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"
60 struct btrfs_iget_args {
62 struct btrfs_root *root;
65 static const struct inode_operations btrfs_dir_inode_operations;
66 static const struct inode_operations btrfs_symlink_inode_operations;
67 static const struct inode_operations btrfs_dir_ro_inode_operations;
68 static const struct inode_operations btrfs_special_inode_operations;
69 static const struct inode_operations btrfs_file_inode_operations;
70 static const struct address_space_operations btrfs_aops;
71 static const struct address_space_operations btrfs_symlink_aops;
72 static const struct file_operations btrfs_dir_file_operations;
73 static struct extent_io_ops btrfs_extent_io_ops;
75 static struct kmem_cache *btrfs_inode_cachep;
76 static struct kmem_cache *btrfs_delalloc_work_cachep;
77 struct kmem_cache *btrfs_trans_handle_cachep;
78 struct kmem_cache *btrfs_transaction_cachep;
79 struct kmem_cache *btrfs_path_cachep;
80 struct kmem_cache *btrfs_free_space_cachep;
83 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
84 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
85 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
86 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
87 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
88 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
89 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
90 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
93 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
94 static int btrfs_truncate(struct inode *inode);
95 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
96 static noinline int cow_file_range(struct inode *inode,
97 struct page *locked_page,
98 u64 start, u64 end, int *page_started,
99 unsigned long *nr_written, int unlock);
100 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
101 u64 len, u64 orig_start,
102 u64 block_start, u64 block_len,
103 u64 orig_block_len, u64 ram_bytes,
106 static int btrfs_dirty_inode(struct inode *inode);
108 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
109 struct inode *inode, struct inode *dir,
110 const struct qstr *qstr)
114 err = btrfs_init_acl(trans, inode, dir);
116 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
121 * this does all the hard work for inserting an inline extent into
122 * the btree. The caller should have done a btrfs_drop_extents so that
123 * no overlapping inline items exist in the btree
125 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
126 struct btrfs_root *root, struct inode *inode,
127 u64 start, size_t size, size_t compressed_size,
129 struct page **compressed_pages)
131 struct btrfs_key key;
132 struct btrfs_path *path;
133 struct extent_buffer *leaf;
134 struct page *page = NULL;
137 struct btrfs_file_extent_item *ei;
140 size_t cur_size = size;
142 unsigned long offset;
144 if (compressed_size && compressed_pages)
145 cur_size = compressed_size;
147 path = btrfs_alloc_path();
151 path->leave_spinning = 1;
153 key.objectid = btrfs_ino(inode);
155 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
156 datasize = btrfs_file_extent_calc_inline_size(cur_size);
158 inode_add_bytes(inode, size);
159 ret = btrfs_insert_empty_item(trans, root, path, &key,
165 leaf = path->nodes[0];
166 ei = btrfs_item_ptr(leaf, path->slots[0],
167 struct btrfs_file_extent_item);
168 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
169 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
170 btrfs_set_file_extent_encryption(leaf, ei, 0);
171 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
172 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
173 ptr = btrfs_file_extent_inline_start(ei);
175 if (compress_type != BTRFS_COMPRESS_NONE) {
178 while (compressed_size > 0) {
179 cpage = compressed_pages[i];
180 cur_size = min_t(unsigned long, compressed_size,
183 kaddr = kmap_atomic(cpage);
184 write_extent_buffer(leaf, kaddr, ptr, cur_size);
185 kunmap_atomic(kaddr);
189 compressed_size -= cur_size;
191 btrfs_set_file_extent_compression(leaf, ei,
194 page = find_get_page(inode->i_mapping,
195 start >> PAGE_CACHE_SHIFT);
196 btrfs_set_file_extent_compression(leaf, ei, 0);
197 kaddr = kmap_atomic(page);
198 offset = start & (PAGE_CACHE_SIZE - 1);
199 write_extent_buffer(leaf, kaddr + offset, ptr, size);
200 kunmap_atomic(kaddr);
201 page_cache_release(page);
203 btrfs_mark_buffer_dirty(leaf);
204 btrfs_free_path(path);
207 * we're an inline extent, so nobody can
208 * extend the file past i_size without locking
209 * a page we already have locked.
211 * We must do any isize and inode updates
212 * before we unlock the pages. Otherwise we
213 * could end up racing with unlink.
215 BTRFS_I(inode)->disk_i_size = inode->i_size;
216 ret = btrfs_update_inode(trans, root, inode);
220 btrfs_free_path(path);
226 * conditionally insert an inline extent into the file. This
227 * does the checks required to make sure the data is small enough
228 * to fit as an inline extent.
230 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
231 struct btrfs_root *root,
232 struct inode *inode, u64 start, u64 end,
233 size_t compressed_size, int compress_type,
234 struct page **compressed_pages)
236 u64 isize = i_size_read(inode);
237 u64 actual_end = min(end + 1, isize);
238 u64 inline_len = actual_end - start;
239 u64 aligned_end = ALIGN(end, root->sectorsize);
240 u64 data_len = inline_len;
244 data_len = compressed_size;
247 actual_end >= PAGE_CACHE_SIZE ||
248 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
250 (actual_end & (root->sectorsize - 1)) == 0) ||
252 data_len > root->fs_info->max_inline) {
256 ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
260 if (isize > actual_end)
261 inline_len = min_t(u64, isize, actual_end);
262 ret = insert_inline_extent(trans, root, inode, start,
263 inline_len, compressed_size,
264 compress_type, compressed_pages);
265 if (ret && ret != -ENOSPC) {
266 btrfs_abort_transaction(trans, root, ret);
268 } else if (ret == -ENOSPC) {
272 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
273 btrfs_delalloc_release_metadata(inode, end + 1 - start);
274 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
278 struct async_extent {
283 unsigned long nr_pages;
285 struct list_head list;
290 struct btrfs_root *root;
291 struct page *locked_page;
294 struct list_head extents;
295 struct btrfs_work work;
298 static noinline int add_async_extent(struct async_cow *cow,
299 u64 start, u64 ram_size,
302 unsigned long nr_pages,
305 struct async_extent *async_extent;
307 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
308 BUG_ON(!async_extent); /* -ENOMEM */
309 async_extent->start = start;
310 async_extent->ram_size = ram_size;
311 async_extent->compressed_size = compressed_size;
312 async_extent->pages = pages;
313 async_extent->nr_pages = nr_pages;
314 async_extent->compress_type = compress_type;
315 list_add_tail(&async_extent->list, &cow->extents);
320 * we create compressed extents in two phases. The first
321 * phase compresses a range of pages that have already been
322 * locked (both pages and state bits are locked).
324 * This is done inside an ordered work queue, and the compression
325 * is spread across many cpus. The actual IO submission is step
326 * two, and the ordered work queue takes care of making sure that
327 * happens in the same order things were put onto the queue by
328 * writepages and friends.
330 * If this code finds it can't get good compression, it puts an
331 * entry onto the work queue to write the uncompressed bytes. This
332 * makes sure that both compressed inodes and uncompressed inodes
333 * are written in the same order that the flusher thread sent them
336 static noinline int compress_file_range(struct inode *inode,
337 struct page *locked_page,
339 struct async_cow *async_cow,
342 struct btrfs_root *root = BTRFS_I(inode)->root;
343 struct btrfs_trans_handle *trans;
345 u64 blocksize = root->sectorsize;
347 u64 isize = i_size_read(inode);
349 struct page **pages = NULL;
350 unsigned long nr_pages;
351 unsigned long nr_pages_ret = 0;
352 unsigned long total_compressed = 0;
353 unsigned long total_in = 0;
354 unsigned long max_compressed = 128 * 1024;
355 unsigned long max_uncompressed = 128 * 1024;
358 int compress_type = root->fs_info->compress_type;
361 /* if this is a small write inside eof, kick off a defrag */
362 if ((end - start + 1) < 16 * 1024 &&
363 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
364 btrfs_add_inode_defrag(NULL, inode);
366 actual_end = min_t(u64, isize, end + 1);
369 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
370 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
373 * we don't want to send crud past the end of i_size through
374 * compression, that's just a waste of CPU time. So, if the
375 * end of the file is before the start of our current
376 * requested range of bytes, we bail out to the uncompressed
377 * cleanup code that can deal with all of this.
379 * It isn't really the fastest way to fix things, but this is a
380 * very uncommon corner.
382 if (actual_end <= start)
383 goto cleanup_and_bail_uncompressed;
385 total_compressed = actual_end - start;
387 /* we want to make sure that amount of ram required to uncompress
388 * an extent is reasonable, so we limit the total size in ram
389 * of a compressed extent to 128k. This is a crucial number
390 * because it also controls how easily we can spread reads across
391 * cpus for decompression.
393 * We also want to make sure the amount of IO required to do
394 * a random read is reasonably small, so we limit the size of
395 * a compressed extent to 128k.
397 total_compressed = min(total_compressed, max_uncompressed);
398 num_bytes = ALIGN(end - start + 1, blocksize);
399 num_bytes = max(blocksize, num_bytes);
404 * we do compression for mount -o compress and when the
405 * inode has not been flagged as nocompress. This flag can
406 * change at any time if we discover bad compression ratios.
408 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
409 (btrfs_test_opt(root, COMPRESS) ||
410 (BTRFS_I(inode)->force_compress) ||
411 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
413 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
415 /* just bail out to the uncompressed code */
419 if (BTRFS_I(inode)->force_compress)
420 compress_type = BTRFS_I(inode)->force_compress;
423 * we need to call clear_page_dirty_for_io on each
424 * page in the range. Otherwise applications with the file
425 * mmap'd can wander in and change the page contents while
426 * we are compressing them.
428 * If the compression fails for any reason, we set the pages
429 * dirty again later on.
431 extent_range_clear_dirty_for_io(inode, start, end);
433 ret = btrfs_compress_pages(compress_type,
434 inode->i_mapping, start,
435 total_compressed, pages,
436 nr_pages, &nr_pages_ret,
442 unsigned long offset = total_compressed &
443 (PAGE_CACHE_SIZE - 1);
444 struct page *page = pages[nr_pages_ret - 1];
447 /* zero the tail end of the last page, we might be
448 * sending it down to disk
451 kaddr = kmap_atomic(page);
452 memset(kaddr + offset, 0,
453 PAGE_CACHE_SIZE - offset);
454 kunmap_atomic(kaddr);
461 trans = btrfs_join_transaction(root);
463 ret = PTR_ERR(trans);
465 goto cleanup_and_out;
467 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
469 /* lets try to make an inline extent */
470 if (ret || total_in < (actual_end - start)) {
471 /* we didn't compress the entire range, try
472 * to make an uncompressed inline extent.
474 ret = cow_file_range_inline(trans, root, inode,
475 start, end, 0, 0, NULL);
477 /* try making a compressed inline extent */
478 ret = cow_file_range_inline(trans, root, inode,
481 compress_type, pages);
485 * inline extent creation worked or returned error,
486 * we don't need to create any more async work items.
487 * Unlock and free up our temp pages.
489 extent_clear_unlock_delalloc(inode,
490 &BTRFS_I(inode)->io_tree,
492 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
493 EXTENT_CLEAR_DELALLOC |
494 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
496 btrfs_end_transaction(trans, root);
499 btrfs_end_transaction(trans, root);
504 * we aren't doing an inline extent round the compressed size
505 * up to a block size boundary so the allocator does sane
508 total_compressed = ALIGN(total_compressed, blocksize);
511 * one last check to make sure the compression is really a
512 * win, compare the page count read with the blocks on disk
514 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
515 if (total_compressed >= total_in) {
518 num_bytes = total_in;
521 if (!will_compress && pages) {
523 * the compression code ran but failed to make things smaller,
524 * free any pages it allocated and our page pointer array
526 for (i = 0; i < nr_pages_ret; i++) {
527 WARN_ON(pages[i]->mapping);
528 page_cache_release(pages[i]);
532 total_compressed = 0;
535 /* flag the file so we don't compress in the future */
536 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
537 !(BTRFS_I(inode)->force_compress)) {
538 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
544 /* the async work queues will take care of doing actual
545 * allocation on disk for these compressed pages,
546 * and will submit them to the elevator.
548 add_async_extent(async_cow, start, num_bytes,
549 total_compressed, pages, nr_pages_ret,
552 if (start + num_bytes < end) {
559 cleanup_and_bail_uncompressed:
561 * No compression, but we still need to write the pages in
562 * the file we've been given so far. redirty the locked
563 * page if it corresponds to our extent and set things up
564 * for the async work queue to run cow_file_range to do
565 * the normal delalloc dance
567 if (page_offset(locked_page) >= start &&
568 page_offset(locked_page) <= end) {
569 __set_page_dirty_nobuffers(locked_page);
570 /* unlocked later on in the async handlers */
573 extent_range_redirty_for_io(inode, start, end);
574 add_async_extent(async_cow, start, end - start + 1,
575 0, NULL, 0, BTRFS_COMPRESS_NONE);
583 for (i = 0; i < nr_pages_ret; i++) {
584 WARN_ON(pages[i]->mapping);
585 page_cache_release(pages[i]);
592 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
594 EXTENT_CLEAR_UNLOCK_PAGE |
596 EXTENT_CLEAR_DELALLOC |
597 EXTENT_SET_WRITEBACK |
598 EXTENT_END_WRITEBACK);
599 if (!trans || IS_ERR(trans))
600 btrfs_error(root->fs_info, ret, "Failed to join transaction");
602 btrfs_abort_transaction(trans, root, ret);
607 * phase two of compressed writeback. This is the ordered portion
608 * of the code, which only gets called in the order the work was
609 * queued. We walk all the async extents created by compress_file_range
610 * and send them down to the disk.
612 static noinline int submit_compressed_extents(struct inode *inode,
613 struct async_cow *async_cow)
615 struct async_extent *async_extent;
617 struct btrfs_trans_handle *trans;
618 struct btrfs_key ins;
619 struct extent_map *em;
620 struct btrfs_root *root = BTRFS_I(inode)->root;
621 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
622 struct extent_io_tree *io_tree;
625 if (list_empty(&async_cow->extents))
629 while (!list_empty(&async_cow->extents)) {
630 async_extent = list_entry(async_cow->extents.next,
631 struct async_extent, list);
632 list_del(&async_extent->list);
634 io_tree = &BTRFS_I(inode)->io_tree;
637 /* did the compression code fall back to uncompressed IO? */
638 if (!async_extent->pages) {
639 int page_started = 0;
640 unsigned long nr_written = 0;
642 lock_extent(io_tree, async_extent->start,
643 async_extent->start +
644 async_extent->ram_size - 1);
646 /* allocate blocks */
647 ret = cow_file_range(inode, async_cow->locked_page,
649 async_extent->start +
650 async_extent->ram_size - 1,
651 &page_started, &nr_written, 0);
656 * if page_started, cow_file_range inserted an
657 * inline extent and took care of all the unlocking
658 * and IO for us. Otherwise, we need to submit
659 * all those pages down to the drive.
661 if (!page_started && !ret)
662 extent_write_locked_range(io_tree,
663 inode, async_extent->start,
664 async_extent->start +
665 async_extent->ram_size - 1,
669 unlock_page(async_cow->locked_page);
675 lock_extent(io_tree, async_extent->start,
676 async_extent->start + async_extent->ram_size - 1);
678 trans = btrfs_join_transaction(root);
680 ret = PTR_ERR(trans);
682 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
683 ret = btrfs_reserve_extent(trans, root,
684 async_extent->compressed_size,
685 async_extent->compressed_size,
686 0, alloc_hint, &ins, 1);
687 if (ret && ret != -ENOSPC)
688 btrfs_abort_transaction(trans, root, ret);
689 btrfs_end_transaction(trans, root);
695 for (i = 0; i < async_extent->nr_pages; i++) {
696 WARN_ON(async_extent->pages[i]->mapping);
697 page_cache_release(async_extent->pages[i]);
699 kfree(async_extent->pages);
700 async_extent->nr_pages = 0;
701 async_extent->pages = NULL;
709 * here we're doing allocation and writeback of the
712 btrfs_drop_extent_cache(inode, async_extent->start,
713 async_extent->start +
714 async_extent->ram_size - 1, 0);
716 em = alloc_extent_map();
719 goto out_free_reserve;
721 em->start = async_extent->start;
722 em->len = async_extent->ram_size;
723 em->orig_start = em->start;
724 em->mod_start = em->start;
725 em->mod_len = em->len;
727 em->block_start = ins.objectid;
728 em->block_len = ins.offset;
729 em->orig_block_len = ins.offset;
730 em->ram_bytes = async_extent->ram_size;
731 em->bdev = root->fs_info->fs_devices->latest_bdev;
732 em->compress_type = async_extent->compress_type;
733 set_bit(EXTENT_FLAG_PINNED, &em->flags);
734 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
738 write_lock(&em_tree->lock);
739 ret = add_extent_mapping(em_tree, em, 1);
740 write_unlock(&em_tree->lock);
741 if (ret != -EEXIST) {
745 btrfs_drop_extent_cache(inode, async_extent->start,
746 async_extent->start +
747 async_extent->ram_size - 1, 0);
751 goto out_free_reserve;
753 ret = btrfs_add_ordered_extent_compress(inode,
756 async_extent->ram_size,
758 BTRFS_ORDERED_COMPRESSED,
759 async_extent->compress_type);
761 goto out_free_reserve;
764 * clear dirty, set writeback and unlock the pages.
766 extent_clear_unlock_delalloc(inode,
767 &BTRFS_I(inode)->io_tree,
769 async_extent->start +
770 async_extent->ram_size - 1,
771 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
772 EXTENT_CLEAR_UNLOCK |
773 EXTENT_CLEAR_DELALLOC |
774 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
776 ret = btrfs_submit_compressed_write(inode,
778 async_extent->ram_size,
780 ins.offset, async_extent->pages,
781 async_extent->nr_pages);
782 alloc_hint = ins.objectid + ins.offset;
792 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
794 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
796 async_extent->start +
797 async_extent->ram_size - 1,
798 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
799 EXTENT_CLEAR_UNLOCK |
800 EXTENT_CLEAR_DELALLOC |
802 EXTENT_SET_WRITEBACK |
803 EXTENT_END_WRITEBACK);
808 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
811 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
812 struct extent_map *em;
815 read_lock(&em_tree->lock);
816 em = search_extent_mapping(em_tree, start, num_bytes);
819 * if block start isn't an actual block number then find the
820 * first block in this inode and use that as a hint. If that
821 * block is also bogus then just don't worry about it.
823 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
825 em = search_extent_mapping(em_tree, 0, 0);
826 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
827 alloc_hint = em->block_start;
831 alloc_hint = em->block_start;
835 read_unlock(&em_tree->lock);
841 * when extent_io.c finds a delayed allocation range in the file,
842 * the call backs end up in this code. The basic idea is to
843 * allocate extents on disk for the range, and create ordered data structs
844 * in ram to track those extents.
846 * locked_page is the page that writepage had locked already. We use
847 * it to make sure we don't do extra locks or unlocks.
849 * *page_started is set to one if we unlock locked_page and do everything
850 * required to start IO on it. It may be clean and already done with
853 static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
855 struct btrfs_root *root,
856 struct page *locked_page,
857 u64 start, u64 end, int *page_started,
858 unsigned long *nr_written,
863 unsigned long ram_size;
866 u64 blocksize = root->sectorsize;
867 struct btrfs_key ins;
868 struct extent_map *em;
869 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
872 BUG_ON(btrfs_is_free_space_inode(inode));
874 num_bytes = ALIGN(end - start + 1, blocksize);
875 num_bytes = max(blocksize, num_bytes);
876 disk_num_bytes = num_bytes;
878 /* if this is a small write inside eof, kick off defrag */
879 if (num_bytes < 64 * 1024 &&
880 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
881 btrfs_add_inode_defrag(trans, inode);
884 /* lets try to make an inline extent */
885 ret = cow_file_range_inline(trans, root, inode,
886 start, end, 0, 0, NULL);
888 extent_clear_unlock_delalloc(inode,
889 &BTRFS_I(inode)->io_tree,
891 EXTENT_CLEAR_UNLOCK_PAGE |
892 EXTENT_CLEAR_UNLOCK |
893 EXTENT_CLEAR_DELALLOC |
895 EXTENT_SET_WRITEBACK |
896 EXTENT_END_WRITEBACK);
898 *nr_written = *nr_written +
899 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
902 } else if (ret < 0) {
903 btrfs_abort_transaction(trans, root, ret);
908 BUG_ON(disk_num_bytes >
909 btrfs_super_total_bytes(root->fs_info->super_copy));
911 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
912 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
914 while (disk_num_bytes > 0) {
917 cur_alloc_size = disk_num_bytes;
918 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
919 root->sectorsize, 0, alloc_hint,
922 btrfs_abort_transaction(trans, root, ret);
926 em = alloc_extent_map();
932 em->orig_start = em->start;
933 ram_size = ins.offset;
934 em->len = ins.offset;
935 em->mod_start = em->start;
936 em->mod_len = em->len;
938 em->block_start = ins.objectid;
939 em->block_len = ins.offset;
940 em->orig_block_len = ins.offset;
941 em->ram_bytes = ram_size;
942 em->bdev = root->fs_info->fs_devices->latest_bdev;
943 set_bit(EXTENT_FLAG_PINNED, &em->flags);
947 write_lock(&em_tree->lock);
948 ret = add_extent_mapping(em_tree, em, 1);
949 write_unlock(&em_tree->lock);
950 if (ret != -EEXIST) {
954 btrfs_drop_extent_cache(inode, start,
955 start + ram_size - 1, 0);
960 cur_alloc_size = ins.offset;
961 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
962 ram_size, cur_alloc_size, 0);
966 if (root->root_key.objectid ==
967 BTRFS_DATA_RELOC_TREE_OBJECTID) {
968 ret = btrfs_reloc_clone_csums(inode, start,
971 btrfs_abort_transaction(trans, root, ret);
976 if (disk_num_bytes < cur_alloc_size)
979 /* we're not doing compressed IO, don't unlock the first
980 * page (which the caller expects to stay locked), don't
981 * clear any dirty bits and don't set any writeback bits
983 * Do set the Private2 bit so we know this page was properly
984 * setup for writepage
986 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
987 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
990 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
991 start, start + ram_size - 1,
993 disk_num_bytes -= cur_alloc_size;
994 num_bytes -= cur_alloc_size;
995 alloc_hint = ins.objectid + ins.offset;
996 start += cur_alloc_size;
1002 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
1004 extent_clear_unlock_delalloc(inode,
1005 &BTRFS_I(inode)->io_tree,
1006 start, end, locked_page,
1007 EXTENT_CLEAR_UNLOCK_PAGE |
1008 EXTENT_CLEAR_UNLOCK |
1009 EXTENT_CLEAR_DELALLOC |
1010 EXTENT_CLEAR_DIRTY |
1011 EXTENT_SET_WRITEBACK |
1012 EXTENT_END_WRITEBACK);
1017 static noinline int cow_file_range(struct inode *inode,
1018 struct page *locked_page,
1019 u64 start, u64 end, int *page_started,
1020 unsigned long *nr_written,
1023 struct btrfs_trans_handle *trans;
1024 struct btrfs_root *root = BTRFS_I(inode)->root;
1027 trans = btrfs_join_transaction(root);
1028 if (IS_ERR(trans)) {
1029 extent_clear_unlock_delalloc(inode,
1030 &BTRFS_I(inode)->io_tree,
1031 start, end, locked_page,
1032 EXTENT_CLEAR_UNLOCK_PAGE |
1033 EXTENT_CLEAR_UNLOCK |
1034 EXTENT_CLEAR_DELALLOC |
1035 EXTENT_CLEAR_DIRTY |
1036 EXTENT_SET_WRITEBACK |
1037 EXTENT_END_WRITEBACK);
1038 return PTR_ERR(trans);
1040 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1042 ret = __cow_file_range(trans, inode, root, locked_page, start, end,
1043 page_started, nr_written, unlock);
1045 btrfs_end_transaction(trans, root);
1051 * work queue call back to started compression on a file and pages
1053 static noinline void async_cow_start(struct btrfs_work *work)
1055 struct async_cow *async_cow;
1057 async_cow = container_of(work, struct async_cow, work);
1059 compress_file_range(async_cow->inode, async_cow->locked_page,
1060 async_cow->start, async_cow->end, async_cow,
1062 if (num_added == 0) {
1063 btrfs_add_delayed_iput(async_cow->inode);
1064 async_cow->inode = NULL;
1069 * work queue call back to submit previously compressed pages
1071 static noinline void async_cow_submit(struct btrfs_work *work)
1073 struct async_cow *async_cow;
1074 struct btrfs_root *root;
1075 unsigned long nr_pages;
1077 async_cow = container_of(work, struct async_cow, work);
1079 root = async_cow->root;
1080 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1083 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1085 waitqueue_active(&root->fs_info->async_submit_wait))
1086 wake_up(&root->fs_info->async_submit_wait);
1088 if (async_cow->inode)
1089 submit_compressed_extents(async_cow->inode, async_cow);
1092 static noinline void async_cow_free(struct btrfs_work *work)
1094 struct async_cow *async_cow;
1095 async_cow = container_of(work, struct async_cow, work);
1096 if (async_cow->inode)
1097 btrfs_add_delayed_iput(async_cow->inode);
1101 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1102 u64 start, u64 end, int *page_started,
1103 unsigned long *nr_written)
1105 struct async_cow *async_cow;
1106 struct btrfs_root *root = BTRFS_I(inode)->root;
1107 unsigned long nr_pages;
1109 int limit = 10 * 1024 * 1024;
1111 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1112 1, 0, NULL, GFP_NOFS);
1113 while (start < end) {
1114 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1115 BUG_ON(!async_cow); /* -ENOMEM */
1116 async_cow->inode = igrab(inode);
1117 async_cow->root = root;
1118 async_cow->locked_page = locked_page;
1119 async_cow->start = start;
1121 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1124 cur_end = min(end, start + 512 * 1024 - 1);
1126 async_cow->end = cur_end;
1127 INIT_LIST_HEAD(&async_cow->extents);
1129 async_cow->work.func = async_cow_start;
1130 async_cow->work.ordered_func = async_cow_submit;
1131 async_cow->work.ordered_free = async_cow_free;
1132 async_cow->work.flags = 0;
1134 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1136 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1138 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1141 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1142 wait_event(root->fs_info->async_submit_wait,
1143 (atomic_read(&root->fs_info->async_delalloc_pages) <
1147 while (atomic_read(&root->fs_info->async_submit_draining) &&
1148 atomic_read(&root->fs_info->async_delalloc_pages)) {
1149 wait_event(root->fs_info->async_submit_wait,
1150 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1154 *nr_written += nr_pages;
1155 start = cur_end + 1;
1161 static noinline int csum_exist_in_range(struct btrfs_root *root,
1162 u64 bytenr, u64 num_bytes)
1165 struct btrfs_ordered_sum *sums;
1168 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1169 bytenr + num_bytes - 1, &list, 0);
1170 if (ret == 0 && list_empty(&list))
1173 while (!list_empty(&list)) {
1174 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1175 list_del(&sums->list);
1182 * when nowcow writeback call back. This checks for snapshots or COW copies
1183 * of the extents that exist in the file, and COWs the file as required.
1185 * If no cow copies or snapshots exist, we write directly to the existing
1188 static noinline int run_delalloc_nocow(struct inode *inode,
1189 struct page *locked_page,
1190 u64 start, u64 end, int *page_started, int force,
1191 unsigned long *nr_written)
1193 struct btrfs_root *root = BTRFS_I(inode)->root;
1194 struct btrfs_trans_handle *trans;
1195 struct extent_buffer *leaf;
1196 struct btrfs_path *path;
1197 struct btrfs_file_extent_item *fi;
1198 struct btrfs_key found_key;
1213 u64 ino = btrfs_ino(inode);
1215 path = btrfs_alloc_path();
1217 extent_clear_unlock_delalloc(inode,
1218 &BTRFS_I(inode)->io_tree,
1219 start, end, locked_page,
1220 EXTENT_CLEAR_UNLOCK_PAGE |
1221 EXTENT_CLEAR_UNLOCK |
1222 EXTENT_CLEAR_DELALLOC |
1223 EXTENT_CLEAR_DIRTY |
1224 EXTENT_SET_WRITEBACK |
1225 EXTENT_END_WRITEBACK);
1229 nolock = btrfs_is_free_space_inode(inode);
1232 trans = btrfs_join_transaction_nolock(root);
1234 trans = btrfs_join_transaction(root);
1236 if (IS_ERR(trans)) {
1237 extent_clear_unlock_delalloc(inode,
1238 &BTRFS_I(inode)->io_tree,
1239 start, end, locked_page,
1240 EXTENT_CLEAR_UNLOCK_PAGE |
1241 EXTENT_CLEAR_UNLOCK |
1242 EXTENT_CLEAR_DELALLOC |
1243 EXTENT_CLEAR_DIRTY |
1244 EXTENT_SET_WRITEBACK |
1245 EXTENT_END_WRITEBACK);
1246 btrfs_free_path(path);
1247 return PTR_ERR(trans);
1250 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1252 cow_start = (u64)-1;
1255 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1258 btrfs_abort_transaction(trans, root, ret);
1261 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1262 leaf = path->nodes[0];
1263 btrfs_item_key_to_cpu(leaf, &found_key,
1264 path->slots[0] - 1);
1265 if (found_key.objectid == ino &&
1266 found_key.type == BTRFS_EXTENT_DATA_KEY)
1271 leaf = path->nodes[0];
1272 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1273 ret = btrfs_next_leaf(root, path);
1275 btrfs_abort_transaction(trans, root, ret);
1280 leaf = path->nodes[0];
1286 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1288 if (found_key.objectid > ino ||
1289 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1290 found_key.offset > end)
1293 if (found_key.offset > cur_offset) {
1294 extent_end = found_key.offset;
1299 fi = btrfs_item_ptr(leaf, path->slots[0],
1300 struct btrfs_file_extent_item);
1301 extent_type = btrfs_file_extent_type(leaf, fi);
1303 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1304 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1305 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1306 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1307 extent_offset = btrfs_file_extent_offset(leaf, fi);
1308 extent_end = found_key.offset +
1309 btrfs_file_extent_num_bytes(leaf, fi);
1311 btrfs_file_extent_disk_num_bytes(leaf, fi);
1312 if (extent_end <= start) {
1316 if (disk_bytenr == 0)
1318 if (btrfs_file_extent_compression(leaf, fi) ||
1319 btrfs_file_extent_encryption(leaf, fi) ||
1320 btrfs_file_extent_other_encoding(leaf, fi))
1322 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1324 if (btrfs_extent_readonly(root, disk_bytenr))
1326 if (btrfs_cross_ref_exist(trans, root, ino,
1328 extent_offset, disk_bytenr))
1330 disk_bytenr += extent_offset;
1331 disk_bytenr += cur_offset - found_key.offset;
1332 num_bytes = min(end + 1, extent_end) - cur_offset;
1334 * force cow if csum exists in the range.
1335 * this ensure that csum for a given extent are
1336 * either valid or do not exist.
1338 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1341 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1342 extent_end = found_key.offset +
1343 btrfs_file_extent_inline_len(leaf, fi);
1344 extent_end = ALIGN(extent_end, root->sectorsize);
1349 if (extent_end <= start) {
1354 if (cow_start == (u64)-1)
1355 cow_start = cur_offset;
1356 cur_offset = extent_end;
1357 if (cur_offset > end)
1363 btrfs_release_path(path);
1364 if (cow_start != (u64)-1) {
1365 ret = __cow_file_range(trans, inode, root, locked_page,
1366 cow_start, found_key.offset - 1,
1367 page_started, nr_written, 1);
1369 btrfs_abort_transaction(trans, root, ret);
1372 cow_start = (u64)-1;
1375 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1376 struct extent_map *em;
1377 struct extent_map_tree *em_tree;
1378 em_tree = &BTRFS_I(inode)->extent_tree;
1379 em = alloc_extent_map();
1380 BUG_ON(!em); /* -ENOMEM */
1381 em->start = cur_offset;
1382 em->orig_start = found_key.offset - extent_offset;
1383 em->len = num_bytes;
1384 em->block_len = num_bytes;
1385 em->block_start = disk_bytenr;
1386 em->orig_block_len = disk_num_bytes;
1387 em->ram_bytes = ram_bytes;
1388 em->bdev = root->fs_info->fs_devices->latest_bdev;
1389 em->mod_start = em->start;
1390 em->mod_len = em->len;
1391 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1392 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1393 em->generation = -1;
1395 write_lock(&em_tree->lock);
1396 ret = add_extent_mapping(em_tree, em, 1);
1397 write_unlock(&em_tree->lock);
1398 if (ret != -EEXIST) {
1399 free_extent_map(em);
1402 btrfs_drop_extent_cache(inode, em->start,
1403 em->start + em->len - 1, 0);
1405 type = BTRFS_ORDERED_PREALLOC;
1407 type = BTRFS_ORDERED_NOCOW;
1410 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1411 num_bytes, num_bytes, type);
1412 BUG_ON(ret); /* -ENOMEM */
1414 if (root->root_key.objectid ==
1415 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1416 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1419 btrfs_abort_transaction(trans, root, ret);
1424 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1425 cur_offset, cur_offset + num_bytes - 1,
1426 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1427 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1428 EXTENT_SET_PRIVATE2);
1429 cur_offset = extent_end;
1430 if (cur_offset > end)
1433 btrfs_release_path(path);
1435 if (cur_offset <= end && cow_start == (u64)-1) {
1436 cow_start = cur_offset;
1440 if (cow_start != (u64)-1) {
1441 ret = __cow_file_range(trans, inode, root, locked_page,
1443 page_started, nr_written, 1);
1445 btrfs_abort_transaction(trans, root, ret);
1451 err = btrfs_end_transaction(trans, root);
1455 if (ret && cur_offset < end)
1456 extent_clear_unlock_delalloc(inode,
1457 &BTRFS_I(inode)->io_tree,
1458 cur_offset, end, locked_page,
1459 EXTENT_CLEAR_UNLOCK_PAGE |
1460 EXTENT_CLEAR_UNLOCK |
1461 EXTENT_CLEAR_DELALLOC |
1462 EXTENT_CLEAR_DIRTY |
1463 EXTENT_SET_WRITEBACK |
1464 EXTENT_END_WRITEBACK);
1466 btrfs_free_path(path);
1471 * extent_io.c call back to do delayed allocation processing
1473 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1474 u64 start, u64 end, int *page_started,
1475 unsigned long *nr_written)
1478 struct btrfs_root *root = BTRFS_I(inode)->root;
1480 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1481 ret = run_delalloc_nocow(inode, locked_page, start, end,
1482 page_started, 1, nr_written);
1483 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1484 ret = run_delalloc_nocow(inode, locked_page, start, end,
1485 page_started, 0, nr_written);
1486 } else if (!btrfs_test_opt(root, COMPRESS) &&
1487 !(BTRFS_I(inode)->force_compress) &&
1488 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1489 ret = cow_file_range(inode, locked_page, start, end,
1490 page_started, nr_written, 1);
1492 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1493 &BTRFS_I(inode)->runtime_flags);
1494 ret = cow_file_range_async(inode, locked_page, start, end,
1495 page_started, nr_written);
1500 static void btrfs_split_extent_hook(struct inode *inode,
1501 struct extent_state *orig, u64 split)
1503 /* not delalloc, ignore it */
1504 if (!(orig->state & EXTENT_DELALLOC))
1507 spin_lock(&BTRFS_I(inode)->lock);
1508 BTRFS_I(inode)->outstanding_extents++;
1509 spin_unlock(&BTRFS_I(inode)->lock);
1513 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1514 * extents so we can keep track of new extents that are just merged onto old
1515 * extents, such as when we are doing sequential writes, so we can properly
1516 * account for the metadata space we'll need.
1518 static void btrfs_merge_extent_hook(struct inode *inode,
1519 struct extent_state *new,
1520 struct extent_state *other)
1522 /* not delalloc, ignore it */
1523 if (!(other->state & EXTENT_DELALLOC))
1526 spin_lock(&BTRFS_I(inode)->lock);
1527 BTRFS_I(inode)->outstanding_extents--;
1528 spin_unlock(&BTRFS_I(inode)->lock);
1532 * extent_io.c set_bit_hook, used to track delayed allocation
1533 * bytes in this file, and to maintain the list of inodes that
1534 * have pending delalloc work to be done.
1536 static void btrfs_set_bit_hook(struct inode *inode,
1537 struct extent_state *state, unsigned long *bits)
1541 * set_bit and clear bit hooks normally require _irqsave/restore
1542 * but in this case, we are only testing for the DELALLOC
1543 * bit, which is only set or cleared with irqs on
1545 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1546 struct btrfs_root *root = BTRFS_I(inode)->root;
1547 u64 len = state->end + 1 - state->start;
1548 bool do_list = !btrfs_is_free_space_inode(inode);
1550 if (*bits & EXTENT_FIRST_DELALLOC) {
1551 *bits &= ~EXTENT_FIRST_DELALLOC;
1553 spin_lock(&BTRFS_I(inode)->lock);
1554 BTRFS_I(inode)->outstanding_extents++;
1555 spin_unlock(&BTRFS_I(inode)->lock);
1558 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1559 root->fs_info->delalloc_batch);
1560 spin_lock(&BTRFS_I(inode)->lock);
1561 BTRFS_I(inode)->delalloc_bytes += len;
1562 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1563 &BTRFS_I(inode)->runtime_flags)) {
1564 spin_lock(&root->fs_info->delalloc_lock);
1565 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1566 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1567 &root->fs_info->delalloc_inodes);
1568 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1569 &BTRFS_I(inode)->runtime_flags);
1571 spin_unlock(&root->fs_info->delalloc_lock);
1573 spin_unlock(&BTRFS_I(inode)->lock);
1578 * extent_io.c clear_bit_hook, see set_bit_hook for why
1580 static void btrfs_clear_bit_hook(struct inode *inode,
1581 struct extent_state *state,
1582 unsigned long *bits)
1585 * set_bit and clear bit hooks normally require _irqsave/restore
1586 * but in this case, we are only testing for the DELALLOC
1587 * bit, which is only set or cleared with irqs on
1589 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1590 struct btrfs_root *root = BTRFS_I(inode)->root;
1591 u64 len = state->end + 1 - state->start;
1592 bool do_list = !btrfs_is_free_space_inode(inode);
1594 if (*bits & EXTENT_FIRST_DELALLOC) {
1595 *bits &= ~EXTENT_FIRST_DELALLOC;
1596 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1597 spin_lock(&BTRFS_I(inode)->lock);
1598 BTRFS_I(inode)->outstanding_extents--;
1599 spin_unlock(&BTRFS_I(inode)->lock);
1602 if (*bits & EXTENT_DO_ACCOUNTING)
1603 btrfs_delalloc_release_metadata(inode, len);
1605 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1607 btrfs_free_reserved_data_space(inode, len);
1609 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1610 root->fs_info->delalloc_batch);
1611 spin_lock(&BTRFS_I(inode)->lock);
1612 BTRFS_I(inode)->delalloc_bytes -= len;
1613 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1614 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1615 &BTRFS_I(inode)->runtime_flags)) {
1616 spin_lock(&root->fs_info->delalloc_lock);
1617 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1618 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1619 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1620 &BTRFS_I(inode)->runtime_flags);
1622 spin_unlock(&root->fs_info->delalloc_lock);
1624 spin_unlock(&BTRFS_I(inode)->lock);
1629 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1630 * we don't create bios that span stripes or chunks
1632 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1633 size_t size, struct bio *bio,
1634 unsigned long bio_flags)
1636 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1637 u64 logical = (u64)bio->bi_sector << 9;
1642 if (bio_flags & EXTENT_BIO_COMPRESSED)
1645 length = bio->bi_size;
1646 map_length = length;
1647 ret = btrfs_map_block(root->fs_info, rw, logical,
1648 &map_length, NULL, 0);
1649 /* Will always return 0 with map_multi == NULL */
1651 if (map_length < length + size)
1657 * in order to insert checksums into the metadata in large chunks,
1658 * we wait until bio submission time. All the pages in the bio are
1659 * checksummed and sums are attached onto the ordered extent record.
1661 * At IO completion time the cums attached on the ordered extent record
1662 * are inserted into the btree
1664 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1665 struct bio *bio, int mirror_num,
1666 unsigned long bio_flags,
1669 struct btrfs_root *root = BTRFS_I(inode)->root;
1672 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1673 BUG_ON(ret); /* -ENOMEM */
1678 * in order to insert checksums into the metadata in large chunks,
1679 * we wait until bio submission time. All the pages in the bio are
1680 * checksummed and sums are attached onto the ordered extent record.
1682 * At IO completion time the cums attached on the ordered extent record
1683 * are inserted into the btree
1685 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1686 int mirror_num, unsigned long bio_flags,
1689 struct btrfs_root *root = BTRFS_I(inode)->root;
1692 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1694 bio_endio(bio, ret);
1699 * extent_io.c submission hook. This does the right thing for csum calculation
1700 * on write, or reading the csums from the tree before a read
1702 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1703 int mirror_num, unsigned long bio_flags,
1706 struct btrfs_root *root = BTRFS_I(inode)->root;
1710 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1712 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1714 if (btrfs_is_free_space_inode(inode))
1717 if (!(rw & REQ_WRITE)) {
1718 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1722 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1723 ret = btrfs_submit_compressed_read(inode, bio,
1727 } else if (!skip_sum) {
1728 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1733 } else if (async && !skip_sum) {
1734 /* csum items have already been cloned */
1735 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1737 /* we're doing a write, do the async checksumming */
1738 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1739 inode, rw, bio, mirror_num,
1740 bio_flags, bio_offset,
1741 __btrfs_submit_bio_start,
1742 __btrfs_submit_bio_done);
1744 } else if (!skip_sum) {
1745 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1751 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1755 bio_endio(bio, ret);
1760 * given a list of ordered sums record them in the inode. This happens
1761 * at IO completion time based on sums calculated at bio submission time.
1763 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1764 struct inode *inode, u64 file_offset,
1765 struct list_head *list)
1767 struct btrfs_ordered_sum *sum;
1769 list_for_each_entry(sum, list, list) {
1770 trans->adding_csums = 1;
1771 btrfs_csum_file_blocks(trans,
1772 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1773 trans->adding_csums = 0;
1778 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1779 struct extent_state **cached_state)
1781 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1782 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1783 cached_state, GFP_NOFS);
1786 /* see btrfs_writepage_start_hook for details on why this is required */
1787 struct btrfs_writepage_fixup {
1789 struct btrfs_work work;
1792 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1794 struct btrfs_writepage_fixup *fixup;
1795 struct btrfs_ordered_extent *ordered;
1796 struct extent_state *cached_state = NULL;
1798 struct inode *inode;
1803 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1807 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1808 ClearPageChecked(page);
1812 inode = page->mapping->host;
1813 page_start = page_offset(page);
1814 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1816 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1819 /* already ordered? We're done */
1820 if (PagePrivate2(page))
1823 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1825 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1826 page_end, &cached_state, GFP_NOFS);
1828 btrfs_start_ordered_extent(inode, ordered, 1);
1829 btrfs_put_ordered_extent(ordered);
1833 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1835 mapping_set_error(page->mapping, ret);
1836 end_extent_writepage(page, ret, page_start, page_end);
1837 ClearPageChecked(page);
1841 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1842 ClearPageChecked(page);
1843 set_page_dirty(page);
1845 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1846 &cached_state, GFP_NOFS);
1849 page_cache_release(page);
1854 * There are a few paths in the higher layers of the kernel that directly
1855 * set the page dirty bit without asking the filesystem if it is a
1856 * good idea. This causes problems because we want to make sure COW
1857 * properly happens and the data=ordered rules are followed.
1859 * In our case any range that doesn't have the ORDERED bit set
1860 * hasn't been properly setup for IO. We kick off an async process
1861 * to fix it up. The async helper will wait for ordered extents, set
1862 * the delalloc bit and make it safe to write the page.
1864 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1866 struct inode *inode = page->mapping->host;
1867 struct btrfs_writepage_fixup *fixup;
1868 struct btrfs_root *root = BTRFS_I(inode)->root;
1870 /* this page is properly in the ordered list */
1871 if (TestClearPagePrivate2(page))
1874 if (PageChecked(page))
1877 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1881 SetPageChecked(page);
1882 page_cache_get(page);
1883 fixup->work.func = btrfs_writepage_fixup_worker;
1885 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1889 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1890 struct inode *inode, u64 file_pos,
1891 u64 disk_bytenr, u64 disk_num_bytes,
1892 u64 num_bytes, u64 ram_bytes,
1893 u8 compression, u8 encryption,
1894 u16 other_encoding, int extent_type)
1896 struct btrfs_root *root = BTRFS_I(inode)->root;
1897 struct btrfs_file_extent_item *fi;
1898 struct btrfs_path *path;
1899 struct extent_buffer *leaf;
1900 struct btrfs_key ins;
1903 path = btrfs_alloc_path();
1907 path->leave_spinning = 1;
1910 * we may be replacing one extent in the tree with another.
1911 * The new extent is pinned in the extent map, and we don't want
1912 * to drop it from the cache until it is completely in the btree.
1914 * So, tell btrfs_drop_extents to leave this extent in the cache.
1915 * the caller is expected to unpin it and allow it to be merged
1918 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1919 file_pos + num_bytes, 0);
1923 ins.objectid = btrfs_ino(inode);
1924 ins.offset = file_pos;
1925 ins.type = BTRFS_EXTENT_DATA_KEY;
1926 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1929 leaf = path->nodes[0];
1930 fi = btrfs_item_ptr(leaf, path->slots[0],
1931 struct btrfs_file_extent_item);
1932 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1933 btrfs_set_file_extent_type(leaf, fi, extent_type);
1934 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1935 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1936 btrfs_set_file_extent_offset(leaf, fi, 0);
1937 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1938 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1939 btrfs_set_file_extent_compression(leaf, fi, compression);
1940 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1941 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1943 btrfs_mark_buffer_dirty(leaf);
1944 btrfs_release_path(path);
1946 inode_add_bytes(inode, num_bytes);
1948 ins.objectid = disk_bytenr;
1949 ins.offset = disk_num_bytes;
1950 ins.type = BTRFS_EXTENT_ITEM_KEY;
1951 ret = btrfs_alloc_reserved_file_extent(trans, root,
1952 root->root_key.objectid,
1953 btrfs_ino(inode), file_pos, &ins);
1955 btrfs_free_path(path);
1960 /* snapshot-aware defrag */
1961 struct sa_defrag_extent_backref {
1962 struct rb_node node;
1963 struct old_sa_defrag_extent *old;
1972 struct old_sa_defrag_extent {
1973 struct list_head list;
1974 struct new_sa_defrag_extent *new;
1983 struct new_sa_defrag_extent {
1984 struct rb_root root;
1985 struct list_head head;
1986 struct btrfs_path *path;
1987 struct inode *inode;
1995 static int backref_comp(struct sa_defrag_extent_backref *b1,
1996 struct sa_defrag_extent_backref *b2)
1998 if (b1->root_id < b2->root_id)
2000 else if (b1->root_id > b2->root_id)
2003 if (b1->inum < b2->inum)
2005 else if (b1->inum > b2->inum)
2008 if (b1->file_pos < b2->file_pos)
2010 else if (b1->file_pos > b2->file_pos)
2014 * [------------------------------] ===> (a range of space)
2015 * |<--->| |<---->| =============> (fs/file tree A)
2016 * |<---------------------------->| ===> (fs/file tree B)
2018 * A range of space can refer to two file extents in one tree while
2019 * refer to only one file extent in another tree.
2021 * So we may process a disk offset more than one time(two extents in A)
2022 * and locate at the same extent(one extent in B), then insert two same
2023 * backrefs(both refer to the extent in B).
2028 static void backref_insert(struct rb_root *root,
2029 struct sa_defrag_extent_backref *backref)
2031 struct rb_node **p = &root->rb_node;
2032 struct rb_node *parent = NULL;
2033 struct sa_defrag_extent_backref *entry;
2038 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2040 ret = backref_comp(backref, entry);
2044 p = &(*p)->rb_right;
2047 rb_link_node(&backref->node, parent, p);
2048 rb_insert_color(&backref->node, root);
2052 * Note the backref might has changed, and in this case we just return 0.
2054 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2057 struct btrfs_file_extent_item *extent;
2058 struct btrfs_fs_info *fs_info;
2059 struct old_sa_defrag_extent *old = ctx;
2060 struct new_sa_defrag_extent *new = old->new;
2061 struct btrfs_path *path = new->path;
2062 struct btrfs_key key;
2063 struct btrfs_root *root;
2064 struct sa_defrag_extent_backref *backref;
2065 struct extent_buffer *leaf;
2066 struct inode *inode = new->inode;
2072 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2073 inum == btrfs_ino(inode))
2076 key.objectid = root_id;
2077 key.type = BTRFS_ROOT_ITEM_KEY;
2078 key.offset = (u64)-1;
2080 fs_info = BTRFS_I(inode)->root->fs_info;
2081 root = btrfs_read_fs_root_no_name(fs_info, &key);
2083 if (PTR_ERR(root) == -ENOENT)
2086 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2087 inum, offset, root_id);
2088 return PTR_ERR(root);
2091 key.objectid = inum;
2092 key.type = BTRFS_EXTENT_DATA_KEY;
2093 if (offset > (u64)-1 << 32)
2096 key.offset = offset;
2098 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2107 leaf = path->nodes[0];
2108 slot = path->slots[0];
2110 if (slot >= btrfs_header_nritems(leaf)) {
2111 ret = btrfs_next_leaf(root, path);
2114 } else if (ret > 0) {
2123 btrfs_item_key_to_cpu(leaf, &key, slot);
2125 if (key.objectid > inum)
2128 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2131 extent = btrfs_item_ptr(leaf, slot,
2132 struct btrfs_file_extent_item);
2134 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2137 extent_offset = btrfs_file_extent_offset(leaf, extent);
2138 if (key.offset - extent_offset != offset)
2141 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2142 if (extent_offset >= old->extent_offset + old->offset +
2143 old->len || extent_offset + num_bytes <=
2144 old->extent_offset + old->offset)
2150 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2156 backref->root_id = root_id;
2157 backref->inum = inum;
2158 backref->file_pos = offset + extent_offset;
2159 backref->num_bytes = num_bytes;
2160 backref->extent_offset = extent_offset;
2161 backref->generation = btrfs_file_extent_generation(leaf, extent);
2163 backref_insert(&new->root, backref);
2166 btrfs_release_path(path);
2171 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2172 struct new_sa_defrag_extent *new)
2174 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2175 struct old_sa_defrag_extent *old, *tmp;
2180 list_for_each_entry_safe(old, tmp, &new->head, list) {
2181 ret = iterate_inodes_from_logical(old->bytenr, fs_info,
2182 path, record_one_backref,
2184 BUG_ON(ret < 0 && ret != -ENOENT);
2186 /* no backref to be processed for this extent */
2188 list_del(&old->list);
2193 if (list_empty(&new->head))
2199 static int relink_is_mergable(struct extent_buffer *leaf,
2200 struct btrfs_file_extent_item *fi,
2203 if (btrfs_file_extent_disk_bytenr(leaf, fi) != disk_bytenr)
2206 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2209 if (btrfs_file_extent_compression(leaf, fi) ||
2210 btrfs_file_extent_encryption(leaf, fi) ||
2211 btrfs_file_extent_other_encoding(leaf, fi))
2218 * Note the backref might has changed, and in this case we just return 0.
2220 static noinline int relink_extent_backref(struct btrfs_path *path,
2221 struct sa_defrag_extent_backref *prev,
2222 struct sa_defrag_extent_backref *backref)
2224 struct btrfs_file_extent_item *extent;
2225 struct btrfs_file_extent_item *item;
2226 struct btrfs_ordered_extent *ordered;
2227 struct btrfs_trans_handle *trans;
2228 struct btrfs_fs_info *fs_info;
2229 struct btrfs_root *root;
2230 struct btrfs_key key;
2231 struct extent_buffer *leaf;
2232 struct old_sa_defrag_extent *old = backref->old;
2233 struct new_sa_defrag_extent *new = old->new;
2234 struct inode *src_inode = new->inode;
2235 struct inode *inode;
2236 struct extent_state *cached = NULL;
2245 if (prev && prev->root_id == backref->root_id &&
2246 prev->inum == backref->inum &&
2247 prev->file_pos + prev->num_bytes == backref->file_pos)
2250 /* step 1: get root */
2251 key.objectid = backref->root_id;
2252 key.type = BTRFS_ROOT_ITEM_KEY;
2253 key.offset = (u64)-1;
2255 fs_info = BTRFS_I(src_inode)->root->fs_info;
2256 index = srcu_read_lock(&fs_info->subvol_srcu);
2258 root = btrfs_read_fs_root_no_name(fs_info, &key);
2260 srcu_read_unlock(&fs_info->subvol_srcu, index);
2261 if (PTR_ERR(root) == -ENOENT)
2263 return PTR_ERR(root);
2266 /* step 2: get inode */
2267 key.objectid = backref->inum;
2268 key.type = BTRFS_INODE_ITEM_KEY;
2271 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2272 if (IS_ERR(inode)) {
2273 srcu_read_unlock(&fs_info->subvol_srcu, index);
2277 srcu_read_unlock(&fs_info->subvol_srcu, index);
2279 /* step 3: relink backref */
2280 lock_start = backref->file_pos;
2281 lock_end = backref->file_pos + backref->num_bytes - 1;
2282 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2285 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2287 btrfs_put_ordered_extent(ordered);
2291 trans = btrfs_join_transaction(root);
2292 if (IS_ERR(trans)) {
2293 ret = PTR_ERR(trans);
2297 key.objectid = backref->inum;
2298 key.type = BTRFS_EXTENT_DATA_KEY;
2299 key.offset = backref->file_pos;
2301 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2304 } else if (ret > 0) {
2309 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2310 struct btrfs_file_extent_item);
2312 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2313 backref->generation)
2316 btrfs_release_path(path);
2318 start = backref->file_pos;
2319 if (backref->extent_offset < old->extent_offset + old->offset)
2320 start += old->extent_offset + old->offset -
2321 backref->extent_offset;
2323 len = min(backref->extent_offset + backref->num_bytes,
2324 old->extent_offset + old->offset + old->len);
2325 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2327 ret = btrfs_drop_extents(trans, root, inode, start,
2332 key.objectid = btrfs_ino(inode);
2333 key.type = BTRFS_EXTENT_DATA_KEY;
2336 path->leave_spinning = 1;
2338 struct btrfs_file_extent_item *fi;
2340 struct btrfs_key found_key;
2342 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
2347 leaf = path->nodes[0];
2348 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2350 fi = btrfs_item_ptr(leaf, path->slots[0],
2351 struct btrfs_file_extent_item);
2352 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2354 if (relink_is_mergable(leaf, fi, new->bytenr) &&
2355 extent_len + found_key.offset == start) {
2356 btrfs_set_file_extent_num_bytes(leaf, fi,
2358 btrfs_mark_buffer_dirty(leaf);
2359 inode_add_bytes(inode, len);
2365 btrfs_release_path(path);
2370 ret = btrfs_insert_empty_item(trans, root, path, &key,
2373 btrfs_abort_transaction(trans, root, ret);
2377 leaf = path->nodes[0];
2378 item = btrfs_item_ptr(leaf, path->slots[0],
2379 struct btrfs_file_extent_item);
2380 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2381 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2382 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2383 btrfs_set_file_extent_num_bytes(leaf, item, len);
2384 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2385 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2386 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2387 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2388 btrfs_set_file_extent_encryption(leaf, item, 0);
2389 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2391 btrfs_mark_buffer_dirty(leaf);
2392 inode_add_bytes(inode, len);
2393 btrfs_release_path(path);
2395 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2397 backref->root_id, backref->inum,
2398 new->file_pos, 0); /* start - extent_offset */
2400 btrfs_abort_transaction(trans, root, ret);
2406 btrfs_release_path(path);
2407 path->leave_spinning = 0;
2408 btrfs_end_transaction(trans, root);
2410 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2416 static void relink_file_extents(struct new_sa_defrag_extent *new)
2418 struct btrfs_path *path;
2419 struct old_sa_defrag_extent *old, *tmp;
2420 struct sa_defrag_extent_backref *backref;
2421 struct sa_defrag_extent_backref *prev = NULL;
2422 struct inode *inode;
2423 struct btrfs_root *root;
2424 struct rb_node *node;
2428 root = BTRFS_I(inode)->root;
2430 path = btrfs_alloc_path();
2434 if (!record_extent_backrefs(path, new)) {
2435 btrfs_free_path(path);
2438 btrfs_release_path(path);
2441 node = rb_first(&new->root);
2444 rb_erase(node, &new->root);
2446 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2448 ret = relink_extent_backref(path, prev, backref);
2461 btrfs_free_path(path);
2463 list_for_each_entry_safe(old, tmp, &new->head, list) {
2464 list_del(&old->list);
2468 atomic_dec(&root->fs_info->defrag_running);
2469 wake_up(&root->fs_info->transaction_wait);
2474 static struct new_sa_defrag_extent *
2475 record_old_file_extents(struct inode *inode,
2476 struct btrfs_ordered_extent *ordered)
2478 struct btrfs_root *root = BTRFS_I(inode)->root;
2479 struct btrfs_path *path;
2480 struct btrfs_key key;
2481 struct old_sa_defrag_extent *old, *tmp;
2482 struct new_sa_defrag_extent *new;
2485 new = kmalloc(sizeof(*new), GFP_NOFS);
2490 new->file_pos = ordered->file_offset;
2491 new->len = ordered->len;
2492 new->bytenr = ordered->start;
2493 new->disk_len = ordered->disk_len;
2494 new->compress_type = ordered->compress_type;
2495 new->root = RB_ROOT;
2496 INIT_LIST_HEAD(&new->head);
2498 path = btrfs_alloc_path();
2502 key.objectid = btrfs_ino(inode);
2503 key.type = BTRFS_EXTENT_DATA_KEY;
2504 key.offset = new->file_pos;
2506 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2509 if (ret > 0 && path->slots[0] > 0)
2512 /* find out all the old extents for the file range */
2514 struct btrfs_file_extent_item *extent;
2515 struct extent_buffer *l;
2524 slot = path->slots[0];
2526 if (slot >= btrfs_header_nritems(l)) {
2527 ret = btrfs_next_leaf(root, path);
2535 btrfs_item_key_to_cpu(l, &key, slot);
2537 if (key.objectid != btrfs_ino(inode))
2539 if (key.type != BTRFS_EXTENT_DATA_KEY)
2541 if (key.offset >= new->file_pos + new->len)
2544 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2546 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2547 if (key.offset + num_bytes < new->file_pos)
2550 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2554 extent_offset = btrfs_file_extent_offset(l, extent);
2556 old = kmalloc(sizeof(*old), GFP_NOFS);
2560 offset = max(new->file_pos, key.offset);
2561 end = min(new->file_pos + new->len, key.offset + num_bytes);
2563 old->bytenr = disk_bytenr;
2564 old->extent_offset = extent_offset;
2565 old->offset = offset - key.offset;
2566 old->len = end - offset;
2569 list_add_tail(&old->list, &new->head);
2575 btrfs_free_path(path);
2576 atomic_inc(&root->fs_info->defrag_running);
2581 list_for_each_entry_safe(old, tmp, &new->head, list) {
2582 list_del(&old->list);
2586 btrfs_free_path(path);
2593 * helper function for btrfs_finish_ordered_io, this
2594 * just reads in some of the csum leaves to prime them into ram
2595 * before we start the transaction. It limits the amount of btree
2596 * reads required while inside the transaction.
2598 /* as ordered data IO finishes, this gets called so we can finish
2599 * an ordered extent if the range of bytes in the file it covers are
2602 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2604 struct inode *inode = ordered_extent->inode;
2605 struct btrfs_root *root = BTRFS_I(inode)->root;
2606 struct btrfs_trans_handle *trans = NULL;
2607 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2608 struct extent_state *cached_state = NULL;
2609 struct new_sa_defrag_extent *new = NULL;
2610 int compress_type = 0;
2614 nolock = btrfs_is_free_space_inode(inode);
2616 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2621 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2622 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2623 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2625 trans = btrfs_join_transaction_nolock(root);
2627 trans = btrfs_join_transaction(root);
2628 if (IS_ERR(trans)) {
2629 ret = PTR_ERR(trans);
2633 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2634 ret = btrfs_update_inode_fallback(trans, root, inode);
2635 if (ret) /* -ENOMEM or corruption */
2636 btrfs_abort_transaction(trans, root, ret);
2640 lock_extent_bits(io_tree, ordered_extent->file_offset,
2641 ordered_extent->file_offset + ordered_extent->len - 1,
2644 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2645 ordered_extent->file_offset + ordered_extent->len - 1,
2646 EXTENT_DEFRAG, 1, cached_state);
2648 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2649 if (last_snapshot >= BTRFS_I(inode)->generation)
2650 /* the inode is shared */
2651 new = record_old_file_extents(inode, ordered_extent);
2653 clear_extent_bit(io_tree, ordered_extent->file_offset,
2654 ordered_extent->file_offset + ordered_extent->len - 1,
2655 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2659 trans = btrfs_join_transaction_nolock(root);
2661 trans = btrfs_join_transaction(root);
2662 if (IS_ERR(trans)) {
2663 ret = PTR_ERR(trans);
2667 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2669 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2670 compress_type = ordered_extent->compress_type;
2671 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2672 BUG_ON(compress_type);
2673 ret = btrfs_mark_extent_written(trans, inode,
2674 ordered_extent->file_offset,
2675 ordered_extent->file_offset +
2676 ordered_extent->len);
2678 BUG_ON(root == root->fs_info->tree_root);
2679 ret = insert_reserved_file_extent(trans, inode,
2680 ordered_extent->file_offset,
2681 ordered_extent->start,
2682 ordered_extent->disk_len,
2683 ordered_extent->len,
2684 ordered_extent->len,
2685 compress_type, 0, 0,
2686 BTRFS_FILE_EXTENT_REG);
2688 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2689 ordered_extent->file_offset, ordered_extent->len,
2692 btrfs_abort_transaction(trans, root, ret);
2696 add_pending_csums(trans, inode, ordered_extent->file_offset,
2697 &ordered_extent->list);
2699 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2700 ret = btrfs_update_inode_fallback(trans, root, inode);
2701 if (ret) { /* -ENOMEM or corruption */
2702 btrfs_abort_transaction(trans, root, ret);
2707 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2708 ordered_extent->file_offset +
2709 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2711 if (root != root->fs_info->tree_root)
2712 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2714 btrfs_end_transaction(trans, root);
2717 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
2718 ordered_extent->file_offset +
2719 ordered_extent->len - 1, NULL, GFP_NOFS);
2722 * If the ordered extent had an IOERR or something else went
2723 * wrong we need to return the space for this ordered extent
2724 * back to the allocator.
2726 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2727 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2728 btrfs_free_reserved_extent(root, ordered_extent->start,
2729 ordered_extent->disk_len);
2734 * This needs to be done to make sure anybody waiting knows we are done
2735 * updating everything for this ordered extent.
2737 btrfs_remove_ordered_extent(inode, ordered_extent);
2739 /* for snapshot-aware defrag */
2741 relink_file_extents(new);
2744 btrfs_put_ordered_extent(ordered_extent);
2745 /* once for the tree */
2746 btrfs_put_ordered_extent(ordered_extent);
2751 static void finish_ordered_fn(struct btrfs_work *work)
2753 struct btrfs_ordered_extent *ordered_extent;
2754 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2755 btrfs_finish_ordered_io(ordered_extent);
2758 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2759 struct extent_state *state, int uptodate)
2761 struct inode *inode = page->mapping->host;
2762 struct btrfs_root *root = BTRFS_I(inode)->root;
2763 struct btrfs_ordered_extent *ordered_extent = NULL;
2764 struct btrfs_workers *workers;
2766 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2768 ClearPagePrivate2(page);
2769 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2770 end - start + 1, uptodate))
2773 ordered_extent->work.func = finish_ordered_fn;
2774 ordered_extent->work.flags = 0;
2776 if (btrfs_is_free_space_inode(inode))
2777 workers = &root->fs_info->endio_freespace_worker;
2779 workers = &root->fs_info->endio_write_workers;
2780 btrfs_queue_worker(workers, &ordered_extent->work);
2786 * when reads are done, we need to check csums to verify the data is correct
2787 * if there's a match, we allow the bio to finish. If not, the code in
2788 * extent_io.c will try to find good copies for us.
2790 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2791 struct extent_state *state, int mirror)
2793 size_t offset = start - page_offset(page);
2794 struct inode *inode = page->mapping->host;
2795 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2797 u64 private = ~(u32)0;
2799 struct btrfs_root *root = BTRFS_I(inode)->root;
2801 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2802 DEFAULT_RATELIMIT_BURST);
2804 if (PageChecked(page)) {
2805 ClearPageChecked(page);
2809 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2812 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2813 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2814 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2819 if (state && state->start == start) {
2820 private = state->private;
2823 ret = get_state_private(io_tree, start, &private);
2825 kaddr = kmap_atomic(page);
2829 csum = btrfs_csum_data(kaddr + offset, csum, end - start + 1);
2830 btrfs_csum_final(csum, (char *)&csum);
2831 if (csum != private)
2834 kunmap_atomic(kaddr);
2839 if (__ratelimit(&_rs))
2840 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u private %llu",
2841 (unsigned long long)btrfs_ino(page->mapping->host),
2842 (unsigned long long)start, csum,
2843 (unsigned long long)private);
2844 memset(kaddr + offset, 1, end - start + 1);
2845 flush_dcache_page(page);
2846 kunmap_atomic(kaddr);
2852 struct delayed_iput {
2853 struct list_head list;
2854 struct inode *inode;
2857 /* JDM: If this is fs-wide, why can't we add a pointer to
2858 * btrfs_inode instead and avoid the allocation? */
2859 void btrfs_add_delayed_iput(struct inode *inode)
2861 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2862 struct delayed_iput *delayed;
2864 if (atomic_add_unless(&inode->i_count, -1, 1))
2867 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2868 delayed->inode = inode;
2870 spin_lock(&fs_info->delayed_iput_lock);
2871 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2872 spin_unlock(&fs_info->delayed_iput_lock);
2875 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2878 struct btrfs_fs_info *fs_info = root->fs_info;
2879 struct delayed_iput *delayed;
2882 spin_lock(&fs_info->delayed_iput_lock);
2883 empty = list_empty(&fs_info->delayed_iputs);
2884 spin_unlock(&fs_info->delayed_iput_lock);
2888 spin_lock(&fs_info->delayed_iput_lock);
2889 list_splice_init(&fs_info->delayed_iputs, &list);
2890 spin_unlock(&fs_info->delayed_iput_lock);
2892 while (!list_empty(&list)) {
2893 delayed = list_entry(list.next, struct delayed_iput, list);
2894 list_del(&delayed->list);
2895 iput(delayed->inode);
2901 * This is called in transaction commit time. If there are no orphan
2902 * files in the subvolume, it removes orphan item and frees block_rsv
2905 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2906 struct btrfs_root *root)
2908 struct btrfs_block_rsv *block_rsv;
2911 if (atomic_read(&root->orphan_inodes) ||
2912 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2915 spin_lock(&root->orphan_lock);
2916 if (atomic_read(&root->orphan_inodes)) {
2917 spin_unlock(&root->orphan_lock);
2921 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2922 spin_unlock(&root->orphan_lock);
2926 block_rsv = root->orphan_block_rsv;
2927 root->orphan_block_rsv = NULL;
2928 spin_unlock(&root->orphan_lock);
2930 if (root->orphan_item_inserted &&
2931 btrfs_root_refs(&root->root_item) > 0) {
2932 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2933 root->root_key.objectid);
2935 root->orphan_item_inserted = 0;
2939 WARN_ON(block_rsv->size > 0);
2940 btrfs_free_block_rsv(root, block_rsv);
2945 * This creates an orphan entry for the given inode in case something goes
2946 * wrong in the middle of an unlink/truncate.
2948 * NOTE: caller of this function should reserve 5 units of metadata for
2951 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2953 struct btrfs_root *root = BTRFS_I(inode)->root;
2954 struct btrfs_block_rsv *block_rsv = NULL;
2959 if (!root->orphan_block_rsv) {
2960 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2965 spin_lock(&root->orphan_lock);
2966 if (!root->orphan_block_rsv) {
2967 root->orphan_block_rsv = block_rsv;
2968 } else if (block_rsv) {
2969 btrfs_free_block_rsv(root, block_rsv);
2973 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2974 &BTRFS_I(inode)->runtime_flags)) {
2977 * For proper ENOSPC handling, we should do orphan
2978 * cleanup when mounting. But this introduces backward
2979 * compatibility issue.
2981 if (!xchg(&root->orphan_item_inserted, 1))
2987 atomic_inc(&root->orphan_inodes);
2990 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2991 &BTRFS_I(inode)->runtime_flags))
2993 spin_unlock(&root->orphan_lock);
2995 /* grab metadata reservation from transaction handle */
2997 ret = btrfs_orphan_reserve_metadata(trans, inode);
2998 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3001 /* insert an orphan item to track this unlinked/truncated file */
3003 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3004 if (ret && ret != -EEXIST) {
3005 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3006 &BTRFS_I(inode)->runtime_flags);
3007 btrfs_abort_transaction(trans, root, ret);
3013 /* insert an orphan item to track subvolume contains orphan files */
3015 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3016 root->root_key.objectid);
3017 if (ret && ret != -EEXIST) {
3018 btrfs_abort_transaction(trans, root, ret);
3026 * We have done the truncate/delete so we can go ahead and remove the orphan
3027 * item for this particular inode.
3029 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3030 struct inode *inode)
3032 struct btrfs_root *root = BTRFS_I(inode)->root;
3033 int delete_item = 0;
3034 int release_rsv = 0;
3037 spin_lock(&root->orphan_lock);
3038 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3039 &BTRFS_I(inode)->runtime_flags))
3042 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3043 &BTRFS_I(inode)->runtime_flags))
3045 spin_unlock(&root->orphan_lock);
3047 if (trans && delete_item) {
3048 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
3049 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3053 btrfs_orphan_release_metadata(inode);
3054 atomic_dec(&root->orphan_inodes);
3061 * this cleans up any orphans that may be left on the list from the last use
3064 int btrfs_orphan_cleanup(struct btrfs_root *root)
3066 struct btrfs_path *path;
3067 struct extent_buffer *leaf;
3068 struct btrfs_key key, found_key;
3069 struct btrfs_trans_handle *trans;
3070 struct inode *inode;
3071 u64 last_objectid = 0;
3072 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3074 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3077 path = btrfs_alloc_path();
3084 key.objectid = BTRFS_ORPHAN_OBJECTID;
3085 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3086 key.offset = (u64)-1;
3089 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3094 * if ret == 0 means we found what we were searching for, which
3095 * is weird, but possible, so only screw with path if we didn't
3096 * find the key and see if we have stuff that matches
3100 if (path->slots[0] == 0)
3105 /* pull out the item */
3106 leaf = path->nodes[0];
3107 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3109 /* make sure the item matches what we want */
3110 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3112 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3115 /* release the path since we're done with it */
3116 btrfs_release_path(path);
3119 * this is where we are basically btrfs_lookup, without the
3120 * crossing root thing. we store the inode number in the
3121 * offset of the orphan item.
3124 if (found_key.offset == last_objectid) {
3125 btrfs_err(root->fs_info,
3126 "Error removing orphan entry, stopping orphan cleanup");
3131 last_objectid = found_key.offset;
3133 found_key.objectid = found_key.offset;
3134 found_key.type = BTRFS_INODE_ITEM_KEY;
3135 found_key.offset = 0;
3136 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3137 ret = PTR_RET(inode);
3138 if (ret && ret != -ESTALE)
3141 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3142 struct btrfs_root *dead_root;
3143 struct btrfs_fs_info *fs_info = root->fs_info;
3144 int is_dead_root = 0;
3147 * this is an orphan in the tree root. Currently these
3148 * could come from 2 sources:
3149 * a) a snapshot deletion in progress
3150 * b) a free space cache inode
3151 * We need to distinguish those two, as the snapshot
3152 * orphan must not get deleted.
3153 * find_dead_roots already ran before us, so if this
3154 * is a snapshot deletion, we should find the root
3155 * in the dead_roots list
3157 spin_lock(&fs_info->trans_lock);
3158 list_for_each_entry(dead_root, &fs_info->dead_roots,
3160 if (dead_root->root_key.objectid ==
3161 found_key.objectid) {
3166 spin_unlock(&fs_info->trans_lock);
3168 /* prevent this orphan from being found again */
3169 key.offset = found_key.objectid - 1;
3174 * Inode is already gone but the orphan item is still there,
3175 * kill the orphan item.
3177 if (ret == -ESTALE) {
3178 trans = btrfs_start_transaction(root, 1);
3179 if (IS_ERR(trans)) {
3180 ret = PTR_ERR(trans);
3183 btrfs_debug(root->fs_info, "auto deleting %Lu",
3184 found_key.objectid);
3185 ret = btrfs_del_orphan_item(trans, root,
3186 found_key.objectid);
3187 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3188 btrfs_end_transaction(trans, root);
3193 * add this inode to the orphan list so btrfs_orphan_del does
3194 * the proper thing when we hit it
3196 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3197 &BTRFS_I(inode)->runtime_flags);
3198 atomic_inc(&root->orphan_inodes);
3200 /* if we have links, this was a truncate, lets do that */
3201 if (inode->i_nlink) {
3202 if (!S_ISREG(inode->i_mode)) {
3209 /* 1 for the orphan item deletion. */
3210 trans = btrfs_start_transaction(root, 1);
3211 if (IS_ERR(trans)) {
3212 ret = PTR_ERR(trans);
3215 ret = btrfs_orphan_add(trans, inode);
3216 btrfs_end_transaction(trans, root);
3220 ret = btrfs_truncate(inode);
3222 btrfs_orphan_del(NULL, inode);
3227 /* this will do delete_inode and everything for us */
3232 /* release the path since we're done with it */
3233 btrfs_release_path(path);
3235 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3237 if (root->orphan_block_rsv)
3238 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3241 if (root->orphan_block_rsv || root->orphan_item_inserted) {
3242 trans = btrfs_join_transaction(root);
3244 btrfs_end_transaction(trans, root);
3248 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3250 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3254 btrfs_crit(root->fs_info,
3255 "could not do orphan cleanup %d", ret);
3256 btrfs_free_path(path);
3261 * very simple check to peek ahead in the leaf looking for xattrs. If we
3262 * don't find any xattrs, we know there can't be any acls.
3264 * slot is the slot the inode is in, objectid is the objectid of the inode
3266 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3267 int slot, u64 objectid)
3269 u32 nritems = btrfs_header_nritems(leaf);
3270 struct btrfs_key found_key;
3274 while (slot < nritems) {
3275 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3277 /* we found a different objectid, there must not be acls */
3278 if (found_key.objectid != objectid)
3281 /* we found an xattr, assume we've got an acl */
3282 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
3286 * we found a key greater than an xattr key, there can't
3287 * be any acls later on
3289 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3296 * it goes inode, inode backrefs, xattrs, extents,
3297 * so if there are a ton of hard links to an inode there can
3298 * be a lot of backrefs. Don't waste time searching too hard,
3299 * this is just an optimization
3304 /* we hit the end of the leaf before we found an xattr or
3305 * something larger than an xattr. We have to assume the inode
3312 * read an inode from the btree into the in-memory inode
3314 static void btrfs_read_locked_inode(struct inode *inode)
3316 struct btrfs_path *path;
3317 struct extent_buffer *leaf;
3318 struct btrfs_inode_item *inode_item;
3319 struct btrfs_timespec *tspec;
3320 struct btrfs_root *root = BTRFS_I(inode)->root;
3321 struct btrfs_key location;
3325 bool filled = false;
3327 ret = btrfs_fill_inode(inode, &rdev);
3331 path = btrfs_alloc_path();
3335 path->leave_spinning = 1;
3336 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3338 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3342 leaf = path->nodes[0];
3347 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3348 struct btrfs_inode_item);
3349 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3350 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3351 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3352 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3353 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3355 tspec = btrfs_inode_atime(inode_item);
3356 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3357 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3359 tspec = btrfs_inode_mtime(inode_item);
3360 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3361 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3363 tspec = btrfs_inode_ctime(inode_item);
3364 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3365 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3367 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3368 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3369 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3372 * If we were modified in the current generation and evicted from memory
3373 * and then re-read we need to do a full sync since we don't have any
3374 * idea about which extents were modified before we were evicted from
3377 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3378 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3379 &BTRFS_I(inode)->runtime_flags);
3381 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3382 inode->i_generation = BTRFS_I(inode)->generation;
3384 rdev = btrfs_inode_rdev(leaf, inode_item);
3386 BTRFS_I(inode)->index_cnt = (u64)-1;
3387 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3390 * try to precache a NULL acl entry for files that don't have
3391 * any xattrs or acls
3393 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3396 cache_no_acl(inode);
3398 btrfs_free_path(path);
3400 switch (inode->i_mode & S_IFMT) {
3402 inode->i_mapping->a_ops = &btrfs_aops;
3403 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3404 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3405 inode->i_fop = &btrfs_file_operations;
3406 inode->i_op = &btrfs_file_inode_operations;
3409 inode->i_fop = &btrfs_dir_file_operations;
3410 if (root == root->fs_info->tree_root)
3411 inode->i_op = &btrfs_dir_ro_inode_operations;
3413 inode->i_op = &btrfs_dir_inode_operations;
3416 inode->i_op = &btrfs_symlink_inode_operations;
3417 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3418 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3421 inode->i_op = &btrfs_special_inode_operations;
3422 init_special_inode(inode, inode->i_mode, rdev);
3426 btrfs_update_iflags(inode);
3430 btrfs_free_path(path);
3431 make_bad_inode(inode);
3435 * given a leaf and an inode, copy the inode fields into the leaf
3437 static void fill_inode_item(struct btrfs_trans_handle *trans,
3438 struct extent_buffer *leaf,
3439 struct btrfs_inode_item *item,
3440 struct inode *inode)
3442 struct btrfs_map_token token;
3444 btrfs_init_map_token(&token);
3446 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3447 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3448 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3450 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3451 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3453 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3454 inode->i_atime.tv_sec, &token);
3455 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3456 inode->i_atime.tv_nsec, &token);
3458 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3459 inode->i_mtime.tv_sec, &token);
3460 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3461 inode->i_mtime.tv_nsec, &token);
3463 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3464 inode->i_ctime.tv_sec, &token);
3465 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3466 inode->i_ctime.tv_nsec, &token);
3468 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3470 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3472 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3473 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3474 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3475 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3476 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3480 * copy everything in the in-memory inode into the btree.
3482 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3483 struct btrfs_root *root, struct inode *inode)
3485 struct btrfs_inode_item *inode_item;
3486 struct btrfs_path *path;
3487 struct extent_buffer *leaf;
3490 path = btrfs_alloc_path();
3494 path->leave_spinning = 1;
3495 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3503 btrfs_unlock_up_safe(path, 1);
3504 leaf = path->nodes[0];
3505 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3506 struct btrfs_inode_item);
3508 fill_inode_item(trans, leaf, inode_item, inode);
3509 btrfs_mark_buffer_dirty(leaf);
3510 btrfs_set_inode_last_trans(trans, inode);
3513 btrfs_free_path(path);
3518 * copy everything in the in-memory inode into the btree.
3520 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3521 struct btrfs_root *root, struct inode *inode)
3526 * If the inode is a free space inode, we can deadlock during commit
3527 * if we put it into the delayed code.
3529 * The data relocation inode should also be directly updated
3532 if (!btrfs_is_free_space_inode(inode)
3533 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3534 btrfs_update_root_times(trans, root);
3536 ret = btrfs_delayed_update_inode(trans, root, inode);
3538 btrfs_set_inode_last_trans(trans, inode);
3542 return btrfs_update_inode_item(trans, root, inode);
3545 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3546 struct btrfs_root *root,
3547 struct inode *inode)
3551 ret = btrfs_update_inode(trans, root, inode);
3553 return btrfs_update_inode_item(trans, root, inode);
3558 * unlink helper that gets used here in inode.c and in the tree logging
3559 * recovery code. It remove a link in a directory with a given name, and
3560 * also drops the back refs in the inode to the directory
3562 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3563 struct btrfs_root *root,
3564 struct inode *dir, struct inode *inode,
3565 const char *name, int name_len)
3567 struct btrfs_path *path;
3569 struct extent_buffer *leaf;
3570 struct btrfs_dir_item *di;
3571 struct btrfs_key key;
3573 u64 ino = btrfs_ino(inode);
3574 u64 dir_ino = btrfs_ino(dir);
3576 path = btrfs_alloc_path();
3582 path->leave_spinning = 1;
3583 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3584 name, name_len, -1);
3593 leaf = path->nodes[0];
3594 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3595 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3598 btrfs_release_path(path);
3600 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3603 btrfs_info(root->fs_info,
3604 "failed to delete reference to %.*s, inode %llu parent %llu",
3606 (unsigned long long)ino, (unsigned long long)dir_ino);
3607 btrfs_abort_transaction(trans, root, ret);
3611 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3613 btrfs_abort_transaction(trans, root, ret);
3617 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3619 if (ret != 0 && ret != -ENOENT) {
3620 btrfs_abort_transaction(trans, root, ret);
3624 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3629 btrfs_abort_transaction(trans, root, ret);
3631 btrfs_free_path(path);
3635 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3636 inode_inc_iversion(inode);
3637 inode_inc_iversion(dir);
3638 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3639 ret = btrfs_update_inode(trans, root, dir);
3644 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3645 struct btrfs_root *root,
3646 struct inode *dir, struct inode *inode,
3647 const char *name, int name_len)
3650 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3652 btrfs_drop_nlink(inode);
3653 ret = btrfs_update_inode(trans, root, inode);
3659 /* helper to check if there is any shared block in the path */
3660 static int check_path_shared(struct btrfs_root *root,
3661 struct btrfs_path *path)
3663 struct extent_buffer *eb;
3667 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
3670 if (!path->nodes[level])
3672 eb = path->nodes[level];
3673 if (!btrfs_block_can_be_shared(root, eb))
3675 ret = btrfs_lookup_extent_info(NULL, root, eb->start, level, 1,
3684 * helper to start transaction for unlink and rmdir.
3686 * unlink and rmdir are special in btrfs, they do not always free space.
3687 * so in enospc case, we should make sure they will free space before
3688 * allowing them to use the global metadata reservation.
3690 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
3691 struct dentry *dentry)
3693 struct btrfs_trans_handle *trans;
3694 struct btrfs_root *root = BTRFS_I(dir)->root;
3695 struct btrfs_path *path;
3696 struct btrfs_dir_item *di;
3697 struct inode *inode = dentry->d_inode;
3702 u64 ino = btrfs_ino(inode);
3703 u64 dir_ino = btrfs_ino(dir);
3706 * 1 for the possible orphan item
3707 * 1 for the dir item
3708 * 1 for the dir index
3709 * 1 for the inode ref
3712 trans = btrfs_start_transaction(root, 5);
3713 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3716 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
3717 return ERR_PTR(-ENOSPC);
3719 /* check if there is someone else holds reference */
3720 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
3721 return ERR_PTR(-ENOSPC);
3723 if (atomic_read(&inode->i_count) > 2)
3724 return ERR_PTR(-ENOSPC);
3726 if (xchg(&root->fs_info->enospc_unlink, 1))
3727 return ERR_PTR(-ENOSPC);
3729 path = btrfs_alloc_path();
3731 root->fs_info->enospc_unlink = 0;
3732 return ERR_PTR(-ENOMEM);
3735 /* 1 for the orphan item */
3736 trans = btrfs_start_transaction(root, 1);
3737 if (IS_ERR(trans)) {
3738 btrfs_free_path(path);
3739 root->fs_info->enospc_unlink = 0;
3743 path->skip_locking = 1;
3744 path->search_commit_root = 1;
3746 ret = btrfs_lookup_inode(trans, root, path,
3747 &BTRFS_I(dir)->location, 0);
3753 if (check_path_shared(root, path))
3758 btrfs_release_path(path);
3760 ret = btrfs_lookup_inode(trans, root, path,
3761 &BTRFS_I(inode)->location, 0);
3767 if (check_path_shared(root, path))
3772 btrfs_release_path(path);
3774 if (ret == 0 && S_ISREG(inode->i_mode)) {
3775 ret = btrfs_lookup_file_extent(trans, root, path,
3781 BUG_ON(ret == 0); /* Corruption */
3782 if (check_path_shared(root, path))
3784 btrfs_release_path(path);
3792 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3793 dentry->d_name.name, dentry->d_name.len, 0);
3799 if (check_path_shared(root, path))
3805 btrfs_release_path(path);
3807 ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3808 dentry->d_name.len, ino, dir_ino, 0,
3815 if (check_path_shared(root, path))
3818 btrfs_release_path(path);
3821 * This is a commit root search, if we can lookup inode item and other
3822 * relative items in the commit root, it means the transaction of
3823 * dir/file creation has been committed, and the dir index item that we
3824 * delay to insert has also been inserted into the commit root. So
3825 * we needn't worry about the delayed insertion of the dir index item
3828 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3829 dentry->d_name.name, dentry->d_name.len, 0);
3834 BUG_ON(ret == -ENOENT);
3835 if (check_path_shared(root, path))
3840 btrfs_free_path(path);
3841 /* Migrate the orphan reservation over */
3843 err = btrfs_block_rsv_migrate(trans->block_rsv,
3844 &root->fs_info->global_block_rsv,
3845 trans->bytes_reserved);
3848 btrfs_end_transaction(trans, root);
3849 root->fs_info->enospc_unlink = 0;
3850 return ERR_PTR(err);
3853 trans->block_rsv = &root->fs_info->global_block_rsv;
3857 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3858 struct btrfs_root *root)
3860 if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3861 btrfs_block_rsv_release(root, trans->block_rsv,
3862 trans->bytes_reserved);
3863 trans->block_rsv = &root->fs_info->trans_block_rsv;
3864 BUG_ON(!root->fs_info->enospc_unlink);
3865 root->fs_info->enospc_unlink = 0;
3867 btrfs_end_transaction(trans, root);
3870 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3872 struct btrfs_root *root = BTRFS_I(dir)->root;
3873 struct btrfs_trans_handle *trans;
3874 struct inode *inode = dentry->d_inode;
3877 trans = __unlink_start_trans(dir, dentry);
3879 return PTR_ERR(trans);
3881 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3883 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3884 dentry->d_name.name, dentry->d_name.len);
3888 if (inode->i_nlink == 0) {
3889 ret = btrfs_orphan_add(trans, inode);
3895 __unlink_end_trans(trans, root);
3896 btrfs_btree_balance_dirty(root);
3900 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3901 struct btrfs_root *root,
3902 struct inode *dir, u64 objectid,
3903 const char *name, int name_len)
3905 struct btrfs_path *path;
3906 struct extent_buffer *leaf;
3907 struct btrfs_dir_item *di;
3908 struct btrfs_key key;
3911 u64 dir_ino = btrfs_ino(dir);
3913 path = btrfs_alloc_path();
3917 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3918 name, name_len, -1);
3919 if (IS_ERR_OR_NULL(di)) {
3927 leaf = path->nodes[0];
3928 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3929 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3930 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3932 btrfs_abort_transaction(trans, root, ret);
3935 btrfs_release_path(path);
3937 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3938 objectid, root->root_key.objectid,
3939 dir_ino, &index, name, name_len);
3941 if (ret != -ENOENT) {
3942 btrfs_abort_transaction(trans, root, ret);
3945 di = btrfs_search_dir_index_item(root, path, dir_ino,
3947 if (IS_ERR_OR_NULL(di)) {
3952 btrfs_abort_transaction(trans, root, ret);
3956 leaf = path->nodes[0];
3957 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3958 btrfs_release_path(path);
3961 btrfs_release_path(path);
3963 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3965 btrfs_abort_transaction(trans, root, ret);
3969 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3970 inode_inc_iversion(dir);
3971 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3972 ret = btrfs_update_inode_fallback(trans, root, dir);
3974 btrfs_abort_transaction(trans, root, ret);
3976 btrfs_free_path(path);
3980 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3982 struct inode *inode = dentry->d_inode;
3984 struct btrfs_root *root = BTRFS_I(dir)->root;
3985 struct btrfs_trans_handle *trans;
3987 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3989 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3992 trans = __unlink_start_trans(dir, dentry);
3994 return PTR_ERR(trans);
3996 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3997 err = btrfs_unlink_subvol(trans, root, dir,
3998 BTRFS_I(inode)->location.objectid,
3999 dentry->d_name.name,
4000 dentry->d_name.len);
4004 err = btrfs_orphan_add(trans, inode);
4008 /* now the directory is empty */
4009 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
4010 dentry->d_name.name, dentry->d_name.len);
4012 btrfs_i_size_write(inode, 0);
4014 __unlink_end_trans(trans, root);
4015 btrfs_btree_balance_dirty(root);
4021 * this can truncate away extent items, csum items and directory items.
4022 * It starts at a high offset and removes keys until it can't find
4023 * any higher than new_size
4025 * csum items that cross the new i_size are truncated to the new size
4028 * min_type is the minimum key type to truncate down to. If set to 0, this
4029 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4031 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4032 struct btrfs_root *root,
4033 struct inode *inode,
4034 u64 new_size, u32 min_type)
4036 struct btrfs_path *path;
4037 struct extent_buffer *leaf;
4038 struct btrfs_file_extent_item *fi;
4039 struct btrfs_key key;
4040 struct btrfs_key found_key;
4041 u64 extent_start = 0;
4042 u64 extent_num_bytes = 0;
4043 u64 extent_offset = 0;
4045 u32 found_type = (u8)-1;
4048 int pending_del_nr = 0;
4049 int pending_del_slot = 0;
4050 int extent_type = -1;
4053 u64 ino = btrfs_ino(inode);
4055 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4057 path = btrfs_alloc_path();
4063 * We want to drop from the next block forward in case this new size is
4064 * not block aligned since we will be keeping the last block of the
4065 * extent just the way it is.
4067 if (root->ref_cows || root == root->fs_info->tree_root)
4068 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4069 root->sectorsize), (u64)-1, 0);
4072 * This function is also used to drop the items in the log tree before
4073 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4074 * it is used to drop the loged items. So we shouldn't kill the delayed
4077 if (min_type == 0 && root == BTRFS_I(inode)->root)
4078 btrfs_kill_delayed_inode_items(inode);
4081 key.offset = (u64)-1;
4085 path->leave_spinning = 1;
4086 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4093 /* there are no items in the tree for us to truncate, we're
4096 if (path->slots[0] == 0)
4103 leaf = path->nodes[0];
4104 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4105 found_type = btrfs_key_type(&found_key);
4107 if (found_key.objectid != ino)
4110 if (found_type < min_type)
4113 item_end = found_key.offset;
4114 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4115 fi = btrfs_item_ptr(leaf, path->slots[0],
4116 struct btrfs_file_extent_item);
4117 extent_type = btrfs_file_extent_type(leaf, fi);
4118 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4120 btrfs_file_extent_num_bytes(leaf, fi);
4121 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4122 item_end += btrfs_file_extent_inline_len(leaf,
4127 if (found_type > min_type) {
4130 if (item_end < new_size)
4132 if (found_key.offset >= new_size)
4138 /* FIXME, shrink the extent if the ref count is only 1 */
4139 if (found_type != BTRFS_EXTENT_DATA_KEY)
4142 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4144 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4146 u64 orig_num_bytes =
4147 btrfs_file_extent_num_bytes(leaf, fi);
4148 extent_num_bytes = ALIGN(new_size -
4151 btrfs_set_file_extent_num_bytes(leaf, fi,
4153 num_dec = (orig_num_bytes -
4155 if (root->ref_cows && extent_start != 0)
4156 inode_sub_bytes(inode, num_dec);
4157 btrfs_mark_buffer_dirty(leaf);
4160 btrfs_file_extent_disk_num_bytes(leaf,
4162 extent_offset = found_key.offset -
4163 btrfs_file_extent_offset(leaf, fi);
4165 /* FIXME blocksize != 4096 */
4166 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4167 if (extent_start != 0) {
4170 inode_sub_bytes(inode, num_dec);
4173 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4175 * we can't truncate inline items that have had
4179 btrfs_file_extent_compression(leaf, fi) == 0 &&
4180 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4181 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4182 u32 size = new_size - found_key.offset;
4184 if (root->ref_cows) {
4185 inode_sub_bytes(inode, item_end + 1 -
4189 btrfs_file_extent_calc_inline_size(size);
4190 btrfs_truncate_item(root, path, size, 1);
4191 } else if (root->ref_cows) {
4192 inode_sub_bytes(inode, item_end + 1 -
4198 if (!pending_del_nr) {
4199 /* no pending yet, add ourselves */
4200 pending_del_slot = path->slots[0];
4202 } else if (pending_del_nr &&
4203 path->slots[0] + 1 == pending_del_slot) {
4204 /* hop on the pending chunk */
4206 pending_del_slot = path->slots[0];
4213 if (found_extent && (root->ref_cows ||
4214 root == root->fs_info->tree_root)) {
4215 btrfs_set_path_blocking(path);
4216 ret = btrfs_free_extent(trans, root, extent_start,
4217 extent_num_bytes, 0,
4218 btrfs_header_owner(leaf),
4219 ino, extent_offset, 0);
4223 if (found_type == BTRFS_INODE_ITEM_KEY)
4226 if (path->slots[0] == 0 ||
4227 path->slots[0] != pending_del_slot) {
4228 if (pending_del_nr) {
4229 ret = btrfs_del_items(trans, root, path,
4233 btrfs_abort_transaction(trans,
4239 btrfs_release_path(path);
4246 if (pending_del_nr) {
4247 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4250 btrfs_abort_transaction(trans, root, ret);
4253 btrfs_free_path(path);
4258 * btrfs_truncate_page - read, zero a chunk and write a page
4259 * @inode - inode that we're zeroing
4260 * @from - the offset to start zeroing
4261 * @len - the length to zero, 0 to zero the entire range respective to the
4263 * @front - zero up to the offset instead of from the offset on
4265 * This will find the page for the "from" offset and cow the page and zero the
4266 * part we want to zero. This is used with truncate and hole punching.
4268 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4271 struct address_space *mapping = inode->i_mapping;
4272 struct btrfs_root *root = BTRFS_I(inode)->root;
4273 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4274 struct btrfs_ordered_extent *ordered;
4275 struct extent_state *cached_state = NULL;
4277 u32 blocksize = root->sectorsize;
4278 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4279 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4281 gfp_t mask = btrfs_alloc_write_mask(mapping);
4286 if ((offset & (blocksize - 1)) == 0 &&
4287 (!len || ((len & (blocksize - 1)) == 0)))
4289 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4294 page = find_or_create_page(mapping, index, mask);
4296 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4301 page_start = page_offset(page);
4302 page_end = page_start + PAGE_CACHE_SIZE - 1;
4304 if (!PageUptodate(page)) {
4305 ret = btrfs_readpage(NULL, page);
4307 if (page->mapping != mapping) {
4309 page_cache_release(page);
4312 if (!PageUptodate(page)) {
4317 wait_on_page_writeback(page);
4319 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4320 set_page_extent_mapped(page);
4322 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4324 unlock_extent_cached(io_tree, page_start, page_end,
4325 &cached_state, GFP_NOFS);
4327 page_cache_release(page);
4328 btrfs_start_ordered_extent(inode, ordered, 1);
4329 btrfs_put_ordered_extent(ordered);
4333 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4334 EXTENT_DIRTY | EXTENT_DELALLOC |
4335 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4336 0, 0, &cached_state, GFP_NOFS);
4338 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4341 unlock_extent_cached(io_tree, page_start, page_end,
4342 &cached_state, GFP_NOFS);
4346 if (offset != PAGE_CACHE_SIZE) {
4348 len = PAGE_CACHE_SIZE - offset;
4351 memset(kaddr, 0, offset);
4353 memset(kaddr + offset, 0, len);
4354 flush_dcache_page(page);
4357 ClearPageChecked(page);
4358 set_page_dirty(page);
4359 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4364 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4366 page_cache_release(page);
4372 * This function puts in dummy file extents for the area we're creating a hole
4373 * for. So if we are truncating this file to a larger size we need to insert
4374 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4375 * the range between oldsize and size
4377 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4379 struct btrfs_trans_handle *trans;
4380 struct btrfs_root *root = BTRFS_I(inode)->root;
4381 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4382 struct extent_map *em = NULL;
4383 struct extent_state *cached_state = NULL;
4384 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4385 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4386 u64 block_end = ALIGN(size, root->sectorsize);
4392 if (size <= hole_start)
4396 struct btrfs_ordered_extent *ordered;
4397 btrfs_wait_ordered_range(inode, hole_start,
4398 block_end - hole_start);
4399 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4401 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
4404 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4405 &cached_state, GFP_NOFS);
4406 btrfs_put_ordered_extent(ordered);
4409 cur_offset = hole_start;
4411 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4412 block_end - cur_offset, 0);
4418 last_byte = min(extent_map_end(em), block_end);
4419 last_byte = ALIGN(last_byte , root->sectorsize);
4420 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4421 struct extent_map *hole_em;
4422 hole_size = last_byte - cur_offset;
4424 trans = btrfs_start_transaction(root, 3);
4425 if (IS_ERR(trans)) {
4426 err = PTR_ERR(trans);
4430 err = btrfs_drop_extents(trans, root, inode,
4432 cur_offset + hole_size, 1);
4434 btrfs_abort_transaction(trans, root, err);
4435 btrfs_end_transaction(trans, root);
4439 err = btrfs_insert_file_extent(trans, root,
4440 btrfs_ino(inode), cur_offset, 0,
4441 0, hole_size, 0, hole_size,
4444 btrfs_abort_transaction(trans, root, err);
4445 btrfs_end_transaction(trans, root);
4449 btrfs_drop_extent_cache(inode, cur_offset,
4450 cur_offset + hole_size - 1, 0);
4451 hole_em = alloc_extent_map();
4453 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4454 &BTRFS_I(inode)->runtime_flags);
4457 hole_em->start = cur_offset;
4458 hole_em->len = hole_size;
4459 hole_em->orig_start = cur_offset;
4461 hole_em->block_start = EXTENT_MAP_HOLE;
4462 hole_em->block_len = 0;
4463 hole_em->orig_block_len = 0;
4464 hole_em->ram_bytes = hole_size;
4465 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4466 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4467 hole_em->generation = trans->transid;
4470 write_lock(&em_tree->lock);
4471 err = add_extent_mapping(em_tree, hole_em, 1);
4472 write_unlock(&em_tree->lock);
4475 btrfs_drop_extent_cache(inode, cur_offset,
4479 free_extent_map(hole_em);
4481 btrfs_update_inode(trans, root, inode);
4482 btrfs_end_transaction(trans, root);
4484 free_extent_map(em);
4486 cur_offset = last_byte;
4487 if (cur_offset >= block_end)
4491 free_extent_map(em);
4492 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4497 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4499 struct btrfs_root *root = BTRFS_I(inode)->root;
4500 struct btrfs_trans_handle *trans;
4501 loff_t oldsize = i_size_read(inode);
4502 loff_t newsize = attr->ia_size;
4503 int mask = attr->ia_valid;
4506 if (newsize == oldsize)
4510 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4511 * special case where we need to update the times despite not having
4512 * these flags set. For all other operations the VFS set these flags
4513 * explicitly if it wants a timestamp update.
4515 if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4516 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4518 if (newsize > oldsize) {
4519 truncate_pagecache(inode, oldsize, newsize);
4520 ret = btrfs_cont_expand(inode, oldsize, newsize);
4524 trans = btrfs_start_transaction(root, 1);
4526 return PTR_ERR(trans);
4528 i_size_write(inode, newsize);
4529 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4530 ret = btrfs_update_inode(trans, root, inode);
4531 btrfs_end_transaction(trans, root);
4535 * We're truncating a file that used to have good data down to
4536 * zero. Make sure it gets into the ordered flush list so that
4537 * any new writes get down to disk quickly.
4540 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4541 &BTRFS_I(inode)->runtime_flags);
4544 * 1 for the orphan item we're going to add
4545 * 1 for the orphan item deletion.
4547 trans = btrfs_start_transaction(root, 2);
4549 return PTR_ERR(trans);
4552 * We need to do this in case we fail at _any_ point during the
4553 * actual truncate. Once we do the truncate_setsize we could
4554 * invalidate pages which forces any outstanding ordered io to
4555 * be instantly completed which will give us extents that need
4556 * to be truncated. If we fail to get an orphan inode down we
4557 * could have left over extents that were never meant to live,
4558 * so we need to garuntee from this point on that everything
4559 * will be consistent.
4561 ret = btrfs_orphan_add(trans, inode);
4562 btrfs_end_transaction(trans, root);
4566 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4567 truncate_setsize(inode, newsize);
4569 /* Disable nonlocked read DIO to avoid the end less truncate */
4570 btrfs_inode_block_unlocked_dio(inode);
4571 inode_dio_wait(inode);
4572 btrfs_inode_resume_unlocked_dio(inode);
4574 ret = btrfs_truncate(inode);
4575 if (ret && inode->i_nlink)
4576 btrfs_orphan_del(NULL, inode);
4582 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4584 struct inode *inode = dentry->d_inode;
4585 struct btrfs_root *root = BTRFS_I(inode)->root;
4588 if (btrfs_root_readonly(root))
4591 err = inode_change_ok(inode, attr);
4595 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4596 err = btrfs_setsize(inode, attr);
4601 if (attr->ia_valid) {
4602 setattr_copy(inode, attr);
4603 inode_inc_iversion(inode);
4604 err = btrfs_dirty_inode(inode);
4606 if (!err && attr->ia_valid & ATTR_MODE)
4607 err = btrfs_acl_chmod(inode);
4613 void btrfs_evict_inode(struct inode *inode)
4615 struct btrfs_trans_handle *trans;
4616 struct btrfs_root *root = BTRFS_I(inode)->root;
4617 struct btrfs_block_rsv *rsv, *global_rsv;
4618 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4621 trace_btrfs_inode_evict(inode);
4623 truncate_inode_pages(&inode->i_data, 0);
4624 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
4625 btrfs_is_free_space_inode(inode)))
4628 if (is_bad_inode(inode)) {
4629 btrfs_orphan_del(NULL, inode);
4632 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4633 btrfs_wait_ordered_range(inode, 0, (u64)-1);
4635 if (root->fs_info->log_root_recovering) {
4636 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4637 &BTRFS_I(inode)->runtime_flags));
4641 if (inode->i_nlink > 0) {
4642 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
4646 ret = btrfs_commit_inode_delayed_inode(inode);
4648 btrfs_orphan_del(NULL, inode);
4652 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4654 btrfs_orphan_del(NULL, inode);
4657 rsv->size = min_size;
4659 global_rsv = &root->fs_info->global_block_rsv;
4661 btrfs_i_size_write(inode, 0);
4664 * This is a bit simpler than btrfs_truncate since we've already
4665 * reserved our space for our orphan item in the unlink, so we just
4666 * need to reserve some slack space in case we add bytes and update
4667 * inode item when doing the truncate.
4670 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4671 BTRFS_RESERVE_FLUSH_LIMIT);
4674 * Try and steal from the global reserve since we will
4675 * likely not use this space anyway, we want to try as
4676 * hard as possible to get this to work.
4679 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4682 btrfs_warn(root->fs_info,
4683 "Could not get space for a delete, will truncate on mount %d",
4685 btrfs_orphan_del(NULL, inode);
4686 btrfs_free_block_rsv(root, rsv);
4690 trans = btrfs_join_transaction(root);
4691 if (IS_ERR(trans)) {
4692 btrfs_orphan_del(NULL, inode);
4693 btrfs_free_block_rsv(root, rsv);
4697 trans->block_rsv = rsv;
4699 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4703 trans->block_rsv = &root->fs_info->trans_block_rsv;
4704 btrfs_end_transaction(trans, root);
4706 btrfs_btree_balance_dirty(root);
4709 btrfs_free_block_rsv(root, rsv);
4712 trans->block_rsv = root->orphan_block_rsv;
4713 ret = btrfs_orphan_del(trans, inode);
4717 trans->block_rsv = &root->fs_info->trans_block_rsv;
4718 if (!(root == root->fs_info->tree_root ||
4719 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4720 btrfs_return_ino(root, btrfs_ino(inode));
4722 btrfs_end_transaction(trans, root);
4723 btrfs_btree_balance_dirty(root);
4725 btrfs_remove_delayed_node(inode);
4731 * this returns the key found in the dir entry in the location pointer.
4732 * If no dir entries were found, location->objectid is 0.
4734 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4735 struct btrfs_key *location)
4737 const char *name = dentry->d_name.name;
4738 int namelen = dentry->d_name.len;
4739 struct btrfs_dir_item *di;
4740 struct btrfs_path *path;
4741 struct btrfs_root *root = BTRFS_I(dir)->root;
4744 path = btrfs_alloc_path();
4748 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4753 if (IS_ERR_OR_NULL(di))
4756 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4758 btrfs_free_path(path);
4761 location->objectid = 0;
4766 * when we hit a tree root in a directory, the btrfs part of the inode
4767 * needs to be changed to reflect the root directory of the tree root. This
4768 * is kind of like crossing a mount point.
4770 static int fixup_tree_root_location(struct btrfs_root *root,
4772 struct dentry *dentry,
4773 struct btrfs_key *location,
4774 struct btrfs_root **sub_root)
4776 struct btrfs_path *path;
4777 struct btrfs_root *new_root;
4778 struct btrfs_root_ref *ref;
4779 struct extent_buffer *leaf;
4783 path = btrfs_alloc_path();
4790 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4791 BTRFS_I(dir)->root->root_key.objectid,
4792 location->objectid);
4799 leaf = path->nodes[0];
4800 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4801 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4802 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4805 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4806 (unsigned long)(ref + 1),
4807 dentry->d_name.len);
4811 btrfs_release_path(path);
4813 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4814 if (IS_ERR(new_root)) {
4815 err = PTR_ERR(new_root);
4819 *sub_root = new_root;
4820 location->objectid = btrfs_root_dirid(&new_root->root_item);
4821 location->type = BTRFS_INODE_ITEM_KEY;
4822 location->offset = 0;
4825 btrfs_free_path(path);
4829 static void inode_tree_add(struct inode *inode)
4831 struct btrfs_root *root = BTRFS_I(inode)->root;
4832 struct btrfs_inode *entry;
4834 struct rb_node *parent;
4835 u64 ino = btrfs_ino(inode);
4837 if (inode_unhashed(inode))
4841 spin_lock(&root->inode_lock);
4842 p = &root->inode_tree.rb_node;
4845 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4847 if (ino < btrfs_ino(&entry->vfs_inode))
4848 p = &parent->rb_left;
4849 else if (ino > btrfs_ino(&entry->vfs_inode))
4850 p = &parent->rb_right;
4852 WARN_ON(!(entry->vfs_inode.i_state &
4853 (I_WILL_FREE | I_FREEING)));
4854 rb_erase(parent, &root->inode_tree);
4855 RB_CLEAR_NODE(parent);
4856 spin_unlock(&root->inode_lock);
4860 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4861 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4862 spin_unlock(&root->inode_lock);
4865 static void inode_tree_del(struct inode *inode)
4867 struct btrfs_root *root = BTRFS_I(inode)->root;
4870 spin_lock(&root->inode_lock);
4871 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4872 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4873 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4874 empty = RB_EMPTY_ROOT(&root->inode_tree);
4876 spin_unlock(&root->inode_lock);
4879 * Free space cache has inodes in the tree root, but the tree root has a
4880 * root_refs of 0, so this could end up dropping the tree root as a
4881 * snapshot, so we need the extra !root->fs_info->tree_root check to
4882 * make sure we don't drop it.
4884 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4885 root != root->fs_info->tree_root) {
4886 synchronize_srcu(&root->fs_info->subvol_srcu);
4887 spin_lock(&root->inode_lock);
4888 empty = RB_EMPTY_ROOT(&root->inode_tree);
4889 spin_unlock(&root->inode_lock);
4891 btrfs_add_dead_root(root);
4895 void btrfs_invalidate_inodes(struct btrfs_root *root)
4897 struct rb_node *node;
4898 struct rb_node *prev;
4899 struct btrfs_inode *entry;
4900 struct inode *inode;
4903 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4905 spin_lock(&root->inode_lock);
4907 node = root->inode_tree.rb_node;
4911 entry = rb_entry(node, struct btrfs_inode, rb_node);
4913 if (objectid < btrfs_ino(&entry->vfs_inode))
4914 node = node->rb_left;
4915 else if (objectid > btrfs_ino(&entry->vfs_inode))
4916 node = node->rb_right;
4922 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4923 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4927 prev = rb_next(prev);
4931 entry = rb_entry(node, struct btrfs_inode, rb_node);
4932 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4933 inode = igrab(&entry->vfs_inode);
4935 spin_unlock(&root->inode_lock);
4936 if (atomic_read(&inode->i_count) > 1)
4937 d_prune_aliases(inode);
4939 * btrfs_drop_inode will have it removed from
4940 * the inode cache when its usage count
4945 spin_lock(&root->inode_lock);
4949 if (cond_resched_lock(&root->inode_lock))
4952 node = rb_next(node);
4954 spin_unlock(&root->inode_lock);
4957 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4959 struct btrfs_iget_args *args = p;
4960 inode->i_ino = args->ino;
4961 BTRFS_I(inode)->root = args->root;
4965 static int btrfs_find_actor(struct inode *inode, void *opaque)
4967 struct btrfs_iget_args *args = opaque;
4968 return args->ino == btrfs_ino(inode) &&
4969 args->root == BTRFS_I(inode)->root;
4972 static struct inode *btrfs_iget_locked(struct super_block *s,
4974 struct btrfs_root *root)
4976 struct inode *inode;
4977 struct btrfs_iget_args args;
4978 args.ino = objectid;
4981 inode = iget5_locked(s, objectid, btrfs_find_actor,
4982 btrfs_init_locked_inode,
4987 /* Get an inode object given its location and corresponding root.
4988 * Returns in *is_new if the inode was read from disk
4990 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4991 struct btrfs_root *root, int *new)
4993 struct inode *inode;
4995 inode = btrfs_iget_locked(s, location->objectid, root);
4997 return ERR_PTR(-ENOMEM);
4999 if (inode->i_state & I_NEW) {
5000 BTRFS_I(inode)->root = root;
5001 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
5002 btrfs_read_locked_inode(inode);
5003 if (!is_bad_inode(inode)) {
5004 inode_tree_add(inode);
5005 unlock_new_inode(inode);
5009 unlock_new_inode(inode);
5011 inode = ERR_PTR(-ESTALE);
5018 static struct inode *new_simple_dir(struct super_block *s,
5019 struct btrfs_key *key,
5020 struct btrfs_root *root)
5022 struct inode *inode = new_inode(s);
5025 return ERR_PTR(-ENOMEM);
5027 BTRFS_I(inode)->root = root;
5028 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5029 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5031 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5032 inode->i_op = &btrfs_dir_ro_inode_operations;
5033 inode->i_fop = &simple_dir_operations;
5034 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5035 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5040 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5042 struct inode *inode;
5043 struct btrfs_root *root = BTRFS_I(dir)->root;
5044 struct btrfs_root *sub_root = root;
5045 struct btrfs_key location;
5049 if (dentry->d_name.len > BTRFS_NAME_LEN)
5050 return ERR_PTR(-ENAMETOOLONG);
5052 ret = btrfs_inode_by_name(dir, dentry, &location);
5054 return ERR_PTR(ret);
5056 if (location.objectid == 0)
5059 if (location.type == BTRFS_INODE_ITEM_KEY) {
5060 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5064 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5066 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5067 ret = fixup_tree_root_location(root, dir, dentry,
5068 &location, &sub_root);
5071 inode = ERR_PTR(ret);
5073 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5075 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5077 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5079 if (!IS_ERR(inode) && root != sub_root) {
5080 down_read(&root->fs_info->cleanup_work_sem);
5081 if (!(inode->i_sb->s_flags & MS_RDONLY))
5082 ret = btrfs_orphan_cleanup(sub_root);
5083 up_read(&root->fs_info->cleanup_work_sem);
5085 inode = ERR_PTR(ret);
5091 static int btrfs_dentry_delete(const struct dentry *dentry)
5093 struct btrfs_root *root;
5094 struct inode *inode = dentry->d_inode;
5096 if (!inode && !IS_ROOT(dentry))
5097 inode = dentry->d_parent->d_inode;
5100 root = BTRFS_I(inode)->root;
5101 if (btrfs_root_refs(&root->root_item) == 0)
5104 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5110 static void btrfs_dentry_release(struct dentry *dentry)
5112 if (dentry->d_fsdata)
5113 kfree(dentry->d_fsdata);
5116 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5121 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
5125 unsigned char btrfs_filetype_table[] = {
5126 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5129 static int btrfs_real_readdir(struct file *filp, void *dirent,
5132 struct inode *inode = file_inode(filp);
5133 struct btrfs_root *root = BTRFS_I(inode)->root;
5134 struct btrfs_item *item;
5135 struct btrfs_dir_item *di;
5136 struct btrfs_key key;
5137 struct btrfs_key found_key;
5138 struct btrfs_path *path;
5139 struct list_head ins_list;
5140 struct list_head del_list;
5142 struct extent_buffer *leaf;
5144 unsigned char d_type;
5149 int key_type = BTRFS_DIR_INDEX_KEY;
5153 int is_curr = 0; /* filp->f_pos points to the current index? */
5155 /* FIXME, use a real flag for deciding about the key type */
5156 if (root->fs_info->tree_root == root)
5157 key_type = BTRFS_DIR_ITEM_KEY;
5159 /* special case for "." */
5160 if (filp->f_pos == 0) {
5161 over = filldir(dirent, ".", 1,
5162 filp->f_pos, btrfs_ino(inode), DT_DIR);
5167 /* special case for .., just use the back ref */
5168 if (filp->f_pos == 1) {
5169 u64 pino = parent_ino(filp->f_path.dentry);
5170 over = filldir(dirent, "..", 2,
5171 filp->f_pos, pino, DT_DIR);
5176 path = btrfs_alloc_path();
5182 if (key_type == BTRFS_DIR_INDEX_KEY) {
5183 INIT_LIST_HEAD(&ins_list);
5184 INIT_LIST_HEAD(&del_list);
5185 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5188 btrfs_set_key_type(&key, key_type);
5189 key.offset = filp->f_pos;
5190 key.objectid = btrfs_ino(inode);
5192 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5197 leaf = path->nodes[0];
5198 slot = path->slots[0];
5199 if (slot >= btrfs_header_nritems(leaf)) {
5200 ret = btrfs_next_leaf(root, path);
5208 item = btrfs_item_nr(leaf, slot);
5209 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5211 if (found_key.objectid != key.objectid)
5213 if (btrfs_key_type(&found_key) != key_type)
5215 if (found_key.offset < filp->f_pos)
5217 if (key_type == BTRFS_DIR_INDEX_KEY &&
5218 btrfs_should_delete_dir_index(&del_list,
5222 filp->f_pos = found_key.offset;
5225 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5227 di_total = btrfs_item_size(leaf, item);
5229 while (di_cur < di_total) {
5230 struct btrfs_key location;
5232 if (verify_dir_item(root, leaf, di))
5235 name_len = btrfs_dir_name_len(leaf, di);
5236 if (name_len <= sizeof(tmp_name)) {
5237 name_ptr = tmp_name;
5239 name_ptr = kmalloc(name_len, GFP_NOFS);
5245 read_extent_buffer(leaf, name_ptr,
5246 (unsigned long)(di + 1), name_len);
5248 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5249 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5252 /* is this a reference to our own snapshot? If so
5255 * In contrast to old kernels, we insert the snapshot's
5256 * dir item and dir index after it has been created, so
5257 * we won't find a reference to our own snapshot. We
5258 * still keep the following code for backward
5261 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5262 location.objectid == root->root_key.objectid) {
5266 over = filldir(dirent, name_ptr, name_len,
5267 found_key.offset, location.objectid,
5271 if (name_ptr != tmp_name)
5276 di_len = btrfs_dir_name_len(leaf, di) +
5277 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5279 di = (struct btrfs_dir_item *)((char *)di + di_len);
5285 if (key_type == BTRFS_DIR_INDEX_KEY) {
5288 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
5294 /* Reached end of directory/root. Bump pos past the last item. */
5295 if (key_type == BTRFS_DIR_INDEX_KEY)
5297 * 32-bit glibc will use getdents64, but then strtol -
5298 * so the last number we can serve is this.
5300 filp->f_pos = 0x7fffffff;
5306 if (key_type == BTRFS_DIR_INDEX_KEY)
5307 btrfs_put_delayed_items(&ins_list, &del_list);
5308 btrfs_free_path(path);
5312 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5314 struct btrfs_root *root = BTRFS_I(inode)->root;
5315 struct btrfs_trans_handle *trans;
5317 bool nolock = false;
5319 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5322 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5325 if (wbc->sync_mode == WB_SYNC_ALL) {
5327 trans = btrfs_join_transaction_nolock(root);
5329 trans = btrfs_join_transaction(root);
5331 return PTR_ERR(trans);
5332 ret = btrfs_commit_transaction(trans, root);
5338 * This is somewhat expensive, updating the tree every time the
5339 * inode changes. But, it is most likely to find the inode in cache.
5340 * FIXME, needs more benchmarking...there are no reasons other than performance
5341 * to keep or drop this code.
5343 static int btrfs_dirty_inode(struct inode *inode)
5345 struct btrfs_root *root = BTRFS_I(inode)->root;
5346 struct btrfs_trans_handle *trans;
5349 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5352 trans = btrfs_join_transaction(root);
5354 return PTR_ERR(trans);
5356 ret = btrfs_update_inode(trans, root, inode);
5357 if (ret && ret == -ENOSPC) {
5358 /* whoops, lets try again with the full transaction */
5359 btrfs_end_transaction(trans, root);
5360 trans = btrfs_start_transaction(root, 1);
5362 return PTR_ERR(trans);
5364 ret = btrfs_update_inode(trans, root, inode);
5366 btrfs_end_transaction(trans, root);
5367 if (BTRFS_I(inode)->delayed_node)
5368 btrfs_balance_delayed_items(root);
5374 * This is a copy of file_update_time. We need this so we can return error on
5375 * ENOSPC for updating the inode in the case of file write and mmap writes.
5377 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5380 struct btrfs_root *root = BTRFS_I(inode)->root;
5382 if (btrfs_root_readonly(root))
5385 if (flags & S_VERSION)
5386 inode_inc_iversion(inode);
5387 if (flags & S_CTIME)
5388 inode->i_ctime = *now;
5389 if (flags & S_MTIME)
5390 inode->i_mtime = *now;
5391 if (flags & S_ATIME)
5392 inode->i_atime = *now;
5393 return btrfs_dirty_inode(inode);
5397 * find the highest existing sequence number in a directory
5398 * and then set the in-memory index_cnt variable to reflect
5399 * free sequence numbers
5401 static int btrfs_set_inode_index_count(struct inode *inode)
5403 struct btrfs_root *root = BTRFS_I(inode)->root;
5404 struct btrfs_key key, found_key;
5405 struct btrfs_path *path;
5406 struct extent_buffer *leaf;
5409 key.objectid = btrfs_ino(inode);
5410 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5411 key.offset = (u64)-1;
5413 path = btrfs_alloc_path();
5417 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5420 /* FIXME: we should be able to handle this */
5426 * MAGIC NUMBER EXPLANATION:
5427 * since we search a directory based on f_pos we have to start at 2
5428 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5429 * else has to start at 2
5431 if (path->slots[0] == 0) {
5432 BTRFS_I(inode)->index_cnt = 2;
5438 leaf = path->nodes[0];
5439 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5441 if (found_key.objectid != btrfs_ino(inode) ||
5442 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5443 BTRFS_I(inode)->index_cnt = 2;
5447 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5449 btrfs_free_path(path);
5454 * helper to find a free sequence number in a given directory. This current
5455 * code is very simple, later versions will do smarter things in the btree
5457 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5461 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5462 ret = btrfs_inode_delayed_dir_index_count(dir);
5464 ret = btrfs_set_inode_index_count(dir);
5470 *index = BTRFS_I(dir)->index_cnt;
5471 BTRFS_I(dir)->index_cnt++;
5476 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5477 struct btrfs_root *root,
5479 const char *name, int name_len,
5480 u64 ref_objectid, u64 objectid,
5481 umode_t mode, u64 *index)
5483 struct inode *inode;
5484 struct btrfs_inode_item *inode_item;
5485 struct btrfs_key *location;
5486 struct btrfs_path *path;
5487 struct btrfs_inode_ref *ref;
5488 struct btrfs_key key[2];
5494 path = btrfs_alloc_path();
5496 return ERR_PTR(-ENOMEM);
5498 inode = new_inode(root->fs_info->sb);
5500 btrfs_free_path(path);
5501 return ERR_PTR(-ENOMEM);
5505 * we have to initialize this early, so we can reclaim the inode
5506 * number if we fail afterwards in this function.
5508 inode->i_ino = objectid;
5511 trace_btrfs_inode_request(dir);
5513 ret = btrfs_set_inode_index(dir, index);
5515 btrfs_free_path(path);
5517 return ERR_PTR(ret);
5521 * index_cnt is ignored for everything but a dir,
5522 * btrfs_get_inode_index_count has an explanation for the magic
5525 BTRFS_I(inode)->index_cnt = 2;
5526 BTRFS_I(inode)->root = root;
5527 BTRFS_I(inode)->generation = trans->transid;
5528 inode->i_generation = BTRFS_I(inode)->generation;
5531 * We could have gotten an inode number from somebody who was fsynced
5532 * and then removed in this same transaction, so let's just set full
5533 * sync since it will be a full sync anyway and this will blow away the
5534 * old info in the log.
5536 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5543 key[0].objectid = objectid;
5544 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5548 * Start new inodes with an inode_ref. This is slightly more
5549 * efficient for small numbers of hard links since they will
5550 * be packed into one item. Extended refs will kick in if we
5551 * add more hard links than can fit in the ref item.
5553 key[1].objectid = objectid;
5554 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5555 key[1].offset = ref_objectid;
5557 sizes[0] = sizeof(struct btrfs_inode_item);
5558 sizes[1] = name_len + sizeof(*ref);
5560 path->leave_spinning = 1;
5561 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5565 inode_init_owner(inode, dir, mode);
5566 inode_set_bytes(inode, 0);
5567 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5568 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5569 struct btrfs_inode_item);
5570 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5571 sizeof(*inode_item));
5572 fill_inode_item(trans, path->nodes[0], inode_item, inode);
5574 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5575 struct btrfs_inode_ref);
5576 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5577 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5578 ptr = (unsigned long)(ref + 1);
5579 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5581 btrfs_mark_buffer_dirty(path->nodes[0]);
5582 btrfs_free_path(path);
5584 location = &BTRFS_I(inode)->location;
5585 location->objectid = objectid;
5586 location->offset = 0;
5587 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5589 btrfs_inherit_iflags(inode, dir);
5591 if (S_ISREG(mode)) {
5592 if (btrfs_test_opt(root, NODATASUM))
5593 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5594 if (btrfs_test_opt(root, NODATACOW))
5595 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5596 BTRFS_INODE_NODATASUM;
5599 insert_inode_hash(inode);
5600 inode_tree_add(inode);
5602 trace_btrfs_inode_new(inode);
5603 btrfs_set_inode_last_trans(trans, inode);
5605 btrfs_update_root_times(trans, root);
5610 BTRFS_I(dir)->index_cnt--;
5611 btrfs_free_path(path);
5613 return ERR_PTR(ret);
5616 static inline u8 btrfs_inode_type(struct inode *inode)
5618 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5622 * utility function to add 'inode' into 'parent_inode' with
5623 * a give name and a given sequence number.
5624 * if 'add_backref' is true, also insert a backref from the
5625 * inode to the parent directory.
5627 int btrfs_add_link(struct btrfs_trans_handle *trans,
5628 struct inode *parent_inode, struct inode *inode,
5629 const char *name, int name_len, int add_backref, u64 index)
5632 struct btrfs_key key;
5633 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5634 u64 ino = btrfs_ino(inode);
5635 u64 parent_ino = btrfs_ino(parent_inode);
5637 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5638 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5641 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5645 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5646 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5647 key.objectid, root->root_key.objectid,
5648 parent_ino, index, name, name_len);
5649 } else if (add_backref) {
5650 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5654 /* Nothing to clean up yet */
5658 ret = btrfs_insert_dir_item(trans, root, name, name_len,
5660 btrfs_inode_type(inode), index);
5661 if (ret == -EEXIST || ret == -EOVERFLOW)
5664 btrfs_abort_transaction(trans, root, ret);
5668 btrfs_i_size_write(parent_inode, parent_inode->i_size +
5670 inode_inc_iversion(parent_inode);
5671 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5672 ret = btrfs_update_inode(trans, root, parent_inode);
5674 btrfs_abort_transaction(trans, root, ret);
5678 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5681 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5682 key.objectid, root->root_key.objectid,
5683 parent_ino, &local_index, name, name_len);
5685 } else if (add_backref) {
5689 err = btrfs_del_inode_ref(trans, root, name, name_len,
5690 ino, parent_ino, &local_index);
5695 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5696 struct inode *dir, struct dentry *dentry,
5697 struct inode *inode, int backref, u64 index)
5699 int err = btrfs_add_link(trans, dir, inode,
5700 dentry->d_name.name, dentry->d_name.len,
5707 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5708 umode_t mode, dev_t rdev)
5710 struct btrfs_trans_handle *trans;
5711 struct btrfs_root *root = BTRFS_I(dir)->root;
5712 struct inode *inode = NULL;
5718 if (!new_valid_dev(rdev))
5722 * 2 for inode item and ref
5724 * 1 for xattr if selinux is on
5726 trans = btrfs_start_transaction(root, 5);
5728 return PTR_ERR(trans);
5730 err = btrfs_find_free_ino(root, &objectid);
5734 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5735 dentry->d_name.len, btrfs_ino(dir), objectid,
5737 if (IS_ERR(inode)) {
5738 err = PTR_ERR(inode);
5742 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5749 * If the active LSM wants to access the inode during
5750 * d_instantiate it needs these. Smack checks to see
5751 * if the filesystem supports xattrs by looking at the
5755 inode->i_op = &btrfs_special_inode_operations;
5756 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5760 init_special_inode(inode, inode->i_mode, rdev);
5761 btrfs_update_inode(trans, root, inode);
5762 d_instantiate(dentry, inode);
5765 btrfs_end_transaction(trans, root);
5766 btrfs_btree_balance_dirty(root);
5768 inode_dec_link_count(inode);
5774 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5775 umode_t mode, bool excl)
5777 struct btrfs_trans_handle *trans;
5778 struct btrfs_root *root = BTRFS_I(dir)->root;
5779 struct inode *inode = NULL;
5780 int drop_inode_on_err = 0;
5786 * 2 for inode item and ref
5788 * 1 for xattr if selinux is on
5790 trans = btrfs_start_transaction(root, 5);
5792 return PTR_ERR(trans);
5794 err = btrfs_find_free_ino(root, &objectid);
5798 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5799 dentry->d_name.len, btrfs_ino(dir), objectid,
5801 if (IS_ERR(inode)) {
5802 err = PTR_ERR(inode);
5805 drop_inode_on_err = 1;
5807 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5811 err = btrfs_update_inode(trans, root, inode);
5816 * If the active LSM wants to access the inode during
5817 * d_instantiate it needs these. Smack checks to see
5818 * if the filesystem supports xattrs by looking at the
5821 inode->i_fop = &btrfs_file_operations;
5822 inode->i_op = &btrfs_file_inode_operations;
5824 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5828 inode->i_mapping->a_ops = &btrfs_aops;
5829 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5830 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5831 d_instantiate(dentry, inode);
5834 btrfs_end_transaction(trans, root);
5835 if (err && drop_inode_on_err) {
5836 inode_dec_link_count(inode);
5839 btrfs_btree_balance_dirty(root);
5843 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5844 struct dentry *dentry)
5846 struct btrfs_trans_handle *trans;
5847 struct btrfs_root *root = BTRFS_I(dir)->root;
5848 struct inode *inode = old_dentry->d_inode;
5853 /* do not allow sys_link's with other subvols of the same device */
5854 if (root->objectid != BTRFS_I(inode)->root->objectid)
5857 if (inode->i_nlink >= BTRFS_LINK_MAX)
5860 err = btrfs_set_inode_index(dir, &index);
5865 * 2 items for inode and inode ref
5866 * 2 items for dir items
5867 * 1 item for parent inode
5869 trans = btrfs_start_transaction(root, 5);
5870 if (IS_ERR(trans)) {
5871 err = PTR_ERR(trans);
5875 btrfs_inc_nlink(inode);
5876 inode_inc_iversion(inode);
5877 inode->i_ctime = CURRENT_TIME;
5879 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5881 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5886 struct dentry *parent = dentry->d_parent;
5887 err = btrfs_update_inode(trans, root, inode);
5890 d_instantiate(dentry, inode);
5891 btrfs_log_new_name(trans, inode, NULL, parent);
5894 btrfs_end_transaction(trans, root);
5897 inode_dec_link_count(inode);
5900 btrfs_btree_balance_dirty(root);
5904 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5906 struct inode *inode = NULL;
5907 struct btrfs_trans_handle *trans;
5908 struct btrfs_root *root = BTRFS_I(dir)->root;
5910 int drop_on_err = 0;
5915 * 2 items for inode and ref
5916 * 2 items for dir items
5917 * 1 for xattr if selinux is on
5919 trans = btrfs_start_transaction(root, 5);
5921 return PTR_ERR(trans);
5923 err = btrfs_find_free_ino(root, &objectid);
5927 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5928 dentry->d_name.len, btrfs_ino(dir), objectid,
5929 S_IFDIR | mode, &index);
5930 if (IS_ERR(inode)) {
5931 err = PTR_ERR(inode);
5937 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5941 inode->i_op = &btrfs_dir_inode_operations;
5942 inode->i_fop = &btrfs_dir_file_operations;
5944 btrfs_i_size_write(inode, 0);
5945 err = btrfs_update_inode(trans, root, inode);
5949 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5950 dentry->d_name.len, 0, index);
5954 d_instantiate(dentry, inode);
5958 btrfs_end_transaction(trans, root);
5961 btrfs_btree_balance_dirty(root);
5965 /* helper for btfs_get_extent. Given an existing extent in the tree,
5966 * and an extent that you want to insert, deal with overlap and insert
5967 * the new extent into the tree.
5969 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5970 struct extent_map *existing,
5971 struct extent_map *em,
5972 u64 map_start, u64 map_len)
5976 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5977 start_diff = map_start - em->start;
5978 em->start = map_start;
5980 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5981 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5982 em->block_start += start_diff;
5983 em->block_len -= start_diff;
5985 return add_extent_mapping(em_tree, em, 0);
5988 static noinline int uncompress_inline(struct btrfs_path *path,
5989 struct inode *inode, struct page *page,
5990 size_t pg_offset, u64 extent_offset,
5991 struct btrfs_file_extent_item *item)
5994 struct extent_buffer *leaf = path->nodes[0];
5997 unsigned long inline_size;
6001 WARN_ON(pg_offset != 0);
6002 compress_type = btrfs_file_extent_compression(leaf, item);
6003 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6004 inline_size = btrfs_file_extent_inline_item_len(leaf,
6005 btrfs_item_nr(leaf, path->slots[0]));
6006 tmp = kmalloc(inline_size, GFP_NOFS);
6009 ptr = btrfs_file_extent_inline_start(item);
6011 read_extent_buffer(leaf, tmp, ptr, inline_size);
6013 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6014 ret = btrfs_decompress(compress_type, tmp, page,
6015 extent_offset, inline_size, max_size);
6017 char *kaddr = kmap_atomic(page);
6018 unsigned long copy_size = min_t(u64,
6019 PAGE_CACHE_SIZE - pg_offset,
6020 max_size - extent_offset);
6021 memset(kaddr + pg_offset, 0, copy_size);
6022 kunmap_atomic(kaddr);
6029 * a bit scary, this does extent mapping from logical file offset to the disk.
6030 * the ugly parts come from merging extents from the disk with the in-ram
6031 * representation. This gets more complex because of the data=ordered code,
6032 * where the in-ram extents might be locked pending data=ordered completion.
6034 * This also copies inline extents directly into the page.
6037 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6038 size_t pg_offset, u64 start, u64 len,
6044 u64 extent_start = 0;
6046 u64 objectid = btrfs_ino(inode);
6048 struct btrfs_path *path = NULL;
6049 struct btrfs_root *root = BTRFS_I(inode)->root;
6050 struct btrfs_file_extent_item *item;
6051 struct extent_buffer *leaf;
6052 struct btrfs_key found_key;
6053 struct extent_map *em = NULL;
6054 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6055 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6056 struct btrfs_trans_handle *trans = NULL;
6060 read_lock(&em_tree->lock);
6061 em = lookup_extent_mapping(em_tree, start, len);
6063 em->bdev = root->fs_info->fs_devices->latest_bdev;
6064 read_unlock(&em_tree->lock);
6067 if (em->start > start || em->start + em->len <= start)
6068 free_extent_map(em);
6069 else if (em->block_start == EXTENT_MAP_INLINE && page)
6070 free_extent_map(em);
6074 em = alloc_extent_map();
6079 em->bdev = root->fs_info->fs_devices->latest_bdev;
6080 em->start = EXTENT_MAP_HOLE;
6081 em->orig_start = EXTENT_MAP_HOLE;
6083 em->block_len = (u64)-1;
6086 path = btrfs_alloc_path();
6092 * Chances are we'll be called again, so go ahead and do
6098 ret = btrfs_lookup_file_extent(trans, root, path,
6099 objectid, start, trans != NULL);
6106 if (path->slots[0] == 0)
6111 leaf = path->nodes[0];
6112 item = btrfs_item_ptr(leaf, path->slots[0],
6113 struct btrfs_file_extent_item);
6114 /* are we inside the extent that was found? */
6115 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6116 found_type = btrfs_key_type(&found_key);
6117 if (found_key.objectid != objectid ||
6118 found_type != BTRFS_EXTENT_DATA_KEY) {
6122 found_type = btrfs_file_extent_type(leaf, item);
6123 extent_start = found_key.offset;
6124 compress_type = btrfs_file_extent_compression(leaf, item);
6125 if (found_type == BTRFS_FILE_EXTENT_REG ||
6126 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6127 extent_end = extent_start +
6128 btrfs_file_extent_num_bytes(leaf, item);
6129 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6131 size = btrfs_file_extent_inline_len(leaf, item);
6132 extent_end = ALIGN(extent_start + size, root->sectorsize);
6135 if (start >= extent_end) {
6137 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6138 ret = btrfs_next_leaf(root, path);
6145 leaf = path->nodes[0];
6147 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6148 if (found_key.objectid != objectid ||
6149 found_key.type != BTRFS_EXTENT_DATA_KEY)
6151 if (start + len <= found_key.offset)
6154 em->orig_start = start;
6155 em->len = found_key.offset - start;
6159 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
6160 if (found_type == BTRFS_FILE_EXTENT_REG ||
6161 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6162 em->start = extent_start;
6163 em->len = extent_end - extent_start;
6164 em->orig_start = extent_start -
6165 btrfs_file_extent_offset(leaf, item);
6166 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6168 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6170 em->block_start = EXTENT_MAP_HOLE;
6173 if (compress_type != BTRFS_COMPRESS_NONE) {
6174 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6175 em->compress_type = compress_type;
6176 em->block_start = bytenr;
6177 em->block_len = em->orig_block_len;
6179 bytenr += btrfs_file_extent_offset(leaf, item);
6180 em->block_start = bytenr;
6181 em->block_len = em->len;
6182 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6183 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6186 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6190 size_t extent_offset;
6193 em->block_start = EXTENT_MAP_INLINE;
6194 if (!page || create) {
6195 em->start = extent_start;
6196 em->len = extent_end - extent_start;
6200 size = btrfs_file_extent_inline_len(leaf, item);
6201 extent_offset = page_offset(page) + pg_offset - extent_start;
6202 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6203 size - extent_offset);
6204 em->start = extent_start + extent_offset;
6205 em->len = ALIGN(copy_size, root->sectorsize);
6206 em->orig_block_len = em->len;
6207 em->orig_start = em->start;
6208 if (compress_type) {
6209 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6210 em->compress_type = compress_type;
6212 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6213 if (create == 0 && !PageUptodate(page)) {
6214 if (btrfs_file_extent_compression(leaf, item) !=
6215 BTRFS_COMPRESS_NONE) {
6216 ret = uncompress_inline(path, inode, page,
6218 extent_offset, item);
6219 BUG_ON(ret); /* -ENOMEM */
6222 read_extent_buffer(leaf, map + pg_offset, ptr,
6224 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6225 memset(map + pg_offset + copy_size, 0,
6226 PAGE_CACHE_SIZE - pg_offset -
6231 flush_dcache_page(page);
6232 } else if (create && PageUptodate(page)) {
6236 free_extent_map(em);
6239 btrfs_release_path(path);
6240 trans = btrfs_join_transaction(root);
6243 return ERR_CAST(trans);
6247 write_extent_buffer(leaf, map + pg_offset, ptr,
6250 btrfs_mark_buffer_dirty(leaf);
6252 set_extent_uptodate(io_tree, em->start,
6253 extent_map_end(em) - 1, NULL, GFP_NOFS);
6256 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6260 em->orig_start = start;
6263 em->block_start = EXTENT_MAP_HOLE;
6264 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6266 btrfs_release_path(path);
6267 if (em->start > start || extent_map_end(em) <= start) {
6268 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6269 (unsigned long long)em->start,
6270 (unsigned long long)em->len,
6271 (unsigned long long)start,
6272 (unsigned long long)len);
6278 write_lock(&em_tree->lock);
6279 ret = add_extent_mapping(em_tree, em, 0);
6280 /* it is possible that someone inserted the extent into the tree
6281 * while we had the lock dropped. It is also possible that
6282 * an overlapping map exists in the tree
6284 if (ret == -EEXIST) {
6285 struct extent_map *existing;
6289 existing = lookup_extent_mapping(em_tree, start, len);
6290 if (existing && (existing->start > start ||
6291 existing->start + existing->len <= start)) {
6292 free_extent_map(existing);
6296 existing = lookup_extent_mapping(em_tree, em->start,
6299 err = merge_extent_mapping(em_tree, existing,
6302 free_extent_map(existing);
6304 free_extent_map(em);
6309 free_extent_map(em);
6313 free_extent_map(em);
6318 write_unlock(&em_tree->lock);
6322 trace_btrfs_get_extent(root, em);
6325 btrfs_free_path(path);
6327 ret = btrfs_end_transaction(trans, root);
6332 free_extent_map(em);
6333 return ERR_PTR(err);
6335 BUG_ON(!em); /* Error is always set */
6339 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6340 size_t pg_offset, u64 start, u64 len,
6343 struct extent_map *em;
6344 struct extent_map *hole_em = NULL;
6345 u64 range_start = start;
6351 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6358 * - a pre-alloc extent,
6359 * there might actually be delalloc bytes behind it.
6361 if (em->block_start != EXTENT_MAP_HOLE &&
6362 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6368 /* check to see if we've wrapped (len == -1 or similar) */
6377 /* ok, we didn't find anything, lets look for delalloc */
6378 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6379 end, len, EXTENT_DELALLOC, 1);
6380 found_end = range_start + found;
6381 if (found_end < range_start)
6382 found_end = (u64)-1;
6385 * we didn't find anything useful, return
6386 * the original results from get_extent()
6388 if (range_start > end || found_end <= start) {
6394 /* adjust the range_start to make sure it doesn't
6395 * go backwards from the start they passed in
6397 range_start = max(start,range_start);
6398 found = found_end - range_start;
6401 u64 hole_start = start;
6404 em = alloc_extent_map();
6410 * when btrfs_get_extent can't find anything it
6411 * returns one huge hole
6413 * make sure what it found really fits our range, and
6414 * adjust to make sure it is based on the start from
6418 u64 calc_end = extent_map_end(hole_em);
6420 if (calc_end <= start || (hole_em->start > end)) {
6421 free_extent_map(hole_em);
6424 hole_start = max(hole_em->start, start);
6425 hole_len = calc_end - hole_start;
6429 if (hole_em && range_start > hole_start) {
6430 /* our hole starts before our delalloc, so we
6431 * have to return just the parts of the hole
6432 * that go until the delalloc starts
6434 em->len = min(hole_len,
6435 range_start - hole_start);
6436 em->start = hole_start;
6437 em->orig_start = hole_start;
6439 * don't adjust block start at all,
6440 * it is fixed at EXTENT_MAP_HOLE
6442 em->block_start = hole_em->block_start;
6443 em->block_len = hole_len;
6444 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6445 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6447 em->start = range_start;
6449 em->orig_start = range_start;
6450 em->block_start = EXTENT_MAP_DELALLOC;
6451 em->block_len = found;
6453 } else if (hole_em) {
6458 free_extent_map(hole_em);
6460 free_extent_map(em);
6461 return ERR_PTR(err);
6466 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6469 struct btrfs_root *root = BTRFS_I(inode)->root;
6470 struct btrfs_trans_handle *trans;
6471 struct extent_map *em;
6472 struct btrfs_key ins;
6476 trans = btrfs_join_transaction(root);
6478 return ERR_CAST(trans);
6480 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
6482 alloc_hint = get_extent_allocation_hint(inode, start, len);
6483 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
6484 alloc_hint, &ins, 1);
6490 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6491 ins.offset, ins.offset, ins.offset, 0);
6495 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6496 ins.offset, ins.offset, 0);
6498 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6502 btrfs_end_transaction(trans, root);
6507 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6508 * block must be cow'd
6510 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
6511 struct inode *inode, u64 offset, u64 *len,
6512 u64 *orig_start, u64 *orig_block_len,
6515 struct btrfs_path *path;
6517 struct extent_buffer *leaf;
6518 struct btrfs_root *root = BTRFS_I(inode)->root;
6519 struct btrfs_file_extent_item *fi;
6520 struct btrfs_key key;
6528 path = btrfs_alloc_path();
6532 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
6537 slot = path->slots[0];
6540 /* can't find the item, must cow */
6547 leaf = path->nodes[0];
6548 btrfs_item_key_to_cpu(leaf, &key, slot);
6549 if (key.objectid != btrfs_ino(inode) ||
6550 key.type != BTRFS_EXTENT_DATA_KEY) {
6551 /* not our file or wrong item type, must cow */
6555 if (key.offset > offset) {
6556 /* Wrong offset, must cow */
6560 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6561 found_type = btrfs_file_extent_type(leaf, fi);
6562 if (found_type != BTRFS_FILE_EXTENT_REG &&
6563 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6564 /* not a regular extent, must cow */
6567 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6568 backref_offset = btrfs_file_extent_offset(leaf, fi);
6570 *orig_start = key.offset - backref_offset;
6571 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6572 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6574 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6575 if (extent_end < offset + *len) {
6576 /* extent doesn't include our full range, must cow */
6580 if (btrfs_extent_readonly(root, disk_bytenr))
6584 * look for other files referencing this extent, if we
6585 * find any we must cow
6587 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6588 key.offset - backref_offset, disk_bytenr))
6592 * adjust disk_bytenr and num_bytes to cover just the bytes
6593 * in this extent we are about to write. If there
6594 * are any csums in that range we have to cow in order
6595 * to keep the csums correct
6597 disk_bytenr += backref_offset;
6598 disk_bytenr += offset - key.offset;
6599 num_bytes = min(offset + *len, extent_end) - offset;
6600 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6603 * all of the above have passed, it is safe to overwrite this extent
6609 btrfs_free_path(path);
6613 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6614 struct extent_state **cached_state, int writing)
6616 struct btrfs_ordered_extent *ordered;
6620 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6623 * We're concerned with the entire range that we're going to be
6624 * doing DIO to, so we need to make sure theres no ordered
6625 * extents in this range.
6627 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6628 lockend - lockstart + 1);
6631 * We need to make sure there are no buffered pages in this
6632 * range either, we could have raced between the invalidate in
6633 * generic_file_direct_write and locking the extent. The
6634 * invalidate needs to happen so that reads after a write do not
6637 if (!ordered && (!writing ||
6638 !test_range_bit(&BTRFS_I(inode)->io_tree,
6639 lockstart, lockend, EXTENT_UPTODATE, 0,
6643 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6644 cached_state, GFP_NOFS);
6647 btrfs_start_ordered_extent(inode, ordered, 1);
6648 btrfs_put_ordered_extent(ordered);
6650 /* Screw you mmap */
6651 ret = filemap_write_and_wait_range(inode->i_mapping,
6658 * If we found a page that couldn't be invalidated just
6659 * fall back to buffered.
6661 ret = invalidate_inode_pages2_range(inode->i_mapping,
6662 lockstart >> PAGE_CACHE_SHIFT,
6663 lockend >> PAGE_CACHE_SHIFT);
6674 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6675 u64 len, u64 orig_start,
6676 u64 block_start, u64 block_len,
6677 u64 orig_block_len, u64 ram_bytes,
6680 struct extent_map_tree *em_tree;
6681 struct extent_map *em;
6682 struct btrfs_root *root = BTRFS_I(inode)->root;
6685 em_tree = &BTRFS_I(inode)->extent_tree;
6686 em = alloc_extent_map();
6688 return ERR_PTR(-ENOMEM);
6691 em->orig_start = orig_start;
6692 em->mod_start = start;
6695 em->block_len = block_len;
6696 em->block_start = block_start;
6697 em->bdev = root->fs_info->fs_devices->latest_bdev;
6698 em->orig_block_len = orig_block_len;
6699 em->ram_bytes = ram_bytes;
6700 em->generation = -1;
6701 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6702 if (type == BTRFS_ORDERED_PREALLOC)
6703 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6706 btrfs_drop_extent_cache(inode, em->start,
6707 em->start + em->len - 1, 0);
6708 write_lock(&em_tree->lock);
6709 ret = add_extent_mapping(em_tree, em, 1);
6710 write_unlock(&em_tree->lock);
6711 } while (ret == -EEXIST);
6714 free_extent_map(em);
6715 return ERR_PTR(ret);
6722 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6723 struct buffer_head *bh_result, int create)
6725 struct extent_map *em;
6726 struct btrfs_root *root = BTRFS_I(inode)->root;
6727 struct extent_state *cached_state = NULL;
6728 u64 start = iblock << inode->i_blkbits;
6729 u64 lockstart, lockend;
6730 u64 len = bh_result->b_size;
6731 struct btrfs_trans_handle *trans;
6732 int unlock_bits = EXTENT_LOCKED;
6736 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6738 len = min_t(u64, len, root->sectorsize);
6741 lockend = start + len - 1;
6744 * If this errors out it's because we couldn't invalidate pagecache for
6745 * this range and we need to fallback to buffered.
6747 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6750 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6757 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6758 * io. INLINE is special, and we could probably kludge it in here, but
6759 * it's still buffered so for safety lets just fall back to the generic
6762 * For COMPRESSED we _have_ to read the entire extent in so we can
6763 * decompress it, so there will be buffering required no matter what we
6764 * do, so go ahead and fallback to buffered.
6766 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6767 * to buffered IO. Don't blame me, this is the price we pay for using
6770 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6771 em->block_start == EXTENT_MAP_INLINE) {
6772 free_extent_map(em);
6777 /* Just a good old fashioned hole, return */
6778 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6779 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6780 free_extent_map(em);
6785 * We don't allocate a new extent in the following cases
6787 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6789 * 2) The extent is marked as PREALLOC. We're good to go here and can
6790 * just use the extent.
6794 len = min(len, em->len - (start - em->start));
6795 lockstart = start + len;
6799 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6800 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6801 em->block_start != EXTENT_MAP_HOLE)) {
6804 u64 block_start, orig_start, orig_block_len, ram_bytes;
6806 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6807 type = BTRFS_ORDERED_PREALLOC;
6809 type = BTRFS_ORDERED_NOCOW;
6810 len = min(len, em->len - (start - em->start));
6811 block_start = em->block_start + (start - em->start);
6814 * we're not going to log anything, but we do need
6815 * to make sure the current transaction stays open
6816 * while we look for nocow cross refs
6818 trans = btrfs_join_transaction(root);
6822 if (can_nocow_odirect(trans, inode, start, &len, &orig_start,
6823 &orig_block_len, &ram_bytes) == 1) {
6824 if (type == BTRFS_ORDERED_PREALLOC) {
6825 free_extent_map(em);
6826 em = create_pinned_em(inode, start, len,
6832 btrfs_end_transaction(trans, root);
6837 ret = btrfs_add_ordered_extent_dio(inode, start,
6838 block_start, len, len, type);
6839 btrfs_end_transaction(trans, root);
6841 free_extent_map(em);
6846 btrfs_end_transaction(trans, root);
6850 * this will cow the extent, reset the len in case we changed
6853 len = bh_result->b_size;
6854 free_extent_map(em);
6855 em = btrfs_new_extent_direct(inode, start, len);
6860 len = min(len, em->len - (start - em->start));
6862 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6864 bh_result->b_size = len;
6865 bh_result->b_bdev = em->bdev;
6866 set_buffer_mapped(bh_result);
6868 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6869 set_buffer_new(bh_result);
6872 * Need to update the i_size under the extent lock so buffered
6873 * readers will get the updated i_size when we unlock.
6875 if (start + len > i_size_read(inode))
6876 i_size_write(inode, start + len);
6878 spin_lock(&BTRFS_I(inode)->lock);
6879 BTRFS_I(inode)->outstanding_extents++;
6880 spin_unlock(&BTRFS_I(inode)->lock);
6882 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6883 lockstart + len - 1, EXTENT_DELALLOC, NULL,
6884 &cached_state, GFP_NOFS);
6889 * In the case of write we need to clear and unlock the entire range,
6890 * in the case of read we need to unlock only the end area that we
6891 * aren't using if there is any left over space.
6893 if (lockstart < lockend) {
6894 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6895 lockend, unlock_bits, 1, 0,
6896 &cached_state, GFP_NOFS);
6898 free_extent_state(cached_state);
6901 free_extent_map(em);
6906 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6907 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6911 struct btrfs_dio_private {
6912 struct inode *inode;
6918 /* number of bios pending for this dio */
6919 atomic_t pending_bios;
6924 /* orig_bio is our btrfs_io_bio */
6925 struct bio *orig_bio;
6927 /* dio_bio came from fs/direct-io.c */
6928 struct bio *dio_bio;
6931 static void btrfs_endio_direct_read(struct bio *bio, int err)
6933 struct btrfs_dio_private *dip = bio->bi_private;
6934 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6935 struct bio_vec *bvec = bio->bi_io_vec;
6936 struct inode *inode = dip->inode;
6937 struct btrfs_root *root = BTRFS_I(inode)->root;
6938 struct bio *dio_bio;
6941 start = dip->logical_offset;
6943 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6944 struct page *page = bvec->bv_page;
6947 u64 private = ~(u32)0;
6948 unsigned long flags;
6950 if (get_state_private(&BTRFS_I(inode)->io_tree,
6953 local_irq_save(flags);
6954 kaddr = kmap_atomic(page);
6955 csum = btrfs_csum_data(kaddr + bvec->bv_offset,
6956 csum, bvec->bv_len);
6957 btrfs_csum_final(csum, (char *)&csum);
6958 kunmap_atomic(kaddr);
6959 local_irq_restore(flags);
6961 flush_dcache_page(bvec->bv_page);
6962 if (csum != private) {
6964 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u private %u",
6965 (unsigned long long)btrfs_ino(inode),
6966 (unsigned long long)start,
6967 csum, (unsigned)private);
6972 start += bvec->bv_len;
6974 } while (bvec <= bvec_end);
6976 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6977 dip->logical_offset + dip->bytes - 1);
6978 dio_bio = dip->dio_bio;
6982 /* If we had a csum failure make sure to clear the uptodate flag */
6984 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6985 dio_end_io(dio_bio, err);
6989 static void btrfs_endio_direct_write(struct bio *bio, int err)
6991 struct btrfs_dio_private *dip = bio->bi_private;
6992 struct inode *inode = dip->inode;
6993 struct btrfs_root *root = BTRFS_I(inode)->root;
6994 struct btrfs_ordered_extent *ordered = NULL;
6995 u64 ordered_offset = dip->logical_offset;
6996 u64 ordered_bytes = dip->bytes;
6997 struct bio *dio_bio;
7003 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7005 ordered_bytes, !err);
7009 ordered->work.func = finish_ordered_fn;
7010 ordered->work.flags = 0;
7011 btrfs_queue_worker(&root->fs_info->endio_write_workers,
7015 * our bio might span multiple ordered extents. If we haven't
7016 * completed the accounting for the whole dio, go back and try again
7018 if (ordered_offset < dip->logical_offset + dip->bytes) {
7019 ordered_bytes = dip->logical_offset + dip->bytes -
7025 dio_bio = dip->dio_bio;
7029 /* If we had an error make sure to clear the uptodate flag */
7031 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7032 dio_end_io(dio_bio, err);
7036 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7037 struct bio *bio, int mirror_num,
7038 unsigned long bio_flags, u64 offset)
7041 struct btrfs_root *root = BTRFS_I(inode)->root;
7042 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7043 BUG_ON(ret); /* -ENOMEM */
7047 static void btrfs_end_dio_bio(struct bio *bio, int err)
7049 struct btrfs_dio_private *dip = bio->bi_private;
7052 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
7053 "sector %#Lx len %u err no %d\n",
7054 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
7055 (unsigned long long)bio->bi_sector, bio->bi_size, err);
7059 * before atomic variable goto zero, we must make sure
7060 * dip->errors is perceived to be set.
7062 smp_mb__before_atomic_dec();
7065 /* if there are more bios still pending for this dio, just exit */
7066 if (!atomic_dec_and_test(&dip->pending_bios))
7070 bio_io_error(dip->orig_bio);
7072 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7073 bio_endio(dip->orig_bio, 0);
7079 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7080 u64 first_sector, gfp_t gfp_flags)
7082 int nr_vecs = bio_get_nr_vecs(bdev);
7083 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7086 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7087 int rw, u64 file_offset, int skip_sum,
7090 int write = rw & REQ_WRITE;
7091 struct btrfs_root *root = BTRFS_I(inode)->root;
7095 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7100 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7108 if (write && async_submit) {
7109 ret = btrfs_wq_submit_bio(root->fs_info,
7110 inode, rw, bio, 0, 0,
7112 __btrfs_submit_bio_start_direct_io,
7113 __btrfs_submit_bio_done);
7117 * If we aren't doing async submit, calculate the csum of the
7120 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7123 } else if (!skip_sum) {
7124 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
7130 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7136 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7139 struct inode *inode = dip->inode;
7140 struct btrfs_root *root = BTRFS_I(inode)->root;
7142 struct bio *orig_bio = dip->orig_bio;
7143 struct bio_vec *bvec = orig_bio->bi_io_vec;
7144 u64 start_sector = orig_bio->bi_sector;
7145 u64 file_offset = dip->logical_offset;
7150 int async_submit = 0;
7152 map_length = orig_bio->bi_size;
7153 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7154 &map_length, NULL, 0);
7159 if (map_length >= orig_bio->bi_size) {
7164 /* async crcs make it difficult to collect full stripe writes. */
7165 if (btrfs_get_alloc_profile(root, 1) &
7166 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7171 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7174 bio->bi_private = dip;
7175 bio->bi_end_io = btrfs_end_dio_bio;
7176 atomic_inc(&dip->pending_bios);
7178 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7179 if (unlikely(map_length < submit_len + bvec->bv_len ||
7180 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7181 bvec->bv_offset) < bvec->bv_len)) {
7183 * inc the count before we submit the bio so
7184 * we know the end IO handler won't happen before
7185 * we inc the count. Otherwise, the dip might get freed
7186 * before we're done setting it up
7188 atomic_inc(&dip->pending_bios);
7189 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7190 file_offset, skip_sum,
7194 atomic_dec(&dip->pending_bios);
7198 start_sector += submit_len >> 9;
7199 file_offset += submit_len;
7204 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7205 start_sector, GFP_NOFS);
7208 bio->bi_private = dip;
7209 bio->bi_end_io = btrfs_end_dio_bio;
7211 map_length = orig_bio->bi_size;
7212 ret = btrfs_map_block(root->fs_info, rw,
7214 &map_length, NULL, 0);
7220 submit_len += bvec->bv_len;
7227 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7236 * before atomic variable goto zero, we must
7237 * make sure dip->errors is perceived to be set.
7239 smp_mb__before_atomic_dec();
7240 if (atomic_dec_and_test(&dip->pending_bios))
7241 bio_io_error(dip->orig_bio);
7243 /* bio_end_io() will handle error, so we needn't return it */
7247 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7248 struct inode *inode, loff_t file_offset)
7250 struct btrfs_root *root = BTRFS_I(inode)->root;
7251 struct btrfs_dio_private *dip;
7252 struct bio_vec *bvec = dio_bio->bi_io_vec;
7255 int write = rw & REQ_WRITE;
7258 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7260 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7267 dip = kmalloc(sizeof(*dip), GFP_NOFS);
7273 dip->private = dio_bio->bi_private;
7274 io_bio->bi_private = dio_bio->bi_private;
7276 dip->logical_offset = file_offset;
7280 dip->bytes += bvec->bv_len;
7282 } while (bvec <= (dio_bio->bi_io_vec + dio_bio->bi_vcnt - 1));
7284 dip->disk_bytenr = (u64)dio_bio->bi_sector << 9;
7285 io_bio->bi_private = dip;
7287 dip->orig_bio = io_bio;
7288 dip->dio_bio = dio_bio;
7289 atomic_set(&dip->pending_bios, 0);
7292 io_bio->bi_end_io = btrfs_endio_direct_write;
7294 io_bio->bi_end_io = btrfs_endio_direct_read;
7296 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7305 * If this is a write, we need to clean up the reserved space and kill
7306 * the ordered extent.
7309 struct btrfs_ordered_extent *ordered;
7310 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7311 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7312 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7313 btrfs_free_reserved_extent(root, ordered->start,
7315 btrfs_put_ordered_extent(ordered);
7316 btrfs_put_ordered_extent(ordered);
7318 bio_endio(dio_bio, ret);
7321 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7322 const struct iovec *iov, loff_t offset,
7323 unsigned long nr_segs)
7329 unsigned blocksize_mask = root->sectorsize - 1;
7330 ssize_t retval = -EINVAL;
7331 loff_t end = offset;
7333 if (offset & blocksize_mask)
7336 /* Check the memory alignment. Blocks cannot straddle pages */
7337 for (seg = 0; seg < nr_segs; seg++) {
7338 addr = (unsigned long)iov[seg].iov_base;
7339 size = iov[seg].iov_len;
7341 if ((addr & blocksize_mask) || (size & blocksize_mask))
7344 /* If this is a write we don't need to check anymore */
7349 * Check to make sure we don't have duplicate iov_base's in this
7350 * iovec, if so return EINVAL, otherwise we'll get csum errors
7351 * when reading back.
7353 for (i = seg + 1; i < nr_segs; i++) {
7354 if (iov[seg].iov_base == iov[i].iov_base)
7363 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7364 const struct iovec *iov, loff_t offset,
7365 unsigned long nr_segs)
7367 struct file *file = iocb->ki_filp;
7368 struct inode *inode = file->f_mapping->host;
7372 bool relock = false;
7375 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7379 atomic_inc(&inode->i_dio_count);
7380 smp_mb__after_atomic_inc();
7383 count = iov_length(iov, nr_segs);
7385 * If the write DIO is beyond the EOF, we need update
7386 * the isize, but it is protected by i_mutex. So we can
7387 * not unlock the i_mutex at this case.
7389 if (offset + count <= inode->i_size) {
7390 mutex_unlock(&inode->i_mutex);
7393 ret = btrfs_delalloc_reserve_space(inode, count);
7396 } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7397 &BTRFS_I(inode)->runtime_flags))) {
7398 inode_dio_done(inode);
7399 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7403 ret = __blockdev_direct_IO(rw, iocb, inode,
7404 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7405 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7406 btrfs_submit_direct, flags);
7408 if (ret < 0 && ret != -EIOCBQUEUED)
7409 btrfs_delalloc_release_space(inode, count);
7410 else if (ret >= 0 && (size_t)ret < count)
7411 btrfs_delalloc_release_space(inode,
7412 count - (size_t)ret);
7414 btrfs_delalloc_release_metadata(inode, 0);
7418 inode_dio_done(inode);
7420 mutex_lock(&inode->i_mutex);
7425 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
7427 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7428 __u64 start, __u64 len)
7432 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7436 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7439 int btrfs_readpage(struct file *file, struct page *page)
7441 struct extent_io_tree *tree;
7442 tree = &BTRFS_I(page->mapping->host)->io_tree;
7443 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7446 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7448 struct extent_io_tree *tree;
7451 if (current->flags & PF_MEMALLOC) {
7452 redirty_page_for_writepage(wbc, page);
7456 tree = &BTRFS_I(page->mapping->host)->io_tree;
7457 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7460 static int btrfs_writepages(struct address_space *mapping,
7461 struct writeback_control *wbc)
7463 struct extent_io_tree *tree;
7465 tree = &BTRFS_I(mapping->host)->io_tree;
7466 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7470 btrfs_readpages(struct file *file, struct address_space *mapping,
7471 struct list_head *pages, unsigned nr_pages)
7473 struct extent_io_tree *tree;
7474 tree = &BTRFS_I(mapping->host)->io_tree;
7475 return extent_readpages(tree, mapping, pages, nr_pages,
7478 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7480 struct extent_io_tree *tree;
7481 struct extent_map_tree *map;
7484 tree = &BTRFS_I(page->mapping->host)->io_tree;
7485 map = &BTRFS_I(page->mapping->host)->extent_tree;
7486 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7488 ClearPagePrivate(page);
7489 set_page_private(page, 0);
7490 page_cache_release(page);
7495 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7497 if (PageWriteback(page) || PageDirty(page))
7499 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7502 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
7504 struct inode *inode = page->mapping->host;
7505 struct extent_io_tree *tree;
7506 struct btrfs_ordered_extent *ordered;
7507 struct extent_state *cached_state = NULL;
7508 u64 page_start = page_offset(page);
7509 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7512 * we have the page locked, so new writeback can't start,
7513 * and the dirty bit won't be cleared while we are here.
7515 * Wait for IO on this page so that we can safely clear
7516 * the PagePrivate2 bit and do ordered accounting
7518 wait_on_page_writeback(page);
7520 tree = &BTRFS_I(inode)->io_tree;
7522 btrfs_releasepage(page, GFP_NOFS);
7525 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7526 ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
7529 * IO on this page will never be started, so we need
7530 * to account for any ordered extents now
7532 clear_extent_bit(tree, page_start, page_end,
7533 EXTENT_DIRTY | EXTENT_DELALLOC |
7534 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7535 EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
7537 * whoever cleared the private bit is responsible
7538 * for the finish_ordered_io
7540 if (TestClearPagePrivate2(page) &&
7541 btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
7542 PAGE_CACHE_SIZE, 1)) {
7543 btrfs_finish_ordered_io(ordered);
7545 btrfs_put_ordered_extent(ordered);
7546 cached_state = NULL;
7547 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7549 clear_extent_bit(tree, page_start, page_end,
7550 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
7551 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
7552 &cached_state, GFP_NOFS);
7553 __btrfs_releasepage(page, GFP_NOFS);
7555 ClearPageChecked(page);
7556 if (PagePrivate(page)) {
7557 ClearPagePrivate(page);
7558 set_page_private(page, 0);
7559 page_cache_release(page);
7564 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7565 * called from a page fault handler when a page is first dirtied. Hence we must
7566 * be careful to check for EOF conditions here. We set the page up correctly
7567 * for a written page which means we get ENOSPC checking when writing into
7568 * holes and correct delalloc and unwritten extent mapping on filesystems that
7569 * support these features.
7571 * We are not allowed to take the i_mutex here so we have to play games to
7572 * protect against truncate races as the page could now be beyond EOF. Because
7573 * vmtruncate() writes the inode size before removing pages, once we have the
7574 * page lock we can determine safely if the page is beyond EOF. If it is not
7575 * beyond EOF, then the page is guaranteed safe against truncation until we
7578 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7580 struct page *page = vmf->page;
7581 struct inode *inode = file_inode(vma->vm_file);
7582 struct btrfs_root *root = BTRFS_I(inode)->root;
7583 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7584 struct btrfs_ordered_extent *ordered;
7585 struct extent_state *cached_state = NULL;
7587 unsigned long zero_start;
7594 sb_start_pagefault(inode->i_sb);
7595 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7597 ret = file_update_time(vma->vm_file);
7603 else /* -ENOSPC, -EIO, etc */
7604 ret = VM_FAULT_SIGBUS;
7610 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7613 size = i_size_read(inode);
7614 page_start = page_offset(page);
7615 page_end = page_start + PAGE_CACHE_SIZE - 1;
7617 if ((page->mapping != inode->i_mapping) ||
7618 (page_start >= size)) {
7619 /* page got truncated out from underneath us */
7622 wait_on_page_writeback(page);
7624 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7625 set_page_extent_mapped(page);
7628 * we can't set the delalloc bits if there are pending ordered
7629 * extents. Drop our locks and wait for them to finish
7631 ordered = btrfs_lookup_ordered_extent(inode, page_start);
7633 unlock_extent_cached(io_tree, page_start, page_end,
7634 &cached_state, GFP_NOFS);
7636 btrfs_start_ordered_extent(inode, ordered, 1);
7637 btrfs_put_ordered_extent(ordered);
7642 * XXX - page_mkwrite gets called every time the page is dirtied, even
7643 * if it was already dirty, so for space accounting reasons we need to
7644 * clear any delalloc bits for the range we are fixing to save. There
7645 * is probably a better way to do this, but for now keep consistent with
7646 * prepare_pages in the normal write path.
7648 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7649 EXTENT_DIRTY | EXTENT_DELALLOC |
7650 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7651 0, 0, &cached_state, GFP_NOFS);
7653 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7656 unlock_extent_cached(io_tree, page_start, page_end,
7657 &cached_state, GFP_NOFS);
7658 ret = VM_FAULT_SIGBUS;
7663 /* page is wholly or partially inside EOF */
7664 if (page_start + PAGE_CACHE_SIZE > size)
7665 zero_start = size & ~PAGE_CACHE_MASK;
7667 zero_start = PAGE_CACHE_SIZE;
7669 if (zero_start != PAGE_CACHE_SIZE) {
7671 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7672 flush_dcache_page(page);
7675 ClearPageChecked(page);
7676 set_page_dirty(page);
7677 SetPageUptodate(page);
7679 BTRFS_I(inode)->last_trans = root->fs_info->generation;
7680 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7681 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7683 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7687 sb_end_pagefault(inode->i_sb);
7688 return VM_FAULT_LOCKED;
7692 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7694 sb_end_pagefault(inode->i_sb);
7698 static int btrfs_truncate(struct inode *inode)
7700 struct btrfs_root *root = BTRFS_I(inode)->root;
7701 struct btrfs_block_rsv *rsv;
7704 struct btrfs_trans_handle *trans;
7705 u64 mask = root->sectorsize - 1;
7706 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7708 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
7712 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
7713 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
7716 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
7717 * 3 things going on here
7719 * 1) We need to reserve space for our orphan item and the space to
7720 * delete our orphan item. Lord knows we don't want to have a dangling
7721 * orphan item because we didn't reserve space to remove it.
7723 * 2) We need to reserve space to update our inode.
7725 * 3) We need to have something to cache all the space that is going to
7726 * be free'd up by the truncate operation, but also have some slack
7727 * space reserved in case it uses space during the truncate (thank you
7728 * very much snapshotting).
7730 * And we need these to all be seperate. The fact is we can use alot of
7731 * space doing the truncate, and we have no earthly idea how much space
7732 * we will use, so we need the truncate reservation to be seperate so it
7733 * doesn't end up using space reserved for updating the inode or
7734 * removing the orphan item. We also need to be able to stop the
7735 * transaction and start a new one, which means we need to be able to
7736 * update the inode several times, and we have no idea of knowing how
7737 * many times that will be, so we can't just reserve 1 item for the
7738 * entirety of the opration, so that has to be done seperately as well.
7739 * Then there is the orphan item, which does indeed need to be held on
7740 * to for the whole operation, and we need nobody to touch this reserved
7741 * space except the orphan code.
7743 * So that leaves us with
7745 * 1) root->orphan_block_rsv - for the orphan deletion.
7746 * 2) rsv - for the truncate reservation, which we will steal from the
7747 * transaction reservation.
7748 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7749 * updating the inode.
7751 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7754 rsv->size = min_size;
7758 * 1 for the truncate slack space
7759 * 1 for updating the inode.
7761 trans = btrfs_start_transaction(root, 2);
7762 if (IS_ERR(trans)) {
7763 err = PTR_ERR(trans);
7767 /* Migrate the slack space for the truncate to our reserve */
7768 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7773 * setattr is responsible for setting the ordered_data_close flag,
7774 * but that is only tested during the last file release. That
7775 * could happen well after the next commit, leaving a great big
7776 * window where new writes may get lost if someone chooses to write
7777 * to this file after truncating to zero
7779 * The inode doesn't have any dirty data here, and so if we commit
7780 * this is a noop. If someone immediately starts writing to the inode
7781 * it is very likely we'll catch some of their writes in this
7782 * transaction, and the commit will find this file on the ordered
7783 * data list with good things to send down.
7785 * This is a best effort solution, there is still a window where
7786 * using truncate to replace the contents of the file will
7787 * end up with a zero length file after a crash.
7789 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7790 &BTRFS_I(inode)->runtime_flags))
7791 btrfs_add_ordered_operation(trans, root, inode);
7794 * So if we truncate and then write and fsync we normally would just
7795 * write the extents that changed, which is a problem if we need to
7796 * first truncate that entire inode. So set this flag so we write out
7797 * all of the extents in the inode to the sync log so we're completely
7800 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7801 trans->block_rsv = rsv;
7804 ret = btrfs_truncate_inode_items(trans, root, inode,
7806 BTRFS_EXTENT_DATA_KEY);
7807 if (ret != -ENOSPC) {
7812 trans->block_rsv = &root->fs_info->trans_block_rsv;
7813 ret = btrfs_update_inode(trans, root, inode);
7819 btrfs_end_transaction(trans, root);
7820 btrfs_btree_balance_dirty(root);
7822 trans = btrfs_start_transaction(root, 2);
7823 if (IS_ERR(trans)) {
7824 ret = err = PTR_ERR(trans);
7829 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7831 BUG_ON(ret); /* shouldn't happen */
7832 trans->block_rsv = rsv;
7835 if (ret == 0 && inode->i_nlink > 0) {
7836 trans->block_rsv = root->orphan_block_rsv;
7837 ret = btrfs_orphan_del(trans, inode);
7843 trans->block_rsv = &root->fs_info->trans_block_rsv;
7844 ret = btrfs_update_inode(trans, root, inode);
7848 ret = btrfs_end_transaction(trans, root);
7849 btrfs_btree_balance_dirty(root);
7853 btrfs_free_block_rsv(root, rsv);
7862 * create a new subvolume directory/inode (helper for the ioctl).
7864 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7865 struct btrfs_root *new_root, u64 new_dirid)
7867 struct inode *inode;
7871 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7872 new_dirid, new_dirid,
7873 S_IFDIR | (~current_umask() & S_IRWXUGO),
7876 return PTR_ERR(inode);
7877 inode->i_op = &btrfs_dir_inode_operations;
7878 inode->i_fop = &btrfs_dir_file_operations;
7880 set_nlink(inode, 1);
7881 btrfs_i_size_write(inode, 0);
7883 err = btrfs_update_inode(trans, new_root, inode);
7889 struct inode *btrfs_alloc_inode(struct super_block *sb)
7891 struct btrfs_inode *ei;
7892 struct inode *inode;
7894 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7901 ei->last_sub_trans = 0;
7902 ei->logged_trans = 0;
7903 ei->delalloc_bytes = 0;
7904 ei->disk_i_size = 0;
7907 ei->index_cnt = (u64)-1;
7908 ei->last_unlink_trans = 0;
7909 ei->last_log_commit = 0;
7911 spin_lock_init(&ei->lock);
7912 ei->outstanding_extents = 0;
7913 ei->reserved_extents = 0;
7915 ei->runtime_flags = 0;
7916 ei->force_compress = BTRFS_COMPRESS_NONE;
7918 ei->delayed_node = NULL;
7920 inode = &ei->vfs_inode;
7921 extent_map_tree_init(&ei->extent_tree);
7922 extent_io_tree_init(&ei->io_tree, &inode->i_data);
7923 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7924 ei->io_tree.track_uptodate = 1;
7925 ei->io_failure_tree.track_uptodate = 1;
7926 atomic_set(&ei->sync_writers, 0);
7927 mutex_init(&ei->log_mutex);
7928 mutex_init(&ei->delalloc_mutex);
7929 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7930 INIT_LIST_HEAD(&ei->delalloc_inodes);
7931 INIT_LIST_HEAD(&ei->ordered_operations);
7932 RB_CLEAR_NODE(&ei->rb_node);
7937 static void btrfs_i_callback(struct rcu_head *head)
7939 struct inode *inode = container_of(head, struct inode, i_rcu);
7940 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7943 void btrfs_destroy_inode(struct inode *inode)
7945 struct btrfs_ordered_extent *ordered;
7946 struct btrfs_root *root = BTRFS_I(inode)->root;
7948 WARN_ON(!hlist_empty(&inode->i_dentry));
7949 WARN_ON(inode->i_data.nrpages);
7950 WARN_ON(BTRFS_I(inode)->outstanding_extents);
7951 WARN_ON(BTRFS_I(inode)->reserved_extents);
7952 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7953 WARN_ON(BTRFS_I(inode)->csum_bytes);
7956 * This can happen where we create an inode, but somebody else also
7957 * created the same inode and we need to destroy the one we already
7964 * Make sure we're properly removed from the ordered operation
7968 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7969 spin_lock(&root->fs_info->ordered_extent_lock);
7970 list_del_init(&BTRFS_I(inode)->ordered_operations);
7971 spin_unlock(&root->fs_info->ordered_extent_lock);
7974 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7975 &BTRFS_I(inode)->runtime_flags)) {
7976 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
7977 (unsigned long long)btrfs_ino(inode));
7978 atomic_dec(&root->orphan_inodes);
7982 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7986 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
7987 (unsigned long long)ordered->file_offset,
7988 (unsigned long long)ordered->len);
7989 btrfs_remove_ordered_extent(inode, ordered);
7990 btrfs_put_ordered_extent(ordered);
7991 btrfs_put_ordered_extent(ordered);
7994 inode_tree_del(inode);
7995 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7997 call_rcu(&inode->i_rcu, btrfs_i_callback);
8000 int btrfs_drop_inode(struct inode *inode)
8002 struct btrfs_root *root = BTRFS_I(inode)->root;
8007 /* the snap/subvol tree is on deleting */
8008 if (btrfs_root_refs(&root->root_item) == 0 &&
8009 root != root->fs_info->tree_root)
8012 return generic_drop_inode(inode);
8015 static void init_once(void *foo)
8017 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8019 inode_init_once(&ei->vfs_inode);
8022 void btrfs_destroy_cachep(void)
8025 * Make sure all delayed rcu free inodes are flushed before we
8029 if (btrfs_inode_cachep)
8030 kmem_cache_destroy(btrfs_inode_cachep);
8031 if (btrfs_trans_handle_cachep)
8032 kmem_cache_destroy(btrfs_trans_handle_cachep);
8033 if (btrfs_transaction_cachep)
8034 kmem_cache_destroy(btrfs_transaction_cachep);
8035 if (btrfs_path_cachep)
8036 kmem_cache_destroy(btrfs_path_cachep);
8037 if (btrfs_free_space_cachep)
8038 kmem_cache_destroy(btrfs_free_space_cachep);
8039 if (btrfs_delalloc_work_cachep)
8040 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8043 int btrfs_init_cachep(void)
8045 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8046 sizeof(struct btrfs_inode), 0,
8047 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8048 if (!btrfs_inode_cachep)
8051 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8052 sizeof(struct btrfs_trans_handle), 0,
8053 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8054 if (!btrfs_trans_handle_cachep)
8057 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8058 sizeof(struct btrfs_transaction), 0,
8059 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8060 if (!btrfs_transaction_cachep)
8063 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8064 sizeof(struct btrfs_path), 0,
8065 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8066 if (!btrfs_path_cachep)
8069 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8070 sizeof(struct btrfs_free_space), 0,
8071 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8072 if (!btrfs_free_space_cachep)
8075 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8076 sizeof(struct btrfs_delalloc_work), 0,
8077 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8079 if (!btrfs_delalloc_work_cachep)
8084 btrfs_destroy_cachep();
8088 static int btrfs_getattr(struct vfsmount *mnt,
8089 struct dentry *dentry, struct kstat *stat)
8092 struct inode *inode = dentry->d_inode;
8093 u32 blocksize = inode->i_sb->s_blocksize;
8095 generic_fillattr(inode, stat);
8096 stat->dev = BTRFS_I(inode)->root->anon_dev;
8097 stat->blksize = PAGE_CACHE_SIZE;
8099 spin_lock(&BTRFS_I(inode)->lock);
8100 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8101 spin_unlock(&BTRFS_I(inode)->lock);
8102 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8103 ALIGN(delalloc_bytes, blocksize)) >> 9;
8107 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8108 struct inode *new_dir, struct dentry *new_dentry)
8110 struct btrfs_trans_handle *trans;
8111 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8112 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8113 struct inode *new_inode = new_dentry->d_inode;
8114 struct inode *old_inode = old_dentry->d_inode;
8115 struct timespec ctime = CURRENT_TIME;
8119 u64 old_ino = btrfs_ino(old_inode);
8121 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8124 /* we only allow rename subvolume link between subvolumes */
8125 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8128 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8129 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8132 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8133 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8137 /* check for collisions, even if the name isn't there */
8138 ret = btrfs_check_dir_item_collision(root, new_dir->i_ino,
8139 new_dentry->d_name.name,
8140 new_dentry->d_name.len);
8143 if (ret == -EEXIST) {
8145 * eexist without a new_inode */
8151 /* maybe -EOVERFLOW */
8158 * we're using rename to replace one file with another.
8159 * and the replacement file is large. Start IO on it now so
8160 * we don't add too much work to the end of the transaction
8162 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8163 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8164 filemap_flush(old_inode->i_mapping);
8166 /* close the racy window with snapshot create/destroy ioctl */
8167 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8168 down_read(&root->fs_info->subvol_sem);
8170 * We want to reserve the absolute worst case amount of items. So if
8171 * both inodes are subvols and we need to unlink them then that would
8172 * require 4 item modifications, but if they are both normal inodes it
8173 * would require 5 item modifications, so we'll assume their normal
8174 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8175 * should cover the worst case number of items we'll modify.
8177 trans = btrfs_start_transaction(root, 11);
8178 if (IS_ERR(trans)) {
8179 ret = PTR_ERR(trans);
8184 btrfs_record_root_in_trans(trans, dest);
8186 ret = btrfs_set_inode_index(new_dir, &index);
8190 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8191 /* force full log commit if subvolume involved. */
8192 root->fs_info->last_trans_log_full_commit = trans->transid;
8194 ret = btrfs_insert_inode_ref(trans, dest,
8195 new_dentry->d_name.name,
8196 new_dentry->d_name.len,
8198 btrfs_ino(new_dir), index);
8202 * this is an ugly little race, but the rename is required
8203 * to make sure that if we crash, the inode is either at the
8204 * old name or the new one. pinning the log transaction lets
8205 * us make sure we don't allow a log commit to come in after
8206 * we unlink the name but before we add the new name back in.
8208 btrfs_pin_log_trans(root);
8211 * make sure the inode gets flushed if it is replacing
8214 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8215 btrfs_add_ordered_operation(trans, root, old_inode);
8217 inode_inc_iversion(old_dir);
8218 inode_inc_iversion(new_dir);
8219 inode_inc_iversion(old_inode);
8220 old_dir->i_ctime = old_dir->i_mtime = ctime;
8221 new_dir->i_ctime = new_dir->i_mtime = ctime;
8222 old_inode->i_ctime = ctime;
8224 if (old_dentry->d_parent != new_dentry->d_parent)
8225 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8227 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8228 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8229 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8230 old_dentry->d_name.name,
8231 old_dentry->d_name.len);
8233 ret = __btrfs_unlink_inode(trans, root, old_dir,
8234 old_dentry->d_inode,
8235 old_dentry->d_name.name,
8236 old_dentry->d_name.len);
8238 ret = btrfs_update_inode(trans, root, old_inode);
8241 btrfs_abort_transaction(trans, root, ret);
8246 inode_inc_iversion(new_inode);
8247 new_inode->i_ctime = CURRENT_TIME;
8248 if (unlikely(btrfs_ino(new_inode) ==
8249 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8250 root_objectid = BTRFS_I(new_inode)->location.objectid;
8251 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8253 new_dentry->d_name.name,
8254 new_dentry->d_name.len);
8255 BUG_ON(new_inode->i_nlink == 0);
8257 ret = btrfs_unlink_inode(trans, dest, new_dir,
8258 new_dentry->d_inode,
8259 new_dentry->d_name.name,
8260 new_dentry->d_name.len);
8262 if (!ret && new_inode->i_nlink == 0) {
8263 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8267 btrfs_abort_transaction(trans, root, ret);
8272 ret = btrfs_add_link(trans, new_dir, old_inode,
8273 new_dentry->d_name.name,
8274 new_dentry->d_name.len, 0, index);
8276 btrfs_abort_transaction(trans, root, ret);
8280 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8281 struct dentry *parent = new_dentry->d_parent;
8282 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8283 btrfs_end_log_trans(root);
8286 btrfs_end_transaction(trans, root);
8288 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8289 up_read(&root->fs_info->subvol_sem);
8294 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8296 struct btrfs_delalloc_work *delalloc_work;
8298 delalloc_work = container_of(work, struct btrfs_delalloc_work,
8300 if (delalloc_work->wait)
8301 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
8303 filemap_flush(delalloc_work->inode->i_mapping);
8305 if (delalloc_work->delay_iput)
8306 btrfs_add_delayed_iput(delalloc_work->inode);
8308 iput(delalloc_work->inode);
8309 complete(&delalloc_work->completion);
8312 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8313 int wait, int delay_iput)
8315 struct btrfs_delalloc_work *work;
8317 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8321 init_completion(&work->completion);
8322 INIT_LIST_HEAD(&work->list);
8323 work->inode = inode;
8325 work->delay_iput = delay_iput;
8326 work->work.func = btrfs_run_delalloc_work;
8331 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8333 wait_for_completion(&work->completion);
8334 kmem_cache_free(btrfs_delalloc_work_cachep, work);
8338 * some fairly slow code that needs optimization. This walks the list
8339 * of all the inodes with pending delalloc and forces them to disk.
8341 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8343 struct btrfs_inode *binode;
8344 struct inode *inode;
8345 struct btrfs_delalloc_work *work, *next;
8346 struct list_head works;
8347 struct list_head splice;
8350 if (root->fs_info->sb->s_flags & MS_RDONLY)
8353 INIT_LIST_HEAD(&works);
8354 INIT_LIST_HEAD(&splice);
8356 spin_lock(&root->fs_info->delalloc_lock);
8357 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
8358 while (!list_empty(&splice)) {
8359 binode = list_entry(splice.next, struct btrfs_inode,
8362 list_del_init(&binode->delalloc_inodes);
8364 inode = igrab(&binode->vfs_inode);
8366 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
8367 &binode->runtime_flags);
8371 list_add_tail(&binode->delalloc_inodes,
8372 &root->fs_info->delalloc_inodes);
8373 spin_unlock(&root->fs_info->delalloc_lock);
8375 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8376 if (unlikely(!work)) {
8380 list_add_tail(&work->list, &works);
8381 btrfs_queue_worker(&root->fs_info->flush_workers,
8385 spin_lock(&root->fs_info->delalloc_lock);
8387 spin_unlock(&root->fs_info->delalloc_lock);
8389 list_for_each_entry_safe(work, next, &works, list) {
8390 list_del_init(&work->list);
8391 btrfs_wait_and_free_delalloc_work(work);
8394 /* the filemap_flush will queue IO into the worker threads, but
8395 * we have to make sure the IO is actually started and that
8396 * ordered extents get created before we return
8398 atomic_inc(&root->fs_info->async_submit_draining);
8399 while (atomic_read(&root->fs_info->nr_async_submits) ||
8400 atomic_read(&root->fs_info->async_delalloc_pages)) {
8401 wait_event(root->fs_info->async_submit_wait,
8402 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8403 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8405 atomic_dec(&root->fs_info->async_submit_draining);
8408 list_for_each_entry_safe(work, next, &works, list) {
8409 list_del_init(&work->list);
8410 btrfs_wait_and_free_delalloc_work(work);
8413 if (!list_empty_careful(&splice)) {
8414 spin_lock(&root->fs_info->delalloc_lock);
8415 list_splice_tail(&splice, &root->fs_info->delalloc_inodes);
8416 spin_unlock(&root->fs_info->delalloc_lock);
8421 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8422 const char *symname)
8424 struct btrfs_trans_handle *trans;
8425 struct btrfs_root *root = BTRFS_I(dir)->root;
8426 struct btrfs_path *path;
8427 struct btrfs_key key;
8428 struct inode *inode = NULL;
8436 struct btrfs_file_extent_item *ei;
8437 struct extent_buffer *leaf;
8439 name_len = strlen(symname) + 1;
8440 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8441 return -ENAMETOOLONG;
8444 * 2 items for inode item and ref
8445 * 2 items for dir items
8446 * 1 item for xattr if selinux is on
8448 trans = btrfs_start_transaction(root, 5);
8450 return PTR_ERR(trans);
8452 err = btrfs_find_free_ino(root, &objectid);
8456 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8457 dentry->d_name.len, btrfs_ino(dir), objectid,
8458 S_IFLNK|S_IRWXUGO, &index);
8459 if (IS_ERR(inode)) {
8460 err = PTR_ERR(inode);
8464 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8471 * If the active LSM wants to access the inode during
8472 * d_instantiate it needs these. Smack checks to see
8473 * if the filesystem supports xattrs by looking at the
8476 inode->i_fop = &btrfs_file_operations;
8477 inode->i_op = &btrfs_file_inode_operations;
8479 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8483 inode->i_mapping->a_ops = &btrfs_aops;
8484 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8485 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8490 path = btrfs_alloc_path();
8496 key.objectid = btrfs_ino(inode);
8498 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8499 datasize = btrfs_file_extent_calc_inline_size(name_len);
8500 err = btrfs_insert_empty_item(trans, root, path, &key,
8504 btrfs_free_path(path);
8507 leaf = path->nodes[0];
8508 ei = btrfs_item_ptr(leaf, path->slots[0],
8509 struct btrfs_file_extent_item);
8510 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8511 btrfs_set_file_extent_type(leaf, ei,
8512 BTRFS_FILE_EXTENT_INLINE);
8513 btrfs_set_file_extent_encryption(leaf, ei, 0);
8514 btrfs_set_file_extent_compression(leaf, ei, 0);
8515 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8516 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8518 ptr = btrfs_file_extent_inline_start(ei);
8519 write_extent_buffer(leaf, symname, ptr, name_len);
8520 btrfs_mark_buffer_dirty(leaf);
8521 btrfs_free_path(path);
8523 inode->i_op = &btrfs_symlink_inode_operations;
8524 inode->i_mapping->a_ops = &btrfs_symlink_aops;
8525 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8526 inode_set_bytes(inode, name_len);
8527 btrfs_i_size_write(inode, name_len - 1);
8528 err = btrfs_update_inode(trans, root, inode);
8534 d_instantiate(dentry, inode);
8535 btrfs_end_transaction(trans, root);
8537 inode_dec_link_count(inode);
8540 btrfs_btree_balance_dirty(root);
8544 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8545 u64 start, u64 num_bytes, u64 min_size,
8546 loff_t actual_len, u64 *alloc_hint,
8547 struct btrfs_trans_handle *trans)
8549 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8550 struct extent_map *em;
8551 struct btrfs_root *root = BTRFS_I(inode)->root;
8552 struct btrfs_key ins;
8553 u64 cur_offset = start;
8557 bool own_trans = true;
8561 while (num_bytes > 0) {
8563 trans = btrfs_start_transaction(root, 3);
8564 if (IS_ERR(trans)) {
8565 ret = PTR_ERR(trans);
8570 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8571 cur_bytes = max(cur_bytes, min_size);
8572 ret = btrfs_reserve_extent(trans, root, cur_bytes,
8573 min_size, 0, *alloc_hint, &ins, 1);
8576 btrfs_end_transaction(trans, root);
8580 ret = insert_reserved_file_extent(trans, inode,
8581 cur_offset, ins.objectid,
8582 ins.offset, ins.offset,
8583 ins.offset, 0, 0, 0,
8584 BTRFS_FILE_EXTENT_PREALLOC);
8586 btrfs_abort_transaction(trans, root, ret);
8588 btrfs_end_transaction(trans, root);
8591 btrfs_drop_extent_cache(inode, cur_offset,
8592 cur_offset + ins.offset -1, 0);
8594 em = alloc_extent_map();
8596 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8597 &BTRFS_I(inode)->runtime_flags);
8601 em->start = cur_offset;
8602 em->orig_start = cur_offset;
8603 em->len = ins.offset;
8604 em->block_start = ins.objectid;
8605 em->block_len = ins.offset;
8606 em->orig_block_len = ins.offset;
8607 em->ram_bytes = ins.offset;
8608 em->bdev = root->fs_info->fs_devices->latest_bdev;
8609 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8610 em->generation = trans->transid;
8613 write_lock(&em_tree->lock);
8614 ret = add_extent_mapping(em_tree, em, 1);
8615 write_unlock(&em_tree->lock);
8618 btrfs_drop_extent_cache(inode, cur_offset,
8619 cur_offset + ins.offset - 1,
8622 free_extent_map(em);
8624 num_bytes -= ins.offset;
8625 cur_offset += ins.offset;
8626 *alloc_hint = ins.objectid + ins.offset;
8628 inode_inc_iversion(inode);
8629 inode->i_ctime = CURRENT_TIME;
8630 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8631 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8632 (actual_len > inode->i_size) &&
8633 (cur_offset > inode->i_size)) {
8634 if (cur_offset > actual_len)
8635 i_size = actual_len;
8637 i_size = cur_offset;
8638 i_size_write(inode, i_size);
8639 btrfs_ordered_update_i_size(inode, i_size, NULL);
8642 ret = btrfs_update_inode(trans, root, inode);
8645 btrfs_abort_transaction(trans, root, ret);
8647 btrfs_end_transaction(trans, root);
8652 btrfs_end_transaction(trans, root);
8657 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8658 u64 start, u64 num_bytes, u64 min_size,
8659 loff_t actual_len, u64 *alloc_hint)
8661 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8662 min_size, actual_len, alloc_hint,
8666 int btrfs_prealloc_file_range_trans(struct inode *inode,
8667 struct btrfs_trans_handle *trans, int mode,
8668 u64 start, u64 num_bytes, u64 min_size,
8669 loff_t actual_len, u64 *alloc_hint)
8671 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8672 min_size, actual_len, alloc_hint, trans);
8675 static int btrfs_set_page_dirty(struct page *page)
8677 return __set_page_dirty_nobuffers(page);
8680 static int btrfs_permission(struct inode *inode, int mask)
8682 struct btrfs_root *root = BTRFS_I(inode)->root;
8683 umode_t mode = inode->i_mode;
8685 if (mask & MAY_WRITE &&
8686 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8687 if (btrfs_root_readonly(root))
8689 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8692 return generic_permission(inode, mask);
8695 static const struct inode_operations btrfs_dir_inode_operations = {
8696 .getattr = btrfs_getattr,
8697 .lookup = btrfs_lookup,
8698 .create = btrfs_create,
8699 .unlink = btrfs_unlink,
8701 .mkdir = btrfs_mkdir,
8702 .rmdir = btrfs_rmdir,
8703 .rename = btrfs_rename,
8704 .symlink = btrfs_symlink,
8705 .setattr = btrfs_setattr,
8706 .mknod = btrfs_mknod,
8707 .setxattr = btrfs_setxattr,
8708 .getxattr = btrfs_getxattr,
8709 .listxattr = btrfs_listxattr,
8710 .removexattr = btrfs_removexattr,
8711 .permission = btrfs_permission,
8712 .get_acl = btrfs_get_acl,
8714 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8715 .lookup = btrfs_lookup,
8716 .permission = btrfs_permission,
8717 .get_acl = btrfs_get_acl,
8720 static const struct file_operations btrfs_dir_file_operations = {
8721 .llseek = generic_file_llseek,
8722 .read = generic_read_dir,
8723 .readdir = btrfs_real_readdir,
8724 .unlocked_ioctl = btrfs_ioctl,
8725 #ifdef CONFIG_COMPAT
8726 .compat_ioctl = btrfs_ioctl,
8728 .release = btrfs_release_file,
8729 .fsync = btrfs_sync_file,
8732 static struct extent_io_ops btrfs_extent_io_ops = {
8733 .fill_delalloc = run_delalloc_range,
8734 .submit_bio_hook = btrfs_submit_bio_hook,
8735 .merge_bio_hook = btrfs_merge_bio_hook,
8736 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
8737 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
8738 .writepage_start_hook = btrfs_writepage_start_hook,
8739 .set_bit_hook = btrfs_set_bit_hook,
8740 .clear_bit_hook = btrfs_clear_bit_hook,
8741 .merge_extent_hook = btrfs_merge_extent_hook,
8742 .split_extent_hook = btrfs_split_extent_hook,
8746 * btrfs doesn't support the bmap operation because swapfiles
8747 * use bmap to make a mapping of extents in the file. They assume
8748 * these extents won't change over the life of the file and they
8749 * use the bmap result to do IO directly to the drive.
8751 * the btrfs bmap call would return logical addresses that aren't
8752 * suitable for IO and they also will change frequently as COW
8753 * operations happen. So, swapfile + btrfs == corruption.
8755 * For now we're avoiding this by dropping bmap.
8757 static const struct address_space_operations btrfs_aops = {
8758 .readpage = btrfs_readpage,
8759 .writepage = btrfs_writepage,
8760 .writepages = btrfs_writepages,
8761 .readpages = btrfs_readpages,
8762 .direct_IO = btrfs_direct_IO,
8763 .invalidatepage = btrfs_invalidatepage,
8764 .releasepage = btrfs_releasepage,
8765 .set_page_dirty = btrfs_set_page_dirty,
8766 .error_remove_page = generic_error_remove_page,
8769 static const struct address_space_operations btrfs_symlink_aops = {
8770 .readpage = btrfs_readpage,
8771 .writepage = btrfs_writepage,
8772 .invalidatepage = btrfs_invalidatepage,
8773 .releasepage = btrfs_releasepage,
8776 static const struct inode_operations btrfs_file_inode_operations = {
8777 .getattr = btrfs_getattr,
8778 .setattr = btrfs_setattr,
8779 .setxattr = btrfs_setxattr,
8780 .getxattr = btrfs_getxattr,
8781 .listxattr = btrfs_listxattr,
8782 .removexattr = btrfs_removexattr,
8783 .permission = btrfs_permission,
8784 .fiemap = btrfs_fiemap,
8785 .get_acl = btrfs_get_acl,
8786 .update_time = btrfs_update_time,
8788 static const struct inode_operations btrfs_special_inode_operations = {
8789 .getattr = btrfs_getattr,
8790 .setattr = btrfs_setattr,
8791 .permission = btrfs_permission,
8792 .setxattr = btrfs_setxattr,
8793 .getxattr = btrfs_getxattr,
8794 .listxattr = btrfs_listxattr,
8795 .removexattr = btrfs_removexattr,
8796 .get_acl = btrfs_get_acl,
8797 .update_time = btrfs_update_time,
8799 static const struct inode_operations btrfs_symlink_inode_operations = {
8800 .readlink = generic_readlink,
8801 .follow_link = page_follow_link_light,
8802 .put_link = page_put_link,
8803 .getattr = btrfs_getattr,
8804 .setattr = btrfs_setattr,
8805 .permission = btrfs_permission,
8806 .setxattr = btrfs_setxattr,
8807 .getxattr = btrfs_getxattr,
8808 .listxattr = btrfs_listxattr,
8809 .removexattr = btrfs_removexattr,
8810 .get_acl = btrfs_get_acl,
8811 .update_time = btrfs_update_time,
8814 const struct dentry_operations btrfs_dentry_operations = {
8815 .d_delete = btrfs_dentry_delete,
8816 .d_release = btrfs_dentry_release,