2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/aio.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/xattr.h>
38 #include <linux/posix_acl.h>
39 #include <linux/falloc.h>
40 #include <linux/slab.h>
41 #include <linux/ratelimit.h>
42 #include <linux/mount.h>
43 #include <linux/btrfs.h>
44 #include <linux/blkdev.h>
45 #include <linux/posix_acl_xattr.h>
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "print-tree.h"
51 #include "ordered-data.h"
55 #include "compression.h"
57 #include "free-space-cache.h"
58 #include "inode-map.h"
63 struct btrfs_iget_args {
64 struct btrfs_key *location;
65 struct btrfs_root *root;
68 static const struct inode_operations btrfs_dir_inode_operations;
69 static const struct inode_operations btrfs_symlink_inode_operations;
70 static const struct inode_operations btrfs_dir_ro_inode_operations;
71 static const struct inode_operations btrfs_special_inode_operations;
72 static const struct inode_operations btrfs_file_inode_operations;
73 static const struct address_space_operations btrfs_aops;
74 static const struct address_space_operations btrfs_symlink_aops;
75 static const struct file_operations btrfs_dir_file_operations;
76 static struct extent_io_ops btrfs_extent_io_ops;
78 static struct kmem_cache *btrfs_inode_cachep;
79 static struct kmem_cache *btrfs_delalloc_work_cachep;
80 struct kmem_cache *btrfs_trans_handle_cachep;
81 struct kmem_cache *btrfs_transaction_cachep;
82 struct kmem_cache *btrfs_path_cachep;
83 struct kmem_cache *btrfs_free_space_cachep;
86 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
87 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
88 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
89 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
90 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
91 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
92 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
93 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
96 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
97 static int btrfs_truncate(struct inode *inode);
98 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
99 static noinline int cow_file_range(struct inode *inode,
100 struct page *locked_page,
101 u64 start, u64 end, int *page_started,
102 unsigned long *nr_written, int unlock);
103 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
104 u64 len, u64 orig_start,
105 u64 block_start, u64 block_len,
106 u64 orig_block_len, u64 ram_bytes,
109 static int btrfs_dirty_inode(struct inode *inode);
111 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
112 struct inode *inode, struct inode *dir,
113 const struct qstr *qstr)
117 err = btrfs_init_acl(trans, inode, dir);
119 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
124 * this does all the hard work for inserting an inline extent into
125 * the btree. The caller should have done a btrfs_drop_extents so that
126 * no overlapping inline items exist in the btree
128 static int insert_inline_extent(struct btrfs_trans_handle *trans,
129 struct btrfs_path *path, int extent_inserted,
130 struct btrfs_root *root, struct inode *inode,
131 u64 start, size_t size, size_t compressed_size,
133 struct page **compressed_pages)
135 struct extent_buffer *leaf;
136 struct page *page = NULL;
139 struct btrfs_file_extent_item *ei;
142 size_t cur_size = size;
143 unsigned long offset;
145 if (compressed_size && compressed_pages)
146 cur_size = compressed_size;
148 inode_add_bytes(inode, size);
150 if (!extent_inserted) {
151 struct btrfs_key key;
154 key.objectid = btrfs_ino(inode);
156 key.type = 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);
351 static inline int inode_need_compress(struct inode *inode)
353 struct btrfs_root *root = BTRFS_I(inode)->root;
356 if (btrfs_test_opt(root, FORCE_COMPRESS))
358 /* bad compression ratios */
359 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
361 if (btrfs_test_opt(root, COMPRESS) ||
362 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
363 BTRFS_I(inode)->force_compress)
369 * we create compressed extents in two phases. The first
370 * phase compresses a range of pages that have already been
371 * locked (both pages and state bits are locked).
373 * This is done inside an ordered work queue, and the compression
374 * is spread across many cpus. The actual IO submission is step
375 * two, and the ordered work queue takes care of making sure that
376 * happens in the same order things were put onto the queue by
377 * writepages and friends.
379 * If this code finds it can't get good compression, it puts an
380 * entry onto the work queue to write the uncompressed bytes. This
381 * makes sure that both compressed inodes and uncompressed inodes
382 * are written in the same order that the flusher thread sent them
385 static noinline void compress_file_range(struct inode *inode,
386 struct page *locked_page,
388 struct async_cow *async_cow,
391 struct btrfs_root *root = BTRFS_I(inode)->root;
393 u64 blocksize = root->sectorsize;
395 u64 isize = i_size_read(inode);
397 struct page **pages = NULL;
398 unsigned long nr_pages;
399 unsigned long nr_pages_ret = 0;
400 unsigned long total_compressed = 0;
401 unsigned long total_in = 0;
402 unsigned long max_compressed = 128 * 1024;
403 unsigned long max_uncompressed = 128 * 1024;
406 int compress_type = root->fs_info->compress_type;
409 /* if this is a small write inside eof, kick off a defrag */
410 if ((end - start + 1) < 16 * 1024 &&
411 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
412 btrfs_add_inode_defrag(NULL, inode);
414 actual_end = min_t(u64, isize, end + 1);
417 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
418 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
421 * we don't want to send crud past the end of i_size through
422 * compression, that's just a waste of CPU time. So, if the
423 * end of the file is before the start of our current
424 * requested range of bytes, we bail out to the uncompressed
425 * cleanup code that can deal with all of this.
427 * It isn't really the fastest way to fix things, but this is a
428 * very uncommon corner.
430 if (actual_end <= start)
431 goto cleanup_and_bail_uncompressed;
433 total_compressed = actual_end - start;
436 * skip compression for a small file range(<=blocksize) that
437 * isn't an inline extent, since it dosen't save disk space at all.
439 if (total_compressed <= blocksize &&
440 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
441 goto cleanup_and_bail_uncompressed;
443 /* we want to make sure that amount of ram required to uncompress
444 * an extent is reasonable, so we limit the total size in ram
445 * of a compressed extent to 128k. This is a crucial number
446 * because it also controls how easily we can spread reads across
447 * cpus for decompression.
449 * We also want to make sure the amount of IO required to do
450 * a random read is reasonably small, so we limit the size of
451 * a compressed extent to 128k.
453 total_compressed = min(total_compressed, max_uncompressed);
454 num_bytes = ALIGN(end - start + 1, blocksize);
455 num_bytes = max(blocksize, num_bytes);
460 * we do compression for mount -o compress and when the
461 * inode has not been flagged as nocompress. This flag can
462 * change at any time if we discover bad compression ratios.
464 if (inode_need_compress(inode)) {
466 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
468 /* just bail out to the uncompressed code */
472 if (BTRFS_I(inode)->force_compress)
473 compress_type = BTRFS_I(inode)->force_compress;
476 * we need to call clear_page_dirty_for_io on each
477 * page in the range. Otherwise applications with the file
478 * mmap'd can wander in and change the page contents while
479 * we are compressing them.
481 * If the compression fails for any reason, we set the pages
482 * dirty again later on.
484 extent_range_clear_dirty_for_io(inode, start, end);
486 ret = btrfs_compress_pages(compress_type,
487 inode->i_mapping, start,
488 total_compressed, pages,
489 nr_pages, &nr_pages_ret,
495 unsigned long offset = total_compressed &
496 (PAGE_CACHE_SIZE - 1);
497 struct page *page = pages[nr_pages_ret - 1];
500 /* zero the tail end of the last page, we might be
501 * sending it down to disk
504 kaddr = kmap_atomic(page);
505 memset(kaddr + offset, 0,
506 PAGE_CACHE_SIZE - offset);
507 kunmap_atomic(kaddr);
514 /* lets try to make an inline extent */
515 if (ret || total_in < (actual_end - start)) {
516 /* we didn't compress the entire range, try
517 * to make an uncompressed inline extent.
519 ret = cow_file_range_inline(root, inode, start, end,
522 /* try making a compressed inline extent */
523 ret = cow_file_range_inline(root, inode, start, end,
525 compress_type, pages);
528 unsigned long clear_flags = EXTENT_DELALLOC |
530 unsigned long page_error_op;
532 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
533 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
536 * inline extent creation worked or returned error,
537 * we don't need to create any more async work items.
538 * Unlock and free up our temp pages.
540 extent_clear_unlock_delalloc(inode, start, end, NULL,
541 clear_flags, PAGE_UNLOCK |
552 * we aren't doing an inline extent round the compressed size
553 * up to a block size boundary so the allocator does sane
556 total_compressed = ALIGN(total_compressed, blocksize);
559 * one last check to make sure the compression is really a
560 * win, compare the page count read with the blocks on disk
562 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
563 if (total_compressed >= total_in) {
566 num_bytes = total_in;
569 if (!will_compress && pages) {
571 * the compression code ran but failed to make things smaller,
572 * free any pages it allocated and our page pointer array
574 for (i = 0; i < nr_pages_ret; i++) {
575 WARN_ON(pages[i]->mapping);
576 page_cache_release(pages[i]);
580 total_compressed = 0;
583 /* flag the file so we don't compress in the future */
584 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
585 !(BTRFS_I(inode)->force_compress)) {
586 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
592 /* the async work queues will take care of doing actual
593 * allocation on disk for these compressed pages,
594 * and will submit them to the elevator.
596 add_async_extent(async_cow, start, num_bytes,
597 total_compressed, pages, nr_pages_ret,
600 if (start + num_bytes < end) {
607 cleanup_and_bail_uncompressed:
609 * No compression, but we still need to write the pages in
610 * the file we've been given so far. redirty the locked
611 * page if it corresponds to our extent and set things up
612 * for the async work queue to run cow_file_range to do
613 * the normal delalloc dance
615 if (page_offset(locked_page) >= start &&
616 page_offset(locked_page) <= end) {
617 __set_page_dirty_nobuffers(locked_page);
618 /* unlocked later on in the async handlers */
621 extent_range_redirty_for_io(inode, start, end);
622 add_async_extent(async_cow, start, end - start + 1,
623 0, NULL, 0, BTRFS_COMPRESS_NONE);
630 for (i = 0; i < nr_pages_ret; i++) {
631 WARN_ON(pages[i]->mapping);
632 page_cache_release(pages[i]);
637 static void free_async_extent_pages(struct async_extent *async_extent)
641 if (!async_extent->pages)
644 for (i = 0; i < async_extent->nr_pages; i++) {
645 WARN_ON(async_extent->pages[i]->mapping);
646 page_cache_release(async_extent->pages[i]);
648 kfree(async_extent->pages);
649 async_extent->nr_pages = 0;
650 async_extent->pages = NULL;
654 * phase two of compressed writeback. This is the ordered portion
655 * of the code, which only gets called in the order the work was
656 * queued. We walk all the async extents created by compress_file_range
657 * and send them down to the disk.
659 static noinline void submit_compressed_extents(struct inode *inode,
660 struct async_cow *async_cow)
662 struct async_extent *async_extent;
664 struct btrfs_key ins;
665 struct extent_map *em;
666 struct btrfs_root *root = BTRFS_I(inode)->root;
667 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
668 struct extent_io_tree *io_tree;
672 while (!list_empty(&async_cow->extents)) {
673 async_extent = list_entry(async_cow->extents.next,
674 struct async_extent, list);
675 list_del(&async_extent->list);
677 io_tree = &BTRFS_I(inode)->io_tree;
680 /* did the compression code fall back to uncompressed IO? */
681 if (!async_extent->pages) {
682 int page_started = 0;
683 unsigned long nr_written = 0;
685 lock_extent(io_tree, async_extent->start,
686 async_extent->start +
687 async_extent->ram_size - 1);
689 /* allocate blocks */
690 ret = cow_file_range(inode, async_cow->locked_page,
692 async_extent->start +
693 async_extent->ram_size - 1,
694 &page_started, &nr_written, 0);
699 * if page_started, cow_file_range inserted an
700 * inline extent and took care of all the unlocking
701 * and IO for us. Otherwise, we need to submit
702 * all those pages down to the drive.
704 if (!page_started && !ret)
705 extent_write_locked_range(io_tree,
706 inode, async_extent->start,
707 async_extent->start +
708 async_extent->ram_size - 1,
712 unlock_page(async_cow->locked_page);
718 lock_extent(io_tree, async_extent->start,
719 async_extent->start + async_extent->ram_size - 1);
721 ret = btrfs_reserve_extent(root,
722 async_extent->compressed_size,
723 async_extent->compressed_size,
724 0, alloc_hint, &ins, 1, 1);
726 free_async_extent_pages(async_extent);
728 if (ret == -ENOSPC) {
729 unlock_extent(io_tree, async_extent->start,
730 async_extent->start +
731 async_extent->ram_size - 1);
734 * we need to redirty the pages if we decide to
735 * fallback to uncompressed IO, otherwise we
736 * will not submit these pages down to lower
739 extent_range_redirty_for_io(inode,
741 async_extent->start +
742 async_extent->ram_size - 1);
750 * here we're doing allocation and writeback of the
753 btrfs_drop_extent_cache(inode, async_extent->start,
754 async_extent->start +
755 async_extent->ram_size - 1, 0);
757 em = alloc_extent_map();
760 goto out_free_reserve;
762 em->start = async_extent->start;
763 em->len = async_extent->ram_size;
764 em->orig_start = em->start;
765 em->mod_start = em->start;
766 em->mod_len = em->len;
768 em->block_start = ins.objectid;
769 em->block_len = ins.offset;
770 em->orig_block_len = ins.offset;
771 em->ram_bytes = async_extent->ram_size;
772 em->bdev = root->fs_info->fs_devices->latest_bdev;
773 em->compress_type = async_extent->compress_type;
774 set_bit(EXTENT_FLAG_PINNED, &em->flags);
775 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
779 write_lock(&em_tree->lock);
780 ret = add_extent_mapping(em_tree, em, 1);
781 write_unlock(&em_tree->lock);
782 if (ret != -EEXIST) {
786 btrfs_drop_extent_cache(inode, async_extent->start,
787 async_extent->start +
788 async_extent->ram_size - 1, 0);
792 goto out_free_reserve;
794 ret = btrfs_add_ordered_extent_compress(inode,
797 async_extent->ram_size,
799 BTRFS_ORDERED_COMPRESSED,
800 async_extent->compress_type);
802 btrfs_drop_extent_cache(inode, async_extent->start,
803 async_extent->start +
804 async_extent->ram_size - 1, 0);
805 goto out_free_reserve;
809 * clear dirty, set writeback and unlock the pages.
811 extent_clear_unlock_delalloc(inode, async_extent->start,
812 async_extent->start +
813 async_extent->ram_size - 1,
814 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
815 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
817 ret = btrfs_submit_compressed_write(inode,
819 async_extent->ram_size,
821 ins.offset, async_extent->pages,
822 async_extent->nr_pages);
824 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
825 struct page *p = async_extent->pages[0];
826 const u64 start = async_extent->start;
827 const u64 end = start + async_extent->ram_size - 1;
829 p->mapping = inode->i_mapping;
830 tree->ops->writepage_end_io_hook(p, start, end,
833 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
836 free_async_extent_pages(async_extent);
838 alloc_hint = ins.objectid + ins.offset;
844 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
846 extent_clear_unlock_delalloc(inode, async_extent->start,
847 async_extent->start +
848 async_extent->ram_size - 1,
849 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
850 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
851 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
852 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
854 free_async_extent_pages(async_extent);
859 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
862 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
863 struct extent_map *em;
866 read_lock(&em_tree->lock);
867 em = search_extent_mapping(em_tree, start, num_bytes);
870 * if block start isn't an actual block number then find the
871 * first block in this inode and use that as a hint. If that
872 * block is also bogus then just don't worry about it.
874 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
876 em = search_extent_mapping(em_tree, 0, 0);
877 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
878 alloc_hint = em->block_start;
882 alloc_hint = em->block_start;
886 read_unlock(&em_tree->lock);
892 * when extent_io.c finds a delayed allocation range in the file,
893 * the call backs end up in this code. The basic idea is to
894 * allocate extents on disk for the range, and create ordered data structs
895 * in ram to track those extents.
897 * locked_page is the page that writepage had locked already. We use
898 * it to make sure we don't do extra locks or unlocks.
900 * *page_started is set to one if we unlock locked_page and do everything
901 * required to start IO on it. It may be clean and already done with
904 static noinline int cow_file_range(struct inode *inode,
905 struct page *locked_page,
906 u64 start, u64 end, int *page_started,
907 unsigned long *nr_written,
910 struct btrfs_root *root = BTRFS_I(inode)->root;
913 unsigned long ram_size;
916 u64 blocksize = root->sectorsize;
917 struct btrfs_key ins;
918 struct extent_map *em;
919 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
922 if (btrfs_is_free_space_inode(inode)) {
928 num_bytes = ALIGN(end - start + 1, blocksize);
929 num_bytes = max(blocksize, num_bytes);
930 disk_num_bytes = num_bytes;
932 /* if this is a small write inside eof, kick off defrag */
933 if (num_bytes < 64 * 1024 &&
934 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
935 btrfs_add_inode_defrag(NULL, inode);
938 /* lets try to make an inline extent */
939 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
942 extent_clear_unlock_delalloc(inode, start, end, NULL,
943 EXTENT_LOCKED | EXTENT_DELALLOC |
944 EXTENT_DEFRAG, PAGE_UNLOCK |
945 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
948 *nr_written = *nr_written +
949 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
952 } else if (ret < 0) {
957 BUG_ON(disk_num_bytes >
958 btrfs_super_total_bytes(root->fs_info->super_copy));
960 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
961 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
963 while (disk_num_bytes > 0) {
966 cur_alloc_size = disk_num_bytes;
967 ret = btrfs_reserve_extent(root, cur_alloc_size,
968 root->sectorsize, 0, alloc_hint,
973 em = alloc_extent_map();
979 em->orig_start = em->start;
980 ram_size = ins.offset;
981 em->len = ins.offset;
982 em->mod_start = em->start;
983 em->mod_len = em->len;
985 em->block_start = ins.objectid;
986 em->block_len = ins.offset;
987 em->orig_block_len = ins.offset;
988 em->ram_bytes = ram_size;
989 em->bdev = root->fs_info->fs_devices->latest_bdev;
990 set_bit(EXTENT_FLAG_PINNED, &em->flags);
994 write_lock(&em_tree->lock);
995 ret = add_extent_mapping(em_tree, em, 1);
996 write_unlock(&em_tree->lock);
997 if (ret != -EEXIST) {
1001 btrfs_drop_extent_cache(inode, start,
1002 start + ram_size - 1, 0);
1007 cur_alloc_size = ins.offset;
1008 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1009 ram_size, cur_alloc_size, 0);
1011 goto out_drop_extent_cache;
1013 if (root->root_key.objectid ==
1014 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1015 ret = btrfs_reloc_clone_csums(inode, start,
1018 goto out_drop_extent_cache;
1021 if (disk_num_bytes < cur_alloc_size)
1024 /* we're not doing compressed IO, don't unlock the first
1025 * page (which the caller expects to stay locked), don't
1026 * clear any dirty bits and don't set any writeback bits
1028 * Do set the Private2 bit so we know this page was properly
1029 * setup for writepage
1031 op = unlock ? PAGE_UNLOCK : 0;
1032 op |= PAGE_SET_PRIVATE2;
1034 extent_clear_unlock_delalloc(inode, start,
1035 start + ram_size - 1, locked_page,
1036 EXTENT_LOCKED | EXTENT_DELALLOC,
1038 disk_num_bytes -= cur_alloc_size;
1039 num_bytes -= cur_alloc_size;
1040 alloc_hint = ins.objectid + ins.offset;
1041 start += cur_alloc_size;
1046 out_drop_extent_cache:
1047 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1049 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1051 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1052 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1053 EXTENT_DELALLOC | EXTENT_DEFRAG,
1054 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1055 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1060 * work queue call back to started compression on a file and pages
1062 static noinline void async_cow_start(struct btrfs_work *work)
1064 struct async_cow *async_cow;
1066 async_cow = container_of(work, struct async_cow, work);
1068 compress_file_range(async_cow->inode, async_cow->locked_page,
1069 async_cow->start, async_cow->end, async_cow,
1071 if (num_added == 0) {
1072 btrfs_add_delayed_iput(async_cow->inode);
1073 async_cow->inode = NULL;
1078 * work queue call back to submit previously compressed pages
1080 static noinline void async_cow_submit(struct btrfs_work *work)
1082 struct async_cow *async_cow;
1083 struct btrfs_root *root;
1084 unsigned long nr_pages;
1086 async_cow = container_of(work, struct async_cow, work);
1088 root = async_cow->root;
1089 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1092 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1094 waitqueue_active(&root->fs_info->async_submit_wait))
1095 wake_up(&root->fs_info->async_submit_wait);
1097 if (async_cow->inode)
1098 submit_compressed_extents(async_cow->inode, async_cow);
1101 static noinline void async_cow_free(struct btrfs_work *work)
1103 struct async_cow *async_cow;
1104 async_cow = container_of(work, struct async_cow, work);
1105 if (async_cow->inode)
1106 btrfs_add_delayed_iput(async_cow->inode);
1110 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1111 u64 start, u64 end, int *page_started,
1112 unsigned long *nr_written)
1114 struct async_cow *async_cow;
1115 struct btrfs_root *root = BTRFS_I(inode)->root;
1116 unsigned long nr_pages;
1118 int limit = 10 * 1024 * 1024;
1120 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1121 1, 0, NULL, GFP_NOFS);
1122 while (start < end) {
1123 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1124 BUG_ON(!async_cow); /* -ENOMEM */
1125 async_cow->inode = igrab(inode);
1126 async_cow->root = root;
1127 async_cow->locked_page = locked_page;
1128 async_cow->start = start;
1130 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1131 !btrfs_test_opt(root, FORCE_COMPRESS))
1134 cur_end = min(end, start + 512 * 1024 - 1);
1136 async_cow->end = cur_end;
1137 INIT_LIST_HEAD(&async_cow->extents);
1139 btrfs_init_work(&async_cow->work,
1140 btrfs_delalloc_helper,
1141 async_cow_start, async_cow_submit,
1144 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1146 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1148 btrfs_queue_work(root->fs_info->delalloc_workers,
1151 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1152 wait_event(root->fs_info->async_submit_wait,
1153 (atomic_read(&root->fs_info->async_delalloc_pages) <
1157 while (atomic_read(&root->fs_info->async_submit_draining) &&
1158 atomic_read(&root->fs_info->async_delalloc_pages)) {
1159 wait_event(root->fs_info->async_submit_wait,
1160 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1164 *nr_written += nr_pages;
1165 start = cur_end + 1;
1171 static noinline int csum_exist_in_range(struct btrfs_root *root,
1172 u64 bytenr, u64 num_bytes)
1175 struct btrfs_ordered_sum *sums;
1178 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1179 bytenr + num_bytes - 1, &list, 0);
1180 if (ret == 0 && list_empty(&list))
1183 while (!list_empty(&list)) {
1184 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1185 list_del(&sums->list);
1192 * when nowcow writeback call back. This checks for snapshots or COW copies
1193 * of the extents that exist in the file, and COWs the file as required.
1195 * If no cow copies or snapshots exist, we write directly to the existing
1198 static noinline int run_delalloc_nocow(struct inode *inode,
1199 struct page *locked_page,
1200 u64 start, u64 end, int *page_started, int force,
1201 unsigned long *nr_written)
1203 struct btrfs_root *root = BTRFS_I(inode)->root;
1204 struct btrfs_trans_handle *trans;
1205 struct extent_buffer *leaf;
1206 struct btrfs_path *path;
1207 struct btrfs_file_extent_item *fi;
1208 struct btrfs_key found_key;
1223 u64 ino = btrfs_ino(inode);
1225 path = btrfs_alloc_path();
1227 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1228 EXTENT_LOCKED | EXTENT_DELALLOC |
1229 EXTENT_DO_ACCOUNTING |
1230 EXTENT_DEFRAG, PAGE_UNLOCK |
1232 PAGE_SET_WRITEBACK |
1233 PAGE_END_WRITEBACK);
1237 nolock = btrfs_is_free_space_inode(inode);
1240 trans = btrfs_join_transaction_nolock(root);
1242 trans = btrfs_join_transaction(root);
1244 if (IS_ERR(trans)) {
1245 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1246 EXTENT_LOCKED | EXTENT_DELALLOC |
1247 EXTENT_DO_ACCOUNTING |
1248 EXTENT_DEFRAG, PAGE_UNLOCK |
1250 PAGE_SET_WRITEBACK |
1251 PAGE_END_WRITEBACK);
1252 btrfs_free_path(path);
1253 return PTR_ERR(trans);
1256 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1258 cow_start = (u64)-1;
1261 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1265 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1266 leaf = path->nodes[0];
1267 btrfs_item_key_to_cpu(leaf, &found_key,
1268 path->slots[0] - 1);
1269 if (found_key.objectid == ino &&
1270 found_key.type == BTRFS_EXTENT_DATA_KEY)
1275 leaf = path->nodes[0];
1276 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1277 ret = btrfs_next_leaf(root, path);
1282 leaf = path->nodes[0];
1288 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1290 if (found_key.objectid > ino ||
1291 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1292 found_key.offset > end)
1295 if (found_key.offset > cur_offset) {
1296 extent_end = found_key.offset;
1301 fi = btrfs_item_ptr(leaf, path->slots[0],
1302 struct btrfs_file_extent_item);
1303 extent_type = btrfs_file_extent_type(leaf, fi);
1305 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1306 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1307 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1308 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1309 extent_offset = btrfs_file_extent_offset(leaf, fi);
1310 extent_end = found_key.offset +
1311 btrfs_file_extent_num_bytes(leaf, fi);
1313 btrfs_file_extent_disk_num_bytes(leaf, fi);
1314 if (extent_end <= start) {
1318 if (disk_bytenr == 0)
1320 if (btrfs_file_extent_compression(leaf, fi) ||
1321 btrfs_file_extent_encryption(leaf, fi) ||
1322 btrfs_file_extent_other_encoding(leaf, fi))
1324 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1326 if (btrfs_extent_readonly(root, disk_bytenr))
1328 if (btrfs_cross_ref_exist(trans, root, ino,
1330 extent_offset, disk_bytenr))
1332 disk_bytenr += extent_offset;
1333 disk_bytenr += cur_offset - found_key.offset;
1334 num_bytes = min(end + 1, extent_end) - cur_offset;
1336 * if there are pending snapshots for this root,
1337 * we fall into common COW way.
1340 err = btrfs_start_write_no_snapshoting(root);
1345 * force cow if csum exists in the range.
1346 * this ensure that csum for a given extent are
1347 * either valid or do not exist.
1349 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1352 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1353 extent_end = found_key.offset +
1354 btrfs_file_extent_inline_len(leaf,
1355 path->slots[0], fi);
1356 extent_end = ALIGN(extent_end, root->sectorsize);
1361 if (extent_end <= start) {
1363 if (!nolock && nocow)
1364 btrfs_end_write_no_snapshoting(root);
1368 if (cow_start == (u64)-1)
1369 cow_start = cur_offset;
1370 cur_offset = extent_end;
1371 if (cur_offset > end)
1377 btrfs_release_path(path);
1378 if (cow_start != (u64)-1) {
1379 ret = cow_file_range(inode, locked_page,
1380 cow_start, found_key.offset - 1,
1381 page_started, nr_written, 1);
1383 if (!nolock && nocow)
1384 btrfs_end_write_no_snapshoting(root);
1387 cow_start = (u64)-1;
1390 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1391 struct extent_map *em;
1392 struct extent_map_tree *em_tree;
1393 em_tree = &BTRFS_I(inode)->extent_tree;
1394 em = alloc_extent_map();
1395 BUG_ON(!em); /* -ENOMEM */
1396 em->start = cur_offset;
1397 em->orig_start = found_key.offset - extent_offset;
1398 em->len = num_bytes;
1399 em->block_len = num_bytes;
1400 em->block_start = disk_bytenr;
1401 em->orig_block_len = disk_num_bytes;
1402 em->ram_bytes = ram_bytes;
1403 em->bdev = root->fs_info->fs_devices->latest_bdev;
1404 em->mod_start = em->start;
1405 em->mod_len = em->len;
1406 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1407 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1408 em->generation = -1;
1410 write_lock(&em_tree->lock);
1411 ret = add_extent_mapping(em_tree, em, 1);
1412 write_unlock(&em_tree->lock);
1413 if (ret != -EEXIST) {
1414 free_extent_map(em);
1417 btrfs_drop_extent_cache(inode, em->start,
1418 em->start + em->len - 1, 0);
1420 type = BTRFS_ORDERED_PREALLOC;
1422 type = BTRFS_ORDERED_NOCOW;
1425 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1426 num_bytes, num_bytes, type);
1427 BUG_ON(ret); /* -ENOMEM */
1429 if (root->root_key.objectid ==
1430 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1431 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1434 if (!nolock && nocow)
1435 btrfs_end_write_no_snapshoting(root);
1440 extent_clear_unlock_delalloc(inode, cur_offset,
1441 cur_offset + num_bytes - 1,
1442 locked_page, EXTENT_LOCKED |
1443 EXTENT_DELALLOC, PAGE_UNLOCK |
1445 if (!nolock && nocow)
1446 btrfs_end_write_no_snapshoting(root);
1447 cur_offset = extent_end;
1448 if (cur_offset > end)
1451 btrfs_release_path(path);
1453 if (cur_offset <= end && cow_start == (u64)-1) {
1454 cow_start = cur_offset;
1458 if (cow_start != (u64)-1) {
1459 ret = cow_file_range(inode, locked_page, cow_start, end,
1460 page_started, nr_written, 1);
1466 err = btrfs_end_transaction(trans, root);
1470 if (ret && cur_offset < end)
1471 extent_clear_unlock_delalloc(inode, cur_offset, end,
1472 locked_page, EXTENT_LOCKED |
1473 EXTENT_DELALLOC | EXTENT_DEFRAG |
1474 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1476 PAGE_SET_WRITEBACK |
1477 PAGE_END_WRITEBACK);
1478 btrfs_free_path(path);
1482 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1485 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1486 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1490 * @defrag_bytes is a hint value, no spinlock held here,
1491 * if is not zero, it means the file is defragging.
1492 * Force cow if given extent needs to be defragged.
1494 if (BTRFS_I(inode)->defrag_bytes &&
1495 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1496 EXTENT_DEFRAG, 0, NULL))
1503 * extent_io.c call back to do delayed allocation processing
1505 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1506 u64 start, u64 end, int *page_started,
1507 unsigned long *nr_written)
1510 int force_cow = need_force_cow(inode, start, end);
1512 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1513 ret = run_delalloc_nocow(inode, locked_page, start, end,
1514 page_started, 1, nr_written);
1515 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1516 ret = run_delalloc_nocow(inode, locked_page, start, end,
1517 page_started, 0, nr_written);
1518 } else if (!inode_need_compress(inode)) {
1519 ret = cow_file_range(inode, locked_page, start, end,
1520 page_started, nr_written, 1);
1522 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1523 &BTRFS_I(inode)->runtime_flags);
1524 ret = cow_file_range_async(inode, locked_page, start, end,
1525 page_started, nr_written);
1530 static void btrfs_split_extent_hook(struct inode *inode,
1531 struct extent_state *orig, u64 split)
1533 /* not delalloc, ignore it */
1534 if (!(orig->state & EXTENT_DELALLOC))
1537 spin_lock(&BTRFS_I(inode)->lock);
1538 BTRFS_I(inode)->outstanding_extents++;
1539 spin_unlock(&BTRFS_I(inode)->lock);
1543 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1544 * extents so we can keep track of new extents that are just merged onto old
1545 * extents, such as when we are doing sequential writes, so we can properly
1546 * account for the metadata space we'll need.
1548 static void btrfs_merge_extent_hook(struct inode *inode,
1549 struct extent_state *new,
1550 struct extent_state *other)
1552 /* not delalloc, ignore it */
1553 if (!(other->state & EXTENT_DELALLOC))
1556 spin_lock(&BTRFS_I(inode)->lock);
1557 BTRFS_I(inode)->outstanding_extents--;
1558 spin_unlock(&BTRFS_I(inode)->lock);
1561 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1562 struct inode *inode)
1564 spin_lock(&root->delalloc_lock);
1565 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1566 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1567 &root->delalloc_inodes);
1568 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1569 &BTRFS_I(inode)->runtime_flags);
1570 root->nr_delalloc_inodes++;
1571 if (root->nr_delalloc_inodes == 1) {
1572 spin_lock(&root->fs_info->delalloc_root_lock);
1573 BUG_ON(!list_empty(&root->delalloc_root));
1574 list_add_tail(&root->delalloc_root,
1575 &root->fs_info->delalloc_roots);
1576 spin_unlock(&root->fs_info->delalloc_root_lock);
1579 spin_unlock(&root->delalloc_lock);
1582 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1583 struct inode *inode)
1585 spin_lock(&root->delalloc_lock);
1586 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1587 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1588 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1589 &BTRFS_I(inode)->runtime_flags);
1590 root->nr_delalloc_inodes--;
1591 if (!root->nr_delalloc_inodes) {
1592 spin_lock(&root->fs_info->delalloc_root_lock);
1593 BUG_ON(list_empty(&root->delalloc_root));
1594 list_del_init(&root->delalloc_root);
1595 spin_unlock(&root->fs_info->delalloc_root_lock);
1598 spin_unlock(&root->delalloc_lock);
1602 * extent_io.c set_bit_hook, used to track delayed allocation
1603 * bytes in this file, and to maintain the list of inodes that
1604 * have pending delalloc work to be done.
1606 static void btrfs_set_bit_hook(struct inode *inode,
1607 struct extent_state *state, unsigned *bits)
1610 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1613 * set_bit and clear bit hooks normally require _irqsave/restore
1614 * but in this case, we are only testing for the DELALLOC
1615 * bit, which is only set or cleared with irqs on
1617 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1618 struct btrfs_root *root = BTRFS_I(inode)->root;
1619 u64 len = state->end + 1 - state->start;
1620 bool do_list = !btrfs_is_free_space_inode(inode);
1622 if (*bits & EXTENT_FIRST_DELALLOC) {
1623 *bits &= ~EXTENT_FIRST_DELALLOC;
1625 spin_lock(&BTRFS_I(inode)->lock);
1626 BTRFS_I(inode)->outstanding_extents++;
1627 spin_unlock(&BTRFS_I(inode)->lock);
1630 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1631 root->fs_info->delalloc_batch);
1632 spin_lock(&BTRFS_I(inode)->lock);
1633 BTRFS_I(inode)->delalloc_bytes += len;
1634 if (*bits & EXTENT_DEFRAG)
1635 BTRFS_I(inode)->defrag_bytes += len;
1636 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1637 &BTRFS_I(inode)->runtime_flags))
1638 btrfs_add_delalloc_inodes(root, inode);
1639 spin_unlock(&BTRFS_I(inode)->lock);
1644 * extent_io.c clear_bit_hook, see set_bit_hook for why
1646 static void btrfs_clear_bit_hook(struct inode *inode,
1647 struct extent_state *state,
1650 u64 len = state->end + 1 - state->start;
1652 spin_lock(&BTRFS_I(inode)->lock);
1653 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1654 BTRFS_I(inode)->defrag_bytes -= len;
1655 spin_unlock(&BTRFS_I(inode)->lock);
1658 * set_bit and clear bit hooks normally require _irqsave/restore
1659 * but in this case, we are only testing for the DELALLOC
1660 * bit, which is only set or cleared with irqs on
1662 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1663 struct btrfs_root *root = BTRFS_I(inode)->root;
1664 bool do_list = !btrfs_is_free_space_inode(inode);
1666 if (*bits & EXTENT_FIRST_DELALLOC) {
1667 *bits &= ~EXTENT_FIRST_DELALLOC;
1668 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1669 spin_lock(&BTRFS_I(inode)->lock);
1670 BTRFS_I(inode)->outstanding_extents--;
1671 spin_unlock(&BTRFS_I(inode)->lock);
1675 * We don't reserve metadata space for space cache inodes so we
1676 * don't need to call dellalloc_release_metadata if there is an
1679 if (*bits & EXTENT_DO_ACCOUNTING &&
1680 root != root->fs_info->tree_root)
1681 btrfs_delalloc_release_metadata(inode, len);
1683 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1684 && do_list && !(state->state & EXTENT_NORESERVE))
1685 btrfs_free_reserved_data_space(inode, len);
1687 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1688 root->fs_info->delalloc_batch);
1689 spin_lock(&BTRFS_I(inode)->lock);
1690 BTRFS_I(inode)->delalloc_bytes -= len;
1691 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1692 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1693 &BTRFS_I(inode)->runtime_flags))
1694 btrfs_del_delalloc_inode(root, inode);
1695 spin_unlock(&BTRFS_I(inode)->lock);
1700 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1701 * we don't create bios that span stripes or chunks
1703 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1704 size_t size, struct bio *bio,
1705 unsigned long bio_flags)
1707 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1708 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1713 if (bio_flags & EXTENT_BIO_COMPRESSED)
1716 length = bio->bi_iter.bi_size;
1717 map_length = length;
1718 ret = btrfs_map_block(root->fs_info, rw, logical,
1719 &map_length, NULL, 0);
1720 /* Will always return 0 with map_multi == NULL */
1722 if (map_length < length + size)
1728 * in order to insert checksums into the metadata in large chunks,
1729 * we wait until bio submission time. All the pages in the bio are
1730 * checksummed and sums are attached onto the ordered extent record.
1732 * At IO completion time the cums attached on the ordered extent record
1733 * are inserted into the btree
1735 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1736 struct bio *bio, int mirror_num,
1737 unsigned long bio_flags,
1740 struct btrfs_root *root = BTRFS_I(inode)->root;
1743 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1744 BUG_ON(ret); /* -ENOMEM */
1749 * in order to insert checksums into the metadata in large chunks,
1750 * we wait until bio submission time. All the pages in the bio are
1751 * checksummed and sums are attached onto the ordered extent record.
1753 * At IO completion time the cums attached on the ordered extent record
1754 * are inserted into the btree
1756 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1757 int mirror_num, unsigned long bio_flags,
1760 struct btrfs_root *root = BTRFS_I(inode)->root;
1763 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1765 bio_endio(bio, ret);
1770 * extent_io.c submission hook. This does the right thing for csum calculation
1771 * on write, or reading the csums from the tree before a read
1773 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1774 int mirror_num, unsigned long bio_flags,
1777 struct btrfs_root *root = BTRFS_I(inode)->root;
1781 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1783 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1785 if (btrfs_is_free_space_inode(inode))
1788 if (!(rw & REQ_WRITE)) {
1789 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1793 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1794 ret = btrfs_submit_compressed_read(inode, bio,
1798 } else if (!skip_sum) {
1799 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1804 } else if (async && !skip_sum) {
1805 /* csum items have already been cloned */
1806 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1808 /* we're doing a write, do the async checksumming */
1809 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1810 inode, rw, bio, mirror_num,
1811 bio_flags, bio_offset,
1812 __btrfs_submit_bio_start,
1813 __btrfs_submit_bio_done);
1815 } else if (!skip_sum) {
1816 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1822 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1826 bio_endio(bio, ret);
1831 * given a list of ordered sums record them in the inode. This happens
1832 * at IO completion time based on sums calculated at bio submission time.
1834 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1835 struct inode *inode, u64 file_offset,
1836 struct list_head *list)
1838 struct btrfs_ordered_sum *sum;
1840 list_for_each_entry(sum, list, list) {
1841 trans->adding_csums = 1;
1842 btrfs_csum_file_blocks(trans,
1843 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1844 trans->adding_csums = 0;
1849 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1850 struct extent_state **cached_state)
1852 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1853 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1854 cached_state, GFP_NOFS);
1857 /* see btrfs_writepage_start_hook for details on why this is required */
1858 struct btrfs_writepage_fixup {
1860 struct btrfs_work work;
1863 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1865 struct btrfs_writepage_fixup *fixup;
1866 struct btrfs_ordered_extent *ordered;
1867 struct extent_state *cached_state = NULL;
1869 struct inode *inode;
1874 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1878 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1879 ClearPageChecked(page);
1883 inode = page->mapping->host;
1884 page_start = page_offset(page);
1885 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1887 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1890 /* already ordered? We're done */
1891 if (PagePrivate2(page))
1894 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1896 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1897 page_end, &cached_state, GFP_NOFS);
1899 btrfs_start_ordered_extent(inode, ordered, 1);
1900 btrfs_put_ordered_extent(ordered);
1904 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1906 mapping_set_error(page->mapping, ret);
1907 end_extent_writepage(page, ret, page_start, page_end);
1908 ClearPageChecked(page);
1912 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1913 ClearPageChecked(page);
1914 set_page_dirty(page);
1916 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1917 &cached_state, GFP_NOFS);
1920 page_cache_release(page);
1925 * There are a few paths in the higher layers of the kernel that directly
1926 * set the page dirty bit without asking the filesystem if it is a
1927 * good idea. This causes problems because we want to make sure COW
1928 * properly happens and the data=ordered rules are followed.
1930 * In our case any range that doesn't have the ORDERED bit set
1931 * hasn't been properly setup for IO. We kick off an async process
1932 * to fix it up. The async helper will wait for ordered extents, set
1933 * the delalloc bit and make it safe to write the page.
1935 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1937 struct inode *inode = page->mapping->host;
1938 struct btrfs_writepage_fixup *fixup;
1939 struct btrfs_root *root = BTRFS_I(inode)->root;
1941 /* this page is properly in the ordered list */
1942 if (TestClearPagePrivate2(page))
1945 if (PageChecked(page))
1948 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1952 SetPageChecked(page);
1953 page_cache_get(page);
1954 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
1955 btrfs_writepage_fixup_worker, NULL, NULL);
1957 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
1961 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1962 struct inode *inode, u64 file_pos,
1963 u64 disk_bytenr, u64 disk_num_bytes,
1964 u64 num_bytes, u64 ram_bytes,
1965 u8 compression, u8 encryption,
1966 u16 other_encoding, int extent_type)
1968 struct btrfs_root *root = BTRFS_I(inode)->root;
1969 struct btrfs_file_extent_item *fi;
1970 struct btrfs_path *path;
1971 struct extent_buffer *leaf;
1972 struct btrfs_key ins;
1973 int extent_inserted = 0;
1976 path = btrfs_alloc_path();
1981 * we may be replacing one extent in the tree with another.
1982 * The new extent is pinned in the extent map, and we don't want
1983 * to drop it from the cache until it is completely in the btree.
1985 * So, tell btrfs_drop_extents to leave this extent in the cache.
1986 * the caller is expected to unpin it and allow it to be merged
1989 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
1990 file_pos + num_bytes, NULL, 0,
1991 1, sizeof(*fi), &extent_inserted);
1995 if (!extent_inserted) {
1996 ins.objectid = btrfs_ino(inode);
1997 ins.offset = file_pos;
1998 ins.type = BTRFS_EXTENT_DATA_KEY;
2000 path->leave_spinning = 1;
2001 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2006 leaf = path->nodes[0];
2007 fi = btrfs_item_ptr(leaf, path->slots[0],
2008 struct btrfs_file_extent_item);
2009 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2010 btrfs_set_file_extent_type(leaf, fi, extent_type);
2011 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2012 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2013 btrfs_set_file_extent_offset(leaf, fi, 0);
2014 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2015 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2016 btrfs_set_file_extent_compression(leaf, fi, compression);
2017 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2018 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2020 btrfs_mark_buffer_dirty(leaf);
2021 btrfs_release_path(path);
2023 inode_add_bytes(inode, num_bytes);
2025 ins.objectid = disk_bytenr;
2026 ins.offset = disk_num_bytes;
2027 ins.type = BTRFS_EXTENT_ITEM_KEY;
2028 ret = btrfs_alloc_reserved_file_extent(trans, root,
2029 root->root_key.objectid,
2030 btrfs_ino(inode), file_pos, &ins);
2032 btrfs_free_path(path);
2037 /* snapshot-aware defrag */
2038 struct sa_defrag_extent_backref {
2039 struct rb_node node;
2040 struct old_sa_defrag_extent *old;
2049 struct old_sa_defrag_extent {
2050 struct list_head list;
2051 struct new_sa_defrag_extent *new;
2060 struct new_sa_defrag_extent {
2061 struct rb_root root;
2062 struct list_head head;
2063 struct btrfs_path *path;
2064 struct inode *inode;
2072 static int backref_comp(struct sa_defrag_extent_backref *b1,
2073 struct sa_defrag_extent_backref *b2)
2075 if (b1->root_id < b2->root_id)
2077 else if (b1->root_id > b2->root_id)
2080 if (b1->inum < b2->inum)
2082 else if (b1->inum > b2->inum)
2085 if (b1->file_pos < b2->file_pos)
2087 else if (b1->file_pos > b2->file_pos)
2091 * [------------------------------] ===> (a range of space)
2092 * |<--->| |<---->| =============> (fs/file tree A)
2093 * |<---------------------------->| ===> (fs/file tree B)
2095 * A range of space can refer to two file extents in one tree while
2096 * refer to only one file extent in another tree.
2098 * So we may process a disk offset more than one time(two extents in A)
2099 * and locate at the same extent(one extent in B), then insert two same
2100 * backrefs(both refer to the extent in B).
2105 static void backref_insert(struct rb_root *root,
2106 struct sa_defrag_extent_backref *backref)
2108 struct rb_node **p = &root->rb_node;
2109 struct rb_node *parent = NULL;
2110 struct sa_defrag_extent_backref *entry;
2115 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2117 ret = backref_comp(backref, entry);
2121 p = &(*p)->rb_right;
2124 rb_link_node(&backref->node, parent, p);
2125 rb_insert_color(&backref->node, root);
2129 * Note the backref might has changed, and in this case we just return 0.
2131 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2134 struct btrfs_file_extent_item *extent;
2135 struct btrfs_fs_info *fs_info;
2136 struct old_sa_defrag_extent *old = ctx;
2137 struct new_sa_defrag_extent *new = old->new;
2138 struct btrfs_path *path = new->path;
2139 struct btrfs_key key;
2140 struct btrfs_root *root;
2141 struct sa_defrag_extent_backref *backref;
2142 struct extent_buffer *leaf;
2143 struct inode *inode = new->inode;
2149 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2150 inum == btrfs_ino(inode))
2153 key.objectid = root_id;
2154 key.type = BTRFS_ROOT_ITEM_KEY;
2155 key.offset = (u64)-1;
2157 fs_info = BTRFS_I(inode)->root->fs_info;
2158 root = btrfs_read_fs_root_no_name(fs_info, &key);
2160 if (PTR_ERR(root) == -ENOENT)
2163 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2164 inum, offset, root_id);
2165 return PTR_ERR(root);
2168 key.objectid = inum;
2169 key.type = BTRFS_EXTENT_DATA_KEY;
2170 if (offset > (u64)-1 << 32)
2173 key.offset = offset;
2175 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2176 if (WARN_ON(ret < 0))
2183 leaf = path->nodes[0];
2184 slot = path->slots[0];
2186 if (slot >= btrfs_header_nritems(leaf)) {
2187 ret = btrfs_next_leaf(root, path);
2190 } else if (ret > 0) {
2199 btrfs_item_key_to_cpu(leaf, &key, slot);
2201 if (key.objectid > inum)
2204 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2207 extent = btrfs_item_ptr(leaf, slot,
2208 struct btrfs_file_extent_item);
2210 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2214 * 'offset' refers to the exact key.offset,
2215 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2216 * (key.offset - extent_offset).
2218 if (key.offset != offset)
2221 extent_offset = btrfs_file_extent_offset(leaf, extent);
2222 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2224 if (extent_offset >= old->extent_offset + old->offset +
2225 old->len || extent_offset + num_bytes <=
2226 old->extent_offset + old->offset)
2231 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2237 backref->root_id = root_id;
2238 backref->inum = inum;
2239 backref->file_pos = offset;
2240 backref->num_bytes = num_bytes;
2241 backref->extent_offset = extent_offset;
2242 backref->generation = btrfs_file_extent_generation(leaf, extent);
2244 backref_insert(&new->root, backref);
2247 btrfs_release_path(path);
2252 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2253 struct new_sa_defrag_extent *new)
2255 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2256 struct old_sa_defrag_extent *old, *tmp;
2261 list_for_each_entry_safe(old, tmp, &new->head, list) {
2262 ret = iterate_inodes_from_logical(old->bytenr +
2263 old->extent_offset, fs_info,
2264 path, record_one_backref,
2266 if (ret < 0 && ret != -ENOENT)
2269 /* no backref to be processed for this extent */
2271 list_del(&old->list);
2276 if (list_empty(&new->head))
2282 static int relink_is_mergable(struct extent_buffer *leaf,
2283 struct btrfs_file_extent_item *fi,
2284 struct new_sa_defrag_extent *new)
2286 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2289 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2292 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2295 if (btrfs_file_extent_encryption(leaf, fi) ||
2296 btrfs_file_extent_other_encoding(leaf, fi))
2303 * Note the backref might has changed, and in this case we just return 0.
2305 static noinline int relink_extent_backref(struct btrfs_path *path,
2306 struct sa_defrag_extent_backref *prev,
2307 struct sa_defrag_extent_backref *backref)
2309 struct btrfs_file_extent_item *extent;
2310 struct btrfs_file_extent_item *item;
2311 struct btrfs_ordered_extent *ordered;
2312 struct btrfs_trans_handle *trans;
2313 struct btrfs_fs_info *fs_info;
2314 struct btrfs_root *root;
2315 struct btrfs_key key;
2316 struct extent_buffer *leaf;
2317 struct old_sa_defrag_extent *old = backref->old;
2318 struct new_sa_defrag_extent *new = old->new;
2319 struct inode *src_inode = new->inode;
2320 struct inode *inode;
2321 struct extent_state *cached = NULL;
2330 if (prev && prev->root_id == backref->root_id &&
2331 prev->inum == backref->inum &&
2332 prev->file_pos + prev->num_bytes == backref->file_pos)
2335 /* step 1: get root */
2336 key.objectid = backref->root_id;
2337 key.type = BTRFS_ROOT_ITEM_KEY;
2338 key.offset = (u64)-1;
2340 fs_info = BTRFS_I(src_inode)->root->fs_info;
2341 index = srcu_read_lock(&fs_info->subvol_srcu);
2343 root = btrfs_read_fs_root_no_name(fs_info, &key);
2345 srcu_read_unlock(&fs_info->subvol_srcu, index);
2346 if (PTR_ERR(root) == -ENOENT)
2348 return PTR_ERR(root);
2351 if (btrfs_root_readonly(root)) {
2352 srcu_read_unlock(&fs_info->subvol_srcu, index);
2356 /* step 2: get inode */
2357 key.objectid = backref->inum;
2358 key.type = BTRFS_INODE_ITEM_KEY;
2361 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2362 if (IS_ERR(inode)) {
2363 srcu_read_unlock(&fs_info->subvol_srcu, index);
2367 srcu_read_unlock(&fs_info->subvol_srcu, index);
2369 /* step 3: relink backref */
2370 lock_start = backref->file_pos;
2371 lock_end = backref->file_pos + backref->num_bytes - 1;
2372 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2375 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2377 btrfs_put_ordered_extent(ordered);
2381 trans = btrfs_join_transaction(root);
2382 if (IS_ERR(trans)) {
2383 ret = PTR_ERR(trans);
2387 key.objectid = backref->inum;
2388 key.type = BTRFS_EXTENT_DATA_KEY;
2389 key.offset = backref->file_pos;
2391 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2394 } else if (ret > 0) {
2399 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2400 struct btrfs_file_extent_item);
2402 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2403 backref->generation)
2406 btrfs_release_path(path);
2408 start = backref->file_pos;
2409 if (backref->extent_offset < old->extent_offset + old->offset)
2410 start += old->extent_offset + old->offset -
2411 backref->extent_offset;
2413 len = min(backref->extent_offset + backref->num_bytes,
2414 old->extent_offset + old->offset + old->len);
2415 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2417 ret = btrfs_drop_extents(trans, root, inode, start,
2422 key.objectid = btrfs_ino(inode);
2423 key.type = BTRFS_EXTENT_DATA_KEY;
2426 path->leave_spinning = 1;
2428 struct btrfs_file_extent_item *fi;
2430 struct btrfs_key found_key;
2432 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2437 leaf = path->nodes[0];
2438 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2440 fi = btrfs_item_ptr(leaf, path->slots[0],
2441 struct btrfs_file_extent_item);
2442 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2444 if (extent_len + found_key.offset == start &&
2445 relink_is_mergable(leaf, fi, new)) {
2446 btrfs_set_file_extent_num_bytes(leaf, fi,
2448 btrfs_mark_buffer_dirty(leaf);
2449 inode_add_bytes(inode, len);
2455 btrfs_release_path(path);
2460 ret = btrfs_insert_empty_item(trans, root, path, &key,
2463 btrfs_abort_transaction(trans, root, ret);
2467 leaf = path->nodes[0];
2468 item = btrfs_item_ptr(leaf, path->slots[0],
2469 struct btrfs_file_extent_item);
2470 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2471 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2472 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2473 btrfs_set_file_extent_num_bytes(leaf, item, len);
2474 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2475 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2476 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2477 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2478 btrfs_set_file_extent_encryption(leaf, item, 0);
2479 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2481 btrfs_mark_buffer_dirty(leaf);
2482 inode_add_bytes(inode, len);
2483 btrfs_release_path(path);
2485 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2487 backref->root_id, backref->inum,
2488 new->file_pos, 0); /* start - extent_offset */
2490 btrfs_abort_transaction(trans, root, ret);
2496 btrfs_release_path(path);
2497 path->leave_spinning = 0;
2498 btrfs_end_transaction(trans, root);
2500 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2506 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2508 struct old_sa_defrag_extent *old, *tmp;
2513 list_for_each_entry_safe(old, tmp, &new->head, list) {
2514 list_del(&old->list);
2520 static void relink_file_extents(struct new_sa_defrag_extent *new)
2522 struct btrfs_path *path;
2523 struct sa_defrag_extent_backref *backref;
2524 struct sa_defrag_extent_backref *prev = NULL;
2525 struct inode *inode;
2526 struct btrfs_root *root;
2527 struct rb_node *node;
2531 root = BTRFS_I(inode)->root;
2533 path = btrfs_alloc_path();
2537 if (!record_extent_backrefs(path, new)) {
2538 btrfs_free_path(path);
2541 btrfs_release_path(path);
2544 node = rb_first(&new->root);
2547 rb_erase(node, &new->root);
2549 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2551 ret = relink_extent_backref(path, prev, backref);
2564 btrfs_free_path(path);
2566 free_sa_defrag_extent(new);
2568 atomic_dec(&root->fs_info->defrag_running);
2569 wake_up(&root->fs_info->transaction_wait);
2572 static struct new_sa_defrag_extent *
2573 record_old_file_extents(struct inode *inode,
2574 struct btrfs_ordered_extent *ordered)
2576 struct btrfs_root *root = BTRFS_I(inode)->root;
2577 struct btrfs_path *path;
2578 struct btrfs_key key;
2579 struct old_sa_defrag_extent *old;
2580 struct new_sa_defrag_extent *new;
2583 new = kmalloc(sizeof(*new), GFP_NOFS);
2588 new->file_pos = ordered->file_offset;
2589 new->len = ordered->len;
2590 new->bytenr = ordered->start;
2591 new->disk_len = ordered->disk_len;
2592 new->compress_type = ordered->compress_type;
2593 new->root = RB_ROOT;
2594 INIT_LIST_HEAD(&new->head);
2596 path = btrfs_alloc_path();
2600 key.objectid = btrfs_ino(inode);
2601 key.type = BTRFS_EXTENT_DATA_KEY;
2602 key.offset = new->file_pos;
2604 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2607 if (ret > 0 && path->slots[0] > 0)
2610 /* find out all the old extents for the file range */
2612 struct btrfs_file_extent_item *extent;
2613 struct extent_buffer *l;
2622 slot = path->slots[0];
2624 if (slot >= btrfs_header_nritems(l)) {
2625 ret = btrfs_next_leaf(root, path);
2633 btrfs_item_key_to_cpu(l, &key, slot);
2635 if (key.objectid != btrfs_ino(inode))
2637 if (key.type != BTRFS_EXTENT_DATA_KEY)
2639 if (key.offset >= new->file_pos + new->len)
2642 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2644 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2645 if (key.offset + num_bytes < new->file_pos)
2648 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2652 extent_offset = btrfs_file_extent_offset(l, extent);
2654 old = kmalloc(sizeof(*old), GFP_NOFS);
2658 offset = max(new->file_pos, key.offset);
2659 end = min(new->file_pos + new->len, key.offset + num_bytes);
2661 old->bytenr = disk_bytenr;
2662 old->extent_offset = extent_offset;
2663 old->offset = offset - key.offset;
2664 old->len = end - offset;
2667 list_add_tail(&old->list, &new->head);
2673 btrfs_free_path(path);
2674 atomic_inc(&root->fs_info->defrag_running);
2679 btrfs_free_path(path);
2681 free_sa_defrag_extent(new);
2685 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2688 struct btrfs_block_group_cache *cache;
2690 cache = btrfs_lookup_block_group(root->fs_info, start);
2693 spin_lock(&cache->lock);
2694 cache->delalloc_bytes -= len;
2695 spin_unlock(&cache->lock);
2697 btrfs_put_block_group(cache);
2700 /* as ordered data IO finishes, this gets called so we can finish
2701 * an ordered extent if the range of bytes in the file it covers are
2704 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2706 struct inode *inode = ordered_extent->inode;
2707 struct btrfs_root *root = BTRFS_I(inode)->root;
2708 struct btrfs_trans_handle *trans = NULL;
2709 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2710 struct extent_state *cached_state = NULL;
2711 struct new_sa_defrag_extent *new = NULL;
2712 int compress_type = 0;
2714 u64 logical_len = ordered_extent->len;
2716 bool truncated = false;
2718 nolock = btrfs_is_free_space_inode(inode);
2720 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2725 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2726 ordered_extent->file_offset +
2727 ordered_extent->len - 1);
2729 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2731 logical_len = ordered_extent->truncated_len;
2732 /* Truncated the entire extent, don't bother adding */
2737 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2738 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2739 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2741 trans = btrfs_join_transaction_nolock(root);
2743 trans = btrfs_join_transaction(root);
2744 if (IS_ERR(trans)) {
2745 ret = PTR_ERR(trans);
2749 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2750 ret = btrfs_update_inode_fallback(trans, root, inode);
2751 if (ret) /* -ENOMEM or corruption */
2752 btrfs_abort_transaction(trans, root, ret);
2756 lock_extent_bits(io_tree, ordered_extent->file_offset,
2757 ordered_extent->file_offset + ordered_extent->len - 1,
2760 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2761 ordered_extent->file_offset + ordered_extent->len - 1,
2762 EXTENT_DEFRAG, 1, cached_state);
2764 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2765 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2766 /* the inode is shared */
2767 new = record_old_file_extents(inode, ordered_extent);
2769 clear_extent_bit(io_tree, ordered_extent->file_offset,
2770 ordered_extent->file_offset + ordered_extent->len - 1,
2771 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2775 trans = btrfs_join_transaction_nolock(root);
2777 trans = btrfs_join_transaction(root);
2778 if (IS_ERR(trans)) {
2779 ret = PTR_ERR(trans);
2784 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2786 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2787 compress_type = ordered_extent->compress_type;
2788 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2789 BUG_ON(compress_type);
2790 ret = btrfs_mark_extent_written(trans, inode,
2791 ordered_extent->file_offset,
2792 ordered_extent->file_offset +
2795 BUG_ON(root == root->fs_info->tree_root);
2796 ret = insert_reserved_file_extent(trans, inode,
2797 ordered_extent->file_offset,
2798 ordered_extent->start,
2799 ordered_extent->disk_len,
2800 logical_len, logical_len,
2801 compress_type, 0, 0,
2802 BTRFS_FILE_EXTENT_REG);
2804 btrfs_release_delalloc_bytes(root,
2805 ordered_extent->start,
2806 ordered_extent->disk_len);
2808 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2809 ordered_extent->file_offset, ordered_extent->len,
2812 btrfs_abort_transaction(trans, root, ret);
2816 add_pending_csums(trans, inode, ordered_extent->file_offset,
2817 &ordered_extent->list);
2819 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2820 ret = btrfs_update_inode_fallback(trans, root, inode);
2821 if (ret) { /* -ENOMEM or corruption */
2822 btrfs_abort_transaction(trans, root, ret);
2827 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2828 ordered_extent->file_offset +
2829 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2831 if (root != root->fs_info->tree_root)
2832 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2834 btrfs_end_transaction(trans, root);
2836 if (ret || truncated) {
2840 start = ordered_extent->file_offset + logical_len;
2842 start = ordered_extent->file_offset;
2843 end = ordered_extent->file_offset + ordered_extent->len - 1;
2844 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2846 /* Drop the cache for the part of the extent we didn't write. */
2847 btrfs_drop_extent_cache(inode, start, end, 0);
2850 * If the ordered extent had an IOERR or something else went
2851 * wrong we need to return the space for this ordered extent
2852 * back to the allocator. We only free the extent in the
2853 * truncated case if we didn't write out the extent at all.
2855 if ((ret || !logical_len) &&
2856 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2857 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2858 btrfs_free_reserved_extent(root, ordered_extent->start,
2859 ordered_extent->disk_len, 1);
2864 * This needs to be done to make sure anybody waiting knows we are done
2865 * updating everything for this ordered extent.
2867 btrfs_remove_ordered_extent(inode, ordered_extent);
2869 /* for snapshot-aware defrag */
2872 free_sa_defrag_extent(new);
2873 atomic_dec(&root->fs_info->defrag_running);
2875 relink_file_extents(new);
2880 btrfs_put_ordered_extent(ordered_extent);
2881 /* once for the tree */
2882 btrfs_put_ordered_extent(ordered_extent);
2887 static void finish_ordered_fn(struct btrfs_work *work)
2889 struct btrfs_ordered_extent *ordered_extent;
2890 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2891 btrfs_finish_ordered_io(ordered_extent);
2894 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2895 struct extent_state *state, int uptodate)
2897 struct inode *inode = page->mapping->host;
2898 struct btrfs_root *root = BTRFS_I(inode)->root;
2899 struct btrfs_ordered_extent *ordered_extent = NULL;
2900 struct btrfs_workqueue *wq;
2901 btrfs_work_func_t func;
2903 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2905 ClearPagePrivate2(page);
2906 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2907 end - start + 1, uptodate))
2910 if (btrfs_is_free_space_inode(inode)) {
2911 wq = root->fs_info->endio_freespace_worker;
2912 func = btrfs_freespace_write_helper;
2914 wq = root->fs_info->endio_write_workers;
2915 func = btrfs_endio_write_helper;
2918 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
2920 btrfs_queue_work(wq, &ordered_extent->work);
2925 static int __readpage_endio_check(struct inode *inode,
2926 struct btrfs_io_bio *io_bio,
2927 int icsum, struct page *page,
2928 int pgoff, u64 start, size_t len)
2933 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2934 DEFAULT_RATELIMIT_BURST);
2936 csum_expected = *(((u32 *)io_bio->csum) + icsum);
2938 kaddr = kmap_atomic(page);
2939 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
2940 btrfs_csum_final(csum, (char *)&csum);
2941 if (csum != csum_expected)
2944 kunmap_atomic(kaddr);
2947 if (__ratelimit(&_rs))
2948 btrfs_warn(BTRFS_I(inode)->root->fs_info,
2949 "csum failed ino %llu off %llu csum %u expected csum %u",
2950 btrfs_ino(inode), start, csum, csum_expected);
2951 memset(kaddr + pgoff, 1, len);
2952 flush_dcache_page(page);
2953 kunmap_atomic(kaddr);
2954 if (csum_expected == 0)
2960 * when reads are done, we need to check csums to verify the data is correct
2961 * if there's a match, we allow the bio to finish. If not, the code in
2962 * extent_io.c will try to find good copies for us.
2964 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2965 u64 phy_offset, struct page *page,
2966 u64 start, u64 end, int mirror)
2968 size_t offset = start - page_offset(page);
2969 struct inode *inode = page->mapping->host;
2970 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2971 struct btrfs_root *root = BTRFS_I(inode)->root;
2973 if (PageChecked(page)) {
2974 ClearPageChecked(page);
2978 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2981 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2982 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2983 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2988 phy_offset >>= inode->i_sb->s_blocksize_bits;
2989 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
2990 start, (size_t)(end - start + 1));
2993 struct delayed_iput {
2994 struct list_head list;
2995 struct inode *inode;
2998 /* JDM: If this is fs-wide, why can't we add a pointer to
2999 * btrfs_inode instead and avoid the allocation? */
3000 void btrfs_add_delayed_iput(struct inode *inode)
3002 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3003 struct delayed_iput *delayed;
3005 if (atomic_add_unless(&inode->i_count, -1, 1))
3008 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
3009 delayed->inode = inode;
3011 spin_lock(&fs_info->delayed_iput_lock);
3012 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
3013 spin_unlock(&fs_info->delayed_iput_lock);
3016 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3019 struct btrfs_fs_info *fs_info = root->fs_info;
3020 struct delayed_iput *delayed;
3023 spin_lock(&fs_info->delayed_iput_lock);
3024 empty = list_empty(&fs_info->delayed_iputs);
3025 spin_unlock(&fs_info->delayed_iput_lock);
3029 spin_lock(&fs_info->delayed_iput_lock);
3030 list_splice_init(&fs_info->delayed_iputs, &list);
3031 spin_unlock(&fs_info->delayed_iput_lock);
3033 while (!list_empty(&list)) {
3034 delayed = list_entry(list.next, struct delayed_iput, list);
3035 list_del(&delayed->list);
3036 iput(delayed->inode);
3042 * This is called in transaction commit time. If there are no orphan
3043 * files in the subvolume, it removes orphan item and frees block_rsv
3046 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3047 struct btrfs_root *root)
3049 struct btrfs_block_rsv *block_rsv;
3052 if (atomic_read(&root->orphan_inodes) ||
3053 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3056 spin_lock(&root->orphan_lock);
3057 if (atomic_read(&root->orphan_inodes)) {
3058 spin_unlock(&root->orphan_lock);
3062 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3063 spin_unlock(&root->orphan_lock);
3067 block_rsv = root->orphan_block_rsv;
3068 root->orphan_block_rsv = NULL;
3069 spin_unlock(&root->orphan_lock);
3071 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3072 btrfs_root_refs(&root->root_item) > 0) {
3073 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3074 root->root_key.objectid);
3076 btrfs_abort_transaction(trans, root, ret);
3078 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3083 WARN_ON(block_rsv->size > 0);
3084 btrfs_free_block_rsv(root, block_rsv);
3089 * This creates an orphan entry for the given inode in case something goes
3090 * wrong in the middle of an unlink/truncate.
3092 * NOTE: caller of this function should reserve 5 units of metadata for
3095 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3097 struct btrfs_root *root = BTRFS_I(inode)->root;
3098 struct btrfs_block_rsv *block_rsv = NULL;
3103 if (!root->orphan_block_rsv) {
3104 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3109 spin_lock(&root->orphan_lock);
3110 if (!root->orphan_block_rsv) {
3111 root->orphan_block_rsv = block_rsv;
3112 } else if (block_rsv) {
3113 btrfs_free_block_rsv(root, block_rsv);
3117 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3118 &BTRFS_I(inode)->runtime_flags)) {
3121 * For proper ENOSPC handling, we should do orphan
3122 * cleanup when mounting. But this introduces backward
3123 * compatibility issue.
3125 if (!xchg(&root->orphan_item_inserted, 1))
3131 atomic_inc(&root->orphan_inodes);
3134 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3135 &BTRFS_I(inode)->runtime_flags))
3137 spin_unlock(&root->orphan_lock);
3139 /* grab metadata reservation from transaction handle */
3141 ret = btrfs_orphan_reserve_metadata(trans, inode);
3142 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3145 /* insert an orphan item to track this unlinked/truncated file */
3147 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3149 atomic_dec(&root->orphan_inodes);
3151 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3152 &BTRFS_I(inode)->runtime_flags);
3153 btrfs_orphan_release_metadata(inode);
3155 if (ret != -EEXIST) {
3156 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3157 &BTRFS_I(inode)->runtime_flags);
3158 btrfs_abort_transaction(trans, root, ret);
3165 /* insert an orphan item to track subvolume contains orphan files */
3167 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3168 root->root_key.objectid);
3169 if (ret && ret != -EEXIST) {
3170 btrfs_abort_transaction(trans, root, ret);
3178 * We have done the truncate/delete so we can go ahead and remove the orphan
3179 * item for this particular inode.
3181 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3182 struct inode *inode)
3184 struct btrfs_root *root = BTRFS_I(inode)->root;
3185 int delete_item = 0;
3186 int release_rsv = 0;
3189 spin_lock(&root->orphan_lock);
3190 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3191 &BTRFS_I(inode)->runtime_flags))
3194 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3195 &BTRFS_I(inode)->runtime_flags))
3197 spin_unlock(&root->orphan_lock);
3200 atomic_dec(&root->orphan_inodes);
3202 ret = btrfs_del_orphan_item(trans, root,
3207 btrfs_orphan_release_metadata(inode);
3213 * this cleans up any orphans that may be left on the list from the last use
3216 int btrfs_orphan_cleanup(struct btrfs_root *root)
3218 struct btrfs_path *path;
3219 struct extent_buffer *leaf;
3220 struct btrfs_key key, found_key;
3221 struct btrfs_trans_handle *trans;
3222 struct inode *inode;
3223 u64 last_objectid = 0;
3224 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3226 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3229 path = btrfs_alloc_path();
3236 key.objectid = BTRFS_ORPHAN_OBJECTID;
3237 key.type = BTRFS_ORPHAN_ITEM_KEY;
3238 key.offset = (u64)-1;
3241 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3246 * if ret == 0 means we found what we were searching for, which
3247 * is weird, but possible, so only screw with path if we didn't
3248 * find the key and see if we have stuff that matches
3252 if (path->slots[0] == 0)
3257 /* pull out the item */
3258 leaf = path->nodes[0];
3259 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3261 /* make sure the item matches what we want */
3262 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3264 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3267 /* release the path since we're done with it */
3268 btrfs_release_path(path);
3271 * this is where we are basically btrfs_lookup, without the
3272 * crossing root thing. we store the inode number in the
3273 * offset of the orphan item.
3276 if (found_key.offset == last_objectid) {
3277 btrfs_err(root->fs_info,
3278 "Error removing orphan entry, stopping orphan cleanup");
3283 last_objectid = found_key.offset;
3285 found_key.objectid = found_key.offset;
3286 found_key.type = BTRFS_INODE_ITEM_KEY;
3287 found_key.offset = 0;
3288 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3289 ret = PTR_ERR_OR_ZERO(inode);
3290 if (ret && ret != -ESTALE)
3293 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3294 struct btrfs_root *dead_root;
3295 struct btrfs_fs_info *fs_info = root->fs_info;
3296 int is_dead_root = 0;
3299 * this is an orphan in the tree root. Currently these
3300 * could come from 2 sources:
3301 * a) a snapshot deletion in progress
3302 * b) a free space cache inode
3303 * We need to distinguish those two, as the snapshot
3304 * orphan must not get deleted.
3305 * find_dead_roots already ran before us, so if this
3306 * is a snapshot deletion, we should find the root
3307 * in the dead_roots list
3309 spin_lock(&fs_info->trans_lock);
3310 list_for_each_entry(dead_root, &fs_info->dead_roots,
3312 if (dead_root->root_key.objectid ==
3313 found_key.objectid) {
3318 spin_unlock(&fs_info->trans_lock);
3320 /* prevent this orphan from being found again */
3321 key.offset = found_key.objectid - 1;
3326 * Inode is already gone but the orphan item is still there,
3327 * kill the orphan item.
3329 if (ret == -ESTALE) {
3330 trans = btrfs_start_transaction(root, 1);
3331 if (IS_ERR(trans)) {
3332 ret = PTR_ERR(trans);
3335 btrfs_debug(root->fs_info, "auto deleting %Lu",
3336 found_key.objectid);
3337 ret = btrfs_del_orphan_item(trans, root,
3338 found_key.objectid);
3339 btrfs_end_transaction(trans, root);
3346 * add this inode to the orphan list so btrfs_orphan_del does
3347 * the proper thing when we hit it
3349 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3350 &BTRFS_I(inode)->runtime_flags);
3351 atomic_inc(&root->orphan_inodes);
3353 /* if we have links, this was a truncate, lets do that */
3354 if (inode->i_nlink) {
3355 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3361 /* 1 for the orphan item deletion. */
3362 trans = btrfs_start_transaction(root, 1);
3363 if (IS_ERR(trans)) {
3365 ret = PTR_ERR(trans);
3368 ret = btrfs_orphan_add(trans, inode);
3369 btrfs_end_transaction(trans, root);
3375 ret = btrfs_truncate(inode);
3377 btrfs_orphan_del(NULL, inode);
3382 /* this will do delete_inode and everything for us */
3387 /* release the path since we're done with it */
3388 btrfs_release_path(path);
3390 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3392 if (root->orphan_block_rsv)
3393 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3396 if (root->orphan_block_rsv ||
3397 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3398 trans = btrfs_join_transaction(root);
3400 btrfs_end_transaction(trans, root);
3404 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3406 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3410 btrfs_err(root->fs_info,
3411 "could not do orphan cleanup %d", ret);
3412 btrfs_free_path(path);
3417 * very simple check to peek ahead in the leaf looking for xattrs. If we
3418 * don't find any xattrs, we know there can't be any acls.
3420 * slot is the slot the inode is in, objectid is the objectid of the inode
3422 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3423 int slot, u64 objectid,
3424 int *first_xattr_slot)
3426 u32 nritems = btrfs_header_nritems(leaf);
3427 struct btrfs_key found_key;
3428 static u64 xattr_access = 0;
3429 static u64 xattr_default = 0;
3432 if (!xattr_access) {
3433 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3434 strlen(POSIX_ACL_XATTR_ACCESS));
3435 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3436 strlen(POSIX_ACL_XATTR_DEFAULT));
3440 *first_xattr_slot = -1;
3441 while (slot < nritems) {
3442 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3444 /* we found a different objectid, there must not be acls */
3445 if (found_key.objectid != objectid)
3448 /* we found an xattr, assume we've got an acl */
3449 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3450 if (*first_xattr_slot == -1)
3451 *first_xattr_slot = slot;
3452 if (found_key.offset == xattr_access ||
3453 found_key.offset == xattr_default)
3458 * we found a key greater than an xattr key, there can't
3459 * be any acls later on
3461 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3468 * it goes inode, inode backrefs, xattrs, extents,
3469 * so if there are a ton of hard links to an inode there can
3470 * be a lot of backrefs. Don't waste time searching too hard,
3471 * this is just an optimization
3476 /* we hit the end of the leaf before we found an xattr or
3477 * something larger than an xattr. We have to assume the inode
3480 if (*first_xattr_slot == -1)
3481 *first_xattr_slot = slot;
3486 * read an inode from the btree into the in-memory inode
3488 static void btrfs_read_locked_inode(struct inode *inode)
3490 struct btrfs_path *path;
3491 struct extent_buffer *leaf;
3492 struct btrfs_inode_item *inode_item;
3493 struct btrfs_root *root = BTRFS_I(inode)->root;
3494 struct btrfs_key location;
3499 bool filled = false;
3500 int first_xattr_slot;
3502 ret = btrfs_fill_inode(inode, &rdev);
3506 path = btrfs_alloc_path();
3510 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3512 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3516 leaf = path->nodes[0];
3521 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3522 struct btrfs_inode_item);
3523 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3524 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3525 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3526 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3527 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3529 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3530 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3532 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3533 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3535 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3536 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3538 BTRFS_I(inode)->i_otime.tv_sec =
3539 btrfs_timespec_sec(leaf, &inode_item->otime);
3540 BTRFS_I(inode)->i_otime.tv_nsec =
3541 btrfs_timespec_nsec(leaf, &inode_item->otime);
3543 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3544 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3545 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3548 * If we were modified in the current generation and evicted from memory
3549 * and then re-read we need to do a full sync since we don't have any
3550 * idea about which extents were modified before we were evicted from
3553 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3554 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3555 &BTRFS_I(inode)->runtime_flags);
3557 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3558 inode->i_generation = BTRFS_I(inode)->generation;
3560 rdev = btrfs_inode_rdev(leaf, inode_item);
3562 BTRFS_I(inode)->index_cnt = (u64)-1;
3563 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3567 if (inode->i_nlink != 1 ||
3568 path->slots[0] >= btrfs_header_nritems(leaf))
3571 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3572 if (location.objectid != btrfs_ino(inode))
3575 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3576 if (location.type == BTRFS_INODE_REF_KEY) {
3577 struct btrfs_inode_ref *ref;
3579 ref = (struct btrfs_inode_ref *)ptr;
3580 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3581 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3582 struct btrfs_inode_extref *extref;
3584 extref = (struct btrfs_inode_extref *)ptr;
3585 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3590 * try to precache a NULL acl entry for files that don't have
3591 * any xattrs or acls
3593 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3594 btrfs_ino(inode), &first_xattr_slot);
3595 if (first_xattr_slot != -1) {
3596 path->slots[0] = first_xattr_slot;
3597 ret = btrfs_load_inode_props(inode, path);
3599 btrfs_err(root->fs_info,
3600 "error loading props for ino %llu (root %llu): %d",
3602 root->root_key.objectid, ret);
3604 btrfs_free_path(path);
3607 cache_no_acl(inode);
3609 switch (inode->i_mode & S_IFMT) {
3611 inode->i_mapping->a_ops = &btrfs_aops;
3612 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3613 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3614 inode->i_fop = &btrfs_file_operations;
3615 inode->i_op = &btrfs_file_inode_operations;
3618 inode->i_fop = &btrfs_dir_file_operations;
3619 if (root == root->fs_info->tree_root)
3620 inode->i_op = &btrfs_dir_ro_inode_operations;
3622 inode->i_op = &btrfs_dir_inode_operations;
3625 inode->i_op = &btrfs_symlink_inode_operations;
3626 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3627 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3630 inode->i_op = &btrfs_special_inode_operations;
3631 init_special_inode(inode, inode->i_mode, rdev);
3635 btrfs_update_iflags(inode);
3639 btrfs_free_path(path);
3640 make_bad_inode(inode);
3644 * given a leaf and an inode, copy the inode fields into the leaf
3646 static void fill_inode_item(struct btrfs_trans_handle *trans,
3647 struct extent_buffer *leaf,
3648 struct btrfs_inode_item *item,
3649 struct inode *inode)
3651 struct btrfs_map_token token;
3653 btrfs_init_map_token(&token);
3655 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3656 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3657 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3659 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3660 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3662 btrfs_set_token_timespec_sec(leaf, &item->atime,
3663 inode->i_atime.tv_sec, &token);
3664 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3665 inode->i_atime.tv_nsec, &token);
3667 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3668 inode->i_mtime.tv_sec, &token);
3669 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3670 inode->i_mtime.tv_nsec, &token);
3672 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3673 inode->i_ctime.tv_sec, &token);
3674 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3675 inode->i_ctime.tv_nsec, &token);
3677 btrfs_set_token_timespec_sec(leaf, &item->otime,
3678 BTRFS_I(inode)->i_otime.tv_sec, &token);
3679 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3680 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3682 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3684 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3686 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3687 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3688 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3689 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3690 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3694 * copy everything in the in-memory inode into the btree.
3696 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3697 struct btrfs_root *root, struct inode *inode)
3699 struct btrfs_inode_item *inode_item;
3700 struct btrfs_path *path;
3701 struct extent_buffer *leaf;
3704 path = btrfs_alloc_path();
3708 path->leave_spinning = 1;
3709 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3717 leaf = path->nodes[0];
3718 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3719 struct btrfs_inode_item);
3721 fill_inode_item(trans, leaf, inode_item, inode);
3722 btrfs_mark_buffer_dirty(leaf);
3723 btrfs_set_inode_last_trans(trans, inode);
3726 btrfs_free_path(path);
3731 * copy everything in the in-memory inode into the btree.
3733 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3734 struct btrfs_root *root, struct inode *inode)
3739 * If the inode is a free space inode, we can deadlock during commit
3740 * if we put it into the delayed code.
3742 * The data relocation inode should also be directly updated
3745 if (!btrfs_is_free_space_inode(inode)
3746 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3747 && !root->fs_info->log_root_recovering) {
3748 btrfs_update_root_times(trans, root);
3750 ret = btrfs_delayed_update_inode(trans, root, inode);
3752 btrfs_set_inode_last_trans(trans, inode);
3756 return btrfs_update_inode_item(trans, root, inode);
3759 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3760 struct btrfs_root *root,
3761 struct inode *inode)
3765 ret = btrfs_update_inode(trans, root, inode);
3767 return btrfs_update_inode_item(trans, root, inode);
3772 * unlink helper that gets used here in inode.c and in the tree logging
3773 * recovery code. It remove a link in a directory with a given name, and
3774 * also drops the back refs in the inode to the directory
3776 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3777 struct btrfs_root *root,
3778 struct inode *dir, struct inode *inode,
3779 const char *name, int name_len)
3781 struct btrfs_path *path;
3783 struct extent_buffer *leaf;
3784 struct btrfs_dir_item *di;
3785 struct btrfs_key key;
3787 u64 ino = btrfs_ino(inode);
3788 u64 dir_ino = btrfs_ino(dir);
3790 path = btrfs_alloc_path();
3796 path->leave_spinning = 1;
3797 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3798 name, name_len, -1);
3807 leaf = path->nodes[0];
3808 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3809 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3812 btrfs_release_path(path);
3815 * If we don't have dir index, we have to get it by looking up
3816 * the inode ref, since we get the inode ref, remove it directly,
3817 * it is unnecessary to do delayed deletion.
3819 * But if we have dir index, needn't search inode ref to get it.
3820 * Since the inode ref is close to the inode item, it is better
3821 * that we delay to delete it, and just do this deletion when
3822 * we update the inode item.
3824 if (BTRFS_I(inode)->dir_index) {
3825 ret = btrfs_delayed_delete_inode_ref(inode);
3827 index = BTRFS_I(inode)->dir_index;
3832 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3835 btrfs_info(root->fs_info,
3836 "failed to delete reference to %.*s, inode %llu parent %llu",
3837 name_len, name, ino, dir_ino);
3838 btrfs_abort_transaction(trans, root, ret);
3842 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3844 btrfs_abort_transaction(trans, root, ret);
3848 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3850 if (ret != 0 && ret != -ENOENT) {
3851 btrfs_abort_transaction(trans, root, ret);
3855 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3860 btrfs_abort_transaction(trans, root, ret);
3862 btrfs_free_path(path);
3866 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3867 inode_inc_iversion(inode);
3868 inode_inc_iversion(dir);
3869 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3870 ret = btrfs_update_inode(trans, root, dir);
3875 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3876 struct btrfs_root *root,
3877 struct inode *dir, struct inode *inode,
3878 const char *name, int name_len)
3881 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3884 ret = btrfs_update_inode(trans, root, inode);
3890 * helper to start transaction for unlink and rmdir.
3892 * unlink and rmdir are special in btrfs, they do not always free space, so
3893 * if we cannot make our reservations the normal way try and see if there is
3894 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3895 * allow the unlink to occur.
3897 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3899 struct btrfs_trans_handle *trans;
3900 struct btrfs_root *root = BTRFS_I(dir)->root;
3904 * 1 for the possible orphan item
3905 * 1 for the dir item
3906 * 1 for the dir index
3907 * 1 for the inode ref
3910 trans = btrfs_start_transaction(root, 5);
3911 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3914 if (PTR_ERR(trans) == -ENOSPC) {
3915 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3917 trans = btrfs_start_transaction(root, 0);
3920 ret = btrfs_cond_migrate_bytes(root->fs_info,
3921 &root->fs_info->trans_block_rsv,
3924 btrfs_end_transaction(trans, root);
3925 return ERR_PTR(ret);
3927 trans->block_rsv = &root->fs_info->trans_block_rsv;
3928 trans->bytes_reserved = num_bytes;
3933 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3935 struct btrfs_root *root = BTRFS_I(dir)->root;
3936 struct btrfs_trans_handle *trans;
3937 struct inode *inode = dentry->d_inode;
3940 trans = __unlink_start_trans(dir);
3942 return PTR_ERR(trans);
3944 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3946 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3947 dentry->d_name.name, dentry->d_name.len);
3951 if (inode->i_nlink == 0) {
3952 ret = btrfs_orphan_add(trans, inode);
3958 btrfs_end_transaction(trans, root);
3959 btrfs_btree_balance_dirty(root);
3963 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3964 struct btrfs_root *root,
3965 struct inode *dir, u64 objectid,
3966 const char *name, int name_len)
3968 struct btrfs_path *path;
3969 struct extent_buffer *leaf;
3970 struct btrfs_dir_item *di;
3971 struct btrfs_key key;
3974 u64 dir_ino = btrfs_ino(dir);
3976 path = btrfs_alloc_path();
3980 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3981 name, name_len, -1);
3982 if (IS_ERR_OR_NULL(di)) {
3990 leaf = path->nodes[0];
3991 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3992 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3993 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3995 btrfs_abort_transaction(trans, root, ret);
3998 btrfs_release_path(path);
4000 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4001 objectid, root->root_key.objectid,
4002 dir_ino, &index, name, name_len);
4004 if (ret != -ENOENT) {
4005 btrfs_abort_transaction(trans, root, ret);
4008 di = btrfs_search_dir_index_item(root, path, dir_ino,
4010 if (IS_ERR_OR_NULL(di)) {
4015 btrfs_abort_transaction(trans, root, ret);
4019 leaf = path->nodes[0];
4020 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4021 btrfs_release_path(path);
4024 btrfs_release_path(path);
4026 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4028 btrfs_abort_transaction(trans, root, ret);
4032 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4033 inode_inc_iversion(dir);
4034 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4035 ret = btrfs_update_inode_fallback(trans, root, dir);
4037 btrfs_abort_transaction(trans, root, ret);
4039 btrfs_free_path(path);
4043 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4045 struct inode *inode = dentry->d_inode;
4047 struct btrfs_root *root = BTRFS_I(dir)->root;
4048 struct btrfs_trans_handle *trans;
4050 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4052 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4055 trans = __unlink_start_trans(dir);
4057 return PTR_ERR(trans);
4059 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4060 err = btrfs_unlink_subvol(trans, root, dir,
4061 BTRFS_I(inode)->location.objectid,
4062 dentry->d_name.name,
4063 dentry->d_name.len);
4067 err = btrfs_orphan_add(trans, inode);
4071 /* now the directory is empty */
4072 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
4073 dentry->d_name.name, dentry->d_name.len);
4075 btrfs_i_size_write(inode, 0);
4077 btrfs_end_transaction(trans, root);
4078 btrfs_btree_balance_dirty(root);
4084 * this can truncate away extent items, csum items and directory items.
4085 * It starts at a high offset and removes keys until it can't find
4086 * any higher than new_size
4088 * csum items that cross the new i_size are truncated to the new size
4091 * min_type is the minimum key type to truncate down to. If set to 0, this
4092 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4094 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4095 struct btrfs_root *root,
4096 struct inode *inode,
4097 u64 new_size, u32 min_type)
4099 struct btrfs_path *path;
4100 struct extent_buffer *leaf;
4101 struct btrfs_file_extent_item *fi;
4102 struct btrfs_key key;
4103 struct btrfs_key found_key;
4104 u64 extent_start = 0;
4105 u64 extent_num_bytes = 0;
4106 u64 extent_offset = 0;
4108 u64 last_size = (u64)-1;
4109 u32 found_type = (u8)-1;
4112 int pending_del_nr = 0;
4113 int pending_del_slot = 0;
4114 int extent_type = -1;
4117 u64 ino = btrfs_ino(inode);
4119 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4121 path = btrfs_alloc_path();
4127 * We want to drop from the next block forward in case this new size is
4128 * not block aligned since we will be keeping the last block of the
4129 * extent just the way it is.
4131 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4132 root == root->fs_info->tree_root)
4133 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4134 root->sectorsize), (u64)-1, 0);
4137 * This function is also used to drop the items in the log tree before
4138 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4139 * it is used to drop the loged items. So we shouldn't kill the delayed
4142 if (min_type == 0 && root == BTRFS_I(inode)->root)
4143 btrfs_kill_delayed_inode_items(inode);
4146 key.offset = (u64)-1;
4150 path->leave_spinning = 1;
4151 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4158 /* there are no items in the tree for us to truncate, we're
4161 if (path->slots[0] == 0)
4168 leaf = path->nodes[0];
4169 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4170 found_type = found_key.type;
4172 if (found_key.objectid != ino)
4175 if (found_type < min_type)
4178 item_end = found_key.offset;
4179 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4180 fi = btrfs_item_ptr(leaf, path->slots[0],
4181 struct btrfs_file_extent_item);
4182 extent_type = btrfs_file_extent_type(leaf, fi);
4183 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4185 btrfs_file_extent_num_bytes(leaf, fi);
4186 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4187 item_end += btrfs_file_extent_inline_len(leaf,
4188 path->slots[0], fi);
4192 if (found_type > min_type) {
4195 if (item_end < new_size)
4197 if (found_key.offset >= new_size)
4203 /* FIXME, shrink the extent if the ref count is only 1 */
4204 if (found_type != BTRFS_EXTENT_DATA_KEY)
4208 last_size = found_key.offset;
4210 last_size = new_size;
4212 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4214 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4216 u64 orig_num_bytes =
4217 btrfs_file_extent_num_bytes(leaf, fi);
4218 extent_num_bytes = ALIGN(new_size -
4221 btrfs_set_file_extent_num_bytes(leaf, fi,
4223 num_dec = (orig_num_bytes -
4225 if (test_bit(BTRFS_ROOT_REF_COWS,
4228 inode_sub_bytes(inode, num_dec);
4229 btrfs_mark_buffer_dirty(leaf);
4232 btrfs_file_extent_disk_num_bytes(leaf,
4234 extent_offset = found_key.offset -
4235 btrfs_file_extent_offset(leaf, fi);
4237 /* FIXME blocksize != 4096 */
4238 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4239 if (extent_start != 0) {
4241 if (test_bit(BTRFS_ROOT_REF_COWS,
4243 inode_sub_bytes(inode, num_dec);
4246 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4248 * we can't truncate inline items that have had
4252 btrfs_file_extent_compression(leaf, fi) == 0 &&
4253 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4254 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4255 u32 size = new_size - found_key.offset;
4257 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4258 inode_sub_bytes(inode, item_end + 1 -
4262 * update the ram bytes to properly reflect
4263 * the new size of our item
4265 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4267 btrfs_file_extent_calc_inline_size(size);
4268 btrfs_truncate_item(root, path, size, 1);
4269 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4271 inode_sub_bytes(inode, item_end + 1 -
4277 if (!pending_del_nr) {
4278 /* no pending yet, add ourselves */
4279 pending_del_slot = path->slots[0];
4281 } else if (pending_del_nr &&
4282 path->slots[0] + 1 == pending_del_slot) {
4283 /* hop on the pending chunk */
4285 pending_del_slot = path->slots[0];
4293 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4294 root == root->fs_info->tree_root)) {
4295 btrfs_set_path_blocking(path);
4296 ret = btrfs_free_extent(trans, root, extent_start,
4297 extent_num_bytes, 0,
4298 btrfs_header_owner(leaf),
4299 ino, extent_offset, 0);
4303 if (found_type == BTRFS_INODE_ITEM_KEY)
4306 if (path->slots[0] == 0 ||
4307 path->slots[0] != pending_del_slot) {
4308 if (pending_del_nr) {
4309 ret = btrfs_del_items(trans, root, path,
4313 btrfs_abort_transaction(trans,
4319 btrfs_release_path(path);
4326 if (pending_del_nr) {
4327 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4330 btrfs_abort_transaction(trans, root, ret);
4333 if (last_size != (u64)-1 &&
4334 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4335 btrfs_ordered_update_i_size(inode, last_size, NULL);
4336 btrfs_free_path(path);
4341 * btrfs_truncate_page - read, zero a chunk and write a page
4342 * @inode - inode that we're zeroing
4343 * @from - the offset to start zeroing
4344 * @len - the length to zero, 0 to zero the entire range respective to the
4346 * @front - zero up to the offset instead of from the offset on
4348 * This will find the page for the "from" offset and cow the page and zero the
4349 * part we want to zero. This is used with truncate and hole punching.
4351 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4354 struct address_space *mapping = inode->i_mapping;
4355 struct btrfs_root *root = BTRFS_I(inode)->root;
4356 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4357 struct btrfs_ordered_extent *ordered;
4358 struct extent_state *cached_state = NULL;
4360 u32 blocksize = root->sectorsize;
4361 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4362 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4364 gfp_t mask = btrfs_alloc_write_mask(mapping);
4369 if ((offset & (blocksize - 1)) == 0 &&
4370 (!len || ((len & (blocksize - 1)) == 0)))
4372 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4377 page = find_or_create_page(mapping, index, mask);
4379 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4384 page_start = page_offset(page);
4385 page_end = page_start + PAGE_CACHE_SIZE - 1;
4387 if (!PageUptodate(page)) {
4388 ret = btrfs_readpage(NULL, page);
4390 if (page->mapping != mapping) {
4392 page_cache_release(page);
4395 if (!PageUptodate(page)) {
4400 wait_on_page_writeback(page);
4402 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4403 set_page_extent_mapped(page);
4405 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4407 unlock_extent_cached(io_tree, page_start, page_end,
4408 &cached_state, GFP_NOFS);
4410 page_cache_release(page);
4411 btrfs_start_ordered_extent(inode, ordered, 1);
4412 btrfs_put_ordered_extent(ordered);
4416 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4417 EXTENT_DIRTY | EXTENT_DELALLOC |
4418 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4419 0, 0, &cached_state, GFP_NOFS);
4421 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4424 unlock_extent_cached(io_tree, page_start, page_end,
4425 &cached_state, GFP_NOFS);
4429 if (offset != PAGE_CACHE_SIZE) {
4431 len = PAGE_CACHE_SIZE - offset;
4434 memset(kaddr, 0, offset);
4436 memset(kaddr + offset, 0, len);
4437 flush_dcache_page(page);
4440 ClearPageChecked(page);
4441 set_page_dirty(page);
4442 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4447 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4449 page_cache_release(page);
4454 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4455 u64 offset, u64 len)
4457 struct btrfs_trans_handle *trans;
4461 * Still need to make sure the inode looks like it's been updated so
4462 * that any holes get logged if we fsync.
4464 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4465 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4466 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4467 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4472 * 1 - for the one we're dropping
4473 * 1 - for the one we're adding
4474 * 1 - for updating the inode.
4476 trans = btrfs_start_transaction(root, 3);
4478 return PTR_ERR(trans);
4480 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4482 btrfs_abort_transaction(trans, root, ret);
4483 btrfs_end_transaction(trans, root);
4487 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4488 0, 0, len, 0, len, 0, 0, 0);
4490 btrfs_abort_transaction(trans, root, ret);
4492 btrfs_update_inode(trans, root, inode);
4493 btrfs_end_transaction(trans, root);
4498 * This function puts in dummy file extents for the area we're creating a hole
4499 * for. So if we are truncating this file to a larger size we need to insert
4500 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4501 * the range between oldsize and size
4503 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4505 struct btrfs_root *root = BTRFS_I(inode)->root;
4506 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4507 struct extent_map *em = NULL;
4508 struct extent_state *cached_state = NULL;
4509 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4510 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4511 u64 block_end = ALIGN(size, root->sectorsize);
4518 * If our size started in the middle of a page we need to zero out the
4519 * rest of the page before we expand the i_size, otherwise we could
4520 * expose stale data.
4522 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4526 if (size <= hole_start)
4530 struct btrfs_ordered_extent *ordered;
4532 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4534 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4535 block_end - hole_start);
4538 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4539 &cached_state, GFP_NOFS);
4540 btrfs_start_ordered_extent(inode, ordered, 1);
4541 btrfs_put_ordered_extent(ordered);
4544 cur_offset = hole_start;
4546 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4547 block_end - cur_offset, 0);
4553 last_byte = min(extent_map_end(em), block_end);
4554 last_byte = ALIGN(last_byte , root->sectorsize);
4555 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4556 struct extent_map *hole_em;
4557 hole_size = last_byte - cur_offset;
4559 err = maybe_insert_hole(root, inode, cur_offset,
4563 btrfs_drop_extent_cache(inode, cur_offset,
4564 cur_offset + hole_size - 1, 0);
4565 hole_em = alloc_extent_map();
4567 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4568 &BTRFS_I(inode)->runtime_flags);
4571 hole_em->start = cur_offset;
4572 hole_em->len = hole_size;
4573 hole_em->orig_start = cur_offset;
4575 hole_em->block_start = EXTENT_MAP_HOLE;
4576 hole_em->block_len = 0;
4577 hole_em->orig_block_len = 0;
4578 hole_em->ram_bytes = hole_size;
4579 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4580 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4581 hole_em->generation = root->fs_info->generation;
4584 write_lock(&em_tree->lock);
4585 err = add_extent_mapping(em_tree, hole_em, 1);
4586 write_unlock(&em_tree->lock);
4589 btrfs_drop_extent_cache(inode, cur_offset,
4593 free_extent_map(hole_em);
4596 free_extent_map(em);
4598 cur_offset = last_byte;
4599 if (cur_offset >= block_end)
4602 free_extent_map(em);
4603 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4608 static int wait_snapshoting_atomic_t(atomic_t *a)
4614 static void wait_for_snapshot_creation(struct btrfs_root *root)
4619 ret = btrfs_start_write_no_snapshoting(root);
4622 wait_on_atomic_t(&root->will_be_snapshoted,
4623 wait_snapshoting_atomic_t,
4624 TASK_UNINTERRUPTIBLE);
4628 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4630 struct btrfs_root *root = BTRFS_I(inode)->root;
4631 struct btrfs_trans_handle *trans;
4632 loff_t oldsize = i_size_read(inode);
4633 loff_t newsize = attr->ia_size;
4634 int mask = attr->ia_valid;
4638 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4639 * special case where we need to update the times despite not having
4640 * these flags set. For all other operations the VFS set these flags
4641 * explicitly if it wants a timestamp update.
4643 if (newsize != oldsize) {
4644 inode_inc_iversion(inode);
4645 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4646 inode->i_ctime = inode->i_mtime =
4647 current_fs_time(inode->i_sb);
4650 if (newsize > oldsize) {
4651 truncate_pagecache(inode, newsize);
4653 * Don't do an expanding truncate while snapshoting is ongoing.
4654 * This is to ensure the snapshot captures a fully consistent
4655 * state of this file - if the snapshot captures this expanding
4656 * truncation, it must capture all writes that happened before
4659 wait_for_snapshot_creation(root);
4660 ret = btrfs_cont_expand(inode, oldsize, newsize);
4662 btrfs_end_write_no_snapshoting(root);
4666 trans = btrfs_start_transaction(root, 1);
4667 if (IS_ERR(trans)) {
4668 btrfs_end_write_no_snapshoting(root);
4669 return PTR_ERR(trans);
4672 i_size_write(inode, newsize);
4673 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4674 ret = btrfs_update_inode(trans, root, inode);
4675 btrfs_end_write_no_snapshoting(root);
4676 btrfs_end_transaction(trans, root);
4680 * We're truncating a file that used to have good data down to
4681 * zero. Make sure it gets into the ordered flush list so that
4682 * any new writes get down to disk quickly.
4685 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4686 &BTRFS_I(inode)->runtime_flags);
4689 * 1 for the orphan item we're going to add
4690 * 1 for the orphan item deletion.
4692 trans = btrfs_start_transaction(root, 2);
4694 return PTR_ERR(trans);
4697 * We need to do this in case we fail at _any_ point during the
4698 * actual truncate. Once we do the truncate_setsize we could
4699 * invalidate pages which forces any outstanding ordered io to
4700 * be instantly completed which will give us extents that need
4701 * to be truncated. If we fail to get an orphan inode down we
4702 * could have left over extents that were never meant to live,
4703 * so we need to garuntee from this point on that everything
4704 * will be consistent.
4706 ret = btrfs_orphan_add(trans, inode);
4707 btrfs_end_transaction(trans, root);
4711 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4712 truncate_setsize(inode, newsize);
4714 /* Disable nonlocked read DIO to avoid the end less truncate */
4715 btrfs_inode_block_unlocked_dio(inode);
4716 inode_dio_wait(inode);
4717 btrfs_inode_resume_unlocked_dio(inode);
4719 ret = btrfs_truncate(inode);
4720 if (ret && inode->i_nlink) {
4724 * failed to truncate, disk_i_size is only adjusted down
4725 * as we remove extents, so it should represent the true
4726 * size of the inode, so reset the in memory size and
4727 * delete our orphan entry.
4729 trans = btrfs_join_transaction(root);
4730 if (IS_ERR(trans)) {
4731 btrfs_orphan_del(NULL, inode);
4734 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4735 err = btrfs_orphan_del(trans, inode);
4737 btrfs_abort_transaction(trans, root, err);
4738 btrfs_end_transaction(trans, root);
4745 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4747 struct inode *inode = dentry->d_inode;
4748 struct btrfs_root *root = BTRFS_I(inode)->root;
4751 if (btrfs_root_readonly(root))
4754 err = inode_change_ok(inode, attr);
4758 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4759 err = btrfs_setsize(inode, attr);
4764 if (attr->ia_valid) {
4765 setattr_copy(inode, attr);
4766 inode_inc_iversion(inode);
4767 err = btrfs_dirty_inode(inode);
4769 if (!err && attr->ia_valid & ATTR_MODE)
4770 err = posix_acl_chmod(inode, inode->i_mode);
4777 * While truncating the inode pages during eviction, we get the VFS calling
4778 * btrfs_invalidatepage() against each page of the inode. This is slow because
4779 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4780 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4781 * extent_state structures over and over, wasting lots of time.
4783 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4784 * those expensive operations on a per page basis and do only the ordered io
4785 * finishing, while we release here the extent_map and extent_state structures,
4786 * without the excessive merging and splitting.
4788 static void evict_inode_truncate_pages(struct inode *inode)
4790 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4791 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4792 struct rb_node *node;
4794 ASSERT(inode->i_state & I_FREEING);
4795 truncate_inode_pages_final(&inode->i_data);
4797 write_lock(&map_tree->lock);
4798 while (!RB_EMPTY_ROOT(&map_tree->map)) {
4799 struct extent_map *em;
4801 node = rb_first(&map_tree->map);
4802 em = rb_entry(node, struct extent_map, rb_node);
4803 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4804 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4805 remove_extent_mapping(map_tree, em);
4806 free_extent_map(em);
4807 if (need_resched()) {
4808 write_unlock(&map_tree->lock);
4810 write_lock(&map_tree->lock);
4813 write_unlock(&map_tree->lock);
4815 spin_lock(&io_tree->lock);
4816 while (!RB_EMPTY_ROOT(&io_tree->state)) {
4817 struct extent_state *state;
4818 struct extent_state *cached_state = NULL;
4820 node = rb_first(&io_tree->state);
4821 state = rb_entry(node, struct extent_state, rb_node);
4822 atomic_inc(&state->refs);
4823 spin_unlock(&io_tree->lock);
4825 lock_extent_bits(io_tree, state->start, state->end,
4827 clear_extent_bit(io_tree, state->start, state->end,
4828 EXTENT_LOCKED | EXTENT_DIRTY |
4829 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
4830 EXTENT_DEFRAG, 1, 1,
4831 &cached_state, GFP_NOFS);
4832 free_extent_state(state);
4835 spin_lock(&io_tree->lock);
4837 spin_unlock(&io_tree->lock);
4840 void btrfs_evict_inode(struct inode *inode)
4842 struct btrfs_trans_handle *trans;
4843 struct btrfs_root *root = BTRFS_I(inode)->root;
4844 struct btrfs_block_rsv *rsv, *global_rsv;
4845 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4848 trace_btrfs_inode_evict(inode);
4850 evict_inode_truncate_pages(inode);
4852 if (inode->i_nlink &&
4853 ((btrfs_root_refs(&root->root_item) != 0 &&
4854 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4855 btrfs_is_free_space_inode(inode)))
4858 if (is_bad_inode(inode)) {
4859 btrfs_orphan_del(NULL, inode);
4862 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4863 btrfs_wait_ordered_range(inode, 0, (u64)-1);
4865 btrfs_free_io_failure_record(inode, 0, (u64)-1);
4867 if (root->fs_info->log_root_recovering) {
4868 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4869 &BTRFS_I(inode)->runtime_flags));
4873 if (inode->i_nlink > 0) {
4874 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4875 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
4879 ret = btrfs_commit_inode_delayed_inode(inode);
4881 btrfs_orphan_del(NULL, inode);
4885 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4887 btrfs_orphan_del(NULL, inode);
4890 rsv->size = min_size;
4892 global_rsv = &root->fs_info->global_block_rsv;
4894 btrfs_i_size_write(inode, 0);
4897 * This is a bit simpler than btrfs_truncate since we've already
4898 * reserved our space for our orphan item in the unlink, so we just
4899 * need to reserve some slack space in case we add bytes and update
4900 * inode item when doing the truncate.
4903 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4904 BTRFS_RESERVE_FLUSH_LIMIT);
4907 * Try and steal from the global reserve since we will
4908 * likely not use this space anyway, we want to try as
4909 * hard as possible to get this to work.
4912 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4915 btrfs_warn(root->fs_info,
4916 "Could not get space for a delete, will truncate on mount %d",
4918 btrfs_orphan_del(NULL, inode);
4919 btrfs_free_block_rsv(root, rsv);
4923 trans = btrfs_join_transaction(root);
4924 if (IS_ERR(trans)) {
4925 btrfs_orphan_del(NULL, inode);
4926 btrfs_free_block_rsv(root, rsv);
4930 trans->block_rsv = rsv;
4932 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4936 trans->block_rsv = &root->fs_info->trans_block_rsv;
4937 btrfs_end_transaction(trans, root);
4939 btrfs_btree_balance_dirty(root);
4942 btrfs_free_block_rsv(root, rsv);
4945 * Errors here aren't a big deal, it just means we leave orphan items
4946 * in the tree. They will be cleaned up on the next mount.
4949 trans->block_rsv = root->orphan_block_rsv;
4950 btrfs_orphan_del(trans, inode);
4952 btrfs_orphan_del(NULL, inode);
4955 trans->block_rsv = &root->fs_info->trans_block_rsv;
4956 if (!(root == root->fs_info->tree_root ||
4957 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4958 btrfs_return_ino(root, btrfs_ino(inode));
4960 btrfs_end_transaction(trans, root);
4961 btrfs_btree_balance_dirty(root);
4963 btrfs_remove_delayed_node(inode);
4969 * this returns the key found in the dir entry in the location pointer.
4970 * If no dir entries were found, location->objectid is 0.
4972 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4973 struct btrfs_key *location)
4975 const char *name = dentry->d_name.name;
4976 int namelen = dentry->d_name.len;
4977 struct btrfs_dir_item *di;
4978 struct btrfs_path *path;
4979 struct btrfs_root *root = BTRFS_I(dir)->root;
4982 path = btrfs_alloc_path();
4986 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4991 if (IS_ERR_OR_NULL(di))
4994 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4996 btrfs_free_path(path);
4999 location->objectid = 0;
5004 * when we hit a tree root in a directory, the btrfs part of the inode
5005 * needs to be changed to reflect the root directory of the tree root. This
5006 * is kind of like crossing a mount point.
5008 static int fixup_tree_root_location(struct btrfs_root *root,
5010 struct dentry *dentry,
5011 struct btrfs_key *location,
5012 struct btrfs_root **sub_root)
5014 struct btrfs_path *path;
5015 struct btrfs_root *new_root;
5016 struct btrfs_root_ref *ref;
5017 struct extent_buffer *leaf;
5018 struct btrfs_key key;
5022 path = btrfs_alloc_path();
5029 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5030 key.type = BTRFS_ROOT_REF_KEY;
5031 key.offset = location->objectid;
5033 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5041 leaf = path->nodes[0];
5042 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5043 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5044 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5047 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5048 (unsigned long)(ref + 1),
5049 dentry->d_name.len);
5053 btrfs_release_path(path);
5055 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5056 if (IS_ERR(new_root)) {
5057 err = PTR_ERR(new_root);
5061 *sub_root = new_root;
5062 location->objectid = btrfs_root_dirid(&new_root->root_item);
5063 location->type = BTRFS_INODE_ITEM_KEY;
5064 location->offset = 0;
5067 btrfs_free_path(path);
5071 static void inode_tree_add(struct inode *inode)
5073 struct btrfs_root *root = BTRFS_I(inode)->root;
5074 struct btrfs_inode *entry;
5076 struct rb_node *parent;
5077 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5078 u64 ino = btrfs_ino(inode);
5080 if (inode_unhashed(inode))
5083 spin_lock(&root->inode_lock);
5084 p = &root->inode_tree.rb_node;
5087 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5089 if (ino < btrfs_ino(&entry->vfs_inode))
5090 p = &parent->rb_left;
5091 else if (ino > btrfs_ino(&entry->vfs_inode))
5092 p = &parent->rb_right;
5094 WARN_ON(!(entry->vfs_inode.i_state &
5095 (I_WILL_FREE | I_FREEING)));
5096 rb_replace_node(parent, new, &root->inode_tree);
5097 RB_CLEAR_NODE(parent);
5098 spin_unlock(&root->inode_lock);
5102 rb_link_node(new, parent, p);
5103 rb_insert_color(new, &root->inode_tree);
5104 spin_unlock(&root->inode_lock);
5107 static void inode_tree_del(struct inode *inode)
5109 struct btrfs_root *root = BTRFS_I(inode)->root;
5112 spin_lock(&root->inode_lock);
5113 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5114 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5115 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5116 empty = RB_EMPTY_ROOT(&root->inode_tree);
5118 spin_unlock(&root->inode_lock);
5120 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5121 synchronize_srcu(&root->fs_info->subvol_srcu);
5122 spin_lock(&root->inode_lock);
5123 empty = RB_EMPTY_ROOT(&root->inode_tree);
5124 spin_unlock(&root->inode_lock);
5126 btrfs_add_dead_root(root);
5130 void btrfs_invalidate_inodes(struct btrfs_root *root)
5132 struct rb_node *node;
5133 struct rb_node *prev;
5134 struct btrfs_inode *entry;
5135 struct inode *inode;
5138 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5139 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5141 spin_lock(&root->inode_lock);
5143 node = root->inode_tree.rb_node;
5147 entry = rb_entry(node, struct btrfs_inode, rb_node);
5149 if (objectid < btrfs_ino(&entry->vfs_inode))
5150 node = node->rb_left;
5151 else if (objectid > btrfs_ino(&entry->vfs_inode))
5152 node = node->rb_right;
5158 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5159 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5163 prev = rb_next(prev);
5167 entry = rb_entry(node, struct btrfs_inode, rb_node);
5168 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5169 inode = igrab(&entry->vfs_inode);
5171 spin_unlock(&root->inode_lock);
5172 if (atomic_read(&inode->i_count) > 1)
5173 d_prune_aliases(inode);
5175 * btrfs_drop_inode will have it removed from
5176 * the inode cache when its usage count
5181 spin_lock(&root->inode_lock);
5185 if (cond_resched_lock(&root->inode_lock))
5188 node = rb_next(node);
5190 spin_unlock(&root->inode_lock);
5193 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5195 struct btrfs_iget_args *args = p;
5196 inode->i_ino = args->location->objectid;
5197 memcpy(&BTRFS_I(inode)->location, args->location,
5198 sizeof(*args->location));
5199 BTRFS_I(inode)->root = args->root;
5203 static int btrfs_find_actor(struct inode *inode, void *opaque)
5205 struct btrfs_iget_args *args = opaque;
5206 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5207 args->root == BTRFS_I(inode)->root;
5210 static struct inode *btrfs_iget_locked(struct super_block *s,
5211 struct btrfs_key *location,
5212 struct btrfs_root *root)
5214 struct inode *inode;
5215 struct btrfs_iget_args args;
5216 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5218 args.location = location;
5221 inode = iget5_locked(s, hashval, btrfs_find_actor,
5222 btrfs_init_locked_inode,
5227 /* Get an inode object given its location and corresponding root.
5228 * Returns in *is_new if the inode was read from disk
5230 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5231 struct btrfs_root *root, int *new)
5233 struct inode *inode;
5235 inode = btrfs_iget_locked(s, location, root);
5237 return ERR_PTR(-ENOMEM);
5239 if (inode->i_state & I_NEW) {
5240 btrfs_read_locked_inode(inode);
5241 if (!is_bad_inode(inode)) {
5242 inode_tree_add(inode);
5243 unlock_new_inode(inode);
5247 unlock_new_inode(inode);
5249 inode = ERR_PTR(-ESTALE);
5256 static struct inode *new_simple_dir(struct super_block *s,
5257 struct btrfs_key *key,
5258 struct btrfs_root *root)
5260 struct inode *inode = new_inode(s);
5263 return ERR_PTR(-ENOMEM);
5265 BTRFS_I(inode)->root = root;
5266 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5267 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5269 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5270 inode->i_op = &btrfs_dir_ro_inode_operations;
5271 inode->i_fop = &simple_dir_operations;
5272 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5273 inode->i_mtime = CURRENT_TIME;
5274 inode->i_atime = inode->i_mtime;
5275 inode->i_ctime = inode->i_mtime;
5276 BTRFS_I(inode)->i_otime = inode->i_mtime;
5281 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5283 struct inode *inode;
5284 struct btrfs_root *root = BTRFS_I(dir)->root;
5285 struct btrfs_root *sub_root = root;
5286 struct btrfs_key location;
5290 if (dentry->d_name.len > BTRFS_NAME_LEN)
5291 return ERR_PTR(-ENAMETOOLONG);
5293 ret = btrfs_inode_by_name(dir, dentry, &location);
5295 return ERR_PTR(ret);
5297 if (location.objectid == 0)
5298 return ERR_PTR(-ENOENT);
5300 if (location.type == BTRFS_INODE_ITEM_KEY) {
5301 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5305 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5307 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5308 ret = fixup_tree_root_location(root, dir, dentry,
5309 &location, &sub_root);
5312 inode = ERR_PTR(ret);
5314 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5316 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5318 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5320 if (!IS_ERR(inode) && root != sub_root) {
5321 down_read(&root->fs_info->cleanup_work_sem);
5322 if (!(inode->i_sb->s_flags & MS_RDONLY))
5323 ret = btrfs_orphan_cleanup(sub_root);
5324 up_read(&root->fs_info->cleanup_work_sem);
5327 inode = ERR_PTR(ret);
5334 static int btrfs_dentry_delete(const struct dentry *dentry)
5336 struct btrfs_root *root;
5337 struct inode *inode = dentry->d_inode;
5339 if (!inode && !IS_ROOT(dentry))
5340 inode = dentry->d_parent->d_inode;
5343 root = BTRFS_I(inode)->root;
5344 if (btrfs_root_refs(&root->root_item) == 0)
5347 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5353 static void btrfs_dentry_release(struct dentry *dentry)
5355 kfree(dentry->d_fsdata);
5358 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5361 struct inode *inode;
5363 inode = btrfs_lookup_dentry(dir, dentry);
5364 if (IS_ERR(inode)) {
5365 if (PTR_ERR(inode) == -ENOENT)
5368 return ERR_CAST(inode);
5371 return d_splice_alias(inode, dentry);
5374 unsigned char btrfs_filetype_table[] = {
5375 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5378 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5380 struct inode *inode = file_inode(file);
5381 struct btrfs_root *root = BTRFS_I(inode)->root;
5382 struct btrfs_item *item;
5383 struct btrfs_dir_item *di;
5384 struct btrfs_key key;
5385 struct btrfs_key found_key;
5386 struct btrfs_path *path;
5387 struct list_head ins_list;
5388 struct list_head del_list;
5390 struct extent_buffer *leaf;
5392 unsigned char d_type;
5397 int key_type = BTRFS_DIR_INDEX_KEY;
5401 int is_curr = 0; /* ctx->pos points to the current index? */
5403 /* FIXME, use a real flag for deciding about the key type */
5404 if (root->fs_info->tree_root == root)
5405 key_type = BTRFS_DIR_ITEM_KEY;
5407 if (!dir_emit_dots(file, ctx))
5410 path = btrfs_alloc_path();
5416 if (key_type == BTRFS_DIR_INDEX_KEY) {
5417 INIT_LIST_HEAD(&ins_list);
5418 INIT_LIST_HEAD(&del_list);
5419 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5422 key.type = key_type;
5423 key.offset = ctx->pos;
5424 key.objectid = btrfs_ino(inode);
5426 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5431 leaf = path->nodes[0];
5432 slot = path->slots[0];
5433 if (slot >= btrfs_header_nritems(leaf)) {
5434 ret = btrfs_next_leaf(root, path);
5442 item = btrfs_item_nr(slot);
5443 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5445 if (found_key.objectid != key.objectid)
5447 if (found_key.type != key_type)
5449 if (found_key.offset < ctx->pos)
5451 if (key_type == BTRFS_DIR_INDEX_KEY &&
5452 btrfs_should_delete_dir_index(&del_list,
5456 ctx->pos = found_key.offset;
5459 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5461 di_total = btrfs_item_size(leaf, item);
5463 while (di_cur < di_total) {
5464 struct btrfs_key location;
5466 if (verify_dir_item(root, leaf, di))
5469 name_len = btrfs_dir_name_len(leaf, di);
5470 if (name_len <= sizeof(tmp_name)) {
5471 name_ptr = tmp_name;
5473 name_ptr = kmalloc(name_len, GFP_NOFS);
5479 read_extent_buffer(leaf, name_ptr,
5480 (unsigned long)(di + 1), name_len);
5482 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5483 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5486 /* is this a reference to our own snapshot? If so
5489 * In contrast to old kernels, we insert the snapshot's
5490 * dir item and dir index after it has been created, so
5491 * we won't find a reference to our own snapshot. We
5492 * still keep the following code for backward
5495 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5496 location.objectid == root->root_key.objectid) {
5500 over = !dir_emit(ctx, name_ptr, name_len,
5501 location.objectid, d_type);
5504 if (name_ptr != tmp_name)
5509 di_len = btrfs_dir_name_len(leaf, di) +
5510 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5512 di = (struct btrfs_dir_item *)((char *)di + di_len);
5518 if (key_type == BTRFS_DIR_INDEX_KEY) {
5521 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5526 /* Reached end of directory/root. Bump pos past the last item. */
5530 * Stop new entries from being returned after we return the last
5533 * New directory entries are assigned a strictly increasing
5534 * offset. This means that new entries created during readdir
5535 * are *guaranteed* to be seen in the future by that readdir.
5536 * This has broken buggy programs which operate on names as
5537 * they're returned by readdir. Until we re-use freed offsets
5538 * we have this hack to stop new entries from being returned
5539 * under the assumption that they'll never reach this huge
5542 * This is being careful not to overflow 32bit loff_t unless the
5543 * last entry requires it because doing so has broken 32bit apps
5546 if (key_type == BTRFS_DIR_INDEX_KEY) {
5547 if (ctx->pos >= INT_MAX)
5548 ctx->pos = LLONG_MAX;
5555 if (key_type == BTRFS_DIR_INDEX_KEY)
5556 btrfs_put_delayed_items(&ins_list, &del_list);
5557 btrfs_free_path(path);
5561 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5563 struct btrfs_root *root = BTRFS_I(inode)->root;
5564 struct btrfs_trans_handle *trans;
5566 bool nolock = false;
5568 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5571 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5574 if (wbc->sync_mode == WB_SYNC_ALL) {
5576 trans = btrfs_join_transaction_nolock(root);
5578 trans = btrfs_join_transaction(root);
5580 return PTR_ERR(trans);
5581 ret = btrfs_commit_transaction(trans, root);
5587 * This is somewhat expensive, updating the tree every time the
5588 * inode changes. But, it is most likely to find the inode in cache.
5589 * FIXME, needs more benchmarking...there are no reasons other than performance
5590 * to keep or drop this code.
5592 static int btrfs_dirty_inode(struct inode *inode)
5594 struct btrfs_root *root = BTRFS_I(inode)->root;
5595 struct btrfs_trans_handle *trans;
5598 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5601 trans = btrfs_join_transaction(root);
5603 return PTR_ERR(trans);
5605 ret = btrfs_update_inode(trans, root, inode);
5606 if (ret && ret == -ENOSPC) {
5607 /* whoops, lets try again with the full transaction */
5608 btrfs_end_transaction(trans, root);
5609 trans = btrfs_start_transaction(root, 1);
5611 return PTR_ERR(trans);
5613 ret = btrfs_update_inode(trans, root, inode);
5615 btrfs_end_transaction(trans, root);
5616 if (BTRFS_I(inode)->delayed_node)
5617 btrfs_balance_delayed_items(root);
5623 * This is a copy of file_update_time. We need this so we can return error on
5624 * ENOSPC for updating the inode in the case of file write and mmap writes.
5626 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5629 struct btrfs_root *root = BTRFS_I(inode)->root;
5631 if (btrfs_root_readonly(root))
5634 if (flags & S_VERSION)
5635 inode_inc_iversion(inode);
5636 if (flags & S_CTIME)
5637 inode->i_ctime = *now;
5638 if (flags & S_MTIME)
5639 inode->i_mtime = *now;
5640 if (flags & S_ATIME)
5641 inode->i_atime = *now;
5642 return btrfs_dirty_inode(inode);
5646 * find the highest existing sequence number in a directory
5647 * and then set the in-memory index_cnt variable to reflect
5648 * free sequence numbers
5650 static int btrfs_set_inode_index_count(struct inode *inode)
5652 struct btrfs_root *root = BTRFS_I(inode)->root;
5653 struct btrfs_key key, found_key;
5654 struct btrfs_path *path;
5655 struct extent_buffer *leaf;
5658 key.objectid = btrfs_ino(inode);
5659 key.type = BTRFS_DIR_INDEX_KEY;
5660 key.offset = (u64)-1;
5662 path = btrfs_alloc_path();
5666 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5669 /* FIXME: we should be able to handle this */
5675 * MAGIC NUMBER EXPLANATION:
5676 * since we search a directory based on f_pos we have to start at 2
5677 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5678 * else has to start at 2
5680 if (path->slots[0] == 0) {
5681 BTRFS_I(inode)->index_cnt = 2;
5687 leaf = path->nodes[0];
5688 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5690 if (found_key.objectid != btrfs_ino(inode) ||
5691 found_key.type != BTRFS_DIR_INDEX_KEY) {
5692 BTRFS_I(inode)->index_cnt = 2;
5696 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5698 btrfs_free_path(path);
5703 * helper to find a free sequence number in a given directory. This current
5704 * code is very simple, later versions will do smarter things in the btree
5706 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5710 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5711 ret = btrfs_inode_delayed_dir_index_count(dir);
5713 ret = btrfs_set_inode_index_count(dir);
5719 *index = BTRFS_I(dir)->index_cnt;
5720 BTRFS_I(dir)->index_cnt++;
5725 static int btrfs_insert_inode_locked(struct inode *inode)
5727 struct btrfs_iget_args args;
5728 args.location = &BTRFS_I(inode)->location;
5729 args.root = BTRFS_I(inode)->root;
5731 return insert_inode_locked4(inode,
5732 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
5733 btrfs_find_actor, &args);
5736 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5737 struct btrfs_root *root,
5739 const char *name, int name_len,
5740 u64 ref_objectid, u64 objectid,
5741 umode_t mode, u64 *index)
5743 struct inode *inode;
5744 struct btrfs_inode_item *inode_item;
5745 struct btrfs_key *location;
5746 struct btrfs_path *path;
5747 struct btrfs_inode_ref *ref;
5748 struct btrfs_key key[2];
5750 int nitems = name ? 2 : 1;
5754 path = btrfs_alloc_path();
5756 return ERR_PTR(-ENOMEM);
5758 inode = new_inode(root->fs_info->sb);
5760 btrfs_free_path(path);
5761 return ERR_PTR(-ENOMEM);
5765 * O_TMPFILE, set link count to 0, so that after this point,
5766 * we fill in an inode item with the correct link count.
5769 set_nlink(inode, 0);
5772 * we have to initialize this early, so we can reclaim the inode
5773 * number if we fail afterwards in this function.
5775 inode->i_ino = objectid;
5778 trace_btrfs_inode_request(dir);
5780 ret = btrfs_set_inode_index(dir, index);
5782 btrfs_free_path(path);
5784 return ERR_PTR(ret);
5790 * index_cnt is ignored for everything but a dir,
5791 * btrfs_get_inode_index_count has an explanation for the magic
5794 BTRFS_I(inode)->index_cnt = 2;
5795 BTRFS_I(inode)->dir_index = *index;
5796 BTRFS_I(inode)->root = root;
5797 BTRFS_I(inode)->generation = trans->transid;
5798 inode->i_generation = BTRFS_I(inode)->generation;
5801 * We could have gotten an inode number from somebody who was fsynced
5802 * and then removed in this same transaction, so let's just set full
5803 * sync since it will be a full sync anyway and this will blow away the
5804 * old info in the log.
5806 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5808 key[0].objectid = objectid;
5809 key[0].type = BTRFS_INODE_ITEM_KEY;
5812 sizes[0] = sizeof(struct btrfs_inode_item);
5816 * Start new inodes with an inode_ref. This is slightly more
5817 * efficient for small numbers of hard links since they will
5818 * be packed into one item. Extended refs will kick in if we
5819 * add more hard links than can fit in the ref item.
5821 key[1].objectid = objectid;
5822 key[1].type = BTRFS_INODE_REF_KEY;
5823 key[1].offset = ref_objectid;
5825 sizes[1] = name_len + sizeof(*ref);
5828 location = &BTRFS_I(inode)->location;
5829 location->objectid = objectid;
5830 location->offset = 0;
5831 location->type = BTRFS_INODE_ITEM_KEY;
5833 ret = btrfs_insert_inode_locked(inode);
5837 path->leave_spinning = 1;
5838 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
5842 inode_init_owner(inode, dir, mode);
5843 inode_set_bytes(inode, 0);
5845 inode->i_mtime = CURRENT_TIME;
5846 inode->i_atime = inode->i_mtime;
5847 inode->i_ctime = inode->i_mtime;
5848 BTRFS_I(inode)->i_otime = inode->i_mtime;
5850 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5851 struct btrfs_inode_item);
5852 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5853 sizeof(*inode_item));
5854 fill_inode_item(trans, path->nodes[0], inode_item, inode);
5857 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5858 struct btrfs_inode_ref);
5859 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5860 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5861 ptr = (unsigned long)(ref + 1);
5862 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5865 btrfs_mark_buffer_dirty(path->nodes[0]);
5866 btrfs_free_path(path);
5868 btrfs_inherit_iflags(inode, dir);
5870 if (S_ISREG(mode)) {
5871 if (btrfs_test_opt(root, NODATASUM))
5872 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5873 if (btrfs_test_opt(root, NODATACOW))
5874 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5875 BTRFS_INODE_NODATASUM;
5878 inode_tree_add(inode);
5880 trace_btrfs_inode_new(inode);
5881 btrfs_set_inode_last_trans(trans, inode);
5883 btrfs_update_root_times(trans, root);
5885 ret = btrfs_inode_inherit_props(trans, inode, dir);
5887 btrfs_err(root->fs_info,
5888 "error inheriting props for ino %llu (root %llu): %d",
5889 btrfs_ino(inode), root->root_key.objectid, ret);
5894 unlock_new_inode(inode);
5897 BTRFS_I(dir)->index_cnt--;
5898 btrfs_free_path(path);
5900 return ERR_PTR(ret);
5903 static inline u8 btrfs_inode_type(struct inode *inode)
5905 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5909 * utility function to add 'inode' into 'parent_inode' with
5910 * a give name and a given sequence number.
5911 * if 'add_backref' is true, also insert a backref from the
5912 * inode to the parent directory.
5914 int btrfs_add_link(struct btrfs_trans_handle *trans,
5915 struct inode *parent_inode, struct inode *inode,
5916 const char *name, int name_len, int add_backref, u64 index)
5919 struct btrfs_key key;
5920 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5921 u64 ino = btrfs_ino(inode);
5922 u64 parent_ino = btrfs_ino(parent_inode);
5924 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5925 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5928 key.type = BTRFS_INODE_ITEM_KEY;
5932 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5933 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5934 key.objectid, root->root_key.objectid,
5935 parent_ino, index, name, name_len);
5936 } else if (add_backref) {
5937 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5941 /* Nothing to clean up yet */
5945 ret = btrfs_insert_dir_item(trans, root, name, name_len,
5947 btrfs_inode_type(inode), index);
5948 if (ret == -EEXIST || ret == -EOVERFLOW)
5951 btrfs_abort_transaction(trans, root, ret);
5955 btrfs_i_size_write(parent_inode, parent_inode->i_size +
5957 inode_inc_iversion(parent_inode);
5958 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5959 ret = btrfs_update_inode(trans, root, parent_inode);
5961 btrfs_abort_transaction(trans, root, ret);
5965 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5968 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5969 key.objectid, root->root_key.objectid,
5970 parent_ino, &local_index, name, name_len);
5972 } else if (add_backref) {
5976 err = btrfs_del_inode_ref(trans, root, name, name_len,
5977 ino, parent_ino, &local_index);
5982 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5983 struct inode *dir, struct dentry *dentry,
5984 struct inode *inode, int backref, u64 index)
5986 int err = btrfs_add_link(trans, dir, inode,
5987 dentry->d_name.name, dentry->d_name.len,
5994 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5995 umode_t mode, dev_t rdev)
5997 struct btrfs_trans_handle *trans;
5998 struct btrfs_root *root = BTRFS_I(dir)->root;
5999 struct inode *inode = NULL;
6005 if (!new_valid_dev(rdev))
6009 * 2 for inode item and ref
6011 * 1 for xattr if selinux is on
6013 trans = btrfs_start_transaction(root, 5);
6015 return PTR_ERR(trans);
6017 err = btrfs_find_free_ino(root, &objectid);
6021 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6022 dentry->d_name.len, btrfs_ino(dir), objectid,
6024 if (IS_ERR(inode)) {
6025 err = PTR_ERR(inode);
6030 * If the active LSM wants to access the inode during
6031 * d_instantiate it needs these. Smack checks to see
6032 * if the filesystem supports xattrs by looking at the
6035 inode->i_op = &btrfs_special_inode_operations;
6036 init_special_inode(inode, inode->i_mode, rdev);
6038 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6040 goto out_unlock_inode;
6042 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6044 goto out_unlock_inode;
6046 btrfs_update_inode(trans, root, inode);
6047 unlock_new_inode(inode);
6048 d_instantiate(dentry, inode);
6052 btrfs_end_transaction(trans, root);
6053 btrfs_balance_delayed_items(root);
6054 btrfs_btree_balance_dirty(root);
6056 inode_dec_link_count(inode);
6063 unlock_new_inode(inode);
6068 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6069 umode_t mode, bool excl)
6071 struct btrfs_trans_handle *trans;
6072 struct btrfs_root *root = BTRFS_I(dir)->root;
6073 struct inode *inode = NULL;
6074 int drop_inode_on_err = 0;
6080 * 2 for inode item and ref
6082 * 1 for xattr if selinux is on
6084 trans = btrfs_start_transaction(root, 5);
6086 return PTR_ERR(trans);
6088 err = btrfs_find_free_ino(root, &objectid);
6092 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6093 dentry->d_name.len, btrfs_ino(dir), objectid,
6095 if (IS_ERR(inode)) {
6096 err = PTR_ERR(inode);
6099 drop_inode_on_err = 1;
6101 * If the active LSM wants to access the inode during
6102 * d_instantiate it needs these. Smack checks to see
6103 * if the filesystem supports xattrs by looking at the
6106 inode->i_fop = &btrfs_file_operations;
6107 inode->i_op = &btrfs_file_inode_operations;
6108 inode->i_mapping->a_ops = &btrfs_aops;
6109 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
6111 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6113 goto out_unlock_inode;
6115 err = btrfs_update_inode(trans, root, inode);
6117 goto out_unlock_inode;
6119 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6121 goto out_unlock_inode;
6123 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6124 unlock_new_inode(inode);
6125 d_instantiate(dentry, inode);
6128 btrfs_end_transaction(trans, root);
6129 if (err && drop_inode_on_err) {
6130 inode_dec_link_count(inode);
6133 btrfs_balance_delayed_items(root);
6134 btrfs_btree_balance_dirty(root);
6138 unlock_new_inode(inode);
6143 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6144 struct dentry *dentry)
6146 struct btrfs_trans_handle *trans;
6147 struct btrfs_root *root = BTRFS_I(dir)->root;
6148 struct inode *inode = old_dentry->d_inode;
6153 /* do not allow sys_link's with other subvols of the same device */
6154 if (root->objectid != BTRFS_I(inode)->root->objectid)
6157 if (inode->i_nlink >= BTRFS_LINK_MAX)
6160 err = btrfs_set_inode_index(dir, &index);
6165 * 2 items for inode and inode ref
6166 * 2 items for dir items
6167 * 1 item for parent inode
6169 trans = btrfs_start_transaction(root, 5);
6170 if (IS_ERR(trans)) {
6171 err = PTR_ERR(trans);
6175 /* There are several dir indexes for this inode, clear the cache. */
6176 BTRFS_I(inode)->dir_index = 0ULL;
6178 inode_inc_iversion(inode);
6179 inode->i_ctime = CURRENT_TIME;
6181 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6183 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6188 struct dentry *parent = dentry->d_parent;
6189 err = btrfs_update_inode(trans, root, inode);
6192 if (inode->i_nlink == 1) {
6194 * If new hard link count is 1, it's a file created
6195 * with open(2) O_TMPFILE flag.
6197 err = btrfs_orphan_del(trans, inode);
6201 d_instantiate(dentry, inode);
6202 btrfs_log_new_name(trans, inode, NULL, parent);
6205 btrfs_end_transaction(trans, root);
6206 btrfs_balance_delayed_items(root);
6209 inode_dec_link_count(inode);
6212 btrfs_btree_balance_dirty(root);
6216 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6218 struct inode *inode = NULL;
6219 struct btrfs_trans_handle *trans;
6220 struct btrfs_root *root = BTRFS_I(dir)->root;
6222 int drop_on_err = 0;
6227 * 2 items for inode and ref
6228 * 2 items for dir items
6229 * 1 for xattr if selinux is on
6231 trans = btrfs_start_transaction(root, 5);
6233 return PTR_ERR(trans);
6235 err = btrfs_find_free_ino(root, &objectid);
6239 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6240 dentry->d_name.len, btrfs_ino(dir), objectid,
6241 S_IFDIR | mode, &index);
6242 if (IS_ERR(inode)) {
6243 err = PTR_ERR(inode);
6248 /* these must be set before we unlock the inode */
6249 inode->i_op = &btrfs_dir_inode_operations;
6250 inode->i_fop = &btrfs_dir_file_operations;
6252 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6254 goto out_fail_inode;
6256 btrfs_i_size_write(inode, 0);
6257 err = btrfs_update_inode(trans, root, inode);
6259 goto out_fail_inode;
6261 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6262 dentry->d_name.len, 0, index);
6264 goto out_fail_inode;
6266 d_instantiate(dentry, inode);
6268 * mkdir is special. We're unlocking after we call d_instantiate
6269 * to avoid a race with nfsd calling d_instantiate.
6271 unlock_new_inode(inode);
6275 btrfs_end_transaction(trans, root);
6277 inode_dec_link_count(inode);
6280 btrfs_balance_delayed_items(root);
6281 btrfs_btree_balance_dirty(root);
6285 unlock_new_inode(inode);
6289 /* Find next extent map of a given extent map, caller needs to ensure locks */
6290 static struct extent_map *next_extent_map(struct extent_map *em)
6292 struct rb_node *next;
6294 next = rb_next(&em->rb_node);
6297 return container_of(next, struct extent_map, rb_node);
6300 static struct extent_map *prev_extent_map(struct extent_map *em)
6302 struct rb_node *prev;
6304 prev = rb_prev(&em->rb_node);
6307 return container_of(prev, struct extent_map, rb_node);
6310 /* helper for btfs_get_extent. Given an existing extent in the tree,
6311 * the existing extent is the nearest extent to map_start,
6312 * and an extent that you want to insert, deal with overlap and insert
6313 * the best fitted new extent into the tree.
6315 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6316 struct extent_map *existing,
6317 struct extent_map *em,
6320 struct extent_map *prev;
6321 struct extent_map *next;
6326 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6328 if (existing->start > map_start) {
6330 prev = prev_extent_map(next);
6333 next = next_extent_map(prev);
6336 start = prev ? extent_map_end(prev) : em->start;
6337 start = max_t(u64, start, em->start);
6338 end = next ? next->start : extent_map_end(em);
6339 end = min_t(u64, end, extent_map_end(em));
6340 start_diff = start - em->start;
6342 em->len = end - start;
6343 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6344 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6345 em->block_start += start_diff;
6346 em->block_len -= start_diff;
6348 return add_extent_mapping(em_tree, em, 0);
6351 static noinline int uncompress_inline(struct btrfs_path *path,
6352 struct inode *inode, struct page *page,
6353 size_t pg_offset, u64 extent_offset,
6354 struct btrfs_file_extent_item *item)
6357 struct extent_buffer *leaf = path->nodes[0];
6360 unsigned long inline_size;
6364 WARN_ON(pg_offset != 0);
6365 compress_type = btrfs_file_extent_compression(leaf, item);
6366 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6367 inline_size = btrfs_file_extent_inline_item_len(leaf,
6368 btrfs_item_nr(path->slots[0]));
6369 tmp = kmalloc(inline_size, GFP_NOFS);
6372 ptr = btrfs_file_extent_inline_start(item);
6374 read_extent_buffer(leaf, tmp, ptr, inline_size);
6376 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6377 ret = btrfs_decompress(compress_type, tmp, page,
6378 extent_offset, inline_size, max_size);
6384 * a bit scary, this does extent mapping from logical file offset to the disk.
6385 * the ugly parts come from merging extents from the disk with the in-ram
6386 * representation. This gets more complex because of the data=ordered code,
6387 * where the in-ram extents might be locked pending data=ordered completion.
6389 * This also copies inline extents directly into the page.
6392 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6393 size_t pg_offset, u64 start, u64 len,
6398 u64 extent_start = 0;
6400 u64 objectid = btrfs_ino(inode);
6402 struct btrfs_path *path = NULL;
6403 struct btrfs_root *root = BTRFS_I(inode)->root;
6404 struct btrfs_file_extent_item *item;
6405 struct extent_buffer *leaf;
6406 struct btrfs_key found_key;
6407 struct extent_map *em = NULL;
6408 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6409 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6410 struct btrfs_trans_handle *trans = NULL;
6411 const bool new_inline = !page || create;
6414 read_lock(&em_tree->lock);
6415 em = lookup_extent_mapping(em_tree, start, len);
6417 em->bdev = root->fs_info->fs_devices->latest_bdev;
6418 read_unlock(&em_tree->lock);
6421 if (em->start > start || em->start + em->len <= start)
6422 free_extent_map(em);
6423 else if (em->block_start == EXTENT_MAP_INLINE && page)
6424 free_extent_map(em);
6428 em = alloc_extent_map();
6433 em->bdev = root->fs_info->fs_devices->latest_bdev;
6434 em->start = EXTENT_MAP_HOLE;
6435 em->orig_start = EXTENT_MAP_HOLE;
6437 em->block_len = (u64)-1;
6440 path = btrfs_alloc_path();
6446 * Chances are we'll be called again, so go ahead and do
6452 ret = btrfs_lookup_file_extent(trans, root, path,
6453 objectid, start, trans != NULL);
6460 if (path->slots[0] == 0)
6465 leaf = path->nodes[0];
6466 item = btrfs_item_ptr(leaf, path->slots[0],
6467 struct btrfs_file_extent_item);
6468 /* are we inside the extent that was found? */
6469 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6470 found_type = found_key.type;
6471 if (found_key.objectid != objectid ||
6472 found_type != BTRFS_EXTENT_DATA_KEY) {
6474 * If we backup past the first extent we want to move forward
6475 * and see if there is an extent in front of us, otherwise we'll
6476 * say there is a hole for our whole search range which can
6483 found_type = btrfs_file_extent_type(leaf, item);
6484 extent_start = found_key.offset;
6485 if (found_type == BTRFS_FILE_EXTENT_REG ||
6486 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6487 extent_end = extent_start +
6488 btrfs_file_extent_num_bytes(leaf, item);
6489 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6491 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6492 extent_end = ALIGN(extent_start + size, root->sectorsize);
6495 if (start >= extent_end) {
6497 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6498 ret = btrfs_next_leaf(root, path);
6505 leaf = path->nodes[0];
6507 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6508 if (found_key.objectid != objectid ||
6509 found_key.type != BTRFS_EXTENT_DATA_KEY)
6511 if (start + len <= found_key.offset)
6513 if (start > found_key.offset)
6516 em->orig_start = start;
6517 em->len = found_key.offset - start;
6521 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6523 if (found_type == BTRFS_FILE_EXTENT_REG ||
6524 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6526 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6530 size_t extent_offset;
6536 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6537 extent_offset = page_offset(page) + pg_offset - extent_start;
6538 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6539 size - extent_offset);
6540 em->start = extent_start + extent_offset;
6541 em->len = ALIGN(copy_size, root->sectorsize);
6542 em->orig_block_len = em->len;
6543 em->orig_start = em->start;
6544 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6545 if (create == 0 && !PageUptodate(page)) {
6546 if (btrfs_file_extent_compression(leaf, item) !=
6547 BTRFS_COMPRESS_NONE) {
6548 ret = uncompress_inline(path, inode, page,
6550 extent_offset, item);
6557 read_extent_buffer(leaf, map + pg_offset, ptr,
6559 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6560 memset(map + pg_offset + copy_size, 0,
6561 PAGE_CACHE_SIZE - pg_offset -
6566 flush_dcache_page(page);
6567 } else if (create && PageUptodate(page)) {
6571 free_extent_map(em);
6574 btrfs_release_path(path);
6575 trans = btrfs_join_transaction(root);
6578 return ERR_CAST(trans);
6582 write_extent_buffer(leaf, map + pg_offset, ptr,
6585 btrfs_mark_buffer_dirty(leaf);
6587 set_extent_uptodate(io_tree, em->start,
6588 extent_map_end(em) - 1, NULL, GFP_NOFS);
6593 em->orig_start = start;
6596 em->block_start = EXTENT_MAP_HOLE;
6597 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6599 btrfs_release_path(path);
6600 if (em->start > start || extent_map_end(em) <= start) {
6601 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6602 em->start, em->len, start, len);
6608 write_lock(&em_tree->lock);
6609 ret = add_extent_mapping(em_tree, em, 0);
6610 /* it is possible that someone inserted the extent into the tree
6611 * while we had the lock dropped. It is also possible that
6612 * an overlapping map exists in the tree
6614 if (ret == -EEXIST) {
6615 struct extent_map *existing;
6619 existing = search_extent_mapping(em_tree, start, len);
6621 * existing will always be non-NULL, since there must be
6622 * extent causing the -EEXIST.
6624 if (start >= extent_map_end(existing) ||
6625 start <= existing->start) {
6627 * The existing extent map is the one nearest to
6628 * the [start, start + len) range which overlaps
6630 err = merge_extent_mapping(em_tree, existing,
6632 free_extent_map(existing);
6634 free_extent_map(em);
6638 free_extent_map(em);
6643 write_unlock(&em_tree->lock);
6646 trace_btrfs_get_extent(root, em);
6649 btrfs_free_path(path);
6651 ret = btrfs_end_transaction(trans, root);
6656 free_extent_map(em);
6657 return ERR_PTR(err);
6659 BUG_ON(!em); /* Error is always set */
6663 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6664 size_t pg_offset, u64 start, u64 len,
6667 struct extent_map *em;
6668 struct extent_map *hole_em = NULL;
6669 u64 range_start = start;
6675 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6682 * - a pre-alloc extent,
6683 * there might actually be delalloc bytes behind it.
6685 if (em->block_start != EXTENT_MAP_HOLE &&
6686 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6692 /* check to see if we've wrapped (len == -1 or similar) */
6701 /* ok, we didn't find anything, lets look for delalloc */
6702 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6703 end, len, EXTENT_DELALLOC, 1);
6704 found_end = range_start + found;
6705 if (found_end < range_start)
6706 found_end = (u64)-1;
6709 * we didn't find anything useful, return
6710 * the original results from get_extent()
6712 if (range_start > end || found_end <= start) {
6718 /* adjust the range_start to make sure it doesn't
6719 * go backwards from the start they passed in
6721 range_start = max(start, range_start);
6722 found = found_end - range_start;
6725 u64 hole_start = start;
6728 em = alloc_extent_map();
6734 * when btrfs_get_extent can't find anything it
6735 * returns one huge hole
6737 * make sure what it found really fits our range, and
6738 * adjust to make sure it is based on the start from
6742 u64 calc_end = extent_map_end(hole_em);
6744 if (calc_end <= start || (hole_em->start > end)) {
6745 free_extent_map(hole_em);
6748 hole_start = max(hole_em->start, start);
6749 hole_len = calc_end - hole_start;
6753 if (hole_em && range_start > hole_start) {
6754 /* our hole starts before our delalloc, so we
6755 * have to return just the parts of the hole
6756 * that go until the delalloc starts
6758 em->len = min(hole_len,
6759 range_start - hole_start);
6760 em->start = hole_start;
6761 em->orig_start = hole_start;
6763 * don't adjust block start at all,
6764 * it is fixed at EXTENT_MAP_HOLE
6766 em->block_start = hole_em->block_start;
6767 em->block_len = hole_len;
6768 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6769 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6771 em->start = range_start;
6773 em->orig_start = range_start;
6774 em->block_start = EXTENT_MAP_DELALLOC;
6775 em->block_len = found;
6777 } else if (hole_em) {
6782 free_extent_map(hole_em);
6784 free_extent_map(em);
6785 return ERR_PTR(err);
6790 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6793 struct btrfs_root *root = BTRFS_I(inode)->root;
6794 struct extent_map *em;
6795 struct btrfs_key ins;
6799 alloc_hint = get_extent_allocation_hint(inode, start, len);
6800 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
6801 alloc_hint, &ins, 1, 1);
6803 return ERR_PTR(ret);
6805 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6806 ins.offset, ins.offset, ins.offset, 0);
6808 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
6812 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6813 ins.offset, ins.offset, 0);
6815 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
6816 free_extent_map(em);
6817 return ERR_PTR(ret);
6824 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6825 * block must be cow'd
6827 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6828 u64 *orig_start, u64 *orig_block_len,
6831 struct btrfs_trans_handle *trans;
6832 struct btrfs_path *path;
6834 struct extent_buffer *leaf;
6835 struct btrfs_root *root = BTRFS_I(inode)->root;
6836 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6837 struct btrfs_file_extent_item *fi;
6838 struct btrfs_key key;
6845 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6847 path = btrfs_alloc_path();
6851 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
6856 slot = path->slots[0];
6859 /* can't find the item, must cow */
6866 leaf = path->nodes[0];
6867 btrfs_item_key_to_cpu(leaf, &key, slot);
6868 if (key.objectid != btrfs_ino(inode) ||
6869 key.type != BTRFS_EXTENT_DATA_KEY) {
6870 /* not our file or wrong item type, must cow */
6874 if (key.offset > offset) {
6875 /* Wrong offset, must cow */
6879 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6880 found_type = btrfs_file_extent_type(leaf, fi);
6881 if (found_type != BTRFS_FILE_EXTENT_REG &&
6882 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6883 /* not a regular extent, must cow */
6887 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6890 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6891 if (extent_end <= offset)
6894 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6895 if (disk_bytenr == 0)
6898 if (btrfs_file_extent_compression(leaf, fi) ||
6899 btrfs_file_extent_encryption(leaf, fi) ||
6900 btrfs_file_extent_other_encoding(leaf, fi))
6903 backref_offset = btrfs_file_extent_offset(leaf, fi);
6906 *orig_start = key.offset - backref_offset;
6907 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6908 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6911 if (btrfs_extent_readonly(root, disk_bytenr))
6914 num_bytes = min(offset + *len, extent_end) - offset;
6915 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6918 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
6919 ret = test_range_bit(io_tree, offset, range_end,
6920 EXTENT_DELALLOC, 0, NULL);
6927 btrfs_release_path(path);
6930 * look for other files referencing this extent, if we
6931 * find any we must cow
6933 trans = btrfs_join_transaction(root);
6934 if (IS_ERR(trans)) {
6939 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6940 key.offset - backref_offset, disk_bytenr);
6941 btrfs_end_transaction(trans, root);
6948 * adjust disk_bytenr and num_bytes to cover just the bytes
6949 * in this extent we are about to write. If there
6950 * are any csums in that range we have to cow in order
6951 * to keep the csums correct
6953 disk_bytenr += backref_offset;
6954 disk_bytenr += offset - key.offset;
6955 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6958 * all of the above have passed, it is safe to overwrite this extent
6964 btrfs_free_path(path);
6968 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
6970 struct radix_tree_root *root = &inode->i_mapping->page_tree;
6972 void **pagep = NULL;
6973 struct page *page = NULL;
6977 start_idx = start >> PAGE_CACHE_SHIFT;
6980 * end is the last byte in the last page. end == start is legal
6982 end_idx = end >> PAGE_CACHE_SHIFT;
6986 /* Most of the code in this while loop is lifted from
6987 * find_get_page. It's been modified to begin searching from a
6988 * page and return just the first page found in that range. If the
6989 * found idx is less than or equal to the end idx then we know that
6990 * a page exists. If no pages are found or if those pages are
6991 * outside of the range then we're fine (yay!) */
6992 while (page == NULL &&
6993 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
6994 page = radix_tree_deref_slot(pagep);
6995 if (unlikely(!page))
6998 if (radix_tree_exception(page)) {
6999 if (radix_tree_deref_retry(page)) {
7004 * Otherwise, shmem/tmpfs must be storing a swap entry
7005 * here as an exceptional entry: so return it without
7006 * attempting to raise page count.
7009 break; /* TODO: Is this relevant for this use case? */
7012 if (!page_cache_get_speculative(page)) {
7018 * Has the page moved?
7019 * This is part of the lockless pagecache protocol. See
7020 * include/linux/pagemap.h for details.
7022 if (unlikely(page != *pagep)) {
7023 page_cache_release(page);
7029 if (page->index <= end_idx)
7031 page_cache_release(page);
7038 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7039 struct extent_state **cached_state, int writing)
7041 struct btrfs_ordered_extent *ordered;
7045 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7048 * We're concerned with the entire range that we're going to be
7049 * doing DIO to, so we need to make sure theres no ordered
7050 * extents in this range.
7052 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7053 lockend - lockstart + 1);
7056 * We need to make sure there are no buffered pages in this
7057 * range either, we could have raced between the invalidate in
7058 * generic_file_direct_write and locking the extent. The
7059 * invalidate needs to happen so that reads after a write do not
7064 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7067 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7068 cached_state, GFP_NOFS);
7071 btrfs_start_ordered_extent(inode, ordered, 1);
7072 btrfs_put_ordered_extent(ordered);
7074 /* Screw you mmap */
7075 ret = btrfs_fdatawrite_range(inode, lockstart, lockend);
7078 ret = filemap_fdatawait_range(inode->i_mapping,
7085 * If we found a page that couldn't be invalidated just
7086 * fall back to buffered.
7088 ret = invalidate_inode_pages2_range(inode->i_mapping,
7089 lockstart >> PAGE_CACHE_SHIFT,
7090 lockend >> PAGE_CACHE_SHIFT);
7101 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7102 u64 len, u64 orig_start,
7103 u64 block_start, u64 block_len,
7104 u64 orig_block_len, u64 ram_bytes,
7107 struct extent_map_tree *em_tree;
7108 struct extent_map *em;
7109 struct btrfs_root *root = BTRFS_I(inode)->root;
7112 em_tree = &BTRFS_I(inode)->extent_tree;
7113 em = alloc_extent_map();
7115 return ERR_PTR(-ENOMEM);
7118 em->orig_start = orig_start;
7119 em->mod_start = start;
7122 em->block_len = block_len;
7123 em->block_start = block_start;
7124 em->bdev = root->fs_info->fs_devices->latest_bdev;
7125 em->orig_block_len = orig_block_len;
7126 em->ram_bytes = ram_bytes;
7127 em->generation = -1;
7128 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7129 if (type == BTRFS_ORDERED_PREALLOC)
7130 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7133 btrfs_drop_extent_cache(inode, em->start,
7134 em->start + em->len - 1, 0);
7135 write_lock(&em_tree->lock);
7136 ret = add_extent_mapping(em_tree, em, 1);
7137 write_unlock(&em_tree->lock);
7138 } while (ret == -EEXIST);
7141 free_extent_map(em);
7142 return ERR_PTR(ret);
7149 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7150 struct buffer_head *bh_result, int create)
7152 struct extent_map *em;
7153 struct btrfs_root *root = BTRFS_I(inode)->root;
7154 struct extent_state *cached_state = NULL;
7155 u64 start = iblock << inode->i_blkbits;
7156 u64 lockstart, lockend;
7157 u64 len = bh_result->b_size;
7159 int unlock_bits = EXTENT_LOCKED;
7163 unlock_bits |= EXTENT_DIRTY;
7165 len = min_t(u64, len, root->sectorsize);
7168 lockend = start + len - 1;
7171 * If this errors out it's because we couldn't invalidate pagecache for
7172 * this range and we need to fallback to buffered.
7174 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
7177 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7184 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7185 * io. INLINE is special, and we could probably kludge it in here, but
7186 * it's still buffered so for safety lets just fall back to the generic
7189 * For COMPRESSED we _have_ to read the entire extent in so we can
7190 * decompress it, so there will be buffering required no matter what we
7191 * do, so go ahead and fallback to buffered.
7193 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7194 * to buffered IO. Don't blame me, this is the price we pay for using
7197 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7198 em->block_start == EXTENT_MAP_INLINE) {
7199 free_extent_map(em);
7204 /* Just a good old fashioned hole, return */
7205 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7206 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7207 free_extent_map(em);
7212 * We don't allocate a new extent in the following cases
7214 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7216 * 2) The extent is marked as PREALLOC. We're good to go here and can
7217 * just use the extent.
7221 len = min(len, em->len - (start - em->start));
7222 lockstart = start + len;
7226 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7227 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7228 em->block_start != EXTENT_MAP_HOLE)) {
7231 u64 block_start, orig_start, orig_block_len, ram_bytes;
7233 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7234 type = BTRFS_ORDERED_PREALLOC;
7236 type = BTRFS_ORDERED_NOCOW;
7237 len = min(len, em->len - (start - em->start));
7238 block_start = em->block_start + (start - em->start);
7240 if (can_nocow_extent(inode, start, &len, &orig_start,
7241 &orig_block_len, &ram_bytes) == 1) {
7242 if (type == BTRFS_ORDERED_PREALLOC) {
7243 free_extent_map(em);
7244 em = create_pinned_em(inode, start, len,
7255 ret = btrfs_add_ordered_extent_dio(inode, start,
7256 block_start, len, len, type);
7258 free_extent_map(em);
7266 * this will cow the extent, reset the len in case we changed
7269 len = bh_result->b_size;
7270 free_extent_map(em);
7271 em = btrfs_new_extent_direct(inode, start, len);
7276 len = min(len, em->len - (start - em->start));
7278 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7280 bh_result->b_size = len;
7281 bh_result->b_bdev = em->bdev;
7282 set_buffer_mapped(bh_result);
7284 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7285 set_buffer_new(bh_result);
7288 * Need to update the i_size under the extent lock so buffered
7289 * readers will get the updated i_size when we unlock.
7291 if (start + len > i_size_read(inode))
7292 i_size_write(inode, start + len);
7294 if (len < orig_len) {
7295 spin_lock(&BTRFS_I(inode)->lock);
7296 BTRFS_I(inode)->outstanding_extents++;
7297 spin_unlock(&BTRFS_I(inode)->lock);
7299 btrfs_free_reserved_data_space(inode, len);
7303 * In the case of write we need to clear and unlock the entire range,
7304 * in the case of read we need to unlock only the end area that we
7305 * aren't using if there is any left over space.
7307 if (lockstart < lockend) {
7308 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7309 lockend, unlock_bits, 1, 0,
7310 &cached_state, GFP_NOFS);
7312 free_extent_state(cached_state);
7315 free_extent_map(em);
7320 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7321 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7325 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7326 int rw, int mirror_num)
7328 struct btrfs_root *root = BTRFS_I(inode)->root;
7331 BUG_ON(rw & REQ_WRITE);
7335 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7336 BTRFS_WQ_ENDIO_DIO_REPAIR);
7340 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7346 static int btrfs_check_dio_repairable(struct inode *inode,
7347 struct bio *failed_bio,
7348 struct io_failure_record *failrec,
7353 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7354 failrec->logical, failrec->len);
7355 if (num_copies == 1) {
7357 * we only have a single copy of the data, so don't bother with
7358 * all the retry and error correction code that follows. no
7359 * matter what the error is, it is very likely to persist.
7361 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7362 num_copies, failrec->this_mirror, failed_mirror);
7366 failrec->failed_mirror = failed_mirror;
7367 failrec->this_mirror++;
7368 if (failrec->this_mirror == failed_mirror)
7369 failrec->this_mirror++;
7371 if (failrec->this_mirror > num_copies) {
7372 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7373 num_copies, failrec->this_mirror, failed_mirror);
7380 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7381 struct page *page, u64 start, u64 end,
7382 int failed_mirror, bio_end_io_t *repair_endio,
7385 struct io_failure_record *failrec;
7391 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7393 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7397 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7400 free_io_failure(inode, failrec);
7404 if (failed_bio->bi_vcnt > 1)
7405 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7407 read_mode = READ_SYNC;
7409 isector = start - btrfs_io_bio(failed_bio)->logical;
7410 isector >>= inode->i_sb->s_blocksize_bits;
7411 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7412 0, isector, repair_endio, repair_arg);
7414 free_io_failure(inode, failrec);
7418 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7419 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7420 read_mode, failrec->this_mirror, failrec->in_validation);
7422 ret = submit_dio_repair_bio(inode, bio, read_mode,
7423 failrec->this_mirror);
7425 free_io_failure(inode, failrec);
7432 struct btrfs_retry_complete {
7433 struct completion done;
7434 struct inode *inode;
7439 static void btrfs_retry_endio_nocsum(struct bio *bio, int err)
7441 struct btrfs_retry_complete *done = bio->bi_private;
7442 struct bio_vec *bvec;
7449 bio_for_each_segment_all(bvec, bio, i)
7450 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7452 complete(&done->done);
7456 static int __btrfs_correct_data_nocsum(struct inode *inode,
7457 struct btrfs_io_bio *io_bio)
7459 struct bio_vec *bvec;
7460 struct btrfs_retry_complete done;
7465 start = io_bio->logical;
7468 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7472 init_completion(&done.done);
7474 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7475 start + bvec->bv_len - 1,
7477 btrfs_retry_endio_nocsum, &done);
7481 wait_for_completion(&done.done);
7483 if (!done.uptodate) {
7484 /* We might have another mirror, so try again */
7488 start += bvec->bv_len;
7494 static void btrfs_retry_endio(struct bio *bio, int err)
7496 struct btrfs_retry_complete *done = bio->bi_private;
7497 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7498 struct bio_vec *bvec;
7507 bio_for_each_segment_all(bvec, bio, i) {
7508 ret = __readpage_endio_check(done->inode, io_bio, i,
7510 done->start, bvec->bv_len);
7512 clean_io_failure(done->inode, done->start,
7518 done->uptodate = uptodate;
7520 complete(&done->done);
7524 static int __btrfs_subio_endio_read(struct inode *inode,
7525 struct btrfs_io_bio *io_bio, int err)
7527 struct bio_vec *bvec;
7528 struct btrfs_retry_complete done;
7535 start = io_bio->logical;
7538 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7539 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7540 0, start, bvec->bv_len);
7546 init_completion(&done.done);
7548 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7549 start + bvec->bv_len - 1,
7551 btrfs_retry_endio, &done);
7557 wait_for_completion(&done.done);
7559 if (!done.uptodate) {
7560 /* We might have another mirror, so try again */
7564 offset += bvec->bv_len;
7565 start += bvec->bv_len;
7571 static int btrfs_subio_endio_read(struct inode *inode,
7572 struct btrfs_io_bio *io_bio, int err)
7574 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7578 return __btrfs_correct_data_nocsum(inode, io_bio);
7582 return __btrfs_subio_endio_read(inode, io_bio, err);
7586 static void btrfs_endio_direct_read(struct bio *bio, int err)
7588 struct btrfs_dio_private *dip = bio->bi_private;
7589 struct inode *inode = dip->inode;
7590 struct bio *dio_bio;
7591 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7593 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
7594 err = btrfs_subio_endio_read(inode, io_bio, err);
7596 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
7597 dip->logical_offset + dip->bytes - 1);
7598 dio_bio = dip->dio_bio;
7602 /* If we had a csum failure make sure to clear the uptodate flag */
7604 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7605 dio_end_io(dio_bio, err);
7608 io_bio->end_io(io_bio, err);
7612 static void btrfs_endio_direct_write(struct bio *bio, int err)
7614 struct btrfs_dio_private *dip = bio->bi_private;
7615 struct inode *inode = dip->inode;
7616 struct btrfs_root *root = BTRFS_I(inode)->root;
7617 struct btrfs_ordered_extent *ordered = NULL;
7618 u64 ordered_offset = dip->logical_offset;
7619 u64 ordered_bytes = dip->bytes;
7620 struct bio *dio_bio;
7626 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7628 ordered_bytes, !err);
7632 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
7633 finish_ordered_fn, NULL, NULL);
7634 btrfs_queue_work(root->fs_info->endio_write_workers,
7638 * our bio might span multiple ordered extents. If we haven't
7639 * completed the accounting for the whole dio, go back and try again
7641 if (ordered_offset < dip->logical_offset + dip->bytes) {
7642 ordered_bytes = dip->logical_offset + dip->bytes -
7648 dio_bio = dip->dio_bio;
7652 /* If we had an error make sure to clear the uptodate flag */
7654 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7655 dio_end_io(dio_bio, err);
7659 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7660 struct bio *bio, int mirror_num,
7661 unsigned long bio_flags, u64 offset)
7664 struct btrfs_root *root = BTRFS_I(inode)->root;
7665 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7666 BUG_ON(ret); /* -ENOMEM */
7670 static void btrfs_end_dio_bio(struct bio *bio, int err)
7672 struct btrfs_dio_private *dip = bio->bi_private;
7675 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7676 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7677 btrfs_ino(dip->inode), bio->bi_rw,
7678 (unsigned long long)bio->bi_iter.bi_sector,
7679 bio->bi_iter.bi_size, err);
7681 if (dip->subio_endio)
7682 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
7688 * before atomic variable goto zero, we must make sure
7689 * dip->errors is perceived to be set.
7691 smp_mb__before_atomic();
7694 /* if there are more bios still pending for this dio, just exit */
7695 if (!atomic_dec_and_test(&dip->pending_bios))
7699 bio_io_error(dip->orig_bio);
7701 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7702 bio_endio(dip->orig_bio, 0);
7708 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7709 u64 first_sector, gfp_t gfp_flags)
7711 int nr_vecs = bio_get_nr_vecs(bdev);
7712 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7715 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
7716 struct inode *inode,
7717 struct btrfs_dio_private *dip,
7721 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7722 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
7726 * We load all the csum data we need when we submit
7727 * the first bio to reduce the csum tree search and
7730 if (dip->logical_offset == file_offset) {
7731 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
7737 if (bio == dip->orig_bio)
7740 file_offset -= dip->logical_offset;
7741 file_offset >>= inode->i_sb->s_blocksize_bits;
7742 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
7747 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7748 int rw, u64 file_offset, int skip_sum,
7751 struct btrfs_dio_private *dip = bio->bi_private;
7752 int write = rw & REQ_WRITE;
7753 struct btrfs_root *root = BTRFS_I(inode)->root;
7757 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7762 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7763 BTRFS_WQ_ENDIO_DATA);
7771 if (write && async_submit) {
7772 ret = btrfs_wq_submit_bio(root->fs_info,
7773 inode, rw, bio, 0, 0,
7775 __btrfs_submit_bio_start_direct_io,
7776 __btrfs_submit_bio_done);
7780 * If we aren't doing async submit, calculate the csum of the
7783 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7787 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
7793 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7799 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7802 struct inode *inode = dip->inode;
7803 struct btrfs_root *root = BTRFS_I(inode)->root;
7805 struct bio *orig_bio = dip->orig_bio;
7806 struct bio_vec *bvec = orig_bio->bi_io_vec;
7807 u64 start_sector = orig_bio->bi_iter.bi_sector;
7808 u64 file_offset = dip->logical_offset;
7813 int async_submit = 0;
7815 map_length = orig_bio->bi_iter.bi_size;
7816 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7817 &map_length, NULL, 0);
7821 if (map_length >= orig_bio->bi_iter.bi_size) {
7823 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
7827 /* async crcs make it difficult to collect full stripe writes. */
7828 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
7833 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7837 bio->bi_private = dip;
7838 bio->bi_end_io = btrfs_end_dio_bio;
7839 btrfs_io_bio(bio)->logical = file_offset;
7840 atomic_inc(&dip->pending_bios);
7842 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7843 if (map_length < submit_len + bvec->bv_len ||
7844 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7845 bvec->bv_offset) < bvec->bv_len) {
7847 * inc the count before we submit the bio so
7848 * we know the end IO handler won't happen before
7849 * we inc the count. Otherwise, the dip might get freed
7850 * before we're done setting it up
7852 atomic_inc(&dip->pending_bios);
7853 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7854 file_offset, skip_sum,
7858 atomic_dec(&dip->pending_bios);
7862 start_sector += submit_len >> 9;
7863 file_offset += submit_len;
7868 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7869 start_sector, GFP_NOFS);
7872 bio->bi_private = dip;
7873 bio->bi_end_io = btrfs_end_dio_bio;
7874 btrfs_io_bio(bio)->logical = file_offset;
7876 map_length = orig_bio->bi_iter.bi_size;
7877 ret = btrfs_map_block(root->fs_info, rw,
7879 &map_length, NULL, 0);
7885 submit_len += bvec->bv_len;
7892 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7901 * before atomic variable goto zero, we must
7902 * make sure dip->errors is perceived to be set.
7904 smp_mb__before_atomic();
7905 if (atomic_dec_and_test(&dip->pending_bios))
7906 bio_io_error(dip->orig_bio);
7908 /* bio_end_io() will handle error, so we needn't return it */
7912 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7913 struct inode *inode, loff_t file_offset)
7915 struct btrfs_root *root = BTRFS_I(inode)->root;
7916 struct btrfs_dio_private *dip;
7918 struct btrfs_io_bio *btrfs_bio;
7920 int write = rw & REQ_WRITE;
7923 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7925 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7931 dip = kzalloc(sizeof(*dip), GFP_NOFS);
7937 dip->private = dio_bio->bi_private;
7939 dip->logical_offset = file_offset;
7940 dip->bytes = dio_bio->bi_iter.bi_size;
7941 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
7942 io_bio->bi_private = dip;
7943 dip->orig_bio = io_bio;
7944 dip->dio_bio = dio_bio;
7945 atomic_set(&dip->pending_bios, 0);
7946 btrfs_bio = btrfs_io_bio(io_bio);
7947 btrfs_bio->logical = file_offset;
7950 io_bio->bi_end_io = btrfs_endio_direct_write;
7952 io_bio->bi_end_io = btrfs_endio_direct_read;
7953 dip->subio_endio = btrfs_subio_endio_read;
7956 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7960 if (btrfs_bio->end_io)
7961 btrfs_bio->end_io(btrfs_bio, ret);
7967 * If this is a write, we need to clean up the reserved space and kill
7968 * the ordered extent.
7971 struct btrfs_ordered_extent *ordered;
7972 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7973 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7974 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7975 btrfs_free_reserved_extent(root, ordered->start,
7976 ordered->disk_len, 1);
7977 btrfs_put_ordered_extent(ordered);
7978 btrfs_put_ordered_extent(ordered);
7980 bio_endio(dio_bio, ret);
7983 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7984 const struct iov_iter *iter, loff_t offset)
7988 unsigned blocksize_mask = root->sectorsize - 1;
7989 ssize_t retval = -EINVAL;
7991 if (offset & blocksize_mask)
7994 if (iov_iter_alignment(iter) & blocksize_mask)
7997 /* If this is a write we don't need to check anymore */
8001 * Check to make sure we don't have duplicate iov_base's in this
8002 * iovec, if so return EINVAL, otherwise we'll get csum errors
8003 * when reading back.
8005 for (seg = 0; seg < iter->nr_segs; seg++) {
8006 for (i = seg + 1; i < iter->nr_segs; i++) {
8007 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8016 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
8017 struct iov_iter *iter, loff_t offset)
8019 struct file *file = iocb->ki_filp;
8020 struct inode *inode = file->f_mapping->host;
8024 bool relock = false;
8027 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iter, offset))
8030 atomic_inc(&inode->i_dio_count);
8031 smp_mb__after_atomic();
8034 * The generic stuff only does filemap_write_and_wait_range, which
8035 * isn't enough if we've written compressed pages to this area, so
8036 * we need to flush the dirty pages again to make absolutely sure
8037 * that any outstanding dirty pages are on disk.
8039 count = iov_iter_count(iter);
8040 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8041 &BTRFS_I(inode)->runtime_flags))
8042 filemap_fdatawrite_range(inode->i_mapping, offset,
8043 offset + count - 1);
8047 * If the write DIO is beyond the EOF, we need update
8048 * the isize, but it is protected by i_mutex. So we can
8049 * not unlock the i_mutex at this case.
8051 if (offset + count <= inode->i_size) {
8052 mutex_unlock(&inode->i_mutex);
8055 ret = btrfs_delalloc_reserve_space(inode, count);
8058 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8059 &BTRFS_I(inode)->runtime_flags)) {
8060 inode_dio_done(inode);
8061 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8065 ret = __blockdev_direct_IO(rw, iocb, inode,
8066 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8067 iter, offset, btrfs_get_blocks_direct, NULL,
8068 btrfs_submit_direct, flags);
8070 if (ret < 0 && ret != -EIOCBQUEUED)
8071 btrfs_delalloc_release_space(inode, count);
8072 else if (ret >= 0 && (size_t)ret < count)
8073 btrfs_delalloc_release_space(inode,
8074 count - (size_t)ret);
8078 inode_dio_done(inode);
8080 mutex_lock(&inode->i_mutex);
8085 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8087 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8088 __u64 start, __u64 len)
8092 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8096 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8099 int btrfs_readpage(struct file *file, struct page *page)
8101 struct extent_io_tree *tree;
8102 tree = &BTRFS_I(page->mapping->host)->io_tree;
8103 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8106 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8108 struct extent_io_tree *tree;
8111 if (current->flags & PF_MEMALLOC) {
8112 redirty_page_for_writepage(wbc, page);
8116 tree = &BTRFS_I(page->mapping->host)->io_tree;
8117 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8120 static int btrfs_writepages(struct address_space *mapping,
8121 struct writeback_control *wbc)
8123 struct extent_io_tree *tree;
8125 tree = &BTRFS_I(mapping->host)->io_tree;
8126 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8130 btrfs_readpages(struct file *file, struct address_space *mapping,
8131 struct list_head *pages, unsigned nr_pages)
8133 struct extent_io_tree *tree;
8134 tree = &BTRFS_I(mapping->host)->io_tree;
8135 return extent_readpages(tree, mapping, pages, nr_pages,
8138 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8140 struct extent_io_tree *tree;
8141 struct extent_map_tree *map;
8144 tree = &BTRFS_I(page->mapping->host)->io_tree;
8145 map = &BTRFS_I(page->mapping->host)->extent_tree;
8146 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8148 ClearPagePrivate(page);
8149 set_page_private(page, 0);
8150 page_cache_release(page);
8155 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8157 if (PageWriteback(page) || PageDirty(page))
8159 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8162 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8163 unsigned int length)
8165 struct inode *inode = page->mapping->host;
8166 struct extent_io_tree *tree;
8167 struct btrfs_ordered_extent *ordered;
8168 struct extent_state *cached_state = NULL;
8169 u64 page_start = page_offset(page);
8170 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
8171 int inode_evicting = inode->i_state & I_FREEING;
8174 * we have the page locked, so new writeback can't start,
8175 * and the dirty bit won't be cleared while we are here.
8177 * Wait for IO on this page so that we can safely clear
8178 * the PagePrivate2 bit and do ordered accounting
8180 wait_on_page_writeback(page);
8182 tree = &BTRFS_I(inode)->io_tree;
8184 btrfs_releasepage(page, GFP_NOFS);
8188 if (!inode_evicting)
8189 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
8190 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8193 * IO on this page will never be started, so we need
8194 * to account for any ordered extents now
8196 if (!inode_evicting)
8197 clear_extent_bit(tree, page_start, page_end,
8198 EXTENT_DIRTY | EXTENT_DELALLOC |
8199 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8200 EXTENT_DEFRAG, 1, 0, &cached_state,
8203 * whoever cleared the private bit is responsible
8204 * for the finish_ordered_io
8206 if (TestClearPagePrivate2(page)) {
8207 struct btrfs_ordered_inode_tree *tree;
8210 tree = &BTRFS_I(inode)->ordered_tree;
8212 spin_lock_irq(&tree->lock);
8213 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8214 new_len = page_start - ordered->file_offset;
8215 if (new_len < ordered->truncated_len)
8216 ordered->truncated_len = new_len;
8217 spin_unlock_irq(&tree->lock);
8219 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8221 PAGE_CACHE_SIZE, 1))
8222 btrfs_finish_ordered_io(ordered);
8224 btrfs_put_ordered_extent(ordered);
8225 if (!inode_evicting) {
8226 cached_state = NULL;
8227 lock_extent_bits(tree, page_start, page_end, 0,
8232 if (!inode_evicting) {
8233 clear_extent_bit(tree, page_start, page_end,
8234 EXTENT_LOCKED | EXTENT_DIRTY |
8235 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8236 EXTENT_DEFRAG, 1, 1,
8237 &cached_state, GFP_NOFS);
8239 __btrfs_releasepage(page, GFP_NOFS);
8242 ClearPageChecked(page);
8243 if (PagePrivate(page)) {
8244 ClearPagePrivate(page);
8245 set_page_private(page, 0);
8246 page_cache_release(page);
8251 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8252 * called from a page fault handler when a page is first dirtied. Hence we must
8253 * be careful to check for EOF conditions here. We set the page up correctly
8254 * for a written page which means we get ENOSPC checking when writing into
8255 * holes and correct delalloc and unwritten extent mapping on filesystems that
8256 * support these features.
8258 * We are not allowed to take the i_mutex here so we have to play games to
8259 * protect against truncate races as the page could now be beyond EOF. Because
8260 * vmtruncate() writes the inode size before removing pages, once we have the
8261 * page lock we can determine safely if the page is beyond EOF. If it is not
8262 * beyond EOF, then the page is guaranteed safe against truncation until we
8265 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8267 struct page *page = vmf->page;
8268 struct inode *inode = file_inode(vma->vm_file);
8269 struct btrfs_root *root = BTRFS_I(inode)->root;
8270 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8271 struct btrfs_ordered_extent *ordered;
8272 struct extent_state *cached_state = NULL;
8274 unsigned long zero_start;
8281 sb_start_pagefault(inode->i_sb);
8282 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
8284 ret = file_update_time(vma->vm_file);
8290 else /* -ENOSPC, -EIO, etc */
8291 ret = VM_FAULT_SIGBUS;
8297 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8300 size = i_size_read(inode);
8301 page_start = page_offset(page);
8302 page_end = page_start + PAGE_CACHE_SIZE - 1;
8304 if ((page->mapping != inode->i_mapping) ||
8305 (page_start >= size)) {
8306 /* page got truncated out from underneath us */
8309 wait_on_page_writeback(page);
8311 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
8312 set_page_extent_mapped(page);
8315 * we can't set the delalloc bits if there are pending ordered
8316 * extents. Drop our locks and wait for them to finish
8318 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8320 unlock_extent_cached(io_tree, page_start, page_end,
8321 &cached_state, GFP_NOFS);
8323 btrfs_start_ordered_extent(inode, ordered, 1);
8324 btrfs_put_ordered_extent(ordered);
8329 * XXX - page_mkwrite gets called every time the page is dirtied, even
8330 * if it was already dirty, so for space accounting reasons we need to
8331 * clear any delalloc bits for the range we are fixing to save. There
8332 * is probably a better way to do this, but for now keep consistent with
8333 * prepare_pages in the normal write path.
8335 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
8336 EXTENT_DIRTY | EXTENT_DELALLOC |
8337 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8338 0, 0, &cached_state, GFP_NOFS);
8340 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
8343 unlock_extent_cached(io_tree, page_start, page_end,
8344 &cached_state, GFP_NOFS);
8345 ret = VM_FAULT_SIGBUS;
8350 /* page is wholly or partially inside EOF */
8351 if (page_start + PAGE_CACHE_SIZE > size)
8352 zero_start = size & ~PAGE_CACHE_MASK;
8354 zero_start = PAGE_CACHE_SIZE;
8356 if (zero_start != PAGE_CACHE_SIZE) {
8358 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
8359 flush_dcache_page(page);
8362 ClearPageChecked(page);
8363 set_page_dirty(page);
8364 SetPageUptodate(page);
8366 BTRFS_I(inode)->last_trans = root->fs_info->generation;
8367 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8368 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8370 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
8374 sb_end_pagefault(inode->i_sb);
8375 return VM_FAULT_LOCKED;
8379 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
8381 sb_end_pagefault(inode->i_sb);
8385 static int btrfs_truncate(struct inode *inode)
8387 struct btrfs_root *root = BTRFS_I(inode)->root;
8388 struct btrfs_block_rsv *rsv;
8391 struct btrfs_trans_handle *trans;
8392 u64 mask = root->sectorsize - 1;
8393 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
8395 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8401 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8402 * 3 things going on here
8404 * 1) We need to reserve space for our orphan item and the space to
8405 * delete our orphan item. Lord knows we don't want to have a dangling
8406 * orphan item because we didn't reserve space to remove it.
8408 * 2) We need to reserve space to update our inode.
8410 * 3) We need to have something to cache all the space that is going to
8411 * be free'd up by the truncate operation, but also have some slack
8412 * space reserved in case it uses space during the truncate (thank you
8413 * very much snapshotting).
8415 * And we need these to all be seperate. The fact is we can use alot of
8416 * space doing the truncate, and we have no earthly idea how much space
8417 * we will use, so we need the truncate reservation to be seperate so it
8418 * doesn't end up using space reserved for updating the inode or
8419 * removing the orphan item. We also need to be able to stop the
8420 * transaction and start a new one, which means we need to be able to
8421 * update the inode several times, and we have no idea of knowing how
8422 * many times that will be, so we can't just reserve 1 item for the
8423 * entirety of the opration, so that has to be done seperately as well.
8424 * Then there is the orphan item, which does indeed need to be held on
8425 * to for the whole operation, and we need nobody to touch this reserved
8426 * space except the orphan code.
8428 * So that leaves us with
8430 * 1) root->orphan_block_rsv - for the orphan deletion.
8431 * 2) rsv - for the truncate reservation, which we will steal from the
8432 * transaction reservation.
8433 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8434 * updating the inode.
8436 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
8439 rsv->size = min_size;
8443 * 1 for the truncate slack space
8444 * 1 for updating the inode.
8446 trans = btrfs_start_transaction(root, 2);
8447 if (IS_ERR(trans)) {
8448 err = PTR_ERR(trans);
8452 /* Migrate the slack space for the truncate to our reserve */
8453 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
8458 * So if we truncate and then write and fsync we normally would just
8459 * write the extents that changed, which is a problem if we need to
8460 * first truncate that entire inode. So set this flag so we write out
8461 * all of the extents in the inode to the sync log so we're completely
8464 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8465 trans->block_rsv = rsv;
8468 ret = btrfs_truncate_inode_items(trans, root, inode,
8470 BTRFS_EXTENT_DATA_KEY);
8471 if (ret != -ENOSPC) {
8476 trans->block_rsv = &root->fs_info->trans_block_rsv;
8477 ret = btrfs_update_inode(trans, root, inode);
8483 btrfs_end_transaction(trans, root);
8484 btrfs_btree_balance_dirty(root);
8486 trans = btrfs_start_transaction(root, 2);
8487 if (IS_ERR(trans)) {
8488 ret = err = PTR_ERR(trans);
8493 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
8495 BUG_ON(ret); /* shouldn't happen */
8496 trans->block_rsv = rsv;
8499 if (ret == 0 && inode->i_nlink > 0) {
8500 trans->block_rsv = root->orphan_block_rsv;
8501 ret = btrfs_orphan_del(trans, inode);
8507 trans->block_rsv = &root->fs_info->trans_block_rsv;
8508 ret = btrfs_update_inode(trans, root, inode);
8512 ret = btrfs_end_transaction(trans, root);
8513 btrfs_btree_balance_dirty(root);
8517 btrfs_free_block_rsv(root, rsv);
8526 * create a new subvolume directory/inode (helper for the ioctl).
8528 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8529 struct btrfs_root *new_root,
8530 struct btrfs_root *parent_root,
8533 struct inode *inode;
8537 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8538 new_dirid, new_dirid,
8539 S_IFDIR | (~current_umask() & S_IRWXUGO),
8542 return PTR_ERR(inode);
8543 inode->i_op = &btrfs_dir_inode_operations;
8544 inode->i_fop = &btrfs_dir_file_operations;
8546 set_nlink(inode, 1);
8547 btrfs_i_size_write(inode, 0);
8548 unlock_new_inode(inode);
8550 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8552 btrfs_err(new_root->fs_info,
8553 "error inheriting subvolume %llu properties: %d",
8554 new_root->root_key.objectid, err);
8556 err = btrfs_update_inode(trans, new_root, inode);
8562 struct inode *btrfs_alloc_inode(struct super_block *sb)
8564 struct btrfs_inode *ei;
8565 struct inode *inode;
8567 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
8574 ei->last_sub_trans = 0;
8575 ei->logged_trans = 0;
8576 ei->delalloc_bytes = 0;
8577 ei->defrag_bytes = 0;
8578 ei->disk_i_size = 0;
8581 ei->index_cnt = (u64)-1;
8583 ei->last_unlink_trans = 0;
8584 ei->last_log_commit = 0;
8586 spin_lock_init(&ei->lock);
8587 ei->outstanding_extents = 0;
8588 ei->reserved_extents = 0;
8590 ei->runtime_flags = 0;
8591 ei->force_compress = BTRFS_COMPRESS_NONE;
8593 ei->delayed_node = NULL;
8595 ei->i_otime.tv_sec = 0;
8596 ei->i_otime.tv_nsec = 0;
8598 inode = &ei->vfs_inode;
8599 extent_map_tree_init(&ei->extent_tree);
8600 extent_io_tree_init(&ei->io_tree, &inode->i_data);
8601 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
8602 ei->io_tree.track_uptodate = 1;
8603 ei->io_failure_tree.track_uptodate = 1;
8604 atomic_set(&ei->sync_writers, 0);
8605 mutex_init(&ei->log_mutex);
8606 mutex_init(&ei->delalloc_mutex);
8607 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8608 INIT_LIST_HEAD(&ei->delalloc_inodes);
8609 RB_CLEAR_NODE(&ei->rb_node);
8614 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8615 void btrfs_test_destroy_inode(struct inode *inode)
8617 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8618 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8622 static void btrfs_i_callback(struct rcu_head *head)
8624 struct inode *inode = container_of(head, struct inode, i_rcu);
8625 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8628 void btrfs_destroy_inode(struct inode *inode)
8630 struct btrfs_ordered_extent *ordered;
8631 struct btrfs_root *root = BTRFS_I(inode)->root;
8633 WARN_ON(!hlist_empty(&inode->i_dentry));
8634 WARN_ON(inode->i_data.nrpages);
8635 WARN_ON(BTRFS_I(inode)->outstanding_extents);
8636 WARN_ON(BTRFS_I(inode)->reserved_extents);
8637 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8638 WARN_ON(BTRFS_I(inode)->csum_bytes);
8639 WARN_ON(BTRFS_I(inode)->defrag_bytes);
8642 * This can happen where we create an inode, but somebody else also
8643 * created the same inode and we need to destroy the one we already
8649 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8650 &BTRFS_I(inode)->runtime_flags)) {
8651 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
8653 atomic_dec(&root->orphan_inodes);
8657 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8661 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
8662 ordered->file_offset, ordered->len);
8663 btrfs_remove_ordered_extent(inode, ordered);
8664 btrfs_put_ordered_extent(ordered);
8665 btrfs_put_ordered_extent(ordered);
8668 inode_tree_del(inode);
8669 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8671 call_rcu(&inode->i_rcu, btrfs_i_callback);
8674 int btrfs_drop_inode(struct inode *inode)
8676 struct btrfs_root *root = BTRFS_I(inode)->root;
8681 /* the snap/subvol tree is on deleting */
8682 if (btrfs_root_refs(&root->root_item) == 0)
8685 return generic_drop_inode(inode);
8688 static void init_once(void *foo)
8690 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8692 inode_init_once(&ei->vfs_inode);
8695 void btrfs_destroy_cachep(void)
8698 * Make sure all delayed rcu free inodes are flushed before we
8702 if (btrfs_inode_cachep)
8703 kmem_cache_destroy(btrfs_inode_cachep);
8704 if (btrfs_trans_handle_cachep)
8705 kmem_cache_destroy(btrfs_trans_handle_cachep);
8706 if (btrfs_transaction_cachep)
8707 kmem_cache_destroy(btrfs_transaction_cachep);
8708 if (btrfs_path_cachep)
8709 kmem_cache_destroy(btrfs_path_cachep);
8710 if (btrfs_free_space_cachep)
8711 kmem_cache_destroy(btrfs_free_space_cachep);
8712 if (btrfs_delalloc_work_cachep)
8713 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8716 int btrfs_init_cachep(void)
8718 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8719 sizeof(struct btrfs_inode), 0,
8720 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8721 if (!btrfs_inode_cachep)
8724 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8725 sizeof(struct btrfs_trans_handle), 0,
8726 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8727 if (!btrfs_trans_handle_cachep)
8730 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8731 sizeof(struct btrfs_transaction), 0,
8732 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8733 if (!btrfs_transaction_cachep)
8736 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8737 sizeof(struct btrfs_path), 0,
8738 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8739 if (!btrfs_path_cachep)
8742 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8743 sizeof(struct btrfs_free_space), 0,
8744 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8745 if (!btrfs_free_space_cachep)
8748 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8749 sizeof(struct btrfs_delalloc_work), 0,
8750 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8752 if (!btrfs_delalloc_work_cachep)
8757 btrfs_destroy_cachep();
8761 static int btrfs_getattr(struct vfsmount *mnt,
8762 struct dentry *dentry, struct kstat *stat)
8765 struct inode *inode = dentry->d_inode;
8766 u32 blocksize = inode->i_sb->s_blocksize;
8768 generic_fillattr(inode, stat);
8769 stat->dev = BTRFS_I(inode)->root->anon_dev;
8770 stat->blksize = PAGE_CACHE_SIZE;
8772 spin_lock(&BTRFS_I(inode)->lock);
8773 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8774 spin_unlock(&BTRFS_I(inode)->lock);
8775 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8776 ALIGN(delalloc_bytes, blocksize)) >> 9;
8780 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8781 struct inode *new_dir, struct dentry *new_dentry)
8783 struct btrfs_trans_handle *trans;
8784 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8785 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8786 struct inode *new_inode = new_dentry->d_inode;
8787 struct inode *old_inode = old_dentry->d_inode;
8788 struct timespec ctime = CURRENT_TIME;
8792 u64 old_ino = btrfs_ino(old_inode);
8794 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8797 /* we only allow rename subvolume link between subvolumes */
8798 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8801 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8802 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8805 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8806 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8810 /* check for collisions, even if the name isn't there */
8811 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
8812 new_dentry->d_name.name,
8813 new_dentry->d_name.len);
8816 if (ret == -EEXIST) {
8818 * eexist without a new_inode */
8819 if (WARN_ON(!new_inode)) {
8823 /* maybe -EOVERFLOW */
8830 * we're using rename to replace one file with another. Start IO on it
8831 * now so we don't add too much work to the end of the transaction
8833 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
8834 filemap_flush(old_inode->i_mapping);
8836 /* close the racy window with snapshot create/destroy ioctl */
8837 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8838 down_read(&root->fs_info->subvol_sem);
8840 * We want to reserve the absolute worst case amount of items. So if
8841 * both inodes are subvols and we need to unlink them then that would
8842 * require 4 item modifications, but if they are both normal inodes it
8843 * would require 5 item modifications, so we'll assume their normal
8844 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8845 * should cover the worst case number of items we'll modify.
8847 trans = btrfs_start_transaction(root, 11);
8848 if (IS_ERR(trans)) {
8849 ret = PTR_ERR(trans);
8854 btrfs_record_root_in_trans(trans, dest);
8856 ret = btrfs_set_inode_index(new_dir, &index);
8860 BTRFS_I(old_inode)->dir_index = 0ULL;
8861 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8862 /* force full log commit if subvolume involved. */
8863 btrfs_set_log_full_commit(root->fs_info, trans);
8865 ret = btrfs_insert_inode_ref(trans, dest,
8866 new_dentry->d_name.name,
8867 new_dentry->d_name.len,
8869 btrfs_ino(new_dir), index);
8873 * this is an ugly little race, but the rename is required
8874 * to make sure that if we crash, the inode is either at the
8875 * old name or the new one. pinning the log transaction lets
8876 * us make sure we don't allow a log commit to come in after
8877 * we unlink the name but before we add the new name back in.
8879 btrfs_pin_log_trans(root);
8882 inode_inc_iversion(old_dir);
8883 inode_inc_iversion(new_dir);
8884 inode_inc_iversion(old_inode);
8885 old_dir->i_ctime = old_dir->i_mtime = ctime;
8886 new_dir->i_ctime = new_dir->i_mtime = ctime;
8887 old_inode->i_ctime = ctime;
8889 if (old_dentry->d_parent != new_dentry->d_parent)
8890 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8892 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8893 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8894 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8895 old_dentry->d_name.name,
8896 old_dentry->d_name.len);
8898 ret = __btrfs_unlink_inode(trans, root, old_dir,
8899 old_dentry->d_inode,
8900 old_dentry->d_name.name,
8901 old_dentry->d_name.len);
8903 ret = btrfs_update_inode(trans, root, old_inode);
8906 btrfs_abort_transaction(trans, root, ret);
8911 inode_inc_iversion(new_inode);
8912 new_inode->i_ctime = CURRENT_TIME;
8913 if (unlikely(btrfs_ino(new_inode) ==
8914 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8915 root_objectid = BTRFS_I(new_inode)->location.objectid;
8916 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8918 new_dentry->d_name.name,
8919 new_dentry->d_name.len);
8920 BUG_ON(new_inode->i_nlink == 0);
8922 ret = btrfs_unlink_inode(trans, dest, new_dir,
8923 new_dentry->d_inode,
8924 new_dentry->d_name.name,
8925 new_dentry->d_name.len);
8927 if (!ret && new_inode->i_nlink == 0)
8928 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8930 btrfs_abort_transaction(trans, root, ret);
8935 ret = btrfs_add_link(trans, new_dir, old_inode,
8936 new_dentry->d_name.name,
8937 new_dentry->d_name.len, 0, index);
8939 btrfs_abort_transaction(trans, root, ret);
8943 if (old_inode->i_nlink == 1)
8944 BTRFS_I(old_inode)->dir_index = index;
8946 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8947 struct dentry *parent = new_dentry->d_parent;
8948 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8949 btrfs_end_log_trans(root);
8952 btrfs_end_transaction(trans, root);
8954 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8955 up_read(&root->fs_info->subvol_sem);
8960 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
8961 struct inode *new_dir, struct dentry *new_dentry,
8964 if (flags & ~RENAME_NOREPLACE)
8967 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
8970 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8972 struct btrfs_delalloc_work *delalloc_work;
8973 struct inode *inode;
8975 delalloc_work = container_of(work, struct btrfs_delalloc_work,
8977 inode = delalloc_work->inode;
8978 if (delalloc_work->wait) {
8979 btrfs_wait_ordered_range(inode, 0, (u64)-1);
8981 filemap_flush(inode->i_mapping);
8982 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8983 &BTRFS_I(inode)->runtime_flags))
8984 filemap_flush(inode->i_mapping);
8987 if (delalloc_work->delay_iput)
8988 btrfs_add_delayed_iput(inode);
8991 complete(&delalloc_work->completion);
8994 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8995 int wait, int delay_iput)
8997 struct btrfs_delalloc_work *work;
8999 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
9003 init_completion(&work->completion);
9004 INIT_LIST_HEAD(&work->list);
9005 work->inode = inode;
9007 work->delay_iput = delay_iput;
9008 WARN_ON_ONCE(!inode);
9009 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9010 btrfs_run_delalloc_work, NULL, NULL);
9015 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9017 wait_for_completion(&work->completion);
9018 kmem_cache_free(btrfs_delalloc_work_cachep, work);
9022 * some fairly slow code that needs optimization. This walks the list
9023 * of all the inodes with pending delalloc and forces them to disk.
9025 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9028 struct btrfs_inode *binode;
9029 struct inode *inode;
9030 struct btrfs_delalloc_work *work, *next;
9031 struct list_head works;
9032 struct list_head splice;
9035 INIT_LIST_HEAD(&works);
9036 INIT_LIST_HEAD(&splice);
9038 mutex_lock(&root->delalloc_mutex);
9039 spin_lock(&root->delalloc_lock);
9040 list_splice_init(&root->delalloc_inodes, &splice);
9041 while (!list_empty(&splice)) {
9042 binode = list_entry(splice.next, struct btrfs_inode,
9045 list_move_tail(&binode->delalloc_inodes,
9046 &root->delalloc_inodes);
9047 inode = igrab(&binode->vfs_inode);
9049 cond_resched_lock(&root->delalloc_lock);
9052 spin_unlock(&root->delalloc_lock);
9054 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
9057 btrfs_add_delayed_iput(inode);
9063 list_add_tail(&work->list, &works);
9064 btrfs_queue_work(root->fs_info->flush_workers,
9067 if (nr != -1 && ret >= nr)
9070 spin_lock(&root->delalloc_lock);
9072 spin_unlock(&root->delalloc_lock);
9075 list_for_each_entry_safe(work, next, &works, list) {
9076 list_del_init(&work->list);
9077 btrfs_wait_and_free_delalloc_work(work);
9080 if (!list_empty_careful(&splice)) {
9081 spin_lock(&root->delalloc_lock);
9082 list_splice_tail(&splice, &root->delalloc_inodes);
9083 spin_unlock(&root->delalloc_lock);
9085 mutex_unlock(&root->delalloc_mutex);
9089 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9093 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
9096 ret = __start_delalloc_inodes(root, delay_iput, -1);
9100 * the filemap_flush will queue IO into the worker threads, but
9101 * we have to make sure the IO is actually started and that
9102 * ordered extents get created before we return
9104 atomic_inc(&root->fs_info->async_submit_draining);
9105 while (atomic_read(&root->fs_info->nr_async_submits) ||
9106 atomic_read(&root->fs_info->async_delalloc_pages)) {
9107 wait_event(root->fs_info->async_submit_wait,
9108 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9109 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
9111 atomic_dec(&root->fs_info->async_submit_draining);
9115 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9118 struct btrfs_root *root;
9119 struct list_head splice;
9122 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9125 INIT_LIST_HEAD(&splice);
9127 mutex_lock(&fs_info->delalloc_root_mutex);
9128 spin_lock(&fs_info->delalloc_root_lock);
9129 list_splice_init(&fs_info->delalloc_roots, &splice);
9130 while (!list_empty(&splice) && nr) {
9131 root = list_first_entry(&splice, struct btrfs_root,
9133 root = btrfs_grab_fs_root(root);
9135 list_move_tail(&root->delalloc_root,
9136 &fs_info->delalloc_roots);
9137 spin_unlock(&fs_info->delalloc_root_lock);
9139 ret = __start_delalloc_inodes(root, delay_iput, nr);
9140 btrfs_put_fs_root(root);
9148 spin_lock(&fs_info->delalloc_root_lock);
9150 spin_unlock(&fs_info->delalloc_root_lock);
9153 atomic_inc(&fs_info->async_submit_draining);
9154 while (atomic_read(&fs_info->nr_async_submits) ||
9155 atomic_read(&fs_info->async_delalloc_pages)) {
9156 wait_event(fs_info->async_submit_wait,
9157 (atomic_read(&fs_info->nr_async_submits) == 0 &&
9158 atomic_read(&fs_info->async_delalloc_pages) == 0));
9160 atomic_dec(&fs_info->async_submit_draining);
9162 if (!list_empty_careful(&splice)) {
9163 spin_lock(&fs_info->delalloc_root_lock);
9164 list_splice_tail(&splice, &fs_info->delalloc_roots);
9165 spin_unlock(&fs_info->delalloc_root_lock);
9167 mutex_unlock(&fs_info->delalloc_root_mutex);
9171 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9172 const char *symname)
9174 struct btrfs_trans_handle *trans;
9175 struct btrfs_root *root = BTRFS_I(dir)->root;
9176 struct btrfs_path *path;
9177 struct btrfs_key key;
9178 struct inode *inode = NULL;
9186 struct btrfs_file_extent_item *ei;
9187 struct extent_buffer *leaf;
9189 name_len = strlen(symname);
9190 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9191 return -ENAMETOOLONG;
9194 * 2 items for inode item and ref
9195 * 2 items for dir items
9196 * 1 item for xattr if selinux is on
9198 trans = btrfs_start_transaction(root, 5);
9200 return PTR_ERR(trans);
9202 err = btrfs_find_free_ino(root, &objectid);
9206 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9207 dentry->d_name.len, btrfs_ino(dir), objectid,
9208 S_IFLNK|S_IRWXUGO, &index);
9209 if (IS_ERR(inode)) {
9210 err = PTR_ERR(inode);
9215 * If the active LSM wants to access the inode during
9216 * d_instantiate it needs these. Smack checks to see
9217 * if the filesystem supports xattrs by looking at the
9220 inode->i_fop = &btrfs_file_operations;
9221 inode->i_op = &btrfs_file_inode_operations;
9222 inode->i_mapping->a_ops = &btrfs_aops;
9223 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
9224 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9226 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9228 goto out_unlock_inode;
9230 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
9232 goto out_unlock_inode;
9234 path = btrfs_alloc_path();
9237 goto out_unlock_inode;
9239 key.objectid = btrfs_ino(inode);
9241 key.type = BTRFS_EXTENT_DATA_KEY;
9242 datasize = btrfs_file_extent_calc_inline_size(name_len);
9243 err = btrfs_insert_empty_item(trans, root, path, &key,
9246 btrfs_free_path(path);
9247 goto out_unlock_inode;
9249 leaf = path->nodes[0];
9250 ei = btrfs_item_ptr(leaf, path->slots[0],
9251 struct btrfs_file_extent_item);
9252 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9253 btrfs_set_file_extent_type(leaf, ei,
9254 BTRFS_FILE_EXTENT_INLINE);
9255 btrfs_set_file_extent_encryption(leaf, ei, 0);
9256 btrfs_set_file_extent_compression(leaf, ei, 0);
9257 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9258 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9260 ptr = btrfs_file_extent_inline_start(ei);
9261 write_extent_buffer(leaf, symname, ptr, name_len);
9262 btrfs_mark_buffer_dirty(leaf);
9263 btrfs_free_path(path);
9265 inode->i_op = &btrfs_symlink_inode_operations;
9266 inode->i_mapping->a_ops = &btrfs_symlink_aops;
9267 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
9268 inode_set_bytes(inode, name_len);
9269 btrfs_i_size_write(inode, name_len);
9270 err = btrfs_update_inode(trans, root, inode);
9273 goto out_unlock_inode;
9276 unlock_new_inode(inode);
9277 d_instantiate(dentry, inode);
9280 btrfs_end_transaction(trans, root);
9282 inode_dec_link_count(inode);
9285 btrfs_btree_balance_dirty(root);
9290 unlock_new_inode(inode);
9294 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9295 u64 start, u64 num_bytes, u64 min_size,
9296 loff_t actual_len, u64 *alloc_hint,
9297 struct btrfs_trans_handle *trans)
9299 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9300 struct extent_map *em;
9301 struct btrfs_root *root = BTRFS_I(inode)->root;
9302 struct btrfs_key ins;
9303 u64 cur_offset = start;
9307 bool own_trans = true;
9311 while (num_bytes > 0) {
9313 trans = btrfs_start_transaction(root, 3);
9314 if (IS_ERR(trans)) {
9315 ret = PTR_ERR(trans);
9320 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
9321 cur_bytes = max(cur_bytes, min_size);
9322 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
9323 *alloc_hint, &ins, 1, 0);
9326 btrfs_end_transaction(trans, root);
9330 ret = insert_reserved_file_extent(trans, inode,
9331 cur_offset, ins.objectid,
9332 ins.offset, ins.offset,
9333 ins.offset, 0, 0, 0,
9334 BTRFS_FILE_EXTENT_PREALLOC);
9336 btrfs_free_reserved_extent(root, ins.objectid,
9338 btrfs_abort_transaction(trans, root, ret);
9340 btrfs_end_transaction(trans, root);
9343 btrfs_drop_extent_cache(inode, cur_offset,
9344 cur_offset + ins.offset -1, 0);
9346 em = alloc_extent_map();
9348 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9349 &BTRFS_I(inode)->runtime_flags);
9353 em->start = cur_offset;
9354 em->orig_start = cur_offset;
9355 em->len = ins.offset;
9356 em->block_start = ins.objectid;
9357 em->block_len = ins.offset;
9358 em->orig_block_len = ins.offset;
9359 em->ram_bytes = ins.offset;
9360 em->bdev = root->fs_info->fs_devices->latest_bdev;
9361 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9362 em->generation = trans->transid;
9365 write_lock(&em_tree->lock);
9366 ret = add_extent_mapping(em_tree, em, 1);
9367 write_unlock(&em_tree->lock);
9370 btrfs_drop_extent_cache(inode, cur_offset,
9371 cur_offset + ins.offset - 1,
9374 free_extent_map(em);
9376 num_bytes -= ins.offset;
9377 cur_offset += ins.offset;
9378 *alloc_hint = ins.objectid + ins.offset;
9380 inode_inc_iversion(inode);
9381 inode->i_ctime = CURRENT_TIME;
9382 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9383 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9384 (actual_len > inode->i_size) &&
9385 (cur_offset > inode->i_size)) {
9386 if (cur_offset > actual_len)
9387 i_size = actual_len;
9389 i_size = cur_offset;
9390 i_size_write(inode, i_size);
9391 btrfs_ordered_update_i_size(inode, i_size, NULL);
9394 ret = btrfs_update_inode(trans, root, inode);
9397 btrfs_abort_transaction(trans, root, ret);
9399 btrfs_end_transaction(trans, root);
9404 btrfs_end_transaction(trans, root);
9409 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9410 u64 start, u64 num_bytes, u64 min_size,
9411 loff_t actual_len, u64 *alloc_hint)
9413 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9414 min_size, actual_len, alloc_hint,
9418 int btrfs_prealloc_file_range_trans(struct inode *inode,
9419 struct btrfs_trans_handle *trans, int mode,
9420 u64 start, u64 num_bytes, u64 min_size,
9421 loff_t actual_len, u64 *alloc_hint)
9423 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9424 min_size, actual_len, alloc_hint, trans);
9427 static int btrfs_set_page_dirty(struct page *page)
9429 return __set_page_dirty_nobuffers(page);
9432 static int btrfs_permission(struct inode *inode, int mask)
9434 struct btrfs_root *root = BTRFS_I(inode)->root;
9435 umode_t mode = inode->i_mode;
9437 if (mask & MAY_WRITE &&
9438 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9439 if (btrfs_root_readonly(root))
9441 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9444 return generic_permission(inode, mask);
9447 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9449 struct btrfs_trans_handle *trans;
9450 struct btrfs_root *root = BTRFS_I(dir)->root;
9451 struct inode *inode = NULL;
9457 * 5 units required for adding orphan entry
9459 trans = btrfs_start_transaction(root, 5);
9461 return PTR_ERR(trans);
9463 ret = btrfs_find_free_ino(root, &objectid);
9467 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9468 btrfs_ino(dir), objectid, mode, &index);
9469 if (IS_ERR(inode)) {
9470 ret = PTR_ERR(inode);
9475 inode->i_fop = &btrfs_file_operations;
9476 inode->i_op = &btrfs_file_inode_operations;
9478 inode->i_mapping->a_ops = &btrfs_aops;
9479 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
9480 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9482 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9486 ret = btrfs_update_inode(trans, root, inode);
9489 ret = btrfs_orphan_add(trans, inode);
9494 * We set number of links to 0 in btrfs_new_inode(), and here we set
9495 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9498 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9500 set_nlink(inode, 1);
9501 unlock_new_inode(inode);
9502 d_tmpfile(dentry, inode);
9503 mark_inode_dirty(inode);
9506 btrfs_end_transaction(trans, root);
9509 btrfs_balance_delayed_items(root);
9510 btrfs_btree_balance_dirty(root);
9514 unlock_new_inode(inode);
9519 /* Inspired by filemap_check_errors() */
9520 int btrfs_inode_check_errors(struct inode *inode)
9524 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
9525 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
9527 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
9528 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
9534 static const struct inode_operations btrfs_dir_inode_operations = {
9535 .getattr = btrfs_getattr,
9536 .lookup = btrfs_lookup,
9537 .create = btrfs_create,
9538 .unlink = btrfs_unlink,
9540 .mkdir = btrfs_mkdir,
9541 .rmdir = btrfs_rmdir,
9542 .rename2 = btrfs_rename2,
9543 .symlink = btrfs_symlink,
9544 .setattr = btrfs_setattr,
9545 .mknod = btrfs_mknod,
9546 .setxattr = btrfs_setxattr,
9547 .getxattr = btrfs_getxattr,
9548 .listxattr = btrfs_listxattr,
9549 .removexattr = btrfs_removexattr,
9550 .permission = btrfs_permission,
9551 .get_acl = btrfs_get_acl,
9552 .set_acl = btrfs_set_acl,
9553 .update_time = btrfs_update_time,
9554 .tmpfile = btrfs_tmpfile,
9556 static const struct inode_operations btrfs_dir_ro_inode_operations = {
9557 .lookup = btrfs_lookup,
9558 .permission = btrfs_permission,
9559 .get_acl = btrfs_get_acl,
9560 .set_acl = btrfs_set_acl,
9561 .update_time = btrfs_update_time,
9564 static const struct file_operations btrfs_dir_file_operations = {
9565 .llseek = generic_file_llseek,
9566 .read = generic_read_dir,
9567 .iterate = btrfs_real_readdir,
9568 .unlocked_ioctl = btrfs_ioctl,
9569 #ifdef CONFIG_COMPAT
9570 .compat_ioctl = btrfs_ioctl,
9572 .release = btrfs_release_file,
9573 .fsync = btrfs_sync_file,
9576 static struct extent_io_ops btrfs_extent_io_ops = {
9577 .fill_delalloc = run_delalloc_range,
9578 .submit_bio_hook = btrfs_submit_bio_hook,
9579 .merge_bio_hook = btrfs_merge_bio_hook,
9580 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
9581 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
9582 .writepage_start_hook = btrfs_writepage_start_hook,
9583 .set_bit_hook = btrfs_set_bit_hook,
9584 .clear_bit_hook = btrfs_clear_bit_hook,
9585 .merge_extent_hook = btrfs_merge_extent_hook,
9586 .split_extent_hook = btrfs_split_extent_hook,
9590 * btrfs doesn't support the bmap operation because swapfiles
9591 * use bmap to make a mapping of extents in the file. They assume
9592 * these extents won't change over the life of the file and they
9593 * use the bmap result to do IO directly to the drive.
9595 * the btrfs bmap call would return logical addresses that aren't
9596 * suitable for IO and they also will change frequently as COW
9597 * operations happen. So, swapfile + btrfs == corruption.
9599 * For now we're avoiding this by dropping bmap.
9601 static const struct address_space_operations btrfs_aops = {
9602 .readpage = btrfs_readpage,
9603 .writepage = btrfs_writepage,
9604 .writepages = btrfs_writepages,
9605 .readpages = btrfs_readpages,
9606 .direct_IO = btrfs_direct_IO,
9607 .invalidatepage = btrfs_invalidatepage,
9608 .releasepage = btrfs_releasepage,
9609 .set_page_dirty = btrfs_set_page_dirty,
9610 .error_remove_page = generic_error_remove_page,
9613 static const struct address_space_operations btrfs_symlink_aops = {
9614 .readpage = btrfs_readpage,
9615 .writepage = btrfs_writepage,
9616 .invalidatepage = btrfs_invalidatepage,
9617 .releasepage = btrfs_releasepage,
9620 static const struct inode_operations btrfs_file_inode_operations = {
9621 .getattr = btrfs_getattr,
9622 .setattr = btrfs_setattr,
9623 .setxattr = btrfs_setxattr,
9624 .getxattr = btrfs_getxattr,
9625 .listxattr = btrfs_listxattr,
9626 .removexattr = btrfs_removexattr,
9627 .permission = btrfs_permission,
9628 .fiemap = btrfs_fiemap,
9629 .get_acl = btrfs_get_acl,
9630 .set_acl = btrfs_set_acl,
9631 .update_time = btrfs_update_time,
9633 static const struct inode_operations btrfs_special_inode_operations = {
9634 .getattr = btrfs_getattr,
9635 .setattr = btrfs_setattr,
9636 .permission = btrfs_permission,
9637 .setxattr = btrfs_setxattr,
9638 .getxattr = btrfs_getxattr,
9639 .listxattr = btrfs_listxattr,
9640 .removexattr = btrfs_removexattr,
9641 .get_acl = btrfs_get_acl,
9642 .set_acl = btrfs_set_acl,
9643 .update_time = btrfs_update_time,
9645 static const struct inode_operations btrfs_symlink_inode_operations = {
9646 .readlink = generic_readlink,
9647 .follow_link = page_follow_link_light,
9648 .put_link = page_put_link,
9649 .getattr = btrfs_getattr,
9650 .setattr = btrfs_setattr,
9651 .permission = btrfs_permission,
9652 .setxattr = btrfs_setxattr,
9653 .getxattr = btrfs_getxattr,
9654 .listxattr = btrfs_listxattr,
9655 .removexattr = btrfs_removexattr,
9656 .update_time = btrfs_update_time,
9659 const struct dentry_operations btrfs_dentry_operations = {
9660 .d_delete = btrfs_dentry_delete,
9661 .d_release = btrfs_dentry_release,