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>
21 #include "transaction.h"
22 #include "btrfs_inode.h"
23 #include "print-tree.h"
28 #include "compression.h"
29 #include "delalloc-space.h"
32 static struct kmem_cache *btrfs_inode_defrag_cachep;
34 * when auto defrag is enabled we
35 * queue up these defrag structs to remember which
36 * inodes need defragging passes
39 struct rb_node rb_node;
43 * transid where the defrag was added, we search for
44 * extents newer than this
51 /* last offset we were able to defrag */
54 /* if we've wrapped around back to zero once already */
58 static int __compare_inode_defrag(struct inode_defrag *defrag1,
59 struct inode_defrag *defrag2)
61 if (defrag1->root > defrag2->root)
63 else if (defrag1->root < defrag2->root)
65 else if (defrag1->ino > defrag2->ino)
67 else if (defrag1->ino < defrag2->ino)
73 /* pop a record for an inode into the defrag tree. The lock
74 * must be held already
76 * If you're inserting a record for an older transid than an
77 * existing record, the transid already in the tree is lowered
79 * If an existing record is found the defrag item you
82 static int __btrfs_add_inode_defrag(struct btrfs_inode *inode,
83 struct inode_defrag *defrag)
85 struct btrfs_fs_info *fs_info = inode->root->fs_info;
86 struct inode_defrag *entry;
88 struct rb_node *parent = NULL;
91 p = &fs_info->defrag_inodes.rb_node;
94 entry = rb_entry(parent, struct inode_defrag, rb_node);
96 ret = __compare_inode_defrag(defrag, entry);
100 p = &parent->rb_right;
102 /* if we're reinserting an entry for
103 * an old defrag run, make sure to
104 * lower the transid of our existing record
106 if (defrag->transid < entry->transid)
107 entry->transid = defrag->transid;
108 if (defrag->last_offset > entry->last_offset)
109 entry->last_offset = defrag->last_offset;
113 set_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags);
114 rb_link_node(&defrag->rb_node, parent, p);
115 rb_insert_color(&defrag->rb_node, &fs_info->defrag_inodes);
119 static inline int __need_auto_defrag(struct btrfs_fs_info *fs_info)
121 if (!btrfs_test_opt(fs_info, AUTO_DEFRAG))
124 if (btrfs_fs_closing(fs_info))
131 * insert a defrag record for this inode if auto defrag is
134 int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
135 struct btrfs_inode *inode)
137 struct btrfs_root *root = inode->root;
138 struct btrfs_fs_info *fs_info = root->fs_info;
139 struct inode_defrag *defrag;
143 if (!__need_auto_defrag(fs_info))
146 if (test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags))
150 transid = trans->transid;
152 transid = inode->root->last_trans;
154 defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS);
158 defrag->ino = btrfs_ino(inode);
159 defrag->transid = transid;
160 defrag->root = root->root_key.objectid;
162 spin_lock(&fs_info->defrag_inodes_lock);
163 if (!test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags)) {
165 * If we set IN_DEFRAG flag and evict the inode from memory,
166 * and then re-read this inode, this new inode doesn't have
167 * IN_DEFRAG flag. At the case, we may find the existed defrag.
169 ret = __btrfs_add_inode_defrag(inode, defrag);
171 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
173 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
175 spin_unlock(&fs_info->defrag_inodes_lock);
180 * Requeue the defrag object. If there is a defrag object that points to
181 * the same inode in the tree, we will merge them together (by
182 * __btrfs_add_inode_defrag()) and free the one that we want to requeue.
184 static void btrfs_requeue_inode_defrag(struct btrfs_inode *inode,
185 struct inode_defrag *defrag)
187 struct btrfs_fs_info *fs_info = inode->root->fs_info;
190 if (!__need_auto_defrag(fs_info))
194 * Here we don't check the IN_DEFRAG flag, because we need merge
197 spin_lock(&fs_info->defrag_inodes_lock);
198 ret = __btrfs_add_inode_defrag(inode, defrag);
199 spin_unlock(&fs_info->defrag_inodes_lock);
204 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
208 * pick the defragable inode that we want, if it doesn't exist, we will get
211 static struct inode_defrag *
212 btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino)
214 struct inode_defrag *entry = NULL;
215 struct inode_defrag tmp;
217 struct rb_node *parent = NULL;
223 spin_lock(&fs_info->defrag_inodes_lock);
224 p = fs_info->defrag_inodes.rb_node;
227 entry = rb_entry(parent, struct inode_defrag, rb_node);
229 ret = __compare_inode_defrag(&tmp, entry);
233 p = parent->rb_right;
238 if (parent && __compare_inode_defrag(&tmp, entry) > 0) {
239 parent = rb_next(parent);
241 entry = rb_entry(parent, struct inode_defrag, rb_node);
247 rb_erase(parent, &fs_info->defrag_inodes);
248 spin_unlock(&fs_info->defrag_inodes_lock);
252 void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info)
254 struct inode_defrag *defrag;
255 struct rb_node *node;
257 spin_lock(&fs_info->defrag_inodes_lock);
258 node = rb_first(&fs_info->defrag_inodes);
260 rb_erase(node, &fs_info->defrag_inodes);
261 defrag = rb_entry(node, struct inode_defrag, rb_node);
262 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
264 cond_resched_lock(&fs_info->defrag_inodes_lock);
266 node = rb_first(&fs_info->defrag_inodes);
268 spin_unlock(&fs_info->defrag_inodes_lock);
271 #define BTRFS_DEFRAG_BATCH 1024
273 static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info,
274 struct inode_defrag *defrag)
276 struct btrfs_root *inode_root;
278 struct btrfs_ioctl_defrag_range_args range;
283 inode_root = btrfs_get_fs_root(fs_info, defrag->root, true);
284 if (IS_ERR(inode_root)) {
285 ret = PTR_ERR(inode_root);
289 inode = btrfs_iget(fs_info->sb, defrag->ino, inode_root);
290 btrfs_put_root(inode_root);
292 ret = PTR_ERR(inode);
296 /* do a chunk of defrag */
297 clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
298 memset(&range, 0, sizeof(range));
300 range.start = defrag->last_offset;
302 sb_start_write(fs_info->sb);
303 num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
305 sb_end_write(fs_info->sb);
307 * if we filled the whole defrag batch, there
308 * must be more work to do. Queue this defrag
311 if (num_defrag == BTRFS_DEFRAG_BATCH) {
312 defrag->last_offset = range.start;
313 btrfs_requeue_inode_defrag(BTRFS_I(inode), defrag);
314 } else if (defrag->last_offset && !defrag->cycled) {
316 * we didn't fill our defrag batch, but
317 * we didn't start at zero. Make sure we loop
318 * around to the start of the file.
320 defrag->last_offset = 0;
322 btrfs_requeue_inode_defrag(BTRFS_I(inode), defrag);
324 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
330 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
335 * run through the list of inodes in the FS that need
338 int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
340 struct inode_defrag *defrag;
342 u64 root_objectid = 0;
344 atomic_inc(&fs_info->defrag_running);
346 /* Pause the auto defragger. */
347 if (test_bit(BTRFS_FS_STATE_REMOUNTING,
351 if (!__need_auto_defrag(fs_info))
354 /* find an inode to defrag */
355 defrag = btrfs_pick_defrag_inode(fs_info, root_objectid,
358 if (root_objectid || first_ino) {
367 first_ino = defrag->ino + 1;
368 root_objectid = defrag->root;
370 __btrfs_run_defrag_inode(fs_info, defrag);
372 atomic_dec(&fs_info->defrag_running);
375 * during unmount, we use the transaction_wait queue to
376 * wait for the defragger to stop
378 wake_up(&fs_info->transaction_wait);
382 /* simple helper to fault in pages and copy. This should go away
383 * and be replaced with calls into generic code.
385 static noinline int btrfs_copy_from_user(loff_t pos, size_t write_bytes,
386 struct page **prepared_pages,
390 size_t total_copied = 0;
392 int offset = offset_in_page(pos);
394 while (write_bytes > 0) {
395 size_t count = min_t(size_t,
396 PAGE_SIZE - offset, write_bytes);
397 struct page *page = prepared_pages[pg];
399 * Copy data from userspace to the current page
401 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
403 /* Flush processor's dcache for this page */
404 flush_dcache_page(page);
407 * if we get a partial write, we can end up with
408 * partially up to date pages. These add
409 * a lot of complexity, so make sure they don't
410 * happen by forcing this copy to be retried.
412 * The rest of the btrfs_file_write code will fall
413 * back to page at a time copies after we return 0.
415 if (!PageUptodate(page) && copied < count)
418 iov_iter_advance(i, copied);
419 write_bytes -= copied;
420 total_copied += copied;
422 /* Return to btrfs_file_write_iter to fault page */
423 if (unlikely(copied == 0))
426 if (copied < PAGE_SIZE - offset) {
437 * unlocks pages after btrfs_file_write is done with them
439 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
442 for (i = 0; i < num_pages; i++) {
443 /* page checked is some magic around finding pages that
444 * have been modified without going through btrfs_set_page_dirty
445 * clear it here. There should be no need to mark the pages
446 * accessed as prepare_pages should have marked them accessed
447 * in prepare_pages via find_or_create_page()
449 ClearPageChecked(pages[i]);
450 unlock_page(pages[i]);
456 * after copy_from_user, pages need to be dirtied and we need to make
457 * sure holes are created between the current EOF and the start of
458 * any next extents (if required).
460 * this also makes the decision about creating an inline extent vs
461 * doing real data extents, marking pages dirty and delalloc as required.
463 int btrfs_dirty_pages(struct btrfs_inode *inode, struct page **pages,
464 size_t num_pages, loff_t pos, size_t write_bytes,
465 struct extent_state **cached)
467 struct btrfs_fs_info *fs_info = inode->root->fs_info;
472 u64 end_of_last_block;
473 u64 end_pos = pos + write_bytes;
474 loff_t isize = i_size_read(&inode->vfs_inode);
475 unsigned int extra_bits = 0;
477 start_pos = pos & ~((u64) fs_info->sectorsize - 1);
478 num_bytes = round_up(write_bytes + pos - start_pos,
479 fs_info->sectorsize);
481 end_of_last_block = start_pos + num_bytes - 1;
484 * The pages may have already been dirty, clear out old accounting so
485 * we can set things up properly
487 clear_extent_bit(&inode->io_tree, start_pos, end_of_last_block,
488 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
491 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
496 for (i = 0; i < num_pages; i++) {
497 struct page *p = pages[i];
504 * we've only changed i_size in ram, and we haven't updated
505 * the disk i_size. There is no need to log the inode
509 i_size_write(&inode->vfs_inode, end_pos);
514 * this drops all the extents in the cache that intersect the range
515 * [start, end]. Existing extents are split as required.
517 void btrfs_drop_extent_cache(struct btrfs_inode *inode, u64 start, u64 end,
520 struct extent_map *em;
521 struct extent_map *split = NULL;
522 struct extent_map *split2 = NULL;
523 struct extent_map_tree *em_tree = &inode->extent_tree;
524 u64 len = end - start + 1;
532 WARN_ON(end < start);
533 if (end == (u64)-1) {
542 split = alloc_extent_map();
544 split2 = alloc_extent_map();
545 if (!split || !split2)
548 write_lock(&em_tree->lock);
549 em = lookup_extent_mapping(em_tree, start, len);
551 write_unlock(&em_tree->lock);
555 gen = em->generation;
556 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
557 if (testend && em->start + em->len >= start + len) {
559 write_unlock(&em_tree->lock);
562 start = em->start + em->len;
564 len = start + len - (em->start + em->len);
566 write_unlock(&em_tree->lock);
569 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
570 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
571 clear_bit(EXTENT_FLAG_LOGGING, &flags);
572 modified = !list_empty(&em->list);
576 if (em->start < start) {
577 split->start = em->start;
578 split->len = start - em->start;
580 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
581 split->orig_start = em->orig_start;
582 split->block_start = em->block_start;
585 split->block_len = em->block_len;
587 split->block_len = split->len;
588 split->orig_block_len = max(split->block_len,
590 split->ram_bytes = em->ram_bytes;
592 split->orig_start = split->start;
593 split->block_len = 0;
594 split->block_start = em->block_start;
595 split->orig_block_len = 0;
596 split->ram_bytes = split->len;
599 split->generation = gen;
600 split->flags = flags;
601 split->compress_type = em->compress_type;
602 replace_extent_mapping(em_tree, em, split, modified);
603 free_extent_map(split);
607 if (testend && em->start + em->len > start + len) {
608 u64 diff = start + len - em->start;
610 split->start = start + len;
611 split->len = em->start + em->len - (start + len);
612 split->flags = flags;
613 split->compress_type = em->compress_type;
614 split->generation = gen;
616 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
617 split->orig_block_len = max(em->block_len,
620 split->ram_bytes = em->ram_bytes;
622 split->block_len = em->block_len;
623 split->block_start = em->block_start;
624 split->orig_start = em->orig_start;
626 split->block_len = split->len;
627 split->block_start = em->block_start
629 split->orig_start = em->orig_start;
632 split->ram_bytes = split->len;
633 split->orig_start = split->start;
634 split->block_len = 0;
635 split->block_start = em->block_start;
636 split->orig_block_len = 0;
639 if (extent_map_in_tree(em)) {
640 replace_extent_mapping(em_tree, em, split,
643 ret = add_extent_mapping(em_tree, split,
645 ASSERT(ret == 0); /* Logic error */
647 free_extent_map(split);
651 if (extent_map_in_tree(em))
652 remove_extent_mapping(em_tree, em);
653 write_unlock(&em_tree->lock);
657 /* once for the tree*/
661 free_extent_map(split);
663 free_extent_map(split2);
667 * this is very complex, but the basic idea is to drop all extents
668 * in the range start - end. hint_block is filled in with a block number
669 * that would be a good hint to the block allocator for this file.
671 * If an extent intersects the range but is not entirely inside the range
672 * it is either truncated or split. Anything entirely inside the range
673 * is deleted from the tree.
675 int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
676 struct btrfs_root *root, struct btrfs_inode *inode,
677 struct btrfs_path *path, u64 start, u64 end,
678 u64 *drop_end, int drop_cache,
680 u32 extent_item_size,
683 struct btrfs_fs_info *fs_info = root->fs_info;
684 struct extent_buffer *leaf;
685 struct btrfs_file_extent_item *fi;
686 struct btrfs_ref ref = { 0 };
687 struct btrfs_key key;
688 struct btrfs_key new_key;
689 struct inode *vfs_inode = &inode->vfs_inode;
690 u64 ino = btrfs_ino(inode);
691 u64 search_start = start;
694 u64 extent_offset = 0;
696 u64 last_end = start;
702 int modify_tree = -1;
705 int leafs_visited = 0;
708 btrfs_drop_extent_cache(inode, start, end - 1, 0);
710 if (start >= inode->disk_i_size && !replace_extent)
713 update_refs = (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID);
716 ret = btrfs_lookup_file_extent(trans, root, path, ino,
717 search_start, modify_tree);
720 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
721 leaf = path->nodes[0];
722 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
723 if (key.objectid == ino &&
724 key.type == BTRFS_EXTENT_DATA_KEY)
730 leaf = path->nodes[0];
731 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
733 ret = btrfs_next_leaf(root, path);
741 leaf = path->nodes[0];
745 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
747 if (key.objectid > ino)
749 if (WARN_ON_ONCE(key.objectid < ino) ||
750 key.type < BTRFS_EXTENT_DATA_KEY) {
755 if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
758 fi = btrfs_item_ptr(leaf, path->slots[0],
759 struct btrfs_file_extent_item);
760 extent_type = btrfs_file_extent_type(leaf, fi);
762 if (extent_type == BTRFS_FILE_EXTENT_REG ||
763 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
764 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
765 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
766 extent_offset = btrfs_file_extent_offset(leaf, fi);
767 extent_end = key.offset +
768 btrfs_file_extent_num_bytes(leaf, fi);
769 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
770 extent_end = key.offset +
771 btrfs_file_extent_ram_bytes(leaf, fi);
778 * Don't skip extent items representing 0 byte lengths. They
779 * used to be created (bug) if while punching holes we hit
780 * -ENOSPC condition. So if we find one here, just ensure we
781 * delete it, otherwise we would insert a new file extent item
782 * with the same key (offset) as that 0 bytes length file
783 * extent item in the call to setup_items_for_insert() later
786 if (extent_end == key.offset && extent_end >= search_start) {
787 last_end = extent_end;
788 goto delete_extent_item;
791 if (extent_end <= search_start) {
797 search_start = max(key.offset, start);
798 if (recow || !modify_tree) {
800 btrfs_release_path(path);
805 * | - range to drop - |
806 * | -------- extent -------- |
808 if (start > key.offset && end < extent_end) {
810 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
815 memcpy(&new_key, &key, sizeof(new_key));
816 new_key.offset = start;
817 ret = btrfs_duplicate_item(trans, root, path,
819 if (ret == -EAGAIN) {
820 btrfs_release_path(path);
826 leaf = path->nodes[0];
827 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
828 struct btrfs_file_extent_item);
829 btrfs_set_file_extent_num_bytes(leaf, fi,
832 fi = btrfs_item_ptr(leaf, path->slots[0],
833 struct btrfs_file_extent_item);
835 extent_offset += start - key.offset;
836 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
837 btrfs_set_file_extent_num_bytes(leaf, fi,
839 btrfs_mark_buffer_dirty(leaf);
841 if (update_refs && disk_bytenr > 0) {
842 btrfs_init_generic_ref(&ref,
843 BTRFS_ADD_DELAYED_REF,
844 disk_bytenr, num_bytes, 0);
845 btrfs_init_data_ref(&ref,
846 root->root_key.objectid,
848 start - extent_offset);
849 ret = btrfs_inc_extent_ref(trans, &ref);
850 BUG_ON(ret); /* -ENOMEM */
855 * From here on out we will have actually dropped something, so
856 * last_end can be updated.
858 last_end = extent_end;
861 * | ---- range to drop ----- |
862 * | -------- extent -------- |
864 if (start <= key.offset && end < extent_end) {
865 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
870 memcpy(&new_key, &key, sizeof(new_key));
871 new_key.offset = end;
872 btrfs_set_item_key_safe(fs_info, path, &new_key);
874 extent_offset += end - key.offset;
875 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
876 btrfs_set_file_extent_num_bytes(leaf, fi,
878 btrfs_mark_buffer_dirty(leaf);
879 if (update_refs && disk_bytenr > 0)
880 inode_sub_bytes(vfs_inode, end - key.offset);
884 search_start = extent_end;
886 * | ---- range to drop ----- |
887 * | -------- extent -------- |
889 if (start > key.offset && end >= extent_end) {
891 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
896 btrfs_set_file_extent_num_bytes(leaf, fi,
898 btrfs_mark_buffer_dirty(leaf);
899 if (update_refs && disk_bytenr > 0)
900 inode_sub_bytes(vfs_inode, extent_end - start);
901 if (end == extent_end)
909 * | ---- range to drop ----- |
910 * | ------ extent ------ |
912 if (start <= key.offset && end >= extent_end) {
915 del_slot = path->slots[0];
918 BUG_ON(del_slot + del_nr != path->slots[0]);
923 extent_type == BTRFS_FILE_EXTENT_INLINE) {
924 inode_sub_bytes(vfs_inode,
925 extent_end - key.offset);
926 extent_end = ALIGN(extent_end,
927 fs_info->sectorsize);
928 } else if (update_refs && disk_bytenr > 0) {
929 btrfs_init_generic_ref(&ref,
930 BTRFS_DROP_DELAYED_REF,
931 disk_bytenr, num_bytes, 0);
932 btrfs_init_data_ref(&ref,
933 root->root_key.objectid,
935 key.offset - extent_offset);
936 ret = btrfs_free_extent(trans, &ref);
937 BUG_ON(ret); /* -ENOMEM */
938 inode_sub_bytes(vfs_inode,
939 extent_end - key.offset);
942 if (end == extent_end)
945 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
950 ret = btrfs_del_items(trans, root, path, del_slot,
953 btrfs_abort_transaction(trans, ret);
960 btrfs_release_path(path);
967 if (!ret && del_nr > 0) {
969 * Set path->slots[0] to first slot, so that after the delete
970 * if items are move off from our leaf to its immediate left or
971 * right neighbor leafs, we end up with a correct and adjusted
972 * path->slots[0] for our insertion (if replace_extent != 0).
974 path->slots[0] = del_slot;
975 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
977 btrfs_abort_transaction(trans, ret);
980 leaf = path->nodes[0];
982 * If btrfs_del_items() was called, it might have deleted a leaf, in
983 * which case it unlocked our path, so check path->locks[0] matches a
986 if (!ret && replace_extent && leafs_visited == 1 &&
987 (path->locks[0] == BTRFS_WRITE_LOCK_BLOCKING ||
988 path->locks[0] == BTRFS_WRITE_LOCK) &&
989 btrfs_leaf_free_space(leaf) >=
990 sizeof(struct btrfs_item) + extent_item_size) {
993 key.type = BTRFS_EXTENT_DATA_KEY;
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 setup_items_for_insert(root, path, &key, &extent_item_size, 1);
1006 if (!replace_extent || !(*key_inserted))
1007 btrfs_release_path(path);
1009 *drop_end = found ? min(end, last_end) : end;
1013 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
1014 struct btrfs_root *root, struct inode *inode, u64 start,
1015 u64 end, int drop_cache)
1017 struct btrfs_path *path;
1020 path = btrfs_alloc_path();
1023 ret = __btrfs_drop_extents(trans, root, BTRFS_I(inode), path, start,
1024 end, NULL, drop_cache, 0, 0, NULL);
1025 btrfs_free_path(path);
1029 static int extent_mergeable(struct extent_buffer *leaf, int slot,
1030 u64 objectid, u64 bytenr, u64 orig_offset,
1031 u64 *start, u64 *end)
1033 struct btrfs_file_extent_item *fi;
1034 struct btrfs_key key;
1037 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1040 btrfs_item_key_to_cpu(leaf, &key, slot);
1041 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
1044 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1045 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
1046 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
1047 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
1048 btrfs_file_extent_compression(leaf, fi) ||
1049 btrfs_file_extent_encryption(leaf, fi) ||
1050 btrfs_file_extent_other_encoding(leaf, fi))
1053 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1054 if ((*start && *start != key.offset) || (*end && *end != extent_end))
1057 *start = key.offset;
1063 * Mark extent in the range start - end as written.
1065 * This changes extent type from 'pre-allocated' to 'regular'. If only
1066 * part of extent is marked as written, the extent will be split into
1069 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
1070 struct btrfs_inode *inode, u64 start, u64 end)
1072 struct btrfs_fs_info *fs_info = trans->fs_info;
1073 struct btrfs_root *root = inode->root;
1074 struct extent_buffer *leaf;
1075 struct btrfs_path *path;
1076 struct btrfs_file_extent_item *fi;
1077 struct btrfs_ref ref = { 0 };
1078 struct btrfs_key key;
1079 struct btrfs_key new_key;
1091 u64 ino = btrfs_ino(inode);
1093 path = btrfs_alloc_path();
1100 key.type = BTRFS_EXTENT_DATA_KEY;
1103 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1106 if (ret > 0 && path->slots[0] > 0)
1109 leaf = path->nodes[0];
1110 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1111 if (key.objectid != ino ||
1112 key.type != BTRFS_EXTENT_DATA_KEY) {
1114 btrfs_abort_transaction(trans, ret);
1117 fi = btrfs_item_ptr(leaf, path->slots[0],
1118 struct btrfs_file_extent_item);
1119 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_PREALLOC) {
1121 btrfs_abort_transaction(trans, ret);
1124 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1125 if (key.offset > start || extent_end < end) {
1127 btrfs_abort_transaction(trans, ret);
1131 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1132 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1133 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1134 memcpy(&new_key, &key, sizeof(new_key));
1136 if (start == key.offset && end < extent_end) {
1139 if (extent_mergeable(leaf, path->slots[0] - 1,
1140 ino, bytenr, orig_offset,
1141 &other_start, &other_end)) {
1142 new_key.offset = end;
1143 btrfs_set_item_key_safe(fs_info, path, &new_key);
1144 fi = btrfs_item_ptr(leaf, path->slots[0],
1145 struct btrfs_file_extent_item);
1146 btrfs_set_file_extent_generation(leaf, fi,
1148 btrfs_set_file_extent_num_bytes(leaf, fi,
1150 btrfs_set_file_extent_offset(leaf, fi,
1152 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1153 struct btrfs_file_extent_item);
1154 btrfs_set_file_extent_generation(leaf, fi,
1156 btrfs_set_file_extent_num_bytes(leaf, fi,
1158 btrfs_mark_buffer_dirty(leaf);
1163 if (start > key.offset && end == extent_end) {
1166 if (extent_mergeable(leaf, path->slots[0] + 1,
1167 ino, bytenr, orig_offset,
1168 &other_start, &other_end)) {
1169 fi = btrfs_item_ptr(leaf, path->slots[0],
1170 struct btrfs_file_extent_item);
1171 btrfs_set_file_extent_num_bytes(leaf, fi,
1172 start - key.offset);
1173 btrfs_set_file_extent_generation(leaf, fi,
1176 new_key.offset = start;
1177 btrfs_set_item_key_safe(fs_info, path, &new_key);
1179 fi = btrfs_item_ptr(leaf, path->slots[0],
1180 struct btrfs_file_extent_item);
1181 btrfs_set_file_extent_generation(leaf, fi,
1183 btrfs_set_file_extent_num_bytes(leaf, fi,
1185 btrfs_set_file_extent_offset(leaf, fi,
1186 start - orig_offset);
1187 btrfs_mark_buffer_dirty(leaf);
1192 while (start > key.offset || end < extent_end) {
1193 if (key.offset == start)
1196 new_key.offset = split;
1197 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1198 if (ret == -EAGAIN) {
1199 btrfs_release_path(path);
1203 btrfs_abort_transaction(trans, ret);
1207 leaf = path->nodes[0];
1208 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1209 struct btrfs_file_extent_item);
1210 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1211 btrfs_set_file_extent_num_bytes(leaf, fi,
1212 split - key.offset);
1214 fi = btrfs_item_ptr(leaf, path->slots[0],
1215 struct btrfs_file_extent_item);
1217 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1218 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1219 btrfs_set_file_extent_num_bytes(leaf, fi,
1220 extent_end - split);
1221 btrfs_mark_buffer_dirty(leaf);
1223 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, bytenr,
1225 btrfs_init_data_ref(&ref, root->root_key.objectid, ino,
1227 ret = btrfs_inc_extent_ref(trans, &ref);
1229 btrfs_abort_transaction(trans, ret);
1233 if (split == start) {
1236 if (start != key.offset) {
1238 btrfs_abort_transaction(trans, ret);
1249 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
1251 btrfs_init_data_ref(&ref, root->root_key.objectid, ino, orig_offset);
1252 if (extent_mergeable(leaf, path->slots[0] + 1,
1253 ino, bytenr, orig_offset,
1254 &other_start, &other_end)) {
1256 btrfs_release_path(path);
1259 extent_end = other_end;
1260 del_slot = path->slots[0] + 1;
1262 ret = btrfs_free_extent(trans, &ref);
1264 btrfs_abort_transaction(trans, ret);
1270 if (extent_mergeable(leaf, path->slots[0] - 1,
1271 ino, bytenr, orig_offset,
1272 &other_start, &other_end)) {
1274 btrfs_release_path(path);
1277 key.offset = other_start;
1278 del_slot = path->slots[0];
1280 ret = btrfs_free_extent(trans, &ref);
1282 btrfs_abort_transaction(trans, ret);
1287 fi = btrfs_item_ptr(leaf, path->slots[0],
1288 struct btrfs_file_extent_item);
1289 btrfs_set_file_extent_type(leaf, fi,
1290 BTRFS_FILE_EXTENT_REG);
1291 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1292 btrfs_mark_buffer_dirty(leaf);
1294 fi = btrfs_item_ptr(leaf, del_slot - 1,
1295 struct btrfs_file_extent_item);
1296 btrfs_set_file_extent_type(leaf, fi,
1297 BTRFS_FILE_EXTENT_REG);
1298 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1299 btrfs_set_file_extent_num_bytes(leaf, fi,
1300 extent_end - key.offset);
1301 btrfs_mark_buffer_dirty(leaf);
1303 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1305 btrfs_abort_transaction(trans, ret);
1310 btrfs_free_path(path);
1315 * on error we return an unlocked page and the error value
1316 * on success we return a locked page and 0
1318 static int prepare_uptodate_page(struct inode *inode,
1319 struct page *page, u64 pos,
1320 bool force_uptodate)
1324 if (((pos & (PAGE_SIZE - 1)) || force_uptodate) &&
1325 !PageUptodate(page)) {
1326 ret = btrfs_readpage(NULL, page);
1330 if (!PageUptodate(page)) {
1334 if (page->mapping != inode->i_mapping) {
1343 * this just gets pages into the page cache and locks them down.
1345 static noinline int prepare_pages(struct inode *inode, struct page **pages,
1346 size_t num_pages, loff_t pos,
1347 size_t write_bytes, bool force_uptodate)
1350 unsigned long index = pos >> PAGE_SHIFT;
1351 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1355 for (i = 0; i < num_pages; i++) {
1357 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1358 mask | __GFP_WRITE);
1366 err = prepare_uptodate_page(inode, pages[i], pos,
1368 if (!err && i == num_pages - 1)
1369 err = prepare_uptodate_page(inode, pages[i],
1370 pos + write_bytes, false);
1373 if (err == -EAGAIN) {
1380 wait_on_page_writeback(pages[i]);
1385 while (faili >= 0) {
1386 unlock_page(pages[faili]);
1387 put_page(pages[faili]);
1395 * This function locks the extent and properly waits for data=ordered extents
1396 * to finish before allowing the pages to be modified if need.
1399 * 1 - the extent is locked
1400 * 0 - the extent is not locked, and everything is OK
1401 * -EAGAIN - need re-prepare the pages
1402 * the other < 0 number - Something wrong happens
1405 lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct page **pages,
1406 size_t num_pages, loff_t pos,
1408 u64 *lockstart, u64 *lockend,
1409 struct extent_state **cached_state)
1411 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1417 start_pos = round_down(pos, fs_info->sectorsize);
1418 last_pos = round_up(pos + write_bytes, fs_info->sectorsize) - 1;
1420 if (start_pos < inode->vfs_inode.i_size) {
1421 struct btrfs_ordered_extent *ordered;
1423 lock_extent_bits(&inode->io_tree, start_pos, last_pos,
1425 ordered = btrfs_lookup_ordered_range(inode, start_pos,
1426 last_pos - start_pos + 1);
1428 ordered->file_offset + ordered->num_bytes > start_pos &&
1429 ordered->file_offset <= last_pos) {
1430 unlock_extent_cached(&inode->io_tree, start_pos,
1431 last_pos, cached_state);
1432 for (i = 0; i < num_pages; i++) {
1433 unlock_page(pages[i]);
1436 btrfs_start_ordered_extent(ordered, 1);
1437 btrfs_put_ordered_extent(ordered);
1441 btrfs_put_ordered_extent(ordered);
1443 *lockstart = start_pos;
1444 *lockend = last_pos;
1449 * It's possible the pages are dirty right now, but we don't want
1450 * to clean them yet because copy_from_user may catch a page fault
1451 * and we might have to fall back to one page at a time. If that
1452 * happens, we'll unlock these pages and we'd have a window where
1453 * reclaim could sneak in and drop the once-dirty page on the floor
1454 * without writing it.
1456 * We have the pages locked and the extent range locked, so there's
1457 * no way someone can start IO on any dirty pages in this range.
1459 * We'll call btrfs_dirty_pages() later on, and that will flip around
1460 * delalloc bits and dirty the pages as required.
1462 for (i = 0; i < num_pages; i++) {
1463 set_page_extent_mapped(pages[i]);
1464 WARN_ON(!PageLocked(pages[i]));
1470 static int check_can_nocow(struct btrfs_inode *inode, loff_t pos,
1471 size_t *write_bytes, bool nowait)
1473 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1474 struct btrfs_root *root = inode->root;
1475 u64 lockstart, lockend;
1479 if (!(inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
1482 if (!nowait && !btrfs_drew_try_write_lock(&root->snapshot_lock))
1485 lockstart = round_down(pos, fs_info->sectorsize);
1486 lockend = round_up(pos + *write_bytes,
1487 fs_info->sectorsize) - 1;
1488 num_bytes = lockend - lockstart + 1;
1491 struct btrfs_ordered_extent *ordered;
1493 if (!try_lock_extent(&inode->io_tree, lockstart, lockend))
1496 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1499 btrfs_put_ordered_extent(ordered);
1504 btrfs_lock_and_flush_ordered_range(inode, lockstart,
1508 ret = can_nocow_extent(&inode->vfs_inode, lockstart, &num_bytes,
1509 NULL, NULL, NULL, false);
1513 btrfs_drew_write_unlock(&root->snapshot_lock);
1515 *write_bytes = min_t(size_t, *write_bytes ,
1516 num_bytes - pos + lockstart);
1519 unlock_extent(&inode->io_tree, lockstart, lockend);
1524 static int check_nocow_nolock(struct btrfs_inode *inode, loff_t pos,
1525 size_t *write_bytes)
1527 return check_can_nocow(inode, pos, write_bytes, true);
1531 * Check if we can do nocow write into the range [@pos, @pos + @write_bytes)
1534 * @write_bytes: The length to write, will be updated to the nocow writeable
1537 * This function will flush ordered extents in the range to ensure proper
1541 * >0 and update @write_bytes if we can do nocow write
1542 * 0 if we can't do nocow write
1543 * -EAGAIN if we can't get the needed lock or there are ordered extents
1544 * for * (nowait == true) case
1545 * <0 if other error happened
1547 * NOTE: Callers need to release the lock by btrfs_check_nocow_unlock().
1549 int btrfs_check_nocow_lock(struct btrfs_inode *inode, loff_t pos,
1550 size_t *write_bytes)
1552 return check_can_nocow(inode, pos, write_bytes, false);
1555 void btrfs_check_nocow_unlock(struct btrfs_inode *inode)
1557 btrfs_drew_write_unlock(&inode->root->snapshot_lock);
1560 static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb,
1563 struct file *file = iocb->ki_filp;
1564 loff_t pos = iocb->ki_pos;
1565 struct inode *inode = file_inode(file);
1566 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1567 struct page **pages = NULL;
1568 struct extent_changeset *data_reserved = NULL;
1569 u64 release_bytes = 0;
1572 size_t num_written = 0;
1575 bool only_release_metadata = false;
1576 bool force_page_uptodate = false;
1578 nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_SIZE),
1579 PAGE_SIZE / (sizeof(struct page *)));
1580 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1581 nrptrs = max(nrptrs, 8);
1582 pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL);
1586 while (iov_iter_count(i) > 0) {
1587 struct extent_state *cached_state = NULL;
1588 size_t offset = offset_in_page(pos);
1589 size_t sector_offset;
1590 size_t write_bytes = min(iov_iter_count(i),
1591 nrptrs * (size_t)PAGE_SIZE -
1593 size_t num_pages = DIV_ROUND_UP(write_bytes + offset,
1595 size_t reserve_bytes;
1598 size_t dirty_sectors;
1602 WARN_ON(num_pages > nrptrs);
1605 * Fault pages before locking them in prepare_pages
1606 * to avoid recursive lock
1608 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1613 only_release_metadata = false;
1614 sector_offset = pos & (fs_info->sectorsize - 1);
1615 reserve_bytes = round_up(write_bytes + sector_offset,
1616 fs_info->sectorsize);
1618 extent_changeset_release(data_reserved);
1619 ret = btrfs_check_data_free_space(BTRFS_I(inode),
1620 &data_reserved, pos,
1623 if (btrfs_check_nocow_lock(BTRFS_I(inode), pos,
1624 &write_bytes) > 0) {
1626 * For nodata cow case, no need to reserve
1629 only_release_metadata = true;
1631 * our prealloc extent may be smaller than
1632 * write_bytes, so scale down.
1634 num_pages = DIV_ROUND_UP(write_bytes + offset,
1636 reserve_bytes = round_up(write_bytes +
1638 fs_info->sectorsize);
1644 WARN_ON(reserve_bytes == 0);
1645 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
1648 if (!only_release_metadata)
1649 btrfs_free_reserved_data_space(BTRFS_I(inode),
1653 btrfs_check_nocow_unlock(BTRFS_I(inode));
1657 release_bytes = reserve_bytes;
1660 * This is going to setup the pages array with the number of
1661 * pages we want, so we don't really need to worry about the
1662 * contents of pages from loop to loop
1664 ret = prepare_pages(inode, pages, num_pages,
1666 force_page_uptodate);
1668 btrfs_delalloc_release_extents(BTRFS_I(inode),
1673 extents_locked = lock_and_cleanup_extent_if_need(
1674 BTRFS_I(inode), pages,
1675 num_pages, pos, write_bytes, &lockstart,
1676 &lockend, &cached_state);
1677 if (extents_locked < 0) {
1678 if (extents_locked == -EAGAIN)
1680 btrfs_delalloc_release_extents(BTRFS_I(inode),
1682 ret = extents_locked;
1686 copied = btrfs_copy_from_user(pos, write_bytes, pages, i);
1688 num_sectors = BTRFS_BYTES_TO_BLKS(fs_info, reserve_bytes);
1689 dirty_sectors = round_up(copied + sector_offset,
1690 fs_info->sectorsize);
1691 dirty_sectors = BTRFS_BYTES_TO_BLKS(fs_info, dirty_sectors);
1694 * if we have trouble faulting in the pages, fall
1695 * back to one page at a time
1697 if (copied < write_bytes)
1701 force_page_uptodate = true;
1705 force_page_uptodate = false;
1706 dirty_pages = DIV_ROUND_UP(copied + offset,
1710 if (num_sectors > dirty_sectors) {
1711 /* release everything except the sectors we dirtied */
1712 release_bytes -= dirty_sectors <<
1713 fs_info->sb->s_blocksize_bits;
1714 if (only_release_metadata) {
1715 btrfs_delalloc_release_metadata(BTRFS_I(inode),
1716 release_bytes, true);
1720 __pos = round_down(pos,
1721 fs_info->sectorsize) +
1722 (dirty_pages << PAGE_SHIFT);
1723 btrfs_delalloc_release_space(BTRFS_I(inode),
1724 data_reserved, __pos,
1725 release_bytes, true);
1729 release_bytes = round_up(copied + sector_offset,
1730 fs_info->sectorsize);
1733 ret = btrfs_dirty_pages(BTRFS_I(inode), pages,
1734 dirty_pages, pos, copied,
1738 * If we have not locked the extent range, because the range's
1739 * start offset is >= i_size, we might still have a non-NULL
1740 * cached extent state, acquired while marking the extent range
1741 * as delalloc through btrfs_dirty_pages(). Therefore free any
1742 * possible cached extent state to avoid a memory leak.
1745 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1746 lockstart, lockend, &cached_state);
1748 free_extent_state(cached_state);
1750 btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes);
1752 btrfs_drop_pages(pages, num_pages);
1757 if (only_release_metadata)
1758 btrfs_check_nocow_unlock(BTRFS_I(inode));
1760 if (only_release_metadata && copied > 0) {
1761 lockstart = round_down(pos,
1762 fs_info->sectorsize);
1763 lockend = round_up(pos + copied,
1764 fs_info->sectorsize) - 1;
1766 set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
1767 lockend, EXTENT_NORESERVE, NULL,
1771 btrfs_drop_pages(pages, num_pages);
1775 balance_dirty_pages_ratelimited(inode->i_mapping);
1778 num_written += copied;
1783 if (release_bytes) {
1784 if (only_release_metadata) {
1785 btrfs_check_nocow_unlock(BTRFS_I(inode));
1786 btrfs_delalloc_release_metadata(BTRFS_I(inode),
1787 release_bytes, true);
1789 btrfs_delalloc_release_space(BTRFS_I(inode),
1791 round_down(pos, fs_info->sectorsize),
1792 release_bytes, true);
1796 extent_changeset_free(data_reserved);
1797 return num_written ? num_written : ret;
1800 static ssize_t __btrfs_direct_write(struct kiocb *iocb, struct iov_iter *from)
1802 struct file *file = iocb->ki_filp;
1803 struct inode *inode = file_inode(file);
1806 ssize_t written_buffered;
1810 written = btrfs_direct_IO(iocb, from);
1812 if (written < 0 || !iov_iter_count(from))
1816 written_buffered = btrfs_buffered_write(iocb, from);
1817 if (written_buffered < 0) {
1818 err = written_buffered;
1822 * Ensure all data is persisted. We want the next direct IO read to be
1823 * able to read what was just written.
1825 endbyte = pos + written_buffered - 1;
1826 err = btrfs_fdatawrite_range(inode, pos, endbyte);
1829 err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte);
1832 written += written_buffered;
1833 iocb->ki_pos = pos + written_buffered;
1834 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_SHIFT,
1835 endbyte >> PAGE_SHIFT);
1837 return written ? written : err;
1840 static void update_time_for_write(struct inode *inode)
1842 struct timespec64 now;
1844 if (IS_NOCMTIME(inode))
1847 now = current_time(inode);
1848 if (!timespec64_equal(&inode->i_mtime, &now))
1849 inode->i_mtime = now;
1851 if (!timespec64_equal(&inode->i_ctime, &now))
1852 inode->i_ctime = now;
1854 if (IS_I_VERSION(inode))
1855 inode_inc_iversion(inode);
1858 static ssize_t btrfs_file_write_iter(struct kiocb *iocb,
1859 struct iov_iter *from)
1861 struct file *file = iocb->ki_filp;
1862 struct inode *inode = file_inode(file);
1863 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1866 ssize_t num_written = 0;
1867 const bool sync = iocb->ki_flags & IOCB_DSYNC;
1874 if (!(iocb->ki_flags & IOCB_DIRECT) &&
1875 (iocb->ki_flags & IOCB_NOWAIT))
1878 if (iocb->ki_flags & IOCB_NOWAIT) {
1879 if (!inode_trylock(inode))
1885 err = generic_write_checks(iocb, from);
1887 inode_unlock(inode);
1892 count = iov_iter_count(from);
1893 if (iocb->ki_flags & IOCB_NOWAIT) {
1894 size_t nocow_bytes = count;
1897 * We will allocate space in case nodatacow is not set,
1900 if (check_nocow_nolock(BTRFS_I(inode), pos, &nocow_bytes)
1902 inode_unlock(inode);
1906 * There are holes in the range or parts of the range that must
1907 * be COWed (shared extents, RO block groups, etc), so just bail
1910 if (nocow_bytes < count) {
1911 inode_unlock(inode);
1916 current->backing_dev_info = inode_to_bdi(inode);
1917 err = file_remove_privs(file);
1919 inode_unlock(inode);
1924 * If BTRFS flips readonly due to some impossible error
1925 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1926 * although we have opened a file as writable, we have
1927 * to stop this write operation to ensure FS consistency.
1929 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
1930 inode_unlock(inode);
1936 * We reserve space for updating the inode when we reserve space for the
1937 * extent we are going to write, so we will enospc out there. We don't
1938 * need to start yet another transaction to update the inode as we will
1939 * update the inode when we finish writing whatever data we write.
1941 update_time_for_write(inode);
1943 start_pos = round_down(pos, fs_info->sectorsize);
1944 oldsize = i_size_read(inode);
1945 if (start_pos > oldsize) {
1946 /* Expand hole size to cover write data, preventing empty gap */
1947 end_pos = round_up(pos + count,
1948 fs_info->sectorsize);
1949 err = btrfs_cont_expand(inode, oldsize, end_pos);
1951 inode_unlock(inode);
1954 if (start_pos > round_up(oldsize, fs_info->sectorsize))
1959 atomic_inc(&BTRFS_I(inode)->sync_writers);
1961 if (iocb->ki_flags & IOCB_DIRECT) {
1963 * 1. We must always clear IOCB_DSYNC in order to not deadlock
1964 * in iomap, as it calls generic_write_sync() in this case.
1965 * 2. If we are async, we can call iomap_dio_complete() either
1968 * 2.1. A worker thread from the last bio completed. In this
1969 * case we need to mark the btrfs_dio_data that it is
1970 * async in order to call generic_write_sync() properly.
1971 * This is handled by setting BTRFS_DIO_SYNC_STUB in the
1972 * current->journal_info.
1973 * 2.2 The submitter context, because all IO completed
1974 * before we exited iomap_dio_rw(). In this case we can
1975 * just re-set the IOCB_DSYNC on the iocb and we'll do
1976 * the sync below. If our ->end_io() gets called and
1977 * current->journal_info is set, then we know we're in
1978 * our current context and we will clear
1979 * current->journal_info to indicate that we need to
1983 ASSERT(current->journal_info == NULL);
1984 iocb->ki_flags &= ~IOCB_DSYNC;
1985 current->journal_info = BTRFS_DIO_SYNC_STUB;
1987 num_written = __btrfs_direct_write(iocb, from);
1990 * As stated above, we cleared journal_info, so we need to do
1991 * the sync ourselves.
1993 if (sync && current->journal_info == NULL)
1994 iocb->ki_flags |= IOCB_DSYNC;
1995 current->journal_info = NULL;
1997 num_written = btrfs_buffered_write(iocb, from);
1998 if (num_written > 0)
1999 iocb->ki_pos = pos + num_written;
2001 pagecache_isize_extended(inode, oldsize,
2002 i_size_read(inode));
2005 inode_unlock(inode);
2007 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
2009 if (num_written > 0)
2010 num_written = generic_write_sync(iocb, num_written);
2013 atomic_dec(&BTRFS_I(inode)->sync_writers);
2015 current->backing_dev_info = NULL;
2016 return num_written ? num_written : err;
2019 int btrfs_release_file(struct inode *inode, struct file *filp)
2021 struct btrfs_file_private *private = filp->private_data;
2023 if (private && private->filldir_buf)
2024 kfree(private->filldir_buf);
2026 filp->private_data = NULL;
2029 * Set by setattr when we are about to truncate a file from a non-zero
2030 * size to a zero size. This tries to flush down new bytes that may
2031 * have been written if the application were using truncate to replace
2034 if (test_and_clear_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
2035 &BTRFS_I(inode)->runtime_flags))
2036 filemap_flush(inode->i_mapping);
2040 static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end)
2043 struct blk_plug plug;
2046 * This is only called in fsync, which would do synchronous writes, so
2047 * a plug can merge adjacent IOs as much as possible. Esp. in case of
2048 * multiple disks using raid profile, a large IO can be split to
2049 * several segments of stripe length (currently 64K).
2051 blk_start_plug(&plug);
2052 atomic_inc(&BTRFS_I(inode)->sync_writers);
2053 ret = btrfs_fdatawrite_range(inode, start, end);
2054 atomic_dec(&BTRFS_I(inode)->sync_writers);
2055 blk_finish_plug(&plug);
2060 static inline bool skip_inode_logging(const struct btrfs_log_ctx *ctx)
2062 struct btrfs_inode *inode = BTRFS_I(ctx->inode);
2063 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2065 if (btrfs_inode_in_log(inode, fs_info->generation) &&
2066 list_empty(&ctx->ordered_extents))
2070 * If we are doing a fast fsync we can not bail out if the inode's
2071 * last_trans is <= then the last committed transaction, because we only
2072 * update the last_trans of the inode during ordered extent completion,
2073 * and for a fast fsync we don't wait for that, we only wait for the
2074 * writeback to complete.
2076 if (inode->last_trans <= fs_info->last_trans_committed &&
2077 (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) ||
2078 list_empty(&ctx->ordered_extents)))
2085 * fsync call for both files and directories. This logs the inode into
2086 * the tree log instead of forcing full commits whenever possible.
2088 * It needs to call filemap_fdatawait so that all ordered extent updates are
2089 * in the metadata btree are up to date for copying to the log.
2091 * It drops the inode mutex before doing the tree log commit. This is an
2092 * important optimization for directories because holding the mutex prevents
2093 * new operations on the dir while we write to disk.
2095 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
2097 struct dentry *dentry = file_dentry(file);
2098 struct inode *inode = d_inode(dentry);
2099 struct btrfs_root *root = BTRFS_I(inode)->root;
2100 struct btrfs_trans_handle *trans;
2101 struct btrfs_log_ctx ctx;
2106 trace_btrfs_sync_file(file, datasync);
2108 btrfs_init_log_ctx(&ctx, inode);
2111 * Always set the range to a full range, otherwise we can get into
2112 * several problems, from missing file extent items to represent holes
2113 * when not using the NO_HOLES feature, to log tree corruption due to
2114 * races between hole detection during logging and completion of ordered
2115 * extents outside the range, to missing checksums due to ordered extents
2116 * for which we flushed only a subset of their pages.
2120 len = (u64)LLONG_MAX + 1;
2123 * We write the dirty pages in the range and wait until they complete
2124 * out of the ->i_mutex. If so, we can flush the dirty pages by
2125 * multi-task, and make the performance up. See
2126 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
2128 ret = start_ordered_ops(inode, start, end);
2135 * We take the dio_sem here because the tree log stuff can race with
2136 * lockless dio writes and get an extent map logged for an extent we
2137 * never waited on. We need it this high up for lockdep reasons.
2139 down_write(&BTRFS_I(inode)->dio_sem);
2141 atomic_inc(&root->log_batch);
2144 * Always check for the full sync flag while holding the inode's lock,
2145 * to avoid races with other tasks. The flag must be either set all the
2146 * time during logging or always off all the time while logging.
2148 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2149 &BTRFS_I(inode)->runtime_flags);
2152 * Before we acquired the inode's lock, someone may have dirtied more
2153 * pages in the target range. We need to make sure that writeback for
2154 * any such pages does not start while we are logging the inode, because
2155 * if it does, any of the following might happen when we are not doing a
2158 * 1) We log an extent after its writeback finishes but before its
2159 * checksums are added to the csum tree, leading to -EIO errors
2160 * when attempting to read the extent after a log replay.
2162 * 2) We can end up logging an extent before its writeback finishes.
2163 * Therefore after the log replay we will have a file extent item
2164 * pointing to an unwritten extent (and no data checksums as well).
2166 * So trigger writeback for any eventual new dirty pages and then we
2167 * wait for all ordered extents to complete below.
2169 ret = start_ordered_ops(inode, start, end);
2171 up_write(&BTRFS_I(inode)->dio_sem);
2172 inode_unlock(inode);
2177 * We have to do this here to avoid the priority inversion of waiting on
2178 * IO of a lower priority task while holding a transaction open.
2180 * For a full fsync we wait for the ordered extents to complete while
2181 * for a fast fsync we wait just for writeback to complete, and then
2182 * attach the ordered extents to the transaction so that a transaction
2183 * commit waits for their completion, to avoid data loss if we fsync,
2184 * the current transaction commits before the ordered extents complete
2185 * and a power failure happens right after that.
2188 ret = btrfs_wait_ordered_range(inode, start, len);
2191 * Get our ordered extents as soon as possible to avoid doing
2192 * checksum lookups in the csum tree, and use instead the
2193 * checksums attached to the ordered extents.
2195 btrfs_get_ordered_extents_for_logging(BTRFS_I(inode),
2196 &ctx.ordered_extents);
2197 ret = filemap_fdatawait_range(inode->i_mapping, start, end);
2201 goto out_release_extents;
2203 atomic_inc(&root->log_batch);
2206 if (skip_inode_logging(&ctx)) {
2208 * We've had everything committed since the last time we were
2209 * modified so clear this flag in case it was set for whatever
2210 * reason, it's no longer relevant.
2212 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2213 &BTRFS_I(inode)->runtime_flags);
2215 * An ordered extent might have started before and completed
2216 * already with io errors, in which case the inode was not
2217 * updated and we end up here. So check the inode's mapping
2218 * for any errors that might have happened since we last
2219 * checked called fsync.
2221 ret = filemap_check_wb_err(inode->i_mapping, file->f_wb_err);
2222 goto out_release_extents;
2226 * We use start here because we will need to wait on the IO to complete
2227 * in btrfs_sync_log, which could require joining a transaction (for
2228 * example checking cross references in the nocow path). If we use join
2229 * here we could get into a situation where we're waiting on IO to
2230 * happen that is blocked on a transaction trying to commit. With start
2231 * we inc the extwriter counter, so we wait for all extwriters to exit
2232 * before we start blocking joiners. This comment is to keep somebody
2233 * from thinking they are super smart and changing this to
2234 * btrfs_join_transaction *cough*Josef*cough*.
2236 trans = btrfs_start_transaction(root, 0);
2237 if (IS_ERR(trans)) {
2238 ret = PTR_ERR(trans);
2239 goto out_release_extents;
2242 ret = btrfs_log_dentry_safe(trans, dentry, &ctx);
2243 btrfs_release_log_ctx_extents(&ctx);
2245 /* Fallthrough and commit/free transaction. */
2249 /* we've logged all the items and now have a consistent
2250 * version of the file in the log. It is possible that
2251 * someone will come in and modify the file, but that's
2252 * fine because the log is consistent on disk, and we
2253 * have references to all of the file's extents
2255 * It is possible that someone will come in and log the
2256 * file again, but that will end up using the synchronization
2257 * inside btrfs_sync_log to keep things safe.
2259 up_write(&BTRFS_I(inode)->dio_sem);
2260 inode_unlock(inode);
2262 if (ret != BTRFS_NO_LOG_SYNC) {
2264 ret = btrfs_sync_log(trans, root, &ctx);
2266 ret = btrfs_end_transaction(trans);
2271 ret = btrfs_wait_ordered_range(inode, start, len);
2273 btrfs_end_transaction(trans);
2277 ret = btrfs_commit_transaction(trans);
2279 ret = btrfs_end_transaction(trans);
2282 ASSERT(list_empty(&ctx.list));
2283 err = file_check_and_advance_wb_err(file);
2286 return ret > 0 ? -EIO : ret;
2288 out_release_extents:
2289 btrfs_release_log_ctx_extents(&ctx);
2290 up_write(&BTRFS_I(inode)->dio_sem);
2291 inode_unlock(inode);
2295 static const struct vm_operations_struct btrfs_file_vm_ops = {
2296 .fault = filemap_fault,
2297 .map_pages = filemap_map_pages,
2298 .page_mkwrite = btrfs_page_mkwrite,
2301 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
2303 struct address_space *mapping = filp->f_mapping;
2305 if (!mapping->a_ops->readpage)
2308 file_accessed(filp);
2309 vma->vm_ops = &btrfs_file_vm_ops;
2314 static int hole_mergeable(struct btrfs_inode *inode, struct extent_buffer *leaf,
2315 int slot, u64 start, u64 end)
2317 struct btrfs_file_extent_item *fi;
2318 struct btrfs_key key;
2320 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2323 btrfs_item_key_to_cpu(leaf, &key, slot);
2324 if (key.objectid != btrfs_ino(inode) ||
2325 key.type != BTRFS_EXTENT_DATA_KEY)
2328 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2330 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2333 if (btrfs_file_extent_disk_bytenr(leaf, fi))
2336 if (key.offset == end)
2338 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2343 static int fill_holes(struct btrfs_trans_handle *trans,
2344 struct btrfs_inode *inode,
2345 struct btrfs_path *path, u64 offset, u64 end)
2347 struct btrfs_fs_info *fs_info = trans->fs_info;
2348 struct btrfs_root *root = inode->root;
2349 struct extent_buffer *leaf;
2350 struct btrfs_file_extent_item *fi;
2351 struct extent_map *hole_em;
2352 struct extent_map_tree *em_tree = &inode->extent_tree;
2353 struct btrfs_key key;
2356 if (btrfs_fs_incompat(fs_info, NO_HOLES))
2359 key.objectid = btrfs_ino(inode);
2360 key.type = BTRFS_EXTENT_DATA_KEY;
2361 key.offset = offset;
2363 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2366 * We should have dropped this offset, so if we find it then
2367 * something has gone horribly wrong.
2374 leaf = path->nodes[0];
2375 if (hole_mergeable(inode, leaf, path->slots[0] - 1, offset, end)) {
2379 fi = btrfs_item_ptr(leaf, path->slots[0],
2380 struct btrfs_file_extent_item);
2381 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2383 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2384 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2385 btrfs_set_file_extent_offset(leaf, fi, 0);
2386 btrfs_mark_buffer_dirty(leaf);
2390 if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) {
2393 key.offset = offset;
2394 btrfs_set_item_key_safe(fs_info, path, &key);
2395 fi = btrfs_item_ptr(leaf, path->slots[0],
2396 struct btrfs_file_extent_item);
2397 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2399 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2400 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2401 btrfs_set_file_extent_offset(leaf, fi, 0);
2402 btrfs_mark_buffer_dirty(leaf);
2405 btrfs_release_path(path);
2407 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode),
2408 offset, 0, 0, end - offset, 0, end - offset, 0, 0, 0);
2413 btrfs_release_path(path);
2415 hole_em = alloc_extent_map();
2417 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2418 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
2420 hole_em->start = offset;
2421 hole_em->len = end - offset;
2422 hole_em->ram_bytes = hole_em->len;
2423 hole_em->orig_start = offset;
2425 hole_em->block_start = EXTENT_MAP_HOLE;
2426 hole_em->block_len = 0;
2427 hole_em->orig_block_len = 0;
2428 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2429 hole_em->generation = trans->transid;
2432 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2433 write_lock(&em_tree->lock);
2434 ret = add_extent_mapping(em_tree, hole_em, 1);
2435 write_unlock(&em_tree->lock);
2436 } while (ret == -EEXIST);
2437 free_extent_map(hole_em);
2439 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2440 &inode->runtime_flags);
2447 * Find a hole extent on given inode and change start/len to the end of hole
2448 * extent.(hole/vacuum extent whose em->start <= start &&
2449 * em->start + em->len > start)
2450 * When a hole extent is found, return 1 and modify start/len.
2452 static int find_first_non_hole(struct inode *inode, u64 *start, u64 *len)
2454 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2455 struct extent_map *em;
2458 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0,
2459 round_down(*start, fs_info->sectorsize),
2460 round_up(*len, fs_info->sectorsize));
2464 /* Hole or vacuum extent(only exists in no-hole mode) */
2465 if (em->block_start == EXTENT_MAP_HOLE) {
2467 *len = em->start + em->len > *start + *len ?
2468 0 : *start + *len - em->start - em->len;
2469 *start = em->start + em->len;
2471 free_extent_map(em);
2475 static int btrfs_punch_hole_lock_range(struct inode *inode,
2476 const u64 lockstart,
2478 struct extent_state **cached_state)
2481 struct btrfs_ordered_extent *ordered;
2484 truncate_pagecache_range(inode, lockstart, lockend);
2486 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2488 ordered = btrfs_lookup_first_ordered_extent(BTRFS_I(inode),
2492 * We need to make sure we have no ordered extents in this range
2493 * and nobody raced in and read a page in this range, if we did
2494 * we need to try again.
2497 (ordered->file_offset + ordered->num_bytes <= lockstart ||
2498 ordered->file_offset > lockend)) &&
2499 !filemap_range_has_page(inode->i_mapping,
2500 lockstart, lockend)) {
2502 btrfs_put_ordered_extent(ordered);
2506 btrfs_put_ordered_extent(ordered);
2507 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2508 lockend, cached_state);
2509 ret = btrfs_wait_ordered_range(inode, lockstart,
2510 lockend - lockstart + 1);
2517 static int btrfs_insert_replace_extent(struct btrfs_trans_handle *trans,
2518 struct inode *inode,
2519 struct btrfs_path *path,
2520 struct btrfs_replace_extent_info *extent_info,
2521 const u64 replace_len)
2523 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2524 struct btrfs_root *root = BTRFS_I(inode)->root;
2525 struct btrfs_file_extent_item *extent;
2526 struct extent_buffer *leaf;
2527 struct btrfs_key key;
2529 struct btrfs_ref ref = { 0 };
2532 if (replace_len == 0)
2535 if (extent_info->disk_offset == 0 &&
2536 btrfs_fs_incompat(fs_info, NO_HOLES))
2539 key.objectid = btrfs_ino(BTRFS_I(inode));
2540 key.type = BTRFS_EXTENT_DATA_KEY;
2541 key.offset = extent_info->file_offset;
2542 ret = btrfs_insert_empty_item(trans, root, path, &key,
2543 sizeof(struct btrfs_file_extent_item));
2546 leaf = path->nodes[0];
2547 slot = path->slots[0];
2548 write_extent_buffer(leaf, extent_info->extent_buf,
2549 btrfs_item_ptr_offset(leaf, slot),
2550 sizeof(struct btrfs_file_extent_item));
2551 extent = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2552 ASSERT(btrfs_file_extent_type(leaf, extent) != BTRFS_FILE_EXTENT_INLINE);
2553 btrfs_set_file_extent_offset(leaf, extent, extent_info->data_offset);
2554 btrfs_set_file_extent_num_bytes(leaf, extent, replace_len);
2555 if (extent_info->is_new_extent)
2556 btrfs_set_file_extent_generation(leaf, extent, trans->transid);
2557 btrfs_mark_buffer_dirty(leaf);
2558 btrfs_release_path(path);
2560 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode),
2561 extent_info->file_offset, replace_len);
2565 /* If it's a hole, nothing more needs to be done. */
2566 if (extent_info->disk_offset == 0)
2569 inode_add_bytes(inode, replace_len);
2571 if (extent_info->is_new_extent && extent_info->insertions == 0) {
2572 key.objectid = extent_info->disk_offset;
2573 key.type = BTRFS_EXTENT_ITEM_KEY;
2574 key.offset = extent_info->disk_len;
2575 ret = btrfs_alloc_reserved_file_extent(trans, root,
2576 btrfs_ino(BTRFS_I(inode)),
2577 extent_info->file_offset,
2578 extent_info->qgroup_reserved,
2583 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
2584 extent_info->disk_offset,
2585 extent_info->disk_len, 0);
2586 ref_offset = extent_info->file_offset - extent_info->data_offset;
2587 btrfs_init_data_ref(&ref, root->root_key.objectid,
2588 btrfs_ino(BTRFS_I(inode)), ref_offset);
2589 ret = btrfs_inc_extent_ref(trans, &ref);
2592 extent_info->insertions++;
2598 * The respective range must have been previously locked, as well as the inode.
2599 * The end offset is inclusive (last byte of the range).
2600 * @extent_info is NULL for fallocate's hole punching and non-NULL when replacing
2601 * the file range with an extent.
2602 * When not punching a hole, we don't want to end up in a state where we dropped
2603 * extents without inserting a new one, so we must abort the transaction to avoid
2606 int btrfs_replace_file_extents(struct inode *inode, struct btrfs_path *path,
2607 const u64 start, const u64 end,
2608 struct btrfs_replace_extent_info *extent_info,
2609 struct btrfs_trans_handle **trans_out)
2611 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2612 u64 min_size = btrfs_calc_insert_metadata_size(fs_info, 1);
2613 u64 ino_size = round_up(inode->i_size, fs_info->sectorsize);
2614 struct btrfs_root *root = BTRFS_I(inode)->root;
2615 struct btrfs_trans_handle *trans = NULL;
2616 struct btrfs_block_rsv *rsv;
2617 unsigned int rsv_count;
2620 u64 len = end - start;
2626 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
2631 rsv->size = btrfs_calc_insert_metadata_size(fs_info, 1);
2635 * 1 - update the inode
2636 * 1 - removing the extents in the range
2637 * 1 - adding the hole extent if no_holes isn't set or if we are
2638 * replacing the range with a new extent
2640 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || extent_info)
2645 trans = btrfs_start_transaction(root, rsv_count);
2646 if (IS_ERR(trans)) {
2647 ret = PTR_ERR(trans);
2652 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
2655 trans->block_rsv = rsv;
2658 while (cur_offset < end) {
2659 ret = __btrfs_drop_extents(trans, root, BTRFS_I(inode), path,
2660 cur_offset, end + 1, &drop_end,
2662 if (ret != -ENOSPC) {
2664 * The only time we don't want to abort is if we are
2665 * attempting to clone a partial inline extent, in which
2666 * case we'll get EOPNOTSUPP. However if we aren't
2667 * clone we need to abort no matter what, because if we
2668 * got EOPNOTSUPP via prealloc then we messed up and
2672 (ret != -EOPNOTSUPP ||
2673 (extent_info && extent_info->is_new_extent)))
2674 btrfs_abort_transaction(trans, ret);
2678 trans->block_rsv = &fs_info->trans_block_rsv;
2680 if (!extent_info && cur_offset < drop_end &&
2681 cur_offset < ino_size) {
2682 ret = fill_holes(trans, BTRFS_I(inode), path,
2683 cur_offset, drop_end);
2686 * If we failed then we didn't insert our hole
2687 * entries for the area we dropped, so now the
2688 * fs is corrupted, so we must abort the
2691 btrfs_abort_transaction(trans, ret);
2694 } else if (!extent_info && cur_offset < drop_end) {
2696 * We are past the i_size here, but since we didn't
2697 * insert holes we need to clear the mapped area so we
2698 * know to not set disk_i_size in this area until a new
2699 * file extent is inserted here.
2701 ret = btrfs_inode_clear_file_extent_range(BTRFS_I(inode),
2702 cur_offset, drop_end - cur_offset);
2705 * We couldn't clear our area, so we could
2706 * presumably adjust up and corrupt the fs, so
2709 btrfs_abort_transaction(trans, ret);
2714 if (extent_info && drop_end > extent_info->file_offset) {
2715 u64 replace_len = drop_end - extent_info->file_offset;
2717 ret = btrfs_insert_replace_extent(trans, inode, path,
2718 extent_info, replace_len);
2720 btrfs_abort_transaction(trans, ret);
2723 extent_info->data_len -= replace_len;
2724 extent_info->data_offset += replace_len;
2725 extent_info->file_offset += replace_len;
2728 cur_offset = drop_end;
2730 ret = btrfs_update_inode(trans, root, inode);
2734 btrfs_end_transaction(trans);
2735 btrfs_btree_balance_dirty(fs_info);
2737 trans = btrfs_start_transaction(root, rsv_count);
2738 if (IS_ERR(trans)) {
2739 ret = PTR_ERR(trans);
2744 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
2745 rsv, min_size, false);
2746 BUG_ON(ret); /* shouldn't happen */
2747 trans->block_rsv = rsv;
2750 ret = find_first_non_hole(inode, &cur_offset, &len);
2751 if (unlikely(ret < 0))
2761 * If we were cloning, force the next fsync to be a full one since we
2762 * we replaced (or just dropped in the case of cloning holes when
2763 * NO_HOLES is enabled) extents and extent maps.
2764 * This is for the sake of simplicity, and cloning into files larger
2765 * than 16Mb would force the full fsync any way (when
2766 * try_release_extent_mapping() is invoked during page cache truncation.
2768 if (extent_info && !extent_info->is_new_extent)
2769 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2770 &BTRFS_I(inode)->runtime_flags);
2775 trans->block_rsv = &fs_info->trans_block_rsv;
2777 * If we are using the NO_HOLES feature we might have had already an
2778 * hole that overlaps a part of the region [lockstart, lockend] and
2779 * ends at (or beyond) lockend. Since we have no file extent items to
2780 * represent holes, drop_end can be less than lockend and so we must
2781 * make sure we have an extent map representing the existing hole (the
2782 * call to __btrfs_drop_extents() might have dropped the existing extent
2783 * map representing the existing hole), otherwise the fast fsync path
2784 * will not record the existence of the hole region
2785 * [existing_hole_start, lockend].
2787 if (drop_end <= end)
2790 * Don't insert file hole extent item if it's for a range beyond eof
2791 * (because it's useless) or if it represents a 0 bytes range (when
2792 * cur_offset == drop_end).
2794 if (!extent_info && cur_offset < ino_size && cur_offset < drop_end) {
2795 ret = fill_holes(trans, BTRFS_I(inode), path,
2796 cur_offset, drop_end);
2798 /* Same comment as above. */
2799 btrfs_abort_transaction(trans, ret);
2802 } else if (!extent_info && cur_offset < drop_end) {
2803 /* See the comment in the loop above for the reasoning here. */
2804 ret = btrfs_inode_clear_file_extent_range(BTRFS_I(inode),
2805 cur_offset, drop_end - cur_offset);
2807 btrfs_abort_transaction(trans, ret);
2813 ret = btrfs_insert_replace_extent(trans, inode, path, extent_info,
2814 extent_info->data_len);
2816 btrfs_abort_transaction(trans, ret);
2825 trans->block_rsv = &fs_info->trans_block_rsv;
2827 btrfs_end_transaction(trans);
2831 btrfs_free_block_rsv(fs_info, rsv);
2836 static int btrfs_punch_hole(struct file *file, loff_t offset, loff_t len)
2838 struct inode *inode = file_inode(file);
2839 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2840 struct btrfs_root *root = BTRFS_I(inode)->root;
2841 struct extent_state *cached_state = NULL;
2842 struct btrfs_path *path;
2843 struct btrfs_trans_handle *trans = NULL;
2848 u64 orig_start = offset;
2852 bool truncated_block = false;
2853 bool updated_inode = false;
2855 ret = btrfs_wait_ordered_range(inode, offset, len);
2860 ino_size = round_up(inode->i_size, fs_info->sectorsize);
2861 ret = find_first_non_hole(inode, &offset, &len);
2863 goto out_only_mutex;
2865 /* Already in a large hole */
2867 goto out_only_mutex;
2870 ret = file_modified(file);
2872 goto out_only_mutex;
2874 lockstart = round_up(offset, btrfs_inode_sectorsize(BTRFS_I(inode)));
2875 lockend = round_down(offset + len,
2876 btrfs_inode_sectorsize(BTRFS_I(inode))) - 1;
2877 same_block = (BTRFS_BYTES_TO_BLKS(fs_info, offset))
2878 == (BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1));
2880 * We needn't truncate any block which is beyond the end of the file
2881 * because we are sure there is no data there.
2884 * Only do this if we are in the same block and we aren't doing the
2887 if (same_block && len < fs_info->sectorsize) {
2888 if (offset < ino_size) {
2889 truncated_block = true;
2890 ret = btrfs_truncate_block(inode, offset, len, 0);
2894 goto out_only_mutex;
2897 /* zero back part of the first block */
2898 if (offset < ino_size) {
2899 truncated_block = true;
2900 ret = btrfs_truncate_block(inode, offset, 0, 0);
2902 inode_unlock(inode);
2907 /* Check the aligned pages after the first unaligned page,
2908 * if offset != orig_start, which means the first unaligned page
2909 * including several following pages are already in holes,
2910 * the extra check can be skipped */
2911 if (offset == orig_start) {
2912 /* after truncate page, check hole again */
2913 len = offset + len - lockstart;
2915 ret = find_first_non_hole(inode, &offset, &len);
2917 goto out_only_mutex;
2920 goto out_only_mutex;
2925 /* Check the tail unaligned part is in a hole */
2926 tail_start = lockend + 1;
2927 tail_len = offset + len - tail_start;
2929 ret = find_first_non_hole(inode, &tail_start, &tail_len);
2930 if (unlikely(ret < 0))
2931 goto out_only_mutex;
2933 /* zero the front end of the last page */
2934 if (tail_start + tail_len < ino_size) {
2935 truncated_block = true;
2936 ret = btrfs_truncate_block(inode,
2937 tail_start + tail_len,
2940 goto out_only_mutex;
2945 if (lockend < lockstart) {
2947 goto out_only_mutex;
2950 ret = btrfs_punch_hole_lock_range(inode, lockstart, lockend,
2953 goto out_only_mutex;
2955 path = btrfs_alloc_path();
2961 ret = btrfs_replace_file_extents(inode, path, lockstart, lockend, NULL,
2963 btrfs_free_path(path);
2967 ASSERT(trans != NULL);
2968 inode_inc_iversion(inode);
2969 inode->i_mtime = inode->i_ctime = current_time(inode);
2970 ret = btrfs_update_inode(trans, root, inode);
2971 updated_inode = true;
2972 btrfs_end_transaction(trans);
2973 btrfs_btree_balance_dirty(fs_info);
2975 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2978 if (!updated_inode && truncated_block && !ret) {
2980 * If we only end up zeroing part of a page, we still need to
2981 * update the inode item, so that all the time fields are
2982 * updated as well as the necessary btrfs inode in memory fields
2983 * for detecting, at fsync time, if the inode isn't yet in the
2984 * log tree or it's there but not up to date.
2986 struct timespec64 now = current_time(inode);
2988 inode_inc_iversion(inode);
2989 inode->i_mtime = now;
2990 inode->i_ctime = now;
2991 trans = btrfs_start_transaction(root, 1);
2992 if (IS_ERR(trans)) {
2993 ret = PTR_ERR(trans);
2997 ret = btrfs_update_inode(trans, root, inode);
2998 ret2 = btrfs_end_transaction(trans);
3003 inode_unlock(inode);
3007 /* Helper structure to record which range is already reserved */
3008 struct falloc_range {
3009 struct list_head list;
3015 * Helper function to add falloc range
3017 * Caller should have locked the larger range of extent containing
3020 static int add_falloc_range(struct list_head *head, u64 start, u64 len)
3022 struct falloc_range *prev = NULL;
3023 struct falloc_range *range = NULL;
3025 if (list_empty(head))
3029 * As fallocate iterate by bytenr order, we only need to check
3032 prev = list_entry(head->prev, struct falloc_range, list);
3033 if (prev->start + prev->len == start) {
3038 range = kmalloc(sizeof(*range), GFP_KERNEL);
3041 range->start = start;
3043 list_add_tail(&range->list, head);
3047 static int btrfs_fallocate_update_isize(struct inode *inode,
3051 struct btrfs_trans_handle *trans;
3052 struct btrfs_root *root = BTRFS_I(inode)->root;
3056 if (mode & FALLOC_FL_KEEP_SIZE || end <= i_size_read(inode))
3059 trans = btrfs_start_transaction(root, 1);
3061 return PTR_ERR(trans);
3063 inode->i_ctime = current_time(inode);
3064 i_size_write(inode, end);
3065 btrfs_inode_safe_disk_i_size_write(inode, 0);
3066 ret = btrfs_update_inode(trans, root, inode);
3067 ret2 = btrfs_end_transaction(trans);
3069 return ret ? ret : ret2;
3073 RANGE_BOUNDARY_WRITTEN_EXTENT,
3074 RANGE_BOUNDARY_PREALLOC_EXTENT,
3075 RANGE_BOUNDARY_HOLE,
3078 static int btrfs_zero_range_check_range_boundary(struct btrfs_inode *inode,
3081 const u64 sectorsize = btrfs_inode_sectorsize(inode);
3082 struct extent_map *em;
3085 offset = round_down(offset, sectorsize);
3086 em = btrfs_get_extent(inode, NULL, 0, offset, sectorsize);
3090 if (em->block_start == EXTENT_MAP_HOLE)
3091 ret = RANGE_BOUNDARY_HOLE;
3092 else if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
3093 ret = RANGE_BOUNDARY_PREALLOC_EXTENT;
3095 ret = RANGE_BOUNDARY_WRITTEN_EXTENT;
3097 free_extent_map(em);
3101 static int btrfs_zero_range(struct inode *inode,
3106 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3107 struct extent_map *em;
3108 struct extent_changeset *data_reserved = NULL;
3111 const u64 sectorsize = btrfs_inode_sectorsize(BTRFS_I(inode));
3112 u64 alloc_start = round_down(offset, sectorsize);
3113 u64 alloc_end = round_up(offset + len, sectorsize);
3114 u64 bytes_to_reserve = 0;
3115 bool space_reserved = false;
3117 inode_dio_wait(inode);
3119 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start,
3120 alloc_end - alloc_start);
3127 * Avoid hole punching and extent allocation for some cases. More cases
3128 * could be considered, but these are unlikely common and we keep things
3129 * as simple as possible for now. Also, intentionally, if the target
3130 * range contains one or more prealloc extents together with regular
3131 * extents and holes, we drop all the existing extents and allocate a
3132 * new prealloc extent, so that we get a larger contiguous disk extent.
3134 if (em->start <= alloc_start &&
3135 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3136 const u64 em_end = em->start + em->len;
3138 if (em_end >= offset + len) {
3140 * The whole range is already a prealloc extent,
3141 * do nothing except updating the inode's i_size if
3144 free_extent_map(em);
3145 ret = btrfs_fallocate_update_isize(inode, offset + len,
3150 * Part of the range is already a prealloc extent, so operate
3151 * only on the remaining part of the range.
3153 alloc_start = em_end;
3154 ASSERT(IS_ALIGNED(alloc_start, sectorsize));
3155 len = offset + len - alloc_start;
3156 offset = alloc_start;
3157 alloc_hint = em->block_start + em->len;
3159 free_extent_map(em);
3161 if (BTRFS_BYTES_TO_BLKS(fs_info, offset) ==
3162 BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)) {
3163 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start,
3170 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3171 free_extent_map(em);
3172 ret = btrfs_fallocate_update_isize(inode, offset + len,
3176 if (len < sectorsize && em->block_start != EXTENT_MAP_HOLE) {
3177 free_extent_map(em);
3178 ret = btrfs_truncate_block(inode, offset, len, 0);
3180 ret = btrfs_fallocate_update_isize(inode,
3185 free_extent_map(em);
3186 alloc_start = round_down(offset, sectorsize);
3187 alloc_end = alloc_start + sectorsize;
3191 alloc_start = round_up(offset, sectorsize);
3192 alloc_end = round_down(offset + len, sectorsize);
3195 * For unaligned ranges, check the pages at the boundaries, they might
3196 * map to an extent, in which case we need to partially zero them, or
3197 * they might map to a hole, in which case we need our allocation range
3200 if (!IS_ALIGNED(offset, sectorsize)) {
3201 ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode),
3205 if (ret == RANGE_BOUNDARY_HOLE) {
3206 alloc_start = round_down(offset, sectorsize);
3208 } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
3209 ret = btrfs_truncate_block(inode, offset, 0, 0);
3217 if (!IS_ALIGNED(offset + len, sectorsize)) {
3218 ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode),
3222 if (ret == RANGE_BOUNDARY_HOLE) {
3223 alloc_end = round_up(offset + len, sectorsize);
3225 } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
3226 ret = btrfs_truncate_block(inode, offset + len, 0, 1);
3235 if (alloc_start < alloc_end) {
3236 struct extent_state *cached_state = NULL;
3237 const u64 lockstart = alloc_start;
3238 const u64 lockend = alloc_end - 1;
3240 bytes_to_reserve = alloc_end - alloc_start;
3241 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
3245 space_reserved = true;
3246 ret = btrfs_punch_hole_lock_range(inode, lockstart, lockend,
3250 ret = btrfs_qgroup_reserve_data(BTRFS_I(inode), &data_reserved,
3251 alloc_start, bytes_to_reserve);
3253 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
3254 lockend, &cached_state);
3257 ret = btrfs_prealloc_file_range(inode, mode, alloc_start,
3258 alloc_end - alloc_start,
3260 offset + len, &alloc_hint);
3261 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
3262 lockend, &cached_state);
3263 /* btrfs_prealloc_file_range releases reserved space on error */
3265 space_reserved = false;
3269 ret = btrfs_fallocate_update_isize(inode, offset + len, mode);
3271 if (ret && space_reserved)
3272 btrfs_free_reserved_data_space(BTRFS_I(inode), data_reserved,
3273 alloc_start, bytes_to_reserve);
3274 extent_changeset_free(data_reserved);
3279 static long btrfs_fallocate(struct file *file, int mode,
3280 loff_t offset, loff_t len)
3282 struct inode *inode = file_inode(file);
3283 struct extent_state *cached_state = NULL;
3284 struct extent_changeset *data_reserved = NULL;
3285 struct falloc_range *range;
3286 struct falloc_range *tmp;
3287 struct list_head reserve_list;
3295 struct extent_map *em;
3296 int blocksize = btrfs_inode_sectorsize(BTRFS_I(inode));
3299 alloc_start = round_down(offset, blocksize);
3300 alloc_end = round_up(offset + len, blocksize);
3301 cur_offset = alloc_start;
3303 /* Make sure we aren't being give some crap mode */
3304 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
3305 FALLOC_FL_ZERO_RANGE))
3308 if (mode & FALLOC_FL_PUNCH_HOLE)
3309 return btrfs_punch_hole(file, offset, len);
3312 * Only trigger disk allocation, don't trigger qgroup reserve
3314 * For qgroup space, it will be checked later.
3316 if (!(mode & FALLOC_FL_ZERO_RANGE)) {
3317 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
3318 alloc_end - alloc_start);
3325 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) {
3326 ret = inode_newsize_ok(inode, offset + len);
3331 ret = file_modified(file);
3336 * TODO: Move these two operations after we have checked
3337 * accurate reserved space, or fallocate can still fail but
3338 * with page truncated or size expanded.
3340 * But that's a minor problem and won't do much harm BTW.
3342 if (alloc_start > inode->i_size) {
3343 ret = btrfs_cont_expand(inode, i_size_read(inode),
3347 } else if (offset + len > inode->i_size) {
3349 * If we are fallocating from the end of the file onward we
3350 * need to zero out the end of the block if i_size lands in the
3351 * middle of a block.
3353 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
3359 * wait for ordered IO before we have any locks. We'll loop again
3360 * below with the locks held.
3362 ret = btrfs_wait_ordered_range(inode, alloc_start,
3363 alloc_end - alloc_start);
3367 if (mode & FALLOC_FL_ZERO_RANGE) {
3368 ret = btrfs_zero_range(inode, offset, len, mode);
3369 inode_unlock(inode);
3373 locked_end = alloc_end - 1;
3375 struct btrfs_ordered_extent *ordered;
3377 /* the extent lock is ordered inside the running
3380 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
3381 locked_end, &cached_state);
3382 ordered = btrfs_lookup_first_ordered_extent(BTRFS_I(inode),
3386 ordered->file_offset + ordered->num_bytes > alloc_start &&
3387 ordered->file_offset < alloc_end) {
3388 btrfs_put_ordered_extent(ordered);
3389 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
3390 alloc_start, locked_end,
3393 * we can't wait on the range with the transaction
3394 * running or with the extent lock held
3396 ret = btrfs_wait_ordered_range(inode, alloc_start,
3397 alloc_end - alloc_start);
3402 btrfs_put_ordered_extent(ordered);
3407 /* First, check if we exceed the qgroup limit */
3408 INIT_LIST_HEAD(&reserve_list);
3409 while (cur_offset < alloc_end) {
3410 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
3411 alloc_end - cur_offset);
3416 last_byte = min(extent_map_end(em), alloc_end);
3417 actual_end = min_t(u64, extent_map_end(em), offset + len);
3418 last_byte = ALIGN(last_byte, blocksize);
3419 if (em->block_start == EXTENT_MAP_HOLE ||
3420 (cur_offset >= inode->i_size &&
3421 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
3422 ret = add_falloc_range(&reserve_list, cur_offset,
3423 last_byte - cur_offset);
3425 free_extent_map(em);
3428 ret = btrfs_qgroup_reserve_data(BTRFS_I(inode),
3429 &data_reserved, cur_offset,
3430 last_byte - cur_offset);
3432 cur_offset = last_byte;
3433 free_extent_map(em);
3438 * Do not need to reserve unwritten extent for this
3439 * range, free reserved data space first, otherwise
3440 * it'll result in false ENOSPC error.
3442 btrfs_free_reserved_data_space(BTRFS_I(inode),
3443 data_reserved, cur_offset,
3444 last_byte - cur_offset);
3446 free_extent_map(em);
3447 cur_offset = last_byte;
3451 * If ret is still 0, means we're OK to fallocate.
3452 * Or just cleanup the list and exit.
3454 list_for_each_entry_safe(range, tmp, &reserve_list, list) {
3456 ret = btrfs_prealloc_file_range(inode, mode,
3458 range->len, i_blocksize(inode),
3459 offset + len, &alloc_hint);
3461 btrfs_free_reserved_data_space(BTRFS_I(inode),
3462 data_reserved, range->start,
3464 list_del(&range->list);
3471 * We didn't need to allocate any more space, but we still extended the
3472 * size of the file so we need to update i_size and the inode item.
3474 ret = btrfs_fallocate_update_isize(inode, actual_end, mode);
3476 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
3479 inode_unlock(inode);
3480 /* Let go of our reservation. */
3481 if (ret != 0 && !(mode & FALLOC_FL_ZERO_RANGE))
3482 btrfs_free_reserved_data_space(BTRFS_I(inode), data_reserved,
3483 cur_offset, alloc_end - cur_offset);
3484 extent_changeset_free(data_reserved);
3488 static loff_t find_desired_extent(struct inode *inode, loff_t offset,
3491 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3492 struct extent_map *em = NULL;
3493 struct extent_state *cached_state = NULL;
3494 loff_t i_size = inode->i_size;
3501 if (i_size == 0 || offset >= i_size)
3505 * offset can be negative, in this case we start finding DATA/HOLE from
3506 * the very start of the file.
3508 start = max_t(loff_t, 0, offset);
3510 lockstart = round_down(start, fs_info->sectorsize);
3511 lockend = round_up(i_size, fs_info->sectorsize);
3512 if (lockend <= lockstart)
3513 lockend = lockstart + fs_info->sectorsize;
3515 len = lockend - lockstart + 1;
3517 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
3520 while (start < i_size) {
3521 em = btrfs_get_extent_fiemap(BTRFS_I(inode), start, len);
3528 if (whence == SEEK_HOLE &&
3529 (em->block_start == EXTENT_MAP_HOLE ||
3530 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
3532 else if (whence == SEEK_DATA &&
3533 (em->block_start != EXTENT_MAP_HOLE &&
3534 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
3537 start = em->start + em->len;
3538 free_extent_map(em);
3542 free_extent_map(em);
3543 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
3548 if (whence == SEEK_DATA && start >= i_size)
3551 offset = min_t(loff_t, start, i_size);
3557 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
3559 struct inode *inode = file->f_mapping->host;
3563 return generic_file_llseek(file, offset, whence);
3566 inode_lock_shared(inode);
3567 offset = find_desired_extent(inode, offset, whence);
3568 inode_unlock_shared(inode);
3575 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
3578 static int btrfs_file_open(struct inode *inode, struct file *filp)
3580 filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
3581 return generic_file_open(inode, filp);
3584 static ssize_t btrfs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
3588 if (iocb->ki_flags & IOCB_DIRECT) {
3589 struct inode *inode = file_inode(iocb->ki_filp);
3591 inode_lock_shared(inode);
3592 ret = btrfs_direct_IO(iocb, to);
3593 inode_unlock_shared(inode);
3594 if (ret < 0 || !iov_iter_count(to) ||
3595 iocb->ki_pos >= i_size_read(file_inode(iocb->ki_filp)))
3599 return generic_file_buffered_read(iocb, to, ret);
3602 const struct file_operations btrfs_file_operations = {
3603 .llseek = btrfs_file_llseek,
3604 .read_iter = btrfs_file_read_iter,
3605 .splice_read = generic_file_splice_read,
3606 .write_iter = btrfs_file_write_iter,
3607 .splice_write = iter_file_splice_write,
3608 .mmap = btrfs_file_mmap,
3609 .open = btrfs_file_open,
3610 .release = btrfs_release_file,
3611 .fsync = btrfs_sync_file,
3612 .fallocate = btrfs_fallocate,
3613 .unlocked_ioctl = btrfs_ioctl,
3614 #ifdef CONFIG_COMPAT
3615 .compat_ioctl = btrfs_compat_ioctl,
3617 .remap_file_range = btrfs_remap_file_range,
3620 void __cold btrfs_auto_defrag_exit(void)
3622 kmem_cache_destroy(btrfs_inode_defrag_cachep);
3625 int __init btrfs_auto_defrag_init(void)
3627 btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
3628 sizeof(struct inode_defrag), 0,
3631 if (!btrfs_inode_defrag_cachep)
3637 int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end)
3642 * So with compression we will find and lock a dirty page and clear the
3643 * first one as dirty, setup an async extent, and immediately return
3644 * with the entire range locked but with nobody actually marked with
3645 * writeback. So we can't just filemap_write_and_wait_range() and
3646 * expect it to work since it will just kick off a thread to do the
3647 * actual work. So we need to call filemap_fdatawrite_range _again_
3648 * since it will wait on the page lock, which won't be unlocked until
3649 * after the pages have been marked as writeback and so we're good to go
3650 * from there. We have to do this otherwise we'll miss the ordered
3651 * extents and that results in badness. Please Josef, do not think you
3652 * know better and pull this out at some point in the future, it is
3653 * right and you are wrong.
3655 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
3656 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
3657 &BTRFS_I(inode)->runtime_flags))
3658 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);