2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/aio.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/xattr.h>
38 #include <linux/posix_acl.h>
39 #include <linux/falloc.h>
40 #include <linux/slab.h>
41 #include <linux/ratelimit.h>
42 #include <linux/mount.h>
43 #include <linux/btrfs.h>
44 #include <linux/blkdev.h>
45 #include <linux/posix_acl_xattr.h>
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "print-tree.h"
51 #include "ordered-data.h"
55 #include "compression.h"
57 #include "free-space-cache.h"
58 #include "inode-map.h"
63 struct btrfs_iget_args {
64 struct btrfs_key *location;
65 struct btrfs_root *root;
68 static const struct inode_operations btrfs_dir_inode_operations;
69 static const struct inode_operations btrfs_symlink_inode_operations;
70 static const struct inode_operations btrfs_dir_ro_inode_operations;
71 static const struct inode_operations btrfs_special_inode_operations;
72 static const struct inode_operations btrfs_file_inode_operations;
73 static const struct address_space_operations btrfs_aops;
74 static const struct address_space_operations btrfs_symlink_aops;
75 static const struct file_operations btrfs_dir_file_operations;
76 static struct extent_io_ops btrfs_extent_io_ops;
78 static struct kmem_cache *btrfs_inode_cachep;
79 static struct kmem_cache *btrfs_delalloc_work_cachep;
80 struct kmem_cache *btrfs_trans_handle_cachep;
81 struct kmem_cache *btrfs_transaction_cachep;
82 struct kmem_cache *btrfs_path_cachep;
83 struct kmem_cache *btrfs_free_space_cachep;
86 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
87 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
88 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
89 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
90 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
91 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
92 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
93 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
96 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
97 static int btrfs_truncate(struct inode *inode);
98 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
99 static noinline int cow_file_range(struct inode *inode,
100 struct page *locked_page,
101 u64 start, u64 end, int *page_started,
102 unsigned long *nr_written, int unlock);
103 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
104 u64 len, u64 orig_start,
105 u64 block_start, u64 block_len,
106 u64 orig_block_len, u64 ram_bytes,
109 static int btrfs_dirty_inode(struct inode *inode);
111 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
112 struct inode *inode, struct inode *dir,
113 const struct qstr *qstr)
117 err = btrfs_init_acl(trans, inode, dir);
119 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
124 * this does all the hard work for inserting an inline extent into
125 * the btree. The caller should have done a btrfs_drop_extents so that
126 * no overlapping inline items exist in the btree
128 static int insert_inline_extent(struct btrfs_trans_handle *trans,
129 struct btrfs_path *path, int extent_inserted,
130 struct btrfs_root *root, struct inode *inode,
131 u64 start, size_t size, size_t compressed_size,
133 struct page **compressed_pages)
135 struct extent_buffer *leaf;
136 struct page *page = NULL;
139 struct btrfs_file_extent_item *ei;
142 size_t cur_size = size;
143 unsigned long offset;
145 if (compressed_size && compressed_pages)
146 cur_size = compressed_size;
148 inode_add_bytes(inode, size);
150 if (!extent_inserted) {
151 struct btrfs_key key;
154 key.objectid = btrfs_ino(inode);
156 key.type = BTRFS_EXTENT_DATA_KEY;
158 datasize = btrfs_file_extent_calc_inline_size(cur_size);
159 path->leave_spinning = 1;
160 ret = btrfs_insert_empty_item(trans, root, path, &key,
167 leaf = path->nodes[0];
168 ei = btrfs_item_ptr(leaf, path->slots[0],
169 struct btrfs_file_extent_item);
170 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
171 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
172 btrfs_set_file_extent_encryption(leaf, ei, 0);
173 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
174 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
175 ptr = btrfs_file_extent_inline_start(ei);
177 if (compress_type != BTRFS_COMPRESS_NONE) {
180 while (compressed_size > 0) {
181 cpage = compressed_pages[i];
182 cur_size = min_t(unsigned long, compressed_size,
185 kaddr = kmap_atomic(cpage);
186 write_extent_buffer(leaf, kaddr, ptr, cur_size);
187 kunmap_atomic(kaddr);
191 compressed_size -= cur_size;
193 btrfs_set_file_extent_compression(leaf, ei,
196 page = find_get_page(inode->i_mapping,
197 start >> PAGE_CACHE_SHIFT);
198 btrfs_set_file_extent_compression(leaf, ei, 0);
199 kaddr = kmap_atomic(page);
200 offset = start & (PAGE_CACHE_SIZE - 1);
201 write_extent_buffer(leaf, kaddr + offset, ptr, size);
202 kunmap_atomic(kaddr);
203 page_cache_release(page);
205 btrfs_mark_buffer_dirty(leaf);
206 btrfs_release_path(path);
209 * we're an inline extent, so nobody can
210 * extend the file past i_size without locking
211 * a page we already have locked.
213 * We must do any isize and inode updates
214 * before we unlock the pages. Otherwise we
215 * could end up racing with unlink.
217 BTRFS_I(inode)->disk_i_size = inode->i_size;
218 ret = btrfs_update_inode(trans, root, inode);
227 * conditionally insert an inline extent into the file. This
228 * does the checks required to make sure the data is small enough
229 * to fit as an inline extent.
231 static noinline int cow_file_range_inline(struct btrfs_root *root,
232 struct inode *inode, u64 start,
233 u64 end, size_t compressed_size,
235 struct page **compressed_pages)
237 struct btrfs_trans_handle *trans;
238 u64 isize = i_size_read(inode);
239 u64 actual_end = min(end + 1, isize);
240 u64 inline_len = actual_end - start;
241 u64 aligned_end = ALIGN(end, root->sectorsize);
242 u64 data_len = inline_len;
244 struct btrfs_path *path;
245 int extent_inserted = 0;
246 u32 extent_item_size;
249 data_len = compressed_size;
252 actual_end > PAGE_CACHE_SIZE ||
253 data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) ||
255 (actual_end & (root->sectorsize - 1)) == 0) ||
257 data_len > root->fs_info->max_inline) {
261 path = btrfs_alloc_path();
265 trans = btrfs_join_transaction(root);
267 btrfs_free_path(path);
268 return PTR_ERR(trans);
270 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
272 if (compressed_size && compressed_pages)
273 extent_item_size = btrfs_file_extent_calc_inline_size(
276 extent_item_size = btrfs_file_extent_calc_inline_size(
279 ret = __btrfs_drop_extents(trans, root, inode, path,
280 start, aligned_end, NULL,
281 1, 1, extent_item_size, &extent_inserted);
283 btrfs_abort_transaction(trans, root, ret);
287 if (isize > actual_end)
288 inline_len = min_t(u64, isize, actual_end);
289 ret = insert_inline_extent(trans, path, extent_inserted,
291 inline_len, compressed_size,
292 compress_type, compressed_pages);
293 if (ret && ret != -ENOSPC) {
294 btrfs_abort_transaction(trans, root, ret);
296 } else if (ret == -ENOSPC) {
301 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
302 btrfs_delalloc_release_metadata(inode, end + 1 - start);
303 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
305 btrfs_free_path(path);
306 btrfs_end_transaction(trans, root);
310 struct async_extent {
315 unsigned long nr_pages;
317 struct list_head list;
322 struct btrfs_root *root;
323 struct page *locked_page;
326 struct list_head extents;
327 struct btrfs_work work;
330 static noinline int add_async_extent(struct async_cow *cow,
331 u64 start, u64 ram_size,
334 unsigned long nr_pages,
337 struct async_extent *async_extent;
339 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
340 BUG_ON(!async_extent); /* -ENOMEM */
341 async_extent->start = start;
342 async_extent->ram_size = ram_size;
343 async_extent->compressed_size = compressed_size;
344 async_extent->pages = pages;
345 async_extent->nr_pages = nr_pages;
346 async_extent->compress_type = compress_type;
347 list_add_tail(&async_extent->list, &cow->extents);
351 static inline int inode_need_compress(struct inode *inode)
353 struct btrfs_root *root = BTRFS_I(inode)->root;
356 if (btrfs_test_opt(root, FORCE_COMPRESS))
358 /* bad compression ratios */
359 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
361 if (btrfs_test_opt(root, COMPRESS) ||
362 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
363 BTRFS_I(inode)->force_compress)
369 * we create compressed extents in two phases. The first
370 * phase compresses a range of pages that have already been
371 * locked (both pages and state bits are locked).
373 * This is done inside an ordered work queue, and the compression
374 * is spread across many cpus. The actual IO submission is step
375 * two, and the ordered work queue takes care of making sure that
376 * happens in the same order things were put onto the queue by
377 * writepages and friends.
379 * If this code finds it can't get good compression, it puts an
380 * entry onto the work queue to write the uncompressed bytes. This
381 * makes sure that both compressed inodes and uncompressed inodes
382 * are written in the same order that the flusher thread sent them
385 static noinline int compress_file_range(struct inode *inode,
386 struct page *locked_page,
388 struct async_cow *async_cow,
391 struct btrfs_root *root = BTRFS_I(inode)->root;
393 u64 blocksize = root->sectorsize;
395 u64 isize = i_size_read(inode);
397 struct page **pages = NULL;
398 unsigned long nr_pages;
399 unsigned long nr_pages_ret = 0;
400 unsigned long total_compressed = 0;
401 unsigned long total_in = 0;
402 unsigned long max_compressed = 128 * 1024;
403 unsigned long max_uncompressed = 128 * 1024;
406 int compress_type = root->fs_info->compress_type;
409 /* if this is a small write inside eof, kick off a defrag */
410 if ((end - start + 1) < 16 * 1024 &&
411 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
412 btrfs_add_inode_defrag(NULL, inode);
414 actual_end = min_t(u64, isize, end + 1);
417 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
418 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
421 * we don't want to send crud past the end of i_size through
422 * compression, that's just a waste of CPU time. So, if the
423 * end of the file is before the start of our current
424 * requested range of bytes, we bail out to the uncompressed
425 * cleanup code that can deal with all of this.
427 * It isn't really the fastest way to fix things, but this is a
428 * very uncommon corner.
430 if (actual_end <= start)
431 goto cleanup_and_bail_uncompressed;
433 total_compressed = actual_end - start;
436 * skip compression for a small file range(<=blocksize) that
437 * isn't an inline extent, since it dosen't save disk space at all.
439 if (total_compressed <= blocksize &&
440 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
441 goto cleanup_and_bail_uncompressed;
443 /* we want to make sure that amount of ram required to uncompress
444 * an extent is reasonable, so we limit the total size in ram
445 * of a compressed extent to 128k. This is a crucial number
446 * because it also controls how easily we can spread reads across
447 * cpus for decompression.
449 * We also want to make sure the amount of IO required to do
450 * a random read is reasonably small, so we limit the size of
451 * a compressed extent to 128k.
453 total_compressed = min(total_compressed, max_uncompressed);
454 num_bytes = ALIGN(end - start + 1, blocksize);
455 num_bytes = max(blocksize, num_bytes);
460 * we do compression for mount -o compress and when the
461 * inode has not been flagged as nocompress. This flag can
462 * change at any time if we discover bad compression ratios.
464 if (inode_need_compress(inode)) {
466 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
468 /* just bail out to the uncompressed code */
472 if (BTRFS_I(inode)->force_compress)
473 compress_type = BTRFS_I(inode)->force_compress;
476 * we need to call clear_page_dirty_for_io on each
477 * page in the range. Otherwise applications with the file
478 * mmap'd can wander in and change the page contents while
479 * we are compressing them.
481 * If the compression fails for any reason, we set the pages
482 * dirty again later on.
484 extent_range_clear_dirty_for_io(inode, start, end);
486 ret = btrfs_compress_pages(compress_type,
487 inode->i_mapping, start,
488 total_compressed, pages,
489 nr_pages, &nr_pages_ret,
495 unsigned long offset = total_compressed &
496 (PAGE_CACHE_SIZE - 1);
497 struct page *page = pages[nr_pages_ret - 1];
500 /* zero the tail end of the last page, we might be
501 * sending it down to disk
504 kaddr = kmap_atomic(page);
505 memset(kaddr + offset, 0,
506 PAGE_CACHE_SIZE - offset);
507 kunmap_atomic(kaddr);
514 /* lets try to make an inline extent */
515 if (ret || total_in < (actual_end - start)) {
516 /* we didn't compress the entire range, try
517 * to make an uncompressed inline extent.
519 ret = cow_file_range_inline(root, inode, start, end,
522 /* try making a compressed inline extent */
523 ret = cow_file_range_inline(root, inode, start, end,
525 compress_type, pages);
528 unsigned long clear_flags = EXTENT_DELALLOC |
530 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
533 * inline extent creation worked or returned error,
534 * we don't need to create any more async work items.
535 * Unlock and free up our temp pages.
537 extent_clear_unlock_delalloc(inode, start, end, NULL,
538 clear_flags, PAGE_UNLOCK |
548 * we aren't doing an inline extent round the compressed size
549 * up to a block size boundary so the allocator does sane
552 total_compressed = ALIGN(total_compressed, blocksize);
555 * one last check to make sure the compression is really a
556 * win, compare the page count read with the blocks on disk
558 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
559 if (total_compressed >= total_in) {
562 num_bytes = total_in;
565 if (!will_compress && pages) {
567 * the compression code ran but failed to make things smaller,
568 * free any pages it allocated and our page pointer array
570 for (i = 0; i < nr_pages_ret; i++) {
571 WARN_ON(pages[i]->mapping);
572 page_cache_release(pages[i]);
576 total_compressed = 0;
579 /* flag the file so we don't compress in the future */
580 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
581 !(BTRFS_I(inode)->force_compress)) {
582 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
588 /* the async work queues will take care of doing actual
589 * allocation on disk for these compressed pages,
590 * and will submit them to the elevator.
592 add_async_extent(async_cow, start, num_bytes,
593 total_compressed, pages, nr_pages_ret,
596 if (start + num_bytes < end) {
603 cleanup_and_bail_uncompressed:
605 * No compression, but we still need to write the pages in
606 * the file we've been given so far. redirty the locked
607 * page if it corresponds to our extent and set things up
608 * for the async work queue to run cow_file_range to do
609 * the normal delalloc dance
611 if (page_offset(locked_page) >= start &&
612 page_offset(locked_page) <= end) {
613 __set_page_dirty_nobuffers(locked_page);
614 /* unlocked later on in the async handlers */
617 extent_range_redirty_for_io(inode, start, end);
618 add_async_extent(async_cow, start, end - start + 1,
619 0, NULL, 0, BTRFS_COMPRESS_NONE);
627 for (i = 0; i < nr_pages_ret; i++) {
628 WARN_ON(pages[i]->mapping);
629 page_cache_release(pages[i]);
636 static void free_async_extent_pages(struct async_extent *async_extent)
640 if (!async_extent->pages)
643 for (i = 0; i < async_extent->nr_pages; i++) {
644 WARN_ON(async_extent->pages[i]->mapping);
645 page_cache_release(async_extent->pages[i]);
647 kfree(async_extent->pages);
648 async_extent->nr_pages = 0;
649 async_extent->pages = NULL;
653 * phase two of compressed writeback. This is the ordered portion
654 * of the code, which only gets called in the order the work was
655 * queued. We walk all the async extents created by compress_file_range
656 * and send them down to the disk.
658 static noinline void submit_compressed_extents(struct inode *inode,
659 struct async_cow *async_cow)
661 struct async_extent *async_extent;
663 struct btrfs_key ins;
664 struct extent_map *em;
665 struct btrfs_root *root = BTRFS_I(inode)->root;
666 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
667 struct extent_io_tree *io_tree;
671 while (!list_empty(&async_cow->extents)) {
672 async_extent = list_entry(async_cow->extents.next,
673 struct async_extent, list);
674 list_del(&async_extent->list);
676 io_tree = &BTRFS_I(inode)->io_tree;
679 /* did the compression code fall back to uncompressed IO? */
680 if (!async_extent->pages) {
681 int page_started = 0;
682 unsigned long nr_written = 0;
684 lock_extent(io_tree, async_extent->start,
685 async_extent->start +
686 async_extent->ram_size - 1);
688 /* allocate blocks */
689 ret = cow_file_range(inode, async_cow->locked_page,
691 async_extent->start +
692 async_extent->ram_size - 1,
693 &page_started, &nr_written, 0);
698 * if page_started, cow_file_range inserted an
699 * inline extent and took care of all the unlocking
700 * and IO for us. Otherwise, we need to submit
701 * all those pages down to the drive.
703 if (!page_started && !ret)
704 extent_write_locked_range(io_tree,
705 inode, async_extent->start,
706 async_extent->start +
707 async_extent->ram_size - 1,
711 unlock_page(async_cow->locked_page);
717 lock_extent(io_tree, async_extent->start,
718 async_extent->start + async_extent->ram_size - 1);
720 ret = btrfs_reserve_extent(root,
721 async_extent->compressed_size,
722 async_extent->compressed_size,
723 0, alloc_hint, &ins, 1, 1);
725 free_async_extent_pages(async_extent);
727 if (ret == -ENOSPC) {
728 unlock_extent(io_tree, async_extent->start,
729 async_extent->start +
730 async_extent->ram_size - 1);
733 * we need to redirty the pages if we decide to
734 * fallback to uncompressed IO, otherwise we
735 * will not submit these pages down to lower
738 extent_range_redirty_for_io(inode,
740 async_extent->start +
741 async_extent->ram_size - 1);
749 * here we're doing allocation and writeback of the
752 btrfs_drop_extent_cache(inode, async_extent->start,
753 async_extent->start +
754 async_extent->ram_size - 1, 0);
756 em = alloc_extent_map();
759 goto out_free_reserve;
761 em->start = async_extent->start;
762 em->len = async_extent->ram_size;
763 em->orig_start = em->start;
764 em->mod_start = em->start;
765 em->mod_len = em->len;
767 em->block_start = ins.objectid;
768 em->block_len = ins.offset;
769 em->orig_block_len = ins.offset;
770 em->ram_bytes = async_extent->ram_size;
771 em->bdev = root->fs_info->fs_devices->latest_bdev;
772 em->compress_type = async_extent->compress_type;
773 set_bit(EXTENT_FLAG_PINNED, &em->flags);
774 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
778 write_lock(&em_tree->lock);
779 ret = add_extent_mapping(em_tree, em, 1);
780 write_unlock(&em_tree->lock);
781 if (ret != -EEXIST) {
785 btrfs_drop_extent_cache(inode, async_extent->start,
786 async_extent->start +
787 async_extent->ram_size - 1, 0);
791 goto out_free_reserve;
793 ret = btrfs_add_ordered_extent_compress(inode,
796 async_extent->ram_size,
798 BTRFS_ORDERED_COMPRESSED,
799 async_extent->compress_type);
801 btrfs_drop_extent_cache(inode, async_extent->start,
802 async_extent->start +
803 async_extent->ram_size - 1, 0);
804 goto out_free_reserve;
808 * clear dirty, set writeback and unlock the pages.
810 extent_clear_unlock_delalloc(inode, async_extent->start,
811 async_extent->start +
812 async_extent->ram_size - 1,
813 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
814 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
816 ret = btrfs_submit_compressed_write(inode,
818 async_extent->ram_size,
820 ins.offset, async_extent->pages,
821 async_extent->nr_pages);
823 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
824 struct page *p = async_extent->pages[0];
825 const u64 start = async_extent->start;
826 const u64 end = start + async_extent->ram_size - 1;
828 p->mapping = inode->i_mapping;
829 tree->ops->writepage_end_io_hook(p, start, end,
832 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
835 free_async_extent_pages(async_extent);
837 alloc_hint = ins.objectid + ins.offset;
843 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
845 extent_clear_unlock_delalloc(inode, async_extent->start,
846 async_extent->start +
847 async_extent->ram_size - 1,
848 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
849 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
850 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
851 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
853 free_async_extent_pages(async_extent);
858 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
861 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
862 struct extent_map *em;
865 read_lock(&em_tree->lock);
866 em = search_extent_mapping(em_tree, start, num_bytes);
869 * if block start isn't an actual block number then find the
870 * first block in this inode and use that as a hint. If that
871 * block is also bogus then just don't worry about it.
873 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
875 em = search_extent_mapping(em_tree, 0, 0);
876 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
877 alloc_hint = em->block_start;
881 alloc_hint = em->block_start;
885 read_unlock(&em_tree->lock);
891 * when extent_io.c finds a delayed allocation range in the file,
892 * the call backs end up in this code. The basic idea is to
893 * allocate extents on disk for the range, and create ordered data structs
894 * in ram to track those extents.
896 * locked_page is the page that writepage had locked already. We use
897 * it to make sure we don't do extra locks or unlocks.
899 * *page_started is set to one if we unlock locked_page and do everything
900 * required to start IO on it. It may be clean and already done with
903 static noinline int cow_file_range(struct inode *inode,
904 struct page *locked_page,
905 u64 start, u64 end, int *page_started,
906 unsigned long *nr_written,
909 struct btrfs_root *root = BTRFS_I(inode)->root;
912 unsigned long ram_size;
915 u64 blocksize = root->sectorsize;
916 struct btrfs_key ins;
917 struct extent_map *em;
918 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
921 if (btrfs_is_free_space_inode(inode)) {
927 num_bytes = ALIGN(end - start + 1, blocksize);
928 num_bytes = max(blocksize, num_bytes);
929 disk_num_bytes = num_bytes;
931 /* if this is a small write inside eof, kick off defrag */
932 if (num_bytes < 64 * 1024 &&
933 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
934 btrfs_add_inode_defrag(NULL, inode);
937 /* lets try to make an inline extent */
938 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
941 extent_clear_unlock_delalloc(inode, start, end, NULL,
942 EXTENT_LOCKED | EXTENT_DELALLOC |
943 EXTENT_DEFRAG, PAGE_UNLOCK |
944 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
947 *nr_written = *nr_written +
948 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
951 } else if (ret < 0) {
956 BUG_ON(disk_num_bytes >
957 btrfs_super_total_bytes(root->fs_info->super_copy));
959 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
960 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
962 while (disk_num_bytes > 0) {
965 cur_alloc_size = disk_num_bytes;
966 ret = btrfs_reserve_extent(root, cur_alloc_size,
967 root->sectorsize, 0, alloc_hint,
972 em = alloc_extent_map();
978 em->orig_start = em->start;
979 ram_size = ins.offset;
980 em->len = ins.offset;
981 em->mod_start = em->start;
982 em->mod_len = em->len;
984 em->block_start = ins.objectid;
985 em->block_len = ins.offset;
986 em->orig_block_len = ins.offset;
987 em->ram_bytes = ram_size;
988 em->bdev = root->fs_info->fs_devices->latest_bdev;
989 set_bit(EXTENT_FLAG_PINNED, &em->flags);
993 write_lock(&em_tree->lock);
994 ret = add_extent_mapping(em_tree, em, 1);
995 write_unlock(&em_tree->lock);
996 if (ret != -EEXIST) {
1000 btrfs_drop_extent_cache(inode, start,
1001 start + ram_size - 1, 0);
1006 cur_alloc_size = ins.offset;
1007 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1008 ram_size, cur_alloc_size, 0);
1010 goto out_drop_extent_cache;
1012 if (root->root_key.objectid ==
1013 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1014 ret = btrfs_reloc_clone_csums(inode, start,
1017 goto out_drop_extent_cache;
1020 if (disk_num_bytes < cur_alloc_size)
1023 /* we're not doing compressed IO, don't unlock the first
1024 * page (which the caller expects to stay locked), don't
1025 * clear any dirty bits and don't set any writeback bits
1027 * Do set the Private2 bit so we know this page was properly
1028 * setup for writepage
1030 op = unlock ? PAGE_UNLOCK : 0;
1031 op |= PAGE_SET_PRIVATE2;
1033 extent_clear_unlock_delalloc(inode, start,
1034 start + ram_size - 1, locked_page,
1035 EXTENT_LOCKED | EXTENT_DELALLOC,
1037 disk_num_bytes -= cur_alloc_size;
1038 num_bytes -= cur_alloc_size;
1039 alloc_hint = ins.objectid + ins.offset;
1040 start += cur_alloc_size;
1045 out_drop_extent_cache:
1046 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1048 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1050 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1051 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1052 EXTENT_DELALLOC | EXTENT_DEFRAG,
1053 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1054 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1059 * work queue call back to started compression on a file and pages
1061 static noinline void async_cow_start(struct btrfs_work *work)
1063 struct async_cow *async_cow;
1065 async_cow = container_of(work, struct async_cow, work);
1067 compress_file_range(async_cow->inode, async_cow->locked_page,
1068 async_cow->start, async_cow->end, async_cow,
1070 if (num_added == 0) {
1071 btrfs_add_delayed_iput(async_cow->inode);
1072 async_cow->inode = NULL;
1077 * work queue call back to submit previously compressed pages
1079 static noinline void async_cow_submit(struct btrfs_work *work)
1081 struct async_cow *async_cow;
1082 struct btrfs_root *root;
1083 unsigned long nr_pages;
1085 async_cow = container_of(work, struct async_cow, work);
1087 root = async_cow->root;
1088 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1091 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1093 waitqueue_active(&root->fs_info->async_submit_wait))
1094 wake_up(&root->fs_info->async_submit_wait);
1096 if (async_cow->inode)
1097 submit_compressed_extents(async_cow->inode, async_cow);
1100 static noinline void async_cow_free(struct btrfs_work *work)
1102 struct async_cow *async_cow;
1103 async_cow = container_of(work, struct async_cow, work);
1104 if (async_cow->inode)
1105 btrfs_add_delayed_iput(async_cow->inode);
1109 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1110 u64 start, u64 end, int *page_started,
1111 unsigned long *nr_written)
1113 struct async_cow *async_cow;
1114 struct btrfs_root *root = BTRFS_I(inode)->root;
1115 unsigned long nr_pages;
1117 int limit = 10 * 1024 * 1024;
1119 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1120 1, 0, NULL, GFP_NOFS);
1121 while (start < end) {
1122 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1123 BUG_ON(!async_cow); /* -ENOMEM */
1124 async_cow->inode = igrab(inode);
1125 async_cow->root = root;
1126 async_cow->locked_page = locked_page;
1127 async_cow->start = start;
1129 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1130 !btrfs_test_opt(root, FORCE_COMPRESS))
1133 cur_end = min(end, start + 512 * 1024 - 1);
1135 async_cow->end = cur_end;
1136 INIT_LIST_HEAD(&async_cow->extents);
1138 btrfs_init_work(&async_cow->work,
1139 btrfs_delalloc_helper,
1140 async_cow_start, async_cow_submit,
1143 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1145 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1147 btrfs_queue_work(root->fs_info->delalloc_workers,
1150 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1151 wait_event(root->fs_info->async_submit_wait,
1152 (atomic_read(&root->fs_info->async_delalloc_pages) <
1156 while (atomic_read(&root->fs_info->async_submit_draining) &&
1157 atomic_read(&root->fs_info->async_delalloc_pages)) {
1158 wait_event(root->fs_info->async_submit_wait,
1159 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1163 *nr_written += nr_pages;
1164 start = cur_end + 1;
1170 static noinline int csum_exist_in_range(struct btrfs_root *root,
1171 u64 bytenr, u64 num_bytes)
1174 struct btrfs_ordered_sum *sums;
1177 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1178 bytenr + num_bytes - 1, &list, 0);
1179 if (ret == 0 && list_empty(&list))
1182 while (!list_empty(&list)) {
1183 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1184 list_del(&sums->list);
1191 * when nowcow writeback call back. This checks for snapshots or COW copies
1192 * of the extents that exist in the file, and COWs the file as required.
1194 * If no cow copies or snapshots exist, we write directly to the existing
1197 static noinline int run_delalloc_nocow(struct inode *inode,
1198 struct page *locked_page,
1199 u64 start, u64 end, int *page_started, int force,
1200 unsigned long *nr_written)
1202 struct btrfs_root *root = BTRFS_I(inode)->root;
1203 struct btrfs_trans_handle *trans;
1204 struct extent_buffer *leaf;
1205 struct btrfs_path *path;
1206 struct btrfs_file_extent_item *fi;
1207 struct btrfs_key found_key;
1222 u64 ino = btrfs_ino(inode);
1224 path = btrfs_alloc_path();
1226 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1227 EXTENT_LOCKED | EXTENT_DELALLOC |
1228 EXTENT_DO_ACCOUNTING |
1229 EXTENT_DEFRAG, PAGE_UNLOCK |
1231 PAGE_SET_WRITEBACK |
1232 PAGE_END_WRITEBACK);
1236 nolock = btrfs_is_free_space_inode(inode);
1239 trans = btrfs_join_transaction_nolock(root);
1241 trans = btrfs_join_transaction(root);
1243 if (IS_ERR(trans)) {
1244 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1245 EXTENT_LOCKED | EXTENT_DELALLOC |
1246 EXTENT_DO_ACCOUNTING |
1247 EXTENT_DEFRAG, PAGE_UNLOCK |
1249 PAGE_SET_WRITEBACK |
1250 PAGE_END_WRITEBACK);
1251 btrfs_free_path(path);
1252 return PTR_ERR(trans);
1255 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1257 cow_start = (u64)-1;
1260 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1264 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1265 leaf = path->nodes[0];
1266 btrfs_item_key_to_cpu(leaf, &found_key,
1267 path->slots[0] - 1);
1268 if (found_key.objectid == ino &&
1269 found_key.type == BTRFS_EXTENT_DATA_KEY)
1274 leaf = path->nodes[0];
1275 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1276 ret = btrfs_next_leaf(root, path);
1281 leaf = path->nodes[0];
1287 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1289 if (found_key.objectid > ino ||
1290 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1291 found_key.offset > end)
1294 if (found_key.offset > cur_offset) {
1295 extent_end = found_key.offset;
1300 fi = btrfs_item_ptr(leaf, path->slots[0],
1301 struct btrfs_file_extent_item);
1302 extent_type = btrfs_file_extent_type(leaf, fi);
1304 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1305 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1306 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1307 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1308 extent_offset = btrfs_file_extent_offset(leaf, fi);
1309 extent_end = found_key.offset +
1310 btrfs_file_extent_num_bytes(leaf, fi);
1312 btrfs_file_extent_disk_num_bytes(leaf, fi);
1313 if (extent_end <= start) {
1317 if (disk_bytenr == 0)
1319 if (btrfs_file_extent_compression(leaf, fi) ||
1320 btrfs_file_extent_encryption(leaf, fi) ||
1321 btrfs_file_extent_other_encoding(leaf, fi))
1323 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1325 if (btrfs_extent_readonly(root, disk_bytenr))
1327 if (btrfs_cross_ref_exist(trans, root, ino,
1329 extent_offset, disk_bytenr))
1331 disk_bytenr += extent_offset;
1332 disk_bytenr += cur_offset - found_key.offset;
1333 num_bytes = min(end + 1, extent_end) - cur_offset;
1335 * if there are pending snapshots for this root,
1336 * we fall into common COW way.
1339 err = btrfs_start_nocow_write(root);
1344 * force cow if csum exists in the range.
1345 * this ensure that csum for a given extent are
1346 * either valid or do not exist.
1348 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1351 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1352 extent_end = found_key.offset +
1353 btrfs_file_extent_inline_len(leaf,
1354 path->slots[0], fi);
1355 extent_end = ALIGN(extent_end, root->sectorsize);
1360 if (extent_end <= start) {
1362 if (!nolock && nocow)
1363 btrfs_end_nocow_write(root);
1367 if (cow_start == (u64)-1)
1368 cow_start = cur_offset;
1369 cur_offset = extent_end;
1370 if (cur_offset > end)
1376 btrfs_release_path(path);
1377 if (cow_start != (u64)-1) {
1378 ret = cow_file_range(inode, locked_page,
1379 cow_start, found_key.offset - 1,
1380 page_started, nr_written, 1);
1382 if (!nolock && nocow)
1383 btrfs_end_nocow_write(root);
1386 cow_start = (u64)-1;
1389 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1390 struct extent_map *em;
1391 struct extent_map_tree *em_tree;
1392 em_tree = &BTRFS_I(inode)->extent_tree;
1393 em = alloc_extent_map();
1394 BUG_ON(!em); /* -ENOMEM */
1395 em->start = cur_offset;
1396 em->orig_start = found_key.offset - extent_offset;
1397 em->len = num_bytes;
1398 em->block_len = num_bytes;
1399 em->block_start = disk_bytenr;
1400 em->orig_block_len = disk_num_bytes;
1401 em->ram_bytes = ram_bytes;
1402 em->bdev = root->fs_info->fs_devices->latest_bdev;
1403 em->mod_start = em->start;
1404 em->mod_len = em->len;
1405 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1406 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1407 em->generation = -1;
1409 write_lock(&em_tree->lock);
1410 ret = add_extent_mapping(em_tree, em, 1);
1411 write_unlock(&em_tree->lock);
1412 if (ret != -EEXIST) {
1413 free_extent_map(em);
1416 btrfs_drop_extent_cache(inode, em->start,
1417 em->start + em->len - 1, 0);
1419 type = BTRFS_ORDERED_PREALLOC;
1421 type = BTRFS_ORDERED_NOCOW;
1424 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1425 num_bytes, num_bytes, type);
1426 BUG_ON(ret); /* -ENOMEM */
1428 if (root->root_key.objectid ==
1429 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1430 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1433 if (!nolock && nocow)
1434 btrfs_end_nocow_write(root);
1439 extent_clear_unlock_delalloc(inode, cur_offset,
1440 cur_offset + num_bytes - 1,
1441 locked_page, EXTENT_LOCKED |
1442 EXTENT_DELALLOC, PAGE_UNLOCK |
1444 if (!nolock && nocow)
1445 btrfs_end_nocow_write(root);
1446 cur_offset = extent_end;
1447 if (cur_offset > end)
1450 btrfs_release_path(path);
1452 if (cur_offset <= end && cow_start == (u64)-1) {
1453 cow_start = cur_offset;
1457 if (cow_start != (u64)-1) {
1458 ret = cow_file_range(inode, locked_page, cow_start, end,
1459 page_started, nr_written, 1);
1465 err = btrfs_end_transaction(trans, root);
1469 if (ret && cur_offset < end)
1470 extent_clear_unlock_delalloc(inode, cur_offset, end,
1471 locked_page, EXTENT_LOCKED |
1472 EXTENT_DELALLOC | EXTENT_DEFRAG |
1473 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1475 PAGE_SET_WRITEBACK |
1476 PAGE_END_WRITEBACK);
1477 btrfs_free_path(path);
1481 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1484 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1485 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1489 * @defrag_bytes is a hint value, no spinlock held here,
1490 * if is not zero, it means the file is defragging.
1491 * Force cow if given extent needs to be defragged.
1493 if (BTRFS_I(inode)->defrag_bytes &&
1494 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1495 EXTENT_DEFRAG, 0, NULL))
1502 * extent_io.c call back to do delayed allocation processing
1504 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1505 u64 start, u64 end, int *page_started,
1506 unsigned long *nr_written)
1509 int force_cow = need_force_cow(inode, start, end);
1511 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1512 ret = run_delalloc_nocow(inode, locked_page, start, end,
1513 page_started, 1, nr_written);
1514 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1515 ret = run_delalloc_nocow(inode, locked_page, start, end,
1516 page_started, 0, nr_written);
1517 } else if (!inode_need_compress(inode)) {
1518 ret = cow_file_range(inode, locked_page, start, end,
1519 page_started, nr_written, 1);
1521 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1522 &BTRFS_I(inode)->runtime_flags);
1523 ret = cow_file_range_async(inode, locked_page, start, end,
1524 page_started, nr_written);
1529 static void btrfs_split_extent_hook(struct inode *inode,
1530 struct extent_state *orig, u64 split)
1532 /* not delalloc, ignore it */
1533 if (!(orig->state & EXTENT_DELALLOC))
1536 spin_lock(&BTRFS_I(inode)->lock);
1537 BTRFS_I(inode)->outstanding_extents++;
1538 spin_unlock(&BTRFS_I(inode)->lock);
1542 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1543 * extents so we can keep track of new extents that are just merged onto old
1544 * extents, such as when we are doing sequential writes, so we can properly
1545 * account for the metadata space we'll need.
1547 static void btrfs_merge_extent_hook(struct inode *inode,
1548 struct extent_state *new,
1549 struct extent_state *other)
1551 /* not delalloc, ignore it */
1552 if (!(other->state & EXTENT_DELALLOC))
1555 spin_lock(&BTRFS_I(inode)->lock);
1556 BTRFS_I(inode)->outstanding_extents--;
1557 spin_unlock(&BTRFS_I(inode)->lock);
1560 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1561 struct inode *inode)
1563 spin_lock(&root->delalloc_lock);
1564 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1565 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1566 &root->delalloc_inodes);
1567 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1568 &BTRFS_I(inode)->runtime_flags);
1569 root->nr_delalloc_inodes++;
1570 if (root->nr_delalloc_inodes == 1) {
1571 spin_lock(&root->fs_info->delalloc_root_lock);
1572 BUG_ON(!list_empty(&root->delalloc_root));
1573 list_add_tail(&root->delalloc_root,
1574 &root->fs_info->delalloc_roots);
1575 spin_unlock(&root->fs_info->delalloc_root_lock);
1578 spin_unlock(&root->delalloc_lock);
1581 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1582 struct inode *inode)
1584 spin_lock(&root->delalloc_lock);
1585 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1586 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1587 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1588 &BTRFS_I(inode)->runtime_flags);
1589 root->nr_delalloc_inodes--;
1590 if (!root->nr_delalloc_inodes) {
1591 spin_lock(&root->fs_info->delalloc_root_lock);
1592 BUG_ON(list_empty(&root->delalloc_root));
1593 list_del_init(&root->delalloc_root);
1594 spin_unlock(&root->fs_info->delalloc_root_lock);
1597 spin_unlock(&root->delalloc_lock);
1601 * extent_io.c set_bit_hook, used to track delayed allocation
1602 * bytes in this file, and to maintain the list of inodes that
1603 * have pending delalloc work to be done.
1605 static void btrfs_set_bit_hook(struct inode *inode,
1606 struct extent_state *state, unsigned long *bits)
1609 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1612 * set_bit and clear bit hooks normally require _irqsave/restore
1613 * but in this case, we are only testing for the DELALLOC
1614 * bit, which is only set or cleared with irqs on
1616 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1617 struct btrfs_root *root = BTRFS_I(inode)->root;
1618 u64 len = state->end + 1 - state->start;
1619 bool do_list = !btrfs_is_free_space_inode(inode);
1621 if (*bits & EXTENT_FIRST_DELALLOC) {
1622 *bits &= ~EXTENT_FIRST_DELALLOC;
1624 spin_lock(&BTRFS_I(inode)->lock);
1625 BTRFS_I(inode)->outstanding_extents++;
1626 spin_unlock(&BTRFS_I(inode)->lock);
1629 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1630 root->fs_info->delalloc_batch);
1631 spin_lock(&BTRFS_I(inode)->lock);
1632 BTRFS_I(inode)->delalloc_bytes += len;
1633 if (*bits & EXTENT_DEFRAG)
1634 BTRFS_I(inode)->defrag_bytes += len;
1635 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1636 &BTRFS_I(inode)->runtime_flags))
1637 btrfs_add_delalloc_inodes(root, inode);
1638 spin_unlock(&BTRFS_I(inode)->lock);
1643 * extent_io.c clear_bit_hook, see set_bit_hook for why
1645 static void btrfs_clear_bit_hook(struct inode *inode,
1646 struct extent_state *state,
1647 unsigned long *bits)
1649 u64 len = state->end + 1 - state->start;
1651 spin_lock(&BTRFS_I(inode)->lock);
1652 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1653 BTRFS_I(inode)->defrag_bytes -= len;
1654 spin_unlock(&BTRFS_I(inode)->lock);
1657 * set_bit and clear bit hooks normally require _irqsave/restore
1658 * but in this case, we are only testing for the DELALLOC
1659 * bit, which is only set or cleared with irqs on
1661 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1662 struct btrfs_root *root = BTRFS_I(inode)->root;
1663 bool do_list = !btrfs_is_free_space_inode(inode);
1665 if (*bits & EXTENT_FIRST_DELALLOC) {
1666 *bits &= ~EXTENT_FIRST_DELALLOC;
1667 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1668 spin_lock(&BTRFS_I(inode)->lock);
1669 BTRFS_I(inode)->outstanding_extents--;
1670 spin_unlock(&BTRFS_I(inode)->lock);
1674 * We don't reserve metadata space for space cache inodes so we
1675 * don't need to call dellalloc_release_metadata if there is an
1678 if (*bits & EXTENT_DO_ACCOUNTING &&
1679 root != root->fs_info->tree_root)
1680 btrfs_delalloc_release_metadata(inode, len);
1682 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1683 && do_list && !(state->state & EXTENT_NORESERVE))
1684 btrfs_free_reserved_data_space(inode, len);
1686 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1687 root->fs_info->delalloc_batch);
1688 spin_lock(&BTRFS_I(inode)->lock);
1689 BTRFS_I(inode)->delalloc_bytes -= len;
1690 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1691 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1692 &BTRFS_I(inode)->runtime_flags))
1693 btrfs_del_delalloc_inode(root, inode);
1694 spin_unlock(&BTRFS_I(inode)->lock);
1699 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1700 * we don't create bios that span stripes or chunks
1702 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1703 size_t size, struct bio *bio,
1704 unsigned long bio_flags)
1706 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1707 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1712 if (bio_flags & EXTENT_BIO_COMPRESSED)
1715 length = bio->bi_iter.bi_size;
1716 map_length = length;
1717 ret = btrfs_map_block(root->fs_info, rw, logical,
1718 &map_length, NULL, 0);
1719 /* Will always return 0 with map_multi == NULL */
1721 if (map_length < length + size)
1727 * in order to insert checksums into the metadata in large chunks,
1728 * we wait until bio submission time. All the pages in the bio are
1729 * checksummed and sums are attached onto the ordered extent record.
1731 * At IO completion time the cums attached on the ordered extent record
1732 * are inserted into the btree
1734 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1735 struct bio *bio, int mirror_num,
1736 unsigned long bio_flags,
1739 struct btrfs_root *root = BTRFS_I(inode)->root;
1742 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1743 BUG_ON(ret); /* -ENOMEM */
1748 * in order to insert checksums into the metadata in large chunks,
1749 * we wait until bio submission time. All the pages in the bio are
1750 * checksummed and sums are attached onto the ordered extent record.
1752 * At IO completion time the cums attached on the ordered extent record
1753 * are inserted into the btree
1755 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1756 int mirror_num, unsigned long bio_flags,
1759 struct btrfs_root *root = BTRFS_I(inode)->root;
1762 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1764 bio_endio(bio, ret);
1769 * extent_io.c submission hook. This does the right thing for csum calculation
1770 * on write, or reading the csums from the tree before a read
1772 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1773 int mirror_num, unsigned long bio_flags,
1776 struct btrfs_root *root = BTRFS_I(inode)->root;
1780 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1782 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1784 if (btrfs_is_free_space_inode(inode))
1787 if (!(rw & REQ_WRITE)) {
1788 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1792 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1793 ret = btrfs_submit_compressed_read(inode, bio,
1797 } else if (!skip_sum) {
1798 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1803 } else if (async && !skip_sum) {
1804 /* csum items have already been cloned */
1805 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1807 /* we're doing a write, do the async checksumming */
1808 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1809 inode, rw, bio, mirror_num,
1810 bio_flags, bio_offset,
1811 __btrfs_submit_bio_start,
1812 __btrfs_submit_bio_done);
1814 } else if (!skip_sum) {
1815 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1821 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1825 bio_endio(bio, ret);
1830 * given a list of ordered sums record them in the inode. This happens
1831 * at IO completion time based on sums calculated at bio submission time.
1833 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1834 struct inode *inode, u64 file_offset,
1835 struct list_head *list)
1837 struct btrfs_ordered_sum *sum;
1839 list_for_each_entry(sum, list, list) {
1840 trans->adding_csums = 1;
1841 btrfs_csum_file_blocks(trans,
1842 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1843 trans->adding_csums = 0;
1848 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1849 struct extent_state **cached_state)
1851 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1852 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1853 cached_state, GFP_NOFS);
1856 /* see btrfs_writepage_start_hook for details on why this is required */
1857 struct btrfs_writepage_fixup {
1859 struct btrfs_work work;
1862 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1864 struct btrfs_writepage_fixup *fixup;
1865 struct btrfs_ordered_extent *ordered;
1866 struct extent_state *cached_state = NULL;
1868 struct inode *inode;
1873 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1877 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1878 ClearPageChecked(page);
1882 inode = page->mapping->host;
1883 page_start = page_offset(page);
1884 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1886 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1889 /* already ordered? We're done */
1890 if (PagePrivate2(page))
1893 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1895 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1896 page_end, &cached_state, GFP_NOFS);
1898 btrfs_start_ordered_extent(inode, ordered, 1);
1899 btrfs_put_ordered_extent(ordered);
1903 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1905 mapping_set_error(page->mapping, ret);
1906 end_extent_writepage(page, ret, page_start, page_end);
1907 ClearPageChecked(page);
1911 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1912 ClearPageChecked(page);
1913 set_page_dirty(page);
1915 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1916 &cached_state, GFP_NOFS);
1919 page_cache_release(page);
1924 * There are a few paths in the higher layers of the kernel that directly
1925 * set the page dirty bit without asking the filesystem if it is a
1926 * good idea. This causes problems because we want to make sure COW
1927 * properly happens and the data=ordered rules are followed.
1929 * In our case any range that doesn't have the ORDERED bit set
1930 * hasn't been properly setup for IO. We kick off an async process
1931 * to fix it up. The async helper will wait for ordered extents, set
1932 * the delalloc bit and make it safe to write the page.
1934 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1936 struct inode *inode = page->mapping->host;
1937 struct btrfs_writepage_fixup *fixup;
1938 struct btrfs_root *root = BTRFS_I(inode)->root;
1940 /* this page is properly in the ordered list */
1941 if (TestClearPagePrivate2(page))
1944 if (PageChecked(page))
1947 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1951 SetPageChecked(page);
1952 page_cache_get(page);
1953 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
1954 btrfs_writepage_fixup_worker, NULL, NULL);
1956 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
1960 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1961 struct inode *inode, u64 file_pos,
1962 u64 disk_bytenr, u64 disk_num_bytes,
1963 u64 num_bytes, u64 ram_bytes,
1964 u8 compression, u8 encryption,
1965 u16 other_encoding, int extent_type)
1967 struct btrfs_root *root = BTRFS_I(inode)->root;
1968 struct btrfs_file_extent_item *fi;
1969 struct btrfs_path *path;
1970 struct extent_buffer *leaf;
1971 struct btrfs_key ins;
1972 int extent_inserted = 0;
1975 path = btrfs_alloc_path();
1980 * we may be replacing one extent in the tree with another.
1981 * The new extent is pinned in the extent map, and we don't want
1982 * to drop it from the cache until it is completely in the btree.
1984 * So, tell btrfs_drop_extents to leave this extent in the cache.
1985 * the caller is expected to unpin it and allow it to be merged
1988 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
1989 file_pos + num_bytes, NULL, 0,
1990 1, sizeof(*fi), &extent_inserted);
1994 if (!extent_inserted) {
1995 ins.objectid = btrfs_ino(inode);
1996 ins.offset = file_pos;
1997 ins.type = BTRFS_EXTENT_DATA_KEY;
1999 path->leave_spinning = 1;
2000 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2005 leaf = path->nodes[0];
2006 fi = btrfs_item_ptr(leaf, path->slots[0],
2007 struct btrfs_file_extent_item);
2008 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2009 btrfs_set_file_extent_type(leaf, fi, extent_type);
2010 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2011 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2012 btrfs_set_file_extent_offset(leaf, fi, 0);
2013 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2014 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2015 btrfs_set_file_extent_compression(leaf, fi, compression);
2016 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2017 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2019 btrfs_mark_buffer_dirty(leaf);
2020 btrfs_release_path(path);
2022 inode_add_bytes(inode, num_bytes);
2024 ins.objectid = disk_bytenr;
2025 ins.offset = disk_num_bytes;
2026 ins.type = BTRFS_EXTENT_ITEM_KEY;
2027 ret = btrfs_alloc_reserved_file_extent(trans, root,
2028 root->root_key.objectid,
2029 btrfs_ino(inode), file_pos, &ins);
2031 btrfs_free_path(path);
2036 /* snapshot-aware defrag */
2037 struct sa_defrag_extent_backref {
2038 struct rb_node node;
2039 struct old_sa_defrag_extent *old;
2048 struct old_sa_defrag_extent {
2049 struct list_head list;
2050 struct new_sa_defrag_extent *new;
2059 struct new_sa_defrag_extent {
2060 struct rb_root root;
2061 struct list_head head;
2062 struct btrfs_path *path;
2063 struct inode *inode;
2071 static int backref_comp(struct sa_defrag_extent_backref *b1,
2072 struct sa_defrag_extent_backref *b2)
2074 if (b1->root_id < b2->root_id)
2076 else if (b1->root_id > b2->root_id)
2079 if (b1->inum < b2->inum)
2081 else if (b1->inum > b2->inum)
2084 if (b1->file_pos < b2->file_pos)
2086 else if (b1->file_pos > b2->file_pos)
2090 * [------------------------------] ===> (a range of space)
2091 * |<--->| |<---->| =============> (fs/file tree A)
2092 * |<---------------------------->| ===> (fs/file tree B)
2094 * A range of space can refer to two file extents in one tree while
2095 * refer to only one file extent in another tree.
2097 * So we may process a disk offset more than one time(two extents in A)
2098 * and locate at the same extent(one extent in B), then insert two same
2099 * backrefs(both refer to the extent in B).
2104 static void backref_insert(struct rb_root *root,
2105 struct sa_defrag_extent_backref *backref)
2107 struct rb_node **p = &root->rb_node;
2108 struct rb_node *parent = NULL;
2109 struct sa_defrag_extent_backref *entry;
2114 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2116 ret = backref_comp(backref, entry);
2120 p = &(*p)->rb_right;
2123 rb_link_node(&backref->node, parent, p);
2124 rb_insert_color(&backref->node, root);
2128 * Note the backref might has changed, and in this case we just return 0.
2130 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2133 struct btrfs_file_extent_item *extent;
2134 struct btrfs_fs_info *fs_info;
2135 struct old_sa_defrag_extent *old = ctx;
2136 struct new_sa_defrag_extent *new = old->new;
2137 struct btrfs_path *path = new->path;
2138 struct btrfs_key key;
2139 struct btrfs_root *root;
2140 struct sa_defrag_extent_backref *backref;
2141 struct extent_buffer *leaf;
2142 struct inode *inode = new->inode;
2148 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2149 inum == btrfs_ino(inode))
2152 key.objectid = root_id;
2153 key.type = BTRFS_ROOT_ITEM_KEY;
2154 key.offset = (u64)-1;
2156 fs_info = BTRFS_I(inode)->root->fs_info;
2157 root = btrfs_read_fs_root_no_name(fs_info, &key);
2159 if (PTR_ERR(root) == -ENOENT)
2162 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2163 inum, offset, root_id);
2164 return PTR_ERR(root);
2167 key.objectid = inum;
2168 key.type = BTRFS_EXTENT_DATA_KEY;
2169 if (offset > (u64)-1 << 32)
2172 key.offset = offset;
2174 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2175 if (WARN_ON(ret < 0))
2182 leaf = path->nodes[0];
2183 slot = path->slots[0];
2185 if (slot >= btrfs_header_nritems(leaf)) {
2186 ret = btrfs_next_leaf(root, path);
2189 } else if (ret > 0) {
2198 btrfs_item_key_to_cpu(leaf, &key, slot);
2200 if (key.objectid > inum)
2203 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2206 extent = btrfs_item_ptr(leaf, slot,
2207 struct btrfs_file_extent_item);
2209 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2213 * 'offset' refers to the exact key.offset,
2214 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2215 * (key.offset - extent_offset).
2217 if (key.offset != offset)
2220 extent_offset = btrfs_file_extent_offset(leaf, extent);
2221 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2223 if (extent_offset >= old->extent_offset + old->offset +
2224 old->len || extent_offset + num_bytes <=
2225 old->extent_offset + old->offset)
2230 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2236 backref->root_id = root_id;
2237 backref->inum = inum;
2238 backref->file_pos = offset;
2239 backref->num_bytes = num_bytes;
2240 backref->extent_offset = extent_offset;
2241 backref->generation = btrfs_file_extent_generation(leaf, extent);
2243 backref_insert(&new->root, backref);
2246 btrfs_release_path(path);
2251 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2252 struct new_sa_defrag_extent *new)
2254 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2255 struct old_sa_defrag_extent *old, *tmp;
2260 list_for_each_entry_safe(old, tmp, &new->head, list) {
2261 ret = iterate_inodes_from_logical(old->bytenr +
2262 old->extent_offset, fs_info,
2263 path, record_one_backref,
2265 if (ret < 0 && ret != -ENOENT)
2268 /* no backref to be processed for this extent */
2270 list_del(&old->list);
2275 if (list_empty(&new->head))
2281 static int relink_is_mergable(struct extent_buffer *leaf,
2282 struct btrfs_file_extent_item *fi,
2283 struct new_sa_defrag_extent *new)
2285 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2288 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2291 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2294 if (btrfs_file_extent_encryption(leaf, fi) ||
2295 btrfs_file_extent_other_encoding(leaf, fi))
2302 * Note the backref might has changed, and in this case we just return 0.
2304 static noinline int relink_extent_backref(struct btrfs_path *path,
2305 struct sa_defrag_extent_backref *prev,
2306 struct sa_defrag_extent_backref *backref)
2308 struct btrfs_file_extent_item *extent;
2309 struct btrfs_file_extent_item *item;
2310 struct btrfs_ordered_extent *ordered;
2311 struct btrfs_trans_handle *trans;
2312 struct btrfs_fs_info *fs_info;
2313 struct btrfs_root *root;
2314 struct btrfs_key key;
2315 struct extent_buffer *leaf;
2316 struct old_sa_defrag_extent *old = backref->old;
2317 struct new_sa_defrag_extent *new = old->new;
2318 struct inode *src_inode = new->inode;
2319 struct inode *inode;
2320 struct extent_state *cached = NULL;
2329 if (prev && prev->root_id == backref->root_id &&
2330 prev->inum == backref->inum &&
2331 prev->file_pos + prev->num_bytes == backref->file_pos)
2334 /* step 1: get root */
2335 key.objectid = backref->root_id;
2336 key.type = BTRFS_ROOT_ITEM_KEY;
2337 key.offset = (u64)-1;
2339 fs_info = BTRFS_I(src_inode)->root->fs_info;
2340 index = srcu_read_lock(&fs_info->subvol_srcu);
2342 root = btrfs_read_fs_root_no_name(fs_info, &key);
2344 srcu_read_unlock(&fs_info->subvol_srcu, index);
2345 if (PTR_ERR(root) == -ENOENT)
2347 return PTR_ERR(root);
2350 if (btrfs_root_readonly(root)) {
2351 srcu_read_unlock(&fs_info->subvol_srcu, index);
2355 /* step 2: get inode */
2356 key.objectid = backref->inum;
2357 key.type = BTRFS_INODE_ITEM_KEY;
2360 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2361 if (IS_ERR(inode)) {
2362 srcu_read_unlock(&fs_info->subvol_srcu, index);
2366 srcu_read_unlock(&fs_info->subvol_srcu, index);
2368 /* step 3: relink backref */
2369 lock_start = backref->file_pos;
2370 lock_end = backref->file_pos + backref->num_bytes - 1;
2371 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2374 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2376 btrfs_put_ordered_extent(ordered);
2380 trans = btrfs_join_transaction(root);
2381 if (IS_ERR(trans)) {
2382 ret = PTR_ERR(trans);
2386 key.objectid = backref->inum;
2387 key.type = BTRFS_EXTENT_DATA_KEY;
2388 key.offset = backref->file_pos;
2390 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2393 } else if (ret > 0) {
2398 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2399 struct btrfs_file_extent_item);
2401 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2402 backref->generation)
2405 btrfs_release_path(path);
2407 start = backref->file_pos;
2408 if (backref->extent_offset < old->extent_offset + old->offset)
2409 start += old->extent_offset + old->offset -
2410 backref->extent_offset;
2412 len = min(backref->extent_offset + backref->num_bytes,
2413 old->extent_offset + old->offset + old->len);
2414 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2416 ret = btrfs_drop_extents(trans, root, inode, start,
2421 key.objectid = btrfs_ino(inode);
2422 key.type = BTRFS_EXTENT_DATA_KEY;
2425 path->leave_spinning = 1;
2427 struct btrfs_file_extent_item *fi;
2429 struct btrfs_key found_key;
2431 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2436 leaf = path->nodes[0];
2437 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2439 fi = btrfs_item_ptr(leaf, path->slots[0],
2440 struct btrfs_file_extent_item);
2441 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2443 if (extent_len + found_key.offset == start &&
2444 relink_is_mergable(leaf, fi, new)) {
2445 btrfs_set_file_extent_num_bytes(leaf, fi,
2447 btrfs_mark_buffer_dirty(leaf);
2448 inode_add_bytes(inode, len);
2454 btrfs_release_path(path);
2459 ret = btrfs_insert_empty_item(trans, root, path, &key,
2462 btrfs_abort_transaction(trans, root, ret);
2466 leaf = path->nodes[0];
2467 item = btrfs_item_ptr(leaf, path->slots[0],
2468 struct btrfs_file_extent_item);
2469 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2470 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2471 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2472 btrfs_set_file_extent_num_bytes(leaf, item, len);
2473 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2474 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2475 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2476 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2477 btrfs_set_file_extent_encryption(leaf, item, 0);
2478 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2480 btrfs_mark_buffer_dirty(leaf);
2481 inode_add_bytes(inode, len);
2482 btrfs_release_path(path);
2484 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2486 backref->root_id, backref->inum,
2487 new->file_pos, 0); /* start - extent_offset */
2489 btrfs_abort_transaction(trans, root, ret);
2495 btrfs_release_path(path);
2496 path->leave_spinning = 0;
2497 btrfs_end_transaction(trans, root);
2499 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2505 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2507 struct old_sa_defrag_extent *old, *tmp;
2512 list_for_each_entry_safe(old, tmp, &new->head, list) {
2513 list_del(&old->list);
2519 static void relink_file_extents(struct new_sa_defrag_extent *new)
2521 struct btrfs_path *path;
2522 struct sa_defrag_extent_backref *backref;
2523 struct sa_defrag_extent_backref *prev = NULL;
2524 struct inode *inode;
2525 struct btrfs_root *root;
2526 struct rb_node *node;
2530 root = BTRFS_I(inode)->root;
2532 path = btrfs_alloc_path();
2536 if (!record_extent_backrefs(path, new)) {
2537 btrfs_free_path(path);
2540 btrfs_release_path(path);
2543 node = rb_first(&new->root);
2546 rb_erase(node, &new->root);
2548 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2550 ret = relink_extent_backref(path, prev, backref);
2563 btrfs_free_path(path);
2565 free_sa_defrag_extent(new);
2567 atomic_dec(&root->fs_info->defrag_running);
2568 wake_up(&root->fs_info->transaction_wait);
2571 static struct new_sa_defrag_extent *
2572 record_old_file_extents(struct inode *inode,
2573 struct btrfs_ordered_extent *ordered)
2575 struct btrfs_root *root = BTRFS_I(inode)->root;
2576 struct btrfs_path *path;
2577 struct btrfs_key key;
2578 struct old_sa_defrag_extent *old;
2579 struct new_sa_defrag_extent *new;
2582 new = kmalloc(sizeof(*new), GFP_NOFS);
2587 new->file_pos = ordered->file_offset;
2588 new->len = ordered->len;
2589 new->bytenr = ordered->start;
2590 new->disk_len = ordered->disk_len;
2591 new->compress_type = ordered->compress_type;
2592 new->root = RB_ROOT;
2593 INIT_LIST_HEAD(&new->head);
2595 path = btrfs_alloc_path();
2599 key.objectid = btrfs_ino(inode);
2600 key.type = BTRFS_EXTENT_DATA_KEY;
2601 key.offset = new->file_pos;
2603 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2606 if (ret > 0 && path->slots[0] > 0)
2609 /* find out all the old extents for the file range */
2611 struct btrfs_file_extent_item *extent;
2612 struct extent_buffer *l;
2621 slot = path->slots[0];
2623 if (slot >= btrfs_header_nritems(l)) {
2624 ret = btrfs_next_leaf(root, path);
2632 btrfs_item_key_to_cpu(l, &key, slot);
2634 if (key.objectid != btrfs_ino(inode))
2636 if (key.type != BTRFS_EXTENT_DATA_KEY)
2638 if (key.offset >= new->file_pos + new->len)
2641 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2643 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2644 if (key.offset + num_bytes < new->file_pos)
2647 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2651 extent_offset = btrfs_file_extent_offset(l, extent);
2653 old = kmalloc(sizeof(*old), GFP_NOFS);
2657 offset = max(new->file_pos, key.offset);
2658 end = min(new->file_pos + new->len, key.offset + num_bytes);
2660 old->bytenr = disk_bytenr;
2661 old->extent_offset = extent_offset;
2662 old->offset = offset - key.offset;
2663 old->len = end - offset;
2666 list_add_tail(&old->list, &new->head);
2672 btrfs_free_path(path);
2673 atomic_inc(&root->fs_info->defrag_running);
2678 btrfs_free_path(path);
2680 free_sa_defrag_extent(new);
2684 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2687 struct btrfs_block_group_cache *cache;
2689 cache = btrfs_lookup_block_group(root->fs_info, start);
2692 spin_lock(&cache->lock);
2693 cache->delalloc_bytes -= len;
2694 spin_unlock(&cache->lock);
2696 btrfs_put_block_group(cache);
2699 /* as ordered data IO finishes, this gets called so we can finish
2700 * an ordered extent if the range of bytes in the file it covers are
2703 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2705 struct inode *inode = ordered_extent->inode;
2706 struct btrfs_root *root = BTRFS_I(inode)->root;
2707 struct btrfs_trans_handle *trans = NULL;
2708 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2709 struct extent_state *cached_state = NULL;
2710 struct new_sa_defrag_extent *new = NULL;
2711 int compress_type = 0;
2713 u64 logical_len = ordered_extent->len;
2715 bool truncated = false;
2717 nolock = btrfs_is_free_space_inode(inode);
2719 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2724 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2725 ordered_extent->file_offset +
2726 ordered_extent->len - 1);
2728 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2730 logical_len = ordered_extent->truncated_len;
2731 /* Truncated the entire extent, don't bother adding */
2736 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2737 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2738 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2740 trans = btrfs_join_transaction_nolock(root);
2742 trans = btrfs_join_transaction(root);
2743 if (IS_ERR(trans)) {
2744 ret = PTR_ERR(trans);
2748 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2749 ret = btrfs_update_inode_fallback(trans, root, inode);
2750 if (ret) /* -ENOMEM or corruption */
2751 btrfs_abort_transaction(trans, root, ret);
2755 lock_extent_bits(io_tree, ordered_extent->file_offset,
2756 ordered_extent->file_offset + ordered_extent->len - 1,
2759 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2760 ordered_extent->file_offset + ordered_extent->len - 1,
2761 EXTENT_DEFRAG, 1, cached_state);
2763 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2764 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2765 /* the inode is shared */
2766 new = record_old_file_extents(inode, ordered_extent);
2768 clear_extent_bit(io_tree, ordered_extent->file_offset,
2769 ordered_extent->file_offset + ordered_extent->len - 1,
2770 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2774 trans = btrfs_join_transaction_nolock(root);
2776 trans = btrfs_join_transaction(root);
2777 if (IS_ERR(trans)) {
2778 ret = PTR_ERR(trans);
2783 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2785 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2786 compress_type = ordered_extent->compress_type;
2787 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2788 BUG_ON(compress_type);
2789 ret = btrfs_mark_extent_written(trans, inode,
2790 ordered_extent->file_offset,
2791 ordered_extent->file_offset +
2794 BUG_ON(root == root->fs_info->tree_root);
2795 ret = insert_reserved_file_extent(trans, inode,
2796 ordered_extent->file_offset,
2797 ordered_extent->start,
2798 ordered_extent->disk_len,
2799 logical_len, logical_len,
2800 compress_type, 0, 0,
2801 BTRFS_FILE_EXTENT_REG);
2803 btrfs_release_delalloc_bytes(root,
2804 ordered_extent->start,
2805 ordered_extent->disk_len);
2807 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2808 ordered_extent->file_offset, ordered_extent->len,
2811 btrfs_abort_transaction(trans, root, ret);
2815 add_pending_csums(trans, inode, ordered_extent->file_offset,
2816 &ordered_extent->list);
2818 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2819 ret = btrfs_update_inode_fallback(trans, root, inode);
2820 if (ret) { /* -ENOMEM or corruption */
2821 btrfs_abort_transaction(trans, root, ret);
2826 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2827 ordered_extent->file_offset +
2828 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2830 if (root != root->fs_info->tree_root)
2831 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2833 btrfs_end_transaction(trans, root);
2835 if (ret || truncated) {
2839 start = ordered_extent->file_offset + logical_len;
2841 start = ordered_extent->file_offset;
2842 end = ordered_extent->file_offset + ordered_extent->len - 1;
2843 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2845 /* Drop the cache for the part of the extent we didn't write. */
2846 btrfs_drop_extent_cache(inode, start, end, 0);
2849 * If the ordered extent had an IOERR or something else went
2850 * wrong we need to return the space for this ordered extent
2851 * back to the allocator. We only free the extent in the
2852 * truncated case if we didn't write out the extent at all.
2854 if ((ret || !logical_len) &&
2855 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2856 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2857 btrfs_free_reserved_extent(root, ordered_extent->start,
2858 ordered_extent->disk_len, 1);
2863 * This needs to be done to make sure anybody waiting knows we are done
2864 * updating everything for this ordered extent.
2866 btrfs_remove_ordered_extent(inode, ordered_extent);
2868 /* for snapshot-aware defrag */
2871 free_sa_defrag_extent(new);
2872 atomic_dec(&root->fs_info->defrag_running);
2874 relink_file_extents(new);
2879 btrfs_put_ordered_extent(ordered_extent);
2880 /* once for the tree */
2881 btrfs_put_ordered_extent(ordered_extent);
2886 static void finish_ordered_fn(struct btrfs_work *work)
2888 struct btrfs_ordered_extent *ordered_extent;
2889 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2890 btrfs_finish_ordered_io(ordered_extent);
2893 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2894 struct extent_state *state, int uptodate)
2896 struct inode *inode = page->mapping->host;
2897 struct btrfs_root *root = BTRFS_I(inode)->root;
2898 struct btrfs_ordered_extent *ordered_extent = NULL;
2899 struct btrfs_workqueue *wq;
2900 btrfs_work_func_t func;
2902 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2904 ClearPagePrivate2(page);
2905 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2906 end - start + 1, uptodate))
2909 if (btrfs_is_free_space_inode(inode)) {
2910 wq = root->fs_info->endio_freespace_worker;
2911 func = btrfs_freespace_write_helper;
2913 wq = root->fs_info->endio_write_workers;
2914 func = btrfs_endio_write_helper;
2917 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
2919 btrfs_queue_work(wq, &ordered_extent->work);
2924 static int __readpage_endio_check(struct inode *inode,
2925 struct btrfs_io_bio *io_bio,
2926 int icsum, struct page *page,
2927 int pgoff, u64 start, size_t len)
2932 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2933 DEFAULT_RATELIMIT_BURST);
2935 csum_expected = *(((u32 *)io_bio->csum) + icsum);
2937 kaddr = kmap_atomic(page);
2938 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
2939 btrfs_csum_final(csum, (char *)&csum);
2940 if (csum != csum_expected)
2943 kunmap_atomic(kaddr);
2946 if (__ratelimit(&_rs))
2947 btrfs_info(BTRFS_I(inode)->root->fs_info,
2948 "csum failed ino %llu off %llu csum %u expected csum %u",
2949 btrfs_ino(inode), start, csum, csum_expected);
2950 memset(kaddr + pgoff, 1, len);
2951 flush_dcache_page(page);
2952 kunmap_atomic(kaddr);
2953 if (csum_expected == 0)
2959 * when reads are done, we need to check csums to verify the data is correct
2960 * if there's a match, we allow the bio to finish. If not, the code in
2961 * extent_io.c will try to find good copies for us.
2963 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2964 u64 phy_offset, struct page *page,
2965 u64 start, u64 end, int mirror)
2967 size_t offset = start - page_offset(page);
2968 struct inode *inode = page->mapping->host;
2969 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2970 struct btrfs_root *root = BTRFS_I(inode)->root;
2972 if (PageChecked(page)) {
2973 ClearPageChecked(page);
2977 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2980 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2981 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2982 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2987 phy_offset >>= inode->i_sb->s_blocksize_bits;
2988 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
2989 start, (size_t)(end - start + 1));
2992 struct delayed_iput {
2993 struct list_head list;
2994 struct inode *inode;
2997 /* JDM: If this is fs-wide, why can't we add a pointer to
2998 * btrfs_inode instead and avoid the allocation? */
2999 void btrfs_add_delayed_iput(struct inode *inode)
3001 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3002 struct delayed_iput *delayed;
3004 if (atomic_add_unless(&inode->i_count, -1, 1))
3007 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
3008 delayed->inode = inode;
3010 spin_lock(&fs_info->delayed_iput_lock);
3011 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
3012 spin_unlock(&fs_info->delayed_iput_lock);
3015 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3018 struct btrfs_fs_info *fs_info = root->fs_info;
3019 struct delayed_iput *delayed;
3022 spin_lock(&fs_info->delayed_iput_lock);
3023 empty = list_empty(&fs_info->delayed_iputs);
3024 spin_unlock(&fs_info->delayed_iput_lock);
3028 spin_lock(&fs_info->delayed_iput_lock);
3029 list_splice_init(&fs_info->delayed_iputs, &list);
3030 spin_unlock(&fs_info->delayed_iput_lock);
3032 while (!list_empty(&list)) {
3033 delayed = list_entry(list.next, struct delayed_iput, list);
3034 list_del(&delayed->list);
3035 iput(delayed->inode);
3041 * This is called in transaction commit time. If there are no orphan
3042 * files in the subvolume, it removes orphan item and frees block_rsv
3045 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3046 struct btrfs_root *root)
3048 struct btrfs_block_rsv *block_rsv;
3051 if (atomic_read(&root->orphan_inodes) ||
3052 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3055 spin_lock(&root->orphan_lock);
3056 if (atomic_read(&root->orphan_inodes)) {
3057 spin_unlock(&root->orphan_lock);
3061 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3062 spin_unlock(&root->orphan_lock);
3066 block_rsv = root->orphan_block_rsv;
3067 root->orphan_block_rsv = NULL;
3068 spin_unlock(&root->orphan_lock);
3070 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3071 btrfs_root_refs(&root->root_item) > 0) {
3072 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3073 root->root_key.objectid);
3075 btrfs_abort_transaction(trans, root, ret);
3077 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3082 WARN_ON(block_rsv->size > 0);
3083 btrfs_free_block_rsv(root, block_rsv);
3088 * This creates an orphan entry for the given inode in case something goes
3089 * wrong in the middle of an unlink/truncate.
3091 * NOTE: caller of this function should reserve 5 units of metadata for
3094 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3096 struct btrfs_root *root = BTRFS_I(inode)->root;
3097 struct btrfs_block_rsv *block_rsv = NULL;
3102 if (!root->orphan_block_rsv) {
3103 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3108 spin_lock(&root->orphan_lock);
3109 if (!root->orphan_block_rsv) {
3110 root->orphan_block_rsv = block_rsv;
3111 } else if (block_rsv) {
3112 btrfs_free_block_rsv(root, block_rsv);
3116 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3117 &BTRFS_I(inode)->runtime_flags)) {
3120 * For proper ENOSPC handling, we should do orphan
3121 * cleanup when mounting. But this introduces backward
3122 * compatibility issue.
3124 if (!xchg(&root->orphan_item_inserted, 1))
3130 atomic_inc(&root->orphan_inodes);
3133 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3134 &BTRFS_I(inode)->runtime_flags))
3136 spin_unlock(&root->orphan_lock);
3138 /* grab metadata reservation from transaction handle */
3140 ret = btrfs_orphan_reserve_metadata(trans, inode);
3141 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3144 /* insert an orphan item to track this unlinked/truncated file */
3146 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3148 atomic_dec(&root->orphan_inodes);
3150 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3151 &BTRFS_I(inode)->runtime_flags);
3152 btrfs_orphan_release_metadata(inode);
3154 if (ret != -EEXIST) {
3155 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3156 &BTRFS_I(inode)->runtime_flags);
3157 btrfs_abort_transaction(trans, root, ret);
3164 /* insert an orphan item to track subvolume contains orphan files */
3166 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3167 root->root_key.objectid);
3168 if (ret && ret != -EEXIST) {
3169 btrfs_abort_transaction(trans, root, ret);
3177 * We have done the truncate/delete so we can go ahead and remove the orphan
3178 * item for this particular inode.
3180 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3181 struct inode *inode)
3183 struct btrfs_root *root = BTRFS_I(inode)->root;
3184 int delete_item = 0;
3185 int release_rsv = 0;
3188 spin_lock(&root->orphan_lock);
3189 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3190 &BTRFS_I(inode)->runtime_flags))
3193 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3194 &BTRFS_I(inode)->runtime_flags))
3196 spin_unlock(&root->orphan_lock);
3199 atomic_dec(&root->orphan_inodes);
3201 ret = btrfs_del_orphan_item(trans, root,
3206 btrfs_orphan_release_metadata(inode);
3212 * this cleans up any orphans that may be left on the list from the last use
3215 int btrfs_orphan_cleanup(struct btrfs_root *root)
3217 struct btrfs_path *path;
3218 struct extent_buffer *leaf;
3219 struct btrfs_key key, found_key;
3220 struct btrfs_trans_handle *trans;
3221 struct inode *inode;
3222 u64 last_objectid = 0;
3223 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3225 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3228 path = btrfs_alloc_path();
3235 key.objectid = BTRFS_ORPHAN_OBJECTID;
3236 key.type = BTRFS_ORPHAN_ITEM_KEY;
3237 key.offset = (u64)-1;
3240 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3245 * if ret == 0 means we found what we were searching for, which
3246 * is weird, but possible, so only screw with path if we didn't
3247 * find the key and see if we have stuff that matches
3251 if (path->slots[0] == 0)
3256 /* pull out the item */
3257 leaf = path->nodes[0];
3258 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3260 /* make sure the item matches what we want */
3261 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3263 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3266 /* release the path since we're done with it */
3267 btrfs_release_path(path);
3270 * this is where we are basically btrfs_lookup, without the
3271 * crossing root thing. we store the inode number in the
3272 * offset of the orphan item.
3275 if (found_key.offset == last_objectid) {
3276 btrfs_err(root->fs_info,
3277 "Error removing orphan entry, stopping orphan cleanup");
3282 last_objectid = found_key.offset;
3284 found_key.objectid = found_key.offset;
3285 found_key.type = BTRFS_INODE_ITEM_KEY;
3286 found_key.offset = 0;
3287 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3288 ret = PTR_ERR_OR_ZERO(inode);
3289 if (ret && ret != -ESTALE)
3292 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3293 struct btrfs_root *dead_root;
3294 struct btrfs_fs_info *fs_info = root->fs_info;
3295 int is_dead_root = 0;
3298 * this is an orphan in the tree root. Currently these
3299 * could come from 2 sources:
3300 * a) a snapshot deletion in progress
3301 * b) a free space cache inode
3302 * We need to distinguish those two, as the snapshot
3303 * orphan must not get deleted.
3304 * find_dead_roots already ran before us, so if this
3305 * is a snapshot deletion, we should find the root
3306 * in the dead_roots list
3308 spin_lock(&fs_info->trans_lock);
3309 list_for_each_entry(dead_root, &fs_info->dead_roots,
3311 if (dead_root->root_key.objectid ==
3312 found_key.objectid) {
3317 spin_unlock(&fs_info->trans_lock);
3319 /* prevent this orphan from being found again */
3320 key.offset = found_key.objectid - 1;
3325 * Inode is already gone but the orphan item is still there,
3326 * kill the orphan item.
3328 if (ret == -ESTALE) {
3329 trans = btrfs_start_transaction(root, 1);
3330 if (IS_ERR(trans)) {
3331 ret = PTR_ERR(trans);
3334 btrfs_debug(root->fs_info, "auto deleting %Lu",
3335 found_key.objectid);
3336 ret = btrfs_del_orphan_item(trans, root,
3337 found_key.objectid);
3338 btrfs_end_transaction(trans, root);
3345 * add this inode to the orphan list so btrfs_orphan_del does
3346 * the proper thing when we hit it
3348 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3349 &BTRFS_I(inode)->runtime_flags);
3350 atomic_inc(&root->orphan_inodes);
3352 /* if we have links, this was a truncate, lets do that */
3353 if (inode->i_nlink) {
3354 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3360 /* 1 for the orphan item deletion. */
3361 trans = btrfs_start_transaction(root, 1);
3362 if (IS_ERR(trans)) {
3364 ret = PTR_ERR(trans);
3367 ret = btrfs_orphan_add(trans, inode);
3368 btrfs_end_transaction(trans, root);
3374 ret = btrfs_truncate(inode);
3376 btrfs_orphan_del(NULL, inode);
3381 /* this will do delete_inode and everything for us */
3386 /* release the path since we're done with it */
3387 btrfs_release_path(path);
3389 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3391 if (root->orphan_block_rsv)
3392 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3395 if (root->orphan_block_rsv ||
3396 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3397 trans = btrfs_join_transaction(root);
3399 btrfs_end_transaction(trans, root);
3403 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3405 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3409 btrfs_crit(root->fs_info,
3410 "could not do orphan cleanup %d", ret);
3411 btrfs_free_path(path);
3416 * very simple check to peek ahead in the leaf looking for xattrs. If we
3417 * don't find any xattrs, we know there can't be any acls.
3419 * slot is the slot the inode is in, objectid is the objectid of the inode
3421 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3422 int slot, u64 objectid,
3423 int *first_xattr_slot)
3425 u32 nritems = btrfs_header_nritems(leaf);
3426 struct btrfs_key found_key;
3427 static u64 xattr_access = 0;
3428 static u64 xattr_default = 0;
3431 if (!xattr_access) {
3432 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3433 strlen(POSIX_ACL_XATTR_ACCESS));
3434 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3435 strlen(POSIX_ACL_XATTR_DEFAULT));
3439 *first_xattr_slot = -1;
3440 while (slot < nritems) {
3441 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3443 /* we found a different objectid, there must not be acls */
3444 if (found_key.objectid != objectid)
3447 /* we found an xattr, assume we've got an acl */
3448 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3449 if (*first_xattr_slot == -1)
3450 *first_xattr_slot = slot;
3451 if (found_key.offset == xattr_access ||
3452 found_key.offset == xattr_default)
3457 * we found a key greater than an xattr key, there can't
3458 * be any acls later on
3460 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3467 * it goes inode, inode backrefs, xattrs, extents,
3468 * so if there are a ton of hard links to an inode there can
3469 * be a lot of backrefs. Don't waste time searching too hard,
3470 * this is just an optimization
3475 /* we hit the end of the leaf before we found an xattr or
3476 * something larger than an xattr. We have to assume the inode
3479 if (*first_xattr_slot == -1)
3480 *first_xattr_slot = slot;
3485 * read an inode from the btree into the in-memory inode
3487 static void btrfs_read_locked_inode(struct inode *inode)
3489 struct btrfs_path *path;
3490 struct extent_buffer *leaf;
3491 struct btrfs_inode_item *inode_item;
3492 struct btrfs_timespec *tspec;
3493 struct btrfs_root *root = BTRFS_I(inode)->root;
3494 struct btrfs_key location;
3499 bool filled = false;
3500 int first_xattr_slot;
3502 ret = btrfs_fill_inode(inode, &rdev);
3506 path = btrfs_alloc_path();
3510 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3512 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3516 leaf = path->nodes[0];
3521 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3522 struct btrfs_inode_item);
3523 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3524 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3525 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3526 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3527 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3529 tspec = btrfs_inode_atime(inode_item);
3530 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3531 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3533 tspec = btrfs_inode_mtime(inode_item);
3534 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3535 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3537 tspec = btrfs_inode_ctime(inode_item);
3538 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3539 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3541 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3542 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3543 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3546 * If we were modified in the current generation and evicted from memory
3547 * and then re-read we need to do a full sync since we don't have any
3548 * idea about which extents were modified before we were evicted from
3551 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3552 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3553 &BTRFS_I(inode)->runtime_flags);
3555 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3556 inode->i_generation = BTRFS_I(inode)->generation;
3558 rdev = btrfs_inode_rdev(leaf, inode_item);
3560 BTRFS_I(inode)->index_cnt = (u64)-1;
3561 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3565 if (inode->i_nlink != 1 ||
3566 path->slots[0] >= btrfs_header_nritems(leaf))
3569 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3570 if (location.objectid != btrfs_ino(inode))
3573 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3574 if (location.type == BTRFS_INODE_REF_KEY) {
3575 struct btrfs_inode_ref *ref;
3577 ref = (struct btrfs_inode_ref *)ptr;
3578 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3579 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3580 struct btrfs_inode_extref *extref;
3582 extref = (struct btrfs_inode_extref *)ptr;
3583 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3588 * try to precache a NULL acl entry for files that don't have
3589 * any xattrs or acls
3591 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3592 btrfs_ino(inode), &first_xattr_slot);
3593 if (first_xattr_slot != -1) {
3594 path->slots[0] = first_xattr_slot;
3595 ret = btrfs_load_inode_props(inode, path);
3597 btrfs_err(root->fs_info,
3598 "error loading props for ino %llu (root %llu): %d",
3600 root->root_key.objectid, ret);
3602 btrfs_free_path(path);
3605 cache_no_acl(inode);
3607 switch (inode->i_mode & S_IFMT) {
3609 inode->i_mapping->a_ops = &btrfs_aops;
3610 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3611 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3612 inode->i_fop = &btrfs_file_operations;
3613 inode->i_op = &btrfs_file_inode_operations;
3616 inode->i_fop = &btrfs_dir_file_operations;
3617 if (root == root->fs_info->tree_root)
3618 inode->i_op = &btrfs_dir_ro_inode_operations;
3620 inode->i_op = &btrfs_dir_inode_operations;
3623 inode->i_op = &btrfs_symlink_inode_operations;
3624 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3625 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3628 inode->i_op = &btrfs_special_inode_operations;
3629 init_special_inode(inode, inode->i_mode, rdev);
3633 btrfs_update_iflags(inode);
3637 btrfs_free_path(path);
3638 make_bad_inode(inode);
3642 * given a leaf and an inode, copy the inode fields into the leaf
3644 static void fill_inode_item(struct btrfs_trans_handle *trans,
3645 struct extent_buffer *leaf,
3646 struct btrfs_inode_item *item,
3647 struct inode *inode)
3649 struct btrfs_map_token token;
3651 btrfs_init_map_token(&token);
3653 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3654 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3655 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3657 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3658 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3660 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3661 inode->i_atime.tv_sec, &token);
3662 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3663 inode->i_atime.tv_nsec, &token);
3665 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3666 inode->i_mtime.tv_sec, &token);
3667 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3668 inode->i_mtime.tv_nsec, &token);
3670 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3671 inode->i_ctime.tv_sec, &token);
3672 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3673 inode->i_ctime.tv_nsec, &token);
3675 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3677 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3679 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3680 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3681 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3682 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3683 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3687 * copy everything in the in-memory inode into the btree.
3689 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3690 struct btrfs_root *root, struct inode *inode)
3692 struct btrfs_inode_item *inode_item;
3693 struct btrfs_path *path;
3694 struct extent_buffer *leaf;
3697 path = btrfs_alloc_path();
3701 path->leave_spinning = 1;
3702 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3710 leaf = path->nodes[0];
3711 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3712 struct btrfs_inode_item);
3714 fill_inode_item(trans, leaf, inode_item, inode);
3715 btrfs_mark_buffer_dirty(leaf);
3716 btrfs_set_inode_last_trans(trans, inode);
3719 btrfs_free_path(path);
3724 * copy everything in the in-memory inode into the btree.
3726 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3727 struct btrfs_root *root, struct inode *inode)
3732 * If the inode is a free space inode, we can deadlock during commit
3733 * if we put it into the delayed code.
3735 * The data relocation inode should also be directly updated
3738 if (!btrfs_is_free_space_inode(inode)
3739 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3740 && !root->fs_info->log_root_recovering) {
3741 btrfs_update_root_times(trans, root);
3743 ret = btrfs_delayed_update_inode(trans, root, inode);
3745 btrfs_set_inode_last_trans(trans, inode);
3749 return btrfs_update_inode_item(trans, root, inode);
3752 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3753 struct btrfs_root *root,
3754 struct inode *inode)
3758 ret = btrfs_update_inode(trans, root, inode);
3760 return btrfs_update_inode_item(trans, root, inode);
3765 * unlink helper that gets used here in inode.c and in the tree logging
3766 * recovery code. It remove a link in a directory with a given name, and
3767 * also drops the back refs in the inode to the directory
3769 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3770 struct btrfs_root *root,
3771 struct inode *dir, struct inode *inode,
3772 const char *name, int name_len)
3774 struct btrfs_path *path;
3776 struct extent_buffer *leaf;
3777 struct btrfs_dir_item *di;
3778 struct btrfs_key key;
3780 u64 ino = btrfs_ino(inode);
3781 u64 dir_ino = btrfs_ino(dir);
3783 path = btrfs_alloc_path();
3789 path->leave_spinning = 1;
3790 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3791 name, name_len, -1);
3800 leaf = path->nodes[0];
3801 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3802 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3805 btrfs_release_path(path);
3808 * If we don't have dir index, we have to get it by looking up
3809 * the inode ref, since we get the inode ref, remove it directly,
3810 * it is unnecessary to do delayed deletion.
3812 * But if we have dir index, needn't search inode ref to get it.
3813 * Since the inode ref is close to the inode item, it is better
3814 * that we delay to delete it, and just do this deletion when
3815 * we update the inode item.
3817 if (BTRFS_I(inode)->dir_index) {
3818 ret = btrfs_delayed_delete_inode_ref(inode);
3820 index = BTRFS_I(inode)->dir_index;
3825 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3828 btrfs_info(root->fs_info,
3829 "failed to delete reference to %.*s, inode %llu parent %llu",
3830 name_len, name, ino, dir_ino);
3831 btrfs_abort_transaction(trans, root, ret);
3835 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3837 btrfs_abort_transaction(trans, root, ret);
3841 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3843 if (ret != 0 && ret != -ENOENT) {
3844 btrfs_abort_transaction(trans, root, ret);
3848 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3853 btrfs_abort_transaction(trans, root, ret);
3855 btrfs_free_path(path);
3859 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3860 inode_inc_iversion(inode);
3861 inode_inc_iversion(dir);
3862 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3863 ret = btrfs_update_inode(trans, root, dir);
3868 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3869 struct btrfs_root *root,
3870 struct inode *dir, struct inode *inode,
3871 const char *name, int name_len)
3874 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3877 ret = btrfs_update_inode(trans, root, inode);
3883 * helper to start transaction for unlink and rmdir.
3885 * unlink and rmdir are special in btrfs, they do not always free space, so
3886 * if we cannot make our reservations the normal way try and see if there is
3887 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3888 * allow the unlink to occur.
3890 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3892 struct btrfs_trans_handle *trans;
3893 struct btrfs_root *root = BTRFS_I(dir)->root;
3897 * 1 for the possible orphan item
3898 * 1 for the dir item
3899 * 1 for the dir index
3900 * 1 for the inode ref
3903 trans = btrfs_start_transaction(root, 5);
3904 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3907 if (PTR_ERR(trans) == -ENOSPC) {
3908 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3910 trans = btrfs_start_transaction(root, 0);
3913 ret = btrfs_cond_migrate_bytes(root->fs_info,
3914 &root->fs_info->trans_block_rsv,
3917 btrfs_end_transaction(trans, root);
3918 return ERR_PTR(ret);
3920 trans->block_rsv = &root->fs_info->trans_block_rsv;
3921 trans->bytes_reserved = num_bytes;
3926 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3928 struct btrfs_root *root = BTRFS_I(dir)->root;
3929 struct btrfs_trans_handle *trans;
3930 struct inode *inode = dentry->d_inode;
3933 trans = __unlink_start_trans(dir);
3935 return PTR_ERR(trans);
3937 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3939 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3940 dentry->d_name.name, dentry->d_name.len);
3944 if (inode->i_nlink == 0) {
3945 ret = btrfs_orphan_add(trans, inode);
3951 btrfs_end_transaction(trans, root);
3952 btrfs_btree_balance_dirty(root);
3956 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3957 struct btrfs_root *root,
3958 struct inode *dir, u64 objectid,
3959 const char *name, int name_len)
3961 struct btrfs_path *path;
3962 struct extent_buffer *leaf;
3963 struct btrfs_dir_item *di;
3964 struct btrfs_key key;
3967 u64 dir_ino = btrfs_ino(dir);
3969 path = btrfs_alloc_path();
3973 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3974 name, name_len, -1);
3975 if (IS_ERR_OR_NULL(di)) {
3983 leaf = path->nodes[0];
3984 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3985 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3986 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3988 btrfs_abort_transaction(trans, root, ret);
3991 btrfs_release_path(path);
3993 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3994 objectid, root->root_key.objectid,
3995 dir_ino, &index, name, name_len);
3997 if (ret != -ENOENT) {
3998 btrfs_abort_transaction(trans, root, ret);
4001 di = btrfs_search_dir_index_item(root, path, dir_ino,
4003 if (IS_ERR_OR_NULL(di)) {
4008 btrfs_abort_transaction(trans, root, ret);
4012 leaf = path->nodes[0];
4013 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4014 btrfs_release_path(path);
4017 btrfs_release_path(path);
4019 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4021 btrfs_abort_transaction(trans, root, ret);
4025 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4026 inode_inc_iversion(dir);
4027 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4028 ret = btrfs_update_inode_fallback(trans, root, dir);
4030 btrfs_abort_transaction(trans, root, ret);
4032 btrfs_free_path(path);
4036 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4038 struct inode *inode = dentry->d_inode;
4040 struct btrfs_root *root = BTRFS_I(dir)->root;
4041 struct btrfs_trans_handle *trans;
4043 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4045 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4048 trans = __unlink_start_trans(dir);
4050 return PTR_ERR(trans);
4052 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4053 err = btrfs_unlink_subvol(trans, root, dir,
4054 BTRFS_I(inode)->location.objectid,
4055 dentry->d_name.name,
4056 dentry->d_name.len);
4060 err = btrfs_orphan_add(trans, inode);
4064 /* now the directory is empty */
4065 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
4066 dentry->d_name.name, dentry->d_name.len);
4068 btrfs_i_size_write(inode, 0);
4070 btrfs_end_transaction(trans, root);
4071 btrfs_btree_balance_dirty(root);
4077 * this can truncate away extent items, csum items and directory items.
4078 * It starts at a high offset and removes keys until it can't find
4079 * any higher than new_size
4081 * csum items that cross the new i_size are truncated to the new size
4084 * min_type is the minimum key type to truncate down to. If set to 0, this
4085 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4087 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4088 struct btrfs_root *root,
4089 struct inode *inode,
4090 u64 new_size, u32 min_type)
4092 struct btrfs_path *path;
4093 struct extent_buffer *leaf;
4094 struct btrfs_file_extent_item *fi;
4095 struct btrfs_key key;
4096 struct btrfs_key found_key;
4097 u64 extent_start = 0;
4098 u64 extent_num_bytes = 0;
4099 u64 extent_offset = 0;
4101 u64 last_size = (u64)-1;
4102 u32 found_type = (u8)-1;
4105 int pending_del_nr = 0;
4106 int pending_del_slot = 0;
4107 int extent_type = -1;
4110 u64 ino = btrfs_ino(inode);
4112 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4114 path = btrfs_alloc_path();
4120 * We want to drop from the next block forward in case this new size is
4121 * not block aligned since we will be keeping the last block of the
4122 * extent just the way it is.
4124 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4125 root == root->fs_info->tree_root)
4126 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4127 root->sectorsize), (u64)-1, 0);
4130 * This function is also used to drop the items in the log tree before
4131 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4132 * it is used to drop the loged items. So we shouldn't kill the delayed
4135 if (min_type == 0 && root == BTRFS_I(inode)->root)
4136 btrfs_kill_delayed_inode_items(inode);
4139 key.offset = (u64)-1;
4143 path->leave_spinning = 1;
4144 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4151 /* there are no items in the tree for us to truncate, we're
4154 if (path->slots[0] == 0)
4161 leaf = path->nodes[0];
4162 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4163 found_type = found_key.type;
4165 if (found_key.objectid != ino)
4168 if (found_type < min_type)
4171 item_end = found_key.offset;
4172 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4173 fi = btrfs_item_ptr(leaf, path->slots[0],
4174 struct btrfs_file_extent_item);
4175 extent_type = btrfs_file_extent_type(leaf, fi);
4176 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4178 btrfs_file_extent_num_bytes(leaf, fi);
4179 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4180 item_end += btrfs_file_extent_inline_len(leaf,
4181 path->slots[0], fi);
4185 if (found_type > min_type) {
4188 if (item_end < new_size)
4190 if (found_key.offset >= new_size)
4196 /* FIXME, shrink the extent if the ref count is only 1 */
4197 if (found_type != BTRFS_EXTENT_DATA_KEY)
4201 last_size = found_key.offset;
4203 last_size = new_size;
4205 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4207 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4209 u64 orig_num_bytes =
4210 btrfs_file_extent_num_bytes(leaf, fi);
4211 extent_num_bytes = ALIGN(new_size -
4214 btrfs_set_file_extent_num_bytes(leaf, fi,
4216 num_dec = (orig_num_bytes -
4218 if (test_bit(BTRFS_ROOT_REF_COWS,
4221 inode_sub_bytes(inode, num_dec);
4222 btrfs_mark_buffer_dirty(leaf);
4225 btrfs_file_extent_disk_num_bytes(leaf,
4227 extent_offset = found_key.offset -
4228 btrfs_file_extent_offset(leaf, fi);
4230 /* FIXME blocksize != 4096 */
4231 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4232 if (extent_start != 0) {
4234 if (test_bit(BTRFS_ROOT_REF_COWS,
4236 inode_sub_bytes(inode, num_dec);
4239 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4241 * we can't truncate inline items that have had
4245 btrfs_file_extent_compression(leaf, fi) == 0 &&
4246 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4247 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4248 u32 size = new_size - found_key.offset;
4250 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4251 inode_sub_bytes(inode, item_end + 1 -
4255 * update the ram bytes to properly reflect
4256 * the new size of our item
4258 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4260 btrfs_file_extent_calc_inline_size(size);
4261 btrfs_truncate_item(root, path, size, 1);
4262 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4264 inode_sub_bytes(inode, item_end + 1 -
4270 if (!pending_del_nr) {
4271 /* no pending yet, add ourselves */
4272 pending_del_slot = path->slots[0];
4274 } else if (pending_del_nr &&
4275 path->slots[0] + 1 == pending_del_slot) {
4276 /* hop on the pending chunk */
4278 pending_del_slot = path->slots[0];
4286 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4287 root == root->fs_info->tree_root)) {
4288 btrfs_set_path_blocking(path);
4289 ret = btrfs_free_extent(trans, root, extent_start,
4290 extent_num_bytes, 0,
4291 btrfs_header_owner(leaf),
4292 ino, extent_offset, 0);
4296 if (found_type == BTRFS_INODE_ITEM_KEY)
4299 if (path->slots[0] == 0 ||
4300 path->slots[0] != pending_del_slot) {
4301 if (pending_del_nr) {
4302 ret = btrfs_del_items(trans, root, path,
4306 btrfs_abort_transaction(trans,
4312 btrfs_release_path(path);
4319 if (pending_del_nr) {
4320 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4323 btrfs_abort_transaction(trans, root, ret);
4326 if (last_size != (u64)-1 &&
4327 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4328 btrfs_ordered_update_i_size(inode, last_size, NULL);
4329 btrfs_free_path(path);
4334 * btrfs_truncate_page - read, zero a chunk and write a page
4335 * @inode - inode that we're zeroing
4336 * @from - the offset to start zeroing
4337 * @len - the length to zero, 0 to zero the entire range respective to the
4339 * @front - zero up to the offset instead of from the offset on
4341 * This will find the page for the "from" offset and cow the page and zero the
4342 * part we want to zero. This is used with truncate and hole punching.
4344 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4347 struct address_space *mapping = inode->i_mapping;
4348 struct btrfs_root *root = BTRFS_I(inode)->root;
4349 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4350 struct btrfs_ordered_extent *ordered;
4351 struct extent_state *cached_state = NULL;
4353 u32 blocksize = root->sectorsize;
4354 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4355 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4357 gfp_t mask = btrfs_alloc_write_mask(mapping);
4362 if ((offset & (blocksize - 1)) == 0 &&
4363 (!len || ((len & (blocksize - 1)) == 0)))
4365 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4370 page = find_or_create_page(mapping, index, mask);
4372 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4377 page_start = page_offset(page);
4378 page_end = page_start + PAGE_CACHE_SIZE - 1;
4380 if (!PageUptodate(page)) {
4381 ret = btrfs_readpage(NULL, page);
4383 if (page->mapping != mapping) {
4385 page_cache_release(page);
4388 if (!PageUptodate(page)) {
4393 wait_on_page_writeback(page);
4395 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4396 set_page_extent_mapped(page);
4398 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4400 unlock_extent_cached(io_tree, page_start, page_end,
4401 &cached_state, GFP_NOFS);
4403 page_cache_release(page);
4404 btrfs_start_ordered_extent(inode, ordered, 1);
4405 btrfs_put_ordered_extent(ordered);
4409 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4410 EXTENT_DIRTY | EXTENT_DELALLOC |
4411 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4412 0, 0, &cached_state, GFP_NOFS);
4414 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4417 unlock_extent_cached(io_tree, page_start, page_end,
4418 &cached_state, GFP_NOFS);
4422 if (offset != PAGE_CACHE_SIZE) {
4424 len = PAGE_CACHE_SIZE - offset;
4427 memset(kaddr, 0, offset);
4429 memset(kaddr + offset, 0, len);
4430 flush_dcache_page(page);
4433 ClearPageChecked(page);
4434 set_page_dirty(page);
4435 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4440 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4442 page_cache_release(page);
4447 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4448 u64 offset, u64 len)
4450 struct btrfs_trans_handle *trans;
4454 * Still need to make sure the inode looks like it's been updated so
4455 * that any holes get logged if we fsync.
4457 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4458 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4459 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4460 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4465 * 1 - for the one we're dropping
4466 * 1 - for the one we're adding
4467 * 1 - for updating the inode.
4469 trans = btrfs_start_transaction(root, 3);
4471 return PTR_ERR(trans);
4473 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4475 btrfs_abort_transaction(trans, root, ret);
4476 btrfs_end_transaction(trans, root);
4480 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4481 0, 0, len, 0, len, 0, 0, 0);
4483 btrfs_abort_transaction(trans, root, ret);
4485 btrfs_update_inode(trans, root, inode);
4486 btrfs_end_transaction(trans, root);
4491 * This function puts in dummy file extents for the area we're creating a hole
4492 * for. So if we are truncating this file to a larger size we need to insert
4493 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4494 * the range between oldsize and size
4496 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4498 struct btrfs_root *root = BTRFS_I(inode)->root;
4499 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4500 struct extent_map *em = NULL;
4501 struct extent_state *cached_state = NULL;
4502 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4503 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4504 u64 block_end = ALIGN(size, root->sectorsize);
4511 * If our size started in the middle of a page we need to zero out the
4512 * rest of the page before we expand the i_size, otherwise we could
4513 * expose stale data.
4515 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4519 if (size <= hole_start)
4523 struct btrfs_ordered_extent *ordered;
4525 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4527 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4528 block_end - hole_start);
4531 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4532 &cached_state, GFP_NOFS);
4533 btrfs_start_ordered_extent(inode, ordered, 1);
4534 btrfs_put_ordered_extent(ordered);
4537 cur_offset = hole_start;
4539 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4540 block_end - cur_offset, 0);
4546 last_byte = min(extent_map_end(em), block_end);
4547 last_byte = ALIGN(last_byte , root->sectorsize);
4548 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4549 struct extent_map *hole_em;
4550 hole_size = last_byte - cur_offset;
4552 err = maybe_insert_hole(root, inode, cur_offset,
4556 btrfs_drop_extent_cache(inode, cur_offset,
4557 cur_offset + hole_size - 1, 0);
4558 hole_em = alloc_extent_map();
4560 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4561 &BTRFS_I(inode)->runtime_flags);
4564 hole_em->start = cur_offset;
4565 hole_em->len = hole_size;
4566 hole_em->orig_start = cur_offset;
4568 hole_em->block_start = EXTENT_MAP_HOLE;
4569 hole_em->block_len = 0;
4570 hole_em->orig_block_len = 0;
4571 hole_em->ram_bytes = hole_size;
4572 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4573 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4574 hole_em->generation = root->fs_info->generation;
4577 write_lock(&em_tree->lock);
4578 err = add_extent_mapping(em_tree, hole_em, 1);
4579 write_unlock(&em_tree->lock);
4582 btrfs_drop_extent_cache(inode, cur_offset,
4586 free_extent_map(hole_em);
4589 free_extent_map(em);
4591 cur_offset = last_byte;
4592 if (cur_offset >= block_end)
4595 free_extent_map(em);
4596 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4601 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4603 struct btrfs_root *root = BTRFS_I(inode)->root;
4604 struct btrfs_trans_handle *trans;
4605 loff_t oldsize = i_size_read(inode);
4606 loff_t newsize = attr->ia_size;
4607 int mask = attr->ia_valid;
4611 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4612 * special case where we need to update the times despite not having
4613 * these flags set. For all other operations the VFS set these flags
4614 * explicitly if it wants a timestamp update.
4616 if (newsize != oldsize) {
4617 inode_inc_iversion(inode);
4618 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4619 inode->i_ctime = inode->i_mtime =
4620 current_fs_time(inode->i_sb);
4623 if (newsize > oldsize) {
4624 truncate_pagecache(inode, newsize);
4625 ret = btrfs_cont_expand(inode, oldsize, newsize);
4629 trans = btrfs_start_transaction(root, 1);
4631 return PTR_ERR(trans);
4633 i_size_write(inode, newsize);
4634 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4635 ret = btrfs_update_inode(trans, root, inode);
4636 btrfs_end_transaction(trans, root);
4640 * We're truncating a file that used to have good data down to
4641 * zero. Make sure it gets into the ordered flush list so that
4642 * any new writes get down to disk quickly.
4645 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4646 &BTRFS_I(inode)->runtime_flags);
4649 * 1 for the orphan item we're going to add
4650 * 1 for the orphan item deletion.
4652 trans = btrfs_start_transaction(root, 2);
4654 return PTR_ERR(trans);
4657 * We need to do this in case we fail at _any_ point during the
4658 * actual truncate. Once we do the truncate_setsize we could
4659 * invalidate pages which forces any outstanding ordered io to
4660 * be instantly completed which will give us extents that need
4661 * to be truncated. If we fail to get an orphan inode down we
4662 * could have left over extents that were never meant to live,
4663 * so we need to garuntee from this point on that everything
4664 * will be consistent.
4666 ret = btrfs_orphan_add(trans, inode);
4667 btrfs_end_transaction(trans, root);
4671 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4672 truncate_setsize(inode, newsize);
4674 /* Disable nonlocked read DIO to avoid the end less truncate */
4675 btrfs_inode_block_unlocked_dio(inode);
4676 inode_dio_wait(inode);
4677 btrfs_inode_resume_unlocked_dio(inode);
4679 ret = btrfs_truncate(inode);
4680 if (ret && inode->i_nlink) {
4684 * failed to truncate, disk_i_size is only adjusted down
4685 * as we remove extents, so it should represent the true
4686 * size of the inode, so reset the in memory size and
4687 * delete our orphan entry.
4689 trans = btrfs_join_transaction(root);
4690 if (IS_ERR(trans)) {
4691 btrfs_orphan_del(NULL, inode);
4694 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4695 err = btrfs_orphan_del(trans, inode);
4697 btrfs_abort_transaction(trans, root, err);
4698 btrfs_end_transaction(trans, root);
4705 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4707 struct inode *inode = dentry->d_inode;
4708 struct btrfs_root *root = BTRFS_I(inode)->root;
4711 if (btrfs_root_readonly(root))
4714 err = inode_change_ok(inode, attr);
4718 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4719 err = btrfs_setsize(inode, attr);
4724 if (attr->ia_valid) {
4725 setattr_copy(inode, attr);
4726 inode_inc_iversion(inode);
4727 err = btrfs_dirty_inode(inode);
4729 if (!err && attr->ia_valid & ATTR_MODE)
4730 err = posix_acl_chmod(inode, inode->i_mode);
4737 * While truncating the inode pages during eviction, we get the VFS calling
4738 * btrfs_invalidatepage() against each page of the inode. This is slow because
4739 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4740 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4741 * extent_state structures over and over, wasting lots of time.
4743 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4744 * those expensive operations on a per page basis and do only the ordered io
4745 * finishing, while we release here the extent_map and extent_state structures,
4746 * without the excessive merging and splitting.
4748 static void evict_inode_truncate_pages(struct inode *inode)
4750 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4751 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4752 struct rb_node *node;
4754 ASSERT(inode->i_state & I_FREEING);
4755 truncate_inode_pages_final(&inode->i_data);
4757 write_lock(&map_tree->lock);
4758 while (!RB_EMPTY_ROOT(&map_tree->map)) {
4759 struct extent_map *em;
4761 node = rb_first(&map_tree->map);
4762 em = rb_entry(node, struct extent_map, rb_node);
4763 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4764 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4765 remove_extent_mapping(map_tree, em);
4766 free_extent_map(em);
4767 if (need_resched()) {
4768 write_unlock(&map_tree->lock);
4770 write_lock(&map_tree->lock);
4773 write_unlock(&map_tree->lock);
4775 spin_lock(&io_tree->lock);
4776 while (!RB_EMPTY_ROOT(&io_tree->state)) {
4777 struct extent_state *state;
4778 struct extent_state *cached_state = NULL;
4780 node = rb_first(&io_tree->state);
4781 state = rb_entry(node, struct extent_state, rb_node);
4782 atomic_inc(&state->refs);
4783 spin_unlock(&io_tree->lock);
4785 lock_extent_bits(io_tree, state->start, state->end,
4787 clear_extent_bit(io_tree, state->start, state->end,
4788 EXTENT_LOCKED | EXTENT_DIRTY |
4789 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
4790 EXTENT_DEFRAG, 1, 1,
4791 &cached_state, GFP_NOFS);
4792 free_extent_state(state);
4795 spin_lock(&io_tree->lock);
4797 spin_unlock(&io_tree->lock);
4800 void btrfs_evict_inode(struct inode *inode)
4802 struct btrfs_trans_handle *trans;
4803 struct btrfs_root *root = BTRFS_I(inode)->root;
4804 struct btrfs_block_rsv *rsv, *global_rsv;
4805 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4808 trace_btrfs_inode_evict(inode);
4810 evict_inode_truncate_pages(inode);
4812 if (inode->i_nlink &&
4813 ((btrfs_root_refs(&root->root_item) != 0 &&
4814 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4815 btrfs_is_free_space_inode(inode)))
4818 if (is_bad_inode(inode)) {
4819 btrfs_orphan_del(NULL, inode);
4822 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4823 btrfs_wait_ordered_range(inode, 0, (u64)-1);
4825 btrfs_free_io_failure_record(inode, 0, (u64)-1);
4827 if (root->fs_info->log_root_recovering) {
4828 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4829 &BTRFS_I(inode)->runtime_flags));
4833 if (inode->i_nlink > 0) {
4834 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4835 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
4839 ret = btrfs_commit_inode_delayed_inode(inode);
4841 btrfs_orphan_del(NULL, inode);
4845 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4847 btrfs_orphan_del(NULL, inode);
4850 rsv->size = min_size;
4852 global_rsv = &root->fs_info->global_block_rsv;
4854 btrfs_i_size_write(inode, 0);
4857 * This is a bit simpler than btrfs_truncate since we've already
4858 * reserved our space for our orphan item in the unlink, so we just
4859 * need to reserve some slack space in case we add bytes and update
4860 * inode item when doing the truncate.
4863 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4864 BTRFS_RESERVE_FLUSH_LIMIT);
4867 * Try and steal from the global reserve since we will
4868 * likely not use this space anyway, we want to try as
4869 * hard as possible to get this to work.
4872 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4875 btrfs_warn(root->fs_info,
4876 "Could not get space for a delete, will truncate on mount %d",
4878 btrfs_orphan_del(NULL, inode);
4879 btrfs_free_block_rsv(root, rsv);
4883 trans = btrfs_join_transaction(root);
4884 if (IS_ERR(trans)) {
4885 btrfs_orphan_del(NULL, inode);
4886 btrfs_free_block_rsv(root, rsv);
4890 trans->block_rsv = rsv;
4892 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4896 trans->block_rsv = &root->fs_info->trans_block_rsv;
4897 btrfs_end_transaction(trans, root);
4899 btrfs_btree_balance_dirty(root);
4902 btrfs_free_block_rsv(root, rsv);
4905 * Errors here aren't a big deal, it just means we leave orphan items
4906 * in the tree. They will be cleaned up on the next mount.
4909 trans->block_rsv = root->orphan_block_rsv;
4910 btrfs_orphan_del(trans, inode);
4912 btrfs_orphan_del(NULL, inode);
4915 trans->block_rsv = &root->fs_info->trans_block_rsv;
4916 if (!(root == root->fs_info->tree_root ||
4917 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4918 btrfs_return_ino(root, btrfs_ino(inode));
4920 btrfs_end_transaction(trans, root);
4921 btrfs_btree_balance_dirty(root);
4923 btrfs_remove_delayed_node(inode);
4929 * this returns the key found in the dir entry in the location pointer.
4930 * If no dir entries were found, location->objectid is 0.
4932 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4933 struct btrfs_key *location)
4935 const char *name = dentry->d_name.name;
4936 int namelen = dentry->d_name.len;
4937 struct btrfs_dir_item *di;
4938 struct btrfs_path *path;
4939 struct btrfs_root *root = BTRFS_I(dir)->root;
4942 path = btrfs_alloc_path();
4946 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4951 if (IS_ERR_OR_NULL(di))
4954 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4956 btrfs_free_path(path);
4959 location->objectid = 0;
4964 * when we hit a tree root in a directory, the btrfs part of the inode
4965 * needs to be changed to reflect the root directory of the tree root. This
4966 * is kind of like crossing a mount point.
4968 static int fixup_tree_root_location(struct btrfs_root *root,
4970 struct dentry *dentry,
4971 struct btrfs_key *location,
4972 struct btrfs_root **sub_root)
4974 struct btrfs_path *path;
4975 struct btrfs_root *new_root;
4976 struct btrfs_root_ref *ref;
4977 struct extent_buffer *leaf;
4981 path = btrfs_alloc_path();
4988 ret = btrfs_find_item(root->fs_info->tree_root, path,
4989 BTRFS_I(dir)->root->root_key.objectid,
4990 location->objectid, BTRFS_ROOT_REF_KEY, NULL);
4997 leaf = path->nodes[0];
4998 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4999 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5000 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5003 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5004 (unsigned long)(ref + 1),
5005 dentry->d_name.len);
5009 btrfs_release_path(path);
5011 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5012 if (IS_ERR(new_root)) {
5013 err = PTR_ERR(new_root);
5017 *sub_root = new_root;
5018 location->objectid = btrfs_root_dirid(&new_root->root_item);
5019 location->type = BTRFS_INODE_ITEM_KEY;
5020 location->offset = 0;
5023 btrfs_free_path(path);
5027 static void inode_tree_add(struct inode *inode)
5029 struct btrfs_root *root = BTRFS_I(inode)->root;
5030 struct btrfs_inode *entry;
5032 struct rb_node *parent;
5033 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5034 u64 ino = btrfs_ino(inode);
5036 if (inode_unhashed(inode))
5039 spin_lock(&root->inode_lock);
5040 p = &root->inode_tree.rb_node;
5043 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5045 if (ino < btrfs_ino(&entry->vfs_inode))
5046 p = &parent->rb_left;
5047 else if (ino > btrfs_ino(&entry->vfs_inode))
5048 p = &parent->rb_right;
5050 WARN_ON(!(entry->vfs_inode.i_state &
5051 (I_WILL_FREE | I_FREEING)));
5052 rb_replace_node(parent, new, &root->inode_tree);
5053 RB_CLEAR_NODE(parent);
5054 spin_unlock(&root->inode_lock);
5058 rb_link_node(new, parent, p);
5059 rb_insert_color(new, &root->inode_tree);
5060 spin_unlock(&root->inode_lock);
5063 static void inode_tree_del(struct inode *inode)
5065 struct btrfs_root *root = BTRFS_I(inode)->root;
5068 spin_lock(&root->inode_lock);
5069 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5070 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5071 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5072 empty = RB_EMPTY_ROOT(&root->inode_tree);
5074 spin_unlock(&root->inode_lock);
5076 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5077 synchronize_srcu(&root->fs_info->subvol_srcu);
5078 spin_lock(&root->inode_lock);
5079 empty = RB_EMPTY_ROOT(&root->inode_tree);
5080 spin_unlock(&root->inode_lock);
5082 btrfs_add_dead_root(root);
5086 void btrfs_invalidate_inodes(struct btrfs_root *root)
5088 struct rb_node *node;
5089 struct rb_node *prev;
5090 struct btrfs_inode *entry;
5091 struct inode *inode;
5094 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5095 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5097 spin_lock(&root->inode_lock);
5099 node = root->inode_tree.rb_node;
5103 entry = rb_entry(node, struct btrfs_inode, rb_node);
5105 if (objectid < btrfs_ino(&entry->vfs_inode))
5106 node = node->rb_left;
5107 else if (objectid > btrfs_ino(&entry->vfs_inode))
5108 node = node->rb_right;
5114 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5115 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5119 prev = rb_next(prev);
5123 entry = rb_entry(node, struct btrfs_inode, rb_node);
5124 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5125 inode = igrab(&entry->vfs_inode);
5127 spin_unlock(&root->inode_lock);
5128 if (atomic_read(&inode->i_count) > 1)
5129 d_prune_aliases(inode);
5131 * btrfs_drop_inode will have it removed from
5132 * the inode cache when its usage count
5137 spin_lock(&root->inode_lock);
5141 if (cond_resched_lock(&root->inode_lock))
5144 node = rb_next(node);
5146 spin_unlock(&root->inode_lock);
5149 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5151 struct btrfs_iget_args *args = p;
5152 inode->i_ino = args->location->objectid;
5153 memcpy(&BTRFS_I(inode)->location, args->location,
5154 sizeof(*args->location));
5155 BTRFS_I(inode)->root = args->root;
5159 static int btrfs_find_actor(struct inode *inode, void *opaque)
5161 struct btrfs_iget_args *args = opaque;
5162 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5163 args->root == BTRFS_I(inode)->root;
5166 static struct inode *btrfs_iget_locked(struct super_block *s,
5167 struct btrfs_key *location,
5168 struct btrfs_root *root)
5170 struct inode *inode;
5171 struct btrfs_iget_args args;
5172 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5174 args.location = location;
5177 inode = iget5_locked(s, hashval, btrfs_find_actor,
5178 btrfs_init_locked_inode,
5183 /* Get an inode object given its location and corresponding root.
5184 * Returns in *is_new if the inode was read from disk
5186 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5187 struct btrfs_root *root, int *new)
5189 struct inode *inode;
5191 inode = btrfs_iget_locked(s, location, root);
5193 return ERR_PTR(-ENOMEM);
5195 if (inode->i_state & I_NEW) {
5196 btrfs_read_locked_inode(inode);
5197 if (!is_bad_inode(inode)) {
5198 inode_tree_add(inode);
5199 unlock_new_inode(inode);
5203 unlock_new_inode(inode);
5205 inode = ERR_PTR(-ESTALE);
5212 static struct inode *new_simple_dir(struct super_block *s,
5213 struct btrfs_key *key,
5214 struct btrfs_root *root)
5216 struct inode *inode = new_inode(s);
5219 return ERR_PTR(-ENOMEM);
5221 BTRFS_I(inode)->root = root;
5222 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5223 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5225 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5226 inode->i_op = &btrfs_dir_ro_inode_operations;
5227 inode->i_fop = &simple_dir_operations;
5228 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5229 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5234 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5236 struct inode *inode;
5237 struct btrfs_root *root = BTRFS_I(dir)->root;
5238 struct btrfs_root *sub_root = root;
5239 struct btrfs_key location;
5243 if (dentry->d_name.len > BTRFS_NAME_LEN)
5244 return ERR_PTR(-ENAMETOOLONG);
5246 ret = btrfs_inode_by_name(dir, dentry, &location);
5248 return ERR_PTR(ret);
5250 if (location.objectid == 0)
5251 return ERR_PTR(-ENOENT);
5253 if (location.type == BTRFS_INODE_ITEM_KEY) {
5254 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5258 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5260 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5261 ret = fixup_tree_root_location(root, dir, dentry,
5262 &location, &sub_root);
5265 inode = ERR_PTR(ret);
5267 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5269 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5271 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5273 if (!IS_ERR(inode) && root != sub_root) {
5274 down_read(&root->fs_info->cleanup_work_sem);
5275 if (!(inode->i_sb->s_flags & MS_RDONLY))
5276 ret = btrfs_orphan_cleanup(sub_root);
5277 up_read(&root->fs_info->cleanup_work_sem);
5280 inode = ERR_PTR(ret);
5287 static int btrfs_dentry_delete(const struct dentry *dentry)
5289 struct btrfs_root *root;
5290 struct inode *inode = dentry->d_inode;
5292 if (!inode && !IS_ROOT(dentry))
5293 inode = dentry->d_parent->d_inode;
5296 root = BTRFS_I(inode)->root;
5297 if (btrfs_root_refs(&root->root_item) == 0)
5300 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5306 static void btrfs_dentry_release(struct dentry *dentry)
5308 kfree(dentry->d_fsdata);
5311 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5314 struct inode *inode;
5316 inode = btrfs_lookup_dentry(dir, dentry);
5317 if (IS_ERR(inode)) {
5318 if (PTR_ERR(inode) == -ENOENT)
5321 return ERR_CAST(inode);
5324 return d_materialise_unique(dentry, inode);
5327 unsigned char btrfs_filetype_table[] = {
5328 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5331 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5333 struct inode *inode = file_inode(file);
5334 struct btrfs_root *root = BTRFS_I(inode)->root;
5335 struct btrfs_item *item;
5336 struct btrfs_dir_item *di;
5337 struct btrfs_key key;
5338 struct btrfs_key found_key;
5339 struct btrfs_path *path;
5340 struct list_head ins_list;
5341 struct list_head del_list;
5343 struct extent_buffer *leaf;
5345 unsigned char d_type;
5350 int key_type = BTRFS_DIR_INDEX_KEY;
5354 int is_curr = 0; /* ctx->pos points to the current index? */
5356 /* FIXME, use a real flag for deciding about the key type */
5357 if (root->fs_info->tree_root == root)
5358 key_type = BTRFS_DIR_ITEM_KEY;
5360 if (!dir_emit_dots(file, ctx))
5363 path = btrfs_alloc_path();
5369 if (key_type == BTRFS_DIR_INDEX_KEY) {
5370 INIT_LIST_HEAD(&ins_list);
5371 INIT_LIST_HEAD(&del_list);
5372 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5375 key.type = key_type;
5376 key.offset = ctx->pos;
5377 key.objectid = btrfs_ino(inode);
5379 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5384 leaf = path->nodes[0];
5385 slot = path->slots[0];
5386 if (slot >= btrfs_header_nritems(leaf)) {
5387 ret = btrfs_next_leaf(root, path);
5395 item = btrfs_item_nr(slot);
5396 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5398 if (found_key.objectid != key.objectid)
5400 if (found_key.type != key_type)
5402 if (found_key.offset < ctx->pos)
5404 if (key_type == BTRFS_DIR_INDEX_KEY &&
5405 btrfs_should_delete_dir_index(&del_list,
5409 ctx->pos = found_key.offset;
5412 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5414 di_total = btrfs_item_size(leaf, item);
5416 while (di_cur < di_total) {
5417 struct btrfs_key location;
5419 if (verify_dir_item(root, leaf, di))
5422 name_len = btrfs_dir_name_len(leaf, di);
5423 if (name_len <= sizeof(tmp_name)) {
5424 name_ptr = tmp_name;
5426 name_ptr = kmalloc(name_len, GFP_NOFS);
5432 read_extent_buffer(leaf, name_ptr,
5433 (unsigned long)(di + 1), name_len);
5435 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5436 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5439 /* is this a reference to our own snapshot? If so
5442 * In contrast to old kernels, we insert the snapshot's
5443 * dir item and dir index after it has been created, so
5444 * we won't find a reference to our own snapshot. We
5445 * still keep the following code for backward
5448 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5449 location.objectid == root->root_key.objectid) {
5453 over = !dir_emit(ctx, name_ptr, name_len,
5454 location.objectid, d_type);
5457 if (name_ptr != tmp_name)
5462 di_len = btrfs_dir_name_len(leaf, di) +
5463 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5465 di = (struct btrfs_dir_item *)((char *)di + di_len);
5471 if (key_type == BTRFS_DIR_INDEX_KEY) {
5474 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5479 /* Reached end of directory/root. Bump pos past the last item. */
5483 * Stop new entries from being returned after we return the last
5486 * New directory entries are assigned a strictly increasing
5487 * offset. This means that new entries created during readdir
5488 * are *guaranteed* to be seen in the future by that readdir.
5489 * This has broken buggy programs which operate on names as
5490 * they're returned by readdir. Until we re-use freed offsets
5491 * we have this hack to stop new entries from being returned
5492 * under the assumption that they'll never reach this huge
5495 * This is being careful not to overflow 32bit loff_t unless the
5496 * last entry requires it because doing so has broken 32bit apps
5499 if (key_type == BTRFS_DIR_INDEX_KEY) {
5500 if (ctx->pos >= INT_MAX)
5501 ctx->pos = LLONG_MAX;
5508 if (key_type == BTRFS_DIR_INDEX_KEY)
5509 btrfs_put_delayed_items(&ins_list, &del_list);
5510 btrfs_free_path(path);
5514 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5516 struct btrfs_root *root = BTRFS_I(inode)->root;
5517 struct btrfs_trans_handle *trans;
5519 bool nolock = false;
5521 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5524 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5527 if (wbc->sync_mode == WB_SYNC_ALL) {
5529 trans = btrfs_join_transaction_nolock(root);
5531 trans = btrfs_join_transaction(root);
5533 return PTR_ERR(trans);
5534 ret = btrfs_commit_transaction(trans, root);
5540 * This is somewhat expensive, updating the tree every time the
5541 * inode changes. But, it is most likely to find the inode in cache.
5542 * FIXME, needs more benchmarking...there are no reasons other than performance
5543 * to keep or drop this code.
5545 static int btrfs_dirty_inode(struct inode *inode)
5547 struct btrfs_root *root = BTRFS_I(inode)->root;
5548 struct btrfs_trans_handle *trans;
5551 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5554 trans = btrfs_join_transaction(root);
5556 return PTR_ERR(trans);
5558 ret = btrfs_update_inode(trans, root, inode);
5559 if (ret && ret == -ENOSPC) {
5560 /* whoops, lets try again with the full transaction */
5561 btrfs_end_transaction(trans, root);
5562 trans = btrfs_start_transaction(root, 1);
5564 return PTR_ERR(trans);
5566 ret = btrfs_update_inode(trans, root, inode);
5568 btrfs_end_transaction(trans, root);
5569 if (BTRFS_I(inode)->delayed_node)
5570 btrfs_balance_delayed_items(root);
5576 * This is a copy of file_update_time. We need this so we can return error on
5577 * ENOSPC for updating the inode in the case of file write and mmap writes.
5579 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5582 struct btrfs_root *root = BTRFS_I(inode)->root;
5584 if (btrfs_root_readonly(root))
5587 if (flags & S_VERSION)
5588 inode_inc_iversion(inode);
5589 if (flags & S_CTIME)
5590 inode->i_ctime = *now;
5591 if (flags & S_MTIME)
5592 inode->i_mtime = *now;
5593 if (flags & S_ATIME)
5594 inode->i_atime = *now;
5595 return btrfs_dirty_inode(inode);
5599 * find the highest existing sequence number in a directory
5600 * and then set the in-memory index_cnt variable to reflect
5601 * free sequence numbers
5603 static int btrfs_set_inode_index_count(struct inode *inode)
5605 struct btrfs_root *root = BTRFS_I(inode)->root;
5606 struct btrfs_key key, found_key;
5607 struct btrfs_path *path;
5608 struct extent_buffer *leaf;
5611 key.objectid = btrfs_ino(inode);
5612 key.type = BTRFS_DIR_INDEX_KEY;
5613 key.offset = (u64)-1;
5615 path = btrfs_alloc_path();
5619 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5622 /* FIXME: we should be able to handle this */
5628 * MAGIC NUMBER EXPLANATION:
5629 * since we search a directory based on f_pos we have to start at 2
5630 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5631 * else has to start at 2
5633 if (path->slots[0] == 0) {
5634 BTRFS_I(inode)->index_cnt = 2;
5640 leaf = path->nodes[0];
5641 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5643 if (found_key.objectid != btrfs_ino(inode) ||
5644 found_key.type != BTRFS_DIR_INDEX_KEY) {
5645 BTRFS_I(inode)->index_cnt = 2;
5649 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5651 btrfs_free_path(path);
5656 * helper to find a free sequence number in a given directory. This current
5657 * code is very simple, later versions will do smarter things in the btree
5659 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5663 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5664 ret = btrfs_inode_delayed_dir_index_count(dir);
5666 ret = btrfs_set_inode_index_count(dir);
5672 *index = BTRFS_I(dir)->index_cnt;
5673 BTRFS_I(dir)->index_cnt++;
5678 static int btrfs_insert_inode_locked(struct inode *inode)
5680 struct btrfs_iget_args args;
5681 args.location = &BTRFS_I(inode)->location;
5682 args.root = BTRFS_I(inode)->root;
5684 return insert_inode_locked4(inode,
5685 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
5686 btrfs_find_actor, &args);
5689 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5690 struct btrfs_root *root,
5692 const char *name, int name_len,
5693 u64 ref_objectid, u64 objectid,
5694 umode_t mode, u64 *index)
5696 struct inode *inode;
5697 struct btrfs_inode_item *inode_item;
5698 struct btrfs_key *location;
5699 struct btrfs_path *path;
5700 struct btrfs_inode_ref *ref;
5701 struct btrfs_key key[2];
5703 int nitems = name ? 2 : 1;
5707 path = btrfs_alloc_path();
5709 return ERR_PTR(-ENOMEM);
5711 inode = new_inode(root->fs_info->sb);
5713 btrfs_free_path(path);
5714 return ERR_PTR(-ENOMEM);
5718 * O_TMPFILE, set link count to 0, so that after this point,
5719 * we fill in an inode item with the correct link count.
5722 set_nlink(inode, 0);
5725 * we have to initialize this early, so we can reclaim the inode
5726 * number if we fail afterwards in this function.
5728 inode->i_ino = objectid;
5731 trace_btrfs_inode_request(dir);
5733 ret = btrfs_set_inode_index(dir, index);
5735 btrfs_free_path(path);
5737 return ERR_PTR(ret);
5743 * index_cnt is ignored for everything but a dir,
5744 * btrfs_get_inode_index_count has an explanation for the magic
5747 BTRFS_I(inode)->index_cnt = 2;
5748 BTRFS_I(inode)->dir_index = *index;
5749 BTRFS_I(inode)->root = root;
5750 BTRFS_I(inode)->generation = trans->transid;
5751 inode->i_generation = BTRFS_I(inode)->generation;
5754 * We could have gotten an inode number from somebody who was fsynced
5755 * and then removed in this same transaction, so let's just set full
5756 * sync since it will be a full sync anyway and this will blow away the
5757 * old info in the log.
5759 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5761 key[0].objectid = objectid;
5762 key[0].type = BTRFS_INODE_ITEM_KEY;
5765 sizes[0] = sizeof(struct btrfs_inode_item);
5769 * Start new inodes with an inode_ref. This is slightly more
5770 * efficient for small numbers of hard links since they will
5771 * be packed into one item. Extended refs will kick in if we
5772 * add more hard links than can fit in the ref item.
5774 key[1].objectid = objectid;
5775 key[1].type = BTRFS_INODE_REF_KEY;
5776 key[1].offset = ref_objectid;
5778 sizes[1] = name_len + sizeof(*ref);
5781 location = &BTRFS_I(inode)->location;
5782 location->objectid = objectid;
5783 location->offset = 0;
5784 location->type = BTRFS_INODE_ITEM_KEY;
5786 ret = btrfs_insert_inode_locked(inode);
5790 path->leave_spinning = 1;
5791 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
5795 inode_init_owner(inode, dir, mode);
5796 inode_set_bytes(inode, 0);
5797 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5798 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5799 struct btrfs_inode_item);
5800 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5801 sizeof(*inode_item));
5802 fill_inode_item(trans, path->nodes[0], inode_item, inode);
5805 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5806 struct btrfs_inode_ref);
5807 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5808 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5809 ptr = (unsigned long)(ref + 1);
5810 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5813 btrfs_mark_buffer_dirty(path->nodes[0]);
5814 btrfs_free_path(path);
5816 btrfs_inherit_iflags(inode, dir);
5818 if (S_ISREG(mode)) {
5819 if (btrfs_test_opt(root, NODATASUM))
5820 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5821 if (btrfs_test_opt(root, NODATACOW))
5822 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5823 BTRFS_INODE_NODATASUM;
5826 inode_tree_add(inode);
5828 trace_btrfs_inode_new(inode);
5829 btrfs_set_inode_last_trans(trans, inode);
5831 btrfs_update_root_times(trans, root);
5833 ret = btrfs_inode_inherit_props(trans, inode, dir);
5835 btrfs_err(root->fs_info,
5836 "error inheriting props for ino %llu (root %llu): %d",
5837 btrfs_ino(inode), root->root_key.objectid, ret);
5842 unlock_new_inode(inode);
5845 BTRFS_I(dir)->index_cnt--;
5846 btrfs_free_path(path);
5848 return ERR_PTR(ret);
5851 static inline u8 btrfs_inode_type(struct inode *inode)
5853 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5857 * utility function to add 'inode' into 'parent_inode' with
5858 * a give name and a given sequence number.
5859 * if 'add_backref' is true, also insert a backref from the
5860 * inode to the parent directory.
5862 int btrfs_add_link(struct btrfs_trans_handle *trans,
5863 struct inode *parent_inode, struct inode *inode,
5864 const char *name, int name_len, int add_backref, u64 index)
5867 struct btrfs_key key;
5868 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5869 u64 ino = btrfs_ino(inode);
5870 u64 parent_ino = btrfs_ino(parent_inode);
5872 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5873 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5876 key.type = BTRFS_INODE_ITEM_KEY;
5880 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5881 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5882 key.objectid, root->root_key.objectid,
5883 parent_ino, index, name, name_len);
5884 } else if (add_backref) {
5885 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5889 /* Nothing to clean up yet */
5893 ret = btrfs_insert_dir_item(trans, root, name, name_len,
5895 btrfs_inode_type(inode), index);
5896 if (ret == -EEXIST || ret == -EOVERFLOW)
5899 btrfs_abort_transaction(trans, root, ret);
5903 btrfs_i_size_write(parent_inode, parent_inode->i_size +
5905 inode_inc_iversion(parent_inode);
5906 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5907 ret = btrfs_update_inode(trans, root, parent_inode);
5909 btrfs_abort_transaction(trans, root, ret);
5913 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5916 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5917 key.objectid, root->root_key.objectid,
5918 parent_ino, &local_index, name, name_len);
5920 } else if (add_backref) {
5924 err = btrfs_del_inode_ref(trans, root, name, name_len,
5925 ino, parent_ino, &local_index);
5930 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5931 struct inode *dir, struct dentry *dentry,
5932 struct inode *inode, int backref, u64 index)
5934 int err = btrfs_add_link(trans, dir, inode,
5935 dentry->d_name.name, dentry->d_name.len,
5942 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5943 umode_t mode, dev_t rdev)
5945 struct btrfs_trans_handle *trans;
5946 struct btrfs_root *root = BTRFS_I(dir)->root;
5947 struct inode *inode = NULL;
5953 if (!new_valid_dev(rdev))
5957 * 2 for inode item and ref
5959 * 1 for xattr if selinux is on
5961 trans = btrfs_start_transaction(root, 5);
5963 return PTR_ERR(trans);
5965 err = btrfs_find_free_ino(root, &objectid);
5969 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5970 dentry->d_name.len, btrfs_ino(dir), objectid,
5972 if (IS_ERR(inode)) {
5973 err = PTR_ERR(inode);
5978 * If the active LSM wants to access the inode during
5979 * d_instantiate it needs these. Smack checks to see
5980 * if the filesystem supports xattrs by looking at the
5983 inode->i_op = &btrfs_special_inode_operations;
5984 init_special_inode(inode, inode->i_mode, rdev);
5986 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5988 goto out_unlock_inode;
5990 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5992 goto out_unlock_inode;
5994 btrfs_update_inode(trans, root, inode);
5995 unlock_new_inode(inode);
5996 d_instantiate(dentry, inode);
6000 btrfs_end_transaction(trans, root);
6001 btrfs_balance_delayed_items(root);
6002 btrfs_btree_balance_dirty(root);
6004 inode_dec_link_count(inode);
6011 unlock_new_inode(inode);
6016 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6017 umode_t mode, bool excl)
6019 struct btrfs_trans_handle *trans;
6020 struct btrfs_root *root = BTRFS_I(dir)->root;
6021 struct inode *inode = NULL;
6022 int drop_inode_on_err = 0;
6028 * 2 for inode item and ref
6030 * 1 for xattr if selinux is on
6032 trans = btrfs_start_transaction(root, 5);
6034 return PTR_ERR(trans);
6036 err = btrfs_find_free_ino(root, &objectid);
6040 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6041 dentry->d_name.len, btrfs_ino(dir), objectid,
6043 if (IS_ERR(inode)) {
6044 err = PTR_ERR(inode);
6047 drop_inode_on_err = 1;
6049 * If the active LSM wants to access the inode during
6050 * d_instantiate it needs these. Smack checks to see
6051 * if the filesystem supports xattrs by looking at the
6054 inode->i_fop = &btrfs_file_operations;
6055 inode->i_op = &btrfs_file_inode_operations;
6056 inode->i_mapping->a_ops = &btrfs_aops;
6057 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
6059 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6061 goto out_unlock_inode;
6063 err = btrfs_update_inode(trans, root, inode);
6065 goto out_unlock_inode;
6067 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6069 goto out_unlock_inode;
6071 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6072 unlock_new_inode(inode);
6073 d_instantiate(dentry, inode);
6076 btrfs_end_transaction(trans, root);
6077 if (err && drop_inode_on_err) {
6078 inode_dec_link_count(inode);
6081 btrfs_balance_delayed_items(root);
6082 btrfs_btree_balance_dirty(root);
6086 unlock_new_inode(inode);
6091 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6092 struct dentry *dentry)
6094 struct btrfs_trans_handle *trans;
6095 struct btrfs_root *root = BTRFS_I(dir)->root;
6096 struct inode *inode = old_dentry->d_inode;
6101 /* do not allow sys_link's with other subvols of the same device */
6102 if (root->objectid != BTRFS_I(inode)->root->objectid)
6105 if (inode->i_nlink >= BTRFS_LINK_MAX)
6108 err = btrfs_set_inode_index(dir, &index);
6113 * 2 items for inode and inode ref
6114 * 2 items for dir items
6115 * 1 item for parent inode
6117 trans = btrfs_start_transaction(root, 5);
6118 if (IS_ERR(trans)) {
6119 err = PTR_ERR(trans);
6123 /* There are several dir indexes for this inode, clear the cache. */
6124 BTRFS_I(inode)->dir_index = 0ULL;
6126 inode_inc_iversion(inode);
6127 inode->i_ctime = CURRENT_TIME;
6129 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6131 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6136 struct dentry *parent = dentry->d_parent;
6137 err = btrfs_update_inode(trans, root, inode);
6140 if (inode->i_nlink == 1) {
6142 * If new hard link count is 1, it's a file created
6143 * with open(2) O_TMPFILE flag.
6145 err = btrfs_orphan_del(trans, inode);
6149 d_instantiate(dentry, inode);
6150 btrfs_log_new_name(trans, inode, NULL, parent);
6153 btrfs_end_transaction(trans, root);
6154 btrfs_balance_delayed_items(root);
6157 inode_dec_link_count(inode);
6160 btrfs_btree_balance_dirty(root);
6164 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6166 struct inode *inode = NULL;
6167 struct btrfs_trans_handle *trans;
6168 struct btrfs_root *root = BTRFS_I(dir)->root;
6170 int drop_on_err = 0;
6175 * 2 items for inode and ref
6176 * 2 items for dir items
6177 * 1 for xattr if selinux is on
6179 trans = btrfs_start_transaction(root, 5);
6181 return PTR_ERR(trans);
6183 err = btrfs_find_free_ino(root, &objectid);
6187 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6188 dentry->d_name.len, btrfs_ino(dir), objectid,
6189 S_IFDIR | mode, &index);
6190 if (IS_ERR(inode)) {
6191 err = PTR_ERR(inode);
6196 /* these must be set before we unlock the inode */
6197 inode->i_op = &btrfs_dir_inode_operations;
6198 inode->i_fop = &btrfs_dir_file_operations;
6200 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6202 goto out_fail_inode;
6204 btrfs_i_size_write(inode, 0);
6205 err = btrfs_update_inode(trans, root, inode);
6207 goto out_fail_inode;
6209 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6210 dentry->d_name.len, 0, index);
6212 goto out_fail_inode;
6214 d_instantiate(dentry, inode);
6216 * mkdir is special. We're unlocking after we call d_instantiate
6217 * to avoid a race with nfsd calling d_instantiate.
6219 unlock_new_inode(inode);
6223 btrfs_end_transaction(trans, root);
6226 btrfs_balance_delayed_items(root);
6227 btrfs_btree_balance_dirty(root);
6231 unlock_new_inode(inode);
6235 /* Find next extent map of a given extent map, caller needs to ensure locks */
6236 static struct extent_map *next_extent_map(struct extent_map *em)
6238 struct rb_node *next;
6240 next = rb_next(&em->rb_node);
6243 return container_of(next, struct extent_map, rb_node);
6246 static struct extent_map *prev_extent_map(struct extent_map *em)
6248 struct rb_node *prev;
6250 prev = rb_prev(&em->rb_node);
6253 return container_of(prev, struct extent_map, rb_node);
6256 /* helper for btfs_get_extent. Given an existing extent in the tree,
6257 * the existing extent is the nearest extent to map_start,
6258 * and an extent that you want to insert, deal with overlap and insert
6259 * the best fitted new extent into the tree.
6261 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6262 struct extent_map *existing,
6263 struct extent_map *em,
6266 struct extent_map *prev;
6267 struct extent_map *next;
6272 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6274 if (existing->start > map_start) {
6276 prev = prev_extent_map(next);
6279 next = next_extent_map(prev);
6282 start = prev ? extent_map_end(prev) : em->start;
6283 start = max_t(u64, start, em->start);
6284 end = next ? next->start : extent_map_end(em);
6285 end = min_t(u64, end, extent_map_end(em));
6286 start_diff = start - em->start;
6288 em->len = end - start;
6289 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6290 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6291 em->block_start += start_diff;
6292 em->block_len -= start_diff;
6294 return add_extent_mapping(em_tree, em, 0);
6297 static noinline int uncompress_inline(struct btrfs_path *path,
6298 struct inode *inode, struct page *page,
6299 size_t pg_offset, u64 extent_offset,
6300 struct btrfs_file_extent_item *item)
6303 struct extent_buffer *leaf = path->nodes[0];
6306 unsigned long inline_size;
6310 WARN_ON(pg_offset != 0);
6311 compress_type = btrfs_file_extent_compression(leaf, item);
6312 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6313 inline_size = btrfs_file_extent_inline_item_len(leaf,
6314 btrfs_item_nr(path->slots[0]));
6315 tmp = kmalloc(inline_size, GFP_NOFS);
6318 ptr = btrfs_file_extent_inline_start(item);
6320 read_extent_buffer(leaf, tmp, ptr, inline_size);
6322 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6323 ret = btrfs_decompress(compress_type, tmp, page,
6324 extent_offset, inline_size, max_size);
6330 * a bit scary, this does extent mapping from logical file offset to the disk.
6331 * the ugly parts come from merging extents from the disk with the in-ram
6332 * representation. This gets more complex because of the data=ordered code,
6333 * where the in-ram extents might be locked pending data=ordered completion.
6335 * This also copies inline extents directly into the page.
6338 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6339 size_t pg_offset, u64 start, u64 len,
6344 u64 extent_start = 0;
6346 u64 objectid = btrfs_ino(inode);
6348 struct btrfs_path *path = NULL;
6349 struct btrfs_root *root = BTRFS_I(inode)->root;
6350 struct btrfs_file_extent_item *item;
6351 struct extent_buffer *leaf;
6352 struct btrfs_key found_key;
6353 struct extent_map *em = NULL;
6354 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6355 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6356 struct btrfs_trans_handle *trans = NULL;
6357 const bool new_inline = !page || create;
6360 read_lock(&em_tree->lock);
6361 em = lookup_extent_mapping(em_tree, start, len);
6363 em->bdev = root->fs_info->fs_devices->latest_bdev;
6364 read_unlock(&em_tree->lock);
6367 if (em->start > start || em->start + em->len <= start)
6368 free_extent_map(em);
6369 else if (em->block_start == EXTENT_MAP_INLINE && page)
6370 free_extent_map(em);
6374 em = alloc_extent_map();
6379 em->bdev = root->fs_info->fs_devices->latest_bdev;
6380 em->start = EXTENT_MAP_HOLE;
6381 em->orig_start = EXTENT_MAP_HOLE;
6383 em->block_len = (u64)-1;
6386 path = btrfs_alloc_path();
6392 * Chances are we'll be called again, so go ahead and do
6398 ret = btrfs_lookup_file_extent(trans, root, path,
6399 objectid, start, trans != NULL);
6406 if (path->slots[0] == 0)
6411 leaf = path->nodes[0];
6412 item = btrfs_item_ptr(leaf, path->slots[0],
6413 struct btrfs_file_extent_item);
6414 /* are we inside the extent that was found? */
6415 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6416 found_type = found_key.type;
6417 if (found_key.objectid != objectid ||
6418 found_type != BTRFS_EXTENT_DATA_KEY) {
6420 * If we backup past the first extent we want to move forward
6421 * and see if there is an extent in front of us, otherwise we'll
6422 * say there is a hole for our whole search range which can
6429 found_type = btrfs_file_extent_type(leaf, item);
6430 extent_start = found_key.offset;
6431 if (found_type == BTRFS_FILE_EXTENT_REG ||
6432 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6433 extent_end = extent_start +
6434 btrfs_file_extent_num_bytes(leaf, item);
6435 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6437 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6438 extent_end = ALIGN(extent_start + size, root->sectorsize);
6441 if (start >= extent_end) {
6443 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6444 ret = btrfs_next_leaf(root, path);
6451 leaf = path->nodes[0];
6453 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6454 if (found_key.objectid != objectid ||
6455 found_key.type != BTRFS_EXTENT_DATA_KEY)
6457 if (start + len <= found_key.offset)
6459 if (start > found_key.offset)
6462 em->orig_start = start;
6463 em->len = found_key.offset - start;
6467 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6469 if (found_type == BTRFS_FILE_EXTENT_REG ||
6470 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6472 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6476 size_t extent_offset;
6482 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6483 extent_offset = page_offset(page) + pg_offset - extent_start;
6484 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6485 size - extent_offset);
6486 em->start = extent_start + extent_offset;
6487 em->len = ALIGN(copy_size, root->sectorsize);
6488 em->orig_block_len = em->len;
6489 em->orig_start = em->start;
6490 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6491 if (create == 0 && !PageUptodate(page)) {
6492 if (btrfs_file_extent_compression(leaf, item) !=
6493 BTRFS_COMPRESS_NONE) {
6494 ret = uncompress_inline(path, inode, page,
6496 extent_offset, item);
6503 read_extent_buffer(leaf, map + pg_offset, ptr,
6505 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6506 memset(map + pg_offset + copy_size, 0,
6507 PAGE_CACHE_SIZE - pg_offset -
6512 flush_dcache_page(page);
6513 } else if (create && PageUptodate(page)) {
6517 free_extent_map(em);
6520 btrfs_release_path(path);
6521 trans = btrfs_join_transaction(root);
6524 return ERR_CAST(trans);
6528 write_extent_buffer(leaf, map + pg_offset, ptr,
6531 btrfs_mark_buffer_dirty(leaf);
6533 set_extent_uptodate(io_tree, em->start,
6534 extent_map_end(em) - 1, NULL, GFP_NOFS);
6539 em->orig_start = start;
6542 em->block_start = EXTENT_MAP_HOLE;
6543 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6545 btrfs_release_path(path);
6546 if (em->start > start || extent_map_end(em) <= start) {
6547 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6548 em->start, em->len, start, len);
6554 write_lock(&em_tree->lock);
6555 ret = add_extent_mapping(em_tree, em, 0);
6556 /* it is possible that someone inserted the extent into the tree
6557 * while we had the lock dropped. It is also possible that
6558 * an overlapping map exists in the tree
6560 if (ret == -EEXIST) {
6561 struct extent_map *existing;
6565 existing = search_extent_mapping(em_tree, start, len);
6567 * existing will always be non-NULL, since there must be
6568 * extent causing the -EEXIST.
6570 if (start >= extent_map_end(existing) ||
6571 start <= existing->start) {
6573 * The existing extent map is the one nearest to
6574 * the [start, start + len) range which overlaps
6576 err = merge_extent_mapping(em_tree, existing,
6578 free_extent_map(existing);
6580 free_extent_map(em);
6584 free_extent_map(em);
6589 write_unlock(&em_tree->lock);
6592 trace_btrfs_get_extent(root, em);
6595 btrfs_free_path(path);
6597 ret = btrfs_end_transaction(trans, root);
6602 free_extent_map(em);
6603 return ERR_PTR(err);
6605 BUG_ON(!em); /* Error is always set */
6609 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6610 size_t pg_offset, u64 start, u64 len,
6613 struct extent_map *em;
6614 struct extent_map *hole_em = NULL;
6615 u64 range_start = start;
6621 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6628 * - a pre-alloc extent,
6629 * there might actually be delalloc bytes behind it.
6631 if (em->block_start != EXTENT_MAP_HOLE &&
6632 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6638 /* check to see if we've wrapped (len == -1 or similar) */
6647 /* ok, we didn't find anything, lets look for delalloc */
6648 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6649 end, len, EXTENT_DELALLOC, 1);
6650 found_end = range_start + found;
6651 if (found_end < range_start)
6652 found_end = (u64)-1;
6655 * we didn't find anything useful, return
6656 * the original results from get_extent()
6658 if (range_start > end || found_end <= start) {
6664 /* adjust the range_start to make sure it doesn't
6665 * go backwards from the start they passed in
6667 range_start = max(start, range_start);
6668 found = found_end - range_start;
6671 u64 hole_start = start;
6674 em = alloc_extent_map();
6680 * when btrfs_get_extent can't find anything it
6681 * returns one huge hole
6683 * make sure what it found really fits our range, and
6684 * adjust to make sure it is based on the start from
6688 u64 calc_end = extent_map_end(hole_em);
6690 if (calc_end <= start || (hole_em->start > end)) {
6691 free_extent_map(hole_em);
6694 hole_start = max(hole_em->start, start);
6695 hole_len = calc_end - hole_start;
6699 if (hole_em && range_start > hole_start) {
6700 /* our hole starts before our delalloc, so we
6701 * have to return just the parts of the hole
6702 * that go until the delalloc starts
6704 em->len = min(hole_len,
6705 range_start - hole_start);
6706 em->start = hole_start;
6707 em->orig_start = hole_start;
6709 * don't adjust block start at all,
6710 * it is fixed at EXTENT_MAP_HOLE
6712 em->block_start = hole_em->block_start;
6713 em->block_len = hole_len;
6714 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6715 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6717 em->start = range_start;
6719 em->orig_start = range_start;
6720 em->block_start = EXTENT_MAP_DELALLOC;
6721 em->block_len = found;
6723 } else if (hole_em) {
6728 free_extent_map(hole_em);
6730 free_extent_map(em);
6731 return ERR_PTR(err);
6736 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6739 struct btrfs_root *root = BTRFS_I(inode)->root;
6740 struct extent_map *em;
6741 struct btrfs_key ins;
6745 alloc_hint = get_extent_allocation_hint(inode, start, len);
6746 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
6747 alloc_hint, &ins, 1, 1);
6749 return ERR_PTR(ret);
6751 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6752 ins.offset, ins.offset, ins.offset, 0);
6754 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
6758 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6759 ins.offset, ins.offset, 0);
6761 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
6762 free_extent_map(em);
6763 return ERR_PTR(ret);
6770 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6771 * block must be cow'd
6773 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6774 u64 *orig_start, u64 *orig_block_len,
6777 struct btrfs_trans_handle *trans;
6778 struct btrfs_path *path;
6780 struct extent_buffer *leaf;
6781 struct btrfs_root *root = BTRFS_I(inode)->root;
6782 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6783 struct btrfs_file_extent_item *fi;
6784 struct btrfs_key key;
6791 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6793 path = btrfs_alloc_path();
6797 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
6802 slot = path->slots[0];
6805 /* can't find the item, must cow */
6812 leaf = path->nodes[0];
6813 btrfs_item_key_to_cpu(leaf, &key, slot);
6814 if (key.objectid != btrfs_ino(inode) ||
6815 key.type != BTRFS_EXTENT_DATA_KEY) {
6816 /* not our file or wrong item type, must cow */
6820 if (key.offset > offset) {
6821 /* Wrong offset, must cow */
6825 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6826 found_type = btrfs_file_extent_type(leaf, fi);
6827 if (found_type != BTRFS_FILE_EXTENT_REG &&
6828 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6829 /* not a regular extent, must cow */
6833 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6836 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6837 if (extent_end <= offset)
6840 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6841 if (disk_bytenr == 0)
6844 if (btrfs_file_extent_compression(leaf, fi) ||
6845 btrfs_file_extent_encryption(leaf, fi) ||
6846 btrfs_file_extent_other_encoding(leaf, fi))
6849 backref_offset = btrfs_file_extent_offset(leaf, fi);
6852 *orig_start = key.offset - backref_offset;
6853 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6854 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6857 if (btrfs_extent_readonly(root, disk_bytenr))
6860 num_bytes = min(offset + *len, extent_end) - offset;
6861 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6864 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
6865 ret = test_range_bit(io_tree, offset, range_end,
6866 EXTENT_DELALLOC, 0, NULL);
6873 btrfs_release_path(path);
6876 * look for other files referencing this extent, if we
6877 * find any we must cow
6879 trans = btrfs_join_transaction(root);
6880 if (IS_ERR(trans)) {
6885 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6886 key.offset - backref_offset, disk_bytenr);
6887 btrfs_end_transaction(trans, root);
6894 * adjust disk_bytenr and num_bytes to cover just the bytes
6895 * in this extent we are about to write. If there
6896 * are any csums in that range we have to cow in order
6897 * to keep the csums correct
6899 disk_bytenr += backref_offset;
6900 disk_bytenr += offset - key.offset;
6901 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6904 * all of the above have passed, it is safe to overwrite this extent
6910 btrfs_free_path(path);
6914 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
6916 struct radix_tree_root *root = &inode->i_mapping->page_tree;
6918 void **pagep = NULL;
6919 struct page *page = NULL;
6923 start_idx = start >> PAGE_CACHE_SHIFT;
6926 * end is the last byte in the last page. end == start is legal
6928 end_idx = end >> PAGE_CACHE_SHIFT;
6932 /* Most of the code in this while loop is lifted from
6933 * find_get_page. It's been modified to begin searching from a
6934 * page and return just the first page found in that range. If the
6935 * found idx is less than or equal to the end idx then we know that
6936 * a page exists. If no pages are found or if those pages are
6937 * outside of the range then we're fine (yay!) */
6938 while (page == NULL &&
6939 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
6940 page = radix_tree_deref_slot(pagep);
6941 if (unlikely(!page))
6944 if (radix_tree_exception(page)) {
6945 if (radix_tree_deref_retry(page)) {
6950 * Otherwise, shmem/tmpfs must be storing a swap entry
6951 * here as an exceptional entry: so return it without
6952 * attempting to raise page count.
6955 break; /* TODO: Is this relevant for this use case? */
6958 if (!page_cache_get_speculative(page)) {
6964 * Has the page moved?
6965 * This is part of the lockless pagecache protocol. See
6966 * include/linux/pagemap.h for details.
6968 if (unlikely(page != *pagep)) {
6969 page_cache_release(page);
6975 if (page->index <= end_idx)
6977 page_cache_release(page);
6984 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6985 struct extent_state **cached_state, int writing)
6987 struct btrfs_ordered_extent *ordered;
6991 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6994 * We're concerned with the entire range that we're going to be
6995 * doing DIO to, so we need to make sure theres no ordered
6996 * extents in this range.
6998 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6999 lockend - lockstart + 1);
7002 * We need to make sure there are no buffered pages in this
7003 * range either, we could have raced between the invalidate in
7004 * generic_file_direct_write and locking the extent. The
7005 * invalidate needs to happen so that reads after a write do not
7010 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7013 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7014 cached_state, GFP_NOFS);
7017 btrfs_start_ordered_extent(inode, ordered, 1);
7018 btrfs_put_ordered_extent(ordered);
7020 /* Screw you mmap */
7021 ret = filemap_write_and_wait_range(inode->i_mapping,
7028 * If we found a page that couldn't be invalidated just
7029 * fall back to buffered.
7031 ret = invalidate_inode_pages2_range(inode->i_mapping,
7032 lockstart >> PAGE_CACHE_SHIFT,
7033 lockend >> PAGE_CACHE_SHIFT);
7044 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7045 u64 len, u64 orig_start,
7046 u64 block_start, u64 block_len,
7047 u64 orig_block_len, u64 ram_bytes,
7050 struct extent_map_tree *em_tree;
7051 struct extent_map *em;
7052 struct btrfs_root *root = BTRFS_I(inode)->root;
7055 em_tree = &BTRFS_I(inode)->extent_tree;
7056 em = alloc_extent_map();
7058 return ERR_PTR(-ENOMEM);
7061 em->orig_start = orig_start;
7062 em->mod_start = start;
7065 em->block_len = block_len;
7066 em->block_start = block_start;
7067 em->bdev = root->fs_info->fs_devices->latest_bdev;
7068 em->orig_block_len = orig_block_len;
7069 em->ram_bytes = ram_bytes;
7070 em->generation = -1;
7071 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7072 if (type == BTRFS_ORDERED_PREALLOC)
7073 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7076 btrfs_drop_extent_cache(inode, em->start,
7077 em->start + em->len - 1, 0);
7078 write_lock(&em_tree->lock);
7079 ret = add_extent_mapping(em_tree, em, 1);
7080 write_unlock(&em_tree->lock);
7081 } while (ret == -EEXIST);
7084 free_extent_map(em);
7085 return ERR_PTR(ret);
7092 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7093 struct buffer_head *bh_result, int create)
7095 struct extent_map *em;
7096 struct btrfs_root *root = BTRFS_I(inode)->root;
7097 struct extent_state *cached_state = NULL;
7098 u64 start = iblock << inode->i_blkbits;
7099 u64 lockstart, lockend;
7100 u64 len = bh_result->b_size;
7101 int unlock_bits = EXTENT_LOCKED;
7105 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
7107 len = min_t(u64, len, root->sectorsize);
7110 lockend = start + len - 1;
7113 * If this errors out it's because we couldn't invalidate pagecache for
7114 * this range and we need to fallback to buffered.
7116 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
7119 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7126 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7127 * io. INLINE is special, and we could probably kludge it in here, but
7128 * it's still buffered so for safety lets just fall back to the generic
7131 * For COMPRESSED we _have_ to read the entire extent in so we can
7132 * decompress it, so there will be buffering required no matter what we
7133 * do, so go ahead and fallback to buffered.
7135 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7136 * to buffered IO. Don't blame me, this is the price we pay for using
7139 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7140 em->block_start == EXTENT_MAP_INLINE) {
7141 free_extent_map(em);
7146 /* Just a good old fashioned hole, return */
7147 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7148 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7149 free_extent_map(em);
7154 * We don't allocate a new extent in the following cases
7156 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7158 * 2) The extent is marked as PREALLOC. We're good to go here and can
7159 * just use the extent.
7163 len = min(len, em->len - (start - em->start));
7164 lockstart = start + len;
7168 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7169 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7170 em->block_start != EXTENT_MAP_HOLE)) {
7173 u64 block_start, orig_start, orig_block_len, ram_bytes;
7175 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7176 type = BTRFS_ORDERED_PREALLOC;
7178 type = BTRFS_ORDERED_NOCOW;
7179 len = min(len, em->len - (start - em->start));
7180 block_start = em->block_start + (start - em->start);
7182 if (can_nocow_extent(inode, start, &len, &orig_start,
7183 &orig_block_len, &ram_bytes) == 1) {
7184 if (type == BTRFS_ORDERED_PREALLOC) {
7185 free_extent_map(em);
7186 em = create_pinned_em(inode, start, len,
7197 ret = btrfs_add_ordered_extent_dio(inode, start,
7198 block_start, len, len, type);
7200 free_extent_map(em);
7208 * this will cow the extent, reset the len in case we changed
7211 len = bh_result->b_size;
7212 free_extent_map(em);
7213 em = btrfs_new_extent_direct(inode, start, len);
7218 len = min(len, em->len - (start - em->start));
7220 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7222 bh_result->b_size = len;
7223 bh_result->b_bdev = em->bdev;
7224 set_buffer_mapped(bh_result);
7226 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7227 set_buffer_new(bh_result);
7230 * Need to update the i_size under the extent lock so buffered
7231 * readers will get the updated i_size when we unlock.
7233 if (start + len > i_size_read(inode))
7234 i_size_write(inode, start + len);
7236 spin_lock(&BTRFS_I(inode)->lock);
7237 BTRFS_I(inode)->outstanding_extents++;
7238 spin_unlock(&BTRFS_I(inode)->lock);
7240 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7241 lockstart + len - 1, EXTENT_DELALLOC, NULL,
7242 &cached_state, GFP_NOFS);
7247 * In the case of write we need to clear and unlock the entire range,
7248 * in the case of read we need to unlock only the end area that we
7249 * aren't using if there is any left over space.
7251 if (lockstart < lockend) {
7252 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7253 lockend, unlock_bits, 1, 0,
7254 &cached_state, GFP_NOFS);
7256 free_extent_state(cached_state);
7259 free_extent_map(em);
7264 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7265 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7269 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7270 int rw, int mirror_num)
7272 struct btrfs_root *root = BTRFS_I(inode)->root;
7275 BUG_ON(rw & REQ_WRITE);
7279 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7280 BTRFS_WQ_ENDIO_DIO_REPAIR);
7284 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7290 static int btrfs_check_dio_repairable(struct inode *inode,
7291 struct bio *failed_bio,
7292 struct io_failure_record *failrec,
7297 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7298 failrec->logical, failrec->len);
7299 if (num_copies == 1) {
7301 * we only have a single copy of the data, so don't bother with
7302 * all the retry and error correction code that follows. no
7303 * matter what the error is, it is very likely to persist.
7305 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7306 num_copies, failrec->this_mirror, failed_mirror);
7310 failrec->failed_mirror = failed_mirror;
7311 failrec->this_mirror++;
7312 if (failrec->this_mirror == failed_mirror)
7313 failrec->this_mirror++;
7315 if (failrec->this_mirror > num_copies) {
7316 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7317 num_copies, failrec->this_mirror, failed_mirror);
7324 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7325 struct page *page, u64 start, u64 end,
7326 int failed_mirror, bio_end_io_t *repair_endio,
7329 struct io_failure_record *failrec;
7335 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7337 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7341 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7344 free_io_failure(inode, failrec);
7348 if (failed_bio->bi_vcnt > 1)
7349 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7351 read_mode = READ_SYNC;
7353 isector = start - btrfs_io_bio(failed_bio)->logical;
7354 isector >>= inode->i_sb->s_blocksize_bits;
7355 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7356 0, isector, repair_endio, repair_arg);
7358 free_io_failure(inode, failrec);
7362 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7363 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7364 read_mode, failrec->this_mirror, failrec->in_validation);
7366 ret = submit_dio_repair_bio(inode, bio, read_mode,
7367 failrec->this_mirror);
7369 free_io_failure(inode, failrec);
7376 struct btrfs_retry_complete {
7377 struct completion done;
7378 struct inode *inode;
7383 static void btrfs_retry_endio_nocsum(struct bio *bio, int err)
7385 struct btrfs_retry_complete *done = bio->bi_private;
7386 struct bio_vec *bvec;
7393 bio_for_each_segment_all(bvec, bio, i)
7394 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7396 complete(&done->done);
7400 static int __btrfs_correct_data_nocsum(struct inode *inode,
7401 struct btrfs_io_bio *io_bio)
7403 struct bio_vec *bvec;
7404 struct btrfs_retry_complete done;
7409 start = io_bio->logical;
7412 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7416 init_completion(&done.done);
7418 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7419 start + bvec->bv_len - 1,
7421 btrfs_retry_endio_nocsum, &done);
7425 wait_for_completion(&done.done);
7427 if (!done.uptodate) {
7428 /* We might have another mirror, so try again */
7432 start += bvec->bv_len;
7438 static void btrfs_retry_endio(struct bio *bio, int err)
7440 struct btrfs_retry_complete *done = bio->bi_private;
7441 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7442 struct bio_vec *bvec;
7451 bio_for_each_segment_all(bvec, bio, i) {
7452 ret = __readpage_endio_check(done->inode, io_bio, i,
7454 done->start, bvec->bv_len);
7456 clean_io_failure(done->inode, done->start,
7462 done->uptodate = uptodate;
7464 complete(&done->done);
7468 static int __btrfs_subio_endio_read(struct inode *inode,
7469 struct btrfs_io_bio *io_bio, int err)
7471 struct bio_vec *bvec;
7472 struct btrfs_retry_complete done;
7479 start = io_bio->logical;
7482 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7483 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7484 0, start, bvec->bv_len);
7490 init_completion(&done.done);
7492 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7493 start + bvec->bv_len - 1,
7495 btrfs_retry_endio, &done);
7501 wait_for_completion(&done.done);
7503 if (!done.uptodate) {
7504 /* We might have another mirror, so try again */
7508 offset += bvec->bv_len;
7509 start += bvec->bv_len;
7515 static int btrfs_subio_endio_read(struct inode *inode,
7516 struct btrfs_io_bio *io_bio, int err)
7518 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7522 return __btrfs_correct_data_nocsum(inode, io_bio);
7526 return __btrfs_subio_endio_read(inode, io_bio, err);
7530 static void btrfs_endio_direct_read(struct bio *bio, int err)
7532 struct btrfs_dio_private *dip = bio->bi_private;
7533 struct inode *inode = dip->inode;
7534 struct bio *dio_bio;
7535 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7537 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
7538 err = btrfs_subio_endio_read(inode, io_bio, err);
7540 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
7541 dip->logical_offset + dip->bytes - 1);
7542 dio_bio = dip->dio_bio;
7546 /* If we had a csum failure make sure to clear the uptodate flag */
7548 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7549 dio_end_io(dio_bio, err);
7552 io_bio->end_io(io_bio, err);
7556 static void btrfs_endio_direct_write(struct bio *bio, int err)
7558 struct btrfs_dio_private *dip = bio->bi_private;
7559 struct inode *inode = dip->inode;
7560 struct btrfs_root *root = BTRFS_I(inode)->root;
7561 struct btrfs_ordered_extent *ordered = NULL;
7562 u64 ordered_offset = dip->logical_offset;
7563 u64 ordered_bytes = dip->bytes;
7564 struct bio *dio_bio;
7570 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7572 ordered_bytes, !err);
7576 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
7577 finish_ordered_fn, NULL, NULL);
7578 btrfs_queue_work(root->fs_info->endio_write_workers,
7582 * our bio might span multiple ordered extents. If we haven't
7583 * completed the accounting for the whole dio, go back and try again
7585 if (ordered_offset < dip->logical_offset + dip->bytes) {
7586 ordered_bytes = dip->logical_offset + dip->bytes -
7592 dio_bio = dip->dio_bio;
7596 /* If we had an error make sure to clear the uptodate flag */
7598 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7599 dio_end_io(dio_bio, err);
7603 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7604 struct bio *bio, int mirror_num,
7605 unsigned long bio_flags, u64 offset)
7608 struct btrfs_root *root = BTRFS_I(inode)->root;
7609 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7610 BUG_ON(ret); /* -ENOMEM */
7614 static void btrfs_end_dio_bio(struct bio *bio, int err)
7616 struct btrfs_dio_private *dip = bio->bi_private;
7619 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7620 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7621 btrfs_ino(dip->inode), bio->bi_rw,
7622 (unsigned long long)bio->bi_iter.bi_sector,
7623 bio->bi_iter.bi_size, err);
7625 if (dip->subio_endio)
7626 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
7632 * before atomic variable goto zero, we must make sure
7633 * dip->errors is perceived to be set.
7635 smp_mb__before_atomic();
7638 /* if there are more bios still pending for this dio, just exit */
7639 if (!atomic_dec_and_test(&dip->pending_bios))
7643 bio_io_error(dip->orig_bio);
7645 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7646 bio_endio(dip->orig_bio, 0);
7652 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7653 u64 first_sector, gfp_t gfp_flags)
7655 int nr_vecs = bio_get_nr_vecs(bdev);
7656 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7659 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
7660 struct inode *inode,
7661 struct btrfs_dio_private *dip,
7665 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7666 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
7670 * We load all the csum data we need when we submit
7671 * the first bio to reduce the csum tree search and
7674 if (dip->logical_offset == file_offset) {
7675 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
7681 if (bio == dip->orig_bio)
7684 file_offset -= dip->logical_offset;
7685 file_offset >>= inode->i_sb->s_blocksize_bits;
7686 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
7691 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7692 int rw, u64 file_offset, int skip_sum,
7695 struct btrfs_dio_private *dip = bio->bi_private;
7696 int write = rw & REQ_WRITE;
7697 struct btrfs_root *root = BTRFS_I(inode)->root;
7701 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7706 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7707 BTRFS_WQ_ENDIO_DATA);
7715 if (write && async_submit) {
7716 ret = btrfs_wq_submit_bio(root->fs_info,
7717 inode, rw, bio, 0, 0,
7719 __btrfs_submit_bio_start_direct_io,
7720 __btrfs_submit_bio_done);
7724 * If we aren't doing async submit, calculate the csum of the
7727 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7731 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
7737 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7743 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7746 struct inode *inode = dip->inode;
7747 struct btrfs_root *root = BTRFS_I(inode)->root;
7749 struct bio *orig_bio = dip->orig_bio;
7750 struct bio_vec *bvec = orig_bio->bi_io_vec;
7751 u64 start_sector = orig_bio->bi_iter.bi_sector;
7752 u64 file_offset = dip->logical_offset;
7757 int async_submit = 0;
7759 map_length = orig_bio->bi_iter.bi_size;
7760 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7761 &map_length, NULL, 0);
7765 if (map_length >= orig_bio->bi_iter.bi_size) {
7767 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
7771 /* async crcs make it difficult to collect full stripe writes. */
7772 if (btrfs_get_alloc_profile(root, 1) &
7773 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7778 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7782 bio->bi_private = dip;
7783 bio->bi_end_io = btrfs_end_dio_bio;
7784 btrfs_io_bio(bio)->logical = file_offset;
7785 atomic_inc(&dip->pending_bios);
7787 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7788 if (map_length < submit_len + bvec->bv_len ||
7789 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7790 bvec->bv_offset) < bvec->bv_len) {
7792 * inc the count before we submit the bio so
7793 * we know the end IO handler won't happen before
7794 * we inc the count. Otherwise, the dip might get freed
7795 * before we're done setting it up
7797 atomic_inc(&dip->pending_bios);
7798 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7799 file_offset, skip_sum,
7803 atomic_dec(&dip->pending_bios);
7807 start_sector += submit_len >> 9;
7808 file_offset += submit_len;
7813 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7814 start_sector, GFP_NOFS);
7817 bio->bi_private = dip;
7818 bio->bi_end_io = btrfs_end_dio_bio;
7819 btrfs_io_bio(bio)->logical = file_offset;
7821 map_length = orig_bio->bi_iter.bi_size;
7822 ret = btrfs_map_block(root->fs_info, rw,
7824 &map_length, NULL, 0);
7830 submit_len += bvec->bv_len;
7837 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7846 * before atomic variable goto zero, we must
7847 * make sure dip->errors is perceived to be set.
7849 smp_mb__before_atomic();
7850 if (atomic_dec_and_test(&dip->pending_bios))
7851 bio_io_error(dip->orig_bio);
7853 /* bio_end_io() will handle error, so we needn't return it */
7857 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7858 struct inode *inode, loff_t file_offset)
7860 struct btrfs_root *root = BTRFS_I(inode)->root;
7861 struct btrfs_dio_private *dip;
7863 struct btrfs_io_bio *btrfs_bio;
7865 int write = rw & REQ_WRITE;
7868 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7870 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7876 dip = kzalloc(sizeof(*dip), GFP_NOFS);
7882 dip->private = dio_bio->bi_private;
7884 dip->logical_offset = file_offset;
7885 dip->bytes = dio_bio->bi_iter.bi_size;
7886 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
7887 io_bio->bi_private = dip;
7888 dip->orig_bio = io_bio;
7889 dip->dio_bio = dio_bio;
7890 atomic_set(&dip->pending_bios, 0);
7891 btrfs_bio = btrfs_io_bio(io_bio);
7892 btrfs_bio->logical = file_offset;
7895 io_bio->bi_end_io = btrfs_endio_direct_write;
7897 io_bio->bi_end_io = btrfs_endio_direct_read;
7898 dip->subio_endio = btrfs_subio_endio_read;
7901 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7905 if (btrfs_bio->end_io)
7906 btrfs_bio->end_io(btrfs_bio, ret);
7912 * If this is a write, we need to clean up the reserved space and kill
7913 * the ordered extent.
7916 struct btrfs_ordered_extent *ordered;
7917 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7918 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7919 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7920 btrfs_free_reserved_extent(root, ordered->start,
7921 ordered->disk_len, 1);
7922 btrfs_put_ordered_extent(ordered);
7923 btrfs_put_ordered_extent(ordered);
7925 bio_endio(dio_bio, ret);
7928 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7929 const struct iov_iter *iter, loff_t offset)
7933 unsigned blocksize_mask = root->sectorsize - 1;
7934 ssize_t retval = -EINVAL;
7936 if (offset & blocksize_mask)
7939 if (iov_iter_alignment(iter) & blocksize_mask)
7942 /* If this is a write we don't need to check anymore */
7946 * Check to make sure we don't have duplicate iov_base's in this
7947 * iovec, if so return EINVAL, otherwise we'll get csum errors
7948 * when reading back.
7950 for (seg = 0; seg < iter->nr_segs; seg++) {
7951 for (i = seg + 1; i < iter->nr_segs; i++) {
7952 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
7961 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7962 struct iov_iter *iter, loff_t offset)
7964 struct file *file = iocb->ki_filp;
7965 struct inode *inode = file->f_mapping->host;
7969 bool relock = false;
7972 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iter, offset))
7975 atomic_inc(&inode->i_dio_count);
7976 smp_mb__after_atomic();
7979 * The generic stuff only does filemap_write_and_wait_range, which
7980 * isn't enough if we've written compressed pages to this area, so
7981 * we need to flush the dirty pages again to make absolutely sure
7982 * that any outstanding dirty pages are on disk.
7984 count = iov_iter_count(iter);
7985 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7986 &BTRFS_I(inode)->runtime_flags))
7987 filemap_fdatawrite_range(inode->i_mapping, offset,
7988 offset + count - 1);
7992 * If the write DIO is beyond the EOF, we need update
7993 * the isize, but it is protected by i_mutex. So we can
7994 * not unlock the i_mutex at this case.
7996 if (offset + count <= inode->i_size) {
7997 mutex_unlock(&inode->i_mutex);
8000 ret = btrfs_delalloc_reserve_space(inode, count);
8003 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8004 &BTRFS_I(inode)->runtime_flags)) {
8005 inode_dio_done(inode);
8006 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8010 ret = __blockdev_direct_IO(rw, iocb, inode,
8011 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8012 iter, offset, btrfs_get_blocks_direct, NULL,
8013 btrfs_submit_direct, flags);
8015 if (ret < 0 && ret != -EIOCBQUEUED)
8016 btrfs_delalloc_release_space(inode, count);
8017 else if (ret >= 0 && (size_t)ret < count)
8018 btrfs_delalloc_release_space(inode,
8019 count - (size_t)ret);
8021 btrfs_delalloc_release_metadata(inode, 0);
8025 inode_dio_done(inode);
8027 mutex_lock(&inode->i_mutex);
8032 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8034 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8035 __u64 start, __u64 len)
8039 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8043 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8046 int btrfs_readpage(struct file *file, struct page *page)
8048 struct extent_io_tree *tree;
8049 tree = &BTRFS_I(page->mapping->host)->io_tree;
8050 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8053 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8055 struct extent_io_tree *tree;
8058 if (current->flags & PF_MEMALLOC) {
8059 redirty_page_for_writepage(wbc, page);
8063 tree = &BTRFS_I(page->mapping->host)->io_tree;
8064 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8067 static int btrfs_writepages(struct address_space *mapping,
8068 struct writeback_control *wbc)
8070 struct extent_io_tree *tree;
8072 tree = &BTRFS_I(mapping->host)->io_tree;
8073 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8077 btrfs_readpages(struct file *file, struct address_space *mapping,
8078 struct list_head *pages, unsigned nr_pages)
8080 struct extent_io_tree *tree;
8081 tree = &BTRFS_I(mapping->host)->io_tree;
8082 return extent_readpages(tree, mapping, pages, nr_pages,
8085 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8087 struct extent_io_tree *tree;
8088 struct extent_map_tree *map;
8091 tree = &BTRFS_I(page->mapping->host)->io_tree;
8092 map = &BTRFS_I(page->mapping->host)->extent_tree;
8093 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8095 ClearPagePrivate(page);
8096 set_page_private(page, 0);
8097 page_cache_release(page);
8102 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8104 if (PageWriteback(page) || PageDirty(page))
8106 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8109 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8110 unsigned int length)
8112 struct inode *inode = page->mapping->host;
8113 struct extent_io_tree *tree;
8114 struct btrfs_ordered_extent *ordered;
8115 struct extent_state *cached_state = NULL;
8116 u64 page_start = page_offset(page);
8117 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
8118 int inode_evicting = inode->i_state & I_FREEING;
8121 * we have the page locked, so new writeback can't start,
8122 * and the dirty bit won't be cleared while we are here.
8124 * Wait for IO on this page so that we can safely clear
8125 * the PagePrivate2 bit and do ordered accounting
8127 wait_on_page_writeback(page);
8129 tree = &BTRFS_I(inode)->io_tree;
8131 btrfs_releasepage(page, GFP_NOFS);
8135 if (!inode_evicting)
8136 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
8137 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8140 * IO on this page will never be started, so we need
8141 * to account for any ordered extents now
8143 if (!inode_evicting)
8144 clear_extent_bit(tree, page_start, page_end,
8145 EXTENT_DIRTY | EXTENT_DELALLOC |
8146 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8147 EXTENT_DEFRAG, 1, 0, &cached_state,
8150 * whoever cleared the private bit is responsible
8151 * for the finish_ordered_io
8153 if (TestClearPagePrivate2(page)) {
8154 struct btrfs_ordered_inode_tree *tree;
8157 tree = &BTRFS_I(inode)->ordered_tree;
8159 spin_lock_irq(&tree->lock);
8160 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8161 new_len = page_start - ordered->file_offset;
8162 if (new_len < ordered->truncated_len)
8163 ordered->truncated_len = new_len;
8164 spin_unlock_irq(&tree->lock);
8166 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8168 PAGE_CACHE_SIZE, 1))
8169 btrfs_finish_ordered_io(ordered);
8171 btrfs_put_ordered_extent(ordered);
8172 if (!inode_evicting) {
8173 cached_state = NULL;
8174 lock_extent_bits(tree, page_start, page_end, 0,
8179 if (!inode_evicting) {
8180 clear_extent_bit(tree, page_start, page_end,
8181 EXTENT_LOCKED | EXTENT_DIRTY |
8182 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8183 EXTENT_DEFRAG, 1, 1,
8184 &cached_state, GFP_NOFS);
8186 __btrfs_releasepage(page, GFP_NOFS);
8189 ClearPageChecked(page);
8190 if (PagePrivate(page)) {
8191 ClearPagePrivate(page);
8192 set_page_private(page, 0);
8193 page_cache_release(page);
8198 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8199 * called from a page fault handler when a page is first dirtied. Hence we must
8200 * be careful to check for EOF conditions here. We set the page up correctly
8201 * for a written page which means we get ENOSPC checking when writing into
8202 * holes and correct delalloc and unwritten extent mapping on filesystems that
8203 * support these features.
8205 * We are not allowed to take the i_mutex here so we have to play games to
8206 * protect against truncate races as the page could now be beyond EOF. Because
8207 * vmtruncate() writes the inode size before removing pages, once we have the
8208 * page lock we can determine safely if the page is beyond EOF. If it is not
8209 * beyond EOF, then the page is guaranteed safe against truncation until we
8212 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8214 struct page *page = vmf->page;
8215 struct inode *inode = file_inode(vma->vm_file);
8216 struct btrfs_root *root = BTRFS_I(inode)->root;
8217 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8218 struct btrfs_ordered_extent *ordered;
8219 struct extent_state *cached_state = NULL;
8221 unsigned long zero_start;
8228 sb_start_pagefault(inode->i_sb);
8229 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
8231 ret = file_update_time(vma->vm_file);
8237 else /* -ENOSPC, -EIO, etc */
8238 ret = VM_FAULT_SIGBUS;
8244 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8247 size = i_size_read(inode);
8248 page_start = page_offset(page);
8249 page_end = page_start + PAGE_CACHE_SIZE - 1;
8251 if ((page->mapping != inode->i_mapping) ||
8252 (page_start >= size)) {
8253 /* page got truncated out from underneath us */
8256 wait_on_page_writeback(page);
8258 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
8259 set_page_extent_mapped(page);
8262 * we can't set the delalloc bits if there are pending ordered
8263 * extents. Drop our locks and wait for them to finish
8265 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8267 unlock_extent_cached(io_tree, page_start, page_end,
8268 &cached_state, GFP_NOFS);
8270 btrfs_start_ordered_extent(inode, ordered, 1);
8271 btrfs_put_ordered_extent(ordered);
8276 * XXX - page_mkwrite gets called every time the page is dirtied, even
8277 * if it was already dirty, so for space accounting reasons we need to
8278 * clear any delalloc bits for the range we are fixing to save. There
8279 * is probably a better way to do this, but for now keep consistent with
8280 * prepare_pages in the normal write path.
8282 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
8283 EXTENT_DIRTY | EXTENT_DELALLOC |
8284 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8285 0, 0, &cached_state, GFP_NOFS);
8287 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
8290 unlock_extent_cached(io_tree, page_start, page_end,
8291 &cached_state, GFP_NOFS);
8292 ret = VM_FAULT_SIGBUS;
8297 /* page is wholly or partially inside EOF */
8298 if (page_start + PAGE_CACHE_SIZE > size)
8299 zero_start = size & ~PAGE_CACHE_MASK;
8301 zero_start = PAGE_CACHE_SIZE;
8303 if (zero_start != PAGE_CACHE_SIZE) {
8305 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
8306 flush_dcache_page(page);
8309 ClearPageChecked(page);
8310 set_page_dirty(page);
8311 SetPageUptodate(page);
8313 BTRFS_I(inode)->last_trans = root->fs_info->generation;
8314 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8315 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8317 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
8321 sb_end_pagefault(inode->i_sb);
8322 return VM_FAULT_LOCKED;
8326 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
8328 sb_end_pagefault(inode->i_sb);
8332 static int btrfs_truncate(struct inode *inode)
8334 struct btrfs_root *root = BTRFS_I(inode)->root;
8335 struct btrfs_block_rsv *rsv;
8338 struct btrfs_trans_handle *trans;
8339 u64 mask = root->sectorsize - 1;
8340 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
8342 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8348 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8349 * 3 things going on here
8351 * 1) We need to reserve space for our orphan item and the space to
8352 * delete our orphan item. Lord knows we don't want to have a dangling
8353 * orphan item because we didn't reserve space to remove it.
8355 * 2) We need to reserve space to update our inode.
8357 * 3) We need to have something to cache all the space that is going to
8358 * be free'd up by the truncate operation, but also have some slack
8359 * space reserved in case it uses space during the truncate (thank you
8360 * very much snapshotting).
8362 * And we need these to all be seperate. The fact is we can use alot of
8363 * space doing the truncate, and we have no earthly idea how much space
8364 * we will use, so we need the truncate reservation to be seperate so it
8365 * doesn't end up using space reserved for updating the inode or
8366 * removing the orphan item. We also need to be able to stop the
8367 * transaction and start a new one, which means we need to be able to
8368 * update the inode several times, and we have no idea of knowing how
8369 * many times that will be, so we can't just reserve 1 item for the
8370 * entirety of the opration, so that has to be done seperately as well.
8371 * Then there is the orphan item, which does indeed need to be held on
8372 * to for the whole operation, and we need nobody to touch this reserved
8373 * space except the orphan code.
8375 * So that leaves us with
8377 * 1) root->orphan_block_rsv - for the orphan deletion.
8378 * 2) rsv - for the truncate reservation, which we will steal from the
8379 * transaction reservation.
8380 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8381 * updating the inode.
8383 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
8386 rsv->size = min_size;
8390 * 1 for the truncate slack space
8391 * 1 for updating the inode.
8393 trans = btrfs_start_transaction(root, 2);
8394 if (IS_ERR(trans)) {
8395 err = PTR_ERR(trans);
8399 /* Migrate the slack space for the truncate to our reserve */
8400 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
8405 * So if we truncate and then write and fsync we normally would just
8406 * write the extents that changed, which is a problem if we need to
8407 * first truncate that entire inode. So set this flag so we write out
8408 * all of the extents in the inode to the sync log so we're completely
8411 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8412 trans->block_rsv = rsv;
8415 ret = btrfs_truncate_inode_items(trans, root, inode,
8417 BTRFS_EXTENT_DATA_KEY);
8418 if (ret != -ENOSPC) {
8423 trans->block_rsv = &root->fs_info->trans_block_rsv;
8424 ret = btrfs_update_inode(trans, root, inode);
8430 btrfs_end_transaction(trans, root);
8431 btrfs_btree_balance_dirty(root);
8433 trans = btrfs_start_transaction(root, 2);
8434 if (IS_ERR(trans)) {
8435 ret = err = PTR_ERR(trans);
8440 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
8442 BUG_ON(ret); /* shouldn't happen */
8443 trans->block_rsv = rsv;
8446 if (ret == 0 && inode->i_nlink > 0) {
8447 trans->block_rsv = root->orphan_block_rsv;
8448 ret = btrfs_orphan_del(trans, inode);
8454 trans->block_rsv = &root->fs_info->trans_block_rsv;
8455 ret = btrfs_update_inode(trans, root, inode);
8459 ret = btrfs_end_transaction(trans, root);
8460 btrfs_btree_balance_dirty(root);
8464 btrfs_free_block_rsv(root, rsv);
8473 * create a new subvolume directory/inode (helper for the ioctl).
8475 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8476 struct btrfs_root *new_root,
8477 struct btrfs_root *parent_root,
8480 struct inode *inode;
8484 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8485 new_dirid, new_dirid,
8486 S_IFDIR | (~current_umask() & S_IRWXUGO),
8489 return PTR_ERR(inode);
8490 inode->i_op = &btrfs_dir_inode_operations;
8491 inode->i_fop = &btrfs_dir_file_operations;
8493 set_nlink(inode, 1);
8494 btrfs_i_size_write(inode, 0);
8495 unlock_new_inode(inode);
8497 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8499 btrfs_err(new_root->fs_info,
8500 "error inheriting subvolume %llu properties: %d",
8501 new_root->root_key.objectid, err);
8503 err = btrfs_update_inode(trans, new_root, inode);
8509 struct inode *btrfs_alloc_inode(struct super_block *sb)
8511 struct btrfs_inode *ei;
8512 struct inode *inode;
8514 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
8521 ei->last_sub_trans = 0;
8522 ei->logged_trans = 0;
8523 ei->delalloc_bytes = 0;
8524 ei->defrag_bytes = 0;
8525 ei->disk_i_size = 0;
8528 ei->index_cnt = (u64)-1;
8530 ei->last_unlink_trans = 0;
8531 ei->last_log_commit = 0;
8533 spin_lock_init(&ei->lock);
8534 ei->outstanding_extents = 0;
8535 ei->reserved_extents = 0;
8537 ei->runtime_flags = 0;
8538 ei->force_compress = BTRFS_COMPRESS_NONE;
8540 ei->delayed_node = NULL;
8542 inode = &ei->vfs_inode;
8543 extent_map_tree_init(&ei->extent_tree);
8544 extent_io_tree_init(&ei->io_tree, &inode->i_data);
8545 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
8546 ei->io_tree.track_uptodate = 1;
8547 ei->io_failure_tree.track_uptodate = 1;
8548 atomic_set(&ei->sync_writers, 0);
8549 mutex_init(&ei->log_mutex);
8550 mutex_init(&ei->delalloc_mutex);
8551 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8552 INIT_LIST_HEAD(&ei->delalloc_inodes);
8553 RB_CLEAR_NODE(&ei->rb_node);
8558 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8559 void btrfs_test_destroy_inode(struct inode *inode)
8561 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8562 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8566 static void btrfs_i_callback(struct rcu_head *head)
8568 struct inode *inode = container_of(head, struct inode, i_rcu);
8569 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8572 void btrfs_destroy_inode(struct inode *inode)
8574 struct btrfs_ordered_extent *ordered;
8575 struct btrfs_root *root = BTRFS_I(inode)->root;
8577 WARN_ON(!hlist_empty(&inode->i_dentry));
8578 WARN_ON(inode->i_data.nrpages);
8579 WARN_ON(BTRFS_I(inode)->outstanding_extents);
8580 WARN_ON(BTRFS_I(inode)->reserved_extents);
8581 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8582 WARN_ON(BTRFS_I(inode)->csum_bytes);
8583 WARN_ON(BTRFS_I(inode)->defrag_bytes);
8586 * This can happen where we create an inode, but somebody else also
8587 * created the same inode and we need to destroy the one we already
8593 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8594 &BTRFS_I(inode)->runtime_flags)) {
8595 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
8597 atomic_dec(&root->orphan_inodes);
8601 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8605 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
8606 ordered->file_offset, ordered->len);
8607 btrfs_remove_ordered_extent(inode, ordered);
8608 btrfs_put_ordered_extent(ordered);
8609 btrfs_put_ordered_extent(ordered);
8612 inode_tree_del(inode);
8613 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8615 call_rcu(&inode->i_rcu, btrfs_i_callback);
8618 int btrfs_drop_inode(struct inode *inode)
8620 struct btrfs_root *root = BTRFS_I(inode)->root;
8625 /* the snap/subvol tree is on deleting */
8626 if (btrfs_root_refs(&root->root_item) == 0)
8629 return generic_drop_inode(inode);
8632 static void init_once(void *foo)
8634 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8636 inode_init_once(&ei->vfs_inode);
8639 void btrfs_destroy_cachep(void)
8642 * Make sure all delayed rcu free inodes are flushed before we
8646 if (btrfs_inode_cachep)
8647 kmem_cache_destroy(btrfs_inode_cachep);
8648 if (btrfs_trans_handle_cachep)
8649 kmem_cache_destroy(btrfs_trans_handle_cachep);
8650 if (btrfs_transaction_cachep)
8651 kmem_cache_destroy(btrfs_transaction_cachep);
8652 if (btrfs_path_cachep)
8653 kmem_cache_destroy(btrfs_path_cachep);
8654 if (btrfs_free_space_cachep)
8655 kmem_cache_destroy(btrfs_free_space_cachep);
8656 if (btrfs_delalloc_work_cachep)
8657 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8660 int btrfs_init_cachep(void)
8662 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8663 sizeof(struct btrfs_inode), 0,
8664 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8665 if (!btrfs_inode_cachep)
8668 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8669 sizeof(struct btrfs_trans_handle), 0,
8670 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8671 if (!btrfs_trans_handle_cachep)
8674 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8675 sizeof(struct btrfs_transaction), 0,
8676 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8677 if (!btrfs_transaction_cachep)
8680 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8681 sizeof(struct btrfs_path), 0,
8682 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8683 if (!btrfs_path_cachep)
8686 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8687 sizeof(struct btrfs_free_space), 0,
8688 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8689 if (!btrfs_free_space_cachep)
8692 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8693 sizeof(struct btrfs_delalloc_work), 0,
8694 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8696 if (!btrfs_delalloc_work_cachep)
8701 btrfs_destroy_cachep();
8705 static int btrfs_getattr(struct vfsmount *mnt,
8706 struct dentry *dentry, struct kstat *stat)
8709 struct inode *inode = dentry->d_inode;
8710 u32 blocksize = inode->i_sb->s_blocksize;
8712 generic_fillattr(inode, stat);
8713 stat->dev = BTRFS_I(inode)->root->anon_dev;
8714 stat->blksize = PAGE_CACHE_SIZE;
8716 spin_lock(&BTRFS_I(inode)->lock);
8717 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8718 spin_unlock(&BTRFS_I(inode)->lock);
8719 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8720 ALIGN(delalloc_bytes, blocksize)) >> 9;
8724 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8725 struct inode *new_dir, struct dentry *new_dentry)
8727 struct btrfs_trans_handle *trans;
8728 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8729 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8730 struct inode *new_inode = new_dentry->d_inode;
8731 struct inode *old_inode = old_dentry->d_inode;
8732 struct timespec ctime = CURRENT_TIME;
8736 u64 old_ino = btrfs_ino(old_inode);
8738 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8741 /* we only allow rename subvolume link between subvolumes */
8742 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8745 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8746 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8749 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8750 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8754 /* check for collisions, even if the name isn't there */
8755 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
8756 new_dentry->d_name.name,
8757 new_dentry->d_name.len);
8760 if (ret == -EEXIST) {
8762 * eexist without a new_inode */
8763 if (WARN_ON(!new_inode)) {
8767 /* maybe -EOVERFLOW */
8774 * we're using rename to replace one file with another. Start IO on it
8775 * now so we don't add too much work to the end of the transaction
8777 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
8778 filemap_flush(old_inode->i_mapping);
8780 /* close the racy window with snapshot create/destroy ioctl */
8781 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8782 down_read(&root->fs_info->subvol_sem);
8784 * We want to reserve the absolute worst case amount of items. So if
8785 * both inodes are subvols and we need to unlink them then that would
8786 * require 4 item modifications, but if they are both normal inodes it
8787 * would require 5 item modifications, so we'll assume their normal
8788 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8789 * should cover the worst case number of items we'll modify.
8791 trans = btrfs_start_transaction(root, 11);
8792 if (IS_ERR(trans)) {
8793 ret = PTR_ERR(trans);
8798 btrfs_record_root_in_trans(trans, dest);
8800 ret = btrfs_set_inode_index(new_dir, &index);
8804 BTRFS_I(old_inode)->dir_index = 0ULL;
8805 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8806 /* force full log commit if subvolume involved. */
8807 btrfs_set_log_full_commit(root->fs_info, trans);
8809 ret = btrfs_insert_inode_ref(trans, dest,
8810 new_dentry->d_name.name,
8811 new_dentry->d_name.len,
8813 btrfs_ino(new_dir), index);
8817 * this is an ugly little race, but the rename is required
8818 * to make sure that if we crash, the inode is either at the
8819 * old name or the new one. pinning the log transaction lets
8820 * us make sure we don't allow a log commit to come in after
8821 * we unlink the name but before we add the new name back in.
8823 btrfs_pin_log_trans(root);
8826 inode_inc_iversion(old_dir);
8827 inode_inc_iversion(new_dir);
8828 inode_inc_iversion(old_inode);
8829 old_dir->i_ctime = old_dir->i_mtime = ctime;
8830 new_dir->i_ctime = new_dir->i_mtime = ctime;
8831 old_inode->i_ctime = ctime;
8833 if (old_dentry->d_parent != new_dentry->d_parent)
8834 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8836 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8837 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8838 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8839 old_dentry->d_name.name,
8840 old_dentry->d_name.len);
8842 ret = __btrfs_unlink_inode(trans, root, old_dir,
8843 old_dentry->d_inode,
8844 old_dentry->d_name.name,
8845 old_dentry->d_name.len);
8847 ret = btrfs_update_inode(trans, root, old_inode);
8850 btrfs_abort_transaction(trans, root, ret);
8855 inode_inc_iversion(new_inode);
8856 new_inode->i_ctime = CURRENT_TIME;
8857 if (unlikely(btrfs_ino(new_inode) ==
8858 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8859 root_objectid = BTRFS_I(new_inode)->location.objectid;
8860 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8862 new_dentry->d_name.name,
8863 new_dentry->d_name.len);
8864 BUG_ON(new_inode->i_nlink == 0);
8866 ret = btrfs_unlink_inode(trans, dest, new_dir,
8867 new_dentry->d_inode,
8868 new_dentry->d_name.name,
8869 new_dentry->d_name.len);
8871 if (!ret && new_inode->i_nlink == 0)
8872 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8874 btrfs_abort_transaction(trans, root, ret);
8879 ret = btrfs_add_link(trans, new_dir, old_inode,
8880 new_dentry->d_name.name,
8881 new_dentry->d_name.len, 0, index);
8883 btrfs_abort_transaction(trans, root, ret);
8887 if (old_inode->i_nlink == 1)
8888 BTRFS_I(old_inode)->dir_index = index;
8890 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8891 struct dentry *parent = new_dentry->d_parent;
8892 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8893 btrfs_end_log_trans(root);
8896 btrfs_end_transaction(trans, root);
8898 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8899 up_read(&root->fs_info->subvol_sem);
8904 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
8905 struct inode *new_dir, struct dentry *new_dentry,
8908 if (flags & ~RENAME_NOREPLACE)
8911 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
8914 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8916 struct btrfs_delalloc_work *delalloc_work;
8917 struct inode *inode;
8919 delalloc_work = container_of(work, struct btrfs_delalloc_work,
8921 inode = delalloc_work->inode;
8922 if (delalloc_work->wait) {
8923 btrfs_wait_ordered_range(inode, 0, (u64)-1);
8925 filemap_flush(inode->i_mapping);
8926 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8927 &BTRFS_I(inode)->runtime_flags))
8928 filemap_flush(inode->i_mapping);
8931 if (delalloc_work->delay_iput)
8932 btrfs_add_delayed_iput(inode);
8935 complete(&delalloc_work->completion);
8938 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8939 int wait, int delay_iput)
8941 struct btrfs_delalloc_work *work;
8943 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8947 init_completion(&work->completion);
8948 INIT_LIST_HEAD(&work->list);
8949 work->inode = inode;
8951 work->delay_iput = delay_iput;
8952 WARN_ON_ONCE(!inode);
8953 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
8954 btrfs_run_delalloc_work, NULL, NULL);
8959 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8961 wait_for_completion(&work->completion);
8962 kmem_cache_free(btrfs_delalloc_work_cachep, work);
8966 * some fairly slow code that needs optimization. This walks the list
8967 * of all the inodes with pending delalloc and forces them to disk.
8969 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
8972 struct btrfs_inode *binode;
8973 struct inode *inode;
8974 struct btrfs_delalloc_work *work, *next;
8975 struct list_head works;
8976 struct list_head splice;
8979 INIT_LIST_HEAD(&works);
8980 INIT_LIST_HEAD(&splice);
8982 mutex_lock(&root->delalloc_mutex);
8983 spin_lock(&root->delalloc_lock);
8984 list_splice_init(&root->delalloc_inodes, &splice);
8985 while (!list_empty(&splice)) {
8986 binode = list_entry(splice.next, struct btrfs_inode,
8989 list_move_tail(&binode->delalloc_inodes,
8990 &root->delalloc_inodes);
8991 inode = igrab(&binode->vfs_inode);
8993 cond_resched_lock(&root->delalloc_lock);
8996 spin_unlock(&root->delalloc_lock);
8998 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
9001 btrfs_add_delayed_iput(inode);
9007 list_add_tail(&work->list, &works);
9008 btrfs_queue_work(root->fs_info->flush_workers,
9011 if (nr != -1 && ret >= nr)
9014 spin_lock(&root->delalloc_lock);
9016 spin_unlock(&root->delalloc_lock);
9019 list_for_each_entry_safe(work, next, &works, list) {
9020 list_del_init(&work->list);
9021 btrfs_wait_and_free_delalloc_work(work);
9024 if (!list_empty_careful(&splice)) {
9025 spin_lock(&root->delalloc_lock);
9026 list_splice_tail(&splice, &root->delalloc_inodes);
9027 spin_unlock(&root->delalloc_lock);
9029 mutex_unlock(&root->delalloc_mutex);
9033 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9037 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
9040 ret = __start_delalloc_inodes(root, delay_iput, -1);
9044 * the filemap_flush will queue IO into the worker threads, but
9045 * we have to make sure the IO is actually started and that
9046 * ordered extents get created before we return
9048 atomic_inc(&root->fs_info->async_submit_draining);
9049 while (atomic_read(&root->fs_info->nr_async_submits) ||
9050 atomic_read(&root->fs_info->async_delalloc_pages)) {
9051 wait_event(root->fs_info->async_submit_wait,
9052 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9053 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
9055 atomic_dec(&root->fs_info->async_submit_draining);
9059 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9062 struct btrfs_root *root;
9063 struct list_head splice;
9066 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9069 INIT_LIST_HEAD(&splice);
9071 mutex_lock(&fs_info->delalloc_root_mutex);
9072 spin_lock(&fs_info->delalloc_root_lock);
9073 list_splice_init(&fs_info->delalloc_roots, &splice);
9074 while (!list_empty(&splice) && nr) {
9075 root = list_first_entry(&splice, struct btrfs_root,
9077 root = btrfs_grab_fs_root(root);
9079 list_move_tail(&root->delalloc_root,
9080 &fs_info->delalloc_roots);
9081 spin_unlock(&fs_info->delalloc_root_lock);
9083 ret = __start_delalloc_inodes(root, delay_iput, nr);
9084 btrfs_put_fs_root(root);
9092 spin_lock(&fs_info->delalloc_root_lock);
9094 spin_unlock(&fs_info->delalloc_root_lock);
9097 atomic_inc(&fs_info->async_submit_draining);
9098 while (atomic_read(&fs_info->nr_async_submits) ||
9099 atomic_read(&fs_info->async_delalloc_pages)) {
9100 wait_event(fs_info->async_submit_wait,
9101 (atomic_read(&fs_info->nr_async_submits) == 0 &&
9102 atomic_read(&fs_info->async_delalloc_pages) == 0));
9104 atomic_dec(&fs_info->async_submit_draining);
9106 if (!list_empty_careful(&splice)) {
9107 spin_lock(&fs_info->delalloc_root_lock);
9108 list_splice_tail(&splice, &fs_info->delalloc_roots);
9109 spin_unlock(&fs_info->delalloc_root_lock);
9111 mutex_unlock(&fs_info->delalloc_root_mutex);
9115 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9116 const char *symname)
9118 struct btrfs_trans_handle *trans;
9119 struct btrfs_root *root = BTRFS_I(dir)->root;
9120 struct btrfs_path *path;
9121 struct btrfs_key key;
9122 struct inode *inode = NULL;
9130 struct btrfs_file_extent_item *ei;
9131 struct extent_buffer *leaf;
9133 name_len = strlen(symname);
9134 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9135 return -ENAMETOOLONG;
9138 * 2 items for inode item and ref
9139 * 2 items for dir items
9140 * 1 item for xattr if selinux is on
9142 trans = btrfs_start_transaction(root, 5);
9144 return PTR_ERR(trans);
9146 err = btrfs_find_free_ino(root, &objectid);
9150 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9151 dentry->d_name.len, btrfs_ino(dir), objectid,
9152 S_IFLNK|S_IRWXUGO, &index);
9153 if (IS_ERR(inode)) {
9154 err = PTR_ERR(inode);
9159 * If the active LSM wants to access the inode during
9160 * d_instantiate it needs these. Smack checks to see
9161 * if the filesystem supports xattrs by looking at the
9164 inode->i_fop = &btrfs_file_operations;
9165 inode->i_op = &btrfs_file_inode_operations;
9166 inode->i_mapping->a_ops = &btrfs_aops;
9167 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
9168 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9170 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9172 goto out_unlock_inode;
9174 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
9176 goto out_unlock_inode;
9178 path = btrfs_alloc_path();
9181 goto out_unlock_inode;
9183 key.objectid = btrfs_ino(inode);
9185 key.type = BTRFS_EXTENT_DATA_KEY;
9186 datasize = btrfs_file_extent_calc_inline_size(name_len);
9187 err = btrfs_insert_empty_item(trans, root, path, &key,
9190 btrfs_free_path(path);
9191 goto out_unlock_inode;
9193 leaf = path->nodes[0];
9194 ei = btrfs_item_ptr(leaf, path->slots[0],
9195 struct btrfs_file_extent_item);
9196 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9197 btrfs_set_file_extent_type(leaf, ei,
9198 BTRFS_FILE_EXTENT_INLINE);
9199 btrfs_set_file_extent_encryption(leaf, ei, 0);
9200 btrfs_set_file_extent_compression(leaf, ei, 0);
9201 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9202 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9204 ptr = btrfs_file_extent_inline_start(ei);
9205 write_extent_buffer(leaf, symname, ptr, name_len);
9206 btrfs_mark_buffer_dirty(leaf);
9207 btrfs_free_path(path);
9209 inode->i_op = &btrfs_symlink_inode_operations;
9210 inode->i_mapping->a_ops = &btrfs_symlink_aops;
9211 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
9212 inode_set_bytes(inode, name_len);
9213 btrfs_i_size_write(inode, name_len);
9214 err = btrfs_update_inode(trans, root, inode);
9217 goto out_unlock_inode;
9220 unlock_new_inode(inode);
9221 d_instantiate(dentry, inode);
9224 btrfs_end_transaction(trans, root);
9226 inode_dec_link_count(inode);
9229 btrfs_btree_balance_dirty(root);
9234 unlock_new_inode(inode);
9238 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9239 u64 start, u64 num_bytes, u64 min_size,
9240 loff_t actual_len, u64 *alloc_hint,
9241 struct btrfs_trans_handle *trans)
9243 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9244 struct extent_map *em;
9245 struct btrfs_root *root = BTRFS_I(inode)->root;
9246 struct btrfs_key ins;
9247 u64 cur_offset = start;
9251 bool own_trans = true;
9255 while (num_bytes > 0) {
9257 trans = btrfs_start_transaction(root, 3);
9258 if (IS_ERR(trans)) {
9259 ret = PTR_ERR(trans);
9264 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
9265 cur_bytes = max(cur_bytes, min_size);
9266 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
9267 *alloc_hint, &ins, 1, 0);
9270 btrfs_end_transaction(trans, root);
9274 ret = insert_reserved_file_extent(trans, inode,
9275 cur_offset, ins.objectid,
9276 ins.offset, ins.offset,
9277 ins.offset, 0, 0, 0,
9278 BTRFS_FILE_EXTENT_PREALLOC);
9280 btrfs_free_reserved_extent(root, ins.objectid,
9282 btrfs_abort_transaction(trans, root, ret);
9284 btrfs_end_transaction(trans, root);
9287 btrfs_drop_extent_cache(inode, cur_offset,
9288 cur_offset + ins.offset -1, 0);
9290 em = alloc_extent_map();
9292 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9293 &BTRFS_I(inode)->runtime_flags);
9297 em->start = cur_offset;
9298 em->orig_start = cur_offset;
9299 em->len = ins.offset;
9300 em->block_start = ins.objectid;
9301 em->block_len = ins.offset;
9302 em->orig_block_len = ins.offset;
9303 em->ram_bytes = ins.offset;
9304 em->bdev = root->fs_info->fs_devices->latest_bdev;
9305 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9306 em->generation = trans->transid;
9309 write_lock(&em_tree->lock);
9310 ret = add_extent_mapping(em_tree, em, 1);
9311 write_unlock(&em_tree->lock);
9314 btrfs_drop_extent_cache(inode, cur_offset,
9315 cur_offset + ins.offset - 1,
9318 free_extent_map(em);
9320 num_bytes -= ins.offset;
9321 cur_offset += ins.offset;
9322 *alloc_hint = ins.objectid + ins.offset;
9324 inode_inc_iversion(inode);
9325 inode->i_ctime = CURRENT_TIME;
9326 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9327 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9328 (actual_len > inode->i_size) &&
9329 (cur_offset > inode->i_size)) {
9330 if (cur_offset > actual_len)
9331 i_size = actual_len;
9333 i_size = cur_offset;
9334 i_size_write(inode, i_size);
9335 btrfs_ordered_update_i_size(inode, i_size, NULL);
9338 ret = btrfs_update_inode(trans, root, inode);
9341 btrfs_abort_transaction(trans, root, ret);
9343 btrfs_end_transaction(trans, root);
9348 btrfs_end_transaction(trans, root);
9353 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9354 u64 start, u64 num_bytes, u64 min_size,
9355 loff_t actual_len, u64 *alloc_hint)
9357 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9358 min_size, actual_len, alloc_hint,
9362 int btrfs_prealloc_file_range_trans(struct inode *inode,
9363 struct btrfs_trans_handle *trans, int mode,
9364 u64 start, u64 num_bytes, u64 min_size,
9365 loff_t actual_len, u64 *alloc_hint)
9367 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9368 min_size, actual_len, alloc_hint, trans);
9371 static int btrfs_set_page_dirty(struct page *page)
9373 return __set_page_dirty_nobuffers(page);
9376 static int btrfs_permission(struct inode *inode, int mask)
9378 struct btrfs_root *root = BTRFS_I(inode)->root;
9379 umode_t mode = inode->i_mode;
9381 if (mask & MAY_WRITE &&
9382 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9383 if (btrfs_root_readonly(root))
9385 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9388 return generic_permission(inode, mask);
9391 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9393 struct btrfs_trans_handle *trans;
9394 struct btrfs_root *root = BTRFS_I(dir)->root;
9395 struct inode *inode = NULL;
9401 * 5 units required for adding orphan entry
9403 trans = btrfs_start_transaction(root, 5);
9405 return PTR_ERR(trans);
9407 ret = btrfs_find_free_ino(root, &objectid);
9411 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9412 btrfs_ino(dir), objectid, mode, &index);
9413 if (IS_ERR(inode)) {
9414 ret = PTR_ERR(inode);
9419 inode->i_fop = &btrfs_file_operations;
9420 inode->i_op = &btrfs_file_inode_operations;
9422 inode->i_mapping->a_ops = &btrfs_aops;
9423 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
9424 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9426 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9430 ret = btrfs_update_inode(trans, root, inode);
9433 ret = btrfs_orphan_add(trans, inode);
9438 * We set number of links to 0 in btrfs_new_inode(), and here we set
9439 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9442 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9444 set_nlink(inode, 1);
9445 unlock_new_inode(inode);
9446 d_tmpfile(dentry, inode);
9447 mark_inode_dirty(inode);
9450 btrfs_end_transaction(trans, root);
9453 btrfs_balance_delayed_items(root);
9454 btrfs_btree_balance_dirty(root);
9458 unlock_new_inode(inode);
9463 static const struct inode_operations btrfs_dir_inode_operations = {
9464 .getattr = btrfs_getattr,
9465 .lookup = btrfs_lookup,
9466 .create = btrfs_create,
9467 .unlink = btrfs_unlink,
9469 .mkdir = btrfs_mkdir,
9470 .rmdir = btrfs_rmdir,
9471 .rename2 = btrfs_rename2,
9472 .symlink = btrfs_symlink,
9473 .setattr = btrfs_setattr,
9474 .mknod = btrfs_mknod,
9475 .setxattr = btrfs_setxattr,
9476 .getxattr = btrfs_getxattr,
9477 .listxattr = btrfs_listxattr,
9478 .removexattr = btrfs_removexattr,
9479 .permission = btrfs_permission,
9480 .get_acl = btrfs_get_acl,
9481 .set_acl = btrfs_set_acl,
9482 .update_time = btrfs_update_time,
9483 .tmpfile = btrfs_tmpfile,
9485 static const struct inode_operations btrfs_dir_ro_inode_operations = {
9486 .lookup = btrfs_lookup,
9487 .permission = btrfs_permission,
9488 .get_acl = btrfs_get_acl,
9489 .set_acl = btrfs_set_acl,
9490 .update_time = btrfs_update_time,
9493 static const struct file_operations btrfs_dir_file_operations = {
9494 .llseek = generic_file_llseek,
9495 .read = generic_read_dir,
9496 .iterate = btrfs_real_readdir,
9497 .unlocked_ioctl = btrfs_ioctl,
9498 #ifdef CONFIG_COMPAT
9499 .compat_ioctl = btrfs_ioctl,
9501 .release = btrfs_release_file,
9502 .fsync = btrfs_sync_file,
9505 static struct extent_io_ops btrfs_extent_io_ops = {
9506 .fill_delalloc = run_delalloc_range,
9507 .submit_bio_hook = btrfs_submit_bio_hook,
9508 .merge_bio_hook = btrfs_merge_bio_hook,
9509 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
9510 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
9511 .writepage_start_hook = btrfs_writepage_start_hook,
9512 .set_bit_hook = btrfs_set_bit_hook,
9513 .clear_bit_hook = btrfs_clear_bit_hook,
9514 .merge_extent_hook = btrfs_merge_extent_hook,
9515 .split_extent_hook = btrfs_split_extent_hook,
9519 * btrfs doesn't support the bmap operation because swapfiles
9520 * use bmap to make a mapping of extents in the file. They assume
9521 * these extents won't change over the life of the file and they
9522 * use the bmap result to do IO directly to the drive.
9524 * the btrfs bmap call would return logical addresses that aren't
9525 * suitable for IO and they also will change frequently as COW
9526 * operations happen. So, swapfile + btrfs == corruption.
9528 * For now we're avoiding this by dropping bmap.
9530 static const struct address_space_operations btrfs_aops = {
9531 .readpage = btrfs_readpage,
9532 .writepage = btrfs_writepage,
9533 .writepages = btrfs_writepages,
9534 .readpages = btrfs_readpages,
9535 .direct_IO = btrfs_direct_IO,
9536 .invalidatepage = btrfs_invalidatepage,
9537 .releasepage = btrfs_releasepage,
9538 .set_page_dirty = btrfs_set_page_dirty,
9539 .error_remove_page = generic_error_remove_page,
9542 static const struct address_space_operations btrfs_symlink_aops = {
9543 .readpage = btrfs_readpage,
9544 .writepage = btrfs_writepage,
9545 .invalidatepage = btrfs_invalidatepage,
9546 .releasepage = btrfs_releasepage,
9549 static const struct inode_operations btrfs_file_inode_operations = {
9550 .getattr = btrfs_getattr,
9551 .setattr = btrfs_setattr,
9552 .setxattr = btrfs_setxattr,
9553 .getxattr = btrfs_getxattr,
9554 .listxattr = btrfs_listxattr,
9555 .removexattr = btrfs_removexattr,
9556 .permission = btrfs_permission,
9557 .fiemap = btrfs_fiemap,
9558 .get_acl = btrfs_get_acl,
9559 .set_acl = btrfs_set_acl,
9560 .update_time = btrfs_update_time,
9562 static const struct inode_operations btrfs_special_inode_operations = {
9563 .getattr = btrfs_getattr,
9564 .setattr = btrfs_setattr,
9565 .permission = btrfs_permission,
9566 .setxattr = btrfs_setxattr,
9567 .getxattr = btrfs_getxattr,
9568 .listxattr = btrfs_listxattr,
9569 .removexattr = btrfs_removexattr,
9570 .get_acl = btrfs_get_acl,
9571 .set_acl = btrfs_set_acl,
9572 .update_time = btrfs_update_time,
9574 static const struct inode_operations btrfs_symlink_inode_operations = {
9575 .readlink = generic_readlink,
9576 .follow_link = page_follow_link_light,
9577 .put_link = page_put_link,
9578 .getattr = btrfs_getattr,
9579 .setattr = btrfs_setattr,
9580 .permission = btrfs_permission,
9581 .setxattr = btrfs_setxattr,
9582 .getxattr = btrfs_getxattr,
9583 .listxattr = btrfs_listxattr,
9584 .removexattr = btrfs_removexattr,
9585 .update_time = btrfs_update_time,
9588 const struct dentry_operations btrfs_dentry_operations = {
9589 .d_delete = btrfs_dentry_delete,
9590 .d_release = btrfs_dentry_release,