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 long *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,
1648 unsigned long *bits)
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_info(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_crit(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_timespec *tspec;
3494 struct btrfs_root *root = BTRFS_I(inode)->root;
3495 struct btrfs_key location;
3500 bool filled = false;
3501 int first_xattr_slot;
3503 ret = btrfs_fill_inode(inode, &rdev);
3507 path = btrfs_alloc_path();
3511 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3513 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3517 leaf = path->nodes[0];
3522 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3523 struct btrfs_inode_item);
3524 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3525 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3526 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3527 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3528 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3530 tspec = btrfs_inode_atime(inode_item);
3531 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3532 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3534 tspec = btrfs_inode_mtime(inode_item);
3535 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3536 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3538 tspec = btrfs_inode_ctime(inode_item);
3539 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3540 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3542 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3543 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3544 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3547 * If we were modified in the current generation and evicted from memory
3548 * and then re-read we need to do a full sync since we don't have any
3549 * idea about which extents were modified before we were evicted from
3552 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3553 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3554 &BTRFS_I(inode)->runtime_flags);
3556 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3557 inode->i_generation = BTRFS_I(inode)->generation;
3559 rdev = btrfs_inode_rdev(leaf, inode_item);
3561 BTRFS_I(inode)->index_cnt = (u64)-1;
3562 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3566 if (inode->i_nlink != 1 ||
3567 path->slots[0] >= btrfs_header_nritems(leaf))
3570 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3571 if (location.objectid != btrfs_ino(inode))
3574 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3575 if (location.type == BTRFS_INODE_REF_KEY) {
3576 struct btrfs_inode_ref *ref;
3578 ref = (struct btrfs_inode_ref *)ptr;
3579 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3580 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3581 struct btrfs_inode_extref *extref;
3583 extref = (struct btrfs_inode_extref *)ptr;
3584 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3589 * try to precache a NULL acl entry for files that don't have
3590 * any xattrs or acls
3592 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3593 btrfs_ino(inode), &first_xattr_slot);
3594 if (first_xattr_slot != -1) {
3595 path->slots[0] = first_xattr_slot;
3596 ret = btrfs_load_inode_props(inode, path);
3598 btrfs_err(root->fs_info,
3599 "error loading props for ino %llu (root %llu): %d",
3601 root->root_key.objectid, ret);
3603 btrfs_free_path(path);
3606 cache_no_acl(inode);
3608 switch (inode->i_mode & S_IFMT) {
3610 inode->i_mapping->a_ops = &btrfs_aops;
3611 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3612 inode->i_fop = &btrfs_file_operations;
3613 inode->i_op = &btrfs_file_inode_operations;
3616 inode->i_fop = &btrfs_dir_file_operations;
3617 if (root == root->fs_info->tree_root)
3618 inode->i_op = &btrfs_dir_ro_inode_operations;
3620 inode->i_op = &btrfs_dir_inode_operations;
3623 inode->i_op = &btrfs_symlink_inode_operations;
3624 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3627 inode->i_op = &btrfs_special_inode_operations;
3628 init_special_inode(inode, inode->i_mode, rdev);
3632 btrfs_update_iflags(inode);
3636 btrfs_free_path(path);
3637 make_bad_inode(inode);
3641 * given a leaf and an inode, copy the inode fields into the leaf
3643 static void fill_inode_item(struct btrfs_trans_handle *trans,
3644 struct extent_buffer *leaf,
3645 struct btrfs_inode_item *item,
3646 struct inode *inode)
3648 struct btrfs_map_token token;
3650 btrfs_init_map_token(&token);
3652 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3653 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3654 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3656 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3657 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3659 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3660 inode->i_atime.tv_sec, &token);
3661 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3662 inode->i_atime.tv_nsec, &token);
3664 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3665 inode->i_mtime.tv_sec, &token);
3666 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3667 inode->i_mtime.tv_nsec, &token);
3669 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3670 inode->i_ctime.tv_sec, &token);
3671 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3672 inode->i_ctime.tv_nsec, &token);
3674 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3676 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3678 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3679 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3680 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3681 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3682 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3686 * copy everything in the in-memory inode into the btree.
3688 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3689 struct btrfs_root *root, struct inode *inode)
3691 struct btrfs_inode_item *inode_item;
3692 struct btrfs_path *path;
3693 struct extent_buffer *leaf;
3696 path = btrfs_alloc_path();
3700 path->leave_spinning = 1;
3701 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3709 leaf = path->nodes[0];
3710 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3711 struct btrfs_inode_item);
3713 fill_inode_item(trans, leaf, inode_item, inode);
3714 btrfs_mark_buffer_dirty(leaf);
3715 btrfs_set_inode_last_trans(trans, inode);
3718 btrfs_free_path(path);
3723 * copy everything in the in-memory inode into the btree.
3725 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3726 struct btrfs_root *root, struct inode *inode)
3731 * If the inode is a free space inode, we can deadlock during commit
3732 * if we put it into the delayed code.
3734 * The data relocation inode should also be directly updated
3737 if (!btrfs_is_free_space_inode(inode)
3738 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3739 && !root->fs_info->log_root_recovering) {
3740 btrfs_update_root_times(trans, root);
3742 ret = btrfs_delayed_update_inode(trans, root, inode);
3744 btrfs_set_inode_last_trans(trans, inode);
3748 return btrfs_update_inode_item(trans, root, inode);
3751 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3752 struct btrfs_root *root,
3753 struct inode *inode)
3757 ret = btrfs_update_inode(trans, root, inode);
3759 return btrfs_update_inode_item(trans, root, inode);
3764 * unlink helper that gets used here in inode.c and in the tree logging
3765 * recovery code. It remove a link in a directory with a given name, and
3766 * also drops the back refs in the inode to the directory
3768 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3769 struct btrfs_root *root,
3770 struct inode *dir, struct inode *inode,
3771 const char *name, int name_len)
3773 struct btrfs_path *path;
3775 struct extent_buffer *leaf;
3776 struct btrfs_dir_item *di;
3777 struct btrfs_key key;
3779 u64 ino = btrfs_ino(inode);
3780 u64 dir_ino = btrfs_ino(dir);
3782 path = btrfs_alloc_path();
3788 path->leave_spinning = 1;
3789 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3790 name, name_len, -1);
3799 leaf = path->nodes[0];
3800 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3801 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3804 btrfs_release_path(path);
3807 * If we don't have dir index, we have to get it by looking up
3808 * the inode ref, since we get the inode ref, remove it directly,
3809 * it is unnecessary to do delayed deletion.
3811 * But if we have dir index, needn't search inode ref to get it.
3812 * Since the inode ref is close to the inode item, it is better
3813 * that we delay to delete it, and just do this deletion when
3814 * we update the inode item.
3816 if (BTRFS_I(inode)->dir_index) {
3817 ret = btrfs_delayed_delete_inode_ref(inode);
3819 index = BTRFS_I(inode)->dir_index;
3824 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3827 btrfs_info(root->fs_info,
3828 "failed to delete reference to %.*s, inode %llu parent %llu",
3829 name_len, name, ino, dir_ino);
3830 btrfs_abort_transaction(trans, root, ret);
3834 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3836 btrfs_abort_transaction(trans, root, ret);
3840 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3842 if (ret != 0 && ret != -ENOENT) {
3843 btrfs_abort_transaction(trans, root, ret);
3847 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3852 btrfs_abort_transaction(trans, root, ret);
3854 btrfs_free_path(path);
3858 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3859 inode_inc_iversion(inode);
3860 inode_inc_iversion(dir);
3861 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3862 ret = btrfs_update_inode(trans, root, dir);
3867 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3868 struct btrfs_root *root,
3869 struct inode *dir, struct inode *inode,
3870 const char *name, int name_len)
3873 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3876 ret = btrfs_update_inode(trans, root, inode);
3882 * helper to start transaction for unlink and rmdir.
3884 * unlink and rmdir are special in btrfs, they do not always free space, so
3885 * if we cannot make our reservations the normal way try and see if there is
3886 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3887 * allow the unlink to occur.
3889 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3891 struct btrfs_trans_handle *trans;
3892 struct btrfs_root *root = BTRFS_I(dir)->root;
3896 * 1 for the possible orphan item
3897 * 1 for the dir item
3898 * 1 for the dir index
3899 * 1 for the inode ref
3902 trans = btrfs_start_transaction(root, 5);
3903 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3906 if (PTR_ERR(trans) == -ENOSPC) {
3907 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3909 trans = btrfs_start_transaction(root, 0);
3912 ret = btrfs_cond_migrate_bytes(root->fs_info,
3913 &root->fs_info->trans_block_rsv,
3916 btrfs_end_transaction(trans, root);
3917 return ERR_PTR(ret);
3919 trans->block_rsv = &root->fs_info->trans_block_rsv;
3920 trans->bytes_reserved = num_bytes;
3925 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3927 struct btrfs_root *root = BTRFS_I(dir)->root;
3928 struct btrfs_trans_handle *trans;
3929 struct inode *inode = dentry->d_inode;
3932 trans = __unlink_start_trans(dir);
3934 return PTR_ERR(trans);
3936 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3938 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3939 dentry->d_name.name, dentry->d_name.len);
3943 if (inode->i_nlink == 0) {
3944 ret = btrfs_orphan_add(trans, inode);
3950 btrfs_end_transaction(trans, root);
3951 btrfs_btree_balance_dirty(root);
3955 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3956 struct btrfs_root *root,
3957 struct inode *dir, u64 objectid,
3958 const char *name, int name_len)
3960 struct btrfs_path *path;
3961 struct extent_buffer *leaf;
3962 struct btrfs_dir_item *di;
3963 struct btrfs_key key;
3966 u64 dir_ino = btrfs_ino(dir);
3968 path = btrfs_alloc_path();
3972 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3973 name, name_len, -1);
3974 if (IS_ERR_OR_NULL(di)) {
3982 leaf = path->nodes[0];
3983 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3984 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3985 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3987 btrfs_abort_transaction(trans, root, ret);
3990 btrfs_release_path(path);
3992 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3993 objectid, root->root_key.objectid,
3994 dir_ino, &index, name, name_len);
3996 if (ret != -ENOENT) {
3997 btrfs_abort_transaction(trans, root, ret);
4000 di = btrfs_search_dir_index_item(root, path, dir_ino,
4002 if (IS_ERR_OR_NULL(di)) {
4007 btrfs_abort_transaction(trans, root, ret);
4011 leaf = path->nodes[0];
4012 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4013 btrfs_release_path(path);
4016 btrfs_release_path(path);
4018 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4020 btrfs_abort_transaction(trans, root, ret);
4024 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4025 inode_inc_iversion(dir);
4026 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4027 ret = btrfs_update_inode_fallback(trans, root, dir);
4029 btrfs_abort_transaction(trans, root, ret);
4031 btrfs_free_path(path);
4035 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4037 struct inode *inode = dentry->d_inode;
4039 struct btrfs_root *root = BTRFS_I(dir)->root;
4040 struct btrfs_trans_handle *trans;
4042 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4044 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4047 trans = __unlink_start_trans(dir);
4049 return PTR_ERR(trans);
4051 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4052 err = btrfs_unlink_subvol(trans, root, dir,
4053 BTRFS_I(inode)->location.objectid,
4054 dentry->d_name.name,
4055 dentry->d_name.len);
4059 err = btrfs_orphan_add(trans, inode);
4063 /* now the directory is empty */
4064 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
4065 dentry->d_name.name, dentry->d_name.len);
4067 btrfs_i_size_write(inode, 0);
4069 btrfs_end_transaction(trans, root);
4070 btrfs_btree_balance_dirty(root);
4076 * this can truncate away extent items, csum items and directory items.
4077 * It starts at a high offset and removes keys until it can't find
4078 * any higher than new_size
4080 * csum items that cross the new i_size are truncated to the new size
4083 * min_type is the minimum key type to truncate down to. If set to 0, this
4084 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4086 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4087 struct btrfs_root *root,
4088 struct inode *inode,
4089 u64 new_size, u32 min_type)
4091 struct btrfs_path *path;
4092 struct extent_buffer *leaf;
4093 struct btrfs_file_extent_item *fi;
4094 struct btrfs_key key;
4095 struct btrfs_key found_key;
4096 u64 extent_start = 0;
4097 u64 extent_num_bytes = 0;
4098 u64 extent_offset = 0;
4100 u64 last_size = (u64)-1;
4101 u32 found_type = (u8)-1;
4104 int pending_del_nr = 0;
4105 int pending_del_slot = 0;
4106 int extent_type = -1;
4109 u64 ino = btrfs_ino(inode);
4111 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4113 path = btrfs_alloc_path();
4119 * We want to drop from the next block forward in case this new size is
4120 * not block aligned since we will be keeping the last block of the
4121 * extent just the way it is.
4123 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4124 root == root->fs_info->tree_root)
4125 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4126 root->sectorsize), (u64)-1, 0);
4129 * This function is also used to drop the items in the log tree before
4130 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4131 * it is used to drop the loged items. So we shouldn't kill the delayed
4134 if (min_type == 0 && root == BTRFS_I(inode)->root)
4135 btrfs_kill_delayed_inode_items(inode);
4138 key.offset = (u64)-1;
4142 path->leave_spinning = 1;
4143 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4150 /* there are no items in the tree for us to truncate, we're
4153 if (path->slots[0] == 0)
4160 leaf = path->nodes[0];
4161 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4162 found_type = found_key.type;
4164 if (found_key.objectid != ino)
4167 if (found_type < min_type)
4170 item_end = found_key.offset;
4171 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4172 fi = btrfs_item_ptr(leaf, path->slots[0],
4173 struct btrfs_file_extent_item);
4174 extent_type = btrfs_file_extent_type(leaf, fi);
4175 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4177 btrfs_file_extent_num_bytes(leaf, fi);
4178 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4179 item_end += btrfs_file_extent_inline_len(leaf,
4180 path->slots[0], fi);
4184 if (found_type > min_type) {
4187 if (item_end < new_size)
4189 if (found_key.offset >= new_size)
4195 /* FIXME, shrink the extent if the ref count is only 1 */
4196 if (found_type != BTRFS_EXTENT_DATA_KEY)
4200 last_size = found_key.offset;
4202 last_size = new_size;
4204 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4206 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4208 u64 orig_num_bytes =
4209 btrfs_file_extent_num_bytes(leaf, fi);
4210 extent_num_bytes = ALIGN(new_size -
4213 btrfs_set_file_extent_num_bytes(leaf, fi,
4215 num_dec = (orig_num_bytes -
4217 if (test_bit(BTRFS_ROOT_REF_COWS,
4220 inode_sub_bytes(inode, num_dec);
4221 btrfs_mark_buffer_dirty(leaf);
4224 btrfs_file_extent_disk_num_bytes(leaf,
4226 extent_offset = found_key.offset -
4227 btrfs_file_extent_offset(leaf, fi);
4229 /* FIXME blocksize != 4096 */
4230 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4231 if (extent_start != 0) {
4233 if (test_bit(BTRFS_ROOT_REF_COWS,
4235 inode_sub_bytes(inode, num_dec);
4238 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4240 * we can't truncate inline items that have had
4244 btrfs_file_extent_compression(leaf, fi) == 0 &&
4245 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4246 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4247 u32 size = new_size - found_key.offset;
4249 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4250 inode_sub_bytes(inode, item_end + 1 -
4254 * update the ram bytes to properly reflect
4255 * the new size of our item
4257 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4259 btrfs_file_extent_calc_inline_size(size);
4260 btrfs_truncate_item(root, path, size, 1);
4261 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4263 inode_sub_bytes(inode, item_end + 1 -
4269 if (!pending_del_nr) {
4270 /* no pending yet, add ourselves */
4271 pending_del_slot = path->slots[0];
4273 } else if (pending_del_nr &&
4274 path->slots[0] + 1 == pending_del_slot) {
4275 /* hop on the pending chunk */
4277 pending_del_slot = path->slots[0];
4285 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4286 root == root->fs_info->tree_root)) {
4287 btrfs_set_path_blocking(path);
4288 ret = btrfs_free_extent(trans, root, extent_start,
4289 extent_num_bytes, 0,
4290 btrfs_header_owner(leaf),
4291 ino, extent_offset, 0);
4295 if (found_type == BTRFS_INODE_ITEM_KEY)
4298 if (path->slots[0] == 0 ||
4299 path->slots[0] != pending_del_slot) {
4300 if (pending_del_nr) {
4301 ret = btrfs_del_items(trans, root, path,
4305 btrfs_abort_transaction(trans,
4311 btrfs_release_path(path);
4318 if (pending_del_nr) {
4319 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4322 btrfs_abort_transaction(trans, root, ret);
4325 if (last_size != (u64)-1 &&
4326 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4327 btrfs_ordered_update_i_size(inode, last_size, NULL);
4328 btrfs_free_path(path);
4333 * btrfs_truncate_page - read, zero a chunk and write a page
4334 * @inode - inode that we're zeroing
4335 * @from - the offset to start zeroing
4336 * @len - the length to zero, 0 to zero the entire range respective to the
4338 * @front - zero up to the offset instead of from the offset on
4340 * This will find the page for the "from" offset and cow the page and zero the
4341 * part we want to zero. This is used with truncate and hole punching.
4343 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4346 struct address_space *mapping = inode->i_mapping;
4347 struct btrfs_root *root = BTRFS_I(inode)->root;
4348 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4349 struct btrfs_ordered_extent *ordered;
4350 struct extent_state *cached_state = NULL;
4352 u32 blocksize = root->sectorsize;
4353 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4354 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4356 gfp_t mask = btrfs_alloc_write_mask(mapping);
4361 if ((offset & (blocksize - 1)) == 0 &&
4362 (!len || ((len & (blocksize - 1)) == 0)))
4364 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4369 page = find_or_create_page(mapping, index, mask);
4371 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4376 page_start = page_offset(page);
4377 page_end = page_start + PAGE_CACHE_SIZE - 1;
4379 if (!PageUptodate(page)) {
4380 ret = btrfs_readpage(NULL, page);
4382 if (page->mapping != mapping) {
4384 page_cache_release(page);
4387 if (!PageUptodate(page)) {
4392 wait_on_page_writeback(page);
4394 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4395 set_page_extent_mapped(page);
4397 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4399 unlock_extent_cached(io_tree, page_start, page_end,
4400 &cached_state, GFP_NOFS);
4402 page_cache_release(page);
4403 btrfs_start_ordered_extent(inode, ordered, 1);
4404 btrfs_put_ordered_extent(ordered);
4408 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4409 EXTENT_DIRTY | EXTENT_DELALLOC |
4410 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4411 0, 0, &cached_state, GFP_NOFS);
4413 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4416 unlock_extent_cached(io_tree, page_start, page_end,
4417 &cached_state, GFP_NOFS);
4421 if (offset != PAGE_CACHE_SIZE) {
4423 len = PAGE_CACHE_SIZE - offset;
4426 memset(kaddr, 0, offset);
4428 memset(kaddr + offset, 0, len);
4429 flush_dcache_page(page);
4432 ClearPageChecked(page);
4433 set_page_dirty(page);
4434 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4439 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4441 page_cache_release(page);
4446 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4447 u64 offset, u64 len)
4449 struct btrfs_trans_handle *trans;
4453 * Still need to make sure the inode looks like it's been updated so
4454 * that any holes get logged if we fsync.
4456 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4457 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4458 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4459 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4464 * 1 - for the one we're dropping
4465 * 1 - for the one we're adding
4466 * 1 - for updating the inode.
4468 trans = btrfs_start_transaction(root, 3);
4470 return PTR_ERR(trans);
4472 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4474 btrfs_abort_transaction(trans, root, ret);
4475 btrfs_end_transaction(trans, root);
4479 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4480 0, 0, len, 0, len, 0, 0, 0);
4482 btrfs_abort_transaction(trans, root, ret);
4484 btrfs_update_inode(trans, root, inode);
4485 btrfs_end_transaction(trans, root);
4490 * This function puts in dummy file extents for the area we're creating a hole
4491 * for. So if we are truncating this file to a larger size we need to insert
4492 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4493 * the range between oldsize and size
4495 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4497 struct btrfs_root *root = BTRFS_I(inode)->root;
4498 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4499 struct extent_map *em = NULL;
4500 struct extent_state *cached_state = NULL;
4501 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4502 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4503 u64 block_end = ALIGN(size, root->sectorsize);
4510 * If our size started in the middle of a page we need to zero out the
4511 * rest of the page before we expand the i_size, otherwise we could
4512 * expose stale data.
4514 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4518 if (size <= hole_start)
4522 struct btrfs_ordered_extent *ordered;
4524 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4526 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4527 block_end - hole_start);
4530 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4531 &cached_state, GFP_NOFS);
4532 btrfs_start_ordered_extent(inode, ordered, 1);
4533 btrfs_put_ordered_extent(ordered);
4536 cur_offset = hole_start;
4538 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4539 block_end - cur_offset, 0);
4545 last_byte = min(extent_map_end(em), block_end);
4546 last_byte = ALIGN(last_byte , root->sectorsize);
4547 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4548 struct extent_map *hole_em;
4549 hole_size = last_byte - cur_offset;
4551 err = maybe_insert_hole(root, inode, cur_offset,
4555 btrfs_drop_extent_cache(inode, cur_offset,
4556 cur_offset + hole_size - 1, 0);
4557 hole_em = alloc_extent_map();
4559 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4560 &BTRFS_I(inode)->runtime_flags);
4563 hole_em->start = cur_offset;
4564 hole_em->len = hole_size;
4565 hole_em->orig_start = cur_offset;
4567 hole_em->block_start = EXTENT_MAP_HOLE;
4568 hole_em->block_len = 0;
4569 hole_em->orig_block_len = 0;
4570 hole_em->ram_bytes = hole_size;
4571 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4572 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4573 hole_em->generation = root->fs_info->generation;
4576 write_lock(&em_tree->lock);
4577 err = add_extent_mapping(em_tree, hole_em, 1);
4578 write_unlock(&em_tree->lock);
4581 btrfs_drop_extent_cache(inode, cur_offset,
4585 free_extent_map(hole_em);
4588 free_extent_map(em);
4590 cur_offset = last_byte;
4591 if (cur_offset >= block_end)
4594 free_extent_map(em);
4595 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4600 static int wait_snapshoting_atomic_t(atomic_t *a)
4606 static void wait_for_snapshot_creation(struct btrfs_root *root)
4611 ret = btrfs_start_write_no_snapshoting(root);
4614 wait_on_atomic_t(&root->will_be_snapshoted,
4615 wait_snapshoting_atomic_t,
4616 TASK_UNINTERRUPTIBLE);
4620 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4622 struct btrfs_root *root = BTRFS_I(inode)->root;
4623 struct btrfs_trans_handle *trans;
4624 loff_t oldsize = i_size_read(inode);
4625 loff_t newsize = attr->ia_size;
4626 int mask = attr->ia_valid;
4630 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4631 * special case where we need to update the times despite not having
4632 * these flags set. For all other operations the VFS set these flags
4633 * explicitly if it wants a timestamp update.
4635 if (newsize != oldsize) {
4636 inode_inc_iversion(inode);
4637 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4638 inode->i_ctime = inode->i_mtime =
4639 current_fs_time(inode->i_sb);
4642 if (newsize > oldsize) {
4643 truncate_pagecache(inode, newsize);
4645 * Don't do an expanding truncate while snapshoting is ongoing.
4646 * This is to ensure the snapshot captures a fully consistent
4647 * state of this file - if the snapshot captures this expanding
4648 * truncation, it must capture all writes that happened before
4651 wait_for_snapshot_creation(root);
4652 ret = btrfs_cont_expand(inode, oldsize, newsize);
4654 btrfs_end_write_no_snapshoting(root);
4658 trans = btrfs_start_transaction(root, 1);
4659 if (IS_ERR(trans)) {
4660 btrfs_end_write_no_snapshoting(root);
4661 return PTR_ERR(trans);
4664 i_size_write(inode, newsize);
4665 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4666 ret = btrfs_update_inode(trans, root, inode);
4667 btrfs_end_write_no_snapshoting(root);
4668 btrfs_end_transaction(trans, root);
4672 * We're truncating a file that used to have good data down to
4673 * zero. Make sure it gets into the ordered flush list so that
4674 * any new writes get down to disk quickly.
4677 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4678 &BTRFS_I(inode)->runtime_flags);
4681 * 1 for the orphan item we're going to add
4682 * 1 for the orphan item deletion.
4684 trans = btrfs_start_transaction(root, 2);
4686 return PTR_ERR(trans);
4689 * We need to do this in case we fail at _any_ point during the
4690 * actual truncate. Once we do the truncate_setsize we could
4691 * invalidate pages which forces any outstanding ordered io to
4692 * be instantly completed which will give us extents that need
4693 * to be truncated. If we fail to get an orphan inode down we
4694 * could have left over extents that were never meant to live,
4695 * so we need to garuntee from this point on that everything
4696 * will be consistent.
4698 ret = btrfs_orphan_add(trans, inode);
4699 btrfs_end_transaction(trans, root);
4703 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4704 truncate_setsize(inode, newsize);
4706 /* Disable nonlocked read DIO to avoid the end less truncate */
4707 btrfs_inode_block_unlocked_dio(inode);
4708 inode_dio_wait(inode);
4709 btrfs_inode_resume_unlocked_dio(inode);
4711 ret = btrfs_truncate(inode);
4712 if (ret && inode->i_nlink) {
4716 * failed to truncate, disk_i_size is only adjusted down
4717 * as we remove extents, so it should represent the true
4718 * size of the inode, so reset the in memory size and
4719 * delete our orphan entry.
4721 trans = btrfs_join_transaction(root);
4722 if (IS_ERR(trans)) {
4723 btrfs_orphan_del(NULL, inode);
4726 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4727 err = btrfs_orphan_del(trans, inode);
4729 btrfs_abort_transaction(trans, root, err);
4730 btrfs_end_transaction(trans, root);
4737 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4739 struct inode *inode = dentry->d_inode;
4740 struct btrfs_root *root = BTRFS_I(inode)->root;
4743 if (btrfs_root_readonly(root))
4746 err = inode_change_ok(inode, attr);
4750 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4751 err = btrfs_setsize(inode, attr);
4756 if (attr->ia_valid) {
4757 setattr_copy(inode, attr);
4758 inode_inc_iversion(inode);
4759 err = btrfs_dirty_inode(inode);
4761 if (!err && attr->ia_valid & ATTR_MODE)
4762 err = posix_acl_chmod(inode, inode->i_mode);
4769 * While truncating the inode pages during eviction, we get the VFS calling
4770 * btrfs_invalidatepage() against each page of the inode. This is slow because
4771 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4772 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4773 * extent_state structures over and over, wasting lots of time.
4775 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4776 * those expensive operations on a per page basis and do only the ordered io
4777 * finishing, while we release here the extent_map and extent_state structures,
4778 * without the excessive merging and splitting.
4780 static void evict_inode_truncate_pages(struct inode *inode)
4782 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4783 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4784 struct rb_node *node;
4786 ASSERT(inode->i_state & I_FREEING);
4787 truncate_inode_pages_final(&inode->i_data);
4789 write_lock(&map_tree->lock);
4790 while (!RB_EMPTY_ROOT(&map_tree->map)) {
4791 struct extent_map *em;
4793 node = rb_first(&map_tree->map);
4794 em = rb_entry(node, struct extent_map, rb_node);
4795 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4796 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4797 remove_extent_mapping(map_tree, em);
4798 free_extent_map(em);
4799 if (need_resched()) {
4800 write_unlock(&map_tree->lock);
4802 write_lock(&map_tree->lock);
4805 write_unlock(&map_tree->lock);
4807 spin_lock(&io_tree->lock);
4808 while (!RB_EMPTY_ROOT(&io_tree->state)) {
4809 struct extent_state *state;
4810 struct extent_state *cached_state = NULL;
4812 node = rb_first(&io_tree->state);
4813 state = rb_entry(node, struct extent_state, rb_node);
4814 atomic_inc(&state->refs);
4815 spin_unlock(&io_tree->lock);
4817 lock_extent_bits(io_tree, state->start, state->end,
4819 clear_extent_bit(io_tree, state->start, state->end,
4820 EXTENT_LOCKED | EXTENT_DIRTY |
4821 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
4822 EXTENT_DEFRAG, 1, 1,
4823 &cached_state, GFP_NOFS);
4824 free_extent_state(state);
4827 spin_lock(&io_tree->lock);
4829 spin_unlock(&io_tree->lock);
4832 void btrfs_evict_inode(struct inode *inode)
4834 struct btrfs_trans_handle *trans;
4835 struct btrfs_root *root = BTRFS_I(inode)->root;
4836 struct btrfs_block_rsv *rsv, *global_rsv;
4837 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4840 trace_btrfs_inode_evict(inode);
4842 evict_inode_truncate_pages(inode);
4844 if (inode->i_nlink &&
4845 ((btrfs_root_refs(&root->root_item) != 0 &&
4846 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4847 btrfs_is_free_space_inode(inode)))
4850 if (is_bad_inode(inode)) {
4851 btrfs_orphan_del(NULL, inode);
4854 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4855 btrfs_wait_ordered_range(inode, 0, (u64)-1);
4857 btrfs_free_io_failure_record(inode, 0, (u64)-1);
4859 if (root->fs_info->log_root_recovering) {
4860 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4861 &BTRFS_I(inode)->runtime_flags));
4865 if (inode->i_nlink > 0) {
4866 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4867 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
4871 ret = btrfs_commit_inode_delayed_inode(inode);
4873 btrfs_orphan_del(NULL, inode);
4877 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4879 btrfs_orphan_del(NULL, inode);
4882 rsv->size = min_size;
4884 global_rsv = &root->fs_info->global_block_rsv;
4886 btrfs_i_size_write(inode, 0);
4889 * This is a bit simpler than btrfs_truncate since we've already
4890 * reserved our space for our orphan item in the unlink, so we just
4891 * need to reserve some slack space in case we add bytes and update
4892 * inode item when doing the truncate.
4895 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4896 BTRFS_RESERVE_FLUSH_LIMIT);
4899 * Try and steal from the global reserve since we will
4900 * likely not use this space anyway, we want to try as
4901 * hard as possible to get this to work.
4904 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4907 btrfs_warn(root->fs_info,
4908 "Could not get space for a delete, will truncate on mount %d",
4910 btrfs_orphan_del(NULL, inode);
4911 btrfs_free_block_rsv(root, rsv);
4915 trans = btrfs_join_transaction(root);
4916 if (IS_ERR(trans)) {
4917 btrfs_orphan_del(NULL, inode);
4918 btrfs_free_block_rsv(root, rsv);
4922 trans->block_rsv = rsv;
4924 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4928 trans->block_rsv = &root->fs_info->trans_block_rsv;
4929 btrfs_end_transaction(trans, root);
4931 btrfs_btree_balance_dirty(root);
4934 btrfs_free_block_rsv(root, rsv);
4937 * Errors here aren't a big deal, it just means we leave orphan items
4938 * in the tree. They will be cleaned up on the next mount.
4941 trans->block_rsv = root->orphan_block_rsv;
4942 btrfs_orphan_del(trans, inode);
4944 btrfs_orphan_del(NULL, inode);
4947 trans->block_rsv = &root->fs_info->trans_block_rsv;
4948 if (!(root == root->fs_info->tree_root ||
4949 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4950 btrfs_return_ino(root, btrfs_ino(inode));
4952 btrfs_end_transaction(trans, root);
4953 btrfs_btree_balance_dirty(root);
4955 btrfs_remove_delayed_node(inode);
4961 * this returns the key found in the dir entry in the location pointer.
4962 * If no dir entries were found, location->objectid is 0.
4964 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4965 struct btrfs_key *location)
4967 const char *name = dentry->d_name.name;
4968 int namelen = dentry->d_name.len;
4969 struct btrfs_dir_item *di;
4970 struct btrfs_path *path;
4971 struct btrfs_root *root = BTRFS_I(dir)->root;
4974 path = btrfs_alloc_path();
4978 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4983 if (IS_ERR_OR_NULL(di))
4986 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4988 btrfs_free_path(path);
4991 location->objectid = 0;
4996 * when we hit a tree root in a directory, the btrfs part of the inode
4997 * needs to be changed to reflect the root directory of the tree root. This
4998 * is kind of like crossing a mount point.
5000 static int fixup_tree_root_location(struct btrfs_root *root,
5002 struct dentry *dentry,
5003 struct btrfs_key *location,
5004 struct btrfs_root **sub_root)
5006 struct btrfs_path *path;
5007 struct btrfs_root *new_root;
5008 struct btrfs_root_ref *ref;
5009 struct extent_buffer *leaf;
5013 path = btrfs_alloc_path();
5020 ret = btrfs_find_item(root->fs_info->tree_root, path,
5021 BTRFS_I(dir)->root->root_key.objectid,
5022 location->objectid, BTRFS_ROOT_REF_KEY, NULL);
5029 leaf = path->nodes[0];
5030 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5031 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5032 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5035 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5036 (unsigned long)(ref + 1),
5037 dentry->d_name.len);
5041 btrfs_release_path(path);
5043 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5044 if (IS_ERR(new_root)) {
5045 err = PTR_ERR(new_root);
5049 *sub_root = new_root;
5050 location->objectid = btrfs_root_dirid(&new_root->root_item);
5051 location->type = BTRFS_INODE_ITEM_KEY;
5052 location->offset = 0;
5055 btrfs_free_path(path);
5059 static void inode_tree_add(struct inode *inode)
5061 struct btrfs_root *root = BTRFS_I(inode)->root;
5062 struct btrfs_inode *entry;
5064 struct rb_node *parent;
5065 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5066 u64 ino = btrfs_ino(inode);
5068 if (inode_unhashed(inode))
5071 spin_lock(&root->inode_lock);
5072 p = &root->inode_tree.rb_node;
5075 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5077 if (ino < btrfs_ino(&entry->vfs_inode))
5078 p = &parent->rb_left;
5079 else if (ino > btrfs_ino(&entry->vfs_inode))
5080 p = &parent->rb_right;
5082 WARN_ON(!(entry->vfs_inode.i_state &
5083 (I_WILL_FREE | I_FREEING)));
5084 rb_replace_node(parent, new, &root->inode_tree);
5085 RB_CLEAR_NODE(parent);
5086 spin_unlock(&root->inode_lock);
5090 rb_link_node(new, parent, p);
5091 rb_insert_color(new, &root->inode_tree);
5092 spin_unlock(&root->inode_lock);
5095 static void inode_tree_del(struct inode *inode)
5097 struct btrfs_root *root = BTRFS_I(inode)->root;
5100 spin_lock(&root->inode_lock);
5101 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5102 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5103 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5104 empty = RB_EMPTY_ROOT(&root->inode_tree);
5106 spin_unlock(&root->inode_lock);
5108 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5109 synchronize_srcu(&root->fs_info->subvol_srcu);
5110 spin_lock(&root->inode_lock);
5111 empty = RB_EMPTY_ROOT(&root->inode_tree);
5112 spin_unlock(&root->inode_lock);
5114 btrfs_add_dead_root(root);
5118 void btrfs_invalidate_inodes(struct btrfs_root *root)
5120 struct rb_node *node;
5121 struct rb_node *prev;
5122 struct btrfs_inode *entry;
5123 struct inode *inode;
5126 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5127 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5129 spin_lock(&root->inode_lock);
5131 node = root->inode_tree.rb_node;
5135 entry = rb_entry(node, struct btrfs_inode, rb_node);
5137 if (objectid < btrfs_ino(&entry->vfs_inode))
5138 node = node->rb_left;
5139 else if (objectid > btrfs_ino(&entry->vfs_inode))
5140 node = node->rb_right;
5146 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5147 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5151 prev = rb_next(prev);
5155 entry = rb_entry(node, struct btrfs_inode, rb_node);
5156 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5157 inode = igrab(&entry->vfs_inode);
5159 spin_unlock(&root->inode_lock);
5160 if (atomic_read(&inode->i_count) > 1)
5161 d_prune_aliases(inode);
5163 * btrfs_drop_inode will have it removed from
5164 * the inode cache when its usage count
5169 spin_lock(&root->inode_lock);
5173 if (cond_resched_lock(&root->inode_lock))
5176 node = rb_next(node);
5178 spin_unlock(&root->inode_lock);
5181 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5183 struct btrfs_iget_args *args = p;
5184 inode->i_ino = args->location->objectid;
5185 memcpy(&BTRFS_I(inode)->location, args->location,
5186 sizeof(*args->location));
5187 BTRFS_I(inode)->root = args->root;
5191 static int btrfs_find_actor(struct inode *inode, void *opaque)
5193 struct btrfs_iget_args *args = opaque;
5194 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5195 args->root == BTRFS_I(inode)->root;
5198 static struct inode *btrfs_iget_locked(struct super_block *s,
5199 struct btrfs_key *location,
5200 struct btrfs_root *root)
5202 struct inode *inode;
5203 struct btrfs_iget_args args;
5204 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5206 args.location = location;
5209 inode = iget5_locked(s, hashval, btrfs_find_actor,
5210 btrfs_init_locked_inode,
5215 /* Get an inode object given its location and corresponding root.
5216 * Returns in *is_new if the inode was read from disk
5218 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5219 struct btrfs_root *root, int *new)
5221 struct inode *inode;
5223 inode = btrfs_iget_locked(s, location, root);
5225 return ERR_PTR(-ENOMEM);
5227 if (inode->i_state & I_NEW) {
5228 btrfs_read_locked_inode(inode);
5229 if (!is_bad_inode(inode)) {
5230 inode_tree_add(inode);
5231 unlock_new_inode(inode);
5235 unlock_new_inode(inode);
5237 inode = ERR_PTR(-ESTALE);
5244 static struct inode *new_simple_dir(struct super_block *s,
5245 struct btrfs_key *key,
5246 struct btrfs_root *root)
5248 struct inode *inode = new_inode(s);
5251 return ERR_PTR(-ENOMEM);
5253 BTRFS_I(inode)->root = root;
5254 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5255 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5257 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5258 inode->i_op = &btrfs_dir_ro_inode_operations;
5259 inode->i_fop = &simple_dir_operations;
5260 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5261 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5266 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5268 struct inode *inode;
5269 struct btrfs_root *root = BTRFS_I(dir)->root;
5270 struct btrfs_root *sub_root = root;
5271 struct btrfs_key location;
5275 if (dentry->d_name.len > BTRFS_NAME_LEN)
5276 return ERR_PTR(-ENAMETOOLONG);
5278 ret = btrfs_inode_by_name(dir, dentry, &location);
5280 return ERR_PTR(ret);
5282 if (location.objectid == 0)
5283 return ERR_PTR(-ENOENT);
5285 if (location.type == BTRFS_INODE_ITEM_KEY) {
5286 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5290 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5292 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5293 ret = fixup_tree_root_location(root, dir, dentry,
5294 &location, &sub_root);
5297 inode = ERR_PTR(ret);
5299 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5301 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5303 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5305 if (!IS_ERR(inode) && root != sub_root) {
5306 down_read(&root->fs_info->cleanup_work_sem);
5307 if (!(inode->i_sb->s_flags & MS_RDONLY))
5308 ret = btrfs_orphan_cleanup(sub_root);
5309 up_read(&root->fs_info->cleanup_work_sem);
5312 inode = ERR_PTR(ret);
5319 static int btrfs_dentry_delete(const struct dentry *dentry)
5321 struct btrfs_root *root;
5322 struct inode *inode = dentry->d_inode;
5324 if (!inode && !IS_ROOT(dentry))
5325 inode = dentry->d_parent->d_inode;
5328 root = BTRFS_I(inode)->root;
5329 if (btrfs_root_refs(&root->root_item) == 0)
5332 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5338 static void btrfs_dentry_release(struct dentry *dentry)
5340 kfree(dentry->d_fsdata);
5343 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5346 struct inode *inode;
5348 inode = btrfs_lookup_dentry(dir, dentry);
5349 if (IS_ERR(inode)) {
5350 if (PTR_ERR(inode) == -ENOENT)
5353 return ERR_CAST(inode);
5356 return d_splice_alias(inode, dentry);
5359 unsigned char btrfs_filetype_table[] = {
5360 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5363 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5365 struct inode *inode = file_inode(file);
5366 struct btrfs_root *root = BTRFS_I(inode)->root;
5367 struct btrfs_item *item;
5368 struct btrfs_dir_item *di;
5369 struct btrfs_key key;
5370 struct btrfs_key found_key;
5371 struct btrfs_path *path;
5372 struct list_head ins_list;
5373 struct list_head del_list;
5375 struct extent_buffer *leaf;
5377 unsigned char d_type;
5382 int key_type = BTRFS_DIR_INDEX_KEY;
5386 int is_curr = 0; /* ctx->pos points to the current index? */
5388 /* FIXME, use a real flag for deciding about the key type */
5389 if (root->fs_info->tree_root == root)
5390 key_type = BTRFS_DIR_ITEM_KEY;
5392 if (!dir_emit_dots(file, ctx))
5395 path = btrfs_alloc_path();
5401 if (key_type == BTRFS_DIR_INDEX_KEY) {
5402 INIT_LIST_HEAD(&ins_list);
5403 INIT_LIST_HEAD(&del_list);
5404 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5407 key.type = key_type;
5408 key.offset = ctx->pos;
5409 key.objectid = btrfs_ino(inode);
5411 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5416 leaf = path->nodes[0];
5417 slot = path->slots[0];
5418 if (slot >= btrfs_header_nritems(leaf)) {
5419 ret = btrfs_next_leaf(root, path);
5427 item = btrfs_item_nr(slot);
5428 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5430 if (found_key.objectid != key.objectid)
5432 if (found_key.type != key_type)
5434 if (found_key.offset < ctx->pos)
5436 if (key_type == BTRFS_DIR_INDEX_KEY &&
5437 btrfs_should_delete_dir_index(&del_list,
5441 ctx->pos = found_key.offset;
5444 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5446 di_total = btrfs_item_size(leaf, item);
5448 while (di_cur < di_total) {
5449 struct btrfs_key location;
5451 if (verify_dir_item(root, leaf, di))
5454 name_len = btrfs_dir_name_len(leaf, di);
5455 if (name_len <= sizeof(tmp_name)) {
5456 name_ptr = tmp_name;
5458 name_ptr = kmalloc(name_len, GFP_NOFS);
5464 read_extent_buffer(leaf, name_ptr,
5465 (unsigned long)(di + 1), name_len);
5467 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5468 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5471 /* is this a reference to our own snapshot? If so
5474 * In contrast to old kernels, we insert the snapshot's
5475 * dir item and dir index after it has been created, so
5476 * we won't find a reference to our own snapshot. We
5477 * still keep the following code for backward
5480 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5481 location.objectid == root->root_key.objectid) {
5485 over = !dir_emit(ctx, name_ptr, name_len,
5486 location.objectid, d_type);
5489 if (name_ptr != tmp_name)
5494 di_len = btrfs_dir_name_len(leaf, di) +
5495 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5497 di = (struct btrfs_dir_item *)((char *)di + di_len);
5503 if (key_type == BTRFS_DIR_INDEX_KEY) {
5506 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5511 /* Reached end of directory/root. Bump pos past the last item. */
5515 * Stop new entries from being returned after we return the last
5518 * New directory entries are assigned a strictly increasing
5519 * offset. This means that new entries created during readdir
5520 * are *guaranteed* to be seen in the future by that readdir.
5521 * This has broken buggy programs which operate on names as
5522 * they're returned by readdir. Until we re-use freed offsets
5523 * we have this hack to stop new entries from being returned
5524 * under the assumption that they'll never reach this huge
5527 * This is being careful not to overflow 32bit loff_t unless the
5528 * last entry requires it because doing so has broken 32bit apps
5531 if (key_type == BTRFS_DIR_INDEX_KEY) {
5532 if (ctx->pos >= INT_MAX)
5533 ctx->pos = LLONG_MAX;
5540 if (key_type == BTRFS_DIR_INDEX_KEY)
5541 btrfs_put_delayed_items(&ins_list, &del_list);
5542 btrfs_free_path(path);
5546 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5548 struct btrfs_root *root = BTRFS_I(inode)->root;
5549 struct btrfs_trans_handle *trans;
5551 bool nolock = false;
5553 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5556 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5559 if (wbc->sync_mode == WB_SYNC_ALL) {
5561 trans = btrfs_join_transaction_nolock(root);
5563 trans = btrfs_join_transaction(root);
5565 return PTR_ERR(trans);
5566 ret = btrfs_commit_transaction(trans, root);
5572 * This is somewhat expensive, updating the tree every time the
5573 * inode changes. But, it is most likely to find the inode in cache.
5574 * FIXME, needs more benchmarking...there are no reasons other than performance
5575 * to keep or drop this code.
5577 static int btrfs_dirty_inode(struct inode *inode)
5579 struct btrfs_root *root = BTRFS_I(inode)->root;
5580 struct btrfs_trans_handle *trans;
5583 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5586 trans = btrfs_join_transaction(root);
5588 return PTR_ERR(trans);
5590 ret = btrfs_update_inode(trans, root, inode);
5591 if (ret && ret == -ENOSPC) {
5592 /* whoops, lets try again with the full transaction */
5593 btrfs_end_transaction(trans, root);
5594 trans = btrfs_start_transaction(root, 1);
5596 return PTR_ERR(trans);
5598 ret = btrfs_update_inode(trans, root, inode);
5600 btrfs_end_transaction(trans, root);
5601 if (BTRFS_I(inode)->delayed_node)
5602 btrfs_balance_delayed_items(root);
5608 * This is a copy of file_update_time. We need this so we can return error on
5609 * ENOSPC for updating the inode in the case of file write and mmap writes.
5611 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5614 struct btrfs_root *root = BTRFS_I(inode)->root;
5616 if (btrfs_root_readonly(root))
5619 if (flags & S_VERSION)
5620 inode_inc_iversion(inode);
5621 if (flags & S_CTIME)
5622 inode->i_ctime = *now;
5623 if (flags & S_MTIME)
5624 inode->i_mtime = *now;
5625 if (flags & S_ATIME)
5626 inode->i_atime = *now;
5627 return btrfs_dirty_inode(inode);
5631 * find the highest existing sequence number in a directory
5632 * and then set the in-memory index_cnt variable to reflect
5633 * free sequence numbers
5635 static int btrfs_set_inode_index_count(struct inode *inode)
5637 struct btrfs_root *root = BTRFS_I(inode)->root;
5638 struct btrfs_key key, found_key;
5639 struct btrfs_path *path;
5640 struct extent_buffer *leaf;
5643 key.objectid = btrfs_ino(inode);
5644 key.type = BTRFS_DIR_INDEX_KEY;
5645 key.offset = (u64)-1;
5647 path = btrfs_alloc_path();
5651 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5654 /* FIXME: we should be able to handle this */
5660 * MAGIC NUMBER EXPLANATION:
5661 * since we search a directory based on f_pos we have to start at 2
5662 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5663 * else has to start at 2
5665 if (path->slots[0] == 0) {
5666 BTRFS_I(inode)->index_cnt = 2;
5672 leaf = path->nodes[0];
5673 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5675 if (found_key.objectid != btrfs_ino(inode) ||
5676 found_key.type != BTRFS_DIR_INDEX_KEY) {
5677 BTRFS_I(inode)->index_cnt = 2;
5681 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5683 btrfs_free_path(path);
5688 * helper to find a free sequence number in a given directory. This current
5689 * code is very simple, later versions will do smarter things in the btree
5691 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5695 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5696 ret = btrfs_inode_delayed_dir_index_count(dir);
5698 ret = btrfs_set_inode_index_count(dir);
5704 *index = BTRFS_I(dir)->index_cnt;
5705 BTRFS_I(dir)->index_cnt++;
5710 static int btrfs_insert_inode_locked(struct inode *inode)
5712 struct btrfs_iget_args args;
5713 args.location = &BTRFS_I(inode)->location;
5714 args.root = BTRFS_I(inode)->root;
5716 return insert_inode_locked4(inode,
5717 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
5718 btrfs_find_actor, &args);
5721 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5722 struct btrfs_root *root,
5724 const char *name, int name_len,
5725 u64 ref_objectid, u64 objectid,
5726 umode_t mode, u64 *index)
5728 struct inode *inode;
5729 struct btrfs_inode_item *inode_item;
5730 struct btrfs_key *location;
5731 struct btrfs_path *path;
5732 struct btrfs_inode_ref *ref;
5733 struct btrfs_key key[2];
5735 int nitems = name ? 2 : 1;
5739 path = btrfs_alloc_path();
5741 return ERR_PTR(-ENOMEM);
5743 inode = new_inode(root->fs_info->sb);
5745 btrfs_free_path(path);
5746 return ERR_PTR(-ENOMEM);
5750 * O_TMPFILE, set link count to 0, so that after this point,
5751 * we fill in an inode item with the correct link count.
5754 set_nlink(inode, 0);
5757 * we have to initialize this early, so we can reclaim the inode
5758 * number if we fail afterwards in this function.
5760 inode->i_ino = objectid;
5763 trace_btrfs_inode_request(dir);
5765 ret = btrfs_set_inode_index(dir, index);
5767 btrfs_free_path(path);
5769 return ERR_PTR(ret);
5775 * index_cnt is ignored for everything but a dir,
5776 * btrfs_get_inode_index_count has an explanation for the magic
5779 BTRFS_I(inode)->index_cnt = 2;
5780 BTRFS_I(inode)->dir_index = *index;
5781 BTRFS_I(inode)->root = root;
5782 BTRFS_I(inode)->generation = trans->transid;
5783 inode->i_generation = BTRFS_I(inode)->generation;
5786 * We could have gotten an inode number from somebody who was fsynced
5787 * and then removed in this same transaction, so let's just set full
5788 * sync since it will be a full sync anyway and this will blow away the
5789 * old info in the log.
5791 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5793 key[0].objectid = objectid;
5794 key[0].type = BTRFS_INODE_ITEM_KEY;
5797 sizes[0] = sizeof(struct btrfs_inode_item);
5801 * Start new inodes with an inode_ref. This is slightly more
5802 * efficient for small numbers of hard links since they will
5803 * be packed into one item. Extended refs will kick in if we
5804 * add more hard links than can fit in the ref item.
5806 key[1].objectid = objectid;
5807 key[1].type = BTRFS_INODE_REF_KEY;
5808 key[1].offset = ref_objectid;
5810 sizes[1] = name_len + sizeof(*ref);
5813 location = &BTRFS_I(inode)->location;
5814 location->objectid = objectid;
5815 location->offset = 0;
5816 location->type = BTRFS_INODE_ITEM_KEY;
5818 ret = btrfs_insert_inode_locked(inode);
5822 path->leave_spinning = 1;
5823 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
5827 inode_init_owner(inode, dir, mode);
5828 inode_set_bytes(inode, 0);
5829 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5830 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5831 struct btrfs_inode_item);
5832 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5833 sizeof(*inode_item));
5834 fill_inode_item(trans, path->nodes[0], inode_item, inode);
5837 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5838 struct btrfs_inode_ref);
5839 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5840 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5841 ptr = (unsigned long)(ref + 1);
5842 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5845 btrfs_mark_buffer_dirty(path->nodes[0]);
5846 btrfs_free_path(path);
5848 btrfs_inherit_iflags(inode, dir);
5850 if (S_ISREG(mode)) {
5851 if (btrfs_test_opt(root, NODATASUM))
5852 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5853 if (btrfs_test_opt(root, NODATACOW))
5854 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5855 BTRFS_INODE_NODATASUM;
5858 inode_tree_add(inode);
5860 trace_btrfs_inode_new(inode);
5861 btrfs_set_inode_last_trans(trans, inode);
5863 btrfs_update_root_times(trans, root);
5865 ret = btrfs_inode_inherit_props(trans, inode, dir);
5867 btrfs_err(root->fs_info,
5868 "error inheriting props for ino %llu (root %llu): %d",
5869 btrfs_ino(inode), root->root_key.objectid, ret);
5874 unlock_new_inode(inode);
5877 BTRFS_I(dir)->index_cnt--;
5878 btrfs_free_path(path);
5880 return ERR_PTR(ret);
5883 static inline u8 btrfs_inode_type(struct inode *inode)
5885 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5889 * utility function to add 'inode' into 'parent_inode' with
5890 * a give name and a given sequence number.
5891 * if 'add_backref' is true, also insert a backref from the
5892 * inode to the parent directory.
5894 int btrfs_add_link(struct btrfs_trans_handle *trans,
5895 struct inode *parent_inode, struct inode *inode,
5896 const char *name, int name_len, int add_backref, u64 index)
5899 struct btrfs_key key;
5900 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5901 u64 ino = btrfs_ino(inode);
5902 u64 parent_ino = btrfs_ino(parent_inode);
5904 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5905 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5908 key.type = BTRFS_INODE_ITEM_KEY;
5912 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5913 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5914 key.objectid, root->root_key.objectid,
5915 parent_ino, index, name, name_len);
5916 } else if (add_backref) {
5917 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5921 /* Nothing to clean up yet */
5925 ret = btrfs_insert_dir_item(trans, root, name, name_len,
5927 btrfs_inode_type(inode), index);
5928 if (ret == -EEXIST || ret == -EOVERFLOW)
5931 btrfs_abort_transaction(trans, root, ret);
5935 btrfs_i_size_write(parent_inode, parent_inode->i_size +
5937 inode_inc_iversion(parent_inode);
5938 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5939 ret = btrfs_update_inode(trans, root, parent_inode);
5941 btrfs_abort_transaction(trans, root, ret);
5945 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5948 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5949 key.objectid, root->root_key.objectid,
5950 parent_ino, &local_index, name, name_len);
5952 } else if (add_backref) {
5956 err = btrfs_del_inode_ref(trans, root, name, name_len,
5957 ino, parent_ino, &local_index);
5962 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5963 struct inode *dir, struct dentry *dentry,
5964 struct inode *inode, int backref, u64 index)
5966 int err = btrfs_add_link(trans, dir, inode,
5967 dentry->d_name.name, dentry->d_name.len,
5974 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5975 umode_t mode, dev_t rdev)
5977 struct btrfs_trans_handle *trans;
5978 struct btrfs_root *root = BTRFS_I(dir)->root;
5979 struct inode *inode = NULL;
5985 if (!new_valid_dev(rdev))
5989 * 2 for inode item and ref
5991 * 1 for xattr if selinux is on
5993 trans = btrfs_start_transaction(root, 5);
5995 return PTR_ERR(trans);
5997 err = btrfs_find_free_ino(root, &objectid);
6001 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6002 dentry->d_name.len, btrfs_ino(dir), objectid,
6004 if (IS_ERR(inode)) {
6005 err = PTR_ERR(inode);
6010 * If the active LSM wants to access the inode during
6011 * d_instantiate it needs these. Smack checks to see
6012 * if the filesystem supports xattrs by looking at the
6015 inode->i_op = &btrfs_special_inode_operations;
6016 init_special_inode(inode, inode->i_mode, rdev);
6018 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6020 goto out_unlock_inode;
6022 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6024 goto out_unlock_inode;
6026 btrfs_update_inode(trans, root, inode);
6027 unlock_new_inode(inode);
6028 d_instantiate(dentry, inode);
6032 btrfs_end_transaction(trans, root);
6033 btrfs_balance_delayed_items(root);
6034 btrfs_btree_balance_dirty(root);
6036 inode_dec_link_count(inode);
6043 unlock_new_inode(inode);
6048 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6049 umode_t mode, bool excl)
6051 struct btrfs_trans_handle *trans;
6052 struct btrfs_root *root = BTRFS_I(dir)->root;
6053 struct inode *inode = NULL;
6054 int drop_inode_on_err = 0;
6060 * 2 for inode item and ref
6062 * 1 for xattr if selinux is on
6064 trans = btrfs_start_transaction(root, 5);
6066 return PTR_ERR(trans);
6068 err = btrfs_find_free_ino(root, &objectid);
6072 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6073 dentry->d_name.len, btrfs_ino(dir), objectid,
6075 if (IS_ERR(inode)) {
6076 err = PTR_ERR(inode);
6079 drop_inode_on_err = 1;
6081 * If the active LSM wants to access the inode during
6082 * d_instantiate it needs these. Smack checks to see
6083 * if the filesystem supports xattrs by looking at the
6086 inode->i_fop = &btrfs_file_operations;
6087 inode->i_op = &btrfs_file_inode_operations;
6088 inode->i_mapping->a_ops = &btrfs_aops;
6090 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6092 goto out_unlock_inode;
6094 err = btrfs_update_inode(trans, root, inode);
6096 goto out_unlock_inode;
6098 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6100 goto out_unlock_inode;
6102 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6103 unlock_new_inode(inode);
6104 d_instantiate(dentry, inode);
6107 btrfs_end_transaction(trans, root);
6108 if (err && drop_inode_on_err) {
6109 inode_dec_link_count(inode);
6112 btrfs_balance_delayed_items(root);
6113 btrfs_btree_balance_dirty(root);
6117 unlock_new_inode(inode);
6122 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6123 struct dentry *dentry)
6125 struct btrfs_trans_handle *trans;
6126 struct btrfs_root *root = BTRFS_I(dir)->root;
6127 struct inode *inode = old_dentry->d_inode;
6132 /* do not allow sys_link's with other subvols of the same device */
6133 if (root->objectid != BTRFS_I(inode)->root->objectid)
6136 if (inode->i_nlink >= BTRFS_LINK_MAX)
6139 err = btrfs_set_inode_index(dir, &index);
6144 * 2 items for inode and inode ref
6145 * 2 items for dir items
6146 * 1 item for parent inode
6148 trans = btrfs_start_transaction(root, 5);
6149 if (IS_ERR(trans)) {
6150 err = PTR_ERR(trans);
6154 /* There are several dir indexes for this inode, clear the cache. */
6155 BTRFS_I(inode)->dir_index = 0ULL;
6157 inode_inc_iversion(inode);
6158 inode->i_ctime = CURRENT_TIME;
6160 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6162 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6167 struct dentry *parent = dentry->d_parent;
6168 err = btrfs_update_inode(trans, root, inode);
6171 if (inode->i_nlink == 1) {
6173 * If new hard link count is 1, it's a file created
6174 * with open(2) O_TMPFILE flag.
6176 err = btrfs_orphan_del(trans, inode);
6180 d_instantiate(dentry, inode);
6181 btrfs_log_new_name(trans, inode, NULL, parent);
6184 btrfs_end_transaction(trans, root);
6185 btrfs_balance_delayed_items(root);
6188 inode_dec_link_count(inode);
6191 btrfs_btree_balance_dirty(root);
6195 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6197 struct inode *inode = NULL;
6198 struct btrfs_trans_handle *trans;
6199 struct btrfs_root *root = BTRFS_I(dir)->root;
6201 int drop_on_err = 0;
6206 * 2 items for inode and ref
6207 * 2 items for dir items
6208 * 1 for xattr if selinux is on
6210 trans = btrfs_start_transaction(root, 5);
6212 return PTR_ERR(trans);
6214 err = btrfs_find_free_ino(root, &objectid);
6218 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6219 dentry->d_name.len, btrfs_ino(dir), objectid,
6220 S_IFDIR | mode, &index);
6221 if (IS_ERR(inode)) {
6222 err = PTR_ERR(inode);
6227 /* these must be set before we unlock the inode */
6228 inode->i_op = &btrfs_dir_inode_operations;
6229 inode->i_fop = &btrfs_dir_file_operations;
6231 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6233 goto out_fail_inode;
6235 btrfs_i_size_write(inode, 0);
6236 err = btrfs_update_inode(trans, root, inode);
6238 goto out_fail_inode;
6240 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6241 dentry->d_name.len, 0, index);
6243 goto out_fail_inode;
6245 d_instantiate(dentry, inode);
6247 * mkdir is special. We're unlocking after we call d_instantiate
6248 * to avoid a race with nfsd calling d_instantiate.
6250 unlock_new_inode(inode);
6254 btrfs_end_transaction(trans, root);
6256 inode_dec_link_count(inode);
6259 btrfs_balance_delayed_items(root);
6260 btrfs_btree_balance_dirty(root);
6264 unlock_new_inode(inode);
6268 /* Find next extent map of a given extent map, caller needs to ensure locks */
6269 static struct extent_map *next_extent_map(struct extent_map *em)
6271 struct rb_node *next;
6273 next = rb_next(&em->rb_node);
6276 return container_of(next, struct extent_map, rb_node);
6279 static struct extent_map *prev_extent_map(struct extent_map *em)
6281 struct rb_node *prev;
6283 prev = rb_prev(&em->rb_node);
6286 return container_of(prev, struct extent_map, rb_node);
6289 /* helper for btfs_get_extent. Given an existing extent in the tree,
6290 * the existing extent is the nearest extent to map_start,
6291 * and an extent that you want to insert, deal with overlap and insert
6292 * the best fitted new extent into the tree.
6294 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6295 struct extent_map *existing,
6296 struct extent_map *em,
6299 struct extent_map *prev;
6300 struct extent_map *next;
6305 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6307 if (existing->start > map_start) {
6309 prev = prev_extent_map(next);
6312 next = next_extent_map(prev);
6315 start = prev ? extent_map_end(prev) : em->start;
6316 start = max_t(u64, start, em->start);
6317 end = next ? next->start : extent_map_end(em);
6318 end = min_t(u64, end, extent_map_end(em));
6319 start_diff = start - em->start;
6321 em->len = end - start;
6322 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6323 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6324 em->block_start += start_diff;
6325 em->block_len -= start_diff;
6327 return add_extent_mapping(em_tree, em, 0);
6330 static noinline int uncompress_inline(struct btrfs_path *path,
6331 struct inode *inode, struct page *page,
6332 size_t pg_offset, u64 extent_offset,
6333 struct btrfs_file_extent_item *item)
6336 struct extent_buffer *leaf = path->nodes[0];
6339 unsigned long inline_size;
6343 WARN_ON(pg_offset != 0);
6344 compress_type = btrfs_file_extent_compression(leaf, item);
6345 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6346 inline_size = btrfs_file_extent_inline_item_len(leaf,
6347 btrfs_item_nr(path->slots[0]));
6348 tmp = kmalloc(inline_size, GFP_NOFS);
6351 ptr = btrfs_file_extent_inline_start(item);
6353 read_extent_buffer(leaf, tmp, ptr, inline_size);
6355 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6356 ret = btrfs_decompress(compress_type, tmp, page,
6357 extent_offset, inline_size, max_size);
6363 * a bit scary, this does extent mapping from logical file offset to the disk.
6364 * the ugly parts come from merging extents from the disk with the in-ram
6365 * representation. This gets more complex because of the data=ordered code,
6366 * where the in-ram extents might be locked pending data=ordered completion.
6368 * This also copies inline extents directly into the page.
6371 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6372 size_t pg_offset, u64 start, u64 len,
6377 u64 extent_start = 0;
6379 u64 objectid = btrfs_ino(inode);
6381 struct btrfs_path *path = NULL;
6382 struct btrfs_root *root = BTRFS_I(inode)->root;
6383 struct btrfs_file_extent_item *item;
6384 struct extent_buffer *leaf;
6385 struct btrfs_key found_key;
6386 struct extent_map *em = NULL;
6387 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6388 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6389 struct btrfs_trans_handle *trans = NULL;
6390 const bool new_inline = !page || create;
6393 read_lock(&em_tree->lock);
6394 em = lookup_extent_mapping(em_tree, start, len);
6396 em->bdev = root->fs_info->fs_devices->latest_bdev;
6397 read_unlock(&em_tree->lock);
6400 if (em->start > start || em->start + em->len <= start)
6401 free_extent_map(em);
6402 else if (em->block_start == EXTENT_MAP_INLINE && page)
6403 free_extent_map(em);
6407 em = alloc_extent_map();
6412 em->bdev = root->fs_info->fs_devices->latest_bdev;
6413 em->start = EXTENT_MAP_HOLE;
6414 em->orig_start = EXTENT_MAP_HOLE;
6416 em->block_len = (u64)-1;
6419 path = btrfs_alloc_path();
6425 * Chances are we'll be called again, so go ahead and do
6431 ret = btrfs_lookup_file_extent(trans, root, path,
6432 objectid, start, trans != NULL);
6439 if (path->slots[0] == 0)
6444 leaf = path->nodes[0];
6445 item = btrfs_item_ptr(leaf, path->slots[0],
6446 struct btrfs_file_extent_item);
6447 /* are we inside the extent that was found? */
6448 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6449 found_type = found_key.type;
6450 if (found_key.objectid != objectid ||
6451 found_type != BTRFS_EXTENT_DATA_KEY) {
6453 * If we backup past the first extent we want to move forward
6454 * and see if there is an extent in front of us, otherwise we'll
6455 * say there is a hole for our whole search range which can
6462 found_type = btrfs_file_extent_type(leaf, item);
6463 extent_start = found_key.offset;
6464 if (found_type == BTRFS_FILE_EXTENT_REG ||
6465 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6466 extent_end = extent_start +
6467 btrfs_file_extent_num_bytes(leaf, item);
6468 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6470 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6471 extent_end = ALIGN(extent_start + size, root->sectorsize);
6474 if (start >= extent_end) {
6476 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6477 ret = btrfs_next_leaf(root, path);
6484 leaf = path->nodes[0];
6486 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6487 if (found_key.objectid != objectid ||
6488 found_key.type != BTRFS_EXTENT_DATA_KEY)
6490 if (start + len <= found_key.offset)
6492 if (start > found_key.offset)
6495 em->orig_start = start;
6496 em->len = found_key.offset - start;
6500 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6502 if (found_type == BTRFS_FILE_EXTENT_REG ||
6503 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6505 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6509 size_t extent_offset;
6515 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6516 extent_offset = page_offset(page) + pg_offset - extent_start;
6517 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6518 size - extent_offset);
6519 em->start = extent_start + extent_offset;
6520 em->len = ALIGN(copy_size, root->sectorsize);
6521 em->orig_block_len = em->len;
6522 em->orig_start = em->start;
6523 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6524 if (create == 0 && !PageUptodate(page)) {
6525 if (btrfs_file_extent_compression(leaf, item) !=
6526 BTRFS_COMPRESS_NONE) {
6527 ret = uncompress_inline(path, inode, page,
6529 extent_offset, item);
6536 read_extent_buffer(leaf, map + pg_offset, ptr,
6538 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6539 memset(map + pg_offset + copy_size, 0,
6540 PAGE_CACHE_SIZE - pg_offset -
6545 flush_dcache_page(page);
6546 } else if (create && PageUptodate(page)) {
6550 free_extent_map(em);
6553 btrfs_release_path(path);
6554 trans = btrfs_join_transaction(root);
6557 return ERR_CAST(trans);
6561 write_extent_buffer(leaf, map + pg_offset, ptr,
6564 btrfs_mark_buffer_dirty(leaf);
6566 set_extent_uptodate(io_tree, em->start,
6567 extent_map_end(em) - 1, NULL, GFP_NOFS);
6572 em->orig_start = start;
6575 em->block_start = EXTENT_MAP_HOLE;
6576 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6578 btrfs_release_path(path);
6579 if (em->start > start || extent_map_end(em) <= start) {
6580 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6581 em->start, em->len, start, len);
6587 write_lock(&em_tree->lock);
6588 ret = add_extent_mapping(em_tree, em, 0);
6589 /* it is possible that someone inserted the extent into the tree
6590 * while we had the lock dropped. It is also possible that
6591 * an overlapping map exists in the tree
6593 if (ret == -EEXIST) {
6594 struct extent_map *existing;
6598 existing = search_extent_mapping(em_tree, start, len);
6600 * existing will always be non-NULL, since there must be
6601 * extent causing the -EEXIST.
6603 if (start >= extent_map_end(existing) ||
6604 start <= existing->start) {
6606 * The existing extent map is the one nearest to
6607 * the [start, start + len) range which overlaps
6609 err = merge_extent_mapping(em_tree, existing,
6611 free_extent_map(existing);
6613 free_extent_map(em);
6617 free_extent_map(em);
6622 write_unlock(&em_tree->lock);
6625 trace_btrfs_get_extent(root, em);
6628 btrfs_free_path(path);
6630 ret = btrfs_end_transaction(trans, root);
6635 free_extent_map(em);
6636 return ERR_PTR(err);
6638 BUG_ON(!em); /* Error is always set */
6642 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6643 size_t pg_offset, u64 start, u64 len,
6646 struct extent_map *em;
6647 struct extent_map *hole_em = NULL;
6648 u64 range_start = start;
6654 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6661 * - a pre-alloc extent,
6662 * there might actually be delalloc bytes behind it.
6664 if (em->block_start != EXTENT_MAP_HOLE &&
6665 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6671 /* check to see if we've wrapped (len == -1 or similar) */
6680 /* ok, we didn't find anything, lets look for delalloc */
6681 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6682 end, len, EXTENT_DELALLOC, 1);
6683 found_end = range_start + found;
6684 if (found_end < range_start)
6685 found_end = (u64)-1;
6688 * we didn't find anything useful, return
6689 * the original results from get_extent()
6691 if (range_start > end || found_end <= start) {
6697 /* adjust the range_start to make sure it doesn't
6698 * go backwards from the start they passed in
6700 range_start = max(start, range_start);
6701 found = found_end - range_start;
6704 u64 hole_start = start;
6707 em = alloc_extent_map();
6713 * when btrfs_get_extent can't find anything it
6714 * returns one huge hole
6716 * make sure what it found really fits our range, and
6717 * adjust to make sure it is based on the start from
6721 u64 calc_end = extent_map_end(hole_em);
6723 if (calc_end <= start || (hole_em->start > end)) {
6724 free_extent_map(hole_em);
6727 hole_start = max(hole_em->start, start);
6728 hole_len = calc_end - hole_start;
6732 if (hole_em && range_start > hole_start) {
6733 /* our hole starts before our delalloc, so we
6734 * have to return just the parts of the hole
6735 * that go until the delalloc starts
6737 em->len = min(hole_len,
6738 range_start - hole_start);
6739 em->start = hole_start;
6740 em->orig_start = hole_start;
6742 * don't adjust block start at all,
6743 * it is fixed at EXTENT_MAP_HOLE
6745 em->block_start = hole_em->block_start;
6746 em->block_len = hole_len;
6747 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6748 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6750 em->start = range_start;
6752 em->orig_start = range_start;
6753 em->block_start = EXTENT_MAP_DELALLOC;
6754 em->block_len = found;
6756 } else if (hole_em) {
6761 free_extent_map(hole_em);
6763 free_extent_map(em);
6764 return ERR_PTR(err);
6769 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6772 struct btrfs_root *root = BTRFS_I(inode)->root;
6773 struct extent_map *em;
6774 struct btrfs_key ins;
6778 alloc_hint = get_extent_allocation_hint(inode, start, len);
6779 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
6780 alloc_hint, &ins, 1, 1);
6782 return ERR_PTR(ret);
6784 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6785 ins.offset, ins.offset, ins.offset, 0);
6787 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
6791 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6792 ins.offset, ins.offset, 0);
6794 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
6795 free_extent_map(em);
6796 return ERR_PTR(ret);
6803 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6804 * block must be cow'd
6806 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6807 u64 *orig_start, u64 *orig_block_len,
6810 struct btrfs_trans_handle *trans;
6811 struct btrfs_path *path;
6813 struct extent_buffer *leaf;
6814 struct btrfs_root *root = BTRFS_I(inode)->root;
6815 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6816 struct btrfs_file_extent_item *fi;
6817 struct btrfs_key key;
6824 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6826 path = btrfs_alloc_path();
6830 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
6835 slot = path->slots[0];
6838 /* can't find the item, must cow */
6845 leaf = path->nodes[0];
6846 btrfs_item_key_to_cpu(leaf, &key, slot);
6847 if (key.objectid != btrfs_ino(inode) ||
6848 key.type != BTRFS_EXTENT_DATA_KEY) {
6849 /* not our file or wrong item type, must cow */
6853 if (key.offset > offset) {
6854 /* Wrong offset, must cow */
6858 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6859 found_type = btrfs_file_extent_type(leaf, fi);
6860 if (found_type != BTRFS_FILE_EXTENT_REG &&
6861 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6862 /* not a regular extent, must cow */
6866 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6869 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6870 if (extent_end <= offset)
6873 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6874 if (disk_bytenr == 0)
6877 if (btrfs_file_extent_compression(leaf, fi) ||
6878 btrfs_file_extent_encryption(leaf, fi) ||
6879 btrfs_file_extent_other_encoding(leaf, fi))
6882 backref_offset = btrfs_file_extent_offset(leaf, fi);
6885 *orig_start = key.offset - backref_offset;
6886 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6887 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6890 if (btrfs_extent_readonly(root, disk_bytenr))
6893 num_bytes = min(offset + *len, extent_end) - offset;
6894 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6897 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
6898 ret = test_range_bit(io_tree, offset, range_end,
6899 EXTENT_DELALLOC, 0, NULL);
6906 btrfs_release_path(path);
6909 * look for other files referencing this extent, if we
6910 * find any we must cow
6912 trans = btrfs_join_transaction(root);
6913 if (IS_ERR(trans)) {
6918 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6919 key.offset - backref_offset, disk_bytenr);
6920 btrfs_end_transaction(trans, root);
6927 * adjust disk_bytenr and num_bytes to cover just the bytes
6928 * in this extent we are about to write. If there
6929 * are any csums in that range we have to cow in order
6930 * to keep the csums correct
6932 disk_bytenr += backref_offset;
6933 disk_bytenr += offset - key.offset;
6934 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6937 * all of the above have passed, it is safe to overwrite this extent
6943 btrfs_free_path(path);
6947 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
6949 struct radix_tree_root *root = &inode->i_mapping->page_tree;
6951 void **pagep = NULL;
6952 struct page *page = NULL;
6956 start_idx = start >> PAGE_CACHE_SHIFT;
6959 * end is the last byte in the last page. end == start is legal
6961 end_idx = end >> PAGE_CACHE_SHIFT;
6965 /* Most of the code in this while loop is lifted from
6966 * find_get_page. It's been modified to begin searching from a
6967 * page and return just the first page found in that range. If the
6968 * found idx is less than or equal to the end idx then we know that
6969 * a page exists. If no pages are found or if those pages are
6970 * outside of the range then we're fine (yay!) */
6971 while (page == NULL &&
6972 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
6973 page = radix_tree_deref_slot(pagep);
6974 if (unlikely(!page))
6977 if (radix_tree_exception(page)) {
6978 if (radix_tree_deref_retry(page)) {
6983 * Otherwise, shmem/tmpfs must be storing a swap entry
6984 * here as an exceptional entry: so return it without
6985 * attempting to raise page count.
6988 break; /* TODO: Is this relevant for this use case? */
6991 if (!page_cache_get_speculative(page)) {
6997 * Has the page moved?
6998 * This is part of the lockless pagecache protocol. See
6999 * include/linux/pagemap.h for details.
7001 if (unlikely(page != *pagep)) {
7002 page_cache_release(page);
7008 if (page->index <= end_idx)
7010 page_cache_release(page);
7017 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7018 struct extent_state **cached_state, int writing)
7020 struct btrfs_ordered_extent *ordered;
7024 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7027 * We're concerned with the entire range that we're going to be
7028 * doing DIO to, so we need to make sure theres no ordered
7029 * extents in this range.
7031 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7032 lockend - lockstart + 1);
7035 * We need to make sure there are no buffered pages in this
7036 * range either, we could have raced between the invalidate in
7037 * generic_file_direct_write and locking the extent. The
7038 * invalidate needs to happen so that reads after a write do not
7043 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7046 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7047 cached_state, GFP_NOFS);
7050 btrfs_start_ordered_extent(inode, ordered, 1);
7051 btrfs_put_ordered_extent(ordered);
7053 /* Screw you mmap */
7054 ret = btrfs_fdatawrite_range(inode, lockstart, lockend);
7057 ret = filemap_fdatawait_range(inode->i_mapping,
7064 * If we found a page that couldn't be invalidated just
7065 * fall back to buffered.
7067 ret = invalidate_inode_pages2_range(inode->i_mapping,
7068 lockstart >> PAGE_CACHE_SHIFT,
7069 lockend >> PAGE_CACHE_SHIFT);
7080 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7081 u64 len, u64 orig_start,
7082 u64 block_start, u64 block_len,
7083 u64 orig_block_len, u64 ram_bytes,
7086 struct extent_map_tree *em_tree;
7087 struct extent_map *em;
7088 struct btrfs_root *root = BTRFS_I(inode)->root;
7091 em_tree = &BTRFS_I(inode)->extent_tree;
7092 em = alloc_extent_map();
7094 return ERR_PTR(-ENOMEM);
7097 em->orig_start = orig_start;
7098 em->mod_start = start;
7101 em->block_len = block_len;
7102 em->block_start = block_start;
7103 em->bdev = root->fs_info->fs_devices->latest_bdev;
7104 em->orig_block_len = orig_block_len;
7105 em->ram_bytes = ram_bytes;
7106 em->generation = -1;
7107 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7108 if (type == BTRFS_ORDERED_PREALLOC)
7109 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7112 btrfs_drop_extent_cache(inode, em->start,
7113 em->start + em->len - 1, 0);
7114 write_lock(&em_tree->lock);
7115 ret = add_extent_mapping(em_tree, em, 1);
7116 write_unlock(&em_tree->lock);
7117 } while (ret == -EEXIST);
7120 free_extent_map(em);
7121 return ERR_PTR(ret);
7128 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7129 struct buffer_head *bh_result, int create)
7131 struct extent_map *em;
7132 struct btrfs_root *root = BTRFS_I(inode)->root;
7133 struct extent_state *cached_state = NULL;
7134 u64 start = iblock << inode->i_blkbits;
7135 u64 lockstart, lockend;
7136 u64 len = bh_result->b_size;
7137 int unlock_bits = EXTENT_LOCKED;
7141 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
7143 len = min_t(u64, len, root->sectorsize);
7146 lockend = start + len - 1;
7149 * If this errors out it's because we couldn't invalidate pagecache for
7150 * this range and we need to fallback to buffered.
7152 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
7155 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7162 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7163 * io. INLINE is special, and we could probably kludge it in here, but
7164 * it's still buffered so for safety lets just fall back to the generic
7167 * For COMPRESSED we _have_ to read the entire extent in so we can
7168 * decompress it, so there will be buffering required no matter what we
7169 * do, so go ahead and fallback to buffered.
7171 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7172 * to buffered IO. Don't blame me, this is the price we pay for using
7175 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7176 em->block_start == EXTENT_MAP_INLINE) {
7177 free_extent_map(em);
7182 /* Just a good old fashioned hole, return */
7183 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7184 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7185 free_extent_map(em);
7190 * We don't allocate a new extent in the following cases
7192 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7194 * 2) The extent is marked as PREALLOC. We're good to go here and can
7195 * just use the extent.
7199 len = min(len, em->len - (start - em->start));
7200 lockstart = start + len;
7204 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7205 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7206 em->block_start != EXTENT_MAP_HOLE)) {
7209 u64 block_start, orig_start, orig_block_len, ram_bytes;
7211 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7212 type = BTRFS_ORDERED_PREALLOC;
7214 type = BTRFS_ORDERED_NOCOW;
7215 len = min(len, em->len - (start - em->start));
7216 block_start = em->block_start + (start - em->start);
7218 if (can_nocow_extent(inode, start, &len, &orig_start,
7219 &orig_block_len, &ram_bytes) == 1) {
7220 if (type == BTRFS_ORDERED_PREALLOC) {
7221 free_extent_map(em);
7222 em = create_pinned_em(inode, start, len,
7233 ret = btrfs_add_ordered_extent_dio(inode, start,
7234 block_start, len, len, type);
7236 free_extent_map(em);
7244 * this will cow the extent, reset the len in case we changed
7247 len = bh_result->b_size;
7248 free_extent_map(em);
7249 em = btrfs_new_extent_direct(inode, start, len);
7254 len = min(len, em->len - (start - em->start));
7256 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7258 bh_result->b_size = len;
7259 bh_result->b_bdev = em->bdev;
7260 set_buffer_mapped(bh_result);
7262 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7263 set_buffer_new(bh_result);
7266 * Need to update the i_size under the extent lock so buffered
7267 * readers will get the updated i_size when we unlock.
7269 if (start + len > i_size_read(inode))
7270 i_size_write(inode, start + len);
7272 spin_lock(&BTRFS_I(inode)->lock);
7273 BTRFS_I(inode)->outstanding_extents++;
7274 spin_unlock(&BTRFS_I(inode)->lock);
7276 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7277 lockstart + len - 1, EXTENT_DELALLOC, NULL,
7278 &cached_state, GFP_NOFS);
7283 * In the case of write we need to clear and unlock the entire range,
7284 * in the case of read we need to unlock only the end area that we
7285 * aren't using if there is any left over space.
7287 if (lockstart < lockend) {
7288 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7289 lockend, unlock_bits, 1, 0,
7290 &cached_state, GFP_NOFS);
7292 free_extent_state(cached_state);
7295 free_extent_map(em);
7300 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7301 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7305 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7306 int rw, int mirror_num)
7308 struct btrfs_root *root = BTRFS_I(inode)->root;
7311 BUG_ON(rw & REQ_WRITE);
7315 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7316 BTRFS_WQ_ENDIO_DIO_REPAIR);
7320 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7326 static int btrfs_check_dio_repairable(struct inode *inode,
7327 struct bio *failed_bio,
7328 struct io_failure_record *failrec,
7333 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7334 failrec->logical, failrec->len);
7335 if (num_copies == 1) {
7337 * we only have a single copy of the data, so don't bother with
7338 * all the retry and error correction code that follows. no
7339 * matter what the error is, it is very likely to persist.
7341 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7342 num_copies, failrec->this_mirror, failed_mirror);
7346 failrec->failed_mirror = failed_mirror;
7347 failrec->this_mirror++;
7348 if (failrec->this_mirror == failed_mirror)
7349 failrec->this_mirror++;
7351 if (failrec->this_mirror > num_copies) {
7352 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7353 num_copies, failrec->this_mirror, failed_mirror);
7360 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7361 struct page *page, u64 start, u64 end,
7362 int failed_mirror, bio_end_io_t *repair_endio,
7365 struct io_failure_record *failrec;
7371 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7373 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7377 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7380 free_io_failure(inode, failrec);
7384 if (failed_bio->bi_vcnt > 1)
7385 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7387 read_mode = READ_SYNC;
7389 isector = start - btrfs_io_bio(failed_bio)->logical;
7390 isector >>= inode->i_sb->s_blocksize_bits;
7391 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7392 0, isector, repair_endio, repair_arg);
7394 free_io_failure(inode, failrec);
7398 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7399 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7400 read_mode, failrec->this_mirror, failrec->in_validation);
7402 ret = submit_dio_repair_bio(inode, bio, read_mode,
7403 failrec->this_mirror);
7405 free_io_failure(inode, failrec);
7412 struct btrfs_retry_complete {
7413 struct completion done;
7414 struct inode *inode;
7419 static void btrfs_retry_endio_nocsum(struct bio *bio, int err)
7421 struct btrfs_retry_complete *done = bio->bi_private;
7422 struct bio_vec *bvec;
7429 bio_for_each_segment_all(bvec, bio, i)
7430 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7432 complete(&done->done);
7436 static int __btrfs_correct_data_nocsum(struct inode *inode,
7437 struct btrfs_io_bio *io_bio)
7439 struct bio_vec *bvec;
7440 struct btrfs_retry_complete done;
7445 start = io_bio->logical;
7448 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7452 init_completion(&done.done);
7454 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7455 start + bvec->bv_len - 1,
7457 btrfs_retry_endio_nocsum, &done);
7461 wait_for_completion(&done.done);
7463 if (!done.uptodate) {
7464 /* We might have another mirror, so try again */
7468 start += bvec->bv_len;
7474 static void btrfs_retry_endio(struct bio *bio, int err)
7476 struct btrfs_retry_complete *done = bio->bi_private;
7477 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7478 struct bio_vec *bvec;
7487 bio_for_each_segment_all(bvec, bio, i) {
7488 ret = __readpage_endio_check(done->inode, io_bio, i,
7490 done->start, bvec->bv_len);
7492 clean_io_failure(done->inode, done->start,
7498 done->uptodate = uptodate;
7500 complete(&done->done);
7504 static int __btrfs_subio_endio_read(struct inode *inode,
7505 struct btrfs_io_bio *io_bio, int err)
7507 struct bio_vec *bvec;
7508 struct btrfs_retry_complete done;
7515 start = io_bio->logical;
7518 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7519 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7520 0, start, bvec->bv_len);
7526 init_completion(&done.done);
7528 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7529 start + bvec->bv_len - 1,
7531 btrfs_retry_endio, &done);
7537 wait_for_completion(&done.done);
7539 if (!done.uptodate) {
7540 /* We might have another mirror, so try again */
7544 offset += bvec->bv_len;
7545 start += bvec->bv_len;
7551 static int btrfs_subio_endio_read(struct inode *inode,
7552 struct btrfs_io_bio *io_bio, int err)
7554 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7558 return __btrfs_correct_data_nocsum(inode, io_bio);
7562 return __btrfs_subio_endio_read(inode, io_bio, err);
7566 static void btrfs_endio_direct_read(struct bio *bio, int err)
7568 struct btrfs_dio_private *dip = bio->bi_private;
7569 struct inode *inode = dip->inode;
7570 struct bio *dio_bio;
7571 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7573 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
7574 err = btrfs_subio_endio_read(inode, io_bio, err);
7576 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
7577 dip->logical_offset + dip->bytes - 1);
7578 dio_bio = dip->dio_bio;
7582 /* If we had a csum failure make sure to clear the uptodate flag */
7584 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7585 dio_end_io(dio_bio, err);
7588 io_bio->end_io(io_bio, err);
7592 static void btrfs_endio_direct_write(struct bio *bio, int err)
7594 struct btrfs_dio_private *dip = bio->bi_private;
7595 struct inode *inode = dip->inode;
7596 struct btrfs_root *root = BTRFS_I(inode)->root;
7597 struct btrfs_ordered_extent *ordered = NULL;
7598 u64 ordered_offset = dip->logical_offset;
7599 u64 ordered_bytes = dip->bytes;
7600 struct bio *dio_bio;
7606 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7608 ordered_bytes, !err);
7612 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
7613 finish_ordered_fn, NULL, NULL);
7614 btrfs_queue_work(root->fs_info->endio_write_workers,
7618 * our bio might span multiple ordered extents. If we haven't
7619 * completed the accounting for the whole dio, go back and try again
7621 if (ordered_offset < dip->logical_offset + dip->bytes) {
7622 ordered_bytes = dip->logical_offset + dip->bytes -
7628 dio_bio = dip->dio_bio;
7632 /* If we had an error make sure to clear the uptodate flag */
7634 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7635 dio_end_io(dio_bio, err);
7639 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7640 struct bio *bio, int mirror_num,
7641 unsigned long bio_flags, u64 offset)
7644 struct btrfs_root *root = BTRFS_I(inode)->root;
7645 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7646 BUG_ON(ret); /* -ENOMEM */
7650 static void btrfs_end_dio_bio(struct bio *bio, int err)
7652 struct btrfs_dio_private *dip = bio->bi_private;
7655 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7656 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7657 btrfs_ino(dip->inode), bio->bi_rw,
7658 (unsigned long long)bio->bi_iter.bi_sector,
7659 bio->bi_iter.bi_size, err);
7661 if (dip->subio_endio)
7662 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
7668 * before atomic variable goto zero, we must make sure
7669 * dip->errors is perceived to be set.
7671 smp_mb__before_atomic();
7674 /* if there are more bios still pending for this dio, just exit */
7675 if (!atomic_dec_and_test(&dip->pending_bios))
7679 bio_io_error(dip->orig_bio);
7681 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7682 bio_endio(dip->orig_bio, 0);
7688 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7689 u64 first_sector, gfp_t gfp_flags)
7691 int nr_vecs = bio_get_nr_vecs(bdev);
7692 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7695 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
7696 struct inode *inode,
7697 struct btrfs_dio_private *dip,
7701 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7702 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
7706 * We load all the csum data we need when we submit
7707 * the first bio to reduce the csum tree search and
7710 if (dip->logical_offset == file_offset) {
7711 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
7717 if (bio == dip->orig_bio)
7720 file_offset -= dip->logical_offset;
7721 file_offset >>= inode->i_sb->s_blocksize_bits;
7722 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
7727 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7728 int rw, u64 file_offset, int skip_sum,
7731 struct btrfs_dio_private *dip = bio->bi_private;
7732 int write = rw & REQ_WRITE;
7733 struct btrfs_root *root = BTRFS_I(inode)->root;
7737 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7742 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7743 BTRFS_WQ_ENDIO_DATA);
7751 if (write && async_submit) {
7752 ret = btrfs_wq_submit_bio(root->fs_info,
7753 inode, rw, bio, 0, 0,
7755 __btrfs_submit_bio_start_direct_io,
7756 __btrfs_submit_bio_done);
7760 * If we aren't doing async submit, calculate the csum of the
7763 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7767 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
7773 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7779 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7782 struct inode *inode = dip->inode;
7783 struct btrfs_root *root = BTRFS_I(inode)->root;
7785 struct bio *orig_bio = dip->orig_bio;
7786 struct bio_vec *bvec = orig_bio->bi_io_vec;
7787 u64 start_sector = orig_bio->bi_iter.bi_sector;
7788 u64 file_offset = dip->logical_offset;
7793 int async_submit = 0;
7795 map_length = orig_bio->bi_iter.bi_size;
7796 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7797 &map_length, NULL, 0);
7801 if (map_length >= orig_bio->bi_iter.bi_size) {
7803 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
7807 /* async crcs make it difficult to collect full stripe writes. */
7808 if (btrfs_get_alloc_profile(root, 1) &
7809 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7814 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7818 bio->bi_private = dip;
7819 bio->bi_end_io = btrfs_end_dio_bio;
7820 btrfs_io_bio(bio)->logical = file_offset;
7821 atomic_inc(&dip->pending_bios);
7823 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7824 if (map_length < submit_len + bvec->bv_len ||
7825 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7826 bvec->bv_offset) < bvec->bv_len) {
7828 * inc the count before we submit the bio so
7829 * we know the end IO handler won't happen before
7830 * we inc the count. Otherwise, the dip might get freed
7831 * before we're done setting it up
7833 atomic_inc(&dip->pending_bios);
7834 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7835 file_offset, skip_sum,
7839 atomic_dec(&dip->pending_bios);
7843 start_sector += submit_len >> 9;
7844 file_offset += submit_len;
7849 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7850 start_sector, GFP_NOFS);
7853 bio->bi_private = dip;
7854 bio->bi_end_io = btrfs_end_dio_bio;
7855 btrfs_io_bio(bio)->logical = file_offset;
7857 map_length = orig_bio->bi_iter.bi_size;
7858 ret = btrfs_map_block(root->fs_info, rw,
7860 &map_length, NULL, 0);
7866 submit_len += bvec->bv_len;
7873 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7882 * before atomic variable goto zero, we must
7883 * make sure dip->errors is perceived to be set.
7885 smp_mb__before_atomic();
7886 if (atomic_dec_and_test(&dip->pending_bios))
7887 bio_io_error(dip->orig_bio);
7889 /* bio_end_io() will handle error, so we needn't return it */
7893 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7894 struct inode *inode, loff_t file_offset)
7896 struct btrfs_root *root = BTRFS_I(inode)->root;
7897 struct btrfs_dio_private *dip;
7899 struct btrfs_io_bio *btrfs_bio;
7901 int write = rw & REQ_WRITE;
7904 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7906 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7912 dip = kzalloc(sizeof(*dip), GFP_NOFS);
7918 dip->private = dio_bio->bi_private;
7920 dip->logical_offset = file_offset;
7921 dip->bytes = dio_bio->bi_iter.bi_size;
7922 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
7923 io_bio->bi_private = dip;
7924 dip->orig_bio = io_bio;
7925 dip->dio_bio = dio_bio;
7926 atomic_set(&dip->pending_bios, 0);
7927 btrfs_bio = btrfs_io_bio(io_bio);
7928 btrfs_bio->logical = file_offset;
7931 io_bio->bi_end_io = btrfs_endio_direct_write;
7933 io_bio->bi_end_io = btrfs_endio_direct_read;
7934 dip->subio_endio = btrfs_subio_endio_read;
7937 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7941 if (btrfs_bio->end_io)
7942 btrfs_bio->end_io(btrfs_bio, ret);
7948 * If this is a write, we need to clean up the reserved space and kill
7949 * the ordered extent.
7952 struct btrfs_ordered_extent *ordered;
7953 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7954 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7955 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7956 btrfs_free_reserved_extent(root, ordered->start,
7957 ordered->disk_len, 1);
7958 btrfs_put_ordered_extent(ordered);
7959 btrfs_put_ordered_extent(ordered);
7961 bio_endio(dio_bio, ret);
7964 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7965 const struct iov_iter *iter, loff_t offset)
7969 unsigned blocksize_mask = root->sectorsize - 1;
7970 ssize_t retval = -EINVAL;
7972 if (offset & blocksize_mask)
7975 if (iov_iter_alignment(iter) & blocksize_mask)
7978 /* If this is a write we don't need to check anymore */
7982 * Check to make sure we don't have duplicate iov_base's in this
7983 * iovec, if so return EINVAL, otherwise we'll get csum errors
7984 * when reading back.
7986 for (seg = 0; seg < iter->nr_segs; seg++) {
7987 for (i = seg + 1; i < iter->nr_segs; i++) {
7988 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
7997 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7998 struct iov_iter *iter, loff_t offset)
8000 struct file *file = iocb->ki_filp;
8001 struct inode *inode = file->f_mapping->host;
8005 bool relock = false;
8008 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iter, offset))
8011 atomic_inc(&inode->i_dio_count);
8012 smp_mb__after_atomic();
8015 * The generic stuff only does filemap_write_and_wait_range, which
8016 * isn't enough if we've written compressed pages to this area, so
8017 * we need to flush the dirty pages again to make absolutely sure
8018 * that any outstanding dirty pages are on disk.
8020 count = iov_iter_count(iter);
8021 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8022 &BTRFS_I(inode)->runtime_flags))
8023 filemap_fdatawrite_range(inode->i_mapping, offset,
8024 offset + count - 1);
8028 * If the write DIO is beyond the EOF, we need update
8029 * the isize, but it is protected by i_mutex. So we can
8030 * not unlock the i_mutex at this case.
8032 if (offset + count <= inode->i_size) {
8033 mutex_unlock(&inode->i_mutex);
8036 ret = btrfs_delalloc_reserve_space(inode, count);
8039 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8040 &BTRFS_I(inode)->runtime_flags)) {
8041 inode_dio_done(inode);
8042 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8046 ret = __blockdev_direct_IO(rw, iocb, inode,
8047 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8048 iter, offset, btrfs_get_blocks_direct, NULL,
8049 btrfs_submit_direct, flags);
8051 if (ret < 0 && ret != -EIOCBQUEUED)
8052 btrfs_delalloc_release_space(inode, count);
8053 else if (ret >= 0 && (size_t)ret < count)
8054 btrfs_delalloc_release_space(inode,
8055 count - (size_t)ret);
8057 btrfs_delalloc_release_metadata(inode, 0);
8061 inode_dio_done(inode);
8063 mutex_lock(&inode->i_mutex);
8068 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8070 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8071 __u64 start, __u64 len)
8075 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8079 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8082 int btrfs_readpage(struct file *file, struct page *page)
8084 struct extent_io_tree *tree;
8085 tree = &BTRFS_I(page->mapping->host)->io_tree;
8086 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8089 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8091 struct extent_io_tree *tree;
8094 if (current->flags & PF_MEMALLOC) {
8095 redirty_page_for_writepage(wbc, page);
8099 tree = &BTRFS_I(page->mapping->host)->io_tree;
8100 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8103 static int btrfs_writepages(struct address_space *mapping,
8104 struct writeback_control *wbc)
8106 struct extent_io_tree *tree;
8108 tree = &BTRFS_I(mapping->host)->io_tree;
8109 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8113 btrfs_readpages(struct file *file, struct address_space *mapping,
8114 struct list_head *pages, unsigned nr_pages)
8116 struct extent_io_tree *tree;
8117 tree = &BTRFS_I(mapping->host)->io_tree;
8118 return extent_readpages(tree, mapping, pages, nr_pages,
8121 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8123 struct extent_io_tree *tree;
8124 struct extent_map_tree *map;
8127 tree = &BTRFS_I(page->mapping->host)->io_tree;
8128 map = &BTRFS_I(page->mapping->host)->extent_tree;
8129 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8131 ClearPagePrivate(page);
8132 set_page_private(page, 0);
8133 page_cache_release(page);
8138 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8140 if (PageWriteback(page) || PageDirty(page))
8142 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8145 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8146 unsigned int length)
8148 struct inode *inode = page->mapping->host;
8149 struct extent_io_tree *tree;
8150 struct btrfs_ordered_extent *ordered;
8151 struct extent_state *cached_state = NULL;
8152 u64 page_start = page_offset(page);
8153 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
8154 int inode_evicting = inode->i_state & I_FREEING;
8157 * we have the page locked, so new writeback can't start,
8158 * and the dirty bit won't be cleared while we are here.
8160 * Wait for IO on this page so that we can safely clear
8161 * the PagePrivate2 bit and do ordered accounting
8163 wait_on_page_writeback(page);
8165 tree = &BTRFS_I(inode)->io_tree;
8167 btrfs_releasepage(page, GFP_NOFS);
8171 if (!inode_evicting)
8172 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
8173 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8176 * IO on this page will never be started, so we need
8177 * to account for any ordered extents now
8179 if (!inode_evicting)
8180 clear_extent_bit(tree, page_start, page_end,
8181 EXTENT_DIRTY | EXTENT_DELALLOC |
8182 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8183 EXTENT_DEFRAG, 1, 0, &cached_state,
8186 * whoever cleared the private bit is responsible
8187 * for the finish_ordered_io
8189 if (TestClearPagePrivate2(page)) {
8190 struct btrfs_ordered_inode_tree *tree;
8193 tree = &BTRFS_I(inode)->ordered_tree;
8195 spin_lock_irq(&tree->lock);
8196 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8197 new_len = page_start - ordered->file_offset;
8198 if (new_len < ordered->truncated_len)
8199 ordered->truncated_len = new_len;
8200 spin_unlock_irq(&tree->lock);
8202 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8204 PAGE_CACHE_SIZE, 1))
8205 btrfs_finish_ordered_io(ordered);
8207 btrfs_put_ordered_extent(ordered);
8208 if (!inode_evicting) {
8209 cached_state = NULL;
8210 lock_extent_bits(tree, page_start, page_end, 0,
8215 if (!inode_evicting) {
8216 clear_extent_bit(tree, page_start, page_end,
8217 EXTENT_LOCKED | EXTENT_DIRTY |
8218 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8219 EXTENT_DEFRAG, 1, 1,
8220 &cached_state, GFP_NOFS);
8222 __btrfs_releasepage(page, GFP_NOFS);
8225 ClearPageChecked(page);
8226 if (PagePrivate(page)) {
8227 ClearPagePrivate(page);
8228 set_page_private(page, 0);
8229 page_cache_release(page);
8234 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8235 * called from a page fault handler when a page is first dirtied. Hence we must
8236 * be careful to check for EOF conditions here. We set the page up correctly
8237 * for a written page which means we get ENOSPC checking when writing into
8238 * holes and correct delalloc and unwritten extent mapping on filesystems that
8239 * support these features.
8241 * We are not allowed to take the i_mutex here so we have to play games to
8242 * protect against truncate races as the page could now be beyond EOF. Because
8243 * vmtruncate() writes the inode size before removing pages, once we have the
8244 * page lock we can determine safely if the page is beyond EOF. If it is not
8245 * beyond EOF, then the page is guaranteed safe against truncation until we
8248 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8250 struct page *page = vmf->page;
8251 struct inode *inode = file_inode(vma->vm_file);
8252 struct btrfs_root *root = BTRFS_I(inode)->root;
8253 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8254 struct btrfs_ordered_extent *ordered;
8255 struct extent_state *cached_state = NULL;
8257 unsigned long zero_start;
8264 sb_start_pagefault(inode->i_sb);
8265 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
8267 ret = file_update_time(vma->vm_file);
8273 else /* -ENOSPC, -EIO, etc */
8274 ret = VM_FAULT_SIGBUS;
8280 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8283 size = i_size_read(inode);
8284 page_start = page_offset(page);
8285 page_end = page_start + PAGE_CACHE_SIZE - 1;
8287 if ((page->mapping != inode->i_mapping) ||
8288 (page_start >= size)) {
8289 /* page got truncated out from underneath us */
8292 wait_on_page_writeback(page);
8294 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
8295 set_page_extent_mapped(page);
8298 * we can't set the delalloc bits if there are pending ordered
8299 * extents. Drop our locks and wait for them to finish
8301 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8303 unlock_extent_cached(io_tree, page_start, page_end,
8304 &cached_state, GFP_NOFS);
8306 btrfs_start_ordered_extent(inode, ordered, 1);
8307 btrfs_put_ordered_extent(ordered);
8312 * XXX - page_mkwrite gets called every time the page is dirtied, even
8313 * if it was already dirty, so for space accounting reasons we need to
8314 * clear any delalloc bits for the range we are fixing to save. There
8315 * is probably a better way to do this, but for now keep consistent with
8316 * prepare_pages in the normal write path.
8318 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
8319 EXTENT_DIRTY | EXTENT_DELALLOC |
8320 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8321 0, 0, &cached_state, GFP_NOFS);
8323 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
8326 unlock_extent_cached(io_tree, page_start, page_end,
8327 &cached_state, GFP_NOFS);
8328 ret = VM_FAULT_SIGBUS;
8333 /* page is wholly or partially inside EOF */
8334 if (page_start + PAGE_CACHE_SIZE > size)
8335 zero_start = size & ~PAGE_CACHE_MASK;
8337 zero_start = PAGE_CACHE_SIZE;
8339 if (zero_start != PAGE_CACHE_SIZE) {
8341 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
8342 flush_dcache_page(page);
8345 ClearPageChecked(page);
8346 set_page_dirty(page);
8347 SetPageUptodate(page);
8349 BTRFS_I(inode)->last_trans = root->fs_info->generation;
8350 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8351 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8353 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
8357 sb_end_pagefault(inode->i_sb);
8358 return VM_FAULT_LOCKED;
8362 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
8364 sb_end_pagefault(inode->i_sb);
8368 static int btrfs_truncate(struct inode *inode)
8370 struct btrfs_root *root = BTRFS_I(inode)->root;
8371 struct btrfs_block_rsv *rsv;
8374 struct btrfs_trans_handle *trans;
8375 u64 mask = root->sectorsize - 1;
8376 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
8378 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8384 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8385 * 3 things going on here
8387 * 1) We need to reserve space for our orphan item and the space to
8388 * delete our orphan item. Lord knows we don't want to have a dangling
8389 * orphan item because we didn't reserve space to remove it.
8391 * 2) We need to reserve space to update our inode.
8393 * 3) We need to have something to cache all the space that is going to
8394 * be free'd up by the truncate operation, but also have some slack
8395 * space reserved in case it uses space during the truncate (thank you
8396 * very much snapshotting).
8398 * And we need these to all be seperate. The fact is we can use alot of
8399 * space doing the truncate, and we have no earthly idea how much space
8400 * we will use, so we need the truncate reservation to be seperate so it
8401 * doesn't end up using space reserved for updating the inode or
8402 * removing the orphan item. We also need to be able to stop the
8403 * transaction and start a new one, which means we need to be able to
8404 * update the inode several times, and we have no idea of knowing how
8405 * many times that will be, so we can't just reserve 1 item for the
8406 * entirety of the opration, so that has to be done seperately as well.
8407 * Then there is the orphan item, which does indeed need to be held on
8408 * to for the whole operation, and we need nobody to touch this reserved
8409 * space except the orphan code.
8411 * So that leaves us with
8413 * 1) root->orphan_block_rsv - for the orphan deletion.
8414 * 2) rsv - for the truncate reservation, which we will steal from the
8415 * transaction reservation.
8416 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8417 * updating the inode.
8419 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
8422 rsv->size = min_size;
8426 * 1 for the truncate slack space
8427 * 1 for updating the inode.
8429 trans = btrfs_start_transaction(root, 2);
8430 if (IS_ERR(trans)) {
8431 err = PTR_ERR(trans);
8435 /* Migrate the slack space for the truncate to our reserve */
8436 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
8441 * So if we truncate and then write and fsync we normally would just
8442 * write the extents that changed, which is a problem if we need to
8443 * first truncate that entire inode. So set this flag so we write out
8444 * all of the extents in the inode to the sync log so we're completely
8447 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8448 trans->block_rsv = rsv;
8451 ret = btrfs_truncate_inode_items(trans, root, inode,
8453 BTRFS_EXTENT_DATA_KEY);
8454 if (ret != -ENOSPC) {
8459 trans->block_rsv = &root->fs_info->trans_block_rsv;
8460 ret = btrfs_update_inode(trans, root, inode);
8466 btrfs_end_transaction(trans, root);
8467 btrfs_btree_balance_dirty(root);
8469 trans = btrfs_start_transaction(root, 2);
8470 if (IS_ERR(trans)) {
8471 ret = err = PTR_ERR(trans);
8476 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
8478 BUG_ON(ret); /* shouldn't happen */
8479 trans->block_rsv = rsv;
8482 if (ret == 0 && inode->i_nlink > 0) {
8483 trans->block_rsv = root->orphan_block_rsv;
8484 ret = btrfs_orphan_del(trans, inode);
8490 trans->block_rsv = &root->fs_info->trans_block_rsv;
8491 ret = btrfs_update_inode(trans, root, inode);
8495 ret = btrfs_end_transaction(trans, root);
8496 btrfs_btree_balance_dirty(root);
8500 btrfs_free_block_rsv(root, rsv);
8509 * create a new subvolume directory/inode (helper for the ioctl).
8511 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8512 struct btrfs_root *new_root,
8513 struct btrfs_root *parent_root,
8516 struct inode *inode;
8520 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8521 new_dirid, new_dirid,
8522 S_IFDIR | (~current_umask() & S_IRWXUGO),
8525 return PTR_ERR(inode);
8526 inode->i_op = &btrfs_dir_inode_operations;
8527 inode->i_fop = &btrfs_dir_file_operations;
8529 set_nlink(inode, 1);
8530 btrfs_i_size_write(inode, 0);
8531 unlock_new_inode(inode);
8533 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8535 btrfs_err(new_root->fs_info,
8536 "error inheriting subvolume %llu properties: %d",
8537 new_root->root_key.objectid, err);
8539 err = btrfs_update_inode(trans, new_root, inode);
8545 struct inode *btrfs_alloc_inode(struct super_block *sb)
8547 struct btrfs_inode *ei;
8548 struct inode *inode;
8550 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
8557 ei->last_sub_trans = 0;
8558 ei->logged_trans = 0;
8559 ei->delalloc_bytes = 0;
8560 ei->defrag_bytes = 0;
8561 ei->disk_i_size = 0;
8564 ei->index_cnt = (u64)-1;
8566 ei->last_unlink_trans = 0;
8567 ei->last_log_commit = 0;
8569 spin_lock_init(&ei->lock);
8570 ei->outstanding_extents = 0;
8571 ei->reserved_extents = 0;
8573 ei->runtime_flags = 0;
8574 ei->force_compress = BTRFS_COMPRESS_NONE;
8576 ei->delayed_node = NULL;
8578 inode = &ei->vfs_inode;
8579 extent_map_tree_init(&ei->extent_tree);
8580 extent_io_tree_init(&ei->io_tree, &inode->i_data);
8581 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
8582 ei->io_tree.track_uptodate = 1;
8583 ei->io_failure_tree.track_uptodate = 1;
8584 atomic_set(&ei->sync_writers, 0);
8585 mutex_init(&ei->log_mutex);
8586 mutex_init(&ei->delalloc_mutex);
8587 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8588 INIT_LIST_HEAD(&ei->delalloc_inodes);
8589 RB_CLEAR_NODE(&ei->rb_node);
8594 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8595 void btrfs_test_destroy_inode(struct inode *inode)
8597 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8598 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8602 static void btrfs_i_callback(struct rcu_head *head)
8604 struct inode *inode = container_of(head, struct inode, i_rcu);
8605 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8608 void btrfs_destroy_inode(struct inode *inode)
8610 struct btrfs_ordered_extent *ordered;
8611 struct btrfs_root *root = BTRFS_I(inode)->root;
8613 WARN_ON(!hlist_empty(&inode->i_dentry));
8614 WARN_ON(inode->i_data.nrpages);
8615 WARN_ON(BTRFS_I(inode)->outstanding_extents);
8616 WARN_ON(BTRFS_I(inode)->reserved_extents);
8617 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8618 WARN_ON(BTRFS_I(inode)->csum_bytes);
8619 WARN_ON(BTRFS_I(inode)->defrag_bytes);
8622 * This can happen where we create an inode, but somebody else also
8623 * created the same inode and we need to destroy the one we already
8629 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8630 &BTRFS_I(inode)->runtime_flags)) {
8631 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
8633 atomic_dec(&root->orphan_inodes);
8637 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8641 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
8642 ordered->file_offset, ordered->len);
8643 btrfs_remove_ordered_extent(inode, ordered);
8644 btrfs_put_ordered_extent(ordered);
8645 btrfs_put_ordered_extent(ordered);
8648 inode_tree_del(inode);
8649 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8651 call_rcu(&inode->i_rcu, btrfs_i_callback);
8654 int btrfs_drop_inode(struct inode *inode)
8656 struct btrfs_root *root = BTRFS_I(inode)->root;
8661 /* the snap/subvol tree is on deleting */
8662 if (btrfs_root_refs(&root->root_item) == 0)
8665 return generic_drop_inode(inode);
8668 static void init_once(void *foo)
8670 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8672 inode_init_once(&ei->vfs_inode);
8675 void btrfs_destroy_cachep(void)
8678 * Make sure all delayed rcu free inodes are flushed before we
8682 if (btrfs_inode_cachep)
8683 kmem_cache_destroy(btrfs_inode_cachep);
8684 if (btrfs_trans_handle_cachep)
8685 kmem_cache_destroy(btrfs_trans_handle_cachep);
8686 if (btrfs_transaction_cachep)
8687 kmem_cache_destroy(btrfs_transaction_cachep);
8688 if (btrfs_path_cachep)
8689 kmem_cache_destroy(btrfs_path_cachep);
8690 if (btrfs_free_space_cachep)
8691 kmem_cache_destroy(btrfs_free_space_cachep);
8692 if (btrfs_delalloc_work_cachep)
8693 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8696 int btrfs_init_cachep(void)
8698 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8699 sizeof(struct btrfs_inode), 0,
8700 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8701 if (!btrfs_inode_cachep)
8704 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8705 sizeof(struct btrfs_trans_handle), 0,
8706 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8707 if (!btrfs_trans_handle_cachep)
8710 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8711 sizeof(struct btrfs_transaction), 0,
8712 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8713 if (!btrfs_transaction_cachep)
8716 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8717 sizeof(struct btrfs_path), 0,
8718 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8719 if (!btrfs_path_cachep)
8722 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8723 sizeof(struct btrfs_free_space), 0,
8724 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8725 if (!btrfs_free_space_cachep)
8728 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8729 sizeof(struct btrfs_delalloc_work), 0,
8730 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8732 if (!btrfs_delalloc_work_cachep)
8737 btrfs_destroy_cachep();
8741 static int btrfs_getattr(struct vfsmount *mnt,
8742 struct dentry *dentry, struct kstat *stat)
8745 struct inode *inode = dentry->d_inode;
8746 u32 blocksize = inode->i_sb->s_blocksize;
8748 generic_fillattr(inode, stat);
8749 stat->dev = BTRFS_I(inode)->root->anon_dev;
8750 stat->blksize = PAGE_CACHE_SIZE;
8752 spin_lock(&BTRFS_I(inode)->lock);
8753 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8754 spin_unlock(&BTRFS_I(inode)->lock);
8755 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8756 ALIGN(delalloc_bytes, blocksize)) >> 9;
8760 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8761 struct inode *new_dir, struct dentry *new_dentry)
8763 struct btrfs_trans_handle *trans;
8764 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8765 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8766 struct inode *new_inode = new_dentry->d_inode;
8767 struct inode *old_inode = old_dentry->d_inode;
8768 struct timespec ctime = CURRENT_TIME;
8772 u64 old_ino = btrfs_ino(old_inode);
8774 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8777 /* we only allow rename subvolume link between subvolumes */
8778 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8781 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8782 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8785 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8786 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8790 /* check for collisions, even if the name isn't there */
8791 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
8792 new_dentry->d_name.name,
8793 new_dentry->d_name.len);
8796 if (ret == -EEXIST) {
8798 * eexist without a new_inode */
8799 if (WARN_ON(!new_inode)) {
8803 /* maybe -EOVERFLOW */
8810 * we're using rename to replace one file with another. Start IO on it
8811 * now so we don't add too much work to the end of the transaction
8813 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
8814 filemap_flush(old_inode->i_mapping);
8816 /* close the racy window with snapshot create/destroy ioctl */
8817 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8818 down_read(&root->fs_info->subvol_sem);
8820 * We want to reserve the absolute worst case amount of items. So if
8821 * both inodes are subvols and we need to unlink them then that would
8822 * require 4 item modifications, but if they are both normal inodes it
8823 * would require 5 item modifications, so we'll assume their normal
8824 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8825 * should cover the worst case number of items we'll modify.
8827 trans = btrfs_start_transaction(root, 11);
8828 if (IS_ERR(trans)) {
8829 ret = PTR_ERR(trans);
8834 btrfs_record_root_in_trans(trans, dest);
8836 ret = btrfs_set_inode_index(new_dir, &index);
8840 BTRFS_I(old_inode)->dir_index = 0ULL;
8841 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8842 /* force full log commit if subvolume involved. */
8843 btrfs_set_log_full_commit(root->fs_info, trans);
8845 ret = btrfs_insert_inode_ref(trans, dest,
8846 new_dentry->d_name.name,
8847 new_dentry->d_name.len,
8849 btrfs_ino(new_dir), index);
8853 * this is an ugly little race, but the rename is required
8854 * to make sure that if we crash, the inode is either at the
8855 * old name or the new one. pinning the log transaction lets
8856 * us make sure we don't allow a log commit to come in after
8857 * we unlink the name but before we add the new name back in.
8859 btrfs_pin_log_trans(root);
8862 inode_inc_iversion(old_dir);
8863 inode_inc_iversion(new_dir);
8864 inode_inc_iversion(old_inode);
8865 old_dir->i_ctime = old_dir->i_mtime = ctime;
8866 new_dir->i_ctime = new_dir->i_mtime = ctime;
8867 old_inode->i_ctime = ctime;
8869 if (old_dentry->d_parent != new_dentry->d_parent)
8870 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8872 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8873 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8874 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8875 old_dentry->d_name.name,
8876 old_dentry->d_name.len);
8878 ret = __btrfs_unlink_inode(trans, root, old_dir,
8879 old_dentry->d_inode,
8880 old_dentry->d_name.name,
8881 old_dentry->d_name.len);
8883 ret = btrfs_update_inode(trans, root, old_inode);
8886 btrfs_abort_transaction(trans, root, ret);
8891 inode_inc_iversion(new_inode);
8892 new_inode->i_ctime = CURRENT_TIME;
8893 if (unlikely(btrfs_ino(new_inode) ==
8894 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8895 root_objectid = BTRFS_I(new_inode)->location.objectid;
8896 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8898 new_dentry->d_name.name,
8899 new_dentry->d_name.len);
8900 BUG_ON(new_inode->i_nlink == 0);
8902 ret = btrfs_unlink_inode(trans, dest, new_dir,
8903 new_dentry->d_inode,
8904 new_dentry->d_name.name,
8905 new_dentry->d_name.len);
8907 if (!ret && new_inode->i_nlink == 0)
8908 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8910 btrfs_abort_transaction(trans, root, ret);
8915 ret = btrfs_add_link(trans, new_dir, old_inode,
8916 new_dentry->d_name.name,
8917 new_dentry->d_name.len, 0, index);
8919 btrfs_abort_transaction(trans, root, ret);
8923 if (old_inode->i_nlink == 1)
8924 BTRFS_I(old_inode)->dir_index = index;
8926 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8927 struct dentry *parent = new_dentry->d_parent;
8928 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8929 btrfs_end_log_trans(root);
8932 btrfs_end_transaction(trans, root);
8934 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8935 up_read(&root->fs_info->subvol_sem);
8940 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
8941 struct inode *new_dir, struct dentry *new_dentry,
8944 if (flags & ~RENAME_NOREPLACE)
8947 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
8950 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8952 struct btrfs_delalloc_work *delalloc_work;
8953 struct inode *inode;
8955 delalloc_work = container_of(work, struct btrfs_delalloc_work,
8957 inode = delalloc_work->inode;
8958 if (delalloc_work->wait) {
8959 btrfs_wait_ordered_range(inode, 0, (u64)-1);
8961 filemap_flush(inode->i_mapping);
8962 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8963 &BTRFS_I(inode)->runtime_flags))
8964 filemap_flush(inode->i_mapping);
8967 if (delalloc_work->delay_iput)
8968 btrfs_add_delayed_iput(inode);
8971 complete(&delalloc_work->completion);
8974 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8975 int wait, int delay_iput)
8977 struct btrfs_delalloc_work *work;
8979 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8983 init_completion(&work->completion);
8984 INIT_LIST_HEAD(&work->list);
8985 work->inode = inode;
8987 work->delay_iput = delay_iput;
8988 WARN_ON_ONCE(!inode);
8989 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
8990 btrfs_run_delalloc_work, NULL, NULL);
8995 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8997 wait_for_completion(&work->completion);
8998 kmem_cache_free(btrfs_delalloc_work_cachep, work);
9002 * some fairly slow code that needs optimization. This walks the list
9003 * of all the inodes with pending delalloc and forces them to disk.
9005 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9008 struct btrfs_inode *binode;
9009 struct inode *inode;
9010 struct btrfs_delalloc_work *work, *next;
9011 struct list_head works;
9012 struct list_head splice;
9015 INIT_LIST_HEAD(&works);
9016 INIT_LIST_HEAD(&splice);
9018 mutex_lock(&root->delalloc_mutex);
9019 spin_lock(&root->delalloc_lock);
9020 list_splice_init(&root->delalloc_inodes, &splice);
9021 while (!list_empty(&splice)) {
9022 binode = list_entry(splice.next, struct btrfs_inode,
9025 list_move_tail(&binode->delalloc_inodes,
9026 &root->delalloc_inodes);
9027 inode = igrab(&binode->vfs_inode);
9029 cond_resched_lock(&root->delalloc_lock);
9032 spin_unlock(&root->delalloc_lock);
9034 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
9037 btrfs_add_delayed_iput(inode);
9043 list_add_tail(&work->list, &works);
9044 btrfs_queue_work(root->fs_info->flush_workers,
9047 if (nr != -1 && ret >= nr)
9050 spin_lock(&root->delalloc_lock);
9052 spin_unlock(&root->delalloc_lock);
9055 list_for_each_entry_safe(work, next, &works, list) {
9056 list_del_init(&work->list);
9057 btrfs_wait_and_free_delalloc_work(work);
9060 if (!list_empty_careful(&splice)) {
9061 spin_lock(&root->delalloc_lock);
9062 list_splice_tail(&splice, &root->delalloc_inodes);
9063 spin_unlock(&root->delalloc_lock);
9065 mutex_unlock(&root->delalloc_mutex);
9069 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9073 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
9076 ret = __start_delalloc_inodes(root, delay_iput, -1);
9080 * the filemap_flush will queue IO into the worker threads, but
9081 * we have to make sure the IO is actually started and that
9082 * ordered extents get created before we return
9084 atomic_inc(&root->fs_info->async_submit_draining);
9085 while (atomic_read(&root->fs_info->nr_async_submits) ||
9086 atomic_read(&root->fs_info->async_delalloc_pages)) {
9087 wait_event(root->fs_info->async_submit_wait,
9088 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9089 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
9091 atomic_dec(&root->fs_info->async_submit_draining);
9095 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9098 struct btrfs_root *root;
9099 struct list_head splice;
9102 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9105 INIT_LIST_HEAD(&splice);
9107 mutex_lock(&fs_info->delalloc_root_mutex);
9108 spin_lock(&fs_info->delalloc_root_lock);
9109 list_splice_init(&fs_info->delalloc_roots, &splice);
9110 while (!list_empty(&splice) && nr) {
9111 root = list_first_entry(&splice, struct btrfs_root,
9113 root = btrfs_grab_fs_root(root);
9115 list_move_tail(&root->delalloc_root,
9116 &fs_info->delalloc_roots);
9117 spin_unlock(&fs_info->delalloc_root_lock);
9119 ret = __start_delalloc_inodes(root, delay_iput, nr);
9120 btrfs_put_fs_root(root);
9128 spin_lock(&fs_info->delalloc_root_lock);
9130 spin_unlock(&fs_info->delalloc_root_lock);
9133 atomic_inc(&fs_info->async_submit_draining);
9134 while (atomic_read(&fs_info->nr_async_submits) ||
9135 atomic_read(&fs_info->async_delalloc_pages)) {
9136 wait_event(fs_info->async_submit_wait,
9137 (atomic_read(&fs_info->nr_async_submits) == 0 &&
9138 atomic_read(&fs_info->async_delalloc_pages) == 0));
9140 atomic_dec(&fs_info->async_submit_draining);
9142 if (!list_empty_careful(&splice)) {
9143 spin_lock(&fs_info->delalloc_root_lock);
9144 list_splice_tail(&splice, &fs_info->delalloc_roots);
9145 spin_unlock(&fs_info->delalloc_root_lock);
9147 mutex_unlock(&fs_info->delalloc_root_mutex);
9151 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9152 const char *symname)
9154 struct btrfs_trans_handle *trans;
9155 struct btrfs_root *root = BTRFS_I(dir)->root;
9156 struct btrfs_path *path;
9157 struct btrfs_key key;
9158 struct inode *inode = NULL;
9166 struct btrfs_file_extent_item *ei;
9167 struct extent_buffer *leaf;
9169 name_len = strlen(symname);
9170 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9171 return -ENAMETOOLONG;
9174 * 2 items for inode item and ref
9175 * 2 items for dir items
9176 * 1 item for xattr if selinux is on
9178 trans = btrfs_start_transaction(root, 5);
9180 return PTR_ERR(trans);
9182 err = btrfs_find_free_ino(root, &objectid);
9186 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9187 dentry->d_name.len, btrfs_ino(dir), objectid,
9188 S_IFLNK|S_IRWXUGO, &index);
9189 if (IS_ERR(inode)) {
9190 err = PTR_ERR(inode);
9195 * If the active LSM wants to access the inode during
9196 * d_instantiate it needs these. Smack checks to see
9197 * if the filesystem supports xattrs by looking at the
9200 inode->i_fop = &btrfs_file_operations;
9201 inode->i_op = &btrfs_file_inode_operations;
9202 inode->i_mapping->a_ops = &btrfs_aops;
9203 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9205 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9207 goto out_unlock_inode;
9209 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
9211 goto out_unlock_inode;
9213 path = btrfs_alloc_path();
9216 goto out_unlock_inode;
9218 key.objectid = btrfs_ino(inode);
9220 key.type = BTRFS_EXTENT_DATA_KEY;
9221 datasize = btrfs_file_extent_calc_inline_size(name_len);
9222 err = btrfs_insert_empty_item(trans, root, path, &key,
9225 btrfs_free_path(path);
9226 goto out_unlock_inode;
9228 leaf = path->nodes[0];
9229 ei = btrfs_item_ptr(leaf, path->slots[0],
9230 struct btrfs_file_extent_item);
9231 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9232 btrfs_set_file_extent_type(leaf, ei,
9233 BTRFS_FILE_EXTENT_INLINE);
9234 btrfs_set_file_extent_encryption(leaf, ei, 0);
9235 btrfs_set_file_extent_compression(leaf, ei, 0);
9236 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9237 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9239 ptr = btrfs_file_extent_inline_start(ei);
9240 write_extent_buffer(leaf, symname, ptr, name_len);
9241 btrfs_mark_buffer_dirty(leaf);
9242 btrfs_free_path(path);
9244 inode->i_op = &btrfs_symlink_inode_operations;
9245 inode->i_mapping->a_ops = &btrfs_symlink_aops;
9246 inode_set_bytes(inode, name_len);
9247 btrfs_i_size_write(inode, name_len);
9248 err = btrfs_update_inode(trans, root, inode);
9251 goto out_unlock_inode;
9254 unlock_new_inode(inode);
9255 d_instantiate(dentry, inode);
9258 btrfs_end_transaction(trans, root);
9260 inode_dec_link_count(inode);
9263 btrfs_btree_balance_dirty(root);
9268 unlock_new_inode(inode);
9272 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9273 u64 start, u64 num_bytes, u64 min_size,
9274 loff_t actual_len, u64 *alloc_hint,
9275 struct btrfs_trans_handle *trans)
9277 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9278 struct extent_map *em;
9279 struct btrfs_root *root = BTRFS_I(inode)->root;
9280 struct btrfs_key ins;
9281 u64 cur_offset = start;
9285 bool own_trans = true;
9289 while (num_bytes > 0) {
9291 trans = btrfs_start_transaction(root, 3);
9292 if (IS_ERR(trans)) {
9293 ret = PTR_ERR(trans);
9298 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
9299 cur_bytes = max(cur_bytes, min_size);
9300 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
9301 *alloc_hint, &ins, 1, 0);
9304 btrfs_end_transaction(trans, root);
9308 ret = insert_reserved_file_extent(trans, inode,
9309 cur_offset, ins.objectid,
9310 ins.offset, ins.offset,
9311 ins.offset, 0, 0, 0,
9312 BTRFS_FILE_EXTENT_PREALLOC);
9314 btrfs_free_reserved_extent(root, ins.objectid,
9316 btrfs_abort_transaction(trans, root, ret);
9318 btrfs_end_transaction(trans, root);
9321 btrfs_drop_extent_cache(inode, cur_offset,
9322 cur_offset + ins.offset -1, 0);
9324 em = alloc_extent_map();
9326 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9327 &BTRFS_I(inode)->runtime_flags);
9331 em->start = cur_offset;
9332 em->orig_start = cur_offset;
9333 em->len = ins.offset;
9334 em->block_start = ins.objectid;
9335 em->block_len = ins.offset;
9336 em->orig_block_len = ins.offset;
9337 em->ram_bytes = ins.offset;
9338 em->bdev = root->fs_info->fs_devices->latest_bdev;
9339 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9340 em->generation = trans->transid;
9343 write_lock(&em_tree->lock);
9344 ret = add_extent_mapping(em_tree, em, 1);
9345 write_unlock(&em_tree->lock);
9348 btrfs_drop_extent_cache(inode, cur_offset,
9349 cur_offset + ins.offset - 1,
9352 free_extent_map(em);
9354 num_bytes -= ins.offset;
9355 cur_offset += ins.offset;
9356 *alloc_hint = ins.objectid + ins.offset;
9358 inode_inc_iversion(inode);
9359 inode->i_ctime = CURRENT_TIME;
9360 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9361 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9362 (actual_len > inode->i_size) &&
9363 (cur_offset > inode->i_size)) {
9364 if (cur_offset > actual_len)
9365 i_size = actual_len;
9367 i_size = cur_offset;
9368 i_size_write(inode, i_size);
9369 btrfs_ordered_update_i_size(inode, i_size, NULL);
9372 ret = btrfs_update_inode(trans, root, inode);
9375 btrfs_abort_transaction(trans, root, ret);
9377 btrfs_end_transaction(trans, root);
9382 btrfs_end_transaction(trans, root);
9387 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9388 u64 start, u64 num_bytes, u64 min_size,
9389 loff_t actual_len, u64 *alloc_hint)
9391 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9392 min_size, actual_len, alloc_hint,
9396 int btrfs_prealloc_file_range_trans(struct inode *inode,
9397 struct btrfs_trans_handle *trans, int mode,
9398 u64 start, u64 num_bytes, u64 min_size,
9399 loff_t actual_len, u64 *alloc_hint)
9401 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9402 min_size, actual_len, alloc_hint, trans);
9405 static int btrfs_set_page_dirty(struct page *page)
9407 return __set_page_dirty_nobuffers(page);
9410 static int btrfs_permission(struct inode *inode, int mask)
9412 struct btrfs_root *root = BTRFS_I(inode)->root;
9413 umode_t mode = inode->i_mode;
9415 if (mask & MAY_WRITE &&
9416 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9417 if (btrfs_root_readonly(root))
9419 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9422 return generic_permission(inode, mask);
9425 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9427 struct btrfs_trans_handle *trans;
9428 struct btrfs_root *root = BTRFS_I(dir)->root;
9429 struct inode *inode = NULL;
9435 * 5 units required for adding orphan entry
9437 trans = btrfs_start_transaction(root, 5);
9439 return PTR_ERR(trans);
9441 ret = btrfs_find_free_ino(root, &objectid);
9445 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9446 btrfs_ino(dir), objectid, mode, &index);
9447 if (IS_ERR(inode)) {
9448 ret = PTR_ERR(inode);
9453 inode->i_fop = &btrfs_file_operations;
9454 inode->i_op = &btrfs_file_inode_operations;
9456 inode->i_mapping->a_ops = &btrfs_aops;
9457 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9459 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9463 ret = btrfs_update_inode(trans, root, inode);
9466 ret = btrfs_orphan_add(trans, inode);
9471 * We set number of links to 0 in btrfs_new_inode(), and here we set
9472 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9475 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9477 set_nlink(inode, 1);
9478 unlock_new_inode(inode);
9479 d_tmpfile(dentry, inode);
9480 mark_inode_dirty(inode);
9483 btrfs_end_transaction(trans, root);
9486 btrfs_balance_delayed_items(root);
9487 btrfs_btree_balance_dirty(root);
9491 unlock_new_inode(inode);
9496 /* Inspired by filemap_check_errors() */
9497 int btrfs_inode_check_errors(struct inode *inode)
9501 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
9502 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
9504 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
9505 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
9511 static const struct inode_operations btrfs_dir_inode_operations = {
9512 .getattr = btrfs_getattr,
9513 .lookup = btrfs_lookup,
9514 .create = btrfs_create,
9515 .unlink = btrfs_unlink,
9517 .mkdir = btrfs_mkdir,
9518 .rmdir = btrfs_rmdir,
9519 .rename2 = btrfs_rename2,
9520 .symlink = btrfs_symlink,
9521 .setattr = btrfs_setattr,
9522 .mknod = btrfs_mknod,
9523 .setxattr = btrfs_setxattr,
9524 .getxattr = btrfs_getxattr,
9525 .listxattr = btrfs_listxattr,
9526 .removexattr = btrfs_removexattr,
9527 .permission = btrfs_permission,
9528 .get_acl = btrfs_get_acl,
9529 .set_acl = btrfs_set_acl,
9530 .update_time = btrfs_update_time,
9531 .tmpfile = btrfs_tmpfile,
9533 static const struct inode_operations btrfs_dir_ro_inode_operations = {
9534 .lookup = btrfs_lookup,
9535 .permission = btrfs_permission,
9536 .get_acl = btrfs_get_acl,
9537 .set_acl = btrfs_set_acl,
9538 .update_time = btrfs_update_time,
9541 static const struct file_operations btrfs_dir_file_operations = {
9542 .llseek = generic_file_llseek,
9543 .read = generic_read_dir,
9544 .iterate = btrfs_real_readdir,
9545 .unlocked_ioctl = btrfs_ioctl,
9546 #ifdef CONFIG_COMPAT
9547 .compat_ioctl = btrfs_ioctl,
9549 .release = btrfs_release_file,
9550 .fsync = btrfs_sync_file,
9553 static struct extent_io_ops btrfs_extent_io_ops = {
9554 .fill_delalloc = run_delalloc_range,
9555 .submit_bio_hook = btrfs_submit_bio_hook,
9556 .merge_bio_hook = btrfs_merge_bio_hook,
9557 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
9558 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
9559 .writepage_start_hook = btrfs_writepage_start_hook,
9560 .set_bit_hook = btrfs_set_bit_hook,
9561 .clear_bit_hook = btrfs_clear_bit_hook,
9562 .merge_extent_hook = btrfs_merge_extent_hook,
9563 .split_extent_hook = btrfs_split_extent_hook,
9567 * btrfs doesn't support the bmap operation because swapfiles
9568 * use bmap to make a mapping of extents in the file. They assume
9569 * these extents won't change over the life of the file and they
9570 * use the bmap result to do IO directly to the drive.
9572 * the btrfs bmap call would return logical addresses that aren't
9573 * suitable for IO and they also will change frequently as COW
9574 * operations happen. So, swapfile + btrfs == corruption.
9576 * For now we're avoiding this by dropping bmap.
9578 static const struct address_space_operations btrfs_aops = {
9579 .readpage = btrfs_readpage,
9580 .writepage = btrfs_writepage,
9581 .writepages = btrfs_writepages,
9582 .readpages = btrfs_readpages,
9583 .direct_IO = btrfs_direct_IO,
9584 .invalidatepage = btrfs_invalidatepage,
9585 .releasepage = btrfs_releasepage,
9586 .set_page_dirty = btrfs_set_page_dirty,
9587 .error_remove_page = generic_error_remove_page,
9590 static const struct address_space_operations btrfs_symlink_aops = {
9591 .readpage = btrfs_readpage,
9592 .writepage = btrfs_writepage,
9593 .invalidatepage = btrfs_invalidatepage,
9594 .releasepage = btrfs_releasepage,
9597 static const struct inode_operations btrfs_file_inode_operations = {
9598 .getattr = btrfs_getattr,
9599 .setattr = btrfs_setattr,
9600 .setxattr = btrfs_setxattr,
9601 .getxattr = btrfs_getxattr,
9602 .listxattr = btrfs_listxattr,
9603 .removexattr = btrfs_removexattr,
9604 .permission = btrfs_permission,
9605 .fiemap = btrfs_fiemap,
9606 .get_acl = btrfs_get_acl,
9607 .set_acl = btrfs_set_acl,
9608 .update_time = btrfs_update_time,
9610 static const struct inode_operations btrfs_special_inode_operations = {
9611 .getattr = btrfs_getattr,
9612 .setattr = btrfs_setattr,
9613 .permission = btrfs_permission,
9614 .setxattr = btrfs_setxattr,
9615 .getxattr = btrfs_getxattr,
9616 .listxattr = btrfs_listxattr,
9617 .removexattr = btrfs_removexattr,
9618 .get_acl = btrfs_get_acl,
9619 .set_acl = btrfs_set_acl,
9620 .update_time = btrfs_update_time,
9622 static const struct inode_operations btrfs_symlink_inode_operations = {
9623 .readlink = generic_readlink,
9624 .follow_link = page_follow_link_light,
9625 .put_link = page_put_link,
9626 .getattr = btrfs_getattr,
9627 .setattr = btrfs_setattr,
9628 .permission = btrfs_permission,
9629 .setxattr = btrfs_setxattr,
9630 .getxattr = btrfs_getxattr,
9631 .listxattr = btrfs_listxattr,
9632 .removexattr = btrfs_removexattr,
9633 .update_time = btrfs_update_time,
9636 const struct dentry_operations btrfs_dentry_operations = {
9637 .d_delete = btrfs_dentry_delete,
9638 .d_release = btrfs_dentry_release,