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>
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"
61 struct btrfs_iget_args {
63 struct btrfs_root *root;
66 static const struct inode_operations btrfs_dir_inode_operations;
67 static const struct inode_operations btrfs_symlink_inode_operations;
68 static const struct inode_operations btrfs_dir_ro_inode_operations;
69 static const struct inode_operations btrfs_special_inode_operations;
70 static const struct inode_operations btrfs_file_inode_operations;
71 static const struct address_space_operations btrfs_aops;
72 static const struct address_space_operations btrfs_symlink_aops;
73 static const struct file_operations btrfs_dir_file_operations;
74 static struct extent_io_ops btrfs_extent_io_ops;
76 static struct kmem_cache *btrfs_inode_cachep;
77 static struct kmem_cache *btrfs_delalloc_work_cachep;
78 struct kmem_cache *btrfs_trans_handle_cachep;
79 struct kmem_cache *btrfs_transaction_cachep;
80 struct kmem_cache *btrfs_path_cachep;
81 struct kmem_cache *btrfs_free_space_cachep;
84 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
85 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
86 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
87 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
88 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
89 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
90 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
91 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
94 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
95 static int btrfs_truncate(struct inode *inode);
96 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
97 static noinline int cow_file_range(struct inode *inode,
98 struct page *locked_page,
99 u64 start, u64 end, int *page_started,
100 unsigned long *nr_written, int unlock);
101 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
102 u64 len, u64 orig_start,
103 u64 block_start, u64 block_len,
104 u64 orig_block_len, u64 ram_bytes,
107 static int btrfs_dirty_inode(struct inode *inode);
109 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
110 struct inode *inode, struct inode *dir,
111 const struct qstr *qstr)
115 err = btrfs_init_acl(trans, inode, dir);
117 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
122 * this does all the hard work for inserting an inline extent into
123 * the btree. The caller should have done a btrfs_drop_extents so that
124 * no overlapping inline items exist in the btree
126 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
127 struct btrfs_root *root, struct inode *inode,
128 u64 start, size_t size, size_t compressed_size,
130 struct page **compressed_pages)
132 struct btrfs_key key;
133 struct btrfs_path *path;
134 struct extent_buffer *leaf;
135 struct page *page = NULL;
138 struct btrfs_file_extent_item *ei;
141 size_t cur_size = size;
143 unsigned long offset;
145 if (compressed_size && compressed_pages)
146 cur_size = compressed_size;
148 path = btrfs_alloc_path();
152 path->leave_spinning = 1;
154 key.objectid = btrfs_ino(inode);
156 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
157 datasize = btrfs_file_extent_calc_inline_size(cur_size);
159 inode_add_bytes(inode, size);
160 ret = btrfs_insert_empty_item(trans, root, path, &key,
166 leaf = path->nodes[0];
167 ei = btrfs_item_ptr(leaf, path->slots[0],
168 struct btrfs_file_extent_item);
169 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
170 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
171 btrfs_set_file_extent_encryption(leaf, ei, 0);
172 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
173 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
174 ptr = btrfs_file_extent_inline_start(ei);
176 if (compress_type != BTRFS_COMPRESS_NONE) {
179 while (compressed_size > 0) {
180 cpage = compressed_pages[i];
181 cur_size = min_t(unsigned long, compressed_size,
184 kaddr = kmap_atomic(cpage);
185 write_extent_buffer(leaf, kaddr, ptr, cur_size);
186 kunmap_atomic(kaddr);
190 compressed_size -= cur_size;
192 btrfs_set_file_extent_compression(leaf, ei,
195 page = find_get_page(inode->i_mapping,
196 start >> PAGE_CACHE_SHIFT);
197 btrfs_set_file_extent_compression(leaf, ei, 0);
198 kaddr = kmap_atomic(page);
199 offset = start & (PAGE_CACHE_SIZE - 1);
200 write_extent_buffer(leaf, kaddr + offset, ptr, size);
201 kunmap_atomic(kaddr);
202 page_cache_release(page);
204 btrfs_mark_buffer_dirty(leaf);
205 btrfs_free_path(path);
208 * we're an inline extent, so nobody can
209 * extend the file past i_size without locking
210 * a page we already have locked.
212 * We must do any isize and inode updates
213 * before we unlock the pages. Otherwise we
214 * could end up racing with unlink.
216 BTRFS_I(inode)->disk_i_size = inode->i_size;
217 ret = btrfs_update_inode(trans, root, inode);
221 btrfs_free_path(path);
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_trans_handle *trans,
232 struct btrfs_root *root,
233 struct inode *inode, u64 start, u64 end,
234 size_t compressed_size, int compress_type,
235 struct page **compressed_pages)
237 u64 isize = i_size_read(inode);
238 u64 actual_end = min(end + 1, isize);
239 u64 inline_len = actual_end - start;
240 u64 aligned_end = ALIGN(end, root->sectorsize);
241 u64 data_len = inline_len;
245 data_len = compressed_size;
248 actual_end >= PAGE_CACHE_SIZE ||
249 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
251 (actual_end & (root->sectorsize - 1)) == 0) ||
253 data_len > root->fs_info->max_inline) {
257 ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
261 if (isize > actual_end)
262 inline_len = min_t(u64, isize, actual_end);
263 ret = insert_inline_extent(trans, root, inode, start,
264 inline_len, compressed_size,
265 compress_type, compressed_pages);
266 if (ret && ret != -ENOSPC) {
267 btrfs_abort_transaction(trans, root, ret);
269 } else if (ret == -ENOSPC) {
273 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
274 btrfs_delalloc_release_metadata(inode, end + 1 - start);
275 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
279 struct async_extent {
284 unsigned long nr_pages;
286 struct list_head list;
291 struct btrfs_root *root;
292 struct page *locked_page;
295 struct list_head extents;
296 struct btrfs_work work;
299 static noinline int add_async_extent(struct async_cow *cow,
300 u64 start, u64 ram_size,
303 unsigned long nr_pages,
306 struct async_extent *async_extent;
308 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
309 BUG_ON(!async_extent); /* -ENOMEM */
310 async_extent->start = start;
311 async_extent->ram_size = ram_size;
312 async_extent->compressed_size = compressed_size;
313 async_extent->pages = pages;
314 async_extent->nr_pages = nr_pages;
315 async_extent->compress_type = compress_type;
316 list_add_tail(&async_extent->list, &cow->extents);
321 * we create compressed extents in two phases. The first
322 * phase compresses a range of pages that have already been
323 * locked (both pages and state bits are locked).
325 * This is done inside an ordered work queue, and the compression
326 * is spread across many cpus. The actual IO submission is step
327 * two, and the ordered work queue takes care of making sure that
328 * happens in the same order things were put onto the queue by
329 * writepages and friends.
331 * If this code finds it can't get good compression, it puts an
332 * entry onto the work queue to write the uncompressed bytes. This
333 * makes sure that both compressed inodes and uncompressed inodes
334 * are written in the same order that the flusher thread sent them
337 static noinline int compress_file_range(struct inode *inode,
338 struct page *locked_page,
340 struct async_cow *async_cow,
343 struct btrfs_root *root = BTRFS_I(inode)->root;
344 struct btrfs_trans_handle *trans;
346 u64 blocksize = root->sectorsize;
348 u64 isize = i_size_read(inode);
350 struct page **pages = NULL;
351 unsigned long nr_pages;
352 unsigned long nr_pages_ret = 0;
353 unsigned long total_compressed = 0;
354 unsigned long total_in = 0;
355 unsigned long max_compressed = 128 * 1024;
356 unsigned long max_uncompressed = 128 * 1024;
359 int compress_type = root->fs_info->compress_type;
362 /* if this is a small write inside eof, kick off a defrag */
363 if ((end - start + 1) < 16 * 1024 &&
364 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
365 btrfs_add_inode_defrag(NULL, inode);
367 actual_end = min_t(u64, isize, end + 1);
370 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
371 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
374 * we don't want to send crud past the end of i_size through
375 * compression, that's just a waste of CPU time. So, if the
376 * end of the file is before the start of our current
377 * requested range of bytes, we bail out to the uncompressed
378 * cleanup code that can deal with all of this.
380 * It isn't really the fastest way to fix things, but this is a
381 * very uncommon corner.
383 if (actual_end <= start)
384 goto cleanup_and_bail_uncompressed;
386 total_compressed = actual_end - start;
388 /* we want to make sure that amount of ram required to uncompress
389 * an extent is reasonable, so we limit the total size in ram
390 * of a compressed extent to 128k. This is a crucial number
391 * because it also controls how easily we can spread reads across
392 * cpus for decompression.
394 * We also want to make sure the amount of IO required to do
395 * a random read is reasonably small, so we limit the size of
396 * a compressed extent to 128k.
398 total_compressed = min(total_compressed, max_uncompressed);
399 num_bytes = ALIGN(end - start + 1, blocksize);
400 num_bytes = max(blocksize, num_bytes);
405 * we do compression for mount -o compress and when the
406 * inode has not been flagged as nocompress. This flag can
407 * change at any time if we discover bad compression ratios.
409 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
410 (btrfs_test_opt(root, COMPRESS) ||
411 (BTRFS_I(inode)->force_compress) ||
412 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
414 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
416 /* just bail out to the uncompressed code */
420 if (BTRFS_I(inode)->force_compress)
421 compress_type = BTRFS_I(inode)->force_compress;
424 * we need to call clear_page_dirty_for_io on each
425 * page in the range. Otherwise applications with the file
426 * mmap'd can wander in and change the page contents while
427 * we are compressing them.
429 * If the compression fails for any reason, we set the pages
430 * dirty again later on.
432 extent_range_clear_dirty_for_io(inode, start, end);
434 ret = btrfs_compress_pages(compress_type,
435 inode->i_mapping, start,
436 total_compressed, pages,
437 nr_pages, &nr_pages_ret,
443 unsigned long offset = total_compressed &
444 (PAGE_CACHE_SIZE - 1);
445 struct page *page = pages[nr_pages_ret - 1];
448 /* zero the tail end of the last page, we might be
449 * sending it down to disk
452 kaddr = kmap_atomic(page);
453 memset(kaddr + offset, 0,
454 PAGE_CACHE_SIZE - offset);
455 kunmap_atomic(kaddr);
462 trans = btrfs_join_transaction(root);
464 ret = PTR_ERR(trans);
466 goto cleanup_and_out;
468 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
470 /* lets try to make an inline extent */
471 if (ret || total_in < (actual_end - start)) {
472 /* we didn't compress the entire range, try
473 * to make an uncompressed inline extent.
475 ret = cow_file_range_inline(trans, root, inode,
476 start, end, 0, 0, NULL);
478 /* try making a compressed inline extent */
479 ret = cow_file_range_inline(trans, root, inode,
482 compress_type, pages);
486 * inline extent creation worked or returned error,
487 * we don't need to create any more async work items.
488 * Unlock and free up our temp pages.
490 extent_clear_unlock_delalloc(inode,
491 &BTRFS_I(inode)->io_tree,
493 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
494 EXTENT_CLEAR_DELALLOC |
495 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
497 btrfs_end_transaction(trans, root);
500 btrfs_end_transaction(trans, root);
505 * we aren't doing an inline extent round the compressed size
506 * up to a block size boundary so the allocator does sane
509 total_compressed = ALIGN(total_compressed, blocksize);
512 * one last check to make sure the compression is really a
513 * win, compare the page count read with the blocks on disk
515 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
516 if (total_compressed >= total_in) {
519 num_bytes = total_in;
522 if (!will_compress && pages) {
524 * the compression code ran but failed to make things smaller,
525 * free any pages it allocated and our page pointer array
527 for (i = 0; i < nr_pages_ret; i++) {
528 WARN_ON(pages[i]->mapping);
529 page_cache_release(pages[i]);
533 total_compressed = 0;
536 /* flag the file so we don't compress in the future */
537 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
538 !(BTRFS_I(inode)->force_compress)) {
539 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
545 /* the async work queues will take care of doing actual
546 * allocation on disk for these compressed pages,
547 * and will submit them to the elevator.
549 add_async_extent(async_cow, start, num_bytes,
550 total_compressed, pages, nr_pages_ret,
553 if (start + num_bytes < end) {
560 cleanup_and_bail_uncompressed:
562 * No compression, but we still need to write the pages in
563 * the file we've been given so far. redirty the locked
564 * page if it corresponds to our extent and set things up
565 * for the async work queue to run cow_file_range to do
566 * the normal delalloc dance
568 if (page_offset(locked_page) >= start &&
569 page_offset(locked_page) <= end) {
570 __set_page_dirty_nobuffers(locked_page);
571 /* unlocked later on in the async handlers */
574 extent_range_redirty_for_io(inode, start, end);
575 add_async_extent(async_cow, start, end - start + 1,
576 0, NULL, 0, BTRFS_COMPRESS_NONE);
584 for (i = 0; i < nr_pages_ret; i++) {
585 WARN_ON(pages[i]->mapping);
586 page_cache_release(pages[i]);
593 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
595 EXTENT_CLEAR_UNLOCK_PAGE |
597 EXTENT_CLEAR_DELALLOC |
598 EXTENT_SET_WRITEBACK |
599 EXTENT_END_WRITEBACK);
600 if (!trans || IS_ERR(trans))
601 btrfs_error(root->fs_info, ret, "Failed to join transaction");
603 btrfs_abort_transaction(trans, root, ret);
608 * phase two of compressed writeback. This is the ordered portion
609 * of the code, which only gets called in the order the work was
610 * queued. We walk all the async extents created by compress_file_range
611 * and send them down to the disk.
613 static noinline int submit_compressed_extents(struct inode *inode,
614 struct async_cow *async_cow)
616 struct async_extent *async_extent;
618 struct btrfs_trans_handle *trans;
619 struct btrfs_key ins;
620 struct extent_map *em;
621 struct btrfs_root *root = BTRFS_I(inode)->root;
622 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
623 struct extent_io_tree *io_tree;
626 if (list_empty(&async_cow->extents))
630 while (!list_empty(&async_cow->extents)) {
631 async_extent = list_entry(async_cow->extents.next,
632 struct async_extent, list);
633 list_del(&async_extent->list);
635 io_tree = &BTRFS_I(inode)->io_tree;
638 /* did the compression code fall back to uncompressed IO? */
639 if (!async_extent->pages) {
640 int page_started = 0;
641 unsigned long nr_written = 0;
643 lock_extent(io_tree, async_extent->start,
644 async_extent->start +
645 async_extent->ram_size - 1);
647 /* allocate blocks */
648 ret = cow_file_range(inode, async_cow->locked_page,
650 async_extent->start +
651 async_extent->ram_size - 1,
652 &page_started, &nr_written, 0);
657 * if page_started, cow_file_range inserted an
658 * inline extent and took care of all the unlocking
659 * and IO for us. Otherwise, we need to submit
660 * all those pages down to the drive.
662 if (!page_started && !ret)
663 extent_write_locked_range(io_tree,
664 inode, async_extent->start,
665 async_extent->start +
666 async_extent->ram_size - 1,
670 unlock_page(async_cow->locked_page);
676 lock_extent(io_tree, async_extent->start,
677 async_extent->start + async_extent->ram_size - 1);
679 trans = btrfs_join_transaction(root);
681 ret = PTR_ERR(trans);
683 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
684 ret = btrfs_reserve_extent(trans, root,
685 async_extent->compressed_size,
686 async_extent->compressed_size,
687 0, alloc_hint, &ins, 1);
688 if (ret && ret != -ENOSPC)
689 btrfs_abort_transaction(trans, root, ret);
690 btrfs_end_transaction(trans, root);
696 for (i = 0; i < async_extent->nr_pages; i++) {
697 WARN_ON(async_extent->pages[i]->mapping);
698 page_cache_release(async_extent->pages[i]);
700 kfree(async_extent->pages);
701 async_extent->nr_pages = 0;
702 async_extent->pages = NULL;
710 * here we're doing allocation and writeback of the
713 btrfs_drop_extent_cache(inode, async_extent->start,
714 async_extent->start +
715 async_extent->ram_size - 1, 0);
717 em = alloc_extent_map();
720 goto out_free_reserve;
722 em->start = async_extent->start;
723 em->len = async_extent->ram_size;
724 em->orig_start = em->start;
725 em->mod_start = em->start;
726 em->mod_len = em->len;
728 em->block_start = ins.objectid;
729 em->block_len = ins.offset;
730 em->orig_block_len = ins.offset;
731 em->ram_bytes = async_extent->ram_size;
732 em->bdev = root->fs_info->fs_devices->latest_bdev;
733 em->compress_type = async_extent->compress_type;
734 set_bit(EXTENT_FLAG_PINNED, &em->flags);
735 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
739 write_lock(&em_tree->lock);
740 ret = add_extent_mapping(em_tree, em, 1);
741 write_unlock(&em_tree->lock);
742 if (ret != -EEXIST) {
746 btrfs_drop_extent_cache(inode, async_extent->start,
747 async_extent->start +
748 async_extent->ram_size - 1, 0);
752 goto out_free_reserve;
754 ret = btrfs_add_ordered_extent_compress(inode,
757 async_extent->ram_size,
759 BTRFS_ORDERED_COMPRESSED,
760 async_extent->compress_type);
762 goto out_free_reserve;
765 * clear dirty, set writeback and unlock the pages.
767 extent_clear_unlock_delalloc(inode,
768 &BTRFS_I(inode)->io_tree,
770 async_extent->start +
771 async_extent->ram_size - 1,
772 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
773 EXTENT_CLEAR_UNLOCK |
774 EXTENT_CLEAR_DELALLOC |
775 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
777 ret = btrfs_submit_compressed_write(inode,
779 async_extent->ram_size,
781 ins.offset, async_extent->pages,
782 async_extent->nr_pages);
783 alloc_hint = ins.objectid + ins.offset;
793 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
795 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
797 async_extent->start +
798 async_extent->ram_size - 1,
799 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
800 EXTENT_CLEAR_UNLOCK |
801 EXTENT_CLEAR_DELALLOC |
803 EXTENT_SET_WRITEBACK |
804 EXTENT_END_WRITEBACK);
809 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
812 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
813 struct extent_map *em;
816 read_lock(&em_tree->lock);
817 em = search_extent_mapping(em_tree, start, num_bytes);
820 * if block start isn't an actual block number then find the
821 * first block in this inode and use that as a hint. If that
822 * block is also bogus then just don't worry about it.
824 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
826 em = search_extent_mapping(em_tree, 0, 0);
827 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
828 alloc_hint = em->block_start;
832 alloc_hint = em->block_start;
836 read_unlock(&em_tree->lock);
842 * when extent_io.c finds a delayed allocation range in the file,
843 * the call backs end up in this code. The basic idea is to
844 * allocate extents on disk for the range, and create ordered data structs
845 * in ram to track those extents.
847 * locked_page is the page that writepage had locked already. We use
848 * it to make sure we don't do extra locks or unlocks.
850 * *page_started is set to one if we unlock locked_page and do everything
851 * required to start IO on it. It may be clean and already done with
854 static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
856 struct btrfs_root *root,
857 struct page *locked_page,
858 u64 start, u64 end, int *page_started,
859 unsigned long *nr_written,
864 unsigned long ram_size;
867 u64 blocksize = root->sectorsize;
868 struct btrfs_key ins;
869 struct extent_map *em;
870 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
873 BUG_ON(btrfs_is_free_space_inode(inode));
875 num_bytes = ALIGN(end - start + 1, blocksize);
876 num_bytes = max(blocksize, num_bytes);
877 disk_num_bytes = num_bytes;
879 /* if this is a small write inside eof, kick off defrag */
880 if (num_bytes < 64 * 1024 &&
881 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
882 btrfs_add_inode_defrag(trans, inode);
885 /* lets try to make an inline extent */
886 ret = cow_file_range_inline(trans, root, inode,
887 start, end, 0, 0, NULL);
889 extent_clear_unlock_delalloc(inode,
890 &BTRFS_I(inode)->io_tree,
892 EXTENT_CLEAR_UNLOCK_PAGE |
893 EXTENT_CLEAR_UNLOCK |
894 EXTENT_CLEAR_DELALLOC |
896 EXTENT_SET_WRITEBACK |
897 EXTENT_END_WRITEBACK);
899 *nr_written = *nr_written +
900 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
903 } else if (ret < 0) {
904 btrfs_abort_transaction(trans, root, ret);
909 BUG_ON(disk_num_bytes >
910 btrfs_super_total_bytes(root->fs_info->super_copy));
912 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
913 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
915 while (disk_num_bytes > 0) {
918 cur_alloc_size = disk_num_bytes;
919 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
920 root->sectorsize, 0, alloc_hint,
923 btrfs_abort_transaction(trans, root, ret);
927 em = alloc_extent_map();
933 em->orig_start = em->start;
934 ram_size = ins.offset;
935 em->len = ins.offset;
936 em->mod_start = em->start;
937 em->mod_len = em->len;
939 em->block_start = ins.objectid;
940 em->block_len = ins.offset;
941 em->orig_block_len = ins.offset;
942 em->ram_bytes = ram_size;
943 em->bdev = root->fs_info->fs_devices->latest_bdev;
944 set_bit(EXTENT_FLAG_PINNED, &em->flags);
948 write_lock(&em_tree->lock);
949 ret = add_extent_mapping(em_tree, em, 1);
950 write_unlock(&em_tree->lock);
951 if (ret != -EEXIST) {
955 btrfs_drop_extent_cache(inode, start,
956 start + ram_size - 1, 0);
961 cur_alloc_size = ins.offset;
962 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
963 ram_size, cur_alloc_size, 0);
967 if (root->root_key.objectid ==
968 BTRFS_DATA_RELOC_TREE_OBJECTID) {
969 ret = btrfs_reloc_clone_csums(inode, start,
972 btrfs_abort_transaction(trans, root, ret);
977 if (disk_num_bytes < cur_alloc_size)
980 /* we're not doing compressed IO, don't unlock the first
981 * page (which the caller expects to stay locked), don't
982 * clear any dirty bits and don't set any writeback bits
984 * Do set the Private2 bit so we know this page was properly
985 * setup for writepage
987 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
988 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
991 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
992 start, start + ram_size - 1,
994 disk_num_bytes -= cur_alloc_size;
995 num_bytes -= cur_alloc_size;
996 alloc_hint = ins.objectid + ins.offset;
997 start += cur_alloc_size;
1003 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
1005 extent_clear_unlock_delalloc(inode,
1006 &BTRFS_I(inode)->io_tree,
1007 start, end, locked_page,
1008 EXTENT_CLEAR_UNLOCK_PAGE |
1009 EXTENT_CLEAR_UNLOCK |
1010 EXTENT_CLEAR_DELALLOC |
1011 EXTENT_CLEAR_DIRTY |
1012 EXTENT_SET_WRITEBACK |
1013 EXTENT_END_WRITEBACK);
1018 static noinline int cow_file_range(struct inode *inode,
1019 struct page *locked_page,
1020 u64 start, u64 end, int *page_started,
1021 unsigned long *nr_written,
1024 struct btrfs_trans_handle *trans;
1025 struct btrfs_root *root = BTRFS_I(inode)->root;
1028 trans = btrfs_join_transaction(root);
1029 if (IS_ERR(trans)) {
1030 extent_clear_unlock_delalloc(inode,
1031 &BTRFS_I(inode)->io_tree,
1032 start, end, locked_page,
1033 EXTENT_CLEAR_UNLOCK_PAGE |
1034 EXTENT_CLEAR_UNLOCK |
1035 EXTENT_CLEAR_DELALLOC |
1036 EXTENT_CLEAR_DIRTY |
1037 EXTENT_SET_WRITEBACK |
1038 EXTENT_END_WRITEBACK);
1039 return PTR_ERR(trans);
1041 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1043 ret = __cow_file_range(trans, inode, root, locked_page, start, end,
1044 page_started, nr_written, unlock);
1046 btrfs_end_transaction(trans, root);
1052 * work queue call back to started compression on a file and pages
1054 static noinline void async_cow_start(struct btrfs_work *work)
1056 struct async_cow *async_cow;
1058 async_cow = container_of(work, struct async_cow, work);
1060 compress_file_range(async_cow->inode, async_cow->locked_page,
1061 async_cow->start, async_cow->end, async_cow,
1063 if (num_added == 0) {
1064 btrfs_add_delayed_iput(async_cow->inode);
1065 async_cow->inode = NULL;
1070 * work queue call back to submit previously compressed pages
1072 static noinline void async_cow_submit(struct btrfs_work *work)
1074 struct async_cow *async_cow;
1075 struct btrfs_root *root;
1076 unsigned long nr_pages;
1078 async_cow = container_of(work, struct async_cow, work);
1080 root = async_cow->root;
1081 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1084 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1086 waitqueue_active(&root->fs_info->async_submit_wait))
1087 wake_up(&root->fs_info->async_submit_wait);
1089 if (async_cow->inode)
1090 submit_compressed_extents(async_cow->inode, async_cow);
1093 static noinline void async_cow_free(struct btrfs_work *work)
1095 struct async_cow *async_cow;
1096 async_cow = container_of(work, struct async_cow, work);
1097 if (async_cow->inode)
1098 btrfs_add_delayed_iput(async_cow->inode);
1102 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1103 u64 start, u64 end, int *page_started,
1104 unsigned long *nr_written)
1106 struct async_cow *async_cow;
1107 struct btrfs_root *root = BTRFS_I(inode)->root;
1108 unsigned long nr_pages;
1110 int limit = 10 * 1024 * 1024;
1112 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1113 1, 0, NULL, GFP_NOFS);
1114 while (start < end) {
1115 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1116 BUG_ON(!async_cow); /* -ENOMEM */
1117 async_cow->inode = igrab(inode);
1118 async_cow->root = root;
1119 async_cow->locked_page = locked_page;
1120 async_cow->start = start;
1122 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1125 cur_end = min(end, start + 512 * 1024 - 1);
1127 async_cow->end = cur_end;
1128 INIT_LIST_HEAD(&async_cow->extents);
1130 async_cow->work.func = async_cow_start;
1131 async_cow->work.ordered_func = async_cow_submit;
1132 async_cow->work.ordered_free = async_cow_free;
1133 async_cow->work.flags = 0;
1135 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1137 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1139 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1142 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1143 wait_event(root->fs_info->async_submit_wait,
1144 (atomic_read(&root->fs_info->async_delalloc_pages) <
1148 while (atomic_read(&root->fs_info->async_submit_draining) &&
1149 atomic_read(&root->fs_info->async_delalloc_pages)) {
1150 wait_event(root->fs_info->async_submit_wait,
1151 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1155 *nr_written += nr_pages;
1156 start = cur_end + 1;
1162 static noinline int csum_exist_in_range(struct btrfs_root *root,
1163 u64 bytenr, u64 num_bytes)
1166 struct btrfs_ordered_sum *sums;
1169 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1170 bytenr + num_bytes - 1, &list, 0);
1171 if (ret == 0 && list_empty(&list))
1174 while (!list_empty(&list)) {
1175 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1176 list_del(&sums->list);
1183 * when nowcow writeback call back. This checks for snapshots or COW copies
1184 * of the extents that exist in the file, and COWs the file as required.
1186 * If no cow copies or snapshots exist, we write directly to the existing
1189 static noinline int run_delalloc_nocow(struct inode *inode,
1190 struct page *locked_page,
1191 u64 start, u64 end, int *page_started, int force,
1192 unsigned long *nr_written)
1194 struct btrfs_root *root = BTRFS_I(inode)->root;
1195 struct btrfs_trans_handle *trans;
1196 struct extent_buffer *leaf;
1197 struct btrfs_path *path;
1198 struct btrfs_file_extent_item *fi;
1199 struct btrfs_key found_key;
1214 u64 ino = btrfs_ino(inode);
1216 path = btrfs_alloc_path();
1218 extent_clear_unlock_delalloc(inode,
1219 &BTRFS_I(inode)->io_tree,
1220 start, end, locked_page,
1221 EXTENT_CLEAR_UNLOCK_PAGE |
1222 EXTENT_CLEAR_UNLOCK |
1223 EXTENT_CLEAR_DELALLOC |
1224 EXTENT_CLEAR_DIRTY |
1225 EXTENT_SET_WRITEBACK |
1226 EXTENT_END_WRITEBACK);
1230 nolock = btrfs_is_free_space_inode(inode);
1233 trans = btrfs_join_transaction_nolock(root);
1235 trans = btrfs_join_transaction(root);
1237 if (IS_ERR(trans)) {
1238 extent_clear_unlock_delalloc(inode,
1239 &BTRFS_I(inode)->io_tree,
1240 start, end, locked_page,
1241 EXTENT_CLEAR_UNLOCK_PAGE |
1242 EXTENT_CLEAR_UNLOCK |
1243 EXTENT_CLEAR_DELALLOC |
1244 EXTENT_CLEAR_DIRTY |
1245 EXTENT_SET_WRITEBACK |
1246 EXTENT_END_WRITEBACK);
1247 btrfs_free_path(path);
1248 return PTR_ERR(trans);
1251 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1253 cow_start = (u64)-1;
1256 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1259 btrfs_abort_transaction(trans, root, ret);
1262 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1263 leaf = path->nodes[0];
1264 btrfs_item_key_to_cpu(leaf, &found_key,
1265 path->slots[0] - 1);
1266 if (found_key.objectid == ino &&
1267 found_key.type == BTRFS_EXTENT_DATA_KEY)
1272 leaf = path->nodes[0];
1273 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1274 ret = btrfs_next_leaf(root, path);
1276 btrfs_abort_transaction(trans, root, ret);
1281 leaf = path->nodes[0];
1287 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1289 if (found_key.objectid > ino ||
1290 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1291 found_key.offset > end)
1294 if (found_key.offset > cur_offset) {
1295 extent_end = found_key.offset;
1300 fi = btrfs_item_ptr(leaf, path->slots[0],
1301 struct btrfs_file_extent_item);
1302 extent_type = btrfs_file_extent_type(leaf, fi);
1304 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1305 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1306 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1307 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1308 extent_offset = btrfs_file_extent_offset(leaf, fi);
1309 extent_end = found_key.offset +
1310 btrfs_file_extent_num_bytes(leaf, fi);
1312 btrfs_file_extent_disk_num_bytes(leaf, fi);
1313 if (extent_end <= start) {
1317 if (disk_bytenr == 0)
1319 if (btrfs_file_extent_compression(leaf, fi) ||
1320 btrfs_file_extent_encryption(leaf, fi) ||
1321 btrfs_file_extent_other_encoding(leaf, fi))
1323 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1325 if (btrfs_extent_readonly(root, disk_bytenr))
1327 if (btrfs_cross_ref_exist(trans, root, ino,
1329 extent_offset, disk_bytenr))
1331 disk_bytenr += extent_offset;
1332 disk_bytenr += cur_offset - found_key.offset;
1333 num_bytes = min(end + 1, extent_end) - cur_offset;
1335 * force cow if csum exists in the range.
1336 * this ensure that csum for a given extent are
1337 * either valid or do not exist.
1339 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1342 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1343 extent_end = found_key.offset +
1344 btrfs_file_extent_inline_len(leaf, fi);
1345 extent_end = ALIGN(extent_end, root->sectorsize);
1350 if (extent_end <= start) {
1355 if (cow_start == (u64)-1)
1356 cow_start = cur_offset;
1357 cur_offset = extent_end;
1358 if (cur_offset > end)
1364 btrfs_release_path(path);
1365 if (cow_start != (u64)-1) {
1366 ret = __cow_file_range(trans, inode, root, locked_page,
1367 cow_start, found_key.offset - 1,
1368 page_started, nr_written, 1);
1370 btrfs_abort_transaction(trans, root, ret);
1373 cow_start = (u64)-1;
1376 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1377 struct extent_map *em;
1378 struct extent_map_tree *em_tree;
1379 em_tree = &BTRFS_I(inode)->extent_tree;
1380 em = alloc_extent_map();
1381 BUG_ON(!em); /* -ENOMEM */
1382 em->start = cur_offset;
1383 em->orig_start = found_key.offset - extent_offset;
1384 em->len = num_bytes;
1385 em->block_len = num_bytes;
1386 em->block_start = disk_bytenr;
1387 em->orig_block_len = disk_num_bytes;
1388 em->ram_bytes = ram_bytes;
1389 em->bdev = root->fs_info->fs_devices->latest_bdev;
1390 em->mod_start = em->start;
1391 em->mod_len = em->len;
1392 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1393 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1394 em->generation = -1;
1396 write_lock(&em_tree->lock);
1397 ret = add_extent_mapping(em_tree, em, 1);
1398 write_unlock(&em_tree->lock);
1399 if (ret != -EEXIST) {
1400 free_extent_map(em);
1403 btrfs_drop_extent_cache(inode, em->start,
1404 em->start + em->len - 1, 0);
1406 type = BTRFS_ORDERED_PREALLOC;
1408 type = BTRFS_ORDERED_NOCOW;
1411 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1412 num_bytes, num_bytes, type);
1413 BUG_ON(ret); /* -ENOMEM */
1415 if (root->root_key.objectid ==
1416 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1417 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1420 btrfs_abort_transaction(trans, root, ret);
1425 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1426 cur_offset, cur_offset + num_bytes - 1,
1427 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1428 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1429 EXTENT_SET_PRIVATE2);
1430 cur_offset = extent_end;
1431 if (cur_offset > end)
1434 btrfs_release_path(path);
1436 if (cur_offset <= end && cow_start == (u64)-1) {
1437 cow_start = cur_offset;
1441 if (cow_start != (u64)-1) {
1442 ret = __cow_file_range(trans, inode, root, locked_page,
1444 page_started, nr_written, 1);
1446 btrfs_abort_transaction(trans, root, ret);
1452 err = btrfs_end_transaction(trans, root);
1456 if (ret && cur_offset < end)
1457 extent_clear_unlock_delalloc(inode,
1458 &BTRFS_I(inode)->io_tree,
1459 cur_offset, end, locked_page,
1460 EXTENT_CLEAR_UNLOCK_PAGE |
1461 EXTENT_CLEAR_UNLOCK |
1462 EXTENT_CLEAR_DELALLOC |
1463 EXTENT_CLEAR_DIRTY |
1464 EXTENT_SET_WRITEBACK |
1465 EXTENT_END_WRITEBACK);
1467 btrfs_free_path(path);
1472 * extent_io.c call back to do delayed allocation processing
1474 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1475 u64 start, u64 end, int *page_started,
1476 unsigned long *nr_written)
1479 struct btrfs_root *root = BTRFS_I(inode)->root;
1481 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1482 ret = run_delalloc_nocow(inode, locked_page, start, end,
1483 page_started, 1, nr_written);
1484 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1485 ret = run_delalloc_nocow(inode, locked_page, start, end,
1486 page_started, 0, nr_written);
1487 } else if (!btrfs_test_opt(root, COMPRESS) &&
1488 !(BTRFS_I(inode)->force_compress) &&
1489 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1490 ret = cow_file_range(inode, locked_page, start, end,
1491 page_started, nr_written, 1);
1493 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1494 &BTRFS_I(inode)->runtime_flags);
1495 ret = cow_file_range_async(inode, locked_page, start, end,
1496 page_started, nr_written);
1501 static void btrfs_split_extent_hook(struct inode *inode,
1502 struct extent_state *orig, u64 split)
1504 /* not delalloc, ignore it */
1505 if (!(orig->state & EXTENT_DELALLOC))
1508 spin_lock(&BTRFS_I(inode)->lock);
1509 BTRFS_I(inode)->outstanding_extents++;
1510 spin_unlock(&BTRFS_I(inode)->lock);
1514 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1515 * extents so we can keep track of new extents that are just merged onto old
1516 * extents, such as when we are doing sequential writes, so we can properly
1517 * account for the metadata space we'll need.
1519 static void btrfs_merge_extent_hook(struct inode *inode,
1520 struct extent_state *new,
1521 struct extent_state *other)
1523 /* not delalloc, ignore it */
1524 if (!(other->state & EXTENT_DELALLOC))
1527 spin_lock(&BTRFS_I(inode)->lock);
1528 BTRFS_I(inode)->outstanding_extents--;
1529 spin_unlock(&BTRFS_I(inode)->lock);
1533 * extent_io.c set_bit_hook, used to track delayed allocation
1534 * bytes in this file, and to maintain the list of inodes that
1535 * have pending delalloc work to be done.
1537 static void btrfs_set_bit_hook(struct inode *inode,
1538 struct extent_state *state, unsigned long *bits)
1542 * set_bit and clear bit hooks normally require _irqsave/restore
1543 * but in this case, we are only testing for the DELALLOC
1544 * bit, which is only set or cleared with irqs on
1546 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1547 struct btrfs_root *root = BTRFS_I(inode)->root;
1548 u64 len = state->end + 1 - state->start;
1549 bool do_list = !btrfs_is_free_space_inode(inode);
1551 if (*bits & EXTENT_FIRST_DELALLOC) {
1552 *bits &= ~EXTENT_FIRST_DELALLOC;
1554 spin_lock(&BTRFS_I(inode)->lock);
1555 BTRFS_I(inode)->outstanding_extents++;
1556 spin_unlock(&BTRFS_I(inode)->lock);
1559 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1560 root->fs_info->delalloc_batch);
1561 spin_lock(&BTRFS_I(inode)->lock);
1562 BTRFS_I(inode)->delalloc_bytes += len;
1563 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1564 &BTRFS_I(inode)->runtime_flags)) {
1565 spin_lock(&root->fs_info->delalloc_lock);
1566 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1567 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1568 &root->fs_info->delalloc_inodes);
1569 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1570 &BTRFS_I(inode)->runtime_flags);
1572 spin_unlock(&root->fs_info->delalloc_lock);
1574 spin_unlock(&BTRFS_I(inode)->lock);
1579 * extent_io.c clear_bit_hook, see set_bit_hook for why
1581 static void btrfs_clear_bit_hook(struct inode *inode,
1582 struct extent_state *state,
1583 unsigned long *bits)
1586 * set_bit and clear bit hooks normally require _irqsave/restore
1587 * but in this case, we are only testing for the DELALLOC
1588 * bit, which is only set or cleared with irqs on
1590 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1591 struct btrfs_root *root = BTRFS_I(inode)->root;
1592 u64 len = state->end + 1 - state->start;
1593 bool do_list = !btrfs_is_free_space_inode(inode);
1595 if (*bits & EXTENT_FIRST_DELALLOC) {
1596 *bits &= ~EXTENT_FIRST_DELALLOC;
1597 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1598 spin_lock(&BTRFS_I(inode)->lock);
1599 BTRFS_I(inode)->outstanding_extents--;
1600 spin_unlock(&BTRFS_I(inode)->lock);
1603 if (*bits & EXTENT_DO_ACCOUNTING)
1604 btrfs_delalloc_release_metadata(inode, len);
1606 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1608 btrfs_free_reserved_data_space(inode, len);
1610 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1611 root->fs_info->delalloc_batch);
1612 spin_lock(&BTRFS_I(inode)->lock);
1613 BTRFS_I(inode)->delalloc_bytes -= len;
1614 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1615 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1616 &BTRFS_I(inode)->runtime_flags)) {
1617 spin_lock(&root->fs_info->delalloc_lock);
1618 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1619 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1620 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1621 &BTRFS_I(inode)->runtime_flags);
1623 spin_unlock(&root->fs_info->delalloc_lock);
1625 spin_unlock(&BTRFS_I(inode)->lock);
1630 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1631 * we don't create bios that span stripes or chunks
1633 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1634 size_t size, struct bio *bio,
1635 unsigned long bio_flags)
1637 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1638 u64 logical = (u64)bio->bi_sector << 9;
1643 if (bio_flags & EXTENT_BIO_COMPRESSED)
1646 length = bio->bi_size;
1647 map_length = length;
1648 ret = btrfs_map_block(root->fs_info, rw, logical,
1649 &map_length, NULL, 0);
1650 /* Will always return 0 with map_multi == NULL */
1652 if (map_length < length + size)
1658 * in order to insert checksums into the metadata in large chunks,
1659 * we wait until bio submission time. All the pages in the bio are
1660 * checksummed and sums are attached onto the ordered extent record.
1662 * At IO completion time the cums attached on the ordered extent record
1663 * are inserted into the btree
1665 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1666 struct bio *bio, int mirror_num,
1667 unsigned long bio_flags,
1670 struct btrfs_root *root = BTRFS_I(inode)->root;
1673 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1674 BUG_ON(ret); /* -ENOMEM */
1679 * in order to insert checksums into the metadata in large chunks,
1680 * we wait until bio submission time. All the pages in the bio are
1681 * checksummed and sums are attached onto the ordered extent record.
1683 * At IO completion time the cums attached on the ordered extent record
1684 * are inserted into the btree
1686 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1687 int mirror_num, unsigned long bio_flags,
1690 struct btrfs_root *root = BTRFS_I(inode)->root;
1693 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1695 bio_endio(bio, ret);
1700 * extent_io.c submission hook. This does the right thing for csum calculation
1701 * on write, or reading the csums from the tree before a read
1703 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1704 int mirror_num, unsigned long bio_flags,
1707 struct btrfs_root *root = BTRFS_I(inode)->root;
1711 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1713 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1715 if (btrfs_is_free_space_inode(inode))
1718 if (!(rw & REQ_WRITE)) {
1719 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1723 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1724 ret = btrfs_submit_compressed_read(inode, bio,
1728 } else if (!skip_sum) {
1729 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1734 } else if (async && !skip_sum) {
1735 /* csum items have already been cloned */
1736 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1738 /* we're doing a write, do the async checksumming */
1739 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1740 inode, rw, bio, mirror_num,
1741 bio_flags, bio_offset,
1742 __btrfs_submit_bio_start,
1743 __btrfs_submit_bio_done);
1745 } else if (!skip_sum) {
1746 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1752 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1756 bio_endio(bio, ret);
1761 * given a list of ordered sums record them in the inode. This happens
1762 * at IO completion time based on sums calculated at bio submission time.
1764 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1765 struct inode *inode, u64 file_offset,
1766 struct list_head *list)
1768 struct btrfs_ordered_sum *sum;
1770 list_for_each_entry(sum, list, list) {
1771 trans->adding_csums = 1;
1772 btrfs_csum_file_blocks(trans,
1773 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1774 trans->adding_csums = 0;
1779 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1780 struct extent_state **cached_state)
1782 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1783 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1784 cached_state, GFP_NOFS);
1787 /* see btrfs_writepage_start_hook for details on why this is required */
1788 struct btrfs_writepage_fixup {
1790 struct btrfs_work work;
1793 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1795 struct btrfs_writepage_fixup *fixup;
1796 struct btrfs_ordered_extent *ordered;
1797 struct extent_state *cached_state = NULL;
1799 struct inode *inode;
1804 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1808 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1809 ClearPageChecked(page);
1813 inode = page->mapping->host;
1814 page_start = page_offset(page);
1815 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1817 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1820 /* already ordered? We're done */
1821 if (PagePrivate2(page))
1824 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1826 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1827 page_end, &cached_state, GFP_NOFS);
1829 btrfs_start_ordered_extent(inode, ordered, 1);
1830 btrfs_put_ordered_extent(ordered);
1834 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1836 mapping_set_error(page->mapping, ret);
1837 end_extent_writepage(page, ret, page_start, page_end);
1838 ClearPageChecked(page);
1842 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1843 ClearPageChecked(page);
1844 set_page_dirty(page);
1846 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1847 &cached_state, GFP_NOFS);
1850 page_cache_release(page);
1855 * There are a few paths in the higher layers of the kernel that directly
1856 * set the page dirty bit without asking the filesystem if it is a
1857 * good idea. This causes problems because we want to make sure COW
1858 * properly happens and the data=ordered rules are followed.
1860 * In our case any range that doesn't have the ORDERED bit set
1861 * hasn't been properly setup for IO. We kick off an async process
1862 * to fix it up. The async helper will wait for ordered extents, set
1863 * the delalloc bit and make it safe to write the page.
1865 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1867 struct inode *inode = page->mapping->host;
1868 struct btrfs_writepage_fixup *fixup;
1869 struct btrfs_root *root = BTRFS_I(inode)->root;
1871 /* this page is properly in the ordered list */
1872 if (TestClearPagePrivate2(page))
1875 if (PageChecked(page))
1878 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1882 SetPageChecked(page);
1883 page_cache_get(page);
1884 fixup->work.func = btrfs_writepage_fixup_worker;
1886 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1890 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1891 struct inode *inode, u64 file_pos,
1892 u64 disk_bytenr, u64 disk_num_bytes,
1893 u64 num_bytes, u64 ram_bytes,
1894 u8 compression, u8 encryption,
1895 u16 other_encoding, int extent_type)
1897 struct btrfs_root *root = BTRFS_I(inode)->root;
1898 struct btrfs_file_extent_item *fi;
1899 struct btrfs_path *path;
1900 struct extent_buffer *leaf;
1901 struct btrfs_key ins;
1904 path = btrfs_alloc_path();
1908 path->leave_spinning = 1;
1911 * we may be replacing one extent in the tree with another.
1912 * The new extent is pinned in the extent map, and we don't want
1913 * to drop it from the cache until it is completely in the btree.
1915 * So, tell btrfs_drop_extents to leave this extent in the cache.
1916 * the caller is expected to unpin it and allow it to be merged
1919 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1920 file_pos + num_bytes, 0);
1924 ins.objectid = btrfs_ino(inode);
1925 ins.offset = file_pos;
1926 ins.type = BTRFS_EXTENT_DATA_KEY;
1927 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1930 leaf = path->nodes[0];
1931 fi = btrfs_item_ptr(leaf, path->slots[0],
1932 struct btrfs_file_extent_item);
1933 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1934 btrfs_set_file_extent_type(leaf, fi, extent_type);
1935 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1936 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1937 btrfs_set_file_extent_offset(leaf, fi, 0);
1938 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1939 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1940 btrfs_set_file_extent_compression(leaf, fi, compression);
1941 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1942 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1944 btrfs_mark_buffer_dirty(leaf);
1945 btrfs_release_path(path);
1947 inode_add_bytes(inode, num_bytes);
1949 ins.objectid = disk_bytenr;
1950 ins.offset = disk_num_bytes;
1951 ins.type = BTRFS_EXTENT_ITEM_KEY;
1952 ret = btrfs_alloc_reserved_file_extent(trans, root,
1953 root->root_key.objectid,
1954 btrfs_ino(inode), file_pos, &ins);
1956 btrfs_free_path(path);
1961 /* snapshot-aware defrag */
1962 struct sa_defrag_extent_backref {
1963 struct rb_node node;
1964 struct old_sa_defrag_extent *old;
1973 struct old_sa_defrag_extent {
1974 struct list_head list;
1975 struct new_sa_defrag_extent *new;
1984 struct new_sa_defrag_extent {
1985 struct rb_root root;
1986 struct list_head head;
1987 struct btrfs_path *path;
1988 struct inode *inode;
1996 static int backref_comp(struct sa_defrag_extent_backref *b1,
1997 struct sa_defrag_extent_backref *b2)
1999 if (b1->root_id < b2->root_id)
2001 else if (b1->root_id > b2->root_id)
2004 if (b1->inum < b2->inum)
2006 else if (b1->inum > b2->inum)
2009 if (b1->file_pos < b2->file_pos)
2011 else if (b1->file_pos > b2->file_pos)
2015 * [------------------------------] ===> (a range of space)
2016 * |<--->| |<---->| =============> (fs/file tree A)
2017 * |<---------------------------->| ===> (fs/file tree B)
2019 * A range of space can refer to two file extents in one tree while
2020 * refer to only one file extent in another tree.
2022 * So we may process a disk offset more than one time(two extents in A)
2023 * and locate at the same extent(one extent in B), then insert two same
2024 * backrefs(both refer to the extent in B).
2029 static void backref_insert(struct rb_root *root,
2030 struct sa_defrag_extent_backref *backref)
2032 struct rb_node **p = &root->rb_node;
2033 struct rb_node *parent = NULL;
2034 struct sa_defrag_extent_backref *entry;
2039 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2041 ret = backref_comp(backref, entry);
2045 p = &(*p)->rb_right;
2048 rb_link_node(&backref->node, parent, p);
2049 rb_insert_color(&backref->node, root);
2053 * Note the backref might has changed, and in this case we just return 0.
2055 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2058 struct btrfs_file_extent_item *extent;
2059 struct btrfs_fs_info *fs_info;
2060 struct old_sa_defrag_extent *old = ctx;
2061 struct new_sa_defrag_extent *new = old->new;
2062 struct btrfs_path *path = new->path;
2063 struct btrfs_key key;
2064 struct btrfs_root *root;
2065 struct sa_defrag_extent_backref *backref;
2066 struct extent_buffer *leaf;
2067 struct inode *inode = new->inode;
2073 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2074 inum == btrfs_ino(inode))
2077 key.objectid = root_id;
2078 key.type = BTRFS_ROOT_ITEM_KEY;
2079 key.offset = (u64)-1;
2081 fs_info = BTRFS_I(inode)->root->fs_info;
2082 root = btrfs_read_fs_root_no_name(fs_info, &key);
2084 if (PTR_ERR(root) == -ENOENT)
2087 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2088 inum, offset, root_id);
2089 return PTR_ERR(root);
2092 key.objectid = inum;
2093 key.type = BTRFS_EXTENT_DATA_KEY;
2094 if (offset > (u64)-1 << 32)
2097 key.offset = offset;
2099 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2108 leaf = path->nodes[0];
2109 slot = path->slots[0];
2111 if (slot >= btrfs_header_nritems(leaf)) {
2112 ret = btrfs_next_leaf(root, path);
2115 } else if (ret > 0) {
2124 btrfs_item_key_to_cpu(leaf, &key, slot);
2126 if (key.objectid > inum)
2129 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2132 extent = btrfs_item_ptr(leaf, slot,
2133 struct btrfs_file_extent_item);
2135 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2138 extent_offset = btrfs_file_extent_offset(leaf, extent);
2139 if (key.offset - extent_offset != offset)
2142 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2143 if (extent_offset >= old->extent_offset + old->offset +
2144 old->len || extent_offset + num_bytes <=
2145 old->extent_offset + old->offset)
2151 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2157 backref->root_id = root_id;
2158 backref->inum = inum;
2159 backref->file_pos = offset + extent_offset;
2160 backref->num_bytes = num_bytes;
2161 backref->extent_offset = extent_offset;
2162 backref->generation = btrfs_file_extent_generation(leaf, extent);
2164 backref_insert(&new->root, backref);
2167 btrfs_release_path(path);
2172 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2173 struct new_sa_defrag_extent *new)
2175 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2176 struct old_sa_defrag_extent *old, *tmp;
2181 list_for_each_entry_safe(old, tmp, &new->head, list) {
2182 ret = iterate_inodes_from_logical(old->bytenr, fs_info,
2183 path, record_one_backref,
2185 BUG_ON(ret < 0 && ret != -ENOENT);
2187 /* no backref to be processed for this extent */
2189 list_del(&old->list);
2194 if (list_empty(&new->head))
2200 static int relink_is_mergable(struct extent_buffer *leaf,
2201 struct btrfs_file_extent_item *fi,
2204 if (btrfs_file_extent_disk_bytenr(leaf, fi) != disk_bytenr)
2207 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2210 if (btrfs_file_extent_compression(leaf, fi) ||
2211 btrfs_file_extent_encryption(leaf, fi) ||
2212 btrfs_file_extent_other_encoding(leaf, fi))
2219 * Note the backref might has changed, and in this case we just return 0.
2221 static noinline int relink_extent_backref(struct btrfs_path *path,
2222 struct sa_defrag_extent_backref *prev,
2223 struct sa_defrag_extent_backref *backref)
2225 struct btrfs_file_extent_item *extent;
2226 struct btrfs_file_extent_item *item;
2227 struct btrfs_ordered_extent *ordered;
2228 struct btrfs_trans_handle *trans;
2229 struct btrfs_fs_info *fs_info;
2230 struct btrfs_root *root;
2231 struct btrfs_key key;
2232 struct extent_buffer *leaf;
2233 struct old_sa_defrag_extent *old = backref->old;
2234 struct new_sa_defrag_extent *new = old->new;
2235 struct inode *src_inode = new->inode;
2236 struct inode *inode;
2237 struct extent_state *cached = NULL;
2246 if (prev && prev->root_id == backref->root_id &&
2247 prev->inum == backref->inum &&
2248 prev->file_pos + prev->num_bytes == backref->file_pos)
2251 /* step 1: get root */
2252 key.objectid = backref->root_id;
2253 key.type = BTRFS_ROOT_ITEM_KEY;
2254 key.offset = (u64)-1;
2256 fs_info = BTRFS_I(src_inode)->root->fs_info;
2257 index = srcu_read_lock(&fs_info->subvol_srcu);
2259 root = btrfs_read_fs_root_no_name(fs_info, &key);
2261 srcu_read_unlock(&fs_info->subvol_srcu, index);
2262 if (PTR_ERR(root) == -ENOENT)
2264 return PTR_ERR(root);
2266 if (btrfs_root_refs(&root->root_item) == 0) {
2267 srcu_read_unlock(&fs_info->subvol_srcu, index);
2268 /* parse ENOENT to 0 */
2272 /* step 2: get inode */
2273 key.objectid = backref->inum;
2274 key.type = BTRFS_INODE_ITEM_KEY;
2277 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2278 if (IS_ERR(inode)) {
2279 srcu_read_unlock(&fs_info->subvol_srcu, index);
2283 srcu_read_unlock(&fs_info->subvol_srcu, index);
2285 /* step 3: relink backref */
2286 lock_start = backref->file_pos;
2287 lock_end = backref->file_pos + backref->num_bytes - 1;
2288 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2291 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2293 btrfs_put_ordered_extent(ordered);
2297 trans = btrfs_join_transaction(root);
2298 if (IS_ERR(trans)) {
2299 ret = PTR_ERR(trans);
2303 key.objectid = backref->inum;
2304 key.type = BTRFS_EXTENT_DATA_KEY;
2305 key.offset = backref->file_pos;
2307 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2310 } else if (ret > 0) {
2315 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2316 struct btrfs_file_extent_item);
2318 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2319 backref->generation)
2322 btrfs_release_path(path);
2324 start = backref->file_pos;
2325 if (backref->extent_offset < old->extent_offset + old->offset)
2326 start += old->extent_offset + old->offset -
2327 backref->extent_offset;
2329 len = min(backref->extent_offset + backref->num_bytes,
2330 old->extent_offset + old->offset + old->len);
2331 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2333 ret = btrfs_drop_extents(trans, root, inode, start,
2338 key.objectid = btrfs_ino(inode);
2339 key.type = BTRFS_EXTENT_DATA_KEY;
2342 path->leave_spinning = 1;
2344 struct btrfs_file_extent_item *fi;
2346 struct btrfs_key found_key;
2348 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
2353 leaf = path->nodes[0];
2354 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2356 fi = btrfs_item_ptr(leaf, path->slots[0],
2357 struct btrfs_file_extent_item);
2358 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2360 if (relink_is_mergable(leaf, fi, new->bytenr) &&
2361 extent_len + found_key.offset == start) {
2362 btrfs_set_file_extent_num_bytes(leaf, fi,
2364 btrfs_mark_buffer_dirty(leaf);
2365 inode_add_bytes(inode, len);
2371 btrfs_release_path(path);
2376 ret = btrfs_insert_empty_item(trans, root, path, &key,
2379 btrfs_abort_transaction(trans, root, ret);
2383 leaf = path->nodes[0];
2384 item = btrfs_item_ptr(leaf, path->slots[0],
2385 struct btrfs_file_extent_item);
2386 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2387 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2388 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2389 btrfs_set_file_extent_num_bytes(leaf, item, len);
2390 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2391 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2392 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2393 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2394 btrfs_set_file_extent_encryption(leaf, item, 0);
2395 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2397 btrfs_mark_buffer_dirty(leaf);
2398 inode_add_bytes(inode, len);
2399 btrfs_release_path(path);
2401 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2403 backref->root_id, backref->inum,
2404 new->file_pos, 0); /* start - extent_offset */
2406 btrfs_abort_transaction(trans, root, ret);
2412 btrfs_release_path(path);
2413 path->leave_spinning = 0;
2414 btrfs_end_transaction(trans, root);
2416 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2422 static void relink_file_extents(struct new_sa_defrag_extent *new)
2424 struct btrfs_path *path;
2425 struct old_sa_defrag_extent *old, *tmp;
2426 struct sa_defrag_extent_backref *backref;
2427 struct sa_defrag_extent_backref *prev = NULL;
2428 struct inode *inode;
2429 struct btrfs_root *root;
2430 struct rb_node *node;
2434 root = BTRFS_I(inode)->root;
2436 path = btrfs_alloc_path();
2440 if (!record_extent_backrefs(path, new)) {
2441 btrfs_free_path(path);
2444 btrfs_release_path(path);
2447 node = rb_first(&new->root);
2450 rb_erase(node, &new->root);
2452 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2454 ret = relink_extent_backref(path, prev, backref);
2467 btrfs_free_path(path);
2469 list_for_each_entry_safe(old, tmp, &new->head, list) {
2470 list_del(&old->list);
2474 atomic_dec(&root->fs_info->defrag_running);
2475 wake_up(&root->fs_info->transaction_wait);
2480 static struct new_sa_defrag_extent *
2481 record_old_file_extents(struct inode *inode,
2482 struct btrfs_ordered_extent *ordered)
2484 struct btrfs_root *root = BTRFS_I(inode)->root;
2485 struct btrfs_path *path;
2486 struct btrfs_key key;
2487 struct old_sa_defrag_extent *old, *tmp;
2488 struct new_sa_defrag_extent *new;
2491 new = kmalloc(sizeof(*new), GFP_NOFS);
2496 new->file_pos = ordered->file_offset;
2497 new->len = ordered->len;
2498 new->bytenr = ordered->start;
2499 new->disk_len = ordered->disk_len;
2500 new->compress_type = ordered->compress_type;
2501 new->root = RB_ROOT;
2502 INIT_LIST_HEAD(&new->head);
2504 path = btrfs_alloc_path();
2508 key.objectid = btrfs_ino(inode);
2509 key.type = BTRFS_EXTENT_DATA_KEY;
2510 key.offset = new->file_pos;
2512 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2515 if (ret > 0 && path->slots[0] > 0)
2518 /* find out all the old extents for the file range */
2520 struct btrfs_file_extent_item *extent;
2521 struct extent_buffer *l;
2530 slot = path->slots[0];
2532 if (slot >= btrfs_header_nritems(l)) {
2533 ret = btrfs_next_leaf(root, path);
2541 btrfs_item_key_to_cpu(l, &key, slot);
2543 if (key.objectid != btrfs_ino(inode))
2545 if (key.type != BTRFS_EXTENT_DATA_KEY)
2547 if (key.offset >= new->file_pos + new->len)
2550 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2552 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2553 if (key.offset + num_bytes < new->file_pos)
2556 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2560 extent_offset = btrfs_file_extent_offset(l, extent);
2562 old = kmalloc(sizeof(*old), GFP_NOFS);
2566 offset = max(new->file_pos, key.offset);
2567 end = min(new->file_pos + new->len, key.offset + num_bytes);
2569 old->bytenr = disk_bytenr;
2570 old->extent_offset = extent_offset;
2571 old->offset = offset - key.offset;
2572 old->len = end - offset;
2575 list_add_tail(&old->list, &new->head);
2581 btrfs_free_path(path);
2582 atomic_inc(&root->fs_info->defrag_running);
2587 list_for_each_entry_safe(old, tmp, &new->head, list) {
2588 list_del(&old->list);
2592 btrfs_free_path(path);
2599 * helper function for btrfs_finish_ordered_io, this
2600 * just reads in some of the csum leaves to prime them into ram
2601 * before we start the transaction. It limits the amount of btree
2602 * reads required while inside the transaction.
2604 /* as ordered data IO finishes, this gets called so we can finish
2605 * an ordered extent if the range of bytes in the file it covers are
2608 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2610 struct inode *inode = ordered_extent->inode;
2611 struct btrfs_root *root = BTRFS_I(inode)->root;
2612 struct btrfs_trans_handle *trans = NULL;
2613 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2614 struct extent_state *cached_state = NULL;
2615 struct new_sa_defrag_extent *new = NULL;
2616 int compress_type = 0;
2620 nolock = btrfs_is_free_space_inode(inode);
2622 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2627 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2628 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2629 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2631 trans = btrfs_join_transaction_nolock(root);
2633 trans = btrfs_join_transaction(root);
2634 if (IS_ERR(trans)) {
2635 ret = PTR_ERR(trans);
2639 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2640 ret = btrfs_update_inode_fallback(trans, root, inode);
2641 if (ret) /* -ENOMEM or corruption */
2642 btrfs_abort_transaction(trans, root, ret);
2646 lock_extent_bits(io_tree, ordered_extent->file_offset,
2647 ordered_extent->file_offset + ordered_extent->len - 1,
2650 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2651 ordered_extent->file_offset + ordered_extent->len - 1,
2652 EXTENT_DEFRAG, 1, cached_state);
2654 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2655 if (last_snapshot >= BTRFS_I(inode)->generation)
2656 /* the inode is shared */
2657 new = record_old_file_extents(inode, ordered_extent);
2659 clear_extent_bit(io_tree, ordered_extent->file_offset,
2660 ordered_extent->file_offset + ordered_extent->len - 1,
2661 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2665 trans = btrfs_join_transaction_nolock(root);
2667 trans = btrfs_join_transaction(root);
2668 if (IS_ERR(trans)) {
2669 ret = PTR_ERR(trans);
2673 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2675 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2676 compress_type = ordered_extent->compress_type;
2677 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2678 BUG_ON(compress_type);
2679 ret = btrfs_mark_extent_written(trans, inode,
2680 ordered_extent->file_offset,
2681 ordered_extent->file_offset +
2682 ordered_extent->len);
2684 BUG_ON(root == root->fs_info->tree_root);
2685 ret = insert_reserved_file_extent(trans, inode,
2686 ordered_extent->file_offset,
2687 ordered_extent->start,
2688 ordered_extent->disk_len,
2689 ordered_extent->len,
2690 ordered_extent->len,
2691 compress_type, 0, 0,
2692 BTRFS_FILE_EXTENT_REG);
2694 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2695 ordered_extent->file_offset, ordered_extent->len,
2698 btrfs_abort_transaction(trans, root, ret);
2702 add_pending_csums(trans, inode, ordered_extent->file_offset,
2703 &ordered_extent->list);
2705 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2706 ret = btrfs_update_inode_fallback(trans, root, inode);
2707 if (ret) { /* -ENOMEM or corruption */
2708 btrfs_abort_transaction(trans, root, ret);
2713 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2714 ordered_extent->file_offset +
2715 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2717 if (root != root->fs_info->tree_root)
2718 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2720 btrfs_end_transaction(trans, root);
2723 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
2724 ordered_extent->file_offset +
2725 ordered_extent->len - 1, NULL, GFP_NOFS);
2728 * If the ordered extent had an IOERR or something else went
2729 * wrong we need to return the space for this ordered extent
2730 * back to the allocator.
2732 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2733 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2734 btrfs_free_reserved_extent(root, ordered_extent->start,
2735 ordered_extent->disk_len);
2740 * This needs to be done to make sure anybody waiting knows we are done
2741 * updating everything for this ordered extent.
2743 btrfs_remove_ordered_extent(inode, ordered_extent);
2745 /* for snapshot-aware defrag */
2747 relink_file_extents(new);
2750 btrfs_put_ordered_extent(ordered_extent);
2751 /* once for the tree */
2752 btrfs_put_ordered_extent(ordered_extent);
2757 static void finish_ordered_fn(struct btrfs_work *work)
2759 struct btrfs_ordered_extent *ordered_extent;
2760 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2761 btrfs_finish_ordered_io(ordered_extent);
2764 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2765 struct extent_state *state, int uptodate)
2767 struct inode *inode = page->mapping->host;
2768 struct btrfs_root *root = BTRFS_I(inode)->root;
2769 struct btrfs_ordered_extent *ordered_extent = NULL;
2770 struct btrfs_workers *workers;
2772 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2774 ClearPagePrivate2(page);
2775 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2776 end - start + 1, uptodate))
2779 ordered_extent->work.func = finish_ordered_fn;
2780 ordered_extent->work.flags = 0;
2782 if (btrfs_is_free_space_inode(inode))
2783 workers = &root->fs_info->endio_freespace_worker;
2785 workers = &root->fs_info->endio_write_workers;
2786 btrfs_queue_worker(workers, &ordered_extent->work);
2792 * when reads are done, we need to check csums to verify the data is correct
2793 * if there's a match, we allow the bio to finish. If not, the code in
2794 * extent_io.c will try to find good copies for us.
2796 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2797 struct extent_state *state, int mirror)
2799 size_t offset = start - page_offset(page);
2800 struct inode *inode = page->mapping->host;
2801 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2803 u64 private = ~(u32)0;
2805 struct btrfs_root *root = BTRFS_I(inode)->root;
2807 static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2808 DEFAULT_RATELIMIT_BURST);
2810 if (PageChecked(page)) {
2811 ClearPageChecked(page);
2815 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2818 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2819 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2820 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2825 if (state && state->start == start) {
2826 private = state->private;
2829 ret = get_state_private(io_tree, start, &private);
2831 kaddr = kmap_atomic(page);
2835 csum = btrfs_csum_data(kaddr + offset, csum, end - start + 1);
2836 btrfs_csum_final(csum, (char *)&csum);
2837 if (csum != private)
2840 kunmap_atomic(kaddr);
2845 if (__ratelimit(&_rs))
2846 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u private %llu",
2847 (unsigned long long)btrfs_ino(page->mapping->host),
2848 (unsigned long long)start, csum,
2849 (unsigned long long)private);
2850 memset(kaddr + offset, 1, end - start + 1);
2851 flush_dcache_page(page);
2852 kunmap_atomic(kaddr);
2858 struct delayed_iput {
2859 struct list_head list;
2860 struct inode *inode;
2863 /* JDM: If this is fs-wide, why can't we add a pointer to
2864 * btrfs_inode instead and avoid the allocation? */
2865 void btrfs_add_delayed_iput(struct inode *inode)
2867 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2868 struct delayed_iput *delayed;
2870 if (atomic_add_unless(&inode->i_count, -1, 1))
2873 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2874 delayed->inode = inode;
2876 spin_lock(&fs_info->delayed_iput_lock);
2877 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2878 spin_unlock(&fs_info->delayed_iput_lock);
2881 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2884 struct btrfs_fs_info *fs_info = root->fs_info;
2885 struct delayed_iput *delayed;
2888 spin_lock(&fs_info->delayed_iput_lock);
2889 empty = list_empty(&fs_info->delayed_iputs);
2890 spin_unlock(&fs_info->delayed_iput_lock);
2894 spin_lock(&fs_info->delayed_iput_lock);
2895 list_splice_init(&fs_info->delayed_iputs, &list);
2896 spin_unlock(&fs_info->delayed_iput_lock);
2898 while (!list_empty(&list)) {
2899 delayed = list_entry(list.next, struct delayed_iput, list);
2900 list_del(&delayed->list);
2901 iput(delayed->inode);
2907 * This is called in transaction commit time. If there are no orphan
2908 * files in the subvolume, it removes orphan item and frees block_rsv
2911 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2912 struct btrfs_root *root)
2914 struct btrfs_block_rsv *block_rsv;
2917 if (atomic_read(&root->orphan_inodes) ||
2918 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2921 spin_lock(&root->orphan_lock);
2922 if (atomic_read(&root->orphan_inodes)) {
2923 spin_unlock(&root->orphan_lock);
2927 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2928 spin_unlock(&root->orphan_lock);
2932 block_rsv = root->orphan_block_rsv;
2933 root->orphan_block_rsv = NULL;
2934 spin_unlock(&root->orphan_lock);
2936 if (root->orphan_item_inserted &&
2937 btrfs_root_refs(&root->root_item) > 0) {
2938 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2939 root->root_key.objectid);
2941 root->orphan_item_inserted = 0;
2945 WARN_ON(block_rsv->size > 0);
2946 btrfs_free_block_rsv(root, block_rsv);
2951 * This creates an orphan entry for the given inode in case something goes
2952 * wrong in the middle of an unlink/truncate.
2954 * NOTE: caller of this function should reserve 5 units of metadata for
2957 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2959 struct btrfs_root *root = BTRFS_I(inode)->root;
2960 struct btrfs_block_rsv *block_rsv = NULL;
2965 if (!root->orphan_block_rsv) {
2966 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2971 spin_lock(&root->orphan_lock);
2972 if (!root->orphan_block_rsv) {
2973 root->orphan_block_rsv = block_rsv;
2974 } else if (block_rsv) {
2975 btrfs_free_block_rsv(root, block_rsv);
2979 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2980 &BTRFS_I(inode)->runtime_flags)) {
2983 * For proper ENOSPC handling, we should do orphan
2984 * cleanup when mounting. But this introduces backward
2985 * compatibility issue.
2987 if (!xchg(&root->orphan_item_inserted, 1))
2993 atomic_inc(&root->orphan_inodes);
2996 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2997 &BTRFS_I(inode)->runtime_flags))
2999 spin_unlock(&root->orphan_lock);
3001 /* grab metadata reservation from transaction handle */
3003 ret = btrfs_orphan_reserve_metadata(trans, inode);
3004 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3007 /* insert an orphan item to track this unlinked/truncated file */
3009 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3010 if (ret && ret != -EEXIST) {
3011 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3012 &BTRFS_I(inode)->runtime_flags);
3013 btrfs_abort_transaction(trans, root, ret);
3019 /* insert an orphan item to track subvolume contains orphan files */
3021 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3022 root->root_key.objectid);
3023 if (ret && ret != -EEXIST) {
3024 btrfs_abort_transaction(trans, root, ret);
3032 * We have done the truncate/delete so we can go ahead and remove the orphan
3033 * item for this particular inode.
3035 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3036 struct inode *inode)
3038 struct btrfs_root *root = BTRFS_I(inode)->root;
3039 int delete_item = 0;
3040 int release_rsv = 0;
3043 spin_lock(&root->orphan_lock);
3044 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3045 &BTRFS_I(inode)->runtime_flags))
3048 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3049 &BTRFS_I(inode)->runtime_flags))
3051 spin_unlock(&root->orphan_lock);
3053 if (trans && delete_item) {
3054 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
3055 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3059 btrfs_orphan_release_metadata(inode);
3060 atomic_dec(&root->orphan_inodes);
3067 * this cleans up any orphans that may be left on the list from the last use
3070 int btrfs_orphan_cleanup(struct btrfs_root *root)
3072 struct btrfs_path *path;
3073 struct extent_buffer *leaf;
3074 struct btrfs_key key, found_key;
3075 struct btrfs_trans_handle *trans;
3076 struct inode *inode;
3077 u64 last_objectid = 0;
3078 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3080 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3083 path = btrfs_alloc_path();
3090 key.objectid = BTRFS_ORPHAN_OBJECTID;
3091 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3092 key.offset = (u64)-1;
3095 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3100 * if ret == 0 means we found what we were searching for, which
3101 * is weird, but possible, so only screw with path if we didn't
3102 * find the key and see if we have stuff that matches
3106 if (path->slots[0] == 0)
3111 /* pull out the item */
3112 leaf = path->nodes[0];
3113 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3115 /* make sure the item matches what we want */
3116 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3118 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3121 /* release the path since we're done with it */
3122 btrfs_release_path(path);
3125 * this is where we are basically btrfs_lookup, without the
3126 * crossing root thing. we store the inode number in the
3127 * offset of the orphan item.
3130 if (found_key.offset == last_objectid) {
3131 btrfs_err(root->fs_info,
3132 "Error removing orphan entry, stopping orphan cleanup");
3137 last_objectid = found_key.offset;
3139 found_key.objectid = found_key.offset;
3140 found_key.type = BTRFS_INODE_ITEM_KEY;
3141 found_key.offset = 0;
3142 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3143 ret = PTR_RET(inode);
3144 if (ret && ret != -ESTALE)
3147 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3148 struct btrfs_root *dead_root;
3149 struct btrfs_fs_info *fs_info = root->fs_info;
3150 int is_dead_root = 0;
3153 * this is an orphan in the tree root. Currently these
3154 * could come from 2 sources:
3155 * a) a snapshot deletion in progress
3156 * b) a free space cache inode
3157 * We need to distinguish those two, as the snapshot
3158 * orphan must not get deleted.
3159 * find_dead_roots already ran before us, so if this
3160 * is a snapshot deletion, we should find the root
3161 * in the dead_roots list
3163 spin_lock(&fs_info->trans_lock);
3164 list_for_each_entry(dead_root, &fs_info->dead_roots,
3166 if (dead_root->root_key.objectid ==
3167 found_key.objectid) {
3172 spin_unlock(&fs_info->trans_lock);
3174 /* prevent this orphan from being found again */
3175 key.offset = found_key.objectid - 1;
3180 * Inode is already gone but the orphan item is still there,
3181 * kill the orphan item.
3183 if (ret == -ESTALE) {
3184 trans = btrfs_start_transaction(root, 1);
3185 if (IS_ERR(trans)) {
3186 ret = PTR_ERR(trans);
3189 btrfs_debug(root->fs_info, "auto deleting %Lu",
3190 found_key.objectid);
3191 ret = btrfs_del_orphan_item(trans, root,
3192 found_key.objectid);
3193 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3194 btrfs_end_transaction(trans, root);
3199 * add this inode to the orphan list so btrfs_orphan_del does
3200 * the proper thing when we hit it
3202 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3203 &BTRFS_I(inode)->runtime_flags);
3204 atomic_inc(&root->orphan_inodes);
3206 /* if we have links, this was a truncate, lets do that */
3207 if (inode->i_nlink) {
3208 if (!S_ISREG(inode->i_mode)) {
3215 /* 1 for the orphan item deletion. */
3216 trans = btrfs_start_transaction(root, 1);
3217 if (IS_ERR(trans)) {
3218 ret = PTR_ERR(trans);
3221 ret = btrfs_orphan_add(trans, inode);
3222 btrfs_end_transaction(trans, root);
3226 ret = btrfs_truncate(inode);
3228 btrfs_orphan_del(NULL, inode);
3233 /* this will do delete_inode and everything for us */
3238 /* release the path since we're done with it */
3239 btrfs_release_path(path);
3241 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3243 if (root->orphan_block_rsv)
3244 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3247 if (root->orphan_block_rsv || root->orphan_item_inserted) {
3248 trans = btrfs_join_transaction(root);
3250 btrfs_end_transaction(trans, root);
3254 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3256 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3260 btrfs_crit(root->fs_info,
3261 "could not do orphan cleanup %d", ret);
3262 btrfs_free_path(path);
3267 * very simple check to peek ahead in the leaf looking for xattrs. If we
3268 * don't find any xattrs, we know there can't be any acls.
3270 * slot is the slot the inode is in, objectid is the objectid of the inode
3272 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3273 int slot, u64 objectid)
3275 u32 nritems = btrfs_header_nritems(leaf);
3276 struct btrfs_key found_key;
3280 while (slot < nritems) {
3281 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3283 /* we found a different objectid, there must not be acls */
3284 if (found_key.objectid != objectid)
3287 /* we found an xattr, assume we've got an acl */
3288 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
3292 * we found a key greater than an xattr key, there can't
3293 * be any acls later on
3295 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3302 * it goes inode, inode backrefs, xattrs, extents,
3303 * so if there are a ton of hard links to an inode there can
3304 * be a lot of backrefs. Don't waste time searching too hard,
3305 * this is just an optimization
3310 /* we hit the end of the leaf before we found an xattr or
3311 * something larger than an xattr. We have to assume the inode
3318 * read an inode from the btree into the in-memory inode
3320 static void btrfs_read_locked_inode(struct inode *inode)
3322 struct btrfs_path *path;
3323 struct extent_buffer *leaf;
3324 struct btrfs_inode_item *inode_item;
3325 struct btrfs_timespec *tspec;
3326 struct btrfs_root *root = BTRFS_I(inode)->root;
3327 struct btrfs_key location;
3331 bool filled = false;
3333 ret = btrfs_fill_inode(inode, &rdev);
3337 path = btrfs_alloc_path();
3341 path->leave_spinning = 1;
3342 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3344 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3348 leaf = path->nodes[0];
3353 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3354 struct btrfs_inode_item);
3355 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3356 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3357 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3358 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3359 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3361 tspec = btrfs_inode_atime(inode_item);
3362 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3363 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3365 tspec = btrfs_inode_mtime(inode_item);
3366 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3367 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3369 tspec = btrfs_inode_ctime(inode_item);
3370 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3371 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3373 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3374 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3375 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3378 * If we were modified in the current generation and evicted from memory
3379 * and then re-read we need to do a full sync since we don't have any
3380 * idea about which extents were modified before we were evicted from
3383 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3384 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3385 &BTRFS_I(inode)->runtime_flags);
3387 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3388 inode->i_generation = BTRFS_I(inode)->generation;
3390 rdev = btrfs_inode_rdev(leaf, inode_item);
3392 BTRFS_I(inode)->index_cnt = (u64)-1;
3393 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3396 * try to precache a NULL acl entry for files that don't have
3397 * any xattrs or acls
3399 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3402 cache_no_acl(inode);
3404 btrfs_free_path(path);
3406 switch (inode->i_mode & S_IFMT) {
3408 inode->i_mapping->a_ops = &btrfs_aops;
3409 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3410 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3411 inode->i_fop = &btrfs_file_operations;
3412 inode->i_op = &btrfs_file_inode_operations;
3415 inode->i_fop = &btrfs_dir_file_operations;
3416 if (root == root->fs_info->tree_root)
3417 inode->i_op = &btrfs_dir_ro_inode_operations;
3419 inode->i_op = &btrfs_dir_inode_operations;
3422 inode->i_op = &btrfs_symlink_inode_operations;
3423 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3424 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3427 inode->i_op = &btrfs_special_inode_operations;
3428 init_special_inode(inode, inode->i_mode, rdev);
3432 btrfs_update_iflags(inode);
3436 btrfs_free_path(path);
3437 make_bad_inode(inode);
3441 * given a leaf and an inode, copy the inode fields into the leaf
3443 static void fill_inode_item(struct btrfs_trans_handle *trans,
3444 struct extent_buffer *leaf,
3445 struct btrfs_inode_item *item,
3446 struct inode *inode)
3448 struct btrfs_map_token token;
3450 btrfs_init_map_token(&token);
3452 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3453 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3454 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3456 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3457 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3459 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3460 inode->i_atime.tv_sec, &token);
3461 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3462 inode->i_atime.tv_nsec, &token);
3464 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3465 inode->i_mtime.tv_sec, &token);
3466 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3467 inode->i_mtime.tv_nsec, &token);
3469 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3470 inode->i_ctime.tv_sec, &token);
3471 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3472 inode->i_ctime.tv_nsec, &token);
3474 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3476 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3478 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3479 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3480 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3481 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3482 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3486 * copy everything in the in-memory inode into the btree.
3488 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3489 struct btrfs_root *root, struct inode *inode)
3491 struct btrfs_inode_item *inode_item;
3492 struct btrfs_path *path;
3493 struct extent_buffer *leaf;
3496 path = btrfs_alloc_path();
3500 path->leave_spinning = 1;
3501 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3509 btrfs_unlock_up_safe(path, 1);
3510 leaf = path->nodes[0];
3511 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3512 struct btrfs_inode_item);
3514 fill_inode_item(trans, leaf, inode_item, inode);
3515 btrfs_mark_buffer_dirty(leaf);
3516 btrfs_set_inode_last_trans(trans, inode);
3519 btrfs_free_path(path);
3524 * copy everything in the in-memory inode into the btree.
3526 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3527 struct btrfs_root *root, struct inode *inode)
3532 * If the inode is a free space inode, we can deadlock during commit
3533 * if we put it into the delayed code.
3535 * The data relocation inode should also be directly updated
3538 if (!btrfs_is_free_space_inode(inode)
3539 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3540 btrfs_update_root_times(trans, root);
3542 ret = btrfs_delayed_update_inode(trans, root, inode);
3544 btrfs_set_inode_last_trans(trans, inode);
3548 return btrfs_update_inode_item(trans, root, inode);
3551 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3552 struct btrfs_root *root,
3553 struct inode *inode)
3557 ret = btrfs_update_inode(trans, root, inode);
3559 return btrfs_update_inode_item(trans, root, inode);
3564 * unlink helper that gets used here in inode.c and in the tree logging
3565 * recovery code. It remove a link in a directory with a given name, and
3566 * also drops the back refs in the inode to the directory
3568 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3569 struct btrfs_root *root,
3570 struct inode *dir, struct inode *inode,
3571 const char *name, int name_len)
3573 struct btrfs_path *path;
3575 struct extent_buffer *leaf;
3576 struct btrfs_dir_item *di;
3577 struct btrfs_key key;
3579 u64 ino = btrfs_ino(inode);
3580 u64 dir_ino = btrfs_ino(dir);
3582 path = btrfs_alloc_path();
3588 path->leave_spinning = 1;
3589 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3590 name, name_len, -1);
3599 leaf = path->nodes[0];
3600 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3601 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3604 btrfs_release_path(path);
3606 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3609 btrfs_info(root->fs_info,
3610 "failed to delete reference to %.*s, inode %llu parent %llu",
3612 (unsigned long long)ino, (unsigned long long)dir_ino);
3613 btrfs_abort_transaction(trans, root, ret);
3617 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3619 btrfs_abort_transaction(trans, root, ret);
3623 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3625 if (ret != 0 && ret != -ENOENT) {
3626 btrfs_abort_transaction(trans, root, ret);
3630 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3635 btrfs_abort_transaction(trans, root, ret);
3637 btrfs_free_path(path);
3641 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3642 inode_inc_iversion(inode);
3643 inode_inc_iversion(dir);
3644 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3645 ret = btrfs_update_inode(trans, root, dir);
3650 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3651 struct btrfs_root *root,
3652 struct inode *dir, struct inode *inode,
3653 const char *name, int name_len)
3656 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3658 btrfs_drop_nlink(inode);
3659 ret = btrfs_update_inode(trans, root, inode);
3665 /* helper to check if there is any shared block in the path */
3666 static int check_path_shared(struct btrfs_root *root,
3667 struct btrfs_path *path)
3669 struct extent_buffer *eb;
3673 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
3676 if (!path->nodes[level])
3678 eb = path->nodes[level];
3679 if (!btrfs_block_can_be_shared(root, eb))
3681 ret = btrfs_lookup_extent_info(NULL, root, eb->start, level, 1,
3690 * helper to start transaction for unlink and rmdir.
3692 * unlink and rmdir are special in btrfs, they do not always free space.
3693 * so in enospc case, we should make sure they will free space before
3694 * allowing them to use the global metadata reservation.
3696 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
3697 struct dentry *dentry)
3699 struct btrfs_trans_handle *trans;
3700 struct btrfs_root *root = BTRFS_I(dir)->root;
3701 struct btrfs_path *path;
3702 struct btrfs_dir_item *di;
3703 struct inode *inode = dentry->d_inode;
3708 u64 ino = btrfs_ino(inode);
3709 u64 dir_ino = btrfs_ino(dir);
3712 * 1 for the possible orphan item
3713 * 1 for the dir item
3714 * 1 for the dir index
3715 * 1 for the inode ref
3718 trans = btrfs_start_transaction(root, 5);
3719 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3722 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
3723 return ERR_PTR(-ENOSPC);
3725 /* check if there is someone else holds reference */
3726 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
3727 return ERR_PTR(-ENOSPC);
3729 if (atomic_read(&inode->i_count) > 2)
3730 return ERR_PTR(-ENOSPC);
3732 if (xchg(&root->fs_info->enospc_unlink, 1))
3733 return ERR_PTR(-ENOSPC);
3735 path = btrfs_alloc_path();
3737 root->fs_info->enospc_unlink = 0;
3738 return ERR_PTR(-ENOMEM);
3741 /* 1 for the orphan item */
3742 trans = btrfs_start_transaction(root, 1);
3743 if (IS_ERR(trans)) {
3744 btrfs_free_path(path);
3745 root->fs_info->enospc_unlink = 0;
3749 path->skip_locking = 1;
3750 path->search_commit_root = 1;
3752 ret = btrfs_lookup_inode(trans, root, path,
3753 &BTRFS_I(dir)->location, 0);
3759 if (check_path_shared(root, path))
3764 btrfs_release_path(path);
3766 ret = btrfs_lookup_inode(trans, root, path,
3767 &BTRFS_I(inode)->location, 0);
3773 if (check_path_shared(root, path))
3778 btrfs_release_path(path);
3780 if (ret == 0 && S_ISREG(inode->i_mode)) {
3781 ret = btrfs_lookup_file_extent(trans, root, path,
3787 BUG_ON(ret == 0); /* Corruption */
3788 if (check_path_shared(root, path))
3790 btrfs_release_path(path);
3798 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3799 dentry->d_name.name, dentry->d_name.len, 0);
3805 if (check_path_shared(root, path))
3811 btrfs_release_path(path);
3813 ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3814 dentry->d_name.len, ino, dir_ino, 0,
3821 if (check_path_shared(root, path))
3824 btrfs_release_path(path);
3827 * This is a commit root search, if we can lookup inode item and other
3828 * relative items in the commit root, it means the transaction of
3829 * dir/file creation has been committed, and the dir index item that we
3830 * delay to insert has also been inserted into the commit root. So
3831 * we needn't worry about the delayed insertion of the dir index item
3834 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3835 dentry->d_name.name, dentry->d_name.len, 0);
3840 BUG_ON(ret == -ENOENT);
3841 if (check_path_shared(root, path))
3846 btrfs_free_path(path);
3847 /* Migrate the orphan reservation over */
3849 err = btrfs_block_rsv_migrate(trans->block_rsv,
3850 &root->fs_info->global_block_rsv,
3851 trans->bytes_reserved);
3854 btrfs_end_transaction(trans, root);
3855 root->fs_info->enospc_unlink = 0;
3856 return ERR_PTR(err);
3859 trans->block_rsv = &root->fs_info->global_block_rsv;
3863 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3864 struct btrfs_root *root)
3866 if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3867 btrfs_block_rsv_release(root, trans->block_rsv,
3868 trans->bytes_reserved);
3869 trans->block_rsv = &root->fs_info->trans_block_rsv;
3870 BUG_ON(!root->fs_info->enospc_unlink);
3871 root->fs_info->enospc_unlink = 0;
3873 btrfs_end_transaction(trans, root);
3876 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3878 struct btrfs_root *root = BTRFS_I(dir)->root;
3879 struct btrfs_trans_handle *trans;
3880 struct inode *inode = dentry->d_inode;
3883 trans = __unlink_start_trans(dir, dentry);
3885 return PTR_ERR(trans);
3887 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3889 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3890 dentry->d_name.name, dentry->d_name.len);
3894 if (inode->i_nlink == 0) {
3895 ret = btrfs_orphan_add(trans, inode);
3901 __unlink_end_trans(trans, root);
3902 btrfs_btree_balance_dirty(root);
3906 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3907 struct btrfs_root *root,
3908 struct inode *dir, u64 objectid,
3909 const char *name, int name_len)
3911 struct btrfs_path *path;
3912 struct extent_buffer *leaf;
3913 struct btrfs_dir_item *di;
3914 struct btrfs_key key;
3917 u64 dir_ino = btrfs_ino(dir);
3919 path = btrfs_alloc_path();
3923 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3924 name, name_len, -1);
3925 if (IS_ERR_OR_NULL(di)) {
3933 leaf = path->nodes[0];
3934 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3935 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3936 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3938 btrfs_abort_transaction(trans, root, ret);
3941 btrfs_release_path(path);
3943 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3944 objectid, root->root_key.objectid,
3945 dir_ino, &index, name, name_len);
3947 if (ret != -ENOENT) {
3948 btrfs_abort_transaction(trans, root, ret);
3951 di = btrfs_search_dir_index_item(root, path, dir_ino,
3953 if (IS_ERR_OR_NULL(di)) {
3958 btrfs_abort_transaction(trans, root, ret);
3962 leaf = path->nodes[0];
3963 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3964 btrfs_release_path(path);
3967 btrfs_release_path(path);
3969 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3971 btrfs_abort_transaction(trans, root, ret);
3975 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3976 inode_inc_iversion(dir);
3977 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3978 ret = btrfs_update_inode_fallback(trans, root, dir);
3980 btrfs_abort_transaction(trans, root, ret);
3982 btrfs_free_path(path);
3986 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3988 struct inode *inode = dentry->d_inode;
3990 struct btrfs_root *root = BTRFS_I(dir)->root;
3991 struct btrfs_trans_handle *trans;
3993 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3995 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3998 trans = __unlink_start_trans(dir, dentry);
4000 return PTR_ERR(trans);
4002 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4003 err = btrfs_unlink_subvol(trans, root, dir,
4004 BTRFS_I(inode)->location.objectid,
4005 dentry->d_name.name,
4006 dentry->d_name.len);
4010 err = btrfs_orphan_add(trans, inode);
4014 /* now the directory is empty */
4015 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
4016 dentry->d_name.name, dentry->d_name.len);
4018 btrfs_i_size_write(inode, 0);
4020 __unlink_end_trans(trans, root);
4021 btrfs_btree_balance_dirty(root);
4027 * this can truncate away extent items, csum items and directory items.
4028 * It starts at a high offset and removes keys until it can't find
4029 * any higher than new_size
4031 * csum items that cross the new i_size are truncated to the new size
4034 * min_type is the minimum key type to truncate down to. If set to 0, this
4035 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4037 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4038 struct btrfs_root *root,
4039 struct inode *inode,
4040 u64 new_size, u32 min_type)
4042 struct btrfs_path *path;
4043 struct extent_buffer *leaf;
4044 struct btrfs_file_extent_item *fi;
4045 struct btrfs_key key;
4046 struct btrfs_key found_key;
4047 u64 extent_start = 0;
4048 u64 extent_num_bytes = 0;
4049 u64 extent_offset = 0;
4051 u32 found_type = (u8)-1;
4054 int pending_del_nr = 0;
4055 int pending_del_slot = 0;
4056 int extent_type = -1;
4059 u64 ino = btrfs_ino(inode);
4061 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4063 path = btrfs_alloc_path();
4069 * We want to drop from the next block forward in case this new size is
4070 * not block aligned since we will be keeping the last block of the
4071 * extent just the way it is.
4073 if (root->ref_cows || root == root->fs_info->tree_root)
4074 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4075 root->sectorsize), (u64)-1, 0);
4078 * This function is also used to drop the items in the log tree before
4079 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4080 * it is used to drop the loged items. So we shouldn't kill the delayed
4083 if (min_type == 0 && root == BTRFS_I(inode)->root)
4084 btrfs_kill_delayed_inode_items(inode);
4087 key.offset = (u64)-1;
4091 path->leave_spinning = 1;
4092 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4099 /* there are no items in the tree for us to truncate, we're
4102 if (path->slots[0] == 0)
4109 leaf = path->nodes[0];
4110 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4111 found_type = btrfs_key_type(&found_key);
4113 if (found_key.objectid != ino)
4116 if (found_type < min_type)
4119 item_end = found_key.offset;
4120 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4121 fi = btrfs_item_ptr(leaf, path->slots[0],
4122 struct btrfs_file_extent_item);
4123 extent_type = btrfs_file_extent_type(leaf, fi);
4124 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4126 btrfs_file_extent_num_bytes(leaf, fi);
4127 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4128 item_end += btrfs_file_extent_inline_len(leaf,
4133 if (found_type > min_type) {
4136 if (item_end < new_size)
4138 if (found_key.offset >= new_size)
4144 /* FIXME, shrink the extent if the ref count is only 1 */
4145 if (found_type != BTRFS_EXTENT_DATA_KEY)
4148 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4150 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4152 u64 orig_num_bytes =
4153 btrfs_file_extent_num_bytes(leaf, fi);
4154 extent_num_bytes = ALIGN(new_size -
4157 btrfs_set_file_extent_num_bytes(leaf, fi,
4159 num_dec = (orig_num_bytes -
4161 if (root->ref_cows && extent_start != 0)
4162 inode_sub_bytes(inode, num_dec);
4163 btrfs_mark_buffer_dirty(leaf);
4166 btrfs_file_extent_disk_num_bytes(leaf,
4168 extent_offset = found_key.offset -
4169 btrfs_file_extent_offset(leaf, fi);
4171 /* FIXME blocksize != 4096 */
4172 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4173 if (extent_start != 0) {
4176 inode_sub_bytes(inode, num_dec);
4179 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4181 * we can't truncate inline items that have had
4185 btrfs_file_extent_compression(leaf, fi) == 0 &&
4186 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4187 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4188 u32 size = new_size - found_key.offset;
4190 if (root->ref_cows) {
4191 inode_sub_bytes(inode, item_end + 1 -
4195 btrfs_file_extent_calc_inline_size(size);
4196 btrfs_truncate_item(root, path, size, 1);
4197 } else if (root->ref_cows) {
4198 inode_sub_bytes(inode, item_end + 1 -
4204 if (!pending_del_nr) {
4205 /* no pending yet, add ourselves */
4206 pending_del_slot = path->slots[0];
4208 } else if (pending_del_nr &&
4209 path->slots[0] + 1 == pending_del_slot) {
4210 /* hop on the pending chunk */
4212 pending_del_slot = path->slots[0];
4219 if (found_extent && (root->ref_cows ||
4220 root == root->fs_info->tree_root)) {
4221 btrfs_set_path_blocking(path);
4222 ret = btrfs_free_extent(trans, root, extent_start,
4223 extent_num_bytes, 0,
4224 btrfs_header_owner(leaf),
4225 ino, extent_offset, 0);
4229 if (found_type == BTRFS_INODE_ITEM_KEY)
4232 if (path->slots[0] == 0 ||
4233 path->slots[0] != pending_del_slot) {
4234 if (pending_del_nr) {
4235 ret = btrfs_del_items(trans, root, path,
4239 btrfs_abort_transaction(trans,
4245 btrfs_release_path(path);
4252 if (pending_del_nr) {
4253 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4256 btrfs_abort_transaction(trans, root, ret);
4259 btrfs_free_path(path);
4264 * btrfs_truncate_page - read, zero a chunk and write a page
4265 * @inode - inode that we're zeroing
4266 * @from - the offset to start zeroing
4267 * @len - the length to zero, 0 to zero the entire range respective to the
4269 * @front - zero up to the offset instead of from the offset on
4271 * This will find the page for the "from" offset and cow the page and zero the
4272 * part we want to zero. This is used with truncate and hole punching.
4274 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4277 struct address_space *mapping = inode->i_mapping;
4278 struct btrfs_root *root = BTRFS_I(inode)->root;
4279 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4280 struct btrfs_ordered_extent *ordered;
4281 struct extent_state *cached_state = NULL;
4283 u32 blocksize = root->sectorsize;
4284 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4285 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4287 gfp_t mask = btrfs_alloc_write_mask(mapping);
4292 if ((offset & (blocksize - 1)) == 0 &&
4293 (!len || ((len & (blocksize - 1)) == 0)))
4295 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4300 page = find_or_create_page(mapping, index, mask);
4302 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4307 page_start = page_offset(page);
4308 page_end = page_start + PAGE_CACHE_SIZE - 1;
4310 if (!PageUptodate(page)) {
4311 ret = btrfs_readpage(NULL, page);
4313 if (page->mapping != mapping) {
4315 page_cache_release(page);
4318 if (!PageUptodate(page)) {
4323 wait_on_page_writeback(page);
4325 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4326 set_page_extent_mapped(page);
4328 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4330 unlock_extent_cached(io_tree, page_start, page_end,
4331 &cached_state, GFP_NOFS);
4333 page_cache_release(page);
4334 btrfs_start_ordered_extent(inode, ordered, 1);
4335 btrfs_put_ordered_extent(ordered);
4339 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4340 EXTENT_DIRTY | EXTENT_DELALLOC |
4341 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4342 0, 0, &cached_state, GFP_NOFS);
4344 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4347 unlock_extent_cached(io_tree, page_start, page_end,
4348 &cached_state, GFP_NOFS);
4352 if (offset != PAGE_CACHE_SIZE) {
4354 len = PAGE_CACHE_SIZE - offset;
4357 memset(kaddr, 0, offset);
4359 memset(kaddr + offset, 0, len);
4360 flush_dcache_page(page);
4363 ClearPageChecked(page);
4364 set_page_dirty(page);
4365 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4370 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4372 page_cache_release(page);
4378 * This function puts in dummy file extents for the area we're creating a hole
4379 * for. So if we are truncating this file to a larger size we need to insert
4380 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4381 * the range between oldsize and size
4383 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4385 struct btrfs_trans_handle *trans;
4386 struct btrfs_root *root = BTRFS_I(inode)->root;
4387 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4388 struct extent_map *em = NULL;
4389 struct extent_state *cached_state = NULL;
4390 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4391 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4392 u64 block_end = ALIGN(size, root->sectorsize);
4398 if (size <= hole_start)
4402 struct btrfs_ordered_extent *ordered;
4403 btrfs_wait_ordered_range(inode, hole_start,
4404 block_end - hole_start);
4405 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4407 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
4410 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4411 &cached_state, GFP_NOFS);
4412 btrfs_put_ordered_extent(ordered);
4415 cur_offset = hole_start;
4417 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4418 block_end - cur_offset, 0);
4424 last_byte = min(extent_map_end(em), block_end);
4425 last_byte = ALIGN(last_byte , root->sectorsize);
4426 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4427 struct extent_map *hole_em;
4428 hole_size = last_byte - cur_offset;
4430 trans = btrfs_start_transaction(root, 3);
4431 if (IS_ERR(trans)) {
4432 err = PTR_ERR(trans);
4436 err = btrfs_drop_extents(trans, root, inode,
4438 cur_offset + hole_size, 1);
4440 btrfs_abort_transaction(trans, root, err);
4441 btrfs_end_transaction(trans, root);
4445 err = btrfs_insert_file_extent(trans, root,
4446 btrfs_ino(inode), cur_offset, 0,
4447 0, hole_size, 0, hole_size,
4450 btrfs_abort_transaction(trans, root, err);
4451 btrfs_end_transaction(trans, root);
4455 btrfs_drop_extent_cache(inode, cur_offset,
4456 cur_offset + hole_size - 1, 0);
4457 hole_em = alloc_extent_map();
4459 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4460 &BTRFS_I(inode)->runtime_flags);
4463 hole_em->start = cur_offset;
4464 hole_em->len = hole_size;
4465 hole_em->orig_start = cur_offset;
4467 hole_em->block_start = EXTENT_MAP_HOLE;
4468 hole_em->block_len = 0;
4469 hole_em->orig_block_len = 0;
4470 hole_em->ram_bytes = hole_size;
4471 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4472 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4473 hole_em->generation = trans->transid;
4476 write_lock(&em_tree->lock);
4477 err = add_extent_mapping(em_tree, hole_em, 1);
4478 write_unlock(&em_tree->lock);
4481 btrfs_drop_extent_cache(inode, cur_offset,
4485 free_extent_map(hole_em);
4487 btrfs_update_inode(trans, root, inode);
4488 btrfs_end_transaction(trans, root);
4490 free_extent_map(em);
4492 cur_offset = last_byte;
4493 if (cur_offset >= block_end)
4497 free_extent_map(em);
4498 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4503 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4505 struct btrfs_root *root = BTRFS_I(inode)->root;
4506 struct btrfs_trans_handle *trans;
4507 loff_t oldsize = i_size_read(inode);
4508 loff_t newsize = attr->ia_size;
4509 int mask = attr->ia_valid;
4512 if (newsize == oldsize)
4516 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4517 * special case where we need to update the times despite not having
4518 * these flags set. For all other operations the VFS set these flags
4519 * explicitly if it wants a timestamp update.
4521 if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4522 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4524 if (newsize > oldsize) {
4525 truncate_pagecache(inode, oldsize, newsize);
4526 ret = btrfs_cont_expand(inode, oldsize, newsize);
4530 trans = btrfs_start_transaction(root, 1);
4532 return PTR_ERR(trans);
4534 i_size_write(inode, newsize);
4535 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4536 ret = btrfs_update_inode(trans, root, inode);
4537 btrfs_end_transaction(trans, root);
4541 * We're truncating a file that used to have good data down to
4542 * zero. Make sure it gets into the ordered flush list so that
4543 * any new writes get down to disk quickly.
4546 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4547 &BTRFS_I(inode)->runtime_flags);
4550 * 1 for the orphan item we're going to add
4551 * 1 for the orphan item deletion.
4553 trans = btrfs_start_transaction(root, 2);
4555 return PTR_ERR(trans);
4558 * We need to do this in case we fail at _any_ point during the
4559 * actual truncate. Once we do the truncate_setsize we could
4560 * invalidate pages which forces any outstanding ordered io to
4561 * be instantly completed which will give us extents that need
4562 * to be truncated. If we fail to get an orphan inode down we
4563 * could have left over extents that were never meant to live,
4564 * so we need to garuntee from this point on that everything
4565 * will be consistent.
4567 ret = btrfs_orphan_add(trans, inode);
4568 btrfs_end_transaction(trans, root);
4572 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4573 truncate_setsize(inode, newsize);
4575 /* Disable nonlocked read DIO to avoid the end less truncate */
4576 btrfs_inode_block_unlocked_dio(inode);
4577 inode_dio_wait(inode);
4578 btrfs_inode_resume_unlocked_dio(inode);
4580 ret = btrfs_truncate(inode);
4581 if (ret && inode->i_nlink)
4582 btrfs_orphan_del(NULL, inode);
4588 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4590 struct inode *inode = dentry->d_inode;
4591 struct btrfs_root *root = BTRFS_I(inode)->root;
4594 if (btrfs_root_readonly(root))
4597 err = inode_change_ok(inode, attr);
4601 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4602 err = btrfs_setsize(inode, attr);
4607 if (attr->ia_valid) {
4608 setattr_copy(inode, attr);
4609 inode_inc_iversion(inode);
4610 err = btrfs_dirty_inode(inode);
4612 if (!err && attr->ia_valid & ATTR_MODE)
4613 err = btrfs_acl_chmod(inode);
4619 void btrfs_evict_inode(struct inode *inode)
4621 struct btrfs_trans_handle *trans;
4622 struct btrfs_root *root = BTRFS_I(inode)->root;
4623 struct btrfs_block_rsv *rsv, *global_rsv;
4624 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4627 trace_btrfs_inode_evict(inode);
4629 truncate_inode_pages(&inode->i_data, 0);
4630 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
4631 btrfs_is_free_space_inode(inode)))
4634 if (is_bad_inode(inode)) {
4635 btrfs_orphan_del(NULL, inode);
4638 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4639 btrfs_wait_ordered_range(inode, 0, (u64)-1);
4641 if (root->fs_info->log_root_recovering) {
4642 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4643 &BTRFS_I(inode)->runtime_flags));
4647 if (inode->i_nlink > 0) {
4648 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
4652 ret = btrfs_commit_inode_delayed_inode(inode);
4654 btrfs_orphan_del(NULL, inode);
4658 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4660 btrfs_orphan_del(NULL, inode);
4663 rsv->size = min_size;
4665 global_rsv = &root->fs_info->global_block_rsv;
4667 btrfs_i_size_write(inode, 0);
4670 * This is a bit simpler than btrfs_truncate since we've already
4671 * reserved our space for our orphan item in the unlink, so we just
4672 * need to reserve some slack space in case we add bytes and update
4673 * inode item when doing the truncate.
4676 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4677 BTRFS_RESERVE_FLUSH_LIMIT);
4680 * Try and steal from the global reserve since we will
4681 * likely not use this space anyway, we want to try as
4682 * hard as possible to get this to work.
4685 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4688 btrfs_warn(root->fs_info,
4689 "Could not get space for a delete, will truncate on mount %d",
4691 btrfs_orphan_del(NULL, inode);
4692 btrfs_free_block_rsv(root, rsv);
4696 trans = btrfs_join_transaction(root);
4697 if (IS_ERR(trans)) {
4698 btrfs_orphan_del(NULL, inode);
4699 btrfs_free_block_rsv(root, rsv);
4703 trans->block_rsv = rsv;
4705 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4709 trans->block_rsv = &root->fs_info->trans_block_rsv;
4710 btrfs_end_transaction(trans, root);
4712 btrfs_btree_balance_dirty(root);
4715 btrfs_free_block_rsv(root, rsv);
4718 trans->block_rsv = root->orphan_block_rsv;
4719 ret = btrfs_orphan_del(trans, inode);
4723 trans->block_rsv = &root->fs_info->trans_block_rsv;
4724 if (!(root == root->fs_info->tree_root ||
4725 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4726 btrfs_return_ino(root, btrfs_ino(inode));
4728 btrfs_end_transaction(trans, root);
4729 btrfs_btree_balance_dirty(root);
4731 btrfs_remove_delayed_node(inode);
4737 * this returns the key found in the dir entry in the location pointer.
4738 * If no dir entries were found, location->objectid is 0.
4740 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4741 struct btrfs_key *location)
4743 const char *name = dentry->d_name.name;
4744 int namelen = dentry->d_name.len;
4745 struct btrfs_dir_item *di;
4746 struct btrfs_path *path;
4747 struct btrfs_root *root = BTRFS_I(dir)->root;
4750 path = btrfs_alloc_path();
4754 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4759 if (IS_ERR_OR_NULL(di))
4762 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4764 btrfs_free_path(path);
4767 location->objectid = 0;
4772 * when we hit a tree root in a directory, the btrfs part of the inode
4773 * needs to be changed to reflect the root directory of the tree root. This
4774 * is kind of like crossing a mount point.
4776 static int fixup_tree_root_location(struct btrfs_root *root,
4778 struct dentry *dentry,
4779 struct btrfs_key *location,
4780 struct btrfs_root **sub_root)
4782 struct btrfs_path *path;
4783 struct btrfs_root *new_root;
4784 struct btrfs_root_ref *ref;
4785 struct extent_buffer *leaf;
4789 path = btrfs_alloc_path();
4796 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4797 BTRFS_I(dir)->root->root_key.objectid,
4798 location->objectid);
4805 leaf = path->nodes[0];
4806 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4807 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4808 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4811 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4812 (unsigned long)(ref + 1),
4813 dentry->d_name.len);
4817 btrfs_release_path(path);
4819 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4820 if (IS_ERR(new_root)) {
4821 err = PTR_ERR(new_root);
4825 if (btrfs_root_refs(&new_root->root_item) == 0) {
4830 *sub_root = new_root;
4831 location->objectid = btrfs_root_dirid(&new_root->root_item);
4832 location->type = BTRFS_INODE_ITEM_KEY;
4833 location->offset = 0;
4836 btrfs_free_path(path);
4840 static void inode_tree_add(struct inode *inode)
4842 struct btrfs_root *root = BTRFS_I(inode)->root;
4843 struct btrfs_inode *entry;
4845 struct rb_node *parent;
4846 u64 ino = btrfs_ino(inode);
4848 if (inode_unhashed(inode))
4852 spin_lock(&root->inode_lock);
4853 p = &root->inode_tree.rb_node;
4856 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4858 if (ino < btrfs_ino(&entry->vfs_inode))
4859 p = &parent->rb_left;
4860 else if (ino > btrfs_ino(&entry->vfs_inode))
4861 p = &parent->rb_right;
4863 WARN_ON(!(entry->vfs_inode.i_state &
4864 (I_WILL_FREE | I_FREEING)));
4865 rb_erase(parent, &root->inode_tree);
4866 RB_CLEAR_NODE(parent);
4867 spin_unlock(&root->inode_lock);
4871 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4872 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4873 spin_unlock(&root->inode_lock);
4876 static void inode_tree_del(struct inode *inode)
4878 struct btrfs_root *root = BTRFS_I(inode)->root;
4881 spin_lock(&root->inode_lock);
4882 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4883 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4884 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4885 empty = RB_EMPTY_ROOT(&root->inode_tree);
4887 spin_unlock(&root->inode_lock);
4890 * Free space cache has inodes in the tree root, but the tree root has a
4891 * root_refs of 0, so this could end up dropping the tree root as a
4892 * snapshot, so we need the extra !root->fs_info->tree_root check to
4893 * make sure we don't drop it.
4895 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4896 root != root->fs_info->tree_root) {
4897 synchronize_srcu(&root->fs_info->subvol_srcu);
4898 spin_lock(&root->inode_lock);
4899 empty = RB_EMPTY_ROOT(&root->inode_tree);
4900 spin_unlock(&root->inode_lock);
4902 btrfs_add_dead_root(root);
4906 void btrfs_invalidate_inodes(struct btrfs_root *root)
4908 struct rb_node *node;
4909 struct rb_node *prev;
4910 struct btrfs_inode *entry;
4911 struct inode *inode;
4914 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4916 spin_lock(&root->inode_lock);
4918 node = root->inode_tree.rb_node;
4922 entry = rb_entry(node, struct btrfs_inode, rb_node);
4924 if (objectid < btrfs_ino(&entry->vfs_inode))
4925 node = node->rb_left;
4926 else if (objectid > btrfs_ino(&entry->vfs_inode))
4927 node = node->rb_right;
4933 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4934 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4938 prev = rb_next(prev);
4942 entry = rb_entry(node, struct btrfs_inode, rb_node);
4943 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4944 inode = igrab(&entry->vfs_inode);
4946 spin_unlock(&root->inode_lock);
4947 if (atomic_read(&inode->i_count) > 1)
4948 d_prune_aliases(inode);
4950 * btrfs_drop_inode will have it removed from
4951 * the inode cache when its usage count
4956 spin_lock(&root->inode_lock);
4960 if (cond_resched_lock(&root->inode_lock))
4963 node = rb_next(node);
4965 spin_unlock(&root->inode_lock);
4968 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4970 struct btrfs_iget_args *args = p;
4971 inode->i_ino = args->ino;
4972 BTRFS_I(inode)->root = args->root;
4976 static int btrfs_find_actor(struct inode *inode, void *opaque)
4978 struct btrfs_iget_args *args = opaque;
4979 return args->ino == btrfs_ino(inode) &&
4980 args->root == BTRFS_I(inode)->root;
4983 static struct inode *btrfs_iget_locked(struct super_block *s,
4985 struct btrfs_root *root)
4987 struct inode *inode;
4988 struct btrfs_iget_args args;
4989 args.ino = objectid;
4992 inode = iget5_locked(s, objectid, btrfs_find_actor,
4993 btrfs_init_locked_inode,
4998 /* Get an inode object given its location and corresponding root.
4999 * Returns in *is_new if the inode was read from disk
5001 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5002 struct btrfs_root *root, int *new)
5004 struct inode *inode;
5006 inode = btrfs_iget_locked(s, location->objectid, root);
5008 return ERR_PTR(-ENOMEM);
5010 if (inode->i_state & I_NEW) {
5011 BTRFS_I(inode)->root = root;
5012 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
5013 btrfs_read_locked_inode(inode);
5014 if (!is_bad_inode(inode)) {
5015 inode_tree_add(inode);
5016 unlock_new_inode(inode);
5020 unlock_new_inode(inode);
5022 inode = ERR_PTR(-ESTALE);
5029 static struct inode *new_simple_dir(struct super_block *s,
5030 struct btrfs_key *key,
5031 struct btrfs_root *root)
5033 struct inode *inode = new_inode(s);
5036 return ERR_PTR(-ENOMEM);
5038 BTRFS_I(inode)->root = root;
5039 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5040 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5042 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5043 inode->i_op = &btrfs_dir_ro_inode_operations;
5044 inode->i_fop = &simple_dir_operations;
5045 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5046 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5051 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5053 struct inode *inode;
5054 struct btrfs_root *root = BTRFS_I(dir)->root;
5055 struct btrfs_root *sub_root = root;
5056 struct btrfs_key location;
5060 if (dentry->d_name.len > BTRFS_NAME_LEN)
5061 return ERR_PTR(-ENAMETOOLONG);
5063 ret = btrfs_inode_by_name(dir, dentry, &location);
5065 return ERR_PTR(ret);
5067 if (location.objectid == 0)
5070 if (location.type == BTRFS_INODE_ITEM_KEY) {
5071 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5075 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5077 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5078 ret = fixup_tree_root_location(root, dir, dentry,
5079 &location, &sub_root);
5082 inode = ERR_PTR(ret);
5084 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5086 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5088 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5090 if (!IS_ERR(inode) && root != sub_root) {
5091 down_read(&root->fs_info->cleanup_work_sem);
5092 if (!(inode->i_sb->s_flags & MS_RDONLY))
5093 ret = btrfs_orphan_cleanup(sub_root);
5094 up_read(&root->fs_info->cleanup_work_sem);
5096 inode = ERR_PTR(ret);
5102 static int btrfs_dentry_delete(const struct dentry *dentry)
5104 struct btrfs_root *root;
5105 struct inode *inode = dentry->d_inode;
5107 if (!inode && !IS_ROOT(dentry))
5108 inode = dentry->d_parent->d_inode;
5111 root = BTRFS_I(inode)->root;
5112 if (btrfs_root_refs(&root->root_item) == 0)
5115 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5121 static void btrfs_dentry_release(struct dentry *dentry)
5123 if (dentry->d_fsdata)
5124 kfree(dentry->d_fsdata);
5127 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5132 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
5136 unsigned char btrfs_filetype_table[] = {
5137 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5140 static int btrfs_real_readdir(struct file *filp, void *dirent,
5143 struct inode *inode = file_inode(filp);
5144 struct btrfs_root *root = BTRFS_I(inode)->root;
5145 struct btrfs_item *item;
5146 struct btrfs_dir_item *di;
5147 struct btrfs_key key;
5148 struct btrfs_key found_key;
5149 struct btrfs_path *path;
5150 struct list_head ins_list;
5151 struct list_head del_list;
5153 struct extent_buffer *leaf;
5155 unsigned char d_type;
5160 int key_type = BTRFS_DIR_INDEX_KEY;
5164 int is_curr = 0; /* filp->f_pos points to the current index? */
5166 /* FIXME, use a real flag for deciding about the key type */
5167 if (root->fs_info->tree_root == root)
5168 key_type = BTRFS_DIR_ITEM_KEY;
5170 /* special case for "." */
5171 if (filp->f_pos == 0) {
5172 over = filldir(dirent, ".", 1,
5173 filp->f_pos, btrfs_ino(inode), DT_DIR);
5178 /* special case for .., just use the back ref */
5179 if (filp->f_pos == 1) {
5180 u64 pino = parent_ino(filp->f_path.dentry);
5181 over = filldir(dirent, "..", 2,
5182 filp->f_pos, pino, DT_DIR);
5187 path = btrfs_alloc_path();
5193 if (key_type == BTRFS_DIR_INDEX_KEY) {
5194 INIT_LIST_HEAD(&ins_list);
5195 INIT_LIST_HEAD(&del_list);
5196 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5199 btrfs_set_key_type(&key, key_type);
5200 key.offset = filp->f_pos;
5201 key.objectid = btrfs_ino(inode);
5203 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5208 leaf = path->nodes[0];
5209 slot = path->slots[0];
5210 if (slot >= btrfs_header_nritems(leaf)) {
5211 ret = btrfs_next_leaf(root, path);
5219 item = btrfs_item_nr(leaf, slot);
5220 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5222 if (found_key.objectid != key.objectid)
5224 if (btrfs_key_type(&found_key) != key_type)
5226 if (found_key.offset < filp->f_pos)
5228 if (key_type == BTRFS_DIR_INDEX_KEY &&
5229 btrfs_should_delete_dir_index(&del_list,
5233 filp->f_pos = found_key.offset;
5236 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5238 di_total = btrfs_item_size(leaf, item);
5240 while (di_cur < di_total) {
5241 struct btrfs_key location;
5243 if (verify_dir_item(root, leaf, di))
5246 name_len = btrfs_dir_name_len(leaf, di);
5247 if (name_len <= sizeof(tmp_name)) {
5248 name_ptr = tmp_name;
5250 name_ptr = kmalloc(name_len, GFP_NOFS);
5256 read_extent_buffer(leaf, name_ptr,
5257 (unsigned long)(di + 1), name_len);
5259 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5260 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5263 /* is this a reference to our own snapshot? If so
5266 * In contrast to old kernels, we insert the snapshot's
5267 * dir item and dir index after it has been created, so
5268 * we won't find a reference to our own snapshot. We
5269 * still keep the following code for backward
5272 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5273 location.objectid == root->root_key.objectid) {
5277 over = filldir(dirent, name_ptr, name_len,
5278 found_key.offset, location.objectid,
5282 if (name_ptr != tmp_name)
5287 di_len = btrfs_dir_name_len(leaf, di) +
5288 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5290 di = (struct btrfs_dir_item *)((char *)di + di_len);
5296 if (key_type == BTRFS_DIR_INDEX_KEY) {
5299 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
5305 /* Reached end of directory/root. Bump pos past the last item. */
5306 if (key_type == BTRFS_DIR_INDEX_KEY)
5308 * 32-bit glibc will use getdents64, but then strtol -
5309 * so the last number we can serve is this.
5311 filp->f_pos = 0x7fffffff;
5317 if (key_type == BTRFS_DIR_INDEX_KEY)
5318 btrfs_put_delayed_items(&ins_list, &del_list);
5319 btrfs_free_path(path);
5323 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5325 struct btrfs_root *root = BTRFS_I(inode)->root;
5326 struct btrfs_trans_handle *trans;
5328 bool nolock = false;
5330 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5333 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5336 if (wbc->sync_mode == WB_SYNC_ALL) {
5338 trans = btrfs_join_transaction_nolock(root);
5340 trans = btrfs_join_transaction(root);
5342 return PTR_ERR(trans);
5343 ret = btrfs_commit_transaction(trans, root);
5349 * This is somewhat expensive, updating the tree every time the
5350 * inode changes. But, it is most likely to find the inode in cache.
5351 * FIXME, needs more benchmarking...there are no reasons other than performance
5352 * to keep or drop this code.
5354 static int btrfs_dirty_inode(struct inode *inode)
5356 struct btrfs_root *root = BTRFS_I(inode)->root;
5357 struct btrfs_trans_handle *trans;
5360 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5363 trans = btrfs_join_transaction(root);
5365 return PTR_ERR(trans);
5367 ret = btrfs_update_inode(trans, root, inode);
5368 if (ret && ret == -ENOSPC) {
5369 /* whoops, lets try again with the full transaction */
5370 btrfs_end_transaction(trans, root);
5371 trans = btrfs_start_transaction(root, 1);
5373 return PTR_ERR(trans);
5375 ret = btrfs_update_inode(trans, root, inode);
5377 btrfs_end_transaction(trans, root);
5378 if (BTRFS_I(inode)->delayed_node)
5379 btrfs_balance_delayed_items(root);
5385 * This is a copy of file_update_time. We need this so we can return error on
5386 * ENOSPC for updating the inode in the case of file write and mmap writes.
5388 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5391 struct btrfs_root *root = BTRFS_I(inode)->root;
5393 if (btrfs_root_readonly(root))
5396 if (flags & S_VERSION)
5397 inode_inc_iversion(inode);
5398 if (flags & S_CTIME)
5399 inode->i_ctime = *now;
5400 if (flags & S_MTIME)
5401 inode->i_mtime = *now;
5402 if (flags & S_ATIME)
5403 inode->i_atime = *now;
5404 return btrfs_dirty_inode(inode);
5408 * find the highest existing sequence number in a directory
5409 * and then set the in-memory index_cnt variable to reflect
5410 * free sequence numbers
5412 static int btrfs_set_inode_index_count(struct inode *inode)
5414 struct btrfs_root *root = BTRFS_I(inode)->root;
5415 struct btrfs_key key, found_key;
5416 struct btrfs_path *path;
5417 struct extent_buffer *leaf;
5420 key.objectid = btrfs_ino(inode);
5421 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5422 key.offset = (u64)-1;
5424 path = btrfs_alloc_path();
5428 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5431 /* FIXME: we should be able to handle this */
5437 * MAGIC NUMBER EXPLANATION:
5438 * since we search a directory based on f_pos we have to start at 2
5439 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5440 * else has to start at 2
5442 if (path->slots[0] == 0) {
5443 BTRFS_I(inode)->index_cnt = 2;
5449 leaf = path->nodes[0];
5450 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5452 if (found_key.objectid != btrfs_ino(inode) ||
5453 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5454 BTRFS_I(inode)->index_cnt = 2;
5458 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5460 btrfs_free_path(path);
5465 * helper to find a free sequence number in a given directory. This current
5466 * code is very simple, later versions will do smarter things in the btree
5468 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5472 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5473 ret = btrfs_inode_delayed_dir_index_count(dir);
5475 ret = btrfs_set_inode_index_count(dir);
5481 *index = BTRFS_I(dir)->index_cnt;
5482 BTRFS_I(dir)->index_cnt++;
5487 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5488 struct btrfs_root *root,
5490 const char *name, int name_len,
5491 u64 ref_objectid, u64 objectid,
5492 umode_t mode, u64 *index)
5494 struct inode *inode;
5495 struct btrfs_inode_item *inode_item;
5496 struct btrfs_key *location;
5497 struct btrfs_path *path;
5498 struct btrfs_inode_ref *ref;
5499 struct btrfs_key key[2];
5505 path = btrfs_alloc_path();
5507 return ERR_PTR(-ENOMEM);
5509 inode = new_inode(root->fs_info->sb);
5511 btrfs_free_path(path);
5512 return ERR_PTR(-ENOMEM);
5516 * we have to initialize this early, so we can reclaim the inode
5517 * number if we fail afterwards in this function.
5519 inode->i_ino = objectid;
5522 trace_btrfs_inode_request(dir);
5524 ret = btrfs_set_inode_index(dir, index);
5526 btrfs_free_path(path);
5528 return ERR_PTR(ret);
5532 * index_cnt is ignored for everything but a dir,
5533 * btrfs_get_inode_index_count has an explanation for the magic
5536 BTRFS_I(inode)->index_cnt = 2;
5537 BTRFS_I(inode)->root = root;
5538 BTRFS_I(inode)->generation = trans->transid;
5539 inode->i_generation = BTRFS_I(inode)->generation;
5542 * We could have gotten an inode number from somebody who was fsynced
5543 * and then removed in this same transaction, so let's just set full
5544 * sync since it will be a full sync anyway and this will blow away the
5545 * old info in the log.
5547 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5554 key[0].objectid = objectid;
5555 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5559 * Start new inodes with an inode_ref. This is slightly more
5560 * efficient for small numbers of hard links since they will
5561 * be packed into one item. Extended refs will kick in if we
5562 * add more hard links than can fit in the ref item.
5564 key[1].objectid = objectid;
5565 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5566 key[1].offset = ref_objectid;
5568 sizes[0] = sizeof(struct btrfs_inode_item);
5569 sizes[1] = name_len + sizeof(*ref);
5571 path->leave_spinning = 1;
5572 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5576 inode_init_owner(inode, dir, mode);
5577 inode_set_bytes(inode, 0);
5578 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5579 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5580 struct btrfs_inode_item);
5581 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5582 sizeof(*inode_item));
5583 fill_inode_item(trans, path->nodes[0], inode_item, inode);
5585 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5586 struct btrfs_inode_ref);
5587 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5588 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5589 ptr = (unsigned long)(ref + 1);
5590 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5592 btrfs_mark_buffer_dirty(path->nodes[0]);
5593 btrfs_free_path(path);
5595 location = &BTRFS_I(inode)->location;
5596 location->objectid = objectid;
5597 location->offset = 0;
5598 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5600 btrfs_inherit_iflags(inode, dir);
5602 if (S_ISREG(mode)) {
5603 if (btrfs_test_opt(root, NODATASUM))
5604 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5605 if (btrfs_test_opt(root, NODATACOW))
5606 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5607 BTRFS_INODE_NODATASUM;
5610 insert_inode_hash(inode);
5611 inode_tree_add(inode);
5613 trace_btrfs_inode_new(inode);
5614 btrfs_set_inode_last_trans(trans, inode);
5616 btrfs_update_root_times(trans, root);
5621 BTRFS_I(dir)->index_cnt--;
5622 btrfs_free_path(path);
5624 return ERR_PTR(ret);
5627 static inline u8 btrfs_inode_type(struct inode *inode)
5629 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5633 * utility function to add 'inode' into 'parent_inode' with
5634 * a give name and a given sequence number.
5635 * if 'add_backref' is true, also insert a backref from the
5636 * inode to the parent directory.
5638 int btrfs_add_link(struct btrfs_trans_handle *trans,
5639 struct inode *parent_inode, struct inode *inode,
5640 const char *name, int name_len, int add_backref, u64 index)
5643 struct btrfs_key key;
5644 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5645 u64 ino = btrfs_ino(inode);
5646 u64 parent_ino = btrfs_ino(parent_inode);
5648 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5649 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5652 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5656 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5657 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5658 key.objectid, root->root_key.objectid,
5659 parent_ino, index, name, name_len);
5660 } else if (add_backref) {
5661 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5665 /* Nothing to clean up yet */
5669 ret = btrfs_insert_dir_item(trans, root, name, name_len,
5671 btrfs_inode_type(inode), index);
5672 if (ret == -EEXIST || ret == -EOVERFLOW)
5675 btrfs_abort_transaction(trans, root, ret);
5679 btrfs_i_size_write(parent_inode, parent_inode->i_size +
5681 inode_inc_iversion(parent_inode);
5682 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5683 ret = btrfs_update_inode(trans, root, parent_inode);
5685 btrfs_abort_transaction(trans, root, ret);
5689 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5692 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5693 key.objectid, root->root_key.objectid,
5694 parent_ino, &local_index, name, name_len);
5696 } else if (add_backref) {
5700 err = btrfs_del_inode_ref(trans, root, name, name_len,
5701 ino, parent_ino, &local_index);
5706 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5707 struct inode *dir, struct dentry *dentry,
5708 struct inode *inode, int backref, u64 index)
5710 int err = btrfs_add_link(trans, dir, inode,
5711 dentry->d_name.name, dentry->d_name.len,
5718 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5719 umode_t mode, dev_t rdev)
5721 struct btrfs_trans_handle *trans;
5722 struct btrfs_root *root = BTRFS_I(dir)->root;
5723 struct inode *inode = NULL;
5729 if (!new_valid_dev(rdev))
5733 * 2 for inode item and ref
5735 * 1 for xattr if selinux is on
5737 trans = btrfs_start_transaction(root, 5);
5739 return PTR_ERR(trans);
5741 err = btrfs_find_free_ino(root, &objectid);
5745 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5746 dentry->d_name.len, btrfs_ino(dir), objectid,
5748 if (IS_ERR(inode)) {
5749 err = PTR_ERR(inode);
5753 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5760 * If the active LSM wants to access the inode during
5761 * d_instantiate it needs these. Smack checks to see
5762 * if the filesystem supports xattrs by looking at the
5766 inode->i_op = &btrfs_special_inode_operations;
5767 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5771 init_special_inode(inode, inode->i_mode, rdev);
5772 btrfs_update_inode(trans, root, inode);
5773 d_instantiate(dentry, inode);
5776 btrfs_end_transaction(trans, root);
5777 btrfs_btree_balance_dirty(root);
5779 inode_dec_link_count(inode);
5785 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5786 umode_t mode, bool excl)
5788 struct btrfs_trans_handle *trans;
5789 struct btrfs_root *root = BTRFS_I(dir)->root;
5790 struct inode *inode = NULL;
5791 int drop_inode_on_err = 0;
5797 * 2 for inode item and ref
5799 * 1 for xattr if selinux is on
5801 trans = btrfs_start_transaction(root, 5);
5803 return PTR_ERR(trans);
5805 err = btrfs_find_free_ino(root, &objectid);
5809 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5810 dentry->d_name.len, btrfs_ino(dir), objectid,
5812 if (IS_ERR(inode)) {
5813 err = PTR_ERR(inode);
5816 drop_inode_on_err = 1;
5818 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5822 err = btrfs_update_inode(trans, root, inode);
5827 * If the active LSM wants to access the inode during
5828 * d_instantiate it needs these. Smack checks to see
5829 * if the filesystem supports xattrs by looking at the
5832 inode->i_fop = &btrfs_file_operations;
5833 inode->i_op = &btrfs_file_inode_operations;
5835 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5839 inode->i_mapping->a_ops = &btrfs_aops;
5840 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5841 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5842 d_instantiate(dentry, inode);
5845 btrfs_end_transaction(trans, root);
5846 if (err && drop_inode_on_err) {
5847 inode_dec_link_count(inode);
5850 btrfs_btree_balance_dirty(root);
5854 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5855 struct dentry *dentry)
5857 struct btrfs_trans_handle *trans;
5858 struct btrfs_root *root = BTRFS_I(dir)->root;
5859 struct inode *inode = old_dentry->d_inode;
5864 /* do not allow sys_link's with other subvols of the same device */
5865 if (root->objectid != BTRFS_I(inode)->root->objectid)
5868 if (inode->i_nlink >= BTRFS_LINK_MAX)
5871 err = btrfs_set_inode_index(dir, &index);
5876 * 2 items for inode and inode ref
5877 * 2 items for dir items
5878 * 1 item for parent inode
5880 trans = btrfs_start_transaction(root, 5);
5881 if (IS_ERR(trans)) {
5882 err = PTR_ERR(trans);
5886 btrfs_inc_nlink(inode);
5887 inode_inc_iversion(inode);
5888 inode->i_ctime = CURRENT_TIME;
5890 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5892 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5897 struct dentry *parent = dentry->d_parent;
5898 err = btrfs_update_inode(trans, root, inode);
5901 d_instantiate(dentry, inode);
5902 btrfs_log_new_name(trans, inode, NULL, parent);
5905 btrfs_end_transaction(trans, root);
5908 inode_dec_link_count(inode);
5911 btrfs_btree_balance_dirty(root);
5915 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5917 struct inode *inode = NULL;
5918 struct btrfs_trans_handle *trans;
5919 struct btrfs_root *root = BTRFS_I(dir)->root;
5921 int drop_on_err = 0;
5926 * 2 items for inode and ref
5927 * 2 items for dir items
5928 * 1 for xattr if selinux is on
5930 trans = btrfs_start_transaction(root, 5);
5932 return PTR_ERR(trans);
5934 err = btrfs_find_free_ino(root, &objectid);
5938 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5939 dentry->d_name.len, btrfs_ino(dir), objectid,
5940 S_IFDIR | mode, &index);
5941 if (IS_ERR(inode)) {
5942 err = PTR_ERR(inode);
5948 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5952 inode->i_op = &btrfs_dir_inode_operations;
5953 inode->i_fop = &btrfs_dir_file_operations;
5955 btrfs_i_size_write(inode, 0);
5956 err = btrfs_update_inode(trans, root, inode);
5960 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5961 dentry->d_name.len, 0, index);
5965 d_instantiate(dentry, inode);
5969 btrfs_end_transaction(trans, root);
5972 btrfs_btree_balance_dirty(root);
5976 /* helper for btfs_get_extent. Given an existing extent in the tree,
5977 * and an extent that you want to insert, deal with overlap and insert
5978 * the new extent into the tree.
5980 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5981 struct extent_map *existing,
5982 struct extent_map *em,
5983 u64 map_start, u64 map_len)
5987 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5988 start_diff = map_start - em->start;
5989 em->start = map_start;
5991 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5992 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5993 em->block_start += start_diff;
5994 em->block_len -= start_diff;
5996 return add_extent_mapping(em_tree, em, 0);
5999 static noinline int uncompress_inline(struct btrfs_path *path,
6000 struct inode *inode, struct page *page,
6001 size_t pg_offset, u64 extent_offset,
6002 struct btrfs_file_extent_item *item)
6005 struct extent_buffer *leaf = path->nodes[0];
6008 unsigned long inline_size;
6012 WARN_ON(pg_offset != 0);
6013 compress_type = btrfs_file_extent_compression(leaf, item);
6014 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6015 inline_size = btrfs_file_extent_inline_item_len(leaf,
6016 btrfs_item_nr(leaf, path->slots[0]));
6017 tmp = kmalloc(inline_size, GFP_NOFS);
6020 ptr = btrfs_file_extent_inline_start(item);
6022 read_extent_buffer(leaf, tmp, ptr, inline_size);
6024 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6025 ret = btrfs_decompress(compress_type, tmp, page,
6026 extent_offset, inline_size, max_size);
6028 char *kaddr = kmap_atomic(page);
6029 unsigned long copy_size = min_t(u64,
6030 PAGE_CACHE_SIZE - pg_offset,
6031 max_size - extent_offset);
6032 memset(kaddr + pg_offset, 0, copy_size);
6033 kunmap_atomic(kaddr);
6040 * a bit scary, this does extent mapping from logical file offset to the disk.
6041 * the ugly parts come from merging extents from the disk with the in-ram
6042 * representation. This gets more complex because of the data=ordered code,
6043 * where the in-ram extents might be locked pending data=ordered completion.
6045 * This also copies inline extents directly into the page.
6048 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6049 size_t pg_offset, u64 start, u64 len,
6055 u64 extent_start = 0;
6057 u64 objectid = btrfs_ino(inode);
6059 struct btrfs_path *path = NULL;
6060 struct btrfs_root *root = BTRFS_I(inode)->root;
6061 struct btrfs_file_extent_item *item;
6062 struct extent_buffer *leaf;
6063 struct btrfs_key found_key;
6064 struct extent_map *em = NULL;
6065 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6066 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6067 struct btrfs_trans_handle *trans = NULL;
6071 read_lock(&em_tree->lock);
6072 em = lookup_extent_mapping(em_tree, start, len);
6074 em->bdev = root->fs_info->fs_devices->latest_bdev;
6075 read_unlock(&em_tree->lock);
6078 if (em->start > start || em->start + em->len <= start)
6079 free_extent_map(em);
6080 else if (em->block_start == EXTENT_MAP_INLINE && page)
6081 free_extent_map(em);
6085 em = alloc_extent_map();
6090 em->bdev = root->fs_info->fs_devices->latest_bdev;
6091 em->start = EXTENT_MAP_HOLE;
6092 em->orig_start = EXTENT_MAP_HOLE;
6094 em->block_len = (u64)-1;
6097 path = btrfs_alloc_path();
6103 * Chances are we'll be called again, so go ahead and do
6109 ret = btrfs_lookup_file_extent(trans, root, path,
6110 objectid, start, trans != NULL);
6117 if (path->slots[0] == 0)
6122 leaf = path->nodes[0];
6123 item = btrfs_item_ptr(leaf, path->slots[0],
6124 struct btrfs_file_extent_item);
6125 /* are we inside the extent that was found? */
6126 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6127 found_type = btrfs_key_type(&found_key);
6128 if (found_key.objectid != objectid ||
6129 found_type != BTRFS_EXTENT_DATA_KEY) {
6133 found_type = btrfs_file_extent_type(leaf, item);
6134 extent_start = found_key.offset;
6135 compress_type = btrfs_file_extent_compression(leaf, item);
6136 if (found_type == BTRFS_FILE_EXTENT_REG ||
6137 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6138 extent_end = extent_start +
6139 btrfs_file_extent_num_bytes(leaf, item);
6140 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6142 size = btrfs_file_extent_inline_len(leaf, item);
6143 extent_end = ALIGN(extent_start + size, root->sectorsize);
6146 if (start >= extent_end) {
6148 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6149 ret = btrfs_next_leaf(root, path);
6156 leaf = path->nodes[0];
6158 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6159 if (found_key.objectid != objectid ||
6160 found_key.type != BTRFS_EXTENT_DATA_KEY)
6162 if (start + len <= found_key.offset)
6165 em->orig_start = start;
6166 em->len = found_key.offset - start;
6170 em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
6171 if (found_type == BTRFS_FILE_EXTENT_REG ||
6172 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6173 em->start = extent_start;
6174 em->len = extent_end - extent_start;
6175 em->orig_start = extent_start -
6176 btrfs_file_extent_offset(leaf, item);
6177 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6179 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6181 em->block_start = EXTENT_MAP_HOLE;
6184 if (compress_type != BTRFS_COMPRESS_NONE) {
6185 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6186 em->compress_type = compress_type;
6187 em->block_start = bytenr;
6188 em->block_len = em->orig_block_len;
6190 bytenr += btrfs_file_extent_offset(leaf, item);
6191 em->block_start = bytenr;
6192 em->block_len = em->len;
6193 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6194 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6197 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6201 size_t extent_offset;
6204 em->block_start = EXTENT_MAP_INLINE;
6205 if (!page || create) {
6206 em->start = extent_start;
6207 em->len = extent_end - extent_start;
6211 size = btrfs_file_extent_inline_len(leaf, item);
6212 extent_offset = page_offset(page) + pg_offset - extent_start;
6213 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6214 size - extent_offset);
6215 em->start = extent_start + extent_offset;
6216 em->len = ALIGN(copy_size, root->sectorsize);
6217 em->orig_block_len = em->len;
6218 em->orig_start = em->start;
6219 if (compress_type) {
6220 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6221 em->compress_type = compress_type;
6223 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6224 if (create == 0 && !PageUptodate(page)) {
6225 if (btrfs_file_extent_compression(leaf, item) !=
6226 BTRFS_COMPRESS_NONE) {
6227 ret = uncompress_inline(path, inode, page,
6229 extent_offset, item);
6230 BUG_ON(ret); /* -ENOMEM */
6233 read_extent_buffer(leaf, map + pg_offset, ptr,
6235 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6236 memset(map + pg_offset + copy_size, 0,
6237 PAGE_CACHE_SIZE - pg_offset -
6242 flush_dcache_page(page);
6243 } else if (create && PageUptodate(page)) {
6247 free_extent_map(em);
6250 btrfs_release_path(path);
6251 trans = btrfs_join_transaction(root);
6254 return ERR_CAST(trans);
6258 write_extent_buffer(leaf, map + pg_offset, ptr,
6261 btrfs_mark_buffer_dirty(leaf);
6263 set_extent_uptodate(io_tree, em->start,
6264 extent_map_end(em) - 1, NULL, GFP_NOFS);
6267 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6271 em->orig_start = start;
6274 em->block_start = EXTENT_MAP_HOLE;
6275 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6277 btrfs_release_path(path);
6278 if (em->start > start || extent_map_end(em) <= start) {
6279 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6280 (unsigned long long)em->start,
6281 (unsigned long long)em->len,
6282 (unsigned long long)start,
6283 (unsigned long long)len);
6289 write_lock(&em_tree->lock);
6290 ret = add_extent_mapping(em_tree, em, 0);
6291 /* it is possible that someone inserted the extent into the tree
6292 * while we had the lock dropped. It is also possible that
6293 * an overlapping map exists in the tree
6295 if (ret == -EEXIST) {
6296 struct extent_map *existing;
6300 existing = lookup_extent_mapping(em_tree, start, len);
6301 if (existing && (existing->start > start ||
6302 existing->start + existing->len <= start)) {
6303 free_extent_map(existing);
6307 existing = lookup_extent_mapping(em_tree, em->start,
6310 err = merge_extent_mapping(em_tree, existing,
6313 free_extent_map(existing);
6315 free_extent_map(em);
6320 free_extent_map(em);
6324 free_extent_map(em);
6329 write_unlock(&em_tree->lock);
6333 trace_btrfs_get_extent(root, em);
6336 btrfs_free_path(path);
6338 ret = btrfs_end_transaction(trans, root);
6343 free_extent_map(em);
6344 return ERR_PTR(err);
6346 BUG_ON(!em); /* Error is always set */
6350 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6351 size_t pg_offset, u64 start, u64 len,
6354 struct extent_map *em;
6355 struct extent_map *hole_em = NULL;
6356 u64 range_start = start;
6362 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6369 * - a pre-alloc extent,
6370 * there might actually be delalloc bytes behind it.
6372 if (em->block_start != EXTENT_MAP_HOLE &&
6373 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6379 /* check to see if we've wrapped (len == -1 or similar) */
6388 /* ok, we didn't find anything, lets look for delalloc */
6389 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6390 end, len, EXTENT_DELALLOC, 1);
6391 found_end = range_start + found;
6392 if (found_end < range_start)
6393 found_end = (u64)-1;
6396 * we didn't find anything useful, return
6397 * the original results from get_extent()
6399 if (range_start > end || found_end <= start) {
6405 /* adjust the range_start to make sure it doesn't
6406 * go backwards from the start they passed in
6408 range_start = max(start,range_start);
6409 found = found_end - range_start;
6412 u64 hole_start = start;
6415 em = alloc_extent_map();
6421 * when btrfs_get_extent can't find anything it
6422 * returns one huge hole
6424 * make sure what it found really fits our range, and
6425 * adjust to make sure it is based on the start from
6429 u64 calc_end = extent_map_end(hole_em);
6431 if (calc_end <= start || (hole_em->start > end)) {
6432 free_extent_map(hole_em);
6435 hole_start = max(hole_em->start, start);
6436 hole_len = calc_end - hole_start;
6440 if (hole_em && range_start > hole_start) {
6441 /* our hole starts before our delalloc, so we
6442 * have to return just the parts of the hole
6443 * that go until the delalloc starts
6445 em->len = min(hole_len,
6446 range_start - hole_start);
6447 em->start = hole_start;
6448 em->orig_start = hole_start;
6450 * don't adjust block start at all,
6451 * it is fixed at EXTENT_MAP_HOLE
6453 em->block_start = hole_em->block_start;
6454 em->block_len = hole_len;
6455 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6456 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6458 em->start = range_start;
6460 em->orig_start = range_start;
6461 em->block_start = EXTENT_MAP_DELALLOC;
6462 em->block_len = found;
6464 } else if (hole_em) {
6469 free_extent_map(hole_em);
6471 free_extent_map(em);
6472 return ERR_PTR(err);
6477 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6480 struct btrfs_root *root = BTRFS_I(inode)->root;
6481 struct btrfs_trans_handle *trans;
6482 struct extent_map *em;
6483 struct btrfs_key ins;
6487 trans = btrfs_join_transaction(root);
6489 return ERR_CAST(trans);
6491 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
6493 alloc_hint = get_extent_allocation_hint(inode, start, len);
6494 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
6495 alloc_hint, &ins, 1);
6501 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6502 ins.offset, ins.offset, ins.offset, 0);
6506 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6507 ins.offset, ins.offset, 0);
6509 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6513 btrfs_end_transaction(trans, root);
6518 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6519 * block must be cow'd
6521 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
6522 struct inode *inode, u64 offset, u64 *len,
6523 u64 *orig_start, u64 *orig_block_len,
6526 struct btrfs_path *path;
6528 struct extent_buffer *leaf;
6529 struct btrfs_root *root = BTRFS_I(inode)->root;
6530 struct btrfs_file_extent_item *fi;
6531 struct btrfs_key key;
6539 path = btrfs_alloc_path();
6543 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
6548 slot = path->slots[0];
6551 /* can't find the item, must cow */
6558 leaf = path->nodes[0];
6559 btrfs_item_key_to_cpu(leaf, &key, slot);
6560 if (key.objectid != btrfs_ino(inode) ||
6561 key.type != BTRFS_EXTENT_DATA_KEY) {
6562 /* not our file or wrong item type, must cow */
6566 if (key.offset > offset) {
6567 /* Wrong offset, must cow */
6571 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6572 found_type = btrfs_file_extent_type(leaf, fi);
6573 if (found_type != BTRFS_FILE_EXTENT_REG &&
6574 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6575 /* not a regular extent, must cow */
6578 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6579 backref_offset = btrfs_file_extent_offset(leaf, fi);
6581 *orig_start = key.offset - backref_offset;
6582 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6583 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6585 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6586 if (extent_end < offset + *len) {
6587 /* extent doesn't include our full range, must cow */
6591 if (btrfs_extent_readonly(root, disk_bytenr))
6595 * look for other files referencing this extent, if we
6596 * find any we must cow
6598 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6599 key.offset - backref_offset, disk_bytenr))
6603 * adjust disk_bytenr and num_bytes to cover just the bytes
6604 * in this extent we are about to write. If there
6605 * are any csums in that range we have to cow in order
6606 * to keep the csums correct
6608 disk_bytenr += backref_offset;
6609 disk_bytenr += offset - key.offset;
6610 num_bytes = min(offset + *len, extent_end) - offset;
6611 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6614 * all of the above have passed, it is safe to overwrite this extent
6620 btrfs_free_path(path);
6624 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6625 struct extent_state **cached_state, int writing)
6627 struct btrfs_ordered_extent *ordered;
6631 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6634 * We're concerned with the entire range that we're going to be
6635 * doing DIO to, so we need to make sure theres no ordered
6636 * extents in this range.
6638 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6639 lockend - lockstart + 1);
6642 * We need to make sure there are no buffered pages in this
6643 * range either, we could have raced between the invalidate in
6644 * generic_file_direct_write and locking the extent. The
6645 * invalidate needs to happen so that reads after a write do not
6648 if (!ordered && (!writing ||
6649 !test_range_bit(&BTRFS_I(inode)->io_tree,
6650 lockstart, lockend, EXTENT_UPTODATE, 0,
6654 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6655 cached_state, GFP_NOFS);
6658 btrfs_start_ordered_extent(inode, ordered, 1);
6659 btrfs_put_ordered_extent(ordered);
6661 /* Screw you mmap */
6662 ret = filemap_write_and_wait_range(inode->i_mapping,
6669 * If we found a page that couldn't be invalidated just
6670 * fall back to buffered.
6672 ret = invalidate_inode_pages2_range(inode->i_mapping,
6673 lockstart >> PAGE_CACHE_SHIFT,
6674 lockend >> PAGE_CACHE_SHIFT);
6685 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6686 u64 len, u64 orig_start,
6687 u64 block_start, u64 block_len,
6688 u64 orig_block_len, u64 ram_bytes,
6691 struct extent_map_tree *em_tree;
6692 struct extent_map *em;
6693 struct btrfs_root *root = BTRFS_I(inode)->root;
6696 em_tree = &BTRFS_I(inode)->extent_tree;
6697 em = alloc_extent_map();
6699 return ERR_PTR(-ENOMEM);
6702 em->orig_start = orig_start;
6703 em->mod_start = start;
6706 em->block_len = block_len;
6707 em->block_start = block_start;
6708 em->bdev = root->fs_info->fs_devices->latest_bdev;
6709 em->orig_block_len = orig_block_len;
6710 em->ram_bytes = ram_bytes;
6711 em->generation = -1;
6712 set_bit(EXTENT_FLAG_PINNED, &em->flags);
6713 if (type == BTRFS_ORDERED_PREALLOC)
6714 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6717 btrfs_drop_extent_cache(inode, em->start,
6718 em->start + em->len - 1, 0);
6719 write_lock(&em_tree->lock);
6720 ret = add_extent_mapping(em_tree, em, 1);
6721 write_unlock(&em_tree->lock);
6722 } while (ret == -EEXIST);
6725 free_extent_map(em);
6726 return ERR_PTR(ret);
6733 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6734 struct buffer_head *bh_result, int create)
6736 struct extent_map *em;
6737 struct btrfs_root *root = BTRFS_I(inode)->root;
6738 struct extent_state *cached_state = NULL;
6739 u64 start = iblock << inode->i_blkbits;
6740 u64 lockstart, lockend;
6741 u64 len = bh_result->b_size;
6742 struct btrfs_trans_handle *trans;
6743 int unlock_bits = EXTENT_LOCKED;
6747 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6749 len = min_t(u64, len, root->sectorsize);
6752 lockend = start + len - 1;
6755 * If this errors out it's because we couldn't invalidate pagecache for
6756 * this range and we need to fallback to buffered.
6758 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6761 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6768 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6769 * io. INLINE is special, and we could probably kludge it in here, but
6770 * it's still buffered so for safety lets just fall back to the generic
6773 * For COMPRESSED we _have_ to read the entire extent in so we can
6774 * decompress it, so there will be buffering required no matter what we
6775 * do, so go ahead and fallback to buffered.
6777 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6778 * to buffered IO. Don't blame me, this is the price we pay for using
6781 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6782 em->block_start == EXTENT_MAP_INLINE) {
6783 free_extent_map(em);
6788 /* Just a good old fashioned hole, return */
6789 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6790 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6791 free_extent_map(em);
6796 * We don't allocate a new extent in the following cases
6798 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6800 * 2) The extent is marked as PREALLOC. We're good to go here and can
6801 * just use the extent.
6805 len = min(len, em->len - (start - em->start));
6806 lockstart = start + len;
6810 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6811 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6812 em->block_start != EXTENT_MAP_HOLE)) {
6815 u64 block_start, orig_start, orig_block_len, ram_bytes;
6817 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6818 type = BTRFS_ORDERED_PREALLOC;
6820 type = BTRFS_ORDERED_NOCOW;
6821 len = min(len, em->len - (start - em->start));
6822 block_start = em->block_start + (start - em->start);
6825 * we're not going to log anything, but we do need
6826 * to make sure the current transaction stays open
6827 * while we look for nocow cross refs
6829 trans = btrfs_join_transaction(root);
6833 if (can_nocow_odirect(trans, inode, start, &len, &orig_start,
6834 &orig_block_len, &ram_bytes) == 1) {
6835 if (type == BTRFS_ORDERED_PREALLOC) {
6836 free_extent_map(em);
6837 em = create_pinned_em(inode, start, len,
6843 btrfs_end_transaction(trans, root);
6848 ret = btrfs_add_ordered_extent_dio(inode, start,
6849 block_start, len, len, type);
6850 btrfs_end_transaction(trans, root);
6852 free_extent_map(em);
6857 btrfs_end_transaction(trans, root);
6861 * this will cow the extent, reset the len in case we changed
6864 len = bh_result->b_size;
6865 free_extent_map(em);
6866 em = btrfs_new_extent_direct(inode, start, len);
6871 len = min(len, em->len - (start - em->start));
6873 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6875 bh_result->b_size = len;
6876 bh_result->b_bdev = em->bdev;
6877 set_buffer_mapped(bh_result);
6879 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6880 set_buffer_new(bh_result);
6883 * Need to update the i_size under the extent lock so buffered
6884 * readers will get the updated i_size when we unlock.
6886 if (start + len > i_size_read(inode))
6887 i_size_write(inode, start + len);
6889 spin_lock(&BTRFS_I(inode)->lock);
6890 BTRFS_I(inode)->outstanding_extents++;
6891 spin_unlock(&BTRFS_I(inode)->lock);
6893 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6894 lockstart + len - 1, EXTENT_DELALLOC, NULL,
6895 &cached_state, GFP_NOFS);
6900 * In the case of write we need to clear and unlock the entire range,
6901 * in the case of read we need to unlock only the end area that we
6902 * aren't using if there is any left over space.
6904 if (lockstart < lockend) {
6905 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6906 lockend, unlock_bits, 1, 0,
6907 &cached_state, GFP_NOFS);
6909 free_extent_state(cached_state);
6912 free_extent_map(em);
6917 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6918 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6922 struct btrfs_dio_private {
6923 struct inode *inode;
6929 /* number of bios pending for this dio */
6930 atomic_t pending_bios;
6935 /* orig_bio is our btrfs_io_bio */
6936 struct bio *orig_bio;
6938 /* dio_bio came from fs/direct-io.c */
6939 struct bio *dio_bio;
6942 static void btrfs_endio_direct_read(struct bio *bio, int err)
6944 struct btrfs_dio_private *dip = bio->bi_private;
6945 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6946 struct bio_vec *bvec = bio->bi_io_vec;
6947 struct inode *inode = dip->inode;
6948 struct btrfs_root *root = BTRFS_I(inode)->root;
6949 struct bio *dio_bio;
6952 start = dip->logical_offset;
6954 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6955 struct page *page = bvec->bv_page;
6958 u64 private = ~(u32)0;
6959 unsigned long flags;
6961 if (get_state_private(&BTRFS_I(inode)->io_tree,
6964 local_irq_save(flags);
6965 kaddr = kmap_atomic(page);
6966 csum = btrfs_csum_data(kaddr + bvec->bv_offset,
6967 csum, bvec->bv_len);
6968 btrfs_csum_final(csum, (char *)&csum);
6969 kunmap_atomic(kaddr);
6970 local_irq_restore(flags);
6972 flush_dcache_page(bvec->bv_page);
6973 if (csum != private) {
6975 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u private %u",
6976 (unsigned long long)btrfs_ino(inode),
6977 (unsigned long long)start,
6978 csum, (unsigned)private);
6983 start += bvec->bv_len;
6985 } while (bvec <= bvec_end);
6987 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6988 dip->logical_offset + dip->bytes - 1);
6989 dio_bio = dip->dio_bio;
6993 /* If we had a csum failure make sure to clear the uptodate flag */
6995 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6996 dio_end_io(dio_bio, err);
7000 static void btrfs_endio_direct_write(struct bio *bio, int err)
7002 struct btrfs_dio_private *dip = bio->bi_private;
7003 struct inode *inode = dip->inode;
7004 struct btrfs_root *root = BTRFS_I(inode)->root;
7005 struct btrfs_ordered_extent *ordered = NULL;
7006 u64 ordered_offset = dip->logical_offset;
7007 u64 ordered_bytes = dip->bytes;
7008 struct bio *dio_bio;
7014 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7016 ordered_bytes, !err);
7020 ordered->work.func = finish_ordered_fn;
7021 ordered->work.flags = 0;
7022 btrfs_queue_worker(&root->fs_info->endio_write_workers,
7026 * our bio might span multiple ordered extents. If we haven't
7027 * completed the accounting for the whole dio, go back and try again
7029 if (ordered_offset < dip->logical_offset + dip->bytes) {
7030 ordered_bytes = dip->logical_offset + dip->bytes -
7036 dio_bio = dip->dio_bio;
7040 /* If we had an error make sure to clear the uptodate flag */
7042 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7043 dio_end_io(dio_bio, err);
7047 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7048 struct bio *bio, int mirror_num,
7049 unsigned long bio_flags, u64 offset)
7052 struct btrfs_root *root = BTRFS_I(inode)->root;
7053 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7054 BUG_ON(ret); /* -ENOMEM */
7058 static void btrfs_end_dio_bio(struct bio *bio, int err)
7060 struct btrfs_dio_private *dip = bio->bi_private;
7063 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
7064 "sector %#Lx len %u err no %d\n",
7065 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
7066 (unsigned long long)bio->bi_sector, bio->bi_size, err);
7070 * before atomic variable goto zero, we must make sure
7071 * dip->errors is perceived to be set.
7073 smp_mb__before_atomic_dec();
7076 /* if there are more bios still pending for this dio, just exit */
7077 if (!atomic_dec_and_test(&dip->pending_bios))
7081 bio_io_error(dip->orig_bio);
7083 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7084 bio_endio(dip->orig_bio, 0);
7090 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7091 u64 first_sector, gfp_t gfp_flags)
7093 int nr_vecs = bio_get_nr_vecs(bdev);
7094 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7097 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7098 int rw, u64 file_offset, int skip_sum,
7101 int write = rw & REQ_WRITE;
7102 struct btrfs_root *root = BTRFS_I(inode)->root;
7106 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7111 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7119 if (write && async_submit) {
7120 ret = btrfs_wq_submit_bio(root->fs_info,
7121 inode, rw, bio, 0, 0,
7123 __btrfs_submit_bio_start_direct_io,
7124 __btrfs_submit_bio_done);
7128 * If we aren't doing async submit, calculate the csum of the
7131 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7134 } else if (!skip_sum) {
7135 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
7141 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7147 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7150 struct inode *inode = dip->inode;
7151 struct btrfs_root *root = BTRFS_I(inode)->root;
7153 struct bio *orig_bio = dip->orig_bio;
7154 struct bio_vec *bvec = orig_bio->bi_io_vec;
7155 u64 start_sector = orig_bio->bi_sector;
7156 u64 file_offset = dip->logical_offset;
7161 int async_submit = 0;
7163 map_length = orig_bio->bi_size;
7164 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7165 &map_length, NULL, 0);
7170 if (map_length >= orig_bio->bi_size) {
7175 /* async crcs make it difficult to collect full stripe writes. */
7176 if (btrfs_get_alloc_profile(root, 1) &
7177 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7182 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7185 bio->bi_private = dip;
7186 bio->bi_end_io = btrfs_end_dio_bio;
7187 atomic_inc(&dip->pending_bios);
7189 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7190 if (unlikely(map_length < submit_len + bvec->bv_len ||
7191 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7192 bvec->bv_offset) < bvec->bv_len)) {
7194 * inc the count before we submit the bio so
7195 * we know the end IO handler won't happen before
7196 * we inc the count. Otherwise, the dip might get freed
7197 * before we're done setting it up
7199 atomic_inc(&dip->pending_bios);
7200 ret = __btrfs_submit_dio_bio(bio, inode, rw,
7201 file_offset, skip_sum,
7205 atomic_dec(&dip->pending_bios);
7209 start_sector += submit_len >> 9;
7210 file_offset += submit_len;
7215 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7216 start_sector, GFP_NOFS);
7219 bio->bi_private = dip;
7220 bio->bi_end_io = btrfs_end_dio_bio;
7222 map_length = orig_bio->bi_size;
7223 ret = btrfs_map_block(root->fs_info, rw,
7225 &map_length, NULL, 0);
7231 submit_len += bvec->bv_len;
7238 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7247 * before atomic variable goto zero, we must
7248 * make sure dip->errors is perceived to be set.
7250 smp_mb__before_atomic_dec();
7251 if (atomic_dec_and_test(&dip->pending_bios))
7252 bio_io_error(dip->orig_bio);
7254 /* bio_end_io() will handle error, so we needn't return it */
7258 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7259 struct inode *inode, loff_t file_offset)
7261 struct btrfs_root *root = BTRFS_I(inode)->root;
7262 struct btrfs_dio_private *dip;
7263 struct bio_vec *bvec = dio_bio->bi_io_vec;
7266 int write = rw & REQ_WRITE;
7269 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7271 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7278 dip = kmalloc(sizeof(*dip), GFP_NOFS);
7284 dip->private = dio_bio->bi_private;
7285 io_bio->bi_private = dio_bio->bi_private;
7287 dip->logical_offset = file_offset;
7291 dip->bytes += bvec->bv_len;
7293 } while (bvec <= (dio_bio->bi_io_vec + dio_bio->bi_vcnt - 1));
7295 dip->disk_bytenr = (u64)dio_bio->bi_sector << 9;
7296 io_bio->bi_private = dip;
7298 dip->orig_bio = io_bio;
7299 dip->dio_bio = dio_bio;
7300 atomic_set(&dip->pending_bios, 0);
7303 io_bio->bi_end_io = btrfs_endio_direct_write;
7305 io_bio->bi_end_io = btrfs_endio_direct_read;
7307 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7316 * If this is a write, we need to clean up the reserved space and kill
7317 * the ordered extent.
7320 struct btrfs_ordered_extent *ordered;
7321 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7322 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7323 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7324 btrfs_free_reserved_extent(root, ordered->start,
7326 btrfs_put_ordered_extent(ordered);
7327 btrfs_put_ordered_extent(ordered);
7329 bio_endio(dio_bio, ret);
7332 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7333 const struct iovec *iov, loff_t offset,
7334 unsigned long nr_segs)
7340 unsigned blocksize_mask = root->sectorsize - 1;
7341 ssize_t retval = -EINVAL;
7342 loff_t end = offset;
7344 if (offset & blocksize_mask)
7347 /* Check the memory alignment. Blocks cannot straddle pages */
7348 for (seg = 0; seg < nr_segs; seg++) {
7349 addr = (unsigned long)iov[seg].iov_base;
7350 size = iov[seg].iov_len;
7352 if ((addr & blocksize_mask) || (size & blocksize_mask))
7355 /* If this is a write we don't need to check anymore */
7360 * Check to make sure we don't have duplicate iov_base's in this
7361 * iovec, if so return EINVAL, otherwise we'll get csum errors
7362 * when reading back.
7364 for (i = seg + 1; i < nr_segs; i++) {
7365 if (iov[seg].iov_base == iov[i].iov_base)
7374 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7375 const struct iovec *iov, loff_t offset,
7376 unsigned long nr_segs)
7378 struct file *file = iocb->ki_filp;
7379 struct inode *inode = file->f_mapping->host;
7383 bool relock = false;
7386 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7390 atomic_inc(&inode->i_dio_count);
7391 smp_mb__after_atomic_inc();
7394 count = iov_length(iov, nr_segs);
7396 * If the write DIO is beyond the EOF, we need update
7397 * the isize, but it is protected by i_mutex. So we can
7398 * not unlock the i_mutex at this case.
7400 if (offset + count <= inode->i_size) {
7401 mutex_unlock(&inode->i_mutex);
7404 ret = btrfs_delalloc_reserve_space(inode, count);
7407 } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7408 &BTRFS_I(inode)->runtime_flags))) {
7409 inode_dio_done(inode);
7410 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7414 ret = __blockdev_direct_IO(rw, iocb, inode,
7415 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7416 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7417 btrfs_submit_direct, flags);
7419 if (ret < 0 && ret != -EIOCBQUEUED)
7420 btrfs_delalloc_release_space(inode, count);
7421 else if (ret >= 0 && (size_t)ret < count)
7422 btrfs_delalloc_release_space(inode,
7423 count - (size_t)ret);
7425 btrfs_delalloc_release_metadata(inode, 0);
7429 inode_dio_done(inode);
7431 mutex_lock(&inode->i_mutex);
7436 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
7438 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7439 __u64 start, __u64 len)
7443 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7447 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7450 int btrfs_readpage(struct file *file, struct page *page)
7452 struct extent_io_tree *tree;
7453 tree = &BTRFS_I(page->mapping->host)->io_tree;
7454 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7457 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7459 struct extent_io_tree *tree;
7462 if (current->flags & PF_MEMALLOC) {
7463 redirty_page_for_writepage(wbc, page);
7467 tree = &BTRFS_I(page->mapping->host)->io_tree;
7468 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7471 static int btrfs_writepages(struct address_space *mapping,
7472 struct writeback_control *wbc)
7474 struct extent_io_tree *tree;
7476 tree = &BTRFS_I(mapping->host)->io_tree;
7477 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7481 btrfs_readpages(struct file *file, struct address_space *mapping,
7482 struct list_head *pages, unsigned nr_pages)
7484 struct extent_io_tree *tree;
7485 tree = &BTRFS_I(mapping->host)->io_tree;
7486 return extent_readpages(tree, mapping, pages, nr_pages,
7489 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7491 struct extent_io_tree *tree;
7492 struct extent_map_tree *map;
7495 tree = &BTRFS_I(page->mapping->host)->io_tree;
7496 map = &BTRFS_I(page->mapping->host)->extent_tree;
7497 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7499 ClearPagePrivate(page);
7500 set_page_private(page, 0);
7501 page_cache_release(page);
7506 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7508 if (PageWriteback(page) || PageDirty(page))
7510 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7513 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
7515 struct inode *inode = page->mapping->host;
7516 struct extent_io_tree *tree;
7517 struct btrfs_ordered_extent *ordered;
7518 struct extent_state *cached_state = NULL;
7519 u64 page_start = page_offset(page);
7520 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7523 * we have the page locked, so new writeback can't start,
7524 * and the dirty bit won't be cleared while we are here.
7526 * Wait for IO on this page so that we can safely clear
7527 * the PagePrivate2 bit and do ordered accounting
7529 wait_on_page_writeback(page);
7531 tree = &BTRFS_I(inode)->io_tree;
7533 btrfs_releasepage(page, GFP_NOFS);
7536 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7537 ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
7540 * IO on this page will never be started, so we need
7541 * to account for any ordered extents now
7543 clear_extent_bit(tree, page_start, page_end,
7544 EXTENT_DIRTY | EXTENT_DELALLOC |
7545 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7546 EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
7548 * whoever cleared the private bit is responsible
7549 * for the finish_ordered_io
7551 if (TestClearPagePrivate2(page) &&
7552 btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
7553 PAGE_CACHE_SIZE, 1)) {
7554 btrfs_finish_ordered_io(ordered);
7556 btrfs_put_ordered_extent(ordered);
7557 cached_state = NULL;
7558 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7560 clear_extent_bit(tree, page_start, page_end,
7561 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
7562 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
7563 &cached_state, GFP_NOFS);
7564 __btrfs_releasepage(page, GFP_NOFS);
7566 ClearPageChecked(page);
7567 if (PagePrivate(page)) {
7568 ClearPagePrivate(page);
7569 set_page_private(page, 0);
7570 page_cache_release(page);
7575 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7576 * called from a page fault handler when a page is first dirtied. Hence we must
7577 * be careful to check for EOF conditions here. We set the page up correctly
7578 * for a written page which means we get ENOSPC checking when writing into
7579 * holes and correct delalloc and unwritten extent mapping on filesystems that
7580 * support these features.
7582 * We are not allowed to take the i_mutex here so we have to play games to
7583 * protect against truncate races as the page could now be beyond EOF. Because
7584 * vmtruncate() writes the inode size before removing pages, once we have the
7585 * page lock we can determine safely if the page is beyond EOF. If it is not
7586 * beyond EOF, then the page is guaranteed safe against truncation until we
7589 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7591 struct page *page = vmf->page;
7592 struct inode *inode = file_inode(vma->vm_file);
7593 struct btrfs_root *root = BTRFS_I(inode)->root;
7594 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7595 struct btrfs_ordered_extent *ordered;
7596 struct extent_state *cached_state = NULL;
7598 unsigned long zero_start;
7605 sb_start_pagefault(inode->i_sb);
7606 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7608 ret = file_update_time(vma->vm_file);
7614 else /* -ENOSPC, -EIO, etc */
7615 ret = VM_FAULT_SIGBUS;
7621 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7624 size = i_size_read(inode);
7625 page_start = page_offset(page);
7626 page_end = page_start + PAGE_CACHE_SIZE - 1;
7628 if ((page->mapping != inode->i_mapping) ||
7629 (page_start >= size)) {
7630 /* page got truncated out from underneath us */
7633 wait_on_page_writeback(page);
7635 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7636 set_page_extent_mapped(page);
7639 * we can't set the delalloc bits if there are pending ordered
7640 * extents. Drop our locks and wait for them to finish
7642 ordered = btrfs_lookup_ordered_extent(inode, page_start);
7644 unlock_extent_cached(io_tree, page_start, page_end,
7645 &cached_state, GFP_NOFS);
7647 btrfs_start_ordered_extent(inode, ordered, 1);
7648 btrfs_put_ordered_extent(ordered);
7653 * XXX - page_mkwrite gets called every time the page is dirtied, even
7654 * if it was already dirty, so for space accounting reasons we need to
7655 * clear any delalloc bits for the range we are fixing to save. There
7656 * is probably a better way to do this, but for now keep consistent with
7657 * prepare_pages in the normal write path.
7659 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7660 EXTENT_DIRTY | EXTENT_DELALLOC |
7661 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7662 0, 0, &cached_state, GFP_NOFS);
7664 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7667 unlock_extent_cached(io_tree, page_start, page_end,
7668 &cached_state, GFP_NOFS);
7669 ret = VM_FAULT_SIGBUS;
7674 /* page is wholly or partially inside EOF */
7675 if (page_start + PAGE_CACHE_SIZE > size)
7676 zero_start = size & ~PAGE_CACHE_MASK;
7678 zero_start = PAGE_CACHE_SIZE;
7680 if (zero_start != PAGE_CACHE_SIZE) {
7682 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7683 flush_dcache_page(page);
7686 ClearPageChecked(page);
7687 set_page_dirty(page);
7688 SetPageUptodate(page);
7690 BTRFS_I(inode)->last_trans = root->fs_info->generation;
7691 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7692 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7694 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7698 sb_end_pagefault(inode->i_sb);
7699 return VM_FAULT_LOCKED;
7703 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7705 sb_end_pagefault(inode->i_sb);
7709 static int btrfs_truncate(struct inode *inode)
7711 struct btrfs_root *root = BTRFS_I(inode)->root;
7712 struct btrfs_block_rsv *rsv;
7715 struct btrfs_trans_handle *trans;
7716 u64 mask = root->sectorsize - 1;
7717 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7719 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
7723 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
7724 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
7727 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
7728 * 3 things going on here
7730 * 1) We need to reserve space for our orphan item and the space to
7731 * delete our orphan item. Lord knows we don't want to have a dangling
7732 * orphan item because we didn't reserve space to remove it.
7734 * 2) We need to reserve space to update our inode.
7736 * 3) We need to have something to cache all the space that is going to
7737 * be free'd up by the truncate operation, but also have some slack
7738 * space reserved in case it uses space during the truncate (thank you
7739 * very much snapshotting).
7741 * And we need these to all be seperate. The fact is we can use alot of
7742 * space doing the truncate, and we have no earthly idea how much space
7743 * we will use, so we need the truncate reservation to be seperate so it
7744 * doesn't end up using space reserved for updating the inode or
7745 * removing the orphan item. We also need to be able to stop the
7746 * transaction and start a new one, which means we need to be able to
7747 * update the inode several times, and we have no idea of knowing how
7748 * many times that will be, so we can't just reserve 1 item for the
7749 * entirety of the opration, so that has to be done seperately as well.
7750 * Then there is the orphan item, which does indeed need to be held on
7751 * to for the whole operation, and we need nobody to touch this reserved
7752 * space except the orphan code.
7754 * So that leaves us with
7756 * 1) root->orphan_block_rsv - for the orphan deletion.
7757 * 2) rsv - for the truncate reservation, which we will steal from the
7758 * transaction reservation.
7759 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7760 * updating the inode.
7762 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7765 rsv->size = min_size;
7769 * 1 for the truncate slack space
7770 * 1 for updating the inode.
7772 trans = btrfs_start_transaction(root, 2);
7773 if (IS_ERR(trans)) {
7774 err = PTR_ERR(trans);
7778 /* Migrate the slack space for the truncate to our reserve */
7779 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7784 * setattr is responsible for setting the ordered_data_close flag,
7785 * but that is only tested during the last file release. That
7786 * could happen well after the next commit, leaving a great big
7787 * window where new writes may get lost if someone chooses to write
7788 * to this file after truncating to zero
7790 * The inode doesn't have any dirty data here, and so if we commit
7791 * this is a noop. If someone immediately starts writing to the inode
7792 * it is very likely we'll catch some of their writes in this
7793 * transaction, and the commit will find this file on the ordered
7794 * data list with good things to send down.
7796 * This is a best effort solution, there is still a window where
7797 * using truncate to replace the contents of the file will
7798 * end up with a zero length file after a crash.
7800 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7801 &BTRFS_I(inode)->runtime_flags))
7802 btrfs_add_ordered_operation(trans, root, inode);
7805 * So if we truncate and then write and fsync we normally would just
7806 * write the extents that changed, which is a problem if we need to
7807 * first truncate that entire inode. So set this flag so we write out
7808 * all of the extents in the inode to the sync log so we're completely
7811 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7812 trans->block_rsv = rsv;
7815 ret = btrfs_truncate_inode_items(trans, root, inode,
7817 BTRFS_EXTENT_DATA_KEY);
7818 if (ret != -ENOSPC) {
7823 trans->block_rsv = &root->fs_info->trans_block_rsv;
7824 ret = btrfs_update_inode(trans, root, inode);
7830 btrfs_end_transaction(trans, root);
7831 btrfs_btree_balance_dirty(root);
7833 trans = btrfs_start_transaction(root, 2);
7834 if (IS_ERR(trans)) {
7835 ret = err = PTR_ERR(trans);
7840 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7842 BUG_ON(ret); /* shouldn't happen */
7843 trans->block_rsv = rsv;
7846 if (ret == 0 && inode->i_nlink > 0) {
7847 trans->block_rsv = root->orphan_block_rsv;
7848 ret = btrfs_orphan_del(trans, inode);
7854 trans->block_rsv = &root->fs_info->trans_block_rsv;
7855 ret = btrfs_update_inode(trans, root, inode);
7859 ret = btrfs_end_transaction(trans, root);
7860 btrfs_btree_balance_dirty(root);
7864 btrfs_free_block_rsv(root, rsv);
7873 * create a new subvolume directory/inode (helper for the ioctl).
7875 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7876 struct btrfs_root *new_root, u64 new_dirid)
7878 struct inode *inode;
7882 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7883 new_dirid, new_dirid,
7884 S_IFDIR | (~current_umask() & S_IRWXUGO),
7887 return PTR_ERR(inode);
7888 inode->i_op = &btrfs_dir_inode_operations;
7889 inode->i_fop = &btrfs_dir_file_operations;
7891 set_nlink(inode, 1);
7892 btrfs_i_size_write(inode, 0);
7894 err = btrfs_update_inode(trans, new_root, inode);
7900 struct inode *btrfs_alloc_inode(struct super_block *sb)
7902 struct btrfs_inode *ei;
7903 struct inode *inode;
7905 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7912 ei->last_sub_trans = 0;
7913 ei->logged_trans = 0;
7914 ei->delalloc_bytes = 0;
7915 ei->disk_i_size = 0;
7918 ei->index_cnt = (u64)-1;
7919 ei->last_unlink_trans = 0;
7920 ei->last_log_commit = 0;
7922 spin_lock_init(&ei->lock);
7923 ei->outstanding_extents = 0;
7924 ei->reserved_extents = 0;
7926 ei->runtime_flags = 0;
7927 ei->force_compress = BTRFS_COMPRESS_NONE;
7929 ei->delayed_node = NULL;
7931 inode = &ei->vfs_inode;
7932 extent_map_tree_init(&ei->extent_tree);
7933 extent_io_tree_init(&ei->io_tree, &inode->i_data);
7934 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7935 ei->io_tree.track_uptodate = 1;
7936 ei->io_failure_tree.track_uptodate = 1;
7937 atomic_set(&ei->sync_writers, 0);
7938 mutex_init(&ei->log_mutex);
7939 mutex_init(&ei->delalloc_mutex);
7940 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7941 INIT_LIST_HEAD(&ei->delalloc_inodes);
7942 INIT_LIST_HEAD(&ei->ordered_operations);
7943 RB_CLEAR_NODE(&ei->rb_node);
7948 static void btrfs_i_callback(struct rcu_head *head)
7950 struct inode *inode = container_of(head, struct inode, i_rcu);
7951 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7954 void btrfs_destroy_inode(struct inode *inode)
7956 struct btrfs_ordered_extent *ordered;
7957 struct btrfs_root *root = BTRFS_I(inode)->root;
7959 WARN_ON(!hlist_empty(&inode->i_dentry));
7960 WARN_ON(inode->i_data.nrpages);
7961 WARN_ON(BTRFS_I(inode)->outstanding_extents);
7962 WARN_ON(BTRFS_I(inode)->reserved_extents);
7963 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7964 WARN_ON(BTRFS_I(inode)->csum_bytes);
7967 * This can happen where we create an inode, but somebody else also
7968 * created the same inode and we need to destroy the one we already
7975 * Make sure we're properly removed from the ordered operation
7979 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7980 spin_lock(&root->fs_info->ordered_extent_lock);
7981 list_del_init(&BTRFS_I(inode)->ordered_operations);
7982 spin_unlock(&root->fs_info->ordered_extent_lock);
7985 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7986 &BTRFS_I(inode)->runtime_flags)) {
7987 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
7988 (unsigned long long)btrfs_ino(inode));
7989 atomic_dec(&root->orphan_inodes);
7993 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7997 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
7998 (unsigned long long)ordered->file_offset,
7999 (unsigned long long)ordered->len);
8000 btrfs_remove_ordered_extent(inode, ordered);
8001 btrfs_put_ordered_extent(ordered);
8002 btrfs_put_ordered_extent(ordered);
8005 inode_tree_del(inode);
8006 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8008 call_rcu(&inode->i_rcu, btrfs_i_callback);
8011 int btrfs_drop_inode(struct inode *inode)
8013 struct btrfs_root *root = BTRFS_I(inode)->root;
8015 /* the snap/subvol tree is on deleting */
8016 if (btrfs_root_refs(&root->root_item) == 0 &&
8017 root != root->fs_info->tree_root)
8020 return generic_drop_inode(inode);
8023 static void init_once(void *foo)
8025 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8027 inode_init_once(&ei->vfs_inode);
8030 void btrfs_destroy_cachep(void)
8033 * Make sure all delayed rcu free inodes are flushed before we
8037 if (btrfs_inode_cachep)
8038 kmem_cache_destroy(btrfs_inode_cachep);
8039 if (btrfs_trans_handle_cachep)
8040 kmem_cache_destroy(btrfs_trans_handle_cachep);
8041 if (btrfs_transaction_cachep)
8042 kmem_cache_destroy(btrfs_transaction_cachep);
8043 if (btrfs_path_cachep)
8044 kmem_cache_destroy(btrfs_path_cachep);
8045 if (btrfs_free_space_cachep)
8046 kmem_cache_destroy(btrfs_free_space_cachep);
8047 if (btrfs_delalloc_work_cachep)
8048 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8051 int btrfs_init_cachep(void)
8053 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8054 sizeof(struct btrfs_inode), 0,
8055 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8056 if (!btrfs_inode_cachep)
8059 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8060 sizeof(struct btrfs_trans_handle), 0,
8061 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8062 if (!btrfs_trans_handle_cachep)
8065 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8066 sizeof(struct btrfs_transaction), 0,
8067 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8068 if (!btrfs_transaction_cachep)
8071 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8072 sizeof(struct btrfs_path), 0,
8073 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8074 if (!btrfs_path_cachep)
8077 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8078 sizeof(struct btrfs_free_space), 0,
8079 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8080 if (!btrfs_free_space_cachep)
8083 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8084 sizeof(struct btrfs_delalloc_work), 0,
8085 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8087 if (!btrfs_delalloc_work_cachep)
8092 btrfs_destroy_cachep();
8096 static int btrfs_getattr(struct vfsmount *mnt,
8097 struct dentry *dentry, struct kstat *stat)
8100 struct inode *inode = dentry->d_inode;
8101 u32 blocksize = inode->i_sb->s_blocksize;
8103 generic_fillattr(inode, stat);
8104 stat->dev = BTRFS_I(inode)->root->anon_dev;
8105 stat->blksize = PAGE_CACHE_SIZE;
8107 spin_lock(&BTRFS_I(inode)->lock);
8108 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8109 spin_unlock(&BTRFS_I(inode)->lock);
8110 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8111 ALIGN(delalloc_bytes, blocksize)) >> 9;
8115 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8116 struct inode *new_dir, struct dentry *new_dentry)
8118 struct btrfs_trans_handle *trans;
8119 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8120 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8121 struct inode *new_inode = new_dentry->d_inode;
8122 struct inode *old_inode = old_dentry->d_inode;
8123 struct timespec ctime = CURRENT_TIME;
8127 u64 old_ino = btrfs_ino(old_inode);
8129 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8132 /* we only allow rename subvolume link between subvolumes */
8133 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8136 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8137 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8140 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8141 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8145 /* check for collisions, even if the name isn't there */
8146 ret = btrfs_check_dir_item_collision(root, new_dir->i_ino,
8147 new_dentry->d_name.name,
8148 new_dentry->d_name.len);
8151 if (ret == -EEXIST) {
8153 * eexist without a new_inode */
8159 /* maybe -EOVERFLOW */
8166 * we're using rename to replace one file with another.
8167 * and the replacement file is large. Start IO on it now so
8168 * we don't add too much work to the end of the transaction
8170 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8171 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8172 filemap_flush(old_inode->i_mapping);
8174 /* close the racy window with snapshot create/destroy ioctl */
8175 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8176 down_read(&root->fs_info->subvol_sem);
8178 * We want to reserve the absolute worst case amount of items. So if
8179 * both inodes are subvols and we need to unlink them then that would
8180 * require 4 item modifications, but if they are both normal inodes it
8181 * would require 5 item modifications, so we'll assume their normal
8182 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8183 * should cover the worst case number of items we'll modify.
8185 trans = btrfs_start_transaction(root, 11);
8186 if (IS_ERR(trans)) {
8187 ret = PTR_ERR(trans);
8192 btrfs_record_root_in_trans(trans, dest);
8194 ret = btrfs_set_inode_index(new_dir, &index);
8198 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8199 /* force full log commit if subvolume involved. */
8200 root->fs_info->last_trans_log_full_commit = trans->transid;
8202 ret = btrfs_insert_inode_ref(trans, dest,
8203 new_dentry->d_name.name,
8204 new_dentry->d_name.len,
8206 btrfs_ino(new_dir), index);
8210 * this is an ugly little race, but the rename is required
8211 * to make sure that if we crash, the inode is either at the
8212 * old name or the new one. pinning the log transaction lets
8213 * us make sure we don't allow a log commit to come in after
8214 * we unlink the name but before we add the new name back in.
8216 btrfs_pin_log_trans(root);
8219 * make sure the inode gets flushed if it is replacing
8222 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8223 btrfs_add_ordered_operation(trans, root, old_inode);
8225 inode_inc_iversion(old_dir);
8226 inode_inc_iversion(new_dir);
8227 inode_inc_iversion(old_inode);
8228 old_dir->i_ctime = old_dir->i_mtime = ctime;
8229 new_dir->i_ctime = new_dir->i_mtime = ctime;
8230 old_inode->i_ctime = ctime;
8232 if (old_dentry->d_parent != new_dentry->d_parent)
8233 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8235 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8236 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8237 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8238 old_dentry->d_name.name,
8239 old_dentry->d_name.len);
8241 ret = __btrfs_unlink_inode(trans, root, old_dir,
8242 old_dentry->d_inode,
8243 old_dentry->d_name.name,
8244 old_dentry->d_name.len);
8246 ret = btrfs_update_inode(trans, root, old_inode);
8249 btrfs_abort_transaction(trans, root, ret);
8254 inode_inc_iversion(new_inode);
8255 new_inode->i_ctime = CURRENT_TIME;
8256 if (unlikely(btrfs_ino(new_inode) ==
8257 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8258 root_objectid = BTRFS_I(new_inode)->location.objectid;
8259 ret = btrfs_unlink_subvol(trans, dest, new_dir,
8261 new_dentry->d_name.name,
8262 new_dentry->d_name.len);
8263 BUG_ON(new_inode->i_nlink == 0);
8265 ret = btrfs_unlink_inode(trans, dest, new_dir,
8266 new_dentry->d_inode,
8267 new_dentry->d_name.name,
8268 new_dentry->d_name.len);
8270 if (!ret && new_inode->i_nlink == 0) {
8271 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8275 btrfs_abort_transaction(trans, root, ret);
8280 ret = btrfs_add_link(trans, new_dir, old_inode,
8281 new_dentry->d_name.name,
8282 new_dentry->d_name.len, 0, index);
8284 btrfs_abort_transaction(trans, root, ret);
8288 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8289 struct dentry *parent = new_dentry->d_parent;
8290 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8291 btrfs_end_log_trans(root);
8294 btrfs_end_transaction(trans, root);
8296 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8297 up_read(&root->fs_info->subvol_sem);
8302 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8304 struct btrfs_delalloc_work *delalloc_work;
8306 delalloc_work = container_of(work, struct btrfs_delalloc_work,
8308 if (delalloc_work->wait)
8309 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
8311 filemap_flush(delalloc_work->inode->i_mapping);
8313 if (delalloc_work->delay_iput)
8314 btrfs_add_delayed_iput(delalloc_work->inode);
8316 iput(delalloc_work->inode);
8317 complete(&delalloc_work->completion);
8320 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8321 int wait, int delay_iput)
8323 struct btrfs_delalloc_work *work;
8325 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8329 init_completion(&work->completion);
8330 INIT_LIST_HEAD(&work->list);
8331 work->inode = inode;
8333 work->delay_iput = delay_iput;
8334 work->work.func = btrfs_run_delalloc_work;
8339 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8341 wait_for_completion(&work->completion);
8342 kmem_cache_free(btrfs_delalloc_work_cachep, work);
8346 * some fairly slow code that needs optimization. This walks the list
8347 * of all the inodes with pending delalloc and forces them to disk.
8349 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8351 struct btrfs_inode *binode;
8352 struct inode *inode;
8353 struct btrfs_delalloc_work *work, *next;
8354 struct list_head works;
8355 struct list_head splice;
8358 if (root->fs_info->sb->s_flags & MS_RDONLY)
8361 INIT_LIST_HEAD(&works);
8362 INIT_LIST_HEAD(&splice);
8364 spin_lock(&root->fs_info->delalloc_lock);
8365 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
8366 while (!list_empty(&splice)) {
8367 binode = list_entry(splice.next, struct btrfs_inode,
8370 list_del_init(&binode->delalloc_inodes);
8372 inode = igrab(&binode->vfs_inode);
8374 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
8375 &binode->runtime_flags);
8379 list_add_tail(&binode->delalloc_inodes,
8380 &root->fs_info->delalloc_inodes);
8381 spin_unlock(&root->fs_info->delalloc_lock);
8383 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8384 if (unlikely(!work)) {
8388 list_add_tail(&work->list, &works);
8389 btrfs_queue_worker(&root->fs_info->flush_workers,
8393 spin_lock(&root->fs_info->delalloc_lock);
8395 spin_unlock(&root->fs_info->delalloc_lock);
8397 list_for_each_entry_safe(work, next, &works, list) {
8398 list_del_init(&work->list);
8399 btrfs_wait_and_free_delalloc_work(work);
8402 /* the filemap_flush will queue IO into the worker threads, but
8403 * we have to make sure the IO is actually started and that
8404 * ordered extents get created before we return
8406 atomic_inc(&root->fs_info->async_submit_draining);
8407 while (atomic_read(&root->fs_info->nr_async_submits) ||
8408 atomic_read(&root->fs_info->async_delalloc_pages)) {
8409 wait_event(root->fs_info->async_submit_wait,
8410 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8411 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8413 atomic_dec(&root->fs_info->async_submit_draining);
8416 list_for_each_entry_safe(work, next, &works, list) {
8417 list_del_init(&work->list);
8418 btrfs_wait_and_free_delalloc_work(work);
8421 if (!list_empty_careful(&splice)) {
8422 spin_lock(&root->fs_info->delalloc_lock);
8423 list_splice_tail(&splice, &root->fs_info->delalloc_inodes);
8424 spin_unlock(&root->fs_info->delalloc_lock);
8429 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8430 const char *symname)
8432 struct btrfs_trans_handle *trans;
8433 struct btrfs_root *root = BTRFS_I(dir)->root;
8434 struct btrfs_path *path;
8435 struct btrfs_key key;
8436 struct inode *inode = NULL;
8444 struct btrfs_file_extent_item *ei;
8445 struct extent_buffer *leaf;
8447 name_len = strlen(symname) + 1;
8448 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8449 return -ENAMETOOLONG;
8452 * 2 items for inode item and ref
8453 * 2 items for dir items
8454 * 1 item for xattr if selinux is on
8456 trans = btrfs_start_transaction(root, 5);
8458 return PTR_ERR(trans);
8460 err = btrfs_find_free_ino(root, &objectid);
8464 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8465 dentry->d_name.len, btrfs_ino(dir), objectid,
8466 S_IFLNK|S_IRWXUGO, &index);
8467 if (IS_ERR(inode)) {
8468 err = PTR_ERR(inode);
8472 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8479 * If the active LSM wants to access the inode during
8480 * d_instantiate it needs these. Smack checks to see
8481 * if the filesystem supports xattrs by looking at the
8484 inode->i_fop = &btrfs_file_operations;
8485 inode->i_op = &btrfs_file_inode_operations;
8487 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8491 inode->i_mapping->a_ops = &btrfs_aops;
8492 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8493 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8498 path = btrfs_alloc_path();
8504 key.objectid = btrfs_ino(inode);
8506 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8507 datasize = btrfs_file_extent_calc_inline_size(name_len);
8508 err = btrfs_insert_empty_item(trans, root, path, &key,
8512 btrfs_free_path(path);
8515 leaf = path->nodes[0];
8516 ei = btrfs_item_ptr(leaf, path->slots[0],
8517 struct btrfs_file_extent_item);
8518 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8519 btrfs_set_file_extent_type(leaf, ei,
8520 BTRFS_FILE_EXTENT_INLINE);
8521 btrfs_set_file_extent_encryption(leaf, ei, 0);
8522 btrfs_set_file_extent_compression(leaf, ei, 0);
8523 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8524 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8526 ptr = btrfs_file_extent_inline_start(ei);
8527 write_extent_buffer(leaf, symname, ptr, name_len);
8528 btrfs_mark_buffer_dirty(leaf);
8529 btrfs_free_path(path);
8531 inode->i_op = &btrfs_symlink_inode_operations;
8532 inode->i_mapping->a_ops = &btrfs_symlink_aops;
8533 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8534 inode_set_bytes(inode, name_len);
8535 btrfs_i_size_write(inode, name_len - 1);
8536 err = btrfs_update_inode(trans, root, inode);
8542 d_instantiate(dentry, inode);
8543 btrfs_end_transaction(trans, root);
8545 inode_dec_link_count(inode);
8548 btrfs_btree_balance_dirty(root);
8552 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8553 u64 start, u64 num_bytes, u64 min_size,
8554 loff_t actual_len, u64 *alloc_hint,
8555 struct btrfs_trans_handle *trans)
8557 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8558 struct extent_map *em;
8559 struct btrfs_root *root = BTRFS_I(inode)->root;
8560 struct btrfs_key ins;
8561 u64 cur_offset = start;
8565 bool own_trans = true;
8569 while (num_bytes > 0) {
8571 trans = btrfs_start_transaction(root, 3);
8572 if (IS_ERR(trans)) {
8573 ret = PTR_ERR(trans);
8578 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8579 cur_bytes = max(cur_bytes, min_size);
8580 ret = btrfs_reserve_extent(trans, root, cur_bytes,
8581 min_size, 0, *alloc_hint, &ins, 1);
8584 btrfs_end_transaction(trans, root);
8588 ret = insert_reserved_file_extent(trans, inode,
8589 cur_offset, ins.objectid,
8590 ins.offset, ins.offset,
8591 ins.offset, 0, 0, 0,
8592 BTRFS_FILE_EXTENT_PREALLOC);
8594 btrfs_abort_transaction(trans, root, ret);
8596 btrfs_end_transaction(trans, root);
8599 btrfs_drop_extent_cache(inode, cur_offset,
8600 cur_offset + ins.offset -1, 0);
8602 em = alloc_extent_map();
8604 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8605 &BTRFS_I(inode)->runtime_flags);
8609 em->start = cur_offset;
8610 em->orig_start = cur_offset;
8611 em->len = ins.offset;
8612 em->block_start = ins.objectid;
8613 em->block_len = ins.offset;
8614 em->orig_block_len = ins.offset;
8615 em->ram_bytes = ins.offset;
8616 em->bdev = root->fs_info->fs_devices->latest_bdev;
8617 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8618 em->generation = trans->transid;
8621 write_lock(&em_tree->lock);
8622 ret = add_extent_mapping(em_tree, em, 1);
8623 write_unlock(&em_tree->lock);
8626 btrfs_drop_extent_cache(inode, cur_offset,
8627 cur_offset + ins.offset - 1,
8630 free_extent_map(em);
8632 num_bytes -= ins.offset;
8633 cur_offset += ins.offset;
8634 *alloc_hint = ins.objectid + ins.offset;
8636 inode_inc_iversion(inode);
8637 inode->i_ctime = CURRENT_TIME;
8638 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8639 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8640 (actual_len > inode->i_size) &&
8641 (cur_offset > inode->i_size)) {
8642 if (cur_offset > actual_len)
8643 i_size = actual_len;
8645 i_size = cur_offset;
8646 i_size_write(inode, i_size);
8647 btrfs_ordered_update_i_size(inode, i_size, NULL);
8650 ret = btrfs_update_inode(trans, root, inode);
8653 btrfs_abort_transaction(trans, root, ret);
8655 btrfs_end_transaction(trans, root);
8660 btrfs_end_transaction(trans, root);
8665 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8666 u64 start, u64 num_bytes, u64 min_size,
8667 loff_t actual_len, u64 *alloc_hint)
8669 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8670 min_size, actual_len, alloc_hint,
8674 int btrfs_prealloc_file_range_trans(struct inode *inode,
8675 struct btrfs_trans_handle *trans, int mode,
8676 u64 start, u64 num_bytes, u64 min_size,
8677 loff_t actual_len, u64 *alloc_hint)
8679 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8680 min_size, actual_len, alloc_hint, trans);
8683 static int btrfs_set_page_dirty(struct page *page)
8685 return __set_page_dirty_nobuffers(page);
8688 static int btrfs_permission(struct inode *inode, int mask)
8690 struct btrfs_root *root = BTRFS_I(inode)->root;
8691 umode_t mode = inode->i_mode;
8693 if (mask & MAY_WRITE &&
8694 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8695 if (btrfs_root_readonly(root))
8697 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8700 return generic_permission(inode, mask);
8703 static const struct inode_operations btrfs_dir_inode_operations = {
8704 .getattr = btrfs_getattr,
8705 .lookup = btrfs_lookup,
8706 .create = btrfs_create,
8707 .unlink = btrfs_unlink,
8709 .mkdir = btrfs_mkdir,
8710 .rmdir = btrfs_rmdir,
8711 .rename = btrfs_rename,
8712 .symlink = btrfs_symlink,
8713 .setattr = btrfs_setattr,
8714 .mknod = btrfs_mknod,
8715 .setxattr = btrfs_setxattr,
8716 .getxattr = btrfs_getxattr,
8717 .listxattr = btrfs_listxattr,
8718 .removexattr = btrfs_removexattr,
8719 .permission = btrfs_permission,
8720 .get_acl = btrfs_get_acl,
8722 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8723 .lookup = btrfs_lookup,
8724 .permission = btrfs_permission,
8725 .get_acl = btrfs_get_acl,
8728 static const struct file_operations btrfs_dir_file_operations = {
8729 .llseek = generic_file_llseek,
8730 .read = generic_read_dir,
8731 .readdir = btrfs_real_readdir,
8732 .unlocked_ioctl = btrfs_ioctl,
8733 #ifdef CONFIG_COMPAT
8734 .compat_ioctl = btrfs_ioctl,
8736 .release = btrfs_release_file,
8737 .fsync = btrfs_sync_file,
8740 static struct extent_io_ops btrfs_extent_io_ops = {
8741 .fill_delalloc = run_delalloc_range,
8742 .submit_bio_hook = btrfs_submit_bio_hook,
8743 .merge_bio_hook = btrfs_merge_bio_hook,
8744 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
8745 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
8746 .writepage_start_hook = btrfs_writepage_start_hook,
8747 .set_bit_hook = btrfs_set_bit_hook,
8748 .clear_bit_hook = btrfs_clear_bit_hook,
8749 .merge_extent_hook = btrfs_merge_extent_hook,
8750 .split_extent_hook = btrfs_split_extent_hook,
8754 * btrfs doesn't support the bmap operation because swapfiles
8755 * use bmap to make a mapping of extents in the file. They assume
8756 * these extents won't change over the life of the file and they
8757 * use the bmap result to do IO directly to the drive.
8759 * the btrfs bmap call would return logical addresses that aren't
8760 * suitable for IO and they also will change frequently as COW
8761 * operations happen. So, swapfile + btrfs == corruption.
8763 * For now we're avoiding this by dropping bmap.
8765 static const struct address_space_operations btrfs_aops = {
8766 .readpage = btrfs_readpage,
8767 .writepage = btrfs_writepage,
8768 .writepages = btrfs_writepages,
8769 .readpages = btrfs_readpages,
8770 .direct_IO = btrfs_direct_IO,
8771 .invalidatepage = btrfs_invalidatepage,
8772 .releasepage = btrfs_releasepage,
8773 .set_page_dirty = btrfs_set_page_dirty,
8774 .error_remove_page = generic_error_remove_page,
8777 static const struct address_space_operations btrfs_symlink_aops = {
8778 .readpage = btrfs_readpage,
8779 .writepage = btrfs_writepage,
8780 .invalidatepage = btrfs_invalidatepage,
8781 .releasepage = btrfs_releasepage,
8784 static const struct inode_operations btrfs_file_inode_operations = {
8785 .getattr = btrfs_getattr,
8786 .setattr = btrfs_setattr,
8787 .setxattr = btrfs_setxattr,
8788 .getxattr = btrfs_getxattr,
8789 .listxattr = btrfs_listxattr,
8790 .removexattr = btrfs_removexattr,
8791 .permission = btrfs_permission,
8792 .fiemap = btrfs_fiemap,
8793 .get_acl = btrfs_get_acl,
8794 .update_time = btrfs_update_time,
8796 static const struct inode_operations btrfs_special_inode_operations = {
8797 .getattr = btrfs_getattr,
8798 .setattr = btrfs_setattr,
8799 .permission = btrfs_permission,
8800 .setxattr = btrfs_setxattr,
8801 .getxattr = btrfs_getxattr,
8802 .listxattr = btrfs_listxattr,
8803 .removexattr = btrfs_removexattr,
8804 .get_acl = btrfs_get_acl,
8805 .update_time = btrfs_update_time,
8807 static const struct inode_operations btrfs_symlink_inode_operations = {
8808 .readlink = generic_readlink,
8809 .follow_link = page_follow_link_light,
8810 .put_link = page_put_link,
8811 .getattr = btrfs_getattr,
8812 .setattr = btrfs_setattr,
8813 .permission = btrfs_permission,
8814 .setxattr = btrfs_setxattr,
8815 .getxattr = btrfs_getxattr,
8816 .listxattr = btrfs_listxattr,
8817 .removexattr = btrfs_removexattr,
8818 .get_acl = btrfs_get_acl,
8819 .update_time = btrfs_update_time,
8822 const struct dentry_operations btrfs_dentry_operations = {
8823 .d_delete = btrfs_dentry_delete,
8824 .d_release = btrfs_dentry_release,