2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/aio.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/xattr.h>
38 #include <linux/posix_acl.h>
39 #include <linux/falloc.h>
40 #include <linux/slab.h>
41 #include <linux/ratelimit.h>
42 #include <linux/mount.h>
43 #include <linux/btrfs.h>
44 #include <linux/blkdev.h>
45 #include <linux/posix_acl_xattr.h>
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "print-tree.h"
51 #include "ordered-data.h"
55 #include "compression.h"
57 #include "free-space-cache.h"
58 #include "inode-map.h"
63 struct btrfs_iget_args {
64 struct btrfs_key *location;
65 struct btrfs_root *root;
68 static const struct inode_operations btrfs_dir_inode_operations;
69 static const struct inode_operations btrfs_symlink_inode_operations;
70 static const struct inode_operations btrfs_dir_ro_inode_operations;
71 static const struct inode_operations btrfs_special_inode_operations;
72 static const struct inode_operations btrfs_file_inode_operations;
73 static const struct address_space_operations btrfs_aops;
74 static const struct address_space_operations btrfs_symlink_aops;
75 static const struct file_operations btrfs_dir_file_operations;
76 static struct extent_io_ops btrfs_extent_io_ops;
78 static struct kmem_cache *btrfs_inode_cachep;
79 static struct kmem_cache *btrfs_delalloc_work_cachep;
80 struct kmem_cache *btrfs_trans_handle_cachep;
81 struct kmem_cache *btrfs_transaction_cachep;
82 struct kmem_cache *btrfs_path_cachep;
83 struct kmem_cache *btrfs_free_space_cachep;
86 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
87 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
88 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
89 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
90 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
91 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
92 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
93 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
96 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
97 static int btrfs_truncate(struct inode *inode);
98 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
99 static noinline int cow_file_range(struct inode *inode,
100 struct page *locked_page,
101 u64 start, u64 end, int *page_started,
102 unsigned long *nr_written, int unlock);
103 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
104 u64 len, u64 orig_start,
105 u64 block_start, u64 block_len,
106 u64 orig_block_len, u64 ram_bytes,
109 static int btrfs_dirty_inode(struct inode *inode);
111 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
112 struct inode *inode, struct inode *dir,
113 const struct qstr *qstr)
117 err = btrfs_init_acl(trans, inode, dir);
119 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
124 * this does all the hard work for inserting an inline extent into
125 * the btree. The caller should have done a btrfs_drop_extents so that
126 * no overlapping inline items exist in the btree
128 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
129 struct btrfs_path *path, int extent_inserted,
130 struct btrfs_root *root, struct inode *inode,
131 u64 start, size_t size, size_t compressed_size,
133 struct page **compressed_pages)
135 struct extent_buffer *leaf;
136 struct page *page = NULL;
139 struct btrfs_file_extent_item *ei;
142 size_t cur_size = size;
143 unsigned long offset;
145 if (compressed_size && compressed_pages)
146 cur_size = compressed_size;
148 inode_add_bytes(inode, size);
150 if (!extent_inserted) {
151 struct btrfs_key key;
154 key.objectid = btrfs_ino(inode);
156 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
158 datasize = btrfs_file_extent_calc_inline_size(cur_size);
159 path->leave_spinning = 1;
160 ret = btrfs_insert_empty_item(trans, root, path, &key,
167 leaf = path->nodes[0];
168 ei = btrfs_item_ptr(leaf, path->slots[0],
169 struct btrfs_file_extent_item);
170 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
171 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
172 btrfs_set_file_extent_encryption(leaf, ei, 0);
173 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
174 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
175 ptr = btrfs_file_extent_inline_start(ei);
177 if (compress_type != BTRFS_COMPRESS_NONE) {
180 while (compressed_size > 0) {
181 cpage = compressed_pages[i];
182 cur_size = min_t(unsigned long, compressed_size,
185 kaddr = kmap_atomic(cpage);
186 write_extent_buffer(leaf, kaddr, ptr, cur_size);
187 kunmap_atomic(kaddr);
191 compressed_size -= cur_size;
193 btrfs_set_file_extent_compression(leaf, ei,
196 page = find_get_page(inode->i_mapping,
197 start >> PAGE_CACHE_SHIFT);
198 btrfs_set_file_extent_compression(leaf, ei, 0);
199 kaddr = kmap_atomic(page);
200 offset = start & (PAGE_CACHE_SIZE - 1);
201 write_extent_buffer(leaf, kaddr + offset, ptr, size);
202 kunmap_atomic(kaddr);
203 page_cache_release(page);
205 btrfs_mark_buffer_dirty(leaf);
206 btrfs_release_path(path);
209 * we're an inline extent, so nobody can
210 * extend the file past i_size without locking
211 * a page we already have locked.
213 * We must do any isize and inode updates
214 * before we unlock the pages. Otherwise we
215 * could end up racing with unlink.
217 BTRFS_I(inode)->disk_i_size = inode->i_size;
218 ret = btrfs_update_inode(trans, root, inode);
227 * conditionally insert an inline extent into the file. This
228 * does the checks required to make sure the data is small enough
229 * to fit as an inline extent.
231 static noinline int cow_file_range_inline(struct btrfs_root *root,
232 struct inode *inode, u64 start,
233 u64 end, size_t compressed_size,
235 struct page **compressed_pages)
237 struct btrfs_trans_handle *trans;
238 u64 isize = i_size_read(inode);
239 u64 actual_end = min(end + 1, isize);
240 u64 inline_len = actual_end - start;
241 u64 aligned_end = ALIGN(end, root->sectorsize);
242 u64 data_len = inline_len;
244 struct btrfs_path *path;
245 int extent_inserted = 0;
246 u32 extent_item_size;
249 data_len = compressed_size;
252 actual_end >= PAGE_CACHE_SIZE ||
253 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
255 (actual_end & (root->sectorsize - 1)) == 0) ||
257 data_len > root->fs_info->max_inline) {
261 path = btrfs_alloc_path();
265 trans = btrfs_join_transaction(root);
267 btrfs_free_path(path);
268 return PTR_ERR(trans);
270 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
272 if (compressed_size && compressed_pages)
273 extent_item_size = btrfs_file_extent_calc_inline_size(
276 extent_item_size = btrfs_file_extent_calc_inline_size(
279 ret = __btrfs_drop_extents(trans, root, inode, path,
280 start, aligned_end, NULL,
281 1, 1, extent_item_size, &extent_inserted);
283 btrfs_abort_transaction(trans, root, ret);
287 if (isize > actual_end)
288 inline_len = min_t(u64, isize, actual_end);
289 ret = insert_inline_extent(trans, path, extent_inserted,
291 inline_len, compressed_size,
292 compress_type, compressed_pages);
293 if (ret && ret != -ENOSPC) {
294 btrfs_abort_transaction(trans, root, ret);
296 } else if (ret == -ENOSPC) {
301 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
302 btrfs_delalloc_release_metadata(inode, end + 1 - start);
303 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
305 btrfs_free_path(path);
306 btrfs_end_transaction(trans, root);
310 struct async_extent {
315 unsigned long nr_pages;
317 struct list_head list;
322 struct btrfs_root *root;
323 struct page *locked_page;
326 struct list_head extents;
327 struct btrfs_work work;
330 static noinline int add_async_extent(struct async_cow *cow,
331 u64 start, u64 ram_size,
334 unsigned long nr_pages,
337 struct async_extent *async_extent;
339 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
340 BUG_ON(!async_extent); /* -ENOMEM */
341 async_extent->start = start;
342 async_extent->ram_size = ram_size;
343 async_extent->compressed_size = compressed_size;
344 async_extent->pages = pages;
345 async_extent->nr_pages = nr_pages;
346 async_extent->compress_type = compress_type;
347 list_add_tail(&async_extent->list, &cow->extents);
352 * we create compressed extents in two phases. The first
353 * phase compresses a range of pages that have already been
354 * locked (both pages and state bits are locked).
356 * This is done inside an ordered work queue, and the compression
357 * is spread across many cpus. The actual IO submission is step
358 * two, and the ordered work queue takes care of making sure that
359 * happens in the same order things were put onto the queue by
360 * writepages and friends.
362 * If this code finds it can't get good compression, it puts an
363 * entry onto the work queue to write the uncompressed bytes. This
364 * makes sure that both compressed inodes and uncompressed inodes
365 * are written in the same order that the flusher thread sent them
368 static noinline int compress_file_range(struct inode *inode,
369 struct page *locked_page,
371 struct async_cow *async_cow,
374 struct btrfs_root *root = BTRFS_I(inode)->root;
376 u64 blocksize = root->sectorsize;
378 u64 isize = i_size_read(inode);
380 struct page **pages = NULL;
381 unsigned long nr_pages;
382 unsigned long nr_pages_ret = 0;
383 unsigned long total_compressed = 0;
384 unsigned long total_in = 0;
385 unsigned long max_compressed = 128 * 1024;
386 unsigned long max_uncompressed = 128 * 1024;
389 int compress_type = root->fs_info->compress_type;
392 /* if this is a small write inside eof, kick off a defrag */
393 if ((end - start + 1) < 16 * 1024 &&
394 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
395 btrfs_add_inode_defrag(NULL, inode);
397 actual_end = min_t(u64, isize, end + 1);
400 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
401 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
404 * we don't want to send crud past the end of i_size through
405 * compression, that's just a waste of CPU time. So, if the
406 * end of the file is before the start of our current
407 * requested range of bytes, we bail out to the uncompressed
408 * cleanup code that can deal with all of this.
410 * It isn't really the fastest way to fix things, but this is a
411 * very uncommon corner.
413 if (actual_end <= start)
414 goto cleanup_and_bail_uncompressed;
416 total_compressed = actual_end - start;
418 /* we want to make sure that amount of ram required to uncompress
419 * an extent is reasonable, so we limit the total size in ram
420 * of a compressed extent to 128k. This is a crucial number
421 * because it also controls how easily we can spread reads across
422 * cpus for decompression.
424 * We also want to make sure the amount of IO required to do
425 * a random read is reasonably small, so we limit the size of
426 * a compressed extent to 128k.
428 total_compressed = min(total_compressed, max_uncompressed);
429 num_bytes = ALIGN(end - start + 1, blocksize);
430 num_bytes = max(blocksize, num_bytes);
435 * we do compression for mount -o compress and when the
436 * inode has not been flagged as nocompress. This flag can
437 * change at any time if we discover bad compression ratios.
439 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
440 (btrfs_test_opt(root, COMPRESS) ||
441 (BTRFS_I(inode)->force_compress) ||
442 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
444 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
446 /* just bail out to the uncompressed code */
450 if (BTRFS_I(inode)->force_compress)
451 compress_type = BTRFS_I(inode)->force_compress;
454 * we need to call clear_page_dirty_for_io on each
455 * page in the range. Otherwise applications with the file
456 * mmap'd can wander in and change the page contents while
457 * we are compressing them.
459 * If the compression fails for any reason, we set the pages
460 * dirty again later on.
462 extent_range_clear_dirty_for_io(inode, start, end);
464 ret = btrfs_compress_pages(compress_type,
465 inode->i_mapping, start,
466 total_compressed, pages,
467 nr_pages, &nr_pages_ret,
473 unsigned long offset = total_compressed &
474 (PAGE_CACHE_SIZE - 1);
475 struct page *page = pages[nr_pages_ret - 1];
478 /* zero the tail end of the last page, we might be
479 * sending it down to disk
482 kaddr = kmap_atomic(page);
483 memset(kaddr + offset, 0,
484 PAGE_CACHE_SIZE - offset);
485 kunmap_atomic(kaddr);
492 /* lets try to make an inline extent */
493 if (ret || total_in < (actual_end - start)) {
494 /* we didn't compress the entire range, try
495 * to make an uncompressed inline extent.
497 ret = cow_file_range_inline(root, inode, start, end,
500 /* try making a compressed inline extent */
501 ret = cow_file_range_inline(root, inode, start, end,
503 compress_type, pages);
506 unsigned long clear_flags = EXTENT_DELALLOC |
508 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
511 * inline extent creation worked or returned error,
512 * we don't need to create any more async work items.
513 * Unlock and free up our temp pages.
515 extent_clear_unlock_delalloc(inode, start, end, NULL,
516 clear_flags, PAGE_UNLOCK |
526 * we aren't doing an inline extent round the compressed size
527 * up to a block size boundary so the allocator does sane
530 total_compressed = ALIGN(total_compressed, blocksize);
533 * one last check to make sure the compression is really a
534 * win, compare the page count read with the blocks on disk
536 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
537 if (total_compressed >= total_in) {
540 num_bytes = total_in;
543 if (!will_compress && pages) {
545 * the compression code ran but failed to make things smaller,
546 * free any pages it allocated and our page pointer array
548 for (i = 0; i < nr_pages_ret; i++) {
549 WARN_ON(pages[i]->mapping);
550 page_cache_release(pages[i]);
554 total_compressed = 0;
557 /* flag the file so we don't compress in the future */
558 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
559 !(BTRFS_I(inode)->force_compress)) {
560 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
566 /* the async work queues will take care of doing actual
567 * allocation on disk for these compressed pages,
568 * and will submit them to the elevator.
570 add_async_extent(async_cow, start, num_bytes,
571 total_compressed, pages, nr_pages_ret,
574 if (start + num_bytes < end) {
581 cleanup_and_bail_uncompressed:
583 * No compression, but we still need to write the pages in
584 * the file we've been given so far. redirty the locked
585 * page if it corresponds to our extent and set things up
586 * for the async work queue to run cow_file_range to do
587 * the normal delalloc dance
589 if (page_offset(locked_page) >= start &&
590 page_offset(locked_page) <= end) {
591 __set_page_dirty_nobuffers(locked_page);
592 /* unlocked later on in the async handlers */
595 extent_range_redirty_for_io(inode, start, end);
596 add_async_extent(async_cow, start, end - start + 1,
597 0, NULL, 0, BTRFS_COMPRESS_NONE);
605 for (i = 0; i < nr_pages_ret; i++) {
606 WARN_ON(pages[i]->mapping);
607 page_cache_release(pages[i]);
615 * phase two of compressed writeback. This is the ordered portion
616 * of the code, which only gets called in the order the work was
617 * queued. We walk all the async extents created by compress_file_range
618 * and send them down to the disk.
620 static noinline int submit_compressed_extents(struct inode *inode,
621 struct async_cow *async_cow)
623 struct async_extent *async_extent;
625 struct btrfs_key ins;
626 struct extent_map *em;
627 struct btrfs_root *root = BTRFS_I(inode)->root;
628 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
629 struct extent_io_tree *io_tree;
632 if (list_empty(&async_cow->extents))
636 while (!list_empty(&async_cow->extents)) {
637 async_extent = list_entry(async_cow->extents.next,
638 struct async_extent, list);
639 list_del(&async_extent->list);
641 io_tree = &BTRFS_I(inode)->io_tree;
644 /* did the compression code fall back to uncompressed IO? */
645 if (!async_extent->pages) {
646 int page_started = 0;
647 unsigned long nr_written = 0;
649 lock_extent(io_tree, async_extent->start,
650 async_extent->start +
651 async_extent->ram_size - 1);
653 /* allocate blocks */
654 ret = cow_file_range(inode, async_cow->locked_page,
656 async_extent->start +
657 async_extent->ram_size - 1,
658 &page_started, &nr_written, 0);
663 * if page_started, cow_file_range inserted an
664 * inline extent and took care of all the unlocking
665 * and IO for us. Otherwise, we need to submit
666 * all those pages down to the drive.
668 if (!page_started && !ret)
669 extent_write_locked_range(io_tree,
670 inode, async_extent->start,
671 async_extent->start +
672 async_extent->ram_size - 1,
676 unlock_page(async_cow->locked_page);
682 lock_extent(io_tree, async_extent->start,
683 async_extent->start + async_extent->ram_size - 1);
685 ret = btrfs_reserve_extent(root,
686 async_extent->compressed_size,
687 async_extent->compressed_size,
688 0, alloc_hint, &ins, 1);
692 for (i = 0; i < async_extent->nr_pages; i++) {
693 WARN_ON(async_extent->pages[i]->mapping);
694 page_cache_release(async_extent->pages[i]);
696 kfree(async_extent->pages);
697 async_extent->nr_pages = 0;
698 async_extent->pages = NULL;
700 if (ret == -ENOSPC) {
701 unlock_extent(io_tree, async_extent->start,
702 async_extent->start +
703 async_extent->ram_size - 1);
710 * here we're doing allocation and writeback of the
713 btrfs_drop_extent_cache(inode, async_extent->start,
714 async_extent->start +
715 async_extent->ram_size - 1, 0);
717 em = alloc_extent_map();
720 goto out_free_reserve;
722 em->start = async_extent->start;
723 em->len = async_extent->ram_size;
724 em->orig_start = em->start;
725 em->mod_start = em->start;
726 em->mod_len = em->len;
728 em->block_start = ins.objectid;
729 em->block_len = ins.offset;
730 em->orig_block_len = ins.offset;
731 em->ram_bytes = async_extent->ram_size;
732 em->bdev = root->fs_info->fs_devices->latest_bdev;
733 em->compress_type = async_extent->compress_type;
734 set_bit(EXTENT_FLAG_PINNED, &em->flags);
735 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
739 write_lock(&em_tree->lock);
740 ret = add_extent_mapping(em_tree, em, 1);
741 write_unlock(&em_tree->lock);
742 if (ret != -EEXIST) {
746 btrfs_drop_extent_cache(inode, async_extent->start,
747 async_extent->start +
748 async_extent->ram_size - 1, 0);
752 goto out_free_reserve;
754 ret = btrfs_add_ordered_extent_compress(inode,
757 async_extent->ram_size,
759 BTRFS_ORDERED_COMPRESSED,
760 async_extent->compress_type);
762 goto out_free_reserve;
765 * clear dirty, set writeback and unlock the pages.
767 extent_clear_unlock_delalloc(inode, async_extent->start,
768 async_extent->start +
769 async_extent->ram_size - 1,
770 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
771 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
773 ret = btrfs_submit_compressed_write(inode,
775 async_extent->ram_size,
777 ins.offset, async_extent->pages,
778 async_extent->nr_pages);
779 alloc_hint = ins.objectid + ins.offset;
789 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
791 extent_clear_unlock_delalloc(inode, async_extent->start,
792 async_extent->start +
793 async_extent->ram_size - 1,
794 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
795 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
796 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
797 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
802 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
805 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
806 struct extent_map *em;
809 read_lock(&em_tree->lock);
810 em = search_extent_mapping(em_tree, start, num_bytes);
813 * if block start isn't an actual block number then find the
814 * first block in this inode and use that as a hint. If that
815 * block is also bogus then just don't worry about it.
817 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
819 em = search_extent_mapping(em_tree, 0, 0);
820 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
821 alloc_hint = em->block_start;
825 alloc_hint = em->block_start;
829 read_unlock(&em_tree->lock);
835 * when extent_io.c finds a delayed allocation range in the file,
836 * the call backs end up in this code. The basic idea is to
837 * allocate extents on disk for the range, and create ordered data structs
838 * in ram to track those extents.
840 * locked_page is the page that writepage had locked already. We use
841 * it to make sure we don't do extra locks or unlocks.
843 * *page_started is set to one if we unlock locked_page and do everything
844 * required to start IO on it. It may be clean and already done with
847 static noinline int cow_file_range(struct inode *inode,
848 struct page *locked_page,
849 u64 start, u64 end, int *page_started,
850 unsigned long *nr_written,
853 struct btrfs_root *root = BTRFS_I(inode)->root;
856 unsigned long ram_size;
859 u64 blocksize = root->sectorsize;
860 struct btrfs_key ins;
861 struct extent_map *em;
862 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
865 if (btrfs_is_free_space_inode(inode)) {
871 num_bytes = ALIGN(end - start + 1, blocksize);
872 num_bytes = max(blocksize, num_bytes);
873 disk_num_bytes = num_bytes;
875 /* if this is a small write inside eof, kick off defrag */
876 if (num_bytes < 64 * 1024 &&
877 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
878 btrfs_add_inode_defrag(NULL, inode);
881 /* lets try to make an inline extent */
882 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
885 extent_clear_unlock_delalloc(inode, start, end, NULL,
886 EXTENT_LOCKED | EXTENT_DELALLOC |
887 EXTENT_DEFRAG, PAGE_UNLOCK |
888 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
891 *nr_written = *nr_written +
892 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
895 } else if (ret < 0) {
900 BUG_ON(disk_num_bytes >
901 btrfs_super_total_bytes(root->fs_info->super_copy));
903 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
904 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
906 while (disk_num_bytes > 0) {
909 cur_alloc_size = disk_num_bytes;
910 ret = btrfs_reserve_extent(root, cur_alloc_size,
911 root->sectorsize, 0, alloc_hint,
916 em = alloc_extent_map();
922 em->orig_start = em->start;
923 ram_size = ins.offset;
924 em->len = ins.offset;
925 em->mod_start = em->start;
926 em->mod_len = em->len;
928 em->block_start = ins.objectid;
929 em->block_len = ins.offset;
930 em->orig_block_len = ins.offset;
931 em->ram_bytes = ram_size;
932 em->bdev = root->fs_info->fs_devices->latest_bdev;
933 set_bit(EXTENT_FLAG_PINNED, &em->flags);
937 write_lock(&em_tree->lock);
938 ret = add_extent_mapping(em_tree, em, 1);
939 write_unlock(&em_tree->lock);
940 if (ret != -EEXIST) {
944 btrfs_drop_extent_cache(inode, start,
945 start + ram_size - 1, 0);
950 cur_alloc_size = ins.offset;
951 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
952 ram_size, cur_alloc_size, 0);
956 if (root->root_key.objectid ==
957 BTRFS_DATA_RELOC_TREE_OBJECTID) {
958 ret = btrfs_reloc_clone_csums(inode, start,
964 if (disk_num_bytes < cur_alloc_size)
967 /* we're not doing compressed IO, don't unlock the first
968 * page (which the caller expects to stay locked), don't
969 * clear any dirty bits and don't set any writeback bits
971 * Do set the Private2 bit so we know this page was properly
972 * setup for writepage
974 op = unlock ? PAGE_UNLOCK : 0;
975 op |= PAGE_SET_PRIVATE2;
977 extent_clear_unlock_delalloc(inode, start,
978 start + ram_size - 1, locked_page,
979 EXTENT_LOCKED | EXTENT_DELALLOC,
981 disk_num_bytes -= cur_alloc_size;
982 num_bytes -= cur_alloc_size;
983 alloc_hint = ins.objectid + ins.offset;
984 start += cur_alloc_size;
990 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
992 extent_clear_unlock_delalloc(inode, start, end, locked_page,
993 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
994 EXTENT_DELALLOC | EXTENT_DEFRAG,
995 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
996 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1001 * work queue call back to started compression on a file and pages
1003 static noinline void async_cow_start(struct btrfs_work *work)
1005 struct async_cow *async_cow;
1007 async_cow = container_of(work, struct async_cow, work);
1009 compress_file_range(async_cow->inode, async_cow->locked_page,
1010 async_cow->start, async_cow->end, async_cow,
1012 if (num_added == 0) {
1013 btrfs_add_delayed_iput(async_cow->inode);
1014 async_cow->inode = NULL;
1019 * work queue call back to submit previously compressed pages
1021 static noinline void async_cow_submit(struct btrfs_work *work)
1023 struct async_cow *async_cow;
1024 struct btrfs_root *root;
1025 unsigned long nr_pages;
1027 async_cow = container_of(work, struct async_cow, work);
1029 root = async_cow->root;
1030 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1033 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1035 waitqueue_active(&root->fs_info->async_submit_wait))
1036 wake_up(&root->fs_info->async_submit_wait);
1038 if (async_cow->inode)
1039 submit_compressed_extents(async_cow->inode, async_cow);
1042 static noinline void async_cow_free(struct btrfs_work *work)
1044 struct async_cow *async_cow;
1045 async_cow = container_of(work, struct async_cow, work);
1046 if (async_cow->inode)
1047 btrfs_add_delayed_iput(async_cow->inode);
1051 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1052 u64 start, u64 end, int *page_started,
1053 unsigned long *nr_written)
1055 struct async_cow *async_cow;
1056 struct btrfs_root *root = BTRFS_I(inode)->root;
1057 unsigned long nr_pages;
1059 int limit = 10 * 1024 * 1024;
1061 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1062 1, 0, NULL, GFP_NOFS);
1063 while (start < end) {
1064 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1065 BUG_ON(!async_cow); /* -ENOMEM */
1066 async_cow->inode = igrab(inode);
1067 async_cow->root = root;
1068 async_cow->locked_page = locked_page;
1069 async_cow->start = start;
1071 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1074 cur_end = min(end, start + 512 * 1024 - 1);
1076 async_cow->end = cur_end;
1077 INIT_LIST_HEAD(&async_cow->extents);
1079 async_cow->work.func = async_cow_start;
1080 async_cow->work.ordered_func = async_cow_submit;
1081 async_cow->work.ordered_free = async_cow_free;
1082 async_cow->work.flags = 0;
1084 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1086 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1088 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1091 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1092 wait_event(root->fs_info->async_submit_wait,
1093 (atomic_read(&root->fs_info->async_delalloc_pages) <
1097 while (atomic_read(&root->fs_info->async_submit_draining) &&
1098 atomic_read(&root->fs_info->async_delalloc_pages)) {
1099 wait_event(root->fs_info->async_submit_wait,
1100 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1104 *nr_written += nr_pages;
1105 start = cur_end + 1;
1111 static noinline int csum_exist_in_range(struct btrfs_root *root,
1112 u64 bytenr, u64 num_bytes)
1115 struct btrfs_ordered_sum *sums;
1118 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1119 bytenr + num_bytes - 1, &list, 0);
1120 if (ret == 0 && list_empty(&list))
1123 while (!list_empty(&list)) {
1124 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1125 list_del(&sums->list);
1132 * when nowcow writeback call back. This checks for snapshots or COW copies
1133 * of the extents that exist in the file, and COWs the file as required.
1135 * If no cow copies or snapshots exist, we write directly to the existing
1138 static noinline int run_delalloc_nocow(struct inode *inode,
1139 struct page *locked_page,
1140 u64 start, u64 end, int *page_started, int force,
1141 unsigned long *nr_written)
1143 struct btrfs_root *root = BTRFS_I(inode)->root;
1144 struct btrfs_trans_handle *trans;
1145 struct extent_buffer *leaf;
1146 struct btrfs_path *path;
1147 struct btrfs_file_extent_item *fi;
1148 struct btrfs_key found_key;
1163 u64 ino = btrfs_ino(inode);
1165 path = btrfs_alloc_path();
1167 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1168 EXTENT_LOCKED | EXTENT_DELALLOC |
1169 EXTENT_DO_ACCOUNTING |
1170 EXTENT_DEFRAG, PAGE_UNLOCK |
1172 PAGE_SET_WRITEBACK |
1173 PAGE_END_WRITEBACK);
1177 nolock = btrfs_is_free_space_inode(inode);
1180 trans = btrfs_join_transaction_nolock(root);
1182 trans = btrfs_join_transaction(root);
1184 if (IS_ERR(trans)) {
1185 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1186 EXTENT_LOCKED | EXTENT_DELALLOC |
1187 EXTENT_DO_ACCOUNTING |
1188 EXTENT_DEFRAG, PAGE_UNLOCK |
1190 PAGE_SET_WRITEBACK |
1191 PAGE_END_WRITEBACK);
1192 btrfs_free_path(path);
1193 return PTR_ERR(trans);
1196 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1198 cow_start = (u64)-1;
1201 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1205 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1206 leaf = path->nodes[0];
1207 btrfs_item_key_to_cpu(leaf, &found_key,
1208 path->slots[0] - 1);
1209 if (found_key.objectid == ino &&
1210 found_key.type == BTRFS_EXTENT_DATA_KEY)
1215 leaf = path->nodes[0];
1216 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1217 ret = btrfs_next_leaf(root, path);
1222 leaf = path->nodes[0];
1228 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1230 if (found_key.objectid > ino ||
1231 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1232 found_key.offset > end)
1235 if (found_key.offset > cur_offset) {
1236 extent_end = found_key.offset;
1241 fi = btrfs_item_ptr(leaf, path->slots[0],
1242 struct btrfs_file_extent_item);
1243 extent_type = btrfs_file_extent_type(leaf, fi);
1245 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1246 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1247 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1248 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1249 extent_offset = btrfs_file_extent_offset(leaf, fi);
1250 extent_end = found_key.offset +
1251 btrfs_file_extent_num_bytes(leaf, fi);
1253 btrfs_file_extent_disk_num_bytes(leaf, fi);
1254 if (extent_end <= start) {
1258 if (disk_bytenr == 0)
1260 if (btrfs_file_extent_compression(leaf, fi) ||
1261 btrfs_file_extent_encryption(leaf, fi) ||
1262 btrfs_file_extent_other_encoding(leaf, fi))
1264 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1266 if (btrfs_extent_readonly(root, disk_bytenr))
1268 if (btrfs_cross_ref_exist(trans, root, ino,
1270 extent_offset, disk_bytenr))
1272 disk_bytenr += extent_offset;
1273 disk_bytenr += cur_offset - found_key.offset;
1274 num_bytes = min(end + 1, extent_end) - cur_offset;
1276 * force cow if csum exists in the range.
1277 * this ensure that csum for a given extent are
1278 * either valid or do not exist.
1280 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1283 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1284 extent_end = found_key.offset +
1285 btrfs_file_extent_inline_len(leaf,
1286 path->slots[0], fi);
1287 extent_end = ALIGN(extent_end, root->sectorsize);
1292 if (extent_end <= start) {
1297 if (cow_start == (u64)-1)
1298 cow_start = cur_offset;
1299 cur_offset = extent_end;
1300 if (cur_offset > end)
1306 btrfs_release_path(path);
1307 if (cow_start != (u64)-1) {
1308 ret = cow_file_range(inode, locked_page,
1309 cow_start, found_key.offset - 1,
1310 page_started, nr_written, 1);
1313 cow_start = (u64)-1;
1316 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1317 struct extent_map *em;
1318 struct extent_map_tree *em_tree;
1319 em_tree = &BTRFS_I(inode)->extent_tree;
1320 em = alloc_extent_map();
1321 BUG_ON(!em); /* -ENOMEM */
1322 em->start = cur_offset;
1323 em->orig_start = found_key.offset - extent_offset;
1324 em->len = num_bytes;
1325 em->block_len = num_bytes;
1326 em->block_start = disk_bytenr;
1327 em->orig_block_len = disk_num_bytes;
1328 em->ram_bytes = ram_bytes;
1329 em->bdev = root->fs_info->fs_devices->latest_bdev;
1330 em->mod_start = em->start;
1331 em->mod_len = em->len;
1332 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1333 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1334 em->generation = -1;
1336 write_lock(&em_tree->lock);
1337 ret = add_extent_mapping(em_tree, em, 1);
1338 write_unlock(&em_tree->lock);
1339 if (ret != -EEXIST) {
1340 free_extent_map(em);
1343 btrfs_drop_extent_cache(inode, em->start,
1344 em->start + em->len - 1, 0);
1346 type = BTRFS_ORDERED_PREALLOC;
1348 type = BTRFS_ORDERED_NOCOW;
1351 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1352 num_bytes, num_bytes, type);
1353 BUG_ON(ret); /* -ENOMEM */
1355 if (root->root_key.objectid ==
1356 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1357 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1363 extent_clear_unlock_delalloc(inode, cur_offset,
1364 cur_offset + num_bytes - 1,
1365 locked_page, EXTENT_LOCKED |
1366 EXTENT_DELALLOC, PAGE_UNLOCK |
1368 cur_offset = extent_end;
1369 if (cur_offset > end)
1372 btrfs_release_path(path);
1374 if (cur_offset <= end && cow_start == (u64)-1) {
1375 cow_start = cur_offset;
1379 if (cow_start != (u64)-1) {
1380 ret = cow_file_range(inode, locked_page, cow_start, end,
1381 page_started, nr_written, 1);
1387 err = btrfs_end_transaction(trans, root);
1391 if (ret && cur_offset < end)
1392 extent_clear_unlock_delalloc(inode, cur_offset, end,
1393 locked_page, EXTENT_LOCKED |
1394 EXTENT_DELALLOC | EXTENT_DEFRAG |
1395 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1397 PAGE_SET_WRITEBACK |
1398 PAGE_END_WRITEBACK);
1399 btrfs_free_path(path);
1404 * extent_io.c call back to do delayed allocation processing
1406 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1407 u64 start, u64 end, int *page_started,
1408 unsigned long *nr_written)
1411 struct btrfs_root *root = BTRFS_I(inode)->root;
1413 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1414 ret = run_delalloc_nocow(inode, locked_page, start, end,
1415 page_started, 1, nr_written);
1416 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1417 ret = run_delalloc_nocow(inode, locked_page, start, end,
1418 page_started, 0, nr_written);
1419 } else if (!btrfs_test_opt(root, COMPRESS) &&
1420 !(BTRFS_I(inode)->force_compress) &&
1421 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1422 ret = cow_file_range(inode, locked_page, start, end,
1423 page_started, nr_written, 1);
1425 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1426 &BTRFS_I(inode)->runtime_flags);
1427 ret = cow_file_range_async(inode, locked_page, start, end,
1428 page_started, nr_written);
1433 static void btrfs_split_extent_hook(struct inode *inode,
1434 struct extent_state *orig, u64 split)
1436 /* not delalloc, ignore it */
1437 if (!(orig->state & EXTENT_DELALLOC))
1440 spin_lock(&BTRFS_I(inode)->lock);
1441 BTRFS_I(inode)->outstanding_extents++;
1442 spin_unlock(&BTRFS_I(inode)->lock);
1446 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1447 * extents so we can keep track of new extents that are just merged onto old
1448 * extents, such as when we are doing sequential writes, so we can properly
1449 * account for the metadata space we'll need.
1451 static void btrfs_merge_extent_hook(struct inode *inode,
1452 struct extent_state *new,
1453 struct extent_state *other)
1455 /* not delalloc, ignore it */
1456 if (!(other->state & EXTENT_DELALLOC))
1459 spin_lock(&BTRFS_I(inode)->lock);
1460 BTRFS_I(inode)->outstanding_extents--;
1461 spin_unlock(&BTRFS_I(inode)->lock);
1464 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1465 struct inode *inode)
1467 spin_lock(&root->delalloc_lock);
1468 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1469 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1470 &root->delalloc_inodes);
1471 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1472 &BTRFS_I(inode)->runtime_flags);
1473 root->nr_delalloc_inodes++;
1474 if (root->nr_delalloc_inodes == 1) {
1475 spin_lock(&root->fs_info->delalloc_root_lock);
1476 BUG_ON(!list_empty(&root->delalloc_root));
1477 list_add_tail(&root->delalloc_root,
1478 &root->fs_info->delalloc_roots);
1479 spin_unlock(&root->fs_info->delalloc_root_lock);
1482 spin_unlock(&root->delalloc_lock);
1485 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1486 struct inode *inode)
1488 spin_lock(&root->delalloc_lock);
1489 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1490 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1491 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1492 &BTRFS_I(inode)->runtime_flags);
1493 root->nr_delalloc_inodes--;
1494 if (!root->nr_delalloc_inodes) {
1495 spin_lock(&root->fs_info->delalloc_root_lock);
1496 BUG_ON(list_empty(&root->delalloc_root));
1497 list_del_init(&root->delalloc_root);
1498 spin_unlock(&root->fs_info->delalloc_root_lock);
1501 spin_unlock(&root->delalloc_lock);
1505 * extent_io.c set_bit_hook, used to track delayed allocation
1506 * bytes in this file, and to maintain the list of inodes that
1507 * have pending delalloc work to be done.
1509 static void btrfs_set_bit_hook(struct inode *inode,
1510 struct extent_state *state, unsigned long *bits)
1514 * set_bit and clear bit hooks normally require _irqsave/restore
1515 * but in this case, we are only testing for the DELALLOC
1516 * bit, which is only set or cleared with irqs on
1518 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1519 struct btrfs_root *root = BTRFS_I(inode)->root;
1520 u64 len = state->end + 1 - state->start;
1521 bool do_list = !btrfs_is_free_space_inode(inode);
1523 if (*bits & EXTENT_FIRST_DELALLOC) {
1524 *bits &= ~EXTENT_FIRST_DELALLOC;
1526 spin_lock(&BTRFS_I(inode)->lock);
1527 BTRFS_I(inode)->outstanding_extents++;
1528 spin_unlock(&BTRFS_I(inode)->lock);
1531 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1532 root->fs_info->delalloc_batch);
1533 spin_lock(&BTRFS_I(inode)->lock);
1534 BTRFS_I(inode)->delalloc_bytes += len;
1535 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1536 &BTRFS_I(inode)->runtime_flags))
1537 btrfs_add_delalloc_inodes(root, inode);
1538 spin_unlock(&BTRFS_I(inode)->lock);
1543 * extent_io.c clear_bit_hook, see set_bit_hook for why
1545 static void btrfs_clear_bit_hook(struct inode *inode,
1546 struct extent_state *state,
1547 unsigned long *bits)
1550 * set_bit and clear bit hooks normally require _irqsave/restore
1551 * but in this case, we are only testing for the DELALLOC
1552 * bit, which is only set or cleared with irqs on
1554 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1555 struct btrfs_root *root = BTRFS_I(inode)->root;
1556 u64 len = state->end + 1 - state->start;
1557 bool do_list = !btrfs_is_free_space_inode(inode);
1559 if (*bits & EXTENT_FIRST_DELALLOC) {
1560 *bits &= ~EXTENT_FIRST_DELALLOC;
1561 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1562 spin_lock(&BTRFS_I(inode)->lock);
1563 BTRFS_I(inode)->outstanding_extents--;
1564 spin_unlock(&BTRFS_I(inode)->lock);
1568 * We don't reserve metadata space for space cache inodes so we
1569 * don't need to call dellalloc_release_metadata if there is an
1572 if (*bits & EXTENT_DO_ACCOUNTING &&
1573 root != root->fs_info->tree_root)
1574 btrfs_delalloc_release_metadata(inode, len);
1576 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1577 && do_list && !(state->state & EXTENT_NORESERVE))
1578 btrfs_free_reserved_data_space(inode, len);
1580 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1581 root->fs_info->delalloc_batch);
1582 spin_lock(&BTRFS_I(inode)->lock);
1583 BTRFS_I(inode)->delalloc_bytes -= len;
1584 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1585 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1586 &BTRFS_I(inode)->runtime_flags))
1587 btrfs_del_delalloc_inode(root, inode);
1588 spin_unlock(&BTRFS_I(inode)->lock);
1593 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1594 * we don't create bios that span stripes or chunks
1596 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1597 size_t size, struct bio *bio,
1598 unsigned long bio_flags)
1600 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1601 u64 logical = (u64)bio->bi_sector << 9;
1606 if (bio_flags & EXTENT_BIO_COMPRESSED)
1609 length = bio->bi_size;
1610 map_length = length;
1611 ret = btrfs_map_block(root->fs_info, rw, logical,
1612 &map_length, NULL, 0);
1613 /* Will always return 0 with map_multi == NULL */
1615 if (map_length < length + size)
1621 * in order to insert checksums into the metadata in large chunks,
1622 * we wait until bio submission time. All the pages in the bio are
1623 * checksummed and sums are attached onto the ordered extent record.
1625 * At IO completion time the cums attached on the ordered extent record
1626 * are inserted into the btree
1628 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1629 struct bio *bio, int mirror_num,
1630 unsigned long bio_flags,
1633 struct btrfs_root *root = BTRFS_I(inode)->root;
1636 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1637 BUG_ON(ret); /* -ENOMEM */
1642 * in order to insert checksums into the metadata in large chunks,
1643 * we wait until bio submission time. All the pages in the bio are
1644 * checksummed and sums are attached onto the ordered extent record.
1646 * At IO completion time the cums attached on the ordered extent record
1647 * are inserted into the btree
1649 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1650 int mirror_num, unsigned long bio_flags,
1653 struct btrfs_root *root = BTRFS_I(inode)->root;
1656 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1658 bio_endio(bio, ret);
1663 * extent_io.c submission hook. This does the right thing for csum calculation
1664 * on write, or reading the csums from the tree before a read
1666 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1667 int mirror_num, unsigned long bio_flags,
1670 struct btrfs_root *root = BTRFS_I(inode)->root;
1674 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1676 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1678 if (btrfs_is_free_space_inode(inode))
1681 if (!(rw & REQ_WRITE)) {
1682 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1686 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1687 ret = btrfs_submit_compressed_read(inode, bio,
1691 } else if (!skip_sum) {
1692 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1697 } else if (async && !skip_sum) {
1698 /* csum items have already been cloned */
1699 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1701 /* we're doing a write, do the async checksumming */
1702 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1703 inode, rw, bio, mirror_num,
1704 bio_flags, bio_offset,
1705 __btrfs_submit_bio_start,
1706 __btrfs_submit_bio_done);
1708 } else if (!skip_sum) {
1709 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1715 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1719 bio_endio(bio, ret);
1724 * given a list of ordered sums record them in the inode. This happens
1725 * at IO completion time based on sums calculated at bio submission time.
1727 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1728 struct inode *inode, u64 file_offset,
1729 struct list_head *list)
1731 struct btrfs_ordered_sum *sum;
1733 list_for_each_entry(sum, list, list) {
1734 trans->adding_csums = 1;
1735 btrfs_csum_file_blocks(trans,
1736 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1737 trans->adding_csums = 0;
1742 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1743 struct extent_state **cached_state)
1745 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1746 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1747 cached_state, GFP_NOFS);
1750 /* see btrfs_writepage_start_hook for details on why this is required */
1751 struct btrfs_writepage_fixup {
1753 struct btrfs_work work;
1756 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1758 struct btrfs_writepage_fixup *fixup;
1759 struct btrfs_ordered_extent *ordered;
1760 struct extent_state *cached_state = NULL;
1762 struct inode *inode;
1767 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1771 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1772 ClearPageChecked(page);
1776 inode = page->mapping->host;
1777 page_start = page_offset(page);
1778 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1780 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1783 /* already ordered? We're done */
1784 if (PagePrivate2(page))
1787 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1789 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1790 page_end, &cached_state, GFP_NOFS);
1792 btrfs_start_ordered_extent(inode, ordered, 1);
1793 btrfs_put_ordered_extent(ordered);
1797 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1799 mapping_set_error(page->mapping, ret);
1800 end_extent_writepage(page, ret, page_start, page_end);
1801 ClearPageChecked(page);
1805 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1806 ClearPageChecked(page);
1807 set_page_dirty(page);
1809 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1810 &cached_state, GFP_NOFS);
1813 page_cache_release(page);
1818 * There are a few paths in the higher layers of the kernel that directly
1819 * set the page dirty bit without asking the filesystem if it is a
1820 * good idea. This causes problems because we want to make sure COW
1821 * properly happens and the data=ordered rules are followed.
1823 * In our case any range that doesn't have the ORDERED bit set
1824 * hasn't been properly setup for IO. We kick off an async process
1825 * to fix it up. The async helper will wait for ordered extents, set
1826 * the delalloc bit and make it safe to write the page.
1828 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1830 struct inode *inode = page->mapping->host;
1831 struct btrfs_writepage_fixup *fixup;
1832 struct btrfs_root *root = BTRFS_I(inode)->root;
1834 /* this page is properly in the ordered list */
1835 if (TestClearPagePrivate2(page))
1838 if (PageChecked(page))
1841 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1845 SetPageChecked(page);
1846 page_cache_get(page);
1847 fixup->work.func = btrfs_writepage_fixup_worker;
1849 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1853 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1854 struct inode *inode, u64 file_pos,
1855 u64 disk_bytenr, u64 disk_num_bytes,
1856 u64 num_bytes, u64 ram_bytes,
1857 u8 compression, u8 encryption,
1858 u16 other_encoding, int extent_type)
1860 struct btrfs_root *root = BTRFS_I(inode)->root;
1861 struct btrfs_file_extent_item *fi;
1862 struct btrfs_path *path;
1863 struct extent_buffer *leaf;
1864 struct btrfs_key ins;
1865 int extent_inserted = 0;
1868 path = btrfs_alloc_path();
1873 * we may be replacing one extent in the tree with another.
1874 * The new extent is pinned in the extent map, and we don't want
1875 * to drop it from the cache until it is completely in the btree.
1877 * So, tell btrfs_drop_extents to leave this extent in the cache.
1878 * the caller is expected to unpin it and allow it to be merged
1881 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
1882 file_pos + num_bytes, NULL, 0,
1883 1, sizeof(*fi), &extent_inserted);
1887 if (!extent_inserted) {
1888 ins.objectid = btrfs_ino(inode);
1889 ins.offset = file_pos;
1890 ins.type = BTRFS_EXTENT_DATA_KEY;
1892 path->leave_spinning = 1;
1893 ret = btrfs_insert_empty_item(trans, root, path, &ins,
1898 leaf = path->nodes[0];
1899 fi = btrfs_item_ptr(leaf, path->slots[0],
1900 struct btrfs_file_extent_item);
1901 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1902 btrfs_set_file_extent_type(leaf, fi, extent_type);
1903 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1904 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1905 btrfs_set_file_extent_offset(leaf, fi, 0);
1906 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1907 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1908 btrfs_set_file_extent_compression(leaf, fi, compression);
1909 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1910 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1912 btrfs_mark_buffer_dirty(leaf);
1913 btrfs_release_path(path);
1915 inode_add_bytes(inode, num_bytes);
1917 ins.objectid = disk_bytenr;
1918 ins.offset = disk_num_bytes;
1919 ins.type = BTRFS_EXTENT_ITEM_KEY;
1920 ret = btrfs_alloc_reserved_file_extent(trans, root,
1921 root->root_key.objectid,
1922 btrfs_ino(inode), file_pos, &ins);
1924 btrfs_free_path(path);
1929 /* snapshot-aware defrag */
1930 struct sa_defrag_extent_backref {
1931 struct rb_node node;
1932 struct old_sa_defrag_extent *old;
1941 struct old_sa_defrag_extent {
1942 struct list_head list;
1943 struct new_sa_defrag_extent *new;
1952 struct new_sa_defrag_extent {
1953 struct rb_root root;
1954 struct list_head head;
1955 struct btrfs_path *path;
1956 struct inode *inode;
1964 static int backref_comp(struct sa_defrag_extent_backref *b1,
1965 struct sa_defrag_extent_backref *b2)
1967 if (b1->root_id < b2->root_id)
1969 else if (b1->root_id > b2->root_id)
1972 if (b1->inum < b2->inum)
1974 else if (b1->inum > b2->inum)
1977 if (b1->file_pos < b2->file_pos)
1979 else if (b1->file_pos > b2->file_pos)
1983 * [------------------------------] ===> (a range of space)
1984 * |<--->| |<---->| =============> (fs/file tree A)
1985 * |<---------------------------->| ===> (fs/file tree B)
1987 * A range of space can refer to two file extents in one tree while
1988 * refer to only one file extent in another tree.
1990 * So we may process a disk offset more than one time(two extents in A)
1991 * and locate at the same extent(one extent in B), then insert two same
1992 * backrefs(both refer to the extent in B).
1997 static void backref_insert(struct rb_root *root,
1998 struct sa_defrag_extent_backref *backref)
2000 struct rb_node **p = &root->rb_node;
2001 struct rb_node *parent = NULL;
2002 struct sa_defrag_extent_backref *entry;
2007 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2009 ret = backref_comp(backref, entry);
2013 p = &(*p)->rb_right;
2016 rb_link_node(&backref->node, parent, p);
2017 rb_insert_color(&backref->node, root);
2021 * Note the backref might has changed, and in this case we just return 0.
2023 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2026 struct btrfs_file_extent_item *extent;
2027 struct btrfs_fs_info *fs_info;
2028 struct old_sa_defrag_extent *old = ctx;
2029 struct new_sa_defrag_extent *new = old->new;
2030 struct btrfs_path *path = new->path;
2031 struct btrfs_key key;
2032 struct btrfs_root *root;
2033 struct sa_defrag_extent_backref *backref;
2034 struct extent_buffer *leaf;
2035 struct inode *inode = new->inode;
2041 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2042 inum == btrfs_ino(inode))
2045 key.objectid = root_id;
2046 key.type = BTRFS_ROOT_ITEM_KEY;
2047 key.offset = (u64)-1;
2049 fs_info = BTRFS_I(inode)->root->fs_info;
2050 root = btrfs_read_fs_root_no_name(fs_info, &key);
2052 if (PTR_ERR(root) == -ENOENT)
2055 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2056 inum, offset, root_id);
2057 return PTR_ERR(root);
2060 key.objectid = inum;
2061 key.type = BTRFS_EXTENT_DATA_KEY;
2062 if (offset > (u64)-1 << 32)
2065 key.offset = offset;
2067 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2068 if (WARN_ON(ret < 0))
2075 leaf = path->nodes[0];
2076 slot = path->slots[0];
2078 if (slot >= btrfs_header_nritems(leaf)) {
2079 ret = btrfs_next_leaf(root, path);
2082 } else if (ret > 0) {
2091 btrfs_item_key_to_cpu(leaf, &key, slot);
2093 if (key.objectid > inum)
2096 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2099 extent = btrfs_item_ptr(leaf, slot,
2100 struct btrfs_file_extent_item);
2102 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2106 * 'offset' refers to the exact key.offset,
2107 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2108 * (key.offset - extent_offset).
2110 if (key.offset != offset)
2113 extent_offset = btrfs_file_extent_offset(leaf, extent);
2114 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2116 if (extent_offset >= old->extent_offset + old->offset +
2117 old->len || extent_offset + num_bytes <=
2118 old->extent_offset + old->offset)
2123 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2129 backref->root_id = root_id;
2130 backref->inum = inum;
2131 backref->file_pos = offset;
2132 backref->num_bytes = num_bytes;
2133 backref->extent_offset = extent_offset;
2134 backref->generation = btrfs_file_extent_generation(leaf, extent);
2136 backref_insert(&new->root, backref);
2139 btrfs_release_path(path);
2144 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2145 struct new_sa_defrag_extent *new)
2147 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2148 struct old_sa_defrag_extent *old, *tmp;
2153 list_for_each_entry_safe(old, tmp, &new->head, list) {
2154 ret = iterate_inodes_from_logical(old->bytenr +
2155 old->extent_offset, fs_info,
2156 path, record_one_backref,
2158 if (ret < 0 && ret != -ENOENT)
2161 /* no backref to be processed for this extent */
2163 list_del(&old->list);
2168 if (list_empty(&new->head))
2174 static int relink_is_mergable(struct extent_buffer *leaf,
2175 struct btrfs_file_extent_item *fi,
2176 struct new_sa_defrag_extent *new)
2178 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2181 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2184 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2187 if (btrfs_file_extent_encryption(leaf, fi) ||
2188 btrfs_file_extent_other_encoding(leaf, fi))
2195 * Note the backref might has changed, and in this case we just return 0.
2197 static noinline int relink_extent_backref(struct btrfs_path *path,
2198 struct sa_defrag_extent_backref *prev,
2199 struct sa_defrag_extent_backref *backref)
2201 struct btrfs_file_extent_item *extent;
2202 struct btrfs_file_extent_item *item;
2203 struct btrfs_ordered_extent *ordered;
2204 struct btrfs_trans_handle *trans;
2205 struct btrfs_fs_info *fs_info;
2206 struct btrfs_root *root;
2207 struct btrfs_key key;
2208 struct extent_buffer *leaf;
2209 struct old_sa_defrag_extent *old = backref->old;
2210 struct new_sa_defrag_extent *new = old->new;
2211 struct inode *src_inode = new->inode;
2212 struct inode *inode;
2213 struct extent_state *cached = NULL;
2222 if (prev && prev->root_id == backref->root_id &&
2223 prev->inum == backref->inum &&
2224 prev->file_pos + prev->num_bytes == backref->file_pos)
2227 /* step 1: get root */
2228 key.objectid = backref->root_id;
2229 key.type = BTRFS_ROOT_ITEM_KEY;
2230 key.offset = (u64)-1;
2232 fs_info = BTRFS_I(src_inode)->root->fs_info;
2233 index = srcu_read_lock(&fs_info->subvol_srcu);
2235 root = btrfs_read_fs_root_no_name(fs_info, &key);
2237 srcu_read_unlock(&fs_info->subvol_srcu, index);
2238 if (PTR_ERR(root) == -ENOENT)
2240 return PTR_ERR(root);
2243 if (btrfs_root_readonly(root)) {
2244 srcu_read_unlock(&fs_info->subvol_srcu, index);
2248 /* step 2: get inode */
2249 key.objectid = backref->inum;
2250 key.type = BTRFS_INODE_ITEM_KEY;
2253 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2254 if (IS_ERR(inode)) {
2255 srcu_read_unlock(&fs_info->subvol_srcu, index);
2259 srcu_read_unlock(&fs_info->subvol_srcu, index);
2261 /* step 3: relink backref */
2262 lock_start = backref->file_pos;
2263 lock_end = backref->file_pos + backref->num_bytes - 1;
2264 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2267 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2269 btrfs_put_ordered_extent(ordered);
2273 trans = btrfs_join_transaction(root);
2274 if (IS_ERR(trans)) {
2275 ret = PTR_ERR(trans);
2279 key.objectid = backref->inum;
2280 key.type = BTRFS_EXTENT_DATA_KEY;
2281 key.offset = backref->file_pos;
2283 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2286 } else if (ret > 0) {
2291 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2292 struct btrfs_file_extent_item);
2294 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2295 backref->generation)
2298 btrfs_release_path(path);
2300 start = backref->file_pos;
2301 if (backref->extent_offset < old->extent_offset + old->offset)
2302 start += old->extent_offset + old->offset -
2303 backref->extent_offset;
2305 len = min(backref->extent_offset + backref->num_bytes,
2306 old->extent_offset + old->offset + old->len);
2307 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2309 ret = btrfs_drop_extents(trans, root, inode, start,
2314 key.objectid = btrfs_ino(inode);
2315 key.type = BTRFS_EXTENT_DATA_KEY;
2318 path->leave_spinning = 1;
2320 struct btrfs_file_extent_item *fi;
2322 struct btrfs_key found_key;
2324 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2329 leaf = path->nodes[0];
2330 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2332 fi = btrfs_item_ptr(leaf, path->slots[0],
2333 struct btrfs_file_extent_item);
2334 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2336 if (extent_len + found_key.offset == start &&
2337 relink_is_mergable(leaf, fi, new)) {
2338 btrfs_set_file_extent_num_bytes(leaf, fi,
2340 btrfs_mark_buffer_dirty(leaf);
2341 inode_add_bytes(inode, len);
2347 btrfs_release_path(path);
2352 ret = btrfs_insert_empty_item(trans, root, path, &key,
2355 btrfs_abort_transaction(trans, root, ret);
2359 leaf = path->nodes[0];
2360 item = btrfs_item_ptr(leaf, path->slots[0],
2361 struct btrfs_file_extent_item);
2362 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2363 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2364 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2365 btrfs_set_file_extent_num_bytes(leaf, item, len);
2366 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2367 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2368 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2369 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2370 btrfs_set_file_extent_encryption(leaf, item, 0);
2371 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2373 btrfs_mark_buffer_dirty(leaf);
2374 inode_add_bytes(inode, len);
2375 btrfs_release_path(path);
2377 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2379 backref->root_id, backref->inum,
2380 new->file_pos, 0); /* start - extent_offset */
2382 btrfs_abort_transaction(trans, root, ret);
2388 btrfs_release_path(path);
2389 path->leave_spinning = 0;
2390 btrfs_end_transaction(trans, root);
2392 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2398 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2400 struct old_sa_defrag_extent *old, *tmp;
2405 list_for_each_entry_safe(old, tmp, &new->head, list) {
2406 list_del(&old->list);
2412 static void relink_file_extents(struct new_sa_defrag_extent *new)
2414 struct btrfs_path *path;
2415 struct sa_defrag_extent_backref *backref;
2416 struct sa_defrag_extent_backref *prev = NULL;
2417 struct inode *inode;
2418 struct btrfs_root *root;
2419 struct rb_node *node;
2423 root = BTRFS_I(inode)->root;
2425 path = btrfs_alloc_path();
2429 if (!record_extent_backrefs(path, new)) {
2430 btrfs_free_path(path);
2433 btrfs_release_path(path);
2436 node = rb_first(&new->root);
2439 rb_erase(node, &new->root);
2441 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2443 ret = relink_extent_backref(path, prev, backref);
2456 btrfs_free_path(path);
2458 free_sa_defrag_extent(new);
2460 atomic_dec(&root->fs_info->defrag_running);
2461 wake_up(&root->fs_info->transaction_wait);
2464 static struct new_sa_defrag_extent *
2465 record_old_file_extents(struct inode *inode,
2466 struct btrfs_ordered_extent *ordered)
2468 struct btrfs_root *root = BTRFS_I(inode)->root;
2469 struct btrfs_path *path;
2470 struct btrfs_key key;
2471 struct old_sa_defrag_extent *old;
2472 struct new_sa_defrag_extent *new;
2475 new = kmalloc(sizeof(*new), GFP_NOFS);
2480 new->file_pos = ordered->file_offset;
2481 new->len = ordered->len;
2482 new->bytenr = ordered->start;
2483 new->disk_len = ordered->disk_len;
2484 new->compress_type = ordered->compress_type;
2485 new->root = RB_ROOT;
2486 INIT_LIST_HEAD(&new->head);
2488 path = btrfs_alloc_path();
2492 key.objectid = btrfs_ino(inode);
2493 key.type = BTRFS_EXTENT_DATA_KEY;
2494 key.offset = new->file_pos;
2496 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2499 if (ret > 0 && path->slots[0] > 0)
2502 /* find out all the old extents for the file range */
2504 struct btrfs_file_extent_item *extent;
2505 struct extent_buffer *l;
2514 slot = path->slots[0];
2516 if (slot >= btrfs_header_nritems(l)) {
2517 ret = btrfs_next_leaf(root, path);
2525 btrfs_item_key_to_cpu(l, &key, slot);
2527 if (key.objectid != btrfs_ino(inode))
2529 if (key.type != BTRFS_EXTENT_DATA_KEY)
2531 if (key.offset >= new->file_pos + new->len)
2534 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2536 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2537 if (key.offset + num_bytes < new->file_pos)
2540 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2544 extent_offset = btrfs_file_extent_offset(l, extent);
2546 old = kmalloc(sizeof(*old), GFP_NOFS);
2550 offset = max(new->file_pos, key.offset);
2551 end = min(new->file_pos + new->len, key.offset + num_bytes);
2553 old->bytenr = disk_bytenr;
2554 old->extent_offset = extent_offset;
2555 old->offset = offset - key.offset;
2556 old->len = end - offset;
2559 list_add_tail(&old->list, &new->head);
2565 btrfs_free_path(path);
2566 atomic_inc(&root->fs_info->defrag_running);
2571 btrfs_free_path(path);
2573 free_sa_defrag_extent(new);
2577 /* as ordered data IO finishes, this gets called so we can finish
2578 * an ordered extent if the range of bytes in the file it covers are
2581 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2583 struct inode *inode = ordered_extent->inode;
2584 struct btrfs_root *root = BTRFS_I(inode)->root;
2585 struct btrfs_trans_handle *trans = NULL;
2586 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2587 struct extent_state *cached_state = NULL;
2588 struct new_sa_defrag_extent *new = NULL;
2589 int compress_type = 0;
2591 u64 logical_len = ordered_extent->len;
2593 bool truncated = false;
2595 nolock = btrfs_is_free_space_inode(inode);
2597 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2602 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2604 logical_len = ordered_extent->truncated_len;
2605 /* Truncated the entire extent, don't bother adding */
2610 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2611 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2612 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2614 trans = btrfs_join_transaction_nolock(root);
2616 trans = btrfs_join_transaction(root);
2617 if (IS_ERR(trans)) {
2618 ret = PTR_ERR(trans);
2622 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2623 ret = btrfs_update_inode_fallback(trans, root, inode);
2624 if (ret) /* -ENOMEM or corruption */
2625 btrfs_abort_transaction(trans, root, ret);
2629 lock_extent_bits(io_tree, ordered_extent->file_offset,
2630 ordered_extent->file_offset + ordered_extent->len - 1,
2633 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2634 ordered_extent->file_offset + ordered_extent->len - 1,
2635 EXTENT_DEFRAG, 1, cached_state);
2637 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2638 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2639 /* the inode is shared */
2640 new = record_old_file_extents(inode, ordered_extent);
2642 clear_extent_bit(io_tree, ordered_extent->file_offset,
2643 ordered_extent->file_offset + ordered_extent->len - 1,
2644 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2648 trans = btrfs_join_transaction_nolock(root);
2650 trans = btrfs_join_transaction(root);
2651 if (IS_ERR(trans)) {
2652 ret = PTR_ERR(trans);
2656 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2658 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2659 compress_type = ordered_extent->compress_type;
2660 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2661 BUG_ON(compress_type);
2662 ret = btrfs_mark_extent_written(trans, inode,
2663 ordered_extent->file_offset,
2664 ordered_extent->file_offset +
2667 BUG_ON(root == root->fs_info->tree_root);
2668 ret = insert_reserved_file_extent(trans, inode,
2669 ordered_extent->file_offset,
2670 ordered_extent->start,
2671 ordered_extent->disk_len,
2672 logical_len, logical_len,
2673 compress_type, 0, 0,
2674 BTRFS_FILE_EXTENT_REG);
2676 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2677 ordered_extent->file_offset, ordered_extent->len,
2680 btrfs_abort_transaction(trans, root, ret);
2684 add_pending_csums(trans, inode, ordered_extent->file_offset,
2685 &ordered_extent->list);
2687 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2688 ret = btrfs_update_inode_fallback(trans, root, inode);
2689 if (ret) { /* -ENOMEM or corruption */
2690 btrfs_abort_transaction(trans, root, ret);
2695 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2696 ordered_extent->file_offset +
2697 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2699 if (root != root->fs_info->tree_root)
2700 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2702 btrfs_end_transaction(trans, root);
2704 if (ret || truncated) {
2708 start = ordered_extent->file_offset + logical_len;
2710 start = ordered_extent->file_offset;
2711 end = ordered_extent->file_offset + ordered_extent->len - 1;
2712 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2714 /* Drop the cache for the part of the extent we didn't write. */
2715 btrfs_drop_extent_cache(inode, start, end, 0);
2718 * If the ordered extent had an IOERR or something else went
2719 * wrong we need to return the space for this ordered extent
2720 * back to the allocator. We only free the extent in the
2721 * truncated case if we didn't write out the extent at all.
2723 if ((ret || !logical_len) &&
2724 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2725 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2726 btrfs_free_reserved_extent(root, ordered_extent->start,
2727 ordered_extent->disk_len);
2732 * This needs to be done to make sure anybody waiting knows we are done
2733 * updating everything for this ordered extent.
2735 btrfs_remove_ordered_extent(inode, ordered_extent);
2737 /* for snapshot-aware defrag */
2740 free_sa_defrag_extent(new);
2741 atomic_dec(&root->fs_info->defrag_running);
2743 relink_file_extents(new);
2748 btrfs_put_ordered_extent(ordered_extent);
2749 /* once for the tree */
2750 btrfs_put_ordered_extent(ordered_extent);
2755 static void finish_ordered_fn(struct btrfs_work *work)
2757 struct btrfs_ordered_extent *ordered_extent;
2758 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2759 btrfs_finish_ordered_io(ordered_extent);
2762 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2763 struct extent_state *state, int uptodate)
2765 struct inode *inode = page->mapping->host;
2766 struct btrfs_root *root = BTRFS_I(inode)->root;
2767 struct btrfs_ordered_extent *ordered_extent = NULL;
2768 struct btrfs_workers *workers;
2770 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2772 ClearPagePrivate2(page);
2773 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2774 end - start + 1, uptodate))
2777 ordered_extent->work.func = finish_ordered_fn;
2778 ordered_extent->work.flags = 0;
2780 if (btrfs_is_free_space_inode(inode))
2781 workers = &root->fs_info->endio_freespace_worker;
2783 workers = &root->fs_info->endio_write_workers;
2784 btrfs_queue_worker(workers, &ordered_extent->work);
2790 * when reads are done, we need to check csums to verify the data is correct
2791 * if there's a match, we allow the bio to finish. If not, the code in
2792 * extent_io.c will try to find good copies for us.
2794 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2795 u64 phy_offset, struct page *page,
2796 u64 start, u64 end, int mirror)
2798 size_t offset = start - page_offset(page);
2799 struct inode *inode = page->mapping->host;
2800 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2802 struct btrfs_root *root = BTRFS_I(inode)->root;
2805 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2806 DEFAULT_RATELIMIT_BURST);
2808 if (PageChecked(page)) {
2809 ClearPageChecked(page);
2813 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2816 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2817 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2818 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2823 phy_offset >>= inode->i_sb->s_blocksize_bits;
2824 csum_expected = *(((u32 *)io_bio->csum) + phy_offset);
2826 kaddr = kmap_atomic(page);
2827 csum = btrfs_csum_data(kaddr + offset, csum, end - start + 1);
2828 btrfs_csum_final(csum, (char *)&csum);
2829 if (csum != csum_expected)
2832 kunmap_atomic(kaddr);
2837 if (__ratelimit(&_rs))
2838 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
2839 btrfs_ino(page->mapping->host), start, csum, csum_expected);
2840 memset(kaddr + offset, 1, end - start + 1);
2841 flush_dcache_page(page);
2842 kunmap_atomic(kaddr);
2843 if (csum_expected == 0)
2848 struct delayed_iput {
2849 struct list_head list;
2850 struct inode *inode;
2853 /* JDM: If this is fs-wide, why can't we add a pointer to
2854 * btrfs_inode instead and avoid the allocation? */
2855 void btrfs_add_delayed_iput(struct inode *inode)
2857 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2858 struct delayed_iput *delayed;
2860 if (atomic_add_unless(&inode->i_count, -1, 1))
2863 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2864 delayed->inode = inode;
2866 spin_lock(&fs_info->delayed_iput_lock);
2867 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2868 spin_unlock(&fs_info->delayed_iput_lock);
2871 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2874 struct btrfs_fs_info *fs_info = root->fs_info;
2875 struct delayed_iput *delayed;
2878 spin_lock(&fs_info->delayed_iput_lock);
2879 empty = list_empty(&fs_info->delayed_iputs);
2880 spin_unlock(&fs_info->delayed_iput_lock);
2884 spin_lock(&fs_info->delayed_iput_lock);
2885 list_splice_init(&fs_info->delayed_iputs, &list);
2886 spin_unlock(&fs_info->delayed_iput_lock);
2888 while (!list_empty(&list)) {
2889 delayed = list_entry(list.next, struct delayed_iput, list);
2890 list_del(&delayed->list);
2891 iput(delayed->inode);
2897 * This is called in transaction commit time. If there are no orphan
2898 * files in the subvolume, it removes orphan item and frees block_rsv
2901 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2902 struct btrfs_root *root)
2904 struct btrfs_block_rsv *block_rsv;
2907 if (atomic_read(&root->orphan_inodes) ||
2908 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2911 spin_lock(&root->orphan_lock);
2912 if (atomic_read(&root->orphan_inodes)) {
2913 spin_unlock(&root->orphan_lock);
2917 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2918 spin_unlock(&root->orphan_lock);
2922 block_rsv = root->orphan_block_rsv;
2923 root->orphan_block_rsv = NULL;
2924 spin_unlock(&root->orphan_lock);
2926 if (root->orphan_item_inserted &&
2927 btrfs_root_refs(&root->root_item) > 0) {
2928 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2929 root->root_key.objectid);
2931 btrfs_abort_transaction(trans, root, ret);
2933 root->orphan_item_inserted = 0;
2937 WARN_ON(block_rsv->size > 0);
2938 btrfs_free_block_rsv(root, block_rsv);
2943 * This creates an orphan entry for the given inode in case something goes
2944 * wrong in the middle of an unlink/truncate.
2946 * NOTE: caller of this function should reserve 5 units of metadata for
2949 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2951 struct btrfs_root *root = BTRFS_I(inode)->root;
2952 struct btrfs_block_rsv *block_rsv = NULL;
2957 if (!root->orphan_block_rsv) {
2958 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2963 spin_lock(&root->orphan_lock);
2964 if (!root->orphan_block_rsv) {
2965 root->orphan_block_rsv = block_rsv;
2966 } else if (block_rsv) {
2967 btrfs_free_block_rsv(root, block_rsv);
2971 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2972 &BTRFS_I(inode)->runtime_flags)) {
2975 * For proper ENOSPC handling, we should do orphan
2976 * cleanup when mounting. But this introduces backward
2977 * compatibility issue.
2979 if (!xchg(&root->orphan_item_inserted, 1))
2985 atomic_inc(&root->orphan_inodes);
2988 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2989 &BTRFS_I(inode)->runtime_flags))
2991 spin_unlock(&root->orphan_lock);
2993 /* grab metadata reservation from transaction handle */
2995 ret = btrfs_orphan_reserve_metadata(trans, inode);
2996 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2999 /* insert an orphan item to track this unlinked/truncated file */
3001 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3003 atomic_dec(&root->orphan_inodes);
3005 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3006 &BTRFS_I(inode)->runtime_flags);
3007 btrfs_orphan_release_metadata(inode);
3009 if (ret != -EEXIST) {
3010 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3011 &BTRFS_I(inode)->runtime_flags);
3012 btrfs_abort_transaction(trans, root, ret);
3019 /* insert an orphan item to track subvolume contains orphan files */
3021 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3022 root->root_key.objectid);
3023 if (ret && ret != -EEXIST) {
3024 btrfs_abort_transaction(trans, root, ret);
3032 * We have done the truncate/delete so we can go ahead and remove the orphan
3033 * item for this particular inode.
3035 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3036 struct inode *inode)
3038 struct btrfs_root *root = BTRFS_I(inode)->root;
3039 int delete_item = 0;
3040 int release_rsv = 0;
3043 spin_lock(&root->orphan_lock);
3044 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3045 &BTRFS_I(inode)->runtime_flags))
3048 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3049 &BTRFS_I(inode)->runtime_flags))
3051 spin_unlock(&root->orphan_lock);
3054 atomic_dec(&root->orphan_inodes);
3056 ret = btrfs_del_orphan_item(trans, root,
3061 btrfs_orphan_release_metadata(inode);
3067 * this cleans up any orphans that may be left on the list from the last use
3070 int btrfs_orphan_cleanup(struct btrfs_root *root)
3072 struct btrfs_path *path;
3073 struct extent_buffer *leaf;
3074 struct btrfs_key key, found_key;
3075 struct btrfs_trans_handle *trans;
3076 struct inode *inode;
3077 u64 last_objectid = 0;
3078 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3080 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3083 path = btrfs_alloc_path();
3090 key.objectid = BTRFS_ORPHAN_OBJECTID;
3091 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3092 key.offset = (u64)-1;
3095 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3100 * if ret == 0 means we found what we were searching for, which
3101 * is weird, but possible, so only screw with path if we didn't
3102 * find the key and see if we have stuff that matches
3106 if (path->slots[0] == 0)
3111 /* pull out the item */
3112 leaf = path->nodes[0];
3113 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3115 /* make sure the item matches what we want */
3116 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3118 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3121 /* release the path since we're done with it */
3122 btrfs_release_path(path);
3125 * this is where we are basically btrfs_lookup, without the
3126 * crossing root thing. we store the inode number in the
3127 * offset of the orphan item.
3130 if (found_key.offset == last_objectid) {
3131 btrfs_err(root->fs_info,
3132 "Error removing orphan entry, stopping orphan cleanup");
3137 last_objectid = found_key.offset;
3139 found_key.objectid = found_key.offset;
3140 found_key.type = BTRFS_INODE_ITEM_KEY;
3141 found_key.offset = 0;
3142 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3143 ret = PTR_ERR_OR_ZERO(inode);
3144 if (ret && ret != -ESTALE)
3147 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3148 struct btrfs_root *dead_root;
3149 struct btrfs_fs_info *fs_info = root->fs_info;
3150 int is_dead_root = 0;
3153 * this is an orphan in the tree root. Currently these
3154 * could come from 2 sources:
3155 * a) a snapshot deletion in progress
3156 * b) a free space cache inode
3157 * We need to distinguish those two, as the snapshot
3158 * orphan must not get deleted.
3159 * find_dead_roots already ran before us, so if this
3160 * is a snapshot deletion, we should find the root
3161 * in the dead_roots list
3163 spin_lock(&fs_info->trans_lock);
3164 list_for_each_entry(dead_root, &fs_info->dead_roots,
3166 if (dead_root->root_key.objectid ==
3167 found_key.objectid) {
3172 spin_unlock(&fs_info->trans_lock);
3174 /* prevent this orphan from being found again */
3175 key.offset = found_key.objectid - 1;
3180 * Inode is already gone but the orphan item is still there,
3181 * kill the orphan item.
3183 if (ret == -ESTALE) {
3184 trans = btrfs_start_transaction(root, 1);
3185 if (IS_ERR(trans)) {
3186 ret = PTR_ERR(trans);
3189 btrfs_debug(root->fs_info, "auto deleting %Lu",
3190 found_key.objectid);
3191 ret = btrfs_del_orphan_item(trans, root,
3192 found_key.objectid);
3193 btrfs_end_transaction(trans, root);
3200 * add this inode to the orphan list so btrfs_orphan_del does
3201 * the proper thing when we hit it
3203 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3204 &BTRFS_I(inode)->runtime_flags);
3205 atomic_inc(&root->orphan_inodes);
3207 /* if we have links, this was a truncate, lets do that */
3208 if (inode->i_nlink) {
3209 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3215 /* 1 for the orphan item deletion. */
3216 trans = btrfs_start_transaction(root, 1);
3217 if (IS_ERR(trans)) {
3219 ret = PTR_ERR(trans);
3222 ret = btrfs_orphan_add(trans, inode);
3223 btrfs_end_transaction(trans, root);
3229 ret = btrfs_truncate(inode);
3231 btrfs_orphan_del(NULL, inode);
3236 /* this will do delete_inode and everything for us */
3241 /* release the path since we're done with it */
3242 btrfs_release_path(path);
3244 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3246 if (root->orphan_block_rsv)
3247 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3250 if (root->orphan_block_rsv || root->orphan_item_inserted) {
3251 trans = btrfs_join_transaction(root);
3253 btrfs_end_transaction(trans, root);
3257 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3259 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3263 btrfs_crit(root->fs_info,
3264 "could not do orphan cleanup %d", ret);
3265 btrfs_free_path(path);
3270 * very simple check to peek ahead in the leaf looking for xattrs. If we
3271 * don't find any xattrs, we know there can't be any acls.
3273 * slot is the slot the inode is in, objectid is the objectid of the inode
3275 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3276 int slot, u64 objectid,
3277 int *first_xattr_slot)
3279 u32 nritems = btrfs_header_nritems(leaf);
3280 struct btrfs_key found_key;
3281 static u64 xattr_access = 0;
3282 static u64 xattr_default = 0;
3285 if (!xattr_access) {
3286 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3287 strlen(POSIX_ACL_XATTR_ACCESS));
3288 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3289 strlen(POSIX_ACL_XATTR_DEFAULT));
3293 *first_xattr_slot = -1;
3294 while (slot < nritems) {
3295 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3297 /* we found a different objectid, there must not be acls */
3298 if (found_key.objectid != objectid)
3301 /* we found an xattr, assume we've got an acl */
3302 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3303 if (*first_xattr_slot == -1)
3304 *first_xattr_slot = slot;
3305 if (found_key.offset == xattr_access ||
3306 found_key.offset == xattr_default)
3311 * we found a key greater than an xattr key, there can't
3312 * be any acls later on
3314 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3321 * it goes inode, inode backrefs, xattrs, extents,
3322 * so if there are a ton of hard links to an inode there can
3323 * be a lot of backrefs. Don't waste time searching too hard,
3324 * this is just an optimization
3329 /* we hit the end of the leaf before we found an xattr or
3330 * something larger than an xattr. We have to assume the inode
3333 if (*first_xattr_slot == -1)
3334 *first_xattr_slot = slot;
3339 * read an inode from the btree into the in-memory inode
3341 static void btrfs_read_locked_inode(struct inode *inode)
3343 struct btrfs_path *path;
3344 struct extent_buffer *leaf;
3345 struct btrfs_inode_item *inode_item;
3346 struct btrfs_timespec *tspec;
3347 struct btrfs_root *root = BTRFS_I(inode)->root;
3348 struct btrfs_key location;
3353 bool filled = false;
3354 int first_xattr_slot;
3356 ret = btrfs_fill_inode(inode, &rdev);
3360 path = btrfs_alloc_path();
3364 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3366 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3370 leaf = path->nodes[0];
3375 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3376 struct btrfs_inode_item);
3377 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3378 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3379 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3380 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3381 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3383 tspec = btrfs_inode_atime(inode_item);
3384 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3385 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3387 tspec = btrfs_inode_mtime(inode_item);
3388 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3389 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3391 tspec = btrfs_inode_ctime(inode_item);
3392 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3393 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3395 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3396 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3397 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3400 * If we were modified in the current generation and evicted from memory
3401 * and then re-read we need to do a full sync since we don't have any
3402 * idea about which extents were modified before we were evicted from
3405 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3406 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3407 &BTRFS_I(inode)->runtime_flags);
3409 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3410 inode->i_generation = BTRFS_I(inode)->generation;
3412 rdev = btrfs_inode_rdev(leaf, inode_item);
3414 BTRFS_I(inode)->index_cnt = (u64)-1;
3415 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3419 if (inode->i_nlink != 1 ||
3420 path->slots[0] >= btrfs_header_nritems(leaf))
3423 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3424 if (location.objectid != btrfs_ino(inode))
3427 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3428 if (location.type == BTRFS_INODE_REF_KEY) {
3429 struct btrfs_inode_ref *ref;
3431 ref = (struct btrfs_inode_ref *)ptr;
3432 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3433 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3434 struct btrfs_inode_extref *extref;
3436 extref = (struct btrfs_inode_extref *)ptr;
3437 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3442 * try to precache a NULL acl entry for files that don't have
3443 * any xattrs or acls
3445 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3446 btrfs_ino(inode), &first_xattr_slot);
3447 if (first_xattr_slot != -1) {
3448 path->slots[0] = first_xattr_slot;
3449 ret = btrfs_load_inode_props(inode, path);
3451 btrfs_err(root->fs_info,
3452 "error loading props for ino %llu (root %llu): %d\n",
3454 root->root_key.objectid, ret);
3456 btrfs_free_path(path);
3459 cache_no_acl(inode);
3461 switch (inode->i_mode & S_IFMT) {
3463 inode->i_mapping->a_ops = &btrfs_aops;
3464 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3465 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3466 inode->i_fop = &btrfs_file_operations;
3467 inode->i_op = &btrfs_file_inode_operations;
3470 inode->i_fop = &btrfs_dir_file_operations;
3471 if (root == root->fs_info->tree_root)
3472 inode->i_op = &btrfs_dir_ro_inode_operations;
3474 inode->i_op = &btrfs_dir_inode_operations;
3477 inode->i_op = &btrfs_symlink_inode_operations;
3478 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3479 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3482 inode->i_op = &btrfs_special_inode_operations;
3483 init_special_inode(inode, inode->i_mode, rdev);
3487 btrfs_update_iflags(inode);
3491 btrfs_free_path(path);
3492 make_bad_inode(inode);
3496 * given a leaf and an inode, copy the inode fields into the leaf
3498 static void fill_inode_item(struct btrfs_trans_handle *trans,
3499 struct extent_buffer *leaf,
3500 struct btrfs_inode_item *item,
3501 struct inode *inode)
3503 struct btrfs_map_token token;
3505 btrfs_init_map_token(&token);
3507 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3508 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3509 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3511 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3512 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3514 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3515 inode->i_atime.tv_sec, &token);
3516 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3517 inode->i_atime.tv_nsec, &token);
3519 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3520 inode->i_mtime.tv_sec, &token);
3521 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3522 inode->i_mtime.tv_nsec, &token);
3524 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3525 inode->i_ctime.tv_sec, &token);
3526 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3527 inode->i_ctime.tv_nsec, &token);
3529 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3531 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3533 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3534 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3535 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3536 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3537 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3541 * copy everything in the in-memory inode into the btree.
3543 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3544 struct btrfs_root *root, struct inode *inode)
3546 struct btrfs_inode_item *inode_item;
3547 struct btrfs_path *path;
3548 struct extent_buffer *leaf;
3551 path = btrfs_alloc_path();
3555 path->leave_spinning = 1;
3556 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3564 leaf = path->nodes[0];
3565 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3566 struct btrfs_inode_item);
3568 fill_inode_item(trans, leaf, inode_item, inode);
3569 btrfs_mark_buffer_dirty(leaf);
3570 btrfs_set_inode_last_trans(trans, inode);
3573 btrfs_free_path(path);
3578 * copy everything in the in-memory inode into the btree.
3580 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3581 struct btrfs_root *root, struct inode *inode)
3586 * If the inode is a free space inode, we can deadlock during commit
3587 * if we put it into the delayed code.
3589 * The data relocation inode should also be directly updated
3592 if (!btrfs_is_free_space_inode(inode)
3593 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3594 btrfs_update_root_times(trans, root);
3596 ret = btrfs_delayed_update_inode(trans, root, inode);
3598 btrfs_set_inode_last_trans(trans, inode);
3602 return btrfs_update_inode_item(trans, root, inode);
3605 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3606 struct btrfs_root *root,
3607 struct inode *inode)
3611 ret = btrfs_update_inode(trans, root, inode);
3613 return btrfs_update_inode_item(trans, root, inode);
3618 * unlink helper that gets used here in inode.c and in the tree logging
3619 * recovery code. It remove a link in a directory with a given name, and
3620 * also drops the back refs in the inode to the directory
3622 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3623 struct btrfs_root *root,
3624 struct inode *dir, struct inode *inode,
3625 const char *name, int name_len)
3627 struct btrfs_path *path;
3629 struct extent_buffer *leaf;
3630 struct btrfs_dir_item *di;
3631 struct btrfs_key key;
3633 u64 ino = btrfs_ino(inode);
3634 u64 dir_ino = btrfs_ino(dir);
3636 path = btrfs_alloc_path();
3642 path->leave_spinning = 1;
3643 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3644 name, name_len, -1);
3653 leaf = path->nodes[0];
3654 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3655 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3658 btrfs_release_path(path);
3661 * If we don't have dir index, we have to get it by looking up
3662 * the inode ref, since we get the inode ref, remove it directly,
3663 * it is unnecessary to do delayed deletion.
3665 * But if we have dir index, needn't search inode ref to get it.
3666 * Since the inode ref is close to the inode item, it is better
3667 * that we delay to delete it, and just do this deletion when
3668 * we update the inode item.
3670 if (BTRFS_I(inode)->dir_index) {
3671 ret = btrfs_delayed_delete_inode_ref(inode);
3673 index = BTRFS_I(inode)->dir_index;
3678 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3681 btrfs_info(root->fs_info,
3682 "failed to delete reference to %.*s, inode %llu parent %llu",
3683 name_len, name, ino, dir_ino);
3684 btrfs_abort_transaction(trans, root, ret);
3688 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3690 btrfs_abort_transaction(trans, root, ret);
3694 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3696 if (ret != 0 && ret != -ENOENT) {
3697 btrfs_abort_transaction(trans, root, ret);
3701 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3706 btrfs_abort_transaction(trans, root, ret);
3708 btrfs_free_path(path);
3712 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3713 inode_inc_iversion(inode);
3714 inode_inc_iversion(dir);
3715 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3716 ret = btrfs_update_inode(trans, root, dir);
3721 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3722 struct btrfs_root *root,
3723 struct inode *dir, struct inode *inode,
3724 const char *name, int name_len)
3727 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3730 ret = btrfs_update_inode(trans, root, inode);
3736 * helper to start transaction for unlink and rmdir.
3738 * unlink and rmdir are special in btrfs, they do not always free space, so
3739 * if we cannot make our reservations the normal way try and see if there is
3740 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3741 * allow the unlink to occur.
3743 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3745 struct btrfs_trans_handle *trans;
3746 struct btrfs_root *root = BTRFS_I(dir)->root;
3750 * 1 for the possible orphan item
3751 * 1 for the dir item
3752 * 1 for the dir index
3753 * 1 for the inode ref
3756 trans = btrfs_start_transaction(root, 5);
3757 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3760 if (PTR_ERR(trans) == -ENOSPC) {
3761 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3763 trans = btrfs_start_transaction(root, 0);
3766 ret = btrfs_cond_migrate_bytes(root->fs_info,
3767 &root->fs_info->trans_block_rsv,
3770 btrfs_end_transaction(trans, root);
3771 return ERR_PTR(ret);
3773 trans->block_rsv = &root->fs_info->trans_block_rsv;
3774 trans->bytes_reserved = num_bytes;
3779 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3781 struct btrfs_root *root = BTRFS_I(dir)->root;
3782 struct btrfs_trans_handle *trans;
3783 struct inode *inode = dentry->d_inode;
3786 trans = __unlink_start_trans(dir);
3788 return PTR_ERR(trans);
3790 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3792 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3793 dentry->d_name.name, dentry->d_name.len);
3797 if (inode->i_nlink == 0) {
3798 ret = btrfs_orphan_add(trans, inode);
3804 btrfs_end_transaction(trans, root);
3805 btrfs_btree_balance_dirty(root);
3809 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3810 struct btrfs_root *root,
3811 struct inode *dir, u64 objectid,
3812 const char *name, int name_len)
3814 struct btrfs_path *path;
3815 struct extent_buffer *leaf;
3816 struct btrfs_dir_item *di;
3817 struct btrfs_key key;
3820 u64 dir_ino = btrfs_ino(dir);
3822 path = btrfs_alloc_path();
3826 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3827 name, name_len, -1);
3828 if (IS_ERR_OR_NULL(di)) {
3836 leaf = path->nodes[0];
3837 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3838 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3839 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3841 btrfs_abort_transaction(trans, root, ret);
3844 btrfs_release_path(path);
3846 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3847 objectid, root->root_key.objectid,
3848 dir_ino, &index, name, name_len);
3850 if (ret != -ENOENT) {
3851 btrfs_abort_transaction(trans, root, ret);
3854 di = btrfs_search_dir_index_item(root, path, dir_ino,
3856 if (IS_ERR_OR_NULL(di)) {
3861 btrfs_abort_transaction(trans, root, ret);
3865 leaf = path->nodes[0];
3866 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3867 btrfs_release_path(path);
3870 btrfs_release_path(path);
3872 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3874 btrfs_abort_transaction(trans, root, ret);
3878 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3879 inode_inc_iversion(dir);
3880 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3881 ret = btrfs_update_inode_fallback(trans, root, dir);
3883 btrfs_abort_transaction(trans, root, ret);
3885 btrfs_free_path(path);
3889 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3891 struct inode *inode = dentry->d_inode;
3893 struct btrfs_root *root = BTRFS_I(dir)->root;
3894 struct btrfs_trans_handle *trans;
3896 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3898 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3901 trans = __unlink_start_trans(dir);
3903 return PTR_ERR(trans);
3905 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3906 err = btrfs_unlink_subvol(trans, root, dir,
3907 BTRFS_I(inode)->location.objectid,
3908 dentry->d_name.name,
3909 dentry->d_name.len);
3913 err = btrfs_orphan_add(trans, inode);
3917 /* now the directory is empty */
3918 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3919 dentry->d_name.name, dentry->d_name.len);
3921 btrfs_i_size_write(inode, 0);
3923 btrfs_end_transaction(trans, root);
3924 btrfs_btree_balance_dirty(root);
3930 * this can truncate away extent items, csum items and directory items.
3931 * It starts at a high offset and removes keys until it can't find
3932 * any higher than new_size
3934 * csum items that cross the new i_size are truncated to the new size
3937 * min_type is the minimum key type to truncate down to. If set to 0, this
3938 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3940 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3941 struct btrfs_root *root,
3942 struct inode *inode,
3943 u64 new_size, u32 min_type)
3945 struct btrfs_path *path;
3946 struct extent_buffer *leaf;
3947 struct btrfs_file_extent_item *fi;
3948 struct btrfs_key key;
3949 struct btrfs_key found_key;
3950 u64 extent_start = 0;
3951 u64 extent_num_bytes = 0;
3952 u64 extent_offset = 0;
3954 u64 last_size = (u64)-1;
3955 u32 found_type = (u8)-1;
3958 int pending_del_nr = 0;
3959 int pending_del_slot = 0;
3960 int extent_type = -1;
3963 u64 ino = btrfs_ino(inode);
3965 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3967 path = btrfs_alloc_path();
3973 * We want to drop from the next block forward in case this new size is
3974 * not block aligned since we will be keeping the last block of the
3975 * extent just the way it is.
3977 if (root->ref_cows || root == root->fs_info->tree_root)
3978 btrfs_drop_extent_cache(inode, ALIGN(new_size,
3979 root->sectorsize), (u64)-1, 0);
3982 * This function is also used to drop the items in the log tree before
3983 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3984 * it is used to drop the loged items. So we shouldn't kill the delayed
3987 if (min_type == 0 && root == BTRFS_I(inode)->root)
3988 btrfs_kill_delayed_inode_items(inode);
3991 key.offset = (u64)-1;
3995 path->leave_spinning = 1;
3996 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4003 /* there are no items in the tree for us to truncate, we're
4006 if (path->slots[0] == 0)
4013 leaf = path->nodes[0];
4014 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4015 found_type = btrfs_key_type(&found_key);
4017 if (found_key.objectid != ino)
4020 if (found_type < min_type)
4023 item_end = found_key.offset;
4024 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4025 fi = btrfs_item_ptr(leaf, path->slots[0],
4026 struct btrfs_file_extent_item);
4027 extent_type = btrfs_file_extent_type(leaf, fi);
4028 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4030 btrfs_file_extent_num_bytes(leaf, fi);
4031 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4032 item_end += btrfs_file_extent_inline_len(leaf,
4033 path->slots[0], fi);
4037 if (found_type > min_type) {
4040 if (item_end < new_size)
4042 if (found_key.offset >= new_size)
4048 /* FIXME, shrink the extent if the ref count is only 1 */
4049 if (found_type != BTRFS_EXTENT_DATA_KEY)
4053 last_size = found_key.offset;
4055 last_size = new_size;
4057 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4059 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4061 u64 orig_num_bytes =
4062 btrfs_file_extent_num_bytes(leaf, fi);
4063 extent_num_bytes = ALIGN(new_size -
4066 btrfs_set_file_extent_num_bytes(leaf, fi,
4068 num_dec = (orig_num_bytes -
4070 if (root->ref_cows && extent_start != 0)
4071 inode_sub_bytes(inode, num_dec);
4072 btrfs_mark_buffer_dirty(leaf);
4075 btrfs_file_extent_disk_num_bytes(leaf,
4077 extent_offset = found_key.offset -
4078 btrfs_file_extent_offset(leaf, fi);
4080 /* FIXME blocksize != 4096 */
4081 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4082 if (extent_start != 0) {
4085 inode_sub_bytes(inode, num_dec);
4088 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4090 * we can't truncate inline items that have had
4094 btrfs_file_extent_compression(leaf, fi) == 0 &&
4095 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4096 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4097 u32 size = new_size - found_key.offset;
4099 if (root->ref_cows) {
4100 inode_sub_bytes(inode, item_end + 1 -
4105 * update the ram bytes to properly reflect
4106 * the new size of our item
4108 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4110 btrfs_file_extent_calc_inline_size(size);
4111 btrfs_truncate_item(root, path, size, 1);
4112 } else if (root->ref_cows) {
4113 inode_sub_bytes(inode, item_end + 1 -
4119 if (!pending_del_nr) {
4120 /* no pending yet, add ourselves */
4121 pending_del_slot = path->slots[0];
4123 } else if (pending_del_nr &&
4124 path->slots[0] + 1 == pending_del_slot) {
4125 /* hop on the pending chunk */
4127 pending_del_slot = path->slots[0];
4134 if (found_extent && (root->ref_cows ||
4135 root == root->fs_info->tree_root)) {
4136 btrfs_set_path_blocking(path);
4137 ret = btrfs_free_extent(trans, root, extent_start,
4138 extent_num_bytes, 0,
4139 btrfs_header_owner(leaf),
4140 ino, extent_offset, 0);
4144 if (found_type == BTRFS_INODE_ITEM_KEY)
4147 if (path->slots[0] == 0 ||
4148 path->slots[0] != pending_del_slot) {
4149 if (pending_del_nr) {
4150 ret = btrfs_del_items(trans, root, path,
4154 btrfs_abort_transaction(trans,
4160 btrfs_release_path(path);
4167 if (pending_del_nr) {
4168 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4171 btrfs_abort_transaction(trans, root, ret);
4174 if (last_size != (u64)-1)
4175 btrfs_ordered_update_i_size(inode, last_size, NULL);
4176 btrfs_free_path(path);
4181 * btrfs_truncate_page - read, zero a chunk and write a page
4182 * @inode - inode that we're zeroing
4183 * @from - the offset to start zeroing
4184 * @len - the length to zero, 0 to zero the entire range respective to the
4186 * @front - zero up to the offset instead of from the offset on
4188 * This will find the page for the "from" offset and cow the page and zero the
4189 * part we want to zero. This is used with truncate and hole punching.
4191 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4194 struct address_space *mapping = inode->i_mapping;
4195 struct btrfs_root *root = BTRFS_I(inode)->root;
4196 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4197 struct btrfs_ordered_extent *ordered;
4198 struct extent_state *cached_state = NULL;
4200 u32 blocksize = root->sectorsize;
4201 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4202 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4204 gfp_t mask = btrfs_alloc_write_mask(mapping);
4209 if ((offset & (blocksize - 1)) == 0 &&
4210 (!len || ((len & (blocksize - 1)) == 0)))
4212 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4217 page = find_or_create_page(mapping, index, mask);
4219 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4224 page_start = page_offset(page);
4225 page_end = page_start + PAGE_CACHE_SIZE - 1;
4227 if (!PageUptodate(page)) {
4228 ret = btrfs_readpage(NULL, page);
4230 if (page->mapping != mapping) {
4232 page_cache_release(page);
4235 if (!PageUptodate(page)) {
4240 wait_on_page_writeback(page);
4242 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4243 set_page_extent_mapped(page);
4245 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4247 unlock_extent_cached(io_tree, page_start, page_end,
4248 &cached_state, GFP_NOFS);
4250 page_cache_release(page);
4251 btrfs_start_ordered_extent(inode, ordered, 1);
4252 btrfs_put_ordered_extent(ordered);
4256 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4257 EXTENT_DIRTY | EXTENT_DELALLOC |
4258 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4259 0, 0, &cached_state, GFP_NOFS);
4261 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4264 unlock_extent_cached(io_tree, page_start, page_end,
4265 &cached_state, GFP_NOFS);
4269 if (offset != PAGE_CACHE_SIZE) {
4271 len = PAGE_CACHE_SIZE - offset;
4274 memset(kaddr, 0, offset);
4276 memset(kaddr + offset, 0, len);
4277 flush_dcache_page(page);
4280 ClearPageChecked(page);
4281 set_page_dirty(page);
4282 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4287 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4289 page_cache_release(page);
4294 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4295 u64 offset, u64 len)
4297 struct btrfs_trans_handle *trans;
4301 * Still need to make sure the inode looks like it's been updated so
4302 * that any holes get logged if we fsync.
4304 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4305 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4306 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4307 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4312 * 1 - for the one we're dropping
4313 * 1 - for the one we're adding
4314 * 1 - for updating the inode.
4316 trans = btrfs_start_transaction(root, 3);
4318 return PTR_ERR(trans);
4320 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4322 btrfs_abort_transaction(trans, root, ret);
4323 btrfs_end_transaction(trans, root);
4327 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4328 0, 0, len, 0, len, 0, 0, 0);
4330 btrfs_abort_transaction(trans, root, ret);
4332 btrfs_update_inode(trans, root, inode);
4333 btrfs_end_transaction(trans, root);
4338 * This function puts in dummy file extents for the area we're creating a hole
4339 * for. So if we are truncating this file to a larger size we need to insert
4340 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4341 * the range between oldsize and size
4343 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4345 struct btrfs_root *root = BTRFS_I(inode)->root;
4346 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4347 struct extent_map *em = NULL;
4348 struct extent_state *cached_state = NULL;
4349 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4350 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4351 u64 block_end = ALIGN(size, root->sectorsize);
4358 * If our size started in the middle of a page we need to zero out the
4359 * rest of the page before we expand the i_size, otherwise we could
4360 * expose stale data.
4362 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4366 if (size <= hole_start)
4370 struct btrfs_ordered_extent *ordered;
4372 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4374 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4375 block_end - hole_start);
4378 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4379 &cached_state, GFP_NOFS);
4380 btrfs_start_ordered_extent(inode, ordered, 1);
4381 btrfs_put_ordered_extent(ordered);
4384 cur_offset = hole_start;
4386 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4387 block_end - cur_offset, 0);
4393 last_byte = min(extent_map_end(em), block_end);
4394 last_byte = ALIGN(last_byte , root->sectorsize);
4395 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4396 struct extent_map *hole_em;
4397 hole_size = last_byte - cur_offset;
4399 err = maybe_insert_hole(root, inode, cur_offset,
4403 btrfs_drop_extent_cache(inode, cur_offset,
4404 cur_offset + hole_size - 1, 0);
4405 hole_em = alloc_extent_map();
4407 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4408 &BTRFS_I(inode)->runtime_flags);
4411 hole_em->start = cur_offset;
4412 hole_em->len = hole_size;
4413 hole_em->orig_start = cur_offset;
4415 hole_em->block_start = EXTENT_MAP_HOLE;
4416 hole_em->block_len = 0;
4417 hole_em->orig_block_len = 0;
4418 hole_em->ram_bytes = hole_size;
4419 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4420 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4421 hole_em->generation = root->fs_info->generation;
4424 write_lock(&em_tree->lock);
4425 err = add_extent_mapping(em_tree, hole_em, 1);
4426 write_unlock(&em_tree->lock);
4429 btrfs_drop_extent_cache(inode, cur_offset,
4433 free_extent_map(hole_em);
4436 free_extent_map(em);
4438 cur_offset = last_byte;
4439 if (cur_offset >= block_end)
4442 free_extent_map(em);
4443 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4448 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4450 struct btrfs_root *root = BTRFS_I(inode)->root;
4451 struct btrfs_trans_handle *trans;
4452 loff_t oldsize = i_size_read(inode);
4453 loff_t newsize = attr->ia_size;
4454 int mask = attr->ia_valid;
4458 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4459 * special case where we need to update the times despite not having
4460 * these flags set. For all other operations the VFS set these flags
4461 * explicitly if it wants a timestamp update.
4463 if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4464 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4466 if (newsize > oldsize) {
4467 truncate_pagecache(inode, newsize);
4468 ret = btrfs_cont_expand(inode, oldsize, newsize);
4472 trans = btrfs_start_transaction(root, 1);
4474 return PTR_ERR(trans);
4476 i_size_write(inode, newsize);
4477 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4478 ret = btrfs_update_inode(trans, root, inode);
4479 btrfs_end_transaction(trans, root);
4483 * We're truncating a file that used to have good data down to
4484 * zero. Make sure it gets into the ordered flush list so that
4485 * any new writes get down to disk quickly.
4488 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4489 &BTRFS_I(inode)->runtime_flags);
4492 * 1 for the orphan item we're going to add
4493 * 1 for the orphan item deletion.
4495 trans = btrfs_start_transaction(root, 2);
4497 return PTR_ERR(trans);
4500 * We need to do this in case we fail at _any_ point during the
4501 * actual truncate. Once we do the truncate_setsize we could
4502 * invalidate pages which forces any outstanding ordered io to
4503 * be instantly completed which will give us extents that need
4504 * to be truncated. If we fail to get an orphan inode down we
4505 * could have left over extents that were never meant to live,
4506 * so we need to garuntee from this point on that everything
4507 * will be consistent.
4509 ret = btrfs_orphan_add(trans, inode);
4510 btrfs_end_transaction(trans, root);
4514 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4515 truncate_setsize(inode, newsize);
4517 /* Disable nonlocked read DIO to avoid the end less truncate */
4518 btrfs_inode_block_unlocked_dio(inode);
4519 inode_dio_wait(inode);
4520 btrfs_inode_resume_unlocked_dio(inode);
4522 ret = btrfs_truncate(inode);
4523 if (ret && inode->i_nlink) {
4527 * failed to truncate, disk_i_size is only adjusted down
4528 * as we remove extents, so it should represent the true
4529 * size of the inode, so reset the in memory size and
4530 * delete our orphan entry.
4532 trans = btrfs_join_transaction(root);
4533 if (IS_ERR(trans)) {
4534 btrfs_orphan_del(NULL, inode);
4537 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4538 err = btrfs_orphan_del(trans, inode);
4540 btrfs_abort_transaction(trans, root, err);
4541 btrfs_end_transaction(trans, root);
4548 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4550 struct inode *inode = dentry->d_inode;
4551 struct btrfs_root *root = BTRFS_I(inode)->root;
4554 if (btrfs_root_readonly(root))
4557 err = inode_change_ok(inode, attr);
4561 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4562 err = btrfs_setsize(inode, attr);
4567 if (attr->ia_valid) {
4568 setattr_copy(inode, attr);
4569 inode_inc_iversion(inode);
4570 err = btrfs_dirty_inode(inode);
4572 if (!err && attr->ia_valid & ATTR_MODE)
4573 err = btrfs_acl_chmod(inode);
4580 * While truncating the inode pages during eviction, we get the VFS calling
4581 * btrfs_invalidatepage() against each page of the inode. This is slow because
4582 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4583 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4584 * extent_state structures over and over, wasting lots of time.
4586 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4587 * those expensive operations on a per page basis and do only the ordered io
4588 * finishing, while we release here the extent_map and extent_state structures,
4589 * without the excessive merging and splitting.
4591 static void evict_inode_truncate_pages(struct inode *inode)
4593 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4594 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4595 struct rb_node *node;
4597 ASSERT(inode->i_state & I_FREEING);
4598 truncate_inode_pages(&inode->i_data, 0);
4600 write_lock(&map_tree->lock);
4601 while (!RB_EMPTY_ROOT(&map_tree->map)) {
4602 struct extent_map *em;
4604 node = rb_first(&map_tree->map);
4605 em = rb_entry(node, struct extent_map, rb_node);
4606 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4607 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4608 remove_extent_mapping(map_tree, em);
4609 free_extent_map(em);
4611 write_unlock(&map_tree->lock);
4613 spin_lock(&io_tree->lock);
4614 while (!RB_EMPTY_ROOT(&io_tree->state)) {
4615 struct extent_state *state;
4616 struct extent_state *cached_state = NULL;
4618 node = rb_first(&io_tree->state);
4619 state = rb_entry(node, struct extent_state, rb_node);
4620 atomic_inc(&state->refs);
4621 spin_unlock(&io_tree->lock);
4623 lock_extent_bits(io_tree, state->start, state->end,
4625 clear_extent_bit(io_tree, state->start, state->end,
4626 EXTENT_LOCKED | EXTENT_DIRTY |
4627 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
4628 EXTENT_DEFRAG, 1, 1,
4629 &cached_state, GFP_NOFS);
4630 free_extent_state(state);
4632 spin_lock(&io_tree->lock);
4634 spin_unlock(&io_tree->lock);
4637 void btrfs_evict_inode(struct inode *inode)
4639 struct btrfs_trans_handle *trans;
4640 struct btrfs_root *root = BTRFS_I(inode)->root;
4641 struct btrfs_block_rsv *rsv, *global_rsv;
4642 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4645 trace_btrfs_inode_evict(inode);
4647 evict_inode_truncate_pages(inode);
4649 if (inode->i_nlink &&
4650 ((btrfs_root_refs(&root->root_item) != 0 &&
4651 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4652 btrfs_is_free_space_inode(inode)))
4655 if (is_bad_inode(inode)) {
4656 btrfs_orphan_del(NULL, inode);
4659 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4660 btrfs_wait_ordered_range(inode, 0, (u64)-1);
4662 if (root->fs_info->log_root_recovering) {
4663 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4664 &BTRFS_I(inode)->runtime_flags));
4668 if (inode->i_nlink > 0) {
4669 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4670 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
4674 ret = btrfs_commit_inode_delayed_inode(inode);
4676 btrfs_orphan_del(NULL, inode);
4680 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4682 btrfs_orphan_del(NULL, inode);
4685 rsv->size = min_size;
4687 global_rsv = &root->fs_info->global_block_rsv;
4689 btrfs_i_size_write(inode, 0);
4692 * This is a bit simpler than btrfs_truncate since we've already
4693 * reserved our space for our orphan item in the unlink, so we just
4694 * need to reserve some slack space in case we add bytes and update
4695 * inode item when doing the truncate.
4698 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4699 BTRFS_RESERVE_FLUSH_LIMIT);
4702 * Try and steal from the global reserve since we will
4703 * likely not use this space anyway, we want to try as
4704 * hard as possible to get this to work.
4707 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4710 btrfs_warn(root->fs_info,
4711 "Could not get space for a delete, will truncate on mount %d",
4713 btrfs_orphan_del(NULL, inode);
4714 btrfs_free_block_rsv(root, rsv);
4718 trans = btrfs_join_transaction(root);
4719 if (IS_ERR(trans)) {
4720 btrfs_orphan_del(NULL, inode);
4721 btrfs_free_block_rsv(root, rsv);
4725 trans->block_rsv = rsv;
4727 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4731 trans->block_rsv = &root->fs_info->trans_block_rsv;
4732 btrfs_end_transaction(trans, root);
4734 btrfs_btree_balance_dirty(root);
4737 btrfs_free_block_rsv(root, rsv);
4740 * Errors here aren't a big deal, it just means we leave orphan items
4741 * in the tree. They will be cleaned up on the next mount.
4744 trans->block_rsv = root->orphan_block_rsv;
4745 btrfs_orphan_del(trans, inode);
4747 btrfs_orphan_del(NULL, inode);
4750 trans->block_rsv = &root->fs_info->trans_block_rsv;
4751 if (!(root == root->fs_info->tree_root ||
4752 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4753 btrfs_return_ino(root, btrfs_ino(inode));
4755 btrfs_end_transaction(trans, root);
4756 btrfs_btree_balance_dirty(root);
4758 btrfs_remove_delayed_node(inode);
4764 * this returns the key found in the dir entry in the location pointer.
4765 * If no dir entries were found, location->objectid is 0.
4767 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4768 struct btrfs_key *location)
4770 const char *name = dentry->d_name.name;
4771 int namelen = dentry->d_name.len;
4772 struct btrfs_dir_item *di;
4773 struct btrfs_path *path;
4774 struct btrfs_root *root = BTRFS_I(dir)->root;
4777 path = btrfs_alloc_path();
4781 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4786 if (IS_ERR_OR_NULL(di))
4789 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4791 btrfs_free_path(path);
4794 location->objectid = 0;
4799 * when we hit a tree root in a directory, the btrfs part of the inode
4800 * needs to be changed to reflect the root directory of the tree root. This
4801 * is kind of like crossing a mount point.
4803 static int fixup_tree_root_location(struct btrfs_root *root,
4805 struct dentry *dentry,
4806 struct btrfs_key *location,
4807 struct btrfs_root **sub_root)
4809 struct btrfs_path *path;
4810 struct btrfs_root *new_root;
4811 struct btrfs_root_ref *ref;
4812 struct extent_buffer *leaf;
4816 path = btrfs_alloc_path();
4823 ret = btrfs_find_item(root->fs_info->tree_root, path,
4824 BTRFS_I(dir)->root->root_key.objectid,
4825 location->objectid, BTRFS_ROOT_REF_KEY, NULL);
4832 leaf = path->nodes[0];
4833 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4834 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4835 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4838 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4839 (unsigned long)(ref + 1),
4840 dentry->d_name.len);
4844 btrfs_release_path(path);
4846 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4847 if (IS_ERR(new_root)) {
4848 err = PTR_ERR(new_root);
4852 *sub_root = new_root;
4853 location->objectid = btrfs_root_dirid(&new_root->root_item);
4854 location->type = BTRFS_INODE_ITEM_KEY;
4855 location->offset = 0;
4858 btrfs_free_path(path);
4862 static void inode_tree_add(struct inode *inode)
4864 struct btrfs_root *root = BTRFS_I(inode)->root;
4865 struct btrfs_inode *entry;
4867 struct rb_node *parent;
4868 struct rb_node *new = &BTRFS_I(inode)->rb_node;
4869 u64 ino = btrfs_ino(inode);
4871 if (inode_unhashed(inode))
4874 spin_lock(&root->inode_lock);
4875 p = &root->inode_tree.rb_node;
4878 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4880 if (ino < btrfs_ino(&entry->vfs_inode))
4881 p = &parent->rb_left;
4882 else if (ino > btrfs_ino(&entry->vfs_inode))
4883 p = &parent->rb_right;
4885 WARN_ON(!(entry->vfs_inode.i_state &
4886 (I_WILL_FREE | I_FREEING)));
4887 rb_replace_node(parent, new, &root->inode_tree);
4888 RB_CLEAR_NODE(parent);
4889 spin_unlock(&root->inode_lock);
4893 rb_link_node(new, parent, p);
4894 rb_insert_color(new, &root->inode_tree);
4895 spin_unlock(&root->inode_lock);
4898 static void inode_tree_del(struct inode *inode)
4900 struct btrfs_root *root = BTRFS_I(inode)->root;
4903 spin_lock(&root->inode_lock);
4904 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4905 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4906 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4907 empty = RB_EMPTY_ROOT(&root->inode_tree);
4909 spin_unlock(&root->inode_lock);
4911 if (empty && btrfs_root_refs(&root->root_item) == 0) {
4912 synchronize_srcu(&root->fs_info->subvol_srcu);
4913 spin_lock(&root->inode_lock);
4914 empty = RB_EMPTY_ROOT(&root->inode_tree);
4915 spin_unlock(&root->inode_lock);
4917 btrfs_add_dead_root(root);
4921 void btrfs_invalidate_inodes(struct btrfs_root *root)
4923 struct rb_node *node;
4924 struct rb_node *prev;
4925 struct btrfs_inode *entry;
4926 struct inode *inode;
4929 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4931 spin_lock(&root->inode_lock);
4933 node = root->inode_tree.rb_node;
4937 entry = rb_entry(node, struct btrfs_inode, rb_node);
4939 if (objectid < btrfs_ino(&entry->vfs_inode))
4940 node = node->rb_left;
4941 else if (objectid > btrfs_ino(&entry->vfs_inode))
4942 node = node->rb_right;
4948 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4949 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4953 prev = rb_next(prev);
4957 entry = rb_entry(node, struct btrfs_inode, rb_node);
4958 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4959 inode = igrab(&entry->vfs_inode);
4961 spin_unlock(&root->inode_lock);
4962 if (atomic_read(&inode->i_count) > 1)
4963 d_prune_aliases(inode);
4965 * btrfs_drop_inode will have it removed from
4966 * the inode cache when its usage count
4971 spin_lock(&root->inode_lock);
4975 if (cond_resched_lock(&root->inode_lock))
4978 node = rb_next(node);
4980 spin_unlock(&root->inode_lock);
4983 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4985 struct btrfs_iget_args *args = p;
4986 inode->i_ino = args->location->objectid;
4987 memcpy(&BTRFS_I(inode)->location, args->location,
4988 sizeof(*args->location));
4989 BTRFS_I(inode)->root = args->root;
4993 static int btrfs_find_actor(struct inode *inode, void *opaque)
4995 struct btrfs_iget_args *args = opaque;
4996 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
4997 args->root == BTRFS_I(inode)->root;
5000 static struct inode *btrfs_iget_locked(struct super_block *s,
5001 struct btrfs_key *location,
5002 struct btrfs_root *root)
5004 struct inode *inode;
5005 struct btrfs_iget_args args;
5006 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5008 args.location = location;
5011 inode = iget5_locked(s, hashval, btrfs_find_actor,
5012 btrfs_init_locked_inode,
5017 /* Get an inode object given its location and corresponding root.
5018 * Returns in *is_new if the inode was read from disk
5020 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5021 struct btrfs_root *root, int *new)
5023 struct inode *inode;
5025 inode = btrfs_iget_locked(s, location, root);
5027 return ERR_PTR(-ENOMEM);
5029 if (inode->i_state & I_NEW) {
5030 btrfs_read_locked_inode(inode);
5031 if (!is_bad_inode(inode)) {
5032 inode_tree_add(inode);
5033 unlock_new_inode(inode);
5037 unlock_new_inode(inode);
5039 inode = ERR_PTR(-ESTALE);
5046 static struct inode *new_simple_dir(struct super_block *s,
5047 struct btrfs_key *key,
5048 struct btrfs_root *root)
5050 struct inode *inode = new_inode(s);
5053 return ERR_PTR(-ENOMEM);
5055 BTRFS_I(inode)->root = root;
5056 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5057 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5059 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5060 inode->i_op = &btrfs_dir_ro_inode_operations;
5061 inode->i_fop = &simple_dir_operations;
5062 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5063 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5068 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5070 struct inode *inode;
5071 struct btrfs_root *root = BTRFS_I(dir)->root;
5072 struct btrfs_root *sub_root = root;
5073 struct btrfs_key location;
5077 if (dentry->d_name.len > BTRFS_NAME_LEN)
5078 return ERR_PTR(-ENAMETOOLONG);
5080 ret = btrfs_inode_by_name(dir, dentry, &location);
5082 return ERR_PTR(ret);
5084 if (location.objectid == 0)
5085 return ERR_PTR(-ENOENT);
5087 if (location.type == BTRFS_INODE_ITEM_KEY) {
5088 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5092 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5094 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5095 ret = fixup_tree_root_location(root, dir, dentry,
5096 &location, &sub_root);
5099 inode = ERR_PTR(ret);
5101 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5103 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5105 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5107 if (!IS_ERR(inode) && root != sub_root) {
5108 down_read(&root->fs_info->cleanup_work_sem);
5109 if (!(inode->i_sb->s_flags & MS_RDONLY))
5110 ret = btrfs_orphan_cleanup(sub_root);
5111 up_read(&root->fs_info->cleanup_work_sem);
5114 inode = ERR_PTR(ret);
5121 static int btrfs_dentry_delete(const struct dentry *dentry)
5123 struct btrfs_root *root;
5124 struct inode *inode = dentry->d_inode;
5126 if (!inode && !IS_ROOT(dentry))
5127 inode = dentry->d_parent->d_inode;
5130 root = BTRFS_I(inode)->root;
5131 if (btrfs_root_refs(&root->root_item) == 0)
5134 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5140 static void btrfs_dentry_release(struct dentry *dentry)
5142 if (dentry->d_fsdata)
5143 kfree(dentry->d_fsdata);
5146 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5149 struct inode *inode;
5151 inode = btrfs_lookup_dentry(dir, dentry);
5152 if (IS_ERR(inode)) {
5153 if (PTR_ERR(inode) == -ENOENT)
5156 return ERR_CAST(inode);
5159 return d_materialise_unique(dentry, inode);
5162 unsigned char btrfs_filetype_table[] = {
5163 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5166 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5168 struct inode *inode = file_inode(file);
5169 struct btrfs_root *root = BTRFS_I(inode)->root;
5170 struct btrfs_item *item;
5171 struct btrfs_dir_item *di;
5172 struct btrfs_key key;
5173 struct btrfs_key found_key;
5174 struct btrfs_path *path;
5175 struct list_head ins_list;
5176 struct list_head del_list;
5178 struct extent_buffer *leaf;
5180 unsigned char d_type;
5185 int key_type = BTRFS_DIR_INDEX_KEY;
5189 int is_curr = 0; /* ctx->pos points to the current index? */
5191 /* FIXME, use a real flag for deciding about the key type */
5192 if (root->fs_info->tree_root == root)
5193 key_type = BTRFS_DIR_ITEM_KEY;
5195 if (!dir_emit_dots(file, ctx))
5198 path = btrfs_alloc_path();
5204 if (key_type == BTRFS_DIR_INDEX_KEY) {
5205 INIT_LIST_HEAD(&ins_list);
5206 INIT_LIST_HEAD(&del_list);
5207 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5210 btrfs_set_key_type(&key, key_type);
5211 key.offset = ctx->pos;
5212 key.objectid = btrfs_ino(inode);
5214 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5219 leaf = path->nodes[0];
5220 slot = path->slots[0];
5221 if (slot >= btrfs_header_nritems(leaf)) {
5222 ret = btrfs_next_leaf(root, path);
5230 item = btrfs_item_nr(slot);
5231 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5233 if (found_key.objectid != key.objectid)
5235 if (btrfs_key_type(&found_key) != key_type)
5237 if (found_key.offset < ctx->pos)
5239 if (key_type == BTRFS_DIR_INDEX_KEY &&
5240 btrfs_should_delete_dir_index(&del_list,
5244 ctx->pos = found_key.offset;
5247 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5249 di_total = btrfs_item_size(leaf, item);
5251 while (di_cur < di_total) {
5252 struct btrfs_key location;
5254 if (verify_dir_item(root, leaf, di))
5257 name_len = btrfs_dir_name_len(leaf, di);
5258 if (name_len <= sizeof(tmp_name)) {
5259 name_ptr = tmp_name;
5261 name_ptr = kmalloc(name_len, GFP_NOFS);
5267 read_extent_buffer(leaf, name_ptr,
5268 (unsigned long)(di + 1), name_len);
5270 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5271 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5274 /* is this a reference to our own snapshot? If so
5277 * In contrast to old kernels, we insert the snapshot's
5278 * dir item and dir index after it has been created, so
5279 * we won't find a reference to our own snapshot. We
5280 * still keep the following code for backward
5283 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5284 location.objectid == root->root_key.objectid) {
5288 over = !dir_emit(ctx, name_ptr, name_len,
5289 location.objectid, d_type);
5292 if (name_ptr != tmp_name)
5297 di_len = btrfs_dir_name_len(leaf, di) +
5298 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5300 di = (struct btrfs_dir_item *)((char *)di + di_len);
5306 if (key_type == BTRFS_DIR_INDEX_KEY) {
5309 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5314 /* Reached end of directory/root. Bump pos past the last item. */
5318 * Stop new entries from being returned after we return the last
5321 * New directory entries are assigned a strictly increasing
5322 * offset. This means that new entries created during readdir
5323 * are *guaranteed* to be seen in the future by that readdir.
5324 * This has broken buggy programs which operate on names as
5325 * they're returned by readdir. Until we re-use freed offsets
5326 * we have this hack to stop new entries from being returned
5327 * under the assumption that they'll never reach this huge
5330 * This is being careful not to overflow 32bit loff_t unless the
5331 * last entry requires it because doing so has broken 32bit apps
5334 if (key_type == BTRFS_DIR_INDEX_KEY) {
5335 if (ctx->pos >= INT_MAX)
5336 ctx->pos = LLONG_MAX;
5343 if (key_type == BTRFS_DIR_INDEX_KEY)
5344 btrfs_put_delayed_items(&ins_list, &del_list);
5345 btrfs_free_path(path);
5349 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5351 struct btrfs_root *root = BTRFS_I(inode)->root;
5352 struct btrfs_trans_handle *trans;
5354 bool nolock = false;
5356 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5359 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5362 if (wbc->sync_mode == WB_SYNC_ALL) {
5364 trans = btrfs_join_transaction_nolock(root);
5366 trans = btrfs_join_transaction(root);
5368 return PTR_ERR(trans);
5369 ret = btrfs_commit_transaction(trans, root);
5375 * This is somewhat expensive, updating the tree every time the
5376 * inode changes. But, it is most likely to find the inode in cache.
5377 * FIXME, needs more benchmarking...there are no reasons other than performance
5378 * to keep or drop this code.
5380 static int btrfs_dirty_inode(struct inode *inode)
5382 struct btrfs_root *root = BTRFS_I(inode)->root;
5383 struct btrfs_trans_handle *trans;
5386 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5389 trans = btrfs_join_transaction(root);
5391 return PTR_ERR(trans);
5393 ret = btrfs_update_inode(trans, root, inode);
5394 if (ret && ret == -ENOSPC) {
5395 /* whoops, lets try again with the full transaction */
5396 btrfs_end_transaction(trans, root);
5397 trans = btrfs_start_transaction(root, 1);
5399 return PTR_ERR(trans);
5401 ret = btrfs_update_inode(trans, root, inode);
5403 btrfs_end_transaction(trans, root);
5404 if (BTRFS_I(inode)->delayed_node)
5405 btrfs_balance_delayed_items(root);
5411 * This is a copy of file_update_time. We need this so we can return error on
5412 * ENOSPC for updating the inode in the case of file write and mmap writes.
5414 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5417 struct btrfs_root *root = BTRFS_I(inode)->root;
5419 if (btrfs_root_readonly(root))
5422 if (flags & S_VERSION)
5423 inode_inc_iversion(inode);
5424 if (flags & S_CTIME)
5425 inode->i_ctime = *now;
5426 if (flags & S_MTIME)
5427 inode->i_mtime = *now;
5428 if (flags & S_ATIME)
5429 inode->i_atime = *now;
5430 return btrfs_dirty_inode(inode);
5434 * find the highest existing sequence number in a directory
5435 * and then set the in-memory index_cnt variable to reflect
5436 * free sequence numbers
5438 static int btrfs_set_inode_index_count(struct inode *inode)
5440 struct btrfs_root *root = BTRFS_I(inode)->root;
5441 struct btrfs_key key, found_key;
5442 struct btrfs_path *path;
5443 struct extent_buffer *leaf;
5446 key.objectid = btrfs_ino(inode);
5447 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5448 key.offset = (u64)-1;
5450 path = btrfs_alloc_path();
5454 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5457 /* FIXME: we should be able to handle this */
5463 * MAGIC NUMBER EXPLANATION:
5464 * since we search a directory based on f_pos we have to start at 2
5465 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5466 * else has to start at 2
5468 if (path->slots[0] == 0) {
5469 BTRFS_I(inode)->index_cnt = 2;
5475 leaf = path->nodes[0];
5476 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5478 if (found_key.objectid != btrfs_ino(inode) ||
5479 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5480 BTRFS_I(inode)->index_cnt = 2;
5484 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5486 btrfs_free_path(path);
5491 * helper to find a free sequence number in a given directory. This current
5492 * code is very simple, later versions will do smarter things in the btree
5494 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5498 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5499 ret = btrfs_inode_delayed_dir_index_count(dir);
5501 ret = btrfs_set_inode_index_count(dir);
5507 *index = BTRFS_I(dir)->index_cnt;
5508 BTRFS_I(dir)->index_cnt++;
5513 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5514 struct btrfs_root *root,
5516 const char *name, int name_len,
5517 u64 ref_objectid, u64 objectid,
5518 umode_t mode, u64 *index)
5520 struct inode *inode;
5521 struct btrfs_inode_item *inode_item;
5522 struct btrfs_key *location;
5523 struct btrfs_path *path;
5524 struct btrfs_inode_ref *ref;
5525 struct btrfs_key key[2];
5530 path = btrfs_alloc_path();
5532 return ERR_PTR(-ENOMEM);
5534 inode = new_inode(root->fs_info->sb);
5536 btrfs_free_path(path);
5537 return ERR_PTR(-ENOMEM);
5541 * we have to initialize this early, so we can reclaim the inode
5542 * number if we fail afterwards in this function.
5544 inode->i_ino = objectid;
5547 trace_btrfs_inode_request(dir);
5549 ret = btrfs_set_inode_index(dir, index);
5551 btrfs_free_path(path);
5553 return ERR_PTR(ret);
5557 * index_cnt is ignored for everything but a dir,
5558 * btrfs_get_inode_index_count has an explanation for the magic
5561 BTRFS_I(inode)->index_cnt = 2;
5562 BTRFS_I(inode)->dir_index = *index;
5563 BTRFS_I(inode)->root = root;
5564 BTRFS_I(inode)->generation = trans->transid;
5565 inode->i_generation = BTRFS_I(inode)->generation;
5568 * We could have gotten an inode number from somebody who was fsynced
5569 * and then removed in this same transaction, so let's just set full
5570 * sync since it will be a full sync anyway and this will blow away the
5571 * old info in the log.
5573 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5575 key[0].objectid = objectid;
5576 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5580 * Start new inodes with an inode_ref. This is slightly more
5581 * efficient for small numbers of hard links since they will
5582 * be packed into one item. Extended refs will kick in if we
5583 * add more hard links than can fit in the ref item.
5585 key[1].objectid = objectid;
5586 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5587 key[1].offset = ref_objectid;
5589 sizes[0] = sizeof(struct btrfs_inode_item);
5590 sizes[1] = name_len + sizeof(*ref);
5592 path->leave_spinning = 1;
5593 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5597 inode_init_owner(inode, dir, mode);
5598 inode_set_bytes(inode, 0);
5599 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5600 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5601 struct btrfs_inode_item);
5602 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5603 sizeof(*inode_item));
5604 fill_inode_item(trans, path->nodes[0], inode_item, inode);
5606 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5607 struct btrfs_inode_ref);
5608 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5609 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5610 ptr = (unsigned long)(ref + 1);
5611 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5613 btrfs_mark_buffer_dirty(path->nodes[0]);
5614 btrfs_free_path(path);
5616 location = &BTRFS_I(inode)->location;
5617 location->objectid = objectid;
5618 location->offset = 0;
5619 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5621 btrfs_inherit_iflags(inode, dir);
5623 if (S_ISREG(mode)) {
5624 if (btrfs_test_opt(root, NODATASUM))
5625 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5626 if (btrfs_test_opt(root, NODATACOW))
5627 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5628 BTRFS_INODE_NODATASUM;
5631 btrfs_insert_inode_hash(inode);
5632 inode_tree_add(inode);
5634 trace_btrfs_inode_new(inode);
5635 btrfs_set_inode_last_trans(trans, inode);
5637 btrfs_update_root_times(trans, root);
5639 ret = btrfs_inode_inherit_props(trans, inode, dir);
5641 btrfs_err(root->fs_info,
5642 "error inheriting props for ino %llu (root %llu): %d",
5643 btrfs_ino(inode), root->root_key.objectid, ret);
5648 BTRFS_I(dir)->index_cnt--;
5649 btrfs_free_path(path);
5651 return ERR_PTR(ret);
5654 static inline u8 btrfs_inode_type(struct inode *inode)
5656 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5660 * utility function to add 'inode' into 'parent_inode' with
5661 * a give name and a given sequence number.
5662 * if 'add_backref' is true, also insert a backref from the
5663 * inode to the parent directory.
5665 int btrfs_add_link(struct btrfs_trans_handle *trans,
5666 struct inode *parent_inode, struct inode *inode,
5667 const char *name, int name_len, int add_backref, u64 index)
5670 struct btrfs_key key;
5671 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5672 u64 ino = btrfs_ino(inode);
5673 u64 parent_ino = btrfs_ino(parent_inode);
5675 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5676 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5679 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5683 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5684 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5685 key.objectid, root->root_key.objectid,
5686 parent_ino, index, name, name_len);
5687 } else if (add_backref) {
5688 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5692 /* Nothing to clean up yet */
5696 ret = btrfs_insert_dir_item(trans, root, name, name_len,
5698 btrfs_inode_type(inode), index);
5699 if (ret == -EEXIST || ret == -EOVERFLOW)
5702 btrfs_abort_transaction(trans, root, ret);
5706 btrfs_i_size_write(parent_inode, parent_inode->i_size +
5708 inode_inc_iversion(parent_inode);
5709 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5710 ret = btrfs_update_inode(trans, root, parent_inode);
5712 btrfs_abort_transaction(trans, root, ret);
5716 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5719 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5720 key.objectid, root->root_key.objectid,
5721 parent_ino, &local_index, name, name_len);
5723 } else if (add_backref) {
5727 err = btrfs_del_inode_ref(trans, root, name, name_len,
5728 ino, parent_ino, &local_index);
5733 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5734 struct inode *dir, struct dentry *dentry,
5735 struct inode *inode, int backref, u64 index)
5737 int err = btrfs_add_link(trans, dir, inode,
5738 dentry->d_name.name, dentry->d_name.len,
5745 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5746 umode_t mode, dev_t rdev)
5748 struct btrfs_trans_handle *trans;
5749 struct btrfs_root *root = BTRFS_I(dir)->root;
5750 struct inode *inode = NULL;
5756 if (!new_valid_dev(rdev))
5760 * 2 for inode item and ref
5762 * 1 for xattr if selinux is on
5764 trans = btrfs_start_transaction(root, 5);
5766 return PTR_ERR(trans);
5768 err = btrfs_find_free_ino(root, &objectid);
5772 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5773 dentry->d_name.len, btrfs_ino(dir), objectid,
5775 if (IS_ERR(inode)) {
5776 err = PTR_ERR(inode);
5780 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5787 * If the active LSM wants to access the inode during
5788 * d_instantiate it needs these. Smack checks to see
5789 * if the filesystem supports xattrs by looking at the
5793 inode->i_op = &btrfs_special_inode_operations;
5794 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5798 init_special_inode(inode, inode->i_mode, rdev);
5799 btrfs_update_inode(trans, root, inode);
5800 d_instantiate(dentry, inode);
5803 btrfs_end_transaction(trans, root);
5804 btrfs_balance_delayed_items(root);
5805 btrfs_btree_balance_dirty(root);
5807 inode_dec_link_count(inode);
5813 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5814 umode_t mode, bool excl)
5816 struct btrfs_trans_handle *trans;
5817 struct btrfs_root *root = BTRFS_I(dir)->root;
5818 struct inode *inode = NULL;
5819 int drop_inode_on_err = 0;
5825 * 2 for inode item and ref
5827 * 1 for xattr if selinux is on
5829 trans = btrfs_start_transaction(root, 5);
5831 return PTR_ERR(trans);
5833 err = btrfs_find_free_ino(root, &objectid);
5837 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5838 dentry->d_name.len, btrfs_ino(dir), objectid,
5840 if (IS_ERR(inode)) {
5841 err = PTR_ERR(inode);
5844 drop_inode_on_err = 1;
5846 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5850 err = btrfs_update_inode(trans, root, inode);
5855 * If the active LSM wants to access the inode during
5856 * d_instantiate it needs these. Smack checks to see
5857 * if the filesystem supports xattrs by looking at the
5860 inode->i_fop = &btrfs_file_operations;
5861 inode->i_op = &btrfs_file_inode_operations;
5863 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5867 inode->i_mapping->a_ops = &btrfs_aops;
5868 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5869 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5870 d_instantiate(dentry, inode);
5873 btrfs_end_transaction(trans, root);
5874 if (err && drop_inode_on_err) {
5875 inode_dec_link_count(inode);
5878 btrfs_balance_delayed_items(root);
5879 btrfs_btree_balance_dirty(root);
5883 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5884 struct dentry *dentry)
5886 struct btrfs_trans_handle *trans;
5887 struct btrfs_root *root = BTRFS_I(dir)->root;
5888 struct inode *inode = old_dentry->d_inode;
5893 /* do not allow sys_link's with other subvols of the same device */
5894 if (root->objectid != BTRFS_I(inode)->root->objectid)
5897 if (inode->i_nlink >= BTRFS_LINK_MAX)
5900 err = btrfs_set_inode_index(dir, &index);
5905 * 2 items for inode and inode ref
5906 * 2 items for dir items
5907 * 1 item for parent inode
5909 trans = btrfs_start_transaction(root, 5);
5910 if (IS_ERR(trans)) {
5911 err = PTR_ERR(trans);
5915 /* There are several dir indexes for this inode, clear the cache. */
5916 BTRFS_I(inode)->dir_index = 0ULL;
5918 inode_inc_iversion(inode);
5919 inode->i_ctime = CURRENT_TIME;
5921 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5923 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5928 struct dentry *parent = dentry->d_parent;
5929 err = btrfs_update_inode(trans, root, inode);
5932 d_instantiate(dentry, inode);
5933 btrfs_log_new_name(trans, inode, NULL, parent);
5936 btrfs_end_transaction(trans, root);
5937 btrfs_balance_delayed_items(root);
5940 inode_dec_link_count(inode);
5943 btrfs_btree_balance_dirty(root);
5947 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5949 struct inode *inode = NULL;
5950 struct btrfs_trans_handle *trans;
5951 struct btrfs_root *root = BTRFS_I(dir)->root;
5953 int drop_on_err = 0;
5958 * 2 items for inode and ref
5959 * 2 items for dir items
5960 * 1 for xattr if selinux is on
5962 trans = btrfs_start_transaction(root, 5);
5964 return PTR_ERR(trans);
5966 err = btrfs_find_free_ino(root, &objectid);
5970 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5971 dentry->d_name.len, btrfs_ino(dir), objectid,
5972 S_IFDIR | mode, &index);
5973 if (IS_ERR(inode)) {
5974 err = PTR_ERR(inode);
5980 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5984 inode->i_op = &btrfs_dir_inode_operations;
5985 inode->i_fop = &btrfs_dir_file_operations;
5987 btrfs_i_size_write(inode, 0);
5988 err = btrfs_update_inode(trans, root, inode);
5992 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5993 dentry->d_name.len, 0, index);
5997 d_instantiate(dentry, inode);
6001 btrfs_end_transaction(trans, root);
6004 btrfs_balance_delayed_items(root);
6005 btrfs_btree_balance_dirty(root);
6009 /* helper for btfs_get_extent. Given an existing extent in the tree,
6010 * and an extent that you want to insert, deal with overlap and insert
6011 * the new extent into the tree.
6013 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6014 struct extent_map *existing,
6015 struct extent_map *em,
6016 u64 map_start, u64 map_len)
6020 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6021 start_diff = map_start - em->start;
6022 em->start = map_start;
6024 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6025 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6026 em->block_start += start_diff;
6027 em->block_len -= start_diff;
6029 return add_extent_mapping(em_tree, em, 0);
6032 static noinline int uncompress_inline(struct btrfs_path *path,
6033 struct inode *inode, struct page *page,
6034 size_t pg_offset, u64 extent_offset,
6035 struct btrfs_file_extent_item *item)
6038 struct extent_buffer *leaf = path->nodes[0];
6041 unsigned long inline_size;
6045 WARN_ON(pg_offset != 0);
6046 compress_type = btrfs_file_extent_compression(leaf, item);
6047 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6048 inline_size = btrfs_file_extent_inline_item_len(leaf,
6049 btrfs_item_nr(path->slots[0]));
6050 tmp = kmalloc(inline_size, GFP_NOFS);
6053 ptr = btrfs_file_extent_inline_start(item);
6055 read_extent_buffer(leaf, tmp, ptr, inline_size);
6057 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6058 ret = btrfs_decompress(compress_type, tmp, page,
6059 extent_offset, inline_size, max_size);
6061 char *kaddr = kmap_atomic(page);
6062 unsigned long copy_size = min_t(u64,
6063 PAGE_CACHE_SIZE - pg_offset,
6064 max_size - extent_offset);
6065 memset(kaddr + pg_offset, 0, copy_size);
6066 kunmap_atomic(kaddr);
6073 * a bit scary, this does extent mapping from logical file offset to the disk.
6074 * the ugly parts come from merging extents from the disk with the in-ram
6075 * representation. This gets more complex because of the data=ordered code,
6076 * where the in-ram extents might be locked pending data=ordered completion.
6078 * This also copies inline extents directly into the page.
6081 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6082 size_t pg_offset, u64 start, u64 len,
6088 u64 extent_start = 0;
6090 u64 objectid = btrfs_ino(inode);
6092 struct btrfs_path *path = NULL;
6093 struct btrfs_root *root = BTRFS_I(inode)->root;
6094 struct btrfs_file_extent_item *item;
6095 struct extent_buffer *leaf;
6096 struct btrfs_key found_key;
6097 struct extent_map *em = NULL;
6098 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6099 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6100 struct btrfs_trans_handle *trans = NULL;
6104 read_lock(&em_tree->lock);
6105 em = lookup_extent_mapping(em_tree, start, len);
6107 em->bdev = root->fs_info->fs_devices->latest_bdev;
6108 read_unlock(&em_tree->lock);
6111 if (em->start > start || em->start + em->len <= start)
6112 free_extent_map(em);
6113 else if (em->block_start == EXTENT_MAP_INLINE && page)
6114 free_extent_map(em);
6118 em = alloc_extent_map();
6123 em->bdev = root->fs_info->fs_devices->latest_bdev;
6124 em->start = EXTENT_MAP_HOLE;
6125 em->orig_start = EXTENT_MAP_HOLE;
6127 em->block_len = (u64)-1;
6130 path = btrfs_alloc_path();
6136 * Chances are we'll be called again, so go ahead and do
6142 ret = btrfs_lookup_file_extent(trans, root, path,
6143 objectid, start, trans != NULL);
6150 if (path->slots[0] == 0)
6155 leaf = path->nodes[0];
6156 item = btrfs_item_ptr(leaf, path->slots[0],
6157 struct btrfs_file_extent_item);
6158 /* are we inside the extent that was found? */
6159 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6160 found_type = btrfs_key_type(&found_key);
6161 if (found_key.objectid != objectid ||
6162 found_type != BTRFS_EXTENT_DATA_KEY) {
6164 * If we backup past the first extent we want to move forward
6165 * and see if there is an extent in front of us, otherwise we'll
6166 * say there is a hole for our whole search range which can
6173 found_type = btrfs_file_extent_type(leaf, item);
6174 extent_start = found_key.offset;
6175 compress_type = btrfs_file_extent_compression(leaf, item);
6176 if (found_type == BTRFS_FILE_EXTENT_REG ||
6177 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6178 extent_end = extent_start +
6179 btrfs_file_extent_num_bytes(leaf, item);
6180 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6182 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6183 extent_end = ALIGN(extent_start + size, root->sectorsize);
6186 if (start >= extent_end) {
6188 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6189 ret = btrfs_next_leaf(root, path);
6196 leaf = path->nodes[0];
6198 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6199 if (found_key.objectid != objectid ||
6200 found_key.type != BTRFS_EXTENT_DATA_KEY)
6202 if (start + len <= found_key.offset)
6205 em->orig_start = start;
6206 em->len = found_key.offset - start;
6210 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
6211 if (found_type == BTRFS_FILE_EXTENT_REG ||
6212 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6213 em->start = extent_start;
6214 em->len = extent_end - extent_start;
6215 em->orig_start = extent_start -
6216 btrfs_file_extent_offset(leaf, item);
6217 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6219 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6221 em->block_start = EXTENT_MAP_HOLE;
6224 if (compress_type != BTRFS_COMPRESS_NONE) {
6225 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6226 em->compress_type = compress_type;
6227 em->block_start = bytenr;
6228 em->block_len = em->orig_block_len;
6230 bytenr += btrfs_file_extent_offset(leaf, item);
6231 em->block_start = bytenr;
6232 em->block_len = em->len;
6233 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6234 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6237 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6241 size_t extent_offset;
6244 em->block_start = EXTENT_MAP_INLINE;
6245 if (!page || create) {
6246 em->start = extent_start;
6247 em->len = extent_end - extent_start;
6251 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6252 extent_offset = page_offset(page) + pg_offset - extent_start;
6253 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6254 size - extent_offset);
6255 em->start = extent_start + extent_offset;
6256 em->len = ALIGN(copy_size, root->sectorsize);
6257 em->orig_block_len = em->len;
6258 em->orig_start = em->start;
6259 if (compress_type) {
6260 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6261 em->compress_type = compress_type;
6263 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6264 if (create == 0 && !PageUptodate(page)) {
6265 if (btrfs_file_extent_compression(leaf, item) !=
6266 BTRFS_COMPRESS_NONE) {
6267 ret = uncompress_inline(path, inode, page,
6269 extent_offset, item);
6270 BUG_ON(ret); /* -ENOMEM */
6273 read_extent_buffer(leaf, map + pg_offset, ptr,
6275 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6276 memset(map + pg_offset + copy_size, 0,
6277 PAGE_CACHE_SIZE - pg_offset -
6282 flush_dcache_page(page);
6283 } else if (create && PageUptodate(page)) {
6287 free_extent_map(em);
6290 btrfs_release_path(path);
6291 trans = btrfs_join_transaction(root);
6294 return ERR_CAST(trans);
6298 write_extent_buffer(leaf, map + pg_offset, ptr,
6301 btrfs_mark_buffer_dirty(leaf);
6303 set_extent_uptodate(io_tree, em->start,
6304 extent_map_end(em) - 1, NULL, GFP_NOFS);
6307 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6311 em->orig_start = start;
6314 em->block_start = EXTENT_MAP_HOLE;
6315 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6317 btrfs_release_path(path);
6318 if (em->start > start || extent_map_end(em) <= start) {
6319 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6320 em->start, em->len, start, len);
6326 write_lock(&em_tree->lock);
6327 ret = add_extent_mapping(em_tree, em, 0);
6328 /* it is possible that someone inserted the extent into the tree
6329 * while we had the lock dropped. It is also possible that
6330 * an overlapping map exists in the tree
6332 if (ret == -EEXIST) {
6333 struct extent_map *existing;
6337 existing = lookup_extent_mapping(em_tree, start, len);
6338 if (existing && (existing->start > start ||
6339 existing->start + existing->len <= start)) {
6340 free_extent_map(existing);
6344 existing = lookup_extent_mapping(em_tree, em->start,
6347 err = merge_extent_mapping(em_tree, existing,
6350 free_extent_map(existing);
6352 free_extent_map(em);
6357 free_extent_map(em);
6361 free_extent_map(em);
6366 write_unlock(&em_tree->lock);
6369 trace_btrfs_get_extent(root, em);
6372 btrfs_free_path(path);
6374 ret = btrfs_end_transaction(trans, root);
6379 free_extent_map(em);
6380 return ERR_PTR(err);
6382 BUG_ON(!em); /* Error is always set */
6386 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6387 size_t pg_offset, u64 start, u64 len,
6390 struct extent_map *em;
6391 struct extent_map *hole_em = NULL;
6392 u64 range_start = start;
6398 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6405 * - a pre-alloc extent,
6406 * there might actually be delalloc bytes behind it.
6408 if (em->block_start != EXTENT_MAP_HOLE &&
6409 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6415 /* check to see if we've wrapped (len == -1 or similar) */
6424 /* ok, we didn't find anything, lets look for delalloc */
6425 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6426 end, len, EXTENT_DELALLOC, 1);
6427 found_end = range_start + found;
6428 if (found_end < range_start)
6429 found_end = (u64)-1;
6432 * we didn't find anything useful, return
6433 * the original results from get_extent()
6435 if (range_start > end || found_end <= start) {
6441 /* adjust the range_start to make sure it doesn't
6442 * go backwards from the start they passed in
6444 range_start = max(start, range_start);
6445 found = found_end - range_start;
6448 u64 hole_start = start;
6451 em = alloc_extent_map();
6457 * when btrfs_get_extent can't find anything it
6458 * returns one huge hole
6460 * make sure what it found really fits our range, and
6461 * adjust to make sure it is based on the start from
6465 u64 calc_end = extent_map_end(hole_em);
6467 if (calc_end <= start || (hole_em->start > end)) {
6468 free_extent_map(hole_em);
6471 hole_start = max(hole_em->start, start);
6472 hole_len = calc_end - hole_start;
6476 if (hole_em && range_start > hole_start) {
6477 /* our hole starts before our delalloc, so we
6478 * have to return just the parts of the hole
6479 * that go until the delalloc starts
6481 em->len = min(hole_len,
6482 range_start - hole_start);
6483 em->start = hole_start;
6484 em->orig_start = hole_start;
6486 * don't adjust block start at all,
6487 * it is fixed at EXTENT_MAP_HOLE
6489 em->block_start = hole_em->block_start;
6490 em->block_len = hole_len;
6491 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6492 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6494 em->start = range_start;
6496 em->orig_start = range_start;
6497 em->block_start = EXTENT_MAP_DELALLOC;
6498 em->block_len = found;
6500 } else if (hole_em) {
6505 free_extent_map(hole_em);
6507 free_extent_map(em);
6508 return ERR_PTR(err);
6513 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6516 struct btrfs_root *root = BTRFS_I(inode)->root;
6517 struct extent_map *em;
6518 struct btrfs_key ins;
6522 alloc_hint = get_extent_allocation_hint(inode, start, len);
6523 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
6524 alloc_hint, &ins, 1);
6526 return ERR_PTR(ret);
6528 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6529 ins.offset, ins.offset, ins.offset, 0);
6531 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6535 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6536 ins.offset, ins.offset, 0);
6538 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6539 free_extent_map(em);
6540 return ERR_PTR(ret);
6547 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6548 * block must be cow'd
6550 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6551 u64 *orig_start, u64 *orig_block_len,
6554 struct btrfs_trans_handle *trans;
6555 struct btrfs_path *path;
6557 struct extent_buffer *leaf;
6558 struct btrfs_root *root = BTRFS_I(inode)->root;
6559 struct btrfs_file_extent_item *fi;
6560 struct btrfs_key key;
6567 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6569 path = btrfs_alloc_path();
6573 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
6578 slot = path->slots[0];
6581 /* can't find the item, must cow */
6588 leaf = path->nodes[0];
6589 btrfs_item_key_to_cpu(leaf, &key, slot);
6590 if (key.objectid != btrfs_ino(inode) ||
6591 key.type != BTRFS_EXTENT_DATA_KEY) {
6592 /* not our file or wrong item type, must cow */
6596 if (key.offset > offset) {
6597 /* Wrong offset, must cow */
6601 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6602 found_type = btrfs_file_extent_type(leaf, fi);
6603 if (found_type != BTRFS_FILE_EXTENT_REG &&
6604 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6605 /* not a regular extent, must cow */
6609 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6612 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6613 if (extent_end <= offset)
6616 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6617 if (disk_bytenr == 0)
6620 if (btrfs_file_extent_compression(leaf, fi) ||
6621 btrfs_file_extent_encryption(leaf, fi) ||
6622 btrfs_file_extent_other_encoding(leaf, fi))
6625 backref_offset = btrfs_file_extent_offset(leaf, fi);
6628 *orig_start = key.offset - backref_offset;
6629 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6630 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6633 if (btrfs_extent_readonly(root, disk_bytenr))
6635 btrfs_release_path(path);
6638 * look for other files referencing this extent, if we
6639 * find any we must cow
6641 trans = btrfs_join_transaction(root);
6642 if (IS_ERR(trans)) {
6647 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6648 key.offset - backref_offset, disk_bytenr);
6649 btrfs_end_transaction(trans, root);
6656 * adjust disk_bytenr and num_bytes to cover just the bytes
6657 * in this extent we are about to write. If there
6658 * are any csums in that range we have to cow in order
6659 * to keep the csums correct
6661 disk_bytenr += backref_offset;
6662 disk_bytenr += offset - key.offset;
6663 num_bytes = min(offset + *len, extent_end) - offset;
6664 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6667 * all of the above have passed, it is safe to overwrite this extent
6673 btrfs_free_path(path);
6677 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6678 struct extent_state **cached_state, int writing)
6680 struct btrfs_ordered_extent *ordered;
6684 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6687 * We're concerned with the entire range that we're going to be
6688 * doing DIO to, so we need to make sure theres no ordered
6689 * extents in this range.
6691 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6692 lockend - lockstart + 1);
6695 * We need to make sure there are no buffered pages in this
6696 * range either, we could have raced between the invalidate in
6697 * generic_file_direct_write and locking the extent. The
6698 * invalidate needs to happen so that reads after a write do not
6701 if (!ordered && (!writing ||
6702 !test_range_bit(&BTRFS_I(inode)->io_tree,
6703 lockstart, lockend, EXTENT_UPTODATE, 0,
6707 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6708 cached_state, GFP_NOFS);
6711 btrfs_start_ordered_extent(inode, ordered, 1);
6712 btrfs_put_ordered_extent(ordered);
6714 /* Screw you mmap */
6715 ret = filemap_write_and_wait_range(inode->i_mapping,
6722 * If we found a page that couldn't be invalidated just
6723 * fall back to buffered.
6725 ret = invalidate_inode_pages2_range(inode->i_mapping,
6726 lockstart >> PAGE_CACHE_SHIFT,
6727 lockend >> PAGE_CACHE_SHIFT);
6738 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6739 u64 len, u64 orig_start,
6740 u64 block_start, u64 block_len,
6741 u64 orig_block_len, u64 ram_bytes,
6744 struct extent_map_tree *em_tree;
6745 struct extent_map *em;
6746 struct btrfs_root *root = BTRFS_I(inode)->root;
6749 em_tree = &BTRFS_I(inode)->extent_tree;
6750 em = alloc_extent_map();
6752 return ERR_PTR(-ENOMEM);
6755 em->orig_start = orig_start;
6756 em->mod_start = start;
6759 em->block_len = block_len;
6760 em->block_start = block_start;
6761 em->bdev = root->fs_info->fs_devices->latest_bdev;
6762 em->orig_block_len = orig_block_len;
6763 em->ram_bytes = ram_bytes;
6764 em->generation = -1;
6765 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6766 if (type == BTRFS_ORDERED_PREALLOC)
6767 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6770 btrfs_drop_extent_cache(inode, em->start,
6771 em->start + em->len - 1, 0);
6772 write_lock(&em_tree->lock);
6773 ret = add_extent_mapping(em_tree, em, 1);
6774 write_unlock(&em_tree->lock);
6775 } while (ret == -EEXIST);
6778 free_extent_map(em);
6779 return ERR_PTR(ret);
6786 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6787 struct buffer_head *bh_result, int create)
6789 struct extent_map *em;
6790 struct btrfs_root *root = BTRFS_I(inode)->root;
6791 struct extent_state *cached_state = NULL;
6792 u64 start = iblock << inode->i_blkbits;
6793 u64 lockstart, lockend;
6794 u64 len = bh_result->b_size;
6795 int unlock_bits = EXTENT_LOCKED;
6799 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6801 len = min_t(u64, len, root->sectorsize);
6804 lockend = start + len - 1;
6807 * If this errors out it's because we couldn't invalidate pagecache for
6808 * this range and we need to fallback to buffered.
6810 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6813 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6820 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6821 * io. INLINE is special, and we could probably kludge it in here, but
6822 * it's still buffered so for safety lets just fall back to the generic
6825 * For COMPRESSED we _have_ to read the entire extent in so we can
6826 * decompress it, so there will be buffering required no matter what we
6827 * do, so go ahead and fallback to buffered.
6829 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6830 * to buffered IO. Don't blame me, this is the price we pay for using
6833 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6834 em->block_start == EXTENT_MAP_INLINE) {
6835 free_extent_map(em);
6840 /* Just a good old fashioned hole, return */
6841 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6842 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6843 free_extent_map(em);
6848 * We don't allocate a new extent in the following cases
6850 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6852 * 2) The extent is marked as PREALLOC. We're good to go here and can
6853 * just use the extent.
6857 len = min(len, em->len - (start - em->start));
6858 lockstart = start + len;
6862 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6863 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6864 em->block_start != EXTENT_MAP_HOLE)) {
6867 u64 block_start, orig_start, orig_block_len, ram_bytes;
6869 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6870 type = BTRFS_ORDERED_PREALLOC;
6872 type = BTRFS_ORDERED_NOCOW;
6873 len = min(len, em->len - (start - em->start));
6874 block_start = em->block_start + (start - em->start);
6876 if (can_nocow_extent(inode, start, &len, &orig_start,
6877 &orig_block_len, &ram_bytes) == 1) {
6878 if (type == BTRFS_ORDERED_PREALLOC) {
6879 free_extent_map(em);
6880 em = create_pinned_em(inode, start, len,
6889 ret = btrfs_add_ordered_extent_dio(inode, start,
6890 block_start, len, len, type);
6892 free_extent_map(em);
6900 * this will cow the extent, reset the len in case we changed
6903 len = bh_result->b_size;
6904 free_extent_map(em);
6905 em = btrfs_new_extent_direct(inode, start, len);
6910 len = min(len, em->len - (start - em->start));
6912 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6914 bh_result->b_size = len;
6915 bh_result->b_bdev = em->bdev;
6916 set_buffer_mapped(bh_result);
6918 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6919 set_buffer_new(bh_result);
6922 * Need to update the i_size under the extent lock so buffered
6923 * readers will get the updated i_size when we unlock.
6925 if (start + len > i_size_read(inode))
6926 i_size_write(inode, start + len);
6928 spin_lock(&BTRFS_I(inode)->lock);
6929 BTRFS_I(inode)->outstanding_extents++;
6930 spin_unlock(&BTRFS_I(inode)->lock);
6932 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6933 lockstart + len - 1, EXTENT_DELALLOC, NULL,
6934 &cached_state, GFP_NOFS);
6939 * In the case of write we need to clear and unlock the entire range,
6940 * in the case of read we need to unlock only the end area that we
6941 * aren't using if there is any left over space.
6943 if (lockstart < lockend) {
6944 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6945 lockend, unlock_bits, 1, 0,
6946 &cached_state, GFP_NOFS);
6948 free_extent_state(cached_state);
6951 free_extent_map(em);
6956 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6957 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6961 static void btrfs_endio_direct_read(struct bio *bio, int err)
6963 struct btrfs_dio_private *dip = bio->bi_private;
6964 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6965 struct bio_vec *bvec = bio->bi_io_vec;
6966 struct inode *inode = dip->inode;
6967 struct btrfs_root *root = BTRFS_I(inode)->root;
6968 struct bio *dio_bio;
6969 u32 *csums = (u32 *)dip->csum;
6973 start = dip->logical_offset;
6975 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6976 struct page *page = bvec->bv_page;
6979 unsigned long flags;
6981 local_irq_save(flags);
6982 kaddr = kmap_atomic(page);
6983 csum = btrfs_csum_data(kaddr + bvec->bv_offset,
6984 csum, bvec->bv_len);
6985 btrfs_csum_final(csum, (char *)&csum);
6986 kunmap_atomic(kaddr);
6987 local_irq_restore(flags);
6989 flush_dcache_page(bvec->bv_page);
6990 if (csum != csums[index]) {
6991 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
6992 btrfs_ino(inode), start, csum,
6998 start += bvec->bv_len;
7001 } while (bvec <= bvec_end);
7003 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
7004 dip->logical_offset + dip->bytes - 1);
7005 dio_bio = dip->dio_bio;
7009 /* If we had a csum failure make sure to clear the uptodate flag */
7011 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7012 dio_end_io(dio_bio, err);
7016 static void btrfs_endio_direct_write(struct bio *bio, int err)
7018 struct btrfs_dio_private *dip = bio->bi_private;
7019 struct inode *inode = dip->inode;
7020 struct btrfs_root *root = BTRFS_I(inode)->root;
7021 struct btrfs_ordered_extent *ordered = NULL;
7022 u64 ordered_offset = dip->logical_offset;
7023 u64 ordered_bytes = dip->bytes;
7024 struct bio *dio_bio;
7030 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7032 ordered_bytes, !err);
7036 ordered->work.func = finish_ordered_fn;
7037 ordered->work.flags = 0;
7038 btrfs_queue_worker(&root->fs_info->endio_write_workers,
7042 * our bio might span multiple ordered extents. If we haven't
7043 * completed the accounting for the whole dio, go back and try again
7045 if (ordered_offset < dip->logical_offset + dip->bytes) {
7046 ordered_bytes = dip->logical_offset + dip->bytes -
7052 dio_bio = dip->dio_bio;
7056 /* If we had an error make sure to clear the uptodate flag */
7058 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7059 dio_end_io(dio_bio, err);
7063 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7064 struct bio *bio, int mirror_num,
7065 unsigned long bio_flags, u64 offset)
7068 struct btrfs_root *root = BTRFS_I(inode)->root;
7069 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7070 BUG_ON(ret); /* -ENOMEM */
7074 static void btrfs_end_dio_bio(struct bio *bio, int err)
7076 struct btrfs_dio_private *dip = bio->bi_private;
7079 btrfs_err(BTRFS_I(dip->inode)->root->fs_info,
7080 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7081 btrfs_ino(dip->inode), bio->bi_rw,
7082 (unsigned long long)bio->bi_sector, bio->bi_size, err);
7086 * before atomic variable goto zero, we must make sure
7087 * dip->errors is perceived to be set.
7089 smp_mb__before_atomic_dec();
7092 /* if there are more bios still pending for this dio, just exit */
7093 if (!atomic_dec_and_test(&dip->pending_bios))
7097 bio_io_error(dip->orig_bio);
7099 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7100 bio_endio(dip->orig_bio, 0);
7106 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7107 u64 first_sector, gfp_t gfp_flags)
7109 int nr_vecs = bio_get_nr_vecs(bdev);
7110 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7113 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7114 int rw, u64 file_offset, int skip_sum,
7117 struct btrfs_dio_private *dip = bio->bi_private;
7118 int write = rw & REQ_WRITE;
7119 struct btrfs_root *root = BTRFS_I(inode)->root;
7123 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7128 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7136 if (write && async_submit) {
7137 ret = btrfs_wq_submit_bio(root->fs_info,
7138 inode, rw, bio, 0, 0,
7140 __btrfs_submit_bio_start_direct_io,
7141 __btrfs_submit_bio_done);
7145 * If we aren't doing async submit, calculate the csum of the
7148 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7151 } else if (!skip_sum) {
7152 ret = btrfs_lookup_bio_sums_dio(root, inode, dip, bio,
7159 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7165 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7168 struct inode *inode = dip->inode;
7169 struct btrfs_root *root = BTRFS_I(inode)->root;
7171 struct bio *orig_bio = dip->orig_bio;
7172 struct bio_vec *bvec = orig_bio->bi_io_vec;
7173 u64 start_sector = orig_bio->bi_sector;
7174 u64 file_offset = dip->logical_offset;
7179 int async_submit = 0;
7181 map_length = orig_bio->bi_size;
7182 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7183 &map_length, NULL, 0);
7189 if (map_length >= orig_bio->bi_size) {
7194 /* async crcs make it difficult to collect full stripe writes. */
7195 if (btrfs_get_alloc_profile(root, 1) &
7196 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7201 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7204 bio->bi_private = dip;
7205 bio->bi_end_io = btrfs_end_dio_bio;
7206 atomic_inc(&dip->pending_bios);
7208 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7209 if (unlikely(map_length < submit_len + bvec->bv_len ||
7210 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7211 bvec->bv_offset) < bvec->bv_len)) {
7213 * inc the count before we submit the bio so
7214 * we know the end IO handler won't happen before
7215 * we inc the count. Otherwise, the dip might get freed
7216 * before we're done setting it up
7218 atomic_inc(&dip->pending_bios);
7219 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7220 file_offset, skip_sum,
7224 atomic_dec(&dip->pending_bios);
7228 start_sector += submit_len >> 9;
7229 file_offset += submit_len;
7234 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7235 start_sector, GFP_NOFS);
7238 bio->bi_private = dip;
7239 bio->bi_end_io = btrfs_end_dio_bio;
7241 map_length = orig_bio->bi_size;
7242 ret = btrfs_map_block(root->fs_info, rw,
7244 &map_length, NULL, 0);
7250 submit_len += bvec->bv_len;
7257 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7266 * before atomic variable goto zero, we must
7267 * make sure dip->errors is perceived to be set.
7269 smp_mb__before_atomic_dec();
7270 if (atomic_dec_and_test(&dip->pending_bios))
7271 bio_io_error(dip->orig_bio);
7273 /* bio_end_io() will handle error, so we needn't return it */
7277 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7278 struct inode *inode, loff_t file_offset)
7280 struct btrfs_root *root = BTRFS_I(inode)->root;
7281 struct btrfs_dio_private *dip;
7285 int write = rw & REQ_WRITE;
7289 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7291 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7297 if (!skip_sum && !write) {
7298 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
7299 sum_len = dio_bio->bi_size >> inode->i_sb->s_blocksize_bits;
7300 sum_len *= csum_size;
7305 dip = kmalloc(sizeof(*dip) + sum_len, GFP_NOFS);
7311 dip->private = dio_bio->bi_private;
7313 dip->logical_offset = file_offset;
7314 dip->bytes = dio_bio->bi_size;
7315 dip->disk_bytenr = (u64)dio_bio->bi_sector << 9;
7316 io_bio->bi_private = dip;
7318 dip->orig_bio = io_bio;
7319 dip->dio_bio = dio_bio;
7320 atomic_set(&dip->pending_bios, 0);
7323 io_bio->bi_end_io = btrfs_endio_direct_write;
7325 io_bio->bi_end_io = btrfs_endio_direct_read;
7327 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7336 * If this is a write, we need to clean up the reserved space and kill
7337 * the ordered extent.
7340 struct btrfs_ordered_extent *ordered;
7341 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7342 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7343 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7344 btrfs_free_reserved_extent(root, ordered->start,
7346 btrfs_put_ordered_extent(ordered);
7347 btrfs_put_ordered_extent(ordered);
7349 bio_endio(dio_bio, ret);
7352 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7353 const struct iovec *iov, loff_t offset,
7354 unsigned long nr_segs)
7360 unsigned blocksize_mask = root->sectorsize - 1;
7361 ssize_t retval = -EINVAL;
7362 loff_t end = offset;
7364 if (offset & blocksize_mask)
7367 /* Check the memory alignment. Blocks cannot straddle pages */
7368 for (seg = 0; seg < nr_segs; seg++) {
7369 addr = (unsigned long)iov[seg].iov_base;
7370 size = iov[seg].iov_len;
7372 if ((addr & blocksize_mask) || (size & blocksize_mask))
7375 /* If this is a write we don't need to check anymore */
7380 * Check to make sure we don't have duplicate iov_base's in this
7381 * iovec, if so return EINVAL, otherwise we'll get csum errors
7382 * when reading back.
7384 for (i = seg + 1; i < nr_segs; i++) {
7385 if (iov[seg].iov_base == iov[i].iov_base)
7394 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7395 const struct iovec *iov, loff_t offset,
7396 unsigned long nr_segs)
7398 struct file *file = iocb->ki_filp;
7399 struct inode *inode = file->f_mapping->host;
7403 bool relock = false;
7406 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7410 atomic_inc(&inode->i_dio_count);
7411 smp_mb__after_atomic_inc();
7414 * The generic stuff only does filemap_write_and_wait_range, which isn't
7415 * enough if we've written compressed pages to this area, so we need to
7416 * call btrfs_wait_ordered_range to make absolutely sure that any
7417 * outstanding dirty pages are on disk.
7419 count = iov_length(iov, nr_segs);
7420 ret = btrfs_wait_ordered_range(inode, offset, count);
7426 * If the write DIO is beyond the EOF, we need update
7427 * the isize, but it is protected by i_mutex. So we can
7428 * not unlock the i_mutex at this case.
7430 if (offset + count <= inode->i_size) {
7431 mutex_unlock(&inode->i_mutex);
7434 ret = btrfs_delalloc_reserve_space(inode, count);
7437 } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7438 &BTRFS_I(inode)->runtime_flags))) {
7439 inode_dio_done(inode);
7440 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7444 ret = __blockdev_direct_IO(rw, iocb, inode,
7445 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7446 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7447 btrfs_submit_direct, flags);
7449 if (ret < 0 && ret != -EIOCBQUEUED)
7450 btrfs_delalloc_release_space(inode, count);
7451 else if (ret >= 0 && (size_t)ret < count)
7452 btrfs_delalloc_release_space(inode,
7453 count - (size_t)ret);
7455 btrfs_delalloc_release_metadata(inode, 0);
7459 inode_dio_done(inode);
7461 mutex_lock(&inode->i_mutex);
7466 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
7468 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7469 __u64 start, __u64 len)
7473 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7477 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7480 int btrfs_readpage(struct file *file, struct page *page)
7482 struct extent_io_tree *tree;
7483 tree = &BTRFS_I(page->mapping->host)->io_tree;
7484 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7487 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7489 struct extent_io_tree *tree;
7492 if (current->flags & PF_MEMALLOC) {
7493 redirty_page_for_writepage(wbc, page);
7497 tree = &BTRFS_I(page->mapping->host)->io_tree;
7498 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7501 static int btrfs_writepages(struct address_space *mapping,
7502 struct writeback_control *wbc)
7504 struct extent_io_tree *tree;
7506 tree = &BTRFS_I(mapping->host)->io_tree;
7507 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7511 btrfs_readpages(struct file *file, struct address_space *mapping,
7512 struct list_head *pages, unsigned nr_pages)
7514 struct extent_io_tree *tree;
7515 tree = &BTRFS_I(mapping->host)->io_tree;
7516 return extent_readpages(tree, mapping, pages, nr_pages,
7519 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7521 struct extent_io_tree *tree;
7522 struct extent_map_tree *map;
7525 tree = &BTRFS_I(page->mapping->host)->io_tree;
7526 map = &BTRFS_I(page->mapping->host)->extent_tree;
7527 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7529 ClearPagePrivate(page);
7530 set_page_private(page, 0);
7531 page_cache_release(page);
7536 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7538 if (PageWriteback(page) || PageDirty(page))
7540 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7543 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
7544 unsigned int length)
7546 struct inode *inode = page->mapping->host;
7547 struct extent_io_tree *tree;
7548 struct btrfs_ordered_extent *ordered;
7549 struct extent_state *cached_state = NULL;
7550 u64 page_start = page_offset(page);
7551 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7552 int inode_evicting = inode->i_state & I_FREEING;
7555 * we have the page locked, so new writeback can't start,
7556 * and the dirty bit won't be cleared while we are here.
7558 * Wait for IO on this page so that we can safely clear
7559 * the PagePrivate2 bit and do ordered accounting
7561 wait_on_page_writeback(page);
7563 tree = &BTRFS_I(inode)->io_tree;
7565 btrfs_releasepage(page, GFP_NOFS);
7569 if (!inode_evicting)
7570 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7571 ordered = btrfs_lookup_ordered_extent(inode, page_start);
7574 * IO on this page will never be started, so we need
7575 * to account for any ordered extents now
7577 if (!inode_evicting)
7578 clear_extent_bit(tree, page_start, page_end,
7579 EXTENT_DIRTY | EXTENT_DELALLOC |
7580 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7581 EXTENT_DEFRAG, 1, 0, &cached_state,
7584 * whoever cleared the private bit is responsible
7585 * for the finish_ordered_io
7587 if (TestClearPagePrivate2(page)) {
7588 struct btrfs_ordered_inode_tree *tree;
7591 tree = &BTRFS_I(inode)->ordered_tree;
7593 spin_lock_irq(&tree->lock);
7594 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
7595 new_len = page_start - ordered->file_offset;
7596 if (new_len < ordered->truncated_len)
7597 ordered->truncated_len = new_len;
7598 spin_unlock_irq(&tree->lock);
7600 if (btrfs_dec_test_ordered_pending(inode, &ordered,
7602 PAGE_CACHE_SIZE, 1))
7603 btrfs_finish_ordered_io(ordered);
7605 btrfs_put_ordered_extent(ordered);
7606 if (!inode_evicting) {
7607 cached_state = NULL;
7608 lock_extent_bits(tree, page_start, page_end, 0,
7613 if (!inode_evicting) {
7614 clear_extent_bit(tree, page_start, page_end,
7615 EXTENT_LOCKED | EXTENT_DIRTY |
7616 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
7617 EXTENT_DEFRAG, 1, 1,
7618 &cached_state, GFP_NOFS);
7620 __btrfs_releasepage(page, GFP_NOFS);
7623 ClearPageChecked(page);
7624 if (PagePrivate(page)) {
7625 ClearPagePrivate(page);
7626 set_page_private(page, 0);
7627 page_cache_release(page);
7632 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7633 * called from a page fault handler when a page is first dirtied. Hence we must
7634 * be careful to check for EOF conditions here. We set the page up correctly
7635 * for a written page which means we get ENOSPC checking when writing into
7636 * holes and correct delalloc and unwritten extent mapping on filesystems that
7637 * support these features.
7639 * We are not allowed to take the i_mutex here so we have to play games to
7640 * protect against truncate races as the page could now be beyond EOF. Because
7641 * vmtruncate() writes the inode size before removing pages, once we have the
7642 * page lock we can determine safely if the page is beyond EOF. If it is not
7643 * beyond EOF, then the page is guaranteed safe against truncation until we
7646 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7648 struct page *page = vmf->page;
7649 struct inode *inode = file_inode(vma->vm_file);
7650 struct btrfs_root *root = BTRFS_I(inode)->root;
7651 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7652 struct btrfs_ordered_extent *ordered;
7653 struct extent_state *cached_state = NULL;
7655 unsigned long zero_start;
7662 sb_start_pagefault(inode->i_sb);
7663 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7665 ret = file_update_time(vma->vm_file);
7671 else /* -ENOSPC, -EIO, etc */
7672 ret = VM_FAULT_SIGBUS;
7678 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7681 size = i_size_read(inode);
7682 page_start = page_offset(page);
7683 page_end = page_start + PAGE_CACHE_SIZE - 1;
7685 if ((page->mapping != inode->i_mapping) ||
7686 (page_start >= size)) {
7687 /* page got truncated out from underneath us */
7690 wait_on_page_writeback(page);
7692 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7693 set_page_extent_mapped(page);
7696 * we can't set the delalloc bits if there are pending ordered
7697 * extents. Drop our locks and wait for them to finish
7699 ordered = btrfs_lookup_ordered_extent(inode, page_start);
7701 unlock_extent_cached(io_tree, page_start, page_end,
7702 &cached_state, GFP_NOFS);
7704 btrfs_start_ordered_extent(inode, ordered, 1);
7705 btrfs_put_ordered_extent(ordered);
7710 * XXX - page_mkwrite gets called every time the page is dirtied, even
7711 * if it was already dirty, so for space accounting reasons we need to
7712 * clear any delalloc bits for the range we are fixing to save. There
7713 * is probably a better way to do this, but for now keep consistent with
7714 * prepare_pages in the normal write path.
7716 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7717 EXTENT_DIRTY | EXTENT_DELALLOC |
7718 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7719 0, 0, &cached_state, GFP_NOFS);
7721 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7724 unlock_extent_cached(io_tree, page_start, page_end,
7725 &cached_state, GFP_NOFS);
7726 ret = VM_FAULT_SIGBUS;
7731 /* page is wholly or partially inside EOF */
7732 if (page_start + PAGE_CACHE_SIZE > size)
7733 zero_start = size & ~PAGE_CACHE_MASK;
7735 zero_start = PAGE_CACHE_SIZE;
7737 if (zero_start != PAGE_CACHE_SIZE) {
7739 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7740 flush_dcache_page(page);
7743 ClearPageChecked(page);
7744 set_page_dirty(page);
7745 SetPageUptodate(page);
7747 BTRFS_I(inode)->last_trans = root->fs_info->generation;
7748 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7749 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7751 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7755 sb_end_pagefault(inode->i_sb);
7756 return VM_FAULT_LOCKED;
7760 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7762 sb_end_pagefault(inode->i_sb);
7766 static int btrfs_truncate(struct inode *inode)
7768 struct btrfs_root *root = BTRFS_I(inode)->root;
7769 struct btrfs_block_rsv *rsv;
7772 struct btrfs_trans_handle *trans;
7773 u64 mask = root->sectorsize - 1;
7774 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7776 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
7782 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
7783 * 3 things going on here
7785 * 1) We need to reserve space for our orphan item and the space to
7786 * delete our orphan item. Lord knows we don't want to have a dangling
7787 * orphan item because we didn't reserve space to remove it.
7789 * 2) We need to reserve space to update our inode.
7791 * 3) We need to have something to cache all the space that is going to
7792 * be free'd up by the truncate operation, but also have some slack
7793 * space reserved in case it uses space during the truncate (thank you
7794 * very much snapshotting).
7796 * And we need these to all be seperate. The fact is we can use alot of
7797 * space doing the truncate, and we have no earthly idea how much space
7798 * we will use, so we need the truncate reservation to be seperate so it
7799 * doesn't end up using space reserved for updating the inode or
7800 * removing the orphan item. We also need to be able to stop the
7801 * transaction and start a new one, which means we need to be able to
7802 * update the inode several times, and we have no idea of knowing how
7803 * many times that will be, so we can't just reserve 1 item for the
7804 * entirety of the opration, so that has to be done seperately as well.
7805 * Then there is the orphan item, which does indeed need to be held on
7806 * to for the whole operation, and we need nobody to touch this reserved
7807 * space except the orphan code.
7809 * So that leaves us with
7811 * 1) root->orphan_block_rsv - for the orphan deletion.
7812 * 2) rsv - for the truncate reservation, which we will steal from the
7813 * transaction reservation.
7814 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7815 * updating the inode.
7817 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7820 rsv->size = min_size;
7824 * 1 for the truncate slack space
7825 * 1 for updating the inode.
7827 trans = btrfs_start_transaction(root, 2);
7828 if (IS_ERR(trans)) {
7829 err = PTR_ERR(trans);
7833 /* Migrate the slack space for the truncate to our reserve */
7834 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7839 * setattr is responsible for setting the ordered_data_close flag,
7840 * but that is only tested during the last file release. That
7841 * could happen well after the next commit, leaving a great big
7842 * window where new writes may get lost if someone chooses to write
7843 * to this file after truncating to zero
7845 * The inode doesn't have any dirty data here, and so if we commit
7846 * this is a noop. If someone immediately starts writing to the inode
7847 * it is very likely we'll catch some of their writes in this
7848 * transaction, and the commit will find this file on the ordered
7849 * data list with good things to send down.
7851 * This is a best effort solution, there is still a window where
7852 * using truncate to replace the contents of the file will
7853 * end up with a zero length file after a crash.
7855 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7856 &BTRFS_I(inode)->runtime_flags))
7857 btrfs_add_ordered_operation(trans, root, inode);
7860 * So if we truncate and then write and fsync we normally would just
7861 * write the extents that changed, which is a problem if we need to
7862 * first truncate that entire inode. So set this flag so we write out
7863 * all of the extents in the inode to the sync log so we're completely
7866 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7867 trans->block_rsv = rsv;
7870 ret = btrfs_truncate_inode_items(trans, root, inode,
7872 BTRFS_EXTENT_DATA_KEY);
7873 if (ret != -ENOSPC) {
7878 trans->block_rsv = &root->fs_info->trans_block_rsv;
7879 ret = btrfs_update_inode(trans, root, inode);
7885 btrfs_end_transaction(trans, root);
7886 btrfs_btree_balance_dirty(root);
7888 trans = btrfs_start_transaction(root, 2);
7889 if (IS_ERR(trans)) {
7890 ret = err = PTR_ERR(trans);
7895 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7897 BUG_ON(ret); /* shouldn't happen */
7898 trans->block_rsv = rsv;
7901 if (ret == 0 && inode->i_nlink > 0) {
7902 trans->block_rsv = root->orphan_block_rsv;
7903 ret = btrfs_orphan_del(trans, inode);
7909 trans->block_rsv = &root->fs_info->trans_block_rsv;
7910 ret = btrfs_update_inode(trans, root, inode);
7914 ret = btrfs_end_transaction(trans, root);
7915 btrfs_btree_balance_dirty(root);
7919 btrfs_free_block_rsv(root, rsv);
7928 * create a new subvolume directory/inode (helper for the ioctl).
7930 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7931 struct btrfs_root *new_root,
7932 struct btrfs_root *parent_root,
7935 struct inode *inode;
7939 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7940 new_dirid, new_dirid,
7941 S_IFDIR | (~current_umask() & S_IRWXUGO),
7944 return PTR_ERR(inode);
7945 inode->i_op = &btrfs_dir_inode_operations;
7946 inode->i_fop = &btrfs_dir_file_operations;
7948 set_nlink(inode, 1);
7949 btrfs_i_size_write(inode, 0);
7951 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
7953 btrfs_err(new_root->fs_info,
7954 "error inheriting subvolume %llu properties: %d\n",
7955 new_root->root_key.objectid, err);
7957 err = btrfs_update_inode(trans, new_root, inode);
7963 struct inode *btrfs_alloc_inode(struct super_block *sb)
7965 struct btrfs_inode *ei;
7966 struct inode *inode;
7968 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7975 ei->last_sub_trans = 0;
7976 ei->logged_trans = 0;
7977 ei->delalloc_bytes = 0;
7978 ei->disk_i_size = 0;
7981 ei->index_cnt = (u64)-1;
7983 ei->last_unlink_trans = 0;
7984 ei->last_log_commit = 0;
7986 spin_lock_init(&ei->lock);
7987 ei->outstanding_extents = 0;
7988 ei->reserved_extents = 0;
7990 ei->runtime_flags = 0;
7991 ei->force_compress = BTRFS_COMPRESS_NONE;
7993 ei->delayed_node = NULL;
7995 inode = &ei->vfs_inode;
7996 extent_map_tree_init(&ei->extent_tree);
7997 extent_io_tree_init(&ei->io_tree, &inode->i_data);
7998 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7999 ei->io_tree.track_uptodate = 1;
8000 ei->io_failure_tree.track_uptodate = 1;
8001 atomic_set(&ei->sync_writers, 0);
8002 mutex_init(&ei->log_mutex);
8003 mutex_init(&ei->delalloc_mutex);
8004 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8005 INIT_LIST_HEAD(&ei->delalloc_inodes);
8006 INIT_LIST_HEAD(&ei->ordered_operations);
8007 RB_CLEAR_NODE(&ei->rb_node);
8012 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8013 void btrfs_test_destroy_inode(struct inode *inode)
8015 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8016 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8020 static void btrfs_i_callback(struct rcu_head *head)
8022 struct inode *inode = container_of(head, struct inode, i_rcu);
8023 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8026 void btrfs_destroy_inode(struct inode *inode)
8028 struct btrfs_ordered_extent *ordered;
8029 struct btrfs_root *root = BTRFS_I(inode)->root;
8031 WARN_ON(!hlist_empty(&inode->i_dentry));
8032 WARN_ON(inode->i_data.nrpages);
8033 WARN_ON(BTRFS_I(inode)->outstanding_extents);
8034 WARN_ON(BTRFS_I(inode)->reserved_extents);
8035 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8036 WARN_ON(BTRFS_I(inode)->csum_bytes);
8039 * This can happen where we create an inode, but somebody else also
8040 * created the same inode and we need to destroy the one we already
8047 * Make sure we're properly removed from the ordered operation
8051 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
8052 spin_lock(&root->fs_info->ordered_root_lock);
8053 list_del_init(&BTRFS_I(inode)->ordered_operations);
8054 spin_unlock(&root->fs_info->ordered_root_lock);
8057 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8058 &BTRFS_I(inode)->runtime_flags)) {
8059 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
8061 atomic_dec(&root->orphan_inodes);
8065 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8069 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
8070 ordered->file_offset, ordered->len);
8071 btrfs_remove_ordered_extent(inode, ordered);
8072 btrfs_put_ordered_extent(ordered);
8073 btrfs_put_ordered_extent(ordered);
8076 inode_tree_del(inode);
8077 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8079 call_rcu(&inode->i_rcu, btrfs_i_callback);
8082 int btrfs_drop_inode(struct inode *inode)
8084 struct btrfs_root *root = BTRFS_I(inode)->root;
8089 /* the snap/subvol tree is on deleting */
8090 if (btrfs_root_refs(&root->root_item) == 0)
8093 return generic_drop_inode(inode);
8096 static void init_once(void *foo)
8098 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8100 inode_init_once(&ei->vfs_inode);
8103 void btrfs_destroy_cachep(void)
8106 * Make sure all delayed rcu free inodes are flushed before we
8110 if (btrfs_inode_cachep)
8111 kmem_cache_destroy(btrfs_inode_cachep);
8112 if (btrfs_trans_handle_cachep)
8113 kmem_cache_destroy(btrfs_trans_handle_cachep);
8114 if (btrfs_transaction_cachep)
8115 kmem_cache_destroy(btrfs_transaction_cachep);
8116 if (btrfs_path_cachep)
8117 kmem_cache_destroy(btrfs_path_cachep);
8118 if (btrfs_free_space_cachep)
8119 kmem_cache_destroy(btrfs_free_space_cachep);
8120 if (btrfs_delalloc_work_cachep)
8121 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8124 int btrfs_init_cachep(void)
8126 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8127 sizeof(struct btrfs_inode), 0,
8128 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8129 if (!btrfs_inode_cachep)
8132 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8133 sizeof(struct btrfs_trans_handle), 0,
8134 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8135 if (!btrfs_trans_handle_cachep)
8138 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8139 sizeof(struct btrfs_transaction), 0,
8140 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8141 if (!btrfs_transaction_cachep)
8144 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8145 sizeof(struct btrfs_path), 0,
8146 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8147 if (!btrfs_path_cachep)
8150 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8151 sizeof(struct btrfs_free_space), 0,
8152 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8153 if (!btrfs_free_space_cachep)
8156 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8157 sizeof(struct btrfs_delalloc_work), 0,
8158 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8160 if (!btrfs_delalloc_work_cachep)
8165 btrfs_destroy_cachep();
8169 static int btrfs_getattr(struct vfsmount *mnt,
8170 struct dentry *dentry, struct kstat *stat)
8173 struct inode *inode = dentry->d_inode;
8174 u32 blocksize = inode->i_sb->s_blocksize;
8176 generic_fillattr(inode, stat);
8177 stat->dev = BTRFS_I(inode)->root->anon_dev;
8178 stat->blksize = PAGE_CACHE_SIZE;
8180 spin_lock(&BTRFS_I(inode)->lock);
8181 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8182 spin_unlock(&BTRFS_I(inode)->lock);
8183 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8184 ALIGN(delalloc_bytes, blocksize)) >> 9;
8188 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8189 struct inode *new_dir, struct dentry *new_dentry)
8191 struct btrfs_trans_handle *trans;
8192 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8193 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8194 struct inode *new_inode = new_dentry->d_inode;
8195 struct inode *old_inode = old_dentry->d_inode;
8196 struct timespec ctime = CURRENT_TIME;
8200 u64 old_ino = btrfs_ino(old_inode);
8202 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8205 /* we only allow rename subvolume link between subvolumes */
8206 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8209 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8210 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8213 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8214 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8218 /* check for collisions, even if the name isn't there */
8219 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
8220 new_dentry->d_name.name,
8221 new_dentry->d_name.len);
8224 if (ret == -EEXIST) {
8226 * eexist without a new_inode */
8227 if (WARN_ON(!new_inode)) {
8231 /* maybe -EOVERFLOW */
8238 * we're using rename to replace one file with another.
8239 * and the replacement file is large. Start IO on it now so
8240 * we don't add too much work to the end of the transaction
8242 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8243 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8244 filemap_flush(old_inode->i_mapping);
8246 /* close the racy window with snapshot create/destroy ioctl */
8247 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8248 down_read(&root->fs_info->subvol_sem);
8250 * We want to reserve the absolute worst case amount of items. So if
8251 * both inodes are subvols and we need to unlink them then that would
8252 * require 4 item modifications, but if they are both normal inodes it
8253 * would require 5 item modifications, so we'll assume their normal
8254 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8255 * should cover the worst case number of items we'll modify.
8257 trans = btrfs_start_transaction(root, 11);
8258 if (IS_ERR(trans)) {
8259 ret = PTR_ERR(trans);
8264 btrfs_record_root_in_trans(trans, dest);
8266 ret = btrfs_set_inode_index(new_dir, &index);
8270 BTRFS_I(old_inode)->dir_index = 0ULL;
8271 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8272 /* force full log commit if subvolume involved. */
8273 root->fs_info->last_trans_log_full_commit = trans->transid;
8275 ret = btrfs_insert_inode_ref(trans, dest,
8276 new_dentry->d_name.name,
8277 new_dentry->d_name.len,
8279 btrfs_ino(new_dir), index);
8283 * this is an ugly little race, but the rename is required
8284 * to make sure that if we crash, the inode is either at the
8285 * old name or the new one. pinning the log transaction lets
8286 * us make sure we don't allow a log commit to come in after
8287 * we unlink the name but before we add the new name back in.
8289 btrfs_pin_log_trans(root);
8292 * make sure the inode gets flushed if it is replacing
8295 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8296 btrfs_add_ordered_operation(trans, root, old_inode);
8298 inode_inc_iversion(old_dir);
8299 inode_inc_iversion(new_dir);
8300 inode_inc_iversion(old_inode);
8301 old_dir->i_ctime = old_dir->i_mtime = ctime;
8302 new_dir->i_ctime = new_dir->i_mtime = ctime;
8303 old_inode->i_ctime = ctime;
8305 if (old_dentry->d_parent != new_dentry->d_parent)
8306 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8308 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8309 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8310 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8311 old_dentry->d_name.name,
8312 old_dentry->d_name.len);
8314 ret = __btrfs_unlink_inode(trans, root, old_dir,
8315 old_dentry->d_inode,
8316 old_dentry->d_name.name,
8317 old_dentry->d_name.len);
8319 ret = btrfs_update_inode(trans, root, old_inode);
8322 btrfs_abort_transaction(trans, root, ret);
8327 inode_inc_iversion(new_inode);
8328 new_inode->i_ctime = CURRENT_TIME;
8329 if (unlikely(btrfs_ino(new_inode) ==
8330 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8331 root_objectid = BTRFS_I(new_inode)->location.objectid;
8332 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8334 new_dentry->d_name.name,
8335 new_dentry->d_name.len);
8336 BUG_ON(new_inode->i_nlink == 0);
8338 ret = btrfs_unlink_inode(trans, dest, new_dir,
8339 new_dentry->d_inode,
8340 new_dentry->d_name.name,
8341 new_dentry->d_name.len);
8343 if (!ret && new_inode->i_nlink == 0)
8344 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8346 btrfs_abort_transaction(trans, root, ret);
8351 ret = btrfs_add_link(trans, new_dir, old_inode,
8352 new_dentry->d_name.name,
8353 new_dentry->d_name.len, 0, index);
8355 btrfs_abort_transaction(trans, root, ret);
8359 if (old_inode->i_nlink == 1)
8360 BTRFS_I(old_inode)->dir_index = index;
8362 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8363 struct dentry *parent = new_dentry->d_parent;
8364 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8365 btrfs_end_log_trans(root);
8368 btrfs_end_transaction(trans, root);
8370 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8371 up_read(&root->fs_info->subvol_sem);
8376 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8378 struct btrfs_delalloc_work *delalloc_work;
8379 struct inode *inode;
8381 delalloc_work = container_of(work, struct btrfs_delalloc_work,
8383 inode = delalloc_work->inode;
8384 if (delalloc_work->wait) {
8385 btrfs_wait_ordered_range(inode, 0, (u64)-1);
8387 filemap_flush(inode->i_mapping);
8388 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8389 &BTRFS_I(inode)->runtime_flags))
8390 filemap_flush(inode->i_mapping);
8393 if (delalloc_work->delay_iput)
8394 btrfs_add_delayed_iput(inode);
8397 complete(&delalloc_work->completion);
8400 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8401 int wait, int delay_iput)
8403 struct btrfs_delalloc_work *work;
8405 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8409 init_completion(&work->completion);
8410 INIT_LIST_HEAD(&work->list);
8411 work->inode = inode;
8413 work->delay_iput = delay_iput;
8414 work->work.func = btrfs_run_delalloc_work;
8419 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8421 wait_for_completion(&work->completion);
8422 kmem_cache_free(btrfs_delalloc_work_cachep, work);
8426 * some fairly slow code that needs optimization. This walks the list
8427 * of all the inodes with pending delalloc and forces them to disk.
8429 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8431 struct btrfs_inode *binode;
8432 struct inode *inode;
8433 struct btrfs_delalloc_work *work, *next;
8434 struct list_head works;
8435 struct list_head splice;
8438 INIT_LIST_HEAD(&works);
8439 INIT_LIST_HEAD(&splice);
8441 spin_lock(&root->delalloc_lock);
8442 list_splice_init(&root->delalloc_inodes, &splice);
8443 while (!list_empty(&splice)) {
8444 binode = list_entry(splice.next, struct btrfs_inode,
8447 list_move_tail(&binode->delalloc_inodes,
8448 &root->delalloc_inodes);
8449 inode = igrab(&binode->vfs_inode);
8451 cond_resched_lock(&root->delalloc_lock);
8454 spin_unlock(&root->delalloc_lock);
8456 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8457 if (unlikely(!work)) {
8459 btrfs_add_delayed_iput(inode);
8465 list_add_tail(&work->list, &works);
8466 btrfs_queue_worker(&root->fs_info->flush_workers,
8470 spin_lock(&root->delalloc_lock);
8472 spin_unlock(&root->delalloc_lock);
8474 list_for_each_entry_safe(work, next, &works, list) {
8475 list_del_init(&work->list);
8476 btrfs_wait_and_free_delalloc_work(work);
8480 list_for_each_entry_safe(work, next, &works, list) {
8481 list_del_init(&work->list);
8482 btrfs_wait_and_free_delalloc_work(work);
8485 if (!list_empty_careful(&splice)) {
8486 spin_lock(&root->delalloc_lock);
8487 list_splice_tail(&splice, &root->delalloc_inodes);
8488 spin_unlock(&root->delalloc_lock);
8493 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8497 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
8500 ret = __start_delalloc_inodes(root, delay_iput);
8502 * the filemap_flush will queue IO into the worker threads, but
8503 * we have to make sure the IO is actually started and that
8504 * ordered extents get created before we return
8506 atomic_inc(&root->fs_info->async_submit_draining);
8507 while (atomic_read(&root->fs_info->nr_async_submits) ||
8508 atomic_read(&root->fs_info->async_delalloc_pages)) {
8509 wait_event(root->fs_info->async_submit_wait,
8510 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8511 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8513 atomic_dec(&root->fs_info->async_submit_draining);
8517 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput)
8519 struct btrfs_root *root;
8520 struct list_head splice;
8523 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
8526 INIT_LIST_HEAD(&splice);
8528 spin_lock(&fs_info->delalloc_root_lock);
8529 list_splice_init(&fs_info->delalloc_roots, &splice);
8530 while (!list_empty(&splice)) {
8531 root = list_first_entry(&splice, struct btrfs_root,
8533 root = btrfs_grab_fs_root(root);
8535 list_move_tail(&root->delalloc_root,
8536 &fs_info->delalloc_roots);
8537 spin_unlock(&fs_info->delalloc_root_lock);
8539 ret = __start_delalloc_inodes(root, delay_iput);
8540 btrfs_put_fs_root(root);
8544 spin_lock(&fs_info->delalloc_root_lock);
8546 spin_unlock(&fs_info->delalloc_root_lock);
8548 atomic_inc(&fs_info->async_submit_draining);
8549 while (atomic_read(&fs_info->nr_async_submits) ||
8550 atomic_read(&fs_info->async_delalloc_pages)) {
8551 wait_event(fs_info->async_submit_wait,
8552 (atomic_read(&fs_info->nr_async_submits) == 0 &&
8553 atomic_read(&fs_info->async_delalloc_pages) == 0));
8555 atomic_dec(&fs_info->async_submit_draining);
8558 if (!list_empty_careful(&splice)) {
8559 spin_lock(&fs_info->delalloc_root_lock);
8560 list_splice_tail(&splice, &fs_info->delalloc_roots);
8561 spin_unlock(&fs_info->delalloc_root_lock);
8566 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8567 const char *symname)
8569 struct btrfs_trans_handle *trans;
8570 struct btrfs_root *root = BTRFS_I(dir)->root;
8571 struct btrfs_path *path;
8572 struct btrfs_key key;
8573 struct inode *inode = NULL;
8581 struct btrfs_file_extent_item *ei;
8582 struct extent_buffer *leaf;
8584 name_len = strlen(symname);
8585 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8586 return -ENAMETOOLONG;
8589 * 2 items for inode item and ref
8590 * 2 items for dir items
8591 * 1 item for xattr if selinux is on
8593 trans = btrfs_start_transaction(root, 5);
8595 return PTR_ERR(trans);
8597 err = btrfs_find_free_ino(root, &objectid);
8601 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8602 dentry->d_name.len, btrfs_ino(dir), objectid,
8603 S_IFLNK|S_IRWXUGO, &index);
8604 if (IS_ERR(inode)) {
8605 err = PTR_ERR(inode);
8609 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8616 * If the active LSM wants to access the inode during
8617 * d_instantiate it needs these. Smack checks to see
8618 * if the filesystem supports xattrs by looking at the
8621 inode->i_fop = &btrfs_file_operations;
8622 inode->i_op = &btrfs_file_inode_operations;
8624 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8628 inode->i_mapping->a_ops = &btrfs_aops;
8629 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8630 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8635 path = btrfs_alloc_path();
8641 key.objectid = btrfs_ino(inode);
8643 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8644 datasize = btrfs_file_extent_calc_inline_size(name_len);
8645 err = btrfs_insert_empty_item(trans, root, path, &key,
8649 btrfs_free_path(path);
8652 leaf = path->nodes[0];
8653 ei = btrfs_item_ptr(leaf, path->slots[0],
8654 struct btrfs_file_extent_item);
8655 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8656 btrfs_set_file_extent_type(leaf, ei,
8657 BTRFS_FILE_EXTENT_INLINE);
8658 btrfs_set_file_extent_encryption(leaf, ei, 0);
8659 btrfs_set_file_extent_compression(leaf, ei, 0);
8660 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8661 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8663 ptr = btrfs_file_extent_inline_start(ei);
8664 write_extent_buffer(leaf, symname, ptr, name_len);
8665 btrfs_mark_buffer_dirty(leaf);
8666 btrfs_free_path(path);
8668 inode->i_op = &btrfs_symlink_inode_operations;
8669 inode->i_mapping->a_ops = &btrfs_symlink_aops;
8670 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8671 inode_set_bytes(inode, name_len);
8672 btrfs_i_size_write(inode, name_len);
8673 err = btrfs_update_inode(trans, root, inode);
8679 d_instantiate(dentry, inode);
8680 btrfs_end_transaction(trans, root);
8682 inode_dec_link_count(inode);
8685 btrfs_btree_balance_dirty(root);
8689 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8690 u64 start, u64 num_bytes, u64 min_size,
8691 loff_t actual_len, u64 *alloc_hint,
8692 struct btrfs_trans_handle *trans)
8694 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8695 struct extent_map *em;
8696 struct btrfs_root *root = BTRFS_I(inode)->root;
8697 struct btrfs_key ins;
8698 u64 cur_offset = start;
8702 bool own_trans = true;
8706 while (num_bytes > 0) {
8708 trans = btrfs_start_transaction(root, 3);
8709 if (IS_ERR(trans)) {
8710 ret = PTR_ERR(trans);
8715 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8716 cur_bytes = max(cur_bytes, min_size);
8717 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
8718 *alloc_hint, &ins, 1);
8721 btrfs_end_transaction(trans, root);
8725 ret = insert_reserved_file_extent(trans, inode,
8726 cur_offset, ins.objectid,
8727 ins.offset, ins.offset,
8728 ins.offset, 0, 0, 0,
8729 BTRFS_FILE_EXTENT_PREALLOC);
8731 btrfs_free_reserved_extent(root, ins.objectid,
8733 btrfs_abort_transaction(trans, root, ret);
8735 btrfs_end_transaction(trans, root);
8738 btrfs_drop_extent_cache(inode, cur_offset,
8739 cur_offset + ins.offset -1, 0);
8741 em = alloc_extent_map();
8743 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8744 &BTRFS_I(inode)->runtime_flags);
8748 em->start = cur_offset;
8749 em->orig_start = cur_offset;
8750 em->len = ins.offset;
8751 em->block_start = ins.objectid;
8752 em->block_len = ins.offset;
8753 em->orig_block_len = ins.offset;
8754 em->ram_bytes = ins.offset;
8755 em->bdev = root->fs_info->fs_devices->latest_bdev;
8756 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8757 em->generation = trans->transid;
8760 write_lock(&em_tree->lock);
8761 ret = add_extent_mapping(em_tree, em, 1);
8762 write_unlock(&em_tree->lock);
8765 btrfs_drop_extent_cache(inode, cur_offset,
8766 cur_offset + ins.offset - 1,
8769 free_extent_map(em);
8771 num_bytes -= ins.offset;
8772 cur_offset += ins.offset;
8773 *alloc_hint = ins.objectid + ins.offset;
8775 inode_inc_iversion(inode);
8776 inode->i_ctime = CURRENT_TIME;
8777 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8778 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8779 (actual_len > inode->i_size) &&
8780 (cur_offset > inode->i_size)) {
8781 if (cur_offset > actual_len)
8782 i_size = actual_len;
8784 i_size = cur_offset;
8785 i_size_write(inode, i_size);
8786 btrfs_ordered_update_i_size(inode, i_size, NULL);
8789 ret = btrfs_update_inode(trans, root, inode);
8792 btrfs_abort_transaction(trans, root, ret);
8794 btrfs_end_transaction(trans, root);
8799 btrfs_end_transaction(trans, root);
8804 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8805 u64 start, u64 num_bytes, u64 min_size,
8806 loff_t actual_len, u64 *alloc_hint)
8808 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8809 min_size, actual_len, alloc_hint,
8813 int btrfs_prealloc_file_range_trans(struct inode *inode,
8814 struct btrfs_trans_handle *trans, int mode,
8815 u64 start, u64 num_bytes, u64 min_size,
8816 loff_t actual_len, u64 *alloc_hint)
8818 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8819 min_size, actual_len, alloc_hint, trans);
8822 static int btrfs_set_page_dirty(struct page *page)
8824 return __set_page_dirty_nobuffers(page);
8827 static int btrfs_permission(struct inode *inode, int mask)
8829 struct btrfs_root *root = BTRFS_I(inode)->root;
8830 umode_t mode = inode->i_mode;
8832 if (mask & MAY_WRITE &&
8833 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8834 if (btrfs_root_readonly(root))
8836 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8839 return generic_permission(inode, mask);
8842 static const struct inode_operations btrfs_dir_inode_operations = {
8843 .getattr = btrfs_getattr,
8844 .lookup = btrfs_lookup,
8845 .create = btrfs_create,
8846 .unlink = btrfs_unlink,
8848 .mkdir = btrfs_mkdir,
8849 .rmdir = btrfs_rmdir,
8850 .rename = btrfs_rename,
8851 .symlink = btrfs_symlink,
8852 .setattr = btrfs_setattr,
8853 .mknod = btrfs_mknod,
8854 .setxattr = btrfs_setxattr,
8855 .getxattr = btrfs_getxattr,
8856 .listxattr = btrfs_listxattr,
8857 .removexattr = btrfs_removexattr,
8858 .permission = btrfs_permission,
8859 .get_acl = btrfs_get_acl,
8860 .update_time = btrfs_update_time,
8862 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8863 .lookup = btrfs_lookup,
8864 .permission = btrfs_permission,
8865 .get_acl = btrfs_get_acl,
8866 .update_time = btrfs_update_time,
8869 static const struct file_operations btrfs_dir_file_operations = {
8870 .llseek = generic_file_llseek,
8871 .read = generic_read_dir,
8872 .iterate = btrfs_real_readdir,
8873 .unlocked_ioctl = btrfs_ioctl,
8874 #ifdef CONFIG_COMPAT
8875 .compat_ioctl = btrfs_ioctl,
8877 .release = btrfs_release_file,
8878 .fsync = btrfs_sync_file,
8881 static struct extent_io_ops btrfs_extent_io_ops = {
8882 .fill_delalloc = run_delalloc_range,
8883 .submit_bio_hook = btrfs_submit_bio_hook,
8884 .merge_bio_hook = btrfs_merge_bio_hook,
8885 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
8886 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
8887 .writepage_start_hook = btrfs_writepage_start_hook,
8888 .set_bit_hook = btrfs_set_bit_hook,
8889 .clear_bit_hook = btrfs_clear_bit_hook,
8890 .merge_extent_hook = btrfs_merge_extent_hook,
8891 .split_extent_hook = btrfs_split_extent_hook,
8895 * btrfs doesn't support the bmap operation because swapfiles
8896 * use bmap to make a mapping of extents in the file. They assume
8897 * these extents won't change over the life of the file and they
8898 * use the bmap result to do IO directly to the drive.
8900 * the btrfs bmap call would return logical addresses that aren't
8901 * suitable for IO and they also will change frequently as COW
8902 * operations happen. So, swapfile + btrfs == corruption.
8904 * For now we're avoiding this by dropping bmap.
8906 static const struct address_space_operations btrfs_aops = {
8907 .readpage = btrfs_readpage,
8908 .writepage = btrfs_writepage,
8909 .writepages = btrfs_writepages,
8910 .readpages = btrfs_readpages,
8911 .direct_IO = btrfs_direct_IO,
8912 .invalidatepage = btrfs_invalidatepage,
8913 .releasepage = btrfs_releasepage,
8914 .set_page_dirty = btrfs_set_page_dirty,
8915 .error_remove_page = generic_error_remove_page,
8918 static const struct address_space_operations btrfs_symlink_aops = {
8919 .readpage = btrfs_readpage,
8920 .writepage = btrfs_writepage,
8921 .invalidatepage = btrfs_invalidatepage,
8922 .releasepage = btrfs_releasepage,
8925 static const struct inode_operations btrfs_file_inode_operations = {
8926 .getattr = btrfs_getattr,
8927 .setattr = btrfs_setattr,
8928 .setxattr = btrfs_setxattr,
8929 .getxattr = btrfs_getxattr,
8930 .listxattr = btrfs_listxattr,
8931 .removexattr = btrfs_removexattr,
8932 .permission = btrfs_permission,
8933 .fiemap = btrfs_fiemap,
8934 .get_acl = btrfs_get_acl,
8935 .update_time = btrfs_update_time,
8937 static const struct inode_operations btrfs_special_inode_operations = {
8938 .getattr = btrfs_getattr,
8939 .setattr = btrfs_setattr,
8940 .permission = btrfs_permission,
8941 .setxattr = btrfs_setxattr,
8942 .getxattr = btrfs_getxattr,
8943 .listxattr = btrfs_listxattr,
8944 .removexattr = btrfs_removexattr,
8945 .get_acl = btrfs_get_acl,
8946 .update_time = btrfs_update_time,
8948 static const struct inode_operations btrfs_symlink_inode_operations = {
8949 .readlink = generic_readlink,
8950 .follow_link = page_follow_link_light,
8951 .put_link = page_put_link,
8952 .getattr = btrfs_getattr,
8953 .setattr = btrfs_setattr,
8954 .permission = btrfs_permission,
8955 .setxattr = btrfs_setxattr,
8956 .getxattr = btrfs_getxattr,
8957 .listxattr = btrfs_listxattr,
8958 .removexattr = btrfs_removexattr,
8959 .get_acl = btrfs_get_acl,
8960 .update_time = btrfs_update_time,
8963 const struct dentry_operations btrfs_dentry_operations = {
8964 .d_delete = btrfs_dentry_delete,
8965 .d_release = btrfs_dentry_release,