2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include <linux/mount.h>
45 #include "transaction.h"
46 #include "btrfs_inode.h"
48 #include "print-tree.h"
49 #include "ordered-data.h"
53 #include "compression.h"
55 #include "free-space-cache.h"
56 #include "inode-map.h"
58 struct btrfs_iget_args {
60 struct btrfs_root *root;
63 static const struct inode_operations btrfs_dir_inode_operations;
64 static const struct inode_operations btrfs_symlink_inode_operations;
65 static const struct inode_operations btrfs_dir_ro_inode_operations;
66 static const struct inode_operations btrfs_special_inode_operations;
67 static const struct inode_operations btrfs_file_inode_operations;
68 static const struct address_space_operations btrfs_aops;
69 static const struct address_space_operations btrfs_symlink_aops;
70 static const struct file_operations btrfs_dir_file_operations;
71 static struct extent_io_ops btrfs_extent_io_ops;
73 static struct kmem_cache *btrfs_inode_cachep;
74 struct kmem_cache *btrfs_trans_handle_cachep;
75 struct kmem_cache *btrfs_transaction_cachep;
76 struct kmem_cache *btrfs_path_cachep;
77 struct kmem_cache *btrfs_free_space_cachep;
80 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
81 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
82 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
83 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
84 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
85 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
86 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
87 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
90 static int btrfs_setsize(struct inode *inode, loff_t newsize);
91 static int btrfs_truncate(struct inode *inode);
92 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
93 static noinline int cow_file_range(struct inode *inode,
94 struct page *locked_page,
95 u64 start, u64 end, int *page_started,
96 unsigned long *nr_written, int unlock);
97 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
98 struct btrfs_root *root, struct inode *inode);
100 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
101 struct inode *inode, struct inode *dir,
102 const struct qstr *qstr)
106 err = btrfs_init_acl(trans, inode, dir);
108 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
113 * this does all the hard work for inserting an inline extent into
114 * the btree. The caller should have done a btrfs_drop_extents so that
115 * no overlapping inline items exist in the btree
117 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
118 struct btrfs_root *root, struct inode *inode,
119 u64 start, size_t size, size_t compressed_size,
121 struct page **compressed_pages)
123 struct btrfs_key key;
124 struct btrfs_path *path;
125 struct extent_buffer *leaf;
126 struct page *page = NULL;
129 struct btrfs_file_extent_item *ei;
132 size_t cur_size = size;
134 unsigned long offset;
136 if (compressed_size && compressed_pages)
137 cur_size = compressed_size;
139 path = btrfs_alloc_path();
143 path->leave_spinning = 1;
145 key.objectid = btrfs_ino(inode);
147 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
148 datasize = btrfs_file_extent_calc_inline_size(cur_size);
150 inode_add_bytes(inode, size);
151 ret = btrfs_insert_empty_item(trans, root, path, &key,
157 leaf = path->nodes[0];
158 ei = btrfs_item_ptr(leaf, path->slots[0],
159 struct btrfs_file_extent_item);
160 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
161 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
162 btrfs_set_file_extent_encryption(leaf, ei, 0);
163 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
164 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
165 ptr = btrfs_file_extent_inline_start(ei);
167 if (compress_type != BTRFS_COMPRESS_NONE) {
170 while (compressed_size > 0) {
171 cpage = compressed_pages[i];
172 cur_size = min_t(unsigned long, compressed_size,
175 kaddr = kmap_atomic(cpage);
176 write_extent_buffer(leaf, kaddr, ptr, cur_size);
177 kunmap_atomic(kaddr);
181 compressed_size -= cur_size;
183 btrfs_set_file_extent_compression(leaf, ei,
186 page = find_get_page(inode->i_mapping,
187 start >> PAGE_CACHE_SHIFT);
188 btrfs_set_file_extent_compression(leaf, ei, 0);
189 kaddr = kmap_atomic(page);
190 offset = start & (PAGE_CACHE_SIZE - 1);
191 write_extent_buffer(leaf, kaddr + offset, ptr, size);
192 kunmap_atomic(kaddr);
193 page_cache_release(page);
195 btrfs_mark_buffer_dirty(leaf);
196 btrfs_free_path(path);
199 * we're an inline extent, so nobody can
200 * extend the file past i_size without locking
201 * a page we already have locked.
203 * We must do any isize and inode updates
204 * before we unlock the pages. Otherwise we
205 * could end up racing with unlink.
207 BTRFS_I(inode)->disk_i_size = inode->i_size;
208 ret = btrfs_update_inode(trans, root, inode);
212 btrfs_free_path(path);
218 * conditionally insert an inline extent into the file. This
219 * does the checks required to make sure the data is small enough
220 * to fit as an inline extent.
222 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
223 struct btrfs_root *root,
224 struct inode *inode, u64 start, u64 end,
225 size_t compressed_size, int compress_type,
226 struct page **compressed_pages)
228 u64 isize = i_size_read(inode);
229 u64 actual_end = min(end + 1, isize);
230 u64 inline_len = actual_end - start;
231 u64 aligned_end = (end + root->sectorsize - 1) &
232 ~((u64)root->sectorsize - 1);
234 u64 data_len = inline_len;
238 data_len = compressed_size;
241 actual_end >= PAGE_CACHE_SIZE ||
242 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
244 (actual_end & (root->sectorsize - 1)) == 0) ||
246 data_len > root->fs_info->max_inline) {
250 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
255 if (isize > actual_end)
256 inline_len = min_t(u64, isize, actual_end);
257 ret = insert_inline_extent(trans, root, inode, start,
258 inline_len, compressed_size,
259 compress_type, compressed_pages);
261 btrfs_abort_transaction(trans, root, ret);
264 btrfs_delalloc_release_metadata(inode, end + 1 - start);
265 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
269 struct async_extent {
274 unsigned long nr_pages;
276 struct list_head list;
281 struct btrfs_root *root;
282 struct page *locked_page;
285 struct list_head extents;
286 struct btrfs_work work;
289 static noinline int add_async_extent(struct async_cow *cow,
290 u64 start, u64 ram_size,
293 unsigned long nr_pages,
296 struct async_extent *async_extent;
298 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
299 BUG_ON(!async_extent); /* -ENOMEM */
300 async_extent->start = start;
301 async_extent->ram_size = ram_size;
302 async_extent->compressed_size = compressed_size;
303 async_extent->pages = pages;
304 async_extent->nr_pages = nr_pages;
305 async_extent->compress_type = compress_type;
306 list_add_tail(&async_extent->list, &cow->extents);
311 * we create compressed extents in two phases. The first
312 * phase compresses a range of pages that have already been
313 * locked (both pages and state bits are locked).
315 * This is done inside an ordered work queue, and the compression
316 * is spread across many cpus. The actual IO submission is step
317 * two, and the ordered work queue takes care of making sure that
318 * happens in the same order things were put onto the queue by
319 * writepages and friends.
321 * If this code finds it can't get good compression, it puts an
322 * entry onto the work queue to write the uncompressed bytes. This
323 * makes sure that both compressed inodes and uncompressed inodes
324 * are written in the same order that pdflush sent them down.
326 static noinline int compress_file_range(struct inode *inode,
327 struct page *locked_page,
329 struct async_cow *async_cow,
332 struct btrfs_root *root = BTRFS_I(inode)->root;
333 struct btrfs_trans_handle *trans;
335 u64 blocksize = root->sectorsize;
337 u64 isize = i_size_read(inode);
339 struct page **pages = NULL;
340 unsigned long nr_pages;
341 unsigned long nr_pages_ret = 0;
342 unsigned long total_compressed = 0;
343 unsigned long total_in = 0;
344 unsigned long max_compressed = 128 * 1024;
345 unsigned long max_uncompressed = 128 * 1024;
348 int compress_type = root->fs_info->compress_type;
350 /* if this is a small write inside eof, kick off a defrag */
351 if ((end - start + 1) < 16 * 1024 &&
352 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
353 btrfs_add_inode_defrag(NULL, inode);
355 actual_end = min_t(u64, isize, end + 1);
358 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
359 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
362 * we don't want to send crud past the end of i_size through
363 * compression, that's just a waste of CPU time. So, if the
364 * end of the file is before the start of our current
365 * requested range of bytes, we bail out to the uncompressed
366 * cleanup code that can deal with all of this.
368 * It isn't really the fastest way to fix things, but this is a
369 * very uncommon corner.
371 if (actual_end <= start)
372 goto cleanup_and_bail_uncompressed;
374 total_compressed = actual_end - start;
376 /* we want to make sure that amount of ram required to uncompress
377 * an extent is reasonable, so we limit the total size in ram
378 * of a compressed extent to 128k. This is a crucial number
379 * because it also controls how easily we can spread reads across
380 * cpus for decompression.
382 * We also want to make sure the amount of IO required to do
383 * a random read is reasonably small, so we limit the size of
384 * a compressed extent to 128k.
386 total_compressed = min(total_compressed, max_uncompressed);
387 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
388 num_bytes = max(blocksize, num_bytes);
393 * we do compression for mount -o compress and when the
394 * inode has not been flagged as nocompress. This flag can
395 * change at any time if we discover bad compression ratios.
397 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
398 (btrfs_test_opt(root, COMPRESS) ||
399 (BTRFS_I(inode)->force_compress) ||
400 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
402 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
404 /* just bail out to the uncompressed code */
408 if (BTRFS_I(inode)->force_compress)
409 compress_type = BTRFS_I(inode)->force_compress;
411 ret = btrfs_compress_pages(compress_type,
412 inode->i_mapping, start,
413 total_compressed, pages,
414 nr_pages, &nr_pages_ret,
420 unsigned long offset = total_compressed &
421 (PAGE_CACHE_SIZE - 1);
422 struct page *page = pages[nr_pages_ret - 1];
425 /* zero the tail end of the last page, we might be
426 * sending it down to disk
429 kaddr = kmap_atomic(page);
430 memset(kaddr + offset, 0,
431 PAGE_CACHE_SIZE - offset);
432 kunmap_atomic(kaddr);
439 trans = btrfs_join_transaction(root);
441 ret = PTR_ERR(trans);
443 goto cleanup_and_out;
445 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
447 /* lets try to make an inline extent */
448 if (ret || total_in < (actual_end - start)) {
449 /* we didn't compress the entire range, try
450 * to make an uncompressed inline extent.
452 ret = cow_file_range_inline(trans, root, inode,
453 start, end, 0, 0, NULL);
455 /* try making a compressed inline extent */
456 ret = cow_file_range_inline(trans, root, inode,
459 compress_type, pages);
463 * inline extent creation worked or returned error,
464 * we don't need to create any more async work items.
465 * Unlock and free up our temp pages.
467 extent_clear_unlock_delalloc(inode,
468 &BTRFS_I(inode)->io_tree,
470 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
471 EXTENT_CLEAR_DELALLOC |
472 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
474 btrfs_end_transaction(trans, root);
477 btrfs_end_transaction(trans, root);
482 * we aren't doing an inline extent round the compressed size
483 * up to a block size boundary so the allocator does sane
486 total_compressed = (total_compressed + blocksize - 1) &
490 * one last check to make sure the compression is really a
491 * win, compare the page count read with the blocks on disk
493 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
494 ~(PAGE_CACHE_SIZE - 1);
495 if (total_compressed >= total_in) {
498 num_bytes = total_in;
501 if (!will_compress && pages) {
503 * the compression code ran but failed to make things smaller,
504 * free any pages it allocated and our page pointer array
506 for (i = 0; i < nr_pages_ret; i++) {
507 WARN_ON(pages[i]->mapping);
508 page_cache_release(pages[i]);
512 total_compressed = 0;
515 /* flag the file so we don't compress in the future */
516 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
517 !(BTRFS_I(inode)->force_compress)) {
518 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
524 /* the async work queues will take care of doing actual
525 * allocation on disk for these compressed pages,
526 * and will submit them to the elevator.
528 add_async_extent(async_cow, start, num_bytes,
529 total_compressed, pages, nr_pages_ret,
532 if (start + num_bytes < end) {
539 cleanup_and_bail_uncompressed:
541 * No compression, but we still need to write the pages in
542 * the file we've been given so far. redirty the locked
543 * page if it corresponds to our extent and set things up
544 * for the async work queue to run cow_file_range to do
545 * the normal delalloc dance
547 if (page_offset(locked_page) >= start &&
548 page_offset(locked_page) <= end) {
549 __set_page_dirty_nobuffers(locked_page);
550 /* unlocked later on in the async handlers */
552 add_async_extent(async_cow, start, end - start + 1,
553 0, NULL, 0, BTRFS_COMPRESS_NONE);
561 for (i = 0; i < nr_pages_ret; i++) {
562 WARN_ON(pages[i]->mapping);
563 page_cache_release(pages[i]);
570 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
572 EXTENT_CLEAR_UNLOCK_PAGE |
574 EXTENT_CLEAR_DELALLOC |
575 EXTENT_SET_WRITEBACK |
576 EXTENT_END_WRITEBACK);
577 if (!trans || IS_ERR(trans))
578 btrfs_error(root->fs_info, ret, "Failed to join transaction");
580 btrfs_abort_transaction(trans, root, ret);
585 * phase two of compressed writeback. This is the ordered portion
586 * of the code, which only gets called in the order the work was
587 * queued. We walk all the async extents created by compress_file_range
588 * and send them down to the disk.
590 static noinline int submit_compressed_extents(struct inode *inode,
591 struct async_cow *async_cow)
593 struct async_extent *async_extent;
595 struct btrfs_trans_handle *trans;
596 struct btrfs_key ins;
597 struct extent_map *em;
598 struct btrfs_root *root = BTRFS_I(inode)->root;
599 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
600 struct extent_io_tree *io_tree;
603 if (list_empty(&async_cow->extents))
607 while (!list_empty(&async_cow->extents)) {
608 async_extent = list_entry(async_cow->extents.next,
609 struct async_extent, list);
610 list_del(&async_extent->list);
612 io_tree = &BTRFS_I(inode)->io_tree;
615 /* did the compression code fall back to uncompressed IO? */
616 if (!async_extent->pages) {
617 int page_started = 0;
618 unsigned long nr_written = 0;
620 lock_extent(io_tree, async_extent->start,
621 async_extent->start +
622 async_extent->ram_size - 1);
624 /* allocate blocks */
625 ret = cow_file_range(inode, async_cow->locked_page,
627 async_extent->start +
628 async_extent->ram_size - 1,
629 &page_started, &nr_written, 0);
634 * if page_started, cow_file_range inserted an
635 * inline extent and took care of all the unlocking
636 * and IO for us. Otherwise, we need to submit
637 * all those pages down to the drive.
639 if (!page_started && !ret)
640 extent_write_locked_range(io_tree,
641 inode, async_extent->start,
642 async_extent->start +
643 async_extent->ram_size - 1,
651 lock_extent(io_tree, async_extent->start,
652 async_extent->start + async_extent->ram_size - 1);
654 trans = btrfs_join_transaction(root);
656 ret = PTR_ERR(trans);
658 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
659 ret = btrfs_reserve_extent(trans, root,
660 async_extent->compressed_size,
661 async_extent->compressed_size,
662 0, alloc_hint, &ins, 1);
664 btrfs_abort_transaction(trans, root, ret);
665 btrfs_end_transaction(trans, root);
670 for (i = 0; i < async_extent->nr_pages; i++) {
671 WARN_ON(async_extent->pages[i]->mapping);
672 page_cache_release(async_extent->pages[i]);
674 kfree(async_extent->pages);
675 async_extent->nr_pages = 0;
676 async_extent->pages = NULL;
677 unlock_extent(io_tree, async_extent->start,
678 async_extent->start +
679 async_extent->ram_size - 1);
682 goto out_free; /* JDM: Requeue? */
686 * here we're doing allocation and writeback of the
689 btrfs_drop_extent_cache(inode, async_extent->start,
690 async_extent->start +
691 async_extent->ram_size - 1, 0);
693 em = alloc_extent_map();
694 BUG_ON(!em); /* -ENOMEM */
695 em->start = async_extent->start;
696 em->len = async_extent->ram_size;
697 em->orig_start = em->start;
699 em->block_start = ins.objectid;
700 em->block_len = ins.offset;
701 em->bdev = root->fs_info->fs_devices->latest_bdev;
702 em->compress_type = async_extent->compress_type;
703 set_bit(EXTENT_FLAG_PINNED, &em->flags);
704 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
707 write_lock(&em_tree->lock);
708 ret = add_extent_mapping(em_tree, em);
709 write_unlock(&em_tree->lock);
710 if (ret != -EEXIST) {
714 btrfs_drop_extent_cache(inode, async_extent->start,
715 async_extent->start +
716 async_extent->ram_size - 1, 0);
719 ret = btrfs_add_ordered_extent_compress(inode,
722 async_extent->ram_size,
724 BTRFS_ORDERED_COMPRESSED,
725 async_extent->compress_type);
726 BUG_ON(ret); /* -ENOMEM */
729 * clear dirty, set writeback and unlock the pages.
731 extent_clear_unlock_delalloc(inode,
732 &BTRFS_I(inode)->io_tree,
734 async_extent->start +
735 async_extent->ram_size - 1,
736 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
737 EXTENT_CLEAR_UNLOCK |
738 EXTENT_CLEAR_DELALLOC |
739 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
741 ret = btrfs_submit_compressed_write(inode,
743 async_extent->ram_size,
745 ins.offset, async_extent->pages,
746 async_extent->nr_pages);
748 BUG_ON(ret); /* -ENOMEM */
749 alloc_hint = ins.objectid + ins.offset;
761 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
764 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
765 struct extent_map *em;
768 read_lock(&em_tree->lock);
769 em = search_extent_mapping(em_tree, start, num_bytes);
772 * if block start isn't an actual block number then find the
773 * first block in this inode and use that as a hint. If that
774 * block is also bogus then just don't worry about it.
776 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
778 em = search_extent_mapping(em_tree, 0, 0);
779 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
780 alloc_hint = em->block_start;
784 alloc_hint = em->block_start;
788 read_unlock(&em_tree->lock);
794 * when extent_io.c finds a delayed allocation range in the file,
795 * the call backs end up in this code. The basic idea is to
796 * allocate extents on disk for the range, and create ordered data structs
797 * in ram to track those extents.
799 * locked_page is the page that writepage had locked already. We use
800 * it to make sure we don't do extra locks or unlocks.
802 * *page_started is set to one if we unlock locked_page and do everything
803 * required to start IO on it. It may be clean and already done with
806 static noinline int cow_file_range(struct inode *inode,
807 struct page *locked_page,
808 u64 start, u64 end, int *page_started,
809 unsigned long *nr_written,
812 struct btrfs_root *root = BTRFS_I(inode)->root;
813 struct btrfs_trans_handle *trans;
816 unsigned long ram_size;
819 u64 blocksize = root->sectorsize;
820 struct btrfs_key ins;
821 struct extent_map *em;
822 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
825 BUG_ON(btrfs_is_free_space_inode(root, inode));
826 trans = btrfs_join_transaction(root);
828 extent_clear_unlock_delalloc(inode,
829 &BTRFS_I(inode)->io_tree,
831 EXTENT_CLEAR_UNLOCK_PAGE |
832 EXTENT_CLEAR_UNLOCK |
833 EXTENT_CLEAR_DELALLOC |
835 EXTENT_SET_WRITEBACK |
836 EXTENT_END_WRITEBACK);
837 return PTR_ERR(trans);
839 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
841 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
842 num_bytes = max(blocksize, num_bytes);
843 disk_num_bytes = num_bytes;
846 /* if this is a small write inside eof, kick off defrag */
847 if (num_bytes < 64 * 1024 &&
848 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
849 btrfs_add_inode_defrag(trans, inode);
852 /* lets try to make an inline extent */
853 ret = cow_file_range_inline(trans, root, inode,
854 start, end, 0, 0, NULL);
856 extent_clear_unlock_delalloc(inode,
857 &BTRFS_I(inode)->io_tree,
859 EXTENT_CLEAR_UNLOCK_PAGE |
860 EXTENT_CLEAR_UNLOCK |
861 EXTENT_CLEAR_DELALLOC |
863 EXTENT_SET_WRITEBACK |
864 EXTENT_END_WRITEBACK);
866 *nr_written = *nr_written +
867 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
870 } else if (ret < 0) {
871 btrfs_abort_transaction(trans, root, ret);
876 BUG_ON(disk_num_bytes >
877 btrfs_super_total_bytes(root->fs_info->super_copy));
879 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
880 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
882 while (disk_num_bytes > 0) {
885 cur_alloc_size = disk_num_bytes;
886 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
887 root->sectorsize, 0, alloc_hint,
890 btrfs_abort_transaction(trans, root, ret);
894 em = alloc_extent_map();
895 BUG_ON(!em); /* -ENOMEM */
897 em->orig_start = em->start;
898 ram_size = ins.offset;
899 em->len = ins.offset;
901 em->block_start = ins.objectid;
902 em->block_len = ins.offset;
903 em->bdev = root->fs_info->fs_devices->latest_bdev;
904 set_bit(EXTENT_FLAG_PINNED, &em->flags);
907 write_lock(&em_tree->lock);
908 ret = add_extent_mapping(em_tree, em);
909 write_unlock(&em_tree->lock);
910 if (ret != -EEXIST) {
914 btrfs_drop_extent_cache(inode, start,
915 start + ram_size - 1, 0);
918 cur_alloc_size = ins.offset;
919 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
920 ram_size, cur_alloc_size, 0);
921 BUG_ON(ret); /* -ENOMEM */
923 if (root->root_key.objectid ==
924 BTRFS_DATA_RELOC_TREE_OBJECTID) {
925 ret = btrfs_reloc_clone_csums(inode, start,
928 btrfs_abort_transaction(trans, root, ret);
933 if (disk_num_bytes < cur_alloc_size)
936 /* we're not doing compressed IO, don't unlock the first
937 * page (which the caller expects to stay locked), don't
938 * clear any dirty bits and don't set any writeback bits
940 * Do set the Private2 bit so we know this page was properly
941 * setup for writepage
943 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
944 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
947 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
948 start, start + ram_size - 1,
950 disk_num_bytes -= cur_alloc_size;
951 num_bytes -= cur_alloc_size;
952 alloc_hint = ins.objectid + ins.offset;
953 start += cur_alloc_size;
957 btrfs_end_transaction(trans, root);
961 extent_clear_unlock_delalloc(inode,
962 &BTRFS_I(inode)->io_tree,
964 EXTENT_CLEAR_UNLOCK_PAGE |
965 EXTENT_CLEAR_UNLOCK |
966 EXTENT_CLEAR_DELALLOC |
968 EXTENT_SET_WRITEBACK |
969 EXTENT_END_WRITEBACK);
975 * work queue call back to started compression on a file and pages
977 static noinline void async_cow_start(struct btrfs_work *work)
979 struct async_cow *async_cow;
981 async_cow = container_of(work, struct async_cow, work);
983 compress_file_range(async_cow->inode, async_cow->locked_page,
984 async_cow->start, async_cow->end, async_cow,
987 async_cow->inode = NULL;
991 * work queue call back to submit previously compressed pages
993 static noinline void async_cow_submit(struct btrfs_work *work)
995 struct async_cow *async_cow;
996 struct btrfs_root *root;
997 unsigned long nr_pages;
999 async_cow = container_of(work, struct async_cow, work);
1001 root = async_cow->root;
1002 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1005 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
1007 if (atomic_read(&root->fs_info->async_delalloc_pages) <
1009 waitqueue_active(&root->fs_info->async_submit_wait))
1010 wake_up(&root->fs_info->async_submit_wait);
1012 if (async_cow->inode)
1013 submit_compressed_extents(async_cow->inode, async_cow);
1016 static noinline void async_cow_free(struct btrfs_work *work)
1018 struct async_cow *async_cow;
1019 async_cow = container_of(work, struct async_cow, work);
1023 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1024 u64 start, u64 end, int *page_started,
1025 unsigned long *nr_written)
1027 struct async_cow *async_cow;
1028 struct btrfs_root *root = BTRFS_I(inode)->root;
1029 unsigned long nr_pages;
1031 int limit = 10 * 1024 * 1042;
1033 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1034 1, 0, NULL, GFP_NOFS);
1035 while (start < end) {
1036 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1037 BUG_ON(!async_cow); /* -ENOMEM */
1038 async_cow->inode = inode;
1039 async_cow->root = root;
1040 async_cow->locked_page = locked_page;
1041 async_cow->start = start;
1043 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1046 cur_end = min(end, start + 512 * 1024 - 1);
1048 async_cow->end = cur_end;
1049 INIT_LIST_HEAD(&async_cow->extents);
1051 async_cow->work.func = async_cow_start;
1052 async_cow->work.ordered_func = async_cow_submit;
1053 async_cow->work.ordered_free = async_cow_free;
1054 async_cow->work.flags = 0;
1056 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1058 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1060 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1063 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1064 wait_event(root->fs_info->async_submit_wait,
1065 (atomic_read(&root->fs_info->async_delalloc_pages) <
1069 while (atomic_read(&root->fs_info->async_submit_draining) &&
1070 atomic_read(&root->fs_info->async_delalloc_pages)) {
1071 wait_event(root->fs_info->async_submit_wait,
1072 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1076 *nr_written += nr_pages;
1077 start = cur_end + 1;
1083 static noinline int csum_exist_in_range(struct btrfs_root *root,
1084 u64 bytenr, u64 num_bytes)
1087 struct btrfs_ordered_sum *sums;
1090 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1091 bytenr + num_bytes - 1, &list, 0);
1092 if (ret == 0 && list_empty(&list))
1095 while (!list_empty(&list)) {
1096 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1097 list_del(&sums->list);
1104 * when nowcow writeback call back. This checks for snapshots or COW copies
1105 * of the extents that exist in the file, and COWs the file as required.
1107 * If no cow copies or snapshots exist, we write directly to the existing
1110 static noinline int run_delalloc_nocow(struct inode *inode,
1111 struct page *locked_page,
1112 u64 start, u64 end, int *page_started, int force,
1113 unsigned long *nr_written)
1115 struct btrfs_root *root = BTRFS_I(inode)->root;
1116 struct btrfs_trans_handle *trans;
1117 struct extent_buffer *leaf;
1118 struct btrfs_path *path;
1119 struct btrfs_file_extent_item *fi;
1120 struct btrfs_key found_key;
1133 u64 ino = btrfs_ino(inode);
1135 path = btrfs_alloc_path();
1139 nolock = btrfs_is_free_space_inode(root, inode);
1142 trans = btrfs_join_transaction_nolock(root);
1144 trans = btrfs_join_transaction(root);
1146 if (IS_ERR(trans)) {
1147 btrfs_free_path(path);
1148 return PTR_ERR(trans);
1151 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1153 cow_start = (u64)-1;
1156 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1159 btrfs_abort_transaction(trans, root, ret);
1162 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1163 leaf = path->nodes[0];
1164 btrfs_item_key_to_cpu(leaf, &found_key,
1165 path->slots[0] - 1);
1166 if (found_key.objectid == ino &&
1167 found_key.type == BTRFS_EXTENT_DATA_KEY)
1172 leaf = path->nodes[0];
1173 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1174 ret = btrfs_next_leaf(root, path);
1176 btrfs_abort_transaction(trans, root, ret);
1181 leaf = path->nodes[0];
1187 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1189 if (found_key.objectid > ino ||
1190 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1191 found_key.offset > end)
1194 if (found_key.offset > cur_offset) {
1195 extent_end = found_key.offset;
1200 fi = btrfs_item_ptr(leaf, path->slots[0],
1201 struct btrfs_file_extent_item);
1202 extent_type = btrfs_file_extent_type(leaf, fi);
1204 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1205 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1206 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1207 extent_offset = btrfs_file_extent_offset(leaf, fi);
1208 extent_end = found_key.offset +
1209 btrfs_file_extent_num_bytes(leaf, fi);
1210 if (extent_end <= start) {
1214 if (disk_bytenr == 0)
1216 if (btrfs_file_extent_compression(leaf, fi) ||
1217 btrfs_file_extent_encryption(leaf, fi) ||
1218 btrfs_file_extent_other_encoding(leaf, fi))
1220 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1222 if (btrfs_extent_readonly(root, disk_bytenr))
1224 if (btrfs_cross_ref_exist(trans, root, ino,
1226 extent_offset, disk_bytenr))
1228 disk_bytenr += extent_offset;
1229 disk_bytenr += cur_offset - found_key.offset;
1230 num_bytes = min(end + 1, extent_end) - cur_offset;
1232 * force cow if csum exists in the range.
1233 * this ensure that csum for a given extent are
1234 * either valid or do not exist.
1236 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1239 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1240 extent_end = found_key.offset +
1241 btrfs_file_extent_inline_len(leaf, fi);
1242 extent_end = ALIGN(extent_end, root->sectorsize);
1247 if (extent_end <= start) {
1252 if (cow_start == (u64)-1)
1253 cow_start = cur_offset;
1254 cur_offset = extent_end;
1255 if (cur_offset > end)
1261 btrfs_release_path(path);
1262 if (cow_start != (u64)-1) {
1263 ret = cow_file_range(inode, locked_page, cow_start,
1264 found_key.offset - 1, page_started,
1267 btrfs_abort_transaction(trans, root, ret);
1270 cow_start = (u64)-1;
1273 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1274 struct extent_map *em;
1275 struct extent_map_tree *em_tree;
1276 em_tree = &BTRFS_I(inode)->extent_tree;
1277 em = alloc_extent_map();
1278 BUG_ON(!em); /* -ENOMEM */
1279 em->start = cur_offset;
1280 em->orig_start = em->start;
1281 em->len = num_bytes;
1282 em->block_len = num_bytes;
1283 em->block_start = disk_bytenr;
1284 em->bdev = root->fs_info->fs_devices->latest_bdev;
1285 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1287 write_lock(&em_tree->lock);
1288 ret = add_extent_mapping(em_tree, em);
1289 write_unlock(&em_tree->lock);
1290 if (ret != -EEXIST) {
1291 free_extent_map(em);
1294 btrfs_drop_extent_cache(inode, em->start,
1295 em->start + em->len - 1, 0);
1297 type = BTRFS_ORDERED_PREALLOC;
1299 type = BTRFS_ORDERED_NOCOW;
1302 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1303 num_bytes, num_bytes, type);
1304 BUG_ON(ret); /* -ENOMEM */
1306 if (root->root_key.objectid ==
1307 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1308 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1311 btrfs_abort_transaction(trans, root, ret);
1316 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1317 cur_offset, cur_offset + num_bytes - 1,
1318 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1319 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1320 EXTENT_SET_PRIVATE2);
1321 cur_offset = extent_end;
1322 if (cur_offset > end)
1325 btrfs_release_path(path);
1327 if (cur_offset <= end && cow_start == (u64)-1)
1328 cow_start = cur_offset;
1329 if (cow_start != (u64)-1) {
1330 ret = cow_file_range(inode, locked_page, cow_start, end,
1331 page_started, nr_written, 1);
1333 btrfs_abort_transaction(trans, root, ret);
1340 err = btrfs_end_transaction_nolock(trans, root);
1342 err = btrfs_end_transaction(trans, root);
1347 btrfs_free_path(path);
1352 * extent_io.c call back to do delayed allocation processing
1354 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1355 u64 start, u64 end, int *page_started,
1356 unsigned long *nr_written)
1359 struct btrfs_root *root = BTRFS_I(inode)->root;
1361 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1362 ret = run_delalloc_nocow(inode, locked_page, start, end,
1363 page_started, 1, nr_written);
1364 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1365 ret = run_delalloc_nocow(inode, locked_page, start, end,
1366 page_started, 0, nr_written);
1367 else if (!btrfs_test_opt(root, COMPRESS) &&
1368 !(BTRFS_I(inode)->force_compress) &&
1369 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1370 ret = cow_file_range(inode, locked_page, start, end,
1371 page_started, nr_written, 1);
1373 ret = cow_file_range_async(inode, locked_page, start, end,
1374 page_started, nr_written);
1378 static void btrfs_split_extent_hook(struct inode *inode,
1379 struct extent_state *orig, u64 split)
1381 /* not delalloc, ignore it */
1382 if (!(orig->state & EXTENT_DELALLOC))
1385 spin_lock(&BTRFS_I(inode)->lock);
1386 BTRFS_I(inode)->outstanding_extents++;
1387 spin_unlock(&BTRFS_I(inode)->lock);
1391 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1392 * extents so we can keep track of new extents that are just merged onto old
1393 * extents, such as when we are doing sequential writes, so we can properly
1394 * account for the metadata space we'll need.
1396 static void btrfs_merge_extent_hook(struct inode *inode,
1397 struct extent_state *new,
1398 struct extent_state *other)
1400 /* not delalloc, ignore it */
1401 if (!(other->state & EXTENT_DELALLOC))
1404 spin_lock(&BTRFS_I(inode)->lock);
1405 BTRFS_I(inode)->outstanding_extents--;
1406 spin_unlock(&BTRFS_I(inode)->lock);
1410 * extent_io.c set_bit_hook, used to track delayed allocation
1411 * bytes in this file, and to maintain the list of inodes that
1412 * have pending delalloc work to be done.
1414 static void btrfs_set_bit_hook(struct inode *inode,
1415 struct extent_state *state, int *bits)
1419 * set_bit and clear bit hooks normally require _irqsave/restore
1420 * but in this case, we are only testing for the DELALLOC
1421 * bit, which is only set or cleared with irqs on
1423 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1424 struct btrfs_root *root = BTRFS_I(inode)->root;
1425 u64 len = state->end + 1 - state->start;
1426 bool do_list = !btrfs_is_free_space_inode(root, inode);
1428 if (*bits & EXTENT_FIRST_DELALLOC) {
1429 *bits &= ~EXTENT_FIRST_DELALLOC;
1431 spin_lock(&BTRFS_I(inode)->lock);
1432 BTRFS_I(inode)->outstanding_extents++;
1433 spin_unlock(&BTRFS_I(inode)->lock);
1436 spin_lock(&root->fs_info->delalloc_lock);
1437 BTRFS_I(inode)->delalloc_bytes += len;
1438 root->fs_info->delalloc_bytes += len;
1439 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1440 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1441 &root->fs_info->delalloc_inodes);
1443 spin_unlock(&root->fs_info->delalloc_lock);
1448 * extent_io.c clear_bit_hook, see set_bit_hook for why
1450 static void btrfs_clear_bit_hook(struct inode *inode,
1451 struct extent_state *state, int *bits)
1454 * set_bit and clear bit hooks normally require _irqsave/restore
1455 * but in this case, we are only testing for the DELALLOC
1456 * bit, which is only set or cleared with irqs on
1458 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1459 struct btrfs_root *root = BTRFS_I(inode)->root;
1460 u64 len = state->end + 1 - state->start;
1461 bool do_list = !btrfs_is_free_space_inode(root, inode);
1463 if (*bits & EXTENT_FIRST_DELALLOC) {
1464 *bits &= ~EXTENT_FIRST_DELALLOC;
1465 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1466 spin_lock(&BTRFS_I(inode)->lock);
1467 BTRFS_I(inode)->outstanding_extents--;
1468 spin_unlock(&BTRFS_I(inode)->lock);
1471 if (*bits & EXTENT_DO_ACCOUNTING)
1472 btrfs_delalloc_release_metadata(inode, len);
1474 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1476 btrfs_free_reserved_data_space(inode, len);
1478 spin_lock(&root->fs_info->delalloc_lock);
1479 root->fs_info->delalloc_bytes -= len;
1480 BTRFS_I(inode)->delalloc_bytes -= len;
1482 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1483 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1484 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1486 spin_unlock(&root->fs_info->delalloc_lock);
1491 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1492 * we don't create bios that span stripes or chunks
1494 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1495 size_t size, struct bio *bio,
1496 unsigned long bio_flags)
1498 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1499 struct btrfs_mapping_tree *map_tree;
1500 u64 logical = (u64)bio->bi_sector << 9;
1505 if (bio_flags & EXTENT_BIO_COMPRESSED)
1508 length = bio->bi_size;
1509 map_tree = &root->fs_info->mapping_tree;
1510 map_length = length;
1511 ret = btrfs_map_block(map_tree, READ, logical,
1512 &map_length, NULL, 0);
1513 /* Will always return 0 or 1 with map_multi == NULL */
1515 if (map_length < length + size)
1521 * in order to insert checksums into the metadata in large chunks,
1522 * we wait until bio submission time. All the pages in the bio are
1523 * checksummed and sums are attached onto the ordered extent record.
1525 * At IO completion time the cums attached on the ordered extent record
1526 * are inserted into the btree
1528 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1529 struct bio *bio, int mirror_num,
1530 unsigned long bio_flags,
1533 struct btrfs_root *root = BTRFS_I(inode)->root;
1536 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1537 BUG_ON(ret); /* -ENOMEM */
1542 * in order to insert checksums into the metadata in large chunks,
1543 * we wait until bio submission time. All the pages in the bio are
1544 * checksummed and sums are attached onto the ordered extent record.
1546 * At IO completion time the cums attached on the ordered extent record
1547 * are inserted into the btree
1549 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1550 int mirror_num, unsigned long bio_flags,
1553 struct btrfs_root *root = BTRFS_I(inode)->root;
1554 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1558 * extent_io.c submission hook. This does the right thing for csum calculation
1559 * on write, or reading the csums from the tree before a read
1561 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1562 int mirror_num, unsigned long bio_flags,
1565 struct btrfs_root *root = BTRFS_I(inode)->root;
1570 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1572 if (btrfs_is_free_space_inode(root, inode))
1575 if (!(rw & REQ_WRITE)) {
1576 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1580 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1581 return btrfs_submit_compressed_read(inode, bio,
1582 mirror_num, bio_flags);
1583 } else if (!skip_sum) {
1584 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1589 } else if (!skip_sum) {
1590 /* csum items have already been cloned */
1591 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1593 /* we're doing a write, do the async checksumming */
1594 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1595 inode, rw, bio, mirror_num,
1596 bio_flags, bio_offset,
1597 __btrfs_submit_bio_start,
1598 __btrfs_submit_bio_done);
1602 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1606 * given a list of ordered sums record them in the inode. This happens
1607 * at IO completion time based on sums calculated at bio submission time.
1609 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1610 struct inode *inode, u64 file_offset,
1611 struct list_head *list)
1613 struct btrfs_ordered_sum *sum;
1615 list_for_each_entry(sum, list, list) {
1616 btrfs_csum_file_blocks(trans,
1617 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1622 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1623 struct extent_state **cached_state)
1625 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1627 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1628 cached_state, GFP_NOFS);
1631 /* see btrfs_writepage_start_hook for details on why this is required */
1632 struct btrfs_writepage_fixup {
1634 struct btrfs_work work;
1637 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1639 struct btrfs_writepage_fixup *fixup;
1640 struct btrfs_ordered_extent *ordered;
1641 struct extent_state *cached_state = NULL;
1643 struct inode *inode;
1648 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1652 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1653 ClearPageChecked(page);
1657 inode = page->mapping->host;
1658 page_start = page_offset(page);
1659 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1661 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1664 /* already ordered? We're done */
1665 if (PagePrivate2(page))
1668 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1670 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1671 page_end, &cached_state, GFP_NOFS);
1673 btrfs_start_ordered_extent(inode, ordered, 1);
1674 btrfs_put_ordered_extent(ordered);
1678 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1680 mapping_set_error(page->mapping, ret);
1681 end_extent_writepage(page, ret, page_start, page_end);
1682 ClearPageChecked(page);
1686 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1687 ClearPageChecked(page);
1688 set_page_dirty(page);
1690 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1691 &cached_state, GFP_NOFS);
1694 page_cache_release(page);
1699 * There are a few paths in the higher layers of the kernel that directly
1700 * set the page dirty bit without asking the filesystem if it is a
1701 * good idea. This causes problems because we want to make sure COW
1702 * properly happens and the data=ordered rules are followed.
1704 * In our case any range that doesn't have the ORDERED bit set
1705 * hasn't been properly setup for IO. We kick off an async process
1706 * to fix it up. The async helper will wait for ordered extents, set
1707 * the delalloc bit and make it safe to write the page.
1709 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1711 struct inode *inode = page->mapping->host;
1712 struct btrfs_writepage_fixup *fixup;
1713 struct btrfs_root *root = BTRFS_I(inode)->root;
1715 /* this page is properly in the ordered list */
1716 if (TestClearPagePrivate2(page))
1719 if (PageChecked(page))
1722 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1726 SetPageChecked(page);
1727 page_cache_get(page);
1728 fixup->work.func = btrfs_writepage_fixup_worker;
1730 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1734 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1735 struct inode *inode, u64 file_pos,
1736 u64 disk_bytenr, u64 disk_num_bytes,
1737 u64 num_bytes, u64 ram_bytes,
1738 u8 compression, u8 encryption,
1739 u16 other_encoding, int extent_type)
1741 struct btrfs_root *root = BTRFS_I(inode)->root;
1742 struct btrfs_file_extent_item *fi;
1743 struct btrfs_path *path;
1744 struct extent_buffer *leaf;
1745 struct btrfs_key ins;
1749 path = btrfs_alloc_path();
1753 path->leave_spinning = 1;
1756 * we may be replacing one extent in the tree with another.
1757 * The new extent is pinned in the extent map, and we don't want
1758 * to drop it from the cache until it is completely in the btree.
1760 * So, tell btrfs_drop_extents to leave this extent in the cache.
1761 * the caller is expected to unpin it and allow it to be merged
1764 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1769 ins.objectid = btrfs_ino(inode);
1770 ins.offset = file_pos;
1771 ins.type = BTRFS_EXTENT_DATA_KEY;
1772 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1775 leaf = path->nodes[0];
1776 fi = btrfs_item_ptr(leaf, path->slots[0],
1777 struct btrfs_file_extent_item);
1778 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1779 btrfs_set_file_extent_type(leaf, fi, extent_type);
1780 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1781 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1782 btrfs_set_file_extent_offset(leaf, fi, 0);
1783 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1784 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1785 btrfs_set_file_extent_compression(leaf, fi, compression);
1786 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1787 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1789 btrfs_unlock_up_safe(path, 1);
1790 btrfs_set_lock_blocking(leaf);
1792 btrfs_mark_buffer_dirty(leaf);
1794 inode_add_bytes(inode, num_bytes);
1796 ins.objectid = disk_bytenr;
1797 ins.offset = disk_num_bytes;
1798 ins.type = BTRFS_EXTENT_ITEM_KEY;
1799 ret = btrfs_alloc_reserved_file_extent(trans, root,
1800 root->root_key.objectid,
1801 btrfs_ino(inode), file_pos, &ins);
1803 btrfs_free_path(path);
1809 * helper function for btrfs_finish_ordered_io, this
1810 * just reads in some of the csum leaves to prime them into ram
1811 * before we start the transaction. It limits the amount of btree
1812 * reads required while inside the transaction.
1814 /* as ordered data IO finishes, this gets called so we can finish
1815 * an ordered extent if the range of bytes in the file it covers are
1818 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
1820 struct inode *inode = ordered_extent->inode;
1821 struct btrfs_root *root = BTRFS_I(inode)->root;
1822 struct btrfs_trans_handle *trans = NULL;
1823 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1824 struct extent_state *cached_state = NULL;
1825 int compress_type = 0;
1829 nolock = btrfs_is_free_space_inode(root, inode);
1831 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1836 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1837 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1838 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1841 trans = btrfs_join_transaction_nolock(root);
1843 trans = btrfs_join_transaction(root);
1845 return PTR_ERR(trans);
1846 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1847 ret = btrfs_update_inode_fallback(trans, root, inode);
1848 if (ret) /* -ENOMEM or corruption */
1849 btrfs_abort_transaction(trans, root, ret);
1854 lock_extent_bits(io_tree, ordered_extent->file_offset,
1855 ordered_extent->file_offset + ordered_extent->len - 1,
1859 trans = btrfs_join_transaction_nolock(root);
1861 trans = btrfs_join_transaction(root);
1862 if (IS_ERR(trans)) {
1863 ret = PTR_ERR(trans);
1867 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1869 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1870 compress_type = ordered_extent->compress_type;
1871 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1872 BUG_ON(compress_type);
1873 ret = btrfs_mark_extent_written(trans, inode,
1874 ordered_extent->file_offset,
1875 ordered_extent->file_offset +
1876 ordered_extent->len);
1878 BUG_ON(root == root->fs_info->tree_root);
1879 ret = insert_reserved_file_extent(trans, inode,
1880 ordered_extent->file_offset,
1881 ordered_extent->start,
1882 ordered_extent->disk_len,
1883 ordered_extent->len,
1884 ordered_extent->len,
1885 compress_type, 0, 0,
1886 BTRFS_FILE_EXTENT_REG);
1887 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1888 ordered_extent->file_offset,
1889 ordered_extent->len);
1893 btrfs_abort_transaction(trans, root, ret);
1897 add_pending_csums(trans, inode, ordered_extent->file_offset,
1898 &ordered_extent->list);
1900 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1901 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1902 ret = btrfs_update_inode_fallback(trans, root, inode);
1903 if (ret) { /* -ENOMEM or corruption */
1904 btrfs_abort_transaction(trans, root, ret);
1910 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1911 ordered_extent->file_offset +
1912 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1914 if (root != root->fs_info->tree_root)
1915 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1918 btrfs_end_transaction_nolock(trans, root);
1920 btrfs_end_transaction(trans, root);
1924 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
1925 ordered_extent->file_offset +
1926 ordered_extent->len - 1, NULL, GFP_NOFS);
1929 * This needs to be dont to make sure anybody waiting knows we are done
1930 * upating everything for this ordered extent.
1932 btrfs_remove_ordered_extent(inode, ordered_extent);
1935 btrfs_put_ordered_extent(ordered_extent);
1936 /* once for the tree */
1937 btrfs_put_ordered_extent(ordered_extent);
1942 static void finish_ordered_fn(struct btrfs_work *work)
1944 struct btrfs_ordered_extent *ordered_extent;
1945 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
1946 btrfs_finish_ordered_io(ordered_extent);
1949 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1950 struct extent_state *state, int uptodate)
1952 struct inode *inode = page->mapping->host;
1953 struct btrfs_root *root = BTRFS_I(inode)->root;
1954 struct btrfs_ordered_extent *ordered_extent = NULL;
1955 struct btrfs_workers *workers;
1957 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1959 ClearPagePrivate2(page);
1960 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1961 end - start + 1, uptodate))
1964 ordered_extent->work.func = finish_ordered_fn;
1965 ordered_extent->work.flags = 0;
1967 if (btrfs_is_free_space_inode(root, inode))
1968 workers = &root->fs_info->endio_freespace_worker;
1970 workers = &root->fs_info->endio_write_workers;
1971 btrfs_queue_worker(workers, &ordered_extent->work);
1977 * when reads are done, we need to check csums to verify the data is correct
1978 * if there's a match, we allow the bio to finish. If not, the code in
1979 * extent_io.c will try to find good copies for us.
1981 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1982 struct extent_state *state, int mirror)
1984 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1985 struct inode *inode = page->mapping->host;
1986 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1988 u64 private = ~(u32)0;
1990 struct btrfs_root *root = BTRFS_I(inode)->root;
1993 if (PageChecked(page)) {
1994 ClearPageChecked(page);
1998 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2001 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2002 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2003 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2008 if (state && state->start == start) {
2009 private = state->private;
2012 ret = get_state_private(io_tree, start, &private);
2014 kaddr = kmap_atomic(page);
2018 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
2019 btrfs_csum_final(csum, (char *)&csum);
2020 if (csum != private)
2023 kunmap_atomic(kaddr);
2028 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2030 (unsigned long long)btrfs_ino(page->mapping->host),
2031 (unsigned long long)start, csum,
2032 (unsigned long long)private);
2033 memset(kaddr + offset, 1, end - start + 1);
2034 flush_dcache_page(page);
2035 kunmap_atomic(kaddr);
2041 struct delayed_iput {
2042 struct list_head list;
2043 struct inode *inode;
2046 /* JDM: If this is fs-wide, why can't we add a pointer to
2047 * btrfs_inode instead and avoid the allocation? */
2048 void btrfs_add_delayed_iput(struct inode *inode)
2050 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2051 struct delayed_iput *delayed;
2053 if (atomic_add_unless(&inode->i_count, -1, 1))
2056 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2057 delayed->inode = inode;
2059 spin_lock(&fs_info->delayed_iput_lock);
2060 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2061 spin_unlock(&fs_info->delayed_iput_lock);
2064 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2067 struct btrfs_fs_info *fs_info = root->fs_info;
2068 struct delayed_iput *delayed;
2071 spin_lock(&fs_info->delayed_iput_lock);
2072 empty = list_empty(&fs_info->delayed_iputs);
2073 spin_unlock(&fs_info->delayed_iput_lock);
2077 down_read(&root->fs_info->cleanup_work_sem);
2078 spin_lock(&fs_info->delayed_iput_lock);
2079 list_splice_init(&fs_info->delayed_iputs, &list);
2080 spin_unlock(&fs_info->delayed_iput_lock);
2082 while (!list_empty(&list)) {
2083 delayed = list_entry(list.next, struct delayed_iput, list);
2084 list_del(&delayed->list);
2085 iput(delayed->inode);
2088 up_read(&root->fs_info->cleanup_work_sem);
2091 enum btrfs_orphan_cleanup_state {
2092 ORPHAN_CLEANUP_STARTED = 1,
2093 ORPHAN_CLEANUP_DONE = 2,
2097 * This is called in transaction commit time. If there are no orphan
2098 * files in the subvolume, it removes orphan item and frees block_rsv
2101 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2102 struct btrfs_root *root)
2104 struct btrfs_block_rsv *block_rsv;
2107 if (!list_empty(&root->orphan_list) ||
2108 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2111 spin_lock(&root->orphan_lock);
2112 if (!list_empty(&root->orphan_list)) {
2113 spin_unlock(&root->orphan_lock);
2117 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2118 spin_unlock(&root->orphan_lock);
2122 block_rsv = root->orphan_block_rsv;
2123 root->orphan_block_rsv = NULL;
2124 spin_unlock(&root->orphan_lock);
2126 if (root->orphan_item_inserted &&
2127 btrfs_root_refs(&root->root_item) > 0) {
2128 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2129 root->root_key.objectid);
2131 root->orphan_item_inserted = 0;
2135 WARN_ON(block_rsv->size > 0);
2136 btrfs_free_block_rsv(root, block_rsv);
2141 * This creates an orphan entry for the given inode in case something goes
2142 * wrong in the middle of an unlink/truncate.
2144 * NOTE: caller of this function should reserve 5 units of metadata for
2147 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2149 struct btrfs_root *root = BTRFS_I(inode)->root;
2150 struct btrfs_block_rsv *block_rsv = NULL;
2155 if (!root->orphan_block_rsv) {
2156 block_rsv = btrfs_alloc_block_rsv(root);
2161 spin_lock(&root->orphan_lock);
2162 if (!root->orphan_block_rsv) {
2163 root->orphan_block_rsv = block_rsv;
2164 } else if (block_rsv) {
2165 btrfs_free_block_rsv(root, block_rsv);
2169 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2170 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2173 * For proper ENOSPC handling, we should do orphan
2174 * cleanup when mounting. But this introduces backward
2175 * compatibility issue.
2177 if (!xchg(&root->orphan_item_inserted, 1))
2185 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2186 &BTRFS_I(inode)->runtime_flags))
2188 spin_unlock(&root->orphan_lock);
2190 /* grab metadata reservation from transaction handle */
2192 ret = btrfs_orphan_reserve_metadata(trans, inode);
2193 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2196 /* insert an orphan item to track this unlinked/truncated file */
2198 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2199 if (ret && ret != -EEXIST) {
2200 btrfs_abort_transaction(trans, root, ret);
2206 /* insert an orphan item to track subvolume contains orphan files */
2208 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2209 root->root_key.objectid);
2210 if (ret && ret != -EEXIST) {
2211 btrfs_abort_transaction(trans, root, ret);
2219 * We have done the truncate/delete so we can go ahead and remove the orphan
2220 * item for this particular inode.
2222 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2224 struct btrfs_root *root = BTRFS_I(inode)->root;
2225 int delete_item = 0;
2226 int release_rsv = 0;
2229 spin_lock(&root->orphan_lock);
2230 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2231 list_del_init(&BTRFS_I(inode)->i_orphan);
2235 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2236 &BTRFS_I(inode)->runtime_flags))
2238 spin_unlock(&root->orphan_lock);
2240 if (trans && delete_item) {
2241 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2242 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2246 btrfs_orphan_release_metadata(inode);
2252 * this cleans up any orphans that may be left on the list from the last use
2255 int btrfs_orphan_cleanup(struct btrfs_root *root)
2257 struct btrfs_path *path;
2258 struct extent_buffer *leaf;
2259 struct btrfs_key key, found_key;
2260 struct btrfs_trans_handle *trans;
2261 struct inode *inode;
2262 u64 last_objectid = 0;
2263 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2265 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2268 path = btrfs_alloc_path();
2275 key.objectid = BTRFS_ORPHAN_OBJECTID;
2276 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2277 key.offset = (u64)-1;
2280 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2285 * if ret == 0 means we found what we were searching for, which
2286 * is weird, but possible, so only screw with path if we didn't
2287 * find the key and see if we have stuff that matches
2291 if (path->slots[0] == 0)
2296 /* pull out the item */
2297 leaf = path->nodes[0];
2298 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2300 /* make sure the item matches what we want */
2301 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2303 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2306 /* release the path since we're done with it */
2307 btrfs_release_path(path);
2310 * this is where we are basically btrfs_lookup, without the
2311 * crossing root thing. we store the inode number in the
2312 * offset of the orphan item.
2315 if (found_key.offset == last_objectid) {
2316 printk(KERN_ERR "btrfs: Error removing orphan entry, "
2317 "stopping orphan cleanup\n");
2322 last_objectid = found_key.offset;
2324 found_key.objectid = found_key.offset;
2325 found_key.type = BTRFS_INODE_ITEM_KEY;
2326 found_key.offset = 0;
2327 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2328 ret = PTR_RET(inode);
2329 if (ret && ret != -ESTALE)
2332 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2333 struct btrfs_root *dead_root;
2334 struct btrfs_fs_info *fs_info = root->fs_info;
2335 int is_dead_root = 0;
2338 * this is an orphan in the tree root. Currently these
2339 * could come from 2 sources:
2340 * a) a snapshot deletion in progress
2341 * b) a free space cache inode
2342 * We need to distinguish those two, as the snapshot
2343 * orphan must not get deleted.
2344 * find_dead_roots already ran before us, so if this
2345 * is a snapshot deletion, we should find the root
2346 * in the dead_roots list
2348 spin_lock(&fs_info->trans_lock);
2349 list_for_each_entry(dead_root, &fs_info->dead_roots,
2351 if (dead_root->root_key.objectid ==
2352 found_key.objectid) {
2357 spin_unlock(&fs_info->trans_lock);
2359 /* prevent this orphan from being found again */
2360 key.offset = found_key.objectid - 1;
2365 * Inode is already gone but the orphan item is still there,
2366 * kill the orphan item.
2368 if (ret == -ESTALE) {
2369 trans = btrfs_start_transaction(root, 1);
2370 if (IS_ERR(trans)) {
2371 ret = PTR_ERR(trans);
2374 ret = btrfs_del_orphan_item(trans, root,
2375 found_key.objectid);
2376 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2377 btrfs_end_transaction(trans, root);
2382 * add this inode to the orphan list so btrfs_orphan_del does
2383 * the proper thing when we hit it
2385 spin_lock(&root->orphan_lock);
2386 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2387 spin_unlock(&root->orphan_lock);
2389 /* if we have links, this was a truncate, lets do that */
2390 if (inode->i_nlink) {
2391 if (!S_ISREG(inode->i_mode)) {
2397 ret = btrfs_truncate(inode);
2402 /* this will do delete_inode and everything for us */
2407 /* release the path since we're done with it */
2408 btrfs_release_path(path);
2410 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2412 if (root->orphan_block_rsv)
2413 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2416 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2417 trans = btrfs_join_transaction(root);
2419 btrfs_end_transaction(trans, root);
2423 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2425 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2429 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2430 btrfs_free_path(path);
2435 * very simple check to peek ahead in the leaf looking for xattrs. If we
2436 * don't find any xattrs, we know there can't be any acls.
2438 * slot is the slot the inode is in, objectid is the objectid of the inode
2440 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2441 int slot, u64 objectid)
2443 u32 nritems = btrfs_header_nritems(leaf);
2444 struct btrfs_key found_key;
2448 while (slot < nritems) {
2449 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2451 /* we found a different objectid, there must not be acls */
2452 if (found_key.objectid != objectid)
2455 /* we found an xattr, assume we've got an acl */
2456 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2460 * we found a key greater than an xattr key, there can't
2461 * be any acls later on
2463 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2470 * it goes inode, inode backrefs, xattrs, extents,
2471 * so if there are a ton of hard links to an inode there can
2472 * be a lot of backrefs. Don't waste time searching too hard,
2473 * this is just an optimization
2478 /* we hit the end of the leaf before we found an xattr or
2479 * something larger than an xattr. We have to assume the inode
2486 * read an inode from the btree into the in-memory inode
2488 static void btrfs_read_locked_inode(struct inode *inode)
2490 struct btrfs_path *path;
2491 struct extent_buffer *leaf;
2492 struct btrfs_inode_item *inode_item;
2493 struct btrfs_timespec *tspec;
2494 struct btrfs_root *root = BTRFS_I(inode)->root;
2495 struct btrfs_key location;
2499 bool filled = false;
2501 ret = btrfs_fill_inode(inode, &rdev);
2505 path = btrfs_alloc_path();
2509 path->leave_spinning = 1;
2510 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2512 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2516 leaf = path->nodes[0];
2521 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2522 struct btrfs_inode_item);
2523 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2524 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2525 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2526 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2527 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2529 tspec = btrfs_inode_atime(inode_item);
2530 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2531 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2533 tspec = btrfs_inode_mtime(inode_item);
2534 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2535 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2537 tspec = btrfs_inode_ctime(inode_item);
2538 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2539 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2541 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2542 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2543 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2544 inode->i_generation = BTRFS_I(inode)->generation;
2546 rdev = btrfs_inode_rdev(leaf, inode_item);
2548 BTRFS_I(inode)->index_cnt = (u64)-1;
2549 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2552 * try to precache a NULL acl entry for files that don't have
2553 * any xattrs or acls
2555 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2558 cache_no_acl(inode);
2560 btrfs_free_path(path);
2562 switch (inode->i_mode & S_IFMT) {
2564 inode->i_mapping->a_ops = &btrfs_aops;
2565 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2566 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2567 inode->i_fop = &btrfs_file_operations;
2568 inode->i_op = &btrfs_file_inode_operations;
2571 inode->i_fop = &btrfs_dir_file_operations;
2572 if (root == root->fs_info->tree_root)
2573 inode->i_op = &btrfs_dir_ro_inode_operations;
2575 inode->i_op = &btrfs_dir_inode_operations;
2578 inode->i_op = &btrfs_symlink_inode_operations;
2579 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2580 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2583 inode->i_op = &btrfs_special_inode_operations;
2584 init_special_inode(inode, inode->i_mode, rdev);
2588 btrfs_update_iflags(inode);
2592 btrfs_free_path(path);
2593 make_bad_inode(inode);
2597 * given a leaf and an inode, copy the inode fields into the leaf
2599 static void fill_inode_item(struct btrfs_trans_handle *trans,
2600 struct extent_buffer *leaf,
2601 struct btrfs_inode_item *item,
2602 struct inode *inode)
2604 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2605 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2606 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2607 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2608 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2610 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2611 inode->i_atime.tv_sec);
2612 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2613 inode->i_atime.tv_nsec);
2615 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2616 inode->i_mtime.tv_sec);
2617 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2618 inode->i_mtime.tv_nsec);
2620 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2621 inode->i_ctime.tv_sec);
2622 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2623 inode->i_ctime.tv_nsec);
2625 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2626 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2627 btrfs_set_inode_sequence(leaf, item, inode->i_version);
2628 btrfs_set_inode_transid(leaf, item, trans->transid);
2629 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2630 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2631 btrfs_set_inode_block_group(leaf, item, 0);
2635 * copy everything in the in-memory inode into the btree.
2637 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2638 struct btrfs_root *root, struct inode *inode)
2640 struct btrfs_inode_item *inode_item;
2641 struct btrfs_path *path;
2642 struct extent_buffer *leaf;
2645 path = btrfs_alloc_path();
2649 path->leave_spinning = 1;
2650 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2658 btrfs_unlock_up_safe(path, 1);
2659 leaf = path->nodes[0];
2660 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2661 struct btrfs_inode_item);
2663 fill_inode_item(trans, leaf, inode_item, inode);
2664 btrfs_mark_buffer_dirty(leaf);
2665 btrfs_set_inode_last_trans(trans, inode);
2668 btrfs_free_path(path);
2673 * copy everything in the in-memory inode into the btree.
2675 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2676 struct btrfs_root *root, struct inode *inode)
2681 * If the inode is a free space inode, we can deadlock during commit
2682 * if we put it into the delayed code.
2684 * The data relocation inode should also be directly updated
2687 if (!btrfs_is_free_space_inode(root, inode)
2688 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2689 ret = btrfs_delayed_update_inode(trans, root, inode);
2691 btrfs_set_inode_last_trans(trans, inode);
2695 return btrfs_update_inode_item(trans, root, inode);
2698 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2699 struct btrfs_root *root, struct inode *inode)
2703 ret = btrfs_update_inode(trans, root, inode);
2705 return btrfs_update_inode_item(trans, root, inode);
2710 * unlink helper that gets used here in inode.c and in the tree logging
2711 * recovery code. It remove a link in a directory with a given name, and
2712 * also drops the back refs in the inode to the directory
2714 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2715 struct btrfs_root *root,
2716 struct inode *dir, struct inode *inode,
2717 const char *name, int name_len)
2719 struct btrfs_path *path;
2721 struct extent_buffer *leaf;
2722 struct btrfs_dir_item *di;
2723 struct btrfs_key key;
2725 u64 ino = btrfs_ino(inode);
2726 u64 dir_ino = btrfs_ino(dir);
2728 path = btrfs_alloc_path();
2734 path->leave_spinning = 1;
2735 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2736 name, name_len, -1);
2745 leaf = path->nodes[0];
2746 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2747 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2750 btrfs_release_path(path);
2752 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2755 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2756 "inode %llu parent %llu\n", name_len, name,
2757 (unsigned long long)ino, (unsigned long long)dir_ino);
2758 btrfs_abort_transaction(trans, root, ret);
2762 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2764 btrfs_abort_transaction(trans, root, ret);
2768 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2770 if (ret != 0 && ret != -ENOENT) {
2771 btrfs_abort_transaction(trans, root, ret);
2775 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2780 btrfs_free_path(path);
2784 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2785 inode_inc_iversion(inode);
2786 inode_inc_iversion(dir);
2787 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2788 btrfs_update_inode(trans, root, dir);
2793 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2794 struct btrfs_root *root,
2795 struct inode *dir, struct inode *inode,
2796 const char *name, int name_len)
2799 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2801 btrfs_drop_nlink(inode);
2802 ret = btrfs_update_inode(trans, root, inode);
2808 /* helper to check if there is any shared block in the path */
2809 static int check_path_shared(struct btrfs_root *root,
2810 struct btrfs_path *path)
2812 struct extent_buffer *eb;
2816 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2819 if (!path->nodes[level])
2821 eb = path->nodes[level];
2822 if (!btrfs_block_can_be_shared(root, eb))
2824 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2833 * helper to start transaction for unlink and rmdir.
2835 * unlink and rmdir are special in btrfs, they do not always free space.
2836 * so in enospc case, we should make sure they will free space before
2837 * allowing them to use the global metadata reservation.
2839 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2840 struct dentry *dentry)
2842 struct btrfs_trans_handle *trans;
2843 struct btrfs_root *root = BTRFS_I(dir)->root;
2844 struct btrfs_path *path;
2845 struct btrfs_inode_ref *ref;
2846 struct btrfs_dir_item *di;
2847 struct inode *inode = dentry->d_inode;
2852 u64 ino = btrfs_ino(inode);
2853 u64 dir_ino = btrfs_ino(dir);
2856 * 1 for the possible orphan item
2857 * 1 for the dir item
2858 * 1 for the dir index
2859 * 1 for the inode ref
2860 * 1 for the inode ref in the tree log
2861 * 2 for the dir entries in the log
2864 trans = btrfs_start_transaction(root, 8);
2865 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2868 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2869 return ERR_PTR(-ENOSPC);
2871 /* check if there is someone else holds reference */
2872 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2873 return ERR_PTR(-ENOSPC);
2875 if (atomic_read(&inode->i_count) > 2)
2876 return ERR_PTR(-ENOSPC);
2878 if (xchg(&root->fs_info->enospc_unlink, 1))
2879 return ERR_PTR(-ENOSPC);
2881 path = btrfs_alloc_path();
2883 root->fs_info->enospc_unlink = 0;
2884 return ERR_PTR(-ENOMEM);
2887 /* 1 for the orphan item */
2888 trans = btrfs_start_transaction(root, 1);
2889 if (IS_ERR(trans)) {
2890 btrfs_free_path(path);
2891 root->fs_info->enospc_unlink = 0;
2895 path->skip_locking = 1;
2896 path->search_commit_root = 1;
2898 ret = btrfs_lookup_inode(trans, root, path,
2899 &BTRFS_I(dir)->location, 0);
2905 if (check_path_shared(root, path))
2910 btrfs_release_path(path);
2912 ret = btrfs_lookup_inode(trans, root, path,
2913 &BTRFS_I(inode)->location, 0);
2919 if (check_path_shared(root, path))
2924 btrfs_release_path(path);
2926 if (ret == 0 && S_ISREG(inode->i_mode)) {
2927 ret = btrfs_lookup_file_extent(trans, root, path,
2933 BUG_ON(ret == 0); /* Corruption */
2934 if (check_path_shared(root, path))
2936 btrfs_release_path(path);
2944 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2945 dentry->d_name.name, dentry->d_name.len, 0);
2951 if (check_path_shared(root, path))
2957 btrfs_release_path(path);
2959 ref = btrfs_lookup_inode_ref(trans, root, path,
2960 dentry->d_name.name, dentry->d_name.len,
2966 BUG_ON(!ref); /* Logic error */
2967 if (check_path_shared(root, path))
2969 index = btrfs_inode_ref_index(path->nodes[0], ref);
2970 btrfs_release_path(path);
2973 * This is a commit root search, if we can lookup inode item and other
2974 * relative items in the commit root, it means the transaction of
2975 * dir/file creation has been committed, and the dir index item that we
2976 * delay to insert has also been inserted into the commit root. So
2977 * we needn't worry about the delayed insertion of the dir index item
2980 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2981 dentry->d_name.name, dentry->d_name.len, 0);
2986 BUG_ON(ret == -ENOENT);
2987 if (check_path_shared(root, path))
2992 btrfs_free_path(path);
2993 /* Migrate the orphan reservation over */
2995 err = btrfs_block_rsv_migrate(trans->block_rsv,
2996 &root->fs_info->global_block_rsv,
2997 trans->bytes_reserved);
3000 btrfs_end_transaction(trans, root);
3001 root->fs_info->enospc_unlink = 0;
3002 return ERR_PTR(err);
3005 trans->block_rsv = &root->fs_info->global_block_rsv;
3009 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3010 struct btrfs_root *root)
3012 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
3013 btrfs_block_rsv_release(root, trans->block_rsv,
3014 trans->bytes_reserved);
3015 trans->block_rsv = &root->fs_info->trans_block_rsv;
3016 BUG_ON(!root->fs_info->enospc_unlink);
3017 root->fs_info->enospc_unlink = 0;
3019 btrfs_end_transaction(trans, root);
3022 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3024 struct btrfs_root *root = BTRFS_I(dir)->root;
3025 struct btrfs_trans_handle *trans;
3026 struct inode *inode = dentry->d_inode;
3028 unsigned long nr = 0;
3030 trans = __unlink_start_trans(dir, dentry);
3032 return PTR_ERR(trans);
3034 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3036 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3037 dentry->d_name.name, dentry->d_name.len);
3041 if (inode->i_nlink == 0) {
3042 ret = btrfs_orphan_add(trans, inode);
3048 nr = trans->blocks_used;
3049 __unlink_end_trans(trans, root);
3050 btrfs_btree_balance_dirty(root, nr);
3054 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3055 struct btrfs_root *root,
3056 struct inode *dir, u64 objectid,
3057 const char *name, int name_len)
3059 struct btrfs_path *path;
3060 struct extent_buffer *leaf;
3061 struct btrfs_dir_item *di;
3062 struct btrfs_key key;
3065 u64 dir_ino = btrfs_ino(dir);
3067 path = btrfs_alloc_path();
3071 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3072 name, name_len, -1);
3073 if (IS_ERR_OR_NULL(di)) {
3081 leaf = path->nodes[0];
3082 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3083 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3084 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3086 btrfs_abort_transaction(trans, root, ret);
3089 btrfs_release_path(path);
3091 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3092 objectid, root->root_key.objectid,
3093 dir_ino, &index, name, name_len);
3095 if (ret != -ENOENT) {
3096 btrfs_abort_transaction(trans, root, ret);
3099 di = btrfs_search_dir_index_item(root, path, dir_ino,
3101 if (IS_ERR_OR_NULL(di)) {
3106 btrfs_abort_transaction(trans, root, ret);
3110 leaf = path->nodes[0];
3111 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3112 btrfs_release_path(path);
3115 btrfs_release_path(path);
3117 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3119 btrfs_abort_transaction(trans, root, ret);
3123 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3124 inode_inc_iversion(dir);
3125 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3126 ret = btrfs_update_inode(trans, root, dir);
3128 btrfs_abort_transaction(trans, root, ret);
3130 btrfs_free_path(path);
3134 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3136 struct inode *inode = dentry->d_inode;
3138 struct btrfs_root *root = BTRFS_I(dir)->root;
3139 struct btrfs_trans_handle *trans;
3140 unsigned long nr = 0;
3142 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3143 btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3146 trans = __unlink_start_trans(dir, dentry);
3148 return PTR_ERR(trans);
3150 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3151 err = btrfs_unlink_subvol(trans, root, dir,
3152 BTRFS_I(inode)->location.objectid,
3153 dentry->d_name.name,
3154 dentry->d_name.len);
3158 err = btrfs_orphan_add(trans, inode);
3162 /* now the directory is empty */
3163 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3164 dentry->d_name.name, dentry->d_name.len);
3166 btrfs_i_size_write(inode, 0);
3168 nr = trans->blocks_used;
3169 __unlink_end_trans(trans, root);
3170 btrfs_btree_balance_dirty(root, nr);
3176 * this can truncate away extent items, csum items and directory items.
3177 * It starts at a high offset and removes keys until it can't find
3178 * any higher than new_size
3180 * csum items that cross the new i_size are truncated to the new size
3183 * min_type is the minimum key type to truncate down to. If set to 0, this
3184 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3186 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3187 struct btrfs_root *root,
3188 struct inode *inode,
3189 u64 new_size, u32 min_type)
3191 struct btrfs_path *path;
3192 struct extent_buffer *leaf;
3193 struct btrfs_file_extent_item *fi;
3194 struct btrfs_key key;
3195 struct btrfs_key found_key;
3196 u64 extent_start = 0;
3197 u64 extent_num_bytes = 0;
3198 u64 extent_offset = 0;
3200 u64 mask = root->sectorsize - 1;
3201 u32 found_type = (u8)-1;
3204 int pending_del_nr = 0;
3205 int pending_del_slot = 0;
3206 int extent_type = -1;
3209 u64 ino = btrfs_ino(inode);
3211 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3213 path = btrfs_alloc_path();
3218 if (root->ref_cows || root == root->fs_info->tree_root)
3219 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3222 * This function is also used to drop the items in the log tree before
3223 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3224 * it is used to drop the loged items. So we shouldn't kill the delayed
3227 if (min_type == 0 && root == BTRFS_I(inode)->root)
3228 btrfs_kill_delayed_inode_items(inode);
3231 key.offset = (u64)-1;
3235 path->leave_spinning = 1;
3236 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3243 /* there are no items in the tree for us to truncate, we're
3246 if (path->slots[0] == 0)
3253 leaf = path->nodes[0];
3254 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3255 found_type = btrfs_key_type(&found_key);
3257 if (found_key.objectid != ino)
3260 if (found_type < min_type)
3263 item_end = found_key.offset;
3264 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3265 fi = btrfs_item_ptr(leaf, path->slots[0],
3266 struct btrfs_file_extent_item);
3267 extent_type = btrfs_file_extent_type(leaf, fi);
3268 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3270 btrfs_file_extent_num_bytes(leaf, fi);
3271 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3272 item_end += btrfs_file_extent_inline_len(leaf,
3277 if (found_type > min_type) {
3280 if (item_end < new_size)
3282 if (found_key.offset >= new_size)
3288 /* FIXME, shrink the extent if the ref count is only 1 */
3289 if (found_type != BTRFS_EXTENT_DATA_KEY)
3292 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3294 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3296 u64 orig_num_bytes =
3297 btrfs_file_extent_num_bytes(leaf, fi);
3298 extent_num_bytes = new_size -
3299 found_key.offset + root->sectorsize - 1;
3300 extent_num_bytes = extent_num_bytes &
3301 ~((u64)root->sectorsize - 1);
3302 btrfs_set_file_extent_num_bytes(leaf, fi,
3304 num_dec = (orig_num_bytes -
3306 if (root->ref_cows && extent_start != 0)
3307 inode_sub_bytes(inode, num_dec);
3308 btrfs_mark_buffer_dirty(leaf);
3311 btrfs_file_extent_disk_num_bytes(leaf,
3313 extent_offset = found_key.offset -
3314 btrfs_file_extent_offset(leaf, fi);
3316 /* FIXME blocksize != 4096 */
3317 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3318 if (extent_start != 0) {
3321 inode_sub_bytes(inode, num_dec);
3324 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3326 * we can't truncate inline items that have had
3330 btrfs_file_extent_compression(leaf, fi) == 0 &&
3331 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3332 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3333 u32 size = new_size - found_key.offset;
3335 if (root->ref_cows) {
3336 inode_sub_bytes(inode, item_end + 1 -
3340 btrfs_file_extent_calc_inline_size(size);
3341 btrfs_truncate_item(trans, root, path,
3343 } else if (root->ref_cows) {
3344 inode_sub_bytes(inode, item_end + 1 -
3350 if (!pending_del_nr) {
3351 /* no pending yet, add ourselves */
3352 pending_del_slot = path->slots[0];
3354 } else if (pending_del_nr &&
3355 path->slots[0] + 1 == pending_del_slot) {
3356 /* hop on the pending chunk */
3358 pending_del_slot = path->slots[0];
3365 if (found_extent && (root->ref_cows ||
3366 root == root->fs_info->tree_root)) {
3367 btrfs_set_path_blocking(path);
3368 ret = btrfs_free_extent(trans, root, extent_start,
3369 extent_num_bytes, 0,
3370 btrfs_header_owner(leaf),
3371 ino, extent_offset, 0);
3375 if (found_type == BTRFS_INODE_ITEM_KEY)
3378 if (path->slots[0] == 0 ||
3379 path->slots[0] != pending_del_slot) {
3380 if (root->ref_cows &&
3381 BTRFS_I(inode)->location.objectid !=
3382 BTRFS_FREE_INO_OBJECTID) {
3386 if (pending_del_nr) {
3387 ret = btrfs_del_items(trans, root, path,
3391 btrfs_abort_transaction(trans,
3397 btrfs_release_path(path);
3404 if (pending_del_nr) {
3405 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3408 btrfs_abort_transaction(trans, root, ret);
3411 btrfs_free_path(path);
3416 * taken from block_truncate_page, but does cow as it zeros out
3417 * any bytes left in the last page in the file.
3419 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3421 struct inode *inode = mapping->host;
3422 struct btrfs_root *root = BTRFS_I(inode)->root;
3423 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3424 struct btrfs_ordered_extent *ordered;
3425 struct extent_state *cached_state = NULL;
3427 u32 blocksize = root->sectorsize;
3428 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3429 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3431 gfp_t mask = btrfs_alloc_write_mask(mapping);
3436 if ((offset & (blocksize - 1)) == 0)
3438 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3444 page = find_or_create_page(mapping, index, mask);
3446 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3450 page_start = page_offset(page);
3451 page_end = page_start + PAGE_CACHE_SIZE - 1;
3453 if (!PageUptodate(page)) {
3454 ret = btrfs_readpage(NULL, page);
3456 if (page->mapping != mapping) {
3458 page_cache_release(page);
3461 if (!PageUptodate(page)) {
3466 wait_on_page_writeback(page);
3468 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3469 set_page_extent_mapped(page);
3471 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3473 unlock_extent_cached(io_tree, page_start, page_end,
3474 &cached_state, GFP_NOFS);
3476 page_cache_release(page);
3477 btrfs_start_ordered_extent(inode, ordered, 1);
3478 btrfs_put_ordered_extent(ordered);
3482 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3483 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3484 0, 0, &cached_state, GFP_NOFS);
3486 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3489 unlock_extent_cached(io_tree, page_start, page_end,
3490 &cached_state, GFP_NOFS);
3495 if (offset != PAGE_CACHE_SIZE) {
3497 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3498 flush_dcache_page(page);
3501 ClearPageChecked(page);
3502 set_page_dirty(page);
3503 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3508 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3510 page_cache_release(page);
3516 * This function puts in dummy file extents for the area we're creating a hole
3517 * for. So if we are truncating this file to a larger size we need to insert
3518 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3519 * the range between oldsize and size
3521 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3523 struct btrfs_trans_handle *trans;
3524 struct btrfs_root *root = BTRFS_I(inode)->root;
3525 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3526 struct extent_map *em = NULL;
3527 struct extent_state *cached_state = NULL;
3528 u64 mask = root->sectorsize - 1;
3529 u64 hole_start = (oldsize + mask) & ~mask;
3530 u64 block_end = (size + mask) & ~mask;
3536 if (size <= hole_start)
3540 struct btrfs_ordered_extent *ordered;
3541 btrfs_wait_ordered_range(inode, hole_start,
3542 block_end - hole_start);
3543 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3545 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3548 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3549 &cached_state, GFP_NOFS);
3550 btrfs_put_ordered_extent(ordered);
3553 cur_offset = hole_start;
3555 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3556 block_end - cur_offset, 0);
3561 last_byte = min(extent_map_end(em), block_end);
3562 last_byte = (last_byte + mask) & ~mask;
3563 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3565 hole_size = last_byte - cur_offset;
3567 trans = btrfs_start_transaction(root, 3);
3568 if (IS_ERR(trans)) {
3569 err = PTR_ERR(trans);
3573 err = btrfs_drop_extents(trans, inode, cur_offset,
3574 cur_offset + hole_size,
3577 btrfs_abort_transaction(trans, root, err);
3578 btrfs_end_transaction(trans, root);
3582 err = btrfs_insert_file_extent(trans, root,
3583 btrfs_ino(inode), cur_offset, 0,
3584 0, hole_size, 0, hole_size,
3587 btrfs_abort_transaction(trans, root, err);
3588 btrfs_end_transaction(trans, root);
3592 btrfs_drop_extent_cache(inode, hole_start,
3595 btrfs_update_inode(trans, root, inode);
3596 btrfs_end_transaction(trans, root);
3598 free_extent_map(em);
3600 cur_offset = last_byte;
3601 if (cur_offset >= block_end)
3605 free_extent_map(em);
3606 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3611 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3613 struct btrfs_root *root = BTRFS_I(inode)->root;
3614 struct btrfs_trans_handle *trans;
3615 loff_t oldsize = i_size_read(inode);
3618 if (newsize == oldsize)
3621 if (newsize > oldsize) {
3622 truncate_pagecache(inode, oldsize, newsize);
3623 ret = btrfs_cont_expand(inode, oldsize, newsize);
3627 trans = btrfs_start_transaction(root, 1);
3629 return PTR_ERR(trans);
3631 i_size_write(inode, newsize);
3632 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3633 ret = btrfs_update_inode(trans, root, inode);
3634 btrfs_end_transaction(trans, root);
3638 * We're truncating a file that used to have good data down to
3639 * zero. Make sure it gets into the ordered flush list so that
3640 * any new writes get down to disk quickly.
3643 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3644 &BTRFS_I(inode)->runtime_flags);
3646 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3647 truncate_setsize(inode, newsize);
3648 ret = btrfs_truncate(inode);
3654 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3656 struct inode *inode = dentry->d_inode;
3657 struct btrfs_root *root = BTRFS_I(inode)->root;
3660 if (btrfs_root_readonly(root))
3663 err = inode_change_ok(inode, attr);
3667 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3668 err = btrfs_setsize(inode, attr->ia_size);
3673 if (attr->ia_valid) {
3674 setattr_copy(inode, attr);
3675 inode_inc_iversion(inode);
3676 err = btrfs_dirty_inode(inode);
3678 if (!err && attr->ia_valid & ATTR_MODE)
3679 err = btrfs_acl_chmod(inode);
3685 void btrfs_evict_inode(struct inode *inode)
3687 struct btrfs_trans_handle *trans;
3688 struct btrfs_root *root = BTRFS_I(inode)->root;
3689 struct btrfs_block_rsv *rsv, *global_rsv;
3690 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3694 trace_btrfs_inode_evict(inode);
3696 truncate_inode_pages(&inode->i_data, 0);
3697 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3698 btrfs_is_free_space_inode(root, inode)))
3701 if (is_bad_inode(inode)) {
3702 btrfs_orphan_del(NULL, inode);
3705 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3706 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3708 if (root->fs_info->log_root_recovering) {
3709 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3713 if (inode->i_nlink > 0) {
3714 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3718 rsv = btrfs_alloc_block_rsv(root);
3720 btrfs_orphan_del(NULL, inode);
3723 rsv->size = min_size;
3724 global_rsv = &root->fs_info->global_block_rsv;
3726 btrfs_i_size_write(inode, 0);
3729 * This is a bit simpler than btrfs_truncate since
3731 * 1) We've already reserved our space for our orphan item in the
3733 * 2) We're going to delete the inode item, so we don't need to update
3736 * So we just need to reserve some slack space in case we add bytes when
3737 * doing the truncate.
3740 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
3743 * Try and steal from the global reserve since we will
3744 * likely not use this space anyway, we want to try as
3745 * hard as possible to get this to work.
3748 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3751 printk(KERN_WARNING "Could not get space for a "
3752 "delete, will truncate on mount %d\n", ret);
3753 btrfs_orphan_del(NULL, inode);
3754 btrfs_free_block_rsv(root, rsv);
3758 trans = btrfs_start_transaction(root, 0);
3759 if (IS_ERR(trans)) {
3760 btrfs_orphan_del(NULL, inode);
3761 btrfs_free_block_rsv(root, rsv);
3765 trans->block_rsv = rsv;
3767 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3771 nr = trans->blocks_used;
3772 btrfs_end_transaction(trans, root);
3774 btrfs_btree_balance_dirty(root, nr);
3777 btrfs_free_block_rsv(root, rsv);
3780 trans->block_rsv = root->orphan_block_rsv;
3781 ret = btrfs_orphan_del(trans, inode);
3785 trans->block_rsv = &root->fs_info->trans_block_rsv;
3786 if (!(root == root->fs_info->tree_root ||
3787 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3788 btrfs_return_ino(root, btrfs_ino(inode));
3790 nr = trans->blocks_used;
3791 btrfs_end_transaction(trans, root);
3792 btrfs_btree_balance_dirty(root, nr);
3794 end_writeback(inode);
3799 * this returns the key found in the dir entry in the location pointer.
3800 * If no dir entries were found, location->objectid is 0.
3802 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3803 struct btrfs_key *location)
3805 const char *name = dentry->d_name.name;
3806 int namelen = dentry->d_name.len;
3807 struct btrfs_dir_item *di;
3808 struct btrfs_path *path;
3809 struct btrfs_root *root = BTRFS_I(dir)->root;
3812 path = btrfs_alloc_path();
3816 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3821 if (IS_ERR_OR_NULL(di))
3824 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3826 btrfs_free_path(path);
3829 location->objectid = 0;
3834 * when we hit a tree root in a directory, the btrfs part of the inode
3835 * needs to be changed to reflect the root directory of the tree root. This
3836 * is kind of like crossing a mount point.
3838 static int fixup_tree_root_location(struct btrfs_root *root,
3840 struct dentry *dentry,
3841 struct btrfs_key *location,
3842 struct btrfs_root **sub_root)
3844 struct btrfs_path *path;
3845 struct btrfs_root *new_root;
3846 struct btrfs_root_ref *ref;
3847 struct extent_buffer *leaf;
3851 path = btrfs_alloc_path();
3858 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3859 BTRFS_I(dir)->root->root_key.objectid,
3860 location->objectid);
3867 leaf = path->nodes[0];
3868 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3869 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3870 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3873 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3874 (unsigned long)(ref + 1),
3875 dentry->d_name.len);
3879 btrfs_release_path(path);
3881 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3882 if (IS_ERR(new_root)) {
3883 err = PTR_ERR(new_root);
3887 if (btrfs_root_refs(&new_root->root_item) == 0) {
3892 *sub_root = new_root;
3893 location->objectid = btrfs_root_dirid(&new_root->root_item);
3894 location->type = BTRFS_INODE_ITEM_KEY;
3895 location->offset = 0;
3898 btrfs_free_path(path);
3902 static void inode_tree_add(struct inode *inode)
3904 struct btrfs_root *root = BTRFS_I(inode)->root;
3905 struct btrfs_inode *entry;
3907 struct rb_node *parent;
3908 u64 ino = btrfs_ino(inode);
3910 p = &root->inode_tree.rb_node;
3913 if (inode_unhashed(inode))
3916 spin_lock(&root->inode_lock);
3919 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3921 if (ino < btrfs_ino(&entry->vfs_inode))
3922 p = &parent->rb_left;
3923 else if (ino > btrfs_ino(&entry->vfs_inode))
3924 p = &parent->rb_right;
3926 WARN_ON(!(entry->vfs_inode.i_state &
3927 (I_WILL_FREE | I_FREEING)));
3928 rb_erase(parent, &root->inode_tree);
3929 RB_CLEAR_NODE(parent);
3930 spin_unlock(&root->inode_lock);
3934 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3935 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3936 spin_unlock(&root->inode_lock);
3939 static void inode_tree_del(struct inode *inode)
3941 struct btrfs_root *root = BTRFS_I(inode)->root;
3944 spin_lock(&root->inode_lock);
3945 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3946 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3947 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3948 empty = RB_EMPTY_ROOT(&root->inode_tree);
3950 spin_unlock(&root->inode_lock);
3953 * Free space cache has inodes in the tree root, but the tree root has a
3954 * root_refs of 0, so this could end up dropping the tree root as a
3955 * snapshot, so we need the extra !root->fs_info->tree_root check to
3956 * make sure we don't drop it.
3958 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3959 root != root->fs_info->tree_root) {
3960 synchronize_srcu(&root->fs_info->subvol_srcu);
3961 spin_lock(&root->inode_lock);
3962 empty = RB_EMPTY_ROOT(&root->inode_tree);
3963 spin_unlock(&root->inode_lock);
3965 btrfs_add_dead_root(root);
3969 void btrfs_invalidate_inodes(struct btrfs_root *root)
3971 struct rb_node *node;
3972 struct rb_node *prev;
3973 struct btrfs_inode *entry;
3974 struct inode *inode;
3977 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3979 spin_lock(&root->inode_lock);
3981 node = root->inode_tree.rb_node;
3985 entry = rb_entry(node, struct btrfs_inode, rb_node);
3987 if (objectid < btrfs_ino(&entry->vfs_inode))
3988 node = node->rb_left;
3989 else if (objectid > btrfs_ino(&entry->vfs_inode))
3990 node = node->rb_right;
3996 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3997 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4001 prev = rb_next(prev);
4005 entry = rb_entry(node, struct btrfs_inode, rb_node);
4006 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4007 inode = igrab(&entry->vfs_inode);
4009 spin_unlock(&root->inode_lock);
4010 if (atomic_read(&inode->i_count) > 1)
4011 d_prune_aliases(inode);
4013 * btrfs_drop_inode will have it removed from
4014 * the inode cache when its usage count
4019 spin_lock(&root->inode_lock);
4023 if (cond_resched_lock(&root->inode_lock))
4026 node = rb_next(node);
4028 spin_unlock(&root->inode_lock);
4031 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4033 struct btrfs_iget_args *args = p;
4034 inode->i_ino = args->ino;
4035 BTRFS_I(inode)->root = args->root;
4036 btrfs_set_inode_space_info(args->root, inode);
4040 static int btrfs_find_actor(struct inode *inode, void *opaque)
4042 struct btrfs_iget_args *args = opaque;
4043 return args->ino == btrfs_ino(inode) &&
4044 args->root == BTRFS_I(inode)->root;
4047 static struct inode *btrfs_iget_locked(struct super_block *s,
4049 struct btrfs_root *root)
4051 struct inode *inode;
4052 struct btrfs_iget_args args;
4053 args.ino = objectid;
4056 inode = iget5_locked(s, objectid, btrfs_find_actor,
4057 btrfs_init_locked_inode,
4062 /* Get an inode object given its location and corresponding root.
4063 * Returns in *is_new if the inode was read from disk
4065 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4066 struct btrfs_root *root, int *new)
4068 struct inode *inode;
4070 inode = btrfs_iget_locked(s, location->objectid, root);
4072 return ERR_PTR(-ENOMEM);
4074 if (inode->i_state & I_NEW) {
4075 BTRFS_I(inode)->root = root;
4076 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4077 btrfs_read_locked_inode(inode);
4078 if (!is_bad_inode(inode)) {
4079 inode_tree_add(inode);
4080 unlock_new_inode(inode);
4084 unlock_new_inode(inode);
4086 inode = ERR_PTR(-ESTALE);
4093 static struct inode *new_simple_dir(struct super_block *s,
4094 struct btrfs_key *key,
4095 struct btrfs_root *root)
4097 struct inode *inode = new_inode(s);
4100 return ERR_PTR(-ENOMEM);
4102 BTRFS_I(inode)->root = root;
4103 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4104 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4106 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4107 inode->i_op = &btrfs_dir_ro_inode_operations;
4108 inode->i_fop = &simple_dir_operations;
4109 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4110 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4115 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4117 struct inode *inode;
4118 struct btrfs_root *root = BTRFS_I(dir)->root;
4119 struct btrfs_root *sub_root = root;
4120 struct btrfs_key location;
4124 if (dentry->d_name.len > BTRFS_NAME_LEN)
4125 return ERR_PTR(-ENAMETOOLONG);
4127 if (unlikely(d_need_lookup(dentry))) {
4128 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4129 kfree(dentry->d_fsdata);
4130 dentry->d_fsdata = NULL;
4131 /* This thing is hashed, drop it for now */
4134 ret = btrfs_inode_by_name(dir, dentry, &location);
4138 return ERR_PTR(ret);
4140 if (location.objectid == 0)
4143 if (location.type == BTRFS_INODE_ITEM_KEY) {
4144 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4148 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4150 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4151 ret = fixup_tree_root_location(root, dir, dentry,
4152 &location, &sub_root);
4155 inode = ERR_PTR(ret);
4157 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4159 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4161 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4163 if (!IS_ERR(inode) && root != sub_root) {
4164 down_read(&root->fs_info->cleanup_work_sem);
4165 if (!(inode->i_sb->s_flags & MS_RDONLY))
4166 ret = btrfs_orphan_cleanup(sub_root);
4167 up_read(&root->fs_info->cleanup_work_sem);
4169 inode = ERR_PTR(ret);
4175 static int btrfs_dentry_delete(const struct dentry *dentry)
4177 struct btrfs_root *root;
4178 struct inode *inode = dentry->d_inode;
4180 if (!inode && !IS_ROOT(dentry))
4181 inode = dentry->d_parent->d_inode;
4184 root = BTRFS_I(inode)->root;
4185 if (btrfs_root_refs(&root->root_item) == 0)
4188 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4194 static void btrfs_dentry_release(struct dentry *dentry)
4196 if (dentry->d_fsdata)
4197 kfree(dentry->d_fsdata);
4200 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4201 struct nameidata *nd)
4205 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4206 if (unlikely(d_need_lookup(dentry))) {
4207 spin_lock(&dentry->d_lock);
4208 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4209 spin_unlock(&dentry->d_lock);
4214 unsigned char btrfs_filetype_table[] = {
4215 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4218 static int btrfs_real_readdir(struct file *filp, void *dirent,
4221 struct inode *inode = filp->f_dentry->d_inode;
4222 struct btrfs_root *root = BTRFS_I(inode)->root;
4223 struct btrfs_item *item;
4224 struct btrfs_dir_item *di;
4225 struct btrfs_key key;
4226 struct btrfs_key found_key;
4227 struct btrfs_path *path;
4228 struct list_head ins_list;
4229 struct list_head del_list;
4231 struct extent_buffer *leaf;
4233 unsigned char d_type;
4238 int key_type = BTRFS_DIR_INDEX_KEY;
4242 int is_curr = 0; /* filp->f_pos points to the current index? */
4244 /* FIXME, use a real flag for deciding about the key type */
4245 if (root->fs_info->tree_root == root)
4246 key_type = BTRFS_DIR_ITEM_KEY;
4248 /* special case for "." */
4249 if (filp->f_pos == 0) {
4250 over = filldir(dirent, ".", 1,
4251 filp->f_pos, btrfs_ino(inode), DT_DIR);
4256 /* special case for .., just use the back ref */
4257 if (filp->f_pos == 1) {
4258 u64 pino = parent_ino(filp->f_path.dentry);
4259 over = filldir(dirent, "..", 2,
4260 filp->f_pos, pino, DT_DIR);
4265 path = btrfs_alloc_path();
4271 if (key_type == BTRFS_DIR_INDEX_KEY) {
4272 INIT_LIST_HEAD(&ins_list);
4273 INIT_LIST_HEAD(&del_list);
4274 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4277 btrfs_set_key_type(&key, key_type);
4278 key.offset = filp->f_pos;
4279 key.objectid = btrfs_ino(inode);
4281 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4286 leaf = path->nodes[0];
4287 slot = path->slots[0];
4288 if (slot >= btrfs_header_nritems(leaf)) {
4289 ret = btrfs_next_leaf(root, path);
4297 item = btrfs_item_nr(leaf, slot);
4298 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4300 if (found_key.objectid != key.objectid)
4302 if (btrfs_key_type(&found_key) != key_type)
4304 if (found_key.offset < filp->f_pos)
4306 if (key_type == BTRFS_DIR_INDEX_KEY &&
4307 btrfs_should_delete_dir_index(&del_list,
4311 filp->f_pos = found_key.offset;
4314 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4316 di_total = btrfs_item_size(leaf, item);
4318 while (di_cur < di_total) {
4319 struct btrfs_key location;
4321 if (verify_dir_item(root, leaf, di))
4324 name_len = btrfs_dir_name_len(leaf, di);
4325 if (name_len <= sizeof(tmp_name)) {
4326 name_ptr = tmp_name;
4328 name_ptr = kmalloc(name_len, GFP_NOFS);
4334 read_extent_buffer(leaf, name_ptr,
4335 (unsigned long)(di + 1), name_len);
4337 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4338 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4341 /* is this a reference to our own snapshot? If so
4344 * In contrast to old kernels, we insert the snapshot's
4345 * dir item and dir index after it has been created, so
4346 * we won't find a reference to our own snapshot. We
4347 * still keep the following code for backward
4350 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4351 location.objectid == root->root_key.objectid) {
4355 over = filldir(dirent, name_ptr, name_len,
4356 found_key.offset, location.objectid,
4360 if (name_ptr != tmp_name)
4365 di_len = btrfs_dir_name_len(leaf, di) +
4366 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4368 di = (struct btrfs_dir_item *)((char *)di + di_len);
4374 if (key_type == BTRFS_DIR_INDEX_KEY) {
4377 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4383 /* Reached end of directory/root. Bump pos past the last item. */
4384 if (key_type == BTRFS_DIR_INDEX_KEY)
4386 * 32-bit glibc will use getdents64, but then strtol -
4387 * so the last number we can serve is this.
4389 filp->f_pos = 0x7fffffff;
4395 if (key_type == BTRFS_DIR_INDEX_KEY)
4396 btrfs_put_delayed_items(&ins_list, &del_list);
4397 btrfs_free_path(path);
4401 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4403 struct btrfs_root *root = BTRFS_I(inode)->root;
4404 struct btrfs_trans_handle *trans;
4406 bool nolock = false;
4408 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4411 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4414 if (wbc->sync_mode == WB_SYNC_ALL) {
4416 trans = btrfs_join_transaction_nolock(root);
4418 trans = btrfs_join_transaction(root);
4420 return PTR_ERR(trans);
4422 ret = btrfs_end_transaction_nolock(trans, root);
4424 ret = btrfs_commit_transaction(trans, root);
4430 * This is somewhat expensive, updating the tree every time the
4431 * inode changes. But, it is most likely to find the inode in cache.
4432 * FIXME, needs more benchmarking...there are no reasons other than performance
4433 * to keep or drop this code.
4435 int btrfs_dirty_inode(struct inode *inode)
4437 struct btrfs_root *root = BTRFS_I(inode)->root;
4438 struct btrfs_trans_handle *trans;
4441 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4444 trans = btrfs_join_transaction(root);
4446 return PTR_ERR(trans);
4448 ret = btrfs_update_inode(trans, root, inode);
4449 if (ret && ret == -ENOSPC) {
4450 /* whoops, lets try again with the full transaction */
4451 btrfs_end_transaction(trans, root);
4452 trans = btrfs_start_transaction(root, 1);
4454 return PTR_ERR(trans);
4456 ret = btrfs_update_inode(trans, root, inode);
4458 btrfs_end_transaction(trans, root);
4459 if (BTRFS_I(inode)->delayed_node)
4460 btrfs_balance_delayed_items(root);
4466 * This is a copy of file_update_time. We need this so we can return error on
4467 * ENOSPC for updating the inode in the case of file write and mmap writes.
4469 int btrfs_update_time(struct file *file)
4471 struct inode *inode = file->f_path.dentry->d_inode;
4472 struct timespec now;
4474 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
4476 /* First try to exhaust all avenues to not sync */
4477 if (IS_NOCMTIME(inode))
4480 now = current_fs_time(inode->i_sb);
4481 if (!timespec_equal(&inode->i_mtime, &now))
4484 if (!timespec_equal(&inode->i_ctime, &now))
4487 if (IS_I_VERSION(inode))
4488 sync_it |= S_VERSION;
4493 /* Finally allowed to write? Takes lock. */
4494 if (mnt_want_write_file(file))
4497 /* Only change inode inside the lock region */
4498 if (sync_it & S_VERSION)
4499 inode_inc_iversion(inode);
4500 if (sync_it & S_CTIME)
4501 inode->i_ctime = now;
4502 if (sync_it & S_MTIME)
4503 inode->i_mtime = now;
4504 ret = btrfs_dirty_inode(inode);
4506 mark_inode_dirty_sync(inode);
4507 mnt_drop_write(file->f_path.mnt);
4512 * find the highest existing sequence number in a directory
4513 * and then set the in-memory index_cnt variable to reflect
4514 * free sequence numbers
4516 static int btrfs_set_inode_index_count(struct inode *inode)
4518 struct btrfs_root *root = BTRFS_I(inode)->root;
4519 struct btrfs_key key, found_key;
4520 struct btrfs_path *path;
4521 struct extent_buffer *leaf;
4524 key.objectid = btrfs_ino(inode);
4525 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4526 key.offset = (u64)-1;
4528 path = btrfs_alloc_path();
4532 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4535 /* FIXME: we should be able to handle this */
4541 * MAGIC NUMBER EXPLANATION:
4542 * since we search a directory based on f_pos we have to start at 2
4543 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4544 * else has to start at 2
4546 if (path->slots[0] == 0) {
4547 BTRFS_I(inode)->index_cnt = 2;
4553 leaf = path->nodes[0];
4554 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4556 if (found_key.objectid != btrfs_ino(inode) ||
4557 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4558 BTRFS_I(inode)->index_cnt = 2;
4562 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4564 btrfs_free_path(path);
4569 * helper to find a free sequence number in a given directory. This current
4570 * code is very simple, later versions will do smarter things in the btree
4572 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4576 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4577 ret = btrfs_inode_delayed_dir_index_count(dir);
4579 ret = btrfs_set_inode_index_count(dir);
4585 *index = BTRFS_I(dir)->index_cnt;
4586 BTRFS_I(dir)->index_cnt++;
4591 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4592 struct btrfs_root *root,
4594 const char *name, int name_len,
4595 u64 ref_objectid, u64 objectid,
4596 umode_t mode, u64 *index)
4598 struct inode *inode;
4599 struct btrfs_inode_item *inode_item;
4600 struct btrfs_key *location;
4601 struct btrfs_path *path;
4602 struct btrfs_inode_ref *ref;
4603 struct btrfs_key key[2];
4609 path = btrfs_alloc_path();
4611 return ERR_PTR(-ENOMEM);
4613 inode = new_inode(root->fs_info->sb);
4615 btrfs_free_path(path);
4616 return ERR_PTR(-ENOMEM);
4620 * we have to initialize this early, so we can reclaim the inode
4621 * number if we fail afterwards in this function.
4623 inode->i_ino = objectid;
4626 trace_btrfs_inode_request(dir);
4628 ret = btrfs_set_inode_index(dir, index);
4630 btrfs_free_path(path);
4632 return ERR_PTR(ret);
4636 * index_cnt is ignored for everything but a dir,
4637 * btrfs_get_inode_index_count has an explanation for the magic
4640 BTRFS_I(inode)->index_cnt = 2;
4641 BTRFS_I(inode)->root = root;
4642 BTRFS_I(inode)->generation = trans->transid;
4643 inode->i_generation = BTRFS_I(inode)->generation;
4644 btrfs_set_inode_space_info(root, inode);
4651 key[0].objectid = objectid;
4652 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4655 key[1].objectid = objectid;
4656 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4657 key[1].offset = ref_objectid;
4659 sizes[0] = sizeof(struct btrfs_inode_item);
4660 sizes[1] = name_len + sizeof(*ref);
4662 path->leave_spinning = 1;
4663 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4667 inode_init_owner(inode, dir, mode);
4668 inode_set_bytes(inode, 0);
4669 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4670 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4671 struct btrfs_inode_item);
4672 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4674 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4675 struct btrfs_inode_ref);
4676 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4677 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4678 ptr = (unsigned long)(ref + 1);
4679 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4681 btrfs_mark_buffer_dirty(path->nodes[0]);
4682 btrfs_free_path(path);
4684 location = &BTRFS_I(inode)->location;
4685 location->objectid = objectid;
4686 location->offset = 0;
4687 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4689 btrfs_inherit_iflags(inode, dir);
4691 if (S_ISREG(mode)) {
4692 if (btrfs_test_opt(root, NODATASUM))
4693 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4694 if (btrfs_test_opt(root, NODATACOW) ||
4695 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4696 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4699 insert_inode_hash(inode);
4700 inode_tree_add(inode);
4702 trace_btrfs_inode_new(inode);
4703 btrfs_set_inode_last_trans(trans, inode);
4708 BTRFS_I(dir)->index_cnt--;
4709 btrfs_free_path(path);
4711 return ERR_PTR(ret);
4714 static inline u8 btrfs_inode_type(struct inode *inode)
4716 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4720 * utility function to add 'inode' into 'parent_inode' with
4721 * a give name and a given sequence number.
4722 * if 'add_backref' is true, also insert a backref from the
4723 * inode to the parent directory.
4725 int btrfs_add_link(struct btrfs_trans_handle *trans,
4726 struct inode *parent_inode, struct inode *inode,
4727 const char *name, int name_len, int add_backref, u64 index)
4730 struct btrfs_key key;
4731 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4732 u64 ino = btrfs_ino(inode);
4733 u64 parent_ino = btrfs_ino(parent_inode);
4735 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4736 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4739 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4743 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4744 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4745 key.objectid, root->root_key.objectid,
4746 parent_ino, index, name, name_len);
4747 } else if (add_backref) {
4748 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4752 /* Nothing to clean up yet */
4756 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4758 btrfs_inode_type(inode), index);
4762 btrfs_abort_transaction(trans, root, ret);
4766 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4768 inode_inc_iversion(parent_inode);
4769 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4770 ret = btrfs_update_inode(trans, root, parent_inode);
4772 btrfs_abort_transaction(trans, root, ret);
4776 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4779 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4780 key.objectid, root->root_key.objectid,
4781 parent_ino, &local_index, name, name_len);
4783 } else if (add_backref) {
4787 err = btrfs_del_inode_ref(trans, root, name, name_len,
4788 ino, parent_ino, &local_index);
4793 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4794 struct inode *dir, struct dentry *dentry,
4795 struct inode *inode, int backref, u64 index)
4797 int err = btrfs_add_link(trans, dir, inode,
4798 dentry->d_name.name, dentry->d_name.len,
4805 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4806 umode_t mode, dev_t rdev)
4808 struct btrfs_trans_handle *trans;
4809 struct btrfs_root *root = BTRFS_I(dir)->root;
4810 struct inode *inode = NULL;
4814 unsigned long nr = 0;
4817 if (!new_valid_dev(rdev))
4821 * 2 for inode item and ref
4823 * 1 for xattr if selinux is on
4825 trans = btrfs_start_transaction(root, 5);
4827 return PTR_ERR(trans);
4829 err = btrfs_find_free_ino(root, &objectid);
4833 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4834 dentry->d_name.len, btrfs_ino(dir), objectid,
4836 if (IS_ERR(inode)) {
4837 err = PTR_ERR(inode);
4841 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4848 * If the active LSM wants to access the inode during
4849 * d_instantiate it needs these. Smack checks to see
4850 * if the filesystem supports xattrs by looking at the
4854 inode->i_op = &btrfs_special_inode_operations;
4855 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4859 init_special_inode(inode, inode->i_mode, rdev);
4860 btrfs_update_inode(trans, root, inode);
4861 d_instantiate(dentry, inode);
4864 nr = trans->blocks_used;
4865 btrfs_end_transaction(trans, root);
4866 btrfs_btree_balance_dirty(root, nr);
4868 inode_dec_link_count(inode);
4874 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4875 umode_t mode, struct nameidata *nd)
4877 struct btrfs_trans_handle *trans;
4878 struct btrfs_root *root = BTRFS_I(dir)->root;
4879 struct inode *inode = NULL;
4882 unsigned long nr = 0;
4887 * 2 for inode item and ref
4889 * 1 for xattr if selinux is on
4891 trans = btrfs_start_transaction(root, 5);
4893 return PTR_ERR(trans);
4895 err = btrfs_find_free_ino(root, &objectid);
4899 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4900 dentry->d_name.len, btrfs_ino(dir), objectid,
4902 if (IS_ERR(inode)) {
4903 err = PTR_ERR(inode);
4907 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4914 * If the active LSM wants to access the inode during
4915 * d_instantiate it needs these. Smack checks to see
4916 * if the filesystem supports xattrs by looking at the
4919 inode->i_fop = &btrfs_file_operations;
4920 inode->i_op = &btrfs_file_inode_operations;
4922 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4926 inode->i_mapping->a_ops = &btrfs_aops;
4927 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4928 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4929 d_instantiate(dentry, inode);
4932 nr = trans->blocks_used;
4933 btrfs_end_transaction(trans, root);
4935 inode_dec_link_count(inode);
4938 btrfs_btree_balance_dirty(root, nr);
4942 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4943 struct dentry *dentry)
4945 struct btrfs_trans_handle *trans;
4946 struct btrfs_root *root = BTRFS_I(dir)->root;
4947 struct inode *inode = old_dentry->d_inode;
4949 unsigned long nr = 0;
4953 /* do not allow sys_link's with other subvols of the same device */
4954 if (root->objectid != BTRFS_I(inode)->root->objectid)
4957 if (inode->i_nlink == ~0U)
4960 err = btrfs_set_inode_index(dir, &index);
4965 * 2 items for inode and inode ref
4966 * 2 items for dir items
4967 * 1 item for parent inode
4969 trans = btrfs_start_transaction(root, 5);
4970 if (IS_ERR(trans)) {
4971 err = PTR_ERR(trans);
4975 btrfs_inc_nlink(inode);
4976 inode_inc_iversion(inode);
4977 inode->i_ctime = CURRENT_TIME;
4980 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4985 struct dentry *parent = dentry->d_parent;
4986 err = btrfs_update_inode(trans, root, inode);
4989 d_instantiate(dentry, inode);
4990 btrfs_log_new_name(trans, inode, NULL, parent);
4993 nr = trans->blocks_used;
4994 btrfs_end_transaction(trans, root);
4997 inode_dec_link_count(inode);
5000 btrfs_btree_balance_dirty(root, nr);
5004 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5006 struct inode *inode = NULL;
5007 struct btrfs_trans_handle *trans;
5008 struct btrfs_root *root = BTRFS_I(dir)->root;
5010 int drop_on_err = 0;
5013 unsigned long nr = 1;
5016 * 2 items for inode and ref
5017 * 2 items for dir items
5018 * 1 for xattr if selinux is on
5020 trans = btrfs_start_transaction(root, 5);
5022 return PTR_ERR(trans);
5024 err = btrfs_find_free_ino(root, &objectid);
5028 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5029 dentry->d_name.len, btrfs_ino(dir), objectid,
5030 S_IFDIR | mode, &index);
5031 if (IS_ERR(inode)) {
5032 err = PTR_ERR(inode);
5038 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5042 inode->i_op = &btrfs_dir_inode_operations;
5043 inode->i_fop = &btrfs_dir_file_operations;
5045 btrfs_i_size_write(inode, 0);
5046 err = btrfs_update_inode(trans, root, inode);
5050 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5051 dentry->d_name.len, 0, index);
5055 d_instantiate(dentry, inode);
5059 nr = trans->blocks_used;
5060 btrfs_end_transaction(trans, root);
5063 btrfs_btree_balance_dirty(root, nr);
5067 /* helper for btfs_get_extent. Given an existing extent in the tree,
5068 * and an extent that you want to insert, deal with overlap and insert
5069 * the new extent into the tree.
5071 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5072 struct extent_map *existing,
5073 struct extent_map *em,
5074 u64 map_start, u64 map_len)
5078 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5079 start_diff = map_start - em->start;
5080 em->start = map_start;
5082 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5083 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5084 em->block_start += start_diff;
5085 em->block_len -= start_diff;
5087 return add_extent_mapping(em_tree, em);
5090 static noinline int uncompress_inline(struct btrfs_path *path,
5091 struct inode *inode, struct page *page,
5092 size_t pg_offset, u64 extent_offset,
5093 struct btrfs_file_extent_item *item)
5096 struct extent_buffer *leaf = path->nodes[0];
5099 unsigned long inline_size;
5103 WARN_ON(pg_offset != 0);
5104 compress_type = btrfs_file_extent_compression(leaf, item);
5105 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5106 inline_size = btrfs_file_extent_inline_item_len(leaf,
5107 btrfs_item_nr(leaf, path->slots[0]));
5108 tmp = kmalloc(inline_size, GFP_NOFS);
5111 ptr = btrfs_file_extent_inline_start(item);
5113 read_extent_buffer(leaf, tmp, ptr, inline_size);
5115 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5116 ret = btrfs_decompress(compress_type, tmp, page,
5117 extent_offset, inline_size, max_size);
5119 char *kaddr = kmap_atomic(page);
5120 unsigned long copy_size = min_t(u64,
5121 PAGE_CACHE_SIZE - pg_offset,
5122 max_size - extent_offset);
5123 memset(kaddr + pg_offset, 0, copy_size);
5124 kunmap_atomic(kaddr);
5131 * a bit scary, this does extent mapping from logical file offset to the disk.
5132 * the ugly parts come from merging extents from the disk with the in-ram
5133 * representation. This gets more complex because of the data=ordered code,
5134 * where the in-ram extents might be locked pending data=ordered completion.
5136 * This also copies inline extents directly into the page.
5139 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5140 size_t pg_offset, u64 start, u64 len,
5146 u64 extent_start = 0;
5148 u64 objectid = btrfs_ino(inode);
5150 struct btrfs_path *path = NULL;
5151 struct btrfs_root *root = BTRFS_I(inode)->root;
5152 struct btrfs_file_extent_item *item;
5153 struct extent_buffer *leaf;
5154 struct btrfs_key found_key;
5155 struct extent_map *em = NULL;
5156 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5157 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5158 struct btrfs_trans_handle *trans = NULL;
5162 read_lock(&em_tree->lock);
5163 em = lookup_extent_mapping(em_tree, start, len);
5165 em->bdev = root->fs_info->fs_devices->latest_bdev;
5166 read_unlock(&em_tree->lock);
5169 if (em->start > start || em->start + em->len <= start)
5170 free_extent_map(em);
5171 else if (em->block_start == EXTENT_MAP_INLINE && page)
5172 free_extent_map(em);
5176 em = alloc_extent_map();
5181 em->bdev = root->fs_info->fs_devices->latest_bdev;
5182 em->start = EXTENT_MAP_HOLE;
5183 em->orig_start = EXTENT_MAP_HOLE;
5185 em->block_len = (u64)-1;
5188 path = btrfs_alloc_path();
5194 * Chances are we'll be called again, so go ahead and do
5200 ret = btrfs_lookup_file_extent(trans, root, path,
5201 objectid, start, trans != NULL);
5208 if (path->slots[0] == 0)
5213 leaf = path->nodes[0];
5214 item = btrfs_item_ptr(leaf, path->slots[0],
5215 struct btrfs_file_extent_item);
5216 /* are we inside the extent that was found? */
5217 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5218 found_type = btrfs_key_type(&found_key);
5219 if (found_key.objectid != objectid ||
5220 found_type != BTRFS_EXTENT_DATA_KEY) {
5224 found_type = btrfs_file_extent_type(leaf, item);
5225 extent_start = found_key.offset;
5226 compress_type = btrfs_file_extent_compression(leaf, item);
5227 if (found_type == BTRFS_FILE_EXTENT_REG ||
5228 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5229 extent_end = extent_start +
5230 btrfs_file_extent_num_bytes(leaf, item);
5231 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5233 size = btrfs_file_extent_inline_len(leaf, item);
5234 extent_end = (extent_start + size + root->sectorsize - 1) &
5235 ~((u64)root->sectorsize - 1);
5238 if (start >= extent_end) {
5240 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5241 ret = btrfs_next_leaf(root, path);
5248 leaf = path->nodes[0];
5250 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5251 if (found_key.objectid != objectid ||
5252 found_key.type != BTRFS_EXTENT_DATA_KEY)
5254 if (start + len <= found_key.offset)
5257 em->len = found_key.offset - start;
5261 if (found_type == BTRFS_FILE_EXTENT_REG ||
5262 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5263 em->start = extent_start;
5264 em->len = extent_end - extent_start;
5265 em->orig_start = extent_start -
5266 btrfs_file_extent_offset(leaf, item);
5267 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5269 em->block_start = EXTENT_MAP_HOLE;
5272 if (compress_type != BTRFS_COMPRESS_NONE) {
5273 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5274 em->compress_type = compress_type;
5275 em->block_start = bytenr;
5276 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5279 bytenr += btrfs_file_extent_offset(leaf, item);
5280 em->block_start = bytenr;
5281 em->block_len = em->len;
5282 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5283 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5286 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5290 size_t extent_offset;
5293 em->block_start = EXTENT_MAP_INLINE;
5294 if (!page || create) {
5295 em->start = extent_start;
5296 em->len = extent_end - extent_start;
5300 size = btrfs_file_extent_inline_len(leaf, item);
5301 extent_offset = page_offset(page) + pg_offset - extent_start;
5302 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5303 size - extent_offset);
5304 em->start = extent_start + extent_offset;
5305 em->len = (copy_size + root->sectorsize - 1) &
5306 ~((u64)root->sectorsize - 1);
5307 em->orig_start = EXTENT_MAP_INLINE;
5308 if (compress_type) {
5309 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5310 em->compress_type = compress_type;
5312 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5313 if (create == 0 && !PageUptodate(page)) {
5314 if (btrfs_file_extent_compression(leaf, item) !=
5315 BTRFS_COMPRESS_NONE) {
5316 ret = uncompress_inline(path, inode, page,
5318 extent_offset, item);
5319 BUG_ON(ret); /* -ENOMEM */
5322 read_extent_buffer(leaf, map + pg_offset, ptr,
5324 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5325 memset(map + pg_offset + copy_size, 0,
5326 PAGE_CACHE_SIZE - pg_offset -
5331 flush_dcache_page(page);
5332 } else if (create && PageUptodate(page)) {
5336 free_extent_map(em);
5339 btrfs_release_path(path);
5340 trans = btrfs_join_transaction(root);
5343 return ERR_CAST(trans);
5347 write_extent_buffer(leaf, map + pg_offset, ptr,
5350 btrfs_mark_buffer_dirty(leaf);
5352 set_extent_uptodate(io_tree, em->start,
5353 extent_map_end(em) - 1, NULL, GFP_NOFS);
5356 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5363 em->block_start = EXTENT_MAP_HOLE;
5364 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5366 btrfs_release_path(path);
5367 if (em->start > start || extent_map_end(em) <= start) {
5368 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5369 "[%llu %llu]\n", (unsigned long long)em->start,
5370 (unsigned long long)em->len,
5371 (unsigned long long)start,
5372 (unsigned long long)len);
5378 write_lock(&em_tree->lock);
5379 ret = add_extent_mapping(em_tree, em);
5380 /* it is possible that someone inserted the extent into the tree
5381 * while we had the lock dropped. It is also possible that
5382 * an overlapping map exists in the tree
5384 if (ret == -EEXIST) {
5385 struct extent_map *existing;
5389 existing = lookup_extent_mapping(em_tree, start, len);
5390 if (existing && (existing->start > start ||
5391 existing->start + existing->len <= start)) {
5392 free_extent_map(existing);
5396 existing = lookup_extent_mapping(em_tree, em->start,
5399 err = merge_extent_mapping(em_tree, existing,
5402 free_extent_map(existing);
5404 free_extent_map(em);
5409 free_extent_map(em);
5413 free_extent_map(em);
5418 write_unlock(&em_tree->lock);
5421 trace_btrfs_get_extent(root, em);
5424 btrfs_free_path(path);
5426 ret = btrfs_end_transaction(trans, root);
5431 free_extent_map(em);
5432 return ERR_PTR(err);
5434 BUG_ON(!em); /* Error is always set */
5438 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5439 size_t pg_offset, u64 start, u64 len,
5442 struct extent_map *em;
5443 struct extent_map *hole_em = NULL;
5444 u64 range_start = start;
5450 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5455 * if our em maps to a hole, there might
5456 * actually be delalloc bytes behind it
5458 if (em->block_start != EXTENT_MAP_HOLE)
5464 /* check to see if we've wrapped (len == -1 or similar) */
5473 /* ok, we didn't find anything, lets look for delalloc */
5474 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5475 end, len, EXTENT_DELALLOC, 1);
5476 found_end = range_start + found;
5477 if (found_end < range_start)
5478 found_end = (u64)-1;
5481 * we didn't find anything useful, return
5482 * the original results from get_extent()
5484 if (range_start > end || found_end <= start) {
5490 /* adjust the range_start to make sure it doesn't
5491 * go backwards from the start they passed in
5493 range_start = max(start,range_start);
5494 found = found_end - range_start;
5497 u64 hole_start = start;
5500 em = alloc_extent_map();
5506 * when btrfs_get_extent can't find anything it
5507 * returns one huge hole
5509 * make sure what it found really fits our range, and
5510 * adjust to make sure it is based on the start from
5514 u64 calc_end = extent_map_end(hole_em);
5516 if (calc_end <= start || (hole_em->start > end)) {
5517 free_extent_map(hole_em);
5520 hole_start = max(hole_em->start, start);
5521 hole_len = calc_end - hole_start;
5525 if (hole_em && range_start > hole_start) {
5526 /* our hole starts before our delalloc, so we
5527 * have to return just the parts of the hole
5528 * that go until the delalloc starts
5530 em->len = min(hole_len,
5531 range_start - hole_start);
5532 em->start = hole_start;
5533 em->orig_start = hole_start;
5535 * don't adjust block start at all,
5536 * it is fixed at EXTENT_MAP_HOLE
5538 em->block_start = hole_em->block_start;
5539 em->block_len = hole_len;
5541 em->start = range_start;
5543 em->orig_start = range_start;
5544 em->block_start = EXTENT_MAP_DELALLOC;
5545 em->block_len = found;
5547 } else if (hole_em) {
5552 free_extent_map(hole_em);
5554 free_extent_map(em);
5555 return ERR_PTR(err);
5560 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5561 struct extent_map *em,
5564 struct btrfs_root *root = BTRFS_I(inode)->root;
5565 struct btrfs_trans_handle *trans;
5566 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5567 struct btrfs_key ins;
5570 bool insert = false;
5573 * Ok if the extent map we looked up is a hole and is for the exact
5574 * range we want, there is no reason to allocate a new one, however if
5575 * it is not right then we need to free this one and drop the cache for
5578 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5580 free_extent_map(em);
5583 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5586 trans = btrfs_join_transaction(root);
5588 return ERR_CAST(trans);
5590 if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5591 btrfs_add_inode_defrag(trans, inode);
5593 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5595 alloc_hint = get_extent_allocation_hint(inode, start, len);
5596 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5597 alloc_hint, &ins, 1);
5604 em = alloc_extent_map();
5606 em = ERR_PTR(-ENOMEM);
5612 em->orig_start = em->start;
5613 em->len = ins.offset;
5615 em->block_start = ins.objectid;
5616 em->block_len = ins.offset;
5617 em->bdev = root->fs_info->fs_devices->latest_bdev;
5620 * We need to do this because if we're using the original em we searched
5621 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5624 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5627 write_lock(&em_tree->lock);
5628 ret = add_extent_mapping(em_tree, em);
5629 write_unlock(&em_tree->lock);
5632 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5635 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5636 ins.offset, ins.offset, 0);
5638 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5642 btrfs_end_transaction(trans, root);
5647 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5648 * block must be cow'd
5650 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5651 struct inode *inode, u64 offset, u64 len)
5653 struct btrfs_path *path;
5655 struct extent_buffer *leaf;
5656 struct btrfs_root *root = BTRFS_I(inode)->root;
5657 struct btrfs_file_extent_item *fi;
5658 struct btrfs_key key;
5666 path = btrfs_alloc_path();
5670 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5675 slot = path->slots[0];
5678 /* can't find the item, must cow */
5685 leaf = path->nodes[0];
5686 btrfs_item_key_to_cpu(leaf, &key, slot);
5687 if (key.objectid != btrfs_ino(inode) ||
5688 key.type != BTRFS_EXTENT_DATA_KEY) {
5689 /* not our file or wrong item type, must cow */
5693 if (key.offset > offset) {
5694 /* Wrong offset, must cow */
5698 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5699 found_type = btrfs_file_extent_type(leaf, fi);
5700 if (found_type != BTRFS_FILE_EXTENT_REG &&
5701 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5702 /* not a regular extent, must cow */
5705 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5706 backref_offset = btrfs_file_extent_offset(leaf, fi);
5708 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5709 if (extent_end < offset + len) {
5710 /* extent doesn't include our full range, must cow */
5714 if (btrfs_extent_readonly(root, disk_bytenr))
5718 * look for other files referencing this extent, if we
5719 * find any we must cow
5721 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5722 key.offset - backref_offset, disk_bytenr))
5726 * adjust disk_bytenr and num_bytes to cover just the bytes
5727 * in this extent we are about to write. If there
5728 * are any csums in that range we have to cow in order
5729 * to keep the csums correct
5731 disk_bytenr += backref_offset;
5732 disk_bytenr += offset - key.offset;
5733 num_bytes = min(offset + len, extent_end) - offset;
5734 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5737 * all of the above have passed, it is safe to overwrite this extent
5742 btrfs_free_path(path);
5746 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5747 struct buffer_head *bh_result, int create)
5749 struct extent_map *em;
5750 struct btrfs_root *root = BTRFS_I(inode)->root;
5751 u64 start = iblock << inode->i_blkbits;
5752 u64 len = bh_result->b_size;
5753 struct btrfs_trans_handle *trans;
5755 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5760 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5761 * io. INLINE is special, and we could probably kludge it in here, but
5762 * it's still buffered so for safety lets just fall back to the generic
5765 * For COMPRESSED we _have_ to read the entire extent in so we can
5766 * decompress it, so there will be buffering required no matter what we
5767 * do, so go ahead and fallback to buffered.
5769 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5770 * to buffered IO. Don't blame me, this is the price we pay for using
5773 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5774 em->block_start == EXTENT_MAP_INLINE) {
5775 free_extent_map(em);
5779 /* Just a good old fashioned hole, return */
5780 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5781 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5782 free_extent_map(em);
5783 /* DIO will do one hole at a time, so just unlock a sector */
5784 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5785 start + root->sectorsize - 1);
5790 * We don't allocate a new extent in the following cases
5792 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5794 * 2) The extent is marked as PREALLOC. We're good to go here and can
5795 * just use the extent.
5799 len = em->len - (start - em->start);
5803 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5804 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5805 em->block_start != EXTENT_MAP_HOLE)) {
5810 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5811 type = BTRFS_ORDERED_PREALLOC;
5813 type = BTRFS_ORDERED_NOCOW;
5814 len = min(len, em->len - (start - em->start));
5815 block_start = em->block_start + (start - em->start);
5818 * we're not going to log anything, but we do need
5819 * to make sure the current transaction stays open
5820 * while we look for nocow cross refs
5822 trans = btrfs_join_transaction(root);
5826 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5827 ret = btrfs_add_ordered_extent_dio(inode, start,
5828 block_start, len, len, type);
5829 btrfs_end_transaction(trans, root);
5831 free_extent_map(em);
5836 btrfs_end_transaction(trans, root);
5840 * this will cow the extent, reset the len in case we changed
5843 len = bh_result->b_size;
5844 em = btrfs_new_extent_direct(inode, em, start, len);
5847 len = min(len, em->len - (start - em->start));
5849 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5850 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5853 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5855 bh_result->b_size = len;
5856 bh_result->b_bdev = em->bdev;
5857 set_buffer_mapped(bh_result);
5858 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5859 set_buffer_new(bh_result);
5861 free_extent_map(em);
5866 struct btrfs_dio_private {
5867 struct inode *inode;
5874 /* number of bios pending for this dio */
5875 atomic_t pending_bios;
5880 struct bio *orig_bio;
5883 static void btrfs_endio_direct_read(struct bio *bio, int err)
5885 struct btrfs_dio_private *dip = bio->bi_private;
5886 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5887 struct bio_vec *bvec = bio->bi_io_vec;
5888 struct inode *inode = dip->inode;
5889 struct btrfs_root *root = BTRFS_I(inode)->root;
5891 u32 *private = dip->csums;
5893 start = dip->logical_offset;
5895 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5896 struct page *page = bvec->bv_page;
5899 unsigned long flags;
5901 local_irq_save(flags);
5902 kaddr = kmap_atomic(page);
5903 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5904 csum, bvec->bv_len);
5905 btrfs_csum_final(csum, (char *)&csum);
5906 kunmap_atomic(kaddr);
5907 local_irq_restore(flags);
5909 flush_dcache_page(bvec->bv_page);
5910 if (csum != *private) {
5911 printk(KERN_ERR "btrfs csum failed ino %llu off"
5912 " %llu csum %u private %u\n",
5913 (unsigned long long)btrfs_ino(inode),
5914 (unsigned long long)start,
5920 start += bvec->bv_len;
5923 } while (bvec <= bvec_end);
5925 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5926 dip->logical_offset + dip->bytes - 1);
5927 bio->bi_private = dip->private;
5932 /* If we had a csum failure make sure to clear the uptodate flag */
5934 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5935 dio_end_io(bio, err);
5938 static void btrfs_endio_direct_write(struct bio *bio, int err)
5940 struct btrfs_dio_private *dip = bio->bi_private;
5941 struct inode *inode = dip->inode;
5942 struct btrfs_root *root = BTRFS_I(inode)->root;
5943 struct btrfs_ordered_extent *ordered = NULL;
5944 u64 ordered_offset = dip->logical_offset;
5945 u64 ordered_bytes = dip->bytes;
5951 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5953 ordered_bytes, !err);
5957 ordered->work.func = finish_ordered_fn;
5958 ordered->work.flags = 0;
5959 btrfs_queue_worker(&root->fs_info->endio_write_workers,
5963 * our bio might span multiple ordered extents. If we haven't
5964 * completed the accounting for the whole dio, go back and try again
5966 if (ordered_offset < dip->logical_offset + dip->bytes) {
5967 ordered_bytes = dip->logical_offset + dip->bytes -
5973 bio->bi_private = dip->private;
5977 /* If we had an error make sure to clear the uptodate flag */
5979 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5980 dio_end_io(bio, err);
5983 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5984 struct bio *bio, int mirror_num,
5985 unsigned long bio_flags, u64 offset)
5988 struct btrfs_root *root = BTRFS_I(inode)->root;
5989 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5990 BUG_ON(ret); /* -ENOMEM */
5994 static void btrfs_end_dio_bio(struct bio *bio, int err)
5996 struct btrfs_dio_private *dip = bio->bi_private;
5999 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6000 "sector %#Lx len %u err no %d\n",
6001 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6002 (unsigned long long)bio->bi_sector, bio->bi_size, err);
6006 * before atomic variable goto zero, we must make sure
6007 * dip->errors is perceived to be set.
6009 smp_mb__before_atomic_dec();
6012 /* if there are more bios still pending for this dio, just exit */
6013 if (!atomic_dec_and_test(&dip->pending_bios))
6017 bio_io_error(dip->orig_bio);
6019 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6020 bio_endio(dip->orig_bio, 0);
6026 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6027 u64 first_sector, gfp_t gfp_flags)
6029 int nr_vecs = bio_get_nr_vecs(bdev);
6030 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6033 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6034 int rw, u64 file_offset, int skip_sum,
6035 u32 *csums, int async_submit)
6037 int write = rw & REQ_WRITE;
6038 struct btrfs_root *root = BTRFS_I(inode)->root;
6044 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6052 if (write && async_submit) {
6053 ret = btrfs_wq_submit_bio(root->fs_info,
6054 inode, rw, bio, 0, 0,
6056 __btrfs_submit_bio_start_direct_io,
6057 __btrfs_submit_bio_done);
6061 * If we aren't doing async submit, calculate the csum of the
6064 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6067 } else if (!skip_sum) {
6068 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
6069 file_offset, csums);
6075 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6081 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6084 struct inode *inode = dip->inode;
6085 struct btrfs_root *root = BTRFS_I(inode)->root;
6086 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6088 struct bio *orig_bio = dip->orig_bio;
6089 struct bio_vec *bvec = orig_bio->bi_io_vec;
6090 u64 start_sector = orig_bio->bi_sector;
6091 u64 file_offset = dip->logical_offset;
6095 u32 *csums = dip->csums;
6097 int async_submit = 0;
6098 int write = rw & REQ_WRITE;
6100 map_length = orig_bio->bi_size;
6101 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6102 &map_length, NULL, 0);
6108 if (map_length >= orig_bio->bi_size) {
6114 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6117 bio->bi_private = dip;
6118 bio->bi_end_io = btrfs_end_dio_bio;
6119 atomic_inc(&dip->pending_bios);
6121 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6122 if (unlikely(map_length < submit_len + bvec->bv_len ||
6123 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6124 bvec->bv_offset) < bvec->bv_len)) {
6126 * inc the count before we submit the bio so
6127 * we know the end IO handler won't happen before
6128 * we inc the count. Otherwise, the dip might get freed
6129 * before we're done setting it up
6131 atomic_inc(&dip->pending_bios);
6132 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6133 file_offset, skip_sum,
6134 csums, async_submit);
6137 atomic_dec(&dip->pending_bios);
6141 /* Write's use the ordered csums */
6142 if (!write && !skip_sum)
6143 csums = csums + nr_pages;
6144 start_sector += submit_len >> 9;
6145 file_offset += submit_len;
6150 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6151 start_sector, GFP_NOFS);
6154 bio->bi_private = dip;
6155 bio->bi_end_io = btrfs_end_dio_bio;
6157 map_length = orig_bio->bi_size;
6158 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6159 &map_length, NULL, 0);
6165 submit_len += bvec->bv_len;
6172 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6173 csums, async_submit);
6181 * before atomic variable goto zero, we must
6182 * make sure dip->errors is perceived to be set.
6184 smp_mb__before_atomic_dec();
6185 if (atomic_dec_and_test(&dip->pending_bios))
6186 bio_io_error(dip->orig_bio);
6188 /* bio_end_io() will handle error, so we needn't return it */
6192 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6195 struct btrfs_root *root = BTRFS_I(inode)->root;
6196 struct btrfs_dio_private *dip;
6197 struct bio_vec *bvec = bio->bi_io_vec;
6199 int write = rw & REQ_WRITE;
6202 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6204 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6211 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6212 if (!write && !skip_sum) {
6213 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6221 dip->private = bio->bi_private;
6223 dip->logical_offset = file_offset;
6227 dip->bytes += bvec->bv_len;
6229 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6231 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6232 bio->bi_private = dip;
6234 dip->orig_bio = bio;
6235 atomic_set(&dip->pending_bios, 0);
6238 bio->bi_end_io = btrfs_endio_direct_write;
6240 bio->bi_end_io = btrfs_endio_direct_read;
6242 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6247 * If this is a write, we need to clean up the reserved space and kill
6248 * the ordered extent.
6251 struct btrfs_ordered_extent *ordered;
6252 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6253 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6254 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6255 btrfs_free_reserved_extent(root, ordered->start,
6257 btrfs_put_ordered_extent(ordered);
6258 btrfs_put_ordered_extent(ordered);
6260 bio_endio(bio, ret);
6263 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6264 const struct iovec *iov, loff_t offset,
6265 unsigned long nr_segs)
6271 unsigned blocksize_mask = root->sectorsize - 1;
6272 ssize_t retval = -EINVAL;
6273 loff_t end = offset;
6275 if (offset & blocksize_mask)
6278 /* Check the memory alignment. Blocks cannot straddle pages */
6279 for (seg = 0; seg < nr_segs; seg++) {
6280 addr = (unsigned long)iov[seg].iov_base;
6281 size = iov[seg].iov_len;
6283 if ((addr & blocksize_mask) || (size & blocksize_mask))
6286 /* If this is a write we don't need to check anymore */
6291 * Check to make sure we don't have duplicate iov_base's in this
6292 * iovec, if so return EINVAL, otherwise we'll get csum errors
6293 * when reading back.
6295 for (i = seg + 1; i < nr_segs; i++) {
6296 if (iov[seg].iov_base == iov[i].iov_base)
6304 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6305 const struct iovec *iov, loff_t offset,
6306 unsigned long nr_segs)
6308 struct file *file = iocb->ki_filp;
6309 struct inode *inode = file->f_mapping->host;
6310 struct btrfs_ordered_extent *ordered;
6311 struct extent_state *cached_state = NULL;
6312 u64 lockstart, lockend;
6314 int writing = rw & WRITE;
6316 size_t count = iov_length(iov, nr_segs);
6318 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6324 lockend = offset + count - 1;
6327 ret = btrfs_delalloc_reserve_space(inode, count);
6333 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6336 * We're concerned with the entire range that we're going to be
6337 * doing DIO to, so we need to make sure theres no ordered
6338 * extents in this range.
6340 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6341 lockend - lockstart + 1);
6344 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6345 &cached_state, GFP_NOFS);
6346 btrfs_start_ordered_extent(inode, ordered, 1);
6347 btrfs_put_ordered_extent(ordered);
6352 * we don't use btrfs_set_extent_delalloc because we don't want
6353 * the dirty or uptodate bits
6356 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6357 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6358 EXTENT_DELALLOC, NULL, &cached_state,
6361 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6362 lockend, EXTENT_LOCKED | write_bits,
6363 1, 0, &cached_state, GFP_NOFS);
6368 free_extent_state(cached_state);
6369 cached_state = NULL;
6371 ret = __blockdev_direct_IO(rw, iocb, inode,
6372 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6373 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6374 btrfs_submit_direct, 0);
6376 if (ret < 0 && ret != -EIOCBQUEUED) {
6377 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6378 offset + iov_length(iov, nr_segs) - 1,
6379 EXTENT_LOCKED | write_bits, 1, 0,
6380 &cached_state, GFP_NOFS);
6381 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6383 * We're falling back to buffered, unlock the section we didn't
6386 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6387 offset + iov_length(iov, nr_segs) - 1,
6388 EXTENT_LOCKED | write_bits, 1, 0,
6389 &cached_state, GFP_NOFS);
6392 free_extent_state(cached_state);
6396 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6397 __u64 start, __u64 len)
6399 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6402 int btrfs_readpage(struct file *file, struct page *page)
6404 struct extent_io_tree *tree;
6405 tree = &BTRFS_I(page->mapping->host)->io_tree;
6406 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6409 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6411 struct extent_io_tree *tree;
6414 if (current->flags & PF_MEMALLOC) {
6415 redirty_page_for_writepage(wbc, page);
6419 tree = &BTRFS_I(page->mapping->host)->io_tree;
6420 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6423 int btrfs_writepages(struct address_space *mapping,
6424 struct writeback_control *wbc)
6426 struct extent_io_tree *tree;
6428 tree = &BTRFS_I(mapping->host)->io_tree;
6429 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6433 btrfs_readpages(struct file *file, struct address_space *mapping,
6434 struct list_head *pages, unsigned nr_pages)
6436 struct extent_io_tree *tree;
6437 tree = &BTRFS_I(mapping->host)->io_tree;
6438 return extent_readpages(tree, mapping, pages, nr_pages,
6441 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6443 struct extent_io_tree *tree;
6444 struct extent_map_tree *map;
6447 tree = &BTRFS_I(page->mapping->host)->io_tree;
6448 map = &BTRFS_I(page->mapping->host)->extent_tree;
6449 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6451 ClearPagePrivate(page);
6452 set_page_private(page, 0);
6453 page_cache_release(page);
6458 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6460 if (PageWriteback(page) || PageDirty(page))
6462 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6465 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6467 struct inode *inode = page->mapping->host;
6468 struct extent_io_tree *tree;
6469 struct btrfs_ordered_extent *ordered;
6470 struct extent_state *cached_state = NULL;
6471 u64 page_start = page_offset(page);
6472 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6475 * we have the page locked, so new writeback can't start,
6476 * and the dirty bit won't be cleared while we are here.
6478 * Wait for IO on this page so that we can safely clear
6479 * the PagePrivate2 bit and do ordered accounting
6481 wait_on_page_writeback(page);
6483 tree = &BTRFS_I(inode)->io_tree;
6485 btrfs_releasepage(page, GFP_NOFS);
6488 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6489 ordered = btrfs_lookup_ordered_extent(inode,
6493 * IO on this page will never be started, so we need
6494 * to account for any ordered extents now
6496 clear_extent_bit(tree, page_start, page_end,
6497 EXTENT_DIRTY | EXTENT_DELALLOC |
6498 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6499 &cached_state, GFP_NOFS);
6501 * whoever cleared the private bit is responsible
6502 * for the finish_ordered_io
6504 if (TestClearPagePrivate2(page) &&
6505 btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6506 PAGE_CACHE_SIZE, 1)) {
6507 btrfs_finish_ordered_io(ordered);
6509 btrfs_put_ordered_extent(ordered);
6510 cached_state = NULL;
6511 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6513 clear_extent_bit(tree, page_start, page_end,
6514 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6515 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6516 __btrfs_releasepage(page, GFP_NOFS);
6518 ClearPageChecked(page);
6519 if (PagePrivate(page)) {
6520 ClearPagePrivate(page);
6521 set_page_private(page, 0);
6522 page_cache_release(page);
6527 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6528 * called from a page fault handler when a page is first dirtied. Hence we must
6529 * be careful to check for EOF conditions here. We set the page up correctly
6530 * for a written page which means we get ENOSPC checking when writing into
6531 * holes and correct delalloc and unwritten extent mapping on filesystems that
6532 * support these features.
6534 * We are not allowed to take the i_mutex here so we have to play games to
6535 * protect against truncate races as the page could now be beyond EOF. Because
6536 * vmtruncate() writes the inode size before removing pages, once we have the
6537 * page lock we can determine safely if the page is beyond EOF. If it is not
6538 * beyond EOF, then the page is guaranteed safe against truncation until we
6541 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6543 struct page *page = vmf->page;
6544 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6545 struct btrfs_root *root = BTRFS_I(inode)->root;
6546 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6547 struct btrfs_ordered_extent *ordered;
6548 struct extent_state *cached_state = NULL;
6550 unsigned long zero_start;
6557 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6559 ret = btrfs_update_time(vma->vm_file);
6565 else /* -ENOSPC, -EIO, etc */
6566 ret = VM_FAULT_SIGBUS;
6572 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6575 size = i_size_read(inode);
6576 page_start = page_offset(page);
6577 page_end = page_start + PAGE_CACHE_SIZE - 1;
6579 if ((page->mapping != inode->i_mapping) ||
6580 (page_start >= size)) {
6581 /* page got truncated out from underneath us */
6584 wait_on_page_writeback(page);
6586 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6587 set_page_extent_mapped(page);
6590 * we can't set the delalloc bits if there are pending ordered
6591 * extents. Drop our locks and wait for them to finish
6593 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6595 unlock_extent_cached(io_tree, page_start, page_end,
6596 &cached_state, GFP_NOFS);
6598 btrfs_start_ordered_extent(inode, ordered, 1);
6599 btrfs_put_ordered_extent(ordered);
6604 * XXX - page_mkwrite gets called every time the page is dirtied, even
6605 * if it was already dirty, so for space accounting reasons we need to
6606 * clear any delalloc bits for the range we are fixing to save. There
6607 * is probably a better way to do this, but for now keep consistent with
6608 * prepare_pages in the normal write path.
6610 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6611 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6612 0, 0, &cached_state, GFP_NOFS);
6614 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6617 unlock_extent_cached(io_tree, page_start, page_end,
6618 &cached_state, GFP_NOFS);
6619 ret = VM_FAULT_SIGBUS;
6624 /* page is wholly or partially inside EOF */
6625 if (page_start + PAGE_CACHE_SIZE > size)
6626 zero_start = size & ~PAGE_CACHE_MASK;
6628 zero_start = PAGE_CACHE_SIZE;
6630 if (zero_start != PAGE_CACHE_SIZE) {
6632 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6633 flush_dcache_page(page);
6636 ClearPageChecked(page);
6637 set_page_dirty(page);
6638 SetPageUptodate(page);
6640 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6641 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6643 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6647 return VM_FAULT_LOCKED;
6650 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6655 static int btrfs_truncate(struct inode *inode)
6657 struct btrfs_root *root = BTRFS_I(inode)->root;
6658 struct btrfs_block_rsv *rsv;
6661 struct btrfs_trans_handle *trans;
6663 u64 mask = root->sectorsize - 1;
6664 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6666 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6670 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6671 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6674 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6675 * 3 things going on here
6677 * 1) We need to reserve space for our orphan item and the space to
6678 * delete our orphan item. Lord knows we don't want to have a dangling
6679 * orphan item because we didn't reserve space to remove it.
6681 * 2) We need to reserve space to update our inode.
6683 * 3) We need to have something to cache all the space that is going to
6684 * be free'd up by the truncate operation, but also have some slack
6685 * space reserved in case it uses space during the truncate (thank you
6686 * very much snapshotting).
6688 * And we need these to all be seperate. The fact is we can use alot of
6689 * space doing the truncate, and we have no earthly idea how much space
6690 * we will use, so we need the truncate reservation to be seperate so it
6691 * doesn't end up using space reserved for updating the inode or
6692 * removing the orphan item. We also need to be able to stop the
6693 * transaction and start a new one, which means we need to be able to
6694 * update the inode several times, and we have no idea of knowing how
6695 * many times that will be, so we can't just reserve 1 item for the
6696 * entirety of the opration, so that has to be done seperately as well.
6697 * Then there is the orphan item, which does indeed need to be held on
6698 * to for the whole operation, and we need nobody to touch this reserved
6699 * space except the orphan code.
6701 * So that leaves us with
6703 * 1) root->orphan_block_rsv - for the orphan deletion.
6704 * 2) rsv - for the truncate reservation, which we will steal from the
6705 * transaction reservation.
6706 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6707 * updating the inode.
6709 rsv = btrfs_alloc_block_rsv(root);
6712 rsv->size = min_size;
6715 * 1 for the truncate slack space
6716 * 1 for the orphan item we're going to add
6717 * 1 for the orphan item deletion
6718 * 1 for updating the inode.
6720 trans = btrfs_start_transaction(root, 4);
6721 if (IS_ERR(trans)) {
6722 err = PTR_ERR(trans);
6726 /* Migrate the slack space for the truncate to our reserve */
6727 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6731 ret = btrfs_orphan_add(trans, inode);
6733 btrfs_end_transaction(trans, root);
6738 * setattr is responsible for setting the ordered_data_close flag,
6739 * but that is only tested during the last file release. That
6740 * could happen well after the next commit, leaving a great big
6741 * window where new writes may get lost if someone chooses to write
6742 * to this file after truncating to zero
6744 * The inode doesn't have any dirty data here, and so if we commit
6745 * this is a noop. If someone immediately starts writing to the inode
6746 * it is very likely we'll catch some of their writes in this
6747 * transaction, and the commit will find this file on the ordered
6748 * data list with good things to send down.
6750 * This is a best effort solution, there is still a window where
6751 * using truncate to replace the contents of the file will
6752 * end up with a zero length file after a crash.
6754 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
6755 &BTRFS_I(inode)->runtime_flags))
6756 btrfs_add_ordered_operation(trans, root, inode);
6759 ret = btrfs_block_rsv_refill(root, rsv, min_size);
6762 * This can only happen with the original transaction we
6763 * started above, every other time we shouldn't have a
6764 * transaction started yet.
6773 /* Just need the 1 for updating the inode */
6774 trans = btrfs_start_transaction(root, 1);
6775 if (IS_ERR(trans)) {
6776 ret = err = PTR_ERR(trans);
6782 trans->block_rsv = rsv;
6784 ret = btrfs_truncate_inode_items(trans, root, inode,
6786 BTRFS_EXTENT_DATA_KEY);
6787 if (ret != -EAGAIN) {
6792 trans->block_rsv = &root->fs_info->trans_block_rsv;
6793 ret = btrfs_update_inode(trans, root, inode);
6799 nr = trans->blocks_used;
6800 btrfs_end_transaction(trans, root);
6802 btrfs_btree_balance_dirty(root, nr);
6805 if (ret == 0 && inode->i_nlink > 0) {
6806 trans->block_rsv = root->orphan_block_rsv;
6807 ret = btrfs_orphan_del(trans, inode);
6810 } else if (ret && inode->i_nlink > 0) {
6812 * Failed to do the truncate, remove us from the in memory
6815 ret = btrfs_orphan_del(NULL, inode);
6819 trans->block_rsv = &root->fs_info->trans_block_rsv;
6820 ret = btrfs_update_inode(trans, root, inode);
6824 nr = trans->blocks_used;
6825 ret = btrfs_end_transaction(trans, root);
6826 btrfs_btree_balance_dirty(root, nr);
6830 btrfs_free_block_rsv(root, rsv);
6839 * create a new subvolume directory/inode (helper for the ioctl).
6841 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6842 struct btrfs_root *new_root, u64 new_dirid)
6844 struct inode *inode;
6848 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
6849 new_dirid, new_dirid,
6850 S_IFDIR | (~current_umask() & S_IRWXUGO),
6853 return PTR_ERR(inode);
6854 inode->i_op = &btrfs_dir_inode_operations;
6855 inode->i_fop = &btrfs_dir_file_operations;
6857 set_nlink(inode, 1);
6858 btrfs_i_size_write(inode, 0);
6860 err = btrfs_update_inode(trans, new_root, inode);
6866 struct inode *btrfs_alloc_inode(struct super_block *sb)
6868 struct btrfs_inode *ei;
6869 struct inode *inode;
6871 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6876 ei->space_info = NULL;
6879 ei->last_sub_trans = 0;
6880 ei->logged_trans = 0;
6881 ei->delalloc_bytes = 0;
6882 ei->disk_i_size = 0;
6885 ei->index_cnt = (u64)-1;
6886 ei->last_unlink_trans = 0;
6888 spin_lock_init(&ei->lock);
6889 ei->outstanding_extents = 0;
6890 ei->reserved_extents = 0;
6892 ei->runtime_flags = 0;
6893 ei->force_compress = BTRFS_COMPRESS_NONE;
6895 ei->delayed_node = NULL;
6897 inode = &ei->vfs_inode;
6898 extent_map_tree_init(&ei->extent_tree);
6899 extent_io_tree_init(&ei->io_tree, &inode->i_data);
6900 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6901 ei->io_tree.track_uptodate = 1;
6902 ei->io_failure_tree.track_uptodate = 1;
6903 mutex_init(&ei->log_mutex);
6904 mutex_init(&ei->delalloc_mutex);
6905 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6906 INIT_LIST_HEAD(&ei->i_orphan);
6907 INIT_LIST_HEAD(&ei->delalloc_inodes);
6908 INIT_LIST_HEAD(&ei->ordered_operations);
6909 RB_CLEAR_NODE(&ei->rb_node);
6914 static void btrfs_i_callback(struct rcu_head *head)
6916 struct inode *inode = container_of(head, struct inode, i_rcu);
6917 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6920 void btrfs_destroy_inode(struct inode *inode)
6922 struct btrfs_ordered_extent *ordered;
6923 struct btrfs_root *root = BTRFS_I(inode)->root;
6925 WARN_ON(!list_empty(&inode->i_dentry));
6926 WARN_ON(inode->i_data.nrpages);
6927 WARN_ON(BTRFS_I(inode)->outstanding_extents);
6928 WARN_ON(BTRFS_I(inode)->reserved_extents);
6929 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6930 WARN_ON(BTRFS_I(inode)->csum_bytes);
6933 * This can happen where we create an inode, but somebody else also
6934 * created the same inode and we need to destroy the one we already
6941 * Make sure we're properly removed from the ordered operation
6945 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6946 spin_lock(&root->fs_info->ordered_extent_lock);
6947 list_del_init(&BTRFS_I(inode)->ordered_operations);
6948 spin_unlock(&root->fs_info->ordered_extent_lock);
6951 spin_lock(&root->orphan_lock);
6952 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6953 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6954 (unsigned long long)btrfs_ino(inode));
6955 list_del_init(&BTRFS_I(inode)->i_orphan);
6957 spin_unlock(&root->orphan_lock);
6960 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6964 printk(KERN_ERR "btrfs found ordered "
6965 "extent %llu %llu on inode cleanup\n",
6966 (unsigned long long)ordered->file_offset,
6967 (unsigned long long)ordered->len);
6968 btrfs_remove_ordered_extent(inode, ordered);
6969 btrfs_put_ordered_extent(ordered);
6970 btrfs_put_ordered_extent(ordered);
6973 inode_tree_del(inode);
6974 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6976 btrfs_remove_delayed_node(inode);
6977 call_rcu(&inode->i_rcu, btrfs_i_callback);
6980 int btrfs_drop_inode(struct inode *inode)
6982 struct btrfs_root *root = BTRFS_I(inode)->root;
6984 if (btrfs_root_refs(&root->root_item) == 0 &&
6985 !btrfs_is_free_space_inode(root, inode))
6988 return generic_drop_inode(inode);
6991 static void init_once(void *foo)
6993 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6995 inode_init_once(&ei->vfs_inode);
6998 void btrfs_destroy_cachep(void)
7000 if (btrfs_inode_cachep)
7001 kmem_cache_destroy(btrfs_inode_cachep);
7002 if (btrfs_trans_handle_cachep)
7003 kmem_cache_destroy(btrfs_trans_handle_cachep);
7004 if (btrfs_transaction_cachep)
7005 kmem_cache_destroy(btrfs_transaction_cachep);
7006 if (btrfs_path_cachep)
7007 kmem_cache_destroy(btrfs_path_cachep);
7008 if (btrfs_free_space_cachep)
7009 kmem_cache_destroy(btrfs_free_space_cachep);
7012 int btrfs_init_cachep(void)
7014 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
7015 sizeof(struct btrfs_inode), 0,
7016 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7017 if (!btrfs_inode_cachep)
7020 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
7021 sizeof(struct btrfs_trans_handle), 0,
7022 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7023 if (!btrfs_trans_handle_cachep)
7026 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
7027 sizeof(struct btrfs_transaction), 0,
7028 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7029 if (!btrfs_transaction_cachep)
7032 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
7033 sizeof(struct btrfs_path), 0,
7034 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7035 if (!btrfs_path_cachep)
7038 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
7039 sizeof(struct btrfs_free_space), 0,
7040 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7041 if (!btrfs_free_space_cachep)
7046 btrfs_destroy_cachep();
7050 static int btrfs_getattr(struct vfsmount *mnt,
7051 struct dentry *dentry, struct kstat *stat)
7053 struct inode *inode = dentry->d_inode;
7054 u32 blocksize = inode->i_sb->s_blocksize;
7056 generic_fillattr(inode, stat);
7057 stat->dev = BTRFS_I(inode)->root->anon_dev;
7058 stat->blksize = PAGE_CACHE_SIZE;
7059 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7060 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
7065 * If a file is moved, it will inherit the cow and compression flags of the new
7068 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7070 struct btrfs_inode *b_dir = BTRFS_I(dir);
7071 struct btrfs_inode *b_inode = BTRFS_I(inode);
7073 if (b_dir->flags & BTRFS_INODE_NODATACOW)
7074 b_inode->flags |= BTRFS_INODE_NODATACOW;
7076 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7078 if (b_dir->flags & BTRFS_INODE_COMPRESS)
7079 b_inode->flags |= BTRFS_INODE_COMPRESS;
7081 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
7084 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7085 struct inode *new_dir, struct dentry *new_dentry)
7087 struct btrfs_trans_handle *trans;
7088 struct btrfs_root *root = BTRFS_I(old_dir)->root;
7089 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7090 struct inode *new_inode = new_dentry->d_inode;
7091 struct inode *old_inode = old_dentry->d_inode;
7092 struct timespec ctime = CURRENT_TIME;
7096 u64 old_ino = btrfs_ino(old_inode);
7098 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7101 /* we only allow rename subvolume link between subvolumes */
7102 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7105 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7106 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7109 if (S_ISDIR(old_inode->i_mode) && new_inode &&
7110 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7113 * we're using rename to replace one file with another.
7114 * and the replacement file is large. Start IO on it now so
7115 * we don't add too much work to the end of the transaction
7117 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7118 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7119 filemap_flush(old_inode->i_mapping);
7121 /* close the racy window with snapshot create/destroy ioctl */
7122 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7123 down_read(&root->fs_info->subvol_sem);
7125 * We want to reserve the absolute worst case amount of items. So if
7126 * both inodes are subvols and we need to unlink them then that would
7127 * require 4 item modifications, but if they are both normal inodes it
7128 * would require 5 item modifications, so we'll assume their normal
7129 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7130 * should cover the worst case number of items we'll modify.
7132 trans = btrfs_start_transaction(root, 20);
7133 if (IS_ERR(trans)) {
7134 ret = PTR_ERR(trans);
7139 btrfs_record_root_in_trans(trans, dest);
7141 ret = btrfs_set_inode_index(new_dir, &index);
7145 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7146 /* force full log commit if subvolume involved. */
7147 root->fs_info->last_trans_log_full_commit = trans->transid;
7149 ret = btrfs_insert_inode_ref(trans, dest,
7150 new_dentry->d_name.name,
7151 new_dentry->d_name.len,
7153 btrfs_ino(new_dir), index);
7157 * this is an ugly little race, but the rename is required
7158 * to make sure that if we crash, the inode is either at the
7159 * old name or the new one. pinning the log transaction lets
7160 * us make sure we don't allow a log commit to come in after
7161 * we unlink the name but before we add the new name back in.
7163 btrfs_pin_log_trans(root);
7166 * make sure the inode gets flushed if it is replacing
7169 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7170 btrfs_add_ordered_operation(trans, root, old_inode);
7172 inode_inc_iversion(old_dir);
7173 inode_inc_iversion(new_dir);
7174 inode_inc_iversion(old_inode);
7175 old_dir->i_ctime = old_dir->i_mtime = ctime;
7176 new_dir->i_ctime = new_dir->i_mtime = ctime;
7177 old_inode->i_ctime = ctime;
7179 if (old_dentry->d_parent != new_dentry->d_parent)
7180 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7182 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7183 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7184 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7185 old_dentry->d_name.name,
7186 old_dentry->d_name.len);
7188 ret = __btrfs_unlink_inode(trans, root, old_dir,
7189 old_dentry->d_inode,
7190 old_dentry->d_name.name,
7191 old_dentry->d_name.len);
7193 ret = btrfs_update_inode(trans, root, old_inode);
7196 btrfs_abort_transaction(trans, root, ret);
7201 inode_inc_iversion(new_inode);
7202 new_inode->i_ctime = CURRENT_TIME;
7203 if (unlikely(btrfs_ino(new_inode) ==
7204 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7205 root_objectid = BTRFS_I(new_inode)->location.objectid;
7206 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7208 new_dentry->d_name.name,
7209 new_dentry->d_name.len);
7210 BUG_ON(new_inode->i_nlink == 0);
7212 ret = btrfs_unlink_inode(trans, dest, new_dir,
7213 new_dentry->d_inode,
7214 new_dentry->d_name.name,
7215 new_dentry->d_name.len);
7217 if (!ret && new_inode->i_nlink == 0) {
7218 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7222 btrfs_abort_transaction(trans, root, ret);
7227 fixup_inode_flags(new_dir, old_inode);
7229 ret = btrfs_add_link(trans, new_dir, old_inode,
7230 new_dentry->d_name.name,
7231 new_dentry->d_name.len, 0, index);
7233 btrfs_abort_transaction(trans, root, ret);
7237 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7238 struct dentry *parent = new_dentry->d_parent;
7239 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7240 btrfs_end_log_trans(root);
7243 btrfs_end_transaction(trans, root);
7245 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7246 up_read(&root->fs_info->subvol_sem);
7252 * some fairly slow code that needs optimization. This walks the list
7253 * of all the inodes with pending delalloc and forces them to disk.
7255 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7257 struct list_head *head = &root->fs_info->delalloc_inodes;
7258 struct btrfs_inode *binode;
7259 struct inode *inode;
7261 if (root->fs_info->sb->s_flags & MS_RDONLY)
7264 spin_lock(&root->fs_info->delalloc_lock);
7265 while (!list_empty(head)) {
7266 binode = list_entry(head->next, struct btrfs_inode,
7268 inode = igrab(&binode->vfs_inode);
7270 list_del_init(&binode->delalloc_inodes);
7271 spin_unlock(&root->fs_info->delalloc_lock);
7273 filemap_flush(inode->i_mapping);
7275 btrfs_add_delayed_iput(inode);
7280 spin_lock(&root->fs_info->delalloc_lock);
7282 spin_unlock(&root->fs_info->delalloc_lock);
7284 /* the filemap_flush will queue IO into the worker threads, but
7285 * we have to make sure the IO is actually started and that
7286 * ordered extents get created before we return
7288 atomic_inc(&root->fs_info->async_submit_draining);
7289 while (atomic_read(&root->fs_info->nr_async_submits) ||
7290 atomic_read(&root->fs_info->async_delalloc_pages)) {
7291 wait_event(root->fs_info->async_submit_wait,
7292 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7293 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7295 atomic_dec(&root->fs_info->async_submit_draining);
7299 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7300 const char *symname)
7302 struct btrfs_trans_handle *trans;
7303 struct btrfs_root *root = BTRFS_I(dir)->root;
7304 struct btrfs_path *path;
7305 struct btrfs_key key;
7306 struct inode *inode = NULL;
7314 struct btrfs_file_extent_item *ei;
7315 struct extent_buffer *leaf;
7316 unsigned long nr = 0;
7318 name_len = strlen(symname) + 1;
7319 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7320 return -ENAMETOOLONG;
7323 * 2 items for inode item and ref
7324 * 2 items for dir items
7325 * 1 item for xattr if selinux is on
7327 trans = btrfs_start_transaction(root, 5);
7329 return PTR_ERR(trans);
7331 err = btrfs_find_free_ino(root, &objectid);
7335 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7336 dentry->d_name.len, btrfs_ino(dir), objectid,
7337 S_IFLNK|S_IRWXUGO, &index);
7338 if (IS_ERR(inode)) {
7339 err = PTR_ERR(inode);
7343 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7350 * If the active LSM wants to access the inode during
7351 * d_instantiate it needs these. Smack checks to see
7352 * if the filesystem supports xattrs by looking at the
7355 inode->i_fop = &btrfs_file_operations;
7356 inode->i_op = &btrfs_file_inode_operations;
7358 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7362 inode->i_mapping->a_ops = &btrfs_aops;
7363 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7364 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7369 path = btrfs_alloc_path();
7375 key.objectid = btrfs_ino(inode);
7377 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7378 datasize = btrfs_file_extent_calc_inline_size(name_len);
7379 err = btrfs_insert_empty_item(trans, root, path, &key,
7383 btrfs_free_path(path);
7386 leaf = path->nodes[0];
7387 ei = btrfs_item_ptr(leaf, path->slots[0],
7388 struct btrfs_file_extent_item);
7389 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7390 btrfs_set_file_extent_type(leaf, ei,
7391 BTRFS_FILE_EXTENT_INLINE);
7392 btrfs_set_file_extent_encryption(leaf, ei, 0);
7393 btrfs_set_file_extent_compression(leaf, ei, 0);
7394 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7395 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7397 ptr = btrfs_file_extent_inline_start(ei);
7398 write_extent_buffer(leaf, symname, ptr, name_len);
7399 btrfs_mark_buffer_dirty(leaf);
7400 btrfs_free_path(path);
7402 inode->i_op = &btrfs_symlink_inode_operations;
7403 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7404 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7405 inode_set_bytes(inode, name_len);
7406 btrfs_i_size_write(inode, name_len - 1);
7407 err = btrfs_update_inode(trans, root, inode);
7413 d_instantiate(dentry, inode);
7414 nr = trans->blocks_used;
7415 btrfs_end_transaction(trans, root);
7417 inode_dec_link_count(inode);
7420 btrfs_btree_balance_dirty(root, nr);
7424 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7425 u64 start, u64 num_bytes, u64 min_size,
7426 loff_t actual_len, u64 *alloc_hint,
7427 struct btrfs_trans_handle *trans)
7429 struct btrfs_root *root = BTRFS_I(inode)->root;
7430 struct btrfs_key ins;
7431 u64 cur_offset = start;
7434 bool own_trans = true;
7438 while (num_bytes > 0) {
7440 trans = btrfs_start_transaction(root, 3);
7441 if (IS_ERR(trans)) {
7442 ret = PTR_ERR(trans);
7447 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7448 0, *alloc_hint, &ins, 1);
7451 btrfs_end_transaction(trans, root);
7455 ret = insert_reserved_file_extent(trans, inode,
7456 cur_offset, ins.objectid,
7457 ins.offset, ins.offset,
7458 ins.offset, 0, 0, 0,
7459 BTRFS_FILE_EXTENT_PREALLOC);
7461 btrfs_abort_transaction(trans, root, ret);
7463 btrfs_end_transaction(trans, root);
7466 btrfs_drop_extent_cache(inode, cur_offset,
7467 cur_offset + ins.offset -1, 0);
7469 num_bytes -= ins.offset;
7470 cur_offset += ins.offset;
7471 *alloc_hint = ins.objectid + ins.offset;
7473 inode_inc_iversion(inode);
7474 inode->i_ctime = CURRENT_TIME;
7475 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7476 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7477 (actual_len > inode->i_size) &&
7478 (cur_offset > inode->i_size)) {
7479 if (cur_offset > actual_len)
7480 i_size = actual_len;
7482 i_size = cur_offset;
7483 i_size_write(inode, i_size);
7484 btrfs_ordered_update_i_size(inode, i_size, NULL);
7487 ret = btrfs_update_inode(trans, root, inode);
7490 btrfs_abort_transaction(trans, root, ret);
7492 btrfs_end_transaction(trans, root);
7497 btrfs_end_transaction(trans, root);
7502 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7503 u64 start, u64 num_bytes, u64 min_size,
7504 loff_t actual_len, u64 *alloc_hint)
7506 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7507 min_size, actual_len, alloc_hint,
7511 int btrfs_prealloc_file_range_trans(struct inode *inode,
7512 struct btrfs_trans_handle *trans, int mode,
7513 u64 start, u64 num_bytes, u64 min_size,
7514 loff_t actual_len, u64 *alloc_hint)
7516 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7517 min_size, actual_len, alloc_hint, trans);
7520 static int btrfs_set_page_dirty(struct page *page)
7522 return __set_page_dirty_nobuffers(page);
7525 static int btrfs_permission(struct inode *inode, int mask)
7527 struct btrfs_root *root = BTRFS_I(inode)->root;
7528 umode_t mode = inode->i_mode;
7530 if (mask & MAY_WRITE &&
7531 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7532 if (btrfs_root_readonly(root))
7534 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7537 return generic_permission(inode, mask);
7540 static const struct inode_operations btrfs_dir_inode_operations = {
7541 .getattr = btrfs_getattr,
7542 .lookup = btrfs_lookup,
7543 .create = btrfs_create,
7544 .unlink = btrfs_unlink,
7546 .mkdir = btrfs_mkdir,
7547 .rmdir = btrfs_rmdir,
7548 .rename = btrfs_rename,
7549 .symlink = btrfs_symlink,
7550 .setattr = btrfs_setattr,
7551 .mknod = btrfs_mknod,
7552 .setxattr = btrfs_setxattr,
7553 .getxattr = btrfs_getxattr,
7554 .listxattr = btrfs_listxattr,
7555 .removexattr = btrfs_removexattr,
7556 .permission = btrfs_permission,
7557 .get_acl = btrfs_get_acl,
7559 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7560 .lookup = btrfs_lookup,
7561 .permission = btrfs_permission,
7562 .get_acl = btrfs_get_acl,
7565 static const struct file_operations btrfs_dir_file_operations = {
7566 .llseek = generic_file_llseek,
7567 .read = generic_read_dir,
7568 .readdir = btrfs_real_readdir,
7569 .unlocked_ioctl = btrfs_ioctl,
7570 #ifdef CONFIG_COMPAT
7571 .compat_ioctl = btrfs_ioctl,
7573 .release = btrfs_release_file,
7574 .fsync = btrfs_sync_file,
7577 static struct extent_io_ops btrfs_extent_io_ops = {
7578 .fill_delalloc = run_delalloc_range,
7579 .submit_bio_hook = btrfs_submit_bio_hook,
7580 .merge_bio_hook = btrfs_merge_bio_hook,
7581 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7582 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7583 .writepage_start_hook = btrfs_writepage_start_hook,
7584 .set_bit_hook = btrfs_set_bit_hook,
7585 .clear_bit_hook = btrfs_clear_bit_hook,
7586 .merge_extent_hook = btrfs_merge_extent_hook,
7587 .split_extent_hook = btrfs_split_extent_hook,
7591 * btrfs doesn't support the bmap operation because swapfiles
7592 * use bmap to make a mapping of extents in the file. They assume
7593 * these extents won't change over the life of the file and they
7594 * use the bmap result to do IO directly to the drive.
7596 * the btrfs bmap call would return logical addresses that aren't
7597 * suitable for IO and they also will change frequently as COW
7598 * operations happen. So, swapfile + btrfs == corruption.
7600 * For now we're avoiding this by dropping bmap.
7602 static const struct address_space_operations btrfs_aops = {
7603 .readpage = btrfs_readpage,
7604 .writepage = btrfs_writepage,
7605 .writepages = btrfs_writepages,
7606 .readpages = btrfs_readpages,
7607 .direct_IO = btrfs_direct_IO,
7608 .invalidatepage = btrfs_invalidatepage,
7609 .releasepage = btrfs_releasepage,
7610 .set_page_dirty = btrfs_set_page_dirty,
7611 .error_remove_page = generic_error_remove_page,
7614 static const struct address_space_operations btrfs_symlink_aops = {
7615 .readpage = btrfs_readpage,
7616 .writepage = btrfs_writepage,
7617 .invalidatepage = btrfs_invalidatepage,
7618 .releasepage = btrfs_releasepage,
7621 static const struct inode_operations btrfs_file_inode_operations = {
7622 .getattr = btrfs_getattr,
7623 .setattr = btrfs_setattr,
7624 .setxattr = btrfs_setxattr,
7625 .getxattr = btrfs_getxattr,
7626 .listxattr = btrfs_listxattr,
7627 .removexattr = btrfs_removexattr,
7628 .permission = btrfs_permission,
7629 .fiemap = btrfs_fiemap,
7630 .get_acl = btrfs_get_acl,
7632 static const struct inode_operations btrfs_special_inode_operations = {
7633 .getattr = btrfs_getattr,
7634 .setattr = btrfs_setattr,
7635 .permission = btrfs_permission,
7636 .setxattr = btrfs_setxattr,
7637 .getxattr = btrfs_getxattr,
7638 .listxattr = btrfs_listxattr,
7639 .removexattr = btrfs_removexattr,
7640 .get_acl = btrfs_get_acl,
7642 static const struct inode_operations btrfs_symlink_inode_operations = {
7643 .readlink = generic_readlink,
7644 .follow_link = page_follow_link_light,
7645 .put_link = page_put_link,
7646 .getattr = btrfs_getattr,
7647 .setattr = btrfs_setattr,
7648 .permission = btrfs_permission,
7649 .setxattr = btrfs_setxattr,
7650 .getxattr = btrfs_getxattr,
7651 .listxattr = btrfs_listxattr,
7652 .removexattr = btrfs_removexattr,
7653 .get_acl = btrfs_get_acl,
7656 const struct dentry_operations btrfs_dentry_operations = {
7657 .d_delete = btrfs_dentry_delete,
7658 .d_release = btrfs_dentry_release,