Loading...
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 "direct-io.h"
22#include "disk-io.h"
23#include "transaction.h"
24#include "btrfs_inode.h"
25#include "tree-log.h"
26#include "locking.h"
27#include "qgroup.h"
28#include "compression.h"
29#include "delalloc-space.h"
30#include "reflink.h"
31#include "subpage.h"
32#include "fs.h"
33#include "accessors.h"
34#include "extent-tree.h"
35#include "file-item.h"
36#include "ioctl.h"
37#include "file.h"
38#include "super.h"
39
40/*
41 * Helper to fault in page and copy. This should go away and be replaced with
42 * calls into generic code.
43 */
44static noinline int btrfs_copy_from_user(loff_t pos, size_t write_bytes,
45 struct folio *folio, struct iov_iter *i)
46{
47 size_t copied = 0;
48 size_t total_copied = 0;
49 int offset = offset_in_page(pos);
50
51 while (write_bytes > 0) {
52 size_t count = min_t(size_t, PAGE_SIZE - offset, write_bytes);
53 /*
54 * Copy data from userspace to the current page
55 */
56 copied = copy_folio_from_iter_atomic(folio, offset, count, i);
57
58 /* Flush processor's dcache for this page */
59 flush_dcache_folio(folio);
60
61 /*
62 * if we get a partial write, we can end up with
63 * partially up to date page. These add
64 * a lot of complexity, so make sure they don't
65 * happen by forcing this copy to be retried.
66 *
67 * The rest of the btrfs_file_write code will fall
68 * back to page at a time copies after we return 0.
69 */
70 if (unlikely(copied < count)) {
71 if (!folio_test_uptodate(folio)) {
72 iov_iter_revert(i, copied);
73 copied = 0;
74 }
75 if (!copied)
76 break;
77 }
78
79 write_bytes -= copied;
80 total_copied += copied;
81 offset += copied;
82 }
83 return total_copied;
84}
85
86/*
87 * Unlock folio after btrfs_file_write() is done with it.
88 */
89static void btrfs_drop_folio(struct btrfs_fs_info *fs_info, struct folio *folio,
90 u64 pos, u64 copied)
91{
92 u64 block_start = round_down(pos, fs_info->sectorsize);
93 u64 block_len = round_up(pos + copied, fs_info->sectorsize) - block_start;
94
95 ASSERT(block_len <= U32_MAX);
96 /*
97 * Folio checked is some magic around finding folios that have been
98 * modified without going through btrfs_dirty_folio(). Clear it here.
99 * There should be no need to mark the pages accessed as
100 * prepare_one_folio() should have marked them accessed in
101 * prepare_one_folio() via find_or_create_page()
102 */
103 btrfs_folio_clamp_clear_checked(fs_info, folio, block_start, block_len);
104 folio_unlock(folio);
105 folio_put(folio);
106}
107
108/*
109 * After btrfs_copy_from_user(), update the following things for delalloc:
110 * - Mark newly dirtied folio as DELALLOC in the io tree.
111 * Used to advise which range is to be written back.
112 * - Mark modified folio as Uptodate/Dirty and not needing COW fixup
113 * - Update inode size for past EOF write
114 */
115int btrfs_dirty_folio(struct btrfs_inode *inode, struct folio *folio, loff_t pos,
116 size_t write_bytes, struct extent_state **cached, bool noreserve)
117{
118 struct btrfs_fs_info *fs_info = inode->root->fs_info;
119 int ret = 0;
120 u64 num_bytes;
121 u64 start_pos;
122 u64 end_of_last_block;
123 u64 end_pos = pos + write_bytes;
124 loff_t isize = i_size_read(&inode->vfs_inode);
125 unsigned int extra_bits = 0;
126
127 if (write_bytes == 0)
128 return 0;
129
130 if (noreserve)
131 extra_bits |= EXTENT_NORESERVE;
132
133 start_pos = round_down(pos, fs_info->sectorsize);
134 num_bytes = round_up(write_bytes + pos - start_pos,
135 fs_info->sectorsize);
136 ASSERT(num_bytes <= U32_MAX);
137 ASSERT(folio_pos(folio) <= pos &&
138 folio_pos(folio) + folio_size(folio) >= pos + write_bytes);
139
140 end_of_last_block = start_pos + num_bytes - 1;
141
142 /*
143 * The pages may have already been dirty, clear out old accounting so
144 * we can set things up properly
145 */
146 clear_extent_bit(&inode->io_tree, start_pos, end_of_last_block,
147 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
148 cached);
149
150 ret = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
151 extra_bits, cached);
152 if (ret)
153 return ret;
154
155 btrfs_folio_clamp_set_uptodate(fs_info, folio, start_pos, num_bytes);
156 btrfs_folio_clamp_clear_checked(fs_info, folio, start_pos, num_bytes);
157 btrfs_folio_clamp_set_dirty(fs_info, folio, start_pos, num_bytes);
158
159 /*
160 * we've only changed i_size in ram, and we haven't updated
161 * the disk i_size. There is no need to log the inode
162 * at this time.
163 */
164 if (end_pos > isize)
165 i_size_write(&inode->vfs_inode, end_pos);
166 return 0;
167}
168
169/*
170 * this is very complex, but the basic idea is to drop all extents
171 * in the range start - end. hint_block is filled in with a block number
172 * that would be a good hint to the block allocator for this file.
173 *
174 * If an extent intersects the range but is not entirely inside the range
175 * it is either truncated or split. Anything entirely inside the range
176 * is deleted from the tree.
177 *
178 * Note: the VFS' inode number of bytes is not updated, it's up to the caller
179 * to deal with that. We set the field 'bytes_found' of the arguments structure
180 * with the number of allocated bytes found in the target range, so that the
181 * caller can update the inode's number of bytes in an atomic way when
182 * replacing extents in a range to avoid races with stat(2).
183 */
184int btrfs_drop_extents(struct btrfs_trans_handle *trans,
185 struct btrfs_root *root, struct btrfs_inode *inode,
186 struct btrfs_drop_extents_args *args)
187{
188 struct btrfs_fs_info *fs_info = root->fs_info;
189 struct extent_buffer *leaf;
190 struct btrfs_file_extent_item *fi;
191 struct btrfs_key key;
192 struct btrfs_key new_key;
193 u64 ino = btrfs_ino(inode);
194 u64 search_start = args->start;
195 u64 disk_bytenr = 0;
196 u64 num_bytes = 0;
197 u64 extent_offset = 0;
198 u64 extent_end = 0;
199 u64 last_end = args->start;
200 int del_nr = 0;
201 int del_slot = 0;
202 int extent_type;
203 int recow;
204 int ret;
205 int modify_tree = -1;
206 int update_refs;
207 int found = 0;
208 struct btrfs_path *path = args->path;
209
210 args->bytes_found = 0;
211 args->extent_inserted = false;
212
213 /* Must always have a path if ->replace_extent is true */
214 ASSERT(!(args->replace_extent && !args->path));
215
216 if (!path) {
217 path = btrfs_alloc_path();
218 if (!path) {
219 ret = -ENOMEM;
220 goto out;
221 }
222 }
223
224 if (args->drop_cache)
225 btrfs_drop_extent_map_range(inode, args->start, args->end - 1, false);
226
227 if (data_race(args->start >= inode->disk_i_size) && !args->replace_extent)
228 modify_tree = 0;
229
230 update_refs = (btrfs_root_id(root) != BTRFS_TREE_LOG_OBJECTID);
231 while (1) {
232 recow = 0;
233 ret = btrfs_lookup_file_extent(trans, root, path, ino,
234 search_start, modify_tree);
235 if (ret < 0)
236 break;
237 if (ret > 0 && path->slots[0] > 0 && search_start == args->start) {
238 leaf = path->nodes[0];
239 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
240 if (key.objectid == ino &&
241 key.type == BTRFS_EXTENT_DATA_KEY)
242 path->slots[0]--;
243 }
244 ret = 0;
245next_slot:
246 leaf = path->nodes[0];
247 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
248 BUG_ON(del_nr > 0);
249 ret = btrfs_next_leaf(root, path);
250 if (ret < 0)
251 break;
252 if (ret > 0) {
253 ret = 0;
254 break;
255 }
256 leaf = path->nodes[0];
257 recow = 1;
258 }
259
260 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
261
262 if (key.objectid > ino)
263 break;
264 if (WARN_ON_ONCE(key.objectid < ino) ||
265 key.type < BTRFS_EXTENT_DATA_KEY) {
266 ASSERT(del_nr == 0);
267 path->slots[0]++;
268 goto next_slot;
269 }
270 if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= args->end)
271 break;
272
273 fi = btrfs_item_ptr(leaf, path->slots[0],
274 struct btrfs_file_extent_item);
275 extent_type = btrfs_file_extent_type(leaf, fi);
276
277 if (extent_type == BTRFS_FILE_EXTENT_REG ||
278 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
279 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
280 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
281 extent_offset = btrfs_file_extent_offset(leaf, fi);
282 extent_end = key.offset +
283 btrfs_file_extent_num_bytes(leaf, fi);
284 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
285 extent_end = key.offset +
286 btrfs_file_extent_ram_bytes(leaf, fi);
287 } else {
288 /* can't happen */
289 BUG();
290 }
291
292 /*
293 * Don't skip extent items representing 0 byte lengths. They
294 * used to be created (bug) if while punching holes we hit
295 * -ENOSPC condition. So if we find one here, just ensure we
296 * delete it, otherwise we would insert a new file extent item
297 * with the same key (offset) as that 0 bytes length file
298 * extent item in the call to setup_items_for_insert() later
299 * in this function.
300 */
301 if (extent_end == key.offset && extent_end >= search_start) {
302 last_end = extent_end;
303 goto delete_extent_item;
304 }
305
306 if (extent_end <= search_start) {
307 path->slots[0]++;
308 goto next_slot;
309 }
310
311 found = 1;
312 search_start = max(key.offset, args->start);
313 if (recow || !modify_tree) {
314 modify_tree = -1;
315 btrfs_release_path(path);
316 continue;
317 }
318
319 /*
320 * | - range to drop - |
321 * | -------- extent -------- |
322 */
323 if (args->start > key.offset && args->end < extent_end) {
324 BUG_ON(del_nr > 0);
325 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
326 ret = -EOPNOTSUPP;
327 break;
328 }
329
330 memcpy(&new_key, &key, sizeof(new_key));
331 new_key.offset = args->start;
332 ret = btrfs_duplicate_item(trans, root, path,
333 &new_key);
334 if (ret == -EAGAIN) {
335 btrfs_release_path(path);
336 continue;
337 }
338 if (ret < 0)
339 break;
340
341 leaf = path->nodes[0];
342 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
343 struct btrfs_file_extent_item);
344 btrfs_set_file_extent_num_bytes(leaf, fi,
345 args->start - key.offset);
346
347 fi = btrfs_item_ptr(leaf, path->slots[0],
348 struct btrfs_file_extent_item);
349
350 extent_offset += args->start - key.offset;
351 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
352 btrfs_set_file_extent_num_bytes(leaf, fi,
353 extent_end - args->start);
354 btrfs_mark_buffer_dirty(trans, leaf);
355
356 if (update_refs && disk_bytenr > 0) {
357 struct btrfs_ref ref = {
358 .action = BTRFS_ADD_DELAYED_REF,
359 .bytenr = disk_bytenr,
360 .num_bytes = num_bytes,
361 .parent = 0,
362 .owning_root = btrfs_root_id(root),
363 .ref_root = btrfs_root_id(root),
364 };
365 btrfs_init_data_ref(&ref, new_key.objectid,
366 args->start - extent_offset,
367 0, false);
368 ret = btrfs_inc_extent_ref(trans, &ref);
369 if (ret) {
370 btrfs_abort_transaction(trans, ret);
371 break;
372 }
373 }
374 key.offset = args->start;
375 }
376 /*
377 * From here on out we will have actually dropped something, so
378 * last_end can be updated.
379 */
380 last_end = extent_end;
381
382 /*
383 * | ---- range to drop ----- |
384 * | -------- extent -------- |
385 */
386 if (args->start <= key.offset && args->end < extent_end) {
387 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
388 ret = -EOPNOTSUPP;
389 break;
390 }
391
392 memcpy(&new_key, &key, sizeof(new_key));
393 new_key.offset = args->end;
394 btrfs_set_item_key_safe(trans, path, &new_key);
395
396 extent_offset += args->end - key.offset;
397 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
398 btrfs_set_file_extent_num_bytes(leaf, fi,
399 extent_end - args->end);
400 btrfs_mark_buffer_dirty(trans, leaf);
401 if (update_refs && disk_bytenr > 0)
402 args->bytes_found += args->end - key.offset;
403 break;
404 }
405
406 search_start = extent_end;
407 /*
408 * | ---- range to drop ----- |
409 * | -------- extent -------- |
410 */
411 if (args->start > key.offset && args->end >= extent_end) {
412 BUG_ON(del_nr > 0);
413 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
414 ret = -EOPNOTSUPP;
415 break;
416 }
417
418 btrfs_set_file_extent_num_bytes(leaf, fi,
419 args->start - key.offset);
420 btrfs_mark_buffer_dirty(trans, leaf);
421 if (update_refs && disk_bytenr > 0)
422 args->bytes_found += extent_end - args->start;
423 if (args->end == extent_end)
424 break;
425
426 path->slots[0]++;
427 goto next_slot;
428 }
429
430 /*
431 * | ---- range to drop ----- |
432 * | ------ extent ------ |
433 */
434 if (args->start <= key.offset && args->end >= extent_end) {
435delete_extent_item:
436 if (del_nr == 0) {
437 del_slot = path->slots[0];
438 del_nr = 1;
439 } else {
440 BUG_ON(del_slot + del_nr != path->slots[0]);
441 del_nr++;
442 }
443
444 if (update_refs &&
445 extent_type == BTRFS_FILE_EXTENT_INLINE) {
446 args->bytes_found += extent_end - key.offset;
447 extent_end = ALIGN(extent_end,
448 fs_info->sectorsize);
449 } else if (update_refs && disk_bytenr > 0) {
450 struct btrfs_ref ref = {
451 .action = BTRFS_DROP_DELAYED_REF,
452 .bytenr = disk_bytenr,
453 .num_bytes = num_bytes,
454 .parent = 0,
455 .owning_root = btrfs_root_id(root),
456 .ref_root = btrfs_root_id(root),
457 };
458 btrfs_init_data_ref(&ref, key.objectid,
459 key.offset - extent_offset,
460 0, false);
461 ret = btrfs_free_extent(trans, &ref);
462 if (ret) {
463 btrfs_abort_transaction(trans, ret);
464 break;
465 }
466 args->bytes_found += extent_end - key.offset;
467 }
468
469 if (args->end == extent_end)
470 break;
471
472 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
473 path->slots[0]++;
474 goto next_slot;
475 }
476
477 ret = btrfs_del_items(trans, root, path, del_slot,
478 del_nr);
479 if (ret) {
480 btrfs_abort_transaction(trans, ret);
481 break;
482 }
483
484 del_nr = 0;
485 del_slot = 0;
486
487 btrfs_release_path(path);
488 continue;
489 }
490
491 BUG();
492 }
493
494 if (!ret && del_nr > 0) {
495 /*
496 * Set path->slots[0] to first slot, so that after the delete
497 * if items are move off from our leaf to its immediate left or
498 * right neighbor leafs, we end up with a correct and adjusted
499 * path->slots[0] for our insertion (if args->replace_extent).
500 */
501 path->slots[0] = del_slot;
502 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
503 if (ret)
504 btrfs_abort_transaction(trans, ret);
505 }
506
507 leaf = path->nodes[0];
508 /*
509 * If btrfs_del_items() was called, it might have deleted a leaf, in
510 * which case it unlocked our path, so check path->locks[0] matches a
511 * write lock.
512 */
513 if (!ret && args->replace_extent &&
514 path->locks[0] == BTRFS_WRITE_LOCK &&
515 btrfs_leaf_free_space(leaf) >=
516 sizeof(struct btrfs_item) + args->extent_item_size) {
517
518 key.objectid = ino;
519 key.type = BTRFS_EXTENT_DATA_KEY;
520 key.offset = args->start;
521 if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
522 struct btrfs_key slot_key;
523
524 btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
525 if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
526 path->slots[0]++;
527 }
528 btrfs_setup_item_for_insert(trans, root, path, &key,
529 args->extent_item_size);
530 args->extent_inserted = true;
531 }
532
533 if (!args->path)
534 btrfs_free_path(path);
535 else if (!args->extent_inserted)
536 btrfs_release_path(path);
537out:
538 args->drop_end = found ? min(args->end, last_end) : args->end;
539
540 return ret;
541}
542
543static int extent_mergeable(struct extent_buffer *leaf, int slot,
544 u64 objectid, u64 bytenr, u64 orig_offset,
545 u64 *start, u64 *end)
546{
547 struct btrfs_file_extent_item *fi;
548 struct btrfs_key key;
549 u64 extent_end;
550
551 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
552 return 0;
553
554 btrfs_item_key_to_cpu(leaf, &key, slot);
555 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
556 return 0;
557
558 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
559 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
560 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
561 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
562 btrfs_file_extent_compression(leaf, fi) ||
563 btrfs_file_extent_encryption(leaf, fi) ||
564 btrfs_file_extent_other_encoding(leaf, fi))
565 return 0;
566
567 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
568 if ((*start && *start != key.offset) || (*end && *end != extent_end))
569 return 0;
570
571 *start = key.offset;
572 *end = extent_end;
573 return 1;
574}
575
576/*
577 * Mark extent in the range start - end as written.
578 *
579 * This changes extent type from 'pre-allocated' to 'regular'. If only
580 * part of extent is marked as written, the extent will be split into
581 * two or three.
582 */
583int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
584 struct btrfs_inode *inode, u64 start, u64 end)
585{
586 struct btrfs_root *root = inode->root;
587 struct extent_buffer *leaf;
588 struct btrfs_path *path;
589 struct btrfs_file_extent_item *fi;
590 struct btrfs_ref ref = { 0 };
591 struct btrfs_key key;
592 struct btrfs_key new_key;
593 u64 bytenr;
594 u64 num_bytes;
595 u64 extent_end;
596 u64 orig_offset;
597 u64 other_start;
598 u64 other_end;
599 u64 split;
600 int del_nr = 0;
601 int del_slot = 0;
602 int recow;
603 int ret = 0;
604 u64 ino = btrfs_ino(inode);
605
606 path = btrfs_alloc_path();
607 if (!path)
608 return -ENOMEM;
609again:
610 recow = 0;
611 split = start;
612 key.objectid = ino;
613 key.type = BTRFS_EXTENT_DATA_KEY;
614 key.offset = split;
615
616 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
617 if (ret < 0)
618 goto out;
619 if (ret > 0 && path->slots[0] > 0)
620 path->slots[0]--;
621
622 leaf = path->nodes[0];
623 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
624 if (key.objectid != ino ||
625 key.type != BTRFS_EXTENT_DATA_KEY) {
626 ret = -EINVAL;
627 btrfs_abort_transaction(trans, ret);
628 goto out;
629 }
630 fi = btrfs_item_ptr(leaf, path->slots[0],
631 struct btrfs_file_extent_item);
632 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_PREALLOC) {
633 ret = -EINVAL;
634 btrfs_abort_transaction(trans, ret);
635 goto out;
636 }
637 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
638 if (key.offset > start || extent_end < end) {
639 ret = -EINVAL;
640 btrfs_abort_transaction(trans, ret);
641 goto out;
642 }
643
644 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
645 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
646 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
647 memcpy(&new_key, &key, sizeof(new_key));
648
649 if (start == key.offset && end < extent_end) {
650 other_start = 0;
651 other_end = start;
652 if (extent_mergeable(leaf, path->slots[0] - 1,
653 ino, bytenr, orig_offset,
654 &other_start, &other_end)) {
655 new_key.offset = end;
656 btrfs_set_item_key_safe(trans, path, &new_key);
657 fi = btrfs_item_ptr(leaf, path->slots[0],
658 struct btrfs_file_extent_item);
659 btrfs_set_file_extent_generation(leaf, fi,
660 trans->transid);
661 btrfs_set_file_extent_num_bytes(leaf, fi,
662 extent_end - end);
663 btrfs_set_file_extent_offset(leaf, fi,
664 end - orig_offset);
665 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
666 struct btrfs_file_extent_item);
667 btrfs_set_file_extent_generation(leaf, fi,
668 trans->transid);
669 btrfs_set_file_extent_num_bytes(leaf, fi,
670 end - other_start);
671 btrfs_mark_buffer_dirty(trans, leaf);
672 goto out;
673 }
674 }
675
676 if (start > key.offset && end == extent_end) {
677 other_start = end;
678 other_end = 0;
679 if (extent_mergeable(leaf, path->slots[0] + 1,
680 ino, bytenr, orig_offset,
681 &other_start, &other_end)) {
682 fi = btrfs_item_ptr(leaf, path->slots[0],
683 struct btrfs_file_extent_item);
684 btrfs_set_file_extent_num_bytes(leaf, fi,
685 start - key.offset);
686 btrfs_set_file_extent_generation(leaf, fi,
687 trans->transid);
688 path->slots[0]++;
689 new_key.offset = start;
690 btrfs_set_item_key_safe(trans, path, &new_key);
691
692 fi = btrfs_item_ptr(leaf, path->slots[0],
693 struct btrfs_file_extent_item);
694 btrfs_set_file_extent_generation(leaf, fi,
695 trans->transid);
696 btrfs_set_file_extent_num_bytes(leaf, fi,
697 other_end - start);
698 btrfs_set_file_extent_offset(leaf, fi,
699 start - orig_offset);
700 btrfs_mark_buffer_dirty(trans, leaf);
701 goto out;
702 }
703 }
704
705 while (start > key.offset || end < extent_end) {
706 if (key.offset == start)
707 split = end;
708
709 new_key.offset = split;
710 ret = btrfs_duplicate_item(trans, root, path, &new_key);
711 if (ret == -EAGAIN) {
712 btrfs_release_path(path);
713 goto again;
714 }
715 if (ret < 0) {
716 btrfs_abort_transaction(trans, ret);
717 goto out;
718 }
719
720 leaf = path->nodes[0];
721 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
722 struct btrfs_file_extent_item);
723 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
724 btrfs_set_file_extent_num_bytes(leaf, fi,
725 split - key.offset);
726
727 fi = btrfs_item_ptr(leaf, path->slots[0],
728 struct btrfs_file_extent_item);
729
730 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
731 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
732 btrfs_set_file_extent_num_bytes(leaf, fi,
733 extent_end - split);
734 btrfs_mark_buffer_dirty(trans, leaf);
735
736 ref.action = BTRFS_ADD_DELAYED_REF;
737 ref.bytenr = bytenr;
738 ref.num_bytes = num_bytes;
739 ref.parent = 0;
740 ref.owning_root = btrfs_root_id(root);
741 ref.ref_root = btrfs_root_id(root);
742 btrfs_init_data_ref(&ref, ino, orig_offset, 0, false);
743 ret = btrfs_inc_extent_ref(trans, &ref);
744 if (ret) {
745 btrfs_abort_transaction(trans, ret);
746 goto out;
747 }
748
749 if (split == start) {
750 key.offset = start;
751 } else {
752 if (start != key.offset) {
753 ret = -EINVAL;
754 btrfs_abort_transaction(trans, ret);
755 goto out;
756 }
757 path->slots[0]--;
758 extent_end = end;
759 }
760 recow = 1;
761 }
762
763 other_start = end;
764 other_end = 0;
765
766 ref.action = BTRFS_DROP_DELAYED_REF;
767 ref.bytenr = bytenr;
768 ref.num_bytes = num_bytes;
769 ref.parent = 0;
770 ref.owning_root = btrfs_root_id(root);
771 ref.ref_root = btrfs_root_id(root);
772 btrfs_init_data_ref(&ref, ino, orig_offset, 0, false);
773 if (extent_mergeable(leaf, path->slots[0] + 1,
774 ino, bytenr, orig_offset,
775 &other_start, &other_end)) {
776 if (recow) {
777 btrfs_release_path(path);
778 goto again;
779 }
780 extent_end = other_end;
781 del_slot = path->slots[0] + 1;
782 del_nr++;
783 ret = btrfs_free_extent(trans, &ref);
784 if (ret) {
785 btrfs_abort_transaction(trans, ret);
786 goto out;
787 }
788 }
789 other_start = 0;
790 other_end = start;
791 if (extent_mergeable(leaf, path->slots[0] - 1,
792 ino, bytenr, orig_offset,
793 &other_start, &other_end)) {
794 if (recow) {
795 btrfs_release_path(path);
796 goto again;
797 }
798 key.offset = other_start;
799 del_slot = path->slots[0];
800 del_nr++;
801 ret = btrfs_free_extent(trans, &ref);
802 if (ret) {
803 btrfs_abort_transaction(trans, ret);
804 goto out;
805 }
806 }
807 if (del_nr == 0) {
808 fi = btrfs_item_ptr(leaf, path->slots[0],
809 struct btrfs_file_extent_item);
810 btrfs_set_file_extent_type(leaf, fi,
811 BTRFS_FILE_EXTENT_REG);
812 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
813 btrfs_mark_buffer_dirty(trans, leaf);
814 } else {
815 fi = btrfs_item_ptr(leaf, del_slot - 1,
816 struct btrfs_file_extent_item);
817 btrfs_set_file_extent_type(leaf, fi,
818 BTRFS_FILE_EXTENT_REG);
819 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
820 btrfs_set_file_extent_num_bytes(leaf, fi,
821 extent_end - key.offset);
822 btrfs_mark_buffer_dirty(trans, leaf);
823
824 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
825 if (ret < 0) {
826 btrfs_abort_transaction(trans, ret);
827 goto out;
828 }
829 }
830out:
831 btrfs_free_path(path);
832 return ret;
833}
834
835/*
836 * On error return an unlocked folio and the error value
837 * On success return a locked folio and 0
838 */
839static int prepare_uptodate_folio(struct inode *inode, struct folio *folio, u64 pos,
840 u64 len, bool force_uptodate)
841{
842 u64 clamp_start = max_t(u64, pos, folio_pos(folio));
843 u64 clamp_end = min_t(u64, pos + len, folio_pos(folio) + folio_size(folio));
844 int ret = 0;
845
846 if (folio_test_uptodate(folio))
847 return 0;
848
849 if (!force_uptodate &&
850 IS_ALIGNED(clamp_start, PAGE_SIZE) &&
851 IS_ALIGNED(clamp_end, PAGE_SIZE))
852 return 0;
853
854 ret = btrfs_read_folio(NULL, folio);
855 if (ret)
856 return ret;
857 folio_lock(folio);
858 if (!folio_test_uptodate(folio)) {
859 folio_unlock(folio);
860 return -EIO;
861 }
862
863 /*
864 * Since btrfs_read_folio() will unlock the folio before it returns,
865 * there is a window where btrfs_release_folio() can be called to
866 * release the page. Here we check both inode mapping and page
867 * private to make sure the page was not released.
868 *
869 * The private flag check is essential for subpage as we need to store
870 * extra bitmap using folio private.
871 */
872 if (folio->mapping != inode->i_mapping || !folio_test_private(folio)) {
873 folio_unlock(folio);
874 return -EAGAIN;
875 }
876 return 0;
877}
878
879static gfp_t get_prepare_gfp_flags(struct inode *inode, bool nowait)
880{
881 gfp_t gfp;
882
883 gfp = btrfs_alloc_write_mask(inode->i_mapping);
884 if (nowait) {
885 gfp &= ~__GFP_DIRECT_RECLAIM;
886 gfp |= GFP_NOWAIT;
887 }
888
889 return gfp;
890}
891
892/*
893 * Get folio into the page cache and lock it.
894 */
895static noinline int prepare_one_folio(struct inode *inode, struct folio **folio_ret,
896 loff_t pos, size_t write_bytes,
897 bool force_uptodate, bool nowait)
898{
899 unsigned long index = pos >> PAGE_SHIFT;
900 gfp_t mask = get_prepare_gfp_flags(inode, nowait);
901 fgf_t fgp_flags = (nowait ? FGP_WRITEBEGIN | FGP_NOWAIT : FGP_WRITEBEGIN);
902 struct folio *folio;
903 int ret = 0;
904
905again:
906 folio = __filemap_get_folio(inode->i_mapping, index, fgp_flags, mask);
907 if (IS_ERR(folio)) {
908 if (nowait)
909 ret = -EAGAIN;
910 else
911 ret = PTR_ERR(folio);
912 return ret;
913 }
914 folio_wait_writeback(folio);
915 /* Only support page sized folio yet. */
916 ASSERT(folio_order(folio) == 0);
917 ret = set_folio_extent_mapped(folio);
918 if (ret < 0) {
919 folio_unlock(folio);
920 folio_put(folio);
921 return ret;
922 }
923 ret = prepare_uptodate_folio(inode, folio, pos, write_bytes, force_uptodate);
924 if (ret) {
925 /* The folio is already unlocked. */
926 folio_put(folio);
927 if (!nowait && ret == -EAGAIN) {
928 ret = 0;
929 goto again;
930 }
931 return ret;
932 }
933 *folio_ret = folio;
934 return 0;
935}
936
937/*
938 * Locks the extent and properly waits for data=ordered extents to finish
939 * before allowing the folios to be modified if need.
940 *
941 * Return:
942 * 1 - the extent is locked
943 * 0 - the extent is not locked, and everything is OK
944 * -EAGAIN - need to prepare the folios again
945 */
946static noinline int
947lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct folio *folio,
948 loff_t pos, size_t write_bytes,
949 u64 *lockstart, u64 *lockend, bool nowait,
950 struct extent_state **cached_state)
951{
952 struct btrfs_fs_info *fs_info = inode->root->fs_info;
953 u64 start_pos;
954 u64 last_pos;
955 int ret = 0;
956
957 start_pos = round_down(pos, fs_info->sectorsize);
958 last_pos = round_up(pos + write_bytes, fs_info->sectorsize) - 1;
959
960 if (start_pos < inode->vfs_inode.i_size) {
961 struct btrfs_ordered_extent *ordered;
962
963 if (nowait) {
964 if (!try_lock_extent(&inode->io_tree, start_pos, last_pos,
965 cached_state)) {
966 folio_unlock(folio);
967 folio_put(folio);
968 return -EAGAIN;
969 }
970 } else {
971 lock_extent(&inode->io_tree, start_pos, last_pos, cached_state);
972 }
973
974 ordered = btrfs_lookup_ordered_range(inode, start_pos,
975 last_pos - start_pos + 1);
976 if (ordered &&
977 ordered->file_offset + ordered->num_bytes > start_pos &&
978 ordered->file_offset <= last_pos) {
979 unlock_extent(&inode->io_tree, start_pos, last_pos,
980 cached_state);
981 folio_unlock(folio);
982 folio_put(folio);
983 btrfs_start_ordered_extent(ordered);
984 btrfs_put_ordered_extent(ordered);
985 return -EAGAIN;
986 }
987 if (ordered)
988 btrfs_put_ordered_extent(ordered);
989
990 *lockstart = start_pos;
991 *lockend = last_pos;
992 ret = 1;
993 }
994
995 /*
996 * We should be called after prepare_one_folio() which should have locked
997 * all pages in the range.
998 */
999 WARN_ON(!folio_test_locked(folio));
1000
1001 return ret;
1002}
1003
1004/*
1005 * Check if we can do nocow write into the range [@pos, @pos + @write_bytes)
1006 *
1007 * @pos: File offset.
1008 * @write_bytes: The length to write, will be updated to the nocow writeable
1009 * range.
1010 *
1011 * This function will flush ordered extents in the range to ensure proper
1012 * nocow checks.
1013 *
1014 * Return:
1015 * > 0 If we can nocow, and updates @write_bytes.
1016 * 0 If we can't do a nocow write.
1017 * -EAGAIN If we can't do a nocow write because snapshoting of the inode's
1018 * root is in progress.
1019 * < 0 If an error happened.
1020 *
1021 * NOTE: Callers need to call btrfs_check_nocow_unlock() if we return > 0.
1022 */
1023int btrfs_check_nocow_lock(struct btrfs_inode *inode, loff_t pos,
1024 size_t *write_bytes, bool nowait)
1025{
1026 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1027 struct btrfs_root *root = inode->root;
1028 struct extent_state *cached_state = NULL;
1029 u64 lockstart, lockend;
1030 u64 num_bytes;
1031 int ret;
1032
1033 if (!(inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
1034 return 0;
1035
1036 if (!btrfs_drew_try_write_lock(&root->snapshot_lock))
1037 return -EAGAIN;
1038
1039 lockstart = round_down(pos, fs_info->sectorsize);
1040 lockend = round_up(pos + *write_bytes,
1041 fs_info->sectorsize) - 1;
1042 num_bytes = lockend - lockstart + 1;
1043
1044 if (nowait) {
1045 if (!btrfs_try_lock_ordered_range(inode, lockstart, lockend,
1046 &cached_state)) {
1047 btrfs_drew_write_unlock(&root->snapshot_lock);
1048 return -EAGAIN;
1049 }
1050 } else {
1051 btrfs_lock_and_flush_ordered_range(inode, lockstart, lockend,
1052 &cached_state);
1053 }
1054 ret = can_nocow_extent(&inode->vfs_inode, lockstart, &num_bytes,
1055 NULL, nowait, false);
1056 if (ret <= 0)
1057 btrfs_drew_write_unlock(&root->snapshot_lock);
1058 else
1059 *write_bytes = min_t(size_t, *write_bytes ,
1060 num_bytes - pos + lockstart);
1061 unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
1062
1063 return ret;
1064}
1065
1066void btrfs_check_nocow_unlock(struct btrfs_inode *inode)
1067{
1068 btrfs_drew_write_unlock(&inode->root->snapshot_lock);
1069}
1070
1071int btrfs_write_check(struct kiocb *iocb, size_t count)
1072{
1073 struct file *file = iocb->ki_filp;
1074 struct inode *inode = file_inode(file);
1075 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
1076 loff_t pos = iocb->ki_pos;
1077 int ret;
1078 loff_t oldsize;
1079
1080 /*
1081 * Quickly bail out on NOWAIT writes if we don't have the nodatacow or
1082 * prealloc flags, as without those flags we always have to COW. We will
1083 * later check if we can really COW into the target range (using
1084 * can_nocow_extent() at btrfs_get_blocks_direct_write()).
1085 */
1086 if ((iocb->ki_flags & IOCB_NOWAIT) &&
1087 !(BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
1088 return -EAGAIN;
1089
1090 ret = file_remove_privs(file);
1091 if (ret)
1092 return ret;
1093
1094 /*
1095 * We reserve space for updating the inode when we reserve space for the
1096 * extent we are going to write, so we will enospc out there. We don't
1097 * need to start yet another transaction to update the inode as we will
1098 * update the inode when we finish writing whatever data we write.
1099 */
1100 if (!IS_NOCMTIME(inode)) {
1101 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
1102 inode_inc_iversion(inode);
1103 }
1104
1105 oldsize = i_size_read(inode);
1106 if (pos > oldsize) {
1107 /* Expand hole size to cover write data, preventing empty gap */
1108 loff_t end_pos = round_up(pos + count, fs_info->sectorsize);
1109
1110 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, end_pos);
1111 if (ret)
1112 return ret;
1113 }
1114
1115 return 0;
1116}
1117
1118ssize_t btrfs_buffered_write(struct kiocb *iocb, struct iov_iter *i)
1119{
1120 struct file *file = iocb->ki_filp;
1121 loff_t pos;
1122 struct inode *inode = file_inode(file);
1123 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
1124 struct extent_changeset *data_reserved = NULL;
1125 u64 release_bytes = 0;
1126 u64 lockstart;
1127 u64 lockend;
1128 size_t num_written = 0;
1129 ssize_t ret;
1130 loff_t old_isize;
1131 unsigned int ilock_flags = 0;
1132 const bool nowait = (iocb->ki_flags & IOCB_NOWAIT);
1133 unsigned int bdp_flags = (nowait ? BDP_ASYNC : 0);
1134 bool only_release_metadata = false;
1135
1136 if (nowait)
1137 ilock_flags |= BTRFS_ILOCK_TRY;
1138
1139 ret = btrfs_inode_lock(BTRFS_I(inode), ilock_flags);
1140 if (ret < 0)
1141 return ret;
1142
1143 /*
1144 * We can only trust the isize with inode lock held, or it can race with
1145 * other buffered writes and cause incorrect call of
1146 * pagecache_isize_extended() to overwrite existing data.
1147 */
1148 old_isize = i_size_read(inode);
1149
1150 ret = generic_write_checks(iocb, i);
1151 if (ret <= 0)
1152 goto out;
1153
1154 ret = btrfs_write_check(iocb, ret);
1155 if (ret < 0)
1156 goto out;
1157
1158 pos = iocb->ki_pos;
1159 while (iov_iter_count(i) > 0) {
1160 struct extent_state *cached_state = NULL;
1161 size_t offset = offset_in_page(pos);
1162 size_t sector_offset;
1163 size_t write_bytes = min(iov_iter_count(i), PAGE_SIZE - offset);
1164 size_t reserve_bytes;
1165 size_t copied;
1166 size_t dirty_sectors;
1167 size_t num_sectors;
1168 struct folio *folio = NULL;
1169 int extents_locked;
1170 bool force_page_uptodate = false;
1171
1172 /*
1173 * Fault pages before locking them in prepare_one_folio()
1174 * to avoid recursive lock
1175 */
1176 if (unlikely(fault_in_iov_iter_readable(i, write_bytes))) {
1177 ret = -EFAULT;
1178 break;
1179 }
1180
1181 only_release_metadata = false;
1182 sector_offset = pos & (fs_info->sectorsize - 1);
1183
1184 extent_changeset_release(data_reserved);
1185 ret = btrfs_check_data_free_space(BTRFS_I(inode),
1186 &data_reserved, pos,
1187 write_bytes, nowait);
1188 if (ret < 0) {
1189 int can_nocow;
1190
1191 if (nowait && (ret == -ENOSPC || ret == -EAGAIN)) {
1192 ret = -EAGAIN;
1193 break;
1194 }
1195
1196 /*
1197 * If we don't have to COW at the offset, reserve
1198 * metadata only. write_bytes may get smaller than
1199 * requested here.
1200 */
1201 can_nocow = btrfs_check_nocow_lock(BTRFS_I(inode), pos,
1202 &write_bytes, nowait);
1203 if (can_nocow < 0)
1204 ret = can_nocow;
1205 if (can_nocow > 0)
1206 ret = 0;
1207 if (ret)
1208 break;
1209 only_release_metadata = true;
1210 }
1211
1212 reserve_bytes = round_up(write_bytes + sector_offset,
1213 fs_info->sectorsize);
1214 WARN_ON(reserve_bytes == 0);
1215 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
1216 reserve_bytes,
1217 reserve_bytes, nowait);
1218 if (ret) {
1219 if (!only_release_metadata)
1220 btrfs_free_reserved_data_space(BTRFS_I(inode),
1221 data_reserved, pos,
1222 write_bytes);
1223 else
1224 btrfs_check_nocow_unlock(BTRFS_I(inode));
1225
1226 if (nowait && ret == -ENOSPC)
1227 ret = -EAGAIN;
1228 break;
1229 }
1230
1231 release_bytes = reserve_bytes;
1232again:
1233 ret = balance_dirty_pages_ratelimited_flags(inode->i_mapping, bdp_flags);
1234 if (ret) {
1235 btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes);
1236 break;
1237 }
1238
1239 ret = prepare_one_folio(inode, &folio, pos, write_bytes,
1240 force_page_uptodate, false);
1241 if (ret) {
1242 btrfs_delalloc_release_extents(BTRFS_I(inode),
1243 reserve_bytes);
1244 break;
1245 }
1246
1247 extents_locked = lock_and_cleanup_extent_if_need(BTRFS_I(inode),
1248 folio, pos, write_bytes, &lockstart,
1249 &lockend, nowait, &cached_state);
1250 if (extents_locked < 0) {
1251 if (!nowait && extents_locked == -EAGAIN)
1252 goto again;
1253
1254 btrfs_delalloc_release_extents(BTRFS_I(inode),
1255 reserve_bytes);
1256 ret = extents_locked;
1257 break;
1258 }
1259
1260 copied = btrfs_copy_from_user(pos, write_bytes, folio, i);
1261
1262 num_sectors = BTRFS_BYTES_TO_BLKS(fs_info, reserve_bytes);
1263 dirty_sectors = round_up(copied + sector_offset,
1264 fs_info->sectorsize);
1265 dirty_sectors = BTRFS_BYTES_TO_BLKS(fs_info, dirty_sectors);
1266
1267 if (copied == 0) {
1268 force_page_uptodate = true;
1269 dirty_sectors = 0;
1270 } else {
1271 force_page_uptodate = false;
1272 }
1273
1274 if (num_sectors > dirty_sectors) {
1275 /* release everything except the sectors we dirtied */
1276 release_bytes -= dirty_sectors << fs_info->sectorsize_bits;
1277 if (only_release_metadata) {
1278 btrfs_delalloc_release_metadata(BTRFS_I(inode),
1279 release_bytes, true);
1280 } else {
1281 u64 release_start = round_up(pos + copied,
1282 fs_info->sectorsize);
1283 btrfs_delalloc_release_space(BTRFS_I(inode),
1284 data_reserved, release_start,
1285 release_bytes, true);
1286 }
1287 }
1288
1289 release_bytes = round_up(copied + sector_offset,
1290 fs_info->sectorsize);
1291
1292 ret = btrfs_dirty_folio(BTRFS_I(inode), folio, pos, copied,
1293 &cached_state, only_release_metadata);
1294
1295 /*
1296 * If we have not locked the extent range, because the range's
1297 * start offset is >= i_size, we might still have a non-NULL
1298 * cached extent state, acquired while marking the extent range
1299 * as delalloc through btrfs_dirty_page(). Therefore free any
1300 * possible cached extent state to avoid a memory leak.
1301 */
1302 if (extents_locked)
1303 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart,
1304 lockend, &cached_state);
1305 else
1306 free_extent_state(cached_state);
1307
1308 btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes);
1309 if (ret) {
1310 btrfs_drop_folio(fs_info, folio, pos, copied);
1311 break;
1312 }
1313
1314 release_bytes = 0;
1315 if (only_release_metadata)
1316 btrfs_check_nocow_unlock(BTRFS_I(inode));
1317
1318 btrfs_drop_folio(fs_info, folio, pos, copied);
1319
1320 cond_resched();
1321
1322 pos += copied;
1323 num_written += copied;
1324 }
1325
1326 if (release_bytes) {
1327 if (only_release_metadata) {
1328 btrfs_check_nocow_unlock(BTRFS_I(inode));
1329 btrfs_delalloc_release_metadata(BTRFS_I(inode),
1330 release_bytes, true);
1331 } else {
1332 btrfs_delalloc_release_space(BTRFS_I(inode),
1333 data_reserved,
1334 round_down(pos, fs_info->sectorsize),
1335 release_bytes, true);
1336 }
1337 }
1338
1339 extent_changeset_free(data_reserved);
1340 if (num_written > 0) {
1341 pagecache_isize_extended(inode, old_isize, iocb->ki_pos);
1342 iocb->ki_pos += num_written;
1343 }
1344out:
1345 btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1346 return num_written ? num_written : ret;
1347}
1348
1349static ssize_t btrfs_encoded_write(struct kiocb *iocb, struct iov_iter *from,
1350 const struct btrfs_ioctl_encoded_io_args *encoded)
1351{
1352 struct file *file = iocb->ki_filp;
1353 struct inode *inode = file_inode(file);
1354 loff_t count;
1355 ssize_t ret;
1356
1357 btrfs_inode_lock(BTRFS_I(inode), 0);
1358 count = encoded->len;
1359 ret = generic_write_checks_count(iocb, &count);
1360 if (ret == 0 && count != encoded->len) {
1361 /*
1362 * The write got truncated by generic_write_checks_count(). We
1363 * can't do a partial encoded write.
1364 */
1365 ret = -EFBIG;
1366 }
1367 if (ret || encoded->len == 0)
1368 goto out;
1369
1370 ret = btrfs_write_check(iocb, encoded->len);
1371 if (ret < 0)
1372 goto out;
1373
1374 ret = btrfs_do_encoded_write(iocb, from, encoded);
1375out:
1376 btrfs_inode_unlock(BTRFS_I(inode), 0);
1377 return ret;
1378}
1379
1380ssize_t btrfs_do_write_iter(struct kiocb *iocb, struct iov_iter *from,
1381 const struct btrfs_ioctl_encoded_io_args *encoded)
1382{
1383 struct file *file = iocb->ki_filp;
1384 struct btrfs_inode *inode = BTRFS_I(file_inode(file));
1385 ssize_t num_written, num_sync;
1386
1387 /*
1388 * If the fs flips readonly due to some impossible error, although we
1389 * have opened a file as writable, we have to stop this write operation
1390 * to ensure consistency.
1391 */
1392 if (BTRFS_FS_ERROR(inode->root->fs_info))
1393 return -EROFS;
1394
1395 if (encoded && (iocb->ki_flags & IOCB_NOWAIT))
1396 return -EOPNOTSUPP;
1397
1398 if (encoded) {
1399 num_written = btrfs_encoded_write(iocb, from, encoded);
1400 num_sync = encoded->len;
1401 } else if (iocb->ki_flags & IOCB_DIRECT) {
1402 num_written = btrfs_direct_write(iocb, from);
1403 num_sync = num_written;
1404 } else {
1405 num_written = btrfs_buffered_write(iocb, from);
1406 num_sync = num_written;
1407 }
1408
1409 btrfs_set_inode_last_sub_trans(inode);
1410
1411 if (num_sync > 0) {
1412 num_sync = generic_write_sync(iocb, num_sync);
1413 if (num_sync < 0)
1414 num_written = num_sync;
1415 }
1416
1417 return num_written;
1418}
1419
1420static ssize_t btrfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1421{
1422 return btrfs_do_write_iter(iocb, from, NULL);
1423}
1424
1425int btrfs_release_file(struct inode *inode, struct file *filp)
1426{
1427 struct btrfs_file_private *private = filp->private_data;
1428
1429 if (private) {
1430 kfree(private->filldir_buf);
1431 free_extent_state(private->llseek_cached_state);
1432 kfree(private);
1433 filp->private_data = NULL;
1434 }
1435
1436 /*
1437 * Set by setattr when we are about to truncate a file from a non-zero
1438 * size to a zero size. This tries to flush down new bytes that may
1439 * have been written if the application were using truncate to replace
1440 * a file in place.
1441 */
1442 if (test_and_clear_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
1443 &BTRFS_I(inode)->runtime_flags))
1444 filemap_flush(inode->i_mapping);
1445 return 0;
1446}
1447
1448static int start_ordered_ops(struct btrfs_inode *inode, loff_t start, loff_t end)
1449{
1450 int ret;
1451 struct blk_plug plug;
1452
1453 /*
1454 * This is only called in fsync, which would do synchronous writes, so
1455 * a plug can merge adjacent IOs as much as possible. Esp. in case of
1456 * multiple disks using raid profile, a large IO can be split to
1457 * several segments of stripe length (currently 64K).
1458 */
1459 blk_start_plug(&plug);
1460 ret = btrfs_fdatawrite_range(inode, start, end);
1461 blk_finish_plug(&plug);
1462
1463 return ret;
1464}
1465
1466static inline bool skip_inode_logging(const struct btrfs_log_ctx *ctx)
1467{
1468 struct btrfs_inode *inode = ctx->inode;
1469 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1470
1471 if (btrfs_inode_in_log(inode, btrfs_get_fs_generation(fs_info)) &&
1472 list_empty(&ctx->ordered_extents))
1473 return true;
1474
1475 /*
1476 * If we are doing a fast fsync we can not bail out if the inode's
1477 * last_trans is <= then the last committed transaction, because we only
1478 * update the last_trans of the inode during ordered extent completion,
1479 * and for a fast fsync we don't wait for that, we only wait for the
1480 * writeback to complete.
1481 */
1482 if (inode->last_trans <= btrfs_get_last_trans_committed(fs_info) &&
1483 (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) ||
1484 list_empty(&ctx->ordered_extents)))
1485 return true;
1486
1487 return false;
1488}
1489
1490/*
1491 * fsync call for both files and directories. This logs the inode into
1492 * the tree log instead of forcing full commits whenever possible.
1493 *
1494 * It needs to call filemap_fdatawait so that all ordered extent updates are
1495 * in the metadata btree are up to date for copying to the log.
1496 *
1497 * It drops the inode mutex before doing the tree log commit. This is an
1498 * important optimization for directories because holding the mutex prevents
1499 * new operations on the dir while we write to disk.
1500 */
1501int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1502{
1503 struct dentry *dentry = file_dentry(file);
1504 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
1505 struct btrfs_root *root = inode->root;
1506 struct btrfs_fs_info *fs_info = root->fs_info;
1507 struct btrfs_trans_handle *trans;
1508 struct btrfs_log_ctx ctx;
1509 int ret = 0, err;
1510 u64 len;
1511 bool full_sync;
1512 bool skip_ilock = false;
1513
1514 if (current->journal_info == BTRFS_TRANS_DIO_WRITE_STUB) {
1515 skip_ilock = true;
1516 current->journal_info = NULL;
1517 btrfs_assert_inode_locked(inode);
1518 }
1519
1520 trace_btrfs_sync_file(file, datasync);
1521
1522 btrfs_init_log_ctx(&ctx, inode);
1523
1524 /*
1525 * Always set the range to a full range, otherwise we can get into
1526 * several problems, from missing file extent items to represent holes
1527 * when not using the NO_HOLES feature, to log tree corruption due to
1528 * races between hole detection during logging and completion of ordered
1529 * extents outside the range, to missing checksums due to ordered extents
1530 * for which we flushed only a subset of their pages.
1531 */
1532 start = 0;
1533 end = LLONG_MAX;
1534 len = (u64)LLONG_MAX + 1;
1535
1536 /*
1537 * We write the dirty pages in the range and wait until they complete
1538 * out of the ->i_mutex. If so, we can flush the dirty pages by
1539 * multi-task, and make the performance up. See
1540 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1541 */
1542 ret = start_ordered_ops(inode, start, end);
1543 if (ret)
1544 goto out;
1545
1546 if (skip_ilock)
1547 down_write(&inode->i_mmap_lock);
1548 else
1549 btrfs_inode_lock(inode, BTRFS_ILOCK_MMAP);
1550
1551 atomic_inc(&root->log_batch);
1552
1553 /*
1554 * Before we acquired the inode's lock and the mmap lock, someone may
1555 * have dirtied more pages in the target range. We need to make sure
1556 * that writeback for any such pages does not start while we are logging
1557 * the inode, because if it does, any of the following might happen when
1558 * we are not doing a full inode sync:
1559 *
1560 * 1) We log an extent after its writeback finishes but before its
1561 * checksums are added to the csum tree, leading to -EIO errors
1562 * when attempting to read the extent after a log replay.
1563 *
1564 * 2) We can end up logging an extent before its writeback finishes.
1565 * Therefore after the log replay we will have a file extent item
1566 * pointing to an unwritten extent (and no data checksums as well).
1567 *
1568 * So trigger writeback for any eventual new dirty pages and then we
1569 * wait for all ordered extents to complete below.
1570 */
1571 ret = start_ordered_ops(inode, start, end);
1572 if (ret) {
1573 if (skip_ilock)
1574 up_write(&inode->i_mmap_lock);
1575 else
1576 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
1577 goto out;
1578 }
1579
1580 /*
1581 * Always check for the full sync flag while holding the inode's lock,
1582 * to avoid races with other tasks. The flag must be either set all the
1583 * time during logging or always off all the time while logging.
1584 * We check the flag here after starting delalloc above, because when
1585 * running delalloc the full sync flag may be set if we need to drop
1586 * extra extent map ranges due to temporary memory allocation failures.
1587 */
1588 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
1589
1590 /*
1591 * We have to do this here to avoid the priority inversion of waiting on
1592 * IO of a lower priority task while holding a transaction open.
1593 *
1594 * For a full fsync we wait for the ordered extents to complete while
1595 * for a fast fsync we wait just for writeback to complete, and then
1596 * attach the ordered extents to the transaction so that a transaction
1597 * commit waits for their completion, to avoid data loss if we fsync,
1598 * the current transaction commits before the ordered extents complete
1599 * and a power failure happens right after that.
1600 *
1601 * For zoned filesystem, if a write IO uses a ZONE_APPEND command, the
1602 * logical address recorded in the ordered extent may change. We need
1603 * to wait for the IO to stabilize the logical address.
1604 */
1605 if (full_sync || btrfs_is_zoned(fs_info)) {
1606 ret = btrfs_wait_ordered_range(inode, start, len);
1607 clear_bit(BTRFS_INODE_COW_WRITE_ERROR, &inode->runtime_flags);
1608 } else {
1609 /*
1610 * Get our ordered extents as soon as possible to avoid doing
1611 * checksum lookups in the csum tree, and use instead the
1612 * checksums attached to the ordered extents.
1613 */
1614 btrfs_get_ordered_extents_for_logging(inode, &ctx.ordered_extents);
1615 ret = filemap_fdatawait_range(inode->vfs_inode.i_mapping, start, end);
1616 if (ret)
1617 goto out_release_extents;
1618
1619 /*
1620 * Check and clear the BTRFS_INODE_COW_WRITE_ERROR now after
1621 * starting and waiting for writeback, because for buffered IO
1622 * it may have been set during the end IO callback
1623 * (end_bbio_data_write() -> btrfs_finish_ordered_extent()) in
1624 * case an error happened and we need to wait for ordered
1625 * extents to complete so that any extent maps that point to
1626 * unwritten locations are dropped and we don't log them.
1627 */
1628 if (test_and_clear_bit(BTRFS_INODE_COW_WRITE_ERROR, &inode->runtime_flags))
1629 ret = btrfs_wait_ordered_range(inode, start, len);
1630 }
1631
1632 if (ret)
1633 goto out_release_extents;
1634
1635 atomic_inc(&root->log_batch);
1636
1637 if (skip_inode_logging(&ctx)) {
1638 /*
1639 * We've had everything committed since the last time we were
1640 * modified so clear this flag in case it was set for whatever
1641 * reason, it's no longer relevant.
1642 */
1643 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
1644 /*
1645 * An ordered extent might have started before and completed
1646 * already with io errors, in which case the inode was not
1647 * updated and we end up here. So check the inode's mapping
1648 * for any errors that might have happened since we last
1649 * checked called fsync.
1650 */
1651 ret = filemap_check_wb_err(inode->vfs_inode.i_mapping, file->f_wb_err);
1652 goto out_release_extents;
1653 }
1654
1655 btrfs_init_log_ctx_scratch_eb(&ctx);
1656
1657 /*
1658 * We use start here because we will need to wait on the IO to complete
1659 * in btrfs_sync_log, which could require joining a transaction (for
1660 * example checking cross references in the nocow path). If we use join
1661 * here we could get into a situation where we're waiting on IO to
1662 * happen that is blocked on a transaction trying to commit. With start
1663 * we inc the extwriter counter, so we wait for all extwriters to exit
1664 * before we start blocking joiners. This comment is to keep somebody
1665 * from thinking they are super smart and changing this to
1666 * btrfs_join_transaction *cough*Josef*cough*.
1667 */
1668 trans = btrfs_start_transaction(root, 0);
1669 if (IS_ERR(trans)) {
1670 ret = PTR_ERR(trans);
1671 goto out_release_extents;
1672 }
1673 trans->in_fsync = true;
1674
1675 ret = btrfs_log_dentry_safe(trans, dentry, &ctx);
1676 /*
1677 * Scratch eb no longer needed, release before syncing log or commit
1678 * transaction, to avoid holding unnecessary memory during such long
1679 * operations.
1680 */
1681 if (ctx.scratch_eb) {
1682 free_extent_buffer(ctx.scratch_eb);
1683 ctx.scratch_eb = NULL;
1684 }
1685 btrfs_release_log_ctx_extents(&ctx);
1686 if (ret < 0) {
1687 /* Fallthrough and commit/free transaction. */
1688 ret = BTRFS_LOG_FORCE_COMMIT;
1689 }
1690
1691 /* we've logged all the items and now have a consistent
1692 * version of the file in the log. It is possible that
1693 * someone will come in and modify the file, but that's
1694 * fine because the log is consistent on disk, and we
1695 * have references to all of the file's extents
1696 *
1697 * It is possible that someone will come in and log the
1698 * file again, but that will end up using the synchronization
1699 * inside btrfs_sync_log to keep things safe.
1700 */
1701 if (skip_ilock)
1702 up_write(&inode->i_mmap_lock);
1703 else
1704 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
1705
1706 if (ret == BTRFS_NO_LOG_SYNC) {
1707 ret = btrfs_end_transaction(trans);
1708 goto out;
1709 }
1710
1711 /* We successfully logged the inode, attempt to sync the log. */
1712 if (!ret) {
1713 ret = btrfs_sync_log(trans, root, &ctx);
1714 if (!ret) {
1715 ret = btrfs_end_transaction(trans);
1716 goto out;
1717 }
1718 }
1719
1720 /*
1721 * At this point we need to commit the transaction because we had
1722 * btrfs_need_log_full_commit() or some other error.
1723 *
1724 * If we didn't do a full sync we have to stop the trans handle, wait on
1725 * the ordered extents, start it again and commit the transaction. If
1726 * we attempt to wait on the ordered extents here we could deadlock with
1727 * something like fallocate() that is holding the extent lock trying to
1728 * start a transaction while some other thread is trying to commit the
1729 * transaction while we (fsync) are currently holding the transaction
1730 * open.
1731 */
1732 if (!full_sync) {
1733 ret = btrfs_end_transaction(trans);
1734 if (ret)
1735 goto out;
1736 ret = btrfs_wait_ordered_range(inode, start, len);
1737 if (ret)
1738 goto out;
1739
1740 /*
1741 * This is safe to use here because we're only interested in
1742 * making sure the transaction that had the ordered extents is
1743 * committed. We aren't waiting on anything past this point,
1744 * we're purely getting the transaction and committing it.
1745 */
1746 trans = btrfs_attach_transaction_barrier(root);
1747 if (IS_ERR(trans)) {
1748 ret = PTR_ERR(trans);
1749
1750 /*
1751 * We committed the transaction and there's no currently
1752 * running transaction, this means everything we care
1753 * about made it to disk and we are done.
1754 */
1755 if (ret == -ENOENT)
1756 ret = 0;
1757 goto out;
1758 }
1759 }
1760
1761 ret = btrfs_commit_transaction(trans);
1762out:
1763 free_extent_buffer(ctx.scratch_eb);
1764 ASSERT(list_empty(&ctx.list));
1765 ASSERT(list_empty(&ctx.conflict_inodes));
1766 err = file_check_and_advance_wb_err(file);
1767 if (!ret)
1768 ret = err;
1769 return ret > 0 ? -EIO : ret;
1770
1771out_release_extents:
1772 btrfs_release_log_ctx_extents(&ctx);
1773 if (skip_ilock)
1774 up_write(&inode->i_mmap_lock);
1775 else
1776 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
1777 goto out;
1778}
1779
1780/*
1781 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
1782 * called from a page fault handler when a page is first dirtied. Hence we must
1783 * be careful to check for EOF conditions here. We set the page up correctly
1784 * for a written page which means we get ENOSPC checking when writing into
1785 * holes and correct delalloc and unwritten extent mapping on filesystems that
1786 * support these features.
1787 *
1788 * We are not allowed to take the i_mutex here so we have to play games to
1789 * protect against truncate races as the page could now be beyond EOF. Because
1790 * truncate_setsize() writes the inode size before removing pages, once we have
1791 * the page lock we can determine safely if the page is beyond EOF. If it is not
1792 * beyond EOF, then the page is guaranteed safe against truncation until we
1793 * unlock the page.
1794 */
1795static vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
1796{
1797 struct page *page = vmf->page;
1798 struct folio *folio = page_folio(page);
1799 struct inode *inode = file_inode(vmf->vma->vm_file);
1800 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
1801 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1802 struct btrfs_ordered_extent *ordered;
1803 struct extent_state *cached_state = NULL;
1804 struct extent_changeset *data_reserved = NULL;
1805 unsigned long zero_start;
1806 loff_t size;
1807 vm_fault_t ret;
1808 int ret2;
1809 int reserved = 0;
1810 u64 reserved_space;
1811 u64 page_start;
1812 u64 page_end;
1813 u64 end;
1814
1815 ASSERT(folio_order(folio) == 0);
1816
1817 reserved_space = PAGE_SIZE;
1818
1819 sb_start_pagefault(inode->i_sb);
1820 page_start = folio_pos(folio);
1821 page_end = page_start + folio_size(folio) - 1;
1822 end = page_end;
1823
1824 /*
1825 * Reserving delalloc space after obtaining the page lock can lead to
1826 * deadlock. For example, if a dirty page is locked by this function
1827 * and the call to btrfs_delalloc_reserve_space() ends up triggering
1828 * dirty page write out, then the btrfs_writepages() function could
1829 * end up waiting indefinitely to get a lock on the page currently
1830 * being processed by btrfs_page_mkwrite() function.
1831 */
1832 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
1833 page_start, reserved_space);
1834 if (!ret2) {
1835 ret2 = file_update_time(vmf->vma->vm_file);
1836 reserved = 1;
1837 }
1838 if (ret2) {
1839 ret = vmf_error(ret2);
1840 if (reserved)
1841 goto out;
1842 goto out_noreserve;
1843 }
1844
1845 /* Make the VM retry the fault. */
1846 ret = VM_FAULT_NOPAGE;
1847again:
1848 down_read(&BTRFS_I(inode)->i_mmap_lock);
1849 folio_lock(folio);
1850 size = i_size_read(inode);
1851
1852 if ((folio->mapping != inode->i_mapping) ||
1853 (page_start >= size)) {
1854 /* Page got truncated out from underneath us. */
1855 goto out_unlock;
1856 }
1857 folio_wait_writeback(folio);
1858
1859 lock_extent(io_tree, page_start, page_end, &cached_state);
1860 ret2 = set_folio_extent_mapped(folio);
1861 if (ret2 < 0) {
1862 ret = vmf_error(ret2);
1863 unlock_extent(io_tree, page_start, page_end, &cached_state);
1864 goto out_unlock;
1865 }
1866
1867 /*
1868 * We can't set the delalloc bits if there are pending ordered
1869 * extents. Drop our locks and wait for them to finish.
1870 */
1871 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start, PAGE_SIZE);
1872 if (ordered) {
1873 unlock_extent(io_tree, page_start, page_end, &cached_state);
1874 folio_unlock(folio);
1875 up_read(&BTRFS_I(inode)->i_mmap_lock);
1876 btrfs_start_ordered_extent(ordered);
1877 btrfs_put_ordered_extent(ordered);
1878 goto again;
1879 }
1880
1881 if (folio->index == ((size - 1) >> PAGE_SHIFT)) {
1882 reserved_space = round_up(size - page_start, fs_info->sectorsize);
1883 if (reserved_space < PAGE_SIZE) {
1884 end = page_start + reserved_space - 1;
1885 btrfs_delalloc_release_space(BTRFS_I(inode),
1886 data_reserved, page_start,
1887 PAGE_SIZE - reserved_space, true);
1888 }
1889 }
1890
1891 /*
1892 * page_mkwrite gets called when the page is firstly dirtied after it's
1893 * faulted in, but write(2) could also dirty a page and set delalloc
1894 * bits, thus in this case for space account reason, we still need to
1895 * clear any delalloc bits within this page range since we have to
1896 * reserve data&meta space before lock_page() (see above comments).
1897 */
1898 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
1899 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
1900 EXTENT_DEFRAG, &cached_state);
1901
1902 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
1903 &cached_state);
1904 if (ret2) {
1905 unlock_extent(io_tree, page_start, page_end, &cached_state);
1906 ret = VM_FAULT_SIGBUS;
1907 goto out_unlock;
1908 }
1909
1910 /* Page is wholly or partially inside EOF. */
1911 if (page_start + folio_size(folio) > size)
1912 zero_start = offset_in_folio(folio, size);
1913 else
1914 zero_start = PAGE_SIZE;
1915
1916 if (zero_start != PAGE_SIZE)
1917 folio_zero_range(folio, zero_start, folio_size(folio) - zero_start);
1918
1919 btrfs_folio_clear_checked(fs_info, folio, page_start, PAGE_SIZE);
1920 btrfs_folio_set_dirty(fs_info, folio, page_start, end + 1 - page_start);
1921 btrfs_folio_set_uptodate(fs_info, folio, page_start, end + 1 - page_start);
1922
1923 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
1924
1925 unlock_extent(io_tree, page_start, page_end, &cached_state);
1926 up_read(&BTRFS_I(inode)->i_mmap_lock);
1927
1928 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
1929 sb_end_pagefault(inode->i_sb);
1930 extent_changeset_free(data_reserved);
1931 return VM_FAULT_LOCKED;
1932
1933out_unlock:
1934 folio_unlock(folio);
1935 up_read(&BTRFS_I(inode)->i_mmap_lock);
1936out:
1937 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
1938 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
1939 reserved_space, (ret != 0));
1940out_noreserve:
1941 sb_end_pagefault(inode->i_sb);
1942 extent_changeset_free(data_reserved);
1943 return ret;
1944}
1945
1946static const struct vm_operations_struct btrfs_file_vm_ops = {
1947 .fault = filemap_fault,
1948 .map_pages = filemap_map_pages,
1949 .page_mkwrite = btrfs_page_mkwrite,
1950};
1951
1952static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1953{
1954 struct address_space *mapping = filp->f_mapping;
1955
1956 if (!mapping->a_ops->read_folio)
1957 return -ENOEXEC;
1958
1959 file_accessed(filp);
1960 vma->vm_ops = &btrfs_file_vm_ops;
1961
1962 return 0;
1963}
1964
1965static int hole_mergeable(struct btrfs_inode *inode, struct extent_buffer *leaf,
1966 int slot, u64 start, u64 end)
1967{
1968 struct btrfs_file_extent_item *fi;
1969 struct btrfs_key key;
1970
1971 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1972 return 0;
1973
1974 btrfs_item_key_to_cpu(leaf, &key, slot);
1975 if (key.objectid != btrfs_ino(inode) ||
1976 key.type != BTRFS_EXTENT_DATA_KEY)
1977 return 0;
1978
1979 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1980
1981 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
1982 return 0;
1983
1984 if (btrfs_file_extent_disk_bytenr(leaf, fi))
1985 return 0;
1986
1987 if (key.offset == end)
1988 return 1;
1989 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
1990 return 1;
1991 return 0;
1992}
1993
1994static int fill_holes(struct btrfs_trans_handle *trans,
1995 struct btrfs_inode *inode,
1996 struct btrfs_path *path, u64 offset, u64 end)
1997{
1998 struct btrfs_fs_info *fs_info = trans->fs_info;
1999 struct btrfs_root *root = inode->root;
2000 struct extent_buffer *leaf;
2001 struct btrfs_file_extent_item *fi;
2002 struct extent_map *hole_em;
2003 struct btrfs_key key;
2004 int ret;
2005
2006 if (btrfs_fs_incompat(fs_info, NO_HOLES))
2007 goto out;
2008
2009 key.objectid = btrfs_ino(inode);
2010 key.type = BTRFS_EXTENT_DATA_KEY;
2011 key.offset = offset;
2012
2013 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2014 if (ret <= 0) {
2015 /*
2016 * We should have dropped this offset, so if we find it then
2017 * something has gone horribly wrong.
2018 */
2019 if (ret == 0)
2020 ret = -EINVAL;
2021 return ret;
2022 }
2023
2024 leaf = path->nodes[0];
2025 if (hole_mergeable(inode, leaf, path->slots[0] - 1, offset, end)) {
2026 u64 num_bytes;
2027
2028 path->slots[0]--;
2029 fi = btrfs_item_ptr(leaf, path->slots[0],
2030 struct btrfs_file_extent_item);
2031 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2032 end - offset;
2033 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2034 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2035 btrfs_set_file_extent_offset(leaf, fi, 0);
2036 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2037 btrfs_mark_buffer_dirty(trans, leaf);
2038 goto out;
2039 }
2040
2041 if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) {
2042 u64 num_bytes;
2043
2044 key.offset = offset;
2045 btrfs_set_item_key_safe(trans, path, &key);
2046 fi = btrfs_item_ptr(leaf, path->slots[0],
2047 struct btrfs_file_extent_item);
2048 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2049 offset;
2050 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2051 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2052 btrfs_set_file_extent_offset(leaf, fi, 0);
2053 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2054 btrfs_mark_buffer_dirty(trans, leaf);
2055 goto out;
2056 }
2057 btrfs_release_path(path);
2058
2059 ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset,
2060 end - offset);
2061 if (ret)
2062 return ret;
2063
2064out:
2065 btrfs_release_path(path);
2066
2067 hole_em = alloc_extent_map();
2068 if (!hole_em) {
2069 btrfs_drop_extent_map_range(inode, offset, end - 1, false);
2070 btrfs_set_inode_full_sync(inode);
2071 } else {
2072 hole_em->start = offset;
2073 hole_em->len = end - offset;
2074 hole_em->ram_bytes = hole_em->len;
2075
2076 hole_em->disk_bytenr = EXTENT_MAP_HOLE;
2077 hole_em->disk_num_bytes = 0;
2078 hole_em->generation = trans->transid;
2079
2080 ret = btrfs_replace_extent_map_range(inode, hole_em, true);
2081 free_extent_map(hole_em);
2082 if (ret)
2083 btrfs_set_inode_full_sync(inode);
2084 }
2085
2086 return 0;
2087}
2088
2089/*
2090 * Find a hole extent on given inode and change start/len to the end of hole
2091 * extent.(hole/vacuum extent whose em->start <= start &&
2092 * em->start + em->len > start)
2093 * When a hole extent is found, return 1 and modify start/len.
2094 */
2095static int find_first_non_hole(struct btrfs_inode *inode, u64 *start, u64 *len)
2096{
2097 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2098 struct extent_map *em;
2099 int ret = 0;
2100
2101 em = btrfs_get_extent(inode, NULL,
2102 round_down(*start, fs_info->sectorsize),
2103 round_up(*len, fs_info->sectorsize));
2104 if (IS_ERR(em))
2105 return PTR_ERR(em);
2106
2107 /* Hole or vacuum extent(only exists in no-hole mode) */
2108 if (em->disk_bytenr == EXTENT_MAP_HOLE) {
2109 ret = 1;
2110 *len = em->start + em->len > *start + *len ?
2111 0 : *start + *len - em->start - em->len;
2112 *start = em->start + em->len;
2113 }
2114 free_extent_map(em);
2115 return ret;
2116}
2117
2118static void btrfs_punch_hole_lock_range(struct inode *inode,
2119 const u64 lockstart,
2120 const u64 lockend,
2121 struct extent_state **cached_state)
2122{
2123 /*
2124 * For subpage case, if the range is not at page boundary, we could
2125 * have pages at the leading/tailing part of the range.
2126 * This could lead to dead loop since filemap_range_has_page()
2127 * will always return true.
2128 * So here we need to do extra page alignment for
2129 * filemap_range_has_page().
2130 */
2131 const u64 page_lockstart = round_up(lockstart, PAGE_SIZE);
2132 const u64 page_lockend = round_down(lockend + 1, PAGE_SIZE) - 1;
2133
2134 while (1) {
2135 truncate_pagecache_range(inode, lockstart, lockend);
2136
2137 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2138 cached_state);
2139 /*
2140 * We can't have ordered extents in the range, nor dirty/writeback
2141 * pages, because we have locked the inode's VFS lock in exclusive
2142 * mode, we have locked the inode's i_mmap_lock in exclusive mode,
2143 * we have flushed all delalloc in the range and we have waited
2144 * for any ordered extents in the range to complete.
2145 * We can race with anyone reading pages from this range, so after
2146 * locking the range check if we have pages in the range, and if
2147 * we do, unlock the range and retry.
2148 */
2149 if (!filemap_range_has_page(inode->i_mapping, page_lockstart,
2150 page_lockend))
2151 break;
2152
2153 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2154 cached_state);
2155 }
2156
2157 btrfs_assert_inode_range_clean(BTRFS_I(inode), lockstart, lockend);
2158}
2159
2160static int btrfs_insert_replace_extent(struct btrfs_trans_handle *trans,
2161 struct btrfs_inode *inode,
2162 struct btrfs_path *path,
2163 struct btrfs_replace_extent_info *extent_info,
2164 const u64 replace_len,
2165 const u64 bytes_to_drop)
2166{
2167 struct btrfs_fs_info *fs_info = trans->fs_info;
2168 struct btrfs_root *root = inode->root;
2169 struct btrfs_file_extent_item *extent;
2170 struct extent_buffer *leaf;
2171 struct btrfs_key key;
2172 int slot;
2173 int ret;
2174
2175 if (replace_len == 0)
2176 return 0;
2177
2178 if (extent_info->disk_offset == 0 &&
2179 btrfs_fs_incompat(fs_info, NO_HOLES)) {
2180 btrfs_update_inode_bytes(inode, 0, bytes_to_drop);
2181 return 0;
2182 }
2183
2184 key.objectid = btrfs_ino(inode);
2185 key.type = BTRFS_EXTENT_DATA_KEY;
2186 key.offset = extent_info->file_offset;
2187 ret = btrfs_insert_empty_item(trans, root, path, &key,
2188 sizeof(struct btrfs_file_extent_item));
2189 if (ret)
2190 return ret;
2191 leaf = path->nodes[0];
2192 slot = path->slots[0];
2193 write_extent_buffer(leaf, extent_info->extent_buf,
2194 btrfs_item_ptr_offset(leaf, slot),
2195 sizeof(struct btrfs_file_extent_item));
2196 extent = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2197 ASSERT(btrfs_file_extent_type(leaf, extent) != BTRFS_FILE_EXTENT_INLINE);
2198 btrfs_set_file_extent_offset(leaf, extent, extent_info->data_offset);
2199 btrfs_set_file_extent_num_bytes(leaf, extent, replace_len);
2200 if (extent_info->is_new_extent)
2201 btrfs_set_file_extent_generation(leaf, extent, trans->transid);
2202 btrfs_mark_buffer_dirty(trans, leaf);
2203 btrfs_release_path(path);
2204
2205 ret = btrfs_inode_set_file_extent_range(inode, extent_info->file_offset,
2206 replace_len);
2207 if (ret)
2208 return ret;
2209
2210 /* If it's a hole, nothing more needs to be done. */
2211 if (extent_info->disk_offset == 0) {
2212 btrfs_update_inode_bytes(inode, 0, bytes_to_drop);
2213 return 0;
2214 }
2215
2216 btrfs_update_inode_bytes(inode, replace_len, bytes_to_drop);
2217
2218 if (extent_info->is_new_extent && extent_info->insertions == 0) {
2219 key.objectid = extent_info->disk_offset;
2220 key.type = BTRFS_EXTENT_ITEM_KEY;
2221 key.offset = extent_info->disk_len;
2222 ret = btrfs_alloc_reserved_file_extent(trans, root,
2223 btrfs_ino(inode),
2224 extent_info->file_offset,
2225 extent_info->qgroup_reserved,
2226 &key);
2227 } else {
2228 struct btrfs_ref ref = {
2229 .action = BTRFS_ADD_DELAYED_REF,
2230 .bytenr = extent_info->disk_offset,
2231 .num_bytes = extent_info->disk_len,
2232 .owning_root = btrfs_root_id(root),
2233 .ref_root = btrfs_root_id(root),
2234 };
2235 u64 ref_offset;
2236
2237 ref_offset = extent_info->file_offset - extent_info->data_offset;
2238 btrfs_init_data_ref(&ref, btrfs_ino(inode), ref_offset, 0, false);
2239 ret = btrfs_inc_extent_ref(trans, &ref);
2240 }
2241
2242 extent_info->insertions++;
2243
2244 return ret;
2245}
2246
2247/*
2248 * The respective range must have been previously locked, as well as the inode.
2249 * The end offset is inclusive (last byte of the range).
2250 * @extent_info is NULL for fallocate's hole punching and non-NULL when replacing
2251 * the file range with an extent.
2252 * When not punching a hole, we don't want to end up in a state where we dropped
2253 * extents without inserting a new one, so we must abort the transaction to avoid
2254 * a corruption.
2255 */
2256int btrfs_replace_file_extents(struct btrfs_inode *inode,
2257 struct btrfs_path *path, const u64 start,
2258 const u64 end,
2259 struct btrfs_replace_extent_info *extent_info,
2260 struct btrfs_trans_handle **trans_out)
2261{
2262 struct btrfs_drop_extents_args drop_args = { 0 };
2263 struct btrfs_root *root = inode->root;
2264 struct btrfs_fs_info *fs_info = root->fs_info;
2265 u64 min_size = btrfs_calc_insert_metadata_size(fs_info, 1);
2266 u64 ino_size = round_up(inode->vfs_inode.i_size, fs_info->sectorsize);
2267 struct btrfs_trans_handle *trans = NULL;
2268 struct btrfs_block_rsv *rsv;
2269 unsigned int rsv_count;
2270 u64 cur_offset;
2271 u64 len = end - start;
2272 int ret = 0;
2273
2274 if (end <= start)
2275 return -EINVAL;
2276
2277 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
2278 if (!rsv) {
2279 ret = -ENOMEM;
2280 goto out;
2281 }
2282 rsv->size = btrfs_calc_insert_metadata_size(fs_info, 1);
2283 rsv->failfast = true;
2284
2285 /*
2286 * 1 - update the inode
2287 * 1 - removing the extents in the range
2288 * 1 - adding the hole extent if no_holes isn't set or if we are
2289 * replacing the range with a new extent
2290 */
2291 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || extent_info)
2292 rsv_count = 3;
2293 else
2294 rsv_count = 2;
2295
2296 trans = btrfs_start_transaction(root, rsv_count);
2297 if (IS_ERR(trans)) {
2298 ret = PTR_ERR(trans);
2299 trans = NULL;
2300 goto out_free;
2301 }
2302
2303 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
2304 min_size, false);
2305 if (WARN_ON(ret))
2306 goto out_trans;
2307 trans->block_rsv = rsv;
2308
2309 cur_offset = start;
2310 drop_args.path = path;
2311 drop_args.end = end + 1;
2312 drop_args.drop_cache = true;
2313 while (cur_offset < end) {
2314 drop_args.start = cur_offset;
2315 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2316 /* If we are punching a hole decrement the inode's byte count */
2317 if (!extent_info)
2318 btrfs_update_inode_bytes(inode, 0,
2319 drop_args.bytes_found);
2320 if (ret != -ENOSPC) {
2321 /*
2322 * The only time we don't want to abort is if we are
2323 * attempting to clone a partial inline extent, in which
2324 * case we'll get EOPNOTSUPP. However if we aren't
2325 * clone we need to abort no matter what, because if we
2326 * got EOPNOTSUPP via prealloc then we messed up and
2327 * need to abort.
2328 */
2329 if (ret &&
2330 (ret != -EOPNOTSUPP ||
2331 (extent_info && extent_info->is_new_extent)))
2332 btrfs_abort_transaction(trans, ret);
2333 break;
2334 }
2335
2336 trans->block_rsv = &fs_info->trans_block_rsv;
2337
2338 if (!extent_info && cur_offset < drop_args.drop_end &&
2339 cur_offset < ino_size) {
2340 ret = fill_holes(trans, inode, path, cur_offset,
2341 drop_args.drop_end);
2342 if (ret) {
2343 /*
2344 * If we failed then we didn't insert our hole
2345 * entries for the area we dropped, so now the
2346 * fs is corrupted, so we must abort the
2347 * transaction.
2348 */
2349 btrfs_abort_transaction(trans, ret);
2350 break;
2351 }
2352 } else if (!extent_info && cur_offset < drop_args.drop_end) {
2353 /*
2354 * We are past the i_size here, but since we didn't
2355 * insert holes we need to clear the mapped area so we
2356 * know to not set disk_i_size in this area until a new
2357 * file extent is inserted here.
2358 */
2359 ret = btrfs_inode_clear_file_extent_range(inode,
2360 cur_offset,
2361 drop_args.drop_end - cur_offset);
2362 if (ret) {
2363 /*
2364 * We couldn't clear our area, so we could
2365 * presumably adjust up and corrupt the fs, so
2366 * we need to abort.
2367 */
2368 btrfs_abort_transaction(trans, ret);
2369 break;
2370 }
2371 }
2372
2373 if (extent_info &&
2374 drop_args.drop_end > extent_info->file_offset) {
2375 u64 replace_len = drop_args.drop_end -
2376 extent_info->file_offset;
2377
2378 ret = btrfs_insert_replace_extent(trans, inode, path,
2379 extent_info, replace_len,
2380 drop_args.bytes_found);
2381 if (ret) {
2382 btrfs_abort_transaction(trans, ret);
2383 break;
2384 }
2385 extent_info->data_len -= replace_len;
2386 extent_info->data_offset += replace_len;
2387 extent_info->file_offset += replace_len;
2388 }
2389
2390 /*
2391 * We are releasing our handle on the transaction, balance the
2392 * dirty pages of the btree inode and flush delayed items, and
2393 * then get a new transaction handle, which may now point to a
2394 * new transaction in case someone else may have committed the
2395 * transaction we used to replace/drop file extent items. So
2396 * bump the inode's iversion and update mtime and ctime except
2397 * if we are called from a dedupe context. This is because a
2398 * power failure/crash may happen after the transaction is
2399 * committed and before we finish replacing/dropping all the
2400 * file extent items we need.
2401 */
2402 inode_inc_iversion(&inode->vfs_inode);
2403
2404 if (!extent_info || extent_info->update_times)
2405 inode_set_mtime_to_ts(&inode->vfs_inode,
2406 inode_set_ctime_current(&inode->vfs_inode));
2407
2408 ret = btrfs_update_inode(trans, inode);
2409 if (ret)
2410 break;
2411
2412 btrfs_end_transaction(trans);
2413 btrfs_btree_balance_dirty(fs_info);
2414
2415 trans = btrfs_start_transaction(root, rsv_count);
2416 if (IS_ERR(trans)) {
2417 ret = PTR_ERR(trans);
2418 trans = NULL;
2419 break;
2420 }
2421
2422 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
2423 rsv, min_size, false);
2424 if (WARN_ON(ret))
2425 break;
2426 trans->block_rsv = rsv;
2427
2428 cur_offset = drop_args.drop_end;
2429 len = end - cur_offset;
2430 if (!extent_info && len) {
2431 ret = find_first_non_hole(inode, &cur_offset, &len);
2432 if (unlikely(ret < 0))
2433 break;
2434 if (ret && !len) {
2435 ret = 0;
2436 break;
2437 }
2438 }
2439 }
2440
2441 /*
2442 * If we were cloning, force the next fsync to be a full one since we
2443 * we replaced (or just dropped in the case of cloning holes when
2444 * NO_HOLES is enabled) file extent items and did not setup new extent
2445 * maps for the replacement extents (or holes).
2446 */
2447 if (extent_info && !extent_info->is_new_extent)
2448 btrfs_set_inode_full_sync(inode);
2449
2450 if (ret)
2451 goto out_trans;
2452
2453 trans->block_rsv = &fs_info->trans_block_rsv;
2454 /*
2455 * If we are using the NO_HOLES feature we might have had already an
2456 * hole that overlaps a part of the region [lockstart, lockend] and
2457 * ends at (or beyond) lockend. Since we have no file extent items to
2458 * represent holes, drop_end can be less than lockend and so we must
2459 * make sure we have an extent map representing the existing hole (the
2460 * call to __btrfs_drop_extents() might have dropped the existing extent
2461 * map representing the existing hole), otherwise the fast fsync path
2462 * will not record the existence of the hole region
2463 * [existing_hole_start, lockend].
2464 */
2465 if (drop_args.drop_end <= end)
2466 drop_args.drop_end = end + 1;
2467 /*
2468 * Don't insert file hole extent item if it's for a range beyond eof
2469 * (because it's useless) or if it represents a 0 bytes range (when
2470 * cur_offset == drop_end).
2471 */
2472 if (!extent_info && cur_offset < ino_size &&
2473 cur_offset < drop_args.drop_end) {
2474 ret = fill_holes(trans, inode, path, cur_offset,
2475 drop_args.drop_end);
2476 if (ret) {
2477 /* Same comment as above. */
2478 btrfs_abort_transaction(trans, ret);
2479 goto out_trans;
2480 }
2481 } else if (!extent_info && cur_offset < drop_args.drop_end) {
2482 /* See the comment in the loop above for the reasoning here. */
2483 ret = btrfs_inode_clear_file_extent_range(inode, cur_offset,
2484 drop_args.drop_end - cur_offset);
2485 if (ret) {
2486 btrfs_abort_transaction(trans, ret);
2487 goto out_trans;
2488 }
2489
2490 }
2491 if (extent_info) {
2492 ret = btrfs_insert_replace_extent(trans, inode, path,
2493 extent_info, extent_info->data_len,
2494 drop_args.bytes_found);
2495 if (ret) {
2496 btrfs_abort_transaction(trans, ret);
2497 goto out_trans;
2498 }
2499 }
2500
2501out_trans:
2502 if (!trans)
2503 goto out_free;
2504
2505 trans->block_rsv = &fs_info->trans_block_rsv;
2506 if (ret)
2507 btrfs_end_transaction(trans);
2508 else
2509 *trans_out = trans;
2510out_free:
2511 btrfs_free_block_rsv(fs_info, rsv);
2512out:
2513 return ret;
2514}
2515
2516static int btrfs_punch_hole(struct file *file, loff_t offset, loff_t len)
2517{
2518 struct inode *inode = file_inode(file);
2519 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
2520 struct btrfs_root *root = BTRFS_I(inode)->root;
2521 struct extent_state *cached_state = NULL;
2522 struct btrfs_path *path;
2523 struct btrfs_trans_handle *trans = NULL;
2524 u64 lockstart;
2525 u64 lockend;
2526 u64 tail_start;
2527 u64 tail_len;
2528 u64 orig_start = offset;
2529 int ret = 0;
2530 bool same_block;
2531 u64 ino_size;
2532 bool truncated_block = false;
2533 bool updated_inode = false;
2534
2535 btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
2536
2537 ret = btrfs_wait_ordered_range(BTRFS_I(inode), offset, len);
2538 if (ret)
2539 goto out_only_mutex;
2540
2541 ino_size = round_up(inode->i_size, fs_info->sectorsize);
2542 ret = find_first_non_hole(BTRFS_I(inode), &offset, &len);
2543 if (ret < 0)
2544 goto out_only_mutex;
2545 if (ret && !len) {
2546 /* Already in a large hole */
2547 ret = 0;
2548 goto out_only_mutex;
2549 }
2550
2551 ret = file_modified(file);
2552 if (ret)
2553 goto out_only_mutex;
2554
2555 lockstart = round_up(offset, fs_info->sectorsize);
2556 lockend = round_down(offset + len, fs_info->sectorsize) - 1;
2557 same_block = (BTRFS_BYTES_TO_BLKS(fs_info, offset))
2558 == (BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1));
2559 /*
2560 * We needn't truncate any block which is beyond the end of the file
2561 * because we are sure there is no data there.
2562 */
2563 /*
2564 * Only do this if we are in the same block and we aren't doing the
2565 * entire block.
2566 */
2567 if (same_block && len < fs_info->sectorsize) {
2568 if (offset < ino_size) {
2569 truncated_block = true;
2570 ret = btrfs_truncate_block(BTRFS_I(inode), offset, len,
2571 0);
2572 } else {
2573 ret = 0;
2574 }
2575 goto out_only_mutex;
2576 }
2577
2578 /* zero back part of the first block */
2579 if (offset < ino_size) {
2580 truncated_block = true;
2581 ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0);
2582 if (ret) {
2583 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
2584 return ret;
2585 }
2586 }
2587
2588 /* Check the aligned pages after the first unaligned page,
2589 * if offset != orig_start, which means the first unaligned page
2590 * including several following pages are already in holes,
2591 * the extra check can be skipped */
2592 if (offset == orig_start) {
2593 /* after truncate page, check hole again */
2594 len = offset + len - lockstart;
2595 offset = lockstart;
2596 ret = find_first_non_hole(BTRFS_I(inode), &offset, &len);
2597 if (ret < 0)
2598 goto out_only_mutex;
2599 if (ret && !len) {
2600 ret = 0;
2601 goto out_only_mutex;
2602 }
2603 lockstart = offset;
2604 }
2605
2606 /* Check the tail unaligned part is in a hole */
2607 tail_start = lockend + 1;
2608 tail_len = offset + len - tail_start;
2609 if (tail_len) {
2610 ret = find_first_non_hole(BTRFS_I(inode), &tail_start, &tail_len);
2611 if (unlikely(ret < 0))
2612 goto out_only_mutex;
2613 if (!ret) {
2614 /* zero the front end of the last page */
2615 if (tail_start + tail_len < ino_size) {
2616 truncated_block = true;
2617 ret = btrfs_truncate_block(BTRFS_I(inode),
2618 tail_start + tail_len,
2619 0, 1);
2620 if (ret)
2621 goto out_only_mutex;
2622 }
2623 }
2624 }
2625
2626 if (lockend < lockstart) {
2627 ret = 0;
2628 goto out_only_mutex;
2629 }
2630
2631 btrfs_punch_hole_lock_range(inode, lockstart, lockend, &cached_state);
2632
2633 path = btrfs_alloc_path();
2634 if (!path) {
2635 ret = -ENOMEM;
2636 goto out;
2637 }
2638
2639 ret = btrfs_replace_file_extents(BTRFS_I(inode), path, lockstart,
2640 lockend, NULL, &trans);
2641 btrfs_free_path(path);
2642 if (ret)
2643 goto out;
2644
2645 ASSERT(trans != NULL);
2646 inode_inc_iversion(inode);
2647 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
2648 ret = btrfs_update_inode(trans, BTRFS_I(inode));
2649 updated_inode = true;
2650 btrfs_end_transaction(trans);
2651 btrfs_btree_balance_dirty(fs_info);
2652out:
2653 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2654 &cached_state);
2655out_only_mutex:
2656 if (!updated_inode && truncated_block && !ret) {
2657 /*
2658 * If we only end up zeroing part of a page, we still need to
2659 * update the inode item, so that all the time fields are
2660 * updated as well as the necessary btrfs inode in memory fields
2661 * for detecting, at fsync time, if the inode isn't yet in the
2662 * log tree or it's there but not up to date.
2663 */
2664 struct timespec64 now = inode_set_ctime_current(inode);
2665
2666 inode_inc_iversion(inode);
2667 inode_set_mtime_to_ts(inode, now);
2668 trans = btrfs_start_transaction(root, 1);
2669 if (IS_ERR(trans)) {
2670 ret = PTR_ERR(trans);
2671 } else {
2672 int ret2;
2673
2674 ret = btrfs_update_inode(trans, BTRFS_I(inode));
2675 ret2 = btrfs_end_transaction(trans);
2676 if (!ret)
2677 ret = ret2;
2678 }
2679 }
2680 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
2681 return ret;
2682}
2683
2684/* Helper structure to record which range is already reserved */
2685struct falloc_range {
2686 struct list_head list;
2687 u64 start;
2688 u64 len;
2689};
2690
2691/*
2692 * Helper function to add falloc range
2693 *
2694 * Caller should have locked the larger range of extent containing
2695 * [start, len)
2696 */
2697static int add_falloc_range(struct list_head *head, u64 start, u64 len)
2698{
2699 struct falloc_range *range = NULL;
2700
2701 if (!list_empty(head)) {
2702 /*
2703 * As fallocate iterates by bytenr order, we only need to check
2704 * the last range.
2705 */
2706 range = list_last_entry(head, struct falloc_range, list);
2707 if (range->start + range->len == start) {
2708 range->len += len;
2709 return 0;
2710 }
2711 }
2712
2713 range = kmalloc(sizeof(*range), GFP_KERNEL);
2714 if (!range)
2715 return -ENOMEM;
2716 range->start = start;
2717 range->len = len;
2718 list_add_tail(&range->list, head);
2719 return 0;
2720}
2721
2722static int btrfs_fallocate_update_isize(struct inode *inode,
2723 const u64 end,
2724 const int mode)
2725{
2726 struct btrfs_trans_handle *trans;
2727 struct btrfs_root *root = BTRFS_I(inode)->root;
2728 int ret;
2729 int ret2;
2730
2731 if (mode & FALLOC_FL_KEEP_SIZE || end <= i_size_read(inode))
2732 return 0;
2733
2734 trans = btrfs_start_transaction(root, 1);
2735 if (IS_ERR(trans))
2736 return PTR_ERR(trans);
2737
2738 inode_set_ctime_current(inode);
2739 i_size_write(inode, end);
2740 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
2741 ret = btrfs_update_inode(trans, BTRFS_I(inode));
2742 ret2 = btrfs_end_transaction(trans);
2743
2744 return ret ? ret : ret2;
2745}
2746
2747enum {
2748 RANGE_BOUNDARY_WRITTEN_EXTENT,
2749 RANGE_BOUNDARY_PREALLOC_EXTENT,
2750 RANGE_BOUNDARY_HOLE,
2751};
2752
2753static int btrfs_zero_range_check_range_boundary(struct btrfs_inode *inode,
2754 u64 offset)
2755{
2756 const u64 sectorsize = inode->root->fs_info->sectorsize;
2757 struct extent_map *em;
2758 int ret;
2759
2760 offset = round_down(offset, sectorsize);
2761 em = btrfs_get_extent(inode, NULL, offset, sectorsize);
2762 if (IS_ERR(em))
2763 return PTR_ERR(em);
2764
2765 if (em->disk_bytenr == EXTENT_MAP_HOLE)
2766 ret = RANGE_BOUNDARY_HOLE;
2767 else if (em->flags & EXTENT_FLAG_PREALLOC)
2768 ret = RANGE_BOUNDARY_PREALLOC_EXTENT;
2769 else
2770 ret = RANGE_BOUNDARY_WRITTEN_EXTENT;
2771
2772 free_extent_map(em);
2773 return ret;
2774}
2775
2776static int btrfs_zero_range(struct inode *inode,
2777 loff_t offset,
2778 loff_t len,
2779 const int mode)
2780{
2781 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2782 struct extent_map *em;
2783 struct extent_changeset *data_reserved = NULL;
2784 int ret;
2785 u64 alloc_hint = 0;
2786 const u64 sectorsize = fs_info->sectorsize;
2787 u64 alloc_start = round_down(offset, sectorsize);
2788 u64 alloc_end = round_up(offset + len, sectorsize);
2789 u64 bytes_to_reserve = 0;
2790 bool space_reserved = false;
2791
2792 em = btrfs_get_extent(BTRFS_I(inode), NULL, alloc_start,
2793 alloc_end - alloc_start);
2794 if (IS_ERR(em)) {
2795 ret = PTR_ERR(em);
2796 goto out;
2797 }
2798
2799 /*
2800 * Avoid hole punching and extent allocation for some cases. More cases
2801 * could be considered, but these are unlikely common and we keep things
2802 * as simple as possible for now. Also, intentionally, if the target
2803 * range contains one or more prealloc extents together with regular
2804 * extents and holes, we drop all the existing extents and allocate a
2805 * new prealloc extent, so that we get a larger contiguous disk extent.
2806 */
2807 if (em->start <= alloc_start && (em->flags & EXTENT_FLAG_PREALLOC)) {
2808 const u64 em_end = em->start + em->len;
2809
2810 if (em_end >= offset + len) {
2811 /*
2812 * The whole range is already a prealloc extent,
2813 * do nothing except updating the inode's i_size if
2814 * needed.
2815 */
2816 free_extent_map(em);
2817 ret = btrfs_fallocate_update_isize(inode, offset + len,
2818 mode);
2819 goto out;
2820 }
2821 /*
2822 * Part of the range is already a prealloc extent, so operate
2823 * only on the remaining part of the range.
2824 */
2825 alloc_start = em_end;
2826 ASSERT(IS_ALIGNED(alloc_start, sectorsize));
2827 len = offset + len - alloc_start;
2828 offset = alloc_start;
2829 alloc_hint = extent_map_block_start(em) + em->len;
2830 }
2831 free_extent_map(em);
2832
2833 if (BTRFS_BYTES_TO_BLKS(fs_info, offset) ==
2834 BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)) {
2835 em = btrfs_get_extent(BTRFS_I(inode), NULL, alloc_start, sectorsize);
2836 if (IS_ERR(em)) {
2837 ret = PTR_ERR(em);
2838 goto out;
2839 }
2840
2841 if (em->flags & EXTENT_FLAG_PREALLOC) {
2842 free_extent_map(em);
2843 ret = btrfs_fallocate_update_isize(inode, offset + len,
2844 mode);
2845 goto out;
2846 }
2847 if (len < sectorsize && em->disk_bytenr != EXTENT_MAP_HOLE) {
2848 free_extent_map(em);
2849 ret = btrfs_truncate_block(BTRFS_I(inode), offset, len,
2850 0);
2851 if (!ret)
2852 ret = btrfs_fallocate_update_isize(inode,
2853 offset + len,
2854 mode);
2855 return ret;
2856 }
2857 free_extent_map(em);
2858 alloc_start = round_down(offset, sectorsize);
2859 alloc_end = alloc_start + sectorsize;
2860 goto reserve_space;
2861 }
2862
2863 alloc_start = round_up(offset, sectorsize);
2864 alloc_end = round_down(offset + len, sectorsize);
2865
2866 /*
2867 * For unaligned ranges, check the pages at the boundaries, they might
2868 * map to an extent, in which case we need to partially zero them, or
2869 * they might map to a hole, in which case we need our allocation range
2870 * to cover them.
2871 */
2872 if (!IS_ALIGNED(offset, sectorsize)) {
2873 ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode),
2874 offset);
2875 if (ret < 0)
2876 goto out;
2877 if (ret == RANGE_BOUNDARY_HOLE) {
2878 alloc_start = round_down(offset, sectorsize);
2879 ret = 0;
2880 } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
2881 ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0);
2882 if (ret)
2883 goto out;
2884 } else {
2885 ret = 0;
2886 }
2887 }
2888
2889 if (!IS_ALIGNED(offset + len, sectorsize)) {
2890 ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode),
2891 offset + len);
2892 if (ret < 0)
2893 goto out;
2894 if (ret == RANGE_BOUNDARY_HOLE) {
2895 alloc_end = round_up(offset + len, sectorsize);
2896 ret = 0;
2897 } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
2898 ret = btrfs_truncate_block(BTRFS_I(inode), offset + len,
2899 0, 1);
2900 if (ret)
2901 goto out;
2902 } else {
2903 ret = 0;
2904 }
2905 }
2906
2907reserve_space:
2908 if (alloc_start < alloc_end) {
2909 struct extent_state *cached_state = NULL;
2910 const u64 lockstart = alloc_start;
2911 const u64 lockend = alloc_end - 1;
2912
2913 bytes_to_reserve = alloc_end - alloc_start;
2914 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
2915 bytes_to_reserve);
2916 if (ret < 0)
2917 goto out;
2918 space_reserved = true;
2919 btrfs_punch_hole_lock_range(inode, lockstart, lockend,
2920 &cached_state);
2921 ret = btrfs_qgroup_reserve_data(BTRFS_I(inode), &data_reserved,
2922 alloc_start, bytes_to_reserve);
2923 if (ret) {
2924 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart,
2925 lockend, &cached_state);
2926 goto out;
2927 }
2928 ret = btrfs_prealloc_file_range(inode, mode, alloc_start,
2929 alloc_end - alloc_start,
2930 fs_info->sectorsize,
2931 offset + len, &alloc_hint);
2932 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2933 &cached_state);
2934 /* btrfs_prealloc_file_range releases reserved space on error */
2935 if (ret) {
2936 space_reserved = false;
2937 goto out;
2938 }
2939 }
2940 ret = btrfs_fallocate_update_isize(inode, offset + len, mode);
2941 out:
2942 if (ret && space_reserved)
2943 btrfs_free_reserved_data_space(BTRFS_I(inode), data_reserved,
2944 alloc_start, bytes_to_reserve);
2945 extent_changeset_free(data_reserved);
2946
2947 return ret;
2948}
2949
2950static long btrfs_fallocate(struct file *file, int mode,
2951 loff_t offset, loff_t len)
2952{
2953 struct inode *inode = file_inode(file);
2954 struct extent_state *cached_state = NULL;
2955 struct extent_changeset *data_reserved = NULL;
2956 struct falloc_range *range;
2957 struct falloc_range *tmp;
2958 LIST_HEAD(reserve_list);
2959 u64 cur_offset;
2960 u64 last_byte;
2961 u64 alloc_start;
2962 u64 alloc_end;
2963 u64 alloc_hint = 0;
2964 u64 locked_end;
2965 u64 actual_end = 0;
2966 u64 data_space_needed = 0;
2967 u64 data_space_reserved = 0;
2968 u64 qgroup_reserved = 0;
2969 struct extent_map *em;
2970 int blocksize = BTRFS_I(inode)->root->fs_info->sectorsize;
2971 int ret;
2972
2973 /* Do not allow fallocate in ZONED mode */
2974 if (btrfs_is_zoned(inode_to_fs_info(inode)))
2975 return -EOPNOTSUPP;
2976
2977 alloc_start = round_down(offset, blocksize);
2978 alloc_end = round_up(offset + len, blocksize);
2979 cur_offset = alloc_start;
2980
2981 /* Make sure we aren't being give some crap mode */
2982 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
2983 FALLOC_FL_ZERO_RANGE))
2984 return -EOPNOTSUPP;
2985
2986 if (mode & FALLOC_FL_PUNCH_HOLE)
2987 return btrfs_punch_hole(file, offset, len);
2988
2989 btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
2990
2991 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) {
2992 ret = inode_newsize_ok(inode, offset + len);
2993 if (ret)
2994 goto out;
2995 }
2996
2997 ret = file_modified(file);
2998 if (ret)
2999 goto out;
3000
3001 /*
3002 * TODO: Move these two operations after we have checked
3003 * accurate reserved space, or fallocate can still fail but
3004 * with page truncated or size expanded.
3005 *
3006 * But that's a minor problem and won't do much harm BTW.
3007 */
3008 if (alloc_start > inode->i_size) {
3009 ret = btrfs_cont_expand(BTRFS_I(inode), i_size_read(inode),
3010 alloc_start);
3011 if (ret)
3012 goto out;
3013 } else if (offset + len > inode->i_size) {
3014 /*
3015 * If we are fallocating from the end of the file onward we
3016 * need to zero out the end of the block if i_size lands in the
3017 * middle of a block.
3018 */
3019 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
3020 if (ret)
3021 goto out;
3022 }
3023
3024 /*
3025 * We have locked the inode at the VFS level (in exclusive mode) and we
3026 * have locked the i_mmap_lock lock (in exclusive mode). Now before
3027 * locking the file range, flush all dealloc in the range and wait for
3028 * all ordered extents in the range to complete. After this we can lock
3029 * the file range and, due to the previous locking we did, we know there
3030 * can't be more delalloc or ordered extents in the range.
3031 */
3032 ret = btrfs_wait_ordered_range(BTRFS_I(inode), alloc_start,
3033 alloc_end - alloc_start);
3034 if (ret)
3035 goto out;
3036
3037 if (mode & FALLOC_FL_ZERO_RANGE) {
3038 ret = btrfs_zero_range(inode, offset, len, mode);
3039 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
3040 return ret;
3041 }
3042
3043 locked_end = alloc_end - 1;
3044 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
3045 &cached_state);
3046
3047 btrfs_assert_inode_range_clean(BTRFS_I(inode), alloc_start, locked_end);
3048
3049 /* First, check if we exceed the qgroup limit */
3050 while (cur_offset < alloc_end) {
3051 em = btrfs_get_extent(BTRFS_I(inode), NULL, cur_offset,
3052 alloc_end - cur_offset);
3053 if (IS_ERR(em)) {
3054 ret = PTR_ERR(em);
3055 break;
3056 }
3057 last_byte = min(extent_map_end(em), alloc_end);
3058 actual_end = min_t(u64, extent_map_end(em), offset + len);
3059 last_byte = ALIGN(last_byte, blocksize);
3060 if (em->disk_bytenr == EXTENT_MAP_HOLE ||
3061 (cur_offset >= inode->i_size &&
3062 !(em->flags & EXTENT_FLAG_PREALLOC))) {
3063 const u64 range_len = last_byte - cur_offset;
3064
3065 ret = add_falloc_range(&reserve_list, cur_offset, range_len);
3066 if (ret < 0) {
3067 free_extent_map(em);
3068 break;
3069 }
3070 ret = btrfs_qgroup_reserve_data(BTRFS_I(inode),
3071 &data_reserved, cur_offset, range_len);
3072 if (ret < 0) {
3073 free_extent_map(em);
3074 break;
3075 }
3076 qgroup_reserved += range_len;
3077 data_space_needed += range_len;
3078 }
3079 free_extent_map(em);
3080 cur_offset = last_byte;
3081 }
3082
3083 if (!ret && data_space_needed > 0) {
3084 /*
3085 * We are safe to reserve space here as we can't have delalloc
3086 * in the range, see above.
3087 */
3088 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
3089 data_space_needed);
3090 if (!ret)
3091 data_space_reserved = data_space_needed;
3092 }
3093
3094 /*
3095 * If ret is still 0, means we're OK to fallocate.
3096 * Or just cleanup the list and exit.
3097 */
3098 list_for_each_entry_safe(range, tmp, &reserve_list, list) {
3099 if (!ret) {
3100 ret = btrfs_prealloc_file_range(inode, mode,
3101 range->start,
3102 range->len, blocksize,
3103 offset + len, &alloc_hint);
3104 /*
3105 * btrfs_prealloc_file_range() releases space even
3106 * if it returns an error.
3107 */
3108 data_space_reserved -= range->len;
3109 qgroup_reserved -= range->len;
3110 } else if (data_space_reserved > 0) {
3111 btrfs_free_reserved_data_space(BTRFS_I(inode),
3112 data_reserved, range->start,
3113 range->len);
3114 data_space_reserved -= range->len;
3115 qgroup_reserved -= range->len;
3116 } else if (qgroup_reserved > 0) {
3117 btrfs_qgroup_free_data(BTRFS_I(inode), data_reserved,
3118 range->start, range->len, NULL);
3119 qgroup_reserved -= range->len;
3120 }
3121 list_del(&range->list);
3122 kfree(range);
3123 }
3124 if (ret < 0)
3125 goto out_unlock;
3126
3127 /*
3128 * We didn't need to allocate any more space, but we still extended the
3129 * size of the file so we need to update i_size and the inode item.
3130 */
3131 ret = btrfs_fallocate_update_isize(inode, actual_end, mode);
3132out_unlock:
3133 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
3134 &cached_state);
3135out:
3136 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
3137 extent_changeset_free(data_reserved);
3138 return ret;
3139}
3140
3141/*
3142 * Helper for btrfs_find_delalloc_in_range(). Find a subrange in a given range
3143 * that has unflushed and/or flushing delalloc. There might be other adjacent
3144 * subranges after the one it found, so btrfs_find_delalloc_in_range() keeps
3145 * looping while it gets adjacent subranges, and merging them together.
3146 */
3147static bool find_delalloc_subrange(struct btrfs_inode *inode, u64 start, u64 end,
3148 struct extent_state **cached_state,
3149 bool *search_io_tree,
3150 u64 *delalloc_start_ret, u64 *delalloc_end_ret)
3151{
3152 u64 len = end + 1 - start;
3153 u64 delalloc_len = 0;
3154 struct btrfs_ordered_extent *oe;
3155 u64 oe_start;
3156 u64 oe_end;
3157
3158 /*
3159 * Search the io tree first for EXTENT_DELALLOC. If we find any, it
3160 * means we have delalloc (dirty pages) for which writeback has not
3161 * started yet.
3162 */
3163 if (*search_io_tree) {
3164 spin_lock(&inode->lock);
3165 if (inode->delalloc_bytes > 0) {
3166 spin_unlock(&inode->lock);
3167 *delalloc_start_ret = start;
3168 delalloc_len = count_range_bits(&inode->io_tree,
3169 delalloc_start_ret, end,
3170 len, EXTENT_DELALLOC, 1,
3171 cached_state);
3172 } else {
3173 spin_unlock(&inode->lock);
3174 }
3175 }
3176
3177 if (delalloc_len > 0) {
3178 /*
3179 * If delalloc was found then *delalloc_start_ret has a sector size
3180 * aligned value (rounded down).
3181 */
3182 *delalloc_end_ret = *delalloc_start_ret + delalloc_len - 1;
3183
3184 if (*delalloc_start_ret == start) {
3185 /* Delalloc for the whole range, nothing more to do. */
3186 if (*delalloc_end_ret == end)
3187 return true;
3188 /* Else trim our search range for ordered extents. */
3189 start = *delalloc_end_ret + 1;
3190 len = end + 1 - start;
3191 }
3192 } else {
3193 /* No delalloc, future calls don't need to search again. */
3194 *search_io_tree = false;
3195 }
3196
3197 /*
3198 * Now also check if there's any ordered extent in the range.
3199 * We do this because:
3200 *
3201 * 1) When delalloc is flushed, the file range is locked, we clear the
3202 * EXTENT_DELALLOC bit from the io tree and create an extent map and
3203 * an ordered extent for the write. So we might just have been called
3204 * after delalloc is flushed and before the ordered extent completes
3205 * and inserts the new file extent item in the subvolume's btree;
3206 *
3207 * 2) We may have an ordered extent created by flushing delalloc for a
3208 * subrange that starts before the subrange we found marked with
3209 * EXTENT_DELALLOC in the io tree.
3210 *
3211 * We could also use the extent map tree to find such delalloc that is
3212 * being flushed, but using the ordered extents tree is more efficient
3213 * because it's usually much smaller as ordered extents are removed from
3214 * the tree once they complete. With the extent maps, we mau have them
3215 * in the extent map tree for a very long time, and they were either
3216 * created by previous writes or loaded by read operations.
3217 */
3218 oe = btrfs_lookup_first_ordered_range(inode, start, len);
3219 if (!oe)
3220 return (delalloc_len > 0);
3221
3222 /* The ordered extent may span beyond our search range. */
3223 oe_start = max(oe->file_offset, start);
3224 oe_end = min(oe->file_offset + oe->num_bytes - 1, end);
3225
3226 btrfs_put_ordered_extent(oe);
3227
3228 /* Don't have unflushed delalloc, return the ordered extent range. */
3229 if (delalloc_len == 0) {
3230 *delalloc_start_ret = oe_start;
3231 *delalloc_end_ret = oe_end;
3232 return true;
3233 }
3234
3235 /*
3236 * We have both unflushed delalloc (io_tree) and an ordered extent.
3237 * If the ranges are adjacent returned a combined range, otherwise
3238 * return the leftmost range.
3239 */
3240 if (oe_start < *delalloc_start_ret) {
3241 if (oe_end < *delalloc_start_ret)
3242 *delalloc_end_ret = oe_end;
3243 *delalloc_start_ret = oe_start;
3244 } else if (*delalloc_end_ret + 1 == oe_start) {
3245 *delalloc_end_ret = oe_end;
3246 }
3247
3248 return true;
3249}
3250
3251/*
3252 * Check if there's delalloc in a given range.
3253 *
3254 * @inode: The inode.
3255 * @start: The start offset of the range. It does not need to be
3256 * sector size aligned.
3257 * @end: The end offset (inclusive value) of the search range.
3258 * It does not need to be sector size aligned.
3259 * @cached_state: Extent state record used for speeding up delalloc
3260 * searches in the inode's io_tree. Can be NULL.
3261 * @delalloc_start_ret: Output argument, set to the start offset of the
3262 * subrange found with delalloc (may not be sector size
3263 * aligned).
3264 * @delalloc_end_ret: Output argument, set to he end offset (inclusive value)
3265 * of the subrange found with delalloc.
3266 *
3267 * Returns true if a subrange with delalloc is found within the given range, and
3268 * if so it sets @delalloc_start_ret and @delalloc_end_ret with the start and
3269 * end offsets of the subrange.
3270 */
3271bool btrfs_find_delalloc_in_range(struct btrfs_inode *inode, u64 start, u64 end,
3272 struct extent_state **cached_state,
3273 u64 *delalloc_start_ret, u64 *delalloc_end_ret)
3274{
3275 u64 cur_offset = round_down(start, inode->root->fs_info->sectorsize);
3276 u64 prev_delalloc_end = 0;
3277 bool search_io_tree = true;
3278 bool ret = false;
3279
3280 while (cur_offset <= end) {
3281 u64 delalloc_start;
3282 u64 delalloc_end;
3283 bool delalloc;
3284
3285 delalloc = find_delalloc_subrange(inode, cur_offset, end,
3286 cached_state, &search_io_tree,
3287 &delalloc_start,
3288 &delalloc_end);
3289 if (!delalloc)
3290 break;
3291
3292 if (prev_delalloc_end == 0) {
3293 /* First subrange found. */
3294 *delalloc_start_ret = max(delalloc_start, start);
3295 *delalloc_end_ret = delalloc_end;
3296 ret = true;
3297 } else if (delalloc_start == prev_delalloc_end + 1) {
3298 /* Subrange adjacent to the previous one, merge them. */
3299 *delalloc_end_ret = delalloc_end;
3300 } else {
3301 /* Subrange not adjacent to the previous one, exit. */
3302 break;
3303 }
3304
3305 prev_delalloc_end = delalloc_end;
3306 cur_offset = delalloc_end + 1;
3307 cond_resched();
3308 }
3309
3310 return ret;
3311}
3312
3313/*
3314 * Check if there's a hole or delalloc range in a range representing a hole (or
3315 * prealloc extent) found in the inode's subvolume btree.
3316 *
3317 * @inode: The inode.
3318 * @whence: Seek mode (SEEK_DATA or SEEK_HOLE).
3319 * @start: Start offset of the hole region. It does not need to be sector
3320 * size aligned.
3321 * @end: End offset (inclusive value) of the hole region. It does not
3322 * need to be sector size aligned.
3323 * @start_ret: Return parameter, used to set the start of the subrange in the
3324 * hole that matches the search criteria (seek mode), if such
3325 * subrange is found (return value of the function is true).
3326 * The value returned here may not be sector size aligned.
3327 *
3328 * Returns true if a subrange matching the given seek mode is found, and if one
3329 * is found, it updates @start_ret with the start of the subrange.
3330 */
3331static bool find_desired_extent_in_hole(struct btrfs_inode *inode, int whence,
3332 struct extent_state **cached_state,
3333 u64 start, u64 end, u64 *start_ret)
3334{
3335 u64 delalloc_start;
3336 u64 delalloc_end;
3337 bool delalloc;
3338
3339 delalloc = btrfs_find_delalloc_in_range(inode, start, end, cached_state,
3340 &delalloc_start, &delalloc_end);
3341 if (delalloc && whence == SEEK_DATA) {
3342 *start_ret = delalloc_start;
3343 return true;
3344 }
3345
3346 if (delalloc && whence == SEEK_HOLE) {
3347 /*
3348 * We found delalloc but it starts after out start offset. So we
3349 * have a hole between our start offset and the delalloc start.
3350 */
3351 if (start < delalloc_start) {
3352 *start_ret = start;
3353 return true;
3354 }
3355 /*
3356 * Delalloc range starts at our start offset.
3357 * If the delalloc range's length is smaller than our range,
3358 * then it means we have a hole that starts where the delalloc
3359 * subrange ends.
3360 */
3361 if (delalloc_end < end) {
3362 *start_ret = delalloc_end + 1;
3363 return true;
3364 }
3365
3366 /* There's delalloc for the whole range. */
3367 return false;
3368 }
3369
3370 if (!delalloc && whence == SEEK_HOLE) {
3371 *start_ret = start;
3372 return true;
3373 }
3374
3375 /*
3376 * No delalloc in the range and we are seeking for data. The caller has
3377 * to iterate to the next extent item in the subvolume btree.
3378 */
3379 return false;
3380}
3381
3382static loff_t find_desired_extent(struct file *file, loff_t offset, int whence)
3383{
3384 struct btrfs_inode *inode = BTRFS_I(file->f_mapping->host);
3385 struct btrfs_file_private *private;
3386 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3387 struct extent_state *cached_state = NULL;
3388 struct extent_state **delalloc_cached_state;
3389 const loff_t i_size = i_size_read(&inode->vfs_inode);
3390 const u64 ino = btrfs_ino(inode);
3391 struct btrfs_root *root = inode->root;
3392 struct btrfs_path *path;
3393 struct btrfs_key key;
3394 u64 last_extent_end;
3395 u64 lockstart;
3396 u64 lockend;
3397 u64 start;
3398 int ret;
3399 bool found = false;
3400
3401 if (i_size == 0 || offset >= i_size)
3402 return -ENXIO;
3403
3404 /*
3405 * Quick path. If the inode has no prealloc extents and its number of
3406 * bytes used matches its i_size, then it can not have holes.
3407 */
3408 if (whence == SEEK_HOLE &&
3409 !(inode->flags & BTRFS_INODE_PREALLOC) &&
3410 inode_get_bytes(&inode->vfs_inode) == i_size)
3411 return i_size;
3412
3413 spin_lock(&inode->lock);
3414 private = file->private_data;
3415 spin_unlock(&inode->lock);
3416
3417 if (private && private->owner_task != current) {
3418 /*
3419 * Not allocated by us, don't use it as its cached state is used
3420 * by the task that allocated it and we don't want neither to
3421 * mess with it nor get incorrect results because it reflects an
3422 * invalid state for the current task.
3423 */
3424 private = NULL;
3425 } else if (!private) {
3426 private = kzalloc(sizeof(*private), GFP_KERNEL);
3427 /*
3428 * No worries if memory allocation failed.
3429 * The private structure is used only for speeding up multiple
3430 * lseek SEEK_HOLE/DATA calls to a file when there's delalloc,
3431 * so everything will still be correct.
3432 */
3433 if (private) {
3434 bool free = false;
3435
3436 private->owner_task = current;
3437
3438 spin_lock(&inode->lock);
3439 if (file->private_data)
3440 free = true;
3441 else
3442 file->private_data = private;
3443 spin_unlock(&inode->lock);
3444
3445 if (free) {
3446 kfree(private);
3447 private = NULL;
3448 }
3449 }
3450 }
3451
3452 if (private)
3453 delalloc_cached_state = &private->llseek_cached_state;
3454 else
3455 delalloc_cached_state = NULL;
3456
3457 /*
3458 * offset can be negative, in this case we start finding DATA/HOLE from
3459 * the very start of the file.
3460 */
3461 start = max_t(loff_t, 0, offset);
3462
3463 lockstart = round_down(start, fs_info->sectorsize);
3464 lockend = round_up(i_size, fs_info->sectorsize);
3465 if (lockend <= lockstart)
3466 lockend = lockstart + fs_info->sectorsize;
3467 lockend--;
3468
3469 path = btrfs_alloc_path();
3470 if (!path)
3471 return -ENOMEM;
3472 path->reada = READA_FORWARD;
3473
3474 key.objectid = ino;
3475 key.type = BTRFS_EXTENT_DATA_KEY;
3476 key.offset = start;
3477
3478 last_extent_end = lockstart;
3479
3480 lock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
3481
3482 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3483 if (ret < 0) {
3484 goto out;
3485 } else if (ret > 0 && path->slots[0] > 0) {
3486 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
3487 if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY)
3488 path->slots[0]--;
3489 }
3490
3491 while (start < i_size) {
3492 struct extent_buffer *leaf = path->nodes[0];
3493 struct btrfs_file_extent_item *extent;
3494 u64 extent_end;
3495 u8 type;
3496
3497 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3498 ret = btrfs_next_leaf(root, path);
3499 if (ret < 0)
3500 goto out;
3501 else if (ret > 0)
3502 break;
3503
3504 leaf = path->nodes[0];
3505 }
3506
3507 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3508 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
3509 break;
3510
3511 extent_end = btrfs_file_extent_end(path);
3512
3513 /*
3514 * In the first iteration we may have a slot that points to an
3515 * extent that ends before our start offset, so skip it.
3516 */
3517 if (extent_end <= start) {
3518 path->slots[0]++;
3519 continue;
3520 }
3521
3522 /* We have an implicit hole, NO_HOLES feature is likely set. */
3523 if (last_extent_end < key.offset) {
3524 u64 search_start = last_extent_end;
3525 u64 found_start;
3526
3527 /*
3528 * First iteration, @start matches @offset and it's
3529 * within the hole.
3530 */
3531 if (start == offset)
3532 search_start = offset;
3533
3534 found = find_desired_extent_in_hole(inode, whence,
3535 delalloc_cached_state,
3536 search_start,
3537 key.offset - 1,
3538 &found_start);
3539 if (found) {
3540 start = found_start;
3541 break;
3542 }
3543 /*
3544 * Didn't find data or a hole (due to delalloc) in the
3545 * implicit hole range, so need to analyze the extent.
3546 */
3547 }
3548
3549 extent = btrfs_item_ptr(leaf, path->slots[0],
3550 struct btrfs_file_extent_item);
3551 type = btrfs_file_extent_type(leaf, extent);
3552
3553 /*
3554 * Can't access the extent's disk_bytenr field if this is an
3555 * inline extent, since at that offset, it's where the extent
3556 * data starts.
3557 */
3558 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
3559 (type == BTRFS_FILE_EXTENT_REG &&
3560 btrfs_file_extent_disk_bytenr(leaf, extent) == 0)) {
3561 /*
3562 * Explicit hole or prealloc extent, search for delalloc.
3563 * A prealloc extent is treated like a hole.
3564 */
3565 u64 search_start = key.offset;
3566 u64 found_start;
3567
3568 /*
3569 * First iteration, @start matches @offset and it's
3570 * within the hole.
3571 */
3572 if (start == offset)
3573 search_start = offset;
3574
3575 found = find_desired_extent_in_hole(inode, whence,
3576 delalloc_cached_state,
3577 search_start,
3578 extent_end - 1,
3579 &found_start);
3580 if (found) {
3581 start = found_start;
3582 break;
3583 }
3584 /*
3585 * Didn't find data or a hole (due to delalloc) in the
3586 * implicit hole range, so need to analyze the next
3587 * extent item.
3588 */
3589 } else {
3590 /*
3591 * Found a regular or inline extent.
3592 * If we are seeking for data, adjust the start offset
3593 * and stop, we're done.
3594 */
3595 if (whence == SEEK_DATA) {
3596 start = max_t(u64, key.offset, offset);
3597 found = true;
3598 break;
3599 }
3600 /*
3601 * Else, we are seeking for a hole, check the next file
3602 * extent item.
3603 */
3604 }
3605
3606 start = extent_end;
3607 last_extent_end = extent_end;
3608 path->slots[0]++;
3609 if (fatal_signal_pending(current)) {
3610 ret = -EINTR;
3611 goto out;
3612 }
3613 cond_resched();
3614 }
3615
3616 /* We have an implicit hole from the last extent found up to i_size. */
3617 if (!found && start < i_size) {
3618 found = find_desired_extent_in_hole(inode, whence,
3619 delalloc_cached_state, start,
3620 i_size - 1, &start);
3621 if (!found)
3622 start = i_size;
3623 }
3624
3625out:
3626 unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
3627 btrfs_free_path(path);
3628
3629 if (ret < 0)
3630 return ret;
3631
3632 if (whence == SEEK_DATA && start >= i_size)
3633 return -ENXIO;
3634
3635 return min_t(loff_t, start, i_size);
3636}
3637
3638static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
3639{
3640 struct inode *inode = file->f_mapping->host;
3641
3642 switch (whence) {
3643 default:
3644 return generic_file_llseek(file, offset, whence);
3645 case SEEK_DATA:
3646 case SEEK_HOLE:
3647 btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_SHARED);
3648 offset = find_desired_extent(file, offset, whence);
3649 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_SHARED);
3650 break;
3651 }
3652
3653 if (offset < 0)
3654 return offset;
3655
3656 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
3657}
3658
3659static int btrfs_file_open(struct inode *inode, struct file *filp)
3660{
3661 int ret;
3662
3663 filp->f_mode |= FMODE_NOWAIT | FMODE_CAN_ODIRECT;
3664
3665 ret = fsverity_file_open(inode, filp);
3666 if (ret)
3667 return ret;
3668 return generic_file_open(inode, filp);
3669}
3670
3671static ssize_t btrfs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
3672{
3673 ssize_t ret = 0;
3674
3675 if (iocb->ki_flags & IOCB_DIRECT) {
3676 ret = btrfs_direct_read(iocb, to);
3677 if (ret < 0 || !iov_iter_count(to) ||
3678 iocb->ki_pos >= i_size_read(file_inode(iocb->ki_filp)))
3679 return ret;
3680 }
3681
3682 return filemap_read(iocb, to, ret);
3683}
3684
3685const struct file_operations btrfs_file_operations = {
3686 .llseek = btrfs_file_llseek,
3687 .read_iter = btrfs_file_read_iter,
3688 .splice_read = filemap_splice_read,
3689 .write_iter = btrfs_file_write_iter,
3690 .splice_write = iter_file_splice_write,
3691 .mmap = btrfs_file_mmap,
3692 .open = btrfs_file_open,
3693 .release = btrfs_release_file,
3694 .get_unmapped_area = thp_get_unmapped_area,
3695 .fsync = btrfs_sync_file,
3696 .fallocate = btrfs_fallocate,
3697 .unlocked_ioctl = btrfs_ioctl,
3698#ifdef CONFIG_COMPAT
3699 .compat_ioctl = btrfs_compat_ioctl,
3700#endif
3701 .remap_file_range = btrfs_remap_file_range,
3702 .uring_cmd = btrfs_uring_cmd,
3703 .fop_flags = FOP_BUFFER_RASYNC | FOP_BUFFER_WASYNC,
3704};
3705
3706int btrfs_fdatawrite_range(struct btrfs_inode *inode, loff_t start, loff_t end)
3707{
3708 struct address_space *mapping = inode->vfs_inode.i_mapping;
3709 int ret;
3710
3711 /*
3712 * So with compression we will find and lock a dirty page and clear the
3713 * first one as dirty, setup an async extent, and immediately return
3714 * with the entire range locked but with nobody actually marked with
3715 * writeback. So we can't just filemap_write_and_wait_range() and
3716 * expect it to work since it will just kick off a thread to do the
3717 * actual work. So we need to call filemap_fdatawrite_range _again_
3718 * since it will wait on the page lock, which won't be unlocked until
3719 * after the pages have been marked as writeback and so we're good to go
3720 * from there. We have to do this otherwise we'll miss the ordered
3721 * extents and that results in badness. Please Josef, do not think you
3722 * know better and pull this out at some point in the future, it is
3723 * right and you are wrong.
3724 */
3725 ret = filemap_fdatawrite_range(mapping, start, end);
3726 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags))
3727 ret = filemap_fdatawrite_range(mapping, start, end);
3728
3729 return ret;
3730}
1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#include <linux/fs.h>
20#include <linux/pagemap.h>
21#include <linux/highmem.h>
22#include <linux/time.h>
23#include <linux/init.h>
24#include <linux/string.h>
25#include <linux/backing-dev.h>
26#include <linux/mpage.h>
27#include <linux/aio.h>
28#include <linux/falloc.h>
29#include <linux/swap.h>
30#include <linux/writeback.h>
31#include <linux/statfs.h>
32#include <linux/compat.h>
33#include <linux/slab.h>
34#include <linux/btrfs.h>
35#include "ctree.h"
36#include "disk-io.h"
37#include "transaction.h"
38#include "btrfs_inode.h"
39#include "print-tree.h"
40#include "tree-log.h"
41#include "locking.h"
42#include "volumes.h"
43
44static struct kmem_cache *btrfs_inode_defrag_cachep;
45/*
46 * when auto defrag is enabled we
47 * queue up these defrag structs to remember which
48 * inodes need defragging passes
49 */
50struct inode_defrag {
51 struct rb_node rb_node;
52 /* objectid */
53 u64 ino;
54 /*
55 * transid where the defrag was added, we search for
56 * extents newer than this
57 */
58 u64 transid;
59
60 /* root objectid */
61 u64 root;
62
63 /* last offset we were able to defrag */
64 u64 last_offset;
65
66 /* if we've wrapped around back to zero once already */
67 int cycled;
68};
69
70static int __compare_inode_defrag(struct inode_defrag *defrag1,
71 struct inode_defrag *defrag2)
72{
73 if (defrag1->root > defrag2->root)
74 return 1;
75 else if (defrag1->root < defrag2->root)
76 return -1;
77 else if (defrag1->ino > defrag2->ino)
78 return 1;
79 else if (defrag1->ino < defrag2->ino)
80 return -1;
81 else
82 return 0;
83}
84
85/* pop a record for an inode into the defrag tree. The lock
86 * must be held already
87 *
88 * If you're inserting a record for an older transid than an
89 * existing record, the transid already in the tree is lowered
90 *
91 * If an existing record is found the defrag item you
92 * pass in is freed
93 */
94static int __btrfs_add_inode_defrag(struct inode *inode,
95 struct inode_defrag *defrag)
96{
97 struct btrfs_root *root = BTRFS_I(inode)->root;
98 struct inode_defrag *entry;
99 struct rb_node **p;
100 struct rb_node *parent = NULL;
101 int ret;
102
103 p = &root->fs_info->defrag_inodes.rb_node;
104 while (*p) {
105 parent = *p;
106 entry = rb_entry(parent, struct inode_defrag, rb_node);
107
108 ret = __compare_inode_defrag(defrag, entry);
109 if (ret < 0)
110 p = &parent->rb_left;
111 else if (ret > 0)
112 p = &parent->rb_right;
113 else {
114 /* if we're reinserting an entry for
115 * an old defrag run, make sure to
116 * lower the transid of our existing record
117 */
118 if (defrag->transid < entry->transid)
119 entry->transid = defrag->transid;
120 if (defrag->last_offset > entry->last_offset)
121 entry->last_offset = defrag->last_offset;
122 return -EEXIST;
123 }
124 }
125 set_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
126 rb_link_node(&defrag->rb_node, parent, p);
127 rb_insert_color(&defrag->rb_node, &root->fs_info->defrag_inodes);
128 return 0;
129}
130
131static inline int __need_auto_defrag(struct btrfs_root *root)
132{
133 if (!btrfs_test_opt(root, AUTO_DEFRAG))
134 return 0;
135
136 if (btrfs_fs_closing(root->fs_info))
137 return 0;
138
139 return 1;
140}
141
142/*
143 * insert a defrag record for this inode if auto defrag is
144 * enabled
145 */
146int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
147 struct inode *inode)
148{
149 struct btrfs_root *root = BTRFS_I(inode)->root;
150 struct inode_defrag *defrag;
151 u64 transid;
152 int ret;
153
154 if (!__need_auto_defrag(root))
155 return 0;
156
157 if (test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags))
158 return 0;
159
160 if (trans)
161 transid = trans->transid;
162 else
163 transid = BTRFS_I(inode)->root->last_trans;
164
165 defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS);
166 if (!defrag)
167 return -ENOMEM;
168
169 defrag->ino = btrfs_ino(inode);
170 defrag->transid = transid;
171 defrag->root = root->root_key.objectid;
172
173 spin_lock(&root->fs_info->defrag_inodes_lock);
174 if (!test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags)) {
175 /*
176 * If we set IN_DEFRAG flag and evict the inode from memory,
177 * and then re-read this inode, this new inode doesn't have
178 * IN_DEFRAG flag. At the case, we may find the existed defrag.
179 */
180 ret = __btrfs_add_inode_defrag(inode, defrag);
181 if (ret)
182 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
183 } else {
184 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
185 }
186 spin_unlock(&root->fs_info->defrag_inodes_lock);
187 return 0;
188}
189
190/*
191 * Requeue the defrag object. If there is a defrag object that points to
192 * the same inode in the tree, we will merge them together (by
193 * __btrfs_add_inode_defrag()) and free the one that we want to requeue.
194 */
195static void btrfs_requeue_inode_defrag(struct inode *inode,
196 struct inode_defrag *defrag)
197{
198 struct btrfs_root *root = BTRFS_I(inode)->root;
199 int ret;
200
201 if (!__need_auto_defrag(root))
202 goto out;
203
204 /*
205 * Here we don't check the IN_DEFRAG flag, because we need merge
206 * them together.
207 */
208 spin_lock(&root->fs_info->defrag_inodes_lock);
209 ret = __btrfs_add_inode_defrag(inode, defrag);
210 spin_unlock(&root->fs_info->defrag_inodes_lock);
211 if (ret)
212 goto out;
213 return;
214out:
215 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
216}
217
218/*
219 * pick the defragable inode that we want, if it doesn't exist, we will get
220 * the next one.
221 */
222static struct inode_defrag *
223btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino)
224{
225 struct inode_defrag *entry = NULL;
226 struct inode_defrag tmp;
227 struct rb_node *p;
228 struct rb_node *parent = NULL;
229 int ret;
230
231 tmp.ino = ino;
232 tmp.root = root;
233
234 spin_lock(&fs_info->defrag_inodes_lock);
235 p = fs_info->defrag_inodes.rb_node;
236 while (p) {
237 parent = p;
238 entry = rb_entry(parent, struct inode_defrag, rb_node);
239
240 ret = __compare_inode_defrag(&tmp, entry);
241 if (ret < 0)
242 p = parent->rb_left;
243 else if (ret > 0)
244 p = parent->rb_right;
245 else
246 goto out;
247 }
248
249 if (parent && __compare_inode_defrag(&tmp, entry) > 0) {
250 parent = rb_next(parent);
251 if (parent)
252 entry = rb_entry(parent, struct inode_defrag, rb_node);
253 else
254 entry = NULL;
255 }
256out:
257 if (entry)
258 rb_erase(parent, &fs_info->defrag_inodes);
259 spin_unlock(&fs_info->defrag_inodes_lock);
260 return entry;
261}
262
263void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info)
264{
265 struct inode_defrag *defrag;
266 struct rb_node *node;
267
268 spin_lock(&fs_info->defrag_inodes_lock);
269 node = rb_first(&fs_info->defrag_inodes);
270 while (node) {
271 rb_erase(node, &fs_info->defrag_inodes);
272 defrag = rb_entry(node, struct inode_defrag, rb_node);
273 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
274
275 if (need_resched()) {
276 spin_unlock(&fs_info->defrag_inodes_lock);
277 cond_resched();
278 spin_lock(&fs_info->defrag_inodes_lock);
279 }
280
281 node = rb_first(&fs_info->defrag_inodes);
282 }
283 spin_unlock(&fs_info->defrag_inodes_lock);
284}
285
286#define BTRFS_DEFRAG_BATCH 1024
287
288static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info,
289 struct inode_defrag *defrag)
290{
291 struct btrfs_root *inode_root;
292 struct inode *inode;
293 struct btrfs_key key;
294 struct btrfs_ioctl_defrag_range_args range;
295 int num_defrag;
296 int index;
297 int ret;
298
299 /* get the inode */
300 key.objectid = defrag->root;
301 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
302 key.offset = (u64)-1;
303
304 index = srcu_read_lock(&fs_info->subvol_srcu);
305
306 inode_root = btrfs_read_fs_root_no_name(fs_info, &key);
307 if (IS_ERR(inode_root)) {
308 ret = PTR_ERR(inode_root);
309 goto cleanup;
310 }
311
312 key.objectid = defrag->ino;
313 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
314 key.offset = 0;
315 inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL);
316 if (IS_ERR(inode)) {
317 ret = PTR_ERR(inode);
318 goto cleanup;
319 }
320 srcu_read_unlock(&fs_info->subvol_srcu, index);
321
322 /* do a chunk of defrag */
323 clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
324 memset(&range, 0, sizeof(range));
325 range.len = (u64)-1;
326 range.start = defrag->last_offset;
327
328 sb_start_write(fs_info->sb);
329 num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
330 BTRFS_DEFRAG_BATCH);
331 sb_end_write(fs_info->sb);
332 /*
333 * if we filled the whole defrag batch, there
334 * must be more work to do. Queue this defrag
335 * again
336 */
337 if (num_defrag == BTRFS_DEFRAG_BATCH) {
338 defrag->last_offset = range.start;
339 btrfs_requeue_inode_defrag(inode, defrag);
340 } else if (defrag->last_offset && !defrag->cycled) {
341 /*
342 * we didn't fill our defrag batch, but
343 * we didn't start at zero. Make sure we loop
344 * around to the start of the file.
345 */
346 defrag->last_offset = 0;
347 defrag->cycled = 1;
348 btrfs_requeue_inode_defrag(inode, defrag);
349 } else {
350 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
351 }
352
353 iput(inode);
354 return 0;
355cleanup:
356 srcu_read_unlock(&fs_info->subvol_srcu, index);
357 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
358 return ret;
359}
360
361/*
362 * run through the list of inodes in the FS that need
363 * defragging
364 */
365int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
366{
367 struct inode_defrag *defrag;
368 u64 first_ino = 0;
369 u64 root_objectid = 0;
370
371 atomic_inc(&fs_info->defrag_running);
372 while (1) {
373 /* Pause the auto defragger. */
374 if (test_bit(BTRFS_FS_STATE_REMOUNTING,
375 &fs_info->fs_state))
376 break;
377
378 if (!__need_auto_defrag(fs_info->tree_root))
379 break;
380
381 /* find an inode to defrag */
382 defrag = btrfs_pick_defrag_inode(fs_info, root_objectid,
383 first_ino);
384 if (!defrag) {
385 if (root_objectid || first_ino) {
386 root_objectid = 0;
387 first_ino = 0;
388 continue;
389 } else {
390 break;
391 }
392 }
393
394 first_ino = defrag->ino + 1;
395 root_objectid = defrag->root;
396
397 __btrfs_run_defrag_inode(fs_info, defrag);
398 }
399 atomic_dec(&fs_info->defrag_running);
400
401 /*
402 * during unmount, we use the transaction_wait queue to
403 * wait for the defragger to stop
404 */
405 wake_up(&fs_info->transaction_wait);
406 return 0;
407}
408
409/* simple helper to fault in pages and copy. This should go away
410 * and be replaced with calls into generic code.
411 */
412static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
413 size_t write_bytes,
414 struct page **prepared_pages,
415 struct iov_iter *i)
416{
417 size_t copied = 0;
418 size_t total_copied = 0;
419 int pg = 0;
420 int offset = pos & (PAGE_CACHE_SIZE - 1);
421
422 while (write_bytes > 0) {
423 size_t count = min_t(size_t,
424 PAGE_CACHE_SIZE - offset, write_bytes);
425 struct page *page = prepared_pages[pg];
426 /*
427 * Copy data from userspace to the current page
428 */
429 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
430
431 /* Flush processor's dcache for this page */
432 flush_dcache_page(page);
433
434 /*
435 * if we get a partial write, we can end up with
436 * partially up to date pages. These add
437 * a lot of complexity, so make sure they don't
438 * happen by forcing this copy to be retried.
439 *
440 * The rest of the btrfs_file_write code will fall
441 * back to page at a time copies after we return 0.
442 */
443 if (!PageUptodate(page) && copied < count)
444 copied = 0;
445
446 iov_iter_advance(i, copied);
447 write_bytes -= copied;
448 total_copied += copied;
449
450 /* Return to btrfs_file_aio_write to fault page */
451 if (unlikely(copied == 0))
452 break;
453
454 if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
455 offset += copied;
456 } else {
457 pg++;
458 offset = 0;
459 }
460 }
461 return total_copied;
462}
463
464/*
465 * unlocks pages after btrfs_file_write is done with them
466 */
467static void btrfs_drop_pages(struct page **pages, size_t num_pages)
468{
469 size_t i;
470 for (i = 0; i < num_pages; i++) {
471 /* page checked is some magic around finding pages that
472 * have been modified without going through btrfs_set_page_dirty
473 * clear it here
474 */
475 ClearPageChecked(pages[i]);
476 unlock_page(pages[i]);
477 mark_page_accessed(pages[i]);
478 page_cache_release(pages[i]);
479 }
480}
481
482/*
483 * after copy_from_user, pages need to be dirtied and we need to make
484 * sure holes are created between the current EOF and the start of
485 * any next extents (if required).
486 *
487 * this also makes the decision about creating an inline extent vs
488 * doing real data extents, marking pages dirty and delalloc as required.
489 */
490int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode,
491 struct page **pages, size_t num_pages,
492 loff_t pos, size_t write_bytes,
493 struct extent_state **cached)
494{
495 int err = 0;
496 int i;
497 u64 num_bytes;
498 u64 start_pos;
499 u64 end_of_last_block;
500 u64 end_pos = pos + write_bytes;
501 loff_t isize = i_size_read(inode);
502
503 start_pos = pos & ~((u64)root->sectorsize - 1);
504 num_bytes = ALIGN(write_bytes + pos - start_pos, root->sectorsize);
505
506 end_of_last_block = start_pos + num_bytes - 1;
507 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
508 cached);
509 if (err)
510 return err;
511
512 for (i = 0; i < num_pages; i++) {
513 struct page *p = pages[i];
514 SetPageUptodate(p);
515 ClearPageChecked(p);
516 set_page_dirty(p);
517 }
518
519 /*
520 * we've only changed i_size in ram, and we haven't updated
521 * the disk i_size. There is no need to log the inode
522 * at this time.
523 */
524 if (end_pos > isize)
525 i_size_write(inode, end_pos);
526 return 0;
527}
528
529/*
530 * this drops all the extents in the cache that intersect the range
531 * [start, end]. Existing extents are split as required.
532 */
533void btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
534 int skip_pinned)
535{
536 struct extent_map *em;
537 struct extent_map *split = NULL;
538 struct extent_map *split2 = NULL;
539 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
540 u64 len = end - start + 1;
541 u64 gen;
542 int ret;
543 int testend = 1;
544 unsigned long flags;
545 int compressed = 0;
546 bool modified;
547
548 WARN_ON(end < start);
549 if (end == (u64)-1) {
550 len = (u64)-1;
551 testend = 0;
552 }
553 while (1) {
554 int no_splits = 0;
555
556 modified = false;
557 if (!split)
558 split = alloc_extent_map();
559 if (!split2)
560 split2 = alloc_extent_map();
561 if (!split || !split2)
562 no_splits = 1;
563
564 write_lock(&em_tree->lock);
565 em = lookup_extent_mapping(em_tree, start, len);
566 if (!em) {
567 write_unlock(&em_tree->lock);
568 break;
569 }
570 flags = em->flags;
571 gen = em->generation;
572 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
573 if (testend && em->start + em->len >= start + len) {
574 free_extent_map(em);
575 write_unlock(&em_tree->lock);
576 break;
577 }
578 start = em->start + em->len;
579 if (testend)
580 len = start + len - (em->start + em->len);
581 free_extent_map(em);
582 write_unlock(&em_tree->lock);
583 continue;
584 }
585 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
586 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
587 clear_bit(EXTENT_FLAG_LOGGING, &flags);
588 modified = !list_empty(&em->list);
589 if (no_splits)
590 goto next;
591
592 if (em->start < start) {
593 split->start = em->start;
594 split->len = start - em->start;
595
596 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
597 split->orig_start = em->orig_start;
598 split->block_start = em->block_start;
599
600 if (compressed)
601 split->block_len = em->block_len;
602 else
603 split->block_len = split->len;
604 split->orig_block_len = max(split->block_len,
605 em->orig_block_len);
606 split->ram_bytes = em->ram_bytes;
607 } else {
608 split->orig_start = split->start;
609 split->block_len = 0;
610 split->block_start = em->block_start;
611 split->orig_block_len = 0;
612 split->ram_bytes = split->len;
613 }
614
615 split->generation = gen;
616 split->bdev = em->bdev;
617 split->flags = flags;
618 split->compress_type = em->compress_type;
619 replace_extent_mapping(em_tree, em, split, modified);
620 free_extent_map(split);
621 split = split2;
622 split2 = NULL;
623 }
624 if (testend && em->start + em->len > start + len) {
625 u64 diff = start + len - em->start;
626
627 split->start = start + len;
628 split->len = em->start + em->len - (start + len);
629 split->bdev = em->bdev;
630 split->flags = flags;
631 split->compress_type = em->compress_type;
632 split->generation = gen;
633
634 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
635 split->orig_block_len = max(em->block_len,
636 em->orig_block_len);
637
638 split->ram_bytes = em->ram_bytes;
639 if (compressed) {
640 split->block_len = em->block_len;
641 split->block_start = em->block_start;
642 split->orig_start = em->orig_start;
643 } else {
644 split->block_len = split->len;
645 split->block_start = em->block_start
646 + diff;
647 split->orig_start = em->orig_start;
648 }
649 } else {
650 split->ram_bytes = split->len;
651 split->orig_start = split->start;
652 split->block_len = 0;
653 split->block_start = em->block_start;
654 split->orig_block_len = 0;
655 }
656
657 if (extent_map_in_tree(em)) {
658 replace_extent_mapping(em_tree, em, split,
659 modified);
660 } else {
661 ret = add_extent_mapping(em_tree, split,
662 modified);
663 ASSERT(ret == 0); /* Logic error */
664 }
665 free_extent_map(split);
666 split = NULL;
667 }
668next:
669 if (extent_map_in_tree(em))
670 remove_extent_mapping(em_tree, em);
671 write_unlock(&em_tree->lock);
672
673 /* once for us */
674 free_extent_map(em);
675 /* once for the tree*/
676 free_extent_map(em);
677 }
678 if (split)
679 free_extent_map(split);
680 if (split2)
681 free_extent_map(split2);
682}
683
684/*
685 * this is very complex, but the basic idea is to drop all extents
686 * in the range start - end. hint_block is filled in with a block number
687 * that would be a good hint to the block allocator for this file.
688 *
689 * If an extent intersects the range but is not entirely inside the range
690 * it is either truncated or split. Anything entirely inside the range
691 * is deleted from the tree.
692 */
693int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
694 struct btrfs_root *root, struct inode *inode,
695 struct btrfs_path *path, u64 start, u64 end,
696 u64 *drop_end, int drop_cache,
697 int replace_extent,
698 u32 extent_item_size,
699 int *key_inserted)
700{
701 struct extent_buffer *leaf;
702 struct btrfs_file_extent_item *fi;
703 struct btrfs_key key;
704 struct btrfs_key new_key;
705 u64 ino = btrfs_ino(inode);
706 u64 search_start = start;
707 u64 disk_bytenr = 0;
708 u64 num_bytes = 0;
709 u64 extent_offset = 0;
710 u64 extent_end = 0;
711 int del_nr = 0;
712 int del_slot = 0;
713 int extent_type;
714 int recow;
715 int ret;
716 int modify_tree = -1;
717 int update_refs = (root->ref_cows || root == root->fs_info->tree_root);
718 int found = 0;
719 int leafs_visited = 0;
720
721 if (drop_cache)
722 btrfs_drop_extent_cache(inode, start, end - 1, 0);
723
724 if (start >= BTRFS_I(inode)->disk_i_size && !replace_extent)
725 modify_tree = 0;
726
727 while (1) {
728 recow = 0;
729 ret = btrfs_lookup_file_extent(trans, root, path, ino,
730 search_start, modify_tree);
731 if (ret < 0)
732 break;
733 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
734 leaf = path->nodes[0];
735 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
736 if (key.objectid == ino &&
737 key.type == BTRFS_EXTENT_DATA_KEY)
738 path->slots[0]--;
739 }
740 ret = 0;
741 leafs_visited++;
742next_slot:
743 leaf = path->nodes[0];
744 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
745 BUG_ON(del_nr > 0);
746 ret = btrfs_next_leaf(root, path);
747 if (ret < 0)
748 break;
749 if (ret > 0) {
750 ret = 0;
751 break;
752 }
753 leafs_visited++;
754 leaf = path->nodes[0];
755 recow = 1;
756 }
757
758 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
759 if (key.objectid > ino ||
760 key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
761 break;
762
763 fi = btrfs_item_ptr(leaf, path->slots[0],
764 struct btrfs_file_extent_item);
765 extent_type = btrfs_file_extent_type(leaf, fi);
766
767 if (extent_type == BTRFS_FILE_EXTENT_REG ||
768 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
769 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
770 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
771 extent_offset = btrfs_file_extent_offset(leaf, fi);
772 extent_end = key.offset +
773 btrfs_file_extent_num_bytes(leaf, fi);
774 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
775 extent_end = key.offset +
776 btrfs_file_extent_inline_len(leaf,
777 path->slots[0], fi);
778 } else {
779 WARN_ON(1);
780 extent_end = search_start;
781 }
782
783 if (extent_end <= search_start) {
784 path->slots[0]++;
785 goto next_slot;
786 }
787
788 found = 1;
789 search_start = max(key.offset, start);
790 if (recow || !modify_tree) {
791 modify_tree = -1;
792 btrfs_release_path(path);
793 continue;
794 }
795
796 /*
797 * | - range to drop - |
798 * | -------- extent -------- |
799 */
800 if (start > key.offset && end < extent_end) {
801 BUG_ON(del_nr > 0);
802 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
803 ret = -EOPNOTSUPP;
804 break;
805 }
806
807 memcpy(&new_key, &key, sizeof(new_key));
808 new_key.offset = start;
809 ret = btrfs_duplicate_item(trans, root, path,
810 &new_key);
811 if (ret == -EAGAIN) {
812 btrfs_release_path(path);
813 continue;
814 }
815 if (ret < 0)
816 break;
817
818 leaf = path->nodes[0];
819 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
820 struct btrfs_file_extent_item);
821 btrfs_set_file_extent_num_bytes(leaf, fi,
822 start - key.offset);
823
824 fi = btrfs_item_ptr(leaf, path->slots[0],
825 struct btrfs_file_extent_item);
826
827 extent_offset += start - key.offset;
828 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
829 btrfs_set_file_extent_num_bytes(leaf, fi,
830 extent_end - start);
831 btrfs_mark_buffer_dirty(leaf);
832
833 if (update_refs && disk_bytenr > 0) {
834 ret = btrfs_inc_extent_ref(trans, root,
835 disk_bytenr, num_bytes, 0,
836 root->root_key.objectid,
837 new_key.objectid,
838 start - extent_offset, 0);
839 BUG_ON(ret); /* -ENOMEM */
840 }
841 key.offset = start;
842 }
843 /*
844 * | ---- range to drop ----- |
845 * | -------- extent -------- |
846 */
847 if (start <= key.offset && end < extent_end) {
848 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
849 ret = -EOPNOTSUPP;
850 break;
851 }
852
853 memcpy(&new_key, &key, sizeof(new_key));
854 new_key.offset = end;
855 btrfs_set_item_key_safe(root, path, &new_key);
856
857 extent_offset += end - key.offset;
858 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
859 btrfs_set_file_extent_num_bytes(leaf, fi,
860 extent_end - end);
861 btrfs_mark_buffer_dirty(leaf);
862 if (update_refs && disk_bytenr > 0)
863 inode_sub_bytes(inode, end - key.offset);
864 break;
865 }
866
867 search_start = extent_end;
868 /*
869 * | ---- range to drop ----- |
870 * | -------- extent -------- |
871 */
872 if (start > key.offset && end >= extent_end) {
873 BUG_ON(del_nr > 0);
874 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
875 ret = -EOPNOTSUPP;
876 break;
877 }
878
879 btrfs_set_file_extent_num_bytes(leaf, fi,
880 start - key.offset);
881 btrfs_mark_buffer_dirty(leaf);
882 if (update_refs && disk_bytenr > 0)
883 inode_sub_bytes(inode, extent_end - start);
884 if (end == extent_end)
885 break;
886
887 path->slots[0]++;
888 goto next_slot;
889 }
890
891 /*
892 * | ---- range to drop ----- |
893 * | ------ extent ------ |
894 */
895 if (start <= key.offset && end >= extent_end) {
896 if (del_nr == 0) {
897 del_slot = path->slots[0];
898 del_nr = 1;
899 } else {
900 BUG_ON(del_slot + del_nr != path->slots[0]);
901 del_nr++;
902 }
903
904 if (update_refs &&
905 extent_type == BTRFS_FILE_EXTENT_INLINE) {
906 inode_sub_bytes(inode,
907 extent_end - key.offset);
908 extent_end = ALIGN(extent_end,
909 root->sectorsize);
910 } else if (update_refs && disk_bytenr > 0) {
911 ret = btrfs_free_extent(trans, root,
912 disk_bytenr, num_bytes, 0,
913 root->root_key.objectid,
914 key.objectid, key.offset -
915 extent_offset, 0);
916 BUG_ON(ret); /* -ENOMEM */
917 inode_sub_bytes(inode,
918 extent_end - key.offset);
919 }
920
921 if (end == extent_end)
922 break;
923
924 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
925 path->slots[0]++;
926 goto next_slot;
927 }
928
929 ret = btrfs_del_items(trans, root, path, del_slot,
930 del_nr);
931 if (ret) {
932 btrfs_abort_transaction(trans, root, ret);
933 break;
934 }
935
936 del_nr = 0;
937 del_slot = 0;
938
939 btrfs_release_path(path);
940 continue;
941 }
942
943 BUG_ON(1);
944 }
945
946 if (!ret && del_nr > 0) {
947 /*
948 * Set path->slots[0] to first slot, so that after the delete
949 * if items are move off from our leaf to its immediate left or
950 * right neighbor leafs, we end up with a correct and adjusted
951 * path->slots[0] for our insertion (if replace_extent != 0).
952 */
953 path->slots[0] = del_slot;
954 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
955 if (ret)
956 btrfs_abort_transaction(trans, root, ret);
957 }
958
959 leaf = path->nodes[0];
960 /*
961 * If btrfs_del_items() was called, it might have deleted a leaf, in
962 * which case it unlocked our path, so check path->locks[0] matches a
963 * write lock.
964 */
965 if (!ret && replace_extent && leafs_visited == 1 &&
966 (path->locks[0] == BTRFS_WRITE_LOCK_BLOCKING ||
967 path->locks[0] == BTRFS_WRITE_LOCK) &&
968 btrfs_leaf_free_space(root, leaf) >=
969 sizeof(struct btrfs_item) + extent_item_size) {
970
971 key.objectid = ino;
972 key.type = BTRFS_EXTENT_DATA_KEY;
973 key.offset = start;
974 if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
975 struct btrfs_key slot_key;
976
977 btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
978 if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
979 path->slots[0]++;
980 }
981 setup_items_for_insert(root, path, &key,
982 &extent_item_size,
983 extent_item_size,
984 sizeof(struct btrfs_item) +
985 extent_item_size, 1);
986 *key_inserted = 1;
987 }
988
989 if (!replace_extent || !(*key_inserted))
990 btrfs_release_path(path);
991 if (drop_end)
992 *drop_end = found ? min(end, extent_end) : end;
993 return ret;
994}
995
996int btrfs_drop_extents(struct btrfs_trans_handle *trans,
997 struct btrfs_root *root, struct inode *inode, u64 start,
998 u64 end, int drop_cache)
999{
1000 struct btrfs_path *path;
1001 int ret;
1002
1003 path = btrfs_alloc_path();
1004 if (!path)
1005 return -ENOMEM;
1006 ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
1007 drop_cache, 0, 0, NULL);
1008 btrfs_free_path(path);
1009 return ret;
1010}
1011
1012static int extent_mergeable(struct extent_buffer *leaf, int slot,
1013 u64 objectid, u64 bytenr, u64 orig_offset,
1014 u64 *start, u64 *end)
1015{
1016 struct btrfs_file_extent_item *fi;
1017 struct btrfs_key key;
1018 u64 extent_end;
1019
1020 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1021 return 0;
1022
1023 btrfs_item_key_to_cpu(leaf, &key, slot);
1024 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
1025 return 0;
1026
1027 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1028 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
1029 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
1030 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
1031 btrfs_file_extent_compression(leaf, fi) ||
1032 btrfs_file_extent_encryption(leaf, fi) ||
1033 btrfs_file_extent_other_encoding(leaf, fi))
1034 return 0;
1035
1036 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1037 if ((*start && *start != key.offset) || (*end && *end != extent_end))
1038 return 0;
1039
1040 *start = key.offset;
1041 *end = extent_end;
1042 return 1;
1043}
1044
1045/*
1046 * Mark extent in the range start - end as written.
1047 *
1048 * This changes extent type from 'pre-allocated' to 'regular'. If only
1049 * part of extent is marked as written, the extent will be split into
1050 * two or three.
1051 */
1052int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
1053 struct inode *inode, u64 start, u64 end)
1054{
1055 struct btrfs_root *root = BTRFS_I(inode)->root;
1056 struct extent_buffer *leaf;
1057 struct btrfs_path *path;
1058 struct btrfs_file_extent_item *fi;
1059 struct btrfs_key key;
1060 struct btrfs_key new_key;
1061 u64 bytenr;
1062 u64 num_bytes;
1063 u64 extent_end;
1064 u64 orig_offset;
1065 u64 other_start;
1066 u64 other_end;
1067 u64 split;
1068 int del_nr = 0;
1069 int del_slot = 0;
1070 int recow;
1071 int ret;
1072 u64 ino = btrfs_ino(inode);
1073
1074 path = btrfs_alloc_path();
1075 if (!path)
1076 return -ENOMEM;
1077again:
1078 recow = 0;
1079 split = start;
1080 key.objectid = ino;
1081 key.type = BTRFS_EXTENT_DATA_KEY;
1082 key.offset = split;
1083
1084 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1085 if (ret < 0)
1086 goto out;
1087 if (ret > 0 && path->slots[0] > 0)
1088 path->slots[0]--;
1089
1090 leaf = path->nodes[0];
1091 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1092 BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
1093 fi = btrfs_item_ptr(leaf, path->slots[0],
1094 struct btrfs_file_extent_item);
1095 BUG_ON(btrfs_file_extent_type(leaf, fi) !=
1096 BTRFS_FILE_EXTENT_PREALLOC);
1097 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1098 BUG_ON(key.offset > start || extent_end < end);
1099
1100 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1101 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1102 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1103 memcpy(&new_key, &key, sizeof(new_key));
1104
1105 if (start == key.offset && end < extent_end) {
1106 other_start = 0;
1107 other_end = start;
1108 if (extent_mergeable(leaf, path->slots[0] - 1,
1109 ino, bytenr, orig_offset,
1110 &other_start, &other_end)) {
1111 new_key.offset = end;
1112 btrfs_set_item_key_safe(root, path, &new_key);
1113 fi = btrfs_item_ptr(leaf, path->slots[0],
1114 struct btrfs_file_extent_item);
1115 btrfs_set_file_extent_generation(leaf, fi,
1116 trans->transid);
1117 btrfs_set_file_extent_num_bytes(leaf, fi,
1118 extent_end - end);
1119 btrfs_set_file_extent_offset(leaf, fi,
1120 end - orig_offset);
1121 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1122 struct btrfs_file_extent_item);
1123 btrfs_set_file_extent_generation(leaf, fi,
1124 trans->transid);
1125 btrfs_set_file_extent_num_bytes(leaf, fi,
1126 end - other_start);
1127 btrfs_mark_buffer_dirty(leaf);
1128 goto out;
1129 }
1130 }
1131
1132 if (start > key.offset && end == extent_end) {
1133 other_start = end;
1134 other_end = 0;
1135 if (extent_mergeable(leaf, path->slots[0] + 1,
1136 ino, bytenr, orig_offset,
1137 &other_start, &other_end)) {
1138 fi = btrfs_item_ptr(leaf, path->slots[0],
1139 struct btrfs_file_extent_item);
1140 btrfs_set_file_extent_num_bytes(leaf, fi,
1141 start - key.offset);
1142 btrfs_set_file_extent_generation(leaf, fi,
1143 trans->transid);
1144 path->slots[0]++;
1145 new_key.offset = start;
1146 btrfs_set_item_key_safe(root, path, &new_key);
1147
1148 fi = btrfs_item_ptr(leaf, path->slots[0],
1149 struct btrfs_file_extent_item);
1150 btrfs_set_file_extent_generation(leaf, fi,
1151 trans->transid);
1152 btrfs_set_file_extent_num_bytes(leaf, fi,
1153 other_end - start);
1154 btrfs_set_file_extent_offset(leaf, fi,
1155 start - orig_offset);
1156 btrfs_mark_buffer_dirty(leaf);
1157 goto out;
1158 }
1159 }
1160
1161 while (start > key.offset || end < extent_end) {
1162 if (key.offset == start)
1163 split = end;
1164
1165 new_key.offset = split;
1166 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1167 if (ret == -EAGAIN) {
1168 btrfs_release_path(path);
1169 goto again;
1170 }
1171 if (ret < 0) {
1172 btrfs_abort_transaction(trans, root, ret);
1173 goto out;
1174 }
1175
1176 leaf = path->nodes[0];
1177 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1178 struct btrfs_file_extent_item);
1179 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1180 btrfs_set_file_extent_num_bytes(leaf, fi,
1181 split - key.offset);
1182
1183 fi = btrfs_item_ptr(leaf, path->slots[0],
1184 struct btrfs_file_extent_item);
1185
1186 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1187 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1188 btrfs_set_file_extent_num_bytes(leaf, fi,
1189 extent_end - split);
1190 btrfs_mark_buffer_dirty(leaf);
1191
1192 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
1193 root->root_key.objectid,
1194 ino, orig_offset, 0);
1195 BUG_ON(ret); /* -ENOMEM */
1196
1197 if (split == start) {
1198 key.offset = start;
1199 } else {
1200 BUG_ON(start != key.offset);
1201 path->slots[0]--;
1202 extent_end = end;
1203 }
1204 recow = 1;
1205 }
1206
1207 other_start = end;
1208 other_end = 0;
1209 if (extent_mergeable(leaf, path->slots[0] + 1,
1210 ino, bytenr, orig_offset,
1211 &other_start, &other_end)) {
1212 if (recow) {
1213 btrfs_release_path(path);
1214 goto again;
1215 }
1216 extent_end = other_end;
1217 del_slot = path->slots[0] + 1;
1218 del_nr++;
1219 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1220 0, root->root_key.objectid,
1221 ino, orig_offset, 0);
1222 BUG_ON(ret); /* -ENOMEM */
1223 }
1224 other_start = 0;
1225 other_end = start;
1226 if (extent_mergeable(leaf, path->slots[0] - 1,
1227 ino, bytenr, orig_offset,
1228 &other_start, &other_end)) {
1229 if (recow) {
1230 btrfs_release_path(path);
1231 goto again;
1232 }
1233 key.offset = other_start;
1234 del_slot = path->slots[0];
1235 del_nr++;
1236 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1237 0, root->root_key.objectid,
1238 ino, orig_offset, 0);
1239 BUG_ON(ret); /* -ENOMEM */
1240 }
1241 if (del_nr == 0) {
1242 fi = btrfs_item_ptr(leaf, path->slots[0],
1243 struct btrfs_file_extent_item);
1244 btrfs_set_file_extent_type(leaf, fi,
1245 BTRFS_FILE_EXTENT_REG);
1246 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1247 btrfs_mark_buffer_dirty(leaf);
1248 } else {
1249 fi = btrfs_item_ptr(leaf, del_slot - 1,
1250 struct btrfs_file_extent_item);
1251 btrfs_set_file_extent_type(leaf, fi,
1252 BTRFS_FILE_EXTENT_REG);
1253 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1254 btrfs_set_file_extent_num_bytes(leaf, fi,
1255 extent_end - key.offset);
1256 btrfs_mark_buffer_dirty(leaf);
1257
1258 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1259 if (ret < 0) {
1260 btrfs_abort_transaction(trans, root, ret);
1261 goto out;
1262 }
1263 }
1264out:
1265 btrfs_free_path(path);
1266 return 0;
1267}
1268
1269/*
1270 * on error we return an unlocked page and the error value
1271 * on success we return a locked page and 0
1272 */
1273static int prepare_uptodate_page(struct page *page, u64 pos,
1274 bool force_uptodate)
1275{
1276 int ret = 0;
1277
1278 if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1279 !PageUptodate(page)) {
1280 ret = btrfs_readpage(NULL, page);
1281 if (ret)
1282 return ret;
1283 lock_page(page);
1284 if (!PageUptodate(page)) {
1285 unlock_page(page);
1286 return -EIO;
1287 }
1288 }
1289 return 0;
1290}
1291
1292/*
1293 * this just gets pages into the page cache and locks them down.
1294 */
1295static noinline int prepare_pages(struct inode *inode, struct page **pages,
1296 size_t num_pages, loff_t pos,
1297 size_t write_bytes, bool force_uptodate)
1298{
1299 int i;
1300 unsigned long index = pos >> PAGE_CACHE_SHIFT;
1301 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1302 int err = 0;
1303 int faili;
1304
1305 for (i = 0; i < num_pages; i++) {
1306 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1307 mask | __GFP_WRITE);
1308 if (!pages[i]) {
1309 faili = i - 1;
1310 err = -ENOMEM;
1311 goto fail;
1312 }
1313
1314 if (i == 0)
1315 err = prepare_uptodate_page(pages[i], pos,
1316 force_uptodate);
1317 if (i == num_pages - 1)
1318 err = prepare_uptodate_page(pages[i],
1319 pos + write_bytes, false);
1320 if (err) {
1321 page_cache_release(pages[i]);
1322 faili = i - 1;
1323 goto fail;
1324 }
1325 wait_on_page_writeback(pages[i]);
1326 }
1327
1328 return 0;
1329fail:
1330 while (faili >= 0) {
1331 unlock_page(pages[faili]);
1332 page_cache_release(pages[faili]);
1333 faili--;
1334 }
1335 return err;
1336
1337}
1338
1339/*
1340 * This function locks the extent and properly waits for data=ordered extents
1341 * to finish before allowing the pages to be modified if need.
1342 *
1343 * The return value:
1344 * 1 - the extent is locked
1345 * 0 - the extent is not locked, and everything is OK
1346 * -EAGAIN - need re-prepare the pages
1347 * the other < 0 number - Something wrong happens
1348 */
1349static noinline int
1350lock_and_cleanup_extent_if_need(struct inode *inode, struct page **pages,
1351 size_t num_pages, loff_t pos,
1352 u64 *lockstart, u64 *lockend,
1353 struct extent_state **cached_state)
1354{
1355 u64 start_pos;
1356 u64 last_pos;
1357 int i;
1358 int ret = 0;
1359
1360 start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
1361 last_pos = start_pos + ((u64)num_pages << PAGE_CACHE_SHIFT) - 1;
1362
1363 if (start_pos < inode->i_size) {
1364 struct btrfs_ordered_extent *ordered;
1365 lock_extent_bits(&BTRFS_I(inode)->io_tree,
1366 start_pos, last_pos, 0, cached_state);
1367 ordered = btrfs_lookup_ordered_range(inode, start_pos,
1368 last_pos - start_pos + 1);
1369 if (ordered &&
1370 ordered->file_offset + ordered->len > start_pos &&
1371 ordered->file_offset <= last_pos) {
1372 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1373 start_pos, last_pos,
1374 cached_state, GFP_NOFS);
1375 for (i = 0; i < num_pages; i++) {
1376 unlock_page(pages[i]);
1377 page_cache_release(pages[i]);
1378 }
1379 btrfs_start_ordered_extent(inode, ordered, 1);
1380 btrfs_put_ordered_extent(ordered);
1381 return -EAGAIN;
1382 }
1383 if (ordered)
1384 btrfs_put_ordered_extent(ordered);
1385
1386 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1387 last_pos, EXTENT_DIRTY | EXTENT_DELALLOC |
1388 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1389 0, 0, cached_state, GFP_NOFS);
1390 *lockstart = start_pos;
1391 *lockend = last_pos;
1392 ret = 1;
1393 }
1394
1395 for (i = 0; i < num_pages; i++) {
1396 if (clear_page_dirty_for_io(pages[i]))
1397 account_page_redirty(pages[i]);
1398 set_page_extent_mapped(pages[i]);
1399 WARN_ON(!PageLocked(pages[i]));
1400 }
1401
1402 return ret;
1403}
1404
1405static noinline int check_can_nocow(struct inode *inode, loff_t pos,
1406 size_t *write_bytes)
1407{
1408 struct btrfs_root *root = BTRFS_I(inode)->root;
1409 struct btrfs_ordered_extent *ordered;
1410 u64 lockstart, lockend;
1411 u64 num_bytes;
1412 int ret;
1413
1414 ret = btrfs_start_nocow_write(root);
1415 if (!ret)
1416 return -ENOSPC;
1417
1418 lockstart = round_down(pos, root->sectorsize);
1419 lockend = round_up(pos + *write_bytes, root->sectorsize) - 1;
1420
1421 while (1) {
1422 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1423 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1424 lockend - lockstart + 1);
1425 if (!ordered) {
1426 break;
1427 }
1428 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1429 btrfs_start_ordered_extent(inode, ordered, 1);
1430 btrfs_put_ordered_extent(ordered);
1431 }
1432
1433 num_bytes = lockend - lockstart + 1;
1434 ret = can_nocow_extent(inode, lockstart, &num_bytes, NULL, NULL, NULL);
1435 if (ret <= 0) {
1436 ret = 0;
1437 btrfs_end_nocow_write(root);
1438 } else {
1439 *write_bytes = min_t(size_t, *write_bytes ,
1440 num_bytes - pos + lockstart);
1441 }
1442
1443 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend);
1444
1445 return ret;
1446}
1447
1448static noinline ssize_t __btrfs_buffered_write(struct file *file,
1449 struct iov_iter *i,
1450 loff_t pos)
1451{
1452 struct inode *inode = file_inode(file);
1453 struct btrfs_root *root = BTRFS_I(inode)->root;
1454 struct page **pages = NULL;
1455 struct extent_state *cached_state = NULL;
1456 u64 release_bytes = 0;
1457 u64 lockstart;
1458 u64 lockend;
1459 unsigned long first_index;
1460 size_t num_written = 0;
1461 int nrptrs;
1462 int ret = 0;
1463 bool only_release_metadata = false;
1464 bool force_page_uptodate = false;
1465 bool need_unlock;
1466
1467 nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
1468 PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
1469 (sizeof(struct page *)));
1470 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1471 nrptrs = max(nrptrs, 8);
1472 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1473 if (!pages)
1474 return -ENOMEM;
1475
1476 first_index = pos >> PAGE_CACHE_SHIFT;
1477
1478 while (iov_iter_count(i) > 0) {
1479 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1480 size_t write_bytes = min(iov_iter_count(i),
1481 nrptrs * (size_t)PAGE_CACHE_SIZE -
1482 offset);
1483 size_t num_pages = (write_bytes + offset +
1484 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1485 size_t reserve_bytes;
1486 size_t dirty_pages;
1487 size_t copied;
1488
1489 WARN_ON(num_pages > nrptrs);
1490
1491 /*
1492 * Fault pages before locking them in prepare_pages
1493 * to avoid recursive lock
1494 */
1495 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1496 ret = -EFAULT;
1497 break;
1498 }
1499
1500 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1501 ret = btrfs_check_data_free_space(inode, reserve_bytes);
1502 if (ret == -ENOSPC &&
1503 (BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1504 BTRFS_INODE_PREALLOC))) {
1505 ret = check_can_nocow(inode, pos, &write_bytes);
1506 if (ret > 0) {
1507 only_release_metadata = true;
1508 /*
1509 * our prealloc extent may be smaller than
1510 * write_bytes, so scale down.
1511 */
1512 num_pages = (write_bytes + offset +
1513 PAGE_CACHE_SIZE - 1) >>
1514 PAGE_CACHE_SHIFT;
1515 reserve_bytes = num_pages << PAGE_CACHE_SHIFT;
1516 ret = 0;
1517 } else {
1518 ret = -ENOSPC;
1519 }
1520 }
1521
1522 if (ret)
1523 break;
1524
1525 ret = btrfs_delalloc_reserve_metadata(inode, reserve_bytes);
1526 if (ret) {
1527 if (!only_release_metadata)
1528 btrfs_free_reserved_data_space(inode,
1529 reserve_bytes);
1530 else
1531 btrfs_end_nocow_write(root);
1532 break;
1533 }
1534
1535 release_bytes = reserve_bytes;
1536 need_unlock = false;
1537again:
1538 /*
1539 * This is going to setup the pages array with the number of
1540 * pages we want, so we don't really need to worry about the
1541 * contents of pages from loop to loop
1542 */
1543 ret = prepare_pages(inode, pages, num_pages,
1544 pos, write_bytes,
1545 force_page_uptodate);
1546 if (ret)
1547 break;
1548
1549 ret = lock_and_cleanup_extent_if_need(inode, pages, num_pages,
1550 pos, &lockstart, &lockend,
1551 &cached_state);
1552 if (ret < 0) {
1553 if (ret == -EAGAIN)
1554 goto again;
1555 break;
1556 } else if (ret > 0) {
1557 need_unlock = true;
1558 ret = 0;
1559 }
1560
1561 copied = btrfs_copy_from_user(pos, num_pages,
1562 write_bytes, pages, i);
1563
1564 /*
1565 * if we have trouble faulting in the pages, fall
1566 * back to one page at a time
1567 */
1568 if (copied < write_bytes)
1569 nrptrs = 1;
1570
1571 if (copied == 0) {
1572 force_page_uptodate = true;
1573 dirty_pages = 0;
1574 } else {
1575 force_page_uptodate = false;
1576 dirty_pages = (copied + offset +
1577 PAGE_CACHE_SIZE - 1) >>
1578 PAGE_CACHE_SHIFT;
1579 }
1580
1581 /*
1582 * If we had a short copy we need to release the excess delaloc
1583 * bytes we reserved. We need to increment outstanding_extents
1584 * because btrfs_delalloc_release_space will decrement it, but
1585 * we still have an outstanding extent for the chunk we actually
1586 * managed to copy.
1587 */
1588 if (num_pages > dirty_pages) {
1589 release_bytes = (num_pages - dirty_pages) <<
1590 PAGE_CACHE_SHIFT;
1591 if (copied > 0) {
1592 spin_lock(&BTRFS_I(inode)->lock);
1593 BTRFS_I(inode)->outstanding_extents++;
1594 spin_unlock(&BTRFS_I(inode)->lock);
1595 }
1596 if (only_release_metadata)
1597 btrfs_delalloc_release_metadata(inode,
1598 release_bytes);
1599 else
1600 btrfs_delalloc_release_space(inode,
1601 release_bytes);
1602 }
1603
1604 release_bytes = dirty_pages << PAGE_CACHE_SHIFT;
1605
1606 if (copied > 0)
1607 ret = btrfs_dirty_pages(root, inode, pages,
1608 dirty_pages, pos, copied,
1609 NULL);
1610 if (need_unlock)
1611 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1612 lockstart, lockend, &cached_state,
1613 GFP_NOFS);
1614 if (ret) {
1615 btrfs_drop_pages(pages, num_pages);
1616 break;
1617 }
1618
1619 release_bytes = 0;
1620 if (only_release_metadata)
1621 btrfs_end_nocow_write(root);
1622
1623 if (only_release_metadata && copied > 0) {
1624 u64 lockstart = round_down(pos, root->sectorsize);
1625 u64 lockend = lockstart +
1626 (dirty_pages << PAGE_CACHE_SHIFT) - 1;
1627
1628 set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
1629 lockend, EXTENT_NORESERVE, NULL,
1630 NULL, GFP_NOFS);
1631 only_release_metadata = false;
1632 }
1633
1634 btrfs_drop_pages(pages, num_pages);
1635
1636 cond_resched();
1637
1638 balance_dirty_pages_ratelimited(inode->i_mapping);
1639 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1640 btrfs_btree_balance_dirty(root);
1641
1642 pos += copied;
1643 num_written += copied;
1644 }
1645
1646 kfree(pages);
1647
1648 if (release_bytes) {
1649 if (only_release_metadata) {
1650 btrfs_end_nocow_write(root);
1651 btrfs_delalloc_release_metadata(inode, release_bytes);
1652 } else {
1653 btrfs_delalloc_release_space(inode, release_bytes);
1654 }
1655 }
1656
1657 return num_written ? num_written : ret;
1658}
1659
1660static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1661 const struct iovec *iov,
1662 unsigned long nr_segs, loff_t pos,
1663 size_t count, size_t ocount)
1664{
1665 struct file *file = iocb->ki_filp;
1666 struct iov_iter i;
1667 ssize_t written;
1668 ssize_t written_buffered;
1669 loff_t endbyte;
1670 int err;
1671
1672 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
1673 count, ocount);
1674
1675 if (written < 0 || written == count)
1676 return written;
1677
1678 pos += written;
1679 count -= written;
1680 iov_iter_init(&i, iov, nr_segs, count, written);
1681 written_buffered = __btrfs_buffered_write(file, &i, pos);
1682 if (written_buffered < 0) {
1683 err = written_buffered;
1684 goto out;
1685 }
1686 endbyte = pos + written_buffered - 1;
1687 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1688 if (err)
1689 goto out;
1690 written += written_buffered;
1691 iocb->ki_pos = pos + written_buffered;
1692 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1693 endbyte >> PAGE_CACHE_SHIFT);
1694out:
1695 return written ? written : err;
1696}
1697
1698static void update_time_for_write(struct inode *inode)
1699{
1700 struct timespec now;
1701
1702 if (IS_NOCMTIME(inode))
1703 return;
1704
1705 now = current_fs_time(inode->i_sb);
1706 if (!timespec_equal(&inode->i_mtime, &now))
1707 inode->i_mtime = now;
1708
1709 if (!timespec_equal(&inode->i_ctime, &now))
1710 inode->i_ctime = now;
1711
1712 if (IS_I_VERSION(inode))
1713 inode_inc_iversion(inode);
1714}
1715
1716static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1717 const struct iovec *iov,
1718 unsigned long nr_segs, loff_t pos)
1719{
1720 struct file *file = iocb->ki_filp;
1721 struct inode *inode = file_inode(file);
1722 struct btrfs_root *root = BTRFS_I(inode)->root;
1723 u64 start_pos;
1724 u64 end_pos;
1725 ssize_t num_written = 0;
1726 ssize_t err = 0;
1727 size_t count, ocount;
1728 bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
1729
1730 mutex_lock(&inode->i_mutex);
1731
1732 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1733 if (err) {
1734 mutex_unlock(&inode->i_mutex);
1735 goto out;
1736 }
1737 count = ocount;
1738
1739 current->backing_dev_info = inode->i_mapping->backing_dev_info;
1740 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1741 if (err) {
1742 mutex_unlock(&inode->i_mutex);
1743 goto out;
1744 }
1745
1746 if (count == 0) {
1747 mutex_unlock(&inode->i_mutex);
1748 goto out;
1749 }
1750
1751 err = file_remove_suid(file);
1752 if (err) {
1753 mutex_unlock(&inode->i_mutex);
1754 goto out;
1755 }
1756
1757 /*
1758 * If BTRFS flips readonly due to some impossible error
1759 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1760 * although we have opened a file as writable, we have
1761 * to stop this write operation to ensure FS consistency.
1762 */
1763 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) {
1764 mutex_unlock(&inode->i_mutex);
1765 err = -EROFS;
1766 goto out;
1767 }
1768
1769 /*
1770 * We reserve space for updating the inode when we reserve space for the
1771 * extent we are going to write, so we will enospc out there. We don't
1772 * need to start yet another transaction to update the inode as we will
1773 * update the inode when we finish writing whatever data we write.
1774 */
1775 update_time_for_write(inode);
1776
1777 start_pos = round_down(pos, root->sectorsize);
1778 if (start_pos > i_size_read(inode)) {
1779 /* Expand hole size to cover write data, preventing empty gap */
1780 end_pos = round_up(pos + count, root->sectorsize);
1781 err = btrfs_cont_expand(inode, i_size_read(inode), end_pos);
1782 if (err) {
1783 mutex_unlock(&inode->i_mutex);
1784 goto out;
1785 }
1786 }
1787
1788 if (sync)
1789 atomic_inc(&BTRFS_I(inode)->sync_writers);
1790
1791 if (unlikely(file->f_flags & O_DIRECT)) {
1792 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1793 pos, count, ocount);
1794 } else {
1795 struct iov_iter i;
1796
1797 iov_iter_init(&i, iov, nr_segs, count, num_written);
1798
1799 num_written = __btrfs_buffered_write(file, &i, pos);
1800 if (num_written > 0)
1801 iocb->ki_pos = pos + num_written;
1802 }
1803
1804 mutex_unlock(&inode->i_mutex);
1805
1806 /*
1807 * we want to make sure fsync finds this change
1808 * but we haven't joined a transaction running right now.
1809 *
1810 * Later on, someone is sure to update the inode and get the
1811 * real transid recorded.
1812 *
1813 * We set last_trans now to the fs_info generation + 1,
1814 * this will either be one more than the running transaction
1815 * or the generation used for the next transaction if there isn't
1816 * one running right now.
1817 *
1818 * We also have to set last_sub_trans to the current log transid,
1819 * otherwise subsequent syncs to a file that's been synced in this
1820 * transaction will appear to have already occured.
1821 */
1822 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1823 BTRFS_I(inode)->last_sub_trans = root->log_transid;
1824 if (num_written > 0) {
1825 err = generic_write_sync(file, pos, num_written);
1826 if (err < 0)
1827 num_written = err;
1828 }
1829
1830 if (sync)
1831 atomic_dec(&BTRFS_I(inode)->sync_writers);
1832out:
1833 current->backing_dev_info = NULL;
1834 return num_written ? num_written : err;
1835}
1836
1837int btrfs_release_file(struct inode *inode, struct file *filp)
1838{
1839 /*
1840 * ordered_data_close is set by settattr when we are about to truncate
1841 * a file from a non-zero size to a zero size. This tries to
1842 * flush down new bytes that may have been written if the
1843 * application were using truncate to replace a file in place.
1844 */
1845 if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
1846 &BTRFS_I(inode)->runtime_flags)) {
1847 struct btrfs_trans_handle *trans;
1848 struct btrfs_root *root = BTRFS_I(inode)->root;
1849
1850 /*
1851 * We need to block on a committing transaction to keep us from
1852 * throwing a ordered operation on to the list and causing
1853 * something like sync to deadlock trying to flush out this
1854 * inode.
1855 */
1856 trans = btrfs_start_transaction(root, 0);
1857 if (IS_ERR(trans))
1858 return PTR_ERR(trans);
1859 btrfs_add_ordered_operation(trans, BTRFS_I(inode)->root, inode);
1860 btrfs_end_transaction(trans, root);
1861 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1862 filemap_flush(inode->i_mapping);
1863 }
1864 if (filp->private_data)
1865 btrfs_ioctl_trans_end(filp);
1866 return 0;
1867}
1868
1869/*
1870 * fsync call for both files and directories. This logs the inode into
1871 * the tree log instead of forcing full commits whenever possible.
1872 *
1873 * It needs to call filemap_fdatawait so that all ordered extent updates are
1874 * in the metadata btree are up to date for copying to the log.
1875 *
1876 * It drops the inode mutex before doing the tree log commit. This is an
1877 * important optimization for directories because holding the mutex prevents
1878 * new operations on the dir while we write to disk.
1879 */
1880int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1881{
1882 struct dentry *dentry = file->f_path.dentry;
1883 struct inode *inode = dentry->d_inode;
1884 struct btrfs_root *root = BTRFS_I(inode)->root;
1885 struct btrfs_trans_handle *trans;
1886 struct btrfs_log_ctx ctx;
1887 int ret = 0;
1888 bool full_sync = 0;
1889
1890 trace_btrfs_sync_file(file, datasync);
1891
1892 /*
1893 * We write the dirty pages in the range and wait until they complete
1894 * out of the ->i_mutex. If so, we can flush the dirty pages by
1895 * multi-task, and make the performance up. See
1896 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1897 */
1898 atomic_inc(&BTRFS_I(inode)->sync_writers);
1899 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1900 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1901 &BTRFS_I(inode)->runtime_flags))
1902 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
1903 atomic_dec(&BTRFS_I(inode)->sync_writers);
1904 if (ret)
1905 return ret;
1906
1907 mutex_lock(&inode->i_mutex);
1908
1909 /*
1910 * We flush the dirty pages again to avoid some dirty pages in the
1911 * range being left.
1912 */
1913 atomic_inc(&root->log_batch);
1914 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1915 &BTRFS_I(inode)->runtime_flags);
1916 if (full_sync) {
1917 ret = btrfs_wait_ordered_range(inode, start, end - start + 1);
1918 if (ret) {
1919 mutex_unlock(&inode->i_mutex);
1920 goto out;
1921 }
1922 }
1923 atomic_inc(&root->log_batch);
1924
1925 /*
1926 * check the transaction that last modified this inode
1927 * and see if its already been committed
1928 */
1929 if (!BTRFS_I(inode)->last_trans) {
1930 mutex_unlock(&inode->i_mutex);
1931 goto out;
1932 }
1933
1934 /*
1935 * if the last transaction that changed this file was before
1936 * the current transaction, we can bail out now without any
1937 * syncing
1938 */
1939 smp_mb();
1940 if (btrfs_inode_in_log(inode, root->fs_info->generation) ||
1941 BTRFS_I(inode)->last_trans <=
1942 root->fs_info->last_trans_committed) {
1943 BTRFS_I(inode)->last_trans = 0;
1944
1945 /*
1946 * We'v had everything committed since the last time we were
1947 * modified so clear this flag in case it was set for whatever
1948 * reason, it's no longer relevant.
1949 */
1950 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1951 &BTRFS_I(inode)->runtime_flags);
1952 mutex_unlock(&inode->i_mutex);
1953 goto out;
1954 }
1955
1956 /*
1957 * ok we haven't committed the transaction yet, lets do a commit
1958 */
1959 if (file->private_data)
1960 btrfs_ioctl_trans_end(file);
1961
1962 /*
1963 * We use start here because we will need to wait on the IO to complete
1964 * in btrfs_sync_log, which could require joining a transaction (for
1965 * example checking cross references in the nocow path). If we use join
1966 * here we could get into a situation where we're waiting on IO to
1967 * happen that is blocked on a transaction trying to commit. With start
1968 * we inc the extwriter counter, so we wait for all extwriters to exit
1969 * before we start blocking join'ers. This comment is to keep somebody
1970 * from thinking they are super smart and changing this to
1971 * btrfs_join_transaction *cough*Josef*cough*.
1972 */
1973 trans = btrfs_start_transaction(root, 0);
1974 if (IS_ERR(trans)) {
1975 ret = PTR_ERR(trans);
1976 mutex_unlock(&inode->i_mutex);
1977 goto out;
1978 }
1979 trans->sync = true;
1980
1981 btrfs_init_log_ctx(&ctx);
1982
1983 ret = btrfs_log_dentry_safe(trans, root, dentry, &ctx);
1984 if (ret < 0) {
1985 /* Fallthrough and commit/free transaction. */
1986 ret = 1;
1987 }
1988
1989 /* we've logged all the items and now have a consistent
1990 * version of the file in the log. It is possible that
1991 * someone will come in and modify the file, but that's
1992 * fine because the log is consistent on disk, and we
1993 * have references to all of the file's extents
1994 *
1995 * It is possible that someone will come in and log the
1996 * file again, but that will end up using the synchronization
1997 * inside btrfs_sync_log to keep things safe.
1998 */
1999 mutex_unlock(&inode->i_mutex);
2000
2001 if (ret != BTRFS_NO_LOG_SYNC) {
2002 if (!ret) {
2003 ret = btrfs_sync_log(trans, root, &ctx);
2004 if (!ret) {
2005 ret = btrfs_end_transaction(trans, root);
2006 goto out;
2007 }
2008 }
2009 if (!full_sync) {
2010 ret = btrfs_wait_ordered_range(inode, start,
2011 end - start + 1);
2012 if (ret)
2013 goto out;
2014 }
2015 ret = btrfs_commit_transaction(trans, root);
2016 } else {
2017 ret = btrfs_end_transaction(trans, root);
2018 }
2019out:
2020 return ret > 0 ? -EIO : ret;
2021}
2022
2023static const struct vm_operations_struct btrfs_file_vm_ops = {
2024 .fault = filemap_fault,
2025 .map_pages = filemap_map_pages,
2026 .page_mkwrite = btrfs_page_mkwrite,
2027 .remap_pages = generic_file_remap_pages,
2028};
2029
2030static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
2031{
2032 struct address_space *mapping = filp->f_mapping;
2033
2034 if (!mapping->a_ops->readpage)
2035 return -ENOEXEC;
2036
2037 file_accessed(filp);
2038 vma->vm_ops = &btrfs_file_vm_ops;
2039
2040 return 0;
2041}
2042
2043static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf,
2044 int slot, u64 start, u64 end)
2045{
2046 struct btrfs_file_extent_item *fi;
2047 struct btrfs_key key;
2048
2049 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2050 return 0;
2051
2052 btrfs_item_key_to_cpu(leaf, &key, slot);
2053 if (key.objectid != btrfs_ino(inode) ||
2054 key.type != BTRFS_EXTENT_DATA_KEY)
2055 return 0;
2056
2057 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2058
2059 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2060 return 0;
2061
2062 if (btrfs_file_extent_disk_bytenr(leaf, fi))
2063 return 0;
2064
2065 if (key.offset == end)
2066 return 1;
2067 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2068 return 1;
2069 return 0;
2070}
2071
2072static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode,
2073 struct btrfs_path *path, u64 offset, u64 end)
2074{
2075 struct btrfs_root *root = BTRFS_I(inode)->root;
2076 struct extent_buffer *leaf;
2077 struct btrfs_file_extent_item *fi;
2078 struct extent_map *hole_em;
2079 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2080 struct btrfs_key key;
2081 int ret;
2082
2083 if (btrfs_fs_incompat(root->fs_info, NO_HOLES))
2084 goto out;
2085
2086 key.objectid = btrfs_ino(inode);
2087 key.type = BTRFS_EXTENT_DATA_KEY;
2088 key.offset = offset;
2089
2090 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2091 if (ret < 0)
2092 return ret;
2093 BUG_ON(!ret);
2094
2095 leaf = path->nodes[0];
2096 if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) {
2097 u64 num_bytes;
2098
2099 path->slots[0]--;
2100 fi = btrfs_item_ptr(leaf, path->slots[0],
2101 struct btrfs_file_extent_item);
2102 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2103 end - offset;
2104 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2105 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2106 btrfs_set_file_extent_offset(leaf, fi, 0);
2107 btrfs_mark_buffer_dirty(leaf);
2108 goto out;
2109 }
2110
2111 if (hole_mergeable(inode, leaf, path->slots[0]+1, offset, end)) {
2112 u64 num_bytes;
2113
2114 path->slots[0]++;
2115 key.offset = offset;
2116 btrfs_set_item_key_safe(root, path, &key);
2117 fi = btrfs_item_ptr(leaf, path->slots[0],
2118 struct btrfs_file_extent_item);
2119 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2120 offset;
2121 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2122 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2123 btrfs_set_file_extent_offset(leaf, fi, 0);
2124 btrfs_mark_buffer_dirty(leaf);
2125 goto out;
2126 }
2127 btrfs_release_path(path);
2128
2129 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
2130 0, 0, end - offset, 0, end - offset,
2131 0, 0, 0);
2132 if (ret)
2133 return ret;
2134
2135out:
2136 btrfs_release_path(path);
2137
2138 hole_em = alloc_extent_map();
2139 if (!hole_em) {
2140 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2141 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2142 &BTRFS_I(inode)->runtime_flags);
2143 } else {
2144 hole_em->start = offset;
2145 hole_em->len = end - offset;
2146 hole_em->ram_bytes = hole_em->len;
2147 hole_em->orig_start = offset;
2148
2149 hole_em->block_start = EXTENT_MAP_HOLE;
2150 hole_em->block_len = 0;
2151 hole_em->orig_block_len = 0;
2152 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
2153 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2154 hole_em->generation = trans->transid;
2155
2156 do {
2157 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2158 write_lock(&em_tree->lock);
2159 ret = add_extent_mapping(em_tree, hole_em, 1);
2160 write_unlock(&em_tree->lock);
2161 } while (ret == -EEXIST);
2162 free_extent_map(hole_em);
2163 if (ret)
2164 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2165 &BTRFS_I(inode)->runtime_flags);
2166 }
2167
2168 return 0;
2169}
2170
2171static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2172{
2173 struct btrfs_root *root = BTRFS_I(inode)->root;
2174 struct extent_state *cached_state = NULL;
2175 struct btrfs_path *path;
2176 struct btrfs_block_rsv *rsv;
2177 struct btrfs_trans_handle *trans;
2178 u64 lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize);
2179 u64 lockend = round_down(offset + len,
2180 BTRFS_I(inode)->root->sectorsize) - 1;
2181 u64 cur_offset = lockstart;
2182 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
2183 u64 drop_end;
2184 int ret = 0;
2185 int err = 0;
2186 int rsv_count;
2187 bool same_page = ((offset >> PAGE_CACHE_SHIFT) ==
2188 ((offset + len - 1) >> PAGE_CACHE_SHIFT));
2189 bool no_holes = btrfs_fs_incompat(root->fs_info, NO_HOLES);
2190 u64 ino_size = round_up(inode->i_size, PAGE_CACHE_SIZE);
2191
2192 ret = btrfs_wait_ordered_range(inode, offset, len);
2193 if (ret)
2194 return ret;
2195
2196 mutex_lock(&inode->i_mutex);
2197 /*
2198 * We needn't truncate any page which is beyond the end of the file
2199 * because we are sure there is no data there.
2200 */
2201 /*
2202 * Only do this if we are in the same page and we aren't doing the
2203 * entire page.
2204 */
2205 if (same_page && len < PAGE_CACHE_SIZE) {
2206 if (offset < ino_size)
2207 ret = btrfs_truncate_page(inode, offset, len, 0);
2208 mutex_unlock(&inode->i_mutex);
2209 return ret;
2210 }
2211
2212 /* zero back part of the first page */
2213 if (offset < ino_size) {
2214 ret = btrfs_truncate_page(inode, offset, 0, 0);
2215 if (ret) {
2216 mutex_unlock(&inode->i_mutex);
2217 return ret;
2218 }
2219 }
2220
2221 /* zero the front end of the last page */
2222 if (offset + len < ino_size) {
2223 ret = btrfs_truncate_page(inode, offset + len, 0, 1);
2224 if (ret) {
2225 mutex_unlock(&inode->i_mutex);
2226 return ret;
2227 }
2228 }
2229
2230 if (lockend < lockstart) {
2231 mutex_unlock(&inode->i_mutex);
2232 return 0;
2233 }
2234
2235 while (1) {
2236 struct btrfs_ordered_extent *ordered;
2237
2238 truncate_pagecache_range(inode, lockstart, lockend);
2239
2240 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2241 0, &cached_state);
2242 ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
2243
2244 /*
2245 * We need to make sure we have no ordered extents in this range
2246 * and nobody raced in and read a page in this range, if we did
2247 * we need to try again.
2248 */
2249 if ((!ordered ||
2250 (ordered->file_offset + ordered->len <= lockstart ||
2251 ordered->file_offset > lockend)) &&
2252 !test_range_bit(&BTRFS_I(inode)->io_tree, lockstart,
2253 lockend, EXTENT_UPTODATE, 0,
2254 cached_state)) {
2255 if (ordered)
2256 btrfs_put_ordered_extent(ordered);
2257 break;
2258 }
2259 if (ordered)
2260 btrfs_put_ordered_extent(ordered);
2261 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2262 lockend, &cached_state, GFP_NOFS);
2263 ret = btrfs_wait_ordered_range(inode, lockstart,
2264 lockend - lockstart + 1);
2265 if (ret) {
2266 mutex_unlock(&inode->i_mutex);
2267 return ret;
2268 }
2269 }
2270
2271 path = btrfs_alloc_path();
2272 if (!path) {
2273 ret = -ENOMEM;
2274 goto out;
2275 }
2276
2277 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2278 if (!rsv) {
2279 ret = -ENOMEM;
2280 goto out_free;
2281 }
2282 rsv->size = btrfs_calc_trunc_metadata_size(root, 1);
2283 rsv->failfast = 1;
2284
2285 /*
2286 * 1 - update the inode
2287 * 1 - removing the extents in the range
2288 * 1 - adding the hole extent if no_holes isn't set
2289 */
2290 rsv_count = no_holes ? 2 : 3;
2291 trans = btrfs_start_transaction(root, rsv_count);
2292 if (IS_ERR(trans)) {
2293 err = PTR_ERR(trans);
2294 goto out_free;
2295 }
2296
2297 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
2298 min_size);
2299 BUG_ON(ret);
2300 trans->block_rsv = rsv;
2301
2302 while (cur_offset < lockend) {
2303 ret = __btrfs_drop_extents(trans, root, inode, path,
2304 cur_offset, lockend + 1,
2305 &drop_end, 1, 0, 0, NULL);
2306 if (ret != -ENOSPC)
2307 break;
2308
2309 trans->block_rsv = &root->fs_info->trans_block_rsv;
2310
2311 if (cur_offset < ino_size) {
2312 ret = fill_holes(trans, inode, path, cur_offset,
2313 drop_end);
2314 if (ret) {
2315 err = ret;
2316 break;
2317 }
2318 }
2319
2320 cur_offset = drop_end;
2321
2322 ret = btrfs_update_inode(trans, root, inode);
2323 if (ret) {
2324 err = ret;
2325 break;
2326 }
2327
2328 btrfs_end_transaction(trans, root);
2329 btrfs_btree_balance_dirty(root);
2330
2331 trans = btrfs_start_transaction(root, rsv_count);
2332 if (IS_ERR(trans)) {
2333 ret = PTR_ERR(trans);
2334 trans = NULL;
2335 break;
2336 }
2337
2338 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
2339 rsv, min_size);
2340 BUG_ON(ret); /* shouldn't happen */
2341 trans->block_rsv = rsv;
2342 }
2343
2344 if (ret) {
2345 err = ret;
2346 goto out_trans;
2347 }
2348
2349 trans->block_rsv = &root->fs_info->trans_block_rsv;
2350 if (cur_offset < ino_size) {
2351 ret = fill_holes(trans, inode, path, cur_offset, drop_end);
2352 if (ret) {
2353 err = ret;
2354 goto out_trans;
2355 }
2356 }
2357
2358out_trans:
2359 if (!trans)
2360 goto out_free;
2361
2362 inode_inc_iversion(inode);
2363 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2364
2365 trans->block_rsv = &root->fs_info->trans_block_rsv;
2366 ret = btrfs_update_inode(trans, root, inode);
2367 btrfs_end_transaction(trans, root);
2368 btrfs_btree_balance_dirty(root);
2369out_free:
2370 btrfs_free_path(path);
2371 btrfs_free_block_rsv(root, rsv);
2372out:
2373 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2374 &cached_state, GFP_NOFS);
2375 mutex_unlock(&inode->i_mutex);
2376 if (ret && !err)
2377 err = ret;
2378 return err;
2379}
2380
2381static long btrfs_fallocate(struct file *file, int mode,
2382 loff_t offset, loff_t len)
2383{
2384 struct inode *inode = file_inode(file);
2385 struct extent_state *cached_state = NULL;
2386 struct btrfs_root *root = BTRFS_I(inode)->root;
2387 u64 cur_offset;
2388 u64 last_byte;
2389 u64 alloc_start;
2390 u64 alloc_end;
2391 u64 alloc_hint = 0;
2392 u64 locked_end;
2393 struct extent_map *em;
2394 int blocksize = BTRFS_I(inode)->root->sectorsize;
2395 int ret;
2396
2397 alloc_start = round_down(offset, blocksize);
2398 alloc_end = round_up(offset + len, blocksize);
2399
2400 /* Make sure we aren't being give some crap mode */
2401 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2402 return -EOPNOTSUPP;
2403
2404 if (mode & FALLOC_FL_PUNCH_HOLE)
2405 return btrfs_punch_hole(inode, offset, len);
2406
2407 /*
2408 * Make sure we have enough space before we do the
2409 * allocation.
2410 */
2411 ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
2412 if (ret)
2413 return ret;
2414 if (root->fs_info->quota_enabled) {
2415 ret = btrfs_qgroup_reserve(root, alloc_end - alloc_start);
2416 if (ret)
2417 goto out_reserve_fail;
2418 }
2419
2420 mutex_lock(&inode->i_mutex);
2421 ret = inode_newsize_ok(inode, alloc_end);
2422 if (ret)
2423 goto out;
2424
2425 if (alloc_start > inode->i_size) {
2426 ret = btrfs_cont_expand(inode, i_size_read(inode),
2427 alloc_start);
2428 if (ret)
2429 goto out;
2430 } else {
2431 /*
2432 * If we are fallocating from the end of the file onward we
2433 * need to zero out the end of the page if i_size lands in the
2434 * middle of a page.
2435 */
2436 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
2437 if (ret)
2438 goto out;
2439 }
2440
2441 /*
2442 * wait for ordered IO before we have any locks. We'll loop again
2443 * below with the locks held.
2444 */
2445 ret = btrfs_wait_ordered_range(inode, alloc_start,
2446 alloc_end - alloc_start);
2447 if (ret)
2448 goto out;
2449
2450 locked_end = alloc_end - 1;
2451 while (1) {
2452 struct btrfs_ordered_extent *ordered;
2453
2454 /* the extent lock is ordered inside the running
2455 * transaction
2456 */
2457 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2458 locked_end, 0, &cached_state);
2459 ordered = btrfs_lookup_first_ordered_extent(inode,
2460 alloc_end - 1);
2461 if (ordered &&
2462 ordered->file_offset + ordered->len > alloc_start &&
2463 ordered->file_offset < alloc_end) {
2464 btrfs_put_ordered_extent(ordered);
2465 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
2466 alloc_start, locked_end,
2467 &cached_state, GFP_NOFS);
2468 /*
2469 * we can't wait on the range with the transaction
2470 * running or with the extent lock held
2471 */
2472 ret = btrfs_wait_ordered_range(inode, alloc_start,
2473 alloc_end - alloc_start);
2474 if (ret)
2475 goto out;
2476 } else {
2477 if (ordered)
2478 btrfs_put_ordered_extent(ordered);
2479 break;
2480 }
2481 }
2482
2483 cur_offset = alloc_start;
2484 while (1) {
2485 u64 actual_end;
2486
2487 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2488 alloc_end - cur_offset, 0);
2489 if (IS_ERR_OR_NULL(em)) {
2490 if (!em)
2491 ret = -ENOMEM;
2492 else
2493 ret = PTR_ERR(em);
2494 break;
2495 }
2496 last_byte = min(extent_map_end(em), alloc_end);
2497 actual_end = min_t(u64, extent_map_end(em), offset + len);
2498 last_byte = ALIGN(last_byte, blocksize);
2499
2500 if (em->block_start == EXTENT_MAP_HOLE ||
2501 (cur_offset >= inode->i_size &&
2502 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2503 ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
2504 last_byte - cur_offset,
2505 1 << inode->i_blkbits,
2506 offset + len,
2507 &alloc_hint);
2508
2509 if (ret < 0) {
2510 free_extent_map(em);
2511 break;
2512 }
2513 } else if (actual_end > inode->i_size &&
2514 !(mode & FALLOC_FL_KEEP_SIZE)) {
2515 /*
2516 * We didn't need to allocate any more space, but we
2517 * still extended the size of the file so we need to
2518 * update i_size.
2519 */
2520 inode->i_ctime = CURRENT_TIME;
2521 i_size_write(inode, actual_end);
2522 btrfs_ordered_update_i_size(inode, actual_end, NULL);
2523 }
2524 free_extent_map(em);
2525
2526 cur_offset = last_byte;
2527 if (cur_offset >= alloc_end) {
2528 ret = 0;
2529 break;
2530 }
2531 }
2532 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
2533 &cached_state, GFP_NOFS);
2534out:
2535 mutex_unlock(&inode->i_mutex);
2536 if (root->fs_info->quota_enabled)
2537 btrfs_qgroup_free(root, alloc_end - alloc_start);
2538out_reserve_fail:
2539 /* Let go of our reservation. */
2540 btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
2541 return ret;
2542}
2543
2544static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
2545{
2546 struct btrfs_root *root = BTRFS_I(inode)->root;
2547 struct extent_map *em = NULL;
2548 struct extent_state *cached_state = NULL;
2549 u64 lockstart = *offset;
2550 u64 lockend = i_size_read(inode);
2551 u64 start = *offset;
2552 u64 len = i_size_read(inode);
2553 int ret = 0;
2554
2555 lockend = max_t(u64, root->sectorsize, lockend);
2556 if (lockend <= lockstart)
2557 lockend = lockstart + root->sectorsize;
2558
2559 lockend--;
2560 len = lockend - lockstart + 1;
2561
2562 len = max_t(u64, len, root->sectorsize);
2563 if (inode->i_size == 0)
2564 return -ENXIO;
2565
2566 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
2567 &cached_state);
2568
2569 while (start < inode->i_size) {
2570 em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
2571 if (IS_ERR(em)) {
2572 ret = PTR_ERR(em);
2573 em = NULL;
2574 break;
2575 }
2576
2577 if (whence == SEEK_HOLE &&
2578 (em->block_start == EXTENT_MAP_HOLE ||
2579 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2580 break;
2581 else if (whence == SEEK_DATA &&
2582 (em->block_start != EXTENT_MAP_HOLE &&
2583 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
2584 break;
2585
2586 start = em->start + em->len;
2587 free_extent_map(em);
2588 em = NULL;
2589 cond_resched();
2590 }
2591 free_extent_map(em);
2592 if (!ret) {
2593 if (whence == SEEK_DATA && start >= inode->i_size)
2594 ret = -ENXIO;
2595 else
2596 *offset = min_t(loff_t, start, inode->i_size);
2597 }
2598 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2599 &cached_state, GFP_NOFS);
2600 return ret;
2601}
2602
2603static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
2604{
2605 struct inode *inode = file->f_mapping->host;
2606 int ret;
2607
2608 mutex_lock(&inode->i_mutex);
2609 switch (whence) {
2610 case SEEK_END:
2611 case SEEK_CUR:
2612 offset = generic_file_llseek(file, offset, whence);
2613 goto out;
2614 case SEEK_DATA:
2615 case SEEK_HOLE:
2616 if (offset >= i_size_read(inode)) {
2617 mutex_unlock(&inode->i_mutex);
2618 return -ENXIO;
2619 }
2620
2621 ret = find_desired_extent(inode, &offset, whence);
2622 if (ret) {
2623 mutex_unlock(&inode->i_mutex);
2624 return ret;
2625 }
2626 }
2627
2628 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2629out:
2630 mutex_unlock(&inode->i_mutex);
2631 return offset;
2632}
2633
2634const struct file_operations btrfs_file_operations = {
2635 .llseek = btrfs_file_llseek,
2636 .read = do_sync_read,
2637 .write = do_sync_write,
2638 .aio_read = generic_file_aio_read,
2639 .splice_read = generic_file_splice_read,
2640 .aio_write = btrfs_file_aio_write,
2641 .mmap = btrfs_file_mmap,
2642 .open = generic_file_open,
2643 .release = btrfs_release_file,
2644 .fsync = btrfs_sync_file,
2645 .fallocate = btrfs_fallocate,
2646 .unlocked_ioctl = btrfs_ioctl,
2647#ifdef CONFIG_COMPAT
2648 .compat_ioctl = btrfs_ioctl,
2649#endif
2650};
2651
2652void btrfs_auto_defrag_exit(void)
2653{
2654 if (btrfs_inode_defrag_cachep)
2655 kmem_cache_destroy(btrfs_inode_defrag_cachep);
2656}
2657
2658int btrfs_auto_defrag_init(void)
2659{
2660 btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
2661 sizeof(struct inode_defrag), 0,
2662 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2663 NULL);
2664 if (!btrfs_inode_defrag_cachep)
2665 return -ENOMEM;
2666
2667 return 0;
2668}