Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0
2 : /*
3 : * Copyright (C) 2007 Oracle. All rights reserved.
4 : */
5 :
6 : #include <linux/fs.h>
7 : #include <linux/pagemap.h>
8 : #include <linux/time.h>
9 : #include <linux/init.h>
10 : #include <linux/string.h>
11 : #include <linux/backing-dev.h>
12 : #include <linux/falloc.h>
13 : #include <linux/writeback.h>
14 : #include <linux/compat.h>
15 : #include <linux/slab.h>
16 : #include <linux/btrfs.h>
17 : #include <linux/uio.h>
18 : #include <linux/iversion.h>
19 : #include <linux/fsverity.h>
20 : #include "ctree.h"
21 : #include "disk-io.h"
22 : #include "transaction.h"
23 : #include "btrfs_inode.h"
24 : #include "print-tree.h"
25 : #include "tree-log.h"
26 : #include "locking.h"
27 : #include "volumes.h"
28 : #include "qgroup.h"
29 : #include "compression.h"
30 : #include "delalloc-space.h"
31 : #include "reflink.h"
32 : #include "subpage.h"
33 : #include "fs.h"
34 : #include "accessors.h"
35 : #include "extent-tree.h"
36 : #include "file-item.h"
37 : #include "ioctl.h"
38 : #include "file.h"
39 : #include "super.h"
40 :
41 : /* simple helper to fault in pages and copy. This should go away
42 : * and be replaced with calls into generic code.
43 : */
44 27590953 : static noinline int btrfs_copy_from_user(loff_t pos, size_t write_bytes,
45 : struct page **prepared_pages,
46 : struct iov_iter *i)
47 : {
48 27590953 : size_t copied = 0;
49 27590953 : size_t total_copied = 0;
50 27590953 : int pg = 0;
51 27590953 : int offset = offset_in_page(pos);
52 :
53 96764648 : while (write_bytes > 0) {
54 69122737 : size_t count = min_t(size_t,
55 : PAGE_SIZE - offset, write_bytes);
56 69122737 : struct page *page = prepared_pages[pg];
57 : /*
58 : * Copy data from userspace to the current page
59 : */
60 69122737 : copied = copy_page_from_iter_atomic(page, offset, count, i);
61 :
62 : /* Flush processor's dcache for this page */
63 69173695 : flush_dcache_page(page);
64 :
65 : /*
66 : * if we get a partial write, we can end up with
67 : * partially up to date pages. These add
68 : * a lot of complexity, so make sure they don't
69 : * happen by forcing this copy to be retried.
70 : *
71 : * The rest of the btrfs_file_write code will fall
72 : * back to page at a time copies after we return 0.
73 : */
74 69173695 : if (unlikely(copied < count)) {
75 0 : if (!PageUptodate(page)) {
76 0 : iov_iter_revert(i, copied);
77 0 : copied = 0;
78 : }
79 0 : if (!copied)
80 : break;
81 : }
82 :
83 69173695 : write_bytes -= copied;
84 69173695 : total_copied += copied;
85 69173695 : offset += copied;
86 69173695 : if (offset == PAGE_SIZE) {
87 58170087 : pg++;
88 58170087 : offset = 0;
89 : }
90 : }
91 27641911 : return total_copied;
92 : }
93 :
94 : /*
95 : * unlocks pages after btrfs_file_write is done with them
96 : */
97 27658521 : static void btrfs_drop_pages(struct btrfs_fs_info *fs_info,
98 : struct page **pages, size_t num_pages,
99 : u64 pos, u64 copied)
100 : {
101 27658521 : size_t i;
102 27658521 : u64 block_start = round_down(pos, fs_info->sectorsize);
103 27658521 : u64 block_len = round_up(pos + copied, fs_info->sectorsize) - block_start;
104 :
105 27658521 : ASSERT(block_len <= U32_MAX);
106 96839433 : for (i = 0; i < num_pages; i++) {
107 : /* page checked is some magic around finding pages that
108 : * have been modified without going through btrfs_set_page_dirty
109 : * clear it here. There should be no need to mark the pages
110 : * accessed as prepare_pages should have marked them accessed
111 : * in prepare_pages via find_or_create_page()
112 : */
113 69179671 : btrfs_page_clamp_clear_checked(fs_info, pages[i], block_start,
114 : block_len);
115 69188337 : unlock_page(pages[i]);
116 69190227 : put_page(pages[i]);
117 : }
118 27659762 : }
119 :
120 : /*
121 : * After btrfs_copy_from_user(), update the following things for delalloc:
122 : * - Mark newly dirtied pages as DELALLOC in the io tree.
123 : * Used to advise which range is to be written back.
124 : * - Mark modified pages as Uptodate/Dirty and not needing COW fixup
125 : * - Update inode size for past EOF write
126 : */
127 27631232 : int btrfs_dirty_pages(struct btrfs_inode *inode, struct page **pages,
128 : size_t num_pages, loff_t pos, size_t write_bytes,
129 : struct extent_state **cached, bool noreserve)
130 : {
131 27631232 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
132 27631232 : int err = 0;
133 27631232 : int i;
134 27631232 : u64 num_bytes;
135 27631232 : u64 start_pos;
136 27631232 : u64 end_of_last_block;
137 27631232 : u64 end_pos = pos + write_bytes;
138 27631232 : loff_t isize = i_size_read(&inode->vfs_inode);
139 27631232 : unsigned int extra_bits = 0;
140 :
141 27631232 : if (write_bytes == 0)
142 : return 0;
143 :
144 27631232 : if (noreserve)
145 5294 : extra_bits |= EXTENT_NORESERVE;
146 :
147 27631232 : start_pos = round_down(pos, fs_info->sectorsize);
148 27631232 : num_bytes = round_up(write_bytes + pos - start_pos,
149 : fs_info->sectorsize);
150 27631232 : ASSERT(num_bytes <= U32_MAX);
151 :
152 27631232 : end_of_last_block = start_pos + num_bytes - 1;
153 :
154 : /*
155 : * The pages may have already been dirty, clear out old accounting so
156 : * we can set things up properly
157 : */
158 27631232 : clear_extent_bit(&inode->io_tree, start_pos, end_of_last_block,
159 : EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
160 : cached);
161 :
162 27645101 : err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
163 : extra_bits, cached);
164 27626387 : if (err)
165 : return err;
166 :
167 96789670 : for (i = 0; i < num_pages; i++) {
168 69145152 : struct page *p = pages[i];
169 :
170 69145152 : btrfs_page_clamp_set_uptodate(fs_info, p, start_pos, num_bytes);
171 69172490 : btrfs_page_clamp_clear_checked(fs_info, p, start_pos, num_bytes);
172 69170636 : btrfs_page_clamp_set_dirty(fs_info, p, start_pos, num_bytes);
173 : }
174 :
175 : /*
176 : * we've only changed i_size in ram, and we haven't updated
177 : * the disk i_size. There is no need to log the inode
178 : * at this time.
179 : */
180 27644518 : if (end_pos > isize)
181 20123834 : i_size_write(&inode->vfs_inode, end_pos);
182 : return 0;
183 : }
184 :
185 : /*
186 : * this is very complex, but the basic idea is to drop all extents
187 : * in the range start - end. hint_block is filled in with a block number
188 : * that would be a good hint to the block allocator for this file.
189 : *
190 : * If an extent intersects the range but is not entirely inside the range
191 : * it is either truncated or split. Anything entirely inside the range
192 : * is deleted from the tree.
193 : *
194 : * Note: the VFS' inode number of bytes is not updated, it's up to the caller
195 : * to deal with that. We set the field 'bytes_found' of the arguments structure
196 : * with the number of allocated bytes found in the target range, so that the
197 : * caller can update the inode's number of bytes in an atomic way when
198 : * replacing extents in a range to avoid races with stat(2).
199 : */
200 23835518 : int btrfs_drop_extents(struct btrfs_trans_handle *trans,
201 : struct btrfs_root *root, struct btrfs_inode *inode,
202 : struct btrfs_drop_extents_args *args)
203 : {
204 23835518 : struct btrfs_fs_info *fs_info = root->fs_info;
205 23835518 : struct extent_buffer *leaf;
206 23835518 : struct btrfs_file_extent_item *fi;
207 23835518 : struct btrfs_ref ref = { 0 };
208 23835518 : struct btrfs_key key;
209 23835518 : struct btrfs_key new_key;
210 23835518 : u64 ino = btrfs_ino(inode);
211 23835518 : u64 search_start = args->start;
212 23835518 : u64 disk_bytenr = 0;
213 23835518 : u64 num_bytes = 0;
214 23835518 : u64 extent_offset = 0;
215 23835518 : u64 extent_end = 0;
216 23835518 : u64 last_end = args->start;
217 23835518 : int del_nr = 0;
218 23835518 : int del_slot = 0;
219 23835518 : int extent_type;
220 23835518 : int recow;
221 23835518 : int ret;
222 23835518 : int modify_tree = -1;
223 23835518 : int update_refs;
224 23835518 : int found = 0;
225 23835518 : struct btrfs_path *path = args->path;
226 :
227 23835518 : args->bytes_found = 0;
228 23835518 : args->extent_inserted = false;
229 :
230 : /* Must always have a path if ->replace_extent is true */
231 23835518 : ASSERT(!(args->replace_extent && !args->path));
232 :
233 23835518 : if (!path) {
234 98812 : path = btrfs_alloc_path();
235 98812 : if (!path) {
236 0 : ret = -ENOMEM;
237 0 : goto out;
238 : }
239 : }
240 :
241 23835518 : if (args->drop_cache)
242 20385806 : btrfs_drop_extent_map_range(inode, args->start, args->end - 1, false);
243 :
244 23835536 : if (args->start >= inode->disk_i_size && !args->replace_extent)
245 14865107 : modify_tree = 0;
246 :
247 23835536 : update_refs = (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID);
248 23878485 : while (1) {
249 23878485 : recow = 0;
250 23878485 : ret = btrfs_lookup_file_extent(trans, root, path, ino,
251 : search_start, modify_tree);
252 23879427 : if (ret < 0)
253 : break;
254 23879418 : if (ret > 0 && path->slots[0] > 0 && search_start == args->start) {
255 22146098 : leaf = path->nodes[0];
256 22146098 : btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
257 22146080 : if (key.objectid == ino &&
258 22146037 : key.type == BTRFS_EXTENT_DATA_KEY)
259 20363744 : path->slots[0]--;
260 : }
261 : ret = 0;
262 : next_slot:
263 42778620 : leaf = path->nodes[0];
264 42778620 : if (path->slots[0] >= btrfs_header_nritems(leaf)) {
265 15293848 : BUG_ON(del_nr > 0);
266 15293848 : ret = btrfs_next_leaf(root, path);
267 15293847 : if (ret < 0)
268 : break;
269 15293847 : if (ret > 0) {
270 : ret = 0;
271 : break;
272 : }
273 27659 : leaf = path->nodes[0];
274 27659 : recow = 1;
275 : }
276 :
277 27512431 : btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
278 :
279 27512515 : if (key.objectid > ino)
280 : break;
281 25557152 : if (WARN_ON_ONCE(key.objectid < ino) ||
282 25557152 : key.type < BTRFS_EXTENT_DATA_KEY) {
283 0 : ASSERT(del_nr == 0);
284 0 : path->slots[0]++;
285 0 : goto next_slot;
286 : }
287 25557152 : if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= args->end)
288 : break;
289 :
290 23579027 : fi = btrfs_item_ptr(leaf, path->slots[0],
291 : struct btrfs_file_extent_item);
292 23578958 : extent_type = btrfs_file_extent_type(leaf, fi);
293 :
294 23578903 : if (extent_type == BTRFS_FILE_EXTENT_REG ||
295 : extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
296 23336885 : disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
297 23336836 : num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
298 23336862 : extent_offset = btrfs_file_extent_offset(leaf, fi);
299 23336868 : extent_end = key.offset +
300 : btrfs_file_extent_num_bytes(leaf, fi);
301 242018 : } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
302 242018 : extent_end = key.offset +
303 : btrfs_file_extent_ram_bytes(leaf, fi);
304 : } else {
305 : /* can't happen */
306 0 : BUG();
307 : }
308 :
309 : /*
310 : * Don't skip extent items representing 0 byte lengths. They
311 : * used to be created (bug) if while punching holes we hit
312 : * -ENOSPC condition. So if we find one here, just ensure we
313 : * delete it, otherwise we would insert a new file extent item
314 : * with the same key (offset) as that 0 bytes length file
315 : * extent item in the call to setup_items_for_insert() later
316 : * in this function.
317 : */
318 23578942 : if (extent_end == key.offset && extent_end >= search_start) {
319 0 : last_end = extent_end;
320 0 : goto delete_extent_item;
321 : }
322 :
323 23578942 : if (extent_end <= search_start) {
324 17391955 : path->slots[0]++;
325 17391955 : goto next_slot;
326 : }
327 :
328 6186987 : found = 1;
329 6186987 : search_start = max(key.offset, args->start);
330 6186987 : if (recow || !modify_tree) {
331 9086 : modify_tree = -1;
332 9086 : btrfs_release_path(path);
333 9086 : continue;
334 : }
335 :
336 : /*
337 : * | - range to drop - |
338 : * | -------- extent -------- |
339 : */
340 6177901 : if (args->start > key.offset && args->end < extent_end) {
341 2391895 : BUG_ON(del_nr > 0);
342 2391895 : if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
343 : ret = -EOPNOTSUPP;
344 : break;
345 : }
346 :
347 2391895 : memcpy(&new_key, &key, sizeof(new_key));
348 2391895 : new_key.offset = args->start;
349 2391895 : ret = btrfs_duplicate_item(trans, root, path,
350 : &new_key);
351 2391891 : if (ret == -EAGAIN) {
352 7 : btrfs_release_path(path);
353 7 : continue;
354 : }
355 2391884 : if (ret < 0)
356 : break;
357 :
358 2391884 : leaf = path->nodes[0];
359 2391884 : fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
360 : struct btrfs_file_extent_item);
361 2391883 : btrfs_set_file_extent_num_bytes(leaf, fi,
362 2391883 : args->start - key.offset);
363 :
364 2391876 : fi = btrfs_item_ptr(leaf, path->slots[0],
365 : struct btrfs_file_extent_item);
366 :
367 2391876 : extent_offset += args->start - key.offset;
368 2391876 : btrfs_set_file_extent_offset(leaf, fi, extent_offset);
369 2391878 : btrfs_set_file_extent_num_bytes(leaf, fi,
370 2391878 : extent_end - args->start);
371 2391875 : btrfs_mark_buffer_dirty(leaf);
372 :
373 2391885 : if (update_refs && disk_bytenr > 0) {
374 2347243 : btrfs_init_generic_ref(&ref,
375 : BTRFS_ADD_DELAYED_REF,
376 : disk_bytenr, num_bytes, 0);
377 2347243 : btrfs_init_data_ref(&ref,
378 : root->root_key.objectid,
379 : new_key.objectid,
380 2347243 : args->start - extent_offset,
381 : 0, false);
382 2347243 : ret = btrfs_inc_extent_ref(trans, &ref);
383 2347246 : if (ret) {
384 0 : btrfs_abort_transaction(trans, ret);
385 0 : break;
386 : }
387 : }
388 2391888 : key.offset = args->start;
389 : }
390 : /*
391 : * From here on out we will have actually dropped something, so
392 : * last_end can be updated.
393 : */
394 6177894 : last_end = extent_end;
395 :
396 : /*
397 : * | ---- range to drop ----- |
398 : * | -------- extent -------- |
399 : */
400 6177894 : if (args->start <= key.offset && args->end < extent_end) {
401 3229182 : if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
402 : ret = -EOPNOTSUPP;
403 : break;
404 : }
405 :
406 3229184 : memcpy(&new_key, &key, sizeof(new_key));
407 3229184 : new_key.offset = args->end;
408 3229184 : btrfs_set_item_key_safe(fs_info, path, &new_key);
409 :
410 3229181 : extent_offset += args->end - key.offset;
411 3229181 : btrfs_set_file_extent_offset(leaf, fi, extent_offset);
412 3229181 : btrfs_set_file_extent_num_bytes(leaf, fi,
413 3229181 : extent_end - args->end);
414 3229180 : btrfs_mark_buffer_dirty(leaf);
415 3229176 : if (update_refs && disk_bytenr > 0)
416 3075953 : args->bytes_found += args->end - key.offset;
417 : break;
418 : }
419 :
420 2948712 : search_start = extent_end;
421 : /*
422 : * | ---- range to drop ----- |
423 : * | -------- extent -------- |
424 : */
425 2948712 : if (args->start > key.offset && args->end >= extent_end) {
426 578551 : BUG_ON(del_nr > 0);
427 578551 : if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
428 : ret = -EOPNOTSUPP;
429 : break;
430 : }
431 :
432 578551 : btrfs_set_file_extent_num_bytes(leaf, fi,
433 : args->start - key.offset);
434 578551 : btrfs_mark_buffer_dirty(leaf);
435 578562 : if (update_refs && disk_bytenr > 0)
436 502358 : args->bytes_found += extent_end - args->start;
437 578562 : if (args->end == extent_end)
438 : break;
439 :
440 323463 : path->slots[0]++;
441 323463 : goto next_slot;
442 : }
443 :
444 : /*
445 : * | ---- range to drop ----- |
446 : * | ------ extent ------ |
447 : */
448 2370161 : if (args->start <= key.offset && args->end >= extent_end) {
449 2370161 : delete_extent_item:
450 2370161 : if (del_nr == 0) {
451 1622157 : del_slot = path->slots[0];
452 1622157 : del_nr = 1;
453 : } else {
454 748004 : BUG_ON(del_slot + del_nr != path->slots[0]);
455 748004 : del_nr++;
456 : }
457 :
458 2370161 : if (update_refs &&
459 2370161 : extent_type == BTRFS_FILE_EXTENT_INLINE) {
460 241901 : args->bytes_found += extent_end - key.offset;
461 241901 : extent_end = ALIGN(extent_end,
462 : fs_info->sectorsize);
463 2128260 : } else if (update_refs && disk_bytenr > 0) {
464 1889305 : btrfs_init_generic_ref(&ref,
465 : BTRFS_DROP_DELAYED_REF,
466 : disk_bytenr, num_bytes, 0);
467 1889305 : btrfs_init_data_ref(&ref,
468 : root->root_key.objectid,
469 : key.objectid,
470 1889305 : key.offset - extent_offset, 0,
471 : false);
472 1889305 : ret = btrfs_free_extent(trans, &ref);
473 1889306 : if (ret) {
474 0 : btrfs_abort_transaction(trans, ret);
475 0 : break;
476 : }
477 1889306 : args->bytes_found += extent_end - key.offset;
478 : }
479 :
480 2370162 : if (args->end == extent_end)
481 : break;
482 :
483 1217658 : if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
484 1183802 : path->slots[0]++;
485 1183802 : goto next_slot;
486 : }
487 :
488 33856 : ret = btrfs_del_items(trans, root, path, del_slot,
489 : del_nr);
490 33856 : if (ret) {
491 0 : btrfs_abort_transaction(trans, ret);
492 0 : break;
493 : }
494 :
495 33856 : del_nr = 0;
496 33856 : del_slot = 0;
497 :
498 33856 : btrfs_release_path(path);
499 33856 : continue;
500 : }
501 :
502 0 : BUG();
503 : }
504 :
505 23836462 : if (!ret && del_nr > 0) {
506 : /*
507 : * Set path->slots[0] to first slot, so that after the delete
508 : * if items are move off from our leaf to its immediate left or
509 : * right neighbor leafs, we end up with a correct and adjusted
510 : * path->slots[0] for our insertion (if args->replace_extent).
511 : */
512 1588304 : path->slots[0] = del_slot;
513 1588304 : ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
514 1588311 : if (ret)
515 0 : btrfs_abort_transaction(trans, ret);
516 : }
517 :
518 23836469 : leaf = path->nodes[0];
519 : /*
520 : * If btrfs_del_items() was called, it might have deleted a leaf, in
521 : * which case it unlocked our path, so check path->locks[0] matches a
522 : * write lock.
523 : */
524 23836469 : if (!ret && args->replace_extent &&
525 4540621 : path->locks[0] == BTRFS_WRITE_LOCK &&
526 4074426 : btrfs_leaf_free_space(leaf) >=
527 4074443 : sizeof(struct btrfs_item) + args->extent_item_size) {
528 :
529 3807701 : key.objectid = ino;
530 3807701 : key.type = BTRFS_EXTENT_DATA_KEY;
531 3807701 : key.offset = args->start;
532 3807701 : if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
533 2624539 : struct btrfs_key slot_key;
534 :
535 2624539 : btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
536 2624608 : if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
537 227963 : path->slots[0]++;
538 : }
539 3807770 : btrfs_setup_item_for_insert(root, path, &key, args->extent_item_size);
540 3807672 : args->extent_inserted = true;
541 : }
542 :
543 23836457 : if (!args->path)
544 98812 : btrfs_free_path(path);
545 23737645 : else if (!args->extent_inserted)
546 19929878 : btrfs_release_path(path);
547 3807767 : out:
548 23836406 : args->drop_end = found ? min(args->end, last_end) : args->end;
549 :
550 23836406 : return ret;
551 : }
552 :
553 645975 : static int extent_mergeable(struct extent_buffer *leaf, int slot,
554 : u64 objectid, u64 bytenr, u64 orig_offset,
555 : u64 *start, u64 *end)
556 : {
557 645975 : struct btrfs_file_extent_item *fi;
558 645975 : struct btrfs_key key;
559 645975 : u64 extent_end;
560 :
561 645975 : if (slot < 0 || slot >= btrfs_header_nritems(leaf))
562 : return 0;
563 :
564 612066 : btrfs_item_key_to_cpu(leaf, &key, slot);
565 612065 : if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
566 : return 0;
567 :
568 592234 : fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
569 865072 : if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
570 62627 : btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
571 125245 : btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
572 62618 : btrfs_file_extent_compression(leaf, fi) ||
573 62618 : btrfs_file_extent_encryption(leaf, fi) ||
574 : btrfs_file_extent_other_encoding(leaf, fi))
575 529615 : return 0;
576 :
577 62618 : extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
578 62618 : if ((*start && *start != key.offset) || (*end && *end != extent_end))
579 : return 0;
580 :
581 54252 : *start = key.offset;
582 54252 : *end = extent_end;
583 54252 : return 1;
584 : }
585 :
586 : /*
587 : * Mark extent in the range start - end as written.
588 : *
589 : * This changes extent type from 'pre-allocated' to 'regular'. If only
590 : * part of extent is marked as written, the extent will be split into
591 : * two or three.
592 : */
593 285567 : int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
594 : struct btrfs_inode *inode, u64 start, u64 end)
595 : {
596 285567 : struct btrfs_fs_info *fs_info = trans->fs_info;
597 285567 : struct btrfs_root *root = inode->root;
598 285567 : struct extent_buffer *leaf;
599 285567 : struct btrfs_path *path;
600 285567 : struct btrfs_file_extent_item *fi;
601 285567 : struct btrfs_ref ref = { 0 };
602 285567 : struct btrfs_key key;
603 285567 : struct btrfs_key new_key;
604 285567 : u64 bytenr;
605 285567 : u64 num_bytes;
606 285567 : u64 extent_end;
607 285567 : u64 orig_offset;
608 285567 : u64 other_start;
609 285567 : u64 other_end;
610 285567 : u64 split;
611 285567 : int del_nr = 0;
612 285567 : int del_slot = 0;
613 285567 : int recow;
614 285567 : int ret = 0;
615 285567 : u64 ino = btrfs_ino(inode);
616 :
617 285567 : path = btrfs_alloc_path();
618 285563 : if (!path)
619 : return -ENOMEM;
620 285563 : again:
621 285593 : recow = 0;
622 285593 : split = start;
623 285593 : key.objectid = ino;
624 285593 : key.type = BTRFS_EXTENT_DATA_KEY;
625 285593 : key.offset = split;
626 :
627 285593 : ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
628 285604 : if (ret < 0)
629 0 : goto out;
630 285604 : if (ret > 0 && path->slots[0] > 0)
631 137975 : path->slots[0]--;
632 :
633 285604 : leaf = path->nodes[0];
634 285604 : btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
635 285604 : if (key.objectid != ino ||
636 285604 : key.type != BTRFS_EXTENT_DATA_KEY) {
637 0 : ret = -EINVAL;
638 0 : btrfs_abort_transaction(trans, ret);
639 0 : goto out;
640 : }
641 285604 : fi = btrfs_item_ptr(leaf, path->slots[0],
642 : struct btrfs_file_extent_item);
643 285604 : if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_PREALLOC) {
644 0 : ret = -EINVAL;
645 0 : btrfs_abort_transaction(trans, ret);
646 0 : goto out;
647 : }
648 285604 : extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
649 285604 : if (key.offset > start || extent_end < end) {
650 0 : ret = -EINVAL;
651 0 : btrfs_abort_transaction(trans, ret);
652 0 : goto out;
653 : }
654 :
655 285604 : bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
656 285604 : num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
657 285604 : orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
658 285604 : memcpy(&new_key, &key, sizeof(new_key));
659 :
660 285604 : if (start == key.offset && end < extent_end) {
661 66370 : other_start = 0;
662 66370 : other_end = start;
663 66370 : if (extent_mergeable(leaf, path->slots[0] - 1,
664 : ino, bytenr, orig_offset,
665 : &other_start, &other_end)) {
666 29615 : new_key.offset = end;
667 29615 : btrfs_set_item_key_safe(fs_info, path, &new_key);
668 29615 : fi = btrfs_item_ptr(leaf, path->slots[0],
669 : struct btrfs_file_extent_item);
670 29615 : btrfs_set_file_extent_generation(leaf, fi,
671 : trans->transid);
672 29615 : btrfs_set_file_extent_num_bytes(leaf, fi,
673 : extent_end - end);
674 29615 : btrfs_set_file_extent_offset(leaf, fi,
675 : end - orig_offset);
676 29615 : fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
677 : struct btrfs_file_extent_item);
678 29615 : btrfs_set_file_extent_generation(leaf, fi,
679 : trans->transid);
680 29615 : btrfs_set_file_extent_num_bytes(leaf, fi,
681 : end - other_start);
682 29615 : btrfs_mark_buffer_dirty(leaf);
683 29615 : goto out;
684 : }
685 : }
686 :
687 255989 : if (start > key.offset && end == extent_end) {
688 84320 : other_start = end;
689 84320 : other_end = 0;
690 84320 : if (extent_mergeable(leaf, path->slots[0] + 1,
691 : ino, bytenr, orig_offset,
692 : &other_start, &other_end)) {
693 8319 : fi = btrfs_item_ptr(leaf, path->slots[0],
694 : struct btrfs_file_extent_item);
695 8319 : btrfs_set_file_extent_num_bytes(leaf, fi,
696 8319 : start - key.offset);
697 8319 : btrfs_set_file_extent_generation(leaf, fi,
698 : trans->transid);
699 8319 : path->slots[0]++;
700 8319 : new_key.offset = start;
701 8319 : btrfs_set_item_key_safe(fs_info, path, &new_key);
702 :
703 8319 : fi = btrfs_item_ptr(leaf, path->slots[0],
704 : struct btrfs_file_extent_item);
705 8319 : btrfs_set_file_extent_generation(leaf, fi,
706 : trans->transid);
707 8319 : btrfs_set_file_extent_num_bytes(leaf, fi,
708 : other_end - start);
709 8319 : btrfs_set_file_extent_offset(leaf, fi,
710 : start - orig_offset);
711 8319 : btrfs_mark_buffer_dirty(leaf);
712 8319 : goto out;
713 : }
714 : }
715 :
716 467700 : while (start > key.offset || end < extent_end) {
717 220055 : if (key.offset == start)
718 90399 : split = end;
719 :
720 220055 : new_key.offset = split;
721 220055 : ret = btrfs_duplicate_item(trans, root, path, &new_key);
722 220053 : if (ret == -EAGAIN) {
723 25 : btrfs_release_path(path);
724 25 : goto again;
725 : }
726 220028 : if (ret < 0) {
727 0 : btrfs_abort_transaction(trans, ret);
728 0 : goto out;
729 : }
730 :
731 220028 : leaf = path->nodes[0];
732 220028 : fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
733 : struct btrfs_file_extent_item);
734 220028 : btrfs_set_file_extent_generation(leaf, fi, trans->transid);
735 220028 : btrfs_set_file_extent_num_bytes(leaf, fi,
736 220028 : split - key.offset);
737 :
738 220028 : fi = btrfs_item_ptr(leaf, path->slots[0],
739 : struct btrfs_file_extent_item);
740 :
741 220028 : btrfs_set_file_extent_generation(leaf, fi, trans->transid);
742 220029 : btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
743 220029 : btrfs_set_file_extent_num_bytes(leaf, fi,
744 : extent_end - split);
745 220028 : btrfs_mark_buffer_dirty(leaf);
746 :
747 220030 : btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, bytenr,
748 : num_bytes, 0);
749 220030 : btrfs_init_data_ref(&ref, root->root_key.objectid, ino,
750 : orig_offset, 0, false);
751 220030 : ret = btrfs_inc_extent_ref(trans, &ref);
752 220030 : if (ret) {
753 0 : btrfs_abort_transaction(trans, ret);
754 0 : goto out;
755 : }
756 :
757 220030 : if (split == start) {
758 129645 : key.offset = start;
759 : } else {
760 90385 : if (start != key.offset) {
761 0 : ret = -EINVAL;
762 0 : btrfs_abort_transaction(trans, ret);
763 0 : goto out;
764 : }
765 90385 : path->slots[0]--;
766 90385 : extent_end = end;
767 : }
768 : recow = 1;
769 : }
770 :
771 247645 : other_start = end;
772 247645 : other_end = 0;
773 247645 : btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
774 : num_bytes, 0);
775 247645 : btrfs_init_data_ref(&ref, root->root_key.objectid, ino, orig_offset,
776 : 0, false);
777 247645 : if (extent_mergeable(leaf, path->slots[0] + 1,
778 : ino, bytenr, orig_offset,
779 : &other_start, &other_end)) {
780 8424 : if (recow) {
781 2 : btrfs_release_path(path);
782 2 : goto again;
783 : }
784 8422 : extent_end = other_end;
785 8422 : del_slot = path->slots[0] + 1;
786 8422 : del_nr++;
787 8422 : ret = btrfs_free_extent(trans, &ref);
788 8422 : if (ret) {
789 0 : btrfs_abort_transaction(trans, ret);
790 0 : goto out;
791 : }
792 : }
793 247642 : other_start = 0;
794 247642 : other_end = start;
795 247642 : if (extent_mergeable(leaf, path->slots[0] - 1,
796 : ino, bytenr, orig_offset,
797 : &other_start, &other_end)) {
798 7894 : if (recow) {
799 3 : btrfs_release_path(path);
800 3 : goto again;
801 : }
802 7891 : key.offset = other_start;
803 7891 : del_slot = path->slots[0];
804 7891 : del_nr++;
805 7891 : ret = btrfs_free_extent(trans, &ref);
806 7891 : if (ret) {
807 0 : btrfs_abort_transaction(trans, ret);
808 0 : goto out;
809 : }
810 : }
811 247639 : if (del_nr == 0) {
812 234487 : fi = btrfs_item_ptr(leaf, path->slots[0],
813 : struct btrfs_file_extent_item);
814 234487 : btrfs_set_file_extent_type(leaf, fi,
815 : BTRFS_FILE_EXTENT_REG);
816 234487 : btrfs_set_file_extent_generation(leaf, fi, trans->transid);
817 234487 : btrfs_mark_buffer_dirty(leaf);
818 : } else {
819 13152 : fi = btrfs_item_ptr(leaf, del_slot - 1,
820 : struct btrfs_file_extent_item);
821 13152 : btrfs_set_file_extent_type(leaf, fi,
822 : BTRFS_FILE_EXTENT_REG);
823 13152 : btrfs_set_file_extent_generation(leaf, fi, trans->transid);
824 13152 : btrfs_set_file_extent_num_bytes(leaf, fi,
825 13152 : extent_end - key.offset);
826 13152 : btrfs_mark_buffer_dirty(leaf);
827 :
828 13152 : ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
829 13152 : if (ret < 0) {
830 0 : btrfs_abort_transaction(trans, ret);
831 0 : goto out;
832 : }
833 : }
834 13152 : out:
835 285574 : btrfs_free_path(path);
836 285574 : return ret;
837 : }
838 :
839 : /*
840 : * on error we return an unlocked page and the error value
841 : * on success we return a locked page and 0
842 : */
843 55188671 : static int prepare_uptodate_page(struct inode *inode,
844 : struct page *page, u64 pos,
845 : bool force_uptodate)
846 : {
847 55188671 : struct folio *folio = page_folio(page);
848 55197885 : int ret = 0;
849 :
850 74794234 : if (((pos & (PAGE_SIZE - 1)) || force_uptodate) &&
851 19603212 : !PageUptodate(page)) {
852 2988599 : ret = btrfs_read_folio(NULL, folio);
853 2986598 : if (ret)
854 : return ret;
855 2986287 : lock_page(page);
856 2986844 : if (!PageUptodate(page)) {
857 0 : unlock_page(page);
858 0 : return -EIO;
859 : }
860 :
861 : /*
862 : * Since btrfs_read_folio() will unlock the folio before it
863 : * returns, there is a window where btrfs_release_folio() can be
864 : * called to release the page. Here we check both inode
865 : * mapping and PagePrivate() to make sure the page was not
866 : * released.
867 : *
868 : * The private flag check is essential for subpage as we need
869 : * to store extra bitmap using page->private.
870 : */
871 2987316 : if (page->mapping != inode->i_mapping || !PagePrivate(page)) {
872 6 : unlock_page(page);
873 6 : return -EAGAIN;
874 : }
875 : }
876 : return 0;
877 : }
878 :
879 : static unsigned int get_prepare_fgp_flags(bool nowait)
880 : {
881 27643606 : unsigned int fgp_flags = FGP_LOCK | FGP_ACCESSED | FGP_CREAT;
882 :
883 27643606 : if (nowait)
884 0 : fgp_flags |= FGP_NOWAIT;
885 :
886 27643606 : return fgp_flags;
887 : }
888 :
889 : static gfp_t get_prepare_gfp_flags(struct inode *inode, bool nowait)
890 : {
891 27643606 : gfp_t gfp;
892 :
893 27643606 : gfp = btrfs_alloc_write_mask(inode->i_mapping);
894 27643606 : if (nowait) {
895 0 : gfp &= ~__GFP_DIRECT_RECLAIM;
896 0 : gfp |= GFP_NOWAIT;
897 : }
898 :
899 27643606 : return gfp;
900 : }
901 :
902 : /*
903 : * this just gets pages into the page cache and locks them down.
904 : */
905 27643606 : static noinline int prepare_pages(struct inode *inode, struct page **pages,
906 : size_t num_pages, loff_t pos,
907 : size_t write_bytes, bool force_uptodate,
908 : bool nowait)
909 : {
910 27643606 : int i;
911 27643606 : unsigned long index = pos >> PAGE_SHIFT;
912 27643606 : gfp_t mask = get_prepare_gfp_flags(inode, nowait);
913 27643606 : unsigned int fgp_flags = get_prepare_fgp_flags(nowait);
914 27643606 : int err = 0;
915 27643606 : int faili;
916 :
917 96813919 : for (i = 0; i < num_pages; i++) {
918 69200433 : again:
919 69200433 : pages[i] = pagecache_get_page(inode->i_mapping, index + i,
920 : fgp_flags, mask | __GFP_WRITE);
921 69178129 : if (!pages[i]) {
922 0 : faili = i - 1;
923 0 : if (nowait)
924 : err = -EAGAIN;
925 : else
926 0 : err = -ENOMEM;
927 0 : goto fail;
928 : }
929 :
930 69178129 : err = set_page_extent_mapped(pages[i]);
931 69176735 : if (err < 0) {
932 0 : faili = i;
933 0 : goto fail;
934 : }
935 :
936 69176735 : if (i == 0)
937 27614772 : err = prepare_uptodate_page(inode, pages[i], pos,
938 : force_uptodate);
939 69173545 : if (!err && i == num_pages - 1)
940 27615016 : err = prepare_uptodate_page(inode, pages[i],
941 : pos + write_bytes, false);
942 69182275 : if (err) {
943 6 : put_page(pages[i]);
944 6 : if (!nowait && err == -EAGAIN) {
945 6 : err = 0;
946 6 : goto again;
947 : }
948 0 : faili = i - 1;
949 0 : goto fail;
950 : }
951 69182269 : wait_on_page_writeback(pages[i]);
952 : }
953 :
954 : return 0;
955 0 : fail:
956 0 : while (faili >= 0) {
957 0 : unlock_page(pages[faili]);
958 0 : put_page(pages[faili]);
959 0 : faili--;
960 : }
961 : return err;
962 :
963 : }
964 :
965 : /*
966 : * This function locks the extent and properly waits for data=ordered extents
967 : * to finish before allowing the pages to be modified if need.
968 : *
969 : * The return value:
970 : * 1 - the extent is locked
971 : * 0 - the extent is not locked, and everything is OK
972 : * -EAGAIN - need re-prepare the pages
973 : * the other < 0 number - Something wrong happens
974 : */
975 : static noinline int
976 27606561 : lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct page **pages,
977 : size_t num_pages, loff_t pos,
978 : size_t write_bytes,
979 : u64 *lockstart, u64 *lockend, bool nowait,
980 : struct extent_state **cached_state)
981 : {
982 27606561 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
983 27606561 : u64 start_pos;
984 27606561 : u64 last_pos;
985 27606561 : int i;
986 27606561 : int ret = 0;
987 :
988 27606561 : start_pos = round_down(pos, fs_info->sectorsize);
989 27606561 : last_pos = round_up(pos + write_bytes, fs_info->sectorsize) - 1;
990 :
991 27606561 : if (start_pos < inode->vfs_inode.i_size) {
992 14749713 : struct btrfs_ordered_extent *ordered;
993 :
994 14749713 : if (nowait) {
995 0 : if (!try_lock_extent(&inode->io_tree, start_pos, last_pos,
996 : cached_state)) {
997 0 : for (i = 0; i < num_pages; i++) {
998 0 : unlock_page(pages[i]);
999 0 : put_page(pages[i]);
1000 0 : pages[i] = NULL;
1001 : }
1002 :
1003 : return -EAGAIN;
1004 : }
1005 : } else {
1006 14749713 : lock_extent(&inode->io_tree, start_pos, last_pos, cached_state);
1007 : }
1008 :
1009 14755851 : ordered = btrfs_lookup_ordered_range(inode, start_pos,
1010 14755851 : last_pos - start_pos + 1);
1011 14767716 : if (ordered &&
1012 2005 : ordered->file_offset + ordered->num_bytes > start_pos &&
1013 : ordered->file_offset <= last_pos) {
1014 2005 : unlock_extent(&inode->io_tree, start_pos, last_pos,
1015 : cached_state);
1016 36838 : for (i = 0; i < num_pages; i++) {
1017 32828 : unlock_page(pages[i]);
1018 32828 : put_page(pages[i]);
1019 : }
1020 2005 : btrfs_start_ordered_extent(ordered);
1021 2005 : btrfs_put_ordered_extent(ordered);
1022 2005 : return -EAGAIN;
1023 : }
1024 14765711 : if (ordered)
1025 0 : btrfs_put_ordered_extent(ordered);
1026 :
1027 14761768 : *lockstart = start_pos;
1028 14761768 : *lockend = last_pos;
1029 14761768 : ret = 1;
1030 : }
1031 :
1032 : /*
1033 : * We should be called after prepare_pages() which should have locked
1034 : * all pages in the range.
1035 : */
1036 96722388 : for (i = 0; i < num_pages; i++)
1037 69125153 : WARN_ON(!PageLocked(pages[i]));
1038 :
1039 : return ret;
1040 : }
1041 :
1042 : /*
1043 : * Check if we can do nocow write into the range [@pos, @pos + @write_bytes)
1044 : *
1045 : * @pos: File offset.
1046 : * @write_bytes: The length to write, will be updated to the nocow writeable
1047 : * range.
1048 : *
1049 : * This function will flush ordered extents in the range to ensure proper
1050 : * nocow checks.
1051 : *
1052 : * Return:
1053 : * > 0 If we can nocow, and updates @write_bytes.
1054 : * 0 If we can't do a nocow write.
1055 : * -EAGAIN If we can't do a nocow write because snapshoting of the inode's
1056 : * root is in progress.
1057 : * < 0 If an error happened.
1058 : *
1059 : * NOTE: Callers need to call btrfs_check_nocow_unlock() if we return > 0.
1060 : */
1061 68276 : int btrfs_check_nocow_lock(struct btrfs_inode *inode, loff_t pos,
1062 : size_t *write_bytes, bool nowait)
1063 : {
1064 68276 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
1065 68276 : struct btrfs_root *root = inode->root;
1066 68276 : struct extent_state *cached_state = NULL;
1067 68276 : u64 lockstart, lockend;
1068 68276 : u64 num_bytes;
1069 68276 : int ret;
1070 :
1071 68276 : if (!(inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
1072 : return 0;
1073 :
1074 7549 : if (!btrfs_drew_try_write_lock(&root->snapshot_lock))
1075 : return -EAGAIN;
1076 :
1077 7555 : lockstart = round_down(pos, fs_info->sectorsize);
1078 7555 : lockend = round_up(pos + *write_bytes,
1079 : fs_info->sectorsize) - 1;
1080 7555 : num_bytes = lockend - lockstart + 1;
1081 :
1082 7555 : if (nowait) {
1083 0 : if (!btrfs_try_lock_ordered_range(inode, lockstart, lockend,
1084 : &cached_state)) {
1085 0 : btrfs_drew_write_unlock(&root->snapshot_lock);
1086 0 : return -EAGAIN;
1087 : }
1088 : } else {
1089 7555 : btrfs_lock_and_flush_ordered_range(inode, lockstart, lockend,
1090 : &cached_state);
1091 : }
1092 7550 : ret = can_nocow_extent(&inode->vfs_inode, lockstart, &num_bytes,
1093 : NULL, NULL, NULL, nowait, false);
1094 7552 : if (ret <= 0)
1095 2192 : btrfs_drew_write_unlock(&root->snapshot_lock);
1096 : else
1097 5360 : *write_bytes = min_t(size_t, *write_bytes ,
1098 : num_bytes - pos + lockstart);
1099 7556 : unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
1100 :
1101 7556 : return ret;
1102 : }
1103 :
1104 49 : void btrfs_check_nocow_unlock(struct btrfs_inode *inode)
1105 : {
1106 49 : btrfs_drew_write_unlock(&inode->root->snapshot_lock);
1107 5311 : }
1108 :
1109 27439564 : static void update_time_for_write(struct inode *inode)
1110 : {
1111 27439564 : struct timespec64 now;
1112 :
1113 27439564 : if (IS_NOCMTIME(inode))
1114 : return;
1115 :
1116 27440163 : now = current_time(inode);
1117 27442608 : if (!timespec64_equal(&inode->i_mtime, &now))
1118 757641 : inode->i_mtime = now;
1119 :
1120 27442608 : if (!timespec64_equal(&inode->i_ctime, &now))
1121 767001 : inode->i_ctime = now;
1122 :
1123 27442608 : if (IS_I_VERSION(inode))
1124 27430944 : inode_inc_iversion(inode);
1125 : }
1126 :
1127 27431385 : static int btrfs_write_check(struct kiocb *iocb, struct iov_iter *from,
1128 : size_t count)
1129 : {
1130 27431385 : struct file *file = iocb->ki_filp;
1131 27431385 : struct inode *inode = file_inode(file);
1132 27431385 : struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1133 27431385 : loff_t pos = iocb->ki_pos;
1134 27431385 : int ret;
1135 27431385 : loff_t oldsize;
1136 27431385 : loff_t start_pos;
1137 :
1138 : /*
1139 : * Quickly bail out on NOWAIT writes if we don't have the nodatacow or
1140 : * prealloc flags, as without those flags we always have to COW. We will
1141 : * later check if we can really COW into the target range (using
1142 : * can_nocow_extent() at btrfs_get_blocks_direct_write()).
1143 : */
1144 27431385 : if ((iocb->ki_flags & IOCB_NOWAIT) &&
1145 7 : !(BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
1146 : return -EAGAIN;
1147 :
1148 27431383 : ret = file_remove_privs(file);
1149 27436721 : if (ret)
1150 : return ret;
1151 :
1152 : /*
1153 : * We reserve space for updating the inode when we reserve space for the
1154 : * extent we are going to write, so we will enospc out there. We don't
1155 : * need to start yet another transaction to update the inode as we will
1156 : * update the inode when we finish writing whatever data we write.
1157 : */
1158 27440911 : update_time_for_write(inode);
1159 :
1160 27473743 : start_pos = round_down(pos, fs_info->sectorsize);
1161 27473743 : oldsize = i_size_read(inode);
1162 27473743 : if (start_pos > oldsize) {
1163 : /* Expand hole size to cover write data, preventing empty gap */
1164 1385188 : loff_t end_pos = round_up(pos + count, fs_info->sectorsize);
1165 :
1166 1385188 : ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, end_pos);
1167 1387166 : if (ret)
1168 1412 : return ret;
1169 : }
1170 :
1171 : return 0;
1172 : }
1173 :
1174 26393393 : static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb,
1175 : struct iov_iter *i)
1176 : {
1177 26393393 : struct file *file = iocb->ki_filp;
1178 26393393 : loff_t pos;
1179 26393393 : struct inode *inode = file_inode(file);
1180 26393393 : struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1181 26393393 : struct page **pages = NULL;
1182 26393393 : struct extent_changeset *data_reserved = NULL;
1183 26393393 : u64 release_bytes = 0;
1184 26393393 : u64 lockstart;
1185 26393393 : u64 lockend;
1186 26393393 : size_t num_written = 0;
1187 26393393 : int nrptrs;
1188 26393393 : ssize_t ret;
1189 26393393 : bool only_release_metadata = false;
1190 26393393 : bool force_page_uptodate = false;
1191 26393393 : loff_t old_isize = i_size_read(inode);
1192 26393393 : unsigned int ilock_flags = 0;
1193 26393393 : const bool nowait = (iocb->ki_flags & IOCB_NOWAIT);
1194 26393393 : unsigned int bdp_flags = (nowait ? BDP_ASYNC : 0);
1195 :
1196 26393393 : if (nowait)
1197 0 : ilock_flags |= BTRFS_ILOCK_TRY;
1198 :
1199 26393393 : ret = btrfs_inode_lock(BTRFS_I(inode), ilock_flags);
1200 26381999 : if (ret < 0)
1201 : return ret;
1202 :
1203 26381999 : ret = generic_write_checks(iocb, i);
1204 26365470 : if (ret <= 0)
1205 21 : goto out;
1206 :
1207 26365449 : ret = btrfs_write_check(iocb, i, ret);
1208 26419562 : if (ret < 0)
1209 1055 : goto out;
1210 :
1211 26418507 : pos = iocb->ki_pos;
1212 26418507 : nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_SIZE),
1213 : PAGE_SIZE / (sizeof(struct page *)));
1214 26418507 : nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1215 26418507 : nrptrs = max(nrptrs, 8);
1216 26418507 : pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL);
1217 26347358 : if (!pages) {
1218 0 : ret = -ENOMEM;
1219 0 : goto out;
1220 : }
1221 :
1222 54006174 : while (iov_iter_count(i) > 0) {
1223 27642377 : struct extent_state *cached_state = NULL;
1224 27642377 : size_t offset = offset_in_page(pos);
1225 27642377 : size_t sector_offset;
1226 27642377 : size_t write_bytes = min(iov_iter_count(i),
1227 : nrptrs * (size_t)PAGE_SIZE -
1228 : offset);
1229 27642377 : size_t num_pages;
1230 27642377 : size_t reserve_bytes;
1231 27642377 : size_t dirty_pages;
1232 27642377 : size_t copied;
1233 27642377 : size_t dirty_sectors;
1234 27642377 : size_t num_sectors;
1235 27642377 : int extents_locked;
1236 :
1237 : /*
1238 : * Fault pages before locking them in prepare_pages
1239 : * to avoid recursive lock
1240 : */
1241 27642377 : if (unlikely(fault_in_iov_iter_readable(i, write_bytes))) {
1242 : ret = -EFAULT;
1243 61293 : break;
1244 : }
1245 :
1246 27659272 : only_release_metadata = false;
1247 27659272 : sector_offset = pos & (fs_info->sectorsize - 1);
1248 :
1249 27659272 : extent_changeset_release(data_reserved);
1250 27659272 : ret = btrfs_check_data_free_space(BTRFS_I(inode),
1251 : &data_reserved, pos,
1252 : write_bytes, nowait);
1253 27721929 : if (ret < 0) {
1254 66432 : int can_nocow;
1255 :
1256 66432 : if (nowait && (ret == -ENOSPC || ret == -EAGAIN)) {
1257 : ret = -EAGAIN;
1258 : break;
1259 : }
1260 :
1261 : /*
1262 : * If we don't have to COW at the offset, reserve
1263 : * metadata only. write_bytes may get smaller than
1264 : * requested here.
1265 : */
1266 66432 : can_nocow = btrfs_check_nocow_lock(BTRFS_I(inode), pos,
1267 : &write_bytes, nowait);
1268 66458 : if (can_nocow < 0)
1269 0 : ret = can_nocow;
1270 66458 : if (can_nocow > 0)
1271 : ret = 0;
1272 : if (ret)
1273 : break;
1274 : only_release_metadata = true;
1275 : }
1276 :
1277 27660818 : num_pages = DIV_ROUND_UP(write_bytes + offset, PAGE_SIZE);
1278 27660818 : WARN_ON(num_pages > nrptrs);
1279 27660818 : reserve_bytes = round_up(write_bytes + sector_offset,
1280 : fs_info->sectorsize);
1281 27660818 : WARN_ON(reserve_bytes == 0);
1282 27660818 : ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
1283 : reserve_bytes,
1284 : reserve_bytes, nowait);
1285 27652937 : if (ret) {
1286 156 : if (!only_release_metadata)
1287 139 : btrfs_free_reserved_data_space(BTRFS_I(inode),
1288 : data_reserved, pos,
1289 : write_bytes);
1290 : else
1291 17 : btrfs_check_nocow_unlock(BTRFS_I(inode));
1292 :
1293 156 : if (nowait && ret == -ENOSPC)
1294 0 : ret = -EAGAIN;
1295 : break;
1296 : }
1297 :
1298 : release_bytes = reserve_bytes;
1299 27654786 : again:
1300 27654786 : ret = balance_dirty_pages_ratelimited_flags(inode->i_mapping, bdp_flags);
1301 27631119 : if (ret) {
1302 0 : btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes);
1303 0 : break;
1304 : }
1305 :
1306 : /*
1307 : * This is going to setup the pages array with the number of
1308 : * pages we want, so we don't really need to worry about the
1309 : * contents of pages from loop to loop
1310 : */
1311 27631119 : ret = prepare_pages(inode, pages, num_pages,
1312 : pos, write_bytes, force_page_uptodate, false);
1313 27620639 : if (ret) {
1314 0 : btrfs_delalloc_release_extents(BTRFS_I(inode),
1315 : reserve_bytes);
1316 0 : break;
1317 : }
1318 :
1319 27620639 : extents_locked = lock_and_cleanup_extent_if_need(
1320 : BTRFS_I(inode), pages,
1321 : num_pages, pos, write_bytes, &lockstart,
1322 : &lockend, nowait, &cached_state);
1323 27585150 : if (extents_locked < 0) {
1324 2005 : if (!nowait && extents_locked == -EAGAIN)
1325 2005 : goto again;
1326 :
1327 0 : btrfs_delalloc_release_extents(BTRFS_I(inode),
1328 : reserve_bytes);
1329 0 : ret = extents_locked;
1330 0 : break;
1331 : }
1332 :
1333 27583145 : copied = btrfs_copy_from_user(pos, write_bytes, pages, i);
1334 :
1335 27606301 : num_sectors = BTRFS_BYTES_TO_BLKS(fs_info, reserve_bytes);
1336 27606301 : dirty_sectors = round_up(copied + sector_offset,
1337 : fs_info->sectorsize);
1338 27606301 : dirty_sectors = BTRFS_BYTES_TO_BLKS(fs_info, dirty_sectors);
1339 :
1340 : /*
1341 : * if we have trouble faulting in the pages, fall
1342 : * back to one page at a time
1343 : */
1344 27606301 : if (copied < write_bytes)
1345 0 : nrptrs = 1;
1346 :
1347 27606301 : if (copied == 0) {
1348 : force_page_uptodate = true;
1349 : dirty_sectors = 0;
1350 : dirty_pages = 0;
1351 : } else {
1352 27619553 : force_page_uptodate = false;
1353 27619553 : dirty_pages = DIV_ROUND_UP(copied + offset,
1354 : PAGE_SIZE);
1355 : }
1356 :
1357 27606301 : if (num_sectors > dirty_sectors) {
1358 : /* release everything except the sectors we dirtied */
1359 0 : release_bytes -= dirty_sectors << fs_info->sectorsize_bits;
1360 0 : if (only_release_metadata) {
1361 0 : btrfs_delalloc_release_metadata(BTRFS_I(inode),
1362 : release_bytes, true);
1363 : } else {
1364 0 : u64 __pos;
1365 :
1366 0 : __pos = round_down(pos,
1367 : fs_info->sectorsize) +
1368 0 : (dirty_pages << PAGE_SHIFT);
1369 0 : btrfs_delalloc_release_space(BTRFS_I(inode),
1370 : data_reserved, __pos,
1371 : release_bytes, true);
1372 : }
1373 : }
1374 :
1375 27606301 : release_bytes = round_up(copied + sector_offset,
1376 : fs_info->sectorsize);
1377 :
1378 27606301 : ret = btrfs_dirty_pages(BTRFS_I(inode), pages,
1379 : dirty_pages, pos, copied,
1380 : &cached_state, only_release_metadata);
1381 :
1382 : /*
1383 : * If we have not locked the extent range, because the range's
1384 : * start offset is >= i_size, we might still have a non-NULL
1385 : * cached extent state, acquired while marking the extent range
1386 : * as delalloc through btrfs_dirty_pages(). Therefore free any
1387 : * possible cached extent state to avoid a memory leak.
1388 : */
1389 27630729 : if (extents_locked)
1390 14753397 : unlock_extent(&BTRFS_I(inode)->io_tree, lockstart,
1391 : lockend, &cached_state);
1392 : else
1393 12877332 : free_extent_state(cached_state);
1394 :
1395 27637369 : btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes);
1396 27658116 : if (ret) {
1397 0 : btrfs_drop_pages(fs_info, pages, num_pages, pos, copied);
1398 0 : break;
1399 : }
1400 :
1401 27658116 : release_bytes = 0;
1402 27658116 : if (only_release_metadata)
1403 5294 : btrfs_check_nocow_unlock(BTRFS_I(inode));
1404 :
1405 27658116 : btrfs_drop_pages(fs_info, pages, num_pages, pos, copied);
1406 :
1407 27658022 : cond_resched();
1408 :
1409 27658816 : pos += copied;
1410 27658816 : num_written += copied;
1411 : }
1412 :
1413 26425090 : kfree(pages);
1414 :
1415 26429622 : if (release_bytes) {
1416 0 : if (only_release_metadata) {
1417 0 : btrfs_check_nocow_unlock(BTRFS_I(inode));
1418 0 : btrfs_delalloc_release_metadata(BTRFS_I(inode),
1419 : release_bytes, true);
1420 : } else {
1421 0 : btrfs_delalloc_release_space(BTRFS_I(inode),
1422 : data_reserved,
1423 0 : round_down(pos, fs_info->sectorsize),
1424 : release_bytes, true);
1425 : }
1426 : }
1427 :
1428 26429622 : extent_changeset_free(data_reserved);
1429 26407861 : if (num_written > 0) {
1430 26346846 : pagecache_isize_extended(inode, old_isize, iocb->ki_pos);
1431 26333444 : iocb->ki_pos += num_written;
1432 : }
1433 61015 : out:
1434 26395535 : btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1435 26408074 : return num_written ? num_written : ret;
1436 : }
1437 :
1438 4119603 : static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
1439 : const struct iov_iter *iter, loff_t offset)
1440 : {
1441 4119603 : const u32 blocksize_mask = fs_info->sectorsize - 1;
1442 :
1443 4119603 : if (offset & blocksize_mask)
1444 : return -EINVAL;
1445 :
1446 3822309 : if (iov_iter_alignment(iter) & blocksize_mask)
1447 2049930 : return -EINVAL;
1448 :
1449 : return 0;
1450 : }
1451 :
1452 1058505 : static ssize_t btrfs_direct_write(struct kiocb *iocb, struct iov_iter *from)
1453 : {
1454 1058505 : struct file *file = iocb->ki_filp;
1455 1058505 : struct inode *inode = file_inode(file);
1456 1058505 : struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1457 1058505 : loff_t pos;
1458 1058505 : ssize_t written = 0;
1459 1058505 : ssize_t written_buffered;
1460 1058505 : size_t prev_left = 0;
1461 1058505 : loff_t endbyte;
1462 1058505 : ssize_t err;
1463 1058505 : unsigned int ilock_flags = 0;
1464 1058505 : struct iomap_dio *dio;
1465 :
1466 1058505 : if (iocb->ki_flags & IOCB_NOWAIT)
1467 7 : ilock_flags |= BTRFS_ILOCK_TRY;
1468 :
1469 : /* If the write DIO is within EOF, use a shared lock */
1470 1058505 : if (iocb->ki_pos + iov_iter_count(from) <= i_size_read(inode))
1471 768755 : ilock_flags |= BTRFS_ILOCK_SHARED;
1472 :
1473 289750 : relock:
1474 1058528 : err = btrfs_inode_lock(BTRFS_I(inode), ilock_flags);
1475 1058787 : if (err < 0)
1476 0 : return err;
1477 :
1478 1058787 : err = generic_write_checks(iocb, from);
1479 1058870 : if (err <= 0) {
1480 1 : btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1481 1 : return err;
1482 : }
1483 :
1484 1058869 : err = btrfs_write_check(iocb, from, err);
1485 1058876 : if (err < 0) {
1486 359 : btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1487 359 : goto out;
1488 : }
1489 :
1490 1058517 : pos = iocb->ki_pos;
1491 : /*
1492 : * Re-check since file size may have changed just before taking the
1493 : * lock or pos may have changed because of O_APPEND in generic_write_check()
1494 : */
1495 1058517 : if ((ilock_flags & BTRFS_ILOCK_SHARED) &&
1496 769150 : pos + iov_iter_count(from) > i_size_read(inode)) {
1497 12 : btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1498 12 : ilock_flags &= ~BTRFS_ILOCK_SHARED;
1499 12 : goto relock;
1500 : }
1501 :
1502 1058505 : if (check_direct_IO(fs_info, from, pos)) {
1503 360650 : btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1504 360695 : goto buffered;
1505 : }
1506 :
1507 : /*
1508 : * The iov_iter can be mapped to the same file range we are writing to.
1509 : * If that's the case, then we will deadlock in the iomap code, because
1510 : * it first calls our callback btrfs_dio_iomap_begin(), which will create
1511 : * an ordered extent, and after that it will fault in the pages that the
1512 : * iov_iter refers to. During the fault in we end up in the readahead
1513 : * pages code (starting at btrfs_readahead()), which will lock the range,
1514 : * find that ordered extent and then wait for it to complete (at
1515 : * btrfs_lock_and_flush_ordered_range()), resulting in a deadlock since
1516 : * obviously the ordered extent can never complete as we didn't submit
1517 : * yet the respective bio(s). This always happens when the buffer is
1518 : * memory mapped to the same file range, since the iomap DIO code always
1519 : * invalidates pages in the target file range (after starting and waiting
1520 : * for any writeback).
1521 : *
1522 : * So here we disable page faults in the iov_iter and then retry if we
1523 : * got -EFAULT, faulting in the pages before the retry.
1524 : */
1525 697389 : from->nofault = true;
1526 697389 : dio = btrfs_dio_write(iocb, from, written);
1527 697841 : from->nofault = false;
1528 :
1529 : /*
1530 : * iomap_dio_complete() will call btrfs_sync_file() if we have a dsync
1531 : * iocb, and that needs to lock the inode. So unlock it before calling
1532 : * iomap_dio_complete() to avoid a deadlock.
1533 : */
1534 697841 : btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1535 :
1536 1395478 : if (IS_ERR_OR_NULL(dio))
1537 308378 : err = PTR_ERR_OR_ZERO(dio);
1538 : else
1539 543548 : err = iomap_dio_complete(dio);
1540 :
1541 : /* No increment (+=) because iomap returns a cumulative value. */
1542 697725 : if (err > 0)
1543 519801 : written = err;
1544 :
1545 697725 : if (iov_iter_count(from) > 0 && (err == -EFAULT || err > 0)) {
1546 113 : const size_t left = iov_iter_count(from);
1547 : /*
1548 : * We have more data left to write. Try to fault in as many as
1549 : * possible of the remainder pages and retry. We do this without
1550 : * releasing and locking again the inode, to prevent races with
1551 : * truncate.
1552 : *
1553 : * Also, in case the iov refers to pages in the file range of the
1554 : * file we want to write to (due to a mmap), we could enter an
1555 : * infinite loop if we retry after faulting the pages in, since
1556 : * iomap will invalidate any pages in the range early on, before
1557 : * it tries to fault in the pages of the iov. So we keep track of
1558 : * how much was left of iov in the previous EFAULT and fallback
1559 : * to buffered IO in case we haven't made any progress.
1560 : */
1561 113 : if (left == prev_left) {
1562 : err = -ENOTBLK;
1563 : } else {
1564 112 : fault_in_iov_iter_readable(from, left);
1565 11 : prev_left = left;
1566 11 : goto relock;
1567 : }
1568 : }
1569 :
1570 : /*
1571 : * If 'err' is -ENOTBLK or we have not written all data, then it means
1572 : * we must fallback to buffered IO.
1573 : */
1574 697613 : if ((err < 0 && err != -ENOTBLK) || !iov_iter_count(from))
1575 697133 : goto out;
1576 :
1577 480 : buffered:
1578 : /*
1579 : * If we are in a NOWAIT context, then return -EAGAIN to signal the caller
1580 : * it must retry the operation in a context where blocking is acceptable,
1581 : * because even if we end up not blocking during the buffered IO attempt
1582 : * below, we will block when flushing and waiting for the IO.
1583 : */
1584 361175 : if (iocb->ki_flags & IOCB_NOWAIT) {
1585 0 : err = -EAGAIN;
1586 0 : goto out;
1587 : }
1588 :
1589 361175 : pos = iocb->ki_pos;
1590 361175 : written_buffered = btrfs_buffered_write(iocb, from);
1591 361381 : if (written_buffered < 0) {
1592 0 : err = written_buffered;
1593 0 : goto out;
1594 : }
1595 : /*
1596 : * Ensure all data is persisted. We want the next direct IO read to be
1597 : * able to read what was just written.
1598 : */
1599 361381 : endbyte = pos + written_buffered - 1;
1600 361381 : err = btrfs_fdatawrite_range(inode, pos, endbyte);
1601 361429 : if (err)
1602 0 : goto out;
1603 361429 : err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte);
1604 361358 : if (err)
1605 10 : goto out;
1606 361348 : written += written_buffered;
1607 361348 : iocb->ki_pos = pos + written_buffered;
1608 361348 : invalidate_mapping_pages(file->f_mapping, pos >> PAGE_SHIFT,
1609 361348 : endbyte >> PAGE_SHIFT);
1610 1058800 : out:
1611 1058800 : return err < 0 ? err : written;
1612 : }
1613 :
1614 0 : static ssize_t btrfs_encoded_write(struct kiocb *iocb, struct iov_iter *from,
1615 : const struct btrfs_ioctl_encoded_io_args *encoded)
1616 : {
1617 0 : struct file *file = iocb->ki_filp;
1618 0 : struct inode *inode = file_inode(file);
1619 0 : loff_t count;
1620 0 : ssize_t ret;
1621 :
1622 0 : btrfs_inode_lock(BTRFS_I(inode), 0);
1623 0 : count = encoded->len;
1624 0 : ret = generic_write_checks_count(iocb, &count);
1625 0 : if (ret == 0 && count != encoded->len) {
1626 : /*
1627 : * The write got truncated by generic_write_checks_count(). We
1628 : * can't do a partial encoded write.
1629 : */
1630 : ret = -EFBIG;
1631 : }
1632 0 : if (ret || encoded->len == 0)
1633 0 : goto out;
1634 :
1635 0 : ret = btrfs_write_check(iocb, from, encoded->len);
1636 0 : if (ret < 0)
1637 0 : goto out;
1638 :
1639 0 : ret = btrfs_do_encoded_write(iocb, from, encoded);
1640 0 : out:
1641 0 : btrfs_inode_unlock(BTRFS_I(inode), 0);
1642 0 : return ret;
1643 : }
1644 :
1645 27112012 : ssize_t btrfs_do_write_iter(struct kiocb *iocb, struct iov_iter *from,
1646 : const struct btrfs_ioctl_encoded_io_args *encoded)
1647 : {
1648 27112012 : struct file *file = iocb->ki_filp;
1649 27112012 : struct btrfs_inode *inode = BTRFS_I(file_inode(file));
1650 27112012 : ssize_t num_written, num_sync;
1651 :
1652 : /*
1653 : * If the fs flips readonly due to some impossible error, although we
1654 : * have opened a file as writable, we have to stop this write operation
1655 : * to ensure consistency.
1656 : */
1657 27112012 : if (BTRFS_FS_ERROR(inode->root->fs_info))
1658 : return -EROFS;
1659 :
1660 27112012 : if (encoded && (iocb->ki_flags & IOCB_NOWAIT))
1661 : return -EOPNOTSUPP;
1662 :
1663 27112012 : if (encoded) {
1664 0 : num_written = btrfs_encoded_write(iocb, from, encoded);
1665 0 : num_sync = encoded->len;
1666 27112012 : } else if (iocb->ki_flags & IOCB_DIRECT) {
1667 1058584 : num_written = btrfs_direct_write(iocb, from);
1668 1058584 : num_sync = num_written;
1669 : } else {
1670 26053428 : num_written = btrfs_buffered_write(iocb, from);
1671 26053428 : num_sync = num_written;
1672 : }
1673 :
1674 27111856 : btrfs_set_inode_last_sub_trans(inode);
1675 :
1676 27120302 : if (num_sync > 0) {
1677 26877332 : num_sync = generic_write_sync(iocb, num_sync);
1678 26867689 : if (num_sync < 0)
1679 0 : num_written = num_sync;
1680 : }
1681 :
1682 : return num_written;
1683 : }
1684 :
1685 27107125 : static ssize_t btrfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1686 : {
1687 27107125 : return btrfs_do_write_iter(iocb, from, NULL);
1688 : }
1689 :
1690 7572374 : int btrfs_release_file(struct inode *inode, struct file *filp)
1691 : {
1692 7572374 : struct btrfs_file_private *private = filp->private_data;
1693 :
1694 7572374 : if (private) {
1695 156464 : kfree(private->filldir_buf);
1696 156467 : free_extent_state(private->llseek_cached_state);
1697 156451 : kfree(private);
1698 156462 : filp->private_data = NULL;
1699 : }
1700 :
1701 : /*
1702 : * Set by setattr when we are about to truncate a file from a non-zero
1703 : * size to a zero size. This tries to flush down new bytes that may
1704 : * have been written if the application were using truncate to replace
1705 : * a file in place.
1706 : */
1707 7579498 : if (test_and_clear_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
1708 7572372 : &BTRFS_I(inode)->runtime_flags))
1709 64730 : filemap_flush(inode->i_mapping);
1710 7579487 : return 0;
1711 : }
1712 :
1713 551464 : static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end)
1714 : {
1715 551464 : int ret;
1716 551464 : struct blk_plug plug;
1717 :
1718 : /*
1719 : * This is only called in fsync, which would do synchronous writes, so
1720 : * a plug can merge adjacent IOs as much as possible. Esp. in case of
1721 : * multiple disks using raid profile, a large IO can be split to
1722 : * several segments of stripe length (currently 64K).
1723 : */
1724 551464 : blk_start_plug(&plug);
1725 551463 : ret = btrfs_fdatawrite_range(inode, start, end);
1726 551531 : blk_finish_plug(&plug);
1727 :
1728 551513 : return ret;
1729 : }
1730 :
1731 275769 : static inline bool skip_inode_logging(const struct btrfs_log_ctx *ctx)
1732 : {
1733 275769 : struct btrfs_inode *inode = BTRFS_I(ctx->inode);
1734 275769 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
1735 :
1736 275769 : if (btrfs_inode_in_log(inode, fs_info->generation) &&
1737 2148 : list_empty(&ctx->ordered_extents))
1738 : return true;
1739 :
1740 : /*
1741 : * If we are doing a fast fsync we can not bail out if the inode's
1742 : * last_trans is <= then the last committed transaction, because we only
1743 : * update the last_trans of the inode during ordered extent completion,
1744 : * and for a fast fsync we don't wait for that, we only wait for the
1745 : * writeback to complete.
1746 : */
1747 273686 : if (inode->last_trans <= fs_info->last_trans_committed &&
1748 18063 : (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) ||
1749 3330 : list_empty(&ctx->ordered_extents)))
1750 17748 : return true;
1751 :
1752 : return false;
1753 : }
1754 :
1755 : /*
1756 : * fsync call for both files and directories. This logs the inode into
1757 : * the tree log instead of forcing full commits whenever possible.
1758 : *
1759 : * It needs to call filemap_fdatawait so that all ordered extent updates are
1760 : * in the metadata btree are up to date for copying to the log.
1761 : *
1762 : * It drops the inode mutex before doing the tree log commit. This is an
1763 : * important optimization for directories because holding the mutex prevents
1764 : * new operations on the dir while we write to disk.
1765 : */
1766 275724 : int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1767 : {
1768 275724 : struct dentry *dentry = file_dentry(file);
1769 275687 : struct inode *inode = d_inode(dentry);
1770 275687 : struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1771 275687 : struct btrfs_root *root = BTRFS_I(inode)->root;
1772 275687 : struct btrfs_trans_handle *trans;
1773 275687 : struct btrfs_log_ctx ctx;
1774 275687 : int ret = 0, err;
1775 275687 : u64 len;
1776 275687 : bool full_sync;
1777 :
1778 275687 : trace_btrfs_sync_file(file, datasync);
1779 :
1780 275687 : btrfs_init_log_ctx(&ctx, inode);
1781 :
1782 : /*
1783 : * Always set the range to a full range, otherwise we can get into
1784 : * several problems, from missing file extent items to represent holes
1785 : * when not using the NO_HOLES feature, to log tree corruption due to
1786 : * races between hole detection during logging and completion of ordered
1787 : * extents outside the range, to missing checksums due to ordered extents
1788 : * for which we flushed only a subset of their pages.
1789 : */
1790 275687 : start = 0;
1791 275687 : end = LLONG_MAX;
1792 275687 : len = (u64)LLONG_MAX + 1;
1793 :
1794 : /*
1795 : * We write the dirty pages in the range and wait until they complete
1796 : * out of the ->i_mutex. If so, we can flush the dirty pages by
1797 : * multi-task, and make the performance up. See
1798 : * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1799 : */
1800 275687 : ret = start_ordered_ops(inode, start, end);
1801 275744 : if (ret)
1802 0 : goto out;
1803 :
1804 275744 : btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
1805 :
1806 275758 : atomic_inc(&root->log_batch);
1807 :
1808 : /*
1809 : * Before we acquired the inode's lock and the mmap lock, someone may
1810 : * have dirtied more pages in the target range. We need to make sure
1811 : * that writeback for any such pages does not start while we are logging
1812 : * the inode, because if it does, any of the following might happen when
1813 : * we are not doing a full inode sync:
1814 : *
1815 : * 1) We log an extent after its writeback finishes but before its
1816 : * checksums are added to the csum tree, leading to -EIO errors
1817 : * when attempting to read the extent after a log replay.
1818 : *
1819 : * 2) We can end up logging an extent before its writeback finishes.
1820 : * Therefore after the log replay we will have a file extent item
1821 : * pointing to an unwritten extent (and no data checksums as well).
1822 : *
1823 : * So trigger writeback for any eventual new dirty pages and then we
1824 : * wait for all ordered extents to complete below.
1825 : */
1826 275774 : ret = start_ordered_ops(inode, start, end);
1827 275771 : if (ret) {
1828 0 : btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
1829 0 : goto out;
1830 : }
1831 :
1832 : /*
1833 : * Always check for the full sync flag while holding the inode's lock,
1834 : * to avoid races with other tasks. The flag must be either set all the
1835 : * time during logging or always off all the time while logging.
1836 : * We check the flag here after starting delalloc above, because when
1837 : * running delalloc the full sync flag may be set if we need to drop
1838 : * extra extent map ranges due to temporary memory allocation failures.
1839 : */
1840 275771 : full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1841 : &BTRFS_I(inode)->runtime_flags);
1842 :
1843 : /*
1844 : * We have to do this here to avoid the priority inversion of waiting on
1845 : * IO of a lower priority task while holding a transaction open.
1846 : *
1847 : * For a full fsync we wait for the ordered extents to complete while
1848 : * for a fast fsync we wait just for writeback to complete, and then
1849 : * attach the ordered extents to the transaction so that a transaction
1850 : * commit waits for their completion, to avoid data loss if we fsync,
1851 : * the current transaction commits before the ordered extents complete
1852 : * and a power failure happens right after that.
1853 : *
1854 : * For zoned filesystem, if a write IO uses a ZONE_APPEND command, the
1855 : * logical address recorded in the ordered extent may change. We need
1856 : * to wait for the IO to stabilize the logical address.
1857 : */
1858 275771 : if (full_sync || btrfs_is_zoned(fs_info)) {
1859 130342 : ret = btrfs_wait_ordered_range(inode, start, len);
1860 : } else {
1861 : /*
1862 : * Get our ordered extents as soon as possible to avoid doing
1863 : * checksum lookups in the csum tree, and use instead the
1864 : * checksums attached to the ordered extents.
1865 : */
1866 145429 : btrfs_get_ordered_extents_for_logging(BTRFS_I(inode),
1867 : &ctx.ordered_extents);
1868 145426 : ret = filemap_fdatawait_range(inode->i_mapping, start, end);
1869 : }
1870 :
1871 275775 : if (ret)
1872 10 : goto out_release_extents;
1873 :
1874 275765 : atomic_inc(&root->log_batch);
1875 :
1876 275768 : smp_mb();
1877 275769 : if (skip_inode_logging(&ctx)) {
1878 : /*
1879 : * We've had everything committed since the last time we were
1880 : * modified so clear this flag in case it was set for whatever
1881 : * reason, it's no longer relevant.
1882 : */
1883 19828 : clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1884 19828 : &BTRFS_I(inode)->runtime_flags);
1885 : /*
1886 : * An ordered extent might have started before and completed
1887 : * already with io errors, in which case the inode was not
1888 : * updated and we end up here. So check the inode's mapping
1889 : * for any errors that might have happened since we last
1890 : * checked called fsync.
1891 : */
1892 19828 : ret = filemap_check_wb_err(inode->i_mapping, file->f_wb_err);
1893 19828 : goto out_release_extents;
1894 : }
1895 :
1896 : /*
1897 : * We use start here because we will need to wait on the IO to complete
1898 : * in btrfs_sync_log, which could require joining a transaction (for
1899 : * example checking cross references in the nocow path). If we use join
1900 : * here we could get into a situation where we're waiting on IO to
1901 : * happen that is blocked on a transaction trying to commit. With start
1902 : * we inc the extwriter counter, so we wait for all extwriters to exit
1903 : * before we start blocking joiners. This comment is to keep somebody
1904 : * from thinking they are super smart and changing this to
1905 : * btrfs_join_transaction *cough*Josef*cough*.
1906 : */
1907 255937 : trans = btrfs_start_transaction(root, 0);
1908 255939 : if (IS_ERR(trans)) {
1909 0 : ret = PTR_ERR(trans);
1910 0 : goto out_release_extents;
1911 : }
1912 255939 : trans->in_fsync = true;
1913 :
1914 255939 : ret = btrfs_log_dentry_safe(trans, dentry, &ctx);
1915 255911 : btrfs_release_log_ctx_extents(&ctx);
1916 255893 : if (ret < 0) {
1917 : /* Fallthrough and commit/free transaction. */
1918 470 : ret = BTRFS_LOG_FORCE_COMMIT;
1919 : }
1920 :
1921 : /* we've logged all the items and now have a consistent
1922 : * version of the file in the log. It is possible that
1923 : * someone will come in and modify the file, but that's
1924 : * fine because the log is consistent on disk, and we
1925 : * have references to all of the file's extents
1926 : *
1927 : * It is possible that someone will come in and log the
1928 : * file again, but that will end up using the synchronization
1929 : * inside btrfs_sync_log to keep things safe.
1930 : */
1931 255893 : btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
1932 :
1933 255919 : if (ret == BTRFS_NO_LOG_SYNC) {
1934 1 : ret = btrfs_end_transaction(trans);
1935 1 : goto out;
1936 : }
1937 :
1938 : /* We successfully logged the inode, attempt to sync the log. */
1939 255918 : if (!ret) {
1940 255448 : ret = btrfs_sync_log(trans, root, &ctx);
1941 255467 : if (!ret) {
1942 255276 : ret = btrfs_end_transaction(trans);
1943 255270 : goto out;
1944 : }
1945 : }
1946 :
1947 : /*
1948 : * At this point we need to commit the transaction because we had
1949 : * btrfs_need_log_full_commit() or some other error.
1950 : *
1951 : * If we didn't do a full sync we have to stop the trans handle, wait on
1952 : * the ordered extents, start it again and commit the transaction. If
1953 : * we attempt to wait on the ordered extents here we could deadlock with
1954 : * something like fallocate() that is holding the extent lock trying to
1955 : * start a transaction while some other thread is trying to commit the
1956 : * transaction while we (fsync) are currently holding the transaction
1957 : * open.
1958 : */
1959 661 : if (!full_sync) {
1960 117 : ret = btrfs_end_transaction(trans);
1961 117 : if (ret)
1962 0 : goto out;
1963 117 : ret = btrfs_wait_ordered_range(inode, start, len);
1964 117 : if (ret)
1965 0 : goto out;
1966 :
1967 : /*
1968 : * This is safe to use here because we're only interested in
1969 : * making sure the transaction that had the ordered extents is
1970 : * committed. We aren't waiting on anything past this point,
1971 : * we're purely getting the transaction and committing it.
1972 : */
1973 117 : trans = btrfs_attach_transaction_barrier(root);
1974 117 : if (IS_ERR(trans)) {
1975 21 : ret = PTR_ERR(trans);
1976 :
1977 : /*
1978 : * We committed the transaction and there's no currently
1979 : * running transaction, this means everything we care
1980 : * about made it to disk and we are done.
1981 : */
1982 21 : if (ret == -ENOENT)
1983 21 : ret = 0;
1984 21 : goto out;
1985 : }
1986 : }
1987 :
1988 640 : ret = btrfs_commit_transaction(trans);
1989 275768 : out:
1990 275768 : ASSERT(list_empty(&ctx.list));
1991 275768 : ASSERT(list_empty(&ctx.conflict_inodes));
1992 275768 : err = file_check_and_advance_wb_err(file);
1993 275774 : if (!ret)
1994 275763 : ret = err;
1995 275774 : return ret > 0 ? -EIO : ret;
1996 :
1997 19838 : out_release_extents:
1998 19838 : btrfs_release_log_ctx_extents(&ctx);
1999 19838 : btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
2000 19837 : goto out;
2001 : }
2002 :
2003 : static const struct vm_operations_struct btrfs_file_vm_ops = {
2004 : .fault = filemap_fault,
2005 : .map_pages = filemap_map_pages,
2006 : .page_mkwrite = btrfs_page_mkwrite,
2007 : };
2008 :
2009 434508 : static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
2010 : {
2011 434508 : struct address_space *mapping = filp->f_mapping;
2012 :
2013 434508 : if (!mapping->a_ops->read_folio)
2014 : return -ENOEXEC;
2015 :
2016 434508 : file_accessed(filp);
2017 434504 : vma->vm_ops = &btrfs_file_vm_ops;
2018 :
2019 434504 : return 0;
2020 : }
2021 :
2022 0 : static int hole_mergeable(struct btrfs_inode *inode, struct extent_buffer *leaf,
2023 : int slot, u64 start, u64 end)
2024 : {
2025 0 : struct btrfs_file_extent_item *fi;
2026 0 : struct btrfs_key key;
2027 :
2028 0 : if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2029 : return 0;
2030 :
2031 0 : btrfs_item_key_to_cpu(leaf, &key, slot);
2032 0 : if (key.objectid != btrfs_ino(inode) ||
2033 0 : key.type != BTRFS_EXTENT_DATA_KEY)
2034 : return 0;
2035 :
2036 0 : fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2037 :
2038 0 : if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2039 : return 0;
2040 :
2041 0 : if (btrfs_file_extent_disk_bytenr(leaf, fi))
2042 : return 0;
2043 :
2044 0 : if (key.offset == end)
2045 : return 1;
2046 0 : if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2047 0 : return 1;
2048 : return 0;
2049 : }
2050 :
2051 3365002 : static int fill_holes(struct btrfs_trans_handle *trans,
2052 : struct btrfs_inode *inode,
2053 : struct btrfs_path *path, u64 offset, u64 end)
2054 : {
2055 3365002 : struct btrfs_fs_info *fs_info = trans->fs_info;
2056 3365002 : struct btrfs_root *root = inode->root;
2057 3365002 : struct extent_buffer *leaf;
2058 3365002 : struct btrfs_file_extent_item *fi;
2059 3365002 : struct extent_map *hole_em;
2060 3365002 : struct btrfs_key key;
2061 3365002 : int ret;
2062 :
2063 3365002 : if (btrfs_fs_incompat(fs_info, NO_HOLES))
2064 3365002 : goto out;
2065 :
2066 0 : key.objectid = btrfs_ino(inode);
2067 0 : key.type = BTRFS_EXTENT_DATA_KEY;
2068 0 : key.offset = offset;
2069 :
2070 0 : ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2071 0 : if (ret <= 0) {
2072 : /*
2073 : * We should have dropped this offset, so if we find it then
2074 : * something has gone horribly wrong.
2075 : */
2076 0 : if (ret == 0)
2077 0 : ret = -EINVAL;
2078 0 : return ret;
2079 : }
2080 :
2081 0 : leaf = path->nodes[0];
2082 0 : if (hole_mergeable(inode, leaf, path->slots[0] - 1, offset, end)) {
2083 0 : u64 num_bytes;
2084 :
2085 0 : path->slots[0]--;
2086 0 : fi = btrfs_item_ptr(leaf, path->slots[0],
2087 : struct btrfs_file_extent_item);
2088 0 : num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2089 : end - offset;
2090 0 : btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2091 0 : btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2092 0 : btrfs_set_file_extent_offset(leaf, fi, 0);
2093 0 : btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2094 0 : btrfs_mark_buffer_dirty(leaf);
2095 0 : goto out;
2096 : }
2097 :
2098 0 : if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) {
2099 0 : u64 num_bytes;
2100 :
2101 0 : key.offset = offset;
2102 0 : btrfs_set_item_key_safe(fs_info, path, &key);
2103 0 : fi = btrfs_item_ptr(leaf, path->slots[0],
2104 : struct btrfs_file_extent_item);
2105 0 : num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2106 : offset;
2107 0 : btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2108 0 : btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2109 0 : btrfs_set_file_extent_offset(leaf, fi, 0);
2110 0 : btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2111 0 : btrfs_mark_buffer_dirty(leaf);
2112 0 : goto out;
2113 : }
2114 0 : btrfs_release_path(path);
2115 :
2116 0 : ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset,
2117 : end - offset);
2118 0 : if (ret)
2119 : return ret;
2120 :
2121 0 : out:
2122 3365002 : btrfs_release_path(path);
2123 :
2124 3365022 : hole_em = alloc_extent_map();
2125 3365028 : if (!hole_em) {
2126 0 : btrfs_drop_extent_map_range(inode, offset, end - 1, false);
2127 0 : btrfs_set_inode_full_sync(inode);
2128 : } else {
2129 3365028 : hole_em->start = offset;
2130 3365028 : hole_em->len = end - offset;
2131 3365028 : hole_em->ram_bytes = hole_em->len;
2132 3365028 : hole_em->orig_start = offset;
2133 :
2134 3365028 : hole_em->block_start = EXTENT_MAP_HOLE;
2135 3365028 : hole_em->block_len = 0;
2136 3365028 : hole_em->orig_block_len = 0;
2137 3365028 : hole_em->compress_type = BTRFS_COMPRESS_NONE;
2138 3365028 : hole_em->generation = trans->transid;
2139 :
2140 3365028 : ret = btrfs_replace_extent_map_range(inode, hole_em, true);
2141 3365052 : free_extent_map(hole_em);
2142 3365088 : if (ret)
2143 0 : btrfs_set_inode_full_sync(inode);
2144 : }
2145 :
2146 : return 0;
2147 : }
2148 :
2149 : /*
2150 : * Find a hole extent on given inode and change start/len to the end of hole
2151 : * extent.(hole/vacuum extent whose em->start <= start &&
2152 : * em->start + em->len > start)
2153 : * When a hole extent is found, return 1 and modify start/len.
2154 : */
2155 4806739 : static int find_first_non_hole(struct btrfs_inode *inode, u64 *start, u64 *len)
2156 : {
2157 4806739 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
2158 4806739 : struct extent_map *em;
2159 4806739 : int ret = 0;
2160 :
2161 4806739 : em = btrfs_get_extent(inode, NULL, 0,
2162 4806739 : round_down(*start, fs_info->sectorsize),
2163 4806739 : round_up(*len, fs_info->sectorsize));
2164 4806791 : if (IS_ERR(em))
2165 0 : return PTR_ERR(em);
2166 :
2167 : /* Hole or vacuum extent(only exists in no-hole mode) */
2168 4806791 : if (em->block_start == EXTENT_MAP_HOLE) {
2169 150402 : ret = 1;
2170 300804 : *len = em->start + em->len > *start + *len ?
2171 150402 : 0 : *start + *len - em->start - em->len;
2172 150402 : *start = em->start + em->len;
2173 : }
2174 4806791 : free_extent_map(em);
2175 4806791 : return ret;
2176 : }
2177 :
2178 2498213 : static void btrfs_punch_hole_lock_range(struct inode *inode,
2179 : const u64 lockstart,
2180 : const u64 lockend,
2181 : struct extent_state **cached_state)
2182 : {
2183 : /*
2184 : * For subpage case, if the range is not at page boundary, we could
2185 : * have pages at the leading/tailing part of the range.
2186 : * This could lead to dead loop since filemap_range_has_page()
2187 : * will always return true.
2188 : * So here we need to do extra page alignment for
2189 : * filemap_range_has_page().
2190 : */
2191 2498213 : const u64 page_lockstart = round_up(lockstart, PAGE_SIZE);
2192 2498213 : const u64 page_lockend = round_down(lockend + 1, PAGE_SIZE) - 1;
2193 :
2194 2498213 : while (1) {
2195 2498213 : truncate_pagecache_range(inode, lockstart, lockend);
2196 :
2197 2498163 : lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2198 : cached_state);
2199 : /*
2200 : * We can't have ordered extents in the range, nor dirty/writeback
2201 : * pages, because we have locked the inode's VFS lock in exclusive
2202 : * mode, we have locked the inode's i_mmap_lock in exclusive mode,
2203 : * we have flushed all delalloc in the range and we have waited
2204 : * for any ordered extents in the range to complete.
2205 : * We can race with anyone reading pages from this range, so after
2206 : * locking the range check if we have pages in the range, and if
2207 : * we do, unlock the range and retry.
2208 : */
2209 2498211 : if (!filemap_range_has_page(inode->i_mapping, page_lockstart,
2210 : page_lockend))
2211 : break;
2212 :
2213 0 : unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2214 : cached_state);
2215 : }
2216 :
2217 2498184 : btrfs_assert_inode_range_clean(BTRFS_I(inode), lockstart, lockend);
2218 2498155 : }
2219 :
2220 15793201 : static int btrfs_insert_replace_extent(struct btrfs_trans_handle *trans,
2221 : struct btrfs_inode *inode,
2222 : struct btrfs_path *path,
2223 : struct btrfs_replace_extent_info *extent_info,
2224 : const u64 replace_len,
2225 : const u64 bytes_to_drop)
2226 : {
2227 15793201 : struct btrfs_fs_info *fs_info = trans->fs_info;
2228 15793201 : struct btrfs_root *root = inode->root;
2229 15793201 : struct btrfs_file_extent_item *extent;
2230 15793201 : struct extent_buffer *leaf;
2231 15793201 : struct btrfs_key key;
2232 15793201 : int slot;
2233 15793201 : struct btrfs_ref ref = { 0 };
2234 15793201 : int ret;
2235 :
2236 15793201 : if (replace_len == 0)
2237 : return 0;
2238 :
2239 15793201 : if (extent_info->disk_offset == 0 &&
2240 11 : btrfs_fs_incompat(fs_info, NO_HOLES)) {
2241 0 : btrfs_update_inode_bytes(inode, 0, bytes_to_drop);
2242 0 : return 0;
2243 : }
2244 :
2245 15793201 : key.objectid = btrfs_ino(inode);
2246 15793201 : key.type = BTRFS_EXTENT_DATA_KEY;
2247 15793201 : key.offset = extent_info->file_offset;
2248 15793201 : ret = btrfs_insert_empty_item(trans, root, path, &key,
2249 : sizeof(struct btrfs_file_extent_item));
2250 15793204 : if (ret)
2251 : return ret;
2252 15793203 : leaf = path->nodes[0];
2253 15793203 : slot = path->slots[0];
2254 15793203 : write_extent_buffer(leaf, extent_info->extent_buf,
2255 15793203 : btrfs_item_ptr_offset(leaf, slot),
2256 : sizeof(struct btrfs_file_extent_item));
2257 15793202 : extent = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2258 15793202 : ASSERT(btrfs_file_extent_type(leaf, extent) != BTRFS_FILE_EXTENT_INLINE);
2259 15793198 : btrfs_set_file_extent_offset(leaf, extent, extent_info->data_offset);
2260 15793198 : btrfs_set_file_extent_num_bytes(leaf, extent, replace_len);
2261 15793199 : if (extent_info->is_new_extent)
2262 435240 : btrfs_set_file_extent_generation(leaf, extent, trans->transid);
2263 15793199 : btrfs_mark_buffer_dirty(leaf);
2264 15793203 : btrfs_release_path(path);
2265 :
2266 15793206 : ret = btrfs_inode_set_file_extent_range(inode, extent_info->file_offset,
2267 : replace_len);
2268 15793207 : if (ret)
2269 : return ret;
2270 :
2271 : /* If it's a hole, nothing more needs to be done. */
2272 15793207 : if (extent_info->disk_offset == 0) {
2273 11 : btrfs_update_inode_bytes(inode, 0, bytes_to_drop);
2274 11 : return 0;
2275 : }
2276 :
2277 15793196 : btrfs_update_inode_bytes(inode, replace_len, bytes_to_drop);
2278 :
2279 15793192 : if (extent_info->is_new_extent && extent_info->insertions == 0) {
2280 435237 : key.objectid = extent_info->disk_offset;
2281 435237 : key.type = BTRFS_EXTENT_ITEM_KEY;
2282 435237 : key.offset = extent_info->disk_len;
2283 435237 : ret = btrfs_alloc_reserved_file_extent(trans, root,
2284 : btrfs_ino(inode),
2285 : extent_info->file_offset,
2286 435237 : extent_info->qgroup_reserved,
2287 : &key);
2288 : } else {
2289 15357955 : u64 ref_offset;
2290 :
2291 15357955 : btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
2292 : extent_info->disk_offset,
2293 : extent_info->disk_len, 0);
2294 15357955 : ref_offset = extent_info->file_offset - extent_info->data_offset;
2295 15357955 : btrfs_init_data_ref(&ref, root->root_key.objectid,
2296 : btrfs_ino(inode), ref_offset, 0, false);
2297 15357955 : ret = btrfs_inc_extent_ref(trans, &ref);
2298 : }
2299 :
2300 15793196 : extent_info->insertions++;
2301 :
2302 15793196 : return ret;
2303 : }
2304 :
2305 : /*
2306 : * The respective range must have been previously locked, as well as the inode.
2307 : * The end offset is inclusive (last byte of the range).
2308 : * @extent_info is NULL for fallocate's hole punching and non-NULL when replacing
2309 : * the file range with an extent.
2310 : * When not punching a hole, we don't want to end up in a state where we dropped
2311 : * extents without inserting a new one, so we must abort the transaction to avoid
2312 : * a corruption.
2313 : */
2314 19199023 : int btrfs_replace_file_extents(struct btrfs_inode *inode,
2315 : struct btrfs_path *path, const u64 start,
2316 : const u64 end,
2317 : struct btrfs_replace_extent_info *extent_info,
2318 : struct btrfs_trans_handle **trans_out)
2319 : {
2320 19199023 : struct btrfs_drop_extents_args drop_args = { 0 };
2321 19199023 : struct btrfs_root *root = inode->root;
2322 19199023 : struct btrfs_fs_info *fs_info = root->fs_info;
2323 19199023 : u64 min_size = btrfs_calc_insert_metadata_size(fs_info, 1);
2324 19199023 : u64 ino_size = round_up(inode->vfs_inode.i_size, fs_info->sectorsize);
2325 19199023 : struct btrfs_trans_handle *trans = NULL;
2326 19199023 : struct btrfs_block_rsv *rsv;
2327 19199023 : unsigned int rsv_count;
2328 19199023 : u64 cur_offset;
2329 19199023 : u64 len = end - start;
2330 19199023 : int ret = 0;
2331 :
2332 19199023 : if (end <= start)
2333 : return -EINVAL;
2334 :
2335 19199023 : rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
2336 19198781 : if (!rsv) {
2337 0 : ret = -ENOMEM;
2338 0 : goto out;
2339 : }
2340 19198781 : rsv->size = btrfs_calc_insert_metadata_size(fs_info, 1);
2341 19198781 : rsv->failfast = true;
2342 :
2343 : /*
2344 : * 1 - update the inode
2345 : * 1 - removing the extents in the range
2346 : * 1 - adding the hole extent if no_holes isn't set or if we are
2347 : * replacing the range with a new extent
2348 : */
2349 19198781 : if (!btrfs_fs_incompat(fs_info, NO_HOLES) || extent_info)
2350 : rsv_count = 3;
2351 : else
2352 3405389 : rsv_count = 2;
2353 :
2354 19198781 : trans = btrfs_start_transaction(root, rsv_count);
2355 19199066 : if (IS_ERR(trans)) {
2356 2099 : ret = PTR_ERR(trans);
2357 2099 : trans = NULL;
2358 2099 : goto out_free;
2359 : }
2360 :
2361 19196967 : ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
2362 : min_size, false);
2363 19196940 : if (WARN_ON(ret))
2364 0 : goto out_trans;
2365 19196940 : trans->block_rsv = rsv;
2366 :
2367 19196940 : cur_offset = start;
2368 19196940 : drop_args.path = path;
2369 19196940 : drop_args.end = end + 1;
2370 19196940 : drop_args.drop_cache = true;
2371 19196949 : while (cur_offset < end) {
2372 19196949 : drop_args.start = cur_offset;
2373 19196949 : ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2374 : /* If we are punching a hole decrement the inode's byte count */
2375 19196938 : if (!extent_info)
2376 3403784 : btrfs_update_inode_bytes(inode, 0,
2377 : drop_args.bytes_found);
2378 19196885 : if (ret != -ENOSPC) {
2379 : /*
2380 : * The only time we don't want to abort is if we are
2381 : * attempting to clone a partial inline extent, in which
2382 : * case we'll get EOPNOTSUPP. However if we aren't
2383 : * clone we need to abort no matter what, because if we
2384 : * got EOPNOTSUPP via prealloc then we messed up and
2385 : * need to abort.
2386 : */
2387 19196876 : if (ret &&
2388 0 : (ret != -EOPNOTSUPP ||
2389 0 : (extent_info && extent_info->is_new_extent)))
2390 0 : btrfs_abort_transaction(trans, ret);
2391 : break;
2392 : }
2393 :
2394 9 : trans->block_rsv = &fs_info->trans_block_rsv;
2395 :
2396 9 : if (!extent_info && cur_offset < drop_args.drop_end &&
2397 : cur_offset < ino_size) {
2398 0 : ret = fill_holes(trans, inode, path, cur_offset,
2399 : drop_args.drop_end);
2400 0 : if (ret) {
2401 : /*
2402 : * If we failed then we didn't insert our hole
2403 : * entries for the area we dropped, so now the
2404 : * fs is corrupted, so we must abort the
2405 : * transaction.
2406 : */
2407 0 : btrfs_abort_transaction(trans, ret);
2408 0 : break;
2409 : }
2410 9 : } else if (!extent_info && cur_offset < drop_args.drop_end) {
2411 : /*
2412 : * We are past the i_size here, but since we didn't
2413 : * insert holes we need to clear the mapped area so we
2414 : * know to not set disk_i_size in this area until a new
2415 : * file extent is inserted here.
2416 : */
2417 0 : ret = btrfs_inode_clear_file_extent_range(inode,
2418 : cur_offset,
2419 : drop_args.drop_end - cur_offset);
2420 0 : if (ret) {
2421 : /*
2422 : * We couldn't clear our area, so we could
2423 : * presumably adjust up and corrupt the fs, so
2424 : * we need to abort.
2425 : */
2426 0 : btrfs_abort_transaction(trans, ret);
2427 0 : break;
2428 : }
2429 : }
2430 :
2431 9 : if (extent_info &&
2432 9 : drop_args.drop_end > extent_info->file_offset) {
2433 9 : u64 replace_len = drop_args.drop_end -
2434 : extent_info->file_offset;
2435 :
2436 9 : ret = btrfs_insert_replace_extent(trans, inode, path,
2437 : extent_info, replace_len,
2438 : drop_args.bytes_found);
2439 9 : if (ret) {
2440 0 : btrfs_abort_transaction(trans, ret);
2441 0 : break;
2442 : }
2443 9 : extent_info->data_len -= replace_len;
2444 9 : extent_info->data_offset += replace_len;
2445 9 : extent_info->file_offset += replace_len;
2446 : }
2447 :
2448 : /*
2449 : * We are releasing our handle on the transaction, balance the
2450 : * dirty pages of the btree inode and flush delayed items, and
2451 : * then get a new transaction handle, which may now point to a
2452 : * new transaction in case someone else may have committed the
2453 : * transaction we used to replace/drop file extent items. So
2454 : * bump the inode's iversion and update mtime and ctime except
2455 : * if we are called from a dedupe context. This is because a
2456 : * power failure/crash may happen after the transaction is
2457 : * committed and before we finish replacing/dropping all the
2458 : * file extent items we need.
2459 : */
2460 9 : inode_inc_iversion(&inode->vfs_inode);
2461 :
2462 9 : if (!extent_info || extent_info->update_times) {
2463 9 : inode->vfs_inode.i_mtime = current_time(&inode->vfs_inode);
2464 9 : inode->vfs_inode.i_ctime = inode->vfs_inode.i_mtime;
2465 : }
2466 :
2467 9 : ret = btrfs_update_inode(trans, root, inode);
2468 9 : if (ret)
2469 : break;
2470 :
2471 9 : btrfs_end_transaction(trans);
2472 9 : btrfs_btree_balance_dirty(fs_info);
2473 :
2474 9 : trans = btrfs_start_transaction(root, rsv_count);
2475 9 : if (IS_ERR(trans)) {
2476 0 : ret = PTR_ERR(trans);
2477 0 : trans = NULL;
2478 0 : break;
2479 : }
2480 :
2481 9 : ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
2482 : rsv, min_size, false);
2483 0 : if (WARN_ON(ret))
2484 : break;
2485 9 : trans->block_rsv = rsv;
2486 :
2487 9 : cur_offset = drop_args.drop_end;
2488 9 : len = end - cur_offset;
2489 9 : if (!extent_info && len) {
2490 0 : ret = find_first_non_hole(inode, &cur_offset, &len);
2491 0 : if (unlikely(ret < 0))
2492 : break;
2493 0 : if (ret && !len) {
2494 : ret = 0;
2495 : break;
2496 : }
2497 : }
2498 : }
2499 :
2500 : /*
2501 : * If we were cloning, force the next fsync to be a full one since we
2502 : * we replaced (or just dropped in the case of cloning holes when
2503 : * NO_HOLES is enabled) file extent items and did not setup new extent
2504 : * maps for the replacement extents (or holes).
2505 : */
2506 19196823 : if (extent_info && !extent_info->is_new_extent)
2507 15357958 : btrfs_set_inode_full_sync(inode);
2508 :
2509 19196824 : if (ret)
2510 0 : goto out_trans;
2511 :
2512 19196824 : trans->block_rsv = &fs_info->trans_block_rsv;
2513 : /*
2514 : * If we are using the NO_HOLES feature we might have had already an
2515 : * hole that overlaps a part of the region [lockstart, lockend] and
2516 : * ends at (or beyond) lockend. Since we have no file extent items to
2517 : * represent holes, drop_end can be less than lockend and so we must
2518 : * make sure we have an extent map representing the existing hole (the
2519 : * call to __btrfs_drop_extents() might have dropped the existing extent
2520 : * map representing the existing hole), otherwise the fast fsync path
2521 : * will not record the existence of the hole region
2522 : * [existing_hole_start, lockend].
2523 : */
2524 19196824 : if (drop_args.drop_end <= end)
2525 231523 : drop_args.drop_end = end + 1;
2526 : /*
2527 : * Don't insert file hole extent item if it's for a range beyond eof
2528 : * (because it's useless) or if it represents a 0 bytes range (when
2529 : * cur_offset == drop_end).
2530 : */
2531 19196824 : if (!extent_info && cur_offset < ino_size &&
2532 3364972 : cur_offset < drop_args.drop_end) {
2533 3364956 : ret = fill_holes(trans, inode, path, cur_offset,
2534 : drop_args.drop_end);
2535 3365075 : if (ret) {
2536 : /* Same comment as above. */
2537 0 : btrfs_abort_transaction(trans, ret);
2538 0 : goto out_trans;
2539 : }
2540 15831868 : } else if (!extent_info && cur_offset < drop_args.drop_end) {
2541 : /* See the comment in the loop above for the reasoning here. */
2542 38686 : ret = btrfs_inode_clear_file_extent_range(inode, cur_offset,
2543 : drop_args.drop_end - cur_offset);
2544 38686 : if (ret) {
2545 0 : btrfs_abort_transaction(trans, ret);
2546 0 : goto out_trans;
2547 : }
2548 :
2549 : }
2550 19196943 : if (extent_info) {
2551 15793195 : ret = btrfs_insert_replace_extent(trans, inode, path,
2552 : extent_info, extent_info->data_len,
2553 : drop_args.bytes_found);
2554 15793198 : if (ret) {
2555 0 : btrfs_abort_transaction(trans, ret);
2556 0 : goto out_trans;
2557 : }
2558 : }
2559 :
2560 19196946 : out_trans:
2561 19196946 : if (!trans)
2562 0 : goto out_free;
2563 :
2564 19196946 : trans->block_rsv = &fs_info->trans_block_rsv;
2565 19196946 : if (ret)
2566 0 : btrfs_end_transaction(trans);
2567 : else
2568 19196946 : *trans_out = trans;
2569 19199045 : out_free:
2570 19199045 : btrfs_free_block_rsv(fs_info, rsv);
2571 : out:
2572 : return ret;
2573 : }
2574 :
2575 2386368 : static int btrfs_punch_hole(struct file *file, loff_t offset, loff_t len)
2576 : {
2577 2386368 : struct inode *inode = file_inode(file);
2578 2386368 : struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2579 2386368 : struct btrfs_root *root = BTRFS_I(inode)->root;
2580 2386368 : struct extent_state *cached_state = NULL;
2581 2386368 : struct btrfs_path *path;
2582 2386368 : struct btrfs_trans_handle *trans = NULL;
2583 2386368 : u64 lockstart;
2584 2386368 : u64 lockend;
2585 2386368 : u64 tail_start;
2586 2386368 : u64 tail_len;
2587 2386368 : u64 orig_start = offset;
2588 2386368 : int ret = 0;
2589 2386368 : bool same_block;
2590 2386368 : u64 ino_size;
2591 2386368 : bool truncated_block = false;
2592 2386368 : bool updated_inode = false;
2593 :
2594 2386368 : btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
2595 :
2596 2386402 : ret = btrfs_wait_ordered_range(inode, offset, len);
2597 2386384 : if (ret)
2598 0 : goto out_only_mutex;
2599 :
2600 2386384 : ino_size = round_up(inode->i_size, fs_info->sectorsize);
2601 2386384 : ret = find_first_non_hole(BTRFS_I(inode), &offset, &len);
2602 2386382 : if (ret < 0)
2603 0 : goto out_only_mutex;
2604 2386382 : if (ret && !len) {
2605 : /* Already in a large hole */
2606 47840 : ret = 0;
2607 47840 : goto out_only_mutex;
2608 : }
2609 :
2610 2338542 : ret = file_modified(file);
2611 2338498 : if (ret)
2612 56 : goto out_only_mutex;
2613 :
2614 2338442 : lockstart = round_up(offset, fs_info->sectorsize);
2615 2338442 : lockend = round_down(offset + len, fs_info->sectorsize) - 1;
2616 2338442 : same_block = (BTRFS_BYTES_TO_BLKS(fs_info, offset))
2617 2338442 : == (BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1));
2618 : /*
2619 : * We needn't truncate any block which is beyond the end of the file
2620 : * because we are sure there is no data there.
2621 : */
2622 : /*
2623 : * Only do this if we are in the same block and we aren't doing the
2624 : * entire block.
2625 : */
2626 2338442 : if (same_block && len < fs_info->sectorsize) {
2627 9848 : if (offset < ino_size) {
2628 7042 : truncated_block = true;
2629 7042 : ret = btrfs_truncate_block(BTRFS_I(inode), offset, len,
2630 : 0);
2631 : } else {
2632 : ret = 0;
2633 : }
2634 9848 : goto out_only_mutex;
2635 : }
2636 :
2637 : /* zero back part of the first block */
2638 2328594 : if (offset < ino_size) {
2639 2328061 : truncated_block = true;
2640 2328061 : ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0);
2641 2328023 : if (ret) {
2642 35 : btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
2643 35 : return ret;
2644 : }
2645 : }
2646 :
2647 : /* Check the aligned pages after the first unaligned page,
2648 : * if offset != orig_start, which means the first unaligned page
2649 : * including several following pages are already in holes,
2650 : * the extra check can be skipped */
2651 2328521 : if (offset == orig_start) {
2652 : /* after truncate page, check hole again */
2653 2281747 : len = offset + len - lockstart;
2654 2281747 : offset = lockstart;
2655 2281747 : ret = find_first_non_hole(BTRFS_I(inode), &offset, &len);
2656 2281810 : if (ret < 0)
2657 0 : goto out_only_mutex;
2658 2281810 : if (ret && !len) {
2659 2354 : ret = 0;
2660 2354 : goto out_only_mutex;
2661 : }
2662 2279456 : lockstart = offset;
2663 : }
2664 :
2665 : /* Check the tail unaligned part is in a hole */
2666 2326230 : tail_start = lockend + 1;
2667 2326230 : tail_len = offset + len - tail_start;
2668 2326230 : if (tail_len) {
2669 138603 : ret = find_first_non_hole(BTRFS_I(inode), &tail_start, &tail_len);
2670 138603 : if (unlikely(ret < 0))
2671 0 : goto out_only_mutex;
2672 138603 : if (!ret) {
2673 : /* zero the front end of the last page */
2674 95556 : if (tail_start + tail_len < ino_size) {
2675 95364 : truncated_block = true;
2676 95364 : ret = btrfs_truncate_block(BTRFS_I(inode),
2677 : tail_start + tail_len,
2678 : 0, 1);
2679 95364 : if (ret)
2680 10 : goto out_only_mutex;
2681 : }
2682 : }
2683 : }
2684 :
2685 2326220 : if (lockend < lockstart) {
2686 5171 : ret = 0;
2687 5171 : goto out_only_mutex;
2688 : }
2689 :
2690 2321049 : btrfs_punch_hole_lock_range(inode, lockstart, lockend, &cached_state);
2691 :
2692 2321025 : path = btrfs_alloc_path();
2693 2321070 : if (!path) {
2694 0 : ret = -ENOMEM;
2695 0 : goto out;
2696 : }
2697 :
2698 2321070 : ret = btrfs_replace_file_extents(BTRFS_I(inode), path, lockstart,
2699 : lockend, NULL, &trans);
2700 2321098 : btrfs_free_path(path);
2701 2321063 : if (ret)
2702 41 : goto out;
2703 :
2704 2321022 : ASSERT(trans != NULL);
2705 2321022 : inode_inc_iversion(inode);
2706 2321052 : inode->i_mtime = current_time(inode);
2707 2321051 : inode->i_ctime = inode->i_mtime;
2708 2321051 : ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
2709 2321062 : updated_inode = true;
2710 2321062 : btrfs_end_transaction(trans);
2711 2321061 : btrfs_btree_balance_dirty(fs_info);
2712 2321066 : out:
2713 2321066 : unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2714 : &cached_state);
2715 2386371 : out_only_mutex:
2716 2386371 : if (!updated_inode && truncated_block && !ret) {
2717 : /*
2718 : * If we only end up zeroing part of a page, we still need to
2719 : * update the inode item, so that all the time fields are
2720 : * updated as well as the necessary btrfs inode in memory fields
2721 : * for detecting, at fsync time, if the inode isn't yet in the
2722 : * log tree or it's there but not up to date.
2723 : */
2724 14566 : struct timespec64 now = current_time(inode);
2725 :
2726 14566 : inode_inc_iversion(inode);
2727 14566 : inode->i_mtime = now;
2728 14566 : inode->i_ctime = now;
2729 14566 : trans = btrfs_start_transaction(root, 1);
2730 14566 : if (IS_ERR(trans)) {
2731 0 : ret = PTR_ERR(trans);
2732 : } else {
2733 14566 : int ret2;
2734 :
2735 14566 : ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
2736 14566 : ret2 = btrfs_end_transaction(trans);
2737 14566 : if (!ret)
2738 14566 : ret = ret2;
2739 : }
2740 : }
2741 2386371 : btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
2742 2386371 : return ret;
2743 : }
2744 :
2745 : /* Helper structure to record which range is already reserved */
2746 : struct falloc_range {
2747 : struct list_head list;
2748 : u64 start;
2749 : u64 len;
2750 : };
2751 :
2752 : /*
2753 : * Helper function to add falloc range
2754 : *
2755 : * Caller should have locked the larger range of extent containing
2756 : * [start, len)
2757 : */
2758 312240 : static int add_falloc_range(struct list_head *head, u64 start, u64 len)
2759 : {
2760 312240 : struct falloc_range *range = NULL;
2761 :
2762 312240 : if (!list_empty(head)) {
2763 : /*
2764 : * As fallocate iterates by bytenr order, we only need to check
2765 : * the last range.
2766 : */
2767 110563 : range = list_last_entry(head, struct falloc_range, list);
2768 110563 : if (range->start + range->len == start) {
2769 63289 : range->len += len;
2770 63289 : return 0;
2771 : }
2772 : }
2773 :
2774 248951 : range = kmalloc(sizeof(*range), GFP_KERNEL);
2775 248951 : if (!range)
2776 : return -ENOMEM;
2777 248951 : range->start = start;
2778 248951 : range->len = len;
2779 248951 : list_add_tail(&range->list, head);
2780 248951 : return 0;
2781 : }
2782 :
2783 692278 : static int btrfs_fallocate_update_isize(struct inode *inode,
2784 : const u64 end,
2785 : const int mode)
2786 : {
2787 692278 : struct btrfs_trans_handle *trans;
2788 692278 : struct btrfs_root *root = BTRFS_I(inode)->root;
2789 692278 : int ret;
2790 692278 : int ret2;
2791 :
2792 692278 : if (mode & FALLOC_FL_KEEP_SIZE || end <= i_size_read(inode))
2793 : return 0;
2794 :
2795 266275 : trans = btrfs_start_transaction(root, 1);
2796 266275 : if (IS_ERR(trans))
2797 0 : return PTR_ERR(trans);
2798 :
2799 266275 : inode->i_ctime = current_time(inode);
2800 266275 : i_size_write(inode, end);
2801 266275 : btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
2802 266275 : ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
2803 266275 : ret2 = btrfs_end_transaction(trans);
2804 :
2805 266275 : return ret ? ret : ret2;
2806 : }
2807 :
2808 : enum {
2809 : RANGE_BOUNDARY_WRITTEN_EXTENT,
2810 : RANGE_BOUNDARY_PREALLOC_EXTENT,
2811 : RANGE_BOUNDARY_HOLE,
2812 : };
2813 :
2814 328375 : static int btrfs_zero_range_check_range_boundary(struct btrfs_inode *inode,
2815 : u64 offset)
2816 : {
2817 328375 : const u64 sectorsize = inode->root->fs_info->sectorsize;
2818 328375 : struct extent_map *em;
2819 328375 : int ret;
2820 :
2821 328375 : offset = round_down(offset, sectorsize);
2822 328375 : em = btrfs_get_extent(inode, NULL, 0, offset, sectorsize);
2823 328375 : if (IS_ERR(em))
2824 0 : return PTR_ERR(em);
2825 :
2826 328375 : if (em->block_start == EXTENT_MAP_HOLE)
2827 : ret = RANGE_BOUNDARY_HOLE;
2828 250450 : else if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2829 : ret = RANGE_BOUNDARY_PREALLOC_EXTENT;
2830 : else
2831 103761 : ret = RANGE_BOUNDARY_WRITTEN_EXTENT;
2832 :
2833 328375 : free_extent_map(em);
2834 328375 : return ret;
2835 : }
2836 :
2837 189177 : static int btrfs_zero_range(struct inode *inode,
2838 : loff_t offset,
2839 : loff_t len,
2840 : const int mode)
2841 : {
2842 189177 : struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2843 189177 : struct extent_map *em;
2844 189177 : struct extent_changeset *data_reserved = NULL;
2845 189177 : int ret;
2846 189177 : u64 alloc_hint = 0;
2847 189177 : const u64 sectorsize = fs_info->sectorsize;
2848 189177 : u64 alloc_start = round_down(offset, sectorsize);
2849 189177 : u64 alloc_end = round_up(offset + len, sectorsize);
2850 189177 : u64 bytes_to_reserve = 0;
2851 189177 : bool space_reserved = false;
2852 :
2853 189177 : em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start,
2854 : alloc_end - alloc_start);
2855 189177 : if (IS_ERR(em)) {
2856 0 : ret = PTR_ERR(em);
2857 0 : goto out;
2858 : }
2859 :
2860 : /*
2861 : * Avoid hole punching and extent allocation for some cases. More cases
2862 : * could be considered, but these are unlikely common and we keep things
2863 : * as simple as possible for now. Also, intentionally, if the target
2864 : * range contains one or more prealloc extents together with regular
2865 : * extents and holes, we drop all the existing extents and allocate a
2866 : * new prealloc extent, so that we get a larger contiguous disk extent.
2867 : */
2868 378354 : if (em->start <= alloc_start &&
2869 189177 : test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
2870 29999 : const u64 em_end = em->start + em->len;
2871 :
2872 29999 : if (em_end >= offset + len) {
2873 : /*
2874 : * The whole range is already a prealloc extent,
2875 : * do nothing except updating the inode's i_size if
2876 : * needed.
2877 : */
2878 7310 : free_extent_map(em);
2879 7310 : ret = btrfs_fallocate_update_isize(inode, offset + len,
2880 : mode);
2881 7310 : goto out;
2882 : }
2883 : /*
2884 : * Part of the range is already a prealloc extent, so operate
2885 : * only on the remaining part of the range.
2886 : */
2887 22689 : alloc_start = em_end;
2888 22689 : ASSERT(IS_ALIGNED(alloc_start, sectorsize));
2889 22689 : len = offset + len - alloc_start;
2890 22689 : offset = alloc_start;
2891 22689 : alloc_hint = em->block_start + em->len;
2892 : }
2893 181867 : free_extent_map(em);
2894 :
2895 181867 : if (BTRFS_BYTES_TO_BLKS(fs_info, offset) ==
2896 181867 : BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)) {
2897 5352 : em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start,
2898 : sectorsize);
2899 5352 : if (IS_ERR(em)) {
2900 0 : ret = PTR_ERR(em);
2901 0 : goto out;
2902 : }
2903 :
2904 10704 : if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
2905 287 : free_extent_map(em);
2906 287 : ret = btrfs_fallocate_update_isize(inode, offset + len,
2907 : mode);
2908 287 : goto out;
2909 : }
2910 5065 : if (len < sectorsize && em->block_start != EXTENT_MAP_HOLE) {
2911 1787 : free_extent_map(em);
2912 1787 : ret = btrfs_truncate_block(BTRFS_I(inode), offset, len,
2913 : 0);
2914 1787 : if (!ret)
2915 1787 : ret = btrfs_fallocate_update_isize(inode,
2916 : offset + len,
2917 : mode);
2918 1787 : return ret;
2919 : }
2920 3278 : free_extent_map(em);
2921 3278 : alloc_start = round_down(offset, sectorsize);
2922 3278 : alloc_end = alloc_start + sectorsize;
2923 3278 : goto reserve_space;
2924 : }
2925 :
2926 176515 : alloc_start = round_up(offset, sectorsize);
2927 176515 : alloc_end = round_down(offset + len, sectorsize);
2928 :
2929 : /*
2930 : * For unaligned ranges, check the pages at the boundaries, they might
2931 : * map to an extent, in which case we need to partially zero them, or
2932 : * they might map to a hole, in which case we need our allocation range
2933 : * to cover them.
2934 : */
2935 176515 : if (!IS_ALIGNED(offset, sectorsize)) {
2936 154908 : ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode),
2937 : offset);
2938 154908 : if (ret < 0)
2939 0 : goto out;
2940 154908 : if (ret == RANGE_BOUNDARY_HOLE) {
2941 98951 : alloc_start = round_down(offset, sectorsize);
2942 98951 : ret = 0;
2943 55957 : } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
2944 55957 : ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0);
2945 55957 : if (ret)
2946 8 : goto out;
2947 : } else {
2948 : ret = 0;
2949 : }
2950 : }
2951 :
2952 176507 : if (!IS_ALIGNED(offset + len, sectorsize)) {
2953 173467 : ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode),
2954 : offset + len);
2955 173467 : if (ret < 0)
2956 0 : goto out;
2957 173467 : if (ret == RANGE_BOUNDARY_HOLE) {
2958 104199 : alloc_end = round_up(offset + len, sectorsize);
2959 104199 : ret = 0;
2960 69268 : } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
2961 47804 : ret = btrfs_truncate_block(BTRFS_I(inode), offset + len,
2962 : 0, 1);
2963 47804 : if (ret)
2964 6 : goto out;
2965 : } else {
2966 : ret = 0;
2967 : }
2968 : }
2969 :
2970 3040 : reserve_space:
2971 179779 : if (alloc_start < alloc_end) {
2972 177577 : struct extent_state *cached_state = NULL;
2973 177577 : const u64 lockstart = alloc_start;
2974 177577 : const u64 lockend = alloc_end - 1;
2975 :
2976 177577 : bytes_to_reserve = alloc_end - alloc_start;
2977 177577 : ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
2978 : bytes_to_reserve);
2979 177576 : if (ret < 0)
2980 503 : goto out;
2981 177144 : space_reserved = true;
2982 177144 : btrfs_punch_hole_lock_range(inode, lockstart, lockend,
2983 : &cached_state);
2984 177144 : ret = btrfs_qgroup_reserve_data(BTRFS_I(inode), &data_reserved,
2985 : alloc_start, bytes_to_reserve);
2986 177144 : if (ret) {
2987 70 : unlock_extent(&BTRFS_I(inode)->io_tree, lockstart,
2988 : lockend, &cached_state);
2989 70 : goto out;
2990 : }
2991 177074 : ret = btrfs_prealloc_file_range(inode, mode, alloc_start,
2992 : alloc_end - alloc_start,
2993 : i_blocksize(inode),
2994 : offset + len, &alloc_hint);
2995 177074 : unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2996 : &cached_state);
2997 : /* btrfs_prealloc_file_range releases reserved space on error */
2998 177074 : if (ret) {
2999 1 : space_reserved = false;
3000 1 : goto out;
3001 : }
3002 : }
3003 179275 : ret = btrfs_fallocate_update_isize(inode, offset + len, mode);
3004 187389 : out:
3005 187389 : if (ret && space_reserved)
3006 70 : btrfs_free_reserved_data_space(BTRFS_I(inode), data_reserved,
3007 : alloc_start, bytes_to_reserve);
3008 187389 : extent_changeset_free(data_reserved);
3009 :
3010 187389 : return ret;
3011 : }
3012 :
3013 3095024 : static long btrfs_fallocate(struct file *file, int mode,
3014 : loff_t offset, loff_t len)
3015 : {
3016 3095024 : struct inode *inode = file_inode(file);
3017 3095024 : struct extent_state *cached_state = NULL;
3018 3095024 : struct extent_changeset *data_reserved = NULL;
3019 3095024 : struct falloc_range *range;
3020 3095024 : struct falloc_range *tmp;
3021 3095024 : struct list_head reserve_list;
3022 3095024 : u64 cur_offset;
3023 3095024 : u64 last_byte;
3024 3095024 : u64 alloc_start;
3025 3095024 : u64 alloc_end;
3026 3095024 : u64 alloc_hint = 0;
3027 3095024 : u64 locked_end;
3028 3095024 : u64 actual_end = 0;
3029 3095024 : u64 data_space_needed = 0;
3030 3095024 : u64 data_space_reserved = 0;
3031 3095024 : u64 qgroup_reserved = 0;
3032 3095024 : struct extent_map *em;
3033 3095024 : int blocksize = BTRFS_I(inode)->root->fs_info->sectorsize;
3034 3095024 : int ret;
3035 :
3036 : /* Do not allow fallocate in ZONED mode */
3037 3095024 : if (btrfs_is_zoned(btrfs_sb(inode->i_sb)))
3038 : return -EOPNOTSUPP;
3039 :
3040 3095024 : alloc_start = round_down(offset, blocksize);
3041 3095024 : alloc_end = round_up(offset + len, blocksize);
3042 3095024 : cur_offset = alloc_start;
3043 :
3044 : /* Make sure we aren't being give some crap mode */
3045 3095024 : if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
3046 : FALLOC_FL_ZERO_RANGE))
3047 : return -EOPNOTSUPP;
3048 :
3049 3084991 : if (mode & FALLOC_FL_PUNCH_HOLE)
3050 2386372 : return btrfs_punch_hole(file, offset, len);
3051 :
3052 698619 : btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
3053 :
3054 698619 : if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) {
3055 358446 : ret = inode_newsize_ok(inode, offset + len);
3056 358446 : if (ret)
3057 1 : goto out;
3058 : }
3059 :
3060 698618 : ret = file_modified(file);
3061 698619 : if (ret)
3062 139 : goto out;
3063 :
3064 : /*
3065 : * TODO: Move these two operations after we have checked
3066 : * accurate reserved space, or fallocate can still fail but
3067 : * with page truncated or size expanded.
3068 : *
3069 : * But that's a minor problem and won't do much harm BTW.
3070 : */
3071 698480 : if (alloc_start > inode->i_size) {
3072 93206 : ret = btrfs_cont_expand(BTRFS_I(inode), i_size_read(inode),
3073 : alloc_start);
3074 93207 : if (ret)
3075 206 : goto out;
3076 605274 : } else if (offset + len > inode->i_size) {
3077 : /*
3078 : * If we are fallocating from the end of the file onward we
3079 : * need to zero out the end of the block if i_size lands in the
3080 : * middle of a block.
3081 : */
3082 325986 : ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
3083 325986 : if (ret)
3084 87 : goto out;
3085 : }
3086 :
3087 : /*
3088 : * We have locked the inode at the VFS level (in exclusive mode) and we
3089 : * have locked the i_mmap_lock lock (in exclusive mode). Now before
3090 : * locking the file range, flush all dealloc in the range and wait for
3091 : * all ordered extents in the range to complete. After this we can lock
3092 : * the file range and, due to the previous locking we did, we know there
3093 : * can't be more delalloc or ordered extents in the range.
3094 : */
3095 698188 : ret = btrfs_wait_ordered_range(inode, alloc_start,
3096 : alloc_end - alloc_start);
3097 698188 : if (ret)
3098 0 : goto out;
3099 :
3100 698188 : if (mode & FALLOC_FL_ZERO_RANGE) {
3101 189177 : ret = btrfs_zero_range(inode, offset, len, mode);
3102 189175 : btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
3103 189177 : return ret;
3104 : }
3105 :
3106 509011 : locked_end = alloc_end - 1;
3107 509011 : lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
3108 : &cached_state);
3109 :
3110 509011 : btrfs_assert_inode_range_clean(BTRFS_I(inode), alloc_start, locked_end);
3111 :
3112 : /* First, check if we exceed the qgroup limit */
3113 509011 : INIT_LIST_HEAD(&reserve_list);
3114 1638213 : while (cur_offset < alloc_end) {
3115 1129341 : em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
3116 : alloc_end - cur_offset);
3117 1129341 : if (IS_ERR(em)) {
3118 0 : ret = PTR_ERR(em);
3119 0 : break;
3120 : }
3121 1129341 : last_byte = min(extent_map_end(em), alloc_end);
3122 1129341 : actual_end = min_t(u64, extent_map_end(em), offset + len);
3123 1129341 : last_byte = ALIGN(last_byte, blocksize);
3124 1129341 : if (em->block_start == EXTENT_MAP_HOLE ||
3125 822163 : (cur_offset >= inode->i_size &&
3126 5062 : !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
3127 312240 : const u64 range_len = last_byte - cur_offset;
3128 :
3129 312240 : ret = add_falloc_range(&reserve_list, cur_offset, range_len);
3130 312240 : if (ret < 0) {
3131 0 : free_extent_map(em);
3132 0 : break;
3133 : }
3134 312240 : ret = btrfs_qgroup_reserve_data(BTRFS_I(inode),
3135 : &data_reserved, cur_offset, range_len);
3136 312240 : if (ret < 0) {
3137 139 : free_extent_map(em);
3138 139 : break;
3139 : }
3140 312101 : qgroup_reserved += range_len;
3141 312101 : data_space_needed += range_len;
3142 : }
3143 1129202 : free_extent_map(em);
3144 1129202 : cur_offset = last_byte;
3145 : }
3146 :
3147 509011 : if (!ret && data_space_needed > 0) {
3148 : /*
3149 : * We are safe to reserve space here as we can't have delalloc
3150 : * in the range, see above.
3151 : */
3152 201538 : ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
3153 : data_space_needed);
3154 201536 : if (!ret)
3155 196290 : data_space_reserved = data_space_needed;
3156 : }
3157 :
3158 : /*
3159 : * If ret is still 0, means we're OK to fallocate.
3160 : * Or just cleanup the list and exit.
3161 : */
3162 757962 : list_for_each_entry_safe(range, tmp, &reserve_list, list) {
3163 248951 : if (!ret) {
3164 232960 : ret = btrfs_prealloc_file_range(inode, mode,
3165 : range->start,
3166 : range->len, i_blocksize(inode),
3167 : offset + len, &alloc_hint);
3168 : /*
3169 : * btrfs_prealloc_file_range() releases space even
3170 : * if it returns an error.
3171 : */
3172 232960 : data_space_reserved -= range->len;
3173 232960 : qgroup_reserved -= range->len;
3174 15991 : } else if (data_space_reserved > 0) {
3175 0 : btrfs_free_reserved_data_space(BTRFS_I(inode),
3176 : data_reserved, range->start,
3177 : range->len);
3178 0 : data_space_reserved -= range->len;
3179 0 : qgroup_reserved -= range->len;
3180 15991 : } else if (qgroup_reserved > 0) {
3181 15865 : btrfs_qgroup_free_data(BTRFS_I(inode), data_reserved,
3182 : range->start, range->len);
3183 15864 : qgroup_reserved -= range->len;
3184 : }
3185 248950 : list_del(&range->list);
3186 248951 : kfree(range);
3187 : }
3188 509011 : if (ret < 0)
3189 5392 : goto out_unlock;
3190 :
3191 : /*
3192 : * We didn't need to allocate any more space, but we still extended the
3193 : * size of the file so we need to update i_size and the inode item.
3194 : */
3195 503619 : ret = btrfs_fallocate_update_isize(inode, actual_end, mode);
3196 509011 : out_unlock:
3197 509011 : unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
3198 : &cached_state);
3199 509443 : out:
3200 509443 : btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
3201 509442 : extent_changeset_free(data_reserved);
3202 509442 : return ret;
3203 : }
3204 :
3205 : /*
3206 : * Helper for btrfs_find_delalloc_in_range(). Find a subrange in a given range
3207 : * that has unflushed and/or flushing delalloc. There might be other adjacent
3208 : * subranges after the one it found, so btrfs_find_delalloc_in_range() keeps
3209 : * looping while it gets adjacent subranges, and merging them together.
3210 : */
3211 590666 : static bool find_delalloc_subrange(struct btrfs_inode *inode, u64 start, u64 end,
3212 : struct extent_state **cached_state,
3213 : bool *search_io_tree,
3214 : u64 *delalloc_start_ret, u64 *delalloc_end_ret)
3215 : {
3216 590666 : u64 len = end + 1 - start;
3217 590666 : u64 delalloc_len = 0;
3218 590666 : struct btrfs_ordered_extent *oe;
3219 590666 : u64 oe_start;
3220 590666 : u64 oe_end;
3221 :
3222 : /*
3223 : * Search the io tree first for EXTENT_DELALLOC. If we find any, it
3224 : * means we have delalloc (dirty pages) for which writeback has not
3225 : * started yet.
3226 : */
3227 590666 : if (*search_io_tree) {
3228 587137 : spin_lock(&inode->lock);
3229 587190 : if (inode->delalloc_bytes > 0) {
3230 222493 : spin_unlock(&inode->lock);
3231 222493 : *delalloc_start_ret = start;
3232 222493 : delalloc_len = count_range_bits(&inode->io_tree,
3233 : delalloc_start_ret, end,
3234 : len, EXTENT_DELALLOC, 1,
3235 : cached_state);
3236 : } else {
3237 364697 : spin_unlock(&inode->lock);
3238 : }
3239 : }
3240 :
3241 587190 : if (delalloc_len > 0) {
3242 : /*
3243 : * If delalloc was found then *delalloc_start_ret has a sector size
3244 : * aligned value (rounded down).
3245 : */
3246 37740 : *delalloc_end_ret = *delalloc_start_ret + delalloc_len - 1;
3247 :
3248 37740 : if (*delalloc_start_ret == start) {
3249 : /* Delalloc for the whole range, nothing more to do. */
3250 33053 : if (*delalloc_end_ret == end)
3251 : return true;
3252 : /* Else trim our search range for ordered extents. */
3253 9078 : start = *delalloc_end_ret + 1;
3254 9078 : len = end + 1 - start;
3255 : }
3256 : } else {
3257 : /* No delalloc, future calls don't need to search again. */
3258 552984 : *search_io_tree = false;
3259 : }
3260 :
3261 : /*
3262 : * Now also check if there's any ordered extent in the range.
3263 : * We do this because:
3264 : *
3265 : * 1) When delalloc is flushed, the file range is locked, we clear the
3266 : * EXTENT_DELALLOC bit from the io tree and create an extent map and
3267 : * an ordered extent for the write. So we might just have been called
3268 : * after delalloc is flushed and before the ordered extent completes
3269 : * and inserts the new file extent item in the subvolume's btree;
3270 : *
3271 : * 2) We may have an ordered extent created by flushing delalloc for a
3272 : * subrange that starts before the subrange we found marked with
3273 : * EXTENT_DELALLOC in the io tree.
3274 : *
3275 : * We could also use the extent map tree to find such delalloc that is
3276 : * being flushed, but using the ordered extents tree is more efficient
3277 : * because it's usually much smaller as ordered extents are removed from
3278 : * the tree once they complete. With the extent maps, we mau have them
3279 : * in the extent map tree for a very long time, and they were either
3280 : * created by previous writes or loaded by read operations.
3281 : */
3282 566749 : oe = btrfs_lookup_first_ordered_range(inode, start, len);
3283 566747 : if (!oe)
3284 550834 : return (delalloc_len > 0);
3285 :
3286 : /* The ordered extent may span beyond our search range. */
3287 15913 : oe_start = max(oe->file_offset, start);
3288 15913 : oe_end = min(oe->file_offset + oe->num_bytes - 1, end);
3289 :
3290 15913 : btrfs_put_ordered_extent(oe);
3291 :
3292 : /* Don't have unflushed delalloc, return the ordered extent range. */
3293 15913 : if (delalloc_len == 0) {
3294 11531 : *delalloc_start_ret = oe_start;
3295 11531 : *delalloc_end_ret = oe_end;
3296 11531 : return true;
3297 : }
3298 :
3299 : /*
3300 : * We have both unflushed delalloc (io_tree) and an ordered extent.
3301 : * If the ranges are adjacent returned a combined range, otherwise
3302 : * return the leftmost range.
3303 : */
3304 4382 : if (oe_start < *delalloc_start_ret) {
3305 4378 : if (oe_end < *delalloc_start_ret)
3306 4378 : *delalloc_end_ret = oe_end;
3307 4378 : *delalloc_start_ret = oe_start;
3308 4 : } else if (*delalloc_end_ret + 1 == oe_start) {
3309 0 : *delalloc_end_ret = oe_end;
3310 : }
3311 :
3312 : return true;
3313 : }
3314 :
3315 : /*
3316 : * Check if there's delalloc in a given range.
3317 : *
3318 : * @inode: The inode.
3319 : * @start: The start offset of the range. It does not need to be
3320 : * sector size aligned.
3321 : * @end: The end offset (inclusive value) of the search range.
3322 : * It does not need to be sector size aligned.
3323 : * @cached_state: Extent state record used for speeding up delalloc
3324 : * searches in the inode's io_tree. Can be NULL.
3325 : * @delalloc_start_ret: Output argument, set to the start offset of the
3326 : * subrange found with delalloc (may not be sector size
3327 : * aligned).
3328 : * @delalloc_end_ret: Output argument, set to he end offset (inclusive value)
3329 : * of the subrange found with delalloc.
3330 : *
3331 : * Returns true if a subrange with delalloc is found within the given range, and
3332 : * if so it sets @delalloc_start_ret and @delalloc_end_ret with the start and
3333 : * end offsets of the subrange.
3334 : */
3335 573539 : bool btrfs_find_delalloc_in_range(struct btrfs_inode *inode, u64 start, u64 end,
3336 : struct extent_state **cached_state,
3337 : u64 *delalloc_start_ret, u64 *delalloc_end_ret)
3338 : {
3339 573539 : u64 cur_offset = round_down(start, inode->root->fs_info->sectorsize);
3340 573539 : u64 prev_delalloc_end = 0;
3341 573539 : bool search_io_tree = true;
3342 573539 : bool ret = false;
3343 :
3344 622709 : while (cur_offset <= end) {
3345 590701 : u64 delalloc_start;
3346 590701 : u64 delalloc_end;
3347 590701 : bool delalloc;
3348 :
3349 590701 : delalloc = find_delalloc_subrange(inode, cur_offset, end,
3350 : cached_state, &search_io_tree,
3351 : &delalloc_start,
3352 : &delalloc_end);
3353 590721 : if (!delalloc)
3354 : break;
3355 :
3356 49271 : if (prev_delalloc_end == 0) {
3357 : /* First subrange found. */
3358 41276 : *delalloc_start_ret = max(delalloc_start, start);
3359 41276 : *delalloc_end_ret = delalloc_end;
3360 41276 : ret = true;
3361 7995 : } else if (delalloc_start == prev_delalloc_end + 1) {
3362 : /* Subrange adjacent to the previous one, merge them. */
3363 7894 : *delalloc_end_ret = delalloc_end;
3364 : } else {
3365 : /* Subrange not adjacent to the previous one, exit. */
3366 : break;
3367 : }
3368 :
3369 49170 : prev_delalloc_end = delalloc_end;
3370 49170 : cur_offset = delalloc_end + 1;
3371 49170 : cond_resched();
3372 : }
3373 :
3374 573559 : return ret;
3375 : }
3376 :
3377 : /*
3378 : * Check if there's a hole or delalloc range in a range representing a hole (or
3379 : * prealloc extent) found in the inode's subvolume btree.
3380 : *
3381 : * @inode: The inode.
3382 : * @whence: Seek mode (SEEK_DATA or SEEK_HOLE).
3383 : * @start: Start offset of the hole region. It does not need to be sector
3384 : * size aligned.
3385 : * @end: End offset (inclusive value) of the hole region. It does not
3386 : * need to be sector size aligned.
3387 : * @start_ret: Return parameter, used to set the start of the subrange in the
3388 : * hole that matches the search criteria (seek mode), if such
3389 : * subrange is found (return value of the function is true).
3390 : * The value returned here may not be sector size aligned.
3391 : *
3392 : * Returns true if a subrange matching the given seek mode is found, and if one
3393 : * is found, it updates @start_ret with the start of the subrange.
3394 : */
3395 500 : static bool find_desired_extent_in_hole(struct btrfs_inode *inode, int whence,
3396 : struct extent_state **cached_state,
3397 : u64 start, u64 end, u64 *start_ret)
3398 : {
3399 500 : u64 delalloc_start;
3400 500 : u64 delalloc_end;
3401 500 : bool delalloc;
3402 :
3403 500 : delalloc = btrfs_find_delalloc_in_range(inode, start, end, cached_state,
3404 : &delalloc_start, &delalloc_end);
3405 500 : if (delalloc && whence == SEEK_DATA) {
3406 315 : *start_ret = delalloc_start;
3407 315 : return true;
3408 : }
3409 :
3410 185 : if (delalloc && whence == SEEK_HOLE) {
3411 : /*
3412 : * We found delalloc but it starts after out start offset. So we
3413 : * have a hole between our start offset and the delalloc start.
3414 : */
3415 106 : if (start < delalloc_start) {
3416 15 : *start_ret = start;
3417 15 : return true;
3418 : }
3419 : /*
3420 : * Delalloc range starts at our start offset.
3421 : * If the delalloc range's length is smaller than our range,
3422 : * then it means we have a hole that starts where the delalloc
3423 : * subrange ends.
3424 : */
3425 91 : if (delalloc_end < end) {
3426 72 : *start_ret = delalloc_end + 1;
3427 72 : return true;
3428 : }
3429 :
3430 : /* There's delalloc for the whole range. */
3431 : return false;
3432 : }
3433 :
3434 79 : if (!delalloc && whence == SEEK_HOLE) {
3435 6 : *start_ret = start;
3436 6 : return true;
3437 : }
3438 :
3439 : /*
3440 : * No delalloc in the range and we are seeking for data. The caller has
3441 : * to iterate to the next extent item in the subvolume btree.
3442 : */
3443 : return false;
3444 : }
3445 :
3446 474 : static loff_t find_desired_extent(struct file *file, loff_t offset, int whence)
3447 : {
3448 474 : struct btrfs_inode *inode = BTRFS_I(file->f_mapping->host);
3449 474 : struct btrfs_file_private *private = file->private_data;
3450 474 : struct btrfs_fs_info *fs_info = inode->root->fs_info;
3451 474 : struct extent_state *cached_state = NULL;
3452 474 : struct extent_state **delalloc_cached_state;
3453 474 : const loff_t i_size = i_size_read(&inode->vfs_inode);
3454 474 : const u64 ino = btrfs_ino(inode);
3455 474 : struct btrfs_root *root = inode->root;
3456 474 : struct btrfs_path *path;
3457 474 : struct btrfs_key key;
3458 474 : u64 last_extent_end;
3459 474 : u64 lockstart;
3460 474 : u64 lockend;
3461 474 : u64 start;
3462 474 : int ret;
3463 474 : bool found = false;
3464 :
3465 474 : if (i_size == 0 || offset >= i_size)
3466 : return -ENXIO;
3467 :
3468 : /*
3469 : * Quick path. If the inode has no prealloc extents and its number of
3470 : * bytes used matches its i_size, then it can not have holes.
3471 : */
3472 447 : if (whence == SEEK_HOLE &&
3473 185 : !(inode->flags & BTRFS_INODE_PREALLOC) &&
3474 76 : inode_get_bytes(&inode->vfs_inode) == i_size)
3475 : return i_size;
3476 :
3477 447 : if (!private) {
3478 39 : private = kzalloc(sizeof(*private), GFP_KERNEL);
3479 : /*
3480 : * No worries if memory allocation failed.
3481 : * The private structure is used only for speeding up multiple
3482 : * lseek SEEK_HOLE/DATA calls to a file when there's delalloc,
3483 : * so everything will still be correct.
3484 : */
3485 39 : file->private_data = private;
3486 : }
3487 :
3488 447 : if (private)
3489 447 : delalloc_cached_state = &private->llseek_cached_state;
3490 : else
3491 : delalloc_cached_state = NULL;
3492 :
3493 : /*
3494 : * offset can be negative, in this case we start finding DATA/HOLE from
3495 : * the very start of the file.
3496 : */
3497 447 : start = max_t(loff_t, 0, offset);
3498 :
3499 447 : lockstart = round_down(start, fs_info->sectorsize);
3500 447 : lockend = round_up(i_size, fs_info->sectorsize);
3501 447 : if (lockend <= lockstart)
3502 0 : lockend = lockstart + fs_info->sectorsize;
3503 447 : lockend--;
3504 :
3505 447 : path = btrfs_alloc_path();
3506 447 : if (!path)
3507 : return -ENOMEM;
3508 447 : path->reada = READA_FORWARD;
3509 :
3510 447 : key.objectid = ino;
3511 447 : key.type = BTRFS_EXTENT_DATA_KEY;
3512 447 : key.offset = start;
3513 :
3514 447 : last_extent_end = lockstart;
3515 :
3516 447 : lock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
3517 :
3518 447 : ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3519 447 : if (ret < 0) {
3520 0 : goto out;
3521 447 : } else if (ret > 0 && path->slots[0] > 0) {
3522 397 : btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
3523 397 : if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY)
3524 33 : path->slots[0]--;
3525 : }
3526 :
3527 501 : while (start < i_size) {
3528 484 : struct extent_buffer *leaf = path->nodes[0];
3529 484 : struct btrfs_file_extent_item *extent;
3530 484 : u64 extent_end;
3531 484 : u8 type;
3532 :
3533 484 : if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3534 317 : ret = btrfs_next_leaf(root, path);
3535 317 : if (ret < 0)
3536 0 : goto out;
3537 317 : else if (ret > 0)
3538 : break;
3539 :
3540 0 : leaf = path->nodes[0];
3541 : }
3542 :
3543 167 : btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3544 167 : if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
3545 : break;
3546 :
3547 122 : extent_end = btrfs_file_extent_end(path);
3548 :
3549 : /*
3550 : * In the first iteration we may have a slot that points to an
3551 : * extent that ends before our start offset, so skip it.
3552 : */
3553 122 : if (extent_end <= start) {
3554 1 : path->slots[0]++;
3555 1 : continue;
3556 : }
3557 :
3558 : /* We have an implicit hole, NO_HOLES feature is likely set. */
3559 121 : if (last_extent_end < key.offset) {
3560 21 : u64 search_start = last_extent_end;
3561 21 : u64 found_start;
3562 :
3563 : /*
3564 : * First iteration, @start matches @offset and it's
3565 : * within the hole.
3566 : */
3567 21 : if (start == offset)
3568 6 : search_start = offset;
3569 :
3570 21 : found = find_desired_extent_in_hole(inode, whence,
3571 : delalloc_cached_state,
3572 : search_start,
3573 : key.offset - 1,
3574 : &found_start);
3575 21 : if (found) {
3576 2 : start = found_start;
3577 2 : break;
3578 : }
3579 : /*
3580 : * Didn't find data or a hole (due to delalloc) in the
3581 : * implicit hole range, so need to analyze the extent.
3582 : */
3583 : }
3584 :
3585 119 : extent = btrfs_item_ptr(leaf, path->slots[0],
3586 : struct btrfs_file_extent_item);
3587 119 : type = btrfs_file_extent_type(leaf, extent);
3588 :
3589 : /*
3590 : * Can't access the extent's disk_bytenr field if this is an
3591 : * inline extent, since at that offset, it's where the extent
3592 : * data starts.
3593 : */
3594 119 : if (type == BTRFS_FILE_EXTENT_PREALLOC ||
3595 1 : (type == BTRFS_FILE_EXTENT_REG &&
3596 53 : btrfs_file_extent_disk_bytenr(leaf, extent) == 0)) {
3597 : /*
3598 : * Explicit hole or prealloc extent, search for delalloc.
3599 : * A prealloc extent is treated like a hole.
3600 : */
3601 117 : u64 search_start = key.offset;
3602 117 : u64 found_start;
3603 :
3604 : /*
3605 : * First iteration, @start matches @offset and it's
3606 : * within the hole.
3607 : */
3608 117 : if (start == offset)
3609 84 : search_start = offset;
3610 :
3611 117 : found = find_desired_extent_in_hole(inode, whence,
3612 : delalloc_cached_state,
3613 : search_start,
3614 : extent_end - 1,
3615 : &found_start);
3616 117 : if (found) {
3617 64 : start = found_start;
3618 64 : break;
3619 : }
3620 : /*
3621 : * Didn't find data or a hole (due to delalloc) in the
3622 : * implicit hole range, so need to analyze the next
3623 : * extent item.
3624 : */
3625 : } else {
3626 : /*
3627 : * Found a regular or inline extent.
3628 : * If we are seeking for data, adjust the start offset
3629 : * and stop, we're done.
3630 : */
3631 2 : if (whence == SEEK_DATA) {
3632 2 : start = max_t(u64, key.offset, offset);
3633 2 : found = true;
3634 2 : break;
3635 : }
3636 : /*
3637 : * Else, we are seeking for a hole, check the next file
3638 : * extent item.
3639 : */
3640 : }
3641 :
3642 53 : start = extent_end;
3643 53 : last_extent_end = extent_end;
3644 53 : path->slots[0]++;
3645 53 : if (fatal_signal_pending(current)) {
3646 0 : ret = -EINTR;
3647 0 : goto out;
3648 : }
3649 53 : cond_resched();
3650 : }
3651 :
3652 : /* We have an implicit hole from the last extent found up to i_size. */
3653 447 : if (!found && start < i_size) {
3654 362 : found = find_desired_extent_in_hole(inode, whence,
3655 : delalloc_cached_state, start,
3656 : i_size - 1, &start);
3657 362 : if (!found)
3658 20 : start = i_size;
3659 : }
3660 :
3661 427 : out:
3662 447 : unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
3663 447 : btrfs_free_path(path);
3664 :
3665 447 : if (ret < 0)
3666 0 : return ret;
3667 :
3668 447 : if (whence == SEEK_DATA && start >= i_size)
3669 : return -ENXIO;
3670 :
3671 426 : return min_t(loff_t, start, i_size);
3672 : }
3673 :
3674 4690320 : static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
3675 : {
3676 4690320 : struct inode *inode = file->f_mapping->host;
3677 :
3678 4690320 : switch (whence) {
3679 4689846 : default:
3680 4689846 : return generic_file_llseek(file, offset, whence);
3681 : case SEEK_DATA:
3682 : case SEEK_HOLE:
3683 474 : btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_SHARED);
3684 474 : offset = find_desired_extent(file, offset, whence);
3685 474 : btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_SHARED);
3686 474 : break;
3687 : }
3688 :
3689 474 : if (offset < 0)
3690 : return offset;
3691 :
3692 426 : return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
3693 : }
3694 :
3695 7421540 : static int btrfs_file_open(struct inode *inode, struct file *filp)
3696 : {
3697 7421540 : int ret;
3698 :
3699 7421540 : filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC | FMODE_BUF_WASYNC |
3700 : FMODE_CAN_ODIRECT;
3701 :
3702 7421540 : ret = fsverity_file_open(inode, filp);
3703 7421540 : if (ret)
3704 : return ret;
3705 7421540 : return generic_file_open(inode, filp);
3706 : }
3707 :
3708 3060482 : static int check_direct_read(struct btrfs_fs_info *fs_info,
3709 : const struct iov_iter *iter, loff_t offset)
3710 : {
3711 3060482 : int ret;
3712 3060482 : int i, seg;
3713 :
3714 3060482 : ret = check_direct_IO(fs_info, iter, offset);
3715 3061163 : if (ret < 0)
3716 : return ret;
3717 :
3718 1074718 : if (!iter_is_iovec(iter))
3719 : return 0;
3720 :
3721 0 : for (seg = 0; seg < iter->nr_segs; seg++) {
3722 0 : for (i = seg + 1; i < iter->nr_segs; i++) {
3723 0 : const struct iovec *iov1 = iter_iov(iter) + seg;
3724 0 : const struct iovec *iov2 = iter_iov(iter) + i;
3725 :
3726 0 : if (iov1->iov_base == iov2->iov_base)
3727 : return -EINVAL;
3728 : }
3729 : }
3730 : return 0;
3731 : }
3732 :
3733 3060599 : static ssize_t btrfs_direct_read(struct kiocb *iocb, struct iov_iter *to)
3734 : {
3735 3060599 : struct inode *inode = file_inode(iocb->ki_filp);
3736 3060599 : size_t prev_left = 0;
3737 3060599 : ssize_t read = 0;
3738 3060599 : ssize_t ret;
3739 :
3740 3060599 : if (fsverity_active(inode))
3741 : return 0;
3742 :
3743 3060599 : if (check_direct_read(btrfs_sb(inode->i_sb), to, iocb->ki_pos))
3744 : return 0;
3745 :
3746 1074730 : btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_SHARED);
3747 1076776 : again:
3748 : /*
3749 : * This is similar to what we do for direct IO writes, see the comment
3750 : * at btrfs_direct_write(), but we also disable page faults in addition
3751 : * to disabling them only at the iov_iter level. This is because when
3752 : * reading from a hole or prealloc extent, iomap calls iov_iter_zero(),
3753 : * which can still trigger page fault ins despite having set ->nofault
3754 : * to true of our 'to' iov_iter.
3755 : *
3756 : * The difference to direct IO writes is that we deadlock when trying
3757 : * to lock the extent range in the inode's tree during he page reads
3758 : * triggered by the fault in (while for writes it is due to waiting for
3759 : * our own ordered extent). This is because for direct IO reads,
3760 : * btrfs_dio_iomap_begin() returns with the extent range locked, which
3761 : * is only unlocked in the endio callback (end_bio_extent_readpage()).
3762 : */
3763 1076776 : pagefault_disable();
3764 1076764 : to->nofault = true;
3765 1076764 : ret = btrfs_dio_read(iocb, to, read);
3766 1077116 : to->nofault = false;
3767 1077116 : pagefault_enable();
3768 :
3769 : /* No increment (+=) because iomap returns a cumulative value. */
3770 1077079 : if (ret > 0)
3771 172999 : read = ret;
3772 :
3773 1077079 : if (iov_iter_count(to) > 0 && (ret == -EFAULT || ret > 0)) {
3774 2135 : const size_t left = iov_iter_count(to);
3775 :
3776 2135 : if (left == prev_left) {
3777 : /*
3778 : * We didn't make any progress since the last attempt,
3779 : * fallback to a buffered read for the remainder of the
3780 : * range. This is just to avoid any possibility of looping
3781 : * for too long.
3782 : */
3783 : ret = read;
3784 : } else {
3785 : /*
3786 : * We made some progress since the last retry or this is
3787 : * the first time we are retrying. Fault in as many pages
3788 : * as possible and retry.
3789 : */
3790 2135 : fault_in_iov_iter_writeable(to, left);
3791 2107 : prev_left = left;
3792 2107 : goto again;
3793 : }
3794 : }
3795 1074944 : btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_SHARED);
3796 1074995 : return ret < 0 ? ret : read;
3797 : }
3798 :
3799 25375696 : static ssize_t btrfs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
3800 : {
3801 25375696 : ssize_t ret = 0;
3802 :
3803 25375696 : if (iocb->ki_flags & IOCB_DIRECT) {
3804 3061225 : ret = btrfs_direct_read(iocb, to);
3805 3061406 : if (ret < 0 || !iov_iter_count(to) ||
3806 2755234 : iocb->ki_pos >= i_size_read(file_inode(iocb->ki_filp)))
3807 : return ret;
3808 : }
3809 :
3810 23023801 : return filemap_read(iocb, to, ret);
3811 : }
3812 :
3813 : const struct file_operations btrfs_file_operations = {
3814 : .llseek = btrfs_file_llseek,
3815 : .read_iter = btrfs_file_read_iter,
3816 : .splice_read = filemap_splice_read,
3817 : .write_iter = btrfs_file_write_iter,
3818 : .splice_write = iter_file_splice_write,
3819 : .mmap = btrfs_file_mmap,
3820 : .open = btrfs_file_open,
3821 : .release = btrfs_release_file,
3822 : .get_unmapped_area = thp_get_unmapped_area,
3823 : .fsync = btrfs_sync_file,
3824 : .fallocate = btrfs_fallocate,
3825 : .unlocked_ioctl = btrfs_ioctl,
3826 : #ifdef CONFIG_COMPAT
3827 : .compat_ioctl = btrfs_compat_ioctl,
3828 : #endif
3829 : .remap_file_range = btrfs_remap_file_range,
3830 : };
3831 :
3832 9472792 : int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end)
3833 : {
3834 9472792 : int ret;
3835 :
3836 : /*
3837 : * So with compression we will find and lock a dirty page and clear the
3838 : * first one as dirty, setup an async extent, and immediately return
3839 : * with the entire range locked but with nobody actually marked with
3840 : * writeback. So we can't just filemap_write_and_wait_range() and
3841 : * expect it to work since it will just kick off a thread to do the
3842 : * actual work. So we need to call filemap_fdatawrite_range _again_
3843 : * since it will wait on the page lock, which won't be unlocked until
3844 : * after the pages have been marked as writeback and so we're good to go
3845 : * from there. We have to do this otherwise we'll miss the ordered
3846 : * extents and that results in badness. Please Josef, do not think you
3847 : * know better and pull this out at some point in the future, it is
3848 : * right and you are wrong.
3849 : */
3850 9472792 : ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
3851 9473545 : if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
3852 : &BTRFS_I(inode)->runtime_flags))
3853 23986 : ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
3854 :
3855 9473545 : return ret;
3856 : }
|
