1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
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 <linux/iomap.h>
23 #include "transaction.h"
24 #include "btrfs_inode.h"
28 #include "compression.h"
29 #include "delalloc-space.h"
33 #include "accessors.h"
34 #include "extent-tree.h"
35 #include "file-item.h"
40 /* simple helper to fault in pages and copy. This should go away
41 * and be replaced with calls into generic code.
43 static noinline int btrfs_copy_from_user(loff_t pos, size_t write_bytes,
44 struct page **prepared_pages,
48 size_t total_copied = 0;
50 int offset = offset_in_page(pos);
52 while (write_bytes > 0) {
53 size_t count = min_t(size_t,
54 PAGE_SIZE - offset, write_bytes);
55 struct page *page = prepared_pages[pg];
57 * Copy data from userspace to the current page
59 copied = copy_page_from_iter_atomic(page, offset, count, i);
61 /* Flush processor's dcache for this page */
62 flush_dcache_page(page);
65 * if we get a partial write, we can end up with
66 * partially up to date pages. These add
67 * a lot of complexity, so make sure they don't
68 * happen by forcing this copy to be retried.
70 * The rest of the btrfs_file_write code will fall
71 * back to page at a time copies after we return 0.
73 if (unlikely(copied < count)) {
74 if (!PageUptodate(page)) {
75 iov_iter_revert(i, copied);
82 write_bytes -= copied;
83 total_copied += copied;
85 if (offset == PAGE_SIZE) {
94 * unlocks pages after btrfs_file_write is done with them
96 static void btrfs_drop_pages(struct btrfs_fs_info *fs_info,
97 struct page **pages, size_t num_pages,
101 u64 block_start = round_down(pos, fs_info->sectorsize);
102 u64 block_len = round_up(pos + copied, fs_info->sectorsize) - block_start;
104 ASSERT(block_len <= U32_MAX);
105 for (i = 0; i < num_pages; i++) {
106 /* page checked is some magic around finding pages that
107 * have been modified without going through btrfs_set_page_dirty
108 * clear it here. There should be no need to mark the pages
109 * accessed as prepare_pages should have marked them accessed
110 * in prepare_pages via find_or_create_page()
112 btrfs_folio_clamp_clear_checked(fs_info, page_folio(pages[i]),
113 block_start, block_len);
114 unlock_page(pages[i]);
120 * After btrfs_copy_from_user(), update the following things for delalloc:
121 * - Mark newly dirtied pages as DELALLOC in the io tree.
122 * Used to advise which range is to be written back.
123 * - Mark modified pages as Uptodate/Dirty and not needing COW fixup
124 * - Update inode size for past EOF write
126 int btrfs_dirty_pages(struct btrfs_inode *inode, struct page **pages,
127 size_t num_pages, loff_t pos, size_t write_bytes,
128 struct extent_state **cached, bool noreserve)
130 struct btrfs_fs_info *fs_info = inode->root->fs_info;
135 u64 end_of_last_block;
136 u64 end_pos = pos + write_bytes;
137 loff_t isize = i_size_read(&inode->vfs_inode);
138 unsigned int extra_bits = 0;
140 if (write_bytes == 0)
144 extra_bits |= EXTENT_NORESERVE;
146 start_pos = round_down(pos, fs_info->sectorsize);
147 num_bytes = round_up(write_bytes + pos - start_pos,
148 fs_info->sectorsize);
149 ASSERT(num_bytes <= U32_MAX);
151 end_of_last_block = start_pos + num_bytes - 1;
154 * The pages may have already been dirty, clear out old accounting so
155 * we can set things up properly
157 clear_extent_bit(&inode->io_tree, start_pos, end_of_last_block,
158 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
161 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
166 for (i = 0; i < num_pages; i++) {
167 struct page *p = pages[i];
169 btrfs_folio_clamp_set_uptodate(fs_info, page_folio(p),
170 start_pos, num_bytes);
171 btrfs_folio_clamp_clear_checked(fs_info, page_folio(p),
172 start_pos, num_bytes);
173 btrfs_folio_clamp_set_dirty(fs_info, page_folio(p),
174 start_pos, num_bytes);
178 * we've only changed i_size in ram, and we haven't updated
179 * the disk i_size. There is no need to log the inode
183 i_size_write(&inode->vfs_inode, end_pos);
188 * this is very complex, but the basic idea is to drop all extents
189 * in the range start - end. hint_block is filled in with a block number
190 * that would be a good hint to the block allocator for this file.
192 * If an extent intersects the range but is not entirely inside the range
193 * it is either truncated or split. Anything entirely inside the range
194 * is deleted from the tree.
196 * Note: the VFS' inode number of bytes is not updated, it's up to the caller
197 * to deal with that. We set the field 'bytes_found' of the arguments structure
198 * with the number of allocated bytes found in the target range, so that the
199 * caller can update the inode's number of bytes in an atomic way when
200 * replacing extents in a range to avoid races with stat(2).
202 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
203 struct btrfs_root *root, struct btrfs_inode *inode,
204 struct btrfs_drop_extents_args *args)
206 struct btrfs_fs_info *fs_info = root->fs_info;
207 struct extent_buffer *leaf;
208 struct btrfs_file_extent_item *fi;
209 struct btrfs_ref ref = { 0 };
210 struct btrfs_key key;
211 struct btrfs_key new_key;
212 u64 ino = btrfs_ino(inode);
213 u64 search_start = args->start;
216 u64 extent_offset = 0;
218 u64 last_end = args->start;
224 int modify_tree = -1;
227 struct btrfs_path *path = args->path;
229 args->bytes_found = 0;
230 args->extent_inserted = false;
232 /* Must always have a path if ->replace_extent is true */
233 ASSERT(!(args->replace_extent && !args->path));
236 path = btrfs_alloc_path();
243 if (args->drop_cache)
244 btrfs_drop_extent_map_range(inode, args->start, args->end - 1, false);
246 if (args->start >= inode->disk_i_size && !args->replace_extent)
249 update_refs = (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID);
252 ret = btrfs_lookup_file_extent(trans, root, path, ino,
253 search_start, modify_tree);
256 if (ret > 0 && path->slots[0] > 0 && search_start == args->start) {
257 leaf = path->nodes[0];
258 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
259 if (key.objectid == ino &&
260 key.type == BTRFS_EXTENT_DATA_KEY)
265 leaf = path->nodes[0];
266 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
268 ret = btrfs_next_leaf(root, path);
275 leaf = path->nodes[0];
279 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
281 if (key.objectid > ino)
283 if (WARN_ON_ONCE(key.objectid < ino) ||
284 key.type < BTRFS_EXTENT_DATA_KEY) {
289 if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= args->end)
292 fi = btrfs_item_ptr(leaf, path->slots[0],
293 struct btrfs_file_extent_item);
294 extent_type = btrfs_file_extent_type(leaf, fi);
296 if (extent_type == BTRFS_FILE_EXTENT_REG ||
297 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
298 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
299 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
300 extent_offset = btrfs_file_extent_offset(leaf, fi);
301 extent_end = key.offset +
302 btrfs_file_extent_num_bytes(leaf, fi);
303 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
304 extent_end = key.offset +
305 btrfs_file_extent_ram_bytes(leaf, fi);
312 * Don't skip extent items representing 0 byte lengths. They
313 * used to be created (bug) if while punching holes we hit
314 * -ENOSPC condition. So if we find one here, just ensure we
315 * delete it, otherwise we would insert a new file extent item
316 * with the same key (offset) as that 0 bytes length file
317 * extent item in the call to setup_items_for_insert() later
320 if (extent_end == key.offset && extent_end >= search_start) {
321 last_end = extent_end;
322 goto delete_extent_item;
325 if (extent_end <= search_start) {
331 search_start = max(key.offset, args->start);
332 if (recow || !modify_tree) {
334 btrfs_release_path(path);
339 * | - range to drop - |
340 * | -------- extent -------- |
342 if (args->start > key.offset && args->end < extent_end) {
344 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
349 memcpy(&new_key, &key, sizeof(new_key));
350 new_key.offset = args->start;
351 ret = btrfs_duplicate_item(trans, root, path,
353 if (ret == -EAGAIN) {
354 btrfs_release_path(path);
360 leaf = path->nodes[0];
361 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
362 struct btrfs_file_extent_item);
363 btrfs_set_file_extent_num_bytes(leaf, fi,
364 args->start - key.offset);
366 fi = btrfs_item_ptr(leaf, path->slots[0],
367 struct btrfs_file_extent_item);
369 extent_offset += args->start - key.offset;
370 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
371 btrfs_set_file_extent_num_bytes(leaf, fi,
372 extent_end - args->start);
373 btrfs_mark_buffer_dirty(trans, leaf);
375 if (update_refs && disk_bytenr > 0) {
376 btrfs_init_generic_ref(&ref,
377 BTRFS_ADD_DELAYED_REF,
378 disk_bytenr, num_bytes, 0,
379 root->root_key.objectid);
380 btrfs_init_data_ref(&ref,
381 root->root_key.objectid,
383 args->start - extent_offset,
385 ret = btrfs_inc_extent_ref(trans, &ref);
387 btrfs_abort_transaction(trans, ret);
391 key.offset = args->start;
394 * From here on out we will have actually dropped something, so
395 * last_end can be updated.
397 last_end = extent_end;
400 * | ---- range to drop ----- |
401 * | -------- extent -------- |
403 if (args->start <= key.offset && args->end < extent_end) {
404 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
409 memcpy(&new_key, &key, sizeof(new_key));
410 new_key.offset = args->end;
411 btrfs_set_item_key_safe(trans, path, &new_key);
413 extent_offset += args->end - key.offset;
414 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
415 btrfs_set_file_extent_num_bytes(leaf, fi,
416 extent_end - args->end);
417 btrfs_mark_buffer_dirty(trans, leaf);
418 if (update_refs && disk_bytenr > 0)
419 args->bytes_found += args->end - key.offset;
423 search_start = extent_end;
425 * | ---- range to drop ----- |
426 * | -------- extent -------- |
428 if (args->start > key.offset && args->end >= extent_end) {
430 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
435 btrfs_set_file_extent_num_bytes(leaf, fi,
436 args->start - key.offset);
437 btrfs_mark_buffer_dirty(trans, leaf);
438 if (update_refs && disk_bytenr > 0)
439 args->bytes_found += extent_end - args->start;
440 if (args->end == extent_end)
448 * | ---- range to drop ----- |
449 * | ------ extent ------ |
451 if (args->start <= key.offset && args->end >= extent_end) {
454 del_slot = path->slots[0];
457 BUG_ON(del_slot + del_nr != path->slots[0]);
462 extent_type == BTRFS_FILE_EXTENT_INLINE) {
463 args->bytes_found += extent_end - key.offset;
464 extent_end = ALIGN(extent_end,
465 fs_info->sectorsize);
466 } else if (update_refs && disk_bytenr > 0) {
467 btrfs_init_generic_ref(&ref,
468 BTRFS_DROP_DELAYED_REF,
469 disk_bytenr, num_bytes, 0,
470 root->root_key.objectid);
471 btrfs_init_data_ref(&ref,
472 root->root_key.objectid,
474 key.offset - extent_offset, 0,
476 ret = btrfs_free_extent(trans, &ref);
478 btrfs_abort_transaction(trans, ret);
481 args->bytes_found += extent_end - key.offset;
484 if (args->end == extent_end)
487 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
492 ret = btrfs_del_items(trans, root, path, del_slot,
495 btrfs_abort_transaction(trans, ret);
502 btrfs_release_path(path);
509 if (!ret && del_nr > 0) {
511 * Set path->slots[0] to first slot, so that after the delete
512 * if items are move off from our leaf to its immediate left or
513 * right neighbor leafs, we end up with a correct and adjusted
514 * path->slots[0] for our insertion (if args->replace_extent).
516 path->slots[0] = del_slot;
517 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
519 btrfs_abort_transaction(trans, ret);
522 leaf = path->nodes[0];
524 * If btrfs_del_items() was called, it might have deleted a leaf, in
525 * which case it unlocked our path, so check path->locks[0] matches a
528 if (!ret && args->replace_extent &&
529 path->locks[0] == BTRFS_WRITE_LOCK &&
530 btrfs_leaf_free_space(leaf) >=
531 sizeof(struct btrfs_item) + args->extent_item_size) {
534 key.type = BTRFS_EXTENT_DATA_KEY;
535 key.offset = args->start;
536 if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
537 struct btrfs_key slot_key;
539 btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
540 if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
543 btrfs_setup_item_for_insert(trans, root, path, &key,
544 args->extent_item_size);
545 args->extent_inserted = true;
549 btrfs_free_path(path);
550 else if (!args->extent_inserted)
551 btrfs_release_path(path);
553 args->drop_end = found ? min(args->end, last_end) : args->end;
558 static int extent_mergeable(struct extent_buffer *leaf, int slot,
559 u64 objectid, u64 bytenr, u64 orig_offset,
560 u64 *start, u64 *end)
562 struct btrfs_file_extent_item *fi;
563 struct btrfs_key key;
566 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
569 btrfs_item_key_to_cpu(leaf, &key, slot);
570 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
573 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
574 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
575 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
576 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
577 btrfs_file_extent_compression(leaf, fi) ||
578 btrfs_file_extent_encryption(leaf, fi) ||
579 btrfs_file_extent_other_encoding(leaf, fi))
582 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
583 if ((*start && *start != key.offset) || (*end && *end != extent_end))
592 * Mark extent in the range start - end as written.
594 * This changes extent type from 'pre-allocated' to 'regular'. If only
595 * part of extent is marked as written, the extent will be split into
598 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
599 struct btrfs_inode *inode, u64 start, u64 end)
601 struct btrfs_root *root = inode->root;
602 struct extent_buffer *leaf;
603 struct btrfs_path *path;
604 struct btrfs_file_extent_item *fi;
605 struct btrfs_ref ref = { 0 };
606 struct btrfs_key key;
607 struct btrfs_key new_key;
619 u64 ino = btrfs_ino(inode);
621 path = btrfs_alloc_path();
628 key.type = BTRFS_EXTENT_DATA_KEY;
631 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
634 if (ret > 0 && path->slots[0] > 0)
637 leaf = path->nodes[0];
638 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
639 if (key.objectid != ino ||
640 key.type != BTRFS_EXTENT_DATA_KEY) {
642 btrfs_abort_transaction(trans, ret);
645 fi = btrfs_item_ptr(leaf, path->slots[0],
646 struct btrfs_file_extent_item);
647 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_PREALLOC) {
649 btrfs_abort_transaction(trans, ret);
652 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
653 if (key.offset > start || extent_end < end) {
655 btrfs_abort_transaction(trans, ret);
659 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
660 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
661 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
662 memcpy(&new_key, &key, sizeof(new_key));
664 if (start == key.offset && end < extent_end) {
667 if (extent_mergeable(leaf, path->slots[0] - 1,
668 ino, bytenr, orig_offset,
669 &other_start, &other_end)) {
670 new_key.offset = end;
671 btrfs_set_item_key_safe(trans, path, &new_key);
672 fi = btrfs_item_ptr(leaf, path->slots[0],
673 struct btrfs_file_extent_item);
674 btrfs_set_file_extent_generation(leaf, fi,
676 btrfs_set_file_extent_num_bytes(leaf, fi,
678 btrfs_set_file_extent_offset(leaf, fi,
680 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
681 struct btrfs_file_extent_item);
682 btrfs_set_file_extent_generation(leaf, fi,
684 btrfs_set_file_extent_num_bytes(leaf, fi,
686 btrfs_mark_buffer_dirty(trans, leaf);
691 if (start > key.offset && end == extent_end) {
694 if (extent_mergeable(leaf, path->slots[0] + 1,
695 ino, bytenr, orig_offset,
696 &other_start, &other_end)) {
697 fi = btrfs_item_ptr(leaf, path->slots[0],
698 struct btrfs_file_extent_item);
699 btrfs_set_file_extent_num_bytes(leaf, fi,
701 btrfs_set_file_extent_generation(leaf, fi,
704 new_key.offset = start;
705 btrfs_set_item_key_safe(trans, path, &new_key);
707 fi = btrfs_item_ptr(leaf, path->slots[0],
708 struct btrfs_file_extent_item);
709 btrfs_set_file_extent_generation(leaf, fi,
711 btrfs_set_file_extent_num_bytes(leaf, fi,
713 btrfs_set_file_extent_offset(leaf, fi,
714 start - orig_offset);
715 btrfs_mark_buffer_dirty(trans, leaf);
720 while (start > key.offset || end < extent_end) {
721 if (key.offset == start)
724 new_key.offset = split;
725 ret = btrfs_duplicate_item(trans, root, path, &new_key);
726 if (ret == -EAGAIN) {
727 btrfs_release_path(path);
731 btrfs_abort_transaction(trans, ret);
735 leaf = path->nodes[0];
736 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
737 struct btrfs_file_extent_item);
738 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
739 btrfs_set_file_extent_num_bytes(leaf, fi,
742 fi = btrfs_item_ptr(leaf, path->slots[0],
743 struct btrfs_file_extent_item);
745 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
746 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
747 btrfs_set_file_extent_num_bytes(leaf, fi,
749 btrfs_mark_buffer_dirty(trans, leaf);
751 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, bytenr,
752 num_bytes, 0, root->root_key.objectid);
753 btrfs_init_data_ref(&ref, root->root_key.objectid, ino,
754 orig_offset, 0, false);
755 ret = btrfs_inc_extent_ref(trans, &ref);
757 btrfs_abort_transaction(trans, ret);
761 if (split == start) {
764 if (start != key.offset) {
766 btrfs_abort_transaction(trans, ret);
777 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
778 num_bytes, 0, root->root_key.objectid);
779 btrfs_init_data_ref(&ref, root->root_key.objectid, ino, orig_offset,
781 if (extent_mergeable(leaf, path->slots[0] + 1,
782 ino, bytenr, orig_offset,
783 &other_start, &other_end)) {
785 btrfs_release_path(path);
788 extent_end = other_end;
789 del_slot = path->slots[0] + 1;
791 ret = btrfs_free_extent(trans, &ref);
793 btrfs_abort_transaction(trans, ret);
799 if (extent_mergeable(leaf, path->slots[0] - 1,
800 ino, bytenr, orig_offset,
801 &other_start, &other_end)) {
803 btrfs_release_path(path);
806 key.offset = other_start;
807 del_slot = path->slots[0];
809 ret = btrfs_free_extent(trans, &ref);
811 btrfs_abort_transaction(trans, ret);
816 fi = btrfs_item_ptr(leaf, path->slots[0],
817 struct btrfs_file_extent_item);
818 btrfs_set_file_extent_type(leaf, fi,
819 BTRFS_FILE_EXTENT_REG);
820 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
821 btrfs_mark_buffer_dirty(trans, leaf);
823 fi = btrfs_item_ptr(leaf, del_slot - 1,
824 struct btrfs_file_extent_item);
825 btrfs_set_file_extent_type(leaf, fi,
826 BTRFS_FILE_EXTENT_REG);
827 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
828 btrfs_set_file_extent_num_bytes(leaf, fi,
829 extent_end - key.offset);
830 btrfs_mark_buffer_dirty(trans, leaf);
832 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
834 btrfs_abort_transaction(trans, ret);
839 btrfs_free_path(path);
844 * on error we return an unlocked page and the error value
845 * on success we return a locked page and 0
847 static int prepare_uptodate_page(struct inode *inode,
848 struct page *page, u64 pos,
851 struct folio *folio = page_folio(page);
854 if (((pos & (PAGE_SIZE - 1)) || force_uptodate) &&
855 !PageUptodate(page)) {
856 ret = btrfs_read_folio(NULL, folio);
860 if (!PageUptodate(page)) {
866 * Since btrfs_read_folio() will unlock the folio before it
867 * returns, there is a window where btrfs_release_folio() can be
868 * called to release the page. Here we check both inode
869 * mapping and PagePrivate() to make sure the page was not
872 * The private flag check is essential for subpage as we need
873 * to store extra bitmap using folio private.
875 if (page->mapping != inode->i_mapping || !folio_test_private(folio)) {
883 static fgf_t get_prepare_fgp_flags(bool nowait)
885 fgf_t fgp_flags = FGP_LOCK | FGP_ACCESSED | FGP_CREAT;
888 fgp_flags |= FGP_NOWAIT;
893 static gfp_t get_prepare_gfp_flags(struct inode *inode, bool nowait)
897 gfp = btrfs_alloc_write_mask(inode->i_mapping);
899 gfp &= ~__GFP_DIRECT_RECLAIM;
907 * this just gets pages into the page cache and locks them down.
909 static noinline int prepare_pages(struct inode *inode, struct page **pages,
910 size_t num_pages, loff_t pos,
911 size_t write_bytes, bool force_uptodate,
915 unsigned long index = pos >> PAGE_SHIFT;
916 gfp_t mask = get_prepare_gfp_flags(inode, nowait);
917 fgf_t fgp_flags = get_prepare_fgp_flags(nowait);
921 for (i = 0; i < num_pages; i++) {
923 pages[i] = pagecache_get_page(inode->i_mapping, index + i,
924 fgp_flags, mask | __GFP_WRITE);
934 err = set_page_extent_mapped(pages[i]);
941 err = prepare_uptodate_page(inode, pages[i], pos,
943 if (!err && i == num_pages - 1)
944 err = prepare_uptodate_page(inode, pages[i],
945 pos + write_bytes, false);
948 if (!nowait && err == -EAGAIN) {
955 wait_on_page_writeback(pages[i]);
961 unlock_page(pages[faili]);
962 put_page(pages[faili]);
970 * This function locks the extent and properly waits for data=ordered extents
971 * to finish before allowing the pages to be modified if need.
974 * 1 - the extent is locked
975 * 0 - the extent is not locked, and everything is OK
976 * -EAGAIN - need re-prepare the pages
977 * the other < 0 number - Something wrong happens
980 lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct page **pages,
981 size_t num_pages, loff_t pos,
983 u64 *lockstart, u64 *lockend, bool nowait,
984 struct extent_state **cached_state)
986 struct btrfs_fs_info *fs_info = inode->root->fs_info;
992 start_pos = round_down(pos, fs_info->sectorsize);
993 last_pos = round_up(pos + write_bytes, fs_info->sectorsize) - 1;
995 if (start_pos < inode->vfs_inode.i_size) {
996 struct btrfs_ordered_extent *ordered;
999 if (!try_lock_extent(&inode->io_tree, start_pos, last_pos,
1001 for (i = 0; i < num_pages; i++) {
1002 unlock_page(pages[i]);
1010 lock_extent(&inode->io_tree, start_pos, last_pos, cached_state);
1013 ordered = btrfs_lookup_ordered_range(inode, start_pos,
1014 last_pos - start_pos + 1);
1016 ordered->file_offset + ordered->num_bytes > start_pos &&
1017 ordered->file_offset <= last_pos) {
1018 unlock_extent(&inode->io_tree, start_pos, last_pos,
1020 for (i = 0; i < num_pages; i++) {
1021 unlock_page(pages[i]);
1024 btrfs_start_ordered_extent(ordered);
1025 btrfs_put_ordered_extent(ordered);
1029 btrfs_put_ordered_extent(ordered);
1031 *lockstart = start_pos;
1032 *lockend = last_pos;
1037 * We should be called after prepare_pages() which should have locked
1038 * all pages in the range.
1040 for (i = 0; i < num_pages; i++)
1041 WARN_ON(!PageLocked(pages[i]));
1047 * Check if we can do nocow write into the range [@pos, @pos + @write_bytes)
1049 * @pos: File offset.
1050 * @write_bytes: The length to write, will be updated to the nocow writeable
1053 * This function will flush ordered extents in the range to ensure proper
1057 * > 0 If we can nocow, and updates @write_bytes.
1058 * 0 If we can't do a nocow write.
1059 * -EAGAIN If we can't do a nocow write because snapshoting of the inode's
1060 * root is in progress.
1061 * < 0 If an error happened.
1063 * NOTE: Callers need to call btrfs_check_nocow_unlock() if we return > 0.
1065 int btrfs_check_nocow_lock(struct btrfs_inode *inode, loff_t pos,
1066 size_t *write_bytes, bool nowait)
1068 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1069 struct btrfs_root *root = inode->root;
1070 struct extent_state *cached_state = NULL;
1071 u64 lockstart, lockend;
1075 if (!(inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
1078 if (!btrfs_drew_try_write_lock(&root->snapshot_lock))
1081 lockstart = round_down(pos, fs_info->sectorsize);
1082 lockend = round_up(pos + *write_bytes,
1083 fs_info->sectorsize) - 1;
1084 num_bytes = lockend - lockstart + 1;
1087 if (!btrfs_try_lock_ordered_range(inode, lockstart, lockend,
1089 btrfs_drew_write_unlock(&root->snapshot_lock);
1093 btrfs_lock_and_flush_ordered_range(inode, lockstart, lockend,
1096 ret = can_nocow_extent(&inode->vfs_inode, lockstart, &num_bytes,
1097 NULL, NULL, NULL, nowait, false);
1099 btrfs_drew_write_unlock(&root->snapshot_lock);
1101 *write_bytes = min_t(size_t, *write_bytes ,
1102 num_bytes - pos + lockstart);
1103 unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
1108 void btrfs_check_nocow_unlock(struct btrfs_inode *inode)
1110 btrfs_drew_write_unlock(&inode->root->snapshot_lock);
1113 static void update_time_for_write(struct inode *inode)
1115 struct timespec64 now, ts;
1117 if (IS_NOCMTIME(inode))
1120 now = current_time(inode);
1121 ts = inode_get_mtime(inode);
1122 if (!timespec64_equal(&ts, &now))
1123 inode_set_mtime_to_ts(inode, now);
1125 ts = inode_get_ctime(inode);
1126 if (!timespec64_equal(&ts, &now))
1127 inode_set_ctime_to_ts(inode, now);
1129 if (IS_I_VERSION(inode))
1130 inode_inc_iversion(inode);
1133 static int btrfs_write_check(struct kiocb *iocb, struct iov_iter *from,
1136 struct file *file = iocb->ki_filp;
1137 struct inode *inode = file_inode(file);
1138 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
1139 loff_t pos = iocb->ki_pos;
1145 * Quickly bail out on NOWAIT writes if we don't have the nodatacow or
1146 * prealloc flags, as without those flags we always have to COW. We will
1147 * later check if we can really COW into the target range (using
1148 * can_nocow_extent() at btrfs_get_blocks_direct_write()).
1150 if ((iocb->ki_flags & IOCB_NOWAIT) &&
1151 !(BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
1154 ret = file_remove_privs(file);
1159 * We reserve space for updating the inode when we reserve space for the
1160 * extent we are going to write, so we will enospc out there. We don't
1161 * need to start yet another transaction to update the inode as we will
1162 * update the inode when we finish writing whatever data we write.
1164 update_time_for_write(inode);
1166 start_pos = round_down(pos, fs_info->sectorsize);
1167 oldsize = i_size_read(inode);
1168 if (start_pos > oldsize) {
1169 /* Expand hole size to cover write data, preventing empty gap */
1170 loff_t end_pos = round_up(pos + count, fs_info->sectorsize);
1172 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, end_pos);
1180 static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb,
1183 struct file *file = iocb->ki_filp;
1185 struct inode *inode = file_inode(file);
1186 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
1187 struct page **pages = NULL;
1188 struct extent_changeset *data_reserved = NULL;
1189 u64 release_bytes = 0;
1192 size_t num_written = 0;
1195 bool only_release_metadata = false;
1196 bool force_page_uptodate = false;
1197 loff_t old_isize = i_size_read(inode);
1198 unsigned int ilock_flags = 0;
1199 const bool nowait = (iocb->ki_flags & IOCB_NOWAIT);
1200 unsigned int bdp_flags = (nowait ? BDP_ASYNC : 0);
1203 ilock_flags |= BTRFS_ILOCK_TRY;
1205 ret = btrfs_inode_lock(BTRFS_I(inode), ilock_flags);
1209 ret = generic_write_checks(iocb, i);
1213 ret = btrfs_write_check(iocb, i, ret);
1218 nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_SIZE),
1219 PAGE_SIZE / (sizeof(struct page *)));
1220 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1221 nrptrs = max(nrptrs, 8);
1222 pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL);
1228 while (iov_iter_count(i) > 0) {
1229 struct extent_state *cached_state = NULL;
1230 size_t offset = offset_in_page(pos);
1231 size_t sector_offset;
1232 size_t write_bytes = min(iov_iter_count(i),
1233 nrptrs * (size_t)PAGE_SIZE -
1236 size_t reserve_bytes;
1239 size_t dirty_sectors;
1244 * Fault pages before locking them in prepare_pages
1245 * to avoid recursive lock
1247 if (unlikely(fault_in_iov_iter_readable(i, write_bytes))) {
1252 only_release_metadata = false;
1253 sector_offset = pos & (fs_info->sectorsize - 1);
1255 extent_changeset_release(data_reserved);
1256 ret = btrfs_check_data_free_space(BTRFS_I(inode),
1257 &data_reserved, pos,
1258 write_bytes, nowait);
1262 if (nowait && (ret == -ENOSPC || ret == -EAGAIN)) {
1268 * If we don't have to COW at the offset, reserve
1269 * metadata only. write_bytes may get smaller than
1272 can_nocow = btrfs_check_nocow_lock(BTRFS_I(inode), pos,
1273 &write_bytes, nowait);
1280 only_release_metadata = true;
1283 num_pages = DIV_ROUND_UP(write_bytes + offset, PAGE_SIZE);
1284 WARN_ON(num_pages > nrptrs);
1285 reserve_bytes = round_up(write_bytes + sector_offset,
1286 fs_info->sectorsize);
1287 WARN_ON(reserve_bytes == 0);
1288 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
1290 reserve_bytes, nowait);
1292 if (!only_release_metadata)
1293 btrfs_free_reserved_data_space(BTRFS_I(inode),
1297 btrfs_check_nocow_unlock(BTRFS_I(inode));
1299 if (nowait && ret == -ENOSPC)
1304 release_bytes = reserve_bytes;
1306 ret = balance_dirty_pages_ratelimited_flags(inode->i_mapping, bdp_flags);
1308 btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes);
1313 * This is going to setup the pages array with the number of
1314 * pages we want, so we don't really need to worry about the
1315 * contents of pages from loop to loop
1317 ret = prepare_pages(inode, pages, num_pages,
1318 pos, write_bytes, force_page_uptodate, false);
1320 btrfs_delalloc_release_extents(BTRFS_I(inode),
1325 extents_locked = lock_and_cleanup_extent_if_need(
1326 BTRFS_I(inode), pages,
1327 num_pages, pos, write_bytes, &lockstart,
1328 &lockend, nowait, &cached_state);
1329 if (extents_locked < 0) {
1330 if (!nowait && extents_locked == -EAGAIN)
1333 btrfs_delalloc_release_extents(BTRFS_I(inode),
1335 ret = extents_locked;
1339 copied = btrfs_copy_from_user(pos, write_bytes, pages, i);
1341 num_sectors = BTRFS_BYTES_TO_BLKS(fs_info, reserve_bytes);
1342 dirty_sectors = round_up(copied + sector_offset,
1343 fs_info->sectorsize);
1344 dirty_sectors = BTRFS_BYTES_TO_BLKS(fs_info, dirty_sectors);
1347 * if we have trouble faulting in the pages, fall
1348 * back to one page at a time
1350 if (copied < write_bytes)
1354 force_page_uptodate = true;
1358 force_page_uptodate = false;
1359 dirty_pages = DIV_ROUND_UP(copied + offset,
1363 if (num_sectors > dirty_sectors) {
1364 /* release everything except the sectors we dirtied */
1365 release_bytes -= dirty_sectors << fs_info->sectorsize_bits;
1366 if (only_release_metadata) {
1367 btrfs_delalloc_release_metadata(BTRFS_I(inode),
1368 release_bytes, true);
1372 __pos = round_down(pos,
1373 fs_info->sectorsize) +
1374 (dirty_pages << PAGE_SHIFT);
1375 btrfs_delalloc_release_space(BTRFS_I(inode),
1376 data_reserved, __pos,
1377 release_bytes, true);
1381 release_bytes = round_up(copied + sector_offset,
1382 fs_info->sectorsize);
1384 ret = btrfs_dirty_pages(BTRFS_I(inode), pages,
1385 dirty_pages, pos, copied,
1386 &cached_state, only_release_metadata);
1389 * If we have not locked the extent range, because the range's
1390 * start offset is >= i_size, we might still have a non-NULL
1391 * cached extent state, acquired while marking the extent range
1392 * as delalloc through btrfs_dirty_pages(). Therefore free any
1393 * possible cached extent state to avoid a memory leak.
1396 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart,
1397 lockend, &cached_state);
1399 free_extent_state(cached_state);
1401 btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes);
1403 btrfs_drop_pages(fs_info, pages, num_pages, pos, copied);
1408 if (only_release_metadata)
1409 btrfs_check_nocow_unlock(BTRFS_I(inode));
1411 btrfs_drop_pages(fs_info, pages, num_pages, pos, copied);
1416 num_written += copied;
1421 if (release_bytes) {
1422 if (only_release_metadata) {
1423 btrfs_check_nocow_unlock(BTRFS_I(inode));
1424 btrfs_delalloc_release_metadata(BTRFS_I(inode),
1425 release_bytes, true);
1427 btrfs_delalloc_release_space(BTRFS_I(inode),
1429 round_down(pos, fs_info->sectorsize),
1430 release_bytes, true);
1434 extent_changeset_free(data_reserved);
1435 if (num_written > 0) {
1436 pagecache_isize_extended(inode, old_isize, iocb->ki_pos);
1437 iocb->ki_pos += num_written;
1440 btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1441 return num_written ? num_written : ret;
1444 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
1445 const struct iov_iter *iter, loff_t offset)
1447 const u32 blocksize_mask = fs_info->sectorsize - 1;
1449 if (offset & blocksize_mask)
1452 if (iov_iter_alignment(iter) & blocksize_mask)
1458 static ssize_t btrfs_direct_write(struct kiocb *iocb, struct iov_iter *from)
1460 struct file *file = iocb->ki_filp;
1461 struct inode *inode = file_inode(file);
1462 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
1464 ssize_t written = 0;
1465 ssize_t written_buffered;
1466 size_t prev_left = 0;
1469 unsigned int ilock_flags = 0;
1470 struct iomap_dio *dio;
1472 if (iocb->ki_flags & IOCB_NOWAIT)
1473 ilock_flags |= BTRFS_ILOCK_TRY;
1476 * If the write DIO is within EOF, use a shared lock and also only if
1477 * security bits will likely not be dropped by file_remove_privs() called
1478 * from btrfs_write_check(). Either will need to be rechecked after the
1479 * lock was acquired.
1481 if (iocb->ki_pos + iov_iter_count(from) <= i_size_read(inode) && IS_NOSEC(inode))
1482 ilock_flags |= BTRFS_ILOCK_SHARED;
1485 err = btrfs_inode_lock(BTRFS_I(inode), ilock_flags);
1489 /* Shared lock cannot be used with security bits set. */
1490 if ((ilock_flags & BTRFS_ILOCK_SHARED) && !IS_NOSEC(inode)) {
1491 btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1492 ilock_flags &= ~BTRFS_ILOCK_SHARED;
1496 err = generic_write_checks(iocb, from);
1498 btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1502 err = btrfs_write_check(iocb, from, err);
1504 btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1510 * Re-check since file size may have changed just before taking the
1511 * lock or pos may have changed because of O_APPEND in generic_write_check()
1513 if ((ilock_flags & BTRFS_ILOCK_SHARED) &&
1514 pos + iov_iter_count(from) > i_size_read(inode)) {
1515 btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1516 ilock_flags &= ~BTRFS_ILOCK_SHARED;
1520 if (check_direct_IO(fs_info, from, pos)) {
1521 btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1526 * The iov_iter can be mapped to the same file range we are writing to.
1527 * If that's the case, then we will deadlock in the iomap code, because
1528 * it first calls our callback btrfs_dio_iomap_begin(), which will create
1529 * an ordered extent, and after that it will fault in the pages that the
1530 * iov_iter refers to. During the fault in we end up in the readahead
1531 * pages code (starting at btrfs_readahead()), which will lock the range,
1532 * find that ordered extent and then wait for it to complete (at
1533 * btrfs_lock_and_flush_ordered_range()), resulting in a deadlock since
1534 * obviously the ordered extent can never complete as we didn't submit
1535 * yet the respective bio(s). This always happens when the buffer is
1536 * memory mapped to the same file range, since the iomap DIO code always
1537 * invalidates pages in the target file range (after starting and waiting
1538 * for any writeback).
1540 * So here we disable page faults in the iov_iter and then retry if we
1541 * got -EFAULT, faulting in the pages before the retry.
1543 from->nofault = true;
1544 dio = btrfs_dio_write(iocb, from, written);
1545 from->nofault = false;
1548 * iomap_dio_complete() will call btrfs_sync_file() if we have a dsync
1549 * iocb, and that needs to lock the inode. So unlock it before calling
1550 * iomap_dio_complete() to avoid a deadlock.
1552 btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1554 if (IS_ERR_OR_NULL(dio))
1555 err = PTR_ERR_OR_ZERO(dio);
1557 err = iomap_dio_complete(dio);
1559 /* No increment (+=) because iomap returns a cumulative value. */
1563 if (iov_iter_count(from) > 0 && (err == -EFAULT || err > 0)) {
1564 const size_t left = iov_iter_count(from);
1566 * We have more data left to write. Try to fault in as many as
1567 * possible of the remainder pages and retry. We do this without
1568 * releasing and locking again the inode, to prevent races with
1571 * Also, in case the iov refers to pages in the file range of the
1572 * file we want to write to (due to a mmap), we could enter an
1573 * infinite loop if we retry after faulting the pages in, since
1574 * iomap will invalidate any pages in the range early on, before
1575 * it tries to fault in the pages of the iov. So we keep track of
1576 * how much was left of iov in the previous EFAULT and fallback
1577 * to buffered IO in case we haven't made any progress.
1579 if (left == prev_left) {
1582 fault_in_iov_iter_readable(from, left);
1589 * If 'err' is -ENOTBLK or we have not written all data, then it means
1590 * we must fallback to buffered IO.
1592 if ((err < 0 && err != -ENOTBLK) || !iov_iter_count(from))
1597 * If we are in a NOWAIT context, then return -EAGAIN to signal the caller
1598 * it must retry the operation in a context where blocking is acceptable,
1599 * because even if we end up not blocking during the buffered IO attempt
1600 * below, we will block when flushing and waiting for the IO.
1602 if (iocb->ki_flags & IOCB_NOWAIT) {
1608 written_buffered = btrfs_buffered_write(iocb, from);
1609 if (written_buffered < 0) {
1610 err = written_buffered;
1614 * Ensure all data is persisted. We want the next direct IO read to be
1615 * able to read what was just written.
1617 endbyte = pos + written_buffered - 1;
1618 err = btrfs_fdatawrite_range(inode, pos, endbyte);
1621 err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte);
1624 written += written_buffered;
1625 iocb->ki_pos = pos + written_buffered;
1626 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_SHIFT,
1627 endbyte >> PAGE_SHIFT);
1629 return err < 0 ? err : written;
1632 static ssize_t btrfs_encoded_write(struct kiocb *iocb, struct iov_iter *from,
1633 const struct btrfs_ioctl_encoded_io_args *encoded)
1635 struct file *file = iocb->ki_filp;
1636 struct inode *inode = file_inode(file);
1640 btrfs_inode_lock(BTRFS_I(inode), 0);
1641 count = encoded->len;
1642 ret = generic_write_checks_count(iocb, &count);
1643 if (ret == 0 && count != encoded->len) {
1645 * The write got truncated by generic_write_checks_count(). We
1646 * can't do a partial encoded write.
1650 if (ret || encoded->len == 0)
1653 ret = btrfs_write_check(iocb, from, encoded->len);
1657 ret = btrfs_do_encoded_write(iocb, from, encoded);
1659 btrfs_inode_unlock(BTRFS_I(inode), 0);
1663 ssize_t btrfs_do_write_iter(struct kiocb *iocb, struct iov_iter *from,
1664 const struct btrfs_ioctl_encoded_io_args *encoded)
1666 struct file *file = iocb->ki_filp;
1667 struct btrfs_inode *inode = BTRFS_I(file_inode(file));
1668 ssize_t num_written, num_sync;
1671 * If the fs flips readonly due to some impossible error, although we
1672 * have opened a file as writable, we have to stop this write operation
1673 * to ensure consistency.
1675 if (BTRFS_FS_ERROR(inode->root->fs_info))
1678 if (encoded && (iocb->ki_flags & IOCB_NOWAIT))
1682 num_written = btrfs_encoded_write(iocb, from, encoded);
1683 num_sync = encoded->len;
1684 } else if (iocb->ki_flags & IOCB_DIRECT) {
1685 num_written = btrfs_direct_write(iocb, from);
1686 num_sync = num_written;
1688 num_written = btrfs_buffered_write(iocb, from);
1689 num_sync = num_written;
1692 btrfs_set_inode_last_sub_trans(inode);
1695 num_sync = generic_write_sync(iocb, num_sync);
1697 num_written = num_sync;
1703 static ssize_t btrfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1705 return btrfs_do_write_iter(iocb, from, NULL);
1708 int btrfs_release_file(struct inode *inode, struct file *filp)
1710 struct btrfs_file_private *private = filp->private_data;
1713 kfree(private->filldir_buf);
1714 free_extent_state(private->llseek_cached_state);
1716 filp->private_data = NULL;
1720 * Set by setattr when we are about to truncate a file from a non-zero
1721 * size to a zero size. This tries to flush down new bytes that may
1722 * have been written if the application were using truncate to replace
1725 if (test_and_clear_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
1726 &BTRFS_I(inode)->runtime_flags))
1727 filemap_flush(inode->i_mapping);
1731 static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end)
1734 struct blk_plug plug;
1737 * This is only called in fsync, which would do synchronous writes, so
1738 * a plug can merge adjacent IOs as much as possible. Esp. in case of
1739 * multiple disks using raid profile, a large IO can be split to
1740 * several segments of stripe length (currently 64K).
1742 blk_start_plug(&plug);
1743 ret = btrfs_fdatawrite_range(inode, start, end);
1744 blk_finish_plug(&plug);
1749 static inline bool skip_inode_logging(const struct btrfs_log_ctx *ctx)
1751 struct btrfs_inode *inode = BTRFS_I(ctx->inode);
1752 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1754 if (btrfs_inode_in_log(inode, btrfs_get_fs_generation(fs_info)) &&
1755 list_empty(&ctx->ordered_extents))
1759 * If we are doing a fast fsync we can not bail out if the inode's
1760 * last_trans is <= then the last committed transaction, because we only
1761 * update the last_trans of the inode during ordered extent completion,
1762 * and for a fast fsync we don't wait for that, we only wait for the
1763 * writeback to complete.
1765 if (inode->last_trans <= btrfs_get_last_trans_committed(fs_info) &&
1766 (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) ||
1767 list_empty(&ctx->ordered_extents)))
1774 * fsync call for both files and directories. This logs the inode into
1775 * the tree log instead of forcing full commits whenever possible.
1777 * It needs to call filemap_fdatawait so that all ordered extent updates are
1778 * in the metadata btree are up to date for copying to the log.
1780 * It drops the inode mutex before doing the tree log commit. This is an
1781 * important optimization for directories because holding the mutex prevents
1782 * new operations on the dir while we write to disk.
1784 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1786 struct dentry *dentry = file_dentry(file);
1787 struct inode *inode = d_inode(dentry);
1788 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
1789 struct btrfs_root *root = BTRFS_I(inode)->root;
1790 struct btrfs_trans_handle *trans;
1791 struct btrfs_log_ctx ctx;
1796 trace_btrfs_sync_file(file, datasync);
1798 btrfs_init_log_ctx(&ctx, inode);
1801 * Always set the range to a full range, otherwise we can get into
1802 * several problems, from missing file extent items to represent holes
1803 * when not using the NO_HOLES feature, to log tree corruption due to
1804 * races between hole detection during logging and completion of ordered
1805 * extents outside the range, to missing checksums due to ordered extents
1806 * for which we flushed only a subset of their pages.
1810 len = (u64)LLONG_MAX + 1;
1813 * We write the dirty pages in the range and wait until they complete
1814 * out of the ->i_mutex. If so, we can flush the dirty pages by
1815 * multi-task, and make the performance up. See
1816 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1818 ret = start_ordered_ops(inode, start, end);
1822 btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
1824 atomic_inc(&root->log_batch);
1827 * Before we acquired the inode's lock and the mmap lock, someone may
1828 * have dirtied more pages in the target range. We need to make sure
1829 * that writeback for any such pages does not start while we are logging
1830 * the inode, because if it does, any of the following might happen when
1831 * we are not doing a full inode sync:
1833 * 1) We log an extent after its writeback finishes but before its
1834 * checksums are added to the csum tree, leading to -EIO errors
1835 * when attempting to read the extent after a log replay.
1837 * 2) We can end up logging an extent before its writeback finishes.
1838 * Therefore after the log replay we will have a file extent item
1839 * pointing to an unwritten extent (and no data checksums as well).
1841 * So trigger writeback for any eventual new dirty pages and then we
1842 * wait for all ordered extents to complete below.
1844 ret = start_ordered_ops(inode, start, end);
1846 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
1851 * Always check for the full sync flag while holding the inode's lock,
1852 * to avoid races with other tasks. The flag must be either set all the
1853 * time during logging or always off all the time while logging.
1854 * We check the flag here after starting delalloc above, because when
1855 * running delalloc the full sync flag may be set if we need to drop
1856 * extra extent map ranges due to temporary memory allocation failures.
1858 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1859 &BTRFS_I(inode)->runtime_flags);
1862 * We have to do this here to avoid the priority inversion of waiting on
1863 * IO of a lower priority task while holding a transaction open.
1865 * For a full fsync we wait for the ordered extents to complete while
1866 * for a fast fsync we wait just for writeback to complete, and then
1867 * attach the ordered extents to the transaction so that a transaction
1868 * commit waits for their completion, to avoid data loss if we fsync,
1869 * the current transaction commits before the ordered extents complete
1870 * and a power failure happens right after that.
1872 * For zoned filesystem, if a write IO uses a ZONE_APPEND command, the
1873 * logical address recorded in the ordered extent may change. We need
1874 * to wait for the IO to stabilize the logical address.
1876 if (full_sync || btrfs_is_zoned(fs_info)) {
1877 ret = btrfs_wait_ordered_range(inode, start, len);
1880 * Get our ordered extents as soon as possible to avoid doing
1881 * checksum lookups in the csum tree, and use instead the
1882 * checksums attached to the ordered extents.
1884 btrfs_get_ordered_extents_for_logging(BTRFS_I(inode),
1885 &ctx.ordered_extents);
1886 ret = filemap_fdatawait_range(inode->i_mapping, start, end);
1890 goto out_release_extents;
1892 atomic_inc(&root->log_batch);
1894 if (skip_inode_logging(&ctx)) {
1896 * We've had everything committed since the last time we were
1897 * modified so clear this flag in case it was set for whatever
1898 * reason, it's no longer relevant.
1900 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1901 &BTRFS_I(inode)->runtime_flags);
1903 * An ordered extent might have started before and completed
1904 * already with io errors, in which case the inode was not
1905 * updated and we end up here. So check the inode's mapping
1906 * for any errors that might have happened since we last
1907 * checked called fsync.
1909 ret = filemap_check_wb_err(inode->i_mapping, file->f_wb_err);
1910 goto out_release_extents;
1913 btrfs_init_log_ctx_scratch_eb(&ctx);
1916 * We use start here because we will need to wait on the IO to complete
1917 * in btrfs_sync_log, which could require joining a transaction (for
1918 * example checking cross references in the nocow path). If we use join
1919 * here we could get into a situation where we're waiting on IO to
1920 * happen that is blocked on a transaction trying to commit. With start
1921 * we inc the extwriter counter, so we wait for all extwriters to exit
1922 * before we start blocking joiners. This comment is to keep somebody
1923 * from thinking they are super smart and changing this to
1924 * btrfs_join_transaction *cough*Josef*cough*.
1926 trans = btrfs_start_transaction(root, 0);
1927 if (IS_ERR(trans)) {
1928 ret = PTR_ERR(trans);
1929 goto out_release_extents;
1931 trans->in_fsync = true;
1933 ret = btrfs_log_dentry_safe(trans, dentry, &ctx);
1935 * Scratch eb no longer needed, release before syncing log or commit
1936 * transaction, to avoid holding unnecessary memory during such long
1939 if (ctx.scratch_eb) {
1940 free_extent_buffer(ctx.scratch_eb);
1941 ctx.scratch_eb = NULL;
1943 btrfs_release_log_ctx_extents(&ctx);
1945 /* Fallthrough and commit/free transaction. */
1946 ret = BTRFS_LOG_FORCE_COMMIT;
1949 /* we've logged all the items and now have a consistent
1950 * version of the file in the log. It is possible that
1951 * someone will come in and modify the file, but that's
1952 * fine because the log is consistent on disk, and we
1953 * have references to all of the file's extents
1955 * It is possible that someone will come in and log the
1956 * file again, but that will end up using the synchronization
1957 * inside btrfs_sync_log to keep things safe.
1959 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
1961 if (ret == BTRFS_NO_LOG_SYNC) {
1962 ret = btrfs_end_transaction(trans);
1966 /* We successfully logged the inode, attempt to sync the log. */
1968 ret = btrfs_sync_log(trans, root, &ctx);
1970 ret = btrfs_end_transaction(trans);
1976 * At this point we need to commit the transaction because we had
1977 * btrfs_need_log_full_commit() or some other error.
1979 * If we didn't do a full sync we have to stop the trans handle, wait on
1980 * the ordered extents, start it again and commit the transaction. If
1981 * we attempt to wait on the ordered extents here we could deadlock with
1982 * something like fallocate() that is holding the extent lock trying to
1983 * start a transaction while some other thread is trying to commit the
1984 * transaction while we (fsync) are currently holding the transaction
1988 ret = btrfs_end_transaction(trans);
1991 ret = btrfs_wait_ordered_range(inode, start, len);
1996 * This is safe to use here because we're only interested in
1997 * making sure the transaction that had the ordered extents is
1998 * committed. We aren't waiting on anything past this point,
1999 * we're purely getting the transaction and committing it.
2001 trans = btrfs_attach_transaction_barrier(root);
2002 if (IS_ERR(trans)) {
2003 ret = PTR_ERR(trans);
2006 * We committed the transaction and there's no currently
2007 * running transaction, this means everything we care
2008 * about made it to disk and we are done.
2016 ret = btrfs_commit_transaction(trans);
2018 free_extent_buffer(ctx.scratch_eb);
2019 ASSERT(list_empty(&ctx.list));
2020 ASSERT(list_empty(&ctx.conflict_inodes));
2021 err = file_check_and_advance_wb_err(file);
2024 return ret > 0 ? -EIO : ret;
2026 out_release_extents:
2027 btrfs_release_log_ctx_extents(&ctx);
2028 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
2032 static const struct vm_operations_struct btrfs_file_vm_ops = {
2033 .fault = filemap_fault,
2034 .map_pages = filemap_map_pages,
2035 .page_mkwrite = btrfs_page_mkwrite,
2038 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
2040 struct address_space *mapping = filp->f_mapping;
2042 if (!mapping->a_ops->read_folio)
2045 file_accessed(filp);
2046 vma->vm_ops = &btrfs_file_vm_ops;
2051 static int hole_mergeable(struct btrfs_inode *inode, struct extent_buffer *leaf,
2052 int slot, u64 start, u64 end)
2054 struct btrfs_file_extent_item *fi;
2055 struct btrfs_key key;
2057 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2060 btrfs_item_key_to_cpu(leaf, &key, slot);
2061 if (key.objectid != btrfs_ino(inode) ||
2062 key.type != BTRFS_EXTENT_DATA_KEY)
2065 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2067 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2070 if (btrfs_file_extent_disk_bytenr(leaf, fi))
2073 if (key.offset == end)
2075 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2080 static int fill_holes(struct btrfs_trans_handle *trans,
2081 struct btrfs_inode *inode,
2082 struct btrfs_path *path, u64 offset, u64 end)
2084 struct btrfs_fs_info *fs_info = trans->fs_info;
2085 struct btrfs_root *root = inode->root;
2086 struct extent_buffer *leaf;
2087 struct btrfs_file_extent_item *fi;
2088 struct extent_map *hole_em;
2089 struct btrfs_key key;
2092 if (btrfs_fs_incompat(fs_info, NO_HOLES))
2095 key.objectid = btrfs_ino(inode);
2096 key.type = BTRFS_EXTENT_DATA_KEY;
2097 key.offset = offset;
2099 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2102 * We should have dropped this offset, so if we find it then
2103 * something has gone horribly wrong.
2110 leaf = path->nodes[0];
2111 if (hole_mergeable(inode, leaf, path->slots[0] - 1, offset, end)) {
2115 fi = btrfs_item_ptr(leaf, path->slots[0],
2116 struct btrfs_file_extent_item);
2117 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2119 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2120 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2121 btrfs_set_file_extent_offset(leaf, fi, 0);
2122 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2123 btrfs_mark_buffer_dirty(trans, leaf);
2127 if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) {
2130 key.offset = offset;
2131 btrfs_set_item_key_safe(trans, path, &key);
2132 fi = btrfs_item_ptr(leaf, path->slots[0],
2133 struct btrfs_file_extent_item);
2134 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2136 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2137 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2138 btrfs_set_file_extent_offset(leaf, fi, 0);
2139 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2140 btrfs_mark_buffer_dirty(trans, leaf);
2143 btrfs_release_path(path);
2145 ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset,
2151 btrfs_release_path(path);
2153 hole_em = alloc_extent_map();
2155 btrfs_drop_extent_map_range(inode, offset, end - 1, false);
2156 btrfs_set_inode_full_sync(inode);
2158 hole_em->start = offset;
2159 hole_em->len = end - offset;
2160 hole_em->ram_bytes = hole_em->len;
2161 hole_em->orig_start = offset;
2163 hole_em->block_start = EXTENT_MAP_HOLE;
2164 hole_em->block_len = 0;
2165 hole_em->orig_block_len = 0;
2166 hole_em->generation = trans->transid;
2168 ret = btrfs_replace_extent_map_range(inode, hole_em, true);
2169 free_extent_map(hole_em);
2171 btrfs_set_inode_full_sync(inode);
2178 * Find a hole extent on given inode and change start/len to the end of hole
2179 * extent.(hole/vacuum extent whose em->start <= start &&
2180 * em->start + em->len > start)
2181 * When a hole extent is found, return 1 and modify start/len.
2183 static int find_first_non_hole(struct btrfs_inode *inode, u64 *start, u64 *len)
2185 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2186 struct extent_map *em;
2189 em = btrfs_get_extent(inode, NULL,
2190 round_down(*start, fs_info->sectorsize),
2191 round_up(*len, fs_info->sectorsize));
2195 /* Hole or vacuum extent(only exists in no-hole mode) */
2196 if (em->block_start == EXTENT_MAP_HOLE) {
2198 *len = em->start + em->len > *start + *len ?
2199 0 : *start + *len - em->start - em->len;
2200 *start = em->start + em->len;
2202 free_extent_map(em);
2206 static void btrfs_punch_hole_lock_range(struct inode *inode,
2207 const u64 lockstart,
2209 struct extent_state **cached_state)
2212 * For subpage case, if the range is not at page boundary, we could
2213 * have pages at the leading/tailing part of the range.
2214 * This could lead to dead loop since filemap_range_has_page()
2215 * will always return true.
2216 * So here we need to do extra page alignment for
2217 * filemap_range_has_page().
2219 const u64 page_lockstart = round_up(lockstart, PAGE_SIZE);
2220 const u64 page_lockend = round_down(lockend + 1, PAGE_SIZE) - 1;
2223 truncate_pagecache_range(inode, lockstart, lockend);
2225 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2228 * We can't have ordered extents in the range, nor dirty/writeback
2229 * pages, because we have locked the inode's VFS lock in exclusive
2230 * mode, we have locked the inode's i_mmap_lock in exclusive mode,
2231 * we have flushed all delalloc in the range and we have waited
2232 * for any ordered extents in the range to complete.
2233 * We can race with anyone reading pages from this range, so after
2234 * locking the range check if we have pages in the range, and if
2235 * we do, unlock the range and retry.
2237 if (!filemap_range_has_page(inode->i_mapping, page_lockstart,
2241 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2245 btrfs_assert_inode_range_clean(BTRFS_I(inode), lockstart, lockend);
2248 static int btrfs_insert_replace_extent(struct btrfs_trans_handle *trans,
2249 struct btrfs_inode *inode,
2250 struct btrfs_path *path,
2251 struct btrfs_replace_extent_info *extent_info,
2252 const u64 replace_len,
2253 const u64 bytes_to_drop)
2255 struct btrfs_fs_info *fs_info = trans->fs_info;
2256 struct btrfs_root *root = inode->root;
2257 struct btrfs_file_extent_item *extent;
2258 struct extent_buffer *leaf;
2259 struct btrfs_key key;
2261 struct btrfs_ref ref = { 0 };
2264 if (replace_len == 0)
2267 if (extent_info->disk_offset == 0 &&
2268 btrfs_fs_incompat(fs_info, NO_HOLES)) {
2269 btrfs_update_inode_bytes(inode, 0, bytes_to_drop);
2273 key.objectid = btrfs_ino(inode);
2274 key.type = BTRFS_EXTENT_DATA_KEY;
2275 key.offset = extent_info->file_offset;
2276 ret = btrfs_insert_empty_item(trans, root, path, &key,
2277 sizeof(struct btrfs_file_extent_item));
2280 leaf = path->nodes[0];
2281 slot = path->slots[0];
2282 write_extent_buffer(leaf, extent_info->extent_buf,
2283 btrfs_item_ptr_offset(leaf, slot),
2284 sizeof(struct btrfs_file_extent_item));
2285 extent = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2286 ASSERT(btrfs_file_extent_type(leaf, extent) != BTRFS_FILE_EXTENT_INLINE);
2287 btrfs_set_file_extent_offset(leaf, extent, extent_info->data_offset);
2288 btrfs_set_file_extent_num_bytes(leaf, extent, replace_len);
2289 if (extent_info->is_new_extent)
2290 btrfs_set_file_extent_generation(leaf, extent, trans->transid);
2291 btrfs_mark_buffer_dirty(trans, leaf);
2292 btrfs_release_path(path);
2294 ret = btrfs_inode_set_file_extent_range(inode, extent_info->file_offset,
2299 /* If it's a hole, nothing more needs to be done. */
2300 if (extent_info->disk_offset == 0) {
2301 btrfs_update_inode_bytes(inode, 0, bytes_to_drop);
2305 btrfs_update_inode_bytes(inode, replace_len, bytes_to_drop);
2307 if (extent_info->is_new_extent && extent_info->insertions == 0) {
2308 key.objectid = extent_info->disk_offset;
2309 key.type = BTRFS_EXTENT_ITEM_KEY;
2310 key.offset = extent_info->disk_len;
2311 ret = btrfs_alloc_reserved_file_extent(trans, root,
2313 extent_info->file_offset,
2314 extent_info->qgroup_reserved,
2319 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
2320 extent_info->disk_offset,
2321 extent_info->disk_len, 0,
2322 root->root_key.objectid);
2323 ref_offset = extent_info->file_offset - extent_info->data_offset;
2324 btrfs_init_data_ref(&ref, root->root_key.objectid,
2325 btrfs_ino(inode), ref_offset, 0, false);
2326 ret = btrfs_inc_extent_ref(trans, &ref);
2329 extent_info->insertions++;
2335 * The respective range must have been previously locked, as well as the inode.
2336 * The end offset is inclusive (last byte of the range).
2337 * @extent_info is NULL for fallocate's hole punching and non-NULL when replacing
2338 * the file range with an extent.
2339 * When not punching a hole, we don't want to end up in a state where we dropped
2340 * extents without inserting a new one, so we must abort the transaction to avoid
2343 int btrfs_replace_file_extents(struct btrfs_inode *inode,
2344 struct btrfs_path *path, const u64 start,
2346 struct btrfs_replace_extent_info *extent_info,
2347 struct btrfs_trans_handle **trans_out)
2349 struct btrfs_drop_extents_args drop_args = { 0 };
2350 struct btrfs_root *root = inode->root;
2351 struct btrfs_fs_info *fs_info = root->fs_info;
2352 u64 min_size = btrfs_calc_insert_metadata_size(fs_info, 1);
2353 u64 ino_size = round_up(inode->vfs_inode.i_size, fs_info->sectorsize);
2354 struct btrfs_trans_handle *trans = NULL;
2355 struct btrfs_block_rsv *rsv;
2356 unsigned int rsv_count;
2358 u64 len = end - start;
2364 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
2369 rsv->size = btrfs_calc_insert_metadata_size(fs_info, 1);
2370 rsv->failfast = true;
2373 * 1 - update the inode
2374 * 1 - removing the extents in the range
2375 * 1 - adding the hole extent if no_holes isn't set or if we are
2376 * replacing the range with a new extent
2378 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || extent_info)
2383 trans = btrfs_start_transaction(root, rsv_count);
2384 if (IS_ERR(trans)) {
2385 ret = PTR_ERR(trans);
2390 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
2394 trans->block_rsv = rsv;
2397 drop_args.path = path;
2398 drop_args.end = end + 1;
2399 drop_args.drop_cache = true;
2400 while (cur_offset < end) {
2401 drop_args.start = cur_offset;
2402 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2403 /* If we are punching a hole decrement the inode's byte count */
2405 btrfs_update_inode_bytes(inode, 0,
2406 drop_args.bytes_found);
2407 if (ret != -ENOSPC) {
2409 * The only time we don't want to abort is if we are
2410 * attempting to clone a partial inline extent, in which
2411 * case we'll get EOPNOTSUPP. However if we aren't
2412 * clone we need to abort no matter what, because if we
2413 * got EOPNOTSUPP via prealloc then we messed up and
2417 (ret != -EOPNOTSUPP ||
2418 (extent_info && extent_info->is_new_extent)))
2419 btrfs_abort_transaction(trans, ret);
2423 trans->block_rsv = &fs_info->trans_block_rsv;
2425 if (!extent_info && cur_offset < drop_args.drop_end &&
2426 cur_offset < ino_size) {
2427 ret = fill_holes(trans, inode, path, cur_offset,
2428 drop_args.drop_end);
2431 * If we failed then we didn't insert our hole
2432 * entries for the area we dropped, so now the
2433 * fs is corrupted, so we must abort the
2436 btrfs_abort_transaction(trans, ret);
2439 } else if (!extent_info && cur_offset < drop_args.drop_end) {
2441 * We are past the i_size here, but since we didn't
2442 * insert holes we need to clear the mapped area so we
2443 * know to not set disk_i_size in this area until a new
2444 * file extent is inserted here.
2446 ret = btrfs_inode_clear_file_extent_range(inode,
2448 drop_args.drop_end - cur_offset);
2451 * We couldn't clear our area, so we could
2452 * presumably adjust up and corrupt the fs, so
2455 btrfs_abort_transaction(trans, ret);
2461 drop_args.drop_end > extent_info->file_offset) {
2462 u64 replace_len = drop_args.drop_end -
2463 extent_info->file_offset;
2465 ret = btrfs_insert_replace_extent(trans, inode, path,
2466 extent_info, replace_len,
2467 drop_args.bytes_found);
2469 btrfs_abort_transaction(trans, ret);
2472 extent_info->data_len -= replace_len;
2473 extent_info->data_offset += replace_len;
2474 extent_info->file_offset += replace_len;
2478 * We are releasing our handle on the transaction, balance the
2479 * dirty pages of the btree inode and flush delayed items, and
2480 * then get a new transaction handle, which may now point to a
2481 * new transaction in case someone else may have committed the
2482 * transaction we used to replace/drop file extent items. So
2483 * bump the inode's iversion and update mtime and ctime except
2484 * if we are called from a dedupe context. This is because a
2485 * power failure/crash may happen after the transaction is
2486 * committed and before we finish replacing/dropping all the
2487 * file extent items we need.
2489 inode_inc_iversion(&inode->vfs_inode);
2491 if (!extent_info || extent_info->update_times)
2492 inode_set_mtime_to_ts(&inode->vfs_inode,
2493 inode_set_ctime_current(&inode->vfs_inode));
2495 ret = btrfs_update_inode(trans, inode);
2499 btrfs_end_transaction(trans);
2500 btrfs_btree_balance_dirty(fs_info);
2502 trans = btrfs_start_transaction(root, rsv_count);
2503 if (IS_ERR(trans)) {
2504 ret = PTR_ERR(trans);
2509 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
2510 rsv, min_size, false);
2513 trans->block_rsv = rsv;
2515 cur_offset = drop_args.drop_end;
2516 len = end - cur_offset;
2517 if (!extent_info && len) {
2518 ret = find_first_non_hole(inode, &cur_offset, &len);
2519 if (unlikely(ret < 0))
2529 * If we were cloning, force the next fsync to be a full one since we
2530 * we replaced (or just dropped in the case of cloning holes when
2531 * NO_HOLES is enabled) file extent items and did not setup new extent
2532 * maps for the replacement extents (or holes).
2534 if (extent_info && !extent_info->is_new_extent)
2535 btrfs_set_inode_full_sync(inode);
2540 trans->block_rsv = &fs_info->trans_block_rsv;
2542 * If we are using the NO_HOLES feature we might have had already an
2543 * hole that overlaps a part of the region [lockstart, lockend] and
2544 * ends at (or beyond) lockend. Since we have no file extent items to
2545 * represent holes, drop_end can be less than lockend and so we must
2546 * make sure we have an extent map representing the existing hole (the
2547 * call to __btrfs_drop_extents() might have dropped the existing extent
2548 * map representing the existing hole), otherwise the fast fsync path
2549 * will not record the existence of the hole region
2550 * [existing_hole_start, lockend].
2552 if (drop_args.drop_end <= end)
2553 drop_args.drop_end = end + 1;
2555 * Don't insert file hole extent item if it's for a range beyond eof
2556 * (because it's useless) or if it represents a 0 bytes range (when
2557 * cur_offset == drop_end).
2559 if (!extent_info && cur_offset < ino_size &&
2560 cur_offset < drop_args.drop_end) {
2561 ret = fill_holes(trans, inode, path, cur_offset,
2562 drop_args.drop_end);
2564 /* Same comment as above. */
2565 btrfs_abort_transaction(trans, ret);
2568 } else if (!extent_info && cur_offset < drop_args.drop_end) {
2569 /* See the comment in the loop above for the reasoning here. */
2570 ret = btrfs_inode_clear_file_extent_range(inode, cur_offset,
2571 drop_args.drop_end - cur_offset);
2573 btrfs_abort_transaction(trans, ret);
2579 ret = btrfs_insert_replace_extent(trans, inode, path,
2580 extent_info, extent_info->data_len,
2581 drop_args.bytes_found);
2583 btrfs_abort_transaction(trans, ret);
2592 trans->block_rsv = &fs_info->trans_block_rsv;
2594 btrfs_end_transaction(trans);
2598 btrfs_free_block_rsv(fs_info, rsv);
2603 static int btrfs_punch_hole(struct file *file, loff_t offset, loff_t len)
2605 struct inode *inode = file_inode(file);
2606 struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
2607 struct btrfs_root *root = BTRFS_I(inode)->root;
2608 struct extent_state *cached_state = NULL;
2609 struct btrfs_path *path;
2610 struct btrfs_trans_handle *trans = NULL;
2615 u64 orig_start = offset;
2619 bool truncated_block = false;
2620 bool updated_inode = false;
2622 btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
2624 ret = btrfs_wait_ordered_range(inode, offset, len);
2626 goto out_only_mutex;
2628 ino_size = round_up(inode->i_size, fs_info->sectorsize);
2629 ret = find_first_non_hole(BTRFS_I(inode), &offset, &len);
2631 goto out_only_mutex;
2633 /* Already in a large hole */
2635 goto out_only_mutex;
2638 ret = file_modified(file);
2640 goto out_only_mutex;
2642 lockstart = round_up(offset, fs_info->sectorsize);
2643 lockend = round_down(offset + len, fs_info->sectorsize) - 1;
2644 same_block = (BTRFS_BYTES_TO_BLKS(fs_info, offset))
2645 == (BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1));
2647 * We needn't truncate any block which is beyond the end of the file
2648 * because we are sure there is no data there.
2651 * Only do this if we are in the same block and we aren't doing the
2654 if (same_block && len < fs_info->sectorsize) {
2655 if (offset < ino_size) {
2656 truncated_block = true;
2657 ret = btrfs_truncate_block(BTRFS_I(inode), offset, len,
2662 goto out_only_mutex;
2665 /* zero back part of the first block */
2666 if (offset < ino_size) {
2667 truncated_block = true;
2668 ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0);
2670 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
2675 /* Check the aligned pages after the first unaligned page,
2676 * if offset != orig_start, which means the first unaligned page
2677 * including several following pages are already in holes,
2678 * the extra check can be skipped */
2679 if (offset == orig_start) {
2680 /* after truncate page, check hole again */
2681 len = offset + len - lockstart;
2683 ret = find_first_non_hole(BTRFS_I(inode), &offset, &len);
2685 goto out_only_mutex;
2688 goto out_only_mutex;
2693 /* Check the tail unaligned part is in a hole */
2694 tail_start = lockend + 1;
2695 tail_len = offset + len - tail_start;
2697 ret = find_first_non_hole(BTRFS_I(inode), &tail_start, &tail_len);
2698 if (unlikely(ret < 0))
2699 goto out_only_mutex;
2701 /* zero the front end of the last page */
2702 if (tail_start + tail_len < ino_size) {
2703 truncated_block = true;
2704 ret = btrfs_truncate_block(BTRFS_I(inode),
2705 tail_start + tail_len,
2708 goto out_only_mutex;
2713 if (lockend < lockstart) {
2715 goto out_only_mutex;
2718 btrfs_punch_hole_lock_range(inode, lockstart, lockend, &cached_state);
2720 path = btrfs_alloc_path();
2726 ret = btrfs_replace_file_extents(BTRFS_I(inode), path, lockstart,
2727 lockend, NULL, &trans);
2728 btrfs_free_path(path);
2732 ASSERT(trans != NULL);
2733 inode_inc_iversion(inode);
2734 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
2735 ret = btrfs_update_inode(trans, BTRFS_I(inode));
2736 updated_inode = true;
2737 btrfs_end_transaction(trans);
2738 btrfs_btree_balance_dirty(fs_info);
2740 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2743 if (!updated_inode && truncated_block && !ret) {
2745 * If we only end up zeroing part of a page, we still need to
2746 * update the inode item, so that all the time fields are
2747 * updated as well as the necessary btrfs inode in memory fields
2748 * for detecting, at fsync time, if the inode isn't yet in the
2749 * log tree or it's there but not up to date.
2751 struct timespec64 now = inode_set_ctime_current(inode);
2753 inode_inc_iversion(inode);
2754 inode_set_mtime_to_ts(inode, now);
2755 trans = btrfs_start_transaction(root, 1);
2756 if (IS_ERR(trans)) {
2757 ret = PTR_ERR(trans);
2761 ret = btrfs_update_inode(trans, BTRFS_I(inode));
2762 ret2 = btrfs_end_transaction(trans);
2767 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
2771 /* Helper structure to record which range is already reserved */
2772 struct falloc_range {
2773 struct list_head list;
2779 * Helper function to add falloc range
2781 * Caller should have locked the larger range of extent containing
2784 static int add_falloc_range(struct list_head *head, u64 start, u64 len)
2786 struct falloc_range *range = NULL;
2788 if (!list_empty(head)) {
2790 * As fallocate iterates by bytenr order, we only need to check
2793 range = list_last_entry(head, struct falloc_range, list);
2794 if (range->start + range->len == start) {
2800 range = kmalloc(sizeof(*range), GFP_KERNEL);
2803 range->start = start;
2805 list_add_tail(&range->list, head);
2809 static int btrfs_fallocate_update_isize(struct inode *inode,
2813 struct btrfs_trans_handle *trans;
2814 struct btrfs_root *root = BTRFS_I(inode)->root;
2818 if (mode & FALLOC_FL_KEEP_SIZE || end <= i_size_read(inode))
2821 trans = btrfs_start_transaction(root, 1);
2823 return PTR_ERR(trans);
2825 inode_set_ctime_current(inode);
2826 i_size_write(inode, end);
2827 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
2828 ret = btrfs_update_inode(trans, BTRFS_I(inode));
2829 ret2 = btrfs_end_transaction(trans);
2831 return ret ? ret : ret2;
2835 RANGE_BOUNDARY_WRITTEN_EXTENT,
2836 RANGE_BOUNDARY_PREALLOC_EXTENT,
2837 RANGE_BOUNDARY_HOLE,
2840 static int btrfs_zero_range_check_range_boundary(struct btrfs_inode *inode,
2843 const u64 sectorsize = inode->root->fs_info->sectorsize;
2844 struct extent_map *em;
2847 offset = round_down(offset, sectorsize);
2848 em = btrfs_get_extent(inode, NULL, offset, sectorsize);
2852 if (em->block_start == EXTENT_MAP_HOLE)
2853 ret = RANGE_BOUNDARY_HOLE;
2854 else if (em->flags & EXTENT_FLAG_PREALLOC)
2855 ret = RANGE_BOUNDARY_PREALLOC_EXTENT;
2857 ret = RANGE_BOUNDARY_WRITTEN_EXTENT;
2859 free_extent_map(em);
2863 static int btrfs_zero_range(struct inode *inode,
2868 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2869 struct extent_map *em;
2870 struct extent_changeset *data_reserved = NULL;
2873 const u64 sectorsize = fs_info->sectorsize;
2874 u64 alloc_start = round_down(offset, sectorsize);
2875 u64 alloc_end = round_up(offset + len, sectorsize);
2876 u64 bytes_to_reserve = 0;
2877 bool space_reserved = false;
2879 em = btrfs_get_extent(BTRFS_I(inode), NULL, alloc_start,
2880 alloc_end - alloc_start);
2887 * Avoid hole punching and extent allocation for some cases. More cases
2888 * could be considered, but these are unlikely common and we keep things
2889 * as simple as possible for now. Also, intentionally, if the target
2890 * range contains one or more prealloc extents together with regular
2891 * extents and holes, we drop all the existing extents and allocate a
2892 * new prealloc extent, so that we get a larger contiguous disk extent.
2894 if (em->start <= alloc_start && (em->flags & EXTENT_FLAG_PREALLOC)) {
2895 const u64 em_end = em->start + em->len;
2897 if (em_end >= offset + len) {
2899 * The whole range is already a prealloc extent,
2900 * do nothing except updating the inode's i_size if
2903 free_extent_map(em);
2904 ret = btrfs_fallocate_update_isize(inode, offset + len,
2909 * Part of the range is already a prealloc extent, so operate
2910 * only on the remaining part of the range.
2912 alloc_start = em_end;
2913 ASSERT(IS_ALIGNED(alloc_start, sectorsize));
2914 len = offset + len - alloc_start;
2915 offset = alloc_start;
2916 alloc_hint = em->block_start + em->len;
2918 free_extent_map(em);
2920 if (BTRFS_BYTES_TO_BLKS(fs_info, offset) ==
2921 BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)) {
2922 em = btrfs_get_extent(BTRFS_I(inode), NULL, alloc_start, sectorsize);
2928 if (em->flags & EXTENT_FLAG_PREALLOC) {
2929 free_extent_map(em);
2930 ret = btrfs_fallocate_update_isize(inode, offset + len,
2934 if (len < sectorsize && em->block_start != EXTENT_MAP_HOLE) {
2935 free_extent_map(em);
2936 ret = btrfs_truncate_block(BTRFS_I(inode), offset, len,
2939 ret = btrfs_fallocate_update_isize(inode,
2944 free_extent_map(em);
2945 alloc_start = round_down(offset, sectorsize);
2946 alloc_end = alloc_start + sectorsize;
2950 alloc_start = round_up(offset, sectorsize);
2951 alloc_end = round_down(offset + len, sectorsize);
2954 * For unaligned ranges, check the pages at the boundaries, they might
2955 * map to an extent, in which case we need to partially zero them, or
2956 * they might map to a hole, in which case we need our allocation range
2959 if (!IS_ALIGNED(offset, sectorsize)) {
2960 ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode),
2964 if (ret == RANGE_BOUNDARY_HOLE) {
2965 alloc_start = round_down(offset, sectorsize);
2967 } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
2968 ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0);
2976 if (!IS_ALIGNED(offset + len, sectorsize)) {
2977 ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode),
2981 if (ret == RANGE_BOUNDARY_HOLE) {
2982 alloc_end = round_up(offset + len, sectorsize);
2984 } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
2985 ret = btrfs_truncate_block(BTRFS_I(inode), offset + len,
2995 if (alloc_start < alloc_end) {
2996 struct extent_state *cached_state = NULL;
2997 const u64 lockstart = alloc_start;
2998 const u64 lockend = alloc_end - 1;
3000 bytes_to_reserve = alloc_end - alloc_start;
3001 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
3005 space_reserved = true;
3006 btrfs_punch_hole_lock_range(inode, lockstart, lockend,
3008 ret = btrfs_qgroup_reserve_data(BTRFS_I(inode), &data_reserved,
3009 alloc_start, bytes_to_reserve);
3011 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart,
3012 lockend, &cached_state);
3015 ret = btrfs_prealloc_file_range(inode, mode, alloc_start,
3016 alloc_end - alloc_start,
3017 fs_info->sectorsize,
3018 offset + len, &alloc_hint);
3019 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
3021 /* btrfs_prealloc_file_range releases reserved space on error */
3023 space_reserved = false;
3027 ret = btrfs_fallocate_update_isize(inode, offset + len, mode);
3029 if (ret && space_reserved)
3030 btrfs_free_reserved_data_space(BTRFS_I(inode), data_reserved,
3031 alloc_start, bytes_to_reserve);
3032 extent_changeset_free(data_reserved);
3037 static long btrfs_fallocate(struct file *file, int mode,
3038 loff_t offset, loff_t len)
3040 struct inode *inode = file_inode(file);
3041 struct extent_state *cached_state = NULL;
3042 struct extent_changeset *data_reserved = NULL;
3043 struct falloc_range *range;
3044 struct falloc_range *tmp;
3045 LIST_HEAD(reserve_list);
3053 u64 data_space_needed = 0;
3054 u64 data_space_reserved = 0;
3055 u64 qgroup_reserved = 0;
3056 struct extent_map *em;
3057 int blocksize = BTRFS_I(inode)->root->fs_info->sectorsize;
3060 /* Do not allow fallocate in ZONED mode */
3061 if (btrfs_is_zoned(inode_to_fs_info(inode)))
3064 alloc_start = round_down(offset, blocksize);
3065 alloc_end = round_up(offset + len, blocksize);
3066 cur_offset = alloc_start;
3068 /* Make sure we aren't being give some crap mode */
3069 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
3070 FALLOC_FL_ZERO_RANGE))
3073 if (mode & FALLOC_FL_PUNCH_HOLE)
3074 return btrfs_punch_hole(file, offset, len);
3076 btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
3078 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) {
3079 ret = inode_newsize_ok(inode, offset + len);
3084 ret = file_modified(file);
3089 * TODO: Move these two operations after we have checked
3090 * accurate reserved space, or fallocate can still fail but
3091 * with page truncated or size expanded.
3093 * But that's a minor problem and won't do much harm BTW.
3095 if (alloc_start > inode->i_size) {
3096 ret = btrfs_cont_expand(BTRFS_I(inode), i_size_read(inode),
3100 } else if (offset + len > inode->i_size) {
3102 * If we are fallocating from the end of the file onward we
3103 * need to zero out the end of the block if i_size lands in the
3104 * middle of a block.
3106 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
3112 * We have locked the inode at the VFS level (in exclusive mode) and we
3113 * have locked the i_mmap_lock lock (in exclusive mode). Now before
3114 * locking the file range, flush all dealloc in the range and wait for
3115 * all ordered extents in the range to complete. After this we can lock
3116 * the file range and, due to the previous locking we did, we know there
3117 * can't be more delalloc or ordered extents in the range.
3119 ret = btrfs_wait_ordered_range(inode, alloc_start,
3120 alloc_end - alloc_start);
3124 if (mode & FALLOC_FL_ZERO_RANGE) {
3125 ret = btrfs_zero_range(inode, offset, len, mode);
3126 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
3130 locked_end = alloc_end - 1;
3131 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
3134 btrfs_assert_inode_range_clean(BTRFS_I(inode), alloc_start, locked_end);
3136 /* First, check if we exceed the qgroup limit */
3137 while (cur_offset < alloc_end) {
3138 em = btrfs_get_extent(BTRFS_I(inode), NULL, cur_offset,
3139 alloc_end - cur_offset);
3144 last_byte = min(extent_map_end(em), alloc_end);
3145 actual_end = min_t(u64, extent_map_end(em), offset + len);
3146 last_byte = ALIGN(last_byte, blocksize);
3147 if (em->block_start == EXTENT_MAP_HOLE ||
3148 (cur_offset >= inode->i_size &&
3149 !(em->flags & EXTENT_FLAG_PREALLOC))) {
3150 const u64 range_len = last_byte - cur_offset;
3152 ret = add_falloc_range(&reserve_list, cur_offset, range_len);
3154 free_extent_map(em);
3157 ret = btrfs_qgroup_reserve_data(BTRFS_I(inode),
3158 &data_reserved, cur_offset, range_len);
3160 free_extent_map(em);
3163 qgroup_reserved += range_len;
3164 data_space_needed += range_len;
3166 free_extent_map(em);
3167 cur_offset = last_byte;
3170 if (!ret && data_space_needed > 0) {
3172 * We are safe to reserve space here as we can't have delalloc
3173 * in the range, see above.
3175 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
3178 data_space_reserved = data_space_needed;
3182 * If ret is still 0, means we're OK to fallocate.
3183 * Or just cleanup the list and exit.
3185 list_for_each_entry_safe(range, tmp, &reserve_list, list) {
3187 ret = btrfs_prealloc_file_range(inode, mode,
3189 range->len, blocksize,
3190 offset + len, &alloc_hint);
3192 * btrfs_prealloc_file_range() releases space even
3193 * if it returns an error.
3195 data_space_reserved -= range->len;
3196 qgroup_reserved -= range->len;
3197 } else if (data_space_reserved > 0) {
3198 btrfs_free_reserved_data_space(BTRFS_I(inode),
3199 data_reserved, range->start,
3201 data_space_reserved -= range->len;
3202 qgroup_reserved -= range->len;
3203 } else if (qgroup_reserved > 0) {
3204 btrfs_qgroup_free_data(BTRFS_I(inode), data_reserved,
3205 range->start, range->len, NULL);
3206 qgroup_reserved -= range->len;
3208 list_del(&range->list);
3215 * We didn't need to allocate any more space, but we still extended the
3216 * size of the file so we need to update i_size and the inode item.
3218 ret = btrfs_fallocate_update_isize(inode, actual_end, mode);
3220 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
3223 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
3224 extent_changeset_free(data_reserved);
3229 * Helper for btrfs_find_delalloc_in_range(). Find a subrange in a given range
3230 * that has unflushed and/or flushing delalloc. There might be other adjacent
3231 * subranges after the one it found, so btrfs_find_delalloc_in_range() keeps
3232 * looping while it gets adjacent subranges, and merging them together.
3234 static bool find_delalloc_subrange(struct btrfs_inode *inode, u64 start, u64 end,
3235 struct extent_state **cached_state,
3236 bool *search_io_tree,
3237 u64 *delalloc_start_ret, u64 *delalloc_end_ret)
3239 u64 len = end + 1 - start;
3240 u64 delalloc_len = 0;
3241 struct btrfs_ordered_extent *oe;
3246 * Search the io tree first for EXTENT_DELALLOC. If we find any, it
3247 * means we have delalloc (dirty pages) for which writeback has not
3250 if (*search_io_tree) {
3251 spin_lock(&inode->lock);
3252 if (inode->delalloc_bytes > 0) {
3253 spin_unlock(&inode->lock);
3254 *delalloc_start_ret = start;
3255 delalloc_len = count_range_bits(&inode->io_tree,
3256 delalloc_start_ret, end,
3257 len, EXTENT_DELALLOC, 1,
3260 spin_unlock(&inode->lock);
3264 if (delalloc_len > 0) {
3266 * If delalloc was found then *delalloc_start_ret has a sector size
3267 * aligned value (rounded down).
3269 *delalloc_end_ret = *delalloc_start_ret + delalloc_len - 1;
3271 if (*delalloc_start_ret == start) {
3272 /* Delalloc for the whole range, nothing more to do. */
3273 if (*delalloc_end_ret == end)
3275 /* Else trim our search range for ordered extents. */
3276 start = *delalloc_end_ret + 1;
3277 len = end + 1 - start;
3280 /* No delalloc, future calls don't need to search again. */
3281 *search_io_tree = false;
3285 * Now also check if there's any ordered extent in the range.
3286 * We do this because:
3288 * 1) When delalloc is flushed, the file range is locked, we clear the
3289 * EXTENT_DELALLOC bit from the io tree and create an extent map and
3290 * an ordered extent for the write. So we might just have been called
3291 * after delalloc is flushed and before the ordered extent completes
3292 * and inserts the new file extent item in the subvolume's btree;
3294 * 2) We may have an ordered extent created by flushing delalloc for a
3295 * subrange that starts before the subrange we found marked with
3296 * EXTENT_DELALLOC in the io tree.
3298 * We could also use the extent map tree to find such delalloc that is
3299 * being flushed, but using the ordered extents tree is more efficient
3300 * because it's usually much smaller as ordered extents are removed from
3301 * the tree once they complete. With the extent maps, we mau have them
3302 * in the extent map tree for a very long time, and they were either
3303 * created by previous writes or loaded by read operations.
3305 oe = btrfs_lookup_first_ordered_range(inode, start, len);
3307 return (delalloc_len > 0);
3309 /* The ordered extent may span beyond our search range. */
3310 oe_start = max(oe->file_offset, start);
3311 oe_end = min(oe->file_offset + oe->num_bytes - 1, end);
3313 btrfs_put_ordered_extent(oe);
3315 /* Don't have unflushed delalloc, return the ordered extent range. */
3316 if (delalloc_len == 0) {
3317 *delalloc_start_ret = oe_start;
3318 *delalloc_end_ret = oe_end;
3323 * We have both unflushed delalloc (io_tree) and an ordered extent.
3324 * If the ranges are adjacent returned a combined range, otherwise
3325 * return the leftmost range.
3327 if (oe_start < *delalloc_start_ret) {
3328 if (oe_end < *delalloc_start_ret)
3329 *delalloc_end_ret = oe_end;
3330 *delalloc_start_ret = oe_start;
3331 } else if (*delalloc_end_ret + 1 == oe_start) {
3332 *delalloc_end_ret = oe_end;
3339 * Check if there's delalloc in a given range.
3341 * @inode: The inode.
3342 * @start: The start offset of the range. It does not need to be
3343 * sector size aligned.
3344 * @end: The end offset (inclusive value) of the search range.
3345 * It does not need to be sector size aligned.
3346 * @cached_state: Extent state record used for speeding up delalloc
3347 * searches in the inode's io_tree. Can be NULL.
3348 * @delalloc_start_ret: Output argument, set to the start offset of the
3349 * subrange found with delalloc (may not be sector size
3351 * @delalloc_end_ret: Output argument, set to he end offset (inclusive value)
3352 * of the subrange found with delalloc.
3354 * Returns true if a subrange with delalloc is found within the given range, and
3355 * if so it sets @delalloc_start_ret and @delalloc_end_ret with the start and
3356 * end offsets of the subrange.
3358 bool btrfs_find_delalloc_in_range(struct btrfs_inode *inode, u64 start, u64 end,
3359 struct extent_state **cached_state,
3360 u64 *delalloc_start_ret, u64 *delalloc_end_ret)
3362 u64 cur_offset = round_down(start, inode->root->fs_info->sectorsize);
3363 u64 prev_delalloc_end = 0;
3364 bool search_io_tree = true;
3367 while (cur_offset <= end) {
3372 delalloc = find_delalloc_subrange(inode, cur_offset, end,
3373 cached_state, &search_io_tree,
3379 if (prev_delalloc_end == 0) {
3380 /* First subrange found. */
3381 *delalloc_start_ret = max(delalloc_start, start);
3382 *delalloc_end_ret = delalloc_end;
3384 } else if (delalloc_start == prev_delalloc_end + 1) {
3385 /* Subrange adjacent to the previous one, merge them. */
3386 *delalloc_end_ret = delalloc_end;
3388 /* Subrange not adjacent to the previous one, exit. */
3392 prev_delalloc_end = delalloc_end;
3393 cur_offset = delalloc_end + 1;
3401 * Check if there's a hole or delalloc range in a range representing a hole (or
3402 * prealloc extent) found in the inode's subvolume btree.
3404 * @inode: The inode.
3405 * @whence: Seek mode (SEEK_DATA or SEEK_HOLE).
3406 * @start: Start offset of the hole region. It does not need to be sector
3408 * @end: End offset (inclusive value) of the hole region. It does not
3409 * need to be sector size aligned.
3410 * @start_ret: Return parameter, used to set the start of the subrange in the
3411 * hole that matches the search criteria (seek mode), if such
3412 * subrange is found (return value of the function is true).
3413 * The value returned here may not be sector size aligned.
3415 * Returns true if a subrange matching the given seek mode is found, and if one
3416 * is found, it updates @start_ret with the start of the subrange.
3418 static bool find_desired_extent_in_hole(struct btrfs_inode *inode, int whence,
3419 struct extent_state **cached_state,
3420 u64 start, u64 end, u64 *start_ret)
3426 delalloc = btrfs_find_delalloc_in_range(inode, start, end, cached_state,
3427 &delalloc_start, &delalloc_end);
3428 if (delalloc && whence == SEEK_DATA) {
3429 *start_ret = delalloc_start;
3433 if (delalloc && whence == SEEK_HOLE) {
3435 * We found delalloc but it starts after out start offset. So we
3436 * have a hole between our start offset and the delalloc start.
3438 if (start < delalloc_start) {
3443 * Delalloc range starts at our start offset.
3444 * If the delalloc range's length is smaller than our range,
3445 * then it means we have a hole that starts where the delalloc
3448 if (delalloc_end < end) {
3449 *start_ret = delalloc_end + 1;
3453 /* There's delalloc for the whole range. */
3457 if (!delalloc && whence == SEEK_HOLE) {
3463 * No delalloc in the range and we are seeking for data. The caller has
3464 * to iterate to the next extent item in the subvolume btree.
3469 static loff_t find_desired_extent(struct file *file, loff_t offset, int whence)
3471 struct btrfs_inode *inode = BTRFS_I(file->f_mapping->host);
3472 struct btrfs_file_private *private = file->private_data;
3473 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3474 struct extent_state *cached_state = NULL;
3475 struct extent_state **delalloc_cached_state;
3476 const loff_t i_size = i_size_read(&inode->vfs_inode);
3477 const u64 ino = btrfs_ino(inode);
3478 struct btrfs_root *root = inode->root;
3479 struct btrfs_path *path;
3480 struct btrfs_key key;
3481 u64 last_extent_end;
3488 if (i_size == 0 || offset >= i_size)
3492 * Quick path. If the inode has no prealloc extents and its number of
3493 * bytes used matches its i_size, then it can not have holes.
3495 if (whence == SEEK_HOLE &&
3496 !(inode->flags & BTRFS_INODE_PREALLOC) &&
3497 inode_get_bytes(&inode->vfs_inode) == i_size)
3501 private = kzalloc(sizeof(*private), GFP_KERNEL);
3503 * No worries if memory allocation failed.
3504 * The private structure is used only for speeding up multiple
3505 * lseek SEEK_HOLE/DATA calls to a file when there's delalloc,
3506 * so everything will still be correct.
3508 file->private_data = private;
3512 delalloc_cached_state = &private->llseek_cached_state;
3514 delalloc_cached_state = NULL;
3517 * offset can be negative, in this case we start finding DATA/HOLE from
3518 * the very start of the file.
3520 start = max_t(loff_t, 0, offset);
3522 lockstart = round_down(start, fs_info->sectorsize);
3523 lockend = round_up(i_size, fs_info->sectorsize);
3524 if (lockend <= lockstart)
3525 lockend = lockstart + fs_info->sectorsize;
3528 path = btrfs_alloc_path();
3531 path->reada = READA_FORWARD;
3534 key.type = BTRFS_EXTENT_DATA_KEY;
3537 last_extent_end = lockstart;
3539 lock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
3541 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3544 } else if (ret > 0 && path->slots[0] > 0) {
3545 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
3546 if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY)
3550 while (start < i_size) {
3551 struct extent_buffer *leaf = path->nodes[0];
3552 struct btrfs_file_extent_item *extent;
3556 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3557 ret = btrfs_next_leaf(root, path);
3563 leaf = path->nodes[0];
3566 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3567 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
3570 extent_end = btrfs_file_extent_end(path);
3573 * In the first iteration we may have a slot that points to an
3574 * extent that ends before our start offset, so skip it.
3576 if (extent_end <= start) {
3581 /* We have an implicit hole, NO_HOLES feature is likely set. */
3582 if (last_extent_end < key.offset) {
3583 u64 search_start = last_extent_end;
3587 * First iteration, @start matches @offset and it's
3590 if (start == offset)
3591 search_start = offset;
3593 found = find_desired_extent_in_hole(inode, whence,
3594 delalloc_cached_state,
3599 start = found_start;
3603 * Didn't find data or a hole (due to delalloc) in the
3604 * implicit hole range, so need to analyze the extent.
3608 extent = btrfs_item_ptr(leaf, path->slots[0],
3609 struct btrfs_file_extent_item);
3610 type = btrfs_file_extent_type(leaf, extent);
3613 * Can't access the extent's disk_bytenr field if this is an
3614 * inline extent, since at that offset, it's where the extent
3617 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
3618 (type == BTRFS_FILE_EXTENT_REG &&
3619 btrfs_file_extent_disk_bytenr(leaf, extent) == 0)) {
3621 * Explicit hole or prealloc extent, search for delalloc.
3622 * A prealloc extent is treated like a hole.
3624 u64 search_start = key.offset;
3628 * First iteration, @start matches @offset and it's
3631 if (start == offset)
3632 search_start = offset;
3634 found = find_desired_extent_in_hole(inode, whence,
3635 delalloc_cached_state,
3640 start = found_start;
3644 * Didn't find data or a hole (due to delalloc) in the
3645 * implicit hole range, so need to analyze the next
3650 * Found a regular or inline extent.
3651 * If we are seeking for data, adjust the start offset
3652 * and stop, we're done.
3654 if (whence == SEEK_DATA) {
3655 start = max_t(u64, key.offset, offset);
3660 * Else, we are seeking for a hole, check the next file
3666 last_extent_end = extent_end;
3668 if (fatal_signal_pending(current)) {
3675 /* We have an implicit hole from the last extent found up to i_size. */
3676 if (!found && start < i_size) {
3677 found = find_desired_extent_in_hole(inode, whence,
3678 delalloc_cached_state, start,
3679 i_size - 1, &start);
3685 unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
3686 btrfs_free_path(path);
3691 if (whence == SEEK_DATA && start >= i_size)
3694 return min_t(loff_t, start, i_size);
3697 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
3699 struct inode *inode = file->f_mapping->host;
3703 return generic_file_llseek(file, offset, whence);
3706 btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_SHARED);
3707 offset = find_desired_extent(file, offset, whence);
3708 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_SHARED);
3715 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
3718 static int btrfs_file_open(struct inode *inode, struct file *filp)
3722 filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC | FMODE_BUF_WASYNC |
3725 ret = fsverity_file_open(inode, filp);
3728 return generic_file_open(inode, filp);
3731 static int check_direct_read(struct btrfs_fs_info *fs_info,
3732 const struct iov_iter *iter, loff_t offset)
3737 ret = check_direct_IO(fs_info, iter, offset);
3741 if (!iter_is_iovec(iter))
3744 for (seg = 0; seg < iter->nr_segs; seg++) {
3745 for (i = seg + 1; i < iter->nr_segs; i++) {
3746 const struct iovec *iov1 = iter_iov(iter) + seg;
3747 const struct iovec *iov2 = iter_iov(iter) + i;
3749 if (iov1->iov_base == iov2->iov_base)
3756 static ssize_t btrfs_direct_read(struct kiocb *iocb, struct iov_iter *to)
3758 struct inode *inode = file_inode(iocb->ki_filp);
3759 size_t prev_left = 0;
3763 if (fsverity_active(inode))
3766 if (check_direct_read(inode_to_fs_info(inode), to, iocb->ki_pos))
3769 btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_SHARED);
3772 * This is similar to what we do for direct IO writes, see the comment
3773 * at btrfs_direct_write(), but we also disable page faults in addition
3774 * to disabling them only at the iov_iter level. This is because when
3775 * reading from a hole or prealloc extent, iomap calls iov_iter_zero(),
3776 * which can still trigger page fault ins despite having set ->nofault
3777 * to true of our 'to' iov_iter.
3779 * The difference to direct IO writes is that we deadlock when trying
3780 * to lock the extent range in the inode's tree during he page reads
3781 * triggered by the fault in (while for writes it is due to waiting for
3782 * our own ordered extent). This is because for direct IO reads,
3783 * btrfs_dio_iomap_begin() returns with the extent range locked, which
3784 * is only unlocked in the endio callback (end_bio_extent_readpage()).
3786 pagefault_disable();
3788 ret = btrfs_dio_read(iocb, to, read);
3789 to->nofault = false;
3792 /* No increment (+=) because iomap returns a cumulative value. */
3796 if (iov_iter_count(to) > 0 && (ret == -EFAULT || ret > 0)) {
3797 const size_t left = iov_iter_count(to);
3799 if (left == prev_left) {
3801 * We didn't make any progress since the last attempt,
3802 * fallback to a buffered read for the remainder of the
3803 * range. This is just to avoid any possibility of looping
3809 * We made some progress since the last retry or this is
3810 * the first time we are retrying. Fault in as many pages
3811 * as possible and retry.
3813 fault_in_iov_iter_writeable(to, left);
3818 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_SHARED);
3819 return ret < 0 ? ret : read;
3822 static ssize_t btrfs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
3826 if (iocb->ki_flags & IOCB_DIRECT) {
3827 ret = btrfs_direct_read(iocb, to);
3828 if (ret < 0 || !iov_iter_count(to) ||
3829 iocb->ki_pos >= i_size_read(file_inode(iocb->ki_filp)))
3833 return filemap_read(iocb, to, ret);
3836 const struct file_operations btrfs_file_operations = {
3837 .llseek = btrfs_file_llseek,
3838 .read_iter = btrfs_file_read_iter,
3839 .splice_read = filemap_splice_read,
3840 .write_iter = btrfs_file_write_iter,
3841 .splice_write = iter_file_splice_write,
3842 .mmap = btrfs_file_mmap,
3843 .open = btrfs_file_open,
3844 .release = btrfs_release_file,
3845 .get_unmapped_area = thp_get_unmapped_area,
3846 .fsync = btrfs_sync_file,
3847 .fallocate = btrfs_fallocate,
3848 .unlocked_ioctl = btrfs_ioctl,
3849 #ifdef CONFIG_COMPAT
3850 .compat_ioctl = btrfs_compat_ioctl,
3852 .remap_file_range = btrfs_remap_file_range,
3855 int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end)
3860 * So with compression we will find and lock a dirty page and clear the
3861 * first one as dirty, setup an async extent, and immediately return
3862 * with the entire range locked but with nobody actually marked with
3863 * writeback. So we can't just filemap_write_and_wait_range() and
3864 * expect it to work since it will just kick off a thread to do the
3865 * actual work. So we need to call filemap_fdatawrite_range _again_
3866 * since it will wait on the page lock, which won't be unlocked until
3867 * after the pages have been marked as writeback and so we're good to go
3868 * from there. We have to do this otherwise we'll miss the ordered
3869 * extents and that results in badness. Please Josef, do not think you
3870 * know better and pull this out at some point in the future, it is
3871 * right and you are wrong.
3873 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
3874 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
3875 &BTRFS_I(inode)->runtime_flags))
3876 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);