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>
49 #include "transaction.h"
50 #include "btrfs_inode.h"
51 #include "print-tree.h"
52 #include "ordered-data.h"
56 #include "compression.h"
58 #include "free-space-cache.h"
59 #include "inode-map.h"
63 struct btrfs_iget_args {
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_root *root, struct inode *inode,
130 u64 start, size_t size, size_t compressed_size,
132 struct page **compressed_pages)
134 struct btrfs_key key;
135 struct btrfs_path *path;
136 struct extent_buffer *leaf;
137 struct page *page = NULL;
140 struct btrfs_file_extent_item *ei;
143 size_t cur_size = size;
145 unsigned long offset;
147 if (compressed_size && compressed_pages)
148 cur_size = compressed_size;
150 path = btrfs_alloc_path();
154 path->leave_spinning = 1;
156 key.objectid = btrfs_ino(inode);
158 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
159 datasize = btrfs_file_extent_calc_inline_size(cur_size);
161 inode_add_bytes(inode, size);
162 ret = btrfs_insert_empty_item(trans, root, path, &key,
168 leaf = path->nodes[0];
169 ei = btrfs_item_ptr(leaf, path->slots[0],
170 struct btrfs_file_extent_item);
171 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
172 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
173 btrfs_set_file_extent_encryption(leaf, ei, 0);
174 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
175 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
176 ptr = btrfs_file_extent_inline_start(ei);
178 if (compress_type != BTRFS_COMPRESS_NONE) {
181 while (compressed_size > 0) {
182 cpage = compressed_pages[i];
183 cur_size = min_t(unsigned long, compressed_size,
186 kaddr = kmap_atomic(cpage);
187 write_extent_buffer(leaf, kaddr, ptr, cur_size);
188 kunmap_atomic(kaddr);
192 compressed_size -= cur_size;
194 btrfs_set_file_extent_compression(leaf, ei,
197 page = find_get_page(inode->i_mapping,
198 start >> PAGE_CACHE_SHIFT);
199 btrfs_set_file_extent_compression(leaf, ei, 0);
200 kaddr = kmap_atomic(page);
201 offset = start & (PAGE_CACHE_SIZE - 1);
202 write_extent_buffer(leaf, kaddr + offset, ptr, size);
203 kunmap_atomic(kaddr);
204 page_cache_release(page);
206 btrfs_mark_buffer_dirty(leaf);
207 btrfs_free_path(path);
210 * we're an inline extent, so nobody can
211 * extend the file past i_size without locking
212 * a page we already have locked.
214 * We must do any isize and inode updates
215 * before we unlock the pages. Otherwise we
216 * could end up racing with unlink.
218 BTRFS_I(inode)->disk_i_size = inode->i_size;
219 ret = btrfs_update_inode(trans, root, inode);
223 btrfs_free_path(path);
229 * conditionally insert an inline extent into the file. This
230 * does the checks required to make sure the data is small enough
231 * to fit as an inline extent.
233 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
234 struct btrfs_root *root,
235 struct inode *inode, u64 start, u64 end,
236 size_t compressed_size, int compress_type,
237 struct page **compressed_pages)
239 u64 isize = i_size_read(inode);
240 u64 actual_end = min(end + 1, isize);
241 u64 inline_len = actual_end - start;
242 u64 aligned_end = ALIGN(end, root->sectorsize);
243 u64 data_len = inline_len;
247 data_len = compressed_size;
250 actual_end >= PAGE_CACHE_SIZE ||
251 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
253 (actual_end & (root->sectorsize - 1)) == 0) ||
255 data_len > root->fs_info->max_inline) {
259 ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
263 if (isize > actual_end)
264 inline_len = min_t(u64, isize, actual_end);
265 ret = insert_inline_extent(trans, root, inode, start,
266 inline_len, compressed_size,
267 compress_type, compressed_pages);
268 if (ret && ret != -ENOSPC) {
269 btrfs_abort_transaction(trans, root, ret);
271 } else if (ret == -ENOSPC) {
275 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
276 btrfs_delalloc_release_metadata(inode, end + 1 - start);
277 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
281 struct async_extent {
286 unsigned long nr_pages;
288 struct list_head list;
293 struct btrfs_root *root;
294 struct page *locked_page;
297 struct list_head extents;
298 struct btrfs_work work;
301 static noinline int add_async_extent(struct async_cow *cow,
302 u64 start, u64 ram_size,
305 unsigned long nr_pages,
308 struct async_extent *async_extent;
310 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
311 BUG_ON(!async_extent); /* -ENOMEM */
312 async_extent->start = start;
313 async_extent->ram_size = ram_size;
314 async_extent->compressed_size = compressed_size;
315 async_extent->pages = pages;
316 async_extent->nr_pages = nr_pages;
317 async_extent->compress_type = compress_type;
318 list_add_tail(&async_extent->list, &cow->extents);
323 * we create compressed extents in two phases. The first
324 * phase compresses a range of pages that have already been
325 * locked (both pages and state bits are locked).
327 * This is done inside an ordered work queue, and the compression
328 * is spread across many cpus. The actual IO submission is step
329 * two, and the ordered work queue takes care of making sure that
330 * happens in the same order things were put onto the queue by
331 * writepages and friends.
333 * If this code finds it can't get good compression, it puts an
334 * entry onto the work queue to write the uncompressed bytes. This
335 * makes sure that both compressed inodes and uncompressed inodes
336 * are written in the same order that the flusher thread sent them
339 static noinline int compress_file_range(struct inode *inode,
340 struct page *locked_page,
342 struct async_cow *async_cow,
345 struct btrfs_root *root = BTRFS_I(inode)->root;
346 struct btrfs_trans_handle *trans;
348 u64 blocksize = root->sectorsize;
350 u64 isize = i_size_read(inode);
352 struct page **pages = NULL;
353 unsigned long nr_pages;
354 unsigned long nr_pages_ret = 0;
355 unsigned long total_compressed = 0;
356 unsigned long total_in = 0;
357 unsigned long max_compressed = 128 * 1024;
358 unsigned long max_uncompressed = 128 * 1024;
361 int compress_type = root->fs_info->compress_type;
364 /* if this is a small write inside eof, kick off a defrag */
365 if ((end - start + 1) < 16 * 1024 &&
366 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
367 btrfs_add_inode_defrag(NULL, inode);
369 actual_end = min_t(u64, isize, end + 1);
372 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
373 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
376 * we don't want to send crud past the end of i_size through
377 * compression, that's just a waste of CPU time. So, if the
378 * end of the file is before the start of our current
379 * requested range of bytes, we bail out to the uncompressed
380 * cleanup code that can deal with all of this.
382 * It isn't really the fastest way to fix things, but this is a
383 * very uncommon corner.
385 if (actual_end <= start)
386 goto cleanup_and_bail_uncompressed;
388 total_compressed = actual_end - start;
390 /* we want to make sure that amount of ram required to uncompress
391 * an extent is reasonable, so we limit the total size in ram
392 * of a compressed extent to 128k. This is a crucial number
393 * because it also controls how easily we can spread reads across
394 * cpus for decompression.
396 * We also want to make sure the amount of IO required to do
397 * a random read is reasonably small, so we limit the size of
398 * a compressed extent to 128k.
400 total_compressed = min(total_compressed, max_uncompressed);
401 num_bytes = ALIGN(end - start + 1, blocksize);
402 num_bytes = max(blocksize, num_bytes);
407 * we do compression for mount -o compress and when the
408 * inode has not been flagged as nocompress. This flag can
409 * change at any time if we discover bad compression ratios.
411 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
412 (btrfs_test_opt(root, COMPRESS) ||
413 (BTRFS_I(inode)->force_compress) ||
414 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
416 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
418 /* just bail out to the uncompressed code */
422 if (BTRFS_I(inode)->force_compress)
423 compress_type = BTRFS_I(inode)->force_compress;
426 * we need to call clear_page_dirty_for_io on each
427 * page in the range. Otherwise applications with the file
428 * mmap'd can wander in and change the page contents while
429 * we are compressing them.
431 * If the compression fails for any reason, we set the pages
432 * dirty again later on.
434 extent_range_clear_dirty_for_io(inode, start, end);
436 ret = btrfs_compress_pages(compress_type,
437 inode->i_mapping, start,
438 total_compressed, pages,
439 nr_pages, &nr_pages_ret,
445 unsigned long offset = total_compressed &
446 (PAGE_CACHE_SIZE - 1);
447 struct page *page = pages[nr_pages_ret - 1];
450 /* zero the tail end of the last page, we might be
451 * sending it down to disk
454 kaddr = kmap_atomic(page);
455 memset(kaddr + offset, 0,
456 PAGE_CACHE_SIZE - offset);
457 kunmap_atomic(kaddr);
464 trans = btrfs_join_transaction(root);
466 ret = PTR_ERR(trans);
468 goto cleanup_and_out;
470 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
472 /* lets try to make an inline extent */
473 if (ret || total_in < (actual_end - start)) {
474 /* we didn't compress the entire range, try
475 * to make an uncompressed inline extent.
477 ret = cow_file_range_inline(trans, root, inode,
478 start, end, 0, 0, NULL);
480 /* try making a compressed inline extent */
481 ret = cow_file_range_inline(trans, root, inode,
484 compress_type, pages);
488 * inline extent creation worked or returned error,
489 * we don't need to create any more async work items.
490 * Unlock and free up our temp pages.
492 extent_clear_unlock_delalloc(inode,
493 &BTRFS_I(inode)->io_tree,
495 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
496 EXTENT_CLEAR_DELALLOC |
497 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
499 btrfs_end_transaction(trans, root);
502 btrfs_end_transaction(trans, root);
507 * we aren't doing an inline extent round the compressed size
508 * up to a block size boundary so the allocator does sane
511 total_compressed = ALIGN(total_compressed, blocksize);
514 * one last check to make sure the compression is really a
515 * win, compare the page count read with the blocks on disk
517 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
518 if (total_compressed >= total_in) {
521 num_bytes = total_in;
524 if (!will_compress && pages) {
526 * the compression code ran but failed to make things smaller,
527 * free any pages it allocated and our page pointer array
529 for (i = 0; i < nr_pages_ret; i++) {
530 WARN_ON(pages[i]->mapping);
531 page_cache_release(pages[i]);
535 total_compressed = 0;
538 /* flag the file so we don't compress in the future */
539 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
540 !(BTRFS_I(inode)->force_compress)) {
541 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
547 /* the async work queues will take care of doing actual
548 * allocation on disk for these compressed pages,
549 * and will submit them to the elevator.
551 add_async_extent(async_cow, start, num_bytes,
552 total_compressed, pages, nr_pages_ret,
555 if (start + num_bytes < end) {
562 cleanup_and_bail_uncompressed:
564 * No compression, but we still need to write the pages in
565 * the file we've been given so far. redirty the locked
566 * page if it corresponds to our extent and set things up
567 * for the async work queue to run cow_file_range to do
568 * the normal delalloc dance
570 if (page_offset(locked_page) >= start &&
571 page_offset(locked_page) <= end) {
572 __set_page_dirty_nobuffers(locked_page);
573 /* unlocked later on in the async handlers */
576 extent_range_redirty_for_io(inode, start, end);
577 add_async_extent(async_cow, start, end - start + 1,
578 0, NULL, 0, BTRFS_COMPRESS_NONE);
586 for (i = 0; i < nr_pages_ret; i++) {
587 WARN_ON(pages[i]->mapping);
588 page_cache_release(pages[i]);
595 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
597 EXTENT_CLEAR_UNLOCK_PAGE |
599 EXTENT_CLEAR_DELALLOC |
600 EXTENT_SET_WRITEBACK |
601 EXTENT_END_WRITEBACK);
602 if (!trans || IS_ERR(trans))
603 btrfs_error(root->fs_info, ret, "Failed to join transaction");
605 btrfs_abort_transaction(trans, root, ret);
610 * phase two of compressed writeback. This is the ordered portion
611 * of the code, which only gets called in the order the work was
612 * queued. We walk all the async extents created by compress_file_range
613 * and send them down to the disk.
615 static noinline int submit_compressed_extents(struct inode *inode,
616 struct async_cow *async_cow)
618 struct async_extent *async_extent;
620 struct btrfs_trans_handle *trans;
621 struct btrfs_key ins;
622 struct extent_map *em;
623 struct btrfs_root *root = BTRFS_I(inode)->root;
624 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
625 struct extent_io_tree *io_tree;
628 if (list_empty(&async_cow->extents))
632 while (!list_empty(&async_cow->extents)) {
633 async_extent = list_entry(async_cow->extents.next,
634 struct async_extent, list);
635 list_del(&async_extent->list);
637 io_tree = &BTRFS_I(inode)->io_tree;
640 /* did the compression code fall back to uncompressed IO? */
641 if (!async_extent->pages) {
642 int page_started = 0;
643 unsigned long nr_written = 0;
645 lock_extent(io_tree, async_extent->start,
646 async_extent->start +
647 async_extent->ram_size - 1);
649 /* allocate blocks */
650 ret = cow_file_range(inode, async_cow->locked_page,
652 async_extent->start +
653 async_extent->ram_size - 1,
654 &page_started, &nr_written, 0);
659 * if page_started, cow_file_range inserted an
660 * inline extent and took care of all the unlocking
661 * and IO for us. Otherwise, we need to submit
662 * all those pages down to the drive.
664 if (!page_started && !ret)
665 extent_write_locked_range(io_tree,
666 inode, async_extent->start,
667 async_extent->start +
668 async_extent->ram_size - 1,
672 unlock_page(async_cow->locked_page);
678 lock_extent(io_tree, async_extent->start,
679 async_extent->start + async_extent->ram_size - 1);
681 trans = btrfs_join_transaction(root);
683 ret = PTR_ERR(trans);
685 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
686 ret = btrfs_reserve_extent(trans, root,
687 async_extent->compressed_size,
688 async_extent->compressed_size,
689 0, alloc_hint, &ins, 1);
690 if (ret && ret != -ENOSPC)
691 btrfs_abort_transaction(trans, root, ret);
692 btrfs_end_transaction(trans, root);
698 for (i = 0; i < async_extent->nr_pages; i++) {
699 WARN_ON(async_extent->pages[i]->mapping);
700 page_cache_release(async_extent->pages[i]);
702 kfree(async_extent->pages);
703 async_extent->nr_pages = 0;
704 async_extent->pages = NULL;
706 if (ret == -ENOSPC) {
707 unlock_extent(io_tree, async_extent->start,
708 async_extent->start +
709 async_extent->ram_size - 1);
716 * here we're doing allocation and writeback of the
719 btrfs_drop_extent_cache(inode, async_extent->start,
720 async_extent->start +
721 async_extent->ram_size - 1, 0);
723 em = alloc_extent_map();
726 goto out_free_reserve;
728 em->start = async_extent->start;
729 em->len = async_extent->ram_size;
730 em->orig_start = em->start;
731 em->mod_start = em->start;
732 em->mod_len = em->len;
734 em->block_start = ins.objectid;
735 em->block_len = ins.offset;
736 em->orig_block_len = ins.offset;
737 em->ram_bytes = async_extent->ram_size;
738 em->bdev = root->fs_info->fs_devices->latest_bdev;
739 em->compress_type = async_extent->compress_type;
740 set_bit(EXTENT_FLAG_PINNED, &em->flags);
741 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
745 write_lock(&em_tree->lock);
746 ret = add_extent_mapping(em_tree, em, 1);
747 write_unlock(&em_tree->lock);
748 if (ret != -EEXIST) {
752 btrfs_drop_extent_cache(inode, async_extent->start,
753 async_extent->start +
754 async_extent->ram_size - 1, 0);
758 goto out_free_reserve;
760 ret = btrfs_add_ordered_extent_compress(inode,
763 async_extent->ram_size,
765 BTRFS_ORDERED_COMPRESSED,
766 async_extent->compress_type);
768 goto out_free_reserve;
771 * clear dirty, set writeback and unlock the pages.
773 extent_clear_unlock_delalloc(inode,
774 &BTRFS_I(inode)->io_tree,
776 async_extent->start +
777 async_extent->ram_size - 1,
778 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
779 EXTENT_CLEAR_UNLOCK |
780 EXTENT_CLEAR_DELALLOC |
781 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
783 ret = btrfs_submit_compressed_write(inode,
785 async_extent->ram_size,
787 ins.offset, async_extent->pages,
788 async_extent->nr_pages);
789 alloc_hint = ins.objectid + ins.offset;
799 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
801 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
803 async_extent->start +
804 async_extent->ram_size - 1,
805 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
806 EXTENT_CLEAR_UNLOCK |
807 EXTENT_CLEAR_DELALLOC |
809 EXTENT_SET_WRITEBACK |
810 EXTENT_END_WRITEBACK);
815 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
818 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
819 struct extent_map *em;
822 read_lock(&em_tree->lock);
823 em = search_extent_mapping(em_tree, start, num_bytes);
826 * if block start isn't an actual block number then find the
827 * first block in this inode and use that as a hint. If that
828 * block is also bogus then just don't worry about it.
830 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
832 em = search_extent_mapping(em_tree, 0, 0);
833 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
834 alloc_hint = em->block_start;
838 alloc_hint = em->block_start;
842 read_unlock(&em_tree->lock);
848 * when extent_io.c finds a delayed allocation range in the file,
849 * the call backs end up in this code. The basic idea is to
850 * allocate extents on disk for the range, and create ordered data structs
851 * in ram to track those extents.
853 * locked_page is the page that writepage had locked already. We use
854 * it to make sure we don't do extra locks or unlocks.
856 * *page_started is set to one if we unlock locked_page and do everything
857 * required to start IO on it. It may be clean and already done with
860 static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
862 struct btrfs_root *root,
863 struct page *locked_page,
864 u64 start, u64 end, int *page_started,
865 unsigned long *nr_written,
870 unsigned long ram_size;
873 u64 blocksize = root->sectorsize;
874 struct btrfs_key ins;
875 struct extent_map *em;
876 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
879 BUG_ON(btrfs_is_free_space_inode(inode));
881 num_bytes = ALIGN(end - start + 1, blocksize);
882 num_bytes = max(blocksize, num_bytes);
883 disk_num_bytes = num_bytes;
885 /* if this is a small write inside eof, kick off defrag */
886 if (num_bytes < 64 * 1024 &&
887 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
888 btrfs_add_inode_defrag(trans, inode);
891 /* lets try to make an inline extent */
892 ret = cow_file_range_inline(trans, root, inode,
893 start, end, 0, 0, NULL);
895 extent_clear_unlock_delalloc(inode,
896 &BTRFS_I(inode)->io_tree,
898 EXTENT_CLEAR_UNLOCK_PAGE |
899 EXTENT_CLEAR_UNLOCK |
900 EXTENT_CLEAR_DELALLOC |
902 EXTENT_SET_WRITEBACK |
903 EXTENT_END_WRITEBACK);
905 *nr_written = *nr_written +
906 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
909 } else if (ret < 0) {
910 btrfs_abort_transaction(trans, root, ret);
915 BUG_ON(disk_num_bytes >
916 btrfs_super_total_bytes(root->fs_info->super_copy));
918 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
919 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
921 while (disk_num_bytes > 0) {
924 cur_alloc_size = disk_num_bytes;
925 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
926 root->sectorsize, 0, alloc_hint,
929 btrfs_abort_transaction(trans, root, ret);
933 em = alloc_extent_map();
939 em->orig_start = em->start;
940 ram_size = ins.offset;
941 em->len = ins.offset;
942 em->mod_start = em->start;
943 em->mod_len = em->len;
945 em->block_start = ins.objectid;
946 em->block_len = ins.offset;
947 em->orig_block_len = ins.offset;
948 em->ram_bytes = ram_size;
949 em->bdev = root->fs_info->fs_devices->latest_bdev;
950 set_bit(EXTENT_FLAG_PINNED, &em->flags);
954 write_lock(&em_tree->lock);
955 ret = add_extent_mapping(em_tree, em, 1);
956 write_unlock(&em_tree->lock);
957 if (ret != -EEXIST) {
961 btrfs_drop_extent_cache(inode, start,
962 start + ram_size - 1, 0);
967 cur_alloc_size = ins.offset;
968 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
969 ram_size, cur_alloc_size, 0);
973 if (root->root_key.objectid ==
974 BTRFS_DATA_RELOC_TREE_OBJECTID) {
975 ret = btrfs_reloc_clone_csums(inode, start,
978 btrfs_abort_transaction(trans, root, ret);
983 if (disk_num_bytes < cur_alloc_size)
986 /* we're not doing compressed IO, don't unlock the first
987 * page (which the caller expects to stay locked), don't
988 * clear any dirty bits and don't set any writeback bits
990 * Do set the Private2 bit so we know this page was properly
991 * setup for writepage
993 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
994 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
997 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
998 start, start + ram_size - 1,
1000 disk_num_bytes -= cur_alloc_size;
1001 num_bytes -= cur_alloc_size;
1002 alloc_hint = ins.objectid + ins.offset;
1003 start += cur_alloc_size;
1009 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
1011 extent_clear_unlock_delalloc(inode,
1012 &BTRFS_I(inode)->io_tree,
1013 start, end, locked_page,
1014 EXTENT_CLEAR_UNLOCK_PAGE |
1015 EXTENT_CLEAR_UNLOCK |
1016 EXTENT_CLEAR_DELALLOC |
1017 EXTENT_CLEAR_DIRTY |
1018 EXTENT_SET_WRITEBACK |
1019 EXTENT_END_WRITEBACK);
1024 static noinline int cow_file_range(struct inode *inode,
1025 struct page *locked_page,
1026 u64 start, u64 end, int *page_started,
1027 unsigned long *nr_written,
1030 struct btrfs_trans_handle *trans;
1031 struct btrfs_root *root = BTRFS_I(inode)->root;
1034 trans = btrfs_join_transaction(root);
1035 if (IS_ERR(trans)) {
1036 extent_clear_unlock_delalloc(inode,
1037 &BTRFS_I(inode)->io_tree,
1038 start, end, locked_page,
1039 EXTENT_CLEAR_UNLOCK_PAGE |
1040 EXTENT_CLEAR_UNLOCK |
1041 EXTENT_CLEAR_DELALLOC |
1042 EXTENT_CLEAR_DIRTY |
1043 EXTENT_SET_WRITEBACK |
1044 EXTENT_END_WRITEBACK);
1045 return PTR_ERR(trans);
1047 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1049 ret = __cow_file_range(trans, inode, root, locked_page, start, end,
1050 page_started, nr_written, unlock);
1052 btrfs_end_transaction(trans, root);
1058 * work queue call back to started compression on a file and pages
1060 static noinline void async_cow_start(struct btrfs_work *work)
1062 struct async_cow *async_cow;
1064 async_cow = container_of(work, struct async_cow, work);
1066 compress_file_range(async_cow->inode, async_cow->locked_page,
1067 async_cow->start, async_cow->end, async_cow,
1069 if (num_added == 0) {
1070 btrfs_add_delayed_iput(async_cow->inode);
1071 async_cow->inode = NULL;
1076 * work queue call back to submit previously compressed pages
1078 static noinline void async_cow_submit(struct btrfs_work *work)
1080 struct async_cow *async_cow;
1081 struct btrfs_root *root;
1082 unsigned long nr_pages;
1084 async_cow = container_of(work, struct async_cow, work);
1086 root = async_cow->root;
1087 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1090 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1092 waitqueue_active(&root->fs_info->async_submit_wait))
1093 wake_up(&root->fs_info->async_submit_wait);
1095 if (async_cow->inode)
1096 submit_compressed_extents(async_cow->inode, async_cow);
1099 static noinline void async_cow_free(struct btrfs_work *work)
1101 struct async_cow *async_cow;
1102 async_cow = container_of(work, struct async_cow, work);
1103 if (async_cow->inode)
1104 btrfs_add_delayed_iput(async_cow->inode);
1108 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1109 u64 start, u64 end, int *page_started,
1110 unsigned long *nr_written)
1112 struct async_cow *async_cow;
1113 struct btrfs_root *root = BTRFS_I(inode)->root;
1114 unsigned long nr_pages;
1116 int limit = 10 * 1024 * 1024;
1118 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1119 1, 0, NULL, GFP_NOFS);
1120 while (start < end) {
1121 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1122 BUG_ON(!async_cow); /* -ENOMEM */
1123 async_cow->inode = igrab(inode);
1124 async_cow->root = root;
1125 async_cow->locked_page = locked_page;
1126 async_cow->start = start;
1128 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1131 cur_end = min(end, start + 512 * 1024 - 1);
1133 async_cow->end = cur_end;
1134 INIT_LIST_HEAD(&async_cow->extents);
1136 async_cow->work.func = async_cow_start;
1137 async_cow->work.ordered_func = async_cow_submit;
1138 async_cow->work.ordered_free = async_cow_free;
1139 async_cow->work.flags = 0;
1141 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1143 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1145 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1148 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1149 wait_event(root->fs_info->async_submit_wait,
1150 (atomic_read(&root->fs_info->async_delalloc_pages) <
1154 while (atomic_read(&root->fs_info->async_submit_draining) &&
1155 atomic_read(&root->fs_info->async_delalloc_pages)) {
1156 wait_event(root->fs_info->async_submit_wait,
1157 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1161 *nr_written += nr_pages;
1162 start = cur_end + 1;
1168 static noinline int csum_exist_in_range(struct btrfs_root *root,
1169 u64 bytenr, u64 num_bytes)
1172 struct btrfs_ordered_sum *sums;
1175 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1176 bytenr + num_bytes - 1, &list, 0);
1177 if (ret == 0 && list_empty(&list))
1180 while (!list_empty(&list)) {
1181 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1182 list_del(&sums->list);
1189 * when nowcow writeback call back. This checks for snapshots or COW copies
1190 * of the extents that exist in the file, and COWs the file as required.
1192 * If no cow copies or snapshots exist, we write directly to the existing
1195 static noinline int run_delalloc_nocow(struct inode *inode,
1196 struct page *locked_page,
1197 u64 start, u64 end, int *page_started, int force,
1198 unsigned long *nr_written)
1200 struct btrfs_root *root = BTRFS_I(inode)->root;
1201 struct btrfs_trans_handle *trans;
1202 struct extent_buffer *leaf;
1203 struct btrfs_path *path;
1204 struct btrfs_file_extent_item *fi;
1205 struct btrfs_key found_key;
1220 u64 ino = btrfs_ino(inode);
1222 path = btrfs_alloc_path();
1224 extent_clear_unlock_delalloc(inode,
1225 &BTRFS_I(inode)->io_tree,
1226 start, end, locked_page,
1227 EXTENT_CLEAR_UNLOCK_PAGE |
1228 EXTENT_CLEAR_UNLOCK |
1229 EXTENT_CLEAR_DELALLOC |
1230 EXTENT_CLEAR_DIRTY |
1231 EXTENT_SET_WRITEBACK |
1232 EXTENT_END_WRITEBACK);
1236 nolock = btrfs_is_free_space_inode(inode);
1239 trans = btrfs_join_transaction_nolock(root);
1241 trans = btrfs_join_transaction(root);
1243 if (IS_ERR(trans)) {
1244 extent_clear_unlock_delalloc(inode,
1245 &BTRFS_I(inode)->io_tree,
1246 start, end, locked_page,
1247 EXTENT_CLEAR_UNLOCK_PAGE |
1248 EXTENT_CLEAR_UNLOCK |
1249 EXTENT_CLEAR_DELALLOC |
1250 EXTENT_CLEAR_DIRTY |
1251 EXTENT_SET_WRITEBACK |
1252 EXTENT_END_WRITEBACK);
1253 btrfs_free_path(path);
1254 return PTR_ERR(trans);
1257 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1259 cow_start = (u64)-1;
1262 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1265 btrfs_abort_transaction(trans, root, ret);
1268 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1269 leaf = path->nodes[0];
1270 btrfs_item_key_to_cpu(leaf, &found_key,
1271 path->slots[0] - 1);
1272 if (found_key.objectid == ino &&
1273 found_key.type == BTRFS_EXTENT_DATA_KEY)
1278 leaf = path->nodes[0];
1279 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1280 ret = btrfs_next_leaf(root, path);
1282 btrfs_abort_transaction(trans, root, ret);
1287 leaf = path->nodes[0];
1293 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1295 if (found_key.objectid > ino ||
1296 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1297 found_key.offset > end)
1300 if (found_key.offset > cur_offset) {
1301 extent_end = found_key.offset;
1306 fi = btrfs_item_ptr(leaf, path->slots[0],
1307 struct btrfs_file_extent_item);
1308 extent_type = btrfs_file_extent_type(leaf, fi);
1310 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1311 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1312 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1313 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1314 extent_offset = btrfs_file_extent_offset(leaf, fi);
1315 extent_end = found_key.offset +
1316 btrfs_file_extent_num_bytes(leaf, fi);
1318 btrfs_file_extent_disk_num_bytes(leaf, fi);
1319 if (extent_end <= start) {
1323 if (disk_bytenr == 0)
1325 if (btrfs_file_extent_compression(leaf, fi) ||
1326 btrfs_file_extent_encryption(leaf, fi) ||
1327 btrfs_file_extent_other_encoding(leaf, fi))
1329 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1331 if (btrfs_extent_readonly(root, disk_bytenr))
1333 if (btrfs_cross_ref_exist(trans, root, ino,
1335 extent_offset, disk_bytenr))
1337 disk_bytenr += extent_offset;
1338 disk_bytenr += cur_offset - found_key.offset;
1339 num_bytes = min(end + 1, extent_end) - cur_offset;
1341 * force cow if csum exists in the range.
1342 * this ensure that csum for a given extent are
1343 * either valid or do not exist.
1345 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1348 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1349 extent_end = found_key.offset +
1350 btrfs_file_extent_inline_len(leaf, fi);
1351 extent_end = ALIGN(extent_end, root->sectorsize);
1356 if (extent_end <= start) {
1361 if (cow_start == (u64)-1)
1362 cow_start = cur_offset;
1363 cur_offset = extent_end;
1364 if (cur_offset > end)
1370 btrfs_release_path(path);
1371 if (cow_start != (u64)-1) {
1372 ret = __cow_file_range(trans, inode, root, locked_page,
1373 cow_start, found_key.offset - 1,
1374 page_started, nr_written, 1);
1376 btrfs_abort_transaction(trans, root, ret);
1379 cow_start = (u64)-1;
1382 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1383 struct extent_map *em;
1384 struct extent_map_tree *em_tree;
1385 em_tree = &BTRFS_I(inode)->extent_tree;
1386 em = alloc_extent_map();
1387 BUG_ON(!em); /* -ENOMEM */
1388 em->start = cur_offset;
1389 em->orig_start = found_key.offset - extent_offset;
1390 em->len = num_bytes;
1391 em->block_len = num_bytes;
1392 em->block_start = disk_bytenr;
1393 em->orig_block_len = disk_num_bytes;
1394 em->ram_bytes = ram_bytes;
1395 em->bdev = root->fs_info->fs_devices->latest_bdev;
1396 em->mod_start = em->start;
1397 em->mod_len = em->len;
1398 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1399 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1400 em->generation = -1;
1402 write_lock(&em_tree->lock);
1403 ret = add_extent_mapping(em_tree, em, 1);
1404 write_unlock(&em_tree->lock);
1405 if (ret != -EEXIST) {
1406 free_extent_map(em);
1409 btrfs_drop_extent_cache(inode, em->start,
1410 em->start + em->len - 1, 0);
1412 type = BTRFS_ORDERED_PREALLOC;
1414 type = BTRFS_ORDERED_NOCOW;
1417 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1418 num_bytes, num_bytes, type);
1419 BUG_ON(ret); /* -ENOMEM */
1421 if (root->root_key.objectid ==
1422 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1423 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1426 btrfs_abort_transaction(trans, root, ret);
1431 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1432 cur_offset, cur_offset + num_bytes - 1,
1433 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1434 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1435 EXTENT_SET_PRIVATE2);
1436 cur_offset = extent_end;
1437 if (cur_offset > end)
1440 btrfs_release_path(path);
1442 if (cur_offset <= end && cow_start == (u64)-1) {
1443 cow_start = cur_offset;
1447 if (cow_start != (u64)-1) {
1448 ret = __cow_file_range(trans, inode, root, locked_page,
1450 page_started, nr_written, 1);
1452 btrfs_abort_transaction(trans, root, ret);
1458 err = btrfs_end_transaction(trans, root);
1462 if (ret && cur_offset < end)
1463 extent_clear_unlock_delalloc(inode,
1464 &BTRFS_I(inode)->io_tree,
1465 cur_offset, end, locked_page,
1466 EXTENT_CLEAR_UNLOCK_PAGE |
1467 EXTENT_CLEAR_UNLOCK |
1468 EXTENT_CLEAR_DELALLOC |
1469 EXTENT_CLEAR_DIRTY |
1470 EXTENT_SET_WRITEBACK |
1471 EXTENT_END_WRITEBACK);
1473 btrfs_free_path(path);
1478 * extent_io.c call back to do delayed allocation processing
1480 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1481 u64 start, u64 end, int *page_started,
1482 unsigned long *nr_written)
1485 struct btrfs_root *root = BTRFS_I(inode)->root;
1487 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1488 ret = run_delalloc_nocow(inode, locked_page, start, end,
1489 page_started, 1, nr_written);
1490 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1491 ret = run_delalloc_nocow(inode, locked_page, start, end,
1492 page_started, 0, nr_written);
1493 } else if (!btrfs_test_opt(root, COMPRESS) &&
1494 !(BTRFS_I(inode)->force_compress) &&
1495 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1496 ret = cow_file_range(inode, locked_page, start, end,
1497 page_started, nr_written, 1);
1499 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1500 &BTRFS_I(inode)->runtime_flags);
1501 ret = cow_file_range_async(inode, locked_page, start, end,
1502 page_started, nr_written);
1507 static void btrfs_split_extent_hook(struct inode *inode,
1508 struct extent_state *orig, u64 split)
1510 /* not delalloc, ignore it */
1511 if (!(orig->state & EXTENT_DELALLOC))
1514 spin_lock(&BTRFS_I(inode)->lock);
1515 BTRFS_I(inode)->outstanding_extents++;
1516 spin_unlock(&BTRFS_I(inode)->lock);
1520 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1521 * extents so we can keep track of new extents that are just merged onto old
1522 * extents, such as when we are doing sequential writes, so we can properly
1523 * account for the metadata space we'll need.
1525 static void btrfs_merge_extent_hook(struct inode *inode,
1526 struct extent_state *new,
1527 struct extent_state *other)
1529 /* not delalloc, ignore it */
1530 if (!(other->state & EXTENT_DELALLOC))
1533 spin_lock(&BTRFS_I(inode)->lock);
1534 BTRFS_I(inode)->outstanding_extents--;
1535 spin_unlock(&BTRFS_I(inode)->lock);
1538 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1539 struct inode *inode)
1541 spin_lock(&root->delalloc_lock);
1542 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1543 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1544 &root->delalloc_inodes);
1545 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1546 &BTRFS_I(inode)->runtime_flags);
1547 root->nr_delalloc_inodes++;
1548 if (root->nr_delalloc_inodes == 1) {
1549 spin_lock(&root->fs_info->delalloc_root_lock);
1550 BUG_ON(!list_empty(&root->delalloc_root));
1551 list_add_tail(&root->delalloc_root,
1552 &root->fs_info->delalloc_roots);
1553 spin_unlock(&root->fs_info->delalloc_root_lock);
1556 spin_unlock(&root->delalloc_lock);
1559 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1560 struct inode *inode)
1562 spin_lock(&root->delalloc_lock);
1563 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1564 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1565 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1566 &BTRFS_I(inode)->runtime_flags);
1567 root->nr_delalloc_inodes--;
1568 if (!root->nr_delalloc_inodes) {
1569 spin_lock(&root->fs_info->delalloc_root_lock);
1570 BUG_ON(list_empty(&root->delalloc_root));
1571 list_del_init(&root->delalloc_root);
1572 spin_unlock(&root->fs_info->delalloc_root_lock);
1575 spin_unlock(&root->delalloc_lock);
1579 * extent_io.c set_bit_hook, used to track delayed allocation
1580 * bytes in this file, and to maintain the list of inodes that
1581 * have pending delalloc work to be done.
1583 static void btrfs_set_bit_hook(struct inode *inode,
1584 struct extent_state *state, unsigned long *bits)
1588 * set_bit and clear bit hooks normally require _irqsave/restore
1589 * but in this case, we are only testing for the DELALLOC
1590 * bit, which is only set or cleared with irqs on
1592 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1593 struct btrfs_root *root = BTRFS_I(inode)->root;
1594 u64 len = state->end + 1 - state->start;
1595 bool do_list = !btrfs_is_free_space_inode(inode);
1597 if (*bits & EXTENT_FIRST_DELALLOC) {
1598 *bits &= ~EXTENT_FIRST_DELALLOC;
1600 spin_lock(&BTRFS_I(inode)->lock);
1601 BTRFS_I(inode)->outstanding_extents++;
1602 spin_unlock(&BTRFS_I(inode)->lock);
1605 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1606 root->fs_info->delalloc_batch);
1607 spin_lock(&BTRFS_I(inode)->lock);
1608 BTRFS_I(inode)->delalloc_bytes += len;
1609 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1610 &BTRFS_I(inode)->runtime_flags))
1611 btrfs_add_delalloc_inodes(root, inode);
1612 spin_unlock(&BTRFS_I(inode)->lock);
1617 * extent_io.c clear_bit_hook, see set_bit_hook for why
1619 static void btrfs_clear_bit_hook(struct inode *inode,
1620 struct extent_state *state,
1621 unsigned long *bits)
1624 * set_bit and clear bit hooks normally require _irqsave/restore
1625 * but in this case, we are only testing for the DELALLOC
1626 * bit, which is only set or cleared with irqs on
1628 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1629 struct btrfs_root *root = BTRFS_I(inode)->root;
1630 u64 len = state->end + 1 - state->start;
1631 bool do_list = !btrfs_is_free_space_inode(inode);
1633 if (*bits & EXTENT_FIRST_DELALLOC) {
1634 *bits &= ~EXTENT_FIRST_DELALLOC;
1635 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1636 spin_lock(&BTRFS_I(inode)->lock);
1637 BTRFS_I(inode)->outstanding_extents--;
1638 spin_unlock(&BTRFS_I(inode)->lock);
1641 if (*bits & EXTENT_DO_ACCOUNTING)
1642 btrfs_delalloc_release_metadata(inode, len);
1644 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1645 && do_list && !(state->state & EXTENT_NORESERVE))
1646 btrfs_free_reserved_data_space(inode, len);
1648 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1649 root->fs_info->delalloc_batch);
1650 spin_lock(&BTRFS_I(inode)->lock);
1651 BTRFS_I(inode)->delalloc_bytes -= len;
1652 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1653 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1654 &BTRFS_I(inode)->runtime_flags))
1655 btrfs_del_delalloc_inode(root, inode);
1656 spin_unlock(&BTRFS_I(inode)->lock);
1661 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1662 * we don't create bios that span stripes or chunks
1664 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1665 size_t size, struct bio *bio,
1666 unsigned long bio_flags)
1668 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1669 u64 logical = (u64)bio->bi_sector << 9;
1674 if (bio_flags & EXTENT_BIO_COMPRESSED)
1677 length = bio->bi_size;
1678 map_length = length;
1679 ret = btrfs_map_block(root->fs_info, rw, logical,
1680 &map_length, NULL, 0);
1681 /* Will always return 0 with map_multi == NULL */
1683 if (map_length < length + size)
1689 * in order to insert checksums into the metadata in large chunks,
1690 * we wait until bio submission time. All the pages in the bio are
1691 * checksummed and sums are attached onto the ordered extent record.
1693 * At IO completion time the cums attached on the ordered extent record
1694 * are inserted into the btree
1696 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1697 struct bio *bio, int mirror_num,
1698 unsigned long bio_flags,
1701 struct btrfs_root *root = BTRFS_I(inode)->root;
1704 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1705 BUG_ON(ret); /* -ENOMEM */
1710 * in order to insert checksums into the metadata in large chunks,
1711 * we wait until bio submission time. All the pages in the bio are
1712 * checksummed and sums are attached onto the ordered extent record.
1714 * At IO completion time the cums attached on the ordered extent record
1715 * are inserted into the btree
1717 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1718 int mirror_num, unsigned long bio_flags,
1721 struct btrfs_root *root = BTRFS_I(inode)->root;
1724 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1726 bio_endio(bio, ret);
1731 * extent_io.c submission hook. This does the right thing for csum calculation
1732 * on write, or reading the csums from the tree before a read
1734 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1735 int mirror_num, unsigned long bio_flags,
1738 struct btrfs_root *root = BTRFS_I(inode)->root;
1742 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1744 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1746 if (btrfs_is_free_space_inode(inode))
1749 if (!(rw & REQ_WRITE)) {
1750 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1754 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1755 ret = btrfs_submit_compressed_read(inode, bio,
1759 } else if (!skip_sum) {
1760 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1765 } else if (async && !skip_sum) {
1766 /* csum items have already been cloned */
1767 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1769 /* we're doing a write, do the async checksumming */
1770 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1771 inode, rw, bio, mirror_num,
1772 bio_flags, bio_offset,
1773 __btrfs_submit_bio_start,
1774 __btrfs_submit_bio_done);
1776 } else if (!skip_sum) {
1777 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1783 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1787 bio_endio(bio, ret);
1792 * given a list of ordered sums record them in the inode. This happens
1793 * at IO completion time based on sums calculated at bio submission time.
1795 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1796 struct inode *inode, u64 file_offset,
1797 struct list_head *list)
1799 struct btrfs_ordered_sum *sum;
1801 list_for_each_entry(sum, list, list) {
1802 trans->adding_csums = 1;
1803 btrfs_csum_file_blocks(trans,
1804 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1805 trans->adding_csums = 0;
1810 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1811 struct extent_state **cached_state)
1813 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1814 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1815 cached_state, GFP_NOFS);
1818 /* see btrfs_writepage_start_hook for details on why this is required */
1819 struct btrfs_writepage_fixup {
1821 struct btrfs_work work;
1824 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1826 struct btrfs_writepage_fixup *fixup;
1827 struct btrfs_ordered_extent *ordered;
1828 struct extent_state *cached_state = NULL;
1830 struct inode *inode;
1835 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1839 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1840 ClearPageChecked(page);
1844 inode = page->mapping->host;
1845 page_start = page_offset(page);
1846 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1848 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1851 /* already ordered? We're done */
1852 if (PagePrivate2(page))
1855 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1857 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1858 page_end, &cached_state, GFP_NOFS);
1860 btrfs_start_ordered_extent(inode, ordered, 1);
1861 btrfs_put_ordered_extent(ordered);
1865 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1867 mapping_set_error(page->mapping, ret);
1868 end_extent_writepage(page, ret, page_start, page_end);
1869 ClearPageChecked(page);
1873 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1874 ClearPageChecked(page);
1875 set_page_dirty(page);
1877 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1878 &cached_state, GFP_NOFS);
1881 page_cache_release(page);
1886 * There are a few paths in the higher layers of the kernel that directly
1887 * set the page dirty bit without asking the filesystem if it is a
1888 * good idea. This causes problems because we want to make sure COW
1889 * properly happens and the data=ordered rules are followed.
1891 * In our case any range that doesn't have the ORDERED bit set
1892 * hasn't been properly setup for IO. We kick off an async process
1893 * to fix it up. The async helper will wait for ordered extents, set
1894 * the delalloc bit and make it safe to write the page.
1896 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1898 struct inode *inode = page->mapping->host;
1899 struct btrfs_writepage_fixup *fixup;
1900 struct btrfs_root *root = BTRFS_I(inode)->root;
1902 /* this page is properly in the ordered list */
1903 if (TestClearPagePrivate2(page))
1906 if (PageChecked(page))
1909 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1913 SetPageChecked(page);
1914 page_cache_get(page);
1915 fixup->work.func = btrfs_writepage_fixup_worker;
1917 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1921 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1922 struct inode *inode, u64 file_pos,
1923 u64 disk_bytenr, u64 disk_num_bytes,
1924 u64 num_bytes, u64 ram_bytes,
1925 u8 compression, u8 encryption,
1926 u16 other_encoding, int extent_type)
1928 struct btrfs_root *root = BTRFS_I(inode)->root;
1929 struct btrfs_file_extent_item *fi;
1930 struct btrfs_path *path;
1931 struct extent_buffer *leaf;
1932 struct btrfs_key ins;
1935 path = btrfs_alloc_path();
1939 path->leave_spinning = 1;
1942 * we may be replacing one extent in the tree with another.
1943 * The new extent is pinned in the extent map, and we don't want
1944 * to drop it from the cache until it is completely in the btree.
1946 * So, tell btrfs_drop_extents to leave this extent in the cache.
1947 * the caller is expected to unpin it and allow it to be merged
1950 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1951 file_pos + num_bytes, 0);
1955 ins.objectid = btrfs_ino(inode);
1956 ins.offset = file_pos;
1957 ins.type = BTRFS_EXTENT_DATA_KEY;
1958 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1961 leaf = path->nodes[0];
1962 fi = btrfs_item_ptr(leaf, path->slots[0],
1963 struct btrfs_file_extent_item);
1964 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1965 btrfs_set_file_extent_type(leaf, fi, extent_type);
1966 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1967 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1968 btrfs_set_file_extent_offset(leaf, fi, 0);
1969 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1970 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1971 btrfs_set_file_extent_compression(leaf, fi, compression);
1972 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1973 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1975 btrfs_mark_buffer_dirty(leaf);
1976 btrfs_release_path(path);
1978 inode_add_bytes(inode, num_bytes);
1980 ins.objectid = disk_bytenr;
1981 ins.offset = disk_num_bytes;
1982 ins.type = BTRFS_EXTENT_ITEM_KEY;
1983 ret = btrfs_alloc_reserved_file_extent(trans, root,
1984 root->root_key.objectid,
1985 btrfs_ino(inode), file_pos, &ins);
1987 btrfs_free_path(path);
1992 /* snapshot-aware defrag */
1993 struct sa_defrag_extent_backref {
1994 struct rb_node node;
1995 struct old_sa_defrag_extent *old;
2004 struct old_sa_defrag_extent {
2005 struct list_head list;
2006 struct new_sa_defrag_extent *new;
2015 struct new_sa_defrag_extent {
2016 struct rb_root root;
2017 struct list_head head;
2018 struct btrfs_path *path;
2019 struct inode *inode;
2027 static int backref_comp(struct sa_defrag_extent_backref *b1,
2028 struct sa_defrag_extent_backref *b2)
2030 if (b1->root_id < b2->root_id)
2032 else if (b1->root_id > b2->root_id)
2035 if (b1->inum < b2->inum)
2037 else if (b1->inum > b2->inum)
2040 if (b1->file_pos < b2->file_pos)
2042 else if (b1->file_pos > b2->file_pos)
2046 * [------------------------------] ===> (a range of space)
2047 * |<--->| |<---->| =============> (fs/file tree A)
2048 * |<---------------------------->| ===> (fs/file tree B)
2050 * A range of space can refer to two file extents in one tree while
2051 * refer to only one file extent in another tree.
2053 * So we may process a disk offset more than one time(two extents in A)
2054 * and locate at the same extent(one extent in B), then insert two same
2055 * backrefs(both refer to the extent in B).
2060 static void backref_insert(struct rb_root *root,
2061 struct sa_defrag_extent_backref *backref)
2063 struct rb_node **p = &root->rb_node;
2064 struct rb_node *parent = NULL;
2065 struct sa_defrag_extent_backref *entry;
2070 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2072 ret = backref_comp(backref, entry);
2076 p = &(*p)->rb_right;
2079 rb_link_node(&backref->node, parent, p);
2080 rb_insert_color(&backref->node, root);
2084 * Note the backref might has changed, and in this case we just return 0.
2086 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2089 struct btrfs_file_extent_item *extent;
2090 struct btrfs_fs_info *fs_info;
2091 struct old_sa_defrag_extent *old = ctx;
2092 struct new_sa_defrag_extent *new = old->new;
2093 struct btrfs_path *path = new->path;
2094 struct btrfs_key key;
2095 struct btrfs_root *root;
2096 struct sa_defrag_extent_backref *backref;
2097 struct extent_buffer *leaf;
2098 struct inode *inode = new->inode;
2104 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2105 inum == btrfs_ino(inode))
2108 key.objectid = root_id;
2109 key.type = BTRFS_ROOT_ITEM_KEY;
2110 key.offset = (u64)-1;
2112 fs_info = BTRFS_I(inode)->root->fs_info;
2113 root = btrfs_read_fs_root_no_name(fs_info, &key);
2115 if (PTR_ERR(root) == -ENOENT)
2118 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2119 inum, offset, root_id);
2120 return PTR_ERR(root);
2123 key.objectid = inum;
2124 key.type = BTRFS_EXTENT_DATA_KEY;
2125 if (offset > (u64)-1 << 32)
2128 key.offset = offset;
2130 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2139 leaf = path->nodes[0];
2140 slot = path->slots[0];
2142 if (slot >= btrfs_header_nritems(leaf)) {
2143 ret = btrfs_next_leaf(root, path);
2146 } else if (ret > 0) {
2155 btrfs_item_key_to_cpu(leaf, &key, slot);
2157 if (key.objectid > inum)
2160 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2163 extent = btrfs_item_ptr(leaf, slot,
2164 struct btrfs_file_extent_item);
2166 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2170 * 'offset' refers to the exact key.offset,
2171 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2172 * (key.offset - extent_offset).
2174 if (key.offset != offset)
2177 extent_offset = btrfs_file_extent_offset(leaf, extent);
2178 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2180 if (extent_offset >= old->extent_offset + old->offset +
2181 old->len || extent_offset + num_bytes <=
2182 old->extent_offset + old->offset)
2189 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2195 backref->root_id = root_id;
2196 backref->inum = inum;
2197 backref->file_pos = offset;
2198 backref->num_bytes = num_bytes;
2199 backref->extent_offset = extent_offset;
2200 backref->generation = btrfs_file_extent_generation(leaf, extent);
2202 backref_insert(&new->root, backref);
2205 btrfs_release_path(path);
2210 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2211 struct new_sa_defrag_extent *new)
2213 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2214 struct old_sa_defrag_extent *old, *tmp;
2219 list_for_each_entry_safe(old, tmp, &new->head, list) {
2220 ret = iterate_inodes_from_logical(old->bytenr +
2221 old->extent_offset, fs_info,
2222 path, record_one_backref,
2224 BUG_ON(ret < 0 && ret != -ENOENT);
2226 /* no backref to be processed for this extent */
2228 list_del(&old->list);
2233 if (list_empty(&new->head))
2239 static int relink_is_mergable(struct extent_buffer *leaf,
2240 struct btrfs_file_extent_item *fi,
2243 if (btrfs_file_extent_disk_bytenr(leaf, fi) != disk_bytenr)
2246 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2249 if (btrfs_file_extent_compression(leaf, fi) ||
2250 btrfs_file_extent_encryption(leaf, fi) ||
2251 btrfs_file_extent_other_encoding(leaf, fi))
2258 * Note the backref might has changed, and in this case we just return 0.
2260 static noinline int relink_extent_backref(struct btrfs_path *path,
2261 struct sa_defrag_extent_backref *prev,
2262 struct sa_defrag_extent_backref *backref)
2264 struct btrfs_file_extent_item *extent;
2265 struct btrfs_file_extent_item *item;
2266 struct btrfs_ordered_extent *ordered;
2267 struct btrfs_trans_handle *trans;
2268 struct btrfs_fs_info *fs_info;
2269 struct btrfs_root *root;
2270 struct btrfs_key key;
2271 struct extent_buffer *leaf;
2272 struct old_sa_defrag_extent *old = backref->old;
2273 struct new_sa_defrag_extent *new = old->new;
2274 struct inode *src_inode = new->inode;
2275 struct inode *inode;
2276 struct extent_state *cached = NULL;
2285 if (prev && prev->root_id == backref->root_id &&
2286 prev->inum == backref->inum &&
2287 prev->file_pos + prev->num_bytes == backref->file_pos)
2290 /* step 1: get root */
2291 key.objectid = backref->root_id;
2292 key.type = BTRFS_ROOT_ITEM_KEY;
2293 key.offset = (u64)-1;
2295 fs_info = BTRFS_I(src_inode)->root->fs_info;
2296 index = srcu_read_lock(&fs_info->subvol_srcu);
2298 root = btrfs_read_fs_root_no_name(fs_info, &key);
2300 srcu_read_unlock(&fs_info->subvol_srcu, index);
2301 if (PTR_ERR(root) == -ENOENT)
2303 return PTR_ERR(root);
2306 /* step 2: get inode */
2307 key.objectid = backref->inum;
2308 key.type = BTRFS_INODE_ITEM_KEY;
2311 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2312 if (IS_ERR(inode)) {
2313 srcu_read_unlock(&fs_info->subvol_srcu, index);
2317 srcu_read_unlock(&fs_info->subvol_srcu, index);
2319 /* step 3: relink backref */
2320 lock_start = backref->file_pos;
2321 lock_end = backref->file_pos + backref->num_bytes - 1;
2322 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2325 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2327 btrfs_put_ordered_extent(ordered);
2331 trans = btrfs_join_transaction(root);
2332 if (IS_ERR(trans)) {
2333 ret = PTR_ERR(trans);
2337 key.objectid = backref->inum;
2338 key.type = BTRFS_EXTENT_DATA_KEY;
2339 key.offset = backref->file_pos;
2341 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2344 } else if (ret > 0) {
2349 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2350 struct btrfs_file_extent_item);
2352 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2353 backref->generation)
2356 btrfs_release_path(path);
2358 start = backref->file_pos;
2359 if (backref->extent_offset < old->extent_offset + old->offset)
2360 start += old->extent_offset + old->offset -
2361 backref->extent_offset;
2363 len = min(backref->extent_offset + backref->num_bytes,
2364 old->extent_offset + old->offset + old->len);
2365 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2367 ret = btrfs_drop_extents(trans, root, inode, start,
2372 key.objectid = btrfs_ino(inode);
2373 key.type = BTRFS_EXTENT_DATA_KEY;
2376 path->leave_spinning = 1;
2378 struct btrfs_file_extent_item *fi;
2380 struct btrfs_key found_key;
2382 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
2387 leaf = path->nodes[0];
2388 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2390 fi = btrfs_item_ptr(leaf, path->slots[0],
2391 struct btrfs_file_extent_item);
2392 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2394 if (relink_is_mergable(leaf, fi, new->bytenr) &&
2395 extent_len + found_key.offset == start) {
2396 btrfs_set_file_extent_num_bytes(leaf, fi,
2398 btrfs_mark_buffer_dirty(leaf);
2399 inode_add_bytes(inode, len);
2405 btrfs_release_path(path);
2410 ret = btrfs_insert_empty_item(trans, root, path, &key,
2413 btrfs_abort_transaction(trans, root, ret);
2417 leaf = path->nodes[0];
2418 item = btrfs_item_ptr(leaf, path->slots[0],
2419 struct btrfs_file_extent_item);
2420 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2421 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2422 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2423 btrfs_set_file_extent_num_bytes(leaf, item, len);
2424 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2425 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2426 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2427 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2428 btrfs_set_file_extent_encryption(leaf, item, 0);
2429 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2431 btrfs_mark_buffer_dirty(leaf);
2432 inode_add_bytes(inode, len);
2433 btrfs_release_path(path);
2435 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2437 backref->root_id, backref->inum,
2438 new->file_pos, 0); /* start - extent_offset */
2440 btrfs_abort_transaction(trans, root, ret);
2446 btrfs_release_path(path);
2447 path->leave_spinning = 0;
2448 btrfs_end_transaction(trans, root);
2450 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2456 static void relink_file_extents(struct new_sa_defrag_extent *new)
2458 struct btrfs_path *path;
2459 struct old_sa_defrag_extent *old, *tmp;
2460 struct sa_defrag_extent_backref *backref;
2461 struct sa_defrag_extent_backref *prev = NULL;
2462 struct inode *inode;
2463 struct btrfs_root *root;
2464 struct rb_node *node;
2468 root = BTRFS_I(inode)->root;
2470 path = btrfs_alloc_path();
2474 if (!record_extent_backrefs(path, new)) {
2475 btrfs_free_path(path);
2478 btrfs_release_path(path);
2481 node = rb_first(&new->root);
2484 rb_erase(node, &new->root);
2486 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2488 ret = relink_extent_backref(path, prev, backref);
2501 btrfs_free_path(path);
2503 list_for_each_entry_safe(old, tmp, &new->head, list) {
2504 list_del(&old->list);
2508 atomic_dec(&root->fs_info->defrag_running);
2509 wake_up(&root->fs_info->transaction_wait);
2514 static struct new_sa_defrag_extent *
2515 record_old_file_extents(struct inode *inode,
2516 struct btrfs_ordered_extent *ordered)
2518 struct btrfs_root *root = BTRFS_I(inode)->root;
2519 struct btrfs_path *path;
2520 struct btrfs_key key;
2521 struct old_sa_defrag_extent *old, *tmp;
2522 struct new_sa_defrag_extent *new;
2525 new = kmalloc(sizeof(*new), GFP_NOFS);
2530 new->file_pos = ordered->file_offset;
2531 new->len = ordered->len;
2532 new->bytenr = ordered->start;
2533 new->disk_len = ordered->disk_len;
2534 new->compress_type = ordered->compress_type;
2535 new->root = RB_ROOT;
2536 INIT_LIST_HEAD(&new->head);
2538 path = btrfs_alloc_path();
2542 key.objectid = btrfs_ino(inode);
2543 key.type = BTRFS_EXTENT_DATA_KEY;
2544 key.offset = new->file_pos;
2546 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2549 if (ret > 0 && path->slots[0] > 0)
2552 /* find out all the old extents for the file range */
2554 struct btrfs_file_extent_item *extent;
2555 struct extent_buffer *l;
2564 slot = path->slots[0];
2566 if (slot >= btrfs_header_nritems(l)) {
2567 ret = btrfs_next_leaf(root, path);
2575 btrfs_item_key_to_cpu(l, &key, slot);
2577 if (key.objectid != btrfs_ino(inode))
2579 if (key.type != BTRFS_EXTENT_DATA_KEY)
2581 if (key.offset >= new->file_pos + new->len)
2584 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2586 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2587 if (key.offset + num_bytes < new->file_pos)
2590 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2594 extent_offset = btrfs_file_extent_offset(l, extent);
2596 old = kmalloc(sizeof(*old), GFP_NOFS);
2600 offset = max(new->file_pos, key.offset);
2601 end = min(new->file_pos + new->len, key.offset + num_bytes);
2603 old->bytenr = disk_bytenr;
2604 old->extent_offset = extent_offset;
2605 old->offset = offset - key.offset;
2606 old->len = end - offset;
2609 list_add_tail(&old->list, &new->head);
2615 btrfs_free_path(path);
2616 atomic_inc(&root->fs_info->defrag_running);
2621 list_for_each_entry_safe(old, tmp, &new->head, list) {
2622 list_del(&old->list);
2626 btrfs_free_path(path);
2633 * helper function for btrfs_finish_ordered_io, this
2634 * just reads in some of the csum leaves to prime them into ram
2635 * before we start the transaction. It limits the amount of btree
2636 * reads required while inside the transaction.
2638 /* as ordered data IO finishes, this gets called so we can finish
2639 * an ordered extent if the range of bytes in the file it covers are
2642 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2644 struct inode *inode = ordered_extent->inode;
2645 struct btrfs_root *root = BTRFS_I(inode)->root;
2646 struct btrfs_trans_handle *trans = NULL;
2647 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2648 struct extent_state *cached_state = NULL;
2649 struct new_sa_defrag_extent *new = NULL;
2650 int compress_type = 0;
2654 nolock = btrfs_is_free_space_inode(inode);
2656 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2661 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2662 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2663 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2665 trans = btrfs_join_transaction_nolock(root);
2667 trans = btrfs_join_transaction(root);
2668 if (IS_ERR(trans)) {
2669 ret = PTR_ERR(trans);
2673 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2674 ret = btrfs_update_inode_fallback(trans, root, inode);
2675 if (ret) /* -ENOMEM or corruption */
2676 btrfs_abort_transaction(trans, root, ret);
2680 lock_extent_bits(io_tree, ordered_extent->file_offset,
2681 ordered_extent->file_offset + ordered_extent->len - 1,
2684 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2685 ordered_extent->file_offset + ordered_extent->len - 1,
2686 EXTENT_DEFRAG, 1, cached_state);
2688 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2689 if (last_snapshot >= BTRFS_I(inode)->generation)
2690 /* the inode is shared */
2691 new = record_old_file_extents(inode, ordered_extent);
2693 clear_extent_bit(io_tree, ordered_extent->file_offset,
2694 ordered_extent->file_offset + ordered_extent->len - 1,
2695 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2699 trans = btrfs_join_transaction_nolock(root);
2701 trans = btrfs_join_transaction(root);
2702 if (IS_ERR(trans)) {
2703 ret = PTR_ERR(trans);
2707 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2709 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2710 compress_type = ordered_extent->compress_type;
2711 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2712 BUG_ON(compress_type);
2713 ret = btrfs_mark_extent_written(trans, inode,
2714 ordered_extent->file_offset,
2715 ordered_extent->file_offset +
2716 ordered_extent->len);
2718 BUG_ON(root == root->fs_info->tree_root);
2719 ret = insert_reserved_file_extent(trans, inode,
2720 ordered_extent->file_offset,
2721 ordered_extent->start,
2722 ordered_extent->disk_len,
2723 ordered_extent->len,
2724 ordered_extent->len,
2725 compress_type, 0, 0,
2726 BTRFS_FILE_EXTENT_REG);
2728 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2729 ordered_extent->file_offset, ordered_extent->len,
2732 btrfs_abort_transaction(trans, root, ret);
2736 add_pending_csums(trans, inode, ordered_extent->file_offset,
2737 &ordered_extent->list);
2739 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2740 ret = btrfs_update_inode_fallback(trans, root, inode);
2741 if (ret) { /* -ENOMEM or corruption */
2742 btrfs_abort_transaction(trans, root, ret);
2747 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2748 ordered_extent->file_offset +
2749 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2751 if (root != root->fs_info->tree_root)
2752 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2754 btrfs_end_transaction(trans, root);
2757 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
2758 ordered_extent->file_offset +
2759 ordered_extent->len - 1, NULL, GFP_NOFS);
2762 * If the ordered extent had an IOERR or something else went
2763 * wrong we need to return the space for this ordered extent
2764 * back to the allocator.
2766 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2767 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2768 btrfs_free_reserved_extent(root, ordered_extent->start,
2769 ordered_extent->disk_len);
2774 * This needs to be done to make sure anybody waiting knows we are done
2775 * updating everything for this ordered extent.
2777 btrfs_remove_ordered_extent(inode, ordered_extent);
2779 /* for snapshot-aware defrag */
2781 relink_file_extents(new);
2784 btrfs_put_ordered_extent(ordered_extent);
2785 /* once for the tree */
2786 btrfs_put_ordered_extent(ordered_extent);
2791 static void finish_ordered_fn(struct btrfs_work *work)
2793 struct btrfs_ordered_extent *ordered_extent;
2794 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2795 btrfs_finish_ordered_io(ordered_extent);
2798 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2799 struct extent_state *state, int uptodate)
2801 struct inode *inode = page->mapping->host;
2802 struct btrfs_root *root = BTRFS_I(inode)->root;
2803 struct btrfs_ordered_extent *ordered_extent = NULL;
2804 struct btrfs_workers *workers;
2806 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2808 ClearPagePrivate2(page);
2809 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2810 end - start + 1, uptodate))
2813 ordered_extent->work.func = finish_ordered_fn;
2814 ordered_extent->work.flags = 0;
2816 if (btrfs_is_free_space_inode(inode))
2817 workers = &root->fs_info->endio_freespace_worker;
2819 workers = &root->fs_info->endio_write_workers;
2820 btrfs_queue_worker(workers, &ordered_extent->work);
2826 * when reads are done, we need to check csums to verify the data is correct
2827 * if there's a match, we allow the bio to finish. If not, the code in
2828 * extent_io.c will try to find good copies for us.
2830 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2831 struct extent_state *state, int mirror)
2833 size_t offset = start - page_offset(page);
2834 struct inode *inode = page->mapping->host;
2835 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2837 u64 private = ~(u32)0;
2839 struct btrfs_root *root = BTRFS_I(inode)->root;
2841 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2842 DEFAULT_RATELIMIT_BURST);
2844 if (PageChecked(page)) {
2845 ClearPageChecked(page);
2849 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2852 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2853 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2854 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2859 if (state && state->start == start) {
2860 private = state->private;
2863 ret = get_state_private(io_tree, start, &private);
2865 kaddr = kmap_atomic(page);
2869 csum = btrfs_csum_data(kaddr + offset, csum, end - start + 1);
2870 btrfs_csum_final(csum, (char *)&csum);
2871 if (csum != private)
2874 kunmap_atomic(kaddr);
2879 if (__ratelimit(&_rs))
2880 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u private %llu",
2881 (unsigned long long)btrfs_ino(page->mapping->host),
2882 (unsigned long long)start, csum,
2883 (unsigned long long)private);
2884 memset(kaddr + offset, 1, end - start + 1);
2885 flush_dcache_page(page);
2886 kunmap_atomic(kaddr);
2892 struct delayed_iput {
2893 struct list_head list;
2894 struct inode *inode;
2897 /* JDM: If this is fs-wide, why can't we add a pointer to
2898 * btrfs_inode instead and avoid the allocation? */
2899 void btrfs_add_delayed_iput(struct inode *inode)
2901 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2902 struct delayed_iput *delayed;
2904 if (atomic_add_unless(&inode->i_count, -1, 1))
2907 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2908 delayed->inode = inode;
2910 spin_lock(&fs_info->delayed_iput_lock);
2911 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2912 spin_unlock(&fs_info->delayed_iput_lock);
2915 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2918 struct btrfs_fs_info *fs_info = root->fs_info;
2919 struct delayed_iput *delayed;
2922 spin_lock(&fs_info->delayed_iput_lock);
2923 empty = list_empty(&fs_info->delayed_iputs);
2924 spin_unlock(&fs_info->delayed_iput_lock);
2928 spin_lock(&fs_info->delayed_iput_lock);
2929 list_splice_init(&fs_info->delayed_iputs, &list);
2930 spin_unlock(&fs_info->delayed_iput_lock);
2932 while (!list_empty(&list)) {
2933 delayed = list_entry(list.next, struct delayed_iput, list);
2934 list_del(&delayed->list);
2935 iput(delayed->inode);
2941 * This is called in transaction commit time. If there are no orphan
2942 * files in the subvolume, it removes orphan item and frees block_rsv
2945 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2946 struct btrfs_root *root)
2948 struct btrfs_block_rsv *block_rsv;
2951 if (atomic_read(&root->orphan_inodes) ||
2952 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2955 spin_lock(&root->orphan_lock);
2956 if (atomic_read(&root->orphan_inodes)) {
2957 spin_unlock(&root->orphan_lock);
2961 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2962 spin_unlock(&root->orphan_lock);
2966 block_rsv = root->orphan_block_rsv;
2967 root->orphan_block_rsv = NULL;
2968 spin_unlock(&root->orphan_lock);
2970 if (root->orphan_item_inserted &&
2971 btrfs_root_refs(&root->root_item) > 0) {
2972 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2973 root->root_key.objectid);
2975 root->orphan_item_inserted = 0;
2979 WARN_ON(block_rsv->size > 0);
2980 btrfs_free_block_rsv(root, block_rsv);
2985 * This creates an orphan entry for the given inode in case something goes
2986 * wrong in the middle of an unlink/truncate.
2988 * NOTE: caller of this function should reserve 5 units of metadata for
2991 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2993 struct btrfs_root *root = BTRFS_I(inode)->root;
2994 struct btrfs_block_rsv *block_rsv = NULL;
2999 if (!root->orphan_block_rsv) {
3000 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3005 spin_lock(&root->orphan_lock);
3006 if (!root->orphan_block_rsv) {
3007 root->orphan_block_rsv = block_rsv;
3008 } else if (block_rsv) {
3009 btrfs_free_block_rsv(root, block_rsv);
3013 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3014 &BTRFS_I(inode)->runtime_flags)) {
3017 * For proper ENOSPC handling, we should do orphan
3018 * cleanup when mounting. But this introduces backward
3019 * compatibility issue.
3021 if (!xchg(&root->orphan_item_inserted, 1))
3027 atomic_inc(&root->orphan_inodes);
3030 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3031 &BTRFS_I(inode)->runtime_flags))
3033 spin_unlock(&root->orphan_lock);
3035 /* grab metadata reservation from transaction handle */
3037 ret = btrfs_orphan_reserve_metadata(trans, inode);
3038 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3041 /* insert an orphan item to track this unlinked/truncated file */
3043 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3044 if (ret && ret != -EEXIST) {
3045 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3046 &BTRFS_I(inode)->runtime_flags);
3047 btrfs_abort_transaction(trans, root, ret);
3053 /* insert an orphan item to track subvolume contains orphan files */
3055 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3056 root->root_key.objectid);
3057 if (ret && ret != -EEXIST) {
3058 btrfs_abort_transaction(trans, root, ret);
3066 * We have done the truncate/delete so we can go ahead and remove the orphan
3067 * item for this particular inode.
3069 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3070 struct inode *inode)
3072 struct btrfs_root *root = BTRFS_I(inode)->root;
3073 int delete_item = 0;
3074 int release_rsv = 0;
3077 spin_lock(&root->orphan_lock);
3078 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3079 &BTRFS_I(inode)->runtime_flags))
3082 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3083 &BTRFS_I(inode)->runtime_flags))
3085 spin_unlock(&root->orphan_lock);
3087 if (trans && delete_item) {
3088 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
3089 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3093 btrfs_orphan_release_metadata(inode);
3094 atomic_dec(&root->orphan_inodes);
3101 * this cleans up any orphans that may be left on the list from the last use
3104 int btrfs_orphan_cleanup(struct btrfs_root *root)
3106 struct btrfs_path *path;
3107 struct extent_buffer *leaf;
3108 struct btrfs_key key, found_key;
3109 struct btrfs_trans_handle *trans;
3110 struct inode *inode;
3111 u64 last_objectid = 0;
3112 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3114 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3117 path = btrfs_alloc_path();
3124 key.objectid = BTRFS_ORPHAN_OBJECTID;
3125 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3126 key.offset = (u64)-1;
3129 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3134 * if ret == 0 means we found what we were searching for, which
3135 * is weird, but possible, so only screw with path if we didn't
3136 * find the key and see if we have stuff that matches
3140 if (path->slots[0] == 0)
3145 /* pull out the item */
3146 leaf = path->nodes[0];
3147 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3149 /* make sure the item matches what we want */
3150 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3152 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3155 /* release the path since we're done with it */
3156 btrfs_release_path(path);
3159 * this is where we are basically btrfs_lookup, without the
3160 * crossing root thing. we store the inode number in the
3161 * offset of the orphan item.
3164 if (found_key.offset == last_objectid) {
3165 btrfs_err(root->fs_info,
3166 "Error removing orphan entry, stopping orphan cleanup");
3171 last_objectid = found_key.offset;
3173 found_key.objectid = found_key.offset;
3174 found_key.type = BTRFS_INODE_ITEM_KEY;
3175 found_key.offset = 0;
3176 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3177 ret = PTR_ERR_OR_ZERO(inode);
3178 if (ret && ret != -ESTALE)
3181 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3182 struct btrfs_root *dead_root;
3183 struct btrfs_fs_info *fs_info = root->fs_info;
3184 int is_dead_root = 0;
3187 * this is an orphan in the tree root. Currently these
3188 * could come from 2 sources:
3189 * a) a snapshot deletion in progress
3190 * b) a free space cache inode
3191 * We need to distinguish those two, as the snapshot
3192 * orphan must not get deleted.
3193 * find_dead_roots already ran before us, so if this
3194 * is a snapshot deletion, we should find the root
3195 * in the dead_roots list
3197 spin_lock(&fs_info->trans_lock);
3198 list_for_each_entry(dead_root, &fs_info->dead_roots,
3200 if (dead_root->root_key.objectid ==
3201 found_key.objectid) {
3206 spin_unlock(&fs_info->trans_lock);
3208 /* prevent this orphan from being found again */
3209 key.offset = found_key.objectid - 1;
3214 * Inode is already gone but the orphan item is still there,
3215 * kill the orphan item.
3217 if (ret == -ESTALE) {
3218 trans = btrfs_start_transaction(root, 1);
3219 if (IS_ERR(trans)) {
3220 ret = PTR_ERR(trans);
3223 btrfs_debug(root->fs_info, "auto deleting %Lu",
3224 found_key.objectid);
3225 ret = btrfs_del_orphan_item(trans, root,
3226 found_key.objectid);
3227 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3228 btrfs_end_transaction(trans, root);
3233 * add this inode to the orphan list so btrfs_orphan_del does
3234 * the proper thing when we hit it
3236 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3237 &BTRFS_I(inode)->runtime_flags);
3238 atomic_inc(&root->orphan_inodes);
3240 /* if we have links, this was a truncate, lets do that */
3241 if (inode->i_nlink) {
3242 if (!S_ISREG(inode->i_mode)) {
3249 /* 1 for the orphan item deletion. */
3250 trans = btrfs_start_transaction(root, 1);
3251 if (IS_ERR(trans)) {
3253 ret = PTR_ERR(trans);
3256 ret = btrfs_orphan_add(trans, inode);
3257 btrfs_end_transaction(trans, root);
3263 ret = btrfs_truncate(inode);
3265 btrfs_orphan_del(NULL, inode);
3270 /* this will do delete_inode and everything for us */
3275 /* release the path since we're done with it */
3276 btrfs_release_path(path);
3278 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3280 if (root->orphan_block_rsv)
3281 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3284 if (root->orphan_block_rsv || root->orphan_item_inserted) {
3285 trans = btrfs_join_transaction(root);
3287 btrfs_end_transaction(trans, root);
3291 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3293 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3297 btrfs_crit(root->fs_info,
3298 "could not do orphan cleanup %d", ret);
3299 btrfs_free_path(path);
3304 * very simple check to peek ahead in the leaf looking for xattrs. If we
3305 * don't find any xattrs, we know there can't be any acls.
3307 * slot is the slot the inode is in, objectid is the objectid of the inode
3309 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3310 int slot, u64 objectid)
3312 u32 nritems = btrfs_header_nritems(leaf);
3313 struct btrfs_key found_key;
3314 static u64 xattr_access = 0;
3315 static u64 xattr_default = 0;
3318 if (!xattr_access) {
3319 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3320 strlen(POSIX_ACL_XATTR_ACCESS));
3321 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3322 strlen(POSIX_ACL_XATTR_DEFAULT));
3326 while (slot < nritems) {
3327 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3329 /* we found a different objectid, there must not be acls */
3330 if (found_key.objectid != objectid)
3333 /* we found an xattr, assume we've got an acl */
3334 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3335 if (found_key.offset == xattr_access ||
3336 found_key.offset == xattr_default)
3341 * we found a key greater than an xattr key, there can't
3342 * be any acls later on
3344 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3351 * it goes inode, inode backrefs, xattrs, extents,
3352 * so if there are a ton of hard links to an inode there can
3353 * be a lot of backrefs. Don't waste time searching too hard,
3354 * this is just an optimization
3359 /* we hit the end of the leaf before we found an xattr or
3360 * something larger than an xattr. We have to assume the inode
3367 * read an inode from the btree into the in-memory inode
3369 static void btrfs_read_locked_inode(struct inode *inode)
3371 struct btrfs_path *path;
3372 struct extent_buffer *leaf;
3373 struct btrfs_inode_item *inode_item;
3374 struct btrfs_timespec *tspec;
3375 struct btrfs_root *root = BTRFS_I(inode)->root;
3376 struct btrfs_key location;
3380 bool filled = false;
3382 ret = btrfs_fill_inode(inode, &rdev);
3386 path = btrfs_alloc_path();
3390 path->leave_spinning = 1;
3391 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3393 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3397 leaf = path->nodes[0];
3402 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3403 struct btrfs_inode_item);
3404 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3405 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3406 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3407 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3408 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3410 tspec = btrfs_inode_atime(inode_item);
3411 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3412 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3414 tspec = btrfs_inode_mtime(inode_item);
3415 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3416 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3418 tspec = btrfs_inode_ctime(inode_item);
3419 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3420 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3422 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3423 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3424 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3427 * If we were modified in the current generation and evicted from memory
3428 * and then re-read we need to do a full sync since we don't have any
3429 * idea about which extents were modified before we were evicted from
3432 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3433 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3434 &BTRFS_I(inode)->runtime_flags);
3436 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3437 inode->i_generation = BTRFS_I(inode)->generation;
3439 rdev = btrfs_inode_rdev(leaf, inode_item);
3441 BTRFS_I(inode)->index_cnt = (u64)-1;
3442 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3445 * try to precache a NULL acl entry for files that don't have
3446 * any xattrs or acls
3448 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3451 cache_no_acl(inode);
3453 btrfs_free_path(path);
3455 switch (inode->i_mode & S_IFMT) {
3457 inode->i_mapping->a_ops = &btrfs_aops;
3458 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3459 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3460 inode->i_fop = &btrfs_file_operations;
3461 inode->i_op = &btrfs_file_inode_operations;
3464 inode->i_fop = &btrfs_dir_file_operations;
3465 if (root == root->fs_info->tree_root)
3466 inode->i_op = &btrfs_dir_ro_inode_operations;
3468 inode->i_op = &btrfs_dir_inode_operations;
3471 inode->i_op = &btrfs_symlink_inode_operations;
3472 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3473 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3476 inode->i_op = &btrfs_special_inode_operations;
3477 init_special_inode(inode, inode->i_mode, rdev);
3481 btrfs_update_iflags(inode);
3485 btrfs_free_path(path);
3486 make_bad_inode(inode);
3490 * given a leaf and an inode, copy the inode fields into the leaf
3492 static void fill_inode_item(struct btrfs_trans_handle *trans,
3493 struct extent_buffer *leaf,
3494 struct btrfs_inode_item *item,
3495 struct inode *inode)
3497 struct btrfs_map_token token;
3499 btrfs_init_map_token(&token);
3501 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3502 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3503 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3505 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3506 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3508 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3509 inode->i_atime.tv_sec, &token);
3510 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3511 inode->i_atime.tv_nsec, &token);
3513 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3514 inode->i_mtime.tv_sec, &token);
3515 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3516 inode->i_mtime.tv_nsec, &token);
3518 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3519 inode->i_ctime.tv_sec, &token);
3520 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3521 inode->i_ctime.tv_nsec, &token);
3523 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3525 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3527 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3528 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3529 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3530 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3531 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3535 * copy everything in the in-memory inode into the btree.
3537 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3538 struct btrfs_root *root, struct inode *inode)
3540 struct btrfs_inode_item *inode_item;
3541 struct btrfs_path *path;
3542 struct extent_buffer *leaf;
3545 path = btrfs_alloc_path();
3549 path->leave_spinning = 1;
3550 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3558 btrfs_unlock_up_safe(path, 1);
3559 leaf = path->nodes[0];
3560 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3561 struct btrfs_inode_item);
3563 fill_inode_item(trans, leaf, inode_item, inode);
3564 btrfs_mark_buffer_dirty(leaf);
3565 btrfs_set_inode_last_trans(trans, inode);
3568 btrfs_free_path(path);
3573 * copy everything in the in-memory inode into the btree.
3575 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3576 struct btrfs_root *root, struct inode *inode)
3581 * If the inode is a free space inode, we can deadlock during commit
3582 * if we put it into the delayed code.
3584 * The data relocation inode should also be directly updated
3587 if (!btrfs_is_free_space_inode(inode)
3588 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3589 btrfs_update_root_times(trans, root);
3591 ret = btrfs_delayed_update_inode(trans, root, inode);
3593 btrfs_set_inode_last_trans(trans, inode);
3597 return btrfs_update_inode_item(trans, root, inode);
3600 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3601 struct btrfs_root *root,
3602 struct inode *inode)
3606 ret = btrfs_update_inode(trans, root, inode);
3608 return btrfs_update_inode_item(trans, root, inode);
3613 * unlink helper that gets used here in inode.c and in the tree logging
3614 * recovery code. It remove a link in a directory with a given name, and
3615 * also drops the back refs in the inode to the directory
3617 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3618 struct btrfs_root *root,
3619 struct inode *dir, struct inode *inode,
3620 const char *name, int name_len)
3622 struct btrfs_path *path;
3624 struct extent_buffer *leaf;
3625 struct btrfs_dir_item *di;
3626 struct btrfs_key key;
3628 u64 ino = btrfs_ino(inode);
3629 u64 dir_ino = btrfs_ino(dir);
3631 path = btrfs_alloc_path();
3637 path->leave_spinning = 1;
3638 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3639 name, name_len, -1);
3648 leaf = path->nodes[0];
3649 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3650 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3653 btrfs_release_path(path);
3655 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3658 btrfs_info(root->fs_info,
3659 "failed to delete reference to %.*s, inode %llu parent %llu",
3661 (unsigned long long)ino, (unsigned long long)dir_ino);
3662 btrfs_abort_transaction(trans, root, ret);
3666 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3668 btrfs_abort_transaction(trans, root, ret);
3672 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3674 if (ret != 0 && ret != -ENOENT) {
3675 btrfs_abort_transaction(trans, root, ret);
3679 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3684 btrfs_abort_transaction(trans, root, ret);
3686 btrfs_free_path(path);
3690 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3691 inode_inc_iversion(inode);
3692 inode_inc_iversion(dir);
3693 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3694 ret = btrfs_update_inode(trans, root, dir);
3699 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3700 struct btrfs_root *root,
3701 struct inode *dir, struct inode *inode,
3702 const char *name, int name_len)
3705 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3707 btrfs_drop_nlink(inode);
3708 ret = btrfs_update_inode(trans, root, inode);
3714 * helper to start transaction for unlink and rmdir.
3716 * unlink and rmdir are special in btrfs, they do not always free space, so
3717 * if we cannot make our reservations the normal way try and see if there is
3718 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3719 * allow the unlink to occur.
3721 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3723 struct btrfs_trans_handle *trans;
3724 struct btrfs_root *root = BTRFS_I(dir)->root;
3728 * 1 for the possible orphan item
3729 * 1 for the dir item
3730 * 1 for the dir index
3731 * 1 for the inode ref
3734 trans = btrfs_start_transaction(root, 5);
3735 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3738 if (PTR_ERR(trans) == -ENOSPC) {
3739 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3741 trans = btrfs_start_transaction(root, 0);
3744 ret = btrfs_cond_migrate_bytes(root->fs_info,
3745 &root->fs_info->trans_block_rsv,
3748 btrfs_end_transaction(trans, root);
3749 return ERR_PTR(ret);
3751 trans->block_rsv = &root->fs_info->trans_block_rsv;
3752 trans->bytes_reserved = num_bytes;
3757 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3759 struct btrfs_root *root = BTRFS_I(dir)->root;
3760 struct btrfs_trans_handle *trans;
3761 struct inode *inode = dentry->d_inode;
3764 trans = __unlink_start_trans(dir);
3766 return PTR_ERR(trans);
3768 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3770 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3771 dentry->d_name.name, dentry->d_name.len);
3775 if (inode->i_nlink == 0) {
3776 ret = btrfs_orphan_add(trans, inode);
3782 btrfs_end_transaction(trans, root);
3783 btrfs_btree_balance_dirty(root);
3787 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3788 struct btrfs_root *root,
3789 struct inode *dir, u64 objectid,
3790 const char *name, int name_len)
3792 struct btrfs_path *path;
3793 struct extent_buffer *leaf;
3794 struct btrfs_dir_item *di;
3795 struct btrfs_key key;
3798 u64 dir_ino = btrfs_ino(dir);
3800 path = btrfs_alloc_path();
3804 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3805 name, name_len, -1);
3806 if (IS_ERR_OR_NULL(di)) {
3814 leaf = path->nodes[0];
3815 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3816 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3817 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3819 btrfs_abort_transaction(trans, root, ret);
3822 btrfs_release_path(path);
3824 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3825 objectid, root->root_key.objectid,
3826 dir_ino, &index, name, name_len);
3828 if (ret != -ENOENT) {
3829 btrfs_abort_transaction(trans, root, ret);
3832 di = btrfs_search_dir_index_item(root, path, dir_ino,
3834 if (IS_ERR_OR_NULL(di)) {
3839 btrfs_abort_transaction(trans, root, ret);
3843 leaf = path->nodes[0];
3844 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3845 btrfs_release_path(path);
3848 btrfs_release_path(path);
3850 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3852 btrfs_abort_transaction(trans, root, ret);
3856 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3857 inode_inc_iversion(dir);
3858 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3859 ret = btrfs_update_inode_fallback(trans, root, dir);
3861 btrfs_abort_transaction(trans, root, ret);
3863 btrfs_free_path(path);
3867 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3869 struct inode *inode = dentry->d_inode;
3871 struct btrfs_root *root = BTRFS_I(dir)->root;
3872 struct btrfs_trans_handle *trans;
3874 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3876 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3879 trans = __unlink_start_trans(dir);
3881 return PTR_ERR(trans);
3883 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3884 err = btrfs_unlink_subvol(trans, root, dir,
3885 BTRFS_I(inode)->location.objectid,
3886 dentry->d_name.name,
3887 dentry->d_name.len);
3891 err = btrfs_orphan_add(trans, inode);
3895 /* now the directory is empty */
3896 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3897 dentry->d_name.name, dentry->d_name.len);
3899 btrfs_i_size_write(inode, 0);
3901 btrfs_end_transaction(trans, root);
3902 btrfs_btree_balance_dirty(root);
3908 * this can truncate away extent items, csum items and directory items.
3909 * It starts at a high offset and removes keys until it can't find
3910 * any higher than new_size
3912 * csum items that cross the new i_size are truncated to the new size
3915 * min_type is the minimum key type to truncate down to. If set to 0, this
3916 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3918 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3919 struct btrfs_root *root,
3920 struct inode *inode,
3921 u64 new_size, u32 min_type)
3923 struct btrfs_path *path;
3924 struct extent_buffer *leaf;
3925 struct btrfs_file_extent_item *fi;
3926 struct btrfs_key key;
3927 struct btrfs_key found_key;
3928 u64 extent_start = 0;
3929 u64 extent_num_bytes = 0;
3930 u64 extent_offset = 0;
3932 u32 found_type = (u8)-1;
3935 int pending_del_nr = 0;
3936 int pending_del_slot = 0;
3937 int extent_type = -1;
3940 u64 ino = btrfs_ino(inode);
3942 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3944 path = btrfs_alloc_path();
3950 * We want to drop from the next block forward in case this new size is
3951 * not block aligned since we will be keeping the last block of the
3952 * extent just the way it is.
3954 if (root->ref_cows || root == root->fs_info->tree_root)
3955 btrfs_drop_extent_cache(inode, ALIGN(new_size,
3956 root->sectorsize), (u64)-1, 0);
3959 * This function is also used to drop the items in the log tree before
3960 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3961 * it is used to drop the loged items. So we shouldn't kill the delayed
3964 if (min_type == 0 && root == BTRFS_I(inode)->root)
3965 btrfs_kill_delayed_inode_items(inode);
3968 key.offset = (u64)-1;
3972 path->leave_spinning = 1;
3973 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3980 /* there are no items in the tree for us to truncate, we're
3983 if (path->slots[0] == 0)
3990 leaf = path->nodes[0];
3991 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3992 found_type = btrfs_key_type(&found_key);
3994 if (found_key.objectid != ino)
3997 if (found_type < min_type)
4000 item_end = found_key.offset;
4001 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4002 fi = btrfs_item_ptr(leaf, path->slots[0],
4003 struct btrfs_file_extent_item);
4004 extent_type = btrfs_file_extent_type(leaf, fi);
4005 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4007 btrfs_file_extent_num_bytes(leaf, fi);
4008 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4009 item_end += btrfs_file_extent_inline_len(leaf,
4014 if (found_type > min_type) {
4017 if (item_end < new_size)
4019 if (found_key.offset >= new_size)
4025 /* FIXME, shrink the extent if the ref count is only 1 */
4026 if (found_type != BTRFS_EXTENT_DATA_KEY)
4029 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4031 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4033 u64 orig_num_bytes =
4034 btrfs_file_extent_num_bytes(leaf, fi);
4035 extent_num_bytes = ALIGN(new_size -
4038 btrfs_set_file_extent_num_bytes(leaf, fi,
4040 num_dec = (orig_num_bytes -
4042 if (root->ref_cows && extent_start != 0)
4043 inode_sub_bytes(inode, num_dec);
4044 btrfs_mark_buffer_dirty(leaf);
4047 btrfs_file_extent_disk_num_bytes(leaf,
4049 extent_offset = found_key.offset -
4050 btrfs_file_extent_offset(leaf, fi);
4052 /* FIXME blocksize != 4096 */
4053 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4054 if (extent_start != 0) {
4057 inode_sub_bytes(inode, num_dec);
4060 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4062 * we can't truncate inline items that have had
4066 btrfs_file_extent_compression(leaf, fi) == 0 &&
4067 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4068 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4069 u32 size = new_size - found_key.offset;
4071 if (root->ref_cows) {
4072 inode_sub_bytes(inode, item_end + 1 -
4076 btrfs_file_extent_calc_inline_size(size);
4077 btrfs_truncate_item(root, path, size, 1);
4078 } else if (root->ref_cows) {
4079 inode_sub_bytes(inode, item_end + 1 -
4085 if (!pending_del_nr) {
4086 /* no pending yet, add ourselves */
4087 pending_del_slot = path->slots[0];
4089 } else if (pending_del_nr &&
4090 path->slots[0] + 1 == pending_del_slot) {
4091 /* hop on the pending chunk */
4093 pending_del_slot = path->slots[0];
4100 if (found_extent && (root->ref_cows ||
4101 root == root->fs_info->tree_root)) {
4102 btrfs_set_path_blocking(path);
4103 ret = btrfs_free_extent(trans, root, extent_start,
4104 extent_num_bytes, 0,
4105 btrfs_header_owner(leaf),
4106 ino, extent_offset, 0);
4110 if (found_type == BTRFS_INODE_ITEM_KEY)
4113 if (path->slots[0] == 0 ||
4114 path->slots[0] != pending_del_slot) {
4115 if (pending_del_nr) {
4116 ret = btrfs_del_items(trans, root, path,
4120 btrfs_abort_transaction(trans,
4126 btrfs_release_path(path);
4133 if (pending_del_nr) {
4134 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4137 btrfs_abort_transaction(trans, root, ret);
4140 btrfs_free_path(path);
4145 * btrfs_truncate_page - read, zero a chunk and write a page
4146 * @inode - inode that we're zeroing
4147 * @from - the offset to start zeroing
4148 * @len - the length to zero, 0 to zero the entire range respective to the
4150 * @front - zero up to the offset instead of from the offset on
4152 * This will find the page for the "from" offset and cow the page and zero the
4153 * part we want to zero. This is used with truncate and hole punching.
4155 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4158 struct address_space *mapping = inode->i_mapping;
4159 struct btrfs_root *root = BTRFS_I(inode)->root;
4160 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4161 struct btrfs_ordered_extent *ordered;
4162 struct extent_state *cached_state = NULL;
4164 u32 blocksize = root->sectorsize;
4165 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4166 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4168 gfp_t mask = btrfs_alloc_write_mask(mapping);
4173 if ((offset & (blocksize - 1)) == 0 &&
4174 (!len || ((len & (blocksize - 1)) == 0)))
4176 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4181 page = find_or_create_page(mapping, index, mask);
4183 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4188 page_start = page_offset(page);
4189 page_end = page_start + PAGE_CACHE_SIZE - 1;
4191 if (!PageUptodate(page)) {
4192 ret = btrfs_readpage(NULL, page);
4194 if (page->mapping != mapping) {
4196 page_cache_release(page);
4199 if (!PageUptodate(page)) {
4204 wait_on_page_writeback(page);
4206 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4207 set_page_extent_mapped(page);
4209 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4211 unlock_extent_cached(io_tree, page_start, page_end,
4212 &cached_state, GFP_NOFS);
4214 page_cache_release(page);
4215 btrfs_start_ordered_extent(inode, ordered, 1);
4216 btrfs_put_ordered_extent(ordered);
4220 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4221 EXTENT_DIRTY | EXTENT_DELALLOC |
4222 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4223 0, 0, &cached_state, GFP_NOFS);
4225 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4228 unlock_extent_cached(io_tree, page_start, page_end,
4229 &cached_state, GFP_NOFS);
4233 if (offset != PAGE_CACHE_SIZE) {
4235 len = PAGE_CACHE_SIZE - offset;
4238 memset(kaddr, 0, offset);
4240 memset(kaddr + offset, 0, len);
4241 flush_dcache_page(page);
4244 ClearPageChecked(page);
4245 set_page_dirty(page);
4246 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4251 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4253 page_cache_release(page);
4259 * This function puts in dummy file extents for the area we're creating a hole
4260 * for. So if we are truncating this file to a larger size we need to insert
4261 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4262 * the range between oldsize and size
4264 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4266 struct btrfs_trans_handle *trans;
4267 struct btrfs_root *root = BTRFS_I(inode)->root;
4268 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4269 struct extent_map *em = NULL;
4270 struct extent_state *cached_state = NULL;
4271 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4272 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4273 u64 block_end = ALIGN(size, root->sectorsize);
4280 * If our size started in the middle of a page we need to zero out the
4281 * rest of the page before we expand the i_size, otherwise we could
4282 * expose stale data.
4284 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4288 if (size <= hole_start)
4292 struct btrfs_ordered_extent *ordered;
4293 btrfs_wait_ordered_range(inode, hole_start,
4294 block_end - hole_start);
4295 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4297 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
4300 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4301 &cached_state, GFP_NOFS);
4302 btrfs_put_ordered_extent(ordered);
4305 cur_offset = hole_start;
4307 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4308 block_end - cur_offset, 0);
4314 last_byte = min(extent_map_end(em), block_end);
4315 last_byte = ALIGN(last_byte , root->sectorsize);
4316 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4317 struct extent_map *hole_em;
4318 hole_size = last_byte - cur_offset;
4320 trans = btrfs_start_transaction(root, 3);
4321 if (IS_ERR(trans)) {
4322 err = PTR_ERR(trans);
4326 err = btrfs_drop_extents(trans, root, inode,
4328 cur_offset + hole_size, 1);
4330 btrfs_abort_transaction(trans, root, err);
4331 btrfs_end_transaction(trans, root);
4335 err = btrfs_insert_file_extent(trans, root,
4336 btrfs_ino(inode), cur_offset, 0,
4337 0, hole_size, 0, hole_size,
4340 btrfs_abort_transaction(trans, root, err);
4341 btrfs_end_transaction(trans, root);
4345 btrfs_drop_extent_cache(inode, cur_offset,
4346 cur_offset + hole_size - 1, 0);
4347 hole_em = alloc_extent_map();
4349 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4350 &BTRFS_I(inode)->runtime_flags);
4353 hole_em->start = cur_offset;
4354 hole_em->len = hole_size;
4355 hole_em->orig_start = cur_offset;
4357 hole_em->block_start = EXTENT_MAP_HOLE;
4358 hole_em->block_len = 0;
4359 hole_em->orig_block_len = 0;
4360 hole_em->ram_bytes = hole_size;
4361 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4362 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4363 hole_em->generation = trans->transid;
4366 write_lock(&em_tree->lock);
4367 err = add_extent_mapping(em_tree, hole_em, 1);
4368 write_unlock(&em_tree->lock);
4371 btrfs_drop_extent_cache(inode, cur_offset,
4375 free_extent_map(hole_em);
4377 btrfs_update_inode(trans, root, inode);
4378 btrfs_end_transaction(trans, root);
4380 free_extent_map(em);
4382 cur_offset = last_byte;
4383 if (cur_offset >= block_end)
4387 free_extent_map(em);
4388 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4393 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4395 struct btrfs_root *root = BTRFS_I(inode)->root;
4396 struct btrfs_trans_handle *trans;
4397 loff_t oldsize = i_size_read(inode);
4398 loff_t newsize = attr->ia_size;
4399 int mask = attr->ia_valid;
4403 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4404 * special case where we need to update the times despite not having
4405 * these flags set. For all other operations the VFS set these flags
4406 * explicitly if it wants a timestamp update.
4408 if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4409 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4411 if (newsize > oldsize) {
4412 truncate_pagecache(inode, oldsize, newsize);
4413 ret = btrfs_cont_expand(inode, oldsize, newsize);
4417 trans = btrfs_start_transaction(root, 1);
4419 return PTR_ERR(trans);
4421 i_size_write(inode, newsize);
4422 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4423 ret = btrfs_update_inode(trans, root, inode);
4424 btrfs_end_transaction(trans, root);
4428 * We're truncating a file that used to have good data down to
4429 * zero. Make sure it gets into the ordered flush list so that
4430 * any new writes get down to disk quickly.
4433 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4434 &BTRFS_I(inode)->runtime_flags);
4437 * 1 for the orphan item we're going to add
4438 * 1 for the orphan item deletion.
4440 trans = btrfs_start_transaction(root, 2);
4442 return PTR_ERR(trans);
4445 * We need to do this in case we fail at _any_ point during the
4446 * actual truncate. Once we do the truncate_setsize we could
4447 * invalidate pages which forces any outstanding ordered io to
4448 * be instantly completed which will give us extents that need
4449 * to be truncated. If we fail to get an orphan inode down we
4450 * could have left over extents that were never meant to live,
4451 * so we need to garuntee from this point on that everything
4452 * will be consistent.
4454 ret = btrfs_orphan_add(trans, inode);
4455 btrfs_end_transaction(trans, root);
4459 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4460 truncate_setsize(inode, newsize);
4462 /* Disable nonlocked read DIO to avoid the end less truncate */
4463 btrfs_inode_block_unlocked_dio(inode);
4464 inode_dio_wait(inode);
4465 btrfs_inode_resume_unlocked_dio(inode);
4467 ret = btrfs_truncate(inode);
4468 if (ret && inode->i_nlink)
4469 btrfs_orphan_del(NULL, inode);
4475 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4477 struct inode *inode = dentry->d_inode;
4478 struct btrfs_root *root = BTRFS_I(inode)->root;
4481 if (btrfs_root_readonly(root))
4484 err = inode_change_ok(inode, attr);
4488 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4489 err = btrfs_setsize(inode, attr);
4494 if (attr->ia_valid) {
4495 setattr_copy(inode, attr);
4496 inode_inc_iversion(inode);
4497 err = btrfs_dirty_inode(inode);
4499 if (!err && attr->ia_valid & ATTR_MODE)
4500 err = btrfs_acl_chmod(inode);
4506 void btrfs_evict_inode(struct inode *inode)
4508 struct btrfs_trans_handle *trans;
4509 struct btrfs_root *root = BTRFS_I(inode)->root;
4510 struct btrfs_block_rsv *rsv, *global_rsv;
4511 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4514 trace_btrfs_inode_evict(inode);
4516 truncate_inode_pages(&inode->i_data, 0);
4517 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
4518 btrfs_is_free_space_inode(inode)))
4521 if (is_bad_inode(inode)) {
4522 btrfs_orphan_del(NULL, inode);
4525 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4526 btrfs_wait_ordered_range(inode, 0, (u64)-1);
4528 if (root->fs_info->log_root_recovering) {
4529 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4530 &BTRFS_I(inode)->runtime_flags));
4534 if (inode->i_nlink > 0) {
4535 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
4539 ret = btrfs_commit_inode_delayed_inode(inode);
4541 btrfs_orphan_del(NULL, inode);
4545 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4547 btrfs_orphan_del(NULL, inode);
4550 rsv->size = min_size;
4552 global_rsv = &root->fs_info->global_block_rsv;
4554 btrfs_i_size_write(inode, 0);
4557 * This is a bit simpler than btrfs_truncate since we've already
4558 * reserved our space for our orphan item in the unlink, so we just
4559 * need to reserve some slack space in case we add bytes and update
4560 * inode item when doing the truncate.
4563 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4564 BTRFS_RESERVE_FLUSH_LIMIT);
4567 * Try and steal from the global reserve since we will
4568 * likely not use this space anyway, we want to try as
4569 * hard as possible to get this to work.
4572 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4575 btrfs_warn(root->fs_info,
4576 "Could not get space for a delete, will truncate on mount %d",
4578 btrfs_orphan_del(NULL, inode);
4579 btrfs_free_block_rsv(root, rsv);
4583 trans = btrfs_join_transaction(root);
4584 if (IS_ERR(trans)) {
4585 btrfs_orphan_del(NULL, inode);
4586 btrfs_free_block_rsv(root, rsv);
4590 trans->block_rsv = rsv;
4592 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4596 trans->block_rsv = &root->fs_info->trans_block_rsv;
4597 btrfs_end_transaction(trans, root);
4599 btrfs_btree_balance_dirty(root);
4602 btrfs_free_block_rsv(root, rsv);
4605 trans->block_rsv = root->orphan_block_rsv;
4606 ret = btrfs_orphan_del(trans, inode);
4610 trans->block_rsv = &root->fs_info->trans_block_rsv;
4611 if (!(root == root->fs_info->tree_root ||
4612 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4613 btrfs_return_ino(root, btrfs_ino(inode));
4615 btrfs_end_transaction(trans, root);
4616 btrfs_btree_balance_dirty(root);
4618 btrfs_remove_delayed_node(inode);
4624 * this returns the key found in the dir entry in the location pointer.
4625 * If no dir entries were found, location->objectid is 0.
4627 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4628 struct btrfs_key *location)
4630 const char *name = dentry->d_name.name;
4631 int namelen = dentry->d_name.len;
4632 struct btrfs_dir_item *di;
4633 struct btrfs_path *path;
4634 struct btrfs_root *root = BTRFS_I(dir)->root;
4637 path = btrfs_alloc_path();
4641 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4646 if (IS_ERR_OR_NULL(di))
4649 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4651 btrfs_free_path(path);
4654 location->objectid = 0;
4659 * when we hit a tree root in a directory, the btrfs part of the inode
4660 * needs to be changed to reflect the root directory of the tree root. This
4661 * is kind of like crossing a mount point.
4663 static int fixup_tree_root_location(struct btrfs_root *root,
4665 struct dentry *dentry,
4666 struct btrfs_key *location,
4667 struct btrfs_root **sub_root)
4669 struct btrfs_path *path;
4670 struct btrfs_root *new_root;
4671 struct btrfs_root_ref *ref;
4672 struct extent_buffer *leaf;
4676 path = btrfs_alloc_path();
4683 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4684 BTRFS_I(dir)->root->root_key.objectid,
4685 location->objectid);
4692 leaf = path->nodes[0];
4693 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4694 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4695 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4698 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4699 (unsigned long)(ref + 1),
4700 dentry->d_name.len);
4704 btrfs_release_path(path);
4706 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4707 if (IS_ERR(new_root)) {
4708 err = PTR_ERR(new_root);
4712 *sub_root = new_root;
4713 location->objectid = btrfs_root_dirid(&new_root->root_item);
4714 location->type = BTRFS_INODE_ITEM_KEY;
4715 location->offset = 0;
4718 btrfs_free_path(path);
4722 static void inode_tree_add(struct inode *inode)
4724 struct btrfs_root *root = BTRFS_I(inode)->root;
4725 struct btrfs_inode *entry;
4727 struct rb_node *parent;
4728 u64 ino = btrfs_ino(inode);
4730 if (inode_unhashed(inode))
4734 spin_lock(&root->inode_lock);
4735 p = &root->inode_tree.rb_node;
4738 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4740 if (ino < btrfs_ino(&entry->vfs_inode))
4741 p = &parent->rb_left;
4742 else if (ino > btrfs_ino(&entry->vfs_inode))
4743 p = &parent->rb_right;
4745 WARN_ON(!(entry->vfs_inode.i_state &
4746 (I_WILL_FREE | I_FREEING)));
4747 rb_erase(parent, &root->inode_tree);
4748 RB_CLEAR_NODE(parent);
4749 spin_unlock(&root->inode_lock);
4753 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4754 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4755 spin_unlock(&root->inode_lock);
4758 static void inode_tree_del(struct inode *inode)
4760 struct btrfs_root *root = BTRFS_I(inode)->root;
4763 spin_lock(&root->inode_lock);
4764 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4765 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4766 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4767 empty = RB_EMPTY_ROOT(&root->inode_tree);
4769 spin_unlock(&root->inode_lock);
4772 * Free space cache has inodes in the tree root, but the tree root has a
4773 * root_refs of 0, so this could end up dropping the tree root as a
4774 * snapshot, so we need the extra !root->fs_info->tree_root check to
4775 * make sure we don't drop it.
4777 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4778 root != root->fs_info->tree_root) {
4779 synchronize_srcu(&root->fs_info->subvol_srcu);
4780 spin_lock(&root->inode_lock);
4781 empty = RB_EMPTY_ROOT(&root->inode_tree);
4782 spin_unlock(&root->inode_lock);
4784 btrfs_add_dead_root(root);
4788 void btrfs_invalidate_inodes(struct btrfs_root *root)
4790 struct rb_node *node;
4791 struct rb_node *prev;
4792 struct btrfs_inode *entry;
4793 struct inode *inode;
4796 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4798 spin_lock(&root->inode_lock);
4800 node = root->inode_tree.rb_node;
4804 entry = rb_entry(node, struct btrfs_inode, rb_node);
4806 if (objectid < btrfs_ino(&entry->vfs_inode))
4807 node = node->rb_left;
4808 else if (objectid > btrfs_ino(&entry->vfs_inode))
4809 node = node->rb_right;
4815 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4816 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4820 prev = rb_next(prev);
4824 entry = rb_entry(node, struct btrfs_inode, rb_node);
4825 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4826 inode = igrab(&entry->vfs_inode);
4828 spin_unlock(&root->inode_lock);
4829 if (atomic_read(&inode->i_count) > 1)
4830 d_prune_aliases(inode);
4832 * btrfs_drop_inode will have it removed from
4833 * the inode cache when its usage count
4838 spin_lock(&root->inode_lock);
4842 if (cond_resched_lock(&root->inode_lock))
4845 node = rb_next(node);
4847 spin_unlock(&root->inode_lock);
4850 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4852 struct btrfs_iget_args *args = p;
4853 inode->i_ino = args->ino;
4854 BTRFS_I(inode)->root = args->root;
4858 static int btrfs_find_actor(struct inode *inode, void *opaque)
4860 struct btrfs_iget_args *args = opaque;
4861 return args->ino == btrfs_ino(inode) &&
4862 args->root == BTRFS_I(inode)->root;
4865 static struct inode *btrfs_iget_locked(struct super_block *s,
4867 struct btrfs_root *root)
4869 struct inode *inode;
4870 struct btrfs_iget_args args;
4871 args.ino = objectid;
4874 inode = iget5_locked(s, objectid, btrfs_find_actor,
4875 btrfs_init_locked_inode,
4880 /* Get an inode object given its location and corresponding root.
4881 * Returns in *is_new if the inode was read from disk
4883 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4884 struct btrfs_root *root, int *new)
4886 struct inode *inode;
4888 inode = btrfs_iget_locked(s, location->objectid, root);
4890 return ERR_PTR(-ENOMEM);
4892 if (inode->i_state & I_NEW) {
4893 BTRFS_I(inode)->root = root;
4894 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4895 btrfs_read_locked_inode(inode);
4896 if (!is_bad_inode(inode)) {
4897 inode_tree_add(inode);
4898 unlock_new_inode(inode);
4902 unlock_new_inode(inode);
4904 inode = ERR_PTR(-ESTALE);
4911 static struct inode *new_simple_dir(struct super_block *s,
4912 struct btrfs_key *key,
4913 struct btrfs_root *root)
4915 struct inode *inode = new_inode(s);
4918 return ERR_PTR(-ENOMEM);
4920 BTRFS_I(inode)->root = root;
4921 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4922 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4924 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4925 inode->i_op = &btrfs_dir_ro_inode_operations;
4926 inode->i_fop = &simple_dir_operations;
4927 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4928 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4933 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4935 struct inode *inode;
4936 struct btrfs_root *root = BTRFS_I(dir)->root;
4937 struct btrfs_root *sub_root = root;
4938 struct btrfs_key location;
4942 if (dentry->d_name.len > BTRFS_NAME_LEN)
4943 return ERR_PTR(-ENAMETOOLONG);
4945 ret = btrfs_inode_by_name(dir, dentry, &location);
4947 return ERR_PTR(ret);
4949 if (location.objectid == 0)
4952 if (location.type == BTRFS_INODE_ITEM_KEY) {
4953 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4957 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4959 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4960 ret = fixup_tree_root_location(root, dir, dentry,
4961 &location, &sub_root);
4964 inode = ERR_PTR(ret);
4966 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4968 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4970 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4972 if (!IS_ERR(inode) && root != sub_root) {
4973 down_read(&root->fs_info->cleanup_work_sem);
4974 if (!(inode->i_sb->s_flags & MS_RDONLY))
4975 ret = btrfs_orphan_cleanup(sub_root);
4976 up_read(&root->fs_info->cleanup_work_sem);
4979 inode = ERR_PTR(ret);
4986 static int btrfs_dentry_delete(const struct dentry *dentry)
4988 struct btrfs_root *root;
4989 struct inode *inode = dentry->d_inode;
4991 if (!inode && !IS_ROOT(dentry))
4992 inode = dentry->d_parent->d_inode;
4995 root = BTRFS_I(inode)->root;
4996 if (btrfs_root_refs(&root->root_item) == 0)
4999 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5005 static void btrfs_dentry_release(struct dentry *dentry)
5007 if (dentry->d_fsdata)
5008 kfree(dentry->d_fsdata);
5011 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5016 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
5020 unsigned char btrfs_filetype_table[] = {
5021 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5024 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5026 struct inode *inode = file_inode(file);
5027 struct btrfs_root *root = BTRFS_I(inode)->root;
5028 struct btrfs_item *item;
5029 struct btrfs_dir_item *di;
5030 struct btrfs_key key;
5031 struct btrfs_key found_key;
5032 struct btrfs_path *path;
5033 struct list_head ins_list;
5034 struct list_head del_list;
5036 struct extent_buffer *leaf;
5038 unsigned char d_type;
5043 int key_type = BTRFS_DIR_INDEX_KEY;
5047 int is_curr = 0; /* ctx->pos points to the current index? */
5049 /* FIXME, use a real flag for deciding about the key type */
5050 if (root->fs_info->tree_root == root)
5051 key_type = BTRFS_DIR_ITEM_KEY;
5053 if (!dir_emit_dots(file, ctx))
5056 path = btrfs_alloc_path();
5062 if (key_type == BTRFS_DIR_INDEX_KEY) {
5063 INIT_LIST_HEAD(&ins_list);
5064 INIT_LIST_HEAD(&del_list);
5065 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5068 btrfs_set_key_type(&key, key_type);
5069 key.offset = ctx->pos;
5070 key.objectid = btrfs_ino(inode);
5072 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5077 leaf = path->nodes[0];
5078 slot = path->slots[0];
5079 if (slot >= btrfs_header_nritems(leaf)) {
5080 ret = btrfs_next_leaf(root, path);
5088 item = btrfs_item_nr(leaf, slot);
5089 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5091 if (found_key.objectid != key.objectid)
5093 if (btrfs_key_type(&found_key) != key_type)
5095 if (found_key.offset < ctx->pos)
5097 if (key_type == BTRFS_DIR_INDEX_KEY &&
5098 btrfs_should_delete_dir_index(&del_list,
5102 ctx->pos = found_key.offset;
5105 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5107 di_total = btrfs_item_size(leaf, item);
5109 while (di_cur < di_total) {
5110 struct btrfs_key location;
5112 if (verify_dir_item(root, leaf, di))
5115 name_len = btrfs_dir_name_len(leaf, di);
5116 if (name_len <= sizeof(tmp_name)) {
5117 name_ptr = tmp_name;
5119 name_ptr = kmalloc(name_len, GFP_NOFS);
5125 read_extent_buffer(leaf, name_ptr,
5126 (unsigned long)(di + 1), name_len);
5128 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5129 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5132 /* is this a reference to our own snapshot? If so
5135 * In contrast to old kernels, we insert the snapshot's
5136 * dir item and dir index after it has been created, so
5137 * we won't find a reference to our own snapshot. We
5138 * still keep the following code for backward
5141 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5142 location.objectid == root->root_key.objectid) {
5146 over = !dir_emit(ctx, name_ptr, name_len,
5147 location.objectid, d_type);
5150 if (name_ptr != tmp_name)
5155 di_len = btrfs_dir_name_len(leaf, di) +
5156 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5158 di = (struct btrfs_dir_item *)((char *)di + di_len);
5164 if (key_type == BTRFS_DIR_INDEX_KEY) {
5167 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5172 /* Reached end of directory/root. Bump pos past the last item. */
5176 * Stop new entries from being returned after we return the last
5179 * New directory entries are assigned a strictly increasing
5180 * offset. This means that new entries created during readdir
5181 * are *guaranteed* to be seen in the future by that readdir.
5182 * This has broken buggy programs which operate on names as
5183 * they're returned by readdir. Until we re-use freed offsets
5184 * we have this hack to stop new entries from being returned
5185 * under the assumption that they'll never reach this huge
5188 * This is being careful not to overflow 32bit loff_t unless the
5189 * last entry requires it because doing so has broken 32bit apps
5192 if (key_type == BTRFS_DIR_INDEX_KEY) {
5193 if (ctx->pos >= INT_MAX)
5194 ctx->pos = LLONG_MAX;
5201 if (key_type == BTRFS_DIR_INDEX_KEY)
5202 btrfs_put_delayed_items(&ins_list, &del_list);
5203 btrfs_free_path(path);
5207 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5209 struct btrfs_root *root = BTRFS_I(inode)->root;
5210 struct btrfs_trans_handle *trans;
5212 bool nolock = false;
5214 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5217 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5220 if (wbc->sync_mode == WB_SYNC_ALL) {
5222 trans = btrfs_join_transaction_nolock(root);
5224 trans = btrfs_join_transaction(root);
5226 return PTR_ERR(trans);
5227 ret = btrfs_commit_transaction(trans, root);
5233 * This is somewhat expensive, updating the tree every time the
5234 * inode changes. But, it is most likely to find the inode in cache.
5235 * FIXME, needs more benchmarking...there are no reasons other than performance
5236 * to keep or drop this code.
5238 static int btrfs_dirty_inode(struct inode *inode)
5240 struct btrfs_root *root = BTRFS_I(inode)->root;
5241 struct btrfs_trans_handle *trans;
5244 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5247 trans = btrfs_join_transaction(root);
5249 return PTR_ERR(trans);
5251 ret = btrfs_update_inode(trans, root, inode);
5252 if (ret && ret == -ENOSPC) {
5253 /* whoops, lets try again with the full transaction */
5254 btrfs_end_transaction(trans, root);
5255 trans = btrfs_start_transaction(root, 1);
5257 return PTR_ERR(trans);
5259 ret = btrfs_update_inode(trans, root, inode);
5261 btrfs_end_transaction(trans, root);
5262 if (BTRFS_I(inode)->delayed_node)
5263 btrfs_balance_delayed_items(root);
5269 * This is a copy of file_update_time. We need this so we can return error on
5270 * ENOSPC for updating the inode in the case of file write and mmap writes.
5272 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5275 struct btrfs_root *root = BTRFS_I(inode)->root;
5277 if (btrfs_root_readonly(root))
5280 if (flags & S_VERSION)
5281 inode_inc_iversion(inode);
5282 if (flags & S_CTIME)
5283 inode->i_ctime = *now;
5284 if (flags & S_MTIME)
5285 inode->i_mtime = *now;
5286 if (flags & S_ATIME)
5287 inode->i_atime = *now;
5288 return btrfs_dirty_inode(inode);
5292 * find the highest existing sequence number in a directory
5293 * and then set the in-memory index_cnt variable to reflect
5294 * free sequence numbers
5296 static int btrfs_set_inode_index_count(struct inode *inode)
5298 struct btrfs_root *root = BTRFS_I(inode)->root;
5299 struct btrfs_key key, found_key;
5300 struct btrfs_path *path;
5301 struct extent_buffer *leaf;
5304 key.objectid = btrfs_ino(inode);
5305 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5306 key.offset = (u64)-1;
5308 path = btrfs_alloc_path();
5312 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5315 /* FIXME: we should be able to handle this */
5321 * MAGIC NUMBER EXPLANATION:
5322 * since we search a directory based on f_pos we have to start at 2
5323 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5324 * else has to start at 2
5326 if (path->slots[0] == 0) {
5327 BTRFS_I(inode)->index_cnt = 2;
5333 leaf = path->nodes[0];
5334 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5336 if (found_key.objectid != btrfs_ino(inode) ||
5337 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5338 BTRFS_I(inode)->index_cnt = 2;
5342 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5344 btrfs_free_path(path);
5349 * helper to find a free sequence number in a given directory. This current
5350 * code is very simple, later versions will do smarter things in the btree
5352 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5356 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5357 ret = btrfs_inode_delayed_dir_index_count(dir);
5359 ret = btrfs_set_inode_index_count(dir);
5365 *index = BTRFS_I(dir)->index_cnt;
5366 BTRFS_I(dir)->index_cnt++;
5371 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5372 struct btrfs_root *root,
5374 const char *name, int name_len,
5375 u64 ref_objectid, u64 objectid,
5376 umode_t mode, u64 *index)
5378 struct inode *inode;
5379 struct btrfs_inode_item *inode_item;
5380 struct btrfs_key *location;
5381 struct btrfs_path *path;
5382 struct btrfs_inode_ref *ref;
5383 struct btrfs_key key[2];
5389 path = btrfs_alloc_path();
5391 return ERR_PTR(-ENOMEM);
5393 inode = new_inode(root->fs_info->sb);
5395 btrfs_free_path(path);
5396 return ERR_PTR(-ENOMEM);
5400 * we have to initialize this early, so we can reclaim the inode
5401 * number if we fail afterwards in this function.
5403 inode->i_ino = objectid;
5406 trace_btrfs_inode_request(dir);
5408 ret = btrfs_set_inode_index(dir, index);
5410 btrfs_free_path(path);
5412 return ERR_PTR(ret);
5416 * index_cnt is ignored for everything but a dir,
5417 * btrfs_get_inode_index_count has an explanation for the magic
5420 BTRFS_I(inode)->index_cnt = 2;
5421 BTRFS_I(inode)->root = root;
5422 BTRFS_I(inode)->generation = trans->transid;
5423 inode->i_generation = BTRFS_I(inode)->generation;
5426 * We could have gotten an inode number from somebody who was fsynced
5427 * and then removed in this same transaction, so let's just set full
5428 * sync since it will be a full sync anyway and this will blow away the
5429 * old info in the log.
5431 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5438 key[0].objectid = objectid;
5439 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5443 * Start new inodes with an inode_ref. This is slightly more
5444 * efficient for small numbers of hard links since they will
5445 * be packed into one item. Extended refs will kick in if we
5446 * add more hard links than can fit in the ref item.
5448 key[1].objectid = objectid;
5449 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5450 key[1].offset = ref_objectid;
5452 sizes[0] = sizeof(struct btrfs_inode_item);
5453 sizes[1] = name_len + sizeof(*ref);
5455 path->leave_spinning = 1;
5456 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5460 inode_init_owner(inode, dir, mode);
5461 inode_set_bytes(inode, 0);
5462 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5463 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5464 struct btrfs_inode_item);
5465 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5466 sizeof(*inode_item));
5467 fill_inode_item(trans, path->nodes[0], inode_item, inode);
5469 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5470 struct btrfs_inode_ref);
5471 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5472 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5473 ptr = (unsigned long)(ref + 1);
5474 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5476 btrfs_mark_buffer_dirty(path->nodes[0]);
5477 btrfs_free_path(path);
5479 location = &BTRFS_I(inode)->location;
5480 location->objectid = objectid;
5481 location->offset = 0;
5482 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5484 btrfs_inherit_iflags(inode, dir);
5486 if (S_ISREG(mode)) {
5487 if (btrfs_test_opt(root, NODATASUM))
5488 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5489 if (btrfs_test_opt(root, NODATACOW))
5490 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5491 BTRFS_INODE_NODATASUM;
5494 insert_inode_hash(inode);
5495 inode_tree_add(inode);
5497 trace_btrfs_inode_new(inode);
5498 btrfs_set_inode_last_trans(trans, inode);
5500 btrfs_update_root_times(trans, root);
5505 BTRFS_I(dir)->index_cnt--;
5506 btrfs_free_path(path);
5508 return ERR_PTR(ret);
5511 static inline u8 btrfs_inode_type(struct inode *inode)
5513 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5517 * utility function to add 'inode' into 'parent_inode' with
5518 * a give name and a given sequence number.
5519 * if 'add_backref' is true, also insert a backref from the
5520 * inode to the parent directory.
5522 int btrfs_add_link(struct btrfs_trans_handle *trans,
5523 struct inode *parent_inode, struct inode *inode,
5524 const char *name, int name_len, int add_backref, u64 index)
5527 struct btrfs_key key;
5528 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5529 u64 ino = btrfs_ino(inode);
5530 u64 parent_ino = btrfs_ino(parent_inode);
5532 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5533 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5536 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5540 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5541 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5542 key.objectid, root->root_key.objectid,
5543 parent_ino, index, name, name_len);
5544 } else if (add_backref) {
5545 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5549 /* Nothing to clean up yet */
5553 ret = btrfs_insert_dir_item(trans, root, name, name_len,
5555 btrfs_inode_type(inode), index);
5556 if (ret == -EEXIST || ret == -EOVERFLOW)
5559 btrfs_abort_transaction(trans, root, ret);
5563 btrfs_i_size_write(parent_inode, parent_inode->i_size +
5565 inode_inc_iversion(parent_inode);
5566 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5567 ret = btrfs_update_inode(trans, root, parent_inode);
5569 btrfs_abort_transaction(trans, root, ret);
5573 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5576 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5577 key.objectid, root->root_key.objectid,
5578 parent_ino, &local_index, name, name_len);
5580 } else if (add_backref) {
5584 err = btrfs_del_inode_ref(trans, root, name, name_len,
5585 ino, parent_ino, &local_index);
5590 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5591 struct inode *dir, struct dentry *dentry,
5592 struct inode *inode, int backref, u64 index)
5594 int err = btrfs_add_link(trans, dir, inode,
5595 dentry->d_name.name, dentry->d_name.len,
5602 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5603 umode_t mode, dev_t rdev)
5605 struct btrfs_trans_handle *trans;
5606 struct btrfs_root *root = BTRFS_I(dir)->root;
5607 struct inode *inode = NULL;
5613 if (!new_valid_dev(rdev))
5617 * 2 for inode item and ref
5619 * 1 for xattr if selinux is on
5621 trans = btrfs_start_transaction(root, 5);
5623 return PTR_ERR(trans);
5625 err = btrfs_find_free_ino(root, &objectid);
5629 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5630 dentry->d_name.len, btrfs_ino(dir), objectid,
5632 if (IS_ERR(inode)) {
5633 err = PTR_ERR(inode);
5637 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5644 * If the active LSM wants to access the inode during
5645 * d_instantiate it needs these. Smack checks to see
5646 * if the filesystem supports xattrs by looking at the
5650 inode->i_op = &btrfs_special_inode_operations;
5651 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5655 init_special_inode(inode, inode->i_mode, rdev);
5656 btrfs_update_inode(trans, root, inode);
5657 d_instantiate(dentry, inode);
5660 btrfs_end_transaction(trans, root);
5661 btrfs_btree_balance_dirty(root);
5663 inode_dec_link_count(inode);
5669 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5670 umode_t mode, bool excl)
5672 struct btrfs_trans_handle *trans;
5673 struct btrfs_root *root = BTRFS_I(dir)->root;
5674 struct inode *inode = NULL;
5675 int drop_inode_on_err = 0;
5681 * 2 for inode item and ref
5683 * 1 for xattr if selinux is on
5685 trans = btrfs_start_transaction(root, 5);
5687 return PTR_ERR(trans);
5689 err = btrfs_find_free_ino(root, &objectid);
5693 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5694 dentry->d_name.len, btrfs_ino(dir), objectid,
5696 if (IS_ERR(inode)) {
5697 err = PTR_ERR(inode);
5700 drop_inode_on_err = 1;
5702 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5706 err = btrfs_update_inode(trans, root, inode);
5711 * If the active LSM wants to access the inode during
5712 * d_instantiate it needs these. Smack checks to see
5713 * if the filesystem supports xattrs by looking at the
5716 inode->i_fop = &btrfs_file_operations;
5717 inode->i_op = &btrfs_file_inode_operations;
5719 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5723 inode->i_mapping->a_ops = &btrfs_aops;
5724 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5725 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5726 d_instantiate(dentry, inode);
5729 btrfs_end_transaction(trans, root);
5730 if (err && drop_inode_on_err) {
5731 inode_dec_link_count(inode);
5734 btrfs_btree_balance_dirty(root);
5738 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5739 struct dentry *dentry)
5741 struct btrfs_trans_handle *trans;
5742 struct btrfs_root *root = BTRFS_I(dir)->root;
5743 struct inode *inode = old_dentry->d_inode;
5748 /* do not allow sys_link's with other subvols of the same device */
5749 if (root->objectid != BTRFS_I(inode)->root->objectid)
5752 if (inode->i_nlink >= BTRFS_LINK_MAX)
5755 err = btrfs_set_inode_index(dir, &index);
5760 * 2 items for inode and inode ref
5761 * 2 items for dir items
5762 * 1 item for parent inode
5764 trans = btrfs_start_transaction(root, 5);
5765 if (IS_ERR(trans)) {
5766 err = PTR_ERR(trans);
5770 btrfs_inc_nlink(inode);
5771 inode_inc_iversion(inode);
5772 inode->i_ctime = CURRENT_TIME;
5774 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5776 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5781 struct dentry *parent = dentry->d_parent;
5782 err = btrfs_update_inode(trans, root, inode);
5785 d_instantiate(dentry, inode);
5786 btrfs_log_new_name(trans, inode, NULL, parent);
5789 btrfs_end_transaction(trans, root);
5792 inode_dec_link_count(inode);
5795 btrfs_btree_balance_dirty(root);
5799 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5801 struct inode *inode = NULL;
5802 struct btrfs_trans_handle *trans;
5803 struct btrfs_root *root = BTRFS_I(dir)->root;
5805 int drop_on_err = 0;
5810 * 2 items for inode and ref
5811 * 2 items for dir items
5812 * 1 for xattr if selinux is on
5814 trans = btrfs_start_transaction(root, 5);
5816 return PTR_ERR(trans);
5818 err = btrfs_find_free_ino(root, &objectid);
5822 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5823 dentry->d_name.len, btrfs_ino(dir), objectid,
5824 S_IFDIR | mode, &index);
5825 if (IS_ERR(inode)) {
5826 err = PTR_ERR(inode);
5832 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5836 inode->i_op = &btrfs_dir_inode_operations;
5837 inode->i_fop = &btrfs_dir_file_operations;
5839 btrfs_i_size_write(inode, 0);
5840 err = btrfs_update_inode(trans, root, inode);
5844 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5845 dentry->d_name.len, 0, index);
5849 d_instantiate(dentry, inode);
5853 btrfs_end_transaction(trans, root);
5856 btrfs_btree_balance_dirty(root);
5860 /* helper for btfs_get_extent. Given an existing extent in the tree,
5861 * and an extent that you want to insert, deal with overlap and insert
5862 * the new extent into the tree.
5864 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5865 struct extent_map *existing,
5866 struct extent_map *em,
5867 u64 map_start, u64 map_len)
5871 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5872 start_diff = map_start - em->start;
5873 em->start = map_start;
5875 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5876 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5877 em->block_start += start_diff;
5878 em->block_len -= start_diff;
5880 return add_extent_mapping(em_tree, em, 0);
5883 static noinline int uncompress_inline(struct btrfs_path *path,
5884 struct inode *inode, struct page *page,
5885 size_t pg_offset, u64 extent_offset,
5886 struct btrfs_file_extent_item *item)
5889 struct extent_buffer *leaf = path->nodes[0];
5892 unsigned long inline_size;
5896 WARN_ON(pg_offset != 0);
5897 compress_type = btrfs_file_extent_compression(leaf, item);
5898 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5899 inline_size = btrfs_file_extent_inline_item_len(leaf,
5900 btrfs_item_nr(leaf, path->slots[0]));
5901 tmp = kmalloc(inline_size, GFP_NOFS);
5904 ptr = btrfs_file_extent_inline_start(item);
5906 read_extent_buffer(leaf, tmp, ptr, inline_size);
5908 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5909 ret = btrfs_decompress(compress_type, tmp, page,
5910 extent_offset, inline_size, max_size);
5912 char *kaddr = kmap_atomic(page);
5913 unsigned long copy_size = min_t(u64,
5914 PAGE_CACHE_SIZE - pg_offset,
5915 max_size - extent_offset);
5916 memset(kaddr + pg_offset, 0, copy_size);
5917 kunmap_atomic(kaddr);
5924 * a bit scary, this does extent mapping from logical file offset to the disk.
5925 * the ugly parts come from merging extents from the disk with the in-ram
5926 * representation. This gets more complex because of the data=ordered code,
5927 * where the in-ram extents might be locked pending data=ordered completion.
5929 * This also copies inline extents directly into the page.
5932 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5933 size_t pg_offset, u64 start, u64 len,
5939 u64 extent_start = 0;
5941 u64 objectid = btrfs_ino(inode);
5943 struct btrfs_path *path = NULL;
5944 struct btrfs_root *root = BTRFS_I(inode)->root;
5945 struct btrfs_file_extent_item *item;
5946 struct extent_buffer *leaf;
5947 struct btrfs_key found_key;
5948 struct extent_map *em = NULL;
5949 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5950 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5951 struct btrfs_trans_handle *trans = NULL;
5955 read_lock(&em_tree->lock);
5956 em = lookup_extent_mapping(em_tree, start, len);
5958 em->bdev = root->fs_info->fs_devices->latest_bdev;
5959 read_unlock(&em_tree->lock);
5962 if (em->start > start || em->start + em->len <= start)
5963 free_extent_map(em);
5964 else if (em->block_start == EXTENT_MAP_INLINE && page)
5965 free_extent_map(em);
5969 em = alloc_extent_map();
5974 em->bdev = root->fs_info->fs_devices->latest_bdev;
5975 em->start = EXTENT_MAP_HOLE;
5976 em->orig_start = EXTENT_MAP_HOLE;
5978 em->block_len = (u64)-1;
5981 path = btrfs_alloc_path();
5987 * Chances are we'll be called again, so go ahead and do
5993 ret = btrfs_lookup_file_extent(trans, root, path,
5994 objectid, start, trans != NULL);
6001 if (path->slots[0] == 0)
6006 leaf = path->nodes[0];
6007 item = btrfs_item_ptr(leaf, path->slots[0],
6008 struct btrfs_file_extent_item);
6009 /* are we inside the extent that was found? */
6010 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6011 found_type = btrfs_key_type(&found_key);
6012 if (found_key.objectid != objectid ||
6013 found_type != BTRFS_EXTENT_DATA_KEY) {
6017 found_type = btrfs_file_extent_type(leaf, item);
6018 extent_start = found_key.offset;
6019 compress_type = btrfs_file_extent_compression(leaf, item);
6020 if (found_type == BTRFS_FILE_EXTENT_REG ||
6021 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6022 extent_end = extent_start +
6023 btrfs_file_extent_num_bytes(leaf, item);
6024 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6026 size = btrfs_file_extent_inline_len(leaf, item);
6027 extent_end = ALIGN(extent_start + size, root->sectorsize);
6030 if (start >= extent_end) {
6032 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6033 ret = btrfs_next_leaf(root, path);
6040 leaf = path->nodes[0];
6042 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6043 if (found_key.objectid != objectid ||
6044 found_key.type != BTRFS_EXTENT_DATA_KEY)
6046 if (start + len <= found_key.offset)
6049 em->orig_start = start;
6050 em->len = found_key.offset - start;
6054 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
6055 if (found_type == BTRFS_FILE_EXTENT_REG ||
6056 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6057 em->start = extent_start;
6058 em->len = extent_end - extent_start;
6059 em->orig_start = extent_start -
6060 btrfs_file_extent_offset(leaf, item);
6061 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6063 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6065 em->block_start = EXTENT_MAP_HOLE;
6068 if (compress_type != BTRFS_COMPRESS_NONE) {
6069 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6070 em->compress_type = compress_type;
6071 em->block_start = bytenr;
6072 em->block_len = em->orig_block_len;
6074 bytenr += btrfs_file_extent_offset(leaf, item);
6075 em->block_start = bytenr;
6076 em->block_len = em->len;
6077 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6078 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6081 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6085 size_t extent_offset;
6088 em->block_start = EXTENT_MAP_INLINE;
6089 if (!page || create) {
6090 em->start = extent_start;
6091 em->len = extent_end - extent_start;
6095 size = btrfs_file_extent_inline_len(leaf, item);
6096 extent_offset = page_offset(page) + pg_offset - extent_start;
6097 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6098 size - extent_offset);
6099 em->start = extent_start + extent_offset;
6100 em->len = ALIGN(copy_size, root->sectorsize);
6101 em->orig_block_len = em->len;
6102 em->orig_start = em->start;
6103 if (compress_type) {
6104 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6105 em->compress_type = compress_type;
6107 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6108 if (create == 0 && !PageUptodate(page)) {
6109 if (btrfs_file_extent_compression(leaf, item) !=
6110 BTRFS_COMPRESS_NONE) {
6111 ret = uncompress_inline(path, inode, page,
6113 extent_offset, item);
6114 BUG_ON(ret); /* -ENOMEM */
6117 read_extent_buffer(leaf, map + pg_offset, ptr,
6119 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6120 memset(map + pg_offset + copy_size, 0,
6121 PAGE_CACHE_SIZE - pg_offset -
6126 flush_dcache_page(page);
6127 } else if (create && PageUptodate(page)) {
6131 free_extent_map(em);
6134 btrfs_release_path(path);
6135 trans = btrfs_join_transaction(root);
6138 return ERR_CAST(trans);
6142 write_extent_buffer(leaf, map + pg_offset, ptr,
6145 btrfs_mark_buffer_dirty(leaf);
6147 set_extent_uptodate(io_tree, em->start,
6148 extent_map_end(em) - 1, NULL, GFP_NOFS);
6151 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6155 em->orig_start = start;
6158 em->block_start = EXTENT_MAP_HOLE;
6159 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6161 btrfs_release_path(path);
6162 if (em->start > start || extent_map_end(em) <= start) {
6163 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6164 (unsigned long long)em->start,
6165 (unsigned long long)em->len,
6166 (unsigned long long)start,
6167 (unsigned long long)len);
6173 write_lock(&em_tree->lock);
6174 ret = add_extent_mapping(em_tree, em, 0);
6175 /* it is possible that someone inserted the extent into the tree
6176 * while we had the lock dropped. It is also possible that
6177 * an overlapping map exists in the tree
6179 if (ret == -EEXIST) {
6180 struct extent_map *existing;
6184 existing = lookup_extent_mapping(em_tree, start, len);
6185 if (existing && (existing->start > start ||
6186 existing->start + existing->len <= start)) {
6187 free_extent_map(existing);
6191 existing = lookup_extent_mapping(em_tree, em->start,
6194 err = merge_extent_mapping(em_tree, existing,
6197 free_extent_map(existing);
6199 free_extent_map(em);
6204 free_extent_map(em);
6208 free_extent_map(em);
6213 write_unlock(&em_tree->lock);
6217 trace_btrfs_get_extent(root, em);
6220 btrfs_free_path(path);
6222 ret = btrfs_end_transaction(trans, root);
6227 free_extent_map(em);
6228 return ERR_PTR(err);
6230 BUG_ON(!em); /* Error is always set */
6234 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6235 size_t pg_offset, u64 start, u64 len,
6238 struct extent_map *em;
6239 struct extent_map *hole_em = NULL;
6240 u64 range_start = start;
6246 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6253 * - a pre-alloc extent,
6254 * there might actually be delalloc bytes behind it.
6256 if (em->block_start != EXTENT_MAP_HOLE &&
6257 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6263 /* check to see if we've wrapped (len == -1 or similar) */
6272 /* ok, we didn't find anything, lets look for delalloc */
6273 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6274 end, len, EXTENT_DELALLOC, 1);
6275 found_end = range_start + found;
6276 if (found_end < range_start)
6277 found_end = (u64)-1;
6280 * we didn't find anything useful, return
6281 * the original results from get_extent()
6283 if (range_start > end || found_end <= start) {
6289 /* adjust the range_start to make sure it doesn't
6290 * go backwards from the start they passed in
6292 range_start = max(start,range_start);
6293 found = found_end - range_start;
6296 u64 hole_start = start;
6299 em = alloc_extent_map();
6305 * when btrfs_get_extent can't find anything it
6306 * returns one huge hole
6308 * make sure what it found really fits our range, and
6309 * adjust to make sure it is based on the start from
6313 u64 calc_end = extent_map_end(hole_em);
6315 if (calc_end <= start || (hole_em->start > end)) {
6316 free_extent_map(hole_em);
6319 hole_start = max(hole_em->start, start);
6320 hole_len = calc_end - hole_start;
6324 if (hole_em && range_start > hole_start) {
6325 /* our hole starts before our delalloc, so we
6326 * have to return just the parts of the hole
6327 * that go until the delalloc starts
6329 em->len = min(hole_len,
6330 range_start - hole_start);
6331 em->start = hole_start;
6332 em->orig_start = hole_start;
6334 * don't adjust block start at all,
6335 * it is fixed at EXTENT_MAP_HOLE
6337 em->block_start = hole_em->block_start;
6338 em->block_len = hole_len;
6339 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6340 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6342 em->start = range_start;
6344 em->orig_start = range_start;
6345 em->block_start = EXTENT_MAP_DELALLOC;
6346 em->block_len = found;
6348 } else if (hole_em) {
6353 free_extent_map(hole_em);
6355 free_extent_map(em);
6356 return ERR_PTR(err);
6361 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6364 struct btrfs_root *root = BTRFS_I(inode)->root;
6365 struct btrfs_trans_handle *trans;
6366 struct extent_map *em;
6367 struct btrfs_key ins;
6371 trans = btrfs_join_transaction(root);
6373 return ERR_CAST(trans);
6375 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
6377 alloc_hint = get_extent_allocation_hint(inode, start, len);
6378 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
6379 alloc_hint, &ins, 1);
6385 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6386 ins.offset, ins.offset, ins.offset, 0);
6390 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6391 ins.offset, ins.offset, 0);
6393 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6397 btrfs_end_transaction(trans, root);
6402 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6403 * block must be cow'd
6405 noinline int can_nocow_extent(struct btrfs_trans_handle *trans,
6406 struct inode *inode, u64 offset, u64 *len,
6407 u64 *orig_start, u64 *orig_block_len,
6410 struct btrfs_path *path;
6412 struct extent_buffer *leaf;
6413 struct btrfs_root *root = BTRFS_I(inode)->root;
6414 struct btrfs_file_extent_item *fi;
6415 struct btrfs_key key;
6422 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6423 path = btrfs_alloc_path();
6427 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
6432 slot = path->slots[0];
6435 /* can't find the item, must cow */
6442 leaf = path->nodes[0];
6443 btrfs_item_key_to_cpu(leaf, &key, slot);
6444 if (key.objectid != btrfs_ino(inode) ||
6445 key.type != BTRFS_EXTENT_DATA_KEY) {
6446 /* not our file or wrong item type, must cow */
6450 if (key.offset > offset) {
6451 /* Wrong offset, must cow */
6455 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6456 found_type = btrfs_file_extent_type(leaf, fi);
6457 if (found_type != BTRFS_FILE_EXTENT_REG &&
6458 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6459 /* not a regular extent, must cow */
6463 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6466 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6467 if (disk_bytenr == 0)
6470 if (btrfs_file_extent_compression(leaf, fi) ||
6471 btrfs_file_extent_encryption(leaf, fi) ||
6472 btrfs_file_extent_other_encoding(leaf, fi))
6475 backref_offset = btrfs_file_extent_offset(leaf, fi);
6478 *orig_start = key.offset - backref_offset;
6479 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6480 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6483 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6485 if (btrfs_extent_readonly(root, disk_bytenr))
6489 * look for other files referencing this extent, if we
6490 * find any we must cow
6492 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6493 key.offset - backref_offset, disk_bytenr))
6497 * adjust disk_bytenr and num_bytes to cover just the bytes
6498 * in this extent we are about to write. If there
6499 * are any csums in that range we have to cow in order
6500 * to keep the csums correct
6502 disk_bytenr += backref_offset;
6503 disk_bytenr += offset - key.offset;
6504 num_bytes = min(offset + *len, extent_end) - offset;
6505 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6508 * all of the above have passed, it is safe to overwrite this extent
6514 btrfs_free_path(path);
6518 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6519 struct extent_state **cached_state, int writing)
6521 struct btrfs_ordered_extent *ordered;
6525 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6528 * We're concerned with the entire range that we're going to be
6529 * doing DIO to, so we need to make sure theres no ordered
6530 * extents in this range.
6532 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6533 lockend - lockstart + 1);
6536 * We need to make sure there are no buffered pages in this
6537 * range either, we could have raced between the invalidate in
6538 * generic_file_direct_write and locking the extent. The
6539 * invalidate needs to happen so that reads after a write do not
6542 if (!ordered && (!writing ||
6543 !test_range_bit(&BTRFS_I(inode)->io_tree,
6544 lockstart, lockend, EXTENT_UPTODATE, 0,
6548 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6549 cached_state, GFP_NOFS);
6552 btrfs_start_ordered_extent(inode, ordered, 1);
6553 btrfs_put_ordered_extent(ordered);
6555 /* Screw you mmap */
6556 ret = filemap_write_and_wait_range(inode->i_mapping,
6563 * If we found a page that couldn't be invalidated just
6564 * fall back to buffered.
6566 ret = invalidate_inode_pages2_range(inode->i_mapping,
6567 lockstart >> PAGE_CACHE_SHIFT,
6568 lockend >> PAGE_CACHE_SHIFT);
6579 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6580 u64 len, u64 orig_start,
6581 u64 block_start, u64 block_len,
6582 u64 orig_block_len, u64 ram_bytes,
6585 struct extent_map_tree *em_tree;
6586 struct extent_map *em;
6587 struct btrfs_root *root = BTRFS_I(inode)->root;
6590 em_tree = &BTRFS_I(inode)->extent_tree;
6591 em = alloc_extent_map();
6593 return ERR_PTR(-ENOMEM);
6596 em->orig_start = orig_start;
6597 em->mod_start = start;
6600 em->block_len = block_len;
6601 em->block_start = block_start;
6602 em->bdev = root->fs_info->fs_devices->latest_bdev;
6603 em->orig_block_len = orig_block_len;
6604 em->ram_bytes = ram_bytes;
6605 em->generation = -1;
6606 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6607 if (type == BTRFS_ORDERED_PREALLOC)
6608 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6611 btrfs_drop_extent_cache(inode, em->start,
6612 em->start + em->len - 1, 0);
6613 write_lock(&em_tree->lock);
6614 ret = add_extent_mapping(em_tree, em, 1);
6615 write_unlock(&em_tree->lock);
6616 } while (ret == -EEXIST);
6619 free_extent_map(em);
6620 return ERR_PTR(ret);
6627 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6628 struct buffer_head *bh_result, int create)
6630 struct extent_map *em;
6631 struct btrfs_root *root = BTRFS_I(inode)->root;
6632 struct extent_state *cached_state = NULL;
6633 u64 start = iblock << inode->i_blkbits;
6634 u64 lockstart, lockend;
6635 u64 len = bh_result->b_size;
6636 struct btrfs_trans_handle *trans;
6637 int unlock_bits = EXTENT_LOCKED;
6641 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6643 len = min_t(u64, len, root->sectorsize);
6646 lockend = start + len - 1;
6649 * If this errors out it's because we couldn't invalidate pagecache for
6650 * this range and we need to fallback to buffered.
6652 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6655 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6662 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6663 * io. INLINE is special, and we could probably kludge it in here, but
6664 * it's still buffered so for safety lets just fall back to the generic
6667 * For COMPRESSED we _have_ to read the entire extent in so we can
6668 * decompress it, so there will be buffering required no matter what we
6669 * do, so go ahead and fallback to buffered.
6671 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6672 * to buffered IO. Don't blame me, this is the price we pay for using
6675 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6676 em->block_start == EXTENT_MAP_INLINE) {
6677 free_extent_map(em);
6682 /* Just a good old fashioned hole, return */
6683 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6684 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6685 free_extent_map(em);
6690 * We don't allocate a new extent in the following cases
6692 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6694 * 2) The extent is marked as PREALLOC. We're good to go here and can
6695 * just use the extent.
6699 len = min(len, em->len - (start - em->start));
6700 lockstart = start + len;
6704 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6705 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6706 em->block_start != EXTENT_MAP_HOLE)) {
6709 u64 block_start, orig_start, orig_block_len, ram_bytes;
6711 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6712 type = BTRFS_ORDERED_PREALLOC;
6714 type = BTRFS_ORDERED_NOCOW;
6715 len = min(len, em->len - (start - em->start));
6716 block_start = em->block_start + (start - em->start);
6719 * we're not going to log anything, but we do need
6720 * to make sure the current transaction stays open
6721 * while we look for nocow cross refs
6723 trans = btrfs_join_transaction(root);
6727 if (can_nocow_extent(trans, inode, start, &len, &orig_start,
6728 &orig_block_len, &ram_bytes) == 1) {
6729 if (type == BTRFS_ORDERED_PREALLOC) {
6730 free_extent_map(em);
6731 em = create_pinned_em(inode, start, len,
6737 btrfs_end_transaction(trans, root);
6742 ret = btrfs_add_ordered_extent_dio(inode, start,
6743 block_start, len, len, type);
6744 btrfs_end_transaction(trans, root);
6746 free_extent_map(em);
6751 btrfs_end_transaction(trans, root);
6755 * this will cow the extent, reset the len in case we changed
6758 len = bh_result->b_size;
6759 free_extent_map(em);
6760 em = btrfs_new_extent_direct(inode, start, len);
6765 len = min(len, em->len - (start - em->start));
6767 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6769 bh_result->b_size = len;
6770 bh_result->b_bdev = em->bdev;
6771 set_buffer_mapped(bh_result);
6773 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6774 set_buffer_new(bh_result);
6777 * Need to update the i_size under the extent lock so buffered
6778 * readers will get the updated i_size when we unlock.
6780 if (start + len > i_size_read(inode))
6781 i_size_write(inode, start + len);
6783 spin_lock(&BTRFS_I(inode)->lock);
6784 BTRFS_I(inode)->outstanding_extents++;
6785 spin_unlock(&BTRFS_I(inode)->lock);
6787 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6788 lockstart + len - 1, EXTENT_DELALLOC, NULL,
6789 &cached_state, GFP_NOFS);
6794 * In the case of write we need to clear and unlock the entire range,
6795 * in the case of read we need to unlock only the end area that we
6796 * aren't using if there is any left over space.
6798 if (lockstart < lockend) {
6799 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6800 lockend, unlock_bits, 1, 0,
6801 &cached_state, GFP_NOFS);
6803 free_extent_state(cached_state);
6806 free_extent_map(em);
6811 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6812 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6816 struct btrfs_dio_private {
6817 struct inode *inode;
6823 /* number of bios pending for this dio */
6824 atomic_t pending_bios;
6829 /* orig_bio is our btrfs_io_bio */
6830 struct bio *orig_bio;
6832 /* dio_bio came from fs/direct-io.c */
6833 struct bio *dio_bio;
6836 static void btrfs_endio_direct_read(struct bio *bio, int err)
6838 struct btrfs_dio_private *dip = bio->bi_private;
6839 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6840 struct bio_vec *bvec = bio->bi_io_vec;
6841 struct inode *inode = dip->inode;
6842 struct btrfs_root *root = BTRFS_I(inode)->root;
6843 struct bio *dio_bio;
6846 start = dip->logical_offset;
6848 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6849 struct page *page = bvec->bv_page;
6852 u64 private = ~(u32)0;
6853 unsigned long flags;
6855 if (get_state_private(&BTRFS_I(inode)->io_tree,
6858 local_irq_save(flags);
6859 kaddr = kmap_atomic(page);
6860 csum = btrfs_csum_data(kaddr + bvec->bv_offset,
6861 csum, bvec->bv_len);
6862 btrfs_csum_final(csum, (char *)&csum);
6863 kunmap_atomic(kaddr);
6864 local_irq_restore(flags);
6866 flush_dcache_page(bvec->bv_page);
6867 if (csum != private) {
6869 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u private %u",
6870 (unsigned long long)btrfs_ino(inode),
6871 (unsigned long long)start,
6872 csum, (unsigned)private);
6877 start += bvec->bv_len;
6879 } while (bvec <= bvec_end);
6881 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6882 dip->logical_offset + dip->bytes - 1);
6883 dio_bio = dip->dio_bio;
6887 /* If we had a csum failure make sure to clear the uptodate flag */
6889 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6890 dio_end_io(dio_bio, err);
6894 static void btrfs_endio_direct_write(struct bio *bio, int err)
6896 struct btrfs_dio_private *dip = bio->bi_private;
6897 struct inode *inode = dip->inode;
6898 struct btrfs_root *root = BTRFS_I(inode)->root;
6899 struct btrfs_ordered_extent *ordered = NULL;
6900 u64 ordered_offset = dip->logical_offset;
6901 u64 ordered_bytes = dip->bytes;
6902 struct bio *dio_bio;
6908 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6910 ordered_bytes, !err);
6914 ordered->work.func = finish_ordered_fn;
6915 ordered->work.flags = 0;
6916 btrfs_queue_worker(&root->fs_info->endio_write_workers,
6920 * our bio might span multiple ordered extents. If we haven't
6921 * completed the accounting for the whole dio, go back and try again
6923 if (ordered_offset < dip->logical_offset + dip->bytes) {
6924 ordered_bytes = dip->logical_offset + dip->bytes -
6930 dio_bio = dip->dio_bio;
6934 /* If we had an error make sure to clear the uptodate flag */
6936 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6937 dio_end_io(dio_bio, err);
6941 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6942 struct bio *bio, int mirror_num,
6943 unsigned long bio_flags, u64 offset)
6946 struct btrfs_root *root = BTRFS_I(inode)->root;
6947 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6948 BUG_ON(ret); /* -ENOMEM */
6952 static void btrfs_end_dio_bio(struct bio *bio, int err)
6954 struct btrfs_dio_private *dip = bio->bi_private;
6957 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6958 "sector %#Lx len %u err no %d\n",
6959 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6960 (unsigned long long)bio->bi_sector, bio->bi_size, err);
6964 * before atomic variable goto zero, we must make sure
6965 * dip->errors is perceived to be set.
6967 smp_mb__before_atomic_dec();
6970 /* if there are more bios still pending for this dio, just exit */
6971 if (!atomic_dec_and_test(&dip->pending_bios))
6975 bio_io_error(dip->orig_bio);
6977 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
6978 bio_endio(dip->orig_bio, 0);
6984 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6985 u64 first_sector, gfp_t gfp_flags)
6987 int nr_vecs = bio_get_nr_vecs(bdev);
6988 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6991 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6992 int rw, u64 file_offset, int skip_sum,
6995 int write = rw & REQ_WRITE;
6996 struct btrfs_root *root = BTRFS_I(inode)->root;
7000 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7005 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7013 if (write && async_submit) {
7014 ret = btrfs_wq_submit_bio(root->fs_info,
7015 inode, rw, bio, 0, 0,
7017 __btrfs_submit_bio_start_direct_io,
7018 __btrfs_submit_bio_done);
7022 * If we aren't doing async submit, calculate the csum of the
7025 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7028 } else if (!skip_sum) {
7029 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
7035 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7041 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7044 struct inode *inode = dip->inode;
7045 struct btrfs_root *root = BTRFS_I(inode)->root;
7047 struct bio *orig_bio = dip->orig_bio;
7048 struct bio_vec *bvec = orig_bio->bi_io_vec;
7049 u64 start_sector = orig_bio->bi_sector;
7050 u64 file_offset = dip->logical_offset;
7055 int async_submit = 0;
7057 map_length = orig_bio->bi_size;
7058 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7059 &map_length, NULL, 0);
7064 if (map_length >= orig_bio->bi_size) {
7069 /* async crcs make it difficult to collect full stripe writes. */
7070 if (btrfs_get_alloc_profile(root, 1) &
7071 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7076 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7079 bio->bi_private = dip;
7080 bio->bi_end_io = btrfs_end_dio_bio;
7081 atomic_inc(&dip->pending_bios);
7083 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7084 if (unlikely(map_length < submit_len + bvec->bv_len ||
7085 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7086 bvec->bv_offset) < bvec->bv_len)) {
7088 * inc the count before we submit the bio so
7089 * we know the end IO handler won't happen before
7090 * we inc the count. Otherwise, the dip might get freed
7091 * before we're done setting it up
7093 atomic_inc(&dip->pending_bios);
7094 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7095 file_offset, skip_sum,
7099 atomic_dec(&dip->pending_bios);
7103 start_sector += submit_len >> 9;
7104 file_offset += submit_len;
7109 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7110 start_sector, GFP_NOFS);
7113 bio->bi_private = dip;
7114 bio->bi_end_io = btrfs_end_dio_bio;
7116 map_length = orig_bio->bi_size;
7117 ret = btrfs_map_block(root->fs_info, rw,
7119 &map_length, NULL, 0);
7125 submit_len += bvec->bv_len;
7132 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7141 * before atomic variable goto zero, we must
7142 * make sure dip->errors is perceived to be set.
7144 smp_mb__before_atomic_dec();
7145 if (atomic_dec_and_test(&dip->pending_bios))
7146 bio_io_error(dip->orig_bio);
7148 /* bio_end_io() will handle error, so we needn't return it */
7152 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7153 struct inode *inode, loff_t file_offset)
7155 struct btrfs_root *root = BTRFS_I(inode)->root;
7156 struct btrfs_dio_private *dip;
7159 int write = rw & REQ_WRITE;
7162 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7164 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7171 dip = kmalloc(sizeof(*dip), GFP_NOFS);
7177 dip->private = dio_bio->bi_private;
7179 dip->logical_offset = file_offset;
7180 dip->bytes = dio_bio->bi_size;
7181 dip->disk_bytenr = (u64)dio_bio->bi_sector << 9;
7182 io_bio->bi_private = dip;
7184 dip->orig_bio = io_bio;
7185 dip->dio_bio = dio_bio;
7186 atomic_set(&dip->pending_bios, 0);
7189 io_bio->bi_end_io = btrfs_endio_direct_write;
7191 io_bio->bi_end_io = btrfs_endio_direct_read;
7193 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7202 * If this is a write, we need to clean up the reserved space and kill
7203 * the ordered extent.
7206 struct btrfs_ordered_extent *ordered;
7207 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7208 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7209 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7210 btrfs_free_reserved_extent(root, ordered->start,
7212 btrfs_put_ordered_extent(ordered);
7213 btrfs_put_ordered_extent(ordered);
7215 bio_endio(dio_bio, ret);
7218 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7219 const struct iovec *iov, loff_t offset,
7220 unsigned long nr_segs)
7226 unsigned blocksize_mask = root->sectorsize - 1;
7227 ssize_t retval = -EINVAL;
7228 loff_t end = offset;
7230 if (offset & blocksize_mask)
7233 /* Check the memory alignment. Blocks cannot straddle pages */
7234 for (seg = 0; seg < nr_segs; seg++) {
7235 addr = (unsigned long)iov[seg].iov_base;
7236 size = iov[seg].iov_len;
7238 if ((addr & blocksize_mask) || (size & blocksize_mask))
7241 /* If this is a write we don't need to check anymore */
7246 * Check to make sure we don't have duplicate iov_base's in this
7247 * iovec, if so return EINVAL, otherwise we'll get csum errors
7248 * when reading back.
7250 for (i = seg + 1; i < nr_segs; i++) {
7251 if (iov[seg].iov_base == iov[i].iov_base)
7260 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7261 const struct iovec *iov, loff_t offset,
7262 unsigned long nr_segs)
7264 struct file *file = iocb->ki_filp;
7265 struct inode *inode = file->f_mapping->host;
7269 bool relock = false;
7272 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7276 atomic_inc(&inode->i_dio_count);
7277 smp_mb__after_atomic_inc();
7280 * The generic stuff only does filemap_write_and_wait_range, which isn't
7281 * enough if we've written compressed pages to this area, so we need to
7282 * call btrfs_wait_ordered_range to make absolutely sure that any
7283 * outstanding dirty pages are on disk.
7285 count = iov_length(iov, nr_segs);
7286 btrfs_wait_ordered_range(inode, offset, count);
7290 * If the write DIO is beyond the EOF, we need update
7291 * the isize, but it is protected by i_mutex. So we can
7292 * not unlock the i_mutex at this case.
7294 if (offset + count <= inode->i_size) {
7295 mutex_unlock(&inode->i_mutex);
7298 ret = btrfs_delalloc_reserve_space(inode, count);
7301 } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7302 &BTRFS_I(inode)->runtime_flags))) {
7303 inode_dio_done(inode);
7304 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7308 ret = __blockdev_direct_IO(rw, iocb, inode,
7309 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7310 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7311 btrfs_submit_direct, flags);
7313 if (ret < 0 && ret != -EIOCBQUEUED)
7314 btrfs_delalloc_release_space(inode, count);
7315 else if (ret >= 0 && (size_t)ret < count)
7316 btrfs_delalloc_release_space(inode,
7317 count - (size_t)ret);
7319 btrfs_delalloc_release_metadata(inode, 0);
7323 inode_dio_done(inode);
7325 mutex_lock(&inode->i_mutex);
7330 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
7332 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7333 __u64 start, __u64 len)
7337 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7341 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7344 int btrfs_readpage(struct file *file, struct page *page)
7346 struct extent_io_tree *tree;
7347 tree = &BTRFS_I(page->mapping->host)->io_tree;
7348 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7351 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7353 struct extent_io_tree *tree;
7356 if (current->flags & PF_MEMALLOC) {
7357 redirty_page_for_writepage(wbc, page);
7361 tree = &BTRFS_I(page->mapping->host)->io_tree;
7362 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7365 static int btrfs_writepages(struct address_space *mapping,
7366 struct writeback_control *wbc)
7368 struct extent_io_tree *tree;
7370 tree = &BTRFS_I(mapping->host)->io_tree;
7371 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7375 btrfs_readpages(struct file *file, struct address_space *mapping,
7376 struct list_head *pages, unsigned nr_pages)
7378 struct extent_io_tree *tree;
7379 tree = &BTRFS_I(mapping->host)->io_tree;
7380 return extent_readpages(tree, mapping, pages, nr_pages,
7383 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7385 struct extent_io_tree *tree;
7386 struct extent_map_tree *map;
7389 tree = &BTRFS_I(page->mapping->host)->io_tree;
7390 map = &BTRFS_I(page->mapping->host)->extent_tree;
7391 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7393 ClearPagePrivate(page);
7394 set_page_private(page, 0);
7395 page_cache_release(page);
7400 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7402 if (PageWriteback(page) || PageDirty(page))
7404 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7407 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
7408 unsigned int length)
7410 struct inode *inode = page->mapping->host;
7411 struct extent_io_tree *tree;
7412 struct btrfs_ordered_extent *ordered;
7413 struct extent_state *cached_state = NULL;
7414 u64 page_start = page_offset(page);
7415 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7418 * we have the page locked, so new writeback can't start,
7419 * and the dirty bit won't be cleared while we are here.
7421 * Wait for IO on this page so that we can safely clear
7422 * the PagePrivate2 bit and do ordered accounting
7424 wait_on_page_writeback(page);
7426 tree = &BTRFS_I(inode)->io_tree;
7428 btrfs_releasepage(page, GFP_NOFS);
7431 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7432 ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
7435 * IO on this page will never be started, so we need
7436 * to account for any ordered extents now
7438 clear_extent_bit(tree, page_start, page_end,
7439 EXTENT_DIRTY | EXTENT_DELALLOC |
7440 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7441 EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
7443 * whoever cleared the private bit is responsible
7444 * for the finish_ordered_io
7446 if (TestClearPagePrivate2(page) &&
7447 btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
7448 PAGE_CACHE_SIZE, 1)) {
7449 btrfs_finish_ordered_io(ordered);
7451 btrfs_put_ordered_extent(ordered);
7452 cached_state = NULL;
7453 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7455 clear_extent_bit(tree, page_start, page_end,
7456 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
7457 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
7458 &cached_state, GFP_NOFS);
7459 __btrfs_releasepage(page, GFP_NOFS);
7461 ClearPageChecked(page);
7462 if (PagePrivate(page)) {
7463 ClearPagePrivate(page);
7464 set_page_private(page, 0);
7465 page_cache_release(page);
7470 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7471 * called from a page fault handler when a page is first dirtied. Hence we must
7472 * be careful to check for EOF conditions here. We set the page up correctly
7473 * for a written page which means we get ENOSPC checking when writing into
7474 * holes and correct delalloc and unwritten extent mapping on filesystems that
7475 * support these features.
7477 * We are not allowed to take the i_mutex here so we have to play games to
7478 * protect against truncate races as the page could now be beyond EOF. Because
7479 * vmtruncate() writes the inode size before removing pages, once we have the
7480 * page lock we can determine safely if the page is beyond EOF. If it is not
7481 * beyond EOF, then the page is guaranteed safe against truncation until we
7484 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7486 struct page *page = vmf->page;
7487 struct inode *inode = file_inode(vma->vm_file);
7488 struct btrfs_root *root = BTRFS_I(inode)->root;
7489 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7490 struct btrfs_ordered_extent *ordered;
7491 struct extent_state *cached_state = NULL;
7493 unsigned long zero_start;
7500 sb_start_pagefault(inode->i_sb);
7501 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7503 ret = file_update_time(vma->vm_file);
7509 else /* -ENOSPC, -EIO, etc */
7510 ret = VM_FAULT_SIGBUS;
7516 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7519 size = i_size_read(inode);
7520 page_start = page_offset(page);
7521 page_end = page_start + PAGE_CACHE_SIZE - 1;
7523 if ((page->mapping != inode->i_mapping) ||
7524 (page_start >= size)) {
7525 /* page got truncated out from underneath us */
7528 wait_on_page_writeback(page);
7530 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7531 set_page_extent_mapped(page);
7534 * we can't set the delalloc bits if there are pending ordered
7535 * extents. Drop our locks and wait for them to finish
7537 ordered = btrfs_lookup_ordered_extent(inode, page_start);
7539 unlock_extent_cached(io_tree, page_start, page_end,
7540 &cached_state, GFP_NOFS);
7542 btrfs_start_ordered_extent(inode, ordered, 1);
7543 btrfs_put_ordered_extent(ordered);
7548 * XXX - page_mkwrite gets called every time the page is dirtied, even
7549 * if it was already dirty, so for space accounting reasons we need to
7550 * clear any delalloc bits for the range we are fixing to save. There
7551 * is probably a better way to do this, but for now keep consistent with
7552 * prepare_pages in the normal write path.
7554 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7555 EXTENT_DIRTY | EXTENT_DELALLOC |
7556 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7557 0, 0, &cached_state, GFP_NOFS);
7559 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7562 unlock_extent_cached(io_tree, page_start, page_end,
7563 &cached_state, GFP_NOFS);
7564 ret = VM_FAULT_SIGBUS;
7569 /* page is wholly or partially inside EOF */
7570 if (page_start + PAGE_CACHE_SIZE > size)
7571 zero_start = size & ~PAGE_CACHE_MASK;
7573 zero_start = PAGE_CACHE_SIZE;
7575 if (zero_start != PAGE_CACHE_SIZE) {
7577 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7578 flush_dcache_page(page);
7581 ClearPageChecked(page);
7582 set_page_dirty(page);
7583 SetPageUptodate(page);
7585 BTRFS_I(inode)->last_trans = root->fs_info->generation;
7586 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7587 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7589 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7593 sb_end_pagefault(inode->i_sb);
7594 return VM_FAULT_LOCKED;
7598 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7600 sb_end_pagefault(inode->i_sb);
7604 static int btrfs_truncate(struct inode *inode)
7606 struct btrfs_root *root = BTRFS_I(inode)->root;
7607 struct btrfs_block_rsv *rsv;
7610 struct btrfs_trans_handle *trans;
7611 u64 mask = root->sectorsize - 1;
7612 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7614 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
7615 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
7618 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
7619 * 3 things going on here
7621 * 1) We need to reserve space for our orphan item and the space to
7622 * delete our orphan item. Lord knows we don't want to have a dangling
7623 * orphan item because we didn't reserve space to remove it.
7625 * 2) We need to reserve space to update our inode.
7627 * 3) We need to have something to cache all the space that is going to
7628 * be free'd up by the truncate operation, but also have some slack
7629 * space reserved in case it uses space during the truncate (thank you
7630 * very much snapshotting).
7632 * And we need these to all be seperate. The fact is we can use alot of
7633 * space doing the truncate, and we have no earthly idea how much space
7634 * we will use, so we need the truncate reservation to be seperate so it
7635 * doesn't end up using space reserved for updating the inode or
7636 * removing the orphan item. We also need to be able to stop the
7637 * transaction and start a new one, which means we need to be able to
7638 * update the inode several times, and we have no idea of knowing how
7639 * many times that will be, so we can't just reserve 1 item for the
7640 * entirety of the opration, so that has to be done seperately as well.
7641 * Then there is the orphan item, which does indeed need to be held on
7642 * to for the whole operation, and we need nobody to touch this reserved
7643 * space except the orphan code.
7645 * So that leaves us with
7647 * 1) root->orphan_block_rsv - for the orphan deletion.
7648 * 2) rsv - for the truncate reservation, which we will steal from the
7649 * transaction reservation.
7650 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7651 * updating the inode.
7653 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7656 rsv->size = min_size;
7660 * 1 for the truncate slack space
7661 * 1 for updating the inode.
7663 trans = btrfs_start_transaction(root, 2);
7664 if (IS_ERR(trans)) {
7665 err = PTR_ERR(trans);
7669 /* Migrate the slack space for the truncate to our reserve */
7670 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7675 * setattr is responsible for setting the ordered_data_close flag,
7676 * but that is only tested during the last file release. That
7677 * could happen well after the next commit, leaving a great big
7678 * window where new writes may get lost if someone chooses to write
7679 * to this file after truncating to zero
7681 * The inode doesn't have any dirty data here, and so if we commit
7682 * this is a noop. If someone immediately starts writing to the inode
7683 * it is very likely we'll catch some of their writes in this
7684 * transaction, and the commit will find this file on the ordered
7685 * data list with good things to send down.
7687 * This is a best effort solution, there is still a window where
7688 * using truncate to replace the contents of the file will
7689 * end up with a zero length file after a crash.
7691 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7692 &BTRFS_I(inode)->runtime_flags))
7693 btrfs_add_ordered_operation(trans, root, inode);
7696 * So if we truncate and then write and fsync we normally would just
7697 * write the extents that changed, which is a problem if we need to
7698 * first truncate that entire inode. So set this flag so we write out
7699 * all of the extents in the inode to the sync log so we're completely
7702 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7703 trans->block_rsv = rsv;
7706 ret = btrfs_truncate_inode_items(trans, root, inode,
7708 BTRFS_EXTENT_DATA_KEY);
7709 if (ret != -ENOSPC) {
7714 trans->block_rsv = &root->fs_info->trans_block_rsv;
7715 ret = btrfs_update_inode(trans, root, inode);
7721 btrfs_end_transaction(trans, root);
7722 btrfs_btree_balance_dirty(root);
7724 trans = btrfs_start_transaction(root, 2);
7725 if (IS_ERR(trans)) {
7726 ret = err = PTR_ERR(trans);
7731 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7733 BUG_ON(ret); /* shouldn't happen */
7734 trans->block_rsv = rsv;
7737 if (ret == 0 && inode->i_nlink > 0) {
7738 trans->block_rsv = root->orphan_block_rsv;
7739 ret = btrfs_orphan_del(trans, inode);
7745 trans->block_rsv = &root->fs_info->trans_block_rsv;
7746 ret = btrfs_update_inode(trans, root, inode);
7750 ret = btrfs_end_transaction(trans, root);
7751 btrfs_btree_balance_dirty(root);
7755 btrfs_free_block_rsv(root, rsv);
7764 * create a new subvolume directory/inode (helper for the ioctl).
7766 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7767 struct btrfs_root *new_root, u64 new_dirid)
7769 struct inode *inode;
7773 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7774 new_dirid, new_dirid,
7775 S_IFDIR | (~current_umask() & S_IRWXUGO),
7778 return PTR_ERR(inode);
7779 inode->i_op = &btrfs_dir_inode_operations;
7780 inode->i_fop = &btrfs_dir_file_operations;
7782 set_nlink(inode, 1);
7783 btrfs_i_size_write(inode, 0);
7785 err = btrfs_update_inode(trans, new_root, inode);
7791 struct inode *btrfs_alloc_inode(struct super_block *sb)
7793 struct btrfs_inode *ei;
7794 struct inode *inode;
7796 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7803 ei->last_sub_trans = 0;
7804 ei->logged_trans = 0;
7805 ei->delalloc_bytes = 0;
7806 ei->disk_i_size = 0;
7809 ei->index_cnt = (u64)-1;
7810 ei->last_unlink_trans = 0;
7811 ei->last_log_commit = 0;
7813 spin_lock_init(&ei->lock);
7814 ei->outstanding_extents = 0;
7815 ei->reserved_extents = 0;
7817 ei->runtime_flags = 0;
7818 ei->force_compress = BTRFS_COMPRESS_NONE;
7820 ei->delayed_node = NULL;
7822 inode = &ei->vfs_inode;
7823 extent_map_tree_init(&ei->extent_tree);
7824 extent_io_tree_init(&ei->io_tree, &inode->i_data);
7825 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7826 ei->io_tree.track_uptodate = 1;
7827 ei->io_failure_tree.track_uptodate = 1;
7828 atomic_set(&ei->sync_writers, 0);
7829 mutex_init(&ei->log_mutex);
7830 mutex_init(&ei->delalloc_mutex);
7831 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7832 INIT_LIST_HEAD(&ei->delalloc_inodes);
7833 INIT_LIST_HEAD(&ei->ordered_operations);
7834 RB_CLEAR_NODE(&ei->rb_node);
7839 static void btrfs_i_callback(struct rcu_head *head)
7841 struct inode *inode = container_of(head, struct inode, i_rcu);
7842 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7845 void btrfs_destroy_inode(struct inode *inode)
7847 struct btrfs_ordered_extent *ordered;
7848 struct btrfs_root *root = BTRFS_I(inode)->root;
7850 WARN_ON(!hlist_empty(&inode->i_dentry));
7851 WARN_ON(inode->i_data.nrpages);
7852 WARN_ON(BTRFS_I(inode)->outstanding_extents);
7853 WARN_ON(BTRFS_I(inode)->reserved_extents);
7854 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7855 WARN_ON(BTRFS_I(inode)->csum_bytes);
7858 * This can happen where we create an inode, but somebody else also
7859 * created the same inode and we need to destroy the one we already
7866 * Make sure we're properly removed from the ordered operation
7870 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7871 spin_lock(&root->fs_info->ordered_root_lock);
7872 list_del_init(&BTRFS_I(inode)->ordered_operations);
7873 spin_unlock(&root->fs_info->ordered_root_lock);
7876 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7877 &BTRFS_I(inode)->runtime_flags)) {
7878 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
7879 (unsigned long long)btrfs_ino(inode));
7880 atomic_dec(&root->orphan_inodes);
7884 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7888 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
7889 (unsigned long long)ordered->file_offset,
7890 (unsigned long long)ordered->len);
7891 btrfs_remove_ordered_extent(inode, ordered);
7892 btrfs_put_ordered_extent(ordered);
7893 btrfs_put_ordered_extent(ordered);
7896 inode_tree_del(inode);
7897 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7899 call_rcu(&inode->i_rcu, btrfs_i_callback);
7902 int btrfs_drop_inode(struct inode *inode)
7904 struct btrfs_root *root = BTRFS_I(inode)->root;
7909 /* the snap/subvol tree is on deleting */
7910 if (btrfs_root_refs(&root->root_item) == 0 &&
7911 root != root->fs_info->tree_root)
7914 return generic_drop_inode(inode);
7917 static void init_once(void *foo)
7919 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7921 inode_init_once(&ei->vfs_inode);
7924 void btrfs_destroy_cachep(void)
7927 * Make sure all delayed rcu free inodes are flushed before we
7931 if (btrfs_inode_cachep)
7932 kmem_cache_destroy(btrfs_inode_cachep);
7933 if (btrfs_trans_handle_cachep)
7934 kmem_cache_destroy(btrfs_trans_handle_cachep);
7935 if (btrfs_transaction_cachep)
7936 kmem_cache_destroy(btrfs_transaction_cachep);
7937 if (btrfs_path_cachep)
7938 kmem_cache_destroy(btrfs_path_cachep);
7939 if (btrfs_free_space_cachep)
7940 kmem_cache_destroy(btrfs_free_space_cachep);
7941 if (btrfs_delalloc_work_cachep)
7942 kmem_cache_destroy(btrfs_delalloc_work_cachep);
7945 int btrfs_init_cachep(void)
7947 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
7948 sizeof(struct btrfs_inode), 0,
7949 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7950 if (!btrfs_inode_cachep)
7953 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
7954 sizeof(struct btrfs_trans_handle), 0,
7955 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7956 if (!btrfs_trans_handle_cachep)
7959 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
7960 sizeof(struct btrfs_transaction), 0,
7961 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7962 if (!btrfs_transaction_cachep)
7965 btrfs_path_cachep = kmem_cache_create("btrfs_path",
7966 sizeof(struct btrfs_path), 0,
7967 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7968 if (!btrfs_path_cachep)
7971 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
7972 sizeof(struct btrfs_free_space), 0,
7973 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7974 if (!btrfs_free_space_cachep)
7977 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
7978 sizeof(struct btrfs_delalloc_work), 0,
7979 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
7981 if (!btrfs_delalloc_work_cachep)
7986 btrfs_destroy_cachep();
7990 static int btrfs_getattr(struct vfsmount *mnt,
7991 struct dentry *dentry, struct kstat *stat)
7994 struct inode *inode = dentry->d_inode;
7995 u32 blocksize = inode->i_sb->s_blocksize;
7997 generic_fillattr(inode, stat);
7998 stat->dev = BTRFS_I(inode)->root->anon_dev;
7999 stat->blksize = PAGE_CACHE_SIZE;
8001 spin_lock(&BTRFS_I(inode)->lock);
8002 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8003 spin_unlock(&BTRFS_I(inode)->lock);
8004 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8005 ALIGN(delalloc_bytes, blocksize)) >> 9;
8009 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8010 struct inode *new_dir, struct dentry *new_dentry)
8012 struct btrfs_trans_handle *trans;
8013 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8014 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8015 struct inode *new_inode = new_dentry->d_inode;
8016 struct inode *old_inode = old_dentry->d_inode;
8017 struct timespec ctime = CURRENT_TIME;
8021 u64 old_ino = btrfs_ino(old_inode);
8023 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8026 /* we only allow rename subvolume link between subvolumes */
8027 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8030 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8031 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8034 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8035 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8039 /* check for collisions, even if the name isn't there */
8040 ret = btrfs_check_dir_item_collision(root, new_dir->i_ino,
8041 new_dentry->d_name.name,
8042 new_dentry->d_name.len);
8045 if (ret == -EEXIST) {
8047 * eexist without a new_inode */
8053 /* maybe -EOVERFLOW */
8060 * we're using rename to replace one file with another.
8061 * and the replacement file is large. Start IO on it now so
8062 * we don't add too much work to the end of the transaction
8064 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8065 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8066 filemap_flush(old_inode->i_mapping);
8068 /* close the racy window with snapshot create/destroy ioctl */
8069 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8070 down_read(&root->fs_info->subvol_sem);
8072 * We want to reserve the absolute worst case amount of items. So if
8073 * both inodes are subvols and we need to unlink them then that would
8074 * require 4 item modifications, but if they are both normal inodes it
8075 * would require 5 item modifications, so we'll assume their normal
8076 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8077 * should cover the worst case number of items we'll modify.
8079 trans = btrfs_start_transaction(root, 11);
8080 if (IS_ERR(trans)) {
8081 ret = PTR_ERR(trans);
8086 btrfs_record_root_in_trans(trans, dest);
8088 ret = btrfs_set_inode_index(new_dir, &index);
8092 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8093 /* force full log commit if subvolume involved. */
8094 root->fs_info->last_trans_log_full_commit = trans->transid;
8096 ret = btrfs_insert_inode_ref(trans, dest,
8097 new_dentry->d_name.name,
8098 new_dentry->d_name.len,
8100 btrfs_ino(new_dir), index);
8104 * this is an ugly little race, but the rename is required
8105 * to make sure that if we crash, the inode is either at the
8106 * old name or the new one. pinning the log transaction lets
8107 * us make sure we don't allow a log commit to come in after
8108 * we unlink the name but before we add the new name back in.
8110 btrfs_pin_log_trans(root);
8113 * make sure the inode gets flushed if it is replacing
8116 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8117 btrfs_add_ordered_operation(trans, root, old_inode);
8119 inode_inc_iversion(old_dir);
8120 inode_inc_iversion(new_dir);
8121 inode_inc_iversion(old_inode);
8122 old_dir->i_ctime = old_dir->i_mtime = ctime;
8123 new_dir->i_ctime = new_dir->i_mtime = ctime;
8124 old_inode->i_ctime = ctime;
8126 if (old_dentry->d_parent != new_dentry->d_parent)
8127 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8129 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8130 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8131 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8132 old_dentry->d_name.name,
8133 old_dentry->d_name.len);
8135 ret = __btrfs_unlink_inode(trans, root, old_dir,
8136 old_dentry->d_inode,
8137 old_dentry->d_name.name,
8138 old_dentry->d_name.len);
8140 ret = btrfs_update_inode(trans, root, old_inode);
8143 btrfs_abort_transaction(trans, root, ret);
8148 inode_inc_iversion(new_inode);
8149 new_inode->i_ctime = CURRENT_TIME;
8150 if (unlikely(btrfs_ino(new_inode) ==
8151 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8152 root_objectid = BTRFS_I(new_inode)->location.objectid;
8153 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8155 new_dentry->d_name.name,
8156 new_dentry->d_name.len);
8157 BUG_ON(new_inode->i_nlink == 0);
8159 ret = btrfs_unlink_inode(trans, dest, new_dir,
8160 new_dentry->d_inode,
8161 new_dentry->d_name.name,
8162 new_dentry->d_name.len);
8164 if (!ret && new_inode->i_nlink == 0) {
8165 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8169 btrfs_abort_transaction(trans, root, ret);
8174 ret = btrfs_add_link(trans, new_dir, old_inode,
8175 new_dentry->d_name.name,
8176 new_dentry->d_name.len, 0, index);
8178 btrfs_abort_transaction(trans, root, ret);
8182 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8183 struct dentry *parent = new_dentry->d_parent;
8184 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8185 btrfs_end_log_trans(root);
8188 btrfs_end_transaction(trans, root);
8190 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8191 up_read(&root->fs_info->subvol_sem);
8196 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8198 struct btrfs_delalloc_work *delalloc_work;
8200 delalloc_work = container_of(work, struct btrfs_delalloc_work,
8202 if (delalloc_work->wait)
8203 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
8205 filemap_flush(delalloc_work->inode->i_mapping);
8207 if (delalloc_work->delay_iput)
8208 btrfs_add_delayed_iput(delalloc_work->inode);
8210 iput(delalloc_work->inode);
8211 complete(&delalloc_work->completion);
8214 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8215 int wait, int delay_iput)
8217 struct btrfs_delalloc_work *work;
8219 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8223 init_completion(&work->completion);
8224 INIT_LIST_HEAD(&work->list);
8225 work->inode = inode;
8227 work->delay_iput = delay_iput;
8228 work->work.func = btrfs_run_delalloc_work;
8233 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8235 wait_for_completion(&work->completion);
8236 kmem_cache_free(btrfs_delalloc_work_cachep, work);
8240 * some fairly slow code that needs optimization. This walks the list
8241 * of all the inodes with pending delalloc and forces them to disk.
8243 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8245 struct btrfs_inode *binode;
8246 struct inode *inode;
8247 struct btrfs_delalloc_work *work, *next;
8248 struct list_head works;
8249 struct list_head splice;
8252 INIT_LIST_HEAD(&works);
8253 INIT_LIST_HEAD(&splice);
8255 spin_lock(&root->delalloc_lock);
8256 list_splice_init(&root->delalloc_inodes, &splice);
8257 while (!list_empty(&splice)) {
8258 binode = list_entry(splice.next, struct btrfs_inode,
8261 list_move_tail(&binode->delalloc_inodes,
8262 &root->delalloc_inodes);
8263 inode = igrab(&binode->vfs_inode);
8265 cond_resched_lock(&root->delalloc_lock);
8268 spin_unlock(&root->delalloc_lock);
8270 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8271 if (unlikely(!work)) {
8275 list_add_tail(&work->list, &works);
8276 btrfs_queue_worker(&root->fs_info->flush_workers,
8280 spin_lock(&root->delalloc_lock);
8282 spin_unlock(&root->delalloc_lock);
8284 list_for_each_entry_safe(work, next, &works, list) {
8285 list_del_init(&work->list);
8286 btrfs_wait_and_free_delalloc_work(work);
8290 list_for_each_entry_safe(work, next, &works, list) {
8291 list_del_init(&work->list);
8292 btrfs_wait_and_free_delalloc_work(work);
8295 if (!list_empty_careful(&splice)) {
8296 spin_lock(&root->delalloc_lock);
8297 list_splice_tail(&splice, &root->delalloc_inodes);
8298 spin_unlock(&root->delalloc_lock);
8303 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8307 if (root->fs_info->sb->s_flags & MS_RDONLY)
8310 ret = __start_delalloc_inodes(root, delay_iput);
8312 * the filemap_flush will queue IO into the worker threads, but
8313 * we have to make sure the IO is actually started and that
8314 * ordered extents get created before we return
8316 atomic_inc(&root->fs_info->async_submit_draining);
8317 while (atomic_read(&root->fs_info->nr_async_submits) ||
8318 atomic_read(&root->fs_info->async_delalloc_pages)) {
8319 wait_event(root->fs_info->async_submit_wait,
8320 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8321 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8323 atomic_dec(&root->fs_info->async_submit_draining);
8327 int btrfs_start_all_delalloc_inodes(struct btrfs_fs_info *fs_info,
8330 struct btrfs_root *root;
8331 struct list_head splice;
8334 if (fs_info->sb->s_flags & MS_RDONLY)
8337 INIT_LIST_HEAD(&splice);
8339 spin_lock(&fs_info->delalloc_root_lock);
8340 list_splice_init(&fs_info->delalloc_roots, &splice);
8341 while (!list_empty(&splice)) {
8342 root = list_first_entry(&splice, struct btrfs_root,
8344 root = btrfs_grab_fs_root(root);
8346 list_move_tail(&root->delalloc_root,
8347 &fs_info->delalloc_roots);
8348 spin_unlock(&fs_info->delalloc_root_lock);
8350 ret = __start_delalloc_inodes(root, delay_iput);
8351 btrfs_put_fs_root(root);
8355 spin_lock(&fs_info->delalloc_root_lock);
8357 spin_unlock(&fs_info->delalloc_root_lock);
8359 atomic_inc(&fs_info->async_submit_draining);
8360 while (atomic_read(&fs_info->nr_async_submits) ||
8361 atomic_read(&fs_info->async_delalloc_pages)) {
8362 wait_event(fs_info->async_submit_wait,
8363 (atomic_read(&fs_info->nr_async_submits) == 0 &&
8364 atomic_read(&fs_info->async_delalloc_pages) == 0));
8366 atomic_dec(&fs_info->async_submit_draining);
8369 if (!list_empty_careful(&splice)) {
8370 spin_lock(&fs_info->delalloc_root_lock);
8371 list_splice_tail(&splice, &fs_info->delalloc_roots);
8372 spin_unlock(&fs_info->delalloc_root_lock);
8377 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8378 const char *symname)
8380 struct btrfs_trans_handle *trans;
8381 struct btrfs_root *root = BTRFS_I(dir)->root;
8382 struct btrfs_path *path;
8383 struct btrfs_key key;
8384 struct inode *inode = NULL;
8392 struct btrfs_file_extent_item *ei;
8393 struct extent_buffer *leaf;
8395 name_len = strlen(symname) + 1;
8396 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8397 return -ENAMETOOLONG;
8400 * 2 items for inode item and ref
8401 * 2 items for dir items
8402 * 1 item for xattr if selinux is on
8404 trans = btrfs_start_transaction(root, 5);
8406 return PTR_ERR(trans);
8408 err = btrfs_find_free_ino(root, &objectid);
8412 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8413 dentry->d_name.len, btrfs_ino(dir), objectid,
8414 S_IFLNK|S_IRWXUGO, &index);
8415 if (IS_ERR(inode)) {
8416 err = PTR_ERR(inode);
8420 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8427 * If the active LSM wants to access the inode during
8428 * d_instantiate it needs these. Smack checks to see
8429 * if the filesystem supports xattrs by looking at the
8432 inode->i_fop = &btrfs_file_operations;
8433 inode->i_op = &btrfs_file_inode_operations;
8435 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8439 inode->i_mapping->a_ops = &btrfs_aops;
8440 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8441 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8446 path = btrfs_alloc_path();
8452 key.objectid = btrfs_ino(inode);
8454 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8455 datasize = btrfs_file_extent_calc_inline_size(name_len);
8456 err = btrfs_insert_empty_item(trans, root, path, &key,
8460 btrfs_free_path(path);
8463 leaf = path->nodes[0];
8464 ei = btrfs_item_ptr(leaf, path->slots[0],
8465 struct btrfs_file_extent_item);
8466 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8467 btrfs_set_file_extent_type(leaf, ei,
8468 BTRFS_FILE_EXTENT_INLINE);
8469 btrfs_set_file_extent_encryption(leaf, ei, 0);
8470 btrfs_set_file_extent_compression(leaf, ei, 0);
8471 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8472 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8474 ptr = btrfs_file_extent_inline_start(ei);
8475 write_extent_buffer(leaf, symname, ptr, name_len);
8476 btrfs_mark_buffer_dirty(leaf);
8477 btrfs_free_path(path);
8479 inode->i_op = &btrfs_symlink_inode_operations;
8480 inode->i_mapping->a_ops = &btrfs_symlink_aops;
8481 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8482 inode_set_bytes(inode, name_len);
8483 btrfs_i_size_write(inode, name_len - 1);
8484 err = btrfs_update_inode(trans, root, inode);
8490 d_instantiate(dentry, inode);
8491 btrfs_end_transaction(trans, root);
8493 inode_dec_link_count(inode);
8496 btrfs_btree_balance_dirty(root);
8500 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8501 u64 start, u64 num_bytes, u64 min_size,
8502 loff_t actual_len, u64 *alloc_hint,
8503 struct btrfs_trans_handle *trans)
8505 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8506 struct extent_map *em;
8507 struct btrfs_root *root = BTRFS_I(inode)->root;
8508 struct btrfs_key ins;
8509 u64 cur_offset = start;
8513 bool own_trans = true;
8517 while (num_bytes > 0) {
8519 trans = btrfs_start_transaction(root, 3);
8520 if (IS_ERR(trans)) {
8521 ret = PTR_ERR(trans);
8526 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8527 cur_bytes = max(cur_bytes, min_size);
8528 ret = btrfs_reserve_extent(trans, root, cur_bytes,
8529 min_size, 0, *alloc_hint, &ins, 1);
8532 btrfs_end_transaction(trans, root);
8536 ret = insert_reserved_file_extent(trans, inode,
8537 cur_offset, ins.objectid,
8538 ins.offset, ins.offset,
8539 ins.offset, 0, 0, 0,
8540 BTRFS_FILE_EXTENT_PREALLOC);
8542 btrfs_abort_transaction(trans, root, ret);
8544 btrfs_end_transaction(trans, root);
8547 btrfs_drop_extent_cache(inode, cur_offset,
8548 cur_offset + ins.offset -1, 0);
8550 em = alloc_extent_map();
8552 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8553 &BTRFS_I(inode)->runtime_flags);
8557 em->start = cur_offset;
8558 em->orig_start = cur_offset;
8559 em->len = ins.offset;
8560 em->block_start = ins.objectid;
8561 em->block_len = ins.offset;
8562 em->orig_block_len = ins.offset;
8563 em->ram_bytes = ins.offset;
8564 em->bdev = root->fs_info->fs_devices->latest_bdev;
8565 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8566 em->generation = trans->transid;
8569 write_lock(&em_tree->lock);
8570 ret = add_extent_mapping(em_tree, em, 1);
8571 write_unlock(&em_tree->lock);
8574 btrfs_drop_extent_cache(inode, cur_offset,
8575 cur_offset + ins.offset - 1,
8578 free_extent_map(em);
8580 num_bytes -= ins.offset;
8581 cur_offset += ins.offset;
8582 *alloc_hint = ins.objectid + ins.offset;
8584 inode_inc_iversion(inode);
8585 inode->i_ctime = CURRENT_TIME;
8586 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8587 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8588 (actual_len > inode->i_size) &&
8589 (cur_offset > inode->i_size)) {
8590 if (cur_offset > actual_len)
8591 i_size = actual_len;
8593 i_size = cur_offset;
8594 i_size_write(inode, i_size);
8595 btrfs_ordered_update_i_size(inode, i_size, NULL);
8598 ret = btrfs_update_inode(trans, root, inode);
8601 btrfs_abort_transaction(trans, root, ret);
8603 btrfs_end_transaction(trans, root);
8608 btrfs_end_transaction(trans, root);
8613 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8614 u64 start, u64 num_bytes, u64 min_size,
8615 loff_t actual_len, u64 *alloc_hint)
8617 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8618 min_size, actual_len, alloc_hint,
8622 int btrfs_prealloc_file_range_trans(struct inode *inode,
8623 struct btrfs_trans_handle *trans, int mode,
8624 u64 start, u64 num_bytes, u64 min_size,
8625 loff_t actual_len, u64 *alloc_hint)
8627 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8628 min_size, actual_len, alloc_hint, trans);
8631 static int btrfs_set_page_dirty(struct page *page)
8633 return __set_page_dirty_nobuffers(page);
8636 static int btrfs_permission(struct inode *inode, int mask)
8638 struct btrfs_root *root = BTRFS_I(inode)->root;
8639 umode_t mode = inode->i_mode;
8641 if (mask & MAY_WRITE &&
8642 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8643 if (btrfs_root_readonly(root))
8645 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8648 return generic_permission(inode, mask);
8651 static const struct inode_operations btrfs_dir_inode_operations = {
8652 .getattr = btrfs_getattr,
8653 .lookup = btrfs_lookup,
8654 .create = btrfs_create,
8655 .unlink = btrfs_unlink,
8657 .mkdir = btrfs_mkdir,
8658 .rmdir = btrfs_rmdir,
8659 .rename = btrfs_rename,
8660 .symlink = btrfs_symlink,
8661 .setattr = btrfs_setattr,
8662 .mknod = btrfs_mknod,
8663 .setxattr = btrfs_setxattr,
8664 .getxattr = btrfs_getxattr,
8665 .listxattr = btrfs_listxattr,
8666 .removexattr = btrfs_removexattr,
8667 .permission = btrfs_permission,
8668 .get_acl = btrfs_get_acl,
8670 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8671 .lookup = btrfs_lookup,
8672 .permission = btrfs_permission,
8673 .get_acl = btrfs_get_acl,
8676 static const struct file_operations btrfs_dir_file_operations = {
8677 .llseek = generic_file_llseek,
8678 .read = generic_read_dir,
8679 .iterate = btrfs_real_readdir,
8680 .unlocked_ioctl = btrfs_ioctl,
8681 #ifdef CONFIG_COMPAT
8682 .compat_ioctl = btrfs_ioctl,
8684 .release = btrfs_release_file,
8685 .fsync = btrfs_sync_file,
8688 static struct extent_io_ops btrfs_extent_io_ops = {
8689 .fill_delalloc = run_delalloc_range,
8690 .submit_bio_hook = btrfs_submit_bio_hook,
8691 .merge_bio_hook = btrfs_merge_bio_hook,
8692 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
8693 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
8694 .writepage_start_hook = btrfs_writepage_start_hook,
8695 .set_bit_hook = btrfs_set_bit_hook,
8696 .clear_bit_hook = btrfs_clear_bit_hook,
8697 .merge_extent_hook = btrfs_merge_extent_hook,
8698 .split_extent_hook = btrfs_split_extent_hook,
8702 * btrfs doesn't support the bmap operation because swapfiles
8703 * use bmap to make a mapping of extents in the file. They assume
8704 * these extents won't change over the life of the file and they
8705 * use the bmap result to do IO directly to the drive.
8707 * the btrfs bmap call would return logical addresses that aren't
8708 * suitable for IO and they also will change frequently as COW
8709 * operations happen. So, swapfile + btrfs == corruption.
8711 * For now we're avoiding this by dropping bmap.
8713 static const struct address_space_operations btrfs_aops = {
8714 .readpage = btrfs_readpage,
8715 .writepage = btrfs_writepage,
8716 .writepages = btrfs_writepages,
8717 .readpages = btrfs_readpages,
8718 .direct_IO = btrfs_direct_IO,
8719 .invalidatepage = btrfs_invalidatepage,
8720 .releasepage = btrfs_releasepage,
8721 .set_page_dirty = btrfs_set_page_dirty,
8722 .error_remove_page = generic_error_remove_page,
8725 static const struct address_space_operations btrfs_symlink_aops = {
8726 .readpage = btrfs_readpage,
8727 .writepage = btrfs_writepage,
8728 .invalidatepage = btrfs_invalidatepage,
8729 .releasepage = btrfs_releasepage,
8732 static const struct inode_operations btrfs_file_inode_operations = {
8733 .getattr = btrfs_getattr,
8734 .setattr = btrfs_setattr,
8735 .setxattr = btrfs_setxattr,
8736 .getxattr = btrfs_getxattr,
8737 .listxattr = btrfs_listxattr,
8738 .removexattr = btrfs_removexattr,
8739 .permission = btrfs_permission,
8740 .fiemap = btrfs_fiemap,
8741 .get_acl = btrfs_get_acl,
8742 .update_time = btrfs_update_time,
8744 static const struct inode_operations btrfs_special_inode_operations = {
8745 .getattr = btrfs_getattr,
8746 .setattr = btrfs_setattr,
8747 .permission = btrfs_permission,
8748 .setxattr = btrfs_setxattr,
8749 .getxattr = btrfs_getxattr,
8750 .listxattr = btrfs_listxattr,
8751 .removexattr = btrfs_removexattr,
8752 .get_acl = btrfs_get_acl,
8753 .update_time = btrfs_update_time,
8755 static const struct inode_operations btrfs_symlink_inode_operations = {
8756 .readlink = generic_readlink,
8757 .follow_link = page_follow_link_light,
8758 .put_link = page_put_link,
8759 .getattr = btrfs_getattr,
8760 .setattr = btrfs_setattr,
8761 .permission = btrfs_permission,
8762 .setxattr = btrfs_setxattr,
8763 .getxattr = btrfs_getxattr,
8764 .listxattr = btrfs_listxattr,
8765 .removexattr = btrfs_removexattr,
8766 .get_acl = btrfs_get_acl,
8767 .update_time = btrfs_update_time,
8770 const struct dentry_operations btrfs_dentry_operations = {
8771 .d_delete = btrfs_dentry_delete,
8772 .d_release = btrfs_dentry_release,