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/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include <linux/mount.h>
45 #include "transaction.h"
46 #include "btrfs_inode.h"
48 #include "print-tree.h"
49 #include "ordered-data.h"
53 #include "compression.h"
55 #include "free-space-cache.h"
56 #include "inode-map.h"
58 struct btrfs_iget_args {
60 struct btrfs_root *root;
63 static const struct inode_operations btrfs_dir_inode_operations;
64 static const struct inode_operations btrfs_symlink_inode_operations;
65 static const struct inode_operations btrfs_dir_ro_inode_operations;
66 static const struct inode_operations btrfs_special_inode_operations;
67 static const struct inode_operations btrfs_file_inode_operations;
68 static const struct address_space_operations btrfs_aops;
69 static const struct address_space_operations btrfs_symlink_aops;
70 static const struct file_operations btrfs_dir_file_operations;
71 static struct extent_io_ops btrfs_extent_io_ops;
73 static struct kmem_cache *btrfs_inode_cachep;
74 static struct kmem_cache *btrfs_delalloc_work_cachep;
75 struct kmem_cache *btrfs_trans_handle_cachep;
76 struct kmem_cache *btrfs_transaction_cachep;
77 struct kmem_cache *btrfs_path_cachep;
78 struct kmem_cache *btrfs_free_space_cachep;
81 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
82 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
83 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
84 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
85 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
86 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
87 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
88 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
91 static int btrfs_setsize(struct inode *inode, loff_t newsize);
92 static int btrfs_truncate(struct inode *inode);
93 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
94 static noinline int cow_file_range(struct inode *inode,
95 struct page *locked_page,
96 u64 start, u64 end, int *page_started,
97 unsigned long *nr_written, int unlock);
99 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
100 struct inode *inode, struct inode *dir,
101 const struct qstr *qstr)
105 err = btrfs_init_acl(trans, inode, dir);
107 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
112 * this does all the hard work for inserting an inline extent into
113 * the btree. The caller should have done a btrfs_drop_extents so that
114 * no overlapping inline items exist in the btree
116 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
117 struct btrfs_root *root, struct inode *inode,
118 u64 start, size_t size, size_t compressed_size,
120 struct page **compressed_pages)
122 struct btrfs_key key;
123 struct btrfs_path *path;
124 struct extent_buffer *leaf;
125 struct page *page = NULL;
128 struct btrfs_file_extent_item *ei;
131 size_t cur_size = size;
133 unsigned long offset;
135 if (compressed_size && compressed_pages)
136 cur_size = compressed_size;
138 path = btrfs_alloc_path();
142 path->leave_spinning = 1;
144 key.objectid = btrfs_ino(inode);
146 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
147 datasize = btrfs_file_extent_calc_inline_size(cur_size);
149 inode_add_bytes(inode, size);
150 ret = btrfs_insert_empty_item(trans, root, path, &key,
156 leaf = path->nodes[0];
157 ei = btrfs_item_ptr(leaf, path->slots[0],
158 struct btrfs_file_extent_item);
159 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
160 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
161 btrfs_set_file_extent_encryption(leaf, ei, 0);
162 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
163 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
164 ptr = btrfs_file_extent_inline_start(ei);
166 if (compress_type != BTRFS_COMPRESS_NONE) {
169 while (compressed_size > 0) {
170 cpage = compressed_pages[i];
171 cur_size = min_t(unsigned long, compressed_size,
174 kaddr = kmap_atomic(cpage);
175 write_extent_buffer(leaf, kaddr, ptr, cur_size);
176 kunmap_atomic(kaddr);
180 compressed_size -= cur_size;
182 btrfs_set_file_extent_compression(leaf, ei,
185 page = find_get_page(inode->i_mapping,
186 start >> PAGE_CACHE_SHIFT);
187 btrfs_set_file_extent_compression(leaf, ei, 0);
188 kaddr = kmap_atomic(page);
189 offset = start & (PAGE_CACHE_SIZE - 1);
190 write_extent_buffer(leaf, kaddr + offset, ptr, size);
191 kunmap_atomic(kaddr);
192 page_cache_release(page);
194 btrfs_mark_buffer_dirty(leaf);
195 btrfs_free_path(path);
198 * we're an inline extent, so nobody can
199 * extend the file past i_size without locking
200 * a page we already have locked.
202 * We must do any isize and inode updates
203 * before we unlock the pages. Otherwise we
204 * could end up racing with unlink.
206 BTRFS_I(inode)->disk_i_size = inode->i_size;
207 ret = btrfs_update_inode(trans, root, inode);
211 btrfs_free_path(path);
217 * conditionally insert an inline extent into the file. This
218 * does the checks required to make sure the data is small enough
219 * to fit as an inline extent.
221 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
222 struct btrfs_root *root,
223 struct inode *inode, u64 start, u64 end,
224 size_t compressed_size, int compress_type,
225 struct page **compressed_pages)
227 u64 isize = i_size_read(inode);
228 u64 actual_end = min(end + 1, isize);
229 u64 inline_len = actual_end - start;
230 u64 aligned_end = (end + root->sectorsize - 1) &
231 ~((u64)root->sectorsize - 1);
232 u64 data_len = inline_len;
236 data_len = compressed_size;
239 actual_end >= PAGE_CACHE_SIZE ||
240 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
242 (actual_end & (root->sectorsize - 1)) == 0) ||
244 data_len > root->fs_info->max_inline) {
248 ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
252 if (isize > actual_end)
253 inline_len = min_t(u64, isize, actual_end);
254 ret = insert_inline_extent(trans, root, inode, start,
255 inline_len, compressed_size,
256 compress_type, compressed_pages);
257 if (ret && ret != -ENOSPC) {
258 btrfs_abort_transaction(trans, root, ret);
260 } else if (ret == -ENOSPC) {
264 btrfs_delalloc_release_metadata(inode, end + 1 - start);
265 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
269 struct async_extent {
274 unsigned long nr_pages;
276 struct list_head list;
281 struct btrfs_root *root;
282 struct page *locked_page;
285 struct list_head extents;
286 struct btrfs_work work;
289 static noinline int add_async_extent(struct async_cow *cow,
290 u64 start, u64 ram_size,
293 unsigned long nr_pages,
296 struct async_extent *async_extent;
298 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
299 BUG_ON(!async_extent); /* -ENOMEM */
300 async_extent->start = start;
301 async_extent->ram_size = ram_size;
302 async_extent->compressed_size = compressed_size;
303 async_extent->pages = pages;
304 async_extent->nr_pages = nr_pages;
305 async_extent->compress_type = compress_type;
306 list_add_tail(&async_extent->list, &cow->extents);
311 * we create compressed extents in two phases. The first
312 * phase compresses a range of pages that have already been
313 * locked (both pages and state bits are locked).
315 * This is done inside an ordered work queue, and the compression
316 * is spread across many cpus. The actual IO submission is step
317 * two, and the ordered work queue takes care of making sure that
318 * happens in the same order things were put onto the queue by
319 * writepages and friends.
321 * If this code finds it can't get good compression, it puts an
322 * entry onto the work queue to write the uncompressed bytes. This
323 * makes sure that both compressed inodes and uncompressed inodes
324 * are written in the same order that the flusher thread sent them
327 static noinline int compress_file_range(struct inode *inode,
328 struct page *locked_page,
330 struct async_cow *async_cow,
333 struct btrfs_root *root = BTRFS_I(inode)->root;
334 struct btrfs_trans_handle *trans;
336 u64 blocksize = root->sectorsize;
338 u64 isize = i_size_read(inode);
340 struct page **pages = NULL;
341 unsigned long nr_pages;
342 unsigned long nr_pages_ret = 0;
343 unsigned long total_compressed = 0;
344 unsigned long total_in = 0;
345 unsigned long max_compressed = 128 * 1024;
346 unsigned long max_uncompressed = 128 * 1024;
349 int compress_type = root->fs_info->compress_type;
351 /* if this is a small write inside eof, kick off a defrag */
352 if ((end - start + 1) < 16 * 1024 &&
353 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
354 btrfs_add_inode_defrag(NULL, inode);
356 actual_end = min_t(u64, isize, end + 1);
359 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
360 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
363 * we don't want to send crud past the end of i_size through
364 * compression, that's just a waste of CPU time. So, if the
365 * end of the file is before the start of our current
366 * requested range of bytes, we bail out to the uncompressed
367 * cleanup code that can deal with all of this.
369 * It isn't really the fastest way to fix things, but this is a
370 * very uncommon corner.
372 if (actual_end <= start)
373 goto cleanup_and_bail_uncompressed;
375 total_compressed = actual_end - start;
377 /* we want to make sure that amount of ram required to uncompress
378 * an extent is reasonable, so we limit the total size in ram
379 * of a compressed extent to 128k. This is a crucial number
380 * because it also controls how easily we can spread reads across
381 * cpus for decompression.
383 * We also want to make sure the amount of IO required to do
384 * a random read is reasonably small, so we limit the size of
385 * a compressed extent to 128k.
387 total_compressed = min(total_compressed, max_uncompressed);
388 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
389 num_bytes = max(blocksize, num_bytes);
394 * we do compression for mount -o compress and when the
395 * inode has not been flagged as nocompress. This flag can
396 * change at any time if we discover bad compression ratios.
398 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
399 (btrfs_test_opt(root, COMPRESS) ||
400 (BTRFS_I(inode)->force_compress) ||
401 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
403 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
405 /* just bail out to the uncompressed code */
409 if (BTRFS_I(inode)->force_compress)
410 compress_type = BTRFS_I(inode)->force_compress;
412 ret = btrfs_compress_pages(compress_type,
413 inode->i_mapping, start,
414 total_compressed, pages,
415 nr_pages, &nr_pages_ret,
421 unsigned long offset = total_compressed &
422 (PAGE_CACHE_SIZE - 1);
423 struct page *page = pages[nr_pages_ret - 1];
426 /* zero the tail end of the last page, we might be
427 * sending it down to disk
430 kaddr = kmap_atomic(page);
431 memset(kaddr + offset, 0,
432 PAGE_CACHE_SIZE - offset);
433 kunmap_atomic(kaddr);
440 trans = btrfs_join_transaction(root);
442 ret = PTR_ERR(trans);
444 goto cleanup_and_out;
446 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
448 /* lets try to make an inline extent */
449 if (ret || total_in < (actual_end - start)) {
450 /* we didn't compress the entire range, try
451 * to make an uncompressed inline extent.
453 ret = cow_file_range_inline(trans, root, inode,
454 start, end, 0, 0, NULL);
456 /* try making a compressed inline extent */
457 ret = cow_file_range_inline(trans, root, inode,
460 compress_type, pages);
464 * inline extent creation worked or returned error,
465 * we don't need to create any more async work items.
466 * Unlock and free up our temp pages.
468 extent_clear_unlock_delalloc(inode,
469 &BTRFS_I(inode)->io_tree,
471 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
472 EXTENT_CLEAR_DELALLOC |
473 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
475 btrfs_end_transaction(trans, root);
478 btrfs_end_transaction(trans, root);
483 * we aren't doing an inline extent round the compressed size
484 * up to a block size boundary so the allocator does sane
487 total_compressed = (total_compressed + blocksize - 1) &
491 * one last check to make sure the compression is really a
492 * win, compare the page count read with the blocks on disk
494 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
495 ~(PAGE_CACHE_SIZE - 1);
496 if (total_compressed >= total_in) {
499 num_bytes = total_in;
502 if (!will_compress && pages) {
504 * the compression code ran but failed to make things smaller,
505 * free any pages it allocated and our page pointer array
507 for (i = 0; i < nr_pages_ret; i++) {
508 WARN_ON(pages[i]->mapping);
509 page_cache_release(pages[i]);
513 total_compressed = 0;
516 /* flag the file so we don't compress in the future */
517 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
518 !(BTRFS_I(inode)->force_compress)) {
519 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
525 /* the async work queues will take care of doing actual
526 * allocation on disk for these compressed pages,
527 * and will submit them to the elevator.
529 add_async_extent(async_cow, start, num_bytes,
530 total_compressed, pages, nr_pages_ret,
533 if (start + num_bytes < end) {
540 cleanup_and_bail_uncompressed:
542 * No compression, but we still need to write the pages in
543 * the file we've been given so far. redirty the locked
544 * page if it corresponds to our extent and set things up
545 * for the async work queue to run cow_file_range to do
546 * the normal delalloc dance
548 if (page_offset(locked_page) >= start &&
549 page_offset(locked_page) <= end) {
550 __set_page_dirty_nobuffers(locked_page);
551 /* unlocked later on in the async handlers */
553 add_async_extent(async_cow, start, end - start + 1,
554 0, NULL, 0, BTRFS_COMPRESS_NONE);
562 for (i = 0; i < nr_pages_ret; i++) {
563 WARN_ON(pages[i]->mapping);
564 page_cache_release(pages[i]);
571 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
573 EXTENT_CLEAR_UNLOCK_PAGE |
575 EXTENT_CLEAR_DELALLOC |
576 EXTENT_SET_WRITEBACK |
577 EXTENT_END_WRITEBACK);
578 if (!trans || IS_ERR(trans))
579 btrfs_error(root->fs_info, ret, "Failed to join transaction");
581 btrfs_abort_transaction(trans, root, ret);
586 * phase two of compressed writeback. This is the ordered portion
587 * of the code, which only gets called in the order the work was
588 * queued. We walk all the async extents created by compress_file_range
589 * and send them down to the disk.
591 static noinline int submit_compressed_extents(struct inode *inode,
592 struct async_cow *async_cow)
594 struct async_extent *async_extent;
596 struct btrfs_trans_handle *trans;
597 struct btrfs_key ins;
598 struct extent_map *em;
599 struct btrfs_root *root = BTRFS_I(inode)->root;
600 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
601 struct extent_io_tree *io_tree;
604 if (list_empty(&async_cow->extents))
608 while (!list_empty(&async_cow->extents)) {
609 async_extent = list_entry(async_cow->extents.next,
610 struct async_extent, list);
611 list_del(&async_extent->list);
613 io_tree = &BTRFS_I(inode)->io_tree;
616 /* did the compression code fall back to uncompressed IO? */
617 if (!async_extent->pages) {
618 int page_started = 0;
619 unsigned long nr_written = 0;
621 lock_extent(io_tree, async_extent->start,
622 async_extent->start +
623 async_extent->ram_size - 1);
625 /* allocate blocks */
626 ret = cow_file_range(inode, async_cow->locked_page,
628 async_extent->start +
629 async_extent->ram_size - 1,
630 &page_started, &nr_written, 0);
635 * if page_started, cow_file_range inserted an
636 * inline extent and took care of all the unlocking
637 * and IO for us. Otherwise, we need to submit
638 * all those pages down to the drive.
640 if (!page_started && !ret)
641 extent_write_locked_range(io_tree,
642 inode, async_extent->start,
643 async_extent->start +
644 async_extent->ram_size - 1,
652 lock_extent(io_tree, async_extent->start,
653 async_extent->start + async_extent->ram_size - 1);
655 trans = btrfs_join_transaction(root);
657 ret = PTR_ERR(trans);
659 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
660 ret = btrfs_reserve_extent(trans, root,
661 async_extent->compressed_size,
662 async_extent->compressed_size,
663 0, alloc_hint, &ins, 1);
664 if (ret && ret != -ENOSPC)
665 btrfs_abort_transaction(trans, root, ret);
666 btrfs_end_transaction(trans, root);
671 for (i = 0; i < async_extent->nr_pages; i++) {
672 WARN_ON(async_extent->pages[i]->mapping);
673 page_cache_release(async_extent->pages[i]);
675 kfree(async_extent->pages);
676 async_extent->nr_pages = 0;
677 async_extent->pages = NULL;
678 unlock_extent(io_tree, async_extent->start,
679 async_extent->start +
680 async_extent->ram_size - 1);
683 goto out_free; /* JDM: Requeue? */
687 * here we're doing allocation and writeback of the
690 btrfs_drop_extent_cache(inode, async_extent->start,
691 async_extent->start +
692 async_extent->ram_size - 1, 0);
694 em = alloc_extent_map();
695 BUG_ON(!em); /* -ENOMEM */
696 em->start = async_extent->start;
697 em->len = async_extent->ram_size;
698 em->orig_start = em->start;
700 em->block_start = ins.objectid;
701 em->block_len = ins.offset;
702 em->bdev = root->fs_info->fs_devices->latest_bdev;
703 em->compress_type = async_extent->compress_type;
704 set_bit(EXTENT_FLAG_PINNED, &em->flags);
705 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
708 write_lock(&em_tree->lock);
709 ret = add_extent_mapping(em_tree, em);
710 write_unlock(&em_tree->lock);
711 if (ret != -EEXIST) {
715 btrfs_drop_extent_cache(inode, async_extent->start,
716 async_extent->start +
717 async_extent->ram_size - 1, 0);
720 ret = btrfs_add_ordered_extent_compress(inode,
723 async_extent->ram_size,
725 BTRFS_ORDERED_COMPRESSED,
726 async_extent->compress_type);
727 BUG_ON(ret); /* -ENOMEM */
730 * clear dirty, set writeback and unlock the pages.
732 extent_clear_unlock_delalloc(inode,
733 &BTRFS_I(inode)->io_tree,
735 async_extent->start +
736 async_extent->ram_size - 1,
737 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
738 EXTENT_CLEAR_UNLOCK |
739 EXTENT_CLEAR_DELALLOC |
740 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
742 ret = btrfs_submit_compressed_write(inode,
744 async_extent->ram_size,
746 ins.offset, async_extent->pages,
747 async_extent->nr_pages);
749 BUG_ON(ret); /* -ENOMEM */
750 alloc_hint = ins.objectid + ins.offset;
762 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
765 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
766 struct extent_map *em;
769 read_lock(&em_tree->lock);
770 em = search_extent_mapping(em_tree, start, num_bytes);
773 * if block start isn't an actual block number then find the
774 * first block in this inode and use that as a hint. If that
775 * block is also bogus then just don't worry about it.
777 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
779 em = search_extent_mapping(em_tree, 0, 0);
780 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
781 alloc_hint = em->block_start;
785 alloc_hint = em->block_start;
789 read_unlock(&em_tree->lock);
795 * when extent_io.c finds a delayed allocation range in the file,
796 * the call backs end up in this code. The basic idea is to
797 * allocate extents on disk for the range, and create ordered data structs
798 * in ram to track those extents.
800 * locked_page is the page that writepage had locked already. We use
801 * it to make sure we don't do extra locks or unlocks.
803 * *page_started is set to one if we unlock locked_page and do everything
804 * required to start IO on it. It may be clean and already done with
807 static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
809 struct btrfs_root *root,
810 struct page *locked_page,
811 u64 start, u64 end, int *page_started,
812 unsigned long *nr_written,
817 unsigned long ram_size;
820 u64 blocksize = root->sectorsize;
821 struct btrfs_key ins;
822 struct extent_map *em;
823 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
826 BUG_ON(btrfs_is_free_space_inode(inode));
828 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
829 num_bytes = max(blocksize, num_bytes);
830 disk_num_bytes = num_bytes;
832 /* if this is a small write inside eof, kick off defrag */
833 if (num_bytes < 64 * 1024 &&
834 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
835 btrfs_add_inode_defrag(trans, inode);
838 /* lets try to make an inline extent */
839 ret = cow_file_range_inline(trans, root, inode,
840 start, end, 0, 0, NULL);
842 extent_clear_unlock_delalloc(inode,
843 &BTRFS_I(inode)->io_tree,
845 EXTENT_CLEAR_UNLOCK_PAGE |
846 EXTENT_CLEAR_UNLOCK |
847 EXTENT_CLEAR_DELALLOC |
849 EXTENT_SET_WRITEBACK |
850 EXTENT_END_WRITEBACK);
852 *nr_written = *nr_written +
853 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
856 } else if (ret < 0) {
857 btrfs_abort_transaction(trans, root, ret);
862 BUG_ON(disk_num_bytes >
863 btrfs_super_total_bytes(root->fs_info->super_copy));
865 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
866 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
868 while (disk_num_bytes > 0) {
871 cur_alloc_size = disk_num_bytes;
872 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
873 root->sectorsize, 0, alloc_hint,
876 btrfs_abort_transaction(trans, root, ret);
880 em = alloc_extent_map();
881 BUG_ON(!em); /* -ENOMEM */
883 em->orig_start = em->start;
884 ram_size = ins.offset;
885 em->len = ins.offset;
887 em->block_start = ins.objectid;
888 em->block_len = ins.offset;
889 em->bdev = root->fs_info->fs_devices->latest_bdev;
890 set_bit(EXTENT_FLAG_PINNED, &em->flags);
893 write_lock(&em_tree->lock);
894 ret = add_extent_mapping(em_tree, em);
895 write_unlock(&em_tree->lock);
896 if (ret != -EEXIST) {
900 btrfs_drop_extent_cache(inode, start,
901 start + ram_size - 1, 0);
904 cur_alloc_size = ins.offset;
905 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
906 ram_size, cur_alloc_size, 0);
907 BUG_ON(ret); /* -ENOMEM */
909 if (root->root_key.objectid ==
910 BTRFS_DATA_RELOC_TREE_OBJECTID) {
911 ret = btrfs_reloc_clone_csums(inode, start,
914 btrfs_abort_transaction(trans, root, ret);
919 if (disk_num_bytes < cur_alloc_size)
922 /* we're not doing compressed IO, don't unlock the first
923 * page (which the caller expects to stay locked), don't
924 * clear any dirty bits and don't set any writeback bits
926 * Do set the Private2 bit so we know this page was properly
927 * setup for writepage
929 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
930 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
933 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
934 start, start + ram_size - 1,
936 disk_num_bytes -= cur_alloc_size;
937 num_bytes -= cur_alloc_size;
938 alloc_hint = ins.objectid + ins.offset;
939 start += cur_alloc_size;
945 extent_clear_unlock_delalloc(inode,
946 &BTRFS_I(inode)->io_tree,
947 start, end, locked_page,
948 EXTENT_CLEAR_UNLOCK_PAGE |
949 EXTENT_CLEAR_UNLOCK |
950 EXTENT_CLEAR_DELALLOC |
952 EXTENT_SET_WRITEBACK |
953 EXTENT_END_WRITEBACK);
958 static noinline int cow_file_range(struct inode *inode,
959 struct page *locked_page,
960 u64 start, u64 end, int *page_started,
961 unsigned long *nr_written,
964 struct btrfs_trans_handle *trans;
965 struct btrfs_root *root = BTRFS_I(inode)->root;
968 trans = btrfs_join_transaction(root);
970 extent_clear_unlock_delalloc(inode,
971 &BTRFS_I(inode)->io_tree,
972 start, end, locked_page,
973 EXTENT_CLEAR_UNLOCK_PAGE |
974 EXTENT_CLEAR_UNLOCK |
975 EXTENT_CLEAR_DELALLOC |
977 EXTENT_SET_WRITEBACK |
978 EXTENT_END_WRITEBACK);
979 return PTR_ERR(trans);
981 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
983 ret = __cow_file_range(trans, inode, root, locked_page, start, end,
984 page_started, nr_written, unlock);
986 btrfs_end_transaction(trans, root);
992 * work queue call back to started compression on a file and pages
994 static noinline void async_cow_start(struct btrfs_work *work)
996 struct async_cow *async_cow;
998 async_cow = container_of(work, struct async_cow, work);
1000 compress_file_range(async_cow->inode, async_cow->locked_page,
1001 async_cow->start, async_cow->end, async_cow,
1003 if (num_added == 0) {
1004 btrfs_add_delayed_iput(async_cow->inode);
1005 async_cow->inode = NULL;
1010 * work queue call back to submit previously compressed pages
1012 static noinline void async_cow_submit(struct btrfs_work *work)
1014 struct async_cow *async_cow;
1015 struct btrfs_root *root;
1016 unsigned long nr_pages;
1018 async_cow = container_of(work, struct async_cow, work);
1020 root = async_cow->root;
1021 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1024 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1026 waitqueue_active(&root->fs_info->async_submit_wait))
1027 wake_up(&root->fs_info->async_submit_wait);
1029 if (async_cow->inode)
1030 submit_compressed_extents(async_cow->inode, async_cow);
1033 static noinline void async_cow_free(struct btrfs_work *work)
1035 struct async_cow *async_cow;
1036 async_cow = container_of(work, struct async_cow, work);
1037 if (async_cow->inode)
1038 btrfs_add_delayed_iput(async_cow->inode);
1042 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1043 u64 start, u64 end, int *page_started,
1044 unsigned long *nr_written)
1046 struct async_cow *async_cow;
1047 struct btrfs_root *root = BTRFS_I(inode)->root;
1048 unsigned long nr_pages;
1050 int limit = 10 * 1024 * 1024;
1052 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1053 1, 0, NULL, GFP_NOFS);
1054 while (start < end) {
1055 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1056 BUG_ON(!async_cow); /* -ENOMEM */
1057 async_cow->inode = igrab(inode);
1058 async_cow->root = root;
1059 async_cow->locked_page = locked_page;
1060 async_cow->start = start;
1062 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1065 cur_end = min(end, start + 512 * 1024 - 1);
1067 async_cow->end = cur_end;
1068 INIT_LIST_HEAD(&async_cow->extents);
1070 async_cow->work.func = async_cow_start;
1071 async_cow->work.ordered_func = async_cow_submit;
1072 async_cow->work.ordered_free = async_cow_free;
1073 async_cow->work.flags = 0;
1075 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1077 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1079 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1082 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1083 wait_event(root->fs_info->async_submit_wait,
1084 (atomic_read(&root->fs_info->async_delalloc_pages) <
1088 while (atomic_read(&root->fs_info->async_submit_draining) &&
1089 atomic_read(&root->fs_info->async_delalloc_pages)) {
1090 wait_event(root->fs_info->async_submit_wait,
1091 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1095 *nr_written += nr_pages;
1096 start = cur_end + 1;
1102 static noinline int csum_exist_in_range(struct btrfs_root *root,
1103 u64 bytenr, u64 num_bytes)
1106 struct btrfs_ordered_sum *sums;
1109 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1110 bytenr + num_bytes - 1, &list, 0);
1111 if (ret == 0 && list_empty(&list))
1114 while (!list_empty(&list)) {
1115 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1116 list_del(&sums->list);
1123 * when nowcow writeback call back. This checks for snapshots or COW copies
1124 * of the extents that exist in the file, and COWs the file as required.
1126 * If no cow copies or snapshots exist, we write directly to the existing
1129 static noinline int run_delalloc_nocow(struct inode *inode,
1130 struct page *locked_page,
1131 u64 start, u64 end, int *page_started, int force,
1132 unsigned long *nr_written)
1134 struct btrfs_root *root = BTRFS_I(inode)->root;
1135 struct btrfs_trans_handle *trans;
1136 struct extent_buffer *leaf;
1137 struct btrfs_path *path;
1138 struct btrfs_file_extent_item *fi;
1139 struct btrfs_key found_key;
1152 u64 ino = btrfs_ino(inode);
1154 path = btrfs_alloc_path();
1156 extent_clear_unlock_delalloc(inode,
1157 &BTRFS_I(inode)->io_tree,
1158 start, end, locked_page,
1159 EXTENT_CLEAR_UNLOCK_PAGE |
1160 EXTENT_CLEAR_UNLOCK |
1161 EXTENT_CLEAR_DELALLOC |
1162 EXTENT_CLEAR_DIRTY |
1163 EXTENT_SET_WRITEBACK |
1164 EXTENT_END_WRITEBACK);
1168 nolock = btrfs_is_free_space_inode(inode);
1171 trans = btrfs_join_transaction_nolock(root);
1173 trans = btrfs_join_transaction(root);
1175 if (IS_ERR(trans)) {
1176 extent_clear_unlock_delalloc(inode,
1177 &BTRFS_I(inode)->io_tree,
1178 start, end, locked_page,
1179 EXTENT_CLEAR_UNLOCK_PAGE |
1180 EXTENT_CLEAR_UNLOCK |
1181 EXTENT_CLEAR_DELALLOC |
1182 EXTENT_CLEAR_DIRTY |
1183 EXTENT_SET_WRITEBACK |
1184 EXTENT_END_WRITEBACK);
1185 btrfs_free_path(path);
1186 return PTR_ERR(trans);
1189 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1191 cow_start = (u64)-1;
1194 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1197 btrfs_abort_transaction(trans, root, ret);
1200 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1201 leaf = path->nodes[0];
1202 btrfs_item_key_to_cpu(leaf, &found_key,
1203 path->slots[0] - 1);
1204 if (found_key.objectid == ino &&
1205 found_key.type == BTRFS_EXTENT_DATA_KEY)
1210 leaf = path->nodes[0];
1211 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1212 ret = btrfs_next_leaf(root, path);
1214 btrfs_abort_transaction(trans, root, ret);
1219 leaf = path->nodes[0];
1225 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1227 if (found_key.objectid > ino ||
1228 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1229 found_key.offset > end)
1232 if (found_key.offset > cur_offset) {
1233 extent_end = found_key.offset;
1238 fi = btrfs_item_ptr(leaf, path->slots[0],
1239 struct btrfs_file_extent_item);
1240 extent_type = btrfs_file_extent_type(leaf, fi);
1242 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1243 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1244 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1245 extent_offset = btrfs_file_extent_offset(leaf, fi);
1246 extent_end = found_key.offset +
1247 btrfs_file_extent_num_bytes(leaf, fi);
1248 if (extent_end <= start) {
1252 if (disk_bytenr == 0)
1254 if (btrfs_file_extent_compression(leaf, fi) ||
1255 btrfs_file_extent_encryption(leaf, fi) ||
1256 btrfs_file_extent_other_encoding(leaf, fi))
1258 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1260 if (btrfs_extent_readonly(root, disk_bytenr))
1262 if (btrfs_cross_ref_exist(trans, root, ino,
1264 extent_offset, disk_bytenr))
1266 disk_bytenr += extent_offset;
1267 disk_bytenr += cur_offset - found_key.offset;
1268 num_bytes = min(end + 1, extent_end) - cur_offset;
1270 * force cow if csum exists in the range.
1271 * this ensure that csum for a given extent are
1272 * either valid or do not exist.
1274 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1277 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1278 extent_end = found_key.offset +
1279 btrfs_file_extent_inline_len(leaf, fi);
1280 extent_end = ALIGN(extent_end, root->sectorsize);
1285 if (extent_end <= start) {
1290 if (cow_start == (u64)-1)
1291 cow_start = cur_offset;
1292 cur_offset = extent_end;
1293 if (cur_offset > end)
1299 btrfs_release_path(path);
1300 if (cow_start != (u64)-1) {
1301 ret = __cow_file_range(trans, inode, root, locked_page,
1302 cow_start, found_key.offset - 1,
1303 page_started, nr_written, 1);
1305 btrfs_abort_transaction(trans, root, ret);
1308 cow_start = (u64)-1;
1311 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1312 struct extent_map *em;
1313 struct extent_map_tree *em_tree;
1314 em_tree = &BTRFS_I(inode)->extent_tree;
1315 em = alloc_extent_map();
1316 BUG_ON(!em); /* -ENOMEM */
1317 em->start = cur_offset;
1318 em->orig_start = em->start;
1319 em->len = num_bytes;
1320 em->block_len = num_bytes;
1321 em->block_start = disk_bytenr;
1322 em->bdev = root->fs_info->fs_devices->latest_bdev;
1323 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1324 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
1326 write_lock(&em_tree->lock);
1327 ret = add_extent_mapping(em_tree, em);
1328 write_unlock(&em_tree->lock);
1329 if (ret != -EEXIST) {
1330 free_extent_map(em);
1333 btrfs_drop_extent_cache(inode, em->start,
1334 em->start + em->len - 1, 0);
1336 type = BTRFS_ORDERED_PREALLOC;
1338 type = BTRFS_ORDERED_NOCOW;
1341 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1342 num_bytes, num_bytes, type);
1343 BUG_ON(ret); /* -ENOMEM */
1345 if (root->root_key.objectid ==
1346 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1347 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1350 btrfs_abort_transaction(trans, root, ret);
1355 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1356 cur_offset, cur_offset + num_bytes - 1,
1357 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1358 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1359 EXTENT_SET_PRIVATE2);
1360 cur_offset = extent_end;
1361 if (cur_offset > end)
1364 btrfs_release_path(path);
1366 if (cur_offset <= end && cow_start == (u64)-1) {
1367 cow_start = cur_offset;
1371 if (cow_start != (u64)-1) {
1372 ret = __cow_file_range(trans, inode, root, locked_page,
1374 page_started, nr_written, 1);
1376 btrfs_abort_transaction(trans, root, ret);
1382 err = btrfs_end_transaction(trans, root);
1386 if (ret && cur_offset < end)
1387 extent_clear_unlock_delalloc(inode,
1388 &BTRFS_I(inode)->io_tree,
1389 cur_offset, end, locked_page,
1390 EXTENT_CLEAR_UNLOCK_PAGE |
1391 EXTENT_CLEAR_UNLOCK |
1392 EXTENT_CLEAR_DELALLOC |
1393 EXTENT_CLEAR_DIRTY |
1394 EXTENT_SET_WRITEBACK |
1395 EXTENT_END_WRITEBACK);
1397 btrfs_free_path(path);
1402 * extent_io.c call back to do delayed allocation processing
1404 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1405 u64 start, u64 end, int *page_started,
1406 unsigned long *nr_written)
1409 struct btrfs_root *root = BTRFS_I(inode)->root;
1411 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1412 ret = run_delalloc_nocow(inode, locked_page, start, end,
1413 page_started, 1, nr_written);
1414 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1415 ret = run_delalloc_nocow(inode, locked_page, start, end,
1416 page_started, 0, nr_written);
1417 } else if (!btrfs_test_opt(root, COMPRESS) &&
1418 !(BTRFS_I(inode)->force_compress) &&
1419 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1420 ret = cow_file_range(inode, locked_page, start, end,
1421 page_started, nr_written, 1);
1423 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1424 &BTRFS_I(inode)->runtime_flags);
1425 ret = cow_file_range_async(inode, locked_page, start, end,
1426 page_started, nr_written);
1431 static void btrfs_split_extent_hook(struct inode *inode,
1432 struct extent_state *orig, u64 split)
1434 /* not delalloc, ignore it */
1435 if (!(orig->state & EXTENT_DELALLOC))
1438 spin_lock(&BTRFS_I(inode)->lock);
1439 BTRFS_I(inode)->outstanding_extents++;
1440 spin_unlock(&BTRFS_I(inode)->lock);
1444 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1445 * extents so we can keep track of new extents that are just merged onto old
1446 * extents, such as when we are doing sequential writes, so we can properly
1447 * account for the metadata space we'll need.
1449 static void btrfs_merge_extent_hook(struct inode *inode,
1450 struct extent_state *new,
1451 struct extent_state *other)
1453 /* not delalloc, ignore it */
1454 if (!(other->state & EXTENT_DELALLOC))
1457 spin_lock(&BTRFS_I(inode)->lock);
1458 BTRFS_I(inode)->outstanding_extents--;
1459 spin_unlock(&BTRFS_I(inode)->lock);
1463 * extent_io.c set_bit_hook, used to track delayed allocation
1464 * bytes in this file, and to maintain the list of inodes that
1465 * have pending delalloc work to be done.
1467 static void btrfs_set_bit_hook(struct inode *inode,
1468 struct extent_state *state, int *bits)
1472 * set_bit and clear bit hooks normally require _irqsave/restore
1473 * but in this case, we are only testing for the DELALLOC
1474 * bit, which is only set or cleared with irqs on
1476 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1477 struct btrfs_root *root = BTRFS_I(inode)->root;
1478 u64 len = state->end + 1 - state->start;
1479 bool do_list = !btrfs_is_free_space_inode(inode);
1481 if (*bits & EXTENT_FIRST_DELALLOC) {
1482 *bits &= ~EXTENT_FIRST_DELALLOC;
1484 spin_lock(&BTRFS_I(inode)->lock);
1485 BTRFS_I(inode)->outstanding_extents++;
1486 spin_unlock(&BTRFS_I(inode)->lock);
1489 spin_lock(&root->fs_info->delalloc_lock);
1490 BTRFS_I(inode)->delalloc_bytes += len;
1491 root->fs_info->delalloc_bytes += len;
1492 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1493 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1494 &root->fs_info->delalloc_inodes);
1496 spin_unlock(&root->fs_info->delalloc_lock);
1501 * extent_io.c clear_bit_hook, see set_bit_hook for why
1503 static void btrfs_clear_bit_hook(struct inode *inode,
1504 struct extent_state *state, int *bits)
1507 * set_bit and clear bit hooks normally require _irqsave/restore
1508 * but in this case, we are only testing for the DELALLOC
1509 * bit, which is only set or cleared with irqs on
1511 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1512 struct btrfs_root *root = BTRFS_I(inode)->root;
1513 u64 len = state->end + 1 - state->start;
1514 bool do_list = !btrfs_is_free_space_inode(inode);
1516 if (*bits & EXTENT_FIRST_DELALLOC) {
1517 *bits &= ~EXTENT_FIRST_DELALLOC;
1518 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1519 spin_lock(&BTRFS_I(inode)->lock);
1520 BTRFS_I(inode)->outstanding_extents--;
1521 spin_unlock(&BTRFS_I(inode)->lock);
1524 if (*bits & EXTENT_DO_ACCOUNTING)
1525 btrfs_delalloc_release_metadata(inode, len);
1527 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1529 btrfs_free_reserved_data_space(inode, len);
1531 spin_lock(&root->fs_info->delalloc_lock);
1532 root->fs_info->delalloc_bytes -= len;
1533 BTRFS_I(inode)->delalloc_bytes -= len;
1535 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1536 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1537 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1539 spin_unlock(&root->fs_info->delalloc_lock);
1544 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1545 * we don't create bios that span stripes or chunks
1547 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1548 size_t size, struct bio *bio,
1549 unsigned long bio_flags)
1551 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1552 struct btrfs_mapping_tree *map_tree;
1553 u64 logical = (u64)bio->bi_sector << 9;
1558 if (bio_flags & EXTENT_BIO_COMPRESSED)
1561 length = bio->bi_size;
1562 map_tree = &root->fs_info->mapping_tree;
1563 map_length = length;
1564 ret = btrfs_map_block(map_tree, READ, logical,
1565 &map_length, NULL, 0);
1566 /* Will always return 0 or 1 with map_multi == NULL */
1568 if (map_length < length + size)
1574 * in order to insert checksums into the metadata in large chunks,
1575 * we wait until bio submission time. All the pages in the bio are
1576 * checksummed and sums are attached onto the ordered extent record.
1578 * At IO completion time the cums attached on the ordered extent record
1579 * are inserted into the btree
1581 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1582 struct bio *bio, int mirror_num,
1583 unsigned long bio_flags,
1586 struct btrfs_root *root = BTRFS_I(inode)->root;
1589 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1590 BUG_ON(ret); /* -ENOMEM */
1595 * in order to insert checksums into the metadata in large chunks,
1596 * we wait until bio submission time. All the pages in the bio are
1597 * checksummed and sums are attached onto the ordered extent record.
1599 * At IO completion time the cums attached on the ordered extent record
1600 * are inserted into the btree
1602 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1603 int mirror_num, unsigned long bio_flags,
1606 struct btrfs_root *root = BTRFS_I(inode)->root;
1607 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1611 * extent_io.c submission hook. This does the right thing for csum calculation
1612 * on write, or reading the csums from the tree before a read
1614 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1615 int mirror_num, unsigned long bio_flags,
1618 struct btrfs_root *root = BTRFS_I(inode)->root;
1623 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1625 if (btrfs_is_free_space_inode(inode))
1628 if (!(rw & REQ_WRITE)) {
1629 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1633 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1634 return btrfs_submit_compressed_read(inode, bio,
1635 mirror_num, bio_flags);
1636 } else if (!skip_sum) {
1637 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1642 } else if (!skip_sum) {
1643 /* csum items have already been cloned */
1644 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1646 /* we're doing a write, do the async checksumming */
1647 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1648 inode, rw, bio, mirror_num,
1649 bio_flags, bio_offset,
1650 __btrfs_submit_bio_start,
1651 __btrfs_submit_bio_done);
1655 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1659 * given a list of ordered sums record them in the inode. This happens
1660 * at IO completion time based on sums calculated at bio submission time.
1662 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1663 struct inode *inode, u64 file_offset,
1664 struct list_head *list)
1666 struct btrfs_ordered_sum *sum;
1668 list_for_each_entry(sum, list, list) {
1669 btrfs_csum_file_blocks(trans,
1670 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1675 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1676 struct extent_state **cached_state)
1678 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1679 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1680 cached_state, GFP_NOFS);
1683 /* see btrfs_writepage_start_hook for details on why this is required */
1684 struct btrfs_writepage_fixup {
1686 struct btrfs_work work;
1689 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1691 struct btrfs_writepage_fixup *fixup;
1692 struct btrfs_ordered_extent *ordered;
1693 struct extent_state *cached_state = NULL;
1695 struct inode *inode;
1700 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1704 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1705 ClearPageChecked(page);
1709 inode = page->mapping->host;
1710 page_start = page_offset(page);
1711 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1713 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1716 /* already ordered? We're done */
1717 if (PagePrivate2(page))
1720 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1722 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1723 page_end, &cached_state, GFP_NOFS);
1725 btrfs_start_ordered_extent(inode, ordered, 1);
1726 btrfs_put_ordered_extent(ordered);
1730 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1732 mapping_set_error(page->mapping, ret);
1733 end_extent_writepage(page, ret, page_start, page_end);
1734 ClearPageChecked(page);
1738 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1739 ClearPageChecked(page);
1740 set_page_dirty(page);
1742 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1743 &cached_state, GFP_NOFS);
1746 page_cache_release(page);
1751 * There are a few paths in the higher layers of the kernel that directly
1752 * set the page dirty bit without asking the filesystem if it is a
1753 * good idea. This causes problems because we want to make sure COW
1754 * properly happens and the data=ordered rules are followed.
1756 * In our case any range that doesn't have the ORDERED bit set
1757 * hasn't been properly setup for IO. We kick off an async process
1758 * to fix it up. The async helper will wait for ordered extents, set
1759 * the delalloc bit and make it safe to write the page.
1761 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1763 struct inode *inode = page->mapping->host;
1764 struct btrfs_writepage_fixup *fixup;
1765 struct btrfs_root *root = BTRFS_I(inode)->root;
1767 /* this page is properly in the ordered list */
1768 if (TestClearPagePrivate2(page))
1771 if (PageChecked(page))
1774 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1778 SetPageChecked(page);
1779 page_cache_get(page);
1780 fixup->work.func = btrfs_writepage_fixup_worker;
1782 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1786 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1787 struct inode *inode, u64 file_pos,
1788 u64 disk_bytenr, u64 disk_num_bytes,
1789 u64 num_bytes, u64 ram_bytes,
1790 u8 compression, u8 encryption,
1791 u16 other_encoding, int extent_type)
1793 struct btrfs_root *root = BTRFS_I(inode)->root;
1794 struct btrfs_file_extent_item *fi;
1795 struct btrfs_path *path;
1796 struct extent_buffer *leaf;
1797 struct btrfs_key ins;
1800 path = btrfs_alloc_path();
1804 path->leave_spinning = 1;
1807 * we may be replacing one extent in the tree with another.
1808 * The new extent is pinned in the extent map, and we don't want
1809 * to drop it from the cache until it is completely in the btree.
1811 * So, tell btrfs_drop_extents to leave this extent in the cache.
1812 * the caller is expected to unpin it and allow it to be merged
1815 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1816 file_pos + num_bytes, 0);
1820 ins.objectid = btrfs_ino(inode);
1821 ins.offset = file_pos;
1822 ins.type = BTRFS_EXTENT_DATA_KEY;
1823 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1826 leaf = path->nodes[0];
1827 fi = btrfs_item_ptr(leaf, path->slots[0],
1828 struct btrfs_file_extent_item);
1829 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1830 btrfs_set_file_extent_type(leaf, fi, extent_type);
1831 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1832 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1833 btrfs_set_file_extent_offset(leaf, fi, 0);
1834 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1835 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1836 btrfs_set_file_extent_compression(leaf, fi, compression);
1837 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1838 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1840 btrfs_mark_buffer_dirty(leaf);
1841 btrfs_release_path(path);
1843 inode_add_bytes(inode, num_bytes);
1845 ins.objectid = disk_bytenr;
1846 ins.offset = disk_num_bytes;
1847 ins.type = BTRFS_EXTENT_ITEM_KEY;
1848 ret = btrfs_alloc_reserved_file_extent(trans, root,
1849 root->root_key.objectid,
1850 btrfs_ino(inode), file_pos, &ins);
1852 btrfs_free_path(path);
1858 * helper function for btrfs_finish_ordered_io, this
1859 * just reads in some of the csum leaves to prime them into ram
1860 * before we start the transaction. It limits the amount of btree
1861 * reads required while inside the transaction.
1863 /* as ordered data IO finishes, this gets called so we can finish
1864 * an ordered extent if the range of bytes in the file it covers are
1867 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
1869 struct inode *inode = ordered_extent->inode;
1870 struct btrfs_root *root = BTRFS_I(inode)->root;
1871 struct btrfs_trans_handle *trans = NULL;
1872 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1873 struct extent_state *cached_state = NULL;
1874 int compress_type = 0;
1878 nolock = btrfs_is_free_space_inode(inode);
1880 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1885 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1886 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1887 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1890 trans = btrfs_join_transaction_nolock(root);
1892 trans = btrfs_join_transaction(root);
1893 if (IS_ERR(trans)) {
1894 ret = PTR_ERR(trans);
1898 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1899 ret = btrfs_update_inode_fallback(trans, root, inode);
1900 if (ret) /* -ENOMEM or corruption */
1901 btrfs_abort_transaction(trans, root, ret);
1906 lock_extent_bits(io_tree, ordered_extent->file_offset,
1907 ordered_extent->file_offset + ordered_extent->len - 1,
1911 trans = btrfs_join_transaction_nolock(root);
1913 trans = btrfs_join_transaction(root);
1914 if (IS_ERR(trans)) {
1915 ret = PTR_ERR(trans);
1919 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1921 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1922 compress_type = ordered_extent->compress_type;
1923 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1924 BUG_ON(compress_type);
1925 ret = btrfs_mark_extent_written(trans, inode,
1926 ordered_extent->file_offset,
1927 ordered_extent->file_offset +
1928 ordered_extent->len);
1930 BUG_ON(root == root->fs_info->tree_root);
1931 ret = insert_reserved_file_extent(trans, inode,
1932 ordered_extent->file_offset,
1933 ordered_extent->start,
1934 ordered_extent->disk_len,
1935 ordered_extent->len,
1936 ordered_extent->len,
1937 compress_type, 0, 0,
1938 BTRFS_FILE_EXTENT_REG);
1940 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1941 ordered_extent->file_offset, ordered_extent->len,
1944 btrfs_abort_transaction(trans, root, ret);
1948 add_pending_csums(trans, inode, ordered_extent->file_offset,
1949 &ordered_extent->list);
1951 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1952 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1953 ret = btrfs_update_inode_fallback(trans, root, inode);
1954 if (ret) { /* -ENOMEM or corruption */
1955 btrfs_abort_transaction(trans, root, ret);
1959 btrfs_set_inode_last_trans(trans, inode);
1963 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1964 ordered_extent->file_offset +
1965 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1967 if (root != root->fs_info->tree_root)
1968 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1970 btrfs_end_transaction(trans, root);
1973 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
1974 ordered_extent->file_offset +
1975 ordered_extent->len - 1, NULL, GFP_NOFS);
1978 * This needs to be done to make sure anybody waiting knows we are done
1979 * updating everything for this ordered extent.
1981 btrfs_remove_ordered_extent(inode, ordered_extent);
1984 btrfs_put_ordered_extent(ordered_extent);
1985 /* once for the tree */
1986 btrfs_put_ordered_extent(ordered_extent);
1991 static void finish_ordered_fn(struct btrfs_work *work)
1993 struct btrfs_ordered_extent *ordered_extent;
1994 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
1995 btrfs_finish_ordered_io(ordered_extent);
1998 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1999 struct extent_state *state, int uptodate)
2001 struct inode *inode = page->mapping->host;
2002 struct btrfs_root *root = BTRFS_I(inode)->root;
2003 struct btrfs_ordered_extent *ordered_extent = NULL;
2004 struct btrfs_workers *workers;
2006 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2008 ClearPagePrivate2(page);
2009 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2010 end - start + 1, uptodate))
2013 ordered_extent->work.func = finish_ordered_fn;
2014 ordered_extent->work.flags = 0;
2016 if (btrfs_is_free_space_inode(inode))
2017 workers = &root->fs_info->endio_freespace_worker;
2019 workers = &root->fs_info->endio_write_workers;
2020 btrfs_queue_worker(workers, &ordered_extent->work);
2026 * when reads are done, we need to check csums to verify the data is correct
2027 * if there's a match, we allow the bio to finish. If not, the code in
2028 * extent_io.c will try to find good copies for us.
2030 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2031 struct extent_state *state, int mirror)
2033 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
2034 struct inode *inode = page->mapping->host;
2035 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2037 u64 private = ~(u32)0;
2039 struct btrfs_root *root = BTRFS_I(inode)->root;
2042 if (PageChecked(page)) {
2043 ClearPageChecked(page);
2047 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2050 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2051 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2052 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2057 if (state && state->start == start) {
2058 private = state->private;
2061 ret = get_state_private(io_tree, start, &private);
2063 kaddr = kmap_atomic(page);
2067 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
2068 btrfs_csum_final(csum, (char *)&csum);
2069 if (csum != private)
2072 kunmap_atomic(kaddr);
2077 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2079 (unsigned long long)btrfs_ino(page->mapping->host),
2080 (unsigned long long)start, csum,
2081 (unsigned long long)private);
2082 memset(kaddr + offset, 1, end - start + 1);
2083 flush_dcache_page(page);
2084 kunmap_atomic(kaddr);
2090 struct delayed_iput {
2091 struct list_head list;
2092 struct inode *inode;
2095 /* JDM: If this is fs-wide, why can't we add a pointer to
2096 * btrfs_inode instead and avoid the allocation? */
2097 void btrfs_add_delayed_iput(struct inode *inode)
2099 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2100 struct delayed_iput *delayed;
2102 if (atomic_add_unless(&inode->i_count, -1, 1))
2105 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2106 delayed->inode = inode;
2108 spin_lock(&fs_info->delayed_iput_lock);
2109 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2110 spin_unlock(&fs_info->delayed_iput_lock);
2113 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2116 struct btrfs_fs_info *fs_info = root->fs_info;
2117 struct delayed_iput *delayed;
2120 spin_lock(&fs_info->delayed_iput_lock);
2121 empty = list_empty(&fs_info->delayed_iputs);
2122 spin_unlock(&fs_info->delayed_iput_lock);
2126 spin_lock(&fs_info->delayed_iput_lock);
2127 list_splice_init(&fs_info->delayed_iputs, &list);
2128 spin_unlock(&fs_info->delayed_iput_lock);
2130 while (!list_empty(&list)) {
2131 delayed = list_entry(list.next, struct delayed_iput, list);
2132 list_del(&delayed->list);
2133 iput(delayed->inode);
2138 enum btrfs_orphan_cleanup_state {
2139 ORPHAN_CLEANUP_STARTED = 1,
2140 ORPHAN_CLEANUP_DONE = 2,
2144 * This is called in transaction commit time. If there are no orphan
2145 * files in the subvolume, it removes orphan item and frees block_rsv
2148 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2149 struct btrfs_root *root)
2151 struct btrfs_block_rsv *block_rsv;
2154 if (atomic_read(&root->orphan_inodes) ||
2155 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2158 spin_lock(&root->orphan_lock);
2159 if (atomic_read(&root->orphan_inodes)) {
2160 spin_unlock(&root->orphan_lock);
2164 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2165 spin_unlock(&root->orphan_lock);
2169 block_rsv = root->orphan_block_rsv;
2170 root->orphan_block_rsv = NULL;
2171 spin_unlock(&root->orphan_lock);
2173 if (root->orphan_item_inserted &&
2174 btrfs_root_refs(&root->root_item) > 0) {
2175 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2176 root->root_key.objectid);
2178 root->orphan_item_inserted = 0;
2182 WARN_ON(block_rsv->size > 0);
2183 btrfs_free_block_rsv(root, block_rsv);
2188 * This creates an orphan entry for the given inode in case something goes
2189 * wrong in the middle of an unlink/truncate.
2191 * NOTE: caller of this function should reserve 5 units of metadata for
2194 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2196 struct btrfs_root *root = BTRFS_I(inode)->root;
2197 struct btrfs_block_rsv *block_rsv = NULL;
2202 if (!root->orphan_block_rsv) {
2203 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2208 spin_lock(&root->orphan_lock);
2209 if (!root->orphan_block_rsv) {
2210 root->orphan_block_rsv = block_rsv;
2211 } else if (block_rsv) {
2212 btrfs_free_block_rsv(root, block_rsv);
2216 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2217 &BTRFS_I(inode)->runtime_flags)) {
2220 * For proper ENOSPC handling, we should do orphan
2221 * cleanup when mounting. But this introduces backward
2222 * compatibility issue.
2224 if (!xchg(&root->orphan_item_inserted, 1))
2230 atomic_inc(&root->orphan_inodes);
2233 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2234 &BTRFS_I(inode)->runtime_flags))
2236 spin_unlock(&root->orphan_lock);
2238 /* grab metadata reservation from transaction handle */
2240 ret = btrfs_orphan_reserve_metadata(trans, inode);
2241 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2244 /* insert an orphan item to track this unlinked/truncated file */
2246 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2247 if (ret && ret != -EEXIST) {
2248 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2249 &BTRFS_I(inode)->runtime_flags);
2250 btrfs_abort_transaction(trans, root, ret);
2256 /* insert an orphan item to track subvolume contains orphan files */
2258 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2259 root->root_key.objectid);
2260 if (ret && ret != -EEXIST) {
2261 btrfs_abort_transaction(trans, root, ret);
2269 * We have done the truncate/delete so we can go ahead and remove the orphan
2270 * item for this particular inode.
2272 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2274 struct btrfs_root *root = BTRFS_I(inode)->root;
2275 int delete_item = 0;
2276 int release_rsv = 0;
2279 spin_lock(&root->orphan_lock);
2280 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2281 &BTRFS_I(inode)->runtime_flags))
2284 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2285 &BTRFS_I(inode)->runtime_flags))
2287 spin_unlock(&root->orphan_lock);
2289 if (trans && delete_item) {
2290 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2291 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2295 btrfs_orphan_release_metadata(inode);
2296 atomic_dec(&root->orphan_inodes);
2303 * this cleans up any orphans that may be left on the list from the last use
2306 int btrfs_orphan_cleanup(struct btrfs_root *root)
2308 struct btrfs_path *path;
2309 struct extent_buffer *leaf;
2310 struct btrfs_key key, found_key;
2311 struct btrfs_trans_handle *trans;
2312 struct inode *inode;
2313 u64 last_objectid = 0;
2314 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2316 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2319 path = btrfs_alloc_path();
2326 key.objectid = BTRFS_ORPHAN_OBJECTID;
2327 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2328 key.offset = (u64)-1;
2331 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2336 * if ret == 0 means we found what we were searching for, which
2337 * is weird, but possible, so only screw with path if we didn't
2338 * find the key and see if we have stuff that matches
2342 if (path->slots[0] == 0)
2347 /* pull out the item */
2348 leaf = path->nodes[0];
2349 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2351 /* make sure the item matches what we want */
2352 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2354 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2357 /* release the path since we're done with it */
2358 btrfs_release_path(path);
2361 * this is where we are basically btrfs_lookup, without the
2362 * crossing root thing. we store the inode number in the
2363 * offset of the orphan item.
2366 if (found_key.offset == last_objectid) {
2367 printk(KERN_ERR "btrfs: Error removing orphan entry, "
2368 "stopping orphan cleanup\n");
2373 last_objectid = found_key.offset;
2375 found_key.objectid = found_key.offset;
2376 found_key.type = BTRFS_INODE_ITEM_KEY;
2377 found_key.offset = 0;
2378 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2379 ret = PTR_RET(inode);
2380 if (ret && ret != -ESTALE)
2383 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2384 struct btrfs_root *dead_root;
2385 struct btrfs_fs_info *fs_info = root->fs_info;
2386 int is_dead_root = 0;
2389 * this is an orphan in the tree root. Currently these
2390 * could come from 2 sources:
2391 * a) a snapshot deletion in progress
2392 * b) a free space cache inode
2393 * We need to distinguish those two, as the snapshot
2394 * orphan must not get deleted.
2395 * find_dead_roots already ran before us, so if this
2396 * is a snapshot deletion, we should find the root
2397 * in the dead_roots list
2399 spin_lock(&fs_info->trans_lock);
2400 list_for_each_entry(dead_root, &fs_info->dead_roots,
2402 if (dead_root->root_key.objectid ==
2403 found_key.objectid) {
2408 spin_unlock(&fs_info->trans_lock);
2410 /* prevent this orphan from being found again */
2411 key.offset = found_key.objectid - 1;
2416 * Inode is already gone but the orphan item is still there,
2417 * kill the orphan item.
2419 if (ret == -ESTALE) {
2420 trans = btrfs_start_transaction(root, 1);
2421 if (IS_ERR(trans)) {
2422 ret = PTR_ERR(trans);
2425 printk(KERN_ERR "auto deleting %Lu\n",
2426 found_key.objectid);
2427 ret = btrfs_del_orphan_item(trans, root,
2428 found_key.objectid);
2429 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2430 btrfs_end_transaction(trans, root);
2435 * add this inode to the orphan list so btrfs_orphan_del does
2436 * the proper thing when we hit it
2438 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2439 &BTRFS_I(inode)->runtime_flags);
2441 /* if we have links, this was a truncate, lets do that */
2442 if (inode->i_nlink) {
2443 if (!S_ISREG(inode->i_mode)) {
2449 ret = btrfs_truncate(inode);
2454 /* this will do delete_inode and everything for us */
2459 /* release the path since we're done with it */
2460 btrfs_release_path(path);
2462 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2464 if (root->orphan_block_rsv)
2465 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2468 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2469 trans = btrfs_join_transaction(root);
2471 btrfs_end_transaction(trans, root);
2475 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2477 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2481 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2482 btrfs_free_path(path);
2487 * very simple check to peek ahead in the leaf looking for xattrs. If we
2488 * don't find any xattrs, we know there can't be any acls.
2490 * slot is the slot the inode is in, objectid is the objectid of the inode
2492 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2493 int slot, u64 objectid)
2495 u32 nritems = btrfs_header_nritems(leaf);
2496 struct btrfs_key found_key;
2500 while (slot < nritems) {
2501 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2503 /* we found a different objectid, there must not be acls */
2504 if (found_key.objectid != objectid)
2507 /* we found an xattr, assume we've got an acl */
2508 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2512 * we found a key greater than an xattr key, there can't
2513 * be any acls later on
2515 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2522 * it goes inode, inode backrefs, xattrs, extents,
2523 * so if there are a ton of hard links to an inode there can
2524 * be a lot of backrefs. Don't waste time searching too hard,
2525 * this is just an optimization
2530 /* we hit the end of the leaf before we found an xattr or
2531 * something larger than an xattr. We have to assume the inode
2538 * read an inode from the btree into the in-memory inode
2540 static void btrfs_read_locked_inode(struct inode *inode)
2542 struct btrfs_path *path;
2543 struct extent_buffer *leaf;
2544 struct btrfs_inode_item *inode_item;
2545 struct btrfs_timespec *tspec;
2546 struct btrfs_root *root = BTRFS_I(inode)->root;
2547 struct btrfs_key location;
2551 bool filled = false;
2553 ret = btrfs_fill_inode(inode, &rdev);
2557 path = btrfs_alloc_path();
2561 path->leave_spinning = 1;
2562 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2564 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2568 leaf = path->nodes[0];
2573 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2574 struct btrfs_inode_item);
2575 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2576 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2577 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
2578 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
2579 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2581 tspec = btrfs_inode_atime(inode_item);
2582 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2583 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2585 tspec = btrfs_inode_mtime(inode_item);
2586 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2587 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2589 tspec = btrfs_inode_ctime(inode_item);
2590 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2591 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2593 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2594 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2595 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
2598 * If we were modified in the current generation and evicted from memory
2599 * and then re-read we need to do a full sync since we don't have any
2600 * idea about which extents were modified before we were evicted from
2603 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
2604 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2605 &BTRFS_I(inode)->runtime_flags);
2607 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2608 inode->i_generation = BTRFS_I(inode)->generation;
2610 rdev = btrfs_inode_rdev(leaf, inode_item);
2612 BTRFS_I(inode)->index_cnt = (u64)-1;
2613 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2616 * try to precache a NULL acl entry for files that don't have
2617 * any xattrs or acls
2619 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2622 cache_no_acl(inode);
2624 btrfs_free_path(path);
2626 switch (inode->i_mode & S_IFMT) {
2628 inode->i_mapping->a_ops = &btrfs_aops;
2629 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2630 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2631 inode->i_fop = &btrfs_file_operations;
2632 inode->i_op = &btrfs_file_inode_operations;
2635 inode->i_fop = &btrfs_dir_file_operations;
2636 if (root == root->fs_info->tree_root)
2637 inode->i_op = &btrfs_dir_ro_inode_operations;
2639 inode->i_op = &btrfs_dir_inode_operations;
2642 inode->i_op = &btrfs_symlink_inode_operations;
2643 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2644 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2647 inode->i_op = &btrfs_special_inode_operations;
2648 init_special_inode(inode, inode->i_mode, rdev);
2652 btrfs_update_iflags(inode);
2656 btrfs_free_path(path);
2657 make_bad_inode(inode);
2661 * given a leaf and an inode, copy the inode fields into the leaf
2663 static void fill_inode_item(struct btrfs_trans_handle *trans,
2664 struct extent_buffer *leaf,
2665 struct btrfs_inode_item *item,
2666 struct inode *inode)
2668 btrfs_set_inode_uid(leaf, item, i_uid_read(inode));
2669 btrfs_set_inode_gid(leaf, item, i_gid_read(inode));
2670 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2671 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2672 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2674 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2675 inode->i_atime.tv_sec);
2676 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2677 inode->i_atime.tv_nsec);
2679 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2680 inode->i_mtime.tv_sec);
2681 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2682 inode->i_mtime.tv_nsec);
2684 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2685 inode->i_ctime.tv_sec);
2686 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2687 inode->i_ctime.tv_nsec);
2689 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2690 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2691 btrfs_set_inode_sequence(leaf, item, inode->i_version);
2692 btrfs_set_inode_transid(leaf, item, trans->transid);
2693 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2694 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2695 btrfs_set_inode_block_group(leaf, item, 0);
2699 * copy everything in the in-memory inode into the btree.
2701 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2702 struct btrfs_root *root, struct inode *inode)
2704 struct btrfs_inode_item *inode_item;
2705 struct btrfs_path *path;
2706 struct extent_buffer *leaf;
2709 path = btrfs_alloc_path();
2713 path->leave_spinning = 1;
2714 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2722 btrfs_unlock_up_safe(path, 1);
2723 leaf = path->nodes[0];
2724 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2725 struct btrfs_inode_item);
2727 fill_inode_item(trans, leaf, inode_item, inode);
2728 btrfs_mark_buffer_dirty(leaf);
2729 btrfs_set_inode_last_trans(trans, inode);
2732 btrfs_free_path(path);
2737 * copy everything in the in-memory inode into the btree.
2739 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2740 struct btrfs_root *root, struct inode *inode)
2745 * If the inode is a free space inode, we can deadlock during commit
2746 * if we put it into the delayed code.
2748 * The data relocation inode should also be directly updated
2751 if (!btrfs_is_free_space_inode(inode)
2752 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2753 btrfs_update_root_times(trans, root);
2755 ret = btrfs_delayed_update_inode(trans, root, inode);
2757 btrfs_set_inode_last_trans(trans, inode);
2761 return btrfs_update_inode_item(trans, root, inode);
2764 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2765 struct btrfs_root *root,
2766 struct inode *inode)
2770 ret = btrfs_update_inode(trans, root, inode);
2772 return btrfs_update_inode_item(trans, root, inode);
2777 * unlink helper that gets used here in inode.c and in the tree logging
2778 * recovery code. It remove a link in a directory with a given name, and
2779 * also drops the back refs in the inode to the directory
2781 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2782 struct btrfs_root *root,
2783 struct inode *dir, struct inode *inode,
2784 const char *name, int name_len)
2786 struct btrfs_path *path;
2788 struct extent_buffer *leaf;
2789 struct btrfs_dir_item *di;
2790 struct btrfs_key key;
2792 u64 ino = btrfs_ino(inode);
2793 u64 dir_ino = btrfs_ino(dir);
2795 path = btrfs_alloc_path();
2801 path->leave_spinning = 1;
2802 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2803 name, name_len, -1);
2812 leaf = path->nodes[0];
2813 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2814 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2817 btrfs_release_path(path);
2819 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2822 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2823 "inode %llu parent %llu\n", name_len, name,
2824 (unsigned long long)ino, (unsigned long long)dir_ino);
2825 btrfs_abort_transaction(trans, root, ret);
2829 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2831 btrfs_abort_transaction(trans, root, ret);
2835 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2837 if (ret != 0 && ret != -ENOENT) {
2838 btrfs_abort_transaction(trans, root, ret);
2842 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2847 btrfs_free_path(path);
2851 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2852 inode_inc_iversion(inode);
2853 inode_inc_iversion(dir);
2854 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2855 ret = btrfs_update_inode(trans, root, dir);
2860 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2861 struct btrfs_root *root,
2862 struct inode *dir, struct inode *inode,
2863 const char *name, int name_len)
2866 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2868 btrfs_drop_nlink(inode);
2869 ret = btrfs_update_inode(trans, root, inode);
2875 /* helper to check if there is any shared block in the path */
2876 static int check_path_shared(struct btrfs_root *root,
2877 struct btrfs_path *path)
2879 struct extent_buffer *eb;
2883 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2886 if (!path->nodes[level])
2888 eb = path->nodes[level];
2889 if (!btrfs_block_can_be_shared(root, eb))
2891 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2900 * helper to start transaction for unlink and rmdir.
2902 * unlink and rmdir are special in btrfs, they do not always free space.
2903 * so in enospc case, we should make sure they will free space before
2904 * allowing them to use the global metadata reservation.
2906 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2907 struct dentry *dentry)
2909 struct btrfs_trans_handle *trans;
2910 struct btrfs_root *root = BTRFS_I(dir)->root;
2911 struct btrfs_path *path;
2912 struct btrfs_dir_item *di;
2913 struct inode *inode = dentry->d_inode;
2918 u64 ino = btrfs_ino(inode);
2919 u64 dir_ino = btrfs_ino(dir);
2922 * 1 for the possible orphan item
2923 * 1 for the dir item
2924 * 1 for the dir index
2925 * 1 for the inode ref
2926 * 1 for the inode ref in the tree log
2927 * 2 for the dir entries in the log
2930 trans = btrfs_start_transaction(root, 8);
2931 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2934 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2935 return ERR_PTR(-ENOSPC);
2937 /* check if there is someone else holds reference */
2938 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2939 return ERR_PTR(-ENOSPC);
2941 if (atomic_read(&inode->i_count) > 2)
2942 return ERR_PTR(-ENOSPC);
2944 if (xchg(&root->fs_info->enospc_unlink, 1))
2945 return ERR_PTR(-ENOSPC);
2947 path = btrfs_alloc_path();
2949 root->fs_info->enospc_unlink = 0;
2950 return ERR_PTR(-ENOMEM);
2953 /* 1 for the orphan item */
2954 trans = btrfs_start_transaction(root, 1);
2955 if (IS_ERR(trans)) {
2956 btrfs_free_path(path);
2957 root->fs_info->enospc_unlink = 0;
2961 path->skip_locking = 1;
2962 path->search_commit_root = 1;
2964 ret = btrfs_lookup_inode(trans, root, path,
2965 &BTRFS_I(dir)->location, 0);
2971 if (check_path_shared(root, path))
2976 btrfs_release_path(path);
2978 ret = btrfs_lookup_inode(trans, root, path,
2979 &BTRFS_I(inode)->location, 0);
2985 if (check_path_shared(root, path))
2990 btrfs_release_path(path);
2992 if (ret == 0 && S_ISREG(inode->i_mode)) {
2993 ret = btrfs_lookup_file_extent(trans, root, path,
2999 BUG_ON(ret == 0); /* Corruption */
3000 if (check_path_shared(root, path))
3002 btrfs_release_path(path);
3010 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3011 dentry->d_name.name, dentry->d_name.len, 0);
3017 if (check_path_shared(root, path))
3023 btrfs_release_path(path);
3025 ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3026 dentry->d_name.len, ino, dir_ino, 0,
3033 if (check_path_shared(root, path))
3036 btrfs_release_path(path);
3039 * This is a commit root search, if we can lookup inode item and other
3040 * relative items in the commit root, it means the transaction of
3041 * dir/file creation has been committed, and the dir index item that we
3042 * delay to insert has also been inserted into the commit root. So
3043 * we needn't worry about the delayed insertion of the dir index item
3046 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3047 dentry->d_name.name, dentry->d_name.len, 0);
3052 BUG_ON(ret == -ENOENT);
3053 if (check_path_shared(root, path))
3058 btrfs_free_path(path);
3059 /* Migrate the orphan reservation over */
3061 err = btrfs_block_rsv_migrate(trans->block_rsv,
3062 &root->fs_info->global_block_rsv,
3063 trans->bytes_reserved);
3066 btrfs_end_transaction(trans, root);
3067 root->fs_info->enospc_unlink = 0;
3068 return ERR_PTR(err);
3071 trans->block_rsv = &root->fs_info->global_block_rsv;
3075 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3076 struct btrfs_root *root)
3078 if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3079 btrfs_block_rsv_release(root, trans->block_rsv,
3080 trans->bytes_reserved);
3081 trans->block_rsv = &root->fs_info->trans_block_rsv;
3082 BUG_ON(!root->fs_info->enospc_unlink);
3083 root->fs_info->enospc_unlink = 0;
3085 btrfs_end_transaction(trans, root);
3088 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3090 struct btrfs_root *root = BTRFS_I(dir)->root;
3091 struct btrfs_trans_handle *trans;
3092 struct inode *inode = dentry->d_inode;
3095 trans = __unlink_start_trans(dir, dentry);
3097 return PTR_ERR(trans);
3099 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3101 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3102 dentry->d_name.name, dentry->d_name.len);
3106 if (inode->i_nlink == 0) {
3107 ret = btrfs_orphan_add(trans, inode);
3113 __unlink_end_trans(trans, root);
3114 btrfs_btree_balance_dirty(root);
3118 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3119 struct btrfs_root *root,
3120 struct inode *dir, u64 objectid,
3121 const char *name, int name_len)
3123 struct btrfs_path *path;
3124 struct extent_buffer *leaf;
3125 struct btrfs_dir_item *di;
3126 struct btrfs_key key;
3129 u64 dir_ino = btrfs_ino(dir);
3131 path = btrfs_alloc_path();
3135 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3136 name, name_len, -1);
3137 if (IS_ERR_OR_NULL(di)) {
3145 leaf = path->nodes[0];
3146 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3147 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3148 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3150 btrfs_abort_transaction(trans, root, ret);
3153 btrfs_release_path(path);
3155 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3156 objectid, root->root_key.objectid,
3157 dir_ino, &index, name, name_len);
3159 if (ret != -ENOENT) {
3160 btrfs_abort_transaction(trans, root, ret);
3163 di = btrfs_search_dir_index_item(root, path, dir_ino,
3165 if (IS_ERR_OR_NULL(di)) {
3170 btrfs_abort_transaction(trans, root, ret);
3174 leaf = path->nodes[0];
3175 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3176 btrfs_release_path(path);
3179 btrfs_release_path(path);
3181 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3183 btrfs_abort_transaction(trans, root, ret);
3187 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3188 inode_inc_iversion(dir);
3189 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3190 ret = btrfs_update_inode_fallback(trans, root, dir);
3192 btrfs_abort_transaction(trans, root, ret);
3194 btrfs_free_path(path);
3198 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3200 struct inode *inode = dentry->d_inode;
3202 struct btrfs_root *root = BTRFS_I(dir)->root;
3203 struct btrfs_trans_handle *trans;
3205 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3207 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3210 trans = __unlink_start_trans(dir, dentry);
3212 return PTR_ERR(trans);
3214 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3215 err = btrfs_unlink_subvol(trans, root, dir,
3216 BTRFS_I(inode)->location.objectid,
3217 dentry->d_name.name,
3218 dentry->d_name.len);
3222 err = btrfs_orphan_add(trans, inode);
3226 /* now the directory is empty */
3227 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3228 dentry->d_name.name, dentry->d_name.len);
3230 btrfs_i_size_write(inode, 0);
3232 __unlink_end_trans(trans, root);
3233 btrfs_btree_balance_dirty(root);
3239 * this can truncate away extent items, csum items and directory items.
3240 * It starts at a high offset and removes keys until it can't find
3241 * any higher than new_size
3243 * csum items that cross the new i_size are truncated to the new size
3246 * min_type is the minimum key type to truncate down to. If set to 0, this
3247 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3249 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3250 struct btrfs_root *root,
3251 struct inode *inode,
3252 u64 new_size, u32 min_type)
3254 struct btrfs_path *path;
3255 struct extent_buffer *leaf;
3256 struct btrfs_file_extent_item *fi;
3257 struct btrfs_key key;
3258 struct btrfs_key found_key;
3259 u64 extent_start = 0;
3260 u64 extent_num_bytes = 0;
3261 u64 extent_offset = 0;
3263 u64 mask = root->sectorsize - 1;
3264 u32 found_type = (u8)-1;
3267 int pending_del_nr = 0;
3268 int pending_del_slot = 0;
3269 int extent_type = -1;
3272 u64 ino = btrfs_ino(inode);
3274 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3276 path = btrfs_alloc_path();
3282 * We want to drop from the next block forward in case this new size is
3283 * not block aligned since we will be keeping the last block of the
3284 * extent just the way it is.
3286 if (root->ref_cows || root == root->fs_info->tree_root)
3287 btrfs_drop_extent_cache(inode, (new_size + mask) & (~mask), (u64)-1, 0);
3290 * This function is also used to drop the items in the log tree before
3291 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3292 * it is used to drop the loged items. So we shouldn't kill the delayed
3295 if (min_type == 0 && root == BTRFS_I(inode)->root)
3296 btrfs_kill_delayed_inode_items(inode);
3299 key.offset = (u64)-1;
3303 path->leave_spinning = 1;
3304 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3311 /* there are no items in the tree for us to truncate, we're
3314 if (path->slots[0] == 0)
3321 leaf = path->nodes[0];
3322 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3323 found_type = btrfs_key_type(&found_key);
3325 if (found_key.objectid != ino)
3328 if (found_type < min_type)
3331 item_end = found_key.offset;
3332 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3333 fi = btrfs_item_ptr(leaf, path->slots[0],
3334 struct btrfs_file_extent_item);
3335 extent_type = btrfs_file_extent_type(leaf, fi);
3336 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3338 btrfs_file_extent_num_bytes(leaf, fi);
3339 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3340 item_end += btrfs_file_extent_inline_len(leaf,
3345 if (found_type > min_type) {
3348 if (item_end < new_size)
3350 if (found_key.offset >= new_size)
3356 /* FIXME, shrink the extent if the ref count is only 1 */
3357 if (found_type != BTRFS_EXTENT_DATA_KEY)
3360 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3362 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3364 u64 orig_num_bytes =
3365 btrfs_file_extent_num_bytes(leaf, fi);
3366 extent_num_bytes = new_size -
3367 found_key.offset + root->sectorsize - 1;
3368 extent_num_bytes = extent_num_bytes &
3369 ~((u64)root->sectorsize - 1);
3370 btrfs_set_file_extent_num_bytes(leaf, fi,
3372 num_dec = (orig_num_bytes -
3374 if (root->ref_cows && extent_start != 0)
3375 inode_sub_bytes(inode, num_dec);
3376 btrfs_mark_buffer_dirty(leaf);
3379 btrfs_file_extent_disk_num_bytes(leaf,
3381 extent_offset = found_key.offset -
3382 btrfs_file_extent_offset(leaf, fi);
3384 /* FIXME blocksize != 4096 */
3385 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3386 if (extent_start != 0) {
3389 inode_sub_bytes(inode, num_dec);
3392 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3394 * we can't truncate inline items that have had
3398 btrfs_file_extent_compression(leaf, fi) == 0 &&
3399 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3400 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3401 u32 size = new_size - found_key.offset;
3403 if (root->ref_cows) {
3404 inode_sub_bytes(inode, item_end + 1 -
3408 btrfs_file_extent_calc_inline_size(size);
3409 btrfs_truncate_item(trans, root, path,
3411 } else if (root->ref_cows) {
3412 inode_sub_bytes(inode, item_end + 1 -
3418 if (!pending_del_nr) {
3419 /* no pending yet, add ourselves */
3420 pending_del_slot = path->slots[0];
3422 } else if (pending_del_nr &&
3423 path->slots[0] + 1 == pending_del_slot) {
3424 /* hop on the pending chunk */
3426 pending_del_slot = path->slots[0];
3433 if (found_extent && (root->ref_cows ||
3434 root == root->fs_info->tree_root)) {
3435 btrfs_set_path_blocking(path);
3436 ret = btrfs_free_extent(trans, root, extent_start,
3437 extent_num_bytes, 0,
3438 btrfs_header_owner(leaf),
3439 ino, extent_offset, 0);
3443 if (found_type == BTRFS_INODE_ITEM_KEY)
3446 if (path->slots[0] == 0 ||
3447 path->slots[0] != pending_del_slot) {
3448 if (pending_del_nr) {
3449 ret = btrfs_del_items(trans, root, path,
3453 btrfs_abort_transaction(trans,
3459 btrfs_release_path(path);
3466 if (pending_del_nr) {
3467 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3470 btrfs_abort_transaction(trans, root, ret);
3473 btrfs_free_path(path);
3478 * btrfs_truncate_page - read, zero a chunk and write a page
3479 * @inode - inode that we're zeroing
3480 * @from - the offset to start zeroing
3481 * @len - the length to zero, 0 to zero the entire range respective to the
3483 * @front - zero up to the offset instead of from the offset on
3485 * This will find the page for the "from" offset and cow the page and zero the
3486 * part we want to zero. This is used with truncate and hole punching.
3488 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
3491 struct address_space *mapping = inode->i_mapping;
3492 struct btrfs_root *root = BTRFS_I(inode)->root;
3493 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3494 struct btrfs_ordered_extent *ordered;
3495 struct extent_state *cached_state = NULL;
3497 u32 blocksize = root->sectorsize;
3498 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3499 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3501 gfp_t mask = btrfs_alloc_write_mask(mapping);
3506 if ((offset & (blocksize - 1)) == 0 &&
3507 (!len || ((len & (blocksize - 1)) == 0)))
3509 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3515 page = find_or_create_page(mapping, index, mask);
3517 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3521 page_start = page_offset(page);
3522 page_end = page_start + PAGE_CACHE_SIZE - 1;
3524 if (!PageUptodate(page)) {
3525 ret = btrfs_readpage(NULL, page);
3527 if (page->mapping != mapping) {
3529 page_cache_release(page);
3532 if (!PageUptodate(page)) {
3537 wait_on_page_writeback(page);
3539 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3540 set_page_extent_mapped(page);
3542 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3544 unlock_extent_cached(io_tree, page_start, page_end,
3545 &cached_state, GFP_NOFS);
3547 page_cache_release(page);
3548 btrfs_start_ordered_extent(inode, ordered, 1);
3549 btrfs_put_ordered_extent(ordered);
3553 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3554 EXTENT_DIRTY | EXTENT_DELALLOC |
3555 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
3556 0, 0, &cached_state, GFP_NOFS);
3558 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3561 unlock_extent_cached(io_tree, page_start, page_end,
3562 &cached_state, GFP_NOFS);
3567 if (offset != PAGE_CACHE_SIZE) {
3569 len = PAGE_CACHE_SIZE - offset;
3572 memset(kaddr, 0, offset);
3574 memset(kaddr + offset, 0, len);
3575 flush_dcache_page(page);
3578 ClearPageChecked(page);
3579 set_page_dirty(page);
3580 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3585 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3587 page_cache_release(page);
3593 * This function puts in dummy file extents for the area we're creating a hole
3594 * for. So if we are truncating this file to a larger size we need to insert
3595 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3596 * the range between oldsize and size
3598 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3600 struct btrfs_trans_handle *trans;
3601 struct btrfs_root *root = BTRFS_I(inode)->root;
3602 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3603 struct extent_map *em = NULL;
3604 struct extent_state *cached_state = NULL;
3605 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3606 u64 mask = root->sectorsize - 1;
3607 u64 hole_start = (oldsize + mask) & ~mask;
3608 u64 block_end = (size + mask) & ~mask;
3614 if (size <= hole_start)
3618 struct btrfs_ordered_extent *ordered;
3619 btrfs_wait_ordered_range(inode, hole_start,
3620 block_end - hole_start);
3621 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3623 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3626 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3627 &cached_state, GFP_NOFS);
3628 btrfs_put_ordered_extent(ordered);
3631 cur_offset = hole_start;
3633 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3634 block_end - cur_offset, 0);
3639 last_byte = min(extent_map_end(em), block_end);
3640 last_byte = (last_byte + mask) & ~mask;
3641 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3642 struct extent_map *hole_em;
3643 hole_size = last_byte - cur_offset;
3645 trans = btrfs_start_transaction(root, 3);
3646 if (IS_ERR(trans)) {
3647 err = PTR_ERR(trans);
3651 err = btrfs_drop_extents(trans, root, inode,
3653 cur_offset + hole_size, 1);
3655 btrfs_abort_transaction(trans, root, err);
3656 btrfs_end_transaction(trans, root);
3660 err = btrfs_insert_file_extent(trans, root,
3661 btrfs_ino(inode), cur_offset, 0,
3662 0, hole_size, 0, hole_size,
3665 btrfs_abort_transaction(trans, root, err);
3666 btrfs_end_transaction(trans, root);
3670 btrfs_drop_extent_cache(inode, cur_offset,
3671 cur_offset + hole_size - 1, 0);
3672 hole_em = alloc_extent_map();
3674 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3675 &BTRFS_I(inode)->runtime_flags);
3678 hole_em->start = cur_offset;
3679 hole_em->len = hole_size;
3680 hole_em->orig_start = cur_offset;
3682 hole_em->block_start = EXTENT_MAP_HOLE;
3683 hole_em->block_len = 0;
3684 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
3685 hole_em->compress_type = BTRFS_COMPRESS_NONE;
3686 hole_em->generation = trans->transid;
3689 write_lock(&em_tree->lock);
3690 err = add_extent_mapping(em_tree, hole_em);
3692 list_move(&hole_em->list,
3693 &em_tree->modified_extents);
3694 write_unlock(&em_tree->lock);
3697 btrfs_drop_extent_cache(inode, cur_offset,
3701 free_extent_map(hole_em);
3703 btrfs_update_inode(trans, root, inode);
3704 btrfs_end_transaction(trans, root);
3706 free_extent_map(em);
3708 cur_offset = last_byte;
3709 if (cur_offset >= block_end)
3713 free_extent_map(em);
3714 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3719 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3721 struct btrfs_root *root = BTRFS_I(inode)->root;
3722 struct btrfs_trans_handle *trans;
3723 loff_t oldsize = i_size_read(inode);
3726 if (newsize == oldsize)
3729 if (newsize > oldsize) {
3730 truncate_pagecache(inode, oldsize, newsize);
3731 ret = btrfs_cont_expand(inode, oldsize, newsize);
3735 trans = btrfs_start_transaction(root, 1);
3737 return PTR_ERR(trans);
3739 i_size_write(inode, newsize);
3740 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3741 ret = btrfs_update_inode(trans, root, inode);
3742 btrfs_end_transaction(trans, root);
3746 * We're truncating a file that used to have good data down to
3747 * zero. Make sure it gets into the ordered flush list so that
3748 * any new writes get down to disk quickly.
3751 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3752 &BTRFS_I(inode)->runtime_flags);
3754 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3755 truncate_setsize(inode, newsize);
3756 ret = btrfs_truncate(inode);
3762 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3764 struct inode *inode = dentry->d_inode;
3765 struct btrfs_root *root = BTRFS_I(inode)->root;
3768 if (btrfs_root_readonly(root))
3771 err = inode_change_ok(inode, attr);
3775 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3776 err = btrfs_setsize(inode, attr->ia_size);
3781 if (attr->ia_valid) {
3782 setattr_copy(inode, attr);
3783 inode_inc_iversion(inode);
3784 err = btrfs_dirty_inode(inode);
3786 if (!err && attr->ia_valid & ATTR_MODE)
3787 err = btrfs_acl_chmod(inode);
3793 void btrfs_evict_inode(struct inode *inode)
3795 struct btrfs_trans_handle *trans;
3796 struct btrfs_root *root = BTRFS_I(inode)->root;
3797 struct btrfs_block_rsv *rsv, *global_rsv;
3798 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3801 trace_btrfs_inode_evict(inode);
3803 truncate_inode_pages(&inode->i_data, 0);
3804 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3805 btrfs_is_free_space_inode(inode)))
3808 if (is_bad_inode(inode)) {
3809 btrfs_orphan_del(NULL, inode);
3812 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3813 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3815 if (root->fs_info->log_root_recovering) {
3816 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3817 &BTRFS_I(inode)->runtime_flags));
3821 if (inode->i_nlink > 0) {
3822 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3826 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3828 btrfs_orphan_del(NULL, inode);
3831 rsv->size = min_size;
3833 global_rsv = &root->fs_info->global_block_rsv;
3835 btrfs_i_size_write(inode, 0);
3838 * This is a bit simpler than btrfs_truncate since we've already
3839 * reserved our space for our orphan item in the unlink, so we just
3840 * need to reserve some slack space in case we add bytes and update
3841 * inode item when doing the truncate.
3844 ret = btrfs_block_rsv_refill(root, rsv, min_size,
3845 BTRFS_RESERVE_FLUSH_LIMIT);
3848 * Try and steal from the global reserve since we will
3849 * likely not use this space anyway, we want to try as
3850 * hard as possible to get this to work.
3853 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3856 printk(KERN_WARNING "Could not get space for a "
3857 "delete, will truncate on mount %d\n", ret);
3858 btrfs_orphan_del(NULL, inode);
3859 btrfs_free_block_rsv(root, rsv);
3863 trans = btrfs_start_transaction_lflush(root, 1);
3864 if (IS_ERR(trans)) {
3865 btrfs_orphan_del(NULL, inode);
3866 btrfs_free_block_rsv(root, rsv);
3870 trans->block_rsv = rsv;
3872 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3876 trans->block_rsv = &root->fs_info->trans_block_rsv;
3877 ret = btrfs_update_inode(trans, root, inode);
3880 btrfs_end_transaction(trans, root);
3882 btrfs_btree_balance_dirty(root);
3885 btrfs_free_block_rsv(root, rsv);
3888 trans->block_rsv = root->orphan_block_rsv;
3889 ret = btrfs_orphan_del(trans, inode);
3893 trans->block_rsv = &root->fs_info->trans_block_rsv;
3894 if (!(root == root->fs_info->tree_root ||
3895 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3896 btrfs_return_ino(root, btrfs_ino(inode));
3898 btrfs_end_transaction(trans, root);
3899 btrfs_btree_balance_dirty(root);
3906 * this returns the key found in the dir entry in the location pointer.
3907 * If no dir entries were found, location->objectid is 0.
3909 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3910 struct btrfs_key *location)
3912 const char *name = dentry->d_name.name;
3913 int namelen = dentry->d_name.len;
3914 struct btrfs_dir_item *di;
3915 struct btrfs_path *path;
3916 struct btrfs_root *root = BTRFS_I(dir)->root;
3919 path = btrfs_alloc_path();
3923 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3928 if (IS_ERR_OR_NULL(di))
3931 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3933 btrfs_free_path(path);
3936 location->objectid = 0;
3941 * when we hit a tree root in a directory, the btrfs part of the inode
3942 * needs to be changed to reflect the root directory of the tree root. This
3943 * is kind of like crossing a mount point.
3945 static int fixup_tree_root_location(struct btrfs_root *root,
3947 struct dentry *dentry,
3948 struct btrfs_key *location,
3949 struct btrfs_root **sub_root)
3951 struct btrfs_path *path;
3952 struct btrfs_root *new_root;
3953 struct btrfs_root_ref *ref;
3954 struct extent_buffer *leaf;
3958 path = btrfs_alloc_path();
3965 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3966 BTRFS_I(dir)->root->root_key.objectid,
3967 location->objectid);
3974 leaf = path->nodes[0];
3975 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3976 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3977 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3980 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3981 (unsigned long)(ref + 1),
3982 dentry->d_name.len);
3986 btrfs_release_path(path);
3988 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3989 if (IS_ERR(new_root)) {
3990 err = PTR_ERR(new_root);
3994 if (btrfs_root_refs(&new_root->root_item) == 0) {
3999 *sub_root = new_root;
4000 location->objectid = btrfs_root_dirid(&new_root->root_item);
4001 location->type = BTRFS_INODE_ITEM_KEY;
4002 location->offset = 0;
4005 btrfs_free_path(path);
4009 static void inode_tree_add(struct inode *inode)
4011 struct btrfs_root *root = BTRFS_I(inode)->root;
4012 struct btrfs_inode *entry;
4014 struct rb_node *parent;
4015 u64 ino = btrfs_ino(inode);
4017 p = &root->inode_tree.rb_node;
4020 if (inode_unhashed(inode))
4023 spin_lock(&root->inode_lock);
4026 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4028 if (ino < btrfs_ino(&entry->vfs_inode))
4029 p = &parent->rb_left;
4030 else if (ino > btrfs_ino(&entry->vfs_inode))
4031 p = &parent->rb_right;
4033 WARN_ON(!(entry->vfs_inode.i_state &
4034 (I_WILL_FREE | I_FREEING)));
4035 rb_erase(parent, &root->inode_tree);
4036 RB_CLEAR_NODE(parent);
4037 spin_unlock(&root->inode_lock);
4041 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4042 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4043 spin_unlock(&root->inode_lock);
4046 static void inode_tree_del(struct inode *inode)
4048 struct btrfs_root *root = BTRFS_I(inode)->root;
4051 spin_lock(&root->inode_lock);
4052 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4053 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4054 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4055 empty = RB_EMPTY_ROOT(&root->inode_tree);
4057 spin_unlock(&root->inode_lock);
4060 * Free space cache has inodes in the tree root, but the tree root has a
4061 * root_refs of 0, so this could end up dropping the tree root as a
4062 * snapshot, so we need the extra !root->fs_info->tree_root check to
4063 * make sure we don't drop it.
4065 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4066 root != root->fs_info->tree_root) {
4067 synchronize_srcu(&root->fs_info->subvol_srcu);
4068 spin_lock(&root->inode_lock);
4069 empty = RB_EMPTY_ROOT(&root->inode_tree);
4070 spin_unlock(&root->inode_lock);
4072 btrfs_add_dead_root(root);
4076 void btrfs_invalidate_inodes(struct btrfs_root *root)
4078 struct rb_node *node;
4079 struct rb_node *prev;
4080 struct btrfs_inode *entry;
4081 struct inode *inode;
4084 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4086 spin_lock(&root->inode_lock);
4088 node = root->inode_tree.rb_node;
4092 entry = rb_entry(node, struct btrfs_inode, rb_node);
4094 if (objectid < btrfs_ino(&entry->vfs_inode))
4095 node = node->rb_left;
4096 else if (objectid > btrfs_ino(&entry->vfs_inode))
4097 node = node->rb_right;
4103 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4104 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4108 prev = rb_next(prev);
4112 entry = rb_entry(node, struct btrfs_inode, rb_node);
4113 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4114 inode = igrab(&entry->vfs_inode);
4116 spin_unlock(&root->inode_lock);
4117 if (atomic_read(&inode->i_count) > 1)
4118 d_prune_aliases(inode);
4120 * btrfs_drop_inode will have it removed from
4121 * the inode cache when its usage count
4126 spin_lock(&root->inode_lock);
4130 if (cond_resched_lock(&root->inode_lock))
4133 node = rb_next(node);
4135 spin_unlock(&root->inode_lock);
4138 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4140 struct btrfs_iget_args *args = p;
4141 inode->i_ino = args->ino;
4142 BTRFS_I(inode)->root = args->root;
4146 static int btrfs_find_actor(struct inode *inode, void *opaque)
4148 struct btrfs_iget_args *args = opaque;
4149 return args->ino == btrfs_ino(inode) &&
4150 args->root == BTRFS_I(inode)->root;
4153 static struct inode *btrfs_iget_locked(struct super_block *s,
4155 struct btrfs_root *root)
4157 struct inode *inode;
4158 struct btrfs_iget_args args;
4159 args.ino = objectid;
4162 inode = iget5_locked(s, objectid, btrfs_find_actor,
4163 btrfs_init_locked_inode,
4168 /* Get an inode object given its location and corresponding root.
4169 * Returns in *is_new if the inode was read from disk
4171 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4172 struct btrfs_root *root, int *new)
4174 struct inode *inode;
4176 inode = btrfs_iget_locked(s, location->objectid, root);
4178 return ERR_PTR(-ENOMEM);
4180 if (inode->i_state & I_NEW) {
4181 BTRFS_I(inode)->root = root;
4182 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4183 btrfs_read_locked_inode(inode);
4184 if (!is_bad_inode(inode)) {
4185 inode_tree_add(inode);
4186 unlock_new_inode(inode);
4190 unlock_new_inode(inode);
4192 inode = ERR_PTR(-ESTALE);
4199 static struct inode *new_simple_dir(struct super_block *s,
4200 struct btrfs_key *key,
4201 struct btrfs_root *root)
4203 struct inode *inode = new_inode(s);
4206 return ERR_PTR(-ENOMEM);
4208 BTRFS_I(inode)->root = root;
4209 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4210 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4212 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4213 inode->i_op = &btrfs_dir_ro_inode_operations;
4214 inode->i_fop = &simple_dir_operations;
4215 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4216 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4221 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4223 struct inode *inode;
4224 struct btrfs_root *root = BTRFS_I(dir)->root;
4225 struct btrfs_root *sub_root = root;
4226 struct btrfs_key location;
4230 if (dentry->d_name.len > BTRFS_NAME_LEN)
4231 return ERR_PTR(-ENAMETOOLONG);
4233 if (unlikely(d_need_lookup(dentry))) {
4234 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4235 kfree(dentry->d_fsdata);
4236 dentry->d_fsdata = NULL;
4237 /* This thing is hashed, drop it for now */
4240 ret = btrfs_inode_by_name(dir, dentry, &location);
4244 return ERR_PTR(ret);
4246 if (location.objectid == 0)
4249 if (location.type == BTRFS_INODE_ITEM_KEY) {
4250 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4254 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4256 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4257 ret = fixup_tree_root_location(root, dir, dentry,
4258 &location, &sub_root);
4261 inode = ERR_PTR(ret);
4263 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4265 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4267 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4269 if (!IS_ERR(inode) && root != sub_root) {
4270 down_read(&root->fs_info->cleanup_work_sem);
4271 if (!(inode->i_sb->s_flags & MS_RDONLY))
4272 ret = btrfs_orphan_cleanup(sub_root);
4273 up_read(&root->fs_info->cleanup_work_sem);
4275 inode = ERR_PTR(ret);
4281 static int btrfs_dentry_delete(const struct dentry *dentry)
4283 struct btrfs_root *root;
4284 struct inode *inode = dentry->d_inode;
4286 if (!inode && !IS_ROOT(dentry))
4287 inode = dentry->d_parent->d_inode;
4290 root = BTRFS_I(inode)->root;
4291 if (btrfs_root_refs(&root->root_item) == 0)
4294 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4300 static void btrfs_dentry_release(struct dentry *dentry)
4302 if (dentry->d_fsdata)
4303 kfree(dentry->d_fsdata);
4306 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4311 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4312 if (unlikely(d_need_lookup(dentry))) {
4313 spin_lock(&dentry->d_lock);
4314 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4315 spin_unlock(&dentry->d_lock);
4320 unsigned char btrfs_filetype_table[] = {
4321 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4324 static int btrfs_real_readdir(struct file *filp, void *dirent,
4327 struct inode *inode = filp->f_dentry->d_inode;
4328 struct btrfs_root *root = BTRFS_I(inode)->root;
4329 struct btrfs_item *item;
4330 struct btrfs_dir_item *di;
4331 struct btrfs_key key;
4332 struct btrfs_key found_key;
4333 struct btrfs_path *path;
4334 struct list_head ins_list;
4335 struct list_head del_list;
4337 struct extent_buffer *leaf;
4339 unsigned char d_type;
4344 int key_type = BTRFS_DIR_INDEX_KEY;
4348 int is_curr = 0; /* filp->f_pos points to the current index? */
4350 /* FIXME, use a real flag for deciding about the key type */
4351 if (root->fs_info->tree_root == root)
4352 key_type = BTRFS_DIR_ITEM_KEY;
4354 /* special case for "." */
4355 if (filp->f_pos == 0) {
4356 over = filldir(dirent, ".", 1,
4357 filp->f_pos, btrfs_ino(inode), DT_DIR);
4362 /* special case for .., just use the back ref */
4363 if (filp->f_pos == 1) {
4364 u64 pino = parent_ino(filp->f_path.dentry);
4365 over = filldir(dirent, "..", 2,
4366 filp->f_pos, pino, DT_DIR);
4371 path = btrfs_alloc_path();
4377 if (key_type == BTRFS_DIR_INDEX_KEY) {
4378 INIT_LIST_HEAD(&ins_list);
4379 INIT_LIST_HEAD(&del_list);
4380 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4383 btrfs_set_key_type(&key, key_type);
4384 key.offset = filp->f_pos;
4385 key.objectid = btrfs_ino(inode);
4387 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4392 leaf = path->nodes[0];
4393 slot = path->slots[0];
4394 if (slot >= btrfs_header_nritems(leaf)) {
4395 ret = btrfs_next_leaf(root, path);
4403 item = btrfs_item_nr(leaf, slot);
4404 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4406 if (found_key.objectid != key.objectid)
4408 if (btrfs_key_type(&found_key) != key_type)
4410 if (found_key.offset < filp->f_pos)
4412 if (key_type == BTRFS_DIR_INDEX_KEY &&
4413 btrfs_should_delete_dir_index(&del_list,
4417 filp->f_pos = found_key.offset;
4420 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4422 di_total = btrfs_item_size(leaf, item);
4424 while (di_cur < di_total) {
4425 struct btrfs_key location;
4427 if (verify_dir_item(root, leaf, di))
4430 name_len = btrfs_dir_name_len(leaf, di);
4431 if (name_len <= sizeof(tmp_name)) {
4432 name_ptr = tmp_name;
4434 name_ptr = kmalloc(name_len, GFP_NOFS);
4440 read_extent_buffer(leaf, name_ptr,
4441 (unsigned long)(di + 1), name_len);
4443 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4444 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4447 /* is this a reference to our own snapshot? If so
4450 * In contrast to old kernels, we insert the snapshot's
4451 * dir item and dir index after it has been created, so
4452 * we won't find a reference to our own snapshot. We
4453 * still keep the following code for backward
4456 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4457 location.objectid == root->root_key.objectid) {
4461 over = filldir(dirent, name_ptr, name_len,
4462 found_key.offset, location.objectid,
4466 if (name_ptr != tmp_name)
4471 di_len = btrfs_dir_name_len(leaf, di) +
4472 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4474 di = (struct btrfs_dir_item *)((char *)di + di_len);
4480 if (key_type == BTRFS_DIR_INDEX_KEY) {
4483 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4489 /* Reached end of directory/root. Bump pos past the last item. */
4490 if (key_type == BTRFS_DIR_INDEX_KEY)
4492 * 32-bit glibc will use getdents64, but then strtol -
4493 * so the last number we can serve is this.
4495 filp->f_pos = 0x7fffffff;
4501 if (key_type == BTRFS_DIR_INDEX_KEY)
4502 btrfs_put_delayed_items(&ins_list, &del_list);
4503 btrfs_free_path(path);
4507 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4509 struct btrfs_root *root = BTRFS_I(inode)->root;
4510 struct btrfs_trans_handle *trans;
4512 bool nolock = false;
4514 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4517 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
4520 if (wbc->sync_mode == WB_SYNC_ALL) {
4522 trans = btrfs_join_transaction_nolock(root);
4524 trans = btrfs_join_transaction(root);
4526 return PTR_ERR(trans);
4527 ret = btrfs_commit_transaction(trans, root);
4533 * This is somewhat expensive, updating the tree every time the
4534 * inode changes. But, it is most likely to find the inode in cache.
4535 * FIXME, needs more benchmarking...there are no reasons other than performance
4536 * to keep or drop this code.
4538 int btrfs_dirty_inode(struct inode *inode)
4540 struct btrfs_root *root = BTRFS_I(inode)->root;
4541 struct btrfs_trans_handle *trans;
4544 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4547 trans = btrfs_join_transaction(root);
4549 return PTR_ERR(trans);
4551 ret = btrfs_update_inode(trans, root, inode);
4552 if (ret && ret == -ENOSPC) {
4553 /* whoops, lets try again with the full transaction */
4554 btrfs_end_transaction(trans, root);
4555 trans = btrfs_start_transaction(root, 1);
4557 return PTR_ERR(trans);
4559 ret = btrfs_update_inode(trans, root, inode);
4561 btrfs_end_transaction(trans, root);
4562 if (BTRFS_I(inode)->delayed_node)
4563 btrfs_balance_delayed_items(root);
4569 * This is a copy of file_update_time. We need this so we can return error on
4570 * ENOSPC for updating the inode in the case of file write and mmap writes.
4572 static int btrfs_update_time(struct inode *inode, struct timespec *now,
4575 struct btrfs_root *root = BTRFS_I(inode)->root;
4577 if (btrfs_root_readonly(root))
4580 if (flags & S_VERSION)
4581 inode_inc_iversion(inode);
4582 if (flags & S_CTIME)
4583 inode->i_ctime = *now;
4584 if (flags & S_MTIME)
4585 inode->i_mtime = *now;
4586 if (flags & S_ATIME)
4587 inode->i_atime = *now;
4588 return btrfs_dirty_inode(inode);
4592 * find the highest existing sequence number in a directory
4593 * and then set the in-memory index_cnt variable to reflect
4594 * free sequence numbers
4596 static int btrfs_set_inode_index_count(struct inode *inode)
4598 struct btrfs_root *root = BTRFS_I(inode)->root;
4599 struct btrfs_key key, found_key;
4600 struct btrfs_path *path;
4601 struct extent_buffer *leaf;
4604 key.objectid = btrfs_ino(inode);
4605 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4606 key.offset = (u64)-1;
4608 path = btrfs_alloc_path();
4612 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4615 /* FIXME: we should be able to handle this */
4621 * MAGIC NUMBER EXPLANATION:
4622 * since we search a directory based on f_pos we have to start at 2
4623 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4624 * else has to start at 2
4626 if (path->slots[0] == 0) {
4627 BTRFS_I(inode)->index_cnt = 2;
4633 leaf = path->nodes[0];
4634 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4636 if (found_key.objectid != btrfs_ino(inode) ||
4637 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4638 BTRFS_I(inode)->index_cnt = 2;
4642 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4644 btrfs_free_path(path);
4649 * helper to find a free sequence number in a given directory. This current
4650 * code is very simple, later versions will do smarter things in the btree
4652 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4656 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4657 ret = btrfs_inode_delayed_dir_index_count(dir);
4659 ret = btrfs_set_inode_index_count(dir);
4665 *index = BTRFS_I(dir)->index_cnt;
4666 BTRFS_I(dir)->index_cnt++;
4671 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4672 struct btrfs_root *root,
4674 const char *name, int name_len,
4675 u64 ref_objectid, u64 objectid,
4676 umode_t mode, u64 *index)
4678 struct inode *inode;
4679 struct btrfs_inode_item *inode_item;
4680 struct btrfs_key *location;
4681 struct btrfs_path *path;
4682 struct btrfs_inode_ref *ref;
4683 struct btrfs_key key[2];
4689 path = btrfs_alloc_path();
4691 return ERR_PTR(-ENOMEM);
4693 inode = new_inode(root->fs_info->sb);
4695 btrfs_free_path(path);
4696 return ERR_PTR(-ENOMEM);
4700 * we have to initialize this early, so we can reclaim the inode
4701 * number if we fail afterwards in this function.
4703 inode->i_ino = objectid;
4706 trace_btrfs_inode_request(dir);
4708 ret = btrfs_set_inode_index(dir, index);
4710 btrfs_free_path(path);
4712 return ERR_PTR(ret);
4716 * index_cnt is ignored for everything but a dir,
4717 * btrfs_get_inode_index_count has an explanation for the magic
4720 BTRFS_I(inode)->index_cnt = 2;
4721 BTRFS_I(inode)->root = root;
4722 BTRFS_I(inode)->generation = trans->transid;
4723 inode->i_generation = BTRFS_I(inode)->generation;
4726 * We could have gotten an inode number from somebody who was fsynced
4727 * and then removed in this same transaction, so let's just set full
4728 * sync since it will be a full sync anyway and this will blow away the
4729 * old info in the log.
4731 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
4738 key[0].objectid = objectid;
4739 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4743 * Start new inodes with an inode_ref. This is slightly more
4744 * efficient for small numbers of hard links since they will
4745 * be packed into one item. Extended refs will kick in if we
4746 * add more hard links than can fit in the ref item.
4748 key[1].objectid = objectid;
4749 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4750 key[1].offset = ref_objectid;
4752 sizes[0] = sizeof(struct btrfs_inode_item);
4753 sizes[1] = name_len + sizeof(*ref);
4755 path->leave_spinning = 1;
4756 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4760 inode_init_owner(inode, dir, mode);
4761 inode_set_bytes(inode, 0);
4762 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4763 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4764 struct btrfs_inode_item);
4765 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
4766 sizeof(*inode_item));
4767 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4769 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4770 struct btrfs_inode_ref);
4771 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4772 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4773 ptr = (unsigned long)(ref + 1);
4774 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4776 btrfs_mark_buffer_dirty(path->nodes[0]);
4777 btrfs_free_path(path);
4779 location = &BTRFS_I(inode)->location;
4780 location->objectid = objectid;
4781 location->offset = 0;
4782 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4784 btrfs_inherit_iflags(inode, dir);
4786 if (S_ISREG(mode)) {
4787 if (btrfs_test_opt(root, NODATASUM))
4788 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4789 if (btrfs_test_opt(root, NODATACOW) ||
4790 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4791 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4794 insert_inode_hash(inode);
4795 inode_tree_add(inode);
4797 trace_btrfs_inode_new(inode);
4798 btrfs_set_inode_last_trans(trans, inode);
4800 btrfs_update_root_times(trans, root);
4805 BTRFS_I(dir)->index_cnt--;
4806 btrfs_free_path(path);
4808 return ERR_PTR(ret);
4811 static inline u8 btrfs_inode_type(struct inode *inode)
4813 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4817 * utility function to add 'inode' into 'parent_inode' with
4818 * a give name and a given sequence number.
4819 * if 'add_backref' is true, also insert a backref from the
4820 * inode to the parent directory.
4822 int btrfs_add_link(struct btrfs_trans_handle *trans,
4823 struct inode *parent_inode, struct inode *inode,
4824 const char *name, int name_len, int add_backref, u64 index)
4827 struct btrfs_key key;
4828 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4829 u64 ino = btrfs_ino(inode);
4830 u64 parent_ino = btrfs_ino(parent_inode);
4832 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4833 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4836 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4840 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4841 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4842 key.objectid, root->root_key.objectid,
4843 parent_ino, index, name, name_len);
4844 } else if (add_backref) {
4845 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4849 /* Nothing to clean up yet */
4853 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4855 btrfs_inode_type(inode), index);
4859 btrfs_abort_transaction(trans, root, ret);
4863 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4865 inode_inc_iversion(parent_inode);
4866 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4867 ret = btrfs_update_inode(trans, root, parent_inode);
4869 btrfs_abort_transaction(trans, root, ret);
4873 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4876 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4877 key.objectid, root->root_key.objectid,
4878 parent_ino, &local_index, name, name_len);
4880 } else if (add_backref) {
4884 err = btrfs_del_inode_ref(trans, root, name, name_len,
4885 ino, parent_ino, &local_index);
4890 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4891 struct inode *dir, struct dentry *dentry,
4892 struct inode *inode, int backref, u64 index)
4894 int err = btrfs_add_link(trans, dir, inode,
4895 dentry->d_name.name, dentry->d_name.len,
4902 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4903 umode_t mode, dev_t rdev)
4905 struct btrfs_trans_handle *trans;
4906 struct btrfs_root *root = BTRFS_I(dir)->root;
4907 struct inode *inode = NULL;
4913 if (!new_valid_dev(rdev))
4917 * 2 for inode item and ref
4919 * 1 for xattr if selinux is on
4921 trans = btrfs_start_transaction(root, 5);
4923 return PTR_ERR(trans);
4925 err = btrfs_find_free_ino(root, &objectid);
4929 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4930 dentry->d_name.len, btrfs_ino(dir), objectid,
4932 if (IS_ERR(inode)) {
4933 err = PTR_ERR(inode);
4937 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4944 * If the active LSM wants to access the inode during
4945 * d_instantiate it needs these. Smack checks to see
4946 * if the filesystem supports xattrs by looking at the
4950 inode->i_op = &btrfs_special_inode_operations;
4951 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4955 init_special_inode(inode, inode->i_mode, rdev);
4956 btrfs_update_inode(trans, root, inode);
4957 d_instantiate(dentry, inode);
4960 btrfs_end_transaction(trans, root);
4961 btrfs_btree_balance_dirty(root);
4963 inode_dec_link_count(inode);
4969 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4970 umode_t mode, bool excl)
4972 struct btrfs_trans_handle *trans;
4973 struct btrfs_root *root = BTRFS_I(dir)->root;
4974 struct inode *inode = NULL;
4981 * 2 for inode item and ref
4983 * 1 for xattr if selinux is on
4985 trans = btrfs_start_transaction(root, 5);
4987 return PTR_ERR(trans);
4989 err = btrfs_find_free_ino(root, &objectid);
4993 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4994 dentry->d_name.len, btrfs_ino(dir), objectid,
4996 if (IS_ERR(inode)) {
4997 err = PTR_ERR(inode);
5001 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5008 * If the active LSM wants to access the inode during
5009 * d_instantiate it needs these. Smack checks to see
5010 * if the filesystem supports xattrs by looking at the
5013 inode->i_fop = &btrfs_file_operations;
5014 inode->i_op = &btrfs_file_inode_operations;
5016 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5020 inode->i_mapping->a_ops = &btrfs_aops;
5021 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5022 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5023 d_instantiate(dentry, inode);
5026 btrfs_end_transaction(trans, root);
5028 inode_dec_link_count(inode);
5031 btrfs_btree_balance_dirty(root);
5035 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5036 struct dentry *dentry)
5038 struct btrfs_trans_handle *trans;
5039 struct btrfs_root *root = BTRFS_I(dir)->root;
5040 struct inode *inode = old_dentry->d_inode;
5045 /* do not allow sys_link's with other subvols of the same device */
5046 if (root->objectid != BTRFS_I(inode)->root->objectid)
5049 if (inode->i_nlink >= BTRFS_LINK_MAX)
5052 err = btrfs_set_inode_index(dir, &index);
5057 * 2 items for inode and inode ref
5058 * 2 items for dir items
5059 * 1 item for parent inode
5061 trans = btrfs_start_transaction(root, 5);
5062 if (IS_ERR(trans)) {
5063 err = PTR_ERR(trans);
5067 btrfs_inc_nlink(inode);
5068 inode_inc_iversion(inode);
5069 inode->i_ctime = CURRENT_TIME;
5072 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5077 struct dentry *parent = dentry->d_parent;
5078 err = btrfs_update_inode(trans, root, inode);
5081 d_instantiate(dentry, inode);
5082 btrfs_log_new_name(trans, inode, NULL, parent);
5085 btrfs_end_transaction(trans, root);
5088 inode_dec_link_count(inode);
5091 btrfs_btree_balance_dirty(root);
5095 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5097 struct inode *inode = NULL;
5098 struct btrfs_trans_handle *trans;
5099 struct btrfs_root *root = BTRFS_I(dir)->root;
5101 int drop_on_err = 0;
5106 * 2 items for inode and ref
5107 * 2 items for dir items
5108 * 1 for xattr if selinux is on
5110 trans = btrfs_start_transaction(root, 5);
5112 return PTR_ERR(trans);
5114 err = btrfs_find_free_ino(root, &objectid);
5118 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5119 dentry->d_name.len, btrfs_ino(dir), objectid,
5120 S_IFDIR | mode, &index);
5121 if (IS_ERR(inode)) {
5122 err = PTR_ERR(inode);
5128 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5132 inode->i_op = &btrfs_dir_inode_operations;
5133 inode->i_fop = &btrfs_dir_file_operations;
5135 btrfs_i_size_write(inode, 0);
5136 err = btrfs_update_inode(trans, root, inode);
5140 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5141 dentry->d_name.len, 0, index);
5145 d_instantiate(dentry, inode);
5149 btrfs_end_transaction(trans, root);
5152 btrfs_btree_balance_dirty(root);
5156 /* helper for btfs_get_extent. Given an existing extent in the tree,
5157 * and an extent that you want to insert, deal with overlap and insert
5158 * the new extent into the tree.
5160 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5161 struct extent_map *existing,
5162 struct extent_map *em,
5163 u64 map_start, u64 map_len)
5167 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5168 start_diff = map_start - em->start;
5169 em->start = map_start;
5171 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5172 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5173 em->block_start += start_diff;
5174 em->block_len -= start_diff;
5176 return add_extent_mapping(em_tree, em);
5179 static noinline int uncompress_inline(struct btrfs_path *path,
5180 struct inode *inode, struct page *page,
5181 size_t pg_offset, u64 extent_offset,
5182 struct btrfs_file_extent_item *item)
5185 struct extent_buffer *leaf = path->nodes[0];
5188 unsigned long inline_size;
5192 WARN_ON(pg_offset != 0);
5193 compress_type = btrfs_file_extent_compression(leaf, item);
5194 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5195 inline_size = btrfs_file_extent_inline_item_len(leaf,
5196 btrfs_item_nr(leaf, path->slots[0]));
5197 tmp = kmalloc(inline_size, GFP_NOFS);
5200 ptr = btrfs_file_extent_inline_start(item);
5202 read_extent_buffer(leaf, tmp, ptr, inline_size);
5204 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5205 ret = btrfs_decompress(compress_type, tmp, page,
5206 extent_offset, inline_size, max_size);
5208 char *kaddr = kmap_atomic(page);
5209 unsigned long copy_size = min_t(u64,
5210 PAGE_CACHE_SIZE - pg_offset,
5211 max_size - extent_offset);
5212 memset(kaddr + pg_offset, 0, copy_size);
5213 kunmap_atomic(kaddr);
5220 * a bit scary, this does extent mapping from logical file offset to the disk.
5221 * the ugly parts come from merging extents from the disk with the in-ram
5222 * representation. This gets more complex because of the data=ordered code,
5223 * where the in-ram extents might be locked pending data=ordered completion.
5225 * This also copies inline extents directly into the page.
5228 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5229 size_t pg_offset, u64 start, u64 len,
5235 u64 extent_start = 0;
5237 u64 objectid = btrfs_ino(inode);
5239 struct btrfs_path *path = NULL;
5240 struct btrfs_root *root = BTRFS_I(inode)->root;
5241 struct btrfs_file_extent_item *item;
5242 struct extent_buffer *leaf;
5243 struct btrfs_key found_key;
5244 struct extent_map *em = NULL;
5245 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5246 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5247 struct btrfs_trans_handle *trans = NULL;
5251 read_lock(&em_tree->lock);
5252 em = lookup_extent_mapping(em_tree, start, len);
5254 em->bdev = root->fs_info->fs_devices->latest_bdev;
5255 read_unlock(&em_tree->lock);
5258 if (em->start > start || em->start + em->len <= start)
5259 free_extent_map(em);
5260 else if (em->block_start == EXTENT_MAP_INLINE && page)
5261 free_extent_map(em);
5265 em = alloc_extent_map();
5270 em->bdev = root->fs_info->fs_devices->latest_bdev;
5271 em->start = EXTENT_MAP_HOLE;
5272 em->orig_start = EXTENT_MAP_HOLE;
5274 em->block_len = (u64)-1;
5277 path = btrfs_alloc_path();
5283 * Chances are we'll be called again, so go ahead and do
5289 ret = btrfs_lookup_file_extent(trans, root, path,
5290 objectid, start, trans != NULL);
5297 if (path->slots[0] == 0)
5302 leaf = path->nodes[0];
5303 item = btrfs_item_ptr(leaf, path->slots[0],
5304 struct btrfs_file_extent_item);
5305 /* are we inside the extent that was found? */
5306 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5307 found_type = btrfs_key_type(&found_key);
5308 if (found_key.objectid != objectid ||
5309 found_type != BTRFS_EXTENT_DATA_KEY) {
5313 found_type = btrfs_file_extent_type(leaf, item);
5314 extent_start = found_key.offset;
5315 compress_type = btrfs_file_extent_compression(leaf, item);
5316 if (found_type == BTRFS_FILE_EXTENT_REG ||
5317 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5318 extent_end = extent_start +
5319 btrfs_file_extent_num_bytes(leaf, item);
5320 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5322 size = btrfs_file_extent_inline_len(leaf, item);
5323 extent_end = (extent_start + size + root->sectorsize - 1) &
5324 ~((u64)root->sectorsize - 1);
5327 if (start >= extent_end) {
5329 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5330 ret = btrfs_next_leaf(root, path);
5337 leaf = path->nodes[0];
5339 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5340 if (found_key.objectid != objectid ||
5341 found_key.type != BTRFS_EXTENT_DATA_KEY)
5343 if (start + len <= found_key.offset)
5346 em->len = found_key.offset - start;
5350 if (found_type == BTRFS_FILE_EXTENT_REG ||
5351 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5352 em->start = extent_start;
5353 em->len = extent_end - extent_start;
5354 em->orig_start = extent_start -
5355 btrfs_file_extent_offset(leaf, item);
5356 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5358 em->block_start = EXTENT_MAP_HOLE;
5361 if (compress_type != BTRFS_COMPRESS_NONE) {
5362 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5363 em->compress_type = compress_type;
5364 em->block_start = bytenr;
5365 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5368 bytenr += btrfs_file_extent_offset(leaf, item);
5369 em->block_start = bytenr;
5370 em->block_len = em->len;
5371 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5372 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5375 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5379 size_t extent_offset;
5382 em->block_start = EXTENT_MAP_INLINE;
5383 if (!page || create) {
5384 em->start = extent_start;
5385 em->len = extent_end - extent_start;
5389 size = btrfs_file_extent_inline_len(leaf, item);
5390 extent_offset = page_offset(page) + pg_offset - extent_start;
5391 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5392 size - extent_offset);
5393 em->start = extent_start + extent_offset;
5394 em->len = (copy_size + root->sectorsize - 1) &
5395 ~((u64)root->sectorsize - 1);
5396 em->orig_start = EXTENT_MAP_INLINE;
5397 if (compress_type) {
5398 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5399 em->compress_type = compress_type;
5401 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5402 if (create == 0 && !PageUptodate(page)) {
5403 if (btrfs_file_extent_compression(leaf, item) !=
5404 BTRFS_COMPRESS_NONE) {
5405 ret = uncompress_inline(path, inode, page,
5407 extent_offset, item);
5408 BUG_ON(ret); /* -ENOMEM */
5411 read_extent_buffer(leaf, map + pg_offset, ptr,
5413 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5414 memset(map + pg_offset + copy_size, 0,
5415 PAGE_CACHE_SIZE - pg_offset -
5420 flush_dcache_page(page);
5421 } else if (create && PageUptodate(page)) {
5425 free_extent_map(em);
5428 btrfs_release_path(path);
5429 trans = btrfs_join_transaction(root);
5432 return ERR_CAST(trans);
5436 write_extent_buffer(leaf, map + pg_offset, ptr,
5439 btrfs_mark_buffer_dirty(leaf);
5441 set_extent_uptodate(io_tree, em->start,
5442 extent_map_end(em) - 1, NULL, GFP_NOFS);
5445 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
5451 em->block_start = EXTENT_MAP_HOLE;
5452 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5454 btrfs_release_path(path);
5455 if (em->start > start || extent_map_end(em) <= start) {
5456 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5457 "[%llu %llu]\n", (unsigned long long)em->start,
5458 (unsigned long long)em->len,
5459 (unsigned long long)start,
5460 (unsigned long long)len);
5466 write_lock(&em_tree->lock);
5467 ret = add_extent_mapping(em_tree, em);
5468 /* it is possible that someone inserted the extent into the tree
5469 * while we had the lock dropped. It is also possible that
5470 * an overlapping map exists in the tree
5472 if (ret == -EEXIST) {
5473 struct extent_map *existing;
5477 existing = lookup_extent_mapping(em_tree, start, len);
5478 if (existing && (existing->start > start ||
5479 existing->start + existing->len <= start)) {
5480 free_extent_map(existing);
5484 existing = lookup_extent_mapping(em_tree, em->start,
5487 err = merge_extent_mapping(em_tree, existing,
5490 free_extent_map(existing);
5492 free_extent_map(em);
5497 free_extent_map(em);
5501 free_extent_map(em);
5506 write_unlock(&em_tree->lock);
5510 trace_btrfs_get_extent(root, em);
5513 btrfs_free_path(path);
5515 ret = btrfs_end_transaction(trans, root);
5520 free_extent_map(em);
5521 return ERR_PTR(err);
5523 BUG_ON(!em); /* Error is always set */
5527 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5528 size_t pg_offset, u64 start, u64 len,
5531 struct extent_map *em;
5532 struct extent_map *hole_em = NULL;
5533 u64 range_start = start;
5539 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5544 * if our em maps to a hole, there might
5545 * actually be delalloc bytes behind it
5547 if (em->block_start != EXTENT_MAP_HOLE)
5553 /* check to see if we've wrapped (len == -1 or similar) */
5562 /* ok, we didn't find anything, lets look for delalloc */
5563 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5564 end, len, EXTENT_DELALLOC, 1);
5565 found_end = range_start + found;
5566 if (found_end < range_start)
5567 found_end = (u64)-1;
5570 * we didn't find anything useful, return
5571 * the original results from get_extent()
5573 if (range_start > end || found_end <= start) {
5579 /* adjust the range_start to make sure it doesn't
5580 * go backwards from the start they passed in
5582 range_start = max(start,range_start);
5583 found = found_end - range_start;
5586 u64 hole_start = start;
5589 em = alloc_extent_map();
5595 * when btrfs_get_extent can't find anything it
5596 * returns one huge hole
5598 * make sure what it found really fits our range, and
5599 * adjust to make sure it is based on the start from
5603 u64 calc_end = extent_map_end(hole_em);
5605 if (calc_end <= start || (hole_em->start > end)) {
5606 free_extent_map(hole_em);
5609 hole_start = max(hole_em->start, start);
5610 hole_len = calc_end - hole_start;
5614 if (hole_em && range_start > hole_start) {
5615 /* our hole starts before our delalloc, so we
5616 * have to return just the parts of the hole
5617 * that go until the delalloc starts
5619 em->len = min(hole_len,
5620 range_start - hole_start);
5621 em->start = hole_start;
5622 em->orig_start = hole_start;
5624 * don't adjust block start at all,
5625 * it is fixed at EXTENT_MAP_HOLE
5627 em->block_start = hole_em->block_start;
5628 em->block_len = hole_len;
5630 em->start = range_start;
5632 em->orig_start = range_start;
5633 em->block_start = EXTENT_MAP_DELALLOC;
5634 em->block_len = found;
5636 } else if (hole_em) {
5641 free_extent_map(hole_em);
5643 free_extent_map(em);
5644 return ERR_PTR(err);
5649 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5650 struct extent_map *em,
5653 struct btrfs_root *root = BTRFS_I(inode)->root;
5654 struct btrfs_trans_handle *trans;
5655 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5656 struct btrfs_key ins;
5659 bool insert = false;
5662 * Ok if the extent map we looked up is a hole and is for the exact
5663 * range we want, there is no reason to allocate a new one, however if
5664 * it is not right then we need to free this one and drop the cache for
5667 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5669 free_extent_map(em);
5672 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5675 trans = btrfs_join_transaction(root);
5677 return ERR_CAST(trans);
5679 if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5680 btrfs_add_inode_defrag(trans, inode);
5682 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5684 alloc_hint = get_extent_allocation_hint(inode, start, len);
5685 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5686 alloc_hint, &ins, 1);
5693 em = alloc_extent_map();
5695 em = ERR_PTR(-ENOMEM);
5701 em->orig_start = em->start;
5702 em->len = ins.offset;
5704 em->block_start = ins.objectid;
5705 em->block_len = ins.offset;
5706 em->bdev = root->fs_info->fs_devices->latest_bdev;
5709 * We need to do this because if we're using the original em we searched
5710 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5713 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5716 write_lock(&em_tree->lock);
5717 ret = add_extent_mapping(em_tree, em);
5718 write_unlock(&em_tree->lock);
5721 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5724 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5725 ins.offset, ins.offset, 0);
5727 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5731 btrfs_end_transaction(trans, root);
5736 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5737 * block must be cow'd
5739 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5740 struct inode *inode, u64 offset, u64 len)
5742 struct btrfs_path *path;
5744 struct extent_buffer *leaf;
5745 struct btrfs_root *root = BTRFS_I(inode)->root;
5746 struct btrfs_file_extent_item *fi;
5747 struct btrfs_key key;
5755 path = btrfs_alloc_path();
5759 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5764 slot = path->slots[0];
5767 /* can't find the item, must cow */
5774 leaf = path->nodes[0];
5775 btrfs_item_key_to_cpu(leaf, &key, slot);
5776 if (key.objectid != btrfs_ino(inode) ||
5777 key.type != BTRFS_EXTENT_DATA_KEY) {
5778 /* not our file or wrong item type, must cow */
5782 if (key.offset > offset) {
5783 /* Wrong offset, must cow */
5787 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5788 found_type = btrfs_file_extent_type(leaf, fi);
5789 if (found_type != BTRFS_FILE_EXTENT_REG &&
5790 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5791 /* not a regular extent, must cow */
5794 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5795 backref_offset = btrfs_file_extent_offset(leaf, fi);
5797 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5798 if (extent_end < offset + len) {
5799 /* extent doesn't include our full range, must cow */
5803 if (btrfs_extent_readonly(root, disk_bytenr))
5807 * look for other files referencing this extent, if we
5808 * find any we must cow
5810 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5811 key.offset - backref_offset, disk_bytenr))
5815 * adjust disk_bytenr and num_bytes to cover just the bytes
5816 * in this extent we are about to write. If there
5817 * are any csums in that range we have to cow in order
5818 * to keep the csums correct
5820 disk_bytenr += backref_offset;
5821 disk_bytenr += offset - key.offset;
5822 num_bytes = min(offset + len, extent_end) - offset;
5823 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5826 * all of the above have passed, it is safe to overwrite this extent
5831 btrfs_free_path(path);
5835 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
5836 struct extent_state **cached_state, int writing)
5838 struct btrfs_ordered_extent *ordered;
5842 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5845 * We're concerned with the entire range that we're going to be
5846 * doing DIO to, so we need to make sure theres no ordered
5847 * extents in this range.
5849 ordered = btrfs_lookup_ordered_range(inode, lockstart,
5850 lockend - lockstart + 1);
5853 * We need to make sure there are no buffered pages in this
5854 * range either, we could have raced between the invalidate in
5855 * generic_file_direct_write and locking the extent. The
5856 * invalidate needs to happen so that reads after a write do not
5859 if (!ordered && (!writing ||
5860 !test_range_bit(&BTRFS_I(inode)->io_tree,
5861 lockstart, lockend, EXTENT_UPTODATE, 0,
5865 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5866 cached_state, GFP_NOFS);
5869 btrfs_start_ordered_extent(inode, ordered, 1);
5870 btrfs_put_ordered_extent(ordered);
5872 /* Screw you mmap */
5873 ret = filemap_write_and_wait_range(inode->i_mapping,
5880 * If we found a page that couldn't be invalidated just
5881 * fall back to buffered.
5883 ret = invalidate_inode_pages2_range(inode->i_mapping,
5884 lockstart >> PAGE_CACHE_SHIFT,
5885 lockend >> PAGE_CACHE_SHIFT);
5896 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
5897 u64 len, u64 orig_start,
5898 u64 block_start, u64 block_len,
5901 struct extent_map_tree *em_tree;
5902 struct extent_map *em;
5903 struct btrfs_root *root = BTRFS_I(inode)->root;
5906 em_tree = &BTRFS_I(inode)->extent_tree;
5907 em = alloc_extent_map();
5909 return ERR_PTR(-ENOMEM);
5912 em->orig_start = orig_start;
5914 em->block_len = block_len;
5915 em->block_start = block_start;
5916 em->bdev = root->fs_info->fs_devices->latest_bdev;
5917 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5918 if (type == BTRFS_ORDERED_PREALLOC)
5919 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5922 btrfs_drop_extent_cache(inode, em->start,
5923 em->start + em->len - 1, 0);
5924 write_lock(&em_tree->lock);
5925 ret = add_extent_mapping(em_tree, em);
5926 write_unlock(&em_tree->lock);
5927 } while (ret == -EEXIST);
5930 free_extent_map(em);
5931 return ERR_PTR(ret);
5938 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5939 struct buffer_head *bh_result, int create)
5941 struct extent_map *em;
5942 struct btrfs_root *root = BTRFS_I(inode)->root;
5943 struct extent_state *cached_state = NULL;
5944 u64 start = iblock << inode->i_blkbits;
5945 u64 lockstart, lockend;
5946 u64 len = bh_result->b_size;
5947 struct btrfs_trans_handle *trans;
5948 int unlock_bits = EXTENT_LOCKED;
5952 ret = btrfs_delalloc_reserve_space(inode, len);
5955 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
5957 len = min_t(u64, len, root->sectorsize);
5961 lockend = start + len - 1;
5964 * If this errors out it's because we couldn't invalidate pagecache for
5965 * this range and we need to fallback to buffered.
5967 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
5971 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
5972 lockend, EXTENT_DELALLOC, NULL,
5973 &cached_state, GFP_NOFS);
5978 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5985 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5986 * io. INLINE is special, and we could probably kludge it in here, but
5987 * it's still buffered so for safety lets just fall back to the generic
5990 * For COMPRESSED we _have_ to read the entire extent in so we can
5991 * decompress it, so there will be buffering required no matter what we
5992 * do, so go ahead and fallback to buffered.
5994 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5995 * to buffered IO. Don't blame me, this is the price we pay for using
5998 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5999 em->block_start == EXTENT_MAP_INLINE) {
6000 free_extent_map(em);
6005 /* Just a good old fashioned hole, return */
6006 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6007 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6008 free_extent_map(em);
6014 * We don't allocate a new extent in the following cases
6016 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6018 * 2) The extent is marked as PREALLOC. We're good to go here and can
6019 * just use the extent.
6023 len = min(len, em->len - (start - em->start));
6024 lockstart = start + len;
6028 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6029 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6030 em->block_start != EXTENT_MAP_HOLE)) {
6035 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6036 type = BTRFS_ORDERED_PREALLOC;
6038 type = BTRFS_ORDERED_NOCOW;
6039 len = min(len, em->len - (start - em->start));
6040 block_start = em->block_start + (start - em->start);
6043 * we're not going to log anything, but we do need
6044 * to make sure the current transaction stays open
6045 * while we look for nocow cross refs
6047 trans = btrfs_join_transaction(root);
6051 if (can_nocow_odirect(trans, inode, start, len) == 1) {
6052 u64 orig_start = em->start;
6054 if (type == BTRFS_ORDERED_PREALLOC) {
6055 free_extent_map(em);
6056 em = create_pinned_em(inode, start, len,
6058 block_start, len, type);
6060 btrfs_end_transaction(trans, root);
6065 ret = btrfs_add_ordered_extent_dio(inode, start,
6066 block_start, len, len, type);
6067 btrfs_end_transaction(trans, root);
6069 free_extent_map(em);
6074 btrfs_end_transaction(trans, root);
6078 * this will cow the extent, reset the len in case we changed
6081 len = bh_result->b_size;
6082 em = btrfs_new_extent_direct(inode, em, start, len);
6087 len = min(len, em->len - (start - em->start));
6089 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6091 bh_result->b_size = len;
6092 bh_result->b_bdev = em->bdev;
6093 set_buffer_mapped(bh_result);
6095 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6096 set_buffer_new(bh_result);
6099 * Need to update the i_size under the extent lock so buffered
6100 * readers will get the updated i_size when we unlock.
6102 if (start + len > i_size_read(inode))
6103 i_size_write(inode, start + len);
6107 * In the case of write we need to clear and unlock the entire range,
6108 * in the case of read we need to unlock only the end area that we
6109 * aren't using if there is any left over space.
6111 if (lockstart < lockend) {
6112 if (create && len < lockend - lockstart) {
6113 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6114 lockstart + len - 1,
6115 unlock_bits | EXTENT_DEFRAG, 1, 0,
6116 &cached_state, GFP_NOFS);
6118 * Beside unlock, we also need to cleanup reserved space
6119 * for the left range by attaching EXTENT_DO_ACCOUNTING.
6121 clear_extent_bit(&BTRFS_I(inode)->io_tree,
6122 lockstart + len, lockend,
6123 unlock_bits | EXTENT_DO_ACCOUNTING |
6124 EXTENT_DEFRAG, 1, 0, NULL, GFP_NOFS);
6126 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6127 lockend, unlock_bits, 1, 0,
6128 &cached_state, GFP_NOFS);
6131 free_extent_state(cached_state);
6134 free_extent_map(em);
6140 unlock_bits |= EXTENT_DO_ACCOUNTING;
6142 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6143 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6147 struct btrfs_dio_private {
6148 struct inode *inode;
6154 /* number of bios pending for this dio */
6155 atomic_t pending_bios;
6160 struct bio *orig_bio;
6163 static void btrfs_endio_direct_read(struct bio *bio, int err)
6165 struct btrfs_dio_private *dip = bio->bi_private;
6166 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6167 struct bio_vec *bvec = bio->bi_io_vec;
6168 struct inode *inode = dip->inode;
6169 struct btrfs_root *root = BTRFS_I(inode)->root;
6172 start = dip->logical_offset;
6174 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6175 struct page *page = bvec->bv_page;
6178 u64 private = ~(u32)0;
6179 unsigned long flags;
6181 if (get_state_private(&BTRFS_I(inode)->io_tree,
6184 local_irq_save(flags);
6185 kaddr = kmap_atomic(page);
6186 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
6187 csum, bvec->bv_len);
6188 btrfs_csum_final(csum, (char *)&csum);
6189 kunmap_atomic(kaddr);
6190 local_irq_restore(flags);
6192 flush_dcache_page(bvec->bv_page);
6193 if (csum != private) {
6195 printk(KERN_ERR "btrfs csum failed ino %llu off"
6196 " %llu csum %u private %u\n",
6197 (unsigned long long)btrfs_ino(inode),
6198 (unsigned long long)start,
6199 csum, (unsigned)private);
6204 start += bvec->bv_len;
6206 } while (bvec <= bvec_end);
6208 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6209 dip->logical_offset + dip->bytes - 1);
6210 bio->bi_private = dip->private;
6214 /* If we had a csum failure make sure to clear the uptodate flag */
6216 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6217 dio_end_io(bio, err);
6220 static void btrfs_endio_direct_write(struct bio *bio, int err)
6222 struct btrfs_dio_private *dip = bio->bi_private;
6223 struct inode *inode = dip->inode;
6224 struct btrfs_root *root = BTRFS_I(inode)->root;
6225 struct btrfs_ordered_extent *ordered = NULL;
6226 u64 ordered_offset = dip->logical_offset;
6227 u64 ordered_bytes = dip->bytes;
6233 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6235 ordered_bytes, !err);
6239 ordered->work.func = finish_ordered_fn;
6240 ordered->work.flags = 0;
6241 btrfs_queue_worker(&root->fs_info->endio_write_workers,
6245 * our bio might span multiple ordered extents. If we haven't
6246 * completed the accounting for the whole dio, go back and try again
6248 if (ordered_offset < dip->logical_offset + dip->bytes) {
6249 ordered_bytes = dip->logical_offset + dip->bytes -
6255 bio->bi_private = dip->private;
6259 /* If we had an error make sure to clear the uptodate flag */
6261 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6262 dio_end_io(bio, err);
6265 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6266 struct bio *bio, int mirror_num,
6267 unsigned long bio_flags, u64 offset)
6270 struct btrfs_root *root = BTRFS_I(inode)->root;
6271 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6272 BUG_ON(ret); /* -ENOMEM */
6276 static void btrfs_end_dio_bio(struct bio *bio, int err)
6278 struct btrfs_dio_private *dip = bio->bi_private;
6281 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6282 "sector %#Lx len %u err no %d\n",
6283 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6284 (unsigned long long)bio->bi_sector, bio->bi_size, err);
6288 * before atomic variable goto zero, we must make sure
6289 * dip->errors is perceived to be set.
6291 smp_mb__before_atomic_dec();
6294 /* if there are more bios still pending for this dio, just exit */
6295 if (!atomic_dec_and_test(&dip->pending_bios))
6299 bio_io_error(dip->orig_bio);
6301 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6302 bio_endio(dip->orig_bio, 0);
6308 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6309 u64 first_sector, gfp_t gfp_flags)
6311 int nr_vecs = bio_get_nr_vecs(bdev);
6312 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6315 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6316 int rw, u64 file_offset, int skip_sum,
6319 int write = rw & REQ_WRITE;
6320 struct btrfs_root *root = BTRFS_I(inode)->root;
6326 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6334 if (write && async_submit) {
6335 ret = btrfs_wq_submit_bio(root->fs_info,
6336 inode, rw, bio, 0, 0,
6338 __btrfs_submit_bio_start_direct_io,
6339 __btrfs_submit_bio_done);
6343 * If we aren't doing async submit, calculate the csum of the
6346 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6349 } else if (!skip_sum) {
6350 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
6356 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6362 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6365 struct inode *inode = dip->inode;
6366 struct btrfs_root *root = BTRFS_I(inode)->root;
6367 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6369 struct bio *orig_bio = dip->orig_bio;
6370 struct bio_vec *bvec = orig_bio->bi_io_vec;
6371 u64 start_sector = orig_bio->bi_sector;
6372 u64 file_offset = dip->logical_offset;
6377 int async_submit = 0;
6379 map_length = orig_bio->bi_size;
6380 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6381 &map_length, NULL, 0);
6387 if (map_length >= orig_bio->bi_size) {
6393 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6396 bio->bi_private = dip;
6397 bio->bi_end_io = btrfs_end_dio_bio;
6398 atomic_inc(&dip->pending_bios);
6400 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6401 if (unlikely(map_length < submit_len + bvec->bv_len ||
6402 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6403 bvec->bv_offset) < bvec->bv_len)) {
6405 * inc the count before we submit the bio so
6406 * we know the end IO handler won't happen before
6407 * we inc the count. Otherwise, the dip might get freed
6408 * before we're done setting it up
6410 atomic_inc(&dip->pending_bios);
6411 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6412 file_offset, skip_sum,
6416 atomic_dec(&dip->pending_bios);
6420 start_sector += submit_len >> 9;
6421 file_offset += submit_len;
6426 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6427 start_sector, GFP_NOFS);
6430 bio->bi_private = dip;
6431 bio->bi_end_io = btrfs_end_dio_bio;
6433 map_length = orig_bio->bi_size;
6434 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6435 &map_length, NULL, 0);
6441 submit_len += bvec->bv_len;
6448 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6457 * before atomic variable goto zero, we must
6458 * make sure dip->errors is perceived to be set.
6460 smp_mb__before_atomic_dec();
6461 if (atomic_dec_and_test(&dip->pending_bios))
6462 bio_io_error(dip->orig_bio);
6464 /* bio_end_io() will handle error, so we needn't return it */
6468 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6471 struct btrfs_root *root = BTRFS_I(inode)->root;
6472 struct btrfs_dio_private *dip;
6473 struct bio_vec *bvec = bio->bi_io_vec;
6475 int write = rw & REQ_WRITE;
6478 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6480 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6486 dip->private = bio->bi_private;
6488 dip->logical_offset = file_offset;
6492 dip->bytes += bvec->bv_len;
6494 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6496 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6497 bio->bi_private = dip;
6499 dip->orig_bio = bio;
6500 atomic_set(&dip->pending_bios, 0);
6503 bio->bi_end_io = btrfs_endio_direct_write;
6505 bio->bi_end_io = btrfs_endio_direct_read;
6507 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6512 * If this is a write, we need to clean up the reserved space and kill
6513 * the ordered extent.
6516 struct btrfs_ordered_extent *ordered;
6517 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6518 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6519 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6520 btrfs_free_reserved_extent(root, ordered->start,
6522 btrfs_put_ordered_extent(ordered);
6523 btrfs_put_ordered_extent(ordered);
6525 bio_endio(bio, ret);
6528 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6529 const struct iovec *iov, loff_t offset,
6530 unsigned long nr_segs)
6536 unsigned blocksize_mask = root->sectorsize - 1;
6537 ssize_t retval = -EINVAL;
6538 loff_t end = offset;
6540 if (offset & blocksize_mask)
6543 /* Check the memory alignment. Blocks cannot straddle pages */
6544 for (seg = 0; seg < nr_segs; seg++) {
6545 addr = (unsigned long)iov[seg].iov_base;
6546 size = iov[seg].iov_len;
6548 if ((addr & blocksize_mask) || (size & blocksize_mask))
6551 /* If this is a write we don't need to check anymore */
6556 * Check to make sure we don't have duplicate iov_base's in this
6557 * iovec, if so return EINVAL, otherwise we'll get csum errors
6558 * when reading back.
6560 for (i = seg + 1; i < nr_segs; i++) {
6561 if (iov[seg].iov_base == iov[i].iov_base)
6570 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6571 const struct iovec *iov, loff_t offset,
6572 unsigned long nr_segs)
6574 struct file *file = iocb->ki_filp;
6575 struct inode *inode = file->f_mapping->host;
6577 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6581 return __blockdev_direct_IO(rw, iocb, inode,
6582 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6583 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6584 btrfs_submit_direct, 0);
6587 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6588 __u64 start, __u64 len)
6590 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6593 int btrfs_readpage(struct file *file, struct page *page)
6595 struct extent_io_tree *tree;
6596 tree = &BTRFS_I(page->mapping->host)->io_tree;
6597 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6600 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6602 struct extent_io_tree *tree;
6605 if (current->flags & PF_MEMALLOC) {
6606 redirty_page_for_writepage(wbc, page);
6610 tree = &BTRFS_I(page->mapping->host)->io_tree;
6611 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6614 int btrfs_writepages(struct address_space *mapping,
6615 struct writeback_control *wbc)
6617 struct extent_io_tree *tree;
6619 tree = &BTRFS_I(mapping->host)->io_tree;
6620 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6624 btrfs_readpages(struct file *file, struct address_space *mapping,
6625 struct list_head *pages, unsigned nr_pages)
6627 struct extent_io_tree *tree;
6628 tree = &BTRFS_I(mapping->host)->io_tree;
6629 return extent_readpages(tree, mapping, pages, nr_pages,
6632 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6634 struct extent_io_tree *tree;
6635 struct extent_map_tree *map;
6638 tree = &BTRFS_I(page->mapping->host)->io_tree;
6639 map = &BTRFS_I(page->mapping->host)->extent_tree;
6640 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6642 ClearPagePrivate(page);
6643 set_page_private(page, 0);
6644 page_cache_release(page);
6649 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6651 if (PageWriteback(page) || PageDirty(page))
6653 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6656 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6658 struct inode *inode = page->mapping->host;
6659 struct extent_io_tree *tree;
6660 struct btrfs_ordered_extent *ordered;
6661 struct extent_state *cached_state = NULL;
6662 u64 page_start = page_offset(page);
6663 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6666 * we have the page locked, so new writeback can't start,
6667 * and the dirty bit won't be cleared while we are here.
6669 * Wait for IO on this page so that we can safely clear
6670 * the PagePrivate2 bit and do ordered accounting
6672 wait_on_page_writeback(page);
6674 tree = &BTRFS_I(inode)->io_tree;
6676 btrfs_releasepage(page, GFP_NOFS);
6679 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6680 ordered = btrfs_lookup_ordered_extent(inode,
6684 * IO on this page will never be started, so we need
6685 * to account for any ordered extents now
6687 clear_extent_bit(tree, page_start, page_end,
6688 EXTENT_DIRTY | EXTENT_DELALLOC |
6689 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
6690 EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
6692 * whoever cleared the private bit is responsible
6693 * for the finish_ordered_io
6695 if (TestClearPagePrivate2(page) &&
6696 btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6697 PAGE_CACHE_SIZE, 1)) {
6698 btrfs_finish_ordered_io(ordered);
6700 btrfs_put_ordered_extent(ordered);
6701 cached_state = NULL;
6702 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6704 clear_extent_bit(tree, page_start, page_end,
6705 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6706 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
6707 &cached_state, GFP_NOFS);
6708 __btrfs_releasepage(page, GFP_NOFS);
6710 ClearPageChecked(page);
6711 if (PagePrivate(page)) {
6712 ClearPagePrivate(page);
6713 set_page_private(page, 0);
6714 page_cache_release(page);
6719 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6720 * called from a page fault handler when a page is first dirtied. Hence we must
6721 * be careful to check for EOF conditions here. We set the page up correctly
6722 * for a written page which means we get ENOSPC checking when writing into
6723 * holes and correct delalloc and unwritten extent mapping on filesystems that
6724 * support these features.
6726 * We are not allowed to take the i_mutex here so we have to play games to
6727 * protect against truncate races as the page could now be beyond EOF. Because
6728 * vmtruncate() writes the inode size before removing pages, once we have the
6729 * page lock we can determine safely if the page is beyond EOF. If it is not
6730 * beyond EOF, then the page is guaranteed safe against truncation until we
6733 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6735 struct page *page = vmf->page;
6736 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6737 struct btrfs_root *root = BTRFS_I(inode)->root;
6738 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6739 struct btrfs_ordered_extent *ordered;
6740 struct extent_state *cached_state = NULL;
6742 unsigned long zero_start;
6749 sb_start_pagefault(inode->i_sb);
6750 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6752 ret = file_update_time(vma->vm_file);
6758 else /* -ENOSPC, -EIO, etc */
6759 ret = VM_FAULT_SIGBUS;
6765 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6768 size = i_size_read(inode);
6769 page_start = page_offset(page);
6770 page_end = page_start + PAGE_CACHE_SIZE - 1;
6772 if ((page->mapping != inode->i_mapping) ||
6773 (page_start >= size)) {
6774 /* page got truncated out from underneath us */
6777 wait_on_page_writeback(page);
6779 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6780 set_page_extent_mapped(page);
6783 * we can't set the delalloc bits if there are pending ordered
6784 * extents. Drop our locks and wait for them to finish
6786 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6788 unlock_extent_cached(io_tree, page_start, page_end,
6789 &cached_state, GFP_NOFS);
6791 btrfs_start_ordered_extent(inode, ordered, 1);
6792 btrfs_put_ordered_extent(ordered);
6797 * XXX - page_mkwrite gets called every time the page is dirtied, even
6798 * if it was already dirty, so for space accounting reasons we need to
6799 * clear any delalloc bits for the range we are fixing to save. There
6800 * is probably a better way to do this, but for now keep consistent with
6801 * prepare_pages in the normal write path.
6803 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6804 EXTENT_DIRTY | EXTENT_DELALLOC |
6805 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
6806 0, 0, &cached_state, GFP_NOFS);
6808 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6811 unlock_extent_cached(io_tree, page_start, page_end,
6812 &cached_state, GFP_NOFS);
6813 ret = VM_FAULT_SIGBUS;
6818 /* page is wholly or partially inside EOF */
6819 if (page_start + PAGE_CACHE_SIZE > size)
6820 zero_start = size & ~PAGE_CACHE_MASK;
6822 zero_start = PAGE_CACHE_SIZE;
6824 if (zero_start != PAGE_CACHE_SIZE) {
6826 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6827 flush_dcache_page(page);
6830 ClearPageChecked(page);
6831 set_page_dirty(page);
6832 SetPageUptodate(page);
6834 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6835 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6836 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
6838 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6842 sb_end_pagefault(inode->i_sb);
6843 return VM_FAULT_LOCKED;
6847 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6849 sb_end_pagefault(inode->i_sb);
6853 static int btrfs_truncate(struct inode *inode)
6855 struct btrfs_root *root = BTRFS_I(inode)->root;
6856 struct btrfs_block_rsv *rsv;
6859 struct btrfs_trans_handle *trans;
6860 u64 mask = root->sectorsize - 1;
6861 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6863 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
6867 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6868 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6871 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6872 * 3 things going on here
6874 * 1) We need to reserve space for our orphan item and the space to
6875 * delete our orphan item. Lord knows we don't want to have a dangling
6876 * orphan item because we didn't reserve space to remove it.
6878 * 2) We need to reserve space to update our inode.
6880 * 3) We need to have something to cache all the space that is going to
6881 * be free'd up by the truncate operation, but also have some slack
6882 * space reserved in case it uses space during the truncate (thank you
6883 * very much snapshotting).
6885 * And we need these to all be seperate. The fact is we can use alot of
6886 * space doing the truncate, and we have no earthly idea how much space
6887 * we will use, so we need the truncate reservation to be seperate so it
6888 * doesn't end up using space reserved for updating the inode or
6889 * removing the orphan item. We also need to be able to stop the
6890 * transaction and start a new one, which means we need to be able to
6891 * update the inode several times, and we have no idea of knowing how
6892 * many times that will be, so we can't just reserve 1 item for the
6893 * entirety of the opration, so that has to be done seperately as well.
6894 * Then there is the orphan item, which does indeed need to be held on
6895 * to for the whole operation, and we need nobody to touch this reserved
6896 * space except the orphan code.
6898 * So that leaves us with
6900 * 1) root->orphan_block_rsv - for the orphan deletion.
6901 * 2) rsv - for the truncate reservation, which we will steal from the
6902 * transaction reservation.
6903 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6904 * updating the inode.
6906 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
6909 rsv->size = min_size;
6913 * 1 for the truncate slack space
6914 * 1 for the orphan item we're going to add
6915 * 1 for the orphan item deletion
6916 * 1 for updating the inode.
6918 trans = btrfs_start_transaction(root, 4);
6919 if (IS_ERR(trans)) {
6920 err = PTR_ERR(trans);
6924 /* Migrate the slack space for the truncate to our reserve */
6925 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6929 ret = btrfs_orphan_add(trans, inode);
6931 btrfs_end_transaction(trans, root);
6936 * setattr is responsible for setting the ordered_data_close flag,
6937 * but that is only tested during the last file release. That
6938 * could happen well after the next commit, leaving a great big
6939 * window where new writes may get lost if someone chooses to write
6940 * to this file after truncating to zero
6942 * The inode doesn't have any dirty data here, and so if we commit
6943 * this is a noop. If someone immediately starts writing to the inode
6944 * it is very likely we'll catch some of their writes in this
6945 * transaction, and the commit will find this file on the ordered
6946 * data list with good things to send down.
6948 * This is a best effort solution, there is still a window where
6949 * using truncate to replace the contents of the file will
6950 * end up with a zero length file after a crash.
6952 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
6953 &BTRFS_I(inode)->runtime_flags))
6954 btrfs_add_ordered_operation(trans, root, inode);
6957 * So if we truncate and then write and fsync we normally would just
6958 * write the extents that changed, which is a problem if we need to
6959 * first truncate that entire inode. So set this flag so we write out
6960 * all of the extents in the inode to the sync log so we're completely
6963 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6964 trans->block_rsv = rsv;
6967 ret = btrfs_truncate_inode_items(trans, root, inode,
6969 BTRFS_EXTENT_DATA_KEY);
6970 if (ret != -ENOSPC) {
6975 trans->block_rsv = &root->fs_info->trans_block_rsv;
6976 ret = btrfs_update_inode(trans, root, inode);
6982 btrfs_end_transaction(trans, root);
6983 btrfs_btree_balance_dirty(root);
6985 trans = btrfs_start_transaction(root, 2);
6986 if (IS_ERR(trans)) {
6987 ret = err = PTR_ERR(trans);
6992 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
6994 BUG_ON(ret); /* shouldn't happen */
6995 trans->block_rsv = rsv;
6998 if (ret == 0 && inode->i_nlink > 0) {
6999 trans->block_rsv = root->orphan_block_rsv;
7000 ret = btrfs_orphan_del(trans, inode);
7003 } else if (ret && inode->i_nlink > 0) {
7005 * Failed to do the truncate, remove us from the in memory
7008 ret = btrfs_orphan_del(NULL, inode);
7012 trans->block_rsv = &root->fs_info->trans_block_rsv;
7013 ret = btrfs_update_inode(trans, root, inode);
7017 ret = btrfs_end_transaction(trans, root);
7018 btrfs_btree_balance_dirty(root);
7022 btrfs_free_block_rsv(root, rsv);
7031 * create a new subvolume directory/inode (helper for the ioctl).
7033 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7034 struct btrfs_root *new_root, u64 new_dirid)
7036 struct inode *inode;
7040 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7041 new_dirid, new_dirid,
7042 S_IFDIR | (~current_umask() & S_IRWXUGO),
7045 return PTR_ERR(inode);
7046 inode->i_op = &btrfs_dir_inode_operations;
7047 inode->i_fop = &btrfs_dir_file_operations;
7049 set_nlink(inode, 1);
7050 btrfs_i_size_write(inode, 0);
7052 err = btrfs_update_inode(trans, new_root, inode);
7058 struct inode *btrfs_alloc_inode(struct super_block *sb)
7060 struct btrfs_inode *ei;
7061 struct inode *inode;
7063 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7070 ei->last_sub_trans = 0;
7071 ei->logged_trans = 0;
7072 ei->delalloc_bytes = 0;
7073 ei->disk_i_size = 0;
7076 ei->index_cnt = (u64)-1;
7077 ei->last_unlink_trans = 0;
7078 ei->last_log_commit = 0;
7080 spin_lock_init(&ei->lock);
7081 ei->outstanding_extents = 0;
7082 ei->reserved_extents = 0;
7084 ei->runtime_flags = 0;
7085 ei->force_compress = BTRFS_COMPRESS_NONE;
7087 ei->delayed_node = NULL;
7089 inode = &ei->vfs_inode;
7090 extent_map_tree_init(&ei->extent_tree);
7091 extent_io_tree_init(&ei->io_tree, &inode->i_data);
7092 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7093 ei->io_tree.track_uptodate = 1;
7094 ei->io_failure_tree.track_uptodate = 1;
7095 mutex_init(&ei->log_mutex);
7096 mutex_init(&ei->delalloc_mutex);
7097 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7098 INIT_LIST_HEAD(&ei->delalloc_inodes);
7099 INIT_LIST_HEAD(&ei->ordered_operations);
7100 RB_CLEAR_NODE(&ei->rb_node);
7105 static void btrfs_i_callback(struct rcu_head *head)
7107 struct inode *inode = container_of(head, struct inode, i_rcu);
7108 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7111 void btrfs_destroy_inode(struct inode *inode)
7113 struct btrfs_ordered_extent *ordered;
7114 struct btrfs_root *root = BTRFS_I(inode)->root;
7116 WARN_ON(!hlist_empty(&inode->i_dentry));
7117 WARN_ON(inode->i_data.nrpages);
7118 WARN_ON(BTRFS_I(inode)->outstanding_extents);
7119 WARN_ON(BTRFS_I(inode)->reserved_extents);
7120 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7121 WARN_ON(BTRFS_I(inode)->csum_bytes);
7124 * This can happen where we create an inode, but somebody else also
7125 * created the same inode and we need to destroy the one we already
7132 * Make sure we're properly removed from the ordered operation
7136 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7137 spin_lock(&root->fs_info->ordered_extent_lock);
7138 list_del_init(&BTRFS_I(inode)->ordered_operations);
7139 spin_unlock(&root->fs_info->ordered_extent_lock);
7142 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7143 &BTRFS_I(inode)->runtime_flags)) {
7144 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7145 (unsigned long long)btrfs_ino(inode));
7146 atomic_dec(&root->orphan_inodes);
7150 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7154 printk(KERN_ERR "btrfs found ordered "
7155 "extent %llu %llu on inode cleanup\n",
7156 (unsigned long long)ordered->file_offset,
7157 (unsigned long long)ordered->len);
7158 btrfs_remove_ordered_extent(inode, ordered);
7159 btrfs_put_ordered_extent(ordered);
7160 btrfs_put_ordered_extent(ordered);
7163 inode_tree_del(inode);
7164 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7166 btrfs_remove_delayed_node(inode);
7167 call_rcu(&inode->i_rcu, btrfs_i_callback);
7170 int btrfs_drop_inode(struct inode *inode)
7172 struct btrfs_root *root = BTRFS_I(inode)->root;
7174 if (btrfs_root_refs(&root->root_item) == 0 &&
7175 !btrfs_is_free_space_inode(inode))
7178 return generic_drop_inode(inode);
7181 static void init_once(void *foo)
7183 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7185 inode_init_once(&ei->vfs_inode);
7188 void btrfs_destroy_cachep(void)
7191 * Make sure all delayed rcu free inodes are flushed before we
7195 if (btrfs_inode_cachep)
7196 kmem_cache_destroy(btrfs_inode_cachep);
7197 if (btrfs_trans_handle_cachep)
7198 kmem_cache_destroy(btrfs_trans_handle_cachep);
7199 if (btrfs_transaction_cachep)
7200 kmem_cache_destroy(btrfs_transaction_cachep);
7201 if (btrfs_path_cachep)
7202 kmem_cache_destroy(btrfs_path_cachep);
7203 if (btrfs_free_space_cachep)
7204 kmem_cache_destroy(btrfs_free_space_cachep);
7205 if (btrfs_delalloc_work_cachep)
7206 kmem_cache_destroy(btrfs_delalloc_work_cachep);
7209 int btrfs_init_cachep(void)
7211 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
7212 sizeof(struct btrfs_inode), 0,
7213 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7214 if (!btrfs_inode_cachep)
7217 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
7218 sizeof(struct btrfs_trans_handle), 0,
7219 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7220 if (!btrfs_trans_handle_cachep)
7223 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
7224 sizeof(struct btrfs_transaction), 0,
7225 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7226 if (!btrfs_transaction_cachep)
7229 btrfs_path_cachep = kmem_cache_create("btrfs_path",
7230 sizeof(struct btrfs_path), 0,
7231 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7232 if (!btrfs_path_cachep)
7235 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
7236 sizeof(struct btrfs_free_space), 0,
7237 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7238 if (!btrfs_free_space_cachep)
7241 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
7242 sizeof(struct btrfs_delalloc_work), 0,
7243 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
7245 if (!btrfs_delalloc_work_cachep)
7250 btrfs_destroy_cachep();
7254 static int btrfs_getattr(struct vfsmount *mnt,
7255 struct dentry *dentry, struct kstat *stat)
7257 struct inode *inode = dentry->d_inode;
7258 u32 blocksize = inode->i_sb->s_blocksize;
7260 generic_fillattr(inode, stat);
7261 stat->dev = BTRFS_I(inode)->root->anon_dev;
7262 stat->blksize = PAGE_CACHE_SIZE;
7263 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7264 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
7269 * If a file is moved, it will inherit the cow and compression flags of the new
7272 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7274 struct btrfs_inode *b_dir = BTRFS_I(dir);
7275 struct btrfs_inode *b_inode = BTRFS_I(inode);
7277 if (b_dir->flags & BTRFS_INODE_NODATACOW)
7278 b_inode->flags |= BTRFS_INODE_NODATACOW;
7280 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7282 if (b_dir->flags & BTRFS_INODE_COMPRESS) {
7283 b_inode->flags |= BTRFS_INODE_COMPRESS;
7284 b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
7286 b_inode->flags &= ~(BTRFS_INODE_COMPRESS |
7287 BTRFS_INODE_NOCOMPRESS);
7291 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7292 struct inode *new_dir, struct dentry *new_dentry)
7294 struct btrfs_trans_handle *trans;
7295 struct btrfs_root *root = BTRFS_I(old_dir)->root;
7296 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7297 struct inode *new_inode = new_dentry->d_inode;
7298 struct inode *old_inode = old_dentry->d_inode;
7299 struct timespec ctime = CURRENT_TIME;
7303 u64 old_ino = btrfs_ino(old_inode);
7305 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7308 /* we only allow rename subvolume link between subvolumes */
7309 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7312 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7313 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7316 if (S_ISDIR(old_inode->i_mode) && new_inode &&
7317 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7320 * we're using rename to replace one file with another.
7321 * and the replacement file is large. Start IO on it now so
7322 * we don't add too much work to the end of the transaction
7324 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7325 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7326 filemap_flush(old_inode->i_mapping);
7328 /* close the racy window with snapshot create/destroy ioctl */
7329 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7330 down_read(&root->fs_info->subvol_sem);
7332 * We want to reserve the absolute worst case amount of items. So if
7333 * both inodes are subvols and we need to unlink them then that would
7334 * require 4 item modifications, but if they are both normal inodes it
7335 * would require 5 item modifications, so we'll assume their normal
7336 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7337 * should cover the worst case number of items we'll modify.
7339 trans = btrfs_start_transaction(root, 20);
7340 if (IS_ERR(trans)) {
7341 ret = PTR_ERR(trans);
7346 btrfs_record_root_in_trans(trans, dest);
7348 ret = btrfs_set_inode_index(new_dir, &index);
7352 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7353 /* force full log commit if subvolume involved. */
7354 root->fs_info->last_trans_log_full_commit = trans->transid;
7356 ret = btrfs_insert_inode_ref(trans, dest,
7357 new_dentry->d_name.name,
7358 new_dentry->d_name.len,
7360 btrfs_ino(new_dir), index);
7364 * this is an ugly little race, but the rename is required
7365 * to make sure that if we crash, the inode is either at the
7366 * old name or the new one. pinning the log transaction lets
7367 * us make sure we don't allow a log commit to come in after
7368 * we unlink the name but before we add the new name back in.
7370 btrfs_pin_log_trans(root);
7373 * make sure the inode gets flushed if it is replacing
7376 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7377 btrfs_add_ordered_operation(trans, root, old_inode);
7379 inode_inc_iversion(old_dir);
7380 inode_inc_iversion(new_dir);
7381 inode_inc_iversion(old_inode);
7382 old_dir->i_ctime = old_dir->i_mtime = ctime;
7383 new_dir->i_ctime = new_dir->i_mtime = ctime;
7384 old_inode->i_ctime = ctime;
7386 if (old_dentry->d_parent != new_dentry->d_parent)
7387 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7389 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7390 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7391 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7392 old_dentry->d_name.name,
7393 old_dentry->d_name.len);
7395 ret = __btrfs_unlink_inode(trans, root, old_dir,
7396 old_dentry->d_inode,
7397 old_dentry->d_name.name,
7398 old_dentry->d_name.len);
7400 ret = btrfs_update_inode(trans, root, old_inode);
7403 btrfs_abort_transaction(trans, root, ret);
7408 inode_inc_iversion(new_inode);
7409 new_inode->i_ctime = CURRENT_TIME;
7410 if (unlikely(btrfs_ino(new_inode) ==
7411 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7412 root_objectid = BTRFS_I(new_inode)->location.objectid;
7413 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7415 new_dentry->d_name.name,
7416 new_dentry->d_name.len);
7417 BUG_ON(new_inode->i_nlink == 0);
7419 ret = btrfs_unlink_inode(trans, dest, new_dir,
7420 new_dentry->d_inode,
7421 new_dentry->d_name.name,
7422 new_dentry->d_name.len);
7424 if (!ret && new_inode->i_nlink == 0) {
7425 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7429 btrfs_abort_transaction(trans, root, ret);
7434 fixup_inode_flags(new_dir, old_inode);
7436 ret = btrfs_add_link(trans, new_dir, old_inode,
7437 new_dentry->d_name.name,
7438 new_dentry->d_name.len, 0, index);
7440 btrfs_abort_transaction(trans, root, ret);
7444 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7445 struct dentry *parent = new_dentry->d_parent;
7446 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7447 btrfs_end_log_trans(root);
7450 btrfs_end_transaction(trans, root);
7452 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7453 up_read(&root->fs_info->subvol_sem);
7458 static void btrfs_run_delalloc_work(struct btrfs_work *work)
7460 struct btrfs_delalloc_work *delalloc_work;
7462 delalloc_work = container_of(work, struct btrfs_delalloc_work,
7464 if (delalloc_work->wait)
7465 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
7467 filemap_flush(delalloc_work->inode->i_mapping);
7469 if (delalloc_work->delay_iput)
7470 btrfs_add_delayed_iput(delalloc_work->inode);
7472 iput(delalloc_work->inode);
7473 complete(&delalloc_work->completion);
7476 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
7477 int wait, int delay_iput)
7479 struct btrfs_delalloc_work *work;
7481 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
7485 init_completion(&work->completion);
7486 INIT_LIST_HEAD(&work->list);
7487 work->inode = inode;
7489 work->delay_iput = delay_iput;
7490 work->work.func = btrfs_run_delalloc_work;
7495 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
7497 wait_for_completion(&work->completion);
7498 kmem_cache_free(btrfs_delalloc_work_cachep, work);
7502 * some fairly slow code that needs optimization. This walks the list
7503 * of all the inodes with pending delalloc and forces them to disk.
7505 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7507 struct list_head *head = &root->fs_info->delalloc_inodes;
7508 struct btrfs_inode *binode;
7509 struct inode *inode;
7510 struct btrfs_delalloc_work *work, *next;
7511 struct list_head works;
7514 if (root->fs_info->sb->s_flags & MS_RDONLY)
7517 INIT_LIST_HEAD(&works);
7519 spin_lock(&root->fs_info->delalloc_lock);
7520 while (!list_empty(head)) {
7521 binode = list_entry(head->next, struct btrfs_inode,
7523 inode = igrab(&binode->vfs_inode);
7525 list_del_init(&binode->delalloc_inodes);
7526 spin_unlock(&root->fs_info->delalloc_lock);
7528 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
7533 list_add_tail(&work->list, &works);
7534 btrfs_queue_worker(&root->fs_info->flush_workers,
7538 spin_lock(&root->fs_info->delalloc_lock);
7540 spin_unlock(&root->fs_info->delalloc_lock);
7542 /* the filemap_flush will queue IO into the worker threads, but
7543 * we have to make sure the IO is actually started and that
7544 * ordered extents get created before we return
7546 atomic_inc(&root->fs_info->async_submit_draining);
7547 while (atomic_read(&root->fs_info->nr_async_submits) ||
7548 atomic_read(&root->fs_info->async_delalloc_pages)) {
7549 wait_event(root->fs_info->async_submit_wait,
7550 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7551 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7553 atomic_dec(&root->fs_info->async_submit_draining);
7555 list_for_each_entry_safe(work, next, &works, list) {
7556 list_del_init(&work->list);
7557 btrfs_wait_and_free_delalloc_work(work);
7562 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7563 const char *symname)
7565 struct btrfs_trans_handle *trans;
7566 struct btrfs_root *root = BTRFS_I(dir)->root;
7567 struct btrfs_path *path;
7568 struct btrfs_key key;
7569 struct inode *inode = NULL;
7577 struct btrfs_file_extent_item *ei;
7578 struct extent_buffer *leaf;
7580 name_len = strlen(symname) + 1;
7581 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7582 return -ENAMETOOLONG;
7585 * 2 items for inode item and ref
7586 * 2 items for dir items
7587 * 1 item for xattr if selinux is on
7589 trans = btrfs_start_transaction(root, 5);
7591 return PTR_ERR(trans);
7593 err = btrfs_find_free_ino(root, &objectid);
7597 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7598 dentry->d_name.len, btrfs_ino(dir), objectid,
7599 S_IFLNK|S_IRWXUGO, &index);
7600 if (IS_ERR(inode)) {
7601 err = PTR_ERR(inode);
7605 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7612 * If the active LSM wants to access the inode during
7613 * d_instantiate it needs these. Smack checks to see
7614 * if the filesystem supports xattrs by looking at the
7617 inode->i_fop = &btrfs_file_operations;
7618 inode->i_op = &btrfs_file_inode_operations;
7620 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7624 inode->i_mapping->a_ops = &btrfs_aops;
7625 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7626 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7631 path = btrfs_alloc_path();
7637 key.objectid = btrfs_ino(inode);
7639 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7640 datasize = btrfs_file_extent_calc_inline_size(name_len);
7641 err = btrfs_insert_empty_item(trans, root, path, &key,
7645 btrfs_free_path(path);
7648 leaf = path->nodes[0];
7649 ei = btrfs_item_ptr(leaf, path->slots[0],
7650 struct btrfs_file_extent_item);
7651 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7652 btrfs_set_file_extent_type(leaf, ei,
7653 BTRFS_FILE_EXTENT_INLINE);
7654 btrfs_set_file_extent_encryption(leaf, ei, 0);
7655 btrfs_set_file_extent_compression(leaf, ei, 0);
7656 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7657 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7659 ptr = btrfs_file_extent_inline_start(ei);
7660 write_extent_buffer(leaf, symname, ptr, name_len);
7661 btrfs_mark_buffer_dirty(leaf);
7662 btrfs_free_path(path);
7664 inode->i_op = &btrfs_symlink_inode_operations;
7665 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7666 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7667 inode_set_bytes(inode, name_len);
7668 btrfs_i_size_write(inode, name_len - 1);
7669 err = btrfs_update_inode(trans, root, inode);
7675 d_instantiate(dentry, inode);
7676 btrfs_end_transaction(trans, root);
7678 inode_dec_link_count(inode);
7681 btrfs_btree_balance_dirty(root);
7685 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7686 u64 start, u64 num_bytes, u64 min_size,
7687 loff_t actual_len, u64 *alloc_hint,
7688 struct btrfs_trans_handle *trans)
7690 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7691 struct extent_map *em;
7692 struct btrfs_root *root = BTRFS_I(inode)->root;
7693 struct btrfs_key ins;
7694 u64 cur_offset = start;
7697 bool own_trans = true;
7701 while (num_bytes > 0) {
7703 trans = btrfs_start_transaction(root, 3);
7704 if (IS_ERR(trans)) {
7705 ret = PTR_ERR(trans);
7710 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7711 0, *alloc_hint, &ins, 1);
7714 btrfs_end_transaction(trans, root);
7718 ret = insert_reserved_file_extent(trans, inode,
7719 cur_offset, ins.objectid,
7720 ins.offset, ins.offset,
7721 ins.offset, 0, 0, 0,
7722 BTRFS_FILE_EXTENT_PREALLOC);
7724 btrfs_abort_transaction(trans, root, ret);
7726 btrfs_end_transaction(trans, root);
7729 btrfs_drop_extent_cache(inode, cur_offset,
7730 cur_offset + ins.offset -1, 0);
7732 em = alloc_extent_map();
7734 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
7735 &BTRFS_I(inode)->runtime_flags);
7739 em->start = cur_offset;
7740 em->orig_start = cur_offset;
7741 em->len = ins.offset;
7742 em->block_start = ins.objectid;
7743 em->block_len = ins.offset;
7744 em->bdev = root->fs_info->fs_devices->latest_bdev;
7745 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7746 em->generation = trans->transid;
7749 write_lock(&em_tree->lock);
7750 ret = add_extent_mapping(em_tree, em);
7752 list_move(&em->list,
7753 &em_tree->modified_extents);
7754 write_unlock(&em_tree->lock);
7757 btrfs_drop_extent_cache(inode, cur_offset,
7758 cur_offset + ins.offset - 1,
7761 free_extent_map(em);
7763 num_bytes -= ins.offset;
7764 cur_offset += ins.offset;
7765 *alloc_hint = ins.objectid + ins.offset;
7767 inode_inc_iversion(inode);
7768 inode->i_ctime = CURRENT_TIME;
7769 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7770 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7771 (actual_len > inode->i_size) &&
7772 (cur_offset > inode->i_size)) {
7773 if (cur_offset > actual_len)
7774 i_size = actual_len;
7776 i_size = cur_offset;
7777 i_size_write(inode, i_size);
7778 btrfs_ordered_update_i_size(inode, i_size, NULL);
7781 ret = btrfs_update_inode(trans, root, inode);
7784 btrfs_abort_transaction(trans, root, ret);
7786 btrfs_end_transaction(trans, root);
7791 btrfs_end_transaction(trans, root);
7796 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7797 u64 start, u64 num_bytes, u64 min_size,
7798 loff_t actual_len, u64 *alloc_hint)
7800 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7801 min_size, actual_len, alloc_hint,
7805 int btrfs_prealloc_file_range_trans(struct inode *inode,
7806 struct btrfs_trans_handle *trans, int mode,
7807 u64 start, u64 num_bytes, u64 min_size,
7808 loff_t actual_len, u64 *alloc_hint)
7810 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7811 min_size, actual_len, alloc_hint, trans);
7814 static int btrfs_set_page_dirty(struct page *page)
7816 return __set_page_dirty_nobuffers(page);
7819 static int btrfs_permission(struct inode *inode, int mask)
7821 struct btrfs_root *root = BTRFS_I(inode)->root;
7822 umode_t mode = inode->i_mode;
7824 if (mask & MAY_WRITE &&
7825 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7826 if (btrfs_root_readonly(root))
7828 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7831 return generic_permission(inode, mask);
7834 static const struct inode_operations btrfs_dir_inode_operations = {
7835 .getattr = btrfs_getattr,
7836 .lookup = btrfs_lookup,
7837 .create = btrfs_create,
7838 .unlink = btrfs_unlink,
7840 .mkdir = btrfs_mkdir,
7841 .rmdir = btrfs_rmdir,
7842 .rename = btrfs_rename,
7843 .symlink = btrfs_symlink,
7844 .setattr = btrfs_setattr,
7845 .mknod = btrfs_mknod,
7846 .setxattr = btrfs_setxattr,
7847 .getxattr = btrfs_getxattr,
7848 .listxattr = btrfs_listxattr,
7849 .removexattr = btrfs_removexattr,
7850 .permission = btrfs_permission,
7851 .get_acl = btrfs_get_acl,
7853 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7854 .lookup = btrfs_lookup,
7855 .permission = btrfs_permission,
7856 .get_acl = btrfs_get_acl,
7859 static const struct file_operations btrfs_dir_file_operations = {
7860 .llseek = generic_file_llseek,
7861 .read = generic_read_dir,
7862 .readdir = btrfs_real_readdir,
7863 .unlocked_ioctl = btrfs_ioctl,
7864 #ifdef CONFIG_COMPAT
7865 .compat_ioctl = btrfs_ioctl,
7867 .release = btrfs_release_file,
7868 .fsync = btrfs_sync_file,
7871 static struct extent_io_ops btrfs_extent_io_ops = {
7872 .fill_delalloc = run_delalloc_range,
7873 .submit_bio_hook = btrfs_submit_bio_hook,
7874 .merge_bio_hook = btrfs_merge_bio_hook,
7875 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7876 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7877 .writepage_start_hook = btrfs_writepage_start_hook,
7878 .set_bit_hook = btrfs_set_bit_hook,
7879 .clear_bit_hook = btrfs_clear_bit_hook,
7880 .merge_extent_hook = btrfs_merge_extent_hook,
7881 .split_extent_hook = btrfs_split_extent_hook,
7885 * btrfs doesn't support the bmap operation because swapfiles
7886 * use bmap to make a mapping of extents in the file. They assume
7887 * these extents won't change over the life of the file and they
7888 * use the bmap result to do IO directly to the drive.
7890 * the btrfs bmap call would return logical addresses that aren't
7891 * suitable for IO and they also will change frequently as COW
7892 * operations happen. So, swapfile + btrfs == corruption.
7894 * For now we're avoiding this by dropping bmap.
7896 static const struct address_space_operations btrfs_aops = {
7897 .readpage = btrfs_readpage,
7898 .writepage = btrfs_writepage,
7899 .writepages = btrfs_writepages,
7900 .readpages = btrfs_readpages,
7901 .direct_IO = btrfs_direct_IO,
7902 .invalidatepage = btrfs_invalidatepage,
7903 .releasepage = btrfs_releasepage,
7904 .set_page_dirty = btrfs_set_page_dirty,
7905 .error_remove_page = generic_error_remove_page,
7908 static const struct address_space_operations btrfs_symlink_aops = {
7909 .readpage = btrfs_readpage,
7910 .writepage = btrfs_writepage,
7911 .invalidatepage = btrfs_invalidatepage,
7912 .releasepage = btrfs_releasepage,
7915 static const struct inode_operations btrfs_file_inode_operations = {
7916 .getattr = btrfs_getattr,
7917 .setattr = btrfs_setattr,
7918 .setxattr = btrfs_setxattr,
7919 .getxattr = btrfs_getxattr,
7920 .listxattr = btrfs_listxattr,
7921 .removexattr = btrfs_removexattr,
7922 .permission = btrfs_permission,
7923 .fiemap = btrfs_fiemap,
7924 .get_acl = btrfs_get_acl,
7925 .update_time = btrfs_update_time,
7927 static const struct inode_operations btrfs_special_inode_operations = {
7928 .getattr = btrfs_getattr,
7929 .setattr = btrfs_setattr,
7930 .permission = btrfs_permission,
7931 .setxattr = btrfs_setxattr,
7932 .getxattr = btrfs_getxattr,
7933 .listxattr = btrfs_listxattr,
7934 .removexattr = btrfs_removexattr,
7935 .get_acl = btrfs_get_acl,
7936 .update_time = btrfs_update_time,
7938 static const struct inode_operations btrfs_symlink_inode_operations = {
7939 .readlink = generic_readlink,
7940 .follow_link = page_follow_link_light,
7941 .put_link = page_put_link,
7942 .getattr = btrfs_getattr,
7943 .setattr = btrfs_setattr,
7944 .permission = btrfs_permission,
7945 .setxattr = btrfs_setxattr,
7946 .getxattr = btrfs_getxattr,
7947 .listxattr = btrfs_listxattr,
7948 .removexattr = btrfs_removexattr,
7949 .get_acl = btrfs_get_acl,
7950 .update_time = btrfs_update_time,
7953 const struct dentry_operations btrfs_dentry_operations = {
7954 .d_delete = btrfs_dentry_delete,
7955 .d_release = btrfs_dentry_release,