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>
44 #include "transaction.h"
45 #include "btrfs_inode.h"
47 #include "print-tree.h"
49 #include "ordered-data.h"
52 #include "compression.h"
54 #include "free-space-cache.h"
55 #include "inode-map.h"
57 struct btrfs_iget_args {
59 struct btrfs_root *root;
62 static const struct inode_operations btrfs_dir_inode_operations;
63 static const struct inode_operations btrfs_symlink_inode_operations;
64 static const struct inode_operations btrfs_dir_ro_inode_operations;
65 static const struct inode_operations btrfs_special_inode_operations;
66 static const struct inode_operations btrfs_file_inode_operations;
67 static const struct address_space_operations btrfs_aops;
68 static const struct address_space_operations btrfs_symlink_aops;
69 static const struct file_operations btrfs_dir_file_operations;
70 static struct extent_io_ops btrfs_extent_io_ops;
72 static struct kmem_cache *btrfs_inode_cachep;
73 struct kmem_cache *btrfs_trans_handle_cachep;
74 struct kmem_cache *btrfs_transaction_cachep;
75 struct kmem_cache *btrfs_path_cachep;
76 struct kmem_cache *btrfs_free_space_cachep;
79 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
80 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
81 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
82 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
83 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
84 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
85 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
86 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
89 static int btrfs_setsize(struct inode *inode, loff_t newsize);
90 static int btrfs_truncate(struct inode *inode);
91 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
92 static noinline int cow_file_range(struct inode *inode,
93 struct page *locked_page,
94 u64 start, u64 end, int *page_started,
95 unsigned long *nr_written, int unlock);
97 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
98 struct inode *inode, struct inode *dir,
99 const struct qstr *qstr)
103 err = btrfs_init_acl(trans, inode, dir);
105 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
110 * this does all the hard work for inserting an inline extent into
111 * the btree. The caller should have done a btrfs_drop_extents so that
112 * no overlapping inline items exist in the btree
114 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
115 struct btrfs_root *root, struct inode *inode,
116 u64 start, size_t size, size_t compressed_size,
118 struct page **compressed_pages)
120 struct btrfs_key key;
121 struct btrfs_path *path;
122 struct extent_buffer *leaf;
123 struct page *page = NULL;
126 struct btrfs_file_extent_item *ei;
129 size_t cur_size = size;
131 unsigned long offset;
133 if (compressed_size && compressed_pages)
134 cur_size = compressed_size;
136 path = btrfs_alloc_path();
140 path->leave_spinning = 1;
142 key.objectid = btrfs_ino(inode);
144 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
145 datasize = btrfs_file_extent_calc_inline_size(cur_size);
147 inode_add_bytes(inode, size);
148 ret = btrfs_insert_empty_item(trans, root, path, &key,
155 leaf = path->nodes[0];
156 ei = btrfs_item_ptr(leaf, path->slots[0],
157 struct btrfs_file_extent_item);
158 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
159 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
160 btrfs_set_file_extent_encryption(leaf, ei, 0);
161 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
162 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
163 ptr = btrfs_file_extent_inline_start(ei);
165 if (compress_type != BTRFS_COMPRESS_NONE) {
168 while (compressed_size > 0) {
169 cpage = compressed_pages[i];
170 cur_size = min_t(unsigned long, compressed_size,
173 kaddr = kmap_atomic(cpage, KM_USER0);
174 write_extent_buffer(leaf, kaddr, ptr, cur_size);
175 kunmap_atomic(kaddr, KM_USER0);
179 compressed_size -= cur_size;
181 btrfs_set_file_extent_compression(leaf, ei,
184 page = find_get_page(inode->i_mapping,
185 start >> PAGE_CACHE_SHIFT);
186 btrfs_set_file_extent_compression(leaf, ei, 0);
187 kaddr = kmap_atomic(page, KM_USER0);
188 offset = start & (PAGE_CACHE_SIZE - 1);
189 write_extent_buffer(leaf, kaddr + offset, ptr, size);
190 kunmap_atomic(kaddr, KM_USER0);
191 page_cache_release(page);
193 btrfs_mark_buffer_dirty(leaf);
194 btrfs_free_path(path);
197 * we're an inline extent, so nobody can
198 * extend the file past i_size without locking
199 * a page we already have locked.
201 * We must do any isize and inode updates
202 * before we unlock the pages. Otherwise we
203 * could end up racing with unlink.
205 BTRFS_I(inode)->disk_i_size = inode->i_size;
206 btrfs_update_inode(trans, root, inode);
210 btrfs_free_path(path);
216 * conditionally insert an inline extent into the file. This
217 * does the checks required to make sure the data is small enough
218 * to fit as an inline extent.
220 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
221 struct btrfs_root *root,
222 struct inode *inode, u64 start, u64 end,
223 size_t compressed_size, int compress_type,
224 struct page **compressed_pages)
226 u64 isize = i_size_read(inode);
227 u64 actual_end = min(end + 1, isize);
228 u64 inline_len = actual_end - start;
229 u64 aligned_end = (end + root->sectorsize - 1) &
230 ~((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, inode, start, aligned_end,
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);
258 btrfs_delalloc_release_metadata(inode, end + 1 - start);
259 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
263 struct async_extent {
268 unsigned long nr_pages;
270 struct list_head list;
275 struct btrfs_root *root;
276 struct page *locked_page;
279 struct list_head extents;
280 struct btrfs_work work;
283 static noinline int add_async_extent(struct async_cow *cow,
284 u64 start, u64 ram_size,
287 unsigned long nr_pages,
290 struct async_extent *async_extent;
292 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
293 BUG_ON(!async_extent);
294 async_extent->start = start;
295 async_extent->ram_size = ram_size;
296 async_extent->compressed_size = compressed_size;
297 async_extent->pages = pages;
298 async_extent->nr_pages = nr_pages;
299 async_extent->compress_type = compress_type;
300 list_add_tail(&async_extent->list, &cow->extents);
305 * we create compressed extents in two phases. The first
306 * phase compresses a range of pages that have already been
307 * locked (both pages and state bits are locked).
309 * This is done inside an ordered work queue, and the compression
310 * is spread across many cpus. The actual IO submission is step
311 * two, and the ordered work queue takes care of making sure that
312 * happens in the same order things were put onto the queue by
313 * writepages and friends.
315 * If this code finds it can't get good compression, it puts an
316 * entry onto the work queue to write the uncompressed bytes. This
317 * makes sure that both compressed inodes and uncompressed inodes
318 * are written in the same order that pdflush sent them down.
320 static noinline int compress_file_range(struct inode *inode,
321 struct page *locked_page,
323 struct async_cow *async_cow,
326 struct btrfs_root *root = BTRFS_I(inode)->root;
327 struct btrfs_trans_handle *trans;
329 u64 blocksize = root->sectorsize;
331 u64 isize = i_size_read(inode);
333 struct page **pages = NULL;
334 unsigned long nr_pages;
335 unsigned long nr_pages_ret = 0;
336 unsigned long total_compressed = 0;
337 unsigned long total_in = 0;
338 unsigned long max_compressed = 128 * 1024;
339 unsigned long max_uncompressed = 128 * 1024;
342 int compress_type = root->fs_info->compress_type;
344 /* if this is a small write inside eof, kick off a defragbot */
345 if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024)
346 btrfs_add_inode_defrag(NULL, inode);
348 actual_end = min_t(u64, isize, end + 1);
351 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
352 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
355 * we don't want to send crud past the end of i_size through
356 * compression, that's just a waste of CPU time. So, if the
357 * end of the file is before the start of our current
358 * requested range of bytes, we bail out to the uncompressed
359 * cleanup code that can deal with all of this.
361 * It isn't really the fastest way to fix things, but this is a
362 * very uncommon corner.
364 if (actual_end <= start)
365 goto cleanup_and_bail_uncompressed;
367 total_compressed = actual_end - start;
369 /* we want to make sure that amount of ram required to uncompress
370 * an extent is reasonable, so we limit the total size in ram
371 * of a compressed extent to 128k. This is a crucial number
372 * because it also controls how easily we can spread reads across
373 * cpus for decompression.
375 * We also want to make sure the amount of IO required to do
376 * a random read is reasonably small, so we limit the size of
377 * a compressed extent to 128k.
379 total_compressed = min(total_compressed, max_uncompressed);
380 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
381 num_bytes = max(blocksize, num_bytes);
386 * we do compression for mount -o compress and when the
387 * inode has not been flagged as nocompress. This flag can
388 * change at any time if we discover bad compression ratios.
390 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
391 (btrfs_test_opt(root, COMPRESS) ||
392 (BTRFS_I(inode)->force_compress) ||
393 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
395 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
398 if (BTRFS_I(inode)->force_compress)
399 compress_type = BTRFS_I(inode)->force_compress;
401 ret = btrfs_compress_pages(compress_type,
402 inode->i_mapping, start,
403 total_compressed, pages,
404 nr_pages, &nr_pages_ret,
410 unsigned long offset = total_compressed &
411 (PAGE_CACHE_SIZE - 1);
412 struct page *page = pages[nr_pages_ret - 1];
415 /* zero the tail end of the last page, we might be
416 * sending it down to disk
419 kaddr = kmap_atomic(page, KM_USER0);
420 memset(kaddr + offset, 0,
421 PAGE_CACHE_SIZE - offset);
422 kunmap_atomic(kaddr, KM_USER0);
428 trans = btrfs_join_transaction(root);
429 BUG_ON(IS_ERR(trans));
430 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
432 /* lets try to make an inline extent */
433 if (ret || total_in < (actual_end - start)) {
434 /* we didn't compress the entire range, try
435 * to make an uncompressed inline extent.
437 ret = cow_file_range_inline(trans, root, inode,
438 start, end, 0, 0, NULL);
440 /* try making a compressed inline extent */
441 ret = cow_file_range_inline(trans, root, inode,
444 compress_type, pages);
448 * inline extent creation worked, we don't need
449 * to create any more async work items. Unlock
450 * and free up our temp pages.
452 extent_clear_unlock_delalloc(inode,
453 &BTRFS_I(inode)->io_tree,
455 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
456 EXTENT_CLEAR_DELALLOC |
457 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
459 btrfs_end_transaction(trans, root);
462 btrfs_end_transaction(trans, root);
467 * we aren't doing an inline extent round the compressed size
468 * up to a block size boundary so the allocator does sane
471 total_compressed = (total_compressed + blocksize - 1) &
475 * one last check to make sure the compression is really a
476 * win, compare the page count read with the blocks on disk
478 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
479 ~(PAGE_CACHE_SIZE - 1);
480 if (total_compressed >= total_in) {
483 num_bytes = total_in;
486 if (!will_compress && pages) {
488 * the compression code ran but failed to make things smaller,
489 * free any pages it allocated and our page pointer array
491 for (i = 0; i < nr_pages_ret; i++) {
492 WARN_ON(pages[i]->mapping);
493 page_cache_release(pages[i]);
497 total_compressed = 0;
500 /* flag the file so we don't compress in the future */
501 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
502 !(BTRFS_I(inode)->force_compress)) {
503 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
509 /* the async work queues will take care of doing actual
510 * allocation on disk for these compressed pages,
511 * and will submit them to the elevator.
513 add_async_extent(async_cow, start, num_bytes,
514 total_compressed, pages, nr_pages_ret,
517 if (start + num_bytes < end) {
524 cleanup_and_bail_uncompressed:
526 * No compression, but we still need to write the pages in
527 * the file we've been given so far. redirty the locked
528 * page if it corresponds to our extent and set things up
529 * for the async work queue to run cow_file_range to do
530 * the normal delalloc dance
532 if (page_offset(locked_page) >= start &&
533 page_offset(locked_page) <= end) {
534 __set_page_dirty_nobuffers(locked_page);
535 /* unlocked later on in the async handlers */
537 add_async_extent(async_cow, start, end - start + 1,
538 0, NULL, 0, BTRFS_COMPRESS_NONE);
546 for (i = 0; i < nr_pages_ret; i++) {
547 WARN_ON(pages[i]->mapping);
548 page_cache_release(pages[i]);
556 * phase two of compressed writeback. This is the ordered portion
557 * of the code, which only gets called in the order the work was
558 * queued. We walk all the async extents created by compress_file_range
559 * and send them down to the disk.
561 static noinline int submit_compressed_extents(struct inode *inode,
562 struct async_cow *async_cow)
564 struct async_extent *async_extent;
566 struct btrfs_trans_handle *trans;
567 struct btrfs_key ins;
568 struct extent_map *em;
569 struct btrfs_root *root = BTRFS_I(inode)->root;
570 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
571 struct extent_io_tree *io_tree;
574 if (list_empty(&async_cow->extents))
578 while (!list_empty(&async_cow->extents)) {
579 async_extent = list_entry(async_cow->extents.next,
580 struct async_extent, list);
581 list_del(&async_extent->list);
583 io_tree = &BTRFS_I(inode)->io_tree;
586 /* did the compression code fall back to uncompressed IO? */
587 if (!async_extent->pages) {
588 int page_started = 0;
589 unsigned long nr_written = 0;
591 lock_extent(io_tree, async_extent->start,
592 async_extent->start +
593 async_extent->ram_size - 1, GFP_NOFS);
595 /* allocate blocks */
596 ret = cow_file_range(inode, async_cow->locked_page,
598 async_extent->start +
599 async_extent->ram_size - 1,
600 &page_started, &nr_written, 0);
603 * if page_started, cow_file_range inserted an
604 * inline extent and took care of all the unlocking
605 * and IO for us. Otherwise, we need to submit
606 * all those pages down to the drive.
608 if (!page_started && !ret)
609 extent_write_locked_range(io_tree,
610 inode, async_extent->start,
611 async_extent->start +
612 async_extent->ram_size - 1,
620 lock_extent(io_tree, async_extent->start,
621 async_extent->start + async_extent->ram_size - 1,
624 trans = btrfs_join_transaction(root);
625 BUG_ON(IS_ERR(trans));
626 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
627 ret = btrfs_reserve_extent(trans, root,
628 async_extent->compressed_size,
629 async_extent->compressed_size,
632 btrfs_end_transaction(trans, root);
636 for (i = 0; i < async_extent->nr_pages; i++) {
637 WARN_ON(async_extent->pages[i]->mapping);
638 page_cache_release(async_extent->pages[i]);
640 kfree(async_extent->pages);
641 async_extent->nr_pages = 0;
642 async_extent->pages = NULL;
643 unlock_extent(io_tree, async_extent->start,
644 async_extent->start +
645 async_extent->ram_size - 1, GFP_NOFS);
650 * here we're doing allocation and writeback of the
653 btrfs_drop_extent_cache(inode, async_extent->start,
654 async_extent->start +
655 async_extent->ram_size - 1, 0);
657 em = alloc_extent_map();
659 em->start = async_extent->start;
660 em->len = async_extent->ram_size;
661 em->orig_start = em->start;
663 em->block_start = ins.objectid;
664 em->block_len = ins.offset;
665 em->bdev = root->fs_info->fs_devices->latest_bdev;
666 em->compress_type = async_extent->compress_type;
667 set_bit(EXTENT_FLAG_PINNED, &em->flags);
668 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
671 write_lock(&em_tree->lock);
672 ret = add_extent_mapping(em_tree, em);
673 write_unlock(&em_tree->lock);
674 if (ret != -EEXIST) {
678 btrfs_drop_extent_cache(inode, async_extent->start,
679 async_extent->start +
680 async_extent->ram_size - 1, 0);
683 ret = btrfs_add_ordered_extent_compress(inode,
686 async_extent->ram_size,
688 BTRFS_ORDERED_COMPRESSED,
689 async_extent->compress_type);
693 * clear dirty, set writeback and unlock the pages.
695 extent_clear_unlock_delalloc(inode,
696 &BTRFS_I(inode)->io_tree,
698 async_extent->start +
699 async_extent->ram_size - 1,
700 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
701 EXTENT_CLEAR_UNLOCK |
702 EXTENT_CLEAR_DELALLOC |
703 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
705 ret = btrfs_submit_compressed_write(inode,
707 async_extent->ram_size,
709 ins.offset, async_extent->pages,
710 async_extent->nr_pages);
713 alloc_hint = ins.objectid + ins.offset;
721 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
724 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
725 struct extent_map *em;
728 read_lock(&em_tree->lock);
729 em = search_extent_mapping(em_tree, start, num_bytes);
732 * if block start isn't an actual block number then find the
733 * first block in this inode and use that as a hint. If that
734 * block is also bogus then just don't worry about it.
736 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
738 em = search_extent_mapping(em_tree, 0, 0);
739 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
740 alloc_hint = em->block_start;
744 alloc_hint = em->block_start;
748 read_unlock(&em_tree->lock);
754 * when extent_io.c finds a delayed allocation range in the file,
755 * the call backs end up in this code. The basic idea is to
756 * allocate extents on disk for the range, and create ordered data structs
757 * in ram to track those extents.
759 * locked_page is the page that writepage had locked already. We use
760 * it to make sure we don't do extra locks or unlocks.
762 * *page_started is set to one if we unlock locked_page and do everything
763 * required to start IO on it. It may be clean and already done with
766 static noinline int cow_file_range(struct inode *inode,
767 struct page *locked_page,
768 u64 start, u64 end, int *page_started,
769 unsigned long *nr_written,
772 struct btrfs_root *root = BTRFS_I(inode)->root;
773 struct btrfs_trans_handle *trans;
776 unsigned long ram_size;
779 u64 blocksize = root->sectorsize;
780 struct btrfs_key ins;
781 struct extent_map *em;
782 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
785 BUG_ON(btrfs_is_free_space_inode(root, inode));
786 trans = btrfs_join_transaction(root);
787 BUG_ON(IS_ERR(trans));
788 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
790 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
791 num_bytes = max(blocksize, num_bytes);
792 disk_num_bytes = num_bytes;
795 /* if this is a small write inside eof, kick off defrag */
796 if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
797 btrfs_add_inode_defrag(trans, inode);
800 /* lets try to make an inline extent */
801 ret = cow_file_range_inline(trans, root, inode,
802 start, end, 0, 0, NULL);
804 extent_clear_unlock_delalloc(inode,
805 &BTRFS_I(inode)->io_tree,
807 EXTENT_CLEAR_UNLOCK_PAGE |
808 EXTENT_CLEAR_UNLOCK |
809 EXTENT_CLEAR_DELALLOC |
811 EXTENT_SET_WRITEBACK |
812 EXTENT_END_WRITEBACK);
814 *nr_written = *nr_written +
815 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
822 BUG_ON(disk_num_bytes >
823 btrfs_super_total_bytes(&root->fs_info->super_copy));
825 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
826 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
828 while (disk_num_bytes > 0) {
831 cur_alloc_size = disk_num_bytes;
832 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
833 root->sectorsize, 0, alloc_hint,
837 em = alloc_extent_map();
840 em->orig_start = em->start;
841 ram_size = ins.offset;
842 em->len = ins.offset;
844 em->block_start = ins.objectid;
845 em->block_len = ins.offset;
846 em->bdev = root->fs_info->fs_devices->latest_bdev;
847 set_bit(EXTENT_FLAG_PINNED, &em->flags);
850 write_lock(&em_tree->lock);
851 ret = add_extent_mapping(em_tree, em);
852 write_unlock(&em_tree->lock);
853 if (ret != -EEXIST) {
857 btrfs_drop_extent_cache(inode, start,
858 start + ram_size - 1, 0);
861 cur_alloc_size = ins.offset;
862 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
863 ram_size, cur_alloc_size, 0);
866 if (root->root_key.objectid ==
867 BTRFS_DATA_RELOC_TREE_OBJECTID) {
868 ret = btrfs_reloc_clone_csums(inode, start,
873 if (disk_num_bytes < cur_alloc_size)
876 /* we're not doing compressed IO, don't unlock the first
877 * page (which the caller expects to stay locked), don't
878 * clear any dirty bits and don't set any writeback bits
880 * Do set the Private2 bit so we know this page was properly
881 * setup for writepage
883 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
884 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
887 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
888 start, start + ram_size - 1,
890 disk_num_bytes -= cur_alloc_size;
891 num_bytes -= cur_alloc_size;
892 alloc_hint = ins.objectid + ins.offset;
893 start += cur_alloc_size;
897 btrfs_end_transaction(trans, root);
903 * work queue call back to started compression on a file and pages
905 static noinline void async_cow_start(struct btrfs_work *work)
907 struct async_cow *async_cow;
909 async_cow = container_of(work, struct async_cow, work);
911 compress_file_range(async_cow->inode, async_cow->locked_page,
912 async_cow->start, async_cow->end, async_cow,
915 async_cow->inode = NULL;
919 * work queue call back to submit previously compressed pages
921 static noinline void async_cow_submit(struct btrfs_work *work)
923 struct async_cow *async_cow;
924 struct btrfs_root *root;
925 unsigned long nr_pages;
927 async_cow = container_of(work, struct async_cow, work);
929 root = async_cow->root;
930 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
933 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
935 if (atomic_read(&root->fs_info->async_delalloc_pages) <
937 waitqueue_active(&root->fs_info->async_submit_wait))
938 wake_up(&root->fs_info->async_submit_wait);
940 if (async_cow->inode)
941 submit_compressed_extents(async_cow->inode, async_cow);
944 static noinline void async_cow_free(struct btrfs_work *work)
946 struct async_cow *async_cow;
947 async_cow = container_of(work, struct async_cow, work);
951 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
952 u64 start, u64 end, int *page_started,
953 unsigned long *nr_written)
955 struct async_cow *async_cow;
956 struct btrfs_root *root = BTRFS_I(inode)->root;
957 unsigned long nr_pages;
959 int limit = 10 * 1024 * 1042;
961 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
962 1, 0, NULL, GFP_NOFS);
963 while (start < end) {
964 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
966 async_cow->inode = inode;
967 async_cow->root = root;
968 async_cow->locked_page = locked_page;
969 async_cow->start = start;
971 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
974 cur_end = min(end, start + 512 * 1024 - 1);
976 async_cow->end = cur_end;
977 INIT_LIST_HEAD(&async_cow->extents);
979 async_cow->work.func = async_cow_start;
980 async_cow->work.ordered_func = async_cow_submit;
981 async_cow->work.ordered_free = async_cow_free;
982 async_cow->work.flags = 0;
984 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
986 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
988 btrfs_queue_worker(&root->fs_info->delalloc_workers,
991 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
992 wait_event(root->fs_info->async_submit_wait,
993 (atomic_read(&root->fs_info->async_delalloc_pages) <
997 while (atomic_read(&root->fs_info->async_submit_draining) &&
998 atomic_read(&root->fs_info->async_delalloc_pages)) {
999 wait_event(root->fs_info->async_submit_wait,
1000 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1004 *nr_written += nr_pages;
1005 start = cur_end + 1;
1011 static noinline int csum_exist_in_range(struct btrfs_root *root,
1012 u64 bytenr, u64 num_bytes)
1015 struct btrfs_ordered_sum *sums;
1018 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1019 bytenr + num_bytes - 1, &list, 0);
1020 if (ret == 0 && list_empty(&list))
1023 while (!list_empty(&list)) {
1024 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1025 list_del(&sums->list);
1032 * when nowcow writeback call back. This checks for snapshots or COW copies
1033 * of the extents that exist in the file, and COWs the file as required.
1035 * If no cow copies or snapshots exist, we write directly to the existing
1038 static noinline int run_delalloc_nocow(struct inode *inode,
1039 struct page *locked_page,
1040 u64 start, u64 end, int *page_started, int force,
1041 unsigned long *nr_written)
1043 struct btrfs_root *root = BTRFS_I(inode)->root;
1044 struct btrfs_trans_handle *trans;
1045 struct extent_buffer *leaf;
1046 struct btrfs_path *path;
1047 struct btrfs_file_extent_item *fi;
1048 struct btrfs_key found_key;
1061 u64 ino = btrfs_ino(inode);
1063 path = btrfs_alloc_path();
1067 nolock = btrfs_is_free_space_inode(root, inode);
1070 trans = btrfs_join_transaction_nolock(root);
1072 trans = btrfs_join_transaction(root);
1074 BUG_ON(IS_ERR(trans));
1075 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1077 cow_start = (u64)-1;
1080 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1083 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1084 leaf = path->nodes[0];
1085 btrfs_item_key_to_cpu(leaf, &found_key,
1086 path->slots[0] - 1);
1087 if (found_key.objectid == ino &&
1088 found_key.type == BTRFS_EXTENT_DATA_KEY)
1093 leaf = path->nodes[0];
1094 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1095 ret = btrfs_next_leaf(root, path);
1100 leaf = path->nodes[0];
1106 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1108 if (found_key.objectid > ino ||
1109 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1110 found_key.offset > end)
1113 if (found_key.offset > cur_offset) {
1114 extent_end = found_key.offset;
1119 fi = btrfs_item_ptr(leaf, path->slots[0],
1120 struct btrfs_file_extent_item);
1121 extent_type = btrfs_file_extent_type(leaf, fi);
1123 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1124 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1125 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1126 extent_offset = btrfs_file_extent_offset(leaf, fi);
1127 extent_end = found_key.offset +
1128 btrfs_file_extent_num_bytes(leaf, fi);
1129 if (extent_end <= start) {
1133 if (disk_bytenr == 0)
1135 if (btrfs_file_extent_compression(leaf, fi) ||
1136 btrfs_file_extent_encryption(leaf, fi) ||
1137 btrfs_file_extent_other_encoding(leaf, fi))
1139 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1141 if (btrfs_extent_readonly(root, disk_bytenr))
1143 if (btrfs_cross_ref_exist(trans, root, ino,
1145 extent_offset, disk_bytenr))
1147 disk_bytenr += extent_offset;
1148 disk_bytenr += cur_offset - found_key.offset;
1149 num_bytes = min(end + 1, extent_end) - cur_offset;
1151 * force cow if csum exists in the range.
1152 * this ensure that csum for a given extent are
1153 * either valid or do not exist.
1155 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1158 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1159 extent_end = found_key.offset +
1160 btrfs_file_extent_inline_len(leaf, fi);
1161 extent_end = ALIGN(extent_end, root->sectorsize);
1166 if (extent_end <= start) {
1171 if (cow_start == (u64)-1)
1172 cow_start = cur_offset;
1173 cur_offset = extent_end;
1174 if (cur_offset > end)
1180 btrfs_release_path(path);
1181 if (cow_start != (u64)-1) {
1182 ret = cow_file_range(inode, locked_page, cow_start,
1183 found_key.offset - 1, page_started,
1186 cow_start = (u64)-1;
1189 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1190 struct extent_map *em;
1191 struct extent_map_tree *em_tree;
1192 em_tree = &BTRFS_I(inode)->extent_tree;
1193 em = alloc_extent_map();
1195 em->start = cur_offset;
1196 em->orig_start = em->start;
1197 em->len = num_bytes;
1198 em->block_len = num_bytes;
1199 em->block_start = disk_bytenr;
1200 em->bdev = root->fs_info->fs_devices->latest_bdev;
1201 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1203 write_lock(&em_tree->lock);
1204 ret = add_extent_mapping(em_tree, em);
1205 write_unlock(&em_tree->lock);
1206 if (ret != -EEXIST) {
1207 free_extent_map(em);
1210 btrfs_drop_extent_cache(inode, em->start,
1211 em->start + em->len - 1, 0);
1213 type = BTRFS_ORDERED_PREALLOC;
1215 type = BTRFS_ORDERED_NOCOW;
1218 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1219 num_bytes, num_bytes, type);
1222 if (root->root_key.objectid ==
1223 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1224 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1229 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1230 cur_offset, cur_offset + num_bytes - 1,
1231 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1232 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1233 EXTENT_SET_PRIVATE2);
1234 cur_offset = extent_end;
1235 if (cur_offset > end)
1238 btrfs_release_path(path);
1240 if (cur_offset <= end && cow_start == (u64)-1)
1241 cow_start = cur_offset;
1242 if (cow_start != (u64)-1) {
1243 ret = cow_file_range(inode, locked_page, cow_start, end,
1244 page_started, nr_written, 1);
1249 ret = btrfs_end_transaction_nolock(trans, root);
1252 ret = btrfs_end_transaction(trans, root);
1255 btrfs_free_path(path);
1260 * extent_io.c call back to do delayed allocation processing
1262 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1263 u64 start, u64 end, int *page_started,
1264 unsigned long *nr_written)
1267 struct btrfs_root *root = BTRFS_I(inode)->root;
1269 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1270 ret = run_delalloc_nocow(inode, locked_page, start, end,
1271 page_started, 1, nr_written);
1272 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1273 ret = run_delalloc_nocow(inode, locked_page, start, end,
1274 page_started, 0, nr_written);
1275 else if (!btrfs_test_opt(root, COMPRESS) &&
1276 !(BTRFS_I(inode)->force_compress) &&
1277 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1278 ret = cow_file_range(inode, locked_page, start, end,
1279 page_started, nr_written, 1);
1281 ret = cow_file_range_async(inode, locked_page, start, end,
1282 page_started, nr_written);
1286 static void btrfs_split_extent_hook(struct inode *inode,
1287 struct extent_state *orig, u64 split)
1289 /* not delalloc, ignore it */
1290 if (!(orig->state & EXTENT_DELALLOC))
1293 spin_lock(&BTRFS_I(inode)->lock);
1294 BTRFS_I(inode)->outstanding_extents++;
1295 spin_unlock(&BTRFS_I(inode)->lock);
1299 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1300 * extents so we can keep track of new extents that are just merged onto old
1301 * extents, such as when we are doing sequential writes, so we can properly
1302 * account for the metadata space we'll need.
1304 static void btrfs_merge_extent_hook(struct inode *inode,
1305 struct extent_state *new,
1306 struct extent_state *other)
1308 /* not delalloc, ignore it */
1309 if (!(other->state & EXTENT_DELALLOC))
1312 spin_lock(&BTRFS_I(inode)->lock);
1313 BTRFS_I(inode)->outstanding_extents--;
1314 spin_unlock(&BTRFS_I(inode)->lock);
1318 * extent_io.c set_bit_hook, used to track delayed allocation
1319 * bytes in this file, and to maintain the list of inodes that
1320 * have pending delalloc work to be done.
1322 static void btrfs_set_bit_hook(struct inode *inode,
1323 struct extent_state *state, int *bits)
1327 * set_bit and clear bit hooks normally require _irqsave/restore
1328 * but in this case, we are only testing for the DELALLOC
1329 * bit, which is only set or cleared with irqs on
1331 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1332 struct btrfs_root *root = BTRFS_I(inode)->root;
1333 u64 len = state->end + 1 - state->start;
1334 bool do_list = !btrfs_is_free_space_inode(root, inode);
1336 if (*bits & EXTENT_FIRST_DELALLOC) {
1337 *bits &= ~EXTENT_FIRST_DELALLOC;
1339 spin_lock(&BTRFS_I(inode)->lock);
1340 BTRFS_I(inode)->outstanding_extents++;
1341 spin_unlock(&BTRFS_I(inode)->lock);
1344 spin_lock(&root->fs_info->delalloc_lock);
1345 BTRFS_I(inode)->delalloc_bytes += len;
1346 root->fs_info->delalloc_bytes += len;
1347 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1348 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1349 &root->fs_info->delalloc_inodes);
1351 spin_unlock(&root->fs_info->delalloc_lock);
1356 * extent_io.c clear_bit_hook, see set_bit_hook for why
1358 static void btrfs_clear_bit_hook(struct inode *inode,
1359 struct extent_state *state, int *bits)
1362 * set_bit and clear bit hooks normally require _irqsave/restore
1363 * but in this case, we are only testing for the DELALLOC
1364 * bit, which is only set or cleared with irqs on
1366 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1367 struct btrfs_root *root = BTRFS_I(inode)->root;
1368 u64 len = state->end + 1 - state->start;
1369 bool do_list = !btrfs_is_free_space_inode(root, inode);
1371 if (*bits & EXTENT_FIRST_DELALLOC) {
1372 *bits &= ~EXTENT_FIRST_DELALLOC;
1373 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1374 spin_lock(&BTRFS_I(inode)->lock);
1375 BTRFS_I(inode)->outstanding_extents--;
1376 spin_unlock(&BTRFS_I(inode)->lock);
1379 if (*bits & EXTENT_DO_ACCOUNTING)
1380 btrfs_delalloc_release_metadata(inode, len);
1382 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1384 btrfs_free_reserved_data_space(inode, len);
1386 spin_lock(&root->fs_info->delalloc_lock);
1387 root->fs_info->delalloc_bytes -= len;
1388 BTRFS_I(inode)->delalloc_bytes -= len;
1390 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1391 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1392 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1394 spin_unlock(&root->fs_info->delalloc_lock);
1399 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1400 * we don't create bios that span stripes or chunks
1402 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1403 size_t size, struct bio *bio,
1404 unsigned long bio_flags)
1406 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1407 struct btrfs_mapping_tree *map_tree;
1408 u64 logical = (u64)bio->bi_sector << 9;
1413 if (bio_flags & EXTENT_BIO_COMPRESSED)
1416 length = bio->bi_size;
1417 map_tree = &root->fs_info->mapping_tree;
1418 map_length = length;
1419 ret = btrfs_map_block(map_tree, READ, logical,
1420 &map_length, NULL, 0);
1422 if (map_length < length + size)
1428 * in order to insert checksums into the metadata in large chunks,
1429 * we wait until bio submission time. All the pages in the bio are
1430 * checksummed and sums are attached onto the ordered extent record.
1432 * At IO completion time the cums attached on the ordered extent record
1433 * are inserted into the btree
1435 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1436 struct bio *bio, int mirror_num,
1437 unsigned long bio_flags,
1440 struct btrfs_root *root = BTRFS_I(inode)->root;
1443 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1449 * in order to insert checksums into the metadata in large chunks,
1450 * we wait until bio submission time. All the pages in the bio are
1451 * checksummed and sums are attached onto the ordered extent record.
1453 * At IO completion time the cums attached on the ordered extent record
1454 * are inserted into the btree
1456 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1457 int mirror_num, unsigned long bio_flags,
1460 struct btrfs_root *root = BTRFS_I(inode)->root;
1461 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1465 * extent_io.c submission hook. This does the right thing for csum calculation
1466 * on write, or reading the csums from the tree before a read
1468 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1469 int mirror_num, unsigned long bio_flags,
1472 struct btrfs_root *root = BTRFS_I(inode)->root;
1476 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1478 if (btrfs_is_free_space_inode(root, inode))
1479 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1481 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1484 if (!(rw & REQ_WRITE)) {
1485 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1486 return btrfs_submit_compressed_read(inode, bio,
1487 mirror_num, bio_flags);
1488 } else if (!skip_sum) {
1489 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1494 } else if (!skip_sum) {
1495 /* csum items have already been cloned */
1496 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1498 /* we're doing a write, do the async checksumming */
1499 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1500 inode, rw, bio, mirror_num,
1501 bio_flags, bio_offset,
1502 __btrfs_submit_bio_start,
1503 __btrfs_submit_bio_done);
1507 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1511 * given a list of ordered sums record them in the inode. This happens
1512 * at IO completion time based on sums calculated at bio submission time.
1514 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1515 struct inode *inode, u64 file_offset,
1516 struct list_head *list)
1518 struct btrfs_ordered_sum *sum;
1520 list_for_each_entry(sum, list, list) {
1521 btrfs_csum_file_blocks(trans,
1522 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1527 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1528 struct extent_state **cached_state)
1530 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1532 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1533 cached_state, GFP_NOFS);
1536 /* see btrfs_writepage_start_hook for details on why this is required */
1537 struct btrfs_writepage_fixup {
1539 struct btrfs_work work;
1542 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1544 struct btrfs_writepage_fixup *fixup;
1545 struct btrfs_ordered_extent *ordered;
1546 struct extent_state *cached_state = NULL;
1548 struct inode *inode;
1552 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1556 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1557 ClearPageChecked(page);
1561 inode = page->mapping->host;
1562 page_start = page_offset(page);
1563 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1565 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1566 &cached_state, GFP_NOFS);
1568 /* already ordered? We're done */
1569 if (PagePrivate2(page))
1572 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1574 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1575 page_end, &cached_state, GFP_NOFS);
1577 btrfs_start_ordered_extent(inode, ordered, 1);
1582 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1583 ClearPageChecked(page);
1585 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1586 &cached_state, GFP_NOFS);
1589 page_cache_release(page);
1594 * There are a few paths in the higher layers of the kernel that directly
1595 * set the page dirty bit without asking the filesystem if it is a
1596 * good idea. This causes problems because we want to make sure COW
1597 * properly happens and the data=ordered rules are followed.
1599 * In our case any range that doesn't have the ORDERED bit set
1600 * hasn't been properly setup for IO. We kick off an async process
1601 * to fix it up. The async helper will wait for ordered extents, set
1602 * the delalloc bit and make it safe to write the page.
1604 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1606 struct inode *inode = page->mapping->host;
1607 struct btrfs_writepage_fixup *fixup;
1608 struct btrfs_root *root = BTRFS_I(inode)->root;
1610 /* this page is properly in the ordered list */
1611 if (TestClearPagePrivate2(page))
1614 if (PageChecked(page))
1617 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1621 SetPageChecked(page);
1622 page_cache_get(page);
1623 fixup->work.func = btrfs_writepage_fixup_worker;
1625 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1629 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1630 struct inode *inode, u64 file_pos,
1631 u64 disk_bytenr, u64 disk_num_bytes,
1632 u64 num_bytes, u64 ram_bytes,
1633 u8 compression, u8 encryption,
1634 u16 other_encoding, int extent_type)
1636 struct btrfs_root *root = BTRFS_I(inode)->root;
1637 struct btrfs_file_extent_item *fi;
1638 struct btrfs_path *path;
1639 struct extent_buffer *leaf;
1640 struct btrfs_key ins;
1644 path = btrfs_alloc_path();
1648 path->leave_spinning = 1;
1651 * we may be replacing one extent in the tree with another.
1652 * The new extent is pinned in the extent map, and we don't want
1653 * to drop it from the cache until it is completely in the btree.
1655 * So, tell btrfs_drop_extents to leave this extent in the cache.
1656 * the caller is expected to unpin it and allow it to be merged
1659 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1663 ins.objectid = btrfs_ino(inode);
1664 ins.offset = file_pos;
1665 ins.type = BTRFS_EXTENT_DATA_KEY;
1666 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1668 leaf = path->nodes[0];
1669 fi = btrfs_item_ptr(leaf, path->slots[0],
1670 struct btrfs_file_extent_item);
1671 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1672 btrfs_set_file_extent_type(leaf, fi, extent_type);
1673 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1674 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1675 btrfs_set_file_extent_offset(leaf, fi, 0);
1676 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1677 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1678 btrfs_set_file_extent_compression(leaf, fi, compression);
1679 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1680 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1682 btrfs_unlock_up_safe(path, 1);
1683 btrfs_set_lock_blocking(leaf);
1685 btrfs_mark_buffer_dirty(leaf);
1687 inode_add_bytes(inode, num_bytes);
1689 ins.objectid = disk_bytenr;
1690 ins.offset = disk_num_bytes;
1691 ins.type = BTRFS_EXTENT_ITEM_KEY;
1692 ret = btrfs_alloc_reserved_file_extent(trans, root,
1693 root->root_key.objectid,
1694 btrfs_ino(inode), file_pos, &ins);
1696 btrfs_free_path(path);
1702 * helper function for btrfs_finish_ordered_io, this
1703 * just reads in some of the csum leaves to prime them into ram
1704 * before we start the transaction. It limits the amount of btree
1705 * reads required while inside the transaction.
1707 /* as ordered data IO finishes, this gets called so we can finish
1708 * an ordered extent if the range of bytes in the file it covers are
1711 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1713 struct btrfs_root *root = BTRFS_I(inode)->root;
1714 struct btrfs_trans_handle *trans = NULL;
1715 struct btrfs_ordered_extent *ordered_extent = NULL;
1716 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1717 struct extent_state *cached_state = NULL;
1718 int compress_type = 0;
1722 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1726 BUG_ON(!ordered_extent);
1728 nolock = btrfs_is_free_space_inode(root, inode);
1730 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1731 BUG_ON(!list_empty(&ordered_extent->list));
1732 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1735 trans = btrfs_join_transaction_nolock(root);
1737 trans = btrfs_join_transaction(root);
1738 BUG_ON(IS_ERR(trans));
1739 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1740 ret = btrfs_update_inode(trans, root, inode);
1746 lock_extent_bits(io_tree, ordered_extent->file_offset,
1747 ordered_extent->file_offset + ordered_extent->len - 1,
1748 0, &cached_state, GFP_NOFS);
1751 trans = btrfs_join_transaction_nolock(root);
1753 trans = btrfs_join_transaction(root);
1754 BUG_ON(IS_ERR(trans));
1755 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1757 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1758 compress_type = ordered_extent->compress_type;
1759 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1760 BUG_ON(compress_type);
1761 ret = btrfs_mark_extent_written(trans, inode,
1762 ordered_extent->file_offset,
1763 ordered_extent->file_offset +
1764 ordered_extent->len);
1767 BUG_ON(root == root->fs_info->tree_root);
1768 ret = insert_reserved_file_extent(trans, inode,
1769 ordered_extent->file_offset,
1770 ordered_extent->start,
1771 ordered_extent->disk_len,
1772 ordered_extent->len,
1773 ordered_extent->len,
1774 compress_type, 0, 0,
1775 BTRFS_FILE_EXTENT_REG);
1776 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1777 ordered_extent->file_offset,
1778 ordered_extent->len);
1781 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1782 ordered_extent->file_offset +
1783 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1785 add_pending_csums(trans, inode, ordered_extent->file_offset,
1786 &ordered_extent->list);
1788 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1789 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1790 ret = btrfs_update_inode(trans, root, inode);
1797 btrfs_end_transaction_nolock(trans, root);
1799 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1801 btrfs_end_transaction(trans, root);
1805 btrfs_put_ordered_extent(ordered_extent);
1806 /* once for the tree */
1807 btrfs_put_ordered_extent(ordered_extent);
1812 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1813 struct extent_state *state, int uptodate)
1815 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1817 ClearPagePrivate2(page);
1818 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1822 * When IO fails, either with EIO or csum verification fails, we
1823 * try other mirrors that might have a good copy of the data. This
1824 * io_failure_record is used to record state as we go through all the
1825 * mirrors. If another mirror has good data, the page is set up to date
1826 * and things continue. If a good mirror can't be found, the original
1827 * bio end_io callback is called to indicate things have failed.
1829 struct io_failure_record {
1834 unsigned long bio_flags;
1838 static int btrfs_io_failed_hook(struct bio *failed_bio,
1839 struct page *page, u64 start, u64 end,
1840 struct extent_state *state)
1842 struct io_failure_record *failrec = NULL;
1844 struct extent_map *em;
1845 struct inode *inode = page->mapping->host;
1846 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1847 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1854 ret = get_state_private(failure_tree, start, &private);
1856 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1859 failrec->start = start;
1860 failrec->len = end - start + 1;
1861 failrec->last_mirror = 0;
1862 failrec->bio_flags = 0;
1864 read_lock(&em_tree->lock);
1865 em = lookup_extent_mapping(em_tree, start, failrec->len);
1866 if (em && !IS_ERR(em) && (em->start > start ||
1867 em->start + em->len < start)) {
1868 free_extent_map(em);
1871 read_unlock(&em_tree->lock);
1873 if (IS_ERR_OR_NULL(em)) {
1877 logical = start - em->start;
1878 logical = em->block_start + logical;
1879 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1880 logical = em->block_start;
1881 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1882 extent_set_compress_type(&failrec->bio_flags,
1885 failrec->logical = logical;
1886 free_extent_map(em);
1887 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1888 EXTENT_DIRTY, GFP_NOFS);
1889 set_state_private(failure_tree, start,
1890 (u64)(unsigned long)failrec);
1892 failrec = (struct io_failure_record *)(unsigned long)private;
1894 num_copies = btrfs_num_copies(
1895 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1896 failrec->logical, failrec->len);
1897 failrec->last_mirror++;
1899 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1900 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1903 if (state && state->start != failrec->start)
1905 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1907 if (!state || failrec->last_mirror > num_copies) {
1908 set_state_private(failure_tree, failrec->start, 0);
1909 clear_extent_bits(failure_tree, failrec->start,
1910 failrec->start + failrec->len - 1,
1911 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1915 bio = bio_alloc(GFP_NOFS, 1);
1916 bio->bi_private = state;
1917 bio->bi_end_io = failed_bio->bi_end_io;
1918 bio->bi_sector = failrec->logical >> 9;
1919 bio->bi_bdev = failed_bio->bi_bdev;
1922 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1923 if (failed_bio->bi_rw & REQ_WRITE)
1928 ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1929 failrec->last_mirror,
1930 failrec->bio_flags, 0);
1935 * each time an IO finishes, we do a fast check in the IO failure tree
1936 * to see if we need to process or clean up an io_failure_record
1938 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1941 u64 private_failure;
1942 struct io_failure_record *failure;
1946 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1947 (u64)-1, 1, EXTENT_DIRTY, 0)) {
1948 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1949 start, &private_failure);
1951 failure = (struct io_failure_record *)(unsigned long)
1953 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1955 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1957 failure->start + failure->len - 1,
1958 EXTENT_DIRTY | EXTENT_LOCKED,
1967 * when reads are done, we need to check csums to verify the data is correct
1968 * if there's a match, we allow the bio to finish. If not, we go through
1969 * the io_failure_record routines to find good copies
1971 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1972 struct extent_state *state)
1974 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1975 struct inode *inode = page->mapping->host;
1976 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1978 u64 private = ~(u32)0;
1980 struct btrfs_root *root = BTRFS_I(inode)->root;
1983 if (PageChecked(page)) {
1984 ClearPageChecked(page);
1988 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1991 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1992 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1993 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1998 if (state && state->start == start) {
1999 private = state->private;
2002 ret = get_state_private(io_tree, start, &private);
2004 kaddr = kmap_atomic(page, KM_USER0);
2008 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
2009 btrfs_csum_final(csum, (char *)&csum);
2010 if (csum != private)
2013 kunmap_atomic(kaddr, KM_USER0);
2015 /* if the io failure tree for this inode is non-empty,
2016 * check to see if we've recovered from a failed IO
2018 btrfs_clean_io_failures(inode, start);
2022 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2024 (unsigned long long)btrfs_ino(page->mapping->host),
2025 (unsigned long long)start, csum,
2026 (unsigned long long)private);
2027 memset(kaddr + offset, 1, end - start + 1);
2028 flush_dcache_page(page);
2029 kunmap_atomic(kaddr, KM_USER0);
2035 struct delayed_iput {
2036 struct list_head list;
2037 struct inode *inode;
2040 void btrfs_add_delayed_iput(struct inode *inode)
2042 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2043 struct delayed_iput *delayed;
2045 if (atomic_add_unless(&inode->i_count, -1, 1))
2048 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2049 delayed->inode = inode;
2051 spin_lock(&fs_info->delayed_iput_lock);
2052 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2053 spin_unlock(&fs_info->delayed_iput_lock);
2056 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2059 struct btrfs_fs_info *fs_info = root->fs_info;
2060 struct delayed_iput *delayed;
2063 spin_lock(&fs_info->delayed_iput_lock);
2064 empty = list_empty(&fs_info->delayed_iputs);
2065 spin_unlock(&fs_info->delayed_iput_lock);
2069 down_read(&root->fs_info->cleanup_work_sem);
2070 spin_lock(&fs_info->delayed_iput_lock);
2071 list_splice_init(&fs_info->delayed_iputs, &list);
2072 spin_unlock(&fs_info->delayed_iput_lock);
2074 while (!list_empty(&list)) {
2075 delayed = list_entry(list.next, struct delayed_iput, list);
2076 list_del(&delayed->list);
2077 iput(delayed->inode);
2080 up_read(&root->fs_info->cleanup_work_sem);
2084 * calculate extra metadata reservation when snapshotting a subvolume
2085 * contains orphan files.
2087 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2088 struct btrfs_pending_snapshot *pending,
2089 u64 *bytes_to_reserve)
2091 struct btrfs_root *root;
2092 struct btrfs_block_rsv *block_rsv;
2096 root = pending->root;
2097 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2100 block_rsv = root->orphan_block_rsv;
2102 /* orphan block reservation for the snapshot */
2103 num_bytes = block_rsv->size;
2106 * after the snapshot is created, COWing tree blocks may use more
2107 * space than it frees. So we should make sure there is enough
2110 index = trans->transid & 0x1;
2111 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2112 num_bytes += block_rsv->size -
2113 (block_rsv->reserved + block_rsv->freed[index]);
2116 *bytes_to_reserve += num_bytes;
2119 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2120 struct btrfs_pending_snapshot *pending)
2122 struct btrfs_root *root = pending->root;
2123 struct btrfs_root *snap = pending->snap;
2124 struct btrfs_block_rsv *block_rsv;
2129 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2132 /* refill source subvolume's orphan block reservation */
2133 block_rsv = root->orphan_block_rsv;
2134 index = trans->transid & 0x1;
2135 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2136 num_bytes = block_rsv->size -
2137 (block_rsv->reserved + block_rsv->freed[index]);
2138 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2139 root->orphan_block_rsv,
2144 /* setup orphan block reservation for the snapshot */
2145 block_rsv = btrfs_alloc_block_rsv(snap);
2148 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2149 snap->orphan_block_rsv = block_rsv;
2151 num_bytes = root->orphan_block_rsv->size;
2152 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2153 block_rsv, num_bytes);
2157 /* insert orphan item for the snapshot */
2158 WARN_ON(!root->orphan_item_inserted);
2159 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2160 snap->root_key.objectid);
2162 snap->orphan_item_inserted = 1;
2166 enum btrfs_orphan_cleanup_state {
2167 ORPHAN_CLEANUP_STARTED = 1,
2168 ORPHAN_CLEANUP_DONE = 2,
2172 * This is called in transaction commmit time. If there are no orphan
2173 * files in the subvolume, it removes orphan item and frees block_rsv
2176 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2177 struct btrfs_root *root)
2181 if (!list_empty(&root->orphan_list) ||
2182 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2185 if (root->orphan_item_inserted &&
2186 btrfs_root_refs(&root->root_item) > 0) {
2187 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2188 root->root_key.objectid);
2190 root->orphan_item_inserted = 0;
2193 if (root->orphan_block_rsv) {
2194 WARN_ON(root->orphan_block_rsv->size > 0);
2195 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2196 root->orphan_block_rsv = NULL;
2201 * This creates an orphan entry for the given inode in case something goes
2202 * wrong in the middle of an unlink/truncate.
2204 * NOTE: caller of this function should reserve 5 units of metadata for
2207 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2209 struct btrfs_root *root = BTRFS_I(inode)->root;
2210 struct btrfs_block_rsv *block_rsv = NULL;
2215 if (!root->orphan_block_rsv) {
2216 block_rsv = btrfs_alloc_block_rsv(root);
2221 spin_lock(&root->orphan_lock);
2222 if (!root->orphan_block_rsv) {
2223 root->orphan_block_rsv = block_rsv;
2224 } else if (block_rsv) {
2225 btrfs_free_block_rsv(root, block_rsv);
2229 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2230 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2233 * For proper ENOSPC handling, we should do orphan
2234 * cleanup when mounting. But this introduces backward
2235 * compatibility issue.
2237 if (!xchg(&root->orphan_item_inserted, 1))
2245 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2246 BTRFS_I(inode)->orphan_meta_reserved = 1;
2249 spin_unlock(&root->orphan_lock);
2252 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2254 /* grab metadata reservation from transaction handle */
2256 ret = btrfs_orphan_reserve_metadata(trans, inode);
2260 /* insert an orphan item to track this unlinked/truncated file */
2262 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2266 /* insert an orphan item to track subvolume contains orphan files */
2268 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2269 root->root_key.objectid);
2276 * We have done the truncate/delete so we can go ahead and remove the orphan
2277 * item for this particular inode.
2279 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2281 struct btrfs_root *root = BTRFS_I(inode)->root;
2282 int delete_item = 0;
2283 int release_rsv = 0;
2286 spin_lock(&root->orphan_lock);
2287 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2288 list_del_init(&BTRFS_I(inode)->i_orphan);
2292 if (BTRFS_I(inode)->orphan_meta_reserved) {
2293 BTRFS_I(inode)->orphan_meta_reserved = 0;
2296 spin_unlock(&root->orphan_lock);
2298 if (trans && delete_item) {
2299 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2304 btrfs_orphan_release_metadata(inode);
2310 * this cleans up any orphans that may be left on the list from the last use
2313 int btrfs_orphan_cleanup(struct btrfs_root *root)
2315 struct btrfs_path *path;
2316 struct extent_buffer *leaf;
2317 struct btrfs_key key, found_key;
2318 struct btrfs_trans_handle *trans;
2319 struct inode *inode;
2320 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2322 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2325 path = btrfs_alloc_path();
2332 key.objectid = BTRFS_ORPHAN_OBJECTID;
2333 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2334 key.offset = (u64)-1;
2337 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2342 * if ret == 0 means we found what we were searching for, which
2343 * is weird, but possible, so only screw with path if we didn't
2344 * find the key and see if we have stuff that matches
2348 if (path->slots[0] == 0)
2353 /* pull out the item */
2354 leaf = path->nodes[0];
2355 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2357 /* make sure the item matches what we want */
2358 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2360 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2363 /* release the path since we're done with it */
2364 btrfs_release_path(path);
2367 * this is where we are basically btrfs_lookup, without the
2368 * crossing root thing. we store the inode number in the
2369 * offset of the orphan item.
2371 found_key.objectid = found_key.offset;
2372 found_key.type = BTRFS_INODE_ITEM_KEY;
2373 found_key.offset = 0;
2374 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2375 if (IS_ERR(inode)) {
2376 ret = PTR_ERR(inode);
2381 * add this inode to the orphan list so btrfs_orphan_del does
2382 * the proper thing when we hit it
2384 spin_lock(&root->orphan_lock);
2385 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2386 spin_unlock(&root->orphan_lock);
2389 * if this is a bad inode, means we actually succeeded in
2390 * removing the inode, but not the orphan record, which means
2391 * we need to manually delete the orphan since iput will just
2392 * do a destroy_inode
2394 if (is_bad_inode(inode)) {
2395 trans = btrfs_start_transaction(root, 0);
2396 if (IS_ERR(trans)) {
2397 ret = PTR_ERR(trans);
2400 btrfs_orphan_del(trans, inode);
2401 btrfs_end_transaction(trans, root);
2406 /* if we have links, this was a truncate, lets do that */
2407 if (inode->i_nlink) {
2408 if (!S_ISREG(inode->i_mode)) {
2414 ret = btrfs_truncate(inode);
2419 /* this will do delete_inode and everything for us */
2424 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2426 if (root->orphan_block_rsv)
2427 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2430 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2431 trans = btrfs_join_transaction(root);
2433 btrfs_end_transaction(trans, root);
2437 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2439 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2443 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2444 btrfs_free_path(path);
2449 * very simple check to peek ahead in the leaf looking for xattrs. If we
2450 * don't find any xattrs, we know there can't be any acls.
2452 * slot is the slot the inode is in, objectid is the objectid of the inode
2454 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2455 int slot, u64 objectid)
2457 u32 nritems = btrfs_header_nritems(leaf);
2458 struct btrfs_key found_key;
2462 while (slot < nritems) {
2463 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2465 /* we found a different objectid, there must not be acls */
2466 if (found_key.objectid != objectid)
2469 /* we found an xattr, assume we've got an acl */
2470 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2474 * we found a key greater than an xattr key, there can't
2475 * be any acls later on
2477 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2484 * it goes inode, inode backrefs, xattrs, extents,
2485 * so if there are a ton of hard links to an inode there can
2486 * be a lot of backrefs. Don't waste time searching too hard,
2487 * this is just an optimization
2492 /* we hit the end of the leaf before we found an xattr or
2493 * something larger than an xattr. We have to assume the inode
2500 * read an inode from the btree into the in-memory inode
2502 static void btrfs_read_locked_inode(struct inode *inode)
2504 struct btrfs_path *path;
2505 struct extent_buffer *leaf;
2506 struct btrfs_inode_item *inode_item;
2507 struct btrfs_timespec *tspec;
2508 struct btrfs_root *root = BTRFS_I(inode)->root;
2509 struct btrfs_key location;
2513 bool filled = false;
2515 ret = btrfs_fill_inode(inode, &rdev);
2519 path = btrfs_alloc_path();
2523 path->leave_spinning = 1;
2524 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2526 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2530 leaf = path->nodes[0];
2535 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2536 struct btrfs_inode_item);
2537 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2538 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2539 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2540 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2541 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2543 tspec = btrfs_inode_atime(inode_item);
2544 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2545 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2547 tspec = btrfs_inode_mtime(inode_item);
2548 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2549 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2551 tspec = btrfs_inode_ctime(inode_item);
2552 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2553 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2555 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2556 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2557 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2558 inode->i_generation = BTRFS_I(inode)->generation;
2560 rdev = btrfs_inode_rdev(leaf, inode_item);
2562 BTRFS_I(inode)->index_cnt = (u64)-1;
2563 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2566 * try to precache a NULL acl entry for files that don't have
2567 * any xattrs or acls
2569 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2572 cache_no_acl(inode);
2574 btrfs_free_path(path);
2576 switch (inode->i_mode & S_IFMT) {
2578 inode->i_mapping->a_ops = &btrfs_aops;
2579 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2580 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2581 inode->i_fop = &btrfs_file_operations;
2582 inode->i_op = &btrfs_file_inode_operations;
2585 inode->i_fop = &btrfs_dir_file_operations;
2586 if (root == root->fs_info->tree_root)
2587 inode->i_op = &btrfs_dir_ro_inode_operations;
2589 inode->i_op = &btrfs_dir_inode_operations;
2592 inode->i_op = &btrfs_symlink_inode_operations;
2593 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2594 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2597 inode->i_op = &btrfs_special_inode_operations;
2598 init_special_inode(inode, inode->i_mode, rdev);
2602 btrfs_update_iflags(inode);
2606 btrfs_free_path(path);
2607 make_bad_inode(inode);
2611 * given a leaf and an inode, copy the inode fields into the leaf
2613 static void fill_inode_item(struct btrfs_trans_handle *trans,
2614 struct extent_buffer *leaf,
2615 struct btrfs_inode_item *item,
2616 struct inode *inode)
2618 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2619 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2620 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2621 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2622 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2624 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2625 inode->i_atime.tv_sec);
2626 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2627 inode->i_atime.tv_nsec);
2629 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2630 inode->i_mtime.tv_sec);
2631 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2632 inode->i_mtime.tv_nsec);
2634 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2635 inode->i_ctime.tv_sec);
2636 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2637 inode->i_ctime.tv_nsec);
2639 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2640 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2641 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2642 btrfs_set_inode_transid(leaf, item, trans->transid);
2643 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2644 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2645 btrfs_set_inode_block_group(leaf, item, 0);
2649 * copy everything in the in-memory inode into the btree.
2651 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2652 struct btrfs_root *root, struct inode *inode)
2654 struct btrfs_inode_item *inode_item;
2655 struct btrfs_path *path;
2656 struct extent_buffer *leaf;
2660 * If the inode is a free space inode, we can deadlock during commit
2661 * if we put it into the delayed code.
2663 * The data relocation inode should also be directly updated
2666 if (!btrfs_is_free_space_inode(root, inode)
2667 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2668 ret = btrfs_delayed_update_inode(trans, root, inode);
2670 btrfs_set_inode_last_trans(trans, inode);
2674 path = btrfs_alloc_path();
2678 path->leave_spinning = 1;
2679 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2687 btrfs_unlock_up_safe(path, 1);
2688 leaf = path->nodes[0];
2689 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2690 struct btrfs_inode_item);
2692 fill_inode_item(trans, leaf, inode_item, inode);
2693 btrfs_mark_buffer_dirty(leaf);
2694 btrfs_set_inode_last_trans(trans, inode);
2697 btrfs_free_path(path);
2702 * unlink helper that gets used here in inode.c and in the tree logging
2703 * recovery code. It remove a link in a directory with a given name, and
2704 * also drops the back refs in the inode to the directory
2706 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2707 struct btrfs_root *root,
2708 struct inode *dir, struct inode *inode,
2709 const char *name, int name_len)
2711 struct btrfs_path *path;
2713 struct extent_buffer *leaf;
2714 struct btrfs_dir_item *di;
2715 struct btrfs_key key;
2717 u64 ino = btrfs_ino(inode);
2718 u64 dir_ino = btrfs_ino(dir);
2720 path = btrfs_alloc_path();
2726 path->leave_spinning = 1;
2727 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2728 name, name_len, -1);
2737 leaf = path->nodes[0];
2738 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2739 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2742 btrfs_release_path(path);
2744 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2747 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2748 "inode %llu parent %llu\n", name_len, name,
2749 (unsigned long long)ino, (unsigned long long)dir_ino);
2753 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2757 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2759 BUG_ON(ret != 0 && ret != -ENOENT);
2761 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2766 btrfs_free_path(path);
2770 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2771 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2772 btrfs_update_inode(trans, root, dir);
2777 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2778 struct btrfs_root *root,
2779 struct inode *dir, struct inode *inode,
2780 const char *name, int name_len)
2783 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2785 btrfs_drop_nlink(inode);
2786 ret = btrfs_update_inode(trans, root, inode);
2792 /* helper to check if there is any shared block in the path */
2793 static int check_path_shared(struct btrfs_root *root,
2794 struct btrfs_path *path)
2796 struct extent_buffer *eb;
2800 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2803 if (!path->nodes[level])
2805 eb = path->nodes[level];
2806 if (!btrfs_block_can_be_shared(root, eb))
2808 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2817 * helper to start transaction for unlink and rmdir.
2819 * unlink and rmdir are special in btrfs, they do not always free space.
2820 * so in enospc case, we should make sure they will free space before
2821 * allowing them to use the global metadata reservation.
2823 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2824 struct dentry *dentry)
2826 struct btrfs_trans_handle *trans;
2827 struct btrfs_root *root = BTRFS_I(dir)->root;
2828 struct btrfs_path *path;
2829 struct btrfs_inode_ref *ref;
2830 struct btrfs_dir_item *di;
2831 struct inode *inode = dentry->d_inode;
2836 u64 ino = btrfs_ino(inode);
2837 u64 dir_ino = btrfs_ino(dir);
2839 trans = btrfs_start_transaction(root, 10);
2840 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2843 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2844 return ERR_PTR(-ENOSPC);
2846 /* check if there is someone else holds reference */
2847 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2848 return ERR_PTR(-ENOSPC);
2850 if (atomic_read(&inode->i_count) > 2)
2851 return ERR_PTR(-ENOSPC);
2853 if (xchg(&root->fs_info->enospc_unlink, 1))
2854 return ERR_PTR(-ENOSPC);
2856 path = btrfs_alloc_path();
2858 root->fs_info->enospc_unlink = 0;
2859 return ERR_PTR(-ENOMEM);
2862 trans = btrfs_start_transaction(root, 0);
2863 if (IS_ERR(trans)) {
2864 btrfs_free_path(path);
2865 root->fs_info->enospc_unlink = 0;
2869 path->skip_locking = 1;
2870 path->search_commit_root = 1;
2872 ret = btrfs_lookup_inode(trans, root, path,
2873 &BTRFS_I(dir)->location, 0);
2879 if (check_path_shared(root, path))
2884 btrfs_release_path(path);
2886 ret = btrfs_lookup_inode(trans, root, path,
2887 &BTRFS_I(inode)->location, 0);
2893 if (check_path_shared(root, path))
2898 btrfs_release_path(path);
2900 if (ret == 0 && S_ISREG(inode->i_mode)) {
2901 ret = btrfs_lookup_file_extent(trans, root, path,
2908 if (check_path_shared(root, path))
2910 btrfs_release_path(path);
2918 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2919 dentry->d_name.name, dentry->d_name.len, 0);
2925 if (check_path_shared(root, path))
2931 btrfs_release_path(path);
2933 ref = btrfs_lookup_inode_ref(trans, root, path,
2934 dentry->d_name.name, dentry->d_name.len,
2941 if (check_path_shared(root, path))
2943 index = btrfs_inode_ref_index(path->nodes[0], ref);
2944 btrfs_release_path(path);
2947 * This is a commit root search, if we can lookup inode item and other
2948 * relative items in the commit root, it means the transaction of
2949 * dir/file creation has been committed, and the dir index item that we
2950 * delay to insert has also been inserted into the commit root. So
2951 * we needn't worry about the delayed insertion of the dir index item
2954 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2955 dentry->d_name.name, dentry->d_name.len, 0);
2960 BUG_ON(ret == -ENOENT);
2961 if (check_path_shared(root, path))
2966 btrfs_free_path(path);
2968 btrfs_end_transaction(trans, root);
2969 root->fs_info->enospc_unlink = 0;
2970 return ERR_PTR(err);
2973 trans->block_rsv = &root->fs_info->global_block_rsv;
2977 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2978 struct btrfs_root *root)
2980 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2981 BUG_ON(!root->fs_info->enospc_unlink);
2982 root->fs_info->enospc_unlink = 0;
2984 btrfs_end_transaction_throttle(trans, root);
2987 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2989 struct btrfs_root *root = BTRFS_I(dir)->root;
2990 struct btrfs_trans_handle *trans;
2991 struct inode *inode = dentry->d_inode;
2993 unsigned long nr = 0;
2995 trans = __unlink_start_trans(dir, dentry);
2997 return PTR_ERR(trans);
2999 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3001 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3002 dentry->d_name.name, dentry->d_name.len);
3006 if (inode->i_nlink == 0) {
3007 ret = btrfs_orphan_add(trans, inode);
3013 nr = trans->blocks_used;
3014 __unlink_end_trans(trans, root);
3015 btrfs_btree_balance_dirty(root, nr);
3019 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3020 struct btrfs_root *root,
3021 struct inode *dir, u64 objectid,
3022 const char *name, int name_len)
3024 struct btrfs_path *path;
3025 struct extent_buffer *leaf;
3026 struct btrfs_dir_item *di;
3027 struct btrfs_key key;
3030 u64 dir_ino = btrfs_ino(dir);
3032 path = btrfs_alloc_path();
3036 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3037 name, name_len, -1);
3038 BUG_ON(IS_ERR_OR_NULL(di));
3040 leaf = path->nodes[0];
3041 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3042 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3043 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3045 btrfs_release_path(path);
3047 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3048 objectid, root->root_key.objectid,
3049 dir_ino, &index, name, name_len);
3051 BUG_ON(ret != -ENOENT);
3052 di = btrfs_search_dir_index_item(root, path, dir_ino,
3054 BUG_ON(IS_ERR_OR_NULL(di));
3056 leaf = path->nodes[0];
3057 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3058 btrfs_release_path(path);
3061 btrfs_release_path(path);
3063 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3066 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3067 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3068 ret = btrfs_update_inode(trans, root, dir);
3071 btrfs_free_path(path);
3075 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3077 struct inode *inode = dentry->d_inode;
3079 struct btrfs_root *root = BTRFS_I(dir)->root;
3080 struct btrfs_trans_handle *trans;
3081 unsigned long nr = 0;
3083 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3084 btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3087 trans = __unlink_start_trans(dir, dentry);
3089 return PTR_ERR(trans);
3091 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3092 err = btrfs_unlink_subvol(trans, root, dir,
3093 BTRFS_I(inode)->location.objectid,
3094 dentry->d_name.name,
3095 dentry->d_name.len);
3099 err = btrfs_orphan_add(trans, inode);
3103 /* now the directory is empty */
3104 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3105 dentry->d_name.name, dentry->d_name.len);
3107 btrfs_i_size_write(inode, 0);
3109 nr = trans->blocks_used;
3110 __unlink_end_trans(trans, root);
3111 btrfs_btree_balance_dirty(root, nr);
3117 * this can truncate away extent items, csum items and directory items.
3118 * It starts at a high offset and removes keys until it can't find
3119 * any higher than new_size
3121 * csum items that cross the new i_size are truncated to the new size
3124 * min_type is the minimum key type to truncate down to. If set to 0, this
3125 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3127 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3128 struct btrfs_root *root,
3129 struct inode *inode,
3130 u64 new_size, u32 min_type)
3132 struct btrfs_path *path;
3133 struct extent_buffer *leaf;
3134 struct btrfs_file_extent_item *fi;
3135 struct btrfs_key key;
3136 struct btrfs_key found_key;
3137 u64 extent_start = 0;
3138 u64 extent_num_bytes = 0;
3139 u64 extent_offset = 0;
3141 u64 mask = root->sectorsize - 1;
3142 u32 found_type = (u8)-1;
3145 int pending_del_nr = 0;
3146 int pending_del_slot = 0;
3147 int extent_type = -1;
3151 u64 ino = btrfs_ino(inode);
3153 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3155 path = btrfs_alloc_path();
3160 if (root->ref_cows || root == root->fs_info->tree_root)
3161 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3164 * This function is also used to drop the items in the log tree before
3165 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3166 * it is used to drop the loged items. So we shouldn't kill the delayed
3169 if (min_type == 0 && root == BTRFS_I(inode)->root)
3170 btrfs_kill_delayed_inode_items(inode);
3173 key.offset = (u64)-1;
3177 path->leave_spinning = 1;
3178 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3185 /* there are no items in the tree for us to truncate, we're
3188 if (path->slots[0] == 0)
3195 leaf = path->nodes[0];
3196 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3197 found_type = btrfs_key_type(&found_key);
3200 if (found_key.objectid != ino)
3203 if (found_type < min_type)
3206 item_end = found_key.offset;
3207 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3208 fi = btrfs_item_ptr(leaf, path->slots[0],
3209 struct btrfs_file_extent_item);
3210 extent_type = btrfs_file_extent_type(leaf, fi);
3211 encoding = btrfs_file_extent_compression(leaf, fi);
3212 encoding |= btrfs_file_extent_encryption(leaf, fi);
3213 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3215 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3217 btrfs_file_extent_num_bytes(leaf, fi);
3218 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3219 item_end += btrfs_file_extent_inline_len(leaf,
3224 if (found_type > min_type) {
3227 if (item_end < new_size)
3229 if (found_key.offset >= new_size)
3235 /* FIXME, shrink the extent if the ref count is only 1 */
3236 if (found_type != BTRFS_EXTENT_DATA_KEY)
3239 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3241 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3242 if (!del_item && !encoding) {
3243 u64 orig_num_bytes =
3244 btrfs_file_extent_num_bytes(leaf, fi);
3245 extent_num_bytes = new_size -
3246 found_key.offset + root->sectorsize - 1;
3247 extent_num_bytes = extent_num_bytes &
3248 ~((u64)root->sectorsize - 1);
3249 btrfs_set_file_extent_num_bytes(leaf, fi,
3251 num_dec = (orig_num_bytes -
3253 if (root->ref_cows && extent_start != 0)
3254 inode_sub_bytes(inode, num_dec);
3255 btrfs_mark_buffer_dirty(leaf);
3258 btrfs_file_extent_disk_num_bytes(leaf,
3260 extent_offset = found_key.offset -
3261 btrfs_file_extent_offset(leaf, fi);
3263 /* FIXME blocksize != 4096 */
3264 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3265 if (extent_start != 0) {
3268 inode_sub_bytes(inode, num_dec);
3271 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3273 * we can't truncate inline items that have had
3277 btrfs_file_extent_compression(leaf, fi) == 0 &&
3278 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3279 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3280 u32 size = new_size - found_key.offset;
3282 if (root->ref_cows) {
3283 inode_sub_bytes(inode, item_end + 1 -
3287 btrfs_file_extent_calc_inline_size(size);
3288 ret = btrfs_truncate_item(trans, root, path,
3290 } else if (root->ref_cows) {
3291 inode_sub_bytes(inode, item_end + 1 -
3297 if (!pending_del_nr) {
3298 /* no pending yet, add ourselves */
3299 pending_del_slot = path->slots[0];
3301 } else if (pending_del_nr &&
3302 path->slots[0] + 1 == pending_del_slot) {
3303 /* hop on the pending chunk */
3305 pending_del_slot = path->slots[0];
3312 if (found_extent && (root->ref_cows ||
3313 root == root->fs_info->tree_root)) {
3314 btrfs_set_path_blocking(path);
3315 ret = btrfs_free_extent(trans, root, extent_start,
3316 extent_num_bytes, 0,
3317 btrfs_header_owner(leaf),
3318 ino, extent_offset);
3322 if (found_type == BTRFS_INODE_ITEM_KEY)
3325 if (path->slots[0] == 0 ||
3326 path->slots[0] != pending_del_slot) {
3327 if (root->ref_cows &&
3328 BTRFS_I(inode)->location.objectid !=
3329 BTRFS_FREE_INO_OBJECTID) {
3333 if (pending_del_nr) {
3334 ret = btrfs_del_items(trans, root, path,
3340 btrfs_release_path(path);
3347 if (pending_del_nr) {
3348 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3352 btrfs_free_path(path);
3357 * taken from block_truncate_page, but does cow as it zeros out
3358 * any bytes left in the last page in the file.
3360 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3362 struct inode *inode = mapping->host;
3363 struct btrfs_root *root = BTRFS_I(inode)->root;
3364 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3365 struct btrfs_ordered_extent *ordered;
3366 struct extent_state *cached_state = NULL;
3368 u32 blocksize = root->sectorsize;
3369 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3370 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3376 if ((offset & (blocksize - 1)) == 0)
3378 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3384 page = find_or_create_page(mapping, index, GFP_NOFS);
3386 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3390 page_start = page_offset(page);
3391 page_end = page_start + PAGE_CACHE_SIZE - 1;
3393 if (!PageUptodate(page)) {
3394 ret = btrfs_readpage(NULL, page);
3396 if (page->mapping != mapping) {
3398 page_cache_release(page);
3401 if (!PageUptodate(page)) {
3406 wait_on_page_writeback(page);
3408 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3410 set_page_extent_mapped(page);
3412 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3414 unlock_extent_cached(io_tree, page_start, page_end,
3415 &cached_state, GFP_NOFS);
3417 page_cache_release(page);
3418 btrfs_start_ordered_extent(inode, ordered, 1);
3419 btrfs_put_ordered_extent(ordered);
3423 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3424 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3425 0, 0, &cached_state, GFP_NOFS);
3427 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3430 unlock_extent_cached(io_tree, page_start, page_end,
3431 &cached_state, GFP_NOFS);
3436 if (offset != PAGE_CACHE_SIZE) {
3438 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3439 flush_dcache_page(page);
3442 ClearPageChecked(page);
3443 set_page_dirty(page);
3444 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3449 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3451 page_cache_release(page);
3457 * This function puts in dummy file extents for the area we're creating a hole
3458 * for. So if we are truncating this file to a larger size we need to insert
3459 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3460 * the range between oldsize and size
3462 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3464 struct btrfs_trans_handle *trans;
3465 struct btrfs_root *root = BTRFS_I(inode)->root;
3466 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3467 struct extent_map *em = NULL;
3468 struct extent_state *cached_state = NULL;
3469 u64 mask = root->sectorsize - 1;
3470 u64 hole_start = (oldsize + mask) & ~mask;
3471 u64 block_end = (size + mask) & ~mask;
3477 if (size <= hole_start)
3481 struct btrfs_ordered_extent *ordered;
3482 btrfs_wait_ordered_range(inode, hole_start,
3483 block_end - hole_start);
3484 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3485 &cached_state, GFP_NOFS);
3486 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3489 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3490 &cached_state, GFP_NOFS);
3491 btrfs_put_ordered_extent(ordered);
3494 cur_offset = hole_start;
3496 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3497 block_end - cur_offset, 0);
3498 BUG_ON(IS_ERR_OR_NULL(em));
3499 last_byte = min(extent_map_end(em), block_end);
3500 last_byte = (last_byte + mask) & ~mask;
3501 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3503 hole_size = last_byte - cur_offset;
3505 trans = btrfs_start_transaction(root, 2);
3506 if (IS_ERR(trans)) {
3507 err = PTR_ERR(trans);
3511 err = btrfs_drop_extents(trans, inode, cur_offset,
3512 cur_offset + hole_size,
3515 btrfs_end_transaction(trans, root);
3519 err = btrfs_insert_file_extent(trans, root,
3520 btrfs_ino(inode), cur_offset, 0,
3521 0, hole_size, 0, hole_size,
3524 btrfs_end_transaction(trans, root);
3528 btrfs_drop_extent_cache(inode, hole_start,
3531 btrfs_end_transaction(trans, root);
3533 free_extent_map(em);
3535 cur_offset = last_byte;
3536 if (cur_offset >= block_end)
3540 free_extent_map(em);
3541 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3546 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3548 loff_t oldsize = i_size_read(inode);
3551 if (newsize == oldsize)
3554 if (newsize > oldsize) {
3555 i_size_write(inode, newsize);
3556 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3557 truncate_pagecache(inode, oldsize, newsize);
3558 ret = btrfs_cont_expand(inode, oldsize, newsize);
3560 btrfs_setsize(inode, oldsize);
3564 mark_inode_dirty(inode);
3568 * We're truncating a file that used to have good data down to
3569 * zero. Make sure it gets into the ordered flush list so that
3570 * any new writes get down to disk quickly.
3573 BTRFS_I(inode)->ordered_data_close = 1;
3575 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3576 truncate_setsize(inode, newsize);
3577 ret = btrfs_truncate(inode);
3583 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3585 struct inode *inode = dentry->d_inode;
3586 struct btrfs_root *root = BTRFS_I(inode)->root;
3589 if (btrfs_root_readonly(root))
3592 err = inode_change_ok(inode, attr);
3596 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3597 err = btrfs_setsize(inode, attr->ia_size);
3602 if (attr->ia_valid) {
3603 setattr_copy(inode, attr);
3604 mark_inode_dirty(inode);
3606 if (attr->ia_valid & ATTR_MODE)
3607 err = btrfs_acl_chmod(inode);
3613 void btrfs_evict_inode(struct inode *inode)
3615 struct btrfs_trans_handle *trans;
3616 struct btrfs_root *root = BTRFS_I(inode)->root;
3620 trace_btrfs_inode_evict(inode);
3622 truncate_inode_pages(&inode->i_data, 0);
3623 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3624 btrfs_is_free_space_inode(root, inode)))
3627 if (is_bad_inode(inode)) {
3628 btrfs_orphan_del(NULL, inode);
3631 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3632 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3634 if (root->fs_info->log_root_recovering) {
3635 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3639 if (inode->i_nlink > 0) {
3640 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3644 btrfs_i_size_write(inode, 0);
3647 trans = btrfs_join_transaction(root);
3648 BUG_ON(IS_ERR(trans));
3649 trans->block_rsv = root->orphan_block_rsv;
3651 ret = btrfs_block_rsv_check(trans, root,
3652 root->orphan_block_rsv, 0, 5);
3654 BUG_ON(ret != -EAGAIN);
3655 ret = btrfs_commit_transaction(trans, root);
3660 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3664 nr = trans->blocks_used;
3665 btrfs_end_transaction(trans, root);
3667 btrfs_btree_balance_dirty(root, nr);
3672 ret = btrfs_orphan_del(trans, inode);
3676 if (!(root == root->fs_info->tree_root ||
3677 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3678 btrfs_return_ino(root, btrfs_ino(inode));
3680 nr = trans->blocks_used;
3681 btrfs_end_transaction(trans, root);
3682 btrfs_btree_balance_dirty(root, nr);
3684 end_writeback(inode);
3689 * this returns the key found in the dir entry in the location pointer.
3690 * If no dir entries were found, location->objectid is 0.
3692 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3693 struct btrfs_key *location)
3695 const char *name = dentry->d_name.name;
3696 int namelen = dentry->d_name.len;
3697 struct btrfs_dir_item *di;
3698 struct btrfs_path *path;
3699 struct btrfs_root *root = BTRFS_I(dir)->root;
3702 path = btrfs_alloc_path();
3706 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3711 if (IS_ERR_OR_NULL(di))
3714 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3716 btrfs_free_path(path);
3719 location->objectid = 0;
3724 * when we hit a tree root in a directory, the btrfs part of the inode
3725 * needs to be changed to reflect the root directory of the tree root. This
3726 * is kind of like crossing a mount point.
3728 static int fixup_tree_root_location(struct btrfs_root *root,
3730 struct dentry *dentry,
3731 struct btrfs_key *location,
3732 struct btrfs_root **sub_root)
3734 struct btrfs_path *path;
3735 struct btrfs_root *new_root;
3736 struct btrfs_root_ref *ref;
3737 struct extent_buffer *leaf;
3741 path = btrfs_alloc_path();
3748 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3749 BTRFS_I(dir)->root->root_key.objectid,
3750 location->objectid);
3757 leaf = path->nodes[0];
3758 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3759 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3760 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3763 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3764 (unsigned long)(ref + 1),
3765 dentry->d_name.len);
3769 btrfs_release_path(path);
3771 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3772 if (IS_ERR(new_root)) {
3773 err = PTR_ERR(new_root);
3777 if (btrfs_root_refs(&new_root->root_item) == 0) {
3782 *sub_root = new_root;
3783 location->objectid = btrfs_root_dirid(&new_root->root_item);
3784 location->type = BTRFS_INODE_ITEM_KEY;
3785 location->offset = 0;
3788 btrfs_free_path(path);
3792 static void inode_tree_add(struct inode *inode)
3794 struct btrfs_root *root = BTRFS_I(inode)->root;
3795 struct btrfs_inode *entry;
3797 struct rb_node *parent;
3798 u64 ino = btrfs_ino(inode);
3800 p = &root->inode_tree.rb_node;
3803 if (inode_unhashed(inode))
3806 spin_lock(&root->inode_lock);
3809 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3811 if (ino < btrfs_ino(&entry->vfs_inode))
3812 p = &parent->rb_left;
3813 else if (ino > btrfs_ino(&entry->vfs_inode))
3814 p = &parent->rb_right;
3816 WARN_ON(!(entry->vfs_inode.i_state &
3817 (I_WILL_FREE | I_FREEING)));
3818 rb_erase(parent, &root->inode_tree);
3819 RB_CLEAR_NODE(parent);
3820 spin_unlock(&root->inode_lock);
3824 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3825 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3826 spin_unlock(&root->inode_lock);
3829 static void inode_tree_del(struct inode *inode)
3831 struct btrfs_root *root = BTRFS_I(inode)->root;
3834 spin_lock(&root->inode_lock);
3835 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3836 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3837 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3838 empty = RB_EMPTY_ROOT(&root->inode_tree);
3840 spin_unlock(&root->inode_lock);
3843 * Free space cache has inodes in the tree root, but the tree root has a
3844 * root_refs of 0, so this could end up dropping the tree root as a
3845 * snapshot, so we need the extra !root->fs_info->tree_root check to
3846 * make sure we don't drop it.
3848 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3849 root != root->fs_info->tree_root) {
3850 synchronize_srcu(&root->fs_info->subvol_srcu);
3851 spin_lock(&root->inode_lock);
3852 empty = RB_EMPTY_ROOT(&root->inode_tree);
3853 spin_unlock(&root->inode_lock);
3855 btrfs_add_dead_root(root);
3859 int btrfs_invalidate_inodes(struct btrfs_root *root)
3861 struct rb_node *node;
3862 struct rb_node *prev;
3863 struct btrfs_inode *entry;
3864 struct inode *inode;
3867 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3869 spin_lock(&root->inode_lock);
3871 node = root->inode_tree.rb_node;
3875 entry = rb_entry(node, struct btrfs_inode, rb_node);
3877 if (objectid < btrfs_ino(&entry->vfs_inode))
3878 node = node->rb_left;
3879 else if (objectid > btrfs_ino(&entry->vfs_inode))
3880 node = node->rb_right;
3886 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3887 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3891 prev = rb_next(prev);
3895 entry = rb_entry(node, struct btrfs_inode, rb_node);
3896 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3897 inode = igrab(&entry->vfs_inode);
3899 spin_unlock(&root->inode_lock);
3900 if (atomic_read(&inode->i_count) > 1)
3901 d_prune_aliases(inode);
3903 * btrfs_drop_inode will have it removed from
3904 * the inode cache when its usage count
3909 spin_lock(&root->inode_lock);
3913 if (cond_resched_lock(&root->inode_lock))
3916 node = rb_next(node);
3918 spin_unlock(&root->inode_lock);
3922 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3924 struct btrfs_iget_args *args = p;
3925 inode->i_ino = args->ino;
3926 BTRFS_I(inode)->root = args->root;
3927 btrfs_set_inode_space_info(args->root, inode);
3931 static int btrfs_find_actor(struct inode *inode, void *opaque)
3933 struct btrfs_iget_args *args = opaque;
3934 return args->ino == btrfs_ino(inode) &&
3935 args->root == BTRFS_I(inode)->root;
3938 static struct inode *btrfs_iget_locked(struct super_block *s,
3940 struct btrfs_root *root)
3942 struct inode *inode;
3943 struct btrfs_iget_args args;
3944 args.ino = objectid;
3947 inode = iget5_locked(s, objectid, btrfs_find_actor,
3948 btrfs_init_locked_inode,
3953 /* Get an inode object given its location and corresponding root.
3954 * Returns in *is_new if the inode was read from disk
3956 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3957 struct btrfs_root *root, int *new)
3959 struct inode *inode;
3961 inode = btrfs_iget_locked(s, location->objectid, root);
3963 return ERR_PTR(-ENOMEM);
3965 if (inode->i_state & I_NEW) {
3966 BTRFS_I(inode)->root = root;
3967 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3968 btrfs_read_locked_inode(inode);
3969 if (!is_bad_inode(inode)) {
3970 inode_tree_add(inode);
3971 unlock_new_inode(inode);
3975 unlock_new_inode(inode);
3977 inode = ERR_PTR(-ESTALE);
3984 static struct inode *new_simple_dir(struct super_block *s,
3985 struct btrfs_key *key,
3986 struct btrfs_root *root)
3988 struct inode *inode = new_inode(s);
3991 return ERR_PTR(-ENOMEM);
3993 BTRFS_I(inode)->root = root;
3994 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3995 BTRFS_I(inode)->dummy_inode = 1;
3997 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3998 inode->i_op = &simple_dir_inode_operations;
3999 inode->i_fop = &simple_dir_operations;
4000 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4001 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4006 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4008 struct inode *inode;
4009 struct btrfs_root *root = BTRFS_I(dir)->root;
4010 struct btrfs_root *sub_root = root;
4011 struct btrfs_key location;
4015 if (dentry->d_name.len > BTRFS_NAME_LEN)
4016 return ERR_PTR(-ENAMETOOLONG);
4018 if (unlikely(d_need_lookup(dentry))) {
4019 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4020 kfree(dentry->d_fsdata);
4021 dentry->d_fsdata = NULL;
4022 /* This thing is hashed, drop it for now */
4025 ret = btrfs_inode_by_name(dir, dentry, &location);
4029 return ERR_PTR(ret);
4031 if (location.objectid == 0)
4034 if (location.type == BTRFS_INODE_ITEM_KEY) {
4035 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4039 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4041 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4042 ret = fixup_tree_root_location(root, dir, dentry,
4043 &location, &sub_root);
4046 inode = ERR_PTR(ret);
4048 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4050 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4052 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4054 if (!IS_ERR(inode) && root != sub_root) {
4055 down_read(&root->fs_info->cleanup_work_sem);
4056 if (!(inode->i_sb->s_flags & MS_RDONLY))
4057 ret = btrfs_orphan_cleanup(sub_root);
4058 up_read(&root->fs_info->cleanup_work_sem);
4060 inode = ERR_PTR(ret);
4066 static int btrfs_dentry_delete(const struct dentry *dentry)
4068 struct btrfs_root *root;
4070 if (!dentry->d_inode && !IS_ROOT(dentry))
4071 dentry = dentry->d_parent;
4073 if (dentry->d_inode) {
4074 root = BTRFS_I(dentry->d_inode)->root;
4075 if (btrfs_root_refs(&root->root_item) == 0)
4081 static void btrfs_dentry_release(struct dentry *dentry)
4083 if (dentry->d_fsdata)
4084 kfree(dentry->d_fsdata);
4087 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4088 struct nameidata *nd)
4092 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4093 if (unlikely(d_need_lookup(dentry))) {
4094 spin_lock(&dentry->d_lock);
4095 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4096 spin_unlock(&dentry->d_lock);
4101 unsigned char btrfs_filetype_table[] = {
4102 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4105 static int btrfs_real_readdir(struct file *filp, void *dirent,
4108 struct inode *inode = filp->f_dentry->d_inode;
4109 struct btrfs_root *root = BTRFS_I(inode)->root;
4110 struct btrfs_item *item;
4111 struct btrfs_dir_item *di;
4112 struct btrfs_key key;
4113 struct btrfs_key found_key;
4114 struct btrfs_path *path;
4115 struct list_head ins_list;
4116 struct list_head del_list;
4119 struct extent_buffer *leaf;
4121 unsigned char d_type;
4126 int key_type = BTRFS_DIR_INDEX_KEY;
4130 int is_curr = 0; /* filp->f_pos points to the current index? */
4132 /* FIXME, use a real flag for deciding about the key type */
4133 if (root->fs_info->tree_root == root)
4134 key_type = BTRFS_DIR_ITEM_KEY;
4136 /* special case for "." */
4137 if (filp->f_pos == 0) {
4138 over = filldir(dirent, ".", 1,
4139 filp->f_pos, btrfs_ino(inode), DT_DIR);
4144 /* special case for .., just use the back ref */
4145 if (filp->f_pos == 1) {
4146 u64 pino = parent_ino(filp->f_path.dentry);
4147 over = filldir(dirent, "..", 2,
4148 filp->f_pos, pino, DT_DIR);
4153 path = btrfs_alloc_path();
4159 if (key_type == BTRFS_DIR_INDEX_KEY) {
4160 INIT_LIST_HEAD(&ins_list);
4161 INIT_LIST_HEAD(&del_list);
4162 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4165 btrfs_set_key_type(&key, key_type);
4166 key.offset = filp->f_pos;
4167 key.objectid = btrfs_ino(inode);
4169 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4174 leaf = path->nodes[0];
4175 slot = path->slots[0];
4176 if (slot >= btrfs_header_nritems(leaf)) {
4177 ret = btrfs_next_leaf(root, path);
4185 item = btrfs_item_nr(leaf, slot);
4186 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4188 if (found_key.objectid != key.objectid)
4190 if (btrfs_key_type(&found_key) != key_type)
4192 if (found_key.offset < filp->f_pos)
4194 if (key_type == BTRFS_DIR_INDEX_KEY &&
4195 btrfs_should_delete_dir_index(&del_list,
4199 filp->f_pos = found_key.offset;
4202 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4204 di_total = btrfs_item_size(leaf, item);
4206 while (di_cur < di_total) {
4207 struct btrfs_key location;
4210 if (verify_dir_item(root, leaf, di))
4213 name_len = btrfs_dir_name_len(leaf, di);
4214 if (name_len <= sizeof(tmp_name)) {
4215 name_ptr = tmp_name;
4217 name_ptr = kmalloc(name_len, GFP_NOFS);
4223 read_extent_buffer(leaf, name_ptr,
4224 (unsigned long)(di + 1), name_len);
4226 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4227 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4231 q.hash = full_name_hash(q.name, q.len);
4232 tmp = d_lookup(filp->f_dentry, &q);
4234 struct btrfs_key *newkey;
4236 newkey = kzalloc(sizeof(struct btrfs_key),
4240 tmp = d_alloc(filp->f_dentry, &q);
4246 memcpy(newkey, &location,
4247 sizeof(struct btrfs_key));
4248 tmp->d_fsdata = newkey;
4249 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4256 /* is this a reference to our own snapshot? If so
4259 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4260 location.objectid == root->root_key.objectid) {
4264 over = filldir(dirent, name_ptr, name_len,
4265 found_key.offset, location.objectid,
4269 if (name_ptr != tmp_name)
4274 di_len = btrfs_dir_name_len(leaf, di) +
4275 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4277 di = (struct btrfs_dir_item *)((char *)di + di_len);
4283 if (key_type == BTRFS_DIR_INDEX_KEY) {
4286 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4292 /* Reached end of directory/root. Bump pos past the last item. */
4293 if (key_type == BTRFS_DIR_INDEX_KEY)
4295 * 32-bit glibc will use getdents64, but then strtol -
4296 * so the last number we can serve is this.
4298 filp->f_pos = 0x7fffffff;
4304 if (key_type == BTRFS_DIR_INDEX_KEY)
4305 btrfs_put_delayed_items(&ins_list, &del_list);
4306 btrfs_free_path(path);
4310 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4312 struct btrfs_root *root = BTRFS_I(inode)->root;
4313 struct btrfs_trans_handle *trans;
4315 bool nolock = false;
4317 if (BTRFS_I(inode)->dummy_inode)
4320 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4323 if (wbc->sync_mode == WB_SYNC_ALL) {
4325 trans = btrfs_join_transaction_nolock(root);
4327 trans = btrfs_join_transaction(root);
4329 return PTR_ERR(trans);
4331 ret = btrfs_end_transaction_nolock(trans, root);
4333 ret = btrfs_commit_transaction(trans, root);
4339 * This is somewhat expensive, updating the tree every time the
4340 * inode changes. But, it is most likely to find the inode in cache.
4341 * FIXME, needs more benchmarking...there are no reasons other than performance
4342 * to keep or drop this code.
4344 void btrfs_dirty_inode(struct inode *inode, int flags)
4346 struct btrfs_root *root = BTRFS_I(inode)->root;
4347 struct btrfs_trans_handle *trans;
4350 if (BTRFS_I(inode)->dummy_inode)
4353 trans = btrfs_join_transaction(root);
4354 BUG_ON(IS_ERR(trans));
4356 ret = btrfs_update_inode(trans, root, inode);
4357 if (ret && ret == -ENOSPC) {
4358 /* whoops, lets try again with the full transaction */
4359 btrfs_end_transaction(trans, root);
4360 trans = btrfs_start_transaction(root, 1);
4361 if (IS_ERR(trans)) {
4362 printk_ratelimited(KERN_ERR "btrfs: fail to "
4363 "dirty inode %llu error %ld\n",
4364 (unsigned long long)btrfs_ino(inode),
4369 ret = btrfs_update_inode(trans, root, inode);
4371 printk_ratelimited(KERN_ERR "btrfs: fail to "
4372 "dirty inode %llu error %d\n",
4373 (unsigned long long)btrfs_ino(inode),
4377 btrfs_end_transaction(trans, root);
4378 if (BTRFS_I(inode)->delayed_node)
4379 btrfs_balance_delayed_items(root);
4383 * find the highest existing sequence number in a directory
4384 * and then set the in-memory index_cnt variable to reflect
4385 * free sequence numbers
4387 static int btrfs_set_inode_index_count(struct inode *inode)
4389 struct btrfs_root *root = BTRFS_I(inode)->root;
4390 struct btrfs_key key, found_key;
4391 struct btrfs_path *path;
4392 struct extent_buffer *leaf;
4395 key.objectid = btrfs_ino(inode);
4396 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4397 key.offset = (u64)-1;
4399 path = btrfs_alloc_path();
4403 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4406 /* FIXME: we should be able to handle this */
4412 * MAGIC NUMBER EXPLANATION:
4413 * since we search a directory based on f_pos we have to start at 2
4414 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4415 * else has to start at 2
4417 if (path->slots[0] == 0) {
4418 BTRFS_I(inode)->index_cnt = 2;
4424 leaf = path->nodes[0];
4425 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4427 if (found_key.objectid != btrfs_ino(inode) ||
4428 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4429 BTRFS_I(inode)->index_cnt = 2;
4433 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4435 btrfs_free_path(path);
4440 * helper to find a free sequence number in a given directory. This current
4441 * code is very simple, later versions will do smarter things in the btree
4443 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4447 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4448 ret = btrfs_inode_delayed_dir_index_count(dir);
4450 ret = btrfs_set_inode_index_count(dir);
4456 *index = BTRFS_I(dir)->index_cnt;
4457 BTRFS_I(dir)->index_cnt++;
4462 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4463 struct btrfs_root *root,
4465 const char *name, int name_len,
4466 u64 ref_objectid, u64 objectid, int mode,
4469 struct inode *inode;
4470 struct btrfs_inode_item *inode_item;
4471 struct btrfs_key *location;
4472 struct btrfs_path *path;
4473 struct btrfs_inode_ref *ref;
4474 struct btrfs_key key[2];
4480 path = btrfs_alloc_path();
4482 return ERR_PTR(-ENOMEM);
4484 inode = new_inode(root->fs_info->sb);
4486 btrfs_free_path(path);
4487 return ERR_PTR(-ENOMEM);
4491 * we have to initialize this early, so we can reclaim the inode
4492 * number if we fail afterwards in this function.
4494 inode->i_ino = objectid;
4497 trace_btrfs_inode_request(dir);
4499 ret = btrfs_set_inode_index(dir, index);
4501 btrfs_free_path(path);
4503 return ERR_PTR(ret);
4507 * index_cnt is ignored for everything but a dir,
4508 * btrfs_get_inode_index_count has an explanation for the magic
4511 BTRFS_I(inode)->index_cnt = 2;
4512 BTRFS_I(inode)->root = root;
4513 BTRFS_I(inode)->generation = trans->transid;
4514 inode->i_generation = BTRFS_I(inode)->generation;
4515 btrfs_set_inode_space_info(root, inode);
4522 key[0].objectid = objectid;
4523 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4526 key[1].objectid = objectid;
4527 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4528 key[1].offset = ref_objectid;
4530 sizes[0] = sizeof(struct btrfs_inode_item);
4531 sizes[1] = name_len + sizeof(*ref);
4533 path->leave_spinning = 1;
4534 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4538 inode_init_owner(inode, dir, mode);
4539 inode_set_bytes(inode, 0);
4540 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4541 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4542 struct btrfs_inode_item);
4543 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4545 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4546 struct btrfs_inode_ref);
4547 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4548 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4549 ptr = (unsigned long)(ref + 1);
4550 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4552 btrfs_mark_buffer_dirty(path->nodes[0]);
4553 btrfs_free_path(path);
4555 location = &BTRFS_I(inode)->location;
4556 location->objectid = objectid;
4557 location->offset = 0;
4558 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4560 btrfs_inherit_iflags(inode, dir);
4562 if (S_ISREG(mode)) {
4563 if (btrfs_test_opt(root, NODATASUM))
4564 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4565 if (btrfs_test_opt(root, NODATACOW) ||
4566 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4567 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4570 insert_inode_hash(inode);
4571 inode_tree_add(inode);
4573 trace_btrfs_inode_new(inode);
4574 btrfs_set_inode_last_trans(trans, inode);
4579 BTRFS_I(dir)->index_cnt--;
4580 btrfs_free_path(path);
4582 return ERR_PTR(ret);
4585 static inline u8 btrfs_inode_type(struct inode *inode)
4587 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4591 * utility function to add 'inode' into 'parent_inode' with
4592 * a give name and a given sequence number.
4593 * if 'add_backref' is true, also insert a backref from the
4594 * inode to the parent directory.
4596 int btrfs_add_link(struct btrfs_trans_handle *trans,
4597 struct inode *parent_inode, struct inode *inode,
4598 const char *name, int name_len, int add_backref, u64 index)
4601 struct btrfs_key key;
4602 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4603 u64 ino = btrfs_ino(inode);
4604 u64 parent_ino = btrfs_ino(parent_inode);
4606 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4607 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4610 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4614 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4615 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4616 key.objectid, root->root_key.objectid,
4617 parent_ino, index, name, name_len);
4618 } else if (add_backref) {
4619 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4624 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4626 btrfs_inode_type(inode), index);
4629 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4631 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4632 ret = btrfs_update_inode(trans, root, parent_inode);
4637 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4638 struct inode *dir, struct dentry *dentry,
4639 struct inode *inode, int backref, u64 index)
4641 int err = btrfs_add_link(trans, dir, inode,
4642 dentry->d_name.name, dentry->d_name.len,
4645 d_instantiate(dentry, inode);
4653 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4654 int mode, dev_t rdev)
4656 struct btrfs_trans_handle *trans;
4657 struct btrfs_root *root = BTRFS_I(dir)->root;
4658 struct inode *inode = NULL;
4662 unsigned long nr = 0;
4665 if (!new_valid_dev(rdev))
4669 * 2 for inode item and ref
4671 * 1 for xattr if selinux is on
4673 trans = btrfs_start_transaction(root, 5);
4675 return PTR_ERR(trans);
4677 err = btrfs_find_free_ino(root, &objectid);
4681 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4682 dentry->d_name.len, btrfs_ino(dir), objectid,
4684 if (IS_ERR(inode)) {
4685 err = PTR_ERR(inode);
4689 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4695 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4699 inode->i_op = &btrfs_special_inode_operations;
4700 init_special_inode(inode, inode->i_mode, rdev);
4701 btrfs_update_inode(trans, root, inode);
4704 nr = trans->blocks_used;
4705 btrfs_end_transaction_throttle(trans, root);
4706 btrfs_btree_balance_dirty(root, nr);
4708 inode_dec_link_count(inode);
4714 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4715 int mode, struct nameidata *nd)
4717 struct btrfs_trans_handle *trans;
4718 struct btrfs_root *root = BTRFS_I(dir)->root;
4719 struct inode *inode = NULL;
4722 unsigned long nr = 0;
4727 * 2 for inode item and ref
4729 * 1 for xattr if selinux is on
4731 trans = btrfs_start_transaction(root, 5);
4733 return PTR_ERR(trans);
4735 err = btrfs_find_free_ino(root, &objectid);
4739 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4740 dentry->d_name.len, btrfs_ino(dir), objectid,
4742 if (IS_ERR(inode)) {
4743 err = PTR_ERR(inode);
4747 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4753 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4757 inode->i_mapping->a_ops = &btrfs_aops;
4758 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4759 inode->i_fop = &btrfs_file_operations;
4760 inode->i_op = &btrfs_file_inode_operations;
4761 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4764 nr = trans->blocks_used;
4765 btrfs_end_transaction_throttle(trans, root);
4767 inode_dec_link_count(inode);
4770 btrfs_btree_balance_dirty(root, nr);
4774 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4775 struct dentry *dentry)
4777 struct btrfs_trans_handle *trans;
4778 struct btrfs_root *root = BTRFS_I(dir)->root;
4779 struct inode *inode = old_dentry->d_inode;
4781 unsigned long nr = 0;
4785 /* do not allow sys_link's with other subvols of the same device */
4786 if (root->objectid != BTRFS_I(inode)->root->objectid)
4789 if (inode->i_nlink == ~0U)
4792 err = btrfs_set_inode_index(dir, &index);
4797 * 2 items for inode and inode ref
4798 * 2 items for dir items
4799 * 1 item for parent inode
4801 trans = btrfs_start_transaction(root, 5);
4802 if (IS_ERR(trans)) {
4803 err = PTR_ERR(trans);
4807 btrfs_inc_nlink(inode);
4808 inode->i_ctime = CURRENT_TIME;
4811 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4816 struct dentry *parent = dentry->d_parent;
4817 err = btrfs_update_inode(trans, root, inode);
4819 btrfs_log_new_name(trans, inode, NULL, parent);
4822 nr = trans->blocks_used;
4823 btrfs_end_transaction_throttle(trans, root);
4826 inode_dec_link_count(inode);
4829 btrfs_btree_balance_dirty(root, nr);
4833 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4835 struct inode *inode = NULL;
4836 struct btrfs_trans_handle *trans;
4837 struct btrfs_root *root = BTRFS_I(dir)->root;
4839 int drop_on_err = 0;
4842 unsigned long nr = 1;
4845 * 2 items for inode and ref
4846 * 2 items for dir items
4847 * 1 for xattr if selinux is on
4849 trans = btrfs_start_transaction(root, 5);
4851 return PTR_ERR(trans);
4853 err = btrfs_find_free_ino(root, &objectid);
4857 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4858 dentry->d_name.len, btrfs_ino(dir), objectid,
4859 S_IFDIR | mode, &index);
4860 if (IS_ERR(inode)) {
4861 err = PTR_ERR(inode);
4867 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4871 inode->i_op = &btrfs_dir_inode_operations;
4872 inode->i_fop = &btrfs_dir_file_operations;
4874 btrfs_i_size_write(inode, 0);
4875 err = btrfs_update_inode(trans, root, inode);
4879 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4880 dentry->d_name.len, 0, index);
4884 d_instantiate(dentry, inode);
4888 nr = trans->blocks_used;
4889 btrfs_end_transaction_throttle(trans, root);
4892 btrfs_btree_balance_dirty(root, nr);
4896 /* helper for btfs_get_extent. Given an existing extent in the tree,
4897 * and an extent that you want to insert, deal with overlap and insert
4898 * the new extent into the tree.
4900 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4901 struct extent_map *existing,
4902 struct extent_map *em,
4903 u64 map_start, u64 map_len)
4907 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4908 start_diff = map_start - em->start;
4909 em->start = map_start;
4911 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4912 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4913 em->block_start += start_diff;
4914 em->block_len -= start_diff;
4916 return add_extent_mapping(em_tree, em);
4919 static noinline int uncompress_inline(struct btrfs_path *path,
4920 struct inode *inode, struct page *page,
4921 size_t pg_offset, u64 extent_offset,
4922 struct btrfs_file_extent_item *item)
4925 struct extent_buffer *leaf = path->nodes[0];
4928 unsigned long inline_size;
4932 WARN_ON(pg_offset != 0);
4933 compress_type = btrfs_file_extent_compression(leaf, item);
4934 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4935 inline_size = btrfs_file_extent_inline_item_len(leaf,
4936 btrfs_item_nr(leaf, path->slots[0]));
4937 tmp = kmalloc(inline_size, GFP_NOFS);
4940 ptr = btrfs_file_extent_inline_start(item);
4942 read_extent_buffer(leaf, tmp, ptr, inline_size);
4944 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4945 ret = btrfs_decompress(compress_type, tmp, page,
4946 extent_offset, inline_size, max_size);
4948 char *kaddr = kmap_atomic(page, KM_USER0);
4949 unsigned long copy_size = min_t(u64,
4950 PAGE_CACHE_SIZE - pg_offset,
4951 max_size - extent_offset);
4952 memset(kaddr + pg_offset, 0, copy_size);
4953 kunmap_atomic(kaddr, KM_USER0);
4960 * a bit scary, this does extent mapping from logical file offset to the disk.
4961 * the ugly parts come from merging extents from the disk with the in-ram
4962 * representation. This gets more complex because of the data=ordered code,
4963 * where the in-ram extents might be locked pending data=ordered completion.
4965 * This also copies inline extents directly into the page.
4968 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4969 size_t pg_offset, u64 start, u64 len,
4975 u64 extent_start = 0;
4977 u64 objectid = btrfs_ino(inode);
4979 struct btrfs_path *path = NULL;
4980 struct btrfs_root *root = BTRFS_I(inode)->root;
4981 struct btrfs_file_extent_item *item;
4982 struct extent_buffer *leaf;
4983 struct btrfs_key found_key;
4984 struct extent_map *em = NULL;
4985 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4986 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4987 struct btrfs_trans_handle *trans = NULL;
4991 read_lock(&em_tree->lock);
4992 em = lookup_extent_mapping(em_tree, start, len);
4994 em->bdev = root->fs_info->fs_devices->latest_bdev;
4995 read_unlock(&em_tree->lock);
4998 if (em->start > start || em->start + em->len <= start)
4999 free_extent_map(em);
5000 else if (em->block_start == EXTENT_MAP_INLINE && page)
5001 free_extent_map(em);
5005 em = alloc_extent_map();
5010 em->bdev = root->fs_info->fs_devices->latest_bdev;
5011 em->start = EXTENT_MAP_HOLE;
5012 em->orig_start = EXTENT_MAP_HOLE;
5014 em->block_len = (u64)-1;
5017 path = btrfs_alloc_path();
5023 * Chances are we'll be called again, so go ahead and do
5029 ret = btrfs_lookup_file_extent(trans, root, path,
5030 objectid, start, trans != NULL);
5037 if (path->slots[0] == 0)
5042 leaf = path->nodes[0];
5043 item = btrfs_item_ptr(leaf, path->slots[0],
5044 struct btrfs_file_extent_item);
5045 /* are we inside the extent that was found? */
5046 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5047 found_type = btrfs_key_type(&found_key);
5048 if (found_key.objectid != objectid ||
5049 found_type != BTRFS_EXTENT_DATA_KEY) {
5053 found_type = btrfs_file_extent_type(leaf, item);
5054 extent_start = found_key.offset;
5055 compress_type = btrfs_file_extent_compression(leaf, item);
5056 if (found_type == BTRFS_FILE_EXTENT_REG ||
5057 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5058 extent_end = extent_start +
5059 btrfs_file_extent_num_bytes(leaf, item);
5060 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5062 size = btrfs_file_extent_inline_len(leaf, item);
5063 extent_end = (extent_start + size + root->sectorsize - 1) &
5064 ~((u64)root->sectorsize - 1);
5067 if (start >= extent_end) {
5069 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5070 ret = btrfs_next_leaf(root, path);
5077 leaf = path->nodes[0];
5079 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5080 if (found_key.objectid != objectid ||
5081 found_key.type != BTRFS_EXTENT_DATA_KEY)
5083 if (start + len <= found_key.offset)
5086 em->len = found_key.offset - start;
5090 if (found_type == BTRFS_FILE_EXTENT_REG ||
5091 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5092 em->start = extent_start;
5093 em->len = extent_end - extent_start;
5094 em->orig_start = extent_start -
5095 btrfs_file_extent_offset(leaf, item);
5096 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5098 em->block_start = EXTENT_MAP_HOLE;
5101 if (compress_type != BTRFS_COMPRESS_NONE) {
5102 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5103 em->compress_type = compress_type;
5104 em->block_start = bytenr;
5105 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5108 bytenr += btrfs_file_extent_offset(leaf, item);
5109 em->block_start = bytenr;
5110 em->block_len = em->len;
5111 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5112 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5115 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5119 size_t extent_offset;
5122 em->block_start = EXTENT_MAP_INLINE;
5123 if (!page || create) {
5124 em->start = extent_start;
5125 em->len = extent_end - extent_start;
5129 size = btrfs_file_extent_inline_len(leaf, item);
5130 extent_offset = page_offset(page) + pg_offset - extent_start;
5131 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5132 size - extent_offset);
5133 em->start = extent_start + extent_offset;
5134 em->len = (copy_size + root->sectorsize - 1) &
5135 ~((u64)root->sectorsize - 1);
5136 em->orig_start = EXTENT_MAP_INLINE;
5137 if (compress_type) {
5138 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5139 em->compress_type = compress_type;
5141 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5142 if (create == 0 && !PageUptodate(page)) {
5143 if (btrfs_file_extent_compression(leaf, item) !=
5144 BTRFS_COMPRESS_NONE) {
5145 ret = uncompress_inline(path, inode, page,
5147 extent_offset, item);
5151 read_extent_buffer(leaf, map + pg_offset, ptr,
5153 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5154 memset(map + pg_offset + copy_size, 0,
5155 PAGE_CACHE_SIZE - pg_offset -
5160 flush_dcache_page(page);
5161 } else if (create && PageUptodate(page)) {
5165 free_extent_map(em);
5168 btrfs_release_path(path);
5169 trans = btrfs_join_transaction(root);
5172 return ERR_CAST(trans);
5176 write_extent_buffer(leaf, map + pg_offset, ptr,
5179 btrfs_mark_buffer_dirty(leaf);
5181 set_extent_uptodate(io_tree, em->start,
5182 extent_map_end(em) - 1, NULL, GFP_NOFS);
5185 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5192 em->block_start = EXTENT_MAP_HOLE;
5193 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5195 btrfs_release_path(path);
5196 if (em->start > start || extent_map_end(em) <= start) {
5197 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5198 "[%llu %llu]\n", (unsigned long long)em->start,
5199 (unsigned long long)em->len,
5200 (unsigned long long)start,
5201 (unsigned long long)len);
5207 write_lock(&em_tree->lock);
5208 ret = add_extent_mapping(em_tree, em);
5209 /* it is possible that someone inserted the extent into the tree
5210 * while we had the lock dropped. It is also possible that
5211 * an overlapping map exists in the tree
5213 if (ret == -EEXIST) {
5214 struct extent_map *existing;
5218 existing = lookup_extent_mapping(em_tree, start, len);
5219 if (existing && (existing->start > start ||
5220 existing->start + existing->len <= start)) {
5221 free_extent_map(existing);
5225 existing = lookup_extent_mapping(em_tree, em->start,
5228 err = merge_extent_mapping(em_tree, existing,
5231 free_extent_map(existing);
5233 free_extent_map(em);
5238 free_extent_map(em);
5242 free_extent_map(em);
5247 write_unlock(&em_tree->lock);
5250 trace_btrfs_get_extent(root, em);
5253 btrfs_free_path(path);
5255 ret = btrfs_end_transaction(trans, root);
5260 free_extent_map(em);
5261 return ERR_PTR(err);
5266 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5267 size_t pg_offset, u64 start, u64 len,
5270 struct extent_map *em;
5271 struct extent_map *hole_em = NULL;
5272 u64 range_start = start;
5278 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5283 * if our em maps to a hole, there might
5284 * actually be delalloc bytes behind it
5286 if (em->block_start != EXTENT_MAP_HOLE)
5292 /* check to see if we've wrapped (len == -1 or similar) */
5301 /* ok, we didn't find anything, lets look for delalloc */
5302 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5303 end, len, EXTENT_DELALLOC, 1);
5304 found_end = range_start + found;
5305 if (found_end < range_start)
5306 found_end = (u64)-1;
5309 * we didn't find anything useful, return
5310 * the original results from get_extent()
5312 if (range_start > end || found_end <= start) {
5318 /* adjust the range_start to make sure it doesn't
5319 * go backwards from the start they passed in
5321 range_start = max(start,range_start);
5322 found = found_end - range_start;
5325 u64 hole_start = start;
5328 em = alloc_extent_map();
5334 * when btrfs_get_extent can't find anything it
5335 * returns one huge hole
5337 * make sure what it found really fits our range, and
5338 * adjust to make sure it is based on the start from
5342 u64 calc_end = extent_map_end(hole_em);
5344 if (calc_end <= start || (hole_em->start > end)) {
5345 free_extent_map(hole_em);
5348 hole_start = max(hole_em->start, start);
5349 hole_len = calc_end - hole_start;
5353 if (hole_em && range_start > hole_start) {
5354 /* our hole starts before our delalloc, so we
5355 * have to return just the parts of the hole
5356 * that go until the delalloc starts
5358 em->len = min(hole_len,
5359 range_start - hole_start);
5360 em->start = hole_start;
5361 em->orig_start = hole_start;
5363 * don't adjust block start at all,
5364 * it is fixed at EXTENT_MAP_HOLE
5366 em->block_start = hole_em->block_start;
5367 em->block_len = hole_len;
5369 em->start = range_start;
5371 em->orig_start = range_start;
5372 em->block_start = EXTENT_MAP_DELALLOC;
5373 em->block_len = found;
5375 } else if (hole_em) {
5380 free_extent_map(hole_em);
5382 free_extent_map(em);
5383 return ERR_PTR(err);
5388 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5389 struct extent_map *em,
5392 struct btrfs_root *root = BTRFS_I(inode)->root;
5393 struct btrfs_trans_handle *trans;
5394 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5395 struct btrfs_key ins;
5398 bool insert = false;
5401 * Ok if the extent map we looked up is a hole and is for the exact
5402 * range we want, there is no reason to allocate a new one, however if
5403 * it is not right then we need to free this one and drop the cache for
5406 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5408 free_extent_map(em);
5411 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5414 trans = btrfs_join_transaction(root);
5416 return ERR_CAST(trans);
5418 if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5419 btrfs_add_inode_defrag(trans, inode);
5421 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5423 alloc_hint = get_extent_allocation_hint(inode, start, len);
5424 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5425 alloc_hint, (u64)-1, &ins, 1);
5432 em = alloc_extent_map();
5434 em = ERR_PTR(-ENOMEM);
5440 em->orig_start = em->start;
5441 em->len = ins.offset;
5443 em->block_start = ins.objectid;
5444 em->block_len = ins.offset;
5445 em->bdev = root->fs_info->fs_devices->latest_bdev;
5448 * We need to do this because if we're using the original em we searched
5449 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5452 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5455 write_lock(&em_tree->lock);
5456 ret = add_extent_mapping(em_tree, em);
5457 write_unlock(&em_tree->lock);
5460 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5463 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5464 ins.offset, ins.offset, 0);
5466 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5470 btrfs_end_transaction(trans, root);
5475 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5476 * block must be cow'd
5478 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5479 struct inode *inode, u64 offset, u64 len)
5481 struct btrfs_path *path;
5483 struct extent_buffer *leaf;
5484 struct btrfs_root *root = BTRFS_I(inode)->root;
5485 struct btrfs_file_extent_item *fi;
5486 struct btrfs_key key;
5494 path = btrfs_alloc_path();
5498 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5503 slot = path->slots[0];
5506 /* can't find the item, must cow */
5513 leaf = path->nodes[0];
5514 btrfs_item_key_to_cpu(leaf, &key, slot);
5515 if (key.objectid != btrfs_ino(inode) ||
5516 key.type != BTRFS_EXTENT_DATA_KEY) {
5517 /* not our file or wrong item type, must cow */
5521 if (key.offset > offset) {
5522 /* Wrong offset, must cow */
5526 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5527 found_type = btrfs_file_extent_type(leaf, fi);
5528 if (found_type != BTRFS_FILE_EXTENT_REG &&
5529 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5530 /* not a regular extent, must cow */
5533 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5534 backref_offset = btrfs_file_extent_offset(leaf, fi);
5536 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5537 if (extent_end < offset + len) {
5538 /* extent doesn't include our full range, must cow */
5542 if (btrfs_extent_readonly(root, disk_bytenr))
5546 * look for other files referencing this extent, if we
5547 * find any we must cow
5549 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5550 key.offset - backref_offset, disk_bytenr))
5554 * adjust disk_bytenr and num_bytes to cover just the bytes
5555 * in this extent we are about to write. If there
5556 * are any csums in that range we have to cow in order
5557 * to keep the csums correct
5559 disk_bytenr += backref_offset;
5560 disk_bytenr += offset - key.offset;
5561 num_bytes = min(offset + len, extent_end) - offset;
5562 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5565 * all of the above have passed, it is safe to overwrite this extent
5570 btrfs_free_path(path);
5574 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5575 struct buffer_head *bh_result, int create)
5577 struct extent_map *em;
5578 struct btrfs_root *root = BTRFS_I(inode)->root;
5579 u64 start = iblock << inode->i_blkbits;
5580 u64 len = bh_result->b_size;
5581 struct btrfs_trans_handle *trans;
5583 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5588 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5589 * io. INLINE is special, and we could probably kludge it in here, but
5590 * it's still buffered so for safety lets just fall back to the generic
5593 * For COMPRESSED we _have_ to read the entire extent in so we can
5594 * decompress it, so there will be buffering required no matter what we
5595 * do, so go ahead and fallback to buffered.
5597 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5598 * to buffered IO. Don't blame me, this is the price we pay for using
5601 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5602 em->block_start == EXTENT_MAP_INLINE) {
5603 free_extent_map(em);
5607 /* Just a good old fashioned hole, return */
5608 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5609 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5610 free_extent_map(em);
5611 /* DIO will do one hole at a time, so just unlock a sector */
5612 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5613 start + root->sectorsize - 1, GFP_NOFS);
5618 * We don't allocate a new extent in the following cases
5620 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5622 * 2) The extent is marked as PREALLOC. We're good to go here and can
5623 * just use the extent.
5627 len = em->len - (start - em->start);
5631 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5632 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5633 em->block_start != EXTENT_MAP_HOLE)) {
5638 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5639 type = BTRFS_ORDERED_PREALLOC;
5641 type = BTRFS_ORDERED_NOCOW;
5642 len = min(len, em->len - (start - em->start));
5643 block_start = em->block_start + (start - em->start);
5646 * we're not going to log anything, but we do need
5647 * to make sure the current transaction stays open
5648 * while we look for nocow cross refs
5650 trans = btrfs_join_transaction(root);
5654 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5655 ret = btrfs_add_ordered_extent_dio(inode, start,
5656 block_start, len, len, type);
5657 btrfs_end_transaction(trans, root);
5659 free_extent_map(em);
5664 btrfs_end_transaction(trans, root);
5668 * this will cow the extent, reset the len in case we changed
5671 len = bh_result->b_size;
5672 em = btrfs_new_extent_direct(inode, em, start, len);
5675 len = min(len, em->len - (start - em->start));
5677 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5678 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5681 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5683 bh_result->b_size = len;
5684 bh_result->b_bdev = em->bdev;
5685 set_buffer_mapped(bh_result);
5686 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5687 set_buffer_new(bh_result);
5689 free_extent_map(em);
5694 struct btrfs_dio_private {
5695 struct inode *inode;
5702 /* number of bios pending for this dio */
5703 atomic_t pending_bios;
5708 struct bio *orig_bio;
5711 static void btrfs_endio_direct_read(struct bio *bio, int err)
5713 struct btrfs_dio_private *dip = bio->bi_private;
5714 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5715 struct bio_vec *bvec = bio->bi_io_vec;
5716 struct inode *inode = dip->inode;
5717 struct btrfs_root *root = BTRFS_I(inode)->root;
5719 u32 *private = dip->csums;
5721 start = dip->logical_offset;
5723 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5724 struct page *page = bvec->bv_page;
5727 unsigned long flags;
5729 local_irq_save(flags);
5730 kaddr = kmap_atomic(page, KM_IRQ0);
5731 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5732 csum, bvec->bv_len);
5733 btrfs_csum_final(csum, (char *)&csum);
5734 kunmap_atomic(kaddr, KM_IRQ0);
5735 local_irq_restore(flags);
5737 flush_dcache_page(bvec->bv_page);
5738 if (csum != *private) {
5739 printk(KERN_ERR "btrfs csum failed ino %llu off"
5740 " %llu csum %u private %u\n",
5741 (unsigned long long)btrfs_ino(inode),
5742 (unsigned long long)start,
5748 start += bvec->bv_len;
5751 } while (bvec <= bvec_end);
5753 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5754 dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5755 bio->bi_private = dip->private;
5760 /* If we had a csum failure make sure to clear the uptodate flag */
5762 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5763 dio_end_io(bio, err);
5766 static void btrfs_endio_direct_write(struct bio *bio, int err)
5768 struct btrfs_dio_private *dip = bio->bi_private;
5769 struct inode *inode = dip->inode;
5770 struct btrfs_root *root = BTRFS_I(inode)->root;
5771 struct btrfs_trans_handle *trans;
5772 struct btrfs_ordered_extent *ordered = NULL;
5773 struct extent_state *cached_state = NULL;
5774 u64 ordered_offset = dip->logical_offset;
5775 u64 ordered_bytes = dip->bytes;
5781 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5789 trans = btrfs_join_transaction(root);
5790 if (IS_ERR(trans)) {
5794 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5796 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5797 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5799 ret = btrfs_update_inode(trans, root, inode);
5804 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5805 ordered->file_offset + ordered->len - 1, 0,
5806 &cached_state, GFP_NOFS);
5808 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5809 ret = btrfs_mark_extent_written(trans, inode,
5810 ordered->file_offset,
5811 ordered->file_offset +
5818 ret = insert_reserved_file_extent(trans, inode,
5819 ordered->file_offset,
5825 BTRFS_FILE_EXTENT_REG);
5826 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5827 ordered->file_offset, ordered->len);
5835 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5836 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5837 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
5838 btrfs_update_inode(trans, root, inode);
5841 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5842 ordered->file_offset + ordered->len - 1,
5843 &cached_state, GFP_NOFS);
5845 btrfs_delalloc_release_metadata(inode, ordered->len);
5846 btrfs_end_transaction(trans, root);
5847 ordered_offset = ordered->file_offset + ordered->len;
5848 btrfs_put_ordered_extent(ordered);
5849 btrfs_put_ordered_extent(ordered);
5853 * our bio might span multiple ordered extents. If we haven't
5854 * completed the accounting for the whole dio, go back and try again
5856 if (ordered_offset < dip->logical_offset + dip->bytes) {
5857 ordered_bytes = dip->logical_offset + dip->bytes -
5862 bio->bi_private = dip->private;
5867 /* If we had an error make sure to clear the uptodate flag */
5869 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5870 dio_end_io(bio, err);
5873 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5874 struct bio *bio, int mirror_num,
5875 unsigned long bio_flags, u64 offset)
5878 struct btrfs_root *root = BTRFS_I(inode)->root;
5879 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5884 static void btrfs_end_dio_bio(struct bio *bio, int err)
5886 struct btrfs_dio_private *dip = bio->bi_private;
5889 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5890 "sector %#Lx len %u err no %d\n",
5891 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5892 (unsigned long long)bio->bi_sector, bio->bi_size, err);
5896 * before atomic variable goto zero, we must make sure
5897 * dip->errors is perceived to be set.
5899 smp_mb__before_atomic_dec();
5902 /* if there are more bios still pending for this dio, just exit */
5903 if (!atomic_dec_and_test(&dip->pending_bios))
5907 bio_io_error(dip->orig_bio);
5909 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5910 bio_endio(dip->orig_bio, 0);
5916 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5917 u64 first_sector, gfp_t gfp_flags)
5919 int nr_vecs = bio_get_nr_vecs(bdev);
5920 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5923 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5924 int rw, u64 file_offset, int skip_sum,
5925 u32 *csums, int async_submit)
5927 int write = rw & REQ_WRITE;
5928 struct btrfs_root *root = BTRFS_I(inode)->root;
5932 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5939 if (write && async_submit) {
5940 ret = btrfs_wq_submit_bio(root->fs_info,
5941 inode, rw, bio, 0, 0,
5943 __btrfs_submit_bio_start_direct_io,
5944 __btrfs_submit_bio_done);
5948 * If we aren't doing async submit, calculate the csum of the
5951 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5954 } else if (!skip_sum) {
5955 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5956 file_offset, csums);
5962 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5968 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5971 struct inode *inode = dip->inode;
5972 struct btrfs_root *root = BTRFS_I(inode)->root;
5973 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5975 struct bio *orig_bio = dip->orig_bio;
5976 struct bio_vec *bvec = orig_bio->bi_io_vec;
5977 u64 start_sector = orig_bio->bi_sector;
5978 u64 file_offset = dip->logical_offset;
5982 u32 *csums = dip->csums;
5984 int async_submit = 0;
5985 int write = rw & REQ_WRITE;
5987 map_length = orig_bio->bi_size;
5988 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5989 &map_length, NULL, 0);
5995 if (map_length >= orig_bio->bi_size) {
6001 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6004 bio->bi_private = dip;
6005 bio->bi_end_io = btrfs_end_dio_bio;
6006 atomic_inc(&dip->pending_bios);
6008 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6009 if (unlikely(map_length < submit_len + bvec->bv_len ||
6010 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6011 bvec->bv_offset) < bvec->bv_len)) {
6013 * inc the count before we submit the bio so
6014 * we know the end IO handler won't happen before
6015 * we inc the count. Otherwise, the dip might get freed
6016 * before we're done setting it up
6018 atomic_inc(&dip->pending_bios);
6019 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6020 file_offset, skip_sum,
6021 csums, async_submit);
6024 atomic_dec(&dip->pending_bios);
6028 /* Write's use the ordered csums */
6029 if (!write && !skip_sum)
6030 csums = csums + nr_pages;
6031 start_sector += submit_len >> 9;
6032 file_offset += submit_len;
6037 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6038 start_sector, GFP_NOFS);
6041 bio->bi_private = dip;
6042 bio->bi_end_io = btrfs_end_dio_bio;
6044 map_length = orig_bio->bi_size;
6045 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6046 &map_length, NULL, 0);
6052 submit_len += bvec->bv_len;
6059 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6060 csums, async_submit);
6068 * before atomic variable goto zero, we must
6069 * make sure dip->errors is perceived to be set.
6071 smp_mb__before_atomic_dec();
6072 if (atomic_dec_and_test(&dip->pending_bios))
6073 bio_io_error(dip->orig_bio);
6075 /* bio_end_io() will handle error, so we needn't return it */
6079 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6082 struct btrfs_root *root = BTRFS_I(inode)->root;
6083 struct btrfs_dio_private *dip;
6084 struct bio_vec *bvec = bio->bi_io_vec;
6086 int write = rw & REQ_WRITE;
6089 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6091 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6098 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6099 if (!write && !skip_sum) {
6100 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6108 dip->private = bio->bi_private;
6110 dip->logical_offset = file_offset;
6114 dip->bytes += bvec->bv_len;
6116 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6118 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6119 bio->bi_private = dip;
6121 dip->orig_bio = bio;
6122 atomic_set(&dip->pending_bios, 0);
6125 bio->bi_end_io = btrfs_endio_direct_write;
6127 bio->bi_end_io = btrfs_endio_direct_read;
6129 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6134 * If this is a write, we need to clean up the reserved space and kill
6135 * the ordered extent.
6138 struct btrfs_ordered_extent *ordered;
6139 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6140 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6141 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6142 btrfs_free_reserved_extent(root, ordered->start,
6144 btrfs_put_ordered_extent(ordered);
6145 btrfs_put_ordered_extent(ordered);
6147 bio_endio(bio, ret);
6150 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6151 const struct iovec *iov, loff_t offset,
6152 unsigned long nr_segs)
6158 unsigned blocksize_mask = root->sectorsize - 1;
6159 ssize_t retval = -EINVAL;
6160 loff_t end = offset;
6162 if (offset & blocksize_mask)
6165 /* Check the memory alignment. Blocks cannot straddle pages */
6166 for (seg = 0; seg < nr_segs; seg++) {
6167 addr = (unsigned long)iov[seg].iov_base;
6168 size = iov[seg].iov_len;
6170 if ((addr & blocksize_mask) || (size & blocksize_mask))
6173 /* If this is a write we don't need to check anymore */
6178 * Check to make sure we don't have duplicate iov_base's in this
6179 * iovec, if so return EINVAL, otherwise we'll get csum errors
6180 * when reading back.
6182 for (i = seg + 1; i < nr_segs; i++) {
6183 if (iov[seg].iov_base == iov[i].iov_base)
6191 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6192 const struct iovec *iov, loff_t offset,
6193 unsigned long nr_segs)
6195 struct file *file = iocb->ki_filp;
6196 struct inode *inode = file->f_mapping->host;
6197 struct btrfs_ordered_extent *ordered;
6198 struct extent_state *cached_state = NULL;
6199 u64 lockstart, lockend;
6201 int writing = rw & WRITE;
6203 size_t count = iov_length(iov, nr_segs);
6205 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6211 lockend = offset + count - 1;
6214 ret = btrfs_delalloc_reserve_space(inode, count);
6220 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6221 0, &cached_state, GFP_NOFS);
6223 * We're concerned with the entire range that we're going to be
6224 * doing DIO to, so we need to make sure theres no ordered
6225 * extents in this range.
6227 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6228 lockend - lockstart + 1);
6231 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6232 &cached_state, GFP_NOFS);
6233 btrfs_start_ordered_extent(inode, ordered, 1);
6234 btrfs_put_ordered_extent(ordered);
6239 * we don't use btrfs_set_extent_delalloc because we don't want
6240 * the dirty or uptodate bits
6243 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6244 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6245 EXTENT_DELALLOC, 0, NULL, &cached_state,
6248 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6249 lockend, EXTENT_LOCKED | write_bits,
6250 1, 0, &cached_state, GFP_NOFS);
6255 free_extent_state(cached_state);
6256 cached_state = NULL;
6258 ret = __blockdev_direct_IO(rw, iocb, inode,
6259 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6260 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6261 btrfs_submit_direct, 0);
6263 if (ret < 0 && ret != -EIOCBQUEUED) {
6264 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6265 offset + iov_length(iov, nr_segs) - 1,
6266 EXTENT_LOCKED | write_bits, 1, 0,
6267 &cached_state, GFP_NOFS);
6268 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6270 * We're falling back to buffered, unlock the section we didn't
6273 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6274 offset + iov_length(iov, nr_segs) - 1,
6275 EXTENT_LOCKED | write_bits, 1, 0,
6276 &cached_state, GFP_NOFS);
6279 free_extent_state(cached_state);
6283 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6284 __u64 start, __u64 len)
6286 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6289 int btrfs_readpage(struct file *file, struct page *page)
6291 struct extent_io_tree *tree;
6292 tree = &BTRFS_I(page->mapping->host)->io_tree;
6293 return extent_read_full_page(tree, page, btrfs_get_extent);
6296 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6298 struct extent_io_tree *tree;
6301 if (current->flags & PF_MEMALLOC) {
6302 redirty_page_for_writepage(wbc, page);
6306 tree = &BTRFS_I(page->mapping->host)->io_tree;
6307 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6310 int btrfs_writepages(struct address_space *mapping,
6311 struct writeback_control *wbc)
6313 struct extent_io_tree *tree;
6315 tree = &BTRFS_I(mapping->host)->io_tree;
6316 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6320 btrfs_readpages(struct file *file, struct address_space *mapping,
6321 struct list_head *pages, unsigned nr_pages)
6323 struct extent_io_tree *tree;
6324 tree = &BTRFS_I(mapping->host)->io_tree;
6325 return extent_readpages(tree, mapping, pages, nr_pages,
6328 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6330 struct extent_io_tree *tree;
6331 struct extent_map_tree *map;
6334 tree = &BTRFS_I(page->mapping->host)->io_tree;
6335 map = &BTRFS_I(page->mapping->host)->extent_tree;
6336 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6338 ClearPagePrivate(page);
6339 set_page_private(page, 0);
6340 page_cache_release(page);
6345 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6347 if (PageWriteback(page) || PageDirty(page))
6349 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6352 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6354 struct extent_io_tree *tree;
6355 struct btrfs_ordered_extent *ordered;
6356 struct extent_state *cached_state = NULL;
6357 u64 page_start = page_offset(page);
6358 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6362 * we have the page locked, so new writeback can't start,
6363 * and the dirty bit won't be cleared while we are here.
6365 * Wait for IO on this page so that we can safely clear
6366 * the PagePrivate2 bit and do ordered accounting
6368 wait_on_page_writeback(page);
6370 tree = &BTRFS_I(page->mapping->host)->io_tree;
6372 btrfs_releasepage(page, GFP_NOFS);
6375 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6377 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6381 * IO on this page will never be started, so we need
6382 * to account for any ordered extents now
6384 clear_extent_bit(tree, page_start, page_end,
6385 EXTENT_DIRTY | EXTENT_DELALLOC |
6386 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6387 &cached_state, GFP_NOFS);
6389 * whoever cleared the private bit is responsible
6390 * for the finish_ordered_io
6392 if (TestClearPagePrivate2(page)) {
6393 btrfs_finish_ordered_io(page->mapping->host,
6394 page_start, page_end);
6396 btrfs_put_ordered_extent(ordered);
6397 cached_state = NULL;
6398 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6401 clear_extent_bit(tree, page_start, page_end,
6402 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6403 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6404 __btrfs_releasepage(page, GFP_NOFS);
6406 ClearPageChecked(page);
6407 if (PagePrivate(page)) {
6408 ClearPagePrivate(page);
6409 set_page_private(page, 0);
6410 page_cache_release(page);
6415 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6416 * called from a page fault handler when a page is first dirtied. Hence we must
6417 * be careful to check for EOF conditions here. We set the page up correctly
6418 * for a written page which means we get ENOSPC checking when writing into
6419 * holes and correct delalloc and unwritten extent mapping on filesystems that
6420 * support these features.
6422 * We are not allowed to take the i_mutex here so we have to play games to
6423 * protect against truncate races as the page could now be beyond EOF. Because
6424 * vmtruncate() writes the inode size before removing pages, once we have the
6425 * page lock we can determine safely if the page is beyond EOF. If it is not
6426 * beyond EOF, then the page is guaranteed safe against truncation until we
6429 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6431 struct page *page = vmf->page;
6432 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6433 struct btrfs_root *root = BTRFS_I(inode)->root;
6434 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6435 struct btrfs_ordered_extent *ordered;
6436 struct extent_state *cached_state = NULL;
6438 unsigned long zero_start;
6444 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6448 else /* -ENOSPC, -EIO, etc */
6449 ret = VM_FAULT_SIGBUS;
6453 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6456 size = i_size_read(inode);
6457 page_start = page_offset(page);
6458 page_end = page_start + PAGE_CACHE_SIZE - 1;
6460 if ((page->mapping != inode->i_mapping) ||
6461 (page_start >= size)) {
6462 /* page got truncated out from underneath us */
6465 wait_on_page_writeback(page);
6467 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6469 set_page_extent_mapped(page);
6472 * we can't set the delalloc bits if there are pending ordered
6473 * extents. Drop our locks and wait for them to finish
6475 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6477 unlock_extent_cached(io_tree, page_start, page_end,
6478 &cached_state, GFP_NOFS);
6480 btrfs_start_ordered_extent(inode, ordered, 1);
6481 btrfs_put_ordered_extent(ordered);
6486 * XXX - page_mkwrite gets called every time the page is dirtied, even
6487 * if it was already dirty, so for space accounting reasons we need to
6488 * clear any delalloc bits for the range we are fixing to save. There
6489 * is probably a better way to do this, but for now keep consistent with
6490 * prepare_pages in the normal write path.
6492 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6493 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6494 0, 0, &cached_state, GFP_NOFS);
6496 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6499 unlock_extent_cached(io_tree, page_start, page_end,
6500 &cached_state, GFP_NOFS);
6501 ret = VM_FAULT_SIGBUS;
6506 /* page is wholly or partially inside EOF */
6507 if (page_start + PAGE_CACHE_SIZE > size)
6508 zero_start = size & ~PAGE_CACHE_MASK;
6510 zero_start = PAGE_CACHE_SIZE;
6512 if (zero_start != PAGE_CACHE_SIZE) {
6514 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6515 flush_dcache_page(page);
6518 ClearPageChecked(page);
6519 set_page_dirty(page);
6520 SetPageUptodate(page);
6522 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6523 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6525 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6529 return VM_FAULT_LOCKED;
6531 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6536 static int btrfs_truncate(struct inode *inode)
6538 struct btrfs_root *root = BTRFS_I(inode)->root;
6539 struct btrfs_block_rsv *rsv;
6542 struct btrfs_trans_handle *trans;
6544 u64 mask = root->sectorsize - 1;
6546 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6550 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6551 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6554 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6555 * 3 things going on here
6557 * 1) We need to reserve space for our orphan item and the space to
6558 * delete our orphan item. Lord knows we don't want to have a dangling
6559 * orphan item because we didn't reserve space to remove it.
6561 * 2) We need to reserve space to update our inode.
6563 * 3) We need to have something to cache all the space that is going to
6564 * be free'd up by the truncate operation, but also have some slack
6565 * space reserved in case it uses space during the truncate (thank you
6566 * very much snapshotting).
6568 * And we need these to all be seperate. The fact is we can use alot of
6569 * space doing the truncate, and we have no earthly idea how much space
6570 * we will use, so we need the truncate reservation to be seperate so it
6571 * doesn't end up using space reserved for updating the inode or
6572 * removing the orphan item. We also need to be able to stop the
6573 * transaction and start a new one, which means we need to be able to
6574 * update the inode several times, and we have no idea of knowing how
6575 * many times that will be, so we can't just reserve 1 item for the
6576 * entirety of the opration, so that has to be done seperately as well.
6577 * Then there is the orphan item, which does indeed need to be held on
6578 * to for the whole operation, and we need nobody to touch this reserved
6579 * space except the orphan code.
6581 * So that leaves us with
6583 * 1) root->orphan_block_rsv - for the orphan deletion.
6584 * 2) rsv - for the truncate reservation, which we will steal from the
6585 * transaction reservation.
6586 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6587 * updating the inode.
6589 rsv = btrfs_alloc_block_rsv(root);
6592 btrfs_add_durable_block_rsv(root->fs_info, rsv);
6594 trans = btrfs_start_transaction(root, 4);
6595 if (IS_ERR(trans)) {
6596 err = PTR_ERR(trans);
6601 * Reserve space for the truncate process. Truncate should be adding
6602 * space, but if there are snapshots it may end up using space.
6604 ret = btrfs_truncate_reserve_metadata(trans, root, rsv);
6607 ret = btrfs_orphan_add(trans, inode);
6609 btrfs_end_transaction(trans, root);
6613 nr = trans->blocks_used;
6614 btrfs_end_transaction(trans, root);
6615 btrfs_btree_balance_dirty(root, nr);
6618 * Ok so we've already migrated our bytes over for the truncate, so here
6619 * just reserve the one slot we need for updating the inode.
6621 trans = btrfs_start_transaction(root, 1);
6622 if (IS_ERR(trans)) {
6623 err = PTR_ERR(trans);
6626 trans->block_rsv = rsv;
6629 * setattr is responsible for setting the ordered_data_close flag,
6630 * but that is only tested during the last file release. That
6631 * could happen well after the next commit, leaving a great big
6632 * window where new writes may get lost if someone chooses to write
6633 * to this file after truncating to zero
6635 * The inode doesn't have any dirty data here, and so if we commit
6636 * this is a noop. If someone immediately starts writing to the inode
6637 * it is very likely we'll catch some of their writes in this
6638 * transaction, and the commit will find this file on the ordered
6639 * data list with good things to send down.
6641 * This is a best effort solution, there is still a window where
6642 * using truncate to replace the contents of the file will
6643 * end up with a zero length file after a crash.
6645 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6646 btrfs_add_ordered_operation(trans, root, inode);
6650 trans = btrfs_start_transaction(root, 3);
6651 if (IS_ERR(trans)) {
6652 err = PTR_ERR(trans);
6656 ret = btrfs_truncate_reserve_metadata(trans, root,
6660 trans->block_rsv = rsv;
6663 ret = btrfs_truncate_inode_items(trans, root, inode,
6665 BTRFS_EXTENT_DATA_KEY);
6666 if (ret != -EAGAIN) {
6671 trans->block_rsv = &root->fs_info->trans_block_rsv;
6672 ret = btrfs_update_inode(trans, root, inode);
6678 nr = trans->blocks_used;
6679 btrfs_end_transaction(trans, root);
6681 btrfs_btree_balance_dirty(root, nr);
6684 if (ret == 0 && inode->i_nlink > 0) {
6685 trans->block_rsv = root->orphan_block_rsv;
6686 ret = btrfs_orphan_del(trans, inode);
6689 } else if (ret && inode->i_nlink > 0) {
6691 * Failed to do the truncate, remove us from the in memory
6694 ret = btrfs_orphan_del(NULL, inode);
6697 trans->block_rsv = &root->fs_info->trans_block_rsv;
6698 ret = btrfs_update_inode(trans, root, inode);
6702 nr = trans->blocks_used;
6703 ret = btrfs_end_transaction_throttle(trans, root);
6704 btrfs_btree_balance_dirty(root, nr);
6707 btrfs_free_block_rsv(root, rsv);
6716 * create a new subvolume directory/inode (helper for the ioctl).
6718 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6719 struct btrfs_root *new_root, u64 new_dirid)
6721 struct inode *inode;
6725 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6726 new_dirid, S_IFDIR | 0700, &index);
6728 return PTR_ERR(inode);
6729 inode->i_op = &btrfs_dir_inode_operations;
6730 inode->i_fop = &btrfs_dir_file_operations;
6732 set_nlink(inode, 1);
6733 btrfs_i_size_write(inode, 0);
6735 err = btrfs_update_inode(trans, new_root, inode);
6742 struct inode *btrfs_alloc_inode(struct super_block *sb)
6744 struct btrfs_inode *ei;
6745 struct inode *inode;
6747 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6752 ei->space_info = NULL;
6756 ei->last_sub_trans = 0;
6757 ei->logged_trans = 0;
6758 ei->delalloc_bytes = 0;
6759 ei->reserved_bytes = 0;
6760 ei->disk_i_size = 0;
6762 ei->index_cnt = (u64)-1;
6763 ei->last_unlink_trans = 0;
6765 spin_lock_init(&ei->lock);
6766 ei->outstanding_extents = 0;
6767 ei->reserved_extents = 0;
6769 ei->ordered_data_close = 0;
6770 ei->orphan_meta_reserved = 0;
6771 ei->dummy_inode = 0;
6773 ei->force_compress = BTRFS_COMPRESS_NONE;
6775 ei->delayed_node = NULL;
6777 inode = &ei->vfs_inode;
6778 extent_map_tree_init(&ei->extent_tree);
6779 extent_io_tree_init(&ei->io_tree, &inode->i_data);
6780 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6781 mutex_init(&ei->log_mutex);
6782 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6783 INIT_LIST_HEAD(&ei->i_orphan);
6784 INIT_LIST_HEAD(&ei->delalloc_inodes);
6785 INIT_LIST_HEAD(&ei->ordered_operations);
6786 RB_CLEAR_NODE(&ei->rb_node);
6791 static void btrfs_i_callback(struct rcu_head *head)
6793 struct inode *inode = container_of(head, struct inode, i_rcu);
6794 INIT_LIST_HEAD(&inode->i_dentry);
6795 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6798 void btrfs_destroy_inode(struct inode *inode)
6800 struct btrfs_ordered_extent *ordered;
6801 struct btrfs_root *root = BTRFS_I(inode)->root;
6803 WARN_ON(!list_empty(&inode->i_dentry));
6804 WARN_ON(inode->i_data.nrpages);
6805 WARN_ON(BTRFS_I(inode)->outstanding_extents);
6806 WARN_ON(BTRFS_I(inode)->reserved_extents);
6809 * This can happen where we create an inode, but somebody else also
6810 * created the same inode and we need to destroy the one we already
6817 * Make sure we're properly removed from the ordered operation
6821 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6822 spin_lock(&root->fs_info->ordered_extent_lock);
6823 list_del_init(&BTRFS_I(inode)->ordered_operations);
6824 spin_unlock(&root->fs_info->ordered_extent_lock);
6827 spin_lock(&root->orphan_lock);
6828 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6829 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6830 (unsigned long long)btrfs_ino(inode));
6831 list_del_init(&BTRFS_I(inode)->i_orphan);
6833 spin_unlock(&root->orphan_lock);
6836 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6840 printk(KERN_ERR "btrfs found ordered "
6841 "extent %llu %llu on inode cleanup\n",
6842 (unsigned long long)ordered->file_offset,
6843 (unsigned long long)ordered->len);
6844 btrfs_remove_ordered_extent(inode, ordered);
6845 btrfs_put_ordered_extent(ordered);
6846 btrfs_put_ordered_extent(ordered);
6849 inode_tree_del(inode);
6850 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6852 btrfs_remove_delayed_node(inode);
6853 call_rcu(&inode->i_rcu, btrfs_i_callback);
6856 int btrfs_drop_inode(struct inode *inode)
6858 struct btrfs_root *root = BTRFS_I(inode)->root;
6860 if (btrfs_root_refs(&root->root_item) == 0 &&
6861 !btrfs_is_free_space_inode(root, inode))
6864 return generic_drop_inode(inode);
6867 static void init_once(void *foo)
6869 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6871 inode_init_once(&ei->vfs_inode);
6874 void btrfs_destroy_cachep(void)
6876 if (btrfs_inode_cachep)
6877 kmem_cache_destroy(btrfs_inode_cachep);
6878 if (btrfs_trans_handle_cachep)
6879 kmem_cache_destroy(btrfs_trans_handle_cachep);
6880 if (btrfs_transaction_cachep)
6881 kmem_cache_destroy(btrfs_transaction_cachep);
6882 if (btrfs_path_cachep)
6883 kmem_cache_destroy(btrfs_path_cachep);
6884 if (btrfs_free_space_cachep)
6885 kmem_cache_destroy(btrfs_free_space_cachep);
6888 int btrfs_init_cachep(void)
6890 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6891 sizeof(struct btrfs_inode), 0,
6892 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6893 if (!btrfs_inode_cachep)
6896 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6897 sizeof(struct btrfs_trans_handle), 0,
6898 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6899 if (!btrfs_trans_handle_cachep)
6902 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6903 sizeof(struct btrfs_transaction), 0,
6904 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6905 if (!btrfs_transaction_cachep)
6908 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6909 sizeof(struct btrfs_path), 0,
6910 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6911 if (!btrfs_path_cachep)
6914 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6915 sizeof(struct btrfs_free_space), 0,
6916 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6917 if (!btrfs_free_space_cachep)
6922 btrfs_destroy_cachep();
6926 static int btrfs_getattr(struct vfsmount *mnt,
6927 struct dentry *dentry, struct kstat *stat)
6929 struct inode *inode = dentry->d_inode;
6930 generic_fillattr(inode, stat);
6931 stat->dev = BTRFS_I(inode)->root->anon_dev;
6932 stat->blksize = PAGE_CACHE_SIZE;
6933 stat->blocks = (inode_get_bytes(inode) +
6934 BTRFS_I(inode)->delalloc_bytes) >> 9;
6939 * If a file is moved, it will inherit the cow and compression flags of the new
6942 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6944 struct btrfs_inode *b_dir = BTRFS_I(dir);
6945 struct btrfs_inode *b_inode = BTRFS_I(inode);
6947 if (b_dir->flags & BTRFS_INODE_NODATACOW)
6948 b_inode->flags |= BTRFS_INODE_NODATACOW;
6950 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6952 if (b_dir->flags & BTRFS_INODE_COMPRESS)
6953 b_inode->flags |= BTRFS_INODE_COMPRESS;
6955 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6958 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6959 struct inode *new_dir, struct dentry *new_dentry)
6961 struct btrfs_trans_handle *trans;
6962 struct btrfs_root *root = BTRFS_I(old_dir)->root;
6963 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6964 struct inode *new_inode = new_dentry->d_inode;
6965 struct inode *old_inode = old_dentry->d_inode;
6966 struct timespec ctime = CURRENT_TIME;
6970 u64 old_ino = btrfs_ino(old_inode);
6972 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6975 /* we only allow rename subvolume link between subvolumes */
6976 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6979 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6980 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6983 if (S_ISDIR(old_inode->i_mode) && new_inode &&
6984 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6987 * we're using rename to replace one file with another.
6988 * and the replacement file is large. Start IO on it now so
6989 * we don't add too much work to the end of the transaction
6991 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6992 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6993 filemap_flush(old_inode->i_mapping);
6995 /* close the racy window with snapshot create/destroy ioctl */
6996 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6997 down_read(&root->fs_info->subvol_sem);
6999 * We want to reserve the absolute worst case amount of items. So if
7000 * both inodes are subvols and we need to unlink them then that would
7001 * require 4 item modifications, but if they are both normal inodes it
7002 * would require 5 item modifications, so we'll assume their normal
7003 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7004 * should cover the worst case number of items we'll modify.
7006 trans = btrfs_start_transaction(root, 20);
7007 if (IS_ERR(trans)) {
7008 ret = PTR_ERR(trans);
7013 btrfs_record_root_in_trans(trans, dest);
7015 ret = btrfs_set_inode_index(new_dir, &index);
7019 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7020 /* force full log commit if subvolume involved. */
7021 root->fs_info->last_trans_log_full_commit = trans->transid;
7023 ret = btrfs_insert_inode_ref(trans, dest,
7024 new_dentry->d_name.name,
7025 new_dentry->d_name.len,
7027 btrfs_ino(new_dir), index);
7031 * this is an ugly little race, but the rename is required
7032 * to make sure that if we crash, the inode is either at the
7033 * old name or the new one. pinning the log transaction lets
7034 * us make sure we don't allow a log commit to come in after
7035 * we unlink the name but before we add the new name back in.
7037 btrfs_pin_log_trans(root);
7040 * make sure the inode gets flushed if it is replacing
7043 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7044 btrfs_add_ordered_operation(trans, root, old_inode);
7046 old_dir->i_ctime = old_dir->i_mtime = ctime;
7047 new_dir->i_ctime = new_dir->i_mtime = ctime;
7048 old_inode->i_ctime = ctime;
7050 if (old_dentry->d_parent != new_dentry->d_parent)
7051 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7053 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7054 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7055 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7056 old_dentry->d_name.name,
7057 old_dentry->d_name.len);
7059 ret = __btrfs_unlink_inode(trans, root, old_dir,
7060 old_dentry->d_inode,
7061 old_dentry->d_name.name,
7062 old_dentry->d_name.len);
7064 ret = btrfs_update_inode(trans, root, old_inode);
7069 new_inode->i_ctime = CURRENT_TIME;
7070 if (unlikely(btrfs_ino(new_inode) ==
7071 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7072 root_objectid = BTRFS_I(new_inode)->location.objectid;
7073 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7075 new_dentry->d_name.name,
7076 new_dentry->d_name.len);
7077 BUG_ON(new_inode->i_nlink == 0);
7079 ret = btrfs_unlink_inode(trans, dest, new_dir,
7080 new_dentry->d_inode,
7081 new_dentry->d_name.name,
7082 new_dentry->d_name.len);
7085 if (new_inode->i_nlink == 0) {
7086 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7091 fixup_inode_flags(new_dir, old_inode);
7093 ret = btrfs_add_link(trans, new_dir, old_inode,
7094 new_dentry->d_name.name,
7095 new_dentry->d_name.len, 0, index);
7098 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7099 struct dentry *parent = new_dentry->d_parent;
7100 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7101 btrfs_end_log_trans(root);
7104 btrfs_end_transaction_throttle(trans, root);
7106 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7107 up_read(&root->fs_info->subvol_sem);
7113 * some fairly slow code that needs optimization. This walks the list
7114 * of all the inodes with pending delalloc and forces them to disk.
7116 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7118 struct list_head *head = &root->fs_info->delalloc_inodes;
7119 struct btrfs_inode *binode;
7120 struct inode *inode;
7122 if (root->fs_info->sb->s_flags & MS_RDONLY)
7125 spin_lock(&root->fs_info->delalloc_lock);
7126 while (!list_empty(head)) {
7127 binode = list_entry(head->next, struct btrfs_inode,
7129 inode = igrab(&binode->vfs_inode);
7131 list_del_init(&binode->delalloc_inodes);
7132 spin_unlock(&root->fs_info->delalloc_lock);
7134 filemap_flush(inode->i_mapping);
7136 btrfs_add_delayed_iput(inode);
7141 spin_lock(&root->fs_info->delalloc_lock);
7143 spin_unlock(&root->fs_info->delalloc_lock);
7145 /* the filemap_flush will queue IO into the worker threads, but
7146 * we have to make sure the IO is actually started and that
7147 * ordered extents get created before we return
7149 atomic_inc(&root->fs_info->async_submit_draining);
7150 while (atomic_read(&root->fs_info->nr_async_submits) ||
7151 atomic_read(&root->fs_info->async_delalloc_pages)) {
7152 wait_event(root->fs_info->async_submit_wait,
7153 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7154 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7156 atomic_dec(&root->fs_info->async_submit_draining);
7160 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7161 const char *symname)
7163 struct btrfs_trans_handle *trans;
7164 struct btrfs_root *root = BTRFS_I(dir)->root;
7165 struct btrfs_path *path;
7166 struct btrfs_key key;
7167 struct inode *inode = NULL;
7175 struct btrfs_file_extent_item *ei;
7176 struct extent_buffer *leaf;
7177 unsigned long nr = 0;
7179 name_len = strlen(symname) + 1;
7180 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7181 return -ENAMETOOLONG;
7184 * 2 items for inode item and ref
7185 * 2 items for dir items
7186 * 1 item for xattr if selinux is on
7188 trans = btrfs_start_transaction(root, 5);
7190 return PTR_ERR(trans);
7192 err = btrfs_find_free_ino(root, &objectid);
7196 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7197 dentry->d_name.len, btrfs_ino(dir), objectid,
7198 S_IFLNK|S_IRWXUGO, &index);
7199 if (IS_ERR(inode)) {
7200 err = PTR_ERR(inode);
7204 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7210 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7214 inode->i_mapping->a_ops = &btrfs_aops;
7215 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7216 inode->i_fop = &btrfs_file_operations;
7217 inode->i_op = &btrfs_file_inode_operations;
7218 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7223 path = btrfs_alloc_path();
7229 key.objectid = btrfs_ino(inode);
7231 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7232 datasize = btrfs_file_extent_calc_inline_size(name_len);
7233 err = btrfs_insert_empty_item(trans, root, path, &key,
7237 btrfs_free_path(path);
7240 leaf = path->nodes[0];
7241 ei = btrfs_item_ptr(leaf, path->slots[0],
7242 struct btrfs_file_extent_item);
7243 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7244 btrfs_set_file_extent_type(leaf, ei,
7245 BTRFS_FILE_EXTENT_INLINE);
7246 btrfs_set_file_extent_encryption(leaf, ei, 0);
7247 btrfs_set_file_extent_compression(leaf, ei, 0);
7248 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7249 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7251 ptr = btrfs_file_extent_inline_start(ei);
7252 write_extent_buffer(leaf, symname, ptr, name_len);
7253 btrfs_mark_buffer_dirty(leaf);
7254 btrfs_free_path(path);
7256 inode->i_op = &btrfs_symlink_inode_operations;
7257 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7258 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7259 inode_set_bytes(inode, name_len);
7260 btrfs_i_size_write(inode, name_len - 1);
7261 err = btrfs_update_inode(trans, root, inode);
7266 nr = trans->blocks_used;
7267 btrfs_end_transaction_throttle(trans, root);
7269 inode_dec_link_count(inode);
7272 btrfs_btree_balance_dirty(root, nr);
7276 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7277 u64 start, u64 num_bytes, u64 min_size,
7278 loff_t actual_len, u64 *alloc_hint,
7279 struct btrfs_trans_handle *trans)
7281 struct btrfs_root *root = BTRFS_I(inode)->root;
7282 struct btrfs_key ins;
7283 u64 cur_offset = start;
7286 bool own_trans = true;
7290 while (num_bytes > 0) {
7292 trans = btrfs_start_transaction(root, 3);
7293 if (IS_ERR(trans)) {
7294 ret = PTR_ERR(trans);
7299 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7300 0, *alloc_hint, (u64)-1, &ins, 1);
7303 btrfs_end_transaction(trans, root);
7307 ret = insert_reserved_file_extent(trans, inode,
7308 cur_offset, ins.objectid,
7309 ins.offset, ins.offset,
7310 ins.offset, 0, 0, 0,
7311 BTRFS_FILE_EXTENT_PREALLOC);
7313 btrfs_drop_extent_cache(inode, cur_offset,
7314 cur_offset + ins.offset -1, 0);
7316 num_bytes -= ins.offset;
7317 cur_offset += ins.offset;
7318 *alloc_hint = ins.objectid + ins.offset;
7320 inode->i_ctime = CURRENT_TIME;
7321 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7322 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7323 (actual_len > inode->i_size) &&
7324 (cur_offset > inode->i_size)) {
7325 if (cur_offset > actual_len)
7326 i_size = actual_len;
7328 i_size = cur_offset;
7329 i_size_write(inode, i_size);
7330 btrfs_ordered_update_i_size(inode, i_size, NULL);
7333 ret = btrfs_update_inode(trans, root, inode);
7337 btrfs_end_transaction(trans, root);
7342 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7343 u64 start, u64 num_bytes, u64 min_size,
7344 loff_t actual_len, u64 *alloc_hint)
7346 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7347 min_size, actual_len, alloc_hint,
7351 int btrfs_prealloc_file_range_trans(struct inode *inode,
7352 struct btrfs_trans_handle *trans, int mode,
7353 u64 start, u64 num_bytes, u64 min_size,
7354 loff_t actual_len, u64 *alloc_hint)
7356 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7357 min_size, actual_len, alloc_hint, trans);
7360 static int btrfs_set_page_dirty(struct page *page)
7362 return __set_page_dirty_nobuffers(page);
7365 static int btrfs_permission(struct inode *inode, int mask)
7367 struct btrfs_root *root = BTRFS_I(inode)->root;
7368 umode_t mode = inode->i_mode;
7370 if (mask & MAY_WRITE &&
7371 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7372 if (btrfs_root_readonly(root))
7374 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7377 return generic_permission(inode, mask);
7380 static const struct inode_operations btrfs_dir_inode_operations = {
7381 .getattr = btrfs_getattr,
7382 .lookup = btrfs_lookup,
7383 .create = btrfs_create,
7384 .unlink = btrfs_unlink,
7386 .mkdir = btrfs_mkdir,
7387 .rmdir = btrfs_rmdir,
7388 .rename = btrfs_rename,
7389 .symlink = btrfs_symlink,
7390 .setattr = btrfs_setattr,
7391 .mknod = btrfs_mknod,
7392 .setxattr = btrfs_setxattr,
7393 .getxattr = btrfs_getxattr,
7394 .listxattr = btrfs_listxattr,
7395 .removexattr = btrfs_removexattr,
7396 .permission = btrfs_permission,
7397 .get_acl = btrfs_get_acl,
7399 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7400 .lookup = btrfs_lookup,
7401 .permission = btrfs_permission,
7402 .get_acl = btrfs_get_acl,
7405 static const struct file_operations btrfs_dir_file_operations = {
7406 .llseek = generic_file_llseek,
7407 .read = generic_read_dir,
7408 .readdir = btrfs_real_readdir,
7409 .unlocked_ioctl = btrfs_ioctl,
7410 #ifdef CONFIG_COMPAT
7411 .compat_ioctl = btrfs_ioctl,
7413 .release = btrfs_release_file,
7414 .fsync = btrfs_sync_file,
7417 static struct extent_io_ops btrfs_extent_io_ops = {
7418 .fill_delalloc = run_delalloc_range,
7419 .submit_bio_hook = btrfs_submit_bio_hook,
7420 .merge_bio_hook = btrfs_merge_bio_hook,
7421 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7422 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7423 .writepage_start_hook = btrfs_writepage_start_hook,
7424 .readpage_io_failed_hook = btrfs_io_failed_hook,
7425 .set_bit_hook = btrfs_set_bit_hook,
7426 .clear_bit_hook = btrfs_clear_bit_hook,
7427 .merge_extent_hook = btrfs_merge_extent_hook,
7428 .split_extent_hook = btrfs_split_extent_hook,
7432 * btrfs doesn't support the bmap operation because swapfiles
7433 * use bmap to make a mapping of extents in the file. They assume
7434 * these extents won't change over the life of the file and they
7435 * use the bmap result to do IO directly to the drive.
7437 * the btrfs bmap call would return logical addresses that aren't
7438 * suitable for IO and they also will change frequently as COW
7439 * operations happen. So, swapfile + btrfs == corruption.
7441 * For now we're avoiding this by dropping bmap.
7443 static const struct address_space_operations btrfs_aops = {
7444 .readpage = btrfs_readpage,
7445 .writepage = btrfs_writepage,
7446 .writepages = btrfs_writepages,
7447 .readpages = btrfs_readpages,
7448 .direct_IO = btrfs_direct_IO,
7449 .invalidatepage = btrfs_invalidatepage,
7450 .releasepage = btrfs_releasepage,
7451 .set_page_dirty = btrfs_set_page_dirty,
7452 .error_remove_page = generic_error_remove_page,
7455 static const struct address_space_operations btrfs_symlink_aops = {
7456 .readpage = btrfs_readpage,
7457 .writepage = btrfs_writepage,
7458 .invalidatepage = btrfs_invalidatepage,
7459 .releasepage = btrfs_releasepage,
7462 static const struct inode_operations btrfs_file_inode_operations = {
7463 .getattr = btrfs_getattr,
7464 .setattr = btrfs_setattr,
7465 .setxattr = btrfs_setxattr,
7466 .getxattr = btrfs_getxattr,
7467 .listxattr = btrfs_listxattr,
7468 .removexattr = btrfs_removexattr,
7469 .permission = btrfs_permission,
7470 .fiemap = btrfs_fiemap,
7471 .get_acl = btrfs_get_acl,
7473 static const struct inode_operations btrfs_special_inode_operations = {
7474 .getattr = btrfs_getattr,
7475 .setattr = btrfs_setattr,
7476 .permission = btrfs_permission,
7477 .setxattr = btrfs_setxattr,
7478 .getxattr = btrfs_getxattr,
7479 .listxattr = btrfs_listxattr,
7480 .removexattr = btrfs_removexattr,
7481 .get_acl = btrfs_get_acl,
7483 static const struct inode_operations btrfs_symlink_inode_operations = {
7484 .readlink = generic_readlink,
7485 .follow_link = page_follow_link_light,
7486 .put_link = page_put_link,
7487 .getattr = btrfs_getattr,
7488 .permission = btrfs_permission,
7489 .setxattr = btrfs_setxattr,
7490 .getxattr = btrfs_getxattr,
7491 .listxattr = btrfs_listxattr,
7492 .removexattr = btrfs_removexattr,
7493 .get_acl = btrfs_get_acl,
7496 const struct dentry_operations btrfs_dentry_operations = {
7497 .d_delete = btrfs_dentry_delete,
7498 .d_release = btrfs_dentry_release,