1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
7 #include <linux/pagemap.h>
8 #include <linux/time.h>
9 #include <linux/init.h>
10 #include <linux/string.h>
11 #include <linux/backing-dev.h>
12 #include <linux/falloc.h>
13 #include <linux/writeback.h>
14 #include <linux/compat.h>
15 #include <linux/slab.h>
16 #include <linux/btrfs.h>
17 #include <linux/uio.h>
18 #include <linux/iversion.h>
19 #include <linux/fsverity.h>
22 #include "transaction.h"
23 #include "btrfs_inode.h"
24 #include "print-tree.h"
29 #include "compression.h"
30 #include "delalloc-space.h"
34 static struct kmem_cache *btrfs_inode_defrag_cachep;
36 * when auto defrag is enabled we
37 * queue up these defrag structs to remember which
38 * inodes need defragging passes
41 struct rb_node rb_node;
45 * transid where the defrag was added, we search for
46 * extents newer than this
54 * The extent size threshold for autodefrag.
56 * This value is different for compressed/non-compressed extents,
57 * thus needs to be passed from higher layer.
58 * (aka, inode_should_defrag())
63 static int __compare_inode_defrag(struct inode_defrag *defrag1,
64 struct inode_defrag *defrag2)
66 if (defrag1->root > defrag2->root)
68 else if (defrag1->root < defrag2->root)
70 else if (defrag1->ino > defrag2->ino)
72 else if (defrag1->ino < defrag2->ino)
78 /* pop a record for an inode into the defrag tree. The lock
79 * must be held already
81 * If you're inserting a record for an older transid than an
82 * existing record, the transid already in the tree is lowered
84 * If an existing record is found the defrag item you
87 static int __btrfs_add_inode_defrag(struct btrfs_inode *inode,
88 struct inode_defrag *defrag)
90 struct btrfs_fs_info *fs_info = inode->root->fs_info;
91 struct inode_defrag *entry;
93 struct rb_node *parent = NULL;
96 p = &fs_info->defrag_inodes.rb_node;
99 entry = rb_entry(parent, struct inode_defrag, rb_node);
101 ret = __compare_inode_defrag(defrag, entry);
103 p = &parent->rb_left;
105 p = &parent->rb_right;
107 /* if we're reinserting an entry for
108 * an old defrag run, make sure to
109 * lower the transid of our existing record
111 if (defrag->transid < entry->transid)
112 entry->transid = defrag->transid;
113 entry->extent_thresh = min(defrag->extent_thresh,
114 entry->extent_thresh);
118 set_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags);
119 rb_link_node(&defrag->rb_node, parent, p);
120 rb_insert_color(&defrag->rb_node, &fs_info->defrag_inodes);
124 static inline int __need_auto_defrag(struct btrfs_fs_info *fs_info)
126 if (!btrfs_test_opt(fs_info, AUTO_DEFRAG))
129 if (btrfs_fs_closing(fs_info))
136 * insert a defrag record for this inode if auto defrag is
139 int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
140 struct btrfs_inode *inode, u32 extent_thresh)
142 struct btrfs_root *root = inode->root;
143 struct btrfs_fs_info *fs_info = root->fs_info;
144 struct inode_defrag *defrag;
148 if (!__need_auto_defrag(fs_info))
151 if (test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags))
155 transid = trans->transid;
157 transid = inode->root->last_trans;
159 defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS);
163 defrag->ino = btrfs_ino(inode);
164 defrag->transid = transid;
165 defrag->root = root->root_key.objectid;
166 defrag->extent_thresh = extent_thresh;
168 spin_lock(&fs_info->defrag_inodes_lock);
169 if (!test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags)) {
171 * If we set IN_DEFRAG flag and evict the inode from memory,
172 * and then re-read this inode, this new inode doesn't have
173 * IN_DEFRAG flag. At the case, we may find the existed defrag.
175 ret = __btrfs_add_inode_defrag(inode, defrag);
177 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
179 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
181 spin_unlock(&fs_info->defrag_inodes_lock);
186 * pick the defragable inode that we want, if it doesn't exist, we will get
189 static struct inode_defrag *
190 btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino)
192 struct inode_defrag *entry = NULL;
193 struct inode_defrag tmp;
195 struct rb_node *parent = NULL;
201 spin_lock(&fs_info->defrag_inodes_lock);
202 p = fs_info->defrag_inodes.rb_node;
205 entry = rb_entry(parent, struct inode_defrag, rb_node);
207 ret = __compare_inode_defrag(&tmp, entry);
211 p = parent->rb_right;
216 if (parent && __compare_inode_defrag(&tmp, entry) > 0) {
217 parent = rb_next(parent);
219 entry = rb_entry(parent, struct inode_defrag, rb_node);
225 rb_erase(parent, &fs_info->defrag_inodes);
226 spin_unlock(&fs_info->defrag_inodes_lock);
230 void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info)
232 struct inode_defrag *defrag;
233 struct rb_node *node;
235 spin_lock(&fs_info->defrag_inodes_lock);
236 node = rb_first(&fs_info->defrag_inodes);
238 rb_erase(node, &fs_info->defrag_inodes);
239 defrag = rb_entry(node, struct inode_defrag, rb_node);
240 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
242 cond_resched_lock(&fs_info->defrag_inodes_lock);
244 node = rb_first(&fs_info->defrag_inodes);
246 spin_unlock(&fs_info->defrag_inodes_lock);
249 #define BTRFS_DEFRAG_BATCH 1024
251 static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info,
252 struct inode_defrag *defrag)
254 struct btrfs_root *inode_root;
256 struct btrfs_ioctl_defrag_range_args range;
261 if (test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state))
263 if (!__need_auto_defrag(fs_info))
267 inode_root = btrfs_get_fs_root(fs_info, defrag->root, true);
268 if (IS_ERR(inode_root)) {
269 ret = PTR_ERR(inode_root);
273 inode = btrfs_iget(fs_info->sb, defrag->ino, inode_root);
274 btrfs_put_root(inode_root);
276 ret = PTR_ERR(inode);
280 if (cur >= i_size_read(inode)) {
285 /* do a chunk of defrag */
286 clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
287 memset(&range, 0, sizeof(range));
290 range.extent_thresh = defrag->extent_thresh;
292 sb_start_write(fs_info->sb);
293 ret = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
295 sb_end_write(fs_info->sb);
301 cur = max(cur + fs_info->sectorsize, range.start);
305 kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
310 * run through the list of inodes in the FS that need
313 int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
315 struct inode_defrag *defrag;
317 u64 root_objectid = 0;
319 atomic_inc(&fs_info->defrag_running);
321 /* Pause the auto defragger. */
322 if (test_bit(BTRFS_FS_STATE_REMOUNTING,
326 if (!__need_auto_defrag(fs_info))
329 /* find an inode to defrag */
330 defrag = btrfs_pick_defrag_inode(fs_info, root_objectid,
333 if (root_objectid || first_ino) {
342 first_ino = defrag->ino + 1;
343 root_objectid = defrag->root;
345 __btrfs_run_defrag_inode(fs_info, defrag);
347 atomic_dec(&fs_info->defrag_running);
350 * during unmount, we use the transaction_wait queue to
351 * wait for the defragger to stop
353 wake_up(&fs_info->transaction_wait);
357 /* simple helper to fault in pages and copy. This should go away
358 * and be replaced with calls into generic code.
360 static noinline int btrfs_copy_from_user(loff_t pos, size_t write_bytes,
361 struct page **prepared_pages,
365 size_t total_copied = 0;
367 int offset = offset_in_page(pos);
369 while (write_bytes > 0) {
370 size_t count = min_t(size_t,
371 PAGE_SIZE - offset, write_bytes);
372 struct page *page = prepared_pages[pg];
374 * Copy data from userspace to the current page
376 copied = copy_page_from_iter_atomic(page, offset, count, i);
378 /* Flush processor's dcache for this page */
379 flush_dcache_page(page);
382 * if we get a partial write, we can end up with
383 * partially up to date pages. These add
384 * a lot of complexity, so make sure they don't
385 * happen by forcing this copy to be retried.
387 * The rest of the btrfs_file_write code will fall
388 * back to page at a time copies after we return 0.
390 if (unlikely(copied < count)) {
391 if (!PageUptodate(page)) {
392 iov_iter_revert(i, copied);
399 write_bytes -= copied;
400 total_copied += copied;
402 if (offset == PAGE_SIZE) {
411 * unlocks pages after btrfs_file_write is done with them
413 static void btrfs_drop_pages(struct btrfs_fs_info *fs_info,
414 struct page **pages, size_t num_pages,
418 u64 block_start = round_down(pos, fs_info->sectorsize);
419 u64 block_len = round_up(pos + copied, fs_info->sectorsize) - block_start;
421 ASSERT(block_len <= U32_MAX);
422 for (i = 0; i < num_pages; i++) {
423 /* page checked is some magic around finding pages that
424 * have been modified without going through btrfs_set_page_dirty
425 * clear it here. There should be no need to mark the pages
426 * accessed as prepare_pages should have marked them accessed
427 * in prepare_pages via find_or_create_page()
429 btrfs_page_clamp_clear_checked(fs_info, pages[i], block_start,
431 unlock_page(pages[i]);
437 * After btrfs_copy_from_user(), update the following things for delalloc:
438 * - Mark newly dirtied pages as DELALLOC in the io tree.
439 * Used to advise which range is to be written back.
440 * - Mark modified pages as Uptodate/Dirty and not needing COW fixup
441 * - Update inode size for past EOF write
443 int btrfs_dirty_pages(struct btrfs_inode *inode, struct page **pages,
444 size_t num_pages, loff_t pos, size_t write_bytes,
445 struct extent_state **cached, bool noreserve)
447 struct btrfs_fs_info *fs_info = inode->root->fs_info;
452 u64 end_of_last_block;
453 u64 end_pos = pos + write_bytes;
454 loff_t isize = i_size_read(&inode->vfs_inode);
455 unsigned int extra_bits = 0;
457 if (write_bytes == 0)
461 extra_bits |= EXTENT_NORESERVE;
463 start_pos = round_down(pos, fs_info->sectorsize);
464 num_bytes = round_up(write_bytes + pos - start_pos,
465 fs_info->sectorsize);
466 ASSERT(num_bytes <= U32_MAX);
468 end_of_last_block = start_pos + num_bytes - 1;
471 * The pages may have already been dirty, clear out old accounting so
472 * we can set things up properly
474 clear_extent_bit(&inode->io_tree, start_pos, end_of_last_block,
475 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
478 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
483 for (i = 0; i < num_pages; i++) {
484 struct page *p = pages[i];
486 btrfs_page_clamp_set_uptodate(fs_info, p, start_pos, num_bytes);
487 btrfs_page_clamp_clear_checked(fs_info, p, start_pos, num_bytes);
488 btrfs_page_clamp_set_dirty(fs_info, p, start_pos, num_bytes);
492 * we've only changed i_size in ram, and we haven't updated
493 * the disk i_size. There is no need to log the inode
497 i_size_write(&inode->vfs_inode, end_pos);
502 * this is very complex, but the basic idea is to drop all extents
503 * in the range start - end. hint_block is filled in with a block number
504 * that would be a good hint to the block allocator for this file.
506 * If an extent intersects the range but is not entirely inside the range
507 * it is either truncated or split. Anything entirely inside the range
508 * is deleted from the tree.
510 * Note: the VFS' inode number of bytes is not updated, it's up to the caller
511 * to deal with that. We set the field 'bytes_found' of the arguments structure
512 * with the number of allocated bytes found in the target range, so that the
513 * caller can update the inode's number of bytes in an atomic way when
514 * replacing extents in a range to avoid races with stat(2).
516 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
517 struct btrfs_root *root, struct btrfs_inode *inode,
518 struct btrfs_drop_extents_args *args)
520 struct btrfs_fs_info *fs_info = root->fs_info;
521 struct extent_buffer *leaf;
522 struct btrfs_file_extent_item *fi;
523 struct btrfs_ref ref = { 0 };
524 struct btrfs_key key;
525 struct btrfs_key new_key;
526 u64 ino = btrfs_ino(inode);
527 u64 search_start = args->start;
530 u64 extent_offset = 0;
532 u64 last_end = args->start;
538 int modify_tree = -1;
541 struct btrfs_path *path = args->path;
543 args->bytes_found = 0;
544 args->extent_inserted = false;
546 /* Must always have a path if ->replace_extent is true */
547 ASSERT(!(args->replace_extent && !args->path));
550 path = btrfs_alloc_path();
557 if (args->drop_cache)
558 btrfs_drop_extent_map_range(inode, args->start, args->end - 1, false);
560 if (args->start >= inode->disk_i_size && !args->replace_extent)
563 update_refs = (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID);
566 ret = btrfs_lookup_file_extent(trans, root, path, ino,
567 search_start, modify_tree);
570 if (ret > 0 && path->slots[0] > 0 && search_start == args->start) {
571 leaf = path->nodes[0];
572 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
573 if (key.objectid == ino &&
574 key.type == BTRFS_EXTENT_DATA_KEY)
579 leaf = path->nodes[0];
580 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
582 ret = btrfs_next_leaf(root, path);
589 leaf = path->nodes[0];
593 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
595 if (key.objectid > ino)
597 if (WARN_ON_ONCE(key.objectid < ino) ||
598 key.type < BTRFS_EXTENT_DATA_KEY) {
603 if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= args->end)
606 fi = btrfs_item_ptr(leaf, path->slots[0],
607 struct btrfs_file_extent_item);
608 extent_type = btrfs_file_extent_type(leaf, fi);
610 if (extent_type == BTRFS_FILE_EXTENT_REG ||
611 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
612 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
613 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
614 extent_offset = btrfs_file_extent_offset(leaf, fi);
615 extent_end = key.offset +
616 btrfs_file_extent_num_bytes(leaf, fi);
617 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
618 extent_end = key.offset +
619 btrfs_file_extent_ram_bytes(leaf, fi);
626 * Don't skip extent items representing 0 byte lengths. They
627 * used to be created (bug) if while punching holes we hit
628 * -ENOSPC condition. So if we find one here, just ensure we
629 * delete it, otherwise we would insert a new file extent item
630 * with the same key (offset) as that 0 bytes length file
631 * extent item in the call to setup_items_for_insert() later
634 if (extent_end == key.offset && extent_end >= search_start) {
635 last_end = extent_end;
636 goto delete_extent_item;
639 if (extent_end <= search_start) {
645 search_start = max(key.offset, args->start);
646 if (recow || !modify_tree) {
648 btrfs_release_path(path);
653 * | - range to drop - |
654 * | -------- extent -------- |
656 if (args->start > key.offset && args->end < extent_end) {
658 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
663 memcpy(&new_key, &key, sizeof(new_key));
664 new_key.offset = args->start;
665 ret = btrfs_duplicate_item(trans, root, path,
667 if (ret == -EAGAIN) {
668 btrfs_release_path(path);
674 leaf = path->nodes[0];
675 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
676 struct btrfs_file_extent_item);
677 btrfs_set_file_extent_num_bytes(leaf, fi,
678 args->start - key.offset);
680 fi = btrfs_item_ptr(leaf, path->slots[0],
681 struct btrfs_file_extent_item);
683 extent_offset += args->start - key.offset;
684 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
685 btrfs_set_file_extent_num_bytes(leaf, fi,
686 extent_end - args->start);
687 btrfs_mark_buffer_dirty(leaf);
689 if (update_refs && disk_bytenr > 0) {
690 btrfs_init_generic_ref(&ref,
691 BTRFS_ADD_DELAYED_REF,
692 disk_bytenr, num_bytes, 0);
693 btrfs_init_data_ref(&ref,
694 root->root_key.objectid,
696 args->start - extent_offset,
698 ret = btrfs_inc_extent_ref(trans, &ref);
700 btrfs_abort_transaction(trans, ret);
704 key.offset = args->start;
707 * From here on out we will have actually dropped something, so
708 * last_end can be updated.
710 last_end = extent_end;
713 * | ---- range to drop ----- |
714 * | -------- extent -------- |
716 if (args->start <= key.offset && args->end < extent_end) {
717 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
722 memcpy(&new_key, &key, sizeof(new_key));
723 new_key.offset = args->end;
724 btrfs_set_item_key_safe(fs_info, path, &new_key);
726 extent_offset += args->end - key.offset;
727 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
728 btrfs_set_file_extent_num_bytes(leaf, fi,
729 extent_end - args->end);
730 btrfs_mark_buffer_dirty(leaf);
731 if (update_refs && disk_bytenr > 0)
732 args->bytes_found += args->end - key.offset;
736 search_start = extent_end;
738 * | ---- range to drop ----- |
739 * | -------- extent -------- |
741 if (args->start > key.offset && args->end >= extent_end) {
743 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
748 btrfs_set_file_extent_num_bytes(leaf, fi,
749 args->start - key.offset);
750 btrfs_mark_buffer_dirty(leaf);
751 if (update_refs && disk_bytenr > 0)
752 args->bytes_found += extent_end - args->start;
753 if (args->end == extent_end)
761 * | ---- range to drop ----- |
762 * | ------ extent ------ |
764 if (args->start <= key.offset && args->end >= extent_end) {
767 del_slot = path->slots[0];
770 BUG_ON(del_slot + del_nr != path->slots[0]);
775 extent_type == BTRFS_FILE_EXTENT_INLINE) {
776 args->bytes_found += extent_end - key.offset;
777 extent_end = ALIGN(extent_end,
778 fs_info->sectorsize);
779 } else if (update_refs && disk_bytenr > 0) {
780 btrfs_init_generic_ref(&ref,
781 BTRFS_DROP_DELAYED_REF,
782 disk_bytenr, num_bytes, 0);
783 btrfs_init_data_ref(&ref,
784 root->root_key.objectid,
786 key.offset - extent_offset, 0,
788 ret = btrfs_free_extent(trans, &ref);
790 btrfs_abort_transaction(trans, ret);
793 args->bytes_found += extent_end - key.offset;
796 if (args->end == extent_end)
799 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
804 ret = btrfs_del_items(trans, root, path, del_slot,
807 btrfs_abort_transaction(trans, ret);
814 btrfs_release_path(path);
821 if (!ret && del_nr > 0) {
823 * Set path->slots[0] to first slot, so that after the delete
824 * if items are move off from our leaf to its immediate left or
825 * right neighbor leafs, we end up with a correct and adjusted
826 * path->slots[0] for our insertion (if args->replace_extent).
828 path->slots[0] = del_slot;
829 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
831 btrfs_abort_transaction(trans, ret);
834 leaf = path->nodes[0];
836 * If btrfs_del_items() was called, it might have deleted a leaf, in
837 * which case it unlocked our path, so check path->locks[0] matches a
840 if (!ret && args->replace_extent &&
841 path->locks[0] == BTRFS_WRITE_LOCK &&
842 btrfs_leaf_free_space(leaf) >=
843 sizeof(struct btrfs_item) + args->extent_item_size) {
846 key.type = BTRFS_EXTENT_DATA_KEY;
847 key.offset = args->start;
848 if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
849 struct btrfs_key slot_key;
851 btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
852 if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
855 btrfs_setup_item_for_insert(root, path, &key, args->extent_item_size);
856 args->extent_inserted = true;
860 btrfs_free_path(path);
861 else if (!args->extent_inserted)
862 btrfs_release_path(path);
864 args->drop_end = found ? min(args->end, last_end) : args->end;
869 static int extent_mergeable(struct extent_buffer *leaf, int slot,
870 u64 objectid, u64 bytenr, u64 orig_offset,
871 u64 *start, u64 *end)
873 struct btrfs_file_extent_item *fi;
874 struct btrfs_key key;
877 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
880 btrfs_item_key_to_cpu(leaf, &key, slot);
881 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
884 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
885 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
886 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
887 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
888 btrfs_file_extent_compression(leaf, fi) ||
889 btrfs_file_extent_encryption(leaf, fi) ||
890 btrfs_file_extent_other_encoding(leaf, fi))
893 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
894 if ((*start && *start != key.offset) || (*end && *end != extent_end))
903 * Mark extent in the range start - end as written.
905 * This changes extent type from 'pre-allocated' to 'regular'. If only
906 * part of extent is marked as written, the extent will be split into
909 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
910 struct btrfs_inode *inode, u64 start, u64 end)
912 struct btrfs_fs_info *fs_info = trans->fs_info;
913 struct btrfs_root *root = inode->root;
914 struct extent_buffer *leaf;
915 struct btrfs_path *path;
916 struct btrfs_file_extent_item *fi;
917 struct btrfs_ref ref = { 0 };
918 struct btrfs_key key;
919 struct btrfs_key new_key;
931 u64 ino = btrfs_ino(inode);
933 path = btrfs_alloc_path();
940 key.type = BTRFS_EXTENT_DATA_KEY;
943 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
946 if (ret > 0 && path->slots[0] > 0)
949 leaf = path->nodes[0];
950 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
951 if (key.objectid != ino ||
952 key.type != BTRFS_EXTENT_DATA_KEY) {
954 btrfs_abort_transaction(trans, ret);
957 fi = btrfs_item_ptr(leaf, path->slots[0],
958 struct btrfs_file_extent_item);
959 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_PREALLOC) {
961 btrfs_abort_transaction(trans, ret);
964 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
965 if (key.offset > start || extent_end < end) {
967 btrfs_abort_transaction(trans, ret);
971 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
972 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
973 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
974 memcpy(&new_key, &key, sizeof(new_key));
976 if (start == key.offset && end < extent_end) {
979 if (extent_mergeable(leaf, path->slots[0] - 1,
980 ino, bytenr, orig_offset,
981 &other_start, &other_end)) {
982 new_key.offset = end;
983 btrfs_set_item_key_safe(fs_info, path, &new_key);
984 fi = btrfs_item_ptr(leaf, path->slots[0],
985 struct btrfs_file_extent_item);
986 btrfs_set_file_extent_generation(leaf, fi,
988 btrfs_set_file_extent_num_bytes(leaf, fi,
990 btrfs_set_file_extent_offset(leaf, fi,
992 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
993 struct btrfs_file_extent_item);
994 btrfs_set_file_extent_generation(leaf, fi,
996 btrfs_set_file_extent_num_bytes(leaf, fi,
998 btrfs_mark_buffer_dirty(leaf);
1003 if (start > key.offset && end == extent_end) {
1006 if (extent_mergeable(leaf, path->slots[0] + 1,
1007 ino, bytenr, orig_offset,
1008 &other_start, &other_end)) {
1009 fi = btrfs_item_ptr(leaf, path->slots[0],
1010 struct btrfs_file_extent_item);
1011 btrfs_set_file_extent_num_bytes(leaf, fi,
1012 start - key.offset);
1013 btrfs_set_file_extent_generation(leaf, fi,
1016 new_key.offset = start;
1017 btrfs_set_item_key_safe(fs_info, path, &new_key);
1019 fi = btrfs_item_ptr(leaf, path->slots[0],
1020 struct btrfs_file_extent_item);
1021 btrfs_set_file_extent_generation(leaf, fi,
1023 btrfs_set_file_extent_num_bytes(leaf, fi,
1025 btrfs_set_file_extent_offset(leaf, fi,
1026 start - orig_offset);
1027 btrfs_mark_buffer_dirty(leaf);
1032 while (start > key.offset || end < extent_end) {
1033 if (key.offset == start)
1036 new_key.offset = split;
1037 ret = btrfs_duplicate_item(trans, root, path, &new_key);
1038 if (ret == -EAGAIN) {
1039 btrfs_release_path(path);
1043 btrfs_abort_transaction(trans, ret);
1047 leaf = path->nodes[0];
1048 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1049 struct btrfs_file_extent_item);
1050 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1051 btrfs_set_file_extent_num_bytes(leaf, fi,
1052 split - key.offset);
1054 fi = btrfs_item_ptr(leaf, path->slots[0],
1055 struct btrfs_file_extent_item);
1057 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1058 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1059 btrfs_set_file_extent_num_bytes(leaf, fi,
1060 extent_end - split);
1061 btrfs_mark_buffer_dirty(leaf);
1063 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, bytenr,
1065 btrfs_init_data_ref(&ref, root->root_key.objectid, ino,
1066 orig_offset, 0, false);
1067 ret = btrfs_inc_extent_ref(trans, &ref);
1069 btrfs_abort_transaction(trans, ret);
1073 if (split == start) {
1076 if (start != key.offset) {
1078 btrfs_abort_transaction(trans, ret);
1089 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
1091 btrfs_init_data_ref(&ref, root->root_key.objectid, ino, orig_offset,
1093 if (extent_mergeable(leaf, path->slots[0] + 1,
1094 ino, bytenr, orig_offset,
1095 &other_start, &other_end)) {
1097 btrfs_release_path(path);
1100 extent_end = other_end;
1101 del_slot = path->slots[0] + 1;
1103 ret = btrfs_free_extent(trans, &ref);
1105 btrfs_abort_transaction(trans, ret);
1111 if (extent_mergeable(leaf, path->slots[0] - 1,
1112 ino, bytenr, orig_offset,
1113 &other_start, &other_end)) {
1115 btrfs_release_path(path);
1118 key.offset = other_start;
1119 del_slot = path->slots[0];
1121 ret = btrfs_free_extent(trans, &ref);
1123 btrfs_abort_transaction(trans, ret);
1128 fi = btrfs_item_ptr(leaf, path->slots[0],
1129 struct btrfs_file_extent_item);
1130 btrfs_set_file_extent_type(leaf, fi,
1131 BTRFS_FILE_EXTENT_REG);
1132 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1133 btrfs_mark_buffer_dirty(leaf);
1135 fi = btrfs_item_ptr(leaf, del_slot - 1,
1136 struct btrfs_file_extent_item);
1137 btrfs_set_file_extent_type(leaf, fi,
1138 BTRFS_FILE_EXTENT_REG);
1139 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1140 btrfs_set_file_extent_num_bytes(leaf, fi,
1141 extent_end - key.offset);
1142 btrfs_mark_buffer_dirty(leaf);
1144 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1146 btrfs_abort_transaction(trans, ret);
1151 btrfs_free_path(path);
1156 * on error we return an unlocked page and the error value
1157 * on success we return a locked page and 0
1159 static int prepare_uptodate_page(struct inode *inode,
1160 struct page *page, u64 pos,
1161 bool force_uptodate)
1163 struct folio *folio = page_folio(page);
1166 if (((pos & (PAGE_SIZE - 1)) || force_uptodate) &&
1167 !PageUptodate(page)) {
1168 ret = btrfs_read_folio(NULL, folio);
1172 if (!PageUptodate(page)) {
1178 * Since btrfs_read_folio() will unlock the folio before it
1179 * returns, there is a window where btrfs_release_folio() can be
1180 * called to release the page. Here we check both inode
1181 * mapping and PagePrivate() to make sure the page was not
1184 * The private flag check is essential for subpage as we need
1185 * to store extra bitmap using page->private.
1187 if (page->mapping != inode->i_mapping || !PagePrivate(page)) {
1195 static unsigned int get_prepare_fgp_flags(bool nowait)
1197 unsigned int fgp_flags = FGP_LOCK | FGP_ACCESSED | FGP_CREAT;
1200 fgp_flags |= FGP_NOWAIT;
1205 static gfp_t get_prepare_gfp_flags(struct inode *inode, bool nowait)
1209 gfp = btrfs_alloc_write_mask(inode->i_mapping);
1211 gfp &= ~__GFP_DIRECT_RECLAIM;
1219 * this just gets pages into the page cache and locks them down.
1221 static noinline int prepare_pages(struct inode *inode, struct page **pages,
1222 size_t num_pages, loff_t pos,
1223 size_t write_bytes, bool force_uptodate,
1227 unsigned long index = pos >> PAGE_SHIFT;
1228 gfp_t mask = get_prepare_gfp_flags(inode, nowait);
1229 unsigned int fgp_flags = get_prepare_fgp_flags(nowait);
1233 for (i = 0; i < num_pages; i++) {
1235 pages[i] = pagecache_get_page(inode->i_mapping, index + i,
1236 fgp_flags, mask | __GFP_WRITE);
1246 err = set_page_extent_mapped(pages[i]);
1253 err = prepare_uptodate_page(inode, pages[i], pos,
1255 if (!err && i == num_pages - 1)
1256 err = prepare_uptodate_page(inode, pages[i],
1257 pos + write_bytes, false);
1260 if (!nowait && err == -EAGAIN) {
1267 wait_on_page_writeback(pages[i]);
1272 while (faili >= 0) {
1273 unlock_page(pages[faili]);
1274 put_page(pages[faili]);
1282 * This function locks the extent and properly waits for data=ordered extents
1283 * to finish before allowing the pages to be modified if need.
1286 * 1 - the extent is locked
1287 * 0 - the extent is not locked, and everything is OK
1288 * -EAGAIN - need re-prepare the pages
1289 * the other < 0 number - Something wrong happens
1292 lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct page **pages,
1293 size_t num_pages, loff_t pos,
1295 u64 *lockstart, u64 *lockend, bool nowait,
1296 struct extent_state **cached_state)
1298 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1304 start_pos = round_down(pos, fs_info->sectorsize);
1305 last_pos = round_up(pos + write_bytes, fs_info->sectorsize) - 1;
1307 if (start_pos < inode->vfs_inode.i_size) {
1308 struct btrfs_ordered_extent *ordered;
1311 if (!try_lock_extent(&inode->io_tree, start_pos, last_pos)) {
1312 for (i = 0; i < num_pages; i++) {
1313 unlock_page(pages[i]);
1321 lock_extent(&inode->io_tree, start_pos, last_pos, cached_state);
1324 ordered = btrfs_lookup_ordered_range(inode, start_pos,
1325 last_pos - start_pos + 1);
1327 ordered->file_offset + ordered->num_bytes > start_pos &&
1328 ordered->file_offset <= last_pos) {
1329 unlock_extent(&inode->io_tree, start_pos, last_pos,
1331 for (i = 0; i < num_pages; i++) {
1332 unlock_page(pages[i]);
1335 btrfs_start_ordered_extent(ordered, 1);
1336 btrfs_put_ordered_extent(ordered);
1340 btrfs_put_ordered_extent(ordered);
1342 *lockstart = start_pos;
1343 *lockend = last_pos;
1348 * We should be called after prepare_pages() which should have locked
1349 * all pages in the range.
1351 for (i = 0; i < num_pages; i++)
1352 WARN_ON(!PageLocked(pages[i]));
1358 * Check if we can do nocow write into the range [@pos, @pos + @write_bytes)
1360 * @pos: File offset.
1361 * @write_bytes: The length to write, will be updated to the nocow writeable
1364 * This function will flush ordered extents in the range to ensure proper
1368 * > 0 If we can nocow, and updates @write_bytes.
1369 * 0 If we can't do a nocow write.
1370 * -EAGAIN If we can't do a nocow write because snapshoting of the inode's
1371 * root is in progress.
1372 * < 0 If an error happened.
1374 * NOTE: Callers need to call btrfs_check_nocow_unlock() if we return > 0.
1376 int btrfs_check_nocow_lock(struct btrfs_inode *inode, loff_t pos,
1377 size_t *write_bytes, bool nowait)
1379 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1380 struct btrfs_root *root = inode->root;
1381 u64 lockstart, lockend;
1385 if (!(inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
1388 if (!btrfs_drew_try_write_lock(&root->snapshot_lock))
1391 lockstart = round_down(pos, fs_info->sectorsize);
1392 lockend = round_up(pos + *write_bytes,
1393 fs_info->sectorsize) - 1;
1394 num_bytes = lockend - lockstart + 1;
1397 if (!btrfs_try_lock_ordered_range(inode, lockstart, lockend)) {
1398 btrfs_drew_write_unlock(&root->snapshot_lock);
1402 btrfs_lock_and_flush_ordered_range(inode, lockstart, lockend, NULL);
1404 ret = can_nocow_extent(&inode->vfs_inode, lockstart, &num_bytes,
1405 NULL, NULL, NULL, nowait, false);
1407 btrfs_drew_write_unlock(&root->snapshot_lock);
1409 *write_bytes = min_t(size_t, *write_bytes ,
1410 num_bytes - pos + lockstart);
1411 unlock_extent(&inode->io_tree, lockstart, lockend, NULL);
1416 void btrfs_check_nocow_unlock(struct btrfs_inode *inode)
1418 btrfs_drew_write_unlock(&inode->root->snapshot_lock);
1421 static void update_time_for_write(struct inode *inode)
1423 struct timespec64 now;
1425 if (IS_NOCMTIME(inode))
1428 now = current_time(inode);
1429 if (!timespec64_equal(&inode->i_mtime, &now))
1430 inode->i_mtime = now;
1432 if (!timespec64_equal(&inode->i_ctime, &now))
1433 inode->i_ctime = now;
1435 if (IS_I_VERSION(inode))
1436 inode_inc_iversion(inode);
1439 static int btrfs_write_check(struct kiocb *iocb, struct iov_iter *from,
1442 struct file *file = iocb->ki_filp;
1443 struct inode *inode = file_inode(file);
1444 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1445 loff_t pos = iocb->ki_pos;
1451 * Quickly bail out on NOWAIT writes if we don't have the nodatacow or
1452 * prealloc flags, as without those flags we always have to COW. We will
1453 * later check if we can really COW into the target range (using
1454 * can_nocow_extent() at btrfs_get_blocks_direct_write()).
1456 if ((iocb->ki_flags & IOCB_NOWAIT) &&
1457 !(BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
1460 current->backing_dev_info = inode_to_bdi(inode);
1461 ret = file_remove_privs(file);
1466 * We reserve space for updating the inode when we reserve space for the
1467 * extent we are going to write, so we will enospc out there. We don't
1468 * need to start yet another transaction to update the inode as we will
1469 * update the inode when we finish writing whatever data we write.
1471 update_time_for_write(inode);
1473 start_pos = round_down(pos, fs_info->sectorsize);
1474 oldsize = i_size_read(inode);
1475 if (start_pos > oldsize) {
1476 /* Expand hole size to cover write data, preventing empty gap */
1477 loff_t end_pos = round_up(pos + count, fs_info->sectorsize);
1479 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, end_pos);
1481 current->backing_dev_info = NULL;
1489 static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb,
1492 struct file *file = iocb->ki_filp;
1494 struct inode *inode = file_inode(file);
1495 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1496 struct page **pages = NULL;
1497 struct extent_changeset *data_reserved = NULL;
1498 u64 release_bytes = 0;
1501 size_t num_written = 0;
1504 bool only_release_metadata = false;
1505 bool force_page_uptodate = false;
1506 loff_t old_isize = i_size_read(inode);
1507 unsigned int ilock_flags = 0;
1508 const bool nowait = (iocb->ki_flags & IOCB_NOWAIT);
1509 unsigned int bdp_flags = (nowait ? BDP_ASYNC : 0);
1512 ilock_flags |= BTRFS_ILOCK_TRY;
1514 ret = btrfs_inode_lock(inode, ilock_flags);
1518 ret = generic_write_checks(iocb, i);
1522 ret = btrfs_write_check(iocb, i, ret);
1527 nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_SIZE),
1528 PAGE_SIZE / (sizeof(struct page *)));
1529 nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1530 nrptrs = max(nrptrs, 8);
1531 pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL);
1537 while (iov_iter_count(i) > 0) {
1538 struct extent_state *cached_state = NULL;
1539 size_t offset = offset_in_page(pos);
1540 size_t sector_offset;
1541 size_t write_bytes = min(iov_iter_count(i),
1542 nrptrs * (size_t)PAGE_SIZE -
1545 size_t reserve_bytes;
1548 size_t dirty_sectors;
1553 * Fault pages before locking them in prepare_pages
1554 * to avoid recursive lock
1556 if (unlikely(fault_in_iov_iter_readable(i, write_bytes))) {
1561 only_release_metadata = false;
1562 sector_offset = pos & (fs_info->sectorsize - 1);
1564 extent_changeset_release(data_reserved);
1565 ret = btrfs_check_data_free_space(BTRFS_I(inode),
1566 &data_reserved, pos,
1567 write_bytes, nowait);
1571 if (nowait && (ret == -ENOSPC || ret == -EAGAIN)) {
1577 * If we don't have to COW at the offset, reserve
1578 * metadata only. write_bytes may get smaller than
1581 can_nocow = btrfs_check_nocow_lock(BTRFS_I(inode), pos,
1582 &write_bytes, nowait);
1589 only_release_metadata = true;
1592 num_pages = DIV_ROUND_UP(write_bytes + offset, PAGE_SIZE);
1593 WARN_ON(num_pages > nrptrs);
1594 reserve_bytes = round_up(write_bytes + sector_offset,
1595 fs_info->sectorsize);
1596 WARN_ON(reserve_bytes == 0);
1597 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
1599 reserve_bytes, nowait);
1601 if (!only_release_metadata)
1602 btrfs_free_reserved_data_space(BTRFS_I(inode),
1606 btrfs_check_nocow_unlock(BTRFS_I(inode));
1608 if (nowait && ret == -ENOSPC)
1613 release_bytes = reserve_bytes;
1615 ret = balance_dirty_pages_ratelimited_flags(inode->i_mapping, bdp_flags);
1617 btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes);
1622 * This is going to setup the pages array with the number of
1623 * pages we want, so we don't really need to worry about the
1624 * contents of pages from loop to loop
1626 ret = prepare_pages(inode, pages, num_pages,
1627 pos, write_bytes, force_page_uptodate, false);
1629 btrfs_delalloc_release_extents(BTRFS_I(inode),
1634 extents_locked = lock_and_cleanup_extent_if_need(
1635 BTRFS_I(inode), pages,
1636 num_pages, pos, write_bytes, &lockstart,
1637 &lockend, nowait, &cached_state);
1638 if (extents_locked < 0) {
1639 if (!nowait && extents_locked == -EAGAIN)
1642 btrfs_delalloc_release_extents(BTRFS_I(inode),
1644 ret = extents_locked;
1648 copied = btrfs_copy_from_user(pos, write_bytes, pages, i);
1650 num_sectors = BTRFS_BYTES_TO_BLKS(fs_info, reserve_bytes);
1651 dirty_sectors = round_up(copied + sector_offset,
1652 fs_info->sectorsize);
1653 dirty_sectors = BTRFS_BYTES_TO_BLKS(fs_info, dirty_sectors);
1656 * if we have trouble faulting in the pages, fall
1657 * back to one page at a time
1659 if (copied < write_bytes)
1663 force_page_uptodate = true;
1667 force_page_uptodate = false;
1668 dirty_pages = DIV_ROUND_UP(copied + offset,
1672 if (num_sectors > dirty_sectors) {
1673 /* release everything except the sectors we dirtied */
1674 release_bytes -= dirty_sectors << fs_info->sectorsize_bits;
1675 if (only_release_metadata) {
1676 btrfs_delalloc_release_metadata(BTRFS_I(inode),
1677 release_bytes, true);
1681 __pos = round_down(pos,
1682 fs_info->sectorsize) +
1683 (dirty_pages << PAGE_SHIFT);
1684 btrfs_delalloc_release_space(BTRFS_I(inode),
1685 data_reserved, __pos,
1686 release_bytes, true);
1690 release_bytes = round_up(copied + sector_offset,
1691 fs_info->sectorsize);
1693 ret = btrfs_dirty_pages(BTRFS_I(inode), pages,
1694 dirty_pages, pos, copied,
1695 &cached_state, only_release_metadata);
1698 * If we have not locked the extent range, because the range's
1699 * start offset is >= i_size, we might still have a non-NULL
1700 * cached extent state, acquired while marking the extent range
1701 * as delalloc through btrfs_dirty_pages(). Therefore free any
1702 * possible cached extent state to avoid a memory leak.
1705 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart,
1706 lockend, &cached_state);
1708 free_extent_state(cached_state);
1710 btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes);
1712 btrfs_drop_pages(fs_info, pages, num_pages, pos, copied);
1717 if (only_release_metadata)
1718 btrfs_check_nocow_unlock(BTRFS_I(inode));
1720 btrfs_drop_pages(fs_info, pages, num_pages, pos, copied);
1725 num_written += copied;
1730 if (release_bytes) {
1731 if (only_release_metadata) {
1732 btrfs_check_nocow_unlock(BTRFS_I(inode));
1733 btrfs_delalloc_release_metadata(BTRFS_I(inode),
1734 release_bytes, true);
1736 btrfs_delalloc_release_space(BTRFS_I(inode),
1738 round_down(pos, fs_info->sectorsize),
1739 release_bytes, true);
1743 extent_changeset_free(data_reserved);
1744 if (num_written > 0) {
1745 pagecache_isize_extended(inode, old_isize, iocb->ki_pos);
1746 iocb->ki_pos += num_written;
1749 btrfs_inode_unlock(inode, ilock_flags);
1750 return num_written ? num_written : ret;
1753 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
1754 const struct iov_iter *iter, loff_t offset)
1756 const u32 blocksize_mask = fs_info->sectorsize - 1;
1758 if (offset & blocksize_mask)
1761 if (iov_iter_alignment(iter) & blocksize_mask)
1767 static ssize_t btrfs_direct_write(struct kiocb *iocb, struct iov_iter *from)
1769 struct file *file = iocb->ki_filp;
1770 struct inode *inode = file_inode(file);
1771 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1773 ssize_t written = 0;
1774 ssize_t written_buffered;
1775 size_t prev_left = 0;
1778 unsigned int ilock_flags = 0;
1779 struct iomap_dio *dio;
1781 if (iocb->ki_flags & IOCB_NOWAIT)
1782 ilock_flags |= BTRFS_ILOCK_TRY;
1784 /* If the write DIO is within EOF, use a shared lock */
1785 if (iocb->ki_pos + iov_iter_count(from) <= i_size_read(inode))
1786 ilock_flags |= BTRFS_ILOCK_SHARED;
1789 err = btrfs_inode_lock(inode, ilock_flags);
1793 err = generic_write_checks(iocb, from);
1795 btrfs_inode_unlock(inode, ilock_flags);
1799 err = btrfs_write_check(iocb, from, err);
1801 btrfs_inode_unlock(inode, ilock_flags);
1807 * Re-check since file size may have changed just before taking the
1808 * lock or pos may have changed because of O_APPEND in generic_write_check()
1810 if ((ilock_flags & BTRFS_ILOCK_SHARED) &&
1811 pos + iov_iter_count(from) > i_size_read(inode)) {
1812 btrfs_inode_unlock(inode, ilock_flags);
1813 ilock_flags &= ~BTRFS_ILOCK_SHARED;
1817 if (check_direct_IO(fs_info, from, pos)) {
1818 btrfs_inode_unlock(inode, ilock_flags);
1823 * The iov_iter can be mapped to the same file range we are writing to.
1824 * If that's the case, then we will deadlock in the iomap code, because
1825 * it first calls our callback btrfs_dio_iomap_begin(), which will create
1826 * an ordered extent, and after that it will fault in the pages that the
1827 * iov_iter refers to. During the fault in we end up in the readahead
1828 * pages code (starting at btrfs_readahead()), which will lock the range,
1829 * find that ordered extent and then wait for it to complete (at
1830 * btrfs_lock_and_flush_ordered_range()), resulting in a deadlock since
1831 * obviously the ordered extent can never complete as we didn't submit
1832 * yet the respective bio(s). This always happens when the buffer is
1833 * memory mapped to the same file range, since the iomap DIO code always
1834 * invalidates pages in the target file range (after starting and waiting
1835 * for any writeback).
1837 * So here we disable page faults in the iov_iter and then retry if we
1838 * got -EFAULT, faulting in the pages before the retry.
1840 from->nofault = true;
1841 dio = btrfs_dio_write(iocb, from, written);
1842 from->nofault = false;
1845 * iomap_dio_complete() will call btrfs_sync_file() if we have a dsync
1846 * iocb, and that needs to lock the inode. So unlock it before calling
1847 * iomap_dio_complete() to avoid a deadlock.
1849 btrfs_inode_unlock(inode, ilock_flags);
1851 if (IS_ERR_OR_NULL(dio))
1852 err = PTR_ERR_OR_ZERO(dio);
1854 err = iomap_dio_complete(dio);
1856 /* No increment (+=) because iomap returns a cumulative value. */
1860 if (iov_iter_count(from) > 0 && (err == -EFAULT || err > 0)) {
1861 const size_t left = iov_iter_count(from);
1863 * We have more data left to write. Try to fault in as many as
1864 * possible of the remainder pages and retry. We do this without
1865 * releasing and locking again the inode, to prevent races with
1868 * Also, in case the iov refers to pages in the file range of the
1869 * file we want to write to (due to a mmap), we could enter an
1870 * infinite loop if we retry after faulting the pages in, since
1871 * iomap will invalidate any pages in the range early on, before
1872 * it tries to fault in the pages of the iov. So we keep track of
1873 * how much was left of iov in the previous EFAULT and fallback
1874 * to buffered IO in case we haven't made any progress.
1876 if (left == prev_left) {
1879 fault_in_iov_iter_readable(from, left);
1886 * If 'err' is -ENOTBLK or we have not written all data, then it means
1887 * we must fallback to buffered IO.
1889 if ((err < 0 && err != -ENOTBLK) || !iov_iter_count(from))
1894 * If we are in a NOWAIT context, then return -EAGAIN to signal the caller
1895 * it must retry the operation in a context where blocking is acceptable,
1896 * since we currently don't have NOWAIT semantics support for buffered IO
1897 * and may block there for many reasons (reserving space for example).
1899 if (iocb->ki_flags & IOCB_NOWAIT) {
1905 written_buffered = btrfs_buffered_write(iocb, from);
1906 if (written_buffered < 0) {
1907 err = written_buffered;
1911 * Ensure all data is persisted. We want the next direct IO read to be
1912 * able to read what was just written.
1914 endbyte = pos + written_buffered - 1;
1915 err = btrfs_fdatawrite_range(inode, pos, endbyte);
1918 err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte);
1921 written += written_buffered;
1922 iocb->ki_pos = pos + written_buffered;
1923 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_SHIFT,
1924 endbyte >> PAGE_SHIFT);
1926 return err < 0 ? err : written;
1929 static ssize_t btrfs_encoded_write(struct kiocb *iocb, struct iov_iter *from,
1930 const struct btrfs_ioctl_encoded_io_args *encoded)
1932 struct file *file = iocb->ki_filp;
1933 struct inode *inode = file_inode(file);
1937 btrfs_inode_lock(inode, 0);
1938 count = encoded->len;
1939 ret = generic_write_checks_count(iocb, &count);
1940 if (ret == 0 && count != encoded->len) {
1942 * The write got truncated by generic_write_checks_count(). We
1943 * can't do a partial encoded write.
1947 if (ret || encoded->len == 0)
1950 ret = btrfs_write_check(iocb, from, encoded->len);
1954 ret = btrfs_do_encoded_write(iocb, from, encoded);
1956 btrfs_inode_unlock(inode, 0);
1960 ssize_t btrfs_do_write_iter(struct kiocb *iocb, struct iov_iter *from,
1961 const struct btrfs_ioctl_encoded_io_args *encoded)
1963 struct file *file = iocb->ki_filp;
1964 struct btrfs_inode *inode = BTRFS_I(file_inode(file));
1965 ssize_t num_written, num_sync;
1966 const bool sync = iocb_is_dsync(iocb);
1969 * If the fs flips readonly due to some impossible error, although we
1970 * have opened a file as writable, we have to stop this write operation
1971 * to ensure consistency.
1973 if (BTRFS_FS_ERROR(inode->root->fs_info))
1976 if (encoded && (iocb->ki_flags & IOCB_NOWAIT))
1980 atomic_inc(&inode->sync_writers);
1983 num_written = btrfs_encoded_write(iocb, from, encoded);
1984 num_sync = encoded->len;
1985 } else if (iocb->ki_flags & IOCB_DIRECT) {
1986 num_written = btrfs_direct_write(iocb, from);
1987 num_sync = num_written;
1989 num_written = btrfs_buffered_write(iocb, from);
1990 num_sync = num_written;
1993 btrfs_set_inode_last_sub_trans(inode);
1996 num_sync = generic_write_sync(iocb, num_sync);
1998 num_written = num_sync;
2002 atomic_dec(&inode->sync_writers);
2004 current->backing_dev_info = NULL;
2008 static ssize_t btrfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2010 return btrfs_do_write_iter(iocb, from, NULL);
2013 int btrfs_release_file(struct inode *inode, struct file *filp)
2015 struct btrfs_file_private *private = filp->private_data;
2017 if (private && private->filldir_buf)
2018 kfree(private->filldir_buf);
2020 filp->private_data = NULL;
2023 * Set by setattr when we are about to truncate a file from a non-zero
2024 * size to a zero size. This tries to flush down new bytes that may
2025 * have been written if the application were using truncate to replace
2028 if (test_and_clear_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
2029 &BTRFS_I(inode)->runtime_flags))
2030 filemap_flush(inode->i_mapping);
2034 static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end)
2037 struct blk_plug plug;
2040 * This is only called in fsync, which would do synchronous writes, so
2041 * a plug can merge adjacent IOs as much as possible. Esp. in case of
2042 * multiple disks using raid profile, a large IO can be split to
2043 * several segments of stripe length (currently 64K).
2045 blk_start_plug(&plug);
2046 atomic_inc(&BTRFS_I(inode)->sync_writers);
2047 ret = btrfs_fdatawrite_range(inode, start, end);
2048 atomic_dec(&BTRFS_I(inode)->sync_writers);
2049 blk_finish_plug(&plug);
2054 static inline bool skip_inode_logging(const struct btrfs_log_ctx *ctx)
2056 struct btrfs_inode *inode = BTRFS_I(ctx->inode);
2057 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2059 if (btrfs_inode_in_log(inode, fs_info->generation) &&
2060 list_empty(&ctx->ordered_extents))
2064 * If we are doing a fast fsync we can not bail out if the inode's
2065 * last_trans is <= then the last committed transaction, because we only
2066 * update the last_trans of the inode during ordered extent completion,
2067 * and for a fast fsync we don't wait for that, we only wait for the
2068 * writeback to complete.
2070 if (inode->last_trans <= fs_info->last_trans_committed &&
2071 (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) ||
2072 list_empty(&ctx->ordered_extents)))
2079 * fsync call for both files and directories. This logs the inode into
2080 * the tree log instead of forcing full commits whenever possible.
2082 * It needs to call filemap_fdatawait so that all ordered extent updates are
2083 * in the metadata btree are up to date for copying to the log.
2085 * It drops the inode mutex before doing the tree log commit. This is an
2086 * important optimization for directories because holding the mutex prevents
2087 * new operations on the dir while we write to disk.
2089 int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
2091 struct dentry *dentry = file_dentry(file);
2092 struct inode *inode = d_inode(dentry);
2093 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2094 struct btrfs_root *root = BTRFS_I(inode)->root;
2095 struct btrfs_trans_handle *trans;
2096 struct btrfs_log_ctx ctx;
2101 trace_btrfs_sync_file(file, datasync);
2103 btrfs_init_log_ctx(&ctx, inode);
2106 * Always set the range to a full range, otherwise we can get into
2107 * several problems, from missing file extent items to represent holes
2108 * when not using the NO_HOLES feature, to log tree corruption due to
2109 * races between hole detection during logging and completion of ordered
2110 * extents outside the range, to missing checksums due to ordered extents
2111 * for which we flushed only a subset of their pages.
2115 len = (u64)LLONG_MAX + 1;
2118 * We write the dirty pages in the range and wait until they complete
2119 * out of the ->i_mutex. If so, we can flush the dirty pages by
2120 * multi-task, and make the performance up. See
2121 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
2123 ret = start_ordered_ops(inode, start, end);
2127 btrfs_inode_lock(inode, BTRFS_ILOCK_MMAP);
2129 atomic_inc(&root->log_batch);
2132 * Before we acquired the inode's lock and the mmap lock, someone may
2133 * have dirtied more pages in the target range. We need to make sure
2134 * that writeback for any such pages does not start while we are logging
2135 * the inode, because if it does, any of the following might happen when
2136 * we are not doing a full inode sync:
2138 * 1) We log an extent after its writeback finishes but before its
2139 * checksums are added to the csum tree, leading to -EIO errors
2140 * when attempting to read the extent after a log replay.
2142 * 2) We can end up logging an extent before its writeback finishes.
2143 * Therefore after the log replay we will have a file extent item
2144 * pointing to an unwritten extent (and no data checksums as well).
2146 * So trigger writeback for any eventual new dirty pages and then we
2147 * wait for all ordered extents to complete below.
2149 ret = start_ordered_ops(inode, start, end);
2151 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
2156 * Always check for the full sync flag while holding the inode's lock,
2157 * to avoid races with other tasks. The flag must be either set all the
2158 * time during logging or always off all the time while logging.
2159 * We check the flag here after starting delalloc above, because when
2160 * running delalloc the full sync flag may be set if we need to drop
2161 * extra extent map ranges due to temporary memory allocation failures.
2163 full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2164 &BTRFS_I(inode)->runtime_flags);
2167 * We have to do this here to avoid the priority inversion of waiting on
2168 * IO of a lower priority task while holding a transaction open.
2170 * For a full fsync we wait for the ordered extents to complete while
2171 * for a fast fsync we wait just for writeback to complete, and then
2172 * attach the ordered extents to the transaction so that a transaction
2173 * commit waits for their completion, to avoid data loss if we fsync,
2174 * the current transaction commits before the ordered extents complete
2175 * and a power failure happens right after that.
2177 * For zoned filesystem, if a write IO uses a ZONE_APPEND command, the
2178 * logical address recorded in the ordered extent may change. We need
2179 * to wait for the IO to stabilize the logical address.
2181 if (full_sync || btrfs_is_zoned(fs_info)) {
2182 ret = btrfs_wait_ordered_range(inode, start, len);
2185 * Get our ordered extents as soon as possible to avoid doing
2186 * checksum lookups in the csum tree, and use instead the
2187 * checksums attached to the ordered extents.
2189 btrfs_get_ordered_extents_for_logging(BTRFS_I(inode),
2190 &ctx.ordered_extents);
2191 ret = filemap_fdatawait_range(inode->i_mapping, start, end);
2195 goto out_release_extents;
2197 atomic_inc(&root->log_batch);
2200 if (skip_inode_logging(&ctx)) {
2202 * We've had everything committed since the last time we were
2203 * modified so clear this flag in case it was set for whatever
2204 * reason, it's no longer relevant.
2206 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2207 &BTRFS_I(inode)->runtime_flags);
2209 * An ordered extent might have started before and completed
2210 * already with io errors, in which case the inode was not
2211 * updated and we end up here. So check the inode's mapping
2212 * for any errors that might have happened since we last
2213 * checked called fsync.
2215 ret = filemap_check_wb_err(inode->i_mapping, file->f_wb_err);
2216 goto out_release_extents;
2220 * We use start here because we will need to wait on the IO to complete
2221 * in btrfs_sync_log, which could require joining a transaction (for
2222 * example checking cross references in the nocow path). If we use join
2223 * here we could get into a situation where we're waiting on IO to
2224 * happen that is blocked on a transaction trying to commit. With start
2225 * we inc the extwriter counter, so we wait for all extwriters to exit
2226 * before we start blocking joiners. This comment is to keep somebody
2227 * from thinking they are super smart and changing this to
2228 * btrfs_join_transaction *cough*Josef*cough*.
2230 trans = btrfs_start_transaction(root, 0);
2231 if (IS_ERR(trans)) {
2232 ret = PTR_ERR(trans);
2233 goto out_release_extents;
2235 trans->in_fsync = true;
2237 ret = btrfs_log_dentry_safe(trans, dentry, &ctx);
2238 btrfs_release_log_ctx_extents(&ctx);
2240 /* Fallthrough and commit/free transaction. */
2241 ret = BTRFS_LOG_FORCE_COMMIT;
2244 /* we've logged all the items and now have a consistent
2245 * version of the file in the log. It is possible that
2246 * someone will come in and modify the file, but that's
2247 * fine because the log is consistent on disk, and we
2248 * have references to all of the file's extents
2250 * It is possible that someone will come in and log the
2251 * file again, but that will end up using the synchronization
2252 * inside btrfs_sync_log to keep things safe.
2254 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
2256 if (ret == BTRFS_NO_LOG_SYNC) {
2257 ret = btrfs_end_transaction(trans);
2261 /* We successfully logged the inode, attempt to sync the log. */
2263 ret = btrfs_sync_log(trans, root, &ctx);
2265 ret = btrfs_end_transaction(trans);
2271 * At this point we need to commit the transaction because we had
2272 * btrfs_need_log_full_commit() or some other error.
2274 * If we didn't do a full sync we have to stop the trans handle, wait on
2275 * the ordered extents, start it again and commit the transaction. If
2276 * we attempt to wait on the ordered extents here we could deadlock with
2277 * something like fallocate() that is holding the extent lock trying to
2278 * start a transaction while some other thread is trying to commit the
2279 * transaction while we (fsync) are currently holding the transaction
2283 ret = btrfs_end_transaction(trans);
2286 ret = btrfs_wait_ordered_range(inode, start, len);
2291 * This is safe to use here because we're only interested in
2292 * making sure the transaction that had the ordered extents is
2293 * committed. We aren't waiting on anything past this point,
2294 * we're purely getting the transaction and committing it.
2296 trans = btrfs_attach_transaction_barrier(root);
2297 if (IS_ERR(trans)) {
2298 ret = PTR_ERR(trans);
2301 * We committed the transaction and there's no currently
2302 * running transaction, this means everything we care
2303 * about made it to disk and we are done.
2311 ret = btrfs_commit_transaction(trans);
2313 ASSERT(list_empty(&ctx.list));
2314 ASSERT(list_empty(&ctx.conflict_inodes));
2315 err = file_check_and_advance_wb_err(file);
2318 return ret > 0 ? -EIO : ret;
2320 out_release_extents:
2321 btrfs_release_log_ctx_extents(&ctx);
2322 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
2326 static const struct vm_operations_struct btrfs_file_vm_ops = {
2327 .fault = filemap_fault,
2328 .map_pages = filemap_map_pages,
2329 .page_mkwrite = btrfs_page_mkwrite,
2332 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
2334 struct address_space *mapping = filp->f_mapping;
2336 if (!mapping->a_ops->read_folio)
2339 file_accessed(filp);
2340 vma->vm_ops = &btrfs_file_vm_ops;
2345 static int hole_mergeable(struct btrfs_inode *inode, struct extent_buffer *leaf,
2346 int slot, u64 start, u64 end)
2348 struct btrfs_file_extent_item *fi;
2349 struct btrfs_key key;
2351 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2354 btrfs_item_key_to_cpu(leaf, &key, slot);
2355 if (key.objectid != btrfs_ino(inode) ||
2356 key.type != BTRFS_EXTENT_DATA_KEY)
2359 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2361 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2364 if (btrfs_file_extent_disk_bytenr(leaf, fi))
2367 if (key.offset == end)
2369 if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2374 static int fill_holes(struct btrfs_trans_handle *trans,
2375 struct btrfs_inode *inode,
2376 struct btrfs_path *path, u64 offset, u64 end)
2378 struct btrfs_fs_info *fs_info = trans->fs_info;
2379 struct btrfs_root *root = inode->root;
2380 struct extent_buffer *leaf;
2381 struct btrfs_file_extent_item *fi;
2382 struct extent_map *hole_em;
2383 struct btrfs_key key;
2386 if (btrfs_fs_incompat(fs_info, NO_HOLES))
2389 key.objectid = btrfs_ino(inode);
2390 key.type = BTRFS_EXTENT_DATA_KEY;
2391 key.offset = offset;
2393 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2396 * We should have dropped this offset, so if we find it then
2397 * something has gone horribly wrong.
2404 leaf = path->nodes[0];
2405 if (hole_mergeable(inode, leaf, path->slots[0] - 1, offset, end)) {
2409 fi = btrfs_item_ptr(leaf, path->slots[0],
2410 struct btrfs_file_extent_item);
2411 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2413 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2414 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2415 btrfs_set_file_extent_offset(leaf, fi, 0);
2416 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2417 btrfs_mark_buffer_dirty(leaf);
2421 if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) {
2424 key.offset = offset;
2425 btrfs_set_item_key_safe(fs_info, path, &key);
2426 fi = btrfs_item_ptr(leaf, path->slots[0],
2427 struct btrfs_file_extent_item);
2428 num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2430 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2431 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2432 btrfs_set_file_extent_offset(leaf, fi, 0);
2433 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2434 btrfs_mark_buffer_dirty(leaf);
2437 btrfs_release_path(path);
2439 ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset,
2445 btrfs_release_path(path);
2447 hole_em = alloc_extent_map();
2449 btrfs_drop_extent_map_range(inode, offset, end - 1, false);
2450 btrfs_set_inode_full_sync(inode);
2452 hole_em->start = offset;
2453 hole_em->len = end - offset;
2454 hole_em->ram_bytes = hole_em->len;
2455 hole_em->orig_start = offset;
2457 hole_em->block_start = EXTENT_MAP_HOLE;
2458 hole_em->block_len = 0;
2459 hole_em->orig_block_len = 0;
2460 hole_em->compress_type = BTRFS_COMPRESS_NONE;
2461 hole_em->generation = trans->transid;
2463 ret = btrfs_replace_extent_map_range(inode, hole_em, true);
2464 free_extent_map(hole_em);
2466 btrfs_set_inode_full_sync(inode);
2473 * Find a hole extent on given inode and change start/len to the end of hole
2474 * extent.(hole/vacuum extent whose em->start <= start &&
2475 * em->start + em->len > start)
2476 * When a hole extent is found, return 1 and modify start/len.
2478 static int find_first_non_hole(struct btrfs_inode *inode, u64 *start, u64 *len)
2480 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2481 struct extent_map *em;
2484 em = btrfs_get_extent(inode, NULL, 0,
2485 round_down(*start, fs_info->sectorsize),
2486 round_up(*len, fs_info->sectorsize));
2490 /* Hole or vacuum extent(only exists in no-hole mode) */
2491 if (em->block_start == EXTENT_MAP_HOLE) {
2493 *len = em->start + em->len > *start + *len ?
2494 0 : *start + *len - em->start - em->len;
2495 *start = em->start + em->len;
2497 free_extent_map(em);
2501 static void btrfs_punch_hole_lock_range(struct inode *inode,
2502 const u64 lockstart,
2504 struct extent_state **cached_state)
2507 * For subpage case, if the range is not at page boundary, we could
2508 * have pages at the leading/tailing part of the range.
2509 * This could lead to dead loop since filemap_range_has_page()
2510 * will always return true.
2511 * So here we need to do extra page alignment for
2512 * filemap_range_has_page().
2514 const u64 page_lockstart = round_up(lockstart, PAGE_SIZE);
2515 const u64 page_lockend = round_down(lockend + 1, PAGE_SIZE) - 1;
2518 truncate_pagecache_range(inode, lockstart, lockend);
2520 lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2523 * We can't have ordered extents in the range, nor dirty/writeback
2524 * pages, because we have locked the inode's VFS lock in exclusive
2525 * mode, we have locked the inode's i_mmap_lock in exclusive mode,
2526 * we have flushed all delalloc in the range and we have waited
2527 * for any ordered extents in the range to complete.
2528 * We can race with anyone reading pages from this range, so after
2529 * locking the range check if we have pages in the range, and if
2530 * we do, unlock the range and retry.
2532 if (!filemap_range_has_page(inode->i_mapping, page_lockstart,
2536 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2540 btrfs_assert_inode_range_clean(BTRFS_I(inode), lockstart, lockend);
2543 static int btrfs_insert_replace_extent(struct btrfs_trans_handle *trans,
2544 struct btrfs_inode *inode,
2545 struct btrfs_path *path,
2546 struct btrfs_replace_extent_info *extent_info,
2547 const u64 replace_len,
2548 const u64 bytes_to_drop)
2550 struct btrfs_fs_info *fs_info = trans->fs_info;
2551 struct btrfs_root *root = inode->root;
2552 struct btrfs_file_extent_item *extent;
2553 struct extent_buffer *leaf;
2554 struct btrfs_key key;
2556 struct btrfs_ref ref = { 0 };
2559 if (replace_len == 0)
2562 if (extent_info->disk_offset == 0 &&
2563 btrfs_fs_incompat(fs_info, NO_HOLES)) {
2564 btrfs_update_inode_bytes(inode, 0, bytes_to_drop);
2568 key.objectid = btrfs_ino(inode);
2569 key.type = BTRFS_EXTENT_DATA_KEY;
2570 key.offset = extent_info->file_offset;
2571 ret = btrfs_insert_empty_item(trans, root, path, &key,
2572 sizeof(struct btrfs_file_extent_item));
2575 leaf = path->nodes[0];
2576 slot = path->slots[0];
2577 write_extent_buffer(leaf, extent_info->extent_buf,
2578 btrfs_item_ptr_offset(leaf, slot),
2579 sizeof(struct btrfs_file_extent_item));
2580 extent = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2581 ASSERT(btrfs_file_extent_type(leaf, extent) != BTRFS_FILE_EXTENT_INLINE);
2582 btrfs_set_file_extent_offset(leaf, extent, extent_info->data_offset);
2583 btrfs_set_file_extent_num_bytes(leaf, extent, replace_len);
2584 if (extent_info->is_new_extent)
2585 btrfs_set_file_extent_generation(leaf, extent, trans->transid);
2586 btrfs_mark_buffer_dirty(leaf);
2587 btrfs_release_path(path);
2589 ret = btrfs_inode_set_file_extent_range(inode, extent_info->file_offset,
2594 /* If it's a hole, nothing more needs to be done. */
2595 if (extent_info->disk_offset == 0) {
2596 btrfs_update_inode_bytes(inode, 0, bytes_to_drop);
2600 btrfs_update_inode_bytes(inode, replace_len, bytes_to_drop);
2602 if (extent_info->is_new_extent && extent_info->insertions == 0) {
2603 key.objectid = extent_info->disk_offset;
2604 key.type = BTRFS_EXTENT_ITEM_KEY;
2605 key.offset = extent_info->disk_len;
2606 ret = btrfs_alloc_reserved_file_extent(trans, root,
2608 extent_info->file_offset,
2609 extent_info->qgroup_reserved,
2614 btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
2615 extent_info->disk_offset,
2616 extent_info->disk_len, 0);
2617 ref_offset = extent_info->file_offset - extent_info->data_offset;
2618 btrfs_init_data_ref(&ref, root->root_key.objectid,
2619 btrfs_ino(inode), ref_offset, 0, false);
2620 ret = btrfs_inc_extent_ref(trans, &ref);
2623 extent_info->insertions++;
2629 * The respective range must have been previously locked, as well as the inode.
2630 * The end offset is inclusive (last byte of the range).
2631 * @extent_info is NULL for fallocate's hole punching and non-NULL when replacing
2632 * the file range with an extent.
2633 * When not punching a hole, we don't want to end up in a state where we dropped
2634 * extents without inserting a new one, so we must abort the transaction to avoid
2637 int btrfs_replace_file_extents(struct btrfs_inode *inode,
2638 struct btrfs_path *path, const u64 start,
2640 struct btrfs_replace_extent_info *extent_info,
2641 struct btrfs_trans_handle **trans_out)
2643 struct btrfs_drop_extents_args drop_args = { 0 };
2644 struct btrfs_root *root = inode->root;
2645 struct btrfs_fs_info *fs_info = root->fs_info;
2646 u64 min_size = btrfs_calc_insert_metadata_size(fs_info, 1);
2647 u64 ino_size = round_up(inode->vfs_inode.i_size, fs_info->sectorsize);
2648 struct btrfs_trans_handle *trans = NULL;
2649 struct btrfs_block_rsv *rsv;
2650 unsigned int rsv_count;
2652 u64 len = end - start;
2658 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
2663 rsv->size = btrfs_calc_insert_metadata_size(fs_info, 1);
2664 rsv->failfast = true;
2667 * 1 - update the inode
2668 * 1 - removing the extents in the range
2669 * 1 - adding the hole extent if no_holes isn't set or if we are
2670 * replacing the range with a new extent
2672 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || extent_info)
2677 trans = btrfs_start_transaction(root, rsv_count);
2678 if (IS_ERR(trans)) {
2679 ret = PTR_ERR(trans);
2684 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
2688 trans->block_rsv = rsv;
2691 drop_args.path = path;
2692 drop_args.end = end + 1;
2693 drop_args.drop_cache = true;
2694 while (cur_offset < end) {
2695 drop_args.start = cur_offset;
2696 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2697 /* If we are punching a hole decrement the inode's byte count */
2699 btrfs_update_inode_bytes(inode, 0,
2700 drop_args.bytes_found);
2701 if (ret != -ENOSPC) {
2703 * The only time we don't want to abort is if we are
2704 * attempting to clone a partial inline extent, in which
2705 * case we'll get EOPNOTSUPP. However if we aren't
2706 * clone we need to abort no matter what, because if we
2707 * got EOPNOTSUPP via prealloc then we messed up and
2711 (ret != -EOPNOTSUPP ||
2712 (extent_info && extent_info->is_new_extent)))
2713 btrfs_abort_transaction(trans, ret);
2717 trans->block_rsv = &fs_info->trans_block_rsv;
2719 if (!extent_info && cur_offset < drop_args.drop_end &&
2720 cur_offset < ino_size) {
2721 ret = fill_holes(trans, inode, path, cur_offset,
2722 drop_args.drop_end);
2725 * If we failed then we didn't insert our hole
2726 * entries for the area we dropped, so now the
2727 * fs is corrupted, so we must abort the
2730 btrfs_abort_transaction(trans, ret);
2733 } else if (!extent_info && cur_offset < drop_args.drop_end) {
2735 * We are past the i_size here, but since we didn't
2736 * insert holes we need to clear the mapped area so we
2737 * know to not set disk_i_size in this area until a new
2738 * file extent is inserted here.
2740 ret = btrfs_inode_clear_file_extent_range(inode,
2742 drop_args.drop_end - cur_offset);
2745 * We couldn't clear our area, so we could
2746 * presumably adjust up and corrupt the fs, so
2749 btrfs_abort_transaction(trans, ret);
2755 drop_args.drop_end > extent_info->file_offset) {
2756 u64 replace_len = drop_args.drop_end -
2757 extent_info->file_offset;
2759 ret = btrfs_insert_replace_extent(trans, inode, path,
2760 extent_info, replace_len,
2761 drop_args.bytes_found);
2763 btrfs_abort_transaction(trans, ret);
2766 extent_info->data_len -= replace_len;
2767 extent_info->data_offset += replace_len;
2768 extent_info->file_offset += replace_len;
2772 * We are releasing our handle on the transaction, balance the
2773 * dirty pages of the btree inode and flush delayed items, and
2774 * then get a new transaction handle, which may now point to a
2775 * new transaction in case someone else may have committed the
2776 * transaction we used to replace/drop file extent items. So
2777 * bump the inode's iversion and update mtime and ctime except
2778 * if we are called from a dedupe context. This is because a
2779 * power failure/crash may happen after the transaction is
2780 * committed and before we finish replacing/dropping all the
2781 * file extent items we need.
2783 inode_inc_iversion(&inode->vfs_inode);
2785 if (!extent_info || extent_info->update_times) {
2786 inode->vfs_inode.i_mtime = current_time(&inode->vfs_inode);
2787 inode->vfs_inode.i_ctime = inode->vfs_inode.i_mtime;
2790 ret = btrfs_update_inode(trans, root, inode);
2794 btrfs_end_transaction(trans);
2795 btrfs_btree_balance_dirty(fs_info);
2797 trans = btrfs_start_transaction(root, rsv_count);
2798 if (IS_ERR(trans)) {
2799 ret = PTR_ERR(trans);
2804 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
2805 rsv, min_size, false);
2808 trans->block_rsv = rsv;
2810 cur_offset = drop_args.drop_end;
2811 len = end - cur_offset;
2812 if (!extent_info && len) {
2813 ret = find_first_non_hole(inode, &cur_offset, &len);
2814 if (unlikely(ret < 0))
2824 * If we were cloning, force the next fsync to be a full one since we
2825 * we replaced (or just dropped in the case of cloning holes when
2826 * NO_HOLES is enabled) file extent items and did not setup new extent
2827 * maps for the replacement extents (or holes).
2829 if (extent_info && !extent_info->is_new_extent)
2830 btrfs_set_inode_full_sync(inode);
2835 trans->block_rsv = &fs_info->trans_block_rsv;
2837 * If we are using the NO_HOLES feature we might have had already an
2838 * hole that overlaps a part of the region [lockstart, lockend] and
2839 * ends at (or beyond) lockend. Since we have no file extent items to
2840 * represent holes, drop_end can be less than lockend and so we must
2841 * make sure we have an extent map representing the existing hole (the
2842 * call to __btrfs_drop_extents() might have dropped the existing extent
2843 * map representing the existing hole), otherwise the fast fsync path
2844 * will not record the existence of the hole region
2845 * [existing_hole_start, lockend].
2847 if (drop_args.drop_end <= end)
2848 drop_args.drop_end = end + 1;
2850 * Don't insert file hole extent item if it's for a range beyond eof
2851 * (because it's useless) or if it represents a 0 bytes range (when
2852 * cur_offset == drop_end).
2854 if (!extent_info && cur_offset < ino_size &&
2855 cur_offset < drop_args.drop_end) {
2856 ret = fill_holes(trans, inode, path, cur_offset,
2857 drop_args.drop_end);
2859 /* Same comment as above. */
2860 btrfs_abort_transaction(trans, ret);
2863 } else if (!extent_info && cur_offset < drop_args.drop_end) {
2864 /* See the comment in the loop above for the reasoning here. */
2865 ret = btrfs_inode_clear_file_extent_range(inode, cur_offset,
2866 drop_args.drop_end - cur_offset);
2868 btrfs_abort_transaction(trans, ret);
2874 ret = btrfs_insert_replace_extent(trans, inode, path,
2875 extent_info, extent_info->data_len,
2876 drop_args.bytes_found);
2878 btrfs_abort_transaction(trans, ret);
2887 trans->block_rsv = &fs_info->trans_block_rsv;
2889 btrfs_end_transaction(trans);
2893 btrfs_free_block_rsv(fs_info, rsv);
2898 static int btrfs_punch_hole(struct file *file, loff_t offset, loff_t len)
2900 struct inode *inode = file_inode(file);
2901 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2902 struct btrfs_root *root = BTRFS_I(inode)->root;
2903 struct extent_state *cached_state = NULL;
2904 struct btrfs_path *path;
2905 struct btrfs_trans_handle *trans = NULL;
2910 u64 orig_start = offset;
2914 bool truncated_block = false;
2915 bool updated_inode = false;
2917 btrfs_inode_lock(inode, BTRFS_ILOCK_MMAP);
2919 ret = btrfs_wait_ordered_range(inode, offset, len);
2921 goto out_only_mutex;
2923 ino_size = round_up(inode->i_size, fs_info->sectorsize);
2924 ret = find_first_non_hole(BTRFS_I(inode), &offset, &len);
2926 goto out_only_mutex;
2928 /* Already in a large hole */
2930 goto out_only_mutex;
2933 ret = file_modified(file);
2935 goto out_only_mutex;
2937 lockstart = round_up(offset, fs_info->sectorsize);
2938 lockend = round_down(offset + len, fs_info->sectorsize) - 1;
2939 same_block = (BTRFS_BYTES_TO_BLKS(fs_info, offset))
2940 == (BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1));
2942 * We needn't truncate any block which is beyond the end of the file
2943 * because we are sure there is no data there.
2946 * Only do this if we are in the same block and we aren't doing the
2949 if (same_block && len < fs_info->sectorsize) {
2950 if (offset < ino_size) {
2951 truncated_block = true;
2952 ret = btrfs_truncate_block(BTRFS_I(inode), offset, len,
2957 goto out_only_mutex;
2960 /* zero back part of the first block */
2961 if (offset < ino_size) {
2962 truncated_block = true;
2963 ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0);
2965 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
2970 /* Check the aligned pages after the first unaligned page,
2971 * if offset != orig_start, which means the first unaligned page
2972 * including several following pages are already in holes,
2973 * the extra check can be skipped */
2974 if (offset == orig_start) {
2975 /* after truncate page, check hole again */
2976 len = offset + len - lockstart;
2978 ret = find_first_non_hole(BTRFS_I(inode), &offset, &len);
2980 goto out_only_mutex;
2983 goto out_only_mutex;
2988 /* Check the tail unaligned part is in a hole */
2989 tail_start = lockend + 1;
2990 tail_len = offset + len - tail_start;
2992 ret = find_first_non_hole(BTRFS_I(inode), &tail_start, &tail_len);
2993 if (unlikely(ret < 0))
2994 goto out_only_mutex;
2996 /* zero the front end of the last page */
2997 if (tail_start + tail_len < ino_size) {
2998 truncated_block = true;
2999 ret = btrfs_truncate_block(BTRFS_I(inode),
3000 tail_start + tail_len,
3003 goto out_only_mutex;
3008 if (lockend < lockstart) {
3010 goto out_only_mutex;
3013 btrfs_punch_hole_lock_range(inode, lockstart, lockend, &cached_state);
3015 path = btrfs_alloc_path();
3021 ret = btrfs_replace_file_extents(BTRFS_I(inode), path, lockstart,
3022 lockend, NULL, &trans);
3023 btrfs_free_path(path);
3027 ASSERT(trans != NULL);
3028 inode_inc_iversion(inode);
3029 inode->i_mtime = current_time(inode);
3030 inode->i_ctime = inode->i_mtime;
3031 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
3032 updated_inode = true;
3033 btrfs_end_transaction(trans);
3034 btrfs_btree_balance_dirty(fs_info);
3036 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
3039 if (!updated_inode && truncated_block && !ret) {
3041 * If we only end up zeroing part of a page, we still need to
3042 * update the inode item, so that all the time fields are
3043 * updated as well as the necessary btrfs inode in memory fields
3044 * for detecting, at fsync time, if the inode isn't yet in the
3045 * log tree or it's there but not up to date.
3047 struct timespec64 now = current_time(inode);
3049 inode_inc_iversion(inode);
3050 inode->i_mtime = now;
3051 inode->i_ctime = now;
3052 trans = btrfs_start_transaction(root, 1);
3053 if (IS_ERR(trans)) {
3054 ret = PTR_ERR(trans);
3058 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
3059 ret2 = btrfs_end_transaction(trans);
3064 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
3068 /* Helper structure to record which range is already reserved */
3069 struct falloc_range {
3070 struct list_head list;
3076 * Helper function to add falloc range
3078 * Caller should have locked the larger range of extent containing
3081 static int add_falloc_range(struct list_head *head, u64 start, u64 len)
3083 struct falloc_range *range = NULL;
3085 if (!list_empty(head)) {
3087 * As fallocate iterates by bytenr order, we only need to check
3090 range = list_last_entry(head, struct falloc_range, list);
3091 if (range->start + range->len == start) {
3097 range = kmalloc(sizeof(*range), GFP_KERNEL);
3100 range->start = start;
3102 list_add_tail(&range->list, head);
3106 static int btrfs_fallocate_update_isize(struct inode *inode,
3110 struct btrfs_trans_handle *trans;
3111 struct btrfs_root *root = BTRFS_I(inode)->root;
3115 if (mode & FALLOC_FL_KEEP_SIZE || end <= i_size_read(inode))
3118 trans = btrfs_start_transaction(root, 1);
3120 return PTR_ERR(trans);
3122 inode->i_ctime = current_time(inode);
3123 i_size_write(inode, end);
3124 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
3125 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
3126 ret2 = btrfs_end_transaction(trans);
3128 return ret ? ret : ret2;
3132 RANGE_BOUNDARY_WRITTEN_EXTENT,
3133 RANGE_BOUNDARY_PREALLOC_EXTENT,
3134 RANGE_BOUNDARY_HOLE,
3137 static int btrfs_zero_range_check_range_boundary(struct btrfs_inode *inode,
3140 const u64 sectorsize = inode->root->fs_info->sectorsize;
3141 struct extent_map *em;
3144 offset = round_down(offset, sectorsize);
3145 em = btrfs_get_extent(inode, NULL, 0, offset, sectorsize);
3149 if (em->block_start == EXTENT_MAP_HOLE)
3150 ret = RANGE_BOUNDARY_HOLE;
3151 else if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
3152 ret = RANGE_BOUNDARY_PREALLOC_EXTENT;
3154 ret = RANGE_BOUNDARY_WRITTEN_EXTENT;
3156 free_extent_map(em);
3160 static int btrfs_zero_range(struct inode *inode,
3165 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3166 struct extent_map *em;
3167 struct extent_changeset *data_reserved = NULL;
3170 const u64 sectorsize = fs_info->sectorsize;
3171 u64 alloc_start = round_down(offset, sectorsize);
3172 u64 alloc_end = round_up(offset + len, sectorsize);
3173 u64 bytes_to_reserve = 0;
3174 bool space_reserved = false;
3176 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start,
3177 alloc_end - alloc_start);
3184 * Avoid hole punching and extent allocation for some cases. More cases
3185 * could be considered, but these are unlikely common and we keep things
3186 * as simple as possible for now. Also, intentionally, if the target
3187 * range contains one or more prealloc extents together with regular
3188 * extents and holes, we drop all the existing extents and allocate a
3189 * new prealloc extent, so that we get a larger contiguous disk extent.
3191 if (em->start <= alloc_start &&
3192 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3193 const u64 em_end = em->start + em->len;
3195 if (em_end >= offset + len) {
3197 * The whole range is already a prealloc extent,
3198 * do nothing except updating the inode's i_size if
3201 free_extent_map(em);
3202 ret = btrfs_fallocate_update_isize(inode, offset + len,
3207 * Part of the range is already a prealloc extent, so operate
3208 * only on the remaining part of the range.
3210 alloc_start = em_end;
3211 ASSERT(IS_ALIGNED(alloc_start, sectorsize));
3212 len = offset + len - alloc_start;
3213 offset = alloc_start;
3214 alloc_hint = em->block_start + em->len;
3216 free_extent_map(em);
3218 if (BTRFS_BYTES_TO_BLKS(fs_info, offset) ==
3219 BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)) {
3220 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, alloc_start,
3227 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3228 free_extent_map(em);
3229 ret = btrfs_fallocate_update_isize(inode, offset + len,
3233 if (len < sectorsize && em->block_start != EXTENT_MAP_HOLE) {
3234 free_extent_map(em);
3235 ret = btrfs_truncate_block(BTRFS_I(inode), offset, len,
3238 ret = btrfs_fallocate_update_isize(inode,
3243 free_extent_map(em);
3244 alloc_start = round_down(offset, sectorsize);
3245 alloc_end = alloc_start + sectorsize;
3249 alloc_start = round_up(offset, sectorsize);
3250 alloc_end = round_down(offset + len, sectorsize);
3253 * For unaligned ranges, check the pages at the boundaries, they might
3254 * map to an extent, in which case we need to partially zero them, or
3255 * they might map to a hole, in which case we need our allocation range
3258 if (!IS_ALIGNED(offset, sectorsize)) {
3259 ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode),
3263 if (ret == RANGE_BOUNDARY_HOLE) {
3264 alloc_start = round_down(offset, sectorsize);
3266 } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
3267 ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0);
3275 if (!IS_ALIGNED(offset + len, sectorsize)) {
3276 ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode),
3280 if (ret == RANGE_BOUNDARY_HOLE) {
3281 alloc_end = round_up(offset + len, sectorsize);
3283 } else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
3284 ret = btrfs_truncate_block(BTRFS_I(inode), offset + len,
3294 if (alloc_start < alloc_end) {
3295 struct extent_state *cached_state = NULL;
3296 const u64 lockstart = alloc_start;
3297 const u64 lockend = alloc_end - 1;
3299 bytes_to_reserve = alloc_end - alloc_start;
3300 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
3304 space_reserved = true;
3305 btrfs_punch_hole_lock_range(inode, lockstart, lockend,
3307 ret = btrfs_qgroup_reserve_data(BTRFS_I(inode), &data_reserved,
3308 alloc_start, bytes_to_reserve);
3310 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart,
3311 lockend, &cached_state);
3314 ret = btrfs_prealloc_file_range(inode, mode, alloc_start,
3315 alloc_end - alloc_start,
3317 offset + len, &alloc_hint);
3318 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
3320 /* btrfs_prealloc_file_range releases reserved space on error */
3322 space_reserved = false;
3326 ret = btrfs_fallocate_update_isize(inode, offset + len, mode);
3328 if (ret && space_reserved)
3329 btrfs_free_reserved_data_space(BTRFS_I(inode), data_reserved,
3330 alloc_start, bytes_to_reserve);
3331 extent_changeset_free(data_reserved);
3336 static long btrfs_fallocate(struct file *file, int mode,
3337 loff_t offset, loff_t len)
3339 struct inode *inode = file_inode(file);
3340 struct extent_state *cached_state = NULL;
3341 struct extent_changeset *data_reserved = NULL;
3342 struct falloc_range *range;
3343 struct falloc_range *tmp;
3344 struct list_head reserve_list;
3352 u64 data_space_needed = 0;
3353 u64 data_space_reserved = 0;
3354 u64 qgroup_reserved = 0;
3355 struct extent_map *em;
3356 int blocksize = BTRFS_I(inode)->root->fs_info->sectorsize;
3359 /* Do not allow fallocate in ZONED mode */
3360 if (btrfs_is_zoned(btrfs_sb(inode->i_sb)))
3363 alloc_start = round_down(offset, blocksize);
3364 alloc_end = round_up(offset + len, blocksize);
3365 cur_offset = alloc_start;
3367 /* Make sure we aren't being give some crap mode */
3368 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
3369 FALLOC_FL_ZERO_RANGE))
3372 if (mode & FALLOC_FL_PUNCH_HOLE)
3373 return btrfs_punch_hole(file, offset, len);
3375 btrfs_inode_lock(inode, BTRFS_ILOCK_MMAP);
3377 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) {
3378 ret = inode_newsize_ok(inode, offset + len);
3383 ret = file_modified(file);
3388 * TODO: Move these two operations after we have checked
3389 * accurate reserved space, or fallocate can still fail but
3390 * with page truncated or size expanded.
3392 * But that's a minor problem and won't do much harm BTW.
3394 if (alloc_start > inode->i_size) {
3395 ret = btrfs_cont_expand(BTRFS_I(inode), i_size_read(inode),
3399 } else if (offset + len > inode->i_size) {
3401 * If we are fallocating from the end of the file onward we
3402 * need to zero out the end of the block if i_size lands in the
3403 * middle of a block.
3405 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
3411 * We have locked the inode at the VFS level (in exclusive mode) and we
3412 * have locked the i_mmap_lock lock (in exclusive mode). Now before
3413 * locking the file range, flush all dealloc in the range and wait for
3414 * all ordered extents in the range to complete. After this we can lock
3415 * the file range and, due to the previous locking we did, we know there
3416 * can't be more delalloc or ordered extents in the range.
3418 ret = btrfs_wait_ordered_range(inode, alloc_start,
3419 alloc_end - alloc_start);
3423 if (mode & FALLOC_FL_ZERO_RANGE) {
3424 ret = btrfs_zero_range(inode, offset, len, mode);
3425 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
3429 locked_end = alloc_end - 1;
3430 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
3433 btrfs_assert_inode_range_clean(BTRFS_I(inode), alloc_start, locked_end);
3435 /* First, check if we exceed the qgroup limit */
3436 INIT_LIST_HEAD(&reserve_list);
3437 while (cur_offset < alloc_end) {
3438 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
3439 alloc_end - cur_offset);
3444 last_byte = min(extent_map_end(em), alloc_end);
3445 actual_end = min_t(u64, extent_map_end(em), offset + len);
3446 last_byte = ALIGN(last_byte, blocksize);
3447 if (em->block_start == EXTENT_MAP_HOLE ||
3448 (cur_offset >= inode->i_size &&
3449 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
3450 const u64 range_len = last_byte - cur_offset;
3452 ret = add_falloc_range(&reserve_list, cur_offset, range_len);
3454 free_extent_map(em);
3457 ret = btrfs_qgroup_reserve_data(BTRFS_I(inode),
3458 &data_reserved, cur_offset, range_len);
3460 free_extent_map(em);
3463 qgroup_reserved += range_len;
3464 data_space_needed += range_len;
3466 free_extent_map(em);
3467 cur_offset = last_byte;
3470 if (!ret && data_space_needed > 0) {
3472 * We are safe to reserve space here as we can't have delalloc
3473 * in the range, see above.
3475 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
3478 data_space_reserved = data_space_needed;
3482 * If ret is still 0, means we're OK to fallocate.
3483 * Or just cleanup the list and exit.
3485 list_for_each_entry_safe(range, tmp, &reserve_list, list) {
3487 ret = btrfs_prealloc_file_range(inode, mode,
3489 range->len, i_blocksize(inode),
3490 offset + len, &alloc_hint);
3492 * btrfs_prealloc_file_range() releases space even
3493 * if it returns an error.
3495 data_space_reserved -= range->len;
3496 qgroup_reserved -= range->len;
3497 } else if (data_space_reserved > 0) {
3498 btrfs_free_reserved_data_space(BTRFS_I(inode),
3499 data_reserved, range->start,
3501 data_space_reserved -= range->len;
3502 qgroup_reserved -= range->len;
3503 } else if (qgroup_reserved > 0) {
3504 btrfs_qgroup_free_data(BTRFS_I(inode), data_reserved,
3505 range->start, range->len);
3506 qgroup_reserved -= range->len;
3508 list_del(&range->list);
3515 * We didn't need to allocate any more space, but we still extended the
3516 * size of the file so we need to update i_size and the inode item.
3518 ret = btrfs_fallocate_update_isize(inode, actual_end, mode);
3520 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
3523 btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
3524 extent_changeset_free(data_reserved);
3529 * Helper for btrfs_find_delalloc_in_range(). Find a subrange in a given range
3530 * that has unflushed and/or flushing delalloc. There might be other adjacent
3531 * subranges after the one it found, so btrfs_find_delalloc_in_range() keeps
3532 * looping while it gets adjacent subranges, and merging them together.
3534 static bool find_delalloc_subrange(struct btrfs_inode *inode, u64 start, u64 end,
3535 u64 *delalloc_start_ret, u64 *delalloc_end_ret)
3537 const u64 len = end + 1 - start;
3538 struct extent_map_tree *em_tree = &inode->extent_tree;
3539 struct extent_map *em;
3544 * Search the io tree first for EXTENT_DELALLOC. If we find any, it
3545 * means we have delalloc (dirty pages) for which writeback has not
3548 *delalloc_start_ret = start;
3549 delalloc_len = count_range_bits(&inode->io_tree, delalloc_start_ret, end,
3550 len, EXTENT_DELALLOC, 1);
3552 * If delalloc was found then *delalloc_start_ret has a sector size
3553 * aligned value (rounded down).
3555 if (delalloc_len > 0)
3556 *delalloc_end_ret = *delalloc_start_ret + delalloc_len - 1;
3559 * Now also check if there's any extent map in the range that does not
3560 * map to a hole or prealloc extent. We do this because:
3562 * 1) When delalloc is flushed, the file range is locked, we clear the
3563 * EXTENT_DELALLOC bit from the io tree and create an extent map for
3564 * an allocated extent. So we might just have been called after
3565 * delalloc is flushed and before the ordered extent completes and
3566 * inserts the new file extent item in the subvolume's btree;
3568 * 2) We may have an extent map created by flushing delalloc for a
3569 * subrange that starts before the subrange we found marked with
3570 * EXTENT_DELALLOC in the io tree.
3572 read_lock(&em_tree->lock);
3573 em = lookup_extent_mapping(em_tree, start, len);
3574 read_unlock(&em_tree->lock);
3576 /* extent_map_end() returns a non-inclusive end offset. */
3577 em_end = em ? extent_map_end(em) : 0;
3580 * If we have a hole/prealloc extent map, check the next one if this one
3581 * ends before our range's end.
3583 if (em && (em->block_start == EXTENT_MAP_HOLE ||
3584 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) && em_end < end) {
3585 struct extent_map *next_em;
3587 read_lock(&em_tree->lock);
3588 next_em = lookup_extent_mapping(em_tree, em_end, len - em_end);
3589 read_unlock(&em_tree->lock);
3591 free_extent_map(em);
3592 em_end = next_em ? extent_map_end(next_em) : 0;
3596 if (em && (em->block_start == EXTENT_MAP_HOLE ||
3597 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
3598 free_extent_map(em);
3603 * No extent map or one for a hole or prealloc extent. Use the delalloc
3604 * range we found in the io tree if we have one.
3607 return (delalloc_len > 0);
3610 * We don't have any range as EXTENT_DELALLOC in the io tree, so the
3611 * extent map is the only subrange representing delalloc.
3613 if (delalloc_len == 0) {
3614 *delalloc_start_ret = em->start;
3615 *delalloc_end_ret = min(end, em_end - 1);
3616 free_extent_map(em);
3621 * The extent map represents a delalloc range that starts before the
3622 * delalloc range we found in the io tree.
3624 if (em->start < *delalloc_start_ret) {
3625 *delalloc_start_ret = em->start;
3627 * If the ranges are adjacent, return a combined range.
3628 * Otherwise return the extent map's range.
3630 if (em_end < *delalloc_start_ret)
3631 *delalloc_end_ret = min(end, em_end - 1);
3633 free_extent_map(em);
3638 * The extent map starts after the delalloc range we found in the io
3639 * tree. If it's adjacent, return a combined range, otherwise return
3640 * the range found in the io tree.
3642 if (*delalloc_end_ret + 1 == em->start)
3643 *delalloc_end_ret = min(end, em_end - 1);
3645 free_extent_map(em);
3650 * Check if there's delalloc in a given range.
3652 * @inode: The inode.
3653 * @start: The start offset of the range. It does not need to be
3654 * sector size aligned.
3655 * @end: The end offset (inclusive value) of the search range.
3656 * It does not need to be sector size aligned.
3657 * @delalloc_start_ret: Output argument, set to the start offset of the
3658 * subrange found with delalloc (may not be sector size
3660 * @delalloc_end_ret: Output argument, set to he end offset (inclusive value)
3661 * of the subrange found with delalloc.
3663 * Returns true if a subrange with delalloc is found within the given range, and
3664 * if so it sets @delalloc_start_ret and @delalloc_end_ret with the start and
3665 * end offsets of the subrange.
3667 bool btrfs_find_delalloc_in_range(struct btrfs_inode *inode, u64 start, u64 end,
3668 u64 *delalloc_start_ret, u64 *delalloc_end_ret)
3670 u64 cur_offset = round_down(start, inode->root->fs_info->sectorsize);
3671 u64 prev_delalloc_end = 0;
3674 while (cur_offset < end) {
3679 delalloc = find_delalloc_subrange(inode, cur_offset, end,
3685 if (prev_delalloc_end == 0) {
3686 /* First subrange found. */
3687 *delalloc_start_ret = max(delalloc_start, start);
3688 *delalloc_end_ret = delalloc_end;
3690 } else if (delalloc_start == prev_delalloc_end + 1) {
3691 /* Subrange adjacent to the previous one, merge them. */
3692 *delalloc_end_ret = delalloc_end;
3694 /* Subrange not adjacent to the previous one, exit. */
3698 prev_delalloc_end = delalloc_end;
3699 cur_offset = delalloc_end + 1;
3707 * Check if there's a hole or delalloc range in a range representing a hole (or
3708 * prealloc extent) found in the inode's subvolume btree.
3710 * @inode: The inode.
3711 * @whence: Seek mode (SEEK_DATA or SEEK_HOLE).
3712 * @start: Start offset of the hole region. It does not need to be sector
3714 * @end: End offset (inclusive value) of the hole region. It does not
3715 * need to be sector size aligned.
3716 * @start_ret: Return parameter, used to set the start of the subrange in the
3717 * hole that matches the search criteria (seek mode), if such
3718 * subrange is found (return value of the function is true).
3719 * The value returned here may not be sector size aligned.
3721 * Returns true if a subrange matching the given seek mode is found, and if one
3722 * is found, it updates @start_ret with the start of the subrange.
3724 static bool find_desired_extent_in_hole(struct btrfs_inode *inode, int whence,
3725 u64 start, u64 end, u64 *start_ret)
3731 delalloc = btrfs_find_delalloc_in_range(inode, start, end,
3732 &delalloc_start, &delalloc_end);
3733 if (delalloc && whence == SEEK_DATA) {
3734 *start_ret = delalloc_start;
3738 if (delalloc && whence == SEEK_HOLE) {
3740 * We found delalloc but it starts after out start offset. So we
3741 * have a hole between our start offset and the delalloc start.
3743 if (start < delalloc_start) {
3748 * Delalloc range starts at our start offset.
3749 * If the delalloc range's length is smaller than our range,
3750 * then it means we have a hole that starts where the delalloc
3753 if (delalloc_end < end) {
3754 *start_ret = delalloc_end + 1;
3758 /* There's delalloc for the whole range. */
3762 if (!delalloc && whence == SEEK_HOLE) {
3768 * No delalloc in the range and we are seeking for data. The caller has
3769 * to iterate to the next extent item in the subvolume btree.
3774 static loff_t find_desired_extent(struct btrfs_inode *inode, loff_t offset,
3777 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3778 struct extent_state *cached_state = NULL;
3779 const loff_t i_size = i_size_read(&inode->vfs_inode);
3780 const u64 ino = btrfs_ino(inode);
3781 struct btrfs_root *root = inode->root;
3782 struct btrfs_path *path;
3783 struct btrfs_key key;
3784 u64 last_extent_end;
3791 if (i_size == 0 || offset >= i_size)
3795 * Quick path. If the inode has no prealloc extents and its number of
3796 * bytes used matches its i_size, then it can not have holes.
3798 if (whence == SEEK_HOLE &&
3799 !(inode->flags & BTRFS_INODE_PREALLOC) &&
3800 inode_get_bytes(&inode->vfs_inode) == i_size)
3804 * offset can be negative, in this case we start finding DATA/HOLE from
3805 * the very start of the file.
3807 start = max_t(loff_t, 0, offset);
3809 lockstart = round_down(start, fs_info->sectorsize);
3810 lockend = round_up(i_size, fs_info->sectorsize);
3811 if (lockend <= lockstart)
3812 lockend = lockstart + fs_info->sectorsize;
3815 path = btrfs_alloc_path();
3818 path->reada = READA_FORWARD;
3821 key.type = BTRFS_EXTENT_DATA_KEY;
3824 last_extent_end = lockstart;
3826 lock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
3828 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3831 } else if (ret > 0 && path->slots[0] > 0) {
3832 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
3833 if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY)
3837 while (start < i_size) {
3838 struct extent_buffer *leaf = path->nodes[0];
3839 struct btrfs_file_extent_item *extent;
3842 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3843 ret = btrfs_next_leaf(root, path);
3849 leaf = path->nodes[0];
3852 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3853 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
3856 extent_end = btrfs_file_extent_end(path);
3859 * In the first iteration we may have a slot that points to an
3860 * extent that ends before our start offset, so skip it.
3862 if (extent_end <= start) {
3867 /* We have an implicit hole, NO_HOLES feature is likely set. */
3868 if (last_extent_end < key.offset) {
3869 u64 search_start = last_extent_end;
3873 * First iteration, @start matches @offset and it's
3876 if (start == offset)
3877 search_start = offset;
3879 found = find_desired_extent_in_hole(inode, whence,
3884 start = found_start;
3888 * Didn't find data or a hole (due to delalloc) in the
3889 * implicit hole range, so need to analyze the extent.
3893 extent = btrfs_item_ptr(leaf, path->slots[0],
3894 struct btrfs_file_extent_item);
3896 if (btrfs_file_extent_disk_bytenr(leaf, extent) == 0 ||
3897 btrfs_file_extent_type(leaf, extent) ==
3898 BTRFS_FILE_EXTENT_PREALLOC) {
3900 * Explicit hole or prealloc extent, search for delalloc.
3901 * A prealloc extent is treated like a hole.
3903 u64 search_start = key.offset;
3907 * First iteration, @start matches @offset and it's
3910 if (start == offset)
3911 search_start = offset;
3913 found = find_desired_extent_in_hole(inode, whence,
3918 start = found_start;
3922 * Didn't find data or a hole (due to delalloc) in the
3923 * implicit hole range, so need to analyze the next
3928 * Found a regular or inline extent.
3929 * If we are seeking for data, adjust the start offset
3930 * and stop, we're done.
3932 if (whence == SEEK_DATA) {
3933 start = max_t(u64, key.offset, offset);
3938 * Else, we are seeking for a hole, check the next file
3944 last_extent_end = extent_end;
3946 if (fatal_signal_pending(current)) {
3953 /* We have an implicit hole from the last extent found up to i_size. */
3954 if (!found && start < i_size) {
3955 found = find_desired_extent_in_hole(inode, whence, start,
3956 i_size - 1, &start);
3962 unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
3963 btrfs_free_path(path);
3968 if (whence == SEEK_DATA && start >= i_size)
3971 return min_t(loff_t, start, i_size);
3974 static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
3976 struct inode *inode = file->f_mapping->host;
3980 return generic_file_llseek(file, offset, whence);
3983 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
3984 offset = find_desired_extent(BTRFS_I(inode), offset, whence);
3985 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
3992 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
3995 static int btrfs_file_open(struct inode *inode, struct file *filp)
3999 filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC | FMODE_BUF_WASYNC;
4001 ret = fsverity_file_open(inode, filp);
4004 return generic_file_open(inode, filp);
4007 static int check_direct_read(struct btrfs_fs_info *fs_info,
4008 const struct iov_iter *iter, loff_t offset)
4013 ret = check_direct_IO(fs_info, iter, offset);
4017 if (!iter_is_iovec(iter))
4020 for (seg = 0; seg < iter->nr_segs; seg++)
4021 for (i = seg + 1; i < iter->nr_segs; i++)
4022 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
4027 static ssize_t btrfs_direct_read(struct kiocb *iocb, struct iov_iter *to)
4029 struct inode *inode = file_inode(iocb->ki_filp);
4030 size_t prev_left = 0;
4034 if (fsverity_active(inode))
4037 if (check_direct_read(btrfs_sb(inode->i_sb), to, iocb->ki_pos))
4040 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
4043 * This is similar to what we do for direct IO writes, see the comment
4044 * at btrfs_direct_write(), but we also disable page faults in addition
4045 * to disabling them only at the iov_iter level. This is because when
4046 * reading from a hole or prealloc extent, iomap calls iov_iter_zero(),
4047 * which can still trigger page fault ins despite having set ->nofault
4048 * to true of our 'to' iov_iter.
4050 * The difference to direct IO writes is that we deadlock when trying
4051 * to lock the extent range in the inode's tree during he page reads
4052 * triggered by the fault in (while for writes it is due to waiting for
4053 * our own ordered extent). This is because for direct IO reads,
4054 * btrfs_dio_iomap_begin() returns with the extent range locked, which
4055 * is only unlocked in the endio callback (end_bio_extent_readpage()).
4057 pagefault_disable();
4059 ret = btrfs_dio_read(iocb, to, read);
4060 to->nofault = false;
4063 /* No increment (+=) because iomap returns a cumulative value. */
4067 if (iov_iter_count(to) > 0 && (ret == -EFAULT || ret > 0)) {
4068 const size_t left = iov_iter_count(to);
4070 if (left == prev_left) {
4072 * We didn't make any progress since the last attempt,
4073 * fallback to a buffered read for the remainder of the
4074 * range. This is just to avoid any possibility of looping
4080 * We made some progress since the last retry or this is
4081 * the first time we are retrying. Fault in as many pages
4082 * as possible and retry.
4084 fault_in_iov_iter_writeable(to, left);
4089 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
4090 return ret < 0 ? ret : read;
4093 static ssize_t btrfs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
4097 if (iocb->ki_flags & IOCB_DIRECT) {
4098 ret = btrfs_direct_read(iocb, to);
4099 if (ret < 0 || !iov_iter_count(to) ||
4100 iocb->ki_pos >= i_size_read(file_inode(iocb->ki_filp)))
4104 return filemap_read(iocb, to, ret);
4107 const struct file_operations btrfs_file_operations = {
4108 .llseek = btrfs_file_llseek,
4109 .read_iter = btrfs_file_read_iter,
4110 .splice_read = generic_file_splice_read,
4111 .write_iter = btrfs_file_write_iter,
4112 .splice_write = iter_file_splice_write,
4113 .mmap = btrfs_file_mmap,
4114 .open = btrfs_file_open,
4115 .release = btrfs_release_file,
4116 .get_unmapped_area = thp_get_unmapped_area,
4117 .fsync = btrfs_sync_file,
4118 .fallocate = btrfs_fallocate,
4119 .unlocked_ioctl = btrfs_ioctl,
4120 #ifdef CONFIG_COMPAT
4121 .compat_ioctl = btrfs_compat_ioctl,
4123 .remap_file_range = btrfs_remap_file_range,
4126 void __cold btrfs_auto_defrag_exit(void)
4128 kmem_cache_destroy(btrfs_inode_defrag_cachep);
4131 int __init btrfs_auto_defrag_init(void)
4133 btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
4134 sizeof(struct inode_defrag), 0,
4137 if (!btrfs_inode_defrag_cachep)
4143 int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end)
4148 * So with compression we will find and lock a dirty page and clear the
4149 * first one as dirty, setup an async extent, and immediately return
4150 * with the entire range locked but with nobody actually marked with
4151 * writeback. So we can't just filemap_write_and_wait_range() and
4152 * expect it to work since it will just kick off a thread to do the
4153 * actual work. So we need to call filemap_fdatawrite_range _again_
4154 * since it will wait on the page lock, which won't be unlocked until
4155 * after the pages have been marked as writeback and so we're good to go
4156 * from there. We have to do this otherwise we'll miss the ordered
4157 * extents and that results in badness. Please Josef, do not think you
4158 * know better and pull this out at some point in the future, it is
4159 * right and you are wrong.
4161 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
4162 if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
4163 &BTRFS_I(inode)->runtime_flags))
4164 ret = filemap_fdatawrite_range(inode->i_mapping, start, end);