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 noinline 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);
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);
718 * here we're doing allocation and writeback of the
721 btrfs_drop_extent_cache(inode, async_extent->start,
722 async_extent->start +
723 async_extent->ram_size - 1, 0);
725 em = alloc_extent_map();
728 goto out_free_reserve;
730 em->start = async_extent->start;
731 em->len = async_extent->ram_size;
732 em->orig_start = em->start;
733 em->mod_start = em->start;
734 em->mod_len = em->len;
736 em->block_start = ins.objectid;
737 em->block_len = ins.offset;
738 em->orig_block_len = ins.offset;
739 em->ram_bytes = async_extent->ram_size;
740 em->bdev = root->fs_info->fs_devices->latest_bdev;
741 em->compress_type = async_extent->compress_type;
742 set_bit(EXTENT_FLAG_PINNED, &em->flags);
743 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
747 write_lock(&em_tree->lock);
748 ret = add_extent_mapping(em_tree, em, 1);
749 write_unlock(&em_tree->lock);
750 if (ret != -EEXIST) {
754 btrfs_drop_extent_cache(inode, async_extent->start,
755 async_extent->start +
756 async_extent->ram_size - 1, 0);
760 goto out_free_reserve;
762 ret = btrfs_add_ordered_extent_compress(inode,
765 async_extent->ram_size,
767 BTRFS_ORDERED_COMPRESSED,
768 async_extent->compress_type);
770 goto out_free_reserve;
773 * clear dirty, set writeback and unlock the pages.
775 extent_clear_unlock_delalloc(inode, async_extent->start,
776 async_extent->start +
777 async_extent->ram_size - 1,
778 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
779 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
781 ret = btrfs_submit_compressed_write(inode,
783 async_extent->ram_size,
785 ins.offset, async_extent->pages,
786 async_extent->nr_pages);
787 alloc_hint = ins.objectid + ins.offset;
797 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
799 extent_clear_unlock_delalloc(inode, async_extent->start,
800 async_extent->start +
801 async_extent->ram_size - 1,
802 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
803 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
804 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
805 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
810 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
813 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
814 struct extent_map *em;
817 read_lock(&em_tree->lock);
818 em = search_extent_mapping(em_tree, start, num_bytes);
821 * if block start isn't an actual block number then find the
822 * first block in this inode and use that as a hint. If that
823 * block is also bogus then just don't worry about it.
825 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
827 em = search_extent_mapping(em_tree, 0, 0);
828 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
829 alloc_hint = em->block_start;
833 alloc_hint = em->block_start;
837 read_unlock(&em_tree->lock);
843 * when extent_io.c finds a delayed allocation range in the file,
844 * the call backs end up in this code. The basic idea is to
845 * allocate extents on disk for the range, and create ordered data structs
846 * in ram to track those extents.
848 * locked_page is the page that writepage had locked already. We use
849 * it to make sure we don't do extra locks or unlocks.
851 * *page_started is set to one if we unlock locked_page and do everything
852 * required to start IO on it. It may be clean and already done with
855 static noinline int cow_file_range(struct inode *inode,
856 struct page *locked_page,
857 u64 start, u64 end, int *page_started,
858 unsigned long *nr_written,
861 struct btrfs_root *root = BTRFS_I(inode)->root;
864 unsigned long ram_size;
867 u64 blocksize = root->sectorsize;
868 struct btrfs_key ins;
869 struct extent_map *em;
870 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
873 if (btrfs_is_free_space_inode(inode)) {
879 num_bytes = ALIGN(end - start + 1, blocksize);
880 num_bytes = max(blocksize, num_bytes);
881 disk_num_bytes = num_bytes;
883 /* if this is a small write inside eof, kick off defrag */
884 if (num_bytes < 64 * 1024 &&
885 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
886 btrfs_add_inode_defrag(NULL, inode);
889 /* lets try to make an inline extent */
890 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
893 extent_clear_unlock_delalloc(inode, start, end, NULL,
894 EXTENT_LOCKED | EXTENT_DELALLOC |
895 EXTENT_DEFRAG, PAGE_UNLOCK |
896 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
899 *nr_written = *nr_written +
900 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
903 } else if (ret < 0) {
908 BUG_ON(disk_num_bytes >
909 btrfs_super_total_bytes(root->fs_info->super_copy));
911 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
912 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
914 while (disk_num_bytes > 0) {
917 cur_alloc_size = disk_num_bytes;
918 ret = btrfs_reserve_extent(root, cur_alloc_size,
919 root->sectorsize, 0, alloc_hint,
924 em = alloc_extent_map();
930 em->orig_start = em->start;
931 ram_size = ins.offset;
932 em->len = ins.offset;
933 em->mod_start = em->start;
934 em->mod_len = em->len;
936 em->block_start = ins.objectid;
937 em->block_len = ins.offset;
938 em->orig_block_len = ins.offset;
939 em->ram_bytes = ram_size;
940 em->bdev = root->fs_info->fs_devices->latest_bdev;
941 set_bit(EXTENT_FLAG_PINNED, &em->flags);
945 write_lock(&em_tree->lock);
946 ret = add_extent_mapping(em_tree, em, 1);
947 write_unlock(&em_tree->lock);
948 if (ret != -EEXIST) {
952 btrfs_drop_extent_cache(inode, start,
953 start + ram_size - 1, 0);
958 cur_alloc_size = ins.offset;
959 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
960 ram_size, cur_alloc_size, 0);
964 if (root->root_key.objectid ==
965 BTRFS_DATA_RELOC_TREE_OBJECTID) {
966 ret = btrfs_reloc_clone_csums(inode, start,
972 if (disk_num_bytes < cur_alloc_size)
975 /* we're not doing compressed IO, don't unlock the first
976 * page (which the caller expects to stay locked), don't
977 * clear any dirty bits and don't set any writeback bits
979 * Do set the Private2 bit so we know this page was properly
980 * setup for writepage
982 op = unlock ? PAGE_UNLOCK : 0;
983 op |= PAGE_SET_PRIVATE2;
985 extent_clear_unlock_delalloc(inode, start,
986 start + ram_size - 1, locked_page,
987 EXTENT_LOCKED | EXTENT_DELALLOC,
989 disk_num_bytes -= cur_alloc_size;
990 num_bytes -= cur_alloc_size;
991 alloc_hint = ins.objectid + ins.offset;
992 start += cur_alloc_size;
998 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
1000 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1001 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1002 EXTENT_DELALLOC | EXTENT_DEFRAG,
1003 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1004 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1009 * work queue call back to started compression on a file and pages
1011 static noinline void async_cow_start(struct btrfs_work *work)
1013 struct async_cow *async_cow;
1015 async_cow = container_of(work, struct async_cow, work);
1017 compress_file_range(async_cow->inode, async_cow->locked_page,
1018 async_cow->start, async_cow->end, async_cow,
1020 if (num_added == 0) {
1021 btrfs_add_delayed_iput(async_cow->inode);
1022 async_cow->inode = NULL;
1027 * work queue call back to submit previously compressed pages
1029 static noinline void async_cow_submit(struct btrfs_work *work)
1031 struct async_cow *async_cow;
1032 struct btrfs_root *root;
1033 unsigned long nr_pages;
1035 async_cow = container_of(work, struct async_cow, work);
1037 root = async_cow->root;
1038 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1041 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1043 waitqueue_active(&root->fs_info->async_submit_wait))
1044 wake_up(&root->fs_info->async_submit_wait);
1046 if (async_cow->inode)
1047 submit_compressed_extents(async_cow->inode, async_cow);
1050 static noinline void async_cow_free(struct btrfs_work *work)
1052 struct async_cow *async_cow;
1053 async_cow = container_of(work, struct async_cow, work);
1054 if (async_cow->inode)
1055 btrfs_add_delayed_iput(async_cow->inode);
1059 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1060 u64 start, u64 end, int *page_started,
1061 unsigned long *nr_written)
1063 struct async_cow *async_cow;
1064 struct btrfs_root *root = BTRFS_I(inode)->root;
1065 unsigned long nr_pages;
1067 int limit = 10 * 1024 * 1024;
1069 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1070 1, 0, NULL, GFP_NOFS);
1071 while (start < end) {
1072 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1073 BUG_ON(!async_cow); /* -ENOMEM */
1074 async_cow->inode = igrab(inode);
1075 async_cow->root = root;
1076 async_cow->locked_page = locked_page;
1077 async_cow->start = start;
1079 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1082 cur_end = min(end, start + 512 * 1024 - 1);
1084 async_cow->end = cur_end;
1085 INIT_LIST_HEAD(&async_cow->extents);
1087 btrfs_init_work(&async_cow->work, async_cow_start,
1088 async_cow_submit, async_cow_free);
1090 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1092 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1094 btrfs_queue_work(root->fs_info->delalloc_workers,
1097 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1098 wait_event(root->fs_info->async_submit_wait,
1099 (atomic_read(&root->fs_info->async_delalloc_pages) <
1103 while (atomic_read(&root->fs_info->async_submit_draining) &&
1104 atomic_read(&root->fs_info->async_delalloc_pages)) {
1105 wait_event(root->fs_info->async_submit_wait,
1106 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1110 *nr_written += nr_pages;
1111 start = cur_end + 1;
1117 static noinline int csum_exist_in_range(struct btrfs_root *root,
1118 u64 bytenr, u64 num_bytes)
1121 struct btrfs_ordered_sum *sums;
1124 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1125 bytenr + num_bytes - 1, &list, 0);
1126 if (ret == 0 && list_empty(&list))
1129 while (!list_empty(&list)) {
1130 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1131 list_del(&sums->list);
1138 * when nowcow writeback call back. This checks for snapshots or COW copies
1139 * of the extents that exist in the file, and COWs the file as required.
1141 * If no cow copies or snapshots exist, we write directly to the existing
1144 static noinline int run_delalloc_nocow(struct inode *inode,
1145 struct page *locked_page,
1146 u64 start, u64 end, int *page_started, int force,
1147 unsigned long *nr_written)
1149 struct btrfs_root *root = BTRFS_I(inode)->root;
1150 struct btrfs_trans_handle *trans;
1151 struct extent_buffer *leaf;
1152 struct btrfs_path *path;
1153 struct btrfs_file_extent_item *fi;
1154 struct btrfs_key found_key;
1169 u64 ino = btrfs_ino(inode);
1171 path = btrfs_alloc_path();
1173 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1174 EXTENT_LOCKED | EXTENT_DELALLOC |
1175 EXTENT_DO_ACCOUNTING |
1176 EXTENT_DEFRAG, PAGE_UNLOCK |
1178 PAGE_SET_WRITEBACK |
1179 PAGE_END_WRITEBACK);
1183 nolock = btrfs_is_free_space_inode(inode);
1186 trans = btrfs_join_transaction_nolock(root);
1188 trans = btrfs_join_transaction(root);
1190 if (IS_ERR(trans)) {
1191 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1192 EXTENT_LOCKED | EXTENT_DELALLOC |
1193 EXTENT_DO_ACCOUNTING |
1194 EXTENT_DEFRAG, PAGE_UNLOCK |
1196 PAGE_SET_WRITEBACK |
1197 PAGE_END_WRITEBACK);
1198 btrfs_free_path(path);
1199 return PTR_ERR(trans);
1202 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1204 cow_start = (u64)-1;
1207 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1211 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1212 leaf = path->nodes[0];
1213 btrfs_item_key_to_cpu(leaf, &found_key,
1214 path->slots[0] - 1);
1215 if (found_key.objectid == ino &&
1216 found_key.type == BTRFS_EXTENT_DATA_KEY)
1221 leaf = path->nodes[0];
1222 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1223 ret = btrfs_next_leaf(root, path);
1228 leaf = path->nodes[0];
1234 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1236 if (found_key.objectid > ino ||
1237 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1238 found_key.offset > end)
1241 if (found_key.offset > cur_offset) {
1242 extent_end = found_key.offset;
1247 fi = btrfs_item_ptr(leaf, path->slots[0],
1248 struct btrfs_file_extent_item);
1249 extent_type = btrfs_file_extent_type(leaf, fi);
1251 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1252 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1253 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1254 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1255 extent_offset = btrfs_file_extent_offset(leaf, fi);
1256 extent_end = found_key.offset +
1257 btrfs_file_extent_num_bytes(leaf, fi);
1259 btrfs_file_extent_disk_num_bytes(leaf, fi);
1260 if (extent_end <= start) {
1264 if (disk_bytenr == 0)
1266 if (btrfs_file_extent_compression(leaf, fi) ||
1267 btrfs_file_extent_encryption(leaf, fi) ||
1268 btrfs_file_extent_other_encoding(leaf, fi))
1270 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1272 if (btrfs_extent_readonly(root, disk_bytenr))
1274 if (btrfs_cross_ref_exist(trans, root, ino,
1276 extent_offset, disk_bytenr))
1278 disk_bytenr += extent_offset;
1279 disk_bytenr += cur_offset - found_key.offset;
1280 num_bytes = min(end + 1, extent_end) - cur_offset;
1282 * if there are pending snapshots for this root,
1283 * we fall into common COW way.
1286 err = btrfs_start_nocow_write(root);
1291 * force cow if csum exists in the range.
1292 * this ensure that csum for a given extent are
1293 * either valid or do not exist.
1295 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1298 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1299 extent_end = found_key.offset +
1300 btrfs_file_extent_inline_len(leaf,
1301 path->slots[0], fi);
1302 extent_end = ALIGN(extent_end, root->sectorsize);
1307 if (extent_end <= start) {
1309 if (!nolock && nocow)
1310 btrfs_end_nocow_write(root);
1314 if (cow_start == (u64)-1)
1315 cow_start = cur_offset;
1316 cur_offset = extent_end;
1317 if (cur_offset > end)
1323 btrfs_release_path(path);
1324 if (cow_start != (u64)-1) {
1325 ret = cow_file_range(inode, locked_page,
1326 cow_start, found_key.offset - 1,
1327 page_started, nr_written, 1);
1329 if (!nolock && nocow)
1330 btrfs_end_nocow_write(root);
1333 cow_start = (u64)-1;
1336 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1337 struct extent_map *em;
1338 struct extent_map_tree *em_tree;
1339 em_tree = &BTRFS_I(inode)->extent_tree;
1340 em = alloc_extent_map();
1341 BUG_ON(!em); /* -ENOMEM */
1342 em->start = cur_offset;
1343 em->orig_start = found_key.offset - extent_offset;
1344 em->len = num_bytes;
1345 em->block_len = num_bytes;
1346 em->block_start = disk_bytenr;
1347 em->orig_block_len = disk_num_bytes;
1348 em->ram_bytes = ram_bytes;
1349 em->bdev = root->fs_info->fs_devices->latest_bdev;
1350 em->mod_start = em->start;
1351 em->mod_len = em->len;
1352 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1353 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1354 em->generation = -1;
1356 write_lock(&em_tree->lock);
1357 ret = add_extent_mapping(em_tree, em, 1);
1358 write_unlock(&em_tree->lock);
1359 if (ret != -EEXIST) {
1360 free_extent_map(em);
1363 btrfs_drop_extent_cache(inode, em->start,
1364 em->start + em->len - 1, 0);
1366 type = BTRFS_ORDERED_PREALLOC;
1368 type = BTRFS_ORDERED_NOCOW;
1371 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1372 num_bytes, num_bytes, type);
1373 BUG_ON(ret); /* -ENOMEM */
1375 if (root->root_key.objectid ==
1376 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1377 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1380 if (!nolock && nocow)
1381 btrfs_end_nocow_write(root);
1386 extent_clear_unlock_delalloc(inode, cur_offset,
1387 cur_offset + num_bytes - 1,
1388 locked_page, EXTENT_LOCKED |
1389 EXTENT_DELALLOC, PAGE_UNLOCK |
1391 if (!nolock && nocow)
1392 btrfs_end_nocow_write(root);
1393 cur_offset = extent_end;
1394 if (cur_offset > end)
1397 btrfs_release_path(path);
1399 if (cur_offset <= end && cow_start == (u64)-1) {
1400 cow_start = cur_offset;
1404 if (cow_start != (u64)-1) {
1405 ret = cow_file_range(inode, locked_page, cow_start, end,
1406 page_started, nr_written, 1);
1412 err = btrfs_end_transaction(trans, root);
1416 if (ret && cur_offset < end)
1417 extent_clear_unlock_delalloc(inode, cur_offset, end,
1418 locked_page, EXTENT_LOCKED |
1419 EXTENT_DELALLOC | EXTENT_DEFRAG |
1420 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1422 PAGE_SET_WRITEBACK |
1423 PAGE_END_WRITEBACK);
1424 btrfs_free_path(path);
1429 * extent_io.c call back to do delayed allocation processing
1431 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1432 u64 start, u64 end, int *page_started,
1433 unsigned long *nr_written)
1436 struct btrfs_root *root = BTRFS_I(inode)->root;
1438 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1439 ret = run_delalloc_nocow(inode, locked_page, start, end,
1440 page_started, 1, nr_written);
1441 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1442 ret = run_delalloc_nocow(inode, locked_page, start, end,
1443 page_started, 0, nr_written);
1444 } else if (!btrfs_test_opt(root, COMPRESS) &&
1445 !(BTRFS_I(inode)->force_compress) &&
1446 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1447 ret = cow_file_range(inode, locked_page, start, end,
1448 page_started, nr_written, 1);
1450 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1451 &BTRFS_I(inode)->runtime_flags);
1452 ret = cow_file_range_async(inode, locked_page, start, end,
1453 page_started, nr_written);
1458 static void btrfs_split_extent_hook(struct inode *inode,
1459 struct extent_state *orig, u64 split)
1461 /* not delalloc, ignore it */
1462 if (!(orig->state & EXTENT_DELALLOC))
1465 spin_lock(&BTRFS_I(inode)->lock);
1466 BTRFS_I(inode)->outstanding_extents++;
1467 spin_unlock(&BTRFS_I(inode)->lock);
1471 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1472 * extents so we can keep track of new extents that are just merged onto old
1473 * extents, such as when we are doing sequential writes, so we can properly
1474 * account for the metadata space we'll need.
1476 static void btrfs_merge_extent_hook(struct inode *inode,
1477 struct extent_state *new,
1478 struct extent_state *other)
1480 /* not delalloc, ignore it */
1481 if (!(other->state & EXTENT_DELALLOC))
1484 spin_lock(&BTRFS_I(inode)->lock);
1485 BTRFS_I(inode)->outstanding_extents--;
1486 spin_unlock(&BTRFS_I(inode)->lock);
1489 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1490 struct inode *inode)
1492 spin_lock(&root->delalloc_lock);
1493 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1494 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1495 &root->delalloc_inodes);
1496 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1497 &BTRFS_I(inode)->runtime_flags);
1498 root->nr_delalloc_inodes++;
1499 if (root->nr_delalloc_inodes == 1) {
1500 spin_lock(&root->fs_info->delalloc_root_lock);
1501 BUG_ON(!list_empty(&root->delalloc_root));
1502 list_add_tail(&root->delalloc_root,
1503 &root->fs_info->delalloc_roots);
1504 spin_unlock(&root->fs_info->delalloc_root_lock);
1507 spin_unlock(&root->delalloc_lock);
1510 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1511 struct inode *inode)
1513 spin_lock(&root->delalloc_lock);
1514 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1515 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1516 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1517 &BTRFS_I(inode)->runtime_flags);
1518 root->nr_delalloc_inodes--;
1519 if (!root->nr_delalloc_inodes) {
1520 spin_lock(&root->fs_info->delalloc_root_lock);
1521 BUG_ON(list_empty(&root->delalloc_root));
1522 list_del_init(&root->delalloc_root);
1523 spin_unlock(&root->fs_info->delalloc_root_lock);
1526 spin_unlock(&root->delalloc_lock);
1530 * extent_io.c set_bit_hook, used to track delayed allocation
1531 * bytes in this file, and to maintain the list of inodes that
1532 * have pending delalloc work to be done.
1534 static void btrfs_set_bit_hook(struct inode *inode,
1535 struct extent_state *state, unsigned long *bits)
1539 * set_bit and clear bit hooks normally require _irqsave/restore
1540 * but in this case, we are only testing for the DELALLOC
1541 * bit, which is only set or cleared with irqs on
1543 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1544 struct btrfs_root *root = BTRFS_I(inode)->root;
1545 u64 len = state->end + 1 - state->start;
1546 bool do_list = !btrfs_is_free_space_inode(inode);
1548 if (*bits & EXTENT_FIRST_DELALLOC) {
1549 *bits &= ~EXTENT_FIRST_DELALLOC;
1551 spin_lock(&BTRFS_I(inode)->lock);
1552 BTRFS_I(inode)->outstanding_extents++;
1553 spin_unlock(&BTRFS_I(inode)->lock);
1556 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1557 root->fs_info->delalloc_batch);
1558 spin_lock(&BTRFS_I(inode)->lock);
1559 BTRFS_I(inode)->delalloc_bytes += len;
1560 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1561 &BTRFS_I(inode)->runtime_flags))
1562 btrfs_add_delalloc_inodes(root, inode);
1563 spin_unlock(&BTRFS_I(inode)->lock);
1568 * extent_io.c clear_bit_hook, see set_bit_hook for why
1570 static void btrfs_clear_bit_hook(struct inode *inode,
1571 struct extent_state *state,
1572 unsigned long *bits)
1575 * set_bit and clear bit hooks normally require _irqsave/restore
1576 * but in this case, we are only testing for the DELALLOC
1577 * bit, which is only set or cleared with irqs on
1579 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1580 struct btrfs_root *root = BTRFS_I(inode)->root;
1581 u64 len = state->end + 1 - state->start;
1582 bool do_list = !btrfs_is_free_space_inode(inode);
1584 if (*bits & EXTENT_FIRST_DELALLOC) {
1585 *bits &= ~EXTENT_FIRST_DELALLOC;
1586 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1587 spin_lock(&BTRFS_I(inode)->lock);
1588 BTRFS_I(inode)->outstanding_extents--;
1589 spin_unlock(&BTRFS_I(inode)->lock);
1593 * We don't reserve metadata space for space cache inodes so we
1594 * don't need to call dellalloc_release_metadata if there is an
1597 if (*bits & EXTENT_DO_ACCOUNTING &&
1598 root != root->fs_info->tree_root)
1599 btrfs_delalloc_release_metadata(inode, len);
1601 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1602 && do_list && !(state->state & EXTENT_NORESERVE))
1603 btrfs_free_reserved_data_space(inode, len);
1605 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1606 root->fs_info->delalloc_batch);
1607 spin_lock(&BTRFS_I(inode)->lock);
1608 BTRFS_I(inode)->delalloc_bytes -= len;
1609 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1610 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1611 &BTRFS_I(inode)->runtime_flags))
1612 btrfs_del_delalloc_inode(root, inode);
1613 spin_unlock(&BTRFS_I(inode)->lock);
1618 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1619 * we don't create bios that span stripes or chunks
1621 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1622 size_t size, struct bio *bio,
1623 unsigned long bio_flags)
1625 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1626 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1631 if (bio_flags & EXTENT_BIO_COMPRESSED)
1634 length = bio->bi_iter.bi_size;
1635 map_length = length;
1636 ret = btrfs_map_block(root->fs_info, rw, logical,
1637 &map_length, NULL, 0);
1638 /* Will always return 0 with map_multi == NULL */
1640 if (map_length < length + size)
1646 * in order to insert checksums into the metadata in large chunks,
1647 * we wait until bio submission time. All the pages in the bio are
1648 * checksummed and sums are attached onto the ordered extent record.
1650 * At IO completion time the cums attached on the ordered extent record
1651 * are inserted into the btree
1653 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1654 struct bio *bio, int mirror_num,
1655 unsigned long bio_flags,
1658 struct btrfs_root *root = BTRFS_I(inode)->root;
1661 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1662 BUG_ON(ret); /* -ENOMEM */
1667 * in order to insert checksums into the metadata in large chunks,
1668 * we wait until bio submission time. All the pages in the bio are
1669 * checksummed and sums are attached onto the ordered extent record.
1671 * At IO completion time the cums attached on the ordered extent record
1672 * are inserted into the btree
1674 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1675 int mirror_num, unsigned long bio_flags,
1678 struct btrfs_root *root = BTRFS_I(inode)->root;
1681 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1683 bio_endio(bio, ret);
1688 * extent_io.c submission hook. This does the right thing for csum calculation
1689 * on write, or reading the csums from the tree before a read
1691 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1692 int mirror_num, unsigned long bio_flags,
1695 struct btrfs_root *root = BTRFS_I(inode)->root;
1699 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1701 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1703 if (btrfs_is_free_space_inode(inode))
1706 if (!(rw & REQ_WRITE)) {
1707 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1711 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1712 ret = btrfs_submit_compressed_read(inode, bio,
1716 } else if (!skip_sum) {
1717 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1722 } else if (async && !skip_sum) {
1723 /* csum items have already been cloned */
1724 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1726 /* we're doing a write, do the async checksumming */
1727 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1728 inode, rw, bio, mirror_num,
1729 bio_flags, bio_offset,
1730 __btrfs_submit_bio_start,
1731 __btrfs_submit_bio_done);
1733 } else if (!skip_sum) {
1734 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1740 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1744 bio_endio(bio, ret);
1749 * given a list of ordered sums record them in the inode. This happens
1750 * at IO completion time based on sums calculated at bio submission time.
1752 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1753 struct inode *inode, u64 file_offset,
1754 struct list_head *list)
1756 struct btrfs_ordered_sum *sum;
1758 list_for_each_entry(sum, list, list) {
1759 trans->adding_csums = 1;
1760 btrfs_csum_file_blocks(trans,
1761 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1762 trans->adding_csums = 0;
1767 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1768 struct extent_state **cached_state)
1770 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1771 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1772 cached_state, GFP_NOFS);
1775 /* see btrfs_writepage_start_hook for details on why this is required */
1776 struct btrfs_writepage_fixup {
1778 struct btrfs_work work;
1781 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1783 struct btrfs_writepage_fixup *fixup;
1784 struct btrfs_ordered_extent *ordered;
1785 struct extent_state *cached_state = NULL;
1787 struct inode *inode;
1792 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1796 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1797 ClearPageChecked(page);
1801 inode = page->mapping->host;
1802 page_start = page_offset(page);
1803 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1805 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1808 /* already ordered? We're done */
1809 if (PagePrivate2(page))
1812 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1814 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1815 page_end, &cached_state, GFP_NOFS);
1817 btrfs_start_ordered_extent(inode, ordered, 1);
1818 btrfs_put_ordered_extent(ordered);
1822 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1824 mapping_set_error(page->mapping, ret);
1825 end_extent_writepage(page, ret, page_start, page_end);
1826 ClearPageChecked(page);
1830 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1831 ClearPageChecked(page);
1832 set_page_dirty(page);
1834 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1835 &cached_state, GFP_NOFS);
1838 page_cache_release(page);
1843 * There are a few paths in the higher layers of the kernel that directly
1844 * set the page dirty bit without asking the filesystem if it is a
1845 * good idea. This causes problems because we want to make sure COW
1846 * properly happens and the data=ordered rules are followed.
1848 * In our case any range that doesn't have the ORDERED bit set
1849 * hasn't been properly setup for IO. We kick off an async process
1850 * to fix it up. The async helper will wait for ordered extents, set
1851 * the delalloc bit and make it safe to write the page.
1853 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1855 struct inode *inode = page->mapping->host;
1856 struct btrfs_writepage_fixup *fixup;
1857 struct btrfs_root *root = BTRFS_I(inode)->root;
1859 /* this page is properly in the ordered list */
1860 if (TestClearPagePrivate2(page))
1863 if (PageChecked(page))
1866 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1870 SetPageChecked(page);
1871 page_cache_get(page);
1872 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
1874 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
1878 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1879 struct inode *inode, u64 file_pos,
1880 u64 disk_bytenr, u64 disk_num_bytes,
1881 u64 num_bytes, u64 ram_bytes,
1882 u8 compression, u8 encryption,
1883 u16 other_encoding, int extent_type)
1885 struct btrfs_root *root = BTRFS_I(inode)->root;
1886 struct btrfs_file_extent_item *fi;
1887 struct btrfs_path *path;
1888 struct extent_buffer *leaf;
1889 struct btrfs_key ins;
1890 int extent_inserted = 0;
1893 path = btrfs_alloc_path();
1898 * we may be replacing one extent in the tree with another.
1899 * The new extent is pinned in the extent map, and we don't want
1900 * to drop it from the cache until it is completely in the btree.
1902 * So, tell btrfs_drop_extents to leave this extent in the cache.
1903 * the caller is expected to unpin it and allow it to be merged
1906 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
1907 file_pos + num_bytes, NULL, 0,
1908 1, sizeof(*fi), &extent_inserted);
1912 if (!extent_inserted) {
1913 ins.objectid = btrfs_ino(inode);
1914 ins.offset = file_pos;
1915 ins.type = BTRFS_EXTENT_DATA_KEY;
1917 path->leave_spinning = 1;
1918 ret = btrfs_insert_empty_item(trans, root, path, &ins,
1923 leaf = path->nodes[0];
1924 fi = btrfs_item_ptr(leaf, path->slots[0],
1925 struct btrfs_file_extent_item);
1926 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1927 btrfs_set_file_extent_type(leaf, fi, extent_type);
1928 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1929 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1930 btrfs_set_file_extent_offset(leaf, fi, 0);
1931 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1932 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1933 btrfs_set_file_extent_compression(leaf, fi, compression);
1934 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1935 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1937 btrfs_mark_buffer_dirty(leaf);
1938 btrfs_release_path(path);
1940 inode_add_bytes(inode, num_bytes);
1942 ins.objectid = disk_bytenr;
1943 ins.offset = disk_num_bytes;
1944 ins.type = BTRFS_EXTENT_ITEM_KEY;
1945 ret = btrfs_alloc_reserved_file_extent(trans, root,
1946 root->root_key.objectid,
1947 btrfs_ino(inode), file_pos, &ins);
1949 btrfs_free_path(path);
1954 /* snapshot-aware defrag */
1955 struct sa_defrag_extent_backref {
1956 struct rb_node node;
1957 struct old_sa_defrag_extent *old;
1966 struct old_sa_defrag_extent {
1967 struct list_head list;
1968 struct new_sa_defrag_extent *new;
1977 struct new_sa_defrag_extent {
1978 struct rb_root root;
1979 struct list_head head;
1980 struct btrfs_path *path;
1981 struct inode *inode;
1989 static int backref_comp(struct sa_defrag_extent_backref *b1,
1990 struct sa_defrag_extent_backref *b2)
1992 if (b1->root_id < b2->root_id)
1994 else if (b1->root_id > b2->root_id)
1997 if (b1->inum < b2->inum)
1999 else if (b1->inum > b2->inum)
2002 if (b1->file_pos < b2->file_pos)
2004 else if (b1->file_pos > b2->file_pos)
2008 * [------------------------------] ===> (a range of space)
2009 * |<--->| |<---->| =============> (fs/file tree A)
2010 * |<---------------------------->| ===> (fs/file tree B)
2012 * A range of space can refer to two file extents in one tree while
2013 * refer to only one file extent in another tree.
2015 * So we may process a disk offset more than one time(two extents in A)
2016 * and locate at the same extent(one extent in B), then insert two same
2017 * backrefs(both refer to the extent in B).
2022 static void backref_insert(struct rb_root *root,
2023 struct sa_defrag_extent_backref *backref)
2025 struct rb_node **p = &root->rb_node;
2026 struct rb_node *parent = NULL;
2027 struct sa_defrag_extent_backref *entry;
2032 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2034 ret = backref_comp(backref, entry);
2038 p = &(*p)->rb_right;
2041 rb_link_node(&backref->node, parent, p);
2042 rb_insert_color(&backref->node, root);
2046 * Note the backref might has changed, and in this case we just return 0.
2048 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2051 struct btrfs_file_extent_item *extent;
2052 struct btrfs_fs_info *fs_info;
2053 struct old_sa_defrag_extent *old = ctx;
2054 struct new_sa_defrag_extent *new = old->new;
2055 struct btrfs_path *path = new->path;
2056 struct btrfs_key key;
2057 struct btrfs_root *root;
2058 struct sa_defrag_extent_backref *backref;
2059 struct extent_buffer *leaf;
2060 struct inode *inode = new->inode;
2066 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2067 inum == btrfs_ino(inode))
2070 key.objectid = root_id;
2071 key.type = BTRFS_ROOT_ITEM_KEY;
2072 key.offset = (u64)-1;
2074 fs_info = BTRFS_I(inode)->root->fs_info;
2075 root = btrfs_read_fs_root_no_name(fs_info, &key);
2077 if (PTR_ERR(root) == -ENOENT)
2080 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2081 inum, offset, root_id);
2082 return PTR_ERR(root);
2085 key.objectid = inum;
2086 key.type = BTRFS_EXTENT_DATA_KEY;
2087 if (offset > (u64)-1 << 32)
2090 key.offset = offset;
2092 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2093 if (WARN_ON(ret < 0))
2100 leaf = path->nodes[0];
2101 slot = path->slots[0];
2103 if (slot >= btrfs_header_nritems(leaf)) {
2104 ret = btrfs_next_leaf(root, path);
2107 } else if (ret > 0) {
2116 btrfs_item_key_to_cpu(leaf, &key, slot);
2118 if (key.objectid > inum)
2121 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2124 extent = btrfs_item_ptr(leaf, slot,
2125 struct btrfs_file_extent_item);
2127 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2131 * 'offset' refers to the exact key.offset,
2132 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2133 * (key.offset - extent_offset).
2135 if (key.offset != offset)
2138 extent_offset = btrfs_file_extent_offset(leaf, extent);
2139 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2141 if (extent_offset >= old->extent_offset + old->offset +
2142 old->len || extent_offset + num_bytes <=
2143 old->extent_offset + old->offset)
2148 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2154 backref->root_id = root_id;
2155 backref->inum = inum;
2156 backref->file_pos = offset;
2157 backref->num_bytes = num_bytes;
2158 backref->extent_offset = extent_offset;
2159 backref->generation = btrfs_file_extent_generation(leaf, extent);
2161 backref_insert(&new->root, backref);
2164 btrfs_release_path(path);
2169 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2170 struct new_sa_defrag_extent *new)
2172 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2173 struct old_sa_defrag_extent *old, *tmp;
2178 list_for_each_entry_safe(old, tmp, &new->head, list) {
2179 ret = iterate_inodes_from_logical(old->bytenr +
2180 old->extent_offset, fs_info,
2181 path, record_one_backref,
2183 if (ret < 0 && ret != -ENOENT)
2186 /* no backref to be processed for this extent */
2188 list_del(&old->list);
2193 if (list_empty(&new->head))
2199 static int relink_is_mergable(struct extent_buffer *leaf,
2200 struct btrfs_file_extent_item *fi,
2201 struct new_sa_defrag_extent *new)
2203 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2206 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2209 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2212 if (btrfs_file_extent_encryption(leaf, fi) ||
2213 btrfs_file_extent_other_encoding(leaf, fi))
2220 * Note the backref might has changed, and in this case we just return 0.
2222 static noinline int relink_extent_backref(struct btrfs_path *path,
2223 struct sa_defrag_extent_backref *prev,
2224 struct sa_defrag_extent_backref *backref)
2226 struct btrfs_file_extent_item *extent;
2227 struct btrfs_file_extent_item *item;
2228 struct btrfs_ordered_extent *ordered;
2229 struct btrfs_trans_handle *trans;
2230 struct btrfs_fs_info *fs_info;
2231 struct btrfs_root *root;
2232 struct btrfs_key key;
2233 struct extent_buffer *leaf;
2234 struct old_sa_defrag_extent *old = backref->old;
2235 struct new_sa_defrag_extent *new = old->new;
2236 struct inode *src_inode = new->inode;
2237 struct inode *inode;
2238 struct extent_state *cached = NULL;
2247 if (prev && prev->root_id == backref->root_id &&
2248 prev->inum == backref->inum &&
2249 prev->file_pos + prev->num_bytes == backref->file_pos)
2252 /* step 1: get root */
2253 key.objectid = backref->root_id;
2254 key.type = BTRFS_ROOT_ITEM_KEY;
2255 key.offset = (u64)-1;
2257 fs_info = BTRFS_I(src_inode)->root->fs_info;
2258 index = srcu_read_lock(&fs_info->subvol_srcu);
2260 root = btrfs_read_fs_root_no_name(fs_info, &key);
2262 srcu_read_unlock(&fs_info->subvol_srcu, index);
2263 if (PTR_ERR(root) == -ENOENT)
2265 return PTR_ERR(root);
2268 if (btrfs_root_readonly(root)) {
2269 srcu_read_unlock(&fs_info->subvol_srcu, index);
2273 /* step 2: get inode */
2274 key.objectid = backref->inum;
2275 key.type = BTRFS_INODE_ITEM_KEY;
2278 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2279 if (IS_ERR(inode)) {
2280 srcu_read_unlock(&fs_info->subvol_srcu, index);
2284 srcu_read_unlock(&fs_info->subvol_srcu, index);
2286 /* step 3: relink backref */
2287 lock_start = backref->file_pos;
2288 lock_end = backref->file_pos + backref->num_bytes - 1;
2289 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2292 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2294 btrfs_put_ordered_extent(ordered);
2298 trans = btrfs_join_transaction(root);
2299 if (IS_ERR(trans)) {
2300 ret = PTR_ERR(trans);
2304 key.objectid = backref->inum;
2305 key.type = BTRFS_EXTENT_DATA_KEY;
2306 key.offset = backref->file_pos;
2308 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2311 } else if (ret > 0) {
2316 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2317 struct btrfs_file_extent_item);
2319 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2320 backref->generation)
2323 btrfs_release_path(path);
2325 start = backref->file_pos;
2326 if (backref->extent_offset < old->extent_offset + old->offset)
2327 start += old->extent_offset + old->offset -
2328 backref->extent_offset;
2330 len = min(backref->extent_offset + backref->num_bytes,
2331 old->extent_offset + old->offset + old->len);
2332 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2334 ret = btrfs_drop_extents(trans, root, inode, start,
2339 key.objectid = btrfs_ino(inode);
2340 key.type = BTRFS_EXTENT_DATA_KEY;
2343 path->leave_spinning = 1;
2345 struct btrfs_file_extent_item *fi;
2347 struct btrfs_key found_key;
2349 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2354 leaf = path->nodes[0];
2355 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2357 fi = btrfs_item_ptr(leaf, path->slots[0],
2358 struct btrfs_file_extent_item);
2359 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2361 if (extent_len + found_key.offset == start &&
2362 relink_is_mergable(leaf, fi, new)) {
2363 btrfs_set_file_extent_num_bytes(leaf, fi,
2365 btrfs_mark_buffer_dirty(leaf);
2366 inode_add_bytes(inode, len);
2372 btrfs_release_path(path);
2377 ret = btrfs_insert_empty_item(trans, root, path, &key,
2380 btrfs_abort_transaction(trans, root, ret);
2384 leaf = path->nodes[0];
2385 item = btrfs_item_ptr(leaf, path->slots[0],
2386 struct btrfs_file_extent_item);
2387 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2388 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2389 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2390 btrfs_set_file_extent_num_bytes(leaf, item, len);
2391 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2392 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2393 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2394 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2395 btrfs_set_file_extent_encryption(leaf, item, 0);
2396 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2398 btrfs_mark_buffer_dirty(leaf);
2399 inode_add_bytes(inode, len);
2400 btrfs_release_path(path);
2402 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2404 backref->root_id, backref->inum,
2405 new->file_pos, 0); /* start - extent_offset */
2407 btrfs_abort_transaction(trans, root, ret);
2413 btrfs_release_path(path);
2414 path->leave_spinning = 0;
2415 btrfs_end_transaction(trans, root);
2417 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2423 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2425 struct old_sa_defrag_extent *old, *tmp;
2430 list_for_each_entry_safe(old, tmp, &new->head, list) {
2431 list_del(&old->list);
2437 static void relink_file_extents(struct new_sa_defrag_extent *new)
2439 struct btrfs_path *path;
2440 struct sa_defrag_extent_backref *backref;
2441 struct sa_defrag_extent_backref *prev = NULL;
2442 struct inode *inode;
2443 struct btrfs_root *root;
2444 struct rb_node *node;
2448 root = BTRFS_I(inode)->root;
2450 path = btrfs_alloc_path();
2454 if (!record_extent_backrefs(path, new)) {
2455 btrfs_free_path(path);
2458 btrfs_release_path(path);
2461 node = rb_first(&new->root);
2464 rb_erase(node, &new->root);
2466 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2468 ret = relink_extent_backref(path, prev, backref);
2481 btrfs_free_path(path);
2483 free_sa_defrag_extent(new);
2485 atomic_dec(&root->fs_info->defrag_running);
2486 wake_up(&root->fs_info->transaction_wait);
2489 static struct new_sa_defrag_extent *
2490 record_old_file_extents(struct inode *inode,
2491 struct btrfs_ordered_extent *ordered)
2493 struct btrfs_root *root = BTRFS_I(inode)->root;
2494 struct btrfs_path *path;
2495 struct btrfs_key key;
2496 struct old_sa_defrag_extent *old;
2497 struct new_sa_defrag_extent *new;
2500 new = kmalloc(sizeof(*new), GFP_NOFS);
2505 new->file_pos = ordered->file_offset;
2506 new->len = ordered->len;
2507 new->bytenr = ordered->start;
2508 new->disk_len = ordered->disk_len;
2509 new->compress_type = ordered->compress_type;
2510 new->root = RB_ROOT;
2511 INIT_LIST_HEAD(&new->head);
2513 path = btrfs_alloc_path();
2517 key.objectid = btrfs_ino(inode);
2518 key.type = BTRFS_EXTENT_DATA_KEY;
2519 key.offset = new->file_pos;
2521 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2524 if (ret > 0 && path->slots[0] > 0)
2527 /* find out all the old extents for the file range */
2529 struct btrfs_file_extent_item *extent;
2530 struct extent_buffer *l;
2539 slot = path->slots[0];
2541 if (slot >= btrfs_header_nritems(l)) {
2542 ret = btrfs_next_leaf(root, path);
2550 btrfs_item_key_to_cpu(l, &key, slot);
2552 if (key.objectid != btrfs_ino(inode))
2554 if (key.type != BTRFS_EXTENT_DATA_KEY)
2556 if (key.offset >= new->file_pos + new->len)
2559 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2561 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2562 if (key.offset + num_bytes < new->file_pos)
2565 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2569 extent_offset = btrfs_file_extent_offset(l, extent);
2571 old = kmalloc(sizeof(*old), GFP_NOFS);
2575 offset = max(new->file_pos, key.offset);
2576 end = min(new->file_pos + new->len, key.offset + num_bytes);
2578 old->bytenr = disk_bytenr;
2579 old->extent_offset = extent_offset;
2580 old->offset = offset - key.offset;
2581 old->len = end - offset;
2584 list_add_tail(&old->list, &new->head);
2590 btrfs_free_path(path);
2591 atomic_inc(&root->fs_info->defrag_running);
2596 btrfs_free_path(path);
2598 free_sa_defrag_extent(new);
2602 /* as ordered data IO finishes, this gets called so we can finish
2603 * an ordered extent if the range of bytes in the file it covers are
2606 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2608 struct inode *inode = ordered_extent->inode;
2609 struct btrfs_root *root = BTRFS_I(inode)->root;
2610 struct btrfs_trans_handle *trans = NULL;
2611 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2612 struct extent_state *cached_state = NULL;
2613 struct new_sa_defrag_extent *new = NULL;
2614 int compress_type = 0;
2616 u64 logical_len = ordered_extent->len;
2618 bool truncated = false;
2620 nolock = btrfs_is_free_space_inode(inode);
2622 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2627 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2629 logical_len = ordered_extent->truncated_len;
2630 /* Truncated the entire extent, don't bother adding */
2635 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2636 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2637 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2639 trans = btrfs_join_transaction_nolock(root);
2641 trans = btrfs_join_transaction(root);
2642 if (IS_ERR(trans)) {
2643 ret = PTR_ERR(trans);
2647 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2648 ret = btrfs_update_inode_fallback(trans, root, inode);
2649 if (ret) /* -ENOMEM or corruption */
2650 btrfs_abort_transaction(trans, root, ret);
2654 lock_extent_bits(io_tree, ordered_extent->file_offset,
2655 ordered_extent->file_offset + ordered_extent->len - 1,
2658 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2659 ordered_extent->file_offset + ordered_extent->len - 1,
2660 EXTENT_DEFRAG, 1, cached_state);
2662 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2663 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2664 /* the inode is shared */
2665 new = record_old_file_extents(inode, ordered_extent);
2667 clear_extent_bit(io_tree, ordered_extent->file_offset,
2668 ordered_extent->file_offset + ordered_extent->len - 1,
2669 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2673 trans = btrfs_join_transaction_nolock(root);
2675 trans = btrfs_join_transaction(root);
2676 if (IS_ERR(trans)) {
2677 ret = PTR_ERR(trans);
2681 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2683 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2684 compress_type = ordered_extent->compress_type;
2685 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2686 BUG_ON(compress_type);
2687 ret = btrfs_mark_extent_written(trans, inode,
2688 ordered_extent->file_offset,
2689 ordered_extent->file_offset +
2692 BUG_ON(root == root->fs_info->tree_root);
2693 ret = insert_reserved_file_extent(trans, inode,
2694 ordered_extent->file_offset,
2695 ordered_extent->start,
2696 ordered_extent->disk_len,
2697 logical_len, logical_len,
2698 compress_type, 0, 0,
2699 BTRFS_FILE_EXTENT_REG);
2701 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2702 ordered_extent->file_offset, ordered_extent->len,
2705 btrfs_abort_transaction(trans, root, ret);
2709 add_pending_csums(trans, inode, ordered_extent->file_offset,
2710 &ordered_extent->list);
2712 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2713 ret = btrfs_update_inode_fallback(trans, root, inode);
2714 if (ret) { /* -ENOMEM or corruption */
2715 btrfs_abort_transaction(trans, root, ret);
2720 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2721 ordered_extent->file_offset +
2722 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2724 if (root != root->fs_info->tree_root)
2725 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2727 btrfs_end_transaction(trans, root);
2729 if (ret || truncated) {
2733 start = ordered_extent->file_offset + logical_len;
2735 start = ordered_extent->file_offset;
2736 end = ordered_extent->file_offset + ordered_extent->len - 1;
2737 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2739 /* Drop the cache for the part of the extent we didn't write. */
2740 btrfs_drop_extent_cache(inode, start, end, 0);
2743 * If the ordered extent had an IOERR or something else went
2744 * wrong we need to return the space for this ordered extent
2745 * back to the allocator. We only free the extent in the
2746 * truncated case if we didn't write out the extent at all.
2748 if ((ret || !logical_len) &&
2749 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2750 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2751 btrfs_free_reserved_extent(root, ordered_extent->start,
2752 ordered_extent->disk_len);
2757 * This needs to be done to make sure anybody waiting knows we are done
2758 * updating everything for this ordered extent.
2760 btrfs_remove_ordered_extent(inode, ordered_extent);
2762 /* for snapshot-aware defrag */
2765 free_sa_defrag_extent(new);
2766 atomic_dec(&root->fs_info->defrag_running);
2768 relink_file_extents(new);
2773 btrfs_put_ordered_extent(ordered_extent);
2774 /* once for the tree */
2775 btrfs_put_ordered_extent(ordered_extent);
2780 static void finish_ordered_fn(struct btrfs_work *work)
2782 struct btrfs_ordered_extent *ordered_extent;
2783 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2784 btrfs_finish_ordered_io(ordered_extent);
2787 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2788 struct extent_state *state, int uptodate)
2790 struct inode *inode = page->mapping->host;
2791 struct btrfs_root *root = BTRFS_I(inode)->root;
2792 struct btrfs_ordered_extent *ordered_extent = NULL;
2793 struct btrfs_workqueue *workers;
2795 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2797 ClearPagePrivate2(page);
2798 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2799 end - start + 1, uptodate))
2802 btrfs_init_work(&ordered_extent->work, finish_ordered_fn, NULL, NULL);
2804 if (btrfs_is_free_space_inode(inode))
2805 workers = root->fs_info->endio_freespace_worker;
2807 workers = root->fs_info->endio_write_workers;
2808 btrfs_queue_work(workers, &ordered_extent->work);
2814 * when reads are done, we need to check csums to verify the data is correct
2815 * if there's a match, we allow the bio to finish. If not, the code in
2816 * extent_io.c will try to find good copies for us.
2818 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2819 u64 phy_offset, struct page *page,
2820 u64 start, u64 end, int mirror)
2822 size_t offset = start - page_offset(page);
2823 struct inode *inode = page->mapping->host;
2824 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2826 struct btrfs_root *root = BTRFS_I(inode)->root;
2829 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2830 DEFAULT_RATELIMIT_BURST);
2832 if (PageChecked(page)) {
2833 ClearPageChecked(page);
2837 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2840 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2841 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2842 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2847 phy_offset >>= inode->i_sb->s_blocksize_bits;
2848 csum_expected = *(((u32 *)io_bio->csum) + phy_offset);
2850 kaddr = kmap_atomic(page);
2851 csum = btrfs_csum_data(kaddr + offset, csum, end - start + 1);
2852 btrfs_csum_final(csum, (char *)&csum);
2853 if (csum != csum_expected)
2856 kunmap_atomic(kaddr);
2861 if (__ratelimit(&_rs))
2862 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
2863 btrfs_ino(page->mapping->host), start, csum, csum_expected);
2864 memset(kaddr + offset, 1, end - start + 1);
2865 flush_dcache_page(page);
2866 kunmap_atomic(kaddr);
2867 if (csum_expected == 0)
2872 struct delayed_iput {
2873 struct list_head list;
2874 struct inode *inode;
2877 /* JDM: If this is fs-wide, why can't we add a pointer to
2878 * btrfs_inode instead and avoid the allocation? */
2879 void btrfs_add_delayed_iput(struct inode *inode)
2881 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2882 struct delayed_iput *delayed;
2884 if (atomic_add_unless(&inode->i_count, -1, 1))
2887 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2888 delayed->inode = inode;
2890 spin_lock(&fs_info->delayed_iput_lock);
2891 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2892 spin_unlock(&fs_info->delayed_iput_lock);
2895 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2898 struct btrfs_fs_info *fs_info = root->fs_info;
2899 struct delayed_iput *delayed;
2902 spin_lock(&fs_info->delayed_iput_lock);
2903 empty = list_empty(&fs_info->delayed_iputs);
2904 spin_unlock(&fs_info->delayed_iput_lock);
2908 spin_lock(&fs_info->delayed_iput_lock);
2909 list_splice_init(&fs_info->delayed_iputs, &list);
2910 spin_unlock(&fs_info->delayed_iput_lock);
2912 while (!list_empty(&list)) {
2913 delayed = list_entry(list.next, struct delayed_iput, list);
2914 list_del(&delayed->list);
2915 iput(delayed->inode);
2921 * This is called in transaction commit time. If there are no orphan
2922 * files in the subvolume, it removes orphan item and frees block_rsv
2925 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2926 struct btrfs_root *root)
2928 struct btrfs_block_rsv *block_rsv;
2931 if (atomic_read(&root->orphan_inodes) ||
2932 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2935 spin_lock(&root->orphan_lock);
2936 if (atomic_read(&root->orphan_inodes)) {
2937 spin_unlock(&root->orphan_lock);
2941 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2942 spin_unlock(&root->orphan_lock);
2946 block_rsv = root->orphan_block_rsv;
2947 root->orphan_block_rsv = NULL;
2948 spin_unlock(&root->orphan_lock);
2950 if (root->orphan_item_inserted &&
2951 btrfs_root_refs(&root->root_item) > 0) {
2952 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2953 root->root_key.objectid);
2955 btrfs_abort_transaction(trans, root, ret);
2957 root->orphan_item_inserted = 0;
2961 WARN_ON(block_rsv->size > 0);
2962 btrfs_free_block_rsv(root, block_rsv);
2967 * This creates an orphan entry for the given inode in case something goes
2968 * wrong in the middle of an unlink/truncate.
2970 * NOTE: caller of this function should reserve 5 units of metadata for
2973 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2975 struct btrfs_root *root = BTRFS_I(inode)->root;
2976 struct btrfs_block_rsv *block_rsv = NULL;
2981 if (!root->orphan_block_rsv) {
2982 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2987 spin_lock(&root->orphan_lock);
2988 if (!root->orphan_block_rsv) {
2989 root->orphan_block_rsv = block_rsv;
2990 } else if (block_rsv) {
2991 btrfs_free_block_rsv(root, block_rsv);
2995 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2996 &BTRFS_I(inode)->runtime_flags)) {
2999 * For proper ENOSPC handling, we should do orphan
3000 * cleanup when mounting. But this introduces backward
3001 * compatibility issue.
3003 if (!xchg(&root->orphan_item_inserted, 1))
3009 atomic_inc(&root->orphan_inodes);
3012 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3013 &BTRFS_I(inode)->runtime_flags))
3015 spin_unlock(&root->orphan_lock);
3017 /* grab metadata reservation from transaction handle */
3019 ret = btrfs_orphan_reserve_metadata(trans, inode);
3020 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3023 /* insert an orphan item to track this unlinked/truncated file */
3025 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3027 atomic_dec(&root->orphan_inodes);
3029 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3030 &BTRFS_I(inode)->runtime_flags);
3031 btrfs_orphan_release_metadata(inode);
3033 if (ret != -EEXIST) {
3034 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3035 &BTRFS_I(inode)->runtime_flags);
3036 btrfs_abort_transaction(trans, root, ret);
3043 /* insert an orphan item to track subvolume contains orphan files */
3045 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3046 root->root_key.objectid);
3047 if (ret && ret != -EEXIST) {
3048 btrfs_abort_transaction(trans, root, ret);
3056 * We have done the truncate/delete so we can go ahead and remove the orphan
3057 * item for this particular inode.
3059 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3060 struct inode *inode)
3062 struct btrfs_root *root = BTRFS_I(inode)->root;
3063 int delete_item = 0;
3064 int release_rsv = 0;
3067 spin_lock(&root->orphan_lock);
3068 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3069 &BTRFS_I(inode)->runtime_flags))
3072 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3073 &BTRFS_I(inode)->runtime_flags))
3075 spin_unlock(&root->orphan_lock);
3078 atomic_dec(&root->orphan_inodes);
3080 ret = btrfs_del_orphan_item(trans, root,
3085 btrfs_orphan_release_metadata(inode);
3091 * this cleans up any orphans that may be left on the list from the last use
3094 int btrfs_orphan_cleanup(struct btrfs_root *root)
3096 struct btrfs_path *path;
3097 struct extent_buffer *leaf;
3098 struct btrfs_key key, found_key;
3099 struct btrfs_trans_handle *trans;
3100 struct inode *inode;
3101 u64 last_objectid = 0;
3102 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3104 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3107 path = btrfs_alloc_path();
3114 key.objectid = BTRFS_ORPHAN_OBJECTID;
3115 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3116 key.offset = (u64)-1;
3119 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3124 * if ret == 0 means we found what we were searching for, which
3125 * is weird, but possible, so only screw with path if we didn't
3126 * find the key and see if we have stuff that matches
3130 if (path->slots[0] == 0)
3135 /* pull out the item */
3136 leaf = path->nodes[0];
3137 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3139 /* make sure the item matches what we want */
3140 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3142 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3145 /* release the path since we're done with it */
3146 btrfs_release_path(path);
3149 * this is where we are basically btrfs_lookup, without the
3150 * crossing root thing. we store the inode number in the
3151 * offset of the orphan item.
3154 if (found_key.offset == last_objectid) {
3155 btrfs_err(root->fs_info,
3156 "Error removing orphan entry, stopping orphan cleanup");
3161 last_objectid = found_key.offset;
3163 found_key.objectid = found_key.offset;
3164 found_key.type = BTRFS_INODE_ITEM_KEY;
3165 found_key.offset = 0;
3166 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3167 ret = PTR_ERR_OR_ZERO(inode);
3168 if (ret && ret != -ESTALE)
3171 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3172 struct btrfs_root *dead_root;
3173 struct btrfs_fs_info *fs_info = root->fs_info;
3174 int is_dead_root = 0;
3177 * this is an orphan in the tree root. Currently these
3178 * could come from 2 sources:
3179 * a) a snapshot deletion in progress
3180 * b) a free space cache inode
3181 * We need to distinguish those two, as the snapshot
3182 * orphan must not get deleted.
3183 * find_dead_roots already ran before us, so if this
3184 * is a snapshot deletion, we should find the root
3185 * in the dead_roots list
3187 spin_lock(&fs_info->trans_lock);
3188 list_for_each_entry(dead_root, &fs_info->dead_roots,
3190 if (dead_root->root_key.objectid ==
3191 found_key.objectid) {
3196 spin_unlock(&fs_info->trans_lock);
3198 /* prevent this orphan from being found again */
3199 key.offset = found_key.objectid - 1;
3204 * Inode is already gone but the orphan item is still there,
3205 * kill the orphan item.
3207 if (ret == -ESTALE) {
3208 trans = btrfs_start_transaction(root, 1);
3209 if (IS_ERR(trans)) {
3210 ret = PTR_ERR(trans);
3213 btrfs_debug(root->fs_info, "auto deleting %Lu",
3214 found_key.objectid);
3215 ret = btrfs_del_orphan_item(trans, root,
3216 found_key.objectid);
3217 btrfs_end_transaction(trans, root);
3224 * add this inode to the orphan list so btrfs_orphan_del does
3225 * the proper thing when we hit it
3227 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3228 &BTRFS_I(inode)->runtime_flags);
3229 atomic_inc(&root->orphan_inodes);
3231 /* if we have links, this was a truncate, lets do that */
3232 if (inode->i_nlink) {
3233 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3239 /* 1 for the orphan item deletion. */
3240 trans = btrfs_start_transaction(root, 1);
3241 if (IS_ERR(trans)) {
3243 ret = PTR_ERR(trans);
3246 ret = btrfs_orphan_add(trans, inode);
3247 btrfs_end_transaction(trans, root);
3253 ret = btrfs_truncate(inode);
3255 btrfs_orphan_del(NULL, inode);
3260 /* this will do delete_inode and everything for us */
3265 /* release the path since we're done with it */
3266 btrfs_release_path(path);
3268 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3270 if (root->orphan_block_rsv)
3271 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3274 if (root->orphan_block_rsv || root->orphan_item_inserted) {
3275 trans = btrfs_join_transaction(root);
3277 btrfs_end_transaction(trans, root);
3281 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3283 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3287 btrfs_crit(root->fs_info,
3288 "could not do orphan cleanup %d", ret);
3289 btrfs_free_path(path);
3294 * very simple check to peek ahead in the leaf looking for xattrs. If we
3295 * don't find any xattrs, we know there can't be any acls.
3297 * slot is the slot the inode is in, objectid is the objectid of the inode
3299 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3300 int slot, u64 objectid,
3301 int *first_xattr_slot)
3303 u32 nritems = btrfs_header_nritems(leaf);
3304 struct btrfs_key found_key;
3305 static u64 xattr_access = 0;
3306 static u64 xattr_default = 0;
3309 if (!xattr_access) {
3310 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3311 strlen(POSIX_ACL_XATTR_ACCESS));
3312 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3313 strlen(POSIX_ACL_XATTR_DEFAULT));
3317 *first_xattr_slot = -1;
3318 while (slot < nritems) {
3319 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3321 /* we found a different objectid, there must not be acls */
3322 if (found_key.objectid != objectid)
3325 /* we found an xattr, assume we've got an acl */
3326 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3327 if (*first_xattr_slot == -1)
3328 *first_xattr_slot = slot;
3329 if (found_key.offset == xattr_access ||
3330 found_key.offset == xattr_default)
3335 * we found a key greater than an xattr key, there can't
3336 * be any acls later on
3338 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3345 * it goes inode, inode backrefs, xattrs, extents,
3346 * so if there are a ton of hard links to an inode there can
3347 * be a lot of backrefs. Don't waste time searching too hard,
3348 * this is just an optimization
3353 /* we hit the end of the leaf before we found an xattr or
3354 * something larger than an xattr. We have to assume the inode
3357 if (*first_xattr_slot == -1)
3358 *first_xattr_slot = slot;
3363 * read an inode from the btree into the in-memory inode
3365 static void btrfs_read_locked_inode(struct inode *inode)
3367 struct btrfs_path *path;
3368 struct extent_buffer *leaf;
3369 struct btrfs_inode_item *inode_item;
3370 struct btrfs_timespec *tspec;
3371 struct btrfs_root *root = BTRFS_I(inode)->root;
3372 struct btrfs_key location;
3377 bool filled = false;
3378 int first_xattr_slot;
3380 ret = btrfs_fill_inode(inode, &rdev);
3384 path = btrfs_alloc_path();
3388 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3390 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3394 leaf = path->nodes[0];
3399 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3400 struct btrfs_inode_item);
3401 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3402 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3403 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3404 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3405 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3407 tspec = btrfs_inode_atime(inode_item);
3408 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3409 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3411 tspec = btrfs_inode_mtime(inode_item);
3412 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3413 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3415 tspec = btrfs_inode_ctime(inode_item);
3416 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3417 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3419 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3420 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3421 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3424 * If we were modified in the current generation and evicted from memory
3425 * and then re-read we need to do a full sync since we don't have any
3426 * idea about which extents were modified before we were evicted from
3429 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3430 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3431 &BTRFS_I(inode)->runtime_flags);
3433 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3434 inode->i_generation = BTRFS_I(inode)->generation;
3436 rdev = btrfs_inode_rdev(leaf, inode_item);
3438 BTRFS_I(inode)->index_cnt = (u64)-1;
3439 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3443 if (inode->i_nlink != 1 ||
3444 path->slots[0] >= btrfs_header_nritems(leaf))
3447 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3448 if (location.objectid != btrfs_ino(inode))
3451 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3452 if (location.type == BTRFS_INODE_REF_KEY) {
3453 struct btrfs_inode_ref *ref;
3455 ref = (struct btrfs_inode_ref *)ptr;
3456 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3457 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3458 struct btrfs_inode_extref *extref;
3460 extref = (struct btrfs_inode_extref *)ptr;
3461 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3466 * try to precache a NULL acl entry for files that don't have
3467 * any xattrs or acls
3469 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3470 btrfs_ino(inode), &first_xattr_slot);
3471 if (first_xattr_slot != -1) {
3472 path->slots[0] = first_xattr_slot;
3473 ret = btrfs_load_inode_props(inode, path);
3475 btrfs_err(root->fs_info,
3476 "error loading props for ino %llu (root %llu): %d\n",
3478 root->root_key.objectid, ret);
3480 btrfs_free_path(path);
3483 cache_no_acl(inode);
3485 switch (inode->i_mode & S_IFMT) {
3487 inode->i_mapping->a_ops = &btrfs_aops;
3488 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3489 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3490 inode->i_fop = &btrfs_file_operations;
3491 inode->i_op = &btrfs_file_inode_operations;
3494 inode->i_fop = &btrfs_dir_file_operations;
3495 if (root == root->fs_info->tree_root)
3496 inode->i_op = &btrfs_dir_ro_inode_operations;
3498 inode->i_op = &btrfs_dir_inode_operations;
3501 inode->i_op = &btrfs_symlink_inode_operations;
3502 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3503 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3506 inode->i_op = &btrfs_special_inode_operations;
3507 init_special_inode(inode, inode->i_mode, rdev);
3511 btrfs_update_iflags(inode);
3515 btrfs_free_path(path);
3516 make_bad_inode(inode);
3520 * given a leaf and an inode, copy the inode fields into the leaf
3522 static void fill_inode_item(struct btrfs_trans_handle *trans,
3523 struct extent_buffer *leaf,
3524 struct btrfs_inode_item *item,
3525 struct inode *inode)
3527 struct btrfs_map_token token;
3529 btrfs_init_map_token(&token);
3531 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3532 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3533 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3535 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3536 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3538 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3539 inode->i_atime.tv_sec, &token);
3540 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3541 inode->i_atime.tv_nsec, &token);
3543 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3544 inode->i_mtime.tv_sec, &token);
3545 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3546 inode->i_mtime.tv_nsec, &token);
3548 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3549 inode->i_ctime.tv_sec, &token);
3550 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3551 inode->i_ctime.tv_nsec, &token);
3553 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3555 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3557 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3558 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3559 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3560 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3561 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3565 * copy everything in the in-memory inode into the btree.
3567 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3568 struct btrfs_root *root, struct inode *inode)
3570 struct btrfs_inode_item *inode_item;
3571 struct btrfs_path *path;
3572 struct extent_buffer *leaf;
3575 path = btrfs_alloc_path();
3579 path->leave_spinning = 1;
3580 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3588 leaf = path->nodes[0];
3589 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3590 struct btrfs_inode_item);
3592 fill_inode_item(trans, leaf, inode_item, inode);
3593 btrfs_mark_buffer_dirty(leaf);
3594 btrfs_set_inode_last_trans(trans, inode);
3597 btrfs_free_path(path);
3602 * copy everything in the in-memory inode into the btree.
3604 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3605 struct btrfs_root *root, struct inode *inode)
3610 * If the inode is a free space inode, we can deadlock during commit
3611 * if we put it into the delayed code.
3613 * The data relocation inode should also be directly updated
3616 if (!btrfs_is_free_space_inode(inode)
3617 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3618 btrfs_update_root_times(trans, root);
3620 ret = btrfs_delayed_update_inode(trans, root, inode);
3622 btrfs_set_inode_last_trans(trans, inode);
3626 return btrfs_update_inode_item(trans, root, inode);
3629 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3630 struct btrfs_root *root,
3631 struct inode *inode)
3635 ret = btrfs_update_inode(trans, root, inode);
3637 return btrfs_update_inode_item(trans, root, inode);
3642 * unlink helper that gets used here in inode.c and in the tree logging
3643 * recovery code. It remove a link in a directory with a given name, and
3644 * also drops the back refs in the inode to the directory
3646 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3647 struct btrfs_root *root,
3648 struct inode *dir, struct inode *inode,
3649 const char *name, int name_len)
3651 struct btrfs_path *path;
3653 struct extent_buffer *leaf;
3654 struct btrfs_dir_item *di;
3655 struct btrfs_key key;
3657 u64 ino = btrfs_ino(inode);
3658 u64 dir_ino = btrfs_ino(dir);
3660 path = btrfs_alloc_path();
3666 path->leave_spinning = 1;
3667 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3668 name, name_len, -1);
3677 leaf = path->nodes[0];
3678 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3679 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3682 btrfs_release_path(path);
3685 * If we don't have dir index, we have to get it by looking up
3686 * the inode ref, since we get the inode ref, remove it directly,
3687 * it is unnecessary to do delayed deletion.
3689 * But if we have dir index, needn't search inode ref to get it.
3690 * Since the inode ref is close to the inode item, it is better
3691 * that we delay to delete it, and just do this deletion when
3692 * we update the inode item.
3694 if (BTRFS_I(inode)->dir_index) {
3695 ret = btrfs_delayed_delete_inode_ref(inode);
3697 index = BTRFS_I(inode)->dir_index;
3702 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3705 btrfs_info(root->fs_info,
3706 "failed to delete reference to %.*s, inode %llu parent %llu",
3707 name_len, name, ino, dir_ino);
3708 btrfs_abort_transaction(trans, root, ret);
3712 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3714 btrfs_abort_transaction(trans, root, ret);
3718 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3720 if (ret != 0 && ret != -ENOENT) {
3721 btrfs_abort_transaction(trans, root, ret);
3725 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3730 btrfs_abort_transaction(trans, root, ret);
3732 btrfs_free_path(path);
3736 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3737 inode_inc_iversion(inode);
3738 inode_inc_iversion(dir);
3739 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3740 ret = btrfs_update_inode(trans, root, dir);
3745 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3746 struct btrfs_root *root,
3747 struct inode *dir, struct inode *inode,
3748 const char *name, int name_len)
3751 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3754 ret = btrfs_update_inode(trans, root, inode);
3760 * helper to start transaction for unlink and rmdir.
3762 * unlink and rmdir are special in btrfs, they do not always free space, so
3763 * if we cannot make our reservations the normal way try and see if there is
3764 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3765 * allow the unlink to occur.
3767 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3769 struct btrfs_trans_handle *trans;
3770 struct btrfs_root *root = BTRFS_I(dir)->root;
3774 * 1 for the possible orphan item
3775 * 1 for the dir item
3776 * 1 for the dir index
3777 * 1 for the inode ref
3780 trans = btrfs_start_transaction(root, 5);
3781 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3784 if (PTR_ERR(trans) == -ENOSPC) {
3785 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3787 trans = btrfs_start_transaction(root, 0);
3790 ret = btrfs_cond_migrate_bytes(root->fs_info,
3791 &root->fs_info->trans_block_rsv,
3794 btrfs_end_transaction(trans, root);
3795 return ERR_PTR(ret);
3797 trans->block_rsv = &root->fs_info->trans_block_rsv;
3798 trans->bytes_reserved = num_bytes;
3803 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3805 struct btrfs_root *root = BTRFS_I(dir)->root;
3806 struct btrfs_trans_handle *trans;
3807 struct inode *inode = dentry->d_inode;
3810 trans = __unlink_start_trans(dir);
3812 return PTR_ERR(trans);
3814 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3816 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3817 dentry->d_name.name, dentry->d_name.len);
3821 if (inode->i_nlink == 0) {
3822 ret = btrfs_orphan_add(trans, inode);
3828 btrfs_end_transaction(trans, root);
3829 btrfs_btree_balance_dirty(root);
3833 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3834 struct btrfs_root *root,
3835 struct inode *dir, u64 objectid,
3836 const char *name, int name_len)
3838 struct btrfs_path *path;
3839 struct extent_buffer *leaf;
3840 struct btrfs_dir_item *di;
3841 struct btrfs_key key;
3844 u64 dir_ino = btrfs_ino(dir);
3846 path = btrfs_alloc_path();
3850 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3851 name, name_len, -1);
3852 if (IS_ERR_OR_NULL(di)) {
3860 leaf = path->nodes[0];
3861 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3862 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3863 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3865 btrfs_abort_transaction(trans, root, ret);
3868 btrfs_release_path(path);
3870 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3871 objectid, root->root_key.objectid,
3872 dir_ino, &index, name, name_len);
3874 if (ret != -ENOENT) {
3875 btrfs_abort_transaction(trans, root, ret);
3878 di = btrfs_search_dir_index_item(root, path, dir_ino,
3880 if (IS_ERR_OR_NULL(di)) {
3885 btrfs_abort_transaction(trans, root, ret);
3889 leaf = path->nodes[0];
3890 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3891 btrfs_release_path(path);
3894 btrfs_release_path(path);
3896 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3898 btrfs_abort_transaction(trans, root, ret);
3902 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3903 inode_inc_iversion(dir);
3904 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3905 ret = btrfs_update_inode_fallback(trans, root, dir);
3907 btrfs_abort_transaction(trans, root, ret);
3909 btrfs_free_path(path);
3913 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3915 struct inode *inode = dentry->d_inode;
3917 struct btrfs_root *root = BTRFS_I(dir)->root;
3918 struct btrfs_trans_handle *trans;
3920 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3922 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3925 trans = __unlink_start_trans(dir);
3927 return PTR_ERR(trans);
3929 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3930 err = btrfs_unlink_subvol(trans, root, dir,
3931 BTRFS_I(inode)->location.objectid,
3932 dentry->d_name.name,
3933 dentry->d_name.len);
3937 err = btrfs_orphan_add(trans, inode);
3941 /* now the directory is empty */
3942 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3943 dentry->d_name.name, dentry->d_name.len);
3945 btrfs_i_size_write(inode, 0);
3947 btrfs_end_transaction(trans, root);
3948 btrfs_btree_balance_dirty(root);
3954 * this can truncate away extent items, csum items and directory items.
3955 * It starts at a high offset and removes keys until it can't find
3956 * any higher than new_size
3958 * csum items that cross the new i_size are truncated to the new size
3961 * min_type is the minimum key type to truncate down to. If set to 0, this
3962 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3964 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3965 struct btrfs_root *root,
3966 struct inode *inode,
3967 u64 new_size, u32 min_type)
3969 struct btrfs_path *path;
3970 struct extent_buffer *leaf;
3971 struct btrfs_file_extent_item *fi;
3972 struct btrfs_key key;
3973 struct btrfs_key found_key;
3974 u64 extent_start = 0;
3975 u64 extent_num_bytes = 0;
3976 u64 extent_offset = 0;
3978 u64 last_size = (u64)-1;
3979 u32 found_type = (u8)-1;
3982 int pending_del_nr = 0;
3983 int pending_del_slot = 0;
3984 int extent_type = -1;
3987 u64 ino = btrfs_ino(inode);
3989 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3991 path = btrfs_alloc_path();
3997 * We want to drop from the next block forward in case this new size is
3998 * not block aligned since we will be keeping the last block of the
3999 * extent just the way it is.
4001 if (root->ref_cows || root == root->fs_info->tree_root)
4002 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4003 root->sectorsize), (u64)-1, 0);
4006 * This function is also used to drop the items in the log tree before
4007 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4008 * it is used to drop the loged items. So we shouldn't kill the delayed
4011 if (min_type == 0 && root == BTRFS_I(inode)->root)
4012 btrfs_kill_delayed_inode_items(inode);
4015 key.offset = (u64)-1;
4019 path->leave_spinning = 1;
4020 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4027 /* there are no items in the tree for us to truncate, we're
4030 if (path->slots[0] == 0)
4037 leaf = path->nodes[0];
4038 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4039 found_type = btrfs_key_type(&found_key);
4041 if (found_key.objectid != ino)
4044 if (found_type < min_type)
4047 item_end = found_key.offset;
4048 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4049 fi = btrfs_item_ptr(leaf, path->slots[0],
4050 struct btrfs_file_extent_item);
4051 extent_type = btrfs_file_extent_type(leaf, fi);
4052 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4054 btrfs_file_extent_num_bytes(leaf, fi);
4055 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4056 item_end += btrfs_file_extent_inline_len(leaf,
4057 path->slots[0], fi);
4061 if (found_type > min_type) {
4064 if (item_end < new_size)
4066 if (found_key.offset >= new_size)
4072 /* FIXME, shrink the extent if the ref count is only 1 */
4073 if (found_type != BTRFS_EXTENT_DATA_KEY)
4077 last_size = found_key.offset;
4079 last_size = new_size;
4081 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4083 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4085 u64 orig_num_bytes =
4086 btrfs_file_extent_num_bytes(leaf, fi);
4087 extent_num_bytes = ALIGN(new_size -
4090 btrfs_set_file_extent_num_bytes(leaf, fi,
4092 num_dec = (orig_num_bytes -
4094 if (root->ref_cows && extent_start != 0)
4095 inode_sub_bytes(inode, num_dec);
4096 btrfs_mark_buffer_dirty(leaf);
4099 btrfs_file_extent_disk_num_bytes(leaf,
4101 extent_offset = found_key.offset -
4102 btrfs_file_extent_offset(leaf, fi);
4104 /* FIXME blocksize != 4096 */
4105 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4106 if (extent_start != 0) {
4109 inode_sub_bytes(inode, num_dec);
4112 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4114 * we can't truncate inline items that have had
4118 btrfs_file_extent_compression(leaf, fi) == 0 &&
4119 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4120 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4121 u32 size = new_size - found_key.offset;
4123 if (root->ref_cows) {
4124 inode_sub_bytes(inode, item_end + 1 -
4129 * update the ram bytes to properly reflect
4130 * the new size of our item
4132 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4134 btrfs_file_extent_calc_inline_size(size);
4135 btrfs_truncate_item(root, path, size, 1);
4136 } else if (root->ref_cows) {
4137 inode_sub_bytes(inode, item_end + 1 -
4143 if (!pending_del_nr) {
4144 /* no pending yet, add ourselves */
4145 pending_del_slot = path->slots[0];
4147 } else if (pending_del_nr &&
4148 path->slots[0] + 1 == pending_del_slot) {
4149 /* hop on the pending chunk */
4151 pending_del_slot = path->slots[0];
4158 if (found_extent && (root->ref_cows ||
4159 root == root->fs_info->tree_root)) {
4160 btrfs_set_path_blocking(path);
4161 ret = btrfs_free_extent(trans, root, extent_start,
4162 extent_num_bytes, 0,
4163 btrfs_header_owner(leaf),
4164 ino, extent_offset, 0);
4168 if (found_type == BTRFS_INODE_ITEM_KEY)
4171 if (path->slots[0] == 0 ||
4172 path->slots[0] != pending_del_slot) {
4173 if (pending_del_nr) {
4174 ret = btrfs_del_items(trans, root, path,
4178 btrfs_abort_transaction(trans,
4184 btrfs_release_path(path);
4191 if (pending_del_nr) {
4192 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4195 btrfs_abort_transaction(trans, root, ret);
4198 if (last_size != (u64)-1)
4199 btrfs_ordered_update_i_size(inode, last_size, NULL);
4200 btrfs_free_path(path);
4205 * btrfs_truncate_page - read, zero a chunk and write a page
4206 * @inode - inode that we're zeroing
4207 * @from - the offset to start zeroing
4208 * @len - the length to zero, 0 to zero the entire range respective to the
4210 * @front - zero up to the offset instead of from the offset on
4212 * This will find the page for the "from" offset and cow the page and zero the
4213 * part we want to zero. This is used with truncate and hole punching.
4215 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4218 struct address_space *mapping = inode->i_mapping;
4219 struct btrfs_root *root = BTRFS_I(inode)->root;
4220 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4221 struct btrfs_ordered_extent *ordered;
4222 struct extent_state *cached_state = NULL;
4224 u32 blocksize = root->sectorsize;
4225 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4226 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4228 gfp_t mask = btrfs_alloc_write_mask(mapping);
4233 if ((offset & (blocksize - 1)) == 0 &&
4234 (!len || ((len & (blocksize - 1)) == 0)))
4236 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4241 page = find_or_create_page(mapping, index, mask);
4243 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4248 page_start = page_offset(page);
4249 page_end = page_start + PAGE_CACHE_SIZE - 1;
4251 if (!PageUptodate(page)) {
4252 ret = btrfs_readpage(NULL, page);
4254 if (page->mapping != mapping) {
4256 page_cache_release(page);
4259 if (!PageUptodate(page)) {
4264 wait_on_page_writeback(page);
4266 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4267 set_page_extent_mapped(page);
4269 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4271 unlock_extent_cached(io_tree, page_start, page_end,
4272 &cached_state, GFP_NOFS);
4274 page_cache_release(page);
4275 btrfs_start_ordered_extent(inode, ordered, 1);
4276 btrfs_put_ordered_extent(ordered);
4280 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4281 EXTENT_DIRTY | EXTENT_DELALLOC |
4282 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4283 0, 0, &cached_state, GFP_NOFS);
4285 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4288 unlock_extent_cached(io_tree, page_start, page_end,
4289 &cached_state, GFP_NOFS);
4293 if (offset != PAGE_CACHE_SIZE) {
4295 len = PAGE_CACHE_SIZE - offset;
4298 memset(kaddr, 0, offset);
4300 memset(kaddr + offset, 0, len);
4301 flush_dcache_page(page);
4304 ClearPageChecked(page);
4305 set_page_dirty(page);
4306 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4311 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4313 page_cache_release(page);
4318 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4319 u64 offset, u64 len)
4321 struct btrfs_trans_handle *trans;
4325 * Still need to make sure the inode looks like it's been updated so
4326 * that any holes get logged if we fsync.
4328 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4329 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4330 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4331 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4336 * 1 - for the one we're dropping
4337 * 1 - for the one we're adding
4338 * 1 - for updating the inode.
4340 trans = btrfs_start_transaction(root, 3);
4342 return PTR_ERR(trans);
4344 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4346 btrfs_abort_transaction(trans, root, ret);
4347 btrfs_end_transaction(trans, root);
4351 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4352 0, 0, len, 0, len, 0, 0, 0);
4354 btrfs_abort_transaction(trans, root, ret);
4356 btrfs_update_inode(trans, root, inode);
4357 btrfs_end_transaction(trans, root);
4362 * This function puts in dummy file extents for the area we're creating a hole
4363 * for. So if we are truncating this file to a larger size we need to insert
4364 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4365 * the range between oldsize and size
4367 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4369 struct btrfs_root *root = BTRFS_I(inode)->root;
4370 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4371 struct extent_map *em = NULL;
4372 struct extent_state *cached_state = NULL;
4373 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4374 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4375 u64 block_end = ALIGN(size, root->sectorsize);
4382 * If our size started in the middle of a page we need to zero out the
4383 * rest of the page before we expand the i_size, otherwise we could
4384 * expose stale data.
4386 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4390 if (size <= hole_start)
4394 struct btrfs_ordered_extent *ordered;
4396 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4398 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4399 block_end - hole_start);
4402 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4403 &cached_state, GFP_NOFS);
4404 btrfs_start_ordered_extent(inode, ordered, 1);
4405 btrfs_put_ordered_extent(ordered);
4408 cur_offset = hole_start;
4410 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4411 block_end - cur_offset, 0);
4417 last_byte = min(extent_map_end(em), block_end);
4418 last_byte = ALIGN(last_byte , root->sectorsize);
4419 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4420 struct extent_map *hole_em;
4421 hole_size = last_byte - cur_offset;
4423 err = maybe_insert_hole(root, inode, cur_offset,
4427 btrfs_drop_extent_cache(inode, cur_offset,
4428 cur_offset + hole_size - 1, 0);
4429 hole_em = alloc_extent_map();
4431 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4432 &BTRFS_I(inode)->runtime_flags);
4435 hole_em->start = cur_offset;
4436 hole_em->len = hole_size;
4437 hole_em->orig_start = cur_offset;
4439 hole_em->block_start = EXTENT_MAP_HOLE;
4440 hole_em->block_len = 0;
4441 hole_em->orig_block_len = 0;
4442 hole_em->ram_bytes = hole_size;
4443 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4444 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4445 hole_em->generation = root->fs_info->generation;
4448 write_lock(&em_tree->lock);
4449 err = add_extent_mapping(em_tree, hole_em, 1);
4450 write_unlock(&em_tree->lock);
4453 btrfs_drop_extent_cache(inode, cur_offset,
4457 free_extent_map(hole_em);
4460 free_extent_map(em);
4462 cur_offset = last_byte;
4463 if (cur_offset >= block_end)
4466 free_extent_map(em);
4467 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4472 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4474 struct btrfs_root *root = BTRFS_I(inode)->root;
4475 struct btrfs_trans_handle *trans;
4476 loff_t oldsize = i_size_read(inode);
4477 loff_t newsize = attr->ia_size;
4478 int mask = attr->ia_valid;
4482 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4483 * special case where we need to update the times despite not having
4484 * these flags set. For all other operations the VFS set these flags
4485 * explicitly if it wants a timestamp update.
4487 if (newsize != oldsize) {
4488 inode_inc_iversion(inode);
4489 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4490 inode->i_ctime = inode->i_mtime =
4491 current_fs_time(inode->i_sb);
4494 if (newsize > oldsize) {
4495 truncate_pagecache(inode, newsize);
4496 ret = btrfs_cont_expand(inode, oldsize, newsize);
4500 trans = btrfs_start_transaction(root, 1);
4502 return PTR_ERR(trans);
4504 i_size_write(inode, newsize);
4505 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4506 ret = btrfs_update_inode(trans, root, inode);
4507 btrfs_end_transaction(trans, root);
4511 * We're truncating a file that used to have good data down to
4512 * zero. Make sure it gets into the ordered flush list so that
4513 * any new writes get down to disk quickly.
4516 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4517 &BTRFS_I(inode)->runtime_flags);
4520 * 1 for the orphan item we're going to add
4521 * 1 for the orphan item deletion.
4523 trans = btrfs_start_transaction(root, 2);
4525 return PTR_ERR(trans);
4528 * We need to do this in case we fail at _any_ point during the
4529 * actual truncate. Once we do the truncate_setsize we could
4530 * invalidate pages which forces any outstanding ordered io to
4531 * be instantly completed which will give us extents that need
4532 * to be truncated. If we fail to get an orphan inode down we
4533 * could have left over extents that were never meant to live,
4534 * so we need to garuntee from this point on that everything
4535 * will be consistent.
4537 ret = btrfs_orphan_add(trans, inode);
4538 btrfs_end_transaction(trans, root);
4542 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4543 truncate_setsize(inode, newsize);
4545 /* Disable nonlocked read DIO to avoid the end less truncate */
4546 btrfs_inode_block_unlocked_dio(inode);
4547 inode_dio_wait(inode);
4548 btrfs_inode_resume_unlocked_dio(inode);
4550 ret = btrfs_truncate(inode);
4551 if (ret && inode->i_nlink) {
4555 * failed to truncate, disk_i_size is only adjusted down
4556 * as we remove extents, so it should represent the true
4557 * size of the inode, so reset the in memory size and
4558 * delete our orphan entry.
4560 trans = btrfs_join_transaction(root);
4561 if (IS_ERR(trans)) {
4562 btrfs_orphan_del(NULL, inode);
4565 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4566 err = btrfs_orphan_del(trans, inode);
4568 btrfs_abort_transaction(trans, root, err);
4569 btrfs_end_transaction(trans, root);
4576 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4578 struct inode *inode = dentry->d_inode;
4579 struct btrfs_root *root = BTRFS_I(inode)->root;
4582 if (btrfs_root_readonly(root))
4585 err = inode_change_ok(inode, attr);
4589 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4590 err = btrfs_setsize(inode, attr);
4595 if (attr->ia_valid) {
4596 setattr_copy(inode, attr);
4597 inode_inc_iversion(inode);
4598 err = btrfs_dirty_inode(inode);
4600 if (!err && attr->ia_valid & ATTR_MODE)
4601 err = posix_acl_chmod(inode, inode->i_mode);
4608 * While truncating the inode pages during eviction, we get the VFS calling
4609 * btrfs_invalidatepage() against each page of the inode. This is slow because
4610 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4611 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4612 * extent_state structures over and over, wasting lots of time.
4614 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4615 * those expensive operations on a per page basis and do only the ordered io
4616 * finishing, while we release here the extent_map and extent_state structures,
4617 * without the excessive merging and splitting.
4619 static void evict_inode_truncate_pages(struct inode *inode)
4621 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4622 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4623 struct rb_node *node;
4625 ASSERT(inode->i_state & I_FREEING);
4626 truncate_inode_pages_final(&inode->i_data);
4628 write_lock(&map_tree->lock);
4629 while (!RB_EMPTY_ROOT(&map_tree->map)) {
4630 struct extent_map *em;
4632 node = rb_first(&map_tree->map);
4633 em = rb_entry(node, struct extent_map, rb_node);
4634 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4635 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4636 remove_extent_mapping(map_tree, em);
4637 free_extent_map(em);
4639 write_unlock(&map_tree->lock);
4641 spin_lock(&io_tree->lock);
4642 while (!RB_EMPTY_ROOT(&io_tree->state)) {
4643 struct extent_state *state;
4644 struct extent_state *cached_state = NULL;
4646 node = rb_first(&io_tree->state);
4647 state = rb_entry(node, struct extent_state, rb_node);
4648 atomic_inc(&state->refs);
4649 spin_unlock(&io_tree->lock);
4651 lock_extent_bits(io_tree, state->start, state->end,
4653 clear_extent_bit(io_tree, state->start, state->end,
4654 EXTENT_LOCKED | EXTENT_DIRTY |
4655 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
4656 EXTENT_DEFRAG, 1, 1,
4657 &cached_state, GFP_NOFS);
4658 free_extent_state(state);
4660 spin_lock(&io_tree->lock);
4662 spin_unlock(&io_tree->lock);
4665 void btrfs_evict_inode(struct inode *inode)
4667 struct btrfs_trans_handle *trans;
4668 struct btrfs_root *root = BTRFS_I(inode)->root;
4669 struct btrfs_block_rsv *rsv, *global_rsv;
4670 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4673 trace_btrfs_inode_evict(inode);
4675 evict_inode_truncate_pages(inode);
4677 if (inode->i_nlink &&
4678 ((btrfs_root_refs(&root->root_item) != 0 &&
4679 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4680 btrfs_is_free_space_inode(inode)))
4683 if (is_bad_inode(inode)) {
4684 btrfs_orphan_del(NULL, inode);
4687 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4688 btrfs_wait_ordered_range(inode, 0, (u64)-1);
4690 if (root->fs_info->log_root_recovering) {
4691 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4692 &BTRFS_I(inode)->runtime_flags));
4696 if (inode->i_nlink > 0) {
4697 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4698 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
4702 ret = btrfs_commit_inode_delayed_inode(inode);
4704 btrfs_orphan_del(NULL, inode);
4708 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4710 btrfs_orphan_del(NULL, inode);
4713 rsv->size = min_size;
4715 global_rsv = &root->fs_info->global_block_rsv;
4717 btrfs_i_size_write(inode, 0);
4720 * This is a bit simpler than btrfs_truncate since we've already
4721 * reserved our space for our orphan item in the unlink, so we just
4722 * need to reserve some slack space in case we add bytes and update
4723 * inode item when doing the truncate.
4726 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4727 BTRFS_RESERVE_FLUSH_LIMIT);
4730 * Try and steal from the global reserve since we will
4731 * likely not use this space anyway, we want to try as
4732 * hard as possible to get this to work.
4735 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4738 btrfs_warn(root->fs_info,
4739 "Could not get space for a delete, will truncate on mount %d",
4741 btrfs_orphan_del(NULL, inode);
4742 btrfs_free_block_rsv(root, rsv);
4746 trans = btrfs_join_transaction(root);
4747 if (IS_ERR(trans)) {
4748 btrfs_orphan_del(NULL, inode);
4749 btrfs_free_block_rsv(root, rsv);
4753 trans->block_rsv = rsv;
4755 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4759 trans->block_rsv = &root->fs_info->trans_block_rsv;
4760 btrfs_end_transaction(trans, root);
4762 btrfs_btree_balance_dirty(root);
4765 btrfs_free_block_rsv(root, rsv);
4768 * Errors here aren't a big deal, it just means we leave orphan items
4769 * in the tree. They will be cleaned up on the next mount.
4772 trans->block_rsv = root->orphan_block_rsv;
4773 btrfs_orphan_del(trans, inode);
4775 btrfs_orphan_del(NULL, inode);
4778 trans->block_rsv = &root->fs_info->trans_block_rsv;
4779 if (!(root == root->fs_info->tree_root ||
4780 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4781 btrfs_return_ino(root, btrfs_ino(inode));
4783 btrfs_end_transaction(trans, root);
4784 btrfs_btree_balance_dirty(root);
4786 btrfs_remove_delayed_node(inode);
4792 * this returns the key found in the dir entry in the location pointer.
4793 * If no dir entries were found, location->objectid is 0.
4795 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4796 struct btrfs_key *location)
4798 const char *name = dentry->d_name.name;
4799 int namelen = dentry->d_name.len;
4800 struct btrfs_dir_item *di;
4801 struct btrfs_path *path;
4802 struct btrfs_root *root = BTRFS_I(dir)->root;
4805 path = btrfs_alloc_path();
4809 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4814 if (IS_ERR_OR_NULL(di))
4817 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4819 btrfs_free_path(path);
4822 location->objectid = 0;
4827 * when we hit a tree root in a directory, the btrfs part of the inode
4828 * needs to be changed to reflect the root directory of the tree root. This
4829 * is kind of like crossing a mount point.
4831 static int fixup_tree_root_location(struct btrfs_root *root,
4833 struct dentry *dentry,
4834 struct btrfs_key *location,
4835 struct btrfs_root **sub_root)
4837 struct btrfs_path *path;
4838 struct btrfs_root *new_root;
4839 struct btrfs_root_ref *ref;
4840 struct extent_buffer *leaf;
4844 path = btrfs_alloc_path();
4851 ret = btrfs_find_item(root->fs_info->tree_root, path,
4852 BTRFS_I(dir)->root->root_key.objectid,
4853 location->objectid, BTRFS_ROOT_REF_KEY, NULL);
4860 leaf = path->nodes[0];
4861 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4862 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4863 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4866 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4867 (unsigned long)(ref + 1),
4868 dentry->d_name.len);
4872 btrfs_release_path(path);
4874 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4875 if (IS_ERR(new_root)) {
4876 err = PTR_ERR(new_root);
4880 *sub_root = new_root;
4881 location->objectid = btrfs_root_dirid(&new_root->root_item);
4882 location->type = BTRFS_INODE_ITEM_KEY;
4883 location->offset = 0;
4886 btrfs_free_path(path);
4890 static void inode_tree_add(struct inode *inode)
4892 struct btrfs_root *root = BTRFS_I(inode)->root;
4893 struct btrfs_inode *entry;
4895 struct rb_node *parent;
4896 struct rb_node *new = &BTRFS_I(inode)->rb_node;
4897 u64 ino = btrfs_ino(inode);
4899 if (inode_unhashed(inode))
4902 spin_lock(&root->inode_lock);
4903 p = &root->inode_tree.rb_node;
4906 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4908 if (ino < btrfs_ino(&entry->vfs_inode))
4909 p = &parent->rb_left;
4910 else if (ino > btrfs_ino(&entry->vfs_inode))
4911 p = &parent->rb_right;
4913 WARN_ON(!(entry->vfs_inode.i_state &
4914 (I_WILL_FREE | I_FREEING)));
4915 rb_replace_node(parent, new, &root->inode_tree);
4916 RB_CLEAR_NODE(parent);
4917 spin_unlock(&root->inode_lock);
4921 rb_link_node(new, parent, p);
4922 rb_insert_color(new, &root->inode_tree);
4923 spin_unlock(&root->inode_lock);
4926 static void inode_tree_del(struct inode *inode)
4928 struct btrfs_root *root = BTRFS_I(inode)->root;
4931 spin_lock(&root->inode_lock);
4932 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4933 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4934 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4935 empty = RB_EMPTY_ROOT(&root->inode_tree);
4937 spin_unlock(&root->inode_lock);
4939 if (empty && btrfs_root_refs(&root->root_item) == 0) {
4940 synchronize_srcu(&root->fs_info->subvol_srcu);
4941 spin_lock(&root->inode_lock);
4942 empty = RB_EMPTY_ROOT(&root->inode_tree);
4943 spin_unlock(&root->inode_lock);
4945 btrfs_add_dead_root(root);
4949 void btrfs_invalidate_inodes(struct btrfs_root *root)
4951 struct rb_node *node;
4952 struct rb_node *prev;
4953 struct btrfs_inode *entry;
4954 struct inode *inode;
4957 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
4958 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4960 spin_lock(&root->inode_lock);
4962 node = root->inode_tree.rb_node;
4966 entry = rb_entry(node, struct btrfs_inode, rb_node);
4968 if (objectid < btrfs_ino(&entry->vfs_inode))
4969 node = node->rb_left;
4970 else if (objectid > btrfs_ino(&entry->vfs_inode))
4971 node = node->rb_right;
4977 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4978 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4982 prev = rb_next(prev);
4986 entry = rb_entry(node, struct btrfs_inode, rb_node);
4987 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4988 inode = igrab(&entry->vfs_inode);
4990 spin_unlock(&root->inode_lock);
4991 if (atomic_read(&inode->i_count) > 1)
4992 d_prune_aliases(inode);
4994 * btrfs_drop_inode will have it removed from
4995 * the inode cache when its usage count
5000 spin_lock(&root->inode_lock);
5004 if (cond_resched_lock(&root->inode_lock))
5007 node = rb_next(node);
5009 spin_unlock(&root->inode_lock);
5012 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5014 struct btrfs_iget_args *args = p;
5015 inode->i_ino = args->location->objectid;
5016 memcpy(&BTRFS_I(inode)->location, args->location,
5017 sizeof(*args->location));
5018 BTRFS_I(inode)->root = args->root;
5022 static int btrfs_find_actor(struct inode *inode, void *opaque)
5024 struct btrfs_iget_args *args = opaque;
5025 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5026 args->root == BTRFS_I(inode)->root;
5029 static struct inode *btrfs_iget_locked(struct super_block *s,
5030 struct btrfs_key *location,
5031 struct btrfs_root *root)
5033 struct inode *inode;
5034 struct btrfs_iget_args args;
5035 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5037 args.location = location;
5040 inode = iget5_locked(s, hashval, btrfs_find_actor,
5041 btrfs_init_locked_inode,
5046 /* Get an inode object given its location and corresponding root.
5047 * Returns in *is_new if the inode was read from disk
5049 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5050 struct btrfs_root *root, int *new)
5052 struct inode *inode;
5054 inode = btrfs_iget_locked(s, location, root);
5056 return ERR_PTR(-ENOMEM);
5058 if (inode->i_state & I_NEW) {
5059 btrfs_read_locked_inode(inode);
5060 if (!is_bad_inode(inode)) {
5061 inode_tree_add(inode);
5062 unlock_new_inode(inode);
5066 unlock_new_inode(inode);
5068 inode = ERR_PTR(-ESTALE);
5075 static struct inode *new_simple_dir(struct super_block *s,
5076 struct btrfs_key *key,
5077 struct btrfs_root *root)
5079 struct inode *inode = new_inode(s);
5082 return ERR_PTR(-ENOMEM);
5084 BTRFS_I(inode)->root = root;
5085 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5086 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5088 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5089 inode->i_op = &btrfs_dir_ro_inode_operations;
5090 inode->i_fop = &simple_dir_operations;
5091 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5092 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5097 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5099 struct inode *inode;
5100 struct btrfs_root *root = BTRFS_I(dir)->root;
5101 struct btrfs_root *sub_root = root;
5102 struct btrfs_key location;
5106 if (dentry->d_name.len > BTRFS_NAME_LEN)
5107 return ERR_PTR(-ENAMETOOLONG);
5109 ret = btrfs_inode_by_name(dir, dentry, &location);
5111 return ERR_PTR(ret);
5113 if (location.objectid == 0)
5114 return ERR_PTR(-ENOENT);
5116 if (location.type == BTRFS_INODE_ITEM_KEY) {
5117 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5121 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5123 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5124 ret = fixup_tree_root_location(root, dir, dentry,
5125 &location, &sub_root);
5128 inode = ERR_PTR(ret);
5130 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5132 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5134 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5136 if (!IS_ERR(inode) && root != sub_root) {
5137 down_read(&root->fs_info->cleanup_work_sem);
5138 if (!(inode->i_sb->s_flags & MS_RDONLY))
5139 ret = btrfs_orphan_cleanup(sub_root);
5140 up_read(&root->fs_info->cleanup_work_sem);
5143 inode = ERR_PTR(ret);
5150 static int btrfs_dentry_delete(const struct dentry *dentry)
5152 struct btrfs_root *root;
5153 struct inode *inode = dentry->d_inode;
5155 if (!inode && !IS_ROOT(dentry))
5156 inode = dentry->d_parent->d_inode;
5159 root = BTRFS_I(inode)->root;
5160 if (btrfs_root_refs(&root->root_item) == 0)
5163 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5169 static void btrfs_dentry_release(struct dentry *dentry)
5171 if (dentry->d_fsdata)
5172 kfree(dentry->d_fsdata);
5175 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5178 struct inode *inode;
5180 inode = btrfs_lookup_dentry(dir, dentry);
5181 if (IS_ERR(inode)) {
5182 if (PTR_ERR(inode) == -ENOENT)
5185 return ERR_CAST(inode);
5188 return d_materialise_unique(dentry, inode);
5191 unsigned char btrfs_filetype_table[] = {
5192 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5195 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5197 struct inode *inode = file_inode(file);
5198 struct btrfs_root *root = BTRFS_I(inode)->root;
5199 struct btrfs_item *item;
5200 struct btrfs_dir_item *di;
5201 struct btrfs_key key;
5202 struct btrfs_key found_key;
5203 struct btrfs_path *path;
5204 struct list_head ins_list;
5205 struct list_head del_list;
5207 struct extent_buffer *leaf;
5209 unsigned char d_type;
5214 int key_type = BTRFS_DIR_INDEX_KEY;
5218 int is_curr = 0; /* ctx->pos points to the current index? */
5220 /* FIXME, use a real flag for deciding about the key type */
5221 if (root->fs_info->tree_root == root)
5222 key_type = BTRFS_DIR_ITEM_KEY;
5224 if (!dir_emit_dots(file, ctx))
5227 path = btrfs_alloc_path();
5233 if (key_type == BTRFS_DIR_INDEX_KEY) {
5234 INIT_LIST_HEAD(&ins_list);
5235 INIT_LIST_HEAD(&del_list);
5236 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5239 btrfs_set_key_type(&key, key_type);
5240 key.offset = ctx->pos;
5241 key.objectid = btrfs_ino(inode);
5243 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5248 leaf = path->nodes[0];
5249 slot = path->slots[0];
5250 if (slot >= btrfs_header_nritems(leaf)) {
5251 ret = btrfs_next_leaf(root, path);
5259 item = btrfs_item_nr(slot);
5260 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5262 if (found_key.objectid != key.objectid)
5264 if (btrfs_key_type(&found_key) != key_type)
5266 if (found_key.offset < ctx->pos)
5268 if (key_type == BTRFS_DIR_INDEX_KEY &&
5269 btrfs_should_delete_dir_index(&del_list,
5273 ctx->pos = found_key.offset;
5276 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5278 di_total = btrfs_item_size(leaf, item);
5280 while (di_cur < di_total) {
5281 struct btrfs_key location;
5283 if (verify_dir_item(root, leaf, di))
5286 name_len = btrfs_dir_name_len(leaf, di);
5287 if (name_len <= sizeof(tmp_name)) {
5288 name_ptr = tmp_name;
5290 name_ptr = kmalloc(name_len, GFP_NOFS);
5296 read_extent_buffer(leaf, name_ptr,
5297 (unsigned long)(di + 1), name_len);
5299 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5300 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5303 /* is this a reference to our own snapshot? If so
5306 * In contrast to old kernels, we insert the snapshot's
5307 * dir item and dir index after it has been created, so
5308 * we won't find a reference to our own snapshot. We
5309 * still keep the following code for backward
5312 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5313 location.objectid == root->root_key.objectid) {
5317 over = !dir_emit(ctx, name_ptr, name_len,
5318 location.objectid, d_type);
5321 if (name_ptr != tmp_name)
5326 di_len = btrfs_dir_name_len(leaf, di) +
5327 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5329 di = (struct btrfs_dir_item *)((char *)di + di_len);
5335 if (key_type == BTRFS_DIR_INDEX_KEY) {
5338 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5343 /* Reached end of directory/root. Bump pos past the last item. */
5347 * Stop new entries from being returned after we return the last
5350 * New directory entries are assigned a strictly increasing
5351 * offset. This means that new entries created during readdir
5352 * are *guaranteed* to be seen in the future by that readdir.
5353 * This has broken buggy programs which operate on names as
5354 * they're returned by readdir. Until we re-use freed offsets
5355 * we have this hack to stop new entries from being returned
5356 * under the assumption that they'll never reach this huge
5359 * This is being careful not to overflow 32bit loff_t unless the
5360 * last entry requires it because doing so has broken 32bit apps
5363 if (key_type == BTRFS_DIR_INDEX_KEY) {
5364 if (ctx->pos >= INT_MAX)
5365 ctx->pos = LLONG_MAX;
5372 if (key_type == BTRFS_DIR_INDEX_KEY)
5373 btrfs_put_delayed_items(&ins_list, &del_list);
5374 btrfs_free_path(path);
5378 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5380 struct btrfs_root *root = BTRFS_I(inode)->root;
5381 struct btrfs_trans_handle *trans;
5383 bool nolock = false;
5385 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5388 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5391 if (wbc->sync_mode == WB_SYNC_ALL) {
5393 trans = btrfs_join_transaction_nolock(root);
5395 trans = btrfs_join_transaction(root);
5397 return PTR_ERR(trans);
5398 ret = btrfs_commit_transaction(trans, root);
5404 * This is somewhat expensive, updating the tree every time the
5405 * inode changes. But, it is most likely to find the inode in cache.
5406 * FIXME, needs more benchmarking...there are no reasons other than performance
5407 * to keep or drop this code.
5409 static int btrfs_dirty_inode(struct inode *inode)
5411 struct btrfs_root *root = BTRFS_I(inode)->root;
5412 struct btrfs_trans_handle *trans;
5415 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5418 trans = btrfs_join_transaction(root);
5420 return PTR_ERR(trans);
5422 ret = btrfs_update_inode(trans, root, inode);
5423 if (ret && ret == -ENOSPC) {
5424 /* whoops, lets try again with the full transaction */
5425 btrfs_end_transaction(trans, root);
5426 trans = btrfs_start_transaction(root, 1);
5428 return PTR_ERR(trans);
5430 ret = btrfs_update_inode(trans, root, inode);
5432 btrfs_end_transaction(trans, root);
5433 if (BTRFS_I(inode)->delayed_node)
5434 btrfs_balance_delayed_items(root);
5440 * This is a copy of file_update_time. We need this so we can return error on
5441 * ENOSPC for updating the inode in the case of file write and mmap writes.
5443 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5446 struct btrfs_root *root = BTRFS_I(inode)->root;
5448 if (btrfs_root_readonly(root))
5451 if (flags & S_VERSION)
5452 inode_inc_iversion(inode);
5453 if (flags & S_CTIME)
5454 inode->i_ctime = *now;
5455 if (flags & S_MTIME)
5456 inode->i_mtime = *now;
5457 if (flags & S_ATIME)
5458 inode->i_atime = *now;
5459 return btrfs_dirty_inode(inode);
5463 * find the highest existing sequence number in a directory
5464 * and then set the in-memory index_cnt variable to reflect
5465 * free sequence numbers
5467 static int btrfs_set_inode_index_count(struct inode *inode)
5469 struct btrfs_root *root = BTRFS_I(inode)->root;
5470 struct btrfs_key key, found_key;
5471 struct btrfs_path *path;
5472 struct extent_buffer *leaf;
5475 key.objectid = btrfs_ino(inode);
5476 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5477 key.offset = (u64)-1;
5479 path = btrfs_alloc_path();
5483 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5486 /* FIXME: we should be able to handle this */
5492 * MAGIC NUMBER EXPLANATION:
5493 * since we search a directory based on f_pos we have to start at 2
5494 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5495 * else has to start at 2
5497 if (path->slots[0] == 0) {
5498 BTRFS_I(inode)->index_cnt = 2;
5504 leaf = path->nodes[0];
5505 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5507 if (found_key.objectid != btrfs_ino(inode) ||
5508 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5509 BTRFS_I(inode)->index_cnt = 2;
5513 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5515 btrfs_free_path(path);
5520 * helper to find a free sequence number in a given directory. This current
5521 * code is very simple, later versions will do smarter things in the btree
5523 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5527 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5528 ret = btrfs_inode_delayed_dir_index_count(dir);
5530 ret = btrfs_set_inode_index_count(dir);
5536 *index = BTRFS_I(dir)->index_cnt;
5537 BTRFS_I(dir)->index_cnt++;
5542 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5543 struct btrfs_root *root,
5545 const char *name, int name_len,
5546 u64 ref_objectid, u64 objectid,
5547 umode_t mode, u64 *index)
5549 struct inode *inode;
5550 struct btrfs_inode_item *inode_item;
5551 struct btrfs_key *location;
5552 struct btrfs_path *path;
5553 struct btrfs_inode_ref *ref;
5554 struct btrfs_key key[2];
5556 int nitems = name ? 2 : 1;
5560 path = btrfs_alloc_path();
5562 return ERR_PTR(-ENOMEM);
5564 inode = new_inode(root->fs_info->sb);
5566 btrfs_free_path(path);
5567 return ERR_PTR(-ENOMEM);
5571 * we have to initialize this early, so we can reclaim the inode
5572 * number if we fail afterwards in this function.
5574 inode->i_ino = objectid;
5577 trace_btrfs_inode_request(dir);
5579 ret = btrfs_set_inode_index(dir, index);
5581 btrfs_free_path(path);
5583 return ERR_PTR(ret);
5589 * index_cnt is ignored for everything but a dir,
5590 * btrfs_get_inode_index_count has an explanation for the magic
5593 BTRFS_I(inode)->index_cnt = 2;
5594 BTRFS_I(inode)->dir_index = *index;
5595 BTRFS_I(inode)->root = root;
5596 BTRFS_I(inode)->generation = trans->transid;
5597 inode->i_generation = BTRFS_I(inode)->generation;
5600 * We could have gotten an inode number from somebody who was fsynced
5601 * and then removed in this same transaction, so let's just set full
5602 * sync since it will be a full sync anyway and this will blow away the
5603 * old info in the log.
5605 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5607 key[0].objectid = objectid;
5608 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5611 sizes[0] = sizeof(struct btrfs_inode_item);
5615 * Start new inodes with an inode_ref. This is slightly more
5616 * efficient for small numbers of hard links since they will
5617 * be packed into one item. Extended refs will kick in if we
5618 * add more hard links than can fit in the ref item.
5620 key[1].objectid = objectid;
5621 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5622 key[1].offset = ref_objectid;
5624 sizes[1] = name_len + sizeof(*ref);
5627 path->leave_spinning = 1;
5628 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
5632 inode_init_owner(inode, dir, mode);
5633 inode_set_bytes(inode, 0);
5634 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5635 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5636 struct btrfs_inode_item);
5637 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5638 sizeof(*inode_item));
5639 fill_inode_item(trans, path->nodes[0], inode_item, inode);
5642 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5643 struct btrfs_inode_ref);
5644 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5645 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5646 ptr = (unsigned long)(ref + 1);
5647 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5650 btrfs_mark_buffer_dirty(path->nodes[0]);
5651 btrfs_free_path(path);
5653 location = &BTRFS_I(inode)->location;
5654 location->objectid = objectid;
5655 location->offset = 0;
5656 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5658 btrfs_inherit_iflags(inode, dir);
5660 if (S_ISREG(mode)) {
5661 if (btrfs_test_opt(root, NODATASUM))
5662 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5663 if (btrfs_test_opt(root, NODATACOW))
5664 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5665 BTRFS_INODE_NODATASUM;
5668 btrfs_insert_inode_hash(inode);
5669 inode_tree_add(inode);
5671 trace_btrfs_inode_new(inode);
5672 btrfs_set_inode_last_trans(trans, inode);
5674 btrfs_update_root_times(trans, root);
5676 ret = btrfs_inode_inherit_props(trans, inode, dir);
5678 btrfs_err(root->fs_info,
5679 "error inheriting props for ino %llu (root %llu): %d",
5680 btrfs_ino(inode), root->root_key.objectid, ret);
5685 BTRFS_I(dir)->index_cnt--;
5686 btrfs_free_path(path);
5688 return ERR_PTR(ret);
5691 static inline u8 btrfs_inode_type(struct inode *inode)
5693 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5697 * utility function to add 'inode' into 'parent_inode' with
5698 * a give name and a given sequence number.
5699 * if 'add_backref' is true, also insert a backref from the
5700 * inode to the parent directory.
5702 int btrfs_add_link(struct btrfs_trans_handle *trans,
5703 struct inode *parent_inode, struct inode *inode,
5704 const char *name, int name_len, int add_backref, u64 index)
5707 struct btrfs_key key;
5708 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5709 u64 ino = btrfs_ino(inode);
5710 u64 parent_ino = btrfs_ino(parent_inode);
5712 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5713 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5716 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5720 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5721 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5722 key.objectid, root->root_key.objectid,
5723 parent_ino, index, name, name_len);
5724 } else if (add_backref) {
5725 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5729 /* Nothing to clean up yet */
5733 ret = btrfs_insert_dir_item(trans, root, name, name_len,
5735 btrfs_inode_type(inode), index);
5736 if (ret == -EEXIST || ret == -EOVERFLOW)
5739 btrfs_abort_transaction(trans, root, ret);
5743 btrfs_i_size_write(parent_inode, parent_inode->i_size +
5745 inode_inc_iversion(parent_inode);
5746 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5747 ret = btrfs_update_inode(trans, root, parent_inode);
5749 btrfs_abort_transaction(trans, root, ret);
5753 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5756 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5757 key.objectid, root->root_key.objectid,
5758 parent_ino, &local_index, name, name_len);
5760 } else if (add_backref) {
5764 err = btrfs_del_inode_ref(trans, root, name, name_len,
5765 ino, parent_ino, &local_index);
5770 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5771 struct inode *dir, struct dentry *dentry,
5772 struct inode *inode, int backref, u64 index)
5774 int err = btrfs_add_link(trans, dir, inode,
5775 dentry->d_name.name, dentry->d_name.len,
5782 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5783 umode_t mode, dev_t rdev)
5785 struct btrfs_trans_handle *trans;
5786 struct btrfs_root *root = BTRFS_I(dir)->root;
5787 struct inode *inode = NULL;
5793 if (!new_valid_dev(rdev))
5797 * 2 for inode item and ref
5799 * 1 for xattr if selinux is on
5801 trans = btrfs_start_transaction(root, 5);
5803 return PTR_ERR(trans);
5805 err = btrfs_find_free_ino(root, &objectid);
5809 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5810 dentry->d_name.len, btrfs_ino(dir), objectid,
5812 if (IS_ERR(inode)) {
5813 err = PTR_ERR(inode);
5817 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5824 * If the active LSM wants to access the inode during
5825 * d_instantiate it needs these. Smack checks to see
5826 * if the filesystem supports xattrs by looking at the
5830 inode->i_op = &btrfs_special_inode_operations;
5831 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5835 init_special_inode(inode, inode->i_mode, rdev);
5836 btrfs_update_inode(trans, root, inode);
5837 d_instantiate(dentry, inode);
5840 btrfs_end_transaction(trans, root);
5841 btrfs_balance_delayed_items(root);
5842 btrfs_btree_balance_dirty(root);
5844 inode_dec_link_count(inode);
5850 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5851 umode_t mode, bool excl)
5853 struct btrfs_trans_handle *trans;
5854 struct btrfs_root *root = BTRFS_I(dir)->root;
5855 struct inode *inode = NULL;
5856 int drop_inode_on_err = 0;
5862 * 2 for inode item and ref
5864 * 1 for xattr if selinux is on
5866 trans = btrfs_start_transaction(root, 5);
5868 return PTR_ERR(trans);
5870 err = btrfs_find_free_ino(root, &objectid);
5874 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5875 dentry->d_name.len, btrfs_ino(dir), objectid,
5877 if (IS_ERR(inode)) {
5878 err = PTR_ERR(inode);
5881 drop_inode_on_err = 1;
5883 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5887 err = btrfs_update_inode(trans, root, inode);
5892 * If the active LSM wants to access the inode during
5893 * d_instantiate it needs these. Smack checks to see
5894 * if the filesystem supports xattrs by looking at the
5897 inode->i_fop = &btrfs_file_operations;
5898 inode->i_op = &btrfs_file_inode_operations;
5900 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5904 inode->i_mapping->a_ops = &btrfs_aops;
5905 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5906 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5907 d_instantiate(dentry, inode);
5910 btrfs_end_transaction(trans, root);
5911 if (err && drop_inode_on_err) {
5912 inode_dec_link_count(inode);
5915 btrfs_balance_delayed_items(root);
5916 btrfs_btree_balance_dirty(root);
5920 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5921 struct dentry *dentry)
5923 struct btrfs_trans_handle *trans;
5924 struct btrfs_root *root = BTRFS_I(dir)->root;
5925 struct inode *inode = old_dentry->d_inode;
5930 /* do not allow sys_link's with other subvols of the same device */
5931 if (root->objectid != BTRFS_I(inode)->root->objectid)
5934 if (inode->i_nlink >= BTRFS_LINK_MAX)
5937 err = btrfs_set_inode_index(dir, &index);
5942 * 2 items for inode and inode ref
5943 * 2 items for dir items
5944 * 1 item for parent inode
5946 trans = btrfs_start_transaction(root, 5);
5947 if (IS_ERR(trans)) {
5948 err = PTR_ERR(trans);
5952 /* There are several dir indexes for this inode, clear the cache. */
5953 BTRFS_I(inode)->dir_index = 0ULL;
5955 inode_inc_iversion(inode);
5956 inode->i_ctime = CURRENT_TIME;
5958 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5960 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5965 struct dentry *parent = dentry->d_parent;
5966 err = btrfs_update_inode(trans, root, inode);
5969 if (inode->i_nlink == 1) {
5971 * If new hard link count is 1, it's a file created
5972 * with open(2) O_TMPFILE flag.
5974 err = btrfs_orphan_del(trans, inode);
5978 d_instantiate(dentry, inode);
5979 btrfs_log_new_name(trans, inode, NULL, parent);
5982 btrfs_end_transaction(trans, root);
5983 btrfs_balance_delayed_items(root);
5986 inode_dec_link_count(inode);
5989 btrfs_btree_balance_dirty(root);
5993 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5995 struct inode *inode = NULL;
5996 struct btrfs_trans_handle *trans;
5997 struct btrfs_root *root = BTRFS_I(dir)->root;
5999 int drop_on_err = 0;
6004 * 2 items for inode and ref
6005 * 2 items for dir items
6006 * 1 for xattr if selinux is on
6008 trans = btrfs_start_transaction(root, 5);
6010 return PTR_ERR(trans);
6012 err = btrfs_find_free_ino(root, &objectid);
6016 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6017 dentry->d_name.len, btrfs_ino(dir), objectid,
6018 S_IFDIR | mode, &index);
6019 if (IS_ERR(inode)) {
6020 err = PTR_ERR(inode);
6026 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6030 inode->i_op = &btrfs_dir_inode_operations;
6031 inode->i_fop = &btrfs_dir_file_operations;
6033 btrfs_i_size_write(inode, 0);
6034 err = btrfs_update_inode(trans, root, inode);
6038 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6039 dentry->d_name.len, 0, index);
6043 d_instantiate(dentry, inode);
6047 btrfs_end_transaction(trans, root);
6050 btrfs_balance_delayed_items(root);
6051 btrfs_btree_balance_dirty(root);
6055 /* helper for btfs_get_extent. Given an existing extent in the tree,
6056 * and an extent that you want to insert, deal with overlap and insert
6057 * the new extent into the tree.
6059 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6060 struct extent_map *existing,
6061 struct extent_map *em,
6062 u64 map_start, u64 map_len)
6066 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6067 start_diff = map_start - em->start;
6068 em->start = map_start;
6070 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6071 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6072 em->block_start += start_diff;
6073 em->block_len -= start_diff;
6075 return add_extent_mapping(em_tree, em, 0);
6078 static noinline int uncompress_inline(struct btrfs_path *path,
6079 struct inode *inode, struct page *page,
6080 size_t pg_offset, u64 extent_offset,
6081 struct btrfs_file_extent_item *item)
6084 struct extent_buffer *leaf = path->nodes[0];
6087 unsigned long inline_size;
6091 WARN_ON(pg_offset != 0);
6092 compress_type = btrfs_file_extent_compression(leaf, item);
6093 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6094 inline_size = btrfs_file_extent_inline_item_len(leaf,
6095 btrfs_item_nr(path->slots[0]));
6096 tmp = kmalloc(inline_size, GFP_NOFS);
6099 ptr = btrfs_file_extent_inline_start(item);
6101 read_extent_buffer(leaf, tmp, ptr, inline_size);
6103 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6104 ret = btrfs_decompress(compress_type, tmp, page,
6105 extent_offset, inline_size, max_size);
6111 * a bit scary, this does extent mapping from logical file offset to the disk.
6112 * the ugly parts come from merging extents from the disk with the in-ram
6113 * representation. This gets more complex because of the data=ordered code,
6114 * where the in-ram extents might be locked pending data=ordered completion.
6116 * This also copies inline extents directly into the page.
6119 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6120 size_t pg_offset, u64 start, u64 len,
6126 u64 extent_start = 0;
6128 u64 objectid = btrfs_ino(inode);
6130 struct btrfs_path *path = NULL;
6131 struct btrfs_root *root = BTRFS_I(inode)->root;
6132 struct btrfs_file_extent_item *item;
6133 struct extent_buffer *leaf;
6134 struct btrfs_key found_key;
6135 struct extent_map *em = NULL;
6136 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6137 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6138 struct btrfs_trans_handle *trans = NULL;
6142 read_lock(&em_tree->lock);
6143 em = lookup_extent_mapping(em_tree, start, len);
6145 em->bdev = root->fs_info->fs_devices->latest_bdev;
6146 read_unlock(&em_tree->lock);
6149 if (em->start > start || em->start + em->len <= start)
6150 free_extent_map(em);
6151 else if (em->block_start == EXTENT_MAP_INLINE && page)
6152 free_extent_map(em);
6156 em = alloc_extent_map();
6161 em->bdev = root->fs_info->fs_devices->latest_bdev;
6162 em->start = EXTENT_MAP_HOLE;
6163 em->orig_start = EXTENT_MAP_HOLE;
6165 em->block_len = (u64)-1;
6168 path = btrfs_alloc_path();
6174 * Chances are we'll be called again, so go ahead and do
6180 ret = btrfs_lookup_file_extent(trans, root, path,
6181 objectid, start, trans != NULL);
6188 if (path->slots[0] == 0)
6193 leaf = path->nodes[0];
6194 item = btrfs_item_ptr(leaf, path->slots[0],
6195 struct btrfs_file_extent_item);
6196 /* are we inside the extent that was found? */
6197 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6198 found_type = btrfs_key_type(&found_key);
6199 if (found_key.objectid != objectid ||
6200 found_type != BTRFS_EXTENT_DATA_KEY) {
6202 * If we backup past the first extent we want to move forward
6203 * and see if there is an extent in front of us, otherwise we'll
6204 * say there is a hole for our whole search range which can
6211 found_type = btrfs_file_extent_type(leaf, item);
6212 extent_start = found_key.offset;
6213 compress_type = btrfs_file_extent_compression(leaf, item);
6214 if (found_type == BTRFS_FILE_EXTENT_REG ||
6215 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6216 extent_end = extent_start +
6217 btrfs_file_extent_num_bytes(leaf, item);
6218 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6220 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6221 extent_end = ALIGN(extent_start + size, root->sectorsize);
6224 if (start >= extent_end) {
6226 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6227 ret = btrfs_next_leaf(root, path);
6234 leaf = path->nodes[0];
6236 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6237 if (found_key.objectid != objectid ||
6238 found_key.type != BTRFS_EXTENT_DATA_KEY)
6240 if (start + len <= found_key.offset)
6243 em->orig_start = start;
6244 em->len = found_key.offset - start;
6248 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
6249 if (found_type == BTRFS_FILE_EXTENT_REG ||
6250 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6251 em->start = extent_start;
6252 em->len = extent_end - extent_start;
6253 em->orig_start = extent_start -
6254 btrfs_file_extent_offset(leaf, item);
6255 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6257 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6259 em->block_start = EXTENT_MAP_HOLE;
6262 if (compress_type != BTRFS_COMPRESS_NONE) {
6263 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6264 em->compress_type = compress_type;
6265 em->block_start = bytenr;
6266 em->block_len = em->orig_block_len;
6268 bytenr += btrfs_file_extent_offset(leaf, item);
6269 em->block_start = bytenr;
6270 em->block_len = em->len;
6271 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6272 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6275 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6279 size_t extent_offset;
6282 em->block_start = EXTENT_MAP_INLINE;
6283 if (!page || create) {
6284 em->start = extent_start;
6285 em->len = extent_end - extent_start;
6289 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6290 extent_offset = page_offset(page) + pg_offset - extent_start;
6291 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6292 size - extent_offset);
6293 em->start = extent_start + extent_offset;
6294 em->len = ALIGN(copy_size, root->sectorsize);
6295 em->orig_block_len = em->len;
6296 em->orig_start = em->start;
6297 if (compress_type) {
6298 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6299 em->compress_type = compress_type;
6301 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6302 if (create == 0 && !PageUptodate(page)) {
6303 if (btrfs_file_extent_compression(leaf, item) !=
6304 BTRFS_COMPRESS_NONE) {
6305 ret = uncompress_inline(path, inode, page,
6307 extent_offset, item);
6314 read_extent_buffer(leaf, map + pg_offset, ptr,
6316 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6317 memset(map + pg_offset + copy_size, 0,
6318 PAGE_CACHE_SIZE - pg_offset -
6323 flush_dcache_page(page);
6324 } else if (create && PageUptodate(page)) {
6328 free_extent_map(em);
6331 btrfs_release_path(path);
6332 trans = btrfs_join_transaction(root);
6335 return ERR_CAST(trans);
6339 write_extent_buffer(leaf, map + pg_offset, ptr,
6342 btrfs_mark_buffer_dirty(leaf);
6344 set_extent_uptodate(io_tree, em->start,
6345 extent_map_end(em) - 1, NULL, GFP_NOFS);
6348 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6352 em->orig_start = start;
6355 em->block_start = EXTENT_MAP_HOLE;
6356 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6358 btrfs_release_path(path);
6359 if (em->start > start || extent_map_end(em) <= start) {
6360 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6361 em->start, em->len, start, len);
6367 write_lock(&em_tree->lock);
6368 ret = add_extent_mapping(em_tree, em, 0);
6369 /* it is possible that someone inserted the extent into the tree
6370 * while we had the lock dropped. It is also possible that
6371 * an overlapping map exists in the tree
6373 if (ret == -EEXIST) {
6374 struct extent_map *existing;
6378 existing = lookup_extent_mapping(em_tree, start, len);
6379 if (existing && (existing->start > start ||
6380 existing->start + existing->len <= start)) {
6381 free_extent_map(existing);
6385 existing = lookup_extent_mapping(em_tree, em->start,
6388 err = merge_extent_mapping(em_tree, existing,
6391 free_extent_map(existing);
6393 free_extent_map(em);
6398 free_extent_map(em);
6402 free_extent_map(em);
6407 write_unlock(&em_tree->lock);
6410 trace_btrfs_get_extent(root, em);
6413 btrfs_free_path(path);
6415 ret = btrfs_end_transaction(trans, root);
6420 free_extent_map(em);
6421 return ERR_PTR(err);
6423 BUG_ON(!em); /* Error is always set */
6427 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6428 size_t pg_offset, u64 start, u64 len,
6431 struct extent_map *em;
6432 struct extent_map *hole_em = NULL;
6433 u64 range_start = start;
6439 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6446 * - a pre-alloc extent,
6447 * there might actually be delalloc bytes behind it.
6449 if (em->block_start != EXTENT_MAP_HOLE &&
6450 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6456 /* check to see if we've wrapped (len == -1 or similar) */
6465 /* ok, we didn't find anything, lets look for delalloc */
6466 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6467 end, len, EXTENT_DELALLOC, 1);
6468 found_end = range_start + found;
6469 if (found_end < range_start)
6470 found_end = (u64)-1;
6473 * we didn't find anything useful, return
6474 * the original results from get_extent()
6476 if (range_start > end || found_end <= start) {
6482 /* adjust the range_start to make sure it doesn't
6483 * go backwards from the start they passed in
6485 range_start = max(start, range_start);
6486 found = found_end - range_start;
6489 u64 hole_start = start;
6492 em = alloc_extent_map();
6498 * when btrfs_get_extent can't find anything it
6499 * returns one huge hole
6501 * make sure what it found really fits our range, and
6502 * adjust to make sure it is based on the start from
6506 u64 calc_end = extent_map_end(hole_em);
6508 if (calc_end <= start || (hole_em->start > end)) {
6509 free_extent_map(hole_em);
6512 hole_start = max(hole_em->start, start);
6513 hole_len = calc_end - hole_start;
6517 if (hole_em && range_start > hole_start) {
6518 /* our hole starts before our delalloc, so we
6519 * have to return just the parts of the hole
6520 * that go until the delalloc starts
6522 em->len = min(hole_len,
6523 range_start - hole_start);
6524 em->start = hole_start;
6525 em->orig_start = hole_start;
6527 * don't adjust block start at all,
6528 * it is fixed at EXTENT_MAP_HOLE
6530 em->block_start = hole_em->block_start;
6531 em->block_len = hole_len;
6532 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6533 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6535 em->start = range_start;
6537 em->orig_start = range_start;
6538 em->block_start = EXTENT_MAP_DELALLOC;
6539 em->block_len = found;
6541 } else if (hole_em) {
6546 free_extent_map(hole_em);
6548 free_extent_map(em);
6549 return ERR_PTR(err);
6554 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6557 struct btrfs_root *root = BTRFS_I(inode)->root;
6558 struct extent_map *em;
6559 struct btrfs_key ins;
6563 alloc_hint = get_extent_allocation_hint(inode, start, len);
6564 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
6565 alloc_hint, &ins, 1);
6567 return ERR_PTR(ret);
6569 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6570 ins.offset, ins.offset, ins.offset, 0);
6572 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6576 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6577 ins.offset, ins.offset, 0);
6579 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6580 free_extent_map(em);
6581 return ERR_PTR(ret);
6588 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6589 * block must be cow'd
6591 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6592 u64 *orig_start, u64 *orig_block_len,
6595 struct btrfs_trans_handle *trans;
6596 struct btrfs_path *path;
6598 struct extent_buffer *leaf;
6599 struct btrfs_root *root = BTRFS_I(inode)->root;
6600 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6601 struct btrfs_file_extent_item *fi;
6602 struct btrfs_key key;
6609 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6611 path = btrfs_alloc_path();
6615 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
6620 slot = path->slots[0];
6623 /* can't find the item, must cow */
6630 leaf = path->nodes[0];
6631 btrfs_item_key_to_cpu(leaf, &key, slot);
6632 if (key.objectid != btrfs_ino(inode) ||
6633 key.type != BTRFS_EXTENT_DATA_KEY) {
6634 /* not our file or wrong item type, must cow */
6638 if (key.offset > offset) {
6639 /* Wrong offset, must cow */
6643 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6644 found_type = btrfs_file_extent_type(leaf, fi);
6645 if (found_type != BTRFS_FILE_EXTENT_REG &&
6646 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6647 /* not a regular extent, must cow */
6651 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6654 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6655 if (extent_end <= offset)
6658 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6659 if (disk_bytenr == 0)
6662 if (btrfs_file_extent_compression(leaf, fi) ||
6663 btrfs_file_extent_encryption(leaf, fi) ||
6664 btrfs_file_extent_other_encoding(leaf, fi))
6667 backref_offset = btrfs_file_extent_offset(leaf, fi);
6670 *orig_start = key.offset - backref_offset;
6671 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6672 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6675 if (btrfs_extent_readonly(root, disk_bytenr))
6678 num_bytes = min(offset + *len, extent_end) - offset;
6679 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6682 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
6683 ret = test_range_bit(io_tree, offset, range_end,
6684 EXTENT_DELALLOC, 0, NULL);
6691 btrfs_release_path(path);
6694 * look for other files referencing this extent, if we
6695 * find any we must cow
6697 trans = btrfs_join_transaction(root);
6698 if (IS_ERR(trans)) {
6703 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6704 key.offset - backref_offset, disk_bytenr);
6705 btrfs_end_transaction(trans, root);
6712 * adjust disk_bytenr and num_bytes to cover just the bytes
6713 * in this extent we are about to write. If there
6714 * are any csums in that range we have to cow in order
6715 * to keep the csums correct
6717 disk_bytenr += backref_offset;
6718 disk_bytenr += offset - key.offset;
6719 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6722 * all of the above have passed, it is safe to overwrite this extent
6728 btrfs_free_path(path);
6732 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6733 struct extent_state **cached_state, int writing)
6735 struct btrfs_ordered_extent *ordered;
6739 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6742 * We're concerned with the entire range that we're going to be
6743 * doing DIO to, so we need to make sure theres no ordered
6744 * extents in this range.
6746 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6747 lockend - lockstart + 1);
6750 * We need to make sure there are no buffered pages in this
6751 * range either, we could have raced between the invalidate in
6752 * generic_file_direct_write and locking the extent. The
6753 * invalidate needs to happen so that reads after a write do not
6756 if (!ordered && (!writing ||
6757 !test_range_bit(&BTRFS_I(inode)->io_tree,
6758 lockstart, lockend, EXTENT_UPTODATE, 0,
6762 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6763 cached_state, GFP_NOFS);
6766 btrfs_start_ordered_extent(inode, ordered, 1);
6767 btrfs_put_ordered_extent(ordered);
6769 /* Screw you mmap */
6770 ret = filemap_write_and_wait_range(inode->i_mapping,
6777 * If we found a page that couldn't be invalidated just
6778 * fall back to buffered.
6780 ret = invalidate_inode_pages2_range(inode->i_mapping,
6781 lockstart >> PAGE_CACHE_SHIFT,
6782 lockend >> PAGE_CACHE_SHIFT);
6793 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6794 u64 len, u64 orig_start,
6795 u64 block_start, u64 block_len,
6796 u64 orig_block_len, u64 ram_bytes,
6799 struct extent_map_tree *em_tree;
6800 struct extent_map *em;
6801 struct btrfs_root *root = BTRFS_I(inode)->root;
6804 em_tree = &BTRFS_I(inode)->extent_tree;
6805 em = alloc_extent_map();
6807 return ERR_PTR(-ENOMEM);
6810 em->orig_start = orig_start;
6811 em->mod_start = start;
6814 em->block_len = block_len;
6815 em->block_start = block_start;
6816 em->bdev = root->fs_info->fs_devices->latest_bdev;
6817 em->orig_block_len = orig_block_len;
6818 em->ram_bytes = ram_bytes;
6819 em->generation = -1;
6820 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6821 if (type == BTRFS_ORDERED_PREALLOC)
6822 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6825 btrfs_drop_extent_cache(inode, em->start,
6826 em->start + em->len - 1, 0);
6827 write_lock(&em_tree->lock);
6828 ret = add_extent_mapping(em_tree, em, 1);
6829 write_unlock(&em_tree->lock);
6830 } while (ret == -EEXIST);
6833 free_extent_map(em);
6834 return ERR_PTR(ret);
6841 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6842 struct buffer_head *bh_result, int create)
6844 struct extent_map *em;
6845 struct btrfs_root *root = BTRFS_I(inode)->root;
6846 struct extent_state *cached_state = NULL;
6847 u64 start = iblock << inode->i_blkbits;
6848 u64 lockstart, lockend;
6849 u64 len = bh_result->b_size;
6850 int unlock_bits = EXTENT_LOCKED;
6854 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6856 len = min_t(u64, len, root->sectorsize);
6859 lockend = start + len - 1;
6862 * If this errors out it's because we couldn't invalidate pagecache for
6863 * this range and we need to fallback to buffered.
6865 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6868 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6875 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6876 * io. INLINE is special, and we could probably kludge it in here, but
6877 * it's still buffered so for safety lets just fall back to the generic
6880 * For COMPRESSED we _have_ to read the entire extent in so we can
6881 * decompress it, so there will be buffering required no matter what we
6882 * do, so go ahead and fallback to buffered.
6884 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6885 * to buffered IO. Don't blame me, this is the price we pay for using
6888 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6889 em->block_start == EXTENT_MAP_INLINE) {
6890 free_extent_map(em);
6895 /* Just a good old fashioned hole, return */
6896 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6897 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6898 free_extent_map(em);
6903 * We don't allocate a new extent in the following cases
6905 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6907 * 2) The extent is marked as PREALLOC. We're good to go here and can
6908 * just use the extent.
6912 len = min(len, em->len - (start - em->start));
6913 lockstart = start + len;
6917 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6918 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6919 em->block_start != EXTENT_MAP_HOLE)) {
6922 u64 block_start, orig_start, orig_block_len, ram_bytes;
6924 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6925 type = BTRFS_ORDERED_PREALLOC;
6927 type = BTRFS_ORDERED_NOCOW;
6928 len = min(len, em->len - (start - em->start));
6929 block_start = em->block_start + (start - em->start);
6931 if (can_nocow_extent(inode, start, &len, &orig_start,
6932 &orig_block_len, &ram_bytes) == 1) {
6933 if (type == BTRFS_ORDERED_PREALLOC) {
6934 free_extent_map(em);
6935 em = create_pinned_em(inode, start, len,
6944 ret = btrfs_add_ordered_extent_dio(inode, start,
6945 block_start, len, len, type);
6947 free_extent_map(em);
6955 * this will cow the extent, reset the len in case we changed
6958 len = bh_result->b_size;
6959 free_extent_map(em);
6960 em = btrfs_new_extent_direct(inode, start, len);
6965 len = min(len, em->len - (start - em->start));
6967 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6969 bh_result->b_size = len;
6970 bh_result->b_bdev = em->bdev;
6971 set_buffer_mapped(bh_result);
6973 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6974 set_buffer_new(bh_result);
6977 * Need to update the i_size under the extent lock so buffered
6978 * readers will get the updated i_size when we unlock.
6980 if (start + len > i_size_read(inode))
6981 i_size_write(inode, start + len);
6983 spin_lock(&BTRFS_I(inode)->lock);
6984 BTRFS_I(inode)->outstanding_extents++;
6985 spin_unlock(&BTRFS_I(inode)->lock);
6987 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6988 lockstart + len - 1, EXTENT_DELALLOC, NULL,
6989 &cached_state, GFP_NOFS);
6994 * In the case of write we need to clear and unlock the entire range,
6995 * in the case of read we need to unlock only the end area that we
6996 * aren't using if there is any left over space.
6998 if (lockstart < lockend) {
6999 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7000 lockend, unlock_bits, 1, 0,
7001 &cached_state, GFP_NOFS);
7003 free_extent_state(cached_state);
7006 free_extent_map(em);
7011 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7012 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7016 static void btrfs_endio_direct_read(struct bio *bio, int err)
7018 struct btrfs_dio_private *dip = bio->bi_private;
7019 struct bio_vec *bvec;
7020 struct inode *inode = dip->inode;
7021 struct btrfs_root *root = BTRFS_I(inode)->root;
7022 struct bio *dio_bio;
7023 u32 *csums = (u32 *)dip->csum;
7027 start = dip->logical_offset;
7028 bio_for_each_segment_all(bvec, bio, i) {
7029 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
7030 struct page *page = bvec->bv_page;
7033 unsigned long flags;
7035 local_irq_save(flags);
7036 kaddr = kmap_atomic(page);
7037 csum = btrfs_csum_data(kaddr + bvec->bv_offset,
7038 csum, bvec->bv_len);
7039 btrfs_csum_final(csum, (char *)&csum);
7040 kunmap_atomic(kaddr);
7041 local_irq_restore(flags);
7043 flush_dcache_page(bvec->bv_page);
7044 if (csum != csums[i]) {
7045 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
7046 btrfs_ino(inode), start, csum,
7052 start += bvec->bv_len;
7055 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
7056 dip->logical_offset + dip->bytes - 1);
7057 dio_bio = dip->dio_bio;
7061 /* If we had a csum failure make sure to clear the uptodate flag */
7063 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7064 dio_end_io(dio_bio, err);
7068 static void btrfs_endio_direct_write(struct bio *bio, int err)
7070 struct btrfs_dio_private *dip = bio->bi_private;
7071 struct inode *inode = dip->inode;
7072 struct btrfs_root *root = BTRFS_I(inode)->root;
7073 struct btrfs_ordered_extent *ordered = NULL;
7074 u64 ordered_offset = dip->logical_offset;
7075 u64 ordered_bytes = dip->bytes;
7076 struct bio *dio_bio;
7082 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7084 ordered_bytes, !err);
7088 btrfs_init_work(&ordered->work, finish_ordered_fn, NULL, NULL);
7089 btrfs_queue_work(root->fs_info->endio_write_workers,
7093 * our bio might span multiple ordered extents. If we haven't
7094 * completed the accounting for the whole dio, go back and try again
7096 if (ordered_offset < dip->logical_offset + dip->bytes) {
7097 ordered_bytes = dip->logical_offset + dip->bytes -
7103 dio_bio = dip->dio_bio;
7107 /* If we had an error make sure to clear the uptodate flag */
7109 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7110 dio_end_io(dio_bio, err);
7114 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7115 struct bio *bio, int mirror_num,
7116 unsigned long bio_flags, u64 offset)
7119 struct btrfs_root *root = BTRFS_I(inode)->root;
7120 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7121 BUG_ON(ret); /* -ENOMEM */
7125 static void btrfs_end_dio_bio(struct bio *bio, int err)
7127 struct btrfs_dio_private *dip = bio->bi_private;
7130 btrfs_err(BTRFS_I(dip->inode)->root->fs_info,
7131 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7132 btrfs_ino(dip->inode), bio->bi_rw,
7133 (unsigned long long)bio->bi_iter.bi_sector,
7134 bio->bi_iter.bi_size, err);
7138 * before atomic variable goto zero, we must make sure
7139 * dip->errors is perceived to be set.
7141 smp_mb__before_atomic_dec();
7144 /* if there are more bios still pending for this dio, just exit */
7145 if (!atomic_dec_and_test(&dip->pending_bios))
7149 bio_io_error(dip->orig_bio);
7151 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7152 bio_endio(dip->orig_bio, 0);
7158 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7159 u64 first_sector, gfp_t gfp_flags)
7161 int nr_vecs = bio_get_nr_vecs(bdev);
7162 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7165 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7166 int rw, u64 file_offset, int skip_sum,
7169 struct btrfs_dio_private *dip = bio->bi_private;
7170 int write = rw & REQ_WRITE;
7171 struct btrfs_root *root = BTRFS_I(inode)->root;
7175 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7180 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7188 if (write && async_submit) {
7189 ret = btrfs_wq_submit_bio(root->fs_info,
7190 inode, rw, bio, 0, 0,
7192 __btrfs_submit_bio_start_direct_io,
7193 __btrfs_submit_bio_done);
7197 * If we aren't doing async submit, calculate the csum of the
7200 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7203 } else if (!skip_sum) {
7204 ret = btrfs_lookup_bio_sums_dio(root, inode, dip, bio,
7211 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7217 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7220 struct inode *inode = dip->inode;
7221 struct btrfs_root *root = BTRFS_I(inode)->root;
7223 struct bio *orig_bio = dip->orig_bio;
7224 struct bio_vec *bvec = orig_bio->bi_io_vec;
7225 u64 start_sector = orig_bio->bi_iter.bi_sector;
7226 u64 file_offset = dip->logical_offset;
7231 int async_submit = 0;
7233 map_length = orig_bio->bi_iter.bi_size;
7234 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7235 &map_length, NULL, 0);
7241 if (map_length >= orig_bio->bi_iter.bi_size) {
7246 /* async crcs make it difficult to collect full stripe writes. */
7247 if (btrfs_get_alloc_profile(root, 1) &
7248 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7253 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7256 bio->bi_private = dip;
7257 bio->bi_end_io = btrfs_end_dio_bio;
7258 atomic_inc(&dip->pending_bios);
7260 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7261 if (unlikely(map_length < submit_len + bvec->bv_len ||
7262 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7263 bvec->bv_offset) < bvec->bv_len)) {
7265 * inc the count before we submit the bio so
7266 * we know the end IO handler won't happen before
7267 * we inc the count. Otherwise, the dip might get freed
7268 * before we're done setting it up
7270 atomic_inc(&dip->pending_bios);
7271 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7272 file_offset, skip_sum,
7276 atomic_dec(&dip->pending_bios);
7280 start_sector += submit_len >> 9;
7281 file_offset += submit_len;
7286 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7287 start_sector, GFP_NOFS);
7290 bio->bi_private = dip;
7291 bio->bi_end_io = btrfs_end_dio_bio;
7293 map_length = orig_bio->bi_iter.bi_size;
7294 ret = btrfs_map_block(root->fs_info, rw,
7296 &map_length, NULL, 0);
7302 submit_len += bvec->bv_len;
7309 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7318 * before atomic variable goto zero, we must
7319 * make sure dip->errors is perceived to be set.
7321 smp_mb__before_atomic_dec();
7322 if (atomic_dec_and_test(&dip->pending_bios))
7323 bio_io_error(dip->orig_bio);
7325 /* bio_end_io() will handle error, so we needn't return it */
7329 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7330 struct inode *inode, loff_t file_offset)
7332 struct btrfs_root *root = BTRFS_I(inode)->root;
7333 struct btrfs_dio_private *dip;
7337 int write = rw & REQ_WRITE;
7341 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7343 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7349 if (!skip_sum && !write) {
7350 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
7351 sum_len = dio_bio->bi_iter.bi_size >>
7352 inode->i_sb->s_blocksize_bits;
7353 sum_len *= csum_size;
7358 dip = kmalloc(sizeof(*dip) + sum_len, GFP_NOFS);
7364 dip->private = dio_bio->bi_private;
7366 dip->logical_offset = file_offset;
7367 dip->bytes = dio_bio->bi_iter.bi_size;
7368 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
7369 io_bio->bi_private = dip;
7371 dip->orig_bio = io_bio;
7372 dip->dio_bio = dio_bio;
7373 atomic_set(&dip->pending_bios, 0);
7376 io_bio->bi_end_io = btrfs_endio_direct_write;
7378 io_bio->bi_end_io = btrfs_endio_direct_read;
7380 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7389 * If this is a write, we need to clean up the reserved space and kill
7390 * the ordered extent.
7393 struct btrfs_ordered_extent *ordered;
7394 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7395 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7396 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7397 btrfs_free_reserved_extent(root, ordered->start,
7399 btrfs_put_ordered_extent(ordered);
7400 btrfs_put_ordered_extent(ordered);
7402 bio_endio(dio_bio, ret);
7405 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7406 const struct iovec *iov, loff_t offset,
7407 unsigned long nr_segs)
7413 unsigned blocksize_mask = root->sectorsize - 1;
7414 ssize_t retval = -EINVAL;
7415 loff_t end = offset;
7417 if (offset & blocksize_mask)
7420 /* Check the memory alignment. Blocks cannot straddle pages */
7421 for (seg = 0; seg < nr_segs; seg++) {
7422 addr = (unsigned long)iov[seg].iov_base;
7423 size = iov[seg].iov_len;
7425 if ((addr & blocksize_mask) || (size & blocksize_mask))
7428 /* If this is a write we don't need to check anymore */
7433 * Check to make sure we don't have duplicate iov_base's in this
7434 * iovec, if so return EINVAL, otherwise we'll get csum errors
7435 * when reading back.
7437 for (i = seg + 1; i < nr_segs; i++) {
7438 if (iov[seg].iov_base == iov[i].iov_base)
7447 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7448 const struct iovec *iov, loff_t offset,
7449 unsigned long nr_segs)
7451 struct file *file = iocb->ki_filp;
7452 struct inode *inode = file->f_mapping->host;
7456 bool relock = false;
7459 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7463 atomic_inc(&inode->i_dio_count);
7464 smp_mb__after_atomic_inc();
7467 * The generic stuff only does filemap_write_and_wait_range, which
7468 * isn't enough if we've written compressed pages to this area, so
7469 * we need to flush the dirty pages again to make absolutely sure
7470 * that any outstanding dirty pages are on disk.
7472 count = iov_length(iov, nr_segs);
7473 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7474 &BTRFS_I(inode)->runtime_flags))
7475 filemap_fdatawrite_range(inode->i_mapping, offset, count);
7479 * If the write DIO is beyond the EOF, we need update
7480 * the isize, but it is protected by i_mutex. So we can
7481 * not unlock the i_mutex at this case.
7483 if (offset + count <= inode->i_size) {
7484 mutex_unlock(&inode->i_mutex);
7487 ret = btrfs_delalloc_reserve_space(inode, count);
7490 } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7491 &BTRFS_I(inode)->runtime_flags))) {
7492 inode_dio_done(inode);
7493 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7497 ret = __blockdev_direct_IO(rw, iocb, inode,
7498 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7499 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7500 btrfs_submit_direct, flags);
7502 if (ret < 0 && ret != -EIOCBQUEUED)
7503 btrfs_delalloc_release_space(inode, count);
7504 else if (ret >= 0 && (size_t)ret < count)
7505 btrfs_delalloc_release_space(inode,
7506 count - (size_t)ret);
7508 btrfs_delalloc_release_metadata(inode, 0);
7512 inode_dio_done(inode);
7514 mutex_lock(&inode->i_mutex);
7519 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
7521 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7522 __u64 start, __u64 len)
7526 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7530 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7533 int btrfs_readpage(struct file *file, struct page *page)
7535 struct extent_io_tree *tree;
7536 tree = &BTRFS_I(page->mapping->host)->io_tree;
7537 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7540 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7542 struct extent_io_tree *tree;
7545 if (current->flags & PF_MEMALLOC) {
7546 redirty_page_for_writepage(wbc, page);
7550 tree = &BTRFS_I(page->mapping->host)->io_tree;
7551 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7554 static int btrfs_writepages(struct address_space *mapping,
7555 struct writeback_control *wbc)
7557 struct extent_io_tree *tree;
7559 tree = &BTRFS_I(mapping->host)->io_tree;
7560 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7564 btrfs_readpages(struct file *file, struct address_space *mapping,
7565 struct list_head *pages, unsigned nr_pages)
7567 struct extent_io_tree *tree;
7568 tree = &BTRFS_I(mapping->host)->io_tree;
7569 return extent_readpages(tree, mapping, pages, nr_pages,
7572 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7574 struct extent_io_tree *tree;
7575 struct extent_map_tree *map;
7578 tree = &BTRFS_I(page->mapping->host)->io_tree;
7579 map = &BTRFS_I(page->mapping->host)->extent_tree;
7580 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7582 ClearPagePrivate(page);
7583 set_page_private(page, 0);
7584 page_cache_release(page);
7589 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7591 if (PageWriteback(page) || PageDirty(page))
7593 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7596 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
7597 unsigned int length)
7599 struct inode *inode = page->mapping->host;
7600 struct extent_io_tree *tree;
7601 struct btrfs_ordered_extent *ordered;
7602 struct extent_state *cached_state = NULL;
7603 u64 page_start = page_offset(page);
7604 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7605 int inode_evicting = inode->i_state & I_FREEING;
7608 * we have the page locked, so new writeback can't start,
7609 * and the dirty bit won't be cleared while we are here.
7611 * Wait for IO on this page so that we can safely clear
7612 * the PagePrivate2 bit and do ordered accounting
7614 wait_on_page_writeback(page);
7616 tree = &BTRFS_I(inode)->io_tree;
7618 btrfs_releasepage(page, GFP_NOFS);
7622 if (!inode_evicting)
7623 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7624 ordered = btrfs_lookup_ordered_extent(inode, page_start);
7627 * IO on this page will never be started, so we need
7628 * to account for any ordered extents now
7630 if (!inode_evicting)
7631 clear_extent_bit(tree, page_start, page_end,
7632 EXTENT_DIRTY | EXTENT_DELALLOC |
7633 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7634 EXTENT_DEFRAG, 1, 0, &cached_state,
7637 * whoever cleared the private bit is responsible
7638 * for the finish_ordered_io
7640 if (TestClearPagePrivate2(page)) {
7641 struct btrfs_ordered_inode_tree *tree;
7644 tree = &BTRFS_I(inode)->ordered_tree;
7646 spin_lock_irq(&tree->lock);
7647 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
7648 new_len = page_start - ordered->file_offset;
7649 if (new_len < ordered->truncated_len)
7650 ordered->truncated_len = new_len;
7651 spin_unlock_irq(&tree->lock);
7653 if (btrfs_dec_test_ordered_pending(inode, &ordered,
7655 PAGE_CACHE_SIZE, 1))
7656 btrfs_finish_ordered_io(ordered);
7658 btrfs_put_ordered_extent(ordered);
7659 if (!inode_evicting) {
7660 cached_state = NULL;
7661 lock_extent_bits(tree, page_start, page_end, 0,
7666 if (!inode_evicting) {
7667 clear_extent_bit(tree, page_start, page_end,
7668 EXTENT_LOCKED | EXTENT_DIRTY |
7669 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
7670 EXTENT_DEFRAG, 1, 1,
7671 &cached_state, GFP_NOFS);
7673 __btrfs_releasepage(page, GFP_NOFS);
7676 ClearPageChecked(page);
7677 if (PagePrivate(page)) {
7678 ClearPagePrivate(page);
7679 set_page_private(page, 0);
7680 page_cache_release(page);
7685 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7686 * called from a page fault handler when a page is first dirtied. Hence we must
7687 * be careful to check for EOF conditions here. We set the page up correctly
7688 * for a written page which means we get ENOSPC checking when writing into
7689 * holes and correct delalloc and unwritten extent mapping on filesystems that
7690 * support these features.
7692 * We are not allowed to take the i_mutex here so we have to play games to
7693 * protect against truncate races as the page could now be beyond EOF. Because
7694 * vmtruncate() writes the inode size before removing pages, once we have the
7695 * page lock we can determine safely if the page is beyond EOF. If it is not
7696 * beyond EOF, then the page is guaranteed safe against truncation until we
7699 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7701 struct page *page = vmf->page;
7702 struct inode *inode = file_inode(vma->vm_file);
7703 struct btrfs_root *root = BTRFS_I(inode)->root;
7704 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7705 struct btrfs_ordered_extent *ordered;
7706 struct extent_state *cached_state = NULL;
7708 unsigned long zero_start;
7715 sb_start_pagefault(inode->i_sb);
7716 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7718 ret = file_update_time(vma->vm_file);
7724 else /* -ENOSPC, -EIO, etc */
7725 ret = VM_FAULT_SIGBUS;
7731 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7734 size = i_size_read(inode);
7735 page_start = page_offset(page);
7736 page_end = page_start + PAGE_CACHE_SIZE - 1;
7738 if ((page->mapping != inode->i_mapping) ||
7739 (page_start >= size)) {
7740 /* page got truncated out from underneath us */
7743 wait_on_page_writeback(page);
7745 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7746 set_page_extent_mapped(page);
7749 * we can't set the delalloc bits if there are pending ordered
7750 * extents. Drop our locks and wait for them to finish
7752 ordered = btrfs_lookup_ordered_extent(inode, page_start);
7754 unlock_extent_cached(io_tree, page_start, page_end,
7755 &cached_state, GFP_NOFS);
7757 btrfs_start_ordered_extent(inode, ordered, 1);
7758 btrfs_put_ordered_extent(ordered);
7763 * XXX - page_mkwrite gets called every time the page is dirtied, even
7764 * if it was already dirty, so for space accounting reasons we need to
7765 * clear any delalloc bits for the range we are fixing to save. There
7766 * is probably a better way to do this, but for now keep consistent with
7767 * prepare_pages in the normal write path.
7769 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7770 EXTENT_DIRTY | EXTENT_DELALLOC |
7771 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7772 0, 0, &cached_state, GFP_NOFS);
7774 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7777 unlock_extent_cached(io_tree, page_start, page_end,
7778 &cached_state, GFP_NOFS);
7779 ret = VM_FAULT_SIGBUS;
7784 /* page is wholly or partially inside EOF */
7785 if (page_start + PAGE_CACHE_SIZE > size)
7786 zero_start = size & ~PAGE_CACHE_MASK;
7788 zero_start = PAGE_CACHE_SIZE;
7790 if (zero_start != PAGE_CACHE_SIZE) {
7792 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7793 flush_dcache_page(page);
7796 ClearPageChecked(page);
7797 set_page_dirty(page);
7798 SetPageUptodate(page);
7800 BTRFS_I(inode)->last_trans = root->fs_info->generation;
7801 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7802 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7804 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7808 sb_end_pagefault(inode->i_sb);
7809 return VM_FAULT_LOCKED;
7813 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7815 sb_end_pagefault(inode->i_sb);
7819 static int btrfs_truncate(struct inode *inode)
7821 struct btrfs_root *root = BTRFS_I(inode)->root;
7822 struct btrfs_block_rsv *rsv;
7825 struct btrfs_trans_handle *trans;
7826 u64 mask = root->sectorsize - 1;
7827 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7829 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
7835 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
7836 * 3 things going on here
7838 * 1) We need to reserve space for our orphan item and the space to
7839 * delete our orphan item. Lord knows we don't want to have a dangling
7840 * orphan item because we didn't reserve space to remove it.
7842 * 2) We need to reserve space to update our inode.
7844 * 3) We need to have something to cache all the space that is going to
7845 * be free'd up by the truncate operation, but also have some slack
7846 * space reserved in case it uses space during the truncate (thank you
7847 * very much snapshotting).
7849 * And we need these to all be seperate. The fact is we can use alot of
7850 * space doing the truncate, and we have no earthly idea how much space
7851 * we will use, so we need the truncate reservation to be seperate so it
7852 * doesn't end up using space reserved for updating the inode or
7853 * removing the orphan item. We also need to be able to stop the
7854 * transaction and start a new one, which means we need to be able to
7855 * update the inode several times, and we have no idea of knowing how
7856 * many times that will be, so we can't just reserve 1 item for the
7857 * entirety of the opration, so that has to be done seperately as well.
7858 * Then there is the orphan item, which does indeed need to be held on
7859 * to for the whole operation, and we need nobody to touch this reserved
7860 * space except the orphan code.
7862 * So that leaves us with
7864 * 1) root->orphan_block_rsv - for the orphan deletion.
7865 * 2) rsv - for the truncate reservation, which we will steal from the
7866 * transaction reservation.
7867 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7868 * updating the inode.
7870 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7873 rsv->size = min_size;
7877 * 1 for the truncate slack space
7878 * 1 for updating the inode.
7880 trans = btrfs_start_transaction(root, 2);
7881 if (IS_ERR(trans)) {
7882 err = PTR_ERR(trans);
7886 /* Migrate the slack space for the truncate to our reserve */
7887 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7892 * setattr is responsible for setting the ordered_data_close flag,
7893 * but that is only tested during the last file release. That
7894 * could happen well after the next commit, leaving a great big
7895 * window where new writes may get lost if someone chooses to write
7896 * to this file after truncating to zero
7898 * The inode doesn't have any dirty data here, and so if we commit
7899 * this is a noop. If someone immediately starts writing to the inode
7900 * it is very likely we'll catch some of their writes in this
7901 * transaction, and the commit will find this file on the ordered
7902 * data list with good things to send down.
7904 * This is a best effort solution, there is still a window where
7905 * using truncate to replace the contents of the file will
7906 * end up with a zero length file after a crash.
7908 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7909 &BTRFS_I(inode)->runtime_flags))
7910 btrfs_add_ordered_operation(trans, root, inode);
7913 * So if we truncate and then write and fsync we normally would just
7914 * write the extents that changed, which is a problem if we need to
7915 * first truncate that entire inode. So set this flag so we write out
7916 * all of the extents in the inode to the sync log so we're completely
7919 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7920 trans->block_rsv = rsv;
7923 ret = btrfs_truncate_inode_items(trans, root, inode,
7925 BTRFS_EXTENT_DATA_KEY);
7926 if (ret != -ENOSPC) {
7931 trans->block_rsv = &root->fs_info->trans_block_rsv;
7932 ret = btrfs_update_inode(trans, root, inode);
7938 btrfs_end_transaction(trans, root);
7939 btrfs_btree_balance_dirty(root);
7941 trans = btrfs_start_transaction(root, 2);
7942 if (IS_ERR(trans)) {
7943 ret = err = PTR_ERR(trans);
7948 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7950 BUG_ON(ret); /* shouldn't happen */
7951 trans->block_rsv = rsv;
7954 if (ret == 0 && inode->i_nlink > 0) {
7955 trans->block_rsv = root->orphan_block_rsv;
7956 ret = btrfs_orphan_del(trans, inode);
7962 trans->block_rsv = &root->fs_info->trans_block_rsv;
7963 ret = btrfs_update_inode(trans, root, inode);
7967 ret = btrfs_end_transaction(trans, root);
7968 btrfs_btree_balance_dirty(root);
7972 btrfs_free_block_rsv(root, rsv);
7981 * create a new subvolume directory/inode (helper for the ioctl).
7983 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7984 struct btrfs_root *new_root,
7985 struct btrfs_root *parent_root,
7988 struct inode *inode;
7992 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7993 new_dirid, new_dirid,
7994 S_IFDIR | (~current_umask() & S_IRWXUGO),
7997 return PTR_ERR(inode);
7998 inode->i_op = &btrfs_dir_inode_operations;
7999 inode->i_fop = &btrfs_dir_file_operations;
8001 set_nlink(inode, 1);
8002 btrfs_i_size_write(inode, 0);
8004 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8006 btrfs_err(new_root->fs_info,
8007 "error inheriting subvolume %llu properties: %d\n",
8008 new_root->root_key.objectid, err);
8010 err = btrfs_update_inode(trans, new_root, inode);
8016 struct inode *btrfs_alloc_inode(struct super_block *sb)
8018 struct btrfs_inode *ei;
8019 struct inode *inode;
8021 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
8028 ei->last_sub_trans = 0;
8029 ei->logged_trans = 0;
8030 ei->delalloc_bytes = 0;
8031 ei->disk_i_size = 0;
8034 ei->index_cnt = (u64)-1;
8036 ei->last_unlink_trans = 0;
8037 ei->last_log_commit = 0;
8039 spin_lock_init(&ei->lock);
8040 ei->outstanding_extents = 0;
8041 ei->reserved_extents = 0;
8043 ei->runtime_flags = 0;
8044 ei->force_compress = BTRFS_COMPRESS_NONE;
8046 ei->delayed_node = NULL;
8048 inode = &ei->vfs_inode;
8049 extent_map_tree_init(&ei->extent_tree);
8050 extent_io_tree_init(&ei->io_tree, &inode->i_data);
8051 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
8052 ei->io_tree.track_uptodate = 1;
8053 ei->io_failure_tree.track_uptodate = 1;
8054 atomic_set(&ei->sync_writers, 0);
8055 mutex_init(&ei->log_mutex);
8056 mutex_init(&ei->delalloc_mutex);
8057 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8058 INIT_LIST_HEAD(&ei->delalloc_inodes);
8059 INIT_LIST_HEAD(&ei->ordered_operations);
8060 RB_CLEAR_NODE(&ei->rb_node);
8065 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8066 void btrfs_test_destroy_inode(struct inode *inode)
8068 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8069 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8073 static void btrfs_i_callback(struct rcu_head *head)
8075 struct inode *inode = container_of(head, struct inode, i_rcu);
8076 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8079 void btrfs_destroy_inode(struct inode *inode)
8081 struct btrfs_ordered_extent *ordered;
8082 struct btrfs_root *root = BTRFS_I(inode)->root;
8084 WARN_ON(!hlist_empty(&inode->i_dentry));
8085 WARN_ON(inode->i_data.nrpages);
8086 WARN_ON(BTRFS_I(inode)->outstanding_extents);
8087 WARN_ON(BTRFS_I(inode)->reserved_extents);
8088 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8089 WARN_ON(BTRFS_I(inode)->csum_bytes);
8092 * This can happen where we create an inode, but somebody else also
8093 * created the same inode and we need to destroy the one we already
8100 * Make sure we're properly removed from the ordered operation
8104 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
8105 spin_lock(&root->fs_info->ordered_root_lock);
8106 list_del_init(&BTRFS_I(inode)->ordered_operations);
8107 spin_unlock(&root->fs_info->ordered_root_lock);
8110 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8111 &BTRFS_I(inode)->runtime_flags)) {
8112 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
8114 atomic_dec(&root->orphan_inodes);
8118 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8122 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
8123 ordered->file_offset, ordered->len);
8124 btrfs_remove_ordered_extent(inode, ordered);
8125 btrfs_put_ordered_extent(ordered);
8126 btrfs_put_ordered_extent(ordered);
8129 inode_tree_del(inode);
8130 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8132 call_rcu(&inode->i_rcu, btrfs_i_callback);
8135 int btrfs_drop_inode(struct inode *inode)
8137 struct btrfs_root *root = BTRFS_I(inode)->root;
8142 /* the snap/subvol tree is on deleting */
8143 if (btrfs_root_refs(&root->root_item) == 0)
8146 return generic_drop_inode(inode);
8149 static void init_once(void *foo)
8151 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8153 inode_init_once(&ei->vfs_inode);
8156 void btrfs_destroy_cachep(void)
8159 * Make sure all delayed rcu free inodes are flushed before we
8163 if (btrfs_inode_cachep)
8164 kmem_cache_destroy(btrfs_inode_cachep);
8165 if (btrfs_trans_handle_cachep)
8166 kmem_cache_destroy(btrfs_trans_handle_cachep);
8167 if (btrfs_transaction_cachep)
8168 kmem_cache_destroy(btrfs_transaction_cachep);
8169 if (btrfs_path_cachep)
8170 kmem_cache_destroy(btrfs_path_cachep);
8171 if (btrfs_free_space_cachep)
8172 kmem_cache_destroy(btrfs_free_space_cachep);
8173 if (btrfs_delalloc_work_cachep)
8174 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8177 int btrfs_init_cachep(void)
8179 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8180 sizeof(struct btrfs_inode), 0,
8181 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8182 if (!btrfs_inode_cachep)
8185 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8186 sizeof(struct btrfs_trans_handle), 0,
8187 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8188 if (!btrfs_trans_handle_cachep)
8191 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8192 sizeof(struct btrfs_transaction), 0,
8193 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8194 if (!btrfs_transaction_cachep)
8197 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8198 sizeof(struct btrfs_path), 0,
8199 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8200 if (!btrfs_path_cachep)
8203 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8204 sizeof(struct btrfs_free_space), 0,
8205 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8206 if (!btrfs_free_space_cachep)
8209 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8210 sizeof(struct btrfs_delalloc_work), 0,
8211 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8213 if (!btrfs_delalloc_work_cachep)
8218 btrfs_destroy_cachep();
8222 static int btrfs_getattr(struct vfsmount *mnt,
8223 struct dentry *dentry, struct kstat *stat)
8226 struct inode *inode = dentry->d_inode;
8227 u32 blocksize = inode->i_sb->s_blocksize;
8229 generic_fillattr(inode, stat);
8230 stat->dev = BTRFS_I(inode)->root->anon_dev;
8231 stat->blksize = PAGE_CACHE_SIZE;
8233 spin_lock(&BTRFS_I(inode)->lock);
8234 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8235 spin_unlock(&BTRFS_I(inode)->lock);
8236 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8237 ALIGN(delalloc_bytes, blocksize)) >> 9;
8241 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8242 struct inode *new_dir, struct dentry *new_dentry)
8244 struct btrfs_trans_handle *trans;
8245 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8246 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8247 struct inode *new_inode = new_dentry->d_inode;
8248 struct inode *old_inode = old_dentry->d_inode;
8249 struct timespec ctime = CURRENT_TIME;
8253 u64 old_ino = btrfs_ino(old_inode);
8255 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8258 /* we only allow rename subvolume link between subvolumes */
8259 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8262 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8263 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8266 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8267 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8271 /* check for collisions, even if the name isn't there */
8272 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
8273 new_dentry->d_name.name,
8274 new_dentry->d_name.len);
8277 if (ret == -EEXIST) {
8279 * eexist without a new_inode */
8280 if (WARN_ON(!new_inode)) {
8284 /* maybe -EOVERFLOW */
8291 * we're using rename to replace one file with another.
8292 * and the replacement file is large. Start IO on it now so
8293 * we don't add too much work to the end of the transaction
8295 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8296 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8297 filemap_flush(old_inode->i_mapping);
8299 /* close the racy window with snapshot create/destroy ioctl */
8300 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8301 down_read(&root->fs_info->subvol_sem);
8303 * We want to reserve the absolute worst case amount of items. So if
8304 * both inodes are subvols and we need to unlink them then that would
8305 * require 4 item modifications, but if they are both normal inodes it
8306 * would require 5 item modifications, so we'll assume their normal
8307 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8308 * should cover the worst case number of items we'll modify.
8310 trans = btrfs_start_transaction(root, 11);
8311 if (IS_ERR(trans)) {
8312 ret = PTR_ERR(trans);
8317 btrfs_record_root_in_trans(trans, dest);
8319 ret = btrfs_set_inode_index(new_dir, &index);
8323 BTRFS_I(old_inode)->dir_index = 0ULL;
8324 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8325 /* force full log commit if subvolume involved. */
8326 root->fs_info->last_trans_log_full_commit = trans->transid;
8328 ret = btrfs_insert_inode_ref(trans, dest,
8329 new_dentry->d_name.name,
8330 new_dentry->d_name.len,
8332 btrfs_ino(new_dir), index);
8336 * this is an ugly little race, but the rename is required
8337 * to make sure that if we crash, the inode is either at the
8338 * old name or the new one. pinning the log transaction lets
8339 * us make sure we don't allow a log commit to come in after
8340 * we unlink the name but before we add the new name back in.
8342 btrfs_pin_log_trans(root);
8345 * make sure the inode gets flushed if it is replacing
8348 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8349 btrfs_add_ordered_operation(trans, root, old_inode);
8351 inode_inc_iversion(old_dir);
8352 inode_inc_iversion(new_dir);
8353 inode_inc_iversion(old_inode);
8354 old_dir->i_ctime = old_dir->i_mtime = ctime;
8355 new_dir->i_ctime = new_dir->i_mtime = ctime;
8356 old_inode->i_ctime = ctime;
8358 if (old_dentry->d_parent != new_dentry->d_parent)
8359 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8361 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8362 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8363 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8364 old_dentry->d_name.name,
8365 old_dentry->d_name.len);
8367 ret = __btrfs_unlink_inode(trans, root, old_dir,
8368 old_dentry->d_inode,
8369 old_dentry->d_name.name,
8370 old_dentry->d_name.len);
8372 ret = btrfs_update_inode(trans, root, old_inode);
8375 btrfs_abort_transaction(trans, root, ret);
8380 inode_inc_iversion(new_inode);
8381 new_inode->i_ctime = CURRENT_TIME;
8382 if (unlikely(btrfs_ino(new_inode) ==
8383 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8384 root_objectid = BTRFS_I(new_inode)->location.objectid;
8385 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8387 new_dentry->d_name.name,
8388 new_dentry->d_name.len);
8389 BUG_ON(new_inode->i_nlink == 0);
8391 ret = btrfs_unlink_inode(trans, dest, new_dir,
8392 new_dentry->d_inode,
8393 new_dentry->d_name.name,
8394 new_dentry->d_name.len);
8396 if (!ret && new_inode->i_nlink == 0)
8397 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8399 btrfs_abort_transaction(trans, root, ret);
8404 ret = btrfs_add_link(trans, new_dir, old_inode,
8405 new_dentry->d_name.name,
8406 new_dentry->d_name.len, 0, index);
8408 btrfs_abort_transaction(trans, root, ret);
8412 if (old_inode->i_nlink == 1)
8413 BTRFS_I(old_inode)->dir_index = index;
8415 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8416 struct dentry *parent = new_dentry->d_parent;
8417 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8418 btrfs_end_log_trans(root);
8421 btrfs_end_transaction(trans, root);
8423 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8424 up_read(&root->fs_info->subvol_sem);
8429 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8431 struct btrfs_delalloc_work *delalloc_work;
8432 struct inode *inode;
8434 delalloc_work = container_of(work, struct btrfs_delalloc_work,
8436 inode = delalloc_work->inode;
8437 if (delalloc_work->wait) {
8438 btrfs_wait_ordered_range(inode, 0, (u64)-1);
8440 filemap_flush(inode->i_mapping);
8441 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8442 &BTRFS_I(inode)->runtime_flags))
8443 filemap_flush(inode->i_mapping);
8446 if (delalloc_work->delay_iput)
8447 btrfs_add_delayed_iput(inode);
8450 complete(&delalloc_work->completion);
8453 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8454 int wait, int delay_iput)
8456 struct btrfs_delalloc_work *work;
8458 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8462 init_completion(&work->completion);
8463 INIT_LIST_HEAD(&work->list);
8464 work->inode = inode;
8466 work->delay_iput = delay_iput;
8467 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
8472 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8474 wait_for_completion(&work->completion);
8475 kmem_cache_free(btrfs_delalloc_work_cachep, work);
8479 * some fairly slow code that needs optimization. This walks the list
8480 * of all the inodes with pending delalloc and forces them to disk.
8482 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
8485 struct btrfs_inode *binode;
8486 struct inode *inode;
8487 struct btrfs_delalloc_work *work, *next;
8488 struct list_head works;
8489 struct list_head splice;
8492 INIT_LIST_HEAD(&works);
8493 INIT_LIST_HEAD(&splice);
8495 mutex_lock(&root->delalloc_mutex);
8496 spin_lock(&root->delalloc_lock);
8497 list_splice_init(&root->delalloc_inodes, &splice);
8498 while (!list_empty(&splice)) {
8499 binode = list_entry(splice.next, struct btrfs_inode,
8502 list_move_tail(&binode->delalloc_inodes,
8503 &root->delalloc_inodes);
8504 inode = igrab(&binode->vfs_inode);
8506 cond_resched_lock(&root->delalloc_lock);
8509 spin_unlock(&root->delalloc_lock);
8511 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8512 if (unlikely(!work)) {
8514 btrfs_add_delayed_iput(inode);
8520 list_add_tail(&work->list, &works);
8521 btrfs_queue_work(root->fs_info->flush_workers,
8524 if (nr != -1 && ret >= nr)
8527 spin_lock(&root->delalloc_lock);
8529 spin_unlock(&root->delalloc_lock);
8532 list_for_each_entry_safe(work, next, &works, list) {
8533 list_del_init(&work->list);
8534 btrfs_wait_and_free_delalloc_work(work);
8537 if (!list_empty_careful(&splice)) {
8538 spin_lock(&root->delalloc_lock);
8539 list_splice_tail(&splice, &root->delalloc_inodes);
8540 spin_unlock(&root->delalloc_lock);
8542 mutex_unlock(&root->delalloc_mutex);
8546 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8550 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
8553 ret = __start_delalloc_inodes(root, delay_iput, -1);
8557 * the filemap_flush will queue IO into the worker threads, but
8558 * we have to make sure the IO is actually started and that
8559 * ordered extents get created before we return
8561 atomic_inc(&root->fs_info->async_submit_draining);
8562 while (atomic_read(&root->fs_info->nr_async_submits) ||
8563 atomic_read(&root->fs_info->async_delalloc_pages)) {
8564 wait_event(root->fs_info->async_submit_wait,
8565 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8566 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8568 atomic_dec(&root->fs_info->async_submit_draining);
8572 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
8575 struct btrfs_root *root;
8576 struct list_head splice;
8579 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
8582 INIT_LIST_HEAD(&splice);
8584 mutex_lock(&fs_info->delalloc_root_mutex);
8585 spin_lock(&fs_info->delalloc_root_lock);
8586 list_splice_init(&fs_info->delalloc_roots, &splice);
8587 while (!list_empty(&splice) && nr) {
8588 root = list_first_entry(&splice, struct btrfs_root,
8590 root = btrfs_grab_fs_root(root);
8592 list_move_tail(&root->delalloc_root,
8593 &fs_info->delalloc_roots);
8594 spin_unlock(&fs_info->delalloc_root_lock);
8596 ret = __start_delalloc_inodes(root, delay_iput, nr);
8597 btrfs_put_fs_root(root);
8605 spin_lock(&fs_info->delalloc_root_lock);
8607 spin_unlock(&fs_info->delalloc_root_lock);
8610 atomic_inc(&fs_info->async_submit_draining);
8611 while (atomic_read(&fs_info->nr_async_submits) ||
8612 atomic_read(&fs_info->async_delalloc_pages)) {
8613 wait_event(fs_info->async_submit_wait,
8614 (atomic_read(&fs_info->nr_async_submits) == 0 &&
8615 atomic_read(&fs_info->async_delalloc_pages) == 0));
8617 atomic_dec(&fs_info->async_submit_draining);
8619 if (!list_empty_careful(&splice)) {
8620 spin_lock(&fs_info->delalloc_root_lock);
8621 list_splice_tail(&splice, &fs_info->delalloc_roots);
8622 spin_unlock(&fs_info->delalloc_root_lock);
8624 mutex_unlock(&fs_info->delalloc_root_mutex);
8628 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8629 const char *symname)
8631 struct btrfs_trans_handle *trans;
8632 struct btrfs_root *root = BTRFS_I(dir)->root;
8633 struct btrfs_path *path;
8634 struct btrfs_key key;
8635 struct inode *inode = NULL;
8643 struct btrfs_file_extent_item *ei;
8644 struct extent_buffer *leaf;
8646 name_len = strlen(symname);
8647 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8648 return -ENAMETOOLONG;
8651 * 2 items for inode item and ref
8652 * 2 items for dir items
8653 * 1 item for xattr if selinux is on
8655 trans = btrfs_start_transaction(root, 5);
8657 return PTR_ERR(trans);
8659 err = btrfs_find_free_ino(root, &objectid);
8663 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8664 dentry->d_name.len, btrfs_ino(dir), objectid,
8665 S_IFLNK|S_IRWXUGO, &index);
8666 if (IS_ERR(inode)) {
8667 err = PTR_ERR(inode);
8671 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8678 * If the active LSM wants to access the inode during
8679 * d_instantiate it needs these. Smack checks to see
8680 * if the filesystem supports xattrs by looking at the
8683 inode->i_fop = &btrfs_file_operations;
8684 inode->i_op = &btrfs_file_inode_operations;
8686 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8690 inode->i_mapping->a_ops = &btrfs_aops;
8691 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8692 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8697 path = btrfs_alloc_path();
8703 key.objectid = btrfs_ino(inode);
8705 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8706 datasize = btrfs_file_extent_calc_inline_size(name_len);
8707 err = btrfs_insert_empty_item(trans, root, path, &key,
8711 btrfs_free_path(path);
8714 leaf = path->nodes[0];
8715 ei = btrfs_item_ptr(leaf, path->slots[0],
8716 struct btrfs_file_extent_item);
8717 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8718 btrfs_set_file_extent_type(leaf, ei,
8719 BTRFS_FILE_EXTENT_INLINE);
8720 btrfs_set_file_extent_encryption(leaf, ei, 0);
8721 btrfs_set_file_extent_compression(leaf, ei, 0);
8722 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8723 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8725 ptr = btrfs_file_extent_inline_start(ei);
8726 write_extent_buffer(leaf, symname, ptr, name_len);
8727 btrfs_mark_buffer_dirty(leaf);
8728 btrfs_free_path(path);
8730 inode->i_op = &btrfs_symlink_inode_operations;
8731 inode->i_mapping->a_ops = &btrfs_symlink_aops;
8732 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8733 inode_set_bytes(inode, name_len);
8734 btrfs_i_size_write(inode, name_len);
8735 err = btrfs_update_inode(trans, root, inode);
8741 d_instantiate(dentry, inode);
8742 btrfs_end_transaction(trans, root);
8744 inode_dec_link_count(inode);
8747 btrfs_btree_balance_dirty(root);
8751 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8752 u64 start, u64 num_bytes, u64 min_size,
8753 loff_t actual_len, u64 *alloc_hint,
8754 struct btrfs_trans_handle *trans)
8756 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8757 struct extent_map *em;
8758 struct btrfs_root *root = BTRFS_I(inode)->root;
8759 struct btrfs_key ins;
8760 u64 cur_offset = start;
8764 bool own_trans = true;
8768 while (num_bytes > 0) {
8770 trans = btrfs_start_transaction(root, 3);
8771 if (IS_ERR(trans)) {
8772 ret = PTR_ERR(trans);
8777 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8778 cur_bytes = max(cur_bytes, min_size);
8779 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
8780 *alloc_hint, &ins, 1);
8783 btrfs_end_transaction(trans, root);
8787 ret = insert_reserved_file_extent(trans, inode,
8788 cur_offset, ins.objectid,
8789 ins.offset, ins.offset,
8790 ins.offset, 0, 0, 0,
8791 BTRFS_FILE_EXTENT_PREALLOC);
8793 btrfs_free_reserved_extent(root, ins.objectid,
8795 btrfs_abort_transaction(trans, root, ret);
8797 btrfs_end_transaction(trans, root);
8800 btrfs_drop_extent_cache(inode, cur_offset,
8801 cur_offset + ins.offset -1, 0);
8803 em = alloc_extent_map();
8805 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8806 &BTRFS_I(inode)->runtime_flags);
8810 em->start = cur_offset;
8811 em->orig_start = cur_offset;
8812 em->len = ins.offset;
8813 em->block_start = ins.objectid;
8814 em->block_len = ins.offset;
8815 em->orig_block_len = ins.offset;
8816 em->ram_bytes = ins.offset;
8817 em->bdev = root->fs_info->fs_devices->latest_bdev;
8818 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8819 em->generation = trans->transid;
8822 write_lock(&em_tree->lock);
8823 ret = add_extent_mapping(em_tree, em, 1);
8824 write_unlock(&em_tree->lock);
8827 btrfs_drop_extent_cache(inode, cur_offset,
8828 cur_offset + ins.offset - 1,
8831 free_extent_map(em);
8833 num_bytes -= ins.offset;
8834 cur_offset += ins.offset;
8835 *alloc_hint = ins.objectid + ins.offset;
8837 inode_inc_iversion(inode);
8838 inode->i_ctime = CURRENT_TIME;
8839 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8840 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8841 (actual_len > inode->i_size) &&
8842 (cur_offset > inode->i_size)) {
8843 if (cur_offset > actual_len)
8844 i_size = actual_len;
8846 i_size = cur_offset;
8847 i_size_write(inode, i_size);
8848 btrfs_ordered_update_i_size(inode, i_size, NULL);
8851 ret = btrfs_update_inode(trans, root, inode);
8854 btrfs_abort_transaction(trans, root, ret);
8856 btrfs_end_transaction(trans, root);
8861 btrfs_end_transaction(trans, root);
8866 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8867 u64 start, u64 num_bytes, u64 min_size,
8868 loff_t actual_len, u64 *alloc_hint)
8870 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8871 min_size, actual_len, alloc_hint,
8875 int btrfs_prealloc_file_range_trans(struct inode *inode,
8876 struct btrfs_trans_handle *trans, int mode,
8877 u64 start, u64 num_bytes, u64 min_size,
8878 loff_t actual_len, u64 *alloc_hint)
8880 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8881 min_size, actual_len, alloc_hint, trans);
8884 static int btrfs_set_page_dirty(struct page *page)
8886 return __set_page_dirty_nobuffers(page);
8889 static int btrfs_permission(struct inode *inode, int mask)
8891 struct btrfs_root *root = BTRFS_I(inode)->root;
8892 umode_t mode = inode->i_mode;
8894 if (mask & MAY_WRITE &&
8895 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8896 if (btrfs_root_readonly(root))
8898 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8901 return generic_permission(inode, mask);
8904 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
8906 struct btrfs_trans_handle *trans;
8907 struct btrfs_root *root = BTRFS_I(dir)->root;
8908 struct inode *inode = NULL;
8914 * 5 units required for adding orphan entry
8916 trans = btrfs_start_transaction(root, 5);
8918 return PTR_ERR(trans);
8920 ret = btrfs_find_free_ino(root, &objectid);
8924 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
8925 btrfs_ino(dir), objectid, mode, &index);
8926 if (IS_ERR(inode)) {
8927 ret = PTR_ERR(inode);
8932 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
8936 ret = btrfs_update_inode(trans, root, inode);
8940 inode->i_fop = &btrfs_file_operations;
8941 inode->i_op = &btrfs_file_inode_operations;
8943 inode->i_mapping->a_ops = &btrfs_aops;
8944 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8945 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8947 ret = btrfs_orphan_add(trans, inode);
8951 d_tmpfile(dentry, inode);
8952 mark_inode_dirty(inode);
8955 btrfs_end_transaction(trans, root);
8958 btrfs_balance_delayed_items(root);
8959 btrfs_btree_balance_dirty(root);
8964 static const struct inode_operations btrfs_dir_inode_operations = {
8965 .getattr = btrfs_getattr,
8966 .lookup = btrfs_lookup,
8967 .create = btrfs_create,
8968 .unlink = btrfs_unlink,
8970 .mkdir = btrfs_mkdir,
8971 .rmdir = btrfs_rmdir,
8972 .rename = btrfs_rename,
8973 .symlink = btrfs_symlink,
8974 .setattr = btrfs_setattr,
8975 .mknod = btrfs_mknod,
8976 .setxattr = btrfs_setxattr,
8977 .getxattr = btrfs_getxattr,
8978 .listxattr = btrfs_listxattr,
8979 .removexattr = btrfs_removexattr,
8980 .permission = btrfs_permission,
8981 .get_acl = btrfs_get_acl,
8982 .set_acl = btrfs_set_acl,
8983 .update_time = btrfs_update_time,
8984 .tmpfile = btrfs_tmpfile,
8986 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8987 .lookup = btrfs_lookup,
8988 .permission = btrfs_permission,
8989 .get_acl = btrfs_get_acl,
8990 .set_acl = btrfs_set_acl,
8991 .update_time = btrfs_update_time,
8994 static const struct file_operations btrfs_dir_file_operations = {
8995 .llseek = generic_file_llseek,
8996 .read = generic_read_dir,
8997 .iterate = btrfs_real_readdir,
8998 .unlocked_ioctl = btrfs_ioctl,
8999 #ifdef CONFIG_COMPAT
9000 .compat_ioctl = btrfs_ioctl,
9002 .release = btrfs_release_file,
9003 .fsync = btrfs_sync_file,
9006 static struct extent_io_ops btrfs_extent_io_ops = {
9007 .fill_delalloc = run_delalloc_range,
9008 .submit_bio_hook = btrfs_submit_bio_hook,
9009 .merge_bio_hook = btrfs_merge_bio_hook,
9010 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
9011 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
9012 .writepage_start_hook = btrfs_writepage_start_hook,
9013 .set_bit_hook = btrfs_set_bit_hook,
9014 .clear_bit_hook = btrfs_clear_bit_hook,
9015 .merge_extent_hook = btrfs_merge_extent_hook,
9016 .split_extent_hook = btrfs_split_extent_hook,
9020 * btrfs doesn't support the bmap operation because swapfiles
9021 * use bmap to make a mapping of extents in the file. They assume
9022 * these extents won't change over the life of the file and they
9023 * use the bmap result to do IO directly to the drive.
9025 * the btrfs bmap call would return logical addresses that aren't
9026 * suitable for IO and they also will change frequently as COW
9027 * operations happen. So, swapfile + btrfs == corruption.
9029 * For now we're avoiding this by dropping bmap.
9031 static const struct address_space_operations btrfs_aops = {
9032 .readpage = btrfs_readpage,
9033 .writepage = btrfs_writepage,
9034 .writepages = btrfs_writepages,
9035 .readpages = btrfs_readpages,
9036 .direct_IO = btrfs_direct_IO,
9037 .invalidatepage = btrfs_invalidatepage,
9038 .releasepage = btrfs_releasepage,
9039 .set_page_dirty = btrfs_set_page_dirty,
9040 .error_remove_page = generic_error_remove_page,
9043 static const struct address_space_operations btrfs_symlink_aops = {
9044 .readpage = btrfs_readpage,
9045 .writepage = btrfs_writepage,
9046 .invalidatepage = btrfs_invalidatepage,
9047 .releasepage = btrfs_releasepage,
9050 static const struct inode_operations btrfs_file_inode_operations = {
9051 .getattr = btrfs_getattr,
9052 .setattr = btrfs_setattr,
9053 .setxattr = btrfs_setxattr,
9054 .getxattr = btrfs_getxattr,
9055 .listxattr = btrfs_listxattr,
9056 .removexattr = btrfs_removexattr,
9057 .permission = btrfs_permission,
9058 .fiemap = btrfs_fiemap,
9059 .get_acl = btrfs_get_acl,
9060 .set_acl = btrfs_set_acl,
9061 .update_time = btrfs_update_time,
9063 static const struct inode_operations btrfs_special_inode_operations = {
9064 .getattr = btrfs_getattr,
9065 .setattr = btrfs_setattr,
9066 .permission = btrfs_permission,
9067 .setxattr = btrfs_setxattr,
9068 .getxattr = btrfs_getxattr,
9069 .listxattr = btrfs_listxattr,
9070 .removexattr = btrfs_removexattr,
9071 .get_acl = btrfs_get_acl,
9072 .set_acl = btrfs_set_acl,
9073 .update_time = btrfs_update_time,
9075 static const struct inode_operations btrfs_symlink_inode_operations = {
9076 .readlink = generic_readlink,
9077 .follow_link = page_follow_link_light,
9078 .put_link = page_put_link,
9079 .getattr = btrfs_getattr,
9080 .setattr = btrfs_setattr,
9081 .permission = btrfs_permission,
9082 .setxattr = btrfs_setxattr,
9083 .getxattr = btrfs_getxattr,
9084 .listxattr = btrfs_listxattr,
9085 .removexattr = btrfs_removexattr,
9086 .update_time = btrfs_update_time,
9089 const struct dentry_operations btrfs_dentry_operations = {
9090 .d_delete = btrfs_dentry_delete,
9091 .d_release = btrfs_dentry_release,