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 #include "accessors.h"
35 #include "extent-tree.h"
36 #include "file-item.h"
41 /* simple helper to fault in pages and copy. This should go away
42 * and be replaced with calls into generic code.
44 static noinline int btrfs_copy_from_user(loff_t pos, size_t write_bytes,
45 struct page **prepared_pages,
49 size_t total_copied = 0;
51 int offset = offset_in_page(pos);
53 while (write_bytes > 0) {
54 size_t count = min_t(size_t,
55 PAGE_SIZE - offset, write_bytes);
56 struct page *page = prepared_pages[pg];
58 * Copy data from userspace to the current page
60 copied = copy_page_from_iter_atomic(page, offset, count, i);
62 /* Flush processor's dcache for this page */
63 flush_dcache_page(page);
66 * if we get a partial write, we can end up with
67 * partially up to date pages. These add
68 * a lot of complexity, so make sure they don't
69 * happen by forcing this copy to be retried.
71 * The rest of the btrfs_file_write code will fall
72 * back to page at a time copies after we return 0.
74 if (unlikely(copied < count)) {
75 if (!PageUptodate(page)) {
76 iov_iter_revert(i, copied);
83 write_bytes -= copied;
84 total_copied += copied;
86 if (offset == PAGE_SIZE) {
95 * unlocks pages after btrfs_file_write is done with them
97 static void btrfs_drop_pages(struct btrfs_fs_info *fs_info,
98 struct page **pages, size_t num_pages,
102 u64 block_start = round_down(pos, fs_info->sectorsize);
103 u64 block_len = round_up(pos + copied, fs_info->sectorsize) - block_start;
105 ASSERT(block_len <= U32_MAX);
106 for (i = 0; i < num_pages; i++) {
107 /* page checked is some magic around finding pages that
108 * have been modified without going through btrfs_set_page_dirty
109 * clear it here. There should be no need to mark the pages
110 * accessed as prepare_pages should have marked them accessed
111 * in prepare_pages via find_or_create_page()
113 btrfs_page_clamp_clear_checked(fs_info, pages[i], block_start,
115 unlock_page(pages[i]);
121 * After btrfs_copy_from_user(), update the following things for delalloc:
122 * - Mark newly dirtied pages as DELALLOC in the io tree.
123 * Used to advise which range is to be written back.
124 * - Mark modified pages as Uptodate/Dirty and not needing COW fixup
125 * - Update inode size for past EOF write
127 int btrfs_dirty_pages(struct btrfs_inode *inode, struct page **pages,
128 size_t num_pages, loff_t pos, size_t write_bytes,
129 struct extent_state **cached, bool noreserve)
131 struct btrfs_fs_info *fs_info = inode->root->fs_info;
136 u64 end_of_last_block;
137 u64 end_pos = pos + write_bytes;
138 loff_t isize = i_size_read(&inode->vfs_inode);
139 unsigned int extra_bits = 0;
141 if (write_bytes == 0)
145 extra_bits |= EXTENT_NORESERVE;
147 start_pos = round_down(pos, fs_info->sectorsize);
148 num_bytes = round_up(write_bytes + pos - start_pos,
149 fs_info->sectorsize);
150 ASSERT(num_bytes <= U32_MAX);
152 end_of_last_block = start_pos + num_bytes - 1;
155 * The pages may have already been dirty, clear out old accounting so
156 * we can set things up properly
158 clear_extent_bit(&inode->io_tree, start_pos, end_of_last_block,
159 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
162 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
167 for (i = 0; i < num_pages; i++) {
168 struct page *p = pages[i];
170 btrfs_page_clamp_set_uptodate(fs_info, p, start_pos, num_bytes);
171 btrfs_page_clamp_clear_checked(fs_info, p, start_pos, num_bytes);
172 btrfs_page_clamp_set_dirty(fs_info, p, start_pos, num_bytes);
176 * we've only changed i_size in ram, and we haven't updated
177 * the disk i_size. There is no need to log the inode
181 i_size_write(&inode->vfs_inode, end_pos);
186 * this is very complex, but the basic idea is to drop all extents
187 * in the range start - end. hint_block is filled in with a block number
188 * that would be a good hint to the block allocator for this file.
190 * If an extent intersects the range but is not entirely inside the range
191 * it is either truncated or split. Anything entirely inside the range
192 * is deleted from the tree.
194 * Note: the VFS' inode number of bytes is not updated, it's up to the caller
195 * to deal with that. We set the field 'bytes_found' of the arguments structure
196 * with the number of allocated bytes found in the target range, so that the
197 * caller can update the inode's number of bytes in an atomic way when
198 * replacing extents in a range to avoid races with stat(2).
200 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
201 struct btrfs_root *root, struct btrfs_inode *inode,
202 struct btrfs_drop_extents_args *args)
204 struct btrfs_fs_info *fs_info = root->fs_info;
205 struct extent_buffer *leaf;
206 struct btrfs_file_extent_item *fi;
207 struct btrfs_ref ref = { 0 };
208 struct btrfs_key key;
209 struct btrfs_key new_key;
210 u64 ino = btrfs_ino(inode);
211 u64 search_start = args->start;
214 u64 extent_offset = 0;
216 u64 last_end = args->start;
222 int modify_tree = -1;
225 struct btrfs_path *path = args->path;
227 args->bytes_found = 0;
228 args->extent_inserted = false;
230 /* Must always have a path if ->replace_extent is true */
231 ASSERT(!(args->replace_extent && !args->path));
234 path = btrfs_alloc_path();
241 if (args->drop_cache)
242 btrfs_drop_extent_map_range(inode, args->start, args->end - 1, false);
244 if (args->start >= inode->disk_i_size && !args->replace_extent)
247 update_refs = (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID);
250 ret = btrfs_lookup_file_extent(trans, root, path, ino,
251 search_start, modify_tree);
254 if (ret > 0 && path->slots[0] > 0 && search_start == args->start) {
255 leaf = path->nodes[0];
256 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
257 if (key.objectid == ino &&
258 key.type == BTRFS_EXTENT_DATA_KEY)
263 leaf = path->nodes[0];
264 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
266 ret = btrfs_next_leaf(root, path);
273 leaf = path->nodes[0];
277 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
279 if (key.objectid > ino)
281 if (WARN_ON_ONCE(key.objectid < ino) ||
282 key.type < BTRFS_EXTENT_DATA_KEY) {
287 if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= args->end)
290 fi = btrfs_item_ptr(leaf, path->slots[0],
291 struct btrfs_file_extent_item);
292 extent_type = btrfs_file_extent_type(leaf, fi);
294 if (extent_type == BTRFS_FILE_EXTENT_REG ||
295 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
296 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
297 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
298 extent_offset = btrfs_file_extent_offset(leaf, fi);
299 extent_end = key.offset +
300 btrfs_file_extent_num_bytes(leaf, fi);
301 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
302 extent_end = key.offset +
303 btrfs_file_extent_ram_bytes(leaf, fi);
310 * Don't skip extent items representing 0 byte lengths. They
311 * used to be created (bug) if while punching holes we hit
312 * -ENOSPC condition. So if we find one here, just ensure we
313 * delete it, otherwise we would insert a new file extent item
314 * with the same key (offset) as that 0 bytes length file
315 * extent item in the call to setup_items_for_insert() later
318 if (extent_end == key.offset && extent_end >= search_start) {
319 last_end = extent_end;
320 goto delete_extent_item;
323 if (extent_end <= search_start) {
329 search_start = max(key.offset, args->start);
330 if (recow || !modify_tree) {
332 btrfs_release_path(path);
337 * | - range to drop - |
338 * | -------- extent -------- |
340 if (args->start > key.offset && args->end < extent_end) {
342 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
347 memcpy(&new_key, &key, sizeof(new_key));
348 new_key.offset = args->start;
349 ret = btrfs_duplicate_item(trans, root, path,
351 if (ret == -EAGAIN) {
352 btrfs_release_path(path);
358 leaf = path->nodes[0];
359 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
360 struct btrfs_file_extent_item);
361 btrfs_set_file_extent_num_bytes(leaf, fi,
362 args->start - key.offset);
364 fi = btrfs_item_ptr(leaf, path->slots[0],
365 struct btrfs_file_extent_item);
367 extent_offset += args->start - key.offset;
368 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
369 btrfs_set_file_extent_num_bytes(leaf, fi,
370 extent_end - args->start);
371 btrfs_mark_buffer_dirty(leaf);
373 if (update_refs && disk_bytenr > 0) {
374 btrfs_init_generic_ref(&ref,
375 BTRFS_ADD_DELAYED_REF,
376 disk_bytenr, num_bytes, 0);
377 btrfs_init_data_ref(&ref,
378 root->root_key.objectid,
380 args->start - extent_offset,
382 ret = btrfs_inc_extent_ref(trans, &ref);
384 btrfs_abort_transaction(trans, ret);
388 key.offset = args->start;
391 * From here on out we will have actually dropped something, so
392 * last_end can be updated.
394 last_end = extent_end;
397 * | ---- range to drop ----- |
398 * | -------- extent -------- |
400 if (args->start <= key.offset && args->end < extent_end) {
401 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
406 memcpy(&new_key, &key, sizeof(new_key));
407 new_key.offset = args->end;
408 btrfs_set_item_key_safe(fs_info, path, &new_key);
410 extent_offset += args->end - key.offset;
411 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
412 btrfs_set_file_extent_num_bytes(leaf, fi,
413 extent_end - args->end);
414 btrfs_mark_buffer_dirty(leaf);
415 if (update_refs && disk_bytenr > 0)
416 args->bytes_found += args->end - key.offset;
420 search_start = extent_end;
422 * | ---- range to drop ----- |
423 * | -------- extent -------- |
425 if (args->start > key.offset && args->end >= extent_end) {
427 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
432 btrfs_set_file_extent_num_bytes(leaf, fi,
433 args->start - key.offset);
434 btrfs_mark_buffer_dirty(leaf);
435 if (update_refs && disk_bytenr > 0)
436 args->bytes_found += extent_end - args->start;
437 if (args->end == extent_end)
445 * | ---- range to drop ----- |
446 * | ------ extent ------ |
448 if (args->start <= key.offset && args->end >= extent_end) {
451 del_slot = path->slots[0];
454 BUG_ON(del_slot + del_nr != path->slots[0]);
459 extent_type == BTRFS_FILE_EXTENT_INLINE) {
460 args->bytes_found += extent_end - key.offset;
461 extent_end = ALIGN(extent_end,
462 fs_info->sectorsize);
463 } else if (update_refs && disk_bytenr > 0) {
464 btrfs_init_generic_ref(&ref,
465 BTRFS_DROP_DELAYED_REF,
466 disk_bytenr, num_bytes, 0);
467 btrfs_init_data_ref(&ref,
468 root->root_key.objectid,
470 key.offset - extent_offset, 0,
472 ret = btrfs_free_extent(trans, &ref);
474 btrfs_abort_transaction(trans, ret);
477 args->bytes_found += extent_end - key.offset;
480 if (args->end == extent_end)
483 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
488 ret = btrfs_del_items(trans, root, path, del_slot,
491 btrfs_abort_transaction(trans, ret);
498 btrfs_release_path(path);
505 if (!ret && del_nr > 0) {
507 * Set path->slots[0] to first slot, so that after the delete
508 * if items are move off from our leaf to its immediate left or
509 * right neighbor leafs, we end up with a correct and adjusted
510 * path->slots[0] for our insertion (if args->replace_extent).
512 path->slots[0] = del_slot;
513 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
515 btrfs_abort_transaction(trans, ret);
518 leaf = path->nodes[0];
520 * If btrfs_del_items() was called, it might have deleted a leaf, in
521 * which case it unlocked our path, so check path->locks[0] matches a
524 if (!ret && args->replace_extent &&
525 path->locks[0] == BTRFS_WRITE_LOCK &&
526 btrfs_leaf_free_space(leaf) >=
527 sizeof(struct btrfs_item) + args->extent_item_size) {
530 key.type = BTRFS_EXTENT_DATA_KEY;
531 key.offset = args->start;
532 if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
533 struct btrfs_key slot_key;
535 btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
536 if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
539 btrfs_setup_item_for_insert(root, path, &key, args->extent_item_size);
540 args->extent_inserted = true;
544 btrfs_free_path(path);
545 else if (!args->extent_inserted)
546 btrfs_release_path(path);
548 args->drop_end = found ? min(args->end, last_end) : args->end;
553 static int extent_mergeable(struct extent_buffer *leaf, int slot,
554 u64 objectid, u64 bytenr, u64 orig_offset,
555 u64 *start, u64 *end)
557 struct btrfs_file_extent_item *fi;
558 struct btrfs_key key;
561 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
564 btrfs_item_key_to_cpu(leaf, &key, slot);
565 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
568 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
569 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
570 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
571 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
572 btrfs_file_extent_compression(leaf, fi) ||
573 btrfs_file_extent_encryption(leaf, fi) ||
574 btrfs_file_extent_other_encoding(leaf, fi))
577 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
578 if ((*start && *start != key.offset) || (*end && *end != extent_end))
587 * Mark extent in the range start - end as written.
589 * This changes extent type from 'pre-allocated' to 'regular'. If only
590 * part of extent is marked as written, the extent will be split into
593 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
594 struct btrfs_inode *inode, u64 start, u64 end)
596 struct btrfs_fs_info *fs_info = trans->fs_info;
597 struct btrfs_root *root = inode->root;
598 struct extent_buffer *leaf;
599 struct btrfs_path *path;
600 struct btrfs_file_extent_item *fi;
601 struct btrfs_ref ref = { 0 };
602 struct btrfs_key key;
603 struct btrfs_key new_key;
615 u64 ino = btrfs_ino(inode);
617 path = btrfs_alloc_path();
624 key.type = BTRFS_EXTENT_DATA_KEY;
627 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
630 if (ret > 0 && path->slots[0] > 0)
633 leaf = path->nodes[0];
634 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
635 if (key.objectid != ino ||
636 key.type != BTRFS_EXTENT_DATA_KEY) {
638 btrfs_abort_transaction(trans, ret);
641 fi = btrfs_item_ptr(leaf, path->slots[0],
642 struct btrfs_file_extent_item);
643 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_PREALLOC) {
645 btrfs_abort_transaction(trans, ret);
648 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
649 if (key.offset > start || extent_end < end) {
651 btrfs_abort_transaction(trans, ret);
655 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
656 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
657 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
658 memcpy(&new_key, &key, sizeof(new_key));
660 if (start == key.offset && end < extent_end) {
663 if (extent_mergeable(leaf, path->slots[0] - 1,
664 ino, bytenr, orig_offset,
665 &other_start, &other_end)) {
666 new_key.offset = end;
667 btrfs_set_item_key_safe(fs_info, path, &new_key);
668 fi = btrfs_item_ptr(leaf, path->slots[0],
669 struct btrfs_file_extent_item);
670 btrfs_set_file_extent_generation(leaf, fi,
672 btrfs_set_file_extent_num_bytes(leaf, fi,
674 btrfs_set_file_extent_offset(leaf, fi,
676 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
677 struct btrfs_file_extent_item);
678 btrfs_set_file_extent_generation(leaf, fi,
680 btrfs_set_file_extent_num_bytes(leaf, fi,
682 btrfs_mark_buffer_dirty(leaf);
687 if (start > key.offset && end == extent_end) {
690 if (extent_mergeable(leaf, path->slots[0] + 1,
691 ino, bytenr, orig_offset,
692 &other_start, &other_end)) {
693 fi = btrfs_item_ptr(leaf, path->slots[0],
694 struct btrfs_file_extent_item);
695 btrfs_set_file_extent_num_bytes(leaf, fi,
697 btrfs_set_file_extent_generation(leaf, fi,
700 new_key.offset = start;
701 btrfs_set_item_key_safe(fs_info, path, &new_key);
703 fi = btrfs_item_ptr(leaf, path->slots[0],
704 struct btrfs_file_extent_item);
705 btrfs_set_file_extent_generation(leaf, fi,
707 btrfs_set_file_extent_num_bytes(leaf, fi,
709 btrfs_set_file_extent_offset(leaf, fi,
710 start - orig_offset);
711 btrfs_mark_buffer_dirty(leaf);
716 while (start > key.offset || end < extent_end) {
717 if (key.offset == start)
720 new_key.offset = split;
721 ret = btrfs_duplicate_item(trans, root, path, &new_key);
722 if (ret == -EAGAIN) {
723 btrfs_release_path(path);
727 btrfs_abort_transaction(trans, ret);
731 leaf = path->nodes[0];
732 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
733 struct btrfs_file_extent_item);
734 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
735 btrfs_set_file_extent_num_bytes(leaf, fi,
738 fi = btrfs_item_ptr(leaf, path->slots[0],
739 struct btrfs_file_extent_item);
741 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
742 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
743 btrfs_set_file_extent_num_bytes(leaf, fi,
745 btrfs_mark_buffer_dirty(leaf);
747 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, bytenr,
749 btrfs_init_data_ref(&ref, root->root_key.objectid, ino,
750 orig_offset, 0, false);
751 ret = btrfs_inc_extent_ref(trans, &ref);
753 btrfs_abort_transaction(trans, ret);
757 if (split == start) {
760 if (start != key.offset) {
762 btrfs_abort_transaction(trans, ret);
773 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
775 btrfs_init_data_ref(&ref, root->root_key.objectid, ino, orig_offset,
777 if (extent_mergeable(leaf, path->slots[0] + 1,
778 ino, bytenr, orig_offset,
779 &other_start, &other_end)) {
781 btrfs_release_path(path);
784 extent_end = other_end;
785 del_slot = path->slots[0] + 1;
787 ret = btrfs_free_extent(trans, &ref);
789 btrfs_abort_transaction(trans, ret);
795 if (extent_mergeable(leaf, path->slots[0] - 1,
796 ino, bytenr, orig_offset,
797 &other_start, &other_end)) {
799 btrfs_release_path(path);
802 key.offset = other_start;
803 del_slot = path->slots[0];
805 ret = btrfs_free_extent(trans, &ref);
807 btrfs_abort_transaction(trans, ret);
812 fi = btrfs_item_ptr(leaf, path->slots[0],
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_mark_buffer_dirty(leaf);
819 fi = btrfs_item_ptr(leaf, del_slot - 1,
820 struct btrfs_file_extent_item);
821 btrfs_set_file_extent_type(leaf, fi,
822 BTRFS_FILE_EXTENT_REG);
823 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
824 btrfs_set_file_extent_num_bytes(leaf, fi,
825 extent_end - key.offset);
826 btrfs_mark_buffer_dirty(leaf);
828 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
830 btrfs_abort_transaction(trans, ret);
835 btrfs_free_path(path);
840 * on error we return an unlocked page and the error value
841 * on success we return a locked page and 0
843 static int prepare_uptodate_page(struct inode *inode,
844 struct page *page, u64 pos,
847 struct folio *folio = page_folio(page);
850 if (((pos & (PAGE_SIZE - 1)) || force_uptodate) &&
851 !PageUptodate(page)) {
852 ret = btrfs_read_folio(NULL, folio);
856 if (!PageUptodate(page)) {
862 * Since btrfs_read_folio() will unlock the folio before it
863 * returns, there is a window where btrfs_release_folio() can be
864 * called to release the page. Here we check both inode
865 * mapping and PagePrivate() to make sure the page was not
868 * The private flag check is essential for subpage as we need
869 * to store extra bitmap using page->private.
871 if (page->mapping != inode->i_mapping || !PagePrivate(page)) {
879 static fgf_t get_prepare_fgp_flags(bool nowait)
881 fgf_t fgp_flags = FGP_LOCK | FGP_ACCESSED | FGP_CREAT;
884 fgp_flags |= FGP_NOWAIT;
889 static gfp_t get_prepare_gfp_flags(struct inode *inode, bool nowait)
893 gfp = btrfs_alloc_write_mask(inode->i_mapping);
895 gfp &= ~__GFP_DIRECT_RECLAIM;
903 * this just gets pages into the page cache and locks them down.
905 static noinline int prepare_pages(struct inode *inode, struct page **pages,
906 size_t num_pages, loff_t pos,
907 size_t write_bytes, bool force_uptodate,
911 unsigned long index = pos >> PAGE_SHIFT;
912 gfp_t mask = get_prepare_gfp_flags(inode, nowait);
913 fgf_t fgp_flags = get_prepare_fgp_flags(nowait);
917 for (i = 0; i < num_pages; i++) {
919 pages[i] = pagecache_get_page(inode->i_mapping, index + i,
920 fgp_flags, mask | __GFP_WRITE);
930 err = set_page_extent_mapped(pages[i]);
937 err = prepare_uptodate_page(inode, pages[i], pos,
939 if (!err && i == num_pages - 1)
940 err = prepare_uptodate_page(inode, pages[i],
941 pos + write_bytes, false);
944 if (!nowait && err == -EAGAIN) {
951 wait_on_page_writeback(pages[i]);
957 unlock_page(pages[faili]);
958 put_page(pages[faili]);
966 * This function locks the extent and properly waits for data=ordered extents
967 * to finish before allowing the pages to be modified if need.
970 * 1 - the extent is locked
971 * 0 - the extent is not locked, and everything is OK
972 * -EAGAIN - need re-prepare the pages
973 * the other < 0 number - Something wrong happens
976 lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct page **pages,
977 size_t num_pages, loff_t pos,
979 u64 *lockstart, u64 *lockend, bool nowait,
980 struct extent_state **cached_state)
982 struct btrfs_fs_info *fs_info = inode->root->fs_info;
988 start_pos = round_down(pos, fs_info->sectorsize);
989 last_pos = round_up(pos + write_bytes, fs_info->sectorsize) - 1;
991 if (start_pos < inode->vfs_inode.i_size) {
992 struct btrfs_ordered_extent *ordered;
995 if (!try_lock_extent(&inode->io_tree, start_pos, last_pos,
997 for (i = 0; i < num_pages; i++) {
998 unlock_page(pages[i]);
1006 lock_extent(&inode->io_tree, start_pos, last_pos, cached_state);
1009 ordered = btrfs_lookup_ordered_range(inode, start_pos,
1010 last_pos - start_pos + 1);
1012 ordered->file_offset + ordered->num_bytes > start_pos &&
1013 ordered->file_offset <= last_pos) {
1014 unlock_extent(&inode->io_tree, start_pos, last_pos,
1016 for (i = 0; i < num_pages; i++) {
1017 unlock_page(pages[i]);
1020 btrfs_start_ordered_extent(ordered);
1021 btrfs_put_ordered_extent(ordered);
1025 btrfs_put_ordered_extent(ordered);
1027 *lockstart = start_pos;
1028 *lockend = last_pos;
1033 * We should be called after prepare_pages() which should have locked
1034 * all pages in the range.
1036 for (i = 0; i < num_pages; i++)
1037 WARN_ON(!PageLocked(pages[i]));
1043 * Check if we can do nocow write into the range [@pos, @pos + @write_bytes)
1045 * @pos: File offset.
1046 * @write_bytes: The length to write, will be updated to the nocow writeable
1049 * This function will flush ordered extents in the range to ensure proper
1053 * > 0 If we can nocow, and updates @write_bytes.
1054 * 0 If we can't do a nocow write.
1055 * -EAGAIN If we can't do a nocow write because snapshoting of the inode's
1056 * root is in progress.
1057 * < 0 If an error happened.
1059 * NOTE: Callers need to call btrfs_check_nocow_unlock() if we return > 0.
1061 int btrfs_check_nocow_lock(struct btrfs_inode *inode, loff_t pos,
1062 size_t *write_bytes, bool nowait)
1064 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1065 struct btrfs_root *root = inode->root;
1066 struct extent_state *cached_state = NULL;
1067 u64 lockstart, lockend;
1071 if (!(inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
1074 if (!btrfs_drew_try_write_lock(&root->snapshot_lock))
1077 lockstart = round_down(pos, fs_info->sectorsize);
1078 lockend = round_up(pos + *write_bytes,
1079 fs_info->sectorsize) - 1;
1080 num_bytes = lockend - lockstart + 1;
1083 if (!btrfs_try_lock_ordered_range(inode, lockstart, lockend,
1085 btrfs_drew_write_unlock(&root->snapshot_lock);
1089 btrfs_lock_and_flush_ordered_range(inode, lockstart, lockend,
1092 ret = can_nocow_extent(&inode->vfs_inode, lockstart, &num_bytes,
1093 NULL, NULL, NULL, nowait, false);
1095 btrfs_drew_write_unlock(&root->snapshot_lock);
1097 *write_bytes = min_t(size_t, *write_bytes ,
1098 num_bytes - pos + lockstart);
1099 unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
1104 void btrfs_check_nocow_unlock(struct btrfs_inode *inode)
1106 btrfs_drew_write_unlock(&inode->root->snapshot_lock);
1109 static int btrfs_write_check(struct kiocb *iocb, struct iov_iter *from,
1112 struct file *file = iocb->ki_filp;
1113 struct inode *inode = file_inode(file);
1114 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1115 loff_t pos = iocb->ki_pos;
1121 * Quickly bail out on NOWAIT writes if we don't have the nodatacow or
1122 * prealloc flags, as without those flags we always have to COW. We will
1123 * later check if we can really COW into the target range (using
1124 * can_nocow_extent() at btrfs_get_blocks_direct_write()).
1126 if ((iocb->ki_flags & IOCB_NOWAIT) &&
1127 !(BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
1130 ret = file_remove_privs(file);
1135 * We reserve space for updating the inode when we reserve space for the
1136 * extent we are going to write, so we will enospc out there. We don't
1137 * need to start yet another transaction to update the inode as we will
1138 * update the inode when we finish writing whatever data we write.
1140 if (!IS_NOCMTIME(inode)) {
1141 inode->i_mtime = inode_set_ctime_current(inode);
1142 inode_inc_iversion(inode);
1145 start_pos = round_down(pos, fs_info->sectorsize);
1146 oldsize = i_size_read(inode);
1147 if (start_pos > oldsize) {
1148 /* Expand hole size to cover write data, preventing empty gap */
1149 loff_t end_pos = round_up(pos + count, fs_info->sectorsize);
1151 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, end_pos);
1159 static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb,
1162 struct file *file = iocb->ki_filp;
1164 struct inode *inode = file_inode(file);
1165 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1166 struct page **pages = NULL;
1167 struct extent_changeset *data_reserved = NULL;
1168 u64 release_bytes = 0;
1171 size_t num_written = 0;
1174 bool only_release_metadata = false;
1175 bool force_page_uptodate = false;
1176 loff_t old_isize = i_size_read(inode);
1177 unsigned int ilock_flags = 0;
1178 const bool nowait = (iocb->ki_flags & IOCB_NOWAIT);
1179 unsigned int bdp_flags = (nowait ? BDP_ASYNC : 0);
1182 ilock_flags |= BTRFS_ILOCK_TRY;
1184 ret = btrfs_inode_lock(BTRFS_I(inode), ilock_flags);
1188 ret = generic_write_checks(iocb, i);
1192 ret = btrfs_write_check(iocb, i, ret);
1197 nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_SIZE),
1198 PAGE_SIZE / (sizeof(struct page *)));
1199 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1200 nrptrs = max(nrptrs, 8);
1201 pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL);
1207 while (iov_iter_count(i) > 0) {
1208 struct extent_state *cached_state = NULL;
1209 size_t offset = offset_in_page(pos);
1210 size_t sector_offset;
1211 size_t write_bytes = min(iov_iter_count(i),
1212 nrptrs * (size_t)PAGE_SIZE -
1215 size_t reserve_bytes;
1218 size_t dirty_sectors;
1223 * Fault pages before locking them in prepare_pages
1224 * to avoid recursive lock
1226 if (unlikely(fault_in_iov_iter_readable(i, write_bytes))) {
1231 only_release_metadata = false;
1232 sector_offset = pos & (fs_info->sectorsize - 1);
1234 extent_changeset_release(data_reserved);
1235 ret = btrfs_check_data_free_space(BTRFS_I(inode),
1236 &data_reserved, pos,
1237 write_bytes, nowait);
1241 if (nowait && (ret == -ENOSPC || ret == -EAGAIN)) {
1247 * If we don't have to COW at the offset, reserve
1248 * metadata only. write_bytes may get smaller than
1251 can_nocow = btrfs_check_nocow_lock(BTRFS_I(inode), pos,
1252 &write_bytes, nowait);
1259 only_release_metadata = true;
1262 num_pages = DIV_ROUND_UP(write_bytes + offset, PAGE_SIZE);
1263 WARN_ON(num_pages > nrptrs);
1264 reserve_bytes = round_up(write_bytes + sector_offset,
1265 fs_info->sectorsize);
1266 WARN_ON(reserve_bytes == 0);
1267 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
1269 reserve_bytes, nowait);
1271 if (!only_release_metadata)
1272 btrfs_free_reserved_data_space(BTRFS_I(inode),
1276 btrfs_check_nocow_unlock(BTRFS_I(inode));
1278 if (nowait && ret == -ENOSPC)
1283 release_bytes = reserve_bytes;
1285 ret = balance_dirty_pages_ratelimited_flags(inode->i_mapping, bdp_flags);
1287 btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes);
1292 * This is going to setup the pages array with the number of
1293 * pages we want, so we don't really need to worry about the
1294 * contents of pages from loop to loop
1296 ret = prepare_pages(inode, pages, num_pages,
1297 pos, write_bytes, force_page_uptodate, false);
1299 btrfs_delalloc_release_extents(BTRFS_I(inode),
1304 extents_locked = lock_and_cleanup_extent_if_need(
1305 BTRFS_I(inode), pages,
1306 num_pages, pos, write_bytes, &lockstart,
1307 &lockend, nowait, &cached_state);
1308 if (extents_locked < 0) {
1309 if (!nowait && extents_locked == -EAGAIN)
1312 btrfs_delalloc_release_extents(BTRFS_I(inode),
1314 ret = extents_locked;
1318 copied = btrfs_copy_from_user(pos, write_bytes, pages, i);
1320 num_sectors = BTRFS_BYTES_TO_BLKS(fs_info, reserve_bytes);
1321 dirty_sectors = round_up(copied + sector_offset,
1322 fs_info->sectorsize);
1323 dirty_sectors = BTRFS_BYTES_TO_BLKS(fs_info, dirty_sectors);
1326 * if we have trouble faulting in the pages, fall
1327 * back to one page at a time
1329 if (copied < write_bytes)
1333 force_page_uptodate = true;
1337 force_page_uptodate = false;
1338 dirty_pages = DIV_ROUND_UP(copied + offset,
1342 if (num_sectors > dirty_sectors) {
1343 /* release everything except the sectors we dirtied */
1344 release_bytes -= dirty_sectors << fs_info->sectorsize_bits;
1345 if (only_release_metadata) {
1346 btrfs_delalloc_release_metadata(BTRFS_I(inode),
1347 release_bytes, true);
1351 __pos = round_down(pos,
1352 fs_info->sectorsize) +
1353 (dirty_pages << PAGE_SHIFT);
1354 btrfs_delalloc_release_space(BTRFS_I(inode),
1355 data_reserved, __pos,
1356 release_bytes, true);
1360 release_bytes = round_up(copied + sector_offset,
1361 fs_info->sectorsize);
1363 ret = btrfs_dirty_pages(BTRFS_I(inode), pages,
1364 dirty_pages, pos, copied,
1365 &cached_state, only_release_metadata);
1368 * If we have not locked the extent range, because the range's
1369 * start offset is >= i_size, we might still have a non-NULL
1370 * cached extent state, acquired while marking the extent range
1371 * as delalloc through btrfs_dirty_pages(). Therefore free any
1372 * possible cached extent state to avoid a memory leak.
1375 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart,
1376 lockend, &cached_state);
1378 free_extent_state(cached_state);
1380 btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes);
1382 btrfs_drop_pages(fs_info, pages, num_pages, pos, copied);
1387 if (only_release_metadata)
1388 btrfs_check_nocow_unlock(BTRFS_I(inode));
1390 btrfs_drop_pages(fs_info, pages, num_pages, pos, copied);
1395 num_written += copied;
1400 if (release_bytes) {
1401 if (only_release_metadata) {
1402 btrfs_check_nocow_unlock(BTRFS_I(inode));
1403 btrfs_delalloc_release_metadata(BTRFS_I(inode),
1404 release_bytes, true);
1406 btrfs_delalloc_release_space(BTRFS_I(inode),
1408 round_down(pos, fs_info->sectorsize),
1409 release_bytes, true);
1413 extent_changeset_free(data_reserved);
1414 if (num_written > 0) {
1415 pagecache_isize_extended(inode, old_isize, iocb->ki_pos);
1416 iocb->ki_pos += num_written;
1419 btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1420 return num_written ? num_written : ret;
1423 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
1424 const struct iov_iter *iter, loff_t offset)
1426 const u32 blocksize_mask = fs_info->sectorsize - 1;
1428 if (offset & blocksize_mask)
1431 if (iov_iter_alignment(iter) & blocksize_mask)
1437 static ssize_t btrfs_direct_write(struct kiocb *iocb, struct iov_iter *from)
1439 struct file *file = iocb->ki_filp;
1440 struct inode *inode = file_inode(file);
1441 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1443 ssize_t written = 0;
1444 ssize_t written_buffered;
1445 size_t prev_left = 0;
1448 unsigned int ilock_flags = 0;
1449 struct iomap_dio *dio;
1451 if (iocb->ki_flags & IOCB_NOWAIT)
1452 ilock_flags |= BTRFS_ILOCK_TRY;
1454 /* If the write DIO is within EOF, use a shared lock */
1455 if (iocb->ki_pos + iov_iter_count(from) <= i_size_read(inode))
1456 ilock_flags |= BTRFS_ILOCK_SHARED;
1459 err = btrfs_inode_lock(BTRFS_I(inode), ilock_flags);
1463 err = generic_write_checks(iocb, from);
1465 btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1469 err = btrfs_write_check(iocb, from, err);
1471 btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1477 * Re-check since file size may have changed just before taking the
1478 * lock or pos may have changed because of O_APPEND in generic_write_check()
1480 if ((ilock_flags & BTRFS_ILOCK_SHARED) &&
1481 pos + iov_iter_count(from) > i_size_read(inode)) {
1482 btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1483 ilock_flags &= ~BTRFS_ILOCK_SHARED;
1487 if (check_direct_IO(fs_info, from, pos)) {
1488 btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1493 * The iov_iter can be mapped to the same file range we are writing to.
1494 * If that's the case, then we will deadlock in the iomap code, because
1495 * it first calls our callback btrfs_dio_iomap_begin(), which will create
1496 * an ordered extent, and after that it will fault in the pages that the
1497 * iov_iter refers to. During the fault in we end up in the readahead
1498 * pages code (starting at btrfs_readahead()), which will lock the range,
1499 * find that ordered extent and then wait for it to complete (at
1500 * btrfs_lock_and_flush_ordered_range()), resulting in a deadlock since
1501 * obviously the ordered extent can never complete as we didn't submit
1502 * yet the respective bio(s). This always happens when the buffer is
1503 * memory mapped to the same file range, since the iomap DIO code always
1504 * invalidates pages in the target file range (after starting and waiting
1505 * for any writeback).
1507 * So here we disable page faults in the iov_iter and then retry if we
1508 * got -EFAULT, faulting in the pages before the retry.
1510 from->nofault = true;
1511 dio = btrfs_dio_write(iocb, from, written);
1512 from->nofault = false;
1515 * iomap_dio_complete() will call btrfs_sync_file() if we have a dsync
1516 * iocb, and that needs to lock the inode. So unlock it before calling
1517 * iomap_dio_complete() to avoid a deadlock.
1519 btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1521 if (IS_ERR_OR_NULL(dio))
1522 err = PTR_ERR_OR_ZERO(dio);
1524 err = iomap_dio_complete(dio);
1526 /* No increment (+=) because iomap returns a cumulative value. */
1530 if (iov_iter_count(from) > 0 && (err == -EFAULT || err > 0)) {
1531 const size_t left = iov_iter_count(from);
1533 * We have more data left to write. Try to fault in as many as
1534 * possible of the remainder pages and retry. We do this without
1535 * releasing and locking again the inode, to prevent races with
1538 * Also, in case the iov refers to pages in the file range of the
1539 * file we want to write to (due to a mmap), we could enter an
1540 * infinite loop if we retry after faulting the pages in, since
1541 * iomap will invalidate any pages in the range early on, before
1542 * it tries to fault in the pages of the iov. So we keep track of
1543 * how much was left of iov in the previous EFAULT and fallback
1544 * to buffered IO in case we haven't made any progress.
1546 if (left == prev_left) {
1549 fault_in_iov_iter_readable(from, left);
1556 * If 'err' is -ENOTBLK or we have not written all data, then it means
1557 * we must fallback to buffered IO.
1559 if ((err < 0 && err != -ENOTBLK) || !iov_iter_count(from))
1564 * If we are in a NOWAIT context, then return -EAGAIN to signal the caller
1565 * it must retry the operation in a context where blocking is acceptable,
1566 * because even if we end up not blocking during the buffered IO attempt
1567 * below, we will block when flushing and waiting for the IO.
1569 if (iocb->ki_flags & IOCB_NOWAIT) {
1575 written_buffered = btrfs_buffered_write(iocb, from);
1576 if (written_buffered < 0) {
1577 err = written_buffered;
1581 * Ensure all data is persisted. We want the next direct IO read to be
1582 * able to read what was just written.
1584 endbyte = pos + written_buffered - 1;
1585 err = btrfs_fdatawrite_range(inode, pos, endbyte);
1588 err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte);
1591 written += written_buffered;
1592 iocb->ki_pos = pos + written_buffered;
1593 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_SHIFT,
1594 endbyte >> PAGE_SHIFT);
1596 return err < 0 ? err : written;
1599 static ssize_t btrfs_encoded_write(struct kiocb *iocb, struct iov_iter *from,
1600 const struct btrfs_ioctl_encoded_io_args *encoded)
1602 struct file *file = iocb->ki_filp;
1603 struct inode *inode = file_inode(file);
1607 btrfs_inode_lock(BTRFS_I(inode), 0);
1608 count = encoded->len;
1609 ret = generic_write_checks_count(iocb, &count);
1610 if (ret == 0 && count != encoded->len) {
1612 * The write got truncated by generic_write_checks_count(). We
1613 * can't do a partial encoded write.
1617 if (ret || encoded->len == 0)
1620 ret = btrfs_write_check(iocb, from, encoded->len);
1624 ret = btrfs_do_encoded_write(iocb, from, encoded);
1626 btrfs_inode_unlock(BTRFS_I(inode), 0);
1630 ssize_t btrfs_do_write_iter(struct kiocb *iocb, struct iov_iter *from,
1631 const struct btrfs_ioctl_encoded_io_args *encoded)
1633 struct file *file = iocb->ki_filp;
1634 struct btrfs_inode *inode = BTRFS_I(file_inode(file));
1635 ssize_t num_written, num_sync;
1638 * If the fs flips readonly due to some impossible error, although we
1639 * have opened a file as writable, we have to stop this write operation
1640 * to ensure consistency.
1642 if (BTRFS_FS_ERROR(inode->root->fs_info))
1645 if (encoded && (iocb->ki_flags & IOCB_NOWAIT))
1649 num_written = btrfs_encoded_write(iocb, from, encoded);
1650 num_sync = encoded->len;
1651 } else if (iocb->ki_flags & IOCB_DIRECT) {
1652 num_written = btrfs_direct_write(iocb, from);
1653 num_sync = num_written;
1655 num_written = btrfs_buffered_write(iocb, from);
1656 num_sync = num_written;
1659 btrfs_set_inode_last_sub_trans(inode);
1662 num_sync = generic_write_sync(iocb, num_sync);
1664 num_written = num_sync;
1670 static ssize_t btrfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1672 return btrfs_do_write_iter(iocb, from, NULL);
1675 int btrfs_release_file(struct inode *inode, struct file *filp)
1677 struct btrfs_file_private *private = filp->private_data;
1680 kfree(private->filldir_buf);
1681 free_extent_state(private->llseek_cached_state);
1683 filp->private_data = NULL;
1687 * Set by setattr when we are about to truncate a file from a non-zero
1688 * size to a zero size. This tries to flush down new bytes that may
1689 * have been written if the application were using truncate to replace
1692 if (test_and_clear_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
1693 &BTRFS_I(inode)->runtime_flags))
1694 filemap_flush(inode->i_mapping);
1698 static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end)
1701 struct blk_plug plug;
1704 * This is only called in fsync, which would do synchronous writes, so
1705 * a plug can merge adjacent IOs as much as possible. Esp. in case of
1706 * multiple disks using raid profile, a large IO can be split to
1707 * several segments of stripe length (currently 64K).
1709 blk_start_plug(&plug);
1710 ret = btrfs_fdatawrite_range(inode, start, end);
1711 blk_finish_plug(&plug);
1716 static inline bool skip_inode_logging(const struct btrfs_log_ctx *ctx)
1718 struct btrfs_inode *inode = BTRFS_I(ctx->inode);
1719 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1721 if (btrfs_inode_in_log(inode, fs_info->generation) &&
1722 list_empty(&ctx->ordered_extents))
1726 * If we are doing a fast fsync we can not bail out if the inode's
1727 * last_trans is <= then the last committed transaction, because we only
1728 * update the last_trans of the inode during ordered extent completion,
1729 * and for a fast fsync we don't wait for that, we only wait for the
1730 * writeback to complete.
1732 if (inode->last_trans <= fs_info->last_trans_committed &&
1733 (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) ||
1734 list_empty(&ctx->ordered_extents)))
1741 * fsync call for both files and directories. This logs the inode into
1742 * the tree log instead of forcing full commits whenever possible.
1744 * It needs to call filemap_fdatawait so that all ordered extent updates are
1745 * in the metadata btree are up to date for copying to the log.
1747 * It drops the inode mutex before doing the tree log commit. This is an
1748 * important optimization for directories because holding the mutex prevents
1749 * new operations on the dir while we write to disk.
1751 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1753 struct dentry *dentry = file_dentry(file);
1754 struct inode *inode = d_inode(dentry);
1755 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1756 struct btrfs_root *root = BTRFS_I(inode)->root;
1757 struct btrfs_trans_handle *trans;
1758 struct btrfs_log_ctx ctx;
1763 trace_btrfs_sync_file(file, datasync);
1765 btrfs_init_log_ctx(&ctx, inode);
1768 * Always set the range to a full range, otherwise we can get into
1769 * several problems, from missing file extent items to represent holes
1770 * when not using the NO_HOLES feature, to log tree corruption due to
1771 * races between hole detection during logging and completion of ordered
1772 * extents outside the range, to missing checksums due to ordered extents
1773 * for which we flushed only a subset of their pages.
1777 len = (u64)LLONG_MAX + 1;
1780 * We write the dirty pages in the range and wait until they complete
1781 * out of the ->i_mutex. If so, we can flush the dirty pages by
1782 * multi-task, and make the performance up. See
1783 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1785 ret = start_ordered_ops(inode, start, end);
1789 btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
1791 atomic_inc(&root->log_batch);
1794 * Before we acquired the inode's lock and the mmap lock, someone may
1795 * have dirtied more pages in the target range. We need to make sure
1796 * that writeback for any such pages does not start while we are logging
1797 * the inode, because if it does, any of the following might happen when
1798 * we are not doing a full inode sync:
1800 * 1) We log an extent after its writeback finishes but before its
1801 * checksums are added to the csum tree, leading to -EIO errors
1802 * when attempting to read the extent after a log replay.
1804 * 2) We can end up logging an extent before its writeback finishes.
1805 * Therefore after the log replay we will have a file extent item
1806 * pointing to an unwritten extent (and no data checksums as well).
1808 * So trigger writeback for any eventual new dirty pages and then we
1809 * wait for all ordered extents to complete below.
1811 ret = start_ordered_ops(inode, start, end);
1813 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
1818 * Always check for the full sync flag while holding the inode's lock,
1819 * to avoid races with other tasks. The flag must be either set all the
1820 * time during logging or always off all the time while logging.
1821 * We check the flag here after starting delalloc above, because when
1822 * running delalloc the full sync flag may be set if we need to drop
1823 * extra extent map ranges due to temporary memory allocation failures.
1825 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1826 &BTRFS_I(inode)->runtime_flags);
1829 * We have to do this here to avoid the priority inversion of waiting on
1830 * IO of a lower priority task while holding a transaction open.
1832 * For a full fsync we wait for the ordered extents to complete while
1833 * for a fast fsync we wait just for writeback to complete, and then
1834 * attach the ordered extents to the transaction so that a transaction
1835 * commit waits for their completion, to avoid data loss if we fsync,
1836 * the current transaction commits before the ordered extents complete
1837 * and a power failure happens right after that.
1839 * For zoned filesystem, if a write IO uses a ZONE_APPEND command, the
1840 * logical address recorded in the ordered extent may change. We need
1841 * to wait for the IO to stabilize the logical address.
1843 if (full_sync || btrfs_is_zoned(fs_info)) {
1844 ret = btrfs_wait_ordered_range(inode, start, len);
1847 * Get our ordered extents as soon as possible to avoid doing
1848 * checksum lookups in the csum tree, and use instead the
1849 * checksums attached to the ordered extents.
1851 btrfs_get_ordered_extents_for_logging(BTRFS_I(inode),
1852 &ctx.ordered_extents);
1853 ret = filemap_fdatawait_range(inode->i_mapping, start, end);
1857 goto out_release_extents;
1859 atomic_inc(&root->log_batch);
1862 if (skip_inode_logging(&ctx)) {
1864 * We've had everything committed since the last time we were
1865 * modified so clear this flag in case it was set for whatever
1866 * reason, it's no longer relevant.
1868 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1869 &BTRFS_I(inode)->runtime_flags);
1871 * An ordered extent might have started before and completed
1872 * already with io errors, in which case the inode was not
1873 * updated and we end up here. So check the inode's mapping
1874 * for any errors that might have happened since we last
1875 * checked called fsync.
1877 ret = filemap_check_wb_err(inode->i_mapping, file->f_wb_err);
1878 goto out_release_extents;
1882 * We use start here because we will need to wait on the IO to complete
1883 * in btrfs_sync_log, which could require joining a transaction (for
1884 * example checking cross references in the nocow path). If we use join
1885 * here we could get into a situation where we're waiting on IO to
1886 * happen that is blocked on a transaction trying to commit. With start
1887 * we inc the extwriter counter, so we wait for all extwriters to exit
1888 * before we start blocking joiners. This comment is to keep somebody
1889 * from thinking they are super smart and changing this to
1890 * btrfs_join_transaction *cough*Josef*cough*.
1892 trans = btrfs_start_transaction(root, 0);
1893 if (IS_ERR(trans)) {
1894 ret = PTR_ERR(trans);
1895 goto out_release_extents;
1897 trans->in_fsync = true;
1899 ret = btrfs_log_dentry_safe(trans, dentry, &ctx);
1900 btrfs_release_log_ctx_extents(&ctx);
1902 /* Fallthrough and commit/free transaction. */
1903 ret = BTRFS_LOG_FORCE_COMMIT;
1906 /* we've logged all the items and now have a consistent
1907 * version of the file in the log. It is possible that
1908 * someone will come in and modify the file, but that's
1909 * fine because the log is consistent on disk, and we
1910 * have references to all of the file's extents
1912 * It is possible that someone will come in and log the
1913 * file again, but that will end up using the synchronization
1914 * inside btrfs_sync_log to keep things safe.
1916 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
1918 if (ret == BTRFS_NO_LOG_SYNC) {
1919 ret = btrfs_end_transaction(trans);
1923 /* We successfully logged the inode, attempt to sync the log. */
1925 ret = btrfs_sync_log(trans, root, &ctx);
1927 ret = btrfs_end_transaction(trans);
1933 * At this point we need to commit the transaction because we had
1934 * btrfs_need_log_full_commit() or some other error.
1936 * If we didn't do a full sync we have to stop the trans handle, wait on
1937 * the ordered extents, start it again and commit the transaction. If
1938 * we attempt to wait on the ordered extents here we could deadlock with
1939 * something like fallocate() that is holding the extent lock trying to
1940 * start a transaction while some other thread is trying to commit the
1941 * transaction while we (fsync) are currently holding the transaction
1945 ret = btrfs_end_transaction(trans);
1948 ret = btrfs_wait_ordered_range(inode, start, len);
1953 * This is safe to use here because we're only interested in
1954 * making sure the transaction that had the ordered extents is
1955 * committed. We aren't waiting on anything past this point,
1956 * we're purely getting the transaction and committing it.
1958 trans = btrfs_attach_transaction_barrier(root);
1959 if (IS_ERR(trans)) {
1960 ret = PTR_ERR(trans);
1963 * We committed the transaction and there's no currently
1964 * running transaction, this means everything we care
1965 * about made it to disk and we are done.
1973 ret = btrfs_commit_transaction(trans);
1975 ASSERT(list_empty(&ctx.list));
1976 ASSERT(list_empty(&ctx.conflict_inodes));
1977 err = file_check_and_advance_wb_err(file);
1980 return ret > 0 ? -EIO : ret;
1982 out_release_extents:
1983 btrfs_release_log_ctx_extents(&ctx);
1984 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
1988 static const struct vm_operations_struct btrfs_file_vm_ops = {
1989 .fault = filemap_fault,
1990 .map_pages = filemap_map_pages,
1991 .page_mkwrite = btrfs_page_mkwrite,
1994 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1996 struct address_space *mapping = filp->f_mapping;
1998 if (!mapping->a_ops->read_folio)
2001 file_accessed(filp);
2002 vma->vm_ops = &btrfs_file_vm_ops;
2007 static int hole_mergeable(struct btrfs_inode *inode, struct extent_buffer *leaf,
2008 int slot, u64 start, u64 end)
2010 struct btrfs_file_extent_item *fi;
2011 struct btrfs_key key;
2013 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2016 btrfs_item_key_to_cpu(leaf, &key, slot);
2017 if (key.objectid != btrfs_ino(inode) ||
2018 key.type != BTRFS_EXTENT_DATA_KEY)
2021 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2023 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2026 if (btrfs_file_extent_disk_bytenr(leaf, fi))
2029 if (key.offset == end)
2031 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2036 static int fill_holes(struct btrfs_trans_handle *trans,
2037 struct btrfs_inode *inode,
2038 struct btrfs_path *path, u64 offset, u64 end)
2040 struct btrfs_fs_info *fs_info = trans->fs_info;
2041 struct btrfs_root *root = inode->root;
2042 struct extent_buffer *leaf;
2043 struct btrfs_file_extent_item *fi;
2044 struct extent_map *hole_em;
2045 struct btrfs_key key;
2048 if (btrfs_fs_incompat(fs_info, NO_HOLES))
2051 key.objectid = btrfs_ino(inode);
2052 key.type = BTRFS_EXTENT_DATA_KEY;
2053 key.offset = offset;
2055 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2058 * We should have dropped this offset, so if we find it then
2059 * something has gone horribly wrong.
2066 leaf = path->nodes[0];
2067 if (hole_mergeable(inode, leaf, path->slots[0] - 1, offset, end)) {
2071 fi = btrfs_item_ptr(leaf, path->slots[0],
2072 struct btrfs_file_extent_item);
2073 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2075 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2076 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2077 btrfs_set_file_extent_offset(leaf, fi, 0);
2078 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2079 btrfs_mark_buffer_dirty(leaf);
2083 if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) {
2086 key.offset = offset;
2087 btrfs_set_item_key_safe(fs_info, path, &key);
2088 fi = btrfs_item_ptr(leaf, path->slots[0],
2089 struct btrfs_file_extent_item);
2090 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2092 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2093 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2094 btrfs_set_file_extent_offset(leaf, fi, 0);
2095 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2096 btrfs_mark_buffer_dirty(leaf);
2099 btrfs_release_path(path);
2101 ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset,
2107 btrfs_release_path(path);
2109 hole_em = alloc_extent_map();
2111 btrfs_drop_extent_map_range(inode, offset, end - 1, false);
2112 btrfs_set_inode_full_sync(inode);
2114 hole_em->start = offset;
2115 hole_em->len = end - offset;
2116 hole_em->ram_bytes = hole_em->len;
2117 hole_em->orig_start = offset;
2119 hole_em->block_start = EXTENT_MAP_HOLE;
2120 hole_em->block_len = 0;
2121 hole_em->orig_block_len = 0;
2122 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2123 hole_em->generation = trans->transid;
2125 ret = btrfs_replace_extent_map_range(inode, hole_em, true);
2126 free_extent_map(hole_em);
2128 btrfs_set_inode_full_sync(inode);
2135 * Find a hole extent on given inode and change start/len to the end of hole
2136 * extent.(hole/vacuum extent whose em->start <= start &&
2137 * em->start + em->len > start)
2138 * When a hole extent is found, return 1 and modify start/len.
2140 static int find_first_non_hole(struct btrfs_inode *inode, u64 *start, u64 *len)
2142 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2143 struct extent_map *em;
2146 em = btrfs_get_extent(inode, NULL, 0,
2147 round_down(*start, fs_info->sectorsize),
2148 round_up(*len, fs_info->sectorsize));
2152 /* Hole or vacuum extent(only exists in no-hole mode) */
2153 if (em->block_start == EXTENT_MAP_HOLE) {
2155 *len = em->start + em->len > *start + *len ?
2156 0 : *start + *len - em->start - em->len;
2157 *start = em->start + em->len;
2159 free_extent_map(em);
2163 static void btrfs_punch_hole_lock_range(struct inode *inode,
2164 const u64 lockstart,
2166 struct extent_state **cached_state)
2169 * For subpage case, if the range is not at page boundary, we could
2170 * have pages at the leading/tailing part of the range.
2171 * This could lead to dead loop since filemap_range_has_page()
2172 * will always return true.
2173 * So here we need to do extra page alignment for
2174 * filemap_range_has_page().
2176 const u64 page_lockstart = round_up(lockstart, PAGE_SIZE);
2177 const u64 page_lockend = round_down(lockend + 1, PAGE_SIZE) - 1;
2180 truncate_pagecache_range(inode, lockstart, lockend);
2182 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2185 * We can't have ordered extents in the range, nor dirty/writeback
2186 * pages, because we have locked the inode's VFS lock in exclusive
2187 * mode, we have locked the inode's i_mmap_lock in exclusive mode,
2188 * we have flushed all delalloc in the range and we have waited
2189 * for any ordered extents in the range to complete.
2190 * We can race with anyone reading pages from this range, so after
2191 * locking the range check if we have pages in the range, and if
2192 * we do, unlock the range and retry.
2194 if (!filemap_range_has_page(inode->i_mapping, page_lockstart,
2198 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2202 btrfs_assert_inode_range_clean(BTRFS_I(inode), lockstart, lockend);
2205 static int btrfs_insert_replace_extent(struct btrfs_trans_handle *trans,
2206 struct btrfs_inode *inode,
2207 struct btrfs_path *path,
2208 struct btrfs_replace_extent_info *extent_info,
2209 const u64 replace_len,
2210 const u64 bytes_to_drop)
2212 struct btrfs_fs_info *fs_info = trans->fs_info;
2213 struct btrfs_root *root = inode->root;
2214 struct btrfs_file_extent_item *extent;
2215 struct extent_buffer *leaf;
2216 struct btrfs_key key;
2218 struct btrfs_ref ref = { 0 };
2221 if (replace_len == 0)
2224 if (extent_info->disk_offset == 0 &&
2225 btrfs_fs_incompat(fs_info, NO_HOLES)) {
2226 btrfs_update_inode_bytes(inode, 0, bytes_to_drop);
2230 key.objectid = btrfs_ino(inode);
2231 key.type = BTRFS_EXTENT_DATA_KEY;
2232 key.offset = extent_info->file_offset;
2233 ret = btrfs_insert_empty_item(trans, root, path, &key,
2234 sizeof(struct btrfs_file_extent_item));
2237 leaf = path->nodes[0];
2238 slot = path->slots[0];
2239 write_extent_buffer(leaf, extent_info->extent_buf,
2240 btrfs_item_ptr_offset(leaf, slot),
2241 sizeof(struct btrfs_file_extent_item));
2242 extent = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2243 ASSERT(btrfs_file_extent_type(leaf, extent) != BTRFS_FILE_EXTENT_INLINE);
2244 btrfs_set_file_extent_offset(leaf, extent, extent_info->data_offset);
2245 btrfs_set_file_extent_num_bytes(leaf, extent, replace_len);
2246 if (extent_info->is_new_extent)
2247 btrfs_set_file_extent_generation(leaf, extent, trans->transid);
2248 btrfs_mark_buffer_dirty(leaf);
2249 btrfs_release_path(path);
2251 ret = btrfs_inode_set_file_extent_range(inode, extent_info->file_offset,
2256 /* If it's a hole, nothing more needs to be done. */
2257 if (extent_info->disk_offset == 0) {
2258 btrfs_update_inode_bytes(inode, 0, bytes_to_drop);
2262 btrfs_update_inode_bytes(inode, replace_len, bytes_to_drop);
2264 if (extent_info->is_new_extent && extent_info->insertions == 0) {
2265 key.objectid = extent_info->disk_offset;
2266 key.type = BTRFS_EXTENT_ITEM_KEY;
2267 key.offset = extent_info->disk_len;
2268 ret = btrfs_alloc_reserved_file_extent(trans, root,
2270 extent_info->file_offset,
2271 extent_info->qgroup_reserved,
2276 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
2277 extent_info->disk_offset,
2278 extent_info->disk_len, 0);
2279 ref_offset = extent_info->file_offset - extent_info->data_offset;
2280 btrfs_init_data_ref(&ref, root->root_key.objectid,
2281 btrfs_ino(inode), ref_offset, 0, false);
2282 ret = btrfs_inc_extent_ref(trans, &ref);
2285 extent_info->insertions++;
2291 * The respective range must have been previously locked, as well as the inode.
2292 * The end offset is inclusive (last byte of the range).
2293 * @extent_info is NULL for fallocate's hole punching and non-NULL when replacing
2294 * the file range with an extent.
2295 * When not punching a hole, we don't want to end up in a state where we dropped
2296 * extents without inserting a new one, so we must abort the transaction to avoid
2299 int btrfs_replace_file_extents(struct btrfs_inode *inode,
2300 struct btrfs_path *path, const u64 start,
2302 struct btrfs_replace_extent_info *extent_info,
2303 struct btrfs_trans_handle **trans_out)
2305 struct btrfs_drop_extents_args drop_args = { 0 };
2306 struct btrfs_root *root = inode->root;
2307 struct btrfs_fs_info *fs_info = root->fs_info;
2308 u64 min_size = btrfs_calc_insert_metadata_size(fs_info, 1);
2309 u64 ino_size = round_up(inode->vfs_inode.i_size, fs_info->sectorsize);
2310 struct btrfs_trans_handle *trans = NULL;
2311 struct btrfs_block_rsv *rsv;
2312 unsigned int rsv_count;
2314 u64 len = end - start;
2320 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
2325 rsv->size = btrfs_calc_insert_metadata_size(fs_info, 1);
2326 rsv->failfast = true;
2329 * 1 - update the inode
2330 * 1 - removing the extents in the range
2331 * 1 - adding the hole extent if no_holes isn't set or if we are
2332 * replacing the range with a new extent
2334 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || extent_info)
2339 trans = btrfs_start_transaction(root, rsv_count);
2340 if (IS_ERR(trans)) {
2341 ret = PTR_ERR(trans);
2346 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
2350 trans->block_rsv = rsv;
2353 drop_args.path = path;
2354 drop_args.end = end + 1;
2355 drop_args.drop_cache = true;
2356 while (cur_offset < end) {
2357 drop_args.start = cur_offset;
2358 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2359 /* If we are punching a hole decrement the inode's byte count */
2361 btrfs_update_inode_bytes(inode, 0,
2362 drop_args.bytes_found);
2363 if (ret != -ENOSPC) {
2365 * The only time we don't want to abort is if we are
2366 * attempting to clone a partial inline extent, in which
2367 * case we'll get EOPNOTSUPP. However if we aren't
2368 * clone we need to abort no matter what, because if we
2369 * got EOPNOTSUPP via prealloc then we messed up and
2373 (ret != -EOPNOTSUPP ||
2374 (extent_info && extent_info->is_new_extent)))
2375 btrfs_abort_transaction(trans, ret);
2379 trans->block_rsv = &fs_info->trans_block_rsv;
2381 if (!extent_info && cur_offset < drop_args.drop_end &&
2382 cur_offset < ino_size) {
2383 ret = fill_holes(trans, inode, path, cur_offset,
2384 drop_args.drop_end);
2387 * If we failed then we didn't insert our hole
2388 * entries for the area we dropped, so now the
2389 * fs is corrupted, so we must abort the
2392 btrfs_abort_transaction(trans, ret);
2395 } else if (!extent_info && cur_offset < drop_args.drop_end) {
2397 * We are past the i_size here, but since we didn't
2398 * insert holes we need to clear the mapped area so we
2399 * know to not set disk_i_size in this area until a new
2400 * file extent is inserted here.
2402 ret = btrfs_inode_clear_file_extent_range(inode,
2404 drop_args.drop_end - cur_offset);
2407 * We couldn't clear our area, so we could
2408 * presumably adjust up and corrupt the fs, so
2411 btrfs_abort_transaction(trans, ret);
2417 drop_args.drop_end > extent_info->file_offset) {
2418 u64 replace_len = drop_args.drop_end -
2419 extent_info->file_offset;
2421 ret = btrfs_insert_replace_extent(trans, inode, path,
2422 extent_info, replace_len,
2423 drop_args.bytes_found);
2425 btrfs_abort_transaction(trans, ret);
2428 extent_info->data_len -= replace_len;
2429 extent_info->data_offset += replace_len;
2430 extent_info->file_offset += replace_len;
2434 * We are releasing our handle on the transaction, balance the
2435 * dirty pages of the btree inode and flush delayed items, and
2436 * then get a new transaction handle, which may now point to a
2437 * new transaction in case someone else may have committed the
2438 * transaction we used to replace/drop file extent items. So
2439 * bump the inode's iversion and update mtime and ctime except
2440 * if we are called from a dedupe context. This is because a
2441 * power failure/crash may happen after the transaction is
2442 * committed and before we finish replacing/dropping all the
2443 * file extent items we need.
2445 inode_inc_iversion(&inode->vfs_inode);
2447 if (!extent_info || extent_info->update_times)
2448 inode->vfs_inode.i_mtime = inode_set_ctime_current(&inode->vfs_inode);
2450 ret = btrfs_update_inode(trans, root, inode);
2454 btrfs_end_transaction(trans);
2455 btrfs_btree_balance_dirty(fs_info);
2457 trans = btrfs_start_transaction(root, rsv_count);
2458 if (IS_ERR(trans)) {
2459 ret = PTR_ERR(trans);
2464 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
2465 rsv, min_size, false);
2468 trans->block_rsv = rsv;
2470 cur_offset = drop_args.drop_end;
2471 len = end - cur_offset;
2472 if (!extent_info && len) {
2473 ret = find_first_non_hole(inode, &cur_offset, &len);
2474 if (unlikely(ret < 0))
2484 * If we were cloning, force the next fsync to be a full one since we
2485 * we replaced (or just dropped in the case of cloning holes when
2486 * NO_HOLES is enabled) file extent items and did not setup new extent
2487 * maps for the replacement extents (or holes).
2489 if (extent_info && !extent_info->is_new_extent)
2490 btrfs_set_inode_full_sync(inode);
2495 trans->block_rsv = &fs_info->trans_block_rsv;
2497 * If we are using the NO_HOLES feature we might have had already an
2498 * hole that overlaps a part of the region [lockstart, lockend] and
2499 * ends at (or beyond) lockend. Since we have no file extent items to
2500 * represent holes, drop_end can be less than lockend and so we must
2501 * make sure we have an extent map representing the existing hole (the
2502 * call to __btrfs_drop_extents() might have dropped the existing extent
2503 * map representing the existing hole), otherwise the fast fsync path
2504 * will not record the existence of the hole region
2505 * [existing_hole_start, lockend].
2507 if (drop_args.drop_end <= end)
2508 drop_args.drop_end = end + 1;
2510 * Don't insert file hole extent item if it's for a range beyond eof
2511 * (because it's useless) or if it represents a 0 bytes range (when
2512 * cur_offset == drop_end).
2514 if (!extent_info && cur_offset < ino_size &&
2515 cur_offset < drop_args.drop_end) {
2516 ret = fill_holes(trans, inode, path, cur_offset,
2517 drop_args.drop_end);
2519 /* Same comment as above. */
2520 btrfs_abort_transaction(trans, ret);
2523 } else if (!extent_info && cur_offset < drop_args.drop_end) {
2524 /* See the comment in the loop above for the reasoning here. */
2525 ret = btrfs_inode_clear_file_extent_range(inode, cur_offset,
2526 drop_args.drop_end - cur_offset);
2528 btrfs_abort_transaction(trans, ret);
2534 ret = btrfs_insert_replace_extent(trans, inode, path,
2535 extent_info, extent_info->data_len,
2536 drop_args.bytes_found);
2538 btrfs_abort_transaction(trans, ret);
2547 trans->block_rsv = &fs_info->trans_block_rsv;
2549 btrfs_end_transaction(trans);
2553 btrfs_free_block_rsv(fs_info, rsv);
2558 static int btrfs_punch_hole(struct file *file, loff_t offset, loff_t len)
2560 struct inode *inode = file_inode(file);
2561 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2562 struct btrfs_root *root = BTRFS_I(inode)->root;
2563 struct extent_state *cached_state = NULL;
2564 struct btrfs_path *path;
2565 struct btrfs_trans_handle *trans = NULL;
2570 u64 orig_start = offset;
2574 bool truncated_block = false;
2575 bool updated_inode = false;
2577 btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
2579 ret = btrfs_wait_ordered_range(inode, offset, len);
2581 goto out_only_mutex;
2583 ino_size = round_up(inode->i_size, fs_info->sectorsize);
2584 ret = find_first_non_hole(BTRFS_I(inode), &offset, &len);
2586 goto out_only_mutex;
2588 /* Already in a large hole */
2590 goto out_only_mutex;
2593 ret = file_modified(file);
2595 goto out_only_mutex;
2597 lockstart = round_up(offset, fs_info->sectorsize);
2598 lockend = round_down(offset + len, fs_info->sectorsize) - 1;
2599 same_block = (BTRFS_BYTES_TO_BLKS(fs_info, offset))
2600 == (BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1));
2602 * We needn't truncate any block which is beyond the end of the file
2603 * because we are sure there is no data there.
2606 * Only do this if we are in the same block and we aren't doing the
2609 if (same_block && len < fs_info->sectorsize) {
2610 if (offset < ino_size) {
2611 truncated_block = true;
2612 ret = btrfs_truncate_block(BTRFS_I(inode), offset, len,
2617 goto out_only_mutex;
2620 /* zero back part of the first block */
2621 if (offset < ino_size) {
2622 truncated_block = true;
2623 ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0);
2625 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
2630 /* Check the aligned pages after the first unaligned page,
2631 * if offset != orig_start, which means the first unaligned page
2632 * including several following pages are already in holes,
2633 * the extra check can be skipped */
2634 if (offset == orig_start) {
2635 /* after truncate page, check hole again */
2636 len = offset + len - lockstart;
2638 ret = find_first_non_hole(BTRFS_I(inode), &offset, &len);
2640 goto out_only_mutex;
2643 goto out_only_mutex;
2648 /* Check the tail unaligned part is in a hole */
2649 tail_start = lockend + 1;
2650 tail_len = offset + len - tail_start;
2652 ret = find_first_non_hole(BTRFS_I(inode), &tail_start, &tail_len);
2653 if (unlikely(ret < 0))
2654 goto out_only_mutex;
2656 /* zero the front end of the last page */
2657 if (tail_start + tail_len < ino_size) {
2658 truncated_block = true;
2659 ret = btrfs_truncate_block(BTRFS_I(inode),
2660 tail_start + tail_len,
2663 goto out_only_mutex;
2668 if (lockend < lockstart) {
2670 goto out_only_mutex;
2673 btrfs_punch_hole_lock_range(inode, lockstart, lockend, &cached_state);
2675 path = btrfs_alloc_path();
2681 ret = btrfs_replace_file_extents(BTRFS_I(inode), path, lockstart,
2682 lockend, NULL, &trans);
2683 btrfs_free_path(path);
2687 ASSERT(trans != NULL);
2688 inode_inc_iversion(inode);
2689 inode->i_mtime = inode_set_ctime_current(inode);
2690 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
2691 updated_inode = true;
2692 btrfs_end_transaction(trans);
2693 btrfs_btree_balance_dirty(fs_info);
2695 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2698 if (!updated_inode && truncated_block && !ret) {
2700 * If we only end up zeroing part of a page, we still need to
2701 * update the inode item, so that all the time fields are
2702 * updated as well as the necessary btrfs inode in memory fields
2703 * for detecting, at fsync time, if the inode isn't yet in the
2704 * log tree or it's there but not up to date.
2706 struct timespec64 now = inode_set_ctime_current(inode);
2708 inode_inc_iversion(inode);
2709 inode->i_mtime = now;
2710 trans = btrfs_start_transaction(root, 1);
2711 if (IS_ERR(trans)) {
2712 ret = PTR_ERR(trans);
2716 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
2717 ret2 = btrfs_end_transaction(trans);
2722 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
2726 /* Helper structure to record which range is already reserved */
2727 struct falloc_range {
2728 struct list_head list;
2734 * Helper function to add falloc range
2736 * Caller should have locked the larger range of extent containing
2739 static int add_falloc_range(struct list_head *head, u64 start, u64 len)
2741 struct falloc_range *range = NULL;
2743 if (!list_empty(head)) {
2745 * As fallocate iterates by bytenr order, we only need to check
2748 range = list_last_entry(head, struct falloc_range, list);
2749 if (range->start + range->len == start) {
2755 range = kmalloc(sizeof(*range), GFP_KERNEL);
2758 range->start = start;
2760 list_add_tail(&range->list, head);
2764 static int btrfs_fallocate_update_isize(struct inode *inode,
2768 struct btrfs_trans_handle *trans;
2769 struct btrfs_root *root = BTRFS_I(inode)->root;
2773 if (mode & FALLOC_FL_KEEP_SIZE || end <= i_size_read(inode))
2776 trans = btrfs_start_transaction(root, 1);
2778 return PTR_ERR(trans);
2780 inode_set_ctime_current(inode);
2781 i_size_write(inode, end);
2782 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
2783 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
2784 ret2 = btrfs_end_transaction(trans);
2786 return ret ? ret : ret2;
2790 RANGE_BOUNDARY_WRITTEN_EXTENT,
2791 RANGE_BOUNDARY_PREALLOC_EXTENT,
2792 RANGE_BOUNDARY_HOLE,
2795 static int btrfs_zero_range_check_range_boundary(struct btrfs_inode *inode,
2798 const u64 sectorsize = inode->root->fs_info->sectorsize;
2799 struct extent_map *em;
2802 offset = round_down(offset, sectorsize);
2803 em = btrfs_get_extent(inode, NULL, 0, offset, sectorsize);
2807 if (em->block_start == EXTENT_MAP_HOLE)
2808 ret = RANGE_BOUNDARY_HOLE;
2809 else if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2810 ret = RANGE_BOUNDARY_PREALLOC_EXTENT;
2812 ret = RANGE_BOUNDARY_WRITTEN_EXTENT;
2814 free_extent_map(em);
2818 static int btrfs_zero_range(struct inode *inode,
2823 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2824 struct extent_map *em;
2825 struct extent_changeset *data_reserved = NULL;
2828 const u64 sectorsize = fs_info->sectorsize;
2829 u64 alloc_start = round_down(offset, sectorsize);
2830 u64 alloc_end = round_up(offset + len, sectorsize);
2831 u64 bytes_to_reserve = 0;
2832 bool space_reserved = false;
2834 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start,
2835 alloc_end - alloc_start);
2842 * Avoid hole punching and extent allocation for some cases. More cases
2843 * could be considered, but these are unlikely common and we keep things
2844 * as simple as possible for now. Also, intentionally, if the target
2845 * range contains one or more prealloc extents together with regular
2846 * extents and holes, we drop all the existing extents and allocate a
2847 * new prealloc extent, so that we get a larger contiguous disk extent.
2849 if (em->start <= alloc_start &&
2850 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
2851 const u64 em_end = em->start + em->len;
2853 if (em_end >= offset + len) {
2855 * The whole range is already a prealloc extent,
2856 * do nothing except updating the inode's i_size if
2859 free_extent_map(em);
2860 ret = btrfs_fallocate_update_isize(inode, offset + len,
2865 * Part of the range is already a prealloc extent, so operate
2866 * only on the remaining part of the range.
2868 alloc_start = em_end;
2869 ASSERT(IS_ALIGNED(alloc_start, sectorsize));
2870 len = offset + len - alloc_start;
2871 offset = alloc_start;
2872 alloc_hint = em->block_start + em->len;
2874 free_extent_map(em);
2876 if (BTRFS_BYTES_TO_BLKS(fs_info, offset) ==
2877 BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)) {
2878 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start,
2885 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
2886 free_extent_map(em);
2887 ret = btrfs_fallocate_update_isize(inode, offset + len,
2891 if (len < sectorsize && em->block_start != EXTENT_MAP_HOLE) {
2892 free_extent_map(em);
2893 ret = btrfs_truncate_block(BTRFS_I(inode), offset, len,
2896 ret = btrfs_fallocate_update_isize(inode,
2901 free_extent_map(em);
2902 alloc_start = round_down(offset, sectorsize);
2903 alloc_end = alloc_start + sectorsize;
2907 alloc_start = round_up(offset, sectorsize);
2908 alloc_end = round_down(offset + len, sectorsize);
2911 * For unaligned ranges, check the pages at the boundaries, they might
2912 * map to an extent, in which case we need to partially zero them, or
2913 * they might map to a hole, in which case we need our allocation range
2916 if (!IS_ALIGNED(offset, sectorsize)) {
2917 ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode),
2921 if (ret == RANGE_BOUNDARY_HOLE) {
2922 alloc_start = round_down(offset, sectorsize);
2924 } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
2925 ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0);
2933 if (!IS_ALIGNED(offset + len, sectorsize)) {
2934 ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode),
2938 if (ret == RANGE_BOUNDARY_HOLE) {
2939 alloc_end = round_up(offset + len, sectorsize);
2941 } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
2942 ret = btrfs_truncate_block(BTRFS_I(inode), offset + len,
2952 if (alloc_start < alloc_end) {
2953 struct extent_state *cached_state = NULL;
2954 const u64 lockstart = alloc_start;
2955 const u64 lockend = alloc_end - 1;
2957 bytes_to_reserve = alloc_end - alloc_start;
2958 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
2962 space_reserved = true;
2963 btrfs_punch_hole_lock_range(inode, lockstart, lockend,
2965 ret = btrfs_qgroup_reserve_data(BTRFS_I(inode), &data_reserved,
2966 alloc_start, bytes_to_reserve);
2968 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart,
2969 lockend, &cached_state);
2972 ret = btrfs_prealloc_file_range(inode, mode, alloc_start,
2973 alloc_end - alloc_start,
2975 offset + len, &alloc_hint);
2976 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2978 /* btrfs_prealloc_file_range releases reserved space on error */
2980 space_reserved = false;
2984 ret = btrfs_fallocate_update_isize(inode, offset + len, mode);
2986 if (ret && space_reserved)
2987 btrfs_free_reserved_data_space(BTRFS_I(inode), data_reserved,
2988 alloc_start, bytes_to_reserve);
2989 extent_changeset_free(data_reserved);
2994 static long btrfs_fallocate(struct file *file, int mode,
2995 loff_t offset, loff_t len)
2997 struct inode *inode = file_inode(file);
2998 struct extent_state *cached_state = NULL;
2999 struct extent_changeset *data_reserved = NULL;
3000 struct falloc_range *range;
3001 struct falloc_range *tmp;
3002 LIST_HEAD(reserve_list);
3010 u64 data_space_needed = 0;
3011 u64 data_space_reserved = 0;
3012 u64 qgroup_reserved = 0;
3013 struct extent_map *em;
3014 int blocksize = BTRFS_I(inode)->root->fs_info->sectorsize;
3017 /* Do not allow fallocate in ZONED mode */
3018 if (btrfs_is_zoned(btrfs_sb(inode->i_sb)))
3021 alloc_start = round_down(offset, blocksize);
3022 alloc_end = round_up(offset + len, blocksize);
3023 cur_offset = alloc_start;
3025 /* Make sure we aren't being give some crap mode */
3026 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
3027 FALLOC_FL_ZERO_RANGE))
3030 if (mode & FALLOC_FL_PUNCH_HOLE)
3031 return btrfs_punch_hole(file, offset, len);
3033 btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
3035 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) {
3036 ret = inode_newsize_ok(inode, offset + len);
3041 ret = file_modified(file);
3046 * TODO: Move these two operations after we have checked
3047 * accurate reserved space, or fallocate can still fail but
3048 * with page truncated or size expanded.
3050 * But that's a minor problem and won't do much harm BTW.
3052 if (alloc_start > inode->i_size) {
3053 ret = btrfs_cont_expand(BTRFS_I(inode), i_size_read(inode),
3057 } else if (offset + len > inode->i_size) {
3059 * If we are fallocating from the end of the file onward we
3060 * need to zero out the end of the block if i_size lands in the
3061 * middle of a block.
3063 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
3069 * We have locked the inode at the VFS level (in exclusive mode) and we
3070 * have locked the i_mmap_lock lock (in exclusive mode). Now before
3071 * locking the file range, flush all dealloc in the range and wait for
3072 * all ordered extents in the range to complete. After this we can lock
3073 * the file range and, due to the previous locking we did, we know there
3074 * can't be more delalloc or ordered extents in the range.
3076 ret = btrfs_wait_ordered_range(inode, alloc_start,
3077 alloc_end - alloc_start);
3081 if (mode & FALLOC_FL_ZERO_RANGE) {
3082 ret = btrfs_zero_range(inode, offset, len, mode);
3083 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
3087 locked_end = alloc_end - 1;
3088 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
3091 btrfs_assert_inode_range_clean(BTRFS_I(inode), alloc_start, locked_end);
3093 /* First, check if we exceed the qgroup limit */
3094 while (cur_offset < alloc_end) {
3095 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
3096 alloc_end - cur_offset);
3101 last_byte = min(extent_map_end(em), alloc_end);
3102 actual_end = min_t(u64, extent_map_end(em), offset + len);
3103 last_byte = ALIGN(last_byte, blocksize);
3104 if (em->block_start == EXTENT_MAP_HOLE ||
3105 (cur_offset >= inode->i_size &&
3106 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
3107 const u64 range_len = last_byte - cur_offset;
3109 ret = add_falloc_range(&reserve_list, cur_offset, range_len);
3111 free_extent_map(em);
3114 ret = btrfs_qgroup_reserve_data(BTRFS_I(inode),
3115 &data_reserved, cur_offset, range_len);
3117 free_extent_map(em);
3120 qgroup_reserved += range_len;
3121 data_space_needed += range_len;
3123 free_extent_map(em);
3124 cur_offset = last_byte;
3127 if (!ret && data_space_needed > 0) {
3129 * We are safe to reserve space here as we can't have delalloc
3130 * in the range, see above.
3132 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
3135 data_space_reserved = data_space_needed;
3139 * If ret is still 0, means we're OK to fallocate.
3140 * Or just cleanup the list and exit.
3142 list_for_each_entry_safe(range, tmp, &reserve_list, list) {
3144 ret = btrfs_prealloc_file_range(inode, mode,
3146 range->len, i_blocksize(inode),
3147 offset + len, &alloc_hint);
3149 * btrfs_prealloc_file_range() releases space even
3150 * if it returns an error.
3152 data_space_reserved -= range->len;
3153 qgroup_reserved -= range->len;
3154 } else if (data_space_reserved > 0) {
3155 btrfs_free_reserved_data_space(BTRFS_I(inode),
3156 data_reserved, range->start,
3158 data_space_reserved -= range->len;
3159 qgroup_reserved -= range->len;
3160 } else if (qgroup_reserved > 0) {
3161 btrfs_qgroup_free_data(BTRFS_I(inode), data_reserved,
3162 range->start, range->len);
3163 qgroup_reserved -= range->len;
3165 list_del(&range->list);
3172 * We didn't need to allocate any more space, but we still extended the
3173 * size of the file so we need to update i_size and the inode item.
3175 ret = btrfs_fallocate_update_isize(inode, actual_end, mode);
3177 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
3180 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
3181 extent_changeset_free(data_reserved);
3186 * Helper for btrfs_find_delalloc_in_range(). Find a subrange in a given range
3187 * that has unflushed and/or flushing delalloc. There might be other adjacent
3188 * subranges after the one it found, so btrfs_find_delalloc_in_range() keeps
3189 * looping while it gets adjacent subranges, and merging them together.
3191 static bool find_delalloc_subrange(struct btrfs_inode *inode, u64 start, u64 end,
3192 struct extent_state **cached_state,
3193 bool *search_io_tree,
3194 u64 *delalloc_start_ret, u64 *delalloc_end_ret)
3196 u64 len = end + 1 - start;
3197 u64 delalloc_len = 0;
3198 struct btrfs_ordered_extent *oe;
3203 * Search the io tree first for EXTENT_DELALLOC. If we find any, it
3204 * means we have delalloc (dirty pages) for which writeback has not
3207 if (*search_io_tree) {
3208 spin_lock(&inode->lock);
3209 if (inode->delalloc_bytes > 0) {
3210 spin_unlock(&inode->lock);
3211 *delalloc_start_ret = start;
3212 delalloc_len = count_range_bits(&inode->io_tree,
3213 delalloc_start_ret, end,
3214 len, EXTENT_DELALLOC, 1,
3217 spin_unlock(&inode->lock);
3221 if (delalloc_len > 0) {
3223 * If delalloc was found then *delalloc_start_ret has a sector size
3224 * aligned value (rounded down).
3226 *delalloc_end_ret = *delalloc_start_ret + delalloc_len - 1;
3228 if (*delalloc_start_ret == start) {
3229 /* Delalloc for the whole range, nothing more to do. */
3230 if (*delalloc_end_ret == end)
3232 /* Else trim our search range for ordered extents. */
3233 start = *delalloc_end_ret + 1;
3234 len = end + 1 - start;
3237 /* No delalloc, future calls don't need to search again. */
3238 *search_io_tree = false;
3242 * Now also check if there's any ordered extent in the range.
3243 * We do this because:
3245 * 1) When delalloc is flushed, the file range is locked, we clear the
3246 * EXTENT_DELALLOC bit from the io tree and create an extent map and
3247 * an ordered extent for the write. So we might just have been called
3248 * after delalloc is flushed and before the ordered extent completes
3249 * and inserts the new file extent item in the subvolume's btree;
3251 * 2) We may have an ordered extent created by flushing delalloc for a
3252 * subrange that starts before the subrange we found marked with
3253 * EXTENT_DELALLOC in the io tree.
3255 * We could also use the extent map tree to find such delalloc that is
3256 * being flushed, but using the ordered extents tree is more efficient
3257 * because it's usually much smaller as ordered extents are removed from
3258 * the tree once they complete. With the extent maps, we mau have them
3259 * in the extent map tree for a very long time, and they were either
3260 * created by previous writes or loaded by read operations.
3262 oe = btrfs_lookup_first_ordered_range(inode, start, len);
3264 return (delalloc_len > 0);
3266 /* The ordered extent may span beyond our search range. */
3267 oe_start = max(oe->file_offset, start);
3268 oe_end = min(oe->file_offset + oe->num_bytes - 1, end);
3270 btrfs_put_ordered_extent(oe);
3272 /* Don't have unflushed delalloc, return the ordered extent range. */
3273 if (delalloc_len == 0) {
3274 *delalloc_start_ret = oe_start;
3275 *delalloc_end_ret = oe_end;
3280 * We have both unflushed delalloc (io_tree) and an ordered extent.
3281 * If the ranges are adjacent returned a combined range, otherwise
3282 * return the leftmost range.
3284 if (oe_start < *delalloc_start_ret) {
3285 if (oe_end < *delalloc_start_ret)
3286 *delalloc_end_ret = oe_end;
3287 *delalloc_start_ret = oe_start;
3288 } else if (*delalloc_end_ret + 1 == oe_start) {
3289 *delalloc_end_ret = oe_end;
3296 * Check if there's delalloc in a given range.
3298 * @inode: The inode.
3299 * @start: The start offset of the range. It does not need to be
3300 * sector size aligned.
3301 * @end: The end offset (inclusive value) of the search range.
3302 * It does not need to be sector size aligned.
3303 * @cached_state: Extent state record used for speeding up delalloc
3304 * searches in the inode's io_tree. Can be NULL.
3305 * @delalloc_start_ret: Output argument, set to the start offset of the
3306 * subrange found with delalloc (may not be sector size
3308 * @delalloc_end_ret: Output argument, set to he end offset (inclusive value)
3309 * of the subrange found with delalloc.
3311 * Returns true if a subrange with delalloc is found within the given range, and
3312 * if so it sets @delalloc_start_ret and @delalloc_end_ret with the start and
3313 * end offsets of the subrange.
3315 bool btrfs_find_delalloc_in_range(struct btrfs_inode *inode, u64 start, u64 end,
3316 struct extent_state **cached_state,
3317 u64 *delalloc_start_ret, u64 *delalloc_end_ret)
3319 u64 cur_offset = round_down(start, inode->root->fs_info->sectorsize);
3320 u64 prev_delalloc_end = 0;
3321 bool search_io_tree = true;
3324 while (cur_offset <= end) {
3329 delalloc = find_delalloc_subrange(inode, cur_offset, end,
3330 cached_state, &search_io_tree,
3336 if (prev_delalloc_end == 0) {
3337 /* First subrange found. */
3338 *delalloc_start_ret = max(delalloc_start, start);
3339 *delalloc_end_ret = delalloc_end;
3341 } else if (delalloc_start == prev_delalloc_end + 1) {
3342 /* Subrange adjacent to the previous one, merge them. */
3343 *delalloc_end_ret = delalloc_end;
3345 /* Subrange not adjacent to the previous one, exit. */
3349 prev_delalloc_end = delalloc_end;
3350 cur_offset = delalloc_end + 1;
3358 * Check if there's a hole or delalloc range in a range representing a hole (or
3359 * prealloc extent) found in the inode's subvolume btree.
3361 * @inode: The inode.
3362 * @whence: Seek mode (SEEK_DATA or SEEK_HOLE).
3363 * @start: Start offset of the hole region. It does not need to be sector
3365 * @end: End offset (inclusive value) of the hole region. It does not
3366 * need to be sector size aligned.
3367 * @start_ret: Return parameter, used to set the start of the subrange in the
3368 * hole that matches the search criteria (seek mode), if such
3369 * subrange is found (return value of the function is true).
3370 * The value returned here may not be sector size aligned.
3372 * Returns true if a subrange matching the given seek mode is found, and if one
3373 * is found, it updates @start_ret with the start of the subrange.
3375 static bool find_desired_extent_in_hole(struct btrfs_inode *inode, int whence,
3376 struct extent_state **cached_state,
3377 u64 start, u64 end, u64 *start_ret)
3383 delalloc = btrfs_find_delalloc_in_range(inode, start, end, cached_state,
3384 &delalloc_start, &delalloc_end);
3385 if (delalloc && whence == SEEK_DATA) {
3386 *start_ret = delalloc_start;
3390 if (delalloc && whence == SEEK_HOLE) {
3392 * We found delalloc but it starts after out start offset. So we
3393 * have a hole between our start offset and the delalloc start.
3395 if (start < delalloc_start) {
3400 * Delalloc range starts at our start offset.
3401 * If the delalloc range's length is smaller than our range,
3402 * then it means we have a hole that starts where the delalloc
3405 if (delalloc_end < end) {
3406 *start_ret = delalloc_end + 1;
3410 /* There's delalloc for the whole range. */
3414 if (!delalloc && whence == SEEK_HOLE) {
3420 * No delalloc in the range and we are seeking for data. The caller has
3421 * to iterate to the next extent item in the subvolume btree.
3426 static loff_t find_desired_extent(struct file *file, loff_t offset, int whence)
3428 struct btrfs_inode *inode = BTRFS_I(file->f_mapping->host);
3429 struct btrfs_file_private *private = file->private_data;
3430 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3431 struct extent_state *cached_state = NULL;
3432 struct extent_state **delalloc_cached_state;
3433 const loff_t i_size = i_size_read(&inode->vfs_inode);
3434 const u64 ino = btrfs_ino(inode);
3435 struct btrfs_root *root = inode->root;
3436 struct btrfs_path *path;
3437 struct btrfs_key key;
3438 u64 last_extent_end;
3445 if (i_size == 0 || offset >= i_size)
3449 * Quick path. If the inode has no prealloc extents and its number of
3450 * bytes used matches its i_size, then it can not have holes.
3452 if (whence == SEEK_HOLE &&
3453 !(inode->flags & BTRFS_INODE_PREALLOC) &&
3454 inode_get_bytes(&inode->vfs_inode) == i_size)
3458 private = kzalloc(sizeof(*private), GFP_KERNEL);
3460 * No worries if memory allocation failed.
3461 * The private structure is used only for speeding up multiple
3462 * lseek SEEK_HOLE/DATA calls to a file when there's delalloc,
3463 * so everything will still be correct.
3465 file->private_data = private;
3469 delalloc_cached_state = &private->llseek_cached_state;
3471 delalloc_cached_state = NULL;
3474 * offset can be negative, in this case we start finding DATA/HOLE from
3475 * the very start of the file.
3477 start = max_t(loff_t, 0, offset);
3479 lockstart = round_down(start, fs_info->sectorsize);
3480 lockend = round_up(i_size, fs_info->sectorsize);
3481 if (lockend <= lockstart)
3482 lockend = lockstart + fs_info->sectorsize;
3485 path = btrfs_alloc_path();
3488 path->reada = READA_FORWARD;
3491 key.type = BTRFS_EXTENT_DATA_KEY;
3494 last_extent_end = lockstart;
3496 lock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
3498 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3501 } else if (ret > 0 && path->slots[0] > 0) {
3502 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
3503 if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY)
3507 while (start < i_size) {
3508 struct extent_buffer *leaf = path->nodes[0];
3509 struct btrfs_file_extent_item *extent;
3513 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3514 ret = btrfs_next_leaf(root, path);
3520 leaf = path->nodes[0];
3523 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3524 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
3527 extent_end = btrfs_file_extent_end(path);
3530 * In the first iteration we may have a slot that points to an
3531 * extent that ends before our start offset, so skip it.
3533 if (extent_end <= start) {
3538 /* We have an implicit hole, NO_HOLES feature is likely set. */
3539 if (last_extent_end < key.offset) {
3540 u64 search_start = last_extent_end;
3544 * First iteration, @start matches @offset and it's
3547 if (start == offset)
3548 search_start = offset;
3550 found = find_desired_extent_in_hole(inode, whence,
3551 delalloc_cached_state,
3556 start = found_start;
3560 * Didn't find data or a hole (due to delalloc) in the
3561 * implicit hole range, so need to analyze the extent.
3565 extent = btrfs_item_ptr(leaf, path->slots[0],
3566 struct btrfs_file_extent_item);
3567 type = btrfs_file_extent_type(leaf, extent);
3570 * Can't access the extent's disk_bytenr field if this is an
3571 * inline extent, since at that offset, it's where the extent
3574 if (type == BTRFS_FILE_EXTENT_PREALLOC ||
3575 (type == BTRFS_FILE_EXTENT_REG &&
3576 btrfs_file_extent_disk_bytenr(leaf, extent) == 0)) {
3578 * Explicit hole or prealloc extent, search for delalloc.
3579 * A prealloc extent is treated like a hole.
3581 u64 search_start = key.offset;
3585 * First iteration, @start matches @offset and it's
3588 if (start == offset)
3589 search_start = offset;
3591 found = find_desired_extent_in_hole(inode, whence,
3592 delalloc_cached_state,
3597 start = found_start;
3601 * Didn't find data or a hole (due to delalloc) in the
3602 * implicit hole range, so need to analyze the next
3607 * Found a regular or inline extent.
3608 * If we are seeking for data, adjust the start offset
3609 * and stop, we're done.
3611 if (whence == SEEK_DATA) {
3612 start = max_t(u64, key.offset, offset);
3617 * Else, we are seeking for a hole, check the next file
3623 last_extent_end = extent_end;
3625 if (fatal_signal_pending(current)) {
3632 /* We have an implicit hole from the last extent found up to i_size. */
3633 if (!found && start < i_size) {
3634 found = find_desired_extent_in_hole(inode, whence,
3635 delalloc_cached_state, start,
3636 i_size - 1, &start);
3642 unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
3643 btrfs_free_path(path);
3648 if (whence == SEEK_DATA && start >= i_size)
3651 return min_t(loff_t, start, i_size);
3654 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
3656 struct inode *inode = file->f_mapping->host;
3660 return generic_file_llseek(file, offset, whence);
3663 btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_SHARED);
3664 offset = find_desired_extent(file, offset, whence);
3665 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_SHARED);
3672 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
3675 static int btrfs_file_open(struct inode *inode, struct file *filp)
3679 filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC | FMODE_BUF_WASYNC |
3682 ret = fsverity_file_open(inode, filp);
3685 return generic_file_open(inode, filp);
3688 static int check_direct_read(struct btrfs_fs_info *fs_info,
3689 const struct iov_iter *iter, loff_t offset)
3694 ret = check_direct_IO(fs_info, iter, offset);
3698 if (!iter_is_iovec(iter))
3701 for (seg = 0; seg < iter->nr_segs; seg++) {
3702 for (i = seg + 1; i < iter->nr_segs; i++) {
3703 const struct iovec *iov1 = iter_iov(iter) + seg;
3704 const struct iovec *iov2 = iter_iov(iter) + i;
3706 if (iov1->iov_base == iov2->iov_base)
3713 static ssize_t btrfs_direct_read(struct kiocb *iocb, struct iov_iter *to)
3715 struct inode *inode = file_inode(iocb->ki_filp);
3716 size_t prev_left = 0;
3720 if (fsverity_active(inode))
3723 if (check_direct_read(btrfs_sb(inode->i_sb), to, iocb->ki_pos))
3726 btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_SHARED);
3729 * This is similar to what we do for direct IO writes, see the comment
3730 * at btrfs_direct_write(), but we also disable page faults in addition
3731 * to disabling them only at the iov_iter level. This is because when
3732 * reading from a hole or prealloc extent, iomap calls iov_iter_zero(),
3733 * which can still trigger page fault ins despite having set ->nofault
3734 * to true of our 'to' iov_iter.
3736 * The difference to direct IO writes is that we deadlock when trying
3737 * to lock the extent range in the inode's tree during he page reads
3738 * triggered by the fault in (while for writes it is due to waiting for
3739 * our own ordered extent). This is because for direct IO reads,
3740 * btrfs_dio_iomap_begin() returns with the extent range locked, which
3741 * is only unlocked in the endio callback (end_bio_extent_readpage()).
3743 pagefault_disable();
3745 ret = btrfs_dio_read(iocb, to, read);
3746 to->nofault = false;
3749 /* No increment (+=) because iomap returns a cumulative value. */
3753 if (iov_iter_count(to) > 0 && (ret == -EFAULT || ret > 0)) {
3754 const size_t left = iov_iter_count(to);
3756 if (left == prev_left) {
3758 * We didn't make any progress since the last attempt,
3759 * fallback to a buffered read for the remainder of the
3760 * range. This is just to avoid any possibility of looping
3766 * We made some progress since the last retry or this is
3767 * the first time we are retrying. Fault in as many pages
3768 * as possible and retry.
3770 fault_in_iov_iter_writeable(to, left);
3775 btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_SHARED);
3776 return ret < 0 ? ret : read;
3779 static ssize_t btrfs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
3783 if (iocb->ki_flags & IOCB_DIRECT) {
3784 ret = btrfs_direct_read(iocb, to);
3785 if (ret < 0 || !iov_iter_count(to) ||
3786 iocb->ki_pos >= i_size_read(file_inode(iocb->ki_filp)))
3790 return filemap_read(iocb, to, ret);
3793 const struct file_operations btrfs_file_operations = {
3794 .llseek = btrfs_file_llseek,
3795 .read_iter = btrfs_file_read_iter,
3796 .splice_read = filemap_splice_read,
3797 .write_iter = btrfs_file_write_iter,
3798 .splice_write = iter_file_splice_write,
3799 .mmap = btrfs_file_mmap,
3800 .open = btrfs_file_open,
3801 .release = btrfs_release_file,
3802 .get_unmapped_area = thp_get_unmapped_area,
3803 .fsync = btrfs_sync_file,
3804 .fallocate = btrfs_fallocate,
3805 .unlocked_ioctl = btrfs_ioctl,
3806 #ifdef CONFIG_COMPAT
3807 .compat_ioctl = btrfs_compat_ioctl,
3809 .remap_file_range = btrfs_remap_file_range,
3812 int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end)
3817 * So with compression we will find and lock a dirty page and clear the
3818 * first one as dirty, setup an async extent, and immediately return
3819 * with the entire range locked but with nobody actually marked with
3820 * writeback. So we can't just filemap_write_and_wait_range() and
3821 * expect it to work since it will just kick off a thread to do the
3822 * actual work. So we need to call filemap_fdatawrite_range _again_
3823 * since it will wait on the page lock, which won't be unlocked until
3824 * after the pages have been marked as writeback and so we're good to go
3825 * from there. We have to do this otherwise we'll miss the ordered
3826 * extents and that results in badness. Please Josef, do not think you
3827 * know better and pull this out at some point in the future, it is
3828 * right and you are wrong.
3830 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
3831 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
3832 &BTRFS_I(inode)->runtime_flags))
3833 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);