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);
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->start > start || em->start + em->len < start) {
1867 free_extent_map(em);
1870 read_unlock(&em_tree->lock);
1872 if (IS_ERR_OR_NULL(em)) {
1876 logical = start - em->start;
1877 logical = em->block_start + logical;
1878 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1879 logical = em->block_start;
1880 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1881 extent_set_compress_type(&failrec->bio_flags,
1884 failrec->logical = logical;
1885 free_extent_map(em);
1886 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1887 EXTENT_DIRTY, GFP_NOFS);
1888 set_state_private(failure_tree, start,
1889 (u64)(unsigned long)failrec);
1891 failrec = (struct io_failure_record *)(unsigned long)private;
1893 num_copies = btrfs_num_copies(
1894 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1895 failrec->logical, failrec->len);
1896 failrec->last_mirror++;
1898 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1899 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1902 if (state && state->start != failrec->start)
1904 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1906 if (!state || failrec->last_mirror > num_copies) {
1907 set_state_private(failure_tree, failrec->start, 0);
1908 clear_extent_bits(failure_tree, failrec->start,
1909 failrec->start + failrec->len - 1,
1910 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1914 bio = bio_alloc(GFP_NOFS, 1);
1915 bio->bi_private = state;
1916 bio->bi_end_io = failed_bio->bi_end_io;
1917 bio->bi_sector = failrec->logical >> 9;
1918 bio->bi_bdev = failed_bio->bi_bdev;
1921 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1922 if (failed_bio->bi_rw & REQ_WRITE)
1927 ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1928 failrec->last_mirror,
1929 failrec->bio_flags, 0);
1934 * each time an IO finishes, we do a fast check in the IO failure tree
1935 * to see if we need to process or clean up an io_failure_record
1937 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1940 u64 private_failure;
1941 struct io_failure_record *failure;
1945 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1946 (u64)-1, 1, EXTENT_DIRTY, 0)) {
1947 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1948 start, &private_failure);
1950 failure = (struct io_failure_record *)(unsigned long)
1952 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1954 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1956 failure->start + failure->len - 1,
1957 EXTENT_DIRTY | EXTENT_LOCKED,
1966 * when reads are done, we need to check csums to verify the data is correct
1967 * if there's a match, we allow the bio to finish. If not, we go through
1968 * the io_failure_record routines to find good copies
1970 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1971 struct extent_state *state)
1973 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1974 struct inode *inode = page->mapping->host;
1975 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1977 u64 private = ~(u32)0;
1979 struct btrfs_root *root = BTRFS_I(inode)->root;
1982 if (PageChecked(page)) {
1983 ClearPageChecked(page);
1987 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1990 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1991 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1992 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1997 if (state && state->start == start) {
1998 private = state->private;
2001 ret = get_state_private(io_tree, start, &private);
2003 kaddr = kmap_atomic(page, KM_USER0);
2007 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
2008 btrfs_csum_final(csum, (char *)&csum);
2009 if (csum != private)
2012 kunmap_atomic(kaddr, KM_USER0);
2014 /* if the io failure tree for this inode is non-empty,
2015 * check to see if we've recovered from a failed IO
2017 btrfs_clean_io_failures(inode, start);
2021 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2023 (unsigned long long)btrfs_ino(page->mapping->host),
2024 (unsigned long long)start, csum,
2025 (unsigned long long)private);
2026 memset(kaddr + offset, 1, end - start + 1);
2027 flush_dcache_page(page);
2028 kunmap_atomic(kaddr, KM_USER0);
2034 struct delayed_iput {
2035 struct list_head list;
2036 struct inode *inode;
2039 void btrfs_add_delayed_iput(struct inode *inode)
2041 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2042 struct delayed_iput *delayed;
2044 if (atomic_add_unless(&inode->i_count, -1, 1))
2047 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2048 delayed->inode = inode;
2050 spin_lock(&fs_info->delayed_iput_lock);
2051 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2052 spin_unlock(&fs_info->delayed_iput_lock);
2055 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2058 struct btrfs_fs_info *fs_info = root->fs_info;
2059 struct delayed_iput *delayed;
2062 spin_lock(&fs_info->delayed_iput_lock);
2063 empty = list_empty(&fs_info->delayed_iputs);
2064 spin_unlock(&fs_info->delayed_iput_lock);
2068 down_read(&root->fs_info->cleanup_work_sem);
2069 spin_lock(&fs_info->delayed_iput_lock);
2070 list_splice_init(&fs_info->delayed_iputs, &list);
2071 spin_unlock(&fs_info->delayed_iput_lock);
2073 while (!list_empty(&list)) {
2074 delayed = list_entry(list.next, struct delayed_iput, list);
2075 list_del(&delayed->list);
2076 iput(delayed->inode);
2079 up_read(&root->fs_info->cleanup_work_sem);
2083 * calculate extra metadata reservation when snapshotting a subvolume
2084 * contains orphan files.
2086 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2087 struct btrfs_pending_snapshot *pending,
2088 u64 *bytes_to_reserve)
2090 struct btrfs_root *root;
2091 struct btrfs_block_rsv *block_rsv;
2095 root = pending->root;
2096 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2099 block_rsv = root->orphan_block_rsv;
2101 /* orphan block reservation for the snapshot */
2102 num_bytes = block_rsv->size;
2105 * after the snapshot is created, COWing tree blocks may use more
2106 * space than it frees. So we should make sure there is enough
2109 index = trans->transid & 0x1;
2110 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2111 num_bytes += block_rsv->size -
2112 (block_rsv->reserved + block_rsv->freed[index]);
2115 *bytes_to_reserve += num_bytes;
2118 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2119 struct btrfs_pending_snapshot *pending)
2121 struct btrfs_root *root = pending->root;
2122 struct btrfs_root *snap = pending->snap;
2123 struct btrfs_block_rsv *block_rsv;
2128 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2131 /* refill source subvolume's orphan block reservation */
2132 block_rsv = root->orphan_block_rsv;
2133 index = trans->transid & 0x1;
2134 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2135 num_bytes = block_rsv->size -
2136 (block_rsv->reserved + block_rsv->freed[index]);
2137 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2138 root->orphan_block_rsv,
2143 /* setup orphan block reservation for the snapshot */
2144 block_rsv = btrfs_alloc_block_rsv(snap);
2147 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2148 snap->orphan_block_rsv = block_rsv;
2150 num_bytes = root->orphan_block_rsv->size;
2151 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2152 block_rsv, num_bytes);
2156 /* insert orphan item for the snapshot */
2157 WARN_ON(!root->orphan_item_inserted);
2158 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2159 snap->root_key.objectid);
2161 snap->orphan_item_inserted = 1;
2165 enum btrfs_orphan_cleanup_state {
2166 ORPHAN_CLEANUP_STARTED = 1,
2167 ORPHAN_CLEANUP_DONE = 2,
2171 * This is called in transaction commmit time. If there are no orphan
2172 * files in the subvolume, it removes orphan item and frees block_rsv
2175 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2176 struct btrfs_root *root)
2180 if (!list_empty(&root->orphan_list) ||
2181 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2184 if (root->orphan_item_inserted &&
2185 btrfs_root_refs(&root->root_item) > 0) {
2186 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2187 root->root_key.objectid);
2189 root->orphan_item_inserted = 0;
2192 if (root->orphan_block_rsv) {
2193 WARN_ON(root->orphan_block_rsv->size > 0);
2194 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2195 root->orphan_block_rsv = NULL;
2200 * This creates an orphan entry for the given inode in case something goes
2201 * wrong in the middle of an unlink/truncate.
2203 * NOTE: caller of this function should reserve 5 units of metadata for
2206 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2208 struct btrfs_root *root = BTRFS_I(inode)->root;
2209 struct btrfs_block_rsv *block_rsv = NULL;
2214 if (!root->orphan_block_rsv) {
2215 block_rsv = btrfs_alloc_block_rsv(root);
2220 spin_lock(&root->orphan_lock);
2221 if (!root->orphan_block_rsv) {
2222 root->orphan_block_rsv = block_rsv;
2223 } else if (block_rsv) {
2224 btrfs_free_block_rsv(root, block_rsv);
2228 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2229 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2232 * For proper ENOSPC handling, we should do orphan
2233 * cleanup when mounting. But this introduces backward
2234 * compatibility issue.
2236 if (!xchg(&root->orphan_item_inserted, 1))
2244 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2245 BTRFS_I(inode)->orphan_meta_reserved = 1;
2248 spin_unlock(&root->orphan_lock);
2251 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2253 /* grab metadata reservation from transaction handle */
2255 ret = btrfs_orphan_reserve_metadata(trans, inode);
2259 /* insert an orphan item to track this unlinked/truncated file */
2261 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2265 /* insert an orphan item to track subvolume contains orphan files */
2267 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2268 root->root_key.objectid);
2275 * We have done the truncate/delete so we can go ahead and remove the orphan
2276 * item for this particular inode.
2278 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2280 struct btrfs_root *root = BTRFS_I(inode)->root;
2281 int delete_item = 0;
2282 int release_rsv = 0;
2285 spin_lock(&root->orphan_lock);
2286 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2287 list_del_init(&BTRFS_I(inode)->i_orphan);
2291 if (BTRFS_I(inode)->orphan_meta_reserved) {
2292 BTRFS_I(inode)->orphan_meta_reserved = 0;
2295 spin_unlock(&root->orphan_lock);
2297 if (trans && delete_item) {
2298 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2303 btrfs_orphan_release_metadata(inode);
2309 * this cleans up any orphans that may be left on the list from the last use
2312 int btrfs_orphan_cleanup(struct btrfs_root *root)
2314 struct btrfs_path *path;
2315 struct extent_buffer *leaf;
2316 struct btrfs_key key, found_key;
2317 struct btrfs_trans_handle *trans;
2318 struct inode *inode;
2319 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2321 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2324 path = btrfs_alloc_path();
2331 key.objectid = BTRFS_ORPHAN_OBJECTID;
2332 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2333 key.offset = (u64)-1;
2336 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2341 * if ret == 0 means we found what we were searching for, which
2342 * is weird, but possible, so only screw with path if we didn't
2343 * find the key and see if we have stuff that matches
2347 if (path->slots[0] == 0)
2352 /* pull out the item */
2353 leaf = path->nodes[0];
2354 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2356 /* make sure the item matches what we want */
2357 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2359 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2362 /* release the path since we're done with it */
2363 btrfs_release_path(path);
2366 * this is where we are basically btrfs_lookup, without the
2367 * crossing root thing. we store the inode number in the
2368 * offset of the orphan item.
2370 found_key.objectid = found_key.offset;
2371 found_key.type = BTRFS_INODE_ITEM_KEY;
2372 found_key.offset = 0;
2373 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2374 if (IS_ERR(inode)) {
2375 ret = PTR_ERR(inode);
2380 * add this inode to the orphan list so btrfs_orphan_del does
2381 * the proper thing when we hit it
2383 spin_lock(&root->orphan_lock);
2384 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2385 spin_unlock(&root->orphan_lock);
2388 * if this is a bad inode, means we actually succeeded in
2389 * removing the inode, but not the orphan record, which means
2390 * we need to manually delete the orphan since iput will just
2391 * do a destroy_inode
2393 if (is_bad_inode(inode)) {
2394 trans = btrfs_start_transaction(root, 0);
2395 if (IS_ERR(trans)) {
2396 ret = PTR_ERR(trans);
2399 btrfs_orphan_del(trans, inode);
2400 btrfs_end_transaction(trans, root);
2405 /* if we have links, this was a truncate, lets do that */
2406 if (inode->i_nlink) {
2407 if (!S_ISREG(inode->i_mode)) {
2413 ret = btrfs_truncate(inode);
2418 /* this will do delete_inode and everything for us */
2423 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2425 if (root->orphan_block_rsv)
2426 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2429 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2430 trans = btrfs_join_transaction(root);
2432 btrfs_end_transaction(trans, root);
2436 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2438 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2442 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2443 btrfs_free_path(path);
2448 * very simple check to peek ahead in the leaf looking for xattrs. If we
2449 * don't find any xattrs, we know there can't be any acls.
2451 * slot is the slot the inode is in, objectid is the objectid of the inode
2453 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2454 int slot, u64 objectid)
2456 u32 nritems = btrfs_header_nritems(leaf);
2457 struct btrfs_key found_key;
2461 while (slot < nritems) {
2462 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2464 /* we found a different objectid, there must not be acls */
2465 if (found_key.objectid != objectid)
2468 /* we found an xattr, assume we've got an acl */
2469 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2473 * we found a key greater than an xattr key, there can't
2474 * be any acls later on
2476 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2483 * it goes inode, inode backrefs, xattrs, extents,
2484 * so if there are a ton of hard links to an inode there can
2485 * be a lot of backrefs. Don't waste time searching too hard,
2486 * this is just an optimization
2491 /* we hit the end of the leaf before we found an xattr or
2492 * something larger than an xattr. We have to assume the inode
2499 * read an inode from the btree into the in-memory inode
2501 static void btrfs_read_locked_inode(struct inode *inode)
2503 struct btrfs_path *path;
2504 struct extent_buffer *leaf;
2505 struct btrfs_inode_item *inode_item;
2506 struct btrfs_timespec *tspec;
2507 struct btrfs_root *root = BTRFS_I(inode)->root;
2508 struct btrfs_key location;
2512 bool filled = false;
2514 ret = btrfs_fill_inode(inode, &rdev);
2518 path = btrfs_alloc_path();
2522 path->leave_spinning = 1;
2523 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2525 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2529 leaf = path->nodes[0];
2534 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2535 struct btrfs_inode_item);
2536 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2537 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2538 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2539 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2540 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2542 tspec = btrfs_inode_atime(inode_item);
2543 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2544 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2546 tspec = btrfs_inode_mtime(inode_item);
2547 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2548 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2550 tspec = btrfs_inode_ctime(inode_item);
2551 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2552 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2554 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2555 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2556 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2557 inode->i_generation = BTRFS_I(inode)->generation;
2559 rdev = btrfs_inode_rdev(leaf, inode_item);
2561 BTRFS_I(inode)->index_cnt = (u64)-1;
2562 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2565 * try to precache a NULL acl entry for files that don't have
2566 * any xattrs or acls
2568 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2571 cache_no_acl(inode);
2573 btrfs_free_path(path);
2575 switch (inode->i_mode & S_IFMT) {
2577 inode->i_mapping->a_ops = &btrfs_aops;
2578 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2579 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2580 inode->i_fop = &btrfs_file_operations;
2581 inode->i_op = &btrfs_file_inode_operations;
2584 inode->i_fop = &btrfs_dir_file_operations;
2585 if (root == root->fs_info->tree_root)
2586 inode->i_op = &btrfs_dir_ro_inode_operations;
2588 inode->i_op = &btrfs_dir_inode_operations;
2591 inode->i_op = &btrfs_symlink_inode_operations;
2592 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2593 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2596 inode->i_op = &btrfs_special_inode_operations;
2597 init_special_inode(inode, inode->i_mode, rdev);
2601 btrfs_update_iflags(inode);
2605 btrfs_free_path(path);
2606 make_bad_inode(inode);
2610 * given a leaf and an inode, copy the inode fields into the leaf
2612 static void fill_inode_item(struct btrfs_trans_handle *trans,
2613 struct extent_buffer *leaf,
2614 struct btrfs_inode_item *item,
2615 struct inode *inode)
2617 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2618 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2619 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2620 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2621 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2623 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2624 inode->i_atime.tv_sec);
2625 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2626 inode->i_atime.tv_nsec);
2628 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2629 inode->i_mtime.tv_sec);
2630 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2631 inode->i_mtime.tv_nsec);
2633 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2634 inode->i_ctime.tv_sec);
2635 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2636 inode->i_ctime.tv_nsec);
2638 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2639 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2640 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2641 btrfs_set_inode_transid(leaf, item, trans->transid);
2642 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2643 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2644 btrfs_set_inode_block_group(leaf, item, 0);
2648 * copy everything in the in-memory inode into the btree.
2650 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2651 struct btrfs_root *root, struct inode *inode)
2653 struct btrfs_inode_item *inode_item;
2654 struct btrfs_path *path;
2655 struct extent_buffer *leaf;
2659 * If the inode is a free space inode, we can deadlock during commit
2660 * if we put it into the delayed code.
2662 * The data relocation inode should also be directly updated
2665 if (!btrfs_is_free_space_inode(root, inode)
2666 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2667 ret = btrfs_delayed_update_inode(trans, root, inode);
2669 btrfs_set_inode_last_trans(trans, inode);
2673 path = btrfs_alloc_path();
2677 path->leave_spinning = 1;
2678 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2686 btrfs_unlock_up_safe(path, 1);
2687 leaf = path->nodes[0];
2688 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2689 struct btrfs_inode_item);
2691 fill_inode_item(trans, leaf, inode_item, inode);
2692 btrfs_mark_buffer_dirty(leaf);
2693 btrfs_set_inode_last_trans(trans, inode);
2696 btrfs_free_path(path);
2701 * unlink helper that gets used here in inode.c and in the tree logging
2702 * recovery code. It remove a link in a directory with a given name, and
2703 * also drops the back refs in the inode to the directory
2705 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2706 struct btrfs_root *root,
2707 struct inode *dir, struct inode *inode,
2708 const char *name, int name_len)
2710 struct btrfs_path *path;
2712 struct extent_buffer *leaf;
2713 struct btrfs_dir_item *di;
2714 struct btrfs_key key;
2716 u64 ino = btrfs_ino(inode);
2717 u64 dir_ino = btrfs_ino(dir);
2719 path = btrfs_alloc_path();
2725 path->leave_spinning = 1;
2726 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2727 name, name_len, -1);
2736 leaf = path->nodes[0];
2737 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2738 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2741 btrfs_release_path(path);
2743 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2746 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2747 "inode %llu parent %llu\n", name_len, name,
2748 (unsigned long long)ino, (unsigned long long)dir_ino);
2752 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2756 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2758 BUG_ON(ret != 0 && ret != -ENOENT);
2760 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2765 btrfs_free_path(path);
2769 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2770 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2771 btrfs_update_inode(trans, root, dir);
2776 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2777 struct btrfs_root *root,
2778 struct inode *dir, struct inode *inode,
2779 const char *name, int name_len)
2782 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2784 btrfs_drop_nlink(inode);
2785 ret = btrfs_update_inode(trans, root, inode);
2791 /* helper to check if there is any shared block in the path */
2792 static int check_path_shared(struct btrfs_root *root,
2793 struct btrfs_path *path)
2795 struct extent_buffer *eb;
2799 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2802 if (!path->nodes[level])
2804 eb = path->nodes[level];
2805 if (!btrfs_block_can_be_shared(root, eb))
2807 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2816 * helper to start transaction for unlink and rmdir.
2818 * unlink and rmdir are special in btrfs, they do not always free space.
2819 * so in enospc case, we should make sure they will free space before
2820 * allowing them to use the global metadata reservation.
2822 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2823 struct dentry *dentry)
2825 struct btrfs_trans_handle *trans;
2826 struct btrfs_root *root = BTRFS_I(dir)->root;
2827 struct btrfs_path *path;
2828 struct btrfs_inode_ref *ref;
2829 struct btrfs_dir_item *di;
2830 struct inode *inode = dentry->d_inode;
2835 u64 ino = btrfs_ino(inode);
2836 u64 dir_ino = btrfs_ino(dir);
2838 trans = btrfs_start_transaction(root, 10);
2839 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2842 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2843 return ERR_PTR(-ENOSPC);
2845 /* check if there is someone else holds reference */
2846 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2847 return ERR_PTR(-ENOSPC);
2849 if (atomic_read(&inode->i_count) > 2)
2850 return ERR_PTR(-ENOSPC);
2852 if (xchg(&root->fs_info->enospc_unlink, 1))
2853 return ERR_PTR(-ENOSPC);
2855 path = btrfs_alloc_path();
2857 root->fs_info->enospc_unlink = 0;
2858 return ERR_PTR(-ENOMEM);
2861 trans = btrfs_start_transaction(root, 0);
2862 if (IS_ERR(trans)) {
2863 btrfs_free_path(path);
2864 root->fs_info->enospc_unlink = 0;
2868 path->skip_locking = 1;
2869 path->search_commit_root = 1;
2871 ret = btrfs_lookup_inode(trans, root, path,
2872 &BTRFS_I(dir)->location, 0);
2878 if (check_path_shared(root, path))
2883 btrfs_release_path(path);
2885 ret = btrfs_lookup_inode(trans, root, path,
2886 &BTRFS_I(inode)->location, 0);
2892 if (check_path_shared(root, path))
2897 btrfs_release_path(path);
2899 if (ret == 0 && S_ISREG(inode->i_mode)) {
2900 ret = btrfs_lookup_file_extent(trans, root, path,
2907 if (check_path_shared(root, path))
2909 btrfs_release_path(path);
2917 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2918 dentry->d_name.name, dentry->d_name.len, 0);
2924 if (check_path_shared(root, path))
2930 btrfs_release_path(path);
2932 ref = btrfs_lookup_inode_ref(trans, root, path,
2933 dentry->d_name.name, dentry->d_name.len,
2940 if (check_path_shared(root, path))
2942 index = btrfs_inode_ref_index(path->nodes[0], ref);
2943 btrfs_release_path(path);
2946 * This is a commit root search, if we can lookup inode item and other
2947 * relative items in the commit root, it means the transaction of
2948 * dir/file creation has been committed, and the dir index item that we
2949 * delay to insert has also been inserted into the commit root. So
2950 * we needn't worry about the delayed insertion of the dir index item
2953 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2954 dentry->d_name.name, dentry->d_name.len, 0);
2959 BUG_ON(ret == -ENOENT);
2960 if (check_path_shared(root, path))
2965 btrfs_free_path(path);
2967 btrfs_end_transaction(trans, root);
2968 root->fs_info->enospc_unlink = 0;
2969 return ERR_PTR(err);
2972 trans->block_rsv = &root->fs_info->global_block_rsv;
2976 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2977 struct btrfs_root *root)
2979 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2980 BUG_ON(!root->fs_info->enospc_unlink);
2981 root->fs_info->enospc_unlink = 0;
2983 btrfs_end_transaction_throttle(trans, root);
2986 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2988 struct btrfs_root *root = BTRFS_I(dir)->root;
2989 struct btrfs_trans_handle *trans;
2990 struct inode *inode = dentry->d_inode;
2992 unsigned long nr = 0;
2994 trans = __unlink_start_trans(dir, dentry);
2996 return PTR_ERR(trans);
2998 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3000 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3001 dentry->d_name.name, dentry->d_name.len);
3005 if (inode->i_nlink == 0) {
3006 ret = btrfs_orphan_add(trans, inode);
3012 nr = trans->blocks_used;
3013 __unlink_end_trans(trans, root);
3014 btrfs_btree_balance_dirty(root, nr);
3018 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3019 struct btrfs_root *root,
3020 struct inode *dir, u64 objectid,
3021 const char *name, int name_len)
3023 struct btrfs_path *path;
3024 struct extent_buffer *leaf;
3025 struct btrfs_dir_item *di;
3026 struct btrfs_key key;
3029 u64 dir_ino = btrfs_ino(dir);
3031 path = btrfs_alloc_path();
3035 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3036 name, name_len, -1);
3037 BUG_ON(IS_ERR_OR_NULL(di));
3039 leaf = path->nodes[0];
3040 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3041 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3042 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3044 btrfs_release_path(path);
3046 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3047 objectid, root->root_key.objectid,
3048 dir_ino, &index, name, name_len);
3050 BUG_ON(ret != -ENOENT);
3051 di = btrfs_search_dir_index_item(root, path, dir_ino,
3053 BUG_ON(IS_ERR_OR_NULL(di));
3055 leaf = path->nodes[0];
3056 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3057 btrfs_release_path(path);
3060 btrfs_release_path(path);
3062 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3065 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3066 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3067 ret = btrfs_update_inode(trans, root, dir);
3070 btrfs_free_path(path);
3074 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3076 struct inode *inode = dentry->d_inode;
3078 struct btrfs_root *root = BTRFS_I(dir)->root;
3079 struct btrfs_trans_handle *trans;
3080 unsigned long nr = 0;
3082 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3083 btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3086 trans = __unlink_start_trans(dir, dentry);
3088 return PTR_ERR(trans);
3090 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3091 err = btrfs_unlink_subvol(trans, root, dir,
3092 BTRFS_I(inode)->location.objectid,
3093 dentry->d_name.name,
3094 dentry->d_name.len);
3098 err = btrfs_orphan_add(trans, inode);
3102 /* now the directory is empty */
3103 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3104 dentry->d_name.name, dentry->d_name.len);
3106 btrfs_i_size_write(inode, 0);
3108 nr = trans->blocks_used;
3109 __unlink_end_trans(trans, root);
3110 btrfs_btree_balance_dirty(root, nr);
3116 * this can truncate away extent items, csum items and directory items.
3117 * It starts at a high offset and removes keys until it can't find
3118 * any higher than new_size
3120 * csum items that cross the new i_size are truncated to the new size
3123 * min_type is the minimum key type to truncate down to. If set to 0, this
3124 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3126 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3127 struct btrfs_root *root,
3128 struct inode *inode,
3129 u64 new_size, u32 min_type)
3131 struct btrfs_path *path;
3132 struct extent_buffer *leaf;
3133 struct btrfs_file_extent_item *fi;
3134 struct btrfs_key key;
3135 struct btrfs_key found_key;
3136 u64 extent_start = 0;
3137 u64 extent_num_bytes = 0;
3138 u64 extent_offset = 0;
3140 u64 mask = root->sectorsize - 1;
3141 u32 found_type = (u8)-1;
3144 int pending_del_nr = 0;
3145 int pending_del_slot = 0;
3146 int extent_type = -1;
3150 u64 ino = btrfs_ino(inode);
3152 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3154 path = btrfs_alloc_path();
3159 if (root->ref_cows || root == root->fs_info->tree_root)
3160 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3163 * This function is also used to drop the items in the log tree before
3164 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3165 * it is used to drop the loged items. So we shouldn't kill the delayed
3168 if (min_type == 0 && root == BTRFS_I(inode)->root)
3169 btrfs_kill_delayed_inode_items(inode);
3172 key.offset = (u64)-1;
3176 path->leave_spinning = 1;
3177 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3184 /* there are no items in the tree for us to truncate, we're
3187 if (path->slots[0] == 0)
3194 leaf = path->nodes[0];
3195 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3196 found_type = btrfs_key_type(&found_key);
3199 if (found_key.objectid != ino)
3202 if (found_type < min_type)
3205 item_end = found_key.offset;
3206 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3207 fi = btrfs_item_ptr(leaf, path->slots[0],
3208 struct btrfs_file_extent_item);
3209 extent_type = btrfs_file_extent_type(leaf, fi);
3210 encoding = btrfs_file_extent_compression(leaf, fi);
3211 encoding |= btrfs_file_extent_encryption(leaf, fi);
3212 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3214 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3216 btrfs_file_extent_num_bytes(leaf, fi);
3217 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3218 item_end += btrfs_file_extent_inline_len(leaf,
3223 if (found_type > min_type) {
3226 if (item_end < new_size)
3228 if (found_key.offset >= new_size)
3234 /* FIXME, shrink the extent if the ref count is only 1 */
3235 if (found_type != BTRFS_EXTENT_DATA_KEY)
3238 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3240 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3241 if (!del_item && !encoding) {
3242 u64 orig_num_bytes =
3243 btrfs_file_extent_num_bytes(leaf, fi);
3244 extent_num_bytes = new_size -
3245 found_key.offset + root->sectorsize - 1;
3246 extent_num_bytes = extent_num_bytes &
3247 ~((u64)root->sectorsize - 1);
3248 btrfs_set_file_extent_num_bytes(leaf, fi,
3250 num_dec = (orig_num_bytes -
3252 if (root->ref_cows && extent_start != 0)
3253 inode_sub_bytes(inode, num_dec);
3254 btrfs_mark_buffer_dirty(leaf);
3257 btrfs_file_extent_disk_num_bytes(leaf,
3259 extent_offset = found_key.offset -
3260 btrfs_file_extent_offset(leaf, fi);
3262 /* FIXME blocksize != 4096 */
3263 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3264 if (extent_start != 0) {
3267 inode_sub_bytes(inode, num_dec);
3270 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3272 * we can't truncate inline items that have had
3276 btrfs_file_extent_compression(leaf, fi) == 0 &&
3277 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3278 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3279 u32 size = new_size - found_key.offset;
3281 if (root->ref_cows) {
3282 inode_sub_bytes(inode, item_end + 1 -
3286 btrfs_file_extent_calc_inline_size(size);
3287 ret = btrfs_truncate_item(trans, root, path,
3289 } else if (root->ref_cows) {
3290 inode_sub_bytes(inode, item_end + 1 -
3296 if (!pending_del_nr) {
3297 /* no pending yet, add ourselves */
3298 pending_del_slot = path->slots[0];
3300 } else if (pending_del_nr &&
3301 path->slots[0] + 1 == pending_del_slot) {
3302 /* hop on the pending chunk */
3304 pending_del_slot = path->slots[0];
3311 if (found_extent && (root->ref_cows ||
3312 root == root->fs_info->tree_root)) {
3313 btrfs_set_path_blocking(path);
3314 ret = btrfs_free_extent(trans, root, extent_start,
3315 extent_num_bytes, 0,
3316 btrfs_header_owner(leaf),
3317 ino, extent_offset);
3321 if (found_type == BTRFS_INODE_ITEM_KEY)
3324 if (path->slots[0] == 0 ||
3325 path->slots[0] != pending_del_slot) {
3326 if (root->ref_cows &&
3327 BTRFS_I(inode)->location.objectid !=
3328 BTRFS_FREE_INO_OBJECTID) {
3332 if (pending_del_nr) {
3333 ret = btrfs_del_items(trans, root, path,
3339 btrfs_release_path(path);
3346 if (pending_del_nr) {
3347 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3351 btrfs_free_path(path);
3356 * taken from block_truncate_page, but does cow as it zeros out
3357 * any bytes left in the last page in the file.
3359 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3361 struct inode *inode = mapping->host;
3362 struct btrfs_root *root = BTRFS_I(inode)->root;
3363 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3364 struct btrfs_ordered_extent *ordered;
3365 struct extent_state *cached_state = NULL;
3367 u32 blocksize = root->sectorsize;
3368 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3369 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3375 if ((offset & (blocksize - 1)) == 0)
3377 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3383 page = find_or_create_page(mapping, index, GFP_NOFS);
3385 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3389 page_start = page_offset(page);
3390 page_end = page_start + PAGE_CACHE_SIZE - 1;
3392 if (!PageUptodate(page)) {
3393 ret = btrfs_readpage(NULL, page);
3395 if (page->mapping != mapping) {
3397 page_cache_release(page);
3400 if (!PageUptodate(page)) {
3405 wait_on_page_writeback(page);
3407 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3409 set_page_extent_mapped(page);
3411 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3413 unlock_extent_cached(io_tree, page_start, page_end,
3414 &cached_state, GFP_NOFS);
3416 page_cache_release(page);
3417 btrfs_start_ordered_extent(inode, ordered, 1);
3418 btrfs_put_ordered_extent(ordered);
3422 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3423 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3424 0, 0, &cached_state, GFP_NOFS);
3426 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3429 unlock_extent_cached(io_tree, page_start, page_end,
3430 &cached_state, GFP_NOFS);
3435 if (offset != PAGE_CACHE_SIZE) {
3437 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3438 flush_dcache_page(page);
3441 ClearPageChecked(page);
3442 set_page_dirty(page);
3443 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3448 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3450 page_cache_release(page);
3456 * This function puts in dummy file extents for the area we're creating a hole
3457 * for. So if we are truncating this file to a larger size we need to insert
3458 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3459 * the range between oldsize and size
3461 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3463 struct btrfs_trans_handle *trans;
3464 struct btrfs_root *root = BTRFS_I(inode)->root;
3465 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3466 struct extent_map *em = NULL;
3467 struct extent_state *cached_state = NULL;
3468 u64 mask = root->sectorsize - 1;
3469 u64 hole_start = (oldsize + mask) & ~mask;
3470 u64 block_end = (size + mask) & ~mask;
3476 if (size <= hole_start)
3480 struct btrfs_ordered_extent *ordered;
3481 btrfs_wait_ordered_range(inode, hole_start,
3482 block_end - hole_start);
3483 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3484 &cached_state, GFP_NOFS);
3485 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3488 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3489 &cached_state, GFP_NOFS);
3490 btrfs_put_ordered_extent(ordered);
3493 cur_offset = hole_start;
3495 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3496 block_end - cur_offset, 0);
3497 BUG_ON(IS_ERR_OR_NULL(em));
3498 last_byte = min(extent_map_end(em), block_end);
3499 last_byte = (last_byte + mask) & ~mask;
3500 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3502 hole_size = last_byte - cur_offset;
3504 trans = btrfs_start_transaction(root, 2);
3505 if (IS_ERR(trans)) {
3506 err = PTR_ERR(trans);
3510 err = btrfs_drop_extents(trans, inode, cur_offset,
3511 cur_offset + hole_size,
3516 err = btrfs_insert_file_extent(trans, root,
3517 btrfs_ino(inode), cur_offset, 0,
3518 0, hole_size, 0, hole_size,
3523 btrfs_drop_extent_cache(inode, hole_start,
3526 btrfs_end_transaction(trans, root);
3528 free_extent_map(em);
3530 cur_offset = last_byte;
3531 if (cur_offset >= block_end)
3535 free_extent_map(em);
3536 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3541 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3543 loff_t oldsize = i_size_read(inode);
3546 if (newsize == oldsize)
3549 if (newsize > oldsize) {
3550 i_size_write(inode, newsize);
3551 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3552 truncate_pagecache(inode, oldsize, newsize);
3553 ret = btrfs_cont_expand(inode, oldsize, newsize);
3555 btrfs_setsize(inode, oldsize);
3559 mark_inode_dirty(inode);
3563 * We're truncating a file that used to have good data down to
3564 * zero. Make sure it gets into the ordered flush list so that
3565 * any new writes get down to disk quickly.
3568 BTRFS_I(inode)->ordered_data_close = 1;
3570 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3571 truncate_setsize(inode, newsize);
3572 ret = btrfs_truncate(inode);
3578 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3580 struct inode *inode = dentry->d_inode;
3581 struct btrfs_root *root = BTRFS_I(inode)->root;
3584 if (btrfs_root_readonly(root))
3587 err = inode_change_ok(inode, attr);
3591 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3592 err = btrfs_setsize(inode, attr->ia_size);
3597 if (attr->ia_valid) {
3598 setattr_copy(inode, attr);
3599 mark_inode_dirty(inode);
3601 if (attr->ia_valid & ATTR_MODE)
3602 err = btrfs_acl_chmod(inode);
3608 void btrfs_evict_inode(struct inode *inode)
3610 struct btrfs_trans_handle *trans;
3611 struct btrfs_root *root = BTRFS_I(inode)->root;
3615 trace_btrfs_inode_evict(inode);
3617 truncate_inode_pages(&inode->i_data, 0);
3618 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3619 btrfs_is_free_space_inode(root, inode)))
3622 if (is_bad_inode(inode)) {
3623 btrfs_orphan_del(NULL, inode);
3626 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3627 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3629 if (root->fs_info->log_root_recovering) {
3630 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3634 if (inode->i_nlink > 0) {
3635 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3639 btrfs_i_size_write(inode, 0);
3642 trans = btrfs_join_transaction(root);
3643 BUG_ON(IS_ERR(trans));
3644 trans->block_rsv = root->orphan_block_rsv;
3646 ret = btrfs_block_rsv_check(trans, root,
3647 root->orphan_block_rsv, 0, 5);
3649 BUG_ON(ret != -EAGAIN);
3650 ret = btrfs_commit_transaction(trans, root);
3655 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3659 nr = trans->blocks_used;
3660 btrfs_end_transaction(trans, root);
3662 btrfs_btree_balance_dirty(root, nr);
3667 ret = btrfs_orphan_del(trans, inode);
3671 if (!(root == root->fs_info->tree_root ||
3672 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3673 btrfs_return_ino(root, btrfs_ino(inode));
3675 nr = trans->blocks_used;
3676 btrfs_end_transaction(trans, root);
3677 btrfs_btree_balance_dirty(root, nr);
3679 end_writeback(inode);
3684 * this returns the key found in the dir entry in the location pointer.
3685 * If no dir entries were found, location->objectid is 0.
3687 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3688 struct btrfs_key *location)
3690 const char *name = dentry->d_name.name;
3691 int namelen = dentry->d_name.len;
3692 struct btrfs_dir_item *di;
3693 struct btrfs_path *path;
3694 struct btrfs_root *root = BTRFS_I(dir)->root;
3697 path = btrfs_alloc_path();
3701 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3706 if (IS_ERR_OR_NULL(di))
3709 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3711 btrfs_free_path(path);
3714 location->objectid = 0;
3719 * when we hit a tree root in a directory, the btrfs part of the inode
3720 * needs to be changed to reflect the root directory of the tree root. This
3721 * is kind of like crossing a mount point.
3723 static int fixup_tree_root_location(struct btrfs_root *root,
3725 struct dentry *dentry,
3726 struct btrfs_key *location,
3727 struct btrfs_root **sub_root)
3729 struct btrfs_path *path;
3730 struct btrfs_root *new_root;
3731 struct btrfs_root_ref *ref;
3732 struct extent_buffer *leaf;
3736 path = btrfs_alloc_path();
3743 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3744 BTRFS_I(dir)->root->root_key.objectid,
3745 location->objectid);
3752 leaf = path->nodes[0];
3753 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3754 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3755 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3758 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3759 (unsigned long)(ref + 1),
3760 dentry->d_name.len);
3764 btrfs_release_path(path);
3766 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3767 if (IS_ERR(new_root)) {
3768 err = PTR_ERR(new_root);
3772 if (btrfs_root_refs(&new_root->root_item) == 0) {
3777 *sub_root = new_root;
3778 location->objectid = btrfs_root_dirid(&new_root->root_item);
3779 location->type = BTRFS_INODE_ITEM_KEY;
3780 location->offset = 0;
3783 btrfs_free_path(path);
3787 static void inode_tree_add(struct inode *inode)
3789 struct btrfs_root *root = BTRFS_I(inode)->root;
3790 struct btrfs_inode *entry;
3792 struct rb_node *parent;
3793 u64 ino = btrfs_ino(inode);
3795 p = &root->inode_tree.rb_node;
3798 if (inode_unhashed(inode))
3801 spin_lock(&root->inode_lock);
3804 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3806 if (ino < btrfs_ino(&entry->vfs_inode))
3807 p = &parent->rb_left;
3808 else if (ino > btrfs_ino(&entry->vfs_inode))
3809 p = &parent->rb_right;
3811 WARN_ON(!(entry->vfs_inode.i_state &
3812 (I_WILL_FREE | I_FREEING)));
3813 rb_erase(parent, &root->inode_tree);
3814 RB_CLEAR_NODE(parent);
3815 spin_unlock(&root->inode_lock);
3819 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3820 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3821 spin_unlock(&root->inode_lock);
3824 static void inode_tree_del(struct inode *inode)
3826 struct btrfs_root *root = BTRFS_I(inode)->root;
3829 spin_lock(&root->inode_lock);
3830 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3831 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3832 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3833 empty = RB_EMPTY_ROOT(&root->inode_tree);
3835 spin_unlock(&root->inode_lock);
3838 * Free space cache has inodes in the tree root, but the tree root has a
3839 * root_refs of 0, so this could end up dropping the tree root as a
3840 * snapshot, so we need the extra !root->fs_info->tree_root check to
3841 * make sure we don't drop it.
3843 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3844 root != root->fs_info->tree_root) {
3845 synchronize_srcu(&root->fs_info->subvol_srcu);
3846 spin_lock(&root->inode_lock);
3847 empty = RB_EMPTY_ROOT(&root->inode_tree);
3848 spin_unlock(&root->inode_lock);
3850 btrfs_add_dead_root(root);
3854 int btrfs_invalidate_inodes(struct btrfs_root *root)
3856 struct rb_node *node;
3857 struct rb_node *prev;
3858 struct btrfs_inode *entry;
3859 struct inode *inode;
3862 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3864 spin_lock(&root->inode_lock);
3866 node = root->inode_tree.rb_node;
3870 entry = rb_entry(node, struct btrfs_inode, rb_node);
3872 if (objectid < btrfs_ino(&entry->vfs_inode))
3873 node = node->rb_left;
3874 else if (objectid > btrfs_ino(&entry->vfs_inode))
3875 node = node->rb_right;
3881 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3882 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3886 prev = rb_next(prev);
3890 entry = rb_entry(node, struct btrfs_inode, rb_node);
3891 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3892 inode = igrab(&entry->vfs_inode);
3894 spin_unlock(&root->inode_lock);
3895 if (atomic_read(&inode->i_count) > 1)
3896 d_prune_aliases(inode);
3898 * btrfs_drop_inode will have it removed from
3899 * the inode cache when its usage count
3904 spin_lock(&root->inode_lock);
3908 if (cond_resched_lock(&root->inode_lock))
3911 node = rb_next(node);
3913 spin_unlock(&root->inode_lock);
3917 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3919 struct btrfs_iget_args *args = p;
3920 inode->i_ino = args->ino;
3921 BTRFS_I(inode)->root = args->root;
3922 btrfs_set_inode_space_info(args->root, inode);
3926 static int btrfs_find_actor(struct inode *inode, void *opaque)
3928 struct btrfs_iget_args *args = opaque;
3929 return args->ino == btrfs_ino(inode) &&
3930 args->root == BTRFS_I(inode)->root;
3933 static struct inode *btrfs_iget_locked(struct super_block *s,
3935 struct btrfs_root *root)
3937 struct inode *inode;
3938 struct btrfs_iget_args args;
3939 args.ino = objectid;
3942 inode = iget5_locked(s, objectid, btrfs_find_actor,
3943 btrfs_init_locked_inode,
3948 /* Get an inode object given its location and corresponding root.
3949 * Returns in *is_new if the inode was read from disk
3951 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3952 struct btrfs_root *root, int *new)
3954 struct inode *inode;
3957 inode = btrfs_iget_locked(s, location->objectid, root);
3959 return ERR_PTR(-ENOMEM);
3961 if (inode->i_state & I_NEW) {
3962 BTRFS_I(inode)->root = root;
3963 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3964 btrfs_read_locked_inode(inode);
3965 if (!is_bad_inode(inode)) {
3966 inode_tree_add(inode);
3967 unlock_new_inode(inode);
3977 inode = ERR_PTR(-ESTALE);
3983 static struct inode *new_simple_dir(struct super_block *s,
3984 struct btrfs_key *key,
3985 struct btrfs_root *root)
3987 struct inode *inode = new_inode(s);
3990 return ERR_PTR(-ENOMEM);
3992 BTRFS_I(inode)->root = root;
3993 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3994 BTRFS_I(inode)->dummy_inode = 1;
3996 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3997 inode->i_op = &simple_dir_inode_operations;
3998 inode->i_fop = &simple_dir_operations;
3999 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4000 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4005 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4007 struct inode *inode;
4008 struct btrfs_root *root = BTRFS_I(dir)->root;
4009 struct btrfs_root *sub_root = root;
4010 struct btrfs_key location;
4014 if (dentry->d_name.len > BTRFS_NAME_LEN)
4015 return ERR_PTR(-ENAMETOOLONG);
4017 if (unlikely(d_need_lookup(dentry))) {
4018 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4019 kfree(dentry->d_fsdata);
4020 dentry->d_fsdata = NULL;
4021 d_clear_need_lookup(dentry);
4023 ret = btrfs_inode_by_name(dir, dentry, &location);
4027 return ERR_PTR(ret);
4029 if (location.objectid == 0)
4032 if (location.type == BTRFS_INODE_ITEM_KEY) {
4033 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4037 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4039 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4040 ret = fixup_tree_root_location(root, dir, dentry,
4041 &location, &sub_root);
4044 inode = ERR_PTR(ret);
4046 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4048 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4050 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4052 if (!IS_ERR(inode) && root != sub_root) {
4053 down_read(&root->fs_info->cleanup_work_sem);
4054 if (!(inode->i_sb->s_flags & MS_RDONLY))
4055 ret = btrfs_orphan_cleanup(sub_root);
4056 up_read(&root->fs_info->cleanup_work_sem);
4058 inode = ERR_PTR(ret);
4064 static int btrfs_dentry_delete(const struct dentry *dentry)
4066 struct btrfs_root *root;
4068 if (!dentry->d_inode && !IS_ROOT(dentry))
4069 dentry = dentry->d_parent;
4071 if (dentry->d_inode) {
4072 root = BTRFS_I(dentry->d_inode)->root;
4073 if (btrfs_root_refs(&root->root_item) == 0)
4079 static void btrfs_dentry_release(struct dentry *dentry)
4081 if (dentry->d_fsdata)
4082 kfree(dentry->d_fsdata);
4085 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4086 struct nameidata *nd)
4088 return d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4091 unsigned char btrfs_filetype_table[] = {
4092 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4095 static int btrfs_real_readdir(struct file *filp, void *dirent,
4098 struct inode *inode = filp->f_dentry->d_inode;
4099 struct btrfs_root *root = BTRFS_I(inode)->root;
4100 struct btrfs_item *item;
4101 struct btrfs_dir_item *di;
4102 struct btrfs_key key;
4103 struct btrfs_key found_key;
4104 struct btrfs_path *path;
4105 struct list_head ins_list;
4106 struct list_head del_list;
4109 struct extent_buffer *leaf;
4111 unsigned char d_type;
4116 int key_type = BTRFS_DIR_INDEX_KEY;
4120 int is_curr = 0; /* filp->f_pos points to the current index? */
4122 /* FIXME, use a real flag for deciding about the key type */
4123 if (root->fs_info->tree_root == root)
4124 key_type = BTRFS_DIR_ITEM_KEY;
4126 /* special case for "." */
4127 if (filp->f_pos == 0) {
4128 over = filldir(dirent, ".", 1, 1, btrfs_ino(inode), DT_DIR);
4133 /* special case for .., just use the back ref */
4134 if (filp->f_pos == 1) {
4135 u64 pino = parent_ino(filp->f_path.dentry);
4136 over = filldir(dirent, "..", 2,
4142 path = btrfs_alloc_path();
4148 if (key_type == BTRFS_DIR_INDEX_KEY) {
4149 INIT_LIST_HEAD(&ins_list);
4150 INIT_LIST_HEAD(&del_list);
4151 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4154 btrfs_set_key_type(&key, key_type);
4155 key.offset = filp->f_pos;
4156 key.objectid = btrfs_ino(inode);
4158 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4163 leaf = path->nodes[0];
4164 slot = path->slots[0];
4165 if (slot >= btrfs_header_nritems(leaf)) {
4166 ret = btrfs_next_leaf(root, path);
4174 item = btrfs_item_nr(leaf, slot);
4175 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4177 if (found_key.objectid != key.objectid)
4179 if (btrfs_key_type(&found_key) != key_type)
4181 if (found_key.offset < filp->f_pos)
4183 if (key_type == BTRFS_DIR_INDEX_KEY &&
4184 btrfs_should_delete_dir_index(&del_list,
4188 filp->f_pos = found_key.offset;
4191 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4193 di_total = btrfs_item_size(leaf, item);
4195 while (di_cur < di_total) {
4196 struct btrfs_key location;
4199 if (verify_dir_item(root, leaf, di))
4202 name_len = btrfs_dir_name_len(leaf, di);
4203 if (name_len <= sizeof(tmp_name)) {
4204 name_ptr = tmp_name;
4206 name_ptr = kmalloc(name_len, GFP_NOFS);
4212 read_extent_buffer(leaf, name_ptr,
4213 (unsigned long)(di + 1), name_len);
4215 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4216 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4220 q.hash = full_name_hash(q.name, q.len);
4221 tmp = d_lookup(filp->f_dentry, &q);
4223 struct btrfs_key *newkey;
4225 newkey = kzalloc(sizeof(struct btrfs_key),
4229 tmp = d_alloc(filp->f_dentry, &q);
4235 memcpy(newkey, &location,
4236 sizeof(struct btrfs_key));
4237 tmp->d_fsdata = newkey;
4238 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4245 /* is this a reference to our own snapshot? If so
4248 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4249 location.objectid == root->root_key.objectid) {
4253 over = filldir(dirent, name_ptr, name_len,
4254 found_key.offset, location.objectid,
4258 if (name_ptr != tmp_name)
4263 di_len = btrfs_dir_name_len(leaf, di) +
4264 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4266 di = (struct btrfs_dir_item *)((char *)di + di_len);
4272 if (key_type == BTRFS_DIR_INDEX_KEY) {
4275 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4281 /* Reached end of directory/root. Bump pos past the last item. */
4282 if (key_type == BTRFS_DIR_INDEX_KEY)
4284 * 32-bit glibc will use getdents64, but then strtol -
4285 * so the last number we can serve is this.
4287 filp->f_pos = 0x7fffffff;
4293 if (key_type == BTRFS_DIR_INDEX_KEY)
4294 btrfs_put_delayed_items(&ins_list, &del_list);
4295 btrfs_free_path(path);
4299 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4301 struct btrfs_root *root = BTRFS_I(inode)->root;
4302 struct btrfs_trans_handle *trans;
4304 bool nolock = false;
4306 if (BTRFS_I(inode)->dummy_inode)
4309 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4312 if (wbc->sync_mode == WB_SYNC_ALL) {
4314 trans = btrfs_join_transaction_nolock(root);
4316 trans = btrfs_join_transaction(root);
4318 return PTR_ERR(trans);
4320 ret = btrfs_end_transaction_nolock(trans, root);
4322 ret = btrfs_commit_transaction(trans, root);
4328 * This is somewhat expensive, updating the tree every time the
4329 * inode changes. But, it is most likely to find the inode in cache.
4330 * FIXME, needs more benchmarking...there are no reasons other than performance
4331 * to keep or drop this code.
4333 void btrfs_dirty_inode(struct inode *inode, int flags)
4335 struct btrfs_root *root = BTRFS_I(inode)->root;
4336 struct btrfs_trans_handle *trans;
4339 if (BTRFS_I(inode)->dummy_inode)
4342 trans = btrfs_join_transaction(root);
4343 BUG_ON(IS_ERR(trans));
4345 ret = btrfs_update_inode(trans, root, inode);
4346 if (ret && ret == -ENOSPC) {
4347 /* whoops, lets try again with the full transaction */
4348 btrfs_end_transaction(trans, root);
4349 trans = btrfs_start_transaction(root, 1);
4350 if (IS_ERR(trans)) {
4351 printk_ratelimited(KERN_ERR "btrfs: fail to "
4352 "dirty inode %llu error %ld\n",
4353 (unsigned long long)btrfs_ino(inode),
4358 ret = btrfs_update_inode(trans, root, inode);
4360 printk_ratelimited(KERN_ERR "btrfs: fail to "
4361 "dirty inode %llu error %d\n",
4362 (unsigned long long)btrfs_ino(inode),
4366 btrfs_end_transaction(trans, root);
4367 if (BTRFS_I(inode)->delayed_node)
4368 btrfs_balance_delayed_items(root);
4372 * find the highest existing sequence number in a directory
4373 * and then set the in-memory index_cnt variable to reflect
4374 * free sequence numbers
4376 static int btrfs_set_inode_index_count(struct inode *inode)
4378 struct btrfs_root *root = BTRFS_I(inode)->root;
4379 struct btrfs_key key, found_key;
4380 struct btrfs_path *path;
4381 struct extent_buffer *leaf;
4384 key.objectid = btrfs_ino(inode);
4385 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4386 key.offset = (u64)-1;
4388 path = btrfs_alloc_path();
4392 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4395 /* FIXME: we should be able to handle this */
4401 * MAGIC NUMBER EXPLANATION:
4402 * since we search a directory based on f_pos we have to start at 2
4403 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4404 * else has to start at 2
4406 if (path->slots[0] == 0) {
4407 BTRFS_I(inode)->index_cnt = 2;
4413 leaf = path->nodes[0];
4414 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4416 if (found_key.objectid != btrfs_ino(inode) ||
4417 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4418 BTRFS_I(inode)->index_cnt = 2;
4422 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4424 btrfs_free_path(path);
4429 * helper to find a free sequence number in a given directory. This current
4430 * code is very simple, later versions will do smarter things in the btree
4432 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4436 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4437 ret = btrfs_inode_delayed_dir_index_count(dir);
4439 ret = btrfs_set_inode_index_count(dir);
4445 *index = BTRFS_I(dir)->index_cnt;
4446 BTRFS_I(dir)->index_cnt++;
4451 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4452 struct btrfs_root *root,
4454 const char *name, int name_len,
4455 u64 ref_objectid, u64 objectid, int mode,
4458 struct inode *inode;
4459 struct btrfs_inode_item *inode_item;
4460 struct btrfs_key *location;
4461 struct btrfs_path *path;
4462 struct btrfs_inode_ref *ref;
4463 struct btrfs_key key[2];
4469 path = btrfs_alloc_path();
4471 return ERR_PTR(-ENOMEM);
4473 inode = new_inode(root->fs_info->sb);
4475 btrfs_free_path(path);
4476 return ERR_PTR(-ENOMEM);
4480 * we have to initialize this early, so we can reclaim the inode
4481 * number if we fail afterwards in this function.
4483 inode->i_ino = objectid;
4486 trace_btrfs_inode_request(dir);
4488 ret = btrfs_set_inode_index(dir, index);
4490 btrfs_free_path(path);
4492 return ERR_PTR(ret);
4496 * index_cnt is ignored for everything but a dir,
4497 * btrfs_get_inode_index_count has an explanation for the magic
4500 BTRFS_I(inode)->index_cnt = 2;
4501 BTRFS_I(inode)->root = root;
4502 BTRFS_I(inode)->generation = trans->transid;
4503 inode->i_generation = BTRFS_I(inode)->generation;
4504 btrfs_set_inode_space_info(root, inode);
4511 key[0].objectid = objectid;
4512 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4515 key[1].objectid = objectid;
4516 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4517 key[1].offset = ref_objectid;
4519 sizes[0] = sizeof(struct btrfs_inode_item);
4520 sizes[1] = name_len + sizeof(*ref);
4522 path->leave_spinning = 1;
4523 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4527 inode_init_owner(inode, dir, mode);
4528 inode_set_bytes(inode, 0);
4529 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4530 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4531 struct btrfs_inode_item);
4532 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4534 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4535 struct btrfs_inode_ref);
4536 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4537 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4538 ptr = (unsigned long)(ref + 1);
4539 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4541 btrfs_mark_buffer_dirty(path->nodes[0]);
4542 btrfs_free_path(path);
4544 location = &BTRFS_I(inode)->location;
4545 location->objectid = objectid;
4546 location->offset = 0;
4547 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4549 btrfs_inherit_iflags(inode, dir);
4551 if (S_ISREG(mode)) {
4552 if (btrfs_test_opt(root, NODATASUM))
4553 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4554 if (btrfs_test_opt(root, NODATACOW) ||
4555 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4556 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4559 insert_inode_hash(inode);
4560 inode_tree_add(inode);
4562 trace_btrfs_inode_new(inode);
4563 btrfs_set_inode_last_trans(trans, inode);
4568 BTRFS_I(dir)->index_cnt--;
4569 btrfs_free_path(path);
4571 return ERR_PTR(ret);
4574 static inline u8 btrfs_inode_type(struct inode *inode)
4576 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4580 * utility function to add 'inode' into 'parent_inode' with
4581 * a give name and a given sequence number.
4582 * if 'add_backref' is true, also insert a backref from the
4583 * inode to the parent directory.
4585 int btrfs_add_link(struct btrfs_trans_handle *trans,
4586 struct inode *parent_inode, struct inode *inode,
4587 const char *name, int name_len, int add_backref, u64 index)
4590 struct btrfs_key key;
4591 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4592 u64 ino = btrfs_ino(inode);
4593 u64 parent_ino = btrfs_ino(parent_inode);
4595 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4596 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4599 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4603 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4604 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4605 key.objectid, root->root_key.objectid,
4606 parent_ino, index, name, name_len);
4607 } else if (add_backref) {
4608 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4613 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4615 btrfs_inode_type(inode), index);
4618 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4620 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4621 ret = btrfs_update_inode(trans, root, parent_inode);
4626 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4627 struct inode *dir, struct dentry *dentry,
4628 struct inode *inode, int backref, u64 index)
4630 int err = btrfs_add_link(trans, dir, inode,
4631 dentry->d_name.name, dentry->d_name.len,
4634 d_instantiate(dentry, inode);
4642 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4643 int mode, dev_t rdev)
4645 struct btrfs_trans_handle *trans;
4646 struct btrfs_root *root = BTRFS_I(dir)->root;
4647 struct inode *inode = NULL;
4651 unsigned long nr = 0;
4654 if (!new_valid_dev(rdev))
4658 * 2 for inode item and ref
4660 * 1 for xattr if selinux is on
4662 trans = btrfs_start_transaction(root, 5);
4664 return PTR_ERR(trans);
4666 err = btrfs_find_free_ino(root, &objectid);
4670 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4671 dentry->d_name.len, btrfs_ino(dir), objectid,
4673 if (IS_ERR(inode)) {
4674 err = PTR_ERR(inode);
4678 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4684 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4688 inode->i_op = &btrfs_special_inode_operations;
4689 init_special_inode(inode, inode->i_mode, rdev);
4690 btrfs_update_inode(trans, root, inode);
4693 nr = trans->blocks_used;
4694 btrfs_end_transaction_throttle(trans, root);
4695 btrfs_btree_balance_dirty(root, nr);
4697 inode_dec_link_count(inode);
4703 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4704 int mode, struct nameidata *nd)
4706 struct btrfs_trans_handle *trans;
4707 struct btrfs_root *root = BTRFS_I(dir)->root;
4708 struct inode *inode = NULL;
4711 unsigned long nr = 0;
4716 * 2 for inode item and ref
4718 * 1 for xattr if selinux is on
4720 trans = btrfs_start_transaction(root, 5);
4722 return PTR_ERR(trans);
4724 err = btrfs_find_free_ino(root, &objectid);
4728 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4729 dentry->d_name.len, btrfs_ino(dir), objectid,
4731 if (IS_ERR(inode)) {
4732 err = PTR_ERR(inode);
4736 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4742 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4746 inode->i_mapping->a_ops = &btrfs_aops;
4747 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4748 inode->i_fop = &btrfs_file_operations;
4749 inode->i_op = &btrfs_file_inode_operations;
4750 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4753 nr = trans->blocks_used;
4754 btrfs_end_transaction_throttle(trans, root);
4756 inode_dec_link_count(inode);
4759 btrfs_btree_balance_dirty(root, nr);
4763 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4764 struct dentry *dentry)
4766 struct btrfs_trans_handle *trans;
4767 struct btrfs_root *root = BTRFS_I(dir)->root;
4768 struct inode *inode = old_dentry->d_inode;
4770 unsigned long nr = 0;
4774 /* do not allow sys_link's with other subvols of the same device */
4775 if (root->objectid != BTRFS_I(inode)->root->objectid)
4778 if (inode->i_nlink == ~0U)
4781 err = btrfs_set_inode_index(dir, &index);
4786 * 2 items for inode and inode ref
4787 * 2 items for dir items
4788 * 1 item for parent inode
4790 trans = btrfs_start_transaction(root, 5);
4791 if (IS_ERR(trans)) {
4792 err = PTR_ERR(trans);
4796 btrfs_inc_nlink(inode);
4797 inode->i_ctime = CURRENT_TIME;
4800 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4805 struct dentry *parent = dentry->d_parent;
4806 err = btrfs_update_inode(trans, root, inode);
4808 btrfs_log_new_name(trans, inode, NULL, parent);
4811 nr = trans->blocks_used;
4812 btrfs_end_transaction_throttle(trans, root);
4815 inode_dec_link_count(inode);
4818 btrfs_btree_balance_dirty(root, nr);
4822 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4824 struct inode *inode = NULL;
4825 struct btrfs_trans_handle *trans;
4826 struct btrfs_root *root = BTRFS_I(dir)->root;
4828 int drop_on_err = 0;
4831 unsigned long nr = 1;
4834 * 2 items for inode and ref
4835 * 2 items for dir items
4836 * 1 for xattr if selinux is on
4838 trans = btrfs_start_transaction(root, 5);
4840 return PTR_ERR(trans);
4842 err = btrfs_find_free_ino(root, &objectid);
4846 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4847 dentry->d_name.len, btrfs_ino(dir), objectid,
4848 S_IFDIR | mode, &index);
4849 if (IS_ERR(inode)) {
4850 err = PTR_ERR(inode);
4856 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4860 inode->i_op = &btrfs_dir_inode_operations;
4861 inode->i_fop = &btrfs_dir_file_operations;
4863 btrfs_i_size_write(inode, 0);
4864 err = btrfs_update_inode(trans, root, inode);
4868 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4869 dentry->d_name.len, 0, index);
4873 d_instantiate(dentry, inode);
4877 nr = trans->blocks_used;
4878 btrfs_end_transaction_throttle(trans, root);
4881 btrfs_btree_balance_dirty(root, nr);
4885 /* helper for btfs_get_extent. Given an existing extent in the tree,
4886 * and an extent that you want to insert, deal with overlap and insert
4887 * the new extent into the tree.
4889 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4890 struct extent_map *existing,
4891 struct extent_map *em,
4892 u64 map_start, u64 map_len)
4896 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4897 start_diff = map_start - em->start;
4898 em->start = map_start;
4900 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4901 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4902 em->block_start += start_diff;
4903 em->block_len -= start_diff;
4905 return add_extent_mapping(em_tree, em);
4908 static noinline int uncompress_inline(struct btrfs_path *path,
4909 struct inode *inode, struct page *page,
4910 size_t pg_offset, u64 extent_offset,
4911 struct btrfs_file_extent_item *item)
4914 struct extent_buffer *leaf = path->nodes[0];
4917 unsigned long inline_size;
4921 WARN_ON(pg_offset != 0);
4922 compress_type = btrfs_file_extent_compression(leaf, item);
4923 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4924 inline_size = btrfs_file_extent_inline_item_len(leaf,
4925 btrfs_item_nr(leaf, path->slots[0]));
4926 tmp = kmalloc(inline_size, GFP_NOFS);
4929 ptr = btrfs_file_extent_inline_start(item);
4931 read_extent_buffer(leaf, tmp, ptr, inline_size);
4933 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4934 ret = btrfs_decompress(compress_type, tmp, page,
4935 extent_offset, inline_size, max_size);
4937 char *kaddr = kmap_atomic(page, KM_USER0);
4938 unsigned long copy_size = min_t(u64,
4939 PAGE_CACHE_SIZE - pg_offset,
4940 max_size - extent_offset);
4941 memset(kaddr + pg_offset, 0, copy_size);
4942 kunmap_atomic(kaddr, KM_USER0);
4949 * a bit scary, this does extent mapping from logical file offset to the disk.
4950 * the ugly parts come from merging extents from the disk with the in-ram
4951 * representation. This gets more complex because of the data=ordered code,
4952 * where the in-ram extents might be locked pending data=ordered completion.
4954 * This also copies inline extents directly into the page.
4957 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4958 size_t pg_offset, u64 start, u64 len,
4964 u64 extent_start = 0;
4966 u64 objectid = btrfs_ino(inode);
4968 struct btrfs_path *path = NULL;
4969 struct btrfs_root *root = BTRFS_I(inode)->root;
4970 struct btrfs_file_extent_item *item;
4971 struct extent_buffer *leaf;
4972 struct btrfs_key found_key;
4973 struct extent_map *em = NULL;
4974 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4975 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4976 struct btrfs_trans_handle *trans = NULL;
4980 read_lock(&em_tree->lock);
4981 em = lookup_extent_mapping(em_tree, start, len);
4983 em->bdev = root->fs_info->fs_devices->latest_bdev;
4984 read_unlock(&em_tree->lock);
4987 if (em->start > start || em->start + em->len <= start)
4988 free_extent_map(em);
4989 else if (em->block_start == EXTENT_MAP_INLINE && page)
4990 free_extent_map(em);
4994 em = alloc_extent_map();
4999 em->bdev = root->fs_info->fs_devices->latest_bdev;
5000 em->start = EXTENT_MAP_HOLE;
5001 em->orig_start = EXTENT_MAP_HOLE;
5003 em->block_len = (u64)-1;
5006 path = btrfs_alloc_path();
5012 * Chances are we'll be called again, so go ahead and do
5018 ret = btrfs_lookup_file_extent(trans, root, path,
5019 objectid, start, trans != NULL);
5026 if (path->slots[0] == 0)
5031 leaf = path->nodes[0];
5032 item = btrfs_item_ptr(leaf, path->slots[0],
5033 struct btrfs_file_extent_item);
5034 /* are we inside the extent that was found? */
5035 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5036 found_type = btrfs_key_type(&found_key);
5037 if (found_key.objectid != objectid ||
5038 found_type != BTRFS_EXTENT_DATA_KEY) {
5042 found_type = btrfs_file_extent_type(leaf, item);
5043 extent_start = found_key.offset;
5044 compress_type = btrfs_file_extent_compression(leaf, item);
5045 if (found_type == BTRFS_FILE_EXTENT_REG ||
5046 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5047 extent_end = extent_start +
5048 btrfs_file_extent_num_bytes(leaf, item);
5049 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5051 size = btrfs_file_extent_inline_len(leaf, item);
5052 extent_end = (extent_start + size + root->sectorsize - 1) &
5053 ~((u64)root->sectorsize - 1);
5056 if (start >= extent_end) {
5058 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5059 ret = btrfs_next_leaf(root, path);
5066 leaf = path->nodes[0];
5068 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5069 if (found_key.objectid != objectid ||
5070 found_key.type != BTRFS_EXTENT_DATA_KEY)
5072 if (start + len <= found_key.offset)
5075 em->len = found_key.offset - start;
5079 if (found_type == BTRFS_FILE_EXTENT_REG ||
5080 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5081 em->start = extent_start;
5082 em->len = extent_end - extent_start;
5083 em->orig_start = extent_start -
5084 btrfs_file_extent_offset(leaf, item);
5085 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5087 em->block_start = EXTENT_MAP_HOLE;
5090 if (compress_type != BTRFS_COMPRESS_NONE) {
5091 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5092 em->compress_type = compress_type;
5093 em->block_start = bytenr;
5094 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5097 bytenr += btrfs_file_extent_offset(leaf, item);
5098 em->block_start = bytenr;
5099 em->block_len = em->len;
5100 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5101 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5104 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5108 size_t extent_offset;
5111 em->block_start = EXTENT_MAP_INLINE;
5112 if (!page || create) {
5113 em->start = extent_start;
5114 em->len = extent_end - extent_start;
5118 size = btrfs_file_extent_inline_len(leaf, item);
5119 extent_offset = page_offset(page) + pg_offset - extent_start;
5120 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5121 size - extent_offset);
5122 em->start = extent_start + extent_offset;
5123 em->len = (copy_size + root->sectorsize - 1) &
5124 ~((u64)root->sectorsize - 1);
5125 em->orig_start = EXTENT_MAP_INLINE;
5126 if (compress_type) {
5127 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5128 em->compress_type = compress_type;
5130 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5131 if (create == 0 && !PageUptodate(page)) {
5132 if (btrfs_file_extent_compression(leaf, item) !=
5133 BTRFS_COMPRESS_NONE) {
5134 ret = uncompress_inline(path, inode, page,
5136 extent_offset, item);
5140 read_extent_buffer(leaf, map + pg_offset, ptr,
5142 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5143 memset(map + pg_offset + copy_size, 0,
5144 PAGE_CACHE_SIZE - pg_offset -
5149 flush_dcache_page(page);
5150 } else if (create && PageUptodate(page)) {
5154 free_extent_map(em);
5157 btrfs_release_path(path);
5158 trans = btrfs_join_transaction(root);
5161 return ERR_CAST(trans);
5165 write_extent_buffer(leaf, map + pg_offset, ptr,
5168 btrfs_mark_buffer_dirty(leaf);
5170 set_extent_uptodate(io_tree, em->start,
5171 extent_map_end(em) - 1, NULL, GFP_NOFS);
5174 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5181 em->block_start = EXTENT_MAP_HOLE;
5182 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5184 btrfs_release_path(path);
5185 if (em->start > start || extent_map_end(em) <= start) {
5186 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5187 "[%llu %llu]\n", (unsigned long long)em->start,
5188 (unsigned long long)em->len,
5189 (unsigned long long)start,
5190 (unsigned long long)len);
5196 write_lock(&em_tree->lock);
5197 ret = add_extent_mapping(em_tree, em);
5198 /* it is possible that someone inserted the extent into the tree
5199 * while we had the lock dropped. It is also possible that
5200 * an overlapping map exists in the tree
5202 if (ret == -EEXIST) {
5203 struct extent_map *existing;
5207 existing = lookup_extent_mapping(em_tree, start, len);
5208 if (existing && (existing->start > start ||
5209 existing->start + existing->len <= start)) {
5210 free_extent_map(existing);
5214 existing = lookup_extent_mapping(em_tree, em->start,
5217 err = merge_extent_mapping(em_tree, existing,
5220 free_extent_map(existing);
5222 free_extent_map(em);
5227 free_extent_map(em);
5231 free_extent_map(em);
5236 write_unlock(&em_tree->lock);
5239 trace_btrfs_get_extent(root, em);
5242 btrfs_free_path(path);
5244 ret = btrfs_end_transaction(trans, root);
5249 free_extent_map(em);
5250 return ERR_PTR(err);
5255 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5256 size_t pg_offset, u64 start, u64 len,
5259 struct extent_map *em;
5260 struct extent_map *hole_em = NULL;
5261 u64 range_start = start;
5267 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5272 * if our em maps to a hole, there might
5273 * actually be delalloc bytes behind it
5275 if (em->block_start != EXTENT_MAP_HOLE)
5281 /* check to see if we've wrapped (len == -1 or similar) */
5290 /* ok, we didn't find anything, lets look for delalloc */
5291 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5292 end, len, EXTENT_DELALLOC, 1);
5293 found_end = range_start + found;
5294 if (found_end < range_start)
5295 found_end = (u64)-1;
5298 * we didn't find anything useful, return
5299 * the original results from get_extent()
5301 if (range_start > end || found_end <= start) {
5307 /* adjust the range_start to make sure it doesn't
5308 * go backwards from the start they passed in
5310 range_start = max(start,range_start);
5311 found = found_end - range_start;
5314 u64 hole_start = start;
5317 em = alloc_extent_map();
5323 * when btrfs_get_extent can't find anything it
5324 * returns one huge hole
5326 * make sure what it found really fits our range, and
5327 * adjust to make sure it is based on the start from
5331 u64 calc_end = extent_map_end(hole_em);
5333 if (calc_end <= start || (hole_em->start > end)) {
5334 free_extent_map(hole_em);
5337 hole_start = max(hole_em->start, start);
5338 hole_len = calc_end - hole_start;
5342 if (hole_em && range_start > hole_start) {
5343 /* our hole starts before our delalloc, so we
5344 * have to return just the parts of the hole
5345 * that go until the delalloc starts
5347 em->len = min(hole_len,
5348 range_start - hole_start);
5349 em->start = hole_start;
5350 em->orig_start = hole_start;
5352 * don't adjust block start at all,
5353 * it is fixed at EXTENT_MAP_HOLE
5355 em->block_start = hole_em->block_start;
5356 em->block_len = hole_len;
5358 em->start = range_start;
5360 em->orig_start = range_start;
5361 em->block_start = EXTENT_MAP_DELALLOC;
5362 em->block_len = found;
5364 } else if (hole_em) {
5369 free_extent_map(hole_em);
5371 free_extent_map(em);
5372 return ERR_PTR(err);
5377 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5378 struct extent_map *em,
5381 struct btrfs_root *root = BTRFS_I(inode)->root;
5382 struct btrfs_trans_handle *trans;
5383 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5384 struct btrfs_key ins;
5387 bool insert = false;
5390 * Ok if the extent map we looked up is a hole and is for the exact
5391 * range we want, there is no reason to allocate a new one, however if
5392 * it is not right then we need to free this one and drop the cache for
5395 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5397 free_extent_map(em);
5400 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5403 trans = btrfs_join_transaction(root);
5405 return ERR_CAST(trans);
5407 if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5408 btrfs_add_inode_defrag(trans, inode);
5410 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5412 alloc_hint = get_extent_allocation_hint(inode, start, len);
5413 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5414 alloc_hint, (u64)-1, &ins, 1);
5421 em = alloc_extent_map();
5423 em = ERR_PTR(-ENOMEM);
5429 em->orig_start = em->start;
5430 em->len = ins.offset;
5432 em->block_start = ins.objectid;
5433 em->block_len = ins.offset;
5434 em->bdev = root->fs_info->fs_devices->latest_bdev;
5437 * We need to do this because if we're using the original em we searched
5438 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5441 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5444 write_lock(&em_tree->lock);
5445 ret = add_extent_mapping(em_tree, em);
5446 write_unlock(&em_tree->lock);
5449 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5452 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5453 ins.offset, ins.offset, 0);
5455 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5459 btrfs_end_transaction(trans, root);
5464 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5465 * block must be cow'd
5467 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5468 struct inode *inode, u64 offset, u64 len)
5470 struct btrfs_path *path;
5472 struct extent_buffer *leaf;
5473 struct btrfs_root *root = BTRFS_I(inode)->root;
5474 struct btrfs_file_extent_item *fi;
5475 struct btrfs_key key;
5483 path = btrfs_alloc_path();
5487 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5492 slot = path->slots[0];
5495 /* can't find the item, must cow */
5502 leaf = path->nodes[0];
5503 btrfs_item_key_to_cpu(leaf, &key, slot);
5504 if (key.objectid != btrfs_ino(inode) ||
5505 key.type != BTRFS_EXTENT_DATA_KEY) {
5506 /* not our file or wrong item type, must cow */
5510 if (key.offset > offset) {
5511 /* Wrong offset, must cow */
5515 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5516 found_type = btrfs_file_extent_type(leaf, fi);
5517 if (found_type != BTRFS_FILE_EXTENT_REG &&
5518 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5519 /* not a regular extent, must cow */
5522 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5523 backref_offset = btrfs_file_extent_offset(leaf, fi);
5525 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5526 if (extent_end < offset + len) {
5527 /* extent doesn't include our full range, must cow */
5531 if (btrfs_extent_readonly(root, disk_bytenr))
5535 * look for other files referencing this extent, if we
5536 * find any we must cow
5538 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5539 key.offset - backref_offset, disk_bytenr))
5543 * adjust disk_bytenr and num_bytes to cover just the bytes
5544 * in this extent we are about to write. If there
5545 * are any csums in that range we have to cow in order
5546 * to keep the csums correct
5548 disk_bytenr += backref_offset;
5549 disk_bytenr += offset - key.offset;
5550 num_bytes = min(offset + len, extent_end) - offset;
5551 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5554 * all of the above have passed, it is safe to overwrite this extent
5559 btrfs_free_path(path);
5563 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5564 struct buffer_head *bh_result, int create)
5566 struct extent_map *em;
5567 struct btrfs_root *root = BTRFS_I(inode)->root;
5568 u64 start = iblock << inode->i_blkbits;
5569 u64 len = bh_result->b_size;
5570 struct btrfs_trans_handle *trans;
5572 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5577 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5578 * io. INLINE is special, and we could probably kludge it in here, but
5579 * it's still buffered so for safety lets just fall back to the generic
5582 * For COMPRESSED we _have_ to read the entire extent in so we can
5583 * decompress it, so there will be buffering required no matter what we
5584 * do, so go ahead and fallback to buffered.
5586 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5587 * to buffered IO. Don't blame me, this is the price we pay for using
5590 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5591 em->block_start == EXTENT_MAP_INLINE) {
5592 free_extent_map(em);
5596 /* Just a good old fashioned hole, return */
5597 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5598 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5599 free_extent_map(em);
5600 /* DIO will do one hole at a time, so just unlock a sector */
5601 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5602 start + root->sectorsize - 1, GFP_NOFS);
5607 * We don't allocate a new extent in the following cases
5609 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5611 * 2) The extent is marked as PREALLOC. We're good to go here and can
5612 * just use the extent.
5616 len = em->len - (start - em->start);
5620 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5621 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5622 em->block_start != EXTENT_MAP_HOLE)) {
5627 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5628 type = BTRFS_ORDERED_PREALLOC;
5630 type = BTRFS_ORDERED_NOCOW;
5631 len = min(len, em->len - (start - em->start));
5632 block_start = em->block_start + (start - em->start);
5635 * we're not going to log anything, but we do need
5636 * to make sure the current transaction stays open
5637 * while we look for nocow cross refs
5639 trans = btrfs_join_transaction(root);
5643 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5644 ret = btrfs_add_ordered_extent_dio(inode, start,
5645 block_start, len, len, type);
5646 btrfs_end_transaction(trans, root);
5648 free_extent_map(em);
5653 btrfs_end_transaction(trans, root);
5657 * this will cow the extent, reset the len in case we changed
5660 len = bh_result->b_size;
5661 em = btrfs_new_extent_direct(inode, em, start, len);
5664 len = min(len, em->len - (start - em->start));
5666 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5667 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5670 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5672 bh_result->b_size = len;
5673 bh_result->b_bdev = em->bdev;
5674 set_buffer_mapped(bh_result);
5675 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5676 set_buffer_new(bh_result);
5678 free_extent_map(em);
5683 struct btrfs_dio_private {
5684 struct inode *inode;
5691 /* number of bios pending for this dio */
5692 atomic_t pending_bios;
5697 struct bio *orig_bio;
5700 static void btrfs_endio_direct_read(struct bio *bio, int err)
5702 struct btrfs_dio_private *dip = bio->bi_private;
5703 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5704 struct bio_vec *bvec = bio->bi_io_vec;
5705 struct inode *inode = dip->inode;
5706 struct btrfs_root *root = BTRFS_I(inode)->root;
5708 u32 *private = dip->csums;
5710 start = dip->logical_offset;
5712 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5713 struct page *page = bvec->bv_page;
5716 unsigned long flags;
5718 local_irq_save(flags);
5719 kaddr = kmap_atomic(page, KM_IRQ0);
5720 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5721 csum, bvec->bv_len);
5722 btrfs_csum_final(csum, (char *)&csum);
5723 kunmap_atomic(kaddr, KM_IRQ0);
5724 local_irq_restore(flags);
5726 flush_dcache_page(bvec->bv_page);
5727 if (csum != *private) {
5728 printk(KERN_ERR "btrfs csum failed ino %llu off"
5729 " %llu csum %u private %u\n",
5730 (unsigned long long)btrfs_ino(inode),
5731 (unsigned long long)start,
5737 start += bvec->bv_len;
5740 } while (bvec <= bvec_end);
5742 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5743 dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5744 bio->bi_private = dip->private;
5749 /* If we had a csum failure make sure to clear the uptodate flag */
5751 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5752 dio_end_io(bio, err);
5755 static void btrfs_endio_direct_write(struct bio *bio, int err)
5757 struct btrfs_dio_private *dip = bio->bi_private;
5758 struct inode *inode = dip->inode;
5759 struct btrfs_root *root = BTRFS_I(inode)->root;
5760 struct btrfs_trans_handle *trans;
5761 struct btrfs_ordered_extent *ordered = NULL;
5762 struct extent_state *cached_state = NULL;
5763 u64 ordered_offset = dip->logical_offset;
5764 u64 ordered_bytes = dip->bytes;
5770 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5778 trans = btrfs_join_transaction(root);
5779 if (IS_ERR(trans)) {
5783 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5785 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5786 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5788 ret = btrfs_update_inode(trans, root, inode);
5793 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5794 ordered->file_offset + ordered->len - 1, 0,
5795 &cached_state, GFP_NOFS);
5797 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5798 ret = btrfs_mark_extent_written(trans, inode,
5799 ordered->file_offset,
5800 ordered->file_offset +
5807 ret = insert_reserved_file_extent(trans, inode,
5808 ordered->file_offset,
5814 BTRFS_FILE_EXTENT_REG);
5815 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5816 ordered->file_offset, ordered->len);
5824 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5825 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5827 btrfs_update_inode(trans, root, inode);
5830 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5831 ordered->file_offset + ordered->len - 1,
5832 &cached_state, GFP_NOFS);
5834 btrfs_delalloc_release_metadata(inode, ordered->len);
5835 btrfs_end_transaction(trans, root);
5836 ordered_offset = ordered->file_offset + ordered->len;
5837 btrfs_put_ordered_extent(ordered);
5838 btrfs_put_ordered_extent(ordered);
5842 * our bio might span multiple ordered extents. If we haven't
5843 * completed the accounting for the whole dio, go back and try again
5845 if (ordered_offset < dip->logical_offset + dip->bytes) {
5846 ordered_bytes = dip->logical_offset + dip->bytes -
5851 bio->bi_private = dip->private;
5856 /* If we had an error make sure to clear the uptodate flag */
5858 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5859 dio_end_io(bio, err);
5862 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5863 struct bio *bio, int mirror_num,
5864 unsigned long bio_flags, u64 offset)
5867 struct btrfs_root *root = BTRFS_I(inode)->root;
5868 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5873 static void btrfs_end_dio_bio(struct bio *bio, int err)
5875 struct btrfs_dio_private *dip = bio->bi_private;
5878 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5879 "sector %#Lx len %u err no %d\n",
5880 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5881 (unsigned long long)bio->bi_sector, bio->bi_size, err);
5885 * before atomic variable goto zero, we must make sure
5886 * dip->errors is perceived to be set.
5888 smp_mb__before_atomic_dec();
5891 /* if there are more bios still pending for this dio, just exit */
5892 if (!atomic_dec_and_test(&dip->pending_bios))
5896 bio_io_error(dip->orig_bio);
5898 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5899 bio_endio(dip->orig_bio, 0);
5905 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5906 u64 first_sector, gfp_t gfp_flags)
5908 int nr_vecs = bio_get_nr_vecs(bdev);
5909 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5912 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5913 int rw, u64 file_offset, int skip_sum,
5914 u32 *csums, int async_submit)
5916 int write = rw & REQ_WRITE;
5917 struct btrfs_root *root = BTRFS_I(inode)->root;
5921 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5928 if (write && async_submit) {
5929 ret = btrfs_wq_submit_bio(root->fs_info,
5930 inode, rw, bio, 0, 0,
5932 __btrfs_submit_bio_start_direct_io,
5933 __btrfs_submit_bio_done);
5937 * If we aren't doing async submit, calculate the csum of the
5940 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5943 } else if (!skip_sum) {
5944 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5945 file_offset, csums);
5951 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5957 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5960 struct inode *inode = dip->inode;
5961 struct btrfs_root *root = BTRFS_I(inode)->root;
5962 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5964 struct bio *orig_bio = dip->orig_bio;
5965 struct bio_vec *bvec = orig_bio->bi_io_vec;
5966 u64 start_sector = orig_bio->bi_sector;
5967 u64 file_offset = dip->logical_offset;
5971 u32 *csums = dip->csums;
5973 int async_submit = 0;
5974 int write = rw & REQ_WRITE;
5976 map_length = orig_bio->bi_size;
5977 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5978 &map_length, NULL, 0);
5984 if (map_length >= orig_bio->bi_size) {
5990 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5993 bio->bi_private = dip;
5994 bio->bi_end_io = btrfs_end_dio_bio;
5995 atomic_inc(&dip->pending_bios);
5997 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5998 if (unlikely(map_length < submit_len + bvec->bv_len ||
5999 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6000 bvec->bv_offset) < bvec->bv_len)) {
6002 * inc the count before we submit the bio so
6003 * we know the end IO handler won't happen before
6004 * we inc the count. Otherwise, the dip might get freed
6005 * before we're done setting it up
6007 atomic_inc(&dip->pending_bios);
6008 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6009 file_offset, skip_sum,
6010 csums, async_submit);
6013 atomic_dec(&dip->pending_bios);
6017 /* Write's use the ordered csums */
6018 if (!write && !skip_sum)
6019 csums = csums + nr_pages;
6020 start_sector += submit_len >> 9;
6021 file_offset += submit_len;
6026 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6027 start_sector, GFP_NOFS);
6030 bio->bi_private = dip;
6031 bio->bi_end_io = btrfs_end_dio_bio;
6033 map_length = orig_bio->bi_size;
6034 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6035 &map_length, NULL, 0);
6041 submit_len += bvec->bv_len;
6048 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6049 csums, async_submit);
6057 * before atomic variable goto zero, we must
6058 * make sure dip->errors is perceived to be set.
6060 smp_mb__before_atomic_dec();
6061 if (atomic_dec_and_test(&dip->pending_bios))
6062 bio_io_error(dip->orig_bio);
6064 /* bio_end_io() will handle error, so we needn't return it */
6068 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6071 struct btrfs_root *root = BTRFS_I(inode)->root;
6072 struct btrfs_dio_private *dip;
6073 struct bio_vec *bvec = bio->bi_io_vec;
6075 int write = rw & REQ_WRITE;
6078 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6080 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6087 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6088 if (!write && !skip_sum) {
6089 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6097 dip->private = bio->bi_private;
6099 dip->logical_offset = file_offset;
6103 dip->bytes += bvec->bv_len;
6105 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6107 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6108 bio->bi_private = dip;
6110 dip->orig_bio = bio;
6111 atomic_set(&dip->pending_bios, 0);
6114 bio->bi_end_io = btrfs_endio_direct_write;
6116 bio->bi_end_io = btrfs_endio_direct_read;
6118 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6123 * If this is a write, we need to clean up the reserved space and kill
6124 * the ordered extent.
6127 struct btrfs_ordered_extent *ordered;
6128 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6129 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6130 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6131 btrfs_free_reserved_extent(root, ordered->start,
6133 btrfs_put_ordered_extent(ordered);
6134 btrfs_put_ordered_extent(ordered);
6136 bio_endio(bio, ret);
6139 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6140 const struct iovec *iov, loff_t offset,
6141 unsigned long nr_segs)
6147 unsigned blocksize_mask = root->sectorsize - 1;
6148 ssize_t retval = -EINVAL;
6149 loff_t end = offset;
6151 if (offset & blocksize_mask)
6154 /* Check the memory alignment. Blocks cannot straddle pages */
6155 for (seg = 0; seg < nr_segs; seg++) {
6156 addr = (unsigned long)iov[seg].iov_base;
6157 size = iov[seg].iov_len;
6159 if ((addr & blocksize_mask) || (size & blocksize_mask))
6162 /* If this is a write we don't need to check anymore */
6167 * Check to make sure we don't have duplicate iov_base's in this
6168 * iovec, if so return EINVAL, otherwise we'll get csum errors
6169 * when reading back.
6171 for (i = seg + 1; i < nr_segs; i++) {
6172 if (iov[seg].iov_base == iov[i].iov_base)
6180 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6181 const struct iovec *iov, loff_t offset,
6182 unsigned long nr_segs)
6184 struct file *file = iocb->ki_filp;
6185 struct inode *inode = file->f_mapping->host;
6186 struct btrfs_ordered_extent *ordered;
6187 struct extent_state *cached_state = NULL;
6188 u64 lockstart, lockend;
6190 int writing = rw & WRITE;
6192 size_t count = iov_length(iov, nr_segs);
6194 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6200 lockend = offset + count - 1;
6203 ret = btrfs_delalloc_reserve_space(inode, count);
6209 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6210 0, &cached_state, GFP_NOFS);
6212 * We're concerned with the entire range that we're going to be
6213 * doing DIO to, so we need to make sure theres no ordered
6214 * extents in this range.
6216 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6217 lockend - lockstart + 1);
6220 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6221 &cached_state, GFP_NOFS);
6222 btrfs_start_ordered_extent(inode, ordered, 1);
6223 btrfs_put_ordered_extent(ordered);
6228 * we don't use btrfs_set_extent_delalloc because we don't want
6229 * the dirty or uptodate bits
6232 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6233 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6234 EXTENT_DELALLOC, 0, NULL, &cached_state,
6237 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6238 lockend, EXTENT_LOCKED | write_bits,
6239 1, 0, &cached_state, GFP_NOFS);
6244 free_extent_state(cached_state);
6245 cached_state = NULL;
6247 ret = __blockdev_direct_IO(rw, iocb, inode,
6248 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6249 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6250 btrfs_submit_direct, 0);
6252 if (ret < 0 && ret != -EIOCBQUEUED) {
6253 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6254 offset + iov_length(iov, nr_segs) - 1,
6255 EXTENT_LOCKED | write_bits, 1, 0,
6256 &cached_state, GFP_NOFS);
6257 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6259 * We're falling back to buffered, unlock the section we didn't
6262 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6263 offset + iov_length(iov, nr_segs) - 1,
6264 EXTENT_LOCKED | write_bits, 1, 0,
6265 &cached_state, GFP_NOFS);
6268 free_extent_state(cached_state);
6272 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6273 __u64 start, __u64 len)
6275 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6278 int btrfs_readpage(struct file *file, struct page *page)
6280 struct extent_io_tree *tree;
6281 tree = &BTRFS_I(page->mapping->host)->io_tree;
6282 return extent_read_full_page(tree, page, btrfs_get_extent);
6285 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6287 struct extent_io_tree *tree;
6290 if (current->flags & PF_MEMALLOC) {
6291 redirty_page_for_writepage(wbc, page);
6295 tree = &BTRFS_I(page->mapping->host)->io_tree;
6296 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6299 int btrfs_writepages(struct address_space *mapping,
6300 struct writeback_control *wbc)
6302 struct extent_io_tree *tree;
6304 tree = &BTRFS_I(mapping->host)->io_tree;
6305 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6309 btrfs_readpages(struct file *file, struct address_space *mapping,
6310 struct list_head *pages, unsigned nr_pages)
6312 struct extent_io_tree *tree;
6313 tree = &BTRFS_I(mapping->host)->io_tree;
6314 return extent_readpages(tree, mapping, pages, nr_pages,
6317 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6319 struct extent_io_tree *tree;
6320 struct extent_map_tree *map;
6323 tree = &BTRFS_I(page->mapping->host)->io_tree;
6324 map = &BTRFS_I(page->mapping->host)->extent_tree;
6325 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6327 ClearPagePrivate(page);
6328 set_page_private(page, 0);
6329 page_cache_release(page);
6334 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6336 if (PageWriteback(page) || PageDirty(page))
6338 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6341 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6343 struct extent_io_tree *tree;
6344 struct btrfs_ordered_extent *ordered;
6345 struct extent_state *cached_state = NULL;
6346 u64 page_start = page_offset(page);
6347 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6351 * we have the page locked, so new writeback can't start,
6352 * and the dirty bit won't be cleared while we are here.
6354 * Wait for IO on this page so that we can safely clear
6355 * the PagePrivate2 bit and do ordered accounting
6357 wait_on_page_writeback(page);
6359 tree = &BTRFS_I(page->mapping->host)->io_tree;
6361 btrfs_releasepage(page, GFP_NOFS);
6364 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6366 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6370 * IO on this page will never be started, so we need
6371 * to account for any ordered extents now
6373 clear_extent_bit(tree, page_start, page_end,
6374 EXTENT_DIRTY | EXTENT_DELALLOC |
6375 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6376 &cached_state, GFP_NOFS);
6378 * whoever cleared the private bit is responsible
6379 * for the finish_ordered_io
6381 if (TestClearPagePrivate2(page)) {
6382 btrfs_finish_ordered_io(page->mapping->host,
6383 page_start, page_end);
6385 btrfs_put_ordered_extent(ordered);
6386 cached_state = NULL;
6387 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6390 clear_extent_bit(tree, page_start, page_end,
6391 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6392 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6393 __btrfs_releasepage(page, GFP_NOFS);
6395 ClearPageChecked(page);
6396 if (PagePrivate(page)) {
6397 ClearPagePrivate(page);
6398 set_page_private(page, 0);
6399 page_cache_release(page);
6404 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6405 * called from a page fault handler when a page is first dirtied. Hence we must
6406 * be careful to check for EOF conditions here. We set the page up correctly
6407 * for a written page which means we get ENOSPC checking when writing into
6408 * holes and correct delalloc and unwritten extent mapping on filesystems that
6409 * support these features.
6411 * We are not allowed to take the i_mutex here so we have to play games to
6412 * protect against truncate races as the page could now be beyond EOF. Because
6413 * vmtruncate() writes the inode size before removing pages, once we have the
6414 * page lock we can determine safely if the page is beyond EOF. If it is not
6415 * beyond EOF, then the page is guaranteed safe against truncation until we
6418 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6420 struct page *page = vmf->page;
6421 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6422 struct btrfs_root *root = BTRFS_I(inode)->root;
6423 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6424 struct btrfs_ordered_extent *ordered;
6425 struct extent_state *cached_state = NULL;
6427 unsigned long zero_start;
6433 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6437 else /* -ENOSPC, -EIO, etc */
6438 ret = VM_FAULT_SIGBUS;
6442 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6445 size = i_size_read(inode);
6446 page_start = page_offset(page);
6447 page_end = page_start + PAGE_CACHE_SIZE - 1;
6449 if ((page->mapping != inode->i_mapping) ||
6450 (page_start >= size)) {
6451 /* page got truncated out from underneath us */
6454 wait_on_page_writeback(page);
6456 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6458 set_page_extent_mapped(page);
6461 * we can't set the delalloc bits if there are pending ordered
6462 * extents. Drop our locks and wait for them to finish
6464 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6466 unlock_extent_cached(io_tree, page_start, page_end,
6467 &cached_state, GFP_NOFS);
6469 btrfs_start_ordered_extent(inode, ordered, 1);
6470 btrfs_put_ordered_extent(ordered);
6475 * XXX - page_mkwrite gets called every time the page is dirtied, even
6476 * if it was already dirty, so for space accounting reasons we need to
6477 * clear any delalloc bits for the range we are fixing to save. There
6478 * is probably a better way to do this, but for now keep consistent with
6479 * prepare_pages in the normal write path.
6481 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6482 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6483 0, 0, &cached_state, GFP_NOFS);
6485 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6488 unlock_extent_cached(io_tree, page_start, page_end,
6489 &cached_state, GFP_NOFS);
6490 ret = VM_FAULT_SIGBUS;
6495 /* page is wholly or partially inside EOF */
6496 if (page_start + PAGE_CACHE_SIZE > size)
6497 zero_start = size & ~PAGE_CACHE_MASK;
6499 zero_start = PAGE_CACHE_SIZE;
6501 if (zero_start != PAGE_CACHE_SIZE) {
6503 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6504 flush_dcache_page(page);
6507 ClearPageChecked(page);
6508 set_page_dirty(page);
6509 SetPageUptodate(page);
6511 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6512 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6514 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6518 return VM_FAULT_LOCKED;
6520 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6525 static int btrfs_truncate(struct inode *inode)
6527 struct btrfs_root *root = BTRFS_I(inode)->root;
6528 struct btrfs_block_rsv *rsv;
6531 struct btrfs_trans_handle *trans;
6533 u64 mask = root->sectorsize - 1;
6535 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6539 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6540 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6543 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6544 * 3 things going on here
6546 * 1) We need to reserve space for our orphan item and the space to
6547 * delete our orphan item. Lord knows we don't want to have a dangling
6548 * orphan item because we didn't reserve space to remove it.
6550 * 2) We need to reserve space to update our inode.
6552 * 3) We need to have something to cache all the space that is going to
6553 * be free'd up by the truncate operation, but also have some slack
6554 * space reserved in case it uses space during the truncate (thank you
6555 * very much snapshotting).
6557 * And we need these to all be seperate. The fact is we can use alot of
6558 * space doing the truncate, and we have no earthly idea how much space
6559 * we will use, so we need the truncate reservation to be seperate so it
6560 * doesn't end up using space reserved for updating the inode or
6561 * removing the orphan item. We also need to be able to stop the
6562 * transaction and start a new one, which means we need to be able to
6563 * update the inode several times, and we have no idea of knowing how
6564 * many times that will be, so we can't just reserve 1 item for the
6565 * entirety of the opration, so that has to be done seperately as well.
6566 * Then there is the orphan item, which does indeed need to be held on
6567 * to for the whole operation, and we need nobody to touch this reserved
6568 * space except the orphan code.
6570 * So that leaves us with
6572 * 1) root->orphan_block_rsv - for the orphan deletion.
6573 * 2) rsv - for the truncate reservation, which we will steal from the
6574 * transaction reservation.
6575 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6576 * updating the inode.
6578 rsv = btrfs_alloc_block_rsv(root);
6581 btrfs_add_durable_block_rsv(root->fs_info, rsv);
6583 trans = btrfs_start_transaction(root, 4);
6584 if (IS_ERR(trans)) {
6585 err = PTR_ERR(trans);
6590 * Reserve space for the truncate process. Truncate should be adding
6591 * space, but if there are snapshots it may end up using space.
6593 ret = btrfs_truncate_reserve_metadata(trans, root, rsv);
6596 ret = btrfs_orphan_add(trans, inode);
6598 btrfs_end_transaction(trans, root);
6602 nr = trans->blocks_used;
6603 btrfs_end_transaction(trans, root);
6604 btrfs_btree_balance_dirty(root, nr);
6607 * Ok so we've already migrated our bytes over for the truncate, so here
6608 * just reserve the one slot we need for updating the inode.
6610 trans = btrfs_start_transaction(root, 1);
6611 if (IS_ERR(trans)) {
6612 err = PTR_ERR(trans);
6615 trans->block_rsv = rsv;
6618 * setattr is responsible for setting the ordered_data_close flag,
6619 * but that is only tested during the last file release. That
6620 * could happen well after the next commit, leaving a great big
6621 * window where new writes may get lost if someone chooses to write
6622 * to this file after truncating to zero
6624 * The inode doesn't have any dirty data here, and so if we commit
6625 * this is a noop. If someone immediately starts writing to the inode
6626 * it is very likely we'll catch some of their writes in this
6627 * transaction, and the commit will find this file on the ordered
6628 * data list with good things to send down.
6630 * This is a best effort solution, there is still a window where
6631 * using truncate to replace the contents of the file will
6632 * end up with a zero length file after a crash.
6634 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6635 btrfs_add_ordered_operation(trans, root, inode);
6639 trans = btrfs_start_transaction(root, 3);
6640 if (IS_ERR(trans)) {
6641 err = PTR_ERR(trans);
6645 ret = btrfs_truncate_reserve_metadata(trans, root,
6649 trans->block_rsv = rsv;
6652 ret = btrfs_truncate_inode_items(trans, root, inode,
6654 BTRFS_EXTENT_DATA_KEY);
6655 if (ret != -EAGAIN) {
6660 trans->block_rsv = &root->fs_info->trans_block_rsv;
6661 ret = btrfs_update_inode(trans, root, inode);
6667 nr = trans->blocks_used;
6668 btrfs_end_transaction(trans, root);
6670 btrfs_btree_balance_dirty(root, nr);
6673 if (ret == 0 && inode->i_nlink > 0) {
6674 trans->block_rsv = root->orphan_block_rsv;
6675 ret = btrfs_orphan_del(trans, inode);
6678 } else if (ret && inode->i_nlink > 0) {
6680 * Failed to do the truncate, remove us from the in memory
6683 ret = btrfs_orphan_del(NULL, inode);
6686 trans->block_rsv = &root->fs_info->trans_block_rsv;
6687 ret = btrfs_update_inode(trans, root, inode);
6691 nr = trans->blocks_used;
6692 ret = btrfs_end_transaction_throttle(trans, root);
6693 btrfs_btree_balance_dirty(root, nr);
6696 btrfs_free_block_rsv(root, rsv);
6705 * create a new subvolume directory/inode (helper for the ioctl).
6707 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6708 struct btrfs_root *new_root, u64 new_dirid)
6710 struct inode *inode;
6714 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6715 new_dirid, S_IFDIR | 0700, &index);
6717 return PTR_ERR(inode);
6718 inode->i_op = &btrfs_dir_inode_operations;
6719 inode->i_fop = &btrfs_dir_file_operations;
6722 btrfs_i_size_write(inode, 0);
6724 err = btrfs_update_inode(trans, new_root, inode);
6731 struct inode *btrfs_alloc_inode(struct super_block *sb)
6733 struct btrfs_inode *ei;
6734 struct inode *inode;
6736 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6741 ei->space_info = NULL;
6745 ei->last_sub_trans = 0;
6746 ei->logged_trans = 0;
6747 ei->delalloc_bytes = 0;
6748 ei->reserved_bytes = 0;
6749 ei->disk_i_size = 0;
6751 ei->index_cnt = (u64)-1;
6752 ei->last_unlink_trans = 0;
6754 spin_lock_init(&ei->lock);
6755 ei->outstanding_extents = 0;
6756 ei->reserved_extents = 0;
6758 ei->ordered_data_close = 0;
6759 ei->orphan_meta_reserved = 0;
6760 ei->dummy_inode = 0;
6762 ei->force_compress = BTRFS_COMPRESS_NONE;
6764 ei->delayed_node = NULL;
6766 inode = &ei->vfs_inode;
6767 extent_map_tree_init(&ei->extent_tree);
6768 extent_io_tree_init(&ei->io_tree, &inode->i_data);
6769 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6770 mutex_init(&ei->log_mutex);
6771 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6772 INIT_LIST_HEAD(&ei->i_orphan);
6773 INIT_LIST_HEAD(&ei->delalloc_inodes);
6774 INIT_LIST_HEAD(&ei->ordered_operations);
6775 RB_CLEAR_NODE(&ei->rb_node);
6780 static void btrfs_i_callback(struct rcu_head *head)
6782 struct inode *inode = container_of(head, struct inode, i_rcu);
6783 INIT_LIST_HEAD(&inode->i_dentry);
6784 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6787 void btrfs_destroy_inode(struct inode *inode)
6789 struct btrfs_ordered_extent *ordered;
6790 struct btrfs_root *root = BTRFS_I(inode)->root;
6792 WARN_ON(!list_empty(&inode->i_dentry));
6793 WARN_ON(inode->i_data.nrpages);
6794 WARN_ON(BTRFS_I(inode)->outstanding_extents);
6795 WARN_ON(BTRFS_I(inode)->reserved_extents);
6798 * This can happen where we create an inode, but somebody else also
6799 * created the same inode and we need to destroy the one we already
6806 * Make sure we're properly removed from the ordered operation
6810 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6811 spin_lock(&root->fs_info->ordered_extent_lock);
6812 list_del_init(&BTRFS_I(inode)->ordered_operations);
6813 spin_unlock(&root->fs_info->ordered_extent_lock);
6816 spin_lock(&root->orphan_lock);
6817 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6818 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6819 (unsigned long long)btrfs_ino(inode));
6820 list_del_init(&BTRFS_I(inode)->i_orphan);
6822 spin_unlock(&root->orphan_lock);
6825 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6829 printk(KERN_ERR "btrfs found ordered "
6830 "extent %llu %llu on inode cleanup\n",
6831 (unsigned long long)ordered->file_offset,
6832 (unsigned long long)ordered->len);
6833 btrfs_remove_ordered_extent(inode, ordered);
6834 btrfs_put_ordered_extent(ordered);
6835 btrfs_put_ordered_extent(ordered);
6838 inode_tree_del(inode);
6839 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6841 btrfs_remove_delayed_node(inode);
6842 call_rcu(&inode->i_rcu, btrfs_i_callback);
6845 int btrfs_drop_inode(struct inode *inode)
6847 struct btrfs_root *root = BTRFS_I(inode)->root;
6849 if (btrfs_root_refs(&root->root_item) == 0 &&
6850 !btrfs_is_free_space_inode(root, inode))
6853 return generic_drop_inode(inode);
6856 static void init_once(void *foo)
6858 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6860 inode_init_once(&ei->vfs_inode);
6863 void btrfs_destroy_cachep(void)
6865 if (btrfs_inode_cachep)
6866 kmem_cache_destroy(btrfs_inode_cachep);
6867 if (btrfs_trans_handle_cachep)
6868 kmem_cache_destroy(btrfs_trans_handle_cachep);
6869 if (btrfs_transaction_cachep)
6870 kmem_cache_destroy(btrfs_transaction_cachep);
6871 if (btrfs_path_cachep)
6872 kmem_cache_destroy(btrfs_path_cachep);
6873 if (btrfs_free_space_cachep)
6874 kmem_cache_destroy(btrfs_free_space_cachep);
6877 int btrfs_init_cachep(void)
6879 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6880 sizeof(struct btrfs_inode), 0,
6881 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6882 if (!btrfs_inode_cachep)
6885 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6886 sizeof(struct btrfs_trans_handle), 0,
6887 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6888 if (!btrfs_trans_handle_cachep)
6891 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6892 sizeof(struct btrfs_transaction), 0,
6893 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6894 if (!btrfs_transaction_cachep)
6897 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6898 sizeof(struct btrfs_path), 0,
6899 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6900 if (!btrfs_path_cachep)
6903 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6904 sizeof(struct btrfs_free_space), 0,
6905 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6906 if (!btrfs_free_space_cachep)
6911 btrfs_destroy_cachep();
6915 static int btrfs_getattr(struct vfsmount *mnt,
6916 struct dentry *dentry, struct kstat *stat)
6918 struct inode *inode = dentry->d_inode;
6919 generic_fillattr(inode, stat);
6920 stat->dev = BTRFS_I(inode)->root->anon_dev;
6921 stat->blksize = PAGE_CACHE_SIZE;
6922 stat->blocks = (inode_get_bytes(inode) +
6923 BTRFS_I(inode)->delalloc_bytes) >> 9;
6928 * If a file is moved, it will inherit the cow and compression flags of the new
6931 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6933 struct btrfs_inode *b_dir = BTRFS_I(dir);
6934 struct btrfs_inode *b_inode = BTRFS_I(inode);
6936 if (b_dir->flags & BTRFS_INODE_NODATACOW)
6937 b_inode->flags |= BTRFS_INODE_NODATACOW;
6939 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6941 if (b_dir->flags & BTRFS_INODE_COMPRESS)
6942 b_inode->flags |= BTRFS_INODE_COMPRESS;
6944 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6947 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6948 struct inode *new_dir, struct dentry *new_dentry)
6950 struct btrfs_trans_handle *trans;
6951 struct btrfs_root *root = BTRFS_I(old_dir)->root;
6952 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6953 struct inode *new_inode = new_dentry->d_inode;
6954 struct inode *old_inode = old_dentry->d_inode;
6955 struct timespec ctime = CURRENT_TIME;
6959 u64 old_ino = btrfs_ino(old_inode);
6961 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6964 /* we only allow rename subvolume link between subvolumes */
6965 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6968 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6969 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6972 if (S_ISDIR(old_inode->i_mode) && new_inode &&
6973 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6976 * we're using rename to replace one file with another.
6977 * and the replacement file is large. Start IO on it now so
6978 * we don't add too much work to the end of the transaction
6980 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6981 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6982 filemap_flush(old_inode->i_mapping);
6984 /* close the racy window with snapshot create/destroy ioctl */
6985 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6986 down_read(&root->fs_info->subvol_sem);
6988 * We want to reserve the absolute worst case amount of items. So if
6989 * both inodes are subvols and we need to unlink them then that would
6990 * require 4 item modifications, but if they are both normal inodes it
6991 * would require 5 item modifications, so we'll assume their normal
6992 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6993 * should cover the worst case number of items we'll modify.
6995 trans = btrfs_start_transaction(root, 20);
6996 if (IS_ERR(trans)) {
6997 ret = PTR_ERR(trans);
7002 btrfs_record_root_in_trans(trans, dest);
7004 ret = btrfs_set_inode_index(new_dir, &index);
7008 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7009 /* force full log commit if subvolume involved. */
7010 root->fs_info->last_trans_log_full_commit = trans->transid;
7012 ret = btrfs_insert_inode_ref(trans, dest,
7013 new_dentry->d_name.name,
7014 new_dentry->d_name.len,
7016 btrfs_ino(new_dir), index);
7020 * this is an ugly little race, but the rename is required
7021 * to make sure that if we crash, the inode is either at the
7022 * old name or the new one. pinning the log transaction lets
7023 * us make sure we don't allow a log commit to come in after
7024 * we unlink the name but before we add the new name back in.
7026 btrfs_pin_log_trans(root);
7029 * make sure the inode gets flushed if it is replacing
7032 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7033 btrfs_add_ordered_operation(trans, root, old_inode);
7035 old_dir->i_ctime = old_dir->i_mtime = ctime;
7036 new_dir->i_ctime = new_dir->i_mtime = ctime;
7037 old_inode->i_ctime = ctime;
7039 if (old_dentry->d_parent != new_dentry->d_parent)
7040 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7042 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7043 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7044 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7045 old_dentry->d_name.name,
7046 old_dentry->d_name.len);
7048 ret = __btrfs_unlink_inode(trans, root, old_dir,
7049 old_dentry->d_inode,
7050 old_dentry->d_name.name,
7051 old_dentry->d_name.len);
7053 ret = btrfs_update_inode(trans, root, old_inode);
7058 new_inode->i_ctime = CURRENT_TIME;
7059 if (unlikely(btrfs_ino(new_inode) ==
7060 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7061 root_objectid = BTRFS_I(new_inode)->location.objectid;
7062 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7064 new_dentry->d_name.name,
7065 new_dentry->d_name.len);
7066 BUG_ON(new_inode->i_nlink == 0);
7068 ret = btrfs_unlink_inode(trans, dest, new_dir,
7069 new_dentry->d_inode,
7070 new_dentry->d_name.name,
7071 new_dentry->d_name.len);
7074 if (new_inode->i_nlink == 0) {
7075 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7080 fixup_inode_flags(new_dir, old_inode);
7082 ret = btrfs_add_link(trans, new_dir, old_inode,
7083 new_dentry->d_name.name,
7084 new_dentry->d_name.len, 0, index);
7087 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7088 struct dentry *parent = new_dentry->d_parent;
7089 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7090 btrfs_end_log_trans(root);
7093 btrfs_end_transaction_throttle(trans, root);
7095 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7096 up_read(&root->fs_info->subvol_sem);
7102 * some fairly slow code that needs optimization. This walks the list
7103 * of all the inodes with pending delalloc and forces them to disk.
7105 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7107 struct list_head *head = &root->fs_info->delalloc_inodes;
7108 struct btrfs_inode *binode;
7109 struct inode *inode;
7111 if (root->fs_info->sb->s_flags & MS_RDONLY)
7114 spin_lock(&root->fs_info->delalloc_lock);
7115 while (!list_empty(head)) {
7116 binode = list_entry(head->next, struct btrfs_inode,
7118 inode = igrab(&binode->vfs_inode);
7120 list_del_init(&binode->delalloc_inodes);
7121 spin_unlock(&root->fs_info->delalloc_lock);
7123 filemap_flush(inode->i_mapping);
7125 btrfs_add_delayed_iput(inode);
7130 spin_lock(&root->fs_info->delalloc_lock);
7132 spin_unlock(&root->fs_info->delalloc_lock);
7134 /* the filemap_flush will queue IO into the worker threads, but
7135 * we have to make sure the IO is actually started and that
7136 * ordered extents get created before we return
7138 atomic_inc(&root->fs_info->async_submit_draining);
7139 while (atomic_read(&root->fs_info->nr_async_submits) ||
7140 atomic_read(&root->fs_info->async_delalloc_pages)) {
7141 wait_event(root->fs_info->async_submit_wait,
7142 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7143 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7145 atomic_dec(&root->fs_info->async_submit_draining);
7149 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7150 const char *symname)
7152 struct btrfs_trans_handle *trans;
7153 struct btrfs_root *root = BTRFS_I(dir)->root;
7154 struct btrfs_path *path;
7155 struct btrfs_key key;
7156 struct inode *inode = NULL;
7164 struct btrfs_file_extent_item *ei;
7165 struct extent_buffer *leaf;
7166 unsigned long nr = 0;
7168 name_len = strlen(symname) + 1;
7169 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7170 return -ENAMETOOLONG;
7173 * 2 items for inode item and ref
7174 * 2 items for dir items
7175 * 1 item for xattr if selinux is on
7177 trans = btrfs_start_transaction(root, 5);
7179 return PTR_ERR(trans);
7181 err = btrfs_find_free_ino(root, &objectid);
7185 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7186 dentry->d_name.len, btrfs_ino(dir), objectid,
7187 S_IFLNK|S_IRWXUGO, &index);
7188 if (IS_ERR(inode)) {
7189 err = PTR_ERR(inode);
7193 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7199 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7203 inode->i_mapping->a_ops = &btrfs_aops;
7204 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7205 inode->i_fop = &btrfs_file_operations;
7206 inode->i_op = &btrfs_file_inode_operations;
7207 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7212 path = btrfs_alloc_path();
7218 key.objectid = btrfs_ino(inode);
7220 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7221 datasize = btrfs_file_extent_calc_inline_size(name_len);
7222 err = btrfs_insert_empty_item(trans, root, path, &key,
7226 btrfs_free_path(path);
7229 leaf = path->nodes[0];
7230 ei = btrfs_item_ptr(leaf, path->slots[0],
7231 struct btrfs_file_extent_item);
7232 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7233 btrfs_set_file_extent_type(leaf, ei,
7234 BTRFS_FILE_EXTENT_INLINE);
7235 btrfs_set_file_extent_encryption(leaf, ei, 0);
7236 btrfs_set_file_extent_compression(leaf, ei, 0);
7237 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7238 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7240 ptr = btrfs_file_extent_inline_start(ei);
7241 write_extent_buffer(leaf, symname, ptr, name_len);
7242 btrfs_mark_buffer_dirty(leaf);
7243 btrfs_free_path(path);
7245 inode->i_op = &btrfs_symlink_inode_operations;
7246 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7247 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7248 inode_set_bytes(inode, name_len);
7249 btrfs_i_size_write(inode, name_len - 1);
7250 err = btrfs_update_inode(trans, root, inode);
7255 nr = trans->blocks_used;
7256 btrfs_end_transaction_throttle(trans, root);
7258 inode_dec_link_count(inode);
7261 btrfs_btree_balance_dirty(root, nr);
7265 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7266 u64 start, u64 num_bytes, u64 min_size,
7267 loff_t actual_len, u64 *alloc_hint,
7268 struct btrfs_trans_handle *trans)
7270 struct btrfs_root *root = BTRFS_I(inode)->root;
7271 struct btrfs_key ins;
7272 u64 cur_offset = start;
7275 bool own_trans = true;
7279 while (num_bytes > 0) {
7281 trans = btrfs_start_transaction(root, 3);
7282 if (IS_ERR(trans)) {
7283 ret = PTR_ERR(trans);
7288 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7289 0, *alloc_hint, (u64)-1, &ins, 1);
7292 btrfs_end_transaction(trans, root);
7296 ret = insert_reserved_file_extent(trans, inode,
7297 cur_offset, ins.objectid,
7298 ins.offset, ins.offset,
7299 ins.offset, 0, 0, 0,
7300 BTRFS_FILE_EXTENT_PREALLOC);
7302 btrfs_drop_extent_cache(inode, cur_offset,
7303 cur_offset + ins.offset -1, 0);
7305 num_bytes -= ins.offset;
7306 cur_offset += ins.offset;
7307 *alloc_hint = ins.objectid + ins.offset;
7309 inode->i_ctime = CURRENT_TIME;
7310 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7311 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7312 (actual_len > inode->i_size) &&
7313 (cur_offset > inode->i_size)) {
7314 if (cur_offset > actual_len)
7315 i_size = actual_len;
7317 i_size = cur_offset;
7318 i_size_write(inode, i_size);
7319 btrfs_ordered_update_i_size(inode, i_size, NULL);
7322 ret = btrfs_update_inode(trans, root, inode);
7326 btrfs_end_transaction(trans, root);
7331 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7332 u64 start, u64 num_bytes, u64 min_size,
7333 loff_t actual_len, u64 *alloc_hint)
7335 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7336 min_size, actual_len, alloc_hint,
7340 int btrfs_prealloc_file_range_trans(struct inode *inode,
7341 struct btrfs_trans_handle *trans, int mode,
7342 u64 start, u64 num_bytes, u64 min_size,
7343 loff_t actual_len, u64 *alloc_hint)
7345 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7346 min_size, actual_len, alloc_hint, trans);
7349 static int btrfs_set_page_dirty(struct page *page)
7351 return __set_page_dirty_nobuffers(page);
7354 static int btrfs_permission(struct inode *inode, int mask)
7356 struct btrfs_root *root = BTRFS_I(inode)->root;
7357 umode_t mode = inode->i_mode;
7359 if (mask & MAY_WRITE &&
7360 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7361 if (btrfs_root_readonly(root))
7363 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7366 return generic_permission(inode, mask);
7369 static const struct inode_operations btrfs_dir_inode_operations = {
7370 .getattr = btrfs_getattr,
7371 .lookup = btrfs_lookup,
7372 .create = btrfs_create,
7373 .unlink = btrfs_unlink,
7375 .mkdir = btrfs_mkdir,
7376 .rmdir = btrfs_rmdir,
7377 .rename = btrfs_rename,
7378 .symlink = btrfs_symlink,
7379 .setattr = btrfs_setattr,
7380 .mknod = btrfs_mknod,
7381 .setxattr = btrfs_setxattr,
7382 .getxattr = btrfs_getxattr,
7383 .listxattr = btrfs_listxattr,
7384 .removexattr = btrfs_removexattr,
7385 .permission = btrfs_permission,
7386 .get_acl = btrfs_get_acl,
7388 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7389 .lookup = btrfs_lookup,
7390 .permission = btrfs_permission,
7391 .get_acl = btrfs_get_acl,
7394 static const struct file_operations btrfs_dir_file_operations = {
7395 .llseek = generic_file_llseek,
7396 .read = generic_read_dir,
7397 .readdir = btrfs_real_readdir,
7398 .unlocked_ioctl = btrfs_ioctl,
7399 #ifdef CONFIG_COMPAT
7400 .compat_ioctl = btrfs_ioctl,
7402 .release = btrfs_release_file,
7403 .fsync = btrfs_sync_file,
7406 static struct extent_io_ops btrfs_extent_io_ops = {
7407 .fill_delalloc = run_delalloc_range,
7408 .submit_bio_hook = btrfs_submit_bio_hook,
7409 .merge_bio_hook = btrfs_merge_bio_hook,
7410 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7411 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7412 .writepage_start_hook = btrfs_writepage_start_hook,
7413 .readpage_io_failed_hook = btrfs_io_failed_hook,
7414 .set_bit_hook = btrfs_set_bit_hook,
7415 .clear_bit_hook = btrfs_clear_bit_hook,
7416 .merge_extent_hook = btrfs_merge_extent_hook,
7417 .split_extent_hook = btrfs_split_extent_hook,
7421 * btrfs doesn't support the bmap operation because swapfiles
7422 * use bmap to make a mapping of extents in the file. They assume
7423 * these extents won't change over the life of the file and they
7424 * use the bmap result to do IO directly to the drive.
7426 * the btrfs bmap call would return logical addresses that aren't
7427 * suitable for IO and they also will change frequently as COW
7428 * operations happen. So, swapfile + btrfs == corruption.
7430 * For now we're avoiding this by dropping bmap.
7432 static const struct address_space_operations btrfs_aops = {
7433 .readpage = btrfs_readpage,
7434 .writepage = btrfs_writepage,
7435 .writepages = btrfs_writepages,
7436 .readpages = btrfs_readpages,
7437 .direct_IO = btrfs_direct_IO,
7438 .invalidatepage = btrfs_invalidatepage,
7439 .releasepage = btrfs_releasepage,
7440 .set_page_dirty = btrfs_set_page_dirty,
7441 .error_remove_page = generic_error_remove_page,
7444 static const struct address_space_operations btrfs_symlink_aops = {
7445 .readpage = btrfs_readpage,
7446 .writepage = btrfs_writepage,
7447 .invalidatepage = btrfs_invalidatepage,
7448 .releasepage = btrfs_releasepage,
7451 static const struct inode_operations btrfs_file_inode_operations = {
7452 .getattr = btrfs_getattr,
7453 .setattr = btrfs_setattr,
7454 .setxattr = btrfs_setxattr,
7455 .getxattr = btrfs_getxattr,
7456 .listxattr = btrfs_listxattr,
7457 .removexattr = btrfs_removexattr,
7458 .permission = btrfs_permission,
7459 .fiemap = btrfs_fiemap,
7460 .get_acl = btrfs_get_acl,
7462 static const struct inode_operations btrfs_special_inode_operations = {
7463 .getattr = btrfs_getattr,
7464 .setattr = btrfs_setattr,
7465 .permission = btrfs_permission,
7466 .setxattr = btrfs_setxattr,
7467 .getxattr = btrfs_getxattr,
7468 .listxattr = btrfs_listxattr,
7469 .removexattr = btrfs_removexattr,
7470 .get_acl = btrfs_get_acl,
7472 static const struct inode_operations btrfs_symlink_inode_operations = {
7473 .readlink = generic_readlink,
7474 .follow_link = page_follow_link_light,
7475 .put_link = page_put_link,
7476 .getattr = btrfs_getattr,
7477 .permission = btrfs_permission,
7478 .setxattr = btrfs_setxattr,
7479 .getxattr = btrfs_getxattr,
7480 .listxattr = btrfs_listxattr,
7481 .removexattr = btrfs_removexattr,
7482 .get_acl = btrfs_get_acl,
7485 const struct dentry_operations btrfs_dentry_operations = {
7486 .d_delete = btrfs_dentry_delete,
7487 .d_release = btrfs_dentry_release,