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>
22 #include "transaction.h"
23 #include "btrfs_inode.h"
24 #include "print-tree.h"
29 #include "compression.h"
30 #include "delalloc-space.h"
34 /* simple helper to fault in pages and copy. This should go away
35 * and be replaced with calls into generic code.
37 static noinline int btrfs_copy_from_user(loff_t pos, size_t write_bytes,
38 struct page **prepared_pages,
42 size_t total_copied = 0;
44 int offset = offset_in_page(pos);
46 while (write_bytes > 0) {
47 size_t count = min_t(size_t,
48 PAGE_SIZE - offset, write_bytes);
49 struct page *page = prepared_pages[pg];
51 * Copy data from userspace to the current page
53 copied = copy_page_from_iter_atomic(page, offset, count, i);
55 /* Flush processor's dcache for this page */
56 flush_dcache_page(page);
59 * if we get a partial write, we can end up with
60 * partially up to date pages. These add
61 * a lot of complexity, so make sure they don't
62 * happen by forcing this copy to be retried.
64 * The rest of the btrfs_file_write code will fall
65 * back to page at a time copies after we return 0.
67 if (unlikely(copied < count)) {
68 if (!PageUptodate(page)) {
69 iov_iter_revert(i, copied);
76 write_bytes -= copied;
77 total_copied += copied;
79 if (offset == PAGE_SIZE) {
88 * unlocks pages after btrfs_file_write is done with them
90 static void btrfs_drop_pages(struct btrfs_fs_info *fs_info,
91 struct page **pages, size_t num_pages,
95 u64 block_start = round_down(pos, fs_info->sectorsize);
96 u64 block_len = round_up(pos + copied, fs_info->sectorsize) - block_start;
98 ASSERT(block_len <= U32_MAX);
99 for (i = 0; i < num_pages; i++) {
100 /* page checked is some magic around finding pages that
101 * have been modified without going through btrfs_set_page_dirty
102 * clear it here. There should be no need to mark the pages
103 * accessed as prepare_pages should have marked them accessed
104 * in prepare_pages via find_or_create_page()
106 btrfs_page_clamp_clear_checked(fs_info, pages[i], block_start,
108 unlock_page(pages[i]);
114 * After btrfs_copy_from_user(), update the following things for delalloc:
115 * - Mark newly dirtied pages as DELALLOC in the io tree.
116 * Used to advise which range is to be written back.
117 * - Mark modified pages as Uptodate/Dirty and not needing COW fixup
118 * - Update inode size for past EOF write
120 int btrfs_dirty_pages(struct btrfs_inode *inode, struct page **pages,
121 size_t num_pages, loff_t pos, size_t write_bytes,
122 struct extent_state **cached, bool noreserve)
124 struct btrfs_fs_info *fs_info = inode->root->fs_info;
129 u64 end_of_last_block;
130 u64 end_pos = pos + write_bytes;
131 loff_t isize = i_size_read(&inode->vfs_inode);
132 unsigned int extra_bits = 0;
134 if (write_bytes == 0)
138 extra_bits |= EXTENT_NORESERVE;
140 start_pos = round_down(pos, fs_info->sectorsize);
141 num_bytes = round_up(write_bytes + pos - start_pos,
142 fs_info->sectorsize);
143 ASSERT(num_bytes <= U32_MAX);
145 end_of_last_block = start_pos + num_bytes - 1;
148 * The pages may have already been dirty, clear out old accounting so
149 * we can set things up properly
151 clear_extent_bit(&inode->io_tree, start_pos, end_of_last_block,
152 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
155 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
160 for (i = 0; i < num_pages; i++) {
161 struct page *p = pages[i];
163 btrfs_page_clamp_set_uptodate(fs_info, p, start_pos, num_bytes);
164 btrfs_page_clamp_clear_checked(fs_info, p, start_pos, num_bytes);
165 btrfs_page_clamp_set_dirty(fs_info, p, start_pos, num_bytes);
169 * we've only changed i_size in ram, and we haven't updated
170 * the disk i_size. There is no need to log the inode
174 i_size_write(&inode->vfs_inode, end_pos);
179 * this is very complex, but the basic idea is to drop all extents
180 * in the range start - end. hint_block is filled in with a block number
181 * that would be a good hint to the block allocator for this file.
183 * If an extent intersects the range but is not entirely inside the range
184 * it is either truncated or split. Anything entirely inside the range
185 * is deleted from the tree.
187 * Note: the VFS' inode number of bytes is not updated, it's up to the caller
188 * to deal with that. We set the field 'bytes_found' of the arguments structure
189 * with the number of allocated bytes found in the target range, so that the
190 * caller can update the inode's number of bytes in an atomic way when
191 * replacing extents in a range to avoid races with stat(2).
193 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
194 struct btrfs_root *root, struct btrfs_inode *inode,
195 struct btrfs_drop_extents_args *args)
197 struct btrfs_fs_info *fs_info = root->fs_info;
198 struct extent_buffer *leaf;
199 struct btrfs_file_extent_item *fi;
200 struct btrfs_ref ref = { 0 };
201 struct btrfs_key key;
202 struct btrfs_key new_key;
203 u64 ino = btrfs_ino(inode);
204 u64 search_start = args->start;
207 u64 extent_offset = 0;
209 u64 last_end = args->start;
215 int modify_tree = -1;
218 struct btrfs_path *path = args->path;
220 args->bytes_found = 0;
221 args->extent_inserted = false;
223 /* Must always have a path if ->replace_extent is true */
224 ASSERT(!(args->replace_extent && !args->path));
227 path = btrfs_alloc_path();
234 if (args->drop_cache)
235 btrfs_drop_extent_map_range(inode, args->start, args->end - 1, false);
237 if (args->start >= inode->disk_i_size && !args->replace_extent)
240 update_refs = (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID);
243 ret = btrfs_lookup_file_extent(trans, root, path, ino,
244 search_start, modify_tree);
247 if (ret > 0 && path->slots[0] > 0 && search_start == args->start) {
248 leaf = path->nodes[0];
249 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
250 if (key.objectid == ino &&
251 key.type == BTRFS_EXTENT_DATA_KEY)
256 leaf = path->nodes[0];
257 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
259 ret = btrfs_next_leaf(root, path);
266 leaf = path->nodes[0];
270 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
272 if (key.objectid > ino)
274 if (WARN_ON_ONCE(key.objectid < ino) ||
275 key.type < BTRFS_EXTENT_DATA_KEY) {
280 if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= args->end)
283 fi = btrfs_item_ptr(leaf, path->slots[0],
284 struct btrfs_file_extent_item);
285 extent_type = btrfs_file_extent_type(leaf, fi);
287 if (extent_type == BTRFS_FILE_EXTENT_REG ||
288 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
289 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
290 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
291 extent_offset = btrfs_file_extent_offset(leaf, fi);
292 extent_end = key.offset +
293 btrfs_file_extent_num_bytes(leaf, fi);
294 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
295 extent_end = key.offset +
296 btrfs_file_extent_ram_bytes(leaf, fi);
303 * Don't skip extent items representing 0 byte lengths. They
304 * used to be created (bug) if while punching holes we hit
305 * -ENOSPC condition. So if we find one here, just ensure we
306 * delete it, otherwise we would insert a new file extent item
307 * with the same key (offset) as that 0 bytes length file
308 * extent item in the call to setup_items_for_insert() later
311 if (extent_end == key.offset && extent_end >= search_start) {
312 last_end = extent_end;
313 goto delete_extent_item;
316 if (extent_end <= search_start) {
322 search_start = max(key.offset, args->start);
323 if (recow || !modify_tree) {
325 btrfs_release_path(path);
330 * | - range to drop - |
331 * | -------- extent -------- |
333 if (args->start > key.offset && args->end < extent_end) {
335 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
340 memcpy(&new_key, &key, sizeof(new_key));
341 new_key.offset = args->start;
342 ret = btrfs_duplicate_item(trans, root, path,
344 if (ret == -EAGAIN) {
345 btrfs_release_path(path);
351 leaf = path->nodes[0];
352 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
353 struct btrfs_file_extent_item);
354 btrfs_set_file_extent_num_bytes(leaf, fi,
355 args->start - key.offset);
357 fi = btrfs_item_ptr(leaf, path->slots[0],
358 struct btrfs_file_extent_item);
360 extent_offset += args->start - key.offset;
361 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
362 btrfs_set_file_extent_num_bytes(leaf, fi,
363 extent_end - args->start);
364 btrfs_mark_buffer_dirty(leaf);
366 if (update_refs && disk_bytenr > 0) {
367 btrfs_init_generic_ref(&ref,
368 BTRFS_ADD_DELAYED_REF,
369 disk_bytenr, num_bytes, 0);
370 btrfs_init_data_ref(&ref,
371 root->root_key.objectid,
373 args->start - extent_offset,
375 ret = btrfs_inc_extent_ref(trans, &ref);
377 btrfs_abort_transaction(trans, ret);
381 key.offset = args->start;
384 * From here on out we will have actually dropped something, so
385 * last_end can be updated.
387 last_end = extent_end;
390 * | ---- range to drop ----- |
391 * | -------- extent -------- |
393 if (args->start <= key.offset && args->end < extent_end) {
394 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
399 memcpy(&new_key, &key, sizeof(new_key));
400 new_key.offset = args->end;
401 btrfs_set_item_key_safe(fs_info, path, &new_key);
403 extent_offset += args->end - key.offset;
404 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
405 btrfs_set_file_extent_num_bytes(leaf, fi,
406 extent_end - args->end);
407 btrfs_mark_buffer_dirty(leaf);
408 if (update_refs && disk_bytenr > 0)
409 args->bytes_found += args->end - key.offset;
413 search_start = extent_end;
415 * | ---- range to drop ----- |
416 * | -------- extent -------- |
418 if (args->start > key.offset && args->end >= extent_end) {
420 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
425 btrfs_set_file_extent_num_bytes(leaf, fi,
426 args->start - key.offset);
427 btrfs_mark_buffer_dirty(leaf);
428 if (update_refs && disk_bytenr > 0)
429 args->bytes_found += extent_end - args->start;
430 if (args->end == extent_end)
438 * | ---- range to drop ----- |
439 * | ------ extent ------ |
441 if (args->start <= key.offset && args->end >= extent_end) {
444 del_slot = path->slots[0];
447 BUG_ON(del_slot + del_nr != path->slots[0]);
452 extent_type == BTRFS_FILE_EXTENT_INLINE) {
453 args->bytes_found += extent_end - key.offset;
454 extent_end = ALIGN(extent_end,
455 fs_info->sectorsize);
456 } else if (update_refs && disk_bytenr > 0) {
457 btrfs_init_generic_ref(&ref,
458 BTRFS_DROP_DELAYED_REF,
459 disk_bytenr, num_bytes, 0);
460 btrfs_init_data_ref(&ref,
461 root->root_key.objectid,
463 key.offset - extent_offset, 0,
465 ret = btrfs_free_extent(trans, &ref);
467 btrfs_abort_transaction(trans, ret);
470 args->bytes_found += extent_end - key.offset;
473 if (args->end == extent_end)
476 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
481 ret = btrfs_del_items(trans, root, path, del_slot,
484 btrfs_abort_transaction(trans, ret);
491 btrfs_release_path(path);
498 if (!ret && del_nr > 0) {
500 * Set path->slots[0] to first slot, so that after the delete
501 * if items are move off from our leaf to its immediate left or
502 * right neighbor leafs, we end up with a correct and adjusted
503 * path->slots[0] for our insertion (if args->replace_extent).
505 path->slots[0] = del_slot;
506 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
508 btrfs_abort_transaction(trans, ret);
511 leaf = path->nodes[0];
513 * If btrfs_del_items() was called, it might have deleted a leaf, in
514 * which case it unlocked our path, so check path->locks[0] matches a
517 if (!ret && args->replace_extent &&
518 path->locks[0] == BTRFS_WRITE_LOCK &&
519 btrfs_leaf_free_space(leaf) >=
520 sizeof(struct btrfs_item) + args->extent_item_size) {
523 key.type = BTRFS_EXTENT_DATA_KEY;
524 key.offset = args->start;
525 if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
526 struct btrfs_key slot_key;
528 btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
529 if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
532 btrfs_setup_item_for_insert(root, path, &key, args->extent_item_size);
533 args->extent_inserted = true;
537 btrfs_free_path(path);
538 else if (!args->extent_inserted)
539 btrfs_release_path(path);
541 args->drop_end = found ? min(args->end, last_end) : args->end;
546 static int extent_mergeable(struct extent_buffer *leaf, int slot,
547 u64 objectid, u64 bytenr, u64 orig_offset,
548 u64 *start, u64 *end)
550 struct btrfs_file_extent_item *fi;
551 struct btrfs_key key;
554 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
557 btrfs_item_key_to_cpu(leaf, &key, slot);
558 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
561 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
562 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
563 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
564 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
565 btrfs_file_extent_compression(leaf, fi) ||
566 btrfs_file_extent_encryption(leaf, fi) ||
567 btrfs_file_extent_other_encoding(leaf, fi))
570 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
571 if ((*start && *start != key.offset) || (*end && *end != extent_end))
580 * Mark extent in the range start - end as written.
582 * This changes extent type from 'pre-allocated' to 'regular'. If only
583 * part of extent is marked as written, the extent will be split into
586 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
587 struct btrfs_inode *inode, u64 start, u64 end)
589 struct btrfs_fs_info *fs_info = trans->fs_info;
590 struct btrfs_root *root = inode->root;
591 struct extent_buffer *leaf;
592 struct btrfs_path *path;
593 struct btrfs_file_extent_item *fi;
594 struct btrfs_ref ref = { 0 };
595 struct btrfs_key key;
596 struct btrfs_key new_key;
608 u64 ino = btrfs_ino(inode);
610 path = btrfs_alloc_path();
617 key.type = BTRFS_EXTENT_DATA_KEY;
620 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
623 if (ret > 0 && path->slots[0] > 0)
626 leaf = path->nodes[0];
627 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
628 if (key.objectid != ino ||
629 key.type != BTRFS_EXTENT_DATA_KEY) {
631 btrfs_abort_transaction(trans, ret);
634 fi = btrfs_item_ptr(leaf, path->slots[0],
635 struct btrfs_file_extent_item);
636 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_PREALLOC) {
638 btrfs_abort_transaction(trans, ret);
641 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
642 if (key.offset > start || extent_end < end) {
644 btrfs_abort_transaction(trans, ret);
648 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
649 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
650 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
651 memcpy(&new_key, &key, sizeof(new_key));
653 if (start == key.offset && end < extent_end) {
656 if (extent_mergeable(leaf, path->slots[0] - 1,
657 ino, bytenr, orig_offset,
658 &other_start, &other_end)) {
659 new_key.offset = end;
660 btrfs_set_item_key_safe(fs_info, path, &new_key);
661 fi = btrfs_item_ptr(leaf, path->slots[0],
662 struct btrfs_file_extent_item);
663 btrfs_set_file_extent_generation(leaf, fi,
665 btrfs_set_file_extent_num_bytes(leaf, fi,
667 btrfs_set_file_extent_offset(leaf, fi,
669 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
670 struct btrfs_file_extent_item);
671 btrfs_set_file_extent_generation(leaf, fi,
673 btrfs_set_file_extent_num_bytes(leaf, fi,
675 btrfs_mark_buffer_dirty(leaf);
680 if (start > key.offset && end == extent_end) {
683 if (extent_mergeable(leaf, path->slots[0] + 1,
684 ino, bytenr, orig_offset,
685 &other_start, &other_end)) {
686 fi = btrfs_item_ptr(leaf, path->slots[0],
687 struct btrfs_file_extent_item);
688 btrfs_set_file_extent_num_bytes(leaf, fi,
690 btrfs_set_file_extent_generation(leaf, fi,
693 new_key.offset = start;
694 btrfs_set_item_key_safe(fs_info, path, &new_key);
696 fi = btrfs_item_ptr(leaf, path->slots[0],
697 struct btrfs_file_extent_item);
698 btrfs_set_file_extent_generation(leaf, fi,
700 btrfs_set_file_extent_num_bytes(leaf, fi,
702 btrfs_set_file_extent_offset(leaf, fi,
703 start - orig_offset);
704 btrfs_mark_buffer_dirty(leaf);
709 while (start > key.offset || end < extent_end) {
710 if (key.offset == start)
713 new_key.offset = split;
714 ret = btrfs_duplicate_item(trans, root, path, &new_key);
715 if (ret == -EAGAIN) {
716 btrfs_release_path(path);
720 btrfs_abort_transaction(trans, ret);
724 leaf = path->nodes[0];
725 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
726 struct btrfs_file_extent_item);
727 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
728 btrfs_set_file_extent_num_bytes(leaf, fi,
731 fi = btrfs_item_ptr(leaf, path->slots[0],
732 struct btrfs_file_extent_item);
734 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
735 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
736 btrfs_set_file_extent_num_bytes(leaf, fi,
738 btrfs_mark_buffer_dirty(leaf);
740 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, bytenr,
742 btrfs_init_data_ref(&ref, root->root_key.objectid, ino,
743 orig_offset, 0, false);
744 ret = btrfs_inc_extent_ref(trans, &ref);
746 btrfs_abort_transaction(trans, ret);
750 if (split == start) {
753 if (start != key.offset) {
755 btrfs_abort_transaction(trans, ret);
766 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
768 btrfs_init_data_ref(&ref, root->root_key.objectid, ino, orig_offset,
770 if (extent_mergeable(leaf, path->slots[0] + 1,
771 ino, bytenr, orig_offset,
772 &other_start, &other_end)) {
774 btrfs_release_path(path);
777 extent_end = other_end;
778 del_slot = path->slots[0] + 1;
780 ret = btrfs_free_extent(trans, &ref);
782 btrfs_abort_transaction(trans, ret);
788 if (extent_mergeable(leaf, path->slots[0] - 1,
789 ino, bytenr, orig_offset,
790 &other_start, &other_end)) {
792 btrfs_release_path(path);
795 key.offset = other_start;
796 del_slot = path->slots[0];
798 ret = btrfs_free_extent(trans, &ref);
800 btrfs_abort_transaction(trans, ret);
805 fi = btrfs_item_ptr(leaf, path->slots[0],
806 struct btrfs_file_extent_item);
807 btrfs_set_file_extent_type(leaf, fi,
808 BTRFS_FILE_EXTENT_REG);
809 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
810 btrfs_mark_buffer_dirty(leaf);
812 fi = btrfs_item_ptr(leaf, del_slot - 1,
813 struct btrfs_file_extent_item);
814 btrfs_set_file_extent_type(leaf, fi,
815 BTRFS_FILE_EXTENT_REG);
816 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
817 btrfs_set_file_extent_num_bytes(leaf, fi,
818 extent_end - key.offset);
819 btrfs_mark_buffer_dirty(leaf);
821 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
823 btrfs_abort_transaction(trans, ret);
828 btrfs_free_path(path);
833 * on error we return an unlocked page and the error value
834 * on success we return a locked page and 0
836 static int prepare_uptodate_page(struct inode *inode,
837 struct page *page, u64 pos,
840 struct folio *folio = page_folio(page);
843 if (((pos & (PAGE_SIZE - 1)) || force_uptodate) &&
844 !PageUptodate(page)) {
845 ret = btrfs_read_folio(NULL, folio);
849 if (!PageUptodate(page)) {
855 * Since btrfs_read_folio() will unlock the folio before it
856 * returns, there is a window where btrfs_release_folio() can be
857 * called to release the page. Here we check both inode
858 * mapping and PagePrivate() to make sure the page was not
861 * The private flag check is essential for subpage as we need
862 * to store extra bitmap using page->private.
864 if (page->mapping != inode->i_mapping || !PagePrivate(page)) {
872 static unsigned int get_prepare_fgp_flags(bool nowait)
874 unsigned int fgp_flags = FGP_LOCK | FGP_ACCESSED | FGP_CREAT;
877 fgp_flags |= FGP_NOWAIT;
882 static gfp_t get_prepare_gfp_flags(struct inode *inode, bool nowait)
886 gfp = btrfs_alloc_write_mask(inode->i_mapping);
888 gfp &= ~__GFP_DIRECT_RECLAIM;
896 * this just gets pages into the page cache and locks them down.
898 static noinline int prepare_pages(struct inode *inode, struct page **pages,
899 size_t num_pages, loff_t pos,
900 size_t write_bytes, bool force_uptodate,
904 unsigned long index = pos >> PAGE_SHIFT;
905 gfp_t mask = get_prepare_gfp_flags(inode, nowait);
906 unsigned int fgp_flags = get_prepare_fgp_flags(nowait);
910 for (i = 0; i < num_pages; i++) {
912 pages[i] = pagecache_get_page(inode->i_mapping, index + i,
913 fgp_flags, mask | __GFP_WRITE);
923 err = set_page_extent_mapped(pages[i]);
930 err = prepare_uptodate_page(inode, pages[i], pos,
932 if (!err && i == num_pages - 1)
933 err = prepare_uptodate_page(inode, pages[i],
934 pos + write_bytes, false);
937 if (!nowait && err == -EAGAIN) {
944 wait_on_page_writeback(pages[i]);
950 unlock_page(pages[faili]);
951 put_page(pages[faili]);
959 * This function locks the extent and properly waits for data=ordered extents
960 * to finish before allowing the pages to be modified if need.
963 * 1 - the extent is locked
964 * 0 - the extent is not locked, and everything is OK
965 * -EAGAIN - need re-prepare the pages
966 * the other < 0 number - Something wrong happens
969 lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct page **pages,
970 size_t num_pages, loff_t pos,
972 u64 *lockstart, u64 *lockend, bool nowait,
973 struct extent_state **cached_state)
975 struct btrfs_fs_info *fs_info = inode->root->fs_info;
981 start_pos = round_down(pos, fs_info->sectorsize);
982 last_pos = round_up(pos + write_bytes, fs_info->sectorsize) - 1;
984 if (start_pos < inode->vfs_inode.i_size) {
985 struct btrfs_ordered_extent *ordered;
988 if (!try_lock_extent(&inode->io_tree, start_pos, last_pos)) {
989 for (i = 0; i < num_pages; i++) {
990 unlock_page(pages[i]);
998 lock_extent(&inode->io_tree, start_pos, last_pos, cached_state);
1001 ordered = btrfs_lookup_ordered_range(inode, start_pos,
1002 last_pos - start_pos + 1);
1004 ordered->file_offset + ordered->num_bytes > start_pos &&
1005 ordered->file_offset <= last_pos) {
1006 unlock_extent(&inode->io_tree, start_pos, last_pos,
1008 for (i = 0; i < num_pages; i++) {
1009 unlock_page(pages[i]);
1012 btrfs_start_ordered_extent(ordered, 1);
1013 btrfs_put_ordered_extent(ordered);
1017 btrfs_put_ordered_extent(ordered);
1019 *lockstart = start_pos;
1020 *lockend = last_pos;
1025 * We should be called after prepare_pages() which should have locked
1026 * all pages in the range.
1028 for (i = 0; i < num_pages; i++)
1029 WARN_ON(!PageLocked(pages[i]));
1035 * Check if we can do nocow write into the range [@pos, @pos + @write_bytes)
1037 * @pos: File offset.
1038 * @write_bytes: The length to write, will be updated to the nocow writeable
1041 * This function will flush ordered extents in the range to ensure proper
1045 * > 0 If we can nocow, and updates @write_bytes.
1046 * 0 If we can't do a nocow write.
1047 * -EAGAIN If we can't do a nocow write because snapshoting of the inode's
1048 * root is in progress.
1049 * < 0 If an error happened.
1051 * NOTE: Callers need to call btrfs_check_nocow_unlock() if we return > 0.
1053 int btrfs_check_nocow_lock(struct btrfs_inode *inode, loff_t pos,
1054 size_t *write_bytes, bool nowait)
1056 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1057 struct btrfs_root *root = inode->root;
1058 u64 lockstart, lockend;
1062 if (!(inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
1065 if (!btrfs_drew_try_write_lock(&root->snapshot_lock))
1068 lockstart = round_down(pos, fs_info->sectorsize);
1069 lockend = round_up(pos + *write_bytes,
1070 fs_info->sectorsize) - 1;
1071 num_bytes = lockend - lockstart + 1;
1074 if (!btrfs_try_lock_ordered_range(inode, lockstart, lockend)) {
1075 btrfs_drew_write_unlock(&root->snapshot_lock);
1079 btrfs_lock_and_flush_ordered_range(inode, lockstart, lockend, NULL);
1081 ret = can_nocow_extent(&inode->vfs_inode, lockstart, &num_bytes,
1082 NULL, NULL, NULL, nowait, false);
1084 btrfs_drew_write_unlock(&root->snapshot_lock);
1086 *write_bytes = min_t(size_t, *write_bytes ,
1087 num_bytes - pos + lockstart);
1088 unlock_extent(&inode->io_tree, lockstart, lockend, NULL);
1093 void btrfs_check_nocow_unlock(struct btrfs_inode *inode)
1095 btrfs_drew_write_unlock(&inode->root->snapshot_lock);
1098 static void update_time_for_write(struct inode *inode)
1100 struct timespec64 now;
1102 if (IS_NOCMTIME(inode))
1105 now = current_time(inode);
1106 if (!timespec64_equal(&inode->i_mtime, &now))
1107 inode->i_mtime = now;
1109 if (!timespec64_equal(&inode->i_ctime, &now))
1110 inode->i_ctime = now;
1112 if (IS_I_VERSION(inode))
1113 inode_inc_iversion(inode);
1116 static int btrfs_write_check(struct kiocb *iocb, struct iov_iter *from,
1119 struct file *file = iocb->ki_filp;
1120 struct inode *inode = file_inode(file);
1121 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1122 loff_t pos = iocb->ki_pos;
1128 * Quickly bail out on NOWAIT writes if we don't have the nodatacow or
1129 * prealloc flags, as without those flags we always have to COW. We will
1130 * later check if we can really COW into the target range (using
1131 * can_nocow_extent() at btrfs_get_blocks_direct_write()).
1133 if ((iocb->ki_flags & IOCB_NOWAIT) &&
1134 !(BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
1137 current->backing_dev_info = inode_to_bdi(inode);
1138 ret = file_remove_privs(file);
1143 * We reserve space for updating the inode when we reserve space for the
1144 * extent we are going to write, so we will enospc out there. We don't
1145 * need to start yet another transaction to update the inode as we will
1146 * update the inode when we finish writing whatever data we write.
1148 update_time_for_write(inode);
1150 start_pos = round_down(pos, fs_info->sectorsize);
1151 oldsize = i_size_read(inode);
1152 if (start_pos > oldsize) {
1153 /* Expand hole size to cover write data, preventing empty gap */
1154 loff_t end_pos = round_up(pos + count, fs_info->sectorsize);
1156 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, end_pos);
1158 current->backing_dev_info = NULL;
1166 static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb,
1169 struct file *file = iocb->ki_filp;
1171 struct inode *inode = file_inode(file);
1172 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1173 struct page **pages = NULL;
1174 struct extent_changeset *data_reserved = NULL;
1175 u64 release_bytes = 0;
1178 size_t num_written = 0;
1181 bool only_release_metadata = false;
1182 bool force_page_uptodate = false;
1183 loff_t old_isize = i_size_read(inode);
1184 unsigned int ilock_flags = 0;
1185 const bool nowait = (iocb->ki_flags & IOCB_NOWAIT);
1186 unsigned int bdp_flags = (nowait ? BDP_ASYNC : 0);
1189 ilock_flags |= BTRFS_ILOCK_TRY;
1191 ret = btrfs_inode_lock(inode, ilock_flags);
1195 ret = generic_write_checks(iocb, i);
1199 ret = btrfs_write_check(iocb, i, ret);
1204 nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_SIZE),
1205 PAGE_SIZE / (sizeof(struct page *)));
1206 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1207 nrptrs = max(nrptrs, 8);
1208 pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL);
1214 while (iov_iter_count(i) > 0) {
1215 struct extent_state *cached_state = NULL;
1216 size_t offset = offset_in_page(pos);
1217 size_t sector_offset;
1218 size_t write_bytes = min(iov_iter_count(i),
1219 nrptrs * (size_t)PAGE_SIZE -
1222 size_t reserve_bytes;
1225 size_t dirty_sectors;
1230 * Fault pages before locking them in prepare_pages
1231 * to avoid recursive lock
1233 if (unlikely(fault_in_iov_iter_readable(i, write_bytes))) {
1238 only_release_metadata = false;
1239 sector_offset = pos & (fs_info->sectorsize - 1);
1241 extent_changeset_release(data_reserved);
1242 ret = btrfs_check_data_free_space(BTRFS_I(inode),
1243 &data_reserved, pos,
1244 write_bytes, nowait);
1248 if (nowait && (ret == -ENOSPC || ret == -EAGAIN)) {
1254 * If we don't have to COW at the offset, reserve
1255 * metadata only. write_bytes may get smaller than
1258 can_nocow = btrfs_check_nocow_lock(BTRFS_I(inode), pos,
1259 &write_bytes, nowait);
1266 only_release_metadata = true;
1269 num_pages = DIV_ROUND_UP(write_bytes + offset, PAGE_SIZE);
1270 WARN_ON(num_pages > nrptrs);
1271 reserve_bytes = round_up(write_bytes + sector_offset,
1272 fs_info->sectorsize);
1273 WARN_ON(reserve_bytes == 0);
1274 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
1276 reserve_bytes, nowait);
1278 if (!only_release_metadata)
1279 btrfs_free_reserved_data_space(BTRFS_I(inode),
1283 btrfs_check_nocow_unlock(BTRFS_I(inode));
1285 if (nowait && ret == -ENOSPC)
1290 release_bytes = reserve_bytes;
1292 ret = balance_dirty_pages_ratelimited_flags(inode->i_mapping, bdp_flags);
1294 btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes);
1299 * This is going to setup the pages array with the number of
1300 * pages we want, so we don't really need to worry about the
1301 * contents of pages from loop to loop
1303 ret = prepare_pages(inode, pages, num_pages,
1304 pos, write_bytes, force_page_uptodate, false);
1306 btrfs_delalloc_release_extents(BTRFS_I(inode),
1311 extents_locked = lock_and_cleanup_extent_if_need(
1312 BTRFS_I(inode), pages,
1313 num_pages, pos, write_bytes, &lockstart,
1314 &lockend, nowait, &cached_state);
1315 if (extents_locked < 0) {
1316 if (!nowait && extents_locked == -EAGAIN)
1319 btrfs_delalloc_release_extents(BTRFS_I(inode),
1321 ret = extents_locked;
1325 copied = btrfs_copy_from_user(pos, write_bytes, pages, i);
1327 num_sectors = BTRFS_BYTES_TO_BLKS(fs_info, reserve_bytes);
1328 dirty_sectors = round_up(copied + sector_offset,
1329 fs_info->sectorsize);
1330 dirty_sectors = BTRFS_BYTES_TO_BLKS(fs_info, dirty_sectors);
1333 * if we have trouble faulting in the pages, fall
1334 * back to one page at a time
1336 if (copied < write_bytes)
1340 force_page_uptodate = true;
1344 force_page_uptodate = false;
1345 dirty_pages = DIV_ROUND_UP(copied + offset,
1349 if (num_sectors > dirty_sectors) {
1350 /* release everything except the sectors we dirtied */
1351 release_bytes -= dirty_sectors << fs_info->sectorsize_bits;
1352 if (only_release_metadata) {
1353 btrfs_delalloc_release_metadata(BTRFS_I(inode),
1354 release_bytes, true);
1358 __pos = round_down(pos,
1359 fs_info->sectorsize) +
1360 (dirty_pages << PAGE_SHIFT);
1361 btrfs_delalloc_release_space(BTRFS_I(inode),
1362 data_reserved, __pos,
1363 release_bytes, true);
1367 release_bytes = round_up(copied + sector_offset,
1368 fs_info->sectorsize);
1370 ret = btrfs_dirty_pages(BTRFS_I(inode), pages,
1371 dirty_pages, pos, copied,
1372 &cached_state, only_release_metadata);
1375 * If we have not locked the extent range, because the range's
1376 * start offset is >= i_size, we might still have a non-NULL
1377 * cached extent state, acquired while marking the extent range
1378 * as delalloc through btrfs_dirty_pages(). Therefore free any
1379 * possible cached extent state to avoid a memory leak.
1382 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart,
1383 lockend, &cached_state);
1385 free_extent_state(cached_state);
1387 btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes);
1389 btrfs_drop_pages(fs_info, pages, num_pages, pos, copied);
1394 if (only_release_metadata)
1395 btrfs_check_nocow_unlock(BTRFS_I(inode));
1397 btrfs_drop_pages(fs_info, pages, num_pages, pos, copied);
1402 num_written += copied;
1407 if (release_bytes) {
1408 if (only_release_metadata) {
1409 btrfs_check_nocow_unlock(BTRFS_I(inode));
1410 btrfs_delalloc_release_metadata(BTRFS_I(inode),
1411 release_bytes, true);
1413 btrfs_delalloc_release_space(BTRFS_I(inode),
1415 round_down(pos, fs_info->sectorsize),
1416 release_bytes, true);
1420 extent_changeset_free(data_reserved);
1421 if (num_written > 0) {
1422 pagecache_isize_extended(inode, old_isize, iocb->ki_pos);
1423 iocb->ki_pos += num_written;
1426 btrfs_inode_unlock(inode, ilock_flags);
1427 return num_written ? num_written : ret;
1430 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
1431 const struct iov_iter *iter, loff_t offset)
1433 const u32 blocksize_mask = fs_info->sectorsize - 1;
1435 if (offset & blocksize_mask)
1438 if (iov_iter_alignment(iter) & blocksize_mask)
1444 static ssize_t btrfs_direct_write(struct kiocb *iocb, struct iov_iter *from)
1446 struct file *file = iocb->ki_filp;
1447 struct inode *inode = file_inode(file);
1448 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1450 ssize_t written = 0;
1451 ssize_t written_buffered;
1452 size_t prev_left = 0;
1455 unsigned int ilock_flags = 0;
1456 struct iomap_dio *dio;
1458 if (iocb->ki_flags & IOCB_NOWAIT)
1459 ilock_flags |= BTRFS_ILOCK_TRY;
1462 * If the write DIO is within EOF, use a shared lock and also only if
1463 * security bits will likely not be dropped by file_remove_privs() called
1464 * from btrfs_write_check(). Either will need to be rechecked after the
1465 * lock was acquired.
1467 if (iocb->ki_pos + iov_iter_count(from) <= i_size_read(inode) && IS_NOSEC(inode))
1468 ilock_flags |= BTRFS_ILOCK_SHARED;
1471 err = btrfs_inode_lock(inode, ilock_flags);
1475 /* Shared lock cannot be used with security bits set. */
1476 if ((ilock_flags & BTRFS_ILOCK_SHARED) && !IS_NOSEC(inode)) {
1477 btrfs_inode_unlock(inode, ilock_flags);
1478 ilock_flags &= ~BTRFS_ILOCK_SHARED;
1482 err = generic_write_checks(iocb, from);
1484 btrfs_inode_unlock(inode, ilock_flags);
1488 err = btrfs_write_check(iocb, from, err);
1490 btrfs_inode_unlock(inode, ilock_flags);
1496 * Re-check since file size may have changed just before taking the
1497 * lock or pos may have changed because of O_APPEND in generic_write_check()
1499 if ((ilock_flags & BTRFS_ILOCK_SHARED) &&
1500 pos + iov_iter_count(from) > i_size_read(inode)) {
1501 btrfs_inode_unlock(inode, ilock_flags);
1502 ilock_flags &= ~BTRFS_ILOCK_SHARED;
1506 if (check_direct_IO(fs_info, from, pos)) {
1507 btrfs_inode_unlock(inode, ilock_flags);
1512 * The iov_iter can be mapped to the same file range we are writing to.
1513 * If that's the case, then we will deadlock in the iomap code, because
1514 * it first calls our callback btrfs_dio_iomap_begin(), which will create
1515 * an ordered extent, and after that it will fault in the pages that the
1516 * iov_iter refers to. During the fault in we end up in the readahead
1517 * pages code (starting at btrfs_readahead()), which will lock the range,
1518 * find that ordered extent and then wait for it to complete (at
1519 * btrfs_lock_and_flush_ordered_range()), resulting in a deadlock since
1520 * obviously the ordered extent can never complete as we didn't submit
1521 * yet the respective bio(s). This always happens when the buffer is
1522 * memory mapped to the same file range, since the iomap DIO code always
1523 * invalidates pages in the target file range (after starting and waiting
1524 * for any writeback).
1526 * So here we disable page faults in the iov_iter and then retry if we
1527 * got -EFAULT, faulting in the pages before the retry.
1529 from->nofault = true;
1530 dio = btrfs_dio_write(iocb, from, written);
1531 from->nofault = false;
1534 * iomap_dio_complete() will call btrfs_sync_file() if we have a dsync
1535 * iocb, and that needs to lock the inode. So unlock it before calling
1536 * iomap_dio_complete() to avoid a deadlock.
1538 btrfs_inode_unlock(inode, ilock_flags);
1540 if (IS_ERR_OR_NULL(dio))
1541 err = PTR_ERR_OR_ZERO(dio);
1543 err = iomap_dio_complete(dio);
1545 /* No increment (+=) because iomap returns a cumulative value. */
1549 if (iov_iter_count(from) > 0 && (err == -EFAULT || err > 0)) {
1550 const size_t left = iov_iter_count(from);
1552 * We have more data left to write. Try to fault in as many as
1553 * possible of the remainder pages and retry. We do this without
1554 * releasing and locking again the inode, to prevent races with
1557 * Also, in case the iov refers to pages in the file range of the
1558 * file we want to write to (due to a mmap), we could enter an
1559 * infinite loop if we retry after faulting the pages in, since
1560 * iomap will invalidate any pages in the range early on, before
1561 * it tries to fault in the pages of the iov. So we keep track of
1562 * how much was left of iov in the previous EFAULT and fallback
1563 * to buffered IO in case we haven't made any progress.
1565 if (left == prev_left) {
1568 fault_in_iov_iter_readable(from, left);
1575 * If 'err' is -ENOTBLK or we have not written all data, then it means
1576 * we must fallback to buffered IO.
1578 if ((err < 0 && err != -ENOTBLK) || !iov_iter_count(from))
1583 * If we are in a NOWAIT context, then return -EAGAIN to signal the caller
1584 * it must retry the operation in a context where blocking is acceptable,
1585 * since we currently don't have NOWAIT semantics support for buffered IO
1586 * and may block there for many reasons (reserving space for example).
1588 if (iocb->ki_flags & IOCB_NOWAIT) {
1594 written_buffered = btrfs_buffered_write(iocb, from);
1595 if (written_buffered < 0) {
1596 err = written_buffered;
1600 * Ensure all data is persisted. We want the next direct IO read to be
1601 * able to read what was just written.
1603 endbyte = pos + written_buffered - 1;
1604 err = btrfs_fdatawrite_range(inode, pos, endbyte);
1607 err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte);
1610 written += written_buffered;
1611 iocb->ki_pos = pos + written_buffered;
1612 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_SHIFT,
1613 endbyte >> PAGE_SHIFT);
1615 return err < 0 ? err : written;
1618 static ssize_t btrfs_encoded_write(struct kiocb *iocb, struct iov_iter *from,
1619 const struct btrfs_ioctl_encoded_io_args *encoded)
1621 struct file *file = iocb->ki_filp;
1622 struct inode *inode = file_inode(file);
1626 btrfs_inode_lock(inode, 0);
1627 count = encoded->len;
1628 ret = generic_write_checks_count(iocb, &count);
1629 if (ret == 0 && count != encoded->len) {
1631 * The write got truncated by generic_write_checks_count(). We
1632 * can't do a partial encoded write.
1636 if (ret || encoded->len == 0)
1639 ret = btrfs_write_check(iocb, from, encoded->len);
1643 ret = btrfs_do_encoded_write(iocb, from, encoded);
1645 btrfs_inode_unlock(inode, 0);
1649 ssize_t btrfs_do_write_iter(struct kiocb *iocb, struct iov_iter *from,
1650 const struct btrfs_ioctl_encoded_io_args *encoded)
1652 struct file *file = iocb->ki_filp;
1653 struct btrfs_inode *inode = BTRFS_I(file_inode(file));
1654 ssize_t num_written, num_sync;
1655 const bool sync = iocb_is_dsync(iocb);
1658 * If the fs flips readonly due to some impossible error, although we
1659 * have opened a file as writable, we have to stop this write operation
1660 * to ensure consistency.
1662 if (BTRFS_FS_ERROR(inode->root->fs_info))
1665 if (encoded && (iocb->ki_flags & IOCB_NOWAIT))
1669 atomic_inc(&inode->sync_writers);
1672 num_written = btrfs_encoded_write(iocb, from, encoded);
1673 num_sync = encoded->len;
1674 } else if (iocb->ki_flags & IOCB_DIRECT) {
1675 num_written = btrfs_direct_write(iocb, from);
1676 num_sync = num_written;
1678 num_written = btrfs_buffered_write(iocb, from);
1679 num_sync = num_written;
1682 btrfs_set_inode_last_sub_trans(inode);
1685 num_sync = generic_write_sync(iocb, num_sync);
1687 num_written = num_sync;
1691 atomic_dec(&inode->sync_writers);
1693 current->backing_dev_info = NULL;
1697 static ssize_t btrfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1699 return btrfs_do_write_iter(iocb, from, NULL);
1702 int btrfs_release_file(struct inode *inode, struct file *filp)
1704 struct btrfs_file_private *private = filp->private_data;
1706 if (private && private->filldir_buf)
1707 kfree(private->filldir_buf);
1709 filp->private_data = NULL;
1712 * Set by setattr when we are about to truncate a file from a non-zero
1713 * size to a zero size. This tries to flush down new bytes that may
1714 * have been written if the application were using truncate to replace
1717 if (test_and_clear_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
1718 &BTRFS_I(inode)->runtime_flags))
1719 filemap_flush(inode->i_mapping);
1723 static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end)
1726 struct blk_plug plug;
1729 * This is only called in fsync, which would do synchronous writes, so
1730 * a plug can merge adjacent IOs as much as possible. Esp. in case of
1731 * multiple disks using raid profile, a large IO can be split to
1732 * several segments of stripe length (currently 64K).
1734 blk_start_plug(&plug);
1735 atomic_inc(&BTRFS_I(inode)->sync_writers);
1736 ret = btrfs_fdatawrite_range(inode, start, end);
1737 atomic_dec(&BTRFS_I(inode)->sync_writers);
1738 blk_finish_plug(&plug);
1743 static inline bool skip_inode_logging(const struct btrfs_log_ctx *ctx)
1745 struct btrfs_inode *inode = BTRFS_I(ctx->inode);
1746 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1748 if (btrfs_inode_in_log(inode, fs_info->generation) &&
1749 list_empty(&ctx->ordered_extents))
1753 * If we are doing a fast fsync we can not bail out if the inode's
1754 * last_trans is <= then the last committed transaction, because we only
1755 * update the last_trans of the inode during ordered extent completion,
1756 * and for a fast fsync we don't wait for that, we only wait for the
1757 * writeback to complete.
1759 if (inode->last_trans <= fs_info->last_trans_committed &&
1760 (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) ||
1761 list_empty(&ctx->ordered_extents)))
1768 * fsync call for both files and directories. This logs the inode into
1769 * the tree log instead of forcing full commits whenever possible.
1771 * It needs to call filemap_fdatawait so that all ordered extent updates are
1772 * in the metadata btree are up to date for copying to the log.
1774 * It drops the inode mutex before doing the tree log commit. This is an
1775 * important optimization for directories because holding the mutex prevents
1776 * new operations on the dir while we write to disk.
1778 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1780 struct dentry *dentry = file_dentry(file);
1781 struct inode *inode = d_inode(dentry);
1782 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1783 struct btrfs_root *root = BTRFS_I(inode)->root;
1784 struct btrfs_trans_handle *trans;
1785 struct btrfs_log_ctx ctx;
1790 trace_btrfs_sync_file(file, datasync);
1792 btrfs_init_log_ctx(&ctx, inode);
1795 * Always set the range to a full range, otherwise we can get into
1796 * several problems, from missing file extent items to represent holes
1797 * when not using the NO_HOLES feature, to log tree corruption due to
1798 * races between hole detection during logging and completion of ordered
1799 * extents outside the range, to missing checksums due to ordered extents
1800 * for which we flushed only a subset of their pages.
1804 len = (u64)LLONG_MAX + 1;
1807 * We write the dirty pages in the range and wait until they complete
1808 * out of the ->i_mutex. If so, we can flush the dirty pages by
1809 * multi-task, and make the performance up. See
1810 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1812 ret = start_ordered_ops(inode, start, end);
1816 btrfs_inode_lock(inode, BTRFS_ILOCK_MMAP);
1818 atomic_inc(&root->log_batch);
1821 * Before we acquired the inode's lock and the mmap lock, someone may
1822 * have dirtied more pages in the target range. We need to make sure
1823 * that writeback for any such pages does not start while we are logging
1824 * the inode, because if it does, any of the following might happen when
1825 * we are not doing a full inode sync:
1827 * 1) We log an extent after its writeback finishes but before its
1828 * checksums are added to the csum tree, leading to -EIO errors
1829 * when attempting to read the extent after a log replay.
1831 * 2) We can end up logging an extent before its writeback finishes.
1832 * Therefore after the log replay we will have a file extent item
1833 * pointing to an unwritten extent (and no data checksums as well).
1835 * So trigger writeback for any eventual new dirty pages and then we
1836 * wait for all ordered extents to complete below.
1838 ret = start_ordered_ops(inode, start, end);
1840 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
1845 * Always check for the full sync flag while holding the inode's lock,
1846 * to avoid races with other tasks. The flag must be either set all the
1847 * time during logging or always off all the time while logging.
1848 * We check the flag here after starting delalloc above, because when
1849 * running delalloc the full sync flag may be set if we need to drop
1850 * extra extent map ranges due to temporary memory allocation failures.
1852 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1853 &BTRFS_I(inode)->runtime_flags);
1856 * We have to do this here to avoid the priority inversion of waiting on
1857 * IO of a lower priority task while holding a transaction open.
1859 * For a full fsync we wait for the ordered extents to complete while
1860 * for a fast fsync we wait just for writeback to complete, and then
1861 * attach the ordered extents to the transaction so that a transaction
1862 * commit waits for their completion, to avoid data loss if we fsync,
1863 * the current transaction commits before the ordered extents complete
1864 * and a power failure happens right after that.
1866 * For zoned filesystem, if a write IO uses a ZONE_APPEND command, the
1867 * logical address recorded in the ordered extent may change. We need
1868 * to wait for the IO to stabilize the logical address.
1870 if (full_sync || btrfs_is_zoned(fs_info)) {
1871 ret = btrfs_wait_ordered_range(inode, start, len);
1874 * Get our ordered extents as soon as possible to avoid doing
1875 * checksum lookups in the csum tree, and use instead the
1876 * checksums attached to the ordered extents.
1878 btrfs_get_ordered_extents_for_logging(BTRFS_I(inode),
1879 &ctx.ordered_extents);
1880 ret = filemap_fdatawait_range(inode->i_mapping, start, end);
1884 goto out_release_extents;
1886 atomic_inc(&root->log_batch);
1889 if (skip_inode_logging(&ctx)) {
1891 * We've had everything committed since the last time we were
1892 * modified so clear this flag in case it was set for whatever
1893 * reason, it's no longer relevant.
1895 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1896 &BTRFS_I(inode)->runtime_flags);
1898 * An ordered extent might have started before and completed
1899 * already with io errors, in which case the inode was not
1900 * updated and we end up here. So check the inode's mapping
1901 * for any errors that might have happened since we last
1902 * checked called fsync.
1904 ret = filemap_check_wb_err(inode->i_mapping, file->f_wb_err);
1905 goto out_release_extents;
1909 * We use start here because we will need to wait on the IO to complete
1910 * in btrfs_sync_log, which could require joining a transaction (for
1911 * example checking cross references in the nocow path). If we use join
1912 * here we could get into a situation where we're waiting on IO to
1913 * happen that is blocked on a transaction trying to commit. With start
1914 * we inc the extwriter counter, so we wait for all extwriters to exit
1915 * before we start blocking joiners. This comment is to keep somebody
1916 * from thinking they are super smart and changing this to
1917 * btrfs_join_transaction *cough*Josef*cough*.
1919 trans = btrfs_start_transaction(root, 0);
1920 if (IS_ERR(trans)) {
1921 ret = PTR_ERR(trans);
1922 goto out_release_extents;
1924 trans->in_fsync = true;
1926 ret = btrfs_log_dentry_safe(trans, dentry, &ctx);
1927 btrfs_release_log_ctx_extents(&ctx);
1929 /* Fallthrough and commit/free transaction. */
1930 ret = BTRFS_LOG_FORCE_COMMIT;
1933 /* we've logged all the items and now have a consistent
1934 * version of the file in the log. It is possible that
1935 * someone will come in and modify the file, but that's
1936 * fine because the log is consistent on disk, and we
1937 * have references to all of the file's extents
1939 * It is possible that someone will come in and log the
1940 * file again, but that will end up using the synchronization
1941 * inside btrfs_sync_log to keep things safe.
1943 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
1945 if (ret == BTRFS_NO_LOG_SYNC) {
1946 ret = btrfs_end_transaction(trans);
1950 /* We successfully logged the inode, attempt to sync the log. */
1952 ret = btrfs_sync_log(trans, root, &ctx);
1954 ret = btrfs_end_transaction(trans);
1960 * At this point we need to commit the transaction because we had
1961 * btrfs_need_log_full_commit() or some other error.
1963 * If we didn't do a full sync we have to stop the trans handle, wait on
1964 * the ordered extents, start it again and commit the transaction. If
1965 * we attempt to wait on the ordered extents here we could deadlock with
1966 * something like fallocate() that is holding the extent lock trying to
1967 * start a transaction while some other thread is trying to commit the
1968 * transaction while we (fsync) are currently holding the transaction
1972 ret = btrfs_end_transaction(trans);
1975 ret = btrfs_wait_ordered_range(inode, start, len);
1980 * This is safe to use here because we're only interested in
1981 * making sure the transaction that had the ordered extents is
1982 * committed. We aren't waiting on anything past this point,
1983 * we're purely getting the transaction and committing it.
1985 trans = btrfs_attach_transaction_barrier(root);
1986 if (IS_ERR(trans)) {
1987 ret = PTR_ERR(trans);
1990 * We committed the transaction and there's no currently
1991 * running transaction, this means everything we care
1992 * about made it to disk and we are done.
2000 ret = btrfs_commit_transaction(trans);
2002 ASSERT(list_empty(&ctx.list));
2003 ASSERT(list_empty(&ctx.conflict_inodes));
2004 err = file_check_and_advance_wb_err(file);
2007 return ret > 0 ? -EIO : ret;
2009 out_release_extents:
2010 btrfs_release_log_ctx_extents(&ctx);
2011 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
2015 static const struct vm_operations_struct btrfs_file_vm_ops = {
2016 .fault = filemap_fault,
2017 .map_pages = filemap_map_pages,
2018 .page_mkwrite = btrfs_page_mkwrite,
2021 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
2023 struct address_space *mapping = filp->f_mapping;
2025 if (!mapping->a_ops->read_folio)
2028 file_accessed(filp);
2029 vma->vm_ops = &btrfs_file_vm_ops;
2034 static int hole_mergeable(struct btrfs_inode *inode, struct extent_buffer *leaf,
2035 int slot, u64 start, u64 end)
2037 struct btrfs_file_extent_item *fi;
2038 struct btrfs_key key;
2040 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2043 btrfs_item_key_to_cpu(leaf, &key, slot);
2044 if (key.objectid != btrfs_ino(inode) ||
2045 key.type != BTRFS_EXTENT_DATA_KEY)
2048 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2050 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2053 if (btrfs_file_extent_disk_bytenr(leaf, fi))
2056 if (key.offset == end)
2058 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2063 static int fill_holes(struct btrfs_trans_handle *trans,
2064 struct btrfs_inode *inode,
2065 struct btrfs_path *path, u64 offset, u64 end)
2067 struct btrfs_fs_info *fs_info = trans->fs_info;
2068 struct btrfs_root *root = inode->root;
2069 struct extent_buffer *leaf;
2070 struct btrfs_file_extent_item *fi;
2071 struct extent_map *hole_em;
2072 struct btrfs_key key;
2075 if (btrfs_fs_incompat(fs_info, NO_HOLES))
2078 key.objectid = btrfs_ino(inode);
2079 key.type = BTRFS_EXTENT_DATA_KEY;
2080 key.offset = offset;
2082 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2085 * We should have dropped this offset, so if we find it then
2086 * something has gone horribly wrong.
2093 leaf = path->nodes[0];
2094 if (hole_mergeable(inode, leaf, path->slots[0] - 1, offset, end)) {
2098 fi = btrfs_item_ptr(leaf, path->slots[0],
2099 struct btrfs_file_extent_item);
2100 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2102 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2103 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2104 btrfs_set_file_extent_offset(leaf, fi, 0);
2105 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2106 btrfs_mark_buffer_dirty(leaf);
2110 if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) {
2113 key.offset = offset;
2114 btrfs_set_item_key_safe(fs_info, path, &key);
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) + end -
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(leaf);
2126 btrfs_release_path(path);
2128 ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset,
2134 btrfs_release_path(path);
2136 hole_em = alloc_extent_map();
2138 btrfs_drop_extent_map_range(inode, offset, end - 1, false);
2139 btrfs_set_inode_full_sync(inode);
2141 hole_em->start = offset;
2142 hole_em->len = end - offset;
2143 hole_em->ram_bytes = hole_em->len;
2144 hole_em->orig_start = offset;
2146 hole_em->block_start = EXTENT_MAP_HOLE;
2147 hole_em->block_len = 0;
2148 hole_em->orig_block_len = 0;
2149 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2150 hole_em->generation = trans->transid;
2152 ret = btrfs_replace_extent_map_range(inode, hole_em, true);
2153 free_extent_map(hole_em);
2155 btrfs_set_inode_full_sync(inode);
2162 * Find a hole extent on given inode and change start/len to the end of hole
2163 * extent.(hole/vacuum extent whose em->start <= start &&
2164 * em->start + em->len > start)
2165 * When a hole extent is found, return 1 and modify start/len.
2167 static int find_first_non_hole(struct btrfs_inode *inode, u64 *start, u64 *len)
2169 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2170 struct extent_map *em;
2173 em = btrfs_get_extent(inode, NULL, 0,
2174 round_down(*start, fs_info->sectorsize),
2175 round_up(*len, fs_info->sectorsize));
2179 /* Hole or vacuum extent(only exists in no-hole mode) */
2180 if (em->block_start == EXTENT_MAP_HOLE) {
2182 *len = em->start + em->len > *start + *len ?
2183 0 : *start + *len - em->start - em->len;
2184 *start = em->start + em->len;
2186 free_extent_map(em);
2190 static void btrfs_punch_hole_lock_range(struct inode *inode,
2191 const u64 lockstart,
2193 struct extent_state **cached_state)
2196 * For subpage case, if the range is not at page boundary, we could
2197 * have pages at the leading/tailing part of the range.
2198 * This could lead to dead loop since filemap_range_has_page()
2199 * will always return true.
2200 * So here we need to do extra page alignment for
2201 * filemap_range_has_page().
2203 const u64 page_lockstart = round_up(lockstart, PAGE_SIZE);
2204 const u64 page_lockend = round_down(lockend + 1, PAGE_SIZE) - 1;
2207 truncate_pagecache_range(inode, lockstart, lockend);
2209 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2212 * We can't have ordered extents in the range, nor dirty/writeback
2213 * pages, because we have locked the inode's VFS lock in exclusive
2214 * mode, we have locked the inode's i_mmap_lock in exclusive mode,
2215 * we have flushed all delalloc in the range and we have waited
2216 * for any ordered extents in the range to complete.
2217 * We can race with anyone reading pages from this range, so after
2218 * locking the range check if we have pages in the range, and if
2219 * we do, unlock the range and retry.
2221 if (!filemap_range_has_page(inode->i_mapping, page_lockstart,
2225 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2229 btrfs_assert_inode_range_clean(BTRFS_I(inode), lockstart, lockend);
2232 static int btrfs_insert_replace_extent(struct btrfs_trans_handle *trans,
2233 struct btrfs_inode *inode,
2234 struct btrfs_path *path,
2235 struct btrfs_replace_extent_info *extent_info,
2236 const u64 replace_len,
2237 const u64 bytes_to_drop)
2239 struct btrfs_fs_info *fs_info = trans->fs_info;
2240 struct btrfs_root *root = inode->root;
2241 struct btrfs_file_extent_item *extent;
2242 struct extent_buffer *leaf;
2243 struct btrfs_key key;
2245 struct btrfs_ref ref = { 0 };
2248 if (replace_len == 0)
2251 if (extent_info->disk_offset == 0 &&
2252 btrfs_fs_incompat(fs_info, NO_HOLES)) {
2253 btrfs_update_inode_bytes(inode, 0, bytes_to_drop);
2257 key.objectid = btrfs_ino(inode);
2258 key.type = BTRFS_EXTENT_DATA_KEY;
2259 key.offset = extent_info->file_offset;
2260 ret = btrfs_insert_empty_item(trans, root, path, &key,
2261 sizeof(struct btrfs_file_extent_item));
2264 leaf = path->nodes[0];
2265 slot = path->slots[0];
2266 write_extent_buffer(leaf, extent_info->extent_buf,
2267 btrfs_item_ptr_offset(leaf, slot),
2268 sizeof(struct btrfs_file_extent_item));
2269 extent = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2270 ASSERT(btrfs_file_extent_type(leaf, extent) != BTRFS_FILE_EXTENT_INLINE);
2271 btrfs_set_file_extent_offset(leaf, extent, extent_info->data_offset);
2272 btrfs_set_file_extent_num_bytes(leaf, extent, replace_len);
2273 if (extent_info->is_new_extent)
2274 btrfs_set_file_extent_generation(leaf, extent, trans->transid);
2275 btrfs_mark_buffer_dirty(leaf);
2276 btrfs_release_path(path);
2278 ret = btrfs_inode_set_file_extent_range(inode, extent_info->file_offset,
2283 /* If it's a hole, nothing more needs to be done. */
2284 if (extent_info->disk_offset == 0) {
2285 btrfs_update_inode_bytes(inode, 0, bytes_to_drop);
2289 btrfs_update_inode_bytes(inode, replace_len, bytes_to_drop);
2291 if (extent_info->is_new_extent && extent_info->insertions == 0) {
2292 key.objectid = extent_info->disk_offset;
2293 key.type = BTRFS_EXTENT_ITEM_KEY;
2294 key.offset = extent_info->disk_len;
2295 ret = btrfs_alloc_reserved_file_extent(trans, root,
2297 extent_info->file_offset,
2298 extent_info->qgroup_reserved,
2303 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
2304 extent_info->disk_offset,
2305 extent_info->disk_len, 0);
2306 ref_offset = extent_info->file_offset - extent_info->data_offset;
2307 btrfs_init_data_ref(&ref, root->root_key.objectid,
2308 btrfs_ino(inode), ref_offset, 0, false);
2309 ret = btrfs_inc_extent_ref(trans, &ref);
2312 extent_info->insertions++;
2318 * The respective range must have been previously locked, as well as the inode.
2319 * The end offset is inclusive (last byte of the range).
2320 * @extent_info is NULL for fallocate's hole punching and non-NULL when replacing
2321 * the file range with an extent.
2322 * When not punching a hole, we don't want to end up in a state where we dropped
2323 * extents without inserting a new one, so we must abort the transaction to avoid
2326 int btrfs_replace_file_extents(struct btrfs_inode *inode,
2327 struct btrfs_path *path, const u64 start,
2329 struct btrfs_replace_extent_info *extent_info,
2330 struct btrfs_trans_handle **trans_out)
2332 struct btrfs_drop_extents_args drop_args = { 0 };
2333 struct btrfs_root *root = inode->root;
2334 struct btrfs_fs_info *fs_info = root->fs_info;
2335 u64 min_size = btrfs_calc_insert_metadata_size(fs_info, 1);
2336 u64 ino_size = round_up(inode->vfs_inode.i_size, fs_info->sectorsize);
2337 struct btrfs_trans_handle *trans = NULL;
2338 struct btrfs_block_rsv *rsv;
2339 unsigned int rsv_count;
2341 u64 len = end - start;
2347 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
2352 rsv->size = btrfs_calc_insert_metadata_size(fs_info, 1);
2353 rsv->failfast = true;
2356 * 1 - update the inode
2357 * 1 - removing the extents in the range
2358 * 1 - adding the hole extent if no_holes isn't set or if we are
2359 * replacing the range with a new extent
2361 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || extent_info)
2366 trans = btrfs_start_transaction(root, rsv_count);
2367 if (IS_ERR(trans)) {
2368 ret = PTR_ERR(trans);
2373 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
2377 trans->block_rsv = rsv;
2380 drop_args.path = path;
2381 drop_args.end = end + 1;
2382 drop_args.drop_cache = true;
2383 while (cur_offset < end) {
2384 drop_args.start = cur_offset;
2385 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2386 /* If we are punching a hole decrement the inode's byte count */
2388 btrfs_update_inode_bytes(inode, 0,
2389 drop_args.bytes_found);
2390 if (ret != -ENOSPC) {
2392 * The only time we don't want to abort is if we are
2393 * attempting to clone a partial inline extent, in which
2394 * case we'll get EOPNOTSUPP. However if we aren't
2395 * clone we need to abort no matter what, because if we
2396 * got EOPNOTSUPP via prealloc then we messed up and
2400 (ret != -EOPNOTSUPP ||
2401 (extent_info && extent_info->is_new_extent)))
2402 btrfs_abort_transaction(trans, ret);
2406 trans->block_rsv = &fs_info->trans_block_rsv;
2408 if (!extent_info && cur_offset < drop_args.drop_end &&
2409 cur_offset < ino_size) {
2410 ret = fill_holes(trans, inode, path, cur_offset,
2411 drop_args.drop_end);
2414 * If we failed then we didn't insert our hole
2415 * entries for the area we dropped, so now the
2416 * fs is corrupted, so we must abort the
2419 btrfs_abort_transaction(trans, ret);
2422 } else if (!extent_info && cur_offset < drop_args.drop_end) {
2424 * We are past the i_size here, but since we didn't
2425 * insert holes we need to clear the mapped area so we
2426 * know to not set disk_i_size in this area until a new
2427 * file extent is inserted here.
2429 ret = btrfs_inode_clear_file_extent_range(inode,
2431 drop_args.drop_end - cur_offset);
2434 * We couldn't clear our area, so we could
2435 * presumably adjust up and corrupt the fs, so
2438 btrfs_abort_transaction(trans, ret);
2444 drop_args.drop_end > extent_info->file_offset) {
2445 u64 replace_len = drop_args.drop_end -
2446 extent_info->file_offset;
2448 ret = btrfs_insert_replace_extent(trans, inode, path,
2449 extent_info, replace_len,
2450 drop_args.bytes_found);
2452 btrfs_abort_transaction(trans, ret);
2455 extent_info->data_len -= replace_len;
2456 extent_info->data_offset += replace_len;
2457 extent_info->file_offset += replace_len;
2461 * We are releasing our handle on the transaction, balance the
2462 * dirty pages of the btree inode and flush delayed items, and
2463 * then get a new transaction handle, which may now point to a
2464 * new transaction in case someone else may have committed the
2465 * transaction we used to replace/drop file extent items. So
2466 * bump the inode's iversion and update mtime and ctime except
2467 * if we are called from a dedupe context. This is because a
2468 * power failure/crash may happen after the transaction is
2469 * committed and before we finish replacing/dropping all the
2470 * file extent items we need.
2472 inode_inc_iversion(&inode->vfs_inode);
2474 if (!extent_info || extent_info->update_times) {
2475 inode->vfs_inode.i_mtime = current_time(&inode->vfs_inode);
2476 inode->vfs_inode.i_ctime = inode->vfs_inode.i_mtime;
2479 ret = btrfs_update_inode(trans, root, inode);
2483 btrfs_end_transaction(trans);
2484 btrfs_btree_balance_dirty(fs_info);
2486 trans = btrfs_start_transaction(root, rsv_count);
2487 if (IS_ERR(trans)) {
2488 ret = PTR_ERR(trans);
2493 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
2494 rsv, min_size, false);
2497 trans->block_rsv = rsv;
2499 cur_offset = drop_args.drop_end;
2500 len = end - cur_offset;
2501 if (!extent_info && len) {
2502 ret = find_first_non_hole(inode, &cur_offset, &len);
2503 if (unlikely(ret < 0))
2513 * If we were cloning, force the next fsync to be a full one since we
2514 * we replaced (or just dropped in the case of cloning holes when
2515 * NO_HOLES is enabled) file extent items and did not setup new extent
2516 * maps for the replacement extents (or holes).
2518 if (extent_info && !extent_info->is_new_extent)
2519 btrfs_set_inode_full_sync(inode);
2524 trans->block_rsv = &fs_info->trans_block_rsv;
2526 * If we are using the NO_HOLES feature we might have had already an
2527 * hole that overlaps a part of the region [lockstart, lockend] and
2528 * ends at (or beyond) lockend. Since we have no file extent items to
2529 * represent holes, drop_end can be less than lockend and so we must
2530 * make sure we have an extent map representing the existing hole (the
2531 * call to __btrfs_drop_extents() might have dropped the existing extent
2532 * map representing the existing hole), otherwise the fast fsync path
2533 * will not record the existence of the hole region
2534 * [existing_hole_start, lockend].
2536 if (drop_args.drop_end <= end)
2537 drop_args.drop_end = end + 1;
2539 * Don't insert file hole extent item if it's for a range beyond eof
2540 * (because it's useless) or if it represents a 0 bytes range (when
2541 * cur_offset == drop_end).
2543 if (!extent_info && cur_offset < ino_size &&
2544 cur_offset < drop_args.drop_end) {
2545 ret = fill_holes(trans, inode, path, cur_offset,
2546 drop_args.drop_end);
2548 /* Same comment as above. */
2549 btrfs_abort_transaction(trans, ret);
2552 } else if (!extent_info && cur_offset < drop_args.drop_end) {
2553 /* See the comment in the loop above for the reasoning here. */
2554 ret = btrfs_inode_clear_file_extent_range(inode, cur_offset,
2555 drop_args.drop_end - cur_offset);
2557 btrfs_abort_transaction(trans, ret);
2563 ret = btrfs_insert_replace_extent(trans, inode, path,
2564 extent_info, extent_info->data_len,
2565 drop_args.bytes_found);
2567 btrfs_abort_transaction(trans, ret);
2576 trans->block_rsv = &fs_info->trans_block_rsv;
2578 btrfs_end_transaction(trans);
2582 btrfs_free_block_rsv(fs_info, rsv);
2587 static int btrfs_punch_hole(struct file *file, loff_t offset, loff_t len)
2589 struct inode *inode = file_inode(file);
2590 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2591 struct btrfs_root *root = BTRFS_I(inode)->root;
2592 struct extent_state *cached_state = NULL;
2593 struct btrfs_path *path;
2594 struct btrfs_trans_handle *trans = NULL;
2599 u64 orig_start = offset;
2603 bool truncated_block = false;
2604 bool updated_inode = false;
2606 btrfs_inode_lock(inode, BTRFS_ILOCK_MMAP);
2608 ret = btrfs_wait_ordered_range(inode, offset, len);
2610 goto out_only_mutex;
2612 ino_size = round_up(inode->i_size, fs_info->sectorsize);
2613 ret = find_first_non_hole(BTRFS_I(inode), &offset, &len);
2615 goto out_only_mutex;
2617 /* Already in a large hole */
2619 goto out_only_mutex;
2622 ret = file_modified(file);
2624 goto out_only_mutex;
2626 lockstart = round_up(offset, fs_info->sectorsize);
2627 lockend = round_down(offset + len, fs_info->sectorsize) - 1;
2628 same_block = (BTRFS_BYTES_TO_BLKS(fs_info, offset))
2629 == (BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1));
2631 * We needn't truncate any block which is beyond the end of the file
2632 * because we are sure there is no data there.
2635 * Only do this if we are in the same block and we aren't doing the
2638 if (same_block && len < fs_info->sectorsize) {
2639 if (offset < ino_size) {
2640 truncated_block = true;
2641 ret = btrfs_truncate_block(BTRFS_I(inode), offset, len,
2646 goto out_only_mutex;
2649 /* zero back part of the first block */
2650 if (offset < ino_size) {
2651 truncated_block = true;
2652 ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0);
2654 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
2659 /* Check the aligned pages after the first unaligned page,
2660 * if offset != orig_start, which means the first unaligned page
2661 * including several following pages are already in holes,
2662 * the extra check can be skipped */
2663 if (offset == orig_start) {
2664 /* after truncate page, check hole again */
2665 len = offset + len - lockstart;
2667 ret = find_first_non_hole(BTRFS_I(inode), &offset, &len);
2669 goto out_only_mutex;
2672 goto out_only_mutex;
2677 /* Check the tail unaligned part is in a hole */
2678 tail_start = lockend + 1;
2679 tail_len = offset + len - tail_start;
2681 ret = find_first_non_hole(BTRFS_I(inode), &tail_start, &tail_len);
2682 if (unlikely(ret < 0))
2683 goto out_only_mutex;
2685 /* zero the front end of the last page */
2686 if (tail_start + tail_len < ino_size) {
2687 truncated_block = true;
2688 ret = btrfs_truncate_block(BTRFS_I(inode),
2689 tail_start + tail_len,
2692 goto out_only_mutex;
2697 if (lockend < lockstart) {
2699 goto out_only_mutex;
2702 btrfs_punch_hole_lock_range(inode, lockstart, lockend, &cached_state);
2704 path = btrfs_alloc_path();
2710 ret = btrfs_replace_file_extents(BTRFS_I(inode), path, lockstart,
2711 lockend, NULL, &trans);
2712 btrfs_free_path(path);
2716 ASSERT(trans != NULL);
2717 inode_inc_iversion(inode);
2718 inode->i_mtime = current_time(inode);
2719 inode->i_ctime = inode->i_mtime;
2720 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
2721 updated_inode = true;
2722 btrfs_end_transaction(trans);
2723 btrfs_btree_balance_dirty(fs_info);
2725 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2728 if (!updated_inode && truncated_block && !ret) {
2730 * If we only end up zeroing part of a page, we still need to
2731 * update the inode item, so that all the time fields are
2732 * updated as well as the necessary btrfs inode in memory fields
2733 * for detecting, at fsync time, if the inode isn't yet in the
2734 * log tree or it's there but not up to date.
2736 struct timespec64 now = current_time(inode);
2738 inode_inc_iversion(inode);
2739 inode->i_mtime = now;
2740 inode->i_ctime = now;
2741 trans = btrfs_start_transaction(root, 1);
2742 if (IS_ERR(trans)) {
2743 ret = PTR_ERR(trans);
2747 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
2748 ret2 = btrfs_end_transaction(trans);
2753 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
2757 /* Helper structure to record which range is already reserved */
2758 struct falloc_range {
2759 struct list_head list;
2765 * Helper function to add falloc range
2767 * Caller should have locked the larger range of extent containing
2770 static int add_falloc_range(struct list_head *head, u64 start, u64 len)
2772 struct falloc_range *range = NULL;
2774 if (!list_empty(head)) {
2776 * As fallocate iterates by bytenr order, we only need to check
2779 range = list_last_entry(head, struct falloc_range, list);
2780 if (range->start + range->len == start) {
2786 range = kmalloc(sizeof(*range), GFP_KERNEL);
2789 range->start = start;
2791 list_add_tail(&range->list, head);
2795 static int btrfs_fallocate_update_isize(struct inode *inode,
2799 struct btrfs_trans_handle *trans;
2800 struct btrfs_root *root = BTRFS_I(inode)->root;
2804 if (mode & FALLOC_FL_KEEP_SIZE || end <= i_size_read(inode))
2807 trans = btrfs_start_transaction(root, 1);
2809 return PTR_ERR(trans);
2811 inode->i_ctime = current_time(inode);
2812 i_size_write(inode, end);
2813 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
2814 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
2815 ret2 = btrfs_end_transaction(trans);
2817 return ret ? ret : ret2;
2821 RANGE_BOUNDARY_WRITTEN_EXTENT,
2822 RANGE_BOUNDARY_PREALLOC_EXTENT,
2823 RANGE_BOUNDARY_HOLE,
2826 static int btrfs_zero_range_check_range_boundary(struct btrfs_inode *inode,
2829 const u64 sectorsize = inode->root->fs_info->sectorsize;
2830 struct extent_map *em;
2833 offset = round_down(offset, sectorsize);
2834 em = btrfs_get_extent(inode, NULL, 0, offset, sectorsize);
2838 if (em->block_start == EXTENT_MAP_HOLE)
2839 ret = RANGE_BOUNDARY_HOLE;
2840 else if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2841 ret = RANGE_BOUNDARY_PREALLOC_EXTENT;
2843 ret = RANGE_BOUNDARY_WRITTEN_EXTENT;
2845 free_extent_map(em);
2849 static int btrfs_zero_range(struct inode *inode,
2854 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2855 struct extent_map *em;
2856 struct extent_changeset *data_reserved = NULL;
2859 const u64 sectorsize = fs_info->sectorsize;
2860 u64 alloc_start = round_down(offset, sectorsize);
2861 u64 alloc_end = round_up(offset + len, sectorsize);
2862 u64 bytes_to_reserve = 0;
2863 bool space_reserved = false;
2865 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start,
2866 alloc_end - alloc_start);
2873 * Avoid hole punching and extent allocation for some cases. More cases
2874 * could be considered, but these are unlikely common and we keep things
2875 * as simple as possible for now. Also, intentionally, if the target
2876 * range contains one or more prealloc extents together with regular
2877 * extents and holes, we drop all the existing extents and allocate a
2878 * new prealloc extent, so that we get a larger contiguous disk extent.
2880 if (em->start <= alloc_start &&
2881 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
2882 const u64 em_end = em->start + em->len;
2884 if (em_end >= offset + len) {
2886 * The whole range is already a prealloc extent,
2887 * do nothing except updating the inode's i_size if
2890 free_extent_map(em);
2891 ret = btrfs_fallocate_update_isize(inode, offset + len,
2896 * Part of the range is already a prealloc extent, so operate
2897 * only on the remaining part of the range.
2899 alloc_start = em_end;
2900 ASSERT(IS_ALIGNED(alloc_start, sectorsize));
2901 len = offset + len - alloc_start;
2902 offset = alloc_start;
2903 alloc_hint = em->block_start + em->len;
2905 free_extent_map(em);
2907 if (BTRFS_BYTES_TO_BLKS(fs_info, offset) ==
2908 BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)) {
2909 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start,
2916 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
2917 free_extent_map(em);
2918 ret = btrfs_fallocate_update_isize(inode, offset + len,
2922 if (len < sectorsize && em->block_start != EXTENT_MAP_HOLE) {
2923 free_extent_map(em);
2924 ret = btrfs_truncate_block(BTRFS_I(inode), offset, len,
2927 ret = btrfs_fallocate_update_isize(inode,
2932 free_extent_map(em);
2933 alloc_start = round_down(offset, sectorsize);
2934 alloc_end = alloc_start + sectorsize;
2938 alloc_start = round_up(offset, sectorsize);
2939 alloc_end = round_down(offset + len, sectorsize);
2942 * For unaligned ranges, check the pages at the boundaries, they might
2943 * map to an extent, in which case we need to partially zero them, or
2944 * they might map to a hole, in which case we need our allocation range
2947 if (!IS_ALIGNED(offset, sectorsize)) {
2948 ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode),
2952 if (ret == RANGE_BOUNDARY_HOLE) {
2953 alloc_start = round_down(offset, sectorsize);
2955 } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
2956 ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0);
2964 if (!IS_ALIGNED(offset + len, sectorsize)) {
2965 ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode),
2969 if (ret == RANGE_BOUNDARY_HOLE) {
2970 alloc_end = round_up(offset + len, sectorsize);
2972 } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
2973 ret = btrfs_truncate_block(BTRFS_I(inode), offset + len,
2983 if (alloc_start < alloc_end) {
2984 struct extent_state *cached_state = NULL;
2985 const u64 lockstart = alloc_start;
2986 const u64 lockend = alloc_end - 1;
2988 bytes_to_reserve = alloc_end - alloc_start;
2989 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
2993 space_reserved = true;
2994 btrfs_punch_hole_lock_range(inode, lockstart, lockend,
2996 ret = btrfs_qgroup_reserve_data(BTRFS_I(inode), &data_reserved,
2997 alloc_start, bytes_to_reserve);
2999 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart,
3000 lockend, &cached_state);
3003 ret = btrfs_prealloc_file_range(inode, mode, alloc_start,
3004 alloc_end - alloc_start,
3006 offset + len, &alloc_hint);
3007 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
3009 /* btrfs_prealloc_file_range releases reserved space on error */
3011 space_reserved = false;
3015 ret = btrfs_fallocate_update_isize(inode, offset + len, mode);
3017 if (ret && space_reserved)
3018 btrfs_free_reserved_data_space(BTRFS_I(inode), data_reserved,
3019 alloc_start, bytes_to_reserve);
3020 extent_changeset_free(data_reserved);
3025 static long btrfs_fallocate(struct file *file, int mode,
3026 loff_t offset, loff_t len)
3028 struct inode *inode = file_inode(file);
3029 struct extent_state *cached_state = NULL;
3030 struct extent_changeset *data_reserved = NULL;
3031 struct falloc_range *range;
3032 struct falloc_range *tmp;
3033 struct list_head reserve_list;
3041 u64 data_space_needed = 0;
3042 u64 data_space_reserved = 0;
3043 u64 qgroup_reserved = 0;
3044 struct extent_map *em;
3045 int blocksize = BTRFS_I(inode)->root->fs_info->sectorsize;
3048 /* Do not allow fallocate in ZONED mode */
3049 if (btrfs_is_zoned(btrfs_sb(inode->i_sb)))
3052 alloc_start = round_down(offset, blocksize);
3053 alloc_end = round_up(offset + len, blocksize);
3054 cur_offset = alloc_start;
3056 /* Make sure we aren't being give some crap mode */
3057 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
3058 FALLOC_FL_ZERO_RANGE))
3061 if (mode & FALLOC_FL_PUNCH_HOLE)
3062 return btrfs_punch_hole(file, offset, len);
3064 btrfs_inode_lock(inode, BTRFS_ILOCK_MMAP);
3066 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) {
3067 ret = inode_newsize_ok(inode, offset + len);
3072 ret = file_modified(file);
3077 * TODO: Move these two operations after we have checked
3078 * accurate reserved space, or fallocate can still fail but
3079 * with page truncated or size expanded.
3081 * But that's a minor problem and won't do much harm BTW.
3083 if (alloc_start > inode->i_size) {
3084 ret = btrfs_cont_expand(BTRFS_I(inode), i_size_read(inode),
3088 } else if (offset + len > inode->i_size) {
3090 * If we are fallocating from the end of the file onward we
3091 * need to zero out the end of the block if i_size lands in the
3092 * middle of a block.
3094 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
3100 * We have locked the inode at the VFS level (in exclusive mode) and we
3101 * have locked the i_mmap_lock lock (in exclusive mode). Now before
3102 * locking the file range, flush all dealloc in the range and wait for
3103 * all ordered extents in the range to complete. After this we can lock
3104 * the file range and, due to the previous locking we did, we know there
3105 * can't be more delalloc or ordered extents in the range.
3107 ret = btrfs_wait_ordered_range(inode, alloc_start,
3108 alloc_end - alloc_start);
3112 if (mode & FALLOC_FL_ZERO_RANGE) {
3113 ret = btrfs_zero_range(inode, offset, len, mode);
3114 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
3118 locked_end = alloc_end - 1;
3119 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
3122 btrfs_assert_inode_range_clean(BTRFS_I(inode), alloc_start, locked_end);
3124 /* First, check if we exceed the qgroup limit */
3125 INIT_LIST_HEAD(&reserve_list);
3126 while (cur_offset < alloc_end) {
3127 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
3128 alloc_end - cur_offset);
3133 last_byte = min(extent_map_end(em), alloc_end);
3134 actual_end = min_t(u64, extent_map_end(em), offset + len);
3135 last_byte = ALIGN(last_byte, blocksize);
3136 if (em->block_start == EXTENT_MAP_HOLE ||
3137 (cur_offset >= inode->i_size &&
3138 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
3139 const u64 range_len = last_byte - cur_offset;
3141 ret = add_falloc_range(&reserve_list, cur_offset, range_len);
3143 free_extent_map(em);
3146 ret = btrfs_qgroup_reserve_data(BTRFS_I(inode),
3147 &data_reserved, cur_offset, range_len);
3149 free_extent_map(em);
3152 qgroup_reserved += range_len;
3153 data_space_needed += range_len;
3155 free_extent_map(em);
3156 cur_offset = last_byte;
3159 if (!ret && data_space_needed > 0) {
3161 * We are safe to reserve space here as we can't have delalloc
3162 * in the range, see above.
3164 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
3167 data_space_reserved = data_space_needed;
3171 * If ret is still 0, means we're OK to fallocate.
3172 * Or just cleanup the list and exit.
3174 list_for_each_entry_safe(range, tmp, &reserve_list, list) {
3176 ret = btrfs_prealloc_file_range(inode, mode,
3178 range->len, i_blocksize(inode),
3179 offset + len, &alloc_hint);
3181 * btrfs_prealloc_file_range() releases space even
3182 * if it returns an error.
3184 data_space_reserved -= range->len;
3185 qgroup_reserved -= range->len;
3186 } else if (data_space_reserved > 0) {
3187 btrfs_free_reserved_data_space(BTRFS_I(inode),
3188 data_reserved, range->start,
3190 data_space_reserved -= range->len;
3191 qgroup_reserved -= range->len;
3192 } else if (qgroup_reserved > 0) {
3193 btrfs_qgroup_free_data(BTRFS_I(inode), data_reserved,
3194 range->start, range->len, NULL);
3195 qgroup_reserved -= range->len;
3197 list_del(&range->list);
3204 * We didn't need to allocate any more space, but we still extended the
3205 * size of the file so we need to update i_size and the inode item.
3207 ret = btrfs_fallocate_update_isize(inode, actual_end, mode);
3209 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
3212 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
3213 extent_changeset_free(data_reserved);
3218 * Helper for btrfs_find_delalloc_in_range(). Find a subrange in a given range
3219 * that has unflushed and/or flushing delalloc. There might be other adjacent
3220 * subranges after the one it found, so btrfs_find_delalloc_in_range() keeps
3221 * looping while it gets adjacent subranges, and merging them together.
3223 static bool find_delalloc_subrange(struct btrfs_inode *inode, u64 start, u64 end,
3224 u64 *delalloc_start_ret, u64 *delalloc_end_ret)
3226 const u64 len = end + 1 - start;
3227 struct extent_map_tree *em_tree = &inode->extent_tree;
3228 struct extent_map *em;
3233 * Search the io tree first for EXTENT_DELALLOC. If we find any, it
3234 * means we have delalloc (dirty pages) for which writeback has not
3237 *delalloc_start_ret = start;
3238 delalloc_len = count_range_bits(&inode->io_tree, delalloc_start_ret, end,
3239 len, EXTENT_DELALLOC, 1);
3241 * If delalloc was found then *delalloc_start_ret has a sector size
3242 * aligned value (rounded down).
3244 if (delalloc_len > 0)
3245 *delalloc_end_ret = *delalloc_start_ret + delalloc_len - 1;
3248 * Now also check if there's any extent map in the range that does not
3249 * map to a hole or prealloc extent. We do this because:
3251 * 1) When delalloc is flushed, the file range is locked, we clear the
3252 * EXTENT_DELALLOC bit from the io tree and create an extent map for
3253 * an allocated extent. So we might just have been called after
3254 * delalloc is flushed and before the ordered extent completes and
3255 * inserts the new file extent item in the subvolume's btree;
3257 * 2) We may have an extent map created by flushing delalloc for a
3258 * subrange that starts before the subrange we found marked with
3259 * EXTENT_DELALLOC in the io tree.
3261 read_lock(&em_tree->lock);
3262 em = lookup_extent_mapping(em_tree, start, len);
3264 read_unlock(&em_tree->lock);
3265 return (delalloc_len > 0);
3268 /* extent_map_end() returns a non-inclusive end offset. */
3269 em_end = extent_map_end(em);
3272 * If we have a hole/prealloc extent map, check the next one if this one
3273 * ends before our range's end.
3275 if ((em->block_start == EXTENT_MAP_HOLE ||
3276 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) && em_end < end) {
3277 struct extent_map *next_em;
3279 next_em = btrfs_next_extent_map(em_tree, em);
3280 free_extent_map(em);
3283 * There's no next extent map or the next one starts beyond our
3284 * range, return the range found in the io tree (if any).
3286 if (!next_em || next_em->start > end) {
3287 read_unlock(&em_tree->lock);
3288 free_extent_map(next_em);
3289 return (delalloc_len > 0);
3292 em_end = extent_map_end(next_em);
3296 read_unlock(&em_tree->lock);
3299 * We have a hole or prealloc extent that ends at or beyond our range's
3300 * end, return the range found in the io tree (if any).
3302 if (em->block_start == EXTENT_MAP_HOLE ||
3303 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3304 free_extent_map(em);
3305 return (delalloc_len > 0);
3309 * We don't have any range as EXTENT_DELALLOC in the io tree, so the
3310 * extent map is the only subrange representing delalloc.
3312 if (delalloc_len == 0) {
3313 *delalloc_start_ret = em->start;
3314 *delalloc_end_ret = min(end, em_end - 1);
3315 free_extent_map(em);
3320 * The extent map represents a delalloc range that starts before the
3321 * delalloc range we found in the io tree.
3323 if (em->start < *delalloc_start_ret) {
3324 *delalloc_start_ret = em->start;
3326 * If the ranges are adjacent, return a combined range.
3327 * Otherwise return the extent map's range.
3329 if (em_end < *delalloc_start_ret)
3330 *delalloc_end_ret = min(end, em_end - 1);
3332 free_extent_map(em);
3337 * The extent map starts after the delalloc range we found in the io
3338 * tree. If it's adjacent, return a combined range, otherwise return
3339 * the range found in the io tree.
3341 if (*delalloc_end_ret + 1 == em->start)
3342 *delalloc_end_ret = min(end, em_end - 1);
3344 free_extent_map(em);
3349 * Check if there's delalloc in a given range.
3351 * @inode: The inode.
3352 * @start: The start offset of the range. It does not need to be
3353 * sector size aligned.
3354 * @end: The end offset (inclusive value) of the search range.
3355 * It does not need to be sector size aligned.
3356 * @delalloc_start_ret: Output argument, set to the start offset of the
3357 * subrange found with delalloc (may not be sector size
3359 * @delalloc_end_ret: Output argument, set to he end offset (inclusive value)
3360 * of the subrange found with delalloc.
3362 * Returns true if a subrange with delalloc is found within the given range, and
3363 * if so it sets @delalloc_start_ret and @delalloc_end_ret with the start and
3364 * end offsets of the subrange.
3366 bool btrfs_find_delalloc_in_range(struct btrfs_inode *inode, u64 start, u64 end,
3367 u64 *delalloc_start_ret, u64 *delalloc_end_ret)
3369 u64 cur_offset = round_down(start, inode->root->fs_info->sectorsize);
3370 u64 prev_delalloc_end = 0;
3373 while (cur_offset <= end) {
3378 delalloc = find_delalloc_subrange(inode, cur_offset, end,
3384 if (prev_delalloc_end == 0) {
3385 /* First subrange found. */
3386 *delalloc_start_ret = max(delalloc_start, start);
3387 *delalloc_end_ret = delalloc_end;
3389 } else if (delalloc_start == prev_delalloc_end + 1) {
3390 /* Subrange adjacent to the previous one, merge them. */
3391 *delalloc_end_ret = delalloc_end;
3393 /* Subrange not adjacent to the previous one, exit. */
3397 prev_delalloc_end = delalloc_end;
3398 cur_offset = delalloc_end + 1;
3406 * Check if there's a hole or delalloc range in a range representing a hole (or
3407 * prealloc extent) found in the inode's subvolume btree.
3409 * @inode: The inode.
3410 * @whence: Seek mode (SEEK_DATA or SEEK_HOLE).
3411 * @start: Start offset of the hole region. It does not need to be sector
3413 * @end: End offset (inclusive value) of the hole region. It does not
3414 * need to be sector size aligned.
3415 * @start_ret: Return parameter, used to set the start of the subrange in the
3416 * hole that matches the search criteria (seek mode), if such
3417 * subrange is found (return value of the function is true).
3418 * The value returned here may not be sector size aligned.
3420 * Returns true if a subrange matching the given seek mode is found, and if one
3421 * is found, it updates @start_ret with the start of the subrange.
3423 static bool find_desired_extent_in_hole(struct btrfs_inode *inode, int whence,
3424 u64 start, u64 end, u64 *start_ret)
3430 delalloc = btrfs_find_delalloc_in_range(inode, start, end,
3431 &delalloc_start, &delalloc_end);
3432 if (delalloc && whence == SEEK_DATA) {
3433 *start_ret = delalloc_start;
3437 if (delalloc && whence == SEEK_HOLE) {
3439 * We found delalloc but it starts after out start offset. So we
3440 * have a hole between our start offset and the delalloc start.
3442 if (start < delalloc_start) {
3447 * Delalloc range starts at our start offset.
3448 * If the delalloc range's length is smaller than our range,
3449 * then it means we have a hole that starts where the delalloc
3452 if (delalloc_end < end) {
3453 *start_ret = delalloc_end + 1;
3457 /* There's delalloc for the whole range. */
3461 if (!delalloc && whence == SEEK_HOLE) {
3467 * No delalloc in the range and we are seeking for data. The caller has
3468 * to iterate to the next extent item in the subvolume btree.
3473 static loff_t find_desired_extent(struct btrfs_inode *inode, loff_t offset,
3476 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3477 struct extent_state *cached_state = NULL;
3478 const loff_t i_size = i_size_read(&inode->vfs_inode);
3479 const u64 ino = btrfs_ino(inode);
3480 struct btrfs_root *root = inode->root;
3481 struct btrfs_path *path;
3482 struct btrfs_key key;
3483 u64 last_extent_end;
3490 if (i_size == 0 || offset >= i_size)
3494 * Quick path. If the inode has no prealloc extents and its number of
3495 * bytes used matches its i_size, then it can not have holes.
3497 if (whence == SEEK_HOLE &&
3498 !(inode->flags & BTRFS_INODE_PREALLOC) &&
3499 inode_get_bytes(&inode->vfs_inode) == i_size)
3503 * offset can be negative, in this case we start finding DATA/HOLE from
3504 * the very start of the file.
3506 start = max_t(loff_t, 0, offset);
3508 lockstart = round_down(start, fs_info->sectorsize);
3509 lockend = round_up(i_size, fs_info->sectorsize);
3510 if (lockend <= lockstart)
3511 lockend = lockstart + fs_info->sectorsize;
3514 path = btrfs_alloc_path();
3517 path->reada = READA_FORWARD;
3520 key.type = BTRFS_EXTENT_DATA_KEY;
3523 last_extent_end = lockstart;
3525 lock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
3527 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3530 } else if (ret > 0 && path->slots[0] > 0) {
3531 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
3532 if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY)
3536 while (start < i_size) {
3537 struct extent_buffer *leaf = path->nodes[0];
3538 struct btrfs_file_extent_item *extent;
3542 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3543 ret = btrfs_next_leaf(root, path);
3549 leaf = path->nodes[0];
3552 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3553 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
3556 extent_end = btrfs_file_extent_end(path);
3559 * In the first iteration we may have a slot that points to an
3560 * extent that ends before our start offset, so skip it.
3562 if (extent_end <= start) {
3567 /* We have an implicit hole, NO_HOLES feature is likely set. */
3568 if (last_extent_end < key.offset) {
3569 u64 search_start = last_extent_end;
3573 * First iteration, @start matches @offset and it's
3576 if (start == offset)
3577 search_start = offset;
3579 found = find_desired_extent_in_hole(inode, whence,
3584 start = found_start;
3588 * Didn't find data or a hole (due to delalloc) in the
3589 * implicit hole range, so need to analyze the extent.
3593 extent = btrfs_item_ptr(leaf, path->slots[0],
3594 struct btrfs_file_extent_item);
3595 type = btrfs_file_extent_type(leaf, extent);
3598 * Can't access the extent's disk_bytenr field if this is an
3599 * inline extent, since at that offset, it's where the extent
3602 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
3603 (type == BTRFS_FILE_EXTENT_REG &&
3604 btrfs_file_extent_disk_bytenr(leaf, extent) == 0)) {
3606 * Explicit hole or prealloc extent, search for delalloc.
3607 * A prealloc extent is treated like a hole.
3609 u64 search_start = key.offset;
3613 * First iteration, @start matches @offset and it's
3616 if (start == offset)
3617 search_start = offset;
3619 found = find_desired_extent_in_hole(inode, whence,
3624 start = found_start;
3628 * Didn't find data or a hole (due to delalloc) in the
3629 * implicit hole range, so need to analyze the next
3634 * Found a regular or inline extent.
3635 * If we are seeking for data, adjust the start offset
3636 * and stop, we're done.
3638 if (whence == SEEK_DATA) {
3639 start = max_t(u64, key.offset, offset);
3644 * Else, we are seeking for a hole, check the next file
3650 last_extent_end = extent_end;
3652 if (fatal_signal_pending(current)) {
3659 /* We have an implicit hole from the last extent found up to i_size. */
3660 if (!found && start < i_size) {
3661 found = find_desired_extent_in_hole(inode, whence, start,
3662 i_size - 1, &start);
3668 unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
3669 btrfs_free_path(path);
3674 if (whence == SEEK_DATA && start >= i_size)
3677 return min_t(loff_t, start, i_size);
3680 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
3682 struct inode *inode = file->f_mapping->host;
3686 return generic_file_llseek(file, offset, whence);
3689 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
3690 offset = find_desired_extent(BTRFS_I(inode), offset, whence);
3691 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
3698 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
3701 static int btrfs_file_open(struct inode *inode, struct file *filp)
3705 filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC | FMODE_BUF_WASYNC;
3707 ret = fsverity_file_open(inode, filp);
3710 return generic_file_open(inode, filp);
3713 static int check_direct_read(struct btrfs_fs_info *fs_info,
3714 const struct iov_iter *iter, loff_t offset)
3719 ret = check_direct_IO(fs_info, iter, offset);
3723 if (!iter_is_iovec(iter))
3726 for (seg = 0; seg < iter->nr_segs; seg++)
3727 for (i = seg + 1; i < iter->nr_segs; i++)
3728 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
3733 static ssize_t btrfs_direct_read(struct kiocb *iocb, struct iov_iter *to)
3735 struct inode *inode = file_inode(iocb->ki_filp);
3736 size_t prev_left = 0;
3740 if (fsverity_active(inode))
3743 if (check_direct_read(btrfs_sb(inode->i_sb), to, iocb->ki_pos))
3746 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
3749 * This is similar to what we do for direct IO writes, see the comment
3750 * at btrfs_direct_write(), but we also disable page faults in addition
3751 * to disabling them only at the iov_iter level. This is because when
3752 * reading from a hole or prealloc extent, iomap calls iov_iter_zero(),
3753 * which can still trigger page fault ins despite having set ->nofault
3754 * to true of our 'to' iov_iter.
3756 * The difference to direct IO writes is that we deadlock when trying
3757 * to lock the extent range in the inode's tree during he page reads
3758 * triggered by the fault in (while for writes it is due to waiting for
3759 * our own ordered extent). This is because for direct IO reads,
3760 * btrfs_dio_iomap_begin() returns with the extent range locked, which
3761 * is only unlocked in the endio callback (end_bio_extent_readpage()).
3763 pagefault_disable();
3765 ret = btrfs_dio_read(iocb, to, read);
3766 to->nofault = false;
3769 /* No increment (+=) because iomap returns a cumulative value. */
3773 if (iov_iter_count(to) > 0 && (ret == -EFAULT || ret > 0)) {
3774 const size_t left = iov_iter_count(to);
3776 if (left == prev_left) {
3778 * We didn't make any progress since the last attempt,
3779 * fallback to a buffered read for the remainder of the
3780 * range. This is just to avoid any possibility of looping
3786 * We made some progress since the last retry or this is
3787 * the first time we are retrying. Fault in as many pages
3788 * as possible and retry.
3790 fault_in_iov_iter_writeable(to, left);
3795 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
3796 return ret < 0 ? ret : read;
3799 static ssize_t btrfs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
3803 if (iocb->ki_flags & IOCB_DIRECT) {
3804 ret = btrfs_direct_read(iocb, to);
3805 if (ret < 0 || !iov_iter_count(to) ||
3806 iocb->ki_pos >= i_size_read(file_inode(iocb->ki_filp)))
3810 return filemap_read(iocb, to, ret);
3813 const struct file_operations btrfs_file_operations = {
3814 .llseek = btrfs_file_llseek,
3815 .read_iter = btrfs_file_read_iter,
3816 .splice_read = generic_file_splice_read,
3817 .write_iter = btrfs_file_write_iter,
3818 .splice_write = iter_file_splice_write,
3819 .mmap = btrfs_file_mmap,
3820 .open = btrfs_file_open,
3821 .release = btrfs_release_file,
3822 .get_unmapped_area = thp_get_unmapped_area,
3823 .fsync = btrfs_sync_file,
3824 .fallocate = btrfs_fallocate,
3825 .unlocked_ioctl = btrfs_ioctl,
3826 #ifdef CONFIG_COMPAT
3827 .compat_ioctl = btrfs_compat_ioctl,
3829 .remap_file_range = btrfs_remap_file_range,
3832 int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end)
3837 * So with compression we will find and lock a dirty page and clear the
3838 * first one as dirty, setup an async extent, and immediately return
3839 * with the entire range locked but with nobody actually marked with
3840 * writeback. So we can't just filemap_write_and_wait_range() and
3841 * expect it to work since it will just kick off a thread to do the
3842 * actual work. So we need to call filemap_fdatawrite_range _again_
3843 * since it will wait on the page lock, which won't be unlocked until
3844 * after the pages have been marked as writeback and so we're good to go
3845 * from there. We have to do this otherwise we'll miss the ordered
3846 * extents and that results in badness. Please Josef, do not think you
3847 * know better and pull this out at some point in the future, it is
3848 * right and you are wrong.
3850 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
3851 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
3852 &BTRFS_I(inode)->runtime_flags))
3853 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);