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, bool noreserve)
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 if (write_bytes == 0)
481 extra_bits |= EXTENT_NORESERVE;
483 start_pos = round_down(pos, fs_info->sectorsize);
484 num_bytes = round_up(write_bytes + pos - start_pos,
485 fs_info->sectorsize);
487 end_of_last_block = start_pos + num_bytes - 1;
490 * The pages may have already been dirty, clear out old accounting so
491 * we can set things up properly
493 clear_extent_bit(&inode->io_tree, start_pos, end_of_last_block,
494 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
497 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
502 for (i = 0; i < num_pages; i++) {
503 struct page *p = pages[i];
510 * we've only changed i_size in ram, and we haven't updated
511 * the disk i_size. There is no need to log the inode
515 i_size_write(&inode->vfs_inode, end_pos);
520 * this drops all the extents in the cache that intersect the range
521 * [start, end]. Existing extents are split as required.
523 void btrfs_drop_extent_cache(struct btrfs_inode *inode, u64 start, u64 end,
526 struct extent_map *em;
527 struct extent_map *split = NULL;
528 struct extent_map *split2 = NULL;
529 struct extent_map_tree *em_tree = &inode->extent_tree;
530 u64 len = end - start + 1;
538 WARN_ON(end < start);
539 if (end == (u64)-1) {
548 split = alloc_extent_map();
550 split2 = alloc_extent_map();
551 if (!split || !split2)
554 write_lock(&em_tree->lock);
555 em = lookup_extent_mapping(em_tree, start, len);
557 write_unlock(&em_tree->lock);
561 gen = em->generation;
562 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
563 if (testend && em->start + em->len >= start + len) {
565 write_unlock(&em_tree->lock);
568 start = em->start + em->len;
570 len = start + len - (em->start + em->len);
572 write_unlock(&em_tree->lock);
575 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
576 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
577 clear_bit(EXTENT_FLAG_LOGGING, &flags);
578 modified = !list_empty(&em->list);
582 if (em->start < start) {
583 split->start = em->start;
584 split->len = start - em->start;
586 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
587 split->orig_start = em->orig_start;
588 split->block_start = em->block_start;
591 split->block_len = em->block_len;
593 split->block_len = split->len;
594 split->orig_block_len = max(split->block_len,
596 split->ram_bytes = em->ram_bytes;
598 split->orig_start = split->start;
599 split->block_len = 0;
600 split->block_start = em->block_start;
601 split->orig_block_len = 0;
602 split->ram_bytes = split->len;
605 split->generation = gen;
606 split->flags = flags;
607 split->compress_type = em->compress_type;
608 replace_extent_mapping(em_tree, em, split, modified);
609 free_extent_map(split);
613 if (testend && em->start + em->len > start + len) {
614 u64 diff = start + len - em->start;
616 split->start = start + len;
617 split->len = em->start + em->len - (start + len);
618 split->flags = flags;
619 split->compress_type = em->compress_type;
620 split->generation = gen;
622 if (em->block_start < EXTENT_MAP_LAST_BYTE) {
623 split->orig_block_len = max(em->block_len,
626 split->ram_bytes = em->ram_bytes;
628 split->block_len = em->block_len;
629 split->block_start = em->block_start;
630 split->orig_start = em->orig_start;
632 split->block_len = split->len;
633 split->block_start = em->block_start
635 split->orig_start = em->orig_start;
638 split->ram_bytes = split->len;
639 split->orig_start = split->start;
640 split->block_len = 0;
641 split->block_start = em->block_start;
642 split->orig_block_len = 0;
645 if (extent_map_in_tree(em)) {
646 replace_extent_mapping(em_tree, em, split,
649 ret = add_extent_mapping(em_tree, split,
651 ASSERT(ret == 0); /* Logic error */
653 free_extent_map(split);
657 if (extent_map_in_tree(em))
658 remove_extent_mapping(em_tree, em);
659 write_unlock(&em_tree->lock);
663 /* once for the tree*/
667 free_extent_map(split);
669 free_extent_map(split2);
673 * this is very complex, but the basic idea is to drop all extents
674 * in the range start - end. hint_block is filled in with a block number
675 * that would be a good hint to the block allocator for this file.
677 * If an extent intersects the range but is not entirely inside the range
678 * it is either truncated or split. Anything entirely inside the range
679 * is deleted from the tree.
681 int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
682 struct btrfs_root *root, struct btrfs_inode *inode,
683 struct btrfs_path *path, u64 start, u64 end,
684 u64 *drop_end, int drop_cache,
686 u32 extent_item_size,
689 struct btrfs_fs_info *fs_info = root->fs_info;
690 struct extent_buffer *leaf;
691 struct btrfs_file_extent_item *fi;
692 struct btrfs_ref ref = { 0 };
693 struct btrfs_key key;
694 struct btrfs_key new_key;
695 struct inode *vfs_inode = &inode->vfs_inode;
696 u64 ino = btrfs_ino(inode);
697 u64 search_start = start;
700 u64 extent_offset = 0;
702 u64 last_end = start;
708 int modify_tree = -1;
711 int leafs_visited = 0;
714 btrfs_drop_extent_cache(inode, start, end - 1, 0);
716 if (start >= inode->disk_i_size && !replace_extent)
719 update_refs = (test_bit(BTRFS_ROOT_SHAREABLE, &root->state) ||
720 root == fs_info->tree_root);
723 ret = btrfs_lookup_file_extent(trans, root, path, ino,
724 search_start, modify_tree);
727 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
728 leaf = path->nodes[0];
729 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
730 if (key.objectid == ino &&
731 key.type == BTRFS_EXTENT_DATA_KEY)
737 leaf = path->nodes[0];
738 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
740 ret = btrfs_next_leaf(root, path);
748 leaf = path->nodes[0];
752 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
754 if (key.objectid > ino)
756 if (WARN_ON_ONCE(key.objectid < ino) ||
757 key.type < BTRFS_EXTENT_DATA_KEY) {
762 if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
765 fi = btrfs_item_ptr(leaf, path->slots[0],
766 struct btrfs_file_extent_item);
767 extent_type = btrfs_file_extent_type(leaf, fi);
769 if (extent_type == BTRFS_FILE_EXTENT_REG ||
770 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
771 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
772 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
773 extent_offset = btrfs_file_extent_offset(leaf, fi);
774 extent_end = key.offset +
775 btrfs_file_extent_num_bytes(leaf, fi);
776 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
777 extent_end = key.offset +
778 btrfs_file_extent_ram_bytes(leaf, fi);
785 * Don't skip extent items representing 0 byte lengths. They
786 * used to be created (bug) if while punching holes we hit
787 * -ENOSPC condition. So if we find one here, just ensure we
788 * delete it, otherwise we would insert a new file extent item
789 * with the same key (offset) as that 0 bytes length file
790 * extent item in the call to setup_items_for_insert() later
793 if (extent_end == key.offset && extent_end >= search_start) {
794 last_end = extent_end;
795 goto delete_extent_item;
798 if (extent_end <= search_start) {
804 search_start = max(key.offset, start);
805 if (recow || !modify_tree) {
807 btrfs_release_path(path);
812 * | - range to drop - |
813 * | -------- extent -------- |
815 if (start > key.offset && end < extent_end) {
817 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
822 memcpy(&new_key, &key, sizeof(new_key));
823 new_key.offset = start;
824 ret = btrfs_duplicate_item(trans, root, path,
826 if (ret == -EAGAIN) {
827 btrfs_release_path(path);
833 leaf = path->nodes[0];
834 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
835 struct btrfs_file_extent_item);
836 btrfs_set_file_extent_num_bytes(leaf, fi,
839 fi = btrfs_item_ptr(leaf, path->slots[0],
840 struct btrfs_file_extent_item);
842 extent_offset += start - key.offset;
843 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
844 btrfs_set_file_extent_num_bytes(leaf, fi,
846 btrfs_mark_buffer_dirty(leaf);
848 if (update_refs && disk_bytenr > 0) {
849 btrfs_init_generic_ref(&ref,
850 BTRFS_ADD_DELAYED_REF,
851 disk_bytenr, num_bytes, 0);
852 btrfs_init_data_ref(&ref,
853 root->root_key.objectid,
855 start - extent_offset);
856 ret = btrfs_inc_extent_ref(trans, &ref);
857 BUG_ON(ret); /* -ENOMEM */
862 * From here on out we will have actually dropped something, so
863 * last_end can be updated.
865 last_end = extent_end;
868 * | ---- range to drop ----- |
869 * | -------- extent -------- |
871 if (start <= key.offset && end < extent_end) {
872 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
877 memcpy(&new_key, &key, sizeof(new_key));
878 new_key.offset = end;
879 btrfs_set_item_key_safe(fs_info, path, &new_key);
881 extent_offset += end - key.offset;
882 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
883 btrfs_set_file_extent_num_bytes(leaf, fi,
885 btrfs_mark_buffer_dirty(leaf);
886 if (update_refs && disk_bytenr > 0)
887 inode_sub_bytes(vfs_inode, end - key.offset);
891 search_start = extent_end;
893 * | ---- range to drop ----- |
894 * | -------- extent -------- |
896 if (start > key.offset && end >= extent_end) {
898 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
903 btrfs_set_file_extent_num_bytes(leaf, fi,
905 btrfs_mark_buffer_dirty(leaf);
906 if (update_refs && disk_bytenr > 0)
907 inode_sub_bytes(vfs_inode, extent_end - start);
908 if (end == extent_end)
916 * | ---- range to drop ----- |
917 * | ------ extent ------ |
919 if (start <= key.offset && end >= extent_end) {
922 del_slot = path->slots[0];
925 BUG_ON(del_slot + del_nr != path->slots[0]);
930 extent_type == BTRFS_FILE_EXTENT_INLINE) {
931 inode_sub_bytes(vfs_inode,
932 extent_end - key.offset);
933 extent_end = ALIGN(extent_end,
934 fs_info->sectorsize);
935 } else if (update_refs && disk_bytenr > 0) {
936 btrfs_init_generic_ref(&ref,
937 BTRFS_DROP_DELAYED_REF,
938 disk_bytenr, num_bytes, 0);
939 btrfs_init_data_ref(&ref,
940 root->root_key.objectid,
942 key.offset - extent_offset);
943 ret = btrfs_free_extent(trans, &ref);
944 BUG_ON(ret); /* -ENOMEM */
945 inode_sub_bytes(vfs_inode,
946 extent_end - key.offset);
949 if (end == extent_end)
952 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
957 ret = btrfs_del_items(trans, root, path, del_slot,
960 btrfs_abort_transaction(trans, ret);
967 btrfs_release_path(path);
974 if (!ret && del_nr > 0) {
976 * Set path->slots[0] to first slot, so that after the delete
977 * if items are move off from our leaf to its immediate left or
978 * right neighbor leafs, we end up with a correct and adjusted
979 * path->slots[0] for our insertion (if replace_extent != 0).
981 path->slots[0] = del_slot;
982 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
984 btrfs_abort_transaction(trans, ret);
987 leaf = path->nodes[0];
989 * If btrfs_del_items() was called, it might have deleted a leaf, in
990 * which case it unlocked our path, so check path->locks[0] matches a
993 if (!ret && replace_extent && leafs_visited == 1 &&
994 path->locks[0] == BTRFS_WRITE_LOCK &&
995 btrfs_leaf_free_space(leaf) >=
996 sizeof(struct btrfs_item) + extent_item_size) {
999 key.type = BTRFS_EXTENT_DATA_KEY;
1001 if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
1002 struct btrfs_key slot_key;
1004 btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
1005 if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
1008 setup_items_for_insert(root, path, &key, &extent_item_size, 1);
1012 if (!replace_extent || !(*key_inserted))
1013 btrfs_release_path(path);
1015 *drop_end = found ? min(end, last_end) : end;
1019 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
1020 struct btrfs_root *root, struct inode *inode, u64 start,
1021 u64 end, int drop_cache)
1023 struct btrfs_path *path;
1026 path = btrfs_alloc_path();
1029 ret = __btrfs_drop_extents(trans, root, BTRFS_I(inode), path, start,
1030 end, NULL, drop_cache, 0, 0, NULL);
1031 btrfs_free_path(path);
1035 static int extent_mergeable(struct extent_buffer *leaf, int slot,
1036 u64 objectid, u64 bytenr, u64 orig_offset,
1037 u64 *start, u64 *end)
1039 struct btrfs_file_extent_item *fi;
1040 struct btrfs_key key;
1043 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1046 btrfs_item_key_to_cpu(leaf, &key, slot);
1047 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
1050 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1051 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
1052 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
1053 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
1054 btrfs_file_extent_compression(leaf, fi) ||
1055 btrfs_file_extent_encryption(leaf, fi) ||
1056 btrfs_file_extent_other_encoding(leaf, fi))
1059 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1060 if ((*start && *start != key.offset) || (*end && *end != extent_end))
1063 *start = key.offset;
1069 * Mark extent in the range start - end as written.
1071 * This changes extent type from 'pre-allocated' to 'regular'. If only
1072 * part of extent is marked as written, the extent will be split into
1075 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
1076 struct btrfs_inode *inode, u64 start, u64 end)
1078 struct btrfs_fs_info *fs_info = trans->fs_info;
1079 struct btrfs_root *root = inode->root;
1080 struct extent_buffer *leaf;
1081 struct btrfs_path *path;
1082 struct btrfs_file_extent_item *fi;
1083 struct btrfs_ref ref = { 0 };
1084 struct btrfs_key key;
1085 struct btrfs_key new_key;
1097 u64 ino = btrfs_ino(inode);
1099 path = btrfs_alloc_path();
1106 key.type = BTRFS_EXTENT_DATA_KEY;
1109 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1112 if (ret > 0 && path->slots[0] > 0)
1115 leaf = path->nodes[0];
1116 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1117 if (key.objectid != ino ||
1118 key.type != BTRFS_EXTENT_DATA_KEY) {
1120 btrfs_abort_transaction(trans, ret);
1123 fi = btrfs_item_ptr(leaf, path->slots[0],
1124 struct btrfs_file_extent_item);
1125 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_PREALLOC) {
1127 btrfs_abort_transaction(trans, ret);
1130 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1131 if (key.offset > start || extent_end < end) {
1133 btrfs_abort_transaction(trans, ret);
1137 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1138 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1139 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1140 memcpy(&new_key, &key, sizeof(new_key));
1142 if (start == key.offset && end < extent_end) {
1145 if (extent_mergeable(leaf, path->slots[0] - 1,
1146 ino, bytenr, orig_offset,
1147 &other_start, &other_end)) {
1148 new_key.offset = end;
1149 btrfs_set_item_key_safe(fs_info, path, &new_key);
1150 fi = btrfs_item_ptr(leaf, path->slots[0],
1151 struct btrfs_file_extent_item);
1152 btrfs_set_file_extent_generation(leaf, fi,
1154 btrfs_set_file_extent_num_bytes(leaf, fi,
1156 btrfs_set_file_extent_offset(leaf, fi,
1158 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1159 struct btrfs_file_extent_item);
1160 btrfs_set_file_extent_generation(leaf, fi,
1162 btrfs_set_file_extent_num_bytes(leaf, fi,
1164 btrfs_mark_buffer_dirty(leaf);
1169 if (start > key.offset && end == extent_end) {
1172 if (extent_mergeable(leaf, path->slots[0] + 1,
1173 ino, bytenr, orig_offset,
1174 &other_start, &other_end)) {
1175 fi = btrfs_item_ptr(leaf, path->slots[0],
1176 struct btrfs_file_extent_item);
1177 btrfs_set_file_extent_num_bytes(leaf, fi,
1178 start - key.offset);
1179 btrfs_set_file_extent_generation(leaf, fi,
1182 new_key.offset = start;
1183 btrfs_set_item_key_safe(fs_info, path, &new_key);
1185 fi = btrfs_item_ptr(leaf, path->slots[0],
1186 struct btrfs_file_extent_item);
1187 btrfs_set_file_extent_generation(leaf, fi,
1189 btrfs_set_file_extent_num_bytes(leaf, fi,
1191 btrfs_set_file_extent_offset(leaf, fi,
1192 start - orig_offset);
1193 btrfs_mark_buffer_dirty(leaf);
1198 while (start > key.offset || end < extent_end) {
1199 if (key.offset == start)
1202 new_key.offset = split;
1203 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1204 if (ret == -EAGAIN) {
1205 btrfs_release_path(path);
1209 btrfs_abort_transaction(trans, ret);
1213 leaf = path->nodes[0];
1214 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1215 struct btrfs_file_extent_item);
1216 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1217 btrfs_set_file_extent_num_bytes(leaf, fi,
1218 split - key.offset);
1220 fi = btrfs_item_ptr(leaf, path->slots[0],
1221 struct btrfs_file_extent_item);
1223 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1224 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1225 btrfs_set_file_extent_num_bytes(leaf, fi,
1226 extent_end - split);
1227 btrfs_mark_buffer_dirty(leaf);
1229 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, bytenr,
1231 btrfs_init_data_ref(&ref, root->root_key.objectid, ino,
1233 ret = btrfs_inc_extent_ref(trans, &ref);
1235 btrfs_abort_transaction(trans, ret);
1239 if (split == start) {
1242 if (start != key.offset) {
1244 btrfs_abort_transaction(trans, ret);
1255 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
1257 btrfs_init_data_ref(&ref, root->root_key.objectid, ino, orig_offset);
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 extent_end = other_end;
1266 del_slot = path->slots[0] + 1;
1268 ret = btrfs_free_extent(trans, &ref);
1270 btrfs_abort_transaction(trans, ret);
1276 if (extent_mergeable(leaf, path->slots[0] - 1,
1277 ino, bytenr, orig_offset,
1278 &other_start, &other_end)) {
1280 btrfs_release_path(path);
1283 key.offset = other_start;
1284 del_slot = path->slots[0];
1286 ret = btrfs_free_extent(trans, &ref);
1288 btrfs_abort_transaction(trans, ret);
1293 fi = btrfs_item_ptr(leaf, path->slots[0],
1294 struct btrfs_file_extent_item);
1295 btrfs_set_file_extent_type(leaf, fi,
1296 BTRFS_FILE_EXTENT_REG);
1297 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1298 btrfs_mark_buffer_dirty(leaf);
1300 fi = btrfs_item_ptr(leaf, del_slot - 1,
1301 struct btrfs_file_extent_item);
1302 btrfs_set_file_extent_type(leaf, fi,
1303 BTRFS_FILE_EXTENT_REG);
1304 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1305 btrfs_set_file_extent_num_bytes(leaf, fi,
1306 extent_end - key.offset);
1307 btrfs_mark_buffer_dirty(leaf);
1309 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1311 btrfs_abort_transaction(trans, ret);
1316 btrfs_free_path(path);
1321 * on error we return an unlocked page and the error value
1322 * on success we return a locked page and 0
1324 static int prepare_uptodate_page(struct inode *inode,
1325 struct page *page, u64 pos,
1326 bool force_uptodate)
1330 if (((pos & (PAGE_SIZE - 1)) || force_uptodate) &&
1331 !PageUptodate(page)) {
1332 ret = btrfs_readpage(NULL, page);
1336 if (!PageUptodate(page)) {
1340 if (page->mapping != inode->i_mapping) {
1349 * this just gets pages into the page cache and locks them down.
1351 static noinline int prepare_pages(struct inode *inode, struct page **pages,
1352 size_t num_pages, loff_t pos,
1353 size_t write_bytes, bool force_uptodate)
1356 unsigned long index = pos >> PAGE_SHIFT;
1357 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1361 for (i = 0; i < num_pages; i++) {
1363 pages[i] = find_or_create_page(inode->i_mapping, index + i,
1364 mask | __GFP_WRITE);
1372 err = prepare_uptodate_page(inode, pages[i], pos,
1374 if (!err && i == num_pages - 1)
1375 err = prepare_uptodate_page(inode, pages[i],
1376 pos + write_bytes, false);
1379 if (err == -EAGAIN) {
1386 wait_on_page_writeback(pages[i]);
1391 while (faili >= 0) {
1392 unlock_page(pages[faili]);
1393 put_page(pages[faili]);
1401 * This function locks the extent and properly waits for data=ordered extents
1402 * to finish before allowing the pages to be modified if need.
1405 * 1 - the extent is locked
1406 * 0 - the extent is not locked, and everything is OK
1407 * -EAGAIN - need re-prepare the pages
1408 * the other < 0 number - Something wrong happens
1411 lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct page **pages,
1412 size_t num_pages, loff_t pos,
1414 u64 *lockstart, u64 *lockend,
1415 struct extent_state **cached_state)
1417 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1423 start_pos = round_down(pos, fs_info->sectorsize);
1424 last_pos = round_up(pos + write_bytes, fs_info->sectorsize) - 1;
1426 if (start_pos < inode->vfs_inode.i_size) {
1427 struct btrfs_ordered_extent *ordered;
1429 lock_extent_bits(&inode->io_tree, start_pos, last_pos,
1431 ordered = btrfs_lookup_ordered_range(inode, start_pos,
1432 last_pos - start_pos + 1);
1434 ordered->file_offset + ordered->num_bytes > start_pos &&
1435 ordered->file_offset <= last_pos) {
1436 unlock_extent_cached(&inode->io_tree, start_pos,
1437 last_pos, cached_state);
1438 for (i = 0; i < num_pages; i++) {
1439 unlock_page(pages[i]);
1442 btrfs_start_ordered_extent(ordered, 1);
1443 btrfs_put_ordered_extent(ordered);
1447 btrfs_put_ordered_extent(ordered);
1449 *lockstart = start_pos;
1450 *lockend = last_pos;
1455 * It's possible the pages are dirty right now, but we don't want
1456 * to clean them yet because copy_from_user may catch a page fault
1457 * and we might have to fall back to one page at a time. If that
1458 * happens, we'll unlock these pages and we'd have a window where
1459 * reclaim could sneak in and drop the once-dirty page on the floor
1460 * without writing it.
1462 * We have the pages locked and the extent range locked, so there's
1463 * no way someone can start IO on any dirty pages in this range.
1465 * We'll call btrfs_dirty_pages() later on, and that will flip around
1466 * delalloc bits and dirty the pages as required.
1468 for (i = 0; i < num_pages; i++) {
1469 set_page_extent_mapped(pages[i]);
1470 WARN_ON(!PageLocked(pages[i]));
1476 static int check_can_nocow(struct btrfs_inode *inode, loff_t pos,
1477 size_t *write_bytes, bool nowait)
1479 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1480 struct btrfs_root *root = inode->root;
1481 u64 lockstart, lockend;
1485 if (!(inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
1488 if (!nowait && !btrfs_drew_try_write_lock(&root->snapshot_lock))
1491 lockstart = round_down(pos, fs_info->sectorsize);
1492 lockend = round_up(pos + *write_bytes,
1493 fs_info->sectorsize) - 1;
1494 num_bytes = lockend - lockstart + 1;
1497 struct btrfs_ordered_extent *ordered;
1499 if (!try_lock_extent(&inode->io_tree, lockstart, lockend))
1502 ordered = btrfs_lookup_ordered_range(inode, lockstart,
1505 btrfs_put_ordered_extent(ordered);
1510 btrfs_lock_and_flush_ordered_range(inode, lockstart,
1514 ret = can_nocow_extent(&inode->vfs_inode, lockstart, &num_bytes,
1515 NULL, NULL, NULL, false);
1519 btrfs_drew_write_unlock(&root->snapshot_lock);
1521 *write_bytes = min_t(size_t, *write_bytes ,
1522 num_bytes - pos + lockstart);
1525 unlock_extent(&inode->io_tree, lockstart, lockend);
1530 static int check_nocow_nolock(struct btrfs_inode *inode, loff_t pos,
1531 size_t *write_bytes)
1533 return check_can_nocow(inode, pos, write_bytes, true);
1537 * Check if we can do nocow write into the range [@pos, @pos + @write_bytes)
1540 * @write_bytes: The length to write, will be updated to the nocow writeable
1543 * This function will flush ordered extents in the range to ensure proper
1547 * >0 and update @write_bytes if we can do nocow write
1548 * 0 if we can't do nocow write
1549 * -EAGAIN if we can't get the needed lock or there are ordered extents
1550 * for * (nowait == true) case
1551 * <0 if other error happened
1553 * NOTE: Callers need to release the lock by btrfs_check_nocow_unlock().
1555 int btrfs_check_nocow_lock(struct btrfs_inode *inode, loff_t pos,
1556 size_t *write_bytes)
1558 return check_can_nocow(inode, pos, write_bytes, false);
1561 void btrfs_check_nocow_unlock(struct btrfs_inode *inode)
1563 btrfs_drew_write_unlock(&inode->root->snapshot_lock);
1566 static void update_time_for_write(struct inode *inode)
1568 struct timespec64 now;
1570 if (IS_NOCMTIME(inode))
1573 now = current_time(inode);
1574 if (!timespec64_equal(&inode->i_mtime, &now))
1575 inode->i_mtime = now;
1577 if (!timespec64_equal(&inode->i_ctime, &now))
1578 inode->i_ctime = now;
1580 if (IS_I_VERSION(inode))
1581 inode_inc_iversion(inode);
1584 static int btrfs_write_check(struct kiocb *iocb, struct iov_iter *from,
1587 struct file *file = iocb->ki_filp;
1588 struct inode *inode = file_inode(file);
1589 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1590 loff_t pos = iocb->ki_pos;
1595 if (iocb->ki_flags & IOCB_NOWAIT) {
1596 size_t nocow_bytes = count;
1598 /* We will allocate space in case nodatacow is not set, so bail */
1599 if (check_nocow_nolock(BTRFS_I(inode), pos, &nocow_bytes) <= 0)
1602 * There are holes in the range or parts of the range that must
1603 * be COWed (shared extents, RO block groups, etc), so just bail
1606 if (nocow_bytes < count)
1610 current->backing_dev_info = inode_to_bdi(inode);
1611 ret = file_remove_privs(file);
1616 * We reserve space for updating the inode when we reserve space for the
1617 * extent we are going to write, so we will enospc out there. We don't
1618 * need to start yet another transaction to update the inode as we will
1619 * update the inode when we finish writing whatever data we write.
1621 update_time_for_write(inode);
1623 start_pos = round_down(pos, fs_info->sectorsize);
1624 oldsize = i_size_read(inode);
1625 if (start_pos > oldsize) {
1626 /* Expand hole size to cover write data, preventing empty gap */
1627 loff_t end_pos = round_up(pos + count, fs_info->sectorsize);
1629 ret = btrfs_cont_expand(inode, oldsize, end_pos);
1631 current->backing_dev_info = NULL;
1639 static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb,
1642 struct file *file = iocb->ki_filp;
1644 struct inode *inode = file_inode(file);
1645 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1646 struct page **pages = NULL;
1647 struct extent_changeset *data_reserved = NULL;
1648 u64 release_bytes = 0;
1651 size_t num_written = 0;
1654 bool only_release_metadata = false;
1655 bool force_page_uptodate = false;
1656 loff_t old_isize = i_size_read(inode);
1657 unsigned int ilock_flags = 0;
1659 if (iocb->ki_flags & IOCB_NOWAIT)
1660 ilock_flags |= BTRFS_ILOCK_TRY;
1662 ret = btrfs_inode_lock(inode, ilock_flags);
1666 ret = generic_write_checks(iocb, i);
1670 ret = btrfs_write_check(iocb, i, ret);
1675 nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_SIZE),
1676 PAGE_SIZE / (sizeof(struct page *)));
1677 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1678 nrptrs = max(nrptrs, 8);
1679 pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL);
1685 while (iov_iter_count(i) > 0) {
1686 struct extent_state *cached_state = NULL;
1687 size_t offset = offset_in_page(pos);
1688 size_t sector_offset;
1689 size_t write_bytes = min(iov_iter_count(i),
1690 nrptrs * (size_t)PAGE_SIZE -
1693 size_t reserve_bytes;
1696 size_t dirty_sectors;
1701 * Fault pages before locking them in prepare_pages
1702 * to avoid recursive lock
1704 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1709 only_release_metadata = false;
1710 sector_offset = pos & (fs_info->sectorsize - 1);
1712 extent_changeset_release(data_reserved);
1713 ret = btrfs_check_data_free_space(BTRFS_I(inode),
1714 &data_reserved, pos,
1718 * If we don't have to COW at the offset, reserve
1719 * metadata only. write_bytes may get smaller than
1722 if (btrfs_check_nocow_lock(BTRFS_I(inode), pos,
1724 only_release_metadata = true;
1729 num_pages = DIV_ROUND_UP(write_bytes + offset, PAGE_SIZE);
1730 WARN_ON(num_pages > nrptrs);
1731 reserve_bytes = round_up(write_bytes + sector_offset,
1732 fs_info->sectorsize);
1733 WARN_ON(reserve_bytes == 0);
1734 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
1737 if (!only_release_metadata)
1738 btrfs_free_reserved_data_space(BTRFS_I(inode),
1742 btrfs_check_nocow_unlock(BTRFS_I(inode));
1746 release_bytes = reserve_bytes;
1749 * This is going to setup the pages array with the number of
1750 * pages we want, so we don't really need to worry about the
1751 * contents of pages from loop to loop
1753 ret = prepare_pages(inode, pages, num_pages,
1755 force_page_uptodate);
1757 btrfs_delalloc_release_extents(BTRFS_I(inode),
1762 extents_locked = lock_and_cleanup_extent_if_need(
1763 BTRFS_I(inode), pages,
1764 num_pages, pos, write_bytes, &lockstart,
1765 &lockend, &cached_state);
1766 if (extents_locked < 0) {
1767 if (extents_locked == -EAGAIN)
1769 btrfs_delalloc_release_extents(BTRFS_I(inode),
1771 ret = extents_locked;
1775 copied = btrfs_copy_from_user(pos, write_bytes, pages, i);
1777 num_sectors = BTRFS_BYTES_TO_BLKS(fs_info, reserve_bytes);
1778 dirty_sectors = round_up(copied + sector_offset,
1779 fs_info->sectorsize);
1780 dirty_sectors = BTRFS_BYTES_TO_BLKS(fs_info, dirty_sectors);
1783 * if we have trouble faulting in the pages, fall
1784 * back to one page at a time
1786 if (copied < write_bytes)
1790 force_page_uptodate = true;
1794 force_page_uptodate = false;
1795 dirty_pages = DIV_ROUND_UP(copied + offset,
1799 if (num_sectors > dirty_sectors) {
1800 /* release everything except the sectors we dirtied */
1801 release_bytes -= dirty_sectors << fs_info->sectorsize_bits;
1802 if (only_release_metadata) {
1803 btrfs_delalloc_release_metadata(BTRFS_I(inode),
1804 release_bytes, true);
1808 __pos = round_down(pos,
1809 fs_info->sectorsize) +
1810 (dirty_pages << PAGE_SHIFT);
1811 btrfs_delalloc_release_space(BTRFS_I(inode),
1812 data_reserved, __pos,
1813 release_bytes, true);
1817 release_bytes = round_up(copied + sector_offset,
1818 fs_info->sectorsize);
1820 ret = btrfs_dirty_pages(BTRFS_I(inode), pages,
1821 dirty_pages, pos, copied,
1822 &cached_state, only_release_metadata);
1825 * If we have not locked the extent range, because the range's
1826 * start offset is >= i_size, we might still have a non-NULL
1827 * cached extent state, acquired while marking the extent range
1828 * as delalloc through btrfs_dirty_pages(). Therefore free any
1829 * possible cached extent state to avoid a memory leak.
1832 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1833 lockstart, lockend, &cached_state);
1835 free_extent_state(cached_state);
1837 btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes);
1839 btrfs_drop_pages(pages, num_pages);
1844 if (only_release_metadata)
1845 btrfs_check_nocow_unlock(BTRFS_I(inode));
1847 btrfs_drop_pages(pages, num_pages);
1851 balance_dirty_pages_ratelimited(inode->i_mapping);
1854 num_written += copied;
1859 if (release_bytes) {
1860 if (only_release_metadata) {
1861 btrfs_check_nocow_unlock(BTRFS_I(inode));
1862 btrfs_delalloc_release_metadata(BTRFS_I(inode),
1863 release_bytes, true);
1865 btrfs_delalloc_release_space(BTRFS_I(inode),
1867 round_down(pos, fs_info->sectorsize),
1868 release_bytes, true);
1872 extent_changeset_free(data_reserved);
1873 if (num_written > 0) {
1874 pagecache_isize_extended(inode, old_isize, iocb->ki_pos);
1875 iocb->ki_pos += num_written;
1878 btrfs_inode_unlock(inode, ilock_flags);
1879 return num_written ? num_written : ret;
1882 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
1883 const struct iov_iter *iter, loff_t offset)
1885 const u32 blocksize_mask = fs_info->sectorsize - 1;
1887 if (offset & blocksize_mask)
1890 if (iov_iter_alignment(iter) & blocksize_mask)
1896 static ssize_t btrfs_direct_write(struct kiocb *iocb, struct iov_iter *from)
1898 struct file *file = iocb->ki_filp;
1899 struct inode *inode = file_inode(file);
1900 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1902 ssize_t written = 0;
1903 ssize_t written_buffered;
1906 unsigned int ilock_flags = 0;
1907 struct iomap_dio *dio = NULL;
1909 if (iocb->ki_flags & IOCB_NOWAIT)
1910 ilock_flags |= BTRFS_ILOCK_TRY;
1912 /* If the write DIO is within EOF, use a shared lock */
1913 if (iocb->ki_pos + iov_iter_count(from) <= i_size_read(inode))
1914 ilock_flags |= BTRFS_ILOCK_SHARED;
1917 err = btrfs_inode_lock(inode, ilock_flags);
1921 err = generic_write_checks(iocb, from);
1923 btrfs_inode_unlock(inode, ilock_flags);
1927 err = btrfs_write_check(iocb, from, err);
1929 btrfs_inode_unlock(inode, ilock_flags);
1935 * Re-check since file size may have changed just before taking the
1936 * lock or pos may have changed because of O_APPEND in generic_write_check()
1938 if ((ilock_flags & BTRFS_ILOCK_SHARED) &&
1939 pos + iov_iter_count(from) > i_size_read(inode)) {
1940 btrfs_inode_unlock(inode, ilock_flags);
1941 ilock_flags &= ~BTRFS_ILOCK_SHARED;
1945 if (check_direct_IO(fs_info, from, pos)) {
1946 btrfs_inode_unlock(inode, ilock_flags);
1950 dio = __iomap_dio_rw(iocb, from, &btrfs_dio_iomap_ops,
1951 &btrfs_dio_ops, is_sync_kiocb(iocb));
1953 btrfs_inode_unlock(inode, ilock_flags);
1955 if (IS_ERR_OR_NULL(dio)) {
1956 err = PTR_ERR_OR_ZERO(dio);
1957 if (err < 0 && err != -ENOTBLK)
1960 written = iomap_dio_complete(dio);
1963 if (written < 0 || !iov_iter_count(from)) {
1970 written_buffered = btrfs_buffered_write(iocb, from);
1971 if (written_buffered < 0) {
1972 err = written_buffered;
1976 * Ensure all data is persisted. We want the next direct IO read to be
1977 * able to read what was just written.
1979 endbyte = pos + written_buffered - 1;
1980 err = btrfs_fdatawrite_range(inode, pos, endbyte);
1983 err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte);
1986 written += written_buffered;
1987 iocb->ki_pos = pos + written_buffered;
1988 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_SHIFT,
1989 endbyte >> PAGE_SHIFT);
1991 return written ? written : err;
1994 static ssize_t btrfs_file_write_iter(struct kiocb *iocb,
1995 struct iov_iter *from)
1997 struct file *file = iocb->ki_filp;
1998 struct inode *inode = file_inode(file);
1999 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2000 struct btrfs_root *root = BTRFS_I(inode)->root;
2001 ssize_t num_written = 0;
2002 const bool sync = iocb->ki_flags & IOCB_DSYNC;
2005 * If the fs flips readonly due to some impossible error, although we
2006 * have opened a file as writable, we have to stop this write operation
2007 * to ensure consistency.
2009 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
2012 if (!(iocb->ki_flags & IOCB_DIRECT) &&
2013 (iocb->ki_flags & IOCB_NOWAIT))
2017 atomic_inc(&BTRFS_I(inode)->sync_writers);
2019 if (iocb->ki_flags & IOCB_DIRECT)
2020 num_written = btrfs_direct_write(iocb, from);
2022 num_written = btrfs_buffered_write(iocb, from);
2025 * We also have to set last_sub_trans to the current log transid,
2026 * otherwise subsequent syncs to a file that's been synced in this
2027 * transaction will appear to have already occurred.
2029 spin_lock(&BTRFS_I(inode)->lock);
2030 BTRFS_I(inode)->last_sub_trans = root->log_transid;
2031 spin_unlock(&BTRFS_I(inode)->lock);
2032 if (num_written > 0)
2033 num_written = generic_write_sync(iocb, num_written);
2036 atomic_dec(&BTRFS_I(inode)->sync_writers);
2038 current->backing_dev_info = NULL;
2042 int btrfs_release_file(struct inode *inode, struct file *filp)
2044 struct btrfs_file_private *private = filp->private_data;
2046 if (private && private->filldir_buf)
2047 kfree(private->filldir_buf);
2049 filp->private_data = NULL;
2052 * Set by setattr when we are about to truncate a file from a non-zero
2053 * size to a zero size. This tries to flush down new bytes that may
2054 * have been written if the application were using truncate to replace
2057 if (test_and_clear_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
2058 &BTRFS_I(inode)->runtime_flags))
2059 filemap_flush(inode->i_mapping);
2063 static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end)
2066 struct blk_plug plug;
2069 * This is only called in fsync, which would do synchronous writes, so
2070 * a plug can merge adjacent IOs as much as possible. Esp. in case of
2071 * multiple disks using raid profile, a large IO can be split to
2072 * several segments of stripe length (currently 64K).
2074 blk_start_plug(&plug);
2075 atomic_inc(&BTRFS_I(inode)->sync_writers);
2076 ret = btrfs_fdatawrite_range(inode, start, end);
2077 atomic_dec(&BTRFS_I(inode)->sync_writers);
2078 blk_finish_plug(&plug);
2084 * fsync call for both files and directories. This logs the inode into
2085 * the tree log instead of forcing full commits whenever possible.
2087 * It needs to call filemap_fdatawait so that all ordered extent updates are
2088 * in the metadata btree are up to date for copying to the log.
2090 * It drops the inode mutex before doing the tree log commit. This is an
2091 * important optimization for directories because holding the mutex prevents
2092 * new operations on the dir while we write to disk.
2094 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
2096 struct dentry *dentry = file_dentry(file);
2097 struct inode *inode = d_inode(dentry);
2098 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
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);
2134 atomic_inc(&root->log_batch);
2137 * Always check for the full sync flag while holding the inode's lock,
2138 * to avoid races with other tasks. The flag must be either set all the
2139 * time during logging or always off all the time while logging.
2141 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2142 &BTRFS_I(inode)->runtime_flags);
2145 * Before we acquired the inode's lock, someone may have dirtied more
2146 * pages in the target range. We need to make sure that writeback for
2147 * any such pages does not start while we are logging the inode, because
2148 * if it does, any of the following might happen when we are not doing a
2151 * 1) We log an extent after its writeback finishes but before its
2152 * checksums are added to the csum tree, leading to -EIO errors
2153 * when attempting to read the extent after a log replay.
2155 * 2) We can end up logging an extent before its writeback finishes.
2156 * Therefore after the log replay we will have a file extent item
2157 * pointing to an unwritten extent (and no data checksums as well).
2159 * So trigger writeback for any eventual new dirty pages and then we
2160 * wait for all ordered extents to complete below.
2162 ret = start_ordered_ops(inode, start, end);
2164 inode_unlock(inode);
2169 * We have to do this here to avoid the priority inversion of waiting on
2170 * IO of a lower priority task while holding a transaction open.
2172 * For a full fsync we wait for the ordered extents to complete while
2173 * for a fast fsync we wait just for writeback to complete, and then
2174 * attach the ordered extents to the transaction so that a transaction
2175 * commit waits for their completion, to avoid data loss if we fsync,
2176 * the current transaction commits before the ordered extents complete
2177 * and a power failure happens right after that.
2180 ret = btrfs_wait_ordered_range(inode, start, len);
2183 * Get our ordered extents as soon as possible to avoid doing
2184 * checksum lookups in the csum tree, and use instead the
2185 * checksums attached to the ordered extents.
2187 btrfs_get_ordered_extents_for_logging(BTRFS_I(inode),
2188 &ctx.ordered_extents);
2189 ret = filemap_fdatawait_range(inode->i_mapping, start, end);
2193 goto out_release_extents;
2195 atomic_inc(&root->log_batch);
2198 * If we are doing a fast fsync we can not bail out if the inode's
2199 * last_trans is <= then the last committed transaction, because we only
2200 * update the last_trans of the inode during ordered extent completion,
2201 * and for a fast fsync we don't wait for that, we only wait for the
2202 * writeback to complete.
2205 if (btrfs_inode_in_log(BTRFS_I(inode), fs_info->generation) ||
2206 (BTRFS_I(inode)->last_trans <= fs_info->last_trans_committed &&
2207 (full_sync || list_empty(&ctx.ordered_extents)))) {
2209 * We've had everything committed since the last time we were
2210 * modified so clear this flag in case it was set for whatever
2211 * reason, it's no longer relevant.
2213 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2214 &BTRFS_I(inode)->runtime_flags);
2216 * An ordered extent might have started before and completed
2217 * already with io errors, in which case the inode was not
2218 * updated and we end up here. So check the inode's mapping
2219 * for any errors that might have happened since we last
2220 * checked called fsync.
2222 ret = filemap_check_wb_err(inode->i_mapping, file->f_wb_err);
2223 goto out_release_extents;
2227 * We use start here because we will need to wait on the IO to complete
2228 * in btrfs_sync_log, which could require joining a transaction (for
2229 * example checking cross references in the nocow path). If we use join
2230 * here we could get into a situation where we're waiting on IO to
2231 * happen that is blocked on a transaction trying to commit. With start
2232 * we inc the extwriter counter, so we wait for all extwriters to exit
2233 * before we start blocking joiners. This comment is to keep somebody
2234 * from thinking they are super smart and changing this to
2235 * btrfs_join_transaction *cough*Josef*cough*.
2237 trans = btrfs_start_transaction(root, 0);
2238 if (IS_ERR(trans)) {
2239 ret = PTR_ERR(trans);
2240 goto out_release_extents;
2243 ret = btrfs_log_dentry_safe(trans, dentry, &ctx);
2244 btrfs_release_log_ctx_extents(&ctx);
2246 /* Fallthrough and commit/free transaction. */
2250 /* we've logged all the items and now have a consistent
2251 * version of the file in the log. It is possible that
2252 * someone will come in and modify the file, but that's
2253 * fine because the log is consistent on disk, and we
2254 * have references to all of the file's extents
2256 * It is possible that someone will come in and log the
2257 * file again, but that will end up using the synchronization
2258 * inside btrfs_sync_log to keep things safe.
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 inode_unlock(inode);
2294 static const struct vm_operations_struct btrfs_file_vm_ops = {
2295 .fault = filemap_fault,
2296 .map_pages = filemap_map_pages,
2297 .page_mkwrite = btrfs_page_mkwrite,
2300 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
2302 struct address_space *mapping = filp->f_mapping;
2304 if (!mapping->a_ops->readpage)
2307 file_accessed(filp);
2308 vma->vm_ops = &btrfs_file_vm_ops;
2313 static int hole_mergeable(struct btrfs_inode *inode, struct extent_buffer *leaf,
2314 int slot, u64 start, u64 end)
2316 struct btrfs_file_extent_item *fi;
2317 struct btrfs_key key;
2319 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2322 btrfs_item_key_to_cpu(leaf, &key, slot);
2323 if (key.objectid != btrfs_ino(inode) ||
2324 key.type != BTRFS_EXTENT_DATA_KEY)
2327 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2329 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2332 if (btrfs_file_extent_disk_bytenr(leaf, fi))
2335 if (key.offset == end)
2337 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2342 static int fill_holes(struct btrfs_trans_handle *trans,
2343 struct btrfs_inode *inode,
2344 struct btrfs_path *path, u64 offset, u64 end)
2346 struct btrfs_fs_info *fs_info = trans->fs_info;
2347 struct btrfs_root *root = inode->root;
2348 struct extent_buffer *leaf;
2349 struct btrfs_file_extent_item *fi;
2350 struct extent_map *hole_em;
2351 struct extent_map_tree *em_tree = &inode->extent_tree;
2352 struct btrfs_key key;
2355 if (btrfs_fs_incompat(fs_info, NO_HOLES))
2358 key.objectid = btrfs_ino(inode);
2359 key.type = BTRFS_EXTENT_DATA_KEY;
2360 key.offset = offset;
2362 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2365 * We should have dropped this offset, so if we find it then
2366 * something has gone horribly wrong.
2373 leaf = path->nodes[0];
2374 if (hole_mergeable(inode, leaf, path->slots[0] - 1, offset, end)) {
2378 fi = btrfs_item_ptr(leaf, path->slots[0],
2379 struct btrfs_file_extent_item);
2380 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2382 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2383 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2384 btrfs_set_file_extent_offset(leaf, fi, 0);
2385 btrfs_mark_buffer_dirty(leaf);
2389 if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) {
2392 key.offset = offset;
2393 btrfs_set_item_key_safe(fs_info, path, &key);
2394 fi = btrfs_item_ptr(leaf, path->slots[0],
2395 struct btrfs_file_extent_item);
2396 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2398 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2399 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2400 btrfs_set_file_extent_offset(leaf, fi, 0);
2401 btrfs_mark_buffer_dirty(leaf);
2404 btrfs_release_path(path);
2406 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode),
2407 offset, 0, 0, end - offset, 0, end - offset, 0, 0, 0);
2412 btrfs_release_path(path);
2414 hole_em = alloc_extent_map();
2416 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2417 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
2419 hole_em->start = offset;
2420 hole_em->len = end - offset;
2421 hole_em->ram_bytes = hole_em->len;
2422 hole_em->orig_start = offset;
2424 hole_em->block_start = EXTENT_MAP_HOLE;
2425 hole_em->block_len = 0;
2426 hole_em->orig_block_len = 0;
2427 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2428 hole_em->generation = trans->transid;
2431 btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2432 write_lock(&em_tree->lock);
2433 ret = add_extent_mapping(em_tree, hole_em, 1);
2434 write_unlock(&em_tree->lock);
2435 } while (ret == -EEXIST);
2436 free_extent_map(hole_em);
2438 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2439 &inode->runtime_flags);
2446 * Find a hole extent on given inode and change start/len to the end of hole
2447 * extent.(hole/vacuum extent whose em->start <= start &&
2448 * em->start + em->len > start)
2449 * When a hole extent is found, return 1 and modify start/len.
2451 static int find_first_non_hole(struct inode *inode, u64 *start, u64 *len)
2453 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2454 struct extent_map *em;
2457 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0,
2458 round_down(*start, fs_info->sectorsize),
2459 round_up(*len, fs_info->sectorsize));
2463 /* Hole or vacuum extent(only exists in no-hole mode) */
2464 if (em->block_start == EXTENT_MAP_HOLE) {
2466 *len = em->start + em->len > *start + *len ?
2467 0 : *start + *len - em->start - em->len;
2468 *start = em->start + em->len;
2470 free_extent_map(em);
2474 static int btrfs_punch_hole_lock_range(struct inode *inode,
2475 const u64 lockstart,
2477 struct extent_state **cached_state)
2480 struct btrfs_ordered_extent *ordered;
2483 truncate_pagecache_range(inode, lockstart, lockend);
2485 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2487 ordered = btrfs_lookup_first_ordered_extent(BTRFS_I(inode),
2491 * We need to make sure we have no ordered extents in this range
2492 * and nobody raced in and read a page in this range, if we did
2493 * we need to try again.
2496 (ordered->file_offset + ordered->num_bytes <= lockstart ||
2497 ordered->file_offset > lockend)) &&
2498 !filemap_range_has_page(inode->i_mapping,
2499 lockstart, lockend)) {
2501 btrfs_put_ordered_extent(ordered);
2505 btrfs_put_ordered_extent(ordered);
2506 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2507 lockend, cached_state);
2508 ret = btrfs_wait_ordered_range(inode, lockstart,
2509 lockend - lockstart + 1);
2516 static int btrfs_insert_replace_extent(struct btrfs_trans_handle *trans,
2517 struct inode *inode,
2518 struct btrfs_path *path,
2519 struct btrfs_replace_extent_info *extent_info,
2520 const u64 replace_len)
2522 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2523 struct btrfs_root *root = BTRFS_I(inode)->root;
2524 struct btrfs_file_extent_item *extent;
2525 struct extent_buffer *leaf;
2526 struct btrfs_key key;
2528 struct btrfs_ref ref = { 0 };
2531 if (replace_len == 0)
2534 if (extent_info->disk_offset == 0 &&
2535 btrfs_fs_incompat(fs_info, NO_HOLES))
2538 key.objectid = btrfs_ino(BTRFS_I(inode));
2539 key.type = BTRFS_EXTENT_DATA_KEY;
2540 key.offset = extent_info->file_offset;
2541 ret = btrfs_insert_empty_item(trans, root, path, &key,
2542 sizeof(struct btrfs_file_extent_item));
2545 leaf = path->nodes[0];
2546 slot = path->slots[0];
2547 write_extent_buffer(leaf, extent_info->extent_buf,
2548 btrfs_item_ptr_offset(leaf, slot),
2549 sizeof(struct btrfs_file_extent_item));
2550 extent = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2551 ASSERT(btrfs_file_extent_type(leaf, extent) != BTRFS_FILE_EXTENT_INLINE);
2552 btrfs_set_file_extent_offset(leaf, extent, extent_info->data_offset);
2553 btrfs_set_file_extent_num_bytes(leaf, extent, replace_len);
2554 if (extent_info->is_new_extent)
2555 btrfs_set_file_extent_generation(leaf, extent, trans->transid);
2556 btrfs_mark_buffer_dirty(leaf);
2557 btrfs_release_path(path);
2559 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode),
2560 extent_info->file_offset, replace_len);
2564 /* If it's a hole, nothing more needs to be done. */
2565 if (extent_info->disk_offset == 0)
2568 inode_add_bytes(inode, replace_len);
2570 if (extent_info->is_new_extent && extent_info->insertions == 0) {
2571 key.objectid = extent_info->disk_offset;
2572 key.type = BTRFS_EXTENT_ITEM_KEY;
2573 key.offset = extent_info->disk_len;
2574 ret = btrfs_alloc_reserved_file_extent(trans, root,
2575 btrfs_ino(BTRFS_I(inode)),
2576 extent_info->file_offset,
2577 extent_info->qgroup_reserved,
2582 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
2583 extent_info->disk_offset,
2584 extent_info->disk_len, 0);
2585 ref_offset = extent_info->file_offset - extent_info->data_offset;
2586 btrfs_init_data_ref(&ref, root->root_key.objectid,
2587 btrfs_ino(BTRFS_I(inode)), ref_offset);
2588 ret = btrfs_inc_extent_ref(trans, &ref);
2591 extent_info->insertions++;
2597 * The respective range must have been previously locked, as well as the inode.
2598 * The end offset is inclusive (last byte of the range).
2599 * @extent_info is NULL for fallocate's hole punching and non-NULL when replacing
2600 * the file range with an extent.
2601 * When not punching a hole, we don't want to end up in a state where we dropped
2602 * extents without inserting a new one, so we must abort the transaction to avoid
2605 int btrfs_replace_file_extents(struct inode *inode, struct btrfs_path *path,
2606 const u64 start, const u64 end,
2607 struct btrfs_replace_extent_info *extent_info,
2608 struct btrfs_trans_handle **trans_out)
2610 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2611 u64 min_size = btrfs_calc_insert_metadata_size(fs_info, 1);
2612 u64 ino_size = round_up(inode->i_size, fs_info->sectorsize);
2613 struct btrfs_root *root = BTRFS_I(inode)->root;
2614 struct btrfs_trans_handle *trans = NULL;
2615 struct btrfs_block_rsv *rsv;
2616 unsigned int rsv_count;
2619 u64 len = end - start;
2625 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
2630 rsv->size = btrfs_calc_insert_metadata_size(fs_info, 1);
2634 * 1 - update the inode
2635 * 1 - removing the extents in the range
2636 * 1 - adding the hole extent if no_holes isn't set or if we are
2637 * replacing the range with a new extent
2639 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || extent_info)
2644 trans = btrfs_start_transaction(root, rsv_count);
2645 if (IS_ERR(trans)) {
2646 ret = PTR_ERR(trans);
2651 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
2654 trans->block_rsv = rsv;
2657 while (cur_offset < end) {
2658 ret = __btrfs_drop_extents(trans, root, BTRFS_I(inode), path,
2659 cur_offset, end + 1, &drop_end,
2661 if (ret != -ENOSPC) {
2663 * When cloning we want to avoid transaction aborts when
2664 * nothing was done and we are attempting to clone parts
2665 * of inline extents, in such cases -EOPNOTSUPP is
2666 * returned by __btrfs_drop_extents() without having
2667 * changed anything in the file.
2669 if (extent_info && !extent_info->is_new_extent &&
2670 ret && ret != -EOPNOTSUPP)
2671 btrfs_abort_transaction(trans, ret);
2675 trans->block_rsv = &fs_info->trans_block_rsv;
2677 if (!extent_info && cur_offset < drop_end &&
2678 cur_offset < ino_size) {
2679 ret = fill_holes(trans, BTRFS_I(inode), path,
2680 cur_offset, drop_end);
2683 * If we failed then we didn't insert our hole
2684 * entries for the area we dropped, so now the
2685 * fs is corrupted, so we must abort the
2688 btrfs_abort_transaction(trans, ret);
2691 } else if (!extent_info && cur_offset < drop_end) {
2693 * We are past the i_size here, but since we didn't
2694 * insert holes we need to clear the mapped area so we
2695 * know to not set disk_i_size in this area until a new
2696 * file extent is inserted here.
2698 ret = btrfs_inode_clear_file_extent_range(BTRFS_I(inode),
2699 cur_offset, drop_end - cur_offset);
2702 * We couldn't clear our area, so we could
2703 * presumably adjust up and corrupt the fs, so
2706 btrfs_abort_transaction(trans, ret);
2711 if (extent_info && drop_end > extent_info->file_offset) {
2712 u64 replace_len = drop_end - extent_info->file_offset;
2714 ret = btrfs_insert_replace_extent(trans, inode, path,
2715 extent_info, replace_len);
2717 btrfs_abort_transaction(trans, ret);
2720 extent_info->data_len -= replace_len;
2721 extent_info->data_offset += replace_len;
2722 extent_info->file_offset += replace_len;
2725 cur_offset = drop_end;
2727 ret = btrfs_update_inode(trans, root, inode);
2731 btrfs_end_transaction(trans);
2732 btrfs_btree_balance_dirty(fs_info);
2734 trans = btrfs_start_transaction(root, rsv_count);
2735 if (IS_ERR(trans)) {
2736 ret = PTR_ERR(trans);
2741 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
2742 rsv, min_size, false);
2743 BUG_ON(ret); /* shouldn't happen */
2744 trans->block_rsv = rsv;
2747 ret = find_first_non_hole(inode, &cur_offset, &len);
2748 if (unlikely(ret < 0))
2758 * If we were cloning, force the next fsync to be a full one since we
2759 * we replaced (or just dropped in the case of cloning holes when
2760 * NO_HOLES is enabled) extents and extent maps.
2761 * This is for the sake of simplicity, and cloning into files larger
2762 * than 16Mb would force the full fsync any way (when
2763 * try_release_extent_mapping() is invoked during page cache truncation.
2765 if (extent_info && !extent_info->is_new_extent)
2766 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2767 &BTRFS_I(inode)->runtime_flags);
2772 trans->block_rsv = &fs_info->trans_block_rsv;
2774 * If we are using the NO_HOLES feature we might have had already an
2775 * hole that overlaps a part of the region [lockstart, lockend] and
2776 * ends at (or beyond) lockend. Since we have no file extent items to
2777 * represent holes, drop_end can be less than lockend and so we must
2778 * make sure we have an extent map representing the existing hole (the
2779 * call to __btrfs_drop_extents() might have dropped the existing extent
2780 * map representing the existing hole), otherwise the fast fsync path
2781 * will not record the existence of the hole region
2782 * [existing_hole_start, lockend].
2784 if (drop_end <= end)
2787 * Don't insert file hole extent item if it's for a range beyond eof
2788 * (because it's useless) or if it represents a 0 bytes range (when
2789 * cur_offset == drop_end).
2791 if (!extent_info && cur_offset < ino_size && cur_offset < drop_end) {
2792 ret = fill_holes(trans, BTRFS_I(inode), path,
2793 cur_offset, drop_end);
2795 /* Same comment as above. */
2796 btrfs_abort_transaction(trans, ret);
2799 } else if (!extent_info && cur_offset < drop_end) {
2800 /* See the comment in the loop above for the reasoning here. */
2801 ret = btrfs_inode_clear_file_extent_range(BTRFS_I(inode),
2802 cur_offset, drop_end - cur_offset);
2804 btrfs_abort_transaction(trans, ret);
2810 ret = btrfs_insert_replace_extent(trans, inode, path, extent_info,
2811 extent_info->data_len);
2813 btrfs_abort_transaction(trans, ret);
2822 trans->block_rsv = &fs_info->trans_block_rsv;
2824 btrfs_end_transaction(trans);
2828 btrfs_free_block_rsv(fs_info, rsv);
2833 static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2835 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2836 struct btrfs_root *root = BTRFS_I(inode)->root;
2837 struct extent_state *cached_state = NULL;
2838 struct btrfs_path *path;
2839 struct btrfs_trans_handle *trans = NULL;
2844 u64 orig_start = offset;
2848 bool truncated_block = false;
2849 bool updated_inode = false;
2851 ret = btrfs_wait_ordered_range(inode, offset, len);
2856 ino_size = round_up(inode->i_size, fs_info->sectorsize);
2857 ret = find_first_non_hole(inode, &offset, &len);
2859 goto out_only_mutex;
2861 /* Already in a large hole */
2863 goto out_only_mutex;
2866 lockstart = round_up(offset, btrfs_inode_sectorsize(BTRFS_I(inode)));
2867 lockend = round_down(offset + len,
2868 btrfs_inode_sectorsize(BTRFS_I(inode))) - 1;
2869 same_block = (BTRFS_BYTES_TO_BLKS(fs_info, offset))
2870 == (BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1));
2872 * We needn't truncate any block which is beyond the end of the file
2873 * because we are sure there is no data there.
2876 * Only do this if we are in the same block and we aren't doing the
2879 if (same_block && len < fs_info->sectorsize) {
2880 if (offset < ino_size) {
2881 truncated_block = true;
2882 ret = btrfs_truncate_block(inode, offset, len, 0);
2886 goto out_only_mutex;
2889 /* zero back part of the first block */
2890 if (offset < ino_size) {
2891 truncated_block = true;
2892 ret = btrfs_truncate_block(inode, offset, 0, 0);
2894 inode_unlock(inode);
2899 /* Check the aligned pages after the first unaligned page,
2900 * if offset != orig_start, which means the first unaligned page
2901 * including several following pages are already in holes,
2902 * the extra check can be skipped */
2903 if (offset == orig_start) {
2904 /* after truncate page, check hole again */
2905 len = offset + len - lockstart;
2907 ret = find_first_non_hole(inode, &offset, &len);
2909 goto out_only_mutex;
2912 goto out_only_mutex;
2917 /* Check the tail unaligned part is in a hole */
2918 tail_start = lockend + 1;
2919 tail_len = offset + len - tail_start;
2921 ret = find_first_non_hole(inode, &tail_start, &tail_len);
2922 if (unlikely(ret < 0))
2923 goto out_only_mutex;
2925 /* zero the front end of the last page */
2926 if (tail_start + tail_len < ino_size) {
2927 truncated_block = true;
2928 ret = btrfs_truncate_block(inode,
2929 tail_start + tail_len,
2932 goto out_only_mutex;
2937 if (lockend < lockstart) {
2939 goto out_only_mutex;
2942 ret = btrfs_punch_hole_lock_range(inode, lockstart, lockend,
2945 goto out_only_mutex;
2947 path = btrfs_alloc_path();
2953 ret = btrfs_replace_file_extents(inode, path, lockstart, lockend, NULL,
2955 btrfs_free_path(path);
2959 ASSERT(trans != NULL);
2960 inode_inc_iversion(inode);
2961 inode->i_mtime = inode->i_ctime = current_time(inode);
2962 ret = btrfs_update_inode(trans, root, inode);
2963 updated_inode = true;
2964 btrfs_end_transaction(trans);
2965 btrfs_btree_balance_dirty(fs_info);
2967 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2970 if (!updated_inode && truncated_block && !ret) {
2972 * If we only end up zeroing part of a page, we still need to
2973 * update the inode item, so that all the time fields are
2974 * updated as well as the necessary btrfs inode in memory fields
2975 * for detecting, at fsync time, if the inode isn't yet in the
2976 * log tree or it's there but not up to date.
2978 struct timespec64 now = current_time(inode);
2980 inode_inc_iversion(inode);
2981 inode->i_mtime = now;
2982 inode->i_ctime = now;
2983 trans = btrfs_start_transaction(root, 1);
2984 if (IS_ERR(trans)) {
2985 ret = PTR_ERR(trans);
2989 ret = btrfs_update_inode(trans, root, inode);
2990 ret2 = btrfs_end_transaction(trans);
2995 inode_unlock(inode);
2999 /* Helper structure to record which range is already reserved */
3000 struct falloc_range {
3001 struct list_head list;
3007 * Helper function to add falloc range
3009 * Caller should have locked the larger range of extent containing
3012 static int add_falloc_range(struct list_head *head, u64 start, u64 len)
3014 struct falloc_range *prev = NULL;
3015 struct falloc_range *range = NULL;
3017 if (list_empty(head))
3021 * As fallocate iterate by bytenr order, we only need to check
3024 prev = list_entry(head->prev, struct falloc_range, list);
3025 if (prev->start + prev->len == start) {
3030 range = kmalloc(sizeof(*range), GFP_KERNEL);
3033 range->start = start;
3035 list_add_tail(&range->list, head);
3039 static int btrfs_fallocate_update_isize(struct inode *inode,
3043 struct btrfs_trans_handle *trans;
3044 struct btrfs_root *root = BTRFS_I(inode)->root;
3048 if (mode & FALLOC_FL_KEEP_SIZE || end <= i_size_read(inode))
3051 trans = btrfs_start_transaction(root, 1);
3053 return PTR_ERR(trans);
3055 inode->i_ctime = current_time(inode);
3056 i_size_write(inode, end);
3057 btrfs_inode_safe_disk_i_size_write(inode, 0);
3058 ret = btrfs_update_inode(trans, root, inode);
3059 ret2 = btrfs_end_transaction(trans);
3061 return ret ? ret : ret2;
3065 RANGE_BOUNDARY_WRITTEN_EXTENT,
3066 RANGE_BOUNDARY_PREALLOC_EXTENT,
3067 RANGE_BOUNDARY_HOLE,
3070 static int btrfs_zero_range_check_range_boundary(struct btrfs_inode *inode,
3073 const u64 sectorsize = btrfs_inode_sectorsize(inode);
3074 struct extent_map *em;
3077 offset = round_down(offset, sectorsize);
3078 em = btrfs_get_extent(inode, NULL, 0, offset, sectorsize);
3082 if (em->block_start == EXTENT_MAP_HOLE)
3083 ret = RANGE_BOUNDARY_HOLE;
3084 else if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
3085 ret = RANGE_BOUNDARY_PREALLOC_EXTENT;
3087 ret = RANGE_BOUNDARY_WRITTEN_EXTENT;
3089 free_extent_map(em);
3093 static int btrfs_zero_range(struct inode *inode,
3098 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3099 struct extent_map *em;
3100 struct extent_changeset *data_reserved = NULL;
3103 const u64 sectorsize = btrfs_inode_sectorsize(BTRFS_I(inode));
3104 u64 alloc_start = round_down(offset, sectorsize);
3105 u64 alloc_end = round_up(offset + len, sectorsize);
3106 u64 bytes_to_reserve = 0;
3107 bool space_reserved = false;
3109 inode_dio_wait(inode);
3111 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start,
3112 alloc_end - alloc_start);
3119 * Avoid hole punching and extent allocation for some cases. More cases
3120 * could be considered, but these are unlikely common and we keep things
3121 * as simple as possible for now. Also, intentionally, if the target
3122 * range contains one or more prealloc extents together with regular
3123 * extents and holes, we drop all the existing extents and allocate a
3124 * new prealloc extent, so that we get a larger contiguous disk extent.
3126 if (em->start <= alloc_start &&
3127 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3128 const u64 em_end = em->start + em->len;
3130 if (em_end >= offset + len) {
3132 * The whole range is already a prealloc extent,
3133 * do nothing except updating the inode's i_size if
3136 free_extent_map(em);
3137 ret = btrfs_fallocate_update_isize(inode, offset + len,
3142 * Part of the range is already a prealloc extent, so operate
3143 * only on the remaining part of the range.
3145 alloc_start = em_end;
3146 ASSERT(IS_ALIGNED(alloc_start, sectorsize));
3147 len = offset + len - alloc_start;
3148 offset = alloc_start;
3149 alloc_hint = em->block_start + em->len;
3151 free_extent_map(em);
3153 if (BTRFS_BYTES_TO_BLKS(fs_info, offset) ==
3154 BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)) {
3155 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start,
3162 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3163 free_extent_map(em);
3164 ret = btrfs_fallocate_update_isize(inode, offset + len,
3168 if (len < sectorsize && em->block_start != EXTENT_MAP_HOLE) {
3169 free_extent_map(em);
3170 ret = btrfs_truncate_block(inode, offset, len, 0);
3172 ret = btrfs_fallocate_update_isize(inode,
3177 free_extent_map(em);
3178 alloc_start = round_down(offset, sectorsize);
3179 alloc_end = alloc_start + sectorsize;
3183 alloc_start = round_up(offset, sectorsize);
3184 alloc_end = round_down(offset + len, sectorsize);
3187 * For unaligned ranges, check the pages at the boundaries, they might
3188 * map to an extent, in which case we need to partially zero them, or
3189 * they might map to a hole, in which case we need our allocation range
3192 if (!IS_ALIGNED(offset, sectorsize)) {
3193 ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode),
3197 if (ret == RANGE_BOUNDARY_HOLE) {
3198 alloc_start = round_down(offset, sectorsize);
3200 } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
3201 ret = btrfs_truncate_block(inode, offset, 0, 0);
3209 if (!IS_ALIGNED(offset + len, sectorsize)) {
3210 ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode),
3214 if (ret == RANGE_BOUNDARY_HOLE) {
3215 alloc_end = round_up(offset + len, sectorsize);
3217 } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
3218 ret = btrfs_truncate_block(inode, offset + len, 0, 1);
3227 if (alloc_start < alloc_end) {
3228 struct extent_state *cached_state = NULL;
3229 const u64 lockstart = alloc_start;
3230 const u64 lockend = alloc_end - 1;
3232 bytes_to_reserve = alloc_end - alloc_start;
3233 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
3237 space_reserved = true;
3238 ret = btrfs_punch_hole_lock_range(inode, lockstart, lockend,
3242 ret = btrfs_qgroup_reserve_data(BTRFS_I(inode), &data_reserved,
3243 alloc_start, bytes_to_reserve);
3246 ret = btrfs_prealloc_file_range(inode, mode, alloc_start,
3247 alloc_end - alloc_start,
3249 offset + len, &alloc_hint);
3250 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
3251 lockend, &cached_state);
3252 /* btrfs_prealloc_file_range releases reserved space on error */
3254 space_reserved = false;
3258 ret = btrfs_fallocate_update_isize(inode, offset + len, mode);
3260 if (ret && space_reserved)
3261 btrfs_free_reserved_data_space(BTRFS_I(inode), data_reserved,
3262 alloc_start, bytes_to_reserve);
3263 extent_changeset_free(data_reserved);
3268 static long btrfs_fallocate(struct file *file, int mode,
3269 loff_t offset, loff_t len)
3271 struct inode *inode = file_inode(file);
3272 struct extent_state *cached_state = NULL;
3273 struct extent_changeset *data_reserved = NULL;
3274 struct falloc_range *range;
3275 struct falloc_range *tmp;
3276 struct list_head reserve_list;
3284 struct extent_map *em;
3285 int blocksize = btrfs_inode_sectorsize(BTRFS_I(inode));
3288 alloc_start = round_down(offset, blocksize);
3289 alloc_end = round_up(offset + len, blocksize);
3290 cur_offset = alloc_start;
3292 /* Make sure we aren't being give some crap mode */
3293 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
3294 FALLOC_FL_ZERO_RANGE))
3297 if (mode & FALLOC_FL_PUNCH_HOLE)
3298 return btrfs_punch_hole(inode, offset, len);
3301 * Only trigger disk allocation, don't trigger qgroup reserve
3303 * For qgroup space, it will be checked later.
3305 if (!(mode & FALLOC_FL_ZERO_RANGE)) {
3306 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
3307 alloc_end - alloc_start);
3312 btrfs_inode_lock(inode, 0);
3314 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) {
3315 ret = inode_newsize_ok(inode, offset + len);
3321 * TODO: Move these two operations after we have checked
3322 * accurate reserved space, or fallocate can still fail but
3323 * with page truncated or size expanded.
3325 * But that's a minor problem and won't do much harm BTW.
3327 if (alloc_start > inode->i_size) {
3328 ret = btrfs_cont_expand(inode, i_size_read(inode),
3332 } else if (offset + len > inode->i_size) {
3334 * If we are fallocating from the end of the file onward we
3335 * need to zero out the end of the block if i_size lands in the
3336 * middle of a block.
3338 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
3344 * wait for ordered IO before we have any locks. We'll loop again
3345 * below with the locks held.
3347 ret = btrfs_wait_ordered_range(inode, alloc_start,
3348 alloc_end - alloc_start);
3352 if (mode & FALLOC_FL_ZERO_RANGE) {
3353 ret = btrfs_zero_range(inode, offset, len, mode);
3354 inode_unlock(inode);
3358 locked_end = alloc_end - 1;
3360 struct btrfs_ordered_extent *ordered;
3362 /* the extent lock is ordered inside the running
3365 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
3366 locked_end, &cached_state);
3367 ordered = btrfs_lookup_first_ordered_extent(BTRFS_I(inode),
3371 ordered->file_offset + ordered->num_bytes > alloc_start &&
3372 ordered->file_offset < alloc_end) {
3373 btrfs_put_ordered_extent(ordered);
3374 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
3375 alloc_start, locked_end,
3378 * we can't wait on the range with the transaction
3379 * running or with the extent lock held
3381 ret = btrfs_wait_ordered_range(inode, alloc_start,
3382 alloc_end - alloc_start);
3387 btrfs_put_ordered_extent(ordered);
3392 /* First, check if we exceed the qgroup limit */
3393 INIT_LIST_HEAD(&reserve_list);
3394 while (cur_offset < alloc_end) {
3395 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
3396 alloc_end - cur_offset);
3401 last_byte = min(extent_map_end(em), alloc_end);
3402 actual_end = min_t(u64, extent_map_end(em), offset + len);
3403 last_byte = ALIGN(last_byte, blocksize);
3404 if (em->block_start == EXTENT_MAP_HOLE ||
3405 (cur_offset >= inode->i_size &&
3406 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
3407 ret = add_falloc_range(&reserve_list, cur_offset,
3408 last_byte - cur_offset);
3410 free_extent_map(em);
3413 ret = btrfs_qgroup_reserve_data(BTRFS_I(inode),
3414 &data_reserved, cur_offset,
3415 last_byte - cur_offset);
3417 cur_offset = last_byte;
3418 free_extent_map(em);
3423 * Do not need to reserve unwritten extent for this
3424 * range, free reserved data space first, otherwise
3425 * it'll result in false ENOSPC error.
3427 btrfs_free_reserved_data_space(BTRFS_I(inode),
3428 data_reserved, cur_offset,
3429 last_byte - cur_offset);
3431 free_extent_map(em);
3432 cur_offset = last_byte;
3436 * If ret is still 0, means we're OK to fallocate.
3437 * Or just cleanup the list and exit.
3439 list_for_each_entry_safe(range, tmp, &reserve_list, list) {
3441 ret = btrfs_prealloc_file_range(inode, mode,
3443 range->len, i_blocksize(inode),
3444 offset + len, &alloc_hint);
3446 btrfs_free_reserved_data_space(BTRFS_I(inode),
3447 data_reserved, range->start,
3449 list_del(&range->list);
3456 * We didn't need to allocate any more space, but we still extended the
3457 * size of the file so we need to update i_size and the inode item.
3459 ret = btrfs_fallocate_update_isize(inode, actual_end, mode);
3461 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
3464 inode_unlock(inode);
3465 /* Let go of our reservation. */
3466 if (ret != 0 && !(mode & FALLOC_FL_ZERO_RANGE))
3467 btrfs_free_reserved_data_space(BTRFS_I(inode), data_reserved,
3468 cur_offset, alloc_end - cur_offset);
3469 extent_changeset_free(data_reserved);
3473 static loff_t find_desired_extent(struct inode *inode, loff_t offset,
3476 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3477 struct extent_map *em = NULL;
3478 struct extent_state *cached_state = NULL;
3479 loff_t i_size = inode->i_size;
3486 if (i_size == 0 || offset >= i_size)
3490 * offset can be negative, in this case we start finding DATA/HOLE from
3491 * the very start of the file.
3493 start = max_t(loff_t, 0, offset);
3495 lockstart = round_down(start, fs_info->sectorsize);
3496 lockend = round_up(i_size, fs_info->sectorsize);
3497 if (lockend <= lockstart)
3498 lockend = lockstart + fs_info->sectorsize;
3500 len = lockend - lockstart + 1;
3502 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
3505 while (start < i_size) {
3506 em = btrfs_get_extent_fiemap(BTRFS_I(inode), start, len);
3513 if (whence == SEEK_HOLE &&
3514 (em->block_start == EXTENT_MAP_HOLE ||
3515 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
3517 else if (whence == SEEK_DATA &&
3518 (em->block_start != EXTENT_MAP_HOLE &&
3519 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
3522 start = em->start + em->len;
3523 free_extent_map(em);
3527 free_extent_map(em);
3528 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
3533 if (whence == SEEK_DATA && start >= i_size)
3536 offset = min_t(loff_t, start, i_size);
3542 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
3544 struct inode *inode = file->f_mapping->host;
3548 return generic_file_llseek(file, offset, whence);
3551 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
3552 offset = find_desired_extent(inode, offset, whence);
3553 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
3560 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
3563 static int btrfs_file_open(struct inode *inode, struct file *filp)
3565 filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
3566 return generic_file_open(inode, filp);
3569 static int check_direct_read(struct btrfs_fs_info *fs_info,
3570 const struct iov_iter *iter, loff_t offset)
3575 ret = check_direct_IO(fs_info, iter, offset);
3579 if (!iter_is_iovec(iter))
3582 for (seg = 0; seg < iter->nr_segs; seg++)
3583 for (i = seg + 1; i < iter->nr_segs; i++)
3584 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
3589 static ssize_t btrfs_direct_read(struct kiocb *iocb, struct iov_iter *to)
3591 struct inode *inode = file_inode(iocb->ki_filp);
3594 if (check_direct_read(btrfs_sb(inode->i_sb), to, iocb->ki_pos))
3597 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
3598 ret = iomap_dio_rw(iocb, to, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
3599 is_sync_kiocb(iocb));
3600 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
3604 static ssize_t btrfs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
3608 if (iocb->ki_flags & IOCB_DIRECT) {
3609 ret = btrfs_direct_read(iocb, to);
3610 if (ret < 0 || !iov_iter_count(to) ||
3611 iocb->ki_pos >= i_size_read(file_inode(iocb->ki_filp)))
3615 return generic_file_buffered_read(iocb, to, ret);
3618 const struct file_operations btrfs_file_operations = {
3619 .llseek = btrfs_file_llseek,
3620 .read_iter = btrfs_file_read_iter,
3621 .splice_read = generic_file_splice_read,
3622 .write_iter = btrfs_file_write_iter,
3623 .splice_write = iter_file_splice_write,
3624 .mmap = btrfs_file_mmap,
3625 .open = btrfs_file_open,
3626 .release = btrfs_release_file,
3627 .fsync = btrfs_sync_file,
3628 .fallocate = btrfs_fallocate,
3629 .unlocked_ioctl = btrfs_ioctl,
3630 #ifdef CONFIG_COMPAT
3631 .compat_ioctl = btrfs_compat_ioctl,
3633 .remap_file_range = btrfs_remap_file_range,
3636 void __cold btrfs_auto_defrag_exit(void)
3638 kmem_cache_destroy(btrfs_inode_defrag_cachep);
3641 int __init btrfs_auto_defrag_init(void)
3643 btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
3644 sizeof(struct inode_defrag), 0,
3647 if (!btrfs_inode_defrag_cachep)
3653 int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end)
3658 * So with compression we will find and lock a dirty page and clear the
3659 * first one as dirty, setup an async extent, and immediately return
3660 * with the entire range locked but with nobody actually marked with
3661 * writeback. So we can't just filemap_write_and_wait_range() and
3662 * expect it to work since it will just kick off a thread to do the
3663 * actual work. So we need to call filemap_fdatawrite_range _again_
3664 * since it will wait on the page lock, which won't be unlocked until
3665 * after the pages have been marked as writeback and so we're good to go
3666 * from there. We have to do this otherwise we'll miss the ordered
3667 * extents and that results in badness. Please Josef, do not think you
3668 * know better and pull this out at some point in the future, it is
3669 * right and you are wrong.
3671 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
3672 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
3673 &BTRFS_I(inode)->runtime_flags))
3674 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);