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 void 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 unsigned long page_error_op;
532 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
533 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
536 * inline extent creation worked or returned error,
537 * we don't need to create any more async work items.
538 * Unlock and free up our temp pages.
540 extent_clear_unlock_delalloc(inode, start, end, NULL,
541 clear_flags, PAGE_UNLOCK |
552 * we aren't doing an inline extent round the compressed size
553 * up to a block size boundary so the allocator does sane
556 total_compressed = ALIGN(total_compressed, blocksize);
559 * one last check to make sure the compression is really a
560 * win, compare the page count read with the blocks on disk
562 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
563 if (total_compressed >= total_in) {
566 num_bytes = total_in;
569 if (!will_compress && pages) {
571 * the compression code ran but failed to make things smaller,
572 * free any pages it allocated and our page pointer array
574 for (i = 0; i < nr_pages_ret; i++) {
575 WARN_ON(pages[i]->mapping);
576 page_cache_release(pages[i]);
580 total_compressed = 0;
583 /* flag the file so we don't compress in the future */
584 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
585 !(BTRFS_I(inode)->force_compress)) {
586 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
592 /* the async work queues will take care of doing actual
593 * allocation on disk for these compressed pages,
594 * and will submit them to the elevator.
596 add_async_extent(async_cow, start, num_bytes,
597 total_compressed, pages, nr_pages_ret,
600 if (start + num_bytes < end) {
607 cleanup_and_bail_uncompressed:
609 * No compression, but we still need to write the pages in
610 * the file we've been given so far. redirty the locked
611 * page if it corresponds to our extent and set things up
612 * for the async work queue to run cow_file_range to do
613 * the normal delalloc dance
615 if (page_offset(locked_page) >= start &&
616 page_offset(locked_page) <= end) {
617 __set_page_dirty_nobuffers(locked_page);
618 /* unlocked later on in the async handlers */
621 extent_range_redirty_for_io(inode, start, end);
622 add_async_extent(async_cow, start, end - start + 1,
623 0, NULL, 0, BTRFS_COMPRESS_NONE);
630 for (i = 0; i < nr_pages_ret; i++) {
631 WARN_ON(pages[i]->mapping);
632 page_cache_release(pages[i]);
637 static void free_async_extent_pages(struct async_extent *async_extent)
641 if (!async_extent->pages)
644 for (i = 0; i < async_extent->nr_pages; i++) {
645 WARN_ON(async_extent->pages[i]->mapping);
646 page_cache_release(async_extent->pages[i]);
648 kfree(async_extent->pages);
649 async_extent->nr_pages = 0;
650 async_extent->pages = NULL;
654 * phase two of compressed writeback. This is the ordered portion
655 * of the code, which only gets called in the order the work was
656 * queued. We walk all the async extents created by compress_file_range
657 * and send them down to the disk.
659 static noinline void submit_compressed_extents(struct inode *inode,
660 struct async_cow *async_cow)
662 struct async_extent *async_extent;
664 struct btrfs_key ins;
665 struct extent_map *em;
666 struct btrfs_root *root = BTRFS_I(inode)->root;
667 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
668 struct extent_io_tree *io_tree;
672 while (!list_empty(&async_cow->extents)) {
673 async_extent = list_entry(async_cow->extents.next,
674 struct async_extent, list);
675 list_del(&async_extent->list);
677 io_tree = &BTRFS_I(inode)->io_tree;
680 /* did the compression code fall back to uncompressed IO? */
681 if (!async_extent->pages) {
682 int page_started = 0;
683 unsigned long nr_written = 0;
685 lock_extent(io_tree, async_extent->start,
686 async_extent->start +
687 async_extent->ram_size - 1);
689 /* allocate blocks */
690 ret = cow_file_range(inode, async_cow->locked_page,
692 async_extent->start +
693 async_extent->ram_size - 1,
694 &page_started, &nr_written, 0);
699 * if page_started, cow_file_range inserted an
700 * inline extent and took care of all the unlocking
701 * and IO for us. Otherwise, we need to submit
702 * all those pages down to the drive.
704 if (!page_started && !ret)
705 extent_write_locked_range(io_tree,
706 inode, async_extent->start,
707 async_extent->start +
708 async_extent->ram_size - 1,
712 unlock_page(async_cow->locked_page);
718 lock_extent(io_tree, async_extent->start,
719 async_extent->start + async_extent->ram_size - 1);
721 ret = btrfs_reserve_extent(root,
722 async_extent->compressed_size,
723 async_extent->compressed_size,
724 0, alloc_hint, &ins, 1, 1);
726 free_async_extent_pages(async_extent);
728 if (ret == -ENOSPC) {
729 unlock_extent(io_tree, async_extent->start,
730 async_extent->start +
731 async_extent->ram_size - 1);
734 * we need to redirty the pages if we decide to
735 * fallback to uncompressed IO, otherwise we
736 * will not submit these pages down to lower
739 extent_range_redirty_for_io(inode,
741 async_extent->start +
742 async_extent->ram_size - 1);
750 * here we're doing allocation and writeback of the
753 btrfs_drop_extent_cache(inode, async_extent->start,
754 async_extent->start +
755 async_extent->ram_size - 1, 0);
757 em = alloc_extent_map();
760 goto out_free_reserve;
762 em->start = async_extent->start;
763 em->len = async_extent->ram_size;
764 em->orig_start = em->start;
765 em->mod_start = em->start;
766 em->mod_len = em->len;
768 em->block_start = ins.objectid;
769 em->block_len = ins.offset;
770 em->orig_block_len = ins.offset;
771 em->ram_bytes = async_extent->ram_size;
772 em->bdev = root->fs_info->fs_devices->latest_bdev;
773 em->compress_type = async_extent->compress_type;
774 set_bit(EXTENT_FLAG_PINNED, &em->flags);
775 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
779 write_lock(&em_tree->lock);
780 ret = add_extent_mapping(em_tree, em, 1);
781 write_unlock(&em_tree->lock);
782 if (ret != -EEXIST) {
786 btrfs_drop_extent_cache(inode, async_extent->start,
787 async_extent->start +
788 async_extent->ram_size - 1, 0);
792 goto out_free_reserve;
794 ret = btrfs_add_ordered_extent_compress(inode,
797 async_extent->ram_size,
799 BTRFS_ORDERED_COMPRESSED,
800 async_extent->compress_type);
802 btrfs_drop_extent_cache(inode, async_extent->start,
803 async_extent->start +
804 async_extent->ram_size - 1, 0);
805 goto out_free_reserve;
809 * clear dirty, set writeback and unlock the pages.
811 extent_clear_unlock_delalloc(inode, async_extent->start,
812 async_extent->start +
813 async_extent->ram_size - 1,
814 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
815 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
817 ret = btrfs_submit_compressed_write(inode,
819 async_extent->ram_size,
821 ins.offset, async_extent->pages,
822 async_extent->nr_pages);
824 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
825 struct page *p = async_extent->pages[0];
826 const u64 start = async_extent->start;
827 const u64 end = start + async_extent->ram_size - 1;
829 p->mapping = inode->i_mapping;
830 tree->ops->writepage_end_io_hook(p, start, end,
833 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
836 free_async_extent_pages(async_extent);
838 alloc_hint = ins.objectid + ins.offset;
844 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
846 extent_clear_unlock_delalloc(inode, async_extent->start,
847 async_extent->start +
848 async_extent->ram_size - 1,
849 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
850 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
851 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
852 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
854 free_async_extent_pages(async_extent);
859 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
862 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
863 struct extent_map *em;
866 read_lock(&em_tree->lock);
867 em = search_extent_mapping(em_tree, start, num_bytes);
870 * if block start isn't an actual block number then find the
871 * first block in this inode and use that as a hint. If that
872 * block is also bogus then just don't worry about it.
874 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
876 em = search_extent_mapping(em_tree, 0, 0);
877 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
878 alloc_hint = em->block_start;
882 alloc_hint = em->block_start;
886 read_unlock(&em_tree->lock);
892 * when extent_io.c finds a delayed allocation range in the file,
893 * the call backs end up in this code. The basic idea is to
894 * allocate extents on disk for the range, and create ordered data structs
895 * in ram to track those extents.
897 * locked_page is the page that writepage had locked already. We use
898 * it to make sure we don't do extra locks or unlocks.
900 * *page_started is set to one if we unlock locked_page and do everything
901 * required to start IO on it. It may be clean and already done with
904 static noinline int cow_file_range(struct inode *inode,
905 struct page *locked_page,
906 u64 start, u64 end, int *page_started,
907 unsigned long *nr_written,
910 struct btrfs_root *root = BTRFS_I(inode)->root;
913 unsigned long ram_size;
916 u64 blocksize = root->sectorsize;
917 struct btrfs_key ins;
918 struct extent_map *em;
919 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
922 if (btrfs_is_free_space_inode(inode)) {
928 num_bytes = ALIGN(end - start + 1, blocksize);
929 num_bytes = max(blocksize, num_bytes);
930 disk_num_bytes = num_bytes;
932 /* if this is a small write inside eof, kick off defrag */
933 if (num_bytes < 64 * 1024 &&
934 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
935 btrfs_add_inode_defrag(NULL, inode);
938 /* lets try to make an inline extent */
939 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
942 extent_clear_unlock_delalloc(inode, start, end, NULL,
943 EXTENT_LOCKED | EXTENT_DELALLOC |
944 EXTENT_DEFRAG, PAGE_UNLOCK |
945 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
948 *nr_written = *nr_written +
949 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
952 } else if (ret < 0) {
957 BUG_ON(disk_num_bytes >
958 btrfs_super_total_bytes(root->fs_info->super_copy));
960 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
961 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
963 while (disk_num_bytes > 0) {
966 cur_alloc_size = disk_num_bytes;
967 ret = btrfs_reserve_extent(root, cur_alloc_size,
968 root->sectorsize, 0, alloc_hint,
973 em = alloc_extent_map();
979 em->orig_start = em->start;
980 ram_size = ins.offset;
981 em->len = ins.offset;
982 em->mod_start = em->start;
983 em->mod_len = em->len;
985 em->block_start = ins.objectid;
986 em->block_len = ins.offset;
987 em->orig_block_len = ins.offset;
988 em->ram_bytes = ram_size;
989 em->bdev = root->fs_info->fs_devices->latest_bdev;
990 set_bit(EXTENT_FLAG_PINNED, &em->flags);
994 write_lock(&em_tree->lock);
995 ret = add_extent_mapping(em_tree, em, 1);
996 write_unlock(&em_tree->lock);
997 if (ret != -EEXIST) {
1001 btrfs_drop_extent_cache(inode, start,
1002 start + ram_size - 1, 0);
1007 cur_alloc_size = ins.offset;
1008 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1009 ram_size, cur_alloc_size, 0);
1011 goto out_drop_extent_cache;
1013 if (root->root_key.objectid ==
1014 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1015 ret = btrfs_reloc_clone_csums(inode, start,
1018 goto out_drop_extent_cache;
1021 if (disk_num_bytes < cur_alloc_size)
1024 /* we're not doing compressed IO, don't unlock the first
1025 * page (which the caller expects to stay locked), don't
1026 * clear any dirty bits and don't set any writeback bits
1028 * Do set the Private2 bit so we know this page was properly
1029 * setup for writepage
1031 op = unlock ? PAGE_UNLOCK : 0;
1032 op |= PAGE_SET_PRIVATE2;
1034 extent_clear_unlock_delalloc(inode, start,
1035 start + ram_size - 1, locked_page,
1036 EXTENT_LOCKED | EXTENT_DELALLOC,
1038 disk_num_bytes -= cur_alloc_size;
1039 num_bytes -= cur_alloc_size;
1040 alloc_hint = ins.objectid + ins.offset;
1041 start += cur_alloc_size;
1046 out_drop_extent_cache:
1047 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1049 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1051 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1052 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1053 EXTENT_DELALLOC | EXTENT_DEFRAG,
1054 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1055 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1060 * work queue call back to started compression on a file and pages
1062 static noinline void async_cow_start(struct btrfs_work *work)
1064 struct async_cow *async_cow;
1066 async_cow = container_of(work, struct async_cow, work);
1068 compress_file_range(async_cow->inode, async_cow->locked_page,
1069 async_cow->start, async_cow->end, async_cow,
1071 if (num_added == 0) {
1072 btrfs_add_delayed_iput(async_cow->inode);
1073 async_cow->inode = NULL;
1078 * work queue call back to submit previously compressed pages
1080 static noinline void async_cow_submit(struct btrfs_work *work)
1082 struct async_cow *async_cow;
1083 struct btrfs_root *root;
1084 unsigned long nr_pages;
1086 async_cow = container_of(work, struct async_cow, work);
1088 root = async_cow->root;
1089 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1092 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1094 waitqueue_active(&root->fs_info->async_submit_wait))
1095 wake_up(&root->fs_info->async_submit_wait);
1097 if (async_cow->inode)
1098 submit_compressed_extents(async_cow->inode, async_cow);
1101 static noinline void async_cow_free(struct btrfs_work *work)
1103 struct async_cow *async_cow;
1104 async_cow = container_of(work, struct async_cow, work);
1105 if (async_cow->inode)
1106 btrfs_add_delayed_iput(async_cow->inode);
1110 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1111 u64 start, u64 end, int *page_started,
1112 unsigned long *nr_written)
1114 struct async_cow *async_cow;
1115 struct btrfs_root *root = BTRFS_I(inode)->root;
1116 unsigned long nr_pages;
1118 int limit = 10 * 1024 * 1024;
1120 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1121 1, 0, NULL, GFP_NOFS);
1122 while (start < end) {
1123 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1124 BUG_ON(!async_cow); /* -ENOMEM */
1125 async_cow->inode = igrab(inode);
1126 async_cow->root = root;
1127 async_cow->locked_page = locked_page;
1128 async_cow->start = start;
1130 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1131 !btrfs_test_opt(root, FORCE_COMPRESS))
1134 cur_end = min(end, start + 512 * 1024 - 1);
1136 async_cow->end = cur_end;
1137 INIT_LIST_HEAD(&async_cow->extents);
1139 btrfs_init_work(&async_cow->work,
1140 btrfs_delalloc_helper,
1141 async_cow_start, async_cow_submit,
1144 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1146 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1148 btrfs_queue_work(root->fs_info->delalloc_workers,
1151 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1152 wait_event(root->fs_info->async_submit_wait,
1153 (atomic_read(&root->fs_info->async_delalloc_pages) <
1157 while (atomic_read(&root->fs_info->async_submit_draining) &&
1158 atomic_read(&root->fs_info->async_delalloc_pages)) {
1159 wait_event(root->fs_info->async_submit_wait,
1160 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1164 *nr_written += nr_pages;
1165 start = cur_end + 1;
1171 static noinline int csum_exist_in_range(struct btrfs_root *root,
1172 u64 bytenr, u64 num_bytes)
1175 struct btrfs_ordered_sum *sums;
1178 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1179 bytenr + num_bytes - 1, &list, 0);
1180 if (ret == 0 && list_empty(&list))
1183 while (!list_empty(&list)) {
1184 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1185 list_del(&sums->list);
1192 * when nowcow writeback call back. This checks for snapshots or COW copies
1193 * of the extents that exist in the file, and COWs the file as required.
1195 * If no cow copies or snapshots exist, we write directly to the existing
1198 static noinline int run_delalloc_nocow(struct inode *inode,
1199 struct page *locked_page,
1200 u64 start, u64 end, int *page_started, int force,
1201 unsigned long *nr_written)
1203 struct btrfs_root *root = BTRFS_I(inode)->root;
1204 struct btrfs_trans_handle *trans;
1205 struct extent_buffer *leaf;
1206 struct btrfs_path *path;
1207 struct btrfs_file_extent_item *fi;
1208 struct btrfs_key found_key;
1223 u64 ino = btrfs_ino(inode);
1225 path = btrfs_alloc_path();
1227 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1228 EXTENT_LOCKED | EXTENT_DELALLOC |
1229 EXTENT_DO_ACCOUNTING |
1230 EXTENT_DEFRAG, PAGE_UNLOCK |
1232 PAGE_SET_WRITEBACK |
1233 PAGE_END_WRITEBACK);
1237 nolock = btrfs_is_free_space_inode(inode);
1240 trans = btrfs_join_transaction_nolock(root);
1242 trans = btrfs_join_transaction(root);
1244 if (IS_ERR(trans)) {
1245 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1246 EXTENT_LOCKED | EXTENT_DELALLOC |
1247 EXTENT_DO_ACCOUNTING |
1248 EXTENT_DEFRAG, PAGE_UNLOCK |
1250 PAGE_SET_WRITEBACK |
1251 PAGE_END_WRITEBACK);
1252 btrfs_free_path(path);
1253 return PTR_ERR(trans);
1256 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1258 cow_start = (u64)-1;
1261 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1265 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1266 leaf = path->nodes[0];
1267 btrfs_item_key_to_cpu(leaf, &found_key,
1268 path->slots[0] - 1);
1269 if (found_key.objectid == ino &&
1270 found_key.type == BTRFS_EXTENT_DATA_KEY)
1275 leaf = path->nodes[0];
1276 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1277 ret = btrfs_next_leaf(root, path);
1282 leaf = path->nodes[0];
1288 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1290 if (found_key.objectid > ino ||
1291 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1292 found_key.offset > end)
1295 if (found_key.offset > cur_offset) {
1296 extent_end = found_key.offset;
1301 fi = btrfs_item_ptr(leaf, path->slots[0],
1302 struct btrfs_file_extent_item);
1303 extent_type = btrfs_file_extent_type(leaf, fi);
1305 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1306 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1307 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1308 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1309 extent_offset = btrfs_file_extent_offset(leaf, fi);
1310 extent_end = found_key.offset +
1311 btrfs_file_extent_num_bytes(leaf, fi);
1313 btrfs_file_extent_disk_num_bytes(leaf, fi);
1314 if (extent_end <= start) {
1318 if (disk_bytenr == 0)
1320 if (btrfs_file_extent_compression(leaf, fi) ||
1321 btrfs_file_extent_encryption(leaf, fi) ||
1322 btrfs_file_extent_other_encoding(leaf, fi))
1324 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1326 if (btrfs_extent_readonly(root, disk_bytenr))
1328 if (btrfs_cross_ref_exist(trans, root, ino,
1330 extent_offset, disk_bytenr))
1332 disk_bytenr += extent_offset;
1333 disk_bytenr += cur_offset - found_key.offset;
1334 num_bytes = min(end + 1, extent_end) - cur_offset;
1336 * if there are pending snapshots for this root,
1337 * we fall into common COW way.
1340 err = btrfs_start_write_no_snapshoting(root);
1345 * force cow if csum exists in the range.
1346 * this ensure that csum for a given extent are
1347 * either valid or do not exist.
1349 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1352 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1353 extent_end = found_key.offset +
1354 btrfs_file_extent_inline_len(leaf,
1355 path->slots[0], fi);
1356 extent_end = ALIGN(extent_end, root->sectorsize);
1361 if (extent_end <= start) {
1363 if (!nolock && nocow)
1364 btrfs_end_write_no_snapshoting(root);
1368 if (cow_start == (u64)-1)
1369 cow_start = cur_offset;
1370 cur_offset = extent_end;
1371 if (cur_offset > end)
1377 btrfs_release_path(path);
1378 if (cow_start != (u64)-1) {
1379 ret = cow_file_range(inode, locked_page,
1380 cow_start, found_key.offset - 1,
1381 page_started, nr_written, 1);
1383 if (!nolock && nocow)
1384 btrfs_end_write_no_snapshoting(root);
1387 cow_start = (u64)-1;
1390 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1391 struct extent_map *em;
1392 struct extent_map_tree *em_tree;
1393 em_tree = &BTRFS_I(inode)->extent_tree;
1394 em = alloc_extent_map();
1395 BUG_ON(!em); /* -ENOMEM */
1396 em->start = cur_offset;
1397 em->orig_start = found_key.offset - extent_offset;
1398 em->len = num_bytes;
1399 em->block_len = num_bytes;
1400 em->block_start = disk_bytenr;
1401 em->orig_block_len = disk_num_bytes;
1402 em->ram_bytes = ram_bytes;
1403 em->bdev = root->fs_info->fs_devices->latest_bdev;
1404 em->mod_start = em->start;
1405 em->mod_len = em->len;
1406 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1407 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1408 em->generation = -1;
1410 write_lock(&em_tree->lock);
1411 ret = add_extent_mapping(em_tree, em, 1);
1412 write_unlock(&em_tree->lock);
1413 if (ret != -EEXIST) {
1414 free_extent_map(em);
1417 btrfs_drop_extent_cache(inode, em->start,
1418 em->start + em->len - 1, 0);
1420 type = BTRFS_ORDERED_PREALLOC;
1422 type = BTRFS_ORDERED_NOCOW;
1425 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1426 num_bytes, num_bytes, type);
1427 BUG_ON(ret); /* -ENOMEM */
1429 if (root->root_key.objectid ==
1430 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1431 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1434 if (!nolock && nocow)
1435 btrfs_end_write_no_snapshoting(root);
1440 extent_clear_unlock_delalloc(inode, cur_offset,
1441 cur_offset + num_bytes - 1,
1442 locked_page, EXTENT_LOCKED |
1443 EXTENT_DELALLOC, PAGE_UNLOCK |
1445 if (!nolock && nocow)
1446 btrfs_end_write_no_snapshoting(root);
1447 cur_offset = extent_end;
1448 if (cur_offset > end)
1451 btrfs_release_path(path);
1453 if (cur_offset <= end && cow_start == (u64)-1) {
1454 cow_start = cur_offset;
1458 if (cow_start != (u64)-1) {
1459 ret = cow_file_range(inode, locked_page, cow_start, end,
1460 page_started, nr_written, 1);
1466 err = btrfs_end_transaction(trans, root);
1470 if (ret && cur_offset < end)
1471 extent_clear_unlock_delalloc(inode, cur_offset, end,
1472 locked_page, EXTENT_LOCKED |
1473 EXTENT_DELALLOC | EXTENT_DEFRAG |
1474 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1476 PAGE_SET_WRITEBACK |
1477 PAGE_END_WRITEBACK);
1478 btrfs_free_path(path);
1482 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1485 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1486 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1490 * @defrag_bytes is a hint value, no spinlock held here,
1491 * if is not zero, it means the file is defragging.
1492 * Force cow if given extent needs to be defragged.
1494 if (BTRFS_I(inode)->defrag_bytes &&
1495 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1496 EXTENT_DEFRAG, 0, NULL))
1503 * extent_io.c call back to do delayed allocation processing
1505 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1506 u64 start, u64 end, int *page_started,
1507 unsigned long *nr_written)
1510 int force_cow = need_force_cow(inode, start, end);
1512 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1513 ret = run_delalloc_nocow(inode, locked_page, start, end,
1514 page_started, 1, nr_written);
1515 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1516 ret = run_delalloc_nocow(inode, locked_page, start, end,
1517 page_started, 0, nr_written);
1518 } else if (!inode_need_compress(inode)) {
1519 ret = cow_file_range(inode, locked_page, start, end,
1520 page_started, nr_written, 1);
1522 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1523 &BTRFS_I(inode)->runtime_flags);
1524 ret = cow_file_range_async(inode, locked_page, start, end,
1525 page_started, nr_written);
1530 static void btrfs_split_extent_hook(struct inode *inode,
1531 struct extent_state *orig, u64 split)
1535 /* not delalloc, ignore it */
1536 if (!(orig->state & EXTENT_DELALLOC))
1539 size = orig->end - orig->start + 1;
1540 if (size > BTRFS_MAX_EXTENT_SIZE) {
1545 * We need the largest size of the remaining extent to see if we
1546 * need to add a new outstanding extent. Think of the following
1549 * [MEAX_EXTENT_SIZEx2 - 4k][4k]
1551 * The new_size would just be 4k and we'd think we had enough
1552 * outstanding extents for this if we only took one side of the
1553 * split, same goes for the other direction. We need to see if
1554 * the larger size still is the same amount of extents as the
1555 * original size, because if it is we need to add a new
1556 * outstanding extent. But if we split up and the larger size
1557 * is less than the original then we are good to go since we've
1558 * already accounted for the extra extent in our original
1561 new_size = orig->end - split + 1;
1562 if ((split - orig->start) > new_size)
1563 new_size = split - orig->start;
1565 num_extents = div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1566 BTRFS_MAX_EXTENT_SIZE);
1567 if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1568 BTRFS_MAX_EXTENT_SIZE) < num_extents)
1572 spin_lock(&BTRFS_I(inode)->lock);
1573 BTRFS_I(inode)->outstanding_extents++;
1574 spin_unlock(&BTRFS_I(inode)->lock);
1578 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1579 * extents so we can keep track of new extents that are just merged onto old
1580 * extents, such as when we are doing sequential writes, so we can properly
1581 * account for the metadata space we'll need.
1583 static void btrfs_merge_extent_hook(struct inode *inode,
1584 struct extent_state *new,
1585 struct extent_state *other)
1587 u64 new_size, old_size;
1590 /* not delalloc, ignore it */
1591 if (!(other->state & EXTENT_DELALLOC))
1594 old_size = other->end - other->start + 1;
1595 new_size = old_size + (new->end - new->start + 1);
1597 /* we're not bigger than the max, unreserve the space and go */
1598 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1599 spin_lock(&BTRFS_I(inode)->lock);
1600 BTRFS_I(inode)->outstanding_extents--;
1601 spin_unlock(&BTRFS_I(inode)->lock);
1606 * If we grew by another max_extent, just return, we want to keep that
1609 num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1610 BTRFS_MAX_EXTENT_SIZE);
1611 if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1612 BTRFS_MAX_EXTENT_SIZE) > num_extents)
1615 spin_lock(&BTRFS_I(inode)->lock);
1616 BTRFS_I(inode)->outstanding_extents--;
1617 spin_unlock(&BTRFS_I(inode)->lock);
1620 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1621 struct inode *inode)
1623 spin_lock(&root->delalloc_lock);
1624 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1625 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1626 &root->delalloc_inodes);
1627 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1628 &BTRFS_I(inode)->runtime_flags);
1629 root->nr_delalloc_inodes++;
1630 if (root->nr_delalloc_inodes == 1) {
1631 spin_lock(&root->fs_info->delalloc_root_lock);
1632 BUG_ON(!list_empty(&root->delalloc_root));
1633 list_add_tail(&root->delalloc_root,
1634 &root->fs_info->delalloc_roots);
1635 spin_unlock(&root->fs_info->delalloc_root_lock);
1638 spin_unlock(&root->delalloc_lock);
1641 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1642 struct inode *inode)
1644 spin_lock(&root->delalloc_lock);
1645 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1646 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1647 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1648 &BTRFS_I(inode)->runtime_flags);
1649 root->nr_delalloc_inodes--;
1650 if (!root->nr_delalloc_inodes) {
1651 spin_lock(&root->fs_info->delalloc_root_lock);
1652 BUG_ON(list_empty(&root->delalloc_root));
1653 list_del_init(&root->delalloc_root);
1654 spin_unlock(&root->fs_info->delalloc_root_lock);
1657 spin_unlock(&root->delalloc_lock);
1661 * extent_io.c set_bit_hook, used to track delayed allocation
1662 * bytes in this file, and to maintain the list of inodes that
1663 * have pending delalloc work to be done.
1665 static void btrfs_set_bit_hook(struct inode *inode,
1666 struct extent_state *state, unsigned *bits)
1669 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1672 * set_bit and clear bit hooks normally require _irqsave/restore
1673 * but in this case, we are only testing for the DELALLOC
1674 * bit, which is only set or cleared with irqs on
1676 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1677 struct btrfs_root *root = BTRFS_I(inode)->root;
1678 u64 len = state->end + 1 - state->start;
1679 bool do_list = !btrfs_is_free_space_inode(inode);
1681 if (*bits & EXTENT_FIRST_DELALLOC) {
1682 *bits &= ~EXTENT_FIRST_DELALLOC;
1684 spin_lock(&BTRFS_I(inode)->lock);
1685 BTRFS_I(inode)->outstanding_extents++;
1686 spin_unlock(&BTRFS_I(inode)->lock);
1689 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1690 root->fs_info->delalloc_batch);
1691 spin_lock(&BTRFS_I(inode)->lock);
1692 BTRFS_I(inode)->delalloc_bytes += len;
1693 if (*bits & EXTENT_DEFRAG)
1694 BTRFS_I(inode)->defrag_bytes += len;
1695 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1696 &BTRFS_I(inode)->runtime_flags))
1697 btrfs_add_delalloc_inodes(root, inode);
1698 spin_unlock(&BTRFS_I(inode)->lock);
1703 * extent_io.c clear_bit_hook, see set_bit_hook for why
1705 static void btrfs_clear_bit_hook(struct inode *inode,
1706 struct extent_state *state,
1709 u64 len = state->end + 1 - state->start;
1710 u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1711 BTRFS_MAX_EXTENT_SIZE);
1713 spin_lock(&BTRFS_I(inode)->lock);
1714 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1715 BTRFS_I(inode)->defrag_bytes -= len;
1716 spin_unlock(&BTRFS_I(inode)->lock);
1719 * set_bit and clear bit hooks normally require _irqsave/restore
1720 * but in this case, we are only testing for the DELALLOC
1721 * bit, which is only set or cleared with irqs on
1723 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1724 struct btrfs_root *root = BTRFS_I(inode)->root;
1725 bool do_list = !btrfs_is_free_space_inode(inode);
1727 if (*bits & EXTENT_FIRST_DELALLOC) {
1728 *bits &= ~EXTENT_FIRST_DELALLOC;
1729 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1730 spin_lock(&BTRFS_I(inode)->lock);
1731 BTRFS_I(inode)->outstanding_extents -= num_extents;
1732 spin_unlock(&BTRFS_I(inode)->lock);
1736 * We don't reserve metadata space for space cache inodes so we
1737 * don't need to call dellalloc_release_metadata if there is an
1740 if (*bits & EXTENT_DO_ACCOUNTING &&
1741 root != root->fs_info->tree_root)
1742 btrfs_delalloc_release_metadata(inode, len);
1744 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1745 && do_list && !(state->state & EXTENT_NORESERVE))
1746 btrfs_free_reserved_data_space(inode, len);
1748 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1749 root->fs_info->delalloc_batch);
1750 spin_lock(&BTRFS_I(inode)->lock);
1751 BTRFS_I(inode)->delalloc_bytes -= len;
1752 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1753 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1754 &BTRFS_I(inode)->runtime_flags))
1755 btrfs_del_delalloc_inode(root, inode);
1756 spin_unlock(&BTRFS_I(inode)->lock);
1761 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1762 * we don't create bios that span stripes or chunks
1764 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1765 size_t size, struct bio *bio,
1766 unsigned long bio_flags)
1768 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1769 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1774 if (bio_flags & EXTENT_BIO_COMPRESSED)
1777 length = bio->bi_iter.bi_size;
1778 map_length = length;
1779 ret = btrfs_map_block(root->fs_info, rw, logical,
1780 &map_length, NULL, 0);
1781 /* Will always return 0 with map_multi == NULL */
1783 if (map_length < length + size)
1789 * in order to insert checksums into the metadata in large chunks,
1790 * we wait until bio submission time. All the pages in the bio are
1791 * checksummed and sums are attached onto the ordered extent record.
1793 * At IO completion time the cums attached on the ordered extent record
1794 * are inserted into the btree
1796 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1797 struct bio *bio, int mirror_num,
1798 unsigned long bio_flags,
1801 struct btrfs_root *root = BTRFS_I(inode)->root;
1804 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1805 BUG_ON(ret); /* -ENOMEM */
1810 * in order to insert checksums into the metadata in large chunks,
1811 * we wait until bio submission time. All the pages in the bio are
1812 * checksummed and sums are attached onto the ordered extent record.
1814 * At IO completion time the cums attached on the ordered extent record
1815 * are inserted into the btree
1817 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1818 int mirror_num, unsigned long bio_flags,
1821 struct btrfs_root *root = BTRFS_I(inode)->root;
1824 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1826 bio_endio(bio, ret);
1831 * extent_io.c submission hook. This does the right thing for csum calculation
1832 * on write, or reading the csums from the tree before a read
1834 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1835 int mirror_num, unsigned long bio_flags,
1838 struct btrfs_root *root = BTRFS_I(inode)->root;
1842 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1844 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1846 if (btrfs_is_free_space_inode(inode))
1849 if (!(rw & REQ_WRITE)) {
1850 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1854 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1855 ret = btrfs_submit_compressed_read(inode, bio,
1859 } else if (!skip_sum) {
1860 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1865 } else if (async && !skip_sum) {
1866 /* csum items have already been cloned */
1867 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1869 /* we're doing a write, do the async checksumming */
1870 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1871 inode, rw, bio, mirror_num,
1872 bio_flags, bio_offset,
1873 __btrfs_submit_bio_start,
1874 __btrfs_submit_bio_done);
1876 } else if (!skip_sum) {
1877 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1883 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1887 bio_endio(bio, ret);
1892 * given a list of ordered sums record them in the inode. This happens
1893 * at IO completion time based on sums calculated at bio submission time.
1895 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1896 struct inode *inode, u64 file_offset,
1897 struct list_head *list)
1899 struct btrfs_ordered_sum *sum;
1901 list_for_each_entry(sum, list, list) {
1902 trans->adding_csums = 1;
1903 btrfs_csum_file_blocks(trans,
1904 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1905 trans->adding_csums = 0;
1910 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1911 struct extent_state **cached_state)
1913 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1914 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1915 cached_state, GFP_NOFS);
1918 /* see btrfs_writepage_start_hook for details on why this is required */
1919 struct btrfs_writepage_fixup {
1921 struct btrfs_work work;
1924 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1926 struct btrfs_writepage_fixup *fixup;
1927 struct btrfs_ordered_extent *ordered;
1928 struct extent_state *cached_state = NULL;
1930 struct inode *inode;
1935 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1939 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1940 ClearPageChecked(page);
1944 inode = page->mapping->host;
1945 page_start = page_offset(page);
1946 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1948 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1951 /* already ordered? We're done */
1952 if (PagePrivate2(page))
1955 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1957 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1958 page_end, &cached_state, GFP_NOFS);
1960 btrfs_start_ordered_extent(inode, ordered, 1);
1961 btrfs_put_ordered_extent(ordered);
1965 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1967 mapping_set_error(page->mapping, ret);
1968 end_extent_writepage(page, ret, page_start, page_end);
1969 ClearPageChecked(page);
1973 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1974 ClearPageChecked(page);
1975 set_page_dirty(page);
1977 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1978 &cached_state, GFP_NOFS);
1981 page_cache_release(page);
1986 * There are a few paths in the higher layers of the kernel that directly
1987 * set the page dirty bit without asking the filesystem if it is a
1988 * good idea. This causes problems because we want to make sure COW
1989 * properly happens and the data=ordered rules are followed.
1991 * In our case any range that doesn't have the ORDERED bit set
1992 * hasn't been properly setup for IO. We kick off an async process
1993 * to fix it up. The async helper will wait for ordered extents, set
1994 * the delalloc bit and make it safe to write the page.
1996 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1998 struct inode *inode = page->mapping->host;
1999 struct btrfs_writepage_fixup *fixup;
2000 struct btrfs_root *root = BTRFS_I(inode)->root;
2002 /* this page is properly in the ordered list */
2003 if (TestClearPagePrivate2(page))
2006 if (PageChecked(page))
2009 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2013 SetPageChecked(page);
2014 page_cache_get(page);
2015 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2016 btrfs_writepage_fixup_worker, NULL, NULL);
2018 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
2022 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2023 struct inode *inode, u64 file_pos,
2024 u64 disk_bytenr, u64 disk_num_bytes,
2025 u64 num_bytes, u64 ram_bytes,
2026 u8 compression, u8 encryption,
2027 u16 other_encoding, int extent_type)
2029 struct btrfs_root *root = BTRFS_I(inode)->root;
2030 struct btrfs_file_extent_item *fi;
2031 struct btrfs_path *path;
2032 struct extent_buffer *leaf;
2033 struct btrfs_key ins;
2034 int extent_inserted = 0;
2037 path = btrfs_alloc_path();
2042 * we may be replacing one extent in the tree with another.
2043 * The new extent is pinned in the extent map, and we don't want
2044 * to drop it from the cache until it is completely in the btree.
2046 * So, tell btrfs_drop_extents to leave this extent in the cache.
2047 * the caller is expected to unpin it and allow it to be merged
2050 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2051 file_pos + num_bytes, NULL, 0,
2052 1, sizeof(*fi), &extent_inserted);
2056 if (!extent_inserted) {
2057 ins.objectid = btrfs_ino(inode);
2058 ins.offset = file_pos;
2059 ins.type = BTRFS_EXTENT_DATA_KEY;
2061 path->leave_spinning = 1;
2062 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2067 leaf = path->nodes[0];
2068 fi = btrfs_item_ptr(leaf, path->slots[0],
2069 struct btrfs_file_extent_item);
2070 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2071 btrfs_set_file_extent_type(leaf, fi, extent_type);
2072 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2073 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2074 btrfs_set_file_extent_offset(leaf, fi, 0);
2075 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2076 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2077 btrfs_set_file_extent_compression(leaf, fi, compression);
2078 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2079 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2081 btrfs_mark_buffer_dirty(leaf);
2082 btrfs_release_path(path);
2084 inode_add_bytes(inode, num_bytes);
2086 ins.objectid = disk_bytenr;
2087 ins.offset = disk_num_bytes;
2088 ins.type = BTRFS_EXTENT_ITEM_KEY;
2089 ret = btrfs_alloc_reserved_file_extent(trans, root,
2090 root->root_key.objectid,
2091 btrfs_ino(inode), file_pos, &ins);
2093 btrfs_free_path(path);
2098 /* snapshot-aware defrag */
2099 struct sa_defrag_extent_backref {
2100 struct rb_node node;
2101 struct old_sa_defrag_extent *old;
2110 struct old_sa_defrag_extent {
2111 struct list_head list;
2112 struct new_sa_defrag_extent *new;
2121 struct new_sa_defrag_extent {
2122 struct rb_root root;
2123 struct list_head head;
2124 struct btrfs_path *path;
2125 struct inode *inode;
2133 static int backref_comp(struct sa_defrag_extent_backref *b1,
2134 struct sa_defrag_extent_backref *b2)
2136 if (b1->root_id < b2->root_id)
2138 else if (b1->root_id > b2->root_id)
2141 if (b1->inum < b2->inum)
2143 else if (b1->inum > b2->inum)
2146 if (b1->file_pos < b2->file_pos)
2148 else if (b1->file_pos > b2->file_pos)
2152 * [------------------------------] ===> (a range of space)
2153 * |<--->| |<---->| =============> (fs/file tree A)
2154 * |<---------------------------->| ===> (fs/file tree B)
2156 * A range of space can refer to two file extents in one tree while
2157 * refer to only one file extent in another tree.
2159 * So we may process a disk offset more than one time(two extents in A)
2160 * and locate at the same extent(one extent in B), then insert two same
2161 * backrefs(both refer to the extent in B).
2166 static void backref_insert(struct rb_root *root,
2167 struct sa_defrag_extent_backref *backref)
2169 struct rb_node **p = &root->rb_node;
2170 struct rb_node *parent = NULL;
2171 struct sa_defrag_extent_backref *entry;
2176 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2178 ret = backref_comp(backref, entry);
2182 p = &(*p)->rb_right;
2185 rb_link_node(&backref->node, parent, p);
2186 rb_insert_color(&backref->node, root);
2190 * Note the backref might has changed, and in this case we just return 0.
2192 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2195 struct btrfs_file_extent_item *extent;
2196 struct btrfs_fs_info *fs_info;
2197 struct old_sa_defrag_extent *old = ctx;
2198 struct new_sa_defrag_extent *new = old->new;
2199 struct btrfs_path *path = new->path;
2200 struct btrfs_key key;
2201 struct btrfs_root *root;
2202 struct sa_defrag_extent_backref *backref;
2203 struct extent_buffer *leaf;
2204 struct inode *inode = new->inode;
2210 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2211 inum == btrfs_ino(inode))
2214 key.objectid = root_id;
2215 key.type = BTRFS_ROOT_ITEM_KEY;
2216 key.offset = (u64)-1;
2218 fs_info = BTRFS_I(inode)->root->fs_info;
2219 root = btrfs_read_fs_root_no_name(fs_info, &key);
2221 if (PTR_ERR(root) == -ENOENT)
2224 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2225 inum, offset, root_id);
2226 return PTR_ERR(root);
2229 key.objectid = inum;
2230 key.type = BTRFS_EXTENT_DATA_KEY;
2231 if (offset > (u64)-1 << 32)
2234 key.offset = offset;
2236 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2237 if (WARN_ON(ret < 0))
2244 leaf = path->nodes[0];
2245 slot = path->slots[0];
2247 if (slot >= btrfs_header_nritems(leaf)) {
2248 ret = btrfs_next_leaf(root, path);
2251 } else if (ret > 0) {
2260 btrfs_item_key_to_cpu(leaf, &key, slot);
2262 if (key.objectid > inum)
2265 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2268 extent = btrfs_item_ptr(leaf, slot,
2269 struct btrfs_file_extent_item);
2271 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2275 * 'offset' refers to the exact key.offset,
2276 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2277 * (key.offset - extent_offset).
2279 if (key.offset != offset)
2282 extent_offset = btrfs_file_extent_offset(leaf, extent);
2283 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2285 if (extent_offset >= old->extent_offset + old->offset +
2286 old->len || extent_offset + num_bytes <=
2287 old->extent_offset + old->offset)
2292 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2298 backref->root_id = root_id;
2299 backref->inum = inum;
2300 backref->file_pos = offset;
2301 backref->num_bytes = num_bytes;
2302 backref->extent_offset = extent_offset;
2303 backref->generation = btrfs_file_extent_generation(leaf, extent);
2305 backref_insert(&new->root, backref);
2308 btrfs_release_path(path);
2313 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2314 struct new_sa_defrag_extent *new)
2316 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2317 struct old_sa_defrag_extent *old, *tmp;
2322 list_for_each_entry_safe(old, tmp, &new->head, list) {
2323 ret = iterate_inodes_from_logical(old->bytenr +
2324 old->extent_offset, fs_info,
2325 path, record_one_backref,
2327 if (ret < 0 && ret != -ENOENT)
2330 /* no backref to be processed for this extent */
2332 list_del(&old->list);
2337 if (list_empty(&new->head))
2343 static int relink_is_mergable(struct extent_buffer *leaf,
2344 struct btrfs_file_extent_item *fi,
2345 struct new_sa_defrag_extent *new)
2347 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2350 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2353 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2356 if (btrfs_file_extent_encryption(leaf, fi) ||
2357 btrfs_file_extent_other_encoding(leaf, fi))
2364 * Note the backref might has changed, and in this case we just return 0.
2366 static noinline int relink_extent_backref(struct btrfs_path *path,
2367 struct sa_defrag_extent_backref *prev,
2368 struct sa_defrag_extent_backref *backref)
2370 struct btrfs_file_extent_item *extent;
2371 struct btrfs_file_extent_item *item;
2372 struct btrfs_ordered_extent *ordered;
2373 struct btrfs_trans_handle *trans;
2374 struct btrfs_fs_info *fs_info;
2375 struct btrfs_root *root;
2376 struct btrfs_key key;
2377 struct extent_buffer *leaf;
2378 struct old_sa_defrag_extent *old = backref->old;
2379 struct new_sa_defrag_extent *new = old->new;
2380 struct inode *src_inode = new->inode;
2381 struct inode *inode;
2382 struct extent_state *cached = NULL;
2391 if (prev && prev->root_id == backref->root_id &&
2392 prev->inum == backref->inum &&
2393 prev->file_pos + prev->num_bytes == backref->file_pos)
2396 /* step 1: get root */
2397 key.objectid = backref->root_id;
2398 key.type = BTRFS_ROOT_ITEM_KEY;
2399 key.offset = (u64)-1;
2401 fs_info = BTRFS_I(src_inode)->root->fs_info;
2402 index = srcu_read_lock(&fs_info->subvol_srcu);
2404 root = btrfs_read_fs_root_no_name(fs_info, &key);
2406 srcu_read_unlock(&fs_info->subvol_srcu, index);
2407 if (PTR_ERR(root) == -ENOENT)
2409 return PTR_ERR(root);
2412 if (btrfs_root_readonly(root)) {
2413 srcu_read_unlock(&fs_info->subvol_srcu, index);
2417 /* step 2: get inode */
2418 key.objectid = backref->inum;
2419 key.type = BTRFS_INODE_ITEM_KEY;
2422 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2423 if (IS_ERR(inode)) {
2424 srcu_read_unlock(&fs_info->subvol_srcu, index);
2428 srcu_read_unlock(&fs_info->subvol_srcu, index);
2430 /* step 3: relink backref */
2431 lock_start = backref->file_pos;
2432 lock_end = backref->file_pos + backref->num_bytes - 1;
2433 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2436 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2438 btrfs_put_ordered_extent(ordered);
2442 trans = btrfs_join_transaction(root);
2443 if (IS_ERR(trans)) {
2444 ret = PTR_ERR(trans);
2448 key.objectid = backref->inum;
2449 key.type = BTRFS_EXTENT_DATA_KEY;
2450 key.offset = backref->file_pos;
2452 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2455 } else if (ret > 0) {
2460 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2461 struct btrfs_file_extent_item);
2463 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2464 backref->generation)
2467 btrfs_release_path(path);
2469 start = backref->file_pos;
2470 if (backref->extent_offset < old->extent_offset + old->offset)
2471 start += old->extent_offset + old->offset -
2472 backref->extent_offset;
2474 len = min(backref->extent_offset + backref->num_bytes,
2475 old->extent_offset + old->offset + old->len);
2476 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2478 ret = btrfs_drop_extents(trans, root, inode, start,
2483 key.objectid = btrfs_ino(inode);
2484 key.type = BTRFS_EXTENT_DATA_KEY;
2487 path->leave_spinning = 1;
2489 struct btrfs_file_extent_item *fi;
2491 struct btrfs_key found_key;
2493 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2498 leaf = path->nodes[0];
2499 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2501 fi = btrfs_item_ptr(leaf, path->slots[0],
2502 struct btrfs_file_extent_item);
2503 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2505 if (extent_len + found_key.offset == start &&
2506 relink_is_mergable(leaf, fi, new)) {
2507 btrfs_set_file_extent_num_bytes(leaf, fi,
2509 btrfs_mark_buffer_dirty(leaf);
2510 inode_add_bytes(inode, len);
2516 btrfs_release_path(path);
2521 ret = btrfs_insert_empty_item(trans, root, path, &key,
2524 btrfs_abort_transaction(trans, root, ret);
2528 leaf = path->nodes[0];
2529 item = btrfs_item_ptr(leaf, path->slots[0],
2530 struct btrfs_file_extent_item);
2531 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2532 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2533 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2534 btrfs_set_file_extent_num_bytes(leaf, item, len);
2535 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2536 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2537 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2538 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2539 btrfs_set_file_extent_encryption(leaf, item, 0);
2540 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2542 btrfs_mark_buffer_dirty(leaf);
2543 inode_add_bytes(inode, len);
2544 btrfs_release_path(path);
2546 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2548 backref->root_id, backref->inum,
2549 new->file_pos, 0); /* start - extent_offset */
2551 btrfs_abort_transaction(trans, root, ret);
2557 btrfs_release_path(path);
2558 path->leave_spinning = 0;
2559 btrfs_end_transaction(trans, root);
2561 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2567 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2569 struct old_sa_defrag_extent *old, *tmp;
2574 list_for_each_entry_safe(old, tmp, &new->head, list) {
2575 list_del(&old->list);
2581 static void relink_file_extents(struct new_sa_defrag_extent *new)
2583 struct btrfs_path *path;
2584 struct sa_defrag_extent_backref *backref;
2585 struct sa_defrag_extent_backref *prev = NULL;
2586 struct inode *inode;
2587 struct btrfs_root *root;
2588 struct rb_node *node;
2592 root = BTRFS_I(inode)->root;
2594 path = btrfs_alloc_path();
2598 if (!record_extent_backrefs(path, new)) {
2599 btrfs_free_path(path);
2602 btrfs_release_path(path);
2605 node = rb_first(&new->root);
2608 rb_erase(node, &new->root);
2610 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2612 ret = relink_extent_backref(path, prev, backref);
2625 btrfs_free_path(path);
2627 free_sa_defrag_extent(new);
2629 atomic_dec(&root->fs_info->defrag_running);
2630 wake_up(&root->fs_info->transaction_wait);
2633 static struct new_sa_defrag_extent *
2634 record_old_file_extents(struct inode *inode,
2635 struct btrfs_ordered_extent *ordered)
2637 struct btrfs_root *root = BTRFS_I(inode)->root;
2638 struct btrfs_path *path;
2639 struct btrfs_key key;
2640 struct old_sa_defrag_extent *old;
2641 struct new_sa_defrag_extent *new;
2644 new = kmalloc(sizeof(*new), GFP_NOFS);
2649 new->file_pos = ordered->file_offset;
2650 new->len = ordered->len;
2651 new->bytenr = ordered->start;
2652 new->disk_len = ordered->disk_len;
2653 new->compress_type = ordered->compress_type;
2654 new->root = RB_ROOT;
2655 INIT_LIST_HEAD(&new->head);
2657 path = btrfs_alloc_path();
2661 key.objectid = btrfs_ino(inode);
2662 key.type = BTRFS_EXTENT_DATA_KEY;
2663 key.offset = new->file_pos;
2665 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2668 if (ret > 0 && path->slots[0] > 0)
2671 /* find out all the old extents for the file range */
2673 struct btrfs_file_extent_item *extent;
2674 struct extent_buffer *l;
2683 slot = path->slots[0];
2685 if (slot >= btrfs_header_nritems(l)) {
2686 ret = btrfs_next_leaf(root, path);
2694 btrfs_item_key_to_cpu(l, &key, slot);
2696 if (key.objectid != btrfs_ino(inode))
2698 if (key.type != BTRFS_EXTENT_DATA_KEY)
2700 if (key.offset >= new->file_pos + new->len)
2703 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2705 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2706 if (key.offset + num_bytes < new->file_pos)
2709 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2713 extent_offset = btrfs_file_extent_offset(l, extent);
2715 old = kmalloc(sizeof(*old), GFP_NOFS);
2719 offset = max(new->file_pos, key.offset);
2720 end = min(new->file_pos + new->len, key.offset + num_bytes);
2722 old->bytenr = disk_bytenr;
2723 old->extent_offset = extent_offset;
2724 old->offset = offset - key.offset;
2725 old->len = end - offset;
2728 list_add_tail(&old->list, &new->head);
2734 btrfs_free_path(path);
2735 atomic_inc(&root->fs_info->defrag_running);
2740 btrfs_free_path(path);
2742 free_sa_defrag_extent(new);
2746 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2749 struct btrfs_block_group_cache *cache;
2751 cache = btrfs_lookup_block_group(root->fs_info, start);
2754 spin_lock(&cache->lock);
2755 cache->delalloc_bytes -= len;
2756 spin_unlock(&cache->lock);
2758 btrfs_put_block_group(cache);
2761 /* as ordered data IO finishes, this gets called so we can finish
2762 * an ordered extent if the range of bytes in the file it covers are
2765 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2767 struct inode *inode = ordered_extent->inode;
2768 struct btrfs_root *root = BTRFS_I(inode)->root;
2769 struct btrfs_trans_handle *trans = NULL;
2770 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2771 struct extent_state *cached_state = NULL;
2772 struct new_sa_defrag_extent *new = NULL;
2773 int compress_type = 0;
2775 u64 logical_len = ordered_extent->len;
2777 bool truncated = false;
2779 nolock = btrfs_is_free_space_inode(inode);
2781 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2786 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2787 ordered_extent->file_offset +
2788 ordered_extent->len - 1);
2790 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2792 logical_len = ordered_extent->truncated_len;
2793 /* Truncated the entire extent, don't bother adding */
2798 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2799 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2800 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2802 trans = btrfs_join_transaction_nolock(root);
2804 trans = btrfs_join_transaction(root);
2805 if (IS_ERR(trans)) {
2806 ret = PTR_ERR(trans);
2810 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2811 ret = btrfs_update_inode_fallback(trans, root, inode);
2812 if (ret) /* -ENOMEM or corruption */
2813 btrfs_abort_transaction(trans, root, ret);
2817 lock_extent_bits(io_tree, ordered_extent->file_offset,
2818 ordered_extent->file_offset + ordered_extent->len - 1,
2821 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2822 ordered_extent->file_offset + ordered_extent->len - 1,
2823 EXTENT_DEFRAG, 1, cached_state);
2825 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2826 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2827 /* the inode is shared */
2828 new = record_old_file_extents(inode, ordered_extent);
2830 clear_extent_bit(io_tree, ordered_extent->file_offset,
2831 ordered_extent->file_offset + ordered_extent->len - 1,
2832 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2836 trans = btrfs_join_transaction_nolock(root);
2838 trans = btrfs_join_transaction(root);
2839 if (IS_ERR(trans)) {
2840 ret = PTR_ERR(trans);
2845 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2847 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2848 compress_type = ordered_extent->compress_type;
2849 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2850 BUG_ON(compress_type);
2851 ret = btrfs_mark_extent_written(trans, inode,
2852 ordered_extent->file_offset,
2853 ordered_extent->file_offset +
2856 BUG_ON(root == root->fs_info->tree_root);
2857 ret = insert_reserved_file_extent(trans, inode,
2858 ordered_extent->file_offset,
2859 ordered_extent->start,
2860 ordered_extent->disk_len,
2861 logical_len, logical_len,
2862 compress_type, 0, 0,
2863 BTRFS_FILE_EXTENT_REG);
2865 btrfs_release_delalloc_bytes(root,
2866 ordered_extent->start,
2867 ordered_extent->disk_len);
2869 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2870 ordered_extent->file_offset, ordered_extent->len,
2873 btrfs_abort_transaction(trans, root, ret);
2877 add_pending_csums(trans, inode, ordered_extent->file_offset,
2878 &ordered_extent->list);
2880 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2881 ret = btrfs_update_inode_fallback(trans, root, inode);
2882 if (ret) { /* -ENOMEM or corruption */
2883 btrfs_abort_transaction(trans, root, ret);
2888 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2889 ordered_extent->file_offset +
2890 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2892 if (root != root->fs_info->tree_root)
2893 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2895 btrfs_end_transaction(trans, root);
2897 if (ret || truncated) {
2901 start = ordered_extent->file_offset + logical_len;
2903 start = ordered_extent->file_offset;
2904 end = ordered_extent->file_offset + ordered_extent->len - 1;
2905 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2907 /* Drop the cache for the part of the extent we didn't write. */
2908 btrfs_drop_extent_cache(inode, start, end, 0);
2911 * If the ordered extent had an IOERR or something else went
2912 * wrong we need to return the space for this ordered extent
2913 * back to the allocator. We only free the extent in the
2914 * truncated case if we didn't write out the extent at all.
2916 if ((ret || !logical_len) &&
2917 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2918 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2919 btrfs_free_reserved_extent(root, ordered_extent->start,
2920 ordered_extent->disk_len, 1);
2925 * This needs to be done to make sure anybody waiting knows we are done
2926 * updating everything for this ordered extent.
2928 btrfs_remove_ordered_extent(inode, ordered_extent);
2930 /* for snapshot-aware defrag */
2933 free_sa_defrag_extent(new);
2934 atomic_dec(&root->fs_info->defrag_running);
2936 relink_file_extents(new);
2941 btrfs_put_ordered_extent(ordered_extent);
2942 /* once for the tree */
2943 btrfs_put_ordered_extent(ordered_extent);
2948 static void finish_ordered_fn(struct btrfs_work *work)
2950 struct btrfs_ordered_extent *ordered_extent;
2951 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2952 btrfs_finish_ordered_io(ordered_extent);
2955 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2956 struct extent_state *state, int uptodate)
2958 struct inode *inode = page->mapping->host;
2959 struct btrfs_root *root = BTRFS_I(inode)->root;
2960 struct btrfs_ordered_extent *ordered_extent = NULL;
2961 struct btrfs_workqueue *wq;
2962 btrfs_work_func_t func;
2964 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2966 ClearPagePrivate2(page);
2967 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2968 end - start + 1, uptodate))
2971 if (btrfs_is_free_space_inode(inode)) {
2972 wq = root->fs_info->endio_freespace_worker;
2973 func = btrfs_freespace_write_helper;
2975 wq = root->fs_info->endio_write_workers;
2976 func = btrfs_endio_write_helper;
2979 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
2981 btrfs_queue_work(wq, &ordered_extent->work);
2986 static int __readpage_endio_check(struct inode *inode,
2987 struct btrfs_io_bio *io_bio,
2988 int icsum, struct page *page,
2989 int pgoff, u64 start, size_t len)
2994 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2995 DEFAULT_RATELIMIT_BURST);
2997 csum_expected = *(((u32 *)io_bio->csum) + icsum);
2999 kaddr = kmap_atomic(page);
3000 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3001 btrfs_csum_final(csum, (char *)&csum);
3002 if (csum != csum_expected)
3005 kunmap_atomic(kaddr);
3008 if (__ratelimit(&_rs))
3009 btrfs_warn(BTRFS_I(inode)->root->fs_info,
3010 "csum failed ino %llu off %llu csum %u expected csum %u",
3011 btrfs_ino(inode), start, csum, csum_expected);
3012 memset(kaddr + pgoff, 1, len);
3013 flush_dcache_page(page);
3014 kunmap_atomic(kaddr);
3015 if (csum_expected == 0)
3021 * when reads are done, we need to check csums to verify the data is correct
3022 * if there's a match, we allow the bio to finish. If not, the code in
3023 * extent_io.c will try to find good copies for us.
3025 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3026 u64 phy_offset, struct page *page,
3027 u64 start, u64 end, int mirror)
3029 size_t offset = start - page_offset(page);
3030 struct inode *inode = page->mapping->host;
3031 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3032 struct btrfs_root *root = BTRFS_I(inode)->root;
3034 if (PageChecked(page)) {
3035 ClearPageChecked(page);
3039 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3042 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3043 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3044 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
3049 phy_offset >>= inode->i_sb->s_blocksize_bits;
3050 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3051 start, (size_t)(end - start + 1));
3054 struct delayed_iput {
3055 struct list_head list;
3056 struct inode *inode;
3059 /* JDM: If this is fs-wide, why can't we add a pointer to
3060 * btrfs_inode instead and avoid the allocation? */
3061 void btrfs_add_delayed_iput(struct inode *inode)
3063 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3064 struct delayed_iput *delayed;
3066 if (atomic_add_unless(&inode->i_count, -1, 1))
3069 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
3070 delayed->inode = inode;
3072 spin_lock(&fs_info->delayed_iput_lock);
3073 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
3074 spin_unlock(&fs_info->delayed_iput_lock);
3077 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3080 struct btrfs_fs_info *fs_info = root->fs_info;
3081 struct delayed_iput *delayed;
3084 spin_lock(&fs_info->delayed_iput_lock);
3085 empty = list_empty(&fs_info->delayed_iputs);
3086 spin_unlock(&fs_info->delayed_iput_lock);
3090 spin_lock(&fs_info->delayed_iput_lock);
3091 list_splice_init(&fs_info->delayed_iputs, &list);
3092 spin_unlock(&fs_info->delayed_iput_lock);
3094 while (!list_empty(&list)) {
3095 delayed = list_entry(list.next, struct delayed_iput, list);
3096 list_del(&delayed->list);
3097 iput(delayed->inode);
3103 * This is called in transaction commit time. If there are no orphan
3104 * files in the subvolume, it removes orphan item and frees block_rsv
3107 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3108 struct btrfs_root *root)
3110 struct btrfs_block_rsv *block_rsv;
3113 if (atomic_read(&root->orphan_inodes) ||
3114 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3117 spin_lock(&root->orphan_lock);
3118 if (atomic_read(&root->orphan_inodes)) {
3119 spin_unlock(&root->orphan_lock);
3123 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3124 spin_unlock(&root->orphan_lock);
3128 block_rsv = root->orphan_block_rsv;
3129 root->orphan_block_rsv = NULL;
3130 spin_unlock(&root->orphan_lock);
3132 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3133 btrfs_root_refs(&root->root_item) > 0) {
3134 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3135 root->root_key.objectid);
3137 btrfs_abort_transaction(trans, root, ret);
3139 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3144 WARN_ON(block_rsv->size > 0);
3145 btrfs_free_block_rsv(root, block_rsv);
3150 * This creates an orphan entry for the given inode in case something goes
3151 * wrong in the middle of an unlink/truncate.
3153 * NOTE: caller of this function should reserve 5 units of metadata for
3156 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3158 struct btrfs_root *root = BTRFS_I(inode)->root;
3159 struct btrfs_block_rsv *block_rsv = NULL;
3164 if (!root->orphan_block_rsv) {
3165 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3170 spin_lock(&root->orphan_lock);
3171 if (!root->orphan_block_rsv) {
3172 root->orphan_block_rsv = block_rsv;
3173 } else if (block_rsv) {
3174 btrfs_free_block_rsv(root, block_rsv);
3178 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3179 &BTRFS_I(inode)->runtime_flags)) {
3182 * For proper ENOSPC handling, we should do orphan
3183 * cleanup when mounting. But this introduces backward
3184 * compatibility issue.
3186 if (!xchg(&root->orphan_item_inserted, 1))
3192 atomic_inc(&root->orphan_inodes);
3195 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3196 &BTRFS_I(inode)->runtime_flags))
3198 spin_unlock(&root->orphan_lock);
3200 /* grab metadata reservation from transaction handle */
3202 ret = btrfs_orphan_reserve_metadata(trans, inode);
3203 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3206 /* insert an orphan item to track this unlinked/truncated file */
3208 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3210 atomic_dec(&root->orphan_inodes);
3212 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3213 &BTRFS_I(inode)->runtime_flags);
3214 btrfs_orphan_release_metadata(inode);
3216 if (ret != -EEXIST) {
3217 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3218 &BTRFS_I(inode)->runtime_flags);
3219 btrfs_abort_transaction(trans, root, ret);
3226 /* insert an orphan item to track subvolume contains orphan files */
3228 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3229 root->root_key.objectid);
3230 if (ret && ret != -EEXIST) {
3231 btrfs_abort_transaction(trans, root, ret);
3239 * We have done the truncate/delete so we can go ahead and remove the orphan
3240 * item for this particular inode.
3242 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3243 struct inode *inode)
3245 struct btrfs_root *root = BTRFS_I(inode)->root;
3246 int delete_item = 0;
3247 int release_rsv = 0;
3250 spin_lock(&root->orphan_lock);
3251 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3252 &BTRFS_I(inode)->runtime_flags))
3255 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3256 &BTRFS_I(inode)->runtime_flags))
3258 spin_unlock(&root->orphan_lock);
3261 atomic_dec(&root->orphan_inodes);
3263 ret = btrfs_del_orphan_item(trans, root,
3268 btrfs_orphan_release_metadata(inode);
3274 * this cleans up any orphans that may be left on the list from the last use
3277 int btrfs_orphan_cleanup(struct btrfs_root *root)
3279 struct btrfs_path *path;
3280 struct extent_buffer *leaf;
3281 struct btrfs_key key, found_key;
3282 struct btrfs_trans_handle *trans;
3283 struct inode *inode;
3284 u64 last_objectid = 0;
3285 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3287 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3290 path = btrfs_alloc_path();
3297 key.objectid = BTRFS_ORPHAN_OBJECTID;
3298 key.type = BTRFS_ORPHAN_ITEM_KEY;
3299 key.offset = (u64)-1;
3302 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3307 * if ret == 0 means we found what we were searching for, which
3308 * is weird, but possible, so only screw with path if we didn't
3309 * find the key and see if we have stuff that matches
3313 if (path->slots[0] == 0)
3318 /* pull out the item */
3319 leaf = path->nodes[0];
3320 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3322 /* make sure the item matches what we want */
3323 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3325 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3328 /* release the path since we're done with it */
3329 btrfs_release_path(path);
3332 * this is where we are basically btrfs_lookup, without the
3333 * crossing root thing. we store the inode number in the
3334 * offset of the orphan item.
3337 if (found_key.offset == last_objectid) {
3338 btrfs_err(root->fs_info,
3339 "Error removing orphan entry, stopping orphan cleanup");
3344 last_objectid = found_key.offset;
3346 found_key.objectid = found_key.offset;
3347 found_key.type = BTRFS_INODE_ITEM_KEY;
3348 found_key.offset = 0;
3349 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3350 ret = PTR_ERR_OR_ZERO(inode);
3351 if (ret && ret != -ESTALE)
3354 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3355 struct btrfs_root *dead_root;
3356 struct btrfs_fs_info *fs_info = root->fs_info;
3357 int is_dead_root = 0;
3360 * this is an orphan in the tree root. Currently these
3361 * could come from 2 sources:
3362 * a) a snapshot deletion in progress
3363 * b) a free space cache inode
3364 * We need to distinguish those two, as the snapshot
3365 * orphan must not get deleted.
3366 * find_dead_roots already ran before us, so if this
3367 * is a snapshot deletion, we should find the root
3368 * in the dead_roots list
3370 spin_lock(&fs_info->trans_lock);
3371 list_for_each_entry(dead_root, &fs_info->dead_roots,
3373 if (dead_root->root_key.objectid ==
3374 found_key.objectid) {
3379 spin_unlock(&fs_info->trans_lock);
3381 /* prevent this orphan from being found again */
3382 key.offset = found_key.objectid - 1;
3387 * Inode is already gone but the orphan item is still there,
3388 * kill the orphan item.
3390 if (ret == -ESTALE) {
3391 trans = btrfs_start_transaction(root, 1);
3392 if (IS_ERR(trans)) {
3393 ret = PTR_ERR(trans);
3396 btrfs_debug(root->fs_info, "auto deleting %Lu",
3397 found_key.objectid);
3398 ret = btrfs_del_orphan_item(trans, root,
3399 found_key.objectid);
3400 btrfs_end_transaction(trans, root);
3407 * add this inode to the orphan list so btrfs_orphan_del does
3408 * the proper thing when we hit it
3410 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3411 &BTRFS_I(inode)->runtime_flags);
3412 atomic_inc(&root->orphan_inodes);
3414 /* if we have links, this was a truncate, lets do that */
3415 if (inode->i_nlink) {
3416 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3422 /* 1 for the orphan item deletion. */
3423 trans = btrfs_start_transaction(root, 1);
3424 if (IS_ERR(trans)) {
3426 ret = PTR_ERR(trans);
3429 ret = btrfs_orphan_add(trans, inode);
3430 btrfs_end_transaction(trans, root);
3436 ret = btrfs_truncate(inode);
3438 btrfs_orphan_del(NULL, inode);
3443 /* this will do delete_inode and everything for us */
3448 /* release the path since we're done with it */
3449 btrfs_release_path(path);
3451 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3453 if (root->orphan_block_rsv)
3454 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3457 if (root->orphan_block_rsv ||
3458 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3459 trans = btrfs_join_transaction(root);
3461 btrfs_end_transaction(trans, root);
3465 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3467 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3471 btrfs_err(root->fs_info,
3472 "could not do orphan cleanup %d", ret);
3473 btrfs_free_path(path);
3478 * very simple check to peek ahead in the leaf looking for xattrs. If we
3479 * don't find any xattrs, we know there can't be any acls.
3481 * slot is the slot the inode is in, objectid is the objectid of the inode
3483 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3484 int slot, u64 objectid,
3485 int *first_xattr_slot)
3487 u32 nritems = btrfs_header_nritems(leaf);
3488 struct btrfs_key found_key;
3489 static u64 xattr_access = 0;
3490 static u64 xattr_default = 0;
3493 if (!xattr_access) {
3494 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3495 strlen(POSIX_ACL_XATTR_ACCESS));
3496 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3497 strlen(POSIX_ACL_XATTR_DEFAULT));
3501 *first_xattr_slot = -1;
3502 while (slot < nritems) {
3503 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3505 /* we found a different objectid, there must not be acls */
3506 if (found_key.objectid != objectid)
3509 /* we found an xattr, assume we've got an acl */
3510 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3511 if (*first_xattr_slot == -1)
3512 *first_xattr_slot = slot;
3513 if (found_key.offset == xattr_access ||
3514 found_key.offset == xattr_default)
3519 * we found a key greater than an xattr key, there can't
3520 * be any acls later on
3522 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3529 * it goes inode, inode backrefs, xattrs, extents,
3530 * so if there are a ton of hard links to an inode there can
3531 * be a lot of backrefs. Don't waste time searching too hard,
3532 * this is just an optimization
3537 /* we hit the end of the leaf before we found an xattr or
3538 * something larger than an xattr. We have to assume the inode
3541 if (*first_xattr_slot == -1)
3542 *first_xattr_slot = slot;
3547 * read an inode from the btree into the in-memory inode
3549 static void btrfs_read_locked_inode(struct inode *inode)
3551 struct btrfs_path *path;
3552 struct extent_buffer *leaf;
3553 struct btrfs_inode_item *inode_item;
3554 struct btrfs_root *root = BTRFS_I(inode)->root;
3555 struct btrfs_key location;
3560 bool filled = false;
3561 int first_xattr_slot;
3563 ret = btrfs_fill_inode(inode, &rdev);
3567 path = btrfs_alloc_path();
3571 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3573 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3577 leaf = path->nodes[0];
3582 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3583 struct btrfs_inode_item);
3584 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3585 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3586 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3587 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3588 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3590 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3591 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3593 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3594 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3596 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3597 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3599 BTRFS_I(inode)->i_otime.tv_sec =
3600 btrfs_timespec_sec(leaf, &inode_item->otime);
3601 BTRFS_I(inode)->i_otime.tv_nsec =
3602 btrfs_timespec_nsec(leaf, &inode_item->otime);
3604 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3605 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3606 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3609 * If we were modified in the current generation and evicted from memory
3610 * and then re-read we need to do a full sync since we don't have any
3611 * idea about which extents were modified before we were evicted from
3614 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3615 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3616 &BTRFS_I(inode)->runtime_flags);
3618 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3619 inode->i_generation = BTRFS_I(inode)->generation;
3621 rdev = btrfs_inode_rdev(leaf, inode_item);
3623 BTRFS_I(inode)->index_cnt = (u64)-1;
3624 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3628 if (inode->i_nlink != 1 ||
3629 path->slots[0] >= btrfs_header_nritems(leaf))
3632 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3633 if (location.objectid != btrfs_ino(inode))
3636 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3637 if (location.type == BTRFS_INODE_REF_KEY) {
3638 struct btrfs_inode_ref *ref;
3640 ref = (struct btrfs_inode_ref *)ptr;
3641 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3642 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3643 struct btrfs_inode_extref *extref;
3645 extref = (struct btrfs_inode_extref *)ptr;
3646 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3651 * try to precache a NULL acl entry for files that don't have
3652 * any xattrs or acls
3654 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3655 btrfs_ino(inode), &first_xattr_slot);
3656 if (first_xattr_slot != -1) {
3657 path->slots[0] = first_xattr_slot;
3658 ret = btrfs_load_inode_props(inode, path);
3660 btrfs_err(root->fs_info,
3661 "error loading props for ino %llu (root %llu): %d",
3663 root->root_key.objectid, ret);
3665 btrfs_free_path(path);
3668 cache_no_acl(inode);
3670 switch (inode->i_mode & S_IFMT) {
3672 inode->i_mapping->a_ops = &btrfs_aops;
3673 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3674 inode->i_fop = &btrfs_file_operations;
3675 inode->i_op = &btrfs_file_inode_operations;
3678 inode->i_fop = &btrfs_dir_file_operations;
3679 if (root == root->fs_info->tree_root)
3680 inode->i_op = &btrfs_dir_ro_inode_operations;
3682 inode->i_op = &btrfs_dir_inode_operations;
3685 inode->i_op = &btrfs_symlink_inode_operations;
3686 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3689 inode->i_op = &btrfs_special_inode_operations;
3690 init_special_inode(inode, inode->i_mode, rdev);
3694 btrfs_update_iflags(inode);
3698 btrfs_free_path(path);
3699 make_bad_inode(inode);
3703 * given a leaf and an inode, copy the inode fields into the leaf
3705 static void fill_inode_item(struct btrfs_trans_handle *trans,
3706 struct extent_buffer *leaf,
3707 struct btrfs_inode_item *item,
3708 struct inode *inode)
3710 struct btrfs_map_token token;
3712 btrfs_init_map_token(&token);
3714 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3715 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3716 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3718 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3719 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3721 btrfs_set_token_timespec_sec(leaf, &item->atime,
3722 inode->i_atime.tv_sec, &token);
3723 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3724 inode->i_atime.tv_nsec, &token);
3726 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3727 inode->i_mtime.tv_sec, &token);
3728 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3729 inode->i_mtime.tv_nsec, &token);
3731 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3732 inode->i_ctime.tv_sec, &token);
3733 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3734 inode->i_ctime.tv_nsec, &token);
3736 btrfs_set_token_timespec_sec(leaf, &item->otime,
3737 BTRFS_I(inode)->i_otime.tv_sec, &token);
3738 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3739 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3741 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3743 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3745 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3746 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3747 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3748 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3749 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3753 * copy everything in the in-memory inode into the btree.
3755 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3756 struct btrfs_root *root, struct inode *inode)
3758 struct btrfs_inode_item *inode_item;
3759 struct btrfs_path *path;
3760 struct extent_buffer *leaf;
3763 path = btrfs_alloc_path();
3767 path->leave_spinning = 1;
3768 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3776 leaf = path->nodes[0];
3777 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3778 struct btrfs_inode_item);
3780 fill_inode_item(trans, leaf, inode_item, inode);
3781 btrfs_mark_buffer_dirty(leaf);
3782 btrfs_set_inode_last_trans(trans, inode);
3785 btrfs_free_path(path);
3790 * copy everything in the in-memory inode into the btree.
3792 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3793 struct btrfs_root *root, struct inode *inode)
3798 * If the inode is a free space inode, we can deadlock during commit
3799 * if we put it into the delayed code.
3801 * The data relocation inode should also be directly updated
3804 if (!btrfs_is_free_space_inode(inode)
3805 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3806 && !root->fs_info->log_root_recovering) {
3807 btrfs_update_root_times(trans, root);
3809 ret = btrfs_delayed_update_inode(trans, root, inode);
3811 btrfs_set_inode_last_trans(trans, inode);
3815 return btrfs_update_inode_item(trans, root, inode);
3818 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3819 struct btrfs_root *root,
3820 struct inode *inode)
3824 ret = btrfs_update_inode(trans, root, inode);
3826 return btrfs_update_inode_item(trans, root, inode);
3831 * unlink helper that gets used here in inode.c and in the tree logging
3832 * recovery code. It remove a link in a directory with a given name, and
3833 * also drops the back refs in the inode to the directory
3835 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3836 struct btrfs_root *root,
3837 struct inode *dir, struct inode *inode,
3838 const char *name, int name_len)
3840 struct btrfs_path *path;
3842 struct extent_buffer *leaf;
3843 struct btrfs_dir_item *di;
3844 struct btrfs_key key;
3846 u64 ino = btrfs_ino(inode);
3847 u64 dir_ino = btrfs_ino(dir);
3849 path = btrfs_alloc_path();
3855 path->leave_spinning = 1;
3856 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3857 name, name_len, -1);
3866 leaf = path->nodes[0];
3867 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3868 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3871 btrfs_release_path(path);
3874 * If we don't have dir index, we have to get it by looking up
3875 * the inode ref, since we get the inode ref, remove it directly,
3876 * it is unnecessary to do delayed deletion.
3878 * But if we have dir index, needn't search inode ref to get it.
3879 * Since the inode ref is close to the inode item, it is better
3880 * that we delay to delete it, and just do this deletion when
3881 * we update the inode item.
3883 if (BTRFS_I(inode)->dir_index) {
3884 ret = btrfs_delayed_delete_inode_ref(inode);
3886 index = BTRFS_I(inode)->dir_index;
3891 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3894 btrfs_info(root->fs_info,
3895 "failed to delete reference to %.*s, inode %llu parent %llu",
3896 name_len, name, ino, dir_ino);
3897 btrfs_abort_transaction(trans, root, ret);
3901 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3903 btrfs_abort_transaction(trans, root, ret);
3907 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3909 if (ret != 0 && ret != -ENOENT) {
3910 btrfs_abort_transaction(trans, root, ret);
3914 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3919 btrfs_abort_transaction(trans, root, ret);
3921 btrfs_free_path(path);
3925 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3926 inode_inc_iversion(inode);
3927 inode_inc_iversion(dir);
3928 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3929 ret = btrfs_update_inode(trans, root, dir);
3934 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3935 struct btrfs_root *root,
3936 struct inode *dir, struct inode *inode,
3937 const char *name, int name_len)
3940 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3943 ret = btrfs_update_inode(trans, root, inode);
3949 * helper to start transaction for unlink and rmdir.
3951 * unlink and rmdir are special in btrfs, they do not always free space, so
3952 * if we cannot make our reservations the normal way try and see if there is
3953 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3954 * allow the unlink to occur.
3956 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3958 struct btrfs_trans_handle *trans;
3959 struct btrfs_root *root = BTRFS_I(dir)->root;
3963 * 1 for the possible orphan item
3964 * 1 for the dir item
3965 * 1 for the dir index
3966 * 1 for the inode ref
3969 trans = btrfs_start_transaction(root, 5);
3970 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3973 if (PTR_ERR(trans) == -ENOSPC) {
3974 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3976 trans = btrfs_start_transaction(root, 0);
3979 ret = btrfs_cond_migrate_bytes(root->fs_info,
3980 &root->fs_info->trans_block_rsv,
3983 btrfs_end_transaction(trans, root);
3984 return ERR_PTR(ret);
3986 trans->block_rsv = &root->fs_info->trans_block_rsv;
3987 trans->bytes_reserved = num_bytes;
3992 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3994 struct btrfs_root *root = BTRFS_I(dir)->root;
3995 struct btrfs_trans_handle *trans;
3996 struct inode *inode = dentry->d_inode;
3999 trans = __unlink_start_trans(dir);
4001 return PTR_ERR(trans);
4003 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
4005 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
4006 dentry->d_name.name, dentry->d_name.len);
4010 if (inode->i_nlink == 0) {
4011 ret = btrfs_orphan_add(trans, inode);
4017 btrfs_end_transaction(trans, root);
4018 btrfs_btree_balance_dirty(root);
4022 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4023 struct btrfs_root *root,
4024 struct inode *dir, u64 objectid,
4025 const char *name, int name_len)
4027 struct btrfs_path *path;
4028 struct extent_buffer *leaf;
4029 struct btrfs_dir_item *di;
4030 struct btrfs_key key;
4033 u64 dir_ino = btrfs_ino(dir);
4035 path = btrfs_alloc_path();
4039 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4040 name, name_len, -1);
4041 if (IS_ERR_OR_NULL(di)) {
4049 leaf = path->nodes[0];
4050 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4051 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4052 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4054 btrfs_abort_transaction(trans, root, ret);
4057 btrfs_release_path(path);
4059 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4060 objectid, root->root_key.objectid,
4061 dir_ino, &index, name, name_len);
4063 if (ret != -ENOENT) {
4064 btrfs_abort_transaction(trans, root, ret);
4067 di = btrfs_search_dir_index_item(root, path, dir_ino,
4069 if (IS_ERR_OR_NULL(di)) {
4074 btrfs_abort_transaction(trans, root, ret);
4078 leaf = path->nodes[0];
4079 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4080 btrfs_release_path(path);
4083 btrfs_release_path(path);
4085 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4087 btrfs_abort_transaction(trans, root, ret);
4091 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4092 inode_inc_iversion(dir);
4093 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4094 ret = btrfs_update_inode_fallback(trans, root, dir);
4096 btrfs_abort_transaction(trans, root, ret);
4098 btrfs_free_path(path);
4102 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4104 struct inode *inode = dentry->d_inode;
4106 struct btrfs_root *root = BTRFS_I(dir)->root;
4107 struct btrfs_trans_handle *trans;
4109 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4111 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4114 trans = __unlink_start_trans(dir);
4116 return PTR_ERR(trans);
4118 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4119 err = btrfs_unlink_subvol(trans, root, dir,
4120 BTRFS_I(inode)->location.objectid,
4121 dentry->d_name.name,
4122 dentry->d_name.len);
4126 err = btrfs_orphan_add(trans, inode);
4130 /* now the directory is empty */
4131 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
4132 dentry->d_name.name, dentry->d_name.len);
4134 btrfs_i_size_write(inode, 0);
4136 btrfs_end_transaction(trans, root);
4137 btrfs_btree_balance_dirty(root);
4143 * this can truncate away extent items, csum items and directory items.
4144 * It starts at a high offset and removes keys until it can't find
4145 * any higher than new_size
4147 * csum items that cross the new i_size are truncated to the new size
4150 * min_type is the minimum key type to truncate down to. If set to 0, this
4151 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4153 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4154 struct btrfs_root *root,
4155 struct inode *inode,
4156 u64 new_size, u32 min_type)
4158 struct btrfs_path *path;
4159 struct extent_buffer *leaf;
4160 struct btrfs_file_extent_item *fi;
4161 struct btrfs_key key;
4162 struct btrfs_key found_key;
4163 u64 extent_start = 0;
4164 u64 extent_num_bytes = 0;
4165 u64 extent_offset = 0;
4167 u64 last_size = (u64)-1;
4168 u32 found_type = (u8)-1;
4171 int pending_del_nr = 0;
4172 int pending_del_slot = 0;
4173 int extent_type = -1;
4176 u64 ino = btrfs_ino(inode);
4178 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4180 path = btrfs_alloc_path();
4186 * We want to drop from the next block forward in case this new size is
4187 * not block aligned since we will be keeping the last block of the
4188 * extent just the way it is.
4190 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4191 root == root->fs_info->tree_root)
4192 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4193 root->sectorsize), (u64)-1, 0);
4196 * This function is also used to drop the items in the log tree before
4197 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4198 * it is used to drop the loged items. So we shouldn't kill the delayed
4201 if (min_type == 0 && root == BTRFS_I(inode)->root)
4202 btrfs_kill_delayed_inode_items(inode);
4205 key.offset = (u64)-1;
4209 path->leave_spinning = 1;
4210 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4217 /* there are no items in the tree for us to truncate, we're
4220 if (path->slots[0] == 0)
4227 leaf = path->nodes[0];
4228 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4229 found_type = found_key.type;
4231 if (found_key.objectid != ino)
4234 if (found_type < min_type)
4237 item_end = found_key.offset;
4238 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4239 fi = btrfs_item_ptr(leaf, path->slots[0],
4240 struct btrfs_file_extent_item);
4241 extent_type = btrfs_file_extent_type(leaf, fi);
4242 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4244 btrfs_file_extent_num_bytes(leaf, fi);
4245 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4246 item_end += btrfs_file_extent_inline_len(leaf,
4247 path->slots[0], fi);
4251 if (found_type > min_type) {
4254 if (item_end < new_size)
4256 if (found_key.offset >= new_size)
4262 /* FIXME, shrink the extent if the ref count is only 1 */
4263 if (found_type != BTRFS_EXTENT_DATA_KEY)
4267 last_size = found_key.offset;
4269 last_size = new_size;
4271 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4273 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4275 u64 orig_num_bytes =
4276 btrfs_file_extent_num_bytes(leaf, fi);
4277 extent_num_bytes = ALIGN(new_size -
4280 btrfs_set_file_extent_num_bytes(leaf, fi,
4282 num_dec = (orig_num_bytes -
4284 if (test_bit(BTRFS_ROOT_REF_COWS,
4287 inode_sub_bytes(inode, num_dec);
4288 btrfs_mark_buffer_dirty(leaf);
4291 btrfs_file_extent_disk_num_bytes(leaf,
4293 extent_offset = found_key.offset -
4294 btrfs_file_extent_offset(leaf, fi);
4296 /* FIXME blocksize != 4096 */
4297 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4298 if (extent_start != 0) {
4300 if (test_bit(BTRFS_ROOT_REF_COWS,
4302 inode_sub_bytes(inode, num_dec);
4305 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4307 * we can't truncate inline items that have had
4311 btrfs_file_extent_compression(leaf, fi) == 0 &&
4312 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4313 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4314 u32 size = new_size - found_key.offset;
4316 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4317 inode_sub_bytes(inode, item_end + 1 -
4321 * update the ram bytes to properly reflect
4322 * the new size of our item
4324 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4326 btrfs_file_extent_calc_inline_size(size);
4327 btrfs_truncate_item(root, path, size, 1);
4328 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4330 inode_sub_bytes(inode, item_end + 1 -
4336 if (!pending_del_nr) {
4337 /* no pending yet, add ourselves */
4338 pending_del_slot = path->slots[0];
4340 } else if (pending_del_nr &&
4341 path->slots[0] + 1 == pending_del_slot) {
4342 /* hop on the pending chunk */
4344 pending_del_slot = path->slots[0];
4352 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4353 root == root->fs_info->tree_root)) {
4354 btrfs_set_path_blocking(path);
4355 ret = btrfs_free_extent(trans, root, extent_start,
4356 extent_num_bytes, 0,
4357 btrfs_header_owner(leaf),
4358 ino, extent_offset, 0);
4362 if (found_type == BTRFS_INODE_ITEM_KEY)
4365 if (path->slots[0] == 0 ||
4366 path->slots[0] != pending_del_slot) {
4367 if (pending_del_nr) {
4368 ret = btrfs_del_items(trans, root, path,
4372 btrfs_abort_transaction(trans,
4378 btrfs_release_path(path);
4385 if (pending_del_nr) {
4386 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4389 btrfs_abort_transaction(trans, root, ret);
4392 if (last_size != (u64)-1 &&
4393 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4394 btrfs_ordered_update_i_size(inode, last_size, NULL);
4395 btrfs_free_path(path);
4400 * btrfs_truncate_page - read, zero a chunk and write a page
4401 * @inode - inode that we're zeroing
4402 * @from - the offset to start zeroing
4403 * @len - the length to zero, 0 to zero the entire range respective to the
4405 * @front - zero up to the offset instead of from the offset on
4407 * This will find the page for the "from" offset and cow the page and zero the
4408 * part we want to zero. This is used with truncate and hole punching.
4410 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4413 struct address_space *mapping = inode->i_mapping;
4414 struct btrfs_root *root = BTRFS_I(inode)->root;
4415 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4416 struct btrfs_ordered_extent *ordered;
4417 struct extent_state *cached_state = NULL;
4419 u32 blocksize = root->sectorsize;
4420 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4421 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4423 gfp_t mask = btrfs_alloc_write_mask(mapping);
4428 if ((offset & (blocksize - 1)) == 0 &&
4429 (!len || ((len & (blocksize - 1)) == 0)))
4431 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4436 page = find_or_create_page(mapping, index, mask);
4438 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4443 page_start = page_offset(page);
4444 page_end = page_start + PAGE_CACHE_SIZE - 1;
4446 if (!PageUptodate(page)) {
4447 ret = btrfs_readpage(NULL, page);
4449 if (page->mapping != mapping) {
4451 page_cache_release(page);
4454 if (!PageUptodate(page)) {
4459 wait_on_page_writeback(page);
4461 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4462 set_page_extent_mapped(page);
4464 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4466 unlock_extent_cached(io_tree, page_start, page_end,
4467 &cached_state, GFP_NOFS);
4469 page_cache_release(page);
4470 btrfs_start_ordered_extent(inode, ordered, 1);
4471 btrfs_put_ordered_extent(ordered);
4475 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4476 EXTENT_DIRTY | EXTENT_DELALLOC |
4477 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4478 0, 0, &cached_state, GFP_NOFS);
4480 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4483 unlock_extent_cached(io_tree, page_start, page_end,
4484 &cached_state, GFP_NOFS);
4488 if (offset != PAGE_CACHE_SIZE) {
4490 len = PAGE_CACHE_SIZE - offset;
4493 memset(kaddr, 0, offset);
4495 memset(kaddr + offset, 0, len);
4496 flush_dcache_page(page);
4499 ClearPageChecked(page);
4500 set_page_dirty(page);
4501 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4506 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4508 page_cache_release(page);
4513 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4514 u64 offset, u64 len)
4516 struct btrfs_trans_handle *trans;
4520 * Still need to make sure the inode looks like it's been updated so
4521 * that any holes get logged if we fsync.
4523 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4524 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4525 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4526 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4531 * 1 - for the one we're dropping
4532 * 1 - for the one we're adding
4533 * 1 - for updating the inode.
4535 trans = btrfs_start_transaction(root, 3);
4537 return PTR_ERR(trans);
4539 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4541 btrfs_abort_transaction(trans, root, ret);
4542 btrfs_end_transaction(trans, root);
4546 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4547 0, 0, len, 0, len, 0, 0, 0);
4549 btrfs_abort_transaction(trans, root, ret);
4551 btrfs_update_inode(trans, root, inode);
4552 btrfs_end_transaction(trans, root);
4557 * This function puts in dummy file extents for the area we're creating a hole
4558 * for. So if we are truncating this file to a larger size we need to insert
4559 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4560 * the range between oldsize and size
4562 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4564 struct btrfs_root *root = BTRFS_I(inode)->root;
4565 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4566 struct extent_map *em = NULL;
4567 struct extent_state *cached_state = NULL;
4568 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4569 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4570 u64 block_end = ALIGN(size, root->sectorsize);
4577 * If our size started in the middle of a page we need to zero out the
4578 * rest of the page before we expand the i_size, otherwise we could
4579 * expose stale data.
4581 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4585 if (size <= hole_start)
4589 struct btrfs_ordered_extent *ordered;
4591 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4593 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4594 block_end - hole_start);
4597 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4598 &cached_state, GFP_NOFS);
4599 btrfs_start_ordered_extent(inode, ordered, 1);
4600 btrfs_put_ordered_extent(ordered);
4603 cur_offset = hole_start;
4605 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4606 block_end - cur_offset, 0);
4612 last_byte = min(extent_map_end(em), block_end);
4613 last_byte = ALIGN(last_byte , root->sectorsize);
4614 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4615 struct extent_map *hole_em;
4616 hole_size = last_byte - cur_offset;
4618 err = maybe_insert_hole(root, inode, cur_offset,
4622 btrfs_drop_extent_cache(inode, cur_offset,
4623 cur_offset + hole_size - 1, 0);
4624 hole_em = alloc_extent_map();
4626 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4627 &BTRFS_I(inode)->runtime_flags);
4630 hole_em->start = cur_offset;
4631 hole_em->len = hole_size;
4632 hole_em->orig_start = cur_offset;
4634 hole_em->block_start = EXTENT_MAP_HOLE;
4635 hole_em->block_len = 0;
4636 hole_em->orig_block_len = 0;
4637 hole_em->ram_bytes = hole_size;
4638 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4639 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4640 hole_em->generation = root->fs_info->generation;
4643 write_lock(&em_tree->lock);
4644 err = add_extent_mapping(em_tree, hole_em, 1);
4645 write_unlock(&em_tree->lock);
4648 btrfs_drop_extent_cache(inode, cur_offset,
4652 free_extent_map(hole_em);
4655 free_extent_map(em);
4657 cur_offset = last_byte;
4658 if (cur_offset >= block_end)
4661 free_extent_map(em);
4662 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4667 static int wait_snapshoting_atomic_t(atomic_t *a)
4673 static void wait_for_snapshot_creation(struct btrfs_root *root)
4678 ret = btrfs_start_write_no_snapshoting(root);
4681 wait_on_atomic_t(&root->will_be_snapshoted,
4682 wait_snapshoting_atomic_t,
4683 TASK_UNINTERRUPTIBLE);
4687 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4689 struct btrfs_root *root = BTRFS_I(inode)->root;
4690 struct btrfs_trans_handle *trans;
4691 loff_t oldsize = i_size_read(inode);
4692 loff_t newsize = attr->ia_size;
4693 int mask = attr->ia_valid;
4697 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4698 * special case where we need to update the times despite not having
4699 * these flags set. For all other operations the VFS set these flags
4700 * explicitly if it wants a timestamp update.
4702 if (newsize != oldsize) {
4703 inode_inc_iversion(inode);
4704 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4705 inode->i_ctime = inode->i_mtime =
4706 current_fs_time(inode->i_sb);
4709 if (newsize > oldsize) {
4710 truncate_pagecache(inode, newsize);
4712 * Don't do an expanding truncate while snapshoting is ongoing.
4713 * This is to ensure the snapshot captures a fully consistent
4714 * state of this file - if the snapshot captures this expanding
4715 * truncation, it must capture all writes that happened before
4718 wait_for_snapshot_creation(root);
4719 ret = btrfs_cont_expand(inode, oldsize, newsize);
4721 btrfs_end_write_no_snapshoting(root);
4725 trans = btrfs_start_transaction(root, 1);
4726 if (IS_ERR(trans)) {
4727 btrfs_end_write_no_snapshoting(root);
4728 return PTR_ERR(trans);
4731 i_size_write(inode, newsize);
4732 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4733 ret = btrfs_update_inode(trans, root, inode);
4734 btrfs_end_write_no_snapshoting(root);
4735 btrfs_end_transaction(trans, root);
4739 * We're truncating a file that used to have good data down to
4740 * zero. Make sure it gets into the ordered flush list so that
4741 * any new writes get down to disk quickly.
4744 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4745 &BTRFS_I(inode)->runtime_flags);
4748 * 1 for the orphan item we're going to add
4749 * 1 for the orphan item deletion.
4751 trans = btrfs_start_transaction(root, 2);
4753 return PTR_ERR(trans);
4756 * We need to do this in case we fail at _any_ point during the
4757 * actual truncate. Once we do the truncate_setsize we could
4758 * invalidate pages which forces any outstanding ordered io to
4759 * be instantly completed which will give us extents that need
4760 * to be truncated. If we fail to get an orphan inode down we
4761 * could have left over extents that were never meant to live,
4762 * so we need to garuntee from this point on that everything
4763 * will be consistent.
4765 ret = btrfs_orphan_add(trans, inode);
4766 btrfs_end_transaction(trans, root);
4770 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4771 truncate_setsize(inode, newsize);
4773 /* Disable nonlocked read DIO to avoid the end less truncate */
4774 btrfs_inode_block_unlocked_dio(inode);
4775 inode_dio_wait(inode);
4776 btrfs_inode_resume_unlocked_dio(inode);
4778 ret = btrfs_truncate(inode);
4779 if (ret && inode->i_nlink) {
4783 * failed to truncate, disk_i_size is only adjusted down
4784 * as we remove extents, so it should represent the true
4785 * size of the inode, so reset the in memory size and
4786 * delete our orphan entry.
4788 trans = btrfs_join_transaction(root);
4789 if (IS_ERR(trans)) {
4790 btrfs_orphan_del(NULL, inode);
4793 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4794 err = btrfs_orphan_del(trans, inode);
4796 btrfs_abort_transaction(trans, root, err);
4797 btrfs_end_transaction(trans, root);
4804 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4806 struct inode *inode = dentry->d_inode;
4807 struct btrfs_root *root = BTRFS_I(inode)->root;
4810 if (btrfs_root_readonly(root))
4813 err = inode_change_ok(inode, attr);
4817 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4818 err = btrfs_setsize(inode, attr);
4823 if (attr->ia_valid) {
4824 setattr_copy(inode, attr);
4825 inode_inc_iversion(inode);
4826 err = btrfs_dirty_inode(inode);
4828 if (!err && attr->ia_valid & ATTR_MODE)
4829 err = posix_acl_chmod(inode, inode->i_mode);
4836 * While truncating the inode pages during eviction, we get the VFS calling
4837 * btrfs_invalidatepage() against each page of the inode. This is slow because
4838 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4839 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4840 * extent_state structures over and over, wasting lots of time.
4842 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4843 * those expensive operations on a per page basis and do only the ordered io
4844 * finishing, while we release here the extent_map and extent_state structures,
4845 * without the excessive merging and splitting.
4847 static void evict_inode_truncate_pages(struct inode *inode)
4849 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4850 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4851 struct rb_node *node;
4853 ASSERT(inode->i_state & I_FREEING);
4854 truncate_inode_pages_final(&inode->i_data);
4856 write_lock(&map_tree->lock);
4857 while (!RB_EMPTY_ROOT(&map_tree->map)) {
4858 struct extent_map *em;
4860 node = rb_first(&map_tree->map);
4861 em = rb_entry(node, struct extent_map, rb_node);
4862 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4863 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4864 remove_extent_mapping(map_tree, em);
4865 free_extent_map(em);
4866 if (need_resched()) {
4867 write_unlock(&map_tree->lock);
4869 write_lock(&map_tree->lock);
4872 write_unlock(&map_tree->lock);
4874 spin_lock(&io_tree->lock);
4875 while (!RB_EMPTY_ROOT(&io_tree->state)) {
4876 struct extent_state *state;
4877 struct extent_state *cached_state = NULL;
4879 node = rb_first(&io_tree->state);
4880 state = rb_entry(node, struct extent_state, rb_node);
4881 atomic_inc(&state->refs);
4882 spin_unlock(&io_tree->lock);
4884 lock_extent_bits(io_tree, state->start, state->end,
4886 clear_extent_bit(io_tree, state->start, state->end,
4887 EXTENT_LOCKED | EXTENT_DIRTY |
4888 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
4889 EXTENT_DEFRAG, 1, 1,
4890 &cached_state, GFP_NOFS);
4891 free_extent_state(state);
4894 spin_lock(&io_tree->lock);
4896 spin_unlock(&io_tree->lock);
4899 void btrfs_evict_inode(struct inode *inode)
4901 struct btrfs_trans_handle *trans;
4902 struct btrfs_root *root = BTRFS_I(inode)->root;
4903 struct btrfs_block_rsv *rsv, *global_rsv;
4904 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4907 trace_btrfs_inode_evict(inode);
4909 evict_inode_truncate_pages(inode);
4911 if (inode->i_nlink &&
4912 ((btrfs_root_refs(&root->root_item) != 0 &&
4913 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4914 btrfs_is_free_space_inode(inode)))
4917 if (is_bad_inode(inode)) {
4918 btrfs_orphan_del(NULL, inode);
4921 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4922 btrfs_wait_ordered_range(inode, 0, (u64)-1);
4924 btrfs_free_io_failure_record(inode, 0, (u64)-1);
4926 if (root->fs_info->log_root_recovering) {
4927 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4928 &BTRFS_I(inode)->runtime_flags));
4932 if (inode->i_nlink > 0) {
4933 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4934 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
4938 ret = btrfs_commit_inode_delayed_inode(inode);
4940 btrfs_orphan_del(NULL, inode);
4944 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4946 btrfs_orphan_del(NULL, inode);
4949 rsv->size = min_size;
4951 global_rsv = &root->fs_info->global_block_rsv;
4953 btrfs_i_size_write(inode, 0);
4956 * This is a bit simpler than btrfs_truncate since we've already
4957 * reserved our space for our orphan item in the unlink, so we just
4958 * need to reserve some slack space in case we add bytes and update
4959 * inode item when doing the truncate.
4962 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4963 BTRFS_RESERVE_FLUSH_LIMIT);
4966 * Try and steal from the global reserve since we will
4967 * likely not use this space anyway, we want to try as
4968 * hard as possible to get this to work.
4971 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4974 btrfs_warn(root->fs_info,
4975 "Could not get space for a delete, will truncate on mount %d",
4977 btrfs_orphan_del(NULL, inode);
4978 btrfs_free_block_rsv(root, rsv);
4982 trans = btrfs_join_transaction(root);
4983 if (IS_ERR(trans)) {
4984 btrfs_orphan_del(NULL, inode);
4985 btrfs_free_block_rsv(root, rsv);
4989 trans->block_rsv = rsv;
4991 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4995 trans->block_rsv = &root->fs_info->trans_block_rsv;
4996 btrfs_end_transaction(trans, root);
4998 btrfs_btree_balance_dirty(root);
5001 btrfs_free_block_rsv(root, rsv);
5004 * Errors here aren't a big deal, it just means we leave orphan items
5005 * in the tree. They will be cleaned up on the next mount.
5008 trans->block_rsv = root->orphan_block_rsv;
5009 btrfs_orphan_del(trans, inode);
5011 btrfs_orphan_del(NULL, inode);
5014 trans->block_rsv = &root->fs_info->trans_block_rsv;
5015 if (!(root == root->fs_info->tree_root ||
5016 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5017 btrfs_return_ino(root, btrfs_ino(inode));
5019 btrfs_end_transaction(trans, root);
5020 btrfs_btree_balance_dirty(root);
5022 btrfs_remove_delayed_node(inode);
5028 * this returns the key found in the dir entry in the location pointer.
5029 * If no dir entries were found, location->objectid is 0.
5031 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5032 struct btrfs_key *location)
5034 const char *name = dentry->d_name.name;
5035 int namelen = dentry->d_name.len;
5036 struct btrfs_dir_item *di;
5037 struct btrfs_path *path;
5038 struct btrfs_root *root = BTRFS_I(dir)->root;
5041 path = btrfs_alloc_path();
5045 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
5050 if (IS_ERR_OR_NULL(di))
5053 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5055 btrfs_free_path(path);
5058 location->objectid = 0;
5063 * when we hit a tree root in a directory, the btrfs part of the inode
5064 * needs to be changed to reflect the root directory of the tree root. This
5065 * is kind of like crossing a mount point.
5067 static int fixup_tree_root_location(struct btrfs_root *root,
5069 struct dentry *dentry,
5070 struct btrfs_key *location,
5071 struct btrfs_root **sub_root)
5073 struct btrfs_path *path;
5074 struct btrfs_root *new_root;
5075 struct btrfs_root_ref *ref;
5076 struct extent_buffer *leaf;
5077 struct btrfs_key key;
5081 path = btrfs_alloc_path();
5088 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5089 key.type = BTRFS_ROOT_REF_KEY;
5090 key.offset = location->objectid;
5092 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5100 leaf = path->nodes[0];
5101 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5102 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5103 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5106 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5107 (unsigned long)(ref + 1),
5108 dentry->d_name.len);
5112 btrfs_release_path(path);
5114 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5115 if (IS_ERR(new_root)) {
5116 err = PTR_ERR(new_root);
5120 *sub_root = new_root;
5121 location->objectid = btrfs_root_dirid(&new_root->root_item);
5122 location->type = BTRFS_INODE_ITEM_KEY;
5123 location->offset = 0;
5126 btrfs_free_path(path);
5130 static void inode_tree_add(struct inode *inode)
5132 struct btrfs_root *root = BTRFS_I(inode)->root;
5133 struct btrfs_inode *entry;
5135 struct rb_node *parent;
5136 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5137 u64 ino = btrfs_ino(inode);
5139 if (inode_unhashed(inode))
5142 spin_lock(&root->inode_lock);
5143 p = &root->inode_tree.rb_node;
5146 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5148 if (ino < btrfs_ino(&entry->vfs_inode))
5149 p = &parent->rb_left;
5150 else if (ino > btrfs_ino(&entry->vfs_inode))
5151 p = &parent->rb_right;
5153 WARN_ON(!(entry->vfs_inode.i_state &
5154 (I_WILL_FREE | I_FREEING)));
5155 rb_replace_node(parent, new, &root->inode_tree);
5156 RB_CLEAR_NODE(parent);
5157 spin_unlock(&root->inode_lock);
5161 rb_link_node(new, parent, p);
5162 rb_insert_color(new, &root->inode_tree);
5163 spin_unlock(&root->inode_lock);
5166 static void inode_tree_del(struct inode *inode)
5168 struct btrfs_root *root = BTRFS_I(inode)->root;
5171 spin_lock(&root->inode_lock);
5172 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5173 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5174 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5175 empty = RB_EMPTY_ROOT(&root->inode_tree);
5177 spin_unlock(&root->inode_lock);
5179 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5180 synchronize_srcu(&root->fs_info->subvol_srcu);
5181 spin_lock(&root->inode_lock);
5182 empty = RB_EMPTY_ROOT(&root->inode_tree);
5183 spin_unlock(&root->inode_lock);
5185 btrfs_add_dead_root(root);
5189 void btrfs_invalidate_inodes(struct btrfs_root *root)
5191 struct rb_node *node;
5192 struct rb_node *prev;
5193 struct btrfs_inode *entry;
5194 struct inode *inode;
5197 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5198 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5200 spin_lock(&root->inode_lock);
5202 node = root->inode_tree.rb_node;
5206 entry = rb_entry(node, struct btrfs_inode, rb_node);
5208 if (objectid < btrfs_ino(&entry->vfs_inode))
5209 node = node->rb_left;
5210 else if (objectid > btrfs_ino(&entry->vfs_inode))
5211 node = node->rb_right;
5217 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5218 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5222 prev = rb_next(prev);
5226 entry = rb_entry(node, struct btrfs_inode, rb_node);
5227 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5228 inode = igrab(&entry->vfs_inode);
5230 spin_unlock(&root->inode_lock);
5231 if (atomic_read(&inode->i_count) > 1)
5232 d_prune_aliases(inode);
5234 * btrfs_drop_inode will have it removed from
5235 * the inode cache when its usage count
5240 spin_lock(&root->inode_lock);
5244 if (cond_resched_lock(&root->inode_lock))
5247 node = rb_next(node);
5249 spin_unlock(&root->inode_lock);
5252 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5254 struct btrfs_iget_args *args = p;
5255 inode->i_ino = args->location->objectid;
5256 memcpy(&BTRFS_I(inode)->location, args->location,
5257 sizeof(*args->location));
5258 BTRFS_I(inode)->root = args->root;
5262 static int btrfs_find_actor(struct inode *inode, void *opaque)
5264 struct btrfs_iget_args *args = opaque;
5265 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5266 args->root == BTRFS_I(inode)->root;
5269 static struct inode *btrfs_iget_locked(struct super_block *s,
5270 struct btrfs_key *location,
5271 struct btrfs_root *root)
5273 struct inode *inode;
5274 struct btrfs_iget_args args;
5275 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5277 args.location = location;
5280 inode = iget5_locked(s, hashval, btrfs_find_actor,
5281 btrfs_init_locked_inode,
5286 /* Get an inode object given its location and corresponding root.
5287 * Returns in *is_new if the inode was read from disk
5289 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5290 struct btrfs_root *root, int *new)
5292 struct inode *inode;
5294 inode = btrfs_iget_locked(s, location, root);
5296 return ERR_PTR(-ENOMEM);
5298 if (inode->i_state & I_NEW) {
5299 btrfs_read_locked_inode(inode);
5300 if (!is_bad_inode(inode)) {
5301 inode_tree_add(inode);
5302 unlock_new_inode(inode);
5306 unlock_new_inode(inode);
5308 inode = ERR_PTR(-ESTALE);
5315 static struct inode *new_simple_dir(struct super_block *s,
5316 struct btrfs_key *key,
5317 struct btrfs_root *root)
5319 struct inode *inode = new_inode(s);
5322 return ERR_PTR(-ENOMEM);
5324 BTRFS_I(inode)->root = root;
5325 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5326 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5328 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5329 inode->i_op = &btrfs_dir_ro_inode_operations;
5330 inode->i_fop = &simple_dir_operations;
5331 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5332 inode->i_mtime = CURRENT_TIME;
5333 inode->i_atime = inode->i_mtime;
5334 inode->i_ctime = inode->i_mtime;
5335 BTRFS_I(inode)->i_otime = inode->i_mtime;
5340 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5342 struct inode *inode;
5343 struct btrfs_root *root = BTRFS_I(dir)->root;
5344 struct btrfs_root *sub_root = root;
5345 struct btrfs_key location;
5349 if (dentry->d_name.len > BTRFS_NAME_LEN)
5350 return ERR_PTR(-ENAMETOOLONG);
5352 ret = btrfs_inode_by_name(dir, dentry, &location);
5354 return ERR_PTR(ret);
5356 if (location.objectid == 0)
5357 return ERR_PTR(-ENOENT);
5359 if (location.type == BTRFS_INODE_ITEM_KEY) {
5360 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5364 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5366 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5367 ret = fixup_tree_root_location(root, dir, dentry,
5368 &location, &sub_root);
5371 inode = ERR_PTR(ret);
5373 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5375 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5377 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5379 if (!IS_ERR(inode) && root != sub_root) {
5380 down_read(&root->fs_info->cleanup_work_sem);
5381 if (!(inode->i_sb->s_flags & MS_RDONLY))
5382 ret = btrfs_orphan_cleanup(sub_root);
5383 up_read(&root->fs_info->cleanup_work_sem);
5386 inode = ERR_PTR(ret);
5393 static int btrfs_dentry_delete(const struct dentry *dentry)
5395 struct btrfs_root *root;
5396 struct inode *inode = dentry->d_inode;
5398 if (!inode && !IS_ROOT(dentry))
5399 inode = dentry->d_parent->d_inode;
5402 root = BTRFS_I(inode)->root;
5403 if (btrfs_root_refs(&root->root_item) == 0)
5406 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5412 static void btrfs_dentry_release(struct dentry *dentry)
5414 kfree(dentry->d_fsdata);
5417 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5420 struct inode *inode;
5422 inode = btrfs_lookup_dentry(dir, dentry);
5423 if (IS_ERR(inode)) {
5424 if (PTR_ERR(inode) == -ENOENT)
5427 return ERR_CAST(inode);
5430 return d_splice_alias(inode, dentry);
5433 unsigned char btrfs_filetype_table[] = {
5434 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5437 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5439 struct inode *inode = file_inode(file);
5440 struct btrfs_root *root = BTRFS_I(inode)->root;
5441 struct btrfs_item *item;
5442 struct btrfs_dir_item *di;
5443 struct btrfs_key key;
5444 struct btrfs_key found_key;
5445 struct btrfs_path *path;
5446 struct list_head ins_list;
5447 struct list_head del_list;
5449 struct extent_buffer *leaf;
5451 unsigned char d_type;
5456 int key_type = BTRFS_DIR_INDEX_KEY;
5460 int is_curr = 0; /* ctx->pos points to the current index? */
5462 /* FIXME, use a real flag for deciding about the key type */
5463 if (root->fs_info->tree_root == root)
5464 key_type = BTRFS_DIR_ITEM_KEY;
5466 if (!dir_emit_dots(file, ctx))
5469 path = btrfs_alloc_path();
5475 if (key_type == BTRFS_DIR_INDEX_KEY) {
5476 INIT_LIST_HEAD(&ins_list);
5477 INIT_LIST_HEAD(&del_list);
5478 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5481 key.type = key_type;
5482 key.offset = ctx->pos;
5483 key.objectid = btrfs_ino(inode);
5485 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5490 leaf = path->nodes[0];
5491 slot = path->slots[0];
5492 if (slot >= btrfs_header_nritems(leaf)) {
5493 ret = btrfs_next_leaf(root, path);
5501 item = btrfs_item_nr(slot);
5502 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5504 if (found_key.objectid != key.objectid)
5506 if (found_key.type != key_type)
5508 if (found_key.offset < ctx->pos)
5510 if (key_type == BTRFS_DIR_INDEX_KEY &&
5511 btrfs_should_delete_dir_index(&del_list,
5515 ctx->pos = found_key.offset;
5518 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5520 di_total = btrfs_item_size(leaf, item);
5522 while (di_cur < di_total) {
5523 struct btrfs_key location;
5525 if (verify_dir_item(root, leaf, di))
5528 name_len = btrfs_dir_name_len(leaf, di);
5529 if (name_len <= sizeof(tmp_name)) {
5530 name_ptr = tmp_name;
5532 name_ptr = kmalloc(name_len, GFP_NOFS);
5538 read_extent_buffer(leaf, name_ptr,
5539 (unsigned long)(di + 1), name_len);
5541 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5542 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5545 /* is this a reference to our own snapshot? If so
5548 * In contrast to old kernels, we insert the snapshot's
5549 * dir item and dir index after it has been created, so
5550 * we won't find a reference to our own snapshot. We
5551 * still keep the following code for backward
5554 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5555 location.objectid == root->root_key.objectid) {
5559 over = !dir_emit(ctx, name_ptr, name_len,
5560 location.objectid, d_type);
5563 if (name_ptr != tmp_name)
5568 di_len = btrfs_dir_name_len(leaf, di) +
5569 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5571 di = (struct btrfs_dir_item *)((char *)di + di_len);
5577 if (key_type == BTRFS_DIR_INDEX_KEY) {
5580 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5585 /* Reached end of directory/root. Bump pos past the last item. */
5589 * Stop new entries from being returned after we return the last
5592 * New directory entries are assigned a strictly increasing
5593 * offset. This means that new entries created during readdir
5594 * are *guaranteed* to be seen in the future by that readdir.
5595 * This has broken buggy programs which operate on names as
5596 * they're returned by readdir. Until we re-use freed offsets
5597 * we have this hack to stop new entries from being returned
5598 * under the assumption that they'll never reach this huge
5601 * This is being careful not to overflow 32bit loff_t unless the
5602 * last entry requires it because doing so has broken 32bit apps
5605 if (key_type == BTRFS_DIR_INDEX_KEY) {
5606 if (ctx->pos >= INT_MAX)
5607 ctx->pos = LLONG_MAX;
5614 if (key_type == BTRFS_DIR_INDEX_KEY)
5615 btrfs_put_delayed_items(&ins_list, &del_list);
5616 btrfs_free_path(path);
5620 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5622 struct btrfs_root *root = BTRFS_I(inode)->root;
5623 struct btrfs_trans_handle *trans;
5625 bool nolock = false;
5627 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5630 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5633 if (wbc->sync_mode == WB_SYNC_ALL) {
5635 trans = btrfs_join_transaction_nolock(root);
5637 trans = btrfs_join_transaction(root);
5639 return PTR_ERR(trans);
5640 ret = btrfs_commit_transaction(trans, root);
5646 * This is somewhat expensive, updating the tree every time the
5647 * inode changes. But, it is most likely to find the inode in cache.
5648 * FIXME, needs more benchmarking...there are no reasons other than performance
5649 * to keep or drop this code.
5651 static int btrfs_dirty_inode(struct inode *inode)
5653 struct btrfs_root *root = BTRFS_I(inode)->root;
5654 struct btrfs_trans_handle *trans;
5657 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5660 trans = btrfs_join_transaction(root);
5662 return PTR_ERR(trans);
5664 ret = btrfs_update_inode(trans, root, inode);
5665 if (ret && ret == -ENOSPC) {
5666 /* whoops, lets try again with the full transaction */
5667 btrfs_end_transaction(trans, root);
5668 trans = btrfs_start_transaction(root, 1);
5670 return PTR_ERR(trans);
5672 ret = btrfs_update_inode(trans, root, inode);
5674 btrfs_end_transaction(trans, root);
5675 if (BTRFS_I(inode)->delayed_node)
5676 btrfs_balance_delayed_items(root);
5682 * This is a copy of file_update_time. We need this so we can return error on
5683 * ENOSPC for updating the inode in the case of file write and mmap writes.
5685 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5688 struct btrfs_root *root = BTRFS_I(inode)->root;
5690 if (btrfs_root_readonly(root))
5693 if (flags & S_VERSION)
5694 inode_inc_iversion(inode);
5695 if (flags & S_CTIME)
5696 inode->i_ctime = *now;
5697 if (flags & S_MTIME)
5698 inode->i_mtime = *now;
5699 if (flags & S_ATIME)
5700 inode->i_atime = *now;
5701 return btrfs_dirty_inode(inode);
5705 * find the highest existing sequence number in a directory
5706 * and then set the in-memory index_cnt variable to reflect
5707 * free sequence numbers
5709 static int btrfs_set_inode_index_count(struct inode *inode)
5711 struct btrfs_root *root = BTRFS_I(inode)->root;
5712 struct btrfs_key key, found_key;
5713 struct btrfs_path *path;
5714 struct extent_buffer *leaf;
5717 key.objectid = btrfs_ino(inode);
5718 key.type = BTRFS_DIR_INDEX_KEY;
5719 key.offset = (u64)-1;
5721 path = btrfs_alloc_path();
5725 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5728 /* FIXME: we should be able to handle this */
5734 * MAGIC NUMBER EXPLANATION:
5735 * since we search a directory based on f_pos we have to start at 2
5736 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5737 * else has to start at 2
5739 if (path->slots[0] == 0) {
5740 BTRFS_I(inode)->index_cnt = 2;
5746 leaf = path->nodes[0];
5747 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5749 if (found_key.objectid != btrfs_ino(inode) ||
5750 found_key.type != BTRFS_DIR_INDEX_KEY) {
5751 BTRFS_I(inode)->index_cnt = 2;
5755 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5757 btrfs_free_path(path);
5762 * helper to find a free sequence number in a given directory. This current
5763 * code is very simple, later versions will do smarter things in the btree
5765 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5769 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5770 ret = btrfs_inode_delayed_dir_index_count(dir);
5772 ret = btrfs_set_inode_index_count(dir);
5778 *index = BTRFS_I(dir)->index_cnt;
5779 BTRFS_I(dir)->index_cnt++;
5784 static int btrfs_insert_inode_locked(struct inode *inode)
5786 struct btrfs_iget_args args;
5787 args.location = &BTRFS_I(inode)->location;
5788 args.root = BTRFS_I(inode)->root;
5790 return insert_inode_locked4(inode,
5791 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
5792 btrfs_find_actor, &args);
5795 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5796 struct btrfs_root *root,
5798 const char *name, int name_len,
5799 u64 ref_objectid, u64 objectid,
5800 umode_t mode, u64 *index)
5802 struct inode *inode;
5803 struct btrfs_inode_item *inode_item;
5804 struct btrfs_key *location;
5805 struct btrfs_path *path;
5806 struct btrfs_inode_ref *ref;
5807 struct btrfs_key key[2];
5809 int nitems = name ? 2 : 1;
5813 path = btrfs_alloc_path();
5815 return ERR_PTR(-ENOMEM);
5817 inode = new_inode(root->fs_info->sb);
5819 btrfs_free_path(path);
5820 return ERR_PTR(-ENOMEM);
5824 * O_TMPFILE, set link count to 0, so that after this point,
5825 * we fill in an inode item with the correct link count.
5828 set_nlink(inode, 0);
5831 * we have to initialize this early, so we can reclaim the inode
5832 * number if we fail afterwards in this function.
5834 inode->i_ino = objectid;
5837 trace_btrfs_inode_request(dir);
5839 ret = btrfs_set_inode_index(dir, index);
5841 btrfs_free_path(path);
5843 return ERR_PTR(ret);
5849 * index_cnt is ignored for everything but a dir,
5850 * btrfs_get_inode_index_count has an explanation for the magic
5853 BTRFS_I(inode)->index_cnt = 2;
5854 BTRFS_I(inode)->dir_index = *index;
5855 BTRFS_I(inode)->root = root;
5856 BTRFS_I(inode)->generation = trans->transid;
5857 inode->i_generation = BTRFS_I(inode)->generation;
5860 * We could have gotten an inode number from somebody who was fsynced
5861 * and then removed in this same transaction, so let's just set full
5862 * sync since it will be a full sync anyway and this will blow away the
5863 * old info in the log.
5865 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5867 key[0].objectid = objectid;
5868 key[0].type = BTRFS_INODE_ITEM_KEY;
5871 sizes[0] = sizeof(struct btrfs_inode_item);
5875 * Start new inodes with an inode_ref. This is slightly more
5876 * efficient for small numbers of hard links since they will
5877 * be packed into one item. Extended refs will kick in if we
5878 * add more hard links than can fit in the ref item.
5880 key[1].objectid = objectid;
5881 key[1].type = BTRFS_INODE_REF_KEY;
5882 key[1].offset = ref_objectid;
5884 sizes[1] = name_len + sizeof(*ref);
5887 location = &BTRFS_I(inode)->location;
5888 location->objectid = objectid;
5889 location->offset = 0;
5890 location->type = BTRFS_INODE_ITEM_KEY;
5892 ret = btrfs_insert_inode_locked(inode);
5896 path->leave_spinning = 1;
5897 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
5901 inode_init_owner(inode, dir, mode);
5902 inode_set_bytes(inode, 0);
5904 inode->i_mtime = CURRENT_TIME;
5905 inode->i_atime = inode->i_mtime;
5906 inode->i_ctime = inode->i_mtime;
5907 BTRFS_I(inode)->i_otime = inode->i_mtime;
5909 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5910 struct btrfs_inode_item);
5911 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5912 sizeof(*inode_item));
5913 fill_inode_item(trans, path->nodes[0], inode_item, inode);
5916 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5917 struct btrfs_inode_ref);
5918 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5919 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5920 ptr = (unsigned long)(ref + 1);
5921 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5924 btrfs_mark_buffer_dirty(path->nodes[0]);
5925 btrfs_free_path(path);
5927 btrfs_inherit_iflags(inode, dir);
5929 if (S_ISREG(mode)) {
5930 if (btrfs_test_opt(root, NODATASUM))
5931 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5932 if (btrfs_test_opt(root, NODATACOW))
5933 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5934 BTRFS_INODE_NODATASUM;
5937 inode_tree_add(inode);
5939 trace_btrfs_inode_new(inode);
5940 btrfs_set_inode_last_trans(trans, inode);
5942 btrfs_update_root_times(trans, root);
5944 ret = btrfs_inode_inherit_props(trans, inode, dir);
5946 btrfs_err(root->fs_info,
5947 "error inheriting props for ino %llu (root %llu): %d",
5948 btrfs_ino(inode), root->root_key.objectid, ret);
5953 unlock_new_inode(inode);
5956 BTRFS_I(dir)->index_cnt--;
5957 btrfs_free_path(path);
5959 return ERR_PTR(ret);
5962 static inline u8 btrfs_inode_type(struct inode *inode)
5964 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5968 * utility function to add 'inode' into 'parent_inode' with
5969 * a give name and a given sequence number.
5970 * if 'add_backref' is true, also insert a backref from the
5971 * inode to the parent directory.
5973 int btrfs_add_link(struct btrfs_trans_handle *trans,
5974 struct inode *parent_inode, struct inode *inode,
5975 const char *name, int name_len, int add_backref, u64 index)
5978 struct btrfs_key key;
5979 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5980 u64 ino = btrfs_ino(inode);
5981 u64 parent_ino = btrfs_ino(parent_inode);
5983 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5984 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5987 key.type = BTRFS_INODE_ITEM_KEY;
5991 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5992 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5993 key.objectid, root->root_key.objectid,
5994 parent_ino, index, name, name_len);
5995 } else if (add_backref) {
5996 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6000 /* Nothing to clean up yet */
6004 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6006 btrfs_inode_type(inode), index);
6007 if (ret == -EEXIST || ret == -EOVERFLOW)
6010 btrfs_abort_transaction(trans, root, ret);
6014 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6016 inode_inc_iversion(parent_inode);
6017 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
6018 ret = btrfs_update_inode(trans, root, parent_inode);
6020 btrfs_abort_transaction(trans, root, ret);
6024 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6027 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6028 key.objectid, root->root_key.objectid,
6029 parent_ino, &local_index, name, name_len);
6031 } else if (add_backref) {
6035 err = btrfs_del_inode_ref(trans, root, name, name_len,
6036 ino, parent_ino, &local_index);
6041 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6042 struct inode *dir, struct dentry *dentry,
6043 struct inode *inode, int backref, u64 index)
6045 int err = btrfs_add_link(trans, dir, inode,
6046 dentry->d_name.name, dentry->d_name.len,
6053 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6054 umode_t mode, dev_t rdev)
6056 struct btrfs_trans_handle *trans;
6057 struct btrfs_root *root = BTRFS_I(dir)->root;
6058 struct inode *inode = NULL;
6064 if (!new_valid_dev(rdev))
6068 * 2 for inode item and ref
6070 * 1 for xattr if selinux is on
6072 trans = btrfs_start_transaction(root, 5);
6074 return PTR_ERR(trans);
6076 err = btrfs_find_free_ino(root, &objectid);
6080 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6081 dentry->d_name.len, btrfs_ino(dir), objectid,
6083 if (IS_ERR(inode)) {
6084 err = PTR_ERR(inode);
6089 * If the active LSM wants to access the inode during
6090 * d_instantiate it needs these. Smack checks to see
6091 * if the filesystem supports xattrs by looking at the
6094 inode->i_op = &btrfs_special_inode_operations;
6095 init_special_inode(inode, inode->i_mode, rdev);
6097 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6099 goto out_unlock_inode;
6101 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6103 goto out_unlock_inode;
6105 btrfs_update_inode(trans, root, inode);
6106 unlock_new_inode(inode);
6107 d_instantiate(dentry, inode);
6111 btrfs_end_transaction(trans, root);
6112 btrfs_balance_delayed_items(root);
6113 btrfs_btree_balance_dirty(root);
6115 inode_dec_link_count(inode);
6122 unlock_new_inode(inode);
6127 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6128 umode_t mode, bool excl)
6130 struct btrfs_trans_handle *trans;
6131 struct btrfs_root *root = BTRFS_I(dir)->root;
6132 struct inode *inode = NULL;
6133 int drop_inode_on_err = 0;
6139 * 2 for inode item and ref
6141 * 1 for xattr if selinux is on
6143 trans = btrfs_start_transaction(root, 5);
6145 return PTR_ERR(trans);
6147 err = btrfs_find_free_ino(root, &objectid);
6151 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6152 dentry->d_name.len, btrfs_ino(dir), objectid,
6154 if (IS_ERR(inode)) {
6155 err = PTR_ERR(inode);
6158 drop_inode_on_err = 1;
6160 * If the active LSM wants to access the inode during
6161 * d_instantiate it needs these. Smack checks to see
6162 * if the filesystem supports xattrs by looking at the
6165 inode->i_fop = &btrfs_file_operations;
6166 inode->i_op = &btrfs_file_inode_operations;
6167 inode->i_mapping->a_ops = &btrfs_aops;
6169 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6171 goto out_unlock_inode;
6173 err = btrfs_update_inode(trans, root, inode);
6175 goto out_unlock_inode;
6177 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6179 goto out_unlock_inode;
6181 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6182 unlock_new_inode(inode);
6183 d_instantiate(dentry, inode);
6186 btrfs_end_transaction(trans, root);
6187 if (err && drop_inode_on_err) {
6188 inode_dec_link_count(inode);
6191 btrfs_balance_delayed_items(root);
6192 btrfs_btree_balance_dirty(root);
6196 unlock_new_inode(inode);
6201 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6202 struct dentry *dentry)
6204 struct btrfs_trans_handle *trans;
6205 struct btrfs_root *root = BTRFS_I(dir)->root;
6206 struct inode *inode = old_dentry->d_inode;
6211 /* do not allow sys_link's with other subvols of the same device */
6212 if (root->objectid != BTRFS_I(inode)->root->objectid)
6215 if (inode->i_nlink >= BTRFS_LINK_MAX)
6218 err = btrfs_set_inode_index(dir, &index);
6223 * 2 items for inode and inode ref
6224 * 2 items for dir items
6225 * 1 item for parent inode
6227 trans = btrfs_start_transaction(root, 5);
6228 if (IS_ERR(trans)) {
6229 err = PTR_ERR(trans);
6233 /* There are several dir indexes for this inode, clear the cache. */
6234 BTRFS_I(inode)->dir_index = 0ULL;
6236 inode_inc_iversion(inode);
6237 inode->i_ctime = CURRENT_TIME;
6239 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6241 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6246 struct dentry *parent = dentry->d_parent;
6247 err = btrfs_update_inode(trans, root, inode);
6250 if (inode->i_nlink == 1) {
6252 * If new hard link count is 1, it's a file created
6253 * with open(2) O_TMPFILE flag.
6255 err = btrfs_orphan_del(trans, inode);
6259 d_instantiate(dentry, inode);
6260 btrfs_log_new_name(trans, inode, NULL, parent);
6263 btrfs_end_transaction(trans, root);
6264 btrfs_balance_delayed_items(root);
6267 inode_dec_link_count(inode);
6270 btrfs_btree_balance_dirty(root);
6274 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6276 struct inode *inode = NULL;
6277 struct btrfs_trans_handle *trans;
6278 struct btrfs_root *root = BTRFS_I(dir)->root;
6280 int drop_on_err = 0;
6285 * 2 items for inode and ref
6286 * 2 items for dir items
6287 * 1 for xattr if selinux is on
6289 trans = btrfs_start_transaction(root, 5);
6291 return PTR_ERR(trans);
6293 err = btrfs_find_free_ino(root, &objectid);
6297 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6298 dentry->d_name.len, btrfs_ino(dir), objectid,
6299 S_IFDIR | mode, &index);
6300 if (IS_ERR(inode)) {
6301 err = PTR_ERR(inode);
6306 /* these must be set before we unlock the inode */
6307 inode->i_op = &btrfs_dir_inode_operations;
6308 inode->i_fop = &btrfs_dir_file_operations;
6310 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6312 goto out_fail_inode;
6314 btrfs_i_size_write(inode, 0);
6315 err = btrfs_update_inode(trans, root, inode);
6317 goto out_fail_inode;
6319 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6320 dentry->d_name.len, 0, index);
6322 goto out_fail_inode;
6324 d_instantiate(dentry, inode);
6326 * mkdir is special. We're unlocking after we call d_instantiate
6327 * to avoid a race with nfsd calling d_instantiate.
6329 unlock_new_inode(inode);
6333 btrfs_end_transaction(trans, root);
6335 inode_dec_link_count(inode);
6338 btrfs_balance_delayed_items(root);
6339 btrfs_btree_balance_dirty(root);
6343 unlock_new_inode(inode);
6347 /* Find next extent map of a given extent map, caller needs to ensure locks */
6348 static struct extent_map *next_extent_map(struct extent_map *em)
6350 struct rb_node *next;
6352 next = rb_next(&em->rb_node);
6355 return container_of(next, struct extent_map, rb_node);
6358 static struct extent_map *prev_extent_map(struct extent_map *em)
6360 struct rb_node *prev;
6362 prev = rb_prev(&em->rb_node);
6365 return container_of(prev, struct extent_map, rb_node);
6368 /* helper for btfs_get_extent. Given an existing extent in the tree,
6369 * the existing extent is the nearest extent to map_start,
6370 * and an extent that you want to insert, deal with overlap and insert
6371 * the best fitted new extent into the tree.
6373 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6374 struct extent_map *existing,
6375 struct extent_map *em,
6378 struct extent_map *prev;
6379 struct extent_map *next;
6384 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6386 if (existing->start > map_start) {
6388 prev = prev_extent_map(next);
6391 next = next_extent_map(prev);
6394 start = prev ? extent_map_end(prev) : em->start;
6395 start = max_t(u64, start, em->start);
6396 end = next ? next->start : extent_map_end(em);
6397 end = min_t(u64, end, extent_map_end(em));
6398 start_diff = start - em->start;
6400 em->len = end - start;
6401 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6402 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6403 em->block_start += start_diff;
6404 em->block_len -= start_diff;
6406 return add_extent_mapping(em_tree, em, 0);
6409 static noinline int uncompress_inline(struct btrfs_path *path,
6410 struct inode *inode, struct page *page,
6411 size_t pg_offset, u64 extent_offset,
6412 struct btrfs_file_extent_item *item)
6415 struct extent_buffer *leaf = path->nodes[0];
6418 unsigned long inline_size;
6422 WARN_ON(pg_offset != 0);
6423 compress_type = btrfs_file_extent_compression(leaf, item);
6424 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6425 inline_size = btrfs_file_extent_inline_item_len(leaf,
6426 btrfs_item_nr(path->slots[0]));
6427 tmp = kmalloc(inline_size, GFP_NOFS);
6430 ptr = btrfs_file_extent_inline_start(item);
6432 read_extent_buffer(leaf, tmp, ptr, inline_size);
6434 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6435 ret = btrfs_decompress(compress_type, tmp, page,
6436 extent_offset, inline_size, max_size);
6442 * a bit scary, this does extent mapping from logical file offset to the disk.
6443 * the ugly parts come from merging extents from the disk with the in-ram
6444 * representation. This gets more complex because of the data=ordered code,
6445 * where the in-ram extents might be locked pending data=ordered completion.
6447 * This also copies inline extents directly into the page.
6450 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6451 size_t pg_offset, u64 start, u64 len,
6456 u64 extent_start = 0;
6458 u64 objectid = btrfs_ino(inode);
6460 struct btrfs_path *path = NULL;
6461 struct btrfs_root *root = BTRFS_I(inode)->root;
6462 struct btrfs_file_extent_item *item;
6463 struct extent_buffer *leaf;
6464 struct btrfs_key found_key;
6465 struct extent_map *em = NULL;
6466 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6467 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6468 struct btrfs_trans_handle *trans = NULL;
6469 const bool new_inline = !page || create;
6472 read_lock(&em_tree->lock);
6473 em = lookup_extent_mapping(em_tree, start, len);
6475 em->bdev = root->fs_info->fs_devices->latest_bdev;
6476 read_unlock(&em_tree->lock);
6479 if (em->start > start || em->start + em->len <= start)
6480 free_extent_map(em);
6481 else if (em->block_start == EXTENT_MAP_INLINE && page)
6482 free_extent_map(em);
6486 em = alloc_extent_map();
6491 em->bdev = root->fs_info->fs_devices->latest_bdev;
6492 em->start = EXTENT_MAP_HOLE;
6493 em->orig_start = EXTENT_MAP_HOLE;
6495 em->block_len = (u64)-1;
6498 path = btrfs_alloc_path();
6504 * Chances are we'll be called again, so go ahead and do
6510 ret = btrfs_lookup_file_extent(trans, root, path,
6511 objectid, start, trans != NULL);
6518 if (path->slots[0] == 0)
6523 leaf = path->nodes[0];
6524 item = btrfs_item_ptr(leaf, path->slots[0],
6525 struct btrfs_file_extent_item);
6526 /* are we inside the extent that was found? */
6527 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6528 found_type = found_key.type;
6529 if (found_key.objectid != objectid ||
6530 found_type != BTRFS_EXTENT_DATA_KEY) {
6532 * If we backup past the first extent we want to move forward
6533 * and see if there is an extent in front of us, otherwise we'll
6534 * say there is a hole for our whole search range which can
6541 found_type = btrfs_file_extent_type(leaf, item);
6542 extent_start = found_key.offset;
6543 if (found_type == BTRFS_FILE_EXTENT_REG ||
6544 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6545 extent_end = extent_start +
6546 btrfs_file_extent_num_bytes(leaf, item);
6547 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6549 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6550 extent_end = ALIGN(extent_start + size, root->sectorsize);
6553 if (start >= extent_end) {
6555 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6556 ret = btrfs_next_leaf(root, path);
6563 leaf = path->nodes[0];
6565 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6566 if (found_key.objectid != objectid ||
6567 found_key.type != BTRFS_EXTENT_DATA_KEY)
6569 if (start + len <= found_key.offset)
6571 if (start > found_key.offset)
6574 em->orig_start = start;
6575 em->len = found_key.offset - start;
6579 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6581 if (found_type == BTRFS_FILE_EXTENT_REG ||
6582 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6584 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6588 size_t extent_offset;
6594 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6595 extent_offset = page_offset(page) + pg_offset - extent_start;
6596 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6597 size - extent_offset);
6598 em->start = extent_start + extent_offset;
6599 em->len = ALIGN(copy_size, root->sectorsize);
6600 em->orig_block_len = em->len;
6601 em->orig_start = em->start;
6602 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6603 if (create == 0 && !PageUptodate(page)) {
6604 if (btrfs_file_extent_compression(leaf, item) !=
6605 BTRFS_COMPRESS_NONE) {
6606 ret = uncompress_inline(path, inode, page,
6608 extent_offset, item);
6615 read_extent_buffer(leaf, map + pg_offset, ptr,
6617 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6618 memset(map + pg_offset + copy_size, 0,
6619 PAGE_CACHE_SIZE - pg_offset -
6624 flush_dcache_page(page);
6625 } else if (create && PageUptodate(page)) {
6629 free_extent_map(em);
6632 btrfs_release_path(path);
6633 trans = btrfs_join_transaction(root);
6636 return ERR_CAST(trans);
6640 write_extent_buffer(leaf, map + pg_offset, ptr,
6643 btrfs_mark_buffer_dirty(leaf);
6645 set_extent_uptodate(io_tree, em->start,
6646 extent_map_end(em) - 1, NULL, GFP_NOFS);
6651 em->orig_start = start;
6654 em->block_start = EXTENT_MAP_HOLE;
6655 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6657 btrfs_release_path(path);
6658 if (em->start > start || extent_map_end(em) <= start) {
6659 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6660 em->start, em->len, start, len);
6666 write_lock(&em_tree->lock);
6667 ret = add_extent_mapping(em_tree, em, 0);
6668 /* it is possible that someone inserted the extent into the tree
6669 * while we had the lock dropped. It is also possible that
6670 * an overlapping map exists in the tree
6672 if (ret == -EEXIST) {
6673 struct extent_map *existing;
6677 existing = search_extent_mapping(em_tree, start, len);
6679 * existing will always be non-NULL, since there must be
6680 * extent causing the -EEXIST.
6682 if (start >= extent_map_end(existing) ||
6683 start <= existing->start) {
6685 * The existing extent map is the one nearest to
6686 * the [start, start + len) range which overlaps
6688 err = merge_extent_mapping(em_tree, existing,
6690 free_extent_map(existing);
6692 free_extent_map(em);
6696 free_extent_map(em);
6701 write_unlock(&em_tree->lock);
6704 trace_btrfs_get_extent(root, em);
6707 btrfs_free_path(path);
6709 ret = btrfs_end_transaction(trans, root);
6714 free_extent_map(em);
6715 return ERR_PTR(err);
6717 BUG_ON(!em); /* Error is always set */
6721 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6722 size_t pg_offset, u64 start, u64 len,
6725 struct extent_map *em;
6726 struct extent_map *hole_em = NULL;
6727 u64 range_start = start;
6733 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6740 * - a pre-alloc extent,
6741 * there might actually be delalloc bytes behind it.
6743 if (em->block_start != EXTENT_MAP_HOLE &&
6744 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6750 /* check to see if we've wrapped (len == -1 or similar) */
6759 /* ok, we didn't find anything, lets look for delalloc */
6760 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6761 end, len, EXTENT_DELALLOC, 1);
6762 found_end = range_start + found;
6763 if (found_end < range_start)
6764 found_end = (u64)-1;
6767 * we didn't find anything useful, return
6768 * the original results from get_extent()
6770 if (range_start > end || found_end <= start) {
6776 /* adjust the range_start to make sure it doesn't
6777 * go backwards from the start they passed in
6779 range_start = max(start, range_start);
6780 found = found_end - range_start;
6783 u64 hole_start = start;
6786 em = alloc_extent_map();
6792 * when btrfs_get_extent can't find anything it
6793 * returns one huge hole
6795 * make sure what it found really fits our range, and
6796 * adjust to make sure it is based on the start from
6800 u64 calc_end = extent_map_end(hole_em);
6802 if (calc_end <= start || (hole_em->start > end)) {
6803 free_extent_map(hole_em);
6806 hole_start = max(hole_em->start, start);
6807 hole_len = calc_end - hole_start;
6811 if (hole_em && range_start > hole_start) {
6812 /* our hole starts before our delalloc, so we
6813 * have to return just the parts of the hole
6814 * that go until the delalloc starts
6816 em->len = min(hole_len,
6817 range_start - hole_start);
6818 em->start = hole_start;
6819 em->orig_start = hole_start;
6821 * don't adjust block start at all,
6822 * it is fixed at EXTENT_MAP_HOLE
6824 em->block_start = hole_em->block_start;
6825 em->block_len = hole_len;
6826 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6827 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6829 em->start = range_start;
6831 em->orig_start = range_start;
6832 em->block_start = EXTENT_MAP_DELALLOC;
6833 em->block_len = found;
6835 } else if (hole_em) {
6840 free_extent_map(hole_em);
6842 free_extent_map(em);
6843 return ERR_PTR(err);
6848 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6851 struct btrfs_root *root = BTRFS_I(inode)->root;
6852 struct extent_map *em;
6853 struct btrfs_key ins;
6857 alloc_hint = get_extent_allocation_hint(inode, start, len);
6858 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
6859 alloc_hint, &ins, 1, 1);
6861 return ERR_PTR(ret);
6863 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6864 ins.offset, ins.offset, ins.offset, 0);
6866 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
6870 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6871 ins.offset, ins.offset, 0);
6873 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
6874 free_extent_map(em);
6875 return ERR_PTR(ret);
6882 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6883 * block must be cow'd
6885 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6886 u64 *orig_start, u64 *orig_block_len,
6889 struct btrfs_trans_handle *trans;
6890 struct btrfs_path *path;
6892 struct extent_buffer *leaf;
6893 struct btrfs_root *root = BTRFS_I(inode)->root;
6894 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6895 struct btrfs_file_extent_item *fi;
6896 struct btrfs_key key;
6903 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6905 path = btrfs_alloc_path();
6909 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
6914 slot = path->slots[0];
6917 /* can't find the item, must cow */
6924 leaf = path->nodes[0];
6925 btrfs_item_key_to_cpu(leaf, &key, slot);
6926 if (key.objectid != btrfs_ino(inode) ||
6927 key.type != BTRFS_EXTENT_DATA_KEY) {
6928 /* not our file or wrong item type, must cow */
6932 if (key.offset > offset) {
6933 /* Wrong offset, must cow */
6937 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6938 found_type = btrfs_file_extent_type(leaf, fi);
6939 if (found_type != BTRFS_FILE_EXTENT_REG &&
6940 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6941 /* not a regular extent, must cow */
6945 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6948 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6949 if (extent_end <= offset)
6952 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6953 if (disk_bytenr == 0)
6956 if (btrfs_file_extent_compression(leaf, fi) ||
6957 btrfs_file_extent_encryption(leaf, fi) ||
6958 btrfs_file_extent_other_encoding(leaf, fi))
6961 backref_offset = btrfs_file_extent_offset(leaf, fi);
6964 *orig_start = key.offset - backref_offset;
6965 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6966 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6969 if (btrfs_extent_readonly(root, disk_bytenr))
6972 num_bytes = min(offset + *len, extent_end) - offset;
6973 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6976 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
6977 ret = test_range_bit(io_tree, offset, range_end,
6978 EXTENT_DELALLOC, 0, NULL);
6985 btrfs_release_path(path);
6988 * look for other files referencing this extent, if we
6989 * find any we must cow
6991 trans = btrfs_join_transaction(root);
6992 if (IS_ERR(trans)) {
6997 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6998 key.offset - backref_offset, disk_bytenr);
6999 btrfs_end_transaction(trans, root);
7006 * adjust disk_bytenr and num_bytes to cover just the bytes
7007 * in this extent we are about to write. If there
7008 * are any csums in that range we have to cow in order
7009 * to keep the csums correct
7011 disk_bytenr += backref_offset;
7012 disk_bytenr += offset - key.offset;
7013 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7016 * all of the above have passed, it is safe to overwrite this extent
7022 btrfs_free_path(path);
7026 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7028 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7030 void **pagep = NULL;
7031 struct page *page = NULL;
7035 start_idx = start >> PAGE_CACHE_SHIFT;
7038 * end is the last byte in the last page. end == start is legal
7040 end_idx = end >> PAGE_CACHE_SHIFT;
7044 /* Most of the code in this while loop is lifted from
7045 * find_get_page. It's been modified to begin searching from a
7046 * page and return just the first page found in that range. If the
7047 * found idx is less than or equal to the end idx then we know that
7048 * a page exists. If no pages are found or if those pages are
7049 * outside of the range then we're fine (yay!) */
7050 while (page == NULL &&
7051 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7052 page = radix_tree_deref_slot(pagep);
7053 if (unlikely(!page))
7056 if (radix_tree_exception(page)) {
7057 if (radix_tree_deref_retry(page)) {
7062 * Otherwise, shmem/tmpfs must be storing a swap entry
7063 * here as an exceptional entry: so return it without
7064 * attempting to raise page count.
7067 break; /* TODO: Is this relevant for this use case? */
7070 if (!page_cache_get_speculative(page)) {
7076 * Has the page moved?
7077 * This is part of the lockless pagecache protocol. See
7078 * include/linux/pagemap.h for details.
7080 if (unlikely(page != *pagep)) {
7081 page_cache_release(page);
7087 if (page->index <= end_idx)
7089 page_cache_release(page);
7096 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7097 struct extent_state **cached_state, int writing)
7099 struct btrfs_ordered_extent *ordered;
7103 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7106 * We're concerned with the entire range that we're going to be
7107 * doing DIO to, so we need to make sure theres no ordered
7108 * extents in this range.
7110 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7111 lockend - lockstart + 1);
7114 * We need to make sure there are no buffered pages in this
7115 * range either, we could have raced between the invalidate in
7116 * generic_file_direct_write and locking the extent. The
7117 * invalidate needs to happen so that reads after a write do not
7122 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7125 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7126 cached_state, GFP_NOFS);
7129 btrfs_start_ordered_extent(inode, ordered, 1);
7130 btrfs_put_ordered_extent(ordered);
7132 /* Screw you mmap */
7133 ret = btrfs_fdatawrite_range(inode, lockstart, lockend);
7136 ret = filemap_fdatawait_range(inode->i_mapping,
7143 * If we found a page that couldn't be invalidated just
7144 * fall back to buffered.
7146 ret = invalidate_inode_pages2_range(inode->i_mapping,
7147 lockstart >> PAGE_CACHE_SHIFT,
7148 lockend >> PAGE_CACHE_SHIFT);
7159 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7160 u64 len, u64 orig_start,
7161 u64 block_start, u64 block_len,
7162 u64 orig_block_len, u64 ram_bytes,
7165 struct extent_map_tree *em_tree;
7166 struct extent_map *em;
7167 struct btrfs_root *root = BTRFS_I(inode)->root;
7170 em_tree = &BTRFS_I(inode)->extent_tree;
7171 em = alloc_extent_map();
7173 return ERR_PTR(-ENOMEM);
7176 em->orig_start = orig_start;
7177 em->mod_start = start;
7180 em->block_len = block_len;
7181 em->block_start = block_start;
7182 em->bdev = root->fs_info->fs_devices->latest_bdev;
7183 em->orig_block_len = orig_block_len;
7184 em->ram_bytes = ram_bytes;
7185 em->generation = -1;
7186 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7187 if (type == BTRFS_ORDERED_PREALLOC)
7188 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7191 btrfs_drop_extent_cache(inode, em->start,
7192 em->start + em->len - 1, 0);
7193 write_lock(&em_tree->lock);
7194 ret = add_extent_mapping(em_tree, em, 1);
7195 write_unlock(&em_tree->lock);
7196 } while (ret == -EEXIST);
7199 free_extent_map(em);
7200 return ERR_PTR(ret);
7207 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7208 struct buffer_head *bh_result, int create)
7210 struct extent_map *em;
7211 struct btrfs_root *root = BTRFS_I(inode)->root;
7212 struct extent_state *cached_state = NULL;
7213 u64 start = iblock << inode->i_blkbits;
7214 u64 lockstart, lockend;
7215 u64 len = bh_result->b_size;
7217 int unlock_bits = EXTENT_LOCKED;
7221 unlock_bits |= EXTENT_DIRTY;
7223 len = min_t(u64, len, root->sectorsize);
7226 lockend = start + len - 1;
7229 * If this errors out it's because we couldn't invalidate pagecache for
7230 * this range and we need to fallback to buffered.
7232 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
7235 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7242 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7243 * io. INLINE is special, and we could probably kludge it in here, but
7244 * it's still buffered so for safety lets just fall back to the generic
7247 * For COMPRESSED we _have_ to read the entire extent in so we can
7248 * decompress it, so there will be buffering required no matter what we
7249 * do, so go ahead and fallback to buffered.
7251 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7252 * to buffered IO. Don't blame me, this is the price we pay for using
7255 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7256 em->block_start == EXTENT_MAP_INLINE) {
7257 free_extent_map(em);
7262 /* Just a good old fashioned hole, return */
7263 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7264 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7265 free_extent_map(em);
7270 * We don't allocate a new extent in the following cases
7272 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7274 * 2) The extent is marked as PREALLOC. We're good to go here and can
7275 * just use the extent.
7279 len = min(len, em->len - (start - em->start));
7280 lockstart = start + len;
7284 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7285 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7286 em->block_start != EXTENT_MAP_HOLE)) {
7289 u64 block_start, orig_start, orig_block_len, ram_bytes;
7291 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7292 type = BTRFS_ORDERED_PREALLOC;
7294 type = BTRFS_ORDERED_NOCOW;
7295 len = min(len, em->len - (start - em->start));
7296 block_start = em->block_start + (start - em->start);
7298 if (can_nocow_extent(inode, start, &len, &orig_start,
7299 &orig_block_len, &ram_bytes) == 1) {
7300 if (type == BTRFS_ORDERED_PREALLOC) {
7301 free_extent_map(em);
7302 em = create_pinned_em(inode, start, len,
7313 ret = btrfs_add_ordered_extent_dio(inode, start,
7314 block_start, len, len, type);
7316 free_extent_map(em);
7324 * this will cow the extent, reset the len in case we changed
7327 len = bh_result->b_size;
7328 free_extent_map(em);
7329 em = btrfs_new_extent_direct(inode, start, len);
7334 len = min(len, em->len - (start - em->start));
7336 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7338 bh_result->b_size = len;
7339 bh_result->b_bdev = em->bdev;
7340 set_buffer_mapped(bh_result);
7342 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7343 set_buffer_new(bh_result);
7346 * Need to update the i_size under the extent lock so buffered
7347 * readers will get the updated i_size when we unlock.
7349 if (start + len > i_size_read(inode))
7350 i_size_write(inode, start + len);
7352 if (len < orig_len) {
7353 spin_lock(&BTRFS_I(inode)->lock);
7354 BTRFS_I(inode)->outstanding_extents++;
7355 spin_unlock(&BTRFS_I(inode)->lock);
7357 btrfs_free_reserved_data_space(inode, len);
7361 * In the case of write we need to clear and unlock the entire range,
7362 * in the case of read we need to unlock only the end area that we
7363 * aren't using if there is any left over space.
7365 if (lockstart < lockend) {
7366 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7367 lockend, unlock_bits, 1, 0,
7368 &cached_state, GFP_NOFS);
7370 free_extent_state(cached_state);
7373 free_extent_map(em);
7378 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7379 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7383 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7384 int rw, int mirror_num)
7386 struct btrfs_root *root = BTRFS_I(inode)->root;
7389 BUG_ON(rw & REQ_WRITE);
7393 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7394 BTRFS_WQ_ENDIO_DIO_REPAIR);
7398 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7404 static int btrfs_check_dio_repairable(struct inode *inode,
7405 struct bio *failed_bio,
7406 struct io_failure_record *failrec,
7411 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7412 failrec->logical, failrec->len);
7413 if (num_copies == 1) {
7415 * we only have a single copy of the data, so don't bother with
7416 * all the retry and error correction code that follows. no
7417 * matter what the error is, it is very likely to persist.
7419 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7420 num_copies, failrec->this_mirror, failed_mirror);
7424 failrec->failed_mirror = failed_mirror;
7425 failrec->this_mirror++;
7426 if (failrec->this_mirror == failed_mirror)
7427 failrec->this_mirror++;
7429 if (failrec->this_mirror > num_copies) {
7430 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7431 num_copies, failrec->this_mirror, failed_mirror);
7438 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7439 struct page *page, u64 start, u64 end,
7440 int failed_mirror, bio_end_io_t *repair_endio,
7443 struct io_failure_record *failrec;
7449 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7451 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7455 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7458 free_io_failure(inode, failrec);
7462 if (failed_bio->bi_vcnt > 1)
7463 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7465 read_mode = READ_SYNC;
7467 isector = start - btrfs_io_bio(failed_bio)->logical;
7468 isector >>= inode->i_sb->s_blocksize_bits;
7469 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7470 0, isector, repair_endio, repair_arg);
7472 free_io_failure(inode, failrec);
7476 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7477 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7478 read_mode, failrec->this_mirror, failrec->in_validation);
7480 ret = submit_dio_repair_bio(inode, bio, read_mode,
7481 failrec->this_mirror);
7483 free_io_failure(inode, failrec);
7490 struct btrfs_retry_complete {
7491 struct completion done;
7492 struct inode *inode;
7497 static void btrfs_retry_endio_nocsum(struct bio *bio, int err)
7499 struct btrfs_retry_complete *done = bio->bi_private;
7500 struct bio_vec *bvec;
7507 bio_for_each_segment_all(bvec, bio, i)
7508 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7510 complete(&done->done);
7514 static int __btrfs_correct_data_nocsum(struct inode *inode,
7515 struct btrfs_io_bio *io_bio)
7517 struct bio_vec *bvec;
7518 struct btrfs_retry_complete done;
7523 start = io_bio->logical;
7526 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7530 init_completion(&done.done);
7532 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7533 start + bvec->bv_len - 1,
7535 btrfs_retry_endio_nocsum, &done);
7539 wait_for_completion(&done.done);
7541 if (!done.uptodate) {
7542 /* We might have another mirror, so try again */
7546 start += bvec->bv_len;
7552 static void btrfs_retry_endio(struct bio *bio, int err)
7554 struct btrfs_retry_complete *done = bio->bi_private;
7555 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7556 struct bio_vec *bvec;
7565 bio_for_each_segment_all(bvec, bio, i) {
7566 ret = __readpage_endio_check(done->inode, io_bio, i,
7568 done->start, bvec->bv_len);
7570 clean_io_failure(done->inode, done->start,
7576 done->uptodate = uptodate;
7578 complete(&done->done);
7582 static int __btrfs_subio_endio_read(struct inode *inode,
7583 struct btrfs_io_bio *io_bio, int err)
7585 struct bio_vec *bvec;
7586 struct btrfs_retry_complete done;
7593 start = io_bio->logical;
7596 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7597 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7598 0, start, bvec->bv_len);
7604 init_completion(&done.done);
7606 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7607 start + bvec->bv_len - 1,
7609 btrfs_retry_endio, &done);
7615 wait_for_completion(&done.done);
7617 if (!done.uptodate) {
7618 /* We might have another mirror, so try again */
7622 offset += bvec->bv_len;
7623 start += bvec->bv_len;
7629 static int btrfs_subio_endio_read(struct inode *inode,
7630 struct btrfs_io_bio *io_bio, int err)
7632 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7636 return __btrfs_correct_data_nocsum(inode, io_bio);
7640 return __btrfs_subio_endio_read(inode, io_bio, err);
7644 static void btrfs_endio_direct_read(struct bio *bio, int err)
7646 struct btrfs_dio_private *dip = bio->bi_private;
7647 struct inode *inode = dip->inode;
7648 struct bio *dio_bio;
7649 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7651 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
7652 err = btrfs_subio_endio_read(inode, io_bio, err);
7654 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
7655 dip->logical_offset + dip->bytes - 1);
7656 dio_bio = dip->dio_bio;
7660 /* If we had a csum failure make sure to clear the uptodate flag */
7662 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7663 dio_end_io(dio_bio, err);
7666 io_bio->end_io(io_bio, err);
7670 static void btrfs_endio_direct_write(struct bio *bio, int err)
7672 struct btrfs_dio_private *dip = bio->bi_private;
7673 struct inode *inode = dip->inode;
7674 struct btrfs_root *root = BTRFS_I(inode)->root;
7675 struct btrfs_ordered_extent *ordered = NULL;
7676 u64 ordered_offset = dip->logical_offset;
7677 u64 ordered_bytes = dip->bytes;
7678 struct bio *dio_bio;
7684 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7686 ordered_bytes, !err);
7690 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
7691 finish_ordered_fn, NULL, NULL);
7692 btrfs_queue_work(root->fs_info->endio_write_workers,
7696 * our bio might span multiple ordered extents. If we haven't
7697 * completed the accounting for the whole dio, go back and try again
7699 if (ordered_offset < dip->logical_offset + dip->bytes) {
7700 ordered_bytes = dip->logical_offset + dip->bytes -
7706 dio_bio = dip->dio_bio;
7710 /* If we had an error make sure to clear the uptodate flag */
7712 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7713 dio_end_io(dio_bio, err);
7717 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7718 struct bio *bio, int mirror_num,
7719 unsigned long bio_flags, u64 offset)
7722 struct btrfs_root *root = BTRFS_I(inode)->root;
7723 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7724 BUG_ON(ret); /* -ENOMEM */
7728 static void btrfs_end_dio_bio(struct bio *bio, int err)
7730 struct btrfs_dio_private *dip = bio->bi_private;
7733 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7734 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7735 btrfs_ino(dip->inode), bio->bi_rw,
7736 (unsigned long long)bio->bi_iter.bi_sector,
7737 bio->bi_iter.bi_size, err);
7739 if (dip->subio_endio)
7740 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
7746 * before atomic variable goto zero, we must make sure
7747 * dip->errors is perceived to be set.
7749 smp_mb__before_atomic();
7752 /* if there are more bios still pending for this dio, just exit */
7753 if (!atomic_dec_and_test(&dip->pending_bios))
7757 bio_io_error(dip->orig_bio);
7759 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7760 bio_endio(dip->orig_bio, 0);
7766 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7767 u64 first_sector, gfp_t gfp_flags)
7769 int nr_vecs = bio_get_nr_vecs(bdev);
7770 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7773 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
7774 struct inode *inode,
7775 struct btrfs_dio_private *dip,
7779 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7780 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
7784 * We load all the csum data we need when we submit
7785 * the first bio to reduce the csum tree search and
7788 if (dip->logical_offset == file_offset) {
7789 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
7795 if (bio == dip->orig_bio)
7798 file_offset -= dip->logical_offset;
7799 file_offset >>= inode->i_sb->s_blocksize_bits;
7800 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
7805 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7806 int rw, u64 file_offset, int skip_sum,
7809 struct btrfs_dio_private *dip = bio->bi_private;
7810 int write = rw & REQ_WRITE;
7811 struct btrfs_root *root = BTRFS_I(inode)->root;
7815 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7820 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7821 BTRFS_WQ_ENDIO_DATA);
7829 if (write && async_submit) {
7830 ret = btrfs_wq_submit_bio(root->fs_info,
7831 inode, rw, bio, 0, 0,
7833 __btrfs_submit_bio_start_direct_io,
7834 __btrfs_submit_bio_done);
7838 * If we aren't doing async submit, calculate the csum of the
7841 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7845 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
7851 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7857 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7860 struct inode *inode = dip->inode;
7861 struct btrfs_root *root = BTRFS_I(inode)->root;
7863 struct bio *orig_bio = dip->orig_bio;
7864 struct bio_vec *bvec = orig_bio->bi_io_vec;
7865 u64 start_sector = orig_bio->bi_iter.bi_sector;
7866 u64 file_offset = dip->logical_offset;
7871 int async_submit = 0;
7873 map_length = orig_bio->bi_iter.bi_size;
7874 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7875 &map_length, NULL, 0);
7879 if (map_length >= orig_bio->bi_iter.bi_size) {
7881 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
7885 /* async crcs make it difficult to collect full stripe writes. */
7886 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
7891 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7895 bio->bi_private = dip;
7896 bio->bi_end_io = btrfs_end_dio_bio;
7897 btrfs_io_bio(bio)->logical = file_offset;
7898 atomic_inc(&dip->pending_bios);
7900 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7901 if (map_length < submit_len + bvec->bv_len ||
7902 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7903 bvec->bv_offset) < bvec->bv_len) {
7905 * inc the count before we submit the bio so
7906 * we know the end IO handler won't happen before
7907 * we inc the count. Otherwise, the dip might get freed
7908 * before we're done setting it up
7910 atomic_inc(&dip->pending_bios);
7911 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7912 file_offset, skip_sum,
7916 atomic_dec(&dip->pending_bios);
7920 start_sector += submit_len >> 9;
7921 file_offset += submit_len;
7926 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7927 start_sector, GFP_NOFS);
7930 bio->bi_private = dip;
7931 bio->bi_end_io = btrfs_end_dio_bio;
7932 btrfs_io_bio(bio)->logical = file_offset;
7934 map_length = orig_bio->bi_iter.bi_size;
7935 ret = btrfs_map_block(root->fs_info, rw,
7937 &map_length, NULL, 0);
7943 submit_len += bvec->bv_len;
7950 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7959 * before atomic variable goto zero, we must
7960 * make sure dip->errors is perceived to be set.
7962 smp_mb__before_atomic();
7963 if (atomic_dec_and_test(&dip->pending_bios))
7964 bio_io_error(dip->orig_bio);
7966 /* bio_end_io() will handle error, so we needn't return it */
7970 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7971 struct inode *inode, loff_t file_offset)
7973 struct btrfs_root *root = BTRFS_I(inode)->root;
7974 struct btrfs_dio_private *dip;
7976 struct btrfs_io_bio *btrfs_bio;
7978 int write = rw & REQ_WRITE;
7981 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7983 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7989 dip = kzalloc(sizeof(*dip), GFP_NOFS);
7995 dip->private = dio_bio->bi_private;
7997 dip->logical_offset = file_offset;
7998 dip->bytes = dio_bio->bi_iter.bi_size;
7999 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8000 io_bio->bi_private = dip;
8001 dip->orig_bio = io_bio;
8002 dip->dio_bio = dio_bio;
8003 atomic_set(&dip->pending_bios, 0);
8004 btrfs_bio = btrfs_io_bio(io_bio);
8005 btrfs_bio->logical = file_offset;
8008 io_bio->bi_end_io = btrfs_endio_direct_write;
8010 io_bio->bi_end_io = btrfs_endio_direct_read;
8011 dip->subio_endio = btrfs_subio_endio_read;
8014 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8018 if (btrfs_bio->end_io)
8019 btrfs_bio->end_io(btrfs_bio, ret);
8025 * If this is a write, we need to clean up the reserved space and kill
8026 * the ordered extent.
8029 struct btrfs_ordered_extent *ordered;
8030 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
8031 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
8032 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
8033 btrfs_free_reserved_extent(root, ordered->start,
8034 ordered->disk_len, 1);
8035 btrfs_put_ordered_extent(ordered);
8036 btrfs_put_ordered_extent(ordered);
8038 bio_endio(dio_bio, ret);
8041 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
8042 const struct iov_iter *iter, loff_t offset)
8046 unsigned blocksize_mask = root->sectorsize - 1;
8047 ssize_t retval = -EINVAL;
8049 if (offset & blocksize_mask)
8052 if (iov_iter_alignment(iter) & blocksize_mask)
8055 /* If this is a write we don't need to check anymore */
8059 * Check to make sure we don't have duplicate iov_base's in this
8060 * iovec, if so return EINVAL, otherwise we'll get csum errors
8061 * when reading back.
8063 for (seg = 0; seg < iter->nr_segs; seg++) {
8064 for (i = seg + 1; i < iter->nr_segs; i++) {
8065 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8074 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
8075 struct iov_iter *iter, loff_t offset)
8077 struct file *file = iocb->ki_filp;
8078 struct inode *inode = file->f_mapping->host;
8082 bool relock = false;
8085 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iter, offset))
8088 atomic_inc(&inode->i_dio_count);
8089 smp_mb__after_atomic();
8092 * The generic stuff only does filemap_write_and_wait_range, which
8093 * isn't enough if we've written compressed pages to this area, so
8094 * we need to flush the dirty pages again to make absolutely sure
8095 * that any outstanding dirty pages are on disk.
8097 count = iov_iter_count(iter);
8098 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8099 &BTRFS_I(inode)->runtime_flags))
8100 filemap_fdatawrite_range(inode->i_mapping, offset,
8101 offset + count - 1);
8105 * If the write DIO is beyond the EOF, we need update
8106 * the isize, but it is protected by i_mutex. So we can
8107 * not unlock the i_mutex at this case.
8109 if (offset + count <= inode->i_size) {
8110 mutex_unlock(&inode->i_mutex);
8113 ret = btrfs_delalloc_reserve_space(inode, count);
8116 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8117 &BTRFS_I(inode)->runtime_flags)) {
8118 inode_dio_done(inode);
8119 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8123 ret = __blockdev_direct_IO(rw, iocb, inode,
8124 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8125 iter, offset, btrfs_get_blocks_direct, NULL,
8126 btrfs_submit_direct, flags);
8128 if (ret < 0 && ret != -EIOCBQUEUED)
8129 btrfs_delalloc_release_space(inode, count);
8130 else if (ret >= 0 && (size_t)ret < count)
8131 btrfs_delalloc_release_space(inode,
8132 count - (size_t)ret);
8136 inode_dio_done(inode);
8138 mutex_lock(&inode->i_mutex);
8143 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8145 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8146 __u64 start, __u64 len)
8150 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8154 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8157 int btrfs_readpage(struct file *file, struct page *page)
8159 struct extent_io_tree *tree;
8160 tree = &BTRFS_I(page->mapping->host)->io_tree;
8161 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8164 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8166 struct extent_io_tree *tree;
8169 if (current->flags & PF_MEMALLOC) {
8170 redirty_page_for_writepage(wbc, page);
8174 tree = &BTRFS_I(page->mapping->host)->io_tree;
8175 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8178 static int btrfs_writepages(struct address_space *mapping,
8179 struct writeback_control *wbc)
8181 struct extent_io_tree *tree;
8183 tree = &BTRFS_I(mapping->host)->io_tree;
8184 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8188 btrfs_readpages(struct file *file, struct address_space *mapping,
8189 struct list_head *pages, unsigned nr_pages)
8191 struct extent_io_tree *tree;
8192 tree = &BTRFS_I(mapping->host)->io_tree;
8193 return extent_readpages(tree, mapping, pages, nr_pages,
8196 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8198 struct extent_io_tree *tree;
8199 struct extent_map_tree *map;
8202 tree = &BTRFS_I(page->mapping->host)->io_tree;
8203 map = &BTRFS_I(page->mapping->host)->extent_tree;
8204 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8206 ClearPagePrivate(page);
8207 set_page_private(page, 0);
8208 page_cache_release(page);
8213 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8215 if (PageWriteback(page) || PageDirty(page))
8217 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8220 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8221 unsigned int length)
8223 struct inode *inode = page->mapping->host;
8224 struct extent_io_tree *tree;
8225 struct btrfs_ordered_extent *ordered;
8226 struct extent_state *cached_state = NULL;
8227 u64 page_start = page_offset(page);
8228 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
8229 int inode_evicting = inode->i_state & I_FREEING;
8232 * we have the page locked, so new writeback can't start,
8233 * and the dirty bit won't be cleared while we are here.
8235 * Wait for IO on this page so that we can safely clear
8236 * the PagePrivate2 bit and do ordered accounting
8238 wait_on_page_writeback(page);
8240 tree = &BTRFS_I(inode)->io_tree;
8242 btrfs_releasepage(page, GFP_NOFS);
8246 if (!inode_evicting)
8247 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
8248 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8251 * IO on this page will never be started, so we need
8252 * to account for any ordered extents now
8254 if (!inode_evicting)
8255 clear_extent_bit(tree, page_start, page_end,
8256 EXTENT_DIRTY | EXTENT_DELALLOC |
8257 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8258 EXTENT_DEFRAG, 1, 0, &cached_state,
8261 * whoever cleared the private bit is responsible
8262 * for the finish_ordered_io
8264 if (TestClearPagePrivate2(page)) {
8265 struct btrfs_ordered_inode_tree *tree;
8268 tree = &BTRFS_I(inode)->ordered_tree;
8270 spin_lock_irq(&tree->lock);
8271 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8272 new_len = page_start - ordered->file_offset;
8273 if (new_len < ordered->truncated_len)
8274 ordered->truncated_len = new_len;
8275 spin_unlock_irq(&tree->lock);
8277 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8279 PAGE_CACHE_SIZE, 1))
8280 btrfs_finish_ordered_io(ordered);
8282 btrfs_put_ordered_extent(ordered);
8283 if (!inode_evicting) {
8284 cached_state = NULL;
8285 lock_extent_bits(tree, page_start, page_end, 0,
8290 if (!inode_evicting) {
8291 clear_extent_bit(tree, page_start, page_end,
8292 EXTENT_LOCKED | EXTENT_DIRTY |
8293 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8294 EXTENT_DEFRAG, 1, 1,
8295 &cached_state, GFP_NOFS);
8297 __btrfs_releasepage(page, GFP_NOFS);
8300 ClearPageChecked(page);
8301 if (PagePrivate(page)) {
8302 ClearPagePrivate(page);
8303 set_page_private(page, 0);
8304 page_cache_release(page);
8309 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8310 * called from a page fault handler when a page is first dirtied. Hence we must
8311 * be careful to check for EOF conditions here. We set the page up correctly
8312 * for a written page which means we get ENOSPC checking when writing into
8313 * holes and correct delalloc and unwritten extent mapping on filesystems that
8314 * support these features.
8316 * We are not allowed to take the i_mutex here so we have to play games to
8317 * protect against truncate races as the page could now be beyond EOF. Because
8318 * vmtruncate() writes the inode size before removing pages, once we have the
8319 * page lock we can determine safely if the page is beyond EOF. If it is not
8320 * beyond EOF, then the page is guaranteed safe against truncation until we
8323 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8325 struct page *page = vmf->page;
8326 struct inode *inode = file_inode(vma->vm_file);
8327 struct btrfs_root *root = BTRFS_I(inode)->root;
8328 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8329 struct btrfs_ordered_extent *ordered;
8330 struct extent_state *cached_state = NULL;
8332 unsigned long zero_start;
8339 sb_start_pagefault(inode->i_sb);
8340 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
8342 ret = file_update_time(vma->vm_file);
8348 else /* -ENOSPC, -EIO, etc */
8349 ret = VM_FAULT_SIGBUS;
8355 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8358 size = i_size_read(inode);
8359 page_start = page_offset(page);
8360 page_end = page_start + PAGE_CACHE_SIZE - 1;
8362 if ((page->mapping != inode->i_mapping) ||
8363 (page_start >= size)) {
8364 /* page got truncated out from underneath us */
8367 wait_on_page_writeback(page);
8369 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
8370 set_page_extent_mapped(page);
8373 * we can't set the delalloc bits if there are pending ordered
8374 * extents. Drop our locks and wait for them to finish
8376 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8378 unlock_extent_cached(io_tree, page_start, page_end,
8379 &cached_state, GFP_NOFS);
8381 btrfs_start_ordered_extent(inode, ordered, 1);
8382 btrfs_put_ordered_extent(ordered);
8387 * XXX - page_mkwrite gets called every time the page is dirtied, even
8388 * if it was already dirty, so for space accounting reasons we need to
8389 * clear any delalloc bits for the range we are fixing to save. There
8390 * is probably a better way to do this, but for now keep consistent with
8391 * prepare_pages in the normal write path.
8393 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
8394 EXTENT_DIRTY | EXTENT_DELALLOC |
8395 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8396 0, 0, &cached_state, GFP_NOFS);
8398 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
8401 unlock_extent_cached(io_tree, page_start, page_end,
8402 &cached_state, GFP_NOFS);
8403 ret = VM_FAULT_SIGBUS;
8408 /* page is wholly or partially inside EOF */
8409 if (page_start + PAGE_CACHE_SIZE > size)
8410 zero_start = size & ~PAGE_CACHE_MASK;
8412 zero_start = PAGE_CACHE_SIZE;
8414 if (zero_start != PAGE_CACHE_SIZE) {
8416 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
8417 flush_dcache_page(page);
8420 ClearPageChecked(page);
8421 set_page_dirty(page);
8422 SetPageUptodate(page);
8424 BTRFS_I(inode)->last_trans = root->fs_info->generation;
8425 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8426 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8428 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
8432 sb_end_pagefault(inode->i_sb);
8433 return VM_FAULT_LOCKED;
8437 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
8439 sb_end_pagefault(inode->i_sb);
8443 static int btrfs_truncate(struct inode *inode)
8445 struct btrfs_root *root = BTRFS_I(inode)->root;
8446 struct btrfs_block_rsv *rsv;
8449 struct btrfs_trans_handle *trans;
8450 u64 mask = root->sectorsize - 1;
8451 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
8453 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8459 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8460 * 3 things going on here
8462 * 1) We need to reserve space for our orphan item and the space to
8463 * delete our orphan item. Lord knows we don't want to have a dangling
8464 * orphan item because we didn't reserve space to remove it.
8466 * 2) We need to reserve space to update our inode.
8468 * 3) We need to have something to cache all the space that is going to
8469 * be free'd up by the truncate operation, but also have some slack
8470 * space reserved in case it uses space during the truncate (thank you
8471 * very much snapshotting).
8473 * And we need these to all be seperate. The fact is we can use alot of
8474 * space doing the truncate, and we have no earthly idea how much space
8475 * we will use, so we need the truncate reservation to be seperate so it
8476 * doesn't end up using space reserved for updating the inode or
8477 * removing the orphan item. We also need to be able to stop the
8478 * transaction and start a new one, which means we need to be able to
8479 * update the inode several times, and we have no idea of knowing how
8480 * many times that will be, so we can't just reserve 1 item for the
8481 * entirety of the opration, so that has to be done seperately as well.
8482 * Then there is the orphan item, which does indeed need to be held on
8483 * to for the whole operation, and we need nobody to touch this reserved
8484 * space except the orphan code.
8486 * So that leaves us with
8488 * 1) root->orphan_block_rsv - for the orphan deletion.
8489 * 2) rsv - for the truncate reservation, which we will steal from the
8490 * transaction reservation.
8491 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8492 * updating the inode.
8494 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
8497 rsv->size = min_size;
8501 * 1 for the truncate slack space
8502 * 1 for updating the inode.
8504 trans = btrfs_start_transaction(root, 2);
8505 if (IS_ERR(trans)) {
8506 err = PTR_ERR(trans);
8510 /* Migrate the slack space for the truncate to our reserve */
8511 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
8516 * So if we truncate and then write and fsync we normally would just
8517 * write the extents that changed, which is a problem if we need to
8518 * first truncate that entire inode. So set this flag so we write out
8519 * all of the extents in the inode to the sync log so we're completely
8522 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8523 trans->block_rsv = rsv;
8526 ret = btrfs_truncate_inode_items(trans, root, inode,
8528 BTRFS_EXTENT_DATA_KEY);
8529 if (ret != -ENOSPC) {
8534 trans->block_rsv = &root->fs_info->trans_block_rsv;
8535 ret = btrfs_update_inode(trans, root, inode);
8541 btrfs_end_transaction(trans, root);
8542 btrfs_btree_balance_dirty(root);
8544 trans = btrfs_start_transaction(root, 2);
8545 if (IS_ERR(trans)) {
8546 ret = err = PTR_ERR(trans);
8551 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
8553 BUG_ON(ret); /* shouldn't happen */
8554 trans->block_rsv = rsv;
8557 if (ret == 0 && inode->i_nlink > 0) {
8558 trans->block_rsv = root->orphan_block_rsv;
8559 ret = btrfs_orphan_del(trans, inode);
8565 trans->block_rsv = &root->fs_info->trans_block_rsv;
8566 ret = btrfs_update_inode(trans, root, inode);
8570 ret = btrfs_end_transaction(trans, root);
8571 btrfs_btree_balance_dirty(root);
8575 btrfs_free_block_rsv(root, rsv);
8584 * create a new subvolume directory/inode (helper for the ioctl).
8586 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8587 struct btrfs_root *new_root,
8588 struct btrfs_root *parent_root,
8591 struct inode *inode;
8595 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8596 new_dirid, new_dirid,
8597 S_IFDIR | (~current_umask() & S_IRWXUGO),
8600 return PTR_ERR(inode);
8601 inode->i_op = &btrfs_dir_inode_operations;
8602 inode->i_fop = &btrfs_dir_file_operations;
8604 set_nlink(inode, 1);
8605 btrfs_i_size_write(inode, 0);
8606 unlock_new_inode(inode);
8608 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8610 btrfs_err(new_root->fs_info,
8611 "error inheriting subvolume %llu properties: %d",
8612 new_root->root_key.objectid, err);
8614 err = btrfs_update_inode(trans, new_root, inode);
8620 struct inode *btrfs_alloc_inode(struct super_block *sb)
8622 struct btrfs_inode *ei;
8623 struct inode *inode;
8625 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
8632 ei->last_sub_trans = 0;
8633 ei->logged_trans = 0;
8634 ei->delalloc_bytes = 0;
8635 ei->defrag_bytes = 0;
8636 ei->disk_i_size = 0;
8639 ei->index_cnt = (u64)-1;
8641 ei->last_unlink_trans = 0;
8642 ei->last_log_commit = 0;
8644 spin_lock_init(&ei->lock);
8645 ei->outstanding_extents = 0;
8646 ei->reserved_extents = 0;
8648 ei->runtime_flags = 0;
8649 ei->force_compress = BTRFS_COMPRESS_NONE;
8651 ei->delayed_node = NULL;
8653 ei->i_otime.tv_sec = 0;
8654 ei->i_otime.tv_nsec = 0;
8656 inode = &ei->vfs_inode;
8657 extent_map_tree_init(&ei->extent_tree);
8658 extent_io_tree_init(&ei->io_tree, &inode->i_data);
8659 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
8660 ei->io_tree.track_uptodate = 1;
8661 ei->io_failure_tree.track_uptodate = 1;
8662 atomic_set(&ei->sync_writers, 0);
8663 mutex_init(&ei->log_mutex);
8664 mutex_init(&ei->delalloc_mutex);
8665 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8666 INIT_LIST_HEAD(&ei->delalloc_inodes);
8667 RB_CLEAR_NODE(&ei->rb_node);
8672 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8673 void btrfs_test_destroy_inode(struct inode *inode)
8675 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8676 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8680 static void btrfs_i_callback(struct rcu_head *head)
8682 struct inode *inode = container_of(head, struct inode, i_rcu);
8683 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8686 void btrfs_destroy_inode(struct inode *inode)
8688 struct btrfs_ordered_extent *ordered;
8689 struct btrfs_root *root = BTRFS_I(inode)->root;
8691 WARN_ON(!hlist_empty(&inode->i_dentry));
8692 WARN_ON(inode->i_data.nrpages);
8693 WARN_ON(BTRFS_I(inode)->outstanding_extents);
8694 WARN_ON(BTRFS_I(inode)->reserved_extents);
8695 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8696 WARN_ON(BTRFS_I(inode)->csum_bytes);
8697 WARN_ON(BTRFS_I(inode)->defrag_bytes);
8700 * This can happen where we create an inode, but somebody else also
8701 * created the same inode and we need to destroy the one we already
8707 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8708 &BTRFS_I(inode)->runtime_flags)) {
8709 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
8711 atomic_dec(&root->orphan_inodes);
8715 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8719 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
8720 ordered->file_offset, ordered->len);
8721 btrfs_remove_ordered_extent(inode, ordered);
8722 btrfs_put_ordered_extent(ordered);
8723 btrfs_put_ordered_extent(ordered);
8726 inode_tree_del(inode);
8727 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8729 call_rcu(&inode->i_rcu, btrfs_i_callback);
8732 int btrfs_drop_inode(struct inode *inode)
8734 struct btrfs_root *root = BTRFS_I(inode)->root;
8739 /* the snap/subvol tree is on deleting */
8740 if (btrfs_root_refs(&root->root_item) == 0)
8743 return generic_drop_inode(inode);
8746 static void init_once(void *foo)
8748 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8750 inode_init_once(&ei->vfs_inode);
8753 void btrfs_destroy_cachep(void)
8756 * Make sure all delayed rcu free inodes are flushed before we
8760 if (btrfs_inode_cachep)
8761 kmem_cache_destroy(btrfs_inode_cachep);
8762 if (btrfs_trans_handle_cachep)
8763 kmem_cache_destroy(btrfs_trans_handle_cachep);
8764 if (btrfs_transaction_cachep)
8765 kmem_cache_destroy(btrfs_transaction_cachep);
8766 if (btrfs_path_cachep)
8767 kmem_cache_destroy(btrfs_path_cachep);
8768 if (btrfs_free_space_cachep)
8769 kmem_cache_destroy(btrfs_free_space_cachep);
8770 if (btrfs_delalloc_work_cachep)
8771 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8774 int btrfs_init_cachep(void)
8776 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8777 sizeof(struct btrfs_inode), 0,
8778 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8779 if (!btrfs_inode_cachep)
8782 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8783 sizeof(struct btrfs_trans_handle), 0,
8784 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8785 if (!btrfs_trans_handle_cachep)
8788 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8789 sizeof(struct btrfs_transaction), 0,
8790 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8791 if (!btrfs_transaction_cachep)
8794 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8795 sizeof(struct btrfs_path), 0,
8796 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8797 if (!btrfs_path_cachep)
8800 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8801 sizeof(struct btrfs_free_space), 0,
8802 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8803 if (!btrfs_free_space_cachep)
8806 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8807 sizeof(struct btrfs_delalloc_work), 0,
8808 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8810 if (!btrfs_delalloc_work_cachep)
8815 btrfs_destroy_cachep();
8819 static int btrfs_getattr(struct vfsmount *mnt,
8820 struct dentry *dentry, struct kstat *stat)
8823 struct inode *inode = dentry->d_inode;
8824 u32 blocksize = inode->i_sb->s_blocksize;
8826 generic_fillattr(inode, stat);
8827 stat->dev = BTRFS_I(inode)->root->anon_dev;
8828 stat->blksize = PAGE_CACHE_SIZE;
8830 spin_lock(&BTRFS_I(inode)->lock);
8831 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8832 spin_unlock(&BTRFS_I(inode)->lock);
8833 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8834 ALIGN(delalloc_bytes, blocksize)) >> 9;
8838 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8839 struct inode *new_dir, struct dentry *new_dentry)
8841 struct btrfs_trans_handle *trans;
8842 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8843 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8844 struct inode *new_inode = new_dentry->d_inode;
8845 struct inode *old_inode = old_dentry->d_inode;
8846 struct timespec ctime = CURRENT_TIME;
8850 u64 old_ino = btrfs_ino(old_inode);
8852 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8855 /* we only allow rename subvolume link between subvolumes */
8856 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8859 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8860 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8863 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8864 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8868 /* check for collisions, even if the name isn't there */
8869 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
8870 new_dentry->d_name.name,
8871 new_dentry->d_name.len);
8874 if (ret == -EEXIST) {
8876 * eexist without a new_inode */
8877 if (WARN_ON(!new_inode)) {
8881 /* maybe -EOVERFLOW */
8888 * we're using rename to replace one file with another. Start IO on it
8889 * now so we don't add too much work to the end of the transaction
8891 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
8892 filemap_flush(old_inode->i_mapping);
8894 /* close the racy window with snapshot create/destroy ioctl */
8895 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8896 down_read(&root->fs_info->subvol_sem);
8898 * We want to reserve the absolute worst case amount of items. So if
8899 * both inodes are subvols and we need to unlink them then that would
8900 * require 4 item modifications, but if they are both normal inodes it
8901 * would require 5 item modifications, so we'll assume their normal
8902 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8903 * should cover the worst case number of items we'll modify.
8905 trans = btrfs_start_transaction(root, 11);
8906 if (IS_ERR(trans)) {
8907 ret = PTR_ERR(trans);
8912 btrfs_record_root_in_trans(trans, dest);
8914 ret = btrfs_set_inode_index(new_dir, &index);
8918 BTRFS_I(old_inode)->dir_index = 0ULL;
8919 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8920 /* force full log commit if subvolume involved. */
8921 btrfs_set_log_full_commit(root->fs_info, trans);
8923 ret = btrfs_insert_inode_ref(trans, dest,
8924 new_dentry->d_name.name,
8925 new_dentry->d_name.len,
8927 btrfs_ino(new_dir), index);
8931 * this is an ugly little race, but the rename is required
8932 * to make sure that if we crash, the inode is either at the
8933 * old name or the new one. pinning the log transaction lets
8934 * us make sure we don't allow a log commit to come in after
8935 * we unlink the name but before we add the new name back in.
8937 btrfs_pin_log_trans(root);
8940 inode_inc_iversion(old_dir);
8941 inode_inc_iversion(new_dir);
8942 inode_inc_iversion(old_inode);
8943 old_dir->i_ctime = old_dir->i_mtime = ctime;
8944 new_dir->i_ctime = new_dir->i_mtime = ctime;
8945 old_inode->i_ctime = ctime;
8947 if (old_dentry->d_parent != new_dentry->d_parent)
8948 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8950 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8951 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8952 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8953 old_dentry->d_name.name,
8954 old_dentry->d_name.len);
8956 ret = __btrfs_unlink_inode(trans, root, old_dir,
8957 old_dentry->d_inode,
8958 old_dentry->d_name.name,
8959 old_dentry->d_name.len);
8961 ret = btrfs_update_inode(trans, root, old_inode);
8964 btrfs_abort_transaction(trans, root, ret);
8969 inode_inc_iversion(new_inode);
8970 new_inode->i_ctime = CURRENT_TIME;
8971 if (unlikely(btrfs_ino(new_inode) ==
8972 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8973 root_objectid = BTRFS_I(new_inode)->location.objectid;
8974 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8976 new_dentry->d_name.name,
8977 new_dentry->d_name.len);
8978 BUG_ON(new_inode->i_nlink == 0);
8980 ret = btrfs_unlink_inode(trans, dest, new_dir,
8981 new_dentry->d_inode,
8982 new_dentry->d_name.name,
8983 new_dentry->d_name.len);
8985 if (!ret && new_inode->i_nlink == 0)
8986 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8988 btrfs_abort_transaction(trans, root, ret);
8993 ret = btrfs_add_link(trans, new_dir, old_inode,
8994 new_dentry->d_name.name,
8995 new_dentry->d_name.len, 0, index);
8997 btrfs_abort_transaction(trans, root, ret);
9001 if (old_inode->i_nlink == 1)
9002 BTRFS_I(old_inode)->dir_index = index;
9004 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
9005 struct dentry *parent = new_dentry->d_parent;
9006 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9007 btrfs_end_log_trans(root);
9010 btrfs_end_transaction(trans, root);
9012 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9013 up_read(&root->fs_info->subvol_sem);
9018 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9019 struct inode *new_dir, struct dentry *new_dentry,
9022 if (flags & ~RENAME_NOREPLACE)
9025 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
9028 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9030 struct btrfs_delalloc_work *delalloc_work;
9031 struct inode *inode;
9033 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9035 inode = delalloc_work->inode;
9036 if (delalloc_work->wait) {
9037 btrfs_wait_ordered_range(inode, 0, (u64)-1);
9039 filemap_flush(inode->i_mapping);
9040 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9041 &BTRFS_I(inode)->runtime_flags))
9042 filemap_flush(inode->i_mapping);
9045 if (delalloc_work->delay_iput)
9046 btrfs_add_delayed_iput(inode);
9049 complete(&delalloc_work->completion);
9052 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9053 int wait, int delay_iput)
9055 struct btrfs_delalloc_work *work;
9057 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
9061 init_completion(&work->completion);
9062 INIT_LIST_HEAD(&work->list);
9063 work->inode = inode;
9065 work->delay_iput = delay_iput;
9066 WARN_ON_ONCE(!inode);
9067 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9068 btrfs_run_delalloc_work, NULL, NULL);
9073 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9075 wait_for_completion(&work->completion);
9076 kmem_cache_free(btrfs_delalloc_work_cachep, work);
9080 * some fairly slow code that needs optimization. This walks the list
9081 * of all the inodes with pending delalloc and forces them to disk.
9083 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9086 struct btrfs_inode *binode;
9087 struct inode *inode;
9088 struct btrfs_delalloc_work *work, *next;
9089 struct list_head works;
9090 struct list_head splice;
9093 INIT_LIST_HEAD(&works);
9094 INIT_LIST_HEAD(&splice);
9096 mutex_lock(&root->delalloc_mutex);
9097 spin_lock(&root->delalloc_lock);
9098 list_splice_init(&root->delalloc_inodes, &splice);
9099 while (!list_empty(&splice)) {
9100 binode = list_entry(splice.next, struct btrfs_inode,
9103 list_move_tail(&binode->delalloc_inodes,
9104 &root->delalloc_inodes);
9105 inode = igrab(&binode->vfs_inode);
9107 cond_resched_lock(&root->delalloc_lock);
9110 spin_unlock(&root->delalloc_lock);
9112 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
9115 btrfs_add_delayed_iput(inode);
9121 list_add_tail(&work->list, &works);
9122 btrfs_queue_work(root->fs_info->flush_workers,
9125 if (nr != -1 && ret >= nr)
9128 spin_lock(&root->delalloc_lock);
9130 spin_unlock(&root->delalloc_lock);
9133 list_for_each_entry_safe(work, next, &works, list) {
9134 list_del_init(&work->list);
9135 btrfs_wait_and_free_delalloc_work(work);
9138 if (!list_empty_careful(&splice)) {
9139 spin_lock(&root->delalloc_lock);
9140 list_splice_tail(&splice, &root->delalloc_inodes);
9141 spin_unlock(&root->delalloc_lock);
9143 mutex_unlock(&root->delalloc_mutex);
9147 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9151 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
9154 ret = __start_delalloc_inodes(root, delay_iput, -1);
9158 * the filemap_flush will queue IO into the worker threads, but
9159 * we have to make sure the IO is actually started and that
9160 * ordered extents get created before we return
9162 atomic_inc(&root->fs_info->async_submit_draining);
9163 while (atomic_read(&root->fs_info->nr_async_submits) ||
9164 atomic_read(&root->fs_info->async_delalloc_pages)) {
9165 wait_event(root->fs_info->async_submit_wait,
9166 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9167 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
9169 atomic_dec(&root->fs_info->async_submit_draining);
9173 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9176 struct btrfs_root *root;
9177 struct list_head splice;
9180 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9183 INIT_LIST_HEAD(&splice);
9185 mutex_lock(&fs_info->delalloc_root_mutex);
9186 spin_lock(&fs_info->delalloc_root_lock);
9187 list_splice_init(&fs_info->delalloc_roots, &splice);
9188 while (!list_empty(&splice) && nr) {
9189 root = list_first_entry(&splice, struct btrfs_root,
9191 root = btrfs_grab_fs_root(root);
9193 list_move_tail(&root->delalloc_root,
9194 &fs_info->delalloc_roots);
9195 spin_unlock(&fs_info->delalloc_root_lock);
9197 ret = __start_delalloc_inodes(root, delay_iput, nr);
9198 btrfs_put_fs_root(root);
9206 spin_lock(&fs_info->delalloc_root_lock);
9208 spin_unlock(&fs_info->delalloc_root_lock);
9211 atomic_inc(&fs_info->async_submit_draining);
9212 while (atomic_read(&fs_info->nr_async_submits) ||
9213 atomic_read(&fs_info->async_delalloc_pages)) {
9214 wait_event(fs_info->async_submit_wait,
9215 (atomic_read(&fs_info->nr_async_submits) == 0 &&
9216 atomic_read(&fs_info->async_delalloc_pages) == 0));
9218 atomic_dec(&fs_info->async_submit_draining);
9220 if (!list_empty_careful(&splice)) {
9221 spin_lock(&fs_info->delalloc_root_lock);
9222 list_splice_tail(&splice, &fs_info->delalloc_roots);
9223 spin_unlock(&fs_info->delalloc_root_lock);
9225 mutex_unlock(&fs_info->delalloc_root_mutex);
9229 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9230 const char *symname)
9232 struct btrfs_trans_handle *trans;
9233 struct btrfs_root *root = BTRFS_I(dir)->root;
9234 struct btrfs_path *path;
9235 struct btrfs_key key;
9236 struct inode *inode = NULL;
9244 struct btrfs_file_extent_item *ei;
9245 struct extent_buffer *leaf;
9247 name_len = strlen(symname);
9248 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9249 return -ENAMETOOLONG;
9252 * 2 items for inode item and ref
9253 * 2 items for dir items
9254 * 1 item for xattr if selinux is on
9256 trans = btrfs_start_transaction(root, 5);
9258 return PTR_ERR(trans);
9260 err = btrfs_find_free_ino(root, &objectid);
9264 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9265 dentry->d_name.len, btrfs_ino(dir), objectid,
9266 S_IFLNK|S_IRWXUGO, &index);
9267 if (IS_ERR(inode)) {
9268 err = PTR_ERR(inode);
9273 * If the active LSM wants to access the inode during
9274 * d_instantiate it needs these. Smack checks to see
9275 * if the filesystem supports xattrs by looking at the
9278 inode->i_fop = &btrfs_file_operations;
9279 inode->i_op = &btrfs_file_inode_operations;
9280 inode->i_mapping->a_ops = &btrfs_aops;
9281 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9283 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9285 goto out_unlock_inode;
9287 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
9289 goto out_unlock_inode;
9291 path = btrfs_alloc_path();
9294 goto out_unlock_inode;
9296 key.objectid = btrfs_ino(inode);
9298 key.type = BTRFS_EXTENT_DATA_KEY;
9299 datasize = btrfs_file_extent_calc_inline_size(name_len);
9300 err = btrfs_insert_empty_item(trans, root, path, &key,
9303 btrfs_free_path(path);
9304 goto out_unlock_inode;
9306 leaf = path->nodes[0];
9307 ei = btrfs_item_ptr(leaf, path->slots[0],
9308 struct btrfs_file_extent_item);
9309 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9310 btrfs_set_file_extent_type(leaf, ei,
9311 BTRFS_FILE_EXTENT_INLINE);
9312 btrfs_set_file_extent_encryption(leaf, ei, 0);
9313 btrfs_set_file_extent_compression(leaf, ei, 0);
9314 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9315 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9317 ptr = btrfs_file_extent_inline_start(ei);
9318 write_extent_buffer(leaf, symname, ptr, name_len);
9319 btrfs_mark_buffer_dirty(leaf);
9320 btrfs_free_path(path);
9322 inode->i_op = &btrfs_symlink_inode_operations;
9323 inode->i_mapping->a_ops = &btrfs_symlink_aops;
9324 inode_set_bytes(inode, name_len);
9325 btrfs_i_size_write(inode, name_len);
9326 err = btrfs_update_inode(trans, root, inode);
9329 goto out_unlock_inode;
9332 unlock_new_inode(inode);
9333 d_instantiate(dentry, inode);
9336 btrfs_end_transaction(trans, root);
9338 inode_dec_link_count(inode);
9341 btrfs_btree_balance_dirty(root);
9346 unlock_new_inode(inode);
9350 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9351 u64 start, u64 num_bytes, u64 min_size,
9352 loff_t actual_len, u64 *alloc_hint,
9353 struct btrfs_trans_handle *trans)
9355 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9356 struct extent_map *em;
9357 struct btrfs_root *root = BTRFS_I(inode)->root;
9358 struct btrfs_key ins;
9359 u64 cur_offset = start;
9363 bool own_trans = true;
9367 while (num_bytes > 0) {
9369 trans = btrfs_start_transaction(root, 3);
9370 if (IS_ERR(trans)) {
9371 ret = PTR_ERR(trans);
9376 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
9377 cur_bytes = max(cur_bytes, min_size);
9378 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
9379 *alloc_hint, &ins, 1, 0);
9382 btrfs_end_transaction(trans, root);
9386 ret = insert_reserved_file_extent(trans, inode,
9387 cur_offset, ins.objectid,
9388 ins.offset, ins.offset,
9389 ins.offset, 0, 0, 0,
9390 BTRFS_FILE_EXTENT_PREALLOC);
9392 btrfs_free_reserved_extent(root, ins.objectid,
9394 btrfs_abort_transaction(trans, root, ret);
9396 btrfs_end_transaction(trans, root);
9399 btrfs_drop_extent_cache(inode, cur_offset,
9400 cur_offset + ins.offset -1, 0);
9402 em = alloc_extent_map();
9404 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9405 &BTRFS_I(inode)->runtime_flags);
9409 em->start = cur_offset;
9410 em->orig_start = cur_offset;
9411 em->len = ins.offset;
9412 em->block_start = ins.objectid;
9413 em->block_len = ins.offset;
9414 em->orig_block_len = ins.offset;
9415 em->ram_bytes = ins.offset;
9416 em->bdev = root->fs_info->fs_devices->latest_bdev;
9417 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9418 em->generation = trans->transid;
9421 write_lock(&em_tree->lock);
9422 ret = add_extent_mapping(em_tree, em, 1);
9423 write_unlock(&em_tree->lock);
9426 btrfs_drop_extent_cache(inode, cur_offset,
9427 cur_offset + ins.offset - 1,
9430 free_extent_map(em);
9432 num_bytes -= ins.offset;
9433 cur_offset += ins.offset;
9434 *alloc_hint = ins.objectid + ins.offset;
9436 inode_inc_iversion(inode);
9437 inode->i_ctime = CURRENT_TIME;
9438 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9439 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9440 (actual_len > inode->i_size) &&
9441 (cur_offset > inode->i_size)) {
9442 if (cur_offset > actual_len)
9443 i_size = actual_len;
9445 i_size = cur_offset;
9446 i_size_write(inode, i_size);
9447 btrfs_ordered_update_i_size(inode, i_size, NULL);
9450 ret = btrfs_update_inode(trans, root, inode);
9453 btrfs_abort_transaction(trans, root, ret);
9455 btrfs_end_transaction(trans, root);
9460 btrfs_end_transaction(trans, root);
9465 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9466 u64 start, u64 num_bytes, u64 min_size,
9467 loff_t actual_len, u64 *alloc_hint)
9469 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9470 min_size, actual_len, alloc_hint,
9474 int btrfs_prealloc_file_range_trans(struct inode *inode,
9475 struct btrfs_trans_handle *trans, int mode,
9476 u64 start, u64 num_bytes, u64 min_size,
9477 loff_t actual_len, u64 *alloc_hint)
9479 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9480 min_size, actual_len, alloc_hint, trans);
9483 static int btrfs_set_page_dirty(struct page *page)
9485 return __set_page_dirty_nobuffers(page);
9488 static int btrfs_permission(struct inode *inode, int mask)
9490 struct btrfs_root *root = BTRFS_I(inode)->root;
9491 umode_t mode = inode->i_mode;
9493 if (mask & MAY_WRITE &&
9494 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9495 if (btrfs_root_readonly(root))
9497 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9500 return generic_permission(inode, mask);
9503 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9505 struct btrfs_trans_handle *trans;
9506 struct btrfs_root *root = BTRFS_I(dir)->root;
9507 struct inode *inode = NULL;
9513 * 5 units required for adding orphan entry
9515 trans = btrfs_start_transaction(root, 5);
9517 return PTR_ERR(trans);
9519 ret = btrfs_find_free_ino(root, &objectid);
9523 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9524 btrfs_ino(dir), objectid, mode, &index);
9525 if (IS_ERR(inode)) {
9526 ret = PTR_ERR(inode);
9531 inode->i_fop = &btrfs_file_operations;
9532 inode->i_op = &btrfs_file_inode_operations;
9534 inode->i_mapping->a_ops = &btrfs_aops;
9535 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9537 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9541 ret = btrfs_update_inode(trans, root, inode);
9544 ret = btrfs_orphan_add(trans, inode);
9549 * We set number of links to 0 in btrfs_new_inode(), and here we set
9550 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9553 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9555 set_nlink(inode, 1);
9556 unlock_new_inode(inode);
9557 d_tmpfile(dentry, inode);
9558 mark_inode_dirty(inode);
9561 btrfs_end_transaction(trans, root);
9564 btrfs_balance_delayed_items(root);
9565 btrfs_btree_balance_dirty(root);
9569 unlock_new_inode(inode);
9574 /* Inspired by filemap_check_errors() */
9575 int btrfs_inode_check_errors(struct inode *inode)
9579 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
9580 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
9582 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
9583 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
9589 static const struct inode_operations btrfs_dir_inode_operations = {
9590 .getattr = btrfs_getattr,
9591 .lookup = btrfs_lookup,
9592 .create = btrfs_create,
9593 .unlink = btrfs_unlink,
9595 .mkdir = btrfs_mkdir,
9596 .rmdir = btrfs_rmdir,
9597 .rename2 = btrfs_rename2,
9598 .symlink = btrfs_symlink,
9599 .setattr = btrfs_setattr,
9600 .mknod = btrfs_mknod,
9601 .setxattr = btrfs_setxattr,
9602 .getxattr = btrfs_getxattr,
9603 .listxattr = btrfs_listxattr,
9604 .removexattr = btrfs_removexattr,
9605 .permission = btrfs_permission,
9606 .get_acl = btrfs_get_acl,
9607 .set_acl = btrfs_set_acl,
9608 .update_time = btrfs_update_time,
9609 .tmpfile = btrfs_tmpfile,
9611 static const struct inode_operations btrfs_dir_ro_inode_operations = {
9612 .lookup = btrfs_lookup,
9613 .permission = btrfs_permission,
9614 .get_acl = btrfs_get_acl,
9615 .set_acl = btrfs_set_acl,
9616 .update_time = btrfs_update_time,
9619 static const struct file_operations btrfs_dir_file_operations = {
9620 .llseek = generic_file_llseek,
9621 .read = generic_read_dir,
9622 .iterate = btrfs_real_readdir,
9623 .unlocked_ioctl = btrfs_ioctl,
9624 #ifdef CONFIG_COMPAT
9625 .compat_ioctl = btrfs_ioctl,
9627 .release = btrfs_release_file,
9628 .fsync = btrfs_sync_file,
9631 static struct extent_io_ops btrfs_extent_io_ops = {
9632 .fill_delalloc = run_delalloc_range,
9633 .submit_bio_hook = btrfs_submit_bio_hook,
9634 .merge_bio_hook = btrfs_merge_bio_hook,
9635 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
9636 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
9637 .writepage_start_hook = btrfs_writepage_start_hook,
9638 .set_bit_hook = btrfs_set_bit_hook,
9639 .clear_bit_hook = btrfs_clear_bit_hook,
9640 .merge_extent_hook = btrfs_merge_extent_hook,
9641 .split_extent_hook = btrfs_split_extent_hook,
9645 * btrfs doesn't support the bmap operation because swapfiles
9646 * use bmap to make a mapping of extents in the file. They assume
9647 * these extents won't change over the life of the file and they
9648 * use the bmap result to do IO directly to the drive.
9650 * the btrfs bmap call would return logical addresses that aren't
9651 * suitable for IO and they also will change frequently as COW
9652 * operations happen. So, swapfile + btrfs == corruption.
9654 * For now we're avoiding this by dropping bmap.
9656 static const struct address_space_operations btrfs_aops = {
9657 .readpage = btrfs_readpage,
9658 .writepage = btrfs_writepage,
9659 .writepages = btrfs_writepages,
9660 .readpages = btrfs_readpages,
9661 .direct_IO = btrfs_direct_IO,
9662 .invalidatepage = btrfs_invalidatepage,
9663 .releasepage = btrfs_releasepage,
9664 .set_page_dirty = btrfs_set_page_dirty,
9665 .error_remove_page = generic_error_remove_page,
9668 static const struct address_space_operations btrfs_symlink_aops = {
9669 .readpage = btrfs_readpage,
9670 .writepage = btrfs_writepage,
9671 .invalidatepage = btrfs_invalidatepage,
9672 .releasepage = btrfs_releasepage,
9675 static const struct inode_operations btrfs_file_inode_operations = {
9676 .getattr = btrfs_getattr,
9677 .setattr = btrfs_setattr,
9678 .setxattr = btrfs_setxattr,
9679 .getxattr = btrfs_getxattr,
9680 .listxattr = btrfs_listxattr,
9681 .removexattr = btrfs_removexattr,
9682 .permission = btrfs_permission,
9683 .fiemap = btrfs_fiemap,
9684 .get_acl = btrfs_get_acl,
9685 .set_acl = btrfs_set_acl,
9686 .update_time = btrfs_update_time,
9688 static const struct inode_operations btrfs_special_inode_operations = {
9689 .getattr = btrfs_getattr,
9690 .setattr = btrfs_setattr,
9691 .permission = btrfs_permission,
9692 .setxattr = btrfs_setxattr,
9693 .getxattr = btrfs_getxattr,
9694 .listxattr = btrfs_listxattr,
9695 .removexattr = btrfs_removexattr,
9696 .get_acl = btrfs_get_acl,
9697 .set_acl = btrfs_set_acl,
9698 .update_time = btrfs_update_time,
9700 static const struct inode_operations btrfs_symlink_inode_operations = {
9701 .readlink = generic_readlink,
9702 .follow_link = page_follow_link_light,
9703 .put_link = page_put_link,
9704 .getattr = btrfs_getattr,
9705 .setattr = btrfs_setattr,
9706 .permission = btrfs_permission,
9707 .setxattr = btrfs_setxattr,
9708 .getxattr = btrfs_getxattr,
9709 .listxattr = btrfs_listxattr,
9710 .removexattr = btrfs_removexattr,
9711 .update_time = btrfs_update_time,
9714 const struct dentry_operations btrfs_dentry_operations = {
9715 .d_delete = btrfs_dentry_delete,
9716 .d_release = btrfs_dentry_release,