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 static struct kmem_cache *btrfs_inode_defrag_cachep;
36 * when auto defrag is enabled we
37 * queue up these defrag structs to remember which
38 * inodes need defragging passes
41 struct rb_node rb_node;
45 * transid where the defrag was added, we search for
46 * extents newer than this
54 * The extent size threshold for autodefrag.
56 * This value is different for compressed/non-compressed extents,
57 * thus needs to be passed from higher layer.
58 * (aka, inode_should_defrag())
63 static int __compare_inode_defrag(struct inode_defrag *defrag1,
64 struct inode_defrag *defrag2)
66 if (defrag1->root > defrag2->root)
68 else if (defrag1->root < defrag2->root)
70 else if (defrag1->ino > defrag2->ino)
72 else if (defrag1->ino < defrag2->ino)
78 /* pop a record for an inode into the defrag tree. The lock
79 * must be held already
81 * If you're inserting a record for an older transid than an
82 * existing record, the transid already in the tree is lowered
84 * If an existing record is found the defrag item you
87 static int __btrfs_add_inode_defrag(struct btrfs_inode *inode,
88 struct inode_defrag *defrag)
90 struct btrfs_fs_info *fs_info = inode->root->fs_info;
91 struct inode_defrag *entry;
93 struct rb_node *parent = NULL;
96 p = &fs_info->defrag_inodes.rb_node;
99 entry = rb_entry(parent, struct inode_defrag, rb_node);
101 ret = __compare_inode_defrag(defrag, entry);
103 p = &parent->rb_left;
105 p = &parent->rb_right;
107 /* if we're reinserting an entry for
108 * an old defrag run, make sure to
109 * lower the transid of our existing record
111 if (defrag->transid < entry->transid)
112 entry->transid = defrag->transid;
113 entry->extent_thresh = min(defrag->extent_thresh,
114 entry->extent_thresh);
118 set_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags);
119 rb_link_node(&defrag->rb_node, parent, p);
120 rb_insert_color(&defrag->rb_node, &fs_info->defrag_inodes);
124 static inline int __need_auto_defrag(struct btrfs_fs_info *fs_info)
126 if (!btrfs_test_opt(fs_info, AUTO_DEFRAG))
129 if (btrfs_fs_closing(fs_info))
136 * insert a defrag record for this inode if auto defrag is
139 int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
140 struct btrfs_inode *inode, u32 extent_thresh)
142 struct btrfs_root *root = inode->root;
143 struct btrfs_fs_info *fs_info = root->fs_info;
144 struct inode_defrag *defrag;
148 if (!__need_auto_defrag(fs_info))
151 if (test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags))
155 transid = trans->transid;
157 transid = inode->root->last_trans;
159 defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS);
163 defrag->ino = btrfs_ino(inode);
164 defrag->transid = transid;
165 defrag->root = root->root_key.objectid;
166 defrag->extent_thresh = extent_thresh;
168 spin_lock(&fs_info->defrag_inodes_lock);
169 if (!test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags)) {
171 * If we set IN_DEFRAG flag and evict the inode from memory,
172 * and then re-read this inode, this new inode doesn't have
173 * IN_DEFRAG flag. At the case, we may find the existed defrag.
175 ret = __btrfs_add_inode_defrag(inode, defrag);
177 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
179 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
181 spin_unlock(&fs_info->defrag_inodes_lock);
186 * pick the defragable inode that we want, if it doesn't exist, we will get
189 static struct inode_defrag *
190 btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino)
192 struct inode_defrag *entry = NULL;
193 struct inode_defrag tmp;
195 struct rb_node *parent = NULL;
201 spin_lock(&fs_info->defrag_inodes_lock);
202 p = fs_info->defrag_inodes.rb_node;
205 entry = rb_entry(parent, struct inode_defrag, rb_node);
207 ret = __compare_inode_defrag(&tmp, entry);
211 p = parent->rb_right;
216 if (parent && __compare_inode_defrag(&tmp, entry) > 0) {
217 parent = rb_next(parent);
219 entry = rb_entry(parent, struct inode_defrag, rb_node);
225 rb_erase(parent, &fs_info->defrag_inodes);
226 spin_unlock(&fs_info->defrag_inodes_lock);
230 void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info)
232 struct inode_defrag *defrag;
233 struct rb_node *node;
235 spin_lock(&fs_info->defrag_inodes_lock);
236 node = rb_first(&fs_info->defrag_inodes);
238 rb_erase(node, &fs_info->defrag_inodes);
239 defrag = rb_entry(node, struct inode_defrag, rb_node);
240 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
242 cond_resched_lock(&fs_info->defrag_inodes_lock);
244 node = rb_first(&fs_info->defrag_inodes);
246 spin_unlock(&fs_info->defrag_inodes_lock);
249 #define BTRFS_DEFRAG_BATCH 1024
251 static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info,
252 struct inode_defrag *defrag)
254 struct btrfs_root *inode_root;
256 struct btrfs_ioctl_defrag_range_args range;
261 if (test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state))
263 if (!__need_auto_defrag(fs_info))
267 inode_root = btrfs_get_fs_root(fs_info, defrag->root, true);
268 if (IS_ERR(inode_root)) {
269 ret = PTR_ERR(inode_root);
273 inode = btrfs_iget(fs_info->sb, defrag->ino, inode_root);
274 btrfs_put_root(inode_root);
276 ret = PTR_ERR(inode);
280 if (cur >= i_size_read(inode)) {
285 /* do a chunk of defrag */
286 clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
287 memset(&range, 0, sizeof(range));
290 range.extent_thresh = defrag->extent_thresh;
292 sb_start_write(fs_info->sb);
293 ret = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
295 sb_end_write(fs_info->sb);
301 cur = max(cur + fs_info->sectorsize, range.start);
305 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
310 * run through the list of inodes in the FS that need
313 int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
315 struct inode_defrag *defrag;
317 u64 root_objectid = 0;
319 atomic_inc(&fs_info->defrag_running);
321 /* Pause the auto defragger. */
322 if (test_bit(BTRFS_FS_STATE_REMOUNTING,
326 if (!__need_auto_defrag(fs_info))
329 /* find an inode to defrag */
330 defrag = btrfs_pick_defrag_inode(fs_info, root_objectid,
333 if (root_objectid || first_ino) {
342 first_ino = defrag->ino + 1;
343 root_objectid = defrag->root;
345 __btrfs_run_defrag_inode(fs_info, defrag);
347 atomic_dec(&fs_info->defrag_running);
350 * during unmount, we use the transaction_wait queue to
351 * wait for the defragger to stop
353 wake_up(&fs_info->transaction_wait);
357 /* simple helper to fault in pages and copy. This should go away
358 * and be replaced with calls into generic code.
360 static noinline int btrfs_copy_from_user(loff_t pos, size_t write_bytes,
361 struct page **prepared_pages,
365 size_t total_copied = 0;
367 int offset = offset_in_page(pos);
369 while (write_bytes > 0) {
370 size_t count = min_t(size_t,
371 PAGE_SIZE - offset, write_bytes);
372 struct page *page = prepared_pages[pg];
374 * Copy data from userspace to the current page
376 copied = copy_page_from_iter_atomic(page, offset, count, i);
378 /* Flush processor's dcache for this page */
379 flush_dcache_page(page);
382 * if we get a partial write, we can end up with
383 * partially up to date pages. These add
384 * a lot of complexity, so make sure they don't
385 * happen by forcing this copy to be retried.
387 * The rest of the btrfs_file_write code will fall
388 * back to page at a time copies after we return 0.
390 if (unlikely(copied < count)) {
391 if (!PageUptodate(page)) {
392 iov_iter_revert(i, copied);
399 write_bytes -= copied;
400 total_copied += copied;
402 if (offset == PAGE_SIZE) {
411 * unlocks pages after btrfs_file_write is done with them
413 static void btrfs_drop_pages(struct btrfs_fs_info *fs_info,
414 struct page **pages, size_t num_pages,
418 u64 block_start = round_down(pos, fs_info->sectorsize);
419 u64 block_len = round_up(pos + copied, fs_info->sectorsize) - block_start;
421 ASSERT(block_len <= U32_MAX);
422 for (i = 0; i < num_pages; i++) {
423 /* page checked is some magic around finding pages that
424 * have been modified without going through btrfs_set_page_dirty
425 * clear it here. There should be no need to mark the pages
426 * accessed as prepare_pages should have marked them accessed
427 * in prepare_pages via find_or_create_page()
429 btrfs_page_clamp_clear_checked(fs_info, pages[i], block_start,
431 unlock_page(pages[i]);
437 * After btrfs_copy_from_user(), update the following things for delalloc:
438 * - Mark newly dirtied pages as DELALLOC in the io tree.
439 * Used to advise which range is to be written back.
440 * - Mark modified pages as Uptodate/Dirty and not needing COW fixup
441 * - Update inode size for past EOF write
443 int btrfs_dirty_pages(struct btrfs_inode *inode, struct page **pages,
444 size_t num_pages, loff_t pos, size_t write_bytes,
445 struct extent_state **cached, bool noreserve)
447 struct btrfs_fs_info *fs_info = inode->root->fs_info;
452 u64 end_of_last_block;
453 u64 end_pos = pos + write_bytes;
454 loff_t isize = i_size_read(&inode->vfs_inode);
455 unsigned int extra_bits = 0;
457 if (write_bytes == 0)
461 extra_bits |= EXTENT_NORESERVE;
463 start_pos = round_down(pos, fs_info->sectorsize);
464 num_bytes = round_up(write_bytes + pos - start_pos,
465 fs_info->sectorsize);
466 ASSERT(num_bytes <= U32_MAX);
468 end_of_last_block = start_pos + num_bytes - 1;
471 * The pages may have already been dirty, clear out old accounting so
472 * we can set things up properly
474 clear_extent_bit(&inode->io_tree, start_pos, end_of_last_block,
475 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
478 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
483 for (i = 0; i < num_pages; i++) {
484 struct page *p = pages[i];
486 btrfs_page_clamp_set_uptodate(fs_info, p, start_pos, num_bytes);
487 btrfs_page_clamp_clear_checked(fs_info, p, start_pos, num_bytes);
488 btrfs_page_clamp_set_dirty(fs_info, p, start_pos, num_bytes);
492 * we've only changed i_size in ram, and we haven't updated
493 * the disk i_size. There is no need to log the inode
497 i_size_write(&inode->vfs_inode, end_pos);
502 * this drops all the extents in the cache that intersect the range
503 * [start, end]. Existing extents are split as required.
505 void btrfs_drop_extent_cache(struct btrfs_inode *inode, u64 start, u64 end,
508 struct extent_map *em;
509 struct extent_map *split = NULL;
510 struct extent_map *split2 = NULL;
511 struct extent_map_tree *em_tree = &inode->extent_tree;
512 u64 len = end - start + 1;
520 WARN_ON(end < start);
521 if (end == (u64)-1) {
530 split = alloc_extent_map();
532 split2 = alloc_extent_map();
533 if (!split || !split2)
536 write_lock(&em_tree->lock);
537 em = lookup_extent_mapping(em_tree, start, len);
539 write_unlock(&em_tree->lock);
543 gen = em->generation;
544 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
545 if (testend && em->start + em->len >= start + len) {
547 write_unlock(&em_tree->lock);
550 start = em->start + em->len;
552 len = start + len - (em->start + em->len);
554 write_unlock(&em_tree->lock);
557 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
558 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
559 clear_bit(EXTENT_FLAG_LOGGING, &flags);
560 modified = !list_empty(&em->list);
564 if (em->start < start) {
565 split->start = em->start;
566 split->len = start - em->start;
568 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
569 split->orig_start = em->orig_start;
570 split->block_start = em->block_start;
573 split->block_len = em->block_len;
575 split->block_len = split->len;
576 split->orig_block_len = max(split->block_len,
578 split->ram_bytes = em->ram_bytes;
580 split->orig_start = split->start;
581 split->block_len = 0;
582 split->block_start = em->block_start;
583 split->orig_block_len = 0;
584 split->ram_bytes = split->len;
587 split->generation = gen;
588 split->flags = flags;
589 split->compress_type = em->compress_type;
590 replace_extent_mapping(em_tree, em, split, modified);
591 free_extent_map(split);
595 if (testend && em->start + em->len > start + len) {
596 u64 diff = start + len - em->start;
598 split->start = start + len;
599 split->len = em->start + em->len - (start + len);
600 split->flags = flags;
601 split->compress_type = em->compress_type;
602 split->generation = gen;
604 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
605 split->orig_block_len = max(em->block_len,
608 split->ram_bytes = em->ram_bytes;
610 split->block_len = em->block_len;
611 split->block_start = em->block_start;
612 split->orig_start = em->orig_start;
614 split->block_len = split->len;
615 split->block_start = em->block_start
617 split->orig_start = em->orig_start;
620 split->ram_bytes = split->len;
621 split->orig_start = split->start;
622 split->block_len = 0;
623 split->block_start = em->block_start;
624 split->orig_block_len = 0;
627 if (extent_map_in_tree(em)) {
628 replace_extent_mapping(em_tree, em, split,
631 ret = add_extent_mapping(em_tree, split,
633 ASSERT(ret == 0); /* Logic error */
635 free_extent_map(split);
639 if (extent_map_in_tree(em))
640 remove_extent_mapping(em_tree, em);
641 write_unlock(&em_tree->lock);
645 /* once for the tree*/
649 free_extent_map(split);
651 free_extent_map(split2);
655 * this is very complex, but the basic idea is to drop all extents
656 * in the range start - end. hint_block is filled in with a block number
657 * that would be a good hint to the block allocator for this file.
659 * If an extent intersects the range but is not entirely inside the range
660 * it is either truncated or split. Anything entirely inside the range
661 * is deleted from the tree.
663 * Note: the VFS' inode number of bytes is not updated, it's up to the caller
664 * to deal with that. We set the field 'bytes_found' of the arguments structure
665 * with the number of allocated bytes found in the target range, so that the
666 * caller can update the inode's number of bytes in an atomic way when
667 * replacing extents in a range to avoid races with stat(2).
669 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
670 struct btrfs_root *root, struct btrfs_inode *inode,
671 struct btrfs_drop_extents_args *args)
673 struct btrfs_fs_info *fs_info = root->fs_info;
674 struct extent_buffer *leaf;
675 struct btrfs_file_extent_item *fi;
676 struct btrfs_ref ref = { 0 };
677 struct btrfs_key key;
678 struct btrfs_key new_key;
679 u64 ino = btrfs_ino(inode);
680 u64 search_start = args->start;
683 u64 extent_offset = 0;
685 u64 last_end = args->start;
691 int modify_tree = -1;
694 struct btrfs_path *path = args->path;
696 args->bytes_found = 0;
697 args->extent_inserted = false;
699 /* Must always have a path if ->replace_extent is true */
700 ASSERT(!(args->replace_extent && !args->path));
703 path = btrfs_alloc_path();
710 if (args->drop_cache)
711 btrfs_drop_extent_cache(inode, args->start, args->end - 1, 0);
713 if (args->start >= inode->disk_i_size && !args->replace_extent)
716 update_refs = (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID);
719 ret = btrfs_lookup_file_extent(trans, root, path, ino,
720 search_start, modify_tree);
723 if (ret > 0 && path->slots[0] > 0 && search_start == args->start) {
724 leaf = path->nodes[0];
725 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
726 if (key.objectid == ino &&
727 key.type == BTRFS_EXTENT_DATA_KEY)
732 leaf = path->nodes[0];
733 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
735 ret = btrfs_next_leaf(root, path);
742 leaf = path->nodes[0];
746 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
748 if (key.objectid > ino)
750 if (WARN_ON_ONCE(key.objectid < ino) ||
751 key.type < BTRFS_EXTENT_DATA_KEY) {
756 if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= args->end)
759 fi = btrfs_item_ptr(leaf, path->slots[0],
760 struct btrfs_file_extent_item);
761 extent_type = btrfs_file_extent_type(leaf, fi);
763 if (extent_type == BTRFS_FILE_EXTENT_REG ||
764 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
765 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
766 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
767 extent_offset = btrfs_file_extent_offset(leaf, fi);
768 extent_end = key.offset +
769 btrfs_file_extent_num_bytes(leaf, fi);
770 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
771 extent_end = key.offset +
772 btrfs_file_extent_ram_bytes(leaf, fi);
779 * Don't skip extent items representing 0 byte lengths. They
780 * used to be created (bug) if while punching holes we hit
781 * -ENOSPC condition. So if we find one here, just ensure we
782 * delete it, otherwise we would insert a new file extent item
783 * with the same key (offset) as that 0 bytes length file
784 * extent item in the call to setup_items_for_insert() later
787 if (extent_end == key.offset && extent_end >= search_start) {
788 last_end = extent_end;
789 goto delete_extent_item;
792 if (extent_end <= search_start) {
798 search_start = max(key.offset, args->start);
799 if (recow || !modify_tree) {
801 btrfs_release_path(path);
806 * | - range to drop - |
807 * | -------- extent -------- |
809 if (args->start > key.offset && args->end < extent_end) {
811 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
816 memcpy(&new_key, &key, sizeof(new_key));
817 new_key.offset = args->start;
818 ret = btrfs_duplicate_item(trans, root, path,
820 if (ret == -EAGAIN) {
821 btrfs_release_path(path);
827 leaf = path->nodes[0];
828 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
829 struct btrfs_file_extent_item);
830 btrfs_set_file_extent_num_bytes(leaf, fi,
831 args->start - key.offset);
833 fi = btrfs_item_ptr(leaf, path->slots[0],
834 struct btrfs_file_extent_item);
836 extent_offset += args->start - key.offset;
837 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
838 btrfs_set_file_extent_num_bytes(leaf, fi,
839 extent_end - args->start);
840 btrfs_mark_buffer_dirty(leaf);
842 if (update_refs && disk_bytenr > 0) {
843 btrfs_init_generic_ref(&ref,
844 BTRFS_ADD_DELAYED_REF,
845 disk_bytenr, num_bytes, 0);
846 btrfs_init_data_ref(&ref,
847 root->root_key.objectid,
849 args->start - extent_offset,
851 ret = btrfs_inc_extent_ref(trans, &ref);
852 BUG_ON(ret); /* -ENOMEM */
854 key.offset = args->start;
857 * From here on out we will have actually dropped something, so
858 * last_end can be updated.
860 last_end = extent_end;
863 * | ---- range to drop ----- |
864 * | -------- extent -------- |
866 if (args->start <= key.offset && args->end < extent_end) {
867 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
872 memcpy(&new_key, &key, sizeof(new_key));
873 new_key.offset = args->end;
874 btrfs_set_item_key_safe(fs_info, path, &new_key);
876 extent_offset += args->end - key.offset;
877 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
878 btrfs_set_file_extent_num_bytes(leaf, fi,
879 extent_end - args->end);
880 btrfs_mark_buffer_dirty(leaf);
881 if (update_refs && disk_bytenr > 0)
882 args->bytes_found += args->end - key.offset;
886 search_start = extent_end;
888 * | ---- range to drop ----- |
889 * | -------- extent -------- |
891 if (args->start > key.offset && args->end >= extent_end) {
893 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
898 btrfs_set_file_extent_num_bytes(leaf, fi,
899 args->start - key.offset);
900 btrfs_mark_buffer_dirty(leaf);
901 if (update_refs && disk_bytenr > 0)
902 args->bytes_found += extent_end - args->start;
903 if (args->end == extent_end)
911 * | ---- range to drop ----- |
912 * | ------ extent ------ |
914 if (args->start <= key.offset && args->end >= extent_end) {
917 del_slot = path->slots[0];
920 BUG_ON(del_slot + del_nr != path->slots[0]);
925 extent_type == BTRFS_FILE_EXTENT_INLINE) {
926 args->bytes_found += extent_end - key.offset;
927 extent_end = ALIGN(extent_end,
928 fs_info->sectorsize);
929 } else if (update_refs && disk_bytenr > 0) {
930 btrfs_init_generic_ref(&ref,
931 BTRFS_DROP_DELAYED_REF,
932 disk_bytenr, num_bytes, 0);
933 btrfs_init_data_ref(&ref,
934 root->root_key.objectid,
936 key.offset - extent_offset, 0,
938 ret = btrfs_free_extent(trans, &ref);
939 BUG_ON(ret); /* -ENOMEM */
940 args->bytes_found += extent_end - key.offset;
943 if (args->end == extent_end)
946 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
951 ret = btrfs_del_items(trans, root, path, del_slot,
954 btrfs_abort_transaction(trans, ret);
961 btrfs_release_path(path);
968 if (!ret && del_nr > 0) {
970 * Set path->slots[0] to first slot, so that after the delete
971 * if items are move off from our leaf to its immediate left or
972 * right neighbor leafs, we end up with a correct and adjusted
973 * path->slots[0] for our insertion (if args->replace_extent).
975 path->slots[0] = del_slot;
976 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
978 btrfs_abort_transaction(trans, ret);
981 leaf = path->nodes[0];
983 * If btrfs_del_items() was called, it might have deleted a leaf, in
984 * which case it unlocked our path, so check path->locks[0] matches a
987 if (!ret && args->replace_extent &&
988 path->locks[0] == BTRFS_WRITE_LOCK &&
989 btrfs_leaf_free_space(leaf) >=
990 sizeof(struct btrfs_item) + args->extent_item_size) {
993 key.type = BTRFS_EXTENT_DATA_KEY;
994 key.offset = args->start;
995 if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
996 struct btrfs_key slot_key;
998 btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
999 if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
1002 btrfs_setup_item_for_insert(root, path, &key, args->extent_item_size);
1003 args->extent_inserted = true;
1007 btrfs_free_path(path);
1008 else if (!args->extent_inserted)
1009 btrfs_release_path(path);
1011 args->drop_end = found ? min(args->end, last_end) : args->end;
1016 static int extent_mergeable(struct extent_buffer *leaf, int slot,
1017 u64 objectid, u64 bytenr, u64 orig_offset,
1018 u64 *start, u64 *end)
1020 struct btrfs_file_extent_item *fi;
1021 struct btrfs_key key;
1024 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1027 btrfs_item_key_to_cpu(leaf, &key, slot);
1028 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
1031 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1032 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
1033 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
1034 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
1035 btrfs_file_extent_compression(leaf, fi) ||
1036 btrfs_file_extent_encryption(leaf, fi) ||
1037 btrfs_file_extent_other_encoding(leaf, fi))
1040 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1041 if ((*start && *start != key.offset) || (*end && *end != extent_end))
1044 *start = key.offset;
1050 * Mark extent in the range start - end as written.
1052 * This changes extent type from 'pre-allocated' to 'regular'. If only
1053 * part of extent is marked as written, the extent will be split into
1056 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
1057 struct btrfs_inode *inode, u64 start, u64 end)
1059 struct btrfs_fs_info *fs_info = trans->fs_info;
1060 struct btrfs_root *root = inode->root;
1061 struct extent_buffer *leaf;
1062 struct btrfs_path *path;
1063 struct btrfs_file_extent_item *fi;
1064 struct btrfs_ref ref = { 0 };
1065 struct btrfs_key key;
1066 struct btrfs_key new_key;
1078 u64 ino = btrfs_ino(inode);
1080 path = btrfs_alloc_path();
1087 key.type = BTRFS_EXTENT_DATA_KEY;
1090 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1093 if (ret > 0 && path->slots[0] > 0)
1096 leaf = path->nodes[0];
1097 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1098 if (key.objectid != ino ||
1099 key.type != BTRFS_EXTENT_DATA_KEY) {
1101 btrfs_abort_transaction(trans, ret);
1104 fi = btrfs_item_ptr(leaf, path->slots[0],
1105 struct btrfs_file_extent_item);
1106 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_PREALLOC) {
1108 btrfs_abort_transaction(trans, ret);
1111 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1112 if (key.offset > start || extent_end < end) {
1114 btrfs_abort_transaction(trans, ret);
1118 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1119 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1120 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1121 memcpy(&new_key, &key, sizeof(new_key));
1123 if (start == key.offset && end < extent_end) {
1126 if (extent_mergeable(leaf, path->slots[0] - 1,
1127 ino, bytenr, orig_offset,
1128 &other_start, &other_end)) {
1129 new_key.offset = end;
1130 btrfs_set_item_key_safe(fs_info, path, &new_key);
1131 fi = btrfs_item_ptr(leaf, path->slots[0],
1132 struct btrfs_file_extent_item);
1133 btrfs_set_file_extent_generation(leaf, fi,
1135 btrfs_set_file_extent_num_bytes(leaf, fi,
1137 btrfs_set_file_extent_offset(leaf, fi,
1139 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1140 struct btrfs_file_extent_item);
1141 btrfs_set_file_extent_generation(leaf, fi,
1143 btrfs_set_file_extent_num_bytes(leaf, fi,
1145 btrfs_mark_buffer_dirty(leaf);
1150 if (start > key.offset && end == extent_end) {
1153 if (extent_mergeable(leaf, path->slots[0] + 1,
1154 ino, bytenr, orig_offset,
1155 &other_start, &other_end)) {
1156 fi = btrfs_item_ptr(leaf, path->slots[0],
1157 struct btrfs_file_extent_item);
1158 btrfs_set_file_extent_num_bytes(leaf, fi,
1159 start - key.offset);
1160 btrfs_set_file_extent_generation(leaf, fi,
1163 new_key.offset = start;
1164 btrfs_set_item_key_safe(fs_info, path, &new_key);
1166 fi = btrfs_item_ptr(leaf, path->slots[0],
1167 struct btrfs_file_extent_item);
1168 btrfs_set_file_extent_generation(leaf, fi,
1170 btrfs_set_file_extent_num_bytes(leaf, fi,
1172 btrfs_set_file_extent_offset(leaf, fi,
1173 start - orig_offset);
1174 btrfs_mark_buffer_dirty(leaf);
1179 while (start > key.offset || end < extent_end) {
1180 if (key.offset == start)
1183 new_key.offset = split;
1184 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1185 if (ret == -EAGAIN) {
1186 btrfs_release_path(path);
1190 btrfs_abort_transaction(trans, ret);
1194 leaf = path->nodes[0];
1195 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1196 struct btrfs_file_extent_item);
1197 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1198 btrfs_set_file_extent_num_bytes(leaf, fi,
1199 split - key.offset);
1201 fi = btrfs_item_ptr(leaf, path->slots[0],
1202 struct btrfs_file_extent_item);
1204 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1205 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1206 btrfs_set_file_extent_num_bytes(leaf, fi,
1207 extent_end - split);
1208 btrfs_mark_buffer_dirty(leaf);
1210 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, bytenr,
1212 btrfs_init_data_ref(&ref, root->root_key.objectid, ino,
1213 orig_offset, 0, false);
1214 ret = btrfs_inc_extent_ref(trans, &ref);
1216 btrfs_abort_transaction(trans, ret);
1220 if (split == start) {
1223 if (start != key.offset) {
1225 btrfs_abort_transaction(trans, ret);
1236 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
1238 btrfs_init_data_ref(&ref, root->root_key.objectid, ino, orig_offset,
1240 if (extent_mergeable(leaf, path->slots[0] + 1,
1241 ino, bytenr, orig_offset,
1242 &other_start, &other_end)) {
1244 btrfs_release_path(path);
1247 extent_end = other_end;
1248 del_slot = path->slots[0] + 1;
1250 ret = btrfs_free_extent(trans, &ref);
1252 btrfs_abort_transaction(trans, ret);
1258 if (extent_mergeable(leaf, path->slots[0] - 1,
1259 ino, bytenr, orig_offset,
1260 &other_start, &other_end)) {
1262 btrfs_release_path(path);
1265 key.offset = other_start;
1266 del_slot = path->slots[0];
1268 ret = btrfs_free_extent(trans, &ref);
1270 btrfs_abort_transaction(trans, ret);
1275 fi = btrfs_item_ptr(leaf, path->slots[0],
1276 struct btrfs_file_extent_item);
1277 btrfs_set_file_extent_type(leaf, fi,
1278 BTRFS_FILE_EXTENT_REG);
1279 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1280 btrfs_mark_buffer_dirty(leaf);
1282 fi = btrfs_item_ptr(leaf, del_slot - 1,
1283 struct btrfs_file_extent_item);
1284 btrfs_set_file_extent_type(leaf, fi,
1285 BTRFS_FILE_EXTENT_REG);
1286 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1287 btrfs_set_file_extent_num_bytes(leaf, fi,
1288 extent_end - key.offset);
1289 btrfs_mark_buffer_dirty(leaf);
1291 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1293 btrfs_abort_transaction(trans, ret);
1298 btrfs_free_path(path);
1303 * on error we return an unlocked page and the error value
1304 * on success we return a locked page and 0
1306 static int prepare_uptodate_page(struct inode *inode,
1307 struct page *page, u64 pos,
1308 bool force_uptodate)
1312 if (((pos & (PAGE_SIZE - 1)) || force_uptodate) &&
1313 !PageUptodate(page)) {
1314 ret = btrfs_readpage(NULL, page);
1318 if (!PageUptodate(page)) {
1324 * Since btrfs_readpage() will unlock the page before it
1325 * returns, there is a window where btrfs_releasepage() can be
1326 * called to release the page. Here we check both inode
1327 * mapping and PagePrivate() to make sure the page was not
1330 * The private flag check is essential for subpage as we need
1331 * to store extra bitmap using page->private.
1333 if (page->mapping != inode->i_mapping || !PagePrivate(page)) {
1342 * this just gets pages into the page cache and locks them down.
1344 static noinline int prepare_pages(struct inode *inode, struct page **pages,
1345 size_t num_pages, loff_t pos,
1346 size_t write_bytes, bool force_uptodate)
1349 unsigned long index = pos >> PAGE_SHIFT;
1350 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1354 for (i = 0; i < num_pages; i++) {
1356 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1357 mask | __GFP_WRITE);
1364 err = set_page_extent_mapped(pages[i]);
1371 err = prepare_uptodate_page(inode, pages[i], pos,
1373 if (!err && i == num_pages - 1)
1374 err = prepare_uptodate_page(inode, pages[i],
1375 pos + write_bytes, false);
1378 if (err == -EAGAIN) {
1385 wait_on_page_writeback(pages[i]);
1390 while (faili >= 0) {
1391 unlock_page(pages[faili]);
1392 put_page(pages[faili]);
1400 * This function locks the extent and properly waits for data=ordered extents
1401 * to finish before allowing the pages to be modified if need.
1404 * 1 - the extent is locked
1405 * 0 - the extent is not locked, and everything is OK
1406 * -EAGAIN - need re-prepare the pages
1407 * the other < 0 number - Something wrong happens
1410 lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct page **pages,
1411 size_t num_pages, loff_t pos,
1413 u64 *lockstart, u64 *lockend,
1414 struct extent_state **cached_state)
1416 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1422 start_pos = round_down(pos, fs_info->sectorsize);
1423 last_pos = round_up(pos + write_bytes, fs_info->sectorsize) - 1;
1425 if (start_pos < inode->vfs_inode.i_size) {
1426 struct btrfs_ordered_extent *ordered;
1428 lock_extent_bits(&inode->io_tree, start_pos, last_pos,
1430 ordered = btrfs_lookup_ordered_range(inode, start_pos,
1431 last_pos - start_pos + 1);
1433 ordered->file_offset + ordered->num_bytes > start_pos &&
1434 ordered->file_offset <= last_pos) {
1435 unlock_extent_cached(&inode->io_tree, start_pos,
1436 last_pos, cached_state);
1437 for (i = 0; i < num_pages; i++) {
1438 unlock_page(pages[i]);
1441 btrfs_start_ordered_extent(ordered, 1);
1442 btrfs_put_ordered_extent(ordered);
1446 btrfs_put_ordered_extent(ordered);
1448 *lockstart = start_pos;
1449 *lockend = last_pos;
1454 * We should be called after prepare_pages() which should have locked
1455 * all pages in the range.
1457 for (i = 0; i < num_pages; i++)
1458 WARN_ON(!PageLocked(pages[i]));
1464 * Check if we can do nocow write into the range [@pos, @pos + @write_bytes)
1466 * @pos: File offset.
1467 * @write_bytes: The length to write, will be updated to the nocow writeable
1470 * This function will flush ordered extents in the range to ensure proper
1474 * > 0 If we can nocow, and updates @write_bytes.
1475 * 0 If we can't do a nocow write.
1476 * -EAGAIN If we can't do a nocow write because snapshoting of the inode's
1477 * root is in progress.
1478 * < 0 If an error happened.
1480 * NOTE: Callers need to call btrfs_check_nocow_unlock() if we return > 0.
1482 int btrfs_check_nocow_lock(struct btrfs_inode *inode, loff_t pos,
1483 size_t *write_bytes)
1485 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1486 struct btrfs_root *root = inode->root;
1487 u64 lockstart, lockend;
1491 if (!(inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
1494 if (!btrfs_drew_try_write_lock(&root->snapshot_lock))
1497 lockstart = round_down(pos, fs_info->sectorsize);
1498 lockend = round_up(pos + *write_bytes,
1499 fs_info->sectorsize) - 1;
1500 num_bytes = lockend - lockstart + 1;
1502 btrfs_lock_and_flush_ordered_range(inode, lockstart, lockend, NULL);
1503 ret = can_nocow_extent(&inode->vfs_inode, lockstart, &num_bytes,
1504 NULL, NULL, NULL, false);
1507 btrfs_drew_write_unlock(&root->snapshot_lock);
1509 *write_bytes = min_t(size_t, *write_bytes ,
1510 num_bytes - pos + lockstart);
1512 unlock_extent(&inode->io_tree, lockstart, lockend);
1517 void btrfs_check_nocow_unlock(struct btrfs_inode *inode)
1519 btrfs_drew_write_unlock(&inode->root->snapshot_lock);
1522 static void update_time_for_write(struct inode *inode)
1524 struct timespec64 now;
1526 if (IS_NOCMTIME(inode))
1529 now = current_time(inode);
1530 if (!timespec64_equal(&inode->i_mtime, &now))
1531 inode->i_mtime = now;
1533 if (!timespec64_equal(&inode->i_ctime, &now))
1534 inode->i_ctime = now;
1536 if (IS_I_VERSION(inode))
1537 inode_inc_iversion(inode);
1540 static int btrfs_write_check(struct kiocb *iocb, struct iov_iter *from,
1543 struct file *file = iocb->ki_filp;
1544 struct inode *inode = file_inode(file);
1545 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1546 loff_t pos = iocb->ki_pos;
1552 * Quickly bail out on NOWAIT writes if we don't have the nodatacow or
1553 * prealloc flags, as without those flags we always have to COW. We will
1554 * later check if we can really COW into the target range (using
1555 * can_nocow_extent() at btrfs_get_blocks_direct_write()).
1557 if ((iocb->ki_flags & IOCB_NOWAIT) &&
1558 !(BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
1561 current->backing_dev_info = inode_to_bdi(inode);
1562 ret = file_remove_privs(file);
1567 * We reserve space for updating the inode when we reserve space for the
1568 * extent we are going to write, so we will enospc out there. We don't
1569 * need to start yet another transaction to update the inode as we will
1570 * update the inode when we finish writing whatever data we write.
1572 update_time_for_write(inode);
1574 start_pos = round_down(pos, fs_info->sectorsize);
1575 oldsize = i_size_read(inode);
1576 if (start_pos > oldsize) {
1577 /* Expand hole size to cover write data, preventing empty gap */
1578 loff_t end_pos = round_up(pos + count, fs_info->sectorsize);
1580 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, end_pos);
1582 current->backing_dev_info = NULL;
1590 static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb,
1593 struct file *file = iocb->ki_filp;
1595 struct inode *inode = file_inode(file);
1596 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1597 struct page **pages = NULL;
1598 struct extent_changeset *data_reserved = NULL;
1599 u64 release_bytes = 0;
1602 size_t num_written = 0;
1605 bool only_release_metadata = false;
1606 bool force_page_uptodate = false;
1607 loff_t old_isize = i_size_read(inode);
1608 unsigned int ilock_flags = 0;
1610 if (iocb->ki_flags & IOCB_NOWAIT)
1611 ilock_flags |= BTRFS_ILOCK_TRY;
1613 ret = btrfs_inode_lock(inode, ilock_flags);
1617 ret = generic_write_checks(iocb, i);
1621 ret = btrfs_write_check(iocb, i, ret);
1626 nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_SIZE),
1627 PAGE_SIZE / (sizeof(struct page *)));
1628 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1629 nrptrs = max(nrptrs, 8);
1630 pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL);
1636 while (iov_iter_count(i) > 0) {
1637 struct extent_state *cached_state = NULL;
1638 size_t offset = offset_in_page(pos);
1639 size_t sector_offset;
1640 size_t write_bytes = min(iov_iter_count(i),
1641 nrptrs * (size_t)PAGE_SIZE -
1644 size_t reserve_bytes;
1647 size_t dirty_sectors;
1652 * Fault pages before locking them in prepare_pages
1653 * to avoid recursive lock
1655 if (unlikely(fault_in_iov_iter_readable(i, write_bytes))) {
1660 only_release_metadata = false;
1661 sector_offset = pos & (fs_info->sectorsize - 1);
1663 extent_changeset_release(data_reserved);
1664 ret = btrfs_check_data_free_space(BTRFS_I(inode),
1665 &data_reserved, pos,
1669 * If we don't have to COW at the offset, reserve
1670 * metadata only. write_bytes may get smaller than
1673 if (btrfs_check_nocow_lock(BTRFS_I(inode), pos,
1675 only_release_metadata = true;
1680 num_pages = DIV_ROUND_UP(write_bytes + offset, PAGE_SIZE);
1681 WARN_ON(num_pages > nrptrs);
1682 reserve_bytes = round_up(write_bytes + sector_offset,
1683 fs_info->sectorsize);
1684 WARN_ON(reserve_bytes == 0);
1685 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
1687 reserve_bytes, false);
1689 if (!only_release_metadata)
1690 btrfs_free_reserved_data_space(BTRFS_I(inode),
1694 btrfs_check_nocow_unlock(BTRFS_I(inode));
1698 release_bytes = reserve_bytes;
1701 * This is going to setup the pages array with the number of
1702 * pages we want, so we don't really need to worry about the
1703 * contents of pages from loop to loop
1705 ret = prepare_pages(inode, pages, num_pages,
1707 force_page_uptodate);
1709 btrfs_delalloc_release_extents(BTRFS_I(inode),
1714 extents_locked = lock_and_cleanup_extent_if_need(
1715 BTRFS_I(inode), pages,
1716 num_pages, pos, write_bytes, &lockstart,
1717 &lockend, &cached_state);
1718 if (extents_locked < 0) {
1719 if (extents_locked == -EAGAIN)
1721 btrfs_delalloc_release_extents(BTRFS_I(inode),
1723 ret = extents_locked;
1727 copied = btrfs_copy_from_user(pos, write_bytes, pages, i);
1729 num_sectors = BTRFS_BYTES_TO_BLKS(fs_info, reserve_bytes);
1730 dirty_sectors = round_up(copied + sector_offset,
1731 fs_info->sectorsize);
1732 dirty_sectors = BTRFS_BYTES_TO_BLKS(fs_info, dirty_sectors);
1735 * if we have trouble faulting in the pages, fall
1736 * back to one page at a time
1738 if (copied < write_bytes)
1742 force_page_uptodate = true;
1746 force_page_uptodate = false;
1747 dirty_pages = DIV_ROUND_UP(copied + offset,
1751 if (num_sectors > dirty_sectors) {
1752 /* release everything except the sectors we dirtied */
1753 release_bytes -= dirty_sectors << fs_info->sectorsize_bits;
1754 if (only_release_metadata) {
1755 btrfs_delalloc_release_metadata(BTRFS_I(inode),
1756 release_bytes, true);
1760 __pos = round_down(pos,
1761 fs_info->sectorsize) +
1762 (dirty_pages << PAGE_SHIFT);
1763 btrfs_delalloc_release_space(BTRFS_I(inode),
1764 data_reserved, __pos,
1765 release_bytes, true);
1769 release_bytes = round_up(copied + sector_offset,
1770 fs_info->sectorsize);
1772 ret = btrfs_dirty_pages(BTRFS_I(inode), pages,
1773 dirty_pages, pos, copied,
1774 &cached_state, only_release_metadata);
1777 * If we have not locked the extent range, because the range's
1778 * start offset is >= i_size, we might still have a non-NULL
1779 * cached extent state, acquired while marking the extent range
1780 * as delalloc through btrfs_dirty_pages(). Therefore free any
1781 * possible cached extent state to avoid a memory leak.
1784 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1785 lockstart, lockend, &cached_state);
1787 free_extent_state(cached_state);
1789 btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes);
1791 btrfs_drop_pages(fs_info, pages, num_pages, pos, copied);
1796 if (only_release_metadata)
1797 btrfs_check_nocow_unlock(BTRFS_I(inode));
1799 btrfs_drop_pages(fs_info, pages, num_pages, pos, copied);
1803 balance_dirty_pages_ratelimited(inode->i_mapping);
1806 num_written += copied;
1811 if (release_bytes) {
1812 if (only_release_metadata) {
1813 btrfs_check_nocow_unlock(BTRFS_I(inode));
1814 btrfs_delalloc_release_metadata(BTRFS_I(inode),
1815 release_bytes, true);
1817 btrfs_delalloc_release_space(BTRFS_I(inode),
1819 round_down(pos, fs_info->sectorsize),
1820 release_bytes, true);
1824 extent_changeset_free(data_reserved);
1825 if (num_written > 0) {
1826 pagecache_isize_extended(inode, old_isize, iocb->ki_pos);
1827 iocb->ki_pos += num_written;
1830 btrfs_inode_unlock(inode, ilock_flags);
1831 return num_written ? num_written : ret;
1834 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
1835 const struct iov_iter *iter, loff_t offset)
1837 const u32 blocksize_mask = fs_info->sectorsize - 1;
1839 if (offset & blocksize_mask)
1842 if (iov_iter_alignment(iter) & blocksize_mask)
1848 static ssize_t btrfs_direct_write(struct kiocb *iocb, struct iov_iter *from)
1850 const bool is_sync_write = (iocb->ki_flags & IOCB_DSYNC);
1851 struct file *file = iocb->ki_filp;
1852 struct inode *inode = file_inode(file);
1853 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1855 ssize_t written = 0;
1856 ssize_t written_buffered;
1857 size_t prev_left = 0;
1860 unsigned int ilock_flags = 0;
1862 if (iocb->ki_flags & IOCB_NOWAIT)
1863 ilock_flags |= BTRFS_ILOCK_TRY;
1865 /* If the write DIO is within EOF, use a shared lock */
1866 if (iocb->ki_pos + iov_iter_count(from) <= i_size_read(inode))
1867 ilock_flags |= BTRFS_ILOCK_SHARED;
1870 err = btrfs_inode_lock(inode, ilock_flags);
1874 err = generic_write_checks(iocb, from);
1876 btrfs_inode_unlock(inode, ilock_flags);
1880 err = btrfs_write_check(iocb, from, err);
1882 btrfs_inode_unlock(inode, ilock_flags);
1888 * Re-check since file size may have changed just before taking the
1889 * lock or pos may have changed because of O_APPEND in generic_write_check()
1891 if ((ilock_flags & BTRFS_ILOCK_SHARED) &&
1892 pos + iov_iter_count(from) > i_size_read(inode)) {
1893 btrfs_inode_unlock(inode, ilock_flags);
1894 ilock_flags &= ~BTRFS_ILOCK_SHARED;
1898 if (check_direct_IO(fs_info, from, pos)) {
1899 btrfs_inode_unlock(inode, ilock_flags);
1904 * We remove IOCB_DSYNC so that we don't deadlock when iomap_dio_rw()
1905 * calls generic_write_sync() (through iomap_dio_complete()), because
1906 * that results in calling fsync (btrfs_sync_file()) which will try to
1907 * lock the inode in exclusive/write mode.
1910 iocb->ki_flags &= ~IOCB_DSYNC;
1913 * The iov_iter can be mapped to the same file range we are writing to.
1914 * If that's the case, then we will deadlock in the iomap code, because
1915 * it first calls our callback btrfs_dio_iomap_begin(), which will create
1916 * an ordered extent, and after that it will fault in the pages that the
1917 * iov_iter refers to. During the fault in we end up in the readahead
1918 * pages code (starting at btrfs_readahead()), which will lock the range,
1919 * find that ordered extent and then wait for it to complete (at
1920 * btrfs_lock_and_flush_ordered_range()), resulting in a deadlock since
1921 * obviously the ordered extent can never complete as we didn't submit
1922 * yet the respective bio(s). This always happens when the buffer is
1923 * memory mapped to the same file range, since the iomap DIO code always
1924 * invalidates pages in the target file range (after starting and waiting
1925 * for any writeback).
1927 * So here we disable page faults in the iov_iter and then retry if we
1928 * got -EFAULT, faulting in the pages before the retry.
1931 from->nofault = true;
1932 err = btrfs_dio_rw(iocb, from, written);
1933 from->nofault = false;
1935 /* No increment (+=) because iomap returns a cumulative value. */
1939 if (iov_iter_count(from) > 0 && (err == -EFAULT || err > 0)) {
1940 const size_t left = iov_iter_count(from);
1942 * We have more data left to write. Try to fault in as many as
1943 * possible of the remainder pages and retry. We do this without
1944 * releasing and locking again the inode, to prevent races with
1947 * Also, in case the iov refers to pages in the file range of the
1948 * file we want to write to (due to a mmap), we could enter an
1949 * infinite loop if we retry after faulting the pages in, since
1950 * iomap will invalidate any pages in the range early on, before
1951 * it tries to fault in the pages of the iov. So we keep track of
1952 * how much was left of iov in the previous EFAULT and fallback
1953 * to buffered IO in case we haven't made any progress.
1955 if (left == prev_left) {
1958 fault_in_iov_iter_readable(from, left);
1964 btrfs_inode_unlock(inode, ilock_flags);
1967 * Add back IOCB_DSYNC. Our caller, btrfs_file_write_iter(), will do
1968 * the fsync (call generic_write_sync()).
1971 iocb->ki_flags |= IOCB_DSYNC;
1973 /* If 'err' is -ENOTBLK then it means we must fallback to buffered IO. */
1974 if ((err < 0 && err != -ENOTBLK) || !iov_iter_count(from))
1979 written_buffered = btrfs_buffered_write(iocb, from);
1980 if (written_buffered < 0) {
1981 err = written_buffered;
1985 * Ensure all data is persisted. We want the next direct IO read to be
1986 * able to read what was just written.
1988 endbyte = pos + written_buffered - 1;
1989 err = btrfs_fdatawrite_range(inode, pos, endbyte);
1992 err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte);
1995 written += written_buffered;
1996 iocb->ki_pos = pos + written_buffered;
1997 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_SHIFT,
1998 endbyte >> PAGE_SHIFT);
2000 return err < 0 ? err : written;
2003 static ssize_t btrfs_encoded_write(struct kiocb *iocb, struct iov_iter *from,
2004 const struct btrfs_ioctl_encoded_io_args *encoded)
2006 struct file *file = iocb->ki_filp;
2007 struct inode *inode = file_inode(file);
2011 btrfs_inode_lock(inode, 0);
2012 count = encoded->len;
2013 ret = generic_write_checks_count(iocb, &count);
2014 if (ret == 0 && count != encoded->len) {
2016 * The write got truncated by generic_write_checks_count(). We
2017 * can't do a partial encoded write.
2021 if (ret || encoded->len == 0)
2024 ret = btrfs_write_check(iocb, from, encoded->len);
2028 ret = btrfs_do_encoded_write(iocb, from, encoded);
2030 btrfs_inode_unlock(inode, 0);
2034 ssize_t btrfs_do_write_iter(struct kiocb *iocb, struct iov_iter *from,
2035 const struct btrfs_ioctl_encoded_io_args *encoded)
2037 struct file *file = iocb->ki_filp;
2038 struct btrfs_inode *inode = BTRFS_I(file_inode(file));
2039 ssize_t num_written, num_sync;
2040 const bool sync = iocb->ki_flags & IOCB_DSYNC;
2043 * If the fs flips readonly due to some impossible error, although we
2044 * have opened a file as writable, we have to stop this write operation
2045 * to ensure consistency.
2047 if (BTRFS_FS_ERROR(inode->root->fs_info))
2050 if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT))
2054 atomic_inc(&inode->sync_writers);
2057 num_written = btrfs_encoded_write(iocb, from, encoded);
2058 num_sync = encoded->len;
2059 } else if (iocb->ki_flags & IOCB_DIRECT) {
2060 num_written = num_sync = btrfs_direct_write(iocb, from);
2062 num_written = num_sync = btrfs_buffered_write(iocb, from);
2065 btrfs_set_inode_last_sub_trans(inode);
2068 num_sync = generic_write_sync(iocb, num_sync);
2070 num_written = num_sync;
2074 atomic_dec(&inode->sync_writers);
2076 current->backing_dev_info = NULL;
2080 static ssize_t btrfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2082 return btrfs_do_write_iter(iocb, from, NULL);
2085 int btrfs_release_file(struct inode *inode, struct file *filp)
2087 struct btrfs_file_private *private = filp->private_data;
2089 if (private && private->filldir_buf)
2090 kfree(private->filldir_buf);
2092 filp->private_data = NULL;
2095 * Set by setattr when we are about to truncate a file from a non-zero
2096 * size to a zero size. This tries to flush down new bytes that may
2097 * have been written if the application were using truncate to replace
2100 if (test_and_clear_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
2101 &BTRFS_I(inode)->runtime_flags))
2102 filemap_flush(inode->i_mapping);
2106 static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end)
2109 struct blk_plug plug;
2112 * This is only called in fsync, which would do synchronous writes, so
2113 * a plug can merge adjacent IOs as much as possible. Esp. in case of
2114 * multiple disks using raid profile, a large IO can be split to
2115 * several segments of stripe length (currently 64K).
2117 blk_start_plug(&plug);
2118 atomic_inc(&BTRFS_I(inode)->sync_writers);
2119 ret = btrfs_fdatawrite_range(inode, start, end);
2120 atomic_dec(&BTRFS_I(inode)->sync_writers);
2121 blk_finish_plug(&plug);
2126 static inline bool skip_inode_logging(const struct btrfs_log_ctx *ctx)
2128 struct btrfs_inode *inode = BTRFS_I(ctx->inode);
2129 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2131 if (btrfs_inode_in_log(inode, fs_info->generation) &&
2132 list_empty(&ctx->ordered_extents))
2136 * If we are doing a fast fsync we can not bail out if the inode's
2137 * last_trans is <= then the last committed transaction, because we only
2138 * update the last_trans of the inode during ordered extent completion,
2139 * and for a fast fsync we don't wait for that, we only wait for the
2140 * writeback to complete.
2142 if (inode->last_trans <= fs_info->last_trans_committed &&
2143 (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) ||
2144 list_empty(&ctx->ordered_extents)))
2151 * fsync call for both files and directories. This logs the inode into
2152 * the tree log instead of forcing full commits whenever possible.
2154 * It needs to call filemap_fdatawait so that all ordered extent updates are
2155 * in the metadata btree are up to date for copying to the log.
2157 * It drops the inode mutex before doing the tree log commit. This is an
2158 * important optimization for directories because holding the mutex prevents
2159 * new operations on the dir while we write to disk.
2161 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
2163 struct dentry *dentry = file_dentry(file);
2164 struct inode *inode = d_inode(dentry);
2165 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2166 struct btrfs_root *root = BTRFS_I(inode)->root;
2167 struct btrfs_trans_handle *trans;
2168 struct btrfs_log_ctx ctx;
2173 trace_btrfs_sync_file(file, datasync);
2175 btrfs_init_log_ctx(&ctx, inode);
2178 * Always set the range to a full range, otherwise we can get into
2179 * several problems, from missing file extent items to represent holes
2180 * when not using the NO_HOLES feature, to log tree corruption due to
2181 * races between hole detection during logging and completion of ordered
2182 * extents outside the range, to missing checksums due to ordered extents
2183 * for which we flushed only a subset of their pages.
2187 len = (u64)LLONG_MAX + 1;
2190 * We write the dirty pages in the range and wait until they complete
2191 * out of the ->i_mutex. If so, we can flush the dirty pages by
2192 * multi-task, and make the performance up. See
2193 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
2195 ret = start_ordered_ops(inode, start, end);
2199 btrfs_inode_lock(inode, BTRFS_ILOCK_MMAP);
2201 atomic_inc(&root->log_batch);
2204 * Always check for the full sync flag while holding the inode's lock,
2205 * to avoid races with other tasks. The flag must be either set all the
2206 * time during logging or always off all the time while logging.
2208 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2209 &BTRFS_I(inode)->runtime_flags);
2212 * Before we acquired the inode's lock and the mmap lock, someone may
2213 * have dirtied more pages in the target range. We need to make sure
2214 * that writeback for any such pages does not start while we are logging
2215 * the inode, because if it does, any of the following might happen when
2216 * we are not doing a full inode sync:
2218 * 1) We log an extent after its writeback finishes but before its
2219 * checksums are added to the csum tree, leading to -EIO errors
2220 * when attempting to read the extent after a log replay.
2222 * 2) We can end up logging an extent before its writeback finishes.
2223 * Therefore after the log replay we will have a file extent item
2224 * pointing to an unwritten extent (and no data checksums as well).
2226 * So trigger writeback for any eventual new dirty pages and then we
2227 * wait for all ordered extents to complete below.
2229 ret = start_ordered_ops(inode, start, end);
2231 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
2236 * We have to do this here to avoid the priority inversion of waiting on
2237 * IO of a lower priority task while holding a transaction open.
2239 * For a full fsync we wait for the ordered extents to complete while
2240 * for a fast fsync we wait just for writeback to complete, and then
2241 * attach the ordered extents to the transaction so that a transaction
2242 * commit waits for their completion, to avoid data loss if we fsync,
2243 * the current transaction commits before the ordered extents complete
2244 * and a power failure happens right after that.
2246 * For zoned filesystem, if a write IO uses a ZONE_APPEND command, the
2247 * logical address recorded in the ordered extent may change. We need
2248 * to wait for the IO to stabilize the logical address.
2250 if (full_sync || btrfs_is_zoned(fs_info)) {
2251 ret = btrfs_wait_ordered_range(inode, start, len);
2254 * Get our ordered extents as soon as possible to avoid doing
2255 * checksum lookups in the csum tree, and use instead the
2256 * checksums attached to the ordered extents.
2258 btrfs_get_ordered_extents_for_logging(BTRFS_I(inode),
2259 &ctx.ordered_extents);
2260 ret = filemap_fdatawait_range(inode->i_mapping, start, end);
2264 goto out_release_extents;
2266 atomic_inc(&root->log_batch);
2269 if (skip_inode_logging(&ctx)) {
2271 * We've had everything committed since the last time we were
2272 * modified so clear this flag in case it was set for whatever
2273 * reason, it's no longer relevant.
2275 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2276 &BTRFS_I(inode)->runtime_flags);
2278 * An ordered extent might have started before and completed
2279 * already with io errors, in which case the inode was not
2280 * updated and we end up here. So check the inode's mapping
2281 * for any errors that might have happened since we last
2282 * checked called fsync.
2284 ret = filemap_check_wb_err(inode->i_mapping, file->f_wb_err);
2285 goto out_release_extents;
2289 * We use start here because we will need to wait on the IO to complete
2290 * in btrfs_sync_log, which could require joining a transaction (for
2291 * example checking cross references in the nocow path). If we use join
2292 * here we could get into a situation where we're waiting on IO to
2293 * happen that is blocked on a transaction trying to commit. With start
2294 * we inc the extwriter counter, so we wait for all extwriters to exit
2295 * before we start blocking joiners. This comment is to keep somebody
2296 * from thinking they are super smart and changing this to
2297 * btrfs_join_transaction *cough*Josef*cough*.
2299 trans = btrfs_start_transaction(root, 0);
2300 if (IS_ERR(trans)) {
2301 ret = PTR_ERR(trans);
2302 goto out_release_extents;
2304 trans->in_fsync = true;
2306 ret = btrfs_log_dentry_safe(trans, dentry, &ctx);
2307 btrfs_release_log_ctx_extents(&ctx);
2309 /* Fallthrough and commit/free transaction. */
2313 /* we've logged all the items and now have a consistent
2314 * version of the file in the log. It is possible that
2315 * someone will come in and modify the file, but that's
2316 * fine because the log is consistent on disk, and we
2317 * have references to all of the file's extents
2319 * It is possible that someone will come in and log the
2320 * file again, but that will end up using the synchronization
2321 * inside btrfs_sync_log to keep things safe.
2323 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
2325 if (ret != BTRFS_NO_LOG_SYNC) {
2327 ret = btrfs_sync_log(trans, root, &ctx);
2329 ret = btrfs_end_transaction(trans);
2334 ret = btrfs_wait_ordered_range(inode, start, len);
2336 btrfs_end_transaction(trans);
2340 ret = btrfs_commit_transaction(trans);
2342 ret = btrfs_end_transaction(trans);
2345 ASSERT(list_empty(&ctx.list));
2346 err = file_check_and_advance_wb_err(file);
2349 return ret > 0 ? -EIO : ret;
2351 out_release_extents:
2352 btrfs_release_log_ctx_extents(&ctx);
2353 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
2357 static const struct vm_operations_struct btrfs_file_vm_ops = {
2358 .fault = filemap_fault,
2359 .map_pages = filemap_map_pages,
2360 .page_mkwrite = btrfs_page_mkwrite,
2363 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
2365 struct address_space *mapping = filp->f_mapping;
2367 if (!mapping->a_ops->readpage)
2370 file_accessed(filp);
2371 vma->vm_ops = &btrfs_file_vm_ops;
2376 static int hole_mergeable(struct btrfs_inode *inode, struct extent_buffer *leaf,
2377 int slot, u64 start, u64 end)
2379 struct btrfs_file_extent_item *fi;
2380 struct btrfs_key key;
2382 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2385 btrfs_item_key_to_cpu(leaf, &key, slot);
2386 if (key.objectid != btrfs_ino(inode) ||
2387 key.type != BTRFS_EXTENT_DATA_KEY)
2390 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2392 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2395 if (btrfs_file_extent_disk_bytenr(leaf, fi))
2398 if (key.offset == end)
2400 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2405 static int fill_holes(struct btrfs_trans_handle *trans,
2406 struct btrfs_inode *inode,
2407 struct btrfs_path *path, u64 offset, u64 end)
2409 struct btrfs_fs_info *fs_info = trans->fs_info;
2410 struct btrfs_root *root = inode->root;
2411 struct extent_buffer *leaf;
2412 struct btrfs_file_extent_item *fi;
2413 struct extent_map *hole_em;
2414 struct extent_map_tree *em_tree = &inode->extent_tree;
2415 struct btrfs_key key;
2418 if (btrfs_fs_incompat(fs_info, NO_HOLES))
2421 key.objectid = btrfs_ino(inode);
2422 key.type = BTRFS_EXTENT_DATA_KEY;
2423 key.offset = offset;
2425 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2428 * We should have dropped this offset, so if we find it then
2429 * something has gone horribly wrong.
2436 leaf = path->nodes[0];
2437 if (hole_mergeable(inode, leaf, path->slots[0] - 1, offset, end)) {
2441 fi = btrfs_item_ptr(leaf, path->slots[0],
2442 struct btrfs_file_extent_item);
2443 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2445 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2446 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2447 btrfs_set_file_extent_offset(leaf, fi, 0);
2448 btrfs_mark_buffer_dirty(leaf);
2452 if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) {
2455 key.offset = offset;
2456 btrfs_set_item_key_safe(fs_info, path, &key);
2457 fi = btrfs_item_ptr(leaf, path->slots[0],
2458 struct btrfs_file_extent_item);
2459 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2461 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2462 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2463 btrfs_set_file_extent_offset(leaf, fi, 0);
2464 btrfs_mark_buffer_dirty(leaf);
2467 btrfs_release_path(path);
2469 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode),
2470 offset, 0, 0, end - offset, 0, end - offset, 0, 0, 0);
2475 btrfs_release_path(path);
2477 hole_em = alloc_extent_map();
2479 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2480 btrfs_set_inode_full_sync(inode);
2482 hole_em->start = offset;
2483 hole_em->len = end - offset;
2484 hole_em->ram_bytes = hole_em->len;
2485 hole_em->orig_start = offset;
2487 hole_em->block_start = EXTENT_MAP_HOLE;
2488 hole_em->block_len = 0;
2489 hole_em->orig_block_len = 0;
2490 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2491 hole_em->generation = trans->transid;
2494 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2495 write_lock(&em_tree->lock);
2496 ret = add_extent_mapping(em_tree, hole_em, 1);
2497 write_unlock(&em_tree->lock);
2498 } while (ret == -EEXIST);
2499 free_extent_map(hole_em);
2501 btrfs_set_inode_full_sync(inode);
2508 * Find a hole extent on given inode and change start/len to the end of hole
2509 * extent.(hole/vacuum extent whose em->start <= start &&
2510 * em->start + em->len > start)
2511 * When a hole extent is found, return 1 and modify start/len.
2513 static int find_first_non_hole(struct btrfs_inode *inode, u64 *start, u64 *len)
2515 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2516 struct extent_map *em;
2519 em = btrfs_get_extent(inode, NULL, 0,
2520 round_down(*start, fs_info->sectorsize),
2521 round_up(*len, fs_info->sectorsize));
2525 /* Hole or vacuum extent(only exists in no-hole mode) */
2526 if (em->block_start == EXTENT_MAP_HOLE) {
2528 *len = em->start + em->len > *start + *len ?
2529 0 : *start + *len - em->start - em->len;
2530 *start = em->start + em->len;
2532 free_extent_map(em);
2536 static void btrfs_punch_hole_lock_range(struct inode *inode,
2537 const u64 lockstart,
2539 struct extent_state **cached_state)
2542 * For subpage case, if the range is not at page boundary, we could
2543 * have pages at the leading/tailing part of the range.
2544 * This could lead to dead loop since filemap_range_has_page()
2545 * will always return true.
2546 * So here we need to do extra page alignment for
2547 * filemap_range_has_page().
2549 const u64 page_lockstart = round_up(lockstart, PAGE_SIZE);
2550 const u64 page_lockend = round_down(lockend + 1, PAGE_SIZE) - 1;
2553 truncate_pagecache_range(inode, lockstart, lockend);
2555 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2558 * We can't have ordered extents in the range, nor dirty/writeback
2559 * pages, because we have locked the inode's VFS lock in exclusive
2560 * mode, we have locked the inode's i_mmap_lock in exclusive mode,
2561 * we have flushed all delalloc in the range and we have waited
2562 * for any ordered extents in the range to complete.
2563 * We can race with anyone reading pages from this range, so after
2564 * locking the range check if we have pages in the range, and if
2565 * we do, unlock the range and retry.
2567 if (!filemap_range_has_page(inode->i_mapping, page_lockstart,
2571 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2572 lockend, cached_state);
2575 btrfs_assert_inode_range_clean(BTRFS_I(inode), lockstart, lockend);
2578 static int btrfs_insert_replace_extent(struct btrfs_trans_handle *trans,
2579 struct btrfs_inode *inode,
2580 struct btrfs_path *path,
2581 struct btrfs_replace_extent_info *extent_info,
2582 const u64 replace_len,
2583 const u64 bytes_to_drop)
2585 struct btrfs_fs_info *fs_info = trans->fs_info;
2586 struct btrfs_root *root = inode->root;
2587 struct btrfs_file_extent_item *extent;
2588 struct extent_buffer *leaf;
2589 struct btrfs_key key;
2591 struct btrfs_ref ref = { 0 };
2594 if (replace_len == 0)
2597 if (extent_info->disk_offset == 0 &&
2598 btrfs_fs_incompat(fs_info, NO_HOLES)) {
2599 btrfs_update_inode_bytes(inode, 0, bytes_to_drop);
2603 key.objectid = btrfs_ino(inode);
2604 key.type = BTRFS_EXTENT_DATA_KEY;
2605 key.offset = extent_info->file_offset;
2606 ret = btrfs_insert_empty_item(trans, root, path, &key,
2607 sizeof(struct btrfs_file_extent_item));
2610 leaf = path->nodes[0];
2611 slot = path->slots[0];
2612 write_extent_buffer(leaf, extent_info->extent_buf,
2613 btrfs_item_ptr_offset(leaf, slot),
2614 sizeof(struct btrfs_file_extent_item));
2615 extent = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2616 ASSERT(btrfs_file_extent_type(leaf, extent) != BTRFS_FILE_EXTENT_INLINE);
2617 btrfs_set_file_extent_offset(leaf, extent, extent_info->data_offset);
2618 btrfs_set_file_extent_num_bytes(leaf, extent, replace_len);
2619 if (extent_info->is_new_extent)
2620 btrfs_set_file_extent_generation(leaf, extent, trans->transid);
2621 btrfs_mark_buffer_dirty(leaf);
2622 btrfs_release_path(path);
2624 ret = btrfs_inode_set_file_extent_range(inode, extent_info->file_offset,
2629 /* If it's a hole, nothing more needs to be done. */
2630 if (extent_info->disk_offset == 0) {
2631 btrfs_update_inode_bytes(inode, 0, bytes_to_drop);
2635 btrfs_update_inode_bytes(inode, replace_len, bytes_to_drop);
2637 if (extent_info->is_new_extent && extent_info->insertions == 0) {
2638 key.objectid = extent_info->disk_offset;
2639 key.type = BTRFS_EXTENT_ITEM_KEY;
2640 key.offset = extent_info->disk_len;
2641 ret = btrfs_alloc_reserved_file_extent(trans, root,
2643 extent_info->file_offset,
2644 extent_info->qgroup_reserved,
2649 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
2650 extent_info->disk_offset,
2651 extent_info->disk_len, 0);
2652 ref_offset = extent_info->file_offset - extent_info->data_offset;
2653 btrfs_init_data_ref(&ref, root->root_key.objectid,
2654 btrfs_ino(inode), ref_offset, 0, false);
2655 ret = btrfs_inc_extent_ref(trans, &ref);
2658 extent_info->insertions++;
2664 * The respective range must have been previously locked, as well as the inode.
2665 * The end offset is inclusive (last byte of the range).
2666 * @extent_info is NULL for fallocate's hole punching and non-NULL when replacing
2667 * the file range with an extent.
2668 * When not punching a hole, we don't want to end up in a state where we dropped
2669 * extents without inserting a new one, so we must abort the transaction to avoid
2672 int btrfs_replace_file_extents(struct btrfs_inode *inode,
2673 struct btrfs_path *path, const u64 start,
2675 struct btrfs_replace_extent_info *extent_info,
2676 struct btrfs_trans_handle **trans_out)
2678 struct btrfs_drop_extents_args drop_args = { 0 };
2679 struct btrfs_root *root = inode->root;
2680 struct btrfs_fs_info *fs_info = root->fs_info;
2681 u64 min_size = btrfs_calc_insert_metadata_size(fs_info, 1);
2682 u64 ino_size = round_up(inode->vfs_inode.i_size, fs_info->sectorsize);
2683 struct btrfs_trans_handle *trans = NULL;
2684 struct btrfs_block_rsv *rsv;
2685 unsigned int rsv_count;
2687 u64 len = end - start;
2693 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
2698 rsv->size = btrfs_calc_insert_metadata_size(fs_info, 1);
2702 * 1 - update the inode
2703 * 1 - removing the extents in the range
2704 * 1 - adding the hole extent if no_holes isn't set or if we are
2705 * replacing the range with a new extent
2707 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || extent_info)
2712 trans = btrfs_start_transaction(root, rsv_count);
2713 if (IS_ERR(trans)) {
2714 ret = PTR_ERR(trans);
2719 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
2722 trans->block_rsv = rsv;
2725 drop_args.path = path;
2726 drop_args.end = end + 1;
2727 drop_args.drop_cache = true;
2728 while (cur_offset < end) {
2729 drop_args.start = cur_offset;
2730 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2731 /* If we are punching a hole decrement the inode's byte count */
2733 btrfs_update_inode_bytes(inode, 0,
2734 drop_args.bytes_found);
2735 if (ret != -ENOSPC) {
2737 * The only time we don't want to abort is if we are
2738 * attempting to clone a partial inline extent, in which
2739 * case we'll get EOPNOTSUPP. However if we aren't
2740 * clone we need to abort no matter what, because if we
2741 * got EOPNOTSUPP via prealloc then we messed up and
2745 (ret != -EOPNOTSUPP ||
2746 (extent_info && extent_info->is_new_extent)))
2747 btrfs_abort_transaction(trans, ret);
2751 trans->block_rsv = &fs_info->trans_block_rsv;
2753 if (!extent_info && cur_offset < drop_args.drop_end &&
2754 cur_offset < ino_size) {
2755 ret = fill_holes(trans, inode, path, cur_offset,
2756 drop_args.drop_end);
2759 * If we failed then we didn't insert our hole
2760 * entries for the area we dropped, so now the
2761 * fs is corrupted, so we must abort the
2764 btrfs_abort_transaction(trans, ret);
2767 } else if (!extent_info && cur_offset < drop_args.drop_end) {
2769 * We are past the i_size here, but since we didn't
2770 * insert holes we need to clear the mapped area so we
2771 * know to not set disk_i_size in this area until a new
2772 * file extent is inserted here.
2774 ret = btrfs_inode_clear_file_extent_range(inode,
2776 drop_args.drop_end - cur_offset);
2779 * We couldn't clear our area, so we could
2780 * presumably adjust up and corrupt the fs, so
2783 btrfs_abort_transaction(trans, ret);
2789 drop_args.drop_end > extent_info->file_offset) {
2790 u64 replace_len = drop_args.drop_end -
2791 extent_info->file_offset;
2793 ret = btrfs_insert_replace_extent(trans, inode, path,
2794 extent_info, replace_len,
2795 drop_args.bytes_found);
2797 btrfs_abort_transaction(trans, ret);
2800 extent_info->data_len -= replace_len;
2801 extent_info->data_offset += replace_len;
2802 extent_info->file_offset += replace_len;
2806 * We are releasing our handle on the transaction, balance the
2807 * dirty pages of the btree inode and flush delayed items, and
2808 * then get a new transaction handle, which may now point to a
2809 * new transaction in case someone else may have committed the
2810 * transaction we used to replace/drop file extent items. So
2811 * bump the inode's iversion and update mtime and ctime except
2812 * if we are called from a dedupe context. This is because a
2813 * power failure/crash may happen after the transaction is
2814 * committed and before we finish replacing/dropping all the
2815 * file extent items we need.
2817 inode_inc_iversion(&inode->vfs_inode);
2819 if (!extent_info || extent_info->update_times) {
2820 inode->vfs_inode.i_mtime = current_time(&inode->vfs_inode);
2821 inode->vfs_inode.i_ctime = inode->vfs_inode.i_mtime;
2824 ret = btrfs_update_inode(trans, root, inode);
2828 btrfs_end_transaction(trans);
2829 btrfs_btree_balance_dirty(fs_info);
2831 trans = btrfs_start_transaction(root, rsv_count);
2832 if (IS_ERR(trans)) {
2833 ret = PTR_ERR(trans);
2838 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
2839 rsv, min_size, false);
2840 BUG_ON(ret); /* shouldn't happen */
2841 trans->block_rsv = rsv;
2843 cur_offset = drop_args.drop_end;
2844 len = end - cur_offset;
2845 if (!extent_info && len) {
2846 ret = find_first_non_hole(inode, &cur_offset, &len);
2847 if (unlikely(ret < 0))
2857 * If we were cloning, force the next fsync to be a full one since we
2858 * we replaced (or just dropped in the case of cloning holes when
2859 * NO_HOLES is enabled) file extent items and did not setup new extent
2860 * maps for the replacement extents (or holes).
2862 if (extent_info && !extent_info->is_new_extent)
2863 btrfs_set_inode_full_sync(inode);
2868 trans->block_rsv = &fs_info->trans_block_rsv;
2870 * If we are using the NO_HOLES feature we might have had already an
2871 * hole that overlaps a part of the region [lockstart, lockend] and
2872 * ends at (or beyond) lockend. Since we have no file extent items to
2873 * represent holes, drop_end can be less than lockend and so we must
2874 * make sure we have an extent map representing the existing hole (the
2875 * call to __btrfs_drop_extents() might have dropped the existing extent
2876 * map representing the existing hole), otherwise the fast fsync path
2877 * will not record the existence of the hole region
2878 * [existing_hole_start, lockend].
2880 if (drop_args.drop_end <= end)
2881 drop_args.drop_end = end + 1;
2883 * Don't insert file hole extent item if it's for a range beyond eof
2884 * (because it's useless) or if it represents a 0 bytes range (when
2885 * cur_offset == drop_end).
2887 if (!extent_info && cur_offset < ino_size &&
2888 cur_offset < drop_args.drop_end) {
2889 ret = fill_holes(trans, inode, path, cur_offset,
2890 drop_args.drop_end);
2892 /* Same comment as above. */
2893 btrfs_abort_transaction(trans, ret);
2896 } else if (!extent_info && cur_offset < drop_args.drop_end) {
2897 /* See the comment in the loop above for the reasoning here. */
2898 ret = btrfs_inode_clear_file_extent_range(inode, cur_offset,
2899 drop_args.drop_end - cur_offset);
2901 btrfs_abort_transaction(trans, ret);
2907 ret = btrfs_insert_replace_extent(trans, inode, path,
2908 extent_info, extent_info->data_len,
2909 drop_args.bytes_found);
2911 btrfs_abort_transaction(trans, ret);
2920 trans->block_rsv = &fs_info->trans_block_rsv;
2922 btrfs_end_transaction(trans);
2926 btrfs_free_block_rsv(fs_info, rsv);
2931 static int btrfs_punch_hole(struct file *file, loff_t offset, loff_t len)
2933 struct inode *inode = file_inode(file);
2934 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2935 struct btrfs_root *root = BTRFS_I(inode)->root;
2936 struct extent_state *cached_state = NULL;
2937 struct btrfs_path *path;
2938 struct btrfs_trans_handle *trans = NULL;
2943 u64 orig_start = offset;
2947 bool truncated_block = false;
2948 bool updated_inode = false;
2950 btrfs_inode_lock(inode, BTRFS_ILOCK_MMAP);
2952 ret = btrfs_wait_ordered_range(inode, offset, len);
2954 goto out_only_mutex;
2956 ino_size = round_up(inode->i_size, fs_info->sectorsize);
2957 ret = find_first_non_hole(BTRFS_I(inode), &offset, &len);
2959 goto out_only_mutex;
2961 /* Already in a large hole */
2963 goto out_only_mutex;
2966 ret = file_modified(file);
2968 goto out_only_mutex;
2970 lockstart = round_up(offset, btrfs_inode_sectorsize(BTRFS_I(inode)));
2971 lockend = round_down(offset + len,
2972 btrfs_inode_sectorsize(BTRFS_I(inode))) - 1;
2973 same_block = (BTRFS_BYTES_TO_BLKS(fs_info, offset))
2974 == (BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1));
2976 * We needn't truncate any block which is beyond the end of the file
2977 * because we are sure there is no data there.
2980 * Only do this if we are in the same block and we aren't doing the
2983 if (same_block && len < fs_info->sectorsize) {
2984 if (offset < ino_size) {
2985 truncated_block = true;
2986 ret = btrfs_truncate_block(BTRFS_I(inode), offset, len,
2991 goto out_only_mutex;
2994 /* zero back part of the first block */
2995 if (offset < ino_size) {
2996 truncated_block = true;
2997 ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0);
2999 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
3004 /* Check the aligned pages after the first unaligned page,
3005 * if offset != orig_start, which means the first unaligned page
3006 * including several following pages are already in holes,
3007 * the extra check can be skipped */
3008 if (offset == orig_start) {
3009 /* after truncate page, check hole again */
3010 len = offset + len - lockstart;
3012 ret = find_first_non_hole(BTRFS_I(inode), &offset, &len);
3014 goto out_only_mutex;
3017 goto out_only_mutex;
3022 /* Check the tail unaligned part is in a hole */
3023 tail_start = lockend + 1;
3024 tail_len = offset + len - tail_start;
3026 ret = find_first_non_hole(BTRFS_I(inode), &tail_start, &tail_len);
3027 if (unlikely(ret < 0))
3028 goto out_only_mutex;
3030 /* zero the front end of the last page */
3031 if (tail_start + tail_len < ino_size) {
3032 truncated_block = true;
3033 ret = btrfs_truncate_block(BTRFS_I(inode),
3034 tail_start + tail_len,
3037 goto out_only_mutex;
3042 if (lockend < lockstart) {
3044 goto out_only_mutex;
3047 btrfs_punch_hole_lock_range(inode, lockstart, lockend, &cached_state);
3049 path = btrfs_alloc_path();
3055 ret = btrfs_replace_file_extents(BTRFS_I(inode), path, lockstart,
3056 lockend, NULL, &trans);
3057 btrfs_free_path(path);
3061 ASSERT(trans != NULL);
3062 inode_inc_iversion(inode);
3063 inode->i_mtime = inode->i_ctime = current_time(inode);
3064 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
3065 updated_inode = true;
3066 btrfs_end_transaction(trans);
3067 btrfs_btree_balance_dirty(fs_info);
3069 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
3072 if (!updated_inode && truncated_block && !ret) {
3074 * If we only end up zeroing part of a page, we still need to
3075 * update the inode item, so that all the time fields are
3076 * updated as well as the necessary btrfs inode in memory fields
3077 * for detecting, at fsync time, if the inode isn't yet in the
3078 * log tree or it's there but not up to date.
3080 struct timespec64 now = current_time(inode);
3082 inode_inc_iversion(inode);
3083 inode->i_mtime = now;
3084 inode->i_ctime = now;
3085 trans = btrfs_start_transaction(root, 1);
3086 if (IS_ERR(trans)) {
3087 ret = PTR_ERR(trans);
3091 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
3092 ret2 = btrfs_end_transaction(trans);
3097 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
3101 /* Helper structure to record which range is already reserved */
3102 struct falloc_range {
3103 struct list_head list;
3109 * Helper function to add falloc range
3111 * Caller should have locked the larger range of extent containing
3114 static int add_falloc_range(struct list_head *head, u64 start, u64 len)
3116 struct falloc_range *range = NULL;
3118 if (!list_empty(head)) {
3120 * As fallocate iterates by bytenr order, we only need to check
3123 range = list_last_entry(head, struct falloc_range, list);
3124 if (range->start + range->len == start) {
3130 range = kmalloc(sizeof(*range), GFP_KERNEL);
3133 range->start = start;
3135 list_add_tail(&range->list, head);
3139 static int btrfs_fallocate_update_isize(struct inode *inode,
3143 struct btrfs_trans_handle *trans;
3144 struct btrfs_root *root = BTRFS_I(inode)->root;
3148 if (mode & FALLOC_FL_KEEP_SIZE || end <= i_size_read(inode))
3151 trans = btrfs_start_transaction(root, 1);
3153 return PTR_ERR(trans);
3155 inode->i_ctime = current_time(inode);
3156 i_size_write(inode, end);
3157 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
3158 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
3159 ret2 = btrfs_end_transaction(trans);
3161 return ret ? ret : ret2;
3165 RANGE_BOUNDARY_WRITTEN_EXTENT,
3166 RANGE_BOUNDARY_PREALLOC_EXTENT,
3167 RANGE_BOUNDARY_HOLE,
3170 static int btrfs_zero_range_check_range_boundary(struct btrfs_inode *inode,
3173 const u64 sectorsize = btrfs_inode_sectorsize(inode);
3174 struct extent_map *em;
3177 offset = round_down(offset, sectorsize);
3178 em = btrfs_get_extent(inode, NULL, 0, offset, sectorsize);
3182 if (em->block_start == EXTENT_MAP_HOLE)
3183 ret = RANGE_BOUNDARY_HOLE;
3184 else if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
3185 ret = RANGE_BOUNDARY_PREALLOC_EXTENT;
3187 ret = RANGE_BOUNDARY_WRITTEN_EXTENT;
3189 free_extent_map(em);
3193 static int btrfs_zero_range(struct inode *inode,
3198 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3199 struct extent_map *em;
3200 struct extent_changeset *data_reserved = NULL;
3203 const u64 sectorsize = btrfs_inode_sectorsize(BTRFS_I(inode));
3204 u64 alloc_start = round_down(offset, sectorsize);
3205 u64 alloc_end = round_up(offset + len, sectorsize);
3206 u64 bytes_to_reserve = 0;
3207 bool space_reserved = false;
3209 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start,
3210 alloc_end - alloc_start);
3217 * Avoid hole punching and extent allocation for some cases. More cases
3218 * could be considered, but these are unlikely common and we keep things
3219 * as simple as possible for now. Also, intentionally, if the target
3220 * range contains one or more prealloc extents together with regular
3221 * extents and holes, we drop all the existing extents and allocate a
3222 * new prealloc extent, so that we get a larger contiguous disk extent.
3224 if (em->start <= alloc_start &&
3225 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3226 const u64 em_end = em->start + em->len;
3228 if (em_end >= offset + len) {
3230 * The whole range is already a prealloc extent,
3231 * do nothing except updating the inode's i_size if
3234 free_extent_map(em);
3235 ret = btrfs_fallocate_update_isize(inode, offset + len,
3240 * Part of the range is already a prealloc extent, so operate
3241 * only on the remaining part of the range.
3243 alloc_start = em_end;
3244 ASSERT(IS_ALIGNED(alloc_start, sectorsize));
3245 len = offset + len - alloc_start;
3246 offset = alloc_start;
3247 alloc_hint = em->block_start + em->len;
3249 free_extent_map(em);
3251 if (BTRFS_BYTES_TO_BLKS(fs_info, offset) ==
3252 BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)) {
3253 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start,
3260 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3261 free_extent_map(em);
3262 ret = btrfs_fallocate_update_isize(inode, offset + len,
3266 if (len < sectorsize && em->block_start != EXTENT_MAP_HOLE) {
3267 free_extent_map(em);
3268 ret = btrfs_truncate_block(BTRFS_I(inode), offset, len,
3271 ret = btrfs_fallocate_update_isize(inode,
3276 free_extent_map(em);
3277 alloc_start = round_down(offset, sectorsize);
3278 alloc_end = alloc_start + sectorsize;
3282 alloc_start = round_up(offset, sectorsize);
3283 alloc_end = round_down(offset + len, sectorsize);
3286 * For unaligned ranges, check the pages at the boundaries, they might
3287 * map to an extent, in which case we need to partially zero them, or
3288 * they might map to a hole, in which case we need our allocation range
3291 if (!IS_ALIGNED(offset, sectorsize)) {
3292 ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode),
3296 if (ret == RANGE_BOUNDARY_HOLE) {
3297 alloc_start = round_down(offset, sectorsize);
3299 } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
3300 ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0);
3308 if (!IS_ALIGNED(offset + len, sectorsize)) {
3309 ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode),
3313 if (ret == RANGE_BOUNDARY_HOLE) {
3314 alloc_end = round_up(offset + len, sectorsize);
3316 } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
3317 ret = btrfs_truncate_block(BTRFS_I(inode), offset + len,
3327 if (alloc_start < alloc_end) {
3328 struct extent_state *cached_state = NULL;
3329 const u64 lockstart = alloc_start;
3330 const u64 lockend = alloc_end - 1;
3332 bytes_to_reserve = alloc_end - alloc_start;
3333 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
3337 space_reserved = true;
3338 btrfs_punch_hole_lock_range(inode, lockstart, lockend,
3340 ret = btrfs_qgroup_reserve_data(BTRFS_I(inode), &data_reserved,
3341 alloc_start, bytes_to_reserve);
3343 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
3344 lockend, &cached_state);
3347 ret = btrfs_prealloc_file_range(inode, mode, alloc_start,
3348 alloc_end - alloc_start,
3350 offset + len, &alloc_hint);
3351 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
3352 lockend, &cached_state);
3353 /* btrfs_prealloc_file_range releases reserved space on error */
3355 space_reserved = false;
3359 ret = btrfs_fallocate_update_isize(inode, offset + len, mode);
3361 if (ret && space_reserved)
3362 btrfs_free_reserved_data_space(BTRFS_I(inode), data_reserved,
3363 alloc_start, bytes_to_reserve);
3364 extent_changeset_free(data_reserved);
3369 static long btrfs_fallocate(struct file *file, int mode,
3370 loff_t offset, loff_t len)
3372 struct inode *inode = file_inode(file);
3373 struct extent_state *cached_state = NULL;
3374 struct extent_changeset *data_reserved = NULL;
3375 struct falloc_range *range;
3376 struct falloc_range *tmp;
3377 struct list_head reserve_list;
3385 u64 data_space_needed = 0;
3386 u64 data_space_reserved = 0;
3387 u64 qgroup_reserved = 0;
3388 struct extent_map *em;
3389 int blocksize = btrfs_inode_sectorsize(BTRFS_I(inode));
3392 /* Do not allow fallocate in ZONED mode */
3393 if (btrfs_is_zoned(btrfs_sb(inode->i_sb)))
3396 alloc_start = round_down(offset, blocksize);
3397 alloc_end = round_up(offset + len, blocksize);
3398 cur_offset = alloc_start;
3400 /* Make sure we aren't being give some crap mode */
3401 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
3402 FALLOC_FL_ZERO_RANGE))
3405 if (mode & FALLOC_FL_PUNCH_HOLE)
3406 return btrfs_punch_hole(file, offset, len);
3408 btrfs_inode_lock(inode, BTRFS_ILOCK_MMAP);
3410 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) {
3411 ret = inode_newsize_ok(inode, offset + len);
3416 ret = file_modified(file);
3421 * TODO: Move these two operations after we have checked
3422 * accurate reserved space, or fallocate can still fail but
3423 * with page truncated or size expanded.
3425 * But that's a minor problem and won't do much harm BTW.
3427 if (alloc_start > inode->i_size) {
3428 ret = btrfs_cont_expand(BTRFS_I(inode), i_size_read(inode),
3432 } else if (offset + len > inode->i_size) {
3434 * If we are fallocating from the end of the file onward we
3435 * need to zero out the end of the block if i_size lands in the
3436 * middle of a block.
3438 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
3444 * We have locked the inode at the VFS level (in exclusive mode) and we
3445 * have locked the i_mmap_lock lock (in exclusive mode). Now before
3446 * locking the file range, flush all dealloc in the range and wait for
3447 * all ordered extents in the range to complete. After this we can lock
3448 * the file range and, due to the previous locking we did, we know there
3449 * can't be more delalloc or ordered extents in the range.
3451 ret = btrfs_wait_ordered_range(inode, alloc_start,
3452 alloc_end - alloc_start);
3456 if (mode & FALLOC_FL_ZERO_RANGE) {
3457 ret = btrfs_zero_range(inode, offset, len, mode);
3458 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
3462 locked_end = alloc_end - 1;
3463 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
3466 btrfs_assert_inode_range_clean(BTRFS_I(inode), alloc_start, locked_end);
3468 /* First, check if we exceed the qgroup limit */
3469 INIT_LIST_HEAD(&reserve_list);
3470 while (cur_offset < alloc_end) {
3471 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
3472 alloc_end - cur_offset);
3477 last_byte = min(extent_map_end(em), alloc_end);
3478 actual_end = min_t(u64, extent_map_end(em), offset + len);
3479 last_byte = ALIGN(last_byte, blocksize);
3480 if (em->block_start == EXTENT_MAP_HOLE ||
3481 (cur_offset >= inode->i_size &&
3482 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
3483 const u64 range_len = last_byte - cur_offset;
3485 ret = add_falloc_range(&reserve_list, cur_offset, range_len);
3487 free_extent_map(em);
3490 ret = btrfs_qgroup_reserve_data(BTRFS_I(inode),
3491 &data_reserved, cur_offset, range_len);
3493 free_extent_map(em);
3496 qgroup_reserved += range_len;
3497 data_space_needed += range_len;
3499 free_extent_map(em);
3500 cur_offset = last_byte;
3503 if (!ret && data_space_needed > 0) {
3505 * We are safe to reserve space here as we can't have delalloc
3506 * in the range, see above.
3508 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
3511 data_space_reserved = data_space_needed;
3515 * If ret is still 0, means we're OK to fallocate.
3516 * Or just cleanup the list and exit.
3518 list_for_each_entry_safe(range, tmp, &reserve_list, list) {
3520 ret = btrfs_prealloc_file_range(inode, mode,
3522 range->len, i_blocksize(inode),
3523 offset + len, &alloc_hint);
3525 * btrfs_prealloc_file_range() releases space even
3526 * if it returns an error.
3528 data_space_reserved -= range->len;
3529 qgroup_reserved -= range->len;
3530 } else if (data_space_reserved > 0) {
3531 btrfs_free_reserved_data_space(BTRFS_I(inode),
3532 data_reserved, range->start,
3534 data_space_reserved -= range->len;
3535 qgroup_reserved -= range->len;
3536 } else if (qgroup_reserved > 0) {
3537 btrfs_qgroup_free_data(BTRFS_I(inode), data_reserved,
3538 range->start, range->len);
3539 qgroup_reserved -= range->len;
3541 list_del(&range->list);
3548 * We didn't need to allocate any more space, but we still extended the
3549 * size of the file so we need to update i_size and the inode item.
3551 ret = btrfs_fallocate_update_isize(inode, actual_end, mode);
3553 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
3556 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
3557 extent_changeset_free(data_reserved);
3561 static loff_t find_desired_extent(struct btrfs_inode *inode, loff_t offset,
3564 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3565 struct extent_map *em = NULL;
3566 struct extent_state *cached_state = NULL;
3567 loff_t i_size = inode->vfs_inode.i_size;
3574 if (i_size == 0 || offset >= i_size)
3578 * offset can be negative, in this case we start finding DATA/HOLE from
3579 * the very start of the file.
3581 start = max_t(loff_t, 0, offset);
3583 lockstart = round_down(start, fs_info->sectorsize);
3584 lockend = round_up(i_size, fs_info->sectorsize);
3585 if (lockend <= lockstart)
3586 lockend = lockstart + fs_info->sectorsize;
3588 len = lockend - lockstart + 1;
3590 lock_extent_bits(&inode->io_tree, lockstart, lockend, &cached_state);
3592 while (start < i_size) {
3593 em = btrfs_get_extent_fiemap(inode, start, len);
3600 if (whence == SEEK_HOLE &&
3601 (em->block_start == EXTENT_MAP_HOLE ||
3602 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
3604 else if (whence == SEEK_DATA &&
3605 (em->block_start != EXTENT_MAP_HOLE &&
3606 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
3609 start = em->start + em->len;
3610 free_extent_map(em);
3614 free_extent_map(em);
3615 unlock_extent_cached(&inode->io_tree, lockstart, lockend,
3620 if (whence == SEEK_DATA && start >= i_size)
3623 offset = min_t(loff_t, start, i_size);
3629 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
3631 struct inode *inode = file->f_mapping->host;
3635 return generic_file_llseek(file, offset, whence);
3638 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
3639 offset = find_desired_extent(BTRFS_I(inode), offset, whence);
3640 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
3647 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
3650 static int btrfs_file_open(struct inode *inode, struct file *filp)
3654 filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
3656 ret = fsverity_file_open(inode, filp);
3659 return generic_file_open(inode, filp);
3662 static int check_direct_read(struct btrfs_fs_info *fs_info,
3663 const struct iov_iter *iter, loff_t offset)
3668 ret = check_direct_IO(fs_info, iter, offset);
3672 if (!iter_is_iovec(iter))
3675 for (seg = 0; seg < iter->nr_segs; seg++)
3676 for (i = seg + 1; i < iter->nr_segs; i++)
3677 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
3682 static ssize_t btrfs_direct_read(struct kiocb *iocb, struct iov_iter *to)
3684 struct inode *inode = file_inode(iocb->ki_filp);
3685 size_t prev_left = 0;
3689 if (fsverity_active(inode))
3692 if (check_direct_read(btrfs_sb(inode->i_sb), to, iocb->ki_pos))
3695 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
3698 * This is similar to what we do for direct IO writes, see the comment
3699 * at btrfs_direct_write(), but we also disable page faults in addition
3700 * to disabling them only at the iov_iter level. This is because when
3701 * reading from a hole or prealloc extent, iomap calls iov_iter_zero(),
3702 * which can still trigger page fault ins despite having set ->nofault
3703 * to true of our 'to' iov_iter.
3705 * The difference to direct IO writes is that we deadlock when trying
3706 * to lock the extent range in the inode's tree during he page reads
3707 * triggered by the fault in (while for writes it is due to waiting for
3708 * our own ordered extent). This is because for direct IO reads,
3709 * btrfs_dio_iomap_begin() returns with the extent range locked, which
3710 * is only unlocked in the endio callback (end_bio_extent_readpage()).
3712 pagefault_disable();
3714 ret = btrfs_dio_rw(iocb, to, read);
3715 to->nofault = false;
3718 /* No increment (+=) because iomap returns a cumulative value. */
3722 if (iov_iter_count(to) > 0 && (ret == -EFAULT || ret > 0)) {
3723 const size_t left = iov_iter_count(to);
3725 if (left == prev_left) {
3727 * We didn't make any progress since the last attempt,
3728 * fallback to a buffered read for the remainder of the
3729 * range. This is just to avoid any possibility of looping
3735 * We made some progress since the last retry or this is
3736 * the first time we are retrying. Fault in as many pages
3737 * as possible and retry.
3739 fault_in_iov_iter_writeable(to, left);
3744 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
3745 return ret < 0 ? ret : read;
3748 static ssize_t btrfs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
3752 if (iocb->ki_flags & IOCB_DIRECT) {
3753 ret = btrfs_direct_read(iocb, to);
3754 if (ret < 0 || !iov_iter_count(to) ||
3755 iocb->ki_pos >= i_size_read(file_inode(iocb->ki_filp)))
3759 return filemap_read(iocb, to, ret);
3762 const struct file_operations btrfs_file_operations = {
3763 .llseek = btrfs_file_llseek,
3764 .read_iter = btrfs_file_read_iter,
3765 .splice_read = generic_file_splice_read,
3766 .write_iter = btrfs_file_write_iter,
3767 .splice_write = iter_file_splice_write,
3768 .mmap = btrfs_file_mmap,
3769 .open = btrfs_file_open,
3770 .release = btrfs_release_file,
3771 .fsync = btrfs_sync_file,
3772 .fallocate = btrfs_fallocate,
3773 .unlocked_ioctl = btrfs_ioctl,
3774 #ifdef CONFIG_COMPAT
3775 .compat_ioctl = btrfs_compat_ioctl,
3777 .remap_file_range = btrfs_remap_file_range,
3780 void __cold btrfs_auto_defrag_exit(void)
3782 kmem_cache_destroy(btrfs_inode_defrag_cachep);
3785 int __init btrfs_auto_defrag_init(void)
3787 btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
3788 sizeof(struct inode_defrag), 0,
3791 if (!btrfs_inode_defrag_cachep)
3797 int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end)
3802 * So with compression we will find and lock a dirty page and clear the
3803 * first one as dirty, setup an async extent, and immediately return
3804 * with the entire range locked but with nobody actually marked with
3805 * writeback. So we can't just filemap_write_and_wait_range() and
3806 * expect it to work since it will just kick off a thread to do the
3807 * actual work. So we need to call filemap_fdatawrite_range _again_
3808 * since it will wait on the page lock, which won't be unlocked until
3809 * after the pages have been marked as writeback and so we're good to go
3810 * from there. We have to do this otherwise we'll miss the ordered
3811 * extents and that results in badness. Please Josef, do not think you
3812 * know better and pull this out at some point in the future, it is
3813 * right and you are wrong.
3815 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
3816 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
3817 &BTRFS_I(inode)->runtime_flags))
3818 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);