2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include <linux/mount.h>
45 #include "transaction.h"
46 #include "btrfs_inode.h"
48 #include "print-tree.h"
49 #include "ordered-data.h"
53 #include "compression.h"
55 #include "free-space-cache.h"
56 #include "inode-map.h"
58 struct btrfs_iget_args {
60 struct btrfs_root *root;
63 static const struct inode_operations btrfs_dir_inode_operations;
64 static const struct inode_operations btrfs_symlink_inode_operations;
65 static const struct inode_operations btrfs_dir_ro_inode_operations;
66 static const struct inode_operations btrfs_special_inode_operations;
67 static const struct inode_operations btrfs_file_inode_operations;
68 static const struct address_space_operations btrfs_aops;
69 static const struct address_space_operations btrfs_symlink_aops;
70 static const struct file_operations btrfs_dir_file_operations;
71 static struct extent_io_ops btrfs_extent_io_ops;
73 static struct kmem_cache *btrfs_inode_cachep;
74 static struct kmem_cache *btrfs_delalloc_work_cachep;
75 struct kmem_cache *btrfs_trans_handle_cachep;
76 struct kmem_cache *btrfs_transaction_cachep;
77 struct kmem_cache *btrfs_path_cachep;
78 struct kmem_cache *btrfs_free_space_cachep;
81 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
82 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
83 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
84 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
85 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
86 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
87 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
88 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
91 static int btrfs_setsize(struct inode *inode, loff_t newsize);
92 static int btrfs_truncate(struct inode *inode);
93 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
94 static noinline int cow_file_range(struct inode *inode,
95 struct page *locked_page,
96 u64 start, u64 end, int *page_started,
97 unsigned long *nr_written, int unlock);
99 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
100 struct inode *inode, struct inode *dir,
101 const struct qstr *qstr)
105 err = btrfs_init_acl(trans, inode, dir);
107 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
112 * this does all the hard work for inserting an inline extent into
113 * the btree. The caller should have done a btrfs_drop_extents so that
114 * no overlapping inline items exist in the btree
116 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
117 struct btrfs_root *root, struct inode *inode,
118 u64 start, size_t size, size_t compressed_size,
120 struct page **compressed_pages)
122 struct btrfs_key key;
123 struct btrfs_path *path;
124 struct extent_buffer *leaf;
125 struct page *page = NULL;
128 struct btrfs_file_extent_item *ei;
131 size_t cur_size = size;
133 unsigned long offset;
135 if (compressed_size && compressed_pages)
136 cur_size = compressed_size;
138 path = btrfs_alloc_path();
142 path->leave_spinning = 1;
144 key.objectid = btrfs_ino(inode);
146 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
147 datasize = btrfs_file_extent_calc_inline_size(cur_size);
149 inode_add_bytes(inode, size);
150 ret = btrfs_insert_empty_item(trans, root, path, &key,
156 leaf = path->nodes[0];
157 ei = btrfs_item_ptr(leaf, path->slots[0],
158 struct btrfs_file_extent_item);
159 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
160 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
161 btrfs_set_file_extent_encryption(leaf, ei, 0);
162 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
163 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
164 ptr = btrfs_file_extent_inline_start(ei);
166 if (compress_type != BTRFS_COMPRESS_NONE) {
169 while (compressed_size > 0) {
170 cpage = compressed_pages[i];
171 cur_size = min_t(unsigned long, compressed_size,
174 kaddr = kmap_atomic(cpage);
175 write_extent_buffer(leaf, kaddr, ptr, cur_size);
176 kunmap_atomic(kaddr);
180 compressed_size -= cur_size;
182 btrfs_set_file_extent_compression(leaf, ei,
185 page = find_get_page(inode->i_mapping,
186 start >> PAGE_CACHE_SHIFT);
187 btrfs_set_file_extent_compression(leaf, ei, 0);
188 kaddr = kmap_atomic(page);
189 offset = start & (PAGE_CACHE_SIZE - 1);
190 write_extent_buffer(leaf, kaddr + offset, ptr, size);
191 kunmap_atomic(kaddr);
192 page_cache_release(page);
194 btrfs_mark_buffer_dirty(leaf);
195 btrfs_free_path(path);
198 * we're an inline extent, so nobody can
199 * extend the file past i_size without locking
200 * a page we already have locked.
202 * We must do any isize and inode updates
203 * before we unlock the pages. Otherwise we
204 * could end up racing with unlink.
206 BTRFS_I(inode)->disk_i_size = inode->i_size;
207 ret = btrfs_update_inode(trans, root, inode);
211 btrfs_free_path(path);
217 * conditionally insert an inline extent into the file. This
218 * does the checks required to make sure the data is small enough
219 * to fit as an inline extent.
221 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
222 struct btrfs_root *root,
223 struct inode *inode, u64 start, u64 end,
224 size_t compressed_size, int compress_type,
225 struct page **compressed_pages)
227 u64 isize = i_size_read(inode);
228 u64 actual_end = min(end + 1, isize);
229 u64 inline_len = actual_end - start;
230 u64 aligned_end = (end + root->sectorsize - 1) &
231 ~((u64)root->sectorsize - 1);
232 u64 data_len = inline_len;
236 data_len = compressed_size;
239 actual_end >= PAGE_CACHE_SIZE ||
240 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
242 (actual_end & (root->sectorsize - 1)) == 0) ||
244 data_len > root->fs_info->max_inline) {
248 ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
252 if (isize > actual_end)
253 inline_len = min_t(u64, isize, actual_end);
254 ret = insert_inline_extent(trans, root, inode, start,
255 inline_len, compressed_size,
256 compress_type, compressed_pages);
257 if (ret && ret != -ENOSPC) {
258 btrfs_abort_transaction(trans, root, ret);
260 } else if (ret == -ENOSPC) {
264 btrfs_delalloc_release_metadata(inode, end + 1 - start);
265 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
269 struct async_extent {
274 unsigned long nr_pages;
276 struct list_head list;
281 struct btrfs_root *root;
282 struct page *locked_page;
285 struct list_head extents;
286 struct btrfs_work work;
289 static noinline int add_async_extent(struct async_cow *cow,
290 u64 start, u64 ram_size,
293 unsigned long nr_pages,
296 struct async_extent *async_extent;
298 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
299 BUG_ON(!async_extent); /* -ENOMEM */
300 async_extent->start = start;
301 async_extent->ram_size = ram_size;
302 async_extent->compressed_size = compressed_size;
303 async_extent->pages = pages;
304 async_extent->nr_pages = nr_pages;
305 async_extent->compress_type = compress_type;
306 list_add_tail(&async_extent->list, &cow->extents);
311 * we create compressed extents in two phases. The first
312 * phase compresses a range of pages that have already been
313 * locked (both pages and state bits are locked).
315 * This is done inside an ordered work queue, and the compression
316 * is spread across many cpus. The actual IO submission is step
317 * two, and the ordered work queue takes care of making sure that
318 * happens in the same order things were put onto the queue by
319 * writepages and friends.
321 * If this code finds it can't get good compression, it puts an
322 * entry onto the work queue to write the uncompressed bytes. This
323 * makes sure that both compressed inodes and uncompressed inodes
324 * are written in the same order that the flusher thread sent them
327 static noinline int compress_file_range(struct inode *inode,
328 struct page *locked_page,
330 struct async_cow *async_cow,
333 struct btrfs_root *root = BTRFS_I(inode)->root;
334 struct btrfs_trans_handle *trans;
336 u64 blocksize = root->sectorsize;
338 u64 isize = i_size_read(inode);
340 struct page **pages = NULL;
341 unsigned long nr_pages;
342 unsigned long nr_pages_ret = 0;
343 unsigned long total_compressed = 0;
344 unsigned long total_in = 0;
345 unsigned long max_compressed = 128 * 1024;
346 unsigned long max_uncompressed = 128 * 1024;
349 int compress_type = root->fs_info->compress_type;
351 /* if this is a small write inside eof, kick off a defrag */
352 if ((end - start + 1) < 16 * 1024 &&
353 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
354 btrfs_add_inode_defrag(NULL, inode);
356 actual_end = min_t(u64, isize, end + 1);
359 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
360 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
363 * we don't want to send crud past the end of i_size through
364 * compression, that's just a waste of CPU time. So, if the
365 * end of the file is before the start of our current
366 * requested range of bytes, we bail out to the uncompressed
367 * cleanup code that can deal with all of this.
369 * It isn't really the fastest way to fix things, but this is a
370 * very uncommon corner.
372 if (actual_end <= start)
373 goto cleanup_and_bail_uncompressed;
375 total_compressed = actual_end - start;
377 /* we want to make sure that amount of ram required to uncompress
378 * an extent is reasonable, so we limit the total size in ram
379 * of a compressed extent to 128k. This is a crucial number
380 * because it also controls how easily we can spread reads across
381 * cpus for decompression.
383 * We also want to make sure the amount of IO required to do
384 * a random read is reasonably small, so we limit the size of
385 * a compressed extent to 128k.
387 total_compressed = min(total_compressed, max_uncompressed);
388 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
389 num_bytes = max(blocksize, num_bytes);
394 * we do compression for mount -o compress and when the
395 * inode has not been flagged as nocompress. This flag can
396 * change at any time if we discover bad compression ratios.
398 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
399 (btrfs_test_opt(root, COMPRESS) ||
400 (BTRFS_I(inode)->force_compress) ||
401 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
403 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
405 /* just bail out to the uncompressed code */
409 if (BTRFS_I(inode)->force_compress)
410 compress_type = BTRFS_I(inode)->force_compress;
412 ret = btrfs_compress_pages(compress_type,
413 inode->i_mapping, start,
414 total_compressed, pages,
415 nr_pages, &nr_pages_ret,
421 unsigned long offset = total_compressed &
422 (PAGE_CACHE_SIZE - 1);
423 struct page *page = pages[nr_pages_ret - 1];
426 /* zero the tail end of the last page, we might be
427 * sending it down to disk
430 kaddr = kmap_atomic(page);
431 memset(kaddr + offset, 0,
432 PAGE_CACHE_SIZE - offset);
433 kunmap_atomic(kaddr);
440 trans = btrfs_join_transaction(root);
442 ret = PTR_ERR(trans);
444 goto cleanup_and_out;
446 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
448 /* lets try to make an inline extent */
449 if (ret || total_in < (actual_end - start)) {
450 /* we didn't compress the entire range, try
451 * to make an uncompressed inline extent.
453 ret = cow_file_range_inline(trans, root, inode,
454 start, end, 0, 0, NULL);
456 /* try making a compressed inline extent */
457 ret = cow_file_range_inline(trans, root, inode,
460 compress_type, pages);
464 * inline extent creation worked or returned error,
465 * we don't need to create any more async work items.
466 * Unlock and free up our temp pages.
468 extent_clear_unlock_delalloc(inode,
469 &BTRFS_I(inode)->io_tree,
471 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
472 EXTENT_CLEAR_DELALLOC |
473 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
475 btrfs_end_transaction(trans, root);
478 btrfs_end_transaction(trans, root);
483 * we aren't doing an inline extent round the compressed size
484 * up to a block size boundary so the allocator does sane
487 total_compressed = (total_compressed + blocksize - 1) &
491 * one last check to make sure the compression is really a
492 * win, compare the page count read with the blocks on disk
494 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
495 ~(PAGE_CACHE_SIZE - 1);
496 if (total_compressed >= total_in) {
499 num_bytes = total_in;
502 if (!will_compress && pages) {
504 * the compression code ran but failed to make things smaller,
505 * free any pages it allocated and our page pointer array
507 for (i = 0; i < nr_pages_ret; i++) {
508 WARN_ON(pages[i]->mapping);
509 page_cache_release(pages[i]);
513 total_compressed = 0;
516 /* flag the file so we don't compress in the future */
517 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
518 !(BTRFS_I(inode)->force_compress)) {
519 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
525 /* the async work queues will take care of doing actual
526 * allocation on disk for these compressed pages,
527 * and will submit them to the elevator.
529 add_async_extent(async_cow, start, num_bytes,
530 total_compressed, pages, nr_pages_ret,
533 if (start + num_bytes < end) {
540 cleanup_and_bail_uncompressed:
542 * No compression, but we still need to write the pages in
543 * the file we've been given so far. redirty the locked
544 * page if it corresponds to our extent and set things up
545 * for the async work queue to run cow_file_range to do
546 * the normal delalloc dance
548 if (page_offset(locked_page) >= start &&
549 page_offset(locked_page) <= end) {
550 __set_page_dirty_nobuffers(locked_page);
551 /* unlocked later on in the async handlers */
553 add_async_extent(async_cow, start, end - start + 1,
554 0, NULL, 0, BTRFS_COMPRESS_NONE);
562 for (i = 0; i < nr_pages_ret; i++) {
563 WARN_ON(pages[i]->mapping);
564 page_cache_release(pages[i]);
571 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
573 EXTENT_CLEAR_UNLOCK_PAGE |
575 EXTENT_CLEAR_DELALLOC |
576 EXTENT_SET_WRITEBACK |
577 EXTENT_END_WRITEBACK);
578 if (!trans || IS_ERR(trans))
579 btrfs_error(root->fs_info, ret, "Failed to join transaction");
581 btrfs_abort_transaction(trans, root, ret);
586 * phase two of compressed writeback. This is the ordered portion
587 * of the code, which only gets called in the order the work was
588 * queued. We walk all the async extents created by compress_file_range
589 * and send them down to the disk.
591 static noinline int submit_compressed_extents(struct inode *inode,
592 struct async_cow *async_cow)
594 struct async_extent *async_extent;
596 struct btrfs_trans_handle *trans;
597 struct btrfs_key ins;
598 struct extent_map *em;
599 struct btrfs_root *root = BTRFS_I(inode)->root;
600 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
601 struct extent_io_tree *io_tree;
604 if (list_empty(&async_cow->extents))
608 while (!list_empty(&async_cow->extents)) {
609 async_extent = list_entry(async_cow->extents.next,
610 struct async_extent, list);
611 list_del(&async_extent->list);
613 io_tree = &BTRFS_I(inode)->io_tree;
616 /* did the compression code fall back to uncompressed IO? */
617 if (!async_extent->pages) {
618 int page_started = 0;
619 unsigned long nr_written = 0;
621 lock_extent(io_tree, async_extent->start,
622 async_extent->start +
623 async_extent->ram_size - 1);
625 /* allocate blocks */
626 ret = cow_file_range(inode, async_cow->locked_page,
628 async_extent->start +
629 async_extent->ram_size - 1,
630 &page_started, &nr_written, 0);
635 * if page_started, cow_file_range inserted an
636 * inline extent and took care of all the unlocking
637 * and IO for us. Otherwise, we need to submit
638 * all those pages down to the drive.
640 if (!page_started && !ret)
641 extent_write_locked_range(io_tree,
642 inode, async_extent->start,
643 async_extent->start +
644 async_extent->ram_size - 1,
652 lock_extent(io_tree, async_extent->start,
653 async_extent->start + async_extent->ram_size - 1);
655 trans = btrfs_join_transaction(root);
657 ret = PTR_ERR(trans);
659 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
660 ret = btrfs_reserve_extent(trans, root,
661 async_extent->compressed_size,
662 async_extent->compressed_size,
663 0, alloc_hint, &ins, 1);
664 if (ret && ret != -ENOSPC)
665 btrfs_abort_transaction(trans, root, ret);
666 btrfs_end_transaction(trans, root);
671 for (i = 0; i < async_extent->nr_pages; i++) {
672 WARN_ON(async_extent->pages[i]->mapping);
673 page_cache_release(async_extent->pages[i]);
675 kfree(async_extent->pages);
676 async_extent->nr_pages = 0;
677 async_extent->pages = NULL;
678 unlock_extent(io_tree, async_extent->start,
679 async_extent->start +
680 async_extent->ram_size - 1);
683 goto out_free; /* JDM: Requeue? */
687 * here we're doing allocation and writeback of the
690 btrfs_drop_extent_cache(inode, async_extent->start,
691 async_extent->start +
692 async_extent->ram_size - 1, 0);
694 em = alloc_extent_map();
695 BUG_ON(!em); /* -ENOMEM */
696 em->start = async_extent->start;
697 em->len = async_extent->ram_size;
698 em->orig_start = em->start;
700 em->block_start = ins.objectid;
701 em->block_len = ins.offset;
702 em->bdev = root->fs_info->fs_devices->latest_bdev;
703 em->compress_type = async_extent->compress_type;
704 set_bit(EXTENT_FLAG_PINNED, &em->flags);
705 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
708 write_lock(&em_tree->lock);
709 ret = add_extent_mapping(em_tree, em);
710 write_unlock(&em_tree->lock);
711 if (ret != -EEXIST) {
715 btrfs_drop_extent_cache(inode, async_extent->start,
716 async_extent->start +
717 async_extent->ram_size - 1, 0);
720 ret = btrfs_add_ordered_extent_compress(inode,
723 async_extent->ram_size,
725 BTRFS_ORDERED_COMPRESSED,
726 async_extent->compress_type);
727 BUG_ON(ret); /* -ENOMEM */
730 * clear dirty, set writeback and unlock the pages.
732 extent_clear_unlock_delalloc(inode,
733 &BTRFS_I(inode)->io_tree,
735 async_extent->start +
736 async_extent->ram_size - 1,
737 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
738 EXTENT_CLEAR_UNLOCK |
739 EXTENT_CLEAR_DELALLOC |
740 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
742 ret = btrfs_submit_compressed_write(inode,
744 async_extent->ram_size,
746 ins.offset, async_extent->pages,
747 async_extent->nr_pages);
749 BUG_ON(ret); /* -ENOMEM */
750 alloc_hint = ins.objectid + ins.offset;
762 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
765 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
766 struct extent_map *em;
769 read_lock(&em_tree->lock);
770 em = search_extent_mapping(em_tree, start, num_bytes);
773 * if block start isn't an actual block number then find the
774 * first block in this inode and use that as a hint. If that
775 * block is also bogus then just don't worry about it.
777 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
779 em = search_extent_mapping(em_tree, 0, 0);
780 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
781 alloc_hint = em->block_start;
785 alloc_hint = em->block_start;
789 read_unlock(&em_tree->lock);
795 * when extent_io.c finds a delayed allocation range in the file,
796 * the call backs end up in this code. The basic idea is to
797 * allocate extents on disk for the range, and create ordered data structs
798 * in ram to track those extents.
800 * locked_page is the page that writepage had locked already. We use
801 * it to make sure we don't do extra locks or unlocks.
803 * *page_started is set to one if we unlock locked_page and do everything
804 * required to start IO on it. It may be clean and already done with
807 static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
809 struct btrfs_root *root,
810 struct page *locked_page,
811 u64 start, u64 end, int *page_started,
812 unsigned long *nr_written,
817 unsigned long ram_size;
820 u64 blocksize = root->sectorsize;
821 struct btrfs_key ins;
822 struct extent_map *em;
823 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
826 BUG_ON(btrfs_is_free_space_inode(inode));
828 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
829 num_bytes = max(blocksize, num_bytes);
830 disk_num_bytes = num_bytes;
832 /* if this is a small write inside eof, kick off defrag */
833 if (num_bytes < 64 * 1024 &&
834 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
835 btrfs_add_inode_defrag(trans, inode);
838 /* lets try to make an inline extent */
839 ret = cow_file_range_inline(trans, root, inode,
840 start, end, 0, 0, NULL);
842 extent_clear_unlock_delalloc(inode,
843 &BTRFS_I(inode)->io_tree,
845 EXTENT_CLEAR_UNLOCK_PAGE |
846 EXTENT_CLEAR_UNLOCK |
847 EXTENT_CLEAR_DELALLOC |
849 EXTENT_SET_WRITEBACK |
850 EXTENT_END_WRITEBACK);
852 *nr_written = *nr_written +
853 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
856 } else if (ret < 0) {
857 btrfs_abort_transaction(trans, root, ret);
862 BUG_ON(disk_num_bytes >
863 btrfs_super_total_bytes(root->fs_info->super_copy));
865 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
866 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
868 while (disk_num_bytes > 0) {
871 cur_alloc_size = disk_num_bytes;
872 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
873 root->sectorsize, 0, alloc_hint,
876 btrfs_abort_transaction(trans, root, ret);
880 em = alloc_extent_map();
881 BUG_ON(!em); /* -ENOMEM */
883 em->orig_start = em->start;
884 ram_size = ins.offset;
885 em->len = ins.offset;
887 em->block_start = ins.objectid;
888 em->block_len = ins.offset;
889 em->bdev = root->fs_info->fs_devices->latest_bdev;
890 set_bit(EXTENT_FLAG_PINNED, &em->flags);
893 write_lock(&em_tree->lock);
894 ret = add_extent_mapping(em_tree, em);
895 write_unlock(&em_tree->lock);
896 if (ret != -EEXIST) {
900 btrfs_drop_extent_cache(inode, start,
901 start + ram_size - 1, 0);
904 cur_alloc_size = ins.offset;
905 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
906 ram_size, cur_alloc_size, 0);
907 BUG_ON(ret); /* -ENOMEM */
909 if (root->root_key.objectid ==
910 BTRFS_DATA_RELOC_TREE_OBJECTID) {
911 ret = btrfs_reloc_clone_csums(inode, start,
914 btrfs_abort_transaction(trans, root, ret);
919 if (disk_num_bytes < cur_alloc_size)
922 /* we're not doing compressed IO, don't unlock the first
923 * page (which the caller expects to stay locked), don't
924 * clear any dirty bits and don't set any writeback bits
926 * Do set the Private2 bit so we know this page was properly
927 * setup for writepage
929 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
930 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
933 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
934 start, start + ram_size - 1,
936 disk_num_bytes -= cur_alloc_size;
937 num_bytes -= cur_alloc_size;
938 alloc_hint = ins.objectid + ins.offset;
939 start += cur_alloc_size;
945 extent_clear_unlock_delalloc(inode,
946 &BTRFS_I(inode)->io_tree,
947 start, end, locked_page,
948 EXTENT_CLEAR_UNLOCK_PAGE |
949 EXTENT_CLEAR_UNLOCK |
950 EXTENT_CLEAR_DELALLOC |
952 EXTENT_SET_WRITEBACK |
953 EXTENT_END_WRITEBACK);
958 static noinline int cow_file_range(struct inode *inode,
959 struct page *locked_page,
960 u64 start, u64 end, int *page_started,
961 unsigned long *nr_written,
964 struct btrfs_trans_handle *trans;
965 struct btrfs_root *root = BTRFS_I(inode)->root;
968 trans = btrfs_join_transaction(root);
970 extent_clear_unlock_delalloc(inode,
971 &BTRFS_I(inode)->io_tree,
972 start, end, locked_page,
973 EXTENT_CLEAR_UNLOCK_PAGE |
974 EXTENT_CLEAR_UNLOCK |
975 EXTENT_CLEAR_DELALLOC |
977 EXTENT_SET_WRITEBACK |
978 EXTENT_END_WRITEBACK);
979 return PTR_ERR(trans);
981 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
983 ret = __cow_file_range(trans, inode, root, locked_page, start, end,
984 page_started, nr_written, unlock);
986 btrfs_end_transaction(trans, root);
992 * work queue call back to started compression on a file and pages
994 static noinline void async_cow_start(struct btrfs_work *work)
996 struct async_cow *async_cow;
998 async_cow = container_of(work, struct async_cow, work);
1000 compress_file_range(async_cow->inode, async_cow->locked_page,
1001 async_cow->start, async_cow->end, async_cow,
1003 if (num_added == 0) {
1004 btrfs_add_delayed_iput(async_cow->inode);
1005 async_cow->inode = NULL;
1010 * work queue call back to submit previously compressed pages
1012 static noinline void async_cow_submit(struct btrfs_work *work)
1014 struct async_cow *async_cow;
1015 struct btrfs_root *root;
1016 unsigned long nr_pages;
1018 async_cow = container_of(work, struct async_cow, work);
1020 root = async_cow->root;
1021 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1024 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1026 waitqueue_active(&root->fs_info->async_submit_wait))
1027 wake_up(&root->fs_info->async_submit_wait);
1029 if (async_cow->inode)
1030 submit_compressed_extents(async_cow->inode, async_cow);
1033 static noinline void async_cow_free(struct btrfs_work *work)
1035 struct async_cow *async_cow;
1036 async_cow = container_of(work, struct async_cow, work);
1037 if (async_cow->inode)
1038 btrfs_add_delayed_iput(async_cow->inode);
1042 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1043 u64 start, u64 end, int *page_started,
1044 unsigned long *nr_written)
1046 struct async_cow *async_cow;
1047 struct btrfs_root *root = BTRFS_I(inode)->root;
1048 unsigned long nr_pages;
1050 int limit = 10 * 1024 * 1024;
1052 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1053 1, 0, NULL, GFP_NOFS);
1054 while (start < end) {
1055 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1056 BUG_ON(!async_cow); /* -ENOMEM */
1057 async_cow->inode = igrab(inode);
1058 async_cow->root = root;
1059 async_cow->locked_page = locked_page;
1060 async_cow->start = start;
1062 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1065 cur_end = min(end, start + 512 * 1024 - 1);
1067 async_cow->end = cur_end;
1068 INIT_LIST_HEAD(&async_cow->extents);
1070 async_cow->work.func = async_cow_start;
1071 async_cow->work.ordered_func = async_cow_submit;
1072 async_cow->work.ordered_free = async_cow_free;
1073 async_cow->work.flags = 0;
1075 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1077 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1079 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1082 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1083 wait_event(root->fs_info->async_submit_wait,
1084 (atomic_read(&root->fs_info->async_delalloc_pages) <
1088 while (atomic_read(&root->fs_info->async_submit_draining) &&
1089 atomic_read(&root->fs_info->async_delalloc_pages)) {
1090 wait_event(root->fs_info->async_submit_wait,
1091 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1095 *nr_written += nr_pages;
1096 start = cur_end + 1;
1102 static noinline int csum_exist_in_range(struct btrfs_root *root,
1103 u64 bytenr, u64 num_bytes)
1106 struct btrfs_ordered_sum *sums;
1109 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1110 bytenr + num_bytes - 1, &list, 0);
1111 if (ret == 0 && list_empty(&list))
1114 while (!list_empty(&list)) {
1115 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1116 list_del(&sums->list);
1123 * when nowcow writeback call back. This checks for snapshots or COW copies
1124 * of the extents that exist in the file, and COWs the file as required.
1126 * If no cow copies or snapshots exist, we write directly to the existing
1129 static noinline int run_delalloc_nocow(struct inode *inode,
1130 struct page *locked_page,
1131 u64 start, u64 end, int *page_started, int force,
1132 unsigned long *nr_written)
1134 struct btrfs_root *root = BTRFS_I(inode)->root;
1135 struct btrfs_trans_handle *trans;
1136 struct extent_buffer *leaf;
1137 struct btrfs_path *path;
1138 struct btrfs_file_extent_item *fi;
1139 struct btrfs_key found_key;
1152 u64 ino = btrfs_ino(inode);
1154 path = btrfs_alloc_path();
1156 extent_clear_unlock_delalloc(inode,
1157 &BTRFS_I(inode)->io_tree,
1158 start, end, locked_page,
1159 EXTENT_CLEAR_UNLOCK_PAGE |
1160 EXTENT_CLEAR_UNLOCK |
1161 EXTENT_CLEAR_DELALLOC |
1162 EXTENT_CLEAR_DIRTY |
1163 EXTENT_SET_WRITEBACK |
1164 EXTENT_END_WRITEBACK);
1168 nolock = btrfs_is_free_space_inode(inode);
1171 trans = btrfs_join_transaction_nolock(root);
1173 trans = btrfs_join_transaction(root);
1175 if (IS_ERR(trans)) {
1176 extent_clear_unlock_delalloc(inode,
1177 &BTRFS_I(inode)->io_tree,
1178 start, end, locked_page,
1179 EXTENT_CLEAR_UNLOCK_PAGE |
1180 EXTENT_CLEAR_UNLOCK |
1181 EXTENT_CLEAR_DELALLOC |
1182 EXTENT_CLEAR_DIRTY |
1183 EXTENT_SET_WRITEBACK |
1184 EXTENT_END_WRITEBACK);
1185 btrfs_free_path(path);
1186 return PTR_ERR(trans);
1189 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1191 cow_start = (u64)-1;
1194 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1197 btrfs_abort_transaction(trans, root, ret);
1200 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1201 leaf = path->nodes[0];
1202 btrfs_item_key_to_cpu(leaf, &found_key,
1203 path->slots[0] - 1);
1204 if (found_key.objectid == ino &&
1205 found_key.type == BTRFS_EXTENT_DATA_KEY)
1210 leaf = path->nodes[0];
1211 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1212 ret = btrfs_next_leaf(root, path);
1214 btrfs_abort_transaction(trans, root, ret);
1219 leaf = path->nodes[0];
1225 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1227 if (found_key.objectid > ino ||
1228 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1229 found_key.offset > end)
1232 if (found_key.offset > cur_offset) {
1233 extent_end = found_key.offset;
1238 fi = btrfs_item_ptr(leaf, path->slots[0],
1239 struct btrfs_file_extent_item);
1240 extent_type = btrfs_file_extent_type(leaf, fi);
1242 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1243 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1244 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1245 extent_offset = btrfs_file_extent_offset(leaf, fi);
1246 extent_end = found_key.offset +
1247 btrfs_file_extent_num_bytes(leaf, fi);
1248 if (extent_end <= start) {
1252 if (disk_bytenr == 0)
1254 if (btrfs_file_extent_compression(leaf, fi) ||
1255 btrfs_file_extent_encryption(leaf, fi) ||
1256 btrfs_file_extent_other_encoding(leaf, fi))
1258 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1260 if (btrfs_extent_readonly(root, disk_bytenr))
1262 if (btrfs_cross_ref_exist(trans, root, ino,
1264 extent_offset, disk_bytenr))
1266 disk_bytenr += extent_offset;
1267 disk_bytenr += cur_offset - found_key.offset;
1268 num_bytes = min(end + 1, extent_end) - cur_offset;
1270 * force cow if csum exists in the range.
1271 * this ensure that csum for a given extent are
1272 * either valid or do not exist.
1274 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1277 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1278 extent_end = found_key.offset +
1279 btrfs_file_extent_inline_len(leaf, fi);
1280 extent_end = ALIGN(extent_end, root->sectorsize);
1285 if (extent_end <= start) {
1290 if (cow_start == (u64)-1)
1291 cow_start = cur_offset;
1292 cur_offset = extent_end;
1293 if (cur_offset > end)
1299 btrfs_release_path(path);
1300 if (cow_start != (u64)-1) {
1301 ret = __cow_file_range(trans, inode, root, locked_page,
1302 cow_start, found_key.offset - 1,
1303 page_started, nr_written, 1);
1305 btrfs_abort_transaction(trans, root, ret);
1308 cow_start = (u64)-1;
1311 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1312 struct extent_map *em;
1313 struct extent_map_tree *em_tree;
1314 em_tree = &BTRFS_I(inode)->extent_tree;
1315 em = alloc_extent_map();
1316 BUG_ON(!em); /* -ENOMEM */
1317 em->start = cur_offset;
1318 em->orig_start = em->start;
1319 em->len = num_bytes;
1320 em->block_len = num_bytes;
1321 em->block_start = disk_bytenr;
1322 em->bdev = root->fs_info->fs_devices->latest_bdev;
1323 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1324 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
1326 write_lock(&em_tree->lock);
1327 ret = add_extent_mapping(em_tree, em);
1328 write_unlock(&em_tree->lock);
1329 if (ret != -EEXIST) {
1330 free_extent_map(em);
1333 btrfs_drop_extent_cache(inode, em->start,
1334 em->start + em->len - 1, 0);
1336 type = BTRFS_ORDERED_PREALLOC;
1338 type = BTRFS_ORDERED_NOCOW;
1341 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1342 num_bytes, num_bytes, type);
1343 BUG_ON(ret); /* -ENOMEM */
1345 if (root->root_key.objectid ==
1346 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1347 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1350 btrfs_abort_transaction(trans, root, ret);
1355 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1356 cur_offset, cur_offset + num_bytes - 1,
1357 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1358 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1359 EXTENT_SET_PRIVATE2);
1360 cur_offset = extent_end;
1361 if (cur_offset > end)
1364 btrfs_release_path(path);
1366 if (cur_offset <= end && cow_start == (u64)-1) {
1367 cow_start = cur_offset;
1371 if (cow_start != (u64)-1) {
1372 ret = __cow_file_range(trans, inode, root, locked_page,
1374 page_started, nr_written, 1);
1376 btrfs_abort_transaction(trans, root, ret);
1382 err = btrfs_end_transaction(trans, root);
1386 if (ret && cur_offset < end)
1387 extent_clear_unlock_delalloc(inode,
1388 &BTRFS_I(inode)->io_tree,
1389 cur_offset, end, locked_page,
1390 EXTENT_CLEAR_UNLOCK_PAGE |
1391 EXTENT_CLEAR_UNLOCK |
1392 EXTENT_CLEAR_DELALLOC |
1393 EXTENT_CLEAR_DIRTY |
1394 EXTENT_SET_WRITEBACK |
1395 EXTENT_END_WRITEBACK);
1397 btrfs_free_path(path);
1402 * extent_io.c call back to do delayed allocation processing
1404 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1405 u64 start, u64 end, int *page_started,
1406 unsigned long *nr_written)
1409 struct btrfs_root *root = BTRFS_I(inode)->root;
1411 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1412 ret = run_delalloc_nocow(inode, locked_page, start, end,
1413 page_started, 1, nr_written);
1414 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1415 ret = run_delalloc_nocow(inode, locked_page, start, end,
1416 page_started, 0, nr_written);
1417 } else if (!btrfs_test_opt(root, COMPRESS) &&
1418 !(BTRFS_I(inode)->force_compress) &&
1419 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1420 ret = cow_file_range(inode, locked_page, start, end,
1421 page_started, nr_written, 1);
1423 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1424 &BTRFS_I(inode)->runtime_flags);
1425 ret = cow_file_range_async(inode, locked_page, start, end,
1426 page_started, nr_written);
1431 static void btrfs_split_extent_hook(struct inode *inode,
1432 struct extent_state *orig, u64 split)
1434 /* not delalloc, ignore it */
1435 if (!(orig->state & EXTENT_DELALLOC))
1438 spin_lock(&BTRFS_I(inode)->lock);
1439 BTRFS_I(inode)->outstanding_extents++;
1440 spin_unlock(&BTRFS_I(inode)->lock);
1444 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1445 * extents so we can keep track of new extents that are just merged onto old
1446 * extents, such as when we are doing sequential writes, so we can properly
1447 * account for the metadata space we'll need.
1449 static void btrfs_merge_extent_hook(struct inode *inode,
1450 struct extent_state *new,
1451 struct extent_state *other)
1453 /* not delalloc, ignore it */
1454 if (!(other->state & EXTENT_DELALLOC))
1457 spin_lock(&BTRFS_I(inode)->lock);
1458 BTRFS_I(inode)->outstanding_extents--;
1459 spin_unlock(&BTRFS_I(inode)->lock);
1463 * extent_io.c set_bit_hook, used to track delayed allocation
1464 * bytes in this file, and to maintain the list of inodes that
1465 * have pending delalloc work to be done.
1467 static void btrfs_set_bit_hook(struct inode *inode,
1468 struct extent_state *state, int *bits)
1472 * set_bit and clear bit hooks normally require _irqsave/restore
1473 * but in this case, we are only testing for the DELALLOC
1474 * bit, which is only set or cleared with irqs on
1476 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1477 struct btrfs_root *root = BTRFS_I(inode)->root;
1478 u64 len = state->end + 1 - state->start;
1479 bool do_list = !btrfs_is_free_space_inode(inode);
1481 if (*bits & EXTENT_FIRST_DELALLOC) {
1482 *bits &= ~EXTENT_FIRST_DELALLOC;
1484 spin_lock(&BTRFS_I(inode)->lock);
1485 BTRFS_I(inode)->outstanding_extents++;
1486 spin_unlock(&BTRFS_I(inode)->lock);
1489 spin_lock(&root->fs_info->delalloc_lock);
1490 BTRFS_I(inode)->delalloc_bytes += len;
1491 root->fs_info->delalloc_bytes += len;
1492 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1493 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1494 &root->fs_info->delalloc_inodes);
1496 spin_unlock(&root->fs_info->delalloc_lock);
1501 * extent_io.c clear_bit_hook, see set_bit_hook for why
1503 static void btrfs_clear_bit_hook(struct inode *inode,
1504 struct extent_state *state, int *bits)
1507 * set_bit and clear bit hooks normally require _irqsave/restore
1508 * but in this case, we are only testing for the DELALLOC
1509 * bit, which is only set or cleared with irqs on
1511 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1512 struct btrfs_root *root = BTRFS_I(inode)->root;
1513 u64 len = state->end + 1 - state->start;
1514 bool do_list = !btrfs_is_free_space_inode(inode);
1516 if (*bits & EXTENT_FIRST_DELALLOC) {
1517 *bits &= ~EXTENT_FIRST_DELALLOC;
1518 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1519 spin_lock(&BTRFS_I(inode)->lock);
1520 BTRFS_I(inode)->outstanding_extents--;
1521 spin_unlock(&BTRFS_I(inode)->lock);
1524 if (*bits & EXTENT_DO_ACCOUNTING)
1525 btrfs_delalloc_release_metadata(inode, len);
1527 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1529 btrfs_free_reserved_data_space(inode, len);
1531 spin_lock(&root->fs_info->delalloc_lock);
1532 root->fs_info->delalloc_bytes -= len;
1533 BTRFS_I(inode)->delalloc_bytes -= len;
1535 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1536 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1537 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1539 spin_unlock(&root->fs_info->delalloc_lock);
1544 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1545 * we don't create bios that span stripes or chunks
1547 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1548 size_t size, struct bio *bio,
1549 unsigned long bio_flags)
1551 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1552 u64 logical = (u64)bio->bi_sector << 9;
1557 if (bio_flags & EXTENT_BIO_COMPRESSED)
1560 length = bio->bi_size;
1561 map_length = length;
1562 ret = btrfs_map_block(root->fs_info, READ, logical,
1563 &map_length, NULL, 0);
1564 /* Will always return 0 with map_multi == NULL */
1566 if (map_length < length + size)
1572 * in order to insert checksums into the metadata in large chunks,
1573 * we wait until bio submission time. All the pages in the bio are
1574 * checksummed and sums are attached onto the ordered extent record.
1576 * At IO completion time the cums attached on the ordered extent record
1577 * are inserted into the btree
1579 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1580 struct bio *bio, int mirror_num,
1581 unsigned long bio_flags,
1584 struct btrfs_root *root = BTRFS_I(inode)->root;
1587 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1588 BUG_ON(ret); /* -ENOMEM */
1593 * in order to insert checksums into the metadata in large chunks,
1594 * we wait until bio submission time. All the pages in the bio are
1595 * checksummed and sums are attached onto the ordered extent record.
1597 * At IO completion time the cums attached on the ordered extent record
1598 * are inserted into the btree
1600 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1601 int mirror_num, unsigned long bio_flags,
1604 struct btrfs_root *root = BTRFS_I(inode)->root;
1607 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1609 bio_endio(bio, ret);
1614 * extent_io.c submission hook. This does the right thing for csum calculation
1615 * on write, or reading the csums from the tree before a read
1617 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1618 int mirror_num, unsigned long bio_flags,
1621 struct btrfs_root *root = BTRFS_I(inode)->root;
1626 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1628 if (btrfs_is_free_space_inode(inode))
1631 if (!(rw & REQ_WRITE)) {
1632 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1636 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1637 ret = btrfs_submit_compressed_read(inode, bio,
1641 } else if (!skip_sum) {
1642 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1647 } else if (!skip_sum) {
1648 /* csum items have already been cloned */
1649 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1651 /* we're doing a write, do the async checksumming */
1652 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1653 inode, rw, bio, mirror_num,
1654 bio_flags, bio_offset,
1655 __btrfs_submit_bio_start,
1656 __btrfs_submit_bio_done);
1661 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1665 bio_endio(bio, ret);
1670 * given a list of ordered sums record them in the inode. This happens
1671 * at IO completion time based on sums calculated at bio submission time.
1673 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1674 struct inode *inode, u64 file_offset,
1675 struct list_head *list)
1677 struct btrfs_ordered_sum *sum;
1679 list_for_each_entry(sum, list, list) {
1680 btrfs_csum_file_blocks(trans,
1681 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1686 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1687 struct extent_state **cached_state)
1689 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1690 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1691 cached_state, GFP_NOFS);
1694 /* see btrfs_writepage_start_hook for details on why this is required */
1695 struct btrfs_writepage_fixup {
1697 struct btrfs_work work;
1700 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1702 struct btrfs_writepage_fixup *fixup;
1703 struct btrfs_ordered_extent *ordered;
1704 struct extent_state *cached_state = NULL;
1706 struct inode *inode;
1711 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1715 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1716 ClearPageChecked(page);
1720 inode = page->mapping->host;
1721 page_start = page_offset(page);
1722 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1724 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1727 /* already ordered? We're done */
1728 if (PagePrivate2(page))
1731 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1733 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1734 page_end, &cached_state, GFP_NOFS);
1736 btrfs_start_ordered_extent(inode, ordered, 1);
1737 btrfs_put_ordered_extent(ordered);
1741 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1743 mapping_set_error(page->mapping, ret);
1744 end_extent_writepage(page, ret, page_start, page_end);
1745 ClearPageChecked(page);
1749 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1750 ClearPageChecked(page);
1751 set_page_dirty(page);
1753 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1754 &cached_state, GFP_NOFS);
1757 page_cache_release(page);
1762 * There are a few paths in the higher layers of the kernel that directly
1763 * set the page dirty bit without asking the filesystem if it is a
1764 * good idea. This causes problems because we want to make sure COW
1765 * properly happens and the data=ordered rules are followed.
1767 * In our case any range that doesn't have the ORDERED bit set
1768 * hasn't been properly setup for IO. We kick off an async process
1769 * to fix it up. The async helper will wait for ordered extents, set
1770 * the delalloc bit and make it safe to write the page.
1772 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1774 struct inode *inode = page->mapping->host;
1775 struct btrfs_writepage_fixup *fixup;
1776 struct btrfs_root *root = BTRFS_I(inode)->root;
1778 /* this page is properly in the ordered list */
1779 if (TestClearPagePrivate2(page))
1782 if (PageChecked(page))
1785 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1789 SetPageChecked(page);
1790 page_cache_get(page);
1791 fixup->work.func = btrfs_writepage_fixup_worker;
1793 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1797 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1798 struct inode *inode, u64 file_pos,
1799 u64 disk_bytenr, u64 disk_num_bytes,
1800 u64 num_bytes, u64 ram_bytes,
1801 u8 compression, u8 encryption,
1802 u16 other_encoding, int extent_type)
1804 struct btrfs_root *root = BTRFS_I(inode)->root;
1805 struct btrfs_file_extent_item *fi;
1806 struct btrfs_path *path;
1807 struct extent_buffer *leaf;
1808 struct btrfs_key ins;
1811 path = btrfs_alloc_path();
1815 path->leave_spinning = 1;
1818 * we may be replacing one extent in the tree with another.
1819 * The new extent is pinned in the extent map, and we don't want
1820 * to drop it from the cache until it is completely in the btree.
1822 * So, tell btrfs_drop_extents to leave this extent in the cache.
1823 * the caller is expected to unpin it and allow it to be merged
1826 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1827 file_pos + num_bytes, 0);
1831 ins.objectid = btrfs_ino(inode);
1832 ins.offset = file_pos;
1833 ins.type = BTRFS_EXTENT_DATA_KEY;
1834 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1837 leaf = path->nodes[0];
1838 fi = btrfs_item_ptr(leaf, path->slots[0],
1839 struct btrfs_file_extent_item);
1840 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1841 btrfs_set_file_extent_type(leaf, fi, extent_type);
1842 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1843 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1844 btrfs_set_file_extent_offset(leaf, fi, 0);
1845 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1846 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1847 btrfs_set_file_extent_compression(leaf, fi, compression);
1848 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1849 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1851 btrfs_mark_buffer_dirty(leaf);
1852 btrfs_release_path(path);
1854 inode_add_bytes(inode, num_bytes);
1856 ins.objectid = disk_bytenr;
1857 ins.offset = disk_num_bytes;
1858 ins.type = BTRFS_EXTENT_ITEM_KEY;
1859 ret = btrfs_alloc_reserved_file_extent(trans, root,
1860 root->root_key.objectid,
1861 btrfs_ino(inode), file_pos, &ins);
1863 btrfs_free_path(path);
1869 * helper function for btrfs_finish_ordered_io, this
1870 * just reads in some of the csum leaves to prime them into ram
1871 * before we start the transaction. It limits the amount of btree
1872 * reads required while inside the transaction.
1874 /* as ordered data IO finishes, this gets called so we can finish
1875 * an ordered extent if the range of bytes in the file it covers are
1878 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
1880 struct inode *inode = ordered_extent->inode;
1881 struct btrfs_root *root = BTRFS_I(inode)->root;
1882 struct btrfs_trans_handle *trans = NULL;
1883 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1884 struct extent_state *cached_state = NULL;
1885 int compress_type = 0;
1889 nolock = btrfs_is_free_space_inode(inode);
1891 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1896 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1897 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1898 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1901 trans = btrfs_join_transaction_nolock(root);
1903 trans = btrfs_join_transaction(root);
1904 if (IS_ERR(trans)) {
1905 ret = PTR_ERR(trans);
1909 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1910 ret = btrfs_update_inode_fallback(trans, root, inode);
1911 if (ret) /* -ENOMEM or corruption */
1912 btrfs_abort_transaction(trans, root, ret);
1917 lock_extent_bits(io_tree, ordered_extent->file_offset,
1918 ordered_extent->file_offset + ordered_extent->len - 1,
1922 trans = btrfs_join_transaction_nolock(root);
1924 trans = btrfs_join_transaction(root);
1925 if (IS_ERR(trans)) {
1926 ret = PTR_ERR(trans);
1930 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1932 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1933 compress_type = ordered_extent->compress_type;
1934 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1935 BUG_ON(compress_type);
1936 ret = btrfs_mark_extent_written(trans, inode,
1937 ordered_extent->file_offset,
1938 ordered_extent->file_offset +
1939 ordered_extent->len);
1941 BUG_ON(root == root->fs_info->tree_root);
1942 ret = insert_reserved_file_extent(trans, inode,
1943 ordered_extent->file_offset,
1944 ordered_extent->start,
1945 ordered_extent->disk_len,
1946 ordered_extent->len,
1947 ordered_extent->len,
1948 compress_type, 0, 0,
1949 BTRFS_FILE_EXTENT_REG);
1951 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1952 ordered_extent->file_offset, ordered_extent->len,
1955 btrfs_abort_transaction(trans, root, ret);
1959 add_pending_csums(trans, inode, ordered_extent->file_offset,
1960 &ordered_extent->list);
1962 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1963 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1964 ret = btrfs_update_inode_fallback(trans, root, inode);
1965 if (ret) { /* -ENOMEM or corruption */
1966 btrfs_abort_transaction(trans, root, ret);
1970 btrfs_set_inode_last_trans(trans, inode);
1974 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1975 ordered_extent->file_offset +
1976 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1978 if (root != root->fs_info->tree_root)
1979 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1981 btrfs_end_transaction(trans, root);
1984 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
1985 ordered_extent->file_offset +
1986 ordered_extent->len - 1, NULL, GFP_NOFS);
1989 * This needs to be done to make sure anybody waiting knows we are done
1990 * updating everything for this ordered extent.
1992 btrfs_remove_ordered_extent(inode, ordered_extent);
1995 btrfs_put_ordered_extent(ordered_extent);
1996 /* once for the tree */
1997 btrfs_put_ordered_extent(ordered_extent);
2002 static void finish_ordered_fn(struct btrfs_work *work)
2004 struct btrfs_ordered_extent *ordered_extent;
2005 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2006 btrfs_finish_ordered_io(ordered_extent);
2009 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2010 struct extent_state *state, int uptodate)
2012 struct inode *inode = page->mapping->host;
2013 struct btrfs_root *root = BTRFS_I(inode)->root;
2014 struct btrfs_ordered_extent *ordered_extent = NULL;
2015 struct btrfs_workers *workers;
2017 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2019 ClearPagePrivate2(page);
2020 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2021 end - start + 1, uptodate))
2024 ordered_extent->work.func = finish_ordered_fn;
2025 ordered_extent->work.flags = 0;
2027 if (btrfs_is_free_space_inode(inode))
2028 workers = &root->fs_info->endio_freespace_worker;
2030 workers = &root->fs_info->endio_write_workers;
2031 btrfs_queue_worker(workers, &ordered_extent->work);
2037 * when reads are done, we need to check csums to verify the data is correct
2038 * if there's a match, we allow the bio to finish. If not, the code in
2039 * extent_io.c will try to find good copies for us.
2041 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2042 struct extent_state *state, int mirror)
2044 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
2045 struct inode *inode = page->mapping->host;
2046 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2048 u64 private = ~(u32)0;
2050 struct btrfs_root *root = BTRFS_I(inode)->root;
2053 if (PageChecked(page)) {
2054 ClearPageChecked(page);
2058 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2061 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2062 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2063 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2068 if (state && state->start == start) {
2069 private = state->private;
2072 ret = get_state_private(io_tree, start, &private);
2074 kaddr = kmap_atomic(page);
2078 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
2079 btrfs_csum_final(csum, (char *)&csum);
2080 if (csum != private)
2083 kunmap_atomic(kaddr);
2088 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2090 (unsigned long long)btrfs_ino(page->mapping->host),
2091 (unsigned long long)start, csum,
2092 (unsigned long long)private);
2093 memset(kaddr + offset, 1, end - start + 1);
2094 flush_dcache_page(page);
2095 kunmap_atomic(kaddr);
2101 struct delayed_iput {
2102 struct list_head list;
2103 struct inode *inode;
2106 /* JDM: If this is fs-wide, why can't we add a pointer to
2107 * btrfs_inode instead and avoid the allocation? */
2108 void btrfs_add_delayed_iput(struct inode *inode)
2110 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2111 struct delayed_iput *delayed;
2113 if (atomic_add_unless(&inode->i_count, -1, 1))
2116 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2117 delayed->inode = inode;
2119 spin_lock(&fs_info->delayed_iput_lock);
2120 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2121 spin_unlock(&fs_info->delayed_iput_lock);
2124 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2127 struct btrfs_fs_info *fs_info = root->fs_info;
2128 struct delayed_iput *delayed;
2131 spin_lock(&fs_info->delayed_iput_lock);
2132 empty = list_empty(&fs_info->delayed_iputs);
2133 spin_unlock(&fs_info->delayed_iput_lock);
2137 spin_lock(&fs_info->delayed_iput_lock);
2138 list_splice_init(&fs_info->delayed_iputs, &list);
2139 spin_unlock(&fs_info->delayed_iput_lock);
2141 while (!list_empty(&list)) {
2142 delayed = list_entry(list.next, struct delayed_iput, list);
2143 list_del(&delayed->list);
2144 iput(delayed->inode);
2149 enum btrfs_orphan_cleanup_state {
2150 ORPHAN_CLEANUP_STARTED = 1,
2151 ORPHAN_CLEANUP_DONE = 2,
2155 * This is called in transaction commit time. If there are no orphan
2156 * files in the subvolume, it removes orphan item and frees block_rsv
2159 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2160 struct btrfs_root *root)
2162 struct btrfs_block_rsv *block_rsv;
2165 if (atomic_read(&root->orphan_inodes) ||
2166 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2169 spin_lock(&root->orphan_lock);
2170 if (atomic_read(&root->orphan_inodes)) {
2171 spin_unlock(&root->orphan_lock);
2175 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2176 spin_unlock(&root->orphan_lock);
2180 block_rsv = root->orphan_block_rsv;
2181 root->orphan_block_rsv = NULL;
2182 spin_unlock(&root->orphan_lock);
2184 if (root->orphan_item_inserted &&
2185 btrfs_root_refs(&root->root_item) > 0) {
2186 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2187 root->root_key.objectid);
2189 root->orphan_item_inserted = 0;
2193 WARN_ON(block_rsv->size > 0);
2194 btrfs_free_block_rsv(root, block_rsv);
2199 * This creates an orphan entry for the given inode in case something goes
2200 * wrong in the middle of an unlink/truncate.
2202 * NOTE: caller of this function should reserve 5 units of metadata for
2205 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2207 struct btrfs_root *root = BTRFS_I(inode)->root;
2208 struct btrfs_block_rsv *block_rsv = NULL;
2213 if (!root->orphan_block_rsv) {
2214 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2219 spin_lock(&root->orphan_lock);
2220 if (!root->orphan_block_rsv) {
2221 root->orphan_block_rsv = block_rsv;
2222 } else if (block_rsv) {
2223 btrfs_free_block_rsv(root, block_rsv);
2227 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2228 &BTRFS_I(inode)->runtime_flags)) {
2231 * For proper ENOSPC handling, we should do orphan
2232 * cleanup when mounting. But this introduces backward
2233 * compatibility issue.
2235 if (!xchg(&root->orphan_item_inserted, 1))
2241 atomic_inc(&root->orphan_inodes);
2244 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2245 &BTRFS_I(inode)->runtime_flags))
2247 spin_unlock(&root->orphan_lock);
2249 /* grab metadata reservation from transaction handle */
2251 ret = btrfs_orphan_reserve_metadata(trans, inode);
2252 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2255 /* insert an orphan item to track this unlinked/truncated file */
2257 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2258 if (ret && ret != -EEXIST) {
2259 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2260 &BTRFS_I(inode)->runtime_flags);
2261 btrfs_abort_transaction(trans, root, ret);
2267 /* insert an orphan item to track subvolume contains orphan files */
2269 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2270 root->root_key.objectid);
2271 if (ret && ret != -EEXIST) {
2272 btrfs_abort_transaction(trans, root, ret);
2280 * We have done the truncate/delete so we can go ahead and remove the orphan
2281 * item for this particular inode.
2283 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2285 struct btrfs_root *root = BTRFS_I(inode)->root;
2286 int delete_item = 0;
2287 int release_rsv = 0;
2290 spin_lock(&root->orphan_lock);
2291 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2292 &BTRFS_I(inode)->runtime_flags))
2295 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2296 &BTRFS_I(inode)->runtime_flags))
2298 spin_unlock(&root->orphan_lock);
2300 if (trans && delete_item) {
2301 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2302 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2306 btrfs_orphan_release_metadata(inode);
2307 atomic_dec(&root->orphan_inodes);
2314 * this cleans up any orphans that may be left on the list from the last use
2317 int btrfs_orphan_cleanup(struct btrfs_root *root)
2319 struct btrfs_path *path;
2320 struct extent_buffer *leaf;
2321 struct btrfs_key key, found_key;
2322 struct btrfs_trans_handle *trans;
2323 struct inode *inode;
2324 u64 last_objectid = 0;
2325 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2327 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2330 path = btrfs_alloc_path();
2337 key.objectid = BTRFS_ORPHAN_OBJECTID;
2338 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2339 key.offset = (u64)-1;
2342 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2347 * if ret == 0 means we found what we were searching for, which
2348 * is weird, but possible, so only screw with path if we didn't
2349 * find the key and see if we have stuff that matches
2353 if (path->slots[0] == 0)
2358 /* pull out the item */
2359 leaf = path->nodes[0];
2360 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2362 /* make sure the item matches what we want */
2363 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2365 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2368 /* release the path since we're done with it */
2369 btrfs_release_path(path);
2372 * this is where we are basically btrfs_lookup, without the
2373 * crossing root thing. we store the inode number in the
2374 * offset of the orphan item.
2377 if (found_key.offset == last_objectid) {
2378 printk(KERN_ERR "btrfs: Error removing orphan entry, "
2379 "stopping orphan cleanup\n");
2384 last_objectid = found_key.offset;
2386 found_key.objectid = found_key.offset;
2387 found_key.type = BTRFS_INODE_ITEM_KEY;
2388 found_key.offset = 0;
2389 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2390 ret = PTR_RET(inode);
2391 if (ret && ret != -ESTALE)
2394 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2395 struct btrfs_root *dead_root;
2396 struct btrfs_fs_info *fs_info = root->fs_info;
2397 int is_dead_root = 0;
2400 * this is an orphan in the tree root. Currently these
2401 * could come from 2 sources:
2402 * a) a snapshot deletion in progress
2403 * b) a free space cache inode
2404 * We need to distinguish those two, as the snapshot
2405 * orphan must not get deleted.
2406 * find_dead_roots already ran before us, so if this
2407 * is a snapshot deletion, we should find the root
2408 * in the dead_roots list
2410 spin_lock(&fs_info->trans_lock);
2411 list_for_each_entry(dead_root, &fs_info->dead_roots,
2413 if (dead_root->root_key.objectid ==
2414 found_key.objectid) {
2419 spin_unlock(&fs_info->trans_lock);
2421 /* prevent this orphan from being found again */
2422 key.offset = found_key.objectid - 1;
2427 * Inode is already gone but the orphan item is still there,
2428 * kill the orphan item.
2430 if (ret == -ESTALE) {
2431 trans = btrfs_start_transaction(root, 1);
2432 if (IS_ERR(trans)) {
2433 ret = PTR_ERR(trans);
2436 printk(KERN_ERR "auto deleting %Lu\n",
2437 found_key.objectid);
2438 ret = btrfs_del_orphan_item(trans, root,
2439 found_key.objectid);
2440 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2441 btrfs_end_transaction(trans, root);
2446 * add this inode to the orphan list so btrfs_orphan_del does
2447 * the proper thing when we hit it
2449 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2450 &BTRFS_I(inode)->runtime_flags);
2452 /* if we have links, this was a truncate, lets do that */
2453 if (inode->i_nlink) {
2454 if (!S_ISREG(inode->i_mode)) {
2460 ret = btrfs_truncate(inode);
2465 /* this will do delete_inode and everything for us */
2470 /* release the path since we're done with it */
2471 btrfs_release_path(path);
2473 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2475 if (root->orphan_block_rsv)
2476 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2479 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2480 trans = btrfs_join_transaction(root);
2482 btrfs_end_transaction(trans, root);
2486 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2488 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2492 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2493 btrfs_free_path(path);
2498 * very simple check to peek ahead in the leaf looking for xattrs. If we
2499 * don't find any xattrs, we know there can't be any acls.
2501 * slot is the slot the inode is in, objectid is the objectid of the inode
2503 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2504 int slot, u64 objectid)
2506 u32 nritems = btrfs_header_nritems(leaf);
2507 struct btrfs_key found_key;
2511 while (slot < nritems) {
2512 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2514 /* we found a different objectid, there must not be acls */
2515 if (found_key.objectid != objectid)
2518 /* we found an xattr, assume we've got an acl */
2519 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2523 * we found a key greater than an xattr key, there can't
2524 * be any acls later on
2526 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2533 * it goes inode, inode backrefs, xattrs, extents,
2534 * so if there are a ton of hard links to an inode there can
2535 * be a lot of backrefs. Don't waste time searching too hard,
2536 * this is just an optimization
2541 /* we hit the end of the leaf before we found an xattr or
2542 * something larger than an xattr. We have to assume the inode
2549 * read an inode from the btree into the in-memory inode
2551 static void btrfs_read_locked_inode(struct inode *inode)
2553 struct btrfs_path *path;
2554 struct extent_buffer *leaf;
2555 struct btrfs_inode_item *inode_item;
2556 struct btrfs_timespec *tspec;
2557 struct btrfs_root *root = BTRFS_I(inode)->root;
2558 struct btrfs_key location;
2562 bool filled = false;
2564 ret = btrfs_fill_inode(inode, &rdev);
2568 path = btrfs_alloc_path();
2572 path->leave_spinning = 1;
2573 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2575 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2579 leaf = path->nodes[0];
2584 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2585 struct btrfs_inode_item);
2586 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2587 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2588 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
2589 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
2590 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2592 tspec = btrfs_inode_atime(inode_item);
2593 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2594 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2596 tspec = btrfs_inode_mtime(inode_item);
2597 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2598 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2600 tspec = btrfs_inode_ctime(inode_item);
2601 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2602 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2604 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2605 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2606 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
2609 * If we were modified in the current generation and evicted from memory
2610 * and then re-read we need to do a full sync since we don't have any
2611 * idea about which extents were modified before we were evicted from
2614 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
2615 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2616 &BTRFS_I(inode)->runtime_flags);
2618 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2619 inode->i_generation = BTRFS_I(inode)->generation;
2621 rdev = btrfs_inode_rdev(leaf, inode_item);
2623 BTRFS_I(inode)->index_cnt = (u64)-1;
2624 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2627 * try to precache a NULL acl entry for files that don't have
2628 * any xattrs or acls
2630 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2633 cache_no_acl(inode);
2635 btrfs_free_path(path);
2637 switch (inode->i_mode & S_IFMT) {
2639 inode->i_mapping->a_ops = &btrfs_aops;
2640 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2641 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2642 inode->i_fop = &btrfs_file_operations;
2643 inode->i_op = &btrfs_file_inode_operations;
2646 inode->i_fop = &btrfs_dir_file_operations;
2647 if (root == root->fs_info->tree_root)
2648 inode->i_op = &btrfs_dir_ro_inode_operations;
2650 inode->i_op = &btrfs_dir_inode_operations;
2653 inode->i_op = &btrfs_symlink_inode_operations;
2654 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2655 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2658 inode->i_op = &btrfs_special_inode_operations;
2659 init_special_inode(inode, inode->i_mode, rdev);
2663 btrfs_update_iflags(inode);
2667 btrfs_free_path(path);
2668 make_bad_inode(inode);
2672 * given a leaf and an inode, copy the inode fields into the leaf
2674 static void fill_inode_item(struct btrfs_trans_handle *trans,
2675 struct extent_buffer *leaf,
2676 struct btrfs_inode_item *item,
2677 struct inode *inode)
2679 btrfs_set_inode_uid(leaf, item, i_uid_read(inode));
2680 btrfs_set_inode_gid(leaf, item, i_gid_read(inode));
2681 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2682 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2683 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2685 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2686 inode->i_atime.tv_sec);
2687 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2688 inode->i_atime.tv_nsec);
2690 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2691 inode->i_mtime.tv_sec);
2692 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2693 inode->i_mtime.tv_nsec);
2695 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2696 inode->i_ctime.tv_sec);
2697 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2698 inode->i_ctime.tv_nsec);
2700 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2701 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2702 btrfs_set_inode_sequence(leaf, item, inode->i_version);
2703 btrfs_set_inode_transid(leaf, item, trans->transid);
2704 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2705 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2706 btrfs_set_inode_block_group(leaf, item, 0);
2710 * copy everything in the in-memory inode into the btree.
2712 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2713 struct btrfs_root *root, struct inode *inode)
2715 struct btrfs_inode_item *inode_item;
2716 struct btrfs_path *path;
2717 struct extent_buffer *leaf;
2720 path = btrfs_alloc_path();
2724 path->leave_spinning = 1;
2725 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2733 btrfs_unlock_up_safe(path, 1);
2734 leaf = path->nodes[0];
2735 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2736 struct btrfs_inode_item);
2738 fill_inode_item(trans, leaf, inode_item, inode);
2739 btrfs_mark_buffer_dirty(leaf);
2740 btrfs_set_inode_last_trans(trans, inode);
2743 btrfs_free_path(path);
2748 * copy everything in the in-memory inode into the btree.
2750 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2751 struct btrfs_root *root, struct inode *inode)
2756 * If the inode is a free space inode, we can deadlock during commit
2757 * if we put it into the delayed code.
2759 * The data relocation inode should also be directly updated
2762 if (!btrfs_is_free_space_inode(inode)
2763 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2764 btrfs_update_root_times(trans, root);
2766 ret = btrfs_delayed_update_inode(trans, root, inode);
2768 btrfs_set_inode_last_trans(trans, inode);
2772 return btrfs_update_inode_item(trans, root, inode);
2775 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2776 struct btrfs_root *root,
2777 struct inode *inode)
2781 ret = btrfs_update_inode(trans, root, inode);
2783 return btrfs_update_inode_item(trans, root, inode);
2788 * unlink helper that gets used here in inode.c and in the tree logging
2789 * recovery code. It remove a link in a directory with a given name, and
2790 * also drops the back refs in the inode to the directory
2792 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2793 struct btrfs_root *root,
2794 struct inode *dir, struct inode *inode,
2795 const char *name, int name_len)
2797 struct btrfs_path *path;
2799 struct extent_buffer *leaf;
2800 struct btrfs_dir_item *di;
2801 struct btrfs_key key;
2803 u64 ino = btrfs_ino(inode);
2804 u64 dir_ino = btrfs_ino(dir);
2806 path = btrfs_alloc_path();
2812 path->leave_spinning = 1;
2813 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2814 name, name_len, -1);
2823 leaf = path->nodes[0];
2824 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2825 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2828 btrfs_release_path(path);
2830 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2833 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2834 "inode %llu parent %llu\n", name_len, name,
2835 (unsigned long long)ino, (unsigned long long)dir_ino);
2836 btrfs_abort_transaction(trans, root, ret);
2840 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2842 btrfs_abort_transaction(trans, root, ret);
2846 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2848 if (ret != 0 && ret != -ENOENT) {
2849 btrfs_abort_transaction(trans, root, ret);
2853 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2858 btrfs_free_path(path);
2862 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2863 inode_inc_iversion(inode);
2864 inode_inc_iversion(dir);
2865 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2866 ret = btrfs_update_inode(trans, root, dir);
2871 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2872 struct btrfs_root *root,
2873 struct inode *dir, struct inode *inode,
2874 const char *name, int name_len)
2877 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2879 btrfs_drop_nlink(inode);
2880 ret = btrfs_update_inode(trans, root, inode);
2886 /* helper to check if there is any shared block in the path */
2887 static int check_path_shared(struct btrfs_root *root,
2888 struct btrfs_path *path)
2890 struct extent_buffer *eb;
2894 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2897 if (!path->nodes[level])
2899 eb = path->nodes[level];
2900 if (!btrfs_block_can_be_shared(root, eb))
2902 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2911 * helper to start transaction for unlink and rmdir.
2913 * unlink and rmdir are special in btrfs, they do not always free space.
2914 * so in enospc case, we should make sure they will free space before
2915 * allowing them to use the global metadata reservation.
2917 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2918 struct dentry *dentry)
2920 struct btrfs_trans_handle *trans;
2921 struct btrfs_root *root = BTRFS_I(dir)->root;
2922 struct btrfs_path *path;
2923 struct btrfs_dir_item *di;
2924 struct inode *inode = dentry->d_inode;
2929 u64 ino = btrfs_ino(inode);
2930 u64 dir_ino = btrfs_ino(dir);
2933 * 1 for the possible orphan item
2934 * 1 for the dir item
2935 * 1 for the dir index
2936 * 1 for the inode ref
2937 * 1 for the inode ref in the tree log
2938 * 2 for the dir entries in the log
2941 trans = btrfs_start_transaction(root, 8);
2942 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2945 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2946 return ERR_PTR(-ENOSPC);
2948 /* check if there is someone else holds reference */
2949 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2950 return ERR_PTR(-ENOSPC);
2952 if (atomic_read(&inode->i_count) > 2)
2953 return ERR_PTR(-ENOSPC);
2955 if (xchg(&root->fs_info->enospc_unlink, 1))
2956 return ERR_PTR(-ENOSPC);
2958 path = btrfs_alloc_path();
2960 root->fs_info->enospc_unlink = 0;
2961 return ERR_PTR(-ENOMEM);
2964 /* 1 for the orphan item */
2965 trans = btrfs_start_transaction(root, 1);
2966 if (IS_ERR(trans)) {
2967 btrfs_free_path(path);
2968 root->fs_info->enospc_unlink = 0;
2972 path->skip_locking = 1;
2973 path->search_commit_root = 1;
2975 ret = btrfs_lookup_inode(trans, root, path,
2976 &BTRFS_I(dir)->location, 0);
2982 if (check_path_shared(root, path))
2987 btrfs_release_path(path);
2989 ret = btrfs_lookup_inode(trans, root, path,
2990 &BTRFS_I(inode)->location, 0);
2996 if (check_path_shared(root, path))
3001 btrfs_release_path(path);
3003 if (ret == 0 && S_ISREG(inode->i_mode)) {
3004 ret = btrfs_lookup_file_extent(trans, root, path,
3010 BUG_ON(ret == 0); /* Corruption */
3011 if (check_path_shared(root, path))
3013 btrfs_release_path(path);
3021 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3022 dentry->d_name.name, dentry->d_name.len, 0);
3028 if (check_path_shared(root, path))
3034 btrfs_release_path(path);
3036 ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3037 dentry->d_name.len, ino, dir_ino, 0,
3044 if (check_path_shared(root, path))
3047 btrfs_release_path(path);
3050 * This is a commit root search, if we can lookup inode item and other
3051 * relative items in the commit root, it means the transaction of
3052 * dir/file creation has been committed, and the dir index item that we
3053 * delay to insert has also been inserted into the commit root. So
3054 * we needn't worry about the delayed insertion of the dir index item
3057 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3058 dentry->d_name.name, dentry->d_name.len, 0);
3063 BUG_ON(ret == -ENOENT);
3064 if (check_path_shared(root, path))
3069 btrfs_free_path(path);
3070 /* Migrate the orphan reservation over */
3072 err = btrfs_block_rsv_migrate(trans->block_rsv,
3073 &root->fs_info->global_block_rsv,
3074 trans->bytes_reserved);
3077 btrfs_end_transaction(trans, root);
3078 root->fs_info->enospc_unlink = 0;
3079 return ERR_PTR(err);
3082 trans->block_rsv = &root->fs_info->global_block_rsv;
3086 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3087 struct btrfs_root *root)
3089 if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3090 btrfs_block_rsv_release(root, trans->block_rsv,
3091 trans->bytes_reserved);
3092 trans->block_rsv = &root->fs_info->trans_block_rsv;
3093 BUG_ON(!root->fs_info->enospc_unlink);
3094 root->fs_info->enospc_unlink = 0;
3096 btrfs_end_transaction(trans, root);
3099 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3101 struct btrfs_root *root = BTRFS_I(dir)->root;
3102 struct btrfs_trans_handle *trans;
3103 struct inode *inode = dentry->d_inode;
3106 trans = __unlink_start_trans(dir, dentry);
3108 return PTR_ERR(trans);
3110 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3112 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3113 dentry->d_name.name, dentry->d_name.len);
3117 if (inode->i_nlink == 0) {
3118 ret = btrfs_orphan_add(trans, inode);
3124 __unlink_end_trans(trans, root);
3125 btrfs_btree_balance_dirty(root);
3129 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3130 struct btrfs_root *root,
3131 struct inode *dir, u64 objectid,
3132 const char *name, int name_len)
3134 struct btrfs_path *path;
3135 struct extent_buffer *leaf;
3136 struct btrfs_dir_item *di;
3137 struct btrfs_key key;
3140 u64 dir_ino = btrfs_ino(dir);
3142 path = btrfs_alloc_path();
3146 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3147 name, name_len, -1);
3148 if (IS_ERR_OR_NULL(di)) {
3156 leaf = path->nodes[0];
3157 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3158 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3159 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3161 btrfs_abort_transaction(trans, root, ret);
3164 btrfs_release_path(path);
3166 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3167 objectid, root->root_key.objectid,
3168 dir_ino, &index, name, name_len);
3170 if (ret != -ENOENT) {
3171 btrfs_abort_transaction(trans, root, ret);
3174 di = btrfs_search_dir_index_item(root, path, dir_ino,
3176 if (IS_ERR_OR_NULL(di)) {
3181 btrfs_abort_transaction(trans, root, ret);
3185 leaf = path->nodes[0];
3186 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3187 btrfs_release_path(path);
3190 btrfs_release_path(path);
3192 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3194 btrfs_abort_transaction(trans, root, ret);
3198 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3199 inode_inc_iversion(dir);
3200 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3201 ret = btrfs_update_inode_fallback(trans, root, dir);
3203 btrfs_abort_transaction(trans, root, ret);
3205 btrfs_free_path(path);
3209 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3211 struct inode *inode = dentry->d_inode;
3213 struct btrfs_root *root = BTRFS_I(dir)->root;
3214 struct btrfs_trans_handle *trans;
3216 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3218 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3221 trans = __unlink_start_trans(dir, dentry);
3223 return PTR_ERR(trans);
3225 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3226 err = btrfs_unlink_subvol(trans, root, dir,
3227 BTRFS_I(inode)->location.objectid,
3228 dentry->d_name.name,
3229 dentry->d_name.len);
3233 err = btrfs_orphan_add(trans, inode);
3237 /* now the directory is empty */
3238 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3239 dentry->d_name.name, dentry->d_name.len);
3241 btrfs_i_size_write(inode, 0);
3243 __unlink_end_trans(trans, root);
3244 btrfs_btree_balance_dirty(root);
3250 * this can truncate away extent items, csum items and directory items.
3251 * It starts at a high offset and removes keys until it can't find
3252 * any higher than new_size
3254 * csum items that cross the new i_size are truncated to the new size
3257 * min_type is the minimum key type to truncate down to. If set to 0, this
3258 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3260 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3261 struct btrfs_root *root,
3262 struct inode *inode,
3263 u64 new_size, u32 min_type)
3265 struct btrfs_path *path;
3266 struct extent_buffer *leaf;
3267 struct btrfs_file_extent_item *fi;
3268 struct btrfs_key key;
3269 struct btrfs_key found_key;
3270 u64 extent_start = 0;
3271 u64 extent_num_bytes = 0;
3272 u64 extent_offset = 0;
3274 u64 mask = root->sectorsize - 1;
3275 u32 found_type = (u8)-1;
3278 int pending_del_nr = 0;
3279 int pending_del_slot = 0;
3280 int extent_type = -1;
3283 u64 ino = btrfs_ino(inode);
3285 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3287 path = btrfs_alloc_path();
3293 * We want to drop from the next block forward in case this new size is
3294 * not block aligned since we will be keeping the last block of the
3295 * extent just the way it is.
3297 if (root->ref_cows || root == root->fs_info->tree_root)
3298 btrfs_drop_extent_cache(inode, (new_size + mask) & (~mask), (u64)-1, 0);
3301 * This function is also used to drop the items in the log tree before
3302 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3303 * it is used to drop the loged items. So we shouldn't kill the delayed
3306 if (min_type == 0 && root == BTRFS_I(inode)->root)
3307 btrfs_kill_delayed_inode_items(inode);
3310 key.offset = (u64)-1;
3314 path->leave_spinning = 1;
3315 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3322 /* there are no items in the tree for us to truncate, we're
3325 if (path->slots[0] == 0)
3332 leaf = path->nodes[0];
3333 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3334 found_type = btrfs_key_type(&found_key);
3336 if (found_key.objectid != ino)
3339 if (found_type < min_type)
3342 item_end = found_key.offset;
3343 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3344 fi = btrfs_item_ptr(leaf, path->slots[0],
3345 struct btrfs_file_extent_item);
3346 extent_type = btrfs_file_extent_type(leaf, fi);
3347 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3349 btrfs_file_extent_num_bytes(leaf, fi);
3350 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3351 item_end += btrfs_file_extent_inline_len(leaf,
3356 if (found_type > min_type) {
3359 if (item_end < new_size)
3361 if (found_key.offset >= new_size)
3367 /* FIXME, shrink the extent if the ref count is only 1 */
3368 if (found_type != BTRFS_EXTENT_DATA_KEY)
3371 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3373 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3375 u64 orig_num_bytes =
3376 btrfs_file_extent_num_bytes(leaf, fi);
3377 extent_num_bytes = new_size -
3378 found_key.offset + root->sectorsize - 1;
3379 extent_num_bytes = extent_num_bytes &
3380 ~((u64)root->sectorsize - 1);
3381 btrfs_set_file_extent_num_bytes(leaf, fi,
3383 num_dec = (orig_num_bytes -
3385 if (root->ref_cows && extent_start != 0)
3386 inode_sub_bytes(inode, num_dec);
3387 btrfs_mark_buffer_dirty(leaf);
3390 btrfs_file_extent_disk_num_bytes(leaf,
3392 extent_offset = found_key.offset -
3393 btrfs_file_extent_offset(leaf, fi);
3395 /* FIXME blocksize != 4096 */
3396 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3397 if (extent_start != 0) {
3400 inode_sub_bytes(inode, num_dec);
3403 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3405 * we can't truncate inline items that have had
3409 btrfs_file_extent_compression(leaf, fi) == 0 &&
3410 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3411 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3412 u32 size = new_size - found_key.offset;
3414 if (root->ref_cows) {
3415 inode_sub_bytes(inode, item_end + 1 -
3419 btrfs_file_extent_calc_inline_size(size);
3420 btrfs_truncate_item(trans, root, path,
3422 } else if (root->ref_cows) {
3423 inode_sub_bytes(inode, item_end + 1 -
3429 if (!pending_del_nr) {
3430 /* no pending yet, add ourselves */
3431 pending_del_slot = path->slots[0];
3433 } else if (pending_del_nr &&
3434 path->slots[0] + 1 == pending_del_slot) {
3435 /* hop on the pending chunk */
3437 pending_del_slot = path->slots[0];
3444 if (found_extent && (root->ref_cows ||
3445 root == root->fs_info->tree_root)) {
3446 btrfs_set_path_blocking(path);
3447 ret = btrfs_free_extent(trans, root, extent_start,
3448 extent_num_bytes, 0,
3449 btrfs_header_owner(leaf),
3450 ino, extent_offset, 0);
3454 if (found_type == BTRFS_INODE_ITEM_KEY)
3457 if (path->slots[0] == 0 ||
3458 path->slots[0] != pending_del_slot) {
3459 if (pending_del_nr) {
3460 ret = btrfs_del_items(trans, root, path,
3464 btrfs_abort_transaction(trans,
3470 btrfs_release_path(path);
3477 if (pending_del_nr) {
3478 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3481 btrfs_abort_transaction(trans, root, ret);
3484 btrfs_free_path(path);
3489 * btrfs_truncate_page - read, zero a chunk and write a page
3490 * @inode - inode that we're zeroing
3491 * @from - the offset to start zeroing
3492 * @len - the length to zero, 0 to zero the entire range respective to the
3494 * @front - zero up to the offset instead of from the offset on
3496 * This will find the page for the "from" offset and cow the page and zero the
3497 * part we want to zero. This is used with truncate and hole punching.
3499 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
3502 struct address_space *mapping = inode->i_mapping;
3503 struct btrfs_root *root = BTRFS_I(inode)->root;
3504 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3505 struct btrfs_ordered_extent *ordered;
3506 struct extent_state *cached_state = NULL;
3508 u32 blocksize = root->sectorsize;
3509 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3510 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3512 gfp_t mask = btrfs_alloc_write_mask(mapping);
3517 if ((offset & (blocksize - 1)) == 0 &&
3518 (!len || ((len & (blocksize - 1)) == 0)))
3520 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3526 page = find_or_create_page(mapping, index, mask);
3528 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3532 page_start = page_offset(page);
3533 page_end = page_start + PAGE_CACHE_SIZE - 1;
3535 if (!PageUptodate(page)) {
3536 ret = btrfs_readpage(NULL, page);
3538 if (page->mapping != mapping) {
3540 page_cache_release(page);
3543 if (!PageUptodate(page)) {
3548 wait_on_page_writeback(page);
3550 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3551 set_page_extent_mapped(page);
3553 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3555 unlock_extent_cached(io_tree, page_start, page_end,
3556 &cached_state, GFP_NOFS);
3558 page_cache_release(page);
3559 btrfs_start_ordered_extent(inode, ordered, 1);
3560 btrfs_put_ordered_extent(ordered);
3564 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3565 EXTENT_DIRTY | EXTENT_DELALLOC |
3566 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
3567 0, 0, &cached_state, GFP_NOFS);
3569 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3572 unlock_extent_cached(io_tree, page_start, page_end,
3573 &cached_state, GFP_NOFS);
3578 if (offset != PAGE_CACHE_SIZE) {
3580 len = PAGE_CACHE_SIZE - offset;
3583 memset(kaddr, 0, offset);
3585 memset(kaddr + offset, 0, len);
3586 flush_dcache_page(page);
3589 ClearPageChecked(page);
3590 set_page_dirty(page);
3591 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3596 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3598 page_cache_release(page);
3604 * This function puts in dummy file extents for the area we're creating a hole
3605 * for. So if we are truncating this file to a larger size we need to insert
3606 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3607 * the range between oldsize and size
3609 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3611 struct btrfs_trans_handle *trans;
3612 struct btrfs_root *root = BTRFS_I(inode)->root;
3613 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3614 struct extent_map *em = NULL;
3615 struct extent_state *cached_state = NULL;
3616 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3617 u64 mask = root->sectorsize - 1;
3618 u64 hole_start = (oldsize + mask) & ~mask;
3619 u64 block_end = (size + mask) & ~mask;
3625 if (size <= hole_start)
3629 struct btrfs_ordered_extent *ordered;
3630 btrfs_wait_ordered_range(inode, hole_start,
3631 block_end - hole_start);
3632 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3634 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3637 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3638 &cached_state, GFP_NOFS);
3639 btrfs_put_ordered_extent(ordered);
3642 cur_offset = hole_start;
3644 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3645 block_end - cur_offset, 0);
3650 last_byte = min(extent_map_end(em), block_end);
3651 last_byte = (last_byte + mask) & ~mask;
3652 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3653 struct extent_map *hole_em;
3654 hole_size = last_byte - cur_offset;
3656 trans = btrfs_start_transaction(root, 3);
3657 if (IS_ERR(trans)) {
3658 err = PTR_ERR(trans);
3662 err = btrfs_drop_extents(trans, root, inode,
3664 cur_offset + hole_size, 1);
3666 btrfs_abort_transaction(trans, root, err);
3667 btrfs_end_transaction(trans, root);
3671 err = btrfs_insert_file_extent(trans, root,
3672 btrfs_ino(inode), cur_offset, 0,
3673 0, hole_size, 0, hole_size,
3676 btrfs_abort_transaction(trans, root, err);
3677 btrfs_end_transaction(trans, root);
3681 btrfs_drop_extent_cache(inode, cur_offset,
3682 cur_offset + hole_size - 1, 0);
3683 hole_em = alloc_extent_map();
3685 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3686 &BTRFS_I(inode)->runtime_flags);
3689 hole_em->start = cur_offset;
3690 hole_em->len = hole_size;
3691 hole_em->orig_start = cur_offset;
3693 hole_em->block_start = EXTENT_MAP_HOLE;
3694 hole_em->block_len = 0;
3695 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
3696 hole_em->compress_type = BTRFS_COMPRESS_NONE;
3697 hole_em->generation = trans->transid;
3700 write_lock(&em_tree->lock);
3701 err = add_extent_mapping(em_tree, hole_em);
3703 list_move(&hole_em->list,
3704 &em_tree->modified_extents);
3705 write_unlock(&em_tree->lock);
3708 btrfs_drop_extent_cache(inode, cur_offset,
3712 free_extent_map(hole_em);
3714 btrfs_update_inode(trans, root, inode);
3715 btrfs_end_transaction(trans, root);
3717 free_extent_map(em);
3719 cur_offset = last_byte;
3720 if (cur_offset >= block_end)
3724 free_extent_map(em);
3725 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3730 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3732 struct btrfs_root *root = BTRFS_I(inode)->root;
3733 struct btrfs_trans_handle *trans;
3734 loff_t oldsize = i_size_read(inode);
3737 if (newsize == oldsize)
3740 if (newsize > oldsize) {
3741 truncate_pagecache(inode, oldsize, newsize);
3742 ret = btrfs_cont_expand(inode, oldsize, newsize);
3746 trans = btrfs_start_transaction(root, 1);
3748 return PTR_ERR(trans);
3750 i_size_write(inode, newsize);
3751 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3752 ret = btrfs_update_inode(trans, root, inode);
3753 btrfs_end_transaction(trans, root);
3757 * We're truncating a file that used to have good data down to
3758 * zero. Make sure it gets into the ordered flush list so that
3759 * any new writes get down to disk quickly.
3762 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3763 &BTRFS_I(inode)->runtime_flags);
3765 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3766 truncate_setsize(inode, newsize);
3767 ret = btrfs_truncate(inode);
3773 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3775 struct inode *inode = dentry->d_inode;
3776 struct btrfs_root *root = BTRFS_I(inode)->root;
3779 if (btrfs_root_readonly(root))
3782 err = inode_change_ok(inode, attr);
3786 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3787 err = btrfs_setsize(inode, attr->ia_size);
3792 if (attr->ia_valid) {
3793 setattr_copy(inode, attr);
3794 inode_inc_iversion(inode);
3795 err = btrfs_dirty_inode(inode);
3797 if (!err && attr->ia_valid & ATTR_MODE)
3798 err = btrfs_acl_chmod(inode);
3804 void btrfs_evict_inode(struct inode *inode)
3806 struct btrfs_trans_handle *trans;
3807 struct btrfs_root *root = BTRFS_I(inode)->root;
3808 struct btrfs_block_rsv *rsv, *global_rsv;
3809 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3812 trace_btrfs_inode_evict(inode);
3814 truncate_inode_pages(&inode->i_data, 0);
3815 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3816 btrfs_is_free_space_inode(inode)))
3819 if (is_bad_inode(inode)) {
3820 btrfs_orphan_del(NULL, inode);
3823 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3824 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3826 if (root->fs_info->log_root_recovering) {
3827 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3828 &BTRFS_I(inode)->runtime_flags));
3832 if (inode->i_nlink > 0) {
3833 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3837 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3839 btrfs_orphan_del(NULL, inode);
3842 rsv->size = min_size;
3844 global_rsv = &root->fs_info->global_block_rsv;
3846 btrfs_i_size_write(inode, 0);
3849 * This is a bit simpler than btrfs_truncate since we've already
3850 * reserved our space for our orphan item in the unlink, so we just
3851 * need to reserve some slack space in case we add bytes and update
3852 * inode item when doing the truncate.
3855 ret = btrfs_block_rsv_refill(root, rsv, min_size,
3856 BTRFS_RESERVE_FLUSH_LIMIT);
3859 * Try and steal from the global reserve since we will
3860 * likely not use this space anyway, we want to try as
3861 * hard as possible to get this to work.
3864 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3867 printk(KERN_WARNING "Could not get space for a "
3868 "delete, will truncate on mount %d\n", ret);
3869 btrfs_orphan_del(NULL, inode);
3870 btrfs_free_block_rsv(root, rsv);
3874 trans = btrfs_start_transaction_lflush(root, 1);
3875 if (IS_ERR(trans)) {
3876 btrfs_orphan_del(NULL, inode);
3877 btrfs_free_block_rsv(root, rsv);
3881 trans->block_rsv = rsv;
3883 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3887 trans->block_rsv = &root->fs_info->trans_block_rsv;
3888 ret = btrfs_update_inode(trans, root, inode);
3891 btrfs_end_transaction(trans, root);
3893 btrfs_btree_balance_dirty(root);
3896 btrfs_free_block_rsv(root, rsv);
3899 trans->block_rsv = root->orphan_block_rsv;
3900 ret = btrfs_orphan_del(trans, inode);
3904 trans->block_rsv = &root->fs_info->trans_block_rsv;
3905 if (!(root == root->fs_info->tree_root ||
3906 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3907 btrfs_return_ino(root, btrfs_ino(inode));
3909 btrfs_end_transaction(trans, root);
3910 btrfs_btree_balance_dirty(root);
3917 * this returns the key found in the dir entry in the location pointer.
3918 * If no dir entries were found, location->objectid is 0.
3920 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3921 struct btrfs_key *location)
3923 const char *name = dentry->d_name.name;
3924 int namelen = dentry->d_name.len;
3925 struct btrfs_dir_item *di;
3926 struct btrfs_path *path;
3927 struct btrfs_root *root = BTRFS_I(dir)->root;
3930 path = btrfs_alloc_path();
3934 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3939 if (IS_ERR_OR_NULL(di))
3942 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3944 btrfs_free_path(path);
3947 location->objectid = 0;
3952 * when we hit a tree root in a directory, the btrfs part of the inode
3953 * needs to be changed to reflect the root directory of the tree root. This
3954 * is kind of like crossing a mount point.
3956 static int fixup_tree_root_location(struct btrfs_root *root,
3958 struct dentry *dentry,
3959 struct btrfs_key *location,
3960 struct btrfs_root **sub_root)
3962 struct btrfs_path *path;
3963 struct btrfs_root *new_root;
3964 struct btrfs_root_ref *ref;
3965 struct extent_buffer *leaf;
3969 path = btrfs_alloc_path();
3976 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3977 BTRFS_I(dir)->root->root_key.objectid,
3978 location->objectid);
3985 leaf = path->nodes[0];
3986 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3987 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3988 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3991 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3992 (unsigned long)(ref + 1),
3993 dentry->d_name.len);
3997 btrfs_release_path(path);
3999 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4000 if (IS_ERR(new_root)) {
4001 err = PTR_ERR(new_root);
4005 if (btrfs_root_refs(&new_root->root_item) == 0) {
4010 *sub_root = new_root;
4011 location->objectid = btrfs_root_dirid(&new_root->root_item);
4012 location->type = BTRFS_INODE_ITEM_KEY;
4013 location->offset = 0;
4016 btrfs_free_path(path);
4020 static void inode_tree_add(struct inode *inode)
4022 struct btrfs_root *root = BTRFS_I(inode)->root;
4023 struct btrfs_inode *entry;
4025 struct rb_node *parent;
4026 u64 ino = btrfs_ino(inode);
4028 p = &root->inode_tree.rb_node;
4031 if (inode_unhashed(inode))
4034 spin_lock(&root->inode_lock);
4037 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4039 if (ino < btrfs_ino(&entry->vfs_inode))
4040 p = &parent->rb_left;
4041 else if (ino > btrfs_ino(&entry->vfs_inode))
4042 p = &parent->rb_right;
4044 WARN_ON(!(entry->vfs_inode.i_state &
4045 (I_WILL_FREE | I_FREEING)));
4046 rb_erase(parent, &root->inode_tree);
4047 RB_CLEAR_NODE(parent);
4048 spin_unlock(&root->inode_lock);
4052 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4053 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4054 spin_unlock(&root->inode_lock);
4057 static void inode_tree_del(struct inode *inode)
4059 struct btrfs_root *root = BTRFS_I(inode)->root;
4062 spin_lock(&root->inode_lock);
4063 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4064 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4065 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4066 empty = RB_EMPTY_ROOT(&root->inode_tree);
4068 spin_unlock(&root->inode_lock);
4071 * Free space cache has inodes in the tree root, but the tree root has a
4072 * root_refs of 0, so this could end up dropping the tree root as a
4073 * snapshot, so we need the extra !root->fs_info->tree_root check to
4074 * make sure we don't drop it.
4076 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4077 root != root->fs_info->tree_root) {
4078 synchronize_srcu(&root->fs_info->subvol_srcu);
4079 spin_lock(&root->inode_lock);
4080 empty = RB_EMPTY_ROOT(&root->inode_tree);
4081 spin_unlock(&root->inode_lock);
4083 btrfs_add_dead_root(root);
4087 void btrfs_invalidate_inodes(struct btrfs_root *root)
4089 struct rb_node *node;
4090 struct rb_node *prev;
4091 struct btrfs_inode *entry;
4092 struct inode *inode;
4095 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4097 spin_lock(&root->inode_lock);
4099 node = root->inode_tree.rb_node;
4103 entry = rb_entry(node, struct btrfs_inode, rb_node);
4105 if (objectid < btrfs_ino(&entry->vfs_inode))
4106 node = node->rb_left;
4107 else if (objectid > btrfs_ino(&entry->vfs_inode))
4108 node = node->rb_right;
4114 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4115 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4119 prev = rb_next(prev);
4123 entry = rb_entry(node, struct btrfs_inode, rb_node);
4124 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4125 inode = igrab(&entry->vfs_inode);
4127 spin_unlock(&root->inode_lock);
4128 if (atomic_read(&inode->i_count) > 1)
4129 d_prune_aliases(inode);
4131 * btrfs_drop_inode will have it removed from
4132 * the inode cache when its usage count
4137 spin_lock(&root->inode_lock);
4141 if (cond_resched_lock(&root->inode_lock))
4144 node = rb_next(node);
4146 spin_unlock(&root->inode_lock);
4149 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4151 struct btrfs_iget_args *args = p;
4152 inode->i_ino = args->ino;
4153 BTRFS_I(inode)->root = args->root;
4157 static int btrfs_find_actor(struct inode *inode, void *opaque)
4159 struct btrfs_iget_args *args = opaque;
4160 return args->ino == btrfs_ino(inode) &&
4161 args->root == BTRFS_I(inode)->root;
4164 static struct inode *btrfs_iget_locked(struct super_block *s,
4166 struct btrfs_root *root)
4168 struct inode *inode;
4169 struct btrfs_iget_args args;
4170 args.ino = objectid;
4173 inode = iget5_locked(s, objectid, btrfs_find_actor,
4174 btrfs_init_locked_inode,
4179 /* Get an inode object given its location and corresponding root.
4180 * Returns in *is_new if the inode was read from disk
4182 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4183 struct btrfs_root *root, int *new)
4185 struct inode *inode;
4187 inode = btrfs_iget_locked(s, location->objectid, root);
4189 return ERR_PTR(-ENOMEM);
4191 if (inode->i_state & I_NEW) {
4192 BTRFS_I(inode)->root = root;
4193 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4194 btrfs_read_locked_inode(inode);
4195 if (!is_bad_inode(inode)) {
4196 inode_tree_add(inode);
4197 unlock_new_inode(inode);
4201 unlock_new_inode(inode);
4203 inode = ERR_PTR(-ESTALE);
4210 static struct inode *new_simple_dir(struct super_block *s,
4211 struct btrfs_key *key,
4212 struct btrfs_root *root)
4214 struct inode *inode = new_inode(s);
4217 return ERR_PTR(-ENOMEM);
4219 BTRFS_I(inode)->root = root;
4220 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4221 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4223 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4224 inode->i_op = &btrfs_dir_ro_inode_operations;
4225 inode->i_fop = &simple_dir_operations;
4226 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4227 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4232 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4234 struct inode *inode;
4235 struct btrfs_root *root = BTRFS_I(dir)->root;
4236 struct btrfs_root *sub_root = root;
4237 struct btrfs_key location;
4241 if (dentry->d_name.len > BTRFS_NAME_LEN)
4242 return ERR_PTR(-ENAMETOOLONG);
4244 if (unlikely(d_need_lookup(dentry))) {
4245 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4246 kfree(dentry->d_fsdata);
4247 dentry->d_fsdata = NULL;
4248 /* This thing is hashed, drop it for now */
4251 ret = btrfs_inode_by_name(dir, dentry, &location);
4255 return ERR_PTR(ret);
4257 if (location.objectid == 0)
4260 if (location.type == BTRFS_INODE_ITEM_KEY) {
4261 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4265 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4267 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4268 ret = fixup_tree_root_location(root, dir, dentry,
4269 &location, &sub_root);
4272 inode = ERR_PTR(ret);
4274 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4276 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4278 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4280 if (!IS_ERR(inode) && root != sub_root) {
4281 down_read(&root->fs_info->cleanup_work_sem);
4282 if (!(inode->i_sb->s_flags & MS_RDONLY))
4283 ret = btrfs_orphan_cleanup(sub_root);
4284 up_read(&root->fs_info->cleanup_work_sem);
4286 inode = ERR_PTR(ret);
4292 static int btrfs_dentry_delete(const struct dentry *dentry)
4294 struct btrfs_root *root;
4295 struct inode *inode = dentry->d_inode;
4297 if (!inode && !IS_ROOT(dentry))
4298 inode = dentry->d_parent->d_inode;
4301 root = BTRFS_I(inode)->root;
4302 if (btrfs_root_refs(&root->root_item) == 0)
4305 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4311 static void btrfs_dentry_release(struct dentry *dentry)
4313 if (dentry->d_fsdata)
4314 kfree(dentry->d_fsdata);
4317 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4322 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4323 if (unlikely(d_need_lookup(dentry))) {
4324 spin_lock(&dentry->d_lock);
4325 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4326 spin_unlock(&dentry->d_lock);
4331 unsigned char btrfs_filetype_table[] = {
4332 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4335 static int btrfs_real_readdir(struct file *filp, void *dirent,
4338 struct inode *inode = filp->f_dentry->d_inode;
4339 struct btrfs_root *root = BTRFS_I(inode)->root;
4340 struct btrfs_item *item;
4341 struct btrfs_dir_item *di;
4342 struct btrfs_key key;
4343 struct btrfs_key found_key;
4344 struct btrfs_path *path;
4345 struct list_head ins_list;
4346 struct list_head del_list;
4348 struct extent_buffer *leaf;
4350 unsigned char d_type;
4355 int key_type = BTRFS_DIR_INDEX_KEY;
4359 int is_curr = 0; /* filp->f_pos points to the current index? */
4361 /* FIXME, use a real flag for deciding about the key type */
4362 if (root->fs_info->tree_root == root)
4363 key_type = BTRFS_DIR_ITEM_KEY;
4365 /* special case for "." */
4366 if (filp->f_pos == 0) {
4367 over = filldir(dirent, ".", 1,
4368 filp->f_pos, btrfs_ino(inode), DT_DIR);
4373 /* special case for .., just use the back ref */
4374 if (filp->f_pos == 1) {
4375 u64 pino = parent_ino(filp->f_path.dentry);
4376 over = filldir(dirent, "..", 2,
4377 filp->f_pos, pino, DT_DIR);
4382 path = btrfs_alloc_path();
4388 if (key_type == BTRFS_DIR_INDEX_KEY) {
4389 INIT_LIST_HEAD(&ins_list);
4390 INIT_LIST_HEAD(&del_list);
4391 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4394 btrfs_set_key_type(&key, key_type);
4395 key.offset = filp->f_pos;
4396 key.objectid = btrfs_ino(inode);
4398 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4403 leaf = path->nodes[0];
4404 slot = path->slots[0];
4405 if (slot >= btrfs_header_nritems(leaf)) {
4406 ret = btrfs_next_leaf(root, path);
4414 item = btrfs_item_nr(leaf, slot);
4415 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4417 if (found_key.objectid != key.objectid)
4419 if (btrfs_key_type(&found_key) != key_type)
4421 if (found_key.offset < filp->f_pos)
4423 if (key_type == BTRFS_DIR_INDEX_KEY &&
4424 btrfs_should_delete_dir_index(&del_list,
4428 filp->f_pos = found_key.offset;
4431 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4433 di_total = btrfs_item_size(leaf, item);
4435 while (di_cur < di_total) {
4436 struct btrfs_key location;
4438 if (verify_dir_item(root, leaf, di))
4441 name_len = btrfs_dir_name_len(leaf, di);
4442 if (name_len <= sizeof(tmp_name)) {
4443 name_ptr = tmp_name;
4445 name_ptr = kmalloc(name_len, GFP_NOFS);
4451 read_extent_buffer(leaf, name_ptr,
4452 (unsigned long)(di + 1), name_len);
4454 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4455 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4458 /* is this a reference to our own snapshot? If so
4461 * In contrast to old kernels, we insert the snapshot's
4462 * dir item and dir index after it has been created, so
4463 * we won't find a reference to our own snapshot. We
4464 * still keep the following code for backward
4467 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4468 location.objectid == root->root_key.objectid) {
4472 over = filldir(dirent, name_ptr, name_len,
4473 found_key.offset, location.objectid,
4477 if (name_ptr != tmp_name)
4482 di_len = btrfs_dir_name_len(leaf, di) +
4483 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4485 di = (struct btrfs_dir_item *)((char *)di + di_len);
4491 if (key_type == BTRFS_DIR_INDEX_KEY) {
4494 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4500 /* Reached end of directory/root. Bump pos past the last item. */
4501 if (key_type == BTRFS_DIR_INDEX_KEY)
4503 * 32-bit glibc will use getdents64, but then strtol -
4504 * so the last number we can serve is this.
4506 filp->f_pos = 0x7fffffff;
4512 if (key_type == BTRFS_DIR_INDEX_KEY)
4513 btrfs_put_delayed_items(&ins_list, &del_list);
4514 btrfs_free_path(path);
4518 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4520 struct btrfs_root *root = BTRFS_I(inode)->root;
4521 struct btrfs_trans_handle *trans;
4523 bool nolock = false;
4525 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4528 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
4531 if (wbc->sync_mode == WB_SYNC_ALL) {
4533 trans = btrfs_join_transaction_nolock(root);
4535 trans = btrfs_join_transaction(root);
4537 return PTR_ERR(trans);
4538 ret = btrfs_commit_transaction(trans, root);
4544 * This is somewhat expensive, updating the tree every time the
4545 * inode changes. But, it is most likely to find the inode in cache.
4546 * FIXME, needs more benchmarking...there are no reasons other than performance
4547 * to keep or drop this code.
4549 int btrfs_dirty_inode(struct inode *inode)
4551 struct btrfs_root *root = BTRFS_I(inode)->root;
4552 struct btrfs_trans_handle *trans;
4555 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4558 trans = btrfs_join_transaction(root);
4560 return PTR_ERR(trans);
4562 ret = btrfs_update_inode(trans, root, inode);
4563 if (ret && ret == -ENOSPC) {
4564 /* whoops, lets try again with the full transaction */
4565 btrfs_end_transaction(trans, root);
4566 trans = btrfs_start_transaction(root, 1);
4568 return PTR_ERR(trans);
4570 ret = btrfs_update_inode(trans, root, inode);
4572 btrfs_end_transaction(trans, root);
4573 if (BTRFS_I(inode)->delayed_node)
4574 btrfs_balance_delayed_items(root);
4580 * This is a copy of file_update_time. We need this so we can return error on
4581 * ENOSPC for updating the inode in the case of file write and mmap writes.
4583 static int btrfs_update_time(struct inode *inode, struct timespec *now,
4586 struct btrfs_root *root = BTRFS_I(inode)->root;
4588 if (btrfs_root_readonly(root))
4591 if (flags & S_VERSION)
4592 inode_inc_iversion(inode);
4593 if (flags & S_CTIME)
4594 inode->i_ctime = *now;
4595 if (flags & S_MTIME)
4596 inode->i_mtime = *now;
4597 if (flags & S_ATIME)
4598 inode->i_atime = *now;
4599 return btrfs_dirty_inode(inode);
4603 * find the highest existing sequence number in a directory
4604 * and then set the in-memory index_cnt variable to reflect
4605 * free sequence numbers
4607 static int btrfs_set_inode_index_count(struct inode *inode)
4609 struct btrfs_root *root = BTRFS_I(inode)->root;
4610 struct btrfs_key key, found_key;
4611 struct btrfs_path *path;
4612 struct extent_buffer *leaf;
4615 key.objectid = btrfs_ino(inode);
4616 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4617 key.offset = (u64)-1;
4619 path = btrfs_alloc_path();
4623 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4626 /* FIXME: we should be able to handle this */
4632 * MAGIC NUMBER EXPLANATION:
4633 * since we search a directory based on f_pos we have to start at 2
4634 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4635 * else has to start at 2
4637 if (path->slots[0] == 0) {
4638 BTRFS_I(inode)->index_cnt = 2;
4644 leaf = path->nodes[0];
4645 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4647 if (found_key.objectid != btrfs_ino(inode) ||
4648 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4649 BTRFS_I(inode)->index_cnt = 2;
4653 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4655 btrfs_free_path(path);
4660 * helper to find a free sequence number in a given directory. This current
4661 * code is very simple, later versions will do smarter things in the btree
4663 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4667 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4668 ret = btrfs_inode_delayed_dir_index_count(dir);
4670 ret = btrfs_set_inode_index_count(dir);
4676 *index = BTRFS_I(dir)->index_cnt;
4677 BTRFS_I(dir)->index_cnt++;
4682 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4683 struct btrfs_root *root,
4685 const char *name, int name_len,
4686 u64 ref_objectid, u64 objectid,
4687 umode_t mode, u64 *index)
4689 struct inode *inode;
4690 struct btrfs_inode_item *inode_item;
4691 struct btrfs_key *location;
4692 struct btrfs_path *path;
4693 struct btrfs_inode_ref *ref;
4694 struct btrfs_key key[2];
4700 path = btrfs_alloc_path();
4702 return ERR_PTR(-ENOMEM);
4704 inode = new_inode(root->fs_info->sb);
4706 btrfs_free_path(path);
4707 return ERR_PTR(-ENOMEM);
4711 * we have to initialize this early, so we can reclaim the inode
4712 * number if we fail afterwards in this function.
4714 inode->i_ino = objectid;
4717 trace_btrfs_inode_request(dir);
4719 ret = btrfs_set_inode_index(dir, index);
4721 btrfs_free_path(path);
4723 return ERR_PTR(ret);
4727 * index_cnt is ignored for everything but a dir,
4728 * btrfs_get_inode_index_count has an explanation for the magic
4731 BTRFS_I(inode)->index_cnt = 2;
4732 BTRFS_I(inode)->root = root;
4733 BTRFS_I(inode)->generation = trans->transid;
4734 inode->i_generation = BTRFS_I(inode)->generation;
4737 * We could have gotten an inode number from somebody who was fsynced
4738 * and then removed in this same transaction, so let's just set full
4739 * sync since it will be a full sync anyway and this will blow away the
4740 * old info in the log.
4742 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
4749 key[0].objectid = objectid;
4750 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4754 * Start new inodes with an inode_ref. This is slightly more
4755 * efficient for small numbers of hard links since they will
4756 * be packed into one item. Extended refs will kick in if we
4757 * add more hard links than can fit in the ref item.
4759 key[1].objectid = objectid;
4760 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4761 key[1].offset = ref_objectid;
4763 sizes[0] = sizeof(struct btrfs_inode_item);
4764 sizes[1] = name_len + sizeof(*ref);
4766 path->leave_spinning = 1;
4767 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4771 inode_init_owner(inode, dir, mode);
4772 inode_set_bytes(inode, 0);
4773 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4774 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4775 struct btrfs_inode_item);
4776 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
4777 sizeof(*inode_item));
4778 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4780 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4781 struct btrfs_inode_ref);
4782 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4783 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4784 ptr = (unsigned long)(ref + 1);
4785 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4787 btrfs_mark_buffer_dirty(path->nodes[0]);
4788 btrfs_free_path(path);
4790 location = &BTRFS_I(inode)->location;
4791 location->objectid = objectid;
4792 location->offset = 0;
4793 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4795 btrfs_inherit_iflags(inode, dir);
4797 if (S_ISREG(mode)) {
4798 if (btrfs_test_opt(root, NODATASUM))
4799 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4800 if (btrfs_test_opt(root, NODATACOW) ||
4801 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4802 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4805 insert_inode_hash(inode);
4806 inode_tree_add(inode);
4808 trace_btrfs_inode_new(inode);
4809 btrfs_set_inode_last_trans(trans, inode);
4811 btrfs_update_root_times(trans, root);
4816 BTRFS_I(dir)->index_cnt--;
4817 btrfs_free_path(path);
4819 return ERR_PTR(ret);
4822 static inline u8 btrfs_inode_type(struct inode *inode)
4824 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4828 * utility function to add 'inode' into 'parent_inode' with
4829 * a give name and a given sequence number.
4830 * if 'add_backref' is true, also insert a backref from the
4831 * inode to the parent directory.
4833 int btrfs_add_link(struct btrfs_trans_handle *trans,
4834 struct inode *parent_inode, struct inode *inode,
4835 const char *name, int name_len, int add_backref, u64 index)
4838 struct btrfs_key key;
4839 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4840 u64 ino = btrfs_ino(inode);
4841 u64 parent_ino = btrfs_ino(parent_inode);
4843 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4844 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4847 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4851 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4852 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4853 key.objectid, root->root_key.objectid,
4854 parent_ino, index, name, name_len);
4855 } else if (add_backref) {
4856 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4860 /* Nothing to clean up yet */
4864 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4866 btrfs_inode_type(inode), index);
4870 btrfs_abort_transaction(trans, root, ret);
4874 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4876 inode_inc_iversion(parent_inode);
4877 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4878 ret = btrfs_update_inode(trans, root, parent_inode);
4880 btrfs_abort_transaction(trans, root, ret);
4884 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4887 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4888 key.objectid, root->root_key.objectid,
4889 parent_ino, &local_index, name, name_len);
4891 } else if (add_backref) {
4895 err = btrfs_del_inode_ref(trans, root, name, name_len,
4896 ino, parent_ino, &local_index);
4901 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4902 struct inode *dir, struct dentry *dentry,
4903 struct inode *inode, int backref, u64 index)
4905 int err = btrfs_add_link(trans, dir, inode,
4906 dentry->d_name.name, dentry->d_name.len,
4913 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4914 umode_t mode, dev_t rdev)
4916 struct btrfs_trans_handle *trans;
4917 struct btrfs_root *root = BTRFS_I(dir)->root;
4918 struct inode *inode = NULL;
4924 if (!new_valid_dev(rdev))
4928 * 2 for inode item and ref
4930 * 1 for xattr if selinux is on
4932 trans = btrfs_start_transaction(root, 5);
4934 return PTR_ERR(trans);
4936 err = btrfs_find_free_ino(root, &objectid);
4940 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4941 dentry->d_name.len, btrfs_ino(dir), objectid,
4943 if (IS_ERR(inode)) {
4944 err = PTR_ERR(inode);
4948 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4954 err = btrfs_update_inode(trans, root, inode);
4961 * If the active LSM wants to access the inode during
4962 * d_instantiate it needs these. Smack checks to see
4963 * if the filesystem supports xattrs by looking at the
4967 inode->i_op = &btrfs_special_inode_operations;
4968 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4972 init_special_inode(inode, inode->i_mode, rdev);
4973 btrfs_update_inode(trans, root, inode);
4974 d_instantiate(dentry, inode);
4977 btrfs_end_transaction(trans, root);
4978 btrfs_btree_balance_dirty(root);
4980 inode_dec_link_count(inode);
4986 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4987 umode_t mode, bool excl)
4989 struct btrfs_trans_handle *trans;
4990 struct btrfs_root *root = BTRFS_I(dir)->root;
4991 struct inode *inode = NULL;
4998 * 2 for inode item and ref
5000 * 1 for xattr if selinux is on
5002 trans = btrfs_start_transaction(root, 5);
5004 return PTR_ERR(trans);
5006 err = btrfs_find_free_ino(root, &objectid);
5010 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5011 dentry->d_name.len, btrfs_ino(dir), objectid,
5013 if (IS_ERR(inode)) {
5014 err = PTR_ERR(inode);
5018 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5025 * If the active LSM wants to access the inode during
5026 * d_instantiate it needs these. Smack checks to see
5027 * if the filesystem supports xattrs by looking at the
5030 inode->i_fop = &btrfs_file_operations;
5031 inode->i_op = &btrfs_file_inode_operations;
5033 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5037 inode->i_mapping->a_ops = &btrfs_aops;
5038 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5039 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5040 d_instantiate(dentry, inode);
5043 btrfs_end_transaction(trans, root);
5045 inode_dec_link_count(inode);
5048 btrfs_btree_balance_dirty(root);
5052 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5053 struct dentry *dentry)
5055 struct btrfs_trans_handle *trans;
5056 struct btrfs_root *root = BTRFS_I(dir)->root;
5057 struct inode *inode = old_dentry->d_inode;
5062 /* do not allow sys_link's with other subvols of the same device */
5063 if (root->objectid != BTRFS_I(inode)->root->objectid)
5066 if (inode->i_nlink >= BTRFS_LINK_MAX)
5069 err = btrfs_set_inode_index(dir, &index);
5074 * 2 items for inode and inode ref
5075 * 2 items for dir items
5076 * 1 item for parent inode
5078 trans = btrfs_start_transaction(root, 5);
5079 if (IS_ERR(trans)) {
5080 err = PTR_ERR(trans);
5084 btrfs_inc_nlink(inode);
5085 inode_inc_iversion(inode);
5086 inode->i_ctime = CURRENT_TIME;
5089 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5094 struct dentry *parent = dentry->d_parent;
5095 err = btrfs_update_inode(trans, root, inode);
5098 d_instantiate(dentry, inode);
5099 btrfs_log_new_name(trans, inode, NULL, parent);
5102 btrfs_end_transaction(trans, root);
5105 inode_dec_link_count(inode);
5108 btrfs_btree_balance_dirty(root);
5112 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5114 struct inode *inode = NULL;
5115 struct btrfs_trans_handle *trans;
5116 struct btrfs_root *root = BTRFS_I(dir)->root;
5118 int drop_on_err = 0;
5123 * 2 items for inode and ref
5124 * 2 items for dir items
5125 * 1 for xattr if selinux is on
5127 trans = btrfs_start_transaction(root, 5);
5129 return PTR_ERR(trans);
5131 err = btrfs_find_free_ino(root, &objectid);
5135 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5136 dentry->d_name.len, btrfs_ino(dir), objectid,
5137 S_IFDIR | mode, &index);
5138 if (IS_ERR(inode)) {
5139 err = PTR_ERR(inode);
5145 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5149 inode->i_op = &btrfs_dir_inode_operations;
5150 inode->i_fop = &btrfs_dir_file_operations;
5152 btrfs_i_size_write(inode, 0);
5153 err = btrfs_update_inode(trans, root, inode);
5157 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5158 dentry->d_name.len, 0, index);
5162 d_instantiate(dentry, inode);
5166 btrfs_end_transaction(trans, root);
5169 btrfs_btree_balance_dirty(root);
5173 /* helper for btfs_get_extent. Given an existing extent in the tree,
5174 * and an extent that you want to insert, deal with overlap and insert
5175 * the new extent into the tree.
5177 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5178 struct extent_map *existing,
5179 struct extent_map *em,
5180 u64 map_start, u64 map_len)
5184 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5185 start_diff = map_start - em->start;
5186 em->start = map_start;
5188 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5189 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5190 em->block_start += start_diff;
5191 em->block_len -= start_diff;
5193 return add_extent_mapping(em_tree, em);
5196 static noinline int uncompress_inline(struct btrfs_path *path,
5197 struct inode *inode, struct page *page,
5198 size_t pg_offset, u64 extent_offset,
5199 struct btrfs_file_extent_item *item)
5202 struct extent_buffer *leaf = path->nodes[0];
5205 unsigned long inline_size;
5209 WARN_ON(pg_offset != 0);
5210 compress_type = btrfs_file_extent_compression(leaf, item);
5211 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5212 inline_size = btrfs_file_extent_inline_item_len(leaf,
5213 btrfs_item_nr(leaf, path->slots[0]));
5214 tmp = kmalloc(inline_size, GFP_NOFS);
5217 ptr = btrfs_file_extent_inline_start(item);
5219 read_extent_buffer(leaf, tmp, ptr, inline_size);
5221 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5222 ret = btrfs_decompress(compress_type, tmp, page,
5223 extent_offset, inline_size, max_size);
5225 char *kaddr = kmap_atomic(page);
5226 unsigned long copy_size = min_t(u64,
5227 PAGE_CACHE_SIZE - pg_offset,
5228 max_size - extent_offset);
5229 memset(kaddr + pg_offset, 0, copy_size);
5230 kunmap_atomic(kaddr);
5237 * a bit scary, this does extent mapping from logical file offset to the disk.
5238 * the ugly parts come from merging extents from the disk with the in-ram
5239 * representation. This gets more complex because of the data=ordered code,
5240 * where the in-ram extents might be locked pending data=ordered completion.
5242 * This also copies inline extents directly into the page.
5245 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5246 size_t pg_offset, u64 start, u64 len,
5252 u64 extent_start = 0;
5254 u64 objectid = btrfs_ino(inode);
5256 struct btrfs_path *path = NULL;
5257 struct btrfs_root *root = BTRFS_I(inode)->root;
5258 struct btrfs_file_extent_item *item;
5259 struct extent_buffer *leaf;
5260 struct btrfs_key found_key;
5261 struct extent_map *em = NULL;
5262 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5263 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5264 struct btrfs_trans_handle *trans = NULL;
5268 read_lock(&em_tree->lock);
5269 em = lookup_extent_mapping(em_tree, start, len);
5271 em->bdev = root->fs_info->fs_devices->latest_bdev;
5272 read_unlock(&em_tree->lock);
5275 if (em->start > start || em->start + em->len <= start)
5276 free_extent_map(em);
5277 else if (em->block_start == EXTENT_MAP_INLINE && page)
5278 free_extent_map(em);
5282 em = alloc_extent_map();
5287 em->bdev = root->fs_info->fs_devices->latest_bdev;
5288 em->start = EXTENT_MAP_HOLE;
5289 em->orig_start = EXTENT_MAP_HOLE;
5291 em->block_len = (u64)-1;
5294 path = btrfs_alloc_path();
5300 * Chances are we'll be called again, so go ahead and do
5306 ret = btrfs_lookup_file_extent(trans, root, path,
5307 objectid, start, trans != NULL);
5314 if (path->slots[0] == 0)
5319 leaf = path->nodes[0];
5320 item = btrfs_item_ptr(leaf, path->slots[0],
5321 struct btrfs_file_extent_item);
5322 /* are we inside the extent that was found? */
5323 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5324 found_type = btrfs_key_type(&found_key);
5325 if (found_key.objectid != objectid ||
5326 found_type != BTRFS_EXTENT_DATA_KEY) {
5330 found_type = btrfs_file_extent_type(leaf, item);
5331 extent_start = found_key.offset;
5332 compress_type = btrfs_file_extent_compression(leaf, item);
5333 if (found_type == BTRFS_FILE_EXTENT_REG ||
5334 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5335 extent_end = extent_start +
5336 btrfs_file_extent_num_bytes(leaf, item);
5337 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5339 size = btrfs_file_extent_inline_len(leaf, item);
5340 extent_end = (extent_start + size + root->sectorsize - 1) &
5341 ~((u64)root->sectorsize - 1);
5344 if (start >= extent_end) {
5346 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5347 ret = btrfs_next_leaf(root, path);
5354 leaf = path->nodes[0];
5356 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5357 if (found_key.objectid != objectid ||
5358 found_key.type != BTRFS_EXTENT_DATA_KEY)
5360 if (start + len <= found_key.offset)
5363 em->len = found_key.offset - start;
5367 if (found_type == BTRFS_FILE_EXTENT_REG ||
5368 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5369 em->start = extent_start;
5370 em->len = extent_end - extent_start;
5371 em->orig_start = extent_start -
5372 btrfs_file_extent_offset(leaf, item);
5373 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5375 em->block_start = EXTENT_MAP_HOLE;
5378 if (compress_type != BTRFS_COMPRESS_NONE) {
5379 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5380 em->compress_type = compress_type;
5381 em->block_start = bytenr;
5382 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5385 bytenr += btrfs_file_extent_offset(leaf, item);
5386 em->block_start = bytenr;
5387 em->block_len = em->len;
5388 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5389 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5392 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5396 size_t extent_offset;
5399 em->block_start = EXTENT_MAP_INLINE;
5400 if (!page || create) {
5401 em->start = extent_start;
5402 em->len = extent_end - extent_start;
5406 size = btrfs_file_extent_inline_len(leaf, item);
5407 extent_offset = page_offset(page) + pg_offset - extent_start;
5408 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5409 size - extent_offset);
5410 em->start = extent_start + extent_offset;
5411 em->len = (copy_size + root->sectorsize - 1) &
5412 ~((u64)root->sectorsize - 1);
5413 em->orig_start = EXTENT_MAP_INLINE;
5414 if (compress_type) {
5415 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5416 em->compress_type = compress_type;
5418 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5419 if (create == 0 && !PageUptodate(page)) {
5420 if (btrfs_file_extent_compression(leaf, item) !=
5421 BTRFS_COMPRESS_NONE) {
5422 ret = uncompress_inline(path, inode, page,
5424 extent_offset, item);
5425 BUG_ON(ret); /* -ENOMEM */
5428 read_extent_buffer(leaf, map + pg_offset, ptr,
5430 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5431 memset(map + pg_offset + copy_size, 0,
5432 PAGE_CACHE_SIZE - pg_offset -
5437 flush_dcache_page(page);
5438 } else if (create && PageUptodate(page)) {
5442 free_extent_map(em);
5445 btrfs_release_path(path);
5446 trans = btrfs_join_transaction(root);
5449 return ERR_CAST(trans);
5453 write_extent_buffer(leaf, map + pg_offset, ptr,
5456 btrfs_mark_buffer_dirty(leaf);
5458 set_extent_uptodate(io_tree, em->start,
5459 extent_map_end(em) - 1, NULL, GFP_NOFS);
5462 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
5468 em->block_start = EXTENT_MAP_HOLE;
5469 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5471 btrfs_release_path(path);
5472 if (em->start > start || extent_map_end(em) <= start) {
5473 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5474 "[%llu %llu]\n", (unsigned long long)em->start,
5475 (unsigned long long)em->len,
5476 (unsigned long long)start,
5477 (unsigned long long)len);
5483 write_lock(&em_tree->lock);
5484 ret = add_extent_mapping(em_tree, em);
5485 /* it is possible that someone inserted the extent into the tree
5486 * while we had the lock dropped. It is also possible that
5487 * an overlapping map exists in the tree
5489 if (ret == -EEXIST) {
5490 struct extent_map *existing;
5494 existing = lookup_extent_mapping(em_tree, start, len);
5495 if (existing && (existing->start > start ||
5496 existing->start + existing->len <= start)) {
5497 free_extent_map(existing);
5501 existing = lookup_extent_mapping(em_tree, em->start,
5504 err = merge_extent_mapping(em_tree, existing,
5507 free_extent_map(existing);
5509 free_extent_map(em);
5514 free_extent_map(em);
5518 free_extent_map(em);
5523 write_unlock(&em_tree->lock);
5527 trace_btrfs_get_extent(root, em);
5530 btrfs_free_path(path);
5532 ret = btrfs_end_transaction(trans, root);
5537 free_extent_map(em);
5538 return ERR_PTR(err);
5540 BUG_ON(!em); /* Error is always set */
5544 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5545 size_t pg_offset, u64 start, u64 len,
5548 struct extent_map *em;
5549 struct extent_map *hole_em = NULL;
5550 u64 range_start = start;
5556 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5561 * if our em maps to a hole, there might
5562 * actually be delalloc bytes behind it
5564 if (em->block_start != EXTENT_MAP_HOLE)
5570 /* check to see if we've wrapped (len == -1 or similar) */
5579 /* ok, we didn't find anything, lets look for delalloc */
5580 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5581 end, len, EXTENT_DELALLOC, 1);
5582 found_end = range_start + found;
5583 if (found_end < range_start)
5584 found_end = (u64)-1;
5587 * we didn't find anything useful, return
5588 * the original results from get_extent()
5590 if (range_start > end || found_end <= start) {
5596 /* adjust the range_start to make sure it doesn't
5597 * go backwards from the start they passed in
5599 range_start = max(start,range_start);
5600 found = found_end - range_start;
5603 u64 hole_start = start;
5606 em = alloc_extent_map();
5612 * when btrfs_get_extent can't find anything it
5613 * returns one huge hole
5615 * make sure what it found really fits our range, and
5616 * adjust to make sure it is based on the start from
5620 u64 calc_end = extent_map_end(hole_em);
5622 if (calc_end <= start || (hole_em->start > end)) {
5623 free_extent_map(hole_em);
5626 hole_start = max(hole_em->start, start);
5627 hole_len = calc_end - hole_start;
5631 if (hole_em && range_start > hole_start) {
5632 /* our hole starts before our delalloc, so we
5633 * have to return just the parts of the hole
5634 * that go until the delalloc starts
5636 em->len = min(hole_len,
5637 range_start - hole_start);
5638 em->start = hole_start;
5639 em->orig_start = hole_start;
5641 * don't adjust block start at all,
5642 * it is fixed at EXTENT_MAP_HOLE
5644 em->block_start = hole_em->block_start;
5645 em->block_len = hole_len;
5647 em->start = range_start;
5649 em->orig_start = range_start;
5650 em->block_start = EXTENT_MAP_DELALLOC;
5651 em->block_len = found;
5653 } else if (hole_em) {
5658 free_extent_map(hole_em);
5660 free_extent_map(em);
5661 return ERR_PTR(err);
5666 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5667 struct extent_map *em,
5670 struct btrfs_root *root = BTRFS_I(inode)->root;
5671 struct btrfs_trans_handle *trans;
5672 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5673 struct btrfs_key ins;
5676 bool insert = false;
5679 * Ok if the extent map we looked up is a hole and is for the exact
5680 * range we want, there is no reason to allocate a new one, however if
5681 * it is not right then we need to free this one and drop the cache for
5684 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5686 free_extent_map(em);
5689 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5692 trans = btrfs_join_transaction(root);
5694 return ERR_CAST(trans);
5696 if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5697 btrfs_add_inode_defrag(trans, inode);
5699 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5701 alloc_hint = get_extent_allocation_hint(inode, start, len);
5702 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5703 alloc_hint, &ins, 1);
5710 em = alloc_extent_map();
5712 em = ERR_PTR(-ENOMEM);
5718 em->orig_start = em->start;
5719 em->len = ins.offset;
5721 em->block_start = ins.objectid;
5722 em->block_len = ins.offset;
5723 em->bdev = root->fs_info->fs_devices->latest_bdev;
5726 * We need to do this because if we're using the original em we searched
5727 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5730 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5733 write_lock(&em_tree->lock);
5734 ret = add_extent_mapping(em_tree, em);
5735 write_unlock(&em_tree->lock);
5738 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5741 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5742 ins.offset, ins.offset, 0);
5744 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5748 btrfs_end_transaction(trans, root);
5753 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5754 * block must be cow'd
5756 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5757 struct inode *inode, u64 offset, u64 len)
5759 struct btrfs_path *path;
5761 struct extent_buffer *leaf;
5762 struct btrfs_root *root = BTRFS_I(inode)->root;
5763 struct btrfs_file_extent_item *fi;
5764 struct btrfs_key key;
5772 path = btrfs_alloc_path();
5776 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5781 slot = path->slots[0];
5784 /* can't find the item, must cow */
5791 leaf = path->nodes[0];
5792 btrfs_item_key_to_cpu(leaf, &key, slot);
5793 if (key.objectid != btrfs_ino(inode) ||
5794 key.type != BTRFS_EXTENT_DATA_KEY) {
5795 /* not our file or wrong item type, must cow */
5799 if (key.offset > offset) {
5800 /* Wrong offset, must cow */
5804 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5805 found_type = btrfs_file_extent_type(leaf, fi);
5806 if (found_type != BTRFS_FILE_EXTENT_REG &&
5807 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5808 /* not a regular extent, must cow */
5811 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5812 backref_offset = btrfs_file_extent_offset(leaf, fi);
5814 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5815 if (extent_end < offset + len) {
5816 /* extent doesn't include our full range, must cow */
5820 if (btrfs_extent_readonly(root, disk_bytenr))
5824 * look for other files referencing this extent, if we
5825 * find any we must cow
5827 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5828 key.offset - backref_offset, disk_bytenr))
5832 * adjust disk_bytenr and num_bytes to cover just the bytes
5833 * in this extent we are about to write. If there
5834 * are any csums in that range we have to cow in order
5835 * to keep the csums correct
5837 disk_bytenr += backref_offset;
5838 disk_bytenr += offset - key.offset;
5839 num_bytes = min(offset + len, extent_end) - offset;
5840 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5843 * all of the above have passed, it is safe to overwrite this extent
5848 btrfs_free_path(path);
5852 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
5853 struct extent_state **cached_state, int writing)
5855 struct btrfs_ordered_extent *ordered;
5859 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5862 * We're concerned with the entire range that we're going to be
5863 * doing DIO to, so we need to make sure theres no ordered
5864 * extents in this range.
5866 ordered = btrfs_lookup_ordered_range(inode, lockstart,
5867 lockend - lockstart + 1);
5870 * We need to make sure there are no buffered pages in this
5871 * range either, we could have raced between the invalidate in
5872 * generic_file_direct_write and locking the extent. The
5873 * invalidate needs to happen so that reads after a write do not
5876 if (!ordered && (!writing ||
5877 !test_range_bit(&BTRFS_I(inode)->io_tree,
5878 lockstart, lockend, EXTENT_UPTODATE, 0,
5882 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5883 cached_state, GFP_NOFS);
5886 btrfs_start_ordered_extent(inode, ordered, 1);
5887 btrfs_put_ordered_extent(ordered);
5889 /* Screw you mmap */
5890 ret = filemap_write_and_wait_range(inode->i_mapping,
5897 * If we found a page that couldn't be invalidated just
5898 * fall back to buffered.
5900 ret = invalidate_inode_pages2_range(inode->i_mapping,
5901 lockstart >> PAGE_CACHE_SHIFT,
5902 lockend >> PAGE_CACHE_SHIFT);
5913 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
5914 u64 len, u64 orig_start,
5915 u64 block_start, u64 block_len,
5918 struct extent_map_tree *em_tree;
5919 struct extent_map *em;
5920 struct btrfs_root *root = BTRFS_I(inode)->root;
5923 em_tree = &BTRFS_I(inode)->extent_tree;
5924 em = alloc_extent_map();
5926 return ERR_PTR(-ENOMEM);
5929 em->orig_start = orig_start;
5931 em->block_len = block_len;
5932 em->block_start = block_start;
5933 em->bdev = root->fs_info->fs_devices->latest_bdev;
5934 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5935 if (type == BTRFS_ORDERED_PREALLOC)
5936 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5939 btrfs_drop_extent_cache(inode, em->start,
5940 em->start + em->len - 1, 0);
5941 write_lock(&em_tree->lock);
5942 ret = add_extent_mapping(em_tree, em);
5943 write_unlock(&em_tree->lock);
5944 } while (ret == -EEXIST);
5947 free_extent_map(em);
5948 return ERR_PTR(ret);
5955 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5956 struct buffer_head *bh_result, int create)
5958 struct extent_map *em;
5959 struct btrfs_root *root = BTRFS_I(inode)->root;
5960 struct extent_state *cached_state = NULL;
5961 u64 start = iblock << inode->i_blkbits;
5962 u64 lockstart, lockend;
5963 u64 len = bh_result->b_size;
5964 struct btrfs_trans_handle *trans;
5965 int unlock_bits = EXTENT_LOCKED;
5969 ret = btrfs_delalloc_reserve_space(inode, len);
5972 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
5974 len = min_t(u64, len, root->sectorsize);
5978 lockend = start + len - 1;
5981 * If this errors out it's because we couldn't invalidate pagecache for
5982 * this range and we need to fallback to buffered.
5984 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
5988 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
5989 lockend, EXTENT_DELALLOC, NULL,
5990 &cached_state, GFP_NOFS);
5995 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6002 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6003 * io. INLINE is special, and we could probably kludge it in here, but
6004 * it's still buffered so for safety lets just fall back to the generic
6007 * For COMPRESSED we _have_ to read the entire extent in so we can
6008 * decompress it, so there will be buffering required no matter what we
6009 * do, so go ahead and fallback to buffered.
6011 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6012 * to buffered IO. Don't blame me, this is the price we pay for using
6015 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6016 em->block_start == EXTENT_MAP_INLINE) {
6017 free_extent_map(em);
6022 /* Just a good old fashioned hole, return */
6023 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6024 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6025 free_extent_map(em);
6031 * We don't allocate a new extent in the following cases
6033 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6035 * 2) The extent is marked as PREALLOC. We're good to go here and can
6036 * just use the extent.
6040 len = min(len, em->len - (start - em->start));
6041 lockstart = start + len;
6045 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6046 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6047 em->block_start != EXTENT_MAP_HOLE)) {
6052 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6053 type = BTRFS_ORDERED_PREALLOC;
6055 type = BTRFS_ORDERED_NOCOW;
6056 len = min(len, em->len - (start - em->start));
6057 block_start = em->block_start + (start - em->start);
6060 * we're not going to log anything, but we do need
6061 * to make sure the current transaction stays open
6062 * while we look for nocow cross refs
6064 trans = btrfs_join_transaction(root);
6068 if (can_nocow_odirect(trans, inode, start, len) == 1) {
6069 u64 orig_start = em->start;
6071 if (type == BTRFS_ORDERED_PREALLOC) {
6072 free_extent_map(em);
6073 em = create_pinned_em(inode, start, len,
6075 block_start, len, type);
6077 btrfs_end_transaction(trans, root);
6082 ret = btrfs_add_ordered_extent_dio(inode, start,
6083 block_start, len, len, type);
6084 btrfs_end_transaction(trans, root);
6086 free_extent_map(em);
6091 btrfs_end_transaction(trans, root);
6095 * this will cow the extent, reset the len in case we changed
6098 len = bh_result->b_size;
6099 em = btrfs_new_extent_direct(inode, em, start, len);
6104 len = min(len, em->len - (start - em->start));
6106 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6108 bh_result->b_size = len;
6109 bh_result->b_bdev = em->bdev;
6110 set_buffer_mapped(bh_result);
6112 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6113 set_buffer_new(bh_result);
6116 * Need to update the i_size under the extent lock so buffered
6117 * readers will get the updated i_size when we unlock.
6119 if (start + len > i_size_read(inode))
6120 i_size_write(inode, start + len);
6124 * In the case of write we need to clear and unlock the entire range,
6125 * in the case of read we need to unlock only the end area that we
6126 * aren't using if there is any left over space.
6128 if (lockstart < lockend) {
6129 if (create && len < lockend - lockstart) {
6130 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6131 lockstart + len - 1,
6132 unlock_bits | EXTENT_DEFRAG, 1, 0,
6133 &cached_state, GFP_NOFS);
6135 * Beside unlock, we also need to cleanup reserved space
6136 * for the left range by attaching EXTENT_DO_ACCOUNTING.
6138 clear_extent_bit(&BTRFS_I(inode)->io_tree,
6139 lockstart + len, lockend,
6140 unlock_bits | EXTENT_DO_ACCOUNTING |
6141 EXTENT_DEFRAG, 1, 0, NULL, GFP_NOFS);
6143 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6144 lockend, unlock_bits, 1, 0,
6145 &cached_state, GFP_NOFS);
6148 free_extent_state(cached_state);
6151 free_extent_map(em);
6157 unlock_bits |= EXTENT_DO_ACCOUNTING;
6159 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6160 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6164 struct btrfs_dio_private {
6165 struct inode *inode;
6171 /* number of bios pending for this dio */
6172 atomic_t pending_bios;
6177 struct bio *orig_bio;
6180 static void btrfs_endio_direct_read(struct bio *bio, int err)
6182 struct btrfs_dio_private *dip = bio->bi_private;
6183 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6184 struct bio_vec *bvec = bio->bi_io_vec;
6185 struct inode *inode = dip->inode;
6186 struct btrfs_root *root = BTRFS_I(inode)->root;
6189 start = dip->logical_offset;
6191 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6192 struct page *page = bvec->bv_page;
6195 u64 private = ~(u32)0;
6196 unsigned long flags;
6198 if (get_state_private(&BTRFS_I(inode)->io_tree,
6201 local_irq_save(flags);
6202 kaddr = kmap_atomic(page);
6203 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
6204 csum, bvec->bv_len);
6205 btrfs_csum_final(csum, (char *)&csum);
6206 kunmap_atomic(kaddr);
6207 local_irq_restore(flags);
6209 flush_dcache_page(bvec->bv_page);
6210 if (csum != private) {
6212 printk(KERN_ERR "btrfs csum failed ino %llu off"
6213 " %llu csum %u private %u\n",
6214 (unsigned long long)btrfs_ino(inode),
6215 (unsigned long long)start,
6216 csum, (unsigned)private);
6221 start += bvec->bv_len;
6223 } while (bvec <= bvec_end);
6225 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6226 dip->logical_offset + dip->bytes - 1);
6227 bio->bi_private = dip->private;
6231 /* If we had a csum failure make sure to clear the uptodate flag */
6233 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6234 dio_end_io(bio, err);
6237 static void btrfs_endio_direct_write(struct bio *bio, int err)
6239 struct btrfs_dio_private *dip = bio->bi_private;
6240 struct inode *inode = dip->inode;
6241 struct btrfs_root *root = BTRFS_I(inode)->root;
6242 struct btrfs_ordered_extent *ordered = NULL;
6243 u64 ordered_offset = dip->logical_offset;
6244 u64 ordered_bytes = dip->bytes;
6250 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6252 ordered_bytes, !err);
6256 ordered->work.func = finish_ordered_fn;
6257 ordered->work.flags = 0;
6258 btrfs_queue_worker(&root->fs_info->endio_write_workers,
6262 * our bio might span multiple ordered extents. If we haven't
6263 * completed the accounting for the whole dio, go back and try again
6265 if (ordered_offset < dip->logical_offset + dip->bytes) {
6266 ordered_bytes = dip->logical_offset + dip->bytes -
6272 bio->bi_private = dip->private;
6276 /* If we had an error make sure to clear the uptodate flag */
6278 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6279 dio_end_io(bio, err);
6282 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6283 struct bio *bio, int mirror_num,
6284 unsigned long bio_flags, u64 offset)
6287 struct btrfs_root *root = BTRFS_I(inode)->root;
6288 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6289 BUG_ON(ret); /* -ENOMEM */
6293 static void btrfs_end_dio_bio(struct bio *bio, int err)
6295 struct btrfs_dio_private *dip = bio->bi_private;
6298 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6299 "sector %#Lx len %u err no %d\n",
6300 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6301 (unsigned long long)bio->bi_sector, bio->bi_size, err);
6305 * before atomic variable goto zero, we must make sure
6306 * dip->errors is perceived to be set.
6308 smp_mb__before_atomic_dec();
6311 /* if there are more bios still pending for this dio, just exit */
6312 if (!atomic_dec_and_test(&dip->pending_bios))
6316 bio_io_error(dip->orig_bio);
6318 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6319 bio_endio(dip->orig_bio, 0);
6325 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6326 u64 first_sector, gfp_t gfp_flags)
6328 int nr_vecs = bio_get_nr_vecs(bdev);
6329 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6332 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6333 int rw, u64 file_offset, int skip_sum,
6336 int write = rw & REQ_WRITE;
6337 struct btrfs_root *root = BTRFS_I(inode)->root;
6343 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6351 if (write && async_submit) {
6352 ret = btrfs_wq_submit_bio(root->fs_info,
6353 inode, rw, bio, 0, 0,
6355 __btrfs_submit_bio_start_direct_io,
6356 __btrfs_submit_bio_done);
6360 * If we aren't doing async submit, calculate the csum of the
6363 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6366 } else if (!skip_sum) {
6367 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
6373 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6379 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6382 struct inode *inode = dip->inode;
6383 struct btrfs_root *root = BTRFS_I(inode)->root;
6385 struct bio *orig_bio = dip->orig_bio;
6386 struct bio_vec *bvec = orig_bio->bi_io_vec;
6387 u64 start_sector = orig_bio->bi_sector;
6388 u64 file_offset = dip->logical_offset;
6393 int async_submit = 0;
6395 map_length = orig_bio->bi_size;
6396 ret = btrfs_map_block(root->fs_info, READ, start_sector << 9,
6397 &map_length, NULL, 0);
6403 if (map_length >= orig_bio->bi_size) {
6409 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6412 bio->bi_private = dip;
6413 bio->bi_end_io = btrfs_end_dio_bio;
6414 atomic_inc(&dip->pending_bios);
6416 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6417 if (unlikely(map_length < submit_len + bvec->bv_len ||
6418 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6419 bvec->bv_offset) < bvec->bv_len)) {
6421 * inc the count before we submit the bio so
6422 * we know the end IO handler won't happen before
6423 * we inc the count. Otherwise, the dip might get freed
6424 * before we're done setting it up
6426 atomic_inc(&dip->pending_bios);
6427 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6428 file_offset, skip_sum,
6432 atomic_dec(&dip->pending_bios);
6436 start_sector += submit_len >> 9;
6437 file_offset += submit_len;
6442 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6443 start_sector, GFP_NOFS);
6446 bio->bi_private = dip;
6447 bio->bi_end_io = btrfs_end_dio_bio;
6449 map_length = orig_bio->bi_size;
6450 ret = btrfs_map_block(root->fs_info, READ,
6452 &map_length, NULL, 0);
6458 submit_len += bvec->bv_len;
6465 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6474 * before atomic variable goto zero, we must
6475 * make sure dip->errors is perceived to be set.
6477 smp_mb__before_atomic_dec();
6478 if (atomic_dec_and_test(&dip->pending_bios))
6479 bio_io_error(dip->orig_bio);
6481 /* bio_end_io() will handle error, so we needn't return it */
6485 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6488 struct btrfs_root *root = BTRFS_I(inode)->root;
6489 struct btrfs_dio_private *dip;
6490 struct bio_vec *bvec = bio->bi_io_vec;
6492 int write = rw & REQ_WRITE;
6495 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6497 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6503 dip->private = bio->bi_private;
6505 dip->logical_offset = file_offset;
6509 dip->bytes += bvec->bv_len;
6511 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6513 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6514 bio->bi_private = dip;
6516 dip->orig_bio = bio;
6517 atomic_set(&dip->pending_bios, 0);
6520 bio->bi_end_io = btrfs_endio_direct_write;
6522 bio->bi_end_io = btrfs_endio_direct_read;
6524 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6529 * If this is a write, we need to clean up the reserved space and kill
6530 * the ordered extent.
6533 struct btrfs_ordered_extent *ordered;
6534 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6535 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6536 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6537 btrfs_free_reserved_extent(root, ordered->start,
6539 btrfs_put_ordered_extent(ordered);
6540 btrfs_put_ordered_extent(ordered);
6542 bio_endio(bio, ret);
6545 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6546 const struct iovec *iov, loff_t offset,
6547 unsigned long nr_segs)
6553 unsigned blocksize_mask = root->sectorsize - 1;
6554 ssize_t retval = -EINVAL;
6555 loff_t end = offset;
6557 if (offset & blocksize_mask)
6560 /* Check the memory alignment. Blocks cannot straddle pages */
6561 for (seg = 0; seg < nr_segs; seg++) {
6562 addr = (unsigned long)iov[seg].iov_base;
6563 size = iov[seg].iov_len;
6565 if ((addr & blocksize_mask) || (size & blocksize_mask))
6568 /* If this is a write we don't need to check anymore */
6573 * Check to make sure we don't have duplicate iov_base's in this
6574 * iovec, if so return EINVAL, otherwise we'll get csum errors
6575 * when reading back.
6577 for (i = seg + 1; i < nr_segs; i++) {
6578 if (iov[seg].iov_base == iov[i].iov_base)
6587 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6588 const struct iovec *iov, loff_t offset,
6589 unsigned long nr_segs)
6591 struct file *file = iocb->ki_filp;
6592 struct inode *inode = file->f_mapping->host;
6594 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6598 return __blockdev_direct_IO(rw, iocb, inode,
6599 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6600 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6601 btrfs_submit_direct, 0);
6604 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
6606 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6607 __u64 start, __u64 len)
6611 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
6615 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6618 int btrfs_readpage(struct file *file, struct page *page)
6620 struct extent_io_tree *tree;
6621 tree = &BTRFS_I(page->mapping->host)->io_tree;
6622 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6625 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6627 struct extent_io_tree *tree;
6630 if (current->flags & PF_MEMALLOC) {
6631 redirty_page_for_writepage(wbc, page);
6635 tree = &BTRFS_I(page->mapping->host)->io_tree;
6636 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6639 int btrfs_writepages(struct address_space *mapping,
6640 struct writeback_control *wbc)
6642 struct extent_io_tree *tree;
6644 tree = &BTRFS_I(mapping->host)->io_tree;
6645 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6649 btrfs_readpages(struct file *file, struct address_space *mapping,
6650 struct list_head *pages, unsigned nr_pages)
6652 struct extent_io_tree *tree;
6653 tree = &BTRFS_I(mapping->host)->io_tree;
6654 return extent_readpages(tree, mapping, pages, nr_pages,
6657 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6659 struct extent_io_tree *tree;
6660 struct extent_map_tree *map;
6663 tree = &BTRFS_I(page->mapping->host)->io_tree;
6664 map = &BTRFS_I(page->mapping->host)->extent_tree;
6665 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6667 ClearPagePrivate(page);
6668 set_page_private(page, 0);
6669 page_cache_release(page);
6674 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6676 if (PageWriteback(page) || PageDirty(page))
6678 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6681 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6683 struct inode *inode = page->mapping->host;
6684 struct extent_io_tree *tree;
6685 struct btrfs_ordered_extent *ordered;
6686 struct extent_state *cached_state = NULL;
6687 u64 page_start = page_offset(page);
6688 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6691 * we have the page locked, so new writeback can't start,
6692 * and the dirty bit won't be cleared while we are here.
6694 * Wait for IO on this page so that we can safely clear
6695 * the PagePrivate2 bit and do ordered accounting
6697 wait_on_page_writeback(page);
6699 tree = &BTRFS_I(inode)->io_tree;
6701 btrfs_releasepage(page, GFP_NOFS);
6704 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6705 ordered = btrfs_lookup_ordered_extent(inode,
6709 * IO on this page will never be started, so we need
6710 * to account for any ordered extents now
6712 clear_extent_bit(tree, page_start, page_end,
6713 EXTENT_DIRTY | EXTENT_DELALLOC |
6714 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
6715 EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
6717 * whoever cleared the private bit is responsible
6718 * for the finish_ordered_io
6720 if (TestClearPagePrivate2(page) &&
6721 btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6722 PAGE_CACHE_SIZE, 1)) {
6723 btrfs_finish_ordered_io(ordered);
6725 btrfs_put_ordered_extent(ordered);
6726 cached_state = NULL;
6727 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6729 clear_extent_bit(tree, page_start, page_end,
6730 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6731 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
6732 &cached_state, GFP_NOFS);
6733 __btrfs_releasepage(page, GFP_NOFS);
6735 ClearPageChecked(page);
6736 if (PagePrivate(page)) {
6737 ClearPagePrivate(page);
6738 set_page_private(page, 0);
6739 page_cache_release(page);
6744 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6745 * called from a page fault handler when a page is first dirtied. Hence we must
6746 * be careful to check for EOF conditions here. We set the page up correctly
6747 * for a written page which means we get ENOSPC checking when writing into
6748 * holes and correct delalloc and unwritten extent mapping on filesystems that
6749 * support these features.
6751 * We are not allowed to take the i_mutex here so we have to play games to
6752 * protect against truncate races as the page could now be beyond EOF. Because
6753 * vmtruncate() writes the inode size before removing pages, once we have the
6754 * page lock we can determine safely if the page is beyond EOF. If it is not
6755 * beyond EOF, then the page is guaranteed safe against truncation until we
6758 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6760 struct page *page = vmf->page;
6761 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6762 struct btrfs_root *root = BTRFS_I(inode)->root;
6763 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6764 struct btrfs_ordered_extent *ordered;
6765 struct extent_state *cached_state = NULL;
6767 unsigned long zero_start;
6774 sb_start_pagefault(inode->i_sb);
6775 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6777 ret = file_update_time(vma->vm_file);
6783 else /* -ENOSPC, -EIO, etc */
6784 ret = VM_FAULT_SIGBUS;
6790 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6793 size = i_size_read(inode);
6794 page_start = page_offset(page);
6795 page_end = page_start + PAGE_CACHE_SIZE - 1;
6797 if ((page->mapping != inode->i_mapping) ||
6798 (page_start >= size)) {
6799 /* page got truncated out from underneath us */
6802 wait_on_page_writeback(page);
6804 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6805 set_page_extent_mapped(page);
6808 * we can't set the delalloc bits if there are pending ordered
6809 * extents. Drop our locks and wait for them to finish
6811 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6813 unlock_extent_cached(io_tree, page_start, page_end,
6814 &cached_state, GFP_NOFS);
6816 btrfs_start_ordered_extent(inode, ordered, 1);
6817 btrfs_put_ordered_extent(ordered);
6822 * XXX - page_mkwrite gets called every time the page is dirtied, even
6823 * if it was already dirty, so for space accounting reasons we need to
6824 * clear any delalloc bits for the range we are fixing to save. There
6825 * is probably a better way to do this, but for now keep consistent with
6826 * prepare_pages in the normal write path.
6828 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6829 EXTENT_DIRTY | EXTENT_DELALLOC |
6830 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
6831 0, 0, &cached_state, GFP_NOFS);
6833 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6836 unlock_extent_cached(io_tree, page_start, page_end,
6837 &cached_state, GFP_NOFS);
6838 ret = VM_FAULT_SIGBUS;
6843 /* page is wholly or partially inside EOF */
6844 if (page_start + PAGE_CACHE_SIZE > size)
6845 zero_start = size & ~PAGE_CACHE_MASK;
6847 zero_start = PAGE_CACHE_SIZE;
6849 if (zero_start != PAGE_CACHE_SIZE) {
6851 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6852 flush_dcache_page(page);
6855 ClearPageChecked(page);
6856 set_page_dirty(page);
6857 SetPageUptodate(page);
6859 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6860 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6861 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
6863 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6867 sb_end_pagefault(inode->i_sb);
6868 return VM_FAULT_LOCKED;
6872 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6874 sb_end_pagefault(inode->i_sb);
6878 static int btrfs_truncate(struct inode *inode)
6880 struct btrfs_root *root = BTRFS_I(inode)->root;
6881 struct btrfs_block_rsv *rsv;
6884 struct btrfs_trans_handle *trans;
6885 u64 mask = root->sectorsize - 1;
6886 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6888 ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
6892 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6893 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6896 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6897 * 3 things going on here
6899 * 1) We need to reserve space for our orphan item and the space to
6900 * delete our orphan item. Lord knows we don't want to have a dangling
6901 * orphan item because we didn't reserve space to remove it.
6903 * 2) We need to reserve space to update our inode.
6905 * 3) We need to have something to cache all the space that is going to
6906 * be free'd up by the truncate operation, but also have some slack
6907 * space reserved in case it uses space during the truncate (thank you
6908 * very much snapshotting).
6910 * And we need these to all be seperate. The fact is we can use alot of
6911 * space doing the truncate, and we have no earthly idea how much space
6912 * we will use, so we need the truncate reservation to be seperate so it
6913 * doesn't end up using space reserved for updating the inode or
6914 * removing the orphan item. We also need to be able to stop the
6915 * transaction and start a new one, which means we need to be able to
6916 * update the inode several times, and we have no idea of knowing how
6917 * many times that will be, so we can't just reserve 1 item for the
6918 * entirety of the opration, so that has to be done seperately as well.
6919 * Then there is the orphan item, which does indeed need to be held on
6920 * to for the whole operation, and we need nobody to touch this reserved
6921 * space except the orphan code.
6923 * So that leaves us with
6925 * 1) root->orphan_block_rsv - for the orphan deletion.
6926 * 2) rsv - for the truncate reservation, which we will steal from the
6927 * transaction reservation.
6928 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6929 * updating the inode.
6931 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
6934 rsv->size = min_size;
6938 * 1 for the truncate slack space
6939 * 1 for the orphan item we're going to add
6940 * 1 for the orphan item deletion
6941 * 1 for updating the inode.
6943 trans = btrfs_start_transaction(root, 4);
6944 if (IS_ERR(trans)) {
6945 err = PTR_ERR(trans);
6949 /* Migrate the slack space for the truncate to our reserve */
6950 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6954 ret = btrfs_orphan_add(trans, inode);
6956 btrfs_end_transaction(trans, root);
6961 * setattr is responsible for setting the ordered_data_close flag,
6962 * but that is only tested during the last file release. That
6963 * could happen well after the next commit, leaving a great big
6964 * window where new writes may get lost if someone chooses to write
6965 * to this file after truncating to zero
6967 * The inode doesn't have any dirty data here, and so if we commit
6968 * this is a noop. If someone immediately starts writing to the inode
6969 * it is very likely we'll catch some of their writes in this
6970 * transaction, and the commit will find this file on the ordered
6971 * data list with good things to send down.
6973 * This is a best effort solution, there is still a window where
6974 * using truncate to replace the contents of the file will
6975 * end up with a zero length file after a crash.
6977 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
6978 &BTRFS_I(inode)->runtime_flags))
6979 btrfs_add_ordered_operation(trans, root, inode);
6982 * So if we truncate and then write and fsync we normally would just
6983 * write the extents that changed, which is a problem if we need to
6984 * first truncate that entire inode. So set this flag so we write out
6985 * all of the extents in the inode to the sync log so we're completely
6988 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6989 trans->block_rsv = rsv;
6992 ret = btrfs_truncate_inode_items(trans, root, inode,
6994 BTRFS_EXTENT_DATA_KEY);
6995 if (ret != -ENOSPC) {
7000 trans->block_rsv = &root->fs_info->trans_block_rsv;
7001 ret = btrfs_update_inode(trans, root, inode);
7007 btrfs_end_transaction(trans, root);
7008 btrfs_btree_balance_dirty(root);
7010 trans = btrfs_start_transaction(root, 2);
7011 if (IS_ERR(trans)) {
7012 ret = err = PTR_ERR(trans);
7017 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7019 BUG_ON(ret); /* shouldn't happen */
7020 trans->block_rsv = rsv;
7023 if (ret == 0 && inode->i_nlink > 0) {
7024 trans->block_rsv = root->orphan_block_rsv;
7025 ret = btrfs_orphan_del(trans, inode);
7028 } else if (ret && inode->i_nlink > 0) {
7030 * Failed to do the truncate, remove us from the in memory
7033 ret = btrfs_orphan_del(NULL, inode);
7037 trans->block_rsv = &root->fs_info->trans_block_rsv;
7038 ret = btrfs_update_inode(trans, root, inode);
7042 ret = btrfs_end_transaction(trans, root);
7043 btrfs_btree_balance_dirty(root);
7047 btrfs_free_block_rsv(root, rsv);
7056 * create a new subvolume directory/inode (helper for the ioctl).
7058 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7059 struct btrfs_root *new_root, u64 new_dirid)
7061 struct inode *inode;
7065 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7066 new_dirid, new_dirid,
7067 S_IFDIR | (~current_umask() & S_IRWXUGO),
7070 return PTR_ERR(inode);
7071 inode->i_op = &btrfs_dir_inode_operations;
7072 inode->i_fop = &btrfs_dir_file_operations;
7074 set_nlink(inode, 1);
7075 btrfs_i_size_write(inode, 0);
7077 err = btrfs_update_inode(trans, new_root, inode);
7083 struct inode *btrfs_alloc_inode(struct super_block *sb)
7085 struct btrfs_inode *ei;
7086 struct inode *inode;
7088 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7095 ei->last_sub_trans = 0;
7096 ei->logged_trans = 0;
7097 ei->delalloc_bytes = 0;
7098 ei->disk_i_size = 0;
7101 ei->index_cnt = (u64)-1;
7102 ei->last_unlink_trans = 0;
7103 ei->last_log_commit = 0;
7105 spin_lock_init(&ei->lock);
7106 ei->outstanding_extents = 0;
7107 ei->reserved_extents = 0;
7109 ei->runtime_flags = 0;
7110 ei->force_compress = BTRFS_COMPRESS_NONE;
7112 ei->delayed_node = NULL;
7114 inode = &ei->vfs_inode;
7115 extent_map_tree_init(&ei->extent_tree);
7116 extent_io_tree_init(&ei->io_tree, &inode->i_data);
7117 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7118 ei->io_tree.track_uptodate = 1;
7119 ei->io_failure_tree.track_uptodate = 1;
7120 mutex_init(&ei->log_mutex);
7121 mutex_init(&ei->delalloc_mutex);
7122 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7123 INIT_LIST_HEAD(&ei->delalloc_inodes);
7124 INIT_LIST_HEAD(&ei->ordered_operations);
7125 RB_CLEAR_NODE(&ei->rb_node);
7130 static void btrfs_i_callback(struct rcu_head *head)
7132 struct inode *inode = container_of(head, struct inode, i_rcu);
7133 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7136 void btrfs_destroy_inode(struct inode *inode)
7138 struct btrfs_ordered_extent *ordered;
7139 struct btrfs_root *root = BTRFS_I(inode)->root;
7141 WARN_ON(!hlist_empty(&inode->i_dentry));
7142 WARN_ON(inode->i_data.nrpages);
7143 WARN_ON(BTRFS_I(inode)->outstanding_extents);
7144 WARN_ON(BTRFS_I(inode)->reserved_extents);
7145 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7146 WARN_ON(BTRFS_I(inode)->csum_bytes);
7149 * This can happen where we create an inode, but somebody else also
7150 * created the same inode and we need to destroy the one we already
7157 * Make sure we're properly removed from the ordered operation
7161 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7162 spin_lock(&root->fs_info->ordered_extent_lock);
7163 list_del_init(&BTRFS_I(inode)->ordered_operations);
7164 spin_unlock(&root->fs_info->ordered_extent_lock);
7167 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7168 &BTRFS_I(inode)->runtime_flags)) {
7169 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7170 (unsigned long long)btrfs_ino(inode));
7171 atomic_dec(&root->orphan_inodes);
7175 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7179 printk(KERN_ERR "btrfs found ordered "
7180 "extent %llu %llu on inode cleanup\n",
7181 (unsigned long long)ordered->file_offset,
7182 (unsigned long long)ordered->len);
7183 btrfs_remove_ordered_extent(inode, ordered);
7184 btrfs_put_ordered_extent(ordered);
7185 btrfs_put_ordered_extent(ordered);
7188 inode_tree_del(inode);
7189 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7191 btrfs_remove_delayed_node(inode);
7192 call_rcu(&inode->i_rcu, btrfs_i_callback);
7195 int btrfs_drop_inode(struct inode *inode)
7197 struct btrfs_root *root = BTRFS_I(inode)->root;
7199 if (btrfs_root_refs(&root->root_item) == 0 &&
7200 !btrfs_is_free_space_inode(inode))
7203 return generic_drop_inode(inode);
7206 static void init_once(void *foo)
7208 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7210 inode_init_once(&ei->vfs_inode);
7213 void btrfs_destroy_cachep(void)
7216 * Make sure all delayed rcu free inodes are flushed before we
7220 if (btrfs_inode_cachep)
7221 kmem_cache_destroy(btrfs_inode_cachep);
7222 if (btrfs_trans_handle_cachep)
7223 kmem_cache_destroy(btrfs_trans_handle_cachep);
7224 if (btrfs_transaction_cachep)
7225 kmem_cache_destroy(btrfs_transaction_cachep);
7226 if (btrfs_path_cachep)
7227 kmem_cache_destroy(btrfs_path_cachep);
7228 if (btrfs_free_space_cachep)
7229 kmem_cache_destroy(btrfs_free_space_cachep);
7230 if (btrfs_delalloc_work_cachep)
7231 kmem_cache_destroy(btrfs_delalloc_work_cachep);
7234 int btrfs_init_cachep(void)
7236 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
7237 sizeof(struct btrfs_inode), 0,
7238 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7239 if (!btrfs_inode_cachep)
7242 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
7243 sizeof(struct btrfs_trans_handle), 0,
7244 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7245 if (!btrfs_trans_handle_cachep)
7248 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
7249 sizeof(struct btrfs_transaction), 0,
7250 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7251 if (!btrfs_transaction_cachep)
7254 btrfs_path_cachep = kmem_cache_create("btrfs_path",
7255 sizeof(struct btrfs_path), 0,
7256 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7257 if (!btrfs_path_cachep)
7260 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
7261 sizeof(struct btrfs_free_space), 0,
7262 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7263 if (!btrfs_free_space_cachep)
7266 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
7267 sizeof(struct btrfs_delalloc_work), 0,
7268 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
7270 if (!btrfs_delalloc_work_cachep)
7275 btrfs_destroy_cachep();
7279 static int btrfs_getattr(struct vfsmount *mnt,
7280 struct dentry *dentry, struct kstat *stat)
7282 struct inode *inode = dentry->d_inode;
7283 u32 blocksize = inode->i_sb->s_blocksize;
7285 generic_fillattr(inode, stat);
7286 stat->dev = BTRFS_I(inode)->root->anon_dev;
7287 stat->blksize = PAGE_CACHE_SIZE;
7288 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7289 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
7294 * If a file is moved, it will inherit the cow and compression flags of the new
7297 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7299 struct btrfs_inode *b_dir = BTRFS_I(dir);
7300 struct btrfs_inode *b_inode = BTRFS_I(inode);
7302 if (b_dir->flags & BTRFS_INODE_NODATACOW)
7303 b_inode->flags |= BTRFS_INODE_NODATACOW;
7305 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7307 if (b_dir->flags & BTRFS_INODE_COMPRESS) {
7308 b_inode->flags |= BTRFS_INODE_COMPRESS;
7309 b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
7311 b_inode->flags &= ~(BTRFS_INODE_COMPRESS |
7312 BTRFS_INODE_NOCOMPRESS);
7316 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7317 struct inode *new_dir, struct dentry *new_dentry)
7319 struct btrfs_trans_handle *trans;
7320 struct btrfs_root *root = BTRFS_I(old_dir)->root;
7321 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7322 struct inode *new_inode = new_dentry->d_inode;
7323 struct inode *old_inode = old_dentry->d_inode;
7324 struct timespec ctime = CURRENT_TIME;
7328 u64 old_ino = btrfs_ino(old_inode);
7330 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7333 /* we only allow rename subvolume link between subvolumes */
7334 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7337 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7338 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7341 if (S_ISDIR(old_inode->i_mode) && new_inode &&
7342 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7345 * we're using rename to replace one file with another.
7346 * and the replacement file is large. Start IO on it now so
7347 * we don't add too much work to the end of the transaction
7349 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7350 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7351 filemap_flush(old_inode->i_mapping);
7353 /* close the racy window with snapshot create/destroy ioctl */
7354 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7355 down_read(&root->fs_info->subvol_sem);
7357 * We want to reserve the absolute worst case amount of items. So if
7358 * both inodes are subvols and we need to unlink them then that would
7359 * require 4 item modifications, but if they are both normal inodes it
7360 * would require 5 item modifications, so we'll assume their normal
7361 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7362 * should cover the worst case number of items we'll modify.
7364 trans = btrfs_start_transaction(root, 20);
7365 if (IS_ERR(trans)) {
7366 ret = PTR_ERR(trans);
7371 btrfs_record_root_in_trans(trans, dest);
7373 ret = btrfs_set_inode_index(new_dir, &index);
7377 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7378 /* force full log commit if subvolume involved. */
7379 root->fs_info->last_trans_log_full_commit = trans->transid;
7381 ret = btrfs_insert_inode_ref(trans, dest,
7382 new_dentry->d_name.name,
7383 new_dentry->d_name.len,
7385 btrfs_ino(new_dir), index);
7389 * this is an ugly little race, but the rename is required
7390 * to make sure that if we crash, the inode is either at the
7391 * old name or the new one. pinning the log transaction lets
7392 * us make sure we don't allow a log commit to come in after
7393 * we unlink the name but before we add the new name back in.
7395 btrfs_pin_log_trans(root);
7398 * make sure the inode gets flushed if it is replacing
7401 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7402 btrfs_add_ordered_operation(trans, root, old_inode);
7404 inode_inc_iversion(old_dir);
7405 inode_inc_iversion(new_dir);
7406 inode_inc_iversion(old_inode);
7407 old_dir->i_ctime = old_dir->i_mtime = ctime;
7408 new_dir->i_ctime = new_dir->i_mtime = ctime;
7409 old_inode->i_ctime = ctime;
7411 if (old_dentry->d_parent != new_dentry->d_parent)
7412 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7414 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7415 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7416 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7417 old_dentry->d_name.name,
7418 old_dentry->d_name.len);
7420 ret = __btrfs_unlink_inode(trans, root, old_dir,
7421 old_dentry->d_inode,
7422 old_dentry->d_name.name,
7423 old_dentry->d_name.len);
7425 ret = btrfs_update_inode(trans, root, old_inode);
7428 btrfs_abort_transaction(trans, root, ret);
7433 inode_inc_iversion(new_inode);
7434 new_inode->i_ctime = CURRENT_TIME;
7435 if (unlikely(btrfs_ino(new_inode) ==
7436 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7437 root_objectid = BTRFS_I(new_inode)->location.objectid;
7438 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7440 new_dentry->d_name.name,
7441 new_dentry->d_name.len);
7442 BUG_ON(new_inode->i_nlink == 0);
7444 ret = btrfs_unlink_inode(trans, dest, new_dir,
7445 new_dentry->d_inode,
7446 new_dentry->d_name.name,
7447 new_dentry->d_name.len);
7449 if (!ret && new_inode->i_nlink == 0) {
7450 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7454 btrfs_abort_transaction(trans, root, ret);
7459 fixup_inode_flags(new_dir, old_inode);
7461 ret = btrfs_add_link(trans, new_dir, old_inode,
7462 new_dentry->d_name.name,
7463 new_dentry->d_name.len, 0, index);
7465 btrfs_abort_transaction(trans, root, ret);
7469 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7470 struct dentry *parent = new_dentry->d_parent;
7471 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7472 btrfs_end_log_trans(root);
7475 btrfs_end_transaction(trans, root);
7477 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7478 up_read(&root->fs_info->subvol_sem);
7483 static void btrfs_run_delalloc_work(struct btrfs_work *work)
7485 struct btrfs_delalloc_work *delalloc_work;
7487 delalloc_work = container_of(work, struct btrfs_delalloc_work,
7489 if (delalloc_work->wait)
7490 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
7492 filemap_flush(delalloc_work->inode->i_mapping);
7494 if (delalloc_work->delay_iput)
7495 btrfs_add_delayed_iput(delalloc_work->inode);
7497 iput(delalloc_work->inode);
7498 complete(&delalloc_work->completion);
7501 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
7502 int wait, int delay_iput)
7504 struct btrfs_delalloc_work *work;
7506 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
7510 init_completion(&work->completion);
7511 INIT_LIST_HEAD(&work->list);
7512 work->inode = inode;
7514 work->delay_iput = delay_iput;
7515 work->work.func = btrfs_run_delalloc_work;
7520 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
7522 wait_for_completion(&work->completion);
7523 kmem_cache_free(btrfs_delalloc_work_cachep, work);
7527 * some fairly slow code that needs optimization. This walks the list
7528 * of all the inodes with pending delalloc and forces them to disk.
7530 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7532 struct list_head *head = &root->fs_info->delalloc_inodes;
7533 struct btrfs_inode *binode;
7534 struct inode *inode;
7535 struct btrfs_delalloc_work *work, *next;
7536 struct list_head works;
7539 if (root->fs_info->sb->s_flags & MS_RDONLY)
7542 INIT_LIST_HEAD(&works);
7544 spin_lock(&root->fs_info->delalloc_lock);
7545 while (!list_empty(head)) {
7546 binode = list_entry(head->next, struct btrfs_inode,
7548 inode = igrab(&binode->vfs_inode);
7550 list_del_init(&binode->delalloc_inodes);
7551 spin_unlock(&root->fs_info->delalloc_lock);
7553 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
7558 list_add_tail(&work->list, &works);
7559 btrfs_queue_worker(&root->fs_info->flush_workers,
7563 spin_lock(&root->fs_info->delalloc_lock);
7565 spin_unlock(&root->fs_info->delalloc_lock);
7567 /* the filemap_flush will queue IO into the worker threads, but
7568 * we have to make sure the IO is actually started and that
7569 * ordered extents get created before we return
7571 atomic_inc(&root->fs_info->async_submit_draining);
7572 while (atomic_read(&root->fs_info->nr_async_submits) ||
7573 atomic_read(&root->fs_info->async_delalloc_pages)) {
7574 wait_event(root->fs_info->async_submit_wait,
7575 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7576 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7578 atomic_dec(&root->fs_info->async_submit_draining);
7580 list_for_each_entry_safe(work, next, &works, list) {
7581 list_del_init(&work->list);
7582 btrfs_wait_and_free_delalloc_work(work);
7587 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7588 const char *symname)
7590 struct btrfs_trans_handle *trans;
7591 struct btrfs_root *root = BTRFS_I(dir)->root;
7592 struct btrfs_path *path;
7593 struct btrfs_key key;
7594 struct inode *inode = NULL;
7602 struct btrfs_file_extent_item *ei;
7603 struct extent_buffer *leaf;
7605 name_len = strlen(symname) + 1;
7606 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7607 return -ENAMETOOLONG;
7610 * 2 items for inode item and ref
7611 * 2 items for dir items
7612 * 1 item for xattr if selinux is on
7614 trans = btrfs_start_transaction(root, 5);
7616 return PTR_ERR(trans);
7618 err = btrfs_find_free_ino(root, &objectid);
7622 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7623 dentry->d_name.len, btrfs_ino(dir), objectid,
7624 S_IFLNK|S_IRWXUGO, &index);
7625 if (IS_ERR(inode)) {
7626 err = PTR_ERR(inode);
7630 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7637 * If the active LSM wants to access the inode during
7638 * d_instantiate it needs these. Smack checks to see
7639 * if the filesystem supports xattrs by looking at the
7642 inode->i_fop = &btrfs_file_operations;
7643 inode->i_op = &btrfs_file_inode_operations;
7645 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7649 inode->i_mapping->a_ops = &btrfs_aops;
7650 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7651 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7656 path = btrfs_alloc_path();
7662 key.objectid = btrfs_ino(inode);
7664 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7665 datasize = btrfs_file_extent_calc_inline_size(name_len);
7666 err = btrfs_insert_empty_item(trans, root, path, &key,
7670 btrfs_free_path(path);
7673 leaf = path->nodes[0];
7674 ei = btrfs_item_ptr(leaf, path->slots[0],
7675 struct btrfs_file_extent_item);
7676 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7677 btrfs_set_file_extent_type(leaf, ei,
7678 BTRFS_FILE_EXTENT_INLINE);
7679 btrfs_set_file_extent_encryption(leaf, ei, 0);
7680 btrfs_set_file_extent_compression(leaf, ei, 0);
7681 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7682 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7684 ptr = btrfs_file_extent_inline_start(ei);
7685 write_extent_buffer(leaf, symname, ptr, name_len);
7686 btrfs_mark_buffer_dirty(leaf);
7687 btrfs_free_path(path);
7689 inode->i_op = &btrfs_symlink_inode_operations;
7690 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7691 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7692 inode_set_bytes(inode, name_len);
7693 btrfs_i_size_write(inode, name_len - 1);
7694 err = btrfs_update_inode(trans, root, inode);
7700 d_instantiate(dentry, inode);
7701 btrfs_end_transaction(trans, root);
7703 inode_dec_link_count(inode);
7706 btrfs_btree_balance_dirty(root);
7710 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7711 u64 start, u64 num_bytes, u64 min_size,
7712 loff_t actual_len, u64 *alloc_hint,
7713 struct btrfs_trans_handle *trans)
7715 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7716 struct extent_map *em;
7717 struct btrfs_root *root = BTRFS_I(inode)->root;
7718 struct btrfs_key ins;
7719 u64 cur_offset = start;
7722 bool own_trans = true;
7726 while (num_bytes > 0) {
7728 trans = btrfs_start_transaction(root, 3);
7729 if (IS_ERR(trans)) {
7730 ret = PTR_ERR(trans);
7735 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7736 0, *alloc_hint, &ins, 1);
7739 btrfs_end_transaction(trans, root);
7743 ret = insert_reserved_file_extent(trans, inode,
7744 cur_offset, ins.objectid,
7745 ins.offset, ins.offset,
7746 ins.offset, 0, 0, 0,
7747 BTRFS_FILE_EXTENT_PREALLOC);
7749 btrfs_abort_transaction(trans, root, ret);
7751 btrfs_end_transaction(trans, root);
7754 btrfs_drop_extent_cache(inode, cur_offset,
7755 cur_offset + ins.offset -1, 0);
7757 em = alloc_extent_map();
7759 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
7760 &BTRFS_I(inode)->runtime_flags);
7764 em->start = cur_offset;
7765 em->orig_start = cur_offset;
7766 em->len = ins.offset;
7767 em->block_start = ins.objectid;
7768 em->block_len = ins.offset;
7769 em->bdev = root->fs_info->fs_devices->latest_bdev;
7770 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7771 em->generation = trans->transid;
7774 write_lock(&em_tree->lock);
7775 ret = add_extent_mapping(em_tree, em);
7777 list_move(&em->list,
7778 &em_tree->modified_extents);
7779 write_unlock(&em_tree->lock);
7782 btrfs_drop_extent_cache(inode, cur_offset,
7783 cur_offset + ins.offset - 1,
7786 free_extent_map(em);
7788 num_bytes -= ins.offset;
7789 cur_offset += ins.offset;
7790 *alloc_hint = ins.objectid + ins.offset;
7792 inode_inc_iversion(inode);
7793 inode->i_ctime = CURRENT_TIME;
7794 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7795 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7796 (actual_len > inode->i_size) &&
7797 (cur_offset > inode->i_size)) {
7798 if (cur_offset > actual_len)
7799 i_size = actual_len;
7801 i_size = cur_offset;
7802 i_size_write(inode, i_size);
7803 btrfs_ordered_update_i_size(inode, i_size, NULL);
7806 ret = btrfs_update_inode(trans, root, inode);
7809 btrfs_abort_transaction(trans, root, ret);
7811 btrfs_end_transaction(trans, root);
7816 btrfs_end_transaction(trans, root);
7821 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7822 u64 start, u64 num_bytes, u64 min_size,
7823 loff_t actual_len, u64 *alloc_hint)
7825 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7826 min_size, actual_len, alloc_hint,
7830 int btrfs_prealloc_file_range_trans(struct inode *inode,
7831 struct btrfs_trans_handle *trans, int mode,
7832 u64 start, u64 num_bytes, u64 min_size,
7833 loff_t actual_len, u64 *alloc_hint)
7835 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7836 min_size, actual_len, alloc_hint, trans);
7839 static int btrfs_set_page_dirty(struct page *page)
7841 return __set_page_dirty_nobuffers(page);
7844 static int btrfs_permission(struct inode *inode, int mask)
7846 struct btrfs_root *root = BTRFS_I(inode)->root;
7847 umode_t mode = inode->i_mode;
7849 if (mask & MAY_WRITE &&
7850 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7851 if (btrfs_root_readonly(root))
7853 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7856 return generic_permission(inode, mask);
7859 static const struct inode_operations btrfs_dir_inode_operations = {
7860 .getattr = btrfs_getattr,
7861 .lookup = btrfs_lookup,
7862 .create = btrfs_create,
7863 .unlink = btrfs_unlink,
7865 .mkdir = btrfs_mkdir,
7866 .rmdir = btrfs_rmdir,
7867 .rename = btrfs_rename,
7868 .symlink = btrfs_symlink,
7869 .setattr = btrfs_setattr,
7870 .mknod = btrfs_mknod,
7871 .setxattr = btrfs_setxattr,
7872 .getxattr = btrfs_getxattr,
7873 .listxattr = btrfs_listxattr,
7874 .removexattr = btrfs_removexattr,
7875 .permission = btrfs_permission,
7876 .get_acl = btrfs_get_acl,
7878 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7879 .lookup = btrfs_lookup,
7880 .permission = btrfs_permission,
7881 .get_acl = btrfs_get_acl,
7884 static const struct file_operations btrfs_dir_file_operations = {
7885 .llseek = generic_file_llseek,
7886 .read = generic_read_dir,
7887 .readdir = btrfs_real_readdir,
7888 .unlocked_ioctl = btrfs_ioctl,
7889 #ifdef CONFIG_COMPAT
7890 .compat_ioctl = btrfs_ioctl,
7892 .release = btrfs_release_file,
7893 .fsync = btrfs_sync_file,
7896 static struct extent_io_ops btrfs_extent_io_ops = {
7897 .fill_delalloc = run_delalloc_range,
7898 .submit_bio_hook = btrfs_submit_bio_hook,
7899 .merge_bio_hook = btrfs_merge_bio_hook,
7900 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7901 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7902 .writepage_start_hook = btrfs_writepage_start_hook,
7903 .set_bit_hook = btrfs_set_bit_hook,
7904 .clear_bit_hook = btrfs_clear_bit_hook,
7905 .merge_extent_hook = btrfs_merge_extent_hook,
7906 .split_extent_hook = btrfs_split_extent_hook,
7910 * btrfs doesn't support the bmap operation because swapfiles
7911 * use bmap to make a mapping of extents in the file. They assume
7912 * these extents won't change over the life of the file and they
7913 * use the bmap result to do IO directly to the drive.
7915 * the btrfs bmap call would return logical addresses that aren't
7916 * suitable for IO and they also will change frequently as COW
7917 * operations happen. So, swapfile + btrfs == corruption.
7919 * For now we're avoiding this by dropping bmap.
7921 static const struct address_space_operations btrfs_aops = {
7922 .readpage = btrfs_readpage,
7923 .writepage = btrfs_writepage,
7924 .writepages = btrfs_writepages,
7925 .readpages = btrfs_readpages,
7926 .direct_IO = btrfs_direct_IO,
7927 .invalidatepage = btrfs_invalidatepage,
7928 .releasepage = btrfs_releasepage,
7929 .set_page_dirty = btrfs_set_page_dirty,
7930 .error_remove_page = generic_error_remove_page,
7933 static const struct address_space_operations btrfs_symlink_aops = {
7934 .readpage = btrfs_readpage,
7935 .writepage = btrfs_writepage,
7936 .invalidatepage = btrfs_invalidatepage,
7937 .releasepage = btrfs_releasepage,
7940 static const struct inode_operations btrfs_file_inode_operations = {
7941 .getattr = btrfs_getattr,
7942 .setattr = btrfs_setattr,
7943 .setxattr = btrfs_setxattr,
7944 .getxattr = btrfs_getxattr,
7945 .listxattr = btrfs_listxattr,
7946 .removexattr = btrfs_removexattr,
7947 .permission = btrfs_permission,
7948 .fiemap = btrfs_fiemap,
7949 .get_acl = btrfs_get_acl,
7950 .update_time = btrfs_update_time,
7952 static const struct inode_operations btrfs_special_inode_operations = {
7953 .getattr = btrfs_getattr,
7954 .setattr = btrfs_setattr,
7955 .permission = btrfs_permission,
7956 .setxattr = btrfs_setxattr,
7957 .getxattr = btrfs_getxattr,
7958 .listxattr = btrfs_listxattr,
7959 .removexattr = btrfs_removexattr,
7960 .get_acl = btrfs_get_acl,
7961 .update_time = btrfs_update_time,
7963 static const struct inode_operations btrfs_symlink_inode_operations = {
7964 .readlink = generic_readlink,
7965 .follow_link = page_follow_link_light,
7966 .put_link = page_put_link,
7967 .getattr = btrfs_getattr,
7968 .setattr = btrfs_setattr,
7969 .permission = btrfs_permission,
7970 .setxattr = btrfs_setxattr,
7971 .getxattr = btrfs_getxattr,
7972 .listxattr = btrfs_listxattr,
7973 .removexattr = btrfs_removexattr,
7974 .get_acl = btrfs_get_acl,
7975 .update_time = btrfs_update_time,
7978 const struct dentry_operations btrfs_dentry_operations = {
7979 .d_delete = btrfs_dentry_delete,
7980 .d_release = btrfs_dentry_release,