2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include <linux/mount.h>
42 #include <linux/btrfs.h>
43 #include <linux/blkdev.h>
47 #include "transaction.h"
48 #include "btrfs_inode.h"
49 #include "print-tree.h"
50 #include "ordered-data.h"
54 #include "compression.h"
56 #include "free-space-cache.h"
57 #include "inode-map.h"
59 struct btrfs_iget_args {
61 struct btrfs_root *root;
64 static const struct inode_operations btrfs_dir_inode_operations;
65 static const struct inode_operations btrfs_symlink_inode_operations;
66 static const struct inode_operations btrfs_dir_ro_inode_operations;
67 static const struct inode_operations btrfs_special_inode_operations;
68 static const struct inode_operations btrfs_file_inode_operations;
69 static const struct address_space_operations btrfs_aops;
70 static const struct address_space_operations btrfs_symlink_aops;
71 static const struct file_operations btrfs_dir_file_operations;
72 static struct extent_io_ops btrfs_extent_io_ops;
74 static struct kmem_cache *btrfs_inode_cachep;
75 static struct kmem_cache *btrfs_delalloc_work_cachep;
76 struct kmem_cache *btrfs_trans_handle_cachep;
77 struct kmem_cache *btrfs_transaction_cachep;
78 struct kmem_cache *btrfs_path_cachep;
79 struct kmem_cache *btrfs_free_space_cachep;
82 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
83 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
84 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
85 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
86 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
87 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
88 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
89 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
92 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
93 static int btrfs_truncate(struct inode *inode);
94 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
95 static noinline int cow_file_range(struct inode *inode,
96 struct page *locked_page,
97 u64 start, u64 end, int *page_started,
98 unsigned long *nr_written, int unlock);
99 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
100 u64 len, u64 orig_start,
101 u64 block_start, u64 block_len,
102 u64 orig_block_len, int type);
104 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
105 struct inode *inode, struct inode *dir,
106 const struct qstr *qstr)
110 err = btrfs_init_acl(trans, inode, dir);
112 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
117 * this does all the hard work for inserting an inline extent into
118 * the btree. The caller should have done a btrfs_drop_extents so that
119 * no overlapping inline items exist in the btree
121 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
122 struct btrfs_root *root, struct inode *inode,
123 u64 start, size_t size, size_t compressed_size,
125 struct page **compressed_pages)
127 struct btrfs_key key;
128 struct btrfs_path *path;
129 struct extent_buffer *leaf;
130 struct page *page = NULL;
133 struct btrfs_file_extent_item *ei;
136 size_t cur_size = size;
138 unsigned long offset;
140 if (compressed_size && compressed_pages)
141 cur_size = compressed_size;
143 path = btrfs_alloc_path();
147 path->leave_spinning = 1;
149 key.objectid = btrfs_ino(inode);
151 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
152 datasize = btrfs_file_extent_calc_inline_size(cur_size);
154 inode_add_bytes(inode, size);
155 ret = btrfs_insert_empty_item(trans, root, path, &key,
161 leaf = path->nodes[0];
162 ei = btrfs_item_ptr(leaf, path->slots[0],
163 struct btrfs_file_extent_item);
164 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
165 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
166 btrfs_set_file_extent_encryption(leaf, ei, 0);
167 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
168 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
169 ptr = btrfs_file_extent_inline_start(ei);
171 if (compress_type != BTRFS_COMPRESS_NONE) {
174 while (compressed_size > 0) {
175 cpage = compressed_pages[i];
176 cur_size = min_t(unsigned long, compressed_size,
179 kaddr = kmap_atomic(cpage);
180 write_extent_buffer(leaf, kaddr, ptr, cur_size);
181 kunmap_atomic(kaddr);
185 compressed_size -= cur_size;
187 btrfs_set_file_extent_compression(leaf, ei,
190 page = find_get_page(inode->i_mapping,
191 start >> PAGE_CACHE_SHIFT);
192 btrfs_set_file_extent_compression(leaf, ei, 0);
193 kaddr = kmap_atomic(page);
194 offset = start & (PAGE_CACHE_SIZE - 1);
195 write_extent_buffer(leaf, kaddr + offset, ptr, size);
196 kunmap_atomic(kaddr);
197 page_cache_release(page);
199 btrfs_mark_buffer_dirty(leaf);
200 btrfs_free_path(path);
203 * we're an inline extent, so nobody can
204 * extend the file past i_size without locking
205 * a page we already have locked.
207 * We must do any isize and inode updates
208 * before we unlock the pages. Otherwise we
209 * could end up racing with unlink.
211 BTRFS_I(inode)->disk_i_size = inode->i_size;
212 ret = btrfs_update_inode(trans, root, inode);
216 btrfs_free_path(path);
222 * conditionally insert an inline extent into the file. This
223 * does the checks required to make sure the data is small enough
224 * to fit as an inline extent.
226 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
227 struct btrfs_root *root,
228 struct inode *inode, u64 start, u64 end,
229 size_t compressed_size, int compress_type,
230 struct page **compressed_pages)
232 u64 isize = i_size_read(inode);
233 u64 actual_end = min(end + 1, isize);
234 u64 inline_len = actual_end - start;
235 u64 aligned_end = (end + root->sectorsize - 1) &
236 ~((u64)root->sectorsize - 1);
237 u64 data_len = inline_len;
241 data_len = compressed_size;
244 actual_end >= PAGE_CACHE_SIZE ||
245 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
247 (actual_end & (root->sectorsize - 1)) == 0) ||
249 data_len > root->fs_info->max_inline) {
253 ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
257 if (isize > actual_end)
258 inline_len = min_t(u64, isize, actual_end);
259 ret = insert_inline_extent(trans, root, inode, start,
260 inline_len, compressed_size,
261 compress_type, compressed_pages);
262 if (ret && ret != -ENOSPC) {
263 btrfs_abort_transaction(trans, root, ret);
265 } else if (ret == -ENOSPC) {
269 btrfs_delalloc_release_metadata(inode, end + 1 - start);
270 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
274 struct async_extent {
279 unsigned long nr_pages;
281 struct list_head list;
286 struct btrfs_root *root;
287 struct page *locked_page;
290 struct list_head extents;
291 struct btrfs_work work;
294 static noinline int add_async_extent(struct async_cow *cow,
295 u64 start, u64 ram_size,
298 unsigned long nr_pages,
301 struct async_extent *async_extent;
303 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
304 BUG_ON(!async_extent); /* -ENOMEM */
305 async_extent->start = start;
306 async_extent->ram_size = ram_size;
307 async_extent->compressed_size = compressed_size;
308 async_extent->pages = pages;
309 async_extent->nr_pages = nr_pages;
310 async_extent->compress_type = compress_type;
311 list_add_tail(&async_extent->list, &cow->extents);
316 * we create compressed extents in two phases. The first
317 * phase compresses a range of pages that have already been
318 * locked (both pages and state bits are locked).
320 * This is done inside an ordered work queue, and the compression
321 * is spread across many cpus. The actual IO submission is step
322 * two, and the ordered work queue takes care of making sure that
323 * happens in the same order things were put onto the queue by
324 * writepages and friends.
326 * If this code finds it can't get good compression, it puts an
327 * entry onto the work queue to write the uncompressed bytes. This
328 * makes sure that both compressed inodes and uncompressed inodes
329 * are written in the same order that the flusher thread sent them
332 static noinline int compress_file_range(struct inode *inode,
333 struct page *locked_page,
335 struct async_cow *async_cow,
338 struct btrfs_root *root = BTRFS_I(inode)->root;
339 struct btrfs_trans_handle *trans;
341 u64 blocksize = root->sectorsize;
343 u64 isize = i_size_read(inode);
345 struct page **pages = NULL;
346 unsigned long nr_pages;
347 unsigned long nr_pages_ret = 0;
348 unsigned long total_compressed = 0;
349 unsigned long total_in = 0;
350 unsigned long max_compressed = 128 * 1024;
351 unsigned long max_uncompressed = 128 * 1024;
354 int compress_type = root->fs_info->compress_type;
356 /* if this is a small write inside eof, kick off a defrag */
357 if ((end - start + 1) < 16 * 1024 &&
358 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
359 btrfs_add_inode_defrag(NULL, inode);
361 actual_end = min_t(u64, isize, end + 1);
364 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
365 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
368 * we don't want to send crud past the end of i_size through
369 * compression, that's just a waste of CPU time. So, if the
370 * end of the file is before the start of our current
371 * requested range of bytes, we bail out to the uncompressed
372 * cleanup code that can deal with all of this.
374 * It isn't really the fastest way to fix things, but this is a
375 * very uncommon corner.
377 if (actual_end <= start)
378 goto cleanup_and_bail_uncompressed;
380 total_compressed = actual_end - start;
382 /* we want to make sure that amount of ram required to uncompress
383 * an extent is reasonable, so we limit the total size in ram
384 * of a compressed extent to 128k. This is a crucial number
385 * because it also controls how easily we can spread reads across
386 * cpus for decompression.
388 * We also want to make sure the amount of IO required to do
389 * a random read is reasonably small, so we limit the size of
390 * a compressed extent to 128k.
392 total_compressed = min(total_compressed, max_uncompressed);
393 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
394 num_bytes = max(blocksize, num_bytes);
399 * we do compression for mount -o compress and when the
400 * inode has not been flagged as nocompress. This flag can
401 * change at any time if we discover bad compression ratios.
403 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
404 (btrfs_test_opt(root, COMPRESS) ||
405 (BTRFS_I(inode)->force_compress) ||
406 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
408 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
410 /* just bail out to the uncompressed code */
414 if (BTRFS_I(inode)->force_compress)
415 compress_type = BTRFS_I(inode)->force_compress;
417 ret = btrfs_compress_pages(compress_type,
418 inode->i_mapping, start,
419 total_compressed, pages,
420 nr_pages, &nr_pages_ret,
426 unsigned long offset = total_compressed &
427 (PAGE_CACHE_SIZE - 1);
428 struct page *page = pages[nr_pages_ret - 1];
431 /* zero the tail end of the last page, we might be
432 * sending it down to disk
435 kaddr = kmap_atomic(page);
436 memset(kaddr + offset, 0,
437 PAGE_CACHE_SIZE - offset);
438 kunmap_atomic(kaddr);
445 trans = btrfs_join_transaction(root);
447 ret = PTR_ERR(trans);
449 goto cleanup_and_out;
451 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
453 /* lets try to make an inline extent */
454 if (ret || total_in < (actual_end - start)) {
455 /* we didn't compress the entire range, try
456 * to make an uncompressed inline extent.
458 ret = cow_file_range_inline(trans, root, inode,
459 start, end, 0, 0, NULL);
461 /* try making a compressed inline extent */
462 ret = cow_file_range_inline(trans, root, inode,
465 compress_type, pages);
469 * inline extent creation worked or returned error,
470 * we don't need to create any more async work items.
471 * Unlock and free up our temp pages.
473 extent_clear_unlock_delalloc(inode,
474 &BTRFS_I(inode)->io_tree,
476 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
477 EXTENT_CLEAR_DELALLOC |
478 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
480 btrfs_end_transaction(trans, root);
483 btrfs_end_transaction(trans, root);
488 * we aren't doing an inline extent round the compressed size
489 * up to a block size boundary so the allocator does sane
492 total_compressed = (total_compressed + blocksize - 1) &
496 * one last check to make sure the compression is really a
497 * win, compare the page count read with the blocks on disk
499 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
500 ~(PAGE_CACHE_SIZE - 1);
501 if (total_compressed >= total_in) {
504 num_bytes = total_in;
507 if (!will_compress && pages) {
509 * the compression code ran but failed to make things smaller,
510 * free any pages it allocated and our page pointer array
512 for (i = 0; i < nr_pages_ret; i++) {
513 WARN_ON(pages[i]->mapping);
514 page_cache_release(pages[i]);
518 total_compressed = 0;
521 /* flag the file so we don't compress in the future */
522 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
523 !(BTRFS_I(inode)->force_compress)) {
524 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
530 /* the async work queues will take care of doing actual
531 * allocation on disk for these compressed pages,
532 * and will submit them to the elevator.
534 add_async_extent(async_cow, start, num_bytes,
535 total_compressed, pages, nr_pages_ret,
538 if (start + num_bytes < end) {
545 cleanup_and_bail_uncompressed:
547 * No compression, but we still need to write the pages in
548 * the file we've been given so far. redirty the locked
549 * page if it corresponds to our extent and set things up
550 * for the async work queue to run cow_file_range to do
551 * the normal delalloc dance
553 if (page_offset(locked_page) >= start &&
554 page_offset(locked_page) <= end) {
555 __set_page_dirty_nobuffers(locked_page);
556 /* unlocked later on in the async handlers */
558 add_async_extent(async_cow, start, end - start + 1,
559 0, NULL, 0, BTRFS_COMPRESS_NONE);
567 for (i = 0; i < nr_pages_ret; i++) {
568 WARN_ON(pages[i]->mapping);
569 page_cache_release(pages[i]);
576 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
578 EXTENT_CLEAR_UNLOCK_PAGE |
580 EXTENT_CLEAR_DELALLOC |
581 EXTENT_SET_WRITEBACK |
582 EXTENT_END_WRITEBACK);
583 if (!trans || IS_ERR(trans))
584 btrfs_error(root->fs_info, ret, "Failed to join transaction");
586 btrfs_abort_transaction(trans, root, ret);
591 * phase two of compressed writeback. This is the ordered portion
592 * of the code, which only gets called in the order the work was
593 * queued. We walk all the async extents created by compress_file_range
594 * and send them down to the disk.
596 static noinline int submit_compressed_extents(struct inode *inode,
597 struct async_cow *async_cow)
599 struct async_extent *async_extent;
601 struct btrfs_trans_handle *trans;
602 struct btrfs_key ins;
603 struct extent_map *em;
604 struct btrfs_root *root = BTRFS_I(inode)->root;
605 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
606 struct extent_io_tree *io_tree;
609 if (list_empty(&async_cow->extents))
613 while (!list_empty(&async_cow->extents)) {
614 async_extent = list_entry(async_cow->extents.next,
615 struct async_extent, list);
616 list_del(&async_extent->list);
618 io_tree = &BTRFS_I(inode)->io_tree;
621 /* did the compression code fall back to uncompressed IO? */
622 if (!async_extent->pages) {
623 int page_started = 0;
624 unsigned long nr_written = 0;
626 lock_extent(io_tree, async_extent->start,
627 async_extent->start +
628 async_extent->ram_size - 1);
630 /* allocate blocks */
631 ret = cow_file_range(inode, async_cow->locked_page,
633 async_extent->start +
634 async_extent->ram_size - 1,
635 &page_started, &nr_written, 0);
640 * if page_started, cow_file_range inserted an
641 * inline extent and took care of all the unlocking
642 * and IO for us. Otherwise, we need to submit
643 * all those pages down to the drive.
645 if (!page_started && !ret)
646 extent_write_locked_range(io_tree,
647 inode, async_extent->start,
648 async_extent->start +
649 async_extent->ram_size - 1,
653 unlock_page(async_cow->locked_page);
659 lock_extent(io_tree, async_extent->start,
660 async_extent->start + async_extent->ram_size - 1);
662 trans = btrfs_join_transaction(root);
664 ret = PTR_ERR(trans);
666 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
667 ret = btrfs_reserve_extent(trans, root,
668 async_extent->compressed_size,
669 async_extent->compressed_size,
670 0, alloc_hint, &ins, 1);
671 if (ret && ret != -ENOSPC)
672 btrfs_abort_transaction(trans, root, ret);
673 btrfs_end_transaction(trans, root);
679 for (i = 0; i < async_extent->nr_pages; i++) {
680 WARN_ON(async_extent->pages[i]->mapping);
681 page_cache_release(async_extent->pages[i]);
683 kfree(async_extent->pages);
684 async_extent->nr_pages = 0;
685 async_extent->pages = NULL;
693 * here we're doing allocation and writeback of the
696 btrfs_drop_extent_cache(inode, async_extent->start,
697 async_extent->start +
698 async_extent->ram_size - 1, 0);
700 em = alloc_extent_map();
702 goto out_free_reserve;
703 em->start = async_extent->start;
704 em->len = async_extent->ram_size;
705 em->orig_start = em->start;
706 em->mod_start = em->start;
707 em->mod_len = em->len;
709 em->block_start = ins.objectid;
710 em->block_len = ins.offset;
711 em->orig_block_len = ins.offset;
712 em->bdev = root->fs_info->fs_devices->latest_bdev;
713 em->compress_type = async_extent->compress_type;
714 set_bit(EXTENT_FLAG_PINNED, &em->flags);
715 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
719 write_lock(&em_tree->lock);
720 ret = add_extent_mapping(em_tree, em);
723 &em_tree->modified_extents);
724 write_unlock(&em_tree->lock);
725 if (ret != -EEXIST) {
729 btrfs_drop_extent_cache(inode, async_extent->start,
730 async_extent->start +
731 async_extent->ram_size - 1, 0);
735 goto out_free_reserve;
737 ret = btrfs_add_ordered_extent_compress(inode,
740 async_extent->ram_size,
742 BTRFS_ORDERED_COMPRESSED,
743 async_extent->compress_type);
745 goto out_free_reserve;
748 * clear dirty, set writeback and unlock the pages.
750 extent_clear_unlock_delalloc(inode,
751 &BTRFS_I(inode)->io_tree,
753 async_extent->start +
754 async_extent->ram_size - 1,
755 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
756 EXTENT_CLEAR_UNLOCK |
757 EXTENT_CLEAR_DELALLOC |
758 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
760 ret = btrfs_submit_compressed_write(inode,
762 async_extent->ram_size,
764 ins.offset, async_extent->pages,
765 async_extent->nr_pages);
766 alloc_hint = ins.objectid + ins.offset;
776 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
778 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
780 async_extent->start +
781 async_extent->ram_size - 1,
782 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
783 EXTENT_CLEAR_UNLOCK |
784 EXTENT_CLEAR_DELALLOC |
786 EXTENT_SET_WRITEBACK |
787 EXTENT_END_WRITEBACK);
792 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
795 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
796 struct extent_map *em;
799 read_lock(&em_tree->lock);
800 em = search_extent_mapping(em_tree, start, num_bytes);
803 * if block start isn't an actual block number then find the
804 * first block in this inode and use that as a hint. If that
805 * block is also bogus then just don't worry about it.
807 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
809 em = search_extent_mapping(em_tree, 0, 0);
810 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
811 alloc_hint = em->block_start;
815 alloc_hint = em->block_start;
819 read_unlock(&em_tree->lock);
825 * when extent_io.c finds a delayed allocation range in the file,
826 * the call backs end up in this code. The basic idea is to
827 * allocate extents on disk for the range, and create ordered data structs
828 * in ram to track those extents.
830 * locked_page is the page that writepage had locked already. We use
831 * it to make sure we don't do extra locks or unlocks.
833 * *page_started is set to one if we unlock locked_page and do everything
834 * required to start IO on it. It may be clean and already done with
837 static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
839 struct btrfs_root *root,
840 struct page *locked_page,
841 u64 start, u64 end, int *page_started,
842 unsigned long *nr_written,
847 unsigned long ram_size;
850 u64 blocksize = root->sectorsize;
851 struct btrfs_key ins;
852 struct extent_map *em;
853 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
856 BUG_ON(btrfs_is_free_space_inode(inode));
858 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
859 num_bytes = max(blocksize, num_bytes);
860 disk_num_bytes = num_bytes;
862 /* if this is a small write inside eof, kick off defrag */
863 if (num_bytes < 64 * 1024 &&
864 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
865 btrfs_add_inode_defrag(trans, inode);
868 /* lets try to make an inline extent */
869 ret = cow_file_range_inline(trans, root, inode,
870 start, end, 0, 0, NULL);
872 extent_clear_unlock_delalloc(inode,
873 &BTRFS_I(inode)->io_tree,
875 EXTENT_CLEAR_UNLOCK_PAGE |
876 EXTENT_CLEAR_UNLOCK |
877 EXTENT_CLEAR_DELALLOC |
879 EXTENT_SET_WRITEBACK |
880 EXTENT_END_WRITEBACK);
882 *nr_written = *nr_written +
883 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
886 } else if (ret < 0) {
887 btrfs_abort_transaction(trans, root, ret);
892 BUG_ON(disk_num_bytes >
893 btrfs_super_total_bytes(root->fs_info->super_copy));
895 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
896 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
898 while (disk_num_bytes > 0) {
901 cur_alloc_size = disk_num_bytes;
902 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
903 root->sectorsize, 0, alloc_hint,
906 btrfs_abort_transaction(trans, root, ret);
910 em = alloc_extent_map();
911 BUG_ON(!em); /* -ENOMEM */
913 em->orig_start = em->start;
914 ram_size = ins.offset;
915 em->len = ins.offset;
916 em->mod_start = em->start;
917 em->mod_len = em->len;
919 em->block_start = ins.objectid;
920 em->block_len = ins.offset;
921 em->orig_block_len = ins.offset;
922 em->bdev = root->fs_info->fs_devices->latest_bdev;
923 set_bit(EXTENT_FLAG_PINNED, &em->flags);
927 write_lock(&em_tree->lock);
928 ret = add_extent_mapping(em_tree, em);
931 &em_tree->modified_extents);
932 write_unlock(&em_tree->lock);
933 if (ret != -EEXIST) {
937 btrfs_drop_extent_cache(inode, start,
938 start + ram_size - 1, 0);
941 cur_alloc_size = ins.offset;
942 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
943 ram_size, cur_alloc_size, 0);
944 BUG_ON(ret); /* -ENOMEM */
946 if (root->root_key.objectid ==
947 BTRFS_DATA_RELOC_TREE_OBJECTID) {
948 ret = btrfs_reloc_clone_csums(inode, start,
951 btrfs_abort_transaction(trans, root, ret);
956 if (disk_num_bytes < cur_alloc_size)
959 /* we're not doing compressed IO, don't unlock the first
960 * page (which the caller expects to stay locked), don't
961 * clear any dirty bits and don't set any writeback bits
963 * Do set the Private2 bit so we know this page was properly
964 * setup for writepage
966 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
967 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
970 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
971 start, start + ram_size - 1,
973 disk_num_bytes -= cur_alloc_size;
974 num_bytes -= cur_alloc_size;
975 alloc_hint = ins.objectid + ins.offset;
976 start += cur_alloc_size;
982 extent_clear_unlock_delalloc(inode,
983 &BTRFS_I(inode)->io_tree,
984 start, end, locked_page,
985 EXTENT_CLEAR_UNLOCK_PAGE |
986 EXTENT_CLEAR_UNLOCK |
987 EXTENT_CLEAR_DELALLOC |
989 EXTENT_SET_WRITEBACK |
990 EXTENT_END_WRITEBACK);
995 static noinline int cow_file_range(struct inode *inode,
996 struct page *locked_page,
997 u64 start, u64 end, int *page_started,
998 unsigned long *nr_written,
1001 struct btrfs_trans_handle *trans;
1002 struct btrfs_root *root = BTRFS_I(inode)->root;
1005 trans = btrfs_join_transaction(root);
1006 if (IS_ERR(trans)) {
1007 extent_clear_unlock_delalloc(inode,
1008 &BTRFS_I(inode)->io_tree,
1009 start, end, locked_page,
1010 EXTENT_CLEAR_UNLOCK_PAGE |
1011 EXTENT_CLEAR_UNLOCK |
1012 EXTENT_CLEAR_DELALLOC |
1013 EXTENT_CLEAR_DIRTY |
1014 EXTENT_SET_WRITEBACK |
1015 EXTENT_END_WRITEBACK);
1016 return PTR_ERR(trans);
1018 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1020 ret = __cow_file_range(trans, inode, root, locked_page, start, end,
1021 page_started, nr_written, unlock);
1023 btrfs_end_transaction(trans, root);
1029 * work queue call back to started compression on a file and pages
1031 static noinline void async_cow_start(struct btrfs_work *work)
1033 struct async_cow *async_cow;
1035 async_cow = container_of(work, struct async_cow, work);
1037 compress_file_range(async_cow->inode, async_cow->locked_page,
1038 async_cow->start, async_cow->end, async_cow,
1040 if (num_added == 0) {
1041 btrfs_add_delayed_iput(async_cow->inode);
1042 async_cow->inode = NULL;
1047 * work queue call back to submit previously compressed pages
1049 static noinline void async_cow_submit(struct btrfs_work *work)
1051 struct async_cow *async_cow;
1052 struct btrfs_root *root;
1053 unsigned long nr_pages;
1055 async_cow = container_of(work, struct async_cow, work);
1057 root = async_cow->root;
1058 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1061 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1063 waitqueue_active(&root->fs_info->async_submit_wait))
1064 wake_up(&root->fs_info->async_submit_wait);
1066 if (async_cow->inode)
1067 submit_compressed_extents(async_cow->inode, async_cow);
1070 static noinline void async_cow_free(struct btrfs_work *work)
1072 struct async_cow *async_cow;
1073 async_cow = container_of(work, struct async_cow, work);
1074 if (async_cow->inode)
1075 btrfs_add_delayed_iput(async_cow->inode);
1079 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1080 u64 start, u64 end, int *page_started,
1081 unsigned long *nr_written)
1083 struct async_cow *async_cow;
1084 struct btrfs_root *root = BTRFS_I(inode)->root;
1085 unsigned long nr_pages;
1087 int limit = 10 * 1024 * 1024;
1089 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1090 1, 0, NULL, GFP_NOFS);
1091 while (start < end) {
1092 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1093 BUG_ON(!async_cow); /* -ENOMEM */
1094 async_cow->inode = igrab(inode);
1095 async_cow->root = root;
1096 async_cow->locked_page = locked_page;
1097 async_cow->start = start;
1099 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1102 cur_end = min(end, start + 512 * 1024 - 1);
1104 async_cow->end = cur_end;
1105 INIT_LIST_HEAD(&async_cow->extents);
1107 async_cow->work.func = async_cow_start;
1108 async_cow->work.ordered_func = async_cow_submit;
1109 async_cow->work.ordered_free = async_cow_free;
1110 async_cow->work.flags = 0;
1112 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1114 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1116 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1119 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1120 wait_event(root->fs_info->async_submit_wait,
1121 (atomic_read(&root->fs_info->async_delalloc_pages) <
1125 while (atomic_read(&root->fs_info->async_submit_draining) &&
1126 atomic_read(&root->fs_info->async_delalloc_pages)) {
1127 wait_event(root->fs_info->async_submit_wait,
1128 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1132 *nr_written += nr_pages;
1133 start = cur_end + 1;
1139 static noinline int csum_exist_in_range(struct btrfs_root *root,
1140 u64 bytenr, u64 num_bytes)
1143 struct btrfs_ordered_sum *sums;
1146 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1147 bytenr + num_bytes - 1, &list, 0);
1148 if (ret == 0 && list_empty(&list))
1151 while (!list_empty(&list)) {
1152 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1153 list_del(&sums->list);
1160 * when nowcow writeback call back. This checks for snapshots or COW copies
1161 * of the extents that exist in the file, and COWs the file as required.
1163 * If no cow copies or snapshots exist, we write directly to the existing
1166 static noinline int run_delalloc_nocow(struct inode *inode,
1167 struct page *locked_page,
1168 u64 start, u64 end, int *page_started, int force,
1169 unsigned long *nr_written)
1171 struct btrfs_root *root = BTRFS_I(inode)->root;
1172 struct btrfs_trans_handle *trans;
1173 struct extent_buffer *leaf;
1174 struct btrfs_path *path;
1175 struct btrfs_file_extent_item *fi;
1176 struct btrfs_key found_key;
1190 u64 ino = btrfs_ino(inode);
1192 path = btrfs_alloc_path();
1194 extent_clear_unlock_delalloc(inode,
1195 &BTRFS_I(inode)->io_tree,
1196 start, end, locked_page,
1197 EXTENT_CLEAR_UNLOCK_PAGE |
1198 EXTENT_CLEAR_UNLOCK |
1199 EXTENT_CLEAR_DELALLOC |
1200 EXTENT_CLEAR_DIRTY |
1201 EXTENT_SET_WRITEBACK |
1202 EXTENT_END_WRITEBACK);
1206 nolock = btrfs_is_free_space_inode(inode);
1209 trans = btrfs_join_transaction_nolock(root);
1211 trans = btrfs_join_transaction(root);
1213 if (IS_ERR(trans)) {
1214 extent_clear_unlock_delalloc(inode,
1215 &BTRFS_I(inode)->io_tree,
1216 start, end, locked_page,
1217 EXTENT_CLEAR_UNLOCK_PAGE |
1218 EXTENT_CLEAR_UNLOCK |
1219 EXTENT_CLEAR_DELALLOC |
1220 EXTENT_CLEAR_DIRTY |
1221 EXTENT_SET_WRITEBACK |
1222 EXTENT_END_WRITEBACK);
1223 btrfs_free_path(path);
1224 return PTR_ERR(trans);
1227 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1229 cow_start = (u64)-1;
1232 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1235 btrfs_abort_transaction(trans, root, ret);
1238 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1239 leaf = path->nodes[0];
1240 btrfs_item_key_to_cpu(leaf, &found_key,
1241 path->slots[0] - 1);
1242 if (found_key.objectid == ino &&
1243 found_key.type == BTRFS_EXTENT_DATA_KEY)
1248 leaf = path->nodes[0];
1249 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1250 ret = btrfs_next_leaf(root, path);
1252 btrfs_abort_transaction(trans, root, ret);
1257 leaf = path->nodes[0];
1263 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1265 if (found_key.objectid > ino ||
1266 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1267 found_key.offset > end)
1270 if (found_key.offset > cur_offset) {
1271 extent_end = found_key.offset;
1276 fi = btrfs_item_ptr(leaf, path->slots[0],
1277 struct btrfs_file_extent_item);
1278 extent_type = btrfs_file_extent_type(leaf, fi);
1280 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1281 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1282 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1283 extent_offset = btrfs_file_extent_offset(leaf, fi);
1284 extent_end = found_key.offset +
1285 btrfs_file_extent_num_bytes(leaf, fi);
1287 btrfs_file_extent_disk_num_bytes(leaf, fi);
1288 if (extent_end <= start) {
1292 if (disk_bytenr == 0)
1294 if (btrfs_file_extent_compression(leaf, fi) ||
1295 btrfs_file_extent_encryption(leaf, fi) ||
1296 btrfs_file_extent_other_encoding(leaf, fi))
1298 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1300 if (btrfs_extent_readonly(root, disk_bytenr))
1302 if (btrfs_cross_ref_exist(trans, root, ino,
1304 extent_offset, disk_bytenr))
1306 disk_bytenr += extent_offset;
1307 disk_bytenr += cur_offset - found_key.offset;
1308 num_bytes = min(end + 1, extent_end) - cur_offset;
1310 * force cow if csum exists in the range.
1311 * this ensure that csum for a given extent are
1312 * either valid or do not exist.
1314 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1317 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1318 extent_end = found_key.offset +
1319 btrfs_file_extent_inline_len(leaf, fi);
1320 extent_end = ALIGN(extent_end, root->sectorsize);
1325 if (extent_end <= start) {
1330 if (cow_start == (u64)-1)
1331 cow_start = cur_offset;
1332 cur_offset = extent_end;
1333 if (cur_offset > end)
1339 btrfs_release_path(path);
1340 if (cow_start != (u64)-1) {
1341 ret = __cow_file_range(trans, inode, root, locked_page,
1342 cow_start, found_key.offset - 1,
1343 page_started, nr_written, 1);
1345 btrfs_abort_transaction(trans, root, ret);
1348 cow_start = (u64)-1;
1351 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1352 struct extent_map *em;
1353 struct extent_map_tree *em_tree;
1354 em_tree = &BTRFS_I(inode)->extent_tree;
1355 em = alloc_extent_map();
1356 BUG_ON(!em); /* -ENOMEM */
1357 em->start = cur_offset;
1358 em->orig_start = found_key.offset - extent_offset;
1359 em->len = num_bytes;
1360 em->block_len = num_bytes;
1361 em->block_start = disk_bytenr;
1362 em->orig_block_len = disk_num_bytes;
1363 em->bdev = root->fs_info->fs_devices->latest_bdev;
1364 em->mod_start = em->start;
1365 em->mod_len = em->len;
1366 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1367 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1368 em->generation = -1;
1370 write_lock(&em_tree->lock);
1371 ret = add_extent_mapping(em_tree, em);
1373 list_move(&em->list,
1374 &em_tree->modified_extents);
1375 write_unlock(&em_tree->lock);
1376 if (ret != -EEXIST) {
1377 free_extent_map(em);
1380 btrfs_drop_extent_cache(inode, em->start,
1381 em->start + em->len - 1, 0);
1383 type = BTRFS_ORDERED_PREALLOC;
1385 type = BTRFS_ORDERED_NOCOW;
1388 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1389 num_bytes, num_bytes, type);
1390 BUG_ON(ret); /* -ENOMEM */
1392 if (root->root_key.objectid ==
1393 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1394 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1397 btrfs_abort_transaction(trans, root, ret);
1402 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1403 cur_offset, cur_offset + num_bytes - 1,
1404 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1405 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1406 EXTENT_SET_PRIVATE2);
1407 cur_offset = extent_end;
1408 if (cur_offset > end)
1411 btrfs_release_path(path);
1413 if (cur_offset <= end && cow_start == (u64)-1) {
1414 cow_start = cur_offset;
1418 if (cow_start != (u64)-1) {
1419 ret = __cow_file_range(trans, inode, root, locked_page,
1421 page_started, nr_written, 1);
1423 btrfs_abort_transaction(trans, root, ret);
1429 err = btrfs_end_transaction(trans, root);
1433 if (ret && cur_offset < end)
1434 extent_clear_unlock_delalloc(inode,
1435 &BTRFS_I(inode)->io_tree,
1436 cur_offset, end, locked_page,
1437 EXTENT_CLEAR_UNLOCK_PAGE |
1438 EXTENT_CLEAR_UNLOCK |
1439 EXTENT_CLEAR_DELALLOC |
1440 EXTENT_CLEAR_DIRTY |
1441 EXTENT_SET_WRITEBACK |
1442 EXTENT_END_WRITEBACK);
1444 btrfs_free_path(path);
1449 * extent_io.c call back to do delayed allocation processing
1451 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1452 u64 start, u64 end, int *page_started,
1453 unsigned long *nr_written)
1456 struct btrfs_root *root = BTRFS_I(inode)->root;
1458 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1459 ret = run_delalloc_nocow(inode, locked_page, start, end,
1460 page_started, 1, nr_written);
1461 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1462 ret = run_delalloc_nocow(inode, locked_page, start, end,
1463 page_started, 0, nr_written);
1464 } else if (!btrfs_test_opt(root, COMPRESS) &&
1465 !(BTRFS_I(inode)->force_compress) &&
1466 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1467 ret = cow_file_range(inode, locked_page, start, end,
1468 page_started, nr_written, 1);
1470 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1471 &BTRFS_I(inode)->runtime_flags);
1472 ret = cow_file_range_async(inode, locked_page, start, end,
1473 page_started, nr_written);
1478 static void btrfs_split_extent_hook(struct inode *inode,
1479 struct extent_state *orig, u64 split)
1481 /* not delalloc, ignore it */
1482 if (!(orig->state & EXTENT_DELALLOC))
1485 spin_lock(&BTRFS_I(inode)->lock);
1486 BTRFS_I(inode)->outstanding_extents++;
1487 spin_unlock(&BTRFS_I(inode)->lock);
1491 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1492 * extents so we can keep track of new extents that are just merged onto old
1493 * extents, such as when we are doing sequential writes, so we can properly
1494 * account for the metadata space we'll need.
1496 static void btrfs_merge_extent_hook(struct inode *inode,
1497 struct extent_state *new,
1498 struct extent_state *other)
1500 /* not delalloc, ignore it */
1501 if (!(other->state & EXTENT_DELALLOC))
1504 spin_lock(&BTRFS_I(inode)->lock);
1505 BTRFS_I(inode)->outstanding_extents--;
1506 spin_unlock(&BTRFS_I(inode)->lock);
1510 * extent_io.c set_bit_hook, used to track delayed allocation
1511 * bytes in this file, and to maintain the list of inodes that
1512 * have pending delalloc work to be done.
1514 static void btrfs_set_bit_hook(struct inode *inode,
1515 struct extent_state *state, int *bits)
1519 * set_bit and clear bit hooks normally require _irqsave/restore
1520 * but in this case, we are only testing for the DELALLOC
1521 * bit, which is only set or cleared with irqs on
1523 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1524 struct btrfs_root *root = BTRFS_I(inode)->root;
1525 u64 len = state->end + 1 - state->start;
1526 bool do_list = !btrfs_is_free_space_inode(inode);
1528 if (*bits & EXTENT_FIRST_DELALLOC) {
1529 *bits &= ~EXTENT_FIRST_DELALLOC;
1531 spin_lock(&BTRFS_I(inode)->lock);
1532 BTRFS_I(inode)->outstanding_extents++;
1533 spin_unlock(&BTRFS_I(inode)->lock);
1536 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1537 root->fs_info->delalloc_batch);
1538 spin_lock(&BTRFS_I(inode)->lock);
1539 BTRFS_I(inode)->delalloc_bytes += len;
1540 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1541 &BTRFS_I(inode)->runtime_flags)) {
1542 spin_lock(&root->fs_info->delalloc_lock);
1543 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1544 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1545 &root->fs_info->delalloc_inodes);
1546 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1547 &BTRFS_I(inode)->runtime_flags);
1549 spin_unlock(&root->fs_info->delalloc_lock);
1551 spin_unlock(&BTRFS_I(inode)->lock);
1556 * extent_io.c clear_bit_hook, see set_bit_hook for why
1558 static void btrfs_clear_bit_hook(struct inode *inode,
1559 struct extent_state *state, int *bits)
1562 * set_bit and clear bit hooks normally require _irqsave/restore
1563 * but in this case, we are only testing for the DELALLOC
1564 * bit, which is only set or cleared with irqs on
1566 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1567 struct btrfs_root *root = BTRFS_I(inode)->root;
1568 u64 len = state->end + 1 - state->start;
1569 bool do_list = !btrfs_is_free_space_inode(inode);
1571 if (*bits & EXTENT_FIRST_DELALLOC) {
1572 *bits &= ~EXTENT_FIRST_DELALLOC;
1573 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1574 spin_lock(&BTRFS_I(inode)->lock);
1575 BTRFS_I(inode)->outstanding_extents--;
1576 spin_unlock(&BTRFS_I(inode)->lock);
1579 if (*bits & EXTENT_DO_ACCOUNTING)
1580 btrfs_delalloc_release_metadata(inode, len);
1582 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1584 btrfs_free_reserved_data_space(inode, len);
1586 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1587 root->fs_info->delalloc_batch);
1588 spin_lock(&BTRFS_I(inode)->lock);
1589 BTRFS_I(inode)->delalloc_bytes -= len;
1590 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1591 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1592 &BTRFS_I(inode)->runtime_flags)) {
1593 spin_lock(&root->fs_info->delalloc_lock);
1594 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1595 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1596 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1597 &BTRFS_I(inode)->runtime_flags);
1599 spin_unlock(&root->fs_info->delalloc_lock);
1601 spin_unlock(&BTRFS_I(inode)->lock);
1606 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1607 * we don't create bios that span stripes or chunks
1609 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1610 size_t size, struct bio *bio,
1611 unsigned long bio_flags)
1613 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1614 u64 logical = (u64)bio->bi_sector << 9;
1619 if (bio_flags & EXTENT_BIO_COMPRESSED)
1622 length = bio->bi_size;
1623 map_length = length;
1624 ret = btrfs_map_block(root->fs_info, rw, logical,
1625 &map_length, NULL, 0);
1626 /* Will always return 0 with map_multi == NULL */
1628 if (map_length < length + size)
1634 * in order to insert checksums into the metadata in large chunks,
1635 * we wait until bio submission time. All the pages in the bio are
1636 * checksummed and sums are attached onto the ordered extent record.
1638 * At IO completion time the cums attached on the ordered extent record
1639 * are inserted into the btree
1641 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1642 struct bio *bio, int mirror_num,
1643 unsigned long bio_flags,
1646 struct btrfs_root *root = BTRFS_I(inode)->root;
1649 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1650 BUG_ON(ret); /* -ENOMEM */
1655 * in order to insert checksums into the metadata in large chunks,
1656 * we wait until bio submission time. All the pages in the bio are
1657 * checksummed and sums are attached onto the ordered extent record.
1659 * At IO completion time the cums attached on the ordered extent record
1660 * are inserted into the btree
1662 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1663 int mirror_num, unsigned long bio_flags,
1666 struct btrfs_root *root = BTRFS_I(inode)->root;
1669 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1671 bio_endio(bio, ret);
1676 * extent_io.c submission hook. This does the right thing for csum calculation
1677 * on write, or reading the csums from the tree before a read
1679 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1680 int mirror_num, unsigned long bio_flags,
1683 struct btrfs_root *root = BTRFS_I(inode)->root;
1687 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1689 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1691 if (btrfs_is_free_space_inode(inode))
1694 if (!(rw & REQ_WRITE)) {
1695 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1699 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1700 ret = btrfs_submit_compressed_read(inode, bio,
1704 } else if (!skip_sum) {
1705 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1710 } else if (async && !skip_sum) {
1711 /* csum items have already been cloned */
1712 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1714 /* we're doing a write, do the async checksumming */
1715 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1716 inode, rw, bio, mirror_num,
1717 bio_flags, bio_offset,
1718 __btrfs_submit_bio_start,
1719 __btrfs_submit_bio_done);
1721 } else if (!skip_sum) {
1722 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1728 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1732 bio_endio(bio, ret);
1737 * given a list of ordered sums record them in the inode. This happens
1738 * at IO completion time based on sums calculated at bio submission time.
1740 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1741 struct inode *inode, u64 file_offset,
1742 struct list_head *list)
1744 struct btrfs_ordered_sum *sum;
1746 list_for_each_entry(sum, list, list) {
1747 btrfs_csum_file_blocks(trans,
1748 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1753 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1754 struct extent_state **cached_state)
1756 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1757 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1758 cached_state, GFP_NOFS);
1761 /* see btrfs_writepage_start_hook for details on why this is required */
1762 struct btrfs_writepage_fixup {
1764 struct btrfs_work work;
1767 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1769 struct btrfs_writepage_fixup *fixup;
1770 struct btrfs_ordered_extent *ordered;
1771 struct extent_state *cached_state = NULL;
1773 struct inode *inode;
1778 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1782 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1783 ClearPageChecked(page);
1787 inode = page->mapping->host;
1788 page_start = page_offset(page);
1789 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1791 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1794 /* already ordered? We're done */
1795 if (PagePrivate2(page))
1798 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1800 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1801 page_end, &cached_state, GFP_NOFS);
1803 btrfs_start_ordered_extent(inode, ordered, 1);
1804 btrfs_put_ordered_extent(ordered);
1808 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1810 mapping_set_error(page->mapping, ret);
1811 end_extent_writepage(page, ret, page_start, page_end);
1812 ClearPageChecked(page);
1816 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1817 ClearPageChecked(page);
1818 set_page_dirty(page);
1820 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1821 &cached_state, GFP_NOFS);
1824 page_cache_release(page);
1829 * There are a few paths in the higher layers of the kernel that directly
1830 * set the page dirty bit without asking the filesystem if it is a
1831 * good idea. This causes problems because we want to make sure COW
1832 * properly happens and the data=ordered rules are followed.
1834 * In our case any range that doesn't have the ORDERED bit set
1835 * hasn't been properly setup for IO. We kick off an async process
1836 * to fix it up. The async helper will wait for ordered extents, set
1837 * the delalloc bit and make it safe to write the page.
1839 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1841 struct inode *inode = page->mapping->host;
1842 struct btrfs_writepage_fixup *fixup;
1843 struct btrfs_root *root = BTRFS_I(inode)->root;
1845 /* this page is properly in the ordered list */
1846 if (TestClearPagePrivate2(page))
1849 if (PageChecked(page))
1852 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1856 SetPageChecked(page);
1857 page_cache_get(page);
1858 fixup->work.func = btrfs_writepage_fixup_worker;
1860 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1864 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1865 struct inode *inode, u64 file_pos,
1866 u64 disk_bytenr, u64 disk_num_bytes,
1867 u64 num_bytes, u64 ram_bytes,
1868 u8 compression, u8 encryption,
1869 u16 other_encoding, int extent_type)
1871 struct btrfs_root *root = BTRFS_I(inode)->root;
1872 struct btrfs_file_extent_item *fi;
1873 struct btrfs_path *path;
1874 struct extent_buffer *leaf;
1875 struct btrfs_key ins;
1878 path = btrfs_alloc_path();
1882 path->leave_spinning = 1;
1885 * we may be replacing one extent in the tree with another.
1886 * The new extent is pinned in the extent map, and we don't want
1887 * to drop it from the cache until it is completely in the btree.
1889 * So, tell btrfs_drop_extents to leave this extent in the cache.
1890 * the caller is expected to unpin it and allow it to be merged
1893 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1894 file_pos + num_bytes, 0);
1898 ins.objectid = btrfs_ino(inode);
1899 ins.offset = file_pos;
1900 ins.type = BTRFS_EXTENT_DATA_KEY;
1901 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1904 leaf = path->nodes[0];
1905 fi = btrfs_item_ptr(leaf, path->slots[0],
1906 struct btrfs_file_extent_item);
1907 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1908 btrfs_set_file_extent_type(leaf, fi, extent_type);
1909 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1910 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1911 btrfs_set_file_extent_offset(leaf, fi, 0);
1912 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1913 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1914 btrfs_set_file_extent_compression(leaf, fi, compression);
1915 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1916 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1918 btrfs_mark_buffer_dirty(leaf);
1919 btrfs_release_path(path);
1921 inode_add_bytes(inode, num_bytes);
1923 ins.objectid = disk_bytenr;
1924 ins.offset = disk_num_bytes;
1925 ins.type = BTRFS_EXTENT_ITEM_KEY;
1926 ret = btrfs_alloc_reserved_file_extent(trans, root,
1927 root->root_key.objectid,
1928 btrfs_ino(inode), file_pos, &ins);
1930 btrfs_free_path(path);
1936 * helper function for btrfs_finish_ordered_io, this
1937 * just reads in some of the csum leaves to prime them into ram
1938 * before we start the transaction. It limits the amount of btree
1939 * reads required while inside the transaction.
1941 /* as ordered data IO finishes, this gets called so we can finish
1942 * an ordered extent if the range of bytes in the file it covers are
1945 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
1947 struct inode *inode = ordered_extent->inode;
1948 struct btrfs_root *root = BTRFS_I(inode)->root;
1949 struct btrfs_trans_handle *trans = NULL;
1950 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1951 struct extent_state *cached_state = NULL;
1952 int compress_type = 0;
1956 nolock = btrfs_is_free_space_inode(inode);
1958 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1963 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1964 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1965 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1967 trans = btrfs_join_transaction_nolock(root);
1969 trans = btrfs_join_transaction(root);
1970 if (IS_ERR(trans)) {
1971 ret = PTR_ERR(trans);
1975 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1976 ret = btrfs_update_inode_fallback(trans, root, inode);
1977 if (ret) /* -ENOMEM or corruption */
1978 btrfs_abort_transaction(trans, root, ret);
1982 lock_extent_bits(io_tree, ordered_extent->file_offset,
1983 ordered_extent->file_offset + ordered_extent->len - 1,
1987 trans = btrfs_join_transaction_nolock(root);
1989 trans = btrfs_join_transaction(root);
1990 if (IS_ERR(trans)) {
1991 ret = PTR_ERR(trans);
1995 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1997 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1998 compress_type = ordered_extent->compress_type;
1999 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2000 BUG_ON(compress_type);
2001 ret = btrfs_mark_extent_written(trans, inode,
2002 ordered_extent->file_offset,
2003 ordered_extent->file_offset +
2004 ordered_extent->len);
2006 BUG_ON(root == root->fs_info->tree_root);
2007 ret = insert_reserved_file_extent(trans, inode,
2008 ordered_extent->file_offset,
2009 ordered_extent->start,
2010 ordered_extent->disk_len,
2011 ordered_extent->len,
2012 ordered_extent->len,
2013 compress_type, 0, 0,
2014 BTRFS_FILE_EXTENT_REG);
2016 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2017 ordered_extent->file_offset, ordered_extent->len,
2020 btrfs_abort_transaction(trans, root, ret);
2024 add_pending_csums(trans, inode, ordered_extent->file_offset,
2025 &ordered_extent->list);
2027 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2028 ret = btrfs_update_inode_fallback(trans, root, inode);
2029 if (ret) { /* -ENOMEM or corruption */
2030 btrfs_abort_transaction(trans, root, ret);
2035 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2036 ordered_extent->file_offset +
2037 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2039 if (root != root->fs_info->tree_root)
2040 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2042 btrfs_end_transaction(trans, root);
2045 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
2046 ordered_extent->file_offset +
2047 ordered_extent->len - 1, NULL, GFP_NOFS);
2050 * If the ordered extent had an IOERR or something else went
2051 * wrong we need to return the space for this ordered extent
2052 * back to the allocator.
2054 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2055 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2056 btrfs_free_reserved_extent(root, ordered_extent->start,
2057 ordered_extent->disk_len);
2062 * This needs to be done to make sure anybody waiting knows we are done
2063 * updating everything for this ordered extent.
2065 btrfs_remove_ordered_extent(inode, ordered_extent);
2068 btrfs_put_ordered_extent(ordered_extent);
2069 /* once for the tree */
2070 btrfs_put_ordered_extent(ordered_extent);
2075 static void finish_ordered_fn(struct btrfs_work *work)
2077 struct btrfs_ordered_extent *ordered_extent;
2078 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2079 btrfs_finish_ordered_io(ordered_extent);
2082 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2083 struct extent_state *state, int uptodate)
2085 struct inode *inode = page->mapping->host;
2086 struct btrfs_root *root = BTRFS_I(inode)->root;
2087 struct btrfs_ordered_extent *ordered_extent = NULL;
2088 struct btrfs_workers *workers;
2090 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2092 ClearPagePrivate2(page);
2093 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2094 end - start + 1, uptodate))
2097 ordered_extent->work.func = finish_ordered_fn;
2098 ordered_extent->work.flags = 0;
2100 if (btrfs_is_free_space_inode(inode))
2101 workers = &root->fs_info->endio_freespace_worker;
2103 workers = &root->fs_info->endio_write_workers;
2104 btrfs_queue_worker(workers, &ordered_extent->work);
2110 * when reads are done, we need to check csums to verify the data is correct
2111 * if there's a match, we allow the bio to finish. If not, the code in
2112 * extent_io.c will try to find good copies for us.
2114 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2115 struct extent_state *state, int mirror)
2117 size_t offset = start - page_offset(page);
2118 struct inode *inode = page->mapping->host;
2119 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2121 u64 private = ~(u32)0;
2123 struct btrfs_root *root = BTRFS_I(inode)->root;
2126 if (PageChecked(page)) {
2127 ClearPageChecked(page);
2131 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2134 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2135 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2136 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2141 if (state && state->start == start) {
2142 private = state->private;
2145 ret = get_state_private(io_tree, start, &private);
2147 kaddr = kmap_atomic(page);
2151 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
2152 btrfs_csum_final(csum, (char *)&csum);
2153 if (csum != private)
2156 kunmap_atomic(kaddr);
2161 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2163 (unsigned long long)btrfs_ino(page->mapping->host),
2164 (unsigned long long)start, csum,
2165 (unsigned long long)private);
2166 memset(kaddr + offset, 1, end - start + 1);
2167 flush_dcache_page(page);
2168 kunmap_atomic(kaddr);
2174 struct delayed_iput {
2175 struct list_head list;
2176 struct inode *inode;
2179 /* JDM: If this is fs-wide, why can't we add a pointer to
2180 * btrfs_inode instead and avoid the allocation? */
2181 void btrfs_add_delayed_iput(struct inode *inode)
2183 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2184 struct delayed_iput *delayed;
2186 if (atomic_add_unless(&inode->i_count, -1, 1))
2189 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2190 delayed->inode = inode;
2192 spin_lock(&fs_info->delayed_iput_lock);
2193 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2194 spin_unlock(&fs_info->delayed_iput_lock);
2197 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2200 struct btrfs_fs_info *fs_info = root->fs_info;
2201 struct delayed_iput *delayed;
2204 spin_lock(&fs_info->delayed_iput_lock);
2205 empty = list_empty(&fs_info->delayed_iputs);
2206 spin_unlock(&fs_info->delayed_iput_lock);
2210 spin_lock(&fs_info->delayed_iput_lock);
2211 list_splice_init(&fs_info->delayed_iputs, &list);
2212 spin_unlock(&fs_info->delayed_iput_lock);
2214 while (!list_empty(&list)) {
2215 delayed = list_entry(list.next, struct delayed_iput, list);
2216 list_del(&delayed->list);
2217 iput(delayed->inode);
2223 * This is called in transaction commit time. If there are no orphan
2224 * files in the subvolume, it removes orphan item and frees block_rsv
2227 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2228 struct btrfs_root *root)
2230 struct btrfs_block_rsv *block_rsv;
2233 if (atomic_read(&root->orphan_inodes) ||
2234 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2237 spin_lock(&root->orphan_lock);
2238 if (atomic_read(&root->orphan_inodes)) {
2239 spin_unlock(&root->orphan_lock);
2243 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2244 spin_unlock(&root->orphan_lock);
2248 block_rsv = root->orphan_block_rsv;
2249 root->orphan_block_rsv = NULL;
2250 spin_unlock(&root->orphan_lock);
2252 if (root->orphan_item_inserted &&
2253 btrfs_root_refs(&root->root_item) > 0) {
2254 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2255 root->root_key.objectid);
2257 root->orphan_item_inserted = 0;
2261 WARN_ON(block_rsv->size > 0);
2262 btrfs_free_block_rsv(root, block_rsv);
2267 * This creates an orphan entry for the given inode in case something goes
2268 * wrong in the middle of an unlink/truncate.
2270 * NOTE: caller of this function should reserve 5 units of metadata for
2273 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2275 struct btrfs_root *root = BTRFS_I(inode)->root;
2276 struct btrfs_block_rsv *block_rsv = NULL;
2281 if (!root->orphan_block_rsv) {
2282 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2287 spin_lock(&root->orphan_lock);
2288 if (!root->orphan_block_rsv) {
2289 root->orphan_block_rsv = block_rsv;
2290 } else if (block_rsv) {
2291 btrfs_free_block_rsv(root, block_rsv);
2295 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2296 &BTRFS_I(inode)->runtime_flags)) {
2299 * For proper ENOSPC handling, we should do orphan
2300 * cleanup when mounting. But this introduces backward
2301 * compatibility issue.
2303 if (!xchg(&root->orphan_item_inserted, 1))
2309 atomic_inc(&root->orphan_inodes);
2312 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2313 &BTRFS_I(inode)->runtime_flags))
2315 spin_unlock(&root->orphan_lock);
2317 /* grab metadata reservation from transaction handle */
2319 ret = btrfs_orphan_reserve_metadata(trans, inode);
2320 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2323 /* insert an orphan item to track this unlinked/truncated file */
2325 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2326 if (ret && ret != -EEXIST) {
2327 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2328 &BTRFS_I(inode)->runtime_flags);
2329 btrfs_abort_transaction(trans, root, ret);
2335 /* insert an orphan item to track subvolume contains orphan files */
2337 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2338 root->root_key.objectid);
2339 if (ret && ret != -EEXIST) {
2340 btrfs_abort_transaction(trans, root, ret);
2348 * We have done the truncate/delete so we can go ahead and remove the orphan
2349 * item for this particular inode.
2351 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2353 struct btrfs_root *root = BTRFS_I(inode)->root;
2354 int delete_item = 0;
2355 int release_rsv = 0;
2358 spin_lock(&root->orphan_lock);
2359 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2360 &BTRFS_I(inode)->runtime_flags))
2363 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2364 &BTRFS_I(inode)->runtime_flags))
2366 spin_unlock(&root->orphan_lock);
2368 if (trans && delete_item) {
2369 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2370 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2374 btrfs_orphan_release_metadata(inode);
2375 atomic_dec(&root->orphan_inodes);
2382 * this cleans up any orphans that may be left on the list from the last use
2385 int btrfs_orphan_cleanup(struct btrfs_root *root)
2387 struct btrfs_path *path;
2388 struct extent_buffer *leaf;
2389 struct btrfs_key key, found_key;
2390 struct btrfs_trans_handle *trans;
2391 struct inode *inode;
2392 u64 last_objectid = 0;
2393 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2395 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2398 path = btrfs_alloc_path();
2405 key.objectid = BTRFS_ORPHAN_OBJECTID;
2406 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2407 key.offset = (u64)-1;
2410 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2415 * if ret == 0 means we found what we were searching for, which
2416 * is weird, but possible, so only screw with path if we didn't
2417 * find the key and see if we have stuff that matches
2421 if (path->slots[0] == 0)
2426 /* pull out the item */
2427 leaf = path->nodes[0];
2428 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2430 /* make sure the item matches what we want */
2431 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2433 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2436 /* release the path since we're done with it */
2437 btrfs_release_path(path);
2440 * this is where we are basically btrfs_lookup, without the
2441 * crossing root thing. we store the inode number in the
2442 * offset of the orphan item.
2445 if (found_key.offset == last_objectid) {
2446 printk(KERN_ERR "btrfs: Error removing orphan entry, "
2447 "stopping orphan cleanup\n");
2452 last_objectid = found_key.offset;
2454 found_key.objectid = found_key.offset;
2455 found_key.type = BTRFS_INODE_ITEM_KEY;
2456 found_key.offset = 0;
2457 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2458 ret = PTR_RET(inode);
2459 if (ret && ret != -ESTALE)
2462 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2463 struct btrfs_root *dead_root;
2464 struct btrfs_fs_info *fs_info = root->fs_info;
2465 int is_dead_root = 0;
2468 * this is an orphan in the tree root. Currently these
2469 * could come from 2 sources:
2470 * a) a snapshot deletion in progress
2471 * b) a free space cache inode
2472 * We need to distinguish those two, as the snapshot
2473 * orphan must not get deleted.
2474 * find_dead_roots already ran before us, so if this
2475 * is a snapshot deletion, we should find the root
2476 * in the dead_roots list
2478 spin_lock(&fs_info->trans_lock);
2479 list_for_each_entry(dead_root, &fs_info->dead_roots,
2481 if (dead_root->root_key.objectid ==
2482 found_key.objectid) {
2487 spin_unlock(&fs_info->trans_lock);
2489 /* prevent this orphan from being found again */
2490 key.offset = found_key.objectid - 1;
2495 * Inode is already gone but the orphan item is still there,
2496 * kill the orphan item.
2498 if (ret == -ESTALE) {
2499 trans = btrfs_start_transaction(root, 1);
2500 if (IS_ERR(trans)) {
2501 ret = PTR_ERR(trans);
2504 printk(KERN_ERR "auto deleting %Lu\n",
2505 found_key.objectid);
2506 ret = btrfs_del_orphan_item(trans, root,
2507 found_key.objectid);
2508 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2509 btrfs_end_transaction(trans, root);
2514 * add this inode to the orphan list so btrfs_orphan_del does
2515 * the proper thing when we hit it
2517 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2518 &BTRFS_I(inode)->runtime_flags);
2519 atomic_inc(&root->orphan_inodes);
2521 /* if we have links, this was a truncate, lets do that */
2522 if (inode->i_nlink) {
2523 if (!S_ISREG(inode->i_mode)) {
2530 /* 1 for the orphan item deletion. */
2531 trans = btrfs_start_transaction(root, 1);
2532 if (IS_ERR(trans)) {
2533 ret = PTR_ERR(trans);
2536 ret = btrfs_orphan_add(trans, inode);
2537 btrfs_end_transaction(trans, root);
2541 ret = btrfs_truncate(inode);
2543 btrfs_orphan_del(NULL, inode);
2548 /* this will do delete_inode and everything for us */
2553 /* release the path since we're done with it */
2554 btrfs_release_path(path);
2556 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2558 if (root->orphan_block_rsv)
2559 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2562 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2563 trans = btrfs_join_transaction(root);
2565 btrfs_end_transaction(trans, root);
2569 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2571 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2575 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2576 btrfs_free_path(path);
2581 * very simple check to peek ahead in the leaf looking for xattrs. If we
2582 * don't find any xattrs, we know there can't be any acls.
2584 * slot is the slot the inode is in, objectid is the objectid of the inode
2586 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2587 int slot, u64 objectid)
2589 u32 nritems = btrfs_header_nritems(leaf);
2590 struct btrfs_key found_key;
2594 while (slot < nritems) {
2595 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2597 /* we found a different objectid, there must not be acls */
2598 if (found_key.objectid != objectid)
2601 /* we found an xattr, assume we've got an acl */
2602 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2606 * we found a key greater than an xattr key, there can't
2607 * be any acls later on
2609 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2616 * it goes inode, inode backrefs, xattrs, extents,
2617 * so if there are a ton of hard links to an inode there can
2618 * be a lot of backrefs. Don't waste time searching too hard,
2619 * this is just an optimization
2624 /* we hit the end of the leaf before we found an xattr or
2625 * something larger than an xattr. We have to assume the inode
2632 * read an inode from the btree into the in-memory inode
2634 static void btrfs_read_locked_inode(struct inode *inode)
2636 struct btrfs_path *path;
2637 struct extent_buffer *leaf;
2638 struct btrfs_inode_item *inode_item;
2639 struct btrfs_timespec *tspec;
2640 struct btrfs_root *root = BTRFS_I(inode)->root;
2641 struct btrfs_key location;
2645 bool filled = false;
2647 ret = btrfs_fill_inode(inode, &rdev);
2651 path = btrfs_alloc_path();
2655 path->leave_spinning = 1;
2656 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2658 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2662 leaf = path->nodes[0];
2667 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2668 struct btrfs_inode_item);
2669 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2670 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2671 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
2672 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
2673 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2675 tspec = btrfs_inode_atime(inode_item);
2676 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2677 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2679 tspec = btrfs_inode_mtime(inode_item);
2680 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2681 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2683 tspec = btrfs_inode_ctime(inode_item);
2684 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2685 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2687 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2688 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2689 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
2692 * If we were modified in the current generation and evicted from memory
2693 * and then re-read we need to do a full sync since we don't have any
2694 * idea about which extents were modified before we were evicted from
2697 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
2698 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2699 &BTRFS_I(inode)->runtime_flags);
2701 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2702 inode->i_generation = BTRFS_I(inode)->generation;
2704 rdev = btrfs_inode_rdev(leaf, inode_item);
2706 BTRFS_I(inode)->index_cnt = (u64)-1;
2707 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2710 * try to precache a NULL acl entry for files that don't have
2711 * any xattrs or acls
2713 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2716 cache_no_acl(inode);
2718 btrfs_free_path(path);
2720 switch (inode->i_mode & S_IFMT) {
2722 inode->i_mapping->a_ops = &btrfs_aops;
2723 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2724 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2725 inode->i_fop = &btrfs_file_operations;
2726 inode->i_op = &btrfs_file_inode_operations;
2729 inode->i_fop = &btrfs_dir_file_operations;
2730 if (root == root->fs_info->tree_root)
2731 inode->i_op = &btrfs_dir_ro_inode_operations;
2733 inode->i_op = &btrfs_dir_inode_operations;
2736 inode->i_op = &btrfs_symlink_inode_operations;
2737 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2738 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2741 inode->i_op = &btrfs_special_inode_operations;
2742 init_special_inode(inode, inode->i_mode, rdev);
2746 btrfs_update_iflags(inode);
2750 btrfs_free_path(path);
2751 make_bad_inode(inode);
2755 * given a leaf and an inode, copy the inode fields into the leaf
2757 static void fill_inode_item(struct btrfs_trans_handle *trans,
2758 struct extent_buffer *leaf,
2759 struct btrfs_inode_item *item,
2760 struct inode *inode)
2762 struct btrfs_map_token token;
2764 btrfs_init_map_token(&token);
2766 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
2767 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
2768 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
2770 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
2771 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
2773 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
2774 inode->i_atime.tv_sec, &token);
2775 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
2776 inode->i_atime.tv_nsec, &token);
2778 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
2779 inode->i_mtime.tv_sec, &token);
2780 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
2781 inode->i_mtime.tv_nsec, &token);
2783 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
2784 inode->i_ctime.tv_sec, &token);
2785 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
2786 inode->i_ctime.tv_nsec, &token);
2788 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
2790 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
2792 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
2793 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
2794 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
2795 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
2796 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
2800 * copy everything in the in-memory inode into the btree.
2802 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2803 struct btrfs_root *root, struct inode *inode)
2805 struct btrfs_inode_item *inode_item;
2806 struct btrfs_path *path;
2807 struct extent_buffer *leaf;
2810 path = btrfs_alloc_path();
2814 path->leave_spinning = 1;
2815 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2823 btrfs_unlock_up_safe(path, 1);
2824 leaf = path->nodes[0];
2825 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2826 struct btrfs_inode_item);
2828 fill_inode_item(trans, leaf, inode_item, inode);
2829 btrfs_mark_buffer_dirty(leaf);
2830 btrfs_set_inode_last_trans(trans, inode);
2833 btrfs_free_path(path);
2838 * copy everything in the in-memory inode into the btree.
2840 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2841 struct btrfs_root *root, struct inode *inode)
2846 * If the inode is a free space inode, we can deadlock during commit
2847 * if we put it into the delayed code.
2849 * The data relocation inode should also be directly updated
2852 if (!btrfs_is_free_space_inode(inode)
2853 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2854 btrfs_update_root_times(trans, root);
2856 ret = btrfs_delayed_update_inode(trans, root, inode);
2858 btrfs_set_inode_last_trans(trans, inode);
2862 return btrfs_update_inode_item(trans, root, inode);
2865 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2866 struct btrfs_root *root,
2867 struct inode *inode)
2871 ret = btrfs_update_inode(trans, root, inode);
2873 return btrfs_update_inode_item(trans, root, inode);
2878 * unlink helper that gets used here in inode.c and in the tree logging
2879 * recovery code. It remove a link in a directory with a given name, and
2880 * also drops the back refs in the inode to the directory
2882 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2883 struct btrfs_root *root,
2884 struct inode *dir, struct inode *inode,
2885 const char *name, int name_len)
2887 struct btrfs_path *path;
2889 struct extent_buffer *leaf;
2890 struct btrfs_dir_item *di;
2891 struct btrfs_key key;
2893 u64 ino = btrfs_ino(inode);
2894 u64 dir_ino = btrfs_ino(dir);
2896 path = btrfs_alloc_path();
2902 path->leave_spinning = 1;
2903 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2904 name, name_len, -1);
2913 leaf = path->nodes[0];
2914 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2915 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2918 btrfs_release_path(path);
2920 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2923 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2924 "inode %llu parent %llu\n", name_len, name,
2925 (unsigned long long)ino, (unsigned long long)dir_ino);
2926 btrfs_abort_transaction(trans, root, ret);
2930 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2932 btrfs_abort_transaction(trans, root, ret);
2936 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2938 if (ret != 0 && ret != -ENOENT) {
2939 btrfs_abort_transaction(trans, root, ret);
2943 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2948 btrfs_free_path(path);
2952 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2953 inode_inc_iversion(inode);
2954 inode_inc_iversion(dir);
2955 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2956 ret = btrfs_update_inode(trans, root, dir);
2961 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2962 struct btrfs_root *root,
2963 struct inode *dir, struct inode *inode,
2964 const char *name, int name_len)
2967 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2969 btrfs_drop_nlink(inode);
2970 ret = btrfs_update_inode(trans, root, inode);
2976 /* helper to check if there is any shared block in the path */
2977 static int check_path_shared(struct btrfs_root *root,
2978 struct btrfs_path *path)
2980 struct extent_buffer *eb;
2984 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2987 if (!path->nodes[level])
2989 eb = path->nodes[level];
2990 if (!btrfs_block_can_be_shared(root, eb))
2992 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
3001 * helper to start transaction for unlink and rmdir.
3003 * unlink and rmdir are special in btrfs, they do not always free space.
3004 * so in enospc case, we should make sure they will free space before
3005 * allowing them to use the global metadata reservation.
3007 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
3008 struct dentry *dentry)
3010 struct btrfs_trans_handle *trans;
3011 struct btrfs_root *root = BTRFS_I(dir)->root;
3012 struct btrfs_path *path;
3013 struct btrfs_dir_item *di;
3014 struct inode *inode = dentry->d_inode;
3019 u64 ino = btrfs_ino(inode);
3020 u64 dir_ino = btrfs_ino(dir);
3023 * 1 for the possible orphan item
3024 * 1 for the dir item
3025 * 1 for the dir index
3026 * 1 for the inode ref
3027 * 1 for the inode ref in the tree log
3028 * 2 for the dir entries in the log
3031 trans = btrfs_start_transaction(root, 8);
3032 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3035 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
3036 return ERR_PTR(-ENOSPC);
3038 /* check if there is someone else holds reference */
3039 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
3040 return ERR_PTR(-ENOSPC);
3042 if (atomic_read(&inode->i_count) > 2)
3043 return ERR_PTR(-ENOSPC);
3045 if (xchg(&root->fs_info->enospc_unlink, 1))
3046 return ERR_PTR(-ENOSPC);
3048 path = btrfs_alloc_path();
3050 root->fs_info->enospc_unlink = 0;
3051 return ERR_PTR(-ENOMEM);
3054 /* 1 for the orphan item */
3055 trans = btrfs_start_transaction(root, 1);
3056 if (IS_ERR(trans)) {
3057 btrfs_free_path(path);
3058 root->fs_info->enospc_unlink = 0;
3062 path->skip_locking = 1;
3063 path->search_commit_root = 1;
3065 ret = btrfs_lookup_inode(trans, root, path,
3066 &BTRFS_I(dir)->location, 0);
3072 if (check_path_shared(root, path))
3077 btrfs_release_path(path);
3079 ret = btrfs_lookup_inode(trans, root, path,
3080 &BTRFS_I(inode)->location, 0);
3086 if (check_path_shared(root, path))
3091 btrfs_release_path(path);
3093 if (ret == 0 && S_ISREG(inode->i_mode)) {
3094 ret = btrfs_lookup_file_extent(trans, root, path,
3100 BUG_ON(ret == 0); /* Corruption */
3101 if (check_path_shared(root, path))
3103 btrfs_release_path(path);
3111 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3112 dentry->d_name.name, dentry->d_name.len, 0);
3118 if (check_path_shared(root, path))
3124 btrfs_release_path(path);
3126 ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3127 dentry->d_name.len, ino, dir_ino, 0,
3134 if (check_path_shared(root, path))
3137 btrfs_release_path(path);
3140 * This is a commit root search, if we can lookup inode item and other
3141 * relative items in the commit root, it means the transaction of
3142 * dir/file creation has been committed, and the dir index item that we
3143 * delay to insert has also been inserted into the commit root. So
3144 * we needn't worry about the delayed insertion of the dir index item
3147 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3148 dentry->d_name.name, dentry->d_name.len, 0);
3153 BUG_ON(ret == -ENOENT);
3154 if (check_path_shared(root, path))
3159 btrfs_free_path(path);
3160 /* Migrate the orphan reservation over */
3162 err = btrfs_block_rsv_migrate(trans->block_rsv,
3163 &root->fs_info->global_block_rsv,
3164 trans->bytes_reserved);
3167 btrfs_end_transaction(trans, root);
3168 root->fs_info->enospc_unlink = 0;
3169 return ERR_PTR(err);
3172 trans->block_rsv = &root->fs_info->global_block_rsv;
3176 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3177 struct btrfs_root *root)
3179 if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3180 btrfs_block_rsv_release(root, trans->block_rsv,
3181 trans->bytes_reserved);
3182 trans->block_rsv = &root->fs_info->trans_block_rsv;
3183 BUG_ON(!root->fs_info->enospc_unlink);
3184 root->fs_info->enospc_unlink = 0;
3186 btrfs_end_transaction(trans, root);
3189 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3191 struct btrfs_root *root = BTRFS_I(dir)->root;
3192 struct btrfs_trans_handle *trans;
3193 struct inode *inode = dentry->d_inode;
3196 trans = __unlink_start_trans(dir, dentry);
3198 return PTR_ERR(trans);
3200 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3202 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3203 dentry->d_name.name, dentry->d_name.len);
3207 if (inode->i_nlink == 0) {
3208 ret = btrfs_orphan_add(trans, inode);
3214 __unlink_end_trans(trans, root);
3215 btrfs_btree_balance_dirty(root);
3219 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3220 struct btrfs_root *root,
3221 struct inode *dir, u64 objectid,
3222 const char *name, int name_len)
3224 struct btrfs_path *path;
3225 struct extent_buffer *leaf;
3226 struct btrfs_dir_item *di;
3227 struct btrfs_key key;
3230 u64 dir_ino = btrfs_ino(dir);
3232 path = btrfs_alloc_path();
3236 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3237 name, name_len, -1);
3238 if (IS_ERR_OR_NULL(di)) {
3246 leaf = path->nodes[0];
3247 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3248 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3249 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3251 btrfs_abort_transaction(trans, root, ret);
3254 btrfs_release_path(path);
3256 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3257 objectid, root->root_key.objectid,
3258 dir_ino, &index, name, name_len);
3260 if (ret != -ENOENT) {
3261 btrfs_abort_transaction(trans, root, ret);
3264 di = btrfs_search_dir_index_item(root, path, dir_ino,
3266 if (IS_ERR_OR_NULL(di)) {
3271 btrfs_abort_transaction(trans, root, ret);
3275 leaf = path->nodes[0];
3276 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3277 btrfs_release_path(path);
3280 btrfs_release_path(path);
3282 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3284 btrfs_abort_transaction(trans, root, ret);
3288 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3289 inode_inc_iversion(dir);
3290 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3291 ret = btrfs_update_inode_fallback(trans, root, dir);
3293 btrfs_abort_transaction(trans, root, ret);
3295 btrfs_free_path(path);
3299 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3301 struct inode *inode = dentry->d_inode;
3303 struct btrfs_root *root = BTRFS_I(dir)->root;
3304 struct btrfs_trans_handle *trans;
3306 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3308 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3311 trans = __unlink_start_trans(dir, dentry);
3313 return PTR_ERR(trans);
3315 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3316 err = btrfs_unlink_subvol(trans, root, dir,
3317 BTRFS_I(inode)->location.objectid,
3318 dentry->d_name.name,
3319 dentry->d_name.len);
3323 err = btrfs_orphan_add(trans, inode);
3327 /* now the directory is empty */
3328 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3329 dentry->d_name.name, dentry->d_name.len);
3331 btrfs_i_size_write(inode, 0);
3333 __unlink_end_trans(trans, root);
3334 btrfs_btree_balance_dirty(root);
3340 * this can truncate away extent items, csum items and directory items.
3341 * It starts at a high offset and removes keys until it can't find
3342 * any higher than new_size
3344 * csum items that cross the new i_size are truncated to the new size
3347 * min_type is the minimum key type to truncate down to. If set to 0, this
3348 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3350 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3351 struct btrfs_root *root,
3352 struct inode *inode,
3353 u64 new_size, u32 min_type)
3355 struct btrfs_path *path;
3356 struct extent_buffer *leaf;
3357 struct btrfs_file_extent_item *fi;
3358 struct btrfs_key key;
3359 struct btrfs_key found_key;
3360 u64 extent_start = 0;
3361 u64 extent_num_bytes = 0;
3362 u64 extent_offset = 0;
3364 u64 mask = root->sectorsize - 1;
3365 u32 found_type = (u8)-1;
3368 int pending_del_nr = 0;
3369 int pending_del_slot = 0;
3370 int extent_type = -1;
3373 u64 ino = btrfs_ino(inode);
3375 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3377 path = btrfs_alloc_path();
3383 * We want to drop from the next block forward in case this new size is
3384 * not block aligned since we will be keeping the last block of the
3385 * extent just the way it is.
3387 if (root->ref_cows || root == root->fs_info->tree_root)
3388 btrfs_drop_extent_cache(inode, (new_size + mask) & (~mask), (u64)-1, 0);
3391 * This function is also used to drop the items in the log tree before
3392 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3393 * it is used to drop the loged items. So we shouldn't kill the delayed
3396 if (min_type == 0 && root == BTRFS_I(inode)->root)
3397 btrfs_kill_delayed_inode_items(inode);
3400 key.offset = (u64)-1;
3404 path->leave_spinning = 1;
3405 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3412 /* there are no items in the tree for us to truncate, we're
3415 if (path->slots[0] == 0)
3422 leaf = path->nodes[0];
3423 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3424 found_type = btrfs_key_type(&found_key);
3426 if (found_key.objectid != ino)
3429 if (found_type < min_type)
3432 item_end = found_key.offset;
3433 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3434 fi = btrfs_item_ptr(leaf, path->slots[0],
3435 struct btrfs_file_extent_item);
3436 extent_type = btrfs_file_extent_type(leaf, fi);
3437 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3439 btrfs_file_extent_num_bytes(leaf, fi);
3440 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3441 item_end += btrfs_file_extent_inline_len(leaf,
3446 if (found_type > min_type) {
3449 if (item_end < new_size)
3451 if (found_key.offset >= new_size)
3457 /* FIXME, shrink the extent if the ref count is only 1 */
3458 if (found_type != BTRFS_EXTENT_DATA_KEY)
3461 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3463 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3465 u64 orig_num_bytes =
3466 btrfs_file_extent_num_bytes(leaf, fi);
3467 extent_num_bytes = new_size -
3468 found_key.offset + root->sectorsize - 1;
3469 extent_num_bytes = extent_num_bytes &
3470 ~((u64)root->sectorsize - 1);
3471 btrfs_set_file_extent_num_bytes(leaf, fi,
3473 num_dec = (orig_num_bytes -
3475 if (root->ref_cows && extent_start != 0)
3476 inode_sub_bytes(inode, num_dec);
3477 btrfs_mark_buffer_dirty(leaf);
3480 btrfs_file_extent_disk_num_bytes(leaf,
3482 extent_offset = found_key.offset -
3483 btrfs_file_extent_offset(leaf, fi);
3485 /* FIXME blocksize != 4096 */
3486 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3487 if (extent_start != 0) {
3490 inode_sub_bytes(inode, num_dec);
3493 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3495 * we can't truncate inline items that have had
3499 btrfs_file_extent_compression(leaf, fi) == 0 &&
3500 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3501 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3502 u32 size = new_size - found_key.offset;
3504 if (root->ref_cows) {
3505 inode_sub_bytes(inode, item_end + 1 -
3509 btrfs_file_extent_calc_inline_size(size);
3510 btrfs_truncate_item(trans, root, path,
3512 } else if (root->ref_cows) {
3513 inode_sub_bytes(inode, item_end + 1 -
3519 if (!pending_del_nr) {
3520 /* no pending yet, add ourselves */
3521 pending_del_slot = path->slots[0];
3523 } else if (pending_del_nr &&
3524 path->slots[0] + 1 == pending_del_slot) {
3525 /* hop on the pending chunk */
3527 pending_del_slot = path->slots[0];
3534 if (found_extent && (root->ref_cows ||
3535 root == root->fs_info->tree_root)) {
3536 btrfs_set_path_blocking(path);
3537 ret = btrfs_free_extent(trans, root, extent_start,
3538 extent_num_bytes, 0,
3539 btrfs_header_owner(leaf),
3540 ino, extent_offset, 0);
3544 if (found_type == BTRFS_INODE_ITEM_KEY)
3547 if (path->slots[0] == 0 ||
3548 path->slots[0] != pending_del_slot) {
3549 if (pending_del_nr) {
3550 ret = btrfs_del_items(trans, root, path,
3554 btrfs_abort_transaction(trans,
3560 btrfs_release_path(path);
3567 if (pending_del_nr) {
3568 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3571 btrfs_abort_transaction(trans, root, ret);
3574 btrfs_free_path(path);
3579 * btrfs_truncate_page - read, zero a chunk and write a page
3580 * @inode - inode that we're zeroing
3581 * @from - the offset to start zeroing
3582 * @len - the length to zero, 0 to zero the entire range respective to the
3584 * @front - zero up to the offset instead of from the offset on
3586 * This will find the page for the "from" offset and cow the page and zero the
3587 * part we want to zero. This is used with truncate and hole punching.
3589 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
3592 struct address_space *mapping = inode->i_mapping;
3593 struct btrfs_root *root = BTRFS_I(inode)->root;
3594 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3595 struct btrfs_ordered_extent *ordered;
3596 struct extent_state *cached_state = NULL;
3598 u32 blocksize = root->sectorsize;
3599 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3600 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3602 gfp_t mask = btrfs_alloc_write_mask(mapping);
3607 if ((offset & (blocksize - 1)) == 0 &&
3608 (!len || ((len & (blocksize - 1)) == 0)))
3610 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3615 page = find_or_create_page(mapping, index, mask);
3617 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3622 page_start = page_offset(page);
3623 page_end = page_start + PAGE_CACHE_SIZE - 1;
3625 if (!PageUptodate(page)) {
3626 ret = btrfs_readpage(NULL, page);
3628 if (page->mapping != mapping) {
3630 page_cache_release(page);
3633 if (!PageUptodate(page)) {
3638 wait_on_page_writeback(page);
3640 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3641 set_page_extent_mapped(page);
3643 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3645 unlock_extent_cached(io_tree, page_start, page_end,
3646 &cached_state, GFP_NOFS);
3648 page_cache_release(page);
3649 btrfs_start_ordered_extent(inode, ordered, 1);
3650 btrfs_put_ordered_extent(ordered);
3654 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3655 EXTENT_DIRTY | EXTENT_DELALLOC |
3656 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
3657 0, 0, &cached_state, GFP_NOFS);
3659 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3662 unlock_extent_cached(io_tree, page_start, page_end,
3663 &cached_state, GFP_NOFS);
3667 if (offset != PAGE_CACHE_SIZE) {
3669 len = PAGE_CACHE_SIZE - offset;
3672 memset(kaddr, 0, offset);
3674 memset(kaddr + offset, 0, len);
3675 flush_dcache_page(page);
3678 ClearPageChecked(page);
3679 set_page_dirty(page);
3680 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3685 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3687 page_cache_release(page);
3693 * This function puts in dummy file extents for the area we're creating a hole
3694 * for. So if we are truncating this file to a larger size we need to insert
3695 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3696 * the range between oldsize and size
3698 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3700 struct btrfs_trans_handle *trans;
3701 struct btrfs_root *root = BTRFS_I(inode)->root;
3702 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3703 struct extent_map *em = NULL;
3704 struct extent_state *cached_state = NULL;
3705 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3706 u64 mask = root->sectorsize - 1;
3707 u64 hole_start = (oldsize + mask) & ~mask;
3708 u64 block_end = (size + mask) & ~mask;
3714 if (size <= hole_start)
3718 struct btrfs_ordered_extent *ordered;
3719 btrfs_wait_ordered_range(inode, hole_start,
3720 block_end - hole_start);
3721 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3723 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3726 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3727 &cached_state, GFP_NOFS);
3728 btrfs_put_ordered_extent(ordered);
3731 cur_offset = hole_start;
3733 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3734 block_end - cur_offset, 0);
3740 last_byte = min(extent_map_end(em), block_end);
3741 last_byte = (last_byte + mask) & ~mask;
3742 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3743 struct extent_map *hole_em;
3744 hole_size = last_byte - cur_offset;
3746 trans = btrfs_start_transaction(root, 3);
3747 if (IS_ERR(trans)) {
3748 err = PTR_ERR(trans);
3752 err = btrfs_drop_extents(trans, root, inode,
3754 cur_offset + hole_size, 1);
3756 btrfs_abort_transaction(trans, root, err);
3757 btrfs_end_transaction(trans, root);
3761 err = btrfs_insert_file_extent(trans, root,
3762 btrfs_ino(inode), cur_offset, 0,
3763 0, hole_size, 0, hole_size,
3766 btrfs_abort_transaction(trans, root, err);
3767 btrfs_end_transaction(trans, root);
3771 btrfs_drop_extent_cache(inode, cur_offset,
3772 cur_offset + hole_size - 1, 0);
3773 hole_em = alloc_extent_map();
3775 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3776 &BTRFS_I(inode)->runtime_flags);
3779 hole_em->start = cur_offset;
3780 hole_em->len = hole_size;
3781 hole_em->orig_start = cur_offset;
3783 hole_em->block_start = EXTENT_MAP_HOLE;
3784 hole_em->block_len = 0;
3785 hole_em->orig_block_len = 0;
3786 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
3787 hole_em->compress_type = BTRFS_COMPRESS_NONE;
3788 hole_em->generation = trans->transid;
3791 write_lock(&em_tree->lock);
3792 err = add_extent_mapping(em_tree, hole_em);
3794 list_move(&hole_em->list,
3795 &em_tree->modified_extents);
3796 write_unlock(&em_tree->lock);
3799 btrfs_drop_extent_cache(inode, cur_offset,
3803 free_extent_map(hole_em);
3805 btrfs_update_inode(trans, root, inode);
3806 btrfs_end_transaction(trans, root);
3808 free_extent_map(em);
3810 cur_offset = last_byte;
3811 if (cur_offset >= block_end)
3815 free_extent_map(em);
3816 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3821 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
3823 struct btrfs_root *root = BTRFS_I(inode)->root;
3824 struct btrfs_trans_handle *trans;
3825 loff_t oldsize = i_size_read(inode);
3826 loff_t newsize = attr->ia_size;
3827 int mask = attr->ia_valid;
3830 if (newsize == oldsize)
3834 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
3835 * special case where we need to update the times despite not having
3836 * these flags set. For all other operations the VFS set these flags
3837 * explicitly if it wants a timestamp update.
3839 if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
3840 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
3842 if (newsize > oldsize) {
3843 truncate_pagecache(inode, oldsize, newsize);
3844 ret = btrfs_cont_expand(inode, oldsize, newsize);
3848 trans = btrfs_start_transaction(root, 1);
3850 return PTR_ERR(trans);
3852 i_size_write(inode, newsize);
3853 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3854 ret = btrfs_update_inode(trans, root, inode);
3855 btrfs_end_transaction(trans, root);
3859 * We're truncating a file that used to have good data down to
3860 * zero. Make sure it gets into the ordered flush list so that
3861 * any new writes get down to disk quickly.
3864 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3865 &BTRFS_I(inode)->runtime_flags);
3868 * 1 for the orphan item we're going to add
3869 * 1 for the orphan item deletion.
3871 trans = btrfs_start_transaction(root, 2);
3873 return PTR_ERR(trans);
3876 * We need to do this in case we fail at _any_ point during the
3877 * actual truncate. Once we do the truncate_setsize we could
3878 * invalidate pages which forces any outstanding ordered io to
3879 * be instantly completed which will give us extents that need
3880 * to be truncated. If we fail to get an orphan inode down we
3881 * could have left over extents that were never meant to live,
3882 * so we need to garuntee from this point on that everything
3883 * will be consistent.
3885 ret = btrfs_orphan_add(trans, inode);
3886 btrfs_end_transaction(trans, root);
3890 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3891 truncate_setsize(inode, newsize);
3893 /* Disable nonlocked read DIO to avoid the end less truncate */
3894 btrfs_inode_block_unlocked_dio(inode);
3895 inode_dio_wait(inode);
3896 btrfs_inode_resume_unlocked_dio(inode);
3898 ret = btrfs_truncate(inode);
3899 if (ret && inode->i_nlink)
3900 btrfs_orphan_del(NULL, inode);
3906 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3908 struct inode *inode = dentry->d_inode;
3909 struct btrfs_root *root = BTRFS_I(inode)->root;
3912 if (btrfs_root_readonly(root))
3915 err = inode_change_ok(inode, attr);
3919 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3920 err = btrfs_setsize(inode, attr);
3925 if (attr->ia_valid) {
3926 setattr_copy(inode, attr);
3927 inode_inc_iversion(inode);
3928 err = btrfs_dirty_inode(inode);
3930 if (!err && attr->ia_valid & ATTR_MODE)
3931 err = btrfs_acl_chmod(inode);
3937 void btrfs_evict_inode(struct inode *inode)
3939 struct btrfs_trans_handle *trans;
3940 struct btrfs_root *root = BTRFS_I(inode)->root;
3941 struct btrfs_block_rsv *rsv, *global_rsv;
3942 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3945 trace_btrfs_inode_evict(inode);
3947 truncate_inode_pages(&inode->i_data, 0);
3948 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3949 btrfs_is_free_space_inode(inode)))
3952 if (is_bad_inode(inode)) {
3953 btrfs_orphan_del(NULL, inode);
3956 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3957 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3959 if (root->fs_info->log_root_recovering) {
3960 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3961 &BTRFS_I(inode)->runtime_flags));
3965 if (inode->i_nlink > 0) {
3966 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3970 ret = btrfs_commit_inode_delayed_inode(inode);
3972 btrfs_orphan_del(NULL, inode);
3976 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3978 btrfs_orphan_del(NULL, inode);
3981 rsv->size = min_size;
3983 global_rsv = &root->fs_info->global_block_rsv;
3985 btrfs_i_size_write(inode, 0);
3988 * This is a bit simpler than btrfs_truncate since we've already
3989 * reserved our space for our orphan item in the unlink, so we just
3990 * need to reserve some slack space in case we add bytes and update
3991 * inode item when doing the truncate.
3994 ret = btrfs_block_rsv_refill(root, rsv, min_size,
3995 BTRFS_RESERVE_FLUSH_LIMIT);
3998 * Try and steal from the global reserve since we will
3999 * likely not use this space anyway, we want to try as
4000 * hard as possible to get this to work.
4003 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4006 printk(KERN_WARNING "Could not get space for a "
4007 "delete, will truncate on mount %d\n", ret);
4008 btrfs_orphan_del(NULL, inode);
4009 btrfs_free_block_rsv(root, rsv);
4013 trans = btrfs_join_transaction(root);
4014 if (IS_ERR(trans)) {
4015 btrfs_orphan_del(NULL, inode);
4016 btrfs_free_block_rsv(root, rsv);
4020 trans->block_rsv = rsv;
4022 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4026 trans->block_rsv = &root->fs_info->trans_block_rsv;
4027 btrfs_end_transaction(trans, root);
4029 btrfs_btree_balance_dirty(root);
4032 btrfs_free_block_rsv(root, rsv);
4035 trans->block_rsv = root->orphan_block_rsv;
4036 ret = btrfs_orphan_del(trans, inode);
4040 trans->block_rsv = &root->fs_info->trans_block_rsv;
4041 if (!(root == root->fs_info->tree_root ||
4042 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4043 btrfs_return_ino(root, btrfs_ino(inode));
4045 btrfs_end_transaction(trans, root);
4046 btrfs_btree_balance_dirty(root);
4053 * this returns the key found in the dir entry in the location pointer.
4054 * If no dir entries were found, location->objectid is 0.
4056 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4057 struct btrfs_key *location)
4059 const char *name = dentry->d_name.name;
4060 int namelen = dentry->d_name.len;
4061 struct btrfs_dir_item *di;
4062 struct btrfs_path *path;
4063 struct btrfs_root *root = BTRFS_I(dir)->root;
4066 path = btrfs_alloc_path();
4070 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4075 if (IS_ERR_OR_NULL(di))
4078 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4080 btrfs_free_path(path);
4083 location->objectid = 0;
4088 * when we hit a tree root in a directory, the btrfs part of the inode
4089 * needs to be changed to reflect the root directory of the tree root. This
4090 * is kind of like crossing a mount point.
4092 static int fixup_tree_root_location(struct btrfs_root *root,
4094 struct dentry *dentry,
4095 struct btrfs_key *location,
4096 struct btrfs_root **sub_root)
4098 struct btrfs_path *path;
4099 struct btrfs_root *new_root;
4100 struct btrfs_root_ref *ref;
4101 struct extent_buffer *leaf;
4105 path = btrfs_alloc_path();
4112 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4113 BTRFS_I(dir)->root->root_key.objectid,
4114 location->objectid);
4121 leaf = path->nodes[0];
4122 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4123 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4124 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4127 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4128 (unsigned long)(ref + 1),
4129 dentry->d_name.len);
4133 btrfs_release_path(path);
4135 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4136 if (IS_ERR(new_root)) {
4137 err = PTR_ERR(new_root);
4141 if (btrfs_root_refs(&new_root->root_item) == 0) {
4146 *sub_root = new_root;
4147 location->objectid = btrfs_root_dirid(&new_root->root_item);
4148 location->type = BTRFS_INODE_ITEM_KEY;
4149 location->offset = 0;
4152 btrfs_free_path(path);
4156 static void inode_tree_add(struct inode *inode)
4158 struct btrfs_root *root = BTRFS_I(inode)->root;
4159 struct btrfs_inode *entry;
4161 struct rb_node *parent;
4162 u64 ino = btrfs_ino(inode);
4164 p = &root->inode_tree.rb_node;
4167 if (inode_unhashed(inode))
4170 spin_lock(&root->inode_lock);
4173 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4175 if (ino < btrfs_ino(&entry->vfs_inode))
4176 p = &parent->rb_left;
4177 else if (ino > btrfs_ino(&entry->vfs_inode))
4178 p = &parent->rb_right;
4180 WARN_ON(!(entry->vfs_inode.i_state &
4181 (I_WILL_FREE | I_FREEING)));
4182 rb_erase(parent, &root->inode_tree);
4183 RB_CLEAR_NODE(parent);
4184 spin_unlock(&root->inode_lock);
4188 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4189 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4190 spin_unlock(&root->inode_lock);
4193 static void inode_tree_del(struct inode *inode)
4195 struct btrfs_root *root = BTRFS_I(inode)->root;
4198 spin_lock(&root->inode_lock);
4199 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4200 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4201 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4202 empty = RB_EMPTY_ROOT(&root->inode_tree);
4204 spin_unlock(&root->inode_lock);
4207 * Free space cache has inodes in the tree root, but the tree root has a
4208 * root_refs of 0, so this could end up dropping the tree root as a
4209 * snapshot, so we need the extra !root->fs_info->tree_root check to
4210 * make sure we don't drop it.
4212 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4213 root != root->fs_info->tree_root) {
4214 synchronize_srcu(&root->fs_info->subvol_srcu);
4215 spin_lock(&root->inode_lock);
4216 empty = RB_EMPTY_ROOT(&root->inode_tree);
4217 spin_unlock(&root->inode_lock);
4219 btrfs_add_dead_root(root);
4223 void btrfs_invalidate_inodes(struct btrfs_root *root)
4225 struct rb_node *node;
4226 struct rb_node *prev;
4227 struct btrfs_inode *entry;
4228 struct inode *inode;
4231 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4233 spin_lock(&root->inode_lock);
4235 node = root->inode_tree.rb_node;
4239 entry = rb_entry(node, struct btrfs_inode, rb_node);
4241 if (objectid < btrfs_ino(&entry->vfs_inode))
4242 node = node->rb_left;
4243 else if (objectid > btrfs_ino(&entry->vfs_inode))
4244 node = node->rb_right;
4250 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4251 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4255 prev = rb_next(prev);
4259 entry = rb_entry(node, struct btrfs_inode, rb_node);
4260 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4261 inode = igrab(&entry->vfs_inode);
4263 spin_unlock(&root->inode_lock);
4264 if (atomic_read(&inode->i_count) > 1)
4265 d_prune_aliases(inode);
4267 * btrfs_drop_inode will have it removed from
4268 * the inode cache when its usage count
4273 spin_lock(&root->inode_lock);
4277 if (cond_resched_lock(&root->inode_lock))
4280 node = rb_next(node);
4282 spin_unlock(&root->inode_lock);
4285 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4287 struct btrfs_iget_args *args = p;
4288 inode->i_ino = args->ino;
4289 BTRFS_I(inode)->root = args->root;
4293 static int btrfs_find_actor(struct inode *inode, void *opaque)
4295 struct btrfs_iget_args *args = opaque;
4296 return args->ino == btrfs_ino(inode) &&
4297 args->root == BTRFS_I(inode)->root;
4300 static struct inode *btrfs_iget_locked(struct super_block *s,
4302 struct btrfs_root *root)
4304 struct inode *inode;
4305 struct btrfs_iget_args args;
4306 args.ino = objectid;
4309 inode = iget5_locked(s, objectid, btrfs_find_actor,
4310 btrfs_init_locked_inode,
4315 /* Get an inode object given its location and corresponding root.
4316 * Returns in *is_new if the inode was read from disk
4318 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4319 struct btrfs_root *root, int *new)
4321 struct inode *inode;
4323 inode = btrfs_iget_locked(s, location->objectid, root);
4325 return ERR_PTR(-ENOMEM);
4327 if (inode->i_state & I_NEW) {
4328 BTRFS_I(inode)->root = root;
4329 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4330 btrfs_read_locked_inode(inode);
4331 if (!is_bad_inode(inode)) {
4332 inode_tree_add(inode);
4333 unlock_new_inode(inode);
4337 unlock_new_inode(inode);
4339 inode = ERR_PTR(-ESTALE);
4346 static struct inode *new_simple_dir(struct super_block *s,
4347 struct btrfs_key *key,
4348 struct btrfs_root *root)
4350 struct inode *inode = new_inode(s);
4353 return ERR_PTR(-ENOMEM);
4355 BTRFS_I(inode)->root = root;
4356 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4357 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4359 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4360 inode->i_op = &btrfs_dir_ro_inode_operations;
4361 inode->i_fop = &simple_dir_operations;
4362 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4363 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4368 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4370 struct inode *inode;
4371 struct btrfs_root *root = BTRFS_I(dir)->root;
4372 struct btrfs_root *sub_root = root;
4373 struct btrfs_key location;
4377 if (dentry->d_name.len > BTRFS_NAME_LEN)
4378 return ERR_PTR(-ENAMETOOLONG);
4380 ret = btrfs_inode_by_name(dir, dentry, &location);
4382 return ERR_PTR(ret);
4384 if (location.objectid == 0)
4387 if (location.type == BTRFS_INODE_ITEM_KEY) {
4388 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4392 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4394 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4395 ret = fixup_tree_root_location(root, dir, dentry,
4396 &location, &sub_root);
4399 inode = ERR_PTR(ret);
4401 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4403 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4405 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4407 if (!IS_ERR(inode) && root != sub_root) {
4408 down_read(&root->fs_info->cleanup_work_sem);
4409 if (!(inode->i_sb->s_flags & MS_RDONLY))
4410 ret = btrfs_orphan_cleanup(sub_root);
4411 up_read(&root->fs_info->cleanup_work_sem);
4413 inode = ERR_PTR(ret);
4419 static int btrfs_dentry_delete(const struct dentry *dentry)
4421 struct btrfs_root *root;
4422 struct inode *inode = dentry->d_inode;
4424 if (!inode && !IS_ROOT(dentry))
4425 inode = dentry->d_parent->d_inode;
4428 root = BTRFS_I(inode)->root;
4429 if (btrfs_root_refs(&root->root_item) == 0)
4432 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4438 static void btrfs_dentry_release(struct dentry *dentry)
4440 if (dentry->d_fsdata)
4441 kfree(dentry->d_fsdata);
4444 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4449 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4453 unsigned char btrfs_filetype_table[] = {
4454 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4457 static int btrfs_real_readdir(struct file *filp, void *dirent,
4460 struct inode *inode = filp->f_dentry->d_inode;
4461 struct btrfs_root *root = BTRFS_I(inode)->root;
4462 struct btrfs_item *item;
4463 struct btrfs_dir_item *di;
4464 struct btrfs_key key;
4465 struct btrfs_key found_key;
4466 struct btrfs_path *path;
4467 struct list_head ins_list;
4468 struct list_head del_list;
4470 struct extent_buffer *leaf;
4472 unsigned char d_type;
4477 int key_type = BTRFS_DIR_INDEX_KEY;
4481 int is_curr = 0; /* filp->f_pos points to the current index? */
4483 /* FIXME, use a real flag for deciding about the key type */
4484 if (root->fs_info->tree_root == root)
4485 key_type = BTRFS_DIR_ITEM_KEY;
4487 /* special case for "." */
4488 if (filp->f_pos == 0) {
4489 over = filldir(dirent, ".", 1,
4490 filp->f_pos, btrfs_ino(inode), DT_DIR);
4495 /* special case for .., just use the back ref */
4496 if (filp->f_pos == 1) {
4497 u64 pino = parent_ino(filp->f_path.dentry);
4498 over = filldir(dirent, "..", 2,
4499 filp->f_pos, pino, DT_DIR);
4504 path = btrfs_alloc_path();
4510 if (key_type == BTRFS_DIR_INDEX_KEY) {
4511 INIT_LIST_HEAD(&ins_list);
4512 INIT_LIST_HEAD(&del_list);
4513 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4516 btrfs_set_key_type(&key, key_type);
4517 key.offset = filp->f_pos;
4518 key.objectid = btrfs_ino(inode);
4520 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4525 leaf = path->nodes[0];
4526 slot = path->slots[0];
4527 if (slot >= btrfs_header_nritems(leaf)) {
4528 ret = btrfs_next_leaf(root, path);
4536 item = btrfs_item_nr(leaf, slot);
4537 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4539 if (found_key.objectid != key.objectid)
4541 if (btrfs_key_type(&found_key) != key_type)
4543 if (found_key.offset < filp->f_pos)
4545 if (key_type == BTRFS_DIR_INDEX_KEY &&
4546 btrfs_should_delete_dir_index(&del_list,
4550 filp->f_pos = found_key.offset;
4553 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4555 di_total = btrfs_item_size(leaf, item);
4557 while (di_cur < di_total) {
4558 struct btrfs_key location;
4560 if (verify_dir_item(root, leaf, di))
4563 name_len = btrfs_dir_name_len(leaf, di);
4564 if (name_len <= sizeof(tmp_name)) {
4565 name_ptr = tmp_name;
4567 name_ptr = kmalloc(name_len, GFP_NOFS);
4573 read_extent_buffer(leaf, name_ptr,
4574 (unsigned long)(di + 1), name_len);
4576 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4577 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4580 /* is this a reference to our own snapshot? If so
4583 * In contrast to old kernels, we insert the snapshot's
4584 * dir item and dir index after it has been created, so
4585 * we won't find a reference to our own snapshot. We
4586 * still keep the following code for backward
4589 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4590 location.objectid == root->root_key.objectid) {
4594 over = filldir(dirent, name_ptr, name_len,
4595 found_key.offset, location.objectid,
4599 if (name_ptr != tmp_name)
4604 di_len = btrfs_dir_name_len(leaf, di) +
4605 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4607 di = (struct btrfs_dir_item *)((char *)di + di_len);
4613 if (key_type == BTRFS_DIR_INDEX_KEY) {
4616 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4622 /* Reached end of directory/root. Bump pos past the last item. */
4623 if (key_type == BTRFS_DIR_INDEX_KEY)
4625 * 32-bit glibc will use getdents64, but then strtol -
4626 * so the last number we can serve is this.
4628 filp->f_pos = 0x7fffffff;
4634 if (key_type == BTRFS_DIR_INDEX_KEY)
4635 btrfs_put_delayed_items(&ins_list, &del_list);
4636 btrfs_free_path(path);
4640 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4642 struct btrfs_root *root = BTRFS_I(inode)->root;
4643 struct btrfs_trans_handle *trans;
4645 bool nolock = false;
4647 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4650 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
4653 if (wbc->sync_mode == WB_SYNC_ALL) {
4655 trans = btrfs_join_transaction_nolock(root);
4657 trans = btrfs_join_transaction(root);
4659 return PTR_ERR(trans);
4660 ret = btrfs_commit_transaction(trans, root);
4666 * This is somewhat expensive, updating the tree every time the
4667 * inode changes. But, it is most likely to find the inode in cache.
4668 * FIXME, needs more benchmarking...there are no reasons other than performance
4669 * to keep or drop this code.
4671 int btrfs_dirty_inode(struct inode *inode)
4673 struct btrfs_root *root = BTRFS_I(inode)->root;
4674 struct btrfs_trans_handle *trans;
4677 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4680 trans = btrfs_join_transaction(root);
4682 return PTR_ERR(trans);
4684 ret = btrfs_update_inode(trans, root, inode);
4685 if (ret && ret == -ENOSPC) {
4686 /* whoops, lets try again with the full transaction */
4687 btrfs_end_transaction(trans, root);
4688 trans = btrfs_start_transaction(root, 1);
4690 return PTR_ERR(trans);
4692 ret = btrfs_update_inode(trans, root, inode);
4694 btrfs_end_transaction(trans, root);
4695 if (BTRFS_I(inode)->delayed_node)
4696 btrfs_balance_delayed_items(root);
4702 * This is a copy of file_update_time. We need this so we can return error on
4703 * ENOSPC for updating the inode in the case of file write and mmap writes.
4705 static int btrfs_update_time(struct inode *inode, struct timespec *now,
4708 struct btrfs_root *root = BTRFS_I(inode)->root;
4710 if (btrfs_root_readonly(root))
4713 if (flags & S_VERSION)
4714 inode_inc_iversion(inode);
4715 if (flags & S_CTIME)
4716 inode->i_ctime = *now;
4717 if (flags & S_MTIME)
4718 inode->i_mtime = *now;
4719 if (flags & S_ATIME)
4720 inode->i_atime = *now;
4721 return btrfs_dirty_inode(inode);
4725 * find the highest existing sequence number in a directory
4726 * and then set the in-memory index_cnt variable to reflect
4727 * free sequence numbers
4729 static int btrfs_set_inode_index_count(struct inode *inode)
4731 struct btrfs_root *root = BTRFS_I(inode)->root;
4732 struct btrfs_key key, found_key;
4733 struct btrfs_path *path;
4734 struct extent_buffer *leaf;
4737 key.objectid = btrfs_ino(inode);
4738 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4739 key.offset = (u64)-1;
4741 path = btrfs_alloc_path();
4745 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4748 /* FIXME: we should be able to handle this */
4754 * MAGIC NUMBER EXPLANATION:
4755 * since we search a directory based on f_pos we have to start at 2
4756 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4757 * else has to start at 2
4759 if (path->slots[0] == 0) {
4760 BTRFS_I(inode)->index_cnt = 2;
4766 leaf = path->nodes[0];
4767 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4769 if (found_key.objectid != btrfs_ino(inode) ||
4770 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4771 BTRFS_I(inode)->index_cnt = 2;
4775 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4777 btrfs_free_path(path);
4782 * helper to find a free sequence number in a given directory. This current
4783 * code is very simple, later versions will do smarter things in the btree
4785 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4789 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4790 ret = btrfs_inode_delayed_dir_index_count(dir);
4792 ret = btrfs_set_inode_index_count(dir);
4798 *index = BTRFS_I(dir)->index_cnt;
4799 BTRFS_I(dir)->index_cnt++;
4804 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4805 struct btrfs_root *root,
4807 const char *name, int name_len,
4808 u64 ref_objectid, u64 objectid,
4809 umode_t mode, u64 *index)
4811 struct inode *inode;
4812 struct btrfs_inode_item *inode_item;
4813 struct btrfs_key *location;
4814 struct btrfs_path *path;
4815 struct btrfs_inode_ref *ref;
4816 struct btrfs_key key[2];
4822 path = btrfs_alloc_path();
4824 return ERR_PTR(-ENOMEM);
4826 inode = new_inode(root->fs_info->sb);
4828 btrfs_free_path(path);
4829 return ERR_PTR(-ENOMEM);
4833 * we have to initialize this early, so we can reclaim the inode
4834 * number if we fail afterwards in this function.
4836 inode->i_ino = objectid;
4839 trace_btrfs_inode_request(dir);
4841 ret = btrfs_set_inode_index(dir, index);
4843 btrfs_free_path(path);
4845 return ERR_PTR(ret);
4849 * index_cnt is ignored for everything but a dir,
4850 * btrfs_get_inode_index_count has an explanation for the magic
4853 BTRFS_I(inode)->index_cnt = 2;
4854 BTRFS_I(inode)->root = root;
4855 BTRFS_I(inode)->generation = trans->transid;
4856 inode->i_generation = BTRFS_I(inode)->generation;
4859 * We could have gotten an inode number from somebody who was fsynced
4860 * and then removed in this same transaction, so let's just set full
4861 * sync since it will be a full sync anyway and this will blow away the
4862 * old info in the log.
4864 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
4871 key[0].objectid = objectid;
4872 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4876 * Start new inodes with an inode_ref. This is slightly more
4877 * efficient for small numbers of hard links since they will
4878 * be packed into one item. Extended refs will kick in if we
4879 * add more hard links than can fit in the ref item.
4881 key[1].objectid = objectid;
4882 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4883 key[1].offset = ref_objectid;
4885 sizes[0] = sizeof(struct btrfs_inode_item);
4886 sizes[1] = name_len + sizeof(*ref);
4888 path->leave_spinning = 1;
4889 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4893 inode_init_owner(inode, dir, mode);
4894 inode_set_bytes(inode, 0);
4895 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4896 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4897 struct btrfs_inode_item);
4898 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
4899 sizeof(*inode_item));
4900 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4902 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4903 struct btrfs_inode_ref);
4904 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4905 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4906 ptr = (unsigned long)(ref + 1);
4907 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4909 btrfs_mark_buffer_dirty(path->nodes[0]);
4910 btrfs_free_path(path);
4912 location = &BTRFS_I(inode)->location;
4913 location->objectid = objectid;
4914 location->offset = 0;
4915 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4917 btrfs_inherit_iflags(inode, dir);
4919 if (S_ISREG(mode)) {
4920 if (btrfs_test_opt(root, NODATASUM))
4921 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4922 if (btrfs_test_opt(root, NODATACOW))
4923 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4926 insert_inode_hash(inode);
4927 inode_tree_add(inode);
4929 trace_btrfs_inode_new(inode);
4930 btrfs_set_inode_last_trans(trans, inode);
4932 btrfs_update_root_times(trans, root);
4937 BTRFS_I(dir)->index_cnt--;
4938 btrfs_free_path(path);
4940 return ERR_PTR(ret);
4943 static inline u8 btrfs_inode_type(struct inode *inode)
4945 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4949 * utility function to add 'inode' into 'parent_inode' with
4950 * a give name and a given sequence number.
4951 * if 'add_backref' is true, also insert a backref from the
4952 * inode to the parent directory.
4954 int btrfs_add_link(struct btrfs_trans_handle *trans,
4955 struct inode *parent_inode, struct inode *inode,
4956 const char *name, int name_len, int add_backref, u64 index)
4959 struct btrfs_key key;
4960 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4961 u64 ino = btrfs_ino(inode);
4962 u64 parent_ino = btrfs_ino(parent_inode);
4964 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4965 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4968 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4972 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4973 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4974 key.objectid, root->root_key.objectid,
4975 parent_ino, index, name, name_len);
4976 } else if (add_backref) {
4977 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4981 /* Nothing to clean up yet */
4985 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4987 btrfs_inode_type(inode), index);
4988 if (ret == -EEXIST || ret == -EOVERFLOW)
4991 btrfs_abort_transaction(trans, root, ret);
4995 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4997 inode_inc_iversion(parent_inode);
4998 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4999 ret = btrfs_update_inode(trans, root, parent_inode);
5001 btrfs_abort_transaction(trans, root, ret);
5005 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5008 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5009 key.objectid, root->root_key.objectid,
5010 parent_ino, &local_index, name, name_len);
5012 } else if (add_backref) {
5016 err = btrfs_del_inode_ref(trans, root, name, name_len,
5017 ino, parent_ino, &local_index);
5022 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5023 struct inode *dir, struct dentry *dentry,
5024 struct inode *inode, int backref, u64 index)
5026 int err = btrfs_add_link(trans, dir, inode,
5027 dentry->d_name.name, dentry->d_name.len,
5034 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5035 umode_t mode, dev_t rdev)
5037 struct btrfs_trans_handle *trans;
5038 struct btrfs_root *root = BTRFS_I(dir)->root;
5039 struct inode *inode = NULL;
5045 if (!new_valid_dev(rdev))
5049 * 2 for inode item and ref
5051 * 1 for xattr if selinux is on
5053 trans = btrfs_start_transaction(root, 5);
5055 return PTR_ERR(trans);
5057 err = btrfs_find_free_ino(root, &objectid);
5061 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5062 dentry->d_name.len, btrfs_ino(dir), objectid,
5064 if (IS_ERR(inode)) {
5065 err = PTR_ERR(inode);
5069 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5076 * If the active LSM wants to access the inode during
5077 * d_instantiate it needs these. Smack checks to see
5078 * if the filesystem supports xattrs by looking at the
5082 inode->i_op = &btrfs_special_inode_operations;
5083 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5087 init_special_inode(inode, inode->i_mode, rdev);
5088 btrfs_update_inode(trans, root, inode);
5089 d_instantiate(dentry, inode);
5092 btrfs_end_transaction(trans, root);
5093 btrfs_btree_balance_dirty(root);
5095 inode_dec_link_count(inode);
5101 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5102 umode_t mode, bool excl)
5104 struct btrfs_trans_handle *trans;
5105 struct btrfs_root *root = BTRFS_I(dir)->root;
5106 struct inode *inode = NULL;
5107 int drop_inode_on_err = 0;
5113 * 2 for inode item and ref
5115 * 1 for xattr if selinux is on
5117 trans = btrfs_start_transaction(root, 5);
5119 return PTR_ERR(trans);
5121 err = btrfs_find_free_ino(root, &objectid);
5125 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5126 dentry->d_name.len, btrfs_ino(dir), objectid,
5128 if (IS_ERR(inode)) {
5129 err = PTR_ERR(inode);
5132 drop_inode_on_err = 1;
5134 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5138 err = btrfs_update_inode(trans, root, inode);
5143 * If the active LSM wants to access the inode during
5144 * d_instantiate it needs these. Smack checks to see
5145 * if the filesystem supports xattrs by looking at the
5148 inode->i_fop = &btrfs_file_operations;
5149 inode->i_op = &btrfs_file_inode_operations;
5151 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5155 inode->i_mapping->a_ops = &btrfs_aops;
5156 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5157 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5158 d_instantiate(dentry, inode);
5161 btrfs_end_transaction(trans, root);
5162 if (err && drop_inode_on_err) {
5163 inode_dec_link_count(inode);
5166 btrfs_btree_balance_dirty(root);
5170 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5171 struct dentry *dentry)
5173 struct btrfs_trans_handle *trans;
5174 struct btrfs_root *root = BTRFS_I(dir)->root;
5175 struct inode *inode = old_dentry->d_inode;
5180 /* do not allow sys_link's with other subvols of the same device */
5181 if (root->objectid != BTRFS_I(inode)->root->objectid)
5184 if (inode->i_nlink >= BTRFS_LINK_MAX)
5187 err = btrfs_set_inode_index(dir, &index);
5192 * 2 items for inode and inode ref
5193 * 2 items for dir items
5194 * 1 item for parent inode
5196 trans = btrfs_start_transaction(root, 5);
5197 if (IS_ERR(trans)) {
5198 err = PTR_ERR(trans);
5202 btrfs_inc_nlink(inode);
5203 inode_inc_iversion(inode);
5204 inode->i_ctime = CURRENT_TIME;
5206 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5208 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5213 struct dentry *parent = dentry->d_parent;
5214 err = btrfs_update_inode(trans, root, inode);
5217 d_instantiate(dentry, inode);
5218 btrfs_log_new_name(trans, inode, NULL, parent);
5221 btrfs_end_transaction(trans, root);
5224 inode_dec_link_count(inode);
5227 btrfs_btree_balance_dirty(root);
5231 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5233 struct inode *inode = NULL;
5234 struct btrfs_trans_handle *trans;
5235 struct btrfs_root *root = BTRFS_I(dir)->root;
5237 int drop_on_err = 0;
5242 * 2 items for inode and ref
5243 * 2 items for dir items
5244 * 1 for xattr if selinux is on
5246 trans = btrfs_start_transaction(root, 5);
5248 return PTR_ERR(trans);
5250 err = btrfs_find_free_ino(root, &objectid);
5254 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5255 dentry->d_name.len, btrfs_ino(dir), objectid,
5256 S_IFDIR | mode, &index);
5257 if (IS_ERR(inode)) {
5258 err = PTR_ERR(inode);
5264 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5268 inode->i_op = &btrfs_dir_inode_operations;
5269 inode->i_fop = &btrfs_dir_file_operations;
5271 btrfs_i_size_write(inode, 0);
5272 err = btrfs_update_inode(trans, root, inode);
5276 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5277 dentry->d_name.len, 0, index);
5281 d_instantiate(dentry, inode);
5285 btrfs_end_transaction(trans, root);
5288 btrfs_btree_balance_dirty(root);
5292 /* helper for btfs_get_extent. Given an existing extent in the tree,
5293 * and an extent that you want to insert, deal with overlap and insert
5294 * the new extent into the tree.
5296 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5297 struct extent_map *existing,
5298 struct extent_map *em,
5299 u64 map_start, u64 map_len)
5303 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5304 start_diff = map_start - em->start;
5305 em->start = map_start;
5307 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5308 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5309 em->block_start += start_diff;
5310 em->block_len -= start_diff;
5312 return add_extent_mapping(em_tree, em);
5315 static noinline int uncompress_inline(struct btrfs_path *path,
5316 struct inode *inode, struct page *page,
5317 size_t pg_offset, u64 extent_offset,
5318 struct btrfs_file_extent_item *item)
5321 struct extent_buffer *leaf = path->nodes[0];
5324 unsigned long inline_size;
5328 WARN_ON(pg_offset != 0);
5329 compress_type = btrfs_file_extent_compression(leaf, item);
5330 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5331 inline_size = btrfs_file_extent_inline_item_len(leaf,
5332 btrfs_item_nr(leaf, path->slots[0]));
5333 tmp = kmalloc(inline_size, GFP_NOFS);
5336 ptr = btrfs_file_extent_inline_start(item);
5338 read_extent_buffer(leaf, tmp, ptr, inline_size);
5340 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5341 ret = btrfs_decompress(compress_type, tmp, page,
5342 extent_offset, inline_size, max_size);
5344 char *kaddr = kmap_atomic(page);
5345 unsigned long copy_size = min_t(u64,
5346 PAGE_CACHE_SIZE - pg_offset,
5347 max_size - extent_offset);
5348 memset(kaddr + pg_offset, 0, copy_size);
5349 kunmap_atomic(kaddr);
5356 * a bit scary, this does extent mapping from logical file offset to the disk.
5357 * the ugly parts come from merging extents from the disk with the in-ram
5358 * representation. This gets more complex because of the data=ordered code,
5359 * where the in-ram extents might be locked pending data=ordered completion.
5361 * This also copies inline extents directly into the page.
5364 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5365 size_t pg_offset, u64 start, u64 len,
5371 u64 extent_start = 0;
5373 u64 objectid = btrfs_ino(inode);
5375 struct btrfs_path *path = NULL;
5376 struct btrfs_root *root = BTRFS_I(inode)->root;
5377 struct btrfs_file_extent_item *item;
5378 struct extent_buffer *leaf;
5379 struct btrfs_key found_key;
5380 struct extent_map *em = NULL;
5381 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5382 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5383 struct btrfs_trans_handle *trans = NULL;
5387 read_lock(&em_tree->lock);
5388 em = lookup_extent_mapping(em_tree, start, len);
5390 em->bdev = root->fs_info->fs_devices->latest_bdev;
5391 read_unlock(&em_tree->lock);
5394 if (em->start > start || em->start + em->len <= start)
5395 free_extent_map(em);
5396 else if (em->block_start == EXTENT_MAP_INLINE && page)
5397 free_extent_map(em);
5401 em = alloc_extent_map();
5406 em->bdev = root->fs_info->fs_devices->latest_bdev;
5407 em->start = EXTENT_MAP_HOLE;
5408 em->orig_start = EXTENT_MAP_HOLE;
5410 em->block_len = (u64)-1;
5413 path = btrfs_alloc_path();
5419 * Chances are we'll be called again, so go ahead and do
5425 ret = btrfs_lookup_file_extent(trans, root, path,
5426 objectid, start, trans != NULL);
5433 if (path->slots[0] == 0)
5438 leaf = path->nodes[0];
5439 item = btrfs_item_ptr(leaf, path->slots[0],
5440 struct btrfs_file_extent_item);
5441 /* are we inside the extent that was found? */
5442 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5443 found_type = btrfs_key_type(&found_key);
5444 if (found_key.objectid != objectid ||
5445 found_type != BTRFS_EXTENT_DATA_KEY) {
5449 found_type = btrfs_file_extent_type(leaf, item);
5450 extent_start = found_key.offset;
5451 compress_type = btrfs_file_extent_compression(leaf, item);
5452 if (found_type == BTRFS_FILE_EXTENT_REG ||
5453 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5454 extent_end = extent_start +
5455 btrfs_file_extent_num_bytes(leaf, item);
5456 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5458 size = btrfs_file_extent_inline_len(leaf, item);
5459 extent_end = (extent_start + size + root->sectorsize - 1) &
5460 ~((u64)root->sectorsize - 1);
5463 if (start >= extent_end) {
5465 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5466 ret = btrfs_next_leaf(root, path);
5473 leaf = path->nodes[0];
5475 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5476 if (found_key.objectid != objectid ||
5477 found_key.type != BTRFS_EXTENT_DATA_KEY)
5479 if (start + len <= found_key.offset)
5482 em->orig_start = start;
5483 em->len = found_key.offset - start;
5487 if (found_type == BTRFS_FILE_EXTENT_REG ||
5488 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5489 em->start = extent_start;
5490 em->len = extent_end - extent_start;
5491 em->orig_start = extent_start -
5492 btrfs_file_extent_offset(leaf, item);
5493 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
5495 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5497 em->block_start = EXTENT_MAP_HOLE;
5500 if (compress_type != BTRFS_COMPRESS_NONE) {
5501 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5502 em->compress_type = compress_type;
5503 em->block_start = bytenr;
5504 em->block_len = em->orig_block_len;
5506 bytenr += btrfs_file_extent_offset(leaf, item);
5507 em->block_start = bytenr;
5508 em->block_len = em->len;
5509 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5510 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5513 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5517 size_t extent_offset;
5520 em->block_start = EXTENT_MAP_INLINE;
5521 if (!page || create) {
5522 em->start = extent_start;
5523 em->len = extent_end - extent_start;
5527 size = btrfs_file_extent_inline_len(leaf, item);
5528 extent_offset = page_offset(page) + pg_offset - extent_start;
5529 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5530 size - extent_offset);
5531 em->start = extent_start + extent_offset;
5532 em->len = (copy_size + root->sectorsize - 1) &
5533 ~((u64)root->sectorsize - 1);
5534 em->orig_block_len = em->len;
5535 em->orig_start = em->start;
5536 if (compress_type) {
5537 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5538 em->compress_type = compress_type;
5540 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5541 if (create == 0 && !PageUptodate(page)) {
5542 if (btrfs_file_extent_compression(leaf, item) !=
5543 BTRFS_COMPRESS_NONE) {
5544 ret = uncompress_inline(path, inode, page,
5546 extent_offset, item);
5547 BUG_ON(ret); /* -ENOMEM */
5550 read_extent_buffer(leaf, map + pg_offset, ptr,
5552 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5553 memset(map + pg_offset + copy_size, 0,
5554 PAGE_CACHE_SIZE - pg_offset -
5559 flush_dcache_page(page);
5560 } else if (create && PageUptodate(page)) {
5564 free_extent_map(em);
5567 btrfs_release_path(path);
5568 trans = btrfs_join_transaction(root);
5571 return ERR_CAST(trans);
5575 write_extent_buffer(leaf, map + pg_offset, ptr,
5578 btrfs_mark_buffer_dirty(leaf);
5580 set_extent_uptodate(io_tree, em->start,
5581 extent_map_end(em) - 1, NULL, GFP_NOFS);
5584 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
5588 em->orig_start = start;
5591 em->block_start = EXTENT_MAP_HOLE;
5592 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5594 btrfs_release_path(path);
5595 if (em->start > start || extent_map_end(em) <= start) {
5596 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5597 "[%llu %llu]\n", (unsigned long long)em->start,
5598 (unsigned long long)em->len,
5599 (unsigned long long)start,
5600 (unsigned long long)len);
5606 write_lock(&em_tree->lock);
5607 ret = add_extent_mapping(em_tree, em);
5608 /* it is possible that someone inserted the extent into the tree
5609 * while we had the lock dropped. It is also possible that
5610 * an overlapping map exists in the tree
5612 if (ret == -EEXIST) {
5613 struct extent_map *existing;
5617 existing = lookup_extent_mapping(em_tree, start, len);
5618 if (existing && (existing->start > start ||
5619 existing->start + existing->len <= start)) {
5620 free_extent_map(existing);
5624 existing = lookup_extent_mapping(em_tree, em->start,
5627 err = merge_extent_mapping(em_tree, existing,
5630 free_extent_map(existing);
5632 free_extent_map(em);
5637 free_extent_map(em);
5641 free_extent_map(em);
5646 write_unlock(&em_tree->lock);
5650 trace_btrfs_get_extent(root, em);
5653 btrfs_free_path(path);
5655 ret = btrfs_end_transaction(trans, root);
5660 free_extent_map(em);
5661 return ERR_PTR(err);
5663 BUG_ON(!em); /* Error is always set */
5667 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5668 size_t pg_offset, u64 start, u64 len,
5671 struct extent_map *em;
5672 struct extent_map *hole_em = NULL;
5673 u64 range_start = start;
5679 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5686 * - a pre-alloc extent,
5687 * there might actually be delalloc bytes behind it.
5689 if (em->block_start != EXTENT_MAP_HOLE &&
5690 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5696 /* check to see if we've wrapped (len == -1 or similar) */
5705 /* ok, we didn't find anything, lets look for delalloc */
5706 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5707 end, len, EXTENT_DELALLOC, 1);
5708 found_end = range_start + found;
5709 if (found_end < range_start)
5710 found_end = (u64)-1;
5713 * we didn't find anything useful, return
5714 * the original results from get_extent()
5716 if (range_start > end || found_end <= start) {
5722 /* adjust the range_start to make sure it doesn't
5723 * go backwards from the start they passed in
5725 range_start = max(start,range_start);
5726 found = found_end - range_start;
5729 u64 hole_start = start;
5732 em = alloc_extent_map();
5738 * when btrfs_get_extent can't find anything it
5739 * returns one huge hole
5741 * make sure what it found really fits our range, and
5742 * adjust to make sure it is based on the start from
5746 u64 calc_end = extent_map_end(hole_em);
5748 if (calc_end <= start || (hole_em->start > end)) {
5749 free_extent_map(hole_em);
5752 hole_start = max(hole_em->start, start);
5753 hole_len = calc_end - hole_start;
5757 if (hole_em && range_start > hole_start) {
5758 /* our hole starts before our delalloc, so we
5759 * have to return just the parts of the hole
5760 * that go until the delalloc starts
5762 em->len = min(hole_len,
5763 range_start - hole_start);
5764 em->start = hole_start;
5765 em->orig_start = hole_start;
5767 * don't adjust block start at all,
5768 * it is fixed at EXTENT_MAP_HOLE
5770 em->block_start = hole_em->block_start;
5771 em->block_len = hole_len;
5772 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
5773 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5775 em->start = range_start;
5777 em->orig_start = range_start;
5778 em->block_start = EXTENT_MAP_DELALLOC;
5779 em->block_len = found;
5781 } else if (hole_em) {
5786 free_extent_map(hole_em);
5788 free_extent_map(em);
5789 return ERR_PTR(err);
5794 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5797 struct btrfs_root *root = BTRFS_I(inode)->root;
5798 struct btrfs_trans_handle *trans;
5799 struct extent_map *em;
5800 struct btrfs_key ins;
5804 trans = btrfs_join_transaction(root);
5806 return ERR_CAST(trans);
5808 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5810 alloc_hint = get_extent_allocation_hint(inode, start, len);
5811 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5812 alloc_hint, &ins, 1);
5818 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
5819 ins.offset, ins.offset, 0);
5823 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5824 ins.offset, ins.offset, 0);
5826 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5830 btrfs_end_transaction(trans, root);
5835 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5836 * block must be cow'd
5838 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5839 struct inode *inode, u64 offset, u64 len)
5841 struct btrfs_path *path;
5843 struct extent_buffer *leaf;
5844 struct btrfs_root *root = BTRFS_I(inode)->root;
5845 struct btrfs_file_extent_item *fi;
5846 struct btrfs_key key;
5854 path = btrfs_alloc_path();
5858 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5863 slot = path->slots[0];
5866 /* can't find the item, must cow */
5873 leaf = path->nodes[0];
5874 btrfs_item_key_to_cpu(leaf, &key, slot);
5875 if (key.objectid != btrfs_ino(inode) ||
5876 key.type != BTRFS_EXTENT_DATA_KEY) {
5877 /* not our file or wrong item type, must cow */
5881 if (key.offset > offset) {
5882 /* Wrong offset, must cow */
5886 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5887 found_type = btrfs_file_extent_type(leaf, fi);
5888 if (found_type != BTRFS_FILE_EXTENT_REG &&
5889 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5890 /* not a regular extent, must cow */
5893 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5894 backref_offset = btrfs_file_extent_offset(leaf, fi);
5896 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5897 if (extent_end < offset + len) {
5898 /* extent doesn't include our full range, must cow */
5902 if (btrfs_extent_readonly(root, disk_bytenr))
5906 * look for other files referencing this extent, if we
5907 * find any we must cow
5909 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5910 key.offset - backref_offset, disk_bytenr))
5914 * adjust disk_bytenr and num_bytes to cover just the bytes
5915 * in this extent we are about to write. If there
5916 * are any csums in that range we have to cow in order
5917 * to keep the csums correct
5919 disk_bytenr += backref_offset;
5920 disk_bytenr += offset - key.offset;
5921 num_bytes = min(offset + len, extent_end) - offset;
5922 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5925 * all of the above have passed, it is safe to overwrite this extent
5930 btrfs_free_path(path);
5934 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
5935 struct extent_state **cached_state, int writing)
5937 struct btrfs_ordered_extent *ordered;
5941 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5944 * We're concerned with the entire range that we're going to be
5945 * doing DIO to, so we need to make sure theres no ordered
5946 * extents in this range.
5948 ordered = btrfs_lookup_ordered_range(inode, lockstart,
5949 lockend - lockstart + 1);
5952 * We need to make sure there are no buffered pages in this
5953 * range either, we could have raced between the invalidate in
5954 * generic_file_direct_write and locking the extent. The
5955 * invalidate needs to happen so that reads after a write do not
5958 if (!ordered && (!writing ||
5959 !test_range_bit(&BTRFS_I(inode)->io_tree,
5960 lockstart, lockend, EXTENT_UPTODATE, 0,
5964 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5965 cached_state, GFP_NOFS);
5968 btrfs_start_ordered_extent(inode, ordered, 1);
5969 btrfs_put_ordered_extent(ordered);
5971 /* Screw you mmap */
5972 ret = filemap_write_and_wait_range(inode->i_mapping,
5979 * If we found a page that couldn't be invalidated just
5980 * fall back to buffered.
5982 ret = invalidate_inode_pages2_range(inode->i_mapping,
5983 lockstart >> PAGE_CACHE_SHIFT,
5984 lockend >> PAGE_CACHE_SHIFT);
5995 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
5996 u64 len, u64 orig_start,
5997 u64 block_start, u64 block_len,
5998 u64 orig_block_len, int type)
6000 struct extent_map_tree *em_tree;
6001 struct extent_map *em;
6002 struct btrfs_root *root = BTRFS_I(inode)->root;
6005 em_tree = &BTRFS_I(inode)->extent_tree;
6006 em = alloc_extent_map();
6008 return ERR_PTR(-ENOMEM);
6011 em->orig_start = orig_start;
6012 em->mod_start = start;
6015 em->block_len = block_len;
6016 em->block_start = block_start;
6017 em->bdev = root->fs_info->fs_devices->latest_bdev;
6018 em->orig_block_len = orig_block_len;
6019 em->generation = -1;
6020 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6021 if (type == BTRFS_ORDERED_PREALLOC)
6022 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6025 btrfs_drop_extent_cache(inode, em->start,
6026 em->start + em->len - 1, 0);
6027 write_lock(&em_tree->lock);
6028 ret = add_extent_mapping(em_tree, em);
6030 list_move(&em->list,
6031 &em_tree->modified_extents);
6032 write_unlock(&em_tree->lock);
6033 } while (ret == -EEXIST);
6036 free_extent_map(em);
6037 return ERR_PTR(ret);
6044 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6045 struct buffer_head *bh_result, int create)
6047 struct extent_map *em;
6048 struct btrfs_root *root = BTRFS_I(inode)->root;
6049 struct extent_state *cached_state = NULL;
6050 u64 start = iblock << inode->i_blkbits;
6051 u64 lockstart, lockend;
6052 u64 len = bh_result->b_size;
6053 struct btrfs_trans_handle *trans;
6054 int unlock_bits = EXTENT_LOCKED;
6058 spin_lock(&BTRFS_I(inode)->lock);
6059 BTRFS_I(inode)->outstanding_extents++;
6060 spin_unlock(&BTRFS_I(inode)->lock);
6061 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6063 len = min_t(u64, len, root->sectorsize);
6066 lockend = start + len - 1;
6069 * If this errors out it's because we couldn't invalidate pagecache for
6070 * this range and we need to fallback to buffered.
6072 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6075 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6082 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6083 * io. INLINE is special, and we could probably kludge it in here, but
6084 * it's still buffered so for safety lets just fall back to the generic
6087 * For COMPRESSED we _have_ to read the entire extent in so we can
6088 * decompress it, so there will be buffering required no matter what we
6089 * do, so go ahead and fallback to buffered.
6091 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6092 * to buffered IO. Don't blame me, this is the price we pay for using
6095 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6096 em->block_start == EXTENT_MAP_INLINE) {
6097 free_extent_map(em);
6102 /* Just a good old fashioned hole, return */
6103 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6104 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6105 free_extent_map(em);
6110 * We don't allocate a new extent in the following cases
6112 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6114 * 2) The extent is marked as PREALLOC. We're good to go here and can
6115 * just use the extent.
6119 len = min(len, em->len - (start - em->start));
6120 lockstart = start + len;
6124 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6125 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6126 em->block_start != EXTENT_MAP_HOLE)) {
6131 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6132 type = BTRFS_ORDERED_PREALLOC;
6134 type = BTRFS_ORDERED_NOCOW;
6135 len = min(len, em->len - (start - em->start));
6136 block_start = em->block_start + (start - em->start);
6139 * we're not going to log anything, but we do need
6140 * to make sure the current transaction stays open
6141 * while we look for nocow cross refs
6143 trans = btrfs_join_transaction(root);
6147 if (can_nocow_odirect(trans, inode, start, len) == 1) {
6148 u64 orig_start = em->orig_start;
6149 u64 orig_block_len = em->orig_block_len;
6151 if (type == BTRFS_ORDERED_PREALLOC) {
6152 free_extent_map(em);
6153 em = create_pinned_em(inode, start, len,
6156 orig_block_len, type);
6158 btrfs_end_transaction(trans, root);
6163 ret = btrfs_add_ordered_extent_dio(inode, start,
6164 block_start, len, len, type);
6165 btrfs_end_transaction(trans, root);
6167 free_extent_map(em);
6172 btrfs_end_transaction(trans, root);
6176 * this will cow the extent, reset the len in case we changed
6179 len = bh_result->b_size;
6180 free_extent_map(em);
6181 em = btrfs_new_extent_direct(inode, start, len);
6186 len = min(len, em->len - (start - em->start));
6188 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6190 bh_result->b_size = len;
6191 bh_result->b_bdev = em->bdev;
6192 set_buffer_mapped(bh_result);
6194 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6195 set_buffer_new(bh_result);
6198 * Need to update the i_size under the extent lock so buffered
6199 * readers will get the updated i_size when we unlock.
6201 if (start + len > i_size_read(inode))
6202 i_size_write(inode, start + len);
6204 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6205 lockstart + len - 1, EXTENT_DELALLOC, NULL,
6206 &cached_state, GFP_NOFS);
6211 * In the case of write we need to clear and unlock the entire range,
6212 * in the case of read we need to unlock only the end area that we
6213 * aren't using if there is any left over space.
6215 if (lockstart < lockend) {
6216 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6217 lockend, unlock_bits, 1, 0,
6218 &cached_state, GFP_NOFS);
6220 free_extent_state(cached_state);
6223 free_extent_map(em);
6228 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6229 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6233 struct btrfs_dio_private {
6234 struct inode *inode;
6240 /* number of bios pending for this dio */
6241 atomic_t pending_bios;
6246 struct bio *orig_bio;
6249 static void btrfs_endio_direct_read(struct bio *bio, int err)
6251 struct btrfs_dio_private *dip = bio->bi_private;
6252 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6253 struct bio_vec *bvec = bio->bi_io_vec;
6254 struct inode *inode = dip->inode;
6255 struct btrfs_root *root = BTRFS_I(inode)->root;
6258 start = dip->logical_offset;
6260 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6261 struct page *page = bvec->bv_page;
6264 u64 private = ~(u32)0;
6265 unsigned long flags;
6267 if (get_state_private(&BTRFS_I(inode)->io_tree,
6270 local_irq_save(flags);
6271 kaddr = kmap_atomic(page);
6272 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
6273 csum, bvec->bv_len);
6274 btrfs_csum_final(csum, (char *)&csum);
6275 kunmap_atomic(kaddr);
6276 local_irq_restore(flags);
6278 flush_dcache_page(bvec->bv_page);
6279 if (csum != private) {
6281 printk(KERN_ERR "btrfs csum failed ino %llu off"
6282 " %llu csum %u private %u\n",
6283 (unsigned long long)btrfs_ino(inode),
6284 (unsigned long long)start,
6285 csum, (unsigned)private);
6290 start += bvec->bv_len;
6292 } while (bvec <= bvec_end);
6294 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6295 dip->logical_offset + dip->bytes - 1);
6296 bio->bi_private = dip->private;
6300 /* If we had a csum failure make sure to clear the uptodate flag */
6302 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6303 dio_end_io(bio, err);
6306 static void btrfs_endio_direct_write(struct bio *bio, int err)
6308 struct btrfs_dio_private *dip = bio->bi_private;
6309 struct inode *inode = dip->inode;
6310 struct btrfs_root *root = BTRFS_I(inode)->root;
6311 struct btrfs_ordered_extent *ordered = NULL;
6312 u64 ordered_offset = dip->logical_offset;
6313 u64 ordered_bytes = dip->bytes;
6319 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6321 ordered_bytes, !err);
6325 ordered->work.func = finish_ordered_fn;
6326 ordered->work.flags = 0;
6327 btrfs_queue_worker(&root->fs_info->endio_write_workers,
6331 * our bio might span multiple ordered extents. If we haven't
6332 * completed the accounting for the whole dio, go back and try again
6334 if (ordered_offset < dip->logical_offset + dip->bytes) {
6335 ordered_bytes = dip->logical_offset + dip->bytes -
6341 bio->bi_private = dip->private;
6345 /* If we had an error make sure to clear the uptodate flag */
6347 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6348 dio_end_io(bio, err);
6351 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6352 struct bio *bio, int mirror_num,
6353 unsigned long bio_flags, u64 offset)
6356 struct btrfs_root *root = BTRFS_I(inode)->root;
6357 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6358 BUG_ON(ret); /* -ENOMEM */
6362 static void btrfs_end_dio_bio(struct bio *bio, int err)
6364 struct btrfs_dio_private *dip = bio->bi_private;
6367 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6368 "sector %#Lx len %u err no %d\n",
6369 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6370 (unsigned long long)bio->bi_sector, bio->bi_size, err);
6374 * before atomic variable goto zero, we must make sure
6375 * dip->errors is perceived to be set.
6377 smp_mb__before_atomic_dec();
6380 /* if there are more bios still pending for this dio, just exit */
6381 if (!atomic_dec_and_test(&dip->pending_bios))
6385 bio_io_error(dip->orig_bio);
6387 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6388 bio_endio(dip->orig_bio, 0);
6394 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6395 u64 first_sector, gfp_t gfp_flags)
6397 int nr_vecs = bio_get_nr_vecs(bdev);
6398 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6401 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6402 int rw, u64 file_offset, int skip_sum,
6405 int write = rw & REQ_WRITE;
6406 struct btrfs_root *root = BTRFS_I(inode)->root;
6410 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
6415 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6423 if (write && async_submit) {
6424 ret = btrfs_wq_submit_bio(root->fs_info,
6425 inode, rw, bio, 0, 0,
6427 __btrfs_submit_bio_start_direct_io,
6428 __btrfs_submit_bio_done);
6432 * If we aren't doing async submit, calculate the csum of the
6435 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6438 } else if (!skip_sum) {
6439 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
6445 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6451 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6454 struct inode *inode = dip->inode;
6455 struct btrfs_root *root = BTRFS_I(inode)->root;
6457 struct bio *orig_bio = dip->orig_bio;
6458 struct bio_vec *bvec = orig_bio->bi_io_vec;
6459 u64 start_sector = orig_bio->bi_sector;
6460 u64 file_offset = dip->logical_offset;
6465 int async_submit = 0;
6467 map_length = orig_bio->bi_size;
6468 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
6469 &map_length, NULL, 0);
6474 if (map_length >= orig_bio->bi_size) {
6479 /* async crcs make it difficult to collect full stripe writes. */
6480 if (btrfs_get_alloc_profile(root, 1) &
6481 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
6486 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6489 bio->bi_private = dip;
6490 bio->bi_end_io = btrfs_end_dio_bio;
6491 atomic_inc(&dip->pending_bios);
6493 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6494 if (unlikely(map_length < submit_len + bvec->bv_len ||
6495 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6496 bvec->bv_offset) < bvec->bv_len)) {
6498 * inc the count before we submit the bio so
6499 * we know the end IO handler won't happen before
6500 * we inc the count. Otherwise, the dip might get freed
6501 * before we're done setting it up
6503 atomic_inc(&dip->pending_bios);
6504 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6505 file_offset, skip_sum,
6509 atomic_dec(&dip->pending_bios);
6513 start_sector += submit_len >> 9;
6514 file_offset += submit_len;
6519 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6520 start_sector, GFP_NOFS);
6523 bio->bi_private = dip;
6524 bio->bi_end_io = btrfs_end_dio_bio;
6526 map_length = orig_bio->bi_size;
6527 ret = btrfs_map_block(root->fs_info, rw,
6529 &map_length, NULL, 0);
6535 submit_len += bvec->bv_len;
6542 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6551 * before atomic variable goto zero, we must
6552 * make sure dip->errors is perceived to be set.
6554 smp_mb__before_atomic_dec();
6555 if (atomic_dec_and_test(&dip->pending_bios))
6556 bio_io_error(dip->orig_bio);
6558 /* bio_end_io() will handle error, so we needn't return it */
6562 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6565 struct btrfs_root *root = BTRFS_I(inode)->root;
6566 struct btrfs_dio_private *dip;
6567 struct bio_vec *bvec = bio->bi_io_vec;
6569 int write = rw & REQ_WRITE;
6572 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6574 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6580 dip->private = bio->bi_private;
6582 dip->logical_offset = file_offset;
6586 dip->bytes += bvec->bv_len;
6588 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6590 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6591 bio->bi_private = dip;
6593 dip->orig_bio = bio;
6594 atomic_set(&dip->pending_bios, 0);
6597 bio->bi_end_io = btrfs_endio_direct_write;
6599 bio->bi_end_io = btrfs_endio_direct_read;
6601 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6606 * If this is a write, we need to clean up the reserved space and kill
6607 * the ordered extent.
6610 struct btrfs_ordered_extent *ordered;
6611 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6612 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6613 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6614 btrfs_free_reserved_extent(root, ordered->start,
6616 btrfs_put_ordered_extent(ordered);
6617 btrfs_put_ordered_extent(ordered);
6619 bio_endio(bio, ret);
6622 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6623 const struct iovec *iov, loff_t offset,
6624 unsigned long nr_segs)
6630 unsigned blocksize_mask = root->sectorsize - 1;
6631 ssize_t retval = -EINVAL;
6632 loff_t end = offset;
6634 if (offset & blocksize_mask)
6637 /* Check the memory alignment. Blocks cannot straddle pages */
6638 for (seg = 0; seg < nr_segs; seg++) {
6639 addr = (unsigned long)iov[seg].iov_base;
6640 size = iov[seg].iov_len;
6642 if ((addr & blocksize_mask) || (size & blocksize_mask))
6645 /* If this is a write we don't need to check anymore */
6650 * Check to make sure we don't have duplicate iov_base's in this
6651 * iovec, if so return EINVAL, otherwise we'll get csum errors
6652 * when reading back.
6654 for (i = seg + 1; i < nr_segs; i++) {
6655 if (iov[seg].iov_base == iov[i].iov_base)
6664 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6665 const struct iovec *iov, loff_t offset,
6666 unsigned long nr_segs)
6668 struct file *file = iocb->ki_filp;
6669 struct inode *inode = file->f_mapping->host;
6673 bool relock = false;
6676 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6680 atomic_inc(&inode->i_dio_count);
6681 smp_mb__after_atomic_inc();
6684 count = iov_length(iov, nr_segs);
6686 * If the write DIO is beyond the EOF, we need update
6687 * the isize, but it is protected by i_mutex. So we can
6688 * not unlock the i_mutex at this case.
6690 if (offset + count <= inode->i_size) {
6691 mutex_unlock(&inode->i_mutex);
6694 ret = btrfs_delalloc_reserve_space(inode, count);
6697 } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
6698 &BTRFS_I(inode)->runtime_flags))) {
6699 inode_dio_done(inode);
6700 flags = DIO_LOCKING | DIO_SKIP_HOLES;
6704 ret = __blockdev_direct_IO(rw, iocb, inode,
6705 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6706 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6707 btrfs_submit_direct, flags);
6709 if (ret < 0 && ret != -EIOCBQUEUED)
6710 btrfs_delalloc_release_space(inode, count);
6711 else if (ret > 0 && (size_t)ret < count) {
6712 spin_lock(&BTRFS_I(inode)->lock);
6713 BTRFS_I(inode)->outstanding_extents++;
6714 spin_unlock(&BTRFS_I(inode)->lock);
6715 btrfs_delalloc_release_space(inode,
6716 count - (size_t)ret);
6718 btrfs_delalloc_release_metadata(inode, 0);
6722 inode_dio_done(inode);
6724 mutex_lock(&inode->i_mutex);
6729 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
6731 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6732 __u64 start, __u64 len)
6736 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
6740 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6743 int btrfs_readpage(struct file *file, struct page *page)
6745 struct extent_io_tree *tree;
6746 tree = &BTRFS_I(page->mapping->host)->io_tree;
6747 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6750 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6752 struct extent_io_tree *tree;
6755 if (current->flags & PF_MEMALLOC) {
6756 redirty_page_for_writepage(wbc, page);
6760 tree = &BTRFS_I(page->mapping->host)->io_tree;
6761 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6764 int btrfs_writepages(struct address_space *mapping,
6765 struct writeback_control *wbc)
6767 struct extent_io_tree *tree;
6769 tree = &BTRFS_I(mapping->host)->io_tree;
6770 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6774 btrfs_readpages(struct file *file, struct address_space *mapping,
6775 struct list_head *pages, unsigned nr_pages)
6777 struct extent_io_tree *tree;
6778 tree = &BTRFS_I(mapping->host)->io_tree;
6779 return extent_readpages(tree, mapping, pages, nr_pages,
6782 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6784 struct extent_io_tree *tree;
6785 struct extent_map_tree *map;
6788 tree = &BTRFS_I(page->mapping->host)->io_tree;
6789 map = &BTRFS_I(page->mapping->host)->extent_tree;
6790 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6792 ClearPagePrivate(page);
6793 set_page_private(page, 0);
6794 page_cache_release(page);
6799 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6801 if (PageWriteback(page) || PageDirty(page))
6803 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6806 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6808 struct inode *inode = page->mapping->host;
6809 struct extent_io_tree *tree;
6810 struct btrfs_ordered_extent *ordered;
6811 struct extent_state *cached_state = NULL;
6812 u64 page_start = page_offset(page);
6813 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6816 * we have the page locked, so new writeback can't start,
6817 * and the dirty bit won't be cleared while we are here.
6819 * Wait for IO on this page so that we can safely clear
6820 * the PagePrivate2 bit and do ordered accounting
6822 wait_on_page_writeback(page);
6824 tree = &BTRFS_I(inode)->io_tree;
6826 btrfs_releasepage(page, GFP_NOFS);
6829 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6830 ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
6833 * IO on this page will never be started, so we need
6834 * to account for any ordered extents now
6836 clear_extent_bit(tree, page_start, page_end,
6837 EXTENT_DIRTY | EXTENT_DELALLOC |
6838 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
6839 EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
6841 * whoever cleared the private bit is responsible
6842 * for the finish_ordered_io
6844 if (TestClearPagePrivate2(page) &&
6845 btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6846 PAGE_CACHE_SIZE, 1)) {
6847 btrfs_finish_ordered_io(ordered);
6849 btrfs_put_ordered_extent(ordered);
6850 cached_state = NULL;
6851 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6853 clear_extent_bit(tree, page_start, page_end,
6854 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6855 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
6856 &cached_state, GFP_NOFS);
6857 __btrfs_releasepage(page, GFP_NOFS);
6859 ClearPageChecked(page);
6860 if (PagePrivate(page)) {
6861 ClearPagePrivate(page);
6862 set_page_private(page, 0);
6863 page_cache_release(page);
6868 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6869 * called from a page fault handler when a page is first dirtied. Hence we must
6870 * be careful to check for EOF conditions here. We set the page up correctly
6871 * for a written page which means we get ENOSPC checking when writing into
6872 * holes and correct delalloc and unwritten extent mapping on filesystems that
6873 * support these features.
6875 * We are not allowed to take the i_mutex here so we have to play games to
6876 * protect against truncate races as the page could now be beyond EOF. Because
6877 * vmtruncate() writes the inode size before removing pages, once we have the
6878 * page lock we can determine safely if the page is beyond EOF. If it is not
6879 * beyond EOF, then the page is guaranteed safe against truncation until we
6882 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6884 struct page *page = vmf->page;
6885 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6886 struct btrfs_root *root = BTRFS_I(inode)->root;
6887 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6888 struct btrfs_ordered_extent *ordered;
6889 struct extent_state *cached_state = NULL;
6891 unsigned long zero_start;
6898 sb_start_pagefault(inode->i_sb);
6899 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6901 ret = file_update_time(vma->vm_file);
6907 else /* -ENOSPC, -EIO, etc */
6908 ret = VM_FAULT_SIGBUS;
6914 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6917 size = i_size_read(inode);
6918 page_start = page_offset(page);
6919 page_end = page_start + PAGE_CACHE_SIZE - 1;
6921 if ((page->mapping != inode->i_mapping) ||
6922 (page_start >= size)) {
6923 /* page got truncated out from underneath us */
6926 wait_on_page_writeback(page);
6928 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6929 set_page_extent_mapped(page);
6932 * we can't set the delalloc bits if there are pending ordered
6933 * extents. Drop our locks and wait for them to finish
6935 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6937 unlock_extent_cached(io_tree, page_start, page_end,
6938 &cached_state, GFP_NOFS);
6940 btrfs_start_ordered_extent(inode, ordered, 1);
6941 btrfs_put_ordered_extent(ordered);
6946 * XXX - page_mkwrite gets called every time the page is dirtied, even
6947 * if it was already dirty, so for space accounting reasons we need to
6948 * clear any delalloc bits for the range we are fixing to save. There
6949 * is probably a better way to do this, but for now keep consistent with
6950 * prepare_pages in the normal write path.
6952 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6953 EXTENT_DIRTY | EXTENT_DELALLOC |
6954 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
6955 0, 0, &cached_state, GFP_NOFS);
6957 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6960 unlock_extent_cached(io_tree, page_start, page_end,
6961 &cached_state, GFP_NOFS);
6962 ret = VM_FAULT_SIGBUS;
6967 /* page is wholly or partially inside EOF */
6968 if (page_start + PAGE_CACHE_SIZE > size)
6969 zero_start = size & ~PAGE_CACHE_MASK;
6971 zero_start = PAGE_CACHE_SIZE;
6973 if (zero_start != PAGE_CACHE_SIZE) {
6975 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6976 flush_dcache_page(page);
6979 ClearPageChecked(page);
6980 set_page_dirty(page);
6981 SetPageUptodate(page);
6983 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6984 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6985 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
6987 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6991 sb_end_pagefault(inode->i_sb);
6992 return VM_FAULT_LOCKED;
6996 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6998 sb_end_pagefault(inode->i_sb);
7002 static int btrfs_truncate(struct inode *inode)
7004 struct btrfs_root *root = BTRFS_I(inode)->root;
7005 struct btrfs_block_rsv *rsv;
7008 struct btrfs_trans_handle *trans;
7009 u64 mask = root->sectorsize - 1;
7010 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7012 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
7016 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
7017 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
7020 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
7021 * 3 things going on here
7023 * 1) We need to reserve space for our orphan item and the space to
7024 * delete our orphan item. Lord knows we don't want to have a dangling
7025 * orphan item because we didn't reserve space to remove it.
7027 * 2) We need to reserve space to update our inode.
7029 * 3) We need to have something to cache all the space that is going to
7030 * be free'd up by the truncate operation, but also have some slack
7031 * space reserved in case it uses space during the truncate (thank you
7032 * very much snapshotting).
7034 * And we need these to all be seperate. The fact is we can use alot of
7035 * space doing the truncate, and we have no earthly idea how much space
7036 * we will use, so we need the truncate reservation to be seperate so it
7037 * doesn't end up using space reserved for updating the inode or
7038 * removing the orphan item. We also need to be able to stop the
7039 * transaction and start a new one, which means we need to be able to
7040 * update the inode several times, and we have no idea of knowing how
7041 * many times that will be, so we can't just reserve 1 item for the
7042 * entirety of the opration, so that has to be done seperately as well.
7043 * Then there is the orphan item, which does indeed need to be held on
7044 * to for the whole operation, and we need nobody to touch this reserved
7045 * space except the orphan code.
7047 * So that leaves us with
7049 * 1) root->orphan_block_rsv - for the orphan deletion.
7050 * 2) rsv - for the truncate reservation, which we will steal from the
7051 * transaction reservation.
7052 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7053 * updating the inode.
7055 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7058 rsv->size = min_size;
7062 * 1 for the truncate slack space
7063 * 1 for updating the inode.
7065 trans = btrfs_start_transaction(root, 2);
7066 if (IS_ERR(trans)) {
7067 err = PTR_ERR(trans);
7071 /* Migrate the slack space for the truncate to our reserve */
7072 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7077 * setattr is responsible for setting the ordered_data_close flag,
7078 * but that is only tested during the last file release. That
7079 * could happen well after the next commit, leaving a great big
7080 * window where new writes may get lost if someone chooses to write
7081 * to this file after truncating to zero
7083 * The inode doesn't have any dirty data here, and so if we commit
7084 * this is a noop. If someone immediately starts writing to the inode
7085 * it is very likely we'll catch some of their writes in this
7086 * transaction, and the commit will find this file on the ordered
7087 * data list with good things to send down.
7089 * This is a best effort solution, there is still a window where
7090 * using truncate to replace the contents of the file will
7091 * end up with a zero length file after a crash.
7093 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7094 &BTRFS_I(inode)->runtime_flags))
7095 btrfs_add_ordered_operation(trans, root, inode);
7098 * So if we truncate and then write and fsync we normally would just
7099 * write the extents that changed, which is a problem if we need to
7100 * first truncate that entire inode. So set this flag so we write out
7101 * all of the extents in the inode to the sync log so we're completely
7104 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7105 trans->block_rsv = rsv;
7108 ret = btrfs_truncate_inode_items(trans, root, inode,
7110 BTRFS_EXTENT_DATA_KEY);
7111 if (ret != -ENOSPC) {
7116 trans->block_rsv = &root->fs_info->trans_block_rsv;
7117 ret = btrfs_update_inode(trans, root, inode);
7123 btrfs_end_transaction(trans, root);
7124 btrfs_btree_balance_dirty(root);
7126 trans = btrfs_start_transaction(root, 2);
7127 if (IS_ERR(trans)) {
7128 ret = err = PTR_ERR(trans);
7133 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7135 BUG_ON(ret); /* shouldn't happen */
7136 trans->block_rsv = rsv;
7139 if (ret == 0 && inode->i_nlink > 0) {
7140 trans->block_rsv = root->orphan_block_rsv;
7141 ret = btrfs_orphan_del(trans, inode);
7147 trans->block_rsv = &root->fs_info->trans_block_rsv;
7148 ret = btrfs_update_inode(trans, root, inode);
7152 ret = btrfs_end_transaction(trans, root);
7153 btrfs_btree_balance_dirty(root);
7157 btrfs_free_block_rsv(root, rsv);
7166 * create a new subvolume directory/inode (helper for the ioctl).
7168 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7169 struct btrfs_root *new_root, u64 new_dirid)
7171 struct inode *inode;
7175 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7176 new_dirid, new_dirid,
7177 S_IFDIR | (~current_umask() & S_IRWXUGO),
7180 return PTR_ERR(inode);
7181 inode->i_op = &btrfs_dir_inode_operations;
7182 inode->i_fop = &btrfs_dir_file_operations;
7184 set_nlink(inode, 1);
7185 btrfs_i_size_write(inode, 0);
7187 err = btrfs_update_inode(trans, new_root, inode);
7193 struct inode *btrfs_alloc_inode(struct super_block *sb)
7195 struct btrfs_inode *ei;
7196 struct inode *inode;
7198 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7205 ei->last_sub_trans = 0;
7206 ei->logged_trans = 0;
7207 ei->delalloc_bytes = 0;
7208 ei->disk_i_size = 0;
7211 ei->index_cnt = (u64)-1;
7212 ei->last_unlink_trans = 0;
7213 ei->last_log_commit = 0;
7215 spin_lock_init(&ei->lock);
7216 ei->outstanding_extents = 0;
7217 ei->reserved_extents = 0;
7219 ei->runtime_flags = 0;
7220 ei->force_compress = BTRFS_COMPRESS_NONE;
7222 ei->delayed_node = NULL;
7224 inode = &ei->vfs_inode;
7225 extent_map_tree_init(&ei->extent_tree);
7226 extent_io_tree_init(&ei->io_tree, &inode->i_data);
7227 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7228 ei->io_tree.track_uptodate = 1;
7229 ei->io_failure_tree.track_uptodate = 1;
7230 atomic_set(&ei->sync_writers, 0);
7231 mutex_init(&ei->log_mutex);
7232 mutex_init(&ei->delalloc_mutex);
7233 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7234 INIT_LIST_HEAD(&ei->delalloc_inodes);
7235 INIT_LIST_HEAD(&ei->ordered_operations);
7236 RB_CLEAR_NODE(&ei->rb_node);
7241 static void btrfs_i_callback(struct rcu_head *head)
7243 struct inode *inode = container_of(head, struct inode, i_rcu);
7244 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7247 void btrfs_destroy_inode(struct inode *inode)
7249 struct btrfs_ordered_extent *ordered;
7250 struct btrfs_root *root = BTRFS_I(inode)->root;
7252 WARN_ON(!hlist_empty(&inode->i_dentry));
7253 WARN_ON(inode->i_data.nrpages);
7254 WARN_ON(BTRFS_I(inode)->outstanding_extents);
7255 WARN_ON(BTRFS_I(inode)->reserved_extents);
7256 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7257 WARN_ON(BTRFS_I(inode)->csum_bytes);
7260 * This can happen where we create an inode, but somebody else also
7261 * created the same inode and we need to destroy the one we already
7268 * Make sure we're properly removed from the ordered operation
7272 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7273 spin_lock(&root->fs_info->ordered_extent_lock);
7274 list_del_init(&BTRFS_I(inode)->ordered_operations);
7275 spin_unlock(&root->fs_info->ordered_extent_lock);
7278 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7279 &BTRFS_I(inode)->runtime_flags)) {
7280 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7281 (unsigned long long)btrfs_ino(inode));
7282 atomic_dec(&root->orphan_inodes);
7286 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7290 printk(KERN_ERR "btrfs found ordered "
7291 "extent %llu %llu on inode cleanup\n",
7292 (unsigned long long)ordered->file_offset,
7293 (unsigned long long)ordered->len);
7294 btrfs_remove_ordered_extent(inode, ordered);
7295 btrfs_put_ordered_extent(ordered);
7296 btrfs_put_ordered_extent(ordered);
7299 inode_tree_del(inode);
7300 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7302 btrfs_remove_delayed_node(inode);
7303 call_rcu(&inode->i_rcu, btrfs_i_callback);
7306 int btrfs_drop_inode(struct inode *inode)
7308 struct btrfs_root *root = BTRFS_I(inode)->root;
7310 /* the snap/subvol tree is on deleting */
7311 if (btrfs_root_refs(&root->root_item) == 0 &&
7312 root != root->fs_info->tree_root)
7315 return generic_drop_inode(inode);
7318 static void init_once(void *foo)
7320 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7322 inode_init_once(&ei->vfs_inode);
7325 void btrfs_destroy_cachep(void)
7328 * Make sure all delayed rcu free inodes are flushed before we
7332 if (btrfs_inode_cachep)
7333 kmem_cache_destroy(btrfs_inode_cachep);
7334 if (btrfs_trans_handle_cachep)
7335 kmem_cache_destroy(btrfs_trans_handle_cachep);
7336 if (btrfs_transaction_cachep)
7337 kmem_cache_destroy(btrfs_transaction_cachep);
7338 if (btrfs_path_cachep)
7339 kmem_cache_destroy(btrfs_path_cachep);
7340 if (btrfs_free_space_cachep)
7341 kmem_cache_destroy(btrfs_free_space_cachep);
7342 if (btrfs_delalloc_work_cachep)
7343 kmem_cache_destroy(btrfs_delalloc_work_cachep);
7346 int btrfs_init_cachep(void)
7348 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
7349 sizeof(struct btrfs_inode), 0,
7350 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7351 if (!btrfs_inode_cachep)
7354 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
7355 sizeof(struct btrfs_trans_handle), 0,
7356 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7357 if (!btrfs_trans_handle_cachep)
7360 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
7361 sizeof(struct btrfs_transaction), 0,
7362 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7363 if (!btrfs_transaction_cachep)
7366 btrfs_path_cachep = kmem_cache_create("btrfs_path",
7367 sizeof(struct btrfs_path), 0,
7368 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7369 if (!btrfs_path_cachep)
7372 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
7373 sizeof(struct btrfs_free_space), 0,
7374 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7375 if (!btrfs_free_space_cachep)
7378 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
7379 sizeof(struct btrfs_delalloc_work), 0,
7380 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
7382 if (!btrfs_delalloc_work_cachep)
7387 btrfs_destroy_cachep();
7391 static int btrfs_getattr(struct vfsmount *mnt,
7392 struct dentry *dentry, struct kstat *stat)
7395 struct inode *inode = dentry->d_inode;
7396 u32 blocksize = inode->i_sb->s_blocksize;
7398 generic_fillattr(inode, stat);
7399 stat->dev = BTRFS_I(inode)->root->anon_dev;
7400 stat->blksize = PAGE_CACHE_SIZE;
7402 spin_lock(&BTRFS_I(inode)->lock);
7403 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
7404 spin_unlock(&BTRFS_I(inode)->lock);
7405 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7406 ALIGN(delalloc_bytes, blocksize)) >> 9;
7411 * If a file is moved, it will inherit the cow and compression flags of the new
7414 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7416 struct btrfs_inode *b_dir = BTRFS_I(dir);
7417 struct btrfs_inode *b_inode = BTRFS_I(inode);
7419 if (b_dir->flags & BTRFS_INODE_NODATACOW)
7420 b_inode->flags |= BTRFS_INODE_NODATACOW;
7422 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7424 if (b_dir->flags & BTRFS_INODE_COMPRESS) {
7425 b_inode->flags |= BTRFS_INODE_COMPRESS;
7426 b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
7428 b_inode->flags &= ~(BTRFS_INODE_COMPRESS |
7429 BTRFS_INODE_NOCOMPRESS);
7433 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7434 struct inode *new_dir, struct dentry *new_dentry)
7436 struct btrfs_trans_handle *trans;
7437 struct btrfs_root *root = BTRFS_I(old_dir)->root;
7438 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7439 struct inode *new_inode = new_dentry->d_inode;
7440 struct inode *old_inode = old_dentry->d_inode;
7441 struct timespec ctime = CURRENT_TIME;
7445 u64 old_ino = btrfs_ino(old_inode);
7447 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7450 /* we only allow rename subvolume link between subvolumes */
7451 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7454 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7455 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7458 if (S_ISDIR(old_inode->i_mode) && new_inode &&
7459 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7463 /* check for collisions, even if the name isn't there */
7464 ret = btrfs_check_dir_item_collision(root, new_dir->i_ino,
7465 new_dentry->d_name.name,
7466 new_dentry->d_name.len);
7469 if (ret == -EEXIST) {
7471 * eexist without a new_inode */
7477 /* maybe -EOVERFLOW */
7484 * we're using rename to replace one file with another.
7485 * and the replacement file is large. Start IO on it now so
7486 * we don't add too much work to the end of the transaction
7488 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7489 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7490 filemap_flush(old_inode->i_mapping);
7492 /* close the racy window with snapshot create/destroy ioctl */
7493 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7494 down_read(&root->fs_info->subvol_sem);
7496 * We want to reserve the absolute worst case amount of items. So if
7497 * both inodes are subvols and we need to unlink them then that would
7498 * require 4 item modifications, but if they are both normal inodes it
7499 * would require 5 item modifications, so we'll assume their normal
7500 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7501 * should cover the worst case number of items we'll modify.
7503 trans = btrfs_start_transaction(root, 20);
7504 if (IS_ERR(trans)) {
7505 ret = PTR_ERR(trans);
7510 btrfs_record_root_in_trans(trans, dest);
7512 ret = btrfs_set_inode_index(new_dir, &index);
7516 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7517 /* force full log commit if subvolume involved. */
7518 root->fs_info->last_trans_log_full_commit = trans->transid;
7520 ret = btrfs_insert_inode_ref(trans, dest,
7521 new_dentry->d_name.name,
7522 new_dentry->d_name.len,
7524 btrfs_ino(new_dir), index);
7528 * this is an ugly little race, but the rename is required
7529 * to make sure that if we crash, the inode is either at the
7530 * old name or the new one. pinning the log transaction lets
7531 * us make sure we don't allow a log commit to come in after
7532 * we unlink the name but before we add the new name back in.
7534 btrfs_pin_log_trans(root);
7537 * make sure the inode gets flushed if it is replacing
7540 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7541 btrfs_add_ordered_operation(trans, root, old_inode);
7543 inode_inc_iversion(old_dir);
7544 inode_inc_iversion(new_dir);
7545 inode_inc_iversion(old_inode);
7546 old_dir->i_ctime = old_dir->i_mtime = ctime;
7547 new_dir->i_ctime = new_dir->i_mtime = ctime;
7548 old_inode->i_ctime = ctime;
7550 if (old_dentry->d_parent != new_dentry->d_parent)
7551 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7553 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7554 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7555 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7556 old_dentry->d_name.name,
7557 old_dentry->d_name.len);
7559 ret = __btrfs_unlink_inode(trans, root, old_dir,
7560 old_dentry->d_inode,
7561 old_dentry->d_name.name,
7562 old_dentry->d_name.len);
7564 ret = btrfs_update_inode(trans, root, old_inode);
7567 btrfs_abort_transaction(trans, root, ret);
7572 inode_inc_iversion(new_inode);
7573 new_inode->i_ctime = CURRENT_TIME;
7574 if (unlikely(btrfs_ino(new_inode) ==
7575 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7576 root_objectid = BTRFS_I(new_inode)->location.objectid;
7577 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7579 new_dentry->d_name.name,
7580 new_dentry->d_name.len);
7581 BUG_ON(new_inode->i_nlink == 0);
7583 ret = btrfs_unlink_inode(trans, dest, new_dir,
7584 new_dentry->d_inode,
7585 new_dentry->d_name.name,
7586 new_dentry->d_name.len);
7588 if (!ret && new_inode->i_nlink == 0) {
7589 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7593 btrfs_abort_transaction(trans, root, ret);
7598 fixup_inode_flags(new_dir, old_inode);
7600 ret = btrfs_add_link(trans, new_dir, old_inode,
7601 new_dentry->d_name.name,
7602 new_dentry->d_name.len, 0, index);
7604 btrfs_abort_transaction(trans, root, ret);
7608 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7609 struct dentry *parent = new_dentry->d_parent;
7610 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7611 btrfs_end_log_trans(root);
7614 btrfs_end_transaction(trans, root);
7616 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7617 up_read(&root->fs_info->subvol_sem);
7622 static void btrfs_run_delalloc_work(struct btrfs_work *work)
7624 struct btrfs_delalloc_work *delalloc_work;
7626 delalloc_work = container_of(work, struct btrfs_delalloc_work,
7628 if (delalloc_work->wait)
7629 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
7631 filemap_flush(delalloc_work->inode->i_mapping);
7633 if (delalloc_work->delay_iput)
7634 btrfs_add_delayed_iput(delalloc_work->inode);
7636 iput(delalloc_work->inode);
7637 complete(&delalloc_work->completion);
7640 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
7641 int wait, int delay_iput)
7643 struct btrfs_delalloc_work *work;
7645 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
7649 init_completion(&work->completion);
7650 INIT_LIST_HEAD(&work->list);
7651 work->inode = inode;
7653 work->delay_iput = delay_iput;
7654 work->work.func = btrfs_run_delalloc_work;
7659 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
7661 wait_for_completion(&work->completion);
7662 kmem_cache_free(btrfs_delalloc_work_cachep, work);
7666 * some fairly slow code that needs optimization. This walks the list
7667 * of all the inodes with pending delalloc and forces them to disk.
7669 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7671 struct btrfs_inode *binode;
7672 struct inode *inode;
7673 struct btrfs_delalloc_work *work, *next;
7674 struct list_head works;
7675 struct list_head splice;
7678 if (root->fs_info->sb->s_flags & MS_RDONLY)
7681 INIT_LIST_HEAD(&works);
7682 INIT_LIST_HEAD(&splice);
7684 spin_lock(&root->fs_info->delalloc_lock);
7685 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
7686 while (!list_empty(&splice)) {
7687 binode = list_entry(splice.next, struct btrfs_inode,
7690 list_del_init(&binode->delalloc_inodes);
7692 inode = igrab(&binode->vfs_inode);
7694 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
7695 &binode->runtime_flags);
7699 list_add_tail(&binode->delalloc_inodes,
7700 &root->fs_info->delalloc_inodes);
7701 spin_unlock(&root->fs_info->delalloc_lock);
7703 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
7704 if (unlikely(!work)) {
7708 list_add_tail(&work->list, &works);
7709 btrfs_queue_worker(&root->fs_info->flush_workers,
7713 spin_lock(&root->fs_info->delalloc_lock);
7715 spin_unlock(&root->fs_info->delalloc_lock);
7717 list_for_each_entry_safe(work, next, &works, list) {
7718 list_del_init(&work->list);
7719 btrfs_wait_and_free_delalloc_work(work);
7722 /* the filemap_flush will queue IO into the worker threads, but
7723 * we have to make sure the IO is actually started and that
7724 * ordered extents get created before we return
7726 atomic_inc(&root->fs_info->async_submit_draining);
7727 while (atomic_read(&root->fs_info->nr_async_submits) ||
7728 atomic_read(&root->fs_info->async_delalloc_pages)) {
7729 wait_event(root->fs_info->async_submit_wait,
7730 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7731 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7733 atomic_dec(&root->fs_info->async_submit_draining);
7736 list_for_each_entry_safe(work, next, &works, list) {
7737 list_del_init(&work->list);
7738 btrfs_wait_and_free_delalloc_work(work);
7741 if (!list_empty_careful(&splice)) {
7742 spin_lock(&root->fs_info->delalloc_lock);
7743 list_splice_tail(&splice, &root->fs_info->delalloc_inodes);
7744 spin_unlock(&root->fs_info->delalloc_lock);
7749 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7750 const char *symname)
7752 struct btrfs_trans_handle *trans;
7753 struct btrfs_root *root = BTRFS_I(dir)->root;
7754 struct btrfs_path *path;
7755 struct btrfs_key key;
7756 struct inode *inode = NULL;
7764 struct btrfs_file_extent_item *ei;
7765 struct extent_buffer *leaf;
7767 name_len = strlen(symname) + 1;
7768 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7769 return -ENAMETOOLONG;
7772 * 2 items for inode item and ref
7773 * 2 items for dir items
7774 * 1 item for xattr if selinux is on
7776 trans = btrfs_start_transaction(root, 5);
7778 return PTR_ERR(trans);
7780 err = btrfs_find_free_ino(root, &objectid);
7784 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7785 dentry->d_name.len, btrfs_ino(dir), objectid,
7786 S_IFLNK|S_IRWXUGO, &index);
7787 if (IS_ERR(inode)) {
7788 err = PTR_ERR(inode);
7792 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7799 * If the active LSM wants to access the inode during
7800 * d_instantiate it needs these. Smack checks to see
7801 * if the filesystem supports xattrs by looking at the
7804 inode->i_fop = &btrfs_file_operations;
7805 inode->i_op = &btrfs_file_inode_operations;
7807 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7811 inode->i_mapping->a_ops = &btrfs_aops;
7812 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7813 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7818 path = btrfs_alloc_path();
7824 key.objectid = btrfs_ino(inode);
7826 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7827 datasize = btrfs_file_extent_calc_inline_size(name_len);
7828 err = btrfs_insert_empty_item(trans, root, path, &key,
7832 btrfs_free_path(path);
7835 leaf = path->nodes[0];
7836 ei = btrfs_item_ptr(leaf, path->slots[0],
7837 struct btrfs_file_extent_item);
7838 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7839 btrfs_set_file_extent_type(leaf, ei,
7840 BTRFS_FILE_EXTENT_INLINE);
7841 btrfs_set_file_extent_encryption(leaf, ei, 0);
7842 btrfs_set_file_extent_compression(leaf, ei, 0);
7843 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7844 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7846 ptr = btrfs_file_extent_inline_start(ei);
7847 write_extent_buffer(leaf, symname, ptr, name_len);
7848 btrfs_mark_buffer_dirty(leaf);
7849 btrfs_free_path(path);
7851 inode->i_op = &btrfs_symlink_inode_operations;
7852 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7853 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7854 inode_set_bytes(inode, name_len);
7855 btrfs_i_size_write(inode, name_len - 1);
7856 err = btrfs_update_inode(trans, root, inode);
7862 d_instantiate(dentry, inode);
7863 btrfs_end_transaction(trans, root);
7865 inode_dec_link_count(inode);
7868 btrfs_btree_balance_dirty(root);
7872 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7873 u64 start, u64 num_bytes, u64 min_size,
7874 loff_t actual_len, u64 *alloc_hint,
7875 struct btrfs_trans_handle *trans)
7877 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7878 struct extent_map *em;
7879 struct btrfs_root *root = BTRFS_I(inode)->root;
7880 struct btrfs_key ins;
7881 u64 cur_offset = start;
7884 bool own_trans = true;
7888 while (num_bytes > 0) {
7890 trans = btrfs_start_transaction(root, 3);
7891 if (IS_ERR(trans)) {
7892 ret = PTR_ERR(trans);
7897 ret = btrfs_reserve_extent(trans, root,
7898 min(num_bytes, 256ULL * 1024 * 1024),
7899 min_size, 0, *alloc_hint, &ins, 1);
7902 btrfs_end_transaction(trans, root);
7906 ret = insert_reserved_file_extent(trans, inode,
7907 cur_offset, ins.objectid,
7908 ins.offset, ins.offset,
7909 ins.offset, 0, 0, 0,
7910 BTRFS_FILE_EXTENT_PREALLOC);
7912 btrfs_abort_transaction(trans, root, ret);
7914 btrfs_end_transaction(trans, root);
7917 btrfs_drop_extent_cache(inode, cur_offset,
7918 cur_offset + ins.offset -1, 0);
7920 em = alloc_extent_map();
7922 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
7923 &BTRFS_I(inode)->runtime_flags);
7927 em->start = cur_offset;
7928 em->orig_start = cur_offset;
7929 em->len = ins.offset;
7930 em->block_start = ins.objectid;
7931 em->block_len = ins.offset;
7932 em->orig_block_len = ins.offset;
7933 em->bdev = root->fs_info->fs_devices->latest_bdev;
7934 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7935 em->generation = trans->transid;
7938 write_lock(&em_tree->lock);
7939 ret = add_extent_mapping(em_tree, em);
7941 list_move(&em->list,
7942 &em_tree->modified_extents);
7943 write_unlock(&em_tree->lock);
7946 btrfs_drop_extent_cache(inode, cur_offset,
7947 cur_offset + ins.offset - 1,
7950 free_extent_map(em);
7952 num_bytes -= ins.offset;
7953 cur_offset += ins.offset;
7954 *alloc_hint = ins.objectid + ins.offset;
7956 inode_inc_iversion(inode);
7957 inode->i_ctime = CURRENT_TIME;
7958 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7959 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7960 (actual_len > inode->i_size) &&
7961 (cur_offset > inode->i_size)) {
7962 if (cur_offset > actual_len)
7963 i_size = actual_len;
7965 i_size = cur_offset;
7966 i_size_write(inode, i_size);
7967 btrfs_ordered_update_i_size(inode, i_size, NULL);
7970 ret = btrfs_update_inode(trans, root, inode);
7973 btrfs_abort_transaction(trans, root, ret);
7975 btrfs_end_transaction(trans, root);
7980 btrfs_end_transaction(trans, root);
7985 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7986 u64 start, u64 num_bytes, u64 min_size,
7987 loff_t actual_len, u64 *alloc_hint)
7989 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7990 min_size, actual_len, alloc_hint,
7994 int btrfs_prealloc_file_range_trans(struct inode *inode,
7995 struct btrfs_trans_handle *trans, int mode,
7996 u64 start, u64 num_bytes, u64 min_size,
7997 loff_t actual_len, u64 *alloc_hint)
7999 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8000 min_size, actual_len, alloc_hint, trans);
8003 static int btrfs_set_page_dirty(struct page *page)
8005 return __set_page_dirty_nobuffers(page);
8008 static int btrfs_permission(struct inode *inode, int mask)
8010 struct btrfs_root *root = BTRFS_I(inode)->root;
8011 umode_t mode = inode->i_mode;
8013 if (mask & MAY_WRITE &&
8014 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8015 if (btrfs_root_readonly(root))
8017 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8020 return generic_permission(inode, mask);
8023 static const struct inode_operations btrfs_dir_inode_operations = {
8024 .getattr = btrfs_getattr,
8025 .lookup = btrfs_lookup,
8026 .create = btrfs_create,
8027 .unlink = btrfs_unlink,
8029 .mkdir = btrfs_mkdir,
8030 .rmdir = btrfs_rmdir,
8031 .rename = btrfs_rename,
8032 .symlink = btrfs_symlink,
8033 .setattr = btrfs_setattr,
8034 .mknod = btrfs_mknod,
8035 .setxattr = btrfs_setxattr,
8036 .getxattr = btrfs_getxattr,
8037 .listxattr = btrfs_listxattr,
8038 .removexattr = btrfs_removexattr,
8039 .permission = btrfs_permission,
8040 .get_acl = btrfs_get_acl,
8042 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8043 .lookup = btrfs_lookup,
8044 .permission = btrfs_permission,
8045 .get_acl = btrfs_get_acl,
8048 static const struct file_operations btrfs_dir_file_operations = {
8049 .llseek = generic_file_llseek,
8050 .read = generic_read_dir,
8051 .readdir = btrfs_real_readdir,
8052 .unlocked_ioctl = btrfs_ioctl,
8053 #ifdef CONFIG_COMPAT
8054 .compat_ioctl = btrfs_ioctl,
8056 .release = btrfs_release_file,
8057 .fsync = btrfs_sync_file,
8060 static struct extent_io_ops btrfs_extent_io_ops = {
8061 .fill_delalloc = run_delalloc_range,
8062 .submit_bio_hook = btrfs_submit_bio_hook,
8063 .merge_bio_hook = btrfs_merge_bio_hook,
8064 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
8065 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
8066 .writepage_start_hook = btrfs_writepage_start_hook,
8067 .set_bit_hook = btrfs_set_bit_hook,
8068 .clear_bit_hook = btrfs_clear_bit_hook,
8069 .merge_extent_hook = btrfs_merge_extent_hook,
8070 .split_extent_hook = btrfs_split_extent_hook,
8074 * btrfs doesn't support the bmap operation because swapfiles
8075 * use bmap to make a mapping of extents in the file. They assume
8076 * these extents won't change over the life of the file and they
8077 * use the bmap result to do IO directly to the drive.
8079 * the btrfs bmap call would return logical addresses that aren't
8080 * suitable for IO and they also will change frequently as COW
8081 * operations happen. So, swapfile + btrfs == corruption.
8083 * For now we're avoiding this by dropping bmap.
8085 static const struct address_space_operations btrfs_aops = {
8086 .readpage = btrfs_readpage,
8087 .writepage = btrfs_writepage,
8088 .writepages = btrfs_writepages,
8089 .readpages = btrfs_readpages,
8090 .direct_IO = btrfs_direct_IO,
8091 .invalidatepage = btrfs_invalidatepage,
8092 .releasepage = btrfs_releasepage,
8093 .set_page_dirty = btrfs_set_page_dirty,
8094 .error_remove_page = generic_error_remove_page,
8097 static const struct address_space_operations btrfs_symlink_aops = {
8098 .readpage = btrfs_readpage,
8099 .writepage = btrfs_writepage,
8100 .invalidatepage = btrfs_invalidatepage,
8101 .releasepage = btrfs_releasepage,
8104 static const struct inode_operations btrfs_file_inode_operations = {
8105 .getattr = btrfs_getattr,
8106 .setattr = btrfs_setattr,
8107 .setxattr = btrfs_setxattr,
8108 .getxattr = btrfs_getxattr,
8109 .listxattr = btrfs_listxattr,
8110 .removexattr = btrfs_removexattr,
8111 .permission = btrfs_permission,
8112 .fiemap = btrfs_fiemap,
8113 .get_acl = btrfs_get_acl,
8114 .update_time = btrfs_update_time,
8116 static const struct inode_operations btrfs_special_inode_operations = {
8117 .getattr = btrfs_getattr,
8118 .setattr = btrfs_setattr,
8119 .permission = btrfs_permission,
8120 .setxattr = btrfs_setxattr,
8121 .getxattr = btrfs_getxattr,
8122 .listxattr = btrfs_listxattr,
8123 .removexattr = btrfs_removexattr,
8124 .get_acl = btrfs_get_acl,
8125 .update_time = btrfs_update_time,
8127 static const struct inode_operations btrfs_symlink_inode_operations = {
8128 .readlink = generic_readlink,
8129 .follow_link = page_follow_link_light,
8130 .put_link = page_put_link,
8131 .getattr = btrfs_getattr,
8132 .setattr = btrfs_setattr,
8133 .permission = btrfs_permission,
8134 .setxattr = btrfs_setxattr,
8135 .getxattr = btrfs_getxattr,
8136 .listxattr = btrfs_listxattr,
8137 .removexattr = btrfs_removexattr,
8138 .get_acl = btrfs_get_acl,
8139 .update_time = btrfs_update_time,
8142 const struct dentry_operations btrfs_dentry_operations = {
8143 .d_delete = btrfs_dentry_delete,
8144 .d_release = btrfs_dentry_release,