2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include <linux/mount.h>
45 #include "transaction.h"
46 #include "btrfs_inode.h"
48 #include "print-tree.h"
49 #include "ordered-data.h"
53 #include "compression.h"
55 #include "free-space-cache.h"
56 #include "inode-map.h"
58 struct btrfs_iget_args {
60 struct btrfs_root *root;
63 static const struct inode_operations btrfs_dir_inode_operations;
64 static const struct inode_operations btrfs_symlink_inode_operations;
65 static const struct inode_operations btrfs_dir_ro_inode_operations;
66 static const struct inode_operations btrfs_special_inode_operations;
67 static const struct inode_operations btrfs_file_inode_operations;
68 static const struct address_space_operations btrfs_aops;
69 static const struct address_space_operations btrfs_symlink_aops;
70 static const struct file_operations btrfs_dir_file_operations;
71 static struct extent_io_ops btrfs_extent_io_ops;
73 static struct kmem_cache *btrfs_inode_cachep;
74 struct kmem_cache *btrfs_trans_handle_cachep;
75 struct kmem_cache *btrfs_transaction_cachep;
76 struct kmem_cache *btrfs_path_cachep;
77 struct kmem_cache *btrfs_free_space_cachep;
80 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
81 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
82 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
83 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
84 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
85 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
86 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
87 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
90 static int btrfs_setsize(struct inode *inode, loff_t newsize);
91 static int btrfs_truncate(struct inode *inode);
92 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
93 static noinline int cow_file_range(struct inode *inode,
94 struct page *locked_page,
95 u64 start, u64 end, int *page_started,
96 unsigned long *nr_written, int unlock);
97 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
98 struct btrfs_root *root, struct inode *inode);
100 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
101 struct inode *inode, struct inode *dir,
102 const struct qstr *qstr)
106 err = btrfs_init_acl(trans, inode, dir);
108 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
113 * this does all the hard work for inserting an inline extent into
114 * the btree. The caller should have done a btrfs_drop_extents so that
115 * no overlapping inline items exist in the btree
117 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
118 struct btrfs_root *root, struct inode *inode,
119 u64 start, size_t size, size_t compressed_size,
121 struct page **compressed_pages)
123 struct btrfs_key key;
124 struct btrfs_path *path;
125 struct extent_buffer *leaf;
126 struct page *page = NULL;
129 struct btrfs_file_extent_item *ei;
132 size_t cur_size = size;
134 unsigned long offset;
136 if (compressed_size && compressed_pages)
137 cur_size = compressed_size;
139 path = btrfs_alloc_path();
143 path->leave_spinning = 1;
145 key.objectid = btrfs_ino(inode);
147 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
148 datasize = btrfs_file_extent_calc_inline_size(cur_size);
150 inode_add_bytes(inode, size);
151 ret = btrfs_insert_empty_item(trans, root, path, &key,
157 leaf = path->nodes[0];
158 ei = btrfs_item_ptr(leaf, path->slots[0],
159 struct btrfs_file_extent_item);
160 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
161 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
162 btrfs_set_file_extent_encryption(leaf, ei, 0);
163 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
164 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
165 ptr = btrfs_file_extent_inline_start(ei);
167 if (compress_type != BTRFS_COMPRESS_NONE) {
170 while (compressed_size > 0) {
171 cpage = compressed_pages[i];
172 cur_size = min_t(unsigned long, compressed_size,
175 kaddr = kmap_atomic(cpage);
176 write_extent_buffer(leaf, kaddr, ptr, cur_size);
177 kunmap_atomic(kaddr);
181 compressed_size -= cur_size;
183 btrfs_set_file_extent_compression(leaf, ei,
186 page = find_get_page(inode->i_mapping,
187 start >> PAGE_CACHE_SHIFT);
188 btrfs_set_file_extent_compression(leaf, ei, 0);
189 kaddr = kmap_atomic(page);
190 offset = start & (PAGE_CACHE_SIZE - 1);
191 write_extent_buffer(leaf, kaddr + offset, ptr, size);
192 kunmap_atomic(kaddr);
193 page_cache_release(page);
195 btrfs_mark_buffer_dirty(leaf);
196 btrfs_free_path(path);
199 * we're an inline extent, so nobody can
200 * extend the file past i_size without locking
201 * a page we already have locked.
203 * We must do any isize and inode updates
204 * before we unlock the pages. Otherwise we
205 * could end up racing with unlink.
207 BTRFS_I(inode)->disk_i_size = inode->i_size;
208 ret = btrfs_update_inode(trans, root, inode);
212 btrfs_free_path(path);
218 * conditionally insert an inline extent into the file. This
219 * does the checks required to make sure the data is small enough
220 * to fit as an inline extent.
222 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
223 struct btrfs_root *root,
224 struct inode *inode, u64 start, u64 end,
225 size_t compressed_size, int compress_type,
226 struct page **compressed_pages)
228 u64 isize = i_size_read(inode);
229 u64 actual_end = min(end + 1, isize);
230 u64 inline_len = actual_end - start;
231 u64 aligned_end = (end + root->sectorsize - 1) &
232 ~((u64)root->sectorsize - 1);
233 u64 data_len = inline_len;
237 data_len = compressed_size;
240 actual_end >= PAGE_CACHE_SIZE ||
241 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
243 (actual_end & (root->sectorsize - 1)) == 0) ||
245 data_len > root->fs_info->max_inline) {
249 ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
253 if (isize > actual_end)
254 inline_len = min_t(u64, isize, actual_end);
255 ret = insert_inline_extent(trans, root, inode, start,
256 inline_len, compressed_size,
257 compress_type, compressed_pages);
258 if (ret && ret != -ENOSPC) {
259 btrfs_abort_transaction(trans, root, ret);
261 } else if (ret == -ENOSPC) {
265 btrfs_delalloc_release_metadata(inode, end + 1 - start);
266 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
270 struct async_extent {
275 unsigned long nr_pages;
277 struct list_head list;
282 struct btrfs_root *root;
283 struct page *locked_page;
286 struct list_head extents;
287 struct btrfs_work work;
290 static noinline int add_async_extent(struct async_cow *cow,
291 u64 start, u64 ram_size,
294 unsigned long nr_pages,
297 struct async_extent *async_extent;
299 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
300 BUG_ON(!async_extent); /* -ENOMEM */
301 async_extent->start = start;
302 async_extent->ram_size = ram_size;
303 async_extent->compressed_size = compressed_size;
304 async_extent->pages = pages;
305 async_extent->nr_pages = nr_pages;
306 async_extent->compress_type = compress_type;
307 list_add_tail(&async_extent->list, &cow->extents);
312 * we create compressed extents in two phases. The first
313 * phase compresses a range of pages that have already been
314 * locked (both pages and state bits are locked).
316 * This is done inside an ordered work queue, and the compression
317 * is spread across many cpus. The actual IO submission is step
318 * two, and the ordered work queue takes care of making sure that
319 * happens in the same order things were put onto the queue by
320 * writepages and friends.
322 * If this code finds it can't get good compression, it puts an
323 * entry onto the work queue to write the uncompressed bytes. This
324 * makes sure that both compressed inodes and uncompressed inodes
325 * are written in the same order that the flusher thread sent them
328 static noinline int compress_file_range(struct inode *inode,
329 struct page *locked_page,
331 struct async_cow *async_cow,
334 struct btrfs_root *root = BTRFS_I(inode)->root;
335 struct btrfs_trans_handle *trans;
337 u64 blocksize = root->sectorsize;
339 u64 isize = i_size_read(inode);
341 struct page **pages = NULL;
342 unsigned long nr_pages;
343 unsigned long nr_pages_ret = 0;
344 unsigned long total_compressed = 0;
345 unsigned long total_in = 0;
346 unsigned long max_compressed = 128 * 1024;
347 unsigned long max_uncompressed = 128 * 1024;
350 int compress_type = root->fs_info->compress_type;
352 /* if this is a small write inside eof, kick off a defrag */
353 if ((end - start + 1) < 16 * 1024 &&
354 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
355 btrfs_add_inode_defrag(NULL, inode);
357 actual_end = min_t(u64, isize, end + 1);
360 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
361 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
364 * we don't want to send crud past the end of i_size through
365 * compression, that's just a waste of CPU time. So, if the
366 * end of the file is before the start of our current
367 * requested range of bytes, we bail out to the uncompressed
368 * cleanup code that can deal with all of this.
370 * It isn't really the fastest way to fix things, but this is a
371 * very uncommon corner.
373 if (actual_end <= start)
374 goto cleanup_and_bail_uncompressed;
376 total_compressed = actual_end - start;
378 /* we want to make sure that amount of ram required to uncompress
379 * an extent is reasonable, so we limit the total size in ram
380 * of a compressed extent to 128k. This is a crucial number
381 * because it also controls how easily we can spread reads across
382 * cpus for decompression.
384 * We also want to make sure the amount of IO required to do
385 * a random read is reasonably small, so we limit the size of
386 * a compressed extent to 128k.
388 total_compressed = min(total_compressed, max_uncompressed);
389 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
390 num_bytes = max(blocksize, num_bytes);
395 * we do compression for mount -o compress and when the
396 * inode has not been flagged as nocompress. This flag can
397 * change at any time if we discover bad compression ratios.
399 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
400 (btrfs_test_opt(root, COMPRESS) ||
401 (BTRFS_I(inode)->force_compress) ||
402 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
404 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
406 /* just bail out to the uncompressed code */
410 if (BTRFS_I(inode)->force_compress)
411 compress_type = BTRFS_I(inode)->force_compress;
413 ret = btrfs_compress_pages(compress_type,
414 inode->i_mapping, start,
415 total_compressed, pages,
416 nr_pages, &nr_pages_ret,
422 unsigned long offset = total_compressed &
423 (PAGE_CACHE_SIZE - 1);
424 struct page *page = pages[nr_pages_ret - 1];
427 /* zero the tail end of the last page, we might be
428 * sending it down to disk
431 kaddr = kmap_atomic(page);
432 memset(kaddr + offset, 0,
433 PAGE_CACHE_SIZE - offset);
434 kunmap_atomic(kaddr);
441 trans = btrfs_join_transaction(root);
443 ret = PTR_ERR(trans);
445 goto cleanup_and_out;
447 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
449 /* lets try to make an inline extent */
450 if (ret || total_in < (actual_end - start)) {
451 /* we didn't compress the entire range, try
452 * to make an uncompressed inline extent.
454 ret = cow_file_range_inline(trans, root, inode,
455 start, end, 0, 0, NULL);
457 /* try making a compressed inline extent */
458 ret = cow_file_range_inline(trans, root, inode,
461 compress_type, pages);
465 * inline extent creation worked or returned error,
466 * we don't need to create any more async work items.
467 * Unlock and free up our temp pages.
469 extent_clear_unlock_delalloc(inode,
470 &BTRFS_I(inode)->io_tree,
472 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
473 EXTENT_CLEAR_DELALLOC |
474 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
476 btrfs_end_transaction(trans, root);
479 btrfs_end_transaction(trans, root);
484 * we aren't doing an inline extent round the compressed size
485 * up to a block size boundary so the allocator does sane
488 total_compressed = (total_compressed + blocksize - 1) &
492 * one last check to make sure the compression is really a
493 * win, compare the page count read with the blocks on disk
495 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
496 ~(PAGE_CACHE_SIZE - 1);
497 if (total_compressed >= total_in) {
500 num_bytes = total_in;
503 if (!will_compress && pages) {
505 * the compression code ran but failed to make things smaller,
506 * free any pages it allocated and our page pointer array
508 for (i = 0; i < nr_pages_ret; i++) {
509 WARN_ON(pages[i]->mapping);
510 page_cache_release(pages[i]);
514 total_compressed = 0;
517 /* flag the file so we don't compress in the future */
518 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
519 !(BTRFS_I(inode)->force_compress)) {
520 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
526 /* the async work queues will take care of doing actual
527 * allocation on disk for these compressed pages,
528 * and will submit them to the elevator.
530 add_async_extent(async_cow, start, num_bytes,
531 total_compressed, pages, nr_pages_ret,
534 if (start + num_bytes < end) {
541 cleanup_and_bail_uncompressed:
543 * No compression, but we still need to write the pages in
544 * the file we've been given so far. redirty the locked
545 * page if it corresponds to our extent and set things up
546 * for the async work queue to run cow_file_range to do
547 * the normal delalloc dance
549 if (page_offset(locked_page) >= start &&
550 page_offset(locked_page) <= end) {
551 __set_page_dirty_nobuffers(locked_page);
552 /* unlocked later on in the async handlers */
554 add_async_extent(async_cow, start, end - start + 1,
555 0, NULL, 0, BTRFS_COMPRESS_NONE);
563 for (i = 0; i < nr_pages_ret; i++) {
564 WARN_ON(pages[i]->mapping);
565 page_cache_release(pages[i]);
572 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
574 EXTENT_CLEAR_UNLOCK_PAGE |
576 EXTENT_CLEAR_DELALLOC |
577 EXTENT_SET_WRITEBACK |
578 EXTENT_END_WRITEBACK);
579 if (!trans || IS_ERR(trans))
580 btrfs_error(root->fs_info, ret, "Failed to join transaction");
582 btrfs_abort_transaction(trans, root, ret);
587 * phase two of compressed writeback. This is the ordered portion
588 * of the code, which only gets called in the order the work was
589 * queued. We walk all the async extents created by compress_file_range
590 * and send them down to the disk.
592 static noinline int submit_compressed_extents(struct inode *inode,
593 struct async_cow *async_cow)
595 struct async_extent *async_extent;
597 struct btrfs_trans_handle *trans;
598 struct btrfs_key ins;
599 struct extent_map *em;
600 struct btrfs_root *root = BTRFS_I(inode)->root;
601 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
602 struct extent_io_tree *io_tree;
605 if (list_empty(&async_cow->extents))
609 while (!list_empty(&async_cow->extents)) {
610 async_extent = list_entry(async_cow->extents.next,
611 struct async_extent, list);
612 list_del(&async_extent->list);
614 io_tree = &BTRFS_I(inode)->io_tree;
617 /* did the compression code fall back to uncompressed IO? */
618 if (!async_extent->pages) {
619 int page_started = 0;
620 unsigned long nr_written = 0;
622 lock_extent(io_tree, async_extent->start,
623 async_extent->start +
624 async_extent->ram_size - 1);
626 /* allocate blocks */
627 ret = cow_file_range(inode, async_cow->locked_page,
629 async_extent->start +
630 async_extent->ram_size - 1,
631 &page_started, &nr_written, 0);
636 * if page_started, cow_file_range inserted an
637 * inline extent and took care of all the unlocking
638 * and IO for us. Otherwise, we need to submit
639 * all those pages down to the drive.
641 if (!page_started && !ret)
642 extent_write_locked_range(io_tree,
643 inode, async_extent->start,
644 async_extent->start +
645 async_extent->ram_size - 1,
653 lock_extent(io_tree, async_extent->start,
654 async_extent->start + async_extent->ram_size - 1);
656 trans = btrfs_join_transaction(root);
658 ret = PTR_ERR(trans);
660 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
661 ret = btrfs_reserve_extent(trans, root,
662 async_extent->compressed_size,
663 async_extent->compressed_size,
664 0, alloc_hint, &ins, 1);
665 if (ret && ret != -ENOSPC)
666 btrfs_abort_transaction(trans, root, ret);
667 btrfs_end_transaction(trans, root);
672 for (i = 0; i < async_extent->nr_pages; i++) {
673 WARN_ON(async_extent->pages[i]->mapping);
674 page_cache_release(async_extent->pages[i]);
676 kfree(async_extent->pages);
677 async_extent->nr_pages = 0;
678 async_extent->pages = NULL;
679 unlock_extent(io_tree, async_extent->start,
680 async_extent->start +
681 async_extent->ram_size - 1);
684 goto out_free; /* JDM: Requeue? */
688 * here we're doing allocation and writeback of the
691 btrfs_drop_extent_cache(inode, async_extent->start,
692 async_extent->start +
693 async_extent->ram_size - 1, 0);
695 em = alloc_extent_map();
696 BUG_ON(!em); /* -ENOMEM */
697 em->start = async_extent->start;
698 em->len = async_extent->ram_size;
699 em->orig_start = em->start;
701 em->block_start = ins.objectid;
702 em->block_len = ins.offset;
703 em->bdev = root->fs_info->fs_devices->latest_bdev;
704 em->compress_type = async_extent->compress_type;
705 set_bit(EXTENT_FLAG_PINNED, &em->flags);
706 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
709 write_lock(&em_tree->lock);
710 ret = add_extent_mapping(em_tree, em);
711 write_unlock(&em_tree->lock);
712 if (ret != -EEXIST) {
716 btrfs_drop_extent_cache(inode, async_extent->start,
717 async_extent->start +
718 async_extent->ram_size - 1, 0);
721 ret = btrfs_add_ordered_extent_compress(inode,
724 async_extent->ram_size,
726 BTRFS_ORDERED_COMPRESSED,
727 async_extent->compress_type);
728 BUG_ON(ret); /* -ENOMEM */
731 * clear dirty, set writeback and unlock the pages.
733 extent_clear_unlock_delalloc(inode,
734 &BTRFS_I(inode)->io_tree,
736 async_extent->start +
737 async_extent->ram_size - 1,
738 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
739 EXTENT_CLEAR_UNLOCK |
740 EXTENT_CLEAR_DELALLOC |
741 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
743 ret = btrfs_submit_compressed_write(inode,
745 async_extent->ram_size,
747 ins.offset, async_extent->pages,
748 async_extent->nr_pages);
750 BUG_ON(ret); /* -ENOMEM */
751 alloc_hint = ins.objectid + ins.offset;
763 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
766 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
767 struct extent_map *em;
770 read_lock(&em_tree->lock);
771 em = search_extent_mapping(em_tree, start, num_bytes);
774 * if block start isn't an actual block number then find the
775 * first block in this inode and use that as a hint. If that
776 * block is also bogus then just don't worry about it.
778 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
780 em = search_extent_mapping(em_tree, 0, 0);
781 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
782 alloc_hint = em->block_start;
786 alloc_hint = em->block_start;
790 read_unlock(&em_tree->lock);
796 * when extent_io.c finds a delayed allocation range in the file,
797 * the call backs end up in this code. The basic idea is to
798 * allocate extents on disk for the range, and create ordered data structs
799 * in ram to track those extents.
801 * locked_page is the page that writepage had locked already. We use
802 * it to make sure we don't do extra locks or unlocks.
804 * *page_started is set to one if we unlock locked_page and do everything
805 * required to start IO on it. It may be clean and already done with
808 static noinline int cow_file_range(struct inode *inode,
809 struct page *locked_page,
810 u64 start, u64 end, int *page_started,
811 unsigned long *nr_written,
814 struct btrfs_root *root = BTRFS_I(inode)->root;
815 struct btrfs_trans_handle *trans;
818 unsigned long ram_size;
821 u64 blocksize = root->sectorsize;
822 struct btrfs_key ins;
823 struct extent_map *em;
824 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
827 BUG_ON(btrfs_is_free_space_inode(inode));
828 trans = btrfs_join_transaction(root);
830 extent_clear_unlock_delalloc(inode,
831 &BTRFS_I(inode)->io_tree,
832 start, end, locked_page,
833 EXTENT_CLEAR_UNLOCK_PAGE |
834 EXTENT_CLEAR_UNLOCK |
835 EXTENT_CLEAR_DELALLOC |
837 EXTENT_SET_WRITEBACK |
838 EXTENT_END_WRITEBACK);
839 return PTR_ERR(trans);
841 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
843 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
844 num_bytes = max(blocksize, num_bytes);
845 disk_num_bytes = num_bytes;
848 /* if this is a small write inside eof, kick off defrag */
849 if (num_bytes < 64 * 1024 &&
850 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
851 btrfs_add_inode_defrag(trans, inode);
854 /* lets try to make an inline extent */
855 ret = cow_file_range_inline(trans, root, inode,
856 start, end, 0, 0, NULL);
858 extent_clear_unlock_delalloc(inode,
859 &BTRFS_I(inode)->io_tree,
861 EXTENT_CLEAR_UNLOCK_PAGE |
862 EXTENT_CLEAR_UNLOCK |
863 EXTENT_CLEAR_DELALLOC |
865 EXTENT_SET_WRITEBACK |
866 EXTENT_END_WRITEBACK);
868 *nr_written = *nr_written +
869 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
872 } else if (ret < 0) {
873 btrfs_abort_transaction(trans, root, ret);
878 BUG_ON(disk_num_bytes >
879 btrfs_super_total_bytes(root->fs_info->super_copy));
881 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
882 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
884 while (disk_num_bytes > 0) {
887 cur_alloc_size = disk_num_bytes;
888 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
889 root->sectorsize, 0, alloc_hint,
892 btrfs_abort_transaction(trans, root, ret);
896 em = alloc_extent_map();
897 BUG_ON(!em); /* -ENOMEM */
899 em->orig_start = em->start;
900 ram_size = ins.offset;
901 em->len = ins.offset;
903 em->block_start = ins.objectid;
904 em->block_len = ins.offset;
905 em->bdev = root->fs_info->fs_devices->latest_bdev;
906 set_bit(EXTENT_FLAG_PINNED, &em->flags);
909 write_lock(&em_tree->lock);
910 ret = add_extent_mapping(em_tree, em);
911 write_unlock(&em_tree->lock);
912 if (ret != -EEXIST) {
916 btrfs_drop_extent_cache(inode, start,
917 start + ram_size - 1, 0);
920 cur_alloc_size = ins.offset;
921 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
922 ram_size, cur_alloc_size, 0);
923 BUG_ON(ret); /* -ENOMEM */
925 if (root->root_key.objectid ==
926 BTRFS_DATA_RELOC_TREE_OBJECTID) {
927 ret = btrfs_reloc_clone_csums(inode, start,
930 btrfs_abort_transaction(trans, root, ret);
935 if (disk_num_bytes < cur_alloc_size)
938 /* we're not doing compressed IO, don't unlock the first
939 * page (which the caller expects to stay locked), don't
940 * clear any dirty bits and don't set any writeback bits
942 * Do set the Private2 bit so we know this page was properly
943 * setup for writepage
945 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
946 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
949 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
950 start, start + ram_size - 1,
952 disk_num_bytes -= cur_alloc_size;
953 num_bytes -= cur_alloc_size;
954 alloc_hint = ins.objectid + ins.offset;
955 start += cur_alloc_size;
959 btrfs_end_transaction(trans, root);
963 extent_clear_unlock_delalloc(inode,
964 &BTRFS_I(inode)->io_tree,
965 start, end, locked_page,
966 EXTENT_CLEAR_UNLOCK_PAGE |
967 EXTENT_CLEAR_UNLOCK |
968 EXTENT_CLEAR_DELALLOC |
970 EXTENT_SET_WRITEBACK |
971 EXTENT_END_WRITEBACK);
977 * work queue call back to started compression on a file and pages
979 static noinline void async_cow_start(struct btrfs_work *work)
981 struct async_cow *async_cow;
983 async_cow = container_of(work, struct async_cow, work);
985 compress_file_range(async_cow->inode, async_cow->locked_page,
986 async_cow->start, async_cow->end, async_cow,
988 if (num_added == 0) {
989 btrfs_add_delayed_iput(async_cow->inode);
990 async_cow->inode = NULL;
995 * work queue call back to submit previously compressed pages
997 static noinline void async_cow_submit(struct btrfs_work *work)
999 struct async_cow *async_cow;
1000 struct btrfs_root *root;
1001 unsigned long nr_pages;
1003 async_cow = container_of(work, struct async_cow, work);
1005 root = async_cow->root;
1006 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1009 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1011 waitqueue_active(&root->fs_info->async_submit_wait))
1012 wake_up(&root->fs_info->async_submit_wait);
1014 if (async_cow->inode)
1015 submit_compressed_extents(async_cow->inode, async_cow);
1018 static noinline void async_cow_free(struct btrfs_work *work)
1020 struct async_cow *async_cow;
1021 async_cow = container_of(work, struct async_cow, work);
1022 if (async_cow->inode)
1023 btrfs_add_delayed_iput(async_cow->inode);
1027 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1028 u64 start, u64 end, int *page_started,
1029 unsigned long *nr_written)
1031 struct async_cow *async_cow;
1032 struct btrfs_root *root = BTRFS_I(inode)->root;
1033 unsigned long nr_pages;
1035 int limit = 10 * 1024 * 1024;
1037 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1038 1, 0, NULL, GFP_NOFS);
1039 while (start < end) {
1040 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1041 BUG_ON(!async_cow); /* -ENOMEM */
1042 async_cow->inode = igrab(inode);
1043 async_cow->root = root;
1044 async_cow->locked_page = locked_page;
1045 async_cow->start = start;
1047 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1050 cur_end = min(end, start + 512 * 1024 - 1);
1052 async_cow->end = cur_end;
1053 INIT_LIST_HEAD(&async_cow->extents);
1055 async_cow->work.func = async_cow_start;
1056 async_cow->work.ordered_func = async_cow_submit;
1057 async_cow->work.ordered_free = async_cow_free;
1058 async_cow->work.flags = 0;
1060 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1062 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1064 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1067 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1068 wait_event(root->fs_info->async_submit_wait,
1069 (atomic_read(&root->fs_info->async_delalloc_pages) <
1073 while (atomic_read(&root->fs_info->async_submit_draining) &&
1074 atomic_read(&root->fs_info->async_delalloc_pages)) {
1075 wait_event(root->fs_info->async_submit_wait,
1076 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1080 *nr_written += nr_pages;
1081 start = cur_end + 1;
1087 static noinline int csum_exist_in_range(struct btrfs_root *root,
1088 u64 bytenr, u64 num_bytes)
1091 struct btrfs_ordered_sum *sums;
1094 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1095 bytenr + num_bytes - 1, &list, 0);
1096 if (ret == 0 && list_empty(&list))
1099 while (!list_empty(&list)) {
1100 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1101 list_del(&sums->list);
1108 * when nowcow writeback call back. This checks for snapshots or COW copies
1109 * of the extents that exist in the file, and COWs the file as required.
1111 * If no cow copies or snapshots exist, we write directly to the existing
1114 static noinline int run_delalloc_nocow(struct inode *inode,
1115 struct page *locked_page,
1116 u64 start, u64 end, int *page_started, int force,
1117 unsigned long *nr_written)
1119 struct btrfs_root *root = BTRFS_I(inode)->root;
1120 struct btrfs_trans_handle *trans;
1121 struct extent_buffer *leaf;
1122 struct btrfs_path *path;
1123 struct btrfs_file_extent_item *fi;
1124 struct btrfs_key found_key;
1137 u64 ino = btrfs_ino(inode);
1139 path = btrfs_alloc_path();
1141 extent_clear_unlock_delalloc(inode,
1142 &BTRFS_I(inode)->io_tree,
1143 start, end, locked_page,
1144 EXTENT_CLEAR_UNLOCK_PAGE |
1145 EXTENT_CLEAR_UNLOCK |
1146 EXTENT_CLEAR_DELALLOC |
1147 EXTENT_CLEAR_DIRTY |
1148 EXTENT_SET_WRITEBACK |
1149 EXTENT_END_WRITEBACK);
1153 nolock = btrfs_is_free_space_inode(inode);
1156 trans = btrfs_join_transaction_nolock(root);
1158 trans = btrfs_join_transaction(root);
1160 if (IS_ERR(trans)) {
1161 extent_clear_unlock_delalloc(inode,
1162 &BTRFS_I(inode)->io_tree,
1163 start, end, locked_page,
1164 EXTENT_CLEAR_UNLOCK_PAGE |
1165 EXTENT_CLEAR_UNLOCK |
1166 EXTENT_CLEAR_DELALLOC |
1167 EXTENT_CLEAR_DIRTY |
1168 EXTENT_SET_WRITEBACK |
1169 EXTENT_END_WRITEBACK);
1170 btrfs_free_path(path);
1171 return PTR_ERR(trans);
1174 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1176 cow_start = (u64)-1;
1179 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1182 btrfs_abort_transaction(trans, root, ret);
1185 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1186 leaf = path->nodes[0];
1187 btrfs_item_key_to_cpu(leaf, &found_key,
1188 path->slots[0] - 1);
1189 if (found_key.objectid == ino &&
1190 found_key.type == BTRFS_EXTENT_DATA_KEY)
1195 leaf = path->nodes[0];
1196 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1197 ret = btrfs_next_leaf(root, path);
1199 btrfs_abort_transaction(trans, root, ret);
1204 leaf = path->nodes[0];
1210 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1212 if (found_key.objectid > ino ||
1213 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1214 found_key.offset > end)
1217 if (found_key.offset > cur_offset) {
1218 extent_end = found_key.offset;
1223 fi = btrfs_item_ptr(leaf, path->slots[0],
1224 struct btrfs_file_extent_item);
1225 extent_type = btrfs_file_extent_type(leaf, fi);
1227 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1228 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1229 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1230 extent_offset = btrfs_file_extent_offset(leaf, fi);
1231 extent_end = found_key.offset +
1232 btrfs_file_extent_num_bytes(leaf, fi);
1233 if (extent_end <= start) {
1237 if (disk_bytenr == 0)
1239 if (btrfs_file_extent_compression(leaf, fi) ||
1240 btrfs_file_extent_encryption(leaf, fi) ||
1241 btrfs_file_extent_other_encoding(leaf, fi))
1243 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1245 if (btrfs_extent_readonly(root, disk_bytenr))
1247 if (btrfs_cross_ref_exist(trans, root, ino,
1249 extent_offset, disk_bytenr))
1251 disk_bytenr += extent_offset;
1252 disk_bytenr += cur_offset - found_key.offset;
1253 num_bytes = min(end + 1, extent_end) - cur_offset;
1255 * force cow if csum exists in the range.
1256 * this ensure that csum for a given extent are
1257 * either valid or do not exist.
1259 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1262 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1263 extent_end = found_key.offset +
1264 btrfs_file_extent_inline_len(leaf, fi);
1265 extent_end = ALIGN(extent_end, root->sectorsize);
1270 if (extent_end <= start) {
1275 if (cow_start == (u64)-1)
1276 cow_start = cur_offset;
1277 cur_offset = extent_end;
1278 if (cur_offset > end)
1284 btrfs_release_path(path);
1285 if (cow_start != (u64)-1) {
1286 ret = cow_file_range(inode, locked_page, cow_start,
1287 found_key.offset - 1, page_started,
1290 btrfs_abort_transaction(trans, root, ret);
1293 cow_start = (u64)-1;
1296 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1297 struct extent_map *em;
1298 struct extent_map_tree *em_tree;
1299 em_tree = &BTRFS_I(inode)->extent_tree;
1300 em = alloc_extent_map();
1301 BUG_ON(!em); /* -ENOMEM */
1302 em->start = cur_offset;
1303 em->orig_start = em->start;
1304 em->len = num_bytes;
1305 em->block_len = num_bytes;
1306 em->block_start = disk_bytenr;
1307 em->bdev = root->fs_info->fs_devices->latest_bdev;
1308 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1309 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
1311 write_lock(&em_tree->lock);
1312 ret = add_extent_mapping(em_tree, em);
1313 write_unlock(&em_tree->lock);
1314 if (ret != -EEXIST) {
1315 free_extent_map(em);
1318 btrfs_drop_extent_cache(inode, em->start,
1319 em->start + em->len - 1, 0);
1321 type = BTRFS_ORDERED_PREALLOC;
1323 type = BTRFS_ORDERED_NOCOW;
1326 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1327 num_bytes, num_bytes, type);
1328 BUG_ON(ret); /* -ENOMEM */
1330 if (root->root_key.objectid ==
1331 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1332 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1335 btrfs_abort_transaction(trans, root, ret);
1340 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1341 cur_offset, cur_offset + num_bytes - 1,
1342 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1343 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1344 EXTENT_SET_PRIVATE2);
1345 cur_offset = extent_end;
1346 if (cur_offset > end)
1349 btrfs_release_path(path);
1351 if (cur_offset <= end && cow_start == (u64)-1) {
1352 cow_start = cur_offset;
1356 if (cow_start != (u64)-1) {
1357 ret = cow_file_range(inode, locked_page, cow_start, end,
1358 page_started, nr_written, 1);
1360 btrfs_abort_transaction(trans, root, ret);
1367 err = btrfs_end_transaction_nolock(trans, root);
1369 err = btrfs_end_transaction(trans, root);
1374 if (ret && cur_offset < end)
1375 extent_clear_unlock_delalloc(inode,
1376 &BTRFS_I(inode)->io_tree,
1377 cur_offset, end, locked_page,
1378 EXTENT_CLEAR_UNLOCK_PAGE |
1379 EXTENT_CLEAR_UNLOCK |
1380 EXTENT_CLEAR_DELALLOC |
1381 EXTENT_CLEAR_DIRTY |
1382 EXTENT_SET_WRITEBACK |
1383 EXTENT_END_WRITEBACK);
1385 btrfs_free_path(path);
1390 * extent_io.c call back to do delayed allocation processing
1392 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1393 u64 start, u64 end, int *page_started,
1394 unsigned long *nr_written)
1397 struct btrfs_root *root = BTRFS_I(inode)->root;
1399 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1400 ret = run_delalloc_nocow(inode, locked_page, start, end,
1401 page_started, 1, nr_written);
1402 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1403 ret = run_delalloc_nocow(inode, locked_page, start, end,
1404 page_started, 0, nr_written);
1405 } else if (!btrfs_test_opt(root, COMPRESS) &&
1406 !(BTRFS_I(inode)->force_compress) &&
1407 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1408 ret = cow_file_range(inode, locked_page, start, end,
1409 page_started, nr_written, 1);
1411 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1412 &BTRFS_I(inode)->runtime_flags);
1413 ret = cow_file_range_async(inode, locked_page, start, end,
1414 page_started, nr_written);
1419 static void btrfs_split_extent_hook(struct inode *inode,
1420 struct extent_state *orig, u64 split)
1422 /* not delalloc, ignore it */
1423 if (!(orig->state & EXTENT_DELALLOC))
1426 spin_lock(&BTRFS_I(inode)->lock);
1427 BTRFS_I(inode)->outstanding_extents++;
1428 spin_unlock(&BTRFS_I(inode)->lock);
1432 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1433 * extents so we can keep track of new extents that are just merged onto old
1434 * extents, such as when we are doing sequential writes, so we can properly
1435 * account for the metadata space we'll need.
1437 static void btrfs_merge_extent_hook(struct inode *inode,
1438 struct extent_state *new,
1439 struct extent_state *other)
1441 /* not delalloc, ignore it */
1442 if (!(other->state & EXTENT_DELALLOC))
1445 spin_lock(&BTRFS_I(inode)->lock);
1446 BTRFS_I(inode)->outstanding_extents--;
1447 spin_unlock(&BTRFS_I(inode)->lock);
1451 * extent_io.c set_bit_hook, used to track delayed allocation
1452 * bytes in this file, and to maintain the list of inodes that
1453 * have pending delalloc work to be done.
1455 static void btrfs_set_bit_hook(struct inode *inode,
1456 struct extent_state *state, int *bits)
1460 * set_bit and clear bit hooks normally require _irqsave/restore
1461 * but in this case, we are only testing for the DELALLOC
1462 * bit, which is only set or cleared with irqs on
1464 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1465 struct btrfs_root *root = BTRFS_I(inode)->root;
1466 u64 len = state->end + 1 - state->start;
1467 bool do_list = !btrfs_is_free_space_inode(inode);
1469 if (*bits & EXTENT_FIRST_DELALLOC) {
1470 *bits &= ~EXTENT_FIRST_DELALLOC;
1472 spin_lock(&BTRFS_I(inode)->lock);
1473 BTRFS_I(inode)->outstanding_extents++;
1474 spin_unlock(&BTRFS_I(inode)->lock);
1477 spin_lock(&root->fs_info->delalloc_lock);
1478 BTRFS_I(inode)->delalloc_bytes += len;
1479 root->fs_info->delalloc_bytes += len;
1480 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1481 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1482 &root->fs_info->delalloc_inodes);
1484 spin_unlock(&root->fs_info->delalloc_lock);
1489 * extent_io.c clear_bit_hook, see set_bit_hook for why
1491 static void btrfs_clear_bit_hook(struct inode *inode,
1492 struct extent_state *state, int *bits)
1495 * set_bit and clear bit hooks normally require _irqsave/restore
1496 * but in this case, we are only testing for the DELALLOC
1497 * bit, which is only set or cleared with irqs on
1499 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1500 struct btrfs_root *root = BTRFS_I(inode)->root;
1501 u64 len = state->end + 1 - state->start;
1502 bool do_list = !btrfs_is_free_space_inode(inode);
1504 if (*bits & EXTENT_FIRST_DELALLOC) {
1505 *bits &= ~EXTENT_FIRST_DELALLOC;
1506 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1507 spin_lock(&BTRFS_I(inode)->lock);
1508 BTRFS_I(inode)->outstanding_extents--;
1509 spin_unlock(&BTRFS_I(inode)->lock);
1512 if (*bits & EXTENT_DO_ACCOUNTING)
1513 btrfs_delalloc_release_metadata(inode, len);
1515 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1517 btrfs_free_reserved_data_space(inode, len);
1519 spin_lock(&root->fs_info->delalloc_lock);
1520 root->fs_info->delalloc_bytes -= len;
1521 BTRFS_I(inode)->delalloc_bytes -= len;
1523 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1524 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1525 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1527 spin_unlock(&root->fs_info->delalloc_lock);
1532 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1533 * we don't create bios that span stripes or chunks
1535 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1536 size_t size, struct bio *bio,
1537 unsigned long bio_flags)
1539 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1540 struct btrfs_mapping_tree *map_tree;
1541 u64 logical = (u64)bio->bi_sector << 9;
1546 if (bio_flags & EXTENT_BIO_COMPRESSED)
1549 length = bio->bi_size;
1550 map_tree = &root->fs_info->mapping_tree;
1551 map_length = length;
1552 ret = btrfs_map_block(map_tree, READ, logical,
1553 &map_length, NULL, 0);
1554 /* Will always return 0 or 1 with map_multi == NULL */
1556 if (map_length < length + size)
1562 * in order to insert checksums into the metadata in large chunks,
1563 * we wait until bio submission time. All the pages in the bio are
1564 * checksummed and sums are attached onto the ordered extent record.
1566 * At IO completion time the cums attached on the ordered extent record
1567 * are inserted into the btree
1569 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1570 struct bio *bio, int mirror_num,
1571 unsigned long bio_flags,
1574 struct btrfs_root *root = BTRFS_I(inode)->root;
1577 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1578 BUG_ON(ret); /* -ENOMEM */
1583 * in order to insert checksums into the metadata in large chunks,
1584 * we wait until bio submission time. All the pages in the bio are
1585 * checksummed and sums are attached onto the ordered extent record.
1587 * At IO completion time the cums attached on the ordered extent record
1588 * are inserted into the btree
1590 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1591 int mirror_num, unsigned long bio_flags,
1594 struct btrfs_root *root = BTRFS_I(inode)->root;
1595 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1599 * extent_io.c submission hook. This does the right thing for csum calculation
1600 * on write, or reading the csums from the tree before a read
1602 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1603 int mirror_num, unsigned long bio_flags,
1606 struct btrfs_root *root = BTRFS_I(inode)->root;
1611 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1613 if (btrfs_is_free_space_inode(inode))
1616 if (!(rw & REQ_WRITE)) {
1617 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1621 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1622 return btrfs_submit_compressed_read(inode, bio,
1623 mirror_num, bio_flags);
1624 } else if (!skip_sum) {
1625 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1630 } else if (!skip_sum) {
1631 /* csum items have already been cloned */
1632 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1634 /* we're doing a write, do the async checksumming */
1635 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1636 inode, rw, bio, mirror_num,
1637 bio_flags, bio_offset,
1638 __btrfs_submit_bio_start,
1639 __btrfs_submit_bio_done);
1643 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1647 * given a list of ordered sums record them in the inode. This happens
1648 * at IO completion time based on sums calculated at bio submission time.
1650 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1651 struct inode *inode, u64 file_offset,
1652 struct list_head *list)
1654 struct btrfs_ordered_sum *sum;
1656 list_for_each_entry(sum, list, list) {
1657 btrfs_csum_file_blocks(trans,
1658 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1663 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1664 struct extent_state **cached_state)
1666 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1668 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1669 cached_state, GFP_NOFS);
1672 /* see btrfs_writepage_start_hook for details on why this is required */
1673 struct btrfs_writepage_fixup {
1675 struct btrfs_work work;
1678 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1680 struct btrfs_writepage_fixup *fixup;
1681 struct btrfs_ordered_extent *ordered;
1682 struct extent_state *cached_state = NULL;
1684 struct inode *inode;
1689 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1693 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1694 ClearPageChecked(page);
1698 inode = page->mapping->host;
1699 page_start = page_offset(page);
1700 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1702 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1705 /* already ordered? We're done */
1706 if (PagePrivate2(page))
1709 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1711 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1712 page_end, &cached_state, GFP_NOFS);
1714 btrfs_start_ordered_extent(inode, ordered, 1);
1715 btrfs_put_ordered_extent(ordered);
1719 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1721 mapping_set_error(page->mapping, ret);
1722 end_extent_writepage(page, ret, page_start, page_end);
1723 ClearPageChecked(page);
1727 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1728 ClearPageChecked(page);
1729 set_page_dirty(page);
1731 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1732 &cached_state, GFP_NOFS);
1735 page_cache_release(page);
1740 * There are a few paths in the higher layers of the kernel that directly
1741 * set the page dirty bit without asking the filesystem if it is a
1742 * good idea. This causes problems because we want to make sure COW
1743 * properly happens and the data=ordered rules are followed.
1745 * In our case any range that doesn't have the ORDERED bit set
1746 * hasn't been properly setup for IO. We kick off an async process
1747 * to fix it up. The async helper will wait for ordered extents, set
1748 * the delalloc bit and make it safe to write the page.
1750 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1752 struct inode *inode = page->mapping->host;
1753 struct btrfs_writepage_fixup *fixup;
1754 struct btrfs_root *root = BTRFS_I(inode)->root;
1756 /* this page is properly in the ordered list */
1757 if (TestClearPagePrivate2(page))
1760 if (PageChecked(page))
1763 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1767 SetPageChecked(page);
1768 page_cache_get(page);
1769 fixup->work.func = btrfs_writepage_fixup_worker;
1771 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1775 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1776 struct inode *inode, u64 file_pos,
1777 u64 disk_bytenr, u64 disk_num_bytes,
1778 u64 num_bytes, u64 ram_bytes,
1779 u8 compression, u8 encryption,
1780 u16 other_encoding, int extent_type)
1782 struct btrfs_root *root = BTRFS_I(inode)->root;
1783 struct btrfs_file_extent_item *fi;
1784 struct btrfs_path *path;
1785 struct extent_buffer *leaf;
1786 struct btrfs_key ins;
1789 path = btrfs_alloc_path();
1793 path->leave_spinning = 1;
1796 * we may be replacing one extent in the tree with another.
1797 * The new extent is pinned in the extent map, and we don't want
1798 * to drop it from the cache until it is completely in the btree.
1800 * So, tell btrfs_drop_extents to leave this extent in the cache.
1801 * the caller is expected to unpin it and allow it to be merged
1804 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1805 file_pos + num_bytes, 0);
1809 ins.objectid = btrfs_ino(inode);
1810 ins.offset = file_pos;
1811 ins.type = BTRFS_EXTENT_DATA_KEY;
1812 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1815 leaf = path->nodes[0];
1816 fi = btrfs_item_ptr(leaf, path->slots[0],
1817 struct btrfs_file_extent_item);
1818 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1819 btrfs_set_file_extent_type(leaf, fi, extent_type);
1820 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1821 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1822 btrfs_set_file_extent_offset(leaf, fi, 0);
1823 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1824 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1825 btrfs_set_file_extent_compression(leaf, fi, compression);
1826 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1827 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1829 btrfs_unlock_up_safe(path, 1);
1830 btrfs_set_lock_blocking(leaf);
1832 btrfs_mark_buffer_dirty(leaf);
1834 inode_add_bytes(inode, num_bytes);
1836 ins.objectid = disk_bytenr;
1837 ins.offset = disk_num_bytes;
1838 ins.type = BTRFS_EXTENT_ITEM_KEY;
1839 ret = btrfs_alloc_reserved_file_extent(trans, root,
1840 root->root_key.objectid,
1841 btrfs_ino(inode), file_pos, &ins);
1843 btrfs_free_path(path);
1849 * helper function for btrfs_finish_ordered_io, this
1850 * just reads in some of the csum leaves to prime them into ram
1851 * before we start the transaction. It limits the amount of btree
1852 * reads required while inside the transaction.
1854 /* as ordered data IO finishes, this gets called so we can finish
1855 * an ordered extent if the range of bytes in the file it covers are
1858 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
1860 struct inode *inode = ordered_extent->inode;
1861 struct btrfs_root *root = BTRFS_I(inode)->root;
1862 struct btrfs_trans_handle *trans = NULL;
1863 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1864 struct extent_state *cached_state = NULL;
1865 int compress_type = 0;
1869 nolock = btrfs_is_free_space_inode(inode);
1871 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1876 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1877 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1878 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1881 trans = btrfs_join_transaction_nolock(root);
1883 trans = btrfs_join_transaction(root);
1884 if (IS_ERR(trans)) {
1885 ret = PTR_ERR(trans);
1889 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1890 ret = btrfs_update_inode_fallback(trans, root, inode);
1891 if (ret) /* -ENOMEM or corruption */
1892 btrfs_abort_transaction(trans, root, ret);
1897 lock_extent_bits(io_tree, ordered_extent->file_offset,
1898 ordered_extent->file_offset + ordered_extent->len - 1,
1902 trans = btrfs_join_transaction_nolock(root);
1904 trans = btrfs_join_transaction(root);
1905 if (IS_ERR(trans)) {
1906 ret = PTR_ERR(trans);
1910 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1912 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1913 compress_type = ordered_extent->compress_type;
1914 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1915 BUG_ON(compress_type);
1916 ret = btrfs_mark_extent_written(trans, inode,
1917 ordered_extent->file_offset,
1918 ordered_extent->file_offset +
1919 ordered_extent->len);
1921 BUG_ON(root == root->fs_info->tree_root);
1922 ret = insert_reserved_file_extent(trans, inode,
1923 ordered_extent->file_offset,
1924 ordered_extent->start,
1925 ordered_extent->disk_len,
1926 ordered_extent->len,
1927 ordered_extent->len,
1928 compress_type, 0, 0,
1929 BTRFS_FILE_EXTENT_REG);
1931 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1932 ordered_extent->file_offset, ordered_extent->len,
1935 btrfs_abort_transaction(trans, root, ret);
1939 add_pending_csums(trans, inode, ordered_extent->file_offset,
1940 &ordered_extent->list);
1942 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1943 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1944 ret = btrfs_update_inode_fallback(trans, root, inode);
1945 if (ret) { /* -ENOMEM or corruption */
1946 btrfs_abort_transaction(trans, root, ret);
1950 btrfs_set_inode_last_trans(trans, inode);
1954 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1955 ordered_extent->file_offset +
1956 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1958 if (root != root->fs_info->tree_root)
1959 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1962 btrfs_end_transaction_nolock(trans, root);
1964 btrfs_end_transaction(trans, root);
1968 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
1969 ordered_extent->file_offset +
1970 ordered_extent->len - 1, NULL, GFP_NOFS);
1973 * This needs to be dont to make sure anybody waiting knows we are done
1974 * upating everything for this ordered extent.
1976 btrfs_remove_ordered_extent(inode, ordered_extent);
1979 btrfs_put_ordered_extent(ordered_extent);
1980 /* once for the tree */
1981 btrfs_put_ordered_extent(ordered_extent);
1986 static void finish_ordered_fn(struct btrfs_work *work)
1988 struct btrfs_ordered_extent *ordered_extent;
1989 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
1990 btrfs_finish_ordered_io(ordered_extent);
1993 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1994 struct extent_state *state, int uptodate)
1996 struct inode *inode = page->mapping->host;
1997 struct btrfs_root *root = BTRFS_I(inode)->root;
1998 struct btrfs_ordered_extent *ordered_extent = NULL;
1999 struct btrfs_workers *workers;
2001 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2003 ClearPagePrivate2(page);
2004 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2005 end - start + 1, uptodate))
2008 ordered_extent->work.func = finish_ordered_fn;
2009 ordered_extent->work.flags = 0;
2011 if (btrfs_is_free_space_inode(inode))
2012 workers = &root->fs_info->endio_freespace_worker;
2014 workers = &root->fs_info->endio_write_workers;
2015 btrfs_queue_worker(workers, &ordered_extent->work);
2021 * when reads are done, we need to check csums to verify the data is correct
2022 * if there's a match, we allow the bio to finish. If not, the code in
2023 * extent_io.c will try to find good copies for us.
2025 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2026 struct extent_state *state, int mirror)
2028 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
2029 struct inode *inode = page->mapping->host;
2030 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2032 u64 private = ~(u32)0;
2034 struct btrfs_root *root = BTRFS_I(inode)->root;
2037 if (PageChecked(page)) {
2038 ClearPageChecked(page);
2042 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2045 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2046 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2047 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2052 if (state && state->start == start) {
2053 private = state->private;
2056 ret = get_state_private(io_tree, start, &private);
2058 kaddr = kmap_atomic(page);
2062 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
2063 btrfs_csum_final(csum, (char *)&csum);
2064 if (csum != private)
2067 kunmap_atomic(kaddr);
2072 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2074 (unsigned long long)btrfs_ino(page->mapping->host),
2075 (unsigned long long)start, csum,
2076 (unsigned long long)private);
2077 memset(kaddr + offset, 1, end - start + 1);
2078 flush_dcache_page(page);
2079 kunmap_atomic(kaddr);
2085 struct delayed_iput {
2086 struct list_head list;
2087 struct inode *inode;
2090 /* JDM: If this is fs-wide, why can't we add a pointer to
2091 * btrfs_inode instead and avoid the allocation? */
2092 void btrfs_add_delayed_iput(struct inode *inode)
2094 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2095 struct delayed_iput *delayed;
2097 if (atomic_add_unless(&inode->i_count, -1, 1))
2100 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2101 delayed->inode = inode;
2103 spin_lock(&fs_info->delayed_iput_lock);
2104 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2105 spin_unlock(&fs_info->delayed_iput_lock);
2108 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2111 struct btrfs_fs_info *fs_info = root->fs_info;
2112 struct delayed_iput *delayed;
2115 spin_lock(&fs_info->delayed_iput_lock);
2116 empty = list_empty(&fs_info->delayed_iputs);
2117 spin_unlock(&fs_info->delayed_iput_lock);
2121 spin_lock(&fs_info->delayed_iput_lock);
2122 list_splice_init(&fs_info->delayed_iputs, &list);
2123 spin_unlock(&fs_info->delayed_iput_lock);
2125 while (!list_empty(&list)) {
2126 delayed = list_entry(list.next, struct delayed_iput, list);
2127 list_del(&delayed->list);
2128 iput(delayed->inode);
2133 enum btrfs_orphan_cleanup_state {
2134 ORPHAN_CLEANUP_STARTED = 1,
2135 ORPHAN_CLEANUP_DONE = 2,
2139 * This is called in transaction commit time. If there are no orphan
2140 * files in the subvolume, it removes orphan item and frees block_rsv
2143 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2144 struct btrfs_root *root)
2146 struct btrfs_block_rsv *block_rsv;
2149 if (atomic_read(&root->orphan_inodes) ||
2150 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2153 spin_lock(&root->orphan_lock);
2154 if (atomic_read(&root->orphan_inodes)) {
2155 spin_unlock(&root->orphan_lock);
2159 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2160 spin_unlock(&root->orphan_lock);
2164 block_rsv = root->orphan_block_rsv;
2165 root->orphan_block_rsv = NULL;
2166 spin_unlock(&root->orphan_lock);
2168 if (root->orphan_item_inserted &&
2169 btrfs_root_refs(&root->root_item) > 0) {
2170 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2171 root->root_key.objectid);
2173 root->orphan_item_inserted = 0;
2177 WARN_ON(block_rsv->size > 0);
2178 btrfs_free_block_rsv(root, block_rsv);
2183 * This creates an orphan entry for the given inode in case something goes
2184 * wrong in the middle of an unlink/truncate.
2186 * NOTE: caller of this function should reserve 5 units of metadata for
2189 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2191 struct btrfs_root *root = BTRFS_I(inode)->root;
2192 struct btrfs_block_rsv *block_rsv = NULL;
2197 if (!root->orphan_block_rsv) {
2198 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2203 spin_lock(&root->orphan_lock);
2204 if (!root->orphan_block_rsv) {
2205 root->orphan_block_rsv = block_rsv;
2206 } else if (block_rsv) {
2207 btrfs_free_block_rsv(root, block_rsv);
2211 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2212 &BTRFS_I(inode)->runtime_flags)) {
2215 * For proper ENOSPC handling, we should do orphan
2216 * cleanup when mounting. But this introduces backward
2217 * compatibility issue.
2219 if (!xchg(&root->orphan_item_inserted, 1))
2225 atomic_inc(&root->orphan_inodes);
2228 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2229 &BTRFS_I(inode)->runtime_flags))
2231 spin_unlock(&root->orphan_lock);
2233 /* grab metadata reservation from transaction handle */
2235 ret = btrfs_orphan_reserve_metadata(trans, inode);
2236 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2239 /* insert an orphan item to track this unlinked/truncated file */
2241 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2242 if (ret && ret != -EEXIST) {
2243 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2244 &BTRFS_I(inode)->runtime_flags);
2245 btrfs_abort_transaction(trans, root, ret);
2251 /* insert an orphan item to track subvolume contains orphan files */
2253 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2254 root->root_key.objectid);
2255 if (ret && ret != -EEXIST) {
2256 btrfs_abort_transaction(trans, root, ret);
2264 * We have done the truncate/delete so we can go ahead and remove the orphan
2265 * item for this particular inode.
2267 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2269 struct btrfs_root *root = BTRFS_I(inode)->root;
2270 int delete_item = 0;
2271 int release_rsv = 0;
2274 spin_lock(&root->orphan_lock);
2275 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2276 &BTRFS_I(inode)->runtime_flags))
2279 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2280 &BTRFS_I(inode)->runtime_flags))
2282 spin_unlock(&root->orphan_lock);
2284 if (trans && delete_item) {
2285 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2286 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2290 btrfs_orphan_release_metadata(inode);
2291 atomic_dec(&root->orphan_inodes);
2298 * this cleans up any orphans that may be left on the list from the last use
2301 int btrfs_orphan_cleanup(struct btrfs_root *root)
2303 struct btrfs_path *path;
2304 struct extent_buffer *leaf;
2305 struct btrfs_key key, found_key;
2306 struct btrfs_trans_handle *trans;
2307 struct inode *inode;
2308 u64 last_objectid = 0;
2309 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2311 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2314 path = btrfs_alloc_path();
2321 key.objectid = BTRFS_ORPHAN_OBJECTID;
2322 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2323 key.offset = (u64)-1;
2326 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2331 * if ret == 0 means we found what we were searching for, which
2332 * is weird, but possible, so only screw with path if we didn't
2333 * find the key and see if we have stuff that matches
2337 if (path->slots[0] == 0)
2342 /* pull out the item */
2343 leaf = path->nodes[0];
2344 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2346 /* make sure the item matches what we want */
2347 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2349 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2352 /* release the path since we're done with it */
2353 btrfs_release_path(path);
2356 * this is where we are basically btrfs_lookup, without the
2357 * crossing root thing. we store the inode number in the
2358 * offset of the orphan item.
2361 if (found_key.offset == last_objectid) {
2362 printk(KERN_ERR "btrfs: Error removing orphan entry, "
2363 "stopping orphan cleanup\n");
2368 last_objectid = found_key.offset;
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 ret = PTR_RET(inode);
2375 if (ret && ret != -ESTALE)
2378 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2379 struct btrfs_root *dead_root;
2380 struct btrfs_fs_info *fs_info = root->fs_info;
2381 int is_dead_root = 0;
2384 * this is an orphan in the tree root. Currently these
2385 * could come from 2 sources:
2386 * a) a snapshot deletion in progress
2387 * b) a free space cache inode
2388 * We need to distinguish those two, as the snapshot
2389 * orphan must not get deleted.
2390 * find_dead_roots already ran before us, so if this
2391 * is a snapshot deletion, we should find the root
2392 * in the dead_roots list
2394 spin_lock(&fs_info->trans_lock);
2395 list_for_each_entry(dead_root, &fs_info->dead_roots,
2397 if (dead_root->root_key.objectid ==
2398 found_key.objectid) {
2403 spin_unlock(&fs_info->trans_lock);
2405 /* prevent this orphan from being found again */
2406 key.offset = found_key.objectid - 1;
2411 * Inode is already gone but the orphan item is still there,
2412 * kill the orphan item.
2414 if (ret == -ESTALE) {
2415 trans = btrfs_start_transaction(root, 1);
2416 if (IS_ERR(trans)) {
2417 ret = PTR_ERR(trans);
2420 printk(KERN_ERR "auto deleting %Lu\n",
2421 found_key.objectid);
2422 ret = btrfs_del_orphan_item(trans, root,
2423 found_key.objectid);
2424 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2425 btrfs_end_transaction(trans, root);
2430 * add this inode to the orphan list so btrfs_orphan_del does
2431 * the proper thing when we hit it
2433 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2434 &BTRFS_I(inode)->runtime_flags);
2436 /* if we have links, this was a truncate, lets do that */
2437 if (inode->i_nlink) {
2438 if (!S_ISREG(inode->i_mode)) {
2444 ret = btrfs_truncate(inode);
2449 /* this will do delete_inode and everything for us */
2454 /* release the path since we're done with it */
2455 btrfs_release_path(path);
2457 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2459 if (root->orphan_block_rsv)
2460 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2463 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2464 trans = btrfs_join_transaction(root);
2466 btrfs_end_transaction(trans, root);
2470 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2472 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2476 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2477 btrfs_free_path(path);
2482 * very simple check to peek ahead in the leaf looking for xattrs. If we
2483 * don't find any xattrs, we know there can't be any acls.
2485 * slot is the slot the inode is in, objectid is the objectid of the inode
2487 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2488 int slot, u64 objectid)
2490 u32 nritems = btrfs_header_nritems(leaf);
2491 struct btrfs_key found_key;
2495 while (slot < nritems) {
2496 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2498 /* we found a different objectid, there must not be acls */
2499 if (found_key.objectid != objectid)
2502 /* we found an xattr, assume we've got an acl */
2503 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2507 * we found a key greater than an xattr key, there can't
2508 * be any acls later on
2510 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2517 * it goes inode, inode backrefs, xattrs, extents,
2518 * so if there are a ton of hard links to an inode there can
2519 * be a lot of backrefs. Don't waste time searching too hard,
2520 * this is just an optimization
2525 /* we hit the end of the leaf before we found an xattr or
2526 * something larger than an xattr. We have to assume the inode
2533 * read an inode from the btree into the in-memory inode
2535 static void btrfs_read_locked_inode(struct inode *inode)
2537 struct btrfs_path *path;
2538 struct extent_buffer *leaf;
2539 struct btrfs_inode_item *inode_item;
2540 struct btrfs_timespec *tspec;
2541 struct btrfs_root *root = BTRFS_I(inode)->root;
2542 struct btrfs_key location;
2546 bool filled = false;
2548 ret = btrfs_fill_inode(inode, &rdev);
2552 path = btrfs_alloc_path();
2556 path->leave_spinning = 1;
2557 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2559 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2563 leaf = path->nodes[0];
2568 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2569 struct btrfs_inode_item);
2570 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2571 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2572 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2573 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2574 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2576 tspec = btrfs_inode_atime(inode_item);
2577 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2578 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2580 tspec = btrfs_inode_mtime(inode_item);
2581 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2582 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2584 tspec = btrfs_inode_ctime(inode_item);
2585 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2586 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2588 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2589 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2590 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
2593 * If we were modified in the current generation and evicted from memory
2594 * and then re-read we need to do a full sync since we don't have any
2595 * idea about which extents were modified before we were evicted from
2598 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
2599 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2600 &BTRFS_I(inode)->runtime_flags);
2602 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2603 inode->i_generation = BTRFS_I(inode)->generation;
2605 rdev = btrfs_inode_rdev(leaf, inode_item);
2607 BTRFS_I(inode)->index_cnt = (u64)-1;
2608 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2611 * try to precache a NULL acl entry for files that don't have
2612 * any xattrs or acls
2614 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2617 cache_no_acl(inode);
2619 btrfs_free_path(path);
2621 switch (inode->i_mode & S_IFMT) {
2623 inode->i_mapping->a_ops = &btrfs_aops;
2624 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2625 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2626 inode->i_fop = &btrfs_file_operations;
2627 inode->i_op = &btrfs_file_inode_operations;
2630 inode->i_fop = &btrfs_dir_file_operations;
2631 if (root == root->fs_info->tree_root)
2632 inode->i_op = &btrfs_dir_ro_inode_operations;
2634 inode->i_op = &btrfs_dir_inode_operations;
2637 inode->i_op = &btrfs_symlink_inode_operations;
2638 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2639 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2642 inode->i_op = &btrfs_special_inode_operations;
2643 init_special_inode(inode, inode->i_mode, rdev);
2647 btrfs_update_iflags(inode);
2651 btrfs_free_path(path);
2652 make_bad_inode(inode);
2656 * given a leaf and an inode, copy the inode fields into the leaf
2658 static void fill_inode_item(struct btrfs_trans_handle *trans,
2659 struct extent_buffer *leaf,
2660 struct btrfs_inode_item *item,
2661 struct inode *inode)
2663 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2664 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2665 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2666 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2667 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2669 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2670 inode->i_atime.tv_sec);
2671 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2672 inode->i_atime.tv_nsec);
2674 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2675 inode->i_mtime.tv_sec);
2676 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2677 inode->i_mtime.tv_nsec);
2679 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2680 inode->i_ctime.tv_sec);
2681 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2682 inode->i_ctime.tv_nsec);
2684 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2685 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2686 btrfs_set_inode_sequence(leaf, item, inode->i_version);
2687 btrfs_set_inode_transid(leaf, item, trans->transid);
2688 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2689 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2690 btrfs_set_inode_block_group(leaf, item, 0);
2694 * copy everything in the in-memory inode into the btree.
2696 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2697 struct btrfs_root *root, struct inode *inode)
2699 struct btrfs_inode_item *inode_item;
2700 struct btrfs_path *path;
2701 struct extent_buffer *leaf;
2704 path = btrfs_alloc_path();
2708 path->leave_spinning = 1;
2709 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2717 btrfs_unlock_up_safe(path, 1);
2718 leaf = path->nodes[0];
2719 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2720 struct btrfs_inode_item);
2722 fill_inode_item(trans, leaf, inode_item, inode);
2723 btrfs_mark_buffer_dirty(leaf);
2724 btrfs_set_inode_last_trans(trans, inode);
2727 btrfs_free_path(path);
2732 * copy everything in the in-memory inode into the btree.
2734 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2735 struct btrfs_root *root, struct inode *inode)
2740 * If the inode is a free space inode, we can deadlock during commit
2741 * if we put it into the delayed code.
2743 * The data relocation inode should also be directly updated
2746 if (!btrfs_is_free_space_inode(inode)
2747 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2748 btrfs_update_root_times(trans, root);
2750 ret = btrfs_delayed_update_inode(trans, root, inode);
2752 btrfs_set_inode_last_trans(trans, inode);
2756 return btrfs_update_inode_item(trans, root, inode);
2759 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2760 struct btrfs_root *root, struct inode *inode)
2764 ret = btrfs_update_inode(trans, root, inode);
2766 return btrfs_update_inode_item(trans, root, inode);
2771 * unlink helper that gets used here in inode.c and in the tree logging
2772 * recovery code. It remove a link in a directory with a given name, and
2773 * also drops the back refs in the inode to the directory
2775 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2776 struct btrfs_root *root,
2777 struct inode *dir, struct inode *inode,
2778 const char *name, int name_len)
2780 struct btrfs_path *path;
2782 struct extent_buffer *leaf;
2783 struct btrfs_dir_item *di;
2784 struct btrfs_key key;
2786 u64 ino = btrfs_ino(inode);
2787 u64 dir_ino = btrfs_ino(dir);
2789 path = btrfs_alloc_path();
2795 path->leave_spinning = 1;
2796 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2797 name, name_len, -1);
2806 leaf = path->nodes[0];
2807 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2808 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2811 btrfs_release_path(path);
2813 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2816 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2817 "inode %llu parent %llu\n", name_len, name,
2818 (unsigned long long)ino, (unsigned long long)dir_ino);
2819 btrfs_abort_transaction(trans, root, ret);
2823 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2825 btrfs_abort_transaction(trans, root, ret);
2829 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2831 if (ret != 0 && ret != -ENOENT) {
2832 btrfs_abort_transaction(trans, root, ret);
2836 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2841 btrfs_free_path(path);
2845 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2846 inode_inc_iversion(inode);
2847 inode_inc_iversion(dir);
2848 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2849 ret = btrfs_update_inode(trans, root, dir);
2854 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2855 struct btrfs_root *root,
2856 struct inode *dir, struct inode *inode,
2857 const char *name, int name_len)
2860 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2862 btrfs_drop_nlink(inode);
2863 ret = btrfs_update_inode(trans, root, inode);
2869 /* helper to check if there is any shared block in the path */
2870 static int check_path_shared(struct btrfs_root *root,
2871 struct btrfs_path *path)
2873 struct extent_buffer *eb;
2877 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2880 if (!path->nodes[level])
2882 eb = path->nodes[level];
2883 if (!btrfs_block_can_be_shared(root, eb))
2885 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2894 * helper to start transaction for unlink and rmdir.
2896 * unlink and rmdir are special in btrfs, they do not always free space.
2897 * so in enospc case, we should make sure they will free space before
2898 * allowing them to use the global metadata reservation.
2900 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2901 struct dentry *dentry)
2903 struct btrfs_trans_handle *trans;
2904 struct btrfs_root *root = BTRFS_I(dir)->root;
2905 struct btrfs_path *path;
2906 struct btrfs_inode_ref *ref;
2907 struct btrfs_dir_item *di;
2908 struct inode *inode = dentry->d_inode;
2913 u64 ino = btrfs_ino(inode);
2914 u64 dir_ino = btrfs_ino(dir);
2917 * 1 for the possible orphan item
2918 * 1 for the dir item
2919 * 1 for the dir index
2920 * 1 for the inode ref
2921 * 1 for the inode ref in the tree log
2922 * 2 for the dir entries in the log
2925 trans = btrfs_start_transaction(root, 8);
2926 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2929 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2930 return ERR_PTR(-ENOSPC);
2932 /* check if there is someone else holds reference */
2933 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2934 return ERR_PTR(-ENOSPC);
2936 if (atomic_read(&inode->i_count) > 2)
2937 return ERR_PTR(-ENOSPC);
2939 if (xchg(&root->fs_info->enospc_unlink, 1))
2940 return ERR_PTR(-ENOSPC);
2942 path = btrfs_alloc_path();
2944 root->fs_info->enospc_unlink = 0;
2945 return ERR_PTR(-ENOMEM);
2948 /* 1 for the orphan item */
2949 trans = btrfs_start_transaction(root, 1);
2950 if (IS_ERR(trans)) {
2951 btrfs_free_path(path);
2952 root->fs_info->enospc_unlink = 0;
2956 path->skip_locking = 1;
2957 path->search_commit_root = 1;
2959 ret = btrfs_lookup_inode(trans, root, path,
2960 &BTRFS_I(dir)->location, 0);
2966 if (check_path_shared(root, path))
2971 btrfs_release_path(path);
2973 ret = btrfs_lookup_inode(trans, root, path,
2974 &BTRFS_I(inode)->location, 0);
2980 if (check_path_shared(root, path))
2985 btrfs_release_path(path);
2987 if (ret == 0 && S_ISREG(inode->i_mode)) {
2988 ret = btrfs_lookup_file_extent(trans, root, path,
2994 BUG_ON(ret == 0); /* Corruption */
2995 if (check_path_shared(root, path))
2997 btrfs_release_path(path);
3005 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3006 dentry->d_name.name, dentry->d_name.len, 0);
3012 if (check_path_shared(root, path))
3018 btrfs_release_path(path);
3020 ref = btrfs_lookup_inode_ref(trans, root, path,
3021 dentry->d_name.name, dentry->d_name.len,
3027 BUG_ON(!ref); /* Logic error */
3028 if (check_path_shared(root, path))
3030 index = btrfs_inode_ref_index(path->nodes[0], ref);
3031 btrfs_release_path(path);
3034 * This is a commit root search, if we can lookup inode item and other
3035 * relative items in the commit root, it means the transaction of
3036 * dir/file creation has been committed, and the dir index item that we
3037 * delay to insert has also been inserted into the commit root. So
3038 * we needn't worry about the delayed insertion of the dir index item
3041 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3042 dentry->d_name.name, dentry->d_name.len, 0);
3047 BUG_ON(ret == -ENOENT);
3048 if (check_path_shared(root, path))
3053 btrfs_free_path(path);
3054 /* Migrate the orphan reservation over */
3056 err = btrfs_block_rsv_migrate(trans->block_rsv,
3057 &root->fs_info->global_block_rsv,
3058 trans->bytes_reserved);
3061 btrfs_end_transaction(trans, root);
3062 root->fs_info->enospc_unlink = 0;
3063 return ERR_PTR(err);
3066 trans->block_rsv = &root->fs_info->global_block_rsv;
3070 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3071 struct btrfs_root *root)
3073 if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3074 btrfs_block_rsv_release(root, trans->block_rsv,
3075 trans->bytes_reserved);
3076 trans->block_rsv = &root->fs_info->trans_block_rsv;
3077 BUG_ON(!root->fs_info->enospc_unlink);
3078 root->fs_info->enospc_unlink = 0;
3080 btrfs_end_transaction(trans, root);
3083 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3085 struct btrfs_root *root = BTRFS_I(dir)->root;
3086 struct btrfs_trans_handle *trans;
3087 struct inode *inode = dentry->d_inode;
3089 unsigned long nr = 0;
3091 trans = __unlink_start_trans(dir, dentry);
3093 return PTR_ERR(trans);
3095 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3097 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3098 dentry->d_name.name, dentry->d_name.len);
3102 if (inode->i_nlink == 0) {
3103 ret = btrfs_orphan_add(trans, inode);
3109 nr = trans->blocks_used;
3110 __unlink_end_trans(trans, root);
3111 btrfs_btree_balance_dirty(root, nr);
3115 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3116 struct btrfs_root *root,
3117 struct inode *dir, u64 objectid,
3118 const char *name, int name_len)
3120 struct btrfs_path *path;
3121 struct extent_buffer *leaf;
3122 struct btrfs_dir_item *di;
3123 struct btrfs_key key;
3126 u64 dir_ino = btrfs_ino(dir);
3128 path = btrfs_alloc_path();
3132 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3133 name, name_len, -1);
3134 if (IS_ERR_OR_NULL(di)) {
3142 leaf = path->nodes[0];
3143 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3144 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3145 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3147 btrfs_abort_transaction(trans, root, ret);
3150 btrfs_release_path(path);
3152 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3153 objectid, root->root_key.objectid,
3154 dir_ino, &index, name, name_len);
3156 if (ret != -ENOENT) {
3157 btrfs_abort_transaction(trans, root, ret);
3160 di = btrfs_search_dir_index_item(root, path, dir_ino,
3162 if (IS_ERR_OR_NULL(di)) {
3167 btrfs_abort_transaction(trans, root, ret);
3171 leaf = path->nodes[0];
3172 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3173 btrfs_release_path(path);
3176 btrfs_release_path(path);
3178 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3180 btrfs_abort_transaction(trans, root, ret);
3184 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3185 inode_inc_iversion(dir);
3186 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3187 ret = btrfs_update_inode_fallback(trans, root, dir);
3189 btrfs_abort_transaction(trans, root, ret);
3191 btrfs_free_path(path);
3195 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3197 struct inode *inode = dentry->d_inode;
3199 struct btrfs_root *root = BTRFS_I(dir)->root;
3200 struct btrfs_trans_handle *trans;
3201 unsigned long nr = 0;
3203 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3205 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3208 trans = __unlink_start_trans(dir, dentry);
3210 return PTR_ERR(trans);
3212 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3213 err = btrfs_unlink_subvol(trans, root, dir,
3214 BTRFS_I(inode)->location.objectid,
3215 dentry->d_name.name,
3216 dentry->d_name.len);
3220 err = btrfs_orphan_add(trans, inode);
3224 /* now the directory is empty */
3225 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3226 dentry->d_name.name, dentry->d_name.len);
3228 btrfs_i_size_write(inode, 0);
3230 nr = trans->blocks_used;
3231 __unlink_end_trans(trans, root);
3232 btrfs_btree_balance_dirty(root, nr);
3238 * this can truncate away extent items, csum items and directory items.
3239 * It starts at a high offset and removes keys until it can't find
3240 * any higher than new_size
3242 * csum items that cross the new i_size are truncated to the new size
3245 * min_type is the minimum key type to truncate down to. If set to 0, this
3246 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3248 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3249 struct btrfs_root *root,
3250 struct inode *inode,
3251 u64 new_size, u32 min_type)
3253 struct btrfs_path *path;
3254 struct extent_buffer *leaf;
3255 struct btrfs_file_extent_item *fi;
3256 struct btrfs_key key;
3257 struct btrfs_key found_key;
3258 u64 extent_start = 0;
3259 u64 extent_num_bytes = 0;
3260 u64 extent_offset = 0;
3262 u64 mask = root->sectorsize - 1;
3263 u32 found_type = (u8)-1;
3266 int pending_del_nr = 0;
3267 int pending_del_slot = 0;
3268 int extent_type = -1;
3271 u64 ino = btrfs_ino(inode);
3273 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3275 path = btrfs_alloc_path();
3281 * We want to drop from the next block forward in case this new size is
3282 * not block aligned since we will be keeping the last block of the
3283 * extent just the way it is.
3285 if (root->ref_cows || root == root->fs_info->tree_root)
3286 btrfs_drop_extent_cache(inode, (new_size + mask) & (~mask), (u64)-1, 0);
3289 * This function is also used to drop the items in the log tree before
3290 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3291 * it is used to drop the loged items. So we shouldn't kill the delayed
3294 if (min_type == 0 && root == BTRFS_I(inode)->root)
3295 btrfs_kill_delayed_inode_items(inode);
3298 key.offset = (u64)-1;
3302 path->leave_spinning = 1;
3303 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3310 /* there are no items in the tree for us to truncate, we're
3313 if (path->slots[0] == 0)
3320 leaf = path->nodes[0];
3321 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3322 found_type = btrfs_key_type(&found_key);
3324 if (found_key.objectid != ino)
3327 if (found_type < min_type)
3330 item_end = found_key.offset;
3331 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3332 fi = btrfs_item_ptr(leaf, path->slots[0],
3333 struct btrfs_file_extent_item);
3334 extent_type = btrfs_file_extent_type(leaf, fi);
3335 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3337 btrfs_file_extent_num_bytes(leaf, fi);
3338 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3339 item_end += btrfs_file_extent_inline_len(leaf,
3344 if (found_type > min_type) {
3347 if (item_end < new_size)
3349 if (found_key.offset >= new_size)
3355 /* FIXME, shrink the extent if the ref count is only 1 */
3356 if (found_type != BTRFS_EXTENT_DATA_KEY)
3359 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3361 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3363 u64 orig_num_bytes =
3364 btrfs_file_extent_num_bytes(leaf, fi);
3365 extent_num_bytes = new_size -
3366 found_key.offset + root->sectorsize - 1;
3367 extent_num_bytes = extent_num_bytes &
3368 ~((u64)root->sectorsize - 1);
3369 btrfs_set_file_extent_num_bytes(leaf, fi,
3371 num_dec = (orig_num_bytes -
3373 if (root->ref_cows && extent_start != 0)
3374 inode_sub_bytes(inode, num_dec);
3375 btrfs_mark_buffer_dirty(leaf);
3378 btrfs_file_extent_disk_num_bytes(leaf,
3380 extent_offset = found_key.offset -
3381 btrfs_file_extent_offset(leaf, fi);
3383 /* FIXME blocksize != 4096 */
3384 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3385 if (extent_start != 0) {
3388 inode_sub_bytes(inode, num_dec);
3391 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3393 * we can't truncate inline items that have had
3397 btrfs_file_extent_compression(leaf, fi) == 0 &&
3398 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3399 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3400 u32 size = new_size - found_key.offset;
3402 if (root->ref_cows) {
3403 inode_sub_bytes(inode, item_end + 1 -
3407 btrfs_file_extent_calc_inline_size(size);
3408 btrfs_truncate_item(trans, root, path,
3410 } else if (root->ref_cows) {
3411 inode_sub_bytes(inode, item_end + 1 -
3417 if (!pending_del_nr) {
3418 /* no pending yet, add ourselves */
3419 pending_del_slot = path->slots[0];
3421 } else if (pending_del_nr &&
3422 path->slots[0] + 1 == pending_del_slot) {
3423 /* hop on the pending chunk */
3425 pending_del_slot = path->slots[0];
3432 if (found_extent && (root->ref_cows ||
3433 root == root->fs_info->tree_root)) {
3434 btrfs_set_path_blocking(path);
3435 ret = btrfs_free_extent(trans, root, extent_start,
3436 extent_num_bytes, 0,
3437 btrfs_header_owner(leaf),
3438 ino, extent_offset, 0);
3442 if (found_type == BTRFS_INODE_ITEM_KEY)
3445 if (path->slots[0] == 0 ||
3446 path->slots[0] != pending_del_slot) {
3447 if (pending_del_nr) {
3448 ret = btrfs_del_items(trans, root, path,
3452 btrfs_abort_transaction(trans,
3458 btrfs_release_path(path);
3465 if (pending_del_nr) {
3466 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3469 btrfs_abort_transaction(trans, root, ret);
3472 btrfs_free_path(path);
3477 * btrfs_truncate_page - read, zero a chunk and write a page
3478 * @inode - inode that we're zeroing
3479 * @from - the offset to start zeroing
3480 * @len - the length to zero, 0 to zero the entire range respective to the
3482 * @front - zero up to the offset instead of from the offset on
3484 * This will find the page for the "from" offset and cow the page and zero the
3485 * part we want to zero. This is used with truncate and hole punching.
3487 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
3490 struct address_space *mapping = inode->i_mapping;
3491 struct btrfs_root *root = BTRFS_I(inode)->root;
3492 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3493 struct btrfs_ordered_extent *ordered;
3494 struct extent_state *cached_state = NULL;
3496 u32 blocksize = root->sectorsize;
3497 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3498 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3500 gfp_t mask = btrfs_alloc_write_mask(mapping);
3505 if ((offset & (blocksize - 1)) == 0 &&
3506 (!len || ((len & (blocksize - 1)) == 0)))
3508 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3514 page = find_or_create_page(mapping, index, mask);
3516 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3520 page_start = page_offset(page);
3521 page_end = page_start + PAGE_CACHE_SIZE - 1;
3523 if (!PageUptodate(page)) {
3524 ret = btrfs_readpage(NULL, page);
3526 if (page->mapping != mapping) {
3528 page_cache_release(page);
3531 if (!PageUptodate(page)) {
3536 wait_on_page_writeback(page);
3538 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3539 set_page_extent_mapped(page);
3541 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3543 unlock_extent_cached(io_tree, page_start, page_end,
3544 &cached_state, GFP_NOFS);
3546 page_cache_release(page);
3547 btrfs_start_ordered_extent(inode, ordered, 1);
3548 btrfs_put_ordered_extent(ordered);
3552 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3553 EXTENT_DIRTY | EXTENT_DELALLOC |
3554 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
3555 0, 0, &cached_state, GFP_NOFS);
3557 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3560 unlock_extent_cached(io_tree, page_start, page_end,
3561 &cached_state, GFP_NOFS);
3566 if (offset != PAGE_CACHE_SIZE) {
3568 len = PAGE_CACHE_SIZE - offset;
3571 memset(kaddr, 0, offset);
3573 memset(kaddr + offset, 0, len);
3574 flush_dcache_page(page);
3577 ClearPageChecked(page);
3578 set_page_dirty(page);
3579 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3584 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3586 page_cache_release(page);
3592 * This function puts in dummy file extents for the area we're creating a hole
3593 * for. So if we are truncating this file to a larger size we need to insert
3594 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3595 * the range between oldsize and size
3597 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3599 struct btrfs_trans_handle *trans;
3600 struct btrfs_root *root = BTRFS_I(inode)->root;
3601 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3602 struct extent_map *em = NULL;
3603 struct extent_state *cached_state = NULL;
3604 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3605 u64 mask = root->sectorsize - 1;
3606 u64 hole_start = (oldsize + mask) & ~mask;
3607 u64 block_end = (size + mask) & ~mask;
3613 if (size <= hole_start)
3617 struct btrfs_ordered_extent *ordered;
3618 btrfs_wait_ordered_range(inode, hole_start,
3619 block_end - hole_start);
3620 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3622 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3625 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3626 &cached_state, GFP_NOFS);
3627 btrfs_put_ordered_extent(ordered);
3630 cur_offset = hole_start;
3632 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3633 block_end - cur_offset, 0);
3638 last_byte = min(extent_map_end(em), block_end);
3639 last_byte = (last_byte + mask) & ~mask;
3640 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3641 struct extent_map *hole_em;
3642 hole_size = last_byte - cur_offset;
3644 trans = btrfs_start_transaction(root, 3);
3645 if (IS_ERR(trans)) {
3646 err = PTR_ERR(trans);
3650 err = btrfs_drop_extents(trans, root, inode,
3652 cur_offset + hole_size, 1);
3654 btrfs_abort_transaction(trans, root, err);
3655 btrfs_end_transaction(trans, root);
3659 err = btrfs_insert_file_extent(trans, root,
3660 btrfs_ino(inode), cur_offset, 0,
3661 0, hole_size, 0, hole_size,
3664 btrfs_abort_transaction(trans, root, err);
3665 btrfs_end_transaction(trans, root);
3669 btrfs_drop_extent_cache(inode, cur_offset,
3670 cur_offset + hole_size - 1, 0);
3671 hole_em = alloc_extent_map();
3673 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3674 &BTRFS_I(inode)->runtime_flags);
3677 hole_em->start = cur_offset;
3678 hole_em->len = hole_size;
3679 hole_em->orig_start = cur_offset;
3681 hole_em->block_start = EXTENT_MAP_HOLE;
3682 hole_em->block_len = 0;
3683 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
3684 hole_em->compress_type = BTRFS_COMPRESS_NONE;
3685 hole_em->generation = trans->transid;
3688 write_lock(&em_tree->lock);
3689 err = add_extent_mapping(em_tree, hole_em);
3691 list_move(&hole_em->list,
3692 &em_tree->modified_extents);
3693 write_unlock(&em_tree->lock);
3696 btrfs_drop_extent_cache(inode, cur_offset,
3700 free_extent_map(hole_em);
3702 btrfs_update_inode(trans, root, inode);
3703 btrfs_end_transaction(trans, root);
3705 free_extent_map(em);
3707 cur_offset = last_byte;
3708 if (cur_offset >= block_end)
3712 free_extent_map(em);
3713 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3718 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3720 struct btrfs_root *root = BTRFS_I(inode)->root;
3721 struct btrfs_trans_handle *trans;
3722 loff_t oldsize = i_size_read(inode);
3725 if (newsize == oldsize)
3728 if (newsize > oldsize) {
3729 truncate_pagecache(inode, oldsize, newsize);
3730 ret = btrfs_cont_expand(inode, oldsize, newsize);
3734 trans = btrfs_start_transaction(root, 1);
3736 return PTR_ERR(trans);
3738 i_size_write(inode, newsize);
3739 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3740 ret = btrfs_update_inode(trans, root, inode);
3741 btrfs_end_transaction(trans, root);
3745 * We're truncating a file that used to have good data down to
3746 * zero. Make sure it gets into the ordered flush list so that
3747 * any new writes get down to disk quickly.
3750 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3751 &BTRFS_I(inode)->runtime_flags);
3753 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3754 truncate_setsize(inode, newsize);
3755 ret = btrfs_truncate(inode);
3761 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3763 struct inode *inode = dentry->d_inode;
3764 struct btrfs_root *root = BTRFS_I(inode)->root;
3767 if (btrfs_root_readonly(root))
3770 err = inode_change_ok(inode, attr);
3774 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3775 err = btrfs_setsize(inode, attr->ia_size);
3780 if (attr->ia_valid) {
3781 setattr_copy(inode, attr);
3782 inode_inc_iversion(inode);
3783 err = btrfs_dirty_inode(inode);
3785 if (!err && attr->ia_valid & ATTR_MODE)
3786 err = btrfs_acl_chmod(inode);
3792 void btrfs_evict_inode(struct inode *inode)
3794 struct btrfs_trans_handle *trans;
3795 struct btrfs_root *root = BTRFS_I(inode)->root;
3796 struct btrfs_block_rsv *rsv, *global_rsv;
3797 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3801 trace_btrfs_inode_evict(inode);
3803 truncate_inode_pages(&inode->i_data, 0);
3804 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3805 btrfs_is_free_space_inode(inode)))
3808 if (is_bad_inode(inode)) {
3809 btrfs_orphan_del(NULL, inode);
3812 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3813 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3815 if (root->fs_info->log_root_recovering) {
3816 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3817 &BTRFS_I(inode)->runtime_flags));
3821 if (inode->i_nlink > 0) {
3822 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3826 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3828 btrfs_orphan_del(NULL, inode);
3831 rsv->size = min_size;
3833 global_rsv = &root->fs_info->global_block_rsv;
3835 btrfs_i_size_write(inode, 0);
3838 * This is a bit simpler than btrfs_truncate since we've already
3839 * reserved our space for our orphan item in the unlink, so we just
3840 * need to reserve some slack space in case we add bytes and update
3841 * inode item when doing the truncate.
3844 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
3847 * Try and steal from the global reserve since we will
3848 * likely not use this space anyway, we want to try as
3849 * hard as possible to get this to work.
3852 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3855 printk(KERN_WARNING "Could not get space for a "
3856 "delete, will truncate on mount %d\n", ret);
3857 btrfs_orphan_del(NULL, inode);
3858 btrfs_free_block_rsv(root, rsv);
3862 trans = btrfs_start_transaction_noflush(root, 1);
3863 if (IS_ERR(trans)) {
3864 btrfs_orphan_del(NULL, inode);
3865 btrfs_free_block_rsv(root, rsv);
3869 trans->block_rsv = rsv;
3871 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3875 trans->block_rsv = &root->fs_info->trans_block_rsv;
3876 ret = btrfs_update_inode(trans, root, inode);
3879 nr = trans->blocks_used;
3880 btrfs_end_transaction(trans, root);
3882 btrfs_btree_balance_dirty(root, nr);
3885 btrfs_free_block_rsv(root, rsv);
3888 trans->block_rsv = root->orphan_block_rsv;
3889 ret = btrfs_orphan_del(trans, inode);
3893 trans->block_rsv = &root->fs_info->trans_block_rsv;
3894 if (!(root == root->fs_info->tree_root ||
3895 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3896 btrfs_return_ino(root, btrfs_ino(inode));
3898 nr = trans->blocks_used;
3899 btrfs_end_transaction(trans, root);
3900 btrfs_btree_balance_dirty(root, nr);
3907 * this returns the key found in the dir entry in the location pointer.
3908 * If no dir entries were found, location->objectid is 0.
3910 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3911 struct btrfs_key *location)
3913 const char *name = dentry->d_name.name;
3914 int namelen = dentry->d_name.len;
3915 struct btrfs_dir_item *di;
3916 struct btrfs_path *path;
3917 struct btrfs_root *root = BTRFS_I(dir)->root;
3920 path = btrfs_alloc_path();
3924 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3929 if (IS_ERR_OR_NULL(di))
3932 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3934 btrfs_free_path(path);
3937 location->objectid = 0;
3942 * when we hit a tree root in a directory, the btrfs part of the inode
3943 * needs to be changed to reflect the root directory of the tree root. This
3944 * is kind of like crossing a mount point.
3946 static int fixup_tree_root_location(struct btrfs_root *root,
3948 struct dentry *dentry,
3949 struct btrfs_key *location,
3950 struct btrfs_root **sub_root)
3952 struct btrfs_path *path;
3953 struct btrfs_root *new_root;
3954 struct btrfs_root_ref *ref;
3955 struct extent_buffer *leaf;
3959 path = btrfs_alloc_path();
3966 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3967 BTRFS_I(dir)->root->root_key.objectid,
3968 location->objectid);
3975 leaf = path->nodes[0];
3976 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3977 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3978 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3981 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3982 (unsigned long)(ref + 1),
3983 dentry->d_name.len);
3987 btrfs_release_path(path);
3989 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3990 if (IS_ERR(new_root)) {
3991 err = PTR_ERR(new_root);
3995 if (btrfs_root_refs(&new_root->root_item) == 0) {
4000 *sub_root = new_root;
4001 location->objectid = btrfs_root_dirid(&new_root->root_item);
4002 location->type = BTRFS_INODE_ITEM_KEY;
4003 location->offset = 0;
4006 btrfs_free_path(path);
4010 static void inode_tree_add(struct inode *inode)
4012 struct btrfs_root *root = BTRFS_I(inode)->root;
4013 struct btrfs_inode *entry;
4015 struct rb_node *parent;
4016 u64 ino = btrfs_ino(inode);
4018 p = &root->inode_tree.rb_node;
4021 if (inode_unhashed(inode))
4024 spin_lock(&root->inode_lock);
4027 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4029 if (ino < btrfs_ino(&entry->vfs_inode))
4030 p = &parent->rb_left;
4031 else if (ino > btrfs_ino(&entry->vfs_inode))
4032 p = &parent->rb_right;
4034 WARN_ON(!(entry->vfs_inode.i_state &
4035 (I_WILL_FREE | I_FREEING)));
4036 rb_erase(parent, &root->inode_tree);
4037 RB_CLEAR_NODE(parent);
4038 spin_unlock(&root->inode_lock);
4042 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4043 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4044 spin_unlock(&root->inode_lock);
4047 static void inode_tree_del(struct inode *inode)
4049 struct btrfs_root *root = BTRFS_I(inode)->root;
4052 spin_lock(&root->inode_lock);
4053 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4054 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4055 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4056 empty = RB_EMPTY_ROOT(&root->inode_tree);
4058 spin_unlock(&root->inode_lock);
4061 * Free space cache has inodes in the tree root, but the tree root has a
4062 * root_refs of 0, so this could end up dropping the tree root as a
4063 * snapshot, so we need the extra !root->fs_info->tree_root check to
4064 * make sure we don't drop it.
4066 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4067 root != root->fs_info->tree_root) {
4068 synchronize_srcu(&root->fs_info->subvol_srcu);
4069 spin_lock(&root->inode_lock);
4070 empty = RB_EMPTY_ROOT(&root->inode_tree);
4071 spin_unlock(&root->inode_lock);
4073 btrfs_add_dead_root(root);
4077 void btrfs_invalidate_inodes(struct btrfs_root *root)
4079 struct rb_node *node;
4080 struct rb_node *prev;
4081 struct btrfs_inode *entry;
4082 struct inode *inode;
4085 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4087 spin_lock(&root->inode_lock);
4089 node = root->inode_tree.rb_node;
4093 entry = rb_entry(node, struct btrfs_inode, rb_node);
4095 if (objectid < btrfs_ino(&entry->vfs_inode))
4096 node = node->rb_left;
4097 else if (objectid > btrfs_ino(&entry->vfs_inode))
4098 node = node->rb_right;
4104 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4105 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4109 prev = rb_next(prev);
4113 entry = rb_entry(node, struct btrfs_inode, rb_node);
4114 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4115 inode = igrab(&entry->vfs_inode);
4117 spin_unlock(&root->inode_lock);
4118 if (atomic_read(&inode->i_count) > 1)
4119 d_prune_aliases(inode);
4121 * btrfs_drop_inode will have it removed from
4122 * the inode cache when its usage count
4127 spin_lock(&root->inode_lock);
4131 if (cond_resched_lock(&root->inode_lock))
4134 node = rb_next(node);
4136 spin_unlock(&root->inode_lock);
4139 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4141 struct btrfs_iget_args *args = p;
4142 inode->i_ino = args->ino;
4143 BTRFS_I(inode)->root = args->root;
4147 static int btrfs_find_actor(struct inode *inode, void *opaque)
4149 struct btrfs_iget_args *args = opaque;
4150 return args->ino == btrfs_ino(inode) &&
4151 args->root == BTRFS_I(inode)->root;
4154 static struct inode *btrfs_iget_locked(struct super_block *s,
4156 struct btrfs_root *root)
4158 struct inode *inode;
4159 struct btrfs_iget_args args;
4160 args.ino = objectid;
4163 inode = iget5_locked(s, objectid, btrfs_find_actor,
4164 btrfs_init_locked_inode,
4169 /* Get an inode object given its location and corresponding root.
4170 * Returns in *is_new if the inode was read from disk
4172 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4173 struct btrfs_root *root, int *new)
4175 struct inode *inode;
4177 inode = btrfs_iget_locked(s, location->objectid, root);
4179 return ERR_PTR(-ENOMEM);
4181 if (inode->i_state & I_NEW) {
4182 BTRFS_I(inode)->root = root;
4183 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4184 btrfs_read_locked_inode(inode);
4185 if (!is_bad_inode(inode)) {
4186 inode_tree_add(inode);
4187 unlock_new_inode(inode);
4191 unlock_new_inode(inode);
4193 inode = ERR_PTR(-ESTALE);
4200 static struct inode *new_simple_dir(struct super_block *s,
4201 struct btrfs_key *key,
4202 struct btrfs_root *root)
4204 struct inode *inode = new_inode(s);
4207 return ERR_PTR(-ENOMEM);
4209 BTRFS_I(inode)->root = root;
4210 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4211 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4213 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4214 inode->i_op = &btrfs_dir_ro_inode_operations;
4215 inode->i_fop = &simple_dir_operations;
4216 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4217 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4222 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4224 struct inode *inode;
4225 struct btrfs_root *root = BTRFS_I(dir)->root;
4226 struct btrfs_root *sub_root = root;
4227 struct btrfs_key location;
4231 if (dentry->d_name.len > BTRFS_NAME_LEN)
4232 return ERR_PTR(-ENAMETOOLONG);
4234 if (unlikely(d_need_lookup(dentry))) {
4235 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4236 kfree(dentry->d_fsdata);
4237 dentry->d_fsdata = NULL;
4238 /* This thing is hashed, drop it for now */
4241 ret = btrfs_inode_by_name(dir, dentry, &location);
4245 return ERR_PTR(ret);
4247 if (location.objectid == 0)
4250 if (location.type == BTRFS_INODE_ITEM_KEY) {
4251 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4255 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4257 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4258 ret = fixup_tree_root_location(root, dir, dentry,
4259 &location, &sub_root);
4262 inode = ERR_PTR(ret);
4264 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4266 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4268 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4270 if (!IS_ERR(inode) && root != sub_root) {
4271 down_read(&root->fs_info->cleanup_work_sem);
4272 if (!(inode->i_sb->s_flags & MS_RDONLY))
4273 ret = btrfs_orphan_cleanup(sub_root);
4274 up_read(&root->fs_info->cleanup_work_sem);
4276 inode = ERR_PTR(ret);
4282 static int btrfs_dentry_delete(const struct dentry *dentry)
4284 struct btrfs_root *root;
4285 struct inode *inode = dentry->d_inode;
4287 if (!inode && !IS_ROOT(dentry))
4288 inode = dentry->d_parent->d_inode;
4291 root = BTRFS_I(inode)->root;
4292 if (btrfs_root_refs(&root->root_item) == 0)
4295 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4301 static void btrfs_dentry_release(struct dentry *dentry)
4303 if (dentry->d_fsdata)
4304 kfree(dentry->d_fsdata);
4307 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4312 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4313 if (unlikely(d_need_lookup(dentry))) {
4314 spin_lock(&dentry->d_lock);
4315 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4316 spin_unlock(&dentry->d_lock);
4321 unsigned char btrfs_filetype_table[] = {
4322 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4325 static int btrfs_real_readdir(struct file *filp, void *dirent,
4328 struct inode *inode = filp->f_dentry->d_inode;
4329 struct btrfs_root *root = BTRFS_I(inode)->root;
4330 struct btrfs_item *item;
4331 struct btrfs_dir_item *di;
4332 struct btrfs_key key;
4333 struct btrfs_key found_key;
4334 struct btrfs_path *path;
4335 struct list_head ins_list;
4336 struct list_head del_list;
4338 struct extent_buffer *leaf;
4340 unsigned char d_type;
4345 int key_type = BTRFS_DIR_INDEX_KEY;
4349 int is_curr = 0; /* filp->f_pos points to the current index? */
4351 /* FIXME, use a real flag for deciding about the key type */
4352 if (root->fs_info->tree_root == root)
4353 key_type = BTRFS_DIR_ITEM_KEY;
4355 /* special case for "." */
4356 if (filp->f_pos == 0) {
4357 over = filldir(dirent, ".", 1,
4358 filp->f_pos, btrfs_ino(inode), DT_DIR);
4363 /* special case for .., just use the back ref */
4364 if (filp->f_pos == 1) {
4365 u64 pino = parent_ino(filp->f_path.dentry);
4366 over = filldir(dirent, "..", 2,
4367 filp->f_pos, pino, DT_DIR);
4372 path = btrfs_alloc_path();
4378 if (key_type == BTRFS_DIR_INDEX_KEY) {
4379 INIT_LIST_HEAD(&ins_list);
4380 INIT_LIST_HEAD(&del_list);
4381 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4384 btrfs_set_key_type(&key, key_type);
4385 key.offset = filp->f_pos;
4386 key.objectid = btrfs_ino(inode);
4388 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4393 leaf = path->nodes[0];
4394 slot = path->slots[0];
4395 if (slot >= btrfs_header_nritems(leaf)) {
4396 ret = btrfs_next_leaf(root, path);
4404 item = btrfs_item_nr(leaf, slot);
4405 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4407 if (found_key.objectid != key.objectid)
4409 if (btrfs_key_type(&found_key) != key_type)
4411 if (found_key.offset < filp->f_pos)
4413 if (key_type == BTRFS_DIR_INDEX_KEY &&
4414 btrfs_should_delete_dir_index(&del_list,
4418 filp->f_pos = found_key.offset;
4421 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4423 di_total = btrfs_item_size(leaf, item);
4425 while (di_cur < di_total) {
4426 struct btrfs_key location;
4428 if (verify_dir_item(root, leaf, di))
4431 name_len = btrfs_dir_name_len(leaf, di);
4432 if (name_len <= sizeof(tmp_name)) {
4433 name_ptr = tmp_name;
4435 name_ptr = kmalloc(name_len, GFP_NOFS);
4441 read_extent_buffer(leaf, name_ptr,
4442 (unsigned long)(di + 1), name_len);
4444 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4445 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4448 /* is this a reference to our own snapshot? If so
4451 * In contrast to old kernels, we insert the snapshot's
4452 * dir item and dir index after it has been created, so
4453 * we won't find a reference to our own snapshot. We
4454 * still keep the following code for backward
4457 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4458 location.objectid == root->root_key.objectid) {
4462 over = filldir(dirent, name_ptr, name_len,
4463 found_key.offset, location.objectid,
4467 if (name_ptr != tmp_name)
4472 di_len = btrfs_dir_name_len(leaf, di) +
4473 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4475 di = (struct btrfs_dir_item *)((char *)di + di_len);
4481 if (key_type == BTRFS_DIR_INDEX_KEY) {
4484 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4490 /* Reached end of directory/root. Bump pos past the last item. */
4491 if (key_type == BTRFS_DIR_INDEX_KEY)
4493 * 32-bit glibc will use getdents64, but then strtol -
4494 * so the last number we can serve is this.
4496 filp->f_pos = 0x7fffffff;
4502 if (key_type == BTRFS_DIR_INDEX_KEY)
4503 btrfs_put_delayed_items(&ins_list, &del_list);
4504 btrfs_free_path(path);
4508 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4510 struct btrfs_root *root = BTRFS_I(inode)->root;
4511 struct btrfs_trans_handle *trans;
4513 bool nolock = false;
4515 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4518 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
4521 if (wbc->sync_mode == WB_SYNC_ALL) {
4523 trans = btrfs_join_transaction_nolock(root);
4525 trans = btrfs_join_transaction(root);
4527 return PTR_ERR(trans);
4529 ret = btrfs_end_transaction_nolock(trans, root);
4531 ret = btrfs_commit_transaction(trans, root);
4537 * This is somewhat expensive, updating the tree every time the
4538 * inode changes. But, it is most likely to find the inode in cache.
4539 * FIXME, needs more benchmarking...there are no reasons other than performance
4540 * to keep or drop this code.
4542 int btrfs_dirty_inode(struct inode *inode)
4544 struct btrfs_root *root = BTRFS_I(inode)->root;
4545 struct btrfs_trans_handle *trans;
4548 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4551 trans = btrfs_join_transaction(root);
4553 return PTR_ERR(trans);
4555 ret = btrfs_update_inode(trans, root, inode);
4556 if (ret && ret == -ENOSPC) {
4557 /* whoops, lets try again with the full transaction */
4558 btrfs_end_transaction(trans, root);
4559 trans = btrfs_start_transaction(root, 1);
4561 return PTR_ERR(trans);
4563 ret = btrfs_update_inode(trans, root, inode);
4565 btrfs_end_transaction(trans, root);
4566 if (BTRFS_I(inode)->delayed_node)
4567 btrfs_balance_delayed_items(root);
4573 * This is a copy of file_update_time. We need this so we can return error on
4574 * ENOSPC for updating the inode in the case of file write and mmap writes.
4576 static int btrfs_update_time(struct inode *inode, struct timespec *now,
4579 struct btrfs_root *root = BTRFS_I(inode)->root;
4581 if (btrfs_root_readonly(root))
4584 if (flags & S_VERSION)
4585 inode_inc_iversion(inode);
4586 if (flags & S_CTIME)
4587 inode->i_ctime = *now;
4588 if (flags & S_MTIME)
4589 inode->i_mtime = *now;
4590 if (flags & S_ATIME)
4591 inode->i_atime = *now;
4592 return btrfs_dirty_inode(inode);
4596 * find the highest existing sequence number in a directory
4597 * and then set the in-memory index_cnt variable to reflect
4598 * free sequence numbers
4600 static int btrfs_set_inode_index_count(struct inode *inode)
4602 struct btrfs_root *root = BTRFS_I(inode)->root;
4603 struct btrfs_key key, found_key;
4604 struct btrfs_path *path;
4605 struct extent_buffer *leaf;
4608 key.objectid = btrfs_ino(inode);
4609 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4610 key.offset = (u64)-1;
4612 path = btrfs_alloc_path();
4616 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4619 /* FIXME: we should be able to handle this */
4625 * MAGIC NUMBER EXPLANATION:
4626 * since we search a directory based on f_pos we have to start at 2
4627 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4628 * else has to start at 2
4630 if (path->slots[0] == 0) {
4631 BTRFS_I(inode)->index_cnt = 2;
4637 leaf = path->nodes[0];
4638 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4640 if (found_key.objectid != btrfs_ino(inode) ||
4641 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4642 BTRFS_I(inode)->index_cnt = 2;
4646 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4648 btrfs_free_path(path);
4653 * helper to find a free sequence number in a given directory. This current
4654 * code is very simple, later versions will do smarter things in the btree
4656 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4660 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4661 ret = btrfs_inode_delayed_dir_index_count(dir);
4663 ret = btrfs_set_inode_index_count(dir);
4669 *index = BTRFS_I(dir)->index_cnt;
4670 BTRFS_I(dir)->index_cnt++;
4675 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4676 struct btrfs_root *root,
4678 const char *name, int name_len,
4679 u64 ref_objectid, u64 objectid,
4680 umode_t mode, u64 *index)
4682 struct inode *inode;
4683 struct btrfs_inode_item *inode_item;
4684 struct btrfs_key *location;
4685 struct btrfs_path *path;
4686 struct btrfs_inode_ref *ref;
4687 struct btrfs_key key[2];
4693 path = btrfs_alloc_path();
4695 return ERR_PTR(-ENOMEM);
4697 inode = new_inode(root->fs_info->sb);
4699 btrfs_free_path(path);
4700 return ERR_PTR(-ENOMEM);
4704 * we have to initialize this early, so we can reclaim the inode
4705 * number if we fail afterwards in this function.
4707 inode->i_ino = objectid;
4710 trace_btrfs_inode_request(dir);
4712 ret = btrfs_set_inode_index(dir, index);
4714 btrfs_free_path(path);
4716 return ERR_PTR(ret);
4720 * index_cnt is ignored for everything but a dir,
4721 * btrfs_get_inode_index_count has an explanation for the magic
4724 BTRFS_I(inode)->index_cnt = 2;
4725 BTRFS_I(inode)->root = root;
4726 BTRFS_I(inode)->generation = trans->transid;
4727 inode->i_generation = BTRFS_I(inode)->generation;
4730 * We could have gotten an inode number from somebody who was fsynced
4731 * and then removed in this same transaction, so let's just set full
4732 * sync since it will be a full sync anyway and this will blow away the
4733 * old info in the log.
4735 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
4742 key[0].objectid = objectid;
4743 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4746 key[1].objectid = objectid;
4747 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4748 key[1].offset = ref_objectid;
4750 sizes[0] = sizeof(struct btrfs_inode_item);
4751 sizes[1] = name_len + sizeof(*ref);
4753 path->leave_spinning = 1;
4754 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4758 inode_init_owner(inode, dir, mode);
4759 inode_set_bytes(inode, 0);
4760 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4761 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4762 struct btrfs_inode_item);
4763 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
4764 sizeof(*inode_item));
4765 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4767 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4768 struct btrfs_inode_ref);
4769 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4770 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4771 ptr = (unsigned long)(ref + 1);
4772 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4774 btrfs_mark_buffer_dirty(path->nodes[0]);
4775 btrfs_free_path(path);
4777 location = &BTRFS_I(inode)->location;
4778 location->objectid = objectid;
4779 location->offset = 0;
4780 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4782 btrfs_inherit_iflags(inode, dir);
4784 if (S_ISREG(mode)) {
4785 if (btrfs_test_opt(root, NODATASUM))
4786 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4787 if (btrfs_test_opt(root, NODATACOW) ||
4788 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4789 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4792 insert_inode_hash(inode);
4793 inode_tree_add(inode);
4795 trace_btrfs_inode_new(inode);
4796 btrfs_set_inode_last_trans(trans, inode);
4798 btrfs_update_root_times(trans, root);
4803 BTRFS_I(dir)->index_cnt--;
4804 btrfs_free_path(path);
4806 return ERR_PTR(ret);
4809 static inline u8 btrfs_inode_type(struct inode *inode)
4811 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4815 * utility function to add 'inode' into 'parent_inode' with
4816 * a give name and a given sequence number.
4817 * if 'add_backref' is true, also insert a backref from the
4818 * inode to the parent directory.
4820 int btrfs_add_link(struct btrfs_trans_handle *trans,
4821 struct inode *parent_inode, struct inode *inode,
4822 const char *name, int name_len, int add_backref, u64 index)
4825 struct btrfs_key key;
4826 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4827 u64 ino = btrfs_ino(inode);
4828 u64 parent_ino = btrfs_ino(parent_inode);
4830 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4831 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4834 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4838 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4839 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4840 key.objectid, root->root_key.objectid,
4841 parent_ino, index, name, name_len);
4842 } else if (add_backref) {
4843 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4847 /* Nothing to clean up yet */
4851 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4853 btrfs_inode_type(inode), index);
4857 btrfs_abort_transaction(trans, root, ret);
4861 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4863 inode_inc_iversion(parent_inode);
4864 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4865 ret = btrfs_update_inode(trans, root, parent_inode);
4867 btrfs_abort_transaction(trans, root, ret);
4871 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4874 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4875 key.objectid, root->root_key.objectid,
4876 parent_ino, &local_index, name, name_len);
4878 } else if (add_backref) {
4882 err = btrfs_del_inode_ref(trans, root, name, name_len,
4883 ino, parent_ino, &local_index);
4888 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4889 struct inode *dir, struct dentry *dentry,
4890 struct inode *inode, int backref, u64 index)
4892 int err = btrfs_add_link(trans, dir, inode,
4893 dentry->d_name.name, dentry->d_name.len,
4900 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4901 umode_t mode, dev_t rdev)
4903 struct btrfs_trans_handle *trans;
4904 struct btrfs_root *root = BTRFS_I(dir)->root;
4905 struct inode *inode = NULL;
4909 unsigned long nr = 0;
4912 if (!new_valid_dev(rdev))
4916 * 2 for inode item and ref
4918 * 1 for xattr if selinux is on
4920 trans = btrfs_start_transaction(root, 5);
4922 return PTR_ERR(trans);
4924 err = btrfs_find_free_ino(root, &objectid);
4928 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4929 dentry->d_name.len, btrfs_ino(dir), objectid,
4931 if (IS_ERR(inode)) {
4932 err = PTR_ERR(inode);
4936 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4943 * If the active LSM wants to access the inode during
4944 * d_instantiate it needs these. Smack checks to see
4945 * if the filesystem supports xattrs by looking at the
4949 inode->i_op = &btrfs_special_inode_operations;
4950 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4954 init_special_inode(inode, inode->i_mode, rdev);
4955 btrfs_update_inode(trans, root, inode);
4956 d_instantiate(dentry, inode);
4959 nr = trans->blocks_used;
4960 btrfs_end_transaction(trans, root);
4961 btrfs_btree_balance_dirty(root, nr);
4963 inode_dec_link_count(inode);
4969 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4970 umode_t mode, bool excl)
4972 struct btrfs_trans_handle *trans;
4973 struct btrfs_root *root = BTRFS_I(dir)->root;
4974 struct inode *inode = NULL;
4977 unsigned long nr = 0;
4982 * 2 for inode item and ref
4984 * 1 for xattr if selinux is on
4986 trans = btrfs_start_transaction(root, 5);
4988 return PTR_ERR(trans);
4990 err = btrfs_find_free_ino(root, &objectid);
4994 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4995 dentry->d_name.len, btrfs_ino(dir), objectid,
4997 if (IS_ERR(inode)) {
4998 err = PTR_ERR(inode);
5002 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5009 * If the active LSM wants to access the inode during
5010 * d_instantiate it needs these. Smack checks to see
5011 * if the filesystem supports xattrs by looking at the
5014 inode->i_fop = &btrfs_file_operations;
5015 inode->i_op = &btrfs_file_inode_operations;
5017 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5021 inode->i_mapping->a_ops = &btrfs_aops;
5022 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5023 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5024 d_instantiate(dentry, inode);
5027 nr = trans->blocks_used;
5028 btrfs_end_transaction(trans, root);
5030 inode_dec_link_count(inode);
5033 btrfs_btree_balance_dirty(root, nr);
5037 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5038 struct dentry *dentry)
5040 struct btrfs_trans_handle *trans;
5041 struct btrfs_root *root = BTRFS_I(dir)->root;
5042 struct inode *inode = old_dentry->d_inode;
5044 unsigned long nr = 0;
5048 /* do not allow sys_link's with other subvols of the same device */
5049 if (root->objectid != BTRFS_I(inode)->root->objectid)
5052 if (inode->i_nlink == ~0U)
5055 err = btrfs_set_inode_index(dir, &index);
5060 * 2 items for inode and inode ref
5061 * 2 items for dir items
5062 * 1 item for parent inode
5064 trans = btrfs_start_transaction(root, 5);
5065 if (IS_ERR(trans)) {
5066 err = PTR_ERR(trans);
5070 btrfs_inc_nlink(inode);
5071 inode_inc_iversion(inode);
5072 inode->i_ctime = CURRENT_TIME;
5075 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5080 struct dentry *parent = dentry->d_parent;
5081 err = btrfs_update_inode(trans, root, inode);
5084 d_instantiate(dentry, inode);
5085 btrfs_log_new_name(trans, inode, NULL, parent);
5088 nr = trans->blocks_used;
5089 btrfs_end_transaction(trans, root);
5092 inode_dec_link_count(inode);
5095 btrfs_btree_balance_dirty(root, nr);
5099 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5101 struct inode *inode = NULL;
5102 struct btrfs_trans_handle *trans;
5103 struct btrfs_root *root = BTRFS_I(dir)->root;
5105 int drop_on_err = 0;
5108 unsigned long nr = 1;
5111 * 2 items for inode and ref
5112 * 2 items for dir items
5113 * 1 for xattr if selinux is on
5115 trans = btrfs_start_transaction(root, 5);
5117 return PTR_ERR(trans);
5119 err = btrfs_find_free_ino(root, &objectid);
5123 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5124 dentry->d_name.len, btrfs_ino(dir), objectid,
5125 S_IFDIR | mode, &index);
5126 if (IS_ERR(inode)) {
5127 err = PTR_ERR(inode);
5133 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5137 inode->i_op = &btrfs_dir_inode_operations;
5138 inode->i_fop = &btrfs_dir_file_operations;
5140 btrfs_i_size_write(inode, 0);
5141 err = btrfs_update_inode(trans, root, inode);
5145 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5146 dentry->d_name.len, 0, index);
5150 d_instantiate(dentry, inode);
5154 nr = trans->blocks_used;
5155 btrfs_end_transaction(trans, root);
5158 btrfs_btree_balance_dirty(root, nr);
5162 /* helper for btfs_get_extent. Given an existing extent in the tree,
5163 * and an extent that you want to insert, deal with overlap and insert
5164 * the new extent into the tree.
5166 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5167 struct extent_map *existing,
5168 struct extent_map *em,
5169 u64 map_start, u64 map_len)
5173 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5174 start_diff = map_start - em->start;
5175 em->start = map_start;
5177 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5178 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5179 em->block_start += start_diff;
5180 em->block_len -= start_diff;
5182 return add_extent_mapping(em_tree, em);
5185 static noinline int uncompress_inline(struct btrfs_path *path,
5186 struct inode *inode, struct page *page,
5187 size_t pg_offset, u64 extent_offset,
5188 struct btrfs_file_extent_item *item)
5191 struct extent_buffer *leaf = path->nodes[0];
5194 unsigned long inline_size;
5198 WARN_ON(pg_offset != 0);
5199 compress_type = btrfs_file_extent_compression(leaf, item);
5200 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5201 inline_size = btrfs_file_extent_inline_item_len(leaf,
5202 btrfs_item_nr(leaf, path->slots[0]));
5203 tmp = kmalloc(inline_size, GFP_NOFS);
5206 ptr = btrfs_file_extent_inline_start(item);
5208 read_extent_buffer(leaf, tmp, ptr, inline_size);
5210 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5211 ret = btrfs_decompress(compress_type, tmp, page,
5212 extent_offset, inline_size, max_size);
5214 char *kaddr = kmap_atomic(page);
5215 unsigned long copy_size = min_t(u64,
5216 PAGE_CACHE_SIZE - pg_offset,
5217 max_size - extent_offset);
5218 memset(kaddr + pg_offset, 0, copy_size);
5219 kunmap_atomic(kaddr);
5226 * a bit scary, this does extent mapping from logical file offset to the disk.
5227 * the ugly parts come from merging extents from the disk with the in-ram
5228 * representation. This gets more complex because of the data=ordered code,
5229 * where the in-ram extents might be locked pending data=ordered completion.
5231 * This also copies inline extents directly into the page.
5234 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5235 size_t pg_offset, u64 start, u64 len,
5241 u64 extent_start = 0;
5243 u64 objectid = btrfs_ino(inode);
5245 struct btrfs_path *path = NULL;
5246 struct btrfs_root *root = BTRFS_I(inode)->root;
5247 struct btrfs_file_extent_item *item;
5248 struct extent_buffer *leaf;
5249 struct btrfs_key found_key;
5250 struct extent_map *em = NULL;
5251 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5252 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5253 struct btrfs_trans_handle *trans = NULL;
5257 read_lock(&em_tree->lock);
5258 em = lookup_extent_mapping(em_tree, start, len);
5260 em->bdev = root->fs_info->fs_devices->latest_bdev;
5261 read_unlock(&em_tree->lock);
5264 if (em->start > start || em->start + em->len <= start)
5265 free_extent_map(em);
5266 else if (em->block_start == EXTENT_MAP_INLINE && page)
5267 free_extent_map(em);
5271 em = alloc_extent_map();
5276 em->bdev = root->fs_info->fs_devices->latest_bdev;
5277 em->start = EXTENT_MAP_HOLE;
5278 em->orig_start = EXTENT_MAP_HOLE;
5280 em->block_len = (u64)-1;
5283 path = btrfs_alloc_path();
5289 * Chances are we'll be called again, so go ahead and do
5295 ret = btrfs_lookup_file_extent(trans, root, path,
5296 objectid, start, trans != NULL);
5303 if (path->slots[0] == 0)
5308 leaf = path->nodes[0];
5309 item = btrfs_item_ptr(leaf, path->slots[0],
5310 struct btrfs_file_extent_item);
5311 /* are we inside the extent that was found? */
5312 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5313 found_type = btrfs_key_type(&found_key);
5314 if (found_key.objectid != objectid ||
5315 found_type != BTRFS_EXTENT_DATA_KEY) {
5319 found_type = btrfs_file_extent_type(leaf, item);
5320 extent_start = found_key.offset;
5321 compress_type = btrfs_file_extent_compression(leaf, item);
5322 if (found_type == BTRFS_FILE_EXTENT_REG ||
5323 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5324 extent_end = extent_start +
5325 btrfs_file_extent_num_bytes(leaf, item);
5326 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5328 size = btrfs_file_extent_inline_len(leaf, item);
5329 extent_end = (extent_start + size + root->sectorsize - 1) &
5330 ~((u64)root->sectorsize - 1);
5333 if (start >= extent_end) {
5335 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5336 ret = btrfs_next_leaf(root, path);
5343 leaf = path->nodes[0];
5345 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5346 if (found_key.objectid != objectid ||
5347 found_key.type != BTRFS_EXTENT_DATA_KEY)
5349 if (start + len <= found_key.offset)
5352 em->len = found_key.offset - start;
5356 if (found_type == BTRFS_FILE_EXTENT_REG ||
5357 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5358 em->start = extent_start;
5359 em->len = extent_end - extent_start;
5360 em->orig_start = extent_start -
5361 btrfs_file_extent_offset(leaf, item);
5362 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5364 em->block_start = EXTENT_MAP_HOLE;
5367 if (compress_type != BTRFS_COMPRESS_NONE) {
5368 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5369 em->compress_type = compress_type;
5370 em->block_start = bytenr;
5371 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5374 bytenr += btrfs_file_extent_offset(leaf, item);
5375 em->block_start = bytenr;
5376 em->block_len = em->len;
5377 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5378 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5381 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5385 size_t extent_offset;
5388 em->block_start = EXTENT_MAP_INLINE;
5389 if (!page || create) {
5390 em->start = extent_start;
5391 em->len = extent_end - extent_start;
5395 size = btrfs_file_extent_inline_len(leaf, item);
5396 extent_offset = page_offset(page) + pg_offset - extent_start;
5397 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5398 size - extent_offset);
5399 em->start = extent_start + extent_offset;
5400 em->len = (copy_size + root->sectorsize - 1) &
5401 ~((u64)root->sectorsize - 1);
5402 em->orig_start = EXTENT_MAP_INLINE;
5403 if (compress_type) {
5404 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5405 em->compress_type = compress_type;
5407 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5408 if (create == 0 && !PageUptodate(page)) {
5409 if (btrfs_file_extent_compression(leaf, item) !=
5410 BTRFS_COMPRESS_NONE) {
5411 ret = uncompress_inline(path, inode, page,
5413 extent_offset, item);
5414 BUG_ON(ret); /* -ENOMEM */
5417 read_extent_buffer(leaf, map + pg_offset, ptr,
5419 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5420 memset(map + pg_offset + copy_size, 0,
5421 PAGE_CACHE_SIZE - pg_offset -
5426 flush_dcache_page(page);
5427 } else if (create && PageUptodate(page)) {
5431 free_extent_map(em);
5434 btrfs_release_path(path);
5435 trans = btrfs_join_transaction(root);
5438 return ERR_CAST(trans);
5442 write_extent_buffer(leaf, map + pg_offset, ptr,
5445 btrfs_mark_buffer_dirty(leaf);
5447 set_extent_uptodate(io_tree, em->start,
5448 extent_map_end(em) - 1, NULL, GFP_NOFS);
5451 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5458 em->block_start = EXTENT_MAP_HOLE;
5459 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5461 btrfs_release_path(path);
5462 if (em->start > start || extent_map_end(em) <= start) {
5463 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5464 "[%llu %llu]\n", (unsigned long long)em->start,
5465 (unsigned long long)em->len,
5466 (unsigned long long)start,
5467 (unsigned long long)len);
5473 write_lock(&em_tree->lock);
5474 ret = add_extent_mapping(em_tree, em);
5475 /* it is possible that someone inserted the extent into the tree
5476 * while we had the lock dropped. It is also possible that
5477 * an overlapping map exists in the tree
5479 if (ret == -EEXIST) {
5480 struct extent_map *existing;
5484 existing = lookup_extent_mapping(em_tree, start, len);
5485 if (existing && (existing->start > start ||
5486 existing->start + existing->len <= start)) {
5487 free_extent_map(existing);
5491 existing = lookup_extent_mapping(em_tree, em->start,
5494 err = merge_extent_mapping(em_tree, existing,
5497 free_extent_map(existing);
5499 free_extent_map(em);
5504 free_extent_map(em);
5508 free_extent_map(em);
5513 write_unlock(&em_tree->lock);
5516 trace_btrfs_get_extent(root, em);
5519 btrfs_free_path(path);
5521 ret = btrfs_end_transaction(trans, root);
5526 free_extent_map(em);
5527 return ERR_PTR(err);
5529 BUG_ON(!em); /* Error is always set */
5533 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5534 size_t pg_offset, u64 start, u64 len,
5537 struct extent_map *em;
5538 struct extent_map *hole_em = NULL;
5539 u64 range_start = start;
5545 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5550 * if our em maps to a hole, there might
5551 * actually be delalloc bytes behind it
5553 if (em->block_start != EXTENT_MAP_HOLE)
5559 /* check to see if we've wrapped (len == -1 or similar) */
5568 /* ok, we didn't find anything, lets look for delalloc */
5569 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5570 end, len, EXTENT_DELALLOC, 1);
5571 found_end = range_start + found;
5572 if (found_end < range_start)
5573 found_end = (u64)-1;
5576 * we didn't find anything useful, return
5577 * the original results from get_extent()
5579 if (range_start > end || found_end <= start) {
5585 /* adjust the range_start to make sure it doesn't
5586 * go backwards from the start they passed in
5588 range_start = max(start,range_start);
5589 found = found_end - range_start;
5592 u64 hole_start = start;
5595 em = alloc_extent_map();
5601 * when btrfs_get_extent can't find anything it
5602 * returns one huge hole
5604 * make sure what it found really fits our range, and
5605 * adjust to make sure it is based on the start from
5609 u64 calc_end = extent_map_end(hole_em);
5611 if (calc_end <= start || (hole_em->start > end)) {
5612 free_extent_map(hole_em);
5615 hole_start = max(hole_em->start, start);
5616 hole_len = calc_end - hole_start;
5620 if (hole_em && range_start > hole_start) {
5621 /* our hole starts before our delalloc, so we
5622 * have to return just the parts of the hole
5623 * that go until the delalloc starts
5625 em->len = min(hole_len,
5626 range_start - hole_start);
5627 em->start = hole_start;
5628 em->orig_start = hole_start;
5630 * don't adjust block start at all,
5631 * it is fixed at EXTENT_MAP_HOLE
5633 em->block_start = hole_em->block_start;
5634 em->block_len = hole_len;
5636 em->start = range_start;
5638 em->orig_start = range_start;
5639 em->block_start = EXTENT_MAP_DELALLOC;
5640 em->block_len = found;
5642 } else if (hole_em) {
5647 free_extent_map(hole_em);
5649 free_extent_map(em);
5650 return ERR_PTR(err);
5655 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5656 struct extent_map *em,
5659 struct btrfs_root *root = BTRFS_I(inode)->root;
5660 struct btrfs_trans_handle *trans;
5661 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5662 struct btrfs_key ins;
5665 bool insert = false;
5668 * Ok if the extent map we looked up is a hole and is for the exact
5669 * range we want, there is no reason to allocate a new one, however if
5670 * it is not right then we need to free this one and drop the cache for
5673 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5675 free_extent_map(em);
5678 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5681 trans = btrfs_join_transaction(root);
5683 return ERR_CAST(trans);
5685 if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5686 btrfs_add_inode_defrag(trans, inode);
5688 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5690 alloc_hint = get_extent_allocation_hint(inode, start, len);
5691 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5692 alloc_hint, &ins, 1);
5699 em = alloc_extent_map();
5701 em = ERR_PTR(-ENOMEM);
5707 em->orig_start = em->start;
5708 em->len = ins.offset;
5710 em->block_start = ins.objectid;
5711 em->block_len = ins.offset;
5712 em->bdev = root->fs_info->fs_devices->latest_bdev;
5715 * We need to do this because if we're using the original em we searched
5716 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5719 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5722 write_lock(&em_tree->lock);
5723 ret = add_extent_mapping(em_tree, em);
5724 write_unlock(&em_tree->lock);
5727 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5730 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5731 ins.offset, ins.offset, 0);
5733 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5737 btrfs_end_transaction(trans, root);
5742 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5743 * block must be cow'd
5745 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5746 struct inode *inode, u64 offset, u64 len)
5748 struct btrfs_path *path;
5750 struct extent_buffer *leaf;
5751 struct btrfs_root *root = BTRFS_I(inode)->root;
5752 struct btrfs_file_extent_item *fi;
5753 struct btrfs_key key;
5761 path = btrfs_alloc_path();
5765 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5770 slot = path->slots[0];
5773 /* can't find the item, must cow */
5780 leaf = path->nodes[0];
5781 btrfs_item_key_to_cpu(leaf, &key, slot);
5782 if (key.objectid != btrfs_ino(inode) ||
5783 key.type != BTRFS_EXTENT_DATA_KEY) {
5784 /* not our file or wrong item type, must cow */
5788 if (key.offset > offset) {
5789 /* Wrong offset, must cow */
5793 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5794 found_type = btrfs_file_extent_type(leaf, fi);
5795 if (found_type != BTRFS_FILE_EXTENT_REG &&
5796 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5797 /* not a regular extent, must cow */
5800 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5801 backref_offset = btrfs_file_extent_offset(leaf, fi);
5803 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5804 if (extent_end < offset + len) {
5805 /* extent doesn't include our full range, must cow */
5809 if (btrfs_extent_readonly(root, disk_bytenr))
5813 * look for other files referencing this extent, if we
5814 * find any we must cow
5816 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5817 key.offset - backref_offset, disk_bytenr))
5821 * adjust disk_bytenr and num_bytes to cover just the bytes
5822 * in this extent we are about to write. If there
5823 * are any csums in that range we have to cow in order
5824 * to keep the csums correct
5826 disk_bytenr += backref_offset;
5827 disk_bytenr += offset - key.offset;
5828 num_bytes = min(offset + len, extent_end) - offset;
5829 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5832 * all of the above have passed, it is safe to overwrite this extent
5837 btrfs_free_path(path);
5841 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
5842 struct extent_state **cached_state, int writing)
5844 struct btrfs_ordered_extent *ordered;
5848 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5851 * We're concerned with the entire range that we're going to be
5852 * doing DIO to, so we need to make sure theres no ordered
5853 * extents in this range.
5855 ordered = btrfs_lookup_ordered_range(inode, lockstart,
5856 lockend - lockstart + 1);
5859 * We need to make sure there are no buffered pages in this
5860 * range either, we could have raced between the invalidate in
5861 * generic_file_direct_write and locking the extent. The
5862 * invalidate needs to happen so that reads after a write do not
5865 if (!ordered && (!writing ||
5866 !test_range_bit(&BTRFS_I(inode)->io_tree,
5867 lockstart, lockend, EXTENT_UPTODATE, 0,
5871 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5872 cached_state, GFP_NOFS);
5875 btrfs_start_ordered_extent(inode, ordered, 1);
5876 btrfs_put_ordered_extent(ordered);
5878 /* Screw you mmap */
5879 ret = filemap_write_and_wait_range(inode->i_mapping,
5886 * If we found a page that couldn't be invalidated just
5887 * fall back to buffered.
5889 ret = invalidate_inode_pages2_range(inode->i_mapping,
5890 lockstart >> PAGE_CACHE_SHIFT,
5891 lockend >> PAGE_CACHE_SHIFT);
5902 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
5903 u64 len, u64 orig_start,
5904 u64 block_start, u64 block_len,
5907 struct extent_map_tree *em_tree;
5908 struct extent_map *em;
5909 struct btrfs_root *root = BTRFS_I(inode)->root;
5912 em_tree = &BTRFS_I(inode)->extent_tree;
5913 em = alloc_extent_map();
5915 return ERR_PTR(-ENOMEM);
5918 em->orig_start = orig_start;
5920 em->block_len = block_len;
5921 em->block_start = block_start;
5922 em->bdev = root->fs_info->fs_devices->latest_bdev;
5923 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5924 if (type == BTRFS_ORDERED_PREALLOC)
5925 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5928 btrfs_drop_extent_cache(inode, em->start,
5929 em->start + em->len - 1, 0);
5930 write_lock(&em_tree->lock);
5931 ret = add_extent_mapping(em_tree, em);
5932 write_unlock(&em_tree->lock);
5933 } while (ret == -EEXIST);
5936 free_extent_map(em);
5937 return ERR_PTR(ret);
5944 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5945 struct buffer_head *bh_result, int create)
5947 struct extent_map *em;
5948 struct btrfs_root *root = BTRFS_I(inode)->root;
5949 struct extent_state *cached_state = NULL;
5950 u64 start = iblock << inode->i_blkbits;
5951 u64 lockstart, lockend;
5952 u64 len = bh_result->b_size;
5953 struct btrfs_trans_handle *trans;
5954 int unlock_bits = EXTENT_LOCKED;
5958 ret = btrfs_delalloc_reserve_space(inode, len);
5961 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
5963 len = min_t(u64, len, root->sectorsize);
5967 lockend = start + len - 1;
5970 * If this errors out it's because we couldn't invalidate pagecache for
5971 * this range and we need to fallback to buffered.
5973 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
5977 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
5978 lockend, EXTENT_DELALLOC, NULL,
5979 &cached_state, GFP_NOFS);
5984 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5991 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5992 * io. INLINE is special, and we could probably kludge it in here, but
5993 * it's still buffered so for safety lets just fall back to the generic
5996 * For COMPRESSED we _have_ to read the entire extent in so we can
5997 * decompress it, so there will be buffering required no matter what we
5998 * do, so go ahead and fallback to buffered.
6000 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6001 * to buffered IO. Don't blame me, this is the price we pay for using
6004 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6005 em->block_start == EXTENT_MAP_INLINE) {
6006 free_extent_map(em);
6011 /* Just a good old fashioned hole, return */
6012 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6013 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6014 free_extent_map(em);
6020 * We don't allocate a new extent in the following cases
6022 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6024 * 2) The extent is marked as PREALLOC. We're good to go here and can
6025 * just use the extent.
6029 len = min(len, em->len - (start - em->start));
6030 lockstart = start + len;
6034 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6035 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6036 em->block_start != EXTENT_MAP_HOLE)) {
6041 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6042 type = BTRFS_ORDERED_PREALLOC;
6044 type = BTRFS_ORDERED_NOCOW;
6045 len = min(len, em->len - (start - em->start));
6046 block_start = em->block_start + (start - em->start);
6049 * we're not going to log anything, but we do need
6050 * to make sure the current transaction stays open
6051 * while we look for nocow cross refs
6053 trans = btrfs_join_transaction(root);
6057 if (can_nocow_odirect(trans, inode, start, len) == 1) {
6058 u64 orig_start = em->start;
6060 if (type == BTRFS_ORDERED_PREALLOC) {
6061 free_extent_map(em);
6062 em = create_pinned_em(inode, start, len,
6064 block_start, len, type);
6066 btrfs_end_transaction(trans, root);
6071 ret = btrfs_add_ordered_extent_dio(inode, start,
6072 block_start, len, len, type);
6073 btrfs_end_transaction(trans, root);
6075 free_extent_map(em);
6080 btrfs_end_transaction(trans, root);
6084 * this will cow the extent, reset the len in case we changed
6087 len = bh_result->b_size;
6088 em = btrfs_new_extent_direct(inode, em, start, len);
6093 len = min(len, em->len - (start - em->start));
6095 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6097 bh_result->b_size = len;
6098 bh_result->b_bdev = em->bdev;
6099 set_buffer_mapped(bh_result);
6101 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6102 set_buffer_new(bh_result);
6105 * Need to update the i_size under the extent lock so buffered
6106 * readers will get the updated i_size when we unlock.
6108 if (start + len > i_size_read(inode))
6109 i_size_write(inode, start + len);
6113 * In the case of write we need to clear and unlock the entire range,
6114 * in the case of read we need to unlock only the end area that we
6115 * aren't using if there is any left over space.
6117 if (lockstart < lockend) {
6118 if (create && len < lockend - lockstart) {
6119 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6120 lockstart + len - 1,
6121 unlock_bits | EXTENT_DEFRAG, 1, 0,
6122 &cached_state, GFP_NOFS);
6124 * Beside unlock, we also need to cleanup reserved space
6125 * for the left range by attaching EXTENT_DO_ACCOUNTING.
6127 clear_extent_bit(&BTRFS_I(inode)->io_tree,
6128 lockstart + len, lockend,
6129 unlock_bits | EXTENT_DO_ACCOUNTING |
6130 EXTENT_DEFRAG, 1, 0, NULL, GFP_NOFS);
6132 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6133 lockend, unlock_bits, 1, 0,
6134 &cached_state, GFP_NOFS);
6137 free_extent_state(cached_state);
6140 free_extent_map(em);
6146 unlock_bits |= EXTENT_DO_ACCOUNTING;
6148 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6149 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6153 struct btrfs_dio_private {
6154 struct inode *inode;
6160 /* number of bios pending for this dio */
6161 atomic_t pending_bios;
6166 struct bio *orig_bio;
6169 static void btrfs_endio_direct_read(struct bio *bio, int err)
6171 struct btrfs_dio_private *dip = bio->bi_private;
6172 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6173 struct bio_vec *bvec = bio->bi_io_vec;
6174 struct inode *inode = dip->inode;
6175 struct btrfs_root *root = BTRFS_I(inode)->root;
6178 start = dip->logical_offset;
6180 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6181 struct page *page = bvec->bv_page;
6184 u64 private = ~(u32)0;
6185 unsigned long flags;
6187 if (get_state_private(&BTRFS_I(inode)->io_tree,
6190 local_irq_save(flags);
6191 kaddr = kmap_atomic(page);
6192 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
6193 csum, bvec->bv_len);
6194 btrfs_csum_final(csum, (char *)&csum);
6195 kunmap_atomic(kaddr);
6196 local_irq_restore(flags);
6198 flush_dcache_page(bvec->bv_page);
6199 if (csum != private) {
6201 printk(KERN_ERR "btrfs csum failed ino %llu off"
6202 " %llu csum %u private %u\n",
6203 (unsigned long long)btrfs_ino(inode),
6204 (unsigned long long)start,
6205 csum, (unsigned)private);
6210 start += bvec->bv_len;
6212 } while (bvec <= bvec_end);
6214 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6215 dip->logical_offset + dip->bytes - 1);
6216 bio->bi_private = dip->private;
6220 /* If we had a csum failure make sure to clear the uptodate flag */
6222 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6223 dio_end_io(bio, err);
6226 static void btrfs_endio_direct_write(struct bio *bio, int err)
6228 struct btrfs_dio_private *dip = bio->bi_private;
6229 struct inode *inode = dip->inode;
6230 struct btrfs_root *root = BTRFS_I(inode)->root;
6231 struct btrfs_ordered_extent *ordered = NULL;
6232 u64 ordered_offset = dip->logical_offset;
6233 u64 ordered_bytes = dip->bytes;
6239 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6241 ordered_bytes, !err);
6245 ordered->work.func = finish_ordered_fn;
6246 ordered->work.flags = 0;
6247 btrfs_queue_worker(&root->fs_info->endio_write_workers,
6251 * our bio might span multiple ordered extents. If we haven't
6252 * completed the accounting for the whole dio, go back and try again
6254 if (ordered_offset < dip->logical_offset + dip->bytes) {
6255 ordered_bytes = dip->logical_offset + dip->bytes -
6261 bio->bi_private = dip->private;
6265 /* If we had an error make sure to clear the uptodate flag */
6267 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6268 dio_end_io(bio, err);
6271 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6272 struct bio *bio, int mirror_num,
6273 unsigned long bio_flags, u64 offset)
6276 struct btrfs_root *root = BTRFS_I(inode)->root;
6277 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6278 BUG_ON(ret); /* -ENOMEM */
6282 static void btrfs_end_dio_bio(struct bio *bio, int err)
6284 struct btrfs_dio_private *dip = bio->bi_private;
6287 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6288 "sector %#Lx len %u err no %d\n",
6289 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6290 (unsigned long long)bio->bi_sector, bio->bi_size, err);
6294 * before atomic variable goto zero, we must make sure
6295 * dip->errors is perceived to be set.
6297 smp_mb__before_atomic_dec();
6300 /* if there are more bios still pending for this dio, just exit */
6301 if (!atomic_dec_and_test(&dip->pending_bios))
6305 bio_io_error(dip->orig_bio);
6307 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6308 bio_endio(dip->orig_bio, 0);
6314 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6315 u64 first_sector, gfp_t gfp_flags)
6317 int nr_vecs = bio_get_nr_vecs(bdev);
6318 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6321 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6322 int rw, u64 file_offset, int skip_sum,
6325 int write = rw & REQ_WRITE;
6326 struct btrfs_root *root = BTRFS_I(inode)->root;
6332 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6340 if (write && async_submit) {
6341 ret = btrfs_wq_submit_bio(root->fs_info,
6342 inode, rw, bio, 0, 0,
6344 __btrfs_submit_bio_start_direct_io,
6345 __btrfs_submit_bio_done);
6349 * If we aren't doing async submit, calculate the csum of the
6352 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6355 } else if (!skip_sum) {
6356 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
6362 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6368 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6371 struct inode *inode = dip->inode;
6372 struct btrfs_root *root = BTRFS_I(inode)->root;
6373 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6375 struct bio *orig_bio = dip->orig_bio;
6376 struct bio_vec *bvec = orig_bio->bi_io_vec;
6377 u64 start_sector = orig_bio->bi_sector;
6378 u64 file_offset = dip->logical_offset;
6383 int async_submit = 0;
6385 map_length = orig_bio->bi_size;
6386 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6387 &map_length, NULL, 0);
6393 if (map_length >= orig_bio->bi_size) {
6399 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6402 bio->bi_private = dip;
6403 bio->bi_end_io = btrfs_end_dio_bio;
6404 atomic_inc(&dip->pending_bios);
6406 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6407 if (unlikely(map_length < submit_len + bvec->bv_len ||
6408 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6409 bvec->bv_offset) < bvec->bv_len)) {
6411 * inc the count before we submit the bio so
6412 * we know the end IO handler won't happen before
6413 * we inc the count. Otherwise, the dip might get freed
6414 * before we're done setting it up
6416 atomic_inc(&dip->pending_bios);
6417 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6418 file_offset, skip_sum,
6422 atomic_dec(&dip->pending_bios);
6426 start_sector += submit_len >> 9;
6427 file_offset += submit_len;
6432 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6433 start_sector, GFP_NOFS);
6436 bio->bi_private = dip;
6437 bio->bi_end_io = btrfs_end_dio_bio;
6439 map_length = orig_bio->bi_size;
6440 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6441 &map_length, NULL, 0);
6447 submit_len += bvec->bv_len;
6454 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6463 * before atomic variable goto zero, we must
6464 * make sure dip->errors is perceived to be set.
6466 smp_mb__before_atomic_dec();
6467 if (atomic_dec_and_test(&dip->pending_bios))
6468 bio_io_error(dip->orig_bio);
6470 /* bio_end_io() will handle error, so we needn't return it */
6474 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6477 struct btrfs_root *root = BTRFS_I(inode)->root;
6478 struct btrfs_dio_private *dip;
6479 struct bio_vec *bvec = bio->bi_io_vec;
6481 int write = rw & REQ_WRITE;
6484 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6486 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6492 dip->private = bio->bi_private;
6494 dip->logical_offset = file_offset;
6498 dip->bytes += bvec->bv_len;
6500 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6502 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6503 bio->bi_private = dip;
6505 dip->orig_bio = bio;
6506 atomic_set(&dip->pending_bios, 0);
6509 bio->bi_end_io = btrfs_endio_direct_write;
6511 bio->bi_end_io = btrfs_endio_direct_read;
6513 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6518 * If this is a write, we need to clean up the reserved space and kill
6519 * the ordered extent.
6522 struct btrfs_ordered_extent *ordered;
6523 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6524 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6525 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6526 btrfs_free_reserved_extent(root, ordered->start,
6528 btrfs_put_ordered_extent(ordered);
6529 btrfs_put_ordered_extent(ordered);
6531 bio_endio(bio, ret);
6534 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6535 const struct iovec *iov, loff_t offset,
6536 unsigned long nr_segs)
6542 unsigned blocksize_mask = root->sectorsize - 1;
6543 ssize_t retval = -EINVAL;
6544 loff_t end = offset;
6546 if (offset & blocksize_mask)
6549 /* Check the memory alignment. Blocks cannot straddle pages */
6550 for (seg = 0; seg < nr_segs; seg++) {
6551 addr = (unsigned long)iov[seg].iov_base;
6552 size = iov[seg].iov_len;
6554 if ((addr & blocksize_mask) || (size & blocksize_mask))
6557 /* If this is a write we don't need to check anymore */
6562 * Check to make sure we don't have duplicate iov_base's in this
6563 * iovec, if so return EINVAL, otherwise we'll get csum errors
6564 * when reading back.
6566 for (i = seg + 1; i < nr_segs; i++) {
6567 if (iov[seg].iov_base == iov[i].iov_base)
6576 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6577 const struct iovec *iov, loff_t offset,
6578 unsigned long nr_segs)
6580 struct file *file = iocb->ki_filp;
6581 struct inode *inode = file->f_mapping->host;
6583 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6587 return __blockdev_direct_IO(rw, iocb, inode,
6588 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6589 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6590 btrfs_submit_direct, 0);
6593 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6594 __u64 start, __u64 len)
6596 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6599 int btrfs_readpage(struct file *file, struct page *page)
6601 struct extent_io_tree *tree;
6602 tree = &BTRFS_I(page->mapping->host)->io_tree;
6603 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6606 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6608 struct extent_io_tree *tree;
6611 if (current->flags & PF_MEMALLOC) {
6612 redirty_page_for_writepage(wbc, page);
6616 tree = &BTRFS_I(page->mapping->host)->io_tree;
6617 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6620 int btrfs_writepages(struct address_space *mapping,
6621 struct writeback_control *wbc)
6623 struct extent_io_tree *tree;
6625 tree = &BTRFS_I(mapping->host)->io_tree;
6626 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6630 btrfs_readpages(struct file *file, struct address_space *mapping,
6631 struct list_head *pages, unsigned nr_pages)
6633 struct extent_io_tree *tree;
6634 tree = &BTRFS_I(mapping->host)->io_tree;
6635 return extent_readpages(tree, mapping, pages, nr_pages,
6638 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6640 struct extent_io_tree *tree;
6641 struct extent_map_tree *map;
6644 tree = &BTRFS_I(page->mapping->host)->io_tree;
6645 map = &BTRFS_I(page->mapping->host)->extent_tree;
6646 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6648 ClearPagePrivate(page);
6649 set_page_private(page, 0);
6650 page_cache_release(page);
6655 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6657 if (PageWriteback(page) || PageDirty(page))
6659 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6662 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6664 struct inode *inode = page->mapping->host;
6665 struct extent_io_tree *tree;
6666 struct btrfs_ordered_extent *ordered;
6667 struct extent_state *cached_state = NULL;
6668 u64 page_start = page_offset(page);
6669 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6672 * we have the page locked, so new writeback can't start,
6673 * and the dirty bit won't be cleared while we are here.
6675 * Wait for IO on this page so that we can safely clear
6676 * the PagePrivate2 bit and do ordered accounting
6678 wait_on_page_writeback(page);
6680 tree = &BTRFS_I(inode)->io_tree;
6682 btrfs_releasepage(page, GFP_NOFS);
6685 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6686 ordered = btrfs_lookup_ordered_extent(inode,
6690 * IO on this page will never be started, so we need
6691 * to account for any ordered extents now
6693 clear_extent_bit(tree, page_start, page_end,
6694 EXTENT_DIRTY | EXTENT_DELALLOC |
6695 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
6696 EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
6698 * whoever cleared the private bit is responsible
6699 * for the finish_ordered_io
6701 if (TestClearPagePrivate2(page) &&
6702 btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6703 PAGE_CACHE_SIZE, 1)) {
6704 btrfs_finish_ordered_io(ordered);
6706 btrfs_put_ordered_extent(ordered);
6707 cached_state = NULL;
6708 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6710 clear_extent_bit(tree, page_start, page_end,
6711 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6712 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
6713 &cached_state, GFP_NOFS);
6714 __btrfs_releasepage(page, GFP_NOFS);
6716 ClearPageChecked(page);
6717 if (PagePrivate(page)) {
6718 ClearPagePrivate(page);
6719 set_page_private(page, 0);
6720 page_cache_release(page);
6725 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6726 * called from a page fault handler when a page is first dirtied. Hence we must
6727 * be careful to check for EOF conditions here. We set the page up correctly
6728 * for a written page which means we get ENOSPC checking when writing into
6729 * holes and correct delalloc and unwritten extent mapping on filesystems that
6730 * support these features.
6732 * We are not allowed to take the i_mutex here so we have to play games to
6733 * protect against truncate races as the page could now be beyond EOF. Because
6734 * vmtruncate() writes the inode size before removing pages, once we have the
6735 * page lock we can determine safely if the page is beyond EOF. If it is not
6736 * beyond EOF, then the page is guaranteed safe against truncation until we
6739 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6741 struct page *page = vmf->page;
6742 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6743 struct btrfs_root *root = BTRFS_I(inode)->root;
6744 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6745 struct btrfs_ordered_extent *ordered;
6746 struct extent_state *cached_state = NULL;
6748 unsigned long zero_start;
6755 sb_start_pagefault(inode->i_sb);
6756 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6758 ret = file_update_time(vma->vm_file);
6764 else /* -ENOSPC, -EIO, etc */
6765 ret = VM_FAULT_SIGBUS;
6771 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6774 size = i_size_read(inode);
6775 page_start = page_offset(page);
6776 page_end = page_start + PAGE_CACHE_SIZE - 1;
6778 if ((page->mapping != inode->i_mapping) ||
6779 (page_start >= size)) {
6780 /* page got truncated out from underneath us */
6783 wait_on_page_writeback(page);
6785 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6786 set_page_extent_mapped(page);
6789 * we can't set the delalloc bits if there are pending ordered
6790 * extents. Drop our locks and wait for them to finish
6792 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6794 unlock_extent_cached(io_tree, page_start, page_end,
6795 &cached_state, GFP_NOFS);
6797 btrfs_start_ordered_extent(inode, ordered, 1);
6798 btrfs_put_ordered_extent(ordered);
6803 * XXX - page_mkwrite gets called every time the page is dirtied, even
6804 * if it was already dirty, so for space accounting reasons we need to
6805 * clear any delalloc bits for the range we are fixing to save. There
6806 * is probably a better way to do this, but for now keep consistent with
6807 * prepare_pages in the normal write path.
6809 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6810 EXTENT_DIRTY | EXTENT_DELALLOC |
6811 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
6812 0, 0, &cached_state, GFP_NOFS);
6814 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6817 unlock_extent_cached(io_tree, page_start, page_end,
6818 &cached_state, GFP_NOFS);
6819 ret = VM_FAULT_SIGBUS;
6824 /* page is wholly or partially inside EOF */
6825 if (page_start + PAGE_CACHE_SIZE > size)
6826 zero_start = size & ~PAGE_CACHE_MASK;
6828 zero_start = PAGE_CACHE_SIZE;
6830 if (zero_start != PAGE_CACHE_SIZE) {
6832 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6833 flush_dcache_page(page);
6836 ClearPageChecked(page);
6837 set_page_dirty(page);
6838 SetPageUptodate(page);
6840 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6841 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6842 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
6844 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6848 sb_end_pagefault(inode->i_sb);
6849 return VM_FAULT_LOCKED;
6853 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6855 sb_end_pagefault(inode->i_sb);
6859 static int btrfs_truncate(struct inode *inode)
6861 struct btrfs_root *root = BTRFS_I(inode)->root;
6862 struct btrfs_block_rsv *rsv;
6865 struct btrfs_trans_handle *trans;
6867 u64 mask = root->sectorsize - 1;
6868 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6870 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
6874 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6875 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6878 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6879 * 3 things going on here
6881 * 1) We need to reserve space for our orphan item and the space to
6882 * delete our orphan item. Lord knows we don't want to have a dangling
6883 * orphan item because we didn't reserve space to remove it.
6885 * 2) We need to reserve space to update our inode.
6887 * 3) We need to have something to cache all the space that is going to
6888 * be free'd up by the truncate operation, but also have some slack
6889 * space reserved in case it uses space during the truncate (thank you
6890 * very much snapshotting).
6892 * And we need these to all be seperate. The fact is we can use alot of
6893 * space doing the truncate, and we have no earthly idea how much space
6894 * we will use, so we need the truncate reservation to be seperate so it
6895 * doesn't end up using space reserved for updating the inode or
6896 * removing the orphan item. We also need to be able to stop the
6897 * transaction and start a new one, which means we need to be able to
6898 * update the inode several times, and we have no idea of knowing how
6899 * many times that will be, so we can't just reserve 1 item for the
6900 * entirety of the opration, so that has to be done seperately as well.
6901 * Then there is the orphan item, which does indeed need to be held on
6902 * to for the whole operation, and we need nobody to touch this reserved
6903 * space except the orphan code.
6905 * So that leaves us with
6907 * 1) root->orphan_block_rsv - for the orphan deletion.
6908 * 2) rsv - for the truncate reservation, which we will steal from the
6909 * transaction reservation.
6910 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6911 * updating the inode.
6913 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
6916 rsv->size = min_size;
6920 * 1 for the truncate slack space
6921 * 1 for the orphan item we're going to add
6922 * 1 for the orphan item deletion
6923 * 1 for updating the inode.
6925 trans = btrfs_start_transaction(root, 4);
6926 if (IS_ERR(trans)) {
6927 err = PTR_ERR(trans);
6931 /* Migrate the slack space for the truncate to our reserve */
6932 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6936 ret = btrfs_orphan_add(trans, inode);
6938 btrfs_end_transaction(trans, root);
6943 * setattr is responsible for setting the ordered_data_close flag,
6944 * but that is only tested during the last file release. That
6945 * could happen well after the next commit, leaving a great big
6946 * window where new writes may get lost if someone chooses to write
6947 * to this file after truncating to zero
6949 * The inode doesn't have any dirty data here, and so if we commit
6950 * this is a noop. If someone immediately starts writing to the inode
6951 * it is very likely we'll catch some of their writes in this
6952 * transaction, and the commit will find this file on the ordered
6953 * data list with good things to send down.
6955 * This is a best effort solution, there is still a window where
6956 * using truncate to replace the contents of the file will
6957 * end up with a zero length file after a crash.
6959 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
6960 &BTRFS_I(inode)->runtime_flags))
6961 btrfs_add_ordered_operation(trans, root, inode);
6964 * So if we truncate and then write and fsync we normally would just
6965 * write the extents that changed, which is a problem if we need to
6966 * first truncate that entire inode. So set this flag so we write out
6967 * all of the extents in the inode to the sync log so we're completely
6970 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6971 trans->block_rsv = rsv;
6974 ret = btrfs_truncate_inode_items(trans, root, inode,
6976 BTRFS_EXTENT_DATA_KEY);
6977 if (ret != -ENOSPC) {
6982 trans->block_rsv = &root->fs_info->trans_block_rsv;
6983 ret = btrfs_update_inode(trans, root, inode);
6989 nr = trans->blocks_used;
6990 btrfs_end_transaction(trans, root);
6991 btrfs_btree_balance_dirty(root, nr);
6993 trans = btrfs_start_transaction(root, 2);
6994 if (IS_ERR(trans)) {
6995 ret = err = PTR_ERR(trans);
7000 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7002 BUG_ON(ret); /* shouldn't happen */
7003 trans->block_rsv = rsv;
7006 if (ret == 0 && inode->i_nlink > 0) {
7007 trans->block_rsv = root->orphan_block_rsv;
7008 ret = btrfs_orphan_del(trans, inode);
7011 } else if (ret && inode->i_nlink > 0) {
7013 * Failed to do the truncate, remove us from the in memory
7016 ret = btrfs_orphan_del(NULL, inode);
7020 trans->block_rsv = &root->fs_info->trans_block_rsv;
7021 ret = btrfs_update_inode(trans, root, inode);
7025 nr = trans->blocks_used;
7026 ret = btrfs_end_transaction(trans, root);
7027 btrfs_btree_balance_dirty(root, nr);
7031 btrfs_free_block_rsv(root, rsv);
7040 * create a new subvolume directory/inode (helper for the ioctl).
7042 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7043 struct btrfs_root *new_root, u64 new_dirid)
7045 struct inode *inode;
7049 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7050 new_dirid, new_dirid,
7051 S_IFDIR | (~current_umask() & S_IRWXUGO),
7054 return PTR_ERR(inode);
7055 inode->i_op = &btrfs_dir_inode_operations;
7056 inode->i_fop = &btrfs_dir_file_operations;
7058 set_nlink(inode, 1);
7059 btrfs_i_size_write(inode, 0);
7061 err = btrfs_update_inode(trans, new_root, inode);
7067 struct inode *btrfs_alloc_inode(struct super_block *sb)
7069 struct btrfs_inode *ei;
7070 struct inode *inode;
7072 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7079 ei->last_sub_trans = 0;
7080 ei->logged_trans = 0;
7081 ei->delalloc_bytes = 0;
7082 ei->disk_i_size = 0;
7085 ei->index_cnt = (u64)-1;
7086 ei->last_unlink_trans = 0;
7087 ei->last_log_commit = 0;
7089 spin_lock_init(&ei->lock);
7090 ei->outstanding_extents = 0;
7091 ei->reserved_extents = 0;
7093 ei->runtime_flags = 0;
7094 ei->force_compress = BTRFS_COMPRESS_NONE;
7096 ei->delayed_node = NULL;
7098 inode = &ei->vfs_inode;
7099 extent_map_tree_init(&ei->extent_tree);
7100 extent_io_tree_init(&ei->io_tree, &inode->i_data);
7101 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7102 ei->io_tree.track_uptodate = 1;
7103 ei->io_failure_tree.track_uptodate = 1;
7104 mutex_init(&ei->log_mutex);
7105 mutex_init(&ei->delalloc_mutex);
7106 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7107 INIT_LIST_HEAD(&ei->delalloc_inodes);
7108 INIT_LIST_HEAD(&ei->ordered_operations);
7109 RB_CLEAR_NODE(&ei->rb_node);
7114 static void btrfs_i_callback(struct rcu_head *head)
7116 struct inode *inode = container_of(head, struct inode, i_rcu);
7117 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7120 void btrfs_destroy_inode(struct inode *inode)
7122 struct btrfs_ordered_extent *ordered;
7123 struct btrfs_root *root = BTRFS_I(inode)->root;
7125 WARN_ON(!hlist_empty(&inode->i_dentry));
7126 WARN_ON(inode->i_data.nrpages);
7127 WARN_ON(BTRFS_I(inode)->outstanding_extents);
7128 WARN_ON(BTRFS_I(inode)->reserved_extents);
7129 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7130 WARN_ON(BTRFS_I(inode)->csum_bytes);
7133 * This can happen where we create an inode, but somebody else also
7134 * created the same inode and we need to destroy the one we already
7141 * Make sure we're properly removed from the ordered operation
7145 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7146 spin_lock(&root->fs_info->ordered_extent_lock);
7147 list_del_init(&BTRFS_I(inode)->ordered_operations);
7148 spin_unlock(&root->fs_info->ordered_extent_lock);
7151 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7152 &BTRFS_I(inode)->runtime_flags)) {
7153 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7154 (unsigned long long)btrfs_ino(inode));
7155 atomic_dec(&root->orphan_inodes);
7159 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7163 printk(KERN_ERR "btrfs found ordered "
7164 "extent %llu %llu on inode cleanup\n",
7165 (unsigned long long)ordered->file_offset,
7166 (unsigned long long)ordered->len);
7167 btrfs_remove_ordered_extent(inode, ordered);
7168 btrfs_put_ordered_extent(ordered);
7169 btrfs_put_ordered_extent(ordered);
7172 inode_tree_del(inode);
7173 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7175 btrfs_remove_delayed_node(inode);
7176 call_rcu(&inode->i_rcu, btrfs_i_callback);
7179 int btrfs_drop_inode(struct inode *inode)
7181 struct btrfs_root *root = BTRFS_I(inode)->root;
7183 if (btrfs_root_refs(&root->root_item) == 0 &&
7184 !btrfs_is_free_space_inode(inode))
7187 return generic_drop_inode(inode);
7190 static void init_once(void *foo)
7192 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7194 inode_init_once(&ei->vfs_inode);
7197 void btrfs_destroy_cachep(void)
7199 if (btrfs_inode_cachep)
7200 kmem_cache_destroy(btrfs_inode_cachep);
7201 if (btrfs_trans_handle_cachep)
7202 kmem_cache_destroy(btrfs_trans_handle_cachep);
7203 if (btrfs_transaction_cachep)
7204 kmem_cache_destroy(btrfs_transaction_cachep);
7205 if (btrfs_path_cachep)
7206 kmem_cache_destroy(btrfs_path_cachep);
7207 if (btrfs_free_space_cachep)
7208 kmem_cache_destroy(btrfs_free_space_cachep);
7211 int btrfs_init_cachep(void)
7213 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
7214 sizeof(struct btrfs_inode), 0,
7215 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7216 if (!btrfs_inode_cachep)
7219 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
7220 sizeof(struct btrfs_trans_handle), 0,
7221 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7222 if (!btrfs_trans_handle_cachep)
7225 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
7226 sizeof(struct btrfs_transaction), 0,
7227 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7228 if (!btrfs_transaction_cachep)
7231 btrfs_path_cachep = kmem_cache_create("btrfs_path",
7232 sizeof(struct btrfs_path), 0,
7233 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7234 if (!btrfs_path_cachep)
7237 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
7238 sizeof(struct btrfs_free_space), 0,
7239 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7240 if (!btrfs_free_space_cachep)
7245 btrfs_destroy_cachep();
7249 static int btrfs_getattr(struct vfsmount *mnt,
7250 struct dentry *dentry, struct kstat *stat)
7252 struct inode *inode = dentry->d_inode;
7253 u32 blocksize = inode->i_sb->s_blocksize;
7255 generic_fillattr(inode, stat);
7256 stat->dev = BTRFS_I(inode)->root->anon_dev;
7257 stat->blksize = PAGE_CACHE_SIZE;
7258 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7259 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
7264 * If a file is moved, it will inherit the cow and compression flags of the new
7267 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7269 struct btrfs_inode *b_dir = BTRFS_I(dir);
7270 struct btrfs_inode *b_inode = BTRFS_I(inode);
7272 if (b_dir->flags & BTRFS_INODE_NODATACOW)
7273 b_inode->flags |= BTRFS_INODE_NODATACOW;
7275 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7277 if (b_dir->flags & BTRFS_INODE_COMPRESS) {
7278 b_inode->flags |= BTRFS_INODE_COMPRESS;
7279 b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
7281 b_inode->flags &= ~(BTRFS_INODE_COMPRESS |
7282 BTRFS_INODE_NOCOMPRESS);
7286 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7287 struct inode *new_dir, struct dentry *new_dentry)
7289 struct btrfs_trans_handle *trans;
7290 struct btrfs_root *root = BTRFS_I(old_dir)->root;
7291 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7292 struct inode *new_inode = new_dentry->d_inode;
7293 struct inode *old_inode = old_dentry->d_inode;
7294 struct timespec ctime = CURRENT_TIME;
7298 u64 old_ino = btrfs_ino(old_inode);
7300 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7303 /* we only allow rename subvolume link between subvolumes */
7304 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7307 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7308 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7311 if (S_ISDIR(old_inode->i_mode) && new_inode &&
7312 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7315 * we're using rename to replace one file with another.
7316 * and the replacement file is large. Start IO on it now so
7317 * we don't add too much work to the end of the transaction
7319 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7320 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7321 filemap_flush(old_inode->i_mapping);
7323 /* close the racy window with snapshot create/destroy ioctl */
7324 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7325 down_read(&root->fs_info->subvol_sem);
7327 * We want to reserve the absolute worst case amount of items. So if
7328 * both inodes are subvols and we need to unlink them then that would
7329 * require 4 item modifications, but if they are both normal inodes it
7330 * would require 5 item modifications, so we'll assume their normal
7331 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7332 * should cover the worst case number of items we'll modify.
7334 trans = btrfs_start_transaction(root, 20);
7335 if (IS_ERR(trans)) {
7336 ret = PTR_ERR(trans);
7341 btrfs_record_root_in_trans(trans, dest);
7343 ret = btrfs_set_inode_index(new_dir, &index);
7347 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7348 /* force full log commit if subvolume involved. */
7349 root->fs_info->last_trans_log_full_commit = trans->transid;
7351 ret = btrfs_insert_inode_ref(trans, dest,
7352 new_dentry->d_name.name,
7353 new_dentry->d_name.len,
7355 btrfs_ino(new_dir), index);
7359 * this is an ugly little race, but the rename is required
7360 * to make sure that if we crash, the inode is either at the
7361 * old name or the new one. pinning the log transaction lets
7362 * us make sure we don't allow a log commit to come in after
7363 * we unlink the name but before we add the new name back in.
7365 btrfs_pin_log_trans(root);
7368 * make sure the inode gets flushed if it is replacing
7371 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7372 btrfs_add_ordered_operation(trans, root, old_inode);
7374 inode_inc_iversion(old_dir);
7375 inode_inc_iversion(new_dir);
7376 inode_inc_iversion(old_inode);
7377 old_dir->i_ctime = old_dir->i_mtime = ctime;
7378 new_dir->i_ctime = new_dir->i_mtime = ctime;
7379 old_inode->i_ctime = ctime;
7381 if (old_dentry->d_parent != new_dentry->d_parent)
7382 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7384 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7385 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7386 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7387 old_dentry->d_name.name,
7388 old_dentry->d_name.len);
7390 ret = __btrfs_unlink_inode(trans, root, old_dir,
7391 old_dentry->d_inode,
7392 old_dentry->d_name.name,
7393 old_dentry->d_name.len);
7395 ret = btrfs_update_inode(trans, root, old_inode);
7398 btrfs_abort_transaction(trans, root, ret);
7403 inode_inc_iversion(new_inode);
7404 new_inode->i_ctime = CURRENT_TIME;
7405 if (unlikely(btrfs_ino(new_inode) ==
7406 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7407 root_objectid = BTRFS_I(new_inode)->location.objectid;
7408 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7410 new_dentry->d_name.name,
7411 new_dentry->d_name.len);
7412 BUG_ON(new_inode->i_nlink == 0);
7414 ret = btrfs_unlink_inode(trans, dest, new_dir,
7415 new_dentry->d_inode,
7416 new_dentry->d_name.name,
7417 new_dentry->d_name.len);
7419 if (!ret && new_inode->i_nlink == 0) {
7420 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7424 btrfs_abort_transaction(trans, root, ret);
7429 fixup_inode_flags(new_dir, old_inode);
7431 ret = btrfs_add_link(trans, new_dir, old_inode,
7432 new_dentry->d_name.name,
7433 new_dentry->d_name.len, 0, index);
7435 btrfs_abort_transaction(trans, root, ret);
7439 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7440 struct dentry *parent = new_dentry->d_parent;
7441 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7442 btrfs_end_log_trans(root);
7445 btrfs_end_transaction(trans, root);
7447 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7448 up_read(&root->fs_info->subvol_sem);
7454 * some fairly slow code that needs optimization. This walks the list
7455 * of all the inodes with pending delalloc and forces them to disk.
7457 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7459 struct list_head *head = &root->fs_info->delalloc_inodes;
7460 struct btrfs_inode *binode;
7461 struct inode *inode;
7463 if (root->fs_info->sb->s_flags & MS_RDONLY)
7466 spin_lock(&root->fs_info->delalloc_lock);
7467 while (!list_empty(head)) {
7468 binode = list_entry(head->next, struct btrfs_inode,
7470 inode = igrab(&binode->vfs_inode);
7472 list_del_init(&binode->delalloc_inodes);
7473 spin_unlock(&root->fs_info->delalloc_lock);
7475 filemap_flush(inode->i_mapping);
7477 btrfs_add_delayed_iput(inode);
7482 spin_lock(&root->fs_info->delalloc_lock);
7484 spin_unlock(&root->fs_info->delalloc_lock);
7486 /* the filemap_flush will queue IO into the worker threads, but
7487 * we have to make sure the IO is actually started and that
7488 * ordered extents get created before we return
7490 atomic_inc(&root->fs_info->async_submit_draining);
7491 while (atomic_read(&root->fs_info->nr_async_submits) ||
7492 atomic_read(&root->fs_info->async_delalloc_pages)) {
7493 wait_event(root->fs_info->async_submit_wait,
7494 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7495 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7497 atomic_dec(&root->fs_info->async_submit_draining);
7501 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7502 const char *symname)
7504 struct btrfs_trans_handle *trans;
7505 struct btrfs_root *root = BTRFS_I(dir)->root;
7506 struct btrfs_path *path;
7507 struct btrfs_key key;
7508 struct inode *inode = NULL;
7516 struct btrfs_file_extent_item *ei;
7517 struct extent_buffer *leaf;
7518 unsigned long nr = 0;
7520 name_len = strlen(symname) + 1;
7521 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7522 return -ENAMETOOLONG;
7525 * 2 items for inode item and ref
7526 * 2 items for dir items
7527 * 1 item for xattr if selinux is on
7529 trans = btrfs_start_transaction(root, 5);
7531 return PTR_ERR(trans);
7533 err = btrfs_find_free_ino(root, &objectid);
7537 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7538 dentry->d_name.len, btrfs_ino(dir), objectid,
7539 S_IFLNK|S_IRWXUGO, &index);
7540 if (IS_ERR(inode)) {
7541 err = PTR_ERR(inode);
7545 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7552 * If the active LSM wants to access the inode during
7553 * d_instantiate it needs these. Smack checks to see
7554 * if the filesystem supports xattrs by looking at the
7557 inode->i_fop = &btrfs_file_operations;
7558 inode->i_op = &btrfs_file_inode_operations;
7560 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7564 inode->i_mapping->a_ops = &btrfs_aops;
7565 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7566 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7571 path = btrfs_alloc_path();
7577 key.objectid = btrfs_ino(inode);
7579 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7580 datasize = btrfs_file_extent_calc_inline_size(name_len);
7581 err = btrfs_insert_empty_item(trans, root, path, &key,
7585 btrfs_free_path(path);
7588 leaf = path->nodes[0];
7589 ei = btrfs_item_ptr(leaf, path->slots[0],
7590 struct btrfs_file_extent_item);
7591 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7592 btrfs_set_file_extent_type(leaf, ei,
7593 BTRFS_FILE_EXTENT_INLINE);
7594 btrfs_set_file_extent_encryption(leaf, ei, 0);
7595 btrfs_set_file_extent_compression(leaf, ei, 0);
7596 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7597 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7599 ptr = btrfs_file_extent_inline_start(ei);
7600 write_extent_buffer(leaf, symname, ptr, name_len);
7601 btrfs_mark_buffer_dirty(leaf);
7602 btrfs_free_path(path);
7604 inode->i_op = &btrfs_symlink_inode_operations;
7605 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7606 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7607 inode_set_bytes(inode, name_len);
7608 btrfs_i_size_write(inode, name_len - 1);
7609 err = btrfs_update_inode(trans, root, inode);
7615 d_instantiate(dentry, inode);
7616 nr = trans->blocks_used;
7617 btrfs_end_transaction(trans, root);
7619 inode_dec_link_count(inode);
7622 btrfs_btree_balance_dirty(root, nr);
7626 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7627 u64 start, u64 num_bytes, u64 min_size,
7628 loff_t actual_len, u64 *alloc_hint,
7629 struct btrfs_trans_handle *trans)
7631 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7632 struct extent_map *em;
7633 struct btrfs_root *root = BTRFS_I(inode)->root;
7634 struct btrfs_key ins;
7635 u64 cur_offset = start;
7638 bool own_trans = true;
7642 while (num_bytes > 0) {
7644 trans = btrfs_start_transaction(root, 3);
7645 if (IS_ERR(trans)) {
7646 ret = PTR_ERR(trans);
7651 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7652 0, *alloc_hint, &ins, 1);
7655 btrfs_end_transaction(trans, root);
7659 ret = insert_reserved_file_extent(trans, inode,
7660 cur_offset, ins.objectid,
7661 ins.offset, ins.offset,
7662 ins.offset, 0, 0, 0,
7663 BTRFS_FILE_EXTENT_PREALLOC);
7665 btrfs_abort_transaction(trans, root, ret);
7667 btrfs_end_transaction(trans, root);
7670 btrfs_drop_extent_cache(inode, cur_offset,
7671 cur_offset + ins.offset -1, 0);
7673 em = alloc_extent_map();
7675 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
7676 &BTRFS_I(inode)->runtime_flags);
7680 em->start = cur_offset;
7681 em->orig_start = cur_offset;
7682 em->len = ins.offset;
7683 em->block_start = ins.objectid;
7684 em->block_len = ins.offset;
7685 em->bdev = root->fs_info->fs_devices->latest_bdev;
7686 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7687 em->generation = trans->transid;
7690 write_lock(&em_tree->lock);
7691 ret = add_extent_mapping(em_tree, em);
7693 list_move(&em->list,
7694 &em_tree->modified_extents);
7695 write_unlock(&em_tree->lock);
7698 btrfs_drop_extent_cache(inode, cur_offset,
7699 cur_offset + ins.offset - 1,
7702 free_extent_map(em);
7704 num_bytes -= ins.offset;
7705 cur_offset += ins.offset;
7706 *alloc_hint = ins.objectid + ins.offset;
7708 inode_inc_iversion(inode);
7709 inode->i_ctime = CURRENT_TIME;
7710 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7711 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7712 (actual_len > inode->i_size) &&
7713 (cur_offset > inode->i_size)) {
7714 if (cur_offset > actual_len)
7715 i_size = actual_len;
7717 i_size = cur_offset;
7718 i_size_write(inode, i_size);
7719 btrfs_ordered_update_i_size(inode, i_size, NULL);
7722 ret = btrfs_update_inode(trans, root, inode);
7725 btrfs_abort_transaction(trans, root, ret);
7727 btrfs_end_transaction(trans, root);
7732 btrfs_end_transaction(trans, root);
7737 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7738 u64 start, u64 num_bytes, u64 min_size,
7739 loff_t actual_len, u64 *alloc_hint)
7741 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7742 min_size, actual_len, alloc_hint,
7746 int btrfs_prealloc_file_range_trans(struct inode *inode,
7747 struct btrfs_trans_handle *trans, int mode,
7748 u64 start, u64 num_bytes, u64 min_size,
7749 loff_t actual_len, u64 *alloc_hint)
7751 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7752 min_size, actual_len, alloc_hint, trans);
7755 static int btrfs_set_page_dirty(struct page *page)
7757 return __set_page_dirty_nobuffers(page);
7760 static int btrfs_permission(struct inode *inode, int mask)
7762 struct btrfs_root *root = BTRFS_I(inode)->root;
7763 umode_t mode = inode->i_mode;
7765 if (mask & MAY_WRITE &&
7766 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7767 if (btrfs_root_readonly(root))
7769 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7772 return generic_permission(inode, mask);
7775 static const struct inode_operations btrfs_dir_inode_operations = {
7776 .getattr = btrfs_getattr,
7777 .lookup = btrfs_lookup,
7778 .create = btrfs_create,
7779 .unlink = btrfs_unlink,
7781 .mkdir = btrfs_mkdir,
7782 .rmdir = btrfs_rmdir,
7783 .rename = btrfs_rename,
7784 .symlink = btrfs_symlink,
7785 .setattr = btrfs_setattr,
7786 .mknod = btrfs_mknod,
7787 .setxattr = btrfs_setxattr,
7788 .getxattr = btrfs_getxattr,
7789 .listxattr = btrfs_listxattr,
7790 .removexattr = btrfs_removexattr,
7791 .permission = btrfs_permission,
7792 .get_acl = btrfs_get_acl,
7794 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7795 .lookup = btrfs_lookup,
7796 .permission = btrfs_permission,
7797 .get_acl = btrfs_get_acl,
7800 static const struct file_operations btrfs_dir_file_operations = {
7801 .llseek = generic_file_llseek,
7802 .read = generic_read_dir,
7803 .readdir = btrfs_real_readdir,
7804 .unlocked_ioctl = btrfs_ioctl,
7805 #ifdef CONFIG_COMPAT
7806 .compat_ioctl = btrfs_ioctl,
7808 .release = btrfs_release_file,
7809 .fsync = btrfs_sync_file,
7812 static struct extent_io_ops btrfs_extent_io_ops = {
7813 .fill_delalloc = run_delalloc_range,
7814 .submit_bio_hook = btrfs_submit_bio_hook,
7815 .merge_bio_hook = btrfs_merge_bio_hook,
7816 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7817 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7818 .writepage_start_hook = btrfs_writepage_start_hook,
7819 .set_bit_hook = btrfs_set_bit_hook,
7820 .clear_bit_hook = btrfs_clear_bit_hook,
7821 .merge_extent_hook = btrfs_merge_extent_hook,
7822 .split_extent_hook = btrfs_split_extent_hook,
7826 * btrfs doesn't support the bmap operation because swapfiles
7827 * use bmap to make a mapping of extents in the file. They assume
7828 * these extents won't change over the life of the file and they
7829 * use the bmap result to do IO directly to the drive.
7831 * the btrfs bmap call would return logical addresses that aren't
7832 * suitable for IO and they also will change frequently as COW
7833 * operations happen. So, swapfile + btrfs == corruption.
7835 * For now we're avoiding this by dropping bmap.
7837 static const struct address_space_operations btrfs_aops = {
7838 .readpage = btrfs_readpage,
7839 .writepage = btrfs_writepage,
7840 .writepages = btrfs_writepages,
7841 .readpages = btrfs_readpages,
7842 .direct_IO = btrfs_direct_IO,
7843 .invalidatepage = btrfs_invalidatepage,
7844 .releasepage = btrfs_releasepage,
7845 .set_page_dirty = btrfs_set_page_dirty,
7846 .error_remove_page = generic_error_remove_page,
7849 static const struct address_space_operations btrfs_symlink_aops = {
7850 .readpage = btrfs_readpage,
7851 .writepage = btrfs_writepage,
7852 .invalidatepage = btrfs_invalidatepage,
7853 .releasepage = btrfs_releasepage,
7856 static const struct inode_operations btrfs_file_inode_operations = {
7857 .getattr = btrfs_getattr,
7858 .setattr = btrfs_setattr,
7859 .setxattr = btrfs_setxattr,
7860 .getxattr = btrfs_getxattr,
7861 .listxattr = btrfs_listxattr,
7862 .removexattr = btrfs_removexattr,
7863 .permission = btrfs_permission,
7864 .fiemap = btrfs_fiemap,
7865 .get_acl = btrfs_get_acl,
7866 .update_time = btrfs_update_time,
7868 static const struct inode_operations btrfs_special_inode_operations = {
7869 .getattr = btrfs_getattr,
7870 .setattr = btrfs_setattr,
7871 .permission = btrfs_permission,
7872 .setxattr = btrfs_setxattr,
7873 .getxattr = btrfs_getxattr,
7874 .listxattr = btrfs_listxattr,
7875 .removexattr = btrfs_removexattr,
7876 .get_acl = btrfs_get_acl,
7877 .update_time = btrfs_update_time,
7879 static const struct inode_operations btrfs_symlink_inode_operations = {
7880 .readlink = generic_readlink,
7881 .follow_link = page_follow_link_light,
7882 .put_link = page_put_link,
7883 .getattr = btrfs_getattr,
7884 .setattr = btrfs_setattr,
7885 .permission = btrfs_permission,
7886 .setxattr = btrfs_setxattr,
7887 .getxattr = btrfs_getxattr,
7888 .listxattr = btrfs_listxattr,
7889 .removexattr = btrfs_removexattr,
7890 .get_acl = btrfs_get_acl,
7891 .update_time = btrfs_update_time,
7894 const struct dentry_operations btrfs_dentry_operations = {
7895 .d_delete = btrfs_dentry_delete,
7896 .d_release = btrfs_dentry_release,