 CM |
4378 | break; |
4379 | ||
5f39d397 | 4380 | item_end = found_key.offset; |
39279cc3 | 4381 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 4382 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 4383 | struct btrfs_file_extent_item); |
179e29e4 CM |
4384 | extent_type = btrfs_file_extent_type(leaf, fi); |
4385 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 4386 | item_end += |
db94535d | 4387 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 4388 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 4389 | item_end += btrfs_file_extent_inline_len(leaf, |
514ac8ad | 4390 | path->slots[0], fi); |
39279cc3 | 4391 | } |
008630c1 | 4392 | item_end--; |
39279cc3 | 4393 | } |
8082510e YZ |
4394 | if (found_type > min_type) { |
4395 | del_item = 1; | |
4396 | } else { | |
4397 | if (item_end < new_size) | |
b888db2b | 4398 | break; |
8082510e YZ |
4399 | if (found_key.offset >= new_size) |
4400 | del_item = 1; | |
4401 | else | |
4402 | del_item = 0; | |
39279cc3 | 4403 | } |
39279cc3 | 4404 | found_extent = 0; |
39279cc3 | 4405 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
4406 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
4407 | goto delete; | |
4408 | ||
7f4f6e0a JB |
4409 | if (del_item) |
4410 | last_size = found_key.offset; | |
4411 | else | |
4412 | last_size = new_size; | |
4413 | ||
179e29e4 | 4414 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { |
39279cc3 | 4415 | u64 num_dec; |
db94535d | 4416 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 4417 | if (!del_item) { |
db94535d CM |
4418 | u64 orig_num_bytes = |
4419 | btrfs_file_extent_num_bytes(leaf, fi); | |
fda2832f QW |
4420 | extent_num_bytes = ALIGN(new_size - |
4421 | found_key.offset, | |
4422 | root->sectorsize); | |
db94535d CM |
4423 | btrfs_set_file_extent_num_bytes(leaf, fi, |
4424 | extent_num_bytes); | |
4425 | num_dec = (orig_num_bytes - | |
9069218d | 4426 | extent_num_bytes); |
27cdeb70 MX |
4427 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4428 | &root->state) && | |
4429 | extent_start != 0) | |
a76a3cd4 | 4430 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 4431 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 4432 | } else { |
db94535d CM |
4433 | extent_num_bytes = |
4434 | btrfs_file_extent_disk_num_bytes(leaf, | |
4435 | fi); | |
5d4f98a2 YZ |
4436 | extent_offset = found_key.offset - |
4437 | btrfs_file_extent_offset(leaf, fi); | |
4438 | ||
39279cc3 | 4439 | /* FIXME blocksize != 4096 */ |
9069218d | 4440 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
4441 | if (extent_start != 0) { |
4442 | found_extent = 1; | |
27cdeb70 MX |
4443 | if (test_bit(BTRFS_ROOT_REF_COWS, |
4444 | &root->state)) | |
a76a3cd4 | 4445 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 4446 | } |
39279cc3 | 4447 | } |
9069218d | 4448 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
4449 | /* |
4450 | * we can't truncate inline items that have had | |
4451 | * special encodings | |
4452 | */ | |
4453 | if (!del_item && | |
c8b97818 CM |
4454 | btrfs_file_extent_encryption(leaf, fi) == 0 && |
4455 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
514ac8ad CM |
4456 | |
4457 | /* | |
0305cd5f FM |
4458 | * Need to release path in order to truncate a |
4459 | * compressed extent. So delete any accumulated | |
4460 | * extent items so far. | |
514ac8ad | 4461 | */ |
0305cd5f FM |
4462 | if (btrfs_file_extent_compression(leaf, fi) != |
4463 | BTRFS_COMPRESS_NONE && pending_del_nr) { | |
4464 | err = btrfs_del_items(trans, root, path, | |
4465 | pending_del_slot, | |
4466 | pending_del_nr); | |
4467 | if (err) { | |
4468 | btrfs_abort_transaction(trans, | |
4469 | root, | |
4470 | err); | |
4471 | goto error; | |
4472 | } | |
4473 | pending_del_nr = 0; | |
4474 | } | |
4475 | ||
4476 | err = truncate_inline_extent(inode, path, | |
4477 | &found_key, | |
4478 | item_end, | |
4479 | new_size); | |
4480 | if (err) { | |
4481 | btrfs_abort_transaction(trans, | |
4482 | root, err); | |
4483 | goto error; | |
4484 | } | |
27cdeb70 MX |
4485 | } else if (test_bit(BTRFS_ROOT_REF_COWS, |
4486 | &root->state)) { | |
0305cd5f | 4487 | inode_sub_bytes(inode, item_end + 1 - new_size); |
9069218d | 4488 | } |
39279cc3 | 4489 | } |
179e29e4 | 4490 | delete: |
39279cc3 | 4491 | if (del_item) { |
85e21bac CM |
4492 | if (!pending_del_nr) { |
4493 | /* no pending yet, add ourselves */ | |
4494 | pending_del_slot = path->slots[0]; | |
4495 | pending_del_nr = 1; | |
4496 | } else if (pending_del_nr && | |
4497 | path->slots[0] + 1 == pending_del_slot) { | |
4498 | /* hop on the pending chunk */ | |
4499 | pending_del_nr++; | |
4500 | pending_del_slot = path->slots[0]; | |
4501 | } else { | |
d397712b | 4502 | BUG(); |
85e21bac | 4503 | } |
39279cc3 CM |
4504 | } else { |
4505 | break; | |
4506 | } | |
28f75a0e CM |
4507 | should_throttle = 0; |
4508 | ||
27cdeb70 MX |
4509 | if (found_extent && |
4510 | (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || | |
4511 | root == root->fs_info->tree_root)) { | |
b9473439 | 4512 | btrfs_set_path_blocking(path); |
28ed1345 | 4513 | bytes_deleted += extent_num_bytes; |
39279cc3 | 4514 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
4515 | extent_num_bytes, 0, |
4516 | btrfs_header_owner(leaf), | |
66d7e7f0 | 4517 | ino, extent_offset, 0); |
39279cc3 | 4518 | BUG_ON(ret); |
1262133b | 4519 | if (btrfs_should_throttle_delayed_refs(trans, root)) |
28ed1345 CM |
4520 | btrfs_async_run_delayed_refs(root, |
4521 | trans->delayed_ref_updates * 2, 0); | |
28f75a0e CM |
4522 | if (be_nice) { |
4523 | if (truncate_space_check(trans, root, | |
4524 | extent_num_bytes)) { | |
4525 | should_end = 1; | |
4526 | } | |
4527 | if (btrfs_should_throttle_delayed_refs(trans, | |
4528 | root)) { | |
4529 | should_throttle = 1; | |
4530 | } | |
4531 | } | |
39279cc3 | 4532 | } |
85e21bac | 4533 | |
8082510e YZ |
4534 | if (found_type == BTRFS_INODE_ITEM_KEY) |
4535 | break; | |
4536 | ||
4537 | if (path->slots[0] == 0 || | |
1262133b | 4538 | path->slots[0] != pending_del_slot || |
28f75a0e | 4539 | should_throttle || should_end) { |
8082510e YZ |
4540 | if (pending_del_nr) { |
4541 | ret = btrfs_del_items(trans, root, path, | |
4542 | pending_del_slot, | |
4543 | pending_del_nr); | |
79787eaa JM |
4544 | if (ret) { |
4545 | btrfs_abort_transaction(trans, | |
4546 | root, ret); | |
4547 | goto error; | |
4548 | } | |
8082510e YZ |
4549 | pending_del_nr = 0; |
4550 | } | |
b3b4aa74 | 4551 | btrfs_release_path(path); |
28f75a0e | 4552 | if (should_throttle) { |
1262133b JB |
4553 | unsigned long updates = trans->delayed_ref_updates; |
4554 | if (updates) { | |
4555 | trans->delayed_ref_updates = 0; | |
4556 | ret = btrfs_run_delayed_refs(trans, root, updates * 2); | |
4557 | if (ret && !err) | |
4558 | err = ret; | |
4559 | } | |
4560 | } | |
28f75a0e CM |
4561 | /* |
4562 | * if we failed to refill our space rsv, bail out | |
4563 | * and let the transaction restart | |
4564 | */ | |
4565 | if (should_end) { | |
4566 | err = -EAGAIN; | |
4567 | goto error; | |
4568 | } | |
85e21bac | 4569 | goto search_again; |
8082510e YZ |
4570 | } else { |
4571 | path->slots[0]--; | |
85e21bac | 4572 | } |
39279cc3 | 4573 | } |
8082510e | 4574 | out: |
85e21bac CM |
4575 | if (pending_del_nr) { |
4576 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
4577 | pending_del_nr); | |
79787eaa JM |
4578 | if (ret) |
4579 | btrfs_abort_transaction(trans, root, ret); | |
85e21bac | 4580 | } |
79787eaa | 4581 | error: |
c1aa4575 | 4582 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) |
7f4f6e0a | 4583 | btrfs_ordered_update_i_size(inode, last_size, NULL); |
28ed1345 | 4584 | |
39279cc3 | 4585 | btrfs_free_path(path); |
28ed1345 | 4586 | |
28f75a0e | 4587 | if (be_nice && bytes_deleted > 32 * 1024 * 1024) { |
28ed1345 CM |
4588 | unsigned long updates = trans->delayed_ref_updates; |
4589 | if (updates) { | |
4590 | trans->delayed_ref_updates = 0; | |
4591 | ret = btrfs_run_delayed_refs(trans, root, updates * 2); | |
4592 | if (ret && !err) | |
4593 | err = ret; | |
4594 | } | |
4595 | } | |
8082510e | 4596 | return err; |
39279cc3 CM |
4597 | } |
4598 | ||
4599 | /* | |
2aaa6655 JB |
4600 | * btrfs_truncate_page - read, zero a chunk and write a page |
4601 | * @inode - inode that we're zeroing | |
4602 | * @from - the offset to start zeroing | |
4603 | * @len - the length to zero, 0 to zero the entire range respective to the | |
4604 | * offset | |
4605 | * @front - zero up to the offset instead of from the offset on | |
4606 | * | |
4607 | * This will find the page for the "from" offset and cow the page and zero the | |
4608 | * part we want to zero. This is used with truncate and hole punching. | |
39279cc3 | 4609 | */ |
2aaa6655 JB |
4610 | int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len, |
4611 | int front) | |
39279cc3 | 4612 | { |
2aaa6655 | 4613 | struct address_space *mapping = inode->i_mapping; |
db94535d | 4614 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
4615 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
4616 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 4617 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 4618 | char *kaddr; |
db94535d | 4619 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
4620 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
4621 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
4622 | struct page *page; | |
3b16a4e3 | 4623 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 4624 | int ret = 0; |
a52d9a80 | 4625 | u64 page_start; |
e6dcd2dc | 4626 | u64 page_end; |
39279cc3 | 4627 | |
2aaa6655 JB |
4628 | if ((offset & (blocksize - 1)) == 0 && |
4629 | (!len || ((len & (blocksize - 1)) == 0))) | |
39279cc3 | 4630 | goto out; |
0ca1f7ce | 4631 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
5d5e103a JB |
4632 | if (ret) |
4633 | goto out; | |
39279cc3 | 4634 | |
211c17f5 | 4635 | again: |
3b16a4e3 | 4636 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 4637 | if (!page) { |
0ca1f7ce | 4638 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
ac6a2b36 | 4639 | ret = -ENOMEM; |
39279cc3 | 4640 | goto out; |
5d5e103a | 4641 | } |
e6dcd2dc CM |
4642 | |
4643 | page_start = page_offset(page); | |
4644 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
4645 | ||
39279cc3 | 4646 | if (!PageUptodate(page)) { |
9ebefb18 | 4647 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 4648 | lock_page(page); |
211c17f5 CM |
4649 | if (page->mapping != mapping) { |
4650 | unlock_page(page); | |
4651 | page_cache_release(page); | |
4652 | goto again; | |
4653 | } | |
39279cc3 CM |
4654 | if (!PageUptodate(page)) { |
4655 | ret = -EIO; | |
89642229 | 4656 | goto out_unlock; |
39279cc3 CM |
4657 | } |
4658 | } | |
211c17f5 | 4659 | wait_on_page_writeback(page); |
e6dcd2dc | 4660 | |
d0082371 | 4661 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
4662 | set_page_extent_mapped(page); |
4663 | ||
4664 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
4665 | if (ordered) { | |
2ac55d41 JB |
4666 | unlock_extent_cached(io_tree, page_start, page_end, |
4667 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
4668 | unlock_page(page); |
4669 | page_cache_release(page); | |
eb84ae03 | 4670 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
4671 | btrfs_put_ordered_extent(ordered); |
4672 | goto again; | |
4673 | } | |
4674 | ||
2ac55d41 | 4675 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
9e8a4a8b LB |
4676 | EXTENT_DIRTY | EXTENT_DELALLOC | |
4677 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 4678 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 4679 | |
2ac55d41 JB |
4680 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
4681 | &cached_state); | |
9ed74f2d | 4682 | if (ret) { |
2ac55d41 JB |
4683 | unlock_extent_cached(io_tree, page_start, page_end, |
4684 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
4685 | goto out_unlock; |
4686 | } | |
4687 | ||
e6dcd2dc | 4688 | if (offset != PAGE_CACHE_SIZE) { |
2aaa6655 JB |
4689 | if (!len) |
4690 | len = PAGE_CACHE_SIZE - offset; | |
e6dcd2dc | 4691 | kaddr = kmap(page); |
2aaa6655 JB |
4692 | if (front) |
4693 | memset(kaddr, 0, offset); | |
4694 | else | |
4695 | memset(kaddr + offset, 0, len); | |
e6dcd2dc CM |
4696 | flush_dcache_page(page); |
4697 | kunmap(page); | |
4698 | } | |
247e743c | 4699 | ClearPageChecked(page); |
e6dcd2dc | 4700 | set_page_dirty(page); |
2ac55d41 JB |
4701 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, |
4702 | GFP_NOFS); | |
39279cc3 | 4703 | |
89642229 | 4704 | out_unlock: |
5d5e103a | 4705 | if (ret) |
0ca1f7ce | 4706 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 CM |
4707 | unlock_page(page); |
4708 | page_cache_release(page); | |
4709 | out: | |
4710 | return ret; | |
4711 | } | |
4712 | ||
16e7549f JB |
4713 | static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode, |
4714 | u64 offset, u64 len) | |
4715 | { | |
4716 | struct btrfs_trans_handle *trans; | |
4717 | int ret; | |
4718 | ||
4719 | /* | |
4720 | * Still need to make sure the inode looks like it's been updated so | |
4721 | * that any holes get logged if we fsync. | |
4722 | */ | |
4723 | if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) { | |
4724 | BTRFS_I(inode)->last_trans = root->fs_info->generation; | |
4725 | BTRFS_I(inode)->last_sub_trans = root->log_transid; | |
4726 | BTRFS_I(inode)->last_log_commit = root->last_log_commit; | |
4727 | return 0; | |
4728 | } | |
4729 | ||
4730 | /* | |
4731 | * 1 - for the one we're dropping | |
4732 | * 1 - for the one we're adding | |
4733 | * 1 - for updating the inode. | |
4734 | */ | |
4735 | trans = btrfs_start_transaction(root, 3); | |
4736 | if (IS_ERR(trans)) | |
4737 | return PTR_ERR(trans); | |
4738 | ||
4739 | ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1); | |
4740 | if (ret) { | |
4741 | btrfs_abort_transaction(trans, root, ret); | |
4742 | btrfs_end_transaction(trans, root); | |
4743 | return ret; | |
4744 | } | |
4745 | ||
4746 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset, | |
4747 | 0, 0, len, 0, len, 0, 0, 0); | |
4748 | if (ret) | |
4749 | btrfs_abort_transaction(trans, root, ret); | |
4750 | else | |
4751 | btrfs_update_inode(trans, root, inode); | |
4752 | btrfs_end_transaction(trans, root); | |
4753 | return ret; | |
4754 | } | |
4755 | ||
695a0d0d JB |
4756 | /* |
4757 | * This function puts in dummy file extents for the area we're creating a hole | |
4758 | * for. So if we are truncating this file to a larger size we need to insert | |
4759 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
4760 | * the range between oldsize and size | |
4761 | */ | |
a41ad394 | 4762 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 4763 | { |
9036c102 YZ |
4764 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4765 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 4766 | struct extent_map *em = NULL; |
2ac55d41 | 4767 | struct extent_state *cached_state = NULL; |
5dc562c5 | 4768 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
fda2832f QW |
4769 | u64 hole_start = ALIGN(oldsize, root->sectorsize); |
4770 | u64 block_end = ALIGN(size, root->sectorsize); | |
9036c102 YZ |
4771 | u64 last_byte; |
4772 | u64 cur_offset; | |
4773 | u64 hole_size; | |
9ed74f2d | 4774 | int err = 0; |
39279cc3 | 4775 | |
a71754fc JB |
4776 | /* |
4777 | * If our size started in the middle of a page we need to zero out the | |
4778 | * rest of the page before we expand the i_size, otherwise we could | |
4779 | * expose stale data. | |
4780 | */ | |
4781 | err = btrfs_truncate_page(inode, oldsize, 0, 0); | |
4782 | if (err) | |
4783 | return err; | |
4784 | ||
9036c102 YZ |
4785 | if (size <= hole_start) |
4786 | return 0; | |
4787 | ||
9036c102 YZ |
4788 | while (1) { |
4789 | struct btrfs_ordered_extent *ordered; | |
fa7c1494 | 4790 | |
2ac55d41 | 4791 | lock_extent_bits(io_tree, hole_start, block_end - 1, 0, |
d0082371 | 4792 | &cached_state); |
fa7c1494 MX |
4793 | ordered = btrfs_lookup_ordered_range(inode, hole_start, |
4794 | block_end - hole_start); | |
9036c102 YZ |
4795 | if (!ordered) |
4796 | break; | |
2ac55d41 JB |
4797 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
4798 | &cached_state, GFP_NOFS); | |
fa7c1494 | 4799 | btrfs_start_ordered_extent(inode, ordered, 1); |
9036c102 YZ |
4800 | btrfs_put_ordered_extent(ordered); |
4801 | } | |
39279cc3 | 4802 | |
9036c102 YZ |
4803 | cur_offset = hole_start; |
4804 | while (1) { | |
4805 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
4806 | block_end - cur_offset, 0); | |
79787eaa JM |
4807 | if (IS_ERR(em)) { |
4808 | err = PTR_ERR(em); | |
f2767956 | 4809 | em = NULL; |
79787eaa JM |
4810 | break; |
4811 | } | |
9036c102 | 4812 | last_byte = min(extent_map_end(em), block_end); |
fda2832f | 4813 | last_byte = ALIGN(last_byte , root->sectorsize); |
8082510e | 4814 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
5dc562c5 | 4815 | struct extent_map *hole_em; |
9036c102 | 4816 | hole_size = last_byte - cur_offset; |
9ed74f2d | 4817 | |
16e7549f JB |
4818 | err = maybe_insert_hole(root, inode, cur_offset, |
4819 | hole_size); | |
4820 | if (err) | |
3893e33b | 4821 | break; |
5dc562c5 JB |
4822 | btrfs_drop_extent_cache(inode, cur_offset, |
4823 | cur_offset + hole_size - 1, 0); | |
4824 | hole_em = alloc_extent_map(); | |
4825 | if (!hole_em) { | |
4826 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
4827 | &BTRFS_I(inode)->runtime_flags); | |
4828 | goto next; | |
4829 | } | |
4830 | hole_em->start = cur_offset; | |
4831 | hole_em->len = hole_size; | |
4832 | hole_em->orig_start = cur_offset; | |
8082510e | 4833 | |
5dc562c5 JB |
4834 | hole_em->block_start = EXTENT_MAP_HOLE; |
4835 | hole_em->block_len = 0; | |
b4939680 | 4836 | hole_em->orig_block_len = 0; |
cc95bef6 | 4837 | hole_em->ram_bytes = hole_size; |
5dc562c5 JB |
4838 | hole_em->bdev = root->fs_info->fs_devices->latest_bdev; |
4839 | hole_em->compress_type = BTRFS_COMPRESS_NONE; | |
16e7549f | 4840 | hole_em->generation = root->fs_info->generation; |
8082510e | 4841 | |
5dc562c5 JB |
4842 | while (1) { |
4843 | write_lock(&em_tree->lock); | |
09a2a8f9 | 4844 | err = add_extent_mapping(em_tree, hole_em, 1); |
5dc562c5 JB |
4845 | write_unlock(&em_tree->lock); |
4846 | if (err != -EEXIST) | |
4847 | break; | |
4848 | btrfs_drop_extent_cache(inode, cur_offset, | |
4849 | cur_offset + | |
4850 | hole_size - 1, 0); | |
4851 | } | |
4852 | free_extent_map(hole_em); | |
9036c102 | 4853 | } |
16e7549f | 4854 | next: |
9036c102 | 4855 | free_extent_map(em); |
a22285a6 | 4856 | em = NULL; |
9036c102 | 4857 | cur_offset = last_byte; |
8082510e | 4858 | if (cur_offset >= block_end) |
9036c102 YZ |
4859 | break; |
4860 | } | |
a22285a6 | 4861 | free_extent_map(em); |
2ac55d41 JB |
4862 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
4863 | GFP_NOFS); | |
9036c102 YZ |
4864 | return err; |
4865 | } | |
39279cc3 | 4866 | |
9ea24bbe FM |
4867 | static int wait_snapshoting_atomic_t(atomic_t *a) |
4868 | { | |
4869 | schedule(); | |
4870 | return 0; | |
4871 | } | |
4872 | ||
4873 | static void wait_for_snapshot_creation(struct btrfs_root *root) | |
4874 | { | |
4875 | while (true) { | |
4876 | int ret; | |
4877 | ||
4878 | ret = btrfs_start_write_no_snapshoting(root); | |
4879 | if (ret) | |
4880 | break; | |
4881 | wait_on_atomic_t(&root->will_be_snapshoted, | |
4882 | wait_snapshoting_atomic_t, | |
4883 | TASK_UNINTERRUPTIBLE); | |
4884 | } | |
4885 | } | |
4886 | ||
3972f260 | 4887 | static int btrfs_setsize(struct inode *inode, struct iattr *attr) |
8082510e | 4888 | { |
f4a2f4c5 MX |
4889 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4890 | struct btrfs_trans_handle *trans; | |
a41ad394 | 4891 | loff_t oldsize = i_size_read(inode); |
3972f260 ES |
4892 | loff_t newsize = attr->ia_size; |
4893 | int mask = attr->ia_valid; | |
8082510e YZ |
4894 | int ret; |
4895 | ||
3972f260 ES |
4896 | /* |
4897 | * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a | |
4898 | * special case where we need to update the times despite not having | |
4899 | * these flags set. For all other operations the VFS set these flags | |
4900 | * explicitly if it wants a timestamp update. | |
4901 | */ | |
dff6efc3 CH |
4902 | if (newsize != oldsize) { |
4903 | inode_inc_iversion(inode); | |
4904 | if (!(mask & (ATTR_CTIME | ATTR_MTIME))) | |
4905 | inode->i_ctime = inode->i_mtime = | |
4906 | current_fs_time(inode->i_sb); | |
4907 | } | |
3972f260 | 4908 | |
a41ad394 | 4909 | if (newsize > oldsize) { |
7caef267 | 4910 | truncate_pagecache(inode, newsize); |
9ea24bbe FM |
4911 | /* |
4912 | * Don't do an expanding truncate while snapshoting is ongoing. | |
4913 | * This is to ensure the snapshot captures a fully consistent | |
4914 | * state of this file - if the snapshot captures this expanding | |
4915 | * truncation, it must capture all writes that happened before | |
4916 | * this truncation. | |
4917 | */ | |
4918 | wait_for_snapshot_creation(root); | |
a41ad394 | 4919 | ret = btrfs_cont_expand(inode, oldsize, newsize); |
9ea24bbe FM |
4920 | if (ret) { |
4921 | btrfs_end_write_no_snapshoting(root); | |
8082510e | 4922 | return ret; |
9ea24bbe | 4923 | } |
8082510e | 4924 | |
f4a2f4c5 | 4925 | trans = btrfs_start_transaction(root, 1); |
9ea24bbe FM |
4926 | if (IS_ERR(trans)) { |
4927 | btrfs_end_write_no_snapshoting(root); | |
f4a2f4c5 | 4928 | return PTR_ERR(trans); |
9ea24bbe | 4929 | } |
f4a2f4c5 MX |
4930 | |
4931 | i_size_write(inode, newsize); | |
4932 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
4933 | ret = btrfs_update_inode(trans, root, inode); | |
9ea24bbe | 4934 | btrfs_end_write_no_snapshoting(root); |
7ad85bb7 | 4935 | btrfs_end_transaction(trans, root); |
a41ad394 | 4936 | } else { |
8082510e | 4937 | |
a41ad394 JB |
4938 | /* |
4939 | * We're truncating a file that used to have good data down to | |
4940 | * zero. Make sure it gets into the ordered flush list so that | |
4941 | * any new writes get down to disk quickly. | |
4942 | */ | |
4943 | if (newsize == 0) | |
72ac3c0d JB |
4944 | set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
4945 | &BTRFS_I(inode)->runtime_flags); | |
8082510e | 4946 | |
f3fe820c JB |
4947 | /* |
4948 | * 1 for the orphan item we're going to add | |
4949 | * 1 for the orphan item deletion. | |
4950 | */ | |
4951 | trans = btrfs_start_transaction(root, 2); | |
4952 | if (IS_ERR(trans)) | |
4953 | return PTR_ERR(trans); | |
4954 | ||
4955 | /* | |
4956 | * We need to do this in case we fail at _any_ point during the | |
4957 | * actual truncate. Once we do the truncate_setsize we could | |
4958 | * invalidate pages which forces any outstanding ordered io to | |
4959 | * be instantly completed which will give us extents that need | |
4960 | * to be truncated. If we fail to get an orphan inode down we | |
4961 | * could have left over extents that were never meant to live, | |
4962 | * so we need to garuntee from this point on that everything | |
4963 | * will be consistent. | |
4964 | */ | |
4965 | ret = btrfs_orphan_add(trans, inode); | |
4966 | btrfs_end_transaction(trans, root); | |
4967 | if (ret) | |
4968 | return ret; | |
4969 | ||
a41ad394 JB |
4970 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
4971 | truncate_setsize(inode, newsize); | |
2e60a51e MX |
4972 | |
4973 | /* Disable nonlocked read DIO to avoid the end less truncate */ | |
4974 | btrfs_inode_block_unlocked_dio(inode); | |
4975 | inode_dio_wait(inode); | |
4976 | btrfs_inode_resume_unlocked_dio(inode); | |
4977 | ||
a41ad394 | 4978 | ret = btrfs_truncate(inode); |
7f4f6e0a JB |
4979 | if (ret && inode->i_nlink) { |
4980 | int err; | |
4981 | ||
4982 | /* | |
4983 | * failed to truncate, disk_i_size is only adjusted down | |
4984 | * as we remove extents, so it should represent the true | |
4985 | * size of the inode, so reset the in memory size and | |
4986 | * delete our orphan entry. | |
4987 | */ | |
4988 | trans = btrfs_join_transaction(root); | |
4989 | if (IS_ERR(trans)) { | |
4990 | btrfs_orphan_del(NULL, inode); | |
4991 | return ret; | |
4992 | } | |
4993 | i_size_write(inode, BTRFS_I(inode)->disk_i_size); | |
4994 | err = btrfs_orphan_del(trans, inode); | |
4995 | if (err) | |
4996 | btrfs_abort_transaction(trans, root, err); | |
4997 | btrfs_end_transaction(trans, root); | |
4998 | } | |
8082510e YZ |
4999 | } |
5000 | ||
a41ad394 | 5001 | return ret; |
8082510e YZ |
5002 | } |
5003 | ||
9036c102 YZ |
5004 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
5005 | { | |
2b0143b5 | 5006 | struct inode *inode = d_inode(dentry); |
b83cc969 | 5007 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 5008 | int err; |
39279cc3 | 5009 | |
b83cc969 LZ |
5010 | if (btrfs_root_readonly(root)) |
5011 | return -EROFS; | |
5012 | ||
9036c102 YZ |
5013 | err = inode_change_ok(inode, attr); |
5014 | if (err) | |
5015 | return err; | |
2bf5a725 | 5016 | |
5a3f23d5 | 5017 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
3972f260 | 5018 | err = btrfs_setsize(inode, attr); |
8082510e YZ |
5019 | if (err) |
5020 | return err; | |
39279cc3 | 5021 | } |
9036c102 | 5022 | |
1025774c CH |
5023 | if (attr->ia_valid) { |
5024 | setattr_copy(inode, attr); | |
0c4d2d95 | 5025 | inode_inc_iversion(inode); |
22c44fe6 | 5026 | err = btrfs_dirty_inode(inode); |
1025774c | 5027 | |
22c44fe6 | 5028 | if (!err && attr->ia_valid & ATTR_MODE) |
996a710d | 5029 | err = posix_acl_chmod(inode, inode->i_mode); |
1025774c | 5030 | } |
33268eaf | 5031 | |
39279cc3 CM |
5032 | return err; |
5033 | } | |
61295eb8 | 5034 | |
131e404a FDBM |
5035 | /* |
5036 | * While truncating the inode pages during eviction, we get the VFS calling | |
5037 | * btrfs_invalidatepage() against each page of the inode. This is slow because | |
5038 | * the calls to btrfs_invalidatepage() result in a huge amount of calls to | |
5039 | * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting | |
5040 | * extent_state structures over and over, wasting lots of time. | |
5041 | * | |
5042 | * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all | |
5043 | * those expensive operations on a per page basis and do only the ordered io | |
5044 | * finishing, while we release here the extent_map and extent_state structures, | |
5045 | * without the excessive merging and splitting. | |
5046 | */ | |
5047 | static void evict_inode_truncate_pages(struct inode *inode) | |
5048 | { | |
5049 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
5050 | struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree; | |
5051 | struct rb_node *node; | |
5052 | ||
5053 | ASSERT(inode->i_state & I_FREEING); | |
91b0abe3 | 5054 | truncate_inode_pages_final(&inode->i_data); |
131e404a FDBM |
5055 | |
5056 | write_lock(&map_tree->lock); | |
5057 | while (!RB_EMPTY_ROOT(&map_tree->map)) { | |
5058 | struct extent_map *em; | |
5059 | ||
5060 | node = rb_first(&map_tree->map); | |
5061 | em = rb_entry(node, struct extent_map, rb_node); | |
180589ef WS |
5062 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
5063 | clear_bit(EXTENT_FLAG_LOGGING, &em->flags); | |
131e404a FDBM |
5064 | remove_extent_mapping(map_tree, em); |
5065 | free_extent_map(em); | |
7064dd5c FM |
5066 | if (need_resched()) { |
5067 | write_unlock(&map_tree->lock); | |
5068 | cond_resched(); | |
5069 | write_lock(&map_tree->lock); | |
5070 | } | |
131e404a FDBM |
5071 | } |
5072 | write_unlock(&map_tree->lock); | |
5073 | ||
6ca07097 FM |
5074 | /* |
5075 | * Keep looping until we have no more ranges in the io tree. | |
5076 | * We can have ongoing bios started by readpages (called from readahead) | |
9c6429d9 FM |
5077 | * that have their endio callback (extent_io.c:end_bio_extent_readpage) |
5078 | * still in progress (unlocked the pages in the bio but did not yet | |
5079 | * unlocked the ranges in the io tree). Therefore this means some | |
6ca07097 FM |
5080 | * ranges can still be locked and eviction started because before |
5081 | * submitting those bios, which are executed by a separate task (work | |
5082 | * queue kthread), inode references (inode->i_count) were not taken | |
5083 | * (which would be dropped in the end io callback of each bio). | |
5084 | * Therefore here we effectively end up waiting for those bios and | |
5085 | * anyone else holding locked ranges without having bumped the inode's | |
5086 | * reference count - if we don't do it, when they access the inode's | |
5087 | * io_tree to unlock a range it may be too late, leading to an | |
5088 | * use-after-free issue. | |
5089 | */ | |
131e404a FDBM |
5090 | spin_lock(&io_tree->lock); |
5091 | while (!RB_EMPTY_ROOT(&io_tree->state)) { | |
5092 | struct extent_state *state; | |
5093 | struct extent_state *cached_state = NULL; | |
6ca07097 FM |
5094 | u64 start; |
5095 | u64 end; | |
131e404a FDBM |
5096 | |
5097 | node = rb_first(&io_tree->state); | |
5098 | state = rb_entry(node, struct extent_state, rb_node); | |
6ca07097 FM |
5099 | start = state->start; |
5100 | end = state->end; | |
131e404a FDBM |
5101 | spin_unlock(&io_tree->lock); |
5102 | ||
6ca07097 FM |
5103 | lock_extent_bits(io_tree, start, end, 0, &cached_state); |
5104 | clear_extent_bit(io_tree, start, end, | |
131e404a FDBM |
5105 | EXTENT_LOCKED | EXTENT_DIRTY | |
5106 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
5107 | EXTENT_DEFRAG, 1, 1, | |
5108 | &cached_state, GFP_NOFS); | |
131e404a | 5109 | |
7064dd5c | 5110 | cond_resched(); |
131e404a FDBM |
5111 | spin_lock(&io_tree->lock); |
5112 | } | |
5113 | spin_unlock(&io_tree->lock); | |
5114 | } | |
5115 | ||
bd555975 | 5116 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 CM |
5117 | { |
5118 | struct btrfs_trans_handle *trans; | |
5119 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
726c35fa | 5120 | struct btrfs_block_rsv *rsv, *global_rsv; |
3bce876f | 5121 | int steal_from_global = 0; |
07127184 | 5122 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 CM |
5123 | int ret; |
5124 | ||
1abe9b8a | 5125 | trace_btrfs_inode_evict(inode); |
5126 | ||
131e404a FDBM |
5127 | evict_inode_truncate_pages(inode); |
5128 | ||
69e9c6c6 SB |
5129 | if (inode->i_nlink && |
5130 | ((btrfs_root_refs(&root->root_item) != 0 && | |
5131 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) || | |
5132 | btrfs_is_free_space_inode(inode))) | |
bd555975 AV |
5133 | goto no_delete; |
5134 | ||
39279cc3 | 5135 | if (is_bad_inode(inode)) { |
7b128766 | 5136 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
5137 | goto no_delete; |
5138 | } | |
bd555975 | 5139 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
a30e577c JM |
5140 | if (!special_file(inode->i_mode)) |
5141 | btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
5f39d397 | 5142 | |
f612496b MX |
5143 | btrfs_free_io_failure_record(inode, 0, (u64)-1); |
5144 | ||
c71bf099 | 5145 | if (root->fs_info->log_root_recovering) { |
6bf02314 | 5146 | BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
8a35d95f | 5147 | &BTRFS_I(inode)->runtime_flags)); |
c71bf099 YZ |
5148 | goto no_delete; |
5149 | } | |
5150 | ||
76dda93c | 5151 | if (inode->i_nlink > 0) { |
69e9c6c6 SB |
5152 | BUG_ON(btrfs_root_refs(&root->root_item) != 0 && |
5153 | root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID); | |
76dda93c YZ |
5154 | goto no_delete; |
5155 | } | |
5156 | ||
0e8c36a9 MX |
5157 | ret = btrfs_commit_inode_delayed_inode(inode); |
5158 | if (ret) { | |
5159 | btrfs_orphan_del(NULL, inode); | |
5160 | goto no_delete; | |
5161 | } | |
5162 | ||
66d8f3dd | 5163 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
4289a667 JB |
5164 | if (!rsv) { |
5165 | btrfs_orphan_del(NULL, inode); | |
5166 | goto no_delete; | |
5167 | } | |
4a338542 | 5168 | rsv->size = min_size; |
ca7e70f5 | 5169 | rsv->failfast = 1; |
726c35fa | 5170 | global_rsv = &root->fs_info->global_block_rsv; |
4289a667 | 5171 | |
dbe674a9 | 5172 | btrfs_i_size_write(inode, 0); |
5f39d397 | 5173 | |
4289a667 | 5174 | /* |
8407aa46 MX |
5175 | * This is a bit simpler than btrfs_truncate since we've already |
5176 | * reserved our space for our orphan item in the unlink, so we just | |
5177 | * need to reserve some slack space in case we add bytes and update | |
5178 | * inode item when doing the truncate. | |
4289a667 | 5179 | */ |
8082510e | 5180 | while (1) { |
08e007d2 MX |
5181 | ret = btrfs_block_rsv_refill(root, rsv, min_size, |
5182 | BTRFS_RESERVE_FLUSH_LIMIT); | |
726c35fa JB |
5183 | |
5184 | /* | |
5185 | * Try and steal from the global reserve since we will | |
5186 | * likely not use this space anyway, we want to try as | |
5187 | * hard as possible to get this to work. | |
5188 | */ | |
5189 | if (ret) | |
3bce876f JB |
5190 | steal_from_global++; |
5191 | else | |
5192 | steal_from_global = 0; | |
5193 | ret = 0; | |
d68fc57b | 5194 | |
3bce876f JB |
5195 | /* |
5196 | * steal_from_global == 0: we reserved stuff, hooray! | |
5197 | * steal_from_global == 1: we didn't reserve stuff, boo! | |
5198 | * steal_from_global == 2: we've committed, still not a lot of | |
5199 | * room but maybe we'll have room in the global reserve this | |
5200 | * time. | |
5201 | * steal_from_global == 3: abandon all hope! | |
5202 | */ | |
5203 | if (steal_from_global > 2) { | |
c2cf52eb SK |
5204 | btrfs_warn(root->fs_info, |
5205 | "Could not get space for a delete, will truncate on mount %d", | |
5206 | ret); | |
4289a667 JB |
5207 | btrfs_orphan_del(NULL, inode); |
5208 | btrfs_free_block_rsv(root, rsv); | |
5209 | goto no_delete; | |
d68fc57b | 5210 | } |
7b128766 | 5211 | |
0e8c36a9 | 5212 | trans = btrfs_join_transaction(root); |
4289a667 JB |
5213 | if (IS_ERR(trans)) { |
5214 | btrfs_orphan_del(NULL, inode); | |
5215 | btrfs_free_block_rsv(root, rsv); | |
5216 | goto no_delete; | |
d68fc57b | 5217 | } |
7b128766 | 5218 | |
3bce876f JB |
5219 | /* |
5220 | * We can't just steal from the global reserve, we need tomake | |
5221 | * sure there is room to do it, if not we need to commit and try | |
5222 | * again. | |
5223 | */ | |
5224 | if (steal_from_global) { | |
5225 | if (!btrfs_check_space_for_delayed_refs(trans, root)) | |
5226 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, | |
5227 | min_size); | |
5228 | else | |
5229 | ret = -ENOSPC; | |
5230 | } | |
5231 | ||
5232 | /* | |
5233 | * Couldn't steal from the global reserve, we have too much | |
5234 | * pending stuff built up, commit the transaction and try it | |
5235 | * again. | |
5236 | */ | |
5237 | if (ret) { | |
5238 | ret = btrfs_commit_transaction(trans, root); | |
5239 | if (ret) { | |
5240 | btrfs_orphan_del(NULL, inode); | |
5241 | btrfs_free_block_rsv(root, rsv); | |
5242 | goto no_delete; | |
5243 | } | |
5244 | continue; | |
5245 | } else { | |
5246 | steal_from_global = 0; | |
5247 | } | |
5248 | ||
4289a667 JB |
5249 | trans->block_rsv = rsv; |
5250 | ||
d68fc57b | 5251 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
28ed1345 | 5252 | if (ret != -ENOSPC && ret != -EAGAIN) |
8082510e | 5253 | break; |
85e21bac | 5254 | |
8407aa46 | 5255 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e YZ |
5256 | btrfs_end_transaction(trans, root); |
5257 | trans = NULL; | |
b53d3f5d | 5258 | btrfs_btree_balance_dirty(root); |
8082510e | 5259 | } |
5f39d397 | 5260 | |
4289a667 JB |
5261 | btrfs_free_block_rsv(root, rsv); |
5262 | ||
4ef31a45 JB |
5263 | /* |
5264 | * Errors here aren't a big deal, it just means we leave orphan items | |
5265 | * in the tree. They will be cleaned up on the next mount. | |
5266 | */ | |
8082510e | 5267 | if (ret == 0) { |
4289a667 | 5268 | trans->block_rsv = root->orphan_block_rsv; |
4ef31a45 JB |
5269 | btrfs_orphan_del(trans, inode); |
5270 | } else { | |
5271 | btrfs_orphan_del(NULL, inode); | |
8082510e | 5272 | } |
54aa1f4d | 5273 | |
4289a667 | 5274 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
581bb050 LZ |
5275 | if (!(root == root->fs_info->tree_root || |
5276 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) | |
33345d01 | 5277 | btrfs_return_ino(root, btrfs_ino(inode)); |
581bb050 | 5278 | |
54aa1f4d | 5279 | btrfs_end_transaction(trans, root); |
b53d3f5d | 5280 | btrfs_btree_balance_dirty(root); |
39279cc3 | 5281 | no_delete: |
89042e5a | 5282 | btrfs_remove_delayed_node(inode); |
dbd5768f | 5283 | clear_inode(inode); |
8082510e | 5284 | return; |
39279cc3 CM |
5285 | } |
5286 | ||
5287 | /* | |
5288 | * this returns the key found in the dir entry in the location pointer. | |
5289 | * If no dir entries were found, location->objectid is 0. | |
5290 | */ | |
5291 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
5292 | struct btrfs_key *location) | |
5293 | { | |
5294 | const char *name = dentry->d_name.name; | |
5295 | int namelen = dentry->d_name.len; | |
5296 | struct btrfs_dir_item *di; | |
5297 | struct btrfs_path *path; | |
5298 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 5299 | int ret = 0; |
39279cc3 CM |
5300 | |
5301 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
5302 | if (!path) |
5303 | return -ENOMEM; | |
3954401f | 5304 | |
33345d01 | 5305 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name, |
39279cc3 | 5306 | namelen, 0); |
0d9f7f3e Y |
5307 | if (IS_ERR(di)) |
5308 | ret = PTR_ERR(di); | |
d397712b | 5309 | |
c704005d | 5310 | if (IS_ERR_OR_NULL(di)) |
3954401f | 5311 | goto out_err; |
d397712b | 5312 | |
5f39d397 | 5313 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 5314 | out: |
39279cc3 CM |
5315 | btrfs_free_path(path); |
5316 | return ret; | |
3954401f CM |
5317 | out_err: |
5318 | location->objectid = 0; | |
5319 | goto out; | |
39279cc3 CM |
5320 | } |
5321 | ||
5322 | /* | |
5323 | * when we hit a tree root in a directory, the btrfs part of the inode | |
5324 | * needs to be changed to reflect the root directory of the tree root. This | |
5325 | * is kind of like crossing a mount point. | |
5326 | */ | |
5327 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
5328 | struct inode *dir, |
5329 | struct dentry *dentry, | |
5330 | struct btrfs_key *location, | |
5331 | struct btrfs_root **sub_root) | |
39279cc3 | 5332 | { |
4df27c4d YZ |
5333 | struct btrfs_path *path; |
5334 | struct btrfs_root *new_root; | |
5335 | struct btrfs_root_ref *ref; | |
5336 | struct extent_buffer *leaf; | |
1d4c08e0 | 5337 | struct btrfs_key key; |
4df27c4d YZ |
5338 | int ret; |
5339 | int err = 0; | |
39279cc3 | 5340 | |
4df27c4d YZ |
5341 | path = btrfs_alloc_path(); |
5342 | if (!path) { | |
5343 | err = -ENOMEM; | |
5344 | goto out; | |
5345 | } | |
39279cc3 | 5346 | |
4df27c4d | 5347 | err = -ENOENT; |
1d4c08e0 DS |
5348 | key.objectid = BTRFS_I(dir)->root->root_key.objectid; |
5349 | key.type = BTRFS_ROOT_REF_KEY; | |
5350 | key.offset = location->objectid; | |
5351 | ||
5352 | ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path, | |
5353 | 0, 0); | |
4df27c4d YZ |
5354 | if (ret) { |
5355 | if (ret < 0) | |
5356 | err = ret; | |
5357 | goto out; | |
5358 | } | |
39279cc3 | 5359 | |
4df27c4d YZ |
5360 | leaf = path->nodes[0]; |
5361 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
33345d01 | 5362 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) || |
4df27c4d YZ |
5363 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
5364 | goto out; | |
39279cc3 | 5365 | |
4df27c4d YZ |
5366 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
5367 | (unsigned long)(ref + 1), | |
5368 | dentry->d_name.len); | |
5369 | if (ret) | |
5370 | goto out; | |
5371 | ||
b3b4aa74 | 5372 | btrfs_release_path(path); |
4df27c4d YZ |
5373 | |
5374 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
5375 | if (IS_ERR(new_root)) { | |
5376 | err = PTR_ERR(new_root); | |
5377 | goto out; | |
5378 | } | |
5379 | ||
4df27c4d YZ |
5380 | *sub_root = new_root; |
5381 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
5382 | location->type = BTRFS_INODE_ITEM_KEY; | |
5383 | location->offset = 0; | |
5384 | err = 0; | |
5385 | out: | |
5386 | btrfs_free_path(path); | |
5387 | return err; | |
39279cc3 CM |
5388 | } |
5389 | ||
5d4f98a2 YZ |
5390 | static void inode_tree_add(struct inode *inode) |
5391 | { | |
5392 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5393 | struct btrfs_inode *entry; | |
03e860bd FNP |
5394 | struct rb_node **p; |
5395 | struct rb_node *parent; | |
cef21937 | 5396 | struct rb_node *new = &BTRFS_I(inode)->rb_node; |
33345d01 | 5397 | u64 ino = btrfs_ino(inode); |
5d4f98a2 | 5398 | |
1d3382cb | 5399 | if (inode_unhashed(inode)) |
76dda93c | 5400 | return; |
e1409cef | 5401 | parent = NULL; |
5d4f98a2 | 5402 | spin_lock(&root->inode_lock); |
e1409cef | 5403 | p = &root->inode_tree.rb_node; |
5d4f98a2 YZ |
5404 | while (*p) { |
5405 | parent = *p; | |
5406 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
5407 | ||
33345d01 | 5408 | if (ino < btrfs_ino(&entry->vfs_inode)) |
03e860bd | 5409 | p = &parent->rb_left; |
33345d01 | 5410 | else if (ino > btrfs_ino(&entry->vfs_inode)) |
03e860bd | 5411 | p = &parent->rb_right; |
5d4f98a2 YZ |
5412 | else { |
5413 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 5414 | (I_WILL_FREE | I_FREEING))); |
cef21937 | 5415 | rb_replace_node(parent, new, &root->inode_tree); |
03e860bd FNP |
5416 | RB_CLEAR_NODE(parent); |
5417 | spin_unlock(&root->inode_lock); | |
cef21937 | 5418 | return; |
5d4f98a2 YZ |
5419 | } |
5420 | } | |
cef21937 FDBM |
5421 | rb_link_node(new, parent, p); |
5422 | rb_insert_color(new, &root->inode_tree); | |
5d4f98a2 YZ |
5423 | spin_unlock(&root->inode_lock); |
5424 | } | |
5425 | ||
5426 | static void inode_tree_del(struct inode *inode) | |
5427 | { | |
5428 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 5429 | int empty = 0; |
5d4f98a2 | 5430 | |
03e860bd | 5431 | spin_lock(&root->inode_lock); |
5d4f98a2 | 5432 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 5433 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 5434 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 5435 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 5436 | } |
03e860bd | 5437 | spin_unlock(&root->inode_lock); |
76dda93c | 5438 | |
69e9c6c6 | 5439 | if (empty && btrfs_root_refs(&root->root_item) == 0) { |
76dda93c YZ |
5440 | synchronize_srcu(&root->fs_info->subvol_srcu); |
5441 | spin_lock(&root->inode_lock); | |
5442 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
5443 | spin_unlock(&root->inode_lock); | |
5444 | if (empty) | |
5445 | btrfs_add_dead_root(root); | |
5446 | } | |
5447 | } | |
5448 | ||
143bede5 | 5449 | void btrfs_invalidate_inodes(struct btrfs_root *root) |
76dda93c YZ |
5450 | { |
5451 | struct rb_node *node; | |
5452 | struct rb_node *prev; | |
5453 | struct btrfs_inode *entry; | |
5454 | struct inode *inode; | |
5455 | u64 objectid = 0; | |
5456 | ||
7813b3db LB |
5457 | if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
5458 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
76dda93c YZ |
5459 | |
5460 | spin_lock(&root->inode_lock); | |
5461 | again: | |
5462 | node = root->inode_tree.rb_node; | |
5463 | prev = NULL; | |
5464 | while (node) { | |
5465 | prev = node; | |
5466 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
5467 | ||
33345d01 | 5468 | if (objectid < btrfs_ino(&entry->vfs_inode)) |
76dda93c | 5469 | node = node->rb_left; |
33345d01 | 5470 | else if (objectid > btrfs_ino(&entry->vfs_inode)) |
76dda93c YZ |
5471 | node = node->rb_right; |
5472 | else | |
5473 | break; | |
5474 | } | |
5475 | if (!node) { | |
5476 | while (prev) { | |
5477 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
33345d01 | 5478 | if (objectid <= btrfs_ino(&entry->vfs_inode)) { |
76dda93c YZ |
5479 | node = prev; |
5480 | break; | |
5481 | } | |
5482 | prev = rb_next(prev); | |
5483 | } | |
5484 | } | |
5485 | while (node) { | |
5486 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
33345d01 | 5487 | objectid = btrfs_ino(&entry->vfs_inode) + 1; |
76dda93c YZ |
5488 | inode = igrab(&entry->vfs_inode); |
5489 | if (inode) { | |
5490 | spin_unlock(&root->inode_lock); | |
5491 | if (atomic_read(&inode->i_count) > 1) | |
5492 | d_prune_aliases(inode); | |
5493 | /* | |
45321ac5 | 5494 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
5495 | * the inode cache when its usage count |
5496 | * hits zero. | |
5497 | */ | |
5498 | iput(inode); | |
5499 | cond_resched(); | |
5500 | spin_lock(&root->inode_lock); | |
5501 | goto again; | |
5502 | } | |
5503 | ||
5504 | if (cond_resched_lock(&root->inode_lock)) | |
5505 | goto again; | |
5506 | ||
5507 | node = rb_next(node); | |
5508 | } | |
5509 | spin_unlock(&root->inode_lock); | |
5d4f98a2 YZ |
5510 | } |
5511 | ||
e02119d5 CM |
5512 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
5513 | { | |
5514 | struct btrfs_iget_args *args = p; | |
90d3e592 CM |
5515 | inode->i_ino = args->location->objectid; |
5516 | memcpy(&BTRFS_I(inode)->location, args->location, | |
5517 | sizeof(*args->location)); | |
e02119d5 | 5518 | BTRFS_I(inode)->root = args->root; |
39279cc3 CM |
5519 | return 0; |
5520 | } | |
5521 | ||
5522 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
5523 | { | |
5524 | struct btrfs_iget_args *args = opaque; | |
90d3e592 | 5525 | return args->location->objectid == BTRFS_I(inode)->location.objectid && |
d397712b | 5526 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
5527 | } |
5528 | ||
5d4f98a2 | 5529 | static struct inode *btrfs_iget_locked(struct super_block *s, |
90d3e592 | 5530 | struct btrfs_key *location, |
5d4f98a2 | 5531 | struct btrfs_root *root) |
39279cc3 CM |
5532 | { |
5533 | struct inode *inode; | |
5534 | struct btrfs_iget_args args; | |
90d3e592 | 5535 | unsigned long hashval = btrfs_inode_hash(location->objectid, root); |
778ba82b | 5536 | |
90d3e592 | 5537 | args.location = location; |
39279cc3 CM |
5538 | args.root = root; |
5539 | ||
778ba82b | 5540 | inode = iget5_locked(s, hashval, btrfs_find_actor, |
39279cc3 CM |
5541 | btrfs_init_locked_inode, |
5542 | (void *)&args); | |
5543 | return inode; | |
5544 | } | |
5545 | ||
1a54ef8c BR |
5546 | /* Get an inode object given its location and corresponding root. |
5547 | * Returns in *is_new if the inode was read from disk | |
5548 | */ | |
5549 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 5550 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
5551 | { |
5552 | struct inode *inode; | |
5553 | ||
90d3e592 | 5554 | inode = btrfs_iget_locked(s, location, root); |
1a54ef8c | 5555 | if (!inode) |
5d4f98a2 | 5556 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
5557 | |
5558 | if (inode->i_state & I_NEW) { | |
1a54ef8c | 5559 | btrfs_read_locked_inode(inode); |
1748f843 MF |
5560 | if (!is_bad_inode(inode)) { |
5561 | inode_tree_add(inode); | |
5562 | unlock_new_inode(inode); | |
5563 | if (new) | |
5564 | *new = 1; | |
5565 | } else { | |
e0b6d65b ST |
5566 | unlock_new_inode(inode); |
5567 | iput(inode); | |
5568 | inode = ERR_PTR(-ESTALE); | |
1748f843 MF |
5569 | } |
5570 | } | |
5571 | ||
1a54ef8c BR |
5572 | return inode; |
5573 | } | |
5574 | ||
4df27c4d YZ |
5575 | static struct inode *new_simple_dir(struct super_block *s, |
5576 | struct btrfs_key *key, | |
5577 | struct btrfs_root *root) | |
5578 | { | |
5579 | struct inode *inode = new_inode(s); | |
5580 | ||
5581 | if (!inode) | |
5582 | return ERR_PTR(-ENOMEM); | |
5583 | ||
4df27c4d YZ |
5584 | BTRFS_I(inode)->root = root; |
5585 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
72ac3c0d | 5586 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); |
4df27c4d YZ |
5587 | |
5588 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
848cce0d | 5589 | inode->i_op = &btrfs_dir_ro_inode_operations; |
4df27c4d YZ |
5590 | inode->i_fop = &simple_dir_operations; |
5591 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
9cc97d64 | 5592 | inode->i_mtime = CURRENT_TIME; |
5593 | inode->i_atime = inode->i_mtime; | |
5594 | inode->i_ctime = inode->i_mtime; | |
5595 | BTRFS_I(inode)->i_otime = inode->i_mtime; | |
4df27c4d YZ |
5596 | |
5597 | return inode; | |
5598 | } | |
5599 | ||
3de4586c | 5600 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 5601 | { |
d397712b | 5602 | struct inode *inode; |
4df27c4d | 5603 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
5604 | struct btrfs_root *sub_root = root; |
5605 | struct btrfs_key location; | |
76dda93c | 5606 | int index; |
b4aff1f8 | 5607 | int ret = 0; |
39279cc3 CM |
5608 | |
5609 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
5610 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 5611 | |
39e3c955 | 5612 | ret = btrfs_inode_by_name(dir, dentry, &location); |
39279cc3 CM |
5613 | if (ret < 0) |
5614 | return ERR_PTR(ret); | |
5f39d397 | 5615 | |
4df27c4d | 5616 | if (location.objectid == 0) |
5662344b | 5617 | return ERR_PTR(-ENOENT); |
4df27c4d YZ |
5618 | |
5619 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 5620 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
5621 | return inode; |
5622 | } | |
5623 | ||
5624 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
5625 | ||
76dda93c | 5626 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
5627 | ret = fixup_tree_root_location(root, dir, dentry, |
5628 | &location, &sub_root); | |
5629 | if (ret < 0) { | |
5630 | if (ret != -ENOENT) | |
5631 | inode = ERR_PTR(ret); | |
5632 | else | |
5633 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
5634 | } else { | |
73f73415 | 5635 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 5636 | } |
76dda93c YZ |
5637 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
5638 | ||
34d19bad | 5639 | if (!IS_ERR(inode) && root != sub_root) { |
c71bf099 YZ |
5640 | down_read(&root->fs_info->cleanup_work_sem); |
5641 | if (!(inode->i_sb->s_flags & MS_RDONLY)) | |
66b4ffd1 | 5642 | ret = btrfs_orphan_cleanup(sub_root); |
c71bf099 | 5643 | up_read(&root->fs_info->cleanup_work_sem); |
01cd3367 JB |
5644 | if (ret) { |
5645 | iput(inode); | |
66b4ffd1 | 5646 | inode = ERR_PTR(ret); |
01cd3367 | 5647 | } |
c71bf099 YZ |
5648 | } |
5649 | ||
3de4586c CM |
5650 | return inode; |
5651 | } | |
5652 | ||
fe15ce44 | 5653 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
5654 | { |
5655 | struct btrfs_root *root; | |
2b0143b5 | 5656 | struct inode *inode = d_inode(dentry); |
76dda93c | 5657 | |
848cce0d | 5658 | if (!inode && !IS_ROOT(dentry)) |
2b0143b5 | 5659 | inode = d_inode(dentry->d_parent); |
76dda93c | 5660 | |
848cce0d LZ |
5661 | if (inode) { |
5662 | root = BTRFS_I(inode)->root; | |
efefb143 YZ |
5663 | if (btrfs_root_refs(&root->root_item) == 0) |
5664 | return 1; | |
848cce0d LZ |
5665 | |
5666 | if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) | |
5667 | return 1; | |
efefb143 | 5668 | } |
76dda93c YZ |
5669 | return 0; |
5670 | } | |
5671 | ||
b4aff1f8 JB |
5672 | static void btrfs_dentry_release(struct dentry *dentry) |
5673 | { | |
944a4515 | 5674 | kfree(dentry->d_fsdata); |
b4aff1f8 JB |
5675 | } |
5676 | ||
3de4586c | 5677 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
00cd8dd3 | 5678 | unsigned int flags) |
3de4586c | 5679 | { |
5662344b | 5680 | struct inode *inode; |
a66e7cc6 | 5681 | |
5662344b TI |
5682 | inode = btrfs_lookup_dentry(dir, dentry); |
5683 | if (IS_ERR(inode)) { | |
5684 | if (PTR_ERR(inode) == -ENOENT) | |
5685 | inode = NULL; | |
5686 | else | |
5687 | return ERR_CAST(inode); | |
5688 | } | |
5689 | ||
41d28bca | 5690 | return d_splice_alias(inode, dentry); |
39279cc3 CM |
5691 | } |
5692 | ||
16cdcec7 | 5693 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
5694 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
5695 | }; | |
5696 | ||
9cdda8d3 | 5697 | static int btrfs_real_readdir(struct file *file, struct dir_context *ctx) |
39279cc3 | 5698 | { |
9cdda8d3 | 5699 | struct inode *inode = file_inode(file); |
39279cc3 CM |
5700 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5701 | struct btrfs_item *item; | |
5702 | struct btrfs_dir_item *di; | |
5703 | struct btrfs_key key; | |
5f39d397 | 5704 | struct btrfs_key found_key; |
39279cc3 | 5705 | struct btrfs_path *path; |
16cdcec7 MX |
5706 | struct list_head ins_list; |
5707 | struct list_head del_list; | |
39279cc3 | 5708 | int ret; |
5f39d397 | 5709 | struct extent_buffer *leaf; |
39279cc3 | 5710 | int slot; |
39279cc3 CM |
5711 | unsigned char d_type; |
5712 | int over = 0; | |
5713 | u32 di_cur; | |
5714 | u32 di_total; | |
5715 | u32 di_len; | |
5716 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
5717 | char tmp_name[32]; |
5718 | char *name_ptr; | |
5719 | int name_len; | |
9cdda8d3 | 5720 | int is_curr = 0; /* ctx->pos points to the current index? */ |
39279cc3 CM |
5721 | |
5722 | /* FIXME, use a real flag for deciding about the key type */ | |
5723 | if (root->fs_info->tree_root == root) | |
5724 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 5725 | |
9cdda8d3 AV |
5726 | if (!dir_emit_dots(file, ctx)) |
5727 | return 0; | |
5728 | ||
49593bfa | 5729 | path = btrfs_alloc_path(); |
16cdcec7 MX |
5730 | if (!path) |
5731 | return -ENOMEM; | |
ff5714cc | 5732 | |
026fd317 | 5733 | path->reada = 1; |
49593bfa | 5734 | |
16cdcec7 MX |
5735 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
5736 | INIT_LIST_HEAD(&ins_list); | |
5737 | INIT_LIST_HEAD(&del_list); | |
5738 | btrfs_get_delayed_items(inode, &ins_list, &del_list); | |
5739 | } | |
5740 | ||
962a298f | 5741 | key.type = key_type; |
9cdda8d3 | 5742 | key.offset = ctx->pos; |
33345d01 | 5743 | key.objectid = btrfs_ino(inode); |
5f39d397 | 5744 | |
39279cc3 CM |
5745 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5746 | if (ret < 0) | |
5747 | goto err; | |
49593bfa DW |
5748 | |
5749 | while (1) { | |
5f39d397 | 5750 | leaf = path->nodes[0]; |
39279cc3 | 5751 | slot = path->slots[0]; |
b9e03af0 LZ |
5752 | if (slot >= btrfs_header_nritems(leaf)) { |
5753 | ret = btrfs_next_leaf(root, path); | |
5754 | if (ret < 0) | |
5755 | goto err; | |
5756 | else if (ret > 0) | |
5757 | break; | |
5758 | continue; | |
39279cc3 | 5759 | } |
3de4586c | 5760 | |
dd3cc16b | 5761 | item = btrfs_item_nr(slot); |
5f39d397 CM |
5762 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
5763 | ||
5764 | if (found_key.objectid != key.objectid) | |
39279cc3 | 5765 | break; |
962a298f | 5766 | if (found_key.type != key_type) |
39279cc3 | 5767 | break; |
9cdda8d3 | 5768 | if (found_key.offset < ctx->pos) |
b9e03af0 | 5769 | goto next; |
16cdcec7 MX |
5770 | if (key_type == BTRFS_DIR_INDEX_KEY && |
5771 | btrfs_should_delete_dir_index(&del_list, | |
5772 | found_key.offset)) | |
5773 | goto next; | |
5f39d397 | 5774 | |
9cdda8d3 | 5775 | ctx->pos = found_key.offset; |
16cdcec7 | 5776 | is_curr = 1; |
49593bfa | 5777 | |
39279cc3 CM |
5778 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
5779 | di_cur = 0; | |
5f39d397 | 5780 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
5781 | |
5782 | while (di_cur < di_total) { | |
5f39d397 CM |
5783 | struct btrfs_key location; |
5784 | ||
22a94d44 JB |
5785 | if (verify_dir_item(root, leaf, di)) |
5786 | break; | |
5787 | ||
5f39d397 | 5788 | name_len = btrfs_dir_name_len(leaf, di); |
49593bfa | 5789 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
5790 | name_ptr = tmp_name; |
5791 | } else { | |
5792 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
5793 | if (!name_ptr) { |
5794 | ret = -ENOMEM; | |
5795 | goto err; | |
5796 | } | |
5f39d397 CM |
5797 | } |
5798 | read_extent_buffer(leaf, name_ptr, | |
5799 | (unsigned long)(di + 1), name_len); | |
5800 | ||
5801 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
5802 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c | 5803 | |
fede766f | 5804 | |
3de4586c | 5805 | /* is this a reference to our own snapshot? If so |
8c9c2bf7 AJ |
5806 | * skip it. |
5807 | * | |
5808 | * In contrast to old kernels, we insert the snapshot's | |
5809 | * dir item and dir index after it has been created, so | |
5810 | * we won't find a reference to our own snapshot. We | |
5811 | * still keep the following code for backward | |
5812 | * compatibility. | |
3de4586c CM |
5813 | */ |
5814 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
5815 | location.objectid == root->root_key.objectid) { | |
5816 | over = 0; | |
5817 | goto skip; | |
5818 | } | |
9cdda8d3 AV |
5819 | over = !dir_emit(ctx, name_ptr, name_len, |
5820 | location.objectid, d_type); | |
5f39d397 | 5821 | |
3de4586c | 5822 | skip: |
5f39d397 CM |
5823 | if (name_ptr != tmp_name) |
5824 | kfree(name_ptr); | |
5825 | ||
39279cc3 CM |
5826 | if (over) |
5827 | goto nopos; | |
5103e947 | 5828 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 5829 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
5830 | di_cur += di_len; |
5831 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
5832 | } | |
b9e03af0 LZ |
5833 | next: |
5834 | path->slots[0]++; | |
39279cc3 | 5835 | } |
49593bfa | 5836 | |
16cdcec7 MX |
5837 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
5838 | if (is_curr) | |
9cdda8d3 AV |
5839 | ctx->pos++; |
5840 | ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list); | |
16cdcec7 MX |
5841 | if (ret) |
5842 | goto nopos; | |
5843 | } | |
5844 | ||
49593bfa | 5845 | /* Reached end of directory/root. Bump pos past the last item. */ |
db62efbb ZB |
5846 | ctx->pos++; |
5847 | ||
5848 | /* | |
5849 | * Stop new entries from being returned after we return the last | |
5850 | * entry. | |
5851 | * | |
5852 | * New directory entries are assigned a strictly increasing | |
5853 | * offset. This means that new entries created during readdir | |
5854 | * are *guaranteed* to be seen in the future by that readdir. | |
5855 | * This has broken buggy programs which operate on names as | |
5856 | * they're returned by readdir. Until we re-use freed offsets | |
5857 | * we have this hack to stop new entries from being returned | |
5858 | * under the assumption that they'll never reach this huge | |
5859 | * offset. | |
5860 | * | |
5861 | * This is being careful not to overflow 32bit loff_t unless the | |
5862 | * last entry requires it because doing so has broken 32bit apps | |
5863 | * in the past. | |
5864 | */ | |
5865 | if (key_type == BTRFS_DIR_INDEX_KEY) { | |
5866 | if (ctx->pos >= INT_MAX) | |
5867 | ctx->pos = LLONG_MAX; | |
5868 | else | |
5869 | ctx->pos = INT_MAX; | |
5870 | } | |
39279cc3 CM |
5871 | nopos: |
5872 | ret = 0; | |
5873 | err: | |
16cdcec7 MX |
5874 | if (key_type == BTRFS_DIR_INDEX_KEY) |
5875 | btrfs_put_delayed_items(&ins_list, &del_list); | |
39279cc3 | 5876 | btrfs_free_path(path); |
39279cc3 CM |
5877 | return ret; |
5878 | } | |
5879 | ||
a9185b41 | 5880 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
5881 | { |
5882 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5883 | struct btrfs_trans_handle *trans; | |
5884 | int ret = 0; | |
0af3d00b | 5885 | bool nolock = false; |
39279cc3 | 5886 | |
72ac3c0d | 5887 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
4ca8b41e CM |
5888 | return 0; |
5889 | ||
83eea1f1 | 5890 | if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode)) |
82d5902d | 5891 | nolock = true; |
0af3d00b | 5892 | |
a9185b41 | 5893 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b | 5894 | if (nolock) |
7a7eaa40 | 5895 | trans = btrfs_join_transaction_nolock(root); |
0af3d00b | 5896 | else |
7a7eaa40 | 5897 | trans = btrfs_join_transaction(root); |
3612b495 TI |
5898 | if (IS_ERR(trans)) |
5899 | return PTR_ERR(trans); | |
a698d075 | 5900 | ret = btrfs_commit_transaction(trans, root); |
39279cc3 CM |
5901 | } |
5902 | return ret; | |
5903 | } | |
5904 | ||
5905 | /* | |
54aa1f4d | 5906 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
5907 | * inode changes. But, it is most likely to find the inode in cache. |
5908 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
5909 | * to keep or drop this code. | |
5910 | */ | |
48a3b636 | 5911 | static int btrfs_dirty_inode(struct inode *inode) |
39279cc3 CM |
5912 | { |
5913 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5914 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
5915 | int ret; |
5916 | ||
72ac3c0d | 5917 | if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags)) |
22c44fe6 | 5918 | return 0; |
39279cc3 | 5919 | |
7a7eaa40 | 5920 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
5921 | if (IS_ERR(trans)) |
5922 | return PTR_ERR(trans); | |
8929ecfa YZ |
5923 | |
5924 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
5925 | if (ret && ret == -ENOSPC) { |
5926 | /* whoops, lets try again with the full transaction */ | |
5927 | btrfs_end_transaction(trans, root); | |
5928 | trans = btrfs_start_transaction(root, 1); | |
22c44fe6 JB |
5929 | if (IS_ERR(trans)) |
5930 | return PTR_ERR(trans); | |
8929ecfa | 5931 | |
94b60442 | 5932 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 5933 | } |
39279cc3 | 5934 | btrfs_end_transaction(trans, root); |
16cdcec7 MX |
5935 | if (BTRFS_I(inode)->delayed_node) |
5936 | btrfs_balance_delayed_items(root); | |
22c44fe6 JB |
5937 | |
5938 | return ret; | |
5939 | } | |
5940 | ||
5941 | /* | |
5942 | * This is a copy of file_update_time. We need this so we can return error on | |
5943 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
5944 | */ | |
e41f941a JB |
5945 | static int btrfs_update_time(struct inode *inode, struct timespec *now, |
5946 | int flags) | |
22c44fe6 | 5947 | { |
2bc55652 AB |
5948 | struct btrfs_root *root = BTRFS_I(inode)->root; |
5949 | ||
5950 | if (btrfs_root_readonly(root)) | |
5951 | return -EROFS; | |
5952 | ||
e41f941a | 5953 | if (flags & S_VERSION) |
22c44fe6 | 5954 | inode_inc_iversion(inode); |
e41f941a JB |
5955 | if (flags & S_CTIME) |
5956 | inode->i_ctime = *now; | |
5957 | if (flags & S_MTIME) | |
5958 | inode->i_mtime = *now; | |
5959 | if (flags & S_ATIME) | |
5960 | inode->i_atime = *now; | |
5961 | return btrfs_dirty_inode(inode); | |
39279cc3 CM |
5962 | } |
5963 | ||
d352ac68 CM |
5964 | /* |
5965 | * find the highest existing sequence number in a directory | |
5966 | * and then set the in-memory index_cnt variable to reflect | |
5967 | * free sequence numbers | |
5968 | */ | |
aec7477b JB |
5969 | static int btrfs_set_inode_index_count(struct inode *inode) |
5970 | { | |
5971 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5972 | struct btrfs_key key, found_key; | |
5973 | struct btrfs_path *path; | |
5974 | struct extent_buffer *leaf; | |
5975 | int ret; | |
5976 | ||
33345d01 | 5977 | key.objectid = btrfs_ino(inode); |
962a298f | 5978 | key.type = BTRFS_DIR_INDEX_KEY; |
aec7477b JB |
5979 | key.offset = (u64)-1; |
5980 | ||
5981 | path = btrfs_alloc_path(); | |
5982 | if (!path) | |
5983 | return -ENOMEM; | |
5984 | ||
5985 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
5986 | if (ret < 0) | |
5987 | goto out; | |
5988 | /* FIXME: we should be able to handle this */ | |
5989 | if (ret == 0) | |
5990 | goto out; | |
5991 | ret = 0; | |
5992 | ||
5993 | /* | |
5994 | * MAGIC NUMBER EXPLANATION: | |
5995 | * since we search a directory based on f_pos we have to start at 2 | |
5996 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
5997 | * else has to start at 2 | |
5998 | */ | |
5999 | if (path->slots[0] == 0) { | |
6000 | BTRFS_I(inode)->index_cnt = 2; | |
6001 | goto out; | |
6002 | } | |
6003 | ||
6004 | path->slots[0]--; | |
6005 | ||
6006 | leaf = path->nodes[0]; | |
6007 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
6008 | ||
33345d01 | 6009 | if (found_key.objectid != btrfs_ino(inode) || |
962a298f | 6010 | found_key.type != BTRFS_DIR_INDEX_KEY) { |
aec7477b JB |
6011 | BTRFS_I(inode)->index_cnt = 2; |
6012 | goto out; | |
6013 | } | |
6014 | ||
6015 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
6016 | out: | |
6017 | btrfs_free_path(path); | |
6018 | return ret; | |
6019 | } | |
6020 | ||
d352ac68 CM |
6021 | /* |
6022 | * helper to find a free sequence number in a given directory. This current | |
6023 | * code is very simple, later versions will do smarter things in the btree | |
6024 | */ | |
3de4586c | 6025 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
6026 | { |
6027 | int ret = 0; | |
6028 | ||
6029 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
16cdcec7 MX |
6030 | ret = btrfs_inode_delayed_dir_index_count(dir); |
6031 | if (ret) { | |
6032 | ret = btrfs_set_inode_index_count(dir); | |
6033 | if (ret) | |
6034 | return ret; | |
6035 | } | |
aec7477b JB |
6036 | } |
6037 | ||
00e4e6b3 | 6038 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
6039 | BTRFS_I(dir)->index_cnt++; |
6040 | ||
6041 | return ret; | |
6042 | } | |
6043 | ||
b0d5d10f CM |
6044 | static int btrfs_insert_inode_locked(struct inode *inode) |
6045 | { | |
6046 | struct btrfs_iget_args args; | |
6047 | args.location = &BTRFS_I(inode)->location; | |
6048 | args.root = BTRFS_I(inode)->root; | |
6049 | ||
6050 | return insert_inode_locked4(inode, | |
6051 | btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root), | |
6052 | btrfs_find_actor, &args); | |
6053 | } | |
6054 | ||
39279cc3 CM |
6055 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
6056 | struct btrfs_root *root, | |
aec7477b | 6057 | struct inode *dir, |
9c58309d | 6058 | const char *name, int name_len, |
175a4eb7 AV |
6059 | u64 ref_objectid, u64 objectid, |
6060 | umode_t mode, u64 *index) | |
39279cc3 CM |
6061 | { |
6062 | struct inode *inode; | |
5f39d397 | 6063 | struct btrfs_inode_item *inode_item; |
39279cc3 | 6064 | struct btrfs_key *location; |
5f39d397 | 6065 | struct btrfs_path *path; |
9c58309d CM |
6066 | struct btrfs_inode_ref *ref; |
6067 | struct btrfs_key key[2]; | |
6068 | u32 sizes[2]; | |
ef3b9af5 | 6069 | int nitems = name ? 2 : 1; |
9c58309d | 6070 | unsigned long ptr; |
39279cc3 | 6071 | int ret; |
39279cc3 | 6072 | |
5f39d397 | 6073 | path = btrfs_alloc_path(); |
d8926bb3 MF |
6074 | if (!path) |
6075 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 6076 | |
39279cc3 | 6077 | inode = new_inode(root->fs_info->sb); |
8fb27640 YS |
6078 | if (!inode) { |
6079 | btrfs_free_path(path); | |
39279cc3 | 6080 | return ERR_PTR(-ENOMEM); |
8fb27640 | 6081 | } |
39279cc3 | 6082 | |
5762b5c9 FM |
6083 | /* |
6084 | * O_TMPFILE, set link count to 0, so that after this point, | |
6085 | * we fill in an inode item with the correct link count. | |
6086 | */ | |
6087 | if (!name) | |
6088 | set_nlink(inode, 0); | |
6089 | ||
581bb050 LZ |
6090 | /* |
6091 | * we have to initialize this early, so we can reclaim the inode | |
6092 | * number if we fail afterwards in this function. | |
6093 | */ | |
6094 | inode->i_ino = objectid; | |
6095 | ||
ef3b9af5 | 6096 | if (dir && name) { |
1abe9b8a | 6097 | trace_btrfs_inode_request(dir); |
6098 | ||
3de4586c | 6099 | ret = btrfs_set_inode_index(dir, index); |
09771430 | 6100 | if (ret) { |
8fb27640 | 6101 | btrfs_free_path(path); |
09771430 | 6102 | iput(inode); |
aec7477b | 6103 | return ERR_PTR(ret); |
09771430 | 6104 | } |
ef3b9af5 FM |
6105 | } else if (dir) { |
6106 | *index = 0; | |
aec7477b JB |
6107 | } |
6108 | /* | |
6109 | * index_cnt is ignored for everything but a dir, | |
6110 | * btrfs_get_inode_index_count has an explanation for the magic | |
6111 | * number | |
6112 | */ | |
6113 | BTRFS_I(inode)->index_cnt = 2; | |
67de1176 | 6114 | BTRFS_I(inode)->dir_index = *index; |
39279cc3 | 6115 | BTRFS_I(inode)->root = root; |
e02119d5 | 6116 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 6117 | inode->i_generation = BTRFS_I(inode)->generation; |
b888db2b | 6118 | |
5dc562c5 JB |
6119 | /* |
6120 | * We could have gotten an inode number from somebody who was fsynced | |
6121 | * and then removed in this same transaction, so let's just set full | |
6122 | * sync since it will be a full sync anyway and this will blow away the | |
6123 | * old info in the log. | |
6124 | */ | |
6125 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
6126 | ||
9c58309d | 6127 | key[0].objectid = objectid; |
962a298f | 6128 | key[0].type = BTRFS_INODE_ITEM_KEY; |
9c58309d CM |
6129 | key[0].offset = 0; |
6130 | ||
9c58309d | 6131 | sizes[0] = sizeof(struct btrfs_inode_item); |
ef3b9af5 FM |
6132 | |
6133 | if (name) { | |
6134 | /* | |
6135 | * Start new inodes with an inode_ref. This is slightly more | |
6136 | * efficient for small numbers of hard links since they will | |
6137 | * be packed into one item. Extended refs will kick in if we | |
6138 | * add more hard links than can fit in the ref item. | |
6139 | */ | |
6140 | key[1].objectid = objectid; | |
962a298f | 6141 | key[1].type = BTRFS_INODE_REF_KEY; |
ef3b9af5 FM |
6142 | key[1].offset = ref_objectid; |
6143 | ||
6144 | sizes[1] = name_len + sizeof(*ref); | |
6145 | } | |
9c58309d | 6146 | |
b0d5d10f CM |
6147 | location = &BTRFS_I(inode)->location; |
6148 | location->objectid = objectid; | |
6149 | location->offset = 0; | |
962a298f | 6150 | location->type = BTRFS_INODE_ITEM_KEY; |
b0d5d10f CM |
6151 | |
6152 | ret = btrfs_insert_inode_locked(inode); | |
6153 | if (ret < 0) | |
6154 | goto fail; | |
6155 | ||
b9473439 | 6156 | path->leave_spinning = 1; |
ef3b9af5 | 6157 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems); |
9c58309d | 6158 | if (ret != 0) |
b0d5d10f | 6159 | goto fail_unlock; |
5f39d397 | 6160 | |
ecc11fab | 6161 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 6162 | inode_set_bytes(inode, 0); |
9cc97d64 | 6163 | |
6164 | inode->i_mtime = CURRENT_TIME; | |
6165 | inode->i_atime = inode->i_mtime; | |
6166 | inode->i_ctime = inode->i_mtime; | |
6167 | BTRFS_I(inode)->i_otime = inode->i_mtime; | |
6168 | ||
5f39d397 CM |
6169 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
6170 | struct btrfs_inode_item); | |
293f7e07 LZ |
6171 | memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item, |
6172 | sizeof(*inode_item)); | |
e02119d5 | 6173 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d | 6174 | |
ef3b9af5 FM |
6175 | if (name) { |
6176 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
6177 | struct btrfs_inode_ref); | |
6178 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
6179 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); | |
6180 | ptr = (unsigned long)(ref + 1); | |
6181 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
6182 | } | |
9c58309d | 6183 | |
5f39d397 CM |
6184 | btrfs_mark_buffer_dirty(path->nodes[0]); |
6185 | btrfs_free_path(path); | |
6186 | ||
6cbff00f CH |
6187 | btrfs_inherit_iflags(inode, dir); |
6188 | ||
569254b0 | 6189 | if (S_ISREG(mode)) { |
94272164 CM |
6190 | if (btrfs_test_opt(root, NODATASUM)) |
6191 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
213490b3 | 6192 | if (btrfs_test_opt(root, NODATACOW)) |
f2bdf9a8 JB |
6193 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW | |
6194 | BTRFS_INODE_NODATASUM; | |
94272164 CM |
6195 | } |
6196 | ||
5d4f98a2 | 6197 | inode_tree_add(inode); |
1abe9b8a | 6198 | |
6199 | trace_btrfs_inode_new(inode); | |
1973f0fa | 6200 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 6201 | |
8ea05e3a AB |
6202 | btrfs_update_root_times(trans, root); |
6203 | ||
63541927 FDBM |
6204 | ret = btrfs_inode_inherit_props(trans, inode, dir); |
6205 | if (ret) | |
6206 | btrfs_err(root->fs_info, | |
6207 | "error inheriting props for ino %llu (root %llu): %d", | |
6208 | btrfs_ino(inode), root->root_key.objectid, ret); | |
6209 | ||
39279cc3 | 6210 | return inode; |
b0d5d10f CM |
6211 | |
6212 | fail_unlock: | |
6213 | unlock_new_inode(inode); | |
5f39d397 | 6214 | fail: |
ef3b9af5 | 6215 | if (dir && name) |
aec7477b | 6216 | BTRFS_I(dir)->index_cnt--; |
5f39d397 | 6217 | btrfs_free_path(path); |
09771430 | 6218 | iput(inode); |
5f39d397 | 6219 | return ERR_PTR(ret); |
39279cc3 CM |
6220 | } |
6221 | ||
6222 | static inline u8 btrfs_inode_type(struct inode *inode) | |
6223 | { | |
6224 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
6225 | } | |
6226 | ||
d352ac68 CM |
6227 | /* |
6228 | * utility function to add 'inode' into 'parent_inode' with | |
6229 | * a give name and a given sequence number. | |
6230 | * if 'add_backref' is true, also insert a backref from the | |
6231 | * inode to the parent directory. | |
6232 | */ | |
e02119d5 CM |
6233 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
6234 | struct inode *parent_inode, struct inode *inode, | |
6235 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 6236 | { |
4df27c4d | 6237 | int ret = 0; |
39279cc3 | 6238 | struct btrfs_key key; |
e02119d5 | 6239 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
33345d01 LZ |
6240 | u64 ino = btrfs_ino(inode); |
6241 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 6242 | |
33345d01 | 6243 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
6244 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); |
6245 | } else { | |
33345d01 | 6246 | key.objectid = ino; |
962a298f | 6247 | key.type = BTRFS_INODE_ITEM_KEY; |
4df27c4d YZ |
6248 | key.offset = 0; |
6249 | } | |
6250 | ||
33345d01 | 6251 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
6252 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, |
6253 | key.objectid, root->root_key.objectid, | |
33345d01 | 6254 | parent_ino, index, name, name_len); |
4df27c4d | 6255 | } else if (add_backref) { |
33345d01 LZ |
6256 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
6257 | parent_ino, index); | |
4df27c4d | 6258 | } |
39279cc3 | 6259 | |
79787eaa JM |
6260 | /* Nothing to clean up yet */ |
6261 | if (ret) | |
6262 | return ret; | |
4df27c4d | 6263 | |
79787eaa JM |
6264 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
6265 | parent_inode, &key, | |
6266 | btrfs_inode_type(inode), index); | |
9c52057c | 6267 | if (ret == -EEXIST || ret == -EOVERFLOW) |
79787eaa JM |
6268 | goto fail_dir_item; |
6269 | else if (ret) { | |
6270 | btrfs_abort_transaction(trans, root, ret); | |
6271 | return ret; | |
39279cc3 | 6272 | } |
79787eaa JM |
6273 | |
6274 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
6275 | name_len * 2); | |
0c4d2d95 | 6276 | inode_inc_iversion(parent_inode); |
79787eaa JM |
6277 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
6278 | ret = btrfs_update_inode(trans, root, parent_inode); | |
6279 | if (ret) | |
6280 | btrfs_abort_transaction(trans, root, ret); | |
39279cc3 | 6281 | return ret; |
fe66a05a CM |
6282 | |
6283 | fail_dir_item: | |
6284 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { | |
6285 | u64 local_index; | |
6286 | int err; | |
6287 | err = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
6288 | key.objectid, root->root_key.objectid, | |
6289 | parent_ino, &local_index, name, name_len); | |
6290 | ||
6291 | } else if (add_backref) { | |
6292 | u64 local_index; | |
6293 | int err; | |
6294 | ||
6295 | err = btrfs_del_inode_ref(trans, root, name, name_len, | |
6296 | ino, parent_ino, &local_index); | |
6297 | } | |
6298 | return ret; | |
39279cc3 CM |
6299 | } |
6300 | ||
6301 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
a1b075d2 JB |
6302 | struct inode *dir, struct dentry *dentry, |
6303 | struct inode *inode, int backref, u64 index) | |
39279cc3 | 6304 | { |
a1b075d2 JB |
6305 | int err = btrfs_add_link(trans, dir, inode, |
6306 | dentry->d_name.name, dentry->d_name.len, | |
6307 | backref, index); | |
39279cc3 CM |
6308 | if (err > 0) |
6309 | err = -EEXIST; | |
6310 | return err; | |
6311 | } | |
6312 | ||
618e21d5 | 6313 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
1a67aafb | 6314 | umode_t mode, dev_t rdev) |
618e21d5 JB |
6315 | { |
6316 | struct btrfs_trans_handle *trans; | |
6317 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6318 | struct inode *inode = NULL; |
618e21d5 JB |
6319 | int err; |
6320 | int drop_inode = 0; | |
6321 | u64 objectid; | |
00e4e6b3 | 6322 | u64 index = 0; |
618e21d5 JB |
6323 | |
6324 | if (!new_valid_dev(rdev)) | |
6325 | return -EINVAL; | |
6326 | ||
9ed74f2d JB |
6327 | /* |
6328 | * 2 for inode item and ref | |
6329 | * 2 for dir items | |
6330 | * 1 for xattr if selinux is on | |
6331 | */ | |
a22285a6 YZ |
6332 | trans = btrfs_start_transaction(root, 5); |
6333 | if (IS_ERR(trans)) | |
6334 | return PTR_ERR(trans); | |
1832a6d5 | 6335 | |
581bb050 LZ |
6336 | err = btrfs_find_free_ino(root, &objectid); |
6337 | if (err) | |
6338 | goto out_unlock; | |
6339 | ||
aec7477b | 6340 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 6341 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 6342 | mode, &index); |
7cf96da3 TI |
6343 | if (IS_ERR(inode)) { |
6344 | err = PTR_ERR(inode); | |
618e21d5 | 6345 | goto out_unlock; |
7cf96da3 | 6346 | } |
618e21d5 | 6347 | |
ad19db71 CS |
6348 | /* |
6349 | * If the active LSM wants to access the inode during | |
6350 | * d_instantiate it needs these. Smack checks to see | |
6351 | * if the filesystem supports xattrs by looking at the | |
6352 | * ops vector. | |
6353 | */ | |
ad19db71 | 6354 | inode->i_op = &btrfs_special_inode_operations; |
b0d5d10f CM |
6355 | init_special_inode(inode, inode->i_mode, rdev); |
6356 | ||
6357 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
618e21d5 | 6358 | if (err) |
b0d5d10f CM |
6359 | goto out_unlock_inode; |
6360 | ||
6361 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); | |
6362 | if (err) { | |
6363 | goto out_unlock_inode; | |
6364 | } else { | |
1b4ab1bb | 6365 | btrfs_update_inode(trans, root, inode); |
b0d5d10f | 6366 | unlock_new_inode(inode); |
08c422c2 | 6367 | d_instantiate(dentry, inode); |
618e21d5 | 6368 | } |
b0d5d10f | 6369 | |
618e21d5 | 6370 | out_unlock: |
7ad85bb7 | 6371 | btrfs_end_transaction(trans, root); |
c581afc8 | 6372 | btrfs_balance_delayed_items(root); |
b53d3f5d | 6373 | btrfs_btree_balance_dirty(root); |
618e21d5 JB |
6374 | if (drop_inode) { |
6375 | inode_dec_link_count(inode); | |
6376 | iput(inode); | |
6377 | } | |
618e21d5 | 6378 | return err; |
b0d5d10f CM |
6379 | |
6380 | out_unlock_inode: | |
6381 | drop_inode = 1; | |
6382 | unlock_new_inode(inode); | |
6383 | goto out_unlock; | |
6384 | ||
618e21d5 JB |
6385 | } |
6386 | ||
39279cc3 | 6387 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
ebfc3b49 | 6388 | umode_t mode, bool excl) |
39279cc3 CM |
6389 | { |
6390 | struct btrfs_trans_handle *trans; | |
6391 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 6392 | struct inode *inode = NULL; |
43baa579 | 6393 | int drop_inode_on_err = 0; |
a22285a6 | 6394 | int err; |
39279cc3 | 6395 | u64 objectid; |
00e4e6b3 | 6396 | u64 index = 0; |
39279cc3 | 6397 | |
9ed74f2d JB |
6398 | /* |
6399 | * 2 for inode item and ref | |
6400 | * 2 for dir items | |
6401 | * 1 for xattr if selinux is on | |
6402 | */ | |
a22285a6 YZ |
6403 | trans = btrfs_start_transaction(root, 5); |
6404 | if (IS_ERR(trans)) | |
6405 | return PTR_ERR(trans); | |
9ed74f2d | 6406 | |
581bb050 LZ |
6407 | err = btrfs_find_free_ino(root, &objectid); |
6408 | if (err) | |
6409 | goto out_unlock; | |
6410 | ||
aec7477b | 6411 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 6412 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 6413 | mode, &index); |
7cf96da3 TI |
6414 | if (IS_ERR(inode)) { |
6415 | err = PTR_ERR(inode); | |
39279cc3 | 6416 | goto out_unlock; |
7cf96da3 | 6417 | } |
43baa579 | 6418 | drop_inode_on_err = 1; |
ad19db71 CS |
6419 | /* |
6420 | * If the active LSM wants to access the inode during | |
6421 | * d_instantiate it needs these. Smack checks to see | |
6422 | * if the filesystem supports xattrs by looking at the | |
6423 | * ops vector. | |
6424 | */ | |
6425 | inode->i_fop = &btrfs_file_operations; | |
6426 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 6427 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
6428 | |
6429 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
6430 | if (err) | |
6431 | goto out_unlock_inode; | |
6432 | ||
6433 | err = btrfs_update_inode(trans, root, inode); | |
6434 | if (err) | |
6435 | goto out_unlock_inode; | |
ad19db71 | 6436 | |
a1b075d2 | 6437 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 | 6438 | if (err) |
b0d5d10f | 6439 | goto out_unlock_inode; |
43baa579 | 6440 | |
43baa579 | 6441 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
b0d5d10f | 6442 | unlock_new_inode(inode); |
43baa579 FB |
6443 | d_instantiate(dentry, inode); |
6444 | ||
39279cc3 | 6445 | out_unlock: |
7ad85bb7 | 6446 | btrfs_end_transaction(trans, root); |
43baa579 | 6447 | if (err && drop_inode_on_err) { |
39279cc3 CM |
6448 | inode_dec_link_count(inode); |
6449 | iput(inode); | |
6450 | } | |
c581afc8 | 6451 | btrfs_balance_delayed_items(root); |
b53d3f5d | 6452 | btrfs_btree_balance_dirty(root); |
39279cc3 | 6453 | return err; |
b0d5d10f CM |
6454 | |
6455 | out_unlock_inode: | |
6456 | unlock_new_inode(inode); | |
6457 | goto out_unlock; | |
6458 | ||
39279cc3 CM |
6459 | } |
6460 | ||
6461 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
6462 | struct dentry *dentry) | |
6463 | { | |
6464 | struct btrfs_trans_handle *trans; | |
6465 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
2b0143b5 | 6466 | struct inode *inode = d_inode(old_dentry); |
00e4e6b3 | 6467 | u64 index; |
39279cc3 CM |
6468 | int err; |
6469 | int drop_inode = 0; | |
6470 | ||
4a8be425 TH |
6471 | /* do not allow sys_link's with other subvols of the same device */ |
6472 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 6473 | return -EXDEV; |
4a8be425 | 6474 | |
f186373f | 6475 | if (inode->i_nlink >= BTRFS_LINK_MAX) |
c055e99e | 6476 | return -EMLINK; |
4a8be425 | 6477 | |
3de4586c | 6478 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
6479 | if (err) |
6480 | goto fail; | |
6481 | ||
a22285a6 | 6482 | /* |
7e6b6465 | 6483 | * 2 items for inode and inode ref |
a22285a6 | 6484 | * 2 items for dir items |
7e6b6465 | 6485 | * 1 item for parent inode |
a22285a6 | 6486 | */ |
7e6b6465 | 6487 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
6488 | if (IS_ERR(trans)) { |
6489 | err = PTR_ERR(trans); | |
6490 | goto fail; | |
6491 | } | |
5f39d397 | 6492 | |
67de1176 MX |
6493 | /* There are several dir indexes for this inode, clear the cache. */ |
6494 | BTRFS_I(inode)->dir_index = 0ULL; | |
8b558c5f | 6495 | inc_nlink(inode); |
0c4d2d95 | 6496 | inode_inc_iversion(inode); |
3153495d | 6497 | inode->i_ctime = CURRENT_TIME; |
7de9c6ee | 6498 | ihold(inode); |
e9976151 | 6499 | set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags); |
aec7477b | 6500 | |
a1b075d2 | 6501 | err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index); |
5f39d397 | 6502 | |
a5719521 | 6503 | if (err) { |
54aa1f4d | 6504 | drop_inode = 1; |
a5719521 | 6505 | } else { |
10d9f309 | 6506 | struct dentry *parent = dentry->d_parent; |
a5719521 | 6507 | err = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
6508 | if (err) |
6509 | goto fail; | |
ef3b9af5 FM |
6510 | if (inode->i_nlink == 1) { |
6511 | /* | |
6512 | * If new hard link count is 1, it's a file created | |
6513 | * with open(2) O_TMPFILE flag. | |
6514 | */ | |
6515 | err = btrfs_orphan_del(trans, inode); | |
6516 | if (err) | |
6517 | goto fail; | |
6518 | } | |
08c422c2 | 6519 | d_instantiate(dentry, inode); |
6a912213 | 6520 | btrfs_log_new_name(trans, inode, NULL, parent); |
a5719521 | 6521 | } |
39279cc3 | 6522 | |
7ad85bb7 | 6523 | btrfs_end_transaction(trans, root); |
c581afc8 | 6524 | btrfs_balance_delayed_items(root); |
1832a6d5 | 6525 | fail: |
39279cc3 CM |
6526 | if (drop_inode) { |
6527 | inode_dec_link_count(inode); | |
6528 | iput(inode); | |
6529 | } | |
b53d3f5d | 6530 | btrfs_btree_balance_dirty(root); |
39279cc3 CM |
6531 | return err; |
6532 | } | |
6533 | ||
18bb1db3 | 6534 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
39279cc3 | 6535 | { |
b9d86667 | 6536 | struct inode *inode = NULL; |
39279cc3 CM |
6537 | struct btrfs_trans_handle *trans; |
6538 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
6539 | int err = 0; | |
6540 | int drop_on_err = 0; | |
b9d86667 | 6541 | u64 objectid = 0; |
00e4e6b3 | 6542 | u64 index = 0; |
39279cc3 | 6543 | |
9ed74f2d JB |
6544 | /* |
6545 | * 2 items for inode and ref | |
6546 | * 2 items for dir items | |
6547 | * 1 for xattr if selinux is on | |
6548 | */ | |
a22285a6 YZ |
6549 | trans = btrfs_start_transaction(root, 5); |
6550 | if (IS_ERR(trans)) | |
6551 | return PTR_ERR(trans); | |
39279cc3 | 6552 | |
581bb050 LZ |
6553 | err = btrfs_find_free_ino(root, &objectid); |
6554 | if (err) | |
6555 | goto out_fail; | |
6556 | ||
aec7477b | 6557 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 6558 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 6559 | S_IFDIR | mode, &index); |
39279cc3 CM |
6560 | if (IS_ERR(inode)) { |
6561 | err = PTR_ERR(inode); | |
6562 | goto out_fail; | |
6563 | } | |
5f39d397 | 6564 | |
39279cc3 | 6565 | drop_on_err = 1; |
b0d5d10f CM |
6566 | /* these must be set before we unlock the inode */ |
6567 | inode->i_op = &btrfs_dir_inode_operations; | |
6568 | inode->i_fop = &btrfs_dir_file_operations; | |
33268eaf | 6569 | |
2a7dba39 | 6570 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf | 6571 | if (err) |
b0d5d10f | 6572 | goto out_fail_inode; |
39279cc3 | 6573 | |
dbe674a9 | 6574 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
6575 | err = btrfs_update_inode(trans, root, inode); |
6576 | if (err) | |
b0d5d10f | 6577 | goto out_fail_inode; |
5f39d397 | 6578 | |
a1b075d2 JB |
6579 | err = btrfs_add_link(trans, dir, inode, dentry->d_name.name, |
6580 | dentry->d_name.len, 0, index); | |
39279cc3 | 6581 | if (err) |
b0d5d10f | 6582 | goto out_fail_inode; |
5f39d397 | 6583 | |
39279cc3 | 6584 | d_instantiate(dentry, inode); |
b0d5d10f CM |
6585 | /* |
6586 | * mkdir is special. We're unlocking after we call d_instantiate | |
6587 | * to avoid a race with nfsd calling d_instantiate. | |
6588 | */ | |
6589 | unlock_new_inode(inode); | |
39279cc3 | 6590 | drop_on_err = 0; |
39279cc3 CM |
6591 | |
6592 | out_fail: | |
7ad85bb7 | 6593 | btrfs_end_transaction(trans, root); |
c7cfb8a5 WS |
6594 | if (drop_on_err) { |
6595 | inode_dec_link_count(inode); | |
39279cc3 | 6596 | iput(inode); |
c7cfb8a5 | 6597 | } |
c581afc8 | 6598 | btrfs_balance_delayed_items(root); |
b53d3f5d | 6599 | btrfs_btree_balance_dirty(root); |
39279cc3 | 6600 | return err; |
b0d5d10f CM |
6601 | |
6602 | out_fail_inode: | |
6603 | unlock_new_inode(inode); | |
6604 | goto out_fail; | |
39279cc3 CM |
6605 | } |
6606 | ||
e6c4efd8 QW |
6607 | /* Find next extent map of a given extent map, caller needs to ensure locks */ |
6608 | static struct extent_map *next_extent_map(struct extent_map *em) | |
6609 | { | |
6610 | struct rb_node *next; | |
6611 | ||
6612 | next = rb_next(&em->rb_node); | |
6613 | if (!next) | |
6614 | return NULL; | |
6615 | return container_of(next, struct extent_map, rb_node); | |
6616 | } | |
6617 | ||
6618 | static struct extent_map *prev_extent_map(struct extent_map *em) | |
6619 | { | |
6620 | struct rb_node *prev; | |
6621 | ||
6622 | prev = rb_prev(&em->rb_node); | |
6623 | if (!prev) | |
6624 | return NULL; | |
6625 | return container_of(prev, struct extent_map, rb_node); | |
6626 | } | |
6627 | ||
d352ac68 | 6628 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
e6c4efd8 | 6629 | * the existing extent is the nearest extent to map_start, |
d352ac68 | 6630 | * and an extent that you want to insert, deal with overlap and insert |
e6c4efd8 | 6631 | * the best fitted new extent into the tree. |
d352ac68 | 6632 | */ |
3b951516 CM |
6633 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
6634 | struct extent_map *existing, | |
e6dcd2dc | 6635 | struct extent_map *em, |
51f395ad | 6636 | u64 map_start) |
3b951516 | 6637 | { |
e6c4efd8 QW |
6638 | struct extent_map *prev; |
6639 | struct extent_map *next; | |
6640 | u64 start; | |
6641 | u64 end; | |
3b951516 | 6642 | u64 start_diff; |
3b951516 | 6643 | |
e6dcd2dc | 6644 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
e6c4efd8 QW |
6645 | |
6646 | if (existing->start > map_start) { | |
6647 | next = existing; | |
6648 | prev = prev_extent_map(next); | |
6649 | } else { | |
6650 | prev = existing; | |
6651 | next = next_extent_map(prev); | |
6652 | } | |
6653 | ||
6654 | start = prev ? extent_map_end(prev) : em->start; | |
6655 | start = max_t(u64, start, em->start); | |
6656 | end = next ? next->start : extent_map_end(em); | |
6657 | end = min_t(u64, end, extent_map_end(em)); | |
6658 | start_diff = start - em->start; | |
6659 | em->start = start; | |
6660 | em->len = end - start; | |
c8b97818 CM |
6661 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
6662 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 6663 | em->block_start += start_diff; |
c8b97818 CM |
6664 | em->block_len -= start_diff; |
6665 | } | |
09a2a8f9 | 6666 | return add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
6667 | } |
6668 | ||
c8b97818 CM |
6669 | static noinline int uncompress_inline(struct btrfs_path *path, |
6670 | struct inode *inode, struct page *page, | |
6671 | size_t pg_offset, u64 extent_offset, | |
6672 | struct btrfs_file_extent_item *item) | |
6673 | { | |
6674 | int ret; | |
6675 | struct extent_buffer *leaf = path->nodes[0]; | |
6676 | char *tmp; | |
6677 | size_t max_size; | |
6678 | unsigned long inline_size; | |
6679 | unsigned long ptr; | |
261507a0 | 6680 | int compress_type; |
c8b97818 CM |
6681 | |
6682 | WARN_ON(pg_offset != 0); | |
261507a0 | 6683 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
6684 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
6685 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
dd3cc16b | 6686 | btrfs_item_nr(path->slots[0])); |
c8b97818 | 6687 | tmp = kmalloc(inline_size, GFP_NOFS); |
8d413713 TI |
6688 | if (!tmp) |
6689 | return -ENOMEM; | |
c8b97818 CM |
6690 | ptr = btrfs_file_extent_inline_start(item); |
6691 | ||
6692 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
6693 | ||
5b050f04 | 6694 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
261507a0 LZ |
6695 | ret = btrfs_decompress(compress_type, tmp, page, |
6696 | extent_offset, inline_size, max_size); | |
c8b97818 | 6697 | kfree(tmp); |
166ae5a4 | 6698 | return ret; |
c8b97818 CM |
6699 | } |
6700 | ||
d352ac68 CM |
6701 | /* |
6702 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
6703 | * the ugly parts come from merging extents from the disk with the in-ram |
6704 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
6705 | * where the in-ram extents might be locked pending data=ordered completion. |
6706 | * | |
6707 | * This also copies inline extents directly into the page. | |
6708 | */ | |
d397712b | 6709 | |
a52d9a80 | 6710 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 6711 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
6712 | int create) |
6713 | { | |
6714 | int ret; | |
6715 | int err = 0; | |
a52d9a80 CM |
6716 | u64 extent_start = 0; |
6717 | u64 extent_end = 0; | |
33345d01 | 6718 | u64 objectid = btrfs_ino(inode); |
a52d9a80 | 6719 | u32 found_type; |
f421950f | 6720 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
6721 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6722 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
6723 | struct extent_buffer *leaf; |
6724 | struct btrfs_key found_key; | |
a52d9a80 CM |
6725 | struct extent_map *em = NULL; |
6726 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 6727 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 6728 | struct btrfs_trans_handle *trans = NULL; |
7ffbb598 | 6729 | const bool new_inline = !page || create; |
a52d9a80 | 6730 | |
a52d9a80 | 6731 | again: |
890871be | 6732 | read_lock(&em_tree->lock); |
d1310b2e | 6733 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
6734 | if (em) |
6735 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 6736 | read_unlock(&em_tree->lock); |
d1310b2e | 6737 | |
a52d9a80 | 6738 | if (em) { |
e1c4b745 CM |
6739 | if (em->start > start || em->start + em->len <= start) |
6740 | free_extent_map(em); | |
6741 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
6742 | free_extent_map(em); |
6743 | else | |
6744 | goto out; | |
a52d9a80 | 6745 | } |
172ddd60 | 6746 | em = alloc_extent_map(); |
a52d9a80 | 6747 | if (!em) { |
d1310b2e CM |
6748 | err = -ENOMEM; |
6749 | goto out; | |
a52d9a80 | 6750 | } |
e6dcd2dc | 6751 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 6752 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 6753 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 6754 | em->len = (u64)-1; |
c8b97818 | 6755 | em->block_len = (u64)-1; |
f421950f CM |
6756 | |
6757 | if (!path) { | |
6758 | path = btrfs_alloc_path(); | |
026fd317 JB |
6759 | if (!path) { |
6760 | err = -ENOMEM; | |
6761 | goto out; | |
6762 | } | |
6763 | /* | |
6764 | * Chances are we'll be called again, so go ahead and do | |
6765 | * readahead | |
6766 | */ | |
6767 | path->reada = 1; | |
f421950f CM |
6768 | } |
6769 | ||
179e29e4 CM |
6770 | ret = btrfs_lookup_file_extent(trans, root, path, |
6771 | objectid, start, trans != NULL); | |
a52d9a80 CM |
6772 | if (ret < 0) { |
6773 | err = ret; | |
6774 | goto out; | |
6775 | } | |
6776 | ||
6777 | if (ret != 0) { | |
6778 | if (path->slots[0] == 0) | |
6779 | goto not_found; | |
6780 | path->slots[0]--; | |
6781 | } | |
6782 | ||
5f39d397 CM |
6783 | leaf = path->nodes[0]; |
6784 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 6785 | struct btrfs_file_extent_item); |
a52d9a80 | 6786 | /* are we inside the extent that was found? */ |
5f39d397 | 6787 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
962a298f | 6788 | found_type = found_key.type; |
5f39d397 | 6789 | if (found_key.objectid != objectid || |
a52d9a80 | 6790 | found_type != BTRFS_EXTENT_DATA_KEY) { |
25a50341 JB |
6791 | /* |
6792 | * If we backup past the first extent we want to move forward | |
6793 | * and see if there is an extent in front of us, otherwise we'll | |
6794 | * say there is a hole for our whole search range which can | |
6795 | * cause problems. | |
6796 | */ | |
6797 | extent_end = start; | |
6798 | goto next; | |
a52d9a80 CM |
6799 | } |
6800 | ||
5f39d397 CM |
6801 | found_type = btrfs_file_extent_type(leaf, item); |
6802 | extent_start = found_key.offset; | |
d899e052 YZ |
6803 | if (found_type == BTRFS_FILE_EXTENT_REG || |
6804 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 6805 | extent_end = extent_start + |
db94535d | 6806 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
6807 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
6808 | size_t size; | |
514ac8ad | 6809 | size = btrfs_file_extent_inline_len(leaf, path->slots[0], item); |
fda2832f | 6810 | extent_end = ALIGN(extent_start + size, root->sectorsize); |
9036c102 | 6811 | } |
25a50341 | 6812 | next: |
9036c102 YZ |
6813 | if (start >= extent_end) { |
6814 | path->slots[0]++; | |
6815 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
6816 | ret = btrfs_next_leaf(root, path); | |
6817 | if (ret < 0) { | |
6818 | err = ret; | |
6819 | goto out; | |
a52d9a80 | 6820 | } |
9036c102 YZ |
6821 | if (ret > 0) |
6822 | goto not_found; | |
6823 | leaf = path->nodes[0]; | |
a52d9a80 | 6824 | } |
9036c102 YZ |
6825 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
6826 | if (found_key.objectid != objectid || | |
6827 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
6828 | goto not_found; | |
6829 | if (start + len <= found_key.offset) | |
6830 | goto not_found; | |
e2eca69d WS |
6831 | if (start > found_key.offset) |
6832 | goto next; | |
9036c102 | 6833 | em->start = start; |
70c8a91c | 6834 | em->orig_start = start; |
9036c102 YZ |
6835 | em->len = found_key.offset - start; |
6836 | goto not_found_em; | |
6837 | } | |
6838 | ||
7ffbb598 FM |
6839 | btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em); |
6840 | ||
d899e052 YZ |
6841 | if (found_type == BTRFS_FILE_EXTENT_REG || |
6842 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 CM |
6843 | goto insert; |
6844 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 6845 | unsigned long ptr; |
a52d9a80 | 6846 | char *map; |
3326d1b0 CM |
6847 | size_t size; |
6848 | size_t extent_offset; | |
6849 | size_t copy_size; | |
a52d9a80 | 6850 | |
7ffbb598 | 6851 | if (new_inline) |
689f9346 | 6852 | goto out; |
5f39d397 | 6853 | |
514ac8ad | 6854 | size = btrfs_file_extent_inline_len(leaf, path->slots[0], item); |
9036c102 | 6855 | extent_offset = page_offset(page) + pg_offset - extent_start; |
70dec807 | 6856 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 6857 | size - extent_offset); |
3326d1b0 | 6858 | em->start = extent_start + extent_offset; |
fda2832f | 6859 | em->len = ALIGN(copy_size, root->sectorsize); |
b4939680 | 6860 | em->orig_block_len = em->len; |
70c8a91c | 6861 | em->orig_start = em->start; |
689f9346 | 6862 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 6863 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
6864 | if (btrfs_file_extent_compression(leaf, item) != |
6865 | BTRFS_COMPRESS_NONE) { | |
c8b97818 CM |
6866 | ret = uncompress_inline(path, inode, page, |
6867 | pg_offset, | |
6868 | extent_offset, item); | |
166ae5a4 ZB |
6869 | if (ret) { |
6870 | err = ret; | |
6871 | goto out; | |
6872 | } | |
c8b97818 CM |
6873 | } else { |
6874 | map = kmap(page); | |
6875 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
6876 | copy_size); | |
93c82d57 CM |
6877 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
6878 | memset(map + pg_offset + copy_size, 0, | |
6879 | PAGE_CACHE_SIZE - pg_offset - | |
6880 | copy_size); | |
6881 | } | |
c8b97818 CM |
6882 | kunmap(page); |
6883 | } | |
179e29e4 CM |
6884 | flush_dcache_page(page); |
6885 | } else if (create && PageUptodate(page)) { | |
6bf7e080 | 6886 | BUG(); |
179e29e4 CM |
6887 | if (!trans) { |
6888 | kunmap(page); | |
6889 | free_extent_map(em); | |
6890 | em = NULL; | |
ff5714cc | 6891 | |
b3b4aa74 | 6892 | btrfs_release_path(path); |
7a7eaa40 | 6893 | trans = btrfs_join_transaction(root); |
ff5714cc | 6894 | |
3612b495 TI |
6895 | if (IS_ERR(trans)) |
6896 | return ERR_CAST(trans); | |
179e29e4 CM |
6897 | goto again; |
6898 | } | |
c8b97818 | 6899 | map = kmap(page); |
70dec807 | 6900 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 6901 | copy_size); |
c8b97818 | 6902 | kunmap(page); |
179e29e4 | 6903 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 6904 | } |
d1310b2e | 6905 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 6906 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 | 6907 | goto insert; |
a52d9a80 CM |
6908 | } |
6909 | not_found: | |
6910 | em->start = start; | |
70c8a91c | 6911 | em->orig_start = start; |
d1310b2e | 6912 | em->len = len; |
a52d9a80 | 6913 | not_found_em: |
5f39d397 | 6914 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 6915 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 | 6916 | insert: |
b3b4aa74 | 6917 | btrfs_release_path(path); |
d1310b2e | 6918 | if (em->start > start || extent_map_end(em) <= start) { |
c2cf52eb | 6919 | btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]", |
c1c9ff7c | 6920 | em->start, em->len, start, len); |
a52d9a80 CM |
6921 | err = -EIO; |
6922 | goto out; | |
6923 | } | |
d1310b2e CM |
6924 | |
6925 | err = 0; | |
890871be | 6926 | write_lock(&em_tree->lock); |
09a2a8f9 | 6927 | ret = add_extent_mapping(em_tree, em, 0); |
3b951516 CM |
6928 | /* it is possible that someone inserted the extent into the tree |
6929 | * while we had the lock dropped. It is also possible that | |
6930 | * an overlapping map exists in the tree | |
6931 | */ | |
a52d9a80 | 6932 | if (ret == -EEXIST) { |
3b951516 | 6933 | struct extent_map *existing; |
e6dcd2dc CM |
6934 | |
6935 | ret = 0; | |
6936 | ||
e6c4efd8 QW |
6937 | existing = search_extent_mapping(em_tree, start, len); |
6938 | /* | |
6939 | * existing will always be non-NULL, since there must be | |
6940 | * extent causing the -EEXIST. | |
6941 | */ | |
6942 | if (start >= extent_map_end(existing) || | |
32be3a1a | 6943 | start <= existing->start) { |
e6c4efd8 QW |
6944 | /* |
6945 | * The existing extent map is the one nearest to | |
6946 | * the [start, start + len) range which overlaps | |
6947 | */ | |
6948 | err = merge_extent_mapping(em_tree, existing, | |
6949 | em, start); | |
e1c4b745 | 6950 | free_extent_map(existing); |
e6c4efd8 | 6951 | if (err) { |
3b951516 CM |
6952 | free_extent_map(em); |
6953 | em = NULL; | |
6954 | } | |
6955 | } else { | |
6956 | free_extent_map(em); | |
6957 | em = existing; | |
e6dcd2dc | 6958 | err = 0; |
a52d9a80 | 6959 | } |
a52d9a80 | 6960 | } |
890871be | 6961 | write_unlock(&em_tree->lock); |
a52d9a80 | 6962 | out: |
1abe9b8a | 6963 | |
4cd8587c | 6964 | trace_btrfs_get_extent(root, em); |
1abe9b8a | 6965 | |
527afb44 | 6966 | btrfs_free_path(path); |
a52d9a80 CM |
6967 | if (trans) { |
6968 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 6969 | if (!err) |
a52d9a80 CM |
6970 | err = ret; |
6971 | } | |
a52d9a80 CM |
6972 | if (err) { |
6973 | free_extent_map(em); | |
a52d9a80 CM |
6974 | return ERR_PTR(err); |
6975 | } | |
79787eaa | 6976 | BUG_ON(!em); /* Error is always set */ |
a52d9a80 CM |
6977 | return em; |
6978 | } | |
6979 | ||
ec29ed5b CM |
6980 | struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page, |
6981 | size_t pg_offset, u64 start, u64 len, | |
6982 | int create) | |
6983 | { | |
6984 | struct extent_map *em; | |
6985 | struct extent_map *hole_em = NULL; | |
6986 | u64 range_start = start; | |
6987 | u64 end; | |
6988 | u64 found; | |
6989 | u64 found_end; | |
6990 | int err = 0; | |
6991 | ||
6992 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
6993 | if (IS_ERR(em)) | |
6994 | return em; | |
6995 | if (em) { | |
6996 | /* | |
f9e4fb53 LB |
6997 | * if our em maps to |
6998 | * - a hole or | |
6999 | * - a pre-alloc extent, | |
7000 | * there might actually be delalloc bytes behind it. | |
ec29ed5b | 7001 | */ |
f9e4fb53 LB |
7002 | if (em->block_start != EXTENT_MAP_HOLE && |
7003 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
ec29ed5b CM |
7004 | return em; |
7005 | else | |
7006 | hole_em = em; | |
7007 | } | |
7008 | ||
7009 | /* check to see if we've wrapped (len == -1 or similar) */ | |
7010 | end = start + len; | |
7011 | if (end < start) | |
7012 | end = (u64)-1; | |
7013 | else | |
7014 | end -= 1; | |
7015 | ||
7016 | em = NULL; | |
7017 | ||
7018 | /* ok, we didn't find anything, lets look for delalloc */ | |
7019 | found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start, | |
7020 | end, len, EXTENT_DELALLOC, 1); | |
7021 | found_end = range_start + found; | |
7022 | if (found_end < range_start) | |
7023 | found_end = (u64)-1; | |
7024 | ||
7025 | /* | |
7026 | * we didn't find anything useful, return | |
7027 | * the original results from get_extent() | |
7028 | */ | |
7029 | if (range_start > end || found_end <= start) { | |
7030 | em = hole_em; | |
7031 | hole_em = NULL; | |
7032 | goto out; | |
7033 | } | |
7034 | ||
7035 | /* adjust the range_start to make sure it doesn't | |
7036 | * go backwards from the start they passed in | |
7037 | */ | |
67871254 | 7038 | range_start = max(start, range_start); |
ec29ed5b CM |
7039 | found = found_end - range_start; |
7040 | ||
7041 | if (found > 0) { | |
7042 | u64 hole_start = start; | |
7043 | u64 hole_len = len; | |
7044 | ||
172ddd60 | 7045 | em = alloc_extent_map(); |
ec29ed5b CM |
7046 | if (!em) { |
7047 | err = -ENOMEM; | |
7048 | goto out; | |
7049 | } | |
7050 | /* | |
7051 | * when btrfs_get_extent can't find anything it | |
7052 | * returns one huge hole | |
7053 | * | |
7054 | * make sure what it found really fits our range, and | |
7055 | * adjust to make sure it is based on the start from | |
7056 | * the caller | |
7057 | */ | |
7058 | if (hole_em) { | |
7059 | u64 calc_end = extent_map_end(hole_em); | |
7060 | ||
7061 | if (calc_end <= start || (hole_em->start > end)) { | |
7062 | free_extent_map(hole_em); | |
7063 | hole_em = NULL; | |
7064 | } else { | |
7065 | hole_start = max(hole_em->start, start); | |
7066 | hole_len = calc_end - hole_start; | |
7067 | } | |
7068 | } | |
7069 | em->bdev = NULL; | |
7070 | if (hole_em && range_start > hole_start) { | |
7071 | /* our hole starts before our delalloc, so we | |
7072 | * have to return just the parts of the hole | |
7073 | * that go until the delalloc starts | |
7074 | */ | |
7075 | em->len = min(hole_len, | |
7076 | range_start - hole_start); | |
7077 | em->start = hole_start; | |
7078 | em->orig_start = hole_start; | |
7079 | /* | |
7080 | * don't adjust block start at all, | |
7081 | * it is fixed at EXTENT_MAP_HOLE | |
7082 | */ | |
7083 | em->block_start = hole_em->block_start; | |
7084 | em->block_len = hole_len; | |
f9e4fb53 LB |
7085 | if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags)) |
7086 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
ec29ed5b CM |
7087 | } else { |
7088 | em->start = range_start; | |
7089 | em->len = found; | |
7090 | em->orig_start = range_start; | |
7091 | em->block_start = EXTENT_MAP_DELALLOC; | |
7092 | em->block_len = found; | |
7093 | } | |
7094 | } else if (hole_em) { | |
7095 | return hole_em; | |
7096 | } | |
7097 | out: | |
7098 | ||
7099 | free_extent_map(hole_em); | |
7100 | if (err) { | |
7101 | free_extent_map(em); | |
7102 | return ERR_PTR(err); | |
7103 | } | |
7104 | return em; | |
7105 | } | |
7106 | ||
4b46fce2 JB |
7107 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
7108 | u64 start, u64 len) | |
7109 | { | |
7110 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
70c8a91c | 7111 | struct extent_map *em; |
4b46fce2 JB |
7112 | struct btrfs_key ins; |
7113 | u64 alloc_hint; | |
7114 | int ret; | |
4b46fce2 | 7115 | |
4b46fce2 | 7116 | alloc_hint = get_extent_allocation_hint(inode, start, len); |
00361589 | 7117 | ret = btrfs_reserve_extent(root, len, root->sectorsize, 0, |
e570fd27 | 7118 | alloc_hint, &ins, 1, 1); |
00361589 JB |
7119 | if (ret) |
7120 | return ERR_PTR(ret); | |
4b46fce2 | 7121 | |
70c8a91c | 7122 | em = create_pinned_em(inode, start, ins.offset, start, ins.objectid, |
cc95bef6 | 7123 | ins.offset, ins.offset, ins.offset, 0); |
00361589 | 7124 | if (IS_ERR(em)) { |
e570fd27 | 7125 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1); |
00361589 JB |
7126 | return em; |
7127 | } | |
4b46fce2 JB |
7128 | |
7129 | ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid, | |
7130 | ins.offset, ins.offset, 0); | |
7131 | if (ret) { | |
e570fd27 | 7132 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1); |
00361589 JB |
7133 | free_extent_map(em); |
7134 | return ERR_PTR(ret); | |
4b46fce2 | 7135 | } |
00361589 | 7136 | |
4b46fce2 JB |
7137 | return em; |
7138 | } | |
7139 | ||
46bfbb5c CM |
7140 | /* |
7141 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
7142 | * block must be cow'd | |
7143 | */ | |
00361589 | 7144 | noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len, |
7ee9e440 JB |
7145 | u64 *orig_start, u64 *orig_block_len, |
7146 | u64 *ram_bytes) | |
46bfbb5c | 7147 | { |
00361589 | 7148 | struct btrfs_trans_handle *trans; |
46bfbb5c CM |
7149 | struct btrfs_path *path; |
7150 | int ret; | |
7151 | struct extent_buffer *leaf; | |
7152 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7b2b7085 | 7153 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
46bfbb5c CM |
7154 | struct btrfs_file_extent_item *fi; |
7155 | struct btrfs_key key; | |
7156 | u64 disk_bytenr; | |
7157 | u64 backref_offset; | |
7158 | u64 extent_end; | |
7159 | u64 num_bytes; | |
7160 | int slot; | |
7161 | int found_type; | |
7ee9e440 | 7162 | bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW); |
e77751aa | 7163 | |
46bfbb5c CM |
7164 | path = btrfs_alloc_path(); |
7165 | if (!path) | |
7166 | return -ENOMEM; | |
7167 | ||
00361589 | 7168 | ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), |
46bfbb5c CM |
7169 | offset, 0); |
7170 | if (ret < 0) | |
7171 | goto out; | |
7172 | ||
7173 | slot = path->slots[0]; | |
7174 | if (ret == 1) { | |
7175 | if (slot == 0) { | |
7176 | /* can't find the item, must cow */ | |
7177 | ret = 0; | |
7178 | goto out; | |
7179 | } | |
7180 | slot--; | |
7181 | } | |
7182 | ret = 0; | |
7183 | leaf = path->nodes[0]; | |
7184 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
33345d01 | 7185 | if (key.objectid != btrfs_ino(inode) || |
46bfbb5c CM |
7186 | key.type != BTRFS_EXTENT_DATA_KEY) { |
7187 | /* not our file or wrong item type, must cow */ | |
7188 | goto out; | |
7189 | } | |
7190 | ||
7191 | if (key.offset > offset) { | |
7192 | /* Wrong offset, must cow */ | |
7193 | goto out; | |
7194 | } | |
7195 | ||
7196 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
7197 | found_type = btrfs_file_extent_type(leaf, fi); | |
7198 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
7199 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
7200 | /* not a regular extent, must cow */ | |
7201 | goto out; | |
7202 | } | |
7ee9e440 JB |
7203 | |
7204 | if (!nocow && found_type == BTRFS_FILE_EXTENT_REG) | |
7205 | goto out; | |
7206 | ||
e77751aa MX |
7207 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
7208 | if (extent_end <= offset) | |
7209 | goto out; | |
7210 | ||
46bfbb5c | 7211 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
7ee9e440 JB |
7212 | if (disk_bytenr == 0) |
7213 | goto out; | |
7214 | ||
7215 | if (btrfs_file_extent_compression(leaf, fi) || | |
7216 | btrfs_file_extent_encryption(leaf, fi) || | |
7217 | btrfs_file_extent_other_encoding(leaf, fi)) | |
7218 | goto out; | |
7219 | ||
46bfbb5c CM |
7220 | backref_offset = btrfs_file_extent_offset(leaf, fi); |
7221 | ||
7ee9e440 JB |
7222 | if (orig_start) { |
7223 | *orig_start = key.offset - backref_offset; | |
7224 | *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi); | |
7225 | *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi); | |
7226 | } | |
eb384b55 | 7227 | |
46bfbb5c CM |
7228 | if (btrfs_extent_readonly(root, disk_bytenr)) |
7229 | goto out; | |
7b2b7085 MX |
7230 | |
7231 | num_bytes = min(offset + *len, extent_end) - offset; | |
7232 | if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
7233 | u64 range_end; | |
7234 | ||
7235 | range_end = round_up(offset + num_bytes, root->sectorsize) - 1; | |
7236 | ret = test_range_bit(io_tree, offset, range_end, | |
7237 | EXTENT_DELALLOC, 0, NULL); | |
7238 | if (ret) { | |
7239 | ret = -EAGAIN; | |
7240 | goto out; | |
7241 | } | |
7242 | } | |
7243 | ||
1bda19eb | 7244 | btrfs_release_path(path); |
46bfbb5c CM |
7245 | |
7246 | /* | |
7247 | * look for other files referencing this extent, if we | |
7248 | * find any we must cow | |
7249 | */ | |
00361589 JB |
7250 | trans = btrfs_join_transaction(root); |
7251 | if (IS_ERR(trans)) { | |
7252 | ret = 0; | |
46bfbb5c | 7253 | goto out; |
00361589 JB |
7254 | } |
7255 | ||
7256 | ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode), | |
7257 | key.offset - backref_offset, disk_bytenr); | |
7258 | btrfs_end_transaction(trans, root); | |
7259 | if (ret) { | |
7260 | ret = 0; | |
7261 | goto out; | |
7262 | } | |
46bfbb5c CM |
7263 | |
7264 | /* | |
7265 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
7266 | * in this extent we are about to write. If there | |
7267 | * are any csums in that range we have to cow in order | |
7268 | * to keep the csums correct | |
7269 | */ | |
7270 | disk_bytenr += backref_offset; | |
7271 | disk_bytenr += offset - key.offset; | |
46bfbb5c CM |
7272 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) |
7273 | goto out; | |
7274 | /* | |
7275 | * all of the above have passed, it is safe to overwrite this extent | |
7276 | * without cow | |
7277 | */ | |
eb384b55 | 7278 | *len = num_bytes; |
46bfbb5c CM |
7279 | ret = 1; |
7280 | out: | |
7281 | btrfs_free_path(path); | |
7282 | return ret; | |
7283 | } | |
7284 | ||
fc4adbff AG |
7285 | bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end) |
7286 | { | |
7287 | struct radix_tree_root *root = &inode->i_mapping->page_tree; | |
7288 | int found = false; | |
7289 | void **pagep = NULL; | |
7290 | struct page *page = NULL; | |
7291 | int start_idx; | |
7292 | int end_idx; | |
7293 | ||
7294 | start_idx = start >> PAGE_CACHE_SHIFT; | |
7295 | ||
7296 | /* | |
7297 | * end is the last byte in the last page. end == start is legal | |
7298 | */ | |
7299 | end_idx = end >> PAGE_CACHE_SHIFT; | |
7300 | ||
7301 | rcu_read_lock(); | |
7302 | ||
7303 | /* Most of the code in this while loop is lifted from | |
7304 | * find_get_page. It's been modified to begin searching from a | |
7305 | * page and return just the first page found in that range. If the | |
7306 | * found idx is less than or equal to the end idx then we know that | |
7307 | * a page exists. If no pages are found or if those pages are | |
7308 | * outside of the range then we're fine (yay!) */ | |
7309 | while (page == NULL && | |
7310 | radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) { | |
7311 | page = radix_tree_deref_slot(pagep); | |
7312 | if (unlikely(!page)) | |
7313 | break; | |
7314 | ||
7315 | if (radix_tree_exception(page)) { | |
809f9016 FM |
7316 | if (radix_tree_deref_retry(page)) { |
7317 | page = NULL; | |
fc4adbff | 7318 | continue; |
809f9016 | 7319 | } |
fc4adbff AG |
7320 | /* |
7321 | * Otherwise, shmem/tmpfs must be storing a swap entry | |
7322 | * here as an exceptional entry: so return it without | |
7323 | * attempting to raise page count. | |
7324 | */ | |
6fdef6d4 | 7325 | page = NULL; |
fc4adbff AG |
7326 | break; /* TODO: Is this relevant for this use case? */ |
7327 | } | |
7328 | ||
91405151 FM |
7329 | if (!page_cache_get_speculative(page)) { |
7330 | page = NULL; | |
fc4adbff | 7331 | continue; |
91405151 | 7332 | } |
fc4adbff AG |
7333 | |
7334 | /* | |
7335 | * Has the page moved? | |
7336 | * This is part of the lockless pagecache protocol. See | |
7337 | * include/linux/pagemap.h for details. | |
7338 | */ | |
7339 | if (unlikely(page != *pagep)) { | |
7340 | page_cache_release(page); | |
7341 | page = NULL; | |
7342 | } | |
7343 | } | |
7344 | ||
7345 | if (page) { | |
7346 | if (page->index <= end_idx) | |
7347 | found = true; | |
7348 | page_cache_release(page); | |
7349 | } | |
7350 | ||
7351 | rcu_read_unlock(); | |
7352 | return found; | |
7353 | } | |
7354 | ||
eb838e73 JB |
7355 | static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend, |
7356 | struct extent_state **cached_state, int writing) | |
7357 | { | |
7358 | struct btrfs_ordered_extent *ordered; | |
7359 | int ret = 0; | |
7360 | ||
7361 | while (1) { | |
7362 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
7363 | 0, cached_state); | |
7364 | /* | |
7365 | * We're concerned with the entire range that we're going to be | |
7366 | * doing DIO to, so we need to make sure theres no ordered | |
7367 | * extents in this range. | |
7368 | */ | |
7369 | ordered = btrfs_lookup_ordered_range(inode, lockstart, | |
7370 | lockend - lockstart + 1); | |
7371 | ||
7372 | /* | |
7373 | * We need to make sure there are no buffered pages in this | |
7374 | * range either, we could have raced between the invalidate in | |
7375 | * generic_file_direct_write and locking the extent. The | |
7376 | * invalidate needs to happen so that reads after a write do not | |
7377 | * get stale data. | |
7378 | */ | |
fc4adbff AG |
7379 | if (!ordered && |
7380 | (!writing || | |
7381 | !btrfs_page_exists_in_range(inode, lockstart, lockend))) | |
eb838e73 JB |
7382 | break; |
7383 | ||
7384 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
7385 | cached_state, GFP_NOFS); | |
7386 | ||
7387 | if (ordered) { | |
7388 | btrfs_start_ordered_extent(inode, ordered, 1); | |
7389 | btrfs_put_ordered_extent(ordered); | |
7390 | } else { | |
7391 | /* Screw you mmap */ | |
728404da | 7392 | ret = btrfs_fdatawrite_range(inode, lockstart, lockend); |
075bdbdb FM |
7393 | if (ret) |
7394 | break; | |
7395 | ret = filemap_fdatawait_range(inode->i_mapping, | |
7396 | lockstart, | |
7397 | lockend); | |
eb838e73 JB |
7398 | if (ret) |
7399 | break; | |
7400 | ||
7401 | /* | |
7402 | * If we found a page that couldn't be invalidated just | |
7403 | * fall back to buffered. | |
7404 | */ | |
7405 | ret = invalidate_inode_pages2_range(inode->i_mapping, | |
7406 | lockstart >> PAGE_CACHE_SHIFT, | |
7407 | lockend >> PAGE_CACHE_SHIFT); | |
7408 | if (ret) | |
7409 | break; | |
7410 | } | |
7411 | ||
7412 | cond_resched(); | |
7413 | } | |
7414 | ||
7415 | return ret; | |
7416 | } | |
7417 | ||
69ffb543 JB |
7418 | static struct extent_map *create_pinned_em(struct inode *inode, u64 start, |
7419 | u64 len, u64 orig_start, | |
7420 | u64 block_start, u64 block_len, | |
cc95bef6 JB |
7421 | u64 orig_block_len, u64 ram_bytes, |
7422 | int type) | |
69ffb543 JB |
7423 | { |
7424 | struct extent_map_tree *em_tree; | |
7425 | struct extent_map *em; | |
7426 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7427 | int ret; | |
7428 | ||
7429 | em_tree = &BTRFS_I(inode)->extent_tree; | |
7430 | em = alloc_extent_map(); | |
7431 | if (!em) | |
7432 | return ERR_PTR(-ENOMEM); | |
7433 | ||
7434 | em->start = start; | |
7435 | em->orig_start = orig_start; | |
2ab28f32 JB |
7436 | em->mod_start = start; |
7437 | em->mod_len = len; | |
69ffb543 JB |
7438 | em->len = len; |
7439 | em->block_len = block_len; | |
7440 | em->block_start = block_start; | |
7441 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
b4939680 | 7442 | em->orig_block_len = orig_block_len; |
cc95bef6 | 7443 | em->ram_bytes = ram_bytes; |
70c8a91c | 7444 | em->generation = -1; |
69ffb543 JB |
7445 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
7446 | if (type == BTRFS_ORDERED_PREALLOC) | |
b11e234d | 7447 | set_bit(EXTENT_FLAG_FILLING, &em->flags); |
69ffb543 JB |
7448 | |
7449 | do { | |
7450 | btrfs_drop_extent_cache(inode, em->start, | |
7451 | em->start + em->len - 1, 0); | |
7452 | write_lock(&em_tree->lock); | |
09a2a8f9 | 7453 | ret = add_extent_mapping(em_tree, em, 1); |
69ffb543 JB |
7454 | write_unlock(&em_tree->lock); |
7455 | } while (ret == -EEXIST); | |
7456 | ||
7457 | if (ret) { | |
7458 | free_extent_map(em); | |
7459 | return ERR_PTR(ret); | |
7460 | } | |
7461 | ||
7462 | return em; | |
7463 | } | |
7464 | ||
50745b0a | 7465 | struct btrfs_dio_data { |
7466 | u64 outstanding_extents; | |
7467 | u64 reserve; | |
7468 | }; | |
69ffb543 | 7469 | |
4b46fce2 JB |
7470 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
7471 | struct buffer_head *bh_result, int create) | |
7472 | { | |
7473 | struct extent_map *em; | |
7474 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
eb838e73 | 7475 | struct extent_state *cached_state = NULL; |
50745b0a | 7476 | struct btrfs_dio_data *dio_data = NULL; |
4b46fce2 | 7477 | u64 start = iblock << inode->i_blkbits; |
eb838e73 | 7478 | u64 lockstart, lockend; |
4b46fce2 | 7479 | u64 len = bh_result->b_size; |
eb838e73 | 7480 | int unlock_bits = EXTENT_LOCKED; |
0934856d | 7481 | int ret = 0; |
eb838e73 | 7482 | |
172a5049 | 7483 | if (create) |
3266789f | 7484 | unlock_bits |= EXTENT_DIRTY; |
172a5049 | 7485 | else |
c329861d | 7486 | len = min_t(u64, len, root->sectorsize); |
eb838e73 | 7487 | |
c329861d JB |
7488 | lockstart = start; |
7489 | lockend = start + len - 1; | |
7490 | ||
e1cbbfa5 JB |
7491 | if (current->journal_info) { |
7492 | /* | |
7493 | * Need to pull our outstanding extents and set journal_info to NULL so | |
7494 | * that anything that needs to check if there's a transction doesn't get | |
7495 | * confused. | |
7496 | */ | |
50745b0a | 7497 | dio_data = current->journal_info; |
e1cbbfa5 JB |
7498 | current->journal_info = NULL; |
7499 | } | |
7500 | ||
eb838e73 JB |
7501 | /* |
7502 | * If this errors out it's because we couldn't invalidate pagecache for | |
7503 | * this range and we need to fallback to buffered. | |
7504 | */ | |
7505 | if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create)) | |
7506 | return -ENOTBLK; | |
7507 | ||
4b46fce2 | 7508 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); |
eb838e73 JB |
7509 | if (IS_ERR(em)) { |
7510 | ret = PTR_ERR(em); | |
7511 | goto unlock_err; | |
7512 | } | |
4b46fce2 JB |
7513 | |
7514 | /* | |
7515 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
7516 | * io. INLINE is special, and we could probably kludge it in here, but | |
7517 | * it's still buffered so for safety lets just fall back to the generic | |
7518 | * buffered path. | |
7519 | * | |
7520 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
7521 | * decompress it, so there will be buffering required no matter what we | |
7522 | * do, so go ahead and fallback to buffered. | |
7523 | * | |
7524 | * We return -ENOTBLK because thats what makes DIO go ahead and go back | |
7525 | * to buffered IO. Don't blame me, this is the price we pay for using | |
7526 | * the generic code. | |
7527 | */ | |
7528 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
7529 | em->block_start == EXTENT_MAP_INLINE) { | |
7530 | free_extent_map(em); | |
eb838e73 JB |
7531 | ret = -ENOTBLK; |
7532 | goto unlock_err; | |
4b46fce2 JB |
7533 | } |
7534 | ||
7535 | /* Just a good old fashioned hole, return */ | |
7536 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
7537 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
7538 | free_extent_map(em); | |
eb838e73 | 7539 | goto unlock_err; |
4b46fce2 JB |
7540 | } |
7541 | ||
7542 | /* | |
7543 | * We don't allocate a new extent in the following cases | |
7544 | * | |
7545 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
7546 | * existing extent. | |
7547 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
7548 | * just use the extent. | |
7549 | * | |
7550 | */ | |
46bfbb5c | 7551 | if (!create) { |
eb838e73 JB |
7552 | len = min(len, em->len - (start - em->start)); |
7553 | lockstart = start + len; | |
7554 | goto unlock; | |
46bfbb5c | 7555 | } |
4b46fce2 JB |
7556 | |
7557 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
7558 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
7559 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 | 7560 | int type; |
eb384b55 | 7561 | u64 block_start, orig_start, orig_block_len, ram_bytes; |
4b46fce2 JB |
7562 | |
7563 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7564 | type = BTRFS_ORDERED_PREALLOC; | |
7565 | else | |
7566 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 7567 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 7568 | block_start = em->block_start + (start - em->start); |
46bfbb5c | 7569 | |
00361589 | 7570 | if (can_nocow_extent(inode, start, &len, &orig_start, |
7ee9e440 | 7571 | &orig_block_len, &ram_bytes) == 1) { |
69ffb543 JB |
7572 | if (type == BTRFS_ORDERED_PREALLOC) { |
7573 | free_extent_map(em); | |
7574 | em = create_pinned_em(inode, start, len, | |
7575 | orig_start, | |
b4939680 | 7576 | block_start, len, |
cc95bef6 JB |
7577 | orig_block_len, |
7578 | ram_bytes, type); | |
555e1286 FM |
7579 | if (IS_ERR(em)) { |
7580 | ret = PTR_ERR(em); | |
69ffb543 | 7581 | goto unlock_err; |
555e1286 | 7582 | } |
69ffb543 JB |
7583 | } |
7584 | ||
46bfbb5c CM |
7585 | ret = btrfs_add_ordered_extent_dio(inode, start, |
7586 | block_start, len, len, type); | |
46bfbb5c CM |
7587 | if (ret) { |
7588 | free_extent_map(em); | |
eb838e73 | 7589 | goto unlock_err; |
46bfbb5c CM |
7590 | } |
7591 | goto unlock; | |
4b46fce2 | 7592 | } |
4b46fce2 | 7593 | } |
00361589 | 7594 | |
46bfbb5c CM |
7595 | /* |
7596 | * this will cow the extent, reset the len in case we changed | |
7597 | * it above | |
7598 | */ | |
7599 | len = bh_result->b_size; | |
70c8a91c JB |
7600 | free_extent_map(em); |
7601 | em = btrfs_new_extent_direct(inode, start, len); | |
eb838e73 JB |
7602 | if (IS_ERR(em)) { |
7603 | ret = PTR_ERR(em); | |
7604 | goto unlock_err; | |
7605 | } | |
46bfbb5c CM |
7606 | len = min(len, em->len - (start - em->start)); |
7607 | unlock: | |
4b46fce2 JB |
7608 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> |
7609 | inode->i_blkbits; | |
46bfbb5c | 7610 | bh_result->b_size = len; |
4b46fce2 JB |
7611 | bh_result->b_bdev = em->bdev; |
7612 | set_buffer_mapped(bh_result); | |
c3473e83 JB |
7613 | if (create) { |
7614 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
7615 | set_buffer_new(bh_result); | |
7616 | ||
7617 | /* | |
7618 | * Need to update the i_size under the extent lock so buffered | |
7619 | * readers will get the updated i_size when we unlock. | |
7620 | */ | |
7621 | if (start + len > i_size_read(inode)) | |
7622 | i_size_write(inode, start + len); | |
0934856d | 7623 | |
e1cbbfa5 JB |
7624 | /* |
7625 | * If we have an outstanding_extents count still set then we're | |
7626 | * within our reservation, otherwise we need to adjust our inode | |
7627 | * counter appropriately. | |
7628 | */ | |
50745b0a | 7629 | if (dio_data->outstanding_extents) { |
7630 | (dio_data->outstanding_extents)--; | |
e1cbbfa5 | 7631 | } else { |
3e05bde8 JB |
7632 | spin_lock(&BTRFS_I(inode)->lock); |
7633 | BTRFS_I(inode)->outstanding_extents++; | |
7634 | spin_unlock(&BTRFS_I(inode)->lock); | |
7635 | } | |
e1cbbfa5 | 7636 | |
3266789f | 7637 | btrfs_free_reserved_data_space(inode, len); |
50745b0a | 7638 | WARN_ON(dio_data->reserve < len); |
7639 | dio_data->reserve -= len; | |
7640 | current->journal_info = dio_data; | |
c3473e83 | 7641 | } |
4b46fce2 | 7642 | |
eb838e73 JB |
7643 | /* |
7644 | * In the case of write we need to clear and unlock the entire range, | |
7645 | * in the case of read we need to unlock only the end area that we | |
7646 | * aren't using if there is any left over space. | |
7647 | */ | |
24c03fa5 | 7648 | if (lockstart < lockend) { |
0934856d MX |
7649 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, |
7650 | lockend, unlock_bits, 1, 0, | |
7651 | &cached_state, GFP_NOFS); | |
24c03fa5 | 7652 | } else { |
eb838e73 | 7653 | free_extent_state(cached_state); |
24c03fa5 | 7654 | } |
eb838e73 | 7655 | |
4b46fce2 JB |
7656 | free_extent_map(em); |
7657 | ||
7658 | return 0; | |
eb838e73 JB |
7659 | |
7660 | unlock_err: | |
eb838e73 JB |
7661 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
7662 | unlock_bits, 1, 0, &cached_state, GFP_NOFS); | |
50745b0a | 7663 | if (dio_data) |
7664 | current->journal_info = dio_data; | |
eb838e73 | 7665 | return ret; |
4b46fce2 JB |
7666 | } |
7667 | ||
8b110e39 MX |
7668 | static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio, |
7669 | int rw, int mirror_num) | |
7670 | { | |
7671 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
7672 | int ret; | |
7673 | ||
7674 | BUG_ON(rw & REQ_WRITE); | |
7675 | ||
7676 | bio_get(bio); | |
7677 | ||
7678 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, | |
7679 | BTRFS_WQ_ENDIO_DIO_REPAIR); | |
7680 | if (ret) | |
7681 | goto err; | |
7682 | ||
7683 | ret = btrfs_map_bio(root, rw, bio, mirror_num, 0); | |
7684 | err: | |
7685 | bio_put(bio); | |
7686 | return ret; | |
7687 | } | |
7688 | ||
7689 | static int btrfs_check_dio_repairable(struct inode *inode, | |
7690 | struct bio *failed_bio, | |
7691 | struct io_failure_record *failrec, | |
7692 | int failed_mirror) | |
7693 | { | |
7694 | int num_copies; | |
7695 | ||
7696 | num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info, | |
7697 | failrec->logical, failrec->len); | |
7698 | if (num_copies == 1) { | |
7699 | /* | |
7700 | * we only have a single copy of the data, so don't bother with | |
7701 | * all the retry and error correction code that follows. no | |
7702 | * matter what the error is, it is very likely to persist. | |
7703 | */ | |
7704 | pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n", | |
7705 | num_copies, failrec->this_mirror, failed_mirror); | |
7706 | return 0; | |
7707 | } | |
7708 | ||
7709 | failrec->failed_mirror = failed_mirror; | |
7710 | failrec->this_mirror++; | |
7711 | if (failrec->this_mirror == failed_mirror) | |
7712 | failrec->this_mirror++; | |
7713 | ||
7714 | if (failrec->this_mirror > num_copies) { | |
7715 | pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n", | |
7716 | num_copies, failrec->this_mirror, failed_mirror); | |
7717 | return 0; | |
7718 | } | |
7719 | ||
7720 | return 1; | |
7721 | } | |
7722 | ||
7723 | static int dio_read_error(struct inode *inode, struct bio *failed_bio, | |
7724 | struct page *page, u64 start, u64 end, | |
7725 | int failed_mirror, bio_end_io_t *repair_endio, | |
7726 | void *repair_arg) | |
7727 | { | |
7728 | struct io_failure_record *failrec; | |
7729 | struct bio *bio; | |
7730 | int isector; | |
7731 | int read_mode; | |
7732 | int ret; | |
7733 | ||
7734 | BUG_ON(failed_bio->bi_rw & REQ_WRITE); | |
7735 | ||
7736 | ret = btrfs_get_io_failure_record(inode, start, end, &failrec); | |
7737 | if (ret) | |
7738 | return ret; | |
7739 | ||
7740 | ret = btrfs_check_dio_repairable(inode, failed_bio, failrec, | |
7741 | failed_mirror); | |
7742 | if (!ret) { | |
7743 | free_io_failure(inode, failrec); | |
7744 | return -EIO; | |
7745 | } | |
7746 | ||
7747 | if (failed_bio->bi_vcnt > 1) | |
7748 | read_mode = READ_SYNC | REQ_FAILFAST_DEV; | |
7749 | else | |
7750 | read_mode = READ_SYNC; | |
7751 | ||
7752 | isector = start - btrfs_io_bio(failed_bio)->logical; | |
7753 | isector >>= inode->i_sb->s_blocksize_bits; | |
7754 | bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page, | |
7755 | 0, isector, repair_endio, repair_arg); | |
7756 | if (!bio) { | |
7757 | free_io_failure(inode, failrec); | |
7758 | return -EIO; | |
7759 | } | |
7760 | ||
7761 | btrfs_debug(BTRFS_I(inode)->root->fs_info, | |
7762 | "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n", | |
7763 | read_mode, failrec->this_mirror, failrec->in_validation); | |
7764 | ||
7765 | ret = submit_dio_repair_bio(inode, bio, read_mode, | |
7766 | failrec->this_mirror); | |
7767 | if (ret) { | |
7768 | free_io_failure(inode, failrec); | |
7769 | bio_put(bio); | |
7770 | } | |
7771 | ||
7772 | return ret; | |
7773 | } | |
7774 | ||
7775 | struct btrfs_retry_complete { | |
7776 | struct completion done; | |
7777 | struct inode *inode; | |
7778 | u64 start; | |
7779 | int uptodate; | |
7780 | }; | |
7781 | ||
4246a0b6 | 7782 | static void btrfs_retry_endio_nocsum(struct bio *bio) |
8b110e39 MX |
7783 | { |
7784 | struct btrfs_retry_complete *done = bio->bi_private; | |
7785 | struct bio_vec *bvec; | |
7786 | int i; | |
7787 | ||
4246a0b6 | 7788 | if (bio->bi_error) |
8b110e39 MX |
7789 | goto end; |
7790 | ||
7791 | done->uptodate = 1; | |
7792 | bio_for_each_segment_all(bvec, bio, i) | |
7793 | clean_io_failure(done->inode, done->start, bvec->bv_page, 0); | |
7794 | end: | |
7795 | complete(&done->done); | |
7796 | bio_put(bio); | |
7797 | } | |
7798 | ||
7799 | static int __btrfs_correct_data_nocsum(struct inode *inode, | |
7800 | struct btrfs_io_bio *io_bio) | |
4b46fce2 | 7801 | { |
2c30c71b | 7802 | struct bio_vec *bvec; |
8b110e39 | 7803 | struct btrfs_retry_complete done; |
4b46fce2 | 7804 | u64 start; |
2c30c71b | 7805 | int i; |
c1dc0896 | 7806 | int ret; |
4b46fce2 | 7807 | |
8b110e39 MX |
7808 | start = io_bio->logical; |
7809 | done.inode = inode; | |
7810 | ||
7811 | bio_for_each_segment_all(bvec, &io_bio->bio, i) { | |
7812 | try_again: | |
7813 | done.uptodate = 0; | |
7814 | done.start = start; | |
7815 | init_completion(&done.done); | |
7816 | ||
7817 | ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start, | |
7818 | start + bvec->bv_len - 1, | |
7819 | io_bio->mirror_num, | |
7820 | btrfs_retry_endio_nocsum, &done); | |
7821 | if (ret) | |
7822 | return ret; | |
7823 | ||
7824 | wait_for_completion(&done.done); | |
7825 | ||
7826 | if (!done.uptodate) { | |
7827 | /* We might have another mirror, so try again */ | |
7828 | goto try_again; | |
7829 | } | |
7830 | ||
7831 | start += bvec->bv_len; | |
7832 | } | |
7833 | ||
7834 | return 0; | |
7835 | } | |
7836 | ||
4246a0b6 | 7837 | static void btrfs_retry_endio(struct bio *bio) |
8b110e39 MX |
7838 | { |
7839 | struct btrfs_retry_complete *done = bio->bi_private; | |
7840 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
7841 | struct bio_vec *bvec; | |
7842 | int uptodate; | |
7843 | int ret; | |
7844 | int i; | |
7845 | ||
4246a0b6 | 7846 | if (bio->bi_error) |
8b110e39 MX |
7847 | goto end; |
7848 | ||
7849 | uptodate = 1; | |
7850 | bio_for_each_segment_all(bvec, bio, i) { | |
7851 | ret = __readpage_endio_check(done->inode, io_bio, i, | |
7852 | bvec->bv_page, 0, | |
7853 | done->start, bvec->bv_len); | |
7854 | if (!ret) | |
7855 | clean_io_failure(done->inode, done->start, | |
7856 | bvec->bv_page, 0); | |
7857 | else | |
7858 | uptodate = 0; | |
7859 | } | |
7860 | ||
7861 | done->uptodate = uptodate; | |
7862 | end: | |
7863 | complete(&done->done); | |
7864 | bio_put(bio); | |
7865 | } | |
7866 | ||
7867 | static int __btrfs_subio_endio_read(struct inode *inode, | |
7868 | struct btrfs_io_bio *io_bio, int err) | |
7869 | { | |
7870 | struct bio_vec *bvec; | |
7871 | struct btrfs_retry_complete done; | |
7872 | u64 start; | |
7873 | u64 offset = 0; | |
7874 | int i; | |
7875 | int ret; | |
dc380aea | 7876 | |
8b110e39 | 7877 | err = 0; |
c1dc0896 | 7878 | start = io_bio->logical; |
8b110e39 MX |
7879 | done.inode = inode; |
7880 | ||
c1dc0896 | 7881 | bio_for_each_segment_all(bvec, &io_bio->bio, i) { |
dc380aea MX |
7882 | ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page, |
7883 | 0, start, bvec->bv_len); | |
8b110e39 MX |
7884 | if (likely(!ret)) |
7885 | goto next; | |
7886 | try_again: | |
7887 | done.uptodate = 0; | |
7888 | done.start = start; | |
7889 | init_completion(&done.done); | |
7890 | ||
7891 | ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start, | |
7892 | start + bvec->bv_len - 1, | |
7893 | io_bio->mirror_num, | |
7894 | btrfs_retry_endio, &done); | |
7895 | if (ret) { | |
7896 | err = ret; | |
7897 | goto next; | |
7898 | } | |
7899 | ||
7900 | wait_for_completion(&done.done); | |
7901 | ||
7902 | if (!done.uptodate) { | |
7903 | /* We might have another mirror, so try again */ | |
7904 | goto try_again; | |
7905 | } | |
7906 | next: | |
7907 | offset += bvec->bv_len; | |
4b46fce2 | 7908 | start += bvec->bv_len; |
2c30c71b | 7909 | } |
c1dc0896 MX |
7910 | |
7911 | return err; | |
7912 | } | |
7913 | ||
8b110e39 MX |
7914 | static int btrfs_subio_endio_read(struct inode *inode, |
7915 | struct btrfs_io_bio *io_bio, int err) | |
7916 | { | |
7917 | bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
7918 | ||
7919 | if (skip_csum) { | |
7920 | if (unlikely(err)) | |
7921 | return __btrfs_correct_data_nocsum(inode, io_bio); | |
7922 | else | |
7923 | return 0; | |
7924 | } else { | |
7925 | return __btrfs_subio_endio_read(inode, io_bio, err); | |
7926 | } | |
7927 | } | |
7928 | ||
4246a0b6 | 7929 | static void btrfs_endio_direct_read(struct bio *bio) |
c1dc0896 MX |
7930 | { |
7931 | struct btrfs_dio_private *dip = bio->bi_private; | |
7932 | struct inode *inode = dip->inode; | |
7933 | struct bio *dio_bio; | |
7934 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
4246a0b6 | 7935 | int err = bio->bi_error; |
c1dc0896 | 7936 | |
8b110e39 MX |
7937 | if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED) |
7938 | err = btrfs_subio_endio_read(inode, io_bio, err); | |
c1dc0896 | 7939 | |
4b46fce2 | 7940 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, |
d0082371 | 7941 | dip->logical_offset + dip->bytes - 1); |
9be3395b | 7942 | dio_bio = dip->dio_bio; |
4b46fce2 | 7943 | |
4b46fce2 | 7944 | kfree(dip); |
c0da7aa1 | 7945 | |
4246a0b6 | 7946 | dio_end_io(dio_bio, bio->bi_error); |
23ea8e5a MX |
7947 | |
7948 | if (io_bio->end_io) | |
7949 | io_bio->end_io(io_bio, err); | |
9be3395b | 7950 | bio_put(bio); |
4b46fce2 JB |
7951 | } |
7952 | ||
4246a0b6 | 7953 | static void btrfs_endio_direct_write(struct bio *bio) |
4b46fce2 JB |
7954 | { |
7955 | struct btrfs_dio_private *dip = bio->bi_private; | |
7956 | struct inode *inode = dip->inode; | |
7957 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4b46fce2 | 7958 | struct btrfs_ordered_extent *ordered = NULL; |
163cf09c CM |
7959 | u64 ordered_offset = dip->logical_offset; |
7960 | u64 ordered_bytes = dip->bytes; | |
9be3395b | 7961 | struct bio *dio_bio; |
4b46fce2 JB |
7962 | int ret; |
7963 | ||
163cf09c CM |
7964 | again: |
7965 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
7966 | &ordered_offset, | |
4246a0b6 CH |
7967 | ordered_bytes, |
7968 | !bio->bi_error); | |
4b46fce2 | 7969 | if (!ret) |
163cf09c | 7970 | goto out_test; |
4b46fce2 | 7971 | |
9e0af237 LB |
7972 | btrfs_init_work(&ordered->work, btrfs_endio_write_helper, |
7973 | finish_ordered_fn, NULL, NULL); | |
fccb5d86 QW |
7974 | btrfs_queue_work(root->fs_info->endio_write_workers, |
7975 | &ordered->work); | |
163cf09c CM |
7976 | out_test: |
7977 | /* | |
7978 | * our bio might span multiple ordered extents. If we haven't | |
7979 | * completed the accounting for the whole dio, go back and try again | |
7980 | */ | |
7981 | if (ordered_offset < dip->logical_offset + dip->bytes) { | |
7982 | ordered_bytes = dip->logical_offset + dip->bytes - | |
7983 | ordered_offset; | |
5fd02043 | 7984 | ordered = NULL; |
163cf09c CM |
7985 | goto again; |
7986 | } | |
9be3395b | 7987 | dio_bio = dip->dio_bio; |
4b46fce2 | 7988 | |
4b46fce2 | 7989 | kfree(dip); |
c0da7aa1 | 7990 | |
4246a0b6 | 7991 | dio_end_io(dio_bio, bio->bi_error); |
9be3395b | 7992 | bio_put(bio); |
4b46fce2 JB |
7993 | } |
7994 | ||
eaf25d93 CM |
7995 | static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw, |
7996 | struct bio *bio, int mirror_num, | |
7997 | unsigned long bio_flags, u64 offset) | |
7998 | { | |
7999 | int ret; | |
8000 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8001 | ret = btrfs_csum_one_bio(root, inode, bio, offset, 1); | |
79787eaa | 8002 | BUG_ON(ret); /* -ENOMEM */ |
eaf25d93 CM |
8003 | return 0; |
8004 | } | |
8005 | ||
4246a0b6 | 8006 | static void btrfs_end_dio_bio(struct bio *bio) |
e65e1535 MX |
8007 | { |
8008 | struct btrfs_dio_private *dip = bio->bi_private; | |
4246a0b6 | 8009 | int err = bio->bi_error; |
e65e1535 | 8010 | |
8b110e39 MX |
8011 | if (err) |
8012 | btrfs_warn(BTRFS_I(dip->inode)->root->fs_info, | |
8013 | "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d", | |
8014 | btrfs_ino(dip->inode), bio->bi_rw, | |
8015 | (unsigned long long)bio->bi_iter.bi_sector, | |
8016 | bio->bi_iter.bi_size, err); | |
8017 | ||
8018 | if (dip->subio_endio) | |
8019 | err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err); | |
c1dc0896 MX |
8020 | |
8021 | if (err) { | |
e65e1535 MX |
8022 | dip->errors = 1; |
8023 | ||
8024 | /* | |
8025 | * before atomic variable goto zero, we must make sure | |
8026 | * dip->errors is perceived to be set. | |
8027 | */ | |
4e857c58 | 8028 | smp_mb__before_atomic(); |
e65e1535 MX |
8029 | } |
8030 | ||
8031 | /* if there are more bios still pending for this dio, just exit */ | |
8032 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
8033 | goto out; | |
8034 | ||
9be3395b | 8035 | if (dip->errors) { |
e65e1535 | 8036 | bio_io_error(dip->orig_bio); |
9be3395b | 8037 | } else { |
4246a0b6 CH |
8038 | dip->dio_bio->bi_error = 0; |
8039 | bio_endio(dip->orig_bio); | |
e65e1535 MX |
8040 | } |
8041 | out: | |
8042 | bio_put(bio); | |
8043 | } | |
8044 | ||
8045 | static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev, | |
8046 | u64 first_sector, gfp_t gfp_flags) | |
8047 | { | |
da2f0f74 | 8048 | struct bio *bio; |
22365979 | 8049 | bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags); |
da2f0f74 CM |
8050 | if (bio) |
8051 | bio_associate_current(bio); | |
8052 | return bio; | |
e65e1535 MX |
8053 | } |
8054 | ||
c1dc0896 MX |
8055 | static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root, |
8056 | struct inode *inode, | |
8057 | struct btrfs_dio_private *dip, | |
8058 | struct bio *bio, | |
8059 | u64 file_offset) | |
8060 | { | |
8061 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); | |
8062 | struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio); | |
8063 | int ret; | |
8064 | ||
8065 | /* | |
8066 | * We load all the csum data we need when we submit | |
8067 | * the first bio to reduce the csum tree search and | |
8068 | * contention. | |
8069 | */ | |
8070 | if (dip->logical_offset == file_offset) { | |
8071 | ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio, | |
8072 | file_offset); | |
8073 | if (ret) | |
8074 | return ret; | |
8075 | } | |
8076 | ||
8077 | if (bio == dip->orig_bio) | |
8078 | return 0; | |
8079 | ||
8080 | file_offset -= dip->logical_offset; | |
8081 | file_offset >>= inode->i_sb->s_blocksize_bits; | |
8082 | io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset); | |
8083 | ||
8084 | return 0; | |
8085 | } | |
8086 | ||
e65e1535 MX |
8087 | static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, |
8088 | int rw, u64 file_offset, int skip_sum, | |
c329861d | 8089 | int async_submit) |
e65e1535 | 8090 | { |
facc8a22 | 8091 | struct btrfs_dio_private *dip = bio->bi_private; |
e65e1535 MX |
8092 | int write = rw & REQ_WRITE; |
8093 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8094 | int ret; | |
8095 | ||
b812ce28 JB |
8096 | if (async_submit) |
8097 | async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers); | |
8098 | ||
e65e1535 | 8099 | bio_get(bio); |
5fd02043 JB |
8100 | |
8101 | if (!write) { | |
bfebd8b5 DS |
8102 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, |
8103 | BTRFS_WQ_ENDIO_DATA); | |
5fd02043 JB |
8104 | if (ret) |
8105 | goto err; | |
8106 | } | |
e65e1535 | 8107 | |
1ae39938 JB |
8108 | if (skip_sum) |
8109 | goto map; | |
8110 | ||
8111 | if (write && async_submit) { | |
e65e1535 MX |
8112 | ret = btrfs_wq_submit_bio(root->fs_info, |
8113 | inode, rw, bio, 0, 0, | |
8114 | file_offset, | |
8115 | __btrfs_submit_bio_start_direct_io, | |
8116 | __btrfs_submit_bio_done); | |
8117 | goto err; | |
1ae39938 JB |
8118 | } else if (write) { |
8119 | /* | |
8120 | * If we aren't doing async submit, calculate the csum of the | |
8121 | * bio now. | |
8122 | */ | |
8123 | ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1); | |
8124 | if (ret) | |
8125 | goto err; | |
23ea8e5a | 8126 | } else { |
c1dc0896 MX |
8127 | ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio, |
8128 | file_offset); | |
c2db1073 TI |
8129 | if (ret) |
8130 | goto err; | |
8131 | } | |
1ae39938 JB |
8132 | map: |
8133 | ret = btrfs_map_bio(root, rw, bio, 0, async_submit); | |
e65e1535 MX |
8134 | err: |
8135 | bio_put(bio); | |
8136 | return ret; | |
8137 | } | |
8138 | ||
8139 | static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip, | |
8140 | int skip_sum) | |
8141 | { | |
8142 | struct inode *inode = dip->inode; | |
8143 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
e65e1535 MX |
8144 | struct bio *bio; |
8145 | struct bio *orig_bio = dip->orig_bio; | |
8146 | struct bio_vec *bvec = orig_bio->bi_io_vec; | |
4f024f37 | 8147 | u64 start_sector = orig_bio->bi_iter.bi_sector; |
e65e1535 MX |
8148 | u64 file_offset = dip->logical_offset; |
8149 | u64 submit_len = 0; | |
8150 | u64 map_length; | |
8151 | int nr_pages = 0; | |
23ea8e5a | 8152 | int ret; |
1ae39938 | 8153 | int async_submit = 0; |
e65e1535 | 8154 | |
4f024f37 | 8155 | map_length = orig_bio->bi_iter.bi_size; |
53b381b3 | 8156 | ret = btrfs_map_block(root->fs_info, rw, start_sector << 9, |
e65e1535 | 8157 | &map_length, NULL, 0); |
7a5c3c9b | 8158 | if (ret) |
e65e1535 | 8159 | return -EIO; |
facc8a22 | 8160 | |
4f024f37 | 8161 | if (map_length >= orig_bio->bi_iter.bi_size) { |
02f57c7a | 8162 | bio = orig_bio; |
c1dc0896 | 8163 | dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED; |
02f57c7a JB |
8164 | goto submit; |
8165 | } | |
8166 | ||
53b381b3 | 8167 | /* async crcs make it difficult to collect full stripe writes. */ |
ffe2d203 | 8168 | if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 DW |
8169 | async_submit = 0; |
8170 | else | |
8171 | async_submit = 1; | |
8172 | ||
02f57c7a JB |
8173 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS); |
8174 | if (!bio) | |
8175 | return -ENOMEM; | |
7a5c3c9b | 8176 | |
02f57c7a JB |
8177 | bio->bi_private = dip; |
8178 | bio->bi_end_io = btrfs_end_dio_bio; | |
c1dc0896 | 8179 | btrfs_io_bio(bio)->logical = file_offset; |
02f57c7a JB |
8180 | atomic_inc(&dip->pending_bios); |
8181 | ||
e65e1535 | 8182 | while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) { |
ee39b432 | 8183 | if (map_length < submit_len + bvec->bv_len || |
e65e1535 | 8184 | bio_add_page(bio, bvec->bv_page, bvec->bv_len, |
ee39b432 | 8185 | bvec->bv_offset) < bvec->bv_len) { |
e65e1535 MX |
8186 | /* |
8187 | * inc the count before we submit the bio so | |
8188 | * we know the end IO handler won't happen before | |
8189 | * we inc the count. Otherwise, the dip might get freed | |
8190 | * before we're done setting it up | |
8191 | */ | |
8192 | atomic_inc(&dip->pending_bios); | |
8193 | ret = __btrfs_submit_dio_bio(bio, inode, rw, | |
8194 | file_offset, skip_sum, | |
c329861d | 8195 | async_submit); |
e65e1535 MX |
8196 | if (ret) { |
8197 | bio_put(bio); | |
8198 | atomic_dec(&dip->pending_bios); | |
8199 | goto out_err; | |
8200 | } | |
8201 | ||
e65e1535 MX |
8202 | start_sector += submit_len >> 9; |
8203 | file_offset += submit_len; | |
8204 | ||
8205 | submit_len = 0; | |
8206 | nr_pages = 0; | |
8207 | ||
8208 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, | |
8209 | start_sector, GFP_NOFS); | |
8210 | if (!bio) | |
8211 | goto out_err; | |
8212 | bio->bi_private = dip; | |
8213 | bio->bi_end_io = btrfs_end_dio_bio; | |
c1dc0896 | 8214 | btrfs_io_bio(bio)->logical = file_offset; |
e65e1535 | 8215 | |
4f024f37 | 8216 | map_length = orig_bio->bi_iter.bi_size; |
53b381b3 | 8217 | ret = btrfs_map_block(root->fs_info, rw, |
3ec706c8 | 8218 | start_sector << 9, |
e65e1535 MX |
8219 | &map_length, NULL, 0); |
8220 | if (ret) { | |
8221 | bio_put(bio); | |
8222 | goto out_err; | |
8223 | } | |
8224 | } else { | |
8225 | submit_len += bvec->bv_len; | |
67871254 | 8226 | nr_pages++; |
e65e1535 MX |
8227 | bvec++; |
8228 | } | |
8229 | } | |
8230 | ||
02f57c7a | 8231 | submit: |
e65e1535 | 8232 | ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum, |
c329861d | 8233 | async_submit); |
e65e1535 MX |
8234 | if (!ret) |
8235 | return 0; | |
8236 | ||
8237 | bio_put(bio); | |
8238 | out_err: | |
8239 | dip->errors = 1; | |
8240 | /* | |
8241 | * before atomic variable goto zero, we must | |
8242 | * make sure dip->errors is perceived to be set. | |
8243 | */ | |
4e857c58 | 8244 | smp_mb__before_atomic(); |
e65e1535 MX |
8245 | if (atomic_dec_and_test(&dip->pending_bios)) |
8246 | bio_io_error(dip->orig_bio); | |
8247 | ||
8248 | /* bio_end_io() will handle error, so we needn't return it */ | |
8249 | return 0; | |
8250 | } | |
8251 | ||
9be3395b CM |
8252 | static void btrfs_submit_direct(int rw, struct bio *dio_bio, |
8253 | struct inode *inode, loff_t file_offset) | |
4b46fce2 | 8254 | { |
61de718f FM |
8255 | struct btrfs_dio_private *dip = NULL; |
8256 | struct bio *io_bio = NULL; | |
23ea8e5a | 8257 | struct btrfs_io_bio *btrfs_bio; |
4b46fce2 | 8258 | int skip_sum; |
7b6d91da | 8259 | int write = rw & REQ_WRITE; |
4b46fce2 JB |
8260 | int ret = 0; |
8261 | ||
8262 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
8263 | ||
9be3395b | 8264 | io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS); |
9be3395b CM |
8265 | if (!io_bio) { |
8266 | ret = -ENOMEM; | |
8267 | goto free_ordered; | |
8268 | } | |
8269 | ||
c1dc0896 | 8270 | dip = kzalloc(sizeof(*dip), GFP_NOFS); |
4b46fce2 JB |
8271 | if (!dip) { |
8272 | ret = -ENOMEM; | |
61de718f | 8273 | goto free_ordered; |
4b46fce2 | 8274 | } |
4b46fce2 | 8275 | |
9be3395b | 8276 | dip->private = dio_bio->bi_private; |
4b46fce2 JB |
8277 | dip->inode = inode; |
8278 | dip->logical_offset = file_offset; | |
4f024f37 KO |
8279 | dip->bytes = dio_bio->bi_iter.bi_size; |
8280 | dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9; | |
9be3395b | 8281 | io_bio->bi_private = dip; |
9be3395b CM |
8282 | dip->orig_bio = io_bio; |
8283 | dip->dio_bio = dio_bio; | |
e65e1535 | 8284 | atomic_set(&dip->pending_bios, 0); |
c1dc0896 MX |
8285 | btrfs_bio = btrfs_io_bio(io_bio); |
8286 | btrfs_bio->logical = file_offset; | |
4b46fce2 | 8287 | |
c1dc0896 | 8288 | if (write) { |
9be3395b | 8289 | io_bio->bi_end_io = btrfs_endio_direct_write; |
c1dc0896 | 8290 | } else { |
9be3395b | 8291 | io_bio->bi_end_io = btrfs_endio_direct_read; |
c1dc0896 MX |
8292 | dip->subio_endio = btrfs_subio_endio_read; |
8293 | } | |
4b46fce2 | 8294 | |
e65e1535 MX |
8295 | ret = btrfs_submit_direct_hook(rw, dip, skip_sum); |
8296 | if (!ret) | |
eaf25d93 | 8297 | return; |
9be3395b | 8298 | |
23ea8e5a MX |
8299 | if (btrfs_bio->end_io) |
8300 | btrfs_bio->end_io(btrfs_bio, ret); | |
9be3395b | 8301 | |
4b46fce2 JB |
8302 | free_ordered: |
8303 | /* | |
61de718f FM |
8304 | * If we arrived here it means either we failed to submit the dip |
8305 | * or we either failed to clone the dio_bio or failed to allocate the | |
8306 | * dip. If we cloned the dio_bio and allocated the dip, we can just | |
8307 | * call bio_endio against our io_bio so that we get proper resource | |
8308 | * cleanup if we fail to submit the dip, otherwise, we must do the | |
8309 | * same as btrfs_endio_direct_[write|read] because we can't call these | |
8310 | * callbacks - they require an allocated dip and a clone of dio_bio. | |
4b46fce2 | 8311 | */ |
61de718f | 8312 | if (io_bio && dip) { |
4246a0b6 CH |
8313 | io_bio->bi_error = -EIO; |
8314 | bio_endio(io_bio); | |
61de718f FM |
8315 | /* |
8316 | * The end io callbacks free our dip, do the final put on io_bio | |
8317 | * and all the cleanup and final put for dio_bio (through | |
8318 | * dio_end_io()). | |
8319 | */ | |
8320 | dip = NULL; | |
8321 | io_bio = NULL; | |
8322 | } else { | |
8323 | if (write) { | |
8324 | struct btrfs_ordered_extent *ordered; | |
8325 | ||
8326 | ordered = btrfs_lookup_ordered_extent(inode, | |
8327 | file_offset); | |
8328 | set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); | |
8329 | /* | |
8330 | * Decrements our ref on the ordered extent and removes | |
8331 | * the ordered extent from the inode's ordered tree, | |
8332 | * doing all the proper resource cleanup such as for the | |
8333 | * reserved space and waking up any waiters for this | |
8334 | * ordered extent (through btrfs_remove_ordered_extent). | |
8335 | */ | |
8336 | btrfs_finish_ordered_io(ordered); | |
8337 | } else { | |
8338 | unlock_extent(&BTRFS_I(inode)->io_tree, file_offset, | |
8339 | file_offset + dio_bio->bi_iter.bi_size - 1); | |
8340 | } | |
4246a0b6 | 8341 | dio_bio->bi_error = -EIO; |
61de718f FM |
8342 | /* |
8343 | * Releases and cleans up our dio_bio, no need to bio_put() | |
8344 | * nor bio_endio()/bio_io_error() against dio_bio. | |
8345 | */ | |
8346 | dio_end_io(dio_bio, ret); | |
4b46fce2 | 8347 | } |
61de718f FM |
8348 | if (io_bio) |
8349 | bio_put(io_bio); | |
8350 | kfree(dip); | |
4b46fce2 JB |
8351 | } |
8352 | ||
6f673763 | 8353 | static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb, |
28060d5d | 8354 | const struct iov_iter *iter, loff_t offset) |
5a5f79b5 CM |
8355 | { |
8356 | int seg; | |
a1b75f7d | 8357 | int i; |
5a5f79b5 CM |
8358 | unsigned blocksize_mask = root->sectorsize - 1; |
8359 | ssize_t retval = -EINVAL; | |
5a5f79b5 CM |
8360 | |
8361 | if (offset & blocksize_mask) | |
8362 | goto out; | |
8363 | ||
28060d5d AV |
8364 | if (iov_iter_alignment(iter) & blocksize_mask) |
8365 | goto out; | |
a1b75f7d | 8366 | |
28060d5d | 8367 | /* If this is a write we don't need to check anymore */ |
6f673763 | 8368 | if (iov_iter_rw(iter) == WRITE) |
28060d5d AV |
8369 | return 0; |
8370 | /* | |
8371 | * Check to make sure we don't have duplicate iov_base's in this | |
8372 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
8373 | * when reading back. | |
8374 | */ | |
8375 | for (seg = 0; seg < iter->nr_segs; seg++) { | |
8376 | for (i = seg + 1; i < iter->nr_segs; i++) { | |
8377 | if (iter->iov[seg].iov_base == iter->iov[i].iov_base) | |
a1b75f7d JB |
8378 | goto out; |
8379 | } | |
5a5f79b5 CM |
8380 | } |
8381 | retval = 0; | |
8382 | out: | |
8383 | return retval; | |
8384 | } | |
eb838e73 | 8385 | |
22c6186e OS |
8386 | static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter, |
8387 | loff_t offset) | |
16432985 | 8388 | { |
4b46fce2 JB |
8389 | struct file *file = iocb->ki_filp; |
8390 | struct inode *inode = file->f_mapping->host; | |
50745b0a | 8391 | struct btrfs_root *root = BTRFS_I(inode)->root; |
8392 | struct btrfs_dio_data dio_data = { 0 }; | |
0934856d | 8393 | size_t count = 0; |
2e60a51e | 8394 | int flags = 0; |
38851cc1 MX |
8395 | bool wakeup = true; |
8396 | bool relock = false; | |
0934856d | 8397 | ssize_t ret; |
4b46fce2 | 8398 | |
6f673763 | 8399 | if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset)) |
5a5f79b5 | 8400 | return 0; |
3f7c579c | 8401 | |
fe0f07d0 | 8402 | inode_dio_begin(inode); |
4e857c58 | 8403 | smp_mb__after_atomic(); |
38851cc1 | 8404 | |
0e267c44 | 8405 | /* |
41bd9ca4 MX |
8406 | * The generic stuff only does filemap_write_and_wait_range, which |
8407 | * isn't enough if we've written compressed pages to this area, so | |
8408 | * we need to flush the dirty pages again to make absolutely sure | |
8409 | * that any outstanding dirty pages are on disk. | |
0e267c44 | 8410 | */ |
a6cbcd4a | 8411 | count = iov_iter_count(iter); |
41bd9ca4 MX |
8412 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, |
8413 | &BTRFS_I(inode)->runtime_flags)) | |
9a025a08 WS |
8414 | filemap_fdatawrite_range(inode->i_mapping, offset, |
8415 | offset + count - 1); | |
0e267c44 | 8416 | |
6f673763 | 8417 | if (iov_iter_rw(iter) == WRITE) { |
38851cc1 MX |
8418 | /* |
8419 | * If the write DIO is beyond the EOF, we need update | |
8420 | * the isize, but it is protected by i_mutex. So we can | |
8421 | * not unlock the i_mutex at this case. | |
8422 | */ | |
8423 | if (offset + count <= inode->i_size) { | |
8424 | mutex_unlock(&inode->i_mutex); | |
8425 | relock = true; | |
8426 | } | |
0934856d MX |
8427 | ret = btrfs_delalloc_reserve_space(inode, count); |
8428 | if (ret) | |
38851cc1 | 8429 | goto out; |
50745b0a | 8430 | dio_data.outstanding_extents = div64_u64(count + |
e1cbbfa5 JB |
8431 | BTRFS_MAX_EXTENT_SIZE - 1, |
8432 | BTRFS_MAX_EXTENT_SIZE); | |
8433 | ||
8434 | /* | |
8435 | * We need to know how many extents we reserved so that we can | |
8436 | * do the accounting properly if we go over the number we | |
8437 | * originally calculated. Abuse current->journal_info for this. | |
8438 | */ | |
50745b0a | 8439 | dio_data.reserve = round_up(count, root->sectorsize); |
8440 | current->journal_info = &dio_data; | |
ee39b432 DS |
8441 | } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK, |
8442 | &BTRFS_I(inode)->runtime_flags)) { | |
fe0f07d0 | 8443 | inode_dio_end(inode); |
38851cc1 MX |
8444 | flags = DIO_LOCKING | DIO_SKIP_HOLES; |
8445 | wakeup = false; | |
0934856d MX |
8446 | } |
8447 | ||
17f8c842 OS |
8448 | ret = __blockdev_direct_IO(iocb, inode, |
8449 | BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev, | |
8450 | iter, offset, btrfs_get_blocks_direct, NULL, | |
8451 | btrfs_submit_direct, flags); | |
6f673763 | 8452 | if (iov_iter_rw(iter) == WRITE) { |
e1cbbfa5 | 8453 | current->journal_info = NULL; |
ddba1bfc | 8454 | if (ret < 0 && ret != -EIOCBQUEUED) { |
50745b0a | 8455 | if (dio_data.reserve) |
8456 | btrfs_delalloc_release_space(inode, | |
8457 | dio_data.reserve); | |
ddba1bfc | 8458 | } else if (ret >= 0 && (size_t)ret < count) |
0934856d MX |
8459 | btrfs_delalloc_release_space(inode, |
8460 | count - (size_t)ret); | |
0934856d | 8461 | } |
38851cc1 | 8462 | out: |
2e60a51e | 8463 | if (wakeup) |
fe0f07d0 | 8464 | inode_dio_end(inode); |
38851cc1 MX |
8465 | if (relock) |
8466 | mutex_lock(&inode->i_mutex); | |
0934856d MX |
8467 | |
8468 | return ret; | |
16432985 CM |
8469 | } |
8470 | ||
05dadc09 TI |
8471 | #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC) |
8472 | ||
1506fcc8 YS |
8473 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
8474 | __u64 start, __u64 len) | |
8475 | { | |
05dadc09 TI |
8476 | int ret; |
8477 | ||
8478 | ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS); | |
8479 | if (ret) | |
8480 | return ret; | |
8481 | ||
ec29ed5b | 8482 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
8483 | } |
8484 | ||
a52d9a80 | 8485 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 8486 | { |
d1310b2e CM |
8487 | struct extent_io_tree *tree; |
8488 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 8489 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 8490 | } |
1832a6d5 | 8491 | |
a52d9a80 | 8492 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 8493 | { |
d1310b2e | 8494 | struct extent_io_tree *tree; |
b888db2b CM |
8495 | |
8496 | ||
8497 | if (current->flags & PF_MEMALLOC) { | |
8498 | redirty_page_for_writepage(wbc, page); | |
8499 | unlock_page(page); | |
8500 | return 0; | |
8501 | } | |
d1310b2e | 8502 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 8503 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
8504 | } |
8505 | ||
48a3b636 ES |
8506 | static int btrfs_writepages(struct address_space *mapping, |
8507 | struct writeback_control *wbc) | |
b293f02e | 8508 | { |
d1310b2e | 8509 | struct extent_io_tree *tree; |
771ed689 | 8510 | |
d1310b2e | 8511 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
8512 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
8513 | } | |
8514 | ||
3ab2fb5a CM |
8515 | static int |
8516 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
8517 | struct list_head *pages, unsigned nr_pages) | |
8518 | { | |
d1310b2e CM |
8519 | struct extent_io_tree *tree; |
8520 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
8521 | return extent_readpages(tree, mapping, pages, nr_pages, |
8522 | btrfs_get_extent); | |
8523 | } | |
e6dcd2dc | 8524 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 8525 | { |
d1310b2e CM |
8526 | struct extent_io_tree *tree; |
8527 | struct extent_map_tree *map; | |
a52d9a80 | 8528 | int ret; |
8c2383c3 | 8529 | |
d1310b2e CM |
8530 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
8531 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 8532 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
8533 | if (ret == 1) { |
8534 | ClearPagePrivate(page); | |
8535 | set_page_private(page, 0); | |
8536 | page_cache_release(page); | |
39279cc3 | 8537 | } |
a52d9a80 | 8538 | return ret; |
39279cc3 CM |
8539 | } |
8540 | ||
e6dcd2dc CM |
8541 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
8542 | { | |
98509cfc CM |
8543 | if (PageWriteback(page) || PageDirty(page)) |
8544 | return 0; | |
b335b003 | 8545 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
8546 | } |
8547 | ||
d47992f8 LC |
8548 | static void btrfs_invalidatepage(struct page *page, unsigned int offset, |
8549 | unsigned int length) | |
39279cc3 | 8550 | { |
5fd02043 | 8551 | struct inode *inode = page->mapping->host; |
d1310b2e | 8552 | struct extent_io_tree *tree; |
e6dcd2dc | 8553 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 8554 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
8555 | u64 page_start = page_offset(page); |
8556 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
131e404a | 8557 | int inode_evicting = inode->i_state & I_FREEING; |
39279cc3 | 8558 | |
8b62b72b CM |
8559 | /* |
8560 | * we have the page locked, so new writeback can't start, | |
8561 | * and the dirty bit won't be cleared while we are here. | |
8562 | * | |
8563 | * Wait for IO on this page so that we can safely clear | |
8564 | * the PagePrivate2 bit and do ordered accounting | |
8565 | */ | |
e6dcd2dc | 8566 | wait_on_page_writeback(page); |
8b62b72b | 8567 | |
5fd02043 | 8568 | tree = &BTRFS_I(inode)->io_tree; |
e6dcd2dc CM |
8569 | if (offset) { |
8570 | btrfs_releasepage(page, GFP_NOFS); | |
8571 | return; | |
8572 | } | |
131e404a FDBM |
8573 | |
8574 | if (!inode_evicting) | |
8575 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state); | |
8576 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
e6dcd2dc | 8577 | if (ordered) { |
eb84ae03 CM |
8578 | /* |
8579 | * IO on this page will never be started, so we need | |
8580 | * to account for any ordered extents now | |
8581 | */ | |
131e404a FDBM |
8582 | if (!inode_evicting) |
8583 | clear_extent_bit(tree, page_start, page_end, | |
8584 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
8585 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | | |
8586 | EXTENT_DEFRAG, 1, 0, &cached_state, | |
8587 | GFP_NOFS); | |
8b62b72b CM |
8588 | /* |
8589 | * whoever cleared the private bit is responsible | |
8590 | * for the finish_ordered_io | |
8591 | */ | |
77cef2ec JB |
8592 | if (TestClearPagePrivate2(page)) { |
8593 | struct btrfs_ordered_inode_tree *tree; | |
8594 | u64 new_len; | |
8595 | ||
8596 | tree = &BTRFS_I(inode)->ordered_tree; | |
8597 | ||
8598 | spin_lock_irq(&tree->lock); | |
8599 | set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags); | |
8600 | new_len = page_start - ordered->file_offset; | |
8601 | if (new_len < ordered->truncated_len) | |
8602 | ordered->truncated_len = new_len; | |
8603 | spin_unlock_irq(&tree->lock); | |
8604 | ||
8605 | if (btrfs_dec_test_ordered_pending(inode, &ordered, | |
8606 | page_start, | |
8607 | PAGE_CACHE_SIZE, 1)) | |
8608 | btrfs_finish_ordered_io(ordered); | |
8b62b72b | 8609 | } |
e6dcd2dc | 8610 | btrfs_put_ordered_extent(ordered); |
131e404a FDBM |
8611 | if (!inode_evicting) { |
8612 | cached_state = NULL; | |
8613 | lock_extent_bits(tree, page_start, page_end, 0, | |
8614 | &cached_state); | |
8615 | } | |
8616 | } | |
8617 | ||
8618 | if (!inode_evicting) { | |
8619 | clear_extent_bit(tree, page_start, page_end, | |
8620 | EXTENT_LOCKED | EXTENT_DIRTY | | |
8621 | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | | |
8622 | EXTENT_DEFRAG, 1, 1, | |
8623 | &cached_state, GFP_NOFS); | |
8624 | ||
8625 | __btrfs_releasepage(page, GFP_NOFS); | |
e6dcd2dc | 8626 | } |
e6dcd2dc | 8627 | |
4a096752 | 8628 | ClearPageChecked(page); |
9ad6b7bc | 8629 | if (PagePrivate(page)) { |
9ad6b7bc CM |
8630 | ClearPagePrivate(page); |
8631 | set_page_private(page, 0); | |
8632 | page_cache_release(page); | |
8633 | } | |
39279cc3 CM |
8634 | } |
8635 | ||
9ebefb18 CM |
8636 | /* |
8637 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
8638 | * called from a page fault handler when a page is first dirtied. Hence we must | |
8639 | * be careful to check for EOF conditions here. We set the page up correctly | |
8640 | * for a written page which means we get ENOSPC checking when writing into | |
8641 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
8642 | * support these features. | |
8643 | * | |
8644 | * We are not allowed to take the i_mutex here so we have to play games to | |
8645 | * protect against truncate races as the page could now be beyond EOF. Because | |
8646 | * vmtruncate() writes the inode size before removing pages, once we have the | |
8647 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
8648 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
8649 | * unlock the page. | |
8650 | */ | |
c2ec175c | 8651 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 8652 | { |
c2ec175c | 8653 | struct page *page = vmf->page; |
496ad9aa | 8654 | struct inode *inode = file_inode(vma->vm_file); |
1832a6d5 | 8655 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
8656 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
8657 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 8658 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
8659 | char *kaddr; |
8660 | unsigned long zero_start; | |
9ebefb18 | 8661 | loff_t size; |
1832a6d5 | 8662 | int ret; |
9998eb70 | 8663 | int reserved = 0; |
a52d9a80 | 8664 | u64 page_start; |
e6dcd2dc | 8665 | u64 page_end; |
9ebefb18 | 8666 | |
b2b5ef5c | 8667 | sb_start_pagefault(inode->i_sb); |
0ca1f7ce | 8668 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 8669 | if (!ret) { |
e41f941a | 8670 | ret = file_update_time(vma->vm_file); |
9998eb70 CM |
8671 | reserved = 1; |
8672 | } | |
56a76f82 NP |
8673 | if (ret) { |
8674 | if (ret == -ENOMEM) | |
8675 | ret = VM_FAULT_OOM; | |
8676 | else /* -ENOSPC, -EIO, etc */ | |
8677 | ret = VM_FAULT_SIGBUS; | |
9998eb70 CM |
8678 | if (reserved) |
8679 | goto out; | |
8680 | goto out_noreserve; | |
56a76f82 | 8681 | } |
1832a6d5 | 8682 | |
56a76f82 | 8683 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 8684 | again: |
9ebefb18 | 8685 | lock_page(page); |
9ebefb18 | 8686 | size = i_size_read(inode); |
e6dcd2dc CM |
8687 | page_start = page_offset(page); |
8688 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 8689 | |
9ebefb18 | 8690 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 8691 | (page_start >= size)) { |
9ebefb18 CM |
8692 | /* page got truncated out from underneath us */ |
8693 | goto out_unlock; | |
8694 | } | |
e6dcd2dc CM |
8695 | wait_on_page_writeback(page); |
8696 | ||
d0082371 | 8697 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state); |
e6dcd2dc CM |
8698 | set_page_extent_mapped(page); |
8699 | ||
eb84ae03 CM |
8700 | /* |
8701 | * we can't set the delalloc bits if there are pending ordered | |
8702 | * extents. Drop our locks and wait for them to finish | |
8703 | */ | |
e6dcd2dc CM |
8704 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
8705 | if (ordered) { | |
2ac55d41 JB |
8706 | unlock_extent_cached(io_tree, page_start, page_end, |
8707 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 8708 | unlock_page(page); |
eb84ae03 | 8709 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
8710 | btrfs_put_ordered_extent(ordered); |
8711 | goto again; | |
8712 | } | |
8713 | ||
fbf19087 JB |
8714 | /* |
8715 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
8716 | * if it was already dirty, so for space accounting reasons we need to | |
8717 | * clear any delalloc bits for the range we are fixing to save. There | |
8718 | * is probably a better way to do this, but for now keep consistent with | |
8719 | * prepare_pages in the normal write path. | |
8720 | */ | |
2ac55d41 | 8721 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
9e8a4a8b LB |
8722 | EXTENT_DIRTY | EXTENT_DELALLOC | |
8723 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, | |
2ac55d41 | 8724 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 8725 | |
2ac55d41 JB |
8726 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
8727 | &cached_state); | |
9ed74f2d | 8728 | if (ret) { |
2ac55d41 JB |
8729 | unlock_extent_cached(io_tree, page_start, page_end, |
8730 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
8731 | ret = VM_FAULT_SIGBUS; |
8732 | goto out_unlock; | |
8733 | } | |
e6dcd2dc | 8734 | ret = 0; |
9ebefb18 CM |
8735 | |
8736 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 8737 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 8738 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 8739 | else |
e6dcd2dc | 8740 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 8741 | |
e6dcd2dc CM |
8742 | if (zero_start != PAGE_CACHE_SIZE) { |
8743 | kaddr = kmap(page); | |
8744 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
8745 | flush_dcache_page(page); | |
8746 | kunmap(page); | |
8747 | } | |
247e743c | 8748 | ClearPageChecked(page); |
e6dcd2dc | 8749 | set_page_dirty(page); |
50a9b214 | 8750 | SetPageUptodate(page); |
5a3f23d5 | 8751 | |
257c62e1 CM |
8752 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
8753 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
46d8bc34 | 8754 | BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit; |
257c62e1 | 8755 | |
2ac55d41 | 8756 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
8757 | |
8758 | out_unlock: | |
b2b5ef5c JK |
8759 | if (!ret) { |
8760 | sb_end_pagefault(inode->i_sb); | |
50a9b214 | 8761 | return VM_FAULT_LOCKED; |
b2b5ef5c | 8762 | } |
9ebefb18 | 8763 | unlock_page(page); |
1832a6d5 | 8764 | out: |
ec39e180 | 8765 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 8766 | out_noreserve: |
b2b5ef5c | 8767 | sb_end_pagefault(inode->i_sb); |
9ebefb18 CM |
8768 | return ret; |
8769 | } | |
8770 | ||
a41ad394 | 8771 | static int btrfs_truncate(struct inode *inode) |
39279cc3 CM |
8772 | { |
8773 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
fcb80c2a | 8774 | struct btrfs_block_rsv *rsv; |
a71754fc | 8775 | int ret = 0; |
3893e33b | 8776 | int err = 0; |
39279cc3 | 8777 | struct btrfs_trans_handle *trans; |
dbe674a9 | 8778 | u64 mask = root->sectorsize - 1; |
07127184 | 8779 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 | 8780 | |
0ef8b726 JB |
8781 | ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask), |
8782 | (u64)-1); | |
8783 | if (ret) | |
8784 | return ret; | |
39279cc3 | 8785 | |
fcb80c2a JB |
8786 | /* |
8787 | * Yes ladies and gentelment, this is indeed ugly. The fact is we have | |
8788 | * 3 things going on here | |
8789 | * | |
8790 | * 1) We need to reserve space for our orphan item and the space to | |
8791 | * delete our orphan item. Lord knows we don't want to have a dangling | |
8792 | * orphan item because we didn't reserve space to remove it. | |
8793 | * | |
8794 | * 2) We need to reserve space to update our inode. | |
8795 | * | |
8796 | * 3) We need to have something to cache all the space that is going to | |
8797 | * be free'd up by the truncate operation, but also have some slack | |
8798 | * space reserved in case it uses space during the truncate (thank you | |
8799 | * very much snapshotting). | |
8800 | * | |
8801 | * And we need these to all be seperate. The fact is we can use alot of | |
8802 | * space doing the truncate, and we have no earthly idea how much space | |
8803 | * we will use, so we need the truncate reservation to be seperate so it | |
8804 | * doesn't end up using space reserved for updating the inode or | |
8805 | * removing the orphan item. We also need to be able to stop the | |
8806 | * transaction and start a new one, which means we need to be able to | |
8807 | * update the inode several times, and we have no idea of knowing how | |
8808 | * many times that will be, so we can't just reserve 1 item for the | |
8809 | * entirety of the opration, so that has to be done seperately as well. | |
8810 | * Then there is the orphan item, which does indeed need to be held on | |
8811 | * to for the whole operation, and we need nobody to touch this reserved | |
8812 | * space except the orphan code. | |
8813 | * | |
8814 | * So that leaves us with | |
8815 | * | |
8816 | * 1) root->orphan_block_rsv - for the orphan deletion. | |
8817 | * 2) rsv - for the truncate reservation, which we will steal from the | |
8818 | * transaction reservation. | |
8819 | * 3) fs_info->trans_block_rsv - this will have 1 items worth left for | |
8820 | * updating the inode. | |
8821 | */ | |
66d8f3dd | 8822 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
fcb80c2a JB |
8823 | if (!rsv) |
8824 | return -ENOMEM; | |
4a338542 | 8825 | rsv->size = min_size; |
ca7e70f5 | 8826 | rsv->failfast = 1; |
f0cd846e | 8827 | |
907cbceb | 8828 | /* |
07127184 | 8829 | * 1 for the truncate slack space |
907cbceb JB |
8830 | * 1 for updating the inode. |
8831 | */ | |
f3fe820c | 8832 | trans = btrfs_start_transaction(root, 2); |
fcb80c2a JB |
8833 | if (IS_ERR(trans)) { |
8834 | err = PTR_ERR(trans); | |
8835 | goto out; | |
8836 | } | |
f0cd846e | 8837 | |
907cbceb JB |
8838 | /* Migrate the slack space for the truncate to our reserve */ |
8839 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv, | |
8840 | min_size); | |
fcb80c2a | 8841 | BUG_ON(ret); |
f0cd846e | 8842 | |
5dc562c5 JB |
8843 | /* |
8844 | * So if we truncate and then write and fsync we normally would just | |
8845 | * write the extents that changed, which is a problem if we need to | |
8846 | * first truncate that entire inode. So set this flag so we write out | |
8847 | * all of the extents in the inode to the sync log so we're completely | |
8848 | * safe. | |
8849 | */ | |
8850 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags); | |
ca7e70f5 | 8851 | trans->block_rsv = rsv; |
907cbceb | 8852 | |
8082510e YZ |
8853 | while (1) { |
8854 | ret = btrfs_truncate_inode_items(trans, root, inode, | |
8855 | inode->i_size, | |
8856 | BTRFS_EXTENT_DATA_KEY); | |
28ed1345 | 8857 | if (ret != -ENOSPC && ret != -EAGAIN) { |
3893e33b | 8858 | err = ret; |
8082510e | 8859 | break; |
3893e33b | 8860 | } |
39279cc3 | 8861 | |
fcb80c2a | 8862 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e | 8863 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
8864 | if (ret) { |
8865 | err = ret; | |
8866 | break; | |
8867 | } | |
ca7e70f5 | 8868 | |
8082510e | 8869 | btrfs_end_transaction(trans, root); |
b53d3f5d | 8870 | btrfs_btree_balance_dirty(root); |
ca7e70f5 JB |
8871 | |
8872 | trans = btrfs_start_transaction(root, 2); | |
8873 | if (IS_ERR(trans)) { | |
8874 | ret = err = PTR_ERR(trans); | |
8875 | trans = NULL; | |
8876 | break; | |
8877 | } | |
8878 | ||
8879 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, | |
8880 | rsv, min_size); | |
8881 | BUG_ON(ret); /* shouldn't happen */ | |
8882 | trans->block_rsv = rsv; | |
8082510e YZ |
8883 | } |
8884 | ||
8885 | if (ret == 0 && inode->i_nlink > 0) { | |
fcb80c2a | 8886 | trans->block_rsv = root->orphan_block_rsv; |
8082510e | 8887 | ret = btrfs_orphan_del(trans, inode); |
3893e33b JB |
8888 | if (ret) |
8889 | err = ret; | |
8082510e YZ |
8890 | } |
8891 | ||
917c16b2 CM |
8892 | if (trans) { |
8893 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
8894 | ret = btrfs_update_inode(trans, root, inode); | |
8895 | if (ret && !err) | |
8896 | err = ret; | |
7b128766 | 8897 | |
7ad85bb7 | 8898 | ret = btrfs_end_transaction(trans, root); |
b53d3f5d | 8899 | btrfs_btree_balance_dirty(root); |
917c16b2 | 8900 | } |
fcb80c2a JB |
8901 | |
8902 | out: | |
8903 | btrfs_free_block_rsv(root, rsv); | |
8904 | ||
3893e33b JB |
8905 | if (ret && !err) |
8906 | err = ret; | |
a41ad394 | 8907 | |
3893e33b | 8908 | return err; |
39279cc3 CM |
8909 | } |
8910 | ||
d352ac68 CM |
8911 | /* |
8912 | * create a new subvolume directory/inode (helper for the ioctl). | |
8913 | */ | |
d2fb3437 | 8914 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
63541927 FDBM |
8915 | struct btrfs_root *new_root, |
8916 | struct btrfs_root *parent_root, | |
8917 | u64 new_dirid) | |
39279cc3 | 8918 | { |
39279cc3 | 8919 | struct inode *inode; |
76dda93c | 8920 | int err; |
00e4e6b3 | 8921 | u64 index = 0; |
39279cc3 | 8922 | |
12fc9d09 FA |
8923 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
8924 | new_dirid, new_dirid, | |
8925 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
8926 | &index); | |
54aa1f4d | 8927 | if (IS_ERR(inode)) |
f46b5a66 | 8928 | return PTR_ERR(inode); |
39279cc3 CM |
8929 | inode->i_op = &btrfs_dir_inode_operations; |
8930 | inode->i_fop = &btrfs_dir_file_operations; | |
8931 | ||
bfe86848 | 8932 | set_nlink(inode, 1); |
dbe674a9 | 8933 | btrfs_i_size_write(inode, 0); |
b0d5d10f | 8934 | unlock_new_inode(inode); |
3b96362c | 8935 | |
63541927 FDBM |
8936 | err = btrfs_subvol_inherit_props(trans, new_root, parent_root); |
8937 | if (err) | |
8938 | btrfs_err(new_root->fs_info, | |
351fd353 | 8939 | "error inheriting subvolume %llu properties: %d", |
63541927 FDBM |
8940 | new_root->root_key.objectid, err); |
8941 | ||
76dda93c | 8942 | err = btrfs_update_inode(trans, new_root, inode); |
cb8e7090 | 8943 | |
76dda93c | 8944 | iput(inode); |
ce598979 | 8945 | return err; |
39279cc3 CM |
8946 | } |
8947 | ||
39279cc3 CM |
8948 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
8949 | { | |
8950 | struct btrfs_inode *ei; | |
2ead6ae7 | 8951 | struct inode *inode; |
39279cc3 CM |
8952 | |
8953 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
8954 | if (!ei) | |
8955 | return NULL; | |
2ead6ae7 YZ |
8956 | |
8957 | ei->root = NULL; | |
2ead6ae7 | 8958 | ei->generation = 0; |
15ee9bc7 | 8959 | ei->last_trans = 0; |
257c62e1 | 8960 | ei->last_sub_trans = 0; |
e02119d5 | 8961 | ei->logged_trans = 0; |
2ead6ae7 | 8962 | ei->delalloc_bytes = 0; |
47059d93 | 8963 | ei->defrag_bytes = 0; |
2ead6ae7 YZ |
8964 | ei->disk_i_size = 0; |
8965 | ei->flags = 0; | |
7709cde3 | 8966 | ei->csum_bytes = 0; |
2ead6ae7 | 8967 | ei->index_cnt = (u64)-1; |
67de1176 | 8968 | ei->dir_index = 0; |
2ead6ae7 | 8969 | ei->last_unlink_trans = 0; |
46d8bc34 | 8970 | ei->last_log_commit = 0; |
2ead6ae7 | 8971 | |
9e0baf60 JB |
8972 | spin_lock_init(&ei->lock); |
8973 | ei->outstanding_extents = 0; | |
8974 | ei->reserved_extents = 0; | |
2ead6ae7 | 8975 | |
72ac3c0d | 8976 | ei->runtime_flags = 0; |
261507a0 | 8977 | ei->force_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 8978 | |
16cdcec7 MX |
8979 | ei->delayed_node = NULL; |
8980 | ||
9cc97d64 | 8981 | ei->i_otime.tv_sec = 0; |
8982 | ei->i_otime.tv_nsec = 0; | |
8983 | ||
2ead6ae7 | 8984 | inode = &ei->vfs_inode; |
a8067e02 | 8985 | extent_map_tree_init(&ei->extent_tree); |
f993c883 DS |
8986 | extent_io_tree_init(&ei->io_tree, &inode->i_data); |
8987 | extent_io_tree_init(&ei->io_failure_tree, &inode->i_data); | |
0b32f4bb JB |
8988 | ei->io_tree.track_uptodate = 1; |
8989 | ei->io_failure_tree.track_uptodate = 1; | |
b812ce28 | 8990 | atomic_set(&ei->sync_writers, 0); |
2ead6ae7 | 8991 | mutex_init(&ei->log_mutex); |
f248679e | 8992 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 8993 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
2ead6ae7 | 8994 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
2ead6ae7 YZ |
8995 | RB_CLEAR_NODE(&ei->rb_node); |
8996 | ||
8997 | return inode; | |
39279cc3 CM |
8998 | } |
8999 | ||
aaedb55b JB |
9000 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
9001 | void btrfs_test_destroy_inode(struct inode *inode) | |
9002 | { | |
9003 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); | |
9004 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); | |
9005 | } | |
9006 | #endif | |
9007 | ||
fa0d7e3d NP |
9008 | static void btrfs_i_callback(struct rcu_head *head) |
9009 | { | |
9010 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
9011 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); |
9012 | } | |
9013 | ||
39279cc3 CM |
9014 | void btrfs_destroy_inode(struct inode *inode) |
9015 | { | |
e6dcd2dc | 9016 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
9017 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9018 | ||
b3d9b7a3 | 9019 | WARN_ON(!hlist_empty(&inode->i_dentry)); |
39279cc3 | 9020 | WARN_ON(inode->i_data.nrpages); |
9e0baf60 JB |
9021 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
9022 | WARN_ON(BTRFS_I(inode)->reserved_extents); | |
7709cde3 JB |
9023 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
9024 | WARN_ON(BTRFS_I(inode)->csum_bytes); | |
47059d93 | 9025 | WARN_ON(BTRFS_I(inode)->defrag_bytes); |
39279cc3 | 9026 | |
a6dbd429 JB |
9027 | /* |
9028 | * This can happen where we create an inode, but somebody else also | |
9029 | * created the same inode and we need to destroy the one we already | |
9030 | * created. | |
9031 | */ | |
9032 | if (!root) | |
9033 | goto free; | |
9034 | ||
8a35d95f JB |
9035 | if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM, |
9036 | &BTRFS_I(inode)->runtime_flags)) { | |
c2cf52eb | 9037 | btrfs_info(root->fs_info, "inode %llu still on the orphan list", |
c1c9ff7c | 9038 | btrfs_ino(inode)); |
8a35d95f | 9039 | atomic_dec(&root->orphan_inodes); |
7b128766 | 9040 | } |
7b128766 | 9041 | |
d397712b | 9042 | while (1) { |
e6dcd2dc CM |
9043 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
9044 | if (!ordered) | |
9045 | break; | |
9046 | else { | |
c2cf52eb | 9047 | btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup", |
c1c9ff7c | 9048 | ordered->file_offset, ordered->len); |
e6dcd2dc CM |
9049 | btrfs_remove_ordered_extent(inode, ordered); |
9050 | btrfs_put_ordered_extent(ordered); | |
9051 | btrfs_put_ordered_extent(ordered); | |
9052 | } | |
9053 | } | |
5d4f98a2 | 9054 | inode_tree_del(inode); |
5b21f2ed | 9055 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 9056 | free: |
fa0d7e3d | 9057 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
9058 | } |
9059 | ||
45321ac5 | 9060 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
9061 | { |
9062 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 9063 | |
6379ef9f NA |
9064 | if (root == NULL) |
9065 | return 1; | |
9066 | ||
fa6ac876 | 9067 | /* the snap/subvol tree is on deleting */ |
69e9c6c6 | 9068 | if (btrfs_root_refs(&root->root_item) == 0) |
45321ac5 | 9069 | return 1; |
76dda93c | 9070 | else |
45321ac5 | 9071 | return generic_drop_inode(inode); |
76dda93c YZ |
9072 | } |
9073 | ||
0ee0fda0 | 9074 | static void init_once(void *foo) |
39279cc3 CM |
9075 | { |
9076 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
9077 | ||
9078 | inode_init_once(&ei->vfs_inode); | |
9079 | } | |
9080 | ||
9081 | void btrfs_destroy_cachep(void) | |
9082 | { | |
8c0a8537 KS |
9083 | /* |
9084 | * Make sure all delayed rcu free inodes are flushed before we | |
9085 | * destroy cache. | |
9086 | */ | |
9087 | rcu_barrier(); | |
39279cc3 CM |
9088 | if (btrfs_inode_cachep) |
9089 | kmem_cache_destroy(btrfs_inode_cachep); | |
9090 | if (btrfs_trans_handle_cachep) | |
9091 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
9092 | if (btrfs_transaction_cachep) | |
9093 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
9094 | if (btrfs_path_cachep) |
9095 | kmem_cache_destroy(btrfs_path_cachep); | |
dc89e982 JB |
9096 | if (btrfs_free_space_cachep) |
9097 | kmem_cache_destroy(btrfs_free_space_cachep); | |
8ccf6f19 MX |
9098 | if (btrfs_delalloc_work_cachep) |
9099 | kmem_cache_destroy(btrfs_delalloc_work_cachep); | |
39279cc3 CM |
9100 | } |
9101 | ||
9102 | int btrfs_init_cachep(void) | |
9103 | { | |
837e1972 | 9104 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode", |
9601e3f6 CH |
9105 | sizeof(struct btrfs_inode), 0, |
9106 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
9107 | if (!btrfs_inode_cachep) |
9108 | goto fail; | |
9601e3f6 | 9109 | |
837e1972 | 9110 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
9601e3f6 CH |
9111 | sizeof(struct btrfs_trans_handle), 0, |
9112 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
9113 | if (!btrfs_trans_handle_cachep) |
9114 | goto fail; | |
9601e3f6 | 9115 | |
837e1972 | 9116 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction", |
9601e3f6 CH |
9117 | sizeof(struct btrfs_transaction), 0, |
9118 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
9119 | if (!btrfs_transaction_cachep) |
9120 | goto fail; | |
9601e3f6 | 9121 | |
837e1972 | 9122 | btrfs_path_cachep = kmem_cache_create("btrfs_path", |
9601e3f6 CH |
9123 | sizeof(struct btrfs_path), 0, |
9124 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
9125 | if (!btrfs_path_cachep) |
9126 | goto fail; | |
9601e3f6 | 9127 | |
837e1972 | 9128 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space", |
dc89e982 JB |
9129 | sizeof(struct btrfs_free_space), 0, |
9130 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
9131 | if (!btrfs_free_space_cachep) | |
9132 | goto fail; | |
9133 | ||
8ccf6f19 MX |
9134 | btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work", |
9135 | sizeof(struct btrfs_delalloc_work), 0, | |
9136 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, | |
9137 | NULL); | |
9138 | if (!btrfs_delalloc_work_cachep) | |
9139 | goto fail; | |
9140 | ||
39279cc3 CM |
9141 | return 0; |
9142 | fail: | |
9143 | btrfs_destroy_cachep(); | |
9144 | return -ENOMEM; | |
9145 | } | |
9146 | ||
9147 | static int btrfs_getattr(struct vfsmount *mnt, | |
9148 | struct dentry *dentry, struct kstat *stat) | |
9149 | { | |
df0af1a5 | 9150 | u64 delalloc_bytes; |
2b0143b5 | 9151 | struct inode *inode = d_inode(dentry); |
fadc0d8b DS |
9152 | u32 blocksize = inode->i_sb->s_blocksize; |
9153 | ||
39279cc3 | 9154 | generic_fillattr(inode, stat); |
0ee5dc67 | 9155 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
d6667462 | 9156 | stat->blksize = PAGE_CACHE_SIZE; |
df0af1a5 MX |
9157 | |
9158 | spin_lock(&BTRFS_I(inode)->lock); | |
9159 | delalloc_bytes = BTRFS_I(inode)->delalloc_bytes; | |
9160 | spin_unlock(&BTRFS_I(inode)->lock); | |
fadc0d8b | 9161 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
df0af1a5 | 9162 | ALIGN(delalloc_bytes, blocksize)) >> 9; |
39279cc3 CM |
9163 | return 0; |
9164 | } | |
9165 | ||
d397712b CM |
9166 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
9167 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
9168 | { |
9169 | struct btrfs_trans_handle *trans; | |
9170 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 9171 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
2b0143b5 DH |
9172 | struct inode *new_inode = d_inode(new_dentry); |
9173 | struct inode *old_inode = d_inode(old_dentry); | |
39279cc3 | 9174 | struct timespec ctime = CURRENT_TIME; |
00e4e6b3 | 9175 | u64 index = 0; |
4df27c4d | 9176 | u64 root_objectid; |
39279cc3 | 9177 | int ret; |
33345d01 | 9178 | u64 old_ino = btrfs_ino(old_inode); |
39279cc3 | 9179 | |
33345d01 | 9180 | if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
9181 | return -EPERM; |
9182 | ||
4df27c4d | 9183 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 9184 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
9185 | return -EXDEV; |
9186 | ||
33345d01 LZ |
9187 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
9188 | (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 9189 | return -ENOTEMPTY; |
5f39d397 | 9190 | |
4df27c4d YZ |
9191 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
9192 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
9193 | return -ENOTEMPTY; | |
9c52057c CM |
9194 | |
9195 | ||
9196 | /* check for collisions, even if the name isn't there */ | |
4871c158 | 9197 | ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, |
9c52057c CM |
9198 | new_dentry->d_name.name, |
9199 | new_dentry->d_name.len); | |
9200 | ||
9201 | if (ret) { | |
9202 | if (ret == -EEXIST) { | |
9203 | /* we shouldn't get | |
9204 | * eexist without a new_inode */ | |
fae7f21c | 9205 | if (WARN_ON(!new_inode)) { |
9c52057c CM |
9206 | return ret; |
9207 | } | |
9208 | } else { | |
9209 | /* maybe -EOVERFLOW */ | |
9210 | return ret; | |
9211 | } | |
9212 | } | |
9213 | ret = 0; | |
9214 | ||
5a3f23d5 | 9215 | /* |
8d875f95 CM |
9216 | * we're using rename to replace one file with another. Start IO on it |
9217 | * now so we don't add too much work to the end of the transaction | |
5a3f23d5 | 9218 | */ |
8d875f95 | 9219 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size) |
5a3f23d5 CM |
9220 | filemap_flush(old_inode->i_mapping); |
9221 | ||
76dda93c | 9222 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 9223 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 9224 | down_read(&root->fs_info->subvol_sem); |
a22285a6 YZ |
9225 | /* |
9226 | * We want to reserve the absolute worst case amount of items. So if | |
9227 | * both inodes are subvols and we need to unlink them then that would | |
9228 | * require 4 item modifications, but if they are both normal inodes it | |
9229 | * would require 5 item modifications, so we'll assume their normal | |
9230 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items | |
9231 | * should cover the worst case number of items we'll modify. | |
9232 | */ | |
6e137ed3 | 9233 | trans = btrfs_start_transaction(root, 11); |
b44c59a8 JL |
9234 | if (IS_ERR(trans)) { |
9235 | ret = PTR_ERR(trans); | |
9236 | goto out_notrans; | |
9237 | } | |
76dda93c | 9238 | |
4df27c4d YZ |
9239 | if (dest != root) |
9240 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 9241 | |
a5719521 YZ |
9242 | ret = btrfs_set_inode_index(new_dir, &index); |
9243 | if (ret) | |
9244 | goto out_fail; | |
5a3f23d5 | 9245 | |
67de1176 | 9246 | BTRFS_I(old_inode)->dir_index = 0ULL; |
33345d01 | 9247 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d | 9248 | /* force full log commit if subvolume involved. */ |
995946dd | 9249 | btrfs_set_log_full_commit(root->fs_info, trans); |
4df27c4d | 9250 | } else { |
a5719521 YZ |
9251 | ret = btrfs_insert_inode_ref(trans, dest, |
9252 | new_dentry->d_name.name, | |
9253 | new_dentry->d_name.len, | |
33345d01 LZ |
9254 | old_ino, |
9255 | btrfs_ino(new_dir), index); | |
a5719521 YZ |
9256 | if (ret) |
9257 | goto out_fail; | |
4df27c4d YZ |
9258 | /* |
9259 | * this is an ugly little race, but the rename is required | |
9260 | * to make sure that if we crash, the inode is either at the | |
9261 | * old name or the new one. pinning the log transaction lets | |
9262 | * us make sure we don't allow a log commit to come in after | |
9263 | * we unlink the name but before we add the new name back in. | |
9264 | */ | |
9265 | btrfs_pin_log_trans(root); | |
9266 | } | |
5a3f23d5 | 9267 | |
0c4d2d95 JB |
9268 | inode_inc_iversion(old_dir); |
9269 | inode_inc_iversion(new_dir); | |
9270 | inode_inc_iversion(old_inode); | |
39279cc3 CM |
9271 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
9272 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
9273 | old_inode->i_ctime = ctime; | |
5f39d397 | 9274 | |
12fcfd22 CM |
9275 | if (old_dentry->d_parent != new_dentry->d_parent) |
9276 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
9277 | ||
33345d01 | 9278 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
9279 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
9280 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
9281 | old_dentry->d_name.name, | |
9282 | old_dentry->d_name.len); | |
9283 | } else { | |
92986796 | 9284 | ret = __btrfs_unlink_inode(trans, root, old_dir, |
2b0143b5 | 9285 | d_inode(old_dentry), |
92986796 AV |
9286 | old_dentry->d_name.name, |
9287 | old_dentry->d_name.len); | |
9288 | if (!ret) | |
9289 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d | 9290 | } |
79787eaa JM |
9291 | if (ret) { |
9292 | btrfs_abort_transaction(trans, root, ret); | |
9293 | goto out_fail; | |
9294 | } | |
39279cc3 CM |
9295 | |
9296 | if (new_inode) { | |
0c4d2d95 | 9297 | inode_inc_iversion(new_inode); |
39279cc3 | 9298 | new_inode->i_ctime = CURRENT_TIME; |
33345d01 | 9299 | if (unlikely(btrfs_ino(new_inode) == |
4df27c4d YZ |
9300 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
9301 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
9302 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
9303 | root_objectid, | |
9304 | new_dentry->d_name.name, | |
9305 | new_dentry->d_name.len); | |
9306 | BUG_ON(new_inode->i_nlink == 0); | |
9307 | } else { | |
9308 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
2b0143b5 | 9309 | d_inode(new_dentry), |
4df27c4d YZ |
9310 | new_dentry->d_name.name, |
9311 | new_dentry->d_name.len); | |
9312 | } | |
4ef31a45 | 9313 | if (!ret && new_inode->i_nlink == 0) |
2b0143b5 | 9314 | ret = btrfs_orphan_add(trans, d_inode(new_dentry)); |
79787eaa JM |
9315 | if (ret) { |
9316 | btrfs_abort_transaction(trans, root, ret); | |
9317 | goto out_fail; | |
9318 | } | |
39279cc3 | 9319 | } |
aec7477b | 9320 | |
4df27c4d YZ |
9321 | ret = btrfs_add_link(trans, new_dir, old_inode, |
9322 | new_dentry->d_name.name, | |
a5719521 | 9323 | new_dentry->d_name.len, 0, index); |
79787eaa JM |
9324 | if (ret) { |
9325 | btrfs_abort_transaction(trans, root, ret); | |
9326 | goto out_fail; | |
9327 | } | |
39279cc3 | 9328 | |
67de1176 MX |
9329 | if (old_inode->i_nlink == 1) |
9330 | BTRFS_I(old_inode)->dir_index = index; | |
9331 | ||
33345d01 | 9332 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) { |
10d9f309 | 9333 | struct dentry *parent = new_dentry->d_parent; |
6a912213 | 9334 | btrfs_log_new_name(trans, old_inode, old_dir, parent); |
4df27c4d YZ |
9335 | btrfs_end_log_trans(root); |
9336 | } | |
39279cc3 | 9337 | out_fail: |
7ad85bb7 | 9338 | btrfs_end_transaction(trans, root); |
b44c59a8 | 9339 | out_notrans: |
33345d01 | 9340 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 9341 | up_read(&root->fs_info->subvol_sem); |
9ed74f2d | 9342 | |
39279cc3 CM |
9343 | return ret; |
9344 | } | |
9345 | ||
80ace85c MS |
9346 | static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry, |
9347 | struct inode *new_dir, struct dentry *new_dentry, | |
9348 | unsigned int flags) | |
9349 | { | |
9350 | if (flags & ~RENAME_NOREPLACE) | |
9351 | return -EINVAL; | |
9352 | ||
9353 | return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry); | |
9354 | } | |
9355 | ||
8ccf6f19 MX |
9356 | static void btrfs_run_delalloc_work(struct btrfs_work *work) |
9357 | { | |
9358 | struct btrfs_delalloc_work *delalloc_work; | |
9f23e289 | 9359 | struct inode *inode; |
8ccf6f19 MX |
9360 | |
9361 | delalloc_work = container_of(work, struct btrfs_delalloc_work, | |
9362 | work); | |
9f23e289 JB |
9363 | inode = delalloc_work->inode; |
9364 | if (delalloc_work->wait) { | |
9365 | btrfs_wait_ordered_range(inode, 0, (u64)-1); | |
9366 | } else { | |
9367 | filemap_flush(inode->i_mapping); | |
9368 | if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, | |
9369 | &BTRFS_I(inode)->runtime_flags)) | |
9370 | filemap_flush(inode->i_mapping); | |
9371 | } | |
8ccf6f19 MX |
9372 | |
9373 | if (delalloc_work->delay_iput) | |
9f23e289 | 9374 | btrfs_add_delayed_iput(inode); |
8ccf6f19 | 9375 | else |
9f23e289 | 9376 | iput(inode); |
8ccf6f19 MX |
9377 | complete(&delalloc_work->completion); |
9378 | } | |
9379 | ||
9380 | struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode, | |
9381 | int wait, int delay_iput) | |
9382 | { | |
9383 | struct btrfs_delalloc_work *work; | |
9384 | ||
9385 | work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS); | |
9386 | if (!work) | |
9387 | return NULL; | |
9388 | ||
9389 | init_completion(&work->completion); | |
9390 | INIT_LIST_HEAD(&work->list); | |
9391 | work->inode = inode; | |
9392 | work->wait = wait; | |
9393 | work->delay_iput = delay_iput; | |
9e0af237 LB |
9394 | WARN_ON_ONCE(!inode); |
9395 | btrfs_init_work(&work->work, btrfs_flush_delalloc_helper, | |
9396 | btrfs_run_delalloc_work, NULL, NULL); | |
8ccf6f19 MX |
9397 | |
9398 | return work; | |
9399 | } | |
9400 | ||
9401 | void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work) | |
9402 | { | |
9403 | wait_for_completion(&work->completion); | |
9404 | kmem_cache_free(btrfs_delalloc_work_cachep, work); | |
9405 | } | |
9406 | ||
d352ac68 CM |
9407 | /* |
9408 | * some fairly slow code that needs optimization. This walks the list | |
9409 | * of all the inodes with pending delalloc and forces them to disk. | |
9410 | */ | |
6c255e67 MX |
9411 | static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput, |
9412 | int nr) | |
ea8c2819 | 9413 | { |
ea8c2819 | 9414 | struct btrfs_inode *binode; |
5b21f2ed | 9415 | struct inode *inode; |
8ccf6f19 MX |
9416 | struct btrfs_delalloc_work *work, *next; |
9417 | struct list_head works; | |
1eafa6c7 | 9418 | struct list_head splice; |
8ccf6f19 | 9419 | int ret = 0; |
ea8c2819 | 9420 | |
8ccf6f19 | 9421 | INIT_LIST_HEAD(&works); |
1eafa6c7 | 9422 | INIT_LIST_HEAD(&splice); |
63607cc8 | 9423 | |
573bfb72 | 9424 | mutex_lock(&root->delalloc_mutex); |
eb73c1b7 MX |
9425 | spin_lock(&root->delalloc_lock); |
9426 | list_splice_init(&root->delalloc_inodes, &splice); | |
1eafa6c7 MX |
9427 | while (!list_empty(&splice)) { |
9428 | binode = list_entry(splice.next, struct btrfs_inode, | |
ea8c2819 | 9429 | delalloc_inodes); |
1eafa6c7 | 9430 | |
eb73c1b7 MX |
9431 | list_move_tail(&binode->delalloc_inodes, |
9432 | &root->delalloc_inodes); | |
5b21f2ed | 9433 | inode = igrab(&binode->vfs_inode); |
df0af1a5 | 9434 | if (!inode) { |
eb73c1b7 | 9435 | cond_resched_lock(&root->delalloc_lock); |
1eafa6c7 | 9436 | continue; |
df0af1a5 | 9437 | } |
eb73c1b7 | 9438 | spin_unlock(&root->delalloc_lock); |
1eafa6c7 MX |
9439 | |
9440 | work = btrfs_alloc_delalloc_work(inode, 0, delay_iput); | |
5d99a998 | 9441 | if (!work) { |
f4ab9ea7 JB |
9442 | if (delay_iput) |
9443 | btrfs_add_delayed_iput(inode); | |
9444 | else | |
9445 | iput(inode); | |
1eafa6c7 | 9446 | ret = -ENOMEM; |
a1ecaabb | 9447 | goto out; |
5b21f2ed | 9448 | } |
1eafa6c7 | 9449 | list_add_tail(&work->list, &works); |
a44903ab QW |
9450 | btrfs_queue_work(root->fs_info->flush_workers, |
9451 | &work->work); | |
6c255e67 MX |
9452 | ret++; |
9453 | if (nr != -1 && ret >= nr) | |
a1ecaabb | 9454 | goto out; |
5b21f2ed | 9455 | cond_resched(); |
eb73c1b7 | 9456 | spin_lock(&root->delalloc_lock); |
ea8c2819 | 9457 | } |
eb73c1b7 | 9458 | spin_unlock(&root->delalloc_lock); |
8c8bee1d | 9459 | |
a1ecaabb | 9460 | out: |
eb73c1b7 MX |
9461 | list_for_each_entry_safe(work, next, &works, list) { |
9462 | list_del_init(&work->list); | |
9463 | btrfs_wait_and_free_delalloc_work(work); | |
9464 | } | |
9465 | ||
9466 | if (!list_empty_careful(&splice)) { | |
9467 | spin_lock(&root->delalloc_lock); | |
9468 | list_splice_tail(&splice, &root->delalloc_inodes); | |
9469 | spin_unlock(&root->delalloc_lock); | |
9470 | } | |
573bfb72 | 9471 | mutex_unlock(&root->delalloc_mutex); |
eb73c1b7 MX |
9472 | return ret; |
9473 | } | |
1eafa6c7 | 9474 | |
eb73c1b7 MX |
9475 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
9476 | { | |
9477 | int ret; | |
1eafa6c7 | 9478 | |
2c21b4d7 | 9479 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
eb73c1b7 MX |
9480 | return -EROFS; |
9481 | ||
6c255e67 MX |
9482 | ret = __start_delalloc_inodes(root, delay_iput, -1); |
9483 | if (ret > 0) | |
9484 | ret = 0; | |
eb73c1b7 MX |
9485 | /* |
9486 | * the filemap_flush will queue IO into the worker threads, but | |
8c8bee1d CM |
9487 | * we have to make sure the IO is actually started and that |
9488 | * ordered extents get created before we return | |
9489 | */ | |
9490 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 9491 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 9492 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 9493 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
9494 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
9495 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
9496 | } |
9497 | atomic_dec(&root->fs_info->async_submit_draining); | |
eb73c1b7 MX |
9498 | return ret; |
9499 | } | |
9500 | ||
6c255e67 MX |
9501 | int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput, |
9502 | int nr) | |
eb73c1b7 MX |
9503 | { |
9504 | struct btrfs_root *root; | |
9505 | struct list_head splice; | |
9506 | int ret; | |
9507 | ||
2c21b4d7 | 9508 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) |
eb73c1b7 MX |
9509 | return -EROFS; |
9510 | ||
9511 | INIT_LIST_HEAD(&splice); | |
9512 | ||
573bfb72 | 9513 | mutex_lock(&fs_info->delalloc_root_mutex); |
eb73c1b7 MX |
9514 | spin_lock(&fs_info->delalloc_root_lock); |
9515 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
6c255e67 | 9516 | while (!list_empty(&splice) && nr) { |
eb73c1b7 MX |
9517 | root = list_first_entry(&splice, struct btrfs_root, |
9518 | delalloc_root); | |
9519 | root = btrfs_grab_fs_root(root); | |
9520 | BUG_ON(!root); | |
9521 | list_move_tail(&root->delalloc_root, | |
9522 | &fs_info->delalloc_roots); | |
9523 | spin_unlock(&fs_info->delalloc_root_lock); | |
9524 | ||
6c255e67 | 9525 | ret = __start_delalloc_inodes(root, delay_iput, nr); |
eb73c1b7 | 9526 | btrfs_put_fs_root(root); |
6c255e67 | 9527 | if (ret < 0) |
eb73c1b7 MX |
9528 | goto out; |
9529 | ||
6c255e67 MX |
9530 | if (nr != -1) { |
9531 | nr -= ret; | |
9532 | WARN_ON(nr < 0); | |
9533 | } | |
eb73c1b7 | 9534 | spin_lock(&fs_info->delalloc_root_lock); |
8ccf6f19 | 9535 | } |
eb73c1b7 | 9536 | spin_unlock(&fs_info->delalloc_root_lock); |
1eafa6c7 | 9537 | |
6c255e67 | 9538 | ret = 0; |
eb73c1b7 MX |
9539 | atomic_inc(&fs_info->async_submit_draining); |
9540 | while (atomic_read(&fs_info->nr_async_submits) || | |
9541 | atomic_read(&fs_info->async_delalloc_pages)) { | |
9542 | wait_event(fs_info->async_submit_wait, | |
9543 | (atomic_read(&fs_info->nr_async_submits) == 0 && | |
9544 | atomic_read(&fs_info->async_delalloc_pages) == 0)); | |
9545 | } | |
9546 | atomic_dec(&fs_info->async_submit_draining); | |
eb73c1b7 | 9547 | out: |
1eafa6c7 | 9548 | if (!list_empty_careful(&splice)) { |
eb73c1b7 MX |
9549 | spin_lock(&fs_info->delalloc_root_lock); |
9550 | list_splice_tail(&splice, &fs_info->delalloc_roots); | |
9551 | spin_unlock(&fs_info->delalloc_root_lock); | |
1eafa6c7 | 9552 | } |
573bfb72 | 9553 | mutex_unlock(&fs_info->delalloc_root_mutex); |
8ccf6f19 | 9554 | return ret; |
ea8c2819 CM |
9555 | } |
9556 | ||
39279cc3 CM |
9557 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
9558 | const char *symname) | |
9559 | { | |
9560 | struct btrfs_trans_handle *trans; | |
9561 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
9562 | struct btrfs_path *path; | |
9563 | struct btrfs_key key; | |
1832a6d5 | 9564 | struct inode *inode = NULL; |
39279cc3 CM |
9565 | int err; |
9566 | int drop_inode = 0; | |
9567 | u64 objectid; | |
67871254 | 9568 | u64 index = 0; |
39279cc3 CM |
9569 | int name_len; |
9570 | int datasize; | |
5f39d397 | 9571 | unsigned long ptr; |
39279cc3 | 9572 | struct btrfs_file_extent_item *ei; |
5f39d397 | 9573 | struct extent_buffer *leaf; |
39279cc3 | 9574 | |
f06becc4 | 9575 | name_len = strlen(symname); |
39279cc3 CM |
9576 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) |
9577 | return -ENAMETOOLONG; | |
1832a6d5 | 9578 | |
9ed74f2d JB |
9579 | /* |
9580 | * 2 items for inode item and ref | |
9581 | * 2 items for dir items | |
9582 | * 1 item for xattr if selinux is on | |
9583 | */ | |
a22285a6 YZ |
9584 | trans = btrfs_start_transaction(root, 5); |
9585 | if (IS_ERR(trans)) | |
9586 | return PTR_ERR(trans); | |
1832a6d5 | 9587 | |
581bb050 LZ |
9588 | err = btrfs_find_free_ino(root, &objectid); |
9589 | if (err) | |
9590 | goto out_unlock; | |
9591 | ||
aec7477b | 9592 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 9593 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 9594 | S_IFLNK|S_IRWXUGO, &index); |
7cf96da3 TI |
9595 | if (IS_ERR(inode)) { |
9596 | err = PTR_ERR(inode); | |
39279cc3 | 9597 | goto out_unlock; |
7cf96da3 | 9598 | } |
39279cc3 | 9599 | |
ad19db71 CS |
9600 | /* |
9601 | * If the active LSM wants to access the inode during | |
9602 | * d_instantiate it needs these. Smack checks to see | |
9603 | * if the filesystem supports xattrs by looking at the | |
9604 | * ops vector. | |
9605 | */ | |
9606 | inode->i_fop = &btrfs_file_operations; | |
9607 | inode->i_op = &btrfs_file_inode_operations; | |
b0d5d10f | 9608 | inode->i_mapping->a_ops = &btrfs_aops; |
b0d5d10f CM |
9609 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
9610 | ||
9611 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); | |
9612 | if (err) | |
9613 | goto out_unlock_inode; | |
ad19db71 | 9614 | |
a1b075d2 | 9615 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 | 9616 | if (err) |
b0d5d10f | 9617 | goto out_unlock_inode; |
39279cc3 CM |
9618 | |
9619 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
9620 | if (!path) { |
9621 | err = -ENOMEM; | |
b0d5d10f | 9622 | goto out_unlock_inode; |
d8926bb3 | 9623 | } |
33345d01 | 9624 | key.objectid = btrfs_ino(inode); |
39279cc3 | 9625 | key.offset = 0; |
962a298f | 9626 | key.type = BTRFS_EXTENT_DATA_KEY; |
39279cc3 CM |
9627 | datasize = btrfs_file_extent_calc_inline_size(name_len); |
9628 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
9629 | datasize); | |
54aa1f4d | 9630 | if (err) { |
b0839166 | 9631 | btrfs_free_path(path); |
b0d5d10f | 9632 | goto out_unlock_inode; |
54aa1f4d | 9633 | } |
5f39d397 CM |
9634 | leaf = path->nodes[0]; |
9635 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
9636 | struct btrfs_file_extent_item); | |
9637 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
9638 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 9639 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
9640 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
9641 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
9642 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
9643 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
9644 | ||
39279cc3 | 9645 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
9646 | write_extent_buffer(leaf, symname, ptr, name_len); |
9647 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 9648 | btrfs_free_path(path); |
5f39d397 | 9649 | |
39279cc3 CM |
9650 | inode->i_op = &btrfs_symlink_inode_operations; |
9651 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
d899e052 | 9652 | inode_set_bytes(inode, name_len); |
f06becc4 | 9653 | btrfs_i_size_write(inode, name_len); |
54aa1f4d | 9654 | err = btrfs_update_inode(trans, root, inode); |
b0d5d10f | 9655 | if (err) { |
54aa1f4d | 9656 | drop_inode = 1; |
b0d5d10f CM |
9657 | goto out_unlock_inode; |
9658 | } | |
9659 | ||
9660 | unlock_new_inode(inode); | |
9661 | d_instantiate(dentry, inode); | |
39279cc3 CM |
9662 | |
9663 | out_unlock: | |
7ad85bb7 | 9664 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
9665 | if (drop_inode) { |
9666 | inode_dec_link_count(inode); | |
9667 | iput(inode); | |
9668 | } | |
b53d3f5d | 9669 | btrfs_btree_balance_dirty(root); |
39279cc3 | 9670 | return err; |
b0d5d10f CM |
9671 | |
9672 | out_unlock_inode: | |
9673 | drop_inode = 1; | |
9674 | unlock_new_inode(inode); | |
9675 | goto out_unlock; | |
39279cc3 | 9676 | } |
16432985 | 9677 | |
0af3d00b JB |
9678 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
9679 | u64 start, u64 num_bytes, u64 min_size, | |
9680 | loff_t actual_len, u64 *alloc_hint, | |
9681 | struct btrfs_trans_handle *trans) | |
d899e052 | 9682 | { |
5dc562c5 JB |
9683 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
9684 | struct extent_map *em; | |
d899e052 YZ |
9685 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9686 | struct btrfs_key ins; | |
d899e052 | 9687 | u64 cur_offset = start; |
55a61d1d | 9688 | u64 i_size; |
154ea289 | 9689 | u64 cur_bytes; |
d899e052 | 9690 | int ret = 0; |
0af3d00b | 9691 | bool own_trans = true; |
d899e052 | 9692 | |
0af3d00b JB |
9693 | if (trans) |
9694 | own_trans = false; | |
d899e052 | 9695 | while (num_bytes > 0) { |
0af3d00b JB |
9696 | if (own_trans) { |
9697 | trans = btrfs_start_transaction(root, 3); | |
9698 | if (IS_ERR(trans)) { | |
9699 | ret = PTR_ERR(trans); | |
9700 | break; | |
9701 | } | |
5a303d5d YZ |
9702 | } |
9703 | ||
154ea289 CM |
9704 | cur_bytes = min(num_bytes, 256ULL * 1024 * 1024); |
9705 | cur_bytes = max(cur_bytes, min_size); | |
00361589 | 9706 | ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0, |
e570fd27 | 9707 | *alloc_hint, &ins, 1, 0); |
5a303d5d | 9708 | if (ret) { |
0af3d00b JB |
9709 | if (own_trans) |
9710 | btrfs_end_transaction(trans, root); | |
a22285a6 | 9711 | break; |
d899e052 | 9712 | } |
5a303d5d | 9713 | |
d899e052 YZ |
9714 | ret = insert_reserved_file_extent(trans, inode, |
9715 | cur_offset, ins.objectid, | |
9716 | ins.offset, ins.offset, | |
920bbbfb | 9717 | ins.offset, 0, 0, 0, |
d899e052 | 9718 | BTRFS_FILE_EXTENT_PREALLOC); |
79787eaa | 9719 | if (ret) { |
857cc2fc | 9720 | btrfs_free_reserved_extent(root, ins.objectid, |
e570fd27 | 9721 | ins.offset, 0); |
79787eaa JM |
9722 | btrfs_abort_transaction(trans, root, ret); |
9723 | if (own_trans) | |
9724 | btrfs_end_transaction(trans, root); | |
9725 | break; | |
9726 | } | |
31193213 | 9727 | |
a1ed835e CM |
9728 | btrfs_drop_extent_cache(inode, cur_offset, |
9729 | cur_offset + ins.offset -1, 0); | |
5a303d5d | 9730 | |
5dc562c5 JB |
9731 | em = alloc_extent_map(); |
9732 | if (!em) { | |
9733 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, | |
9734 | &BTRFS_I(inode)->runtime_flags); | |
9735 | goto next; | |
9736 | } | |
9737 | ||
9738 | em->start = cur_offset; | |
9739 | em->orig_start = cur_offset; | |
9740 | em->len = ins.offset; | |
9741 | em->block_start = ins.objectid; | |
9742 | em->block_len = ins.offset; | |
b4939680 | 9743 | em->orig_block_len = ins.offset; |
cc95bef6 | 9744 | em->ram_bytes = ins.offset; |
5dc562c5 JB |
9745 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
9746 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
9747 | em->generation = trans->transid; | |
9748 | ||
9749 | while (1) { | |
9750 | write_lock(&em_tree->lock); | |
09a2a8f9 | 9751 | ret = add_extent_mapping(em_tree, em, 1); |
5dc562c5 JB |
9752 | write_unlock(&em_tree->lock); |
9753 | if (ret != -EEXIST) | |
9754 | break; | |
9755 | btrfs_drop_extent_cache(inode, cur_offset, | |
9756 | cur_offset + ins.offset - 1, | |
9757 | 0); | |
9758 | } | |
9759 | free_extent_map(em); | |
9760 | next: | |
d899e052 YZ |
9761 | num_bytes -= ins.offset; |
9762 | cur_offset += ins.offset; | |
efa56464 | 9763 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 9764 | |
0c4d2d95 | 9765 | inode_inc_iversion(inode); |
d899e052 | 9766 | inode->i_ctime = CURRENT_TIME; |
6cbff00f | 9767 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 9768 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
9769 | (actual_len > inode->i_size) && |
9770 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 9771 | if (cur_offset > actual_len) |
55a61d1d | 9772 | i_size = actual_len; |
d1ea6a61 | 9773 | else |
55a61d1d JB |
9774 | i_size = cur_offset; |
9775 | i_size_write(inode, i_size); | |
9776 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
9777 | } |
9778 | ||
d899e052 | 9779 | ret = btrfs_update_inode(trans, root, inode); |
79787eaa JM |
9780 | |
9781 | if (ret) { | |
9782 | btrfs_abort_transaction(trans, root, ret); | |
9783 | if (own_trans) | |
9784 | btrfs_end_transaction(trans, root); | |
9785 | break; | |
9786 | } | |
d899e052 | 9787 | |
0af3d00b JB |
9788 | if (own_trans) |
9789 | btrfs_end_transaction(trans, root); | |
5a303d5d | 9790 | } |
d899e052 YZ |
9791 | return ret; |
9792 | } | |
9793 | ||
0af3d00b JB |
9794 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
9795 | u64 start, u64 num_bytes, u64 min_size, | |
9796 | loff_t actual_len, u64 *alloc_hint) | |
9797 | { | |
9798 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
9799 | min_size, actual_len, alloc_hint, | |
9800 | NULL); | |
9801 | } | |
9802 | ||
9803 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
9804 | struct btrfs_trans_handle *trans, int mode, | |
9805 | u64 start, u64 num_bytes, u64 min_size, | |
9806 | loff_t actual_len, u64 *alloc_hint) | |
9807 | { | |
9808 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
9809 | min_size, actual_len, alloc_hint, trans); | |
9810 | } | |
9811 | ||
e6dcd2dc CM |
9812 | static int btrfs_set_page_dirty(struct page *page) |
9813 | { | |
e6dcd2dc CM |
9814 | return __set_page_dirty_nobuffers(page); |
9815 | } | |
9816 | ||
10556cb2 | 9817 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 9818 | { |
b83cc969 | 9819 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 9820 | umode_t mode = inode->i_mode; |
b83cc969 | 9821 | |
cb6db4e5 JM |
9822 | if (mask & MAY_WRITE && |
9823 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
9824 | if (btrfs_root_readonly(root)) | |
9825 | return -EROFS; | |
9826 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
9827 | return -EACCES; | |
9828 | } | |
2830ba7f | 9829 | return generic_permission(inode, mask); |
fdebe2bd | 9830 | } |
39279cc3 | 9831 | |
ef3b9af5 FM |
9832 | static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode) |
9833 | { | |
9834 | struct btrfs_trans_handle *trans; | |
9835 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
9836 | struct inode *inode = NULL; | |
9837 | u64 objectid; | |
9838 | u64 index; | |
9839 | int ret = 0; | |
9840 | ||
9841 | /* | |
9842 | * 5 units required for adding orphan entry | |
9843 | */ | |
9844 | trans = btrfs_start_transaction(root, 5); | |
9845 | if (IS_ERR(trans)) | |
9846 | return PTR_ERR(trans); | |
9847 | ||
9848 | ret = btrfs_find_free_ino(root, &objectid); | |
9849 | if (ret) | |
9850 | goto out; | |
9851 | ||
9852 | inode = btrfs_new_inode(trans, root, dir, NULL, 0, | |
9853 | btrfs_ino(dir), objectid, mode, &index); | |
9854 | if (IS_ERR(inode)) { | |
9855 | ret = PTR_ERR(inode); | |
9856 | inode = NULL; | |
9857 | goto out; | |
9858 | } | |
9859 | ||
ef3b9af5 FM |
9860 | inode->i_fop = &btrfs_file_operations; |
9861 | inode->i_op = &btrfs_file_inode_operations; | |
9862 | ||
9863 | inode->i_mapping->a_ops = &btrfs_aops; | |
ef3b9af5 FM |
9864 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
9865 | ||
b0d5d10f CM |
9866 | ret = btrfs_init_inode_security(trans, inode, dir, NULL); |
9867 | if (ret) | |
9868 | goto out_inode; | |
9869 | ||
9870 | ret = btrfs_update_inode(trans, root, inode); | |
9871 | if (ret) | |
9872 | goto out_inode; | |
ef3b9af5 FM |
9873 | ret = btrfs_orphan_add(trans, inode); |
9874 | if (ret) | |
b0d5d10f | 9875 | goto out_inode; |
ef3b9af5 | 9876 | |
5762b5c9 FM |
9877 | /* |
9878 | * We set number of links to 0 in btrfs_new_inode(), and here we set | |
9879 | * it to 1 because d_tmpfile() will issue a warning if the count is 0, | |
9880 | * through: | |
9881 | * | |
9882 | * d_tmpfile() -> inode_dec_link_count() -> drop_nlink() | |
9883 | */ | |
9884 | set_nlink(inode, 1); | |
b0d5d10f | 9885 | unlock_new_inode(inode); |
ef3b9af5 FM |
9886 | d_tmpfile(dentry, inode); |
9887 | mark_inode_dirty(inode); | |
9888 | ||
9889 | out: | |
9890 | btrfs_end_transaction(trans, root); | |
9891 | if (ret) | |
9892 | iput(inode); | |
9893 | btrfs_balance_delayed_items(root); | |
9894 | btrfs_btree_balance_dirty(root); | |
ef3b9af5 | 9895 | return ret; |
b0d5d10f CM |
9896 | |
9897 | out_inode: | |
9898 | unlock_new_inode(inode); | |
9899 | goto out; | |
9900 | ||
ef3b9af5 FM |
9901 | } |
9902 | ||
b38ef71c FM |
9903 | /* Inspired by filemap_check_errors() */ |
9904 | int btrfs_inode_check_errors(struct inode *inode) | |
9905 | { | |
9906 | int ret = 0; | |
9907 | ||
9908 | if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) && | |
9909 | test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags)) | |
9910 | ret = -ENOSPC; | |
9911 | if (test_bit(AS_EIO, &inode->i_mapping->flags) && | |
9912 | test_and_clear_bit(AS_EIO, &inode->i_mapping->flags)) | |
9913 | ret = -EIO; | |
9914 | ||
9915 | return ret; | |
9916 | } | |
9917 | ||
6e1d5dcc | 9918 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 9919 | .getattr = btrfs_getattr, |
39279cc3 CM |
9920 | .lookup = btrfs_lookup, |
9921 | .create = btrfs_create, | |
9922 | .unlink = btrfs_unlink, | |
9923 | .link = btrfs_link, | |
9924 | .mkdir = btrfs_mkdir, | |
9925 | .rmdir = btrfs_rmdir, | |
80ace85c | 9926 | .rename2 = btrfs_rename2, |
39279cc3 CM |
9927 | .symlink = btrfs_symlink, |
9928 | .setattr = btrfs_setattr, | |
618e21d5 | 9929 | .mknod = btrfs_mknod, |
95819c05 CH |
9930 | .setxattr = btrfs_setxattr, |
9931 | .getxattr = btrfs_getxattr, | |
5103e947 | 9932 | .listxattr = btrfs_listxattr, |
95819c05 | 9933 | .removexattr = btrfs_removexattr, |
fdebe2bd | 9934 | .permission = btrfs_permission, |
4e34e719 | 9935 | .get_acl = btrfs_get_acl, |
996a710d | 9936 | .set_acl = btrfs_set_acl, |
93fd63c2 | 9937 | .update_time = btrfs_update_time, |
ef3b9af5 | 9938 | .tmpfile = btrfs_tmpfile, |
39279cc3 | 9939 | }; |
6e1d5dcc | 9940 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 9941 | .lookup = btrfs_lookup, |
fdebe2bd | 9942 | .permission = btrfs_permission, |
4e34e719 | 9943 | .get_acl = btrfs_get_acl, |
996a710d | 9944 | .set_acl = btrfs_set_acl, |
93fd63c2 | 9945 | .update_time = btrfs_update_time, |
39279cc3 | 9946 | }; |
76dda93c | 9947 | |
828c0950 | 9948 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
9949 | .llseek = generic_file_llseek, |
9950 | .read = generic_read_dir, | |
9cdda8d3 | 9951 | .iterate = btrfs_real_readdir, |
34287aa3 | 9952 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 9953 | #ifdef CONFIG_COMPAT |
34287aa3 | 9954 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 9955 | #endif |
6bf13c0c | 9956 | .release = btrfs_release_file, |
e02119d5 | 9957 | .fsync = btrfs_sync_file, |
39279cc3 CM |
9958 | }; |
9959 | ||
d1310b2e | 9960 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 9961 | .fill_delalloc = run_delalloc_range, |
065631f6 | 9962 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 9963 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 9964 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 9965 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 9966 | .writepage_start_hook = btrfs_writepage_start_hook, |
b0c68f8b CM |
9967 | .set_bit_hook = btrfs_set_bit_hook, |
9968 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
9969 | .merge_extent_hook = btrfs_merge_extent_hook, |
9970 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
9971 | }; |
9972 | ||
35054394 CM |
9973 | /* |
9974 | * btrfs doesn't support the bmap operation because swapfiles | |
9975 | * use bmap to make a mapping of extents in the file. They assume | |
9976 | * these extents won't change over the life of the file and they | |
9977 | * use the bmap result to do IO directly to the drive. | |
9978 | * | |
9979 | * the btrfs bmap call would return logical addresses that aren't | |
9980 | * suitable for IO and they also will change frequently as COW | |
9981 | * operations happen. So, swapfile + btrfs == corruption. | |
9982 | * | |
9983 | * For now we're avoiding this by dropping bmap. | |
9984 | */ | |
7f09410b | 9985 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
9986 | .readpage = btrfs_readpage, |
9987 | .writepage = btrfs_writepage, | |
b293f02e | 9988 | .writepages = btrfs_writepages, |
3ab2fb5a | 9989 | .readpages = btrfs_readpages, |
16432985 | 9990 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
9991 | .invalidatepage = btrfs_invalidatepage, |
9992 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 9993 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 9994 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
9995 | }; |
9996 | ||
7f09410b | 9997 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
9998 | .readpage = btrfs_readpage, |
9999 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
10000 | .invalidatepage = btrfs_invalidatepage, |
10001 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
10002 | }; |
10003 | ||
6e1d5dcc | 10004 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
10005 | .getattr = btrfs_getattr, |
10006 | .setattr = btrfs_setattr, | |
95819c05 CH |
10007 | .setxattr = btrfs_setxattr, |
10008 | .getxattr = btrfs_getxattr, | |
5103e947 | 10009 | .listxattr = btrfs_listxattr, |
95819c05 | 10010 | .removexattr = btrfs_removexattr, |
fdebe2bd | 10011 | .permission = btrfs_permission, |
1506fcc8 | 10012 | .fiemap = btrfs_fiemap, |
4e34e719 | 10013 | .get_acl = btrfs_get_acl, |
996a710d | 10014 | .set_acl = btrfs_set_acl, |
e41f941a | 10015 | .update_time = btrfs_update_time, |
39279cc3 | 10016 | }; |
6e1d5dcc | 10017 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
10018 | .getattr = btrfs_getattr, |
10019 | .setattr = btrfs_setattr, | |
fdebe2bd | 10020 | .permission = btrfs_permission, |
95819c05 CH |
10021 | .setxattr = btrfs_setxattr, |
10022 | .getxattr = btrfs_getxattr, | |
33268eaf | 10023 | .listxattr = btrfs_listxattr, |
95819c05 | 10024 | .removexattr = btrfs_removexattr, |
4e34e719 | 10025 | .get_acl = btrfs_get_acl, |
996a710d | 10026 | .set_acl = btrfs_set_acl, |
e41f941a | 10027 | .update_time = btrfs_update_time, |
618e21d5 | 10028 | }; |
6e1d5dcc | 10029 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 CM |
10030 | .readlink = generic_readlink, |
10031 | .follow_link = page_follow_link_light, | |
10032 | .put_link = page_put_link, | |
f209561a | 10033 | .getattr = btrfs_getattr, |
22c44fe6 | 10034 | .setattr = btrfs_setattr, |
fdebe2bd | 10035 | .permission = btrfs_permission, |
0279b4cd JO |
10036 | .setxattr = btrfs_setxattr, |
10037 | .getxattr = btrfs_getxattr, | |
10038 | .listxattr = btrfs_listxattr, | |
10039 | .removexattr = btrfs_removexattr, | |
e41f941a | 10040 | .update_time = btrfs_update_time, |
39279cc3 | 10041 | }; |
76dda93c | 10042 | |
82d339d9 | 10043 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c | 10044 | .d_delete = btrfs_dentry_delete, |
b4aff1f8 | 10045 | .d_release = btrfs_dentry_release, |
76dda93c | 10046 | }; |