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/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include <linux/mount.h>
42 #include <linux/btrfs.h>
43 #include <linux/blkdev.h>
44 #include <linux/posix_acl_xattr.h>
45 #include <linux/uio.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"
64 struct btrfs_iget_args {
65 struct btrfs_key *location;
66 struct btrfs_root *root;
69 static const struct inode_operations btrfs_dir_inode_operations;
70 static const struct inode_operations btrfs_symlink_inode_operations;
71 static const struct inode_operations btrfs_dir_ro_inode_operations;
72 static const struct inode_operations btrfs_special_inode_operations;
73 static const struct inode_operations btrfs_file_inode_operations;
74 static const struct address_space_operations btrfs_aops;
75 static const struct address_space_operations btrfs_symlink_aops;
76 static const struct file_operations btrfs_dir_file_operations;
77 static struct extent_io_ops btrfs_extent_io_ops;
79 static struct kmem_cache *btrfs_inode_cachep;
80 static struct kmem_cache *btrfs_delalloc_work_cachep;
81 struct kmem_cache *btrfs_trans_handle_cachep;
82 struct kmem_cache *btrfs_transaction_cachep;
83 struct kmem_cache *btrfs_path_cachep;
84 struct kmem_cache *btrfs_free_space_cachep;
87 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
88 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
89 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
90 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
91 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
92 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
93 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
94 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
97 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
98 static int btrfs_truncate(struct inode *inode);
99 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
100 static noinline int cow_file_range(struct inode *inode,
101 struct page *locked_page,
102 u64 start, u64 end, int *page_started,
103 unsigned long *nr_written, int unlock);
104 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
105 u64 len, u64 orig_start,
106 u64 block_start, u64 block_len,
107 u64 orig_block_len, u64 ram_bytes,
110 static int btrfs_dirty_inode(struct inode *inode);
112 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
113 void btrfs_test_inode_set_ops(struct inode *inode)
115 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
119 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
120 struct inode *inode, struct inode *dir,
121 const struct qstr *qstr)
125 err = btrfs_init_acl(trans, inode, dir);
127 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
132 * this does all the hard work for inserting an inline extent into
133 * the btree. The caller should have done a btrfs_drop_extents so that
134 * no overlapping inline items exist in the btree
136 static int insert_inline_extent(struct btrfs_trans_handle *trans,
137 struct btrfs_path *path, int extent_inserted,
138 struct btrfs_root *root, struct inode *inode,
139 u64 start, size_t size, size_t compressed_size,
141 struct page **compressed_pages)
143 struct extent_buffer *leaf;
144 struct page *page = NULL;
147 struct btrfs_file_extent_item *ei;
150 size_t cur_size = size;
151 unsigned long offset;
153 if (compressed_size && compressed_pages)
154 cur_size = compressed_size;
156 inode_add_bytes(inode, size);
158 if (!extent_inserted) {
159 struct btrfs_key key;
162 key.objectid = btrfs_ino(inode);
164 key.type = BTRFS_EXTENT_DATA_KEY;
166 datasize = btrfs_file_extent_calc_inline_size(cur_size);
167 path->leave_spinning = 1;
168 ret = btrfs_insert_empty_item(trans, root, path, &key,
175 leaf = path->nodes[0];
176 ei = btrfs_item_ptr(leaf, path->slots[0],
177 struct btrfs_file_extent_item);
178 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
179 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
180 btrfs_set_file_extent_encryption(leaf, ei, 0);
181 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
182 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
183 ptr = btrfs_file_extent_inline_start(ei);
185 if (compress_type != BTRFS_COMPRESS_NONE) {
188 while (compressed_size > 0) {
189 cpage = compressed_pages[i];
190 cur_size = min_t(unsigned long, compressed_size,
193 kaddr = kmap_atomic(cpage);
194 write_extent_buffer(leaf, kaddr, ptr, cur_size);
195 kunmap_atomic(kaddr);
199 compressed_size -= cur_size;
201 btrfs_set_file_extent_compression(leaf, ei,
204 page = find_get_page(inode->i_mapping,
205 start >> PAGE_CACHE_SHIFT);
206 btrfs_set_file_extent_compression(leaf, ei, 0);
207 kaddr = kmap_atomic(page);
208 offset = start & (PAGE_CACHE_SIZE - 1);
209 write_extent_buffer(leaf, kaddr + offset, ptr, size);
210 kunmap_atomic(kaddr);
211 page_cache_release(page);
213 btrfs_mark_buffer_dirty(leaf);
214 btrfs_release_path(path);
217 * we're an inline extent, so nobody can
218 * extend the file past i_size without locking
219 * a page we already have locked.
221 * We must do any isize and inode updates
222 * before we unlock the pages. Otherwise we
223 * could end up racing with unlink.
225 BTRFS_I(inode)->disk_i_size = inode->i_size;
226 ret = btrfs_update_inode(trans, root, inode);
235 * conditionally insert an inline extent into the file. This
236 * does the checks required to make sure the data is small enough
237 * to fit as an inline extent.
239 static noinline int cow_file_range_inline(struct btrfs_root *root,
240 struct inode *inode, u64 start,
241 u64 end, size_t compressed_size,
243 struct page **compressed_pages)
245 struct btrfs_trans_handle *trans;
246 u64 isize = i_size_read(inode);
247 u64 actual_end = min(end + 1, isize);
248 u64 inline_len = actual_end - start;
249 u64 aligned_end = ALIGN(end, root->sectorsize);
250 u64 data_len = inline_len;
252 struct btrfs_path *path;
253 int extent_inserted = 0;
254 u32 extent_item_size;
257 data_len = compressed_size;
260 actual_end > PAGE_CACHE_SIZE ||
261 data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) ||
263 (actual_end & (root->sectorsize - 1)) == 0) ||
265 data_len > root->fs_info->max_inline) {
269 path = btrfs_alloc_path();
273 trans = btrfs_join_transaction(root);
275 btrfs_free_path(path);
276 return PTR_ERR(trans);
278 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
280 if (compressed_size && compressed_pages)
281 extent_item_size = btrfs_file_extent_calc_inline_size(
284 extent_item_size = btrfs_file_extent_calc_inline_size(
287 ret = __btrfs_drop_extents(trans, root, inode, path,
288 start, aligned_end, NULL,
289 1, 1, extent_item_size, &extent_inserted);
291 btrfs_abort_transaction(trans, root, ret);
295 if (isize > actual_end)
296 inline_len = min_t(u64, isize, actual_end);
297 ret = insert_inline_extent(trans, path, extent_inserted,
299 inline_len, compressed_size,
300 compress_type, compressed_pages);
301 if (ret && ret != -ENOSPC) {
302 btrfs_abort_transaction(trans, root, ret);
304 } else if (ret == -ENOSPC) {
309 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
310 btrfs_delalloc_release_metadata(inode, end + 1 - start);
311 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
313 btrfs_free_path(path);
314 btrfs_end_transaction(trans, root);
318 struct async_extent {
323 unsigned long nr_pages;
325 struct list_head list;
330 struct btrfs_root *root;
331 struct page *locked_page;
334 struct list_head extents;
335 struct btrfs_work work;
338 static noinline int add_async_extent(struct async_cow *cow,
339 u64 start, u64 ram_size,
342 unsigned long nr_pages,
345 struct async_extent *async_extent;
347 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
348 BUG_ON(!async_extent); /* -ENOMEM */
349 async_extent->start = start;
350 async_extent->ram_size = ram_size;
351 async_extent->compressed_size = compressed_size;
352 async_extent->pages = pages;
353 async_extent->nr_pages = nr_pages;
354 async_extent->compress_type = compress_type;
355 list_add_tail(&async_extent->list, &cow->extents);
359 static inline int inode_need_compress(struct inode *inode)
361 struct btrfs_root *root = BTRFS_I(inode)->root;
364 if (btrfs_test_opt(root, FORCE_COMPRESS))
366 /* bad compression ratios */
367 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
369 if (btrfs_test_opt(root, COMPRESS) ||
370 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
371 BTRFS_I(inode)->force_compress)
377 * we create compressed extents in two phases. The first
378 * phase compresses a range of pages that have already been
379 * locked (both pages and state bits are locked).
381 * This is done inside an ordered work queue, and the compression
382 * is spread across many cpus. The actual IO submission is step
383 * two, and the ordered work queue takes care of making sure that
384 * happens in the same order things were put onto the queue by
385 * writepages and friends.
387 * If this code finds it can't get good compression, it puts an
388 * entry onto the work queue to write the uncompressed bytes. This
389 * makes sure that both compressed inodes and uncompressed inodes
390 * are written in the same order that the flusher thread sent them
393 static noinline void compress_file_range(struct inode *inode,
394 struct page *locked_page,
396 struct async_cow *async_cow,
399 struct btrfs_root *root = BTRFS_I(inode)->root;
401 u64 blocksize = root->sectorsize;
403 u64 isize = i_size_read(inode);
405 struct page **pages = NULL;
406 unsigned long nr_pages;
407 unsigned long nr_pages_ret = 0;
408 unsigned long total_compressed = 0;
409 unsigned long total_in = 0;
410 unsigned long max_compressed = 128 * 1024;
411 unsigned long max_uncompressed = 128 * 1024;
414 int compress_type = root->fs_info->compress_type;
417 /* if this is a small write inside eof, kick off a defrag */
418 if ((end - start + 1) < 16 * 1024 &&
419 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
420 btrfs_add_inode_defrag(NULL, inode);
422 actual_end = min_t(u64, isize, end + 1);
425 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
426 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
429 * we don't want to send crud past the end of i_size through
430 * compression, that's just a waste of CPU time. So, if the
431 * end of the file is before the start of our current
432 * requested range of bytes, we bail out to the uncompressed
433 * cleanup code that can deal with all of this.
435 * It isn't really the fastest way to fix things, but this is a
436 * very uncommon corner.
438 if (actual_end <= start)
439 goto cleanup_and_bail_uncompressed;
441 total_compressed = actual_end - start;
444 * skip compression for a small file range(<=blocksize) that
445 * isn't an inline extent, since it dosen't save disk space at all.
447 if (total_compressed <= blocksize &&
448 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
449 goto cleanup_and_bail_uncompressed;
451 /* we want to make sure that amount of ram required to uncompress
452 * an extent is reasonable, so we limit the total size in ram
453 * of a compressed extent to 128k. This is a crucial number
454 * because it also controls how easily we can spread reads across
455 * cpus for decompression.
457 * We also want to make sure the amount of IO required to do
458 * a random read is reasonably small, so we limit the size of
459 * a compressed extent to 128k.
461 total_compressed = min(total_compressed, max_uncompressed);
462 num_bytes = ALIGN(end - start + 1, blocksize);
463 num_bytes = max(blocksize, num_bytes);
468 * we do compression for mount -o compress and when the
469 * inode has not been flagged as nocompress. This flag can
470 * change at any time if we discover bad compression ratios.
472 if (inode_need_compress(inode)) {
474 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
476 /* just bail out to the uncompressed code */
480 if (BTRFS_I(inode)->force_compress)
481 compress_type = BTRFS_I(inode)->force_compress;
484 * we need to call clear_page_dirty_for_io on each
485 * page in the range. Otherwise applications with the file
486 * mmap'd can wander in and change the page contents while
487 * we are compressing them.
489 * If the compression fails for any reason, we set the pages
490 * dirty again later on.
492 extent_range_clear_dirty_for_io(inode, start, end);
494 ret = btrfs_compress_pages(compress_type,
495 inode->i_mapping, start,
496 total_compressed, pages,
497 nr_pages, &nr_pages_ret,
503 unsigned long offset = total_compressed &
504 (PAGE_CACHE_SIZE - 1);
505 struct page *page = pages[nr_pages_ret - 1];
508 /* zero the tail end of the last page, we might be
509 * sending it down to disk
512 kaddr = kmap_atomic(page);
513 memset(kaddr + offset, 0,
514 PAGE_CACHE_SIZE - offset);
515 kunmap_atomic(kaddr);
522 /* lets try to make an inline extent */
523 if (ret || total_in < (actual_end - start)) {
524 /* we didn't compress the entire range, try
525 * to make an uncompressed inline extent.
527 ret = cow_file_range_inline(root, inode, start, end,
530 /* try making a compressed inline extent */
531 ret = cow_file_range_inline(root, inode, start, end,
533 compress_type, pages);
536 unsigned long clear_flags = EXTENT_DELALLOC |
538 unsigned long page_error_op;
540 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
541 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
544 * inline extent creation worked or returned error,
545 * we don't need to create any more async work items.
546 * Unlock and free up our temp pages.
548 extent_clear_unlock_delalloc(inode, start, end, NULL,
549 clear_flags, PAGE_UNLOCK |
560 * we aren't doing an inline extent round the compressed size
561 * up to a block size boundary so the allocator does sane
564 total_compressed = ALIGN(total_compressed, blocksize);
567 * one last check to make sure the compression is really a
568 * win, compare the page count read with the blocks on disk
570 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
571 if (total_compressed >= total_in) {
574 num_bytes = total_in;
577 if (!will_compress && pages) {
579 * the compression code ran but failed to make things smaller,
580 * free any pages it allocated and our page pointer array
582 for (i = 0; i < nr_pages_ret; i++) {
583 WARN_ON(pages[i]->mapping);
584 page_cache_release(pages[i]);
588 total_compressed = 0;
591 /* flag the file so we don't compress in the future */
592 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
593 !(BTRFS_I(inode)->force_compress)) {
594 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
600 /* the async work queues will take care of doing actual
601 * allocation on disk for these compressed pages,
602 * and will submit them to the elevator.
604 add_async_extent(async_cow, start, num_bytes,
605 total_compressed, pages, nr_pages_ret,
608 if (start + num_bytes < end) {
615 cleanup_and_bail_uncompressed:
617 * No compression, but we still need to write the pages in
618 * the file we've been given so far. redirty the locked
619 * page if it corresponds to our extent and set things up
620 * for the async work queue to run cow_file_range to do
621 * the normal delalloc dance
623 if (page_offset(locked_page) >= start &&
624 page_offset(locked_page) <= end) {
625 __set_page_dirty_nobuffers(locked_page);
626 /* unlocked later on in the async handlers */
629 extent_range_redirty_for_io(inode, start, end);
630 add_async_extent(async_cow, start, end - start + 1,
631 0, NULL, 0, BTRFS_COMPRESS_NONE);
638 for (i = 0; i < nr_pages_ret; i++) {
639 WARN_ON(pages[i]->mapping);
640 page_cache_release(pages[i]);
645 static void free_async_extent_pages(struct async_extent *async_extent)
649 if (!async_extent->pages)
652 for (i = 0; i < async_extent->nr_pages; i++) {
653 WARN_ON(async_extent->pages[i]->mapping);
654 page_cache_release(async_extent->pages[i]);
656 kfree(async_extent->pages);
657 async_extent->nr_pages = 0;
658 async_extent->pages = NULL;
662 * phase two of compressed writeback. This is the ordered portion
663 * of the code, which only gets called in the order the work was
664 * queued. We walk all the async extents created by compress_file_range
665 * and send them down to the disk.
667 static noinline void submit_compressed_extents(struct inode *inode,
668 struct async_cow *async_cow)
670 struct async_extent *async_extent;
672 struct btrfs_key ins;
673 struct extent_map *em;
674 struct btrfs_root *root = BTRFS_I(inode)->root;
675 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
676 struct extent_io_tree *io_tree;
680 while (!list_empty(&async_cow->extents)) {
681 async_extent = list_entry(async_cow->extents.next,
682 struct async_extent, list);
683 list_del(&async_extent->list);
685 io_tree = &BTRFS_I(inode)->io_tree;
688 /* did the compression code fall back to uncompressed IO? */
689 if (!async_extent->pages) {
690 int page_started = 0;
691 unsigned long nr_written = 0;
693 lock_extent(io_tree, async_extent->start,
694 async_extent->start +
695 async_extent->ram_size - 1);
697 /* allocate blocks */
698 ret = cow_file_range(inode, async_cow->locked_page,
700 async_extent->start +
701 async_extent->ram_size - 1,
702 &page_started, &nr_written, 0);
707 * if page_started, cow_file_range inserted an
708 * inline extent and took care of all the unlocking
709 * and IO for us. Otherwise, we need to submit
710 * all those pages down to the drive.
712 if (!page_started && !ret)
713 extent_write_locked_range(io_tree,
714 inode, async_extent->start,
715 async_extent->start +
716 async_extent->ram_size - 1,
720 unlock_page(async_cow->locked_page);
726 lock_extent(io_tree, async_extent->start,
727 async_extent->start + async_extent->ram_size - 1);
729 ret = btrfs_reserve_extent(root,
730 async_extent->compressed_size,
731 async_extent->compressed_size,
732 0, alloc_hint, &ins, 1, 1);
734 free_async_extent_pages(async_extent);
736 if (ret == -ENOSPC) {
737 unlock_extent(io_tree, async_extent->start,
738 async_extent->start +
739 async_extent->ram_size - 1);
742 * we need to redirty the pages if we decide to
743 * fallback to uncompressed IO, otherwise we
744 * will not submit these pages down to lower
747 extent_range_redirty_for_io(inode,
749 async_extent->start +
750 async_extent->ram_size - 1);
757 * here we're doing allocation and writeback of the
760 btrfs_drop_extent_cache(inode, async_extent->start,
761 async_extent->start +
762 async_extent->ram_size - 1, 0);
764 em = alloc_extent_map();
767 goto out_free_reserve;
769 em->start = async_extent->start;
770 em->len = async_extent->ram_size;
771 em->orig_start = em->start;
772 em->mod_start = em->start;
773 em->mod_len = em->len;
775 em->block_start = ins.objectid;
776 em->block_len = ins.offset;
777 em->orig_block_len = ins.offset;
778 em->ram_bytes = async_extent->ram_size;
779 em->bdev = root->fs_info->fs_devices->latest_bdev;
780 em->compress_type = async_extent->compress_type;
781 set_bit(EXTENT_FLAG_PINNED, &em->flags);
782 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
786 write_lock(&em_tree->lock);
787 ret = add_extent_mapping(em_tree, em, 1);
788 write_unlock(&em_tree->lock);
789 if (ret != -EEXIST) {
793 btrfs_drop_extent_cache(inode, async_extent->start,
794 async_extent->start +
795 async_extent->ram_size - 1, 0);
799 goto out_free_reserve;
801 ret = btrfs_add_ordered_extent_compress(inode,
804 async_extent->ram_size,
806 BTRFS_ORDERED_COMPRESSED,
807 async_extent->compress_type);
809 btrfs_drop_extent_cache(inode, async_extent->start,
810 async_extent->start +
811 async_extent->ram_size - 1, 0);
812 goto out_free_reserve;
816 * clear dirty, set writeback and unlock the pages.
818 extent_clear_unlock_delalloc(inode, async_extent->start,
819 async_extent->start +
820 async_extent->ram_size - 1,
821 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
822 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
824 ret = btrfs_submit_compressed_write(inode,
826 async_extent->ram_size,
828 ins.offset, async_extent->pages,
829 async_extent->nr_pages);
831 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
832 struct page *p = async_extent->pages[0];
833 const u64 start = async_extent->start;
834 const u64 end = start + async_extent->ram_size - 1;
836 p->mapping = inode->i_mapping;
837 tree->ops->writepage_end_io_hook(p, start, end,
840 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
843 free_async_extent_pages(async_extent);
845 alloc_hint = ins.objectid + ins.offset;
851 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
853 extent_clear_unlock_delalloc(inode, async_extent->start,
854 async_extent->start +
855 async_extent->ram_size - 1,
856 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
857 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
858 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
859 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
861 free_async_extent_pages(async_extent);
866 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
869 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
870 struct extent_map *em;
873 read_lock(&em_tree->lock);
874 em = search_extent_mapping(em_tree, start, num_bytes);
877 * if block start isn't an actual block number then find the
878 * first block in this inode and use that as a hint. If that
879 * block is also bogus then just don't worry about it.
881 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
883 em = search_extent_mapping(em_tree, 0, 0);
884 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
885 alloc_hint = em->block_start;
889 alloc_hint = em->block_start;
893 read_unlock(&em_tree->lock);
899 * when extent_io.c finds a delayed allocation range in the file,
900 * the call backs end up in this code. The basic idea is to
901 * allocate extents on disk for the range, and create ordered data structs
902 * in ram to track those extents.
904 * locked_page is the page that writepage had locked already. We use
905 * it to make sure we don't do extra locks or unlocks.
907 * *page_started is set to one if we unlock locked_page and do everything
908 * required to start IO on it. It may be clean and already done with
911 static noinline int cow_file_range(struct inode *inode,
912 struct page *locked_page,
913 u64 start, u64 end, int *page_started,
914 unsigned long *nr_written,
917 struct btrfs_root *root = BTRFS_I(inode)->root;
920 unsigned long ram_size;
923 u64 blocksize = root->sectorsize;
924 struct btrfs_key ins;
925 struct extent_map *em;
926 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
929 if (btrfs_is_free_space_inode(inode)) {
935 num_bytes = ALIGN(end - start + 1, blocksize);
936 num_bytes = max(blocksize, num_bytes);
937 disk_num_bytes = num_bytes;
939 /* if this is a small write inside eof, kick off defrag */
940 if (num_bytes < 64 * 1024 &&
941 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
942 btrfs_add_inode_defrag(NULL, inode);
945 /* lets try to make an inline extent */
946 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
949 extent_clear_unlock_delalloc(inode, start, end, NULL,
950 EXTENT_LOCKED | EXTENT_DELALLOC |
951 EXTENT_DEFRAG, PAGE_UNLOCK |
952 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
955 *nr_written = *nr_written +
956 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
959 } else if (ret < 0) {
964 BUG_ON(disk_num_bytes >
965 btrfs_super_total_bytes(root->fs_info->super_copy));
967 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
968 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
970 while (disk_num_bytes > 0) {
973 cur_alloc_size = disk_num_bytes;
974 ret = btrfs_reserve_extent(root, cur_alloc_size,
975 root->sectorsize, 0, alloc_hint,
980 em = alloc_extent_map();
986 em->orig_start = em->start;
987 ram_size = ins.offset;
988 em->len = ins.offset;
989 em->mod_start = em->start;
990 em->mod_len = em->len;
992 em->block_start = ins.objectid;
993 em->block_len = ins.offset;
994 em->orig_block_len = ins.offset;
995 em->ram_bytes = ram_size;
996 em->bdev = root->fs_info->fs_devices->latest_bdev;
997 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1001 write_lock(&em_tree->lock);
1002 ret = add_extent_mapping(em_tree, em, 1);
1003 write_unlock(&em_tree->lock);
1004 if (ret != -EEXIST) {
1005 free_extent_map(em);
1008 btrfs_drop_extent_cache(inode, start,
1009 start + ram_size - 1, 0);
1014 cur_alloc_size = ins.offset;
1015 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1016 ram_size, cur_alloc_size, 0);
1018 goto out_drop_extent_cache;
1020 if (root->root_key.objectid ==
1021 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1022 ret = btrfs_reloc_clone_csums(inode, start,
1025 goto out_drop_extent_cache;
1028 if (disk_num_bytes < cur_alloc_size)
1031 /* we're not doing compressed IO, don't unlock the first
1032 * page (which the caller expects to stay locked), don't
1033 * clear any dirty bits and don't set any writeback bits
1035 * Do set the Private2 bit so we know this page was properly
1036 * setup for writepage
1038 op = unlock ? PAGE_UNLOCK : 0;
1039 op |= PAGE_SET_PRIVATE2;
1041 extent_clear_unlock_delalloc(inode, start,
1042 start + ram_size - 1, locked_page,
1043 EXTENT_LOCKED | EXTENT_DELALLOC,
1045 disk_num_bytes -= cur_alloc_size;
1046 num_bytes -= cur_alloc_size;
1047 alloc_hint = ins.objectid + ins.offset;
1048 start += cur_alloc_size;
1053 out_drop_extent_cache:
1054 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1056 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1058 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1059 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1060 EXTENT_DELALLOC | EXTENT_DEFRAG,
1061 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1062 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1067 * work queue call back to started compression on a file and pages
1069 static noinline void async_cow_start(struct btrfs_work *work)
1071 struct async_cow *async_cow;
1073 async_cow = container_of(work, struct async_cow, work);
1075 compress_file_range(async_cow->inode, async_cow->locked_page,
1076 async_cow->start, async_cow->end, async_cow,
1078 if (num_added == 0) {
1079 btrfs_add_delayed_iput(async_cow->inode);
1080 async_cow->inode = NULL;
1085 * work queue call back to submit previously compressed pages
1087 static noinline void async_cow_submit(struct btrfs_work *work)
1089 struct async_cow *async_cow;
1090 struct btrfs_root *root;
1091 unsigned long nr_pages;
1093 async_cow = container_of(work, struct async_cow, work);
1095 root = async_cow->root;
1096 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1099 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1101 waitqueue_active(&root->fs_info->async_submit_wait))
1102 wake_up(&root->fs_info->async_submit_wait);
1104 if (async_cow->inode)
1105 submit_compressed_extents(async_cow->inode, async_cow);
1108 static noinline void async_cow_free(struct btrfs_work *work)
1110 struct async_cow *async_cow;
1111 async_cow = container_of(work, struct async_cow, work);
1112 if (async_cow->inode)
1113 btrfs_add_delayed_iput(async_cow->inode);
1117 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1118 u64 start, u64 end, int *page_started,
1119 unsigned long *nr_written)
1121 struct async_cow *async_cow;
1122 struct btrfs_root *root = BTRFS_I(inode)->root;
1123 unsigned long nr_pages;
1125 int limit = 10 * 1024 * 1024;
1127 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1128 1, 0, NULL, GFP_NOFS);
1129 while (start < end) {
1130 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1131 BUG_ON(!async_cow); /* -ENOMEM */
1132 async_cow->inode = igrab(inode);
1133 async_cow->root = root;
1134 async_cow->locked_page = locked_page;
1135 async_cow->start = start;
1137 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1138 !btrfs_test_opt(root, FORCE_COMPRESS))
1141 cur_end = min(end, start + 512 * 1024 - 1);
1143 async_cow->end = cur_end;
1144 INIT_LIST_HEAD(&async_cow->extents);
1146 btrfs_init_work(&async_cow->work,
1147 btrfs_delalloc_helper,
1148 async_cow_start, async_cow_submit,
1151 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1153 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1155 btrfs_queue_work(root->fs_info->delalloc_workers,
1158 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1159 wait_event(root->fs_info->async_submit_wait,
1160 (atomic_read(&root->fs_info->async_delalloc_pages) <
1164 while (atomic_read(&root->fs_info->async_submit_draining) &&
1165 atomic_read(&root->fs_info->async_delalloc_pages)) {
1166 wait_event(root->fs_info->async_submit_wait,
1167 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1171 *nr_written += nr_pages;
1172 start = cur_end + 1;
1178 static noinline int csum_exist_in_range(struct btrfs_root *root,
1179 u64 bytenr, u64 num_bytes)
1182 struct btrfs_ordered_sum *sums;
1185 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1186 bytenr + num_bytes - 1, &list, 0);
1187 if (ret == 0 && list_empty(&list))
1190 while (!list_empty(&list)) {
1191 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1192 list_del(&sums->list);
1199 * when nowcow writeback call back. This checks for snapshots or COW copies
1200 * of the extents that exist in the file, and COWs the file as required.
1202 * If no cow copies or snapshots exist, we write directly to the existing
1205 static noinline int run_delalloc_nocow(struct inode *inode,
1206 struct page *locked_page,
1207 u64 start, u64 end, int *page_started, int force,
1208 unsigned long *nr_written)
1210 struct btrfs_root *root = BTRFS_I(inode)->root;
1211 struct btrfs_trans_handle *trans;
1212 struct extent_buffer *leaf;
1213 struct btrfs_path *path;
1214 struct btrfs_file_extent_item *fi;
1215 struct btrfs_key found_key;
1230 u64 ino = btrfs_ino(inode);
1232 path = btrfs_alloc_path();
1234 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1235 EXTENT_LOCKED | EXTENT_DELALLOC |
1236 EXTENT_DO_ACCOUNTING |
1237 EXTENT_DEFRAG, PAGE_UNLOCK |
1239 PAGE_SET_WRITEBACK |
1240 PAGE_END_WRITEBACK);
1244 nolock = btrfs_is_free_space_inode(inode);
1247 trans = btrfs_join_transaction_nolock(root);
1249 trans = btrfs_join_transaction(root);
1251 if (IS_ERR(trans)) {
1252 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1253 EXTENT_LOCKED | EXTENT_DELALLOC |
1254 EXTENT_DO_ACCOUNTING |
1255 EXTENT_DEFRAG, PAGE_UNLOCK |
1257 PAGE_SET_WRITEBACK |
1258 PAGE_END_WRITEBACK);
1259 btrfs_free_path(path);
1260 return PTR_ERR(trans);
1263 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1265 cow_start = (u64)-1;
1268 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1272 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1273 leaf = path->nodes[0];
1274 btrfs_item_key_to_cpu(leaf, &found_key,
1275 path->slots[0] - 1);
1276 if (found_key.objectid == ino &&
1277 found_key.type == BTRFS_EXTENT_DATA_KEY)
1282 leaf = path->nodes[0];
1283 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1284 ret = btrfs_next_leaf(root, path);
1289 leaf = path->nodes[0];
1295 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1297 if (found_key.objectid > ino ||
1298 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1299 found_key.offset > end)
1302 if (found_key.offset > cur_offset) {
1303 extent_end = found_key.offset;
1308 fi = btrfs_item_ptr(leaf, path->slots[0],
1309 struct btrfs_file_extent_item);
1310 extent_type = btrfs_file_extent_type(leaf, fi);
1312 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1313 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1314 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1315 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1316 extent_offset = btrfs_file_extent_offset(leaf, fi);
1317 extent_end = found_key.offset +
1318 btrfs_file_extent_num_bytes(leaf, fi);
1320 btrfs_file_extent_disk_num_bytes(leaf, fi);
1321 if (extent_end <= start) {
1325 if (disk_bytenr == 0)
1327 if (btrfs_file_extent_compression(leaf, fi) ||
1328 btrfs_file_extent_encryption(leaf, fi) ||
1329 btrfs_file_extent_other_encoding(leaf, fi))
1331 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1333 if (btrfs_extent_readonly(root, disk_bytenr))
1335 if (btrfs_cross_ref_exist(trans, root, ino,
1337 extent_offset, disk_bytenr))
1339 disk_bytenr += extent_offset;
1340 disk_bytenr += cur_offset - found_key.offset;
1341 num_bytes = min(end + 1, extent_end) - cur_offset;
1343 * if there are pending snapshots for this root,
1344 * we fall into common COW way.
1347 err = btrfs_start_write_no_snapshoting(root);
1352 * force cow if csum exists in the range.
1353 * this ensure that csum for a given extent are
1354 * either valid or do not exist.
1356 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1359 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1360 extent_end = found_key.offset +
1361 btrfs_file_extent_inline_len(leaf,
1362 path->slots[0], fi);
1363 extent_end = ALIGN(extent_end, root->sectorsize);
1368 if (extent_end <= start) {
1370 if (!nolock && nocow)
1371 btrfs_end_write_no_snapshoting(root);
1375 if (cow_start == (u64)-1)
1376 cow_start = cur_offset;
1377 cur_offset = extent_end;
1378 if (cur_offset > end)
1384 btrfs_release_path(path);
1385 if (cow_start != (u64)-1) {
1386 ret = cow_file_range(inode, locked_page,
1387 cow_start, found_key.offset - 1,
1388 page_started, nr_written, 1);
1390 if (!nolock && nocow)
1391 btrfs_end_write_no_snapshoting(root);
1394 cow_start = (u64)-1;
1397 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1398 struct extent_map *em;
1399 struct extent_map_tree *em_tree;
1400 em_tree = &BTRFS_I(inode)->extent_tree;
1401 em = alloc_extent_map();
1402 BUG_ON(!em); /* -ENOMEM */
1403 em->start = cur_offset;
1404 em->orig_start = found_key.offset - extent_offset;
1405 em->len = num_bytes;
1406 em->block_len = num_bytes;
1407 em->block_start = disk_bytenr;
1408 em->orig_block_len = disk_num_bytes;
1409 em->ram_bytes = ram_bytes;
1410 em->bdev = root->fs_info->fs_devices->latest_bdev;
1411 em->mod_start = em->start;
1412 em->mod_len = em->len;
1413 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1414 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1415 em->generation = -1;
1417 write_lock(&em_tree->lock);
1418 ret = add_extent_mapping(em_tree, em, 1);
1419 write_unlock(&em_tree->lock);
1420 if (ret != -EEXIST) {
1421 free_extent_map(em);
1424 btrfs_drop_extent_cache(inode, em->start,
1425 em->start + em->len - 1, 0);
1427 type = BTRFS_ORDERED_PREALLOC;
1429 type = BTRFS_ORDERED_NOCOW;
1432 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1433 num_bytes, num_bytes, type);
1434 BUG_ON(ret); /* -ENOMEM */
1436 if (root->root_key.objectid ==
1437 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1438 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1441 if (!nolock && nocow)
1442 btrfs_end_write_no_snapshoting(root);
1447 extent_clear_unlock_delalloc(inode, cur_offset,
1448 cur_offset + num_bytes - 1,
1449 locked_page, EXTENT_LOCKED |
1450 EXTENT_DELALLOC, PAGE_UNLOCK |
1452 if (!nolock && nocow)
1453 btrfs_end_write_no_snapshoting(root);
1454 cur_offset = extent_end;
1455 if (cur_offset > end)
1458 btrfs_release_path(path);
1460 if (cur_offset <= end && cow_start == (u64)-1) {
1461 cow_start = cur_offset;
1465 if (cow_start != (u64)-1) {
1466 ret = cow_file_range(inode, locked_page, cow_start, end,
1467 page_started, nr_written, 1);
1473 err = btrfs_end_transaction(trans, root);
1477 if (ret && cur_offset < end)
1478 extent_clear_unlock_delalloc(inode, cur_offset, end,
1479 locked_page, EXTENT_LOCKED |
1480 EXTENT_DELALLOC | EXTENT_DEFRAG |
1481 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1483 PAGE_SET_WRITEBACK |
1484 PAGE_END_WRITEBACK);
1485 btrfs_free_path(path);
1489 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1492 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1493 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1497 * @defrag_bytes is a hint value, no spinlock held here,
1498 * if is not zero, it means the file is defragging.
1499 * Force cow if given extent needs to be defragged.
1501 if (BTRFS_I(inode)->defrag_bytes &&
1502 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1503 EXTENT_DEFRAG, 0, NULL))
1510 * extent_io.c call back to do delayed allocation processing
1512 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1513 u64 start, u64 end, int *page_started,
1514 unsigned long *nr_written)
1517 int force_cow = need_force_cow(inode, start, end);
1519 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1520 ret = run_delalloc_nocow(inode, locked_page, start, end,
1521 page_started, 1, nr_written);
1522 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1523 ret = run_delalloc_nocow(inode, locked_page, start, end,
1524 page_started, 0, nr_written);
1525 } else if (!inode_need_compress(inode)) {
1526 ret = cow_file_range(inode, locked_page, start, end,
1527 page_started, nr_written, 1);
1529 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1530 &BTRFS_I(inode)->runtime_flags);
1531 ret = cow_file_range_async(inode, locked_page, start, end,
1532 page_started, nr_written);
1537 static void btrfs_split_extent_hook(struct inode *inode,
1538 struct extent_state *orig, u64 split)
1542 /* not delalloc, ignore it */
1543 if (!(orig->state & EXTENT_DELALLOC))
1546 size = orig->end - orig->start + 1;
1547 if (size > BTRFS_MAX_EXTENT_SIZE) {
1552 * See the explanation in btrfs_merge_extent_hook, the same
1553 * applies here, just in reverse.
1555 new_size = orig->end - split + 1;
1556 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1557 BTRFS_MAX_EXTENT_SIZE);
1558 new_size = split - orig->start;
1559 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1560 BTRFS_MAX_EXTENT_SIZE);
1561 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1562 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1566 spin_lock(&BTRFS_I(inode)->lock);
1567 BTRFS_I(inode)->outstanding_extents++;
1568 spin_unlock(&BTRFS_I(inode)->lock);
1572 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1573 * extents so we can keep track of new extents that are just merged onto old
1574 * extents, such as when we are doing sequential writes, so we can properly
1575 * account for the metadata space we'll need.
1577 static void btrfs_merge_extent_hook(struct inode *inode,
1578 struct extent_state *new,
1579 struct extent_state *other)
1581 u64 new_size, old_size;
1584 /* not delalloc, ignore it */
1585 if (!(other->state & EXTENT_DELALLOC))
1588 if (new->start > other->start)
1589 new_size = new->end - other->start + 1;
1591 new_size = other->end - new->start + 1;
1593 /* we're not bigger than the max, unreserve the space and go */
1594 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1595 spin_lock(&BTRFS_I(inode)->lock);
1596 BTRFS_I(inode)->outstanding_extents--;
1597 spin_unlock(&BTRFS_I(inode)->lock);
1602 * We have to add up either side to figure out how many extents were
1603 * accounted for before we merged into one big extent. If the number of
1604 * extents we accounted for is <= the amount we need for the new range
1605 * then we can return, otherwise drop. Think of it like this
1609 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1610 * need 2 outstanding extents, on one side we have 1 and the other side
1611 * we have 1 so they are == and we can return. But in this case
1613 * [MAX_SIZE+4k][MAX_SIZE+4k]
1615 * Each range on their own accounts for 2 extents, but merged together
1616 * they are only 3 extents worth of accounting, so we need to drop in
1619 old_size = other->end - other->start + 1;
1620 num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1621 BTRFS_MAX_EXTENT_SIZE);
1622 old_size = new->end - new->start + 1;
1623 num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1624 BTRFS_MAX_EXTENT_SIZE);
1626 if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1627 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1630 spin_lock(&BTRFS_I(inode)->lock);
1631 BTRFS_I(inode)->outstanding_extents--;
1632 spin_unlock(&BTRFS_I(inode)->lock);
1635 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1636 struct inode *inode)
1638 spin_lock(&root->delalloc_lock);
1639 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1640 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1641 &root->delalloc_inodes);
1642 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1643 &BTRFS_I(inode)->runtime_flags);
1644 root->nr_delalloc_inodes++;
1645 if (root->nr_delalloc_inodes == 1) {
1646 spin_lock(&root->fs_info->delalloc_root_lock);
1647 BUG_ON(!list_empty(&root->delalloc_root));
1648 list_add_tail(&root->delalloc_root,
1649 &root->fs_info->delalloc_roots);
1650 spin_unlock(&root->fs_info->delalloc_root_lock);
1653 spin_unlock(&root->delalloc_lock);
1656 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1657 struct inode *inode)
1659 spin_lock(&root->delalloc_lock);
1660 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1661 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1662 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1663 &BTRFS_I(inode)->runtime_flags);
1664 root->nr_delalloc_inodes--;
1665 if (!root->nr_delalloc_inodes) {
1666 spin_lock(&root->fs_info->delalloc_root_lock);
1667 BUG_ON(list_empty(&root->delalloc_root));
1668 list_del_init(&root->delalloc_root);
1669 spin_unlock(&root->fs_info->delalloc_root_lock);
1672 spin_unlock(&root->delalloc_lock);
1676 * extent_io.c set_bit_hook, used to track delayed allocation
1677 * bytes in this file, and to maintain the list of inodes that
1678 * have pending delalloc work to be done.
1680 static void btrfs_set_bit_hook(struct inode *inode,
1681 struct extent_state *state, unsigned *bits)
1684 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1687 * set_bit and clear bit hooks normally require _irqsave/restore
1688 * but in this case, we are only testing for the DELALLOC
1689 * bit, which is only set or cleared with irqs on
1691 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1692 struct btrfs_root *root = BTRFS_I(inode)->root;
1693 u64 len = state->end + 1 - state->start;
1694 bool do_list = !btrfs_is_free_space_inode(inode);
1696 if (*bits & EXTENT_FIRST_DELALLOC) {
1697 *bits &= ~EXTENT_FIRST_DELALLOC;
1699 spin_lock(&BTRFS_I(inode)->lock);
1700 BTRFS_I(inode)->outstanding_extents++;
1701 spin_unlock(&BTRFS_I(inode)->lock);
1704 /* For sanity tests */
1705 if (btrfs_test_is_dummy_root(root))
1708 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1709 root->fs_info->delalloc_batch);
1710 spin_lock(&BTRFS_I(inode)->lock);
1711 BTRFS_I(inode)->delalloc_bytes += len;
1712 if (*bits & EXTENT_DEFRAG)
1713 BTRFS_I(inode)->defrag_bytes += len;
1714 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1715 &BTRFS_I(inode)->runtime_flags))
1716 btrfs_add_delalloc_inodes(root, inode);
1717 spin_unlock(&BTRFS_I(inode)->lock);
1722 * extent_io.c clear_bit_hook, see set_bit_hook for why
1724 static void btrfs_clear_bit_hook(struct inode *inode,
1725 struct extent_state *state,
1728 u64 len = state->end + 1 - state->start;
1729 u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1730 BTRFS_MAX_EXTENT_SIZE);
1732 spin_lock(&BTRFS_I(inode)->lock);
1733 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1734 BTRFS_I(inode)->defrag_bytes -= len;
1735 spin_unlock(&BTRFS_I(inode)->lock);
1738 * set_bit and clear bit hooks normally require _irqsave/restore
1739 * but in this case, we are only testing for the DELALLOC
1740 * bit, which is only set or cleared with irqs on
1742 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1743 struct btrfs_root *root = BTRFS_I(inode)->root;
1744 bool do_list = !btrfs_is_free_space_inode(inode);
1746 if (*bits & EXTENT_FIRST_DELALLOC) {
1747 *bits &= ~EXTENT_FIRST_DELALLOC;
1748 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1749 spin_lock(&BTRFS_I(inode)->lock);
1750 BTRFS_I(inode)->outstanding_extents -= num_extents;
1751 spin_unlock(&BTRFS_I(inode)->lock);
1755 * We don't reserve metadata space for space cache inodes so we
1756 * don't need to call dellalloc_release_metadata if there is an
1759 if (*bits & EXTENT_DO_ACCOUNTING &&
1760 root != root->fs_info->tree_root)
1761 btrfs_delalloc_release_metadata(inode, len);
1763 /* For sanity tests. */
1764 if (btrfs_test_is_dummy_root(root))
1767 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1768 && do_list && !(state->state & EXTENT_NORESERVE))
1769 btrfs_free_reserved_data_space(inode, len);
1771 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1772 root->fs_info->delalloc_batch);
1773 spin_lock(&BTRFS_I(inode)->lock);
1774 BTRFS_I(inode)->delalloc_bytes -= len;
1775 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1776 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1777 &BTRFS_I(inode)->runtime_flags))
1778 btrfs_del_delalloc_inode(root, inode);
1779 spin_unlock(&BTRFS_I(inode)->lock);
1784 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1785 * we don't create bios that span stripes or chunks
1787 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1788 size_t size, struct bio *bio,
1789 unsigned long bio_flags)
1791 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1792 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1797 if (bio_flags & EXTENT_BIO_COMPRESSED)
1800 length = bio->bi_iter.bi_size;
1801 map_length = length;
1802 ret = btrfs_map_block(root->fs_info, rw, logical,
1803 &map_length, NULL, 0);
1804 /* Will always return 0 with map_multi == NULL */
1806 if (map_length < length + size)
1812 * in order to insert checksums into the metadata in large chunks,
1813 * we wait until bio submission time. All the pages in the bio are
1814 * checksummed and sums are attached onto the ordered extent record.
1816 * At IO completion time the cums attached on the ordered extent record
1817 * are inserted into the btree
1819 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1820 struct bio *bio, int mirror_num,
1821 unsigned long bio_flags,
1824 struct btrfs_root *root = BTRFS_I(inode)->root;
1827 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1828 BUG_ON(ret); /* -ENOMEM */
1833 * in order to insert checksums into the metadata in large chunks,
1834 * we wait until bio submission time. All the pages in the bio are
1835 * checksummed and sums are attached onto the ordered extent record.
1837 * At IO completion time the cums attached on the ordered extent record
1838 * are inserted into the btree
1840 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1841 int mirror_num, unsigned long bio_flags,
1844 struct btrfs_root *root = BTRFS_I(inode)->root;
1847 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1849 bio_endio(bio, ret);
1854 * extent_io.c submission hook. This does the right thing for csum calculation
1855 * on write, or reading the csums from the tree before a read
1857 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1858 int mirror_num, unsigned long bio_flags,
1861 struct btrfs_root *root = BTRFS_I(inode)->root;
1865 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1867 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1869 if (btrfs_is_free_space_inode(inode))
1872 if (!(rw & REQ_WRITE)) {
1873 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1877 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1878 ret = btrfs_submit_compressed_read(inode, bio,
1882 } else if (!skip_sum) {
1883 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1888 } else if (async && !skip_sum) {
1889 /* csum items have already been cloned */
1890 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1892 /* we're doing a write, do the async checksumming */
1893 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1894 inode, rw, bio, mirror_num,
1895 bio_flags, bio_offset,
1896 __btrfs_submit_bio_start,
1897 __btrfs_submit_bio_done);
1899 } else if (!skip_sum) {
1900 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1906 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1910 bio_endio(bio, ret);
1915 * given a list of ordered sums record them in the inode. This happens
1916 * at IO completion time based on sums calculated at bio submission time.
1918 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1919 struct inode *inode, u64 file_offset,
1920 struct list_head *list)
1922 struct btrfs_ordered_sum *sum;
1924 list_for_each_entry(sum, list, list) {
1925 trans->adding_csums = 1;
1926 btrfs_csum_file_blocks(trans,
1927 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1928 trans->adding_csums = 0;
1933 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1934 struct extent_state **cached_state)
1936 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1937 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1938 cached_state, GFP_NOFS);
1941 /* see btrfs_writepage_start_hook for details on why this is required */
1942 struct btrfs_writepage_fixup {
1944 struct btrfs_work work;
1947 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1949 struct btrfs_writepage_fixup *fixup;
1950 struct btrfs_ordered_extent *ordered;
1951 struct extent_state *cached_state = NULL;
1953 struct inode *inode;
1958 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1962 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1963 ClearPageChecked(page);
1967 inode = page->mapping->host;
1968 page_start = page_offset(page);
1969 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1971 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1974 /* already ordered? We're done */
1975 if (PagePrivate2(page))
1978 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1980 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1981 page_end, &cached_state, GFP_NOFS);
1983 btrfs_start_ordered_extent(inode, ordered, 1);
1984 btrfs_put_ordered_extent(ordered);
1988 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1990 mapping_set_error(page->mapping, ret);
1991 end_extent_writepage(page, ret, page_start, page_end);
1992 ClearPageChecked(page);
1996 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1997 ClearPageChecked(page);
1998 set_page_dirty(page);
2000 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2001 &cached_state, GFP_NOFS);
2004 page_cache_release(page);
2009 * There are a few paths in the higher layers of the kernel that directly
2010 * set the page dirty bit without asking the filesystem if it is a
2011 * good idea. This causes problems because we want to make sure COW
2012 * properly happens and the data=ordered rules are followed.
2014 * In our case any range that doesn't have the ORDERED bit set
2015 * hasn't been properly setup for IO. We kick off an async process
2016 * to fix it up. The async helper will wait for ordered extents, set
2017 * the delalloc bit and make it safe to write the page.
2019 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2021 struct inode *inode = page->mapping->host;
2022 struct btrfs_writepage_fixup *fixup;
2023 struct btrfs_root *root = BTRFS_I(inode)->root;
2025 /* this page is properly in the ordered list */
2026 if (TestClearPagePrivate2(page))
2029 if (PageChecked(page))
2032 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2036 SetPageChecked(page);
2037 page_cache_get(page);
2038 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2039 btrfs_writepage_fixup_worker, NULL, NULL);
2041 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
2045 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2046 struct inode *inode, u64 file_pos,
2047 u64 disk_bytenr, u64 disk_num_bytes,
2048 u64 num_bytes, u64 ram_bytes,
2049 u8 compression, u8 encryption,
2050 u16 other_encoding, int extent_type)
2052 struct btrfs_root *root = BTRFS_I(inode)->root;
2053 struct btrfs_file_extent_item *fi;
2054 struct btrfs_path *path;
2055 struct extent_buffer *leaf;
2056 struct btrfs_key ins;
2057 int extent_inserted = 0;
2060 path = btrfs_alloc_path();
2065 * we may be replacing one extent in the tree with another.
2066 * The new extent is pinned in the extent map, and we don't want
2067 * to drop it from the cache until it is completely in the btree.
2069 * So, tell btrfs_drop_extents to leave this extent in the cache.
2070 * the caller is expected to unpin it and allow it to be merged
2073 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2074 file_pos + num_bytes, NULL, 0,
2075 1, sizeof(*fi), &extent_inserted);
2079 if (!extent_inserted) {
2080 ins.objectid = btrfs_ino(inode);
2081 ins.offset = file_pos;
2082 ins.type = BTRFS_EXTENT_DATA_KEY;
2084 path->leave_spinning = 1;
2085 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2090 leaf = path->nodes[0];
2091 fi = btrfs_item_ptr(leaf, path->slots[0],
2092 struct btrfs_file_extent_item);
2093 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2094 btrfs_set_file_extent_type(leaf, fi, extent_type);
2095 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2096 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2097 btrfs_set_file_extent_offset(leaf, fi, 0);
2098 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2099 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2100 btrfs_set_file_extent_compression(leaf, fi, compression);
2101 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2102 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2104 btrfs_mark_buffer_dirty(leaf);
2105 btrfs_release_path(path);
2107 inode_add_bytes(inode, num_bytes);
2109 ins.objectid = disk_bytenr;
2110 ins.offset = disk_num_bytes;
2111 ins.type = BTRFS_EXTENT_ITEM_KEY;
2112 ret = btrfs_alloc_reserved_file_extent(trans, root,
2113 root->root_key.objectid,
2114 btrfs_ino(inode), file_pos, &ins);
2116 btrfs_free_path(path);
2121 /* snapshot-aware defrag */
2122 struct sa_defrag_extent_backref {
2123 struct rb_node node;
2124 struct old_sa_defrag_extent *old;
2133 struct old_sa_defrag_extent {
2134 struct list_head list;
2135 struct new_sa_defrag_extent *new;
2144 struct new_sa_defrag_extent {
2145 struct rb_root root;
2146 struct list_head head;
2147 struct btrfs_path *path;
2148 struct inode *inode;
2156 static int backref_comp(struct sa_defrag_extent_backref *b1,
2157 struct sa_defrag_extent_backref *b2)
2159 if (b1->root_id < b2->root_id)
2161 else if (b1->root_id > b2->root_id)
2164 if (b1->inum < b2->inum)
2166 else if (b1->inum > b2->inum)
2169 if (b1->file_pos < b2->file_pos)
2171 else if (b1->file_pos > b2->file_pos)
2175 * [------------------------------] ===> (a range of space)
2176 * |<--->| |<---->| =============> (fs/file tree A)
2177 * |<---------------------------->| ===> (fs/file tree B)
2179 * A range of space can refer to two file extents in one tree while
2180 * refer to only one file extent in another tree.
2182 * So we may process a disk offset more than one time(two extents in A)
2183 * and locate at the same extent(one extent in B), then insert two same
2184 * backrefs(both refer to the extent in B).
2189 static void backref_insert(struct rb_root *root,
2190 struct sa_defrag_extent_backref *backref)
2192 struct rb_node **p = &root->rb_node;
2193 struct rb_node *parent = NULL;
2194 struct sa_defrag_extent_backref *entry;
2199 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2201 ret = backref_comp(backref, entry);
2205 p = &(*p)->rb_right;
2208 rb_link_node(&backref->node, parent, p);
2209 rb_insert_color(&backref->node, root);
2213 * Note the backref might has changed, and in this case we just return 0.
2215 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2218 struct btrfs_file_extent_item *extent;
2219 struct btrfs_fs_info *fs_info;
2220 struct old_sa_defrag_extent *old = ctx;
2221 struct new_sa_defrag_extent *new = old->new;
2222 struct btrfs_path *path = new->path;
2223 struct btrfs_key key;
2224 struct btrfs_root *root;
2225 struct sa_defrag_extent_backref *backref;
2226 struct extent_buffer *leaf;
2227 struct inode *inode = new->inode;
2233 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2234 inum == btrfs_ino(inode))
2237 key.objectid = root_id;
2238 key.type = BTRFS_ROOT_ITEM_KEY;
2239 key.offset = (u64)-1;
2241 fs_info = BTRFS_I(inode)->root->fs_info;
2242 root = btrfs_read_fs_root_no_name(fs_info, &key);
2244 if (PTR_ERR(root) == -ENOENT)
2247 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2248 inum, offset, root_id);
2249 return PTR_ERR(root);
2252 key.objectid = inum;
2253 key.type = BTRFS_EXTENT_DATA_KEY;
2254 if (offset > (u64)-1 << 32)
2257 key.offset = offset;
2259 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2260 if (WARN_ON(ret < 0))
2267 leaf = path->nodes[0];
2268 slot = path->slots[0];
2270 if (slot >= btrfs_header_nritems(leaf)) {
2271 ret = btrfs_next_leaf(root, path);
2274 } else if (ret > 0) {
2283 btrfs_item_key_to_cpu(leaf, &key, slot);
2285 if (key.objectid > inum)
2288 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2291 extent = btrfs_item_ptr(leaf, slot,
2292 struct btrfs_file_extent_item);
2294 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2298 * 'offset' refers to the exact key.offset,
2299 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2300 * (key.offset - extent_offset).
2302 if (key.offset != offset)
2305 extent_offset = btrfs_file_extent_offset(leaf, extent);
2306 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2308 if (extent_offset >= old->extent_offset + old->offset +
2309 old->len || extent_offset + num_bytes <=
2310 old->extent_offset + old->offset)
2315 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2321 backref->root_id = root_id;
2322 backref->inum = inum;
2323 backref->file_pos = offset;
2324 backref->num_bytes = num_bytes;
2325 backref->extent_offset = extent_offset;
2326 backref->generation = btrfs_file_extent_generation(leaf, extent);
2328 backref_insert(&new->root, backref);
2331 btrfs_release_path(path);
2336 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2337 struct new_sa_defrag_extent *new)
2339 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2340 struct old_sa_defrag_extent *old, *tmp;
2345 list_for_each_entry_safe(old, tmp, &new->head, list) {
2346 ret = iterate_inodes_from_logical(old->bytenr +
2347 old->extent_offset, fs_info,
2348 path, record_one_backref,
2350 if (ret < 0 && ret != -ENOENT)
2353 /* no backref to be processed for this extent */
2355 list_del(&old->list);
2360 if (list_empty(&new->head))
2366 static int relink_is_mergable(struct extent_buffer *leaf,
2367 struct btrfs_file_extent_item *fi,
2368 struct new_sa_defrag_extent *new)
2370 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2373 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2376 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2379 if (btrfs_file_extent_encryption(leaf, fi) ||
2380 btrfs_file_extent_other_encoding(leaf, fi))
2387 * Note the backref might has changed, and in this case we just return 0.
2389 static noinline int relink_extent_backref(struct btrfs_path *path,
2390 struct sa_defrag_extent_backref *prev,
2391 struct sa_defrag_extent_backref *backref)
2393 struct btrfs_file_extent_item *extent;
2394 struct btrfs_file_extent_item *item;
2395 struct btrfs_ordered_extent *ordered;
2396 struct btrfs_trans_handle *trans;
2397 struct btrfs_fs_info *fs_info;
2398 struct btrfs_root *root;
2399 struct btrfs_key key;
2400 struct extent_buffer *leaf;
2401 struct old_sa_defrag_extent *old = backref->old;
2402 struct new_sa_defrag_extent *new = old->new;
2403 struct inode *src_inode = new->inode;
2404 struct inode *inode;
2405 struct extent_state *cached = NULL;
2414 if (prev && prev->root_id == backref->root_id &&
2415 prev->inum == backref->inum &&
2416 prev->file_pos + prev->num_bytes == backref->file_pos)
2419 /* step 1: get root */
2420 key.objectid = backref->root_id;
2421 key.type = BTRFS_ROOT_ITEM_KEY;
2422 key.offset = (u64)-1;
2424 fs_info = BTRFS_I(src_inode)->root->fs_info;
2425 index = srcu_read_lock(&fs_info->subvol_srcu);
2427 root = btrfs_read_fs_root_no_name(fs_info, &key);
2429 srcu_read_unlock(&fs_info->subvol_srcu, index);
2430 if (PTR_ERR(root) == -ENOENT)
2432 return PTR_ERR(root);
2435 if (btrfs_root_readonly(root)) {
2436 srcu_read_unlock(&fs_info->subvol_srcu, index);
2440 /* step 2: get inode */
2441 key.objectid = backref->inum;
2442 key.type = BTRFS_INODE_ITEM_KEY;
2445 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2446 if (IS_ERR(inode)) {
2447 srcu_read_unlock(&fs_info->subvol_srcu, index);
2451 srcu_read_unlock(&fs_info->subvol_srcu, index);
2453 /* step 3: relink backref */
2454 lock_start = backref->file_pos;
2455 lock_end = backref->file_pos + backref->num_bytes - 1;
2456 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2459 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2461 btrfs_put_ordered_extent(ordered);
2465 trans = btrfs_join_transaction(root);
2466 if (IS_ERR(trans)) {
2467 ret = PTR_ERR(trans);
2471 key.objectid = backref->inum;
2472 key.type = BTRFS_EXTENT_DATA_KEY;
2473 key.offset = backref->file_pos;
2475 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2478 } else if (ret > 0) {
2483 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2484 struct btrfs_file_extent_item);
2486 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2487 backref->generation)
2490 btrfs_release_path(path);
2492 start = backref->file_pos;
2493 if (backref->extent_offset < old->extent_offset + old->offset)
2494 start += old->extent_offset + old->offset -
2495 backref->extent_offset;
2497 len = min(backref->extent_offset + backref->num_bytes,
2498 old->extent_offset + old->offset + old->len);
2499 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2501 ret = btrfs_drop_extents(trans, root, inode, start,
2506 key.objectid = btrfs_ino(inode);
2507 key.type = BTRFS_EXTENT_DATA_KEY;
2510 path->leave_spinning = 1;
2512 struct btrfs_file_extent_item *fi;
2514 struct btrfs_key found_key;
2516 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2521 leaf = path->nodes[0];
2522 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2524 fi = btrfs_item_ptr(leaf, path->slots[0],
2525 struct btrfs_file_extent_item);
2526 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2528 if (extent_len + found_key.offset == start &&
2529 relink_is_mergable(leaf, fi, new)) {
2530 btrfs_set_file_extent_num_bytes(leaf, fi,
2532 btrfs_mark_buffer_dirty(leaf);
2533 inode_add_bytes(inode, len);
2539 btrfs_release_path(path);
2544 ret = btrfs_insert_empty_item(trans, root, path, &key,
2547 btrfs_abort_transaction(trans, root, ret);
2551 leaf = path->nodes[0];
2552 item = btrfs_item_ptr(leaf, path->slots[0],
2553 struct btrfs_file_extent_item);
2554 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2555 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2556 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2557 btrfs_set_file_extent_num_bytes(leaf, item, len);
2558 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2559 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2560 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2561 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2562 btrfs_set_file_extent_encryption(leaf, item, 0);
2563 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2565 btrfs_mark_buffer_dirty(leaf);
2566 inode_add_bytes(inode, len);
2567 btrfs_release_path(path);
2569 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2571 backref->root_id, backref->inum,
2572 new->file_pos, 0); /* start - extent_offset */
2574 btrfs_abort_transaction(trans, root, ret);
2580 btrfs_release_path(path);
2581 path->leave_spinning = 0;
2582 btrfs_end_transaction(trans, root);
2584 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2590 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2592 struct old_sa_defrag_extent *old, *tmp;
2597 list_for_each_entry_safe(old, tmp, &new->head, list) {
2598 list_del(&old->list);
2604 static void relink_file_extents(struct new_sa_defrag_extent *new)
2606 struct btrfs_path *path;
2607 struct sa_defrag_extent_backref *backref;
2608 struct sa_defrag_extent_backref *prev = NULL;
2609 struct inode *inode;
2610 struct btrfs_root *root;
2611 struct rb_node *node;
2615 root = BTRFS_I(inode)->root;
2617 path = btrfs_alloc_path();
2621 if (!record_extent_backrefs(path, new)) {
2622 btrfs_free_path(path);
2625 btrfs_release_path(path);
2628 node = rb_first(&new->root);
2631 rb_erase(node, &new->root);
2633 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2635 ret = relink_extent_backref(path, prev, backref);
2648 btrfs_free_path(path);
2650 free_sa_defrag_extent(new);
2652 atomic_dec(&root->fs_info->defrag_running);
2653 wake_up(&root->fs_info->transaction_wait);
2656 static struct new_sa_defrag_extent *
2657 record_old_file_extents(struct inode *inode,
2658 struct btrfs_ordered_extent *ordered)
2660 struct btrfs_root *root = BTRFS_I(inode)->root;
2661 struct btrfs_path *path;
2662 struct btrfs_key key;
2663 struct old_sa_defrag_extent *old;
2664 struct new_sa_defrag_extent *new;
2667 new = kmalloc(sizeof(*new), GFP_NOFS);
2672 new->file_pos = ordered->file_offset;
2673 new->len = ordered->len;
2674 new->bytenr = ordered->start;
2675 new->disk_len = ordered->disk_len;
2676 new->compress_type = ordered->compress_type;
2677 new->root = RB_ROOT;
2678 INIT_LIST_HEAD(&new->head);
2680 path = btrfs_alloc_path();
2684 key.objectid = btrfs_ino(inode);
2685 key.type = BTRFS_EXTENT_DATA_KEY;
2686 key.offset = new->file_pos;
2688 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2691 if (ret > 0 && path->slots[0] > 0)
2694 /* find out all the old extents for the file range */
2696 struct btrfs_file_extent_item *extent;
2697 struct extent_buffer *l;
2706 slot = path->slots[0];
2708 if (slot >= btrfs_header_nritems(l)) {
2709 ret = btrfs_next_leaf(root, path);
2717 btrfs_item_key_to_cpu(l, &key, slot);
2719 if (key.objectid != btrfs_ino(inode))
2721 if (key.type != BTRFS_EXTENT_DATA_KEY)
2723 if (key.offset >= new->file_pos + new->len)
2726 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2728 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2729 if (key.offset + num_bytes < new->file_pos)
2732 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2736 extent_offset = btrfs_file_extent_offset(l, extent);
2738 old = kmalloc(sizeof(*old), GFP_NOFS);
2742 offset = max(new->file_pos, key.offset);
2743 end = min(new->file_pos + new->len, key.offset + num_bytes);
2745 old->bytenr = disk_bytenr;
2746 old->extent_offset = extent_offset;
2747 old->offset = offset - key.offset;
2748 old->len = end - offset;
2751 list_add_tail(&old->list, &new->head);
2757 btrfs_free_path(path);
2758 atomic_inc(&root->fs_info->defrag_running);
2763 btrfs_free_path(path);
2765 free_sa_defrag_extent(new);
2769 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2772 struct btrfs_block_group_cache *cache;
2774 cache = btrfs_lookup_block_group(root->fs_info, start);
2777 spin_lock(&cache->lock);
2778 cache->delalloc_bytes -= len;
2779 spin_unlock(&cache->lock);
2781 btrfs_put_block_group(cache);
2784 /* as ordered data IO finishes, this gets called so we can finish
2785 * an ordered extent if the range of bytes in the file it covers are
2788 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2790 struct inode *inode = ordered_extent->inode;
2791 struct btrfs_root *root = BTRFS_I(inode)->root;
2792 struct btrfs_trans_handle *trans = NULL;
2793 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2794 struct extent_state *cached_state = NULL;
2795 struct new_sa_defrag_extent *new = NULL;
2796 int compress_type = 0;
2798 u64 logical_len = ordered_extent->len;
2800 bool truncated = false;
2802 nolock = btrfs_is_free_space_inode(inode);
2804 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2809 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2810 ordered_extent->file_offset +
2811 ordered_extent->len - 1);
2813 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2815 logical_len = ordered_extent->truncated_len;
2816 /* Truncated the entire extent, don't bother adding */
2821 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2822 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2823 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2825 trans = btrfs_join_transaction_nolock(root);
2827 trans = btrfs_join_transaction(root);
2828 if (IS_ERR(trans)) {
2829 ret = PTR_ERR(trans);
2833 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2834 ret = btrfs_update_inode_fallback(trans, root, inode);
2835 if (ret) /* -ENOMEM or corruption */
2836 btrfs_abort_transaction(trans, root, ret);
2840 lock_extent_bits(io_tree, ordered_extent->file_offset,
2841 ordered_extent->file_offset + ordered_extent->len - 1,
2844 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2845 ordered_extent->file_offset + ordered_extent->len - 1,
2846 EXTENT_DEFRAG, 1, cached_state);
2848 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2849 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2850 /* the inode is shared */
2851 new = record_old_file_extents(inode, ordered_extent);
2853 clear_extent_bit(io_tree, ordered_extent->file_offset,
2854 ordered_extent->file_offset + ordered_extent->len - 1,
2855 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2859 trans = btrfs_join_transaction_nolock(root);
2861 trans = btrfs_join_transaction(root);
2862 if (IS_ERR(trans)) {
2863 ret = PTR_ERR(trans);
2868 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2870 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2871 compress_type = ordered_extent->compress_type;
2872 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2873 BUG_ON(compress_type);
2874 ret = btrfs_mark_extent_written(trans, inode,
2875 ordered_extent->file_offset,
2876 ordered_extent->file_offset +
2879 BUG_ON(root == root->fs_info->tree_root);
2880 ret = insert_reserved_file_extent(trans, inode,
2881 ordered_extent->file_offset,
2882 ordered_extent->start,
2883 ordered_extent->disk_len,
2884 logical_len, logical_len,
2885 compress_type, 0, 0,
2886 BTRFS_FILE_EXTENT_REG);
2888 btrfs_release_delalloc_bytes(root,
2889 ordered_extent->start,
2890 ordered_extent->disk_len);
2892 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2893 ordered_extent->file_offset, ordered_extent->len,
2896 btrfs_abort_transaction(trans, root, ret);
2900 add_pending_csums(trans, inode, ordered_extent->file_offset,
2901 &ordered_extent->list);
2903 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2904 ret = btrfs_update_inode_fallback(trans, root, inode);
2905 if (ret) { /* -ENOMEM or corruption */
2906 btrfs_abort_transaction(trans, root, ret);
2911 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2912 ordered_extent->file_offset +
2913 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2915 if (root != root->fs_info->tree_root)
2916 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2918 btrfs_end_transaction(trans, root);
2920 if (ret || truncated) {
2924 start = ordered_extent->file_offset + logical_len;
2926 start = ordered_extent->file_offset;
2927 end = ordered_extent->file_offset + ordered_extent->len - 1;
2928 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2930 /* Drop the cache for the part of the extent we didn't write. */
2931 btrfs_drop_extent_cache(inode, start, end, 0);
2934 * If the ordered extent had an IOERR or something else went
2935 * wrong we need to return the space for this ordered extent
2936 * back to the allocator. We only free the extent in the
2937 * truncated case if we didn't write out the extent at all.
2939 if ((ret || !logical_len) &&
2940 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2941 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2942 btrfs_free_reserved_extent(root, ordered_extent->start,
2943 ordered_extent->disk_len, 1);
2948 * This needs to be done to make sure anybody waiting knows we are done
2949 * updating everything for this ordered extent.
2951 btrfs_remove_ordered_extent(inode, ordered_extent);
2953 /* for snapshot-aware defrag */
2956 free_sa_defrag_extent(new);
2957 atomic_dec(&root->fs_info->defrag_running);
2959 relink_file_extents(new);
2964 btrfs_put_ordered_extent(ordered_extent);
2965 /* once for the tree */
2966 btrfs_put_ordered_extent(ordered_extent);
2971 static void finish_ordered_fn(struct btrfs_work *work)
2973 struct btrfs_ordered_extent *ordered_extent;
2974 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2975 btrfs_finish_ordered_io(ordered_extent);
2978 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2979 struct extent_state *state, int uptodate)
2981 struct inode *inode = page->mapping->host;
2982 struct btrfs_root *root = BTRFS_I(inode)->root;
2983 struct btrfs_ordered_extent *ordered_extent = NULL;
2984 struct btrfs_workqueue *wq;
2985 btrfs_work_func_t func;
2987 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2989 ClearPagePrivate2(page);
2990 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2991 end - start + 1, uptodate))
2994 if (btrfs_is_free_space_inode(inode)) {
2995 wq = root->fs_info->endio_freespace_worker;
2996 func = btrfs_freespace_write_helper;
2998 wq = root->fs_info->endio_write_workers;
2999 func = btrfs_endio_write_helper;
3002 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3004 btrfs_queue_work(wq, &ordered_extent->work);
3009 static int __readpage_endio_check(struct inode *inode,
3010 struct btrfs_io_bio *io_bio,
3011 int icsum, struct page *page,
3012 int pgoff, u64 start, size_t len)
3017 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
3018 DEFAULT_RATELIMIT_BURST);
3020 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3022 kaddr = kmap_atomic(page);
3023 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3024 btrfs_csum_final(csum, (char *)&csum);
3025 if (csum != csum_expected)
3028 kunmap_atomic(kaddr);
3031 if (__ratelimit(&_rs))
3032 btrfs_warn(BTRFS_I(inode)->root->fs_info,
3033 "csum failed ino %llu off %llu csum %u expected csum %u",
3034 btrfs_ino(inode), start, csum, csum_expected);
3035 memset(kaddr + pgoff, 1, len);
3036 flush_dcache_page(page);
3037 kunmap_atomic(kaddr);
3038 if (csum_expected == 0)
3044 * when reads are done, we need to check csums to verify the data is correct
3045 * if there's a match, we allow the bio to finish. If not, the code in
3046 * extent_io.c will try to find good copies for us.
3048 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3049 u64 phy_offset, struct page *page,
3050 u64 start, u64 end, int mirror)
3052 size_t offset = start - page_offset(page);
3053 struct inode *inode = page->mapping->host;
3054 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3055 struct btrfs_root *root = BTRFS_I(inode)->root;
3057 if (PageChecked(page)) {
3058 ClearPageChecked(page);
3062 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3065 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3066 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3067 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
3072 phy_offset >>= inode->i_sb->s_blocksize_bits;
3073 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3074 start, (size_t)(end - start + 1));
3077 struct delayed_iput {
3078 struct list_head list;
3079 struct inode *inode;
3082 /* JDM: If this is fs-wide, why can't we add a pointer to
3083 * btrfs_inode instead and avoid the allocation? */
3084 void btrfs_add_delayed_iput(struct inode *inode)
3086 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3087 struct delayed_iput *delayed;
3089 if (atomic_add_unless(&inode->i_count, -1, 1))
3092 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
3093 delayed->inode = inode;
3095 spin_lock(&fs_info->delayed_iput_lock);
3096 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
3097 spin_unlock(&fs_info->delayed_iput_lock);
3100 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3103 struct btrfs_fs_info *fs_info = root->fs_info;
3104 struct delayed_iput *delayed;
3107 spin_lock(&fs_info->delayed_iput_lock);
3108 empty = list_empty(&fs_info->delayed_iputs);
3109 spin_unlock(&fs_info->delayed_iput_lock);
3113 down_read(&fs_info->delayed_iput_sem);
3115 spin_lock(&fs_info->delayed_iput_lock);
3116 list_splice_init(&fs_info->delayed_iputs, &list);
3117 spin_unlock(&fs_info->delayed_iput_lock);
3119 while (!list_empty(&list)) {
3120 delayed = list_entry(list.next, struct delayed_iput, list);
3121 list_del(&delayed->list);
3122 iput(delayed->inode);
3126 up_read(&root->fs_info->delayed_iput_sem);
3130 * This is called in transaction commit time. If there are no orphan
3131 * files in the subvolume, it removes orphan item and frees block_rsv
3134 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3135 struct btrfs_root *root)
3137 struct btrfs_block_rsv *block_rsv;
3140 if (atomic_read(&root->orphan_inodes) ||
3141 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3144 spin_lock(&root->orphan_lock);
3145 if (atomic_read(&root->orphan_inodes)) {
3146 spin_unlock(&root->orphan_lock);
3150 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3151 spin_unlock(&root->orphan_lock);
3155 block_rsv = root->orphan_block_rsv;
3156 root->orphan_block_rsv = NULL;
3157 spin_unlock(&root->orphan_lock);
3159 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3160 btrfs_root_refs(&root->root_item) > 0) {
3161 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3162 root->root_key.objectid);
3164 btrfs_abort_transaction(trans, root, ret);
3166 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3171 WARN_ON(block_rsv->size > 0);
3172 btrfs_free_block_rsv(root, block_rsv);
3177 * This creates an orphan entry for the given inode in case something goes
3178 * wrong in the middle of an unlink/truncate.
3180 * NOTE: caller of this function should reserve 5 units of metadata for
3183 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3185 struct btrfs_root *root = BTRFS_I(inode)->root;
3186 struct btrfs_block_rsv *block_rsv = NULL;
3191 if (!root->orphan_block_rsv) {
3192 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3197 spin_lock(&root->orphan_lock);
3198 if (!root->orphan_block_rsv) {
3199 root->orphan_block_rsv = block_rsv;
3200 } else if (block_rsv) {
3201 btrfs_free_block_rsv(root, block_rsv);
3205 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3206 &BTRFS_I(inode)->runtime_flags)) {
3209 * For proper ENOSPC handling, we should do orphan
3210 * cleanup when mounting. But this introduces backward
3211 * compatibility issue.
3213 if (!xchg(&root->orphan_item_inserted, 1))
3219 atomic_inc(&root->orphan_inodes);
3222 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3223 &BTRFS_I(inode)->runtime_flags))
3225 spin_unlock(&root->orphan_lock);
3227 /* grab metadata reservation from transaction handle */
3229 ret = btrfs_orphan_reserve_metadata(trans, inode);
3230 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3233 /* insert an orphan item to track this unlinked/truncated file */
3235 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3237 atomic_dec(&root->orphan_inodes);
3239 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3240 &BTRFS_I(inode)->runtime_flags);
3241 btrfs_orphan_release_metadata(inode);
3243 if (ret != -EEXIST) {
3244 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3245 &BTRFS_I(inode)->runtime_flags);
3246 btrfs_abort_transaction(trans, root, ret);
3253 /* insert an orphan item to track subvolume contains orphan files */
3255 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3256 root->root_key.objectid);
3257 if (ret && ret != -EEXIST) {
3258 btrfs_abort_transaction(trans, root, ret);
3266 * We have done the truncate/delete so we can go ahead and remove the orphan
3267 * item for this particular inode.
3269 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3270 struct inode *inode)
3272 struct btrfs_root *root = BTRFS_I(inode)->root;
3273 int delete_item = 0;
3274 int release_rsv = 0;
3277 spin_lock(&root->orphan_lock);
3278 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3279 &BTRFS_I(inode)->runtime_flags))
3282 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3283 &BTRFS_I(inode)->runtime_flags))
3285 spin_unlock(&root->orphan_lock);
3288 atomic_dec(&root->orphan_inodes);
3290 ret = btrfs_del_orphan_item(trans, root,
3295 btrfs_orphan_release_metadata(inode);
3301 * this cleans up any orphans that may be left on the list from the last use
3304 int btrfs_orphan_cleanup(struct btrfs_root *root)
3306 struct btrfs_path *path;
3307 struct extent_buffer *leaf;
3308 struct btrfs_key key, found_key;
3309 struct btrfs_trans_handle *trans;
3310 struct inode *inode;
3311 u64 last_objectid = 0;
3312 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3314 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3317 path = btrfs_alloc_path();
3324 key.objectid = BTRFS_ORPHAN_OBJECTID;
3325 key.type = BTRFS_ORPHAN_ITEM_KEY;
3326 key.offset = (u64)-1;
3329 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3334 * if ret == 0 means we found what we were searching for, which
3335 * is weird, but possible, so only screw with path if we didn't
3336 * find the key and see if we have stuff that matches
3340 if (path->slots[0] == 0)
3345 /* pull out the item */
3346 leaf = path->nodes[0];
3347 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3349 /* make sure the item matches what we want */
3350 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3352 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3355 /* release the path since we're done with it */
3356 btrfs_release_path(path);
3359 * this is where we are basically btrfs_lookup, without the
3360 * crossing root thing. we store the inode number in the
3361 * offset of the orphan item.
3364 if (found_key.offset == last_objectid) {
3365 btrfs_err(root->fs_info,
3366 "Error removing orphan entry, stopping orphan cleanup");
3371 last_objectid = found_key.offset;
3373 found_key.objectid = found_key.offset;
3374 found_key.type = BTRFS_INODE_ITEM_KEY;
3375 found_key.offset = 0;
3376 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3377 ret = PTR_ERR_OR_ZERO(inode);
3378 if (ret && ret != -ESTALE)
3381 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3382 struct btrfs_root *dead_root;
3383 struct btrfs_fs_info *fs_info = root->fs_info;
3384 int is_dead_root = 0;
3387 * this is an orphan in the tree root. Currently these
3388 * could come from 2 sources:
3389 * a) a snapshot deletion in progress
3390 * b) a free space cache inode
3391 * We need to distinguish those two, as the snapshot
3392 * orphan must not get deleted.
3393 * find_dead_roots already ran before us, so if this
3394 * is a snapshot deletion, we should find the root
3395 * in the dead_roots list
3397 spin_lock(&fs_info->trans_lock);
3398 list_for_each_entry(dead_root, &fs_info->dead_roots,
3400 if (dead_root->root_key.objectid ==
3401 found_key.objectid) {
3406 spin_unlock(&fs_info->trans_lock);
3408 /* prevent this orphan from being found again */
3409 key.offset = found_key.objectid - 1;
3414 * Inode is already gone but the orphan item is still there,
3415 * kill the orphan item.
3417 if (ret == -ESTALE) {
3418 trans = btrfs_start_transaction(root, 1);
3419 if (IS_ERR(trans)) {
3420 ret = PTR_ERR(trans);
3423 btrfs_debug(root->fs_info, "auto deleting %Lu",
3424 found_key.objectid);
3425 ret = btrfs_del_orphan_item(trans, root,
3426 found_key.objectid);
3427 btrfs_end_transaction(trans, root);
3434 * add this inode to the orphan list so btrfs_orphan_del does
3435 * the proper thing when we hit it
3437 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3438 &BTRFS_I(inode)->runtime_flags);
3439 atomic_inc(&root->orphan_inodes);
3441 /* if we have links, this was a truncate, lets do that */
3442 if (inode->i_nlink) {
3443 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3449 /* 1 for the orphan item deletion. */
3450 trans = btrfs_start_transaction(root, 1);
3451 if (IS_ERR(trans)) {
3453 ret = PTR_ERR(trans);
3456 ret = btrfs_orphan_add(trans, inode);
3457 btrfs_end_transaction(trans, root);
3463 ret = btrfs_truncate(inode);
3465 btrfs_orphan_del(NULL, inode);
3470 /* this will do delete_inode and everything for us */
3475 /* release the path since we're done with it */
3476 btrfs_release_path(path);
3478 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3480 if (root->orphan_block_rsv)
3481 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3484 if (root->orphan_block_rsv ||
3485 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3486 trans = btrfs_join_transaction(root);
3488 btrfs_end_transaction(trans, root);
3492 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3494 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3498 btrfs_err(root->fs_info,
3499 "could not do orphan cleanup %d", ret);
3500 btrfs_free_path(path);
3505 * very simple check to peek ahead in the leaf looking for xattrs. If we
3506 * don't find any xattrs, we know there can't be any acls.
3508 * slot is the slot the inode is in, objectid is the objectid of the inode
3510 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3511 int slot, u64 objectid,
3512 int *first_xattr_slot)
3514 u32 nritems = btrfs_header_nritems(leaf);
3515 struct btrfs_key found_key;
3516 static u64 xattr_access = 0;
3517 static u64 xattr_default = 0;
3520 if (!xattr_access) {
3521 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3522 strlen(POSIX_ACL_XATTR_ACCESS));
3523 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3524 strlen(POSIX_ACL_XATTR_DEFAULT));
3528 *first_xattr_slot = -1;
3529 while (slot < nritems) {
3530 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3532 /* we found a different objectid, there must not be acls */
3533 if (found_key.objectid != objectid)
3536 /* we found an xattr, assume we've got an acl */
3537 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3538 if (*first_xattr_slot == -1)
3539 *first_xattr_slot = slot;
3540 if (found_key.offset == xattr_access ||
3541 found_key.offset == xattr_default)
3546 * we found a key greater than an xattr key, there can't
3547 * be any acls later on
3549 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3556 * it goes inode, inode backrefs, xattrs, extents,
3557 * so if there are a ton of hard links to an inode there can
3558 * be a lot of backrefs. Don't waste time searching too hard,
3559 * this is just an optimization
3564 /* we hit the end of the leaf before we found an xattr or
3565 * something larger than an xattr. We have to assume the inode
3568 if (*first_xattr_slot == -1)
3569 *first_xattr_slot = slot;
3574 * read an inode from the btree into the in-memory inode
3576 static void btrfs_read_locked_inode(struct inode *inode)
3578 struct btrfs_path *path;
3579 struct extent_buffer *leaf;
3580 struct btrfs_inode_item *inode_item;
3581 struct btrfs_root *root = BTRFS_I(inode)->root;
3582 struct btrfs_key location;
3587 bool filled = false;
3588 int first_xattr_slot;
3590 ret = btrfs_fill_inode(inode, &rdev);
3594 path = btrfs_alloc_path();
3598 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3600 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3604 leaf = path->nodes[0];
3609 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3610 struct btrfs_inode_item);
3611 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3612 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3613 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3614 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3615 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3617 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3618 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3620 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3621 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3623 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3624 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3626 BTRFS_I(inode)->i_otime.tv_sec =
3627 btrfs_timespec_sec(leaf, &inode_item->otime);
3628 BTRFS_I(inode)->i_otime.tv_nsec =
3629 btrfs_timespec_nsec(leaf, &inode_item->otime);
3631 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3632 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3633 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3635 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3636 inode->i_generation = BTRFS_I(inode)->generation;
3638 rdev = btrfs_inode_rdev(leaf, inode_item);
3640 BTRFS_I(inode)->index_cnt = (u64)-1;
3641 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3645 * If we were modified in the current generation and evicted from memory
3646 * and then re-read we need to do a full sync since we don't have any
3647 * idea about which extents were modified before we were evicted from
3650 * This is required for both inode re-read from disk and delayed inode
3651 * in delayed_nodes_tree.
3653 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3654 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3655 &BTRFS_I(inode)->runtime_flags);
3658 if (inode->i_nlink != 1 ||
3659 path->slots[0] >= btrfs_header_nritems(leaf))
3662 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3663 if (location.objectid != btrfs_ino(inode))
3666 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3667 if (location.type == BTRFS_INODE_REF_KEY) {
3668 struct btrfs_inode_ref *ref;
3670 ref = (struct btrfs_inode_ref *)ptr;
3671 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3672 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3673 struct btrfs_inode_extref *extref;
3675 extref = (struct btrfs_inode_extref *)ptr;
3676 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3681 * try to precache a NULL acl entry for files that don't have
3682 * any xattrs or acls
3684 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3685 btrfs_ino(inode), &first_xattr_slot);
3686 if (first_xattr_slot != -1) {
3687 path->slots[0] = first_xattr_slot;
3688 ret = btrfs_load_inode_props(inode, path);
3690 btrfs_err(root->fs_info,
3691 "error loading props for ino %llu (root %llu): %d",
3693 root->root_key.objectid, ret);
3695 btrfs_free_path(path);
3698 cache_no_acl(inode);
3700 switch (inode->i_mode & S_IFMT) {
3702 inode->i_mapping->a_ops = &btrfs_aops;
3703 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3704 inode->i_fop = &btrfs_file_operations;
3705 inode->i_op = &btrfs_file_inode_operations;
3708 inode->i_fop = &btrfs_dir_file_operations;
3709 if (root == root->fs_info->tree_root)
3710 inode->i_op = &btrfs_dir_ro_inode_operations;
3712 inode->i_op = &btrfs_dir_inode_operations;
3715 inode->i_op = &btrfs_symlink_inode_operations;
3716 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3719 inode->i_op = &btrfs_special_inode_operations;
3720 init_special_inode(inode, inode->i_mode, rdev);
3724 btrfs_update_iflags(inode);
3728 btrfs_free_path(path);
3729 make_bad_inode(inode);
3733 * given a leaf and an inode, copy the inode fields into the leaf
3735 static void fill_inode_item(struct btrfs_trans_handle *trans,
3736 struct extent_buffer *leaf,
3737 struct btrfs_inode_item *item,
3738 struct inode *inode)
3740 struct btrfs_map_token token;
3742 btrfs_init_map_token(&token);
3744 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3745 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3746 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3748 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3749 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3751 btrfs_set_token_timespec_sec(leaf, &item->atime,
3752 inode->i_atime.tv_sec, &token);
3753 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3754 inode->i_atime.tv_nsec, &token);
3756 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3757 inode->i_mtime.tv_sec, &token);
3758 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3759 inode->i_mtime.tv_nsec, &token);
3761 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3762 inode->i_ctime.tv_sec, &token);
3763 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3764 inode->i_ctime.tv_nsec, &token);
3766 btrfs_set_token_timespec_sec(leaf, &item->otime,
3767 BTRFS_I(inode)->i_otime.tv_sec, &token);
3768 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3769 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3771 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3773 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3775 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3776 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3777 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3778 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3779 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3783 * copy everything in the in-memory inode into the btree.
3785 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3786 struct btrfs_root *root, struct inode *inode)
3788 struct btrfs_inode_item *inode_item;
3789 struct btrfs_path *path;
3790 struct extent_buffer *leaf;
3793 path = btrfs_alloc_path();
3797 path->leave_spinning = 1;
3798 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3806 leaf = path->nodes[0];
3807 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3808 struct btrfs_inode_item);
3810 fill_inode_item(trans, leaf, inode_item, inode);
3811 btrfs_mark_buffer_dirty(leaf);
3812 btrfs_set_inode_last_trans(trans, inode);
3815 btrfs_free_path(path);
3820 * copy everything in the in-memory inode into the btree.
3822 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3823 struct btrfs_root *root, struct inode *inode)
3828 * If the inode is a free space inode, we can deadlock during commit
3829 * if we put it into the delayed code.
3831 * The data relocation inode should also be directly updated
3834 if (!btrfs_is_free_space_inode(inode)
3835 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3836 && !root->fs_info->log_root_recovering) {
3837 btrfs_update_root_times(trans, root);
3839 ret = btrfs_delayed_update_inode(trans, root, inode);
3841 btrfs_set_inode_last_trans(trans, inode);
3845 return btrfs_update_inode_item(trans, root, inode);
3848 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3849 struct btrfs_root *root,
3850 struct inode *inode)
3854 ret = btrfs_update_inode(trans, root, inode);
3856 return btrfs_update_inode_item(trans, root, inode);
3861 * unlink helper that gets used here in inode.c and in the tree logging
3862 * recovery code. It remove a link in a directory with a given name, and
3863 * also drops the back refs in the inode to the directory
3865 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3866 struct btrfs_root *root,
3867 struct inode *dir, struct inode *inode,
3868 const char *name, int name_len)
3870 struct btrfs_path *path;
3872 struct extent_buffer *leaf;
3873 struct btrfs_dir_item *di;
3874 struct btrfs_key key;
3876 u64 ino = btrfs_ino(inode);
3877 u64 dir_ino = btrfs_ino(dir);
3879 path = btrfs_alloc_path();
3885 path->leave_spinning = 1;
3886 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3887 name, name_len, -1);
3896 leaf = path->nodes[0];
3897 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3898 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3901 btrfs_release_path(path);
3904 * If we don't have dir index, we have to get it by looking up
3905 * the inode ref, since we get the inode ref, remove it directly,
3906 * it is unnecessary to do delayed deletion.
3908 * But if we have dir index, needn't search inode ref to get it.
3909 * Since the inode ref is close to the inode item, it is better
3910 * that we delay to delete it, and just do this deletion when
3911 * we update the inode item.
3913 if (BTRFS_I(inode)->dir_index) {
3914 ret = btrfs_delayed_delete_inode_ref(inode);
3916 index = BTRFS_I(inode)->dir_index;
3921 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3924 btrfs_info(root->fs_info,
3925 "failed to delete reference to %.*s, inode %llu parent %llu",
3926 name_len, name, ino, dir_ino);
3927 btrfs_abort_transaction(trans, root, ret);
3931 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3933 btrfs_abort_transaction(trans, root, ret);
3937 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3939 if (ret != 0 && ret != -ENOENT) {
3940 btrfs_abort_transaction(trans, root, ret);
3944 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3949 btrfs_abort_transaction(trans, root, ret);
3951 btrfs_free_path(path);
3955 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3956 inode_inc_iversion(inode);
3957 inode_inc_iversion(dir);
3958 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3959 ret = btrfs_update_inode(trans, root, dir);
3964 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3965 struct btrfs_root *root,
3966 struct inode *dir, struct inode *inode,
3967 const char *name, int name_len)
3970 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3973 ret = btrfs_update_inode(trans, root, inode);
3979 * helper to start transaction for unlink and rmdir.
3981 * unlink and rmdir are special in btrfs, they do not always free space, so
3982 * if we cannot make our reservations the normal way try and see if there is
3983 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3984 * allow the unlink to occur.
3986 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3988 struct btrfs_trans_handle *trans;
3989 struct btrfs_root *root = BTRFS_I(dir)->root;
3993 * 1 for the possible orphan item
3994 * 1 for the dir item
3995 * 1 for the dir index
3996 * 1 for the inode ref
3999 trans = btrfs_start_transaction(root, 5);
4000 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
4003 if (PTR_ERR(trans) == -ENOSPC) {
4004 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4006 trans = btrfs_start_transaction(root, 0);
4009 ret = btrfs_cond_migrate_bytes(root->fs_info,
4010 &root->fs_info->trans_block_rsv,
4013 btrfs_end_transaction(trans, root);
4014 return ERR_PTR(ret);
4016 trans->block_rsv = &root->fs_info->trans_block_rsv;
4017 trans->bytes_reserved = num_bytes;
4022 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4024 struct btrfs_root *root = BTRFS_I(dir)->root;
4025 struct btrfs_trans_handle *trans;
4026 struct inode *inode = d_inode(dentry);
4029 trans = __unlink_start_trans(dir);
4031 return PTR_ERR(trans);
4033 btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
4035 ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4036 dentry->d_name.name, dentry->d_name.len);
4040 if (inode->i_nlink == 0) {
4041 ret = btrfs_orphan_add(trans, inode);
4047 btrfs_end_transaction(trans, root);
4048 btrfs_btree_balance_dirty(root);
4052 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4053 struct btrfs_root *root,
4054 struct inode *dir, u64 objectid,
4055 const char *name, int name_len)
4057 struct btrfs_path *path;
4058 struct extent_buffer *leaf;
4059 struct btrfs_dir_item *di;
4060 struct btrfs_key key;
4063 u64 dir_ino = btrfs_ino(dir);
4065 path = btrfs_alloc_path();
4069 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4070 name, name_len, -1);
4071 if (IS_ERR_OR_NULL(di)) {
4079 leaf = path->nodes[0];
4080 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4081 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4082 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4084 btrfs_abort_transaction(trans, root, ret);
4087 btrfs_release_path(path);
4089 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4090 objectid, root->root_key.objectid,
4091 dir_ino, &index, name, name_len);
4093 if (ret != -ENOENT) {
4094 btrfs_abort_transaction(trans, root, ret);
4097 di = btrfs_search_dir_index_item(root, path, dir_ino,
4099 if (IS_ERR_OR_NULL(di)) {
4104 btrfs_abort_transaction(trans, root, ret);
4108 leaf = path->nodes[0];
4109 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4110 btrfs_release_path(path);
4113 btrfs_release_path(path);
4115 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4117 btrfs_abort_transaction(trans, root, ret);
4121 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4122 inode_inc_iversion(dir);
4123 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4124 ret = btrfs_update_inode_fallback(trans, root, dir);
4126 btrfs_abort_transaction(trans, root, ret);
4128 btrfs_free_path(path);
4132 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4134 struct inode *inode = d_inode(dentry);
4136 struct btrfs_root *root = BTRFS_I(dir)->root;
4137 struct btrfs_trans_handle *trans;
4139 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4141 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4144 trans = __unlink_start_trans(dir);
4146 return PTR_ERR(trans);
4148 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4149 err = btrfs_unlink_subvol(trans, root, dir,
4150 BTRFS_I(inode)->location.objectid,
4151 dentry->d_name.name,
4152 dentry->d_name.len);
4156 err = btrfs_orphan_add(trans, inode);
4160 /* now the directory is empty */
4161 err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4162 dentry->d_name.name, dentry->d_name.len);
4164 btrfs_i_size_write(inode, 0);
4166 btrfs_end_transaction(trans, root);
4167 btrfs_btree_balance_dirty(root);
4172 static int truncate_space_check(struct btrfs_trans_handle *trans,
4173 struct btrfs_root *root,
4178 bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4179 ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4180 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4182 trans->bytes_reserved += bytes_deleted;
4188 * this can truncate away extent items, csum items and directory items.
4189 * It starts at a high offset and removes keys until it can't find
4190 * any higher than new_size
4192 * csum items that cross the new i_size are truncated to the new size
4195 * min_type is the minimum key type to truncate down to. If set to 0, this
4196 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4198 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4199 struct btrfs_root *root,
4200 struct inode *inode,
4201 u64 new_size, u32 min_type)
4203 struct btrfs_path *path;
4204 struct extent_buffer *leaf;
4205 struct btrfs_file_extent_item *fi;
4206 struct btrfs_key key;
4207 struct btrfs_key found_key;
4208 u64 extent_start = 0;
4209 u64 extent_num_bytes = 0;
4210 u64 extent_offset = 0;
4212 u64 last_size = (u64)-1;
4213 u32 found_type = (u8)-1;
4216 int pending_del_nr = 0;
4217 int pending_del_slot = 0;
4218 int extent_type = -1;
4221 u64 ino = btrfs_ino(inode);
4222 u64 bytes_deleted = 0;
4224 bool should_throttle = 0;
4225 bool should_end = 0;
4227 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4230 * for non-free space inodes and ref cows, we want to back off from
4233 if (!btrfs_is_free_space_inode(inode) &&
4234 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4237 path = btrfs_alloc_path();
4243 * We want to drop from the next block forward in case this new size is
4244 * not block aligned since we will be keeping the last block of the
4245 * extent just the way it is.
4247 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4248 root == root->fs_info->tree_root)
4249 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4250 root->sectorsize), (u64)-1, 0);
4253 * This function is also used to drop the items in the log tree before
4254 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4255 * it is used to drop the loged items. So we shouldn't kill the delayed
4258 if (min_type == 0 && root == BTRFS_I(inode)->root)
4259 btrfs_kill_delayed_inode_items(inode);
4262 key.offset = (u64)-1;
4267 * with a 16K leaf size and 128MB extents, you can actually queue
4268 * up a huge file in a single leaf. Most of the time that
4269 * bytes_deleted is > 0, it will be huge by the time we get here
4271 if (be_nice && bytes_deleted > 32 * 1024 * 1024) {
4272 if (btrfs_should_end_transaction(trans, root)) {
4279 path->leave_spinning = 1;
4280 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4287 /* there are no items in the tree for us to truncate, we're
4290 if (path->slots[0] == 0)
4297 leaf = path->nodes[0];
4298 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4299 found_type = found_key.type;
4301 if (found_key.objectid != ino)
4304 if (found_type < min_type)
4307 item_end = found_key.offset;
4308 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4309 fi = btrfs_item_ptr(leaf, path->slots[0],
4310 struct btrfs_file_extent_item);
4311 extent_type = btrfs_file_extent_type(leaf, fi);
4312 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4314 btrfs_file_extent_num_bytes(leaf, fi);
4315 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4316 item_end += btrfs_file_extent_inline_len(leaf,
4317 path->slots[0], fi);
4321 if (found_type > min_type) {
4324 if (item_end < new_size)
4326 if (found_key.offset >= new_size)
4332 /* FIXME, shrink the extent if the ref count is only 1 */
4333 if (found_type != BTRFS_EXTENT_DATA_KEY)
4337 last_size = found_key.offset;
4339 last_size = new_size;
4341 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4343 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4345 u64 orig_num_bytes =
4346 btrfs_file_extent_num_bytes(leaf, fi);
4347 extent_num_bytes = ALIGN(new_size -
4350 btrfs_set_file_extent_num_bytes(leaf, fi,
4352 num_dec = (orig_num_bytes -
4354 if (test_bit(BTRFS_ROOT_REF_COWS,
4357 inode_sub_bytes(inode, num_dec);
4358 btrfs_mark_buffer_dirty(leaf);
4361 btrfs_file_extent_disk_num_bytes(leaf,
4363 extent_offset = found_key.offset -
4364 btrfs_file_extent_offset(leaf, fi);
4366 /* FIXME blocksize != 4096 */
4367 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4368 if (extent_start != 0) {
4370 if (test_bit(BTRFS_ROOT_REF_COWS,
4372 inode_sub_bytes(inode, num_dec);
4375 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4377 * we can't truncate inline items that have had
4381 btrfs_file_extent_compression(leaf, fi) == 0 &&
4382 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4383 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4384 u32 size = new_size - found_key.offset;
4386 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4387 inode_sub_bytes(inode, item_end + 1 -
4391 * update the ram bytes to properly reflect
4392 * the new size of our item
4394 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4396 btrfs_file_extent_calc_inline_size(size);
4397 btrfs_truncate_item(root, path, size, 1);
4398 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4400 inode_sub_bytes(inode, item_end + 1 -
4406 if (!pending_del_nr) {
4407 /* no pending yet, add ourselves */
4408 pending_del_slot = path->slots[0];
4410 } else if (pending_del_nr &&
4411 path->slots[0] + 1 == pending_del_slot) {
4412 /* hop on the pending chunk */
4414 pending_del_slot = path->slots[0];
4421 should_throttle = 0;
4424 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4425 root == root->fs_info->tree_root)) {
4426 btrfs_set_path_blocking(path);
4427 bytes_deleted += extent_num_bytes;
4428 ret = btrfs_free_extent(trans, root, extent_start,
4429 extent_num_bytes, 0,
4430 btrfs_header_owner(leaf),
4431 ino, extent_offset, 0);
4433 if (btrfs_should_throttle_delayed_refs(trans, root))
4434 btrfs_async_run_delayed_refs(root,
4435 trans->delayed_ref_updates * 2, 0);
4437 if (truncate_space_check(trans, root,
4438 extent_num_bytes)) {
4441 if (btrfs_should_throttle_delayed_refs(trans,
4443 should_throttle = 1;
4448 if (found_type == BTRFS_INODE_ITEM_KEY)
4451 if (path->slots[0] == 0 ||
4452 path->slots[0] != pending_del_slot ||
4453 should_throttle || should_end) {
4454 if (pending_del_nr) {
4455 ret = btrfs_del_items(trans, root, path,
4459 btrfs_abort_transaction(trans,
4465 btrfs_release_path(path);
4466 if (should_throttle) {
4467 unsigned long updates = trans->delayed_ref_updates;
4469 trans->delayed_ref_updates = 0;
4470 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4476 * if we failed to refill our space rsv, bail out
4477 * and let the transaction restart
4489 if (pending_del_nr) {
4490 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4493 btrfs_abort_transaction(trans, root, ret);
4496 if (last_size != (u64)-1 &&
4497 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4498 btrfs_ordered_update_i_size(inode, last_size, NULL);
4500 btrfs_free_path(path);
4502 if (be_nice && bytes_deleted > 32 * 1024 * 1024) {
4503 unsigned long updates = trans->delayed_ref_updates;
4505 trans->delayed_ref_updates = 0;
4506 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4515 * btrfs_truncate_page - read, zero a chunk and write a page
4516 * @inode - inode that we're zeroing
4517 * @from - the offset to start zeroing
4518 * @len - the length to zero, 0 to zero the entire range respective to the
4520 * @front - zero up to the offset instead of from the offset on
4522 * This will find the page for the "from" offset and cow the page and zero the
4523 * part we want to zero. This is used with truncate and hole punching.
4525 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4528 struct address_space *mapping = inode->i_mapping;
4529 struct btrfs_root *root = BTRFS_I(inode)->root;
4530 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4531 struct btrfs_ordered_extent *ordered;
4532 struct extent_state *cached_state = NULL;
4534 u32 blocksize = root->sectorsize;
4535 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4536 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4538 gfp_t mask = btrfs_alloc_write_mask(mapping);
4543 if ((offset & (blocksize - 1)) == 0 &&
4544 (!len || ((len & (blocksize - 1)) == 0)))
4546 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4551 page = find_or_create_page(mapping, index, mask);
4553 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4558 page_start = page_offset(page);
4559 page_end = page_start + PAGE_CACHE_SIZE - 1;
4561 if (!PageUptodate(page)) {
4562 ret = btrfs_readpage(NULL, page);
4564 if (page->mapping != mapping) {
4566 page_cache_release(page);
4569 if (!PageUptodate(page)) {
4574 wait_on_page_writeback(page);
4576 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4577 set_page_extent_mapped(page);
4579 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4581 unlock_extent_cached(io_tree, page_start, page_end,
4582 &cached_state, GFP_NOFS);
4584 page_cache_release(page);
4585 btrfs_start_ordered_extent(inode, ordered, 1);
4586 btrfs_put_ordered_extent(ordered);
4590 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4591 EXTENT_DIRTY | EXTENT_DELALLOC |
4592 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4593 0, 0, &cached_state, GFP_NOFS);
4595 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4598 unlock_extent_cached(io_tree, page_start, page_end,
4599 &cached_state, GFP_NOFS);
4603 if (offset != PAGE_CACHE_SIZE) {
4605 len = PAGE_CACHE_SIZE - offset;
4608 memset(kaddr, 0, offset);
4610 memset(kaddr + offset, 0, len);
4611 flush_dcache_page(page);
4614 ClearPageChecked(page);
4615 set_page_dirty(page);
4616 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4621 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4623 page_cache_release(page);
4628 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4629 u64 offset, u64 len)
4631 struct btrfs_trans_handle *trans;
4635 * Still need to make sure the inode looks like it's been updated so
4636 * that any holes get logged if we fsync.
4638 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4639 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4640 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4641 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4646 * 1 - for the one we're dropping
4647 * 1 - for the one we're adding
4648 * 1 - for updating the inode.
4650 trans = btrfs_start_transaction(root, 3);
4652 return PTR_ERR(trans);
4654 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4656 btrfs_abort_transaction(trans, root, ret);
4657 btrfs_end_transaction(trans, root);
4661 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4662 0, 0, len, 0, len, 0, 0, 0);
4664 btrfs_abort_transaction(trans, root, ret);
4666 btrfs_update_inode(trans, root, inode);
4667 btrfs_end_transaction(trans, root);
4672 * This function puts in dummy file extents for the area we're creating a hole
4673 * for. So if we are truncating this file to a larger size we need to insert
4674 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4675 * the range between oldsize and size
4677 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4679 struct btrfs_root *root = BTRFS_I(inode)->root;
4680 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4681 struct extent_map *em = NULL;
4682 struct extent_state *cached_state = NULL;
4683 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4684 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4685 u64 block_end = ALIGN(size, root->sectorsize);
4692 * If our size started in the middle of a page we need to zero out the
4693 * rest of the page before we expand the i_size, otherwise we could
4694 * expose stale data.
4696 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4700 if (size <= hole_start)
4704 struct btrfs_ordered_extent *ordered;
4706 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4708 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4709 block_end - hole_start);
4712 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4713 &cached_state, GFP_NOFS);
4714 btrfs_start_ordered_extent(inode, ordered, 1);
4715 btrfs_put_ordered_extent(ordered);
4718 cur_offset = hole_start;
4720 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4721 block_end - cur_offset, 0);
4727 last_byte = min(extent_map_end(em), block_end);
4728 last_byte = ALIGN(last_byte , root->sectorsize);
4729 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4730 struct extent_map *hole_em;
4731 hole_size = last_byte - cur_offset;
4733 err = maybe_insert_hole(root, inode, cur_offset,
4737 btrfs_drop_extent_cache(inode, cur_offset,
4738 cur_offset + hole_size - 1, 0);
4739 hole_em = alloc_extent_map();
4741 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4742 &BTRFS_I(inode)->runtime_flags);
4745 hole_em->start = cur_offset;
4746 hole_em->len = hole_size;
4747 hole_em->orig_start = cur_offset;
4749 hole_em->block_start = EXTENT_MAP_HOLE;
4750 hole_em->block_len = 0;
4751 hole_em->orig_block_len = 0;
4752 hole_em->ram_bytes = hole_size;
4753 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4754 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4755 hole_em->generation = root->fs_info->generation;
4758 write_lock(&em_tree->lock);
4759 err = add_extent_mapping(em_tree, hole_em, 1);
4760 write_unlock(&em_tree->lock);
4763 btrfs_drop_extent_cache(inode, cur_offset,
4767 free_extent_map(hole_em);
4770 free_extent_map(em);
4772 cur_offset = last_byte;
4773 if (cur_offset >= block_end)
4776 free_extent_map(em);
4777 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4782 static int wait_snapshoting_atomic_t(atomic_t *a)
4788 static void wait_for_snapshot_creation(struct btrfs_root *root)
4793 ret = btrfs_start_write_no_snapshoting(root);
4796 wait_on_atomic_t(&root->will_be_snapshoted,
4797 wait_snapshoting_atomic_t,
4798 TASK_UNINTERRUPTIBLE);
4802 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4804 struct btrfs_root *root = BTRFS_I(inode)->root;
4805 struct btrfs_trans_handle *trans;
4806 loff_t oldsize = i_size_read(inode);
4807 loff_t newsize = attr->ia_size;
4808 int mask = attr->ia_valid;
4812 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4813 * special case where we need to update the times despite not having
4814 * these flags set. For all other operations the VFS set these flags
4815 * explicitly if it wants a timestamp update.
4817 if (newsize != oldsize) {
4818 inode_inc_iversion(inode);
4819 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4820 inode->i_ctime = inode->i_mtime =
4821 current_fs_time(inode->i_sb);
4824 if (newsize > oldsize) {
4825 truncate_pagecache(inode, newsize);
4827 * Don't do an expanding truncate while snapshoting is ongoing.
4828 * This is to ensure the snapshot captures a fully consistent
4829 * state of this file - if the snapshot captures this expanding
4830 * truncation, it must capture all writes that happened before
4833 wait_for_snapshot_creation(root);
4834 ret = btrfs_cont_expand(inode, oldsize, newsize);
4836 btrfs_end_write_no_snapshoting(root);
4840 trans = btrfs_start_transaction(root, 1);
4841 if (IS_ERR(trans)) {
4842 btrfs_end_write_no_snapshoting(root);
4843 return PTR_ERR(trans);
4846 i_size_write(inode, newsize);
4847 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4848 ret = btrfs_update_inode(trans, root, inode);
4849 btrfs_end_write_no_snapshoting(root);
4850 btrfs_end_transaction(trans, root);
4854 * We're truncating a file that used to have good data down to
4855 * zero. Make sure it gets into the ordered flush list so that
4856 * any new writes get down to disk quickly.
4859 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4860 &BTRFS_I(inode)->runtime_flags);
4863 * 1 for the orphan item we're going to add
4864 * 1 for the orphan item deletion.
4866 trans = btrfs_start_transaction(root, 2);
4868 return PTR_ERR(trans);
4871 * We need to do this in case we fail at _any_ point during the
4872 * actual truncate. Once we do the truncate_setsize we could
4873 * invalidate pages which forces any outstanding ordered io to
4874 * be instantly completed which will give us extents that need
4875 * to be truncated. If we fail to get an orphan inode down we
4876 * could have left over extents that were never meant to live,
4877 * so we need to garuntee from this point on that everything
4878 * will be consistent.
4880 ret = btrfs_orphan_add(trans, inode);
4881 btrfs_end_transaction(trans, root);
4885 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4886 truncate_setsize(inode, newsize);
4888 /* Disable nonlocked read DIO to avoid the end less truncate */
4889 btrfs_inode_block_unlocked_dio(inode);
4890 inode_dio_wait(inode);
4891 btrfs_inode_resume_unlocked_dio(inode);
4893 ret = btrfs_truncate(inode);
4894 if (ret && inode->i_nlink) {
4898 * failed to truncate, disk_i_size is only adjusted down
4899 * as we remove extents, so it should represent the true
4900 * size of the inode, so reset the in memory size and
4901 * delete our orphan entry.
4903 trans = btrfs_join_transaction(root);
4904 if (IS_ERR(trans)) {
4905 btrfs_orphan_del(NULL, inode);
4908 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4909 err = btrfs_orphan_del(trans, inode);
4911 btrfs_abort_transaction(trans, root, err);
4912 btrfs_end_transaction(trans, root);
4919 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4921 struct inode *inode = d_inode(dentry);
4922 struct btrfs_root *root = BTRFS_I(inode)->root;
4925 if (btrfs_root_readonly(root))
4928 err = inode_change_ok(inode, attr);
4932 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4933 err = btrfs_setsize(inode, attr);
4938 if (attr->ia_valid) {
4939 setattr_copy(inode, attr);
4940 inode_inc_iversion(inode);
4941 err = btrfs_dirty_inode(inode);
4943 if (!err && attr->ia_valid & ATTR_MODE)
4944 err = posix_acl_chmod(inode, inode->i_mode);
4951 * While truncating the inode pages during eviction, we get the VFS calling
4952 * btrfs_invalidatepage() against each page of the inode. This is slow because
4953 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4954 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4955 * extent_state structures over and over, wasting lots of time.
4957 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4958 * those expensive operations on a per page basis and do only the ordered io
4959 * finishing, while we release here the extent_map and extent_state structures,
4960 * without the excessive merging and splitting.
4962 static void evict_inode_truncate_pages(struct inode *inode)
4964 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4965 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4966 struct rb_node *node;
4968 ASSERT(inode->i_state & I_FREEING);
4969 truncate_inode_pages_final(&inode->i_data);
4971 write_lock(&map_tree->lock);
4972 while (!RB_EMPTY_ROOT(&map_tree->map)) {
4973 struct extent_map *em;
4975 node = rb_first(&map_tree->map);
4976 em = rb_entry(node, struct extent_map, rb_node);
4977 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4978 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4979 remove_extent_mapping(map_tree, em);
4980 free_extent_map(em);
4981 if (need_resched()) {
4982 write_unlock(&map_tree->lock);
4984 write_lock(&map_tree->lock);
4987 write_unlock(&map_tree->lock);
4990 * Keep looping until we have no more ranges in the io tree.
4991 * We can have ongoing bios started by readpages (called from readahead)
4992 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
4993 * still in progress (unlocked the pages in the bio but did not yet
4994 * unlocked the ranges in the io tree). Therefore this means some
4995 * ranges can still be locked and eviction started because before
4996 * submitting those bios, which are executed by a separate task (work
4997 * queue kthread), inode references (inode->i_count) were not taken
4998 * (which would be dropped in the end io callback of each bio).
4999 * Therefore here we effectively end up waiting for those bios and
5000 * anyone else holding locked ranges without having bumped the inode's
5001 * reference count - if we don't do it, when they access the inode's
5002 * io_tree to unlock a range it may be too late, leading to an
5003 * use-after-free issue.
5005 spin_lock(&io_tree->lock);
5006 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5007 struct extent_state *state;
5008 struct extent_state *cached_state = NULL;
5012 node = rb_first(&io_tree->state);
5013 state = rb_entry(node, struct extent_state, rb_node);
5014 start = state->start;
5016 spin_unlock(&io_tree->lock);
5018 lock_extent_bits(io_tree, start, end, 0, &cached_state);
5019 clear_extent_bit(io_tree, start, end,
5020 EXTENT_LOCKED | EXTENT_DIRTY |
5021 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5022 EXTENT_DEFRAG, 1, 1,
5023 &cached_state, GFP_NOFS);
5026 spin_lock(&io_tree->lock);
5028 spin_unlock(&io_tree->lock);
5031 void btrfs_evict_inode(struct inode *inode)
5033 struct btrfs_trans_handle *trans;
5034 struct btrfs_root *root = BTRFS_I(inode)->root;
5035 struct btrfs_block_rsv *rsv, *global_rsv;
5036 int steal_from_global = 0;
5037 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
5040 trace_btrfs_inode_evict(inode);
5042 evict_inode_truncate_pages(inode);
5044 if (inode->i_nlink &&
5045 ((btrfs_root_refs(&root->root_item) != 0 &&
5046 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5047 btrfs_is_free_space_inode(inode)))
5050 if (is_bad_inode(inode)) {
5051 btrfs_orphan_del(NULL, inode);
5054 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5055 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5057 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5059 if (root->fs_info->log_root_recovering) {
5060 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5061 &BTRFS_I(inode)->runtime_flags));
5065 if (inode->i_nlink > 0) {
5066 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5067 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5071 ret = btrfs_commit_inode_delayed_inode(inode);
5073 btrfs_orphan_del(NULL, inode);
5077 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
5079 btrfs_orphan_del(NULL, inode);
5082 rsv->size = min_size;
5084 global_rsv = &root->fs_info->global_block_rsv;
5086 btrfs_i_size_write(inode, 0);
5089 * This is a bit simpler than btrfs_truncate since we've already
5090 * reserved our space for our orphan item in the unlink, so we just
5091 * need to reserve some slack space in case we add bytes and update
5092 * inode item when doing the truncate.
5095 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5096 BTRFS_RESERVE_FLUSH_LIMIT);
5099 * Try and steal from the global reserve since we will
5100 * likely not use this space anyway, we want to try as
5101 * hard as possible to get this to work.
5104 steal_from_global++;
5106 steal_from_global = 0;
5110 * steal_from_global == 0: we reserved stuff, hooray!
5111 * steal_from_global == 1: we didn't reserve stuff, boo!
5112 * steal_from_global == 2: we've committed, still not a lot of
5113 * room but maybe we'll have room in the global reserve this
5115 * steal_from_global == 3: abandon all hope!
5117 if (steal_from_global > 2) {
5118 btrfs_warn(root->fs_info,
5119 "Could not get space for a delete, will truncate on mount %d",
5121 btrfs_orphan_del(NULL, inode);
5122 btrfs_free_block_rsv(root, rsv);
5126 trans = btrfs_join_transaction(root);
5127 if (IS_ERR(trans)) {
5128 btrfs_orphan_del(NULL, inode);
5129 btrfs_free_block_rsv(root, rsv);
5134 * We can't just steal from the global reserve, we need tomake
5135 * sure there is room to do it, if not we need to commit and try
5138 if (steal_from_global) {
5139 if (!btrfs_check_space_for_delayed_refs(trans, root))
5140 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5147 * Couldn't steal from the global reserve, we have too much
5148 * pending stuff built up, commit the transaction and try it
5152 ret = btrfs_commit_transaction(trans, root);
5154 btrfs_orphan_del(NULL, inode);
5155 btrfs_free_block_rsv(root, rsv);
5160 steal_from_global = 0;
5163 trans->block_rsv = rsv;
5165 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5166 if (ret != -ENOSPC && ret != -EAGAIN)
5169 trans->block_rsv = &root->fs_info->trans_block_rsv;
5170 btrfs_end_transaction(trans, root);
5172 btrfs_btree_balance_dirty(root);
5175 btrfs_free_block_rsv(root, rsv);
5178 * Errors here aren't a big deal, it just means we leave orphan items
5179 * in the tree. They will be cleaned up on the next mount.
5182 trans->block_rsv = root->orphan_block_rsv;
5183 btrfs_orphan_del(trans, inode);
5185 btrfs_orphan_del(NULL, inode);
5188 trans->block_rsv = &root->fs_info->trans_block_rsv;
5189 if (!(root == root->fs_info->tree_root ||
5190 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5191 btrfs_return_ino(root, btrfs_ino(inode));
5193 btrfs_end_transaction(trans, root);
5194 btrfs_btree_balance_dirty(root);
5196 btrfs_remove_delayed_node(inode);
5202 * this returns the key found in the dir entry in the location pointer.
5203 * If no dir entries were found, location->objectid is 0.
5205 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5206 struct btrfs_key *location)
5208 const char *name = dentry->d_name.name;
5209 int namelen = dentry->d_name.len;
5210 struct btrfs_dir_item *di;
5211 struct btrfs_path *path;
5212 struct btrfs_root *root = BTRFS_I(dir)->root;
5215 path = btrfs_alloc_path();
5219 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
5224 if (IS_ERR_OR_NULL(di))
5227 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5229 btrfs_free_path(path);
5232 location->objectid = 0;
5237 * when we hit a tree root in a directory, the btrfs part of the inode
5238 * needs to be changed to reflect the root directory of the tree root. This
5239 * is kind of like crossing a mount point.
5241 static int fixup_tree_root_location(struct btrfs_root *root,
5243 struct dentry *dentry,
5244 struct btrfs_key *location,
5245 struct btrfs_root **sub_root)
5247 struct btrfs_path *path;
5248 struct btrfs_root *new_root;
5249 struct btrfs_root_ref *ref;
5250 struct extent_buffer *leaf;
5251 struct btrfs_key key;
5255 path = btrfs_alloc_path();
5262 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5263 key.type = BTRFS_ROOT_REF_KEY;
5264 key.offset = location->objectid;
5266 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5274 leaf = path->nodes[0];
5275 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5276 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5277 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5280 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5281 (unsigned long)(ref + 1),
5282 dentry->d_name.len);
5286 btrfs_release_path(path);
5288 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5289 if (IS_ERR(new_root)) {
5290 err = PTR_ERR(new_root);
5294 *sub_root = new_root;
5295 location->objectid = btrfs_root_dirid(&new_root->root_item);
5296 location->type = BTRFS_INODE_ITEM_KEY;
5297 location->offset = 0;
5300 btrfs_free_path(path);
5304 static void inode_tree_add(struct inode *inode)
5306 struct btrfs_root *root = BTRFS_I(inode)->root;
5307 struct btrfs_inode *entry;
5309 struct rb_node *parent;
5310 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5311 u64 ino = btrfs_ino(inode);
5313 if (inode_unhashed(inode))
5316 spin_lock(&root->inode_lock);
5317 p = &root->inode_tree.rb_node;
5320 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5322 if (ino < btrfs_ino(&entry->vfs_inode))
5323 p = &parent->rb_left;
5324 else if (ino > btrfs_ino(&entry->vfs_inode))
5325 p = &parent->rb_right;
5327 WARN_ON(!(entry->vfs_inode.i_state &
5328 (I_WILL_FREE | I_FREEING)));
5329 rb_replace_node(parent, new, &root->inode_tree);
5330 RB_CLEAR_NODE(parent);
5331 spin_unlock(&root->inode_lock);
5335 rb_link_node(new, parent, p);
5336 rb_insert_color(new, &root->inode_tree);
5337 spin_unlock(&root->inode_lock);
5340 static void inode_tree_del(struct inode *inode)
5342 struct btrfs_root *root = BTRFS_I(inode)->root;
5345 spin_lock(&root->inode_lock);
5346 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5347 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5348 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5349 empty = RB_EMPTY_ROOT(&root->inode_tree);
5351 spin_unlock(&root->inode_lock);
5353 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5354 synchronize_srcu(&root->fs_info->subvol_srcu);
5355 spin_lock(&root->inode_lock);
5356 empty = RB_EMPTY_ROOT(&root->inode_tree);
5357 spin_unlock(&root->inode_lock);
5359 btrfs_add_dead_root(root);
5363 void btrfs_invalidate_inodes(struct btrfs_root *root)
5365 struct rb_node *node;
5366 struct rb_node *prev;
5367 struct btrfs_inode *entry;
5368 struct inode *inode;
5371 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5372 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5374 spin_lock(&root->inode_lock);
5376 node = root->inode_tree.rb_node;
5380 entry = rb_entry(node, struct btrfs_inode, rb_node);
5382 if (objectid < btrfs_ino(&entry->vfs_inode))
5383 node = node->rb_left;
5384 else if (objectid > btrfs_ino(&entry->vfs_inode))
5385 node = node->rb_right;
5391 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5392 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5396 prev = rb_next(prev);
5400 entry = rb_entry(node, struct btrfs_inode, rb_node);
5401 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5402 inode = igrab(&entry->vfs_inode);
5404 spin_unlock(&root->inode_lock);
5405 if (atomic_read(&inode->i_count) > 1)
5406 d_prune_aliases(inode);
5408 * btrfs_drop_inode will have it removed from
5409 * the inode cache when its usage count
5414 spin_lock(&root->inode_lock);
5418 if (cond_resched_lock(&root->inode_lock))
5421 node = rb_next(node);
5423 spin_unlock(&root->inode_lock);
5426 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5428 struct btrfs_iget_args *args = p;
5429 inode->i_ino = args->location->objectid;
5430 memcpy(&BTRFS_I(inode)->location, args->location,
5431 sizeof(*args->location));
5432 BTRFS_I(inode)->root = args->root;
5436 static int btrfs_find_actor(struct inode *inode, void *opaque)
5438 struct btrfs_iget_args *args = opaque;
5439 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5440 args->root == BTRFS_I(inode)->root;
5443 static struct inode *btrfs_iget_locked(struct super_block *s,
5444 struct btrfs_key *location,
5445 struct btrfs_root *root)
5447 struct inode *inode;
5448 struct btrfs_iget_args args;
5449 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5451 args.location = location;
5454 inode = iget5_locked(s, hashval, btrfs_find_actor,
5455 btrfs_init_locked_inode,
5460 /* Get an inode object given its location and corresponding root.
5461 * Returns in *is_new if the inode was read from disk
5463 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5464 struct btrfs_root *root, int *new)
5466 struct inode *inode;
5468 inode = btrfs_iget_locked(s, location, root);
5470 return ERR_PTR(-ENOMEM);
5472 if (inode->i_state & I_NEW) {
5473 btrfs_read_locked_inode(inode);
5474 if (!is_bad_inode(inode)) {
5475 inode_tree_add(inode);
5476 unlock_new_inode(inode);
5480 unlock_new_inode(inode);
5482 inode = ERR_PTR(-ESTALE);
5489 static struct inode *new_simple_dir(struct super_block *s,
5490 struct btrfs_key *key,
5491 struct btrfs_root *root)
5493 struct inode *inode = new_inode(s);
5496 return ERR_PTR(-ENOMEM);
5498 BTRFS_I(inode)->root = root;
5499 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5500 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5502 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5503 inode->i_op = &btrfs_dir_ro_inode_operations;
5504 inode->i_fop = &simple_dir_operations;
5505 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5506 inode->i_mtime = CURRENT_TIME;
5507 inode->i_atime = inode->i_mtime;
5508 inode->i_ctime = inode->i_mtime;
5509 BTRFS_I(inode)->i_otime = inode->i_mtime;
5514 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5516 struct inode *inode;
5517 struct btrfs_root *root = BTRFS_I(dir)->root;
5518 struct btrfs_root *sub_root = root;
5519 struct btrfs_key location;
5523 if (dentry->d_name.len > BTRFS_NAME_LEN)
5524 return ERR_PTR(-ENAMETOOLONG);
5526 ret = btrfs_inode_by_name(dir, dentry, &location);
5528 return ERR_PTR(ret);
5530 if (location.objectid == 0)
5531 return ERR_PTR(-ENOENT);
5533 if (location.type == BTRFS_INODE_ITEM_KEY) {
5534 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5538 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5540 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5541 ret = fixup_tree_root_location(root, dir, dentry,
5542 &location, &sub_root);
5545 inode = ERR_PTR(ret);
5547 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5549 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5551 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5553 if (!IS_ERR(inode) && root != sub_root) {
5554 down_read(&root->fs_info->cleanup_work_sem);
5555 if (!(inode->i_sb->s_flags & MS_RDONLY))
5556 ret = btrfs_orphan_cleanup(sub_root);
5557 up_read(&root->fs_info->cleanup_work_sem);
5560 inode = ERR_PTR(ret);
5567 static int btrfs_dentry_delete(const struct dentry *dentry)
5569 struct btrfs_root *root;
5570 struct inode *inode = d_inode(dentry);
5572 if (!inode && !IS_ROOT(dentry))
5573 inode = d_inode(dentry->d_parent);
5576 root = BTRFS_I(inode)->root;
5577 if (btrfs_root_refs(&root->root_item) == 0)
5580 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5586 static void btrfs_dentry_release(struct dentry *dentry)
5588 kfree(dentry->d_fsdata);
5591 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5594 struct inode *inode;
5596 inode = btrfs_lookup_dentry(dir, dentry);
5597 if (IS_ERR(inode)) {
5598 if (PTR_ERR(inode) == -ENOENT)
5601 return ERR_CAST(inode);
5604 return d_splice_alias(inode, dentry);
5607 unsigned char btrfs_filetype_table[] = {
5608 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5611 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5613 struct inode *inode = file_inode(file);
5614 struct btrfs_root *root = BTRFS_I(inode)->root;
5615 struct btrfs_item *item;
5616 struct btrfs_dir_item *di;
5617 struct btrfs_key key;
5618 struct btrfs_key found_key;
5619 struct btrfs_path *path;
5620 struct list_head ins_list;
5621 struct list_head del_list;
5623 struct extent_buffer *leaf;
5625 unsigned char d_type;
5630 int key_type = BTRFS_DIR_INDEX_KEY;
5634 int is_curr = 0; /* ctx->pos points to the current index? */
5636 /* FIXME, use a real flag for deciding about the key type */
5637 if (root->fs_info->tree_root == root)
5638 key_type = BTRFS_DIR_ITEM_KEY;
5640 if (!dir_emit_dots(file, ctx))
5643 path = btrfs_alloc_path();
5649 if (key_type == BTRFS_DIR_INDEX_KEY) {
5650 INIT_LIST_HEAD(&ins_list);
5651 INIT_LIST_HEAD(&del_list);
5652 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5655 key.type = key_type;
5656 key.offset = ctx->pos;
5657 key.objectid = btrfs_ino(inode);
5659 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5664 leaf = path->nodes[0];
5665 slot = path->slots[0];
5666 if (slot >= btrfs_header_nritems(leaf)) {
5667 ret = btrfs_next_leaf(root, path);
5675 item = btrfs_item_nr(slot);
5676 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5678 if (found_key.objectid != key.objectid)
5680 if (found_key.type != key_type)
5682 if (found_key.offset < ctx->pos)
5684 if (key_type == BTRFS_DIR_INDEX_KEY &&
5685 btrfs_should_delete_dir_index(&del_list,
5689 ctx->pos = found_key.offset;
5692 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5694 di_total = btrfs_item_size(leaf, item);
5696 while (di_cur < di_total) {
5697 struct btrfs_key location;
5699 if (verify_dir_item(root, leaf, di))
5702 name_len = btrfs_dir_name_len(leaf, di);
5703 if (name_len <= sizeof(tmp_name)) {
5704 name_ptr = tmp_name;
5706 name_ptr = kmalloc(name_len, GFP_NOFS);
5712 read_extent_buffer(leaf, name_ptr,
5713 (unsigned long)(di + 1), name_len);
5715 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5716 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5719 /* is this a reference to our own snapshot? If so
5722 * In contrast to old kernels, we insert the snapshot's
5723 * dir item and dir index after it has been created, so
5724 * we won't find a reference to our own snapshot. We
5725 * still keep the following code for backward
5728 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5729 location.objectid == root->root_key.objectid) {
5733 over = !dir_emit(ctx, name_ptr, name_len,
5734 location.objectid, d_type);
5737 if (name_ptr != tmp_name)
5742 di_len = btrfs_dir_name_len(leaf, di) +
5743 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5745 di = (struct btrfs_dir_item *)((char *)di + di_len);
5751 if (key_type == BTRFS_DIR_INDEX_KEY) {
5754 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5759 /* Reached end of directory/root. Bump pos past the last item. */
5763 * Stop new entries from being returned after we return the last
5766 * New directory entries are assigned a strictly increasing
5767 * offset. This means that new entries created during readdir
5768 * are *guaranteed* to be seen in the future by that readdir.
5769 * This has broken buggy programs which operate on names as
5770 * they're returned by readdir. Until we re-use freed offsets
5771 * we have this hack to stop new entries from being returned
5772 * under the assumption that they'll never reach this huge
5775 * This is being careful not to overflow 32bit loff_t unless the
5776 * last entry requires it because doing so has broken 32bit apps
5779 if (key_type == BTRFS_DIR_INDEX_KEY) {
5780 if (ctx->pos >= INT_MAX)
5781 ctx->pos = LLONG_MAX;
5788 if (key_type == BTRFS_DIR_INDEX_KEY)
5789 btrfs_put_delayed_items(&ins_list, &del_list);
5790 btrfs_free_path(path);
5794 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5796 struct btrfs_root *root = BTRFS_I(inode)->root;
5797 struct btrfs_trans_handle *trans;
5799 bool nolock = false;
5801 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5804 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5807 if (wbc->sync_mode == WB_SYNC_ALL) {
5809 trans = btrfs_join_transaction_nolock(root);
5811 trans = btrfs_join_transaction(root);
5813 return PTR_ERR(trans);
5814 ret = btrfs_commit_transaction(trans, root);
5820 * This is somewhat expensive, updating the tree every time the
5821 * inode changes. But, it is most likely to find the inode in cache.
5822 * FIXME, needs more benchmarking...there are no reasons other than performance
5823 * to keep or drop this code.
5825 static int btrfs_dirty_inode(struct inode *inode)
5827 struct btrfs_root *root = BTRFS_I(inode)->root;
5828 struct btrfs_trans_handle *trans;
5831 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5834 trans = btrfs_join_transaction(root);
5836 return PTR_ERR(trans);
5838 ret = btrfs_update_inode(trans, root, inode);
5839 if (ret && ret == -ENOSPC) {
5840 /* whoops, lets try again with the full transaction */
5841 btrfs_end_transaction(trans, root);
5842 trans = btrfs_start_transaction(root, 1);
5844 return PTR_ERR(trans);
5846 ret = btrfs_update_inode(trans, root, inode);
5848 btrfs_end_transaction(trans, root);
5849 if (BTRFS_I(inode)->delayed_node)
5850 btrfs_balance_delayed_items(root);
5856 * This is a copy of file_update_time. We need this so we can return error on
5857 * ENOSPC for updating the inode in the case of file write and mmap writes.
5859 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5862 struct btrfs_root *root = BTRFS_I(inode)->root;
5864 if (btrfs_root_readonly(root))
5867 if (flags & S_VERSION)
5868 inode_inc_iversion(inode);
5869 if (flags & S_CTIME)
5870 inode->i_ctime = *now;
5871 if (flags & S_MTIME)
5872 inode->i_mtime = *now;
5873 if (flags & S_ATIME)
5874 inode->i_atime = *now;
5875 return btrfs_dirty_inode(inode);
5879 * find the highest existing sequence number in a directory
5880 * and then set the in-memory index_cnt variable to reflect
5881 * free sequence numbers
5883 static int btrfs_set_inode_index_count(struct inode *inode)
5885 struct btrfs_root *root = BTRFS_I(inode)->root;
5886 struct btrfs_key key, found_key;
5887 struct btrfs_path *path;
5888 struct extent_buffer *leaf;
5891 key.objectid = btrfs_ino(inode);
5892 key.type = BTRFS_DIR_INDEX_KEY;
5893 key.offset = (u64)-1;
5895 path = btrfs_alloc_path();
5899 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5902 /* FIXME: we should be able to handle this */
5908 * MAGIC NUMBER EXPLANATION:
5909 * since we search a directory based on f_pos we have to start at 2
5910 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5911 * else has to start at 2
5913 if (path->slots[0] == 0) {
5914 BTRFS_I(inode)->index_cnt = 2;
5920 leaf = path->nodes[0];
5921 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5923 if (found_key.objectid != btrfs_ino(inode) ||
5924 found_key.type != BTRFS_DIR_INDEX_KEY) {
5925 BTRFS_I(inode)->index_cnt = 2;
5929 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5931 btrfs_free_path(path);
5936 * helper to find a free sequence number in a given directory. This current
5937 * code is very simple, later versions will do smarter things in the btree
5939 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5943 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5944 ret = btrfs_inode_delayed_dir_index_count(dir);
5946 ret = btrfs_set_inode_index_count(dir);
5952 *index = BTRFS_I(dir)->index_cnt;
5953 BTRFS_I(dir)->index_cnt++;
5958 static int btrfs_insert_inode_locked(struct inode *inode)
5960 struct btrfs_iget_args args;
5961 args.location = &BTRFS_I(inode)->location;
5962 args.root = BTRFS_I(inode)->root;
5964 return insert_inode_locked4(inode,
5965 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
5966 btrfs_find_actor, &args);
5969 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5970 struct btrfs_root *root,
5972 const char *name, int name_len,
5973 u64 ref_objectid, u64 objectid,
5974 umode_t mode, u64 *index)
5976 struct inode *inode;
5977 struct btrfs_inode_item *inode_item;
5978 struct btrfs_key *location;
5979 struct btrfs_path *path;
5980 struct btrfs_inode_ref *ref;
5981 struct btrfs_key key[2];
5983 int nitems = name ? 2 : 1;
5987 path = btrfs_alloc_path();
5989 return ERR_PTR(-ENOMEM);
5991 inode = new_inode(root->fs_info->sb);
5993 btrfs_free_path(path);
5994 return ERR_PTR(-ENOMEM);
5998 * O_TMPFILE, set link count to 0, so that after this point,
5999 * we fill in an inode item with the correct link count.
6002 set_nlink(inode, 0);
6005 * we have to initialize this early, so we can reclaim the inode
6006 * number if we fail afterwards in this function.
6008 inode->i_ino = objectid;
6011 trace_btrfs_inode_request(dir);
6013 ret = btrfs_set_inode_index(dir, index);
6015 btrfs_free_path(path);
6017 return ERR_PTR(ret);
6023 * index_cnt is ignored for everything but a dir,
6024 * btrfs_get_inode_index_count has an explanation for the magic
6027 BTRFS_I(inode)->index_cnt = 2;
6028 BTRFS_I(inode)->dir_index = *index;
6029 BTRFS_I(inode)->root = root;
6030 BTRFS_I(inode)->generation = trans->transid;
6031 inode->i_generation = BTRFS_I(inode)->generation;
6034 * We could have gotten an inode number from somebody who was fsynced
6035 * and then removed in this same transaction, so let's just set full
6036 * sync since it will be a full sync anyway and this will blow away the
6037 * old info in the log.
6039 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6041 key[0].objectid = objectid;
6042 key[0].type = BTRFS_INODE_ITEM_KEY;
6045 sizes[0] = sizeof(struct btrfs_inode_item);
6049 * Start new inodes with an inode_ref. This is slightly more
6050 * efficient for small numbers of hard links since they will
6051 * be packed into one item. Extended refs will kick in if we
6052 * add more hard links than can fit in the ref item.
6054 key[1].objectid = objectid;
6055 key[1].type = BTRFS_INODE_REF_KEY;
6056 key[1].offset = ref_objectid;
6058 sizes[1] = name_len + sizeof(*ref);
6061 location = &BTRFS_I(inode)->location;
6062 location->objectid = objectid;
6063 location->offset = 0;
6064 location->type = BTRFS_INODE_ITEM_KEY;
6066 ret = btrfs_insert_inode_locked(inode);
6070 path->leave_spinning = 1;
6071 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6075 inode_init_owner(inode, dir, mode);
6076 inode_set_bytes(inode, 0);
6078 inode->i_mtime = CURRENT_TIME;
6079 inode->i_atime = inode->i_mtime;
6080 inode->i_ctime = inode->i_mtime;
6081 BTRFS_I(inode)->i_otime = inode->i_mtime;
6083 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6084 struct btrfs_inode_item);
6085 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6086 sizeof(*inode_item));
6087 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6090 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6091 struct btrfs_inode_ref);
6092 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6093 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6094 ptr = (unsigned long)(ref + 1);
6095 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6098 btrfs_mark_buffer_dirty(path->nodes[0]);
6099 btrfs_free_path(path);
6101 btrfs_inherit_iflags(inode, dir);
6103 if (S_ISREG(mode)) {
6104 if (btrfs_test_opt(root, NODATASUM))
6105 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6106 if (btrfs_test_opt(root, NODATACOW))
6107 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6108 BTRFS_INODE_NODATASUM;
6111 inode_tree_add(inode);
6113 trace_btrfs_inode_new(inode);
6114 btrfs_set_inode_last_trans(trans, inode);
6116 btrfs_update_root_times(trans, root);
6118 ret = btrfs_inode_inherit_props(trans, inode, dir);
6120 btrfs_err(root->fs_info,
6121 "error inheriting props for ino %llu (root %llu): %d",
6122 btrfs_ino(inode), root->root_key.objectid, ret);
6127 unlock_new_inode(inode);
6130 BTRFS_I(dir)->index_cnt--;
6131 btrfs_free_path(path);
6133 return ERR_PTR(ret);
6136 static inline u8 btrfs_inode_type(struct inode *inode)
6138 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6142 * utility function to add 'inode' into 'parent_inode' with
6143 * a give name and a given sequence number.
6144 * if 'add_backref' is true, also insert a backref from the
6145 * inode to the parent directory.
6147 int btrfs_add_link(struct btrfs_trans_handle *trans,
6148 struct inode *parent_inode, struct inode *inode,
6149 const char *name, int name_len, int add_backref, u64 index)
6152 struct btrfs_key key;
6153 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
6154 u64 ino = btrfs_ino(inode);
6155 u64 parent_ino = btrfs_ino(parent_inode);
6157 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6158 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6161 key.type = BTRFS_INODE_ITEM_KEY;
6165 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6166 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6167 key.objectid, root->root_key.objectid,
6168 parent_ino, index, name, name_len);
6169 } else if (add_backref) {
6170 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6174 /* Nothing to clean up yet */
6178 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6180 btrfs_inode_type(inode), index);
6181 if (ret == -EEXIST || ret == -EOVERFLOW)
6184 btrfs_abort_transaction(trans, root, ret);
6188 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6190 inode_inc_iversion(parent_inode);
6191 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
6192 ret = btrfs_update_inode(trans, root, parent_inode);
6194 btrfs_abort_transaction(trans, root, ret);
6198 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6201 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6202 key.objectid, root->root_key.objectid,
6203 parent_ino, &local_index, name, name_len);
6205 } else if (add_backref) {
6209 err = btrfs_del_inode_ref(trans, root, name, name_len,
6210 ino, parent_ino, &local_index);
6215 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6216 struct inode *dir, struct dentry *dentry,
6217 struct inode *inode, int backref, u64 index)
6219 int err = btrfs_add_link(trans, dir, inode,
6220 dentry->d_name.name, dentry->d_name.len,
6227 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6228 umode_t mode, dev_t rdev)
6230 struct btrfs_trans_handle *trans;
6231 struct btrfs_root *root = BTRFS_I(dir)->root;
6232 struct inode *inode = NULL;
6238 if (!new_valid_dev(rdev))
6242 * 2 for inode item and ref
6244 * 1 for xattr if selinux is on
6246 trans = btrfs_start_transaction(root, 5);
6248 return PTR_ERR(trans);
6250 err = btrfs_find_free_ino(root, &objectid);
6254 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6255 dentry->d_name.len, btrfs_ino(dir), objectid,
6257 if (IS_ERR(inode)) {
6258 err = PTR_ERR(inode);
6263 * If the active LSM wants to access the inode during
6264 * d_instantiate it needs these. Smack checks to see
6265 * if the filesystem supports xattrs by looking at the
6268 inode->i_op = &btrfs_special_inode_operations;
6269 init_special_inode(inode, inode->i_mode, rdev);
6271 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6273 goto out_unlock_inode;
6275 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6277 goto out_unlock_inode;
6279 btrfs_update_inode(trans, root, inode);
6280 unlock_new_inode(inode);
6281 d_instantiate(dentry, inode);
6285 btrfs_end_transaction(trans, root);
6286 btrfs_balance_delayed_items(root);
6287 btrfs_btree_balance_dirty(root);
6289 inode_dec_link_count(inode);
6296 unlock_new_inode(inode);
6301 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6302 umode_t mode, bool excl)
6304 struct btrfs_trans_handle *trans;
6305 struct btrfs_root *root = BTRFS_I(dir)->root;
6306 struct inode *inode = NULL;
6307 int drop_inode_on_err = 0;
6313 * 2 for inode item and ref
6315 * 1 for xattr if selinux is on
6317 trans = btrfs_start_transaction(root, 5);
6319 return PTR_ERR(trans);
6321 err = btrfs_find_free_ino(root, &objectid);
6325 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6326 dentry->d_name.len, btrfs_ino(dir), objectid,
6328 if (IS_ERR(inode)) {
6329 err = PTR_ERR(inode);
6332 drop_inode_on_err = 1;
6334 * If the active LSM wants to access the inode during
6335 * d_instantiate it needs these. Smack checks to see
6336 * if the filesystem supports xattrs by looking at the
6339 inode->i_fop = &btrfs_file_operations;
6340 inode->i_op = &btrfs_file_inode_operations;
6341 inode->i_mapping->a_ops = &btrfs_aops;
6343 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6345 goto out_unlock_inode;
6347 err = btrfs_update_inode(trans, root, inode);
6349 goto out_unlock_inode;
6351 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6353 goto out_unlock_inode;
6355 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6356 unlock_new_inode(inode);
6357 d_instantiate(dentry, inode);
6360 btrfs_end_transaction(trans, root);
6361 if (err && drop_inode_on_err) {
6362 inode_dec_link_count(inode);
6365 btrfs_balance_delayed_items(root);
6366 btrfs_btree_balance_dirty(root);
6370 unlock_new_inode(inode);
6375 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6376 struct dentry *dentry)
6378 struct btrfs_trans_handle *trans;
6379 struct btrfs_root *root = BTRFS_I(dir)->root;
6380 struct inode *inode = d_inode(old_dentry);
6385 /* do not allow sys_link's with other subvols of the same device */
6386 if (root->objectid != BTRFS_I(inode)->root->objectid)
6389 if (inode->i_nlink >= BTRFS_LINK_MAX)
6392 err = btrfs_set_inode_index(dir, &index);
6397 * 2 items for inode and inode ref
6398 * 2 items for dir items
6399 * 1 item for parent inode
6401 trans = btrfs_start_transaction(root, 5);
6402 if (IS_ERR(trans)) {
6403 err = PTR_ERR(trans);
6407 /* There are several dir indexes for this inode, clear the cache. */
6408 BTRFS_I(inode)->dir_index = 0ULL;
6410 inode_inc_iversion(inode);
6411 inode->i_ctime = CURRENT_TIME;
6413 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6415 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6420 struct dentry *parent = dentry->d_parent;
6421 err = btrfs_update_inode(trans, root, inode);
6424 if (inode->i_nlink == 1) {
6426 * If new hard link count is 1, it's a file created
6427 * with open(2) O_TMPFILE flag.
6429 err = btrfs_orphan_del(trans, inode);
6433 d_instantiate(dentry, inode);
6434 btrfs_log_new_name(trans, inode, NULL, parent);
6437 btrfs_end_transaction(trans, root);
6438 btrfs_balance_delayed_items(root);
6441 inode_dec_link_count(inode);
6444 btrfs_btree_balance_dirty(root);
6448 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6450 struct inode *inode = NULL;
6451 struct btrfs_trans_handle *trans;
6452 struct btrfs_root *root = BTRFS_I(dir)->root;
6454 int drop_on_err = 0;
6459 * 2 items for inode and ref
6460 * 2 items for dir items
6461 * 1 for xattr if selinux is on
6463 trans = btrfs_start_transaction(root, 5);
6465 return PTR_ERR(trans);
6467 err = btrfs_find_free_ino(root, &objectid);
6471 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6472 dentry->d_name.len, btrfs_ino(dir), objectid,
6473 S_IFDIR | mode, &index);
6474 if (IS_ERR(inode)) {
6475 err = PTR_ERR(inode);
6480 /* these must be set before we unlock the inode */
6481 inode->i_op = &btrfs_dir_inode_operations;
6482 inode->i_fop = &btrfs_dir_file_operations;
6484 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6486 goto out_fail_inode;
6488 btrfs_i_size_write(inode, 0);
6489 err = btrfs_update_inode(trans, root, inode);
6491 goto out_fail_inode;
6493 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6494 dentry->d_name.len, 0, index);
6496 goto out_fail_inode;
6498 d_instantiate(dentry, inode);
6500 * mkdir is special. We're unlocking after we call d_instantiate
6501 * to avoid a race with nfsd calling d_instantiate.
6503 unlock_new_inode(inode);
6507 btrfs_end_transaction(trans, root);
6509 inode_dec_link_count(inode);
6512 btrfs_balance_delayed_items(root);
6513 btrfs_btree_balance_dirty(root);
6517 unlock_new_inode(inode);
6521 /* Find next extent map of a given extent map, caller needs to ensure locks */
6522 static struct extent_map *next_extent_map(struct extent_map *em)
6524 struct rb_node *next;
6526 next = rb_next(&em->rb_node);
6529 return container_of(next, struct extent_map, rb_node);
6532 static struct extent_map *prev_extent_map(struct extent_map *em)
6534 struct rb_node *prev;
6536 prev = rb_prev(&em->rb_node);
6539 return container_of(prev, struct extent_map, rb_node);
6542 /* helper for btfs_get_extent. Given an existing extent in the tree,
6543 * the existing extent is the nearest extent to map_start,
6544 * and an extent that you want to insert, deal with overlap and insert
6545 * the best fitted new extent into the tree.
6547 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6548 struct extent_map *existing,
6549 struct extent_map *em,
6552 struct extent_map *prev;
6553 struct extent_map *next;
6558 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6560 if (existing->start > map_start) {
6562 prev = prev_extent_map(next);
6565 next = next_extent_map(prev);
6568 start = prev ? extent_map_end(prev) : em->start;
6569 start = max_t(u64, start, em->start);
6570 end = next ? next->start : extent_map_end(em);
6571 end = min_t(u64, end, extent_map_end(em));
6572 start_diff = start - em->start;
6574 em->len = end - start;
6575 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6576 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6577 em->block_start += start_diff;
6578 em->block_len -= start_diff;
6580 return add_extent_mapping(em_tree, em, 0);
6583 static noinline int uncompress_inline(struct btrfs_path *path,
6584 struct inode *inode, struct page *page,
6585 size_t pg_offset, u64 extent_offset,
6586 struct btrfs_file_extent_item *item)
6589 struct extent_buffer *leaf = path->nodes[0];
6592 unsigned long inline_size;
6596 WARN_ON(pg_offset != 0);
6597 compress_type = btrfs_file_extent_compression(leaf, item);
6598 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6599 inline_size = btrfs_file_extent_inline_item_len(leaf,
6600 btrfs_item_nr(path->slots[0]));
6601 tmp = kmalloc(inline_size, GFP_NOFS);
6604 ptr = btrfs_file_extent_inline_start(item);
6606 read_extent_buffer(leaf, tmp, ptr, inline_size);
6608 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6609 ret = btrfs_decompress(compress_type, tmp, page,
6610 extent_offset, inline_size, max_size);
6616 * a bit scary, this does extent mapping from logical file offset to the disk.
6617 * the ugly parts come from merging extents from the disk with the in-ram
6618 * representation. This gets more complex because of the data=ordered code,
6619 * where the in-ram extents might be locked pending data=ordered completion.
6621 * This also copies inline extents directly into the page.
6624 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6625 size_t pg_offset, u64 start, u64 len,
6630 u64 extent_start = 0;
6632 u64 objectid = btrfs_ino(inode);
6634 struct btrfs_path *path = NULL;
6635 struct btrfs_root *root = BTRFS_I(inode)->root;
6636 struct btrfs_file_extent_item *item;
6637 struct extent_buffer *leaf;
6638 struct btrfs_key found_key;
6639 struct extent_map *em = NULL;
6640 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6641 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6642 struct btrfs_trans_handle *trans = NULL;
6643 const bool new_inline = !page || create;
6646 read_lock(&em_tree->lock);
6647 em = lookup_extent_mapping(em_tree, start, len);
6649 em->bdev = root->fs_info->fs_devices->latest_bdev;
6650 read_unlock(&em_tree->lock);
6653 if (em->start > start || em->start + em->len <= start)
6654 free_extent_map(em);
6655 else if (em->block_start == EXTENT_MAP_INLINE && page)
6656 free_extent_map(em);
6660 em = alloc_extent_map();
6665 em->bdev = root->fs_info->fs_devices->latest_bdev;
6666 em->start = EXTENT_MAP_HOLE;
6667 em->orig_start = EXTENT_MAP_HOLE;
6669 em->block_len = (u64)-1;
6672 path = btrfs_alloc_path();
6678 * Chances are we'll be called again, so go ahead and do
6684 ret = btrfs_lookup_file_extent(trans, root, path,
6685 objectid, start, trans != NULL);
6692 if (path->slots[0] == 0)
6697 leaf = path->nodes[0];
6698 item = btrfs_item_ptr(leaf, path->slots[0],
6699 struct btrfs_file_extent_item);
6700 /* are we inside the extent that was found? */
6701 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6702 found_type = found_key.type;
6703 if (found_key.objectid != objectid ||
6704 found_type != BTRFS_EXTENT_DATA_KEY) {
6706 * If we backup past the first extent we want to move forward
6707 * and see if there is an extent in front of us, otherwise we'll
6708 * say there is a hole for our whole search range which can
6715 found_type = btrfs_file_extent_type(leaf, item);
6716 extent_start = found_key.offset;
6717 if (found_type == BTRFS_FILE_EXTENT_REG ||
6718 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6719 extent_end = extent_start +
6720 btrfs_file_extent_num_bytes(leaf, item);
6721 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6723 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6724 extent_end = ALIGN(extent_start + size, root->sectorsize);
6727 if (start >= extent_end) {
6729 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6730 ret = btrfs_next_leaf(root, path);
6737 leaf = path->nodes[0];
6739 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6740 if (found_key.objectid != objectid ||
6741 found_key.type != BTRFS_EXTENT_DATA_KEY)
6743 if (start + len <= found_key.offset)
6745 if (start > found_key.offset)
6748 em->orig_start = start;
6749 em->len = found_key.offset - start;
6753 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6755 if (found_type == BTRFS_FILE_EXTENT_REG ||
6756 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6758 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6762 size_t extent_offset;
6768 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6769 extent_offset = page_offset(page) + pg_offset - extent_start;
6770 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6771 size - extent_offset);
6772 em->start = extent_start + extent_offset;
6773 em->len = ALIGN(copy_size, root->sectorsize);
6774 em->orig_block_len = em->len;
6775 em->orig_start = em->start;
6776 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6777 if (create == 0 && !PageUptodate(page)) {
6778 if (btrfs_file_extent_compression(leaf, item) !=
6779 BTRFS_COMPRESS_NONE) {
6780 ret = uncompress_inline(path, inode, page,
6782 extent_offset, item);
6789 read_extent_buffer(leaf, map + pg_offset, ptr,
6791 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6792 memset(map + pg_offset + copy_size, 0,
6793 PAGE_CACHE_SIZE - pg_offset -
6798 flush_dcache_page(page);
6799 } else if (create && PageUptodate(page)) {
6803 free_extent_map(em);
6806 btrfs_release_path(path);
6807 trans = btrfs_join_transaction(root);
6810 return ERR_CAST(trans);
6814 write_extent_buffer(leaf, map + pg_offset, ptr,
6817 btrfs_mark_buffer_dirty(leaf);
6819 set_extent_uptodate(io_tree, em->start,
6820 extent_map_end(em) - 1, NULL, GFP_NOFS);
6825 em->orig_start = start;
6828 em->block_start = EXTENT_MAP_HOLE;
6829 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6831 btrfs_release_path(path);
6832 if (em->start > start || extent_map_end(em) <= start) {
6833 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6834 em->start, em->len, start, len);
6840 write_lock(&em_tree->lock);
6841 ret = add_extent_mapping(em_tree, em, 0);
6842 /* it is possible that someone inserted the extent into the tree
6843 * while we had the lock dropped. It is also possible that
6844 * an overlapping map exists in the tree
6846 if (ret == -EEXIST) {
6847 struct extent_map *existing;
6851 existing = search_extent_mapping(em_tree, start, len);
6853 * existing will always be non-NULL, since there must be
6854 * extent causing the -EEXIST.
6856 if (start >= extent_map_end(existing) ||
6857 start <= existing->start) {
6859 * The existing extent map is the one nearest to
6860 * the [start, start + len) range which overlaps
6862 err = merge_extent_mapping(em_tree, existing,
6864 free_extent_map(existing);
6866 free_extent_map(em);
6870 free_extent_map(em);
6875 write_unlock(&em_tree->lock);
6878 trace_btrfs_get_extent(root, em);
6881 btrfs_free_path(path);
6883 ret = btrfs_end_transaction(trans, root);
6888 free_extent_map(em);
6889 return ERR_PTR(err);
6891 BUG_ON(!em); /* Error is always set */
6895 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6896 size_t pg_offset, u64 start, u64 len,
6899 struct extent_map *em;
6900 struct extent_map *hole_em = NULL;
6901 u64 range_start = start;
6907 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6914 * - a pre-alloc extent,
6915 * there might actually be delalloc bytes behind it.
6917 if (em->block_start != EXTENT_MAP_HOLE &&
6918 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6924 /* check to see if we've wrapped (len == -1 or similar) */
6933 /* ok, we didn't find anything, lets look for delalloc */
6934 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6935 end, len, EXTENT_DELALLOC, 1);
6936 found_end = range_start + found;
6937 if (found_end < range_start)
6938 found_end = (u64)-1;
6941 * we didn't find anything useful, return
6942 * the original results from get_extent()
6944 if (range_start > end || found_end <= start) {
6950 /* adjust the range_start to make sure it doesn't
6951 * go backwards from the start they passed in
6953 range_start = max(start, range_start);
6954 found = found_end - range_start;
6957 u64 hole_start = start;
6960 em = alloc_extent_map();
6966 * when btrfs_get_extent can't find anything it
6967 * returns one huge hole
6969 * make sure what it found really fits our range, and
6970 * adjust to make sure it is based on the start from
6974 u64 calc_end = extent_map_end(hole_em);
6976 if (calc_end <= start || (hole_em->start > end)) {
6977 free_extent_map(hole_em);
6980 hole_start = max(hole_em->start, start);
6981 hole_len = calc_end - hole_start;
6985 if (hole_em && range_start > hole_start) {
6986 /* our hole starts before our delalloc, so we
6987 * have to return just the parts of the hole
6988 * that go until the delalloc starts
6990 em->len = min(hole_len,
6991 range_start - hole_start);
6992 em->start = hole_start;
6993 em->orig_start = hole_start;
6995 * don't adjust block start at all,
6996 * it is fixed at EXTENT_MAP_HOLE
6998 em->block_start = hole_em->block_start;
6999 em->block_len = hole_len;
7000 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7001 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7003 em->start = range_start;
7005 em->orig_start = range_start;
7006 em->block_start = EXTENT_MAP_DELALLOC;
7007 em->block_len = found;
7009 } else if (hole_em) {
7014 free_extent_map(hole_em);
7016 free_extent_map(em);
7017 return ERR_PTR(err);
7022 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7025 struct btrfs_root *root = BTRFS_I(inode)->root;
7026 struct extent_map *em;
7027 struct btrfs_key ins;
7031 alloc_hint = get_extent_allocation_hint(inode, start, len);
7032 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
7033 alloc_hint, &ins, 1, 1);
7035 return ERR_PTR(ret);
7037 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
7038 ins.offset, ins.offset, ins.offset, 0);
7040 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7044 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
7045 ins.offset, ins.offset, 0);
7047 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7048 free_extent_map(em);
7049 return ERR_PTR(ret);
7056 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7057 * block must be cow'd
7059 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7060 u64 *orig_start, u64 *orig_block_len,
7063 struct btrfs_trans_handle *trans;
7064 struct btrfs_path *path;
7066 struct extent_buffer *leaf;
7067 struct btrfs_root *root = BTRFS_I(inode)->root;
7068 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7069 struct btrfs_file_extent_item *fi;
7070 struct btrfs_key key;
7077 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7079 path = btrfs_alloc_path();
7083 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
7088 slot = path->slots[0];
7091 /* can't find the item, must cow */
7098 leaf = path->nodes[0];
7099 btrfs_item_key_to_cpu(leaf, &key, slot);
7100 if (key.objectid != btrfs_ino(inode) ||
7101 key.type != BTRFS_EXTENT_DATA_KEY) {
7102 /* not our file or wrong item type, must cow */
7106 if (key.offset > offset) {
7107 /* Wrong offset, must cow */
7111 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7112 found_type = btrfs_file_extent_type(leaf, fi);
7113 if (found_type != BTRFS_FILE_EXTENT_REG &&
7114 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7115 /* not a regular extent, must cow */
7119 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7122 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7123 if (extent_end <= offset)
7126 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7127 if (disk_bytenr == 0)
7130 if (btrfs_file_extent_compression(leaf, fi) ||
7131 btrfs_file_extent_encryption(leaf, fi) ||
7132 btrfs_file_extent_other_encoding(leaf, fi))
7135 backref_offset = btrfs_file_extent_offset(leaf, fi);
7138 *orig_start = key.offset - backref_offset;
7139 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7140 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7143 if (btrfs_extent_readonly(root, disk_bytenr))
7146 num_bytes = min(offset + *len, extent_end) - offset;
7147 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7150 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7151 ret = test_range_bit(io_tree, offset, range_end,
7152 EXTENT_DELALLOC, 0, NULL);
7159 btrfs_release_path(path);
7162 * look for other files referencing this extent, if we
7163 * find any we must cow
7165 trans = btrfs_join_transaction(root);
7166 if (IS_ERR(trans)) {
7171 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7172 key.offset - backref_offset, disk_bytenr);
7173 btrfs_end_transaction(trans, root);
7180 * adjust disk_bytenr and num_bytes to cover just the bytes
7181 * in this extent we are about to write. If there
7182 * are any csums in that range we have to cow in order
7183 * to keep the csums correct
7185 disk_bytenr += backref_offset;
7186 disk_bytenr += offset - key.offset;
7187 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7190 * all of the above have passed, it is safe to overwrite this extent
7196 btrfs_free_path(path);
7200 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7202 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7204 void **pagep = NULL;
7205 struct page *page = NULL;
7209 start_idx = start >> PAGE_CACHE_SHIFT;
7212 * end is the last byte in the last page. end == start is legal
7214 end_idx = end >> PAGE_CACHE_SHIFT;
7218 /* Most of the code in this while loop is lifted from
7219 * find_get_page. It's been modified to begin searching from a
7220 * page and return just the first page found in that range. If the
7221 * found idx is less than or equal to the end idx then we know that
7222 * a page exists. If no pages are found or if those pages are
7223 * outside of the range then we're fine (yay!) */
7224 while (page == NULL &&
7225 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7226 page = radix_tree_deref_slot(pagep);
7227 if (unlikely(!page))
7230 if (radix_tree_exception(page)) {
7231 if (radix_tree_deref_retry(page)) {
7236 * Otherwise, shmem/tmpfs must be storing a swap entry
7237 * here as an exceptional entry: so return it without
7238 * attempting to raise page count.
7241 break; /* TODO: Is this relevant for this use case? */
7244 if (!page_cache_get_speculative(page)) {
7250 * Has the page moved?
7251 * This is part of the lockless pagecache protocol. See
7252 * include/linux/pagemap.h for details.
7254 if (unlikely(page != *pagep)) {
7255 page_cache_release(page);
7261 if (page->index <= end_idx)
7263 page_cache_release(page);
7270 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7271 struct extent_state **cached_state, int writing)
7273 struct btrfs_ordered_extent *ordered;
7277 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7280 * We're concerned with the entire range that we're going to be
7281 * doing DIO to, so we need to make sure theres no ordered
7282 * extents in this range.
7284 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7285 lockend - lockstart + 1);
7288 * We need to make sure there are no buffered pages in this
7289 * range either, we could have raced between the invalidate in
7290 * generic_file_direct_write and locking the extent. The
7291 * invalidate needs to happen so that reads after a write do not
7296 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7299 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7300 cached_state, GFP_NOFS);
7303 btrfs_start_ordered_extent(inode, ordered, 1);
7304 btrfs_put_ordered_extent(ordered);
7306 /* Screw you mmap */
7307 ret = btrfs_fdatawrite_range(inode, lockstart, lockend);
7310 ret = filemap_fdatawait_range(inode->i_mapping,
7317 * If we found a page that couldn't be invalidated just
7318 * fall back to buffered.
7320 ret = invalidate_inode_pages2_range(inode->i_mapping,
7321 lockstart >> PAGE_CACHE_SHIFT,
7322 lockend >> PAGE_CACHE_SHIFT);
7333 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7334 u64 len, u64 orig_start,
7335 u64 block_start, u64 block_len,
7336 u64 orig_block_len, u64 ram_bytes,
7339 struct extent_map_tree *em_tree;
7340 struct extent_map *em;
7341 struct btrfs_root *root = BTRFS_I(inode)->root;
7344 em_tree = &BTRFS_I(inode)->extent_tree;
7345 em = alloc_extent_map();
7347 return ERR_PTR(-ENOMEM);
7350 em->orig_start = orig_start;
7351 em->mod_start = start;
7354 em->block_len = block_len;
7355 em->block_start = block_start;
7356 em->bdev = root->fs_info->fs_devices->latest_bdev;
7357 em->orig_block_len = orig_block_len;
7358 em->ram_bytes = ram_bytes;
7359 em->generation = -1;
7360 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7361 if (type == BTRFS_ORDERED_PREALLOC)
7362 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7365 btrfs_drop_extent_cache(inode, em->start,
7366 em->start + em->len - 1, 0);
7367 write_lock(&em_tree->lock);
7368 ret = add_extent_mapping(em_tree, em, 1);
7369 write_unlock(&em_tree->lock);
7370 } while (ret == -EEXIST);
7373 free_extent_map(em);
7374 return ERR_PTR(ret);
7381 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7382 struct buffer_head *bh_result, int create)
7384 struct extent_map *em;
7385 struct btrfs_root *root = BTRFS_I(inode)->root;
7386 struct extent_state *cached_state = NULL;
7387 u64 start = iblock << inode->i_blkbits;
7388 u64 lockstart, lockend;
7389 u64 len = bh_result->b_size;
7390 u64 *outstanding_extents = NULL;
7391 int unlock_bits = EXTENT_LOCKED;
7395 unlock_bits |= EXTENT_DIRTY;
7397 len = min_t(u64, len, root->sectorsize);
7400 lockend = start + len - 1;
7402 if (current->journal_info) {
7404 * Need to pull our outstanding extents and set journal_info to NULL so
7405 * that anything that needs to check if there's a transction doesn't get
7408 outstanding_extents = current->journal_info;
7409 current->journal_info = NULL;
7413 * If this errors out it's because we couldn't invalidate pagecache for
7414 * this range and we need to fallback to buffered.
7416 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
7419 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7426 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7427 * io. INLINE is special, and we could probably kludge it in here, but
7428 * it's still buffered so for safety lets just fall back to the generic
7431 * For COMPRESSED we _have_ to read the entire extent in so we can
7432 * decompress it, so there will be buffering required no matter what we
7433 * do, so go ahead and fallback to buffered.
7435 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7436 * to buffered IO. Don't blame me, this is the price we pay for using
7439 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7440 em->block_start == EXTENT_MAP_INLINE) {
7441 free_extent_map(em);
7446 /* Just a good old fashioned hole, return */
7447 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7448 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7449 free_extent_map(em);
7454 * We don't allocate a new extent in the following cases
7456 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7458 * 2) The extent is marked as PREALLOC. We're good to go here and can
7459 * just use the extent.
7463 len = min(len, em->len - (start - em->start));
7464 lockstart = start + len;
7468 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7469 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7470 em->block_start != EXTENT_MAP_HOLE)) {
7472 u64 block_start, orig_start, orig_block_len, ram_bytes;
7474 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7475 type = BTRFS_ORDERED_PREALLOC;
7477 type = BTRFS_ORDERED_NOCOW;
7478 len = min(len, em->len - (start - em->start));
7479 block_start = em->block_start + (start - em->start);
7481 if (can_nocow_extent(inode, start, &len, &orig_start,
7482 &orig_block_len, &ram_bytes) == 1) {
7483 if (type == BTRFS_ORDERED_PREALLOC) {
7484 free_extent_map(em);
7485 em = create_pinned_em(inode, start, len,
7496 ret = btrfs_add_ordered_extent_dio(inode, start,
7497 block_start, len, len, type);
7499 free_extent_map(em);
7507 * this will cow the extent, reset the len in case we changed
7510 len = bh_result->b_size;
7511 free_extent_map(em);
7512 em = btrfs_new_extent_direct(inode, start, len);
7517 len = min(len, em->len - (start - em->start));
7519 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7521 bh_result->b_size = len;
7522 bh_result->b_bdev = em->bdev;
7523 set_buffer_mapped(bh_result);
7525 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7526 set_buffer_new(bh_result);
7529 * Need to update the i_size under the extent lock so buffered
7530 * readers will get the updated i_size when we unlock.
7532 if (start + len > i_size_read(inode))
7533 i_size_write(inode, start + len);
7536 * If we have an outstanding_extents count still set then we're
7537 * within our reservation, otherwise we need to adjust our inode
7538 * counter appropriately.
7540 if (*outstanding_extents) {
7541 (*outstanding_extents)--;
7543 spin_lock(&BTRFS_I(inode)->lock);
7544 BTRFS_I(inode)->outstanding_extents++;
7545 spin_unlock(&BTRFS_I(inode)->lock);
7548 current->journal_info = outstanding_extents;
7549 btrfs_free_reserved_data_space(inode, len);
7550 set_bit(BTRFS_INODE_DIO_READY, &BTRFS_I(inode)->runtime_flags);
7554 * In the case of write we need to clear and unlock the entire range,
7555 * in the case of read we need to unlock only the end area that we
7556 * aren't using if there is any left over space.
7558 if (lockstart < lockend) {
7559 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7560 lockend, unlock_bits, 1, 0,
7561 &cached_state, GFP_NOFS);
7563 free_extent_state(cached_state);
7566 free_extent_map(em);
7571 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7572 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7573 if (outstanding_extents)
7574 current->journal_info = outstanding_extents;
7578 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7579 int rw, int mirror_num)
7581 struct btrfs_root *root = BTRFS_I(inode)->root;
7584 BUG_ON(rw & REQ_WRITE);
7588 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7589 BTRFS_WQ_ENDIO_DIO_REPAIR);
7593 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7599 static int btrfs_check_dio_repairable(struct inode *inode,
7600 struct bio *failed_bio,
7601 struct io_failure_record *failrec,
7606 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7607 failrec->logical, failrec->len);
7608 if (num_copies == 1) {
7610 * we only have a single copy of the data, so don't bother with
7611 * all the retry and error correction code that follows. no
7612 * matter what the error is, it is very likely to persist.
7614 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7615 num_copies, failrec->this_mirror, failed_mirror);
7619 failrec->failed_mirror = failed_mirror;
7620 failrec->this_mirror++;
7621 if (failrec->this_mirror == failed_mirror)
7622 failrec->this_mirror++;
7624 if (failrec->this_mirror > num_copies) {
7625 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7626 num_copies, failrec->this_mirror, failed_mirror);
7633 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7634 struct page *page, u64 start, u64 end,
7635 int failed_mirror, bio_end_io_t *repair_endio,
7638 struct io_failure_record *failrec;
7644 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7646 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7650 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7653 free_io_failure(inode, failrec);
7657 if (failed_bio->bi_vcnt > 1)
7658 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7660 read_mode = READ_SYNC;
7662 isector = start - btrfs_io_bio(failed_bio)->logical;
7663 isector >>= inode->i_sb->s_blocksize_bits;
7664 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7665 0, isector, repair_endio, repair_arg);
7667 free_io_failure(inode, failrec);
7671 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7672 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7673 read_mode, failrec->this_mirror, failrec->in_validation);
7675 ret = submit_dio_repair_bio(inode, bio, read_mode,
7676 failrec->this_mirror);
7678 free_io_failure(inode, failrec);
7685 struct btrfs_retry_complete {
7686 struct completion done;
7687 struct inode *inode;
7692 static void btrfs_retry_endio_nocsum(struct bio *bio, int err)
7694 struct btrfs_retry_complete *done = bio->bi_private;
7695 struct bio_vec *bvec;
7702 bio_for_each_segment_all(bvec, bio, i)
7703 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7705 complete(&done->done);
7709 static int __btrfs_correct_data_nocsum(struct inode *inode,
7710 struct btrfs_io_bio *io_bio)
7712 struct bio_vec *bvec;
7713 struct btrfs_retry_complete done;
7718 start = io_bio->logical;
7721 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7725 init_completion(&done.done);
7727 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7728 start + bvec->bv_len - 1,
7730 btrfs_retry_endio_nocsum, &done);
7734 wait_for_completion(&done.done);
7736 if (!done.uptodate) {
7737 /* We might have another mirror, so try again */
7741 start += bvec->bv_len;
7747 static void btrfs_retry_endio(struct bio *bio, int err)
7749 struct btrfs_retry_complete *done = bio->bi_private;
7750 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7751 struct bio_vec *bvec;
7760 bio_for_each_segment_all(bvec, bio, i) {
7761 ret = __readpage_endio_check(done->inode, io_bio, i,
7763 done->start, bvec->bv_len);
7765 clean_io_failure(done->inode, done->start,
7771 done->uptodate = uptodate;
7773 complete(&done->done);
7777 static int __btrfs_subio_endio_read(struct inode *inode,
7778 struct btrfs_io_bio *io_bio, int err)
7780 struct bio_vec *bvec;
7781 struct btrfs_retry_complete done;
7788 start = io_bio->logical;
7791 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7792 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7793 0, start, bvec->bv_len);
7799 init_completion(&done.done);
7801 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7802 start + bvec->bv_len - 1,
7804 btrfs_retry_endio, &done);
7810 wait_for_completion(&done.done);
7812 if (!done.uptodate) {
7813 /* We might have another mirror, so try again */
7817 offset += bvec->bv_len;
7818 start += bvec->bv_len;
7824 static int btrfs_subio_endio_read(struct inode *inode,
7825 struct btrfs_io_bio *io_bio, int err)
7827 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7831 return __btrfs_correct_data_nocsum(inode, io_bio);
7835 return __btrfs_subio_endio_read(inode, io_bio, err);
7839 static void btrfs_endio_direct_read(struct bio *bio, int err)
7841 struct btrfs_dio_private *dip = bio->bi_private;
7842 struct inode *inode = dip->inode;
7843 struct bio *dio_bio;
7844 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7846 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
7847 err = btrfs_subio_endio_read(inode, io_bio, err);
7849 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
7850 dip->logical_offset + dip->bytes - 1);
7851 dio_bio = dip->dio_bio;
7855 /* If we had a csum failure make sure to clear the uptodate flag */
7857 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7858 dio_end_io(dio_bio, err);
7861 io_bio->end_io(io_bio, err);
7865 static void btrfs_endio_direct_write(struct bio *bio, int err)
7867 struct btrfs_dio_private *dip = bio->bi_private;
7868 struct inode *inode = dip->inode;
7869 struct btrfs_root *root = BTRFS_I(inode)->root;
7870 struct btrfs_ordered_extent *ordered = NULL;
7871 u64 ordered_offset = dip->logical_offset;
7872 u64 ordered_bytes = dip->bytes;
7873 struct bio *dio_bio;
7877 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7879 ordered_bytes, !err);
7883 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
7884 finish_ordered_fn, NULL, NULL);
7885 btrfs_queue_work(root->fs_info->endio_write_workers,
7889 * our bio might span multiple ordered extents. If we haven't
7890 * completed the accounting for the whole dio, go back and try again
7892 if (ordered_offset < dip->logical_offset + dip->bytes) {
7893 ordered_bytes = dip->logical_offset + dip->bytes -
7898 dio_bio = dip->dio_bio;
7902 /* If we had an error make sure to clear the uptodate flag */
7904 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7905 dio_end_io(dio_bio, err);
7909 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7910 struct bio *bio, int mirror_num,
7911 unsigned long bio_flags, u64 offset)
7914 struct btrfs_root *root = BTRFS_I(inode)->root;
7915 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7916 BUG_ON(ret); /* -ENOMEM */
7920 static void btrfs_end_dio_bio(struct bio *bio, int err)
7922 struct btrfs_dio_private *dip = bio->bi_private;
7925 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7926 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7927 btrfs_ino(dip->inode), bio->bi_rw,
7928 (unsigned long long)bio->bi_iter.bi_sector,
7929 bio->bi_iter.bi_size, err);
7931 if (dip->subio_endio)
7932 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
7938 * before atomic variable goto zero, we must make sure
7939 * dip->errors is perceived to be set.
7941 smp_mb__before_atomic();
7944 /* if there are more bios still pending for this dio, just exit */
7945 if (!atomic_dec_and_test(&dip->pending_bios))
7949 bio_io_error(dip->orig_bio);
7951 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7952 bio_endio(dip->orig_bio, 0);
7958 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7959 u64 first_sector, gfp_t gfp_flags)
7961 int nr_vecs = bio_get_nr_vecs(bdev);
7962 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7965 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
7966 struct inode *inode,
7967 struct btrfs_dio_private *dip,
7971 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7972 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
7976 * We load all the csum data we need when we submit
7977 * the first bio to reduce the csum tree search and
7980 if (dip->logical_offset == file_offset) {
7981 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
7987 if (bio == dip->orig_bio)
7990 file_offset -= dip->logical_offset;
7991 file_offset >>= inode->i_sb->s_blocksize_bits;
7992 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
7997 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7998 int rw, u64 file_offset, int skip_sum,
8001 struct btrfs_dio_private *dip = bio->bi_private;
8002 int write = rw & REQ_WRITE;
8003 struct btrfs_root *root = BTRFS_I(inode)->root;
8007 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8012 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
8013 BTRFS_WQ_ENDIO_DATA);
8021 if (write && async_submit) {
8022 ret = btrfs_wq_submit_bio(root->fs_info,
8023 inode, rw, bio, 0, 0,
8025 __btrfs_submit_bio_start_direct_io,
8026 __btrfs_submit_bio_done);
8030 * If we aren't doing async submit, calculate the csum of the
8033 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
8037 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
8043 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
8049 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
8052 struct inode *inode = dip->inode;
8053 struct btrfs_root *root = BTRFS_I(inode)->root;
8055 struct bio *orig_bio = dip->orig_bio;
8056 struct bio_vec *bvec = orig_bio->bi_io_vec;
8057 u64 start_sector = orig_bio->bi_iter.bi_sector;
8058 u64 file_offset = dip->logical_offset;
8063 int async_submit = 0;
8065 map_length = orig_bio->bi_iter.bi_size;
8066 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
8067 &map_length, NULL, 0);
8071 if (map_length >= orig_bio->bi_iter.bi_size) {
8073 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8077 /* async crcs make it difficult to collect full stripe writes. */
8078 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8083 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8087 bio->bi_private = dip;
8088 bio->bi_end_io = btrfs_end_dio_bio;
8089 btrfs_io_bio(bio)->logical = file_offset;
8090 atomic_inc(&dip->pending_bios);
8092 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
8093 if (map_length < submit_len + bvec->bv_len ||
8094 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
8095 bvec->bv_offset) < bvec->bv_len) {
8097 * inc the count before we submit the bio so
8098 * we know the end IO handler won't happen before
8099 * we inc the count. Otherwise, the dip might get freed
8100 * before we're done setting it up
8102 atomic_inc(&dip->pending_bios);
8103 ret = __btrfs_submit_dio_bio(bio, inode, rw,
8104 file_offset, skip_sum,
8108 atomic_dec(&dip->pending_bios);
8112 start_sector += submit_len >> 9;
8113 file_offset += submit_len;
8118 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8119 start_sector, GFP_NOFS);
8122 bio->bi_private = dip;
8123 bio->bi_end_io = btrfs_end_dio_bio;
8124 btrfs_io_bio(bio)->logical = file_offset;
8126 map_length = orig_bio->bi_iter.bi_size;
8127 ret = btrfs_map_block(root->fs_info, rw,
8129 &map_length, NULL, 0);
8135 submit_len += bvec->bv_len;
8142 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
8151 * before atomic variable goto zero, we must
8152 * make sure dip->errors is perceived to be set.
8154 smp_mb__before_atomic();
8155 if (atomic_dec_and_test(&dip->pending_bios))
8156 bio_io_error(dip->orig_bio);
8158 /* bio_end_io() will handle error, so we needn't return it */
8162 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
8163 struct inode *inode, loff_t file_offset)
8165 struct btrfs_dio_private *dip = NULL;
8166 struct bio *io_bio = NULL;
8167 struct btrfs_io_bio *btrfs_bio;
8169 int write = rw & REQ_WRITE;
8172 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8174 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
8180 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8186 dip->private = dio_bio->bi_private;
8188 dip->logical_offset = file_offset;
8189 dip->bytes = dio_bio->bi_iter.bi_size;
8190 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8191 io_bio->bi_private = dip;
8192 dip->orig_bio = io_bio;
8193 dip->dio_bio = dio_bio;
8194 atomic_set(&dip->pending_bios, 0);
8195 btrfs_bio = btrfs_io_bio(io_bio);
8196 btrfs_bio->logical = file_offset;
8199 io_bio->bi_end_io = btrfs_endio_direct_write;
8201 io_bio->bi_end_io = btrfs_endio_direct_read;
8202 dip->subio_endio = btrfs_subio_endio_read;
8205 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8209 if (btrfs_bio->end_io)
8210 btrfs_bio->end_io(btrfs_bio, ret);
8214 * If we arrived here it means either we failed to submit the dip
8215 * or we either failed to clone the dio_bio or failed to allocate the
8216 * dip. If we cloned the dio_bio and allocated the dip, we can just
8217 * call bio_endio against our io_bio so that we get proper resource
8218 * cleanup if we fail to submit the dip, otherwise, we must do the
8219 * same as btrfs_endio_direct_[write|read] because we can't call these
8220 * callbacks - they require an allocated dip and a clone of dio_bio.
8222 if (io_bio && dip) {
8223 bio_endio(io_bio, ret);
8225 * The end io callbacks free our dip, do the final put on io_bio
8226 * and all the cleanup and final put for dio_bio (through
8233 struct btrfs_ordered_extent *ordered;
8235 ordered = btrfs_lookup_ordered_extent(inode,
8237 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
8239 * Decrements our ref on the ordered extent and removes
8240 * the ordered extent from the inode's ordered tree,
8241 * doing all the proper resource cleanup such as for the
8242 * reserved space and waking up any waiters for this
8243 * ordered extent (through btrfs_remove_ordered_extent).
8245 btrfs_finish_ordered_io(ordered);
8247 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8248 file_offset + dio_bio->bi_iter.bi_size - 1);
8250 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
8252 * Releases and cleans up our dio_bio, no need to bio_put()
8253 * nor bio_endio()/bio_io_error() against dio_bio.
8255 dio_end_io(dio_bio, ret);
8262 static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb,
8263 const struct iov_iter *iter, loff_t offset)
8267 unsigned blocksize_mask = root->sectorsize - 1;
8268 ssize_t retval = -EINVAL;
8270 if (offset & blocksize_mask)
8273 if (iov_iter_alignment(iter) & blocksize_mask)
8276 /* If this is a write we don't need to check anymore */
8277 if (iov_iter_rw(iter) == WRITE)
8280 * Check to make sure we don't have duplicate iov_base's in this
8281 * iovec, if so return EINVAL, otherwise we'll get csum errors
8282 * when reading back.
8284 for (seg = 0; seg < iter->nr_segs; seg++) {
8285 for (i = seg + 1; i < iter->nr_segs; i++) {
8286 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8295 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
8298 struct file *file = iocb->ki_filp;
8299 struct inode *inode = file->f_mapping->host;
8300 u64 outstanding_extents = 0;
8304 bool relock = false;
8307 if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset))
8310 inode_dio_begin(inode);
8311 smp_mb__after_atomic();
8314 * The generic stuff only does filemap_write_and_wait_range, which
8315 * isn't enough if we've written compressed pages to this area, so
8316 * we need to flush the dirty pages again to make absolutely sure
8317 * that any outstanding dirty pages are on disk.
8319 count = iov_iter_count(iter);
8320 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8321 &BTRFS_I(inode)->runtime_flags))
8322 filemap_fdatawrite_range(inode->i_mapping, offset,
8323 offset + count - 1);
8325 if (iov_iter_rw(iter) == WRITE) {
8327 * If the write DIO is beyond the EOF, we need update
8328 * the isize, but it is protected by i_mutex. So we can
8329 * not unlock the i_mutex at this case.
8331 if (offset + count <= inode->i_size) {
8332 mutex_unlock(&inode->i_mutex);
8335 ret = btrfs_delalloc_reserve_space(inode, count);
8338 outstanding_extents = div64_u64(count +
8339 BTRFS_MAX_EXTENT_SIZE - 1,
8340 BTRFS_MAX_EXTENT_SIZE);
8343 * We need to know how many extents we reserved so that we can
8344 * do the accounting properly if we go over the number we
8345 * originally calculated. Abuse current->journal_info for this.
8347 current->journal_info = &outstanding_extents;
8348 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8349 &BTRFS_I(inode)->runtime_flags)) {
8350 inode_dio_end(inode);
8351 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8355 ret = __blockdev_direct_IO(iocb, inode,
8356 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8357 iter, offset, btrfs_get_blocks_direct, NULL,
8358 btrfs_submit_direct, flags);
8359 if (iov_iter_rw(iter) == WRITE) {
8360 current->journal_info = NULL;
8361 if (ret < 0 && ret != -EIOCBQUEUED) {
8363 * If the error comes from submitting stage,
8364 * btrfs_get_blocsk_direct() has free'd data space,
8365 * and metadata space will be handled by
8366 * finish_ordered_fn, don't do that again to make
8367 * sure bytes_may_use is correct.
8369 if (!test_and_clear_bit(BTRFS_INODE_DIO_READY,
8370 &BTRFS_I(inode)->runtime_flags))
8371 btrfs_delalloc_release_space(inode, count);
8372 } else if (ret >= 0 && (size_t)ret < count)
8373 btrfs_delalloc_release_space(inode,
8374 count - (size_t)ret);
8378 inode_dio_end(inode);
8380 mutex_lock(&inode->i_mutex);
8385 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8387 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8388 __u64 start, __u64 len)
8392 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8396 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8399 int btrfs_readpage(struct file *file, struct page *page)
8401 struct extent_io_tree *tree;
8402 tree = &BTRFS_I(page->mapping->host)->io_tree;
8403 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8406 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8408 struct extent_io_tree *tree;
8411 if (current->flags & PF_MEMALLOC) {
8412 redirty_page_for_writepage(wbc, page);
8416 tree = &BTRFS_I(page->mapping->host)->io_tree;
8417 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8420 static int btrfs_writepages(struct address_space *mapping,
8421 struct writeback_control *wbc)
8423 struct extent_io_tree *tree;
8425 tree = &BTRFS_I(mapping->host)->io_tree;
8426 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8430 btrfs_readpages(struct file *file, struct address_space *mapping,
8431 struct list_head *pages, unsigned nr_pages)
8433 struct extent_io_tree *tree;
8434 tree = &BTRFS_I(mapping->host)->io_tree;
8435 return extent_readpages(tree, mapping, pages, nr_pages,
8438 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8440 struct extent_io_tree *tree;
8441 struct extent_map_tree *map;
8444 tree = &BTRFS_I(page->mapping->host)->io_tree;
8445 map = &BTRFS_I(page->mapping->host)->extent_tree;
8446 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8448 ClearPagePrivate(page);
8449 set_page_private(page, 0);
8450 page_cache_release(page);
8455 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8457 if (PageWriteback(page) || PageDirty(page))
8459 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8462 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8463 unsigned int length)
8465 struct inode *inode = page->mapping->host;
8466 struct extent_io_tree *tree;
8467 struct btrfs_ordered_extent *ordered;
8468 struct extent_state *cached_state = NULL;
8469 u64 page_start = page_offset(page);
8470 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
8471 int inode_evicting = inode->i_state & I_FREEING;
8474 * we have the page locked, so new writeback can't start,
8475 * and the dirty bit won't be cleared while we are here.
8477 * Wait for IO on this page so that we can safely clear
8478 * the PagePrivate2 bit and do ordered accounting
8480 wait_on_page_writeback(page);
8482 tree = &BTRFS_I(inode)->io_tree;
8484 btrfs_releasepage(page, GFP_NOFS);
8488 if (!inode_evicting)
8489 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
8490 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8493 * IO on this page will never be started, so we need
8494 * to account for any ordered extents now
8496 if (!inode_evicting)
8497 clear_extent_bit(tree, page_start, page_end,
8498 EXTENT_DIRTY | EXTENT_DELALLOC |
8499 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8500 EXTENT_DEFRAG, 1, 0, &cached_state,
8503 * whoever cleared the private bit is responsible
8504 * for the finish_ordered_io
8506 if (TestClearPagePrivate2(page)) {
8507 struct btrfs_ordered_inode_tree *tree;
8510 tree = &BTRFS_I(inode)->ordered_tree;
8512 spin_lock_irq(&tree->lock);
8513 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8514 new_len = page_start - ordered->file_offset;
8515 if (new_len < ordered->truncated_len)
8516 ordered->truncated_len = new_len;
8517 spin_unlock_irq(&tree->lock);
8519 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8521 PAGE_CACHE_SIZE, 1))
8522 btrfs_finish_ordered_io(ordered);
8524 btrfs_put_ordered_extent(ordered);
8525 if (!inode_evicting) {
8526 cached_state = NULL;
8527 lock_extent_bits(tree, page_start, page_end, 0,
8532 if (!inode_evicting) {
8533 clear_extent_bit(tree, page_start, page_end,
8534 EXTENT_LOCKED | EXTENT_DIRTY |
8535 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8536 EXTENT_DEFRAG, 1, 1,
8537 &cached_state, GFP_NOFS);
8539 __btrfs_releasepage(page, GFP_NOFS);
8542 ClearPageChecked(page);
8543 if (PagePrivate(page)) {
8544 ClearPagePrivate(page);
8545 set_page_private(page, 0);
8546 page_cache_release(page);
8551 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8552 * called from a page fault handler when a page is first dirtied. Hence we must
8553 * be careful to check for EOF conditions here. We set the page up correctly
8554 * for a written page which means we get ENOSPC checking when writing into
8555 * holes and correct delalloc and unwritten extent mapping on filesystems that
8556 * support these features.
8558 * We are not allowed to take the i_mutex here so we have to play games to
8559 * protect against truncate races as the page could now be beyond EOF. Because
8560 * vmtruncate() writes the inode size before removing pages, once we have the
8561 * page lock we can determine safely if the page is beyond EOF. If it is not
8562 * beyond EOF, then the page is guaranteed safe against truncation until we
8565 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8567 struct page *page = vmf->page;
8568 struct inode *inode = file_inode(vma->vm_file);
8569 struct btrfs_root *root = BTRFS_I(inode)->root;
8570 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8571 struct btrfs_ordered_extent *ordered;
8572 struct extent_state *cached_state = NULL;
8574 unsigned long zero_start;
8581 sb_start_pagefault(inode->i_sb);
8582 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
8584 ret = file_update_time(vma->vm_file);
8590 else /* -ENOSPC, -EIO, etc */
8591 ret = VM_FAULT_SIGBUS;
8597 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8600 size = i_size_read(inode);
8601 page_start = page_offset(page);
8602 page_end = page_start + PAGE_CACHE_SIZE - 1;
8604 if ((page->mapping != inode->i_mapping) ||
8605 (page_start >= size)) {
8606 /* page got truncated out from underneath us */
8609 wait_on_page_writeback(page);
8611 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
8612 set_page_extent_mapped(page);
8615 * we can't set the delalloc bits if there are pending ordered
8616 * extents. Drop our locks and wait for them to finish
8618 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8620 unlock_extent_cached(io_tree, page_start, page_end,
8621 &cached_state, GFP_NOFS);
8623 btrfs_start_ordered_extent(inode, ordered, 1);
8624 btrfs_put_ordered_extent(ordered);
8629 * XXX - page_mkwrite gets called every time the page is dirtied, even
8630 * if it was already dirty, so for space accounting reasons we need to
8631 * clear any delalloc bits for the range we are fixing to save. There
8632 * is probably a better way to do this, but for now keep consistent with
8633 * prepare_pages in the normal write path.
8635 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
8636 EXTENT_DIRTY | EXTENT_DELALLOC |
8637 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8638 0, 0, &cached_state, GFP_NOFS);
8640 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
8643 unlock_extent_cached(io_tree, page_start, page_end,
8644 &cached_state, GFP_NOFS);
8645 ret = VM_FAULT_SIGBUS;
8650 /* page is wholly or partially inside EOF */
8651 if (page_start + PAGE_CACHE_SIZE > size)
8652 zero_start = size & ~PAGE_CACHE_MASK;
8654 zero_start = PAGE_CACHE_SIZE;
8656 if (zero_start != PAGE_CACHE_SIZE) {
8658 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
8659 flush_dcache_page(page);
8662 ClearPageChecked(page);
8663 set_page_dirty(page);
8664 SetPageUptodate(page);
8666 BTRFS_I(inode)->last_trans = root->fs_info->generation;
8667 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8668 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8670 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
8674 sb_end_pagefault(inode->i_sb);
8675 return VM_FAULT_LOCKED;
8679 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
8681 sb_end_pagefault(inode->i_sb);
8685 static int btrfs_truncate(struct inode *inode)
8687 struct btrfs_root *root = BTRFS_I(inode)->root;
8688 struct btrfs_block_rsv *rsv;
8691 struct btrfs_trans_handle *trans;
8692 u64 mask = root->sectorsize - 1;
8693 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
8695 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8701 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8702 * 3 things going on here
8704 * 1) We need to reserve space for our orphan item and the space to
8705 * delete our orphan item. Lord knows we don't want to have a dangling
8706 * orphan item because we didn't reserve space to remove it.
8708 * 2) We need to reserve space to update our inode.
8710 * 3) We need to have something to cache all the space that is going to
8711 * be free'd up by the truncate operation, but also have some slack
8712 * space reserved in case it uses space during the truncate (thank you
8713 * very much snapshotting).
8715 * And we need these to all be seperate. The fact is we can use alot of
8716 * space doing the truncate, and we have no earthly idea how much space
8717 * we will use, so we need the truncate reservation to be seperate so it
8718 * doesn't end up using space reserved for updating the inode or
8719 * removing the orphan item. We also need to be able to stop the
8720 * transaction and start a new one, which means we need to be able to
8721 * update the inode several times, and we have no idea of knowing how
8722 * many times that will be, so we can't just reserve 1 item for the
8723 * entirety of the opration, so that has to be done seperately as well.
8724 * Then there is the orphan item, which does indeed need to be held on
8725 * to for the whole operation, and we need nobody to touch this reserved
8726 * space except the orphan code.
8728 * So that leaves us with
8730 * 1) root->orphan_block_rsv - for the orphan deletion.
8731 * 2) rsv - for the truncate reservation, which we will steal from the
8732 * transaction reservation.
8733 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8734 * updating the inode.
8736 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
8739 rsv->size = min_size;
8743 * 1 for the truncate slack space
8744 * 1 for updating the inode.
8746 trans = btrfs_start_transaction(root, 2);
8747 if (IS_ERR(trans)) {
8748 err = PTR_ERR(trans);
8752 /* Migrate the slack space for the truncate to our reserve */
8753 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
8758 * So if we truncate and then write and fsync we normally would just
8759 * write the extents that changed, which is a problem if we need to
8760 * first truncate that entire inode. So set this flag so we write out
8761 * all of the extents in the inode to the sync log so we're completely
8764 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8765 trans->block_rsv = rsv;
8768 ret = btrfs_truncate_inode_items(trans, root, inode,
8770 BTRFS_EXTENT_DATA_KEY);
8771 if (ret != -ENOSPC && ret != -EAGAIN) {
8776 trans->block_rsv = &root->fs_info->trans_block_rsv;
8777 ret = btrfs_update_inode(trans, root, inode);
8783 btrfs_end_transaction(trans, root);
8784 btrfs_btree_balance_dirty(root);
8786 trans = btrfs_start_transaction(root, 2);
8787 if (IS_ERR(trans)) {
8788 ret = err = PTR_ERR(trans);
8793 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
8795 BUG_ON(ret); /* shouldn't happen */
8796 trans->block_rsv = rsv;
8799 if (ret == 0 && inode->i_nlink > 0) {
8800 trans->block_rsv = root->orphan_block_rsv;
8801 ret = btrfs_orphan_del(trans, inode);
8807 trans->block_rsv = &root->fs_info->trans_block_rsv;
8808 ret = btrfs_update_inode(trans, root, inode);
8812 ret = btrfs_end_transaction(trans, root);
8813 btrfs_btree_balance_dirty(root);
8817 btrfs_free_block_rsv(root, rsv);
8826 * create a new subvolume directory/inode (helper for the ioctl).
8828 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8829 struct btrfs_root *new_root,
8830 struct btrfs_root *parent_root,
8833 struct inode *inode;
8837 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8838 new_dirid, new_dirid,
8839 S_IFDIR | (~current_umask() & S_IRWXUGO),
8842 return PTR_ERR(inode);
8843 inode->i_op = &btrfs_dir_inode_operations;
8844 inode->i_fop = &btrfs_dir_file_operations;
8846 set_nlink(inode, 1);
8847 btrfs_i_size_write(inode, 0);
8848 unlock_new_inode(inode);
8850 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8852 btrfs_err(new_root->fs_info,
8853 "error inheriting subvolume %llu properties: %d",
8854 new_root->root_key.objectid, err);
8856 err = btrfs_update_inode(trans, new_root, inode);
8862 struct inode *btrfs_alloc_inode(struct super_block *sb)
8864 struct btrfs_inode *ei;
8865 struct inode *inode;
8867 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
8874 ei->last_sub_trans = 0;
8875 ei->logged_trans = 0;
8876 ei->delalloc_bytes = 0;
8877 ei->defrag_bytes = 0;
8878 ei->disk_i_size = 0;
8881 ei->index_cnt = (u64)-1;
8883 ei->last_unlink_trans = 0;
8884 ei->last_log_commit = 0;
8886 spin_lock_init(&ei->lock);
8887 ei->outstanding_extents = 0;
8888 ei->reserved_extents = 0;
8890 ei->runtime_flags = 0;
8891 ei->force_compress = BTRFS_COMPRESS_NONE;
8893 ei->delayed_node = NULL;
8895 ei->i_otime.tv_sec = 0;
8896 ei->i_otime.tv_nsec = 0;
8898 inode = &ei->vfs_inode;
8899 extent_map_tree_init(&ei->extent_tree);
8900 extent_io_tree_init(&ei->io_tree, &inode->i_data);
8901 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
8902 ei->io_tree.track_uptodate = 1;
8903 ei->io_failure_tree.track_uptodate = 1;
8904 atomic_set(&ei->sync_writers, 0);
8905 mutex_init(&ei->log_mutex);
8906 mutex_init(&ei->delalloc_mutex);
8907 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8908 INIT_LIST_HEAD(&ei->delalloc_inodes);
8909 RB_CLEAR_NODE(&ei->rb_node);
8914 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8915 void btrfs_test_destroy_inode(struct inode *inode)
8917 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8918 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8922 static void btrfs_i_callback(struct rcu_head *head)
8924 struct inode *inode = container_of(head, struct inode, i_rcu);
8925 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8928 void btrfs_destroy_inode(struct inode *inode)
8930 struct btrfs_ordered_extent *ordered;
8931 struct btrfs_root *root = BTRFS_I(inode)->root;
8933 WARN_ON(!hlist_empty(&inode->i_dentry));
8934 WARN_ON(inode->i_data.nrpages);
8935 WARN_ON(BTRFS_I(inode)->outstanding_extents);
8936 WARN_ON(BTRFS_I(inode)->reserved_extents);
8937 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8938 WARN_ON(BTRFS_I(inode)->csum_bytes);
8939 WARN_ON(BTRFS_I(inode)->defrag_bytes);
8942 * This can happen where we create an inode, but somebody else also
8943 * created the same inode and we need to destroy the one we already
8949 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8950 &BTRFS_I(inode)->runtime_flags)) {
8951 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
8953 atomic_dec(&root->orphan_inodes);
8957 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8961 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
8962 ordered->file_offset, ordered->len);
8963 btrfs_remove_ordered_extent(inode, ordered);
8964 btrfs_put_ordered_extent(ordered);
8965 btrfs_put_ordered_extent(ordered);
8968 inode_tree_del(inode);
8969 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8971 call_rcu(&inode->i_rcu, btrfs_i_callback);
8974 int btrfs_drop_inode(struct inode *inode)
8976 struct btrfs_root *root = BTRFS_I(inode)->root;
8981 /* the snap/subvol tree is on deleting */
8982 if (btrfs_root_refs(&root->root_item) == 0)
8985 return generic_drop_inode(inode);
8988 static void init_once(void *foo)
8990 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8992 inode_init_once(&ei->vfs_inode);
8995 void btrfs_destroy_cachep(void)
8998 * Make sure all delayed rcu free inodes are flushed before we
9002 if (btrfs_inode_cachep)
9003 kmem_cache_destroy(btrfs_inode_cachep);
9004 if (btrfs_trans_handle_cachep)
9005 kmem_cache_destroy(btrfs_trans_handle_cachep);
9006 if (btrfs_transaction_cachep)
9007 kmem_cache_destroy(btrfs_transaction_cachep);
9008 if (btrfs_path_cachep)
9009 kmem_cache_destroy(btrfs_path_cachep);
9010 if (btrfs_free_space_cachep)
9011 kmem_cache_destroy(btrfs_free_space_cachep);
9012 if (btrfs_delalloc_work_cachep)
9013 kmem_cache_destroy(btrfs_delalloc_work_cachep);
9016 int btrfs_init_cachep(void)
9018 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9019 sizeof(struct btrfs_inode), 0,
9020 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
9021 if (!btrfs_inode_cachep)
9024 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9025 sizeof(struct btrfs_trans_handle), 0,
9026 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9027 if (!btrfs_trans_handle_cachep)
9030 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9031 sizeof(struct btrfs_transaction), 0,
9032 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9033 if (!btrfs_transaction_cachep)
9036 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9037 sizeof(struct btrfs_path), 0,
9038 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9039 if (!btrfs_path_cachep)
9042 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9043 sizeof(struct btrfs_free_space), 0,
9044 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9045 if (!btrfs_free_space_cachep)
9048 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
9049 sizeof(struct btrfs_delalloc_work), 0,
9050 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
9052 if (!btrfs_delalloc_work_cachep)
9057 btrfs_destroy_cachep();
9061 static int btrfs_getattr(struct vfsmount *mnt,
9062 struct dentry *dentry, struct kstat *stat)
9065 struct inode *inode = d_inode(dentry);
9066 u32 blocksize = inode->i_sb->s_blocksize;
9068 generic_fillattr(inode, stat);
9069 stat->dev = BTRFS_I(inode)->root->anon_dev;
9070 stat->blksize = PAGE_CACHE_SIZE;
9072 spin_lock(&BTRFS_I(inode)->lock);
9073 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9074 spin_unlock(&BTRFS_I(inode)->lock);
9075 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9076 ALIGN(delalloc_bytes, blocksize)) >> 9;
9080 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9081 struct inode *new_dir, struct dentry *new_dentry)
9083 struct btrfs_trans_handle *trans;
9084 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9085 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9086 struct inode *new_inode = d_inode(new_dentry);
9087 struct inode *old_inode = d_inode(old_dentry);
9088 struct timespec ctime = CURRENT_TIME;
9092 u64 old_ino = btrfs_ino(old_inode);
9094 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9097 /* we only allow rename subvolume link between subvolumes */
9098 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9101 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9102 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
9105 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9106 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9110 /* check for collisions, even if the name isn't there */
9111 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9112 new_dentry->d_name.name,
9113 new_dentry->d_name.len);
9116 if (ret == -EEXIST) {
9118 * eexist without a new_inode */
9119 if (WARN_ON(!new_inode)) {
9123 /* maybe -EOVERFLOW */
9130 * we're using rename to replace one file with another. Start IO on it
9131 * now so we don't add too much work to the end of the transaction
9133 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9134 filemap_flush(old_inode->i_mapping);
9136 /* close the racy window with snapshot create/destroy ioctl */
9137 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9138 down_read(&root->fs_info->subvol_sem);
9140 * We want to reserve the absolute worst case amount of items. So if
9141 * both inodes are subvols and we need to unlink them then that would
9142 * require 4 item modifications, but if they are both normal inodes it
9143 * would require 5 item modifications, so we'll assume their normal
9144 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9145 * should cover the worst case number of items we'll modify.
9147 trans = btrfs_start_transaction(root, 11);
9148 if (IS_ERR(trans)) {
9149 ret = PTR_ERR(trans);
9154 btrfs_record_root_in_trans(trans, dest);
9156 ret = btrfs_set_inode_index(new_dir, &index);
9160 BTRFS_I(old_inode)->dir_index = 0ULL;
9161 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9162 /* force full log commit if subvolume involved. */
9163 btrfs_set_log_full_commit(root->fs_info, trans);
9165 ret = btrfs_insert_inode_ref(trans, dest,
9166 new_dentry->d_name.name,
9167 new_dentry->d_name.len,
9169 btrfs_ino(new_dir), index);
9173 * this is an ugly little race, but the rename is required
9174 * to make sure that if we crash, the inode is either at the
9175 * old name or the new one. pinning the log transaction lets
9176 * us make sure we don't allow a log commit to come in after
9177 * we unlink the name but before we add the new name back in.
9179 btrfs_pin_log_trans(root);
9182 inode_inc_iversion(old_dir);
9183 inode_inc_iversion(new_dir);
9184 inode_inc_iversion(old_inode);
9185 old_dir->i_ctime = old_dir->i_mtime = ctime;
9186 new_dir->i_ctime = new_dir->i_mtime = ctime;
9187 old_inode->i_ctime = ctime;
9189 if (old_dentry->d_parent != new_dentry->d_parent)
9190 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9192 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9193 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9194 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9195 old_dentry->d_name.name,
9196 old_dentry->d_name.len);
9198 ret = __btrfs_unlink_inode(trans, root, old_dir,
9199 d_inode(old_dentry),
9200 old_dentry->d_name.name,
9201 old_dentry->d_name.len);
9203 ret = btrfs_update_inode(trans, root, old_inode);
9206 btrfs_abort_transaction(trans, root, ret);
9211 inode_inc_iversion(new_inode);
9212 new_inode->i_ctime = CURRENT_TIME;
9213 if (unlikely(btrfs_ino(new_inode) ==
9214 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9215 root_objectid = BTRFS_I(new_inode)->location.objectid;
9216 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9218 new_dentry->d_name.name,
9219 new_dentry->d_name.len);
9220 BUG_ON(new_inode->i_nlink == 0);
9222 ret = btrfs_unlink_inode(trans, dest, new_dir,
9223 d_inode(new_dentry),
9224 new_dentry->d_name.name,
9225 new_dentry->d_name.len);
9227 if (!ret && new_inode->i_nlink == 0)
9228 ret = btrfs_orphan_add(trans, d_inode(new_dentry));
9230 btrfs_abort_transaction(trans, root, ret);
9235 ret = btrfs_add_link(trans, new_dir, old_inode,
9236 new_dentry->d_name.name,
9237 new_dentry->d_name.len, 0, index);
9239 btrfs_abort_transaction(trans, root, ret);
9243 if (old_inode->i_nlink == 1)
9244 BTRFS_I(old_inode)->dir_index = index;
9246 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
9247 struct dentry *parent = new_dentry->d_parent;
9248 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9249 btrfs_end_log_trans(root);
9252 btrfs_end_transaction(trans, root);
9254 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9255 up_read(&root->fs_info->subvol_sem);
9260 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9261 struct inode *new_dir, struct dentry *new_dentry,
9264 if (flags & ~RENAME_NOREPLACE)
9267 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
9270 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9272 struct btrfs_delalloc_work *delalloc_work;
9273 struct inode *inode;
9275 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9277 inode = delalloc_work->inode;
9278 if (delalloc_work->wait) {
9279 btrfs_wait_ordered_range(inode, 0, (u64)-1);
9281 filemap_flush(inode->i_mapping);
9282 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9283 &BTRFS_I(inode)->runtime_flags))
9284 filemap_flush(inode->i_mapping);
9287 if (delalloc_work->delay_iput)
9288 btrfs_add_delayed_iput(inode);
9291 complete(&delalloc_work->completion);
9294 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9295 int wait, int delay_iput)
9297 struct btrfs_delalloc_work *work;
9299 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
9303 init_completion(&work->completion);
9304 INIT_LIST_HEAD(&work->list);
9305 work->inode = inode;
9307 work->delay_iput = delay_iput;
9308 WARN_ON_ONCE(!inode);
9309 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9310 btrfs_run_delalloc_work, NULL, NULL);
9315 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9317 wait_for_completion(&work->completion);
9318 kmem_cache_free(btrfs_delalloc_work_cachep, work);
9322 * some fairly slow code that needs optimization. This walks the list
9323 * of all the inodes with pending delalloc and forces them to disk.
9325 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9328 struct btrfs_inode *binode;
9329 struct inode *inode;
9330 struct btrfs_delalloc_work *work, *next;
9331 struct list_head works;
9332 struct list_head splice;
9335 INIT_LIST_HEAD(&works);
9336 INIT_LIST_HEAD(&splice);
9338 mutex_lock(&root->delalloc_mutex);
9339 spin_lock(&root->delalloc_lock);
9340 list_splice_init(&root->delalloc_inodes, &splice);
9341 while (!list_empty(&splice)) {
9342 binode = list_entry(splice.next, struct btrfs_inode,
9345 list_move_tail(&binode->delalloc_inodes,
9346 &root->delalloc_inodes);
9347 inode = igrab(&binode->vfs_inode);
9349 cond_resched_lock(&root->delalloc_lock);
9352 spin_unlock(&root->delalloc_lock);
9354 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
9357 btrfs_add_delayed_iput(inode);
9363 list_add_tail(&work->list, &works);
9364 btrfs_queue_work(root->fs_info->flush_workers,
9367 if (nr != -1 && ret >= nr)
9370 spin_lock(&root->delalloc_lock);
9372 spin_unlock(&root->delalloc_lock);
9375 list_for_each_entry_safe(work, next, &works, list) {
9376 list_del_init(&work->list);
9377 btrfs_wait_and_free_delalloc_work(work);
9380 if (!list_empty_careful(&splice)) {
9381 spin_lock(&root->delalloc_lock);
9382 list_splice_tail(&splice, &root->delalloc_inodes);
9383 spin_unlock(&root->delalloc_lock);
9385 mutex_unlock(&root->delalloc_mutex);
9389 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9393 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
9396 ret = __start_delalloc_inodes(root, delay_iput, -1);
9400 * the filemap_flush will queue IO into the worker threads, but
9401 * we have to make sure the IO is actually started and that
9402 * ordered extents get created before we return
9404 atomic_inc(&root->fs_info->async_submit_draining);
9405 while (atomic_read(&root->fs_info->nr_async_submits) ||
9406 atomic_read(&root->fs_info->async_delalloc_pages)) {
9407 wait_event(root->fs_info->async_submit_wait,
9408 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9409 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
9411 atomic_dec(&root->fs_info->async_submit_draining);
9415 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9418 struct btrfs_root *root;
9419 struct list_head splice;
9422 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9425 INIT_LIST_HEAD(&splice);
9427 mutex_lock(&fs_info->delalloc_root_mutex);
9428 spin_lock(&fs_info->delalloc_root_lock);
9429 list_splice_init(&fs_info->delalloc_roots, &splice);
9430 while (!list_empty(&splice) && nr) {
9431 root = list_first_entry(&splice, struct btrfs_root,
9433 root = btrfs_grab_fs_root(root);
9435 list_move_tail(&root->delalloc_root,
9436 &fs_info->delalloc_roots);
9437 spin_unlock(&fs_info->delalloc_root_lock);
9439 ret = __start_delalloc_inodes(root, delay_iput, nr);
9440 btrfs_put_fs_root(root);
9448 spin_lock(&fs_info->delalloc_root_lock);
9450 spin_unlock(&fs_info->delalloc_root_lock);
9453 atomic_inc(&fs_info->async_submit_draining);
9454 while (atomic_read(&fs_info->nr_async_submits) ||
9455 atomic_read(&fs_info->async_delalloc_pages)) {
9456 wait_event(fs_info->async_submit_wait,
9457 (atomic_read(&fs_info->nr_async_submits) == 0 &&
9458 atomic_read(&fs_info->async_delalloc_pages) == 0));
9460 atomic_dec(&fs_info->async_submit_draining);
9462 if (!list_empty_careful(&splice)) {
9463 spin_lock(&fs_info->delalloc_root_lock);
9464 list_splice_tail(&splice, &fs_info->delalloc_roots);
9465 spin_unlock(&fs_info->delalloc_root_lock);
9467 mutex_unlock(&fs_info->delalloc_root_mutex);
9471 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9472 const char *symname)
9474 struct btrfs_trans_handle *trans;
9475 struct btrfs_root *root = BTRFS_I(dir)->root;
9476 struct btrfs_path *path;
9477 struct btrfs_key key;
9478 struct inode *inode = NULL;
9486 struct btrfs_file_extent_item *ei;
9487 struct extent_buffer *leaf;
9489 name_len = strlen(symname);
9490 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9491 return -ENAMETOOLONG;
9494 * 2 items for inode item and ref
9495 * 2 items for dir items
9496 * 1 item for xattr if selinux is on
9498 trans = btrfs_start_transaction(root, 5);
9500 return PTR_ERR(trans);
9502 err = btrfs_find_free_ino(root, &objectid);
9506 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9507 dentry->d_name.len, btrfs_ino(dir), objectid,
9508 S_IFLNK|S_IRWXUGO, &index);
9509 if (IS_ERR(inode)) {
9510 err = PTR_ERR(inode);
9515 * If the active LSM wants to access the inode during
9516 * d_instantiate it needs these. Smack checks to see
9517 * if the filesystem supports xattrs by looking at the
9520 inode->i_fop = &btrfs_file_operations;
9521 inode->i_op = &btrfs_file_inode_operations;
9522 inode->i_mapping->a_ops = &btrfs_aops;
9523 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9525 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9527 goto out_unlock_inode;
9529 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
9531 goto out_unlock_inode;
9533 path = btrfs_alloc_path();
9536 goto out_unlock_inode;
9538 key.objectid = btrfs_ino(inode);
9540 key.type = BTRFS_EXTENT_DATA_KEY;
9541 datasize = btrfs_file_extent_calc_inline_size(name_len);
9542 err = btrfs_insert_empty_item(trans, root, path, &key,
9545 btrfs_free_path(path);
9546 goto out_unlock_inode;
9548 leaf = path->nodes[0];
9549 ei = btrfs_item_ptr(leaf, path->slots[0],
9550 struct btrfs_file_extent_item);
9551 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9552 btrfs_set_file_extent_type(leaf, ei,
9553 BTRFS_FILE_EXTENT_INLINE);
9554 btrfs_set_file_extent_encryption(leaf, ei, 0);
9555 btrfs_set_file_extent_compression(leaf, ei, 0);
9556 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9557 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9559 ptr = btrfs_file_extent_inline_start(ei);
9560 write_extent_buffer(leaf, symname, ptr, name_len);
9561 btrfs_mark_buffer_dirty(leaf);
9562 btrfs_free_path(path);
9564 inode->i_op = &btrfs_symlink_inode_operations;
9565 inode->i_mapping->a_ops = &btrfs_symlink_aops;
9566 inode_set_bytes(inode, name_len);
9567 btrfs_i_size_write(inode, name_len);
9568 err = btrfs_update_inode(trans, root, inode);
9571 goto out_unlock_inode;
9574 unlock_new_inode(inode);
9575 d_instantiate(dentry, inode);
9578 btrfs_end_transaction(trans, root);
9580 inode_dec_link_count(inode);
9583 btrfs_btree_balance_dirty(root);
9588 unlock_new_inode(inode);
9592 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9593 u64 start, u64 num_bytes, u64 min_size,
9594 loff_t actual_len, u64 *alloc_hint,
9595 struct btrfs_trans_handle *trans)
9597 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9598 struct extent_map *em;
9599 struct btrfs_root *root = BTRFS_I(inode)->root;
9600 struct btrfs_key ins;
9601 u64 cur_offset = start;
9605 bool own_trans = true;
9609 while (num_bytes > 0) {
9611 trans = btrfs_start_transaction(root, 3);
9612 if (IS_ERR(trans)) {
9613 ret = PTR_ERR(trans);
9618 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
9619 cur_bytes = max(cur_bytes, min_size);
9620 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
9621 *alloc_hint, &ins, 1, 0);
9624 btrfs_end_transaction(trans, root);
9628 ret = insert_reserved_file_extent(trans, inode,
9629 cur_offset, ins.objectid,
9630 ins.offset, ins.offset,
9631 ins.offset, 0, 0, 0,
9632 BTRFS_FILE_EXTENT_PREALLOC);
9634 btrfs_free_reserved_extent(root, ins.objectid,
9636 btrfs_abort_transaction(trans, root, ret);
9638 btrfs_end_transaction(trans, root);
9642 btrfs_drop_extent_cache(inode, cur_offset,
9643 cur_offset + ins.offset -1, 0);
9645 em = alloc_extent_map();
9647 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9648 &BTRFS_I(inode)->runtime_flags);
9652 em->start = cur_offset;
9653 em->orig_start = cur_offset;
9654 em->len = ins.offset;
9655 em->block_start = ins.objectid;
9656 em->block_len = ins.offset;
9657 em->orig_block_len = ins.offset;
9658 em->ram_bytes = ins.offset;
9659 em->bdev = root->fs_info->fs_devices->latest_bdev;
9660 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9661 em->generation = trans->transid;
9664 write_lock(&em_tree->lock);
9665 ret = add_extent_mapping(em_tree, em, 1);
9666 write_unlock(&em_tree->lock);
9669 btrfs_drop_extent_cache(inode, cur_offset,
9670 cur_offset + ins.offset - 1,
9673 free_extent_map(em);
9675 num_bytes -= ins.offset;
9676 cur_offset += ins.offset;
9677 *alloc_hint = ins.objectid + ins.offset;
9679 inode_inc_iversion(inode);
9680 inode->i_ctime = CURRENT_TIME;
9681 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9682 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9683 (actual_len > inode->i_size) &&
9684 (cur_offset > inode->i_size)) {
9685 if (cur_offset > actual_len)
9686 i_size = actual_len;
9688 i_size = cur_offset;
9689 i_size_write(inode, i_size);
9690 btrfs_ordered_update_i_size(inode, i_size, NULL);
9693 ret = btrfs_update_inode(trans, root, inode);
9696 btrfs_abort_transaction(trans, root, ret);
9698 btrfs_end_transaction(trans, root);
9703 btrfs_end_transaction(trans, root);
9708 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9709 u64 start, u64 num_bytes, u64 min_size,
9710 loff_t actual_len, u64 *alloc_hint)
9712 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9713 min_size, actual_len, alloc_hint,
9717 int btrfs_prealloc_file_range_trans(struct inode *inode,
9718 struct btrfs_trans_handle *trans, int mode,
9719 u64 start, u64 num_bytes, u64 min_size,
9720 loff_t actual_len, u64 *alloc_hint)
9722 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9723 min_size, actual_len, alloc_hint, trans);
9726 static int btrfs_set_page_dirty(struct page *page)
9728 return __set_page_dirty_nobuffers(page);
9731 static int btrfs_permission(struct inode *inode, int mask)
9733 struct btrfs_root *root = BTRFS_I(inode)->root;
9734 umode_t mode = inode->i_mode;
9736 if (mask & MAY_WRITE &&
9737 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9738 if (btrfs_root_readonly(root))
9740 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9743 return generic_permission(inode, mask);
9746 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9748 struct btrfs_trans_handle *trans;
9749 struct btrfs_root *root = BTRFS_I(dir)->root;
9750 struct inode *inode = NULL;
9756 * 5 units required for adding orphan entry
9758 trans = btrfs_start_transaction(root, 5);
9760 return PTR_ERR(trans);
9762 ret = btrfs_find_free_ino(root, &objectid);
9766 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9767 btrfs_ino(dir), objectid, mode, &index);
9768 if (IS_ERR(inode)) {
9769 ret = PTR_ERR(inode);
9774 inode->i_fop = &btrfs_file_operations;
9775 inode->i_op = &btrfs_file_inode_operations;
9777 inode->i_mapping->a_ops = &btrfs_aops;
9778 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9780 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9784 ret = btrfs_update_inode(trans, root, inode);
9787 ret = btrfs_orphan_add(trans, inode);
9792 * We set number of links to 0 in btrfs_new_inode(), and here we set
9793 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9796 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9798 set_nlink(inode, 1);
9799 unlock_new_inode(inode);
9800 d_tmpfile(dentry, inode);
9801 mark_inode_dirty(inode);
9804 btrfs_end_transaction(trans, root);
9807 btrfs_balance_delayed_items(root);
9808 btrfs_btree_balance_dirty(root);
9812 unlock_new_inode(inode);
9817 /* Inspired by filemap_check_errors() */
9818 int btrfs_inode_check_errors(struct inode *inode)
9822 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
9823 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
9825 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
9826 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
9832 static const struct inode_operations btrfs_dir_inode_operations = {
9833 .getattr = btrfs_getattr,
9834 .lookup = btrfs_lookup,
9835 .create = btrfs_create,
9836 .unlink = btrfs_unlink,
9838 .mkdir = btrfs_mkdir,
9839 .rmdir = btrfs_rmdir,
9840 .rename2 = btrfs_rename2,
9841 .symlink = btrfs_symlink,
9842 .setattr = btrfs_setattr,
9843 .mknod = btrfs_mknod,
9844 .setxattr = btrfs_setxattr,
9845 .getxattr = btrfs_getxattr,
9846 .listxattr = btrfs_listxattr,
9847 .removexattr = btrfs_removexattr,
9848 .permission = btrfs_permission,
9849 .get_acl = btrfs_get_acl,
9850 .set_acl = btrfs_set_acl,
9851 .update_time = btrfs_update_time,
9852 .tmpfile = btrfs_tmpfile,
9854 static const struct inode_operations btrfs_dir_ro_inode_operations = {
9855 .lookup = btrfs_lookup,
9856 .permission = btrfs_permission,
9857 .get_acl = btrfs_get_acl,
9858 .set_acl = btrfs_set_acl,
9859 .update_time = btrfs_update_time,
9862 static const struct file_operations btrfs_dir_file_operations = {
9863 .llseek = generic_file_llseek,
9864 .read = generic_read_dir,
9865 .iterate = btrfs_real_readdir,
9866 .unlocked_ioctl = btrfs_ioctl,
9867 #ifdef CONFIG_COMPAT
9868 .compat_ioctl = btrfs_ioctl,
9870 .release = btrfs_release_file,
9871 .fsync = btrfs_sync_file,
9874 static struct extent_io_ops btrfs_extent_io_ops = {
9875 .fill_delalloc = run_delalloc_range,
9876 .submit_bio_hook = btrfs_submit_bio_hook,
9877 .merge_bio_hook = btrfs_merge_bio_hook,
9878 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
9879 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
9880 .writepage_start_hook = btrfs_writepage_start_hook,
9881 .set_bit_hook = btrfs_set_bit_hook,
9882 .clear_bit_hook = btrfs_clear_bit_hook,
9883 .merge_extent_hook = btrfs_merge_extent_hook,
9884 .split_extent_hook = btrfs_split_extent_hook,
9888 * btrfs doesn't support the bmap operation because swapfiles
9889 * use bmap to make a mapping of extents in the file. They assume
9890 * these extents won't change over the life of the file and they
9891 * use the bmap result to do IO directly to the drive.
9893 * the btrfs bmap call would return logical addresses that aren't
9894 * suitable for IO and they also will change frequently as COW
9895 * operations happen. So, swapfile + btrfs == corruption.
9897 * For now we're avoiding this by dropping bmap.
9899 static const struct address_space_operations btrfs_aops = {
9900 .readpage = btrfs_readpage,
9901 .writepage = btrfs_writepage,
9902 .writepages = btrfs_writepages,
9903 .readpages = btrfs_readpages,
9904 .direct_IO = btrfs_direct_IO,
9905 .invalidatepage = btrfs_invalidatepage,
9906 .releasepage = btrfs_releasepage,
9907 .set_page_dirty = btrfs_set_page_dirty,
9908 .error_remove_page = generic_error_remove_page,
9911 static const struct address_space_operations btrfs_symlink_aops = {
9912 .readpage = btrfs_readpage,
9913 .writepage = btrfs_writepage,
9914 .invalidatepage = btrfs_invalidatepage,
9915 .releasepage = btrfs_releasepage,
9918 static const struct inode_operations btrfs_file_inode_operations = {
9919 .getattr = btrfs_getattr,
9920 .setattr = btrfs_setattr,
9921 .setxattr = btrfs_setxattr,
9922 .getxattr = btrfs_getxattr,
9923 .listxattr = btrfs_listxattr,
9924 .removexattr = btrfs_removexattr,
9925 .permission = btrfs_permission,
9926 .fiemap = btrfs_fiemap,
9927 .get_acl = btrfs_get_acl,
9928 .set_acl = btrfs_set_acl,
9929 .update_time = btrfs_update_time,
9931 static const struct inode_operations btrfs_special_inode_operations = {
9932 .getattr = btrfs_getattr,
9933 .setattr = btrfs_setattr,
9934 .permission = btrfs_permission,
9935 .setxattr = btrfs_setxattr,
9936 .getxattr = btrfs_getxattr,
9937 .listxattr = btrfs_listxattr,
9938 .removexattr = btrfs_removexattr,
9939 .get_acl = btrfs_get_acl,
9940 .set_acl = btrfs_set_acl,
9941 .update_time = btrfs_update_time,
9943 static const struct inode_operations btrfs_symlink_inode_operations = {
9944 .readlink = generic_readlink,
9945 .follow_link = page_follow_link_light,
9946 .put_link = page_put_link,
9947 .getattr = btrfs_getattr,
9948 .setattr = btrfs_setattr,
9949 .permission = btrfs_permission,
9950 .setxattr = btrfs_setxattr,
9951 .getxattr = btrfs_getxattr,
9952 .listxattr = btrfs_listxattr,
9953 .removexattr = btrfs_removexattr,
9954 .update_time = btrfs_update_time,
9957 const struct dentry_operations btrfs_dentry_operations = {
9958 .d_delete = btrfs_dentry_delete,
9959 .d_release = btrfs_dentry_release,