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 if (old_size < (new->end - new->start + 1))
1596 old_size = (new->end - new->start + 1);
1597 if (new->start > other->start)
1598 new_size = new->end - other->start + 1;
1600 new_size = other->end - new->start + 1;
1602 /* we're not bigger than the max, unreserve the space and go */
1603 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1604 spin_lock(&BTRFS_I(inode)->lock);
1605 BTRFS_I(inode)->outstanding_extents--;
1606 spin_unlock(&BTRFS_I(inode)->lock);
1611 * If we grew by another max_extent, just return, we want to keep that
1614 num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1615 BTRFS_MAX_EXTENT_SIZE);
1616 if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1617 BTRFS_MAX_EXTENT_SIZE) > num_extents)
1620 spin_lock(&BTRFS_I(inode)->lock);
1621 BTRFS_I(inode)->outstanding_extents--;
1622 spin_unlock(&BTRFS_I(inode)->lock);
1625 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1626 struct inode *inode)
1628 spin_lock(&root->delalloc_lock);
1629 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1630 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1631 &root->delalloc_inodes);
1632 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1633 &BTRFS_I(inode)->runtime_flags);
1634 root->nr_delalloc_inodes++;
1635 if (root->nr_delalloc_inodes == 1) {
1636 spin_lock(&root->fs_info->delalloc_root_lock);
1637 BUG_ON(!list_empty(&root->delalloc_root));
1638 list_add_tail(&root->delalloc_root,
1639 &root->fs_info->delalloc_roots);
1640 spin_unlock(&root->fs_info->delalloc_root_lock);
1643 spin_unlock(&root->delalloc_lock);
1646 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1647 struct inode *inode)
1649 spin_lock(&root->delalloc_lock);
1650 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1651 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1652 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1653 &BTRFS_I(inode)->runtime_flags);
1654 root->nr_delalloc_inodes--;
1655 if (!root->nr_delalloc_inodes) {
1656 spin_lock(&root->fs_info->delalloc_root_lock);
1657 BUG_ON(list_empty(&root->delalloc_root));
1658 list_del_init(&root->delalloc_root);
1659 spin_unlock(&root->fs_info->delalloc_root_lock);
1662 spin_unlock(&root->delalloc_lock);
1666 * extent_io.c set_bit_hook, used to track delayed allocation
1667 * bytes in this file, and to maintain the list of inodes that
1668 * have pending delalloc work to be done.
1670 static void btrfs_set_bit_hook(struct inode *inode,
1671 struct extent_state *state, unsigned *bits)
1674 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1677 * set_bit and clear bit hooks normally require _irqsave/restore
1678 * but in this case, we are only testing for the DELALLOC
1679 * bit, which is only set or cleared with irqs on
1681 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1682 struct btrfs_root *root = BTRFS_I(inode)->root;
1683 u64 len = state->end + 1 - state->start;
1684 bool do_list = !btrfs_is_free_space_inode(inode);
1686 if (*bits & EXTENT_FIRST_DELALLOC) {
1687 *bits &= ~EXTENT_FIRST_DELALLOC;
1689 spin_lock(&BTRFS_I(inode)->lock);
1690 BTRFS_I(inode)->outstanding_extents++;
1691 spin_unlock(&BTRFS_I(inode)->lock);
1694 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1695 root->fs_info->delalloc_batch);
1696 spin_lock(&BTRFS_I(inode)->lock);
1697 BTRFS_I(inode)->delalloc_bytes += len;
1698 if (*bits & EXTENT_DEFRAG)
1699 BTRFS_I(inode)->defrag_bytes += len;
1700 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1701 &BTRFS_I(inode)->runtime_flags))
1702 btrfs_add_delalloc_inodes(root, inode);
1703 spin_unlock(&BTRFS_I(inode)->lock);
1708 * extent_io.c clear_bit_hook, see set_bit_hook for why
1710 static void btrfs_clear_bit_hook(struct inode *inode,
1711 struct extent_state *state,
1714 u64 len = state->end + 1 - state->start;
1715 u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1716 BTRFS_MAX_EXTENT_SIZE);
1718 spin_lock(&BTRFS_I(inode)->lock);
1719 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1720 BTRFS_I(inode)->defrag_bytes -= len;
1721 spin_unlock(&BTRFS_I(inode)->lock);
1724 * set_bit and clear bit hooks normally require _irqsave/restore
1725 * but in this case, we are only testing for the DELALLOC
1726 * bit, which is only set or cleared with irqs on
1728 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1729 struct btrfs_root *root = BTRFS_I(inode)->root;
1730 bool do_list = !btrfs_is_free_space_inode(inode);
1732 if (*bits & EXTENT_FIRST_DELALLOC) {
1733 *bits &= ~EXTENT_FIRST_DELALLOC;
1734 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1735 spin_lock(&BTRFS_I(inode)->lock);
1736 BTRFS_I(inode)->outstanding_extents -= num_extents;
1737 spin_unlock(&BTRFS_I(inode)->lock);
1741 * We don't reserve metadata space for space cache inodes so we
1742 * don't need to call dellalloc_release_metadata if there is an
1745 if (*bits & EXTENT_DO_ACCOUNTING &&
1746 root != root->fs_info->tree_root)
1747 btrfs_delalloc_release_metadata(inode, len);
1749 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1750 && do_list && !(state->state & EXTENT_NORESERVE))
1751 btrfs_free_reserved_data_space(inode, len);
1753 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1754 root->fs_info->delalloc_batch);
1755 spin_lock(&BTRFS_I(inode)->lock);
1756 BTRFS_I(inode)->delalloc_bytes -= len;
1757 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1758 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1759 &BTRFS_I(inode)->runtime_flags))
1760 btrfs_del_delalloc_inode(root, inode);
1761 spin_unlock(&BTRFS_I(inode)->lock);
1766 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1767 * we don't create bios that span stripes or chunks
1769 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1770 size_t size, struct bio *bio,
1771 unsigned long bio_flags)
1773 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1774 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1779 if (bio_flags & EXTENT_BIO_COMPRESSED)
1782 length = bio->bi_iter.bi_size;
1783 map_length = length;
1784 ret = btrfs_map_block(root->fs_info, rw, logical,
1785 &map_length, NULL, 0);
1786 /* Will always return 0 with map_multi == NULL */
1788 if (map_length < length + size)
1794 * in order to insert checksums into the metadata in large chunks,
1795 * we wait until bio submission time. All the pages in the bio are
1796 * checksummed and sums are attached onto the ordered extent record.
1798 * At IO completion time the cums attached on the ordered extent record
1799 * are inserted into the btree
1801 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1802 struct bio *bio, int mirror_num,
1803 unsigned long bio_flags,
1806 struct btrfs_root *root = BTRFS_I(inode)->root;
1809 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1810 BUG_ON(ret); /* -ENOMEM */
1815 * in order to insert checksums into the metadata in large chunks,
1816 * we wait until bio submission time. All the pages in the bio are
1817 * checksummed and sums are attached onto the ordered extent record.
1819 * At IO completion time the cums attached on the ordered extent record
1820 * are inserted into the btree
1822 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1823 int mirror_num, unsigned long bio_flags,
1826 struct btrfs_root *root = BTRFS_I(inode)->root;
1829 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1831 bio_endio(bio, ret);
1836 * extent_io.c submission hook. This does the right thing for csum calculation
1837 * on write, or reading the csums from the tree before a read
1839 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1840 int mirror_num, unsigned long bio_flags,
1843 struct btrfs_root *root = BTRFS_I(inode)->root;
1847 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1849 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1851 if (btrfs_is_free_space_inode(inode))
1854 if (!(rw & REQ_WRITE)) {
1855 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1859 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1860 ret = btrfs_submit_compressed_read(inode, bio,
1864 } else if (!skip_sum) {
1865 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1870 } else if (async && !skip_sum) {
1871 /* csum items have already been cloned */
1872 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1874 /* we're doing a write, do the async checksumming */
1875 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1876 inode, rw, bio, mirror_num,
1877 bio_flags, bio_offset,
1878 __btrfs_submit_bio_start,
1879 __btrfs_submit_bio_done);
1881 } else if (!skip_sum) {
1882 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1888 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1892 bio_endio(bio, ret);
1897 * given a list of ordered sums record them in the inode. This happens
1898 * at IO completion time based on sums calculated at bio submission time.
1900 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1901 struct inode *inode, u64 file_offset,
1902 struct list_head *list)
1904 struct btrfs_ordered_sum *sum;
1906 list_for_each_entry(sum, list, list) {
1907 trans->adding_csums = 1;
1908 btrfs_csum_file_blocks(trans,
1909 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1910 trans->adding_csums = 0;
1915 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1916 struct extent_state **cached_state)
1918 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1919 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1920 cached_state, GFP_NOFS);
1923 /* see btrfs_writepage_start_hook for details on why this is required */
1924 struct btrfs_writepage_fixup {
1926 struct btrfs_work work;
1929 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1931 struct btrfs_writepage_fixup *fixup;
1932 struct btrfs_ordered_extent *ordered;
1933 struct extent_state *cached_state = NULL;
1935 struct inode *inode;
1940 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1944 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1945 ClearPageChecked(page);
1949 inode = page->mapping->host;
1950 page_start = page_offset(page);
1951 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1953 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1956 /* already ordered? We're done */
1957 if (PagePrivate2(page))
1960 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1962 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1963 page_end, &cached_state, GFP_NOFS);
1965 btrfs_start_ordered_extent(inode, ordered, 1);
1966 btrfs_put_ordered_extent(ordered);
1970 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1972 mapping_set_error(page->mapping, ret);
1973 end_extent_writepage(page, ret, page_start, page_end);
1974 ClearPageChecked(page);
1978 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1979 ClearPageChecked(page);
1980 set_page_dirty(page);
1982 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1983 &cached_state, GFP_NOFS);
1986 page_cache_release(page);
1991 * There are a few paths in the higher layers of the kernel that directly
1992 * set the page dirty bit without asking the filesystem if it is a
1993 * good idea. This causes problems because we want to make sure COW
1994 * properly happens and the data=ordered rules are followed.
1996 * In our case any range that doesn't have the ORDERED bit set
1997 * hasn't been properly setup for IO. We kick off an async process
1998 * to fix it up. The async helper will wait for ordered extents, set
1999 * the delalloc bit and make it safe to write the page.
2001 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2003 struct inode *inode = page->mapping->host;
2004 struct btrfs_writepage_fixup *fixup;
2005 struct btrfs_root *root = BTRFS_I(inode)->root;
2007 /* this page is properly in the ordered list */
2008 if (TestClearPagePrivate2(page))
2011 if (PageChecked(page))
2014 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2018 SetPageChecked(page);
2019 page_cache_get(page);
2020 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2021 btrfs_writepage_fixup_worker, NULL, NULL);
2023 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
2027 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2028 struct inode *inode, u64 file_pos,
2029 u64 disk_bytenr, u64 disk_num_bytes,
2030 u64 num_bytes, u64 ram_bytes,
2031 u8 compression, u8 encryption,
2032 u16 other_encoding, int extent_type)
2034 struct btrfs_root *root = BTRFS_I(inode)->root;
2035 struct btrfs_file_extent_item *fi;
2036 struct btrfs_path *path;
2037 struct extent_buffer *leaf;
2038 struct btrfs_key ins;
2039 int extent_inserted = 0;
2042 path = btrfs_alloc_path();
2047 * we may be replacing one extent in the tree with another.
2048 * The new extent is pinned in the extent map, and we don't want
2049 * to drop it from the cache until it is completely in the btree.
2051 * So, tell btrfs_drop_extents to leave this extent in the cache.
2052 * the caller is expected to unpin it and allow it to be merged
2055 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2056 file_pos + num_bytes, NULL, 0,
2057 1, sizeof(*fi), &extent_inserted);
2061 if (!extent_inserted) {
2062 ins.objectid = btrfs_ino(inode);
2063 ins.offset = file_pos;
2064 ins.type = BTRFS_EXTENT_DATA_KEY;
2066 path->leave_spinning = 1;
2067 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2072 leaf = path->nodes[0];
2073 fi = btrfs_item_ptr(leaf, path->slots[0],
2074 struct btrfs_file_extent_item);
2075 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2076 btrfs_set_file_extent_type(leaf, fi, extent_type);
2077 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2078 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2079 btrfs_set_file_extent_offset(leaf, fi, 0);
2080 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2081 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2082 btrfs_set_file_extent_compression(leaf, fi, compression);
2083 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2084 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2086 btrfs_mark_buffer_dirty(leaf);
2087 btrfs_release_path(path);
2089 inode_add_bytes(inode, num_bytes);
2091 ins.objectid = disk_bytenr;
2092 ins.offset = disk_num_bytes;
2093 ins.type = BTRFS_EXTENT_ITEM_KEY;
2094 ret = btrfs_alloc_reserved_file_extent(trans, root,
2095 root->root_key.objectid,
2096 btrfs_ino(inode), file_pos, &ins);
2098 btrfs_free_path(path);
2103 /* snapshot-aware defrag */
2104 struct sa_defrag_extent_backref {
2105 struct rb_node node;
2106 struct old_sa_defrag_extent *old;
2115 struct old_sa_defrag_extent {
2116 struct list_head list;
2117 struct new_sa_defrag_extent *new;
2126 struct new_sa_defrag_extent {
2127 struct rb_root root;
2128 struct list_head head;
2129 struct btrfs_path *path;
2130 struct inode *inode;
2138 static int backref_comp(struct sa_defrag_extent_backref *b1,
2139 struct sa_defrag_extent_backref *b2)
2141 if (b1->root_id < b2->root_id)
2143 else if (b1->root_id > b2->root_id)
2146 if (b1->inum < b2->inum)
2148 else if (b1->inum > b2->inum)
2151 if (b1->file_pos < b2->file_pos)
2153 else if (b1->file_pos > b2->file_pos)
2157 * [------------------------------] ===> (a range of space)
2158 * |<--->| |<---->| =============> (fs/file tree A)
2159 * |<---------------------------->| ===> (fs/file tree B)
2161 * A range of space can refer to two file extents in one tree while
2162 * refer to only one file extent in another tree.
2164 * So we may process a disk offset more than one time(two extents in A)
2165 * and locate at the same extent(one extent in B), then insert two same
2166 * backrefs(both refer to the extent in B).
2171 static void backref_insert(struct rb_root *root,
2172 struct sa_defrag_extent_backref *backref)
2174 struct rb_node **p = &root->rb_node;
2175 struct rb_node *parent = NULL;
2176 struct sa_defrag_extent_backref *entry;
2181 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2183 ret = backref_comp(backref, entry);
2187 p = &(*p)->rb_right;
2190 rb_link_node(&backref->node, parent, p);
2191 rb_insert_color(&backref->node, root);
2195 * Note the backref might has changed, and in this case we just return 0.
2197 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2200 struct btrfs_file_extent_item *extent;
2201 struct btrfs_fs_info *fs_info;
2202 struct old_sa_defrag_extent *old = ctx;
2203 struct new_sa_defrag_extent *new = old->new;
2204 struct btrfs_path *path = new->path;
2205 struct btrfs_key key;
2206 struct btrfs_root *root;
2207 struct sa_defrag_extent_backref *backref;
2208 struct extent_buffer *leaf;
2209 struct inode *inode = new->inode;
2215 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2216 inum == btrfs_ino(inode))
2219 key.objectid = root_id;
2220 key.type = BTRFS_ROOT_ITEM_KEY;
2221 key.offset = (u64)-1;
2223 fs_info = BTRFS_I(inode)->root->fs_info;
2224 root = btrfs_read_fs_root_no_name(fs_info, &key);
2226 if (PTR_ERR(root) == -ENOENT)
2229 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2230 inum, offset, root_id);
2231 return PTR_ERR(root);
2234 key.objectid = inum;
2235 key.type = BTRFS_EXTENT_DATA_KEY;
2236 if (offset > (u64)-1 << 32)
2239 key.offset = offset;
2241 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2242 if (WARN_ON(ret < 0))
2249 leaf = path->nodes[0];
2250 slot = path->slots[0];
2252 if (slot >= btrfs_header_nritems(leaf)) {
2253 ret = btrfs_next_leaf(root, path);
2256 } else if (ret > 0) {
2265 btrfs_item_key_to_cpu(leaf, &key, slot);
2267 if (key.objectid > inum)
2270 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2273 extent = btrfs_item_ptr(leaf, slot,
2274 struct btrfs_file_extent_item);
2276 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2280 * 'offset' refers to the exact key.offset,
2281 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2282 * (key.offset - extent_offset).
2284 if (key.offset != offset)
2287 extent_offset = btrfs_file_extent_offset(leaf, extent);
2288 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2290 if (extent_offset >= old->extent_offset + old->offset +
2291 old->len || extent_offset + num_bytes <=
2292 old->extent_offset + old->offset)
2297 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2303 backref->root_id = root_id;
2304 backref->inum = inum;
2305 backref->file_pos = offset;
2306 backref->num_bytes = num_bytes;
2307 backref->extent_offset = extent_offset;
2308 backref->generation = btrfs_file_extent_generation(leaf, extent);
2310 backref_insert(&new->root, backref);
2313 btrfs_release_path(path);
2318 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2319 struct new_sa_defrag_extent *new)
2321 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2322 struct old_sa_defrag_extent *old, *tmp;
2327 list_for_each_entry_safe(old, tmp, &new->head, list) {
2328 ret = iterate_inodes_from_logical(old->bytenr +
2329 old->extent_offset, fs_info,
2330 path, record_one_backref,
2332 if (ret < 0 && ret != -ENOENT)
2335 /* no backref to be processed for this extent */
2337 list_del(&old->list);
2342 if (list_empty(&new->head))
2348 static int relink_is_mergable(struct extent_buffer *leaf,
2349 struct btrfs_file_extent_item *fi,
2350 struct new_sa_defrag_extent *new)
2352 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2355 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2358 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2361 if (btrfs_file_extent_encryption(leaf, fi) ||
2362 btrfs_file_extent_other_encoding(leaf, fi))
2369 * Note the backref might has changed, and in this case we just return 0.
2371 static noinline int relink_extent_backref(struct btrfs_path *path,
2372 struct sa_defrag_extent_backref *prev,
2373 struct sa_defrag_extent_backref *backref)
2375 struct btrfs_file_extent_item *extent;
2376 struct btrfs_file_extent_item *item;
2377 struct btrfs_ordered_extent *ordered;
2378 struct btrfs_trans_handle *trans;
2379 struct btrfs_fs_info *fs_info;
2380 struct btrfs_root *root;
2381 struct btrfs_key key;
2382 struct extent_buffer *leaf;
2383 struct old_sa_defrag_extent *old = backref->old;
2384 struct new_sa_defrag_extent *new = old->new;
2385 struct inode *src_inode = new->inode;
2386 struct inode *inode;
2387 struct extent_state *cached = NULL;
2396 if (prev && prev->root_id == backref->root_id &&
2397 prev->inum == backref->inum &&
2398 prev->file_pos + prev->num_bytes == backref->file_pos)
2401 /* step 1: get root */
2402 key.objectid = backref->root_id;
2403 key.type = BTRFS_ROOT_ITEM_KEY;
2404 key.offset = (u64)-1;
2406 fs_info = BTRFS_I(src_inode)->root->fs_info;
2407 index = srcu_read_lock(&fs_info->subvol_srcu);
2409 root = btrfs_read_fs_root_no_name(fs_info, &key);
2411 srcu_read_unlock(&fs_info->subvol_srcu, index);
2412 if (PTR_ERR(root) == -ENOENT)
2414 return PTR_ERR(root);
2417 if (btrfs_root_readonly(root)) {
2418 srcu_read_unlock(&fs_info->subvol_srcu, index);
2422 /* step 2: get inode */
2423 key.objectid = backref->inum;
2424 key.type = BTRFS_INODE_ITEM_KEY;
2427 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2428 if (IS_ERR(inode)) {
2429 srcu_read_unlock(&fs_info->subvol_srcu, index);
2433 srcu_read_unlock(&fs_info->subvol_srcu, index);
2435 /* step 3: relink backref */
2436 lock_start = backref->file_pos;
2437 lock_end = backref->file_pos + backref->num_bytes - 1;
2438 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2441 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2443 btrfs_put_ordered_extent(ordered);
2447 trans = btrfs_join_transaction(root);
2448 if (IS_ERR(trans)) {
2449 ret = PTR_ERR(trans);
2453 key.objectid = backref->inum;
2454 key.type = BTRFS_EXTENT_DATA_KEY;
2455 key.offset = backref->file_pos;
2457 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2460 } else if (ret > 0) {
2465 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2466 struct btrfs_file_extent_item);
2468 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2469 backref->generation)
2472 btrfs_release_path(path);
2474 start = backref->file_pos;
2475 if (backref->extent_offset < old->extent_offset + old->offset)
2476 start += old->extent_offset + old->offset -
2477 backref->extent_offset;
2479 len = min(backref->extent_offset + backref->num_bytes,
2480 old->extent_offset + old->offset + old->len);
2481 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2483 ret = btrfs_drop_extents(trans, root, inode, start,
2488 key.objectid = btrfs_ino(inode);
2489 key.type = BTRFS_EXTENT_DATA_KEY;
2492 path->leave_spinning = 1;
2494 struct btrfs_file_extent_item *fi;
2496 struct btrfs_key found_key;
2498 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2503 leaf = path->nodes[0];
2504 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2506 fi = btrfs_item_ptr(leaf, path->slots[0],
2507 struct btrfs_file_extent_item);
2508 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2510 if (extent_len + found_key.offset == start &&
2511 relink_is_mergable(leaf, fi, new)) {
2512 btrfs_set_file_extent_num_bytes(leaf, fi,
2514 btrfs_mark_buffer_dirty(leaf);
2515 inode_add_bytes(inode, len);
2521 btrfs_release_path(path);
2526 ret = btrfs_insert_empty_item(trans, root, path, &key,
2529 btrfs_abort_transaction(trans, root, ret);
2533 leaf = path->nodes[0];
2534 item = btrfs_item_ptr(leaf, path->slots[0],
2535 struct btrfs_file_extent_item);
2536 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2537 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2538 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2539 btrfs_set_file_extent_num_bytes(leaf, item, len);
2540 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2541 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2542 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2543 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2544 btrfs_set_file_extent_encryption(leaf, item, 0);
2545 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2547 btrfs_mark_buffer_dirty(leaf);
2548 inode_add_bytes(inode, len);
2549 btrfs_release_path(path);
2551 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2553 backref->root_id, backref->inum,
2554 new->file_pos, 0); /* start - extent_offset */
2556 btrfs_abort_transaction(trans, root, ret);
2562 btrfs_release_path(path);
2563 path->leave_spinning = 0;
2564 btrfs_end_transaction(trans, root);
2566 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2572 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2574 struct old_sa_defrag_extent *old, *tmp;
2579 list_for_each_entry_safe(old, tmp, &new->head, list) {
2580 list_del(&old->list);
2586 static void relink_file_extents(struct new_sa_defrag_extent *new)
2588 struct btrfs_path *path;
2589 struct sa_defrag_extent_backref *backref;
2590 struct sa_defrag_extent_backref *prev = NULL;
2591 struct inode *inode;
2592 struct btrfs_root *root;
2593 struct rb_node *node;
2597 root = BTRFS_I(inode)->root;
2599 path = btrfs_alloc_path();
2603 if (!record_extent_backrefs(path, new)) {
2604 btrfs_free_path(path);
2607 btrfs_release_path(path);
2610 node = rb_first(&new->root);
2613 rb_erase(node, &new->root);
2615 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2617 ret = relink_extent_backref(path, prev, backref);
2630 btrfs_free_path(path);
2632 free_sa_defrag_extent(new);
2634 atomic_dec(&root->fs_info->defrag_running);
2635 wake_up(&root->fs_info->transaction_wait);
2638 static struct new_sa_defrag_extent *
2639 record_old_file_extents(struct inode *inode,
2640 struct btrfs_ordered_extent *ordered)
2642 struct btrfs_root *root = BTRFS_I(inode)->root;
2643 struct btrfs_path *path;
2644 struct btrfs_key key;
2645 struct old_sa_defrag_extent *old;
2646 struct new_sa_defrag_extent *new;
2649 new = kmalloc(sizeof(*new), GFP_NOFS);
2654 new->file_pos = ordered->file_offset;
2655 new->len = ordered->len;
2656 new->bytenr = ordered->start;
2657 new->disk_len = ordered->disk_len;
2658 new->compress_type = ordered->compress_type;
2659 new->root = RB_ROOT;
2660 INIT_LIST_HEAD(&new->head);
2662 path = btrfs_alloc_path();
2666 key.objectid = btrfs_ino(inode);
2667 key.type = BTRFS_EXTENT_DATA_KEY;
2668 key.offset = new->file_pos;
2670 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2673 if (ret > 0 && path->slots[0] > 0)
2676 /* find out all the old extents for the file range */
2678 struct btrfs_file_extent_item *extent;
2679 struct extent_buffer *l;
2688 slot = path->slots[0];
2690 if (slot >= btrfs_header_nritems(l)) {
2691 ret = btrfs_next_leaf(root, path);
2699 btrfs_item_key_to_cpu(l, &key, slot);
2701 if (key.objectid != btrfs_ino(inode))
2703 if (key.type != BTRFS_EXTENT_DATA_KEY)
2705 if (key.offset >= new->file_pos + new->len)
2708 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2710 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2711 if (key.offset + num_bytes < new->file_pos)
2714 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2718 extent_offset = btrfs_file_extent_offset(l, extent);
2720 old = kmalloc(sizeof(*old), GFP_NOFS);
2724 offset = max(new->file_pos, key.offset);
2725 end = min(new->file_pos + new->len, key.offset + num_bytes);
2727 old->bytenr = disk_bytenr;
2728 old->extent_offset = extent_offset;
2729 old->offset = offset - key.offset;
2730 old->len = end - offset;
2733 list_add_tail(&old->list, &new->head);
2739 btrfs_free_path(path);
2740 atomic_inc(&root->fs_info->defrag_running);
2745 btrfs_free_path(path);
2747 free_sa_defrag_extent(new);
2751 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2754 struct btrfs_block_group_cache *cache;
2756 cache = btrfs_lookup_block_group(root->fs_info, start);
2759 spin_lock(&cache->lock);
2760 cache->delalloc_bytes -= len;
2761 spin_unlock(&cache->lock);
2763 btrfs_put_block_group(cache);
2766 /* as ordered data IO finishes, this gets called so we can finish
2767 * an ordered extent if the range of bytes in the file it covers are
2770 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2772 struct inode *inode = ordered_extent->inode;
2773 struct btrfs_root *root = BTRFS_I(inode)->root;
2774 struct btrfs_trans_handle *trans = NULL;
2775 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2776 struct extent_state *cached_state = NULL;
2777 struct new_sa_defrag_extent *new = NULL;
2778 int compress_type = 0;
2780 u64 logical_len = ordered_extent->len;
2782 bool truncated = false;
2784 nolock = btrfs_is_free_space_inode(inode);
2786 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2791 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2792 ordered_extent->file_offset +
2793 ordered_extent->len - 1);
2795 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2797 logical_len = ordered_extent->truncated_len;
2798 /* Truncated the entire extent, don't bother adding */
2803 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2804 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2805 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2807 trans = btrfs_join_transaction_nolock(root);
2809 trans = btrfs_join_transaction(root);
2810 if (IS_ERR(trans)) {
2811 ret = PTR_ERR(trans);
2815 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2816 ret = btrfs_update_inode_fallback(trans, root, inode);
2817 if (ret) /* -ENOMEM or corruption */
2818 btrfs_abort_transaction(trans, root, ret);
2822 lock_extent_bits(io_tree, ordered_extent->file_offset,
2823 ordered_extent->file_offset + ordered_extent->len - 1,
2826 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2827 ordered_extent->file_offset + ordered_extent->len - 1,
2828 EXTENT_DEFRAG, 1, cached_state);
2830 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2831 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2832 /* the inode is shared */
2833 new = record_old_file_extents(inode, ordered_extent);
2835 clear_extent_bit(io_tree, ordered_extent->file_offset,
2836 ordered_extent->file_offset + ordered_extent->len - 1,
2837 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2841 trans = btrfs_join_transaction_nolock(root);
2843 trans = btrfs_join_transaction(root);
2844 if (IS_ERR(trans)) {
2845 ret = PTR_ERR(trans);
2850 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2852 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2853 compress_type = ordered_extent->compress_type;
2854 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2855 BUG_ON(compress_type);
2856 ret = btrfs_mark_extent_written(trans, inode,
2857 ordered_extent->file_offset,
2858 ordered_extent->file_offset +
2861 BUG_ON(root == root->fs_info->tree_root);
2862 ret = insert_reserved_file_extent(trans, inode,
2863 ordered_extent->file_offset,
2864 ordered_extent->start,
2865 ordered_extent->disk_len,
2866 logical_len, logical_len,
2867 compress_type, 0, 0,
2868 BTRFS_FILE_EXTENT_REG);
2870 btrfs_release_delalloc_bytes(root,
2871 ordered_extent->start,
2872 ordered_extent->disk_len);
2874 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2875 ordered_extent->file_offset, ordered_extent->len,
2878 btrfs_abort_transaction(trans, root, ret);
2882 add_pending_csums(trans, inode, ordered_extent->file_offset,
2883 &ordered_extent->list);
2885 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2886 ret = btrfs_update_inode_fallback(trans, root, inode);
2887 if (ret) { /* -ENOMEM or corruption */
2888 btrfs_abort_transaction(trans, root, ret);
2893 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2894 ordered_extent->file_offset +
2895 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2897 if (root != root->fs_info->tree_root)
2898 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2900 btrfs_end_transaction(trans, root);
2902 if (ret || truncated) {
2906 start = ordered_extent->file_offset + logical_len;
2908 start = ordered_extent->file_offset;
2909 end = ordered_extent->file_offset + ordered_extent->len - 1;
2910 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2912 /* Drop the cache for the part of the extent we didn't write. */
2913 btrfs_drop_extent_cache(inode, start, end, 0);
2916 * If the ordered extent had an IOERR or something else went
2917 * wrong we need to return the space for this ordered extent
2918 * back to the allocator. We only free the extent in the
2919 * truncated case if we didn't write out the extent at all.
2921 if ((ret || !logical_len) &&
2922 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2923 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2924 btrfs_free_reserved_extent(root, ordered_extent->start,
2925 ordered_extent->disk_len, 1);
2930 * This needs to be done to make sure anybody waiting knows we are done
2931 * updating everything for this ordered extent.
2933 btrfs_remove_ordered_extent(inode, ordered_extent);
2935 /* for snapshot-aware defrag */
2938 free_sa_defrag_extent(new);
2939 atomic_dec(&root->fs_info->defrag_running);
2941 relink_file_extents(new);
2946 btrfs_put_ordered_extent(ordered_extent);
2947 /* once for the tree */
2948 btrfs_put_ordered_extent(ordered_extent);
2953 static void finish_ordered_fn(struct btrfs_work *work)
2955 struct btrfs_ordered_extent *ordered_extent;
2956 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2957 btrfs_finish_ordered_io(ordered_extent);
2960 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2961 struct extent_state *state, int uptodate)
2963 struct inode *inode = page->mapping->host;
2964 struct btrfs_root *root = BTRFS_I(inode)->root;
2965 struct btrfs_ordered_extent *ordered_extent = NULL;
2966 struct btrfs_workqueue *wq;
2967 btrfs_work_func_t func;
2969 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2971 ClearPagePrivate2(page);
2972 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2973 end - start + 1, uptodate))
2976 if (btrfs_is_free_space_inode(inode)) {
2977 wq = root->fs_info->endio_freespace_worker;
2978 func = btrfs_freespace_write_helper;
2980 wq = root->fs_info->endio_write_workers;
2981 func = btrfs_endio_write_helper;
2984 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
2986 btrfs_queue_work(wq, &ordered_extent->work);
2991 static int __readpage_endio_check(struct inode *inode,
2992 struct btrfs_io_bio *io_bio,
2993 int icsum, struct page *page,
2994 int pgoff, u64 start, size_t len)
2999 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
3000 DEFAULT_RATELIMIT_BURST);
3002 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3004 kaddr = kmap_atomic(page);
3005 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3006 btrfs_csum_final(csum, (char *)&csum);
3007 if (csum != csum_expected)
3010 kunmap_atomic(kaddr);
3013 if (__ratelimit(&_rs))
3014 btrfs_warn(BTRFS_I(inode)->root->fs_info,
3015 "csum failed ino %llu off %llu csum %u expected csum %u",
3016 btrfs_ino(inode), start, csum, csum_expected);
3017 memset(kaddr + pgoff, 1, len);
3018 flush_dcache_page(page);
3019 kunmap_atomic(kaddr);
3020 if (csum_expected == 0)
3026 * when reads are done, we need to check csums to verify the data is correct
3027 * if there's a match, we allow the bio to finish. If not, the code in
3028 * extent_io.c will try to find good copies for us.
3030 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3031 u64 phy_offset, struct page *page,
3032 u64 start, u64 end, int mirror)
3034 size_t offset = start - page_offset(page);
3035 struct inode *inode = page->mapping->host;
3036 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3037 struct btrfs_root *root = BTRFS_I(inode)->root;
3039 if (PageChecked(page)) {
3040 ClearPageChecked(page);
3044 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3047 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3048 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3049 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
3054 phy_offset >>= inode->i_sb->s_blocksize_bits;
3055 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3056 start, (size_t)(end - start + 1));
3059 struct delayed_iput {
3060 struct list_head list;
3061 struct inode *inode;
3064 /* JDM: If this is fs-wide, why can't we add a pointer to
3065 * btrfs_inode instead and avoid the allocation? */
3066 void btrfs_add_delayed_iput(struct inode *inode)
3068 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3069 struct delayed_iput *delayed;
3071 if (atomic_add_unless(&inode->i_count, -1, 1))
3074 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
3075 delayed->inode = inode;
3077 spin_lock(&fs_info->delayed_iput_lock);
3078 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
3079 spin_unlock(&fs_info->delayed_iput_lock);
3082 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3085 struct btrfs_fs_info *fs_info = root->fs_info;
3086 struct delayed_iput *delayed;
3089 spin_lock(&fs_info->delayed_iput_lock);
3090 empty = list_empty(&fs_info->delayed_iputs);
3091 spin_unlock(&fs_info->delayed_iput_lock);
3095 spin_lock(&fs_info->delayed_iput_lock);
3096 list_splice_init(&fs_info->delayed_iputs, &list);
3097 spin_unlock(&fs_info->delayed_iput_lock);
3099 while (!list_empty(&list)) {
3100 delayed = list_entry(list.next, struct delayed_iput, list);
3101 list_del(&delayed->list);
3102 iput(delayed->inode);
3108 * This is called in transaction commit time. If there are no orphan
3109 * files in the subvolume, it removes orphan item and frees block_rsv
3112 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3113 struct btrfs_root *root)
3115 struct btrfs_block_rsv *block_rsv;
3118 if (atomic_read(&root->orphan_inodes) ||
3119 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3122 spin_lock(&root->orphan_lock);
3123 if (atomic_read(&root->orphan_inodes)) {
3124 spin_unlock(&root->orphan_lock);
3128 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3129 spin_unlock(&root->orphan_lock);
3133 block_rsv = root->orphan_block_rsv;
3134 root->orphan_block_rsv = NULL;
3135 spin_unlock(&root->orphan_lock);
3137 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3138 btrfs_root_refs(&root->root_item) > 0) {
3139 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3140 root->root_key.objectid);
3142 btrfs_abort_transaction(trans, root, ret);
3144 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3149 WARN_ON(block_rsv->size > 0);
3150 btrfs_free_block_rsv(root, block_rsv);
3155 * This creates an orphan entry for the given inode in case something goes
3156 * wrong in the middle of an unlink/truncate.
3158 * NOTE: caller of this function should reserve 5 units of metadata for
3161 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3163 struct btrfs_root *root = BTRFS_I(inode)->root;
3164 struct btrfs_block_rsv *block_rsv = NULL;
3169 if (!root->orphan_block_rsv) {
3170 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3175 spin_lock(&root->orphan_lock);
3176 if (!root->orphan_block_rsv) {
3177 root->orphan_block_rsv = block_rsv;
3178 } else if (block_rsv) {
3179 btrfs_free_block_rsv(root, block_rsv);
3183 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3184 &BTRFS_I(inode)->runtime_flags)) {
3187 * For proper ENOSPC handling, we should do orphan
3188 * cleanup when mounting. But this introduces backward
3189 * compatibility issue.
3191 if (!xchg(&root->orphan_item_inserted, 1))
3197 atomic_inc(&root->orphan_inodes);
3200 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3201 &BTRFS_I(inode)->runtime_flags))
3203 spin_unlock(&root->orphan_lock);
3205 /* grab metadata reservation from transaction handle */
3207 ret = btrfs_orphan_reserve_metadata(trans, inode);
3208 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3211 /* insert an orphan item to track this unlinked/truncated file */
3213 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3215 atomic_dec(&root->orphan_inodes);
3217 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3218 &BTRFS_I(inode)->runtime_flags);
3219 btrfs_orphan_release_metadata(inode);
3221 if (ret != -EEXIST) {
3222 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3223 &BTRFS_I(inode)->runtime_flags);
3224 btrfs_abort_transaction(trans, root, ret);
3231 /* insert an orphan item to track subvolume contains orphan files */
3233 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3234 root->root_key.objectid);
3235 if (ret && ret != -EEXIST) {
3236 btrfs_abort_transaction(trans, root, ret);
3244 * We have done the truncate/delete so we can go ahead and remove the orphan
3245 * item for this particular inode.
3247 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3248 struct inode *inode)
3250 struct btrfs_root *root = BTRFS_I(inode)->root;
3251 int delete_item = 0;
3252 int release_rsv = 0;
3255 spin_lock(&root->orphan_lock);
3256 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3257 &BTRFS_I(inode)->runtime_flags))
3260 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3261 &BTRFS_I(inode)->runtime_flags))
3263 spin_unlock(&root->orphan_lock);
3266 atomic_dec(&root->orphan_inodes);
3268 ret = btrfs_del_orphan_item(trans, root,
3273 btrfs_orphan_release_metadata(inode);
3279 * this cleans up any orphans that may be left on the list from the last use
3282 int btrfs_orphan_cleanup(struct btrfs_root *root)
3284 struct btrfs_path *path;
3285 struct extent_buffer *leaf;
3286 struct btrfs_key key, found_key;
3287 struct btrfs_trans_handle *trans;
3288 struct inode *inode;
3289 u64 last_objectid = 0;
3290 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3292 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3295 path = btrfs_alloc_path();
3302 key.objectid = BTRFS_ORPHAN_OBJECTID;
3303 key.type = BTRFS_ORPHAN_ITEM_KEY;
3304 key.offset = (u64)-1;
3307 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3312 * if ret == 0 means we found what we were searching for, which
3313 * is weird, but possible, so only screw with path if we didn't
3314 * find the key and see if we have stuff that matches
3318 if (path->slots[0] == 0)
3323 /* pull out the item */
3324 leaf = path->nodes[0];
3325 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3327 /* make sure the item matches what we want */
3328 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3330 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3333 /* release the path since we're done with it */
3334 btrfs_release_path(path);
3337 * this is where we are basically btrfs_lookup, without the
3338 * crossing root thing. we store the inode number in the
3339 * offset of the orphan item.
3342 if (found_key.offset == last_objectid) {
3343 btrfs_err(root->fs_info,
3344 "Error removing orphan entry, stopping orphan cleanup");
3349 last_objectid = found_key.offset;
3351 found_key.objectid = found_key.offset;
3352 found_key.type = BTRFS_INODE_ITEM_KEY;
3353 found_key.offset = 0;
3354 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3355 ret = PTR_ERR_OR_ZERO(inode);
3356 if (ret && ret != -ESTALE)
3359 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3360 struct btrfs_root *dead_root;
3361 struct btrfs_fs_info *fs_info = root->fs_info;
3362 int is_dead_root = 0;
3365 * this is an orphan in the tree root. Currently these
3366 * could come from 2 sources:
3367 * a) a snapshot deletion in progress
3368 * b) a free space cache inode
3369 * We need to distinguish those two, as the snapshot
3370 * orphan must not get deleted.
3371 * find_dead_roots already ran before us, so if this
3372 * is a snapshot deletion, we should find the root
3373 * in the dead_roots list
3375 spin_lock(&fs_info->trans_lock);
3376 list_for_each_entry(dead_root, &fs_info->dead_roots,
3378 if (dead_root->root_key.objectid ==
3379 found_key.objectid) {
3384 spin_unlock(&fs_info->trans_lock);
3386 /* prevent this orphan from being found again */
3387 key.offset = found_key.objectid - 1;
3392 * Inode is already gone but the orphan item is still there,
3393 * kill the orphan item.
3395 if (ret == -ESTALE) {
3396 trans = btrfs_start_transaction(root, 1);
3397 if (IS_ERR(trans)) {
3398 ret = PTR_ERR(trans);
3401 btrfs_debug(root->fs_info, "auto deleting %Lu",
3402 found_key.objectid);
3403 ret = btrfs_del_orphan_item(trans, root,
3404 found_key.objectid);
3405 btrfs_end_transaction(trans, root);
3412 * add this inode to the orphan list so btrfs_orphan_del does
3413 * the proper thing when we hit it
3415 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3416 &BTRFS_I(inode)->runtime_flags);
3417 atomic_inc(&root->orphan_inodes);
3419 /* if we have links, this was a truncate, lets do that */
3420 if (inode->i_nlink) {
3421 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3427 /* 1 for the orphan item deletion. */
3428 trans = btrfs_start_transaction(root, 1);
3429 if (IS_ERR(trans)) {
3431 ret = PTR_ERR(trans);
3434 ret = btrfs_orphan_add(trans, inode);
3435 btrfs_end_transaction(trans, root);
3441 ret = btrfs_truncate(inode);
3443 btrfs_orphan_del(NULL, inode);
3448 /* this will do delete_inode and everything for us */
3453 /* release the path since we're done with it */
3454 btrfs_release_path(path);
3456 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3458 if (root->orphan_block_rsv)
3459 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3462 if (root->orphan_block_rsv ||
3463 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3464 trans = btrfs_join_transaction(root);
3466 btrfs_end_transaction(trans, root);
3470 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3472 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3476 btrfs_err(root->fs_info,
3477 "could not do orphan cleanup %d", ret);
3478 btrfs_free_path(path);
3483 * very simple check to peek ahead in the leaf looking for xattrs. If we
3484 * don't find any xattrs, we know there can't be any acls.
3486 * slot is the slot the inode is in, objectid is the objectid of the inode
3488 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3489 int slot, u64 objectid,
3490 int *first_xattr_slot)
3492 u32 nritems = btrfs_header_nritems(leaf);
3493 struct btrfs_key found_key;
3494 static u64 xattr_access = 0;
3495 static u64 xattr_default = 0;
3498 if (!xattr_access) {
3499 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3500 strlen(POSIX_ACL_XATTR_ACCESS));
3501 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3502 strlen(POSIX_ACL_XATTR_DEFAULT));
3506 *first_xattr_slot = -1;
3507 while (slot < nritems) {
3508 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3510 /* we found a different objectid, there must not be acls */
3511 if (found_key.objectid != objectid)
3514 /* we found an xattr, assume we've got an acl */
3515 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3516 if (*first_xattr_slot == -1)
3517 *first_xattr_slot = slot;
3518 if (found_key.offset == xattr_access ||
3519 found_key.offset == xattr_default)
3524 * we found a key greater than an xattr key, there can't
3525 * be any acls later on
3527 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3534 * it goes inode, inode backrefs, xattrs, extents,
3535 * so if there are a ton of hard links to an inode there can
3536 * be a lot of backrefs. Don't waste time searching too hard,
3537 * this is just an optimization
3542 /* we hit the end of the leaf before we found an xattr or
3543 * something larger than an xattr. We have to assume the inode
3546 if (*first_xattr_slot == -1)
3547 *first_xattr_slot = slot;
3552 * read an inode from the btree into the in-memory inode
3554 static void btrfs_read_locked_inode(struct inode *inode)
3556 struct btrfs_path *path;
3557 struct extent_buffer *leaf;
3558 struct btrfs_inode_item *inode_item;
3559 struct btrfs_root *root = BTRFS_I(inode)->root;
3560 struct btrfs_key location;
3565 bool filled = false;
3566 int first_xattr_slot;
3568 ret = btrfs_fill_inode(inode, &rdev);
3572 path = btrfs_alloc_path();
3576 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3578 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3582 leaf = path->nodes[0];
3587 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3588 struct btrfs_inode_item);
3589 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3590 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3591 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3592 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3593 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3595 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3596 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3598 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3599 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3601 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3602 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3604 BTRFS_I(inode)->i_otime.tv_sec =
3605 btrfs_timespec_sec(leaf, &inode_item->otime);
3606 BTRFS_I(inode)->i_otime.tv_nsec =
3607 btrfs_timespec_nsec(leaf, &inode_item->otime);
3609 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3610 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3611 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3614 * If we were modified in the current generation and evicted from memory
3615 * and then re-read we need to do a full sync since we don't have any
3616 * idea about which extents were modified before we were evicted from
3619 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3620 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3621 &BTRFS_I(inode)->runtime_flags);
3623 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3624 inode->i_generation = BTRFS_I(inode)->generation;
3626 rdev = btrfs_inode_rdev(leaf, inode_item);
3628 BTRFS_I(inode)->index_cnt = (u64)-1;
3629 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3633 if (inode->i_nlink != 1 ||
3634 path->slots[0] >= btrfs_header_nritems(leaf))
3637 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3638 if (location.objectid != btrfs_ino(inode))
3641 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3642 if (location.type == BTRFS_INODE_REF_KEY) {
3643 struct btrfs_inode_ref *ref;
3645 ref = (struct btrfs_inode_ref *)ptr;
3646 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3647 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3648 struct btrfs_inode_extref *extref;
3650 extref = (struct btrfs_inode_extref *)ptr;
3651 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3656 * try to precache a NULL acl entry for files that don't have
3657 * any xattrs or acls
3659 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3660 btrfs_ino(inode), &first_xattr_slot);
3661 if (first_xattr_slot != -1) {
3662 path->slots[0] = first_xattr_slot;
3663 ret = btrfs_load_inode_props(inode, path);
3665 btrfs_err(root->fs_info,
3666 "error loading props for ino %llu (root %llu): %d",
3668 root->root_key.objectid, ret);
3670 btrfs_free_path(path);
3673 cache_no_acl(inode);
3675 switch (inode->i_mode & S_IFMT) {
3677 inode->i_mapping->a_ops = &btrfs_aops;
3678 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3679 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3680 inode->i_fop = &btrfs_file_operations;
3681 inode->i_op = &btrfs_file_inode_operations;
3684 inode->i_fop = &btrfs_dir_file_operations;
3685 if (root == root->fs_info->tree_root)
3686 inode->i_op = &btrfs_dir_ro_inode_operations;
3688 inode->i_op = &btrfs_dir_inode_operations;
3691 inode->i_op = &btrfs_symlink_inode_operations;
3692 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3693 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3696 inode->i_op = &btrfs_special_inode_operations;
3697 init_special_inode(inode, inode->i_mode, rdev);
3701 btrfs_update_iflags(inode);
3705 btrfs_free_path(path);
3706 make_bad_inode(inode);
3710 * given a leaf and an inode, copy the inode fields into the leaf
3712 static void fill_inode_item(struct btrfs_trans_handle *trans,
3713 struct extent_buffer *leaf,
3714 struct btrfs_inode_item *item,
3715 struct inode *inode)
3717 struct btrfs_map_token token;
3719 btrfs_init_map_token(&token);
3721 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3722 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3723 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3725 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3726 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3728 btrfs_set_token_timespec_sec(leaf, &item->atime,
3729 inode->i_atime.tv_sec, &token);
3730 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3731 inode->i_atime.tv_nsec, &token);
3733 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3734 inode->i_mtime.tv_sec, &token);
3735 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3736 inode->i_mtime.tv_nsec, &token);
3738 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3739 inode->i_ctime.tv_sec, &token);
3740 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3741 inode->i_ctime.tv_nsec, &token);
3743 btrfs_set_token_timespec_sec(leaf, &item->otime,
3744 BTRFS_I(inode)->i_otime.tv_sec, &token);
3745 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3746 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3748 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3750 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3752 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3753 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3754 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3755 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3756 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3760 * copy everything in the in-memory inode into the btree.
3762 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3763 struct btrfs_root *root, struct inode *inode)
3765 struct btrfs_inode_item *inode_item;
3766 struct btrfs_path *path;
3767 struct extent_buffer *leaf;
3770 path = btrfs_alloc_path();
3774 path->leave_spinning = 1;
3775 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3783 leaf = path->nodes[0];
3784 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3785 struct btrfs_inode_item);
3787 fill_inode_item(trans, leaf, inode_item, inode);
3788 btrfs_mark_buffer_dirty(leaf);
3789 btrfs_set_inode_last_trans(trans, inode);
3792 btrfs_free_path(path);
3797 * copy everything in the in-memory inode into the btree.
3799 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3800 struct btrfs_root *root, struct inode *inode)
3805 * If the inode is a free space inode, we can deadlock during commit
3806 * if we put it into the delayed code.
3808 * The data relocation inode should also be directly updated
3811 if (!btrfs_is_free_space_inode(inode)
3812 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3813 && !root->fs_info->log_root_recovering) {
3814 btrfs_update_root_times(trans, root);
3816 ret = btrfs_delayed_update_inode(trans, root, inode);
3818 btrfs_set_inode_last_trans(trans, inode);
3822 return btrfs_update_inode_item(trans, root, inode);
3825 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3826 struct btrfs_root *root,
3827 struct inode *inode)
3831 ret = btrfs_update_inode(trans, root, inode);
3833 return btrfs_update_inode_item(trans, root, inode);
3838 * unlink helper that gets used here in inode.c and in the tree logging
3839 * recovery code. It remove a link in a directory with a given name, and
3840 * also drops the back refs in the inode to the directory
3842 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3843 struct btrfs_root *root,
3844 struct inode *dir, struct inode *inode,
3845 const char *name, int name_len)
3847 struct btrfs_path *path;
3849 struct extent_buffer *leaf;
3850 struct btrfs_dir_item *di;
3851 struct btrfs_key key;
3853 u64 ino = btrfs_ino(inode);
3854 u64 dir_ino = btrfs_ino(dir);
3856 path = btrfs_alloc_path();
3862 path->leave_spinning = 1;
3863 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3864 name, name_len, -1);
3873 leaf = path->nodes[0];
3874 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3875 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3878 btrfs_release_path(path);
3881 * If we don't have dir index, we have to get it by looking up
3882 * the inode ref, since we get the inode ref, remove it directly,
3883 * it is unnecessary to do delayed deletion.
3885 * But if we have dir index, needn't search inode ref to get it.
3886 * Since the inode ref is close to the inode item, it is better
3887 * that we delay to delete it, and just do this deletion when
3888 * we update the inode item.
3890 if (BTRFS_I(inode)->dir_index) {
3891 ret = btrfs_delayed_delete_inode_ref(inode);
3893 index = BTRFS_I(inode)->dir_index;
3898 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3901 btrfs_info(root->fs_info,
3902 "failed to delete reference to %.*s, inode %llu parent %llu",
3903 name_len, name, ino, dir_ino);
3904 btrfs_abort_transaction(trans, root, ret);
3908 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3910 btrfs_abort_transaction(trans, root, ret);
3914 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3916 if (ret != 0 && ret != -ENOENT) {
3917 btrfs_abort_transaction(trans, root, ret);
3921 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3926 btrfs_abort_transaction(trans, root, ret);
3928 btrfs_free_path(path);
3932 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3933 inode_inc_iversion(inode);
3934 inode_inc_iversion(dir);
3935 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3936 ret = btrfs_update_inode(trans, root, dir);
3941 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3942 struct btrfs_root *root,
3943 struct inode *dir, struct inode *inode,
3944 const char *name, int name_len)
3947 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3950 ret = btrfs_update_inode(trans, root, inode);
3956 * helper to start transaction for unlink and rmdir.
3958 * unlink and rmdir are special in btrfs, they do not always free space, so
3959 * if we cannot make our reservations the normal way try and see if there is
3960 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3961 * allow the unlink to occur.
3963 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3965 struct btrfs_trans_handle *trans;
3966 struct btrfs_root *root = BTRFS_I(dir)->root;
3970 * 1 for the possible orphan item
3971 * 1 for the dir item
3972 * 1 for the dir index
3973 * 1 for the inode ref
3976 trans = btrfs_start_transaction(root, 5);
3977 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3980 if (PTR_ERR(trans) == -ENOSPC) {
3981 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3983 trans = btrfs_start_transaction(root, 0);
3986 ret = btrfs_cond_migrate_bytes(root->fs_info,
3987 &root->fs_info->trans_block_rsv,
3990 btrfs_end_transaction(trans, root);
3991 return ERR_PTR(ret);
3993 trans->block_rsv = &root->fs_info->trans_block_rsv;
3994 trans->bytes_reserved = num_bytes;
3999 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4001 struct btrfs_root *root = BTRFS_I(dir)->root;
4002 struct btrfs_trans_handle *trans;
4003 struct inode *inode = dentry->d_inode;
4006 trans = __unlink_start_trans(dir);
4008 return PTR_ERR(trans);
4010 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
4012 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
4013 dentry->d_name.name, dentry->d_name.len);
4017 if (inode->i_nlink == 0) {
4018 ret = btrfs_orphan_add(trans, inode);
4024 btrfs_end_transaction(trans, root);
4025 btrfs_btree_balance_dirty(root);
4029 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4030 struct btrfs_root *root,
4031 struct inode *dir, u64 objectid,
4032 const char *name, int name_len)
4034 struct btrfs_path *path;
4035 struct extent_buffer *leaf;
4036 struct btrfs_dir_item *di;
4037 struct btrfs_key key;
4040 u64 dir_ino = btrfs_ino(dir);
4042 path = btrfs_alloc_path();
4046 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4047 name, name_len, -1);
4048 if (IS_ERR_OR_NULL(di)) {
4056 leaf = path->nodes[0];
4057 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4058 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4059 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4061 btrfs_abort_transaction(trans, root, ret);
4064 btrfs_release_path(path);
4066 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4067 objectid, root->root_key.objectid,
4068 dir_ino, &index, name, name_len);
4070 if (ret != -ENOENT) {
4071 btrfs_abort_transaction(trans, root, ret);
4074 di = btrfs_search_dir_index_item(root, path, dir_ino,
4076 if (IS_ERR_OR_NULL(di)) {
4081 btrfs_abort_transaction(trans, root, ret);
4085 leaf = path->nodes[0];
4086 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4087 btrfs_release_path(path);
4090 btrfs_release_path(path);
4092 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4094 btrfs_abort_transaction(trans, root, ret);
4098 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4099 inode_inc_iversion(dir);
4100 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4101 ret = btrfs_update_inode_fallback(trans, root, dir);
4103 btrfs_abort_transaction(trans, root, ret);
4105 btrfs_free_path(path);
4109 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4111 struct inode *inode = dentry->d_inode;
4113 struct btrfs_root *root = BTRFS_I(dir)->root;
4114 struct btrfs_trans_handle *trans;
4116 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4118 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4121 trans = __unlink_start_trans(dir);
4123 return PTR_ERR(trans);
4125 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4126 err = btrfs_unlink_subvol(trans, root, dir,
4127 BTRFS_I(inode)->location.objectid,
4128 dentry->d_name.name,
4129 dentry->d_name.len);
4133 err = btrfs_orphan_add(trans, inode);
4137 /* now the directory is empty */
4138 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
4139 dentry->d_name.name, dentry->d_name.len);
4141 btrfs_i_size_write(inode, 0);
4143 btrfs_end_transaction(trans, root);
4144 btrfs_btree_balance_dirty(root);
4150 * this can truncate away extent items, csum items and directory items.
4151 * It starts at a high offset and removes keys until it can't find
4152 * any higher than new_size
4154 * csum items that cross the new i_size are truncated to the new size
4157 * min_type is the minimum key type to truncate down to. If set to 0, this
4158 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4160 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4161 struct btrfs_root *root,
4162 struct inode *inode,
4163 u64 new_size, u32 min_type)
4165 struct btrfs_path *path;
4166 struct extent_buffer *leaf;
4167 struct btrfs_file_extent_item *fi;
4168 struct btrfs_key key;
4169 struct btrfs_key found_key;
4170 u64 extent_start = 0;
4171 u64 extent_num_bytes = 0;
4172 u64 extent_offset = 0;
4174 u64 last_size = (u64)-1;
4175 u32 found_type = (u8)-1;
4178 int pending_del_nr = 0;
4179 int pending_del_slot = 0;
4180 int extent_type = -1;
4183 u64 ino = btrfs_ino(inode);
4185 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4187 path = btrfs_alloc_path();
4193 * We want to drop from the next block forward in case this new size is
4194 * not block aligned since we will be keeping the last block of the
4195 * extent just the way it is.
4197 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4198 root == root->fs_info->tree_root)
4199 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4200 root->sectorsize), (u64)-1, 0);
4203 * This function is also used to drop the items in the log tree before
4204 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4205 * it is used to drop the loged items. So we shouldn't kill the delayed
4208 if (min_type == 0 && root == BTRFS_I(inode)->root)
4209 btrfs_kill_delayed_inode_items(inode);
4212 key.offset = (u64)-1;
4216 path->leave_spinning = 1;
4217 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4224 /* there are no items in the tree for us to truncate, we're
4227 if (path->slots[0] == 0)
4234 leaf = path->nodes[0];
4235 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4236 found_type = found_key.type;
4238 if (found_key.objectid != ino)
4241 if (found_type < min_type)
4244 item_end = found_key.offset;
4245 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4246 fi = btrfs_item_ptr(leaf, path->slots[0],
4247 struct btrfs_file_extent_item);
4248 extent_type = btrfs_file_extent_type(leaf, fi);
4249 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4251 btrfs_file_extent_num_bytes(leaf, fi);
4252 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4253 item_end += btrfs_file_extent_inline_len(leaf,
4254 path->slots[0], fi);
4258 if (found_type > min_type) {
4261 if (item_end < new_size)
4263 if (found_key.offset >= new_size)
4269 /* FIXME, shrink the extent if the ref count is only 1 */
4270 if (found_type != BTRFS_EXTENT_DATA_KEY)
4274 last_size = found_key.offset;
4276 last_size = new_size;
4278 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4280 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4282 u64 orig_num_bytes =
4283 btrfs_file_extent_num_bytes(leaf, fi);
4284 extent_num_bytes = ALIGN(new_size -
4287 btrfs_set_file_extent_num_bytes(leaf, fi,
4289 num_dec = (orig_num_bytes -
4291 if (test_bit(BTRFS_ROOT_REF_COWS,
4294 inode_sub_bytes(inode, num_dec);
4295 btrfs_mark_buffer_dirty(leaf);
4298 btrfs_file_extent_disk_num_bytes(leaf,
4300 extent_offset = found_key.offset -
4301 btrfs_file_extent_offset(leaf, fi);
4303 /* FIXME blocksize != 4096 */
4304 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4305 if (extent_start != 0) {
4307 if (test_bit(BTRFS_ROOT_REF_COWS,
4309 inode_sub_bytes(inode, num_dec);
4312 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4314 * we can't truncate inline items that have had
4318 btrfs_file_extent_compression(leaf, fi) == 0 &&
4319 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4320 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4321 u32 size = new_size - found_key.offset;
4323 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4324 inode_sub_bytes(inode, item_end + 1 -
4328 * update the ram bytes to properly reflect
4329 * the new size of our item
4331 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4333 btrfs_file_extent_calc_inline_size(size);
4334 btrfs_truncate_item(root, path, size, 1);
4335 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4337 inode_sub_bytes(inode, item_end + 1 -
4343 if (!pending_del_nr) {
4344 /* no pending yet, add ourselves */
4345 pending_del_slot = path->slots[0];
4347 } else if (pending_del_nr &&
4348 path->slots[0] + 1 == pending_del_slot) {
4349 /* hop on the pending chunk */
4351 pending_del_slot = path->slots[0];
4359 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4360 root == root->fs_info->tree_root)) {
4361 btrfs_set_path_blocking(path);
4362 ret = btrfs_free_extent(trans, root, extent_start,
4363 extent_num_bytes, 0,
4364 btrfs_header_owner(leaf),
4365 ino, extent_offset, 0);
4369 if (found_type == BTRFS_INODE_ITEM_KEY)
4372 if (path->slots[0] == 0 ||
4373 path->slots[0] != pending_del_slot) {
4374 if (pending_del_nr) {
4375 ret = btrfs_del_items(trans, root, path,
4379 btrfs_abort_transaction(trans,
4385 btrfs_release_path(path);
4392 if (pending_del_nr) {
4393 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4396 btrfs_abort_transaction(trans, root, ret);
4399 if (last_size != (u64)-1 &&
4400 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4401 btrfs_ordered_update_i_size(inode, last_size, NULL);
4402 btrfs_free_path(path);
4407 * btrfs_truncate_page - read, zero a chunk and write a page
4408 * @inode - inode that we're zeroing
4409 * @from - the offset to start zeroing
4410 * @len - the length to zero, 0 to zero the entire range respective to the
4412 * @front - zero up to the offset instead of from the offset on
4414 * This will find the page for the "from" offset and cow the page and zero the
4415 * part we want to zero. This is used with truncate and hole punching.
4417 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4420 struct address_space *mapping = inode->i_mapping;
4421 struct btrfs_root *root = BTRFS_I(inode)->root;
4422 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4423 struct btrfs_ordered_extent *ordered;
4424 struct extent_state *cached_state = NULL;
4426 u32 blocksize = root->sectorsize;
4427 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4428 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4430 gfp_t mask = btrfs_alloc_write_mask(mapping);
4435 if ((offset & (blocksize - 1)) == 0 &&
4436 (!len || ((len & (blocksize - 1)) == 0)))
4438 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4443 page = find_or_create_page(mapping, index, mask);
4445 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4450 page_start = page_offset(page);
4451 page_end = page_start + PAGE_CACHE_SIZE - 1;
4453 if (!PageUptodate(page)) {
4454 ret = btrfs_readpage(NULL, page);
4456 if (page->mapping != mapping) {
4458 page_cache_release(page);
4461 if (!PageUptodate(page)) {
4466 wait_on_page_writeback(page);
4468 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4469 set_page_extent_mapped(page);
4471 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4473 unlock_extent_cached(io_tree, page_start, page_end,
4474 &cached_state, GFP_NOFS);
4476 page_cache_release(page);
4477 btrfs_start_ordered_extent(inode, ordered, 1);
4478 btrfs_put_ordered_extent(ordered);
4482 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4483 EXTENT_DIRTY | EXTENT_DELALLOC |
4484 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4485 0, 0, &cached_state, GFP_NOFS);
4487 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4490 unlock_extent_cached(io_tree, page_start, page_end,
4491 &cached_state, GFP_NOFS);
4495 if (offset != PAGE_CACHE_SIZE) {
4497 len = PAGE_CACHE_SIZE - offset;
4500 memset(kaddr, 0, offset);
4502 memset(kaddr + offset, 0, len);
4503 flush_dcache_page(page);
4506 ClearPageChecked(page);
4507 set_page_dirty(page);
4508 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4513 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4515 page_cache_release(page);
4520 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4521 u64 offset, u64 len)
4523 struct btrfs_trans_handle *trans;
4527 * Still need to make sure the inode looks like it's been updated so
4528 * that any holes get logged if we fsync.
4530 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4531 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4532 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4533 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4538 * 1 - for the one we're dropping
4539 * 1 - for the one we're adding
4540 * 1 - for updating the inode.
4542 trans = btrfs_start_transaction(root, 3);
4544 return PTR_ERR(trans);
4546 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4548 btrfs_abort_transaction(trans, root, ret);
4549 btrfs_end_transaction(trans, root);
4553 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4554 0, 0, len, 0, len, 0, 0, 0);
4556 btrfs_abort_transaction(trans, root, ret);
4558 btrfs_update_inode(trans, root, inode);
4559 btrfs_end_transaction(trans, root);
4564 * This function puts in dummy file extents for the area we're creating a hole
4565 * for. So if we are truncating this file to a larger size we need to insert
4566 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4567 * the range between oldsize and size
4569 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4571 struct btrfs_root *root = BTRFS_I(inode)->root;
4572 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4573 struct extent_map *em = NULL;
4574 struct extent_state *cached_state = NULL;
4575 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4576 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4577 u64 block_end = ALIGN(size, root->sectorsize);
4584 * If our size started in the middle of a page we need to zero out the
4585 * rest of the page before we expand the i_size, otherwise we could
4586 * expose stale data.
4588 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4592 if (size <= hole_start)
4596 struct btrfs_ordered_extent *ordered;
4598 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4600 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4601 block_end - hole_start);
4604 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4605 &cached_state, GFP_NOFS);
4606 btrfs_start_ordered_extent(inode, ordered, 1);
4607 btrfs_put_ordered_extent(ordered);
4610 cur_offset = hole_start;
4612 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4613 block_end - cur_offset, 0);
4619 last_byte = min(extent_map_end(em), block_end);
4620 last_byte = ALIGN(last_byte , root->sectorsize);
4621 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4622 struct extent_map *hole_em;
4623 hole_size = last_byte - cur_offset;
4625 err = maybe_insert_hole(root, inode, cur_offset,
4629 btrfs_drop_extent_cache(inode, cur_offset,
4630 cur_offset + hole_size - 1, 0);
4631 hole_em = alloc_extent_map();
4633 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4634 &BTRFS_I(inode)->runtime_flags);
4637 hole_em->start = cur_offset;
4638 hole_em->len = hole_size;
4639 hole_em->orig_start = cur_offset;
4641 hole_em->block_start = EXTENT_MAP_HOLE;
4642 hole_em->block_len = 0;
4643 hole_em->orig_block_len = 0;
4644 hole_em->ram_bytes = hole_size;
4645 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4646 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4647 hole_em->generation = root->fs_info->generation;
4650 write_lock(&em_tree->lock);
4651 err = add_extent_mapping(em_tree, hole_em, 1);
4652 write_unlock(&em_tree->lock);
4655 btrfs_drop_extent_cache(inode, cur_offset,
4659 free_extent_map(hole_em);
4662 free_extent_map(em);
4664 cur_offset = last_byte;
4665 if (cur_offset >= block_end)
4668 free_extent_map(em);
4669 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4674 static int wait_snapshoting_atomic_t(atomic_t *a)
4680 static void wait_for_snapshot_creation(struct btrfs_root *root)
4685 ret = btrfs_start_write_no_snapshoting(root);
4688 wait_on_atomic_t(&root->will_be_snapshoted,
4689 wait_snapshoting_atomic_t,
4690 TASK_UNINTERRUPTIBLE);
4694 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4696 struct btrfs_root *root = BTRFS_I(inode)->root;
4697 struct btrfs_trans_handle *trans;
4698 loff_t oldsize = i_size_read(inode);
4699 loff_t newsize = attr->ia_size;
4700 int mask = attr->ia_valid;
4704 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4705 * special case where we need to update the times despite not having
4706 * these flags set. For all other operations the VFS set these flags
4707 * explicitly if it wants a timestamp update.
4709 if (newsize != oldsize) {
4710 inode_inc_iversion(inode);
4711 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4712 inode->i_ctime = inode->i_mtime =
4713 current_fs_time(inode->i_sb);
4716 if (newsize > oldsize) {
4717 truncate_pagecache(inode, newsize);
4719 * Don't do an expanding truncate while snapshoting is ongoing.
4720 * This is to ensure the snapshot captures a fully consistent
4721 * state of this file - if the snapshot captures this expanding
4722 * truncation, it must capture all writes that happened before
4725 wait_for_snapshot_creation(root);
4726 ret = btrfs_cont_expand(inode, oldsize, newsize);
4728 btrfs_end_write_no_snapshoting(root);
4732 trans = btrfs_start_transaction(root, 1);
4733 if (IS_ERR(trans)) {
4734 btrfs_end_write_no_snapshoting(root);
4735 return PTR_ERR(trans);
4738 i_size_write(inode, newsize);
4739 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4740 ret = btrfs_update_inode(trans, root, inode);
4741 btrfs_end_write_no_snapshoting(root);
4742 btrfs_end_transaction(trans, root);
4746 * We're truncating a file that used to have good data down to
4747 * zero. Make sure it gets into the ordered flush list so that
4748 * any new writes get down to disk quickly.
4751 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4752 &BTRFS_I(inode)->runtime_flags);
4755 * 1 for the orphan item we're going to add
4756 * 1 for the orphan item deletion.
4758 trans = btrfs_start_transaction(root, 2);
4760 return PTR_ERR(trans);
4763 * We need to do this in case we fail at _any_ point during the
4764 * actual truncate. Once we do the truncate_setsize we could
4765 * invalidate pages which forces any outstanding ordered io to
4766 * be instantly completed which will give us extents that need
4767 * to be truncated. If we fail to get an orphan inode down we
4768 * could have left over extents that were never meant to live,
4769 * so we need to garuntee from this point on that everything
4770 * will be consistent.
4772 ret = btrfs_orphan_add(trans, inode);
4773 btrfs_end_transaction(trans, root);
4777 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4778 truncate_setsize(inode, newsize);
4780 /* Disable nonlocked read DIO to avoid the end less truncate */
4781 btrfs_inode_block_unlocked_dio(inode);
4782 inode_dio_wait(inode);
4783 btrfs_inode_resume_unlocked_dio(inode);
4785 ret = btrfs_truncate(inode);
4786 if (ret && inode->i_nlink) {
4790 * failed to truncate, disk_i_size is only adjusted down
4791 * as we remove extents, so it should represent the true
4792 * size of the inode, so reset the in memory size and
4793 * delete our orphan entry.
4795 trans = btrfs_join_transaction(root);
4796 if (IS_ERR(trans)) {
4797 btrfs_orphan_del(NULL, inode);
4800 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4801 err = btrfs_orphan_del(trans, inode);
4803 btrfs_abort_transaction(trans, root, err);
4804 btrfs_end_transaction(trans, root);
4811 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4813 struct inode *inode = dentry->d_inode;
4814 struct btrfs_root *root = BTRFS_I(inode)->root;
4817 if (btrfs_root_readonly(root))
4820 err = inode_change_ok(inode, attr);
4824 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4825 err = btrfs_setsize(inode, attr);
4830 if (attr->ia_valid) {
4831 setattr_copy(inode, attr);
4832 inode_inc_iversion(inode);
4833 err = btrfs_dirty_inode(inode);
4835 if (!err && attr->ia_valid & ATTR_MODE)
4836 err = posix_acl_chmod(inode, inode->i_mode);
4843 * While truncating the inode pages during eviction, we get the VFS calling
4844 * btrfs_invalidatepage() against each page of the inode. This is slow because
4845 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4846 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4847 * extent_state structures over and over, wasting lots of time.
4849 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4850 * those expensive operations on a per page basis and do only the ordered io
4851 * finishing, while we release here the extent_map and extent_state structures,
4852 * without the excessive merging and splitting.
4854 static void evict_inode_truncate_pages(struct inode *inode)
4856 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4857 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4858 struct rb_node *node;
4860 ASSERT(inode->i_state & I_FREEING);
4861 truncate_inode_pages_final(&inode->i_data);
4863 write_lock(&map_tree->lock);
4864 while (!RB_EMPTY_ROOT(&map_tree->map)) {
4865 struct extent_map *em;
4867 node = rb_first(&map_tree->map);
4868 em = rb_entry(node, struct extent_map, rb_node);
4869 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4870 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4871 remove_extent_mapping(map_tree, em);
4872 free_extent_map(em);
4873 if (need_resched()) {
4874 write_unlock(&map_tree->lock);
4876 write_lock(&map_tree->lock);
4879 write_unlock(&map_tree->lock);
4881 spin_lock(&io_tree->lock);
4882 while (!RB_EMPTY_ROOT(&io_tree->state)) {
4883 struct extent_state *state;
4884 struct extent_state *cached_state = NULL;
4886 node = rb_first(&io_tree->state);
4887 state = rb_entry(node, struct extent_state, rb_node);
4888 atomic_inc(&state->refs);
4889 spin_unlock(&io_tree->lock);
4891 lock_extent_bits(io_tree, state->start, state->end,
4893 clear_extent_bit(io_tree, state->start, state->end,
4894 EXTENT_LOCKED | EXTENT_DIRTY |
4895 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
4896 EXTENT_DEFRAG, 1, 1,
4897 &cached_state, GFP_NOFS);
4898 free_extent_state(state);
4901 spin_lock(&io_tree->lock);
4903 spin_unlock(&io_tree->lock);
4906 void btrfs_evict_inode(struct inode *inode)
4908 struct btrfs_trans_handle *trans;
4909 struct btrfs_root *root = BTRFS_I(inode)->root;
4910 struct btrfs_block_rsv *rsv, *global_rsv;
4911 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4914 trace_btrfs_inode_evict(inode);
4916 evict_inode_truncate_pages(inode);
4918 if (inode->i_nlink &&
4919 ((btrfs_root_refs(&root->root_item) != 0 &&
4920 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4921 btrfs_is_free_space_inode(inode)))
4924 if (is_bad_inode(inode)) {
4925 btrfs_orphan_del(NULL, inode);
4928 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4929 btrfs_wait_ordered_range(inode, 0, (u64)-1);
4931 btrfs_free_io_failure_record(inode, 0, (u64)-1);
4933 if (root->fs_info->log_root_recovering) {
4934 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4935 &BTRFS_I(inode)->runtime_flags));
4939 if (inode->i_nlink > 0) {
4940 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4941 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
4945 ret = btrfs_commit_inode_delayed_inode(inode);
4947 btrfs_orphan_del(NULL, inode);
4951 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4953 btrfs_orphan_del(NULL, inode);
4956 rsv->size = min_size;
4958 global_rsv = &root->fs_info->global_block_rsv;
4960 btrfs_i_size_write(inode, 0);
4963 * This is a bit simpler than btrfs_truncate since we've already
4964 * reserved our space for our orphan item in the unlink, so we just
4965 * need to reserve some slack space in case we add bytes and update
4966 * inode item when doing the truncate.
4969 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4970 BTRFS_RESERVE_FLUSH_LIMIT);
4973 * Try and steal from the global reserve since we will
4974 * likely not use this space anyway, we want to try as
4975 * hard as possible to get this to work.
4978 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4981 btrfs_warn(root->fs_info,
4982 "Could not get space for a delete, will truncate on mount %d",
4984 btrfs_orphan_del(NULL, inode);
4985 btrfs_free_block_rsv(root, rsv);
4989 trans = btrfs_join_transaction(root);
4990 if (IS_ERR(trans)) {
4991 btrfs_orphan_del(NULL, inode);
4992 btrfs_free_block_rsv(root, rsv);
4996 trans->block_rsv = rsv;
4998 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5002 trans->block_rsv = &root->fs_info->trans_block_rsv;
5003 btrfs_end_transaction(trans, root);
5005 btrfs_btree_balance_dirty(root);
5008 btrfs_free_block_rsv(root, rsv);
5011 * Errors here aren't a big deal, it just means we leave orphan items
5012 * in the tree. They will be cleaned up on the next mount.
5015 trans->block_rsv = root->orphan_block_rsv;
5016 btrfs_orphan_del(trans, inode);
5018 btrfs_orphan_del(NULL, inode);
5021 trans->block_rsv = &root->fs_info->trans_block_rsv;
5022 if (!(root == root->fs_info->tree_root ||
5023 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5024 btrfs_return_ino(root, btrfs_ino(inode));
5026 btrfs_end_transaction(trans, root);
5027 btrfs_btree_balance_dirty(root);
5029 btrfs_remove_delayed_node(inode);
5035 * this returns the key found in the dir entry in the location pointer.
5036 * If no dir entries were found, location->objectid is 0.
5038 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5039 struct btrfs_key *location)
5041 const char *name = dentry->d_name.name;
5042 int namelen = dentry->d_name.len;
5043 struct btrfs_dir_item *di;
5044 struct btrfs_path *path;
5045 struct btrfs_root *root = BTRFS_I(dir)->root;
5048 path = btrfs_alloc_path();
5052 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
5057 if (IS_ERR_OR_NULL(di))
5060 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5062 btrfs_free_path(path);
5065 location->objectid = 0;
5070 * when we hit a tree root in a directory, the btrfs part of the inode
5071 * needs to be changed to reflect the root directory of the tree root. This
5072 * is kind of like crossing a mount point.
5074 static int fixup_tree_root_location(struct btrfs_root *root,
5076 struct dentry *dentry,
5077 struct btrfs_key *location,
5078 struct btrfs_root **sub_root)
5080 struct btrfs_path *path;
5081 struct btrfs_root *new_root;
5082 struct btrfs_root_ref *ref;
5083 struct extent_buffer *leaf;
5084 struct btrfs_key key;
5088 path = btrfs_alloc_path();
5095 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5096 key.type = BTRFS_ROOT_REF_KEY;
5097 key.offset = location->objectid;
5099 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5107 leaf = path->nodes[0];
5108 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5109 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5110 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5113 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5114 (unsigned long)(ref + 1),
5115 dentry->d_name.len);
5119 btrfs_release_path(path);
5121 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5122 if (IS_ERR(new_root)) {
5123 err = PTR_ERR(new_root);
5127 *sub_root = new_root;
5128 location->objectid = btrfs_root_dirid(&new_root->root_item);
5129 location->type = BTRFS_INODE_ITEM_KEY;
5130 location->offset = 0;
5133 btrfs_free_path(path);
5137 static void inode_tree_add(struct inode *inode)
5139 struct btrfs_root *root = BTRFS_I(inode)->root;
5140 struct btrfs_inode *entry;
5142 struct rb_node *parent;
5143 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5144 u64 ino = btrfs_ino(inode);
5146 if (inode_unhashed(inode))
5149 spin_lock(&root->inode_lock);
5150 p = &root->inode_tree.rb_node;
5153 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5155 if (ino < btrfs_ino(&entry->vfs_inode))
5156 p = &parent->rb_left;
5157 else if (ino > btrfs_ino(&entry->vfs_inode))
5158 p = &parent->rb_right;
5160 WARN_ON(!(entry->vfs_inode.i_state &
5161 (I_WILL_FREE | I_FREEING)));
5162 rb_replace_node(parent, new, &root->inode_tree);
5163 RB_CLEAR_NODE(parent);
5164 spin_unlock(&root->inode_lock);
5168 rb_link_node(new, parent, p);
5169 rb_insert_color(new, &root->inode_tree);
5170 spin_unlock(&root->inode_lock);
5173 static void inode_tree_del(struct inode *inode)
5175 struct btrfs_root *root = BTRFS_I(inode)->root;
5178 spin_lock(&root->inode_lock);
5179 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5180 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5181 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5182 empty = RB_EMPTY_ROOT(&root->inode_tree);
5184 spin_unlock(&root->inode_lock);
5186 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5187 synchronize_srcu(&root->fs_info->subvol_srcu);
5188 spin_lock(&root->inode_lock);
5189 empty = RB_EMPTY_ROOT(&root->inode_tree);
5190 spin_unlock(&root->inode_lock);
5192 btrfs_add_dead_root(root);
5196 void btrfs_invalidate_inodes(struct btrfs_root *root)
5198 struct rb_node *node;
5199 struct rb_node *prev;
5200 struct btrfs_inode *entry;
5201 struct inode *inode;
5204 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5205 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5207 spin_lock(&root->inode_lock);
5209 node = root->inode_tree.rb_node;
5213 entry = rb_entry(node, struct btrfs_inode, rb_node);
5215 if (objectid < btrfs_ino(&entry->vfs_inode))
5216 node = node->rb_left;
5217 else if (objectid > btrfs_ino(&entry->vfs_inode))
5218 node = node->rb_right;
5224 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5225 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5229 prev = rb_next(prev);
5233 entry = rb_entry(node, struct btrfs_inode, rb_node);
5234 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5235 inode = igrab(&entry->vfs_inode);
5237 spin_unlock(&root->inode_lock);
5238 if (atomic_read(&inode->i_count) > 1)
5239 d_prune_aliases(inode);
5241 * btrfs_drop_inode will have it removed from
5242 * the inode cache when its usage count
5247 spin_lock(&root->inode_lock);
5251 if (cond_resched_lock(&root->inode_lock))
5254 node = rb_next(node);
5256 spin_unlock(&root->inode_lock);
5259 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5261 struct btrfs_iget_args *args = p;
5262 inode->i_ino = args->location->objectid;
5263 memcpy(&BTRFS_I(inode)->location, args->location,
5264 sizeof(*args->location));
5265 BTRFS_I(inode)->root = args->root;
5269 static int btrfs_find_actor(struct inode *inode, void *opaque)
5271 struct btrfs_iget_args *args = opaque;
5272 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5273 args->root == BTRFS_I(inode)->root;
5276 static struct inode *btrfs_iget_locked(struct super_block *s,
5277 struct btrfs_key *location,
5278 struct btrfs_root *root)
5280 struct inode *inode;
5281 struct btrfs_iget_args args;
5282 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5284 args.location = location;
5287 inode = iget5_locked(s, hashval, btrfs_find_actor,
5288 btrfs_init_locked_inode,
5293 /* Get an inode object given its location and corresponding root.
5294 * Returns in *is_new if the inode was read from disk
5296 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5297 struct btrfs_root *root, int *new)
5299 struct inode *inode;
5301 inode = btrfs_iget_locked(s, location, root);
5303 return ERR_PTR(-ENOMEM);
5305 if (inode->i_state & I_NEW) {
5306 btrfs_read_locked_inode(inode);
5307 if (!is_bad_inode(inode)) {
5308 inode_tree_add(inode);
5309 unlock_new_inode(inode);
5313 unlock_new_inode(inode);
5315 inode = ERR_PTR(-ESTALE);
5322 static struct inode *new_simple_dir(struct super_block *s,
5323 struct btrfs_key *key,
5324 struct btrfs_root *root)
5326 struct inode *inode = new_inode(s);
5329 return ERR_PTR(-ENOMEM);
5331 BTRFS_I(inode)->root = root;
5332 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5333 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5335 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5336 inode->i_op = &btrfs_dir_ro_inode_operations;
5337 inode->i_fop = &simple_dir_operations;
5338 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5339 inode->i_mtime = CURRENT_TIME;
5340 inode->i_atime = inode->i_mtime;
5341 inode->i_ctime = inode->i_mtime;
5342 BTRFS_I(inode)->i_otime = inode->i_mtime;
5347 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5349 struct inode *inode;
5350 struct btrfs_root *root = BTRFS_I(dir)->root;
5351 struct btrfs_root *sub_root = root;
5352 struct btrfs_key location;
5356 if (dentry->d_name.len > BTRFS_NAME_LEN)
5357 return ERR_PTR(-ENAMETOOLONG);
5359 ret = btrfs_inode_by_name(dir, dentry, &location);
5361 return ERR_PTR(ret);
5363 if (location.objectid == 0)
5364 return ERR_PTR(-ENOENT);
5366 if (location.type == BTRFS_INODE_ITEM_KEY) {
5367 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5371 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5373 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5374 ret = fixup_tree_root_location(root, dir, dentry,
5375 &location, &sub_root);
5378 inode = ERR_PTR(ret);
5380 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5382 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5384 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5386 if (!IS_ERR(inode) && root != sub_root) {
5387 down_read(&root->fs_info->cleanup_work_sem);
5388 if (!(inode->i_sb->s_flags & MS_RDONLY))
5389 ret = btrfs_orphan_cleanup(sub_root);
5390 up_read(&root->fs_info->cleanup_work_sem);
5393 inode = ERR_PTR(ret);
5400 static int btrfs_dentry_delete(const struct dentry *dentry)
5402 struct btrfs_root *root;
5403 struct inode *inode = dentry->d_inode;
5405 if (!inode && !IS_ROOT(dentry))
5406 inode = dentry->d_parent->d_inode;
5409 root = BTRFS_I(inode)->root;
5410 if (btrfs_root_refs(&root->root_item) == 0)
5413 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5419 static void btrfs_dentry_release(struct dentry *dentry)
5421 kfree(dentry->d_fsdata);
5424 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5427 struct inode *inode;
5429 inode = btrfs_lookup_dentry(dir, dentry);
5430 if (IS_ERR(inode)) {
5431 if (PTR_ERR(inode) == -ENOENT)
5434 return ERR_CAST(inode);
5437 return d_splice_alias(inode, dentry);
5440 unsigned char btrfs_filetype_table[] = {
5441 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5444 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5446 struct inode *inode = file_inode(file);
5447 struct btrfs_root *root = BTRFS_I(inode)->root;
5448 struct btrfs_item *item;
5449 struct btrfs_dir_item *di;
5450 struct btrfs_key key;
5451 struct btrfs_key found_key;
5452 struct btrfs_path *path;
5453 struct list_head ins_list;
5454 struct list_head del_list;
5456 struct extent_buffer *leaf;
5458 unsigned char d_type;
5463 int key_type = BTRFS_DIR_INDEX_KEY;
5467 int is_curr = 0; /* ctx->pos points to the current index? */
5469 /* FIXME, use a real flag for deciding about the key type */
5470 if (root->fs_info->tree_root == root)
5471 key_type = BTRFS_DIR_ITEM_KEY;
5473 if (!dir_emit_dots(file, ctx))
5476 path = btrfs_alloc_path();
5482 if (key_type == BTRFS_DIR_INDEX_KEY) {
5483 INIT_LIST_HEAD(&ins_list);
5484 INIT_LIST_HEAD(&del_list);
5485 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5488 key.type = key_type;
5489 key.offset = ctx->pos;
5490 key.objectid = btrfs_ino(inode);
5492 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5497 leaf = path->nodes[0];
5498 slot = path->slots[0];
5499 if (slot >= btrfs_header_nritems(leaf)) {
5500 ret = btrfs_next_leaf(root, path);
5508 item = btrfs_item_nr(slot);
5509 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5511 if (found_key.objectid != key.objectid)
5513 if (found_key.type != key_type)
5515 if (found_key.offset < ctx->pos)
5517 if (key_type == BTRFS_DIR_INDEX_KEY &&
5518 btrfs_should_delete_dir_index(&del_list,
5522 ctx->pos = found_key.offset;
5525 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5527 di_total = btrfs_item_size(leaf, item);
5529 while (di_cur < di_total) {
5530 struct btrfs_key location;
5532 if (verify_dir_item(root, leaf, di))
5535 name_len = btrfs_dir_name_len(leaf, di);
5536 if (name_len <= sizeof(tmp_name)) {
5537 name_ptr = tmp_name;
5539 name_ptr = kmalloc(name_len, GFP_NOFS);
5545 read_extent_buffer(leaf, name_ptr,
5546 (unsigned long)(di + 1), name_len);
5548 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5549 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5552 /* is this a reference to our own snapshot? If so
5555 * In contrast to old kernels, we insert the snapshot's
5556 * dir item and dir index after it has been created, so
5557 * we won't find a reference to our own snapshot. We
5558 * still keep the following code for backward
5561 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5562 location.objectid == root->root_key.objectid) {
5566 over = !dir_emit(ctx, name_ptr, name_len,
5567 location.objectid, d_type);
5570 if (name_ptr != tmp_name)
5575 di_len = btrfs_dir_name_len(leaf, di) +
5576 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5578 di = (struct btrfs_dir_item *)((char *)di + di_len);
5584 if (key_type == BTRFS_DIR_INDEX_KEY) {
5587 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5592 /* Reached end of directory/root. Bump pos past the last item. */
5596 * Stop new entries from being returned after we return the last
5599 * New directory entries are assigned a strictly increasing
5600 * offset. This means that new entries created during readdir
5601 * are *guaranteed* to be seen in the future by that readdir.
5602 * This has broken buggy programs which operate on names as
5603 * they're returned by readdir. Until we re-use freed offsets
5604 * we have this hack to stop new entries from being returned
5605 * under the assumption that they'll never reach this huge
5608 * This is being careful not to overflow 32bit loff_t unless the
5609 * last entry requires it because doing so has broken 32bit apps
5612 if (key_type == BTRFS_DIR_INDEX_KEY) {
5613 if (ctx->pos >= INT_MAX)
5614 ctx->pos = LLONG_MAX;
5621 if (key_type == BTRFS_DIR_INDEX_KEY)
5622 btrfs_put_delayed_items(&ins_list, &del_list);
5623 btrfs_free_path(path);
5627 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5629 struct btrfs_root *root = BTRFS_I(inode)->root;
5630 struct btrfs_trans_handle *trans;
5632 bool nolock = false;
5634 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5637 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5640 if (wbc->sync_mode == WB_SYNC_ALL) {
5642 trans = btrfs_join_transaction_nolock(root);
5644 trans = btrfs_join_transaction(root);
5646 return PTR_ERR(trans);
5647 ret = btrfs_commit_transaction(trans, root);
5653 * This is somewhat expensive, updating the tree every time the
5654 * inode changes. But, it is most likely to find the inode in cache.
5655 * FIXME, needs more benchmarking...there are no reasons other than performance
5656 * to keep or drop this code.
5658 static int btrfs_dirty_inode(struct inode *inode)
5660 struct btrfs_root *root = BTRFS_I(inode)->root;
5661 struct btrfs_trans_handle *trans;
5664 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5667 trans = btrfs_join_transaction(root);
5669 return PTR_ERR(trans);
5671 ret = btrfs_update_inode(trans, root, inode);
5672 if (ret && ret == -ENOSPC) {
5673 /* whoops, lets try again with the full transaction */
5674 btrfs_end_transaction(trans, root);
5675 trans = btrfs_start_transaction(root, 1);
5677 return PTR_ERR(trans);
5679 ret = btrfs_update_inode(trans, root, inode);
5681 btrfs_end_transaction(trans, root);
5682 if (BTRFS_I(inode)->delayed_node)
5683 btrfs_balance_delayed_items(root);
5689 * This is a copy of file_update_time. We need this so we can return error on
5690 * ENOSPC for updating the inode in the case of file write and mmap writes.
5692 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5695 struct btrfs_root *root = BTRFS_I(inode)->root;
5697 if (btrfs_root_readonly(root))
5700 if (flags & S_VERSION)
5701 inode_inc_iversion(inode);
5702 if (flags & S_CTIME)
5703 inode->i_ctime = *now;
5704 if (flags & S_MTIME)
5705 inode->i_mtime = *now;
5706 if (flags & S_ATIME)
5707 inode->i_atime = *now;
5708 return btrfs_dirty_inode(inode);
5712 * find the highest existing sequence number in a directory
5713 * and then set the in-memory index_cnt variable to reflect
5714 * free sequence numbers
5716 static int btrfs_set_inode_index_count(struct inode *inode)
5718 struct btrfs_root *root = BTRFS_I(inode)->root;
5719 struct btrfs_key key, found_key;
5720 struct btrfs_path *path;
5721 struct extent_buffer *leaf;
5724 key.objectid = btrfs_ino(inode);
5725 key.type = BTRFS_DIR_INDEX_KEY;
5726 key.offset = (u64)-1;
5728 path = btrfs_alloc_path();
5732 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5735 /* FIXME: we should be able to handle this */
5741 * MAGIC NUMBER EXPLANATION:
5742 * since we search a directory based on f_pos we have to start at 2
5743 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5744 * else has to start at 2
5746 if (path->slots[0] == 0) {
5747 BTRFS_I(inode)->index_cnt = 2;
5753 leaf = path->nodes[0];
5754 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5756 if (found_key.objectid != btrfs_ino(inode) ||
5757 found_key.type != BTRFS_DIR_INDEX_KEY) {
5758 BTRFS_I(inode)->index_cnt = 2;
5762 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5764 btrfs_free_path(path);
5769 * helper to find a free sequence number in a given directory. This current
5770 * code is very simple, later versions will do smarter things in the btree
5772 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5776 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5777 ret = btrfs_inode_delayed_dir_index_count(dir);
5779 ret = btrfs_set_inode_index_count(dir);
5785 *index = BTRFS_I(dir)->index_cnt;
5786 BTRFS_I(dir)->index_cnt++;
5791 static int btrfs_insert_inode_locked(struct inode *inode)
5793 struct btrfs_iget_args args;
5794 args.location = &BTRFS_I(inode)->location;
5795 args.root = BTRFS_I(inode)->root;
5797 return insert_inode_locked4(inode,
5798 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
5799 btrfs_find_actor, &args);
5802 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5803 struct btrfs_root *root,
5805 const char *name, int name_len,
5806 u64 ref_objectid, u64 objectid,
5807 umode_t mode, u64 *index)
5809 struct inode *inode;
5810 struct btrfs_inode_item *inode_item;
5811 struct btrfs_key *location;
5812 struct btrfs_path *path;
5813 struct btrfs_inode_ref *ref;
5814 struct btrfs_key key[2];
5816 int nitems = name ? 2 : 1;
5820 path = btrfs_alloc_path();
5822 return ERR_PTR(-ENOMEM);
5824 inode = new_inode(root->fs_info->sb);
5826 btrfs_free_path(path);
5827 return ERR_PTR(-ENOMEM);
5831 * O_TMPFILE, set link count to 0, so that after this point,
5832 * we fill in an inode item with the correct link count.
5835 set_nlink(inode, 0);
5838 * we have to initialize this early, so we can reclaim the inode
5839 * number if we fail afterwards in this function.
5841 inode->i_ino = objectid;
5844 trace_btrfs_inode_request(dir);
5846 ret = btrfs_set_inode_index(dir, index);
5848 btrfs_free_path(path);
5850 return ERR_PTR(ret);
5856 * index_cnt is ignored for everything but a dir,
5857 * btrfs_get_inode_index_count has an explanation for the magic
5860 BTRFS_I(inode)->index_cnt = 2;
5861 BTRFS_I(inode)->dir_index = *index;
5862 BTRFS_I(inode)->root = root;
5863 BTRFS_I(inode)->generation = trans->transid;
5864 inode->i_generation = BTRFS_I(inode)->generation;
5867 * We could have gotten an inode number from somebody who was fsynced
5868 * and then removed in this same transaction, so let's just set full
5869 * sync since it will be a full sync anyway and this will blow away the
5870 * old info in the log.
5872 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5874 key[0].objectid = objectid;
5875 key[0].type = BTRFS_INODE_ITEM_KEY;
5878 sizes[0] = sizeof(struct btrfs_inode_item);
5882 * Start new inodes with an inode_ref. This is slightly more
5883 * efficient for small numbers of hard links since they will
5884 * be packed into one item. Extended refs will kick in if we
5885 * add more hard links than can fit in the ref item.
5887 key[1].objectid = objectid;
5888 key[1].type = BTRFS_INODE_REF_KEY;
5889 key[1].offset = ref_objectid;
5891 sizes[1] = name_len + sizeof(*ref);
5894 location = &BTRFS_I(inode)->location;
5895 location->objectid = objectid;
5896 location->offset = 0;
5897 location->type = BTRFS_INODE_ITEM_KEY;
5899 ret = btrfs_insert_inode_locked(inode);
5903 path->leave_spinning = 1;
5904 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
5908 inode_init_owner(inode, dir, mode);
5909 inode_set_bytes(inode, 0);
5911 inode->i_mtime = CURRENT_TIME;
5912 inode->i_atime = inode->i_mtime;
5913 inode->i_ctime = inode->i_mtime;
5914 BTRFS_I(inode)->i_otime = inode->i_mtime;
5916 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5917 struct btrfs_inode_item);
5918 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5919 sizeof(*inode_item));
5920 fill_inode_item(trans, path->nodes[0], inode_item, inode);
5923 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5924 struct btrfs_inode_ref);
5925 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5926 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5927 ptr = (unsigned long)(ref + 1);
5928 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5931 btrfs_mark_buffer_dirty(path->nodes[0]);
5932 btrfs_free_path(path);
5934 btrfs_inherit_iflags(inode, dir);
5936 if (S_ISREG(mode)) {
5937 if (btrfs_test_opt(root, NODATASUM))
5938 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5939 if (btrfs_test_opt(root, NODATACOW))
5940 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5941 BTRFS_INODE_NODATASUM;
5944 inode_tree_add(inode);
5946 trace_btrfs_inode_new(inode);
5947 btrfs_set_inode_last_trans(trans, inode);
5949 btrfs_update_root_times(trans, root);
5951 ret = btrfs_inode_inherit_props(trans, inode, dir);
5953 btrfs_err(root->fs_info,
5954 "error inheriting props for ino %llu (root %llu): %d",
5955 btrfs_ino(inode), root->root_key.objectid, ret);
5960 unlock_new_inode(inode);
5963 BTRFS_I(dir)->index_cnt--;
5964 btrfs_free_path(path);
5966 return ERR_PTR(ret);
5969 static inline u8 btrfs_inode_type(struct inode *inode)
5971 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5975 * utility function to add 'inode' into 'parent_inode' with
5976 * a give name and a given sequence number.
5977 * if 'add_backref' is true, also insert a backref from the
5978 * inode to the parent directory.
5980 int btrfs_add_link(struct btrfs_trans_handle *trans,
5981 struct inode *parent_inode, struct inode *inode,
5982 const char *name, int name_len, int add_backref, u64 index)
5985 struct btrfs_key key;
5986 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5987 u64 ino = btrfs_ino(inode);
5988 u64 parent_ino = btrfs_ino(parent_inode);
5990 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5991 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5994 key.type = BTRFS_INODE_ITEM_KEY;
5998 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5999 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6000 key.objectid, root->root_key.objectid,
6001 parent_ino, index, name, name_len);
6002 } else if (add_backref) {
6003 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6007 /* Nothing to clean up yet */
6011 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6013 btrfs_inode_type(inode), index);
6014 if (ret == -EEXIST || ret == -EOVERFLOW)
6017 btrfs_abort_transaction(trans, root, ret);
6021 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6023 inode_inc_iversion(parent_inode);
6024 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
6025 ret = btrfs_update_inode(trans, root, parent_inode);
6027 btrfs_abort_transaction(trans, root, ret);
6031 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6034 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6035 key.objectid, root->root_key.objectid,
6036 parent_ino, &local_index, name, name_len);
6038 } else if (add_backref) {
6042 err = btrfs_del_inode_ref(trans, root, name, name_len,
6043 ino, parent_ino, &local_index);
6048 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6049 struct inode *dir, struct dentry *dentry,
6050 struct inode *inode, int backref, u64 index)
6052 int err = btrfs_add_link(trans, dir, inode,
6053 dentry->d_name.name, dentry->d_name.len,
6060 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6061 umode_t mode, dev_t rdev)
6063 struct btrfs_trans_handle *trans;
6064 struct btrfs_root *root = BTRFS_I(dir)->root;
6065 struct inode *inode = NULL;
6071 if (!new_valid_dev(rdev))
6075 * 2 for inode item and ref
6077 * 1 for xattr if selinux is on
6079 trans = btrfs_start_transaction(root, 5);
6081 return PTR_ERR(trans);
6083 err = btrfs_find_free_ino(root, &objectid);
6087 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6088 dentry->d_name.len, btrfs_ino(dir), objectid,
6090 if (IS_ERR(inode)) {
6091 err = PTR_ERR(inode);
6096 * If the active LSM wants to access the inode during
6097 * d_instantiate it needs these. Smack checks to see
6098 * if the filesystem supports xattrs by looking at the
6101 inode->i_op = &btrfs_special_inode_operations;
6102 init_special_inode(inode, inode->i_mode, rdev);
6104 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6106 goto out_unlock_inode;
6108 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6110 goto out_unlock_inode;
6112 btrfs_update_inode(trans, root, inode);
6113 unlock_new_inode(inode);
6114 d_instantiate(dentry, inode);
6118 btrfs_end_transaction(trans, root);
6119 btrfs_balance_delayed_items(root);
6120 btrfs_btree_balance_dirty(root);
6122 inode_dec_link_count(inode);
6129 unlock_new_inode(inode);
6134 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6135 umode_t mode, bool excl)
6137 struct btrfs_trans_handle *trans;
6138 struct btrfs_root *root = BTRFS_I(dir)->root;
6139 struct inode *inode = NULL;
6140 int drop_inode_on_err = 0;
6146 * 2 for inode item and ref
6148 * 1 for xattr if selinux is on
6150 trans = btrfs_start_transaction(root, 5);
6152 return PTR_ERR(trans);
6154 err = btrfs_find_free_ino(root, &objectid);
6158 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6159 dentry->d_name.len, btrfs_ino(dir), objectid,
6161 if (IS_ERR(inode)) {
6162 err = PTR_ERR(inode);
6165 drop_inode_on_err = 1;
6167 * If the active LSM wants to access the inode during
6168 * d_instantiate it needs these. Smack checks to see
6169 * if the filesystem supports xattrs by looking at the
6172 inode->i_fop = &btrfs_file_operations;
6173 inode->i_op = &btrfs_file_inode_operations;
6174 inode->i_mapping->a_ops = &btrfs_aops;
6175 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
6177 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6179 goto out_unlock_inode;
6181 err = btrfs_update_inode(trans, root, inode);
6183 goto out_unlock_inode;
6185 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6187 goto out_unlock_inode;
6189 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6190 unlock_new_inode(inode);
6191 d_instantiate(dentry, inode);
6194 btrfs_end_transaction(trans, root);
6195 if (err && drop_inode_on_err) {
6196 inode_dec_link_count(inode);
6199 btrfs_balance_delayed_items(root);
6200 btrfs_btree_balance_dirty(root);
6204 unlock_new_inode(inode);
6209 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6210 struct dentry *dentry)
6212 struct btrfs_trans_handle *trans;
6213 struct btrfs_root *root = BTRFS_I(dir)->root;
6214 struct inode *inode = old_dentry->d_inode;
6219 /* do not allow sys_link's with other subvols of the same device */
6220 if (root->objectid != BTRFS_I(inode)->root->objectid)
6223 if (inode->i_nlink >= BTRFS_LINK_MAX)
6226 err = btrfs_set_inode_index(dir, &index);
6231 * 2 items for inode and inode ref
6232 * 2 items for dir items
6233 * 1 item for parent inode
6235 trans = btrfs_start_transaction(root, 5);
6236 if (IS_ERR(trans)) {
6237 err = PTR_ERR(trans);
6241 /* There are several dir indexes for this inode, clear the cache. */
6242 BTRFS_I(inode)->dir_index = 0ULL;
6244 inode_inc_iversion(inode);
6245 inode->i_ctime = CURRENT_TIME;
6247 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6249 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6254 struct dentry *parent = dentry->d_parent;
6255 err = btrfs_update_inode(trans, root, inode);
6258 if (inode->i_nlink == 1) {
6260 * If new hard link count is 1, it's a file created
6261 * with open(2) O_TMPFILE flag.
6263 err = btrfs_orphan_del(trans, inode);
6267 d_instantiate(dentry, inode);
6268 btrfs_log_new_name(trans, inode, NULL, parent);
6271 btrfs_end_transaction(trans, root);
6272 btrfs_balance_delayed_items(root);
6275 inode_dec_link_count(inode);
6278 btrfs_btree_balance_dirty(root);
6282 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6284 struct inode *inode = NULL;
6285 struct btrfs_trans_handle *trans;
6286 struct btrfs_root *root = BTRFS_I(dir)->root;
6288 int drop_on_err = 0;
6293 * 2 items for inode and ref
6294 * 2 items for dir items
6295 * 1 for xattr if selinux is on
6297 trans = btrfs_start_transaction(root, 5);
6299 return PTR_ERR(trans);
6301 err = btrfs_find_free_ino(root, &objectid);
6305 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6306 dentry->d_name.len, btrfs_ino(dir), objectid,
6307 S_IFDIR | mode, &index);
6308 if (IS_ERR(inode)) {
6309 err = PTR_ERR(inode);
6314 /* these must be set before we unlock the inode */
6315 inode->i_op = &btrfs_dir_inode_operations;
6316 inode->i_fop = &btrfs_dir_file_operations;
6318 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6320 goto out_fail_inode;
6322 btrfs_i_size_write(inode, 0);
6323 err = btrfs_update_inode(trans, root, inode);
6325 goto out_fail_inode;
6327 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6328 dentry->d_name.len, 0, index);
6330 goto out_fail_inode;
6332 d_instantiate(dentry, inode);
6334 * mkdir is special. We're unlocking after we call d_instantiate
6335 * to avoid a race with nfsd calling d_instantiate.
6337 unlock_new_inode(inode);
6341 btrfs_end_transaction(trans, root);
6343 inode_dec_link_count(inode);
6346 btrfs_balance_delayed_items(root);
6347 btrfs_btree_balance_dirty(root);
6351 unlock_new_inode(inode);
6355 /* Find next extent map of a given extent map, caller needs to ensure locks */
6356 static struct extent_map *next_extent_map(struct extent_map *em)
6358 struct rb_node *next;
6360 next = rb_next(&em->rb_node);
6363 return container_of(next, struct extent_map, rb_node);
6366 static struct extent_map *prev_extent_map(struct extent_map *em)
6368 struct rb_node *prev;
6370 prev = rb_prev(&em->rb_node);
6373 return container_of(prev, struct extent_map, rb_node);
6376 /* helper for btfs_get_extent. Given an existing extent in the tree,
6377 * the existing extent is the nearest extent to map_start,
6378 * and an extent that you want to insert, deal with overlap and insert
6379 * the best fitted new extent into the tree.
6381 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6382 struct extent_map *existing,
6383 struct extent_map *em,
6386 struct extent_map *prev;
6387 struct extent_map *next;
6392 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6394 if (existing->start > map_start) {
6396 prev = prev_extent_map(next);
6399 next = next_extent_map(prev);
6402 start = prev ? extent_map_end(prev) : em->start;
6403 start = max_t(u64, start, em->start);
6404 end = next ? next->start : extent_map_end(em);
6405 end = min_t(u64, end, extent_map_end(em));
6406 start_diff = start - em->start;
6408 em->len = end - start;
6409 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6410 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6411 em->block_start += start_diff;
6412 em->block_len -= start_diff;
6414 return add_extent_mapping(em_tree, em, 0);
6417 static noinline int uncompress_inline(struct btrfs_path *path,
6418 struct inode *inode, struct page *page,
6419 size_t pg_offset, u64 extent_offset,
6420 struct btrfs_file_extent_item *item)
6423 struct extent_buffer *leaf = path->nodes[0];
6426 unsigned long inline_size;
6430 WARN_ON(pg_offset != 0);
6431 compress_type = btrfs_file_extent_compression(leaf, item);
6432 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6433 inline_size = btrfs_file_extent_inline_item_len(leaf,
6434 btrfs_item_nr(path->slots[0]));
6435 tmp = kmalloc(inline_size, GFP_NOFS);
6438 ptr = btrfs_file_extent_inline_start(item);
6440 read_extent_buffer(leaf, tmp, ptr, inline_size);
6442 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6443 ret = btrfs_decompress(compress_type, tmp, page,
6444 extent_offset, inline_size, max_size);
6450 * a bit scary, this does extent mapping from logical file offset to the disk.
6451 * the ugly parts come from merging extents from the disk with the in-ram
6452 * representation. This gets more complex because of the data=ordered code,
6453 * where the in-ram extents might be locked pending data=ordered completion.
6455 * This also copies inline extents directly into the page.
6458 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6459 size_t pg_offset, u64 start, u64 len,
6464 u64 extent_start = 0;
6466 u64 objectid = btrfs_ino(inode);
6468 struct btrfs_path *path = NULL;
6469 struct btrfs_root *root = BTRFS_I(inode)->root;
6470 struct btrfs_file_extent_item *item;
6471 struct extent_buffer *leaf;
6472 struct btrfs_key found_key;
6473 struct extent_map *em = NULL;
6474 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6475 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6476 struct btrfs_trans_handle *trans = NULL;
6477 const bool new_inline = !page || create;
6480 read_lock(&em_tree->lock);
6481 em = lookup_extent_mapping(em_tree, start, len);
6483 em->bdev = root->fs_info->fs_devices->latest_bdev;
6484 read_unlock(&em_tree->lock);
6487 if (em->start > start || em->start + em->len <= start)
6488 free_extent_map(em);
6489 else if (em->block_start == EXTENT_MAP_INLINE && page)
6490 free_extent_map(em);
6494 em = alloc_extent_map();
6499 em->bdev = root->fs_info->fs_devices->latest_bdev;
6500 em->start = EXTENT_MAP_HOLE;
6501 em->orig_start = EXTENT_MAP_HOLE;
6503 em->block_len = (u64)-1;
6506 path = btrfs_alloc_path();
6512 * Chances are we'll be called again, so go ahead and do
6518 ret = btrfs_lookup_file_extent(trans, root, path,
6519 objectid, start, trans != NULL);
6526 if (path->slots[0] == 0)
6531 leaf = path->nodes[0];
6532 item = btrfs_item_ptr(leaf, path->slots[0],
6533 struct btrfs_file_extent_item);
6534 /* are we inside the extent that was found? */
6535 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6536 found_type = found_key.type;
6537 if (found_key.objectid != objectid ||
6538 found_type != BTRFS_EXTENT_DATA_KEY) {
6540 * If we backup past the first extent we want to move forward
6541 * and see if there is an extent in front of us, otherwise we'll
6542 * say there is a hole for our whole search range which can
6549 found_type = btrfs_file_extent_type(leaf, item);
6550 extent_start = found_key.offset;
6551 if (found_type == BTRFS_FILE_EXTENT_REG ||
6552 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6553 extent_end = extent_start +
6554 btrfs_file_extent_num_bytes(leaf, item);
6555 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6557 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6558 extent_end = ALIGN(extent_start + size, root->sectorsize);
6561 if (start >= extent_end) {
6563 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6564 ret = btrfs_next_leaf(root, path);
6571 leaf = path->nodes[0];
6573 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6574 if (found_key.objectid != objectid ||
6575 found_key.type != BTRFS_EXTENT_DATA_KEY)
6577 if (start + len <= found_key.offset)
6579 if (start > found_key.offset)
6582 em->orig_start = start;
6583 em->len = found_key.offset - start;
6587 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6589 if (found_type == BTRFS_FILE_EXTENT_REG ||
6590 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6592 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6596 size_t extent_offset;
6602 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6603 extent_offset = page_offset(page) + pg_offset - extent_start;
6604 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6605 size - extent_offset);
6606 em->start = extent_start + extent_offset;
6607 em->len = ALIGN(copy_size, root->sectorsize);
6608 em->orig_block_len = em->len;
6609 em->orig_start = em->start;
6610 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6611 if (create == 0 && !PageUptodate(page)) {
6612 if (btrfs_file_extent_compression(leaf, item) !=
6613 BTRFS_COMPRESS_NONE) {
6614 ret = uncompress_inline(path, inode, page,
6616 extent_offset, item);
6623 read_extent_buffer(leaf, map + pg_offset, ptr,
6625 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6626 memset(map + pg_offset + copy_size, 0,
6627 PAGE_CACHE_SIZE - pg_offset -
6632 flush_dcache_page(page);
6633 } else if (create && PageUptodate(page)) {
6637 free_extent_map(em);
6640 btrfs_release_path(path);
6641 trans = btrfs_join_transaction(root);
6644 return ERR_CAST(trans);
6648 write_extent_buffer(leaf, map + pg_offset, ptr,
6651 btrfs_mark_buffer_dirty(leaf);
6653 set_extent_uptodate(io_tree, em->start,
6654 extent_map_end(em) - 1, NULL, GFP_NOFS);
6659 em->orig_start = start;
6662 em->block_start = EXTENT_MAP_HOLE;
6663 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6665 btrfs_release_path(path);
6666 if (em->start > start || extent_map_end(em) <= start) {
6667 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6668 em->start, em->len, start, len);
6674 write_lock(&em_tree->lock);
6675 ret = add_extent_mapping(em_tree, em, 0);
6676 /* it is possible that someone inserted the extent into the tree
6677 * while we had the lock dropped. It is also possible that
6678 * an overlapping map exists in the tree
6680 if (ret == -EEXIST) {
6681 struct extent_map *existing;
6685 existing = search_extent_mapping(em_tree, start, len);
6687 * existing will always be non-NULL, since there must be
6688 * extent causing the -EEXIST.
6690 if (start >= extent_map_end(existing) ||
6691 start <= existing->start) {
6693 * The existing extent map is the one nearest to
6694 * the [start, start + len) range which overlaps
6696 err = merge_extent_mapping(em_tree, existing,
6698 free_extent_map(existing);
6700 free_extent_map(em);
6704 free_extent_map(em);
6709 write_unlock(&em_tree->lock);
6712 trace_btrfs_get_extent(root, em);
6715 btrfs_free_path(path);
6717 ret = btrfs_end_transaction(trans, root);
6722 free_extent_map(em);
6723 return ERR_PTR(err);
6725 BUG_ON(!em); /* Error is always set */
6729 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6730 size_t pg_offset, u64 start, u64 len,
6733 struct extent_map *em;
6734 struct extent_map *hole_em = NULL;
6735 u64 range_start = start;
6741 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6748 * - a pre-alloc extent,
6749 * there might actually be delalloc bytes behind it.
6751 if (em->block_start != EXTENT_MAP_HOLE &&
6752 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6758 /* check to see if we've wrapped (len == -1 or similar) */
6767 /* ok, we didn't find anything, lets look for delalloc */
6768 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6769 end, len, EXTENT_DELALLOC, 1);
6770 found_end = range_start + found;
6771 if (found_end < range_start)
6772 found_end = (u64)-1;
6775 * we didn't find anything useful, return
6776 * the original results from get_extent()
6778 if (range_start > end || found_end <= start) {
6784 /* adjust the range_start to make sure it doesn't
6785 * go backwards from the start they passed in
6787 range_start = max(start, range_start);
6788 found = found_end - range_start;
6791 u64 hole_start = start;
6794 em = alloc_extent_map();
6800 * when btrfs_get_extent can't find anything it
6801 * returns one huge hole
6803 * make sure what it found really fits our range, and
6804 * adjust to make sure it is based on the start from
6808 u64 calc_end = extent_map_end(hole_em);
6810 if (calc_end <= start || (hole_em->start > end)) {
6811 free_extent_map(hole_em);
6814 hole_start = max(hole_em->start, start);
6815 hole_len = calc_end - hole_start;
6819 if (hole_em && range_start > hole_start) {
6820 /* our hole starts before our delalloc, so we
6821 * have to return just the parts of the hole
6822 * that go until the delalloc starts
6824 em->len = min(hole_len,
6825 range_start - hole_start);
6826 em->start = hole_start;
6827 em->orig_start = hole_start;
6829 * don't adjust block start at all,
6830 * it is fixed at EXTENT_MAP_HOLE
6832 em->block_start = hole_em->block_start;
6833 em->block_len = hole_len;
6834 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6835 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6837 em->start = range_start;
6839 em->orig_start = range_start;
6840 em->block_start = EXTENT_MAP_DELALLOC;
6841 em->block_len = found;
6843 } else if (hole_em) {
6848 free_extent_map(hole_em);
6850 free_extent_map(em);
6851 return ERR_PTR(err);
6856 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6859 struct btrfs_root *root = BTRFS_I(inode)->root;
6860 struct extent_map *em;
6861 struct btrfs_key ins;
6865 alloc_hint = get_extent_allocation_hint(inode, start, len);
6866 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
6867 alloc_hint, &ins, 1, 1);
6869 return ERR_PTR(ret);
6871 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6872 ins.offset, ins.offset, ins.offset, 0);
6874 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
6878 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6879 ins.offset, ins.offset, 0);
6881 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
6882 free_extent_map(em);
6883 return ERR_PTR(ret);
6890 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6891 * block must be cow'd
6893 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6894 u64 *orig_start, u64 *orig_block_len,
6897 struct btrfs_trans_handle *trans;
6898 struct btrfs_path *path;
6900 struct extent_buffer *leaf;
6901 struct btrfs_root *root = BTRFS_I(inode)->root;
6902 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6903 struct btrfs_file_extent_item *fi;
6904 struct btrfs_key key;
6911 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6913 path = btrfs_alloc_path();
6917 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
6922 slot = path->slots[0];
6925 /* can't find the item, must cow */
6932 leaf = path->nodes[0];
6933 btrfs_item_key_to_cpu(leaf, &key, slot);
6934 if (key.objectid != btrfs_ino(inode) ||
6935 key.type != BTRFS_EXTENT_DATA_KEY) {
6936 /* not our file or wrong item type, must cow */
6940 if (key.offset > offset) {
6941 /* Wrong offset, must cow */
6945 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6946 found_type = btrfs_file_extent_type(leaf, fi);
6947 if (found_type != BTRFS_FILE_EXTENT_REG &&
6948 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6949 /* not a regular extent, must cow */
6953 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6956 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6957 if (extent_end <= offset)
6960 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6961 if (disk_bytenr == 0)
6964 if (btrfs_file_extent_compression(leaf, fi) ||
6965 btrfs_file_extent_encryption(leaf, fi) ||
6966 btrfs_file_extent_other_encoding(leaf, fi))
6969 backref_offset = btrfs_file_extent_offset(leaf, fi);
6972 *orig_start = key.offset - backref_offset;
6973 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6974 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6977 if (btrfs_extent_readonly(root, disk_bytenr))
6980 num_bytes = min(offset + *len, extent_end) - offset;
6981 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6984 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
6985 ret = test_range_bit(io_tree, offset, range_end,
6986 EXTENT_DELALLOC, 0, NULL);
6993 btrfs_release_path(path);
6996 * look for other files referencing this extent, if we
6997 * find any we must cow
6999 trans = btrfs_join_transaction(root);
7000 if (IS_ERR(trans)) {
7005 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7006 key.offset - backref_offset, disk_bytenr);
7007 btrfs_end_transaction(trans, root);
7014 * adjust disk_bytenr and num_bytes to cover just the bytes
7015 * in this extent we are about to write. If there
7016 * are any csums in that range we have to cow in order
7017 * to keep the csums correct
7019 disk_bytenr += backref_offset;
7020 disk_bytenr += offset - key.offset;
7021 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7024 * all of the above have passed, it is safe to overwrite this extent
7030 btrfs_free_path(path);
7034 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7036 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7038 void **pagep = NULL;
7039 struct page *page = NULL;
7043 start_idx = start >> PAGE_CACHE_SHIFT;
7046 * end is the last byte in the last page. end == start is legal
7048 end_idx = end >> PAGE_CACHE_SHIFT;
7052 /* Most of the code in this while loop is lifted from
7053 * find_get_page. It's been modified to begin searching from a
7054 * page and return just the first page found in that range. If the
7055 * found idx is less than or equal to the end idx then we know that
7056 * a page exists. If no pages are found or if those pages are
7057 * outside of the range then we're fine (yay!) */
7058 while (page == NULL &&
7059 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7060 page = radix_tree_deref_slot(pagep);
7061 if (unlikely(!page))
7064 if (radix_tree_exception(page)) {
7065 if (radix_tree_deref_retry(page)) {
7070 * Otherwise, shmem/tmpfs must be storing a swap entry
7071 * here as an exceptional entry: so return it without
7072 * attempting to raise page count.
7075 break; /* TODO: Is this relevant for this use case? */
7078 if (!page_cache_get_speculative(page)) {
7084 * Has the page moved?
7085 * This is part of the lockless pagecache protocol. See
7086 * include/linux/pagemap.h for details.
7088 if (unlikely(page != *pagep)) {
7089 page_cache_release(page);
7095 if (page->index <= end_idx)
7097 page_cache_release(page);
7104 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7105 struct extent_state **cached_state, int writing)
7107 struct btrfs_ordered_extent *ordered;
7111 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7114 * We're concerned with the entire range that we're going to be
7115 * doing DIO to, so we need to make sure theres no ordered
7116 * extents in this range.
7118 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7119 lockend - lockstart + 1);
7122 * We need to make sure there are no buffered pages in this
7123 * range either, we could have raced between the invalidate in
7124 * generic_file_direct_write and locking the extent. The
7125 * invalidate needs to happen so that reads after a write do not
7130 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7133 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7134 cached_state, GFP_NOFS);
7137 btrfs_start_ordered_extent(inode, ordered, 1);
7138 btrfs_put_ordered_extent(ordered);
7140 /* Screw you mmap */
7141 ret = btrfs_fdatawrite_range(inode, lockstart, lockend);
7144 ret = filemap_fdatawait_range(inode->i_mapping,
7151 * If we found a page that couldn't be invalidated just
7152 * fall back to buffered.
7154 ret = invalidate_inode_pages2_range(inode->i_mapping,
7155 lockstart >> PAGE_CACHE_SHIFT,
7156 lockend >> PAGE_CACHE_SHIFT);
7167 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7168 u64 len, u64 orig_start,
7169 u64 block_start, u64 block_len,
7170 u64 orig_block_len, u64 ram_bytes,
7173 struct extent_map_tree *em_tree;
7174 struct extent_map *em;
7175 struct btrfs_root *root = BTRFS_I(inode)->root;
7178 em_tree = &BTRFS_I(inode)->extent_tree;
7179 em = alloc_extent_map();
7181 return ERR_PTR(-ENOMEM);
7184 em->orig_start = orig_start;
7185 em->mod_start = start;
7188 em->block_len = block_len;
7189 em->block_start = block_start;
7190 em->bdev = root->fs_info->fs_devices->latest_bdev;
7191 em->orig_block_len = orig_block_len;
7192 em->ram_bytes = ram_bytes;
7193 em->generation = -1;
7194 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7195 if (type == BTRFS_ORDERED_PREALLOC)
7196 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7199 btrfs_drop_extent_cache(inode, em->start,
7200 em->start + em->len - 1, 0);
7201 write_lock(&em_tree->lock);
7202 ret = add_extent_mapping(em_tree, em, 1);
7203 write_unlock(&em_tree->lock);
7204 } while (ret == -EEXIST);
7207 free_extent_map(em);
7208 return ERR_PTR(ret);
7215 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7216 struct buffer_head *bh_result, int create)
7218 struct extent_map *em;
7219 struct btrfs_root *root = BTRFS_I(inode)->root;
7220 struct extent_state *cached_state = NULL;
7221 u64 start = iblock << inode->i_blkbits;
7222 u64 lockstart, lockend;
7223 u64 len = bh_result->b_size;
7225 int unlock_bits = EXTENT_LOCKED;
7229 unlock_bits |= EXTENT_DIRTY;
7231 len = min_t(u64, len, root->sectorsize);
7234 lockend = start + len - 1;
7237 * If this errors out it's because we couldn't invalidate pagecache for
7238 * this range and we need to fallback to buffered.
7240 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
7243 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7250 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7251 * io. INLINE is special, and we could probably kludge it in here, but
7252 * it's still buffered so for safety lets just fall back to the generic
7255 * For COMPRESSED we _have_ to read the entire extent in so we can
7256 * decompress it, so there will be buffering required no matter what we
7257 * do, so go ahead and fallback to buffered.
7259 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7260 * to buffered IO. Don't blame me, this is the price we pay for using
7263 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7264 em->block_start == EXTENT_MAP_INLINE) {
7265 free_extent_map(em);
7270 /* Just a good old fashioned hole, return */
7271 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7272 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7273 free_extent_map(em);
7278 * We don't allocate a new extent in the following cases
7280 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7282 * 2) The extent is marked as PREALLOC. We're good to go here and can
7283 * just use the extent.
7287 len = min(len, em->len - (start - em->start));
7288 lockstart = start + len;
7292 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7293 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7294 em->block_start != EXTENT_MAP_HOLE)) {
7296 u64 block_start, orig_start, orig_block_len, ram_bytes;
7298 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7299 type = BTRFS_ORDERED_PREALLOC;
7301 type = BTRFS_ORDERED_NOCOW;
7302 len = min(len, em->len - (start - em->start));
7303 block_start = em->block_start + (start - em->start);
7305 if (can_nocow_extent(inode, start, &len, &orig_start,
7306 &orig_block_len, &ram_bytes) == 1) {
7307 if (type == BTRFS_ORDERED_PREALLOC) {
7308 free_extent_map(em);
7309 em = create_pinned_em(inode, start, len,
7320 ret = btrfs_add_ordered_extent_dio(inode, start,
7321 block_start, len, len, type);
7323 free_extent_map(em);
7331 * this will cow the extent, reset the len in case we changed
7334 len = bh_result->b_size;
7335 free_extent_map(em);
7336 em = btrfs_new_extent_direct(inode, start, len);
7341 len = min(len, em->len - (start - em->start));
7343 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7345 bh_result->b_size = len;
7346 bh_result->b_bdev = em->bdev;
7347 set_buffer_mapped(bh_result);
7349 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7350 set_buffer_new(bh_result);
7353 * Need to update the i_size under the extent lock so buffered
7354 * readers will get the updated i_size when we unlock.
7356 if (start + len > i_size_read(inode))
7357 i_size_write(inode, start + len);
7359 if (len < orig_len) {
7360 spin_lock(&BTRFS_I(inode)->lock);
7361 BTRFS_I(inode)->outstanding_extents++;
7362 spin_unlock(&BTRFS_I(inode)->lock);
7364 btrfs_free_reserved_data_space(inode, len);
7368 * In the case of write we need to clear and unlock the entire range,
7369 * in the case of read we need to unlock only the end area that we
7370 * aren't using if there is any left over space.
7372 if (lockstart < lockend) {
7373 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7374 lockend, unlock_bits, 1, 0,
7375 &cached_state, GFP_NOFS);
7377 free_extent_state(cached_state);
7380 free_extent_map(em);
7385 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7386 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7390 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7391 int rw, int mirror_num)
7393 struct btrfs_root *root = BTRFS_I(inode)->root;
7396 BUG_ON(rw & REQ_WRITE);
7400 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7401 BTRFS_WQ_ENDIO_DIO_REPAIR);
7405 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7411 static int btrfs_check_dio_repairable(struct inode *inode,
7412 struct bio *failed_bio,
7413 struct io_failure_record *failrec,
7418 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7419 failrec->logical, failrec->len);
7420 if (num_copies == 1) {
7422 * we only have a single copy of the data, so don't bother with
7423 * all the retry and error correction code that follows. no
7424 * matter what the error is, it is very likely to persist.
7426 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7427 num_copies, failrec->this_mirror, failed_mirror);
7431 failrec->failed_mirror = failed_mirror;
7432 failrec->this_mirror++;
7433 if (failrec->this_mirror == failed_mirror)
7434 failrec->this_mirror++;
7436 if (failrec->this_mirror > num_copies) {
7437 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7438 num_copies, failrec->this_mirror, failed_mirror);
7445 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7446 struct page *page, u64 start, u64 end,
7447 int failed_mirror, bio_end_io_t *repair_endio,
7450 struct io_failure_record *failrec;
7456 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7458 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7462 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7465 free_io_failure(inode, failrec);
7469 if (failed_bio->bi_vcnt > 1)
7470 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7472 read_mode = READ_SYNC;
7474 isector = start - btrfs_io_bio(failed_bio)->logical;
7475 isector >>= inode->i_sb->s_blocksize_bits;
7476 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7477 0, isector, repair_endio, repair_arg);
7479 free_io_failure(inode, failrec);
7483 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7484 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7485 read_mode, failrec->this_mirror, failrec->in_validation);
7487 ret = submit_dio_repair_bio(inode, bio, read_mode,
7488 failrec->this_mirror);
7490 free_io_failure(inode, failrec);
7497 struct btrfs_retry_complete {
7498 struct completion done;
7499 struct inode *inode;
7504 static void btrfs_retry_endio_nocsum(struct bio *bio, int err)
7506 struct btrfs_retry_complete *done = bio->bi_private;
7507 struct bio_vec *bvec;
7514 bio_for_each_segment_all(bvec, bio, i)
7515 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7517 complete(&done->done);
7521 static int __btrfs_correct_data_nocsum(struct inode *inode,
7522 struct btrfs_io_bio *io_bio)
7524 struct bio_vec *bvec;
7525 struct btrfs_retry_complete done;
7530 start = io_bio->logical;
7533 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7537 init_completion(&done.done);
7539 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7540 start + bvec->bv_len - 1,
7542 btrfs_retry_endio_nocsum, &done);
7546 wait_for_completion(&done.done);
7548 if (!done.uptodate) {
7549 /* We might have another mirror, so try again */
7553 start += bvec->bv_len;
7559 static void btrfs_retry_endio(struct bio *bio, int err)
7561 struct btrfs_retry_complete *done = bio->bi_private;
7562 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7563 struct bio_vec *bvec;
7572 bio_for_each_segment_all(bvec, bio, i) {
7573 ret = __readpage_endio_check(done->inode, io_bio, i,
7575 done->start, bvec->bv_len);
7577 clean_io_failure(done->inode, done->start,
7583 done->uptodate = uptodate;
7585 complete(&done->done);
7589 static int __btrfs_subio_endio_read(struct inode *inode,
7590 struct btrfs_io_bio *io_bio, int err)
7592 struct bio_vec *bvec;
7593 struct btrfs_retry_complete done;
7600 start = io_bio->logical;
7603 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7604 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7605 0, start, bvec->bv_len);
7611 init_completion(&done.done);
7613 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7614 start + bvec->bv_len - 1,
7616 btrfs_retry_endio, &done);
7622 wait_for_completion(&done.done);
7624 if (!done.uptodate) {
7625 /* We might have another mirror, so try again */
7629 offset += bvec->bv_len;
7630 start += bvec->bv_len;
7636 static int btrfs_subio_endio_read(struct inode *inode,
7637 struct btrfs_io_bio *io_bio, int err)
7639 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7643 return __btrfs_correct_data_nocsum(inode, io_bio);
7647 return __btrfs_subio_endio_read(inode, io_bio, err);
7651 static void btrfs_endio_direct_read(struct bio *bio, int err)
7653 struct btrfs_dio_private *dip = bio->bi_private;
7654 struct inode *inode = dip->inode;
7655 struct bio *dio_bio;
7656 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7658 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
7659 err = btrfs_subio_endio_read(inode, io_bio, err);
7661 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
7662 dip->logical_offset + dip->bytes - 1);
7663 dio_bio = dip->dio_bio;
7667 /* If we had a csum failure make sure to clear the uptodate flag */
7669 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7670 dio_end_io(dio_bio, err);
7673 io_bio->end_io(io_bio, err);
7677 static void btrfs_endio_direct_write(struct bio *bio, int err)
7679 struct btrfs_dio_private *dip = bio->bi_private;
7680 struct inode *inode = dip->inode;
7681 struct btrfs_root *root = BTRFS_I(inode)->root;
7682 struct btrfs_ordered_extent *ordered = NULL;
7683 u64 ordered_offset = dip->logical_offset;
7684 u64 ordered_bytes = dip->bytes;
7685 struct bio *dio_bio;
7691 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7693 ordered_bytes, !err);
7697 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
7698 finish_ordered_fn, NULL, NULL);
7699 btrfs_queue_work(root->fs_info->endio_write_workers,
7703 * our bio might span multiple ordered extents. If we haven't
7704 * completed the accounting for the whole dio, go back and try again
7706 if (ordered_offset < dip->logical_offset + dip->bytes) {
7707 ordered_bytes = dip->logical_offset + dip->bytes -
7713 dio_bio = dip->dio_bio;
7717 /* If we had an error make sure to clear the uptodate flag */
7719 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7720 dio_end_io(dio_bio, err);
7724 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7725 struct bio *bio, int mirror_num,
7726 unsigned long bio_flags, u64 offset)
7729 struct btrfs_root *root = BTRFS_I(inode)->root;
7730 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7731 BUG_ON(ret); /* -ENOMEM */
7735 static void btrfs_end_dio_bio(struct bio *bio, int err)
7737 struct btrfs_dio_private *dip = bio->bi_private;
7740 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7741 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7742 btrfs_ino(dip->inode), bio->bi_rw,
7743 (unsigned long long)bio->bi_iter.bi_sector,
7744 bio->bi_iter.bi_size, err);
7746 if (dip->subio_endio)
7747 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
7753 * before atomic variable goto zero, we must make sure
7754 * dip->errors is perceived to be set.
7756 smp_mb__before_atomic();
7759 /* if there are more bios still pending for this dio, just exit */
7760 if (!atomic_dec_and_test(&dip->pending_bios))
7764 bio_io_error(dip->orig_bio);
7766 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7767 bio_endio(dip->orig_bio, 0);
7773 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7774 u64 first_sector, gfp_t gfp_flags)
7776 int nr_vecs = bio_get_nr_vecs(bdev);
7777 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7780 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
7781 struct inode *inode,
7782 struct btrfs_dio_private *dip,
7786 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7787 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
7791 * We load all the csum data we need when we submit
7792 * the first bio to reduce the csum tree search and
7795 if (dip->logical_offset == file_offset) {
7796 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
7802 if (bio == dip->orig_bio)
7805 file_offset -= dip->logical_offset;
7806 file_offset >>= inode->i_sb->s_blocksize_bits;
7807 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
7812 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7813 int rw, u64 file_offset, int skip_sum,
7816 struct btrfs_dio_private *dip = bio->bi_private;
7817 int write = rw & REQ_WRITE;
7818 struct btrfs_root *root = BTRFS_I(inode)->root;
7822 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7827 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7828 BTRFS_WQ_ENDIO_DATA);
7836 if (write && async_submit) {
7837 ret = btrfs_wq_submit_bio(root->fs_info,
7838 inode, rw, bio, 0, 0,
7840 __btrfs_submit_bio_start_direct_io,
7841 __btrfs_submit_bio_done);
7845 * If we aren't doing async submit, calculate the csum of the
7848 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7852 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
7858 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7864 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7867 struct inode *inode = dip->inode;
7868 struct btrfs_root *root = BTRFS_I(inode)->root;
7870 struct bio *orig_bio = dip->orig_bio;
7871 struct bio_vec *bvec = orig_bio->bi_io_vec;
7872 u64 start_sector = orig_bio->bi_iter.bi_sector;
7873 u64 file_offset = dip->logical_offset;
7878 int async_submit = 0;
7880 map_length = orig_bio->bi_iter.bi_size;
7881 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7882 &map_length, NULL, 0);
7886 if (map_length >= orig_bio->bi_iter.bi_size) {
7888 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
7892 /* async crcs make it difficult to collect full stripe writes. */
7893 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
7898 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7902 bio->bi_private = dip;
7903 bio->bi_end_io = btrfs_end_dio_bio;
7904 btrfs_io_bio(bio)->logical = file_offset;
7905 atomic_inc(&dip->pending_bios);
7907 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7908 if (map_length < submit_len + bvec->bv_len ||
7909 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7910 bvec->bv_offset) < bvec->bv_len) {
7912 * inc the count before we submit the bio so
7913 * we know the end IO handler won't happen before
7914 * we inc the count. Otherwise, the dip might get freed
7915 * before we're done setting it up
7917 atomic_inc(&dip->pending_bios);
7918 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7919 file_offset, skip_sum,
7923 atomic_dec(&dip->pending_bios);
7927 start_sector += submit_len >> 9;
7928 file_offset += submit_len;
7933 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7934 start_sector, GFP_NOFS);
7937 bio->bi_private = dip;
7938 bio->bi_end_io = btrfs_end_dio_bio;
7939 btrfs_io_bio(bio)->logical = file_offset;
7941 map_length = orig_bio->bi_iter.bi_size;
7942 ret = btrfs_map_block(root->fs_info, rw,
7944 &map_length, NULL, 0);
7950 submit_len += bvec->bv_len;
7957 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7966 * before atomic variable goto zero, we must
7967 * make sure dip->errors is perceived to be set.
7969 smp_mb__before_atomic();
7970 if (atomic_dec_and_test(&dip->pending_bios))
7971 bio_io_error(dip->orig_bio);
7973 /* bio_end_io() will handle error, so we needn't return it */
7977 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7978 struct inode *inode, loff_t file_offset)
7980 struct btrfs_root *root = BTRFS_I(inode)->root;
7981 struct btrfs_dio_private *dip;
7983 struct btrfs_io_bio *btrfs_bio;
7985 int write = rw & REQ_WRITE;
7988 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7990 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7996 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8002 dip->private = dio_bio->bi_private;
8004 dip->logical_offset = file_offset;
8005 dip->bytes = dio_bio->bi_iter.bi_size;
8006 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8007 io_bio->bi_private = dip;
8008 dip->orig_bio = io_bio;
8009 dip->dio_bio = dio_bio;
8010 atomic_set(&dip->pending_bios, 0);
8011 btrfs_bio = btrfs_io_bio(io_bio);
8012 btrfs_bio->logical = file_offset;
8015 io_bio->bi_end_io = btrfs_endio_direct_write;
8017 io_bio->bi_end_io = btrfs_endio_direct_read;
8018 dip->subio_endio = btrfs_subio_endio_read;
8021 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8025 if (btrfs_bio->end_io)
8026 btrfs_bio->end_io(btrfs_bio, ret);
8032 * If this is a write, we need to clean up the reserved space and kill
8033 * the ordered extent.
8036 struct btrfs_ordered_extent *ordered;
8037 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
8038 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
8039 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
8040 btrfs_free_reserved_extent(root, ordered->start,
8041 ordered->disk_len, 1);
8042 btrfs_put_ordered_extent(ordered);
8043 btrfs_put_ordered_extent(ordered);
8045 bio_endio(dio_bio, ret);
8048 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
8049 const struct iov_iter *iter, loff_t offset)
8053 unsigned blocksize_mask = root->sectorsize - 1;
8054 ssize_t retval = -EINVAL;
8056 if (offset & blocksize_mask)
8059 if (iov_iter_alignment(iter) & blocksize_mask)
8062 /* If this is a write we don't need to check anymore */
8066 * Check to make sure we don't have duplicate iov_base's in this
8067 * iovec, if so return EINVAL, otherwise we'll get csum errors
8068 * when reading back.
8070 for (seg = 0; seg < iter->nr_segs; seg++) {
8071 for (i = seg + 1; i < iter->nr_segs; i++) {
8072 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8081 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
8082 struct iov_iter *iter, loff_t offset)
8084 struct file *file = iocb->ki_filp;
8085 struct inode *inode = file->f_mapping->host;
8089 bool relock = false;
8092 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iter, offset))
8095 atomic_inc(&inode->i_dio_count);
8096 smp_mb__after_atomic();
8099 * The generic stuff only does filemap_write_and_wait_range, which
8100 * isn't enough if we've written compressed pages to this area, so
8101 * we need to flush the dirty pages again to make absolutely sure
8102 * that any outstanding dirty pages are on disk.
8104 count = iov_iter_count(iter);
8105 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8106 &BTRFS_I(inode)->runtime_flags))
8107 filemap_fdatawrite_range(inode->i_mapping, offset,
8108 offset + count - 1);
8112 * If the write DIO is beyond the EOF, we need update
8113 * the isize, but it is protected by i_mutex. So we can
8114 * not unlock the i_mutex at this case.
8116 if (offset + count <= inode->i_size) {
8117 mutex_unlock(&inode->i_mutex);
8120 ret = btrfs_delalloc_reserve_space(inode, count);
8123 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8124 &BTRFS_I(inode)->runtime_flags)) {
8125 inode_dio_done(inode);
8126 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8130 ret = __blockdev_direct_IO(rw, iocb, inode,
8131 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8132 iter, offset, btrfs_get_blocks_direct, NULL,
8133 btrfs_submit_direct, flags);
8135 if (ret < 0 && ret != -EIOCBQUEUED)
8136 btrfs_delalloc_release_space(inode, count);
8137 else if (ret >= 0 && (size_t)ret < count)
8138 btrfs_delalloc_release_space(inode,
8139 count - (size_t)ret);
8143 inode_dio_done(inode);
8145 mutex_lock(&inode->i_mutex);
8150 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8152 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8153 __u64 start, __u64 len)
8157 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8161 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8164 int btrfs_readpage(struct file *file, struct page *page)
8166 struct extent_io_tree *tree;
8167 tree = &BTRFS_I(page->mapping->host)->io_tree;
8168 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8171 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8173 struct extent_io_tree *tree;
8176 if (current->flags & PF_MEMALLOC) {
8177 redirty_page_for_writepage(wbc, page);
8181 tree = &BTRFS_I(page->mapping->host)->io_tree;
8182 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8185 static int btrfs_writepages(struct address_space *mapping,
8186 struct writeback_control *wbc)
8188 struct extent_io_tree *tree;
8190 tree = &BTRFS_I(mapping->host)->io_tree;
8191 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8195 btrfs_readpages(struct file *file, struct address_space *mapping,
8196 struct list_head *pages, unsigned nr_pages)
8198 struct extent_io_tree *tree;
8199 tree = &BTRFS_I(mapping->host)->io_tree;
8200 return extent_readpages(tree, mapping, pages, nr_pages,
8203 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8205 struct extent_io_tree *tree;
8206 struct extent_map_tree *map;
8209 tree = &BTRFS_I(page->mapping->host)->io_tree;
8210 map = &BTRFS_I(page->mapping->host)->extent_tree;
8211 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8213 ClearPagePrivate(page);
8214 set_page_private(page, 0);
8215 page_cache_release(page);
8220 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8222 if (PageWriteback(page) || PageDirty(page))
8224 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8227 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8228 unsigned int length)
8230 struct inode *inode = page->mapping->host;
8231 struct extent_io_tree *tree;
8232 struct btrfs_ordered_extent *ordered;
8233 struct extent_state *cached_state = NULL;
8234 u64 page_start = page_offset(page);
8235 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
8236 int inode_evicting = inode->i_state & I_FREEING;
8239 * we have the page locked, so new writeback can't start,
8240 * and the dirty bit won't be cleared while we are here.
8242 * Wait for IO on this page so that we can safely clear
8243 * the PagePrivate2 bit and do ordered accounting
8245 wait_on_page_writeback(page);
8247 tree = &BTRFS_I(inode)->io_tree;
8249 btrfs_releasepage(page, GFP_NOFS);
8253 if (!inode_evicting)
8254 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
8255 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8258 * IO on this page will never be started, so we need
8259 * to account for any ordered extents now
8261 if (!inode_evicting)
8262 clear_extent_bit(tree, page_start, page_end,
8263 EXTENT_DIRTY | EXTENT_DELALLOC |
8264 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8265 EXTENT_DEFRAG, 1, 0, &cached_state,
8268 * whoever cleared the private bit is responsible
8269 * for the finish_ordered_io
8271 if (TestClearPagePrivate2(page)) {
8272 struct btrfs_ordered_inode_tree *tree;
8275 tree = &BTRFS_I(inode)->ordered_tree;
8277 spin_lock_irq(&tree->lock);
8278 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8279 new_len = page_start - ordered->file_offset;
8280 if (new_len < ordered->truncated_len)
8281 ordered->truncated_len = new_len;
8282 spin_unlock_irq(&tree->lock);
8284 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8286 PAGE_CACHE_SIZE, 1))
8287 btrfs_finish_ordered_io(ordered);
8289 btrfs_put_ordered_extent(ordered);
8290 if (!inode_evicting) {
8291 cached_state = NULL;
8292 lock_extent_bits(tree, page_start, page_end, 0,
8297 if (!inode_evicting) {
8298 clear_extent_bit(tree, page_start, page_end,
8299 EXTENT_LOCKED | EXTENT_DIRTY |
8300 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8301 EXTENT_DEFRAG, 1, 1,
8302 &cached_state, GFP_NOFS);
8304 __btrfs_releasepage(page, GFP_NOFS);
8307 ClearPageChecked(page);
8308 if (PagePrivate(page)) {
8309 ClearPagePrivate(page);
8310 set_page_private(page, 0);
8311 page_cache_release(page);
8316 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8317 * called from a page fault handler when a page is first dirtied. Hence we must
8318 * be careful to check for EOF conditions here. We set the page up correctly
8319 * for a written page which means we get ENOSPC checking when writing into
8320 * holes and correct delalloc and unwritten extent mapping on filesystems that
8321 * support these features.
8323 * We are not allowed to take the i_mutex here so we have to play games to
8324 * protect against truncate races as the page could now be beyond EOF. Because
8325 * vmtruncate() writes the inode size before removing pages, once we have the
8326 * page lock we can determine safely if the page is beyond EOF. If it is not
8327 * beyond EOF, then the page is guaranteed safe against truncation until we
8330 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8332 struct page *page = vmf->page;
8333 struct inode *inode = file_inode(vma->vm_file);
8334 struct btrfs_root *root = BTRFS_I(inode)->root;
8335 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8336 struct btrfs_ordered_extent *ordered;
8337 struct extent_state *cached_state = NULL;
8339 unsigned long zero_start;
8346 sb_start_pagefault(inode->i_sb);
8347 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
8349 ret = file_update_time(vma->vm_file);
8355 else /* -ENOSPC, -EIO, etc */
8356 ret = VM_FAULT_SIGBUS;
8362 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8365 size = i_size_read(inode);
8366 page_start = page_offset(page);
8367 page_end = page_start + PAGE_CACHE_SIZE - 1;
8369 if ((page->mapping != inode->i_mapping) ||
8370 (page_start >= size)) {
8371 /* page got truncated out from underneath us */
8374 wait_on_page_writeback(page);
8376 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
8377 set_page_extent_mapped(page);
8380 * we can't set the delalloc bits if there are pending ordered
8381 * extents. Drop our locks and wait for them to finish
8383 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8385 unlock_extent_cached(io_tree, page_start, page_end,
8386 &cached_state, GFP_NOFS);
8388 btrfs_start_ordered_extent(inode, ordered, 1);
8389 btrfs_put_ordered_extent(ordered);
8394 * XXX - page_mkwrite gets called every time the page is dirtied, even
8395 * if it was already dirty, so for space accounting reasons we need to
8396 * clear any delalloc bits for the range we are fixing to save. There
8397 * is probably a better way to do this, but for now keep consistent with
8398 * prepare_pages in the normal write path.
8400 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
8401 EXTENT_DIRTY | EXTENT_DELALLOC |
8402 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8403 0, 0, &cached_state, GFP_NOFS);
8405 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
8408 unlock_extent_cached(io_tree, page_start, page_end,
8409 &cached_state, GFP_NOFS);
8410 ret = VM_FAULT_SIGBUS;
8415 /* page is wholly or partially inside EOF */
8416 if (page_start + PAGE_CACHE_SIZE > size)
8417 zero_start = size & ~PAGE_CACHE_MASK;
8419 zero_start = PAGE_CACHE_SIZE;
8421 if (zero_start != PAGE_CACHE_SIZE) {
8423 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
8424 flush_dcache_page(page);
8427 ClearPageChecked(page);
8428 set_page_dirty(page);
8429 SetPageUptodate(page);
8431 BTRFS_I(inode)->last_trans = root->fs_info->generation;
8432 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8433 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8435 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
8439 sb_end_pagefault(inode->i_sb);
8440 return VM_FAULT_LOCKED;
8444 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
8446 sb_end_pagefault(inode->i_sb);
8450 static int btrfs_truncate(struct inode *inode)
8452 struct btrfs_root *root = BTRFS_I(inode)->root;
8453 struct btrfs_block_rsv *rsv;
8456 struct btrfs_trans_handle *trans;
8457 u64 mask = root->sectorsize - 1;
8458 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
8460 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8466 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8467 * 3 things going on here
8469 * 1) We need to reserve space for our orphan item and the space to
8470 * delete our orphan item. Lord knows we don't want to have a dangling
8471 * orphan item because we didn't reserve space to remove it.
8473 * 2) We need to reserve space to update our inode.
8475 * 3) We need to have something to cache all the space that is going to
8476 * be free'd up by the truncate operation, but also have some slack
8477 * space reserved in case it uses space during the truncate (thank you
8478 * very much snapshotting).
8480 * And we need these to all be seperate. The fact is we can use alot of
8481 * space doing the truncate, and we have no earthly idea how much space
8482 * we will use, so we need the truncate reservation to be seperate so it
8483 * doesn't end up using space reserved for updating the inode or
8484 * removing the orphan item. We also need to be able to stop the
8485 * transaction and start a new one, which means we need to be able to
8486 * update the inode several times, and we have no idea of knowing how
8487 * many times that will be, so we can't just reserve 1 item for the
8488 * entirety of the opration, so that has to be done seperately as well.
8489 * Then there is the orphan item, which does indeed need to be held on
8490 * to for the whole operation, and we need nobody to touch this reserved
8491 * space except the orphan code.
8493 * So that leaves us with
8495 * 1) root->orphan_block_rsv - for the orphan deletion.
8496 * 2) rsv - for the truncate reservation, which we will steal from the
8497 * transaction reservation.
8498 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8499 * updating the inode.
8501 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
8504 rsv->size = min_size;
8508 * 1 for the truncate slack space
8509 * 1 for updating the inode.
8511 trans = btrfs_start_transaction(root, 2);
8512 if (IS_ERR(trans)) {
8513 err = PTR_ERR(trans);
8517 /* Migrate the slack space for the truncate to our reserve */
8518 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
8523 * So if we truncate and then write and fsync we normally would just
8524 * write the extents that changed, which is a problem if we need to
8525 * first truncate that entire inode. So set this flag so we write out
8526 * all of the extents in the inode to the sync log so we're completely
8529 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8530 trans->block_rsv = rsv;
8533 ret = btrfs_truncate_inode_items(trans, root, inode,
8535 BTRFS_EXTENT_DATA_KEY);
8536 if (ret != -ENOSPC) {
8541 trans->block_rsv = &root->fs_info->trans_block_rsv;
8542 ret = btrfs_update_inode(trans, root, inode);
8548 btrfs_end_transaction(trans, root);
8549 btrfs_btree_balance_dirty(root);
8551 trans = btrfs_start_transaction(root, 2);
8552 if (IS_ERR(trans)) {
8553 ret = err = PTR_ERR(trans);
8558 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
8560 BUG_ON(ret); /* shouldn't happen */
8561 trans->block_rsv = rsv;
8564 if (ret == 0 && inode->i_nlink > 0) {
8565 trans->block_rsv = root->orphan_block_rsv;
8566 ret = btrfs_orphan_del(trans, inode);
8572 trans->block_rsv = &root->fs_info->trans_block_rsv;
8573 ret = btrfs_update_inode(trans, root, inode);
8577 ret = btrfs_end_transaction(trans, root);
8578 btrfs_btree_balance_dirty(root);
8582 btrfs_free_block_rsv(root, rsv);
8591 * create a new subvolume directory/inode (helper for the ioctl).
8593 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8594 struct btrfs_root *new_root,
8595 struct btrfs_root *parent_root,
8598 struct inode *inode;
8602 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8603 new_dirid, new_dirid,
8604 S_IFDIR | (~current_umask() & S_IRWXUGO),
8607 return PTR_ERR(inode);
8608 inode->i_op = &btrfs_dir_inode_operations;
8609 inode->i_fop = &btrfs_dir_file_operations;
8611 set_nlink(inode, 1);
8612 btrfs_i_size_write(inode, 0);
8613 unlock_new_inode(inode);
8615 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8617 btrfs_err(new_root->fs_info,
8618 "error inheriting subvolume %llu properties: %d",
8619 new_root->root_key.objectid, err);
8621 err = btrfs_update_inode(trans, new_root, inode);
8627 struct inode *btrfs_alloc_inode(struct super_block *sb)
8629 struct btrfs_inode *ei;
8630 struct inode *inode;
8632 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
8639 ei->last_sub_trans = 0;
8640 ei->logged_trans = 0;
8641 ei->delalloc_bytes = 0;
8642 ei->defrag_bytes = 0;
8643 ei->disk_i_size = 0;
8646 ei->index_cnt = (u64)-1;
8648 ei->last_unlink_trans = 0;
8649 ei->last_log_commit = 0;
8651 spin_lock_init(&ei->lock);
8652 ei->outstanding_extents = 0;
8653 ei->reserved_extents = 0;
8655 ei->runtime_flags = 0;
8656 ei->force_compress = BTRFS_COMPRESS_NONE;
8658 ei->delayed_node = NULL;
8660 ei->i_otime.tv_sec = 0;
8661 ei->i_otime.tv_nsec = 0;
8663 inode = &ei->vfs_inode;
8664 extent_map_tree_init(&ei->extent_tree);
8665 extent_io_tree_init(&ei->io_tree, &inode->i_data);
8666 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
8667 ei->io_tree.track_uptodate = 1;
8668 ei->io_failure_tree.track_uptodate = 1;
8669 atomic_set(&ei->sync_writers, 0);
8670 mutex_init(&ei->log_mutex);
8671 mutex_init(&ei->delalloc_mutex);
8672 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8673 INIT_LIST_HEAD(&ei->delalloc_inodes);
8674 RB_CLEAR_NODE(&ei->rb_node);
8679 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8680 void btrfs_test_destroy_inode(struct inode *inode)
8682 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8683 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8687 static void btrfs_i_callback(struct rcu_head *head)
8689 struct inode *inode = container_of(head, struct inode, i_rcu);
8690 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8693 void btrfs_destroy_inode(struct inode *inode)
8695 struct btrfs_ordered_extent *ordered;
8696 struct btrfs_root *root = BTRFS_I(inode)->root;
8698 WARN_ON(!hlist_empty(&inode->i_dentry));
8699 WARN_ON(inode->i_data.nrpages);
8700 WARN_ON(BTRFS_I(inode)->outstanding_extents);
8701 WARN_ON(BTRFS_I(inode)->reserved_extents);
8702 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8703 WARN_ON(BTRFS_I(inode)->csum_bytes);
8704 WARN_ON(BTRFS_I(inode)->defrag_bytes);
8707 * This can happen where we create an inode, but somebody else also
8708 * created the same inode and we need to destroy the one we already
8714 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8715 &BTRFS_I(inode)->runtime_flags)) {
8716 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
8718 atomic_dec(&root->orphan_inodes);
8722 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8726 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
8727 ordered->file_offset, ordered->len);
8728 btrfs_remove_ordered_extent(inode, ordered);
8729 btrfs_put_ordered_extent(ordered);
8730 btrfs_put_ordered_extent(ordered);
8733 inode_tree_del(inode);
8734 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8736 call_rcu(&inode->i_rcu, btrfs_i_callback);
8739 int btrfs_drop_inode(struct inode *inode)
8741 struct btrfs_root *root = BTRFS_I(inode)->root;
8746 /* the snap/subvol tree is on deleting */
8747 if (btrfs_root_refs(&root->root_item) == 0)
8750 return generic_drop_inode(inode);
8753 static void init_once(void *foo)
8755 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8757 inode_init_once(&ei->vfs_inode);
8760 void btrfs_destroy_cachep(void)
8763 * Make sure all delayed rcu free inodes are flushed before we
8767 if (btrfs_inode_cachep)
8768 kmem_cache_destroy(btrfs_inode_cachep);
8769 if (btrfs_trans_handle_cachep)
8770 kmem_cache_destroy(btrfs_trans_handle_cachep);
8771 if (btrfs_transaction_cachep)
8772 kmem_cache_destroy(btrfs_transaction_cachep);
8773 if (btrfs_path_cachep)
8774 kmem_cache_destroy(btrfs_path_cachep);
8775 if (btrfs_free_space_cachep)
8776 kmem_cache_destroy(btrfs_free_space_cachep);
8777 if (btrfs_delalloc_work_cachep)
8778 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8781 int btrfs_init_cachep(void)
8783 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8784 sizeof(struct btrfs_inode), 0,
8785 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8786 if (!btrfs_inode_cachep)
8789 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8790 sizeof(struct btrfs_trans_handle), 0,
8791 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8792 if (!btrfs_trans_handle_cachep)
8795 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8796 sizeof(struct btrfs_transaction), 0,
8797 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8798 if (!btrfs_transaction_cachep)
8801 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8802 sizeof(struct btrfs_path), 0,
8803 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8804 if (!btrfs_path_cachep)
8807 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8808 sizeof(struct btrfs_free_space), 0,
8809 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8810 if (!btrfs_free_space_cachep)
8813 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8814 sizeof(struct btrfs_delalloc_work), 0,
8815 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8817 if (!btrfs_delalloc_work_cachep)
8822 btrfs_destroy_cachep();
8826 static int btrfs_getattr(struct vfsmount *mnt,
8827 struct dentry *dentry, struct kstat *stat)
8830 struct inode *inode = dentry->d_inode;
8831 u32 blocksize = inode->i_sb->s_blocksize;
8833 generic_fillattr(inode, stat);
8834 stat->dev = BTRFS_I(inode)->root->anon_dev;
8835 stat->blksize = PAGE_CACHE_SIZE;
8837 spin_lock(&BTRFS_I(inode)->lock);
8838 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8839 spin_unlock(&BTRFS_I(inode)->lock);
8840 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8841 ALIGN(delalloc_bytes, blocksize)) >> 9;
8845 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8846 struct inode *new_dir, struct dentry *new_dentry)
8848 struct btrfs_trans_handle *trans;
8849 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8850 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8851 struct inode *new_inode = new_dentry->d_inode;
8852 struct inode *old_inode = old_dentry->d_inode;
8853 struct timespec ctime = CURRENT_TIME;
8857 u64 old_ino = btrfs_ino(old_inode);
8859 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8862 /* we only allow rename subvolume link between subvolumes */
8863 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8866 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8867 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8870 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8871 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8875 /* check for collisions, even if the name isn't there */
8876 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
8877 new_dentry->d_name.name,
8878 new_dentry->d_name.len);
8881 if (ret == -EEXIST) {
8883 * eexist without a new_inode */
8884 if (WARN_ON(!new_inode)) {
8888 /* maybe -EOVERFLOW */
8895 * we're using rename to replace one file with another. Start IO on it
8896 * now so we don't add too much work to the end of the transaction
8898 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
8899 filemap_flush(old_inode->i_mapping);
8901 /* close the racy window with snapshot create/destroy ioctl */
8902 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8903 down_read(&root->fs_info->subvol_sem);
8905 * We want to reserve the absolute worst case amount of items. So if
8906 * both inodes are subvols and we need to unlink them then that would
8907 * require 4 item modifications, but if they are both normal inodes it
8908 * would require 5 item modifications, so we'll assume their normal
8909 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8910 * should cover the worst case number of items we'll modify.
8912 trans = btrfs_start_transaction(root, 11);
8913 if (IS_ERR(trans)) {
8914 ret = PTR_ERR(trans);
8919 btrfs_record_root_in_trans(trans, dest);
8921 ret = btrfs_set_inode_index(new_dir, &index);
8925 BTRFS_I(old_inode)->dir_index = 0ULL;
8926 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8927 /* force full log commit if subvolume involved. */
8928 btrfs_set_log_full_commit(root->fs_info, trans);
8930 ret = btrfs_insert_inode_ref(trans, dest,
8931 new_dentry->d_name.name,
8932 new_dentry->d_name.len,
8934 btrfs_ino(new_dir), index);
8938 * this is an ugly little race, but the rename is required
8939 * to make sure that if we crash, the inode is either at the
8940 * old name or the new one. pinning the log transaction lets
8941 * us make sure we don't allow a log commit to come in after
8942 * we unlink the name but before we add the new name back in.
8944 btrfs_pin_log_trans(root);
8947 inode_inc_iversion(old_dir);
8948 inode_inc_iversion(new_dir);
8949 inode_inc_iversion(old_inode);
8950 old_dir->i_ctime = old_dir->i_mtime = ctime;
8951 new_dir->i_ctime = new_dir->i_mtime = ctime;
8952 old_inode->i_ctime = ctime;
8954 if (old_dentry->d_parent != new_dentry->d_parent)
8955 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8957 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8958 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8959 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8960 old_dentry->d_name.name,
8961 old_dentry->d_name.len);
8963 ret = __btrfs_unlink_inode(trans, root, old_dir,
8964 old_dentry->d_inode,
8965 old_dentry->d_name.name,
8966 old_dentry->d_name.len);
8968 ret = btrfs_update_inode(trans, root, old_inode);
8971 btrfs_abort_transaction(trans, root, ret);
8976 inode_inc_iversion(new_inode);
8977 new_inode->i_ctime = CURRENT_TIME;
8978 if (unlikely(btrfs_ino(new_inode) ==
8979 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8980 root_objectid = BTRFS_I(new_inode)->location.objectid;
8981 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8983 new_dentry->d_name.name,
8984 new_dentry->d_name.len);
8985 BUG_ON(new_inode->i_nlink == 0);
8987 ret = btrfs_unlink_inode(trans, dest, new_dir,
8988 new_dentry->d_inode,
8989 new_dentry->d_name.name,
8990 new_dentry->d_name.len);
8992 if (!ret && new_inode->i_nlink == 0)
8993 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8995 btrfs_abort_transaction(trans, root, ret);
9000 ret = btrfs_add_link(trans, new_dir, old_inode,
9001 new_dentry->d_name.name,
9002 new_dentry->d_name.len, 0, index);
9004 btrfs_abort_transaction(trans, root, ret);
9008 if (old_inode->i_nlink == 1)
9009 BTRFS_I(old_inode)->dir_index = index;
9011 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
9012 struct dentry *parent = new_dentry->d_parent;
9013 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9014 btrfs_end_log_trans(root);
9017 btrfs_end_transaction(trans, root);
9019 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9020 up_read(&root->fs_info->subvol_sem);
9025 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9026 struct inode *new_dir, struct dentry *new_dentry,
9029 if (flags & ~RENAME_NOREPLACE)
9032 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
9035 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9037 struct btrfs_delalloc_work *delalloc_work;
9038 struct inode *inode;
9040 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9042 inode = delalloc_work->inode;
9043 if (delalloc_work->wait) {
9044 btrfs_wait_ordered_range(inode, 0, (u64)-1);
9046 filemap_flush(inode->i_mapping);
9047 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9048 &BTRFS_I(inode)->runtime_flags))
9049 filemap_flush(inode->i_mapping);
9052 if (delalloc_work->delay_iput)
9053 btrfs_add_delayed_iput(inode);
9056 complete(&delalloc_work->completion);
9059 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9060 int wait, int delay_iput)
9062 struct btrfs_delalloc_work *work;
9064 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
9068 init_completion(&work->completion);
9069 INIT_LIST_HEAD(&work->list);
9070 work->inode = inode;
9072 work->delay_iput = delay_iput;
9073 WARN_ON_ONCE(!inode);
9074 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9075 btrfs_run_delalloc_work, NULL, NULL);
9080 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9082 wait_for_completion(&work->completion);
9083 kmem_cache_free(btrfs_delalloc_work_cachep, work);
9087 * some fairly slow code that needs optimization. This walks the list
9088 * of all the inodes with pending delalloc and forces them to disk.
9090 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9093 struct btrfs_inode *binode;
9094 struct inode *inode;
9095 struct btrfs_delalloc_work *work, *next;
9096 struct list_head works;
9097 struct list_head splice;
9100 INIT_LIST_HEAD(&works);
9101 INIT_LIST_HEAD(&splice);
9103 mutex_lock(&root->delalloc_mutex);
9104 spin_lock(&root->delalloc_lock);
9105 list_splice_init(&root->delalloc_inodes, &splice);
9106 while (!list_empty(&splice)) {
9107 binode = list_entry(splice.next, struct btrfs_inode,
9110 list_move_tail(&binode->delalloc_inodes,
9111 &root->delalloc_inodes);
9112 inode = igrab(&binode->vfs_inode);
9114 cond_resched_lock(&root->delalloc_lock);
9117 spin_unlock(&root->delalloc_lock);
9119 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
9122 btrfs_add_delayed_iput(inode);
9128 list_add_tail(&work->list, &works);
9129 btrfs_queue_work(root->fs_info->flush_workers,
9132 if (nr != -1 && ret >= nr)
9135 spin_lock(&root->delalloc_lock);
9137 spin_unlock(&root->delalloc_lock);
9140 list_for_each_entry_safe(work, next, &works, list) {
9141 list_del_init(&work->list);
9142 btrfs_wait_and_free_delalloc_work(work);
9145 if (!list_empty_careful(&splice)) {
9146 spin_lock(&root->delalloc_lock);
9147 list_splice_tail(&splice, &root->delalloc_inodes);
9148 spin_unlock(&root->delalloc_lock);
9150 mutex_unlock(&root->delalloc_mutex);
9154 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9158 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
9161 ret = __start_delalloc_inodes(root, delay_iput, -1);
9165 * the filemap_flush will queue IO into the worker threads, but
9166 * we have to make sure the IO is actually started and that
9167 * ordered extents get created before we return
9169 atomic_inc(&root->fs_info->async_submit_draining);
9170 while (atomic_read(&root->fs_info->nr_async_submits) ||
9171 atomic_read(&root->fs_info->async_delalloc_pages)) {
9172 wait_event(root->fs_info->async_submit_wait,
9173 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9174 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
9176 atomic_dec(&root->fs_info->async_submit_draining);
9180 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9183 struct btrfs_root *root;
9184 struct list_head splice;
9187 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9190 INIT_LIST_HEAD(&splice);
9192 mutex_lock(&fs_info->delalloc_root_mutex);
9193 spin_lock(&fs_info->delalloc_root_lock);
9194 list_splice_init(&fs_info->delalloc_roots, &splice);
9195 while (!list_empty(&splice) && nr) {
9196 root = list_first_entry(&splice, struct btrfs_root,
9198 root = btrfs_grab_fs_root(root);
9200 list_move_tail(&root->delalloc_root,
9201 &fs_info->delalloc_roots);
9202 spin_unlock(&fs_info->delalloc_root_lock);
9204 ret = __start_delalloc_inodes(root, delay_iput, nr);
9205 btrfs_put_fs_root(root);
9213 spin_lock(&fs_info->delalloc_root_lock);
9215 spin_unlock(&fs_info->delalloc_root_lock);
9218 atomic_inc(&fs_info->async_submit_draining);
9219 while (atomic_read(&fs_info->nr_async_submits) ||
9220 atomic_read(&fs_info->async_delalloc_pages)) {
9221 wait_event(fs_info->async_submit_wait,
9222 (atomic_read(&fs_info->nr_async_submits) == 0 &&
9223 atomic_read(&fs_info->async_delalloc_pages) == 0));
9225 atomic_dec(&fs_info->async_submit_draining);
9227 if (!list_empty_careful(&splice)) {
9228 spin_lock(&fs_info->delalloc_root_lock);
9229 list_splice_tail(&splice, &fs_info->delalloc_roots);
9230 spin_unlock(&fs_info->delalloc_root_lock);
9232 mutex_unlock(&fs_info->delalloc_root_mutex);
9236 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9237 const char *symname)
9239 struct btrfs_trans_handle *trans;
9240 struct btrfs_root *root = BTRFS_I(dir)->root;
9241 struct btrfs_path *path;
9242 struct btrfs_key key;
9243 struct inode *inode = NULL;
9251 struct btrfs_file_extent_item *ei;
9252 struct extent_buffer *leaf;
9254 name_len = strlen(symname);
9255 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9256 return -ENAMETOOLONG;
9259 * 2 items for inode item and ref
9260 * 2 items for dir items
9261 * 1 item for xattr if selinux is on
9263 trans = btrfs_start_transaction(root, 5);
9265 return PTR_ERR(trans);
9267 err = btrfs_find_free_ino(root, &objectid);
9271 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9272 dentry->d_name.len, btrfs_ino(dir), objectid,
9273 S_IFLNK|S_IRWXUGO, &index);
9274 if (IS_ERR(inode)) {
9275 err = PTR_ERR(inode);
9280 * If the active LSM wants to access the inode during
9281 * d_instantiate it needs these. Smack checks to see
9282 * if the filesystem supports xattrs by looking at the
9285 inode->i_fop = &btrfs_file_operations;
9286 inode->i_op = &btrfs_file_inode_operations;
9287 inode->i_mapping->a_ops = &btrfs_aops;
9288 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
9289 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9291 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9293 goto out_unlock_inode;
9295 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
9297 goto out_unlock_inode;
9299 path = btrfs_alloc_path();
9302 goto out_unlock_inode;
9304 key.objectid = btrfs_ino(inode);
9306 key.type = BTRFS_EXTENT_DATA_KEY;
9307 datasize = btrfs_file_extent_calc_inline_size(name_len);
9308 err = btrfs_insert_empty_item(trans, root, path, &key,
9311 btrfs_free_path(path);
9312 goto out_unlock_inode;
9314 leaf = path->nodes[0];
9315 ei = btrfs_item_ptr(leaf, path->slots[0],
9316 struct btrfs_file_extent_item);
9317 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9318 btrfs_set_file_extent_type(leaf, ei,
9319 BTRFS_FILE_EXTENT_INLINE);
9320 btrfs_set_file_extent_encryption(leaf, ei, 0);
9321 btrfs_set_file_extent_compression(leaf, ei, 0);
9322 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9323 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9325 ptr = btrfs_file_extent_inline_start(ei);
9326 write_extent_buffer(leaf, symname, ptr, name_len);
9327 btrfs_mark_buffer_dirty(leaf);
9328 btrfs_free_path(path);
9330 inode->i_op = &btrfs_symlink_inode_operations;
9331 inode->i_mapping->a_ops = &btrfs_symlink_aops;
9332 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
9333 inode_set_bytes(inode, name_len);
9334 btrfs_i_size_write(inode, name_len);
9335 err = btrfs_update_inode(trans, root, inode);
9338 goto out_unlock_inode;
9341 unlock_new_inode(inode);
9342 d_instantiate(dentry, inode);
9345 btrfs_end_transaction(trans, root);
9347 inode_dec_link_count(inode);
9350 btrfs_btree_balance_dirty(root);
9355 unlock_new_inode(inode);
9359 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9360 u64 start, u64 num_bytes, u64 min_size,
9361 loff_t actual_len, u64 *alloc_hint,
9362 struct btrfs_trans_handle *trans)
9364 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9365 struct extent_map *em;
9366 struct btrfs_root *root = BTRFS_I(inode)->root;
9367 struct btrfs_key ins;
9368 u64 cur_offset = start;
9372 bool own_trans = true;
9376 while (num_bytes > 0) {
9378 trans = btrfs_start_transaction(root, 3);
9379 if (IS_ERR(trans)) {
9380 ret = PTR_ERR(trans);
9385 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
9386 cur_bytes = max(cur_bytes, min_size);
9387 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
9388 *alloc_hint, &ins, 1, 0);
9391 btrfs_end_transaction(trans, root);
9395 ret = insert_reserved_file_extent(trans, inode,
9396 cur_offset, ins.objectid,
9397 ins.offset, ins.offset,
9398 ins.offset, 0, 0, 0,
9399 BTRFS_FILE_EXTENT_PREALLOC);
9401 btrfs_free_reserved_extent(root, ins.objectid,
9403 btrfs_abort_transaction(trans, root, ret);
9405 btrfs_end_transaction(trans, root);
9408 btrfs_drop_extent_cache(inode, cur_offset,
9409 cur_offset + ins.offset -1, 0);
9411 em = alloc_extent_map();
9413 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9414 &BTRFS_I(inode)->runtime_flags);
9418 em->start = cur_offset;
9419 em->orig_start = cur_offset;
9420 em->len = ins.offset;
9421 em->block_start = ins.objectid;
9422 em->block_len = ins.offset;
9423 em->orig_block_len = ins.offset;
9424 em->ram_bytes = ins.offset;
9425 em->bdev = root->fs_info->fs_devices->latest_bdev;
9426 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9427 em->generation = trans->transid;
9430 write_lock(&em_tree->lock);
9431 ret = add_extent_mapping(em_tree, em, 1);
9432 write_unlock(&em_tree->lock);
9435 btrfs_drop_extent_cache(inode, cur_offset,
9436 cur_offset + ins.offset - 1,
9439 free_extent_map(em);
9441 num_bytes -= ins.offset;
9442 cur_offset += ins.offset;
9443 *alloc_hint = ins.objectid + ins.offset;
9445 inode_inc_iversion(inode);
9446 inode->i_ctime = CURRENT_TIME;
9447 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9448 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9449 (actual_len > inode->i_size) &&
9450 (cur_offset > inode->i_size)) {
9451 if (cur_offset > actual_len)
9452 i_size = actual_len;
9454 i_size = cur_offset;
9455 i_size_write(inode, i_size);
9456 btrfs_ordered_update_i_size(inode, i_size, NULL);
9459 ret = btrfs_update_inode(trans, root, inode);
9462 btrfs_abort_transaction(trans, root, ret);
9464 btrfs_end_transaction(trans, root);
9469 btrfs_end_transaction(trans, root);
9474 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9475 u64 start, u64 num_bytes, u64 min_size,
9476 loff_t actual_len, u64 *alloc_hint)
9478 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9479 min_size, actual_len, alloc_hint,
9483 int btrfs_prealloc_file_range_trans(struct inode *inode,
9484 struct btrfs_trans_handle *trans, int mode,
9485 u64 start, u64 num_bytes, u64 min_size,
9486 loff_t actual_len, u64 *alloc_hint)
9488 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9489 min_size, actual_len, alloc_hint, trans);
9492 static int btrfs_set_page_dirty(struct page *page)
9494 return __set_page_dirty_nobuffers(page);
9497 static int btrfs_permission(struct inode *inode, int mask)
9499 struct btrfs_root *root = BTRFS_I(inode)->root;
9500 umode_t mode = inode->i_mode;
9502 if (mask & MAY_WRITE &&
9503 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9504 if (btrfs_root_readonly(root))
9506 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9509 return generic_permission(inode, mask);
9512 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9514 struct btrfs_trans_handle *trans;
9515 struct btrfs_root *root = BTRFS_I(dir)->root;
9516 struct inode *inode = NULL;
9522 * 5 units required for adding orphan entry
9524 trans = btrfs_start_transaction(root, 5);
9526 return PTR_ERR(trans);
9528 ret = btrfs_find_free_ino(root, &objectid);
9532 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9533 btrfs_ino(dir), objectid, mode, &index);
9534 if (IS_ERR(inode)) {
9535 ret = PTR_ERR(inode);
9540 inode->i_fop = &btrfs_file_operations;
9541 inode->i_op = &btrfs_file_inode_operations;
9543 inode->i_mapping->a_ops = &btrfs_aops;
9544 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
9545 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9547 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9551 ret = btrfs_update_inode(trans, root, inode);
9554 ret = btrfs_orphan_add(trans, inode);
9559 * We set number of links to 0 in btrfs_new_inode(), and here we set
9560 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9563 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9565 set_nlink(inode, 1);
9566 unlock_new_inode(inode);
9567 d_tmpfile(dentry, inode);
9568 mark_inode_dirty(inode);
9571 btrfs_end_transaction(trans, root);
9574 btrfs_balance_delayed_items(root);
9575 btrfs_btree_balance_dirty(root);
9579 unlock_new_inode(inode);
9584 /* Inspired by filemap_check_errors() */
9585 int btrfs_inode_check_errors(struct inode *inode)
9589 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
9590 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
9592 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
9593 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
9599 static const struct inode_operations btrfs_dir_inode_operations = {
9600 .getattr = btrfs_getattr,
9601 .lookup = btrfs_lookup,
9602 .create = btrfs_create,
9603 .unlink = btrfs_unlink,
9605 .mkdir = btrfs_mkdir,
9606 .rmdir = btrfs_rmdir,
9607 .rename2 = btrfs_rename2,
9608 .symlink = btrfs_symlink,
9609 .setattr = btrfs_setattr,
9610 .mknod = btrfs_mknod,
9611 .setxattr = btrfs_setxattr,
9612 .getxattr = btrfs_getxattr,
9613 .listxattr = btrfs_listxattr,
9614 .removexattr = btrfs_removexattr,
9615 .permission = btrfs_permission,
9616 .get_acl = btrfs_get_acl,
9617 .set_acl = btrfs_set_acl,
9618 .update_time = btrfs_update_time,
9619 .tmpfile = btrfs_tmpfile,
9621 static const struct inode_operations btrfs_dir_ro_inode_operations = {
9622 .lookup = btrfs_lookup,
9623 .permission = btrfs_permission,
9624 .get_acl = btrfs_get_acl,
9625 .set_acl = btrfs_set_acl,
9626 .update_time = btrfs_update_time,
9629 static const struct file_operations btrfs_dir_file_operations = {
9630 .llseek = generic_file_llseek,
9631 .read = generic_read_dir,
9632 .iterate = btrfs_real_readdir,
9633 .unlocked_ioctl = btrfs_ioctl,
9634 #ifdef CONFIG_COMPAT
9635 .compat_ioctl = btrfs_ioctl,
9637 .release = btrfs_release_file,
9638 .fsync = btrfs_sync_file,
9641 static struct extent_io_ops btrfs_extent_io_ops = {
9642 .fill_delalloc = run_delalloc_range,
9643 .submit_bio_hook = btrfs_submit_bio_hook,
9644 .merge_bio_hook = btrfs_merge_bio_hook,
9645 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
9646 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
9647 .writepage_start_hook = btrfs_writepage_start_hook,
9648 .set_bit_hook = btrfs_set_bit_hook,
9649 .clear_bit_hook = btrfs_clear_bit_hook,
9650 .merge_extent_hook = btrfs_merge_extent_hook,
9651 .split_extent_hook = btrfs_split_extent_hook,
9655 * btrfs doesn't support the bmap operation because swapfiles
9656 * use bmap to make a mapping of extents in the file. They assume
9657 * these extents won't change over the life of the file and they
9658 * use the bmap result to do IO directly to the drive.
9660 * the btrfs bmap call would return logical addresses that aren't
9661 * suitable for IO and they also will change frequently as COW
9662 * operations happen. So, swapfile + btrfs == corruption.
9664 * For now we're avoiding this by dropping bmap.
9666 static const struct address_space_operations btrfs_aops = {
9667 .readpage = btrfs_readpage,
9668 .writepage = btrfs_writepage,
9669 .writepages = btrfs_writepages,
9670 .readpages = btrfs_readpages,
9671 .direct_IO = btrfs_direct_IO,
9672 .invalidatepage = btrfs_invalidatepage,
9673 .releasepage = btrfs_releasepage,
9674 .set_page_dirty = btrfs_set_page_dirty,
9675 .error_remove_page = generic_error_remove_page,
9678 static const struct address_space_operations btrfs_symlink_aops = {
9679 .readpage = btrfs_readpage,
9680 .writepage = btrfs_writepage,
9681 .invalidatepage = btrfs_invalidatepage,
9682 .releasepage = btrfs_releasepage,
9685 static const struct inode_operations btrfs_file_inode_operations = {
9686 .getattr = btrfs_getattr,
9687 .setattr = btrfs_setattr,
9688 .setxattr = btrfs_setxattr,
9689 .getxattr = btrfs_getxattr,
9690 .listxattr = btrfs_listxattr,
9691 .removexattr = btrfs_removexattr,
9692 .permission = btrfs_permission,
9693 .fiemap = btrfs_fiemap,
9694 .get_acl = btrfs_get_acl,
9695 .set_acl = btrfs_set_acl,
9696 .update_time = btrfs_update_time,
9698 static const struct inode_operations btrfs_special_inode_operations = {
9699 .getattr = btrfs_getattr,
9700 .setattr = btrfs_setattr,
9701 .permission = btrfs_permission,
9702 .setxattr = btrfs_setxattr,
9703 .getxattr = btrfs_getxattr,
9704 .listxattr = btrfs_listxattr,
9705 .removexattr = btrfs_removexattr,
9706 .get_acl = btrfs_get_acl,
9707 .set_acl = btrfs_set_acl,
9708 .update_time = btrfs_update_time,
9710 static const struct inode_operations btrfs_symlink_inode_operations = {
9711 .readlink = generic_readlink,
9712 .follow_link = page_follow_link_light,
9713 .put_link = page_put_link,
9714 .getattr = btrfs_getattr,
9715 .setattr = btrfs_setattr,
9716 .permission = btrfs_permission,
9717 .setxattr = btrfs_setxattr,
9718 .getxattr = btrfs_getxattr,
9719 .listxattr = btrfs_listxattr,
9720 .removexattr = btrfs_removexattr,
9721 .update_time = btrfs_update_time,
9724 const struct dentry_operations btrfs_dentry_operations = {
9725 .d_delete = btrfs_dentry_delete,
9726 .d_release = btrfs_dentry_release,