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, int type);
105 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
106 struct inode *inode, struct inode *dir,
107 const struct qstr *qstr)
111 err = btrfs_init_acl(trans, inode, dir);
113 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
118 * this does all the hard work for inserting an inline extent into
119 * the btree. The caller should have done a btrfs_drop_extents so that
120 * no overlapping inline items exist in the btree
122 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
123 struct btrfs_root *root, struct inode *inode,
124 u64 start, size_t size, size_t compressed_size,
126 struct page **compressed_pages)
128 struct btrfs_key key;
129 struct btrfs_path *path;
130 struct extent_buffer *leaf;
131 struct page *page = NULL;
134 struct btrfs_file_extent_item *ei;
137 size_t cur_size = size;
139 unsigned long offset;
141 if (compressed_size && compressed_pages)
142 cur_size = compressed_size;
144 path = btrfs_alloc_path();
148 path->leave_spinning = 1;
150 key.objectid = btrfs_ino(inode);
152 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
153 datasize = btrfs_file_extent_calc_inline_size(cur_size);
155 inode_add_bytes(inode, size);
156 ret = btrfs_insert_empty_item(trans, root, path, &key,
162 leaf = path->nodes[0];
163 ei = btrfs_item_ptr(leaf, path->slots[0],
164 struct btrfs_file_extent_item);
165 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
166 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
167 btrfs_set_file_extent_encryption(leaf, ei, 0);
168 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
169 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
170 ptr = btrfs_file_extent_inline_start(ei);
172 if (compress_type != BTRFS_COMPRESS_NONE) {
175 while (compressed_size > 0) {
176 cpage = compressed_pages[i];
177 cur_size = min_t(unsigned long, compressed_size,
180 kaddr = kmap_atomic(cpage);
181 write_extent_buffer(leaf, kaddr, ptr, cur_size);
182 kunmap_atomic(kaddr);
186 compressed_size -= cur_size;
188 btrfs_set_file_extent_compression(leaf, ei,
191 page = find_get_page(inode->i_mapping,
192 start >> PAGE_CACHE_SHIFT);
193 btrfs_set_file_extent_compression(leaf, ei, 0);
194 kaddr = kmap_atomic(page);
195 offset = start & (PAGE_CACHE_SIZE - 1);
196 write_extent_buffer(leaf, kaddr + offset, ptr, size);
197 kunmap_atomic(kaddr);
198 page_cache_release(page);
200 btrfs_mark_buffer_dirty(leaf);
201 btrfs_free_path(path);
204 * we're an inline extent, so nobody can
205 * extend the file past i_size without locking
206 * a page we already have locked.
208 * We must do any isize and inode updates
209 * before we unlock the pages. Otherwise we
210 * could end up racing with unlink.
212 BTRFS_I(inode)->disk_i_size = inode->i_size;
213 ret = btrfs_update_inode(trans, root, inode);
217 btrfs_free_path(path);
223 * conditionally insert an inline extent into the file. This
224 * does the checks required to make sure the data is small enough
225 * to fit as an inline extent.
227 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
228 struct btrfs_root *root,
229 struct inode *inode, u64 start, u64 end,
230 size_t compressed_size, int compress_type,
231 struct page **compressed_pages)
233 u64 isize = i_size_read(inode);
234 u64 actual_end = min(end + 1, isize);
235 u64 inline_len = actual_end - start;
236 u64 aligned_end = ALIGN(end, root->sectorsize);
237 u64 data_len = inline_len;
241 data_len = compressed_size;
244 actual_end >= PAGE_CACHE_SIZE ||
245 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
247 (actual_end & (root->sectorsize - 1)) == 0) ||
249 data_len > root->fs_info->max_inline) {
253 ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
257 if (isize > actual_end)
258 inline_len = min_t(u64, isize, actual_end);
259 ret = insert_inline_extent(trans, root, inode, start,
260 inline_len, compressed_size,
261 compress_type, compressed_pages);
262 if (ret && ret != -ENOSPC) {
263 btrfs_abort_transaction(trans, root, ret);
265 } else if (ret == -ENOSPC) {
269 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
270 btrfs_delalloc_release_metadata(inode, end + 1 - start);
271 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
275 struct async_extent {
280 unsigned long nr_pages;
282 struct list_head list;
287 struct btrfs_root *root;
288 struct page *locked_page;
291 struct list_head extents;
292 struct btrfs_work work;
295 static noinline int add_async_extent(struct async_cow *cow,
296 u64 start, u64 ram_size,
299 unsigned long nr_pages,
302 struct async_extent *async_extent;
304 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
305 BUG_ON(!async_extent); /* -ENOMEM */
306 async_extent->start = start;
307 async_extent->ram_size = ram_size;
308 async_extent->compressed_size = compressed_size;
309 async_extent->pages = pages;
310 async_extent->nr_pages = nr_pages;
311 async_extent->compress_type = compress_type;
312 list_add_tail(&async_extent->list, &cow->extents);
317 * we create compressed extents in two phases. The first
318 * phase compresses a range of pages that have already been
319 * locked (both pages and state bits are locked).
321 * This is done inside an ordered work queue, and the compression
322 * is spread across many cpus. The actual IO submission is step
323 * two, and the ordered work queue takes care of making sure that
324 * happens in the same order things were put onto the queue by
325 * writepages and friends.
327 * If this code finds it can't get good compression, it puts an
328 * entry onto the work queue to write the uncompressed bytes. This
329 * makes sure that both compressed inodes and uncompressed inodes
330 * are written in the same order that the flusher thread sent them
333 static noinline int compress_file_range(struct inode *inode,
334 struct page *locked_page,
336 struct async_cow *async_cow,
339 struct btrfs_root *root = BTRFS_I(inode)->root;
340 struct btrfs_trans_handle *trans;
342 u64 blocksize = root->sectorsize;
344 u64 isize = i_size_read(inode);
346 struct page **pages = NULL;
347 unsigned long nr_pages;
348 unsigned long nr_pages_ret = 0;
349 unsigned long total_compressed = 0;
350 unsigned long total_in = 0;
351 unsigned long max_compressed = 128 * 1024;
352 unsigned long max_uncompressed = 128 * 1024;
355 int compress_type = root->fs_info->compress_type;
358 /* if this is a small write inside eof, kick off a defrag */
359 if ((end - start + 1) < 16 * 1024 &&
360 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
361 btrfs_add_inode_defrag(NULL, inode);
363 actual_end = min_t(u64, isize, end + 1);
366 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
367 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
370 * we don't want to send crud past the end of i_size through
371 * compression, that's just a waste of CPU time. So, if the
372 * end of the file is before the start of our current
373 * requested range of bytes, we bail out to the uncompressed
374 * cleanup code that can deal with all of this.
376 * It isn't really the fastest way to fix things, but this is a
377 * very uncommon corner.
379 if (actual_end <= start)
380 goto cleanup_and_bail_uncompressed;
382 total_compressed = actual_end - start;
384 /* we want to make sure that amount of ram required to uncompress
385 * an extent is reasonable, so we limit the total size in ram
386 * of a compressed extent to 128k. This is a crucial number
387 * because it also controls how easily we can spread reads across
388 * cpus for decompression.
390 * We also want to make sure the amount of IO required to do
391 * a random read is reasonably small, so we limit the size of
392 * a compressed extent to 128k.
394 total_compressed = min(total_compressed, max_uncompressed);
395 num_bytes = ALIGN(end - start + 1, blocksize);
396 num_bytes = max(blocksize, num_bytes);
401 * we do compression for mount -o compress and when the
402 * inode has not been flagged as nocompress. This flag can
403 * change at any time if we discover bad compression ratios.
405 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
406 (btrfs_test_opt(root, COMPRESS) ||
407 (BTRFS_I(inode)->force_compress) ||
408 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
410 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
412 /* just bail out to the uncompressed code */
416 if (BTRFS_I(inode)->force_compress)
417 compress_type = BTRFS_I(inode)->force_compress;
420 * we need to call clear_page_dirty_for_io on each
421 * page in the range. Otherwise applications with the file
422 * mmap'd can wander in and change the page contents while
423 * we are compressing them.
425 * If the compression fails for any reason, we set the pages
426 * dirty again later on.
428 extent_range_clear_dirty_for_io(inode, start, end);
430 ret = btrfs_compress_pages(compress_type,
431 inode->i_mapping, start,
432 total_compressed, pages,
433 nr_pages, &nr_pages_ret,
439 unsigned long offset = total_compressed &
440 (PAGE_CACHE_SIZE - 1);
441 struct page *page = pages[nr_pages_ret - 1];
444 /* zero the tail end of the last page, we might be
445 * sending it down to disk
448 kaddr = kmap_atomic(page);
449 memset(kaddr + offset, 0,
450 PAGE_CACHE_SIZE - offset);
451 kunmap_atomic(kaddr);
458 trans = btrfs_join_transaction(root);
460 ret = PTR_ERR(trans);
462 goto cleanup_and_out;
464 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
466 /* lets try to make an inline extent */
467 if (ret || total_in < (actual_end - start)) {
468 /* we didn't compress the entire range, try
469 * to make an uncompressed inline extent.
471 ret = cow_file_range_inline(trans, root, inode,
472 start, end, 0, 0, NULL);
474 /* try making a compressed inline extent */
475 ret = cow_file_range_inline(trans, root, inode,
478 compress_type, pages);
482 * inline extent creation worked or returned error,
483 * we don't need to create any more async work items.
484 * Unlock and free up our temp pages.
486 extent_clear_unlock_delalloc(inode,
487 &BTRFS_I(inode)->io_tree,
489 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
490 EXTENT_CLEAR_DELALLOC |
491 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
493 btrfs_end_transaction(trans, root);
496 btrfs_end_transaction(trans, root);
501 * we aren't doing an inline extent round the compressed size
502 * up to a block size boundary so the allocator does sane
505 total_compressed = ALIGN(total_compressed, blocksize);
508 * one last check to make sure the compression is really a
509 * win, compare the page count read with the blocks on disk
511 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
512 if (total_compressed >= total_in) {
515 num_bytes = total_in;
518 if (!will_compress && pages) {
520 * the compression code ran but failed to make things smaller,
521 * free any pages it allocated and our page pointer array
523 for (i = 0; i < nr_pages_ret; i++) {
524 WARN_ON(pages[i]->mapping);
525 page_cache_release(pages[i]);
529 total_compressed = 0;
532 /* flag the file so we don't compress in the future */
533 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
534 !(BTRFS_I(inode)->force_compress)) {
535 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
541 /* the async work queues will take care of doing actual
542 * allocation on disk for these compressed pages,
543 * and will submit them to the elevator.
545 add_async_extent(async_cow, start, num_bytes,
546 total_compressed, pages, nr_pages_ret,
549 if (start + num_bytes < end) {
556 cleanup_and_bail_uncompressed:
558 * No compression, but we still need to write the pages in
559 * the file we've been given so far. redirty the locked
560 * page if it corresponds to our extent and set things up
561 * for the async work queue to run cow_file_range to do
562 * the normal delalloc dance
564 if (page_offset(locked_page) >= start &&
565 page_offset(locked_page) <= end) {
566 __set_page_dirty_nobuffers(locked_page);
567 /* unlocked later on in the async handlers */
570 extent_range_redirty_for_io(inode, start, end);
571 add_async_extent(async_cow, start, end - start + 1,
572 0, NULL, 0, BTRFS_COMPRESS_NONE);
580 for (i = 0; i < nr_pages_ret; i++) {
581 WARN_ON(pages[i]->mapping);
582 page_cache_release(pages[i]);
589 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
591 EXTENT_CLEAR_UNLOCK_PAGE |
593 EXTENT_CLEAR_DELALLOC |
594 EXTENT_SET_WRITEBACK |
595 EXTENT_END_WRITEBACK);
596 if (!trans || IS_ERR(trans))
597 btrfs_error(root->fs_info, ret, "Failed to join transaction");
599 btrfs_abort_transaction(trans, root, ret);
604 * phase two of compressed writeback. This is the ordered portion
605 * of the code, which only gets called in the order the work was
606 * queued. We walk all the async extents created by compress_file_range
607 * and send them down to the disk.
609 static noinline int submit_compressed_extents(struct inode *inode,
610 struct async_cow *async_cow)
612 struct async_extent *async_extent;
614 struct btrfs_trans_handle *trans;
615 struct btrfs_key ins;
616 struct extent_map *em;
617 struct btrfs_root *root = BTRFS_I(inode)->root;
618 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
619 struct extent_io_tree *io_tree;
622 if (list_empty(&async_cow->extents))
626 while (!list_empty(&async_cow->extents)) {
627 async_extent = list_entry(async_cow->extents.next,
628 struct async_extent, list);
629 list_del(&async_extent->list);
631 io_tree = &BTRFS_I(inode)->io_tree;
634 /* did the compression code fall back to uncompressed IO? */
635 if (!async_extent->pages) {
636 int page_started = 0;
637 unsigned long nr_written = 0;
639 lock_extent(io_tree, async_extent->start,
640 async_extent->start +
641 async_extent->ram_size - 1);
643 /* allocate blocks */
644 ret = cow_file_range(inode, async_cow->locked_page,
646 async_extent->start +
647 async_extent->ram_size - 1,
648 &page_started, &nr_written, 0);
653 * if page_started, cow_file_range inserted an
654 * inline extent and took care of all the unlocking
655 * and IO for us. Otherwise, we need to submit
656 * all those pages down to the drive.
658 if (!page_started && !ret)
659 extent_write_locked_range(io_tree,
660 inode, async_extent->start,
661 async_extent->start +
662 async_extent->ram_size - 1,
666 unlock_page(async_cow->locked_page);
672 lock_extent(io_tree, async_extent->start,
673 async_extent->start + async_extent->ram_size - 1);
675 trans = btrfs_join_transaction(root);
677 ret = PTR_ERR(trans);
679 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
680 ret = btrfs_reserve_extent(trans, root,
681 async_extent->compressed_size,
682 async_extent->compressed_size,
683 0, alloc_hint, &ins, 1);
684 if (ret && ret != -ENOSPC)
685 btrfs_abort_transaction(trans, root, ret);
686 btrfs_end_transaction(trans, root);
692 for (i = 0; i < async_extent->nr_pages; i++) {
693 WARN_ON(async_extent->pages[i]->mapping);
694 page_cache_release(async_extent->pages[i]);
696 kfree(async_extent->pages);
697 async_extent->nr_pages = 0;
698 async_extent->pages = NULL;
706 * here we're doing allocation and writeback of the
709 btrfs_drop_extent_cache(inode, async_extent->start,
710 async_extent->start +
711 async_extent->ram_size - 1, 0);
713 em = alloc_extent_map();
715 goto out_free_reserve;
716 em->start = async_extent->start;
717 em->len = async_extent->ram_size;
718 em->orig_start = em->start;
719 em->mod_start = em->start;
720 em->mod_len = em->len;
722 em->block_start = ins.objectid;
723 em->block_len = ins.offset;
724 em->orig_block_len = ins.offset;
725 em->bdev = root->fs_info->fs_devices->latest_bdev;
726 em->compress_type = async_extent->compress_type;
727 set_bit(EXTENT_FLAG_PINNED, &em->flags);
728 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
732 write_lock(&em_tree->lock);
733 ret = add_extent_mapping(em_tree, em);
736 &em_tree->modified_extents);
737 write_unlock(&em_tree->lock);
738 if (ret != -EEXIST) {
742 btrfs_drop_extent_cache(inode, async_extent->start,
743 async_extent->start +
744 async_extent->ram_size - 1, 0);
748 goto out_free_reserve;
750 ret = btrfs_add_ordered_extent_compress(inode,
753 async_extent->ram_size,
755 BTRFS_ORDERED_COMPRESSED,
756 async_extent->compress_type);
758 goto out_free_reserve;
761 * clear dirty, set writeback and unlock the pages.
763 extent_clear_unlock_delalloc(inode,
764 &BTRFS_I(inode)->io_tree,
766 async_extent->start +
767 async_extent->ram_size - 1,
768 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
769 EXTENT_CLEAR_UNLOCK |
770 EXTENT_CLEAR_DELALLOC |
771 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
773 ret = btrfs_submit_compressed_write(inode,
775 async_extent->ram_size,
777 ins.offset, async_extent->pages,
778 async_extent->nr_pages);
779 alloc_hint = ins.objectid + ins.offset;
789 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
791 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
793 async_extent->start +
794 async_extent->ram_size - 1,
795 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
796 EXTENT_CLEAR_UNLOCK |
797 EXTENT_CLEAR_DELALLOC |
799 EXTENT_SET_WRITEBACK |
800 EXTENT_END_WRITEBACK);
805 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
808 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
809 struct extent_map *em;
812 read_lock(&em_tree->lock);
813 em = search_extent_mapping(em_tree, start, num_bytes);
816 * if block start isn't an actual block number then find the
817 * first block in this inode and use that as a hint. If that
818 * block is also bogus then just don't worry about it.
820 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
822 em = search_extent_mapping(em_tree, 0, 0);
823 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
824 alloc_hint = em->block_start;
828 alloc_hint = em->block_start;
832 read_unlock(&em_tree->lock);
838 * when extent_io.c finds a delayed allocation range in the file,
839 * the call backs end up in this code. The basic idea is to
840 * allocate extents on disk for the range, and create ordered data structs
841 * in ram to track those extents.
843 * locked_page is the page that writepage had locked already. We use
844 * it to make sure we don't do extra locks or unlocks.
846 * *page_started is set to one if we unlock locked_page and do everything
847 * required to start IO on it. It may be clean and already done with
850 static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
852 struct btrfs_root *root,
853 struct page *locked_page,
854 u64 start, u64 end, int *page_started,
855 unsigned long *nr_written,
860 unsigned long ram_size;
863 u64 blocksize = root->sectorsize;
864 struct btrfs_key ins;
865 struct extent_map *em;
866 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
869 BUG_ON(btrfs_is_free_space_inode(inode));
871 num_bytes = ALIGN(end - start + 1, blocksize);
872 num_bytes = max(blocksize, num_bytes);
873 disk_num_bytes = num_bytes;
875 /* if this is a small write inside eof, kick off defrag */
876 if (num_bytes < 64 * 1024 &&
877 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
878 btrfs_add_inode_defrag(trans, inode);
881 /* lets try to make an inline extent */
882 ret = cow_file_range_inline(trans, root, inode,
883 start, end, 0, 0, NULL);
885 extent_clear_unlock_delalloc(inode,
886 &BTRFS_I(inode)->io_tree,
888 EXTENT_CLEAR_UNLOCK_PAGE |
889 EXTENT_CLEAR_UNLOCK |
890 EXTENT_CLEAR_DELALLOC |
892 EXTENT_SET_WRITEBACK |
893 EXTENT_END_WRITEBACK);
895 *nr_written = *nr_written +
896 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
899 } else if (ret < 0) {
900 btrfs_abort_transaction(trans, root, ret);
905 BUG_ON(disk_num_bytes >
906 btrfs_super_total_bytes(root->fs_info->super_copy));
908 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
909 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
911 while (disk_num_bytes > 0) {
914 cur_alloc_size = disk_num_bytes;
915 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
916 root->sectorsize, 0, alloc_hint,
919 btrfs_abort_transaction(trans, root, ret);
923 em = alloc_extent_map();
924 BUG_ON(!em); /* -ENOMEM */
926 em->orig_start = em->start;
927 ram_size = ins.offset;
928 em->len = ins.offset;
929 em->mod_start = em->start;
930 em->mod_len = em->len;
932 em->block_start = ins.objectid;
933 em->block_len = ins.offset;
934 em->orig_block_len = ins.offset;
935 em->bdev = root->fs_info->fs_devices->latest_bdev;
936 set_bit(EXTENT_FLAG_PINNED, &em->flags);
940 write_lock(&em_tree->lock);
941 ret = add_extent_mapping(em_tree, em);
944 &em_tree->modified_extents);
945 write_unlock(&em_tree->lock);
946 if (ret != -EEXIST) {
950 btrfs_drop_extent_cache(inode, start,
951 start + ram_size - 1, 0);
954 cur_alloc_size = ins.offset;
955 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
956 ram_size, cur_alloc_size, 0);
957 BUG_ON(ret); /* -ENOMEM */
959 if (root->root_key.objectid ==
960 BTRFS_DATA_RELOC_TREE_OBJECTID) {
961 ret = btrfs_reloc_clone_csums(inode, start,
964 btrfs_abort_transaction(trans, root, ret);
969 if (disk_num_bytes < cur_alloc_size)
972 /* we're not doing compressed IO, don't unlock the first
973 * page (which the caller expects to stay locked), don't
974 * clear any dirty bits and don't set any writeback bits
976 * Do set the Private2 bit so we know this page was properly
977 * setup for writepage
979 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
980 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
983 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
984 start, start + ram_size - 1,
986 disk_num_bytes -= cur_alloc_size;
987 num_bytes -= cur_alloc_size;
988 alloc_hint = ins.objectid + ins.offset;
989 start += cur_alloc_size;
995 extent_clear_unlock_delalloc(inode,
996 &BTRFS_I(inode)->io_tree,
997 start, end, locked_page,
998 EXTENT_CLEAR_UNLOCK_PAGE |
999 EXTENT_CLEAR_UNLOCK |
1000 EXTENT_CLEAR_DELALLOC |
1001 EXTENT_CLEAR_DIRTY |
1002 EXTENT_SET_WRITEBACK |
1003 EXTENT_END_WRITEBACK);
1008 static noinline int cow_file_range(struct inode *inode,
1009 struct page *locked_page,
1010 u64 start, u64 end, int *page_started,
1011 unsigned long *nr_written,
1014 struct btrfs_trans_handle *trans;
1015 struct btrfs_root *root = BTRFS_I(inode)->root;
1018 trans = btrfs_join_transaction(root);
1019 if (IS_ERR(trans)) {
1020 extent_clear_unlock_delalloc(inode,
1021 &BTRFS_I(inode)->io_tree,
1022 start, end, locked_page,
1023 EXTENT_CLEAR_UNLOCK_PAGE |
1024 EXTENT_CLEAR_UNLOCK |
1025 EXTENT_CLEAR_DELALLOC |
1026 EXTENT_CLEAR_DIRTY |
1027 EXTENT_SET_WRITEBACK |
1028 EXTENT_END_WRITEBACK);
1029 return PTR_ERR(trans);
1031 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1033 ret = __cow_file_range(trans, inode, root, locked_page, start, end,
1034 page_started, nr_written, unlock);
1036 btrfs_end_transaction(trans, root);
1042 * work queue call back to started compression on a file and pages
1044 static noinline void async_cow_start(struct btrfs_work *work)
1046 struct async_cow *async_cow;
1048 async_cow = container_of(work, struct async_cow, work);
1050 compress_file_range(async_cow->inode, async_cow->locked_page,
1051 async_cow->start, async_cow->end, async_cow,
1053 if (num_added == 0) {
1054 btrfs_add_delayed_iput(async_cow->inode);
1055 async_cow->inode = NULL;
1060 * work queue call back to submit previously compressed pages
1062 static noinline void async_cow_submit(struct btrfs_work *work)
1064 struct async_cow *async_cow;
1065 struct btrfs_root *root;
1066 unsigned long nr_pages;
1068 async_cow = container_of(work, struct async_cow, work);
1070 root = async_cow->root;
1071 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1074 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1076 waitqueue_active(&root->fs_info->async_submit_wait))
1077 wake_up(&root->fs_info->async_submit_wait);
1079 if (async_cow->inode)
1080 submit_compressed_extents(async_cow->inode, async_cow);
1083 static noinline void async_cow_free(struct btrfs_work *work)
1085 struct async_cow *async_cow;
1086 async_cow = container_of(work, struct async_cow, work);
1087 if (async_cow->inode)
1088 btrfs_add_delayed_iput(async_cow->inode);
1092 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1093 u64 start, u64 end, int *page_started,
1094 unsigned long *nr_written)
1096 struct async_cow *async_cow;
1097 struct btrfs_root *root = BTRFS_I(inode)->root;
1098 unsigned long nr_pages;
1100 int limit = 10 * 1024 * 1024;
1102 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1103 1, 0, NULL, GFP_NOFS);
1104 while (start < end) {
1105 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1106 BUG_ON(!async_cow); /* -ENOMEM */
1107 async_cow->inode = igrab(inode);
1108 async_cow->root = root;
1109 async_cow->locked_page = locked_page;
1110 async_cow->start = start;
1112 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1115 cur_end = min(end, start + 512 * 1024 - 1);
1117 async_cow->end = cur_end;
1118 INIT_LIST_HEAD(&async_cow->extents);
1120 async_cow->work.func = async_cow_start;
1121 async_cow->work.ordered_func = async_cow_submit;
1122 async_cow->work.ordered_free = async_cow_free;
1123 async_cow->work.flags = 0;
1125 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1127 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1129 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1132 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1133 wait_event(root->fs_info->async_submit_wait,
1134 (atomic_read(&root->fs_info->async_delalloc_pages) <
1138 while (atomic_read(&root->fs_info->async_submit_draining) &&
1139 atomic_read(&root->fs_info->async_delalloc_pages)) {
1140 wait_event(root->fs_info->async_submit_wait,
1141 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1145 *nr_written += nr_pages;
1146 start = cur_end + 1;
1152 static noinline int csum_exist_in_range(struct btrfs_root *root,
1153 u64 bytenr, u64 num_bytes)
1156 struct btrfs_ordered_sum *sums;
1159 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1160 bytenr + num_bytes - 1, &list, 0);
1161 if (ret == 0 && list_empty(&list))
1164 while (!list_empty(&list)) {
1165 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1166 list_del(&sums->list);
1173 * when nowcow writeback call back. This checks for snapshots or COW copies
1174 * of the extents that exist in the file, and COWs the file as required.
1176 * If no cow copies or snapshots exist, we write directly to the existing
1179 static noinline int run_delalloc_nocow(struct inode *inode,
1180 struct page *locked_page,
1181 u64 start, u64 end, int *page_started, int force,
1182 unsigned long *nr_written)
1184 struct btrfs_root *root = BTRFS_I(inode)->root;
1185 struct btrfs_trans_handle *trans;
1186 struct extent_buffer *leaf;
1187 struct btrfs_path *path;
1188 struct btrfs_file_extent_item *fi;
1189 struct btrfs_key found_key;
1203 u64 ino = btrfs_ino(inode);
1205 path = btrfs_alloc_path();
1207 extent_clear_unlock_delalloc(inode,
1208 &BTRFS_I(inode)->io_tree,
1209 start, end, locked_page,
1210 EXTENT_CLEAR_UNLOCK_PAGE |
1211 EXTENT_CLEAR_UNLOCK |
1212 EXTENT_CLEAR_DELALLOC |
1213 EXTENT_CLEAR_DIRTY |
1214 EXTENT_SET_WRITEBACK |
1215 EXTENT_END_WRITEBACK);
1219 nolock = btrfs_is_free_space_inode(inode);
1222 trans = btrfs_join_transaction_nolock(root);
1224 trans = btrfs_join_transaction(root);
1226 if (IS_ERR(trans)) {
1227 extent_clear_unlock_delalloc(inode,
1228 &BTRFS_I(inode)->io_tree,
1229 start, end, locked_page,
1230 EXTENT_CLEAR_UNLOCK_PAGE |
1231 EXTENT_CLEAR_UNLOCK |
1232 EXTENT_CLEAR_DELALLOC |
1233 EXTENT_CLEAR_DIRTY |
1234 EXTENT_SET_WRITEBACK |
1235 EXTENT_END_WRITEBACK);
1236 btrfs_free_path(path);
1237 return PTR_ERR(trans);
1240 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1242 cow_start = (u64)-1;
1245 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1248 btrfs_abort_transaction(trans, root, ret);
1251 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1252 leaf = path->nodes[0];
1253 btrfs_item_key_to_cpu(leaf, &found_key,
1254 path->slots[0] - 1);
1255 if (found_key.objectid == ino &&
1256 found_key.type == BTRFS_EXTENT_DATA_KEY)
1261 leaf = path->nodes[0];
1262 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1263 ret = btrfs_next_leaf(root, path);
1265 btrfs_abort_transaction(trans, root, ret);
1270 leaf = path->nodes[0];
1276 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1278 if (found_key.objectid > ino ||
1279 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1280 found_key.offset > end)
1283 if (found_key.offset > cur_offset) {
1284 extent_end = found_key.offset;
1289 fi = btrfs_item_ptr(leaf, path->slots[0],
1290 struct btrfs_file_extent_item);
1291 extent_type = btrfs_file_extent_type(leaf, fi);
1293 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1294 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1295 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1296 extent_offset = btrfs_file_extent_offset(leaf, fi);
1297 extent_end = found_key.offset +
1298 btrfs_file_extent_num_bytes(leaf, fi);
1300 btrfs_file_extent_disk_num_bytes(leaf, fi);
1301 if (extent_end <= start) {
1305 if (disk_bytenr == 0)
1307 if (btrfs_file_extent_compression(leaf, fi) ||
1308 btrfs_file_extent_encryption(leaf, fi) ||
1309 btrfs_file_extent_other_encoding(leaf, fi))
1311 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1313 if (btrfs_extent_readonly(root, disk_bytenr))
1315 if (btrfs_cross_ref_exist(trans, root, ino,
1317 extent_offset, disk_bytenr))
1319 disk_bytenr += extent_offset;
1320 disk_bytenr += cur_offset - found_key.offset;
1321 num_bytes = min(end + 1, extent_end) - cur_offset;
1323 * force cow if csum exists in the range.
1324 * this ensure that csum for a given extent are
1325 * either valid or do not exist.
1327 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1330 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1331 extent_end = found_key.offset +
1332 btrfs_file_extent_inline_len(leaf, fi);
1333 extent_end = ALIGN(extent_end, root->sectorsize);
1338 if (extent_end <= start) {
1343 if (cow_start == (u64)-1)
1344 cow_start = cur_offset;
1345 cur_offset = extent_end;
1346 if (cur_offset > end)
1352 btrfs_release_path(path);
1353 if (cow_start != (u64)-1) {
1354 ret = __cow_file_range(trans, inode, root, locked_page,
1355 cow_start, found_key.offset - 1,
1356 page_started, nr_written, 1);
1358 btrfs_abort_transaction(trans, root, ret);
1361 cow_start = (u64)-1;
1364 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1365 struct extent_map *em;
1366 struct extent_map_tree *em_tree;
1367 em_tree = &BTRFS_I(inode)->extent_tree;
1368 em = alloc_extent_map();
1369 BUG_ON(!em); /* -ENOMEM */
1370 em->start = cur_offset;
1371 em->orig_start = found_key.offset - extent_offset;
1372 em->len = num_bytes;
1373 em->block_len = num_bytes;
1374 em->block_start = disk_bytenr;
1375 em->orig_block_len = disk_num_bytes;
1376 em->bdev = root->fs_info->fs_devices->latest_bdev;
1377 em->mod_start = em->start;
1378 em->mod_len = em->len;
1379 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1380 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1381 em->generation = -1;
1383 write_lock(&em_tree->lock);
1384 ret = add_extent_mapping(em_tree, em);
1386 list_move(&em->list,
1387 &em_tree->modified_extents);
1388 write_unlock(&em_tree->lock);
1389 if (ret != -EEXIST) {
1390 free_extent_map(em);
1393 btrfs_drop_extent_cache(inode, em->start,
1394 em->start + em->len - 1, 0);
1396 type = BTRFS_ORDERED_PREALLOC;
1398 type = BTRFS_ORDERED_NOCOW;
1401 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1402 num_bytes, num_bytes, type);
1403 BUG_ON(ret); /* -ENOMEM */
1405 if (root->root_key.objectid ==
1406 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1407 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1410 btrfs_abort_transaction(trans, root, ret);
1415 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1416 cur_offset, cur_offset + num_bytes - 1,
1417 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1418 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1419 EXTENT_SET_PRIVATE2);
1420 cur_offset = extent_end;
1421 if (cur_offset > end)
1424 btrfs_release_path(path);
1426 if (cur_offset <= end && cow_start == (u64)-1) {
1427 cow_start = cur_offset;
1431 if (cow_start != (u64)-1) {
1432 ret = __cow_file_range(trans, inode, root, locked_page,
1434 page_started, nr_written, 1);
1436 btrfs_abort_transaction(trans, root, ret);
1442 err = btrfs_end_transaction(trans, root);
1446 if (ret && cur_offset < end)
1447 extent_clear_unlock_delalloc(inode,
1448 &BTRFS_I(inode)->io_tree,
1449 cur_offset, end, locked_page,
1450 EXTENT_CLEAR_UNLOCK_PAGE |
1451 EXTENT_CLEAR_UNLOCK |
1452 EXTENT_CLEAR_DELALLOC |
1453 EXTENT_CLEAR_DIRTY |
1454 EXTENT_SET_WRITEBACK |
1455 EXTENT_END_WRITEBACK);
1457 btrfs_free_path(path);
1462 * extent_io.c call back to do delayed allocation processing
1464 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1465 u64 start, u64 end, int *page_started,
1466 unsigned long *nr_written)
1469 struct btrfs_root *root = BTRFS_I(inode)->root;
1471 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1472 ret = run_delalloc_nocow(inode, locked_page, start, end,
1473 page_started, 1, nr_written);
1474 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1475 ret = run_delalloc_nocow(inode, locked_page, start, end,
1476 page_started, 0, nr_written);
1477 } else if (!btrfs_test_opt(root, COMPRESS) &&
1478 !(BTRFS_I(inode)->force_compress) &&
1479 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1480 ret = cow_file_range(inode, locked_page, start, end,
1481 page_started, nr_written, 1);
1483 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1484 &BTRFS_I(inode)->runtime_flags);
1485 ret = cow_file_range_async(inode, locked_page, start, end,
1486 page_started, nr_written);
1491 static void btrfs_split_extent_hook(struct inode *inode,
1492 struct extent_state *orig, u64 split)
1494 /* not delalloc, ignore it */
1495 if (!(orig->state & EXTENT_DELALLOC))
1498 spin_lock(&BTRFS_I(inode)->lock);
1499 BTRFS_I(inode)->outstanding_extents++;
1500 spin_unlock(&BTRFS_I(inode)->lock);
1504 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1505 * extents so we can keep track of new extents that are just merged onto old
1506 * extents, such as when we are doing sequential writes, so we can properly
1507 * account for the metadata space we'll need.
1509 static void btrfs_merge_extent_hook(struct inode *inode,
1510 struct extent_state *new,
1511 struct extent_state *other)
1513 /* not delalloc, ignore it */
1514 if (!(other->state & EXTENT_DELALLOC))
1517 spin_lock(&BTRFS_I(inode)->lock);
1518 BTRFS_I(inode)->outstanding_extents--;
1519 spin_unlock(&BTRFS_I(inode)->lock);
1523 * extent_io.c set_bit_hook, used to track delayed allocation
1524 * bytes in this file, and to maintain the list of inodes that
1525 * have pending delalloc work to be done.
1527 static void btrfs_set_bit_hook(struct inode *inode,
1528 struct extent_state *state, int *bits)
1532 * set_bit and clear bit hooks normally require _irqsave/restore
1533 * but in this case, we are only testing for the DELALLOC
1534 * bit, which is only set or cleared with irqs on
1536 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1537 struct btrfs_root *root = BTRFS_I(inode)->root;
1538 u64 len = state->end + 1 - state->start;
1539 bool do_list = !btrfs_is_free_space_inode(inode);
1541 if (*bits & EXTENT_FIRST_DELALLOC) {
1542 *bits &= ~EXTENT_FIRST_DELALLOC;
1544 spin_lock(&BTRFS_I(inode)->lock);
1545 BTRFS_I(inode)->outstanding_extents++;
1546 spin_unlock(&BTRFS_I(inode)->lock);
1549 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1550 root->fs_info->delalloc_batch);
1551 spin_lock(&BTRFS_I(inode)->lock);
1552 BTRFS_I(inode)->delalloc_bytes += len;
1553 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1554 &BTRFS_I(inode)->runtime_flags)) {
1555 spin_lock(&root->fs_info->delalloc_lock);
1556 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1557 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1558 &root->fs_info->delalloc_inodes);
1559 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1560 &BTRFS_I(inode)->runtime_flags);
1562 spin_unlock(&root->fs_info->delalloc_lock);
1564 spin_unlock(&BTRFS_I(inode)->lock);
1569 * extent_io.c clear_bit_hook, see set_bit_hook for why
1571 static void btrfs_clear_bit_hook(struct inode *inode,
1572 struct extent_state *state, int *bits)
1575 * set_bit and clear bit hooks normally require _irqsave/restore
1576 * but in this case, we are only testing for the DELALLOC
1577 * bit, which is only set or cleared with irqs on
1579 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1580 struct btrfs_root *root = BTRFS_I(inode)->root;
1581 u64 len = state->end + 1 - state->start;
1582 bool do_list = !btrfs_is_free_space_inode(inode);
1584 if (*bits & EXTENT_FIRST_DELALLOC) {
1585 *bits &= ~EXTENT_FIRST_DELALLOC;
1586 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1587 spin_lock(&BTRFS_I(inode)->lock);
1588 BTRFS_I(inode)->outstanding_extents--;
1589 spin_unlock(&BTRFS_I(inode)->lock);
1592 if (*bits & EXTENT_DO_ACCOUNTING)
1593 btrfs_delalloc_release_metadata(inode, len);
1595 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1597 btrfs_free_reserved_data_space(inode, len);
1599 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1600 root->fs_info->delalloc_batch);
1601 spin_lock(&BTRFS_I(inode)->lock);
1602 BTRFS_I(inode)->delalloc_bytes -= len;
1603 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1604 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1605 &BTRFS_I(inode)->runtime_flags)) {
1606 spin_lock(&root->fs_info->delalloc_lock);
1607 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1608 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1609 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1610 &BTRFS_I(inode)->runtime_flags);
1612 spin_unlock(&root->fs_info->delalloc_lock);
1614 spin_unlock(&BTRFS_I(inode)->lock);
1619 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1620 * we don't create bios that span stripes or chunks
1622 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1623 size_t size, struct bio *bio,
1624 unsigned long bio_flags)
1626 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1627 u64 logical = (u64)bio->bi_sector << 9;
1632 if (bio_flags & EXTENT_BIO_COMPRESSED)
1635 length = bio->bi_size;
1636 map_length = length;
1637 ret = btrfs_map_block(root->fs_info, rw, logical,
1638 &map_length, NULL, 0);
1639 /* Will always return 0 with map_multi == NULL */
1641 if (map_length < length + size)
1647 * in order to insert checksums into the metadata in large chunks,
1648 * we wait until bio submission time. All the pages in the bio are
1649 * checksummed and sums are attached onto the ordered extent record.
1651 * At IO completion time the cums attached on the ordered extent record
1652 * are inserted into the btree
1654 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1655 struct bio *bio, int mirror_num,
1656 unsigned long bio_flags,
1659 struct btrfs_root *root = BTRFS_I(inode)->root;
1662 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1663 BUG_ON(ret); /* -ENOMEM */
1668 * in order to insert checksums into the metadata in large chunks,
1669 * we wait until bio submission time. All the pages in the bio are
1670 * checksummed and sums are attached onto the ordered extent record.
1672 * At IO completion time the cums attached on the ordered extent record
1673 * are inserted into the btree
1675 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1676 int mirror_num, unsigned long bio_flags,
1679 struct btrfs_root *root = BTRFS_I(inode)->root;
1682 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1684 bio_endio(bio, ret);
1689 * extent_io.c submission hook. This does the right thing for csum calculation
1690 * on write, or reading the csums from the tree before a read
1692 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1693 int mirror_num, unsigned long bio_flags,
1696 struct btrfs_root *root = BTRFS_I(inode)->root;
1700 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1702 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1704 if (btrfs_is_free_space_inode(inode))
1707 if (!(rw & REQ_WRITE)) {
1708 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1712 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1713 ret = btrfs_submit_compressed_read(inode, bio,
1717 } else if (!skip_sum) {
1718 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1723 } else if (async && !skip_sum) {
1724 /* csum items have already been cloned */
1725 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1727 /* we're doing a write, do the async checksumming */
1728 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1729 inode, rw, bio, mirror_num,
1730 bio_flags, bio_offset,
1731 __btrfs_submit_bio_start,
1732 __btrfs_submit_bio_done);
1734 } else if (!skip_sum) {
1735 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1741 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1745 bio_endio(bio, ret);
1750 * given a list of ordered sums record them in the inode. This happens
1751 * at IO completion time based on sums calculated at bio submission time.
1753 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1754 struct inode *inode, u64 file_offset,
1755 struct list_head *list)
1757 struct btrfs_ordered_sum *sum;
1759 list_for_each_entry(sum, list, list) {
1760 trans->adding_csums = 1;
1761 btrfs_csum_file_blocks(trans,
1762 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1763 trans->adding_csums = 0;
1768 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1769 struct extent_state **cached_state)
1771 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1772 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1773 cached_state, GFP_NOFS);
1776 /* see btrfs_writepage_start_hook for details on why this is required */
1777 struct btrfs_writepage_fixup {
1779 struct btrfs_work work;
1782 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1784 struct btrfs_writepage_fixup *fixup;
1785 struct btrfs_ordered_extent *ordered;
1786 struct extent_state *cached_state = NULL;
1788 struct inode *inode;
1793 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1797 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1798 ClearPageChecked(page);
1802 inode = page->mapping->host;
1803 page_start = page_offset(page);
1804 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1806 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1809 /* already ordered? We're done */
1810 if (PagePrivate2(page))
1813 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1815 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1816 page_end, &cached_state, GFP_NOFS);
1818 btrfs_start_ordered_extent(inode, ordered, 1);
1819 btrfs_put_ordered_extent(ordered);
1823 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1825 mapping_set_error(page->mapping, ret);
1826 end_extent_writepage(page, ret, page_start, page_end);
1827 ClearPageChecked(page);
1831 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1832 ClearPageChecked(page);
1833 set_page_dirty(page);
1835 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1836 &cached_state, GFP_NOFS);
1839 page_cache_release(page);
1844 * There are a few paths in the higher layers of the kernel that directly
1845 * set the page dirty bit without asking the filesystem if it is a
1846 * good idea. This causes problems because we want to make sure COW
1847 * properly happens and the data=ordered rules are followed.
1849 * In our case any range that doesn't have the ORDERED bit set
1850 * hasn't been properly setup for IO. We kick off an async process
1851 * to fix it up. The async helper will wait for ordered extents, set
1852 * the delalloc bit and make it safe to write the page.
1854 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1856 struct inode *inode = page->mapping->host;
1857 struct btrfs_writepage_fixup *fixup;
1858 struct btrfs_root *root = BTRFS_I(inode)->root;
1860 /* this page is properly in the ordered list */
1861 if (TestClearPagePrivate2(page))
1864 if (PageChecked(page))
1867 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1871 SetPageChecked(page);
1872 page_cache_get(page);
1873 fixup->work.func = btrfs_writepage_fixup_worker;
1875 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1879 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1880 struct inode *inode, u64 file_pos,
1881 u64 disk_bytenr, u64 disk_num_bytes,
1882 u64 num_bytes, u64 ram_bytes,
1883 u8 compression, u8 encryption,
1884 u16 other_encoding, int extent_type)
1886 struct btrfs_root *root = BTRFS_I(inode)->root;
1887 struct btrfs_file_extent_item *fi;
1888 struct btrfs_path *path;
1889 struct extent_buffer *leaf;
1890 struct btrfs_key ins;
1893 path = btrfs_alloc_path();
1897 path->leave_spinning = 1;
1900 * we may be replacing one extent in the tree with another.
1901 * The new extent is pinned in the extent map, and we don't want
1902 * to drop it from the cache until it is completely in the btree.
1904 * So, tell btrfs_drop_extents to leave this extent in the cache.
1905 * the caller is expected to unpin it and allow it to be merged
1908 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1909 file_pos + num_bytes, 0);
1913 ins.objectid = btrfs_ino(inode);
1914 ins.offset = file_pos;
1915 ins.type = BTRFS_EXTENT_DATA_KEY;
1916 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1919 leaf = path->nodes[0];
1920 fi = btrfs_item_ptr(leaf, path->slots[0],
1921 struct btrfs_file_extent_item);
1922 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1923 btrfs_set_file_extent_type(leaf, fi, extent_type);
1924 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1925 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1926 btrfs_set_file_extent_offset(leaf, fi, 0);
1927 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1928 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1929 btrfs_set_file_extent_compression(leaf, fi, compression);
1930 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1931 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1933 btrfs_mark_buffer_dirty(leaf);
1934 btrfs_release_path(path);
1936 inode_add_bytes(inode, num_bytes);
1938 ins.objectid = disk_bytenr;
1939 ins.offset = disk_num_bytes;
1940 ins.type = BTRFS_EXTENT_ITEM_KEY;
1941 ret = btrfs_alloc_reserved_file_extent(trans, root,
1942 root->root_key.objectid,
1943 btrfs_ino(inode), file_pos, &ins);
1945 btrfs_free_path(path);
1950 /* snapshot-aware defrag */
1951 struct sa_defrag_extent_backref {
1952 struct rb_node node;
1953 struct old_sa_defrag_extent *old;
1962 struct old_sa_defrag_extent {
1963 struct list_head list;
1964 struct new_sa_defrag_extent *new;
1973 struct new_sa_defrag_extent {
1974 struct rb_root root;
1975 struct list_head head;
1976 struct btrfs_path *path;
1977 struct inode *inode;
1985 static int backref_comp(struct sa_defrag_extent_backref *b1,
1986 struct sa_defrag_extent_backref *b2)
1988 if (b1->root_id < b2->root_id)
1990 else if (b1->root_id > b2->root_id)
1993 if (b1->inum < b2->inum)
1995 else if (b1->inum > b2->inum)
1998 if (b1->file_pos < b2->file_pos)
2000 else if (b1->file_pos > b2->file_pos)
2004 * [------------------------------] ===> (a range of space)
2005 * |<--->| |<---->| =============> (fs/file tree A)
2006 * |<---------------------------->| ===> (fs/file tree B)
2008 * A range of space can refer to two file extents in one tree while
2009 * refer to only one file extent in another tree.
2011 * So we may process a disk offset more than one time(two extents in A)
2012 * and locate at the same extent(one extent in B), then insert two same
2013 * backrefs(both refer to the extent in B).
2018 static void backref_insert(struct rb_root *root,
2019 struct sa_defrag_extent_backref *backref)
2021 struct rb_node **p = &root->rb_node;
2022 struct rb_node *parent = NULL;
2023 struct sa_defrag_extent_backref *entry;
2028 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2030 ret = backref_comp(backref, entry);
2034 p = &(*p)->rb_right;
2037 rb_link_node(&backref->node, parent, p);
2038 rb_insert_color(&backref->node, root);
2042 * Note the backref might has changed, and in this case we just return 0.
2044 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2047 struct btrfs_file_extent_item *extent;
2048 struct btrfs_fs_info *fs_info;
2049 struct old_sa_defrag_extent *old = ctx;
2050 struct new_sa_defrag_extent *new = old->new;
2051 struct btrfs_path *path = new->path;
2052 struct btrfs_key key;
2053 struct btrfs_root *root;
2054 struct sa_defrag_extent_backref *backref;
2055 struct extent_buffer *leaf;
2056 struct inode *inode = new->inode;
2062 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2063 inum == btrfs_ino(inode))
2066 key.objectid = root_id;
2067 key.type = BTRFS_ROOT_ITEM_KEY;
2068 key.offset = (u64)-1;
2070 fs_info = BTRFS_I(inode)->root->fs_info;
2071 root = btrfs_read_fs_root_no_name(fs_info, &key);
2073 if (PTR_ERR(root) == -ENOENT)
2076 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2077 inum, offset, root_id);
2078 return PTR_ERR(root);
2081 key.objectid = inum;
2082 key.type = BTRFS_EXTENT_DATA_KEY;
2083 if (offset > (u64)-1 << 32)
2086 key.offset = offset;
2088 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2097 leaf = path->nodes[0];
2098 slot = path->slots[0];
2100 if (slot >= btrfs_header_nritems(leaf)) {
2101 ret = btrfs_next_leaf(root, path);
2104 } else if (ret > 0) {
2113 btrfs_item_key_to_cpu(leaf, &key, slot);
2115 if (key.objectid > inum)
2118 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2121 extent = btrfs_item_ptr(leaf, slot,
2122 struct btrfs_file_extent_item);
2124 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2127 extent_offset = btrfs_file_extent_offset(leaf, extent);
2128 if (key.offset - extent_offset != offset)
2131 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2132 if (extent_offset >= old->extent_offset + old->offset +
2133 old->len || extent_offset + num_bytes <=
2134 old->extent_offset + old->offset)
2140 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2146 backref->root_id = root_id;
2147 backref->inum = inum;
2148 backref->file_pos = offset + extent_offset;
2149 backref->num_bytes = num_bytes;
2150 backref->extent_offset = extent_offset;
2151 backref->generation = btrfs_file_extent_generation(leaf, extent);
2153 backref_insert(&new->root, backref);
2156 btrfs_release_path(path);
2161 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2162 struct new_sa_defrag_extent *new)
2164 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2165 struct old_sa_defrag_extent *old, *tmp;
2170 list_for_each_entry_safe(old, tmp, &new->head, list) {
2171 ret = iterate_inodes_from_logical(old->bytenr, fs_info,
2172 path, record_one_backref,
2174 BUG_ON(ret < 0 && ret != -ENOENT);
2176 /* no backref to be processed for this extent */
2178 list_del(&old->list);
2183 if (list_empty(&new->head))
2189 static int relink_is_mergable(struct extent_buffer *leaf,
2190 struct btrfs_file_extent_item *fi,
2193 if (btrfs_file_extent_disk_bytenr(leaf, fi) != disk_bytenr)
2196 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2199 if (btrfs_file_extent_compression(leaf, fi) ||
2200 btrfs_file_extent_encryption(leaf, fi) ||
2201 btrfs_file_extent_other_encoding(leaf, fi))
2208 * Note the backref might has changed, and in this case we just return 0.
2210 static noinline int relink_extent_backref(struct btrfs_path *path,
2211 struct sa_defrag_extent_backref *prev,
2212 struct sa_defrag_extent_backref *backref)
2214 struct btrfs_file_extent_item *extent;
2215 struct btrfs_file_extent_item *item;
2216 struct btrfs_ordered_extent *ordered;
2217 struct btrfs_trans_handle *trans;
2218 struct btrfs_fs_info *fs_info;
2219 struct btrfs_root *root;
2220 struct btrfs_key key;
2221 struct extent_buffer *leaf;
2222 struct old_sa_defrag_extent *old = backref->old;
2223 struct new_sa_defrag_extent *new = old->new;
2224 struct inode *src_inode = new->inode;
2225 struct inode *inode;
2226 struct extent_state *cached = NULL;
2235 if (prev && prev->root_id == backref->root_id &&
2236 prev->inum == backref->inum &&
2237 prev->file_pos + prev->num_bytes == backref->file_pos)
2240 /* step 1: get root */
2241 key.objectid = backref->root_id;
2242 key.type = BTRFS_ROOT_ITEM_KEY;
2243 key.offset = (u64)-1;
2245 fs_info = BTRFS_I(src_inode)->root->fs_info;
2246 index = srcu_read_lock(&fs_info->subvol_srcu);
2248 root = btrfs_read_fs_root_no_name(fs_info, &key);
2250 srcu_read_unlock(&fs_info->subvol_srcu, index);
2251 if (PTR_ERR(root) == -ENOENT)
2253 return PTR_ERR(root);
2255 if (btrfs_root_refs(&root->root_item) == 0) {
2256 srcu_read_unlock(&fs_info->subvol_srcu, index);
2257 /* parse ENOENT to 0 */
2261 /* step 2: get inode */
2262 key.objectid = backref->inum;
2263 key.type = BTRFS_INODE_ITEM_KEY;
2266 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2267 if (IS_ERR(inode)) {
2268 srcu_read_unlock(&fs_info->subvol_srcu, index);
2272 srcu_read_unlock(&fs_info->subvol_srcu, index);
2274 /* step 3: relink backref */
2275 lock_start = backref->file_pos;
2276 lock_end = backref->file_pos + backref->num_bytes - 1;
2277 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2280 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2282 btrfs_put_ordered_extent(ordered);
2286 trans = btrfs_join_transaction(root);
2287 if (IS_ERR(trans)) {
2288 ret = PTR_ERR(trans);
2292 key.objectid = backref->inum;
2293 key.type = BTRFS_EXTENT_DATA_KEY;
2294 key.offset = backref->file_pos;
2296 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2299 } else if (ret > 0) {
2304 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2305 struct btrfs_file_extent_item);
2307 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2308 backref->generation)
2311 btrfs_release_path(path);
2313 start = backref->file_pos;
2314 if (backref->extent_offset < old->extent_offset + old->offset)
2315 start += old->extent_offset + old->offset -
2316 backref->extent_offset;
2318 len = min(backref->extent_offset + backref->num_bytes,
2319 old->extent_offset + old->offset + old->len);
2320 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2322 ret = btrfs_drop_extents(trans, root, inode, start,
2327 key.objectid = btrfs_ino(inode);
2328 key.type = BTRFS_EXTENT_DATA_KEY;
2331 path->leave_spinning = 1;
2333 struct btrfs_file_extent_item *fi;
2335 struct btrfs_key found_key;
2337 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
2342 leaf = path->nodes[0];
2343 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2345 fi = btrfs_item_ptr(leaf, path->slots[0],
2346 struct btrfs_file_extent_item);
2347 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2349 if (relink_is_mergable(leaf, fi, new->bytenr) &&
2350 extent_len + found_key.offset == start) {
2351 btrfs_set_file_extent_num_bytes(leaf, fi,
2353 btrfs_mark_buffer_dirty(leaf);
2354 inode_add_bytes(inode, len);
2360 btrfs_release_path(path);
2365 ret = btrfs_insert_empty_item(trans, root, path, &key,
2368 btrfs_abort_transaction(trans, root, ret);
2372 leaf = path->nodes[0];
2373 item = btrfs_item_ptr(leaf, path->slots[0],
2374 struct btrfs_file_extent_item);
2375 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2376 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2377 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2378 btrfs_set_file_extent_num_bytes(leaf, item, len);
2379 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2380 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2381 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2382 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2383 btrfs_set_file_extent_encryption(leaf, item, 0);
2384 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2386 btrfs_mark_buffer_dirty(leaf);
2387 inode_add_bytes(inode, len);
2388 btrfs_release_path(path);
2390 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2392 backref->root_id, backref->inum,
2393 new->file_pos, 0); /* start - extent_offset */
2395 btrfs_abort_transaction(trans, root, ret);
2401 btrfs_release_path(path);
2402 path->leave_spinning = 0;
2403 btrfs_end_transaction(trans, root);
2405 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2411 static void relink_file_extents(struct new_sa_defrag_extent *new)
2413 struct btrfs_path *path;
2414 struct old_sa_defrag_extent *old, *tmp;
2415 struct sa_defrag_extent_backref *backref;
2416 struct sa_defrag_extent_backref *prev = NULL;
2417 struct inode *inode;
2418 struct btrfs_root *root;
2419 struct rb_node *node;
2423 root = BTRFS_I(inode)->root;
2425 path = btrfs_alloc_path();
2429 if (!record_extent_backrefs(path, new)) {
2430 btrfs_free_path(path);
2433 btrfs_release_path(path);
2436 node = rb_first(&new->root);
2439 rb_erase(node, &new->root);
2441 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2443 ret = relink_extent_backref(path, prev, backref);
2456 btrfs_free_path(path);
2458 list_for_each_entry_safe(old, tmp, &new->head, list) {
2459 list_del(&old->list);
2463 atomic_dec(&root->fs_info->defrag_running);
2464 wake_up(&root->fs_info->transaction_wait);
2469 static struct new_sa_defrag_extent *
2470 record_old_file_extents(struct inode *inode,
2471 struct btrfs_ordered_extent *ordered)
2473 struct btrfs_root *root = BTRFS_I(inode)->root;
2474 struct btrfs_path *path;
2475 struct btrfs_key key;
2476 struct old_sa_defrag_extent *old, *tmp;
2477 struct new_sa_defrag_extent *new;
2480 new = kmalloc(sizeof(*new), GFP_NOFS);
2485 new->file_pos = ordered->file_offset;
2486 new->len = ordered->len;
2487 new->bytenr = ordered->start;
2488 new->disk_len = ordered->disk_len;
2489 new->compress_type = ordered->compress_type;
2490 new->root = RB_ROOT;
2491 INIT_LIST_HEAD(&new->head);
2493 path = btrfs_alloc_path();
2497 key.objectid = btrfs_ino(inode);
2498 key.type = BTRFS_EXTENT_DATA_KEY;
2499 key.offset = new->file_pos;
2501 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2504 if (ret > 0 && path->slots[0] > 0)
2507 /* find out all the old extents for the file range */
2509 struct btrfs_file_extent_item *extent;
2510 struct extent_buffer *l;
2519 slot = path->slots[0];
2521 if (slot >= btrfs_header_nritems(l)) {
2522 ret = btrfs_next_leaf(root, path);
2530 btrfs_item_key_to_cpu(l, &key, slot);
2532 if (key.objectid != btrfs_ino(inode))
2534 if (key.type != BTRFS_EXTENT_DATA_KEY)
2536 if (key.offset >= new->file_pos + new->len)
2539 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2541 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2542 if (key.offset + num_bytes < new->file_pos)
2545 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2549 extent_offset = btrfs_file_extent_offset(l, extent);
2551 old = kmalloc(sizeof(*old), GFP_NOFS);
2555 offset = max(new->file_pos, key.offset);
2556 end = min(new->file_pos + new->len, key.offset + num_bytes);
2558 old->bytenr = disk_bytenr;
2559 old->extent_offset = extent_offset;
2560 old->offset = offset - key.offset;
2561 old->len = end - offset;
2564 list_add_tail(&old->list, &new->head);
2570 btrfs_free_path(path);
2571 atomic_inc(&root->fs_info->defrag_running);
2576 list_for_each_entry_safe(old, tmp, &new->head, list) {
2577 list_del(&old->list);
2581 btrfs_free_path(path);
2588 * helper function for btrfs_finish_ordered_io, this
2589 * just reads in some of the csum leaves to prime them into ram
2590 * before we start the transaction. It limits the amount of btree
2591 * reads required while inside the transaction.
2593 /* as ordered data IO finishes, this gets called so we can finish
2594 * an ordered extent if the range of bytes in the file it covers are
2597 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2599 struct inode *inode = ordered_extent->inode;
2600 struct btrfs_root *root = BTRFS_I(inode)->root;
2601 struct btrfs_trans_handle *trans = NULL;
2602 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2603 struct extent_state *cached_state = NULL;
2604 struct new_sa_defrag_extent *new = NULL;
2605 int compress_type = 0;
2609 nolock = btrfs_is_free_space_inode(inode);
2611 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2616 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2617 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2618 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2620 trans = btrfs_join_transaction_nolock(root);
2622 trans = btrfs_join_transaction(root);
2623 if (IS_ERR(trans)) {
2624 ret = PTR_ERR(trans);
2628 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2629 ret = btrfs_update_inode_fallback(trans, root, inode);
2630 if (ret) /* -ENOMEM or corruption */
2631 btrfs_abort_transaction(trans, root, ret);
2635 lock_extent_bits(io_tree, ordered_extent->file_offset,
2636 ordered_extent->file_offset + ordered_extent->len - 1,
2639 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2640 ordered_extent->file_offset + ordered_extent->len - 1,
2641 EXTENT_DEFRAG, 1, cached_state);
2643 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2644 if (last_snapshot >= BTRFS_I(inode)->generation)
2645 /* the inode is shared */
2646 new = record_old_file_extents(inode, ordered_extent);
2648 clear_extent_bit(io_tree, ordered_extent->file_offset,
2649 ordered_extent->file_offset + ordered_extent->len - 1,
2650 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2654 trans = btrfs_join_transaction_nolock(root);
2656 trans = btrfs_join_transaction(root);
2657 if (IS_ERR(trans)) {
2658 ret = PTR_ERR(trans);
2662 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2664 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2665 compress_type = ordered_extent->compress_type;
2666 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2667 BUG_ON(compress_type);
2668 ret = btrfs_mark_extent_written(trans, inode,
2669 ordered_extent->file_offset,
2670 ordered_extent->file_offset +
2671 ordered_extent->len);
2673 BUG_ON(root == root->fs_info->tree_root);
2674 ret = insert_reserved_file_extent(trans, inode,
2675 ordered_extent->file_offset,
2676 ordered_extent->start,
2677 ordered_extent->disk_len,
2678 ordered_extent->len,
2679 ordered_extent->len,
2680 compress_type, 0, 0,
2681 BTRFS_FILE_EXTENT_REG);
2683 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2684 ordered_extent->file_offset, ordered_extent->len,
2687 btrfs_abort_transaction(trans, root, ret);
2691 add_pending_csums(trans, inode, ordered_extent->file_offset,
2692 &ordered_extent->list);
2694 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2695 ret = btrfs_update_inode_fallback(trans, root, inode);
2696 if (ret) { /* -ENOMEM or corruption */
2697 btrfs_abort_transaction(trans, root, ret);
2702 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2703 ordered_extent->file_offset +
2704 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2706 if (root != root->fs_info->tree_root)
2707 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2709 btrfs_end_transaction(trans, root);
2712 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
2713 ordered_extent->file_offset +
2714 ordered_extent->len - 1, NULL, GFP_NOFS);
2717 * If the ordered extent had an IOERR or something else went
2718 * wrong we need to return the space for this ordered extent
2719 * back to the allocator.
2721 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2722 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2723 btrfs_free_reserved_extent(root, ordered_extent->start,
2724 ordered_extent->disk_len);
2729 * This needs to be done to make sure anybody waiting knows we are done
2730 * updating everything for this ordered extent.
2732 btrfs_remove_ordered_extent(inode, ordered_extent);
2734 /* for snapshot-aware defrag */
2736 relink_file_extents(new);
2739 btrfs_put_ordered_extent(ordered_extent);
2740 /* once for the tree */
2741 btrfs_put_ordered_extent(ordered_extent);
2746 static void finish_ordered_fn(struct btrfs_work *work)
2748 struct btrfs_ordered_extent *ordered_extent;
2749 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2750 btrfs_finish_ordered_io(ordered_extent);
2753 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2754 struct extent_state *state, int uptodate)
2756 struct inode *inode = page->mapping->host;
2757 struct btrfs_root *root = BTRFS_I(inode)->root;
2758 struct btrfs_ordered_extent *ordered_extent = NULL;
2759 struct btrfs_workers *workers;
2761 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2763 ClearPagePrivate2(page);
2764 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2765 end - start + 1, uptodate))
2768 ordered_extent->work.func = finish_ordered_fn;
2769 ordered_extent->work.flags = 0;
2771 if (btrfs_is_free_space_inode(inode))
2772 workers = &root->fs_info->endio_freespace_worker;
2774 workers = &root->fs_info->endio_write_workers;
2775 btrfs_queue_worker(workers, &ordered_extent->work);
2781 * when reads are done, we need to check csums to verify the data is correct
2782 * if there's a match, we allow the bio to finish. If not, the code in
2783 * extent_io.c will try to find good copies for us.
2785 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2786 struct extent_state *state, int mirror)
2788 size_t offset = start - page_offset(page);
2789 struct inode *inode = page->mapping->host;
2790 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2792 u64 private = ~(u32)0;
2794 struct btrfs_root *root = BTRFS_I(inode)->root;
2797 if (PageChecked(page)) {
2798 ClearPageChecked(page);
2802 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2805 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2806 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2807 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2812 if (state && state->start == start) {
2813 private = state->private;
2816 ret = get_state_private(io_tree, start, &private);
2818 kaddr = kmap_atomic(page);
2822 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
2823 btrfs_csum_final(csum, (char *)&csum);
2824 if (csum != private)
2827 kunmap_atomic(kaddr);
2832 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2834 (unsigned long long)btrfs_ino(page->mapping->host),
2835 (unsigned long long)start, csum,
2836 (unsigned long long)private);
2837 memset(kaddr + offset, 1, end - start + 1);
2838 flush_dcache_page(page);
2839 kunmap_atomic(kaddr);
2845 struct delayed_iput {
2846 struct list_head list;
2847 struct inode *inode;
2850 /* JDM: If this is fs-wide, why can't we add a pointer to
2851 * btrfs_inode instead and avoid the allocation? */
2852 void btrfs_add_delayed_iput(struct inode *inode)
2854 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2855 struct delayed_iput *delayed;
2857 if (atomic_add_unless(&inode->i_count, -1, 1))
2860 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2861 delayed->inode = inode;
2863 spin_lock(&fs_info->delayed_iput_lock);
2864 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2865 spin_unlock(&fs_info->delayed_iput_lock);
2868 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2871 struct btrfs_fs_info *fs_info = root->fs_info;
2872 struct delayed_iput *delayed;
2875 spin_lock(&fs_info->delayed_iput_lock);
2876 empty = list_empty(&fs_info->delayed_iputs);
2877 spin_unlock(&fs_info->delayed_iput_lock);
2881 spin_lock(&fs_info->delayed_iput_lock);
2882 list_splice_init(&fs_info->delayed_iputs, &list);
2883 spin_unlock(&fs_info->delayed_iput_lock);
2885 while (!list_empty(&list)) {
2886 delayed = list_entry(list.next, struct delayed_iput, list);
2887 list_del(&delayed->list);
2888 iput(delayed->inode);
2894 * This is called in transaction commit time. If there are no orphan
2895 * files in the subvolume, it removes orphan item and frees block_rsv
2898 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2899 struct btrfs_root *root)
2901 struct btrfs_block_rsv *block_rsv;
2904 if (atomic_read(&root->orphan_inodes) ||
2905 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2908 spin_lock(&root->orphan_lock);
2909 if (atomic_read(&root->orphan_inodes)) {
2910 spin_unlock(&root->orphan_lock);
2914 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2915 spin_unlock(&root->orphan_lock);
2919 block_rsv = root->orphan_block_rsv;
2920 root->orphan_block_rsv = NULL;
2921 spin_unlock(&root->orphan_lock);
2923 if (root->orphan_item_inserted &&
2924 btrfs_root_refs(&root->root_item) > 0) {
2925 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2926 root->root_key.objectid);
2928 root->orphan_item_inserted = 0;
2932 WARN_ON(block_rsv->size > 0);
2933 btrfs_free_block_rsv(root, block_rsv);
2938 * This creates an orphan entry for the given inode in case something goes
2939 * wrong in the middle of an unlink/truncate.
2941 * NOTE: caller of this function should reserve 5 units of metadata for
2944 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2946 struct btrfs_root *root = BTRFS_I(inode)->root;
2947 struct btrfs_block_rsv *block_rsv = NULL;
2952 if (!root->orphan_block_rsv) {
2953 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2958 spin_lock(&root->orphan_lock);
2959 if (!root->orphan_block_rsv) {
2960 root->orphan_block_rsv = block_rsv;
2961 } else if (block_rsv) {
2962 btrfs_free_block_rsv(root, block_rsv);
2966 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2967 &BTRFS_I(inode)->runtime_flags)) {
2970 * For proper ENOSPC handling, we should do orphan
2971 * cleanup when mounting. But this introduces backward
2972 * compatibility issue.
2974 if (!xchg(&root->orphan_item_inserted, 1))
2980 atomic_inc(&root->orphan_inodes);
2983 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2984 &BTRFS_I(inode)->runtime_flags))
2986 spin_unlock(&root->orphan_lock);
2988 /* grab metadata reservation from transaction handle */
2990 ret = btrfs_orphan_reserve_metadata(trans, inode);
2991 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2994 /* insert an orphan item to track this unlinked/truncated file */
2996 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2997 if (ret && ret != -EEXIST) {
2998 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2999 &BTRFS_I(inode)->runtime_flags);
3000 btrfs_abort_transaction(trans, root, ret);
3006 /* insert an orphan item to track subvolume contains orphan files */
3008 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3009 root->root_key.objectid);
3010 if (ret && ret != -EEXIST) {
3011 btrfs_abort_transaction(trans, root, ret);
3019 * We have done the truncate/delete so we can go ahead and remove the orphan
3020 * item for this particular inode.
3022 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
3024 struct btrfs_root *root = BTRFS_I(inode)->root;
3025 int delete_item = 0;
3026 int release_rsv = 0;
3029 spin_lock(&root->orphan_lock);
3030 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3031 &BTRFS_I(inode)->runtime_flags))
3034 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3035 &BTRFS_I(inode)->runtime_flags))
3037 spin_unlock(&root->orphan_lock);
3039 if (trans && delete_item) {
3040 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
3041 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3045 btrfs_orphan_release_metadata(inode);
3046 atomic_dec(&root->orphan_inodes);
3053 * this cleans up any orphans that may be left on the list from the last use
3056 int btrfs_orphan_cleanup(struct btrfs_root *root)
3058 struct btrfs_path *path;
3059 struct extent_buffer *leaf;
3060 struct btrfs_key key, found_key;
3061 struct btrfs_trans_handle *trans;
3062 struct inode *inode;
3063 u64 last_objectid = 0;
3064 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3066 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3069 path = btrfs_alloc_path();
3076 key.objectid = BTRFS_ORPHAN_OBJECTID;
3077 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3078 key.offset = (u64)-1;
3081 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3086 * if ret == 0 means we found what we were searching for, which
3087 * is weird, but possible, so only screw with path if we didn't
3088 * find the key and see if we have stuff that matches
3092 if (path->slots[0] == 0)
3097 /* pull out the item */
3098 leaf = path->nodes[0];
3099 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3101 /* make sure the item matches what we want */
3102 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3104 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3107 /* release the path since we're done with it */
3108 btrfs_release_path(path);
3111 * this is where we are basically btrfs_lookup, without the
3112 * crossing root thing. we store the inode number in the
3113 * offset of the orphan item.
3116 if (found_key.offset == last_objectid) {
3117 printk(KERN_ERR "btrfs: Error removing orphan entry, "
3118 "stopping orphan cleanup\n");
3123 last_objectid = found_key.offset;
3125 found_key.objectid = found_key.offset;
3126 found_key.type = BTRFS_INODE_ITEM_KEY;
3127 found_key.offset = 0;
3128 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3129 ret = PTR_RET(inode);
3130 if (ret && ret != -ESTALE)
3133 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3134 struct btrfs_root *dead_root;
3135 struct btrfs_fs_info *fs_info = root->fs_info;
3136 int is_dead_root = 0;
3139 * this is an orphan in the tree root. Currently these
3140 * could come from 2 sources:
3141 * a) a snapshot deletion in progress
3142 * b) a free space cache inode
3143 * We need to distinguish those two, as the snapshot
3144 * orphan must not get deleted.
3145 * find_dead_roots already ran before us, so if this
3146 * is a snapshot deletion, we should find the root
3147 * in the dead_roots list
3149 spin_lock(&fs_info->trans_lock);
3150 list_for_each_entry(dead_root, &fs_info->dead_roots,
3152 if (dead_root->root_key.objectid ==
3153 found_key.objectid) {
3158 spin_unlock(&fs_info->trans_lock);
3160 /* prevent this orphan from being found again */
3161 key.offset = found_key.objectid - 1;
3166 * Inode is already gone but the orphan item is still there,
3167 * kill the orphan item.
3169 if (ret == -ESTALE) {
3170 trans = btrfs_start_transaction(root, 1);
3171 if (IS_ERR(trans)) {
3172 ret = PTR_ERR(trans);
3175 printk(KERN_ERR "auto deleting %Lu\n",
3176 found_key.objectid);
3177 ret = btrfs_del_orphan_item(trans, root,
3178 found_key.objectid);
3179 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3180 btrfs_end_transaction(trans, root);
3185 * add this inode to the orphan list so btrfs_orphan_del does
3186 * the proper thing when we hit it
3188 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3189 &BTRFS_I(inode)->runtime_flags);
3190 atomic_inc(&root->orphan_inodes);
3192 /* if we have links, this was a truncate, lets do that */
3193 if (inode->i_nlink) {
3194 if (!S_ISREG(inode->i_mode)) {
3201 /* 1 for the orphan item deletion. */
3202 trans = btrfs_start_transaction(root, 1);
3203 if (IS_ERR(trans)) {
3204 ret = PTR_ERR(trans);
3207 ret = btrfs_orphan_add(trans, inode);
3208 btrfs_end_transaction(trans, root);
3212 ret = btrfs_truncate(inode);
3214 btrfs_orphan_del(NULL, inode);
3219 /* this will do delete_inode and everything for us */
3224 /* release the path since we're done with it */
3225 btrfs_release_path(path);
3227 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3229 if (root->orphan_block_rsv)
3230 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3233 if (root->orphan_block_rsv || root->orphan_item_inserted) {
3234 trans = btrfs_join_transaction(root);
3236 btrfs_end_transaction(trans, root);
3240 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
3242 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
3246 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
3247 btrfs_free_path(path);
3252 * very simple check to peek ahead in the leaf looking for xattrs. If we
3253 * don't find any xattrs, we know there can't be any acls.
3255 * slot is the slot the inode is in, objectid is the objectid of the inode
3257 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3258 int slot, u64 objectid)
3260 u32 nritems = btrfs_header_nritems(leaf);
3261 struct btrfs_key found_key;
3265 while (slot < nritems) {
3266 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3268 /* we found a different objectid, there must not be acls */
3269 if (found_key.objectid != objectid)
3272 /* we found an xattr, assume we've got an acl */
3273 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
3277 * we found a key greater than an xattr key, there can't
3278 * be any acls later on
3280 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3287 * it goes inode, inode backrefs, xattrs, extents,
3288 * so if there are a ton of hard links to an inode there can
3289 * be a lot of backrefs. Don't waste time searching too hard,
3290 * this is just an optimization
3295 /* we hit the end of the leaf before we found an xattr or
3296 * something larger than an xattr. We have to assume the inode
3303 * read an inode from the btree into the in-memory inode
3305 static void btrfs_read_locked_inode(struct inode *inode)
3307 struct btrfs_path *path;
3308 struct extent_buffer *leaf;
3309 struct btrfs_inode_item *inode_item;
3310 struct btrfs_timespec *tspec;
3311 struct btrfs_root *root = BTRFS_I(inode)->root;
3312 struct btrfs_key location;
3316 bool filled = false;
3318 ret = btrfs_fill_inode(inode, &rdev);
3322 path = btrfs_alloc_path();
3326 path->leave_spinning = 1;
3327 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3329 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3333 leaf = path->nodes[0];
3338 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3339 struct btrfs_inode_item);
3340 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3341 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3342 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3343 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3344 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3346 tspec = btrfs_inode_atime(inode_item);
3347 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3348 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3350 tspec = btrfs_inode_mtime(inode_item);
3351 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3352 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3354 tspec = btrfs_inode_ctime(inode_item);
3355 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3356 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3358 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3359 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3360 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3363 * If we were modified in the current generation and evicted from memory
3364 * and then re-read we need to do a full sync since we don't have any
3365 * idea about which extents were modified before we were evicted from
3368 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3369 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3370 &BTRFS_I(inode)->runtime_flags);
3372 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3373 inode->i_generation = BTRFS_I(inode)->generation;
3375 rdev = btrfs_inode_rdev(leaf, inode_item);
3377 BTRFS_I(inode)->index_cnt = (u64)-1;
3378 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3381 * try to precache a NULL acl entry for files that don't have
3382 * any xattrs or acls
3384 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3387 cache_no_acl(inode);
3389 btrfs_free_path(path);
3391 switch (inode->i_mode & S_IFMT) {
3393 inode->i_mapping->a_ops = &btrfs_aops;
3394 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3395 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3396 inode->i_fop = &btrfs_file_operations;
3397 inode->i_op = &btrfs_file_inode_operations;
3400 inode->i_fop = &btrfs_dir_file_operations;
3401 if (root == root->fs_info->tree_root)
3402 inode->i_op = &btrfs_dir_ro_inode_operations;
3404 inode->i_op = &btrfs_dir_inode_operations;
3407 inode->i_op = &btrfs_symlink_inode_operations;
3408 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3409 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3412 inode->i_op = &btrfs_special_inode_operations;
3413 init_special_inode(inode, inode->i_mode, rdev);
3417 btrfs_update_iflags(inode);
3421 btrfs_free_path(path);
3422 make_bad_inode(inode);
3426 * given a leaf and an inode, copy the inode fields into the leaf
3428 static void fill_inode_item(struct btrfs_trans_handle *trans,
3429 struct extent_buffer *leaf,
3430 struct btrfs_inode_item *item,
3431 struct inode *inode)
3433 struct btrfs_map_token token;
3435 btrfs_init_map_token(&token);
3437 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3438 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3439 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3441 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3442 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3444 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3445 inode->i_atime.tv_sec, &token);
3446 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3447 inode->i_atime.tv_nsec, &token);
3449 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3450 inode->i_mtime.tv_sec, &token);
3451 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3452 inode->i_mtime.tv_nsec, &token);
3454 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3455 inode->i_ctime.tv_sec, &token);
3456 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3457 inode->i_ctime.tv_nsec, &token);
3459 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3461 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3463 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3464 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3465 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3466 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3467 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3471 * copy everything in the in-memory inode into the btree.
3473 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3474 struct btrfs_root *root, struct inode *inode)
3476 struct btrfs_inode_item *inode_item;
3477 struct btrfs_path *path;
3478 struct extent_buffer *leaf;
3481 path = btrfs_alloc_path();
3485 path->leave_spinning = 1;
3486 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3494 btrfs_unlock_up_safe(path, 1);
3495 leaf = path->nodes[0];
3496 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3497 struct btrfs_inode_item);
3499 fill_inode_item(trans, leaf, inode_item, inode);
3500 btrfs_mark_buffer_dirty(leaf);
3501 btrfs_set_inode_last_trans(trans, inode);
3504 btrfs_free_path(path);
3509 * copy everything in the in-memory inode into the btree.
3511 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3512 struct btrfs_root *root, struct inode *inode)
3517 * If the inode is a free space inode, we can deadlock during commit
3518 * if we put it into the delayed code.
3520 * The data relocation inode should also be directly updated
3523 if (!btrfs_is_free_space_inode(inode)
3524 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3525 btrfs_update_root_times(trans, root);
3527 ret = btrfs_delayed_update_inode(trans, root, inode);
3529 btrfs_set_inode_last_trans(trans, inode);
3533 return btrfs_update_inode_item(trans, root, inode);
3536 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3537 struct btrfs_root *root,
3538 struct inode *inode)
3542 ret = btrfs_update_inode(trans, root, inode);
3544 return btrfs_update_inode_item(trans, root, inode);
3549 * unlink helper that gets used here in inode.c and in the tree logging
3550 * recovery code. It remove a link in a directory with a given name, and
3551 * also drops the back refs in the inode to the directory
3553 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3554 struct btrfs_root *root,
3555 struct inode *dir, struct inode *inode,
3556 const char *name, int name_len)
3558 struct btrfs_path *path;
3560 struct extent_buffer *leaf;
3561 struct btrfs_dir_item *di;
3562 struct btrfs_key key;
3564 u64 ino = btrfs_ino(inode);
3565 u64 dir_ino = btrfs_ino(dir);
3567 path = btrfs_alloc_path();
3573 path->leave_spinning = 1;
3574 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3575 name, name_len, -1);
3584 leaf = path->nodes[0];
3585 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3586 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3589 btrfs_release_path(path);
3591 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3594 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
3595 "inode %llu parent %llu\n", name_len, name,
3596 (unsigned long long)ino, (unsigned long long)dir_ino);
3597 btrfs_abort_transaction(trans, root, ret);
3601 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3603 btrfs_abort_transaction(trans, root, ret);
3607 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3609 if (ret != 0 && ret != -ENOENT) {
3610 btrfs_abort_transaction(trans, root, ret);
3614 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3619 btrfs_free_path(path);
3623 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3624 inode_inc_iversion(inode);
3625 inode_inc_iversion(dir);
3626 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3627 ret = btrfs_update_inode(trans, root, dir);
3632 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3633 struct btrfs_root *root,
3634 struct inode *dir, struct inode *inode,
3635 const char *name, int name_len)
3638 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3640 btrfs_drop_nlink(inode);
3641 ret = btrfs_update_inode(trans, root, inode);
3647 /* helper to check if there is any shared block in the path */
3648 static int check_path_shared(struct btrfs_root *root,
3649 struct btrfs_path *path)
3651 struct extent_buffer *eb;
3655 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
3658 if (!path->nodes[level])
3660 eb = path->nodes[level];
3661 if (!btrfs_block_can_be_shared(root, eb))
3663 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
3672 * helper to start transaction for unlink and rmdir.
3674 * unlink and rmdir are special in btrfs, they do not always free space.
3675 * so in enospc case, we should make sure they will free space before
3676 * allowing them to use the global metadata reservation.
3678 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
3679 struct dentry *dentry)
3681 struct btrfs_trans_handle *trans;
3682 struct btrfs_root *root = BTRFS_I(dir)->root;
3683 struct btrfs_path *path;
3684 struct btrfs_dir_item *di;
3685 struct inode *inode = dentry->d_inode;
3690 u64 ino = btrfs_ino(inode);
3691 u64 dir_ino = btrfs_ino(dir);
3694 * 1 for the possible orphan item
3695 * 1 for the dir item
3696 * 1 for the dir index
3697 * 1 for the inode ref
3700 trans = btrfs_start_transaction(root, 5);
3701 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3704 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
3705 return ERR_PTR(-ENOSPC);
3707 /* check if there is someone else holds reference */
3708 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
3709 return ERR_PTR(-ENOSPC);
3711 if (atomic_read(&inode->i_count) > 2)
3712 return ERR_PTR(-ENOSPC);
3714 if (xchg(&root->fs_info->enospc_unlink, 1))
3715 return ERR_PTR(-ENOSPC);
3717 path = btrfs_alloc_path();
3719 root->fs_info->enospc_unlink = 0;
3720 return ERR_PTR(-ENOMEM);
3723 /* 1 for the orphan item */
3724 trans = btrfs_start_transaction(root, 1);
3725 if (IS_ERR(trans)) {
3726 btrfs_free_path(path);
3727 root->fs_info->enospc_unlink = 0;
3731 path->skip_locking = 1;
3732 path->search_commit_root = 1;
3734 ret = btrfs_lookup_inode(trans, root, path,
3735 &BTRFS_I(dir)->location, 0);
3741 if (check_path_shared(root, path))
3746 btrfs_release_path(path);
3748 ret = btrfs_lookup_inode(trans, root, path,
3749 &BTRFS_I(inode)->location, 0);
3755 if (check_path_shared(root, path))
3760 btrfs_release_path(path);
3762 if (ret == 0 && S_ISREG(inode->i_mode)) {
3763 ret = btrfs_lookup_file_extent(trans, root, path,
3769 BUG_ON(ret == 0); /* Corruption */
3770 if (check_path_shared(root, path))
3772 btrfs_release_path(path);
3780 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3781 dentry->d_name.name, dentry->d_name.len, 0);
3787 if (check_path_shared(root, path))
3793 btrfs_release_path(path);
3795 ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3796 dentry->d_name.len, ino, dir_ino, 0,
3803 if (check_path_shared(root, path))
3806 btrfs_release_path(path);
3809 * This is a commit root search, if we can lookup inode item and other
3810 * relative items in the commit root, it means the transaction of
3811 * dir/file creation has been committed, and the dir index item that we
3812 * delay to insert has also been inserted into the commit root. So
3813 * we needn't worry about the delayed insertion of the dir index item
3816 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3817 dentry->d_name.name, dentry->d_name.len, 0);
3822 BUG_ON(ret == -ENOENT);
3823 if (check_path_shared(root, path))
3828 btrfs_free_path(path);
3829 /* Migrate the orphan reservation over */
3831 err = btrfs_block_rsv_migrate(trans->block_rsv,
3832 &root->fs_info->global_block_rsv,
3833 trans->bytes_reserved);
3836 btrfs_end_transaction(trans, root);
3837 root->fs_info->enospc_unlink = 0;
3838 return ERR_PTR(err);
3841 trans->block_rsv = &root->fs_info->global_block_rsv;
3845 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3846 struct btrfs_root *root)
3848 if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3849 btrfs_block_rsv_release(root, trans->block_rsv,
3850 trans->bytes_reserved);
3851 trans->block_rsv = &root->fs_info->trans_block_rsv;
3852 BUG_ON(!root->fs_info->enospc_unlink);
3853 root->fs_info->enospc_unlink = 0;
3855 btrfs_end_transaction(trans, root);
3858 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3860 struct btrfs_root *root = BTRFS_I(dir)->root;
3861 struct btrfs_trans_handle *trans;
3862 struct inode *inode = dentry->d_inode;
3865 trans = __unlink_start_trans(dir, dentry);
3867 return PTR_ERR(trans);
3869 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3871 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3872 dentry->d_name.name, dentry->d_name.len);
3876 if (inode->i_nlink == 0) {
3877 ret = btrfs_orphan_add(trans, inode);
3883 __unlink_end_trans(trans, root);
3884 btrfs_btree_balance_dirty(root);
3888 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3889 struct btrfs_root *root,
3890 struct inode *dir, u64 objectid,
3891 const char *name, int name_len)
3893 struct btrfs_path *path;
3894 struct extent_buffer *leaf;
3895 struct btrfs_dir_item *di;
3896 struct btrfs_key key;
3899 u64 dir_ino = btrfs_ino(dir);
3901 path = btrfs_alloc_path();
3905 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3906 name, name_len, -1);
3907 if (IS_ERR_OR_NULL(di)) {
3915 leaf = path->nodes[0];
3916 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3917 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3918 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3920 btrfs_abort_transaction(trans, root, ret);
3923 btrfs_release_path(path);
3925 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3926 objectid, root->root_key.objectid,
3927 dir_ino, &index, name, name_len);
3929 if (ret != -ENOENT) {
3930 btrfs_abort_transaction(trans, root, ret);
3933 di = btrfs_search_dir_index_item(root, path, dir_ino,
3935 if (IS_ERR_OR_NULL(di)) {
3940 btrfs_abort_transaction(trans, root, ret);
3944 leaf = path->nodes[0];
3945 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3946 btrfs_release_path(path);
3949 btrfs_release_path(path);
3951 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3953 btrfs_abort_transaction(trans, root, ret);
3957 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3958 inode_inc_iversion(dir);
3959 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3960 ret = btrfs_update_inode_fallback(trans, root, dir);
3962 btrfs_abort_transaction(trans, root, ret);
3964 btrfs_free_path(path);
3968 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3970 struct inode *inode = dentry->d_inode;
3972 struct btrfs_root *root = BTRFS_I(dir)->root;
3973 struct btrfs_trans_handle *trans;
3975 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3977 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3980 trans = __unlink_start_trans(dir, dentry);
3982 return PTR_ERR(trans);
3984 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3985 err = btrfs_unlink_subvol(trans, root, dir,
3986 BTRFS_I(inode)->location.objectid,
3987 dentry->d_name.name,
3988 dentry->d_name.len);
3992 err = btrfs_orphan_add(trans, inode);
3996 /* now the directory is empty */
3997 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3998 dentry->d_name.name, dentry->d_name.len);
4000 btrfs_i_size_write(inode, 0);
4002 __unlink_end_trans(trans, root);
4003 btrfs_btree_balance_dirty(root);
4009 * this can truncate away extent items, csum items and directory items.
4010 * It starts at a high offset and removes keys until it can't find
4011 * any higher than new_size
4013 * csum items that cross the new i_size are truncated to the new size
4016 * min_type is the minimum key type to truncate down to. If set to 0, this
4017 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4019 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4020 struct btrfs_root *root,
4021 struct inode *inode,
4022 u64 new_size, u32 min_type)
4024 struct btrfs_path *path;
4025 struct extent_buffer *leaf;
4026 struct btrfs_file_extent_item *fi;
4027 struct btrfs_key key;
4028 struct btrfs_key found_key;
4029 u64 extent_start = 0;
4030 u64 extent_num_bytes = 0;
4031 u64 extent_offset = 0;
4033 u32 found_type = (u8)-1;
4036 int pending_del_nr = 0;
4037 int pending_del_slot = 0;
4038 int extent_type = -1;
4041 u64 ino = btrfs_ino(inode);
4043 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4045 path = btrfs_alloc_path();
4051 * We want to drop from the next block forward in case this new size is
4052 * not block aligned since we will be keeping the last block of the
4053 * extent just the way it is.
4055 if (root->ref_cows || root == root->fs_info->tree_root)
4056 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4057 root->sectorsize), (u64)-1, 0);
4060 * This function is also used to drop the items in the log tree before
4061 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4062 * it is used to drop the loged items. So we shouldn't kill the delayed
4065 if (min_type == 0 && root == BTRFS_I(inode)->root)
4066 btrfs_kill_delayed_inode_items(inode);
4069 key.offset = (u64)-1;
4073 path->leave_spinning = 1;
4074 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4081 /* there are no items in the tree for us to truncate, we're
4084 if (path->slots[0] == 0)
4091 leaf = path->nodes[0];
4092 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4093 found_type = btrfs_key_type(&found_key);
4095 if (found_key.objectid != ino)
4098 if (found_type < min_type)
4101 item_end = found_key.offset;
4102 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4103 fi = btrfs_item_ptr(leaf, path->slots[0],
4104 struct btrfs_file_extent_item);
4105 extent_type = btrfs_file_extent_type(leaf, fi);
4106 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4108 btrfs_file_extent_num_bytes(leaf, fi);
4109 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4110 item_end += btrfs_file_extent_inline_len(leaf,
4115 if (found_type > min_type) {
4118 if (item_end < new_size)
4120 if (found_key.offset >= new_size)
4126 /* FIXME, shrink the extent if the ref count is only 1 */
4127 if (found_type != BTRFS_EXTENT_DATA_KEY)
4130 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4132 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4134 u64 orig_num_bytes =
4135 btrfs_file_extent_num_bytes(leaf, fi);
4136 extent_num_bytes = ALIGN(new_size -
4139 btrfs_set_file_extent_num_bytes(leaf, fi,
4141 num_dec = (orig_num_bytes -
4143 if (root->ref_cows && extent_start != 0)
4144 inode_sub_bytes(inode, num_dec);
4145 btrfs_mark_buffer_dirty(leaf);
4148 btrfs_file_extent_disk_num_bytes(leaf,
4150 extent_offset = found_key.offset -
4151 btrfs_file_extent_offset(leaf, fi);
4153 /* FIXME blocksize != 4096 */
4154 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4155 if (extent_start != 0) {
4158 inode_sub_bytes(inode, num_dec);
4161 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4163 * we can't truncate inline items that have had
4167 btrfs_file_extent_compression(leaf, fi) == 0 &&
4168 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4169 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4170 u32 size = new_size - found_key.offset;
4172 if (root->ref_cows) {
4173 inode_sub_bytes(inode, item_end + 1 -
4177 btrfs_file_extent_calc_inline_size(size);
4178 btrfs_truncate_item(trans, root, path,
4180 } else if (root->ref_cows) {
4181 inode_sub_bytes(inode, item_end + 1 -
4187 if (!pending_del_nr) {
4188 /* no pending yet, add ourselves */
4189 pending_del_slot = path->slots[0];
4191 } else if (pending_del_nr &&
4192 path->slots[0] + 1 == pending_del_slot) {
4193 /* hop on the pending chunk */
4195 pending_del_slot = path->slots[0];
4202 if (found_extent && (root->ref_cows ||
4203 root == root->fs_info->tree_root)) {
4204 btrfs_set_path_blocking(path);
4205 ret = btrfs_free_extent(trans, root, extent_start,
4206 extent_num_bytes, 0,
4207 btrfs_header_owner(leaf),
4208 ino, extent_offset, 0);
4212 if (found_type == BTRFS_INODE_ITEM_KEY)
4215 if (path->slots[0] == 0 ||
4216 path->slots[0] != pending_del_slot) {
4217 if (pending_del_nr) {
4218 ret = btrfs_del_items(trans, root, path,
4222 btrfs_abort_transaction(trans,
4228 btrfs_release_path(path);
4235 if (pending_del_nr) {
4236 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4239 btrfs_abort_transaction(trans, root, ret);
4242 btrfs_free_path(path);
4247 * btrfs_truncate_page - read, zero a chunk and write a page
4248 * @inode - inode that we're zeroing
4249 * @from - the offset to start zeroing
4250 * @len - the length to zero, 0 to zero the entire range respective to the
4252 * @front - zero up to the offset instead of from the offset on
4254 * This will find the page for the "from" offset and cow the page and zero the
4255 * part we want to zero. This is used with truncate and hole punching.
4257 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4260 struct address_space *mapping = inode->i_mapping;
4261 struct btrfs_root *root = BTRFS_I(inode)->root;
4262 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4263 struct btrfs_ordered_extent *ordered;
4264 struct extent_state *cached_state = NULL;
4266 u32 blocksize = root->sectorsize;
4267 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4268 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4270 gfp_t mask = btrfs_alloc_write_mask(mapping);
4275 if ((offset & (blocksize - 1)) == 0 &&
4276 (!len || ((len & (blocksize - 1)) == 0)))
4278 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4283 page = find_or_create_page(mapping, index, mask);
4285 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4290 page_start = page_offset(page);
4291 page_end = page_start + PAGE_CACHE_SIZE - 1;
4293 if (!PageUptodate(page)) {
4294 ret = btrfs_readpage(NULL, page);
4296 if (page->mapping != mapping) {
4298 page_cache_release(page);
4301 if (!PageUptodate(page)) {
4306 wait_on_page_writeback(page);
4308 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4309 set_page_extent_mapped(page);
4311 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4313 unlock_extent_cached(io_tree, page_start, page_end,
4314 &cached_state, GFP_NOFS);
4316 page_cache_release(page);
4317 btrfs_start_ordered_extent(inode, ordered, 1);
4318 btrfs_put_ordered_extent(ordered);
4322 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4323 EXTENT_DIRTY | EXTENT_DELALLOC |
4324 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4325 0, 0, &cached_state, GFP_NOFS);
4327 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4330 unlock_extent_cached(io_tree, page_start, page_end,
4331 &cached_state, GFP_NOFS);
4335 if (offset != PAGE_CACHE_SIZE) {
4337 len = PAGE_CACHE_SIZE - offset;
4340 memset(kaddr, 0, offset);
4342 memset(kaddr + offset, 0, len);
4343 flush_dcache_page(page);
4346 ClearPageChecked(page);
4347 set_page_dirty(page);
4348 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4353 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4355 page_cache_release(page);
4361 * This function puts in dummy file extents for the area we're creating a hole
4362 * for. So if we are truncating this file to a larger size we need to insert
4363 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4364 * the range between oldsize and size
4366 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4368 struct btrfs_trans_handle *trans;
4369 struct btrfs_root *root = BTRFS_I(inode)->root;
4370 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4371 struct extent_map *em = NULL;
4372 struct extent_state *cached_state = NULL;
4373 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4374 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4375 u64 block_end = ALIGN(size, root->sectorsize);
4381 if (size <= hole_start)
4385 struct btrfs_ordered_extent *ordered;
4386 btrfs_wait_ordered_range(inode, hole_start,
4387 block_end - hole_start);
4388 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4390 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
4393 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4394 &cached_state, GFP_NOFS);
4395 btrfs_put_ordered_extent(ordered);
4398 cur_offset = hole_start;
4400 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4401 block_end - cur_offset, 0);
4407 last_byte = min(extent_map_end(em), block_end);
4408 last_byte = ALIGN(last_byte , root->sectorsize);
4409 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4410 struct extent_map *hole_em;
4411 hole_size = last_byte - cur_offset;
4413 trans = btrfs_start_transaction(root, 3);
4414 if (IS_ERR(trans)) {
4415 err = PTR_ERR(trans);
4419 err = btrfs_drop_extents(trans, root, inode,
4421 cur_offset + hole_size, 1);
4423 btrfs_abort_transaction(trans, root, err);
4424 btrfs_end_transaction(trans, root);
4428 err = btrfs_insert_file_extent(trans, root,
4429 btrfs_ino(inode), cur_offset, 0,
4430 0, hole_size, 0, hole_size,
4433 btrfs_abort_transaction(trans, root, err);
4434 btrfs_end_transaction(trans, root);
4438 btrfs_drop_extent_cache(inode, cur_offset,
4439 cur_offset + hole_size - 1, 0);
4440 hole_em = alloc_extent_map();
4442 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4443 &BTRFS_I(inode)->runtime_flags);
4446 hole_em->start = cur_offset;
4447 hole_em->len = hole_size;
4448 hole_em->orig_start = cur_offset;
4450 hole_em->block_start = EXTENT_MAP_HOLE;
4451 hole_em->block_len = 0;
4452 hole_em->orig_block_len = 0;
4453 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4454 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4455 hole_em->generation = trans->transid;
4458 write_lock(&em_tree->lock);
4459 err = add_extent_mapping(em_tree, hole_em);
4461 list_move(&hole_em->list,
4462 &em_tree->modified_extents);
4463 write_unlock(&em_tree->lock);
4466 btrfs_drop_extent_cache(inode, cur_offset,
4470 free_extent_map(hole_em);
4472 btrfs_update_inode(trans, root, inode);
4473 btrfs_end_transaction(trans, root);
4475 free_extent_map(em);
4477 cur_offset = last_byte;
4478 if (cur_offset >= block_end)
4482 free_extent_map(em);
4483 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4488 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4490 struct btrfs_root *root = BTRFS_I(inode)->root;
4491 struct btrfs_trans_handle *trans;
4492 loff_t oldsize = i_size_read(inode);
4493 loff_t newsize = attr->ia_size;
4494 int mask = attr->ia_valid;
4497 if (newsize == oldsize)
4501 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4502 * special case where we need to update the times despite not having
4503 * these flags set. For all other operations the VFS set these flags
4504 * explicitly if it wants a timestamp update.
4506 if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4507 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4509 if (newsize > oldsize) {
4510 truncate_pagecache(inode, oldsize, newsize);
4511 ret = btrfs_cont_expand(inode, oldsize, newsize);
4515 trans = btrfs_start_transaction(root, 1);
4517 return PTR_ERR(trans);
4519 i_size_write(inode, newsize);
4520 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4521 ret = btrfs_update_inode(trans, root, inode);
4522 btrfs_end_transaction(trans, root);
4526 * We're truncating a file that used to have good data down to
4527 * zero. Make sure it gets into the ordered flush list so that
4528 * any new writes get down to disk quickly.
4531 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4532 &BTRFS_I(inode)->runtime_flags);
4535 * 1 for the orphan item we're going to add
4536 * 1 for the orphan item deletion.
4538 trans = btrfs_start_transaction(root, 2);
4540 return PTR_ERR(trans);
4543 * We need to do this in case we fail at _any_ point during the
4544 * actual truncate. Once we do the truncate_setsize we could
4545 * invalidate pages which forces any outstanding ordered io to
4546 * be instantly completed which will give us extents that need
4547 * to be truncated. If we fail to get an orphan inode down we
4548 * could have left over extents that were never meant to live,
4549 * so we need to garuntee from this point on that everything
4550 * will be consistent.
4552 ret = btrfs_orphan_add(trans, inode);
4553 btrfs_end_transaction(trans, root);
4557 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4558 truncate_setsize(inode, newsize);
4560 /* Disable nonlocked read DIO to avoid the end less truncate */
4561 btrfs_inode_block_unlocked_dio(inode);
4562 inode_dio_wait(inode);
4563 btrfs_inode_resume_unlocked_dio(inode);
4565 ret = btrfs_truncate(inode);
4566 if (ret && inode->i_nlink)
4567 btrfs_orphan_del(NULL, inode);
4573 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4575 struct inode *inode = dentry->d_inode;
4576 struct btrfs_root *root = BTRFS_I(inode)->root;
4579 if (btrfs_root_readonly(root))
4582 err = inode_change_ok(inode, attr);
4586 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4587 err = btrfs_setsize(inode, attr);
4592 if (attr->ia_valid) {
4593 setattr_copy(inode, attr);
4594 inode_inc_iversion(inode);
4595 err = btrfs_dirty_inode(inode);
4597 if (!err && attr->ia_valid & ATTR_MODE)
4598 err = btrfs_acl_chmod(inode);
4604 void btrfs_evict_inode(struct inode *inode)
4606 struct btrfs_trans_handle *trans;
4607 struct btrfs_root *root = BTRFS_I(inode)->root;
4608 struct btrfs_block_rsv *rsv, *global_rsv;
4609 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4612 trace_btrfs_inode_evict(inode);
4614 truncate_inode_pages(&inode->i_data, 0);
4615 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
4616 btrfs_is_free_space_inode(inode)))
4619 if (is_bad_inode(inode)) {
4620 btrfs_orphan_del(NULL, inode);
4623 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4624 btrfs_wait_ordered_range(inode, 0, (u64)-1);
4626 if (root->fs_info->log_root_recovering) {
4627 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4628 &BTRFS_I(inode)->runtime_flags));
4632 if (inode->i_nlink > 0) {
4633 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
4637 ret = btrfs_commit_inode_delayed_inode(inode);
4639 btrfs_orphan_del(NULL, inode);
4643 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4645 btrfs_orphan_del(NULL, inode);
4648 rsv->size = min_size;
4650 global_rsv = &root->fs_info->global_block_rsv;
4652 btrfs_i_size_write(inode, 0);
4655 * This is a bit simpler than btrfs_truncate since we've already
4656 * reserved our space for our orphan item in the unlink, so we just
4657 * need to reserve some slack space in case we add bytes and update
4658 * inode item when doing the truncate.
4661 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4662 BTRFS_RESERVE_FLUSH_LIMIT);
4665 * Try and steal from the global reserve since we will
4666 * likely not use this space anyway, we want to try as
4667 * hard as possible to get this to work.
4670 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4673 printk(KERN_WARNING "Could not get space for a "
4674 "delete, will truncate on mount %d\n", ret);
4675 btrfs_orphan_del(NULL, inode);
4676 btrfs_free_block_rsv(root, rsv);
4680 trans = btrfs_join_transaction(root);
4681 if (IS_ERR(trans)) {
4682 btrfs_orphan_del(NULL, inode);
4683 btrfs_free_block_rsv(root, rsv);
4687 trans->block_rsv = rsv;
4689 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4693 trans->block_rsv = &root->fs_info->trans_block_rsv;
4694 btrfs_end_transaction(trans, root);
4696 btrfs_btree_balance_dirty(root);
4699 btrfs_free_block_rsv(root, rsv);
4702 trans->block_rsv = root->orphan_block_rsv;
4703 ret = btrfs_orphan_del(trans, inode);
4707 trans->block_rsv = &root->fs_info->trans_block_rsv;
4708 if (!(root == root->fs_info->tree_root ||
4709 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4710 btrfs_return_ino(root, btrfs_ino(inode));
4712 btrfs_end_transaction(trans, root);
4713 btrfs_btree_balance_dirty(root);
4720 * this returns the key found in the dir entry in the location pointer.
4721 * If no dir entries were found, location->objectid is 0.
4723 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4724 struct btrfs_key *location)
4726 const char *name = dentry->d_name.name;
4727 int namelen = dentry->d_name.len;
4728 struct btrfs_dir_item *di;
4729 struct btrfs_path *path;
4730 struct btrfs_root *root = BTRFS_I(dir)->root;
4733 path = btrfs_alloc_path();
4737 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4742 if (IS_ERR_OR_NULL(di))
4745 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4747 btrfs_free_path(path);
4750 location->objectid = 0;
4755 * when we hit a tree root in a directory, the btrfs part of the inode
4756 * needs to be changed to reflect the root directory of the tree root. This
4757 * is kind of like crossing a mount point.
4759 static int fixup_tree_root_location(struct btrfs_root *root,
4761 struct dentry *dentry,
4762 struct btrfs_key *location,
4763 struct btrfs_root **sub_root)
4765 struct btrfs_path *path;
4766 struct btrfs_root *new_root;
4767 struct btrfs_root_ref *ref;
4768 struct extent_buffer *leaf;
4772 path = btrfs_alloc_path();
4779 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4780 BTRFS_I(dir)->root->root_key.objectid,
4781 location->objectid);
4788 leaf = path->nodes[0];
4789 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4790 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4791 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4794 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4795 (unsigned long)(ref + 1),
4796 dentry->d_name.len);
4800 btrfs_release_path(path);
4802 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4803 if (IS_ERR(new_root)) {
4804 err = PTR_ERR(new_root);
4808 if (btrfs_root_refs(&new_root->root_item) == 0) {
4813 *sub_root = new_root;
4814 location->objectid = btrfs_root_dirid(&new_root->root_item);
4815 location->type = BTRFS_INODE_ITEM_KEY;
4816 location->offset = 0;
4819 btrfs_free_path(path);
4823 static void inode_tree_add(struct inode *inode)
4825 struct btrfs_root *root = BTRFS_I(inode)->root;
4826 struct btrfs_inode *entry;
4828 struct rb_node *parent;
4829 u64 ino = btrfs_ino(inode);
4831 p = &root->inode_tree.rb_node;
4834 if (inode_unhashed(inode))
4837 spin_lock(&root->inode_lock);
4840 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4842 if (ino < btrfs_ino(&entry->vfs_inode))
4843 p = &parent->rb_left;
4844 else if (ino > btrfs_ino(&entry->vfs_inode))
4845 p = &parent->rb_right;
4847 WARN_ON(!(entry->vfs_inode.i_state &
4848 (I_WILL_FREE | I_FREEING)));
4849 rb_erase(parent, &root->inode_tree);
4850 RB_CLEAR_NODE(parent);
4851 spin_unlock(&root->inode_lock);
4855 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4856 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4857 spin_unlock(&root->inode_lock);
4860 static void inode_tree_del(struct inode *inode)
4862 struct btrfs_root *root = BTRFS_I(inode)->root;
4865 spin_lock(&root->inode_lock);
4866 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4867 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4868 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4869 empty = RB_EMPTY_ROOT(&root->inode_tree);
4871 spin_unlock(&root->inode_lock);
4874 * Free space cache has inodes in the tree root, but the tree root has a
4875 * root_refs of 0, so this could end up dropping the tree root as a
4876 * snapshot, so we need the extra !root->fs_info->tree_root check to
4877 * make sure we don't drop it.
4879 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4880 root != root->fs_info->tree_root) {
4881 synchronize_srcu(&root->fs_info->subvol_srcu);
4882 spin_lock(&root->inode_lock);
4883 empty = RB_EMPTY_ROOT(&root->inode_tree);
4884 spin_unlock(&root->inode_lock);
4886 btrfs_add_dead_root(root);
4890 void btrfs_invalidate_inodes(struct btrfs_root *root)
4892 struct rb_node *node;
4893 struct rb_node *prev;
4894 struct btrfs_inode *entry;
4895 struct inode *inode;
4898 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4900 spin_lock(&root->inode_lock);
4902 node = root->inode_tree.rb_node;
4906 entry = rb_entry(node, struct btrfs_inode, rb_node);
4908 if (objectid < btrfs_ino(&entry->vfs_inode))
4909 node = node->rb_left;
4910 else if (objectid > btrfs_ino(&entry->vfs_inode))
4911 node = node->rb_right;
4917 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4918 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4922 prev = rb_next(prev);
4926 entry = rb_entry(node, struct btrfs_inode, rb_node);
4927 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4928 inode = igrab(&entry->vfs_inode);
4930 spin_unlock(&root->inode_lock);
4931 if (atomic_read(&inode->i_count) > 1)
4932 d_prune_aliases(inode);
4934 * btrfs_drop_inode will have it removed from
4935 * the inode cache when its usage count
4940 spin_lock(&root->inode_lock);
4944 if (cond_resched_lock(&root->inode_lock))
4947 node = rb_next(node);
4949 spin_unlock(&root->inode_lock);
4952 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4954 struct btrfs_iget_args *args = p;
4955 inode->i_ino = args->ino;
4956 BTRFS_I(inode)->root = args->root;
4960 static int btrfs_find_actor(struct inode *inode, void *opaque)
4962 struct btrfs_iget_args *args = opaque;
4963 return args->ino == btrfs_ino(inode) &&
4964 args->root == BTRFS_I(inode)->root;
4967 static struct inode *btrfs_iget_locked(struct super_block *s,
4969 struct btrfs_root *root)
4971 struct inode *inode;
4972 struct btrfs_iget_args args;
4973 args.ino = objectid;
4976 inode = iget5_locked(s, objectid, btrfs_find_actor,
4977 btrfs_init_locked_inode,
4982 /* Get an inode object given its location and corresponding root.
4983 * Returns in *is_new if the inode was read from disk
4985 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4986 struct btrfs_root *root, int *new)
4988 struct inode *inode;
4990 inode = btrfs_iget_locked(s, location->objectid, root);
4992 return ERR_PTR(-ENOMEM);
4994 if (inode->i_state & I_NEW) {
4995 BTRFS_I(inode)->root = root;
4996 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4997 btrfs_read_locked_inode(inode);
4998 if (!is_bad_inode(inode)) {
4999 inode_tree_add(inode);
5000 unlock_new_inode(inode);
5004 unlock_new_inode(inode);
5006 inode = ERR_PTR(-ESTALE);
5013 static struct inode *new_simple_dir(struct super_block *s,
5014 struct btrfs_key *key,
5015 struct btrfs_root *root)
5017 struct inode *inode = new_inode(s);
5020 return ERR_PTR(-ENOMEM);
5022 BTRFS_I(inode)->root = root;
5023 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5024 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5026 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5027 inode->i_op = &btrfs_dir_ro_inode_operations;
5028 inode->i_fop = &simple_dir_operations;
5029 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5030 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5035 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5037 struct inode *inode;
5038 struct btrfs_root *root = BTRFS_I(dir)->root;
5039 struct btrfs_root *sub_root = root;
5040 struct btrfs_key location;
5044 if (dentry->d_name.len > BTRFS_NAME_LEN)
5045 return ERR_PTR(-ENAMETOOLONG);
5047 ret = btrfs_inode_by_name(dir, dentry, &location);
5049 return ERR_PTR(ret);
5051 if (location.objectid == 0)
5054 if (location.type == BTRFS_INODE_ITEM_KEY) {
5055 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5059 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5061 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5062 ret = fixup_tree_root_location(root, dir, dentry,
5063 &location, &sub_root);
5066 inode = ERR_PTR(ret);
5068 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5070 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5072 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5074 if (!IS_ERR(inode) && root != sub_root) {
5075 down_read(&root->fs_info->cleanup_work_sem);
5076 if (!(inode->i_sb->s_flags & MS_RDONLY))
5077 ret = btrfs_orphan_cleanup(sub_root);
5078 up_read(&root->fs_info->cleanup_work_sem);
5080 inode = ERR_PTR(ret);
5086 static int btrfs_dentry_delete(const struct dentry *dentry)
5088 struct btrfs_root *root;
5089 struct inode *inode = dentry->d_inode;
5091 if (!inode && !IS_ROOT(dentry))
5092 inode = dentry->d_parent->d_inode;
5095 root = BTRFS_I(inode)->root;
5096 if (btrfs_root_refs(&root->root_item) == 0)
5099 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5105 static void btrfs_dentry_release(struct dentry *dentry)
5107 if (dentry->d_fsdata)
5108 kfree(dentry->d_fsdata);
5111 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5116 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
5120 unsigned char btrfs_filetype_table[] = {
5121 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5124 static int btrfs_real_readdir(struct file *filp, void *dirent,
5127 struct inode *inode = file_inode(filp);
5128 struct btrfs_root *root = BTRFS_I(inode)->root;
5129 struct btrfs_item *item;
5130 struct btrfs_dir_item *di;
5131 struct btrfs_key key;
5132 struct btrfs_key found_key;
5133 struct btrfs_path *path;
5134 struct list_head ins_list;
5135 struct list_head del_list;
5137 struct extent_buffer *leaf;
5139 unsigned char d_type;
5144 int key_type = BTRFS_DIR_INDEX_KEY;
5148 int is_curr = 0; /* filp->f_pos points to the current index? */
5150 /* FIXME, use a real flag for deciding about the key type */
5151 if (root->fs_info->tree_root == root)
5152 key_type = BTRFS_DIR_ITEM_KEY;
5154 /* special case for "." */
5155 if (filp->f_pos == 0) {
5156 over = filldir(dirent, ".", 1,
5157 filp->f_pos, btrfs_ino(inode), DT_DIR);
5162 /* special case for .., just use the back ref */
5163 if (filp->f_pos == 1) {
5164 u64 pino = parent_ino(filp->f_path.dentry);
5165 over = filldir(dirent, "..", 2,
5166 filp->f_pos, pino, DT_DIR);
5171 path = btrfs_alloc_path();
5177 if (key_type == BTRFS_DIR_INDEX_KEY) {
5178 INIT_LIST_HEAD(&ins_list);
5179 INIT_LIST_HEAD(&del_list);
5180 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5183 btrfs_set_key_type(&key, key_type);
5184 key.offset = filp->f_pos;
5185 key.objectid = btrfs_ino(inode);
5187 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5192 leaf = path->nodes[0];
5193 slot = path->slots[0];
5194 if (slot >= btrfs_header_nritems(leaf)) {
5195 ret = btrfs_next_leaf(root, path);
5203 item = btrfs_item_nr(leaf, slot);
5204 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5206 if (found_key.objectid != key.objectid)
5208 if (btrfs_key_type(&found_key) != key_type)
5210 if (found_key.offset < filp->f_pos)
5212 if (key_type == BTRFS_DIR_INDEX_KEY &&
5213 btrfs_should_delete_dir_index(&del_list,
5217 filp->f_pos = found_key.offset;
5220 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5222 di_total = btrfs_item_size(leaf, item);
5224 while (di_cur < di_total) {
5225 struct btrfs_key location;
5227 if (verify_dir_item(root, leaf, di))
5230 name_len = btrfs_dir_name_len(leaf, di);
5231 if (name_len <= sizeof(tmp_name)) {
5232 name_ptr = tmp_name;
5234 name_ptr = kmalloc(name_len, GFP_NOFS);
5240 read_extent_buffer(leaf, name_ptr,
5241 (unsigned long)(di + 1), name_len);
5243 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5244 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5247 /* is this a reference to our own snapshot? If so
5250 * In contrast to old kernels, we insert the snapshot's
5251 * dir item and dir index after it has been created, so
5252 * we won't find a reference to our own snapshot. We
5253 * still keep the following code for backward
5256 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5257 location.objectid == root->root_key.objectid) {
5261 over = filldir(dirent, name_ptr, name_len,
5262 found_key.offset, location.objectid,
5266 if (name_ptr != tmp_name)
5271 di_len = btrfs_dir_name_len(leaf, di) +
5272 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5274 di = (struct btrfs_dir_item *)((char *)di + di_len);
5280 if (key_type == BTRFS_DIR_INDEX_KEY) {
5283 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
5289 /* Reached end of directory/root. Bump pos past the last item. */
5290 if (key_type == BTRFS_DIR_INDEX_KEY)
5292 * 32-bit glibc will use getdents64, but then strtol -
5293 * so the last number we can serve is this.
5295 filp->f_pos = 0x7fffffff;
5301 if (key_type == BTRFS_DIR_INDEX_KEY)
5302 btrfs_put_delayed_items(&ins_list, &del_list);
5303 btrfs_free_path(path);
5307 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5309 struct btrfs_root *root = BTRFS_I(inode)->root;
5310 struct btrfs_trans_handle *trans;
5312 bool nolock = false;
5314 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5317 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5320 if (wbc->sync_mode == WB_SYNC_ALL) {
5322 trans = btrfs_join_transaction_nolock(root);
5324 trans = btrfs_join_transaction(root);
5326 return PTR_ERR(trans);
5327 ret = btrfs_commit_transaction(trans, root);
5333 * This is somewhat expensive, updating the tree every time the
5334 * inode changes. But, it is most likely to find the inode in cache.
5335 * FIXME, needs more benchmarking...there are no reasons other than performance
5336 * to keep or drop this code.
5338 int btrfs_dirty_inode(struct inode *inode)
5340 struct btrfs_root *root = BTRFS_I(inode)->root;
5341 struct btrfs_trans_handle *trans;
5344 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5347 trans = btrfs_join_transaction(root);
5349 return PTR_ERR(trans);
5351 ret = btrfs_update_inode(trans, root, inode);
5352 if (ret && ret == -ENOSPC) {
5353 /* whoops, lets try again with the full transaction */
5354 btrfs_end_transaction(trans, root);
5355 trans = btrfs_start_transaction(root, 1);
5357 return PTR_ERR(trans);
5359 ret = btrfs_update_inode(trans, root, inode);
5361 btrfs_end_transaction(trans, root);
5362 if (BTRFS_I(inode)->delayed_node)
5363 btrfs_balance_delayed_items(root);
5369 * This is a copy of file_update_time. We need this so we can return error on
5370 * ENOSPC for updating the inode in the case of file write and mmap writes.
5372 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5375 struct btrfs_root *root = BTRFS_I(inode)->root;
5377 if (btrfs_root_readonly(root))
5380 if (flags & S_VERSION)
5381 inode_inc_iversion(inode);
5382 if (flags & S_CTIME)
5383 inode->i_ctime = *now;
5384 if (flags & S_MTIME)
5385 inode->i_mtime = *now;
5386 if (flags & S_ATIME)
5387 inode->i_atime = *now;
5388 return btrfs_dirty_inode(inode);
5392 * find the highest existing sequence number in a directory
5393 * and then set the in-memory index_cnt variable to reflect
5394 * free sequence numbers
5396 static int btrfs_set_inode_index_count(struct inode *inode)
5398 struct btrfs_root *root = BTRFS_I(inode)->root;
5399 struct btrfs_key key, found_key;
5400 struct btrfs_path *path;
5401 struct extent_buffer *leaf;
5404 key.objectid = btrfs_ino(inode);
5405 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5406 key.offset = (u64)-1;
5408 path = btrfs_alloc_path();
5412 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5415 /* FIXME: we should be able to handle this */
5421 * MAGIC NUMBER EXPLANATION:
5422 * since we search a directory based on f_pos we have to start at 2
5423 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5424 * else has to start at 2
5426 if (path->slots[0] == 0) {
5427 BTRFS_I(inode)->index_cnt = 2;
5433 leaf = path->nodes[0];
5434 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5436 if (found_key.objectid != btrfs_ino(inode) ||
5437 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5438 BTRFS_I(inode)->index_cnt = 2;
5442 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5444 btrfs_free_path(path);
5449 * helper to find a free sequence number in a given directory. This current
5450 * code is very simple, later versions will do smarter things in the btree
5452 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5456 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5457 ret = btrfs_inode_delayed_dir_index_count(dir);
5459 ret = btrfs_set_inode_index_count(dir);
5465 *index = BTRFS_I(dir)->index_cnt;
5466 BTRFS_I(dir)->index_cnt++;
5471 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5472 struct btrfs_root *root,
5474 const char *name, int name_len,
5475 u64 ref_objectid, u64 objectid,
5476 umode_t mode, u64 *index)
5478 struct inode *inode;
5479 struct btrfs_inode_item *inode_item;
5480 struct btrfs_key *location;
5481 struct btrfs_path *path;
5482 struct btrfs_inode_ref *ref;
5483 struct btrfs_key key[2];
5489 path = btrfs_alloc_path();
5491 return ERR_PTR(-ENOMEM);
5493 inode = new_inode(root->fs_info->sb);
5495 btrfs_free_path(path);
5496 return ERR_PTR(-ENOMEM);
5500 * we have to initialize this early, so we can reclaim the inode
5501 * number if we fail afterwards in this function.
5503 inode->i_ino = objectid;
5506 trace_btrfs_inode_request(dir);
5508 ret = btrfs_set_inode_index(dir, index);
5510 btrfs_free_path(path);
5512 return ERR_PTR(ret);
5516 * index_cnt is ignored for everything but a dir,
5517 * btrfs_get_inode_index_count has an explanation for the magic
5520 BTRFS_I(inode)->index_cnt = 2;
5521 BTRFS_I(inode)->root = root;
5522 BTRFS_I(inode)->generation = trans->transid;
5523 inode->i_generation = BTRFS_I(inode)->generation;
5526 * We could have gotten an inode number from somebody who was fsynced
5527 * and then removed in this same transaction, so let's just set full
5528 * sync since it will be a full sync anyway and this will blow away the
5529 * old info in the log.
5531 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5538 key[0].objectid = objectid;
5539 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5543 * Start new inodes with an inode_ref. This is slightly more
5544 * efficient for small numbers of hard links since they will
5545 * be packed into one item. Extended refs will kick in if we
5546 * add more hard links than can fit in the ref item.
5548 key[1].objectid = objectid;
5549 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5550 key[1].offset = ref_objectid;
5552 sizes[0] = sizeof(struct btrfs_inode_item);
5553 sizes[1] = name_len + sizeof(*ref);
5555 path->leave_spinning = 1;
5556 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5560 inode_init_owner(inode, dir, mode);
5561 inode_set_bytes(inode, 0);
5562 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5563 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5564 struct btrfs_inode_item);
5565 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5566 sizeof(*inode_item));
5567 fill_inode_item(trans, path->nodes[0], inode_item, inode);
5569 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5570 struct btrfs_inode_ref);
5571 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5572 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5573 ptr = (unsigned long)(ref + 1);
5574 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5576 btrfs_mark_buffer_dirty(path->nodes[0]);
5577 btrfs_free_path(path);
5579 location = &BTRFS_I(inode)->location;
5580 location->objectid = objectid;
5581 location->offset = 0;
5582 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5584 btrfs_inherit_iflags(inode, dir);
5586 if (S_ISREG(mode)) {
5587 if (btrfs_test_opt(root, NODATASUM))
5588 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5589 if (btrfs_test_opt(root, NODATACOW))
5590 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5591 BTRFS_INODE_NODATASUM;
5594 insert_inode_hash(inode);
5595 inode_tree_add(inode);
5597 trace_btrfs_inode_new(inode);
5598 btrfs_set_inode_last_trans(trans, inode);
5600 btrfs_update_root_times(trans, root);
5605 BTRFS_I(dir)->index_cnt--;
5606 btrfs_free_path(path);
5608 return ERR_PTR(ret);
5611 static inline u8 btrfs_inode_type(struct inode *inode)
5613 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5617 * utility function to add 'inode' into 'parent_inode' with
5618 * a give name and a given sequence number.
5619 * if 'add_backref' is true, also insert a backref from the
5620 * inode to the parent directory.
5622 int btrfs_add_link(struct btrfs_trans_handle *trans,
5623 struct inode *parent_inode, struct inode *inode,
5624 const char *name, int name_len, int add_backref, u64 index)
5627 struct btrfs_key key;
5628 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5629 u64 ino = btrfs_ino(inode);
5630 u64 parent_ino = btrfs_ino(parent_inode);
5632 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5633 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5636 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5640 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5641 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5642 key.objectid, root->root_key.objectid,
5643 parent_ino, index, name, name_len);
5644 } else if (add_backref) {
5645 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5649 /* Nothing to clean up yet */
5653 ret = btrfs_insert_dir_item(trans, root, name, name_len,
5655 btrfs_inode_type(inode), index);
5656 if (ret == -EEXIST || ret == -EOVERFLOW)
5659 btrfs_abort_transaction(trans, root, ret);
5663 btrfs_i_size_write(parent_inode, parent_inode->i_size +
5665 inode_inc_iversion(parent_inode);
5666 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5667 ret = btrfs_update_inode(trans, root, parent_inode);
5669 btrfs_abort_transaction(trans, root, ret);
5673 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5676 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5677 key.objectid, root->root_key.objectid,
5678 parent_ino, &local_index, name, name_len);
5680 } else if (add_backref) {
5684 err = btrfs_del_inode_ref(trans, root, name, name_len,
5685 ino, parent_ino, &local_index);
5690 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5691 struct inode *dir, struct dentry *dentry,
5692 struct inode *inode, int backref, u64 index)
5694 int err = btrfs_add_link(trans, dir, inode,
5695 dentry->d_name.name, dentry->d_name.len,
5702 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5703 umode_t mode, dev_t rdev)
5705 struct btrfs_trans_handle *trans;
5706 struct btrfs_root *root = BTRFS_I(dir)->root;
5707 struct inode *inode = NULL;
5713 if (!new_valid_dev(rdev))
5717 * 2 for inode item and ref
5719 * 1 for xattr if selinux is on
5721 trans = btrfs_start_transaction(root, 5);
5723 return PTR_ERR(trans);
5725 err = btrfs_find_free_ino(root, &objectid);
5729 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5730 dentry->d_name.len, btrfs_ino(dir), objectid,
5732 if (IS_ERR(inode)) {
5733 err = PTR_ERR(inode);
5737 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5744 * If the active LSM wants to access the inode during
5745 * d_instantiate it needs these. Smack checks to see
5746 * if the filesystem supports xattrs by looking at the
5750 inode->i_op = &btrfs_special_inode_operations;
5751 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5755 init_special_inode(inode, inode->i_mode, rdev);
5756 btrfs_update_inode(trans, root, inode);
5757 d_instantiate(dentry, inode);
5760 btrfs_end_transaction(trans, root);
5761 btrfs_btree_balance_dirty(root);
5763 inode_dec_link_count(inode);
5769 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5770 umode_t mode, bool excl)
5772 struct btrfs_trans_handle *trans;
5773 struct btrfs_root *root = BTRFS_I(dir)->root;
5774 struct inode *inode = NULL;
5775 int drop_inode_on_err = 0;
5781 * 2 for inode item and ref
5783 * 1 for xattr if selinux is on
5785 trans = btrfs_start_transaction(root, 5);
5787 return PTR_ERR(trans);
5789 err = btrfs_find_free_ino(root, &objectid);
5793 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5794 dentry->d_name.len, btrfs_ino(dir), objectid,
5796 if (IS_ERR(inode)) {
5797 err = PTR_ERR(inode);
5800 drop_inode_on_err = 1;
5802 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5806 err = btrfs_update_inode(trans, root, inode);
5811 * If the active LSM wants to access the inode during
5812 * d_instantiate it needs these. Smack checks to see
5813 * if the filesystem supports xattrs by looking at the
5816 inode->i_fop = &btrfs_file_operations;
5817 inode->i_op = &btrfs_file_inode_operations;
5819 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5823 inode->i_mapping->a_ops = &btrfs_aops;
5824 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5825 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5826 d_instantiate(dentry, inode);
5829 btrfs_end_transaction(trans, root);
5830 if (err && drop_inode_on_err) {
5831 inode_dec_link_count(inode);
5834 btrfs_btree_balance_dirty(root);
5838 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5839 struct dentry *dentry)
5841 struct btrfs_trans_handle *trans;
5842 struct btrfs_root *root = BTRFS_I(dir)->root;
5843 struct inode *inode = old_dentry->d_inode;
5848 /* do not allow sys_link's with other subvols of the same device */
5849 if (root->objectid != BTRFS_I(inode)->root->objectid)
5852 if (inode->i_nlink >= BTRFS_LINK_MAX)
5855 err = btrfs_set_inode_index(dir, &index);
5860 * 2 items for inode and inode ref
5861 * 2 items for dir items
5862 * 1 item for parent inode
5864 trans = btrfs_start_transaction(root, 5);
5865 if (IS_ERR(trans)) {
5866 err = PTR_ERR(trans);
5870 btrfs_inc_nlink(inode);
5871 inode_inc_iversion(inode);
5872 inode->i_ctime = CURRENT_TIME;
5874 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5876 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5881 struct dentry *parent = dentry->d_parent;
5882 err = btrfs_update_inode(trans, root, inode);
5885 d_instantiate(dentry, inode);
5886 btrfs_log_new_name(trans, inode, NULL, parent);
5889 btrfs_end_transaction(trans, root);
5892 inode_dec_link_count(inode);
5895 btrfs_btree_balance_dirty(root);
5899 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5901 struct inode *inode = NULL;
5902 struct btrfs_trans_handle *trans;
5903 struct btrfs_root *root = BTRFS_I(dir)->root;
5905 int drop_on_err = 0;
5910 * 2 items for inode and ref
5911 * 2 items for dir items
5912 * 1 for xattr if selinux is on
5914 trans = btrfs_start_transaction(root, 5);
5916 return PTR_ERR(trans);
5918 err = btrfs_find_free_ino(root, &objectid);
5922 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5923 dentry->d_name.len, btrfs_ino(dir), objectid,
5924 S_IFDIR | mode, &index);
5925 if (IS_ERR(inode)) {
5926 err = PTR_ERR(inode);
5932 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5936 inode->i_op = &btrfs_dir_inode_operations;
5937 inode->i_fop = &btrfs_dir_file_operations;
5939 btrfs_i_size_write(inode, 0);
5940 err = btrfs_update_inode(trans, root, inode);
5944 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5945 dentry->d_name.len, 0, index);
5949 d_instantiate(dentry, inode);
5953 btrfs_end_transaction(trans, root);
5956 btrfs_btree_balance_dirty(root);
5960 /* helper for btfs_get_extent. Given an existing extent in the tree,
5961 * and an extent that you want to insert, deal with overlap and insert
5962 * the new extent into the tree.
5964 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5965 struct extent_map *existing,
5966 struct extent_map *em,
5967 u64 map_start, u64 map_len)
5971 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5972 start_diff = map_start - em->start;
5973 em->start = map_start;
5975 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5976 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5977 em->block_start += start_diff;
5978 em->block_len -= start_diff;
5980 return add_extent_mapping(em_tree, em);
5983 static noinline int uncompress_inline(struct btrfs_path *path,
5984 struct inode *inode, struct page *page,
5985 size_t pg_offset, u64 extent_offset,
5986 struct btrfs_file_extent_item *item)
5989 struct extent_buffer *leaf = path->nodes[0];
5992 unsigned long inline_size;
5996 WARN_ON(pg_offset != 0);
5997 compress_type = btrfs_file_extent_compression(leaf, item);
5998 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5999 inline_size = btrfs_file_extent_inline_item_len(leaf,
6000 btrfs_item_nr(leaf, path->slots[0]));
6001 tmp = kmalloc(inline_size, GFP_NOFS);
6004 ptr = btrfs_file_extent_inline_start(item);
6006 read_extent_buffer(leaf, tmp, ptr, inline_size);
6008 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6009 ret = btrfs_decompress(compress_type, tmp, page,
6010 extent_offset, inline_size, max_size);
6012 char *kaddr = kmap_atomic(page);
6013 unsigned long copy_size = min_t(u64,
6014 PAGE_CACHE_SIZE - pg_offset,
6015 max_size - extent_offset);
6016 memset(kaddr + pg_offset, 0, copy_size);
6017 kunmap_atomic(kaddr);
6024 * a bit scary, this does extent mapping from logical file offset to the disk.
6025 * the ugly parts come from merging extents from the disk with the in-ram
6026 * representation. This gets more complex because of the data=ordered code,
6027 * where the in-ram extents might be locked pending data=ordered completion.
6029 * This also copies inline extents directly into the page.
6032 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6033 size_t pg_offset, u64 start, u64 len,
6039 u64 extent_start = 0;
6041 u64 objectid = btrfs_ino(inode);
6043 struct btrfs_path *path = NULL;
6044 struct btrfs_root *root = BTRFS_I(inode)->root;
6045 struct btrfs_file_extent_item *item;
6046 struct extent_buffer *leaf;
6047 struct btrfs_key found_key;
6048 struct extent_map *em = NULL;
6049 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6050 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6051 struct btrfs_trans_handle *trans = NULL;
6055 read_lock(&em_tree->lock);
6056 em = lookup_extent_mapping(em_tree, start, len);
6058 em->bdev = root->fs_info->fs_devices->latest_bdev;
6059 read_unlock(&em_tree->lock);
6062 if (em->start > start || em->start + em->len <= start)
6063 free_extent_map(em);
6064 else if (em->block_start == EXTENT_MAP_INLINE && page)
6065 free_extent_map(em);
6069 em = alloc_extent_map();
6074 em->bdev = root->fs_info->fs_devices->latest_bdev;
6075 em->start = EXTENT_MAP_HOLE;
6076 em->orig_start = EXTENT_MAP_HOLE;
6078 em->block_len = (u64)-1;
6081 path = btrfs_alloc_path();
6087 * Chances are we'll be called again, so go ahead and do
6093 ret = btrfs_lookup_file_extent(trans, root, path,
6094 objectid, start, trans != NULL);
6101 if (path->slots[0] == 0)
6106 leaf = path->nodes[0];
6107 item = btrfs_item_ptr(leaf, path->slots[0],
6108 struct btrfs_file_extent_item);
6109 /* are we inside the extent that was found? */
6110 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6111 found_type = btrfs_key_type(&found_key);
6112 if (found_key.objectid != objectid ||
6113 found_type != BTRFS_EXTENT_DATA_KEY) {
6117 found_type = btrfs_file_extent_type(leaf, item);
6118 extent_start = found_key.offset;
6119 compress_type = btrfs_file_extent_compression(leaf, item);
6120 if (found_type == BTRFS_FILE_EXTENT_REG ||
6121 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6122 extent_end = extent_start +
6123 btrfs_file_extent_num_bytes(leaf, item);
6124 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6126 size = btrfs_file_extent_inline_len(leaf, item);
6127 extent_end = ALIGN(extent_start + size, root->sectorsize);
6130 if (start >= extent_end) {
6132 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6133 ret = btrfs_next_leaf(root, path);
6140 leaf = path->nodes[0];
6142 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6143 if (found_key.objectid != objectid ||
6144 found_key.type != BTRFS_EXTENT_DATA_KEY)
6146 if (start + len <= found_key.offset)
6149 em->orig_start = start;
6150 em->len = found_key.offset - start;
6154 if (found_type == BTRFS_FILE_EXTENT_REG ||
6155 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6156 em->start = extent_start;
6157 em->len = extent_end - extent_start;
6158 em->orig_start = extent_start -
6159 btrfs_file_extent_offset(leaf, item);
6160 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6162 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6164 em->block_start = EXTENT_MAP_HOLE;
6167 if (compress_type != BTRFS_COMPRESS_NONE) {
6168 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6169 em->compress_type = compress_type;
6170 em->block_start = bytenr;
6171 em->block_len = em->orig_block_len;
6173 bytenr += btrfs_file_extent_offset(leaf, item);
6174 em->block_start = bytenr;
6175 em->block_len = em->len;
6176 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6177 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6180 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6184 size_t extent_offset;
6187 em->block_start = EXTENT_MAP_INLINE;
6188 if (!page || create) {
6189 em->start = extent_start;
6190 em->len = extent_end - extent_start;
6194 size = btrfs_file_extent_inline_len(leaf, item);
6195 extent_offset = page_offset(page) + pg_offset - extent_start;
6196 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6197 size - extent_offset);
6198 em->start = extent_start + extent_offset;
6199 em->len = ALIGN(copy_size, root->sectorsize);
6200 em->orig_block_len = em->len;
6201 em->orig_start = em->start;
6202 if (compress_type) {
6203 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6204 em->compress_type = compress_type;
6206 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6207 if (create == 0 && !PageUptodate(page)) {
6208 if (btrfs_file_extent_compression(leaf, item) !=
6209 BTRFS_COMPRESS_NONE) {
6210 ret = uncompress_inline(path, inode, page,
6212 extent_offset, item);
6213 BUG_ON(ret); /* -ENOMEM */
6216 read_extent_buffer(leaf, map + pg_offset, ptr,
6218 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6219 memset(map + pg_offset + copy_size, 0,
6220 PAGE_CACHE_SIZE - pg_offset -
6225 flush_dcache_page(page);
6226 } else if (create && PageUptodate(page)) {
6230 free_extent_map(em);
6233 btrfs_release_path(path);
6234 trans = btrfs_join_transaction(root);
6237 return ERR_CAST(trans);
6241 write_extent_buffer(leaf, map + pg_offset, ptr,
6244 btrfs_mark_buffer_dirty(leaf);
6246 set_extent_uptodate(io_tree, em->start,
6247 extent_map_end(em) - 1, NULL, GFP_NOFS);
6250 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6254 em->orig_start = start;
6257 em->block_start = EXTENT_MAP_HOLE;
6258 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6260 btrfs_release_path(path);
6261 if (em->start > start || extent_map_end(em) <= start) {
6262 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
6263 "[%llu %llu]\n", (unsigned long long)em->start,
6264 (unsigned long long)em->len,
6265 (unsigned long long)start,
6266 (unsigned long long)len);
6272 write_lock(&em_tree->lock);
6273 ret = add_extent_mapping(em_tree, em);
6274 /* it is possible that someone inserted the extent into the tree
6275 * while we had the lock dropped. It is also possible that
6276 * an overlapping map exists in the tree
6278 if (ret == -EEXIST) {
6279 struct extent_map *existing;
6283 existing = lookup_extent_mapping(em_tree, start, len);
6284 if (existing && (existing->start > start ||
6285 existing->start + existing->len <= start)) {
6286 free_extent_map(existing);
6290 existing = lookup_extent_mapping(em_tree, em->start,
6293 err = merge_extent_mapping(em_tree, existing,
6296 free_extent_map(existing);
6298 free_extent_map(em);
6303 free_extent_map(em);
6307 free_extent_map(em);
6312 write_unlock(&em_tree->lock);
6316 trace_btrfs_get_extent(root, em);
6319 btrfs_free_path(path);
6321 ret = btrfs_end_transaction(trans, root);
6326 free_extent_map(em);
6327 return ERR_PTR(err);
6329 BUG_ON(!em); /* Error is always set */
6333 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6334 size_t pg_offset, u64 start, u64 len,
6337 struct extent_map *em;
6338 struct extent_map *hole_em = NULL;
6339 u64 range_start = start;
6345 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6352 * - a pre-alloc extent,
6353 * there might actually be delalloc bytes behind it.
6355 if (em->block_start != EXTENT_MAP_HOLE &&
6356 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6362 /* check to see if we've wrapped (len == -1 or similar) */
6371 /* ok, we didn't find anything, lets look for delalloc */
6372 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6373 end, len, EXTENT_DELALLOC, 1);
6374 found_end = range_start + found;
6375 if (found_end < range_start)
6376 found_end = (u64)-1;
6379 * we didn't find anything useful, return
6380 * the original results from get_extent()
6382 if (range_start > end || found_end <= start) {
6388 /* adjust the range_start to make sure it doesn't
6389 * go backwards from the start they passed in
6391 range_start = max(start,range_start);
6392 found = found_end - range_start;
6395 u64 hole_start = start;
6398 em = alloc_extent_map();
6404 * when btrfs_get_extent can't find anything it
6405 * returns one huge hole
6407 * make sure what it found really fits our range, and
6408 * adjust to make sure it is based on the start from
6412 u64 calc_end = extent_map_end(hole_em);
6414 if (calc_end <= start || (hole_em->start > end)) {
6415 free_extent_map(hole_em);
6418 hole_start = max(hole_em->start, start);
6419 hole_len = calc_end - hole_start;
6423 if (hole_em && range_start > hole_start) {
6424 /* our hole starts before our delalloc, so we
6425 * have to return just the parts of the hole
6426 * that go until the delalloc starts
6428 em->len = min(hole_len,
6429 range_start - hole_start);
6430 em->start = hole_start;
6431 em->orig_start = hole_start;
6433 * don't adjust block start at all,
6434 * it is fixed at EXTENT_MAP_HOLE
6436 em->block_start = hole_em->block_start;
6437 em->block_len = hole_len;
6438 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6439 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6441 em->start = range_start;
6443 em->orig_start = range_start;
6444 em->block_start = EXTENT_MAP_DELALLOC;
6445 em->block_len = found;
6447 } else if (hole_em) {
6452 free_extent_map(hole_em);
6454 free_extent_map(em);
6455 return ERR_PTR(err);
6460 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6463 struct btrfs_root *root = BTRFS_I(inode)->root;
6464 struct btrfs_trans_handle *trans;
6465 struct extent_map *em;
6466 struct btrfs_key ins;
6470 trans = btrfs_join_transaction(root);
6472 return ERR_CAST(trans);
6474 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
6476 alloc_hint = get_extent_allocation_hint(inode, start, len);
6477 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
6478 alloc_hint, &ins, 1);
6484 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6485 ins.offset, ins.offset, 0);
6489 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6490 ins.offset, ins.offset, 0);
6492 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6496 btrfs_end_transaction(trans, root);
6501 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6502 * block must be cow'd
6504 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
6505 struct inode *inode, u64 offset, u64 len)
6507 struct btrfs_path *path;
6509 struct extent_buffer *leaf;
6510 struct btrfs_root *root = BTRFS_I(inode)->root;
6511 struct btrfs_file_extent_item *fi;
6512 struct btrfs_key key;
6520 path = btrfs_alloc_path();
6524 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
6529 slot = path->slots[0];
6532 /* can't find the item, must cow */
6539 leaf = path->nodes[0];
6540 btrfs_item_key_to_cpu(leaf, &key, slot);
6541 if (key.objectid != btrfs_ino(inode) ||
6542 key.type != BTRFS_EXTENT_DATA_KEY) {
6543 /* not our file or wrong item type, must cow */
6547 if (key.offset > offset) {
6548 /* Wrong offset, must cow */
6552 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6553 found_type = btrfs_file_extent_type(leaf, fi);
6554 if (found_type != BTRFS_FILE_EXTENT_REG &&
6555 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6556 /* not a regular extent, must cow */
6559 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6560 backref_offset = btrfs_file_extent_offset(leaf, fi);
6562 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6563 if (extent_end < offset + len) {
6564 /* extent doesn't include our full range, must cow */
6568 if (btrfs_extent_readonly(root, disk_bytenr))
6572 * look for other files referencing this extent, if we
6573 * find any we must cow
6575 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6576 key.offset - backref_offset, disk_bytenr))
6580 * adjust disk_bytenr and num_bytes to cover just the bytes
6581 * in this extent we are about to write. If there
6582 * are any csums in that range we have to cow in order
6583 * to keep the csums correct
6585 disk_bytenr += backref_offset;
6586 disk_bytenr += offset - key.offset;
6587 num_bytes = min(offset + len, extent_end) - offset;
6588 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6591 * all of the above have passed, it is safe to overwrite this extent
6596 btrfs_free_path(path);
6600 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6601 struct extent_state **cached_state, int writing)
6603 struct btrfs_ordered_extent *ordered;
6607 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6610 * We're concerned with the entire range that we're going to be
6611 * doing DIO to, so we need to make sure theres no ordered
6612 * extents in this range.
6614 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6615 lockend - lockstart + 1);
6618 * We need to make sure there are no buffered pages in this
6619 * range either, we could have raced between the invalidate in
6620 * generic_file_direct_write and locking the extent. The
6621 * invalidate needs to happen so that reads after a write do not
6624 if (!ordered && (!writing ||
6625 !test_range_bit(&BTRFS_I(inode)->io_tree,
6626 lockstart, lockend, EXTENT_UPTODATE, 0,
6630 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6631 cached_state, GFP_NOFS);
6634 btrfs_start_ordered_extent(inode, ordered, 1);
6635 btrfs_put_ordered_extent(ordered);
6637 /* Screw you mmap */
6638 ret = filemap_write_and_wait_range(inode->i_mapping,
6645 * If we found a page that couldn't be invalidated just
6646 * fall back to buffered.
6648 ret = invalidate_inode_pages2_range(inode->i_mapping,
6649 lockstart >> PAGE_CACHE_SHIFT,
6650 lockend >> PAGE_CACHE_SHIFT);
6661 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6662 u64 len, u64 orig_start,
6663 u64 block_start, u64 block_len,
6664 u64 orig_block_len, int type)
6666 struct extent_map_tree *em_tree;
6667 struct extent_map *em;
6668 struct btrfs_root *root = BTRFS_I(inode)->root;
6671 em_tree = &BTRFS_I(inode)->extent_tree;
6672 em = alloc_extent_map();
6674 return ERR_PTR(-ENOMEM);
6677 em->orig_start = orig_start;
6678 em->mod_start = start;
6681 em->block_len = block_len;
6682 em->block_start = block_start;
6683 em->bdev = root->fs_info->fs_devices->latest_bdev;
6684 em->orig_block_len = orig_block_len;
6685 em->generation = -1;
6686 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6687 if (type == BTRFS_ORDERED_PREALLOC)
6688 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6691 btrfs_drop_extent_cache(inode, em->start,
6692 em->start + em->len - 1, 0);
6693 write_lock(&em_tree->lock);
6694 ret = add_extent_mapping(em_tree, em);
6696 list_move(&em->list,
6697 &em_tree->modified_extents);
6698 write_unlock(&em_tree->lock);
6699 } while (ret == -EEXIST);
6702 free_extent_map(em);
6703 return ERR_PTR(ret);
6710 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6711 struct buffer_head *bh_result, int create)
6713 struct extent_map *em;
6714 struct btrfs_root *root = BTRFS_I(inode)->root;
6715 struct extent_state *cached_state = NULL;
6716 u64 start = iblock << inode->i_blkbits;
6717 u64 lockstart, lockend;
6718 u64 len = bh_result->b_size;
6719 struct btrfs_trans_handle *trans;
6720 int unlock_bits = EXTENT_LOCKED;
6724 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6726 len = min_t(u64, len, root->sectorsize);
6729 lockend = start + len - 1;
6732 * If this errors out it's because we couldn't invalidate pagecache for
6733 * this range and we need to fallback to buffered.
6735 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6738 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6745 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6746 * io. INLINE is special, and we could probably kludge it in here, but
6747 * it's still buffered so for safety lets just fall back to the generic
6750 * For COMPRESSED we _have_ to read the entire extent in so we can
6751 * decompress it, so there will be buffering required no matter what we
6752 * do, so go ahead and fallback to buffered.
6754 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6755 * to buffered IO. Don't blame me, this is the price we pay for using
6758 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6759 em->block_start == EXTENT_MAP_INLINE) {
6760 free_extent_map(em);
6765 /* Just a good old fashioned hole, return */
6766 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6767 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6768 free_extent_map(em);
6773 * We don't allocate a new extent in the following cases
6775 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6777 * 2) The extent is marked as PREALLOC. We're good to go here and can
6778 * just use the extent.
6782 len = min(len, em->len - (start - em->start));
6783 lockstart = start + len;
6787 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6788 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6789 em->block_start != EXTENT_MAP_HOLE)) {
6794 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6795 type = BTRFS_ORDERED_PREALLOC;
6797 type = BTRFS_ORDERED_NOCOW;
6798 len = min(len, em->len - (start - em->start));
6799 block_start = em->block_start + (start - em->start);
6802 * we're not going to log anything, but we do need
6803 * to make sure the current transaction stays open
6804 * while we look for nocow cross refs
6806 trans = btrfs_join_transaction(root);
6810 if (can_nocow_odirect(trans, inode, start, len) == 1) {
6811 u64 orig_start = em->orig_start;
6812 u64 orig_block_len = em->orig_block_len;
6814 if (type == BTRFS_ORDERED_PREALLOC) {
6815 free_extent_map(em);
6816 em = create_pinned_em(inode, start, len,
6819 orig_block_len, type);
6821 btrfs_end_transaction(trans, root);
6826 ret = btrfs_add_ordered_extent_dio(inode, start,
6827 block_start, len, len, type);
6828 btrfs_end_transaction(trans, root);
6830 free_extent_map(em);
6835 btrfs_end_transaction(trans, root);
6839 * this will cow the extent, reset the len in case we changed
6842 len = bh_result->b_size;
6843 free_extent_map(em);
6844 em = btrfs_new_extent_direct(inode, start, len);
6849 len = min(len, em->len - (start - em->start));
6851 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6853 bh_result->b_size = len;
6854 bh_result->b_bdev = em->bdev;
6855 set_buffer_mapped(bh_result);
6857 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6858 set_buffer_new(bh_result);
6861 * Need to update the i_size under the extent lock so buffered
6862 * readers will get the updated i_size when we unlock.
6864 if (start + len > i_size_read(inode))
6865 i_size_write(inode, start + len);
6867 spin_lock(&BTRFS_I(inode)->lock);
6868 BTRFS_I(inode)->outstanding_extents++;
6869 spin_unlock(&BTRFS_I(inode)->lock);
6871 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6872 lockstart + len - 1, EXTENT_DELALLOC, NULL,
6873 &cached_state, GFP_NOFS);
6878 * In the case of write we need to clear and unlock the entire range,
6879 * in the case of read we need to unlock only the end area that we
6880 * aren't using if there is any left over space.
6882 if (lockstart < lockend) {
6883 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6884 lockend, unlock_bits, 1, 0,
6885 &cached_state, GFP_NOFS);
6887 free_extent_state(cached_state);
6890 free_extent_map(em);
6895 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6896 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6900 struct btrfs_dio_private {
6901 struct inode *inode;
6907 /* number of bios pending for this dio */
6908 atomic_t pending_bios;
6913 struct bio *orig_bio;
6916 static void btrfs_endio_direct_read(struct bio *bio, int err)
6918 struct btrfs_dio_private *dip = bio->bi_private;
6919 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6920 struct bio_vec *bvec = bio->bi_io_vec;
6921 struct inode *inode = dip->inode;
6922 struct btrfs_root *root = BTRFS_I(inode)->root;
6925 start = dip->logical_offset;
6927 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6928 struct page *page = bvec->bv_page;
6931 u64 private = ~(u32)0;
6932 unsigned long flags;
6934 if (get_state_private(&BTRFS_I(inode)->io_tree,
6937 local_irq_save(flags);
6938 kaddr = kmap_atomic(page);
6939 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
6940 csum, bvec->bv_len);
6941 btrfs_csum_final(csum, (char *)&csum);
6942 kunmap_atomic(kaddr);
6943 local_irq_restore(flags);
6945 flush_dcache_page(bvec->bv_page);
6946 if (csum != private) {
6948 printk(KERN_ERR "btrfs csum failed ino %llu off"
6949 " %llu csum %u private %u\n",
6950 (unsigned long long)btrfs_ino(inode),
6951 (unsigned long long)start,
6952 csum, (unsigned)private);
6957 start += bvec->bv_len;
6959 } while (bvec <= bvec_end);
6961 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6962 dip->logical_offset + dip->bytes - 1);
6963 bio->bi_private = dip->private;
6967 /* If we had a csum failure make sure to clear the uptodate flag */
6969 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6970 dio_end_io(bio, err);
6973 static void btrfs_endio_direct_write(struct bio *bio, int err)
6975 struct btrfs_dio_private *dip = bio->bi_private;
6976 struct inode *inode = dip->inode;
6977 struct btrfs_root *root = BTRFS_I(inode)->root;
6978 struct btrfs_ordered_extent *ordered = NULL;
6979 u64 ordered_offset = dip->logical_offset;
6980 u64 ordered_bytes = dip->bytes;
6986 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6988 ordered_bytes, !err);
6992 ordered->work.func = finish_ordered_fn;
6993 ordered->work.flags = 0;
6994 btrfs_queue_worker(&root->fs_info->endio_write_workers,
6998 * our bio might span multiple ordered extents. If we haven't
6999 * completed the accounting for the whole dio, go back and try again
7001 if (ordered_offset < dip->logical_offset + dip->bytes) {
7002 ordered_bytes = dip->logical_offset + dip->bytes -
7008 bio->bi_private = dip->private;
7012 /* If we had an error make sure to clear the uptodate flag */
7014 clear_bit(BIO_UPTODATE, &bio->bi_flags);
7015 dio_end_io(bio, err);
7018 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7019 struct bio *bio, int mirror_num,
7020 unsigned long bio_flags, u64 offset)
7023 struct btrfs_root *root = BTRFS_I(inode)->root;
7024 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7025 BUG_ON(ret); /* -ENOMEM */
7029 static void btrfs_end_dio_bio(struct bio *bio, int err)
7031 struct btrfs_dio_private *dip = bio->bi_private;
7034 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
7035 "sector %#Lx len %u err no %d\n",
7036 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
7037 (unsigned long long)bio->bi_sector, bio->bi_size, err);
7041 * before atomic variable goto zero, we must make sure
7042 * dip->errors is perceived to be set.
7044 smp_mb__before_atomic_dec();
7047 /* if there are more bios still pending for this dio, just exit */
7048 if (!atomic_dec_and_test(&dip->pending_bios))
7052 bio_io_error(dip->orig_bio);
7054 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
7055 bio_endio(dip->orig_bio, 0);
7061 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7062 u64 first_sector, gfp_t gfp_flags)
7064 int nr_vecs = bio_get_nr_vecs(bdev);
7065 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7068 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7069 int rw, u64 file_offset, int skip_sum,
7072 int write = rw & REQ_WRITE;
7073 struct btrfs_root *root = BTRFS_I(inode)->root;
7077 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7082 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7090 if (write && async_submit) {
7091 ret = btrfs_wq_submit_bio(root->fs_info,
7092 inode, rw, bio, 0, 0,
7094 __btrfs_submit_bio_start_direct_io,
7095 __btrfs_submit_bio_done);
7099 * If we aren't doing async submit, calculate the csum of the
7102 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7105 } else if (!skip_sum) {
7106 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
7112 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7118 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7121 struct inode *inode = dip->inode;
7122 struct btrfs_root *root = BTRFS_I(inode)->root;
7124 struct bio *orig_bio = dip->orig_bio;
7125 struct bio_vec *bvec = orig_bio->bi_io_vec;
7126 u64 start_sector = orig_bio->bi_sector;
7127 u64 file_offset = dip->logical_offset;
7132 int async_submit = 0;
7134 map_length = orig_bio->bi_size;
7135 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7136 &map_length, NULL, 0);
7141 if (map_length >= orig_bio->bi_size) {
7146 /* async crcs make it difficult to collect full stripe writes. */
7147 if (btrfs_get_alloc_profile(root, 1) &
7148 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7153 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7156 bio->bi_private = dip;
7157 bio->bi_end_io = btrfs_end_dio_bio;
7158 atomic_inc(&dip->pending_bios);
7160 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7161 if (unlikely(map_length < submit_len + bvec->bv_len ||
7162 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7163 bvec->bv_offset) < bvec->bv_len)) {
7165 * inc the count before we submit the bio so
7166 * we know the end IO handler won't happen before
7167 * we inc the count. Otherwise, the dip might get freed
7168 * before we're done setting it up
7170 atomic_inc(&dip->pending_bios);
7171 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7172 file_offset, skip_sum,
7176 atomic_dec(&dip->pending_bios);
7180 start_sector += submit_len >> 9;
7181 file_offset += submit_len;
7186 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7187 start_sector, GFP_NOFS);
7190 bio->bi_private = dip;
7191 bio->bi_end_io = btrfs_end_dio_bio;
7193 map_length = orig_bio->bi_size;
7194 ret = btrfs_map_block(root->fs_info, rw,
7196 &map_length, NULL, 0);
7202 submit_len += bvec->bv_len;
7209 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7218 * before atomic variable goto zero, we must
7219 * make sure dip->errors is perceived to be set.
7221 smp_mb__before_atomic_dec();
7222 if (atomic_dec_and_test(&dip->pending_bios))
7223 bio_io_error(dip->orig_bio);
7225 /* bio_end_io() will handle error, so we needn't return it */
7229 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
7232 struct btrfs_root *root = BTRFS_I(inode)->root;
7233 struct btrfs_dio_private *dip;
7234 struct bio_vec *bvec = bio->bi_io_vec;
7236 int write = rw & REQ_WRITE;
7239 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7241 dip = kmalloc(sizeof(*dip), GFP_NOFS);
7247 dip->private = bio->bi_private;
7249 dip->logical_offset = file_offset;
7253 dip->bytes += bvec->bv_len;
7255 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
7257 dip->disk_bytenr = (u64)bio->bi_sector << 9;
7258 bio->bi_private = dip;
7260 dip->orig_bio = bio;
7261 atomic_set(&dip->pending_bios, 0);
7264 bio->bi_end_io = btrfs_endio_direct_write;
7266 bio->bi_end_io = btrfs_endio_direct_read;
7268 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7273 * If this is a write, we need to clean up the reserved space and kill
7274 * the ordered extent.
7277 struct btrfs_ordered_extent *ordered;
7278 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7279 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7280 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7281 btrfs_free_reserved_extent(root, ordered->start,
7283 btrfs_put_ordered_extent(ordered);
7284 btrfs_put_ordered_extent(ordered);
7286 bio_endio(bio, ret);
7289 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7290 const struct iovec *iov, loff_t offset,
7291 unsigned long nr_segs)
7297 unsigned blocksize_mask = root->sectorsize - 1;
7298 ssize_t retval = -EINVAL;
7299 loff_t end = offset;
7301 if (offset & blocksize_mask)
7304 /* Check the memory alignment. Blocks cannot straddle pages */
7305 for (seg = 0; seg < nr_segs; seg++) {
7306 addr = (unsigned long)iov[seg].iov_base;
7307 size = iov[seg].iov_len;
7309 if ((addr & blocksize_mask) || (size & blocksize_mask))
7312 /* If this is a write we don't need to check anymore */
7317 * Check to make sure we don't have duplicate iov_base's in this
7318 * iovec, if so return EINVAL, otherwise we'll get csum errors
7319 * when reading back.
7321 for (i = seg + 1; i < nr_segs; i++) {
7322 if (iov[seg].iov_base == iov[i].iov_base)
7331 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7332 const struct iovec *iov, loff_t offset,
7333 unsigned long nr_segs)
7335 struct file *file = iocb->ki_filp;
7336 struct inode *inode = file->f_mapping->host;
7340 bool relock = false;
7343 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7347 atomic_inc(&inode->i_dio_count);
7348 smp_mb__after_atomic_inc();
7351 count = iov_length(iov, nr_segs);
7353 * If the write DIO is beyond the EOF, we need update
7354 * the isize, but it is protected by i_mutex. So we can
7355 * not unlock the i_mutex at this case.
7357 if (offset + count <= inode->i_size) {
7358 mutex_unlock(&inode->i_mutex);
7361 ret = btrfs_delalloc_reserve_space(inode, count);
7364 } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7365 &BTRFS_I(inode)->runtime_flags))) {
7366 inode_dio_done(inode);
7367 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7371 ret = __blockdev_direct_IO(rw, iocb, inode,
7372 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7373 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7374 btrfs_submit_direct, flags);
7376 if (ret < 0 && ret != -EIOCBQUEUED)
7377 btrfs_delalloc_release_space(inode, count);
7378 else if (ret >= 0 && (size_t)ret < count)
7379 btrfs_delalloc_release_space(inode,
7380 count - (size_t)ret);
7382 btrfs_delalloc_release_metadata(inode, 0);
7386 inode_dio_done(inode);
7388 mutex_lock(&inode->i_mutex);
7393 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
7395 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7396 __u64 start, __u64 len)
7400 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7404 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7407 int btrfs_readpage(struct file *file, struct page *page)
7409 struct extent_io_tree *tree;
7410 tree = &BTRFS_I(page->mapping->host)->io_tree;
7411 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7414 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7416 struct extent_io_tree *tree;
7419 if (current->flags & PF_MEMALLOC) {
7420 redirty_page_for_writepage(wbc, page);
7424 tree = &BTRFS_I(page->mapping->host)->io_tree;
7425 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7428 int btrfs_writepages(struct address_space *mapping,
7429 struct writeback_control *wbc)
7431 struct extent_io_tree *tree;
7433 tree = &BTRFS_I(mapping->host)->io_tree;
7434 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7438 btrfs_readpages(struct file *file, struct address_space *mapping,
7439 struct list_head *pages, unsigned nr_pages)
7441 struct extent_io_tree *tree;
7442 tree = &BTRFS_I(mapping->host)->io_tree;
7443 return extent_readpages(tree, mapping, pages, nr_pages,
7446 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7448 struct extent_io_tree *tree;
7449 struct extent_map_tree *map;
7452 tree = &BTRFS_I(page->mapping->host)->io_tree;
7453 map = &BTRFS_I(page->mapping->host)->extent_tree;
7454 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7456 ClearPagePrivate(page);
7457 set_page_private(page, 0);
7458 page_cache_release(page);
7463 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7465 if (PageWriteback(page) || PageDirty(page))
7467 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7470 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
7472 struct inode *inode = page->mapping->host;
7473 struct extent_io_tree *tree;
7474 struct btrfs_ordered_extent *ordered;
7475 struct extent_state *cached_state = NULL;
7476 u64 page_start = page_offset(page);
7477 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7480 * we have the page locked, so new writeback can't start,
7481 * and the dirty bit won't be cleared while we are here.
7483 * Wait for IO on this page so that we can safely clear
7484 * the PagePrivate2 bit and do ordered accounting
7486 wait_on_page_writeback(page);
7488 tree = &BTRFS_I(inode)->io_tree;
7490 btrfs_releasepage(page, GFP_NOFS);
7493 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7494 ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
7497 * IO on this page will never be started, so we need
7498 * to account for any ordered extents now
7500 clear_extent_bit(tree, page_start, page_end,
7501 EXTENT_DIRTY | EXTENT_DELALLOC |
7502 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7503 EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
7505 * whoever cleared the private bit is responsible
7506 * for the finish_ordered_io
7508 if (TestClearPagePrivate2(page) &&
7509 btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
7510 PAGE_CACHE_SIZE, 1)) {
7511 btrfs_finish_ordered_io(ordered);
7513 btrfs_put_ordered_extent(ordered);
7514 cached_state = NULL;
7515 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7517 clear_extent_bit(tree, page_start, page_end,
7518 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
7519 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
7520 &cached_state, GFP_NOFS);
7521 __btrfs_releasepage(page, GFP_NOFS);
7523 ClearPageChecked(page);
7524 if (PagePrivate(page)) {
7525 ClearPagePrivate(page);
7526 set_page_private(page, 0);
7527 page_cache_release(page);
7532 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7533 * called from a page fault handler when a page is first dirtied. Hence we must
7534 * be careful to check for EOF conditions here. We set the page up correctly
7535 * for a written page which means we get ENOSPC checking when writing into
7536 * holes and correct delalloc and unwritten extent mapping on filesystems that
7537 * support these features.
7539 * We are not allowed to take the i_mutex here so we have to play games to
7540 * protect against truncate races as the page could now be beyond EOF. Because
7541 * vmtruncate() writes the inode size before removing pages, once we have the
7542 * page lock we can determine safely if the page is beyond EOF. If it is not
7543 * beyond EOF, then the page is guaranteed safe against truncation until we
7546 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7548 struct page *page = vmf->page;
7549 struct inode *inode = file_inode(vma->vm_file);
7550 struct btrfs_root *root = BTRFS_I(inode)->root;
7551 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7552 struct btrfs_ordered_extent *ordered;
7553 struct extent_state *cached_state = NULL;
7555 unsigned long zero_start;
7562 sb_start_pagefault(inode->i_sb);
7563 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7565 ret = file_update_time(vma->vm_file);
7571 else /* -ENOSPC, -EIO, etc */
7572 ret = VM_FAULT_SIGBUS;
7578 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7581 size = i_size_read(inode);
7582 page_start = page_offset(page);
7583 page_end = page_start + PAGE_CACHE_SIZE - 1;
7585 if ((page->mapping != inode->i_mapping) ||
7586 (page_start >= size)) {
7587 /* page got truncated out from underneath us */
7590 wait_on_page_writeback(page);
7592 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7593 set_page_extent_mapped(page);
7596 * we can't set the delalloc bits if there are pending ordered
7597 * extents. Drop our locks and wait for them to finish
7599 ordered = btrfs_lookup_ordered_extent(inode, page_start);
7601 unlock_extent_cached(io_tree, page_start, page_end,
7602 &cached_state, GFP_NOFS);
7604 btrfs_start_ordered_extent(inode, ordered, 1);
7605 btrfs_put_ordered_extent(ordered);
7610 * XXX - page_mkwrite gets called every time the page is dirtied, even
7611 * if it was already dirty, so for space accounting reasons we need to
7612 * clear any delalloc bits for the range we are fixing to save. There
7613 * is probably a better way to do this, but for now keep consistent with
7614 * prepare_pages in the normal write path.
7616 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7617 EXTENT_DIRTY | EXTENT_DELALLOC |
7618 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7619 0, 0, &cached_state, GFP_NOFS);
7621 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7624 unlock_extent_cached(io_tree, page_start, page_end,
7625 &cached_state, GFP_NOFS);
7626 ret = VM_FAULT_SIGBUS;
7631 /* page is wholly or partially inside EOF */
7632 if (page_start + PAGE_CACHE_SIZE > size)
7633 zero_start = size & ~PAGE_CACHE_MASK;
7635 zero_start = PAGE_CACHE_SIZE;
7637 if (zero_start != PAGE_CACHE_SIZE) {
7639 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7640 flush_dcache_page(page);
7643 ClearPageChecked(page);
7644 set_page_dirty(page);
7645 SetPageUptodate(page);
7647 BTRFS_I(inode)->last_trans = root->fs_info->generation;
7648 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7649 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7651 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7655 sb_end_pagefault(inode->i_sb);
7656 return VM_FAULT_LOCKED;
7660 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7662 sb_end_pagefault(inode->i_sb);
7666 static int btrfs_truncate(struct inode *inode)
7668 struct btrfs_root *root = BTRFS_I(inode)->root;
7669 struct btrfs_block_rsv *rsv;
7672 struct btrfs_trans_handle *trans;
7673 u64 mask = root->sectorsize - 1;
7674 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7676 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
7680 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
7681 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
7684 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
7685 * 3 things going on here
7687 * 1) We need to reserve space for our orphan item and the space to
7688 * delete our orphan item. Lord knows we don't want to have a dangling
7689 * orphan item because we didn't reserve space to remove it.
7691 * 2) We need to reserve space to update our inode.
7693 * 3) We need to have something to cache all the space that is going to
7694 * be free'd up by the truncate operation, but also have some slack
7695 * space reserved in case it uses space during the truncate (thank you
7696 * very much snapshotting).
7698 * And we need these to all be seperate. The fact is we can use alot of
7699 * space doing the truncate, and we have no earthly idea how much space
7700 * we will use, so we need the truncate reservation to be seperate so it
7701 * doesn't end up using space reserved for updating the inode or
7702 * removing the orphan item. We also need to be able to stop the
7703 * transaction and start a new one, which means we need to be able to
7704 * update the inode several times, and we have no idea of knowing how
7705 * many times that will be, so we can't just reserve 1 item for the
7706 * entirety of the opration, so that has to be done seperately as well.
7707 * Then there is the orphan item, which does indeed need to be held on
7708 * to for the whole operation, and we need nobody to touch this reserved
7709 * space except the orphan code.
7711 * So that leaves us with
7713 * 1) root->orphan_block_rsv - for the orphan deletion.
7714 * 2) rsv - for the truncate reservation, which we will steal from the
7715 * transaction reservation.
7716 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7717 * updating the inode.
7719 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7722 rsv->size = min_size;
7726 * 1 for the truncate slack space
7727 * 1 for updating the inode.
7729 trans = btrfs_start_transaction(root, 2);
7730 if (IS_ERR(trans)) {
7731 err = PTR_ERR(trans);
7735 /* Migrate the slack space for the truncate to our reserve */
7736 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7741 * setattr is responsible for setting the ordered_data_close flag,
7742 * but that is only tested during the last file release. That
7743 * could happen well after the next commit, leaving a great big
7744 * window where new writes may get lost if someone chooses to write
7745 * to this file after truncating to zero
7747 * The inode doesn't have any dirty data here, and so if we commit
7748 * this is a noop. If someone immediately starts writing to the inode
7749 * it is very likely we'll catch some of their writes in this
7750 * transaction, and the commit will find this file on the ordered
7751 * data list with good things to send down.
7753 * This is a best effort solution, there is still a window where
7754 * using truncate to replace the contents of the file will
7755 * end up with a zero length file after a crash.
7757 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7758 &BTRFS_I(inode)->runtime_flags))
7759 btrfs_add_ordered_operation(trans, root, inode);
7762 * So if we truncate and then write and fsync we normally would just
7763 * write the extents that changed, which is a problem if we need to
7764 * first truncate that entire inode. So set this flag so we write out
7765 * all of the extents in the inode to the sync log so we're completely
7768 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7769 trans->block_rsv = rsv;
7772 ret = btrfs_truncate_inode_items(trans, root, inode,
7774 BTRFS_EXTENT_DATA_KEY);
7775 if (ret != -ENOSPC) {
7780 trans->block_rsv = &root->fs_info->trans_block_rsv;
7781 ret = btrfs_update_inode(trans, root, inode);
7787 btrfs_end_transaction(trans, root);
7788 btrfs_btree_balance_dirty(root);
7790 trans = btrfs_start_transaction(root, 2);
7791 if (IS_ERR(trans)) {
7792 ret = err = PTR_ERR(trans);
7797 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7799 BUG_ON(ret); /* shouldn't happen */
7800 trans->block_rsv = rsv;
7803 if (ret == 0 && inode->i_nlink > 0) {
7804 trans->block_rsv = root->orphan_block_rsv;
7805 ret = btrfs_orphan_del(trans, inode);
7811 trans->block_rsv = &root->fs_info->trans_block_rsv;
7812 ret = btrfs_update_inode(trans, root, inode);
7816 ret = btrfs_end_transaction(trans, root);
7817 btrfs_btree_balance_dirty(root);
7821 btrfs_free_block_rsv(root, rsv);
7830 * create a new subvolume directory/inode (helper for the ioctl).
7832 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7833 struct btrfs_root *new_root, u64 new_dirid)
7835 struct inode *inode;
7839 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7840 new_dirid, new_dirid,
7841 S_IFDIR | (~current_umask() & S_IRWXUGO),
7844 return PTR_ERR(inode);
7845 inode->i_op = &btrfs_dir_inode_operations;
7846 inode->i_fop = &btrfs_dir_file_operations;
7848 set_nlink(inode, 1);
7849 btrfs_i_size_write(inode, 0);
7851 err = btrfs_update_inode(trans, new_root, inode);
7857 struct inode *btrfs_alloc_inode(struct super_block *sb)
7859 struct btrfs_inode *ei;
7860 struct inode *inode;
7862 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7869 ei->last_sub_trans = 0;
7870 ei->logged_trans = 0;
7871 ei->delalloc_bytes = 0;
7872 ei->disk_i_size = 0;
7875 ei->index_cnt = (u64)-1;
7876 ei->last_unlink_trans = 0;
7877 ei->last_log_commit = 0;
7879 spin_lock_init(&ei->lock);
7880 ei->outstanding_extents = 0;
7881 ei->reserved_extents = 0;
7883 ei->runtime_flags = 0;
7884 ei->force_compress = BTRFS_COMPRESS_NONE;
7886 ei->delayed_node = NULL;
7888 inode = &ei->vfs_inode;
7889 extent_map_tree_init(&ei->extent_tree);
7890 extent_io_tree_init(&ei->io_tree, &inode->i_data);
7891 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7892 ei->io_tree.track_uptodate = 1;
7893 ei->io_failure_tree.track_uptodate = 1;
7894 atomic_set(&ei->sync_writers, 0);
7895 mutex_init(&ei->log_mutex);
7896 mutex_init(&ei->delalloc_mutex);
7897 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7898 INIT_LIST_HEAD(&ei->delalloc_inodes);
7899 INIT_LIST_HEAD(&ei->ordered_operations);
7900 RB_CLEAR_NODE(&ei->rb_node);
7905 static void btrfs_i_callback(struct rcu_head *head)
7907 struct inode *inode = container_of(head, struct inode, i_rcu);
7908 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7911 void btrfs_destroy_inode(struct inode *inode)
7913 struct btrfs_ordered_extent *ordered;
7914 struct btrfs_root *root = BTRFS_I(inode)->root;
7916 WARN_ON(!hlist_empty(&inode->i_dentry));
7917 WARN_ON(inode->i_data.nrpages);
7918 WARN_ON(BTRFS_I(inode)->outstanding_extents);
7919 WARN_ON(BTRFS_I(inode)->reserved_extents);
7920 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7921 WARN_ON(BTRFS_I(inode)->csum_bytes);
7924 * This can happen where we create an inode, but somebody else also
7925 * created the same inode and we need to destroy the one we already
7932 * Make sure we're properly removed from the ordered operation
7936 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7937 spin_lock(&root->fs_info->ordered_extent_lock);
7938 list_del_init(&BTRFS_I(inode)->ordered_operations);
7939 spin_unlock(&root->fs_info->ordered_extent_lock);
7942 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7943 &BTRFS_I(inode)->runtime_flags)) {
7944 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7945 (unsigned long long)btrfs_ino(inode));
7946 atomic_dec(&root->orphan_inodes);
7950 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7954 printk(KERN_ERR "btrfs found ordered "
7955 "extent %llu %llu on inode cleanup\n",
7956 (unsigned long long)ordered->file_offset,
7957 (unsigned long long)ordered->len);
7958 btrfs_remove_ordered_extent(inode, ordered);
7959 btrfs_put_ordered_extent(ordered);
7960 btrfs_put_ordered_extent(ordered);
7963 inode_tree_del(inode);
7964 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7966 btrfs_remove_delayed_node(inode);
7967 call_rcu(&inode->i_rcu, btrfs_i_callback);
7970 int btrfs_drop_inode(struct inode *inode)
7972 struct btrfs_root *root = BTRFS_I(inode)->root;
7974 /* the snap/subvol tree is on deleting */
7975 if (btrfs_root_refs(&root->root_item) == 0 &&
7976 root != root->fs_info->tree_root)
7979 return generic_drop_inode(inode);
7982 static void init_once(void *foo)
7984 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7986 inode_init_once(&ei->vfs_inode);
7989 void btrfs_destroy_cachep(void)
7992 * Make sure all delayed rcu free inodes are flushed before we
7996 if (btrfs_inode_cachep)
7997 kmem_cache_destroy(btrfs_inode_cachep);
7998 if (btrfs_trans_handle_cachep)
7999 kmem_cache_destroy(btrfs_trans_handle_cachep);
8000 if (btrfs_transaction_cachep)
8001 kmem_cache_destroy(btrfs_transaction_cachep);
8002 if (btrfs_path_cachep)
8003 kmem_cache_destroy(btrfs_path_cachep);
8004 if (btrfs_free_space_cachep)
8005 kmem_cache_destroy(btrfs_free_space_cachep);
8006 if (btrfs_delalloc_work_cachep)
8007 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8010 int btrfs_init_cachep(void)
8012 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8013 sizeof(struct btrfs_inode), 0,
8014 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8015 if (!btrfs_inode_cachep)
8018 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8019 sizeof(struct btrfs_trans_handle), 0,
8020 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8021 if (!btrfs_trans_handle_cachep)
8024 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8025 sizeof(struct btrfs_transaction), 0,
8026 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8027 if (!btrfs_transaction_cachep)
8030 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8031 sizeof(struct btrfs_path), 0,
8032 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8033 if (!btrfs_path_cachep)
8036 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8037 sizeof(struct btrfs_free_space), 0,
8038 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8039 if (!btrfs_free_space_cachep)
8042 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8043 sizeof(struct btrfs_delalloc_work), 0,
8044 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8046 if (!btrfs_delalloc_work_cachep)
8051 btrfs_destroy_cachep();
8055 static int btrfs_getattr(struct vfsmount *mnt,
8056 struct dentry *dentry, struct kstat *stat)
8059 struct inode *inode = dentry->d_inode;
8060 u32 blocksize = inode->i_sb->s_blocksize;
8062 generic_fillattr(inode, stat);
8063 stat->dev = BTRFS_I(inode)->root->anon_dev;
8064 stat->blksize = PAGE_CACHE_SIZE;
8066 spin_lock(&BTRFS_I(inode)->lock);
8067 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8068 spin_unlock(&BTRFS_I(inode)->lock);
8069 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8070 ALIGN(delalloc_bytes, blocksize)) >> 9;
8074 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8075 struct inode *new_dir, struct dentry *new_dentry)
8077 struct btrfs_trans_handle *trans;
8078 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8079 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8080 struct inode *new_inode = new_dentry->d_inode;
8081 struct inode *old_inode = old_dentry->d_inode;
8082 struct timespec ctime = CURRENT_TIME;
8086 u64 old_ino = btrfs_ino(old_inode);
8088 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8091 /* we only allow rename subvolume link between subvolumes */
8092 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8095 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8096 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8099 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8100 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8104 /* check for collisions, even if the name isn't there */
8105 ret = btrfs_check_dir_item_collision(root, new_dir->i_ino,
8106 new_dentry->d_name.name,
8107 new_dentry->d_name.len);
8110 if (ret == -EEXIST) {
8112 * eexist without a new_inode */
8118 /* maybe -EOVERFLOW */
8125 * we're using rename to replace one file with another.
8126 * and the replacement file is large. Start IO on it now so
8127 * we don't add too much work to the end of the transaction
8129 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8130 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8131 filemap_flush(old_inode->i_mapping);
8133 /* close the racy window with snapshot create/destroy ioctl */
8134 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8135 down_read(&root->fs_info->subvol_sem);
8137 * We want to reserve the absolute worst case amount of items. So if
8138 * both inodes are subvols and we need to unlink them then that would
8139 * require 4 item modifications, but if they are both normal inodes it
8140 * would require 5 item modifications, so we'll assume their normal
8141 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8142 * should cover the worst case number of items we'll modify.
8144 trans = btrfs_start_transaction(root, 11);
8145 if (IS_ERR(trans)) {
8146 ret = PTR_ERR(trans);
8151 btrfs_record_root_in_trans(trans, dest);
8153 ret = btrfs_set_inode_index(new_dir, &index);
8157 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8158 /* force full log commit if subvolume involved. */
8159 root->fs_info->last_trans_log_full_commit = trans->transid;
8161 ret = btrfs_insert_inode_ref(trans, dest,
8162 new_dentry->d_name.name,
8163 new_dentry->d_name.len,
8165 btrfs_ino(new_dir), index);
8169 * this is an ugly little race, but the rename is required
8170 * to make sure that if we crash, the inode is either at the
8171 * old name or the new one. pinning the log transaction lets
8172 * us make sure we don't allow a log commit to come in after
8173 * we unlink the name but before we add the new name back in.
8175 btrfs_pin_log_trans(root);
8178 * make sure the inode gets flushed if it is replacing
8181 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8182 btrfs_add_ordered_operation(trans, root, old_inode);
8184 inode_inc_iversion(old_dir);
8185 inode_inc_iversion(new_dir);
8186 inode_inc_iversion(old_inode);
8187 old_dir->i_ctime = old_dir->i_mtime = ctime;
8188 new_dir->i_ctime = new_dir->i_mtime = ctime;
8189 old_inode->i_ctime = ctime;
8191 if (old_dentry->d_parent != new_dentry->d_parent)
8192 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8194 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8195 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8196 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8197 old_dentry->d_name.name,
8198 old_dentry->d_name.len);
8200 ret = __btrfs_unlink_inode(trans, root, old_dir,
8201 old_dentry->d_inode,
8202 old_dentry->d_name.name,
8203 old_dentry->d_name.len);
8205 ret = btrfs_update_inode(trans, root, old_inode);
8208 btrfs_abort_transaction(trans, root, ret);
8213 inode_inc_iversion(new_inode);
8214 new_inode->i_ctime = CURRENT_TIME;
8215 if (unlikely(btrfs_ino(new_inode) ==
8216 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8217 root_objectid = BTRFS_I(new_inode)->location.objectid;
8218 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8220 new_dentry->d_name.name,
8221 new_dentry->d_name.len);
8222 BUG_ON(new_inode->i_nlink == 0);
8224 ret = btrfs_unlink_inode(trans, dest, new_dir,
8225 new_dentry->d_inode,
8226 new_dentry->d_name.name,
8227 new_dentry->d_name.len);
8229 if (!ret && new_inode->i_nlink == 0) {
8230 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8234 btrfs_abort_transaction(trans, root, ret);
8239 ret = btrfs_add_link(trans, new_dir, old_inode,
8240 new_dentry->d_name.name,
8241 new_dentry->d_name.len, 0, index);
8243 btrfs_abort_transaction(trans, root, ret);
8247 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8248 struct dentry *parent = new_dentry->d_parent;
8249 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8250 btrfs_end_log_trans(root);
8253 btrfs_end_transaction(trans, root);
8255 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8256 up_read(&root->fs_info->subvol_sem);
8261 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8263 struct btrfs_delalloc_work *delalloc_work;
8265 delalloc_work = container_of(work, struct btrfs_delalloc_work,
8267 if (delalloc_work->wait)
8268 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
8270 filemap_flush(delalloc_work->inode->i_mapping);
8272 if (delalloc_work->delay_iput)
8273 btrfs_add_delayed_iput(delalloc_work->inode);
8275 iput(delalloc_work->inode);
8276 complete(&delalloc_work->completion);
8279 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8280 int wait, int delay_iput)
8282 struct btrfs_delalloc_work *work;
8284 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8288 init_completion(&work->completion);
8289 INIT_LIST_HEAD(&work->list);
8290 work->inode = inode;
8292 work->delay_iput = delay_iput;
8293 work->work.func = btrfs_run_delalloc_work;
8298 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8300 wait_for_completion(&work->completion);
8301 kmem_cache_free(btrfs_delalloc_work_cachep, work);
8305 * some fairly slow code that needs optimization. This walks the list
8306 * of all the inodes with pending delalloc and forces them to disk.
8308 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8310 struct btrfs_inode *binode;
8311 struct inode *inode;
8312 struct btrfs_delalloc_work *work, *next;
8313 struct list_head works;
8314 struct list_head splice;
8317 if (root->fs_info->sb->s_flags & MS_RDONLY)
8320 INIT_LIST_HEAD(&works);
8321 INIT_LIST_HEAD(&splice);
8323 spin_lock(&root->fs_info->delalloc_lock);
8324 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
8325 while (!list_empty(&splice)) {
8326 binode = list_entry(splice.next, struct btrfs_inode,
8329 list_del_init(&binode->delalloc_inodes);
8331 inode = igrab(&binode->vfs_inode);
8333 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
8334 &binode->runtime_flags);
8338 list_add_tail(&binode->delalloc_inodes,
8339 &root->fs_info->delalloc_inodes);
8340 spin_unlock(&root->fs_info->delalloc_lock);
8342 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8343 if (unlikely(!work)) {
8347 list_add_tail(&work->list, &works);
8348 btrfs_queue_worker(&root->fs_info->flush_workers,
8352 spin_lock(&root->fs_info->delalloc_lock);
8354 spin_unlock(&root->fs_info->delalloc_lock);
8356 list_for_each_entry_safe(work, next, &works, list) {
8357 list_del_init(&work->list);
8358 btrfs_wait_and_free_delalloc_work(work);
8361 /* the filemap_flush will queue IO into the worker threads, but
8362 * we have to make sure the IO is actually started and that
8363 * ordered extents get created before we return
8365 atomic_inc(&root->fs_info->async_submit_draining);
8366 while (atomic_read(&root->fs_info->nr_async_submits) ||
8367 atomic_read(&root->fs_info->async_delalloc_pages)) {
8368 wait_event(root->fs_info->async_submit_wait,
8369 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8370 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8372 atomic_dec(&root->fs_info->async_submit_draining);
8375 list_for_each_entry_safe(work, next, &works, list) {
8376 list_del_init(&work->list);
8377 btrfs_wait_and_free_delalloc_work(work);
8380 if (!list_empty_careful(&splice)) {
8381 spin_lock(&root->fs_info->delalloc_lock);
8382 list_splice_tail(&splice, &root->fs_info->delalloc_inodes);
8383 spin_unlock(&root->fs_info->delalloc_lock);
8388 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8389 const char *symname)
8391 struct btrfs_trans_handle *trans;
8392 struct btrfs_root *root = BTRFS_I(dir)->root;
8393 struct btrfs_path *path;
8394 struct btrfs_key key;
8395 struct inode *inode = NULL;
8403 struct btrfs_file_extent_item *ei;
8404 struct extent_buffer *leaf;
8406 name_len = strlen(symname) + 1;
8407 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8408 return -ENAMETOOLONG;
8411 * 2 items for inode item and ref
8412 * 2 items for dir items
8413 * 1 item for xattr if selinux is on
8415 trans = btrfs_start_transaction(root, 5);
8417 return PTR_ERR(trans);
8419 err = btrfs_find_free_ino(root, &objectid);
8423 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8424 dentry->d_name.len, btrfs_ino(dir), objectid,
8425 S_IFLNK|S_IRWXUGO, &index);
8426 if (IS_ERR(inode)) {
8427 err = PTR_ERR(inode);
8431 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8438 * If the active LSM wants to access the inode during
8439 * d_instantiate it needs these. Smack checks to see
8440 * if the filesystem supports xattrs by looking at the
8443 inode->i_fop = &btrfs_file_operations;
8444 inode->i_op = &btrfs_file_inode_operations;
8446 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8450 inode->i_mapping->a_ops = &btrfs_aops;
8451 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8452 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8457 path = btrfs_alloc_path();
8463 key.objectid = btrfs_ino(inode);
8465 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8466 datasize = btrfs_file_extent_calc_inline_size(name_len);
8467 err = btrfs_insert_empty_item(trans, root, path, &key,
8471 btrfs_free_path(path);
8474 leaf = path->nodes[0];
8475 ei = btrfs_item_ptr(leaf, path->slots[0],
8476 struct btrfs_file_extent_item);
8477 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8478 btrfs_set_file_extent_type(leaf, ei,
8479 BTRFS_FILE_EXTENT_INLINE);
8480 btrfs_set_file_extent_encryption(leaf, ei, 0);
8481 btrfs_set_file_extent_compression(leaf, ei, 0);
8482 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8483 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8485 ptr = btrfs_file_extent_inline_start(ei);
8486 write_extent_buffer(leaf, symname, ptr, name_len);
8487 btrfs_mark_buffer_dirty(leaf);
8488 btrfs_free_path(path);
8490 inode->i_op = &btrfs_symlink_inode_operations;
8491 inode->i_mapping->a_ops = &btrfs_symlink_aops;
8492 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8493 inode_set_bytes(inode, name_len);
8494 btrfs_i_size_write(inode, name_len - 1);
8495 err = btrfs_update_inode(trans, root, inode);
8501 d_instantiate(dentry, inode);
8502 btrfs_end_transaction(trans, root);
8504 inode_dec_link_count(inode);
8507 btrfs_btree_balance_dirty(root);
8511 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8512 u64 start, u64 num_bytes, u64 min_size,
8513 loff_t actual_len, u64 *alloc_hint,
8514 struct btrfs_trans_handle *trans)
8516 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8517 struct extent_map *em;
8518 struct btrfs_root *root = BTRFS_I(inode)->root;
8519 struct btrfs_key ins;
8520 u64 cur_offset = start;
8524 bool own_trans = true;
8528 while (num_bytes > 0) {
8530 trans = btrfs_start_transaction(root, 3);
8531 if (IS_ERR(trans)) {
8532 ret = PTR_ERR(trans);
8537 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8538 cur_bytes = max(cur_bytes, min_size);
8539 ret = btrfs_reserve_extent(trans, root, cur_bytes,
8540 min_size, 0, *alloc_hint, &ins, 1);
8543 btrfs_end_transaction(trans, root);
8547 ret = insert_reserved_file_extent(trans, inode,
8548 cur_offset, ins.objectid,
8549 ins.offset, ins.offset,
8550 ins.offset, 0, 0, 0,
8551 BTRFS_FILE_EXTENT_PREALLOC);
8553 btrfs_abort_transaction(trans, root, ret);
8555 btrfs_end_transaction(trans, root);
8558 btrfs_drop_extent_cache(inode, cur_offset,
8559 cur_offset + ins.offset -1, 0);
8561 em = alloc_extent_map();
8563 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8564 &BTRFS_I(inode)->runtime_flags);
8568 em->start = cur_offset;
8569 em->orig_start = cur_offset;
8570 em->len = ins.offset;
8571 em->block_start = ins.objectid;
8572 em->block_len = ins.offset;
8573 em->orig_block_len = ins.offset;
8574 em->bdev = root->fs_info->fs_devices->latest_bdev;
8575 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8576 em->generation = trans->transid;
8579 write_lock(&em_tree->lock);
8580 ret = add_extent_mapping(em_tree, em);
8582 list_move(&em->list,
8583 &em_tree->modified_extents);
8584 write_unlock(&em_tree->lock);
8587 btrfs_drop_extent_cache(inode, cur_offset,
8588 cur_offset + ins.offset - 1,
8591 free_extent_map(em);
8593 num_bytes -= ins.offset;
8594 cur_offset += ins.offset;
8595 *alloc_hint = ins.objectid + ins.offset;
8597 inode_inc_iversion(inode);
8598 inode->i_ctime = CURRENT_TIME;
8599 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8600 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8601 (actual_len > inode->i_size) &&
8602 (cur_offset > inode->i_size)) {
8603 if (cur_offset > actual_len)
8604 i_size = actual_len;
8606 i_size = cur_offset;
8607 i_size_write(inode, i_size);
8608 btrfs_ordered_update_i_size(inode, i_size, NULL);
8611 ret = btrfs_update_inode(trans, root, inode);
8614 btrfs_abort_transaction(trans, root, ret);
8616 btrfs_end_transaction(trans, root);
8621 btrfs_end_transaction(trans, root);
8626 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8627 u64 start, u64 num_bytes, u64 min_size,
8628 loff_t actual_len, u64 *alloc_hint)
8630 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8631 min_size, actual_len, alloc_hint,
8635 int btrfs_prealloc_file_range_trans(struct inode *inode,
8636 struct btrfs_trans_handle *trans, int mode,
8637 u64 start, u64 num_bytes, u64 min_size,
8638 loff_t actual_len, u64 *alloc_hint)
8640 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8641 min_size, actual_len, alloc_hint, trans);
8644 static int btrfs_set_page_dirty(struct page *page)
8646 return __set_page_dirty_nobuffers(page);
8649 static int btrfs_permission(struct inode *inode, int mask)
8651 struct btrfs_root *root = BTRFS_I(inode)->root;
8652 umode_t mode = inode->i_mode;
8654 if (mask & MAY_WRITE &&
8655 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8656 if (btrfs_root_readonly(root))
8658 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8661 return generic_permission(inode, mask);
8664 static const struct inode_operations btrfs_dir_inode_operations = {
8665 .getattr = btrfs_getattr,
8666 .lookup = btrfs_lookup,
8667 .create = btrfs_create,
8668 .unlink = btrfs_unlink,
8670 .mkdir = btrfs_mkdir,
8671 .rmdir = btrfs_rmdir,
8672 .rename = btrfs_rename,
8673 .symlink = btrfs_symlink,
8674 .setattr = btrfs_setattr,
8675 .mknod = btrfs_mknod,
8676 .setxattr = btrfs_setxattr,
8677 .getxattr = btrfs_getxattr,
8678 .listxattr = btrfs_listxattr,
8679 .removexattr = btrfs_removexattr,
8680 .permission = btrfs_permission,
8681 .get_acl = btrfs_get_acl,
8683 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8684 .lookup = btrfs_lookup,
8685 .permission = btrfs_permission,
8686 .get_acl = btrfs_get_acl,
8689 static const struct file_operations btrfs_dir_file_operations = {
8690 .llseek = generic_file_llseek,
8691 .read = generic_read_dir,
8692 .readdir = btrfs_real_readdir,
8693 .unlocked_ioctl = btrfs_ioctl,
8694 #ifdef CONFIG_COMPAT
8695 .compat_ioctl = btrfs_ioctl,
8697 .release = btrfs_release_file,
8698 .fsync = btrfs_sync_file,
8701 static struct extent_io_ops btrfs_extent_io_ops = {
8702 .fill_delalloc = run_delalloc_range,
8703 .submit_bio_hook = btrfs_submit_bio_hook,
8704 .merge_bio_hook = btrfs_merge_bio_hook,
8705 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
8706 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
8707 .writepage_start_hook = btrfs_writepage_start_hook,
8708 .set_bit_hook = btrfs_set_bit_hook,
8709 .clear_bit_hook = btrfs_clear_bit_hook,
8710 .merge_extent_hook = btrfs_merge_extent_hook,
8711 .split_extent_hook = btrfs_split_extent_hook,
8715 * btrfs doesn't support the bmap operation because swapfiles
8716 * use bmap to make a mapping of extents in the file. They assume
8717 * these extents won't change over the life of the file and they
8718 * use the bmap result to do IO directly to the drive.
8720 * the btrfs bmap call would return logical addresses that aren't
8721 * suitable for IO and they also will change frequently as COW
8722 * operations happen. So, swapfile + btrfs == corruption.
8724 * For now we're avoiding this by dropping bmap.
8726 static const struct address_space_operations btrfs_aops = {
8727 .readpage = btrfs_readpage,
8728 .writepage = btrfs_writepage,
8729 .writepages = btrfs_writepages,
8730 .readpages = btrfs_readpages,
8731 .direct_IO = btrfs_direct_IO,
8732 .invalidatepage = btrfs_invalidatepage,
8733 .releasepage = btrfs_releasepage,
8734 .set_page_dirty = btrfs_set_page_dirty,
8735 .error_remove_page = generic_error_remove_page,
8738 static const struct address_space_operations btrfs_symlink_aops = {
8739 .readpage = btrfs_readpage,
8740 .writepage = btrfs_writepage,
8741 .invalidatepage = btrfs_invalidatepage,
8742 .releasepage = btrfs_releasepage,
8745 static const struct inode_operations btrfs_file_inode_operations = {
8746 .getattr = btrfs_getattr,
8747 .setattr = btrfs_setattr,
8748 .setxattr = btrfs_setxattr,
8749 .getxattr = btrfs_getxattr,
8750 .listxattr = btrfs_listxattr,
8751 .removexattr = btrfs_removexattr,
8752 .permission = btrfs_permission,
8753 .fiemap = btrfs_fiemap,
8754 .get_acl = btrfs_get_acl,
8755 .update_time = btrfs_update_time,
8757 static const struct inode_operations btrfs_special_inode_operations = {
8758 .getattr = btrfs_getattr,
8759 .setattr = btrfs_setattr,
8760 .permission = btrfs_permission,
8761 .setxattr = btrfs_setxattr,
8762 .getxattr = btrfs_getxattr,
8763 .listxattr = btrfs_listxattr,
8764 .removexattr = btrfs_removexattr,
8765 .get_acl = btrfs_get_acl,
8766 .update_time = btrfs_update_time,
8768 static const struct inode_operations btrfs_symlink_inode_operations = {
8769 .readlink = generic_readlink,
8770 .follow_link = page_follow_link_light,
8771 .put_link = page_put_link,
8772 .getattr = btrfs_getattr,
8773 .setattr = btrfs_setattr,
8774 .permission = btrfs_permission,
8775 .setxattr = btrfs_setxattr,
8776 .getxattr = btrfs_getxattr,
8777 .listxattr = btrfs_listxattr,
8778 .removexattr = btrfs_removexattr,
8779 .get_acl = btrfs_get_acl,
8780 .update_time = btrfs_update_time,
8783 const struct dentry_operations btrfs_dentry_operations = {
8784 .d_delete = btrfs_dentry_delete,
8785 .d_release = btrfs_dentry_release,