2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/aio.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/xattr.h>
38 #include <linux/posix_acl.h>
39 #include <linux/falloc.h>
40 #include <linux/slab.h>
41 #include <linux/ratelimit.h>
42 #include <linux/mount.h>
43 #include <linux/btrfs.h>
44 #include <linux/blkdev.h>
45 #include <linux/posix_acl_xattr.h>
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "print-tree.h"
51 #include "ordered-data.h"
55 #include "compression.h"
57 #include "free-space-cache.h"
58 #include "inode-map.h"
63 struct btrfs_iget_args {
64 struct btrfs_key *location;
65 struct btrfs_root *root;
68 static const struct inode_operations btrfs_dir_inode_operations;
69 static const struct inode_operations btrfs_symlink_inode_operations;
70 static const struct inode_operations btrfs_dir_ro_inode_operations;
71 static const struct inode_operations btrfs_special_inode_operations;
72 static const struct inode_operations btrfs_file_inode_operations;
73 static const struct address_space_operations btrfs_aops;
74 static const struct address_space_operations btrfs_symlink_aops;
75 static const struct file_operations btrfs_dir_file_operations;
76 static struct extent_io_ops btrfs_extent_io_ops;
78 static struct kmem_cache *btrfs_inode_cachep;
79 static struct kmem_cache *btrfs_delalloc_work_cachep;
80 struct kmem_cache *btrfs_trans_handle_cachep;
81 struct kmem_cache *btrfs_transaction_cachep;
82 struct kmem_cache *btrfs_path_cachep;
83 struct kmem_cache *btrfs_free_space_cachep;
86 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
87 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
88 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
89 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
90 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
91 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
92 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
93 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
96 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
97 static int btrfs_truncate(struct inode *inode);
98 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
99 static noinline int cow_file_range(struct inode *inode,
100 struct page *locked_page,
101 u64 start, u64 end, int *page_started,
102 unsigned long *nr_written, int unlock);
103 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
104 u64 len, u64 orig_start,
105 u64 block_start, u64 block_len,
106 u64 orig_block_len, u64 ram_bytes,
109 static int btrfs_dirty_inode(struct inode *inode);
111 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
112 struct inode *inode, struct inode *dir,
113 const struct qstr *qstr)
117 err = btrfs_init_acl(trans, inode, dir);
119 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
124 * this does all the hard work for inserting an inline extent into
125 * the btree. The caller should have done a btrfs_drop_extents so that
126 * no overlapping inline items exist in the btree
128 static int insert_inline_extent(struct btrfs_trans_handle *trans,
129 struct btrfs_path *path, int extent_inserted,
130 struct btrfs_root *root, struct inode *inode,
131 u64 start, size_t size, size_t compressed_size,
133 struct page **compressed_pages)
135 struct extent_buffer *leaf;
136 struct page *page = NULL;
139 struct btrfs_file_extent_item *ei;
142 size_t cur_size = size;
143 unsigned long offset;
145 if (compressed_size && compressed_pages)
146 cur_size = compressed_size;
148 inode_add_bytes(inode, size);
150 if (!extent_inserted) {
151 struct btrfs_key key;
154 key.objectid = btrfs_ino(inode);
156 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
158 datasize = btrfs_file_extent_calc_inline_size(cur_size);
159 path->leave_spinning = 1;
160 ret = btrfs_insert_empty_item(trans, root, path, &key,
167 leaf = path->nodes[0];
168 ei = btrfs_item_ptr(leaf, path->slots[0],
169 struct btrfs_file_extent_item);
170 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
171 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
172 btrfs_set_file_extent_encryption(leaf, ei, 0);
173 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
174 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
175 ptr = btrfs_file_extent_inline_start(ei);
177 if (compress_type != BTRFS_COMPRESS_NONE) {
180 while (compressed_size > 0) {
181 cpage = compressed_pages[i];
182 cur_size = min_t(unsigned long, compressed_size,
185 kaddr = kmap_atomic(cpage);
186 write_extent_buffer(leaf, kaddr, ptr, cur_size);
187 kunmap_atomic(kaddr);
191 compressed_size -= cur_size;
193 btrfs_set_file_extent_compression(leaf, ei,
196 page = find_get_page(inode->i_mapping,
197 start >> PAGE_CACHE_SHIFT);
198 btrfs_set_file_extent_compression(leaf, ei, 0);
199 kaddr = kmap_atomic(page);
200 offset = start & (PAGE_CACHE_SIZE - 1);
201 write_extent_buffer(leaf, kaddr + offset, ptr, size);
202 kunmap_atomic(kaddr);
203 page_cache_release(page);
205 btrfs_mark_buffer_dirty(leaf);
206 btrfs_release_path(path);
209 * we're an inline extent, so nobody can
210 * extend the file past i_size without locking
211 * a page we already have locked.
213 * We must do any isize and inode updates
214 * before we unlock the pages. Otherwise we
215 * could end up racing with unlink.
217 BTRFS_I(inode)->disk_i_size = inode->i_size;
218 ret = btrfs_update_inode(trans, root, inode);
227 * conditionally insert an inline extent into the file. This
228 * does the checks required to make sure the data is small enough
229 * to fit as an inline extent.
231 static noinline int cow_file_range_inline(struct btrfs_root *root,
232 struct inode *inode, u64 start,
233 u64 end, size_t compressed_size,
235 struct page **compressed_pages)
237 struct btrfs_trans_handle *trans;
238 u64 isize = i_size_read(inode);
239 u64 actual_end = min(end + 1, isize);
240 u64 inline_len = actual_end - start;
241 u64 aligned_end = ALIGN(end, root->sectorsize);
242 u64 data_len = inline_len;
244 struct btrfs_path *path;
245 int extent_inserted = 0;
246 u32 extent_item_size;
249 data_len = compressed_size;
252 actual_end >= PAGE_CACHE_SIZE ||
253 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
255 (actual_end & (root->sectorsize - 1)) == 0) ||
257 data_len > root->fs_info->max_inline) {
261 path = btrfs_alloc_path();
265 trans = btrfs_join_transaction(root);
267 btrfs_free_path(path);
268 return PTR_ERR(trans);
270 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
272 if (compressed_size && compressed_pages)
273 extent_item_size = btrfs_file_extent_calc_inline_size(
276 extent_item_size = btrfs_file_extent_calc_inline_size(
279 ret = __btrfs_drop_extents(trans, root, inode, path,
280 start, aligned_end, NULL,
281 1, 1, extent_item_size, &extent_inserted);
283 btrfs_abort_transaction(trans, root, ret);
287 if (isize > actual_end)
288 inline_len = min_t(u64, isize, actual_end);
289 ret = insert_inline_extent(trans, path, extent_inserted,
291 inline_len, compressed_size,
292 compress_type, compressed_pages);
293 if (ret && ret != -ENOSPC) {
294 btrfs_abort_transaction(trans, root, ret);
296 } else if (ret == -ENOSPC) {
301 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
302 btrfs_delalloc_release_metadata(inode, end + 1 - start);
303 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
305 btrfs_free_path(path);
306 btrfs_end_transaction(trans, root);
310 struct async_extent {
315 unsigned long nr_pages;
317 struct list_head list;
322 struct btrfs_root *root;
323 struct page *locked_page;
326 struct list_head extents;
327 struct btrfs_work work;
330 static noinline int add_async_extent(struct async_cow *cow,
331 u64 start, u64 ram_size,
334 unsigned long nr_pages,
337 struct async_extent *async_extent;
339 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
340 BUG_ON(!async_extent); /* -ENOMEM */
341 async_extent->start = start;
342 async_extent->ram_size = ram_size;
343 async_extent->compressed_size = compressed_size;
344 async_extent->pages = pages;
345 async_extent->nr_pages = nr_pages;
346 async_extent->compress_type = compress_type;
347 list_add_tail(&async_extent->list, &cow->extents);
352 * we create compressed extents in two phases. The first
353 * phase compresses a range of pages that have already been
354 * locked (both pages and state bits are locked).
356 * This is done inside an ordered work queue, and the compression
357 * is spread across many cpus. The actual IO submission is step
358 * two, and the ordered work queue takes care of making sure that
359 * happens in the same order things were put onto the queue by
360 * writepages and friends.
362 * If this code finds it can't get good compression, it puts an
363 * entry onto the work queue to write the uncompressed bytes. This
364 * makes sure that both compressed inodes and uncompressed inodes
365 * are written in the same order that the flusher thread sent them
368 static noinline int compress_file_range(struct inode *inode,
369 struct page *locked_page,
371 struct async_cow *async_cow,
374 struct btrfs_root *root = BTRFS_I(inode)->root;
376 u64 blocksize = root->sectorsize;
378 u64 isize = i_size_read(inode);
380 struct page **pages = NULL;
381 unsigned long nr_pages;
382 unsigned long nr_pages_ret = 0;
383 unsigned long total_compressed = 0;
384 unsigned long total_in = 0;
385 unsigned long max_compressed = 128 * 1024;
386 unsigned long max_uncompressed = 128 * 1024;
389 int compress_type = root->fs_info->compress_type;
392 /* if this is a small write inside eof, kick off a defrag */
393 if ((end - start + 1) < 16 * 1024 &&
394 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
395 btrfs_add_inode_defrag(NULL, inode);
398 * skip compression for a small file range(<=blocksize) that
399 * isn't an inline extent, since it dosen't save disk space at all.
401 if ((end - start + 1) <= blocksize &&
402 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
403 goto cleanup_and_bail_uncompressed;
405 actual_end = min_t(u64, isize, end + 1);
408 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
409 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
412 * we don't want to send crud past the end of i_size through
413 * compression, that's just a waste of CPU time. So, if the
414 * end of the file is before the start of our current
415 * requested range of bytes, we bail out to the uncompressed
416 * cleanup code that can deal with all of this.
418 * It isn't really the fastest way to fix things, but this is a
419 * very uncommon corner.
421 if (actual_end <= start)
422 goto cleanup_and_bail_uncompressed;
424 total_compressed = actual_end - start;
426 /* we want to make sure that amount of ram required to uncompress
427 * an extent is reasonable, so we limit the total size in ram
428 * of a compressed extent to 128k. This is a crucial number
429 * because it also controls how easily we can spread reads across
430 * cpus for decompression.
432 * We also want to make sure the amount of IO required to do
433 * a random read is reasonably small, so we limit the size of
434 * a compressed extent to 128k.
436 total_compressed = min(total_compressed, max_uncompressed);
437 num_bytes = ALIGN(end - start + 1, blocksize);
438 num_bytes = max(blocksize, num_bytes);
443 * we do compression for mount -o compress and when the
444 * inode has not been flagged as nocompress. This flag can
445 * change at any time if we discover bad compression ratios.
447 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
448 (btrfs_test_opt(root, COMPRESS) ||
449 (BTRFS_I(inode)->force_compress) ||
450 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
452 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
454 /* just bail out to the uncompressed code */
458 if (BTRFS_I(inode)->force_compress)
459 compress_type = BTRFS_I(inode)->force_compress;
462 * we need to call clear_page_dirty_for_io on each
463 * page in the range. Otherwise applications with the file
464 * mmap'd can wander in and change the page contents while
465 * we are compressing them.
467 * If the compression fails for any reason, we set the pages
468 * dirty again later on.
470 extent_range_clear_dirty_for_io(inode, start, end);
472 ret = btrfs_compress_pages(compress_type,
473 inode->i_mapping, start,
474 total_compressed, pages,
475 nr_pages, &nr_pages_ret,
481 unsigned long offset = total_compressed &
482 (PAGE_CACHE_SIZE - 1);
483 struct page *page = pages[nr_pages_ret - 1];
486 /* zero the tail end of the last page, we might be
487 * sending it down to disk
490 kaddr = kmap_atomic(page);
491 memset(kaddr + offset, 0,
492 PAGE_CACHE_SIZE - offset);
493 kunmap_atomic(kaddr);
500 /* lets try to make an inline extent */
501 if (ret || total_in < (actual_end - start)) {
502 /* we didn't compress the entire range, try
503 * to make an uncompressed inline extent.
505 ret = cow_file_range_inline(root, inode, start, end,
508 /* try making a compressed inline extent */
509 ret = cow_file_range_inline(root, inode, start, end,
511 compress_type, pages);
514 unsigned long clear_flags = EXTENT_DELALLOC |
516 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
519 * inline extent creation worked or returned error,
520 * we don't need to create any more async work items.
521 * Unlock and free up our temp pages.
523 extent_clear_unlock_delalloc(inode, start, end, NULL,
524 clear_flags, PAGE_UNLOCK |
534 * we aren't doing an inline extent round the compressed size
535 * up to a block size boundary so the allocator does sane
538 total_compressed = ALIGN(total_compressed, blocksize);
541 * one last check to make sure the compression is really a
542 * win, compare the page count read with the blocks on disk
544 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
545 if (total_compressed >= total_in) {
548 num_bytes = total_in;
551 if (!will_compress && pages) {
553 * the compression code ran but failed to make things smaller,
554 * free any pages it allocated and our page pointer array
556 for (i = 0; i < nr_pages_ret; i++) {
557 WARN_ON(pages[i]->mapping);
558 page_cache_release(pages[i]);
562 total_compressed = 0;
565 /* flag the file so we don't compress in the future */
566 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
567 !(BTRFS_I(inode)->force_compress)) {
568 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
574 /* the async work queues will take care of doing actual
575 * allocation on disk for these compressed pages,
576 * and will submit them to the elevator.
578 add_async_extent(async_cow, start, num_bytes,
579 total_compressed, pages, nr_pages_ret,
582 if (start + num_bytes < end) {
589 cleanup_and_bail_uncompressed:
591 * No compression, but we still need to write the pages in
592 * the file we've been given so far. redirty the locked
593 * page if it corresponds to our extent and set things up
594 * for the async work queue to run cow_file_range to do
595 * the normal delalloc dance
597 if (page_offset(locked_page) >= start &&
598 page_offset(locked_page) <= end) {
599 __set_page_dirty_nobuffers(locked_page);
600 /* unlocked later on in the async handlers */
603 extent_range_redirty_for_io(inode, start, end);
604 add_async_extent(async_cow, start, end - start + 1,
605 0, NULL, 0, BTRFS_COMPRESS_NONE);
613 for (i = 0; i < nr_pages_ret; i++) {
614 WARN_ON(pages[i]->mapping);
615 page_cache_release(pages[i]);
623 * phase two of compressed writeback. This is the ordered portion
624 * of the code, which only gets called in the order the work was
625 * queued. We walk all the async extents created by compress_file_range
626 * and send them down to the disk.
628 static noinline int submit_compressed_extents(struct inode *inode,
629 struct async_cow *async_cow)
631 struct async_extent *async_extent;
633 struct btrfs_key ins;
634 struct extent_map *em;
635 struct btrfs_root *root = BTRFS_I(inode)->root;
636 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
637 struct extent_io_tree *io_tree;
640 if (list_empty(&async_cow->extents))
644 while (!list_empty(&async_cow->extents)) {
645 async_extent = list_entry(async_cow->extents.next,
646 struct async_extent, list);
647 list_del(&async_extent->list);
649 io_tree = &BTRFS_I(inode)->io_tree;
652 /* did the compression code fall back to uncompressed IO? */
653 if (!async_extent->pages) {
654 int page_started = 0;
655 unsigned long nr_written = 0;
657 lock_extent(io_tree, async_extent->start,
658 async_extent->start +
659 async_extent->ram_size - 1);
661 /* allocate blocks */
662 ret = cow_file_range(inode, async_cow->locked_page,
664 async_extent->start +
665 async_extent->ram_size - 1,
666 &page_started, &nr_written, 0);
671 * if page_started, cow_file_range inserted an
672 * inline extent and took care of all the unlocking
673 * and IO for us. Otherwise, we need to submit
674 * all those pages down to the drive.
676 if (!page_started && !ret)
677 extent_write_locked_range(io_tree,
678 inode, async_extent->start,
679 async_extent->start +
680 async_extent->ram_size - 1,
684 unlock_page(async_cow->locked_page);
690 lock_extent(io_tree, async_extent->start,
691 async_extent->start + async_extent->ram_size - 1);
693 ret = btrfs_reserve_extent(root,
694 async_extent->compressed_size,
695 async_extent->compressed_size,
696 0, alloc_hint, &ins, 1, 1);
700 for (i = 0; i < async_extent->nr_pages; i++) {
701 WARN_ON(async_extent->pages[i]->mapping);
702 page_cache_release(async_extent->pages[i]);
704 kfree(async_extent->pages);
705 async_extent->nr_pages = 0;
706 async_extent->pages = NULL;
708 if (ret == -ENOSPC) {
709 unlock_extent(io_tree, async_extent->start,
710 async_extent->start +
711 async_extent->ram_size - 1);
714 * we need to redirty the pages if we decide to
715 * fallback to uncompressed IO, otherwise we
716 * will not submit these pages down to lower
719 extent_range_redirty_for_io(inode,
721 async_extent->start +
722 async_extent->ram_size - 1);
730 * here we're doing allocation and writeback of the
733 btrfs_drop_extent_cache(inode, async_extent->start,
734 async_extent->start +
735 async_extent->ram_size - 1, 0);
737 em = alloc_extent_map();
740 goto out_free_reserve;
742 em->start = async_extent->start;
743 em->len = async_extent->ram_size;
744 em->orig_start = em->start;
745 em->mod_start = em->start;
746 em->mod_len = em->len;
748 em->block_start = ins.objectid;
749 em->block_len = ins.offset;
750 em->orig_block_len = ins.offset;
751 em->ram_bytes = async_extent->ram_size;
752 em->bdev = root->fs_info->fs_devices->latest_bdev;
753 em->compress_type = async_extent->compress_type;
754 set_bit(EXTENT_FLAG_PINNED, &em->flags);
755 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
759 write_lock(&em_tree->lock);
760 ret = add_extent_mapping(em_tree, em, 1);
761 write_unlock(&em_tree->lock);
762 if (ret != -EEXIST) {
766 btrfs_drop_extent_cache(inode, async_extent->start,
767 async_extent->start +
768 async_extent->ram_size - 1, 0);
772 goto out_free_reserve;
774 ret = btrfs_add_ordered_extent_compress(inode,
777 async_extent->ram_size,
779 BTRFS_ORDERED_COMPRESSED,
780 async_extent->compress_type);
782 goto out_free_reserve;
785 * clear dirty, set writeback and unlock the pages.
787 extent_clear_unlock_delalloc(inode, async_extent->start,
788 async_extent->start +
789 async_extent->ram_size - 1,
790 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
791 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
793 ret = btrfs_submit_compressed_write(inode,
795 async_extent->ram_size,
797 ins.offset, async_extent->pages,
798 async_extent->nr_pages);
799 alloc_hint = ins.objectid + ins.offset;
809 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
811 extent_clear_unlock_delalloc(inode, async_extent->start,
812 async_extent->start +
813 async_extent->ram_size - 1,
814 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
815 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
816 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
817 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
822 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
825 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
826 struct extent_map *em;
829 read_lock(&em_tree->lock);
830 em = search_extent_mapping(em_tree, start, num_bytes);
833 * if block start isn't an actual block number then find the
834 * first block in this inode and use that as a hint. If that
835 * block is also bogus then just don't worry about it.
837 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
839 em = search_extent_mapping(em_tree, 0, 0);
840 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
841 alloc_hint = em->block_start;
845 alloc_hint = em->block_start;
849 read_unlock(&em_tree->lock);
855 * when extent_io.c finds a delayed allocation range in the file,
856 * the call backs end up in this code. The basic idea is to
857 * allocate extents on disk for the range, and create ordered data structs
858 * in ram to track those extents.
860 * locked_page is the page that writepage had locked already. We use
861 * it to make sure we don't do extra locks or unlocks.
863 * *page_started is set to one if we unlock locked_page and do everything
864 * required to start IO on it. It may be clean and already done with
867 static noinline int cow_file_range(struct inode *inode,
868 struct page *locked_page,
869 u64 start, u64 end, int *page_started,
870 unsigned long *nr_written,
873 struct btrfs_root *root = BTRFS_I(inode)->root;
876 unsigned long ram_size;
879 u64 blocksize = root->sectorsize;
880 struct btrfs_key ins;
881 struct extent_map *em;
882 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
885 if (btrfs_is_free_space_inode(inode)) {
891 num_bytes = ALIGN(end - start + 1, blocksize);
892 num_bytes = max(blocksize, num_bytes);
893 disk_num_bytes = num_bytes;
895 /* if this is a small write inside eof, kick off defrag */
896 if (num_bytes < 64 * 1024 &&
897 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
898 btrfs_add_inode_defrag(NULL, inode);
901 /* lets try to make an inline extent */
902 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
905 extent_clear_unlock_delalloc(inode, start, end, NULL,
906 EXTENT_LOCKED | EXTENT_DELALLOC |
907 EXTENT_DEFRAG, PAGE_UNLOCK |
908 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
911 *nr_written = *nr_written +
912 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
915 } else if (ret < 0) {
920 BUG_ON(disk_num_bytes >
921 btrfs_super_total_bytes(root->fs_info->super_copy));
923 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
924 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
926 while (disk_num_bytes > 0) {
929 cur_alloc_size = disk_num_bytes;
930 ret = btrfs_reserve_extent(root, cur_alloc_size,
931 root->sectorsize, 0, alloc_hint,
936 em = alloc_extent_map();
942 em->orig_start = em->start;
943 ram_size = ins.offset;
944 em->len = ins.offset;
945 em->mod_start = em->start;
946 em->mod_len = em->len;
948 em->block_start = ins.objectid;
949 em->block_len = ins.offset;
950 em->orig_block_len = ins.offset;
951 em->ram_bytes = ram_size;
952 em->bdev = root->fs_info->fs_devices->latest_bdev;
953 set_bit(EXTENT_FLAG_PINNED, &em->flags);
957 write_lock(&em_tree->lock);
958 ret = add_extent_mapping(em_tree, em, 1);
959 write_unlock(&em_tree->lock);
960 if (ret != -EEXIST) {
964 btrfs_drop_extent_cache(inode, start,
965 start + ram_size - 1, 0);
970 cur_alloc_size = ins.offset;
971 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
972 ram_size, cur_alloc_size, 0);
976 if (root->root_key.objectid ==
977 BTRFS_DATA_RELOC_TREE_OBJECTID) {
978 ret = btrfs_reloc_clone_csums(inode, start,
984 if (disk_num_bytes < cur_alloc_size)
987 /* we're not doing compressed IO, don't unlock the first
988 * page (which the caller expects to stay locked), don't
989 * clear any dirty bits and don't set any writeback bits
991 * Do set the Private2 bit so we know this page was properly
992 * setup for writepage
994 op = unlock ? PAGE_UNLOCK : 0;
995 op |= PAGE_SET_PRIVATE2;
997 extent_clear_unlock_delalloc(inode, start,
998 start + ram_size - 1, locked_page,
999 EXTENT_LOCKED | EXTENT_DELALLOC,
1001 disk_num_bytes -= cur_alloc_size;
1002 num_bytes -= cur_alloc_size;
1003 alloc_hint = ins.objectid + ins.offset;
1004 start += cur_alloc_size;
1010 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1012 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1013 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1014 EXTENT_DELALLOC | EXTENT_DEFRAG,
1015 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1016 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1021 * work queue call back to started compression on a file and pages
1023 static noinline void async_cow_start(struct btrfs_work *work)
1025 struct async_cow *async_cow;
1027 async_cow = container_of(work, struct async_cow, work);
1029 compress_file_range(async_cow->inode, async_cow->locked_page,
1030 async_cow->start, async_cow->end, async_cow,
1032 if (num_added == 0) {
1033 btrfs_add_delayed_iput(async_cow->inode);
1034 async_cow->inode = NULL;
1039 * work queue call back to submit previously compressed pages
1041 static noinline void async_cow_submit(struct btrfs_work *work)
1043 struct async_cow *async_cow;
1044 struct btrfs_root *root;
1045 unsigned long nr_pages;
1047 async_cow = container_of(work, struct async_cow, work);
1049 root = async_cow->root;
1050 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1053 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1055 waitqueue_active(&root->fs_info->async_submit_wait))
1056 wake_up(&root->fs_info->async_submit_wait);
1058 if (async_cow->inode)
1059 submit_compressed_extents(async_cow->inode, async_cow);
1062 static noinline void async_cow_free(struct btrfs_work *work)
1064 struct async_cow *async_cow;
1065 async_cow = container_of(work, struct async_cow, work);
1066 if (async_cow->inode)
1067 btrfs_add_delayed_iput(async_cow->inode);
1071 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1072 u64 start, u64 end, int *page_started,
1073 unsigned long *nr_written)
1075 struct async_cow *async_cow;
1076 struct btrfs_root *root = BTRFS_I(inode)->root;
1077 unsigned long nr_pages;
1079 int limit = 10 * 1024 * 1024;
1081 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1082 1, 0, NULL, GFP_NOFS);
1083 while (start < end) {
1084 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1085 BUG_ON(!async_cow); /* -ENOMEM */
1086 async_cow->inode = igrab(inode);
1087 async_cow->root = root;
1088 async_cow->locked_page = locked_page;
1089 async_cow->start = start;
1091 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1094 cur_end = min(end, start + 512 * 1024 - 1);
1096 async_cow->end = cur_end;
1097 INIT_LIST_HEAD(&async_cow->extents);
1099 btrfs_init_work(&async_cow->work, async_cow_start,
1100 async_cow_submit, async_cow_free);
1102 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1104 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1106 btrfs_queue_work(root->fs_info->delalloc_workers,
1109 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1110 wait_event(root->fs_info->async_submit_wait,
1111 (atomic_read(&root->fs_info->async_delalloc_pages) <
1115 while (atomic_read(&root->fs_info->async_submit_draining) &&
1116 atomic_read(&root->fs_info->async_delalloc_pages)) {
1117 wait_event(root->fs_info->async_submit_wait,
1118 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1122 *nr_written += nr_pages;
1123 start = cur_end + 1;
1129 static noinline int csum_exist_in_range(struct btrfs_root *root,
1130 u64 bytenr, u64 num_bytes)
1133 struct btrfs_ordered_sum *sums;
1136 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1137 bytenr + num_bytes - 1, &list, 0);
1138 if (ret == 0 && list_empty(&list))
1141 while (!list_empty(&list)) {
1142 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1143 list_del(&sums->list);
1150 * when nowcow writeback call back. This checks for snapshots or COW copies
1151 * of the extents that exist in the file, and COWs the file as required.
1153 * If no cow copies or snapshots exist, we write directly to the existing
1156 static noinline int run_delalloc_nocow(struct inode *inode,
1157 struct page *locked_page,
1158 u64 start, u64 end, int *page_started, int force,
1159 unsigned long *nr_written)
1161 struct btrfs_root *root = BTRFS_I(inode)->root;
1162 struct btrfs_trans_handle *trans;
1163 struct extent_buffer *leaf;
1164 struct btrfs_path *path;
1165 struct btrfs_file_extent_item *fi;
1166 struct btrfs_key found_key;
1181 u64 ino = btrfs_ino(inode);
1183 path = btrfs_alloc_path();
1185 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1186 EXTENT_LOCKED | EXTENT_DELALLOC |
1187 EXTENT_DO_ACCOUNTING |
1188 EXTENT_DEFRAG, PAGE_UNLOCK |
1190 PAGE_SET_WRITEBACK |
1191 PAGE_END_WRITEBACK);
1195 nolock = btrfs_is_free_space_inode(inode);
1198 trans = btrfs_join_transaction_nolock(root);
1200 trans = btrfs_join_transaction(root);
1202 if (IS_ERR(trans)) {
1203 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1204 EXTENT_LOCKED | EXTENT_DELALLOC |
1205 EXTENT_DO_ACCOUNTING |
1206 EXTENT_DEFRAG, PAGE_UNLOCK |
1208 PAGE_SET_WRITEBACK |
1209 PAGE_END_WRITEBACK);
1210 btrfs_free_path(path);
1211 return PTR_ERR(trans);
1214 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1216 cow_start = (u64)-1;
1219 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1223 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1224 leaf = path->nodes[0];
1225 btrfs_item_key_to_cpu(leaf, &found_key,
1226 path->slots[0] - 1);
1227 if (found_key.objectid == ino &&
1228 found_key.type == BTRFS_EXTENT_DATA_KEY)
1233 leaf = path->nodes[0];
1234 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1235 ret = btrfs_next_leaf(root, path);
1240 leaf = path->nodes[0];
1246 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1248 if (found_key.objectid > ino ||
1249 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1250 found_key.offset > end)
1253 if (found_key.offset > cur_offset) {
1254 extent_end = found_key.offset;
1259 fi = btrfs_item_ptr(leaf, path->slots[0],
1260 struct btrfs_file_extent_item);
1261 extent_type = btrfs_file_extent_type(leaf, fi);
1263 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1264 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1265 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1266 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1267 extent_offset = btrfs_file_extent_offset(leaf, fi);
1268 extent_end = found_key.offset +
1269 btrfs_file_extent_num_bytes(leaf, fi);
1271 btrfs_file_extent_disk_num_bytes(leaf, fi);
1272 if (extent_end <= start) {
1276 if (disk_bytenr == 0)
1278 if (btrfs_file_extent_compression(leaf, fi) ||
1279 btrfs_file_extent_encryption(leaf, fi) ||
1280 btrfs_file_extent_other_encoding(leaf, fi))
1282 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1284 if (btrfs_extent_readonly(root, disk_bytenr))
1286 if (btrfs_cross_ref_exist(trans, root, ino,
1288 extent_offset, disk_bytenr))
1290 disk_bytenr += extent_offset;
1291 disk_bytenr += cur_offset - found_key.offset;
1292 num_bytes = min(end + 1, extent_end) - cur_offset;
1294 * if there are pending snapshots for this root,
1295 * we fall into common COW way.
1298 err = btrfs_start_nocow_write(root);
1303 * force cow if csum exists in the range.
1304 * this ensure that csum for a given extent are
1305 * either valid or do not exist.
1307 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1310 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1311 extent_end = found_key.offset +
1312 btrfs_file_extent_inline_len(leaf,
1313 path->slots[0], fi);
1314 extent_end = ALIGN(extent_end, root->sectorsize);
1319 if (extent_end <= start) {
1321 if (!nolock && nocow)
1322 btrfs_end_nocow_write(root);
1326 if (cow_start == (u64)-1)
1327 cow_start = cur_offset;
1328 cur_offset = extent_end;
1329 if (cur_offset > end)
1335 btrfs_release_path(path);
1336 if (cow_start != (u64)-1) {
1337 ret = cow_file_range(inode, locked_page,
1338 cow_start, found_key.offset - 1,
1339 page_started, nr_written, 1);
1341 if (!nolock && nocow)
1342 btrfs_end_nocow_write(root);
1345 cow_start = (u64)-1;
1348 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1349 struct extent_map *em;
1350 struct extent_map_tree *em_tree;
1351 em_tree = &BTRFS_I(inode)->extent_tree;
1352 em = alloc_extent_map();
1353 BUG_ON(!em); /* -ENOMEM */
1354 em->start = cur_offset;
1355 em->orig_start = found_key.offset - extent_offset;
1356 em->len = num_bytes;
1357 em->block_len = num_bytes;
1358 em->block_start = disk_bytenr;
1359 em->orig_block_len = disk_num_bytes;
1360 em->ram_bytes = ram_bytes;
1361 em->bdev = root->fs_info->fs_devices->latest_bdev;
1362 em->mod_start = em->start;
1363 em->mod_len = em->len;
1364 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1365 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1366 em->generation = -1;
1368 write_lock(&em_tree->lock);
1369 ret = add_extent_mapping(em_tree, em, 1);
1370 write_unlock(&em_tree->lock);
1371 if (ret != -EEXIST) {
1372 free_extent_map(em);
1375 btrfs_drop_extent_cache(inode, em->start,
1376 em->start + em->len - 1, 0);
1378 type = BTRFS_ORDERED_PREALLOC;
1380 type = BTRFS_ORDERED_NOCOW;
1383 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1384 num_bytes, num_bytes, type);
1385 BUG_ON(ret); /* -ENOMEM */
1387 if (root->root_key.objectid ==
1388 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1389 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1392 if (!nolock && nocow)
1393 btrfs_end_nocow_write(root);
1398 extent_clear_unlock_delalloc(inode, cur_offset,
1399 cur_offset + num_bytes - 1,
1400 locked_page, EXTENT_LOCKED |
1401 EXTENT_DELALLOC, PAGE_UNLOCK |
1403 if (!nolock && nocow)
1404 btrfs_end_nocow_write(root);
1405 cur_offset = extent_end;
1406 if (cur_offset > end)
1409 btrfs_release_path(path);
1411 if (cur_offset <= end && cow_start == (u64)-1) {
1412 cow_start = cur_offset;
1416 if (cow_start != (u64)-1) {
1417 ret = cow_file_range(inode, locked_page, cow_start, end,
1418 page_started, nr_written, 1);
1424 err = btrfs_end_transaction(trans, root);
1428 if (ret && cur_offset < end)
1429 extent_clear_unlock_delalloc(inode, cur_offset, end,
1430 locked_page, EXTENT_LOCKED |
1431 EXTENT_DELALLOC | EXTENT_DEFRAG |
1432 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1434 PAGE_SET_WRITEBACK |
1435 PAGE_END_WRITEBACK);
1436 btrfs_free_path(path);
1441 * extent_io.c call back to do delayed allocation processing
1443 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1444 u64 start, u64 end, int *page_started,
1445 unsigned long *nr_written)
1448 struct btrfs_root *root = BTRFS_I(inode)->root;
1450 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1451 ret = run_delalloc_nocow(inode, locked_page, start, end,
1452 page_started, 1, nr_written);
1453 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1454 ret = run_delalloc_nocow(inode, locked_page, start, end,
1455 page_started, 0, nr_written);
1456 } else if (!btrfs_test_opt(root, COMPRESS) &&
1457 !(BTRFS_I(inode)->force_compress) &&
1458 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1459 ret = cow_file_range(inode, locked_page, start, end,
1460 page_started, nr_written, 1);
1462 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1463 &BTRFS_I(inode)->runtime_flags);
1464 ret = cow_file_range_async(inode, locked_page, start, end,
1465 page_started, nr_written);
1470 static void btrfs_split_extent_hook(struct inode *inode,
1471 struct extent_state *orig, u64 split)
1473 /* not delalloc, ignore it */
1474 if (!(orig->state & EXTENT_DELALLOC))
1477 spin_lock(&BTRFS_I(inode)->lock);
1478 BTRFS_I(inode)->outstanding_extents++;
1479 spin_unlock(&BTRFS_I(inode)->lock);
1483 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1484 * extents so we can keep track of new extents that are just merged onto old
1485 * extents, such as when we are doing sequential writes, so we can properly
1486 * account for the metadata space we'll need.
1488 static void btrfs_merge_extent_hook(struct inode *inode,
1489 struct extent_state *new,
1490 struct extent_state *other)
1492 /* not delalloc, ignore it */
1493 if (!(other->state & EXTENT_DELALLOC))
1496 spin_lock(&BTRFS_I(inode)->lock);
1497 BTRFS_I(inode)->outstanding_extents--;
1498 spin_unlock(&BTRFS_I(inode)->lock);
1501 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1502 struct inode *inode)
1504 spin_lock(&root->delalloc_lock);
1505 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1506 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1507 &root->delalloc_inodes);
1508 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1509 &BTRFS_I(inode)->runtime_flags);
1510 root->nr_delalloc_inodes++;
1511 if (root->nr_delalloc_inodes == 1) {
1512 spin_lock(&root->fs_info->delalloc_root_lock);
1513 BUG_ON(!list_empty(&root->delalloc_root));
1514 list_add_tail(&root->delalloc_root,
1515 &root->fs_info->delalloc_roots);
1516 spin_unlock(&root->fs_info->delalloc_root_lock);
1519 spin_unlock(&root->delalloc_lock);
1522 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1523 struct inode *inode)
1525 spin_lock(&root->delalloc_lock);
1526 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1527 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1528 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1529 &BTRFS_I(inode)->runtime_flags);
1530 root->nr_delalloc_inodes--;
1531 if (!root->nr_delalloc_inodes) {
1532 spin_lock(&root->fs_info->delalloc_root_lock);
1533 BUG_ON(list_empty(&root->delalloc_root));
1534 list_del_init(&root->delalloc_root);
1535 spin_unlock(&root->fs_info->delalloc_root_lock);
1538 spin_unlock(&root->delalloc_lock);
1542 * extent_io.c set_bit_hook, used to track delayed allocation
1543 * bytes in this file, and to maintain the list of inodes that
1544 * have pending delalloc work to be done.
1546 static void btrfs_set_bit_hook(struct inode *inode,
1547 struct extent_state *state, unsigned long *bits)
1551 * set_bit and clear bit hooks normally require _irqsave/restore
1552 * but in this case, we are only testing for the DELALLOC
1553 * bit, which is only set or cleared with irqs on
1555 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1556 struct btrfs_root *root = BTRFS_I(inode)->root;
1557 u64 len = state->end + 1 - state->start;
1558 bool do_list = !btrfs_is_free_space_inode(inode);
1560 if (*bits & EXTENT_FIRST_DELALLOC) {
1561 *bits &= ~EXTENT_FIRST_DELALLOC;
1563 spin_lock(&BTRFS_I(inode)->lock);
1564 BTRFS_I(inode)->outstanding_extents++;
1565 spin_unlock(&BTRFS_I(inode)->lock);
1568 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1569 root->fs_info->delalloc_batch);
1570 spin_lock(&BTRFS_I(inode)->lock);
1571 BTRFS_I(inode)->delalloc_bytes += len;
1572 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1573 &BTRFS_I(inode)->runtime_flags))
1574 btrfs_add_delalloc_inodes(root, inode);
1575 spin_unlock(&BTRFS_I(inode)->lock);
1580 * extent_io.c clear_bit_hook, see set_bit_hook for why
1582 static void btrfs_clear_bit_hook(struct inode *inode,
1583 struct extent_state *state,
1584 unsigned long *bits)
1587 * set_bit and clear bit hooks normally require _irqsave/restore
1588 * but in this case, we are only testing for the DELALLOC
1589 * bit, which is only set or cleared with irqs on
1591 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1592 struct btrfs_root *root = BTRFS_I(inode)->root;
1593 u64 len = state->end + 1 - state->start;
1594 bool do_list = !btrfs_is_free_space_inode(inode);
1596 if (*bits & EXTENT_FIRST_DELALLOC) {
1597 *bits &= ~EXTENT_FIRST_DELALLOC;
1598 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1599 spin_lock(&BTRFS_I(inode)->lock);
1600 BTRFS_I(inode)->outstanding_extents--;
1601 spin_unlock(&BTRFS_I(inode)->lock);
1605 * We don't reserve metadata space for space cache inodes so we
1606 * don't need to call dellalloc_release_metadata if there is an
1609 if (*bits & EXTENT_DO_ACCOUNTING &&
1610 root != root->fs_info->tree_root)
1611 btrfs_delalloc_release_metadata(inode, len);
1613 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1614 && do_list && !(state->state & EXTENT_NORESERVE))
1615 btrfs_free_reserved_data_space(inode, len);
1617 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1618 root->fs_info->delalloc_batch);
1619 spin_lock(&BTRFS_I(inode)->lock);
1620 BTRFS_I(inode)->delalloc_bytes -= len;
1621 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1622 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1623 &BTRFS_I(inode)->runtime_flags))
1624 btrfs_del_delalloc_inode(root, inode);
1625 spin_unlock(&BTRFS_I(inode)->lock);
1630 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1631 * we don't create bios that span stripes or chunks
1633 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1634 size_t size, struct bio *bio,
1635 unsigned long bio_flags)
1637 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1638 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1643 if (bio_flags & EXTENT_BIO_COMPRESSED)
1646 length = bio->bi_iter.bi_size;
1647 map_length = length;
1648 ret = btrfs_map_block(root->fs_info, rw, logical,
1649 &map_length, NULL, 0);
1650 /* Will always return 0 with map_multi == NULL */
1652 if (map_length < length + size)
1658 * in order to insert checksums into the metadata in large chunks,
1659 * we wait until bio submission time. All the pages in the bio are
1660 * checksummed and sums are attached onto the ordered extent record.
1662 * At IO completion time the cums attached on the ordered extent record
1663 * are inserted into the btree
1665 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1666 struct bio *bio, int mirror_num,
1667 unsigned long bio_flags,
1670 struct btrfs_root *root = BTRFS_I(inode)->root;
1673 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1674 BUG_ON(ret); /* -ENOMEM */
1679 * in order to insert checksums into the metadata in large chunks,
1680 * we wait until bio submission time. All the pages in the bio are
1681 * checksummed and sums are attached onto the ordered extent record.
1683 * At IO completion time the cums attached on the ordered extent record
1684 * are inserted into the btree
1686 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1687 int mirror_num, unsigned long bio_flags,
1690 struct btrfs_root *root = BTRFS_I(inode)->root;
1693 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1695 bio_endio(bio, ret);
1700 * extent_io.c submission hook. This does the right thing for csum calculation
1701 * on write, or reading the csums from the tree before a read
1703 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1704 int mirror_num, unsigned long bio_flags,
1707 struct btrfs_root *root = BTRFS_I(inode)->root;
1711 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1713 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1715 if (btrfs_is_free_space_inode(inode))
1718 if (!(rw & REQ_WRITE)) {
1719 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1723 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1724 ret = btrfs_submit_compressed_read(inode, bio,
1728 } else if (!skip_sum) {
1729 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1734 } else if (async && !skip_sum) {
1735 /* csum items have already been cloned */
1736 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1738 /* we're doing a write, do the async checksumming */
1739 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1740 inode, rw, bio, mirror_num,
1741 bio_flags, bio_offset,
1742 __btrfs_submit_bio_start,
1743 __btrfs_submit_bio_done);
1745 } else if (!skip_sum) {
1746 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1752 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1756 bio_endio(bio, ret);
1761 * given a list of ordered sums record them in the inode. This happens
1762 * at IO completion time based on sums calculated at bio submission time.
1764 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1765 struct inode *inode, u64 file_offset,
1766 struct list_head *list)
1768 struct btrfs_ordered_sum *sum;
1770 list_for_each_entry(sum, list, list) {
1771 trans->adding_csums = 1;
1772 btrfs_csum_file_blocks(trans,
1773 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1774 trans->adding_csums = 0;
1779 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1780 struct extent_state **cached_state)
1782 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1783 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1784 cached_state, GFP_NOFS);
1787 /* see btrfs_writepage_start_hook for details on why this is required */
1788 struct btrfs_writepage_fixup {
1790 struct btrfs_work work;
1793 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1795 struct btrfs_writepage_fixup *fixup;
1796 struct btrfs_ordered_extent *ordered;
1797 struct extent_state *cached_state = NULL;
1799 struct inode *inode;
1804 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1808 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1809 ClearPageChecked(page);
1813 inode = page->mapping->host;
1814 page_start = page_offset(page);
1815 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1817 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1820 /* already ordered? We're done */
1821 if (PagePrivate2(page))
1824 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1826 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1827 page_end, &cached_state, GFP_NOFS);
1829 btrfs_start_ordered_extent(inode, ordered, 1);
1830 btrfs_put_ordered_extent(ordered);
1834 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1836 mapping_set_error(page->mapping, ret);
1837 end_extent_writepage(page, ret, page_start, page_end);
1838 ClearPageChecked(page);
1842 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1843 ClearPageChecked(page);
1844 set_page_dirty(page);
1846 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1847 &cached_state, GFP_NOFS);
1850 page_cache_release(page);
1855 * There are a few paths in the higher layers of the kernel that directly
1856 * set the page dirty bit without asking the filesystem if it is a
1857 * good idea. This causes problems because we want to make sure COW
1858 * properly happens and the data=ordered rules are followed.
1860 * In our case any range that doesn't have the ORDERED bit set
1861 * hasn't been properly setup for IO. We kick off an async process
1862 * to fix it up. The async helper will wait for ordered extents, set
1863 * the delalloc bit and make it safe to write the page.
1865 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1867 struct inode *inode = page->mapping->host;
1868 struct btrfs_writepage_fixup *fixup;
1869 struct btrfs_root *root = BTRFS_I(inode)->root;
1871 /* this page is properly in the ordered list */
1872 if (TestClearPagePrivate2(page))
1875 if (PageChecked(page))
1878 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1882 SetPageChecked(page);
1883 page_cache_get(page);
1884 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
1886 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
1890 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1891 struct inode *inode, u64 file_pos,
1892 u64 disk_bytenr, u64 disk_num_bytes,
1893 u64 num_bytes, u64 ram_bytes,
1894 u8 compression, u8 encryption,
1895 u16 other_encoding, int extent_type)
1897 struct btrfs_root *root = BTRFS_I(inode)->root;
1898 struct btrfs_file_extent_item *fi;
1899 struct btrfs_path *path;
1900 struct extent_buffer *leaf;
1901 struct btrfs_key ins;
1902 int extent_inserted = 0;
1905 path = btrfs_alloc_path();
1910 * we may be replacing one extent in the tree with another.
1911 * The new extent is pinned in the extent map, and we don't want
1912 * to drop it from the cache until it is completely in the btree.
1914 * So, tell btrfs_drop_extents to leave this extent in the cache.
1915 * the caller is expected to unpin it and allow it to be merged
1918 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
1919 file_pos + num_bytes, NULL, 0,
1920 1, sizeof(*fi), &extent_inserted);
1924 if (!extent_inserted) {
1925 ins.objectid = btrfs_ino(inode);
1926 ins.offset = file_pos;
1927 ins.type = BTRFS_EXTENT_DATA_KEY;
1929 path->leave_spinning = 1;
1930 ret = btrfs_insert_empty_item(trans, root, path, &ins,
1935 leaf = path->nodes[0];
1936 fi = btrfs_item_ptr(leaf, path->slots[0],
1937 struct btrfs_file_extent_item);
1938 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1939 btrfs_set_file_extent_type(leaf, fi, extent_type);
1940 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1941 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1942 btrfs_set_file_extent_offset(leaf, fi, 0);
1943 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1944 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1945 btrfs_set_file_extent_compression(leaf, fi, compression);
1946 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1947 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1949 btrfs_mark_buffer_dirty(leaf);
1950 btrfs_release_path(path);
1952 inode_add_bytes(inode, num_bytes);
1954 ins.objectid = disk_bytenr;
1955 ins.offset = disk_num_bytes;
1956 ins.type = BTRFS_EXTENT_ITEM_KEY;
1957 ret = btrfs_alloc_reserved_file_extent(trans, root,
1958 root->root_key.objectid,
1959 btrfs_ino(inode), file_pos, &ins);
1961 btrfs_free_path(path);
1966 /* snapshot-aware defrag */
1967 struct sa_defrag_extent_backref {
1968 struct rb_node node;
1969 struct old_sa_defrag_extent *old;
1978 struct old_sa_defrag_extent {
1979 struct list_head list;
1980 struct new_sa_defrag_extent *new;
1989 struct new_sa_defrag_extent {
1990 struct rb_root root;
1991 struct list_head head;
1992 struct btrfs_path *path;
1993 struct inode *inode;
2001 static int backref_comp(struct sa_defrag_extent_backref *b1,
2002 struct sa_defrag_extent_backref *b2)
2004 if (b1->root_id < b2->root_id)
2006 else if (b1->root_id > b2->root_id)
2009 if (b1->inum < b2->inum)
2011 else if (b1->inum > b2->inum)
2014 if (b1->file_pos < b2->file_pos)
2016 else if (b1->file_pos > b2->file_pos)
2020 * [------------------------------] ===> (a range of space)
2021 * |<--->| |<---->| =============> (fs/file tree A)
2022 * |<---------------------------->| ===> (fs/file tree B)
2024 * A range of space can refer to two file extents in one tree while
2025 * refer to only one file extent in another tree.
2027 * So we may process a disk offset more than one time(two extents in A)
2028 * and locate at the same extent(one extent in B), then insert two same
2029 * backrefs(both refer to the extent in B).
2034 static void backref_insert(struct rb_root *root,
2035 struct sa_defrag_extent_backref *backref)
2037 struct rb_node **p = &root->rb_node;
2038 struct rb_node *parent = NULL;
2039 struct sa_defrag_extent_backref *entry;
2044 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2046 ret = backref_comp(backref, entry);
2050 p = &(*p)->rb_right;
2053 rb_link_node(&backref->node, parent, p);
2054 rb_insert_color(&backref->node, root);
2058 * Note the backref might has changed, and in this case we just return 0.
2060 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2063 struct btrfs_file_extent_item *extent;
2064 struct btrfs_fs_info *fs_info;
2065 struct old_sa_defrag_extent *old = ctx;
2066 struct new_sa_defrag_extent *new = old->new;
2067 struct btrfs_path *path = new->path;
2068 struct btrfs_key key;
2069 struct btrfs_root *root;
2070 struct sa_defrag_extent_backref *backref;
2071 struct extent_buffer *leaf;
2072 struct inode *inode = new->inode;
2078 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2079 inum == btrfs_ino(inode))
2082 key.objectid = root_id;
2083 key.type = BTRFS_ROOT_ITEM_KEY;
2084 key.offset = (u64)-1;
2086 fs_info = BTRFS_I(inode)->root->fs_info;
2087 root = btrfs_read_fs_root_no_name(fs_info, &key);
2089 if (PTR_ERR(root) == -ENOENT)
2092 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2093 inum, offset, root_id);
2094 return PTR_ERR(root);
2097 key.objectid = inum;
2098 key.type = BTRFS_EXTENT_DATA_KEY;
2099 if (offset > (u64)-1 << 32)
2102 key.offset = offset;
2104 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2105 if (WARN_ON(ret < 0))
2112 leaf = path->nodes[0];
2113 slot = path->slots[0];
2115 if (slot >= btrfs_header_nritems(leaf)) {
2116 ret = btrfs_next_leaf(root, path);
2119 } else if (ret > 0) {
2128 btrfs_item_key_to_cpu(leaf, &key, slot);
2130 if (key.objectid > inum)
2133 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2136 extent = btrfs_item_ptr(leaf, slot,
2137 struct btrfs_file_extent_item);
2139 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2143 * 'offset' refers to the exact key.offset,
2144 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2145 * (key.offset - extent_offset).
2147 if (key.offset != offset)
2150 extent_offset = btrfs_file_extent_offset(leaf, extent);
2151 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2153 if (extent_offset >= old->extent_offset + old->offset +
2154 old->len || extent_offset + num_bytes <=
2155 old->extent_offset + old->offset)
2160 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2166 backref->root_id = root_id;
2167 backref->inum = inum;
2168 backref->file_pos = offset;
2169 backref->num_bytes = num_bytes;
2170 backref->extent_offset = extent_offset;
2171 backref->generation = btrfs_file_extent_generation(leaf, extent);
2173 backref_insert(&new->root, backref);
2176 btrfs_release_path(path);
2181 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2182 struct new_sa_defrag_extent *new)
2184 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2185 struct old_sa_defrag_extent *old, *tmp;
2190 list_for_each_entry_safe(old, tmp, &new->head, list) {
2191 ret = iterate_inodes_from_logical(old->bytenr +
2192 old->extent_offset, fs_info,
2193 path, record_one_backref,
2195 if (ret < 0 && ret != -ENOENT)
2198 /* no backref to be processed for this extent */
2200 list_del(&old->list);
2205 if (list_empty(&new->head))
2211 static int relink_is_mergable(struct extent_buffer *leaf,
2212 struct btrfs_file_extent_item *fi,
2213 struct new_sa_defrag_extent *new)
2215 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2218 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2221 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2224 if (btrfs_file_extent_encryption(leaf, fi) ||
2225 btrfs_file_extent_other_encoding(leaf, fi))
2232 * Note the backref might has changed, and in this case we just return 0.
2234 static noinline int relink_extent_backref(struct btrfs_path *path,
2235 struct sa_defrag_extent_backref *prev,
2236 struct sa_defrag_extent_backref *backref)
2238 struct btrfs_file_extent_item *extent;
2239 struct btrfs_file_extent_item *item;
2240 struct btrfs_ordered_extent *ordered;
2241 struct btrfs_trans_handle *trans;
2242 struct btrfs_fs_info *fs_info;
2243 struct btrfs_root *root;
2244 struct btrfs_key key;
2245 struct extent_buffer *leaf;
2246 struct old_sa_defrag_extent *old = backref->old;
2247 struct new_sa_defrag_extent *new = old->new;
2248 struct inode *src_inode = new->inode;
2249 struct inode *inode;
2250 struct extent_state *cached = NULL;
2259 if (prev && prev->root_id == backref->root_id &&
2260 prev->inum == backref->inum &&
2261 prev->file_pos + prev->num_bytes == backref->file_pos)
2264 /* step 1: get root */
2265 key.objectid = backref->root_id;
2266 key.type = BTRFS_ROOT_ITEM_KEY;
2267 key.offset = (u64)-1;
2269 fs_info = BTRFS_I(src_inode)->root->fs_info;
2270 index = srcu_read_lock(&fs_info->subvol_srcu);
2272 root = btrfs_read_fs_root_no_name(fs_info, &key);
2274 srcu_read_unlock(&fs_info->subvol_srcu, index);
2275 if (PTR_ERR(root) == -ENOENT)
2277 return PTR_ERR(root);
2280 if (btrfs_root_readonly(root)) {
2281 srcu_read_unlock(&fs_info->subvol_srcu, index);
2285 /* step 2: get inode */
2286 key.objectid = backref->inum;
2287 key.type = BTRFS_INODE_ITEM_KEY;
2290 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2291 if (IS_ERR(inode)) {
2292 srcu_read_unlock(&fs_info->subvol_srcu, index);
2296 srcu_read_unlock(&fs_info->subvol_srcu, index);
2298 /* step 3: relink backref */
2299 lock_start = backref->file_pos;
2300 lock_end = backref->file_pos + backref->num_bytes - 1;
2301 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2304 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2306 btrfs_put_ordered_extent(ordered);
2310 trans = btrfs_join_transaction(root);
2311 if (IS_ERR(trans)) {
2312 ret = PTR_ERR(trans);
2316 key.objectid = backref->inum;
2317 key.type = BTRFS_EXTENT_DATA_KEY;
2318 key.offset = backref->file_pos;
2320 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2323 } else if (ret > 0) {
2328 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2329 struct btrfs_file_extent_item);
2331 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2332 backref->generation)
2335 btrfs_release_path(path);
2337 start = backref->file_pos;
2338 if (backref->extent_offset < old->extent_offset + old->offset)
2339 start += old->extent_offset + old->offset -
2340 backref->extent_offset;
2342 len = min(backref->extent_offset + backref->num_bytes,
2343 old->extent_offset + old->offset + old->len);
2344 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2346 ret = btrfs_drop_extents(trans, root, inode, start,
2351 key.objectid = btrfs_ino(inode);
2352 key.type = BTRFS_EXTENT_DATA_KEY;
2355 path->leave_spinning = 1;
2357 struct btrfs_file_extent_item *fi;
2359 struct btrfs_key found_key;
2361 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2366 leaf = path->nodes[0];
2367 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2369 fi = btrfs_item_ptr(leaf, path->slots[0],
2370 struct btrfs_file_extent_item);
2371 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2373 if (extent_len + found_key.offset == start &&
2374 relink_is_mergable(leaf, fi, new)) {
2375 btrfs_set_file_extent_num_bytes(leaf, fi,
2377 btrfs_mark_buffer_dirty(leaf);
2378 inode_add_bytes(inode, len);
2384 btrfs_release_path(path);
2389 ret = btrfs_insert_empty_item(trans, root, path, &key,
2392 btrfs_abort_transaction(trans, root, ret);
2396 leaf = path->nodes[0];
2397 item = btrfs_item_ptr(leaf, path->slots[0],
2398 struct btrfs_file_extent_item);
2399 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2400 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2401 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2402 btrfs_set_file_extent_num_bytes(leaf, item, len);
2403 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2404 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2405 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2406 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2407 btrfs_set_file_extent_encryption(leaf, item, 0);
2408 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2410 btrfs_mark_buffer_dirty(leaf);
2411 inode_add_bytes(inode, len);
2412 btrfs_release_path(path);
2414 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2416 backref->root_id, backref->inum,
2417 new->file_pos, 0); /* start - extent_offset */
2419 btrfs_abort_transaction(trans, root, ret);
2425 btrfs_release_path(path);
2426 path->leave_spinning = 0;
2427 btrfs_end_transaction(trans, root);
2429 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2435 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2437 struct old_sa_defrag_extent *old, *tmp;
2442 list_for_each_entry_safe(old, tmp, &new->head, list) {
2443 list_del(&old->list);
2449 static void relink_file_extents(struct new_sa_defrag_extent *new)
2451 struct btrfs_path *path;
2452 struct sa_defrag_extent_backref *backref;
2453 struct sa_defrag_extent_backref *prev = NULL;
2454 struct inode *inode;
2455 struct btrfs_root *root;
2456 struct rb_node *node;
2460 root = BTRFS_I(inode)->root;
2462 path = btrfs_alloc_path();
2466 if (!record_extent_backrefs(path, new)) {
2467 btrfs_free_path(path);
2470 btrfs_release_path(path);
2473 node = rb_first(&new->root);
2476 rb_erase(node, &new->root);
2478 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2480 ret = relink_extent_backref(path, prev, backref);
2493 btrfs_free_path(path);
2495 free_sa_defrag_extent(new);
2497 atomic_dec(&root->fs_info->defrag_running);
2498 wake_up(&root->fs_info->transaction_wait);
2501 static struct new_sa_defrag_extent *
2502 record_old_file_extents(struct inode *inode,
2503 struct btrfs_ordered_extent *ordered)
2505 struct btrfs_root *root = BTRFS_I(inode)->root;
2506 struct btrfs_path *path;
2507 struct btrfs_key key;
2508 struct old_sa_defrag_extent *old;
2509 struct new_sa_defrag_extent *new;
2512 new = kmalloc(sizeof(*new), GFP_NOFS);
2517 new->file_pos = ordered->file_offset;
2518 new->len = ordered->len;
2519 new->bytenr = ordered->start;
2520 new->disk_len = ordered->disk_len;
2521 new->compress_type = ordered->compress_type;
2522 new->root = RB_ROOT;
2523 INIT_LIST_HEAD(&new->head);
2525 path = btrfs_alloc_path();
2529 key.objectid = btrfs_ino(inode);
2530 key.type = BTRFS_EXTENT_DATA_KEY;
2531 key.offset = new->file_pos;
2533 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2536 if (ret > 0 && path->slots[0] > 0)
2539 /* find out all the old extents for the file range */
2541 struct btrfs_file_extent_item *extent;
2542 struct extent_buffer *l;
2551 slot = path->slots[0];
2553 if (slot >= btrfs_header_nritems(l)) {
2554 ret = btrfs_next_leaf(root, path);
2562 btrfs_item_key_to_cpu(l, &key, slot);
2564 if (key.objectid != btrfs_ino(inode))
2566 if (key.type != BTRFS_EXTENT_DATA_KEY)
2568 if (key.offset >= new->file_pos + new->len)
2571 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2573 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2574 if (key.offset + num_bytes < new->file_pos)
2577 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2581 extent_offset = btrfs_file_extent_offset(l, extent);
2583 old = kmalloc(sizeof(*old), GFP_NOFS);
2587 offset = max(new->file_pos, key.offset);
2588 end = min(new->file_pos + new->len, key.offset + num_bytes);
2590 old->bytenr = disk_bytenr;
2591 old->extent_offset = extent_offset;
2592 old->offset = offset - key.offset;
2593 old->len = end - offset;
2596 list_add_tail(&old->list, &new->head);
2602 btrfs_free_path(path);
2603 atomic_inc(&root->fs_info->defrag_running);
2608 btrfs_free_path(path);
2610 free_sa_defrag_extent(new);
2614 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2617 struct btrfs_block_group_cache *cache;
2619 cache = btrfs_lookup_block_group(root->fs_info, start);
2622 spin_lock(&cache->lock);
2623 cache->delalloc_bytes -= len;
2624 spin_unlock(&cache->lock);
2626 btrfs_put_block_group(cache);
2629 /* as ordered data IO finishes, this gets called so we can finish
2630 * an ordered extent if the range of bytes in the file it covers are
2633 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2635 struct inode *inode = ordered_extent->inode;
2636 struct btrfs_root *root = BTRFS_I(inode)->root;
2637 struct btrfs_trans_handle *trans = NULL;
2638 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2639 struct extent_state *cached_state = NULL;
2640 struct new_sa_defrag_extent *new = NULL;
2641 int compress_type = 0;
2643 u64 logical_len = ordered_extent->len;
2645 bool truncated = false;
2647 nolock = btrfs_is_free_space_inode(inode);
2649 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2654 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2656 logical_len = ordered_extent->truncated_len;
2657 /* Truncated the entire extent, don't bother adding */
2662 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2663 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2664 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2666 trans = btrfs_join_transaction_nolock(root);
2668 trans = btrfs_join_transaction(root);
2669 if (IS_ERR(trans)) {
2670 ret = PTR_ERR(trans);
2674 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2675 ret = btrfs_update_inode_fallback(trans, root, inode);
2676 if (ret) /* -ENOMEM or corruption */
2677 btrfs_abort_transaction(trans, root, ret);
2681 lock_extent_bits(io_tree, ordered_extent->file_offset,
2682 ordered_extent->file_offset + ordered_extent->len - 1,
2685 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2686 ordered_extent->file_offset + ordered_extent->len - 1,
2687 EXTENT_DEFRAG, 1, cached_state);
2689 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2690 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2691 /* the inode is shared */
2692 new = record_old_file_extents(inode, ordered_extent);
2694 clear_extent_bit(io_tree, ordered_extent->file_offset,
2695 ordered_extent->file_offset + ordered_extent->len - 1,
2696 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2700 trans = btrfs_join_transaction_nolock(root);
2702 trans = btrfs_join_transaction(root);
2703 if (IS_ERR(trans)) {
2704 ret = PTR_ERR(trans);
2709 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2711 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2712 compress_type = ordered_extent->compress_type;
2713 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2714 BUG_ON(compress_type);
2715 ret = btrfs_mark_extent_written(trans, inode,
2716 ordered_extent->file_offset,
2717 ordered_extent->file_offset +
2720 BUG_ON(root == root->fs_info->tree_root);
2721 ret = insert_reserved_file_extent(trans, inode,
2722 ordered_extent->file_offset,
2723 ordered_extent->start,
2724 ordered_extent->disk_len,
2725 logical_len, logical_len,
2726 compress_type, 0, 0,
2727 BTRFS_FILE_EXTENT_REG);
2729 btrfs_release_delalloc_bytes(root,
2730 ordered_extent->start,
2731 ordered_extent->disk_len);
2733 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2734 ordered_extent->file_offset, ordered_extent->len,
2737 btrfs_abort_transaction(trans, root, ret);
2741 add_pending_csums(trans, inode, ordered_extent->file_offset,
2742 &ordered_extent->list);
2744 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2745 ret = btrfs_update_inode_fallback(trans, root, inode);
2746 if (ret) { /* -ENOMEM or corruption */
2747 btrfs_abort_transaction(trans, root, ret);
2752 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2753 ordered_extent->file_offset +
2754 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2756 if (root != root->fs_info->tree_root)
2757 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2759 btrfs_end_transaction(trans, root);
2761 if (ret || truncated) {
2765 start = ordered_extent->file_offset + logical_len;
2767 start = ordered_extent->file_offset;
2768 end = ordered_extent->file_offset + ordered_extent->len - 1;
2769 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2771 /* Drop the cache for the part of the extent we didn't write. */
2772 btrfs_drop_extent_cache(inode, start, end, 0);
2775 * If the ordered extent had an IOERR or something else went
2776 * wrong we need to return the space for this ordered extent
2777 * back to the allocator. We only free the extent in the
2778 * truncated case if we didn't write out the extent at all.
2780 if ((ret || !logical_len) &&
2781 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2782 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2783 btrfs_free_reserved_extent(root, ordered_extent->start,
2784 ordered_extent->disk_len, 1);
2789 * This needs to be done to make sure anybody waiting knows we are done
2790 * updating everything for this ordered extent.
2792 btrfs_remove_ordered_extent(inode, ordered_extent);
2794 /* for snapshot-aware defrag */
2797 free_sa_defrag_extent(new);
2798 atomic_dec(&root->fs_info->defrag_running);
2800 relink_file_extents(new);
2805 btrfs_put_ordered_extent(ordered_extent);
2806 /* once for the tree */
2807 btrfs_put_ordered_extent(ordered_extent);
2812 static void finish_ordered_fn(struct btrfs_work *work)
2814 struct btrfs_ordered_extent *ordered_extent;
2815 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2816 btrfs_finish_ordered_io(ordered_extent);
2819 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2820 struct extent_state *state, int uptodate)
2822 struct inode *inode = page->mapping->host;
2823 struct btrfs_root *root = BTRFS_I(inode)->root;
2824 struct btrfs_ordered_extent *ordered_extent = NULL;
2825 struct btrfs_workqueue *workers;
2827 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2829 ClearPagePrivate2(page);
2830 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2831 end - start + 1, uptodate))
2834 btrfs_init_work(&ordered_extent->work, finish_ordered_fn, NULL, NULL);
2836 if (btrfs_is_free_space_inode(inode))
2837 workers = root->fs_info->endio_freespace_worker;
2839 workers = root->fs_info->endio_write_workers;
2840 btrfs_queue_work(workers, &ordered_extent->work);
2846 * when reads are done, we need to check csums to verify the data is correct
2847 * if there's a match, we allow the bio to finish. If not, the code in
2848 * extent_io.c will try to find good copies for us.
2850 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2851 u64 phy_offset, struct page *page,
2852 u64 start, u64 end, int mirror)
2854 size_t offset = start - page_offset(page);
2855 struct inode *inode = page->mapping->host;
2856 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2858 struct btrfs_root *root = BTRFS_I(inode)->root;
2861 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2862 DEFAULT_RATELIMIT_BURST);
2864 if (PageChecked(page)) {
2865 ClearPageChecked(page);
2869 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2872 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2873 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2874 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2879 phy_offset >>= inode->i_sb->s_blocksize_bits;
2880 csum_expected = *(((u32 *)io_bio->csum) + phy_offset);
2882 kaddr = kmap_atomic(page);
2883 csum = btrfs_csum_data(kaddr + offset, csum, end - start + 1);
2884 btrfs_csum_final(csum, (char *)&csum);
2885 if (csum != csum_expected)
2888 kunmap_atomic(kaddr);
2893 if (__ratelimit(&_rs))
2894 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
2895 btrfs_ino(page->mapping->host), start, csum, csum_expected);
2896 memset(kaddr + offset, 1, end - start + 1);
2897 flush_dcache_page(page);
2898 kunmap_atomic(kaddr);
2899 if (csum_expected == 0)
2904 struct delayed_iput {
2905 struct list_head list;
2906 struct inode *inode;
2909 /* JDM: If this is fs-wide, why can't we add a pointer to
2910 * btrfs_inode instead and avoid the allocation? */
2911 void btrfs_add_delayed_iput(struct inode *inode)
2913 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2914 struct delayed_iput *delayed;
2916 if (atomic_add_unless(&inode->i_count, -1, 1))
2919 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2920 delayed->inode = inode;
2922 spin_lock(&fs_info->delayed_iput_lock);
2923 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2924 spin_unlock(&fs_info->delayed_iput_lock);
2927 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2930 struct btrfs_fs_info *fs_info = root->fs_info;
2931 struct delayed_iput *delayed;
2934 spin_lock(&fs_info->delayed_iput_lock);
2935 empty = list_empty(&fs_info->delayed_iputs);
2936 spin_unlock(&fs_info->delayed_iput_lock);
2940 spin_lock(&fs_info->delayed_iput_lock);
2941 list_splice_init(&fs_info->delayed_iputs, &list);
2942 spin_unlock(&fs_info->delayed_iput_lock);
2944 while (!list_empty(&list)) {
2945 delayed = list_entry(list.next, struct delayed_iput, list);
2946 list_del(&delayed->list);
2947 iput(delayed->inode);
2953 * This is called in transaction commit time. If there are no orphan
2954 * files in the subvolume, it removes orphan item and frees block_rsv
2957 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2958 struct btrfs_root *root)
2960 struct btrfs_block_rsv *block_rsv;
2963 if (atomic_read(&root->orphan_inodes) ||
2964 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2967 spin_lock(&root->orphan_lock);
2968 if (atomic_read(&root->orphan_inodes)) {
2969 spin_unlock(&root->orphan_lock);
2973 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2974 spin_unlock(&root->orphan_lock);
2978 block_rsv = root->orphan_block_rsv;
2979 root->orphan_block_rsv = NULL;
2980 spin_unlock(&root->orphan_lock);
2982 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
2983 btrfs_root_refs(&root->root_item) > 0) {
2984 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2985 root->root_key.objectid);
2987 btrfs_abort_transaction(trans, root, ret);
2989 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
2994 WARN_ON(block_rsv->size > 0);
2995 btrfs_free_block_rsv(root, block_rsv);
3000 * This creates an orphan entry for the given inode in case something goes
3001 * wrong in the middle of an unlink/truncate.
3003 * NOTE: caller of this function should reserve 5 units of metadata for
3006 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3008 struct btrfs_root *root = BTRFS_I(inode)->root;
3009 struct btrfs_block_rsv *block_rsv = NULL;
3014 if (!root->orphan_block_rsv) {
3015 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3020 spin_lock(&root->orphan_lock);
3021 if (!root->orphan_block_rsv) {
3022 root->orphan_block_rsv = block_rsv;
3023 } else if (block_rsv) {
3024 btrfs_free_block_rsv(root, block_rsv);
3028 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3029 &BTRFS_I(inode)->runtime_flags)) {
3032 * For proper ENOSPC handling, we should do orphan
3033 * cleanup when mounting. But this introduces backward
3034 * compatibility issue.
3036 if (!xchg(&root->orphan_item_inserted, 1))
3042 atomic_inc(&root->orphan_inodes);
3045 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3046 &BTRFS_I(inode)->runtime_flags))
3048 spin_unlock(&root->orphan_lock);
3050 /* grab metadata reservation from transaction handle */
3052 ret = btrfs_orphan_reserve_metadata(trans, inode);
3053 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3056 /* insert an orphan item to track this unlinked/truncated file */
3058 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3060 atomic_dec(&root->orphan_inodes);
3062 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3063 &BTRFS_I(inode)->runtime_flags);
3064 btrfs_orphan_release_metadata(inode);
3066 if (ret != -EEXIST) {
3067 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3068 &BTRFS_I(inode)->runtime_flags);
3069 btrfs_abort_transaction(trans, root, ret);
3076 /* insert an orphan item to track subvolume contains orphan files */
3078 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3079 root->root_key.objectid);
3080 if (ret && ret != -EEXIST) {
3081 btrfs_abort_transaction(trans, root, ret);
3089 * We have done the truncate/delete so we can go ahead and remove the orphan
3090 * item for this particular inode.
3092 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3093 struct inode *inode)
3095 struct btrfs_root *root = BTRFS_I(inode)->root;
3096 int delete_item = 0;
3097 int release_rsv = 0;
3100 spin_lock(&root->orphan_lock);
3101 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3102 &BTRFS_I(inode)->runtime_flags))
3105 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3106 &BTRFS_I(inode)->runtime_flags))
3108 spin_unlock(&root->orphan_lock);
3111 atomic_dec(&root->orphan_inodes);
3113 ret = btrfs_del_orphan_item(trans, root,
3118 btrfs_orphan_release_metadata(inode);
3124 * this cleans up any orphans that may be left on the list from the last use
3127 int btrfs_orphan_cleanup(struct btrfs_root *root)
3129 struct btrfs_path *path;
3130 struct extent_buffer *leaf;
3131 struct btrfs_key key, found_key;
3132 struct btrfs_trans_handle *trans;
3133 struct inode *inode;
3134 u64 last_objectid = 0;
3135 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3137 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3140 path = btrfs_alloc_path();
3147 key.objectid = BTRFS_ORPHAN_OBJECTID;
3148 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3149 key.offset = (u64)-1;
3152 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3157 * if ret == 0 means we found what we were searching for, which
3158 * is weird, but possible, so only screw with path if we didn't
3159 * find the key and see if we have stuff that matches
3163 if (path->slots[0] == 0)
3168 /* pull out the item */
3169 leaf = path->nodes[0];
3170 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3172 /* make sure the item matches what we want */
3173 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3175 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3178 /* release the path since we're done with it */
3179 btrfs_release_path(path);
3182 * this is where we are basically btrfs_lookup, without the
3183 * crossing root thing. we store the inode number in the
3184 * offset of the orphan item.
3187 if (found_key.offset == last_objectid) {
3188 btrfs_err(root->fs_info,
3189 "Error removing orphan entry, stopping orphan cleanup");
3194 last_objectid = found_key.offset;
3196 found_key.objectid = found_key.offset;
3197 found_key.type = BTRFS_INODE_ITEM_KEY;
3198 found_key.offset = 0;
3199 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3200 ret = PTR_ERR_OR_ZERO(inode);
3201 if (ret && ret != -ESTALE)
3204 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3205 struct btrfs_root *dead_root;
3206 struct btrfs_fs_info *fs_info = root->fs_info;
3207 int is_dead_root = 0;
3210 * this is an orphan in the tree root. Currently these
3211 * could come from 2 sources:
3212 * a) a snapshot deletion in progress
3213 * b) a free space cache inode
3214 * We need to distinguish those two, as the snapshot
3215 * orphan must not get deleted.
3216 * find_dead_roots already ran before us, so if this
3217 * is a snapshot deletion, we should find the root
3218 * in the dead_roots list
3220 spin_lock(&fs_info->trans_lock);
3221 list_for_each_entry(dead_root, &fs_info->dead_roots,
3223 if (dead_root->root_key.objectid ==
3224 found_key.objectid) {
3229 spin_unlock(&fs_info->trans_lock);
3231 /* prevent this orphan from being found again */
3232 key.offset = found_key.objectid - 1;
3237 * Inode is already gone but the orphan item is still there,
3238 * kill the orphan item.
3240 if (ret == -ESTALE) {
3241 trans = btrfs_start_transaction(root, 1);
3242 if (IS_ERR(trans)) {
3243 ret = PTR_ERR(trans);
3246 btrfs_debug(root->fs_info, "auto deleting %Lu",
3247 found_key.objectid);
3248 ret = btrfs_del_orphan_item(trans, root,
3249 found_key.objectid);
3250 btrfs_end_transaction(trans, root);
3257 * add this inode to the orphan list so btrfs_orphan_del does
3258 * the proper thing when we hit it
3260 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3261 &BTRFS_I(inode)->runtime_flags);
3262 atomic_inc(&root->orphan_inodes);
3264 /* if we have links, this was a truncate, lets do that */
3265 if (inode->i_nlink) {
3266 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3272 /* 1 for the orphan item deletion. */
3273 trans = btrfs_start_transaction(root, 1);
3274 if (IS_ERR(trans)) {
3276 ret = PTR_ERR(trans);
3279 ret = btrfs_orphan_add(trans, inode);
3280 btrfs_end_transaction(trans, root);
3286 ret = btrfs_truncate(inode);
3288 btrfs_orphan_del(NULL, inode);
3293 /* this will do delete_inode and everything for us */
3298 /* release the path since we're done with it */
3299 btrfs_release_path(path);
3301 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3303 if (root->orphan_block_rsv)
3304 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3307 if (root->orphan_block_rsv ||
3308 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3309 trans = btrfs_join_transaction(root);
3311 btrfs_end_transaction(trans, root);
3315 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3317 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3321 btrfs_crit(root->fs_info,
3322 "could not do orphan cleanup %d", ret);
3323 btrfs_free_path(path);
3328 * very simple check to peek ahead in the leaf looking for xattrs. If we
3329 * don't find any xattrs, we know there can't be any acls.
3331 * slot is the slot the inode is in, objectid is the objectid of the inode
3333 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3334 int slot, u64 objectid,
3335 int *first_xattr_slot)
3337 u32 nritems = btrfs_header_nritems(leaf);
3338 struct btrfs_key found_key;
3339 static u64 xattr_access = 0;
3340 static u64 xattr_default = 0;
3343 if (!xattr_access) {
3344 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3345 strlen(POSIX_ACL_XATTR_ACCESS));
3346 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3347 strlen(POSIX_ACL_XATTR_DEFAULT));
3351 *first_xattr_slot = -1;
3352 while (slot < nritems) {
3353 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3355 /* we found a different objectid, there must not be acls */
3356 if (found_key.objectid != objectid)
3359 /* we found an xattr, assume we've got an acl */
3360 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3361 if (*first_xattr_slot == -1)
3362 *first_xattr_slot = slot;
3363 if (found_key.offset == xattr_access ||
3364 found_key.offset == xattr_default)
3369 * we found a key greater than an xattr key, there can't
3370 * be any acls later on
3372 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3379 * it goes inode, inode backrefs, xattrs, extents,
3380 * so if there are a ton of hard links to an inode there can
3381 * be a lot of backrefs. Don't waste time searching too hard,
3382 * this is just an optimization
3387 /* we hit the end of the leaf before we found an xattr or
3388 * something larger than an xattr. We have to assume the inode
3391 if (*first_xattr_slot == -1)
3392 *first_xattr_slot = slot;
3397 * read an inode from the btree into the in-memory inode
3399 static void btrfs_read_locked_inode(struct inode *inode)
3401 struct btrfs_path *path;
3402 struct extent_buffer *leaf;
3403 struct btrfs_inode_item *inode_item;
3404 struct btrfs_timespec *tspec;
3405 struct btrfs_root *root = BTRFS_I(inode)->root;
3406 struct btrfs_key location;
3411 bool filled = false;
3412 int first_xattr_slot;
3414 ret = btrfs_fill_inode(inode, &rdev);
3418 path = btrfs_alloc_path();
3422 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3424 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3428 leaf = path->nodes[0];
3433 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3434 struct btrfs_inode_item);
3435 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3436 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3437 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3438 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3439 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3441 tspec = btrfs_inode_atime(inode_item);
3442 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3443 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3445 tspec = btrfs_inode_mtime(inode_item);
3446 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3447 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3449 tspec = btrfs_inode_ctime(inode_item);
3450 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3451 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3453 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3454 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3455 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3458 * If we were modified in the current generation and evicted from memory
3459 * and then re-read we need to do a full sync since we don't have any
3460 * idea about which extents were modified before we were evicted from
3463 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3464 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3465 &BTRFS_I(inode)->runtime_flags);
3467 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3468 inode->i_generation = BTRFS_I(inode)->generation;
3470 rdev = btrfs_inode_rdev(leaf, inode_item);
3472 BTRFS_I(inode)->index_cnt = (u64)-1;
3473 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3477 if (inode->i_nlink != 1 ||
3478 path->slots[0] >= btrfs_header_nritems(leaf))
3481 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3482 if (location.objectid != btrfs_ino(inode))
3485 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3486 if (location.type == BTRFS_INODE_REF_KEY) {
3487 struct btrfs_inode_ref *ref;
3489 ref = (struct btrfs_inode_ref *)ptr;
3490 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3491 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3492 struct btrfs_inode_extref *extref;
3494 extref = (struct btrfs_inode_extref *)ptr;
3495 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3500 * try to precache a NULL acl entry for files that don't have
3501 * any xattrs or acls
3503 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3504 btrfs_ino(inode), &first_xattr_slot);
3505 if (first_xattr_slot != -1) {
3506 path->slots[0] = first_xattr_slot;
3507 ret = btrfs_load_inode_props(inode, path);
3509 btrfs_err(root->fs_info,
3510 "error loading props for ino %llu (root %llu): %d",
3512 root->root_key.objectid, ret);
3514 btrfs_free_path(path);
3517 cache_no_acl(inode);
3519 switch (inode->i_mode & S_IFMT) {
3521 inode->i_mapping->a_ops = &btrfs_aops;
3522 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3523 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3524 inode->i_fop = &btrfs_file_operations;
3525 inode->i_op = &btrfs_file_inode_operations;
3528 inode->i_fop = &btrfs_dir_file_operations;
3529 if (root == root->fs_info->tree_root)
3530 inode->i_op = &btrfs_dir_ro_inode_operations;
3532 inode->i_op = &btrfs_dir_inode_operations;
3535 inode->i_op = &btrfs_symlink_inode_operations;
3536 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3537 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3540 inode->i_op = &btrfs_special_inode_operations;
3541 init_special_inode(inode, inode->i_mode, rdev);
3545 btrfs_update_iflags(inode);
3549 btrfs_free_path(path);
3550 make_bad_inode(inode);
3554 * given a leaf and an inode, copy the inode fields into the leaf
3556 static void fill_inode_item(struct btrfs_trans_handle *trans,
3557 struct extent_buffer *leaf,
3558 struct btrfs_inode_item *item,
3559 struct inode *inode)
3561 struct btrfs_map_token token;
3563 btrfs_init_map_token(&token);
3565 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3566 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3567 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3569 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3570 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3572 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3573 inode->i_atime.tv_sec, &token);
3574 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3575 inode->i_atime.tv_nsec, &token);
3577 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3578 inode->i_mtime.tv_sec, &token);
3579 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3580 inode->i_mtime.tv_nsec, &token);
3582 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3583 inode->i_ctime.tv_sec, &token);
3584 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3585 inode->i_ctime.tv_nsec, &token);
3587 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3589 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3591 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3592 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3593 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3594 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3595 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3599 * copy everything in the in-memory inode into the btree.
3601 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3602 struct btrfs_root *root, struct inode *inode)
3604 struct btrfs_inode_item *inode_item;
3605 struct btrfs_path *path;
3606 struct extent_buffer *leaf;
3609 path = btrfs_alloc_path();
3613 path->leave_spinning = 1;
3614 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3622 leaf = path->nodes[0];
3623 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3624 struct btrfs_inode_item);
3626 fill_inode_item(trans, leaf, inode_item, inode);
3627 btrfs_mark_buffer_dirty(leaf);
3628 btrfs_set_inode_last_trans(trans, inode);
3631 btrfs_free_path(path);
3636 * copy everything in the in-memory inode into the btree.
3638 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3639 struct btrfs_root *root, struct inode *inode)
3644 * If the inode is a free space inode, we can deadlock during commit
3645 * if we put it into the delayed code.
3647 * The data relocation inode should also be directly updated
3650 if (!btrfs_is_free_space_inode(inode)
3651 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3652 btrfs_update_root_times(trans, root);
3654 ret = btrfs_delayed_update_inode(trans, root, inode);
3656 btrfs_set_inode_last_trans(trans, inode);
3660 return btrfs_update_inode_item(trans, root, inode);
3663 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3664 struct btrfs_root *root,
3665 struct inode *inode)
3669 ret = btrfs_update_inode(trans, root, inode);
3671 return btrfs_update_inode_item(trans, root, inode);
3676 * unlink helper that gets used here in inode.c and in the tree logging
3677 * recovery code. It remove a link in a directory with a given name, and
3678 * also drops the back refs in the inode to the directory
3680 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3681 struct btrfs_root *root,
3682 struct inode *dir, struct inode *inode,
3683 const char *name, int name_len)
3685 struct btrfs_path *path;
3687 struct extent_buffer *leaf;
3688 struct btrfs_dir_item *di;
3689 struct btrfs_key key;
3691 u64 ino = btrfs_ino(inode);
3692 u64 dir_ino = btrfs_ino(dir);
3694 path = btrfs_alloc_path();
3700 path->leave_spinning = 1;
3701 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3702 name, name_len, -1);
3711 leaf = path->nodes[0];
3712 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3713 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3716 btrfs_release_path(path);
3719 * If we don't have dir index, we have to get it by looking up
3720 * the inode ref, since we get the inode ref, remove it directly,
3721 * it is unnecessary to do delayed deletion.
3723 * But if we have dir index, needn't search inode ref to get it.
3724 * Since the inode ref is close to the inode item, it is better
3725 * that we delay to delete it, and just do this deletion when
3726 * we update the inode item.
3728 if (BTRFS_I(inode)->dir_index) {
3729 ret = btrfs_delayed_delete_inode_ref(inode);
3731 index = BTRFS_I(inode)->dir_index;
3736 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3739 btrfs_info(root->fs_info,
3740 "failed to delete reference to %.*s, inode %llu parent %llu",
3741 name_len, name, ino, dir_ino);
3742 btrfs_abort_transaction(trans, root, ret);
3746 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3748 btrfs_abort_transaction(trans, root, ret);
3752 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3754 if (ret != 0 && ret != -ENOENT) {
3755 btrfs_abort_transaction(trans, root, ret);
3759 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3764 btrfs_abort_transaction(trans, root, ret);
3766 btrfs_free_path(path);
3770 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3771 inode_inc_iversion(inode);
3772 inode_inc_iversion(dir);
3773 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3774 ret = btrfs_update_inode(trans, root, dir);
3779 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3780 struct btrfs_root *root,
3781 struct inode *dir, struct inode *inode,
3782 const char *name, int name_len)
3785 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3788 ret = btrfs_update_inode(trans, root, inode);
3794 * helper to start transaction for unlink and rmdir.
3796 * unlink and rmdir are special in btrfs, they do not always free space, so
3797 * if we cannot make our reservations the normal way try and see if there is
3798 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3799 * allow the unlink to occur.
3801 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3803 struct btrfs_trans_handle *trans;
3804 struct btrfs_root *root = BTRFS_I(dir)->root;
3808 * 1 for the possible orphan item
3809 * 1 for the dir item
3810 * 1 for the dir index
3811 * 1 for the inode ref
3814 trans = btrfs_start_transaction(root, 5);
3815 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3818 if (PTR_ERR(trans) == -ENOSPC) {
3819 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3821 trans = btrfs_start_transaction(root, 0);
3824 ret = btrfs_cond_migrate_bytes(root->fs_info,
3825 &root->fs_info->trans_block_rsv,
3828 btrfs_end_transaction(trans, root);
3829 return ERR_PTR(ret);
3831 trans->block_rsv = &root->fs_info->trans_block_rsv;
3832 trans->bytes_reserved = num_bytes;
3837 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3839 struct btrfs_root *root = BTRFS_I(dir)->root;
3840 struct btrfs_trans_handle *trans;
3841 struct inode *inode = dentry->d_inode;
3844 trans = __unlink_start_trans(dir);
3846 return PTR_ERR(trans);
3848 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3850 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3851 dentry->d_name.name, dentry->d_name.len);
3855 if (inode->i_nlink == 0) {
3856 ret = btrfs_orphan_add(trans, inode);
3862 btrfs_end_transaction(trans, root);
3863 btrfs_btree_balance_dirty(root);
3867 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3868 struct btrfs_root *root,
3869 struct inode *dir, u64 objectid,
3870 const char *name, int name_len)
3872 struct btrfs_path *path;
3873 struct extent_buffer *leaf;
3874 struct btrfs_dir_item *di;
3875 struct btrfs_key key;
3878 u64 dir_ino = btrfs_ino(dir);
3880 path = btrfs_alloc_path();
3884 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3885 name, name_len, -1);
3886 if (IS_ERR_OR_NULL(di)) {
3894 leaf = path->nodes[0];
3895 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3896 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3897 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3899 btrfs_abort_transaction(trans, root, ret);
3902 btrfs_release_path(path);
3904 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3905 objectid, root->root_key.objectid,
3906 dir_ino, &index, name, name_len);
3908 if (ret != -ENOENT) {
3909 btrfs_abort_transaction(trans, root, ret);
3912 di = btrfs_search_dir_index_item(root, path, dir_ino,
3914 if (IS_ERR_OR_NULL(di)) {
3919 btrfs_abort_transaction(trans, root, ret);
3923 leaf = path->nodes[0];
3924 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3925 btrfs_release_path(path);
3928 btrfs_release_path(path);
3930 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3932 btrfs_abort_transaction(trans, root, ret);
3936 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3937 inode_inc_iversion(dir);
3938 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3939 ret = btrfs_update_inode_fallback(trans, root, dir);
3941 btrfs_abort_transaction(trans, root, ret);
3943 btrfs_free_path(path);
3947 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3949 struct inode *inode = dentry->d_inode;
3951 struct btrfs_root *root = BTRFS_I(dir)->root;
3952 struct btrfs_trans_handle *trans;
3954 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3956 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3959 trans = __unlink_start_trans(dir);
3961 return PTR_ERR(trans);
3963 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3964 err = btrfs_unlink_subvol(trans, root, dir,
3965 BTRFS_I(inode)->location.objectid,
3966 dentry->d_name.name,
3967 dentry->d_name.len);
3971 err = btrfs_orphan_add(trans, inode);
3975 /* now the directory is empty */
3976 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3977 dentry->d_name.name, dentry->d_name.len);
3979 btrfs_i_size_write(inode, 0);
3981 btrfs_end_transaction(trans, root);
3982 btrfs_btree_balance_dirty(root);
3988 * this can truncate away extent items, csum items and directory items.
3989 * It starts at a high offset and removes keys until it can't find
3990 * any higher than new_size
3992 * csum items that cross the new i_size are truncated to the new size
3995 * min_type is the minimum key type to truncate down to. If set to 0, this
3996 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3998 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3999 struct btrfs_root *root,
4000 struct inode *inode,
4001 u64 new_size, u32 min_type)
4003 struct btrfs_path *path;
4004 struct extent_buffer *leaf;
4005 struct btrfs_file_extent_item *fi;
4006 struct btrfs_key key;
4007 struct btrfs_key found_key;
4008 u64 extent_start = 0;
4009 u64 extent_num_bytes = 0;
4010 u64 extent_offset = 0;
4012 u64 last_size = (u64)-1;
4013 u32 found_type = (u8)-1;
4016 int pending_del_nr = 0;
4017 int pending_del_slot = 0;
4018 int extent_type = -1;
4021 u64 ino = btrfs_ino(inode);
4023 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4025 path = btrfs_alloc_path();
4031 * We want to drop from the next block forward in case this new size is
4032 * not block aligned since we will be keeping the last block of the
4033 * extent just the way it is.
4035 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4036 root == root->fs_info->tree_root)
4037 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4038 root->sectorsize), (u64)-1, 0);
4041 * This function is also used to drop the items in the log tree before
4042 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4043 * it is used to drop the loged items. So we shouldn't kill the delayed
4046 if (min_type == 0 && root == BTRFS_I(inode)->root)
4047 btrfs_kill_delayed_inode_items(inode);
4050 key.offset = (u64)-1;
4054 path->leave_spinning = 1;
4055 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4062 /* there are no items in the tree for us to truncate, we're
4065 if (path->slots[0] == 0)
4072 leaf = path->nodes[0];
4073 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4074 found_type = btrfs_key_type(&found_key);
4076 if (found_key.objectid != ino)
4079 if (found_type < min_type)
4082 item_end = found_key.offset;
4083 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4084 fi = btrfs_item_ptr(leaf, path->slots[0],
4085 struct btrfs_file_extent_item);
4086 extent_type = btrfs_file_extent_type(leaf, fi);
4087 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4089 btrfs_file_extent_num_bytes(leaf, fi);
4090 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4091 item_end += btrfs_file_extent_inline_len(leaf,
4092 path->slots[0], fi);
4096 if (found_type > min_type) {
4099 if (item_end < new_size)
4101 if (found_key.offset >= new_size)
4107 /* FIXME, shrink the extent if the ref count is only 1 */
4108 if (found_type != BTRFS_EXTENT_DATA_KEY)
4112 last_size = found_key.offset;
4114 last_size = new_size;
4116 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4118 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4120 u64 orig_num_bytes =
4121 btrfs_file_extent_num_bytes(leaf, fi);
4122 extent_num_bytes = ALIGN(new_size -
4125 btrfs_set_file_extent_num_bytes(leaf, fi,
4127 num_dec = (orig_num_bytes -
4129 if (test_bit(BTRFS_ROOT_REF_COWS,
4132 inode_sub_bytes(inode, num_dec);
4133 btrfs_mark_buffer_dirty(leaf);
4136 btrfs_file_extent_disk_num_bytes(leaf,
4138 extent_offset = found_key.offset -
4139 btrfs_file_extent_offset(leaf, fi);
4141 /* FIXME blocksize != 4096 */
4142 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4143 if (extent_start != 0) {
4145 if (test_bit(BTRFS_ROOT_REF_COWS,
4147 inode_sub_bytes(inode, num_dec);
4150 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4152 * we can't truncate inline items that have had
4156 btrfs_file_extent_compression(leaf, fi) == 0 &&
4157 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4158 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4159 u32 size = new_size - found_key.offset;
4161 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4162 inode_sub_bytes(inode, item_end + 1 -
4166 * update the ram bytes to properly reflect
4167 * the new size of our item
4169 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4171 btrfs_file_extent_calc_inline_size(size);
4172 btrfs_truncate_item(root, path, size, 1);
4173 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4175 inode_sub_bytes(inode, item_end + 1 -
4181 if (!pending_del_nr) {
4182 /* no pending yet, add ourselves */
4183 pending_del_slot = path->slots[0];
4185 } else if (pending_del_nr &&
4186 path->slots[0] + 1 == pending_del_slot) {
4187 /* hop on the pending chunk */
4189 pending_del_slot = path->slots[0];
4197 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4198 root == root->fs_info->tree_root)) {
4199 btrfs_set_path_blocking(path);
4200 ret = btrfs_free_extent(trans, root, extent_start,
4201 extent_num_bytes, 0,
4202 btrfs_header_owner(leaf),
4203 ino, extent_offset, 0);
4207 if (found_type == BTRFS_INODE_ITEM_KEY)
4210 if (path->slots[0] == 0 ||
4211 path->slots[0] != pending_del_slot) {
4212 if (pending_del_nr) {
4213 ret = btrfs_del_items(trans, root, path,
4217 btrfs_abort_transaction(trans,
4223 btrfs_release_path(path);
4230 if (pending_del_nr) {
4231 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4234 btrfs_abort_transaction(trans, root, ret);
4237 if (last_size != (u64)-1)
4238 btrfs_ordered_update_i_size(inode, last_size, NULL);
4239 btrfs_free_path(path);
4244 * btrfs_truncate_page - read, zero a chunk and write a page
4245 * @inode - inode that we're zeroing
4246 * @from - the offset to start zeroing
4247 * @len - the length to zero, 0 to zero the entire range respective to the
4249 * @front - zero up to the offset instead of from the offset on
4251 * This will find the page for the "from" offset and cow the page and zero the
4252 * part we want to zero. This is used with truncate and hole punching.
4254 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4257 struct address_space *mapping = inode->i_mapping;
4258 struct btrfs_root *root = BTRFS_I(inode)->root;
4259 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4260 struct btrfs_ordered_extent *ordered;
4261 struct extent_state *cached_state = NULL;
4263 u32 blocksize = root->sectorsize;
4264 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4265 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4267 gfp_t mask = btrfs_alloc_write_mask(mapping);
4272 if ((offset & (blocksize - 1)) == 0 &&
4273 (!len || ((len & (blocksize - 1)) == 0)))
4275 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4280 page = find_or_create_page(mapping, index, mask);
4282 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4287 page_start = page_offset(page);
4288 page_end = page_start + PAGE_CACHE_SIZE - 1;
4290 if (!PageUptodate(page)) {
4291 ret = btrfs_readpage(NULL, page);
4293 if (page->mapping != mapping) {
4295 page_cache_release(page);
4298 if (!PageUptodate(page)) {
4303 wait_on_page_writeback(page);
4305 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4306 set_page_extent_mapped(page);
4308 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4310 unlock_extent_cached(io_tree, page_start, page_end,
4311 &cached_state, GFP_NOFS);
4313 page_cache_release(page);
4314 btrfs_start_ordered_extent(inode, ordered, 1);
4315 btrfs_put_ordered_extent(ordered);
4319 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4320 EXTENT_DIRTY | EXTENT_DELALLOC |
4321 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4322 0, 0, &cached_state, GFP_NOFS);
4324 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4327 unlock_extent_cached(io_tree, page_start, page_end,
4328 &cached_state, GFP_NOFS);
4332 if (offset != PAGE_CACHE_SIZE) {
4334 len = PAGE_CACHE_SIZE - offset;
4337 memset(kaddr, 0, offset);
4339 memset(kaddr + offset, 0, len);
4340 flush_dcache_page(page);
4343 ClearPageChecked(page);
4344 set_page_dirty(page);
4345 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4350 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4352 page_cache_release(page);
4357 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4358 u64 offset, u64 len)
4360 struct btrfs_trans_handle *trans;
4364 * Still need to make sure the inode looks like it's been updated so
4365 * that any holes get logged if we fsync.
4367 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4368 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4369 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4370 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4375 * 1 - for the one we're dropping
4376 * 1 - for the one we're adding
4377 * 1 - for updating the inode.
4379 trans = btrfs_start_transaction(root, 3);
4381 return PTR_ERR(trans);
4383 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4385 btrfs_abort_transaction(trans, root, ret);
4386 btrfs_end_transaction(trans, root);
4390 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4391 0, 0, len, 0, len, 0, 0, 0);
4393 btrfs_abort_transaction(trans, root, ret);
4395 btrfs_update_inode(trans, root, inode);
4396 btrfs_end_transaction(trans, root);
4401 * This function puts in dummy file extents for the area we're creating a hole
4402 * for. So if we are truncating this file to a larger size we need to insert
4403 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4404 * the range between oldsize and size
4406 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4408 struct btrfs_root *root = BTRFS_I(inode)->root;
4409 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4410 struct extent_map *em = NULL;
4411 struct extent_state *cached_state = NULL;
4412 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4413 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4414 u64 block_end = ALIGN(size, root->sectorsize);
4421 * If our size started in the middle of a page we need to zero out the
4422 * rest of the page before we expand the i_size, otherwise we could
4423 * expose stale data.
4425 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4429 if (size <= hole_start)
4433 struct btrfs_ordered_extent *ordered;
4435 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4437 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4438 block_end - hole_start);
4441 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4442 &cached_state, GFP_NOFS);
4443 btrfs_start_ordered_extent(inode, ordered, 1);
4444 btrfs_put_ordered_extent(ordered);
4447 cur_offset = hole_start;
4449 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4450 block_end - cur_offset, 0);
4456 last_byte = min(extent_map_end(em), block_end);
4457 last_byte = ALIGN(last_byte , root->sectorsize);
4458 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4459 struct extent_map *hole_em;
4460 hole_size = last_byte - cur_offset;
4462 err = maybe_insert_hole(root, inode, cur_offset,
4466 btrfs_drop_extent_cache(inode, cur_offset,
4467 cur_offset + hole_size - 1, 0);
4468 hole_em = alloc_extent_map();
4470 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4471 &BTRFS_I(inode)->runtime_flags);
4474 hole_em->start = cur_offset;
4475 hole_em->len = hole_size;
4476 hole_em->orig_start = cur_offset;
4478 hole_em->block_start = EXTENT_MAP_HOLE;
4479 hole_em->block_len = 0;
4480 hole_em->orig_block_len = 0;
4481 hole_em->ram_bytes = hole_size;
4482 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4483 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4484 hole_em->generation = root->fs_info->generation;
4487 write_lock(&em_tree->lock);
4488 err = add_extent_mapping(em_tree, hole_em, 1);
4489 write_unlock(&em_tree->lock);
4492 btrfs_drop_extent_cache(inode, cur_offset,
4496 free_extent_map(hole_em);
4499 free_extent_map(em);
4501 cur_offset = last_byte;
4502 if (cur_offset >= block_end)
4505 free_extent_map(em);
4506 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4511 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4513 struct btrfs_root *root = BTRFS_I(inode)->root;
4514 struct btrfs_trans_handle *trans;
4515 loff_t oldsize = i_size_read(inode);
4516 loff_t newsize = attr->ia_size;
4517 int mask = attr->ia_valid;
4521 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4522 * special case where we need to update the times despite not having
4523 * these flags set. For all other operations the VFS set these flags
4524 * explicitly if it wants a timestamp update.
4526 if (newsize != oldsize) {
4527 inode_inc_iversion(inode);
4528 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4529 inode->i_ctime = inode->i_mtime =
4530 current_fs_time(inode->i_sb);
4533 if (newsize > oldsize) {
4534 truncate_pagecache(inode, newsize);
4535 ret = btrfs_cont_expand(inode, oldsize, newsize);
4539 trans = btrfs_start_transaction(root, 1);
4541 return PTR_ERR(trans);
4543 i_size_write(inode, newsize);
4544 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4545 ret = btrfs_update_inode(trans, root, inode);
4546 btrfs_end_transaction(trans, root);
4550 * We're truncating a file that used to have good data down to
4551 * zero. Make sure it gets into the ordered flush list so that
4552 * any new writes get down to disk quickly.
4555 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4556 &BTRFS_I(inode)->runtime_flags);
4559 * 1 for the orphan item we're going to add
4560 * 1 for the orphan item deletion.
4562 trans = btrfs_start_transaction(root, 2);
4564 return PTR_ERR(trans);
4567 * We need to do this in case we fail at _any_ point during the
4568 * actual truncate. Once we do the truncate_setsize we could
4569 * invalidate pages which forces any outstanding ordered io to
4570 * be instantly completed which will give us extents that need
4571 * to be truncated. If we fail to get an orphan inode down we
4572 * could have left over extents that were never meant to live,
4573 * so we need to garuntee from this point on that everything
4574 * will be consistent.
4576 ret = btrfs_orphan_add(trans, inode);
4577 btrfs_end_transaction(trans, root);
4581 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4582 truncate_setsize(inode, newsize);
4584 /* Disable nonlocked read DIO to avoid the end less truncate */
4585 btrfs_inode_block_unlocked_dio(inode);
4586 inode_dio_wait(inode);
4587 btrfs_inode_resume_unlocked_dio(inode);
4589 ret = btrfs_truncate(inode);
4590 if (ret && inode->i_nlink) {
4594 * failed to truncate, disk_i_size is only adjusted down
4595 * as we remove extents, so it should represent the true
4596 * size of the inode, so reset the in memory size and
4597 * delete our orphan entry.
4599 trans = btrfs_join_transaction(root);
4600 if (IS_ERR(trans)) {
4601 btrfs_orphan_del(NULL, inode);
4604 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4605 err = btrfs_orphan_del(trans, inode);
4607 btrfs_abort_transaction(trans, root, err);
4608 btrfs_end_transaction(trans, root);
4615 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4617 struct inode *inode = dentry->d_inode;
4618 struct btrfs_root *root = BTRFS_I(inode)->root;
4621 if (btrfs_root_readonly(root))
4624 err = inode_change_ok(inode, attr);
4628 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4629 err = btrfs_setsize(inode, attr);
4634 if (attr->ia_valid) {
4635 setattr_copy(inode, attr);
4636 inode_inc_iversion(inode);
4637 err = btrfs_dirty_inode(inode);
4639 if (!err && attr->ia_valid & ATTR_MODE)
4640 err = posix_acl_chmod(inode, inode->i_mode);
4647 * While truncating the inode pages during eviction, we get the VFS calling
4648 * btrfs_invalidatepage() against each page of the inode. This is slow because
4649 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4650 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4651 * extent_state structures over and over, wasting lots of time.
4653 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4654 * those expensive operations on a per page basis and do only the ordered io
4655 * finishing, while we release here the extent_map and extent_state structures,
4656 * without the excessive merging and splitting.
4658 static void evict_inode_truncate_pages(struct inode *inode)
4660 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4661 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4662 struct rb_node *node;
4664 ASSERT(inode->i_state & I_FREEING);
4665 truncate_inode_pages_final(&inode->i_data);
4667 write_lock(&map_tree->lock);
4668 while (!RB_EMPTY_ROOT(&map_tree->map)) {
4669 struct extent_map *em;
4671 node = rb_first(&map_tree->map);
4672 em = rb_entry(node, struct extent_map, rb_node);
4673 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4674 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4675 remove_extent_mapping(map_tree, em);
4676 free_extent_map(em);
4677 if (need_resched()) {
4678 write_unlock(&map_tree->lock);
4680 write_lock(&map_tree->lock);
4683 write_unlock(&map_tree->lock);
4685 spin_lock(&io_tree->lock);
4686 while (!RB_EMPTY_ROOT(&io_tree->state)) {
4687 struct extent_state *state;
4688 struct extent_state *cached_state = NULL;
4690 node = rb_first(&io_tree->state);
4691 state = rb_entry(node, struct extent_state, rb_node);
4692 atomic_inc(&state->refs);
4693 spin_unlock(&io_tree->lock);
4695 lock_extent_bits(io_tree, state->start, state->end,
4697 clear_extent_bit(io_tree, state->start, state->end,
4698 EXTENT_LOCKED | EXTENT_DIRTY |
4699 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
4700 EXTENT_DEFRAG, 1, 1,
4701 &cached_state, GFP_NOFS);
4702 free_extent_state(state);
4705 spin_lock(&io_tree->lock);
4707 spin_unlock(&io_tree->lock);
4710 void btrfs_evict_inode(struct inode *inode)
4712 struct btrfs_trans_handle *trans;
4713 struct btrfs_root *root = BTRFS_I(inode)->root;
4714 struct btrfs_block_rsv *rsv, *global_rsv;
4715 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4718 trace_btrfs_inode_evict(inode);
4720 evict_inode_truncate_pages(inode);
4722 if (inode->i_nlink &&
4723 ((btrfs_root_refs(&root->root_item) != 0 &&
4724 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4725 btrfs_is_free_space_inode(inode)))
4728 if (is_bad_inode(inode)) {
4729 btrfs_orphan_del(NULL, inode);
4732 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4733 btrfs_wait_ordered_range(inode, 0, (u64)-1);
4735 if (root->fs_info->log_root_recovering) {
4736 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4737 &BTRFS_I(inode)->runtime_flags));
4741 if (inode->i_nlink > 0) {
4742 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4743 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
4747 ret = btrfs_commit_inode_delayed_inode(inode);
4749 btrfs_orphan_del(NULL, inode);
4753 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4755 btrfs_orphan_del(NULL, inode);
4758 rsv->size = min_size;
4760 global_rsv = &root->fs_info->global_block_rsv;
4762 btrfs_i_size_write(inode, 0);
4765 * This is a bit simpler than btrfs_truncate since we've already
4766 * reserved our space for our orphan item in the unlink, so we just
4767 * need to reserve some slack space in case we add bytes and update
4768 * inode item when doing the truncate.
4771 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4772 BTRFS_RESERVE_FLUSH_LIMIT);
4775 * Try and steal from the global reserve since we will
4776 * likely not use this space anyway, we want to try as
4777 * hard as possible to get this to work.
4780 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4783 btrfs_warn(root->fs_info,
4784 "Could not get space for a delete, will truncate on mount %d",
4786 btrfs_orphan_del(NULL, inode);
4787 btrfs_free_block_rsv(root, rsv);
4791 trans = btrfs_join_transaction(root);
4792 if (IS_ERR(trans)) {
4793 btrfs_orphan_del(NULL, inode);
4794 btrfs_free_block_rsv(root, rsv);
4798 trans->block_rsv = rsv;
4800 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4804 trans->block_rsv = &root->fs_info->trans_block_rsv;
4805 btrfs_end_transaction(trans, root);
4807 btrfs_btree_balance_dirty(root);
4810 btrfs_free_block_rsv(root, rsv);
4813 * Errors here aren't a big deal, it just means we leave orphan items
4814 * in the tree. They will be cleaned up on the next mount.
4817 trans->block_rsv = root->orphan_block_rsv;
4818 btrfs_orphan_del(trans, inode);
4820 btrfs_orphan_del(NULL, inode);
4823 trans->block_rsv = &root->fs_info->trans_block_rsv;
4824 if (!(root == root->fs_info->tree_root ||
4825 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4826 btrfs_return_ino(root, btrfs_ino(inode));
4828 btrfs_end_transaction(trans, root);
4829 btrfs_btree_balance_dirty(root);
4831 btrfs_remove_delayed_node(inode);
4837 * this returns the key found in the dir entry in the location pointer.
4838 * If no dir entries were found, location->objectid is 0.
4840 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4841 struct btrfs_key *location)
4843 const char *name = dentry->d_name.name;
4844 int namelen = dentry->d_name.len;
4845 struct btrfs_dir_item *di;
4846 struct btrfs_path *path;
4847 struct btrfs_root *root = BTRFS_I(dir)->root;
4850 path = btrfs_alloc_path();
4854 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4859 if (IS_ERR_OR_NULL(di))
4862 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4864 btrfs_free_path(path);
4867 location->objectid = 0;
4872 * when we hit a tree root in a directory, the btrfs part of the inode
4873 * needs to be changed to reflect the root directory of the tree root. This
4874 * is kind of like crossing a mount point.
4876 static int fixup_tree_root_location(struct btrfs_root *root,
4878 struct dentry *dentry,
4879 struct btrfs_key *location,
4880 struct btrfs_root **sub_root)
4882 struct btrfs_path *path;
4883 struct btrfs_root *new_root;
4884 struct btrfs_root_ref *ref;
4885 struct extent_buffer *leaf;
4889 path = btrfs_alloc_path();
4896 ret = btrfs_find_item(root->fs_info->tree_root, path,
4897 BTRFS_I(dir)->root->root_key.objectid,
4898 location->objectid, BTRFS_ROOT_REF_KEY, NULL);
4905 leaf = path->nodes[0];
4906 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4907 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4908 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4911 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4912 (unsigned long)(ref + 1),
4913 dentry->d_name.len);
4917 btrfs_release_path(path);
4919 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4920 if (IS_ERR(new_root)) {
4921 err = PTR_ERR(new_root);
4925 *sub_root = new_root;
4926 location->objectid = btrfs_root_dirid(&new_root->root_item);
4927 location->type = BTRFS_INODE_ITEM_KEY;
4928 location->offset = 0;
4931 btrfs_free_path(path);
4935 static void inode_tree_add(struct inode *inode)
4937 struct btrfs_root *root = BTRFS_I(inode)->root;
4938 struct btrfs_inode *entry;
4940 struct rb_node *parent;
4941 struct rb_node *new = &BTRFS_I(inode)->rb_node;
4942 u64 ino = btrfs_ino(inode);
4944 if (inode_unhashed(inode))
4947 spin_lock(&root->inode_lock);
4948 p = &root->inode_tree.rb_node;
4951 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4953 if (ino < btrfs_ino(&entry->vfs_inode))
4954 p = &parent->rb_left;
4955 else if (ino > btrfs_ino(&entry->vfs_inode))
4956 p = &parent->rb_right;
4958 WARN_ON(!(entry->vfs_inode.i_state &
4959 (I_WILL_FREE | I_FREEING)));
4960 rb_replace_node(parent, new, &root->inode_tree);
4961 RB_CLEAR_NODE(parent);
4962 spin_unlock(&root->inode_lock);
4966 rb_link_node(new, parent, p);
4967 rb_insert_color(new, &root->inode_tree);
4968 spin_unlock(&root->inode_lock);
4971 static void inode_tree_del(struct inode *inode)
4973 struct btrfs_root *root = BTRFS_I(inode)->root;
4976 spin_lock(&root->inode_lock);
4977 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4978 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4979 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4980 empty = RB_EMPTY_ROOT(&root->inode_tree);
4982 spin_unlock(&root->inode_lock);
4984 if (empty && btrfs_root_refs(&root->root_item) == 0) {
4985 synchronize_srcu(&root->fs_info->subvol_srcu);
4986 spin_lock(&root->inode_lock);
4987 empty = RB_EMPTY_ROOT(&root->inode_tree);
4988 spin_unlock(&root->inode_lock);
4990 btrfs_add_dead_root(root);
4994 void btrfs_invalidate_inodes(struct btrfs_root *root)
4996 struct rb_node *node;
4997 struct rb_node *prev;
4998 struct btrfs_inode *entry;
4999 struct inode *inode;
5002 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5003 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5005 spin_lock(&root->inode_lock);
5007 node = root->inode_tree.rb_node;
5011 entry = rb_entry(node, struct btrfs_inode, rb_node);
5013 if (objectid < btrfs_ino(&entry->vfs_inode))
5014 node = node->rb_left;
5015 else if (objectid > btrfs_ino(&entry->vfs_inode))
5016 node = node->rb_right;
5022 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5023 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5027 prev = rb_next(prev);
5031 entry = rb_entry(node, struct btrfs_inode, rb_node);
5032 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5033 inode = igrab(&entry->vfs_inode);
5035 spin_unlock(&root->inode_lock);
5036 if (atomic_read(&inode->i_count) > 1)
5037 d_prune_aliases(inode);
5039 * btrfs_drop_inode will have it removed from
5040 * the inode cache when its usage count
5045 spin_lock(&root->inode_lock);
5049 if (cond_resched_lock(&root->inode_lock))
5052 node = rb_next(node);
5054 spin_unlock(&root->inode_lock);
5057 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5059 struct btrfs_iget_args *args = p;
5060 inode->i_ino = args->location->objectid;
5061 memcpy(&BTRFS_I(inode)->location, args->location,
5062 sizeof(*args->location));
5063 BTRFS_I(inode)->root = args->root;
5067 static int btrfs_find_actor(struct inode *inode, void *opaque)
5069 struct btrfs_iget_args *args = opaque;
5070 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5071 args->root == BTRFS_I(inode)->root;
5074 static struct inode *btrfs_iget_locked(struct super_block *s,
5075 struct btrfs_key *location,
5076 struct btrfs_root *root)
5078 struct inode *inode;
5079 struct btrfs_iget_args args;
5080 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5082 args.location = location;
5085 inode = iget5_locked(s, hashval, btrfs_find_actor,
5086 btrfs_init_locked_inode,
5091 /* Get an inode object given its location and corresponding root.
5092 * Returns in *is_new if the inode was read from disk
5094 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5095 struct btrfs_root *root, int *new)
5097 struct inode *inode;
5099 inode = btrfs_iget_locked(s, location, root);
5101 return ERR_PTR(-ENOMEM);
5103 if (inode->i_state & I_NEW) {
5104 btrfs_read_locked_inode(inode);
5105 if (!is_bad_inode(inode)) {
5106 inode_tree_add(inode);
5107 unlock_new_inode(inode);
5111 unlock_new_inode(inode);
5113 inode = ERR_PTR(-ESTALE);
5120 static struct inode *new_simple_dir(struct super_block *s,
5121 struct btrfs_key *key,
5122 struct btrfs_root *root)
5124 struct inode *inode = new_inode(s);
5127 return ERR_PTR(-ENOMEM);
5129 BTRFS_I(inode)->root = root;
5130 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5131 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5133 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5134 inode->i_op = &btrfs_dir_ro_inode_operations;
5135 inode->i_fop = &simple_dir_operations;
5136 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5137 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5142 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5144 struct inode *inode;
5145 struct btrfs_root *root = BTRFS_I(dir)->root;
5146 struct btrfs_root *sub_root = root;
5147 struct btrfs_key location;
5151 if (dentry->d_name.len > BTRFS_NAME_LEN)
5152 return ERR_PTR(-ENAMETOOLONG);
5154 ret = btrfs_inode_by_name(dir, dentry, &location);
5156 return ERR_PTR(ret);
5158 if (location.objectid == 0)
5159 return ERR_PTR(-ENOENT);
5161 if (location.type == BTRFS_INODE_ITEM_KEY) {
5162 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5166 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5168 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5169 ret = fixup_tree_root_location(root, dir, dentry,
5170 &location, &sub_root);
5173 inode = ERR_PTR(ret);
5175 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5177 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5179 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5181 if (!IS_ERR(inode) && root != sub_root) {
5182 down_read(&root->fs_info->cleanup_work_sem);
5183 if (!(inode->i_sb->s_flags & MS_RDONLY))
5184 ret = btrfs_orphan_cleanup(sub_root);
5185 up_read(&root->fs_info->cleanup_work_sem);
5188 inode = ERR_PTR(ret);
5191 * If orphan cleanup did remove any orphans, it means the tree
5192 * was modified and therefore the commit root is not the same as
5193 * the current root anymore. This is a problem, because send
5194 * uses the commit root and therefore can see inode items that
5195 * don't exist in the current root anymore, and for example make
5196 * calls to btrfs_iget, which will do tree lookups based on the
5197 * current root and not on the commit root. Those lookups will
5198 * fail, returning a -ESTALE error, and making send fail with
5199 * that error. So make sure a send does not see any orphans we
5200 * have just removed, and that it will see the same inodes
5201 * regardless of whether a transaction commit happened before
5202 * it started (meaning that the commit root will be the same as
5203 * the current root) or not.
5205 if (sub_root->node != sub_root->commit_root) {
5206 u64 sub_flags = btrfs_root_flags(&sub_root->root_item);
5208 if (sub_flags & BTRFS_ROOT_SUBVOL_RDONLY) {
5209 struct extent_buffer *eb;
5212 * Assert we can't have races between dentry
5213 * lookup called through the snapshot creation
5214 * ioctl and the VFS.
5216 ASSERT(mutex_is_locked(&dir->i_mutex));
5218 down_write(&root->fs_info->commit_root_sem);
5219 eb = sub_root->commit_root;
5220 sub_root->commit_root =
5221 btrfs_root_node(sub_root);
5222 up_write(&root->fs_info->commit_root_sem);
5223 free_extent_buffer(eb);
5231 static int btrfs_dentry_delete(const struct dentry *dentry)
5233 struct btrfs_root *root;
5234 struct inode *inode = dentry->d_inode;
5236 if (!inode && !IS_ROOT(dentry))
5237 inode = dentry->d_parent->d_inode;
5240 root = BTRFS_I(inode)->root;
5241 if (btrfs_root_refs(&root->root_item) == 0)
5244 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5250 static void btrfs_dentry_release(struct dentry *dentry)
5252 kfree(dentry->d_fsdata);
5255 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5258 struct inode *inode;
5260 inode = btrfs_lookup_dentry(dir, dentry);
5261 if (IS_ERR(inode)) {
5262 if (PTR_ERR(inode) == -ENOENT)
5265 return ERR_CAST(inode);
5268 return d_materialise_unique(dentry, inode);
5271 unsigned char btrfs_filetype_table[] = {
5272 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5275 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5277 struct inode *inode = file_inode(file);
5278 struct btrfs_root *root = BTRFS_I(inode)->root;
5279 struct btrfs_item *item;
5280 struct btrfs_dir_item *di;
5281 struct btrfs_key key;
5282 struct btrfs_key found_key;
5283 struct btrfs_path *path;
5284 struct list_head ins_list;
5285 struct list_head del_list;
5287 struct extent_buffer *leaf;
5289 unsigned char d_type;
5294 int key_type = BTRFS_DIR_INDEX_KEY;
5298 int is_curr = 0; /* ctx->pos points to the current index? */
5300 /* FIXME, use a real flag for deciding about the key type */
5301 if (root->fs_info->tree_root == root)
5302 key_type = BTRFS_DIR_ITEM_KEY;
5304 if (!dir_emit_dots(file, ctx))
5307 path = btrfs_alloc_path();
5313 if (key_type == BTRFS_DIR_INDEX_KEY) {
5314 INIT_LIST_HEAD(&ins_list);
5315 INIT_LIST_HEAD(&del_list);
5316 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5319 btrfs_set_key_type(&key, key_type);
5320 key.offset = ctx->pos;
5321 key.objectid = btrfs_ino(inode);
5323 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5328 leaf = path->nodes[0];
5329 slot = path->slots[0];
5330 if (slot >= btrfs_header_nritems(leaf)) {
5331 ret = btrfs_next_leaf(root, path);
5339 item = btrfs_item_nr(slot);
5340 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5342 if (found_key.objectid != key.objectid)
5344 if (btrfs_key_type(&found_key) != key_type)
5346 if (found_key.offset < ctx->pos)
5348 if (key_type == BTRFS_DIR_INDEX_KEY &&
5349 btrfs_should_delete_dir_index(&del_list,
5353 ctx->pos = found_key.offset;
5356 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5358 di_total = btrfs_item_size(leaf, item);
5360 while (di_cur < di_total) {
5361 struct btrfs_key location;
5363 if (verify_dir_item(root, leaf, di))
5366 name_len = btrfs_dir_name_len(leaf, di);
5367 if (name_len <= sizeof(tmp_name)) {
5368 name_ptr = tmp_name;
5370 name_ptr = kmalloc(name_len, GFP_NOFS);
5376 read_extent_buffer(leaf, name_ptr,
5377 (unsigned long)(di + 1), name_len);
5379 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5380 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5383 /* is this a reference to our own snapshot? If so
5386 * In contrast to old kernels, we insert the snapshot's
5387 * dir item and dir index after it has been created, so
5388 * we won't find a reference to our own snapshot. We
5389 * still keep the following code for backward
5392 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5393 location.objectid == root->root_key.objectid) {
5397 over = !dir_emit(ctx, name_ptr, name_len,
5398 location.objectid, d_type);
5401 if (name_ptr != tmp_name)
5406 di_len = btrfs_dir_name_len(leaf, di) +
5407 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5409 di = (struct btrfs_dir_item *)((char *)di + di_len);
5415 if (key_type == BTRFS_DIR_INDEX_KEY) {
5418 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5423 /* Reached end of directory/root. Bump pos past the last item. */
5427 * Stop new entries from being returned after we return the last
5430 * New directory entries are assigned a strictly increasing
5431 * offset. This means that new entries created during readdir
5432 * are *guaranteed* to be seen in the future by that readdir.
5433 * This has broken buggy programs which operate on names as
5434 * they're returned by readdir. Until we re-use freed offsets
5435 * we have this hack to stop new entries from being returned
5436 * under the assumption that they'll never reach this huge
5439 * This is being careful not to overflow 32bit loff_t unless the
5440 * last entry requires it because doing so has broken 32bit apps
5443 if (key_type == BTRFS_DIR_INDEX_KEY) {
5444 if (ctx->pos >= INT_MAX)
5445 ctx->pos = LLONG_MAX;
5452 if (key_type == BTRFS_DIR_INDEX_KEY)
5453 btrfs_put_delayed_items(&ins_list, &del_list);
5454 btrfs_free_path(path);
5458 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5460 struct btrfs_root *root = BTRFS_I(inode)->root;
5461 struct btrfs_trans_handle *trans;
5463 bool nolock = false;
5465 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5468 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5471 if (wbc->sync_mode == WB_SYNC_ALL) {
5473 trans = btrfs_join_transaction_nolock(root);
5475 trans = btrfs_join_transaction(root);
5477 return PTR_ERR(trans);
5478 ret = btrfs_commit_transaction(trans, root);
5484 * This is somewhat expensive, updating the tree every time the
5485 * inode changes. But, it is most likely to find the inode in cache.
5486 * FIXME, needs more benchmarking...there are no reasons other than performance
5487 * to keep or drop this code.
5489 static int btrfs_dirty_inode(struct inode *inode)
5491 struct btrfs_root *root = BTRFS_I(inode)->root;
5492 struct btrfs_trans_handle *trans;
5495 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5498 trans = btrfs_join_transaction(root);
5500 return PTR_ERR(trans);
5502 ret = btrfs_update_inode(trans, root, inode);
5503 if (ret && ret == -ENOSPC) {
5504 /* whoops, lets try again with the full transaction */
5505 btrfs_end_transaction(trans, root);
5506 trans = btrfs_start_transaction(root, 1);
5508 return PTR_ERR(trans);
5510 ret = btrfs_update_inode(trans, root, inode);
5512 btrfs_end_transaction(trans, root);
5513 if (BTRFS_I(inode)->delayed_node)
5514 btrfs_balance_delayed_items(root);
5520 * This is a copy of file_update_time. We need this so we can return error on
5521 * ENOSPC for updating the inode in the case of file write and mmap writes.
5523 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5526 struct btrfs_root *root = BTRFS_I(inode)->root;
5528 if (btrfs_root_readonly(root))
5531 if (flags & S_VERSION)
5532 inode_inc_iversion(inode);
5533 if (flags & S_CTIME)
5534 inode->i_ctime = *now;
5535 if (flags & S_MTIME)
5536 inode->i_mtime = *now;
5537 if (flags & S_ATIME)
5538 inode->i_atime = *now;
5539 return btrfs_dirty_inode(inode);
5543 * find the highest existing sequence number in a directory
5544 * and then set the in-memory index_cnt variable to reflect
5545 * free sequence numbers
5547 static int btrfs_set_inode_index_count(struct inode *inode)
5549 struct btrfs_root *root = BTRFS_I(inode)->root;
5550 struct btrfs_key key, found_key;
5551 struct btrfs_path *path;
5552 struct extent_buffer *leaf;
5555 key.objectid = btrfs_ino(inode);
5556 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5557 key.offset = (u64)-1;
5559 path = btrfs_alloc_path();
5563 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5566 /* FIXME: we should be able to handle this */
5572 * MAGIC NUMBER EXPLANATION:
5573 * since we search a directory based on f_pos we have to start at 2
5574 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5575 * else has to start at 2
5577 if (path->slots[0] == 0) {
5578 BTRFS_I(inode)->index_cnt = 2;
5584 leaf = path->nodes[0];
5585 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5587 if (found_key.objectid != btrfs_ino(inode) ||
5588 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5589 BTRFS_I(inode)->index_cnt = 2;
5593 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5595 btrfs_free_path(path);
5600 * helper to find a free sequence number in a given directory. This current
5601 * code is very simple, later versions will do smarter things in the btree
5603 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5607 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5608 ret = btrfs_inode_delayed_dir_index_count(dir);
5610 ret = btrfs_set_inode_index_count(dir);
5616 *index = BTRFS_I(dir)->index_cnt;
5617 BTRFS_I(dir)->index_cnt++;
5622 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5623 struct btrfs_root *root,
5625 const char *name, int name_len,
5626 u64 ref_objectid, u64 objectid,
5627 umode_t mode, u64 *index)
5629 struct inode *inode;
5630 struct btrfs_inode_item *inode_item;
5631 struct btrfs_key *location;
5632 struct btrfs_path *path;
5633 struct btrfs_inode_ref *ref;
5634 struct btrfs_key key[2];
5636 int nitems = name ? 2 : 1;
5640 path = btrfs_alloc_path();
5642 return ERR_PTR(-ENOMEM);
5644 inode = new_inode(root->fs_info->sb);
5646 btrfs_free_path(path);
5647 return ERR_PTR(-ENOMEM);
5651 * O_TMPFILE, set link count to 0, so that after this point,
5652 * we fill in an inode item with the correct link count.
5655 set_nlink(inode, 0);
5658 * we have to initialize this early, so we can reclaim the inode
5659 * number if we fail afterwards in this function.
5661 inode->i_ino = objectid;
5664 trace_btrfs_inode_request(dir);
5666 ret = btrfs_set_inode_index(dir, index);
5668 btrfs_free_path(path);
5670 return ERR_PTR(ret);
5676 * index_cnt is ignored for everything but a dir,
5677 * btrfs_get_inode_index_count has an explanation for the magic
5680 BTRFS_I(inode)->index_cnt = 2;
5681 BTRFS_I(inode)->dir_index = *index;
5682 BTRFS_I(inode)->root = root;
5683 BTRFS_I(inode)->generation = trans->transid;
5684 inode->i_generation = BTRFS_I(inode)->generation;
5687 * We could have gotten an inode number from somebody who was fsynced
5688 * and then removed in this same transaction, so let's just set full
5689 * sync since it will be a full sync anyway and this will blow away the
5690 * old info in the log.
5692 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5694 key[0].objectid = objectid;
5695 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5698 sizes[0] = sizeof(struct btrfs_inode_item);
5702 * Start new inodes with an inode_ref. This is slightly more
5703 * efficient for small numbers of hard links since they will
5704 * be packed into one item. Extended refs will kick in if we
5705 * add more hard links than can fit in the ref item.
5707 key[1].objectid = objectid;
5708 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5709 key[1].offset = ref_objectid;
5711 sizes[1] = name_len + sizeof(*ref);
5714 path->leave_spinning = 1;
5715 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
5719 inode_init_owner(inode, dir, mode);
5720 inode_set_bytes(inode, 0);
5721 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5722 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5723 struct btrfs_inode_item);
5724 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5725 sizeof(*inode_item));
5726 fill_inode_item(trans, path->nodes[0], inode_item, inode);
5729 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5730 struct btrfs_inode_ref);
5731 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5732 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5733 ptr = (unsigned long)(ref + 1);
5734 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5737 btrfs_mark_buffer_dirty(path->nodes[0]);
5738 btrfs_free_path(path);
5740 location = &BTRFS_I(inode)->location;
5741 location->objectid = objectid;
5742 location->offset = 0;
5743 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5745 btrfs_inherit_iflags(inode, dir);
5747 if (S_ISREG(mode)) {
5748 if (btrfs_test_opt(root, NODATASUM))
5749 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5750 if (btrfs_test_opt(root, NODATACOW))
5751 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5752 BTRFS_INODE_NODATASUM;
5755 btrfs_insert_inode_hash(inode);
5756 inode_tree_add(inode);
5758 trace_btrfs_inode_new(inode);
5759 btrfs_set_inode_last_trans(trans, inode);
5761 btrfs_update_root_times(trans, root);
5763 ret = btrfs_inode_inherit_props(trans, inode, dir);
5765 btrfs_err(root->fs_info,
5766 "error inheriting props for ino %llu (root %llu): %d",
5767 btrfs_ino(inode), root->root_key.objectid, ret);
5772 BTRFS_I(dir)->index_cnt--;
5773 btrfs_free_path(path);
5775 return ERR_PTR(ret);
5778 static inline u8 btrfs_inode_type(struct inode *inode)
5780 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5784 * utility function to add 'inode' into 'parent_inode' with
5785 * a give name and a given sequence number.
5786 * if 'add_backref' is true, also insert a backref from the
5787 * inode to the parent directory.
5789 int btrfs_add_link(struct btrfs_trans_handle *trans,
5790 struct inode *parent_inode, struct inode *inode,
5791 const char *name, int name_len, int add_backref, u64 index)
5794 struct btrfs_key key;
5795 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5796 u64 ino = btrfs_ino(inode);
5797 u64 parent_ino = btrfs_ino(parent_inode);
5799 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5800 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5803 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5807 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5808 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5809 key.objectid, root->root_key.objectid,
5810 parent_ino, index, name, name_len);
5811 } else if (add_backref) {
5812 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5816 /* Nothing to clean up yet */
5820 ret = btrfs_insert_dir_item(trans, root, name, name_len,
5822 btrfs_inode_type(inode), index);
5823 if (ret == -EEXIST || ret == -EOVERFLOW)
5826 btrfs_abort_transaction(trans, root, ret);
5830 btrfs_i_size_write(parent_inode, parent_inode->i_size +
5832 inode_inc_iversion(parent_inode);
5833 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5834 ret = btrfs_update_inode(trans, root, parent_inode);
5836 btrfs_abort_transaction(trans, root, ret);
5840 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5843 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5844 key.objectid, root->root_key.objectid,
5845 parent_ino, &local_index, name, name_len);
5847 } else if (add_backref) {
5851 err = btrfs_del_inode_ref(trans, root, name, name_len,
5852 ino, parent_ino, &local_index);
5857 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5858 struct inode *dir, struct dentry *dentry,
5859 struct inode *inode, int backref, u64 index)
5861 int err = btrfs_add_link(trans, dir, inode,
5862 dentry->d_name.name, dentry->d_name.len,
5869 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5870 umode_t mode, dev_t rdev)
5872 struct btrfs_trans_handle *trans;
5873 struct btrfs_root *root = BTRFS_I(dir)->root;
5874 struct inode *inode = NULL;
5880 if (!new_valid_dev(rdev))
5884 * 2 for inode item and ref
5886 * 1 for xattr if selinux is on
5888 trans = btrfs_start_transaction(root, 5);
5890 return PTR_ERR(trans);
5892 err = btrfs_find_free_ino(root, &objectid);
5896 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5897 dentry->d_name.len, btrfs_ino(dir), objectid,
5899 if (IS_ERR(inode)) {
5900 err = PTR_ERR(inode);
5904 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5911 * If the active LSM wants to access the inode during
5912 * d_instantiate it needs these. Smack checks to see
5913 * if the filesystem supports xattrs by looking at the
5917 inode->i_op = &btrfs_special_inode_operations;
5918 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5922 init_special_inode(inode, inode->i_mode, rdev);
5923 btrfs_update_inode(trans, root, inode);
5924 d_instantiate(dentry, inode);
5927 btrfs_end_transaction(trans, root);
5928 btrfs_balance_delayed_items(root);
5929 btrfs_btree_balance_dirty(root);
5931 inode_dec_link_count(inode);
5937 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5938 umode_t mode, bool excl)
5940 struct btrfs_trans_handle *trans;
5941 struct btrfs_root *root = BTRFS_I(dir)->root;
5942 struct inode *inode = NULL;
5943 int drop_inode_on_err = 0;
5949 * 2 for inode item and ref
5951 * 1 for xattr if selinux is on
5953 trans = btrfs_start_transaction(root, 5);
5955 return PTR_ERR(trans);
5957 err = btrfs_find_free_ino(root, &objectid);
5961 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5962 dentry->d_name.len, btrfs_ino(dir), objectid,
5964 if (IS_ERR(inode)) {
5965 err = PTR_ERR(inode);
5968 drop_inode_on_err = 1;
5970 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5974 err = btrfs_update_inode(trans, root, inode);
5979 * If the active LSM wants to access the inode during
5980 * d_instantiate it needs these. Smack checks to see
5981 * if the filesystem supports xattrs by looking at the
5984 inode->i_fop = &btrfs_file_operations;
5985 inode->i_op = &btrfs_file_inode_operations;
5987 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5991 inode->i_mapping->a_ops = &btrfs_aops;
5992 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5993 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5994 d_instantiate(dentry, inode);
5997 btrfs_end_transaction(trans, root);
5998 if (err && drop_inode_on_err) {
5999 inode_dec_link_count(inode);
6002 btrfs_balance_delayed_items(root);
6003 btrfs_btree_balance_dirty(root);
6007 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6008 struct dentry *dentry)
6010 struct btrfs_trans_handle *trans;
6011 struct btrfs_root *root = BTRFS_I(dir)->root;
6012 struct inode *inode = old_dentry->d_inode;
6017 /* do not allow sys_link's with other subvols of the same device */
6018 if (root->objectid != BTRFS_I(inode)->root->objectid)
6021 if (inode->i_nlink >= BTRFS_LINK_MAX)
6024 err = btrfs_set_inode_index(dir, &index);
6029 * 2 items for inode and inode ref
6030 * 2 items for dir items
6031 * 1 item for parent inode
6033 trans = btrfs_start_transaction(root, 5);
6034 if (IS_ERR(trans)) {
6035 err = PTR_ERR(trans);
6039 /* There are several dir indexes for this inode, clear the cache. */
6040 BTRFS_I(inode)->dir_index = 0ULL;
6042 inode_inc_iversion(inode);
6043 inode->i_ctime = CURRENT_TIME;
6045 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6047 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6052 struct dentry *parent = dentry->d_parent;
6053 err = btrfs_update_inode(trans, root, inode);
6056 if (inode->i_nlink == 1) {
6058 * If new hard link count is 1, it's a file created
6059 * with open(2) O_TMPFILE flag.
6061 err = btrfs_orphan_del(trans, inode);
6065 d_instantiate(dentry, inode);
6066 btrfs_log_new_name(trans, inode, NULL, parent);
6069 btrfs_end_transaction(trans, root);
6070 btrfs_balance_delayed_items(root);
6073 inode_dec_link_count(inode);
6076 btrfs_btree_balance_dirty(root);
6080 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6082 struct inode *inode = NULL;
6083 struct btrfs_trans_handle *trans;
6084 struct btrfs_root *root = BTRFS_I(dir)->root;
6086 int drop_on_err = 0;
6091 * 2 items for inode and ref
6092 * 2 items for dir items
6093 * 1 for xattr if selinux is on
6095 trans = btrfs_start_transaction(root, 5);
6097 return PTR_ERR(trans);
6099 err = btrfs_find_free_ino(root, &objectid);
6103 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6104 dentry->d_name.len, btrfs_ino(dir), objectid,
6105 S_IFDIR | mode, &index);
6106 if (IS_ERR(inode)) {
6107 err = PTR_ERR(inode);
6113 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6117 inode->i_op = &btrfs_dir_inode_operations;
6118 inode->i_fop = &btrfs_dir_file_operations;
6120 btrfs_i_size_write(inode, 0);
6121 err = btrfs_update_inode(trans, root, inode);
6125 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6126 dentry->d_name.len, 0, index);
6130 d_instantiate(dentry, inode);
6134 btrfs_end_transaction(trans, root);
6137 btrfs_balance_delayed_items(root);
6138 btrfs_btree_balance_dirty(root);
6142 /* helper for btfs_get_extent. Given an existing extent in the tree,
6143 * and an extent that you want to insert, deal with overlap and insert
6144 * the new extent into the tree.
6146 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6147 struct extent_map *existing,
6148 struct extent_map *em,
6153 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6154 start_diff = map_start - em->start;
6155 em->start = map_start;
6156 em->len = existing->start - em->start;
6157 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6158 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6159 em->block_start += start_diff;
6160 em->block_len -= start_diff;
6162 return add_extent_mapping(em_tree, em, 0);
6165 static noinline int uncompress_inline(struct btrfs_path *path,
6166 struct inode *inode, struct page *page,
6167 size_t pg_offset, u64 extent_offset,
6168 struct btrfs_file_extent_item *item)
6171 struct extent_buffer *leaf = path->nodes[0];
6174 unsigned long inline_size;
6178 WARN_ON(pg_offset != 0);
6179 compress_type = btrfs_file_extent_compression(leaf, item);
6180 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6181 inline_size = btrfs_file_extent_inline_item_len(leaf,
6182 btrfs_item_nr(path->slots[0]));
6183 tmp = kmalloc(inline_size, GFP_NOFS);
6186 ptr = btrfs_file_extent_inline_start(item);
6188 read_extent_buffer(leaf, tmp, ptr, inline_size);
6190 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6191 ret = btrfs_decompress(compress_type, tmp, page,
6192 extent_offset, inline_size, max_size);
6198 * a bit scary, this does extent mapping from logical file offset to the disk.
6199 * the ugly parts come from merging extents from the disk with the in-ram
6200 * representation. This gets more complex because of the data=ordered code,
6201 * where the in-ram extents might be locked pending data=ordered completion.
6203 * This also copies inline extents directly into the page.
6206 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6207 size_t pg_offset, u64 start, u64 len,
6212 u64 extent_start = 0;
6214 u64 objectid = btrfs_ino(inode);
6216 struct btrfs_path *path = NULL;
6217 struct btrfs_root *root = BTRFS_I(inode)->root;
6218 struct btrfs_file_extent_item *item;
6219 struct extent_buffer *leaf;
6220 struct btrfs_key found_key;
6221 struct extent_map *em = NULL;
6222 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6223 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6224 struct btrfs_trans_handle *trans = NULL;
6225 const bool new_inline = !page || create;
6228 read_lock(&em_tree->lock);
6229 em = lookup_extent_mapping(em_tree, start, len);
6231 em->bdev = root->fs_info->fs_devices->latest_bdev;
6232 read_unlock(&em_tree->lock);
6235 if (em->start > start || em->start + em->len <= start)
6236 free_extent_map(em);
6237 else if (em->block_start == EXTENT_MAP_INLINE && page)
6238 free_extent_map(em);
6242 em = alloc_extent_map();
6247 em->bdev = root->fs_info->fs_devices->latest_bdev;
6248 em->start = EXTENT_MAP_HOLE;
6249 em->orig_start = EXTENT_MAP_HOLE;
6251 em->block_len = (u64)-1;
6254 path = btrfs_alloc_path();
6260 * Chances are we'll be called again, so go ahead and do
6266 ret = btrfs_lookup_file_extent(trans, root, path,
6267 objectid, start, trans != NULL);
6274 if (path->slots[0] == 0)
6279 leaf = path->nodes[0];
6280 item = btrfs_item_ptr(leaf, path->slots[0],
6281 struct btrfs_file_extent_item);
6282 /* are we inside the extent that was found? */
6283 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6284 found_type = btrfs_key_type(&found_key);
6285 if (found_key.objectid != objectid ||
6286 found_type != BTRFS_EXTENT_DATA_KEY) {
6288 * If we backup past the first extent we want to move forward
6289 * and see if there is an extent in front of us, otherwise we'll
6290 * say there is a hole for our whole search range which can
6297 found_type = btrfs_file_extent_type(leaf, item);
6298 extent_start = found_key.offset;
6299 if (found_type == BTRFS_FILE_EXTENT_REG ||
6300 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6301 extent_end = extent_start +
6302 btrfs_file_extent_num_bytes(leaf, item);
6303 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6305 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6306 extent_end = ALIGN(extent_start + size, root->sectorsize);
6309 if (start >= extent_end) {
6311 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6312 ret = btrfs_next_leaf(root, path);
6319 leaf = path->nodes[0];
6321 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6322 if (found_key.objectid != objectid ||
6323 found_key.type != BTRFS_EXTENT_DATA_KEY)
6325 if (start + len <= found_key.offset)
6327 if (start > found_key.offset)
6330 em->orig_start = start;
6331 em->len = found_key.offset - start;
6335 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6337 if (found_type == BTRFS_FILE_EXTENT_REG ||
6338 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6340 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6344 size_t extent_offset;
6350 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6351 extent_offset = page_offset(page) + pg_offset - extent_start;
6352 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6353 size - extent_offset);
6354 em->start = extent_start + extent_offset;
6355 em->len = ALIGN(copy_size, root->sectorsize);
6356 em->orig_block_len = em->len;
6357 em->orig_start = em->start;
6358 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6359 if (create == 0 && !PageUptodate(page)) {
6360 if (btrfs_file_extent_compression(leaf, item) !=
6361 BTRFS_COMPRESS_NONE) {
6362 ret = uncompress_inline(path, inode, page,
6364 extent_offset, item);
6371 read_extent_buffer(leaf, map + pg_offset, ptr,
6373 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6374 memset(map + pg_offset + copy_size, 0,
6375 PAGE_CACHE_SIZE - pg_offset -
6380 flush_dcache_page(page);
6381 } else if (create && PageUptodate(page)) {
6385 free_extent_map(em);
6388 btrfs_release_path(path);
6389 trans = btrfs_join_transaction(root);
6392 return ERR_CAST(trans);
6396 write_extent_buffer(leaf, map + pg_offset, ptr,
6399 btrfs_mark_buffer_dirty(leaf);
6401 set_extent_uptodate(io_tree, em->start,
6402 extent_map_end(em) - 1, NULL, GFP_NOFS);
6407 em->orig_start = start;
6410 em->block_start = EXTENT_MAP_HOLE;
6411 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6413 btrfs_release_path(path);
6414 if (em->start > start || extent_map_end(em) <= start) {
6415 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6416 em->start, em->len, start, len);
6422 write_lock(&em_tree->lock);
6423 ret = add_extent_mapping(em_tree, em, 0);
6424 /* it is possible that someone inserted the extent into the tree
6425 * while we had the lock dropped. It is also possible that
6426 * an overlapping map exists in the tree
6428 if (ret == -EEXIST) {
6429 struct extent_map *existing;
6433 existing = lookup_extent_mapping(em_tree, start, len);
6434 if (existing && (existing->start > start ||
6435 existing->start + existing->len <= start)) {
6436 free_extent_map(existing);
6440 existing = lookup_extent_mapping(em_tree, em->start,
6443 err = merge_extent_mapping(em_tree, existing,
6445 free_extent_map(existing);
6447 free_extent_map(em);
6452 free_extent_map(em);
6456 free_extent_map(em);
6461 write_unlock(&em_tree->lock);
6464 trace_btrfs_get_extent(root, em);
6467 btrfs_free_path(path);
6469 ret = btrfs_end_transaction(trans, root);
6474 free_extent_map(em);
6475 return ERR_PTR(err);
6477 BUG_ON(!em); /* Error is always set */
6481 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6482 size_t pg_offset, u64 start, u64 len,
6485 struct extent_map *em;
6486 struct extent_map *hole_em = NULL;
6487 u64 range_start = start;
6493 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6500 * - a pre-alloc extent,
6501 * there might actually be delalloc bytes behind it.
6503 if (em->block_start != EXTENT_MAP_HOLE &&
6504 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6510 /* check to see if we've wrapped (len == -1 or similar) */
6519 /* ok, we didn't find anything, lets look for delalloc */
6520 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6521 end, len, EXTENT_DELALLOC, 1);
6522 found_end = range_start + found;
6523 if (found_end < range_start)
6524 found_end = (u64)-1;
6527 * we didn't find anything useful, return
6528 * the original results from get_extent()
6530 if (range_start > end || found_end <= start) {
6536 /* adjust the range_start to make sure it doesn't
6537 * go backwards from the start they passed in
6539 range_start = max(start, range_start);
6540 found = found_end - range_start;
6543 u64 hole_start = start;
6546 em = alloc_extent_map();
6552 * when btrfs_get_extent can't find anything it
6553 * returns one huge hole
6555 * make sure what it found really fits our range, and
6556 * adjust to make sure it is based on the start from
6560 u64 calc_end = extent_map_end(hole_em);
6562 if (calc_end <= start || (hole_em->start > end)) {
6563 free_extent_map(hole_em);
6566 hole_start = max(hole_em->start, start);
6567 hole_len = calc_end - hole_start;
6571 if (hole_em && range_start > hole_start) {
6572 /* our hole starts before our delalloc, so we
6573 * have to return just the parts of the hole
6574 * that go until the delalloc starts
6576 em->len = min(hole_len,
6577 range_start - hole_start);
6578 em->start = hole_start;
6579 em->orig_start = hole_start;
6581 * don't adjust block start at all,
6582 * it is fixed at EXTENT_MAP_HOLE
6584 em->block_start = hole_em->block_start;
6585 em->block_len = hole_len;
6586 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6587 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6589 em->start = range_start;
6591 em->orig_start = range_start;
6592 em->block_start = EXTENT_MAP_DELALLOC;
6593 em->block_len = found;
6595 } else if (hole_em) {
6600 free_extent_map(hole_em);
6602 free_extent_map(em);
6603 return ERR_PTR(err);
6608 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6611 struct btrfs_root *root = BTRFS_I(inode)->root;
6612 struct extent_map *em;
6613 struct btrfs_key ins;
6617 alloc_hint = get_extent_allocation_hint(inode, start, len);
6618 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
6619 alloc_hint, &ins, 1, 1);
6621 return ERR_PTR(ret);
6623 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6624 ins.offset, ins.offset, ins.offset, 0);
6626 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
6630 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6631 ins.offset, ins.offset, 0);
6633 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
6634 free_extent_map(em);
6635 return ERR_PTR(ret);
6642 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6643 * block must be cow'd
6645 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6646 u64 *orig_start, u64 *orig_block_len,
6649 struct btrfs_trans_handle *trans;
6650 struct btrfs_path *path;
6652 struct extent_buffer *leaf;
6653 struct btrfs_root *root = BTRFS_I(inode)->root;
6654 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6655 struct btrfs_file_extent_item *fi;
6656 struct btrfs_key key;
6663 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6665 path = btrfs_alloc_path();
6669 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
6674 slot = path->slots[0];
6677 /* can't find the item, must cow */
6684 leaf = path->nodes[0];
6685 btrfs_item_key_to_cpu(leaf, &key, slot);
6686 if (key.objectid != btrfs_ino(inode) ||
6687 key.type != BTRFS_EXTENT_DATA_KEY) {
6688 /* not our file or wrong item type, must cow */
6692 if (key.offset > offset) {
6693 /* Wrong offset, must cow */
6697 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6698 found_type = btrfs_file_extent_type(leaf, fi);
6699 if (found_type != BTRFS_FILE_EXTENT_REG &&
6700 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6701 /* not a regular extent, must cow */
6705 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6708 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6709 if (extent_end <= offset)
6712 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6713 if (disk_bytenr == 0)
6716 if (btrfs_file_extent_compression(leaf, fi) ||
6717 btrfs_file_extent_encryption(leaf, fi) ||
6718 btrfs_file_extent_other_encoding(leaf, fi))
6721 backref_offset = btrfs_file_extent_offset(leaf, fi);
6724 *orig_start = key.offset - backref_offset;
6725 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6726 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6729 if (btrfs_extent_readonly(root, disk_bytenr))
6732 num_bytes = min(offset + *len, extent_end) - offset;
6733 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6736 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
6737 ret = test_range_bit(io_tree, offset, range_end,
6738 EXTENT_DELALLOC, 0, NULL);
6745 btrfs_release_path(path);
6748 * look for other files referencing this extent, if we
6749 * find any we must cow
6751 trans = btrfs_join_transaction(root);
6752 if (IS_ERR(trans)) {
6757 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6758 key.offset - backref_offset, disk_bytenr);
6759 btrfs_end_transaction(trans, root);
6766 * adjust disk_bytenr and num_bytes to cover just the bytes
6767 * in this extent we are about to write. If there
6768 * are any csums in that range we have to cow in order
6769 * to keep the csums correct
6771 disk_bytenr += backref_offset;
6772 disk_bytenr += offset - key.offset;
6773 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6776 * all of the above have passed, it is safe to overwrite this extent
6782 btrfs_free_path(path);
6786 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
6788 struct radix_tree_root *root = &inode->i_mapping->page_tree;
6790 void **pagep = NULL;
6791 struct page *page = NULL;
6795 start_idx = start >> PAGE_CACHE_SHIFT;
6798 * end is the last byte in the last page. end == start is legal
6800 end_idx = end >> PAGE_CACHE_SHIFT;
6804 /* Most of the code in this while loop is lifted from
6805 * find_get_page. It's been modified to begin searching from a
6806 * page and return just the first page found in that range. If the
6807 * found idx is less than or equal to the end idx then we know that
6808 * a page exists. If no pages are found or if those pages are
6809 * outside of the range then we're fine (yay!) */
6810 while (page == NULL &&
6811 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
6812 page = radix_tree_deref_slot(pagep);
6813 if (unlikely(!page))
6816 if (radix_tree_exception(page)) {
6817 if (radix_tree_deref_retry(page)) {
6822 * Otherwise, shmem/tmpfs must be storing a swap entry
6823 * here as an exceptional entry: so return it without
6824 * attempting to raise page count.
6827 break; /* TODO: Is this relevant for this use case? */
6830 if (!page_cache_get_speculative(page)) {
6836 * Has the page moved?
6837 * This is part of the lockless pagecache protocol. See
6838 * include/linux/pagemap.h for details.
6840 if (unlikely(page != *pagep)) {
6841 page_cache_release(page);
6847 if (page->index <= end_idx)
6849 page_cache_release(page);
6856 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6857 struct extent_state **cached_state, int writing)
6859 struct btrfs_ordered_extent *ordered;
6863 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6866 * We're concerned with the entire range that we're going to be
6867 * doing DIO to, so we need to make sure theres no ordered
6868 * extents in this range.
6870 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6871 lockend - lockstart + 1);
6874 * We need to make sure there are no buffered pages in this
6875 * range either, we could have raced between the invalidate in
6876 * generic_file_direct_write and locking the extent. The
6877 * invalidate needs to happen so that reads after a write do not
6882 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
6885 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6886 cached_state, GFP_NOFS);
6889 btrfs_start_ordered_extent(inode, ordered, 1);
6890 btrfs_put_ordered_extent(ordered);
6892 /* Screw you mmap */
6893 ret = filemap_write_and_wait_range(inode->i_mapping,
6900 * If we found a page that couldn't be invalidated just
6901 * fall back to buffered.
6903 ret = invalidate_inode_pages2_range(inode->i_mapping,
6904 lockstart >> PAGE_CACHE_SHIFT,
6905 lockend >> PAGE_CACHE_SHIFT);
6916 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6917 u64 len, u64 orig_start,
6918 u64 block_start, u64 block_len,
6919 u64 orig_block_len, u64 ram_bytes,
6922 struct extent_map_tree *em_tree;
6923 struct extent_map *em;
6924 struct btrfs_root *root = BTRFS_I(inode)->root;
6927 em_tree = &BTRFS_I(inode)->extent_tree;
6928 em = alloc_extent_map();
6930 return ERR_PTR(-ENOMEM);
6933 em->orig_start = orig_start;
6934 em->mod_start = start;
6937 em->block_len = block_len;
6938 em->block_start = block_start;
6939 em->bdev = root->fs_info->fs_devices->latest_bdev;
6940 em->orig_block_len = orig_block_len;
6941 em->ram_bytes = ram_bytes;
6942 em->generation = -1;
6943 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6944 if (type == BTRFS_ORDERED_PREALLOC)
6945 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6948 btrfs_drop_extent_cache(inode, em->start,
6949 em->start + em->len - 1, 0);
6950 write_lock(&em_tree->lock);
6951 ret = add_extent_mapping(em_tree, em, 1);
6952 write_unlock(&em_tree->lock);
6953 } while (ret == -EEXIST);
6956 free_extent_map(em);
6957 return ERR_PTR(ret);
6964 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6965 struct buffer_head *bh_result, int create)
6967 struct extent_map *em;
6968 struct btrfs_root *root = BTRFS_I(inode)->root;
6969 struct extent_state *cached_state = NULL;
6970 u64 start = iblock << inode->i_blkbits;
6971 u64 lockstart, lockend;
6972 u64 len = bh_result->b_size;
6973 int unlock_bits = EXTENT_LOCKED;
6977 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6979 len = min_t(u64, len, root->sectorsize);
6982 lockend = start + len - 1;
6985 * If this errors out it's because we couldn't invalidate pagecache for
6986 * this range and we need to fallback to buffered.
6988 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6991 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6998 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6999 * io. INLINE is special, and we could probably kludge it in here, but
7000 * it's still buffered so for safety lets just fall back to the generic
7003 * For COMPRESSED we _have_ to read the entire extent in so we can
7004 * decompress it, so there will be buffering required no matter what we
7005 * do, so go ahead and fallback to buffered.
7007 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7008 * to buffered IO. Don't blame me, this is the price we pay for using
7011 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7012 em->block_start == EXTENT_MAP_INLINE) {
7013 free_extent_map(em);
7018 /* Just a good old fashioned hole, return */
7019 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7020 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7021 free_extent_map(em);
7026 * We don't allocate a new extent in the following cases
7028 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7030 * 2) The extent is marked as PREALLOC. We're good to go here and can
7031 * just use the extent.
7035 len = min(len, em->len - (start - em->start));
7036 lockstart = start + len;
7040 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7041 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7042 em->block_start != EXTENT_MAP_HOLE)) {
7045 u64 block_start, orig_start, orig_block_len, ram_bytes;
7047 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7048 type = BTRFS_ORDERED_PREALLOC;
7050 type = BTRFS_ORDERED_NOCOW;
7051 len = min(len, em->len - (start - em->start));
7052 block_start = em->block_start + (start - em->start);
7054 if (can_nocow_extent(inode, start, &len, &orig_start,
7055 &orig_block_len, &ram_bytes) == 1) {
7056 if (type == BTRFS_ORDERED_PREALLOC) {
7057 free_extent_map(em);
7058 em = create_pinned_em(inode, start, len,
7067 ret = btrfs_add_ordered_extent_dio(inode, start,
7068 block_start, len, len, type);
7070 free_extent_map(em);
7078 * this will cow the extent, reset the len in case we changed
7081 len = bh_result->b_size;
7082 free_extent_map(em);
7083 em = btrfs_new_extent_direct(inode, start, len);
7088 len = min(len, em->len - (start - em->start));
7090 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7092 bh_result->b_size = len;
7093 bh_result->b_bdev = em->bdev;
7094 set_buffer_mapped(bh_result);
7096 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7097 set_buffer_new(bh_result);
7100 * Need to update the i_size under the extent lock so buffered
7101 * readers will get the updated i_size when we unlock.
7103 if (start + len > i_size_read(inode))
7104 i_size_write(inode, start + len);
7106 spin_lock(&BTRFS_I(inode)->lock);
7107 BTRFS_I(inode)->outstanding_extents++;
7108 spin_unlock(&BTRFS_I(inode)->lock);
7110 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7111 lockstart + len - 1, EXTENT_DELALLOC, NULL,
7112 &cached_state, GFP_NOFS);
7117 * In the case of write we need to clear and unlock the entire range,
7118 * in the case of read we need to unlock only the end area that we
7119 * aren't using if there is any left over space.
7121 if (lockstart < lockend) {
7122 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7123 lockend, unlock_bits, 1, 0,
7124 &cached_state, GFP_NOFS);
7126 free_extent_state(cached_state);
7129 free_extent_map(em);
7134 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7135 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7139 static void btrfs_endio_direct_read(struct bio *bio, int err)
7141 struct btrfs_dio_private *dip = bio->bi_private;
7142 struct bio_vec *bvec;
7143 struct inode *inode = dip->inode;
7144 struct btrfs_root *root = BTRFS_I(inode)->root;
7145 struct bio *dio_bio;
7146 u32 *csums = (u32 *)dip->csum;
7150 start = dip->logical_offset;
7151 bio_for_each_segment_all(bvec, bio, i) {
7152 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
7153 struct page *page = bvec->bv_page;
7156 unsigned long flags;
7158 local_irq_save(flags);
7159 kaddr = kmap_atomic(page);
7160 csum = btrfs_csum_data(kaddr + bvec->bv_offset,
7161 csum, bvec->bv_len);
7162 btrfs_csum_final(csum, (char *)&csum);
7163 kunmap_atomic(kaddr);
7164 local_irq_restore(flags);
7166 flush_dcache_page(bvec->bv_page);
7167 if (csum != csums[i]) {
7168 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
7169 btrfs_ino(inode), start, csum,
7175 start += bvec->bv_len;
7178 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
7179 dip->logical_offset + dip->bytes - 1);
7180 dio_bio = dip->dio_bio;
7184 /* If we had a csum failure make sure to clear the uptodate flag */
7186 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7187 dio_end_io(dio_bio, err);
7191 static void btrfs_endio_direct_write(struct bio *bio, int err)
7193 struct btrfs_dio_private *dip = bio->bi_private;
7194 struct inode *inode = dip->inode;
7195 struct btrfs_root *root = BTRFS_I(inode)->root;
7196 struct btrfs_ordered_extent *ordered = NULL;
7197 u64 ordered_offset = dip->logical_offset;
7198 u64 ordered_bytes = dip->bytes;
7199 struct bio *dio_bio;
7205 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7207 ordered_bytes, !err);
7211 btrfs_init_work(&ordered->work, finish_ordered_fn, NULL, NULL);
7212 btrfs_queue_work(root->fs_info->endio_write_workers,
7216 * our bio might span multiple ordered extents. If we haven't
7217 * completed the accounting for the whole dio, go back and try again
7219 if (ordered_offset < dip->logical_offset + dip->bytes) {
7220 ordered_bytes = dip->logical_offset + dip->bytes -
7226 dio_bio = dip->dio_bio;
7230 /* If we had an error make sure to clear the uptodate flag */
7232 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7233 dio_end_io(dio_bio, err);
7237 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7238 struct bio *bio, int mirror_num,
7239 unsigned long bio_flags, u64 offset)
7242 struct btrfs_root *root = BTRFS_I(inode)->root;
7243 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7244 BUG_ON(ret); /* -ENOMEM */
7248 static void btrfs_end_dio_bio(struct bio *bio, int err)
7250 struct btrfs_dio_private *dip = bio->bi_private;
7253 btrfs_err(BTRFS_I(dip->inode)->root->fs_info,
7254 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7255 btrfs_ino(dip->inode), bio->bi_rw,
7256 (unsigned long long)bio->bi_iter.bi_sector,
7257 bio->bi_iter.bi_size, err);
7261 * before atomic variable goto zero, we must make sure
7262 * dip->errors is perceived to be set.
7264 smp_mb__before_atomic();
7267 /* if there are more bios still pending for this dio, just exit */
7268 if (!atomic_dec_and_test(&dip->pending_bios))
7272 bio_io_error(dip->orig_bio);
7274 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7275 bio_endio(dip->orig_bio, 0);
7281 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7282 u64 first_sector, gfp_t gfp_flags)
7284 int nr_vecs = bio_get_nr_vecs(bdev);
7285 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7288 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7289 int rw, u64 file_offset, int skip_sum,
7292 struct btrfs_dio_private *dip = bio->bi_private;
7293 int write = rw & REQ_WRITE;
7294 struct btrfs_root *root = BTRFS_I(inode)->root;
7298 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7303 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7311 if (write && async_submit) {
7312 ret = btrfs_wq_submit_bio(root->fs_info,
7313 inode, rw, bio, 0, 0,
7315 __btrfs_submit_bio_start_direct_io,
7316 __btrfs_submit_bio_done);
7320 * If we aren't doing async submit, calculate the csum of the
7323 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7326 } else if (!skip_sum) {
7327 ret = btrfs_lookup_bio_sums_dio(root, inode, dip, bio,
7334 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7340 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7343 struct inode *inode = dip->inode;
7344 struct btrfs_root *root = BTRFS_I(inode)->root;
7346 struct bio *orig_bio = dip->orig_bio;
7347 struct bio_vec *bvec = orig_bio->bi_io_vec;
7348 u64 start_sector = orig_bio->bi_iter.bi_sector;
7349 u64 file_offset = dip->logical_offset;
7354 int async_submit = 0;
7356 map_length = orig_bio->bi_iter.bi_size;
7357 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7358 &map_length, NULL, 0);
7362 if (map_length >= orig_bio->bi_iter.bi_size) {
7367 /* async crcs make it difficult to collect full stripe writes. */
7368 if (btrfs_get_alloc_profile(root, 1) &
7369 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7374 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7378 bio->bi_private = dip;
7379 bio->bi_end_io = btrfs_end_dio_bio;
7380 atomic_inc(&dip->pending_bios);
7382 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7383 if (unlikely(map_length < submit_len + bvec->bv_len ||
7384 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7385 bvec->bv_offset) < bvec->bv_len)) {
7387 * inc the count before we submit the bio so
7388 * we know the end IO handler won't happen before
7389 * we inc the count. Otherwise, the dip might get freed
7390 * before we're done setting it up
7392 atomic_inc(&dip->pending_bios);
7393 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7394 file_offset, skip_sum,
7398 atomic_dec(&dip->pending_bios);
7402 start_sector += submit_len >> 9;
7403 file_offset += submit_len;
7408 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7409 start_sector, GFP_NOFS);
7412 bio->bi_private = dip;
7413 bio->bi_end_io = btrfs_end_dio_bio;
7415 map_length = orig_bio->bi_iter.bi_size;
7416 ret = btrfs_map_block(root->fs_info, rw,
7418 &map_length, NULL, 0);
7424 submit_len += bvec->bv_len;
7431 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7440 * before atomic variable goto zero, we must
7441 * make sure dip->errors is perceived to be set.
7443 smp_mb__before_atomic();
7444 if (atomic_dec_and_test(&dip->pending_bios))
7445 bio_io_error(dip->orig_bio);
7447 /* bio_end_io() will handle error, so we needn't return it */
7451 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7452 struct inode *inode, loff_t file_offset)
7454 struct btrfs_root *root = BTRFS_I(inode)->root;
7455 struct btrfs_dio_private *dip;
7459 int write = rw & REQ_WRITE;
7463 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7465 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7471 if (!skip_sum && !write) {
7472 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
7473 sum_len = dio_bio->bi_iter.bi_size >>
7474 inode->i_sb->s_blocksize_bits;
7475 sum_len *= csum_size;
7480 dip = kmalloc(sizeof(*dip) + sum_len, GFP_NOFS);
7486 dip->private = dio_bio->bi_private;
7488 dip->logical_offset = file_offset;
7489 dip->bytes = dio_bio->bi_iter.bi_size;
7490 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
7491 io_bio->bi_private = dip;
7493 dip->orig_bio = io_bio;
7494 dip->dio_bio = dio_bio;
7495 atomic_set(&dip->pending_bios, 0);
7498 io_bio->bi_end_io = btrfs_endio_direct_write;
7500 io_bio->bi_end_io = btrfs_endio_direct_read;
7502 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7511 * If this is a write, we need to clean up the reserved space and kill
7512 * the ordered extent.
7515 struct btrfs_ordered_extent *ordered;
7516 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7517 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7518 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7519 btrfs_free_reserved_extent(root, ordered->start,
7520 ordered->disk_len, 1);
7521 btrfs_put_ordered_extent(ordered);
7522 btrfs_put_ordered_extent(ordered);
7524 bio_endio(dio_bio, ret);
7527 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7528 const struct iov_iter *iter, loff_t offset)
7532 unsigned blocksize_mask = root->sectorsize - 1;
7533 ssize_t retval = -EINVAL;
7535 if (offset & blocksize_mask)
7538 if (iov_iter_alignment(iter) & blocksize_mask)
7541 /* If this is a write we don't need to check anymore */
7545 * Check to make sure we don't have duplicate iov_base's in this
7546 * iovec, if so return EINVAL, otherwise we'll get csum errors
7547 * when reading back.
7549 for (seg = 0; seg < iter->nr_segs; seg++) {
7550 for (i = seg + 1; i < iter->nr_segs; i++) {
7551 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
7560 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7561 struct iov_iter *iter, loff_t offset)
7563 struct file *file = iocb->ki_filp;
7564 struct inode *inode = file->f_mapping->host;
7568 bool relock = false;
7571 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iter, offset))
7574 atomic_inc(&inode->i_dio_count);
7575 smp_mb__after_atomic();
7578 * The generic stuff only does filemap_write_and_wait_range, which
7579 * isn't enough if we've written compressed pages to this area, so
7580 * we need to flush the dirty pages again to make absolutely sure
7581 * that any outstanding dirty pages are on disk.
7583 count = iov_iter_count(iter);
7584 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7585 &BTRFS_I(inode)->runtime_flags))
7586 filemap_fdatawrite_range(inode->i_mapping, offset,
7587 offset + count - 1);
7591 * If the write DIO is beyond the EOF, we need update
7592 * the isize, but it is protected by i_mutex. So we can
7593 * not unlock the i_mutex at this case.
7595 if (offset + count <= inode->i_size) {
7596 mutex_unlock(&inode->i_mutex);
7599 ret = btrfs_delalloc_reserve_space(inode, count);
7602 } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7603 &BTRFS_I(inode)->runtime_flags))) {
7604 inode_dio_done(inode);
7605 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7609 ret = __blockdev_direct_IO(rw, iocb, inode,
7610 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7611 iter, offset, btrfs_get_blocks_direct, NULL,
7612 btrfs_submit_direct, flags);
7614 if (ret < 0 && ret != -EIOCBQUEUED)
7615 btrfs_delalloc_release_space(inode, count);
7616 else if (ret >= 0 && (size_t)ret < count)
7617 btrfs_delalloc_release_space(inode,
7618 count - (size_t)ret);
7620 btrfs_delalloc_release_metadata(inode, 0);
7624 inode_dio_done(inode);
7626 mutex_lock(&inode->i_mutex);
7631 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
7633 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7634 __u64 start, __u64 len)
7638 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7642 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7645 int btrfs_readpage(struct file *file, struct page *page)
7647 struct extent_io_tree *tree;
7648 tree = &BTRFS_I(page->mapping->host)->io_tree;
7649 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7652 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7654 struct extent_io_tree *tree;
7657 if (current->flags & PF_MEMALLOC) {
7658 redirty_page_for_writepage(wbc, page);
7662 tree = &BTRFS_I(page->mapping->host)->io_tree;
7663 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7666 static int btrfs_writepages(struct address_space *mapping,
7667 struct writeback_control *wbc)
7669 struct extent_io_tree *tree;
7671 tree = &BTRFS_I(mapping->host)->io_tree;
7672 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7676 btrfs_readpages(struct file *file, struct address_space *mapping,
7677 struct list_head *pages, unsigned nr_pages)
7679 struct extent_io_tree *tree;
7680 tree = &BTRFS_I(mapping->host)->io_tree;
7681 return extent_readpages(tree, mapping, pages, nr_pages,
7684 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7686 struct extent_io_tree *tree;
7687 struct extent_map_tree *map;
7690 tree = &BTRFS_I(page->mapping->host)->io_tree;
7691 map = &BTRFS_I(page->mapping->host)->extent_tree;
7692 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7694 ClearPagePrivate(page);
7695 set_page_private(page, 0);
7696 page_cache_release(page);
7701 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7703 if (PageWriteback(page) || PageDirty(page))
7705 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7708 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
7709 unsigned int length)
7711 struct inode *inode = page->mapping->host;
7712 struct extent_io_tree *tree;
7713 struct btrfs_ordered_extent *ordered;
7714 struct extent_state *cached_state = NULL;
7715 u64 page_start = page_offset(page);
7716 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7717 int inode_evicting = inode->i_state & I_FREEING;
7720 * we have the page locked, so new writeback can't start,
7721 * and the dirty bit won't be cleared while we are here.
7723 * Wait for IO on this page so that we can safely clear
7724 * the PagePrivate2 bit and do ordered accounting
7726 wait_on_page_writeback(page);
7728 tree = &BTRFS_I(inode)->io_tree;
7730 btrfs_releasepage(page, GFP_NOFS);
7734 if (!inode_evicting)
7735 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7736 ordered = btrfs_lookup_ordered_extent(inode, page_start);
7739 * IO on this page will never be started, so we need
7740 * to account for any ordered extents now
7742 if (!inode_evicting)
7743 clear_extent_bit(tree, page_start, page_end,
7744 EXTENT_DIRTY | EXTENT_DELALLOC |
7745 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7746 EXTENT_DEFRAG, 1, 0, &cached_state,
7749 * whoever cleared the private bit is responsible
7750 * for the finish_ordered_io
7752 if (TestClearPagePrivate2(page)) {
7753 struct btrfs_ordered_inode_tree *tree;
7756 tree = &BTRFS_I(inode)->ordered_tree;
7758 spin_lock_irq(&tree->lock);
7759 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
7760 new_len = page_start - ordered->file_offset;
7761 if (new_len < ordered->truncated_len)
7762 ordered->truncated_len = new_len;
7763 spin_unlock_irq(&tree->lock);
7765 if (btrfs_dec_test_ordered_pending(inode, &ordered,
7767 PAGE_CACHE_SIZE, 1))
7768 btrfs_finish_ordered_io(ordered);
7770 btrfs_put_ordered_extent(ordered);
7771 if (!inode_evicting) {
7772 cached_state = NULL;
7773 lock_extent_bits(tree, page_start, page_end, 0,
7778 if (!inode_evicting) {
7779 clear_extent_bit(tree, page_start, page_end,
7780 EXTENT_LOCKED | EXTENT_DIRTY |
7781 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
7782 EXTENT_DEFRAG, 1, 1,
7783 &cached_state, GFP_NOFS);
7785 __btrfs_releasepage(page, GFP_NOFS);
7788 ClearPageChecked(page);
7789 if (PagePrivate(page)) {
7790 ClearPagePrivate(page);
7791 set_page_private(page, 0);
7792 page_cache_release(page);
7797 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7798 * called from a page fault handler when a page is first dirtied. Hence we must
7799 * be careful to check for EOF conditions here. We set the page up correctly
7800 * for a written page which means we get ENOSPC checking when writing into
7801 * holes and correct delalloc and unwritten extent mapping on filesystems that
7802 * support these features.
7804 * We are not allowed to take the i_mutex here so we have to play games to
7805 * protect against truncate races as the page could now be beyond EOF. Because
7806 * vmtruncate() writes the inode size before removing pages, once we have the
7807 * page lock we can determine safely if the page is beyond EOF. If it is not
7808 * beyond EOF, then the page is guaranteed safe against truncation until we
7811 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7813 struct page *page = vmf->page;
7814 struct inode *inode = file_inode(vma->vm_file);
7815 struct btrfs_root *root = BTRFS_I(inode)->root;
7816 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7817 struct btrfs_ordered_extent *ordered;
7818 struct extent_state *cached_state = NULL;
7820 unsigned long zero_start;
7827 sb_start_pagefault(inode->i_sb);
7828 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7830 ret = file_update_time(vma->vm_file);
7836 else /* -ENOSPC, -EIO, etc */
7837 ret = VM_FAULT_SIGBUS;
7843 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7846 size = i_size_read(inode);
7847 page_start = page_offset(page);
7848 page_end = page_start + PAGE_CACHE_SIZE - 1;
7850 if ((page->mapping != inode->i_mapping) ||
7851 (page_start >= size)) {
7852 /* page got truncated out from underneath us */
7855 wait_on_page_writeback(page);
7857 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7858 set_page_extent_mapped(page);
7861 * we can't set the delalloc bits if there are pending ordered
7862 * extents. Drop our locks and wait for them to finish
7864 ordered = btrfs_lookup_ordered_extent(inode, page_start);
7866 unlock_extent_cached(io_tree, page_start, page_end,
7867 &cached_state, GFP_NOFS);
7869 btrfs_start_ordered_extent(inode, ordered, 1);
7870 btrfs_put_ordered_extent(ordered);
7875 * XXX - page_mkwrite gets called every time the page is dirtied, even
7876 * if it was already dirty, so for space accounting reasons we need to
7877 * clear any delalloc bits for the range we are fixing to save. There
7878 * is probably a better way to do this, but for now keep consistent with
7879 * prepare_pages in the normal write path.
7881 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7882 EXTENT_DIRTY | EXTENT_DELALLOC |
7883 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7884 0, 0, &cached_state, GFP_NOFS);
7886 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7889 unlock_extent_cached(io_tree, page_start, page_end,
7890 &cached_state, GFP_NOFS);
7891 ret = VM_FAULT_SIGBUS;
7896 /* page is wholly or partially inside EOF */
7897 if (page_start + PAGE_CACHE_SIZE > size)
7898 zero_start = size & ~PAGE_CACHE_MASK;
7900 zero_start = PAGE_CACHE_SIZE;
7902 if (zero_start != PAGE_CACHE_SIZE) {
7904 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7905 flush_dcache_page(page);
7908 ClearPageChecked(page);
7909 set_page_dirty(page);
7910 SetPageUptodate(page);
7912 BTRFS_I(inode)->last_trans = root->fs_info->generation;
7913 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7914 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7916 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7920 sb_end_pagefault(inode->i_sb);
7921 return VM_FAULT_LOCKED;
7925 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7927 sb_end_pagefault(inode->i_sb);
7931 static int btrfs_truncate(struct inode *inode)
7933 struct btrfs_root *root = BTRFS_I(inode)->root;
7934 struct btrfs_block_rsv *rsv;
7937 struct btrfs_trans_handle *trans;
7938 u64 mask = root->sectorsize - 1;
7939 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7941 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
7947 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
7948 * 3 things going on here
7950 * 1) We need to reserve space for our orphan item and the space to
7951 * delete our orphan item. Lord knows we don't want to have a dangling
7952 * orphan item because we didn't reserve space to remove it.
7954 * 2) We need to reserve space to update our inode.
7956 * 3) We need to have something to cache all the space that is going to
7957 * be free'd up by the truncate operation, but also have some slack
7958 * space reserved in case it uses space during the truncate (thank you
7959 * very much snapshotting).
7961 * And we need these to all be seperate. The fact is we can use alot of
7962 * space doing the truncate, and we have no earthly idea how much space
7963 * we will use, so we need the truncate reservation to be seperate so it
7964 * doesn't end up using space reserved for updating the inode or
7965 * removing the orphan item. We also need to be able to stop the
7966 * transaction and start a new one, which means we need to be able to
7967 * update the inode several times, and we have no idea of knowing how
7968 * many times that will be, so we can't just reserve 1 item for the
7969 * entirety of the opration, so that has to be done seperately as well.
7970 * Then there is the orphan item, which does indeed need to be held on
7971 * to for the whole operation, and we need nobody to touch this reserved
7972 * space except the orphan code.
7974 * So that leaves us with
7976 * 1) root->orphan_block_rsv - for the orphan deletion.
7977 * 2) rsv - for the truncate reservation, which we will steal from the
7978 * transaction reservation.
7979 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7980 * updating the inode.
7982 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7985 rsv->size = min_size;
7989 * 1 for the truncate slack space
7990 * 1 for updating the inode.
7992 trans = btrfs_start_transaction(root, 2);
7993 if (IS_ERR(trans)) {
7994 err = PTR_ERR(trans);
7998 /* Migrate the slack space for the truncate to our reserve */
7999 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
8004 * So if we truncate and then write and fsync we normally would just
8005 * write the extents that changed, which is a problem if we need to
8006 * first truncate that entire inode. So set this flag so we write out
8007 * all of the extents in the inode to the sync log so we're completely
8010 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8011 trans->block_rsv = rsv;
8014 ret = btrfs_truncate_inode_items(trans, root, inode,
8016 BTRFS_EXTENT_DATA_KEY);
8017 if (ret != -ENOSPC) {
8022 trans->block_rsv = &root->fs_info->trans_block_rsv;
8023 ret = btrfs_update_inode(trans, root, inode);
8029 btrfs_end_transaction(trans, root);
8030 btrfs_btree_balance_dirty(root);
8032 trans = btrfs_start_transaction(root, 2);
8033 if (IS_ERR(trans)) {
8034 ret = err = PTR_ERR(trans);
8039 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
8041 BUG_ON(ret); /* shouldn't happen */
8042 trans->block_rsv = rsv;
8045 if (ret == 0 && inode->i_nlink > 0) {
8046 trans->block_rsv = root->orphan_block_rsv;
8047 ret = btrfs_orphan_del(trans, inode);
8053 trans->block_rsv = &root->fs_info->trans_block_rsv;
8054 ret = btrfs_update_inode(trans, root, inode);
8058 ret = btrfs_end_transaction(trans, root);
8059 btrfs_btree_balance_dirty(root);
8063 btrfs_free_block_rsv(root, rsv);
8072 * create a new subvolume directory/inode (helper for the ioctl).
8074 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8075 struct btrfs_root *new_root,
8076 struct btrfs_root *parent_root,
8079 struct inode *inode;
8083 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8084 new_dirid, new_dirid,
8085 S_IFDIR | (~current_umask() & S_IRWXUGO),
8088 return PTR_ERR(inode);
8089 inode->i_op = &btrfs_dir_inode_operations;
8090 inode->i_fop = &btrfs_dir_file_operations;
8092 set_nlink(inode, 1);
8093 btrfs_i_size_write(inode, 0);
8095 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8097 btrfs_err(new_root->fs_info,
8098 "error inheriting subvolume %llu properties: %d",
8099 new_root->root_key.objectid, err);
8101 err = btrfs_update_inode(trans, new_root, inode);
8107 struct inode *btrfs_alloc_inode(struct super_block *sb)
8109 struct btrfs_inode *ei;
8110 struct inode *inode;
8112 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
8119 ei->last_sub_trans = 0;
8120 ei->logged_trans = 0;
8121 ei->delalloc_bytes = 0;
8122 ei->disk_i_size = 0;
8125 ei->index_cnt = (u64)-1;
8127 ei->last_unlink_trans = 0;
8128 ei->last_log_commit = 0;
8130 spin_lock_init(&ei->lock);
8131 ei->outstanding_extents = 0;
8132 ei->reserved_extents = 0;
8134 ei->runtime_flags = 0;
8135 ei->force_compress = BTRFS_COMPRESS_NONE;
8137 ei->delayed_node = NULL;
8139 inode = &ei->vfs_inode;
8140 extent_map_tree_init(&ei->extent_tree);
8141 extent_io_tree_init(&ei->io_tree, &inode->i_data);
8142 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
8143 ei->io_tree.track_uptodate = 1;
8144 ei->io_failure_tree.track_uptodate = 1;
8145 atomic_set(&ei->sync_writers, 0);
8146 mutex_init(&ei->log_mutex);
8147 mutex_init(&ei->delalloc_mutex);
8148 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8149 INIT_LIST_HEAD(&ei->delalloc_inodes);
8150 RB_CLEAR_NODE(&ei->rb_node);
8155 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8156 void btrfs_test_destroy_inode(struct inode *inode)
8158 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8159 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8163 static void btrfs_i_callback(struct rcu_head *head)
8165 struct inode *inode = container_of(head, struct inode, i_rcu);
8166 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8169 void btrfs_destroy_inode(struct inode *inode)
8171 struct btrfs_ordered_extent *ordered;
8172 struct btrfs_root *root = BTRFS_I(inode)->root;
8174 WARN_ON(!hlist_empty(&inode->i_dentry));
8175 WARN_ON(inode->i_data.nrpages);
8176 WARN_ON(BTRFS_I(inode)->outstanding_extents);
8177 WARN_ON(BTRFS_I(inode)->reserved_extents);
8178 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8179 WARN_ON(BTRFS_I(inode)->csum_bytes);
8182 * This can happen where we create an inode, but somebody else also
8183 * created the same inode and we need to destroy the one we already
8189 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8190 &BTRFS_I(inode)->runtime_flags)) {
8191 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
8193 atomic_dec(&root->orphan_inodes);
8197 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8201 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
8202 ordered->file_offset, ordered->len);
8203 btrfs_remove_ordered_extent(inode, ordered);
8204 btrfs_put_ordered_extent(ordered);
8205 btrfs_put_ordered_extent(ordered);
8208 inode_tree_del(inode);
8209 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8211 call_rcu(&inode->i_rcu, btrfs_i_callback);
8214 int btrfs_drop_inode(struct inode *inode)
8216 struct btrfs_root *root = BTRFS_I(inode)->root;
8221 /* the snap/subvol tree is on deleting */
8222 if (btrfs_root_refs(&root->root_item) == 0)
8225 return generic_drop_inode(inode);
8228 static void init_once(void *foo)
8230 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8232 inode_init_once(&ei->vfs_inode);
8235 void btrfs_destroy_cachep(void)
8238 * Make sure all delayed rcu free inodes are flushed before we
8242 if (btrfs_inode_cachep)
8243 kmem_cache_destroy(btrfs_inode_cachep);
8244 if (btrfs_trans_handle_cachep)
8245 kmem_cache_destroy(btrfs_trans_handle_cachep);
8246 if (btrfs_transaction_cachep)
8247 kmem_cache_destroy(btrfs_transaction_cachep);
8248 if (btrfs_path_cachep)
8249 kmem_cache_destroy(btrfs_path_cachep);
8250 if (btrfs_free_space_cachep)
8251 kmem_cache_destroy(btrfs_free_space_cachep);
8252 if (btrfs_delalloc_work_cachep)
8253 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8256 int btrfs_init_cachep(void)
8258 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8259 sizeof(struct btrfs_inode), 0,
8260 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8261 if (!btrfs_inode_cachep)
8264 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8265 sizeof(struct btrfs_trans_handle), 0,
8266 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8267 if (!btrfs_trans_handle_cachep)
8270 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8271 sizeof(struct btrfs_transaction), 0,
8272 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8273 if (!btrfs_transaction_cachep)
8276 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8277 sizeof(struct btrfs_path), 0,
8278 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8279 if (!btrfs_path_cachep)
8282 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8283 sizeof(struct btrfs_free_space), 0,
8284 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8285 if (!btrfs_free_space_cachep)
8288 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8289 sizeof(struct btrfs_delalloc_work), 0,
8290 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8292 if (!btrfs_delalloc_work_cachep)
8297 btrfs_destroy_cachep();
8301 static int btrfs_getattr(struct vfsmount *mnt,
8302 struct dentry *dentry, struct kstat *stat)
8305 struct inode *inode = dentry->d_inode;
8306 u32 blocksize = inode->i_sb->s_blocksize;
8308 generic_fillattr(inode, stat);
8309 stat->dev = BTRFS_I(inode)->root->anon_dev;
8310 stat->blksize = PAGE_CACHE_SIZE;
8312 spin_lock(&BTRFS_I(inode)->lock);
8313 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8314 spin_unlock(&BTRFS_I(inode)->lock);
8315 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8316 ALIGN(delalloc_bytes, blocksize)) >> 9;
8320 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8321 struct inode *new_dir, struct dentry *new_dentry)
8323 struct btrfs_trans_handle *trans;
8324 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8325 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8326 struct inode *new_inode = new_dentry->d_inode;
8327 struct inode *old_inode = old_dentry->d_inode;
8328 struct timespec ctime = CURRENT_TIME;
8332 u64 old_ino = btrfs_ino(old_inode);
8334 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8337 /* we only allow rename subvolume link between subvolumes */
8338 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8341 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8342 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8345 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8346 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8350 /* check for collisions, even if the name isn't there */
8351 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
8352 new_dentry->d_name.name,
8353 new_dentry->d_name.len);
8356 if (ret == -EEXIST) {
8358 * eexist without a new_inode */
8359 if (WARN_ON(!new_inode)) {
8363 /* maybe -EOVERFLOW */
8370 * we're using rename to replace one file with another. Start IO on it
8371 * now so we don't add too much work to the end of the transaction
8373 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
8374 filemap_flush(old_inode->i_mapping);
8376 /* close the racy window with snapshot create/destroy ioctl */
8377 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8378 down_read(&root->fs_info->subvol_sem);
8380 * We want to reserve the absolute worst case amount of items. So if
8381 * both inodes are subvols and we need to unlink them then that would
8382 * require 4 item modifications, but if they are both normal inodes it
8383 * would require 5 item modifications, so we'll assume their normal
8384 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8385 * should cover the worst case number of items we'll modify.
8387 trans = btrfs_start_transaction(root, 11);
8388 if (IS_ERR(trans)) {
8389 ret = PTR_ERR(trans);
8394 btrfs_record_root_in_trans(trans, dest);
8396 ret = btrfs_set_inode_index(new_dir, &index);
8400 BTRFS_I(old_inode)->dir_index = 0ULL;
8401 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8402 /* force full log commit if subvolume involved. */
8403 btrfs_set_log_full_commit(root->fs_info, trans);
8405 ret = btrfs_insert_inode_ref(trans, dest,
8406 new_dentry->d_name.name,
8407 new_dentry->d_name.len,
8409 btrfs_ino(new_dir), index);
8413 * this is an ugly little race, but the rename is required
8414 * to make sure that if we crash, the inode is either at the
8415 * old name or the new one. pinning the log transaction lets
8416 * us make sure we don't allow a log commit to come in after
8417 * we unlink the name but before we add the new name back in.
8419 btrfs_pin_log_trans(root);
8422 inode_inc_iversion(old_dir);
8423 inode_inc_iversion(new_dir);
8424 inode_inc_iversion(old_inode);
8425 old_dir->i_ctime = old_dir->i_mtime = ctime;
8426 new_dir->i_ctime = new_dir->i_mtime = ctime;
8427 old_inode->i_ctime = ctime;
8429 if (old_dentry->d_parent != new_dentry->d_parent)
8430 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8432 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8433 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8434 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8435 old_dentry->d_name.name,
8436 old_dentry->d_name.len);
8438 ret = __btrfs_unlink_inode(trans, root, old_dir,
8439 old_dentry->d_inode,
8440 old_dentry->d_name.name,
8441 old_dentry->d_name.len);
8443 ret = btrfs_update_inode(trans, root, old_inode);
8446 btrfs_abort_transaction(trans, root, ret);
8451 inode_inc_iversion(new_inode);
8452 new_inode->i_ctime = CURRENT_TIME;
8453 if (unlikely(btrfs_ino(new_inode) ==
8454 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8455 root_objectid = BTRFS_I(new_inode)->location.objectid;
8456 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8458 new_dentry->d_name.name,
8459 new_dentry->d_name.len);
8460 BUG_ON(new_inode->i_nlink == 0);
8462 ret = btrfs_unlink_inode(trans, dest, new_dir,
8463 new_dentry->d_inode,
8464 new_dentry->d_name.name,
8465 new_dentry->d_name.len);
8467 if (!ret && new_inode->i_nlink == 0)
8468 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8470 btrfs_abort_transaction(trans, root, ret);
8475 ret = btrfs_add_link(trans, new_dir, old_inode,
8476 new_dentry->d_name.name,
8477 new_dentry->d_name.len, 0, index);
8479 btrfs_abort_transaction(trans, root, ret);
8483 if (old_inode->i_nlink == 1)
8484 BTRFS_I(old_inode)->dir_index = index;
8486 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8487 struct dentry *parent = new_dentry->d_parent;
8488 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8489 btrfs_end_log_trans(root);
8492 btrfs_end_transaction(trans, root);
8494 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8495 up_read(&root->fs_info->subvol_sem);
8500 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8502 struct btrfs_delalloc_work *delalloc_work;
8503 struct inode *inode;
8505 delalloc_work = container_of(work, struct btrfs_delalloc_work,
8507 inode = delalloc_work->inode;
8508 if (delalloc_work->wait) {
8509 btrfs_wait_ordered_range(inode, 0, (u64)-1);
8511 filemap_flush(inode->i_mapping);
8512 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8513 &BTRFS_I(inode)->runtime_flags))
8514 filemap_flush(inode->i_mapping);
8517 if (delalloc_work->delay_iput)
8518 btrfs_add_delayed_iput(inode);
8521 complete(&delalloc_work->completion);
8524 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8525 int wait, int delay_iput)
8527 struct btrfs_delalloc_work *work;
8529 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8533 init_completion(&work->completion);
8534 INIT_LIST_HEAD(&work->list);
8535 work->inode = inode;
8537 work->delay_iput = delay_iput;
8538 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
8543 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8545 wait_for_completion(&work->completion);
8546 kmem_cache_free(btrfs_delalloc_work_cachep, work);
8550 * some fairly slow code that needs optimization. This walks the list
8551 * of all the inodes with pending delalloc and forces them to disk.
8553 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
8556 struct btrfs_inode *binode;
8557 struct inode *inode;
8558 struct btrfs_delalloc_work *work, *next;
8559 struct list_head works;
8560 struct list_head splice;
8563 INIT_LIST_HEAD(&works);
8564 INIT_LIST_HEAD(&splice);
8566 mutex_lock(&root->delalloc_mutex);
8567 spin_lock(&root->delalloc_lock);
8568 list_splice_init(&root->delalloc_inodes, &splice);
8569 while (!list_empty(&splice)) {
8570 binode = list_entry(splice.next, struct btrfs_inode,
8573 list_move_tail(&binode->delalloc_inodes,
8574 &root->delalloc_inodes);
8575 inode = igrab(&binode->vfs_inode);
8577 cond_resched_lock(&root->delalloc_lock);
8580 spin_unlock(&root->delalloc_lock);
8582 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8583 if (unlikely(!work)) {
8585 btrfs_add_delayed_iput(inode);
8591 list_add_tail(&work->list, &works);
8592 btrfs_queue_work(root->fs_info->flush_workers,
8595 if (nr != -1 && ret >= nr)
8598 spin_lock(&root->delalloc_lock);
8600 spin_unlock(&root->delalloc_lock);
8603 list_for_each_entry_safe(work, next, &works, list) {
8604 list_del_init(&work->list);
8605 btrfs_wait_and_free_delalloc_work(work);
8608 if (!list_empty_careful(&splice)) {
8609 spin_lock(&root->delalloc_lock);
8610 list_splice_tail(&splice, &root->delalloc_inodes);
8611 spin_unlock(&root->delalloc_lock);
8613 mutex_unlock(&root->delalloc_mutex);
8617 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8621 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
8624 ret = __start_delalloc_inodes(root, delay_iput, -1);
8628 * the filemap_flush will queue IO into the worker threads, but
8629 * we have to make sure the IO is actually started and that
8630 * ordered extents get created before we return
8632 atomic_inc(&root->fs_info->async_submit_draining);
8633 while (atomic_read(&root->fs_info->nr_async_submits) ||
8634 atomic_read(&root->fs_info->async_delalloc_pages)) {
8635 wait_event(root->fs_info->async_submit_wait,
8636 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8637 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8639 atomic_dec(&root->fs_info->async_submit_draining);
8643 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
8646 struct btrfs_root *root;
8647 struct list_head splice;
8650 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
8653 INIT_LIST_HEAD(&splice);
8655 mutex_lock(&fs_info->delalloc_root_mutex);
8656 spin_lock(&fs_info->delalloc_root_lock);
8657 list_splice_init(&fs_info->delalloc_roots, &splice);
8658 while (!list_empty(&splice) && nr) {
8659 root = list_first_entry(&splice, struct btrfs_root,
8661 root = btrfs_grab_fs_root(root);
8663 list_move_tail(&root->delalloc_root,
8664 &fs_info->delalloc_roots);
8665 spin_unlock(&fs_info->delalloc_root_lock);
8667 ret = __start_delalloc_inodes(root, delay_iput, nr);
8668 btrfs_put_fs_root(root);
8676 spin_lock(&fs_info->delalloc_root_lock);
8678 spin_unlock(&fs_info->delalloc_root_lock);
8681 atomic_inc(&fs_info->async_submit_draining);
8682 while (atomic_read(&fs_info->nr_async_submits) ||
8683 atomic_read(&fs_info->async_delalloc_pages)) {
8684 wait_event(fs_info->async_submit_wait,
8685 (atomic_read(&fs_info->nr_async_submits) == 0 &&
8686 atomic_read(&fs_info->async_delalloc_pages) == 0));
8688 atomic_dec(&fs_info->async_submit_draining);
8690 if (!list_empty_careful(&splice)) {
8691 spin_lock(&fs_info->delalloc_root_lock);
8692 list_splice_tail(&splice, &fs_info->delalloc_roots);
8693 spin_unlock(&fs_info->delalloc_root_lock);
8695 mutex_unlock(&fs_info->delalloc_root_mutex);
8699 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8700 const char *symname)
8702 struct btrfs_trans_handle *trans;
8703 struct btrfs_root *root = BTRFS_I(dir)->root;
8704 struct btrfs_path *path;
8705 struct btrfs_key key;
8706 struct inode *inode = NULL;
8714 struct btrfs_file_extent_item *ei;
8715 struct extent_buffer *leaf;
8717 name_len = strlen(symname);
8718 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8719 return -ENAMETOOLONG;
8722 * 2 items for inode item and ref
8723 * 2 items for dir items
8724 * 1 item for xattr if selinux is on
8726 trans = btrfs_start_transaction(root, 5);
8728 return PTR_ERR(trans);
8730 err = btrfs_find_free_ino(root, &objectid);
8734 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8735 dentry->d_name.len, btrfs_ino(dir), objectid,
8736 S_IFLNK|S_IRWXUGO, &index);
8737 if (IS_ERR(inode)) {
8738 err = PTR_ERR(inode);
8742 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8749 * If the active LSM wants to access the inode during
8750 * d_instantiate it needs these. Smack checks to see
8751 * if the filesystem supports xattrs by looking at the
8754 inode->i_fop = &btrfs_file_operations;
8755 inode->i_op = &btrfs_file_inode_operations;
8757 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8761 inode->i_mapping->a_ops = &btrfs_aops;
8762 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8763 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8768 path = btrfs_alloc_path();
8774 key.objectid = btrfs_ino(inode);
8776 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8777 datasize = btrfs_file_extent_calc_inline_size(name_len);
8778 err = btrfs_insert_empty_item(trans, root, path, &key,
8782 btrfs_free_path(path);
8785 leaf = path->nodes[0];
8786 ei = btrfs_item_ptr(leaf, path->slots[0],
8787 struct btrfs_file_extent_item);
8788 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8789 btrfs_set_file_extent_type(leaf, ei,
8790 BTRFS_FILE_EXTENT_INLINE);
8791 btrfs_set_file_extent_encryption(leaf, ei, 0);
8792 btrfs_set_file_extent_compression(leaf, ei, 0);
8793 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8794 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8796 ptr = btrfs_file_extent_inline_start(ei);
8797 write_extent_buffer(leaf, symname, ptr, name_len);
8798 btrfs_mark_buffer_dirty(leaf);
8799 btrfs_free_path(path);
8801 inode->i_op = &btrfs_symlink_inode_operations;
8802 inode->i_mapping->a_ops = &btrfs_symlink_aops;
8803 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8804 inode_set_bytes(inode, name_len);
8805 btrfs_i_size_write(inode, name_len);
8806 err = btrfs_update_inode(trans, root, inode);
8812 d_instantiate(dentry, inode);
8813 btrfs_end_transaction(trans, root);
8815 inode_dec_link_count(inode);
8818 btrfs_btree_balance_dirty(root);
8822 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8823 u64 start, u64 num_bytes, u64 min_size,
8824 loff_t actual_len, u64 *alloc_hint,
8825 struct btrfs_trans_handle *trans)
8827 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8828 struct extent_map *em;
8829 struct btrfs_root *root = BTRFS_I(inode)->root;
8830 struct btrfs_key ins;
8831 u64 cur_offset = start;
8835 bool own_trans = true;
8839 while (num_bytes > 0) {
8841 trans = btrfs_start_transaction(root, 3);
8842 if (IS_ERR(trans)) {
8843 ret = PTR_ERR(trans);
8848 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8849 cur_bytes = max(cur_bytes, min_size);
8850 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
8851 *alloc_hint, &ins, 1, 0);
8854 btrfs_end_transaction(trans, root);
8858 ret = insert_reserved_file_extent(trans, inode,
8859 cur_offset, ins.objectid,
8860 ins.offset, ins.offset,
8861 ins.offset, 0, 0, 0,
8862 BTRFS_FILE_EXTENT_PREALLOC);
8864 btrfs_free_reserved_extent(root, ins.objectid,
8866 btrfs_abort_transaction(trans, root, ret);
8868 btrfs_end_transaction(trans, root);
8871 btrfs_drop_extent_cache(inode, cur_offset,
8872 cur_offset + ins.offset -1, 0);
8874 em = alloc_extent_map();
8876 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8877 &BTRFS_I(inode)->runtime_flags);
8881 em->start = cur_offset;
8882 em->orig_start = cur_offset;
8883 em->len = ins.offset;
8884 em->block_start = ins.objectid;
8885 em->block_len = ins.offset;
8886 em->orig_block_len = ins.offset;
8887 em->ram_bytes = ins.offset;
8888 em->bdev = root->fs_info->fs_devices->latest_bdev;
8889 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8890 em->generation = trans->transid;
8893 write_lock(&em_tree->lock);
8894 ret = add_extent_mapping(em_tree, em, 1);
8895 write_unlock(&em_tree->lock);
8898 btrfs_drop_extent_cache(inode, cur_offset,
8899 cur_offset + ins.offset - 1,
8902 free_extent_map(em);
8904 num_bytes -= ins.offset;
8905 cur_offset += ins.offset;
8906 *alloc_hint = ins.objectid + ins.offset;
8908 inode_inc_iversion(inode);
8909 inode->i_ctime = CURRENT_TIME;
8910 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8911 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8912 (actual_len > inode->i_size) &&
8913 (cur_offset > inode->i_size)) {
8914 if (cur_offset > actual_len)
8915 i_size = actual_len;
8917 i_size = cur_offset;
8918 i_size_write(inode, i_size);
8919 btrfs_ordered_update_i_size(inode, i_size, NULL);
8922 ret = btrfs_update_inode(trans, root, inode);
8925 btrfs_abort_transaction(trans, root, ret);
8927 btrfs_end_transaction(trans, root);
8932 btrfs_end_transaction(trans, root);
8937 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8938 u64 start, u64 num_bytes, u64 min_size,
8939 loff_t actual_len, u64 *alloc_hint)
8941 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8942 min_size, actual_len, alloc_hint,
8946 int btrfs_prealloc_file_range_trans(struct inode *inode,
8947 struct btrfs_trans_handle *trans, int mode,
8948 u64 start, u64 num_bytes, u64 min_size,
8949 loff_t actual_len, u64 *alloc_hint)
8951 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8952 min_size, actual_len, alloc_hint, trans);
8955 static int btrfs_set_page_dirty(struct page *page)
8957 return __set_page_dirty_nobuffers(page);
8960 static int btrfs_permission(struct inode *inode, int mask)
8962 struct btrfs_root *root = BTRFS_I(inode)->root;
8963 umode_t mode = inode->i_mode;
8965 if (mask & MAY_WRITE &&
8966 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8967 if (btrfs_root_readonly(root))
8969 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8972 return generic_permission(inode, mask);
8975 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
8977 struct btrfs_trans_handle *trans;
8978 struct btrfs_root *root = BTRFS_I(dir)->root;
8979 struct inode *inode = NULL;
8985 * 5 units required for adding orphan entry
8987 trans = btrfs_start_transaction(root, 5);
8989 return PTR_ERR(trans);
8991 ret = btrfs_find_free_ino(root, &objectid);
8995 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
8996 btrfs_ino(dir), objectid, mode, &index);
8997 if (IS_ERR(inode)) {
8998 ret = PTR_ERR(inode);
9003 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9007 ret = btrfs_update_inode(trans, root, inode);
9011 inode->i_fop = &btrfs_file_operations;
9012 inode->i_op = &btrfs_file_inode_operations;
9014 inode->i_mapping->a_ops = &btrfs_aops;
9015 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
9016 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9018 ret = btrfs_orphan_add(trans, inode);
9023 * We set number of links to 0 in btrfs_new_inode(), and here we set
9024 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9027 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9029 set_nlink(inode, 1);
9030 d_tmpfile(dentry, inode);
9031 mark_inode_dirty(inode);
9034 btrfs_end_transaction(trans, root);
9037 btrfs_balance_delayed_items(root);
9038 btrfs_btree_balance_dirty(root);
9043 static const struct inode_operations btrfs_dir_inode_operations = {
9044 .getattr = btrfs_getattr,
9045 .lookup = btrfs_lookup,
9046 .create = btrfs_create,
9047 .unlink = btrfs_unlink,
9049 .mkdir = btrfs_mkdir,
9050 .rmdir = btrfs_rmdir,
9051 .rename = btrfs_rename,
9052 .symlink = btrfs_symlink,
9053 .setattr = btrfs_setattr,
9054 .mknod = btrfs_mknod,
9055 .setxattr = btrfs_setxattr,
9056 .getxattr = btrfs_getxattr,
9057 .listxattr = btrfs_listxattr,
9058 .removexattr = btrfs_removexattr,
9059 .permission = btrfs_permission,
9060 .get_acl = btrfs_get_acl,
9061 .set_acl = btrfs_set_acl,
9062 .update_time = btrfs_update_time,
9063 .tmpfile = btrfs_tmpfile,
9065 static const struct inode_operations btrfs_dir_ro_inode_operations = {
9066 .lookup = btrfs_lookup,
9067 .permission = btrfs_permission,
9068 .get_acl = btrfs_get_acl,
9069 .set_acl = btrfs_set_acl,
9070 .update_time = btrfs_update_time,
9073 static const struct file_operations btrfs_dir_file_operations = {
9074 .llseek = generic_file_llseek,
9075 .read = generic_read_dir,
9076 .iterate = btrfs_real_readdir,
9077 .unlocked_ioctl = btrfs_ioctl,
9078 #ifdef CONFIG_COMPAT
9079 .compat_ioctl = btrfs_ioctl,
9081 .release = btrfs_release_file,
9082 .fsync = btrfs_sync_file,
9085 static struct extent_io_ops btrfs_extent_io_ops = {
9086 .fill_delalloc = run_delalloc_range,
9087 .submit_bio_hook = btrfs_submit_bio_hook,
9088 .merge_bio_hook = btrfs_merge_bio_hook,
9089 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
9090 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
9091 .writepage_start_hook = btrfs_writepage_start_hook,
9092 .set_bit_hook = btrfs_set_bit_hook,
9093 .clear_bit_hook = btrfs_clear_bit_hook,
9094 .merge_extent_hook = btrfs_merge_extent_hook,
9095 .split_extent_hook = btrfs_split_extent_hook,
9099 * btrfs doesn't support the bmap operation because swapfiles
9100 * use bmap to make a mapping of extents in the file. They assume
9101 * these extents won't change over the life of the file and they
9102 * use the bmap result to do IO directly to the drive.
9104 * the btrfs bmap call would return logical addresses that aren't
9105 * suitable for IO and they also will change frequently as COW
9106 * operations happen. So, swapfile + btrfs == corruption.
9108 * For now we're avoiding this by dropping bmap.
9110 static const struct address_space_operations btrfs_aops = {
9111 .readpage = btrfs_readpage,
9112 .writepage = btrfs_writepage,
9113 .writepages = btrfs_writepages,
9114 .readpages = btrfs_readpages,
9115 .direct_IO = btrfs_direct_IO,
9116 .invalidatepage = btrfs_invalidatepage,
9117 .releasepage = btrfs_releasepage,
9118 .set_page_dirty = btrfs_set_page_dirty,
9119 .error_remove_page = generic_error_remove_page,
9122 static const struct address_space_operations btrfs_symlink_aops = {
9123 .readpage = btrfs_readpage,
9124 .writepage = btrfs_writepage,
9125 .invalidatepage = btrfs_invalidatepage,
9126 .releasepage = btrfs_releasepage,
9129 static const struct inode_operations btrfs_file_inode_operations = {
9130 .getattr = btrfs_getattr,
9131 .setattr = btrfs_setattr,
9132 .setxattr = btrfs_setxattr,
9133 .getxattr = btrfs_getxattr,
9134 .listxattr = btrfs_listxattr,
9135 .removexattr = btrfs_removexattr,
9136 .permission = btrfs_permission,
9137 .fiemap = btrfs_fiemap,
9138 .get_acl = btrfs_get_acl,
9139 .set_acl = btrfs_set_acl,
9140 .update_time = btrfs_update_time,
9142 static const struct inode_operations btrfs_special_inode_operations = {
9143 .getattr = btrfs_getattr,
9144 .setattr = btrfs_setattr,
9145 .permission = btrfs_permission,
9146 .setxattr = btrfs_setxattr,
9147 .getxattr = btrfs_getxattr,
9148 .listxattr = btrfs_listxattr,
9149 .removexattr = btrfs_removexattr,
9150 .get_acl = btrfs_get_acl,
9151 .set_acl = btrfs_set_acl,
9152 .update_time = btrfs_update_time,
9154 static const struct inode_operations btrfs_symlink_inode_operations = {
9155 .readlink = generic_readlink,
9156 .follow_link = page_follow_link_light,
9157 .put_link = page_put_link,
9158 .getattr = btrfs_getattr,
9159 .setattr = btrfs_setattr,
9160 .permission = btrfs_permission,
9161 .setxattr = btrfs_setxattr,
9162 .getxattr = btrfs_getxattr,
9163 .listxattr = btrfs_listxattr,
9164 .removexattr = btrfs_removexattr,
9165 .update_time = btrfs_update_time,
9168 const struct dentry_operations btrfs_dentry_operations = {
9169 .d_delete = btrfs_dentry_delete,
9170 .d_release = btrfs_dentry_release,