2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include <linux/mount.h>
45 #include "transaction.h"
46 #include "btrfs_inode.h"
48 #include "print-tree.h"
49 #include "ordered-data.h"
53 #include "compression.h"
55 #include "free-space-cache.h"
56 #include "inode-map.h"
58 struct btrfs_iget_args {
60 struct btrfs_root *root;
63 static const struct inode_operations btrfs_dir_inode_operations;
64 static const struct inode_operations btrfs_symlink_inode_operations;
65 static const struct inode_operations btrfs_dir_ro_inode_operations;
66 static const struct inode_operations btrfs_special_inode_operations;
67 static const struct inode_operations btrfs_file_inode_operations;
68 static const struct address_space_operations btrfs_aops;
69 static const struct address_space_operations btrfs_symlink_aops;
70 static const struct file_operations btrfs_dir_file_operations;
71 static struct extent_io_ops btrfs_extent_io_ops;
73 static struct kmem_cache *btrfs_inode_cachep;
74 struct kmem_cache *btrfs_trans_handle_cachep;
75 struct kmem_cache *btrfs_transaction_cachep;
76 struct kmem_cache *btrfs_path_cachep;
77 struct kmem_cache *btrfs_free_space_cachep;
80 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
81 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
82 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
83 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
84 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
85 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
86 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
87 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
90 static int btrfs_setsize(struct inode *inode, loff_t newsize);
91 static int btrfs_truncate(struct inode *inode);
92 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
93 static noinline int cow_file_range(struct inode *inode,
94 struct page *locked_page,
95 u64 start, u64 end, int *page_started,
96 unsigned long *nr_written, int unlock);
97 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
98 struct btrfs_root *root, struct inode *inode);
100 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
101 struct inode *inode, struct inode *dir,
102 const struct qstr *qstr)
106 err = btrfs_init_acl(trans, inode, dir);
108 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
113 * this does all the hard work for inserting an inline extent into
114 * the btree. The caller should have done a btrfs_drop_extents so that
115 * no overlapping inline items exist in the btree
117 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
118 struct btrfs_root *root, struct inode *inode,
119 u64 start, size_t size, size_t compressed_size,
121 struct page **compressed_pages)
123 struct btrfs_key key;
124 struct btrfs_path *path;
125 struct extent_buffer *leaf;
126 struct page *page = NULL;
129 struct btrfs_file_extent_item *ei;
132 size_t cur_size = size;
134 unsigned long offset;
136 if (compressed_size && compressed_pages)
137 cur_size = compressed_size;
139 path = btrfs_alloc_path();
143 path->leave_spinning = 1;
145 key.objectid = btrfs_ino(inode);
147 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
148 datasize = btrfs_file_extent_calc_inline_size(cur_size);
150 inode_add_bytes(inode, size);
151 ret = btrfs_insert_empty_item(trans, root, path, &key,
158 leaf = path->nodes[0];
159 ei = btrfs_item_ptr(leaf, path->slots[0],
160 struct btrfs_file_extent_item);
161 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
162 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
163 btrfs_set_file_extent_encryption(leaf, ei, 0);
164 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
165 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
166 ptr = btrfs_file_extent_inline_start(ei);
168 if (compress_type != BTRFS_COMPRESS_NONE) {
171 while (compressed_size > 0) {
172 cpage = compressed_pages[i];
173 cur_size = min_t(unsigned long, compressed_size,
176 kaddr = kmap_atomic(cpage, KM_USER0);
177 write_extent_buffer(leaf, kaddr, ptr, cur_size);
178 kunmap_atomic(kaddr, KM_USER0);
182 compressed_size -= cur_size;
184 btrfs_set_file_extent_compression(leaf, ei,
187 page = find_get_page(inode->i_mapping,
188 start >> PAGE_CACHE_SHIFT);
189 btrfs_set_file_extent_compression(leaf, ei, 0);
190 kaddr = kmap_atomic(page, KM_USER0);
191 offset = start & (PAGE_CACHE_SIZE - 1);
192 write_extent_buffer(leaf, kaddr + offset, ptr, size);
193 kunmap_atomic(kaddr, KM_USER0);
194 page_cache_release(page);
196 btrfs_mark_buffer_dirty(leaf);
197 btrfs_free_path(path);
200 * we're an inline extent, so nobody can
201 * extend the file past i_size without locking
202 * a page we already have locked.
204 * We must do any isize and inode updates
205 * before we unlock the pages. Otherwise we
206 * could end up racing with unlink.
208 BTRFS_I(inode)->disk_i_size = inode->i_size;
209 btrfs_update_inode(trans, root, inode);
213 btrfs_free_path(path);
219 * conditionally insert an inline extent into the file. This
220 * does the checks required to make sure the data is small enough
221 * to fit as an inline extent.
223 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
224 struct btrfs_root *root,
225 struct inode *inode, u64 start, u64 end,
226 size_t compressed_size, int compress_type,
227 struct page **compressed_pages)
229 u64 isize = i_size_read(inode);
230 u64 actual_end = min(end + 1, isize);
231 u64 inline_len = actual_end - start;
232 u64 aligned_end = (end + root->sectorsize - 1) &
233 ~((u64)root->sectorsize - 1);
235 u64 data_len = inline_len;
239 data_len = compressed_size;
242 actual_end >= PAGE_CACHE_SIZE ||
243 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
245 (actual_end & (root->sectorsize - 1)) == 0) ||
247 data_len > root->fs_info->max_inline) {
251 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
255 if (isize > actual_end)
256 inline_len = min_t(u64, isize, actual_end);
257 ret = insert_inline_extent(trans, root, inode, start,
258 inline_len, compressed_size,
259 compress_type, compressed_pages);
261 btrfs_delalloc_release_metadata(inode, end + 1 - start);
262 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
266 struct async_extent {
271 unsigned long nr_pages;
273 struct list_head list;
278 struct btrfs_root *root;
279 struct page *locked_page;
282 struct list_head extents;
283 struct btrfs_work work;
286 static noinline int add_async_extent(struct async_cow *cow,
287 u64 start, u64 ram_size,
290 unsigned long nr_pages,
293 struct async_extent *async_extent;
295 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
296 BUG_ON(!async_extent);
297 async_extent->start = start;
298 async_extent->ram_size = ram_size;
299 async_extent->compressed_size = compressed_size;
300 async_extent->pages = pages;
301 async_extent->nr_pages = nr_pages;
302 async_extent->compress_type = compress_type;
303 list_add_tail(&async_extent->list, &cow->extents);
308 * we create compressed extents in two phases. The first
309 * phase compresses a range of pages that have already been
310 * locked (both pages and state bits are locked).
312 * This is done inside an ordered work queue, and the compression
313 * is spread across many cpus. The actual IO submission is step
314 * two, and the ordered work queue takes care of making sure that
315 * happens in the same order things were put onto the queue by
316 * writepages and friends.
318 * If this code finds it can't get good compression, it puts an
319 * entry onto the work queue to write the uncompressed bytes. This
320 * makes sure that both compressed inodes and uncompressed inodes
321 * are written in the same order that pdflush sent them down.
323 static noinline int compress_file_range(struct inode *inode,
324 struct page *locked_page,
326 struct async_cow *async_cow,
329 struct btrfs_root *root = BTRFS_I(inode)->root;
330 struct btrfs_trans_handle *trans;
332 u64 blocksize = root->sectorsize;
334 u64 isize = i_size_read(inode);
336 struct page **pages = NULL;
337 unsigned long nr_pages;
338 unsigned long nr_pages_ret = 0;
339 unsigned long total_compressed = 0;
340 unsigned long total_in = 0;
341 unsigned long max_compressed = 128 * 1024;
342 unsigned long max_uncompressed = 128 * 1024;
345 int compress_type = root->fs_info->compress_type;
347 /* if this is a small write inside eof, kick off a defragbot */
348 if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024)
349 btrfs_add_inode_defrag(NULL, inode);
351 actual_end = min_t(u64, isize, end + 1);
354 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
355 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
358 * we don't want to send crud past the end of i_size through
359 * compression, that's just a waste of CPU time. So, if the
360 * end of the file is before the start of our current
361 * requested range of bytes, we bail out to the uncompressed
362 * cleanup code that can deal with all of this.
364 * It isn't really the fastest way to fix things, but this is a
365 * very uncommon corner.
367 if (actual_end <= start)
368 goto cleanup_and_bail_uncompressed;
370 total_compressed = actual_end - start;
372 /* we want to make sure that amount of ram required to uncompress
373 * an extent is reasonable, so we limit the total size in ram
374 * of a compressed extent to 128k. This is a crucial number
375 * because it also controls how easily we can spread reads across
376 * cpus for decompression.
378 * We also want to make sure the amount of IO required to do
379 * a random read is reasonably small, so we limit the size of
380 * a compressed extent to 128k.
382 total_compressed = min(total_compressed, max_uncompressed);
383 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
384 num_bytes = max(blocksize, num_bytes);
389 * we do compression for mount -o compress and when the
390 * inode has not been flagged as nocompress. This flag can
391 * change at any time if we discover bad compression ratios.
393 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
394 (btrfs_test_opt(root, COMPRESS) ||
395 (BTRFS_I(inode)->force_compress) ||
396 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
398 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
400 /* just bail out to the uncompressed code */
404 if (BTRFS_I(inode)->force_compress)
405 compress_type = BTRFS_I(inode)->force_compress;
407 ret = btrfs_compress_pages(compress_type,
408 inode->i_mapping, start,
409 total_compressed, pages,
410 nr_pages, &nr_pages_ret,
416 unsigned long offset = total_compressed &
417 (PAGE_CACHE_SIZE - 1);
418 struct page *page = pages[nr_pages_ret - 1];
421 /* zero the tail end of the last page, we might be
422 * sending it down to disk
425 kaddr = kmap_atomic(page, KM_USER0);
426 memset(kaddr + offset, 0,
427 PAGE_CACHE_SIZE - offset);
428 kunmap_atomic(kaddr, KM_USER0);
435 trans = btrfs_join_transaction(root);
436 BUG_ON(IS_ERR(trans));
437 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
439 /* lets try to make an inline extent */
440 if (ret || total_in < (actual_end - start)) {
441 /* we didn't compress the entire range, try
442 * to make an uncompressed inline extent.
444 ret = cow_file_range_inline(trans, root, inode,
445 start, end, 0, 0, NULL);
447 /* try making a compressed inline extent */
448 ret = cow_file_range_inline(trans, root, inode,
451 compress_type, pages);
455 * inline extent creation worked, we don't need
456 * to create any more async work items. Unlock
457 * and free up our temp pages.
459 extent_clear_unlock_delalloc(inode,
460 &BTRFS_I(inode)->io_tree,
462 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
463 EXTENT_CLEAR_DELALLOC |
464 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
466 btrfs_end_transaction(trans, root);
469 btrfs_end_transaction(trans, root);
474 * we aren't doing an inline extent round the compressed size
475 * up to a block size boundary so the allocator does sane
478 total_compressed = (total_compressed + blocksize - 1) &
482 * one last check to make sure the compression is really a
483 * win, compare the page count read with the blocks on disk
485 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
486 ~(PAGE_CACHE_SIZE - 1);
487 if (total_compressed >= total_in) {
490 num_bytes = total_in;
493 if (!will_compress && pages) {
495 * the compression code ran but failed to make things smaller,
496 * free any pages it allocated and our page pointer array
498 for (i = 0; i < nr_pages_ret; i++) {
499 WARN_ON(pages[i]->mapping);
500 page_cache_release(pages[i]);
504 total_compressed = 0;
507 /* flag the file so we don't compress in the future */
508 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
509 !(BTRFS_I(inode)->force_compress)) {
510 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
516 /* the async work queues will take care of doing actual
517 * allocation on disk for these compressed pages,
518 * and will submit them to the elevator.
520 add_async_extent(async_cow, start, num_bytes,
521 total_compressed, pages, nr_pages_ret,
524 if (start + num_bytes < end) {
531 cleanup_and_bail_uncompressed:
533 * No compression, but we still need to write the pages in
534 * the file we've been given so far. redirty the locked
535 * page if it corresponds to our extent and set things up
536 * for the async work queue to run cow_file_range to do
537 * the normal delalloc dance
539 if (page_offset(locked_page) >= start &&
540 page_offset(locked_page) <= end) {
541 __set_page_dirty_nobuffers(locked_page);
542 /* unlocked later on in the async handlers */
544 add_async_extent(async_cow, start, end - start + 1,
545 0, NULL, 0, BTRFS_COMPRESS_NONE);
553 for (i = 0; i < nr_pages_ret; i++) {
554 WARN_ON(pages[i]->mapping);
555 page_cache_release(pages[i]);
563 * phase two of compressed writeback. This is the ordered portion
564 * of the code, which only gets called in the order the work was
565 * queued. We walk all the async extents created by compress_file_range
566 * and send them down to the disk.
568 static noinline int submit_compressed_extents(struct inode *inode,
569 struct async_cow *async_cow)
571 struct async_extent *async_extent;
573 struct btrfs_trans_handle *trans;
574 struct btrfs_key ins;
575 struct extent_map *em;
576 struct btrfs_root *root = BTRFS_I(inode)->root;
577 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
578 struct extent_io_tree *io_tree;
581 if (list_empty(&async_cow->extents))
585 while (!list_empty(&async_cow->extents)) {
586 async_extent = list_entry(async_cow->extents.next,
587 struct async_extent, list);
588 list_del(&async_extent->list);
590 io_tree = &BTRFS_I(inode)->io_tree;
593 /* did the compression code fall back to uncompressed IO? */
594 if (!async_extent->pages) {
595 int page_started = 0;
596 unsigned long nr_written = 0;
598 lock_extent(io_tree, async_extent->start,
599 async_extent->start +
600 async_extent->ram_size - 1, GFP_NOFS);
602 /* allocate blocks */
603 ret = cow_file_range(inode, async_cow->locked_page,
605 async_extent->start +
606 async_extent->ram_size - 1,
607 &page_started, &nr_written, 0);
610 * if page_started, cow_file_range inserted an
611 * inline extent and took care of all the unlocking
612 * and IO for us. Otherwise, we need to submit
613 * all those pages down to the drive.
615 if (!page_started && !ret)
616 extent_write_locked_range(io_tree,
617 inode, async_extent->start,
618 async_extent->start +
619 async_extent->ram_size - 1,
627 lock_extent(io_tree, async_extent->start,
628 async_extent->start + async_extent->ram_size - 1,
631 trans = btrfs_join_transaction(root);
632 BUG_ON(IS_ERR(trans));
633 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
634 ret = btrfs_reserve_extent(trans, root,
635 async_extent->compressed_size,
636 async_extent->compressed_size,
637 0, alloc_hint, &ins, 1);
638 btrfs_end_transaction(trans, root);
642 for (i = 0; i < async_extent->nr_pages; i++) {
643 WARN_ON(async_extent->pages[i]->mapping);
644 page_cache_release(async_extent->pages[i]);
646 kfree(async_extent->pages);
647 async_extent->nr_pages = 0;
648 async_extent->pages = NULL;
649 unlock_extent(io_tree, async_extent->start,
650 async_extent->start +
651 async_extent->ram_size - 1, GFP_NOFS);
656 * here we're doing allocation and writeback of the
659 btrfs_drop_extent_cache(inode, async_extent->start,
660 async_extent->start +
661 async_extent->ram_size - 1, 0);
663 em = alloc_extent_map();
665 em->start = async_extent->start;
666 em->len = async_extent->ram_size;
667 em->orig_start = em->start;
669 em->block_start = ins.objectid;
670 em->block_len = ins.offset;
671 em->bdev = root->fs_info->fs_devices->latest_bdev;
672 em->compress_type = async_extent->compress_type;
673 set_bit(EXTENT_FLAG_PINNED, &em->flags);
674 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
677 write_lock(&em_tree->lock);
678 ret = add_extent_mapping(em_tree, em);
679 write_unlock(&em_tree->lock);
680 if (ret != -EEXIST) {
684 btrfs_drop_extent_cache(inode, async_extent->start,
685 async_extent->start +
686 async_extent->ram_size - 1, 0);
689 ret = btrfs_add_ordered_extent_compress(inode,
692 async_extent->ram_size,
694 BTRFS_ORDERED_COMPRESSED,
695 async_extent->compress_type);
699 * clear dirty, set writeback and unlock the pages.
701 extent_clear_unlock_delalloc(inode,
702 &BTRFS_I(inode)->io_tree,
704 async_extent->start +
705 async_extent->ram_size - 1,
706 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
707 EXTENT_CLEAR_UNLOCK |
708 EXTENT_CLEAR_DELALLOC |
709 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
711 ret = btrfs_submit_compressed_write(inode,
713 async_extent->ram_size,
715 ins.offset, async_extent->pages,
716 async_extent->nr_pages);
719 alloc_hint = ins.objectid + ins.offset;
727 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
730 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
731 struct extent_map *em;
734 read_lock(&em_tree->lock);
735 em = search_extent_mapping(em_tree, start, num_bytes);
738 * if block start isn't an actual block number then find the
739 * first block in this inode and use that as a hint. If that
740 * block is also bogus then just don't worry about it.
742 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
744 em = search_extent_mapping(em_tree, 0, 0);
745 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
746 alloc_hint = em->block_start;
750 alloc_hint = em->block_start;
754 read_unlock(&em_tree->lock);
760 * when extent_io.c finds a delayed allocation range in the file,
761 * the call backs end up in this code. The basic idea is to
762 * allocate extents on disk for the range, and create ordered data structs
763 * in ram to track those extents.
765 * locked_page is the page that writepage had locked already. We use
766 * it to make sure we don't do extra locks or unlocks.
768 * *page_started is set to one if we unlock locked_page and do everything
769 * required to start IO on it. It may be clean and already done with
772 static noinline int cow_file_range(struct inode *inode,
773 struct page *locked_page,
774 u64 start, u64 end, int *page_started,
775 unsigned long *nr_written,
778 struct btrfs_root *root = BTRFS_I(inode)->root;
779 struct btrfs_trans_handle *trans;
782 unsigned long ram_size;
785 u64 blocksize = root->sectorsize;
786 struct btrfs_key ins;
787 struct extent_map *em;
788 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
791 BUG_ON(btrfs_is_free_space_inode(root, inode));
792 trans = btrfs_join_transaction(root);
793 BUG_ON(IS_ERR(trans));
794 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
796 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
797 num_bytes = max(blocksize, num_bytes);
798 disk_num_bytes = num_bytes;
801 /* if this is a small write inside eof, kick off defrag */
802 if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
803 btrfs_add_inode_defrag(trans, inode);
806 /* lets try to make an inline extent */
807 ret = cow_file_range_inline(trans, root, inode,
808 start, end, 0, 0, NULL);
810 extent_clear_unlock_delalloc(inode,
811 &BTRFS_I(inode)->io_tree,
813 EXTENT_CLEAR_UNLOCK_PAGE |
814 EXTENT_CLEAR_UNLOCK |
815 EXTENT_CLEAR_DELALLOC |
817 EXTENT_SET_WRITEBACK |
818 EXTENT_END_WRITEBACK);
820 *nr_written = *nr_written +
821 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
828 BUG_ON(disk_num_bytes >
829 btrfs_super_total_bytes(root->fs_info->super_copy));
831 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
832 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
834 while (disk_num_bytes > 0) {
837 cur_alloc_size = disk_num_bytes;
838 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
839 root->sectorsize, 0, alloc_hint,
843 em = alloc_extent_map();
846 em->orig_start = em->start;
847 ram_size = ins.offset;
848 em->len = ins.offset;
850 em->block_start = ins.objectid;
851 em->block_len = ins.offset;
852 em->bdev = root->fs_info->fs_devices->latest_bdev;
853 set_bit(EXTENT_FLAG_PINNED, &em->flags);
856 write_lock(&em_tree->lock);
857 ret = add_extent_mapping(em_tree, em);
858 write_unlock(&em_tree->lock);
859 if (ret != -EEXIST) {
863 btrfs_drop_extent_cache(inode, start,
864 start + ram_size - 1, 0);
867 cur_alloc_size = ins.offset;
868 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
869 ram_size, cur_alloc_size, 0);
872 if (root->root_key.objectid ==
873 BTRFS_DATA_RELOC_TREE_OBJECTID) {
874 ret = btrfs_reloc_clone_csums(inode, start,
879 if (disk_num_bytes < cur_alloc_size)
882 /* we're not doing compressed IO, don't unlock the first
883 * page (which the caller expects to stay locked), don't
884 * clear any dirty bits and don't set any writeback bits
886 * Do set the Private2 bit so we know this page was properly
887 * setup for writepage
889 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
890 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
893 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
894 start, start + ram_size - 1,
896 disk_num_bytes -= cur_alloc_size;
897 num_bytes -= cur_alloc_size;
898 alloc_hint = ins.objectid + ins.offset;
899 start += cur_alloc_size;
903 btrfs_end_transaction(trans, root);
909 * work queue call back to started compression on a file and pages
911 static noinline void async_cow_start(struct btrfs_work *work)
913 struct async_cow *async_cow;
915 async_cow = container_of(work, struct async_cow, work);
917 compress_file_range(async_cow->inode, async_cow->locked_page,
918 async_cow->start, async_cow->end, async_cow,
921 async_cow->inode = NULL;
925 * work queue call back to submit previously compressed pages
927 static noinline void async_cow_submit(struct btrfs_work *work)
929 struct async_cow *async_cow;
930 struct btrfs_root *root;
931 unsigned long nr_pages;
933 async_cow = container_of(work, struct async_cow, work);
935 root = async_cow->root;
936 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
939 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
941 if (atomic_read(&root->fs_info->async_delalloc_pages) <
943 waitqueue_active(&root->fs_info->async_submit_wait))
944 wake_up(&root->fs_info->async_submit_wait);
946 if (async_cow->inode)
947 submit_compressed_extents(async_cow->inode, async_cow);
950 static noinline void async_cow_free(struct btrfs_work *work)
952 struct async_cow *async_cow;
953 async_cow = container_of(work, struct async_cow, work);
957 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
958 u64 start, u64 end, int *page_started,
959 unsigned long *nr_written)
961 struct async_cow *async_cow;
962 struct btrfs_root *root = BTRFS_I(inode)->root;
963 unsigned long nr_pages;
965 int limit = 10 * 1024 * 1042;
967 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
968 1, 0, NULL, GFP_NOFS);
969 while (start < end) {
970 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
972 async_cow->inode = inode;
973 async_cow->root = root;
974 async_cow->locked_page = locked_page;
975 async_cow->start = start;
977 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
980 cur_end = min(end, start + 512 * 1024 - 1);
982 async_cow->end = cur_end;
983 INIT_LIST_HEAD(&async_cow->extents);
985 async_cow->work.func = async_cow_start;
986 async_cow->work.ordered_func = async_cow_submit;
987 async_cow->work.ordered_free = async_cow_free;
988 async_cow->work.flags = 0;
990 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
992 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
994 btrfs_queue_worker(&root->fs_info->delalloc_workers,
997 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
998 wait_event(root->fs_info->async_submit_wait,
999 (atomic_read(&root->fs_info->async_delalloc_pages) <
1003 while (atomic_read(&root->fs_info->async_submit_draining) &&
1004 atomic_read(&root->fs_info->async_delalloc_pages)) {
1005 wait_event(root->fs_info->async_submit_wait,
1006 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1010 *nr_written += nr_pages;
1011 start = cur_end + 1;
1017 static noinline int csum_exist_in_range(struct btrfs_root *root,
1018 u64 bytenr, u64 num_bytes)
1021 struct btrfs_ordered_sum *sums;
1024 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1025 bytenr + num_bytes - 1, &list, 0);
1026 if (ret == 0 && list_empty(&list))
1029 while (!list_empty(&list)) {
1030 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1031 list_del(&sums->list);
1038 * when nowcow writeback call back. This checks for snapshots or COW copies
1039 * of the extents that exist in the file, and COWs the file as required.
1041 * If no cow copies or snapshots exist, we write directly to the existing
1044 static noinline int run_delalloc_nocow(struct inode *inode,
1045 struct page *locked_page,
1046 u64 start, u64 end, int *page_started, int force,
1047 unsigned long *nr_written)
1049 struct btrfs_root *root = BTRFS_I(inode)->root;
1050 struct btrfs_trans_handle *trans;
1051 struct extent_buffer *leaf;
1052 struct btrfs_path *path;
1053 struct btrfs_file_extent_item *fi;
1054 struct btrfs_key found_key;
1067 u64 ino = btrfs_ino(inode);
1069 path = btrfs_alloc_path();
1073 nolock = btrfs_is_free_space_inode(root, inode);
1076 trans = btrfs_join_transaction_nolock(root);
1078 trans = btrfs_join_transaction(root);
1080 BUG_ON(IS_ERR(trans));
1081 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1083 cow_start = (u64)-1;
1086 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1089 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1090 leaf = path->nodes[0];
1091 btrfs_item_key_to_cpu(leaf, &found_key,
1092 path->slots[0] - 1);
1093 if (found_key.objectid == ino &&
1094 found_key.type == BTRFS_EXTENT_DATA_KEY)
1099 leaf = path->nodes[0];
1100 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1101 ret = btrfs_next_leaf(root, path);
1106 leaf = path->nodes[0];
1112 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1114 if (found_key.objectid > ino ||
1115 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1116 found_key.offset > end)
1119 if (found_key.offset > cur_offset) {
1120 extent_end = found_key.offset;
1125 fi = btrfs_item_ptr(leaf, path->slots[0],
1126 struct btrfs_file_extent_item);
1127 extent_type = btrfs_file_extent_type(leaf, fi);
1129 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1130 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1131 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1132 extent_offset = btrfs_file_extent_offset(leaf, fi);
1133 extent_end = found_key.offset +
1134 btrfs_file_extent_num_bytes(leaf, fi);
1135 if (extent_end <= start) {
1139 if (disk_bytenr == 0)
1141 if (btrfs_file_extent_compression(leaf, fi) ||
1142 btrfs_file_extent_encryption(leaf, fi) ||
1143 btrfs_file_extent_other_encoding(leaf, fi))
1145 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1147 if (btrfs_extent_readonly(root, disk_bytenr))
1149 if (btrfs_cross_ref_exist(trans, root, ino,
1151 extent_offset, disk_bytenr))
1153 disk_bytenr += extent_offset;
1154 disk_bytenr += cur_offset - found_key.offset;
1155 num_bytes = min(end + 1, extent_end) - cur_offset;
1157 * force cow if csum exists in the range.
1158 * this ensure that csum for a given extent are
1159 * either valid or do not exist.
1161 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1164 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1165 extent_end = found_key.offset +
1166 btrfs_file_extent_inline_len(leaf, fi);
1167 extent_end = ALIGN(extent_end, root->sectorsize);
1172 if (extent_end <= start) {
1177 if (cow_start == (u64)-1)
1178 cow_start = cur_offset;
1179 cur_offset = extent_end;
1180 if (cur_offset > end)
1186 btrfs_release_path(path);
1187 if (cow_start != (u64)-1) {
1188 ret = cow_file_range(inode, locked_page, cow_start,
1189 found_key.offset - 1, page_started,
1192 cow_start = (u64)-1;
1195 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1196 struct extent_map *em;
1197 struct extent_map_tree *em_tree;
1198 em_tree = &BTRFS_I(inode)->extent_tree;
1199 em = alloc_extent_map();
1201 em->start = cur_offset;
1202 em->orig_start = em->start;
1203 em->len = num_bytes;
1204 em->block_len = num_bytes;
1205 em->block_start = disk_bytenr;
1206 em->bdev = root->fs_info->fs_devices->latest_bdev;
1207 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1209 write_lock(&em_tree->lock);
1210 ret = add_extent_mapping(em_tree, em);
1211 write_unlock(&em_tree->lock);
1212 if (ret != -EEXIST) {
1213 free_extent_map(em);
1216 btrfs_drop_extent_cache(inode, em->start,
1217 em->start + em->len - 1, 0);
1219 type = BTRFS_ORDERED_PREALLOC;
1221 type = BTRFS_ORDERED_NOCOW;
1224 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1225 num_bytes, num_bytes, type);
1228 if (root->root_key.objectid ==
1229 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1230 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1235 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1236 cur_offset, cur_offset + num_bytes - 1,
1237 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1238 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1239 EXTENT_SET_PRIVATE2);
1240 cur_offset = extent_end;
1241 if (cur_offset > end)
1244 btrfs_release_path(path);
1246 if (cur_offset <= end && cow_start == (u64)-1)
1247 cow_start = cur_offset;
1248 if (cow_start != (u64)-1) {
1249 ret = cow_file_range(inode, locked_page, cow_start, end,
1250 page_started, nr_written, 1);
1255 ret = btrfs_end_transaction_nolock(trans, root);
1258 ret = btrfs_end_transaction(trans, root);
1261 btrfs_free_path(path);
1266 * extent_io.c call back to do delayed allocation processing
1268 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1269 u64 start, u64 end, int *page_started,
1270 unsigned long *nr_written)
1273 struct btrfs_root *root = BTRFS_I(inode)->root;
1275 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1276 ret = run_delalloc_nocow(inode, locked_page, start, end,
1277 page_started, 1, nr_written);
1278 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1279 ret = run_delalloc_nocow(inode, locked_page, start, end,
1280 page_started, 0, nr_written);
1281 else if (!btrfs_test_opt(root, COMPRESS) &&
1282 !(BTRFS_I(inode)->force_compress) &&
1283 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1284 ret = cow_file_range(inode, locked_page, start, end,
1285 page_started, nr_written, 1);
1287 ret = cow_file_range_async(inode, locked_page, start, end,
1288 page_started, nr_written);
1292 static void btrfs_split_extent_hook(struct inode *inode,
1293 struct extent_state *orig, u64 split)
1295 /* not delalloc, ignore it */
1296 if (!(orig->state & EXTENT_DELALLOC))
1299 spin_lock(&BTRFS_I(inode)->lock);
1300 BTRFS_I(inode)->outstanding_extents++;
1301 spin_unlock(&BTRFS_I(inode)->lock);
1305 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1306 * extents so we can keep track of new extents that are just merged onto old
1307 * extents, such as when we are doing sequential writes, so we can properly
1308 * account for the metadata space we'll need.
1310 static void btrfs_merge_extent_hook(struct inode *inode,
1311 struct extent_state *new,
1312 struct extent_state *other)
1314 /* not delalloc, ignore it */
1315 if (!(other->state & EXTENT_DELALLOC))
1318 spin_lock(&BTRFS_I(inode)->lock);
1319 BTRFS_I(inode)->outstanding_extents--;
1320 spin_unlock(&BTRFS_I(inode)->lock);
1324 * extent_io.c set_bit_hook, used to track delayed allocation
1325 * bytes in this file, and to maintain the list of inodes that
1326 * have pending delalloc work to be done.
1328 static void btrfs_set_bit_hook(struct inode *inode,
1329 struct extent_state *state, int *bits)
1333 * set_bit and clear bit hooks normally require _irqsave/restore
1334 * but in this case, we are only testing for the DELALLOC
1335 * bit, which is only set or cleared with irqs on
1337 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1338 struct btrfs_root *root = BTRFS_I(inode)->root;
1339 u64 len = state->end + 1 - state->start;
1340 bool do_list = !btrfs_is_free_space_inode(root, inode);
1342 if (*bits & EXTENT_FIRST_DELALLOC) {
1343 *bits &= ~EXTENT_FIRST_DELALLOC;
1345 spin_lock(&BTRFS_I(inode)->lock);
1346 BTRFS_I(inode)->outstanding_extents++;
1347 spin_unlock(&BTRFS_I(inode)->lock);
1350 spin_lock(&root->fs_info->delalloc_lock);
1351 BTRFS_I(inode)->delalloc_bytes += len;
1352 root->fs_info->delalloc_bytes += len;
1353 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1354 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1355 &root->fs_info->delalloc_inodes);
1357 spin_unlock(&root->fs_info->delalloc_lock);
1362 * extent_io.c clear_bit_hook, see set_bit_hook for why
1364 static void btrfs_clear_bit_hook(struct inode *inode,
1365 struct extent_state *state, int *bits)
1368 * set_bit and clear bit hooks normally require _irqsave/restore
1369 * but in this case, we are only testing for the DELALLOC
1370 * bit, which is only set or cleared with irqs on
1372 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1373 struct btrfs_root *root = BTRFS_I(inode)->root;
1374 u64 len = state->end + 1 - state->start;
1375 bool do_list = !btrfs_is_free_space_inode(root, inode);
1377 if (*bits & EXTENT_FIRST_DELALLOC) {
1378 *bits &= ~EXTENT_FIRST_DELALLOC;
1379 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1380 spin_lock(&BTRFS_I(inode)->lock);
1381 BTRFS_I(inode)->outstanding_extents--;
1382 spin_unlock(&BTRFS_I(inode)->lock);
1385 if (*bits & EXTENT_DO_ACCOUNTING)
1386 btrfs_delalloc_release_metadata(inode, len);
1388 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1390 btrfs_free_reserved_data_space(inode, len);
1392 spin_lock(&root->fs_info->delalloc_lock);
1393 root->fs_info->delalloc_bytes -= len;
1394 BTRFS_I(inode)->delalloc_bytes -= len;
1396 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1397 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1398 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1400 spin_unlock(&root->fs_info->delalloc_lock);
1405 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1406 * we don't create bios that span stripes or chunks
1408 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1409 size_t size, struct bio *bio,
1410 unsigned long bio_flags)
1412 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1413 struct btrfs_mapping_tree *map_tree;
1414 u64 logical = (u64)bio->bi_sector << 9;
1419 if (bio_flags & EXTENT_BIO_COMPRESSED)
1422 length = bio->bi_size;
1423 map_tree = &root->fs_info->mapping_tree;
1424 map_length = length;
1425 ret = btrfs_map_block(map_tree, READ, logical,
1426 &map_length, NULL, 0);
1428 if (map_length < length + size)
1434 * in order to insert checksums into the metadata in large chunks,
1435 * we wait until bio submission time. All the pages in the bio are
1436 * checksummed and sums are attached onto the ordered extent record.
1438 * At IO completion time the cums attached on the ordered extent record
1439 * are inserted into the btree
1441 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1442 struct bio *bio, int mirror_num,
1443 unsigned long bio_flags,
1446 struct btrfs_root *root = BTRFS_I(inode)->root;
1449 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1455 * in order to insert checksums into the metadata in large chunks,
1456 * we wait until bio submission time. All the pages in the bio are
1457 * checksummed and sums are attached onto the ordered extent record.
1459 * At IO completion time the cums attached on the ordered extent record
1460 * are inserted into the btree
1462 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1463 int mirror_num, unsigned long bio_flags,
1466 struct btrfs_root *root = BTRFS_I(inode)->root;
1467 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1471 * extent_io.c submission hook. This does the right thing for csum calculation
1472 * on write, or reading the csums from the tree before a read
1474 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1475 int mirror_num, unsigned long bio_flags,
1478 struct btrfs_root *root = BTRFS_I(inode)->root;
1482 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1484 if (btrfs_is_free_space_inode(root, inode))
1485 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1487 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1490 if (!(rw & REQ_WRITE)) {
1491 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1492 return btrfs_submit_compressed_read(inode, bio,
1493 mirror_num, bio_flags);
1494 } else if (!skip_sum) {
1495 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1500 } else if (!skip_sum) {
1501 /* csum items have already been cloned */
1502 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1504 /* we're doing a write, do the async checksumming */
1505 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1506 inode, rw, bio, mirror_num,
1507 bio_flags, bio_offset,
1508 __btrfs_submit_bio_start,
1509 __btrfs_submit_bio_done);
1513 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1517 * given a list of ordered sums record them in the inode. This happens
1518 * at IO completion time based on sums calculated at bio submission time.
1520 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1521 struct inode *inode, u64 file_offset,
1522 struct list_head *list)
1524 struct btrfs_ordered_sum *sum;
1526 list_for_each_entry(sum, list, list) {
1527 btrfs_csum_file_blocks(trans,
1528 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1533 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1534 struct extent_state **cached_state)
1536 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1538 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1539 cached_state, GFP_NOFS);
1542 /* see btrfs_writepage_start_hook for details on why this is required */
1543 struct btrfs_writepage_fixup {
1545 struct btrfs_work work;
1548 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1550 struct btrfs_writepage_fixup *fixup;
1551 struct btrfs_ordered_extent *ordered;
1552 struct extent_state *cached_state = NULL;
1554 struct inode *inode;
1559 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1563 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1564 ClearPageChecked(page);
1568 inode = page->mapping->host;
1569 page_start = page_offset(page);
1570 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1572 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1573 &cached_state, GFP_NOFS);
1575 /* already ordered? We're done */
1576 if (PagePrivate2(page))
1579 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1581 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1582 page_end, &cached_state, GFP_NOFS);
1584 btrfs_start_ordered_extent(inode, ordered, 1);
1585 btrfs_put_ordered_extent(ordered);
1589 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1591 mapping_set_error(page->mapping, ret);
1592 end_extent_writepage(page, ret, page_start, page_end);
1593 ClearPageChecked(page);
1597 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1598 ClearPageChecked(page);
1599 set_page_dirty(page);
1601 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1602 &cached_state, GFP_NOFS);
1605 page_cache_release(page);
1610 * There are a few paths in the higher layers of the kernel that directly
1611 * set the page dirty bit without asking the filesystem if it is a
1612 * good idea. This causes problems because we want to make sure COW
1613 * properly happens and the data=ordered rules are followed.
1615 * In our case any range that doesn't have the ORDERED bit set
1616 * hasn't been properly setup for IO. We kick off an async process
1617 * to fix it up. The async helper will wait for ordered extents, set
1618 * the delalloc bit and make it safe to write the page.
1620 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1622 struct inode *inode = page->mapping->host;
1623 struct btrfs_writepage_fixup *fixup;
1624 struct btrfs_root *root = BTRFS_I(inode)->root;
1626 /* this page is properly in the ordered list */
1627 if (TestClearPagePrivate2(page))
1630 if (PageChecked(page))
1633 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1637 SetPageChecked(page);
1638 page_cache_get(page);
1639 fixup->work.func = btrfs_writepage_fixup_worker;
1641 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1645 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1646 struct inode *inode, u64 file_pos,
1647 u64 disk_bytenr, u64 disk_num_bytes,
1648 u64 num_bytes, u64 ram_bytes,
1649 u8 compression, u8 encryption,
1650 u16 other_encoding, int extent_type)
1652 struct btrfs_root *root = BTRFS_I(inode)->root;
1653 struct btrfs_file_extent_item *fi;
1654 struct btrfs_path *path;
1655 struct extent_buffer *leaf;
1656 struct btrfs_key ins;
1660 path = btrfs_alloc_path();
1664 path->leave_spinning = 1;
1667 * we may be replacing one extent in the tree with another.
1668 * The new extent is pinned in the extent map, and we don't want
1669 * to drop it from the cache until it is completely in the btree.
1671 * So, tell btrfs_drop_extents to leave this extent in the cache.
1672 * the caller is expected to unpin it and allow it to be merged
1675 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1679 ins.objectid = btrfs_ino(inode);
1680 ins.offset = file_pos;
1681 ins.type = BTRFS_EXTENT_DATA_KEY;
1682 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1684 leaf = path->nodes[0];
1685 fi = btrfs_item_ptr(leaf, path->slots[0],
1686 struct btrfs_file_extent_item);
1687 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1688 btrfs_set_file_extent_type(leaf, fi, extent_type);
1689 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1690 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1691 btrfs_set_file_extent_offset(leaf, fi, 0);
1692 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1693 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1694 btrfs_set_file_extent_compression(leaf, fi, compression);
1695 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1696 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1698 btrfs_unlock_up_safe(path, 1);
1699 btrfs_set_lock_blocking(leaf);
1701 btrfs_mark_buffer_dirty(leaf);
1703 inode_add_bytes(inode, num_bytes);
1705 ins.objectid = disk_bytenr;
1706 ins.offset = disk_num_bytes;
1707 ins.type = BTRFS_EXTENT_ITEM_KEY;
1708 ret = btrfs_alloc_reserved_file_extent(trans, root,
1709 root->root_key.objectid,
1710 btrfs_ino(inode), file_pos, &ins);
1712 btrfs_free_path(path);
1718 * helper function for btrfs_finish_ordered_io, this
1719 * just reads in some of the csum leaves to prime them into ram
1720 * before we start the transaction. It limits the amount of btree
1721 * reads required while inside the transaction.
1723 /* as ordered data IO finishes, this gets called so we can finish
1724 * an ordered extent if the range of bytes in the file it covers are
1727 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1729 struct btrfs_root *root = BTRFS_I(inode)->root;
1730 struct btrfs_trans_handle *trans = NULL;
1731 struct btrfs_ordered_extent *ordered_extent = NULL;
1732 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1733 struct extent_state *cached_state = NULL;
1734 int compress_type = 0;
1738 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1742 BUG_ON(!ordered_extent);
1744 nolock = btrfs_is_free_space_inode(root, inode);
1746 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1747 BUG_ON(!list_empty(&ordered_extent->list));
1748 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1751 trans = btrfs_join_transaction_nolock(root);
1753 trans = btrfs_join_transaction(root);
1754 BUG_ON(IS_ERR(trans));
1755 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1756 ret = btrfs_update_inode_fallback(trans, root, inode);
1762 lock_extent_bits(io_tree, ordered_extent->file_offset,
1763 ordered_extent->file_offset + ordered_extent->len - 1,
1764 0, &cached_state, GFP_NOFS);
1767 trans = btrfs_join_transaction_nolock(root);
1769 trans = btrfs_join_transaction(root);
1770 BUG_ON(IS_ERR(trans));
1771 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1773 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1774 compress_type = ordered_extent->compress_type;
1775 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1776 BUG_ON(compress_type);
1777 ret = btrfs_mark_extent_written(trans, inode,
1778 ordered_extent->file_offset,
1779 ordered_extent->file_offset +
1780 ordered_extent->len);
1783 BUG_ON(root == root->fs_info->tree_root);
1784 ret = insert_reserved_file_extent(trans, inode,
1785 ordered_extent->file_offset,
1786 ordered_extent->start,
1787 ordered_extent->disk_len,
1788 ordered_extent->len,
1789 ordered_extent->len,
1790 compress_type, 0, 0,
1791 BTRFS_FILE_EXTENT_REG);
1792 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1793 ordered_extent->file_offset,
1794 ordered_extent->len);
1797 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1798 ordered_extent->file_offset +
1799 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1801 add_pending_csums(trans, inode, ordered_extent->file_offset,
1802 &ordered_extent->list);
1804 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1805 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1806 ret = btrfs_update_inode_fallback(trans, root, inode);
1811 if (root != root->fs_info->tree_root)
1812 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1815 btrfs_end_transaction_nolock(trans, root);
1817 btrfs_end_transaction(trans, root);
1821 btrfs_put_ordered_extent(ordered_extent);
1822 /* once for the tree */
1823 btrfs_put_ordered_extent(ordered_extent);
1828 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1829 struct extent_state *state, int uptodate)
1831 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1833 ClearPagePrivate2(page);
1834 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1838 * when reads are done, we need to check csums to verify the data is correct
1839 * if there's a match, we allow the bio to finish. If not, the code in
1840 * extent_io.c will try to find good copies for us.
1842 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1843 struct extent_state *state)
1845 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1846 struct inode *inode = page->mapping->host;
1847 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1849 u64 private = ~(u32)0;
1851 struct btrfs_root *root = BTRFS_I(inode)->root;
1854 if (PageChecked(page)) {
1855 ClearPageChecked(page);
1859 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1862 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1863 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1864 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1869 if (state && state->start == start) {
1870 private = state->private;
1873 ret = get_state_private(io_tree, start, &private);
1875 kaddr = kmap_atomic(page, KM_USER0);
1879 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1880 btrfs_csum_final(csum, (char *)&csum);
1881 if (csum != private)
1884 kunmap_atomic(kaddr, KM_USER0);
1889 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
1891 (unsigned long long)btrfs_ino(page->mapping->host),
1892 (unsigned long long)start, csum,
1893 (unsigned long long)private);
1894 memset(kaddr + offset, 1, end - start + 1);
1895 flush_dcache_page(page);
1896 kunmap_atomic(kaddr, KM_USER0);
1902 struct delayed_iput {
1903 struct list_head list;
1904 struct inode *inode;
1907 void btrfs_add_delayed_iput(struct inode *inode)
1909 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1910 struct delayed_iput *delayed;
1912 if (atomic_add_unless(&inode->i_count, -1, 1))
1915 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
1916 delayed->inode = inode;
1918 spin_lock(&fs_info->delayed_iput_lock);
1919 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
1920 spin_unlock(&fs_info->delayed_iput_lock);
1923 void btrfs_run_delayed_iputs(struct btrfs_root *root)
1926 struct btrfs_fs_info *fs_info = root->fs_info;
1927 struct delayed_iput *delayed;
1930 spin_lock(&fs_info->delayed_iput_lock);
1931 empty = list_empty(&fs_info->delayed_iputs);
1932 spin_unlock(&fs_info->delayed_iput_lock);
1936 down_read(&root->fs_info->cleanup_work_sem);
1937 spin_lock(&fs_info->delayed_iput_lock);
1938 list_splice_init(&fs_info->delayed_iputs, &list);
1939 spin_unlock(&fs_info->delayed_iput_lock);
1941 while (!list_empty(&list)) {
1942 delayed = list_entry(list.next, struct delayed_iput, list);
1943 list_del(&delayed->list);
1944 iput(delayed->inode);
1947 up_read(&root->fs_info->cleanup_work_sem);
1950 enum btrfs_orphan_cleanup_state {
1951 ORPHAN_CLEANUP_STARTED = 1,
1952 ORPHAN_CLEANUP_DONE = 2,
1956 * This is called in transaction commit time. If there are no orphan
1957 * files in the subvolume, it removes orphan item and frees block_rsv
1960 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
1961 struct btrfs_root *root)
1963 struct btrfs_block_rsv *block_rsv;
1966 if (!list_empty(&root->orphan_list) ||
1967 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
1970 spin_lock(&root->orphan_lock);
1971 if (!list_empty(&root->orphan_list)) {
1972 spin_unlock(&root->orphan_lock);
1976 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
1977 spin_unlock(&root->orphan_lock);
1981 block_rsv = root->orphan_block_rsv;
1982 root->orphan_block_rsv = NULL;
1983 spin_unlock(&root->orphan_lock);
1985 if (root->orphan_item_inserted &&
1986 btrfs_root_refs(&root->root_item) > 0) {
1987 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
1988 root->root_key.objectid);
1990 root->orphan_item_inserted = 0;
1994 WARN_ON(block_rsv->size > 0);
1995 btrfs_free_block_rsv(root, block_rsv);
2000 * This creates an orphan entry for the given inode in case something goes
2001 * wrong in the middle of an unlink/truncate.
2003 * NOTE: caller of this function should reserve 5 units of metadata for
2006 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2008 struct btrfs_root *root = BTRFS_I(inode)->root;
2009 struct btrfs_block_rsv *block_rsv = NULL;
2014 if (!root->orphan_block_rsv) {
2015 block_rsv = btrfs_alloc_block_rsv(root);
2020 spin_lock(&root->orphan_lock);
2021 if (!root->orphan_block_rsv) {
2022 root->orphan_block_rsv = block_rsv;
2023 } else if (block_rsv) {
2024 btrfs_free_block_rsv(root, block_rsv);
2028 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2029 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2032 * For proper ENOSPC handling, we should do orphan
2033 * cleanup when mounting. But this introduces backward
2034 * compatibility issue.
2036 if (!xchg(&root->orphan_item_inserted, 1))
2044 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2045 BTRFS_I(inode)->orphan_meta_reserved = 1;
2048 spin_unlock(&root->orphan_lock);
2050 /* grab metadata reservation from transaction handle */
2052 ret = btrfs_orphan_reserve_metadata(trans, inode);
2056 /* insert an orphan item to track this unlinked/truncated file */
2058 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2059 BUG_ON(ret && ret != -EEXIST);
2062 /* insert an orphan item to track subvolume contains orphan files */
2064 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2065 root->root_key.objectid);
2072 * We have done the truncate/delete so we can go ahead and remove the orphan
2073 * item for this particular inode.
2075 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2077 struct btrfs_root *root = BTRFS_I(inode)->root;
2078 int delete_item = 0;
2079 int release_rsv = 0;
2082 spin_lock(&root->orphan_lock);
2083 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2084 list_del_init(&BTRFS_I(inode)->i_orphan);
2088 if (BTRFS_I(inode)->orphan_meta_reserved) {
2089 BTRFS_I(inode)->orphan_meta_reserved = 0;
2092 spin_unlock(&root->orphan_lock);
2094 if (trans && delete_item) {
2095 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2100 btrfs_orphan_release_metadata(inode);
2106 * this cleans up any orphans that may be left on the list from the last use
2109 int btrfs_orphan_cleanup(struct btrfs_root *root)
2111 struct btrfs_path *path;
2112 struct extent_buffer *leaf;
2113 struct btrfs_key key, found_key;
2114 struct btrfs_trans_handle *trans;
2115 struct inode *inode;
2116 u64 last_objectid = 0;
2117 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2119 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2122 path = btrfs_alloc_path();
2129 key.objectid = BTRFS_ORPHAN_OBJECTID;
2130 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2131 key.offset = (u64)-1;
2134 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2139 * if ret == 0 means we found what we were searching for, which
2140 * is weird, but possible, so only screw with path if we didn't
2141 * find the key and see if we have stuff that matches
2145 if (path->slots[0] == 0)
2150 /* pull out the item */
2151 leaf = path->nodes[0];
2152 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2154 /* make sure the item matches what we want */
2155 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2157 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2160 /* release the path since we're done with it */
2161 btrfs_release_path(path);
2164 * this is where we are basically btrfs_lookup, without the
2165 * crossing root thing. we store the inode number in the
2166 * offset of the orphan item.
2169 if (found_key.offset == last_objectid) {
2170 printk(KERN_ERR "btrfs: Error removing orphan entry, "
2171 "stopping orphan cleanup\n");
2176 last_objectid = found_key.offset;
2178 found_key.objectid = found_key.offset;
2179 found_key.type = BTRFS_INODE_ITEM_KEY;
2180 found_key.offset = 0;
2181 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2182 ret = PTR_RET(inode);
2183 if (ret && ret != -ESTALE)
2186 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2187 struct btrfs_root *dead_root;
2188 struct btrfs_fs_info *fs_info = root->fs_info;
2189 int is_dead_root = 0;
2192 * this is an orphan in the tree root. Currently these
2193 * could come from 2 sources:
2194 * a) a snapshot deletion in progress
2195 * b) a free space cache inode
2196 * We need to distinguish those two, as the snapshot
2197 * orphan must not get deleted.
2198 * find_dead_roots already ran before us, so if this
2199 * is a snapshot deletion, we should find the root
2200 * in the dead_roots list
2202 spin_lock(&fs_info->trans_lock);
2203 list_for_each_entry(dead_root, &fs_info->dead_roots,
2205 if (dead_root->root_key.objectid ==
2206 found_key.objectid) {
2211 spin_unlock(&fs_info->trans_lock);
2213 /* prevent this orphan from being found again */
2214 key.offset = found_key.objectid - 1;
2219 * Inode is already gone but the orphan item is still there,
2220 * kill the orphan item.
2222 if (ret == -ESTALE) {
2223 trans = btrfs_start_transaction(root, 1);
2224 if (IS_ERR(trans)) {
2225 ret = PTR_ERR(trans);
2228 ret = btrfs_del_orphan_item(trans, root,
2229 found_key.objectid);
2231 btrfs_end_transaction(trans, root);
2236 * add this inode to the orphan list so btrfs_orphan_del does
2237 * the proper thing when we hit it
2239 spin_lock(&root->orphan_lock);
2240 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2241 spin_unlock(&root->orphan_lock);
2243 /* if we have links, this was a truncate, lets do that */
2244 if (inode->i_nlink) {
2245 if (!S_ISREG(inode->i_mode)) {
2251 ret = btrfs_truncate(inode);
2256 /* this will do delete_inode and everything for us */
2261 /* release the path since we're done with it */
2262 btrfs_release_path(path);
2264 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2266 if (root->orphan_block_rsv)
2267 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2270 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2271 trans = btrfs_join_transaction(root);
2273 btrfs_end_transaction(trans, root);
2277 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2279 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2283 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2284 btrfs_free_path(path);
2289 * very simple check to peek ahead in the leaf looking for xattrs. If we
2290 * don't find any xattrs, we know there can't be any acls.
2292 * slot is the slot the inode is in, objectid is the objectid of the inode
2294 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2295 int slot, u64 objectid)
2297 u32 nritems = btrfs_header_nritems(leaf);
2298 struct btrfs_key found_key;
2302 while (slot < nritems) {
2303 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2305 /* we found a different objectid, there must not be acls */
2306 if (found_key.objectid != objectid)
2309 /* we found an xattr, assume we've got an acl */
2310 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2314 * we found a key greater than an xattr key, there can't
2315 * be any acls later on
2317 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2324 * it goes inode, inode backrefs, xattrs, extents,
2325 * so if there are a ton of hard links to an inode there can
2326 * be a lot of backrefs. Don't waste time searching too hard,
2327 * this is just an optimization
2332 /* we hit the end of the leaf before we found an xattr or
2333 * something larger than an xattr. We have to assume the inode
2340 * read an inode from the btree into the in-memory inode
2342 static void btrfs_read_locked_inode(struct inode *inode)
2344 struct btrfs_path *path;
2345 struct extent_buffer *leaf;
2346 struct btrfs_inode_item *inode_item;
2347 struct btrfs_timespec *tspec;
2348 struct btrfs_root *root = BTRFS_I(inode)->root;
2349 struct btrfs_key location;
2353 bool filled = false;
2355 ret = btrfs_fill_inode(inode, &rdev);
2359 path = btrfs_alloc_path();
2363 path->leave_spinning = 1;
2364 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2366 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2370 leaf = path->nodes[0];
2375 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2376 struct btrfs_inode_item);
2377 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2378 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2379 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2380 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2381 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2383 tspec = btrfs_inode_atime(inode_item);
2384 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2385 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2387 tspec = btrfs_inode_mtime(inode_item);
2388 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2389 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2391 tspec = btrfs_inode_ctime(inode_item);
2392 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2393 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2395 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2396 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2397 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2398 inode->i_generation = BTRFS_I(inode)->generation;
2400 rdev = btrfs_inode_rdev(leaf, inode_item);
2402 BTRFS_I(inode)->index_cnt = (u64)-1;
2403 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2406 * try to precache a NULL acl entry for files that don't have
2407 * any xattrs or acls
2409 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2412 cache_no_acl(inode);
2414 btrfs_free_path(path);
2416 switch (inode->i_mode & S_IFMT) {
2418 inode->i_mapping->a_ops = &btrfs_aops;
2419 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2420 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2421 inode->i_fop = &btrfs_file_operations;
2422 inode->i_op = &btrfs_file_inode_operations;
2425 inode->i_fop = &btrfs_dir_file_operations;
2426 if (root == root->fs_info->tree_root)
2427 inode->i_op = &btrfs_dir_ro_inode_operations;
2429 inode->i_op = &btrfs_dir_inode_operations;
2432 inode->i_op = &btrfs_symlink_inode_operations;
2433 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2434 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2437 inode->i_op = &btrfs_special_inode_operations;
2438 init_special_inode(inode, inode->i_mode, rdev);
2442 btrfs_update_iflags(inode);
2446 btrfs_free_path(path);
2447 make_bad_inode(inode);
2451 * given a leaf and an inode, copy the inode fields into the leaf
2453 static void fill_inode_item(struct btrfs_trans_handle *trans,
2454 struct extent_buffer *leaf,
2455 struct btrfs_inode_item *item,
2456 struct inode *inode)
2458 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2459 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2460 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2461 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2462 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2464 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2465 inode->i_atime.tv_sec);
2466 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2467 inode->i_atime.tv_nsec);
2469 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2470 inode->i_mtime.tv_sec);
2471 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2472 inode->i_mtime.tv_nsec);
2474 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2475 inode->i_ctime.tv_sec);
2476 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2477 inode->i_ctime.tv_nsec);
2479 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2480 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2481 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2482 btrfs_set_inode_transid(leaf, item, trans->transid);
2483 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2484 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2485 btrfs_set_inode_block_group(leaf, item, 0);
2489 * copy everything in the in-memory inode into the btree.
2491 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2492 struct btrfs_root *root, struct inode *inode)
2494 struct btrfs_inode_item *inode_item;
2495 struct btrfs_path *path;
2496 struct extent_buffer *leaf;
2499 path = btrfs_alloc_path();
2503 path->leave_spinning = 1;
2504 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2512 btrfs_unlock_up_safe(path, 1);
2513 leaf = path->nodes[0];
2514 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2515 struct btrfs_inode_item);
2517 fill_inode_item(trans, leaf, inode_item, inode);
2518 btrfs_mark_buffer_dirty(leaf);
2519 btrfs_set_inode_last_trans(trans, inode);
2522 btrfs_free_path(path);
2527 * copy everything in the in-memory inode into the btree.
2529 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2530 struct btrfs_root *root, struct inode *inode)
2535 * If the inode is a free space inode, we can deadlock during commit
2536 * if we put it into the delayed code.
2538 * The data relocation inode should also be directly updated
2541 if (!btrfs_is_free_space_inode(root, inode)
2542 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2543 ret = btrfs_delayed_update_inode(trans, root, inode);
2545 btrfs_set_inode_last_trans(trans, inode);
2549 return btrfs_update_inode_item(trans, root, inode);
2552 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2553 struct btrfs_root *root, struct inode *inode)
2557 ret = btrfs_update_inode(trans, root, inode);
2559 return btrfs_update_inode_item(trans, root, inode);
2564 * unlink helper that gets used here in inode.c and in the tree logging
2565 * recovery code. It remove a link in a directory with a given name, and
2566 * also drops the back refs in the inode to the directory
2568 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2569 struct btrfs_root *root,
2570 struct inode *dir, struct inode *inode,
2571 const char *name, int name_len)
2573 struct btrfs_path *path;
2575 struct extent_buffer *leaf;
2576 struct btrfs_dir_item *di;
2577 struct btrfs_key key;
2579 u64 ino = btrfs_ino(inode);
2580 u64 dir_ino = btrfs_ino(dir);
2582 path = btrfs_alloc_path();
2588 path->leave_spinning = 1;
2589 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2590 name, name_len, -1);
2599 leaf = path->nodes[0];
2600 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2601 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2604 btrfs_release_path(path);
2606 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2609 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2610 "inode %llu parent %llu\n", name_len, name,
2611 (unsigned long long)ino, (unsigned long long)dir_ino);
2615 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2619 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2621 BUG_ON(ret != 0 && ret != -ENOENT);
2623 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2628 btrfs_free_path(path);
2632 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2633 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2634 btrfs_update_inode(trans, root, dir);
2639 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2640 struct btrfs_root *root,
2641 struct inode *dir, struct inode *inode,
2642 const char *name, int name_len)
2645 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2647 btrfs_drop_nlink(inode);
2648 ret = btrfs_update_inode(trans, root, inode);
2654 /* helper to check if there is any shared block in the path */
2655 static int check_path_shared(struct btrfs_root *root,
2656 struct btrfs_path *path)
2658 struct extent_buffer *eb;
2662 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2665 if (!path->nodes[level])
2667 eb = path->nodes[level];
2668 if (!btrfs_block_can_be_shared(root, eb))
2670 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2679 * helper to start transaction for unlink and rmdir.
2681 * unlink and rmdir are special in btrfs, they do not always free space.
2682 * so in enospc case, we should make sure they will free space before
2683 * allowing them to use the global metadata reservation.
2685 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2686 struct dentry *dentry)
2688 struct btrfs_trans_handle *trans;
2689 struct btrfs_root *root = BTRFS_I(dir)->root;
2690 struct btrfs_path *path;
2691 struct btrfs_inode_ref *ref;
2692 struct btrfs_dir_item *di;
2693 struct inode *inode = dentry->d_inode;
2698 u64 ino = btrfs_ino(inode);
2699 u64 dir_ino = btrfs_ino(dir);
2702 * 1 for the possible orphan item
2703 * 1 for the dir item
2704 * 1 for the dir index
2705 * 1 for the inode ref
2706 * 1 for the inode ref in the tree log
2707 * 2 for the dir entries in the log
2710 trans = btrfs_start_transaction(root, 8);
2711 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2714 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2715 return ERR_PTR(-ENOSPC);
2717 /* check if there is someone else holds reference */
2718 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2719 return ERR_PTR(-ENOSPC);
2721 if (atomic_read(&inode->i_count) > 2)
2722 return ERR_PTR(-ENOSPC);
2724 if (xchg(&root->fs_info->enospc_unlink, 1))
2725 return ERR_PTR(-ENOSPC);
2727 path = btrfs_alloc_path();
2729 root->fs_info->enospc_unlink = 0;
2730 return ERR_PTR(-ENOMEM);
2733 /* 1 for the orphan item */
2734 trans = btrfs_start_transaction(root, 1);
2735 if (IS_ERR(trans)) {
2736 btrfs_free_path(path);
2737 root->fs_info->enospc_unlink = 0;
2741 path->skip_locking = 1;
2742 path->search_commit_root = 1;
2744 ret = btrfs_lookup_inode(trans, root, path,
2745 &BTRFS_I(dir)->location, 0);
2751 if (check_path_shared(root, path))
2756 btrfs_release_path(path);
2758 ret = btrfs_lookup_inode(trans, root, path,
2759 &BTRFS_I(inode)->location, 0);
2765 if (check_path_shared(root, path))
2770 btrfs_release_path(path);
2772 if (ret == 0 && S_ISREG(inode->i_mode)) {
2773 ret = btrfs_lookup_file_extent(trans, root, path,
2780 if (check_path_shared(root, path))
2782 btrfs_release_path(path);
2790 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2791 dentry->d_name.name, dentry->d_name.len, 0);
2797 if (check_path_shared(root, path))
2803 btrfs_release_path(path);
2805 ref = btrfs_lookup_inode_ref(trans, root, path,
2806 dentry->d_name.name, dentry->d_name.len,
2813 if (check_path_shared(root, path))
2815 index = btrfs_inode_ref_index(path->nodes[0], ref);
2816 btrfs_release_path(path);
2819 * This is a commit root search, if we can lookup inode item and other
2820 * relative items in the commit root, it means the transaction of
2821 * dir/file creation has been committed, and the dir index item that we
2822 * delay to insert has also been inserted into the commit root. So
2823 * we needn't worry about the delayed insertion of the dir index item
2826 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2827 dentry->d_name.name, dentry->d_name.len, 0);
2832 BUG_ON(ret == -ENOENT);
2833 if (check_path_shared(root, path))
2838 btrfs_free_path(path);
2839 /* Migrate the orphan reservation over */
2841 err = btrfs_block_rsv_migrate(trans->block_rsv,
2842 &root->fs_info->global_block_rsv,
2843 trans->bytes_reserved);
2846 btrfs_end_transaction(trans, root);
2847 root->fs_info->enospc_unlink = 0;
2848 return ERR_PTR(err);
2851 trans->block_rsv = &root->fs_info->global_block_rsv;
2855 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2856 struct btrfs_root *root)
2858 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2859 btrfs_block_rsv_release(root, trans->block_rsv,
2860 trans->bytes_reserved);
2861 trans->block_rsv = &root->fs_info->trans_block_rsv;
2862 BUG_ON(!root->fs_info->enospc_unlink);
2863 root->fs_info->enospc_unlink = 0;
2865 btrfs_end_transaction(trans, root);
2868 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2870 struct btrfs_root *root = BTRFS_I(dir)->root;
2871 struct btrfs_trans_handle *trans;
2872 struct inode *inode = dentry->d_inode;
2874 unsigned long nr = 0;
2876 trans = __unlink_start_trans(dir, dentry);
2878 return PTR_ERR(trans);
2880 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2882 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2883 dentry->d_name.name, dentry->d_name.len);
2887 if (inode->i_nlink == 0) {
2888 ret = btrfs_orphan_add(trans, inode);
2894 nr = trans->blocks_used;
2895 __unlink_end_trans(trans, root);
2896 btrfs_btree_balance_dirty(root, nr);
2900 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2901 struct btrfs_root *root,
2902 struct inode *dir, u64 objectid,
2903 const char *name, int name_len)
2905 struct btrfs_path *path;
2906 struct extent_buffer *leaf;
2907 struct btrfs_dir_item *di;
2908 struct btrfs_key key;
2911 u64 dir_ino = btrfs_ino(dir);
2913 path = btrfs_alloc_path();
2917 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2918 name, name_len, -1);
2919 BUG_ON(IS_ERR_OR_NULL(di));
2921 leaf = path->nodes[0];
2922 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2923 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2924 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2926 btrfs_release_path(path);
2928 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2929 objectid, root->root_key.objectid,
2930 dir_ino, &index, name, name_len);
2932 BUG_ON(ret != -ENOENT);
2933 di = btrfs_search_dir_index_item(root, path, dir_ino,
2935 BUG_ON(IS_ERR_OR_NULL(di));
2937 leaf = path->nodes[0];
2938 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2939 btrfs_release_path(path);
2942 btrfs_release_path(path);
2944 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2947 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2948 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2949 ret = btrfs_update_inode(trans, root, dir);
2952 btrfs_free_path(path);
2956 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2958 struct inode *inode = dentry->d_inode;
2960 struct btrfs_root *root = BTRFS_I(dir)->root;
2961 struct btrfs_trans_handle *trans;
2962 unsigned long nr = 0;
2964 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2965 btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
2968 trans = __unlink_start_trans(dir, dentry);
2970 return PTR_ERR(trans);
2972 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2973 err = btrfs_unlink_subvol(trans, root, dir,
2974 BTRFS_I(inode)->location.objectid,
2975 dentry->d_name.name,
2976 dentry->d_name.len);
2980 err = btrfs_orphan_add(trans, inode);
2984 /* now the directory is empty */
2985 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2986 dentry->d_name.name, dentry->d_name.len);
2988 btrfs_i_size_write(inode, 0);
2990 nr = trans->blocks_used;
2991 __unlink_end_trans(trans, root);
2992 btrfs_btree_balance_dirty(root, nr);
2998 * this can truncate away extent items, csum items and directory items.
2999 * It starts at a high offset and removes keys until it can't find
3000 * any higher than new_size
3002 * csum items that cross the new i_size are truncated to the new size
3005 * min_type is the minimum key type to truncate down to. If set to 0, this
3006 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3008 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3009 struct btrfs_root *root,
3010 struct inode *inode,
3011 u64 new_size, u32 min_type)
3013 struct btrfs_path *path;
3014 struct extent_buffer *leaf;
3015 struct btrfs_file_extent_item *fi;
3016 struct btrfs_key key;
3017 struct btrfs_key found_key;
3018 u64 extent_start = 0;
3019 u64 extent_num_bytes = 0;
3020 u64 extent_offset = 0;
3022 u64 mask = root->sectorsize - 1;
3023 u32 found_type = (u8)-1;
3026 int pending_del_nr = 0;
3027 int pending_del_slot = 0;
3028 int extent_type = -1;
3031 u64 ino = btrfs_ino(inode);
3033 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3035 path = btrfs_alloc_path();
3040 if (root->ref_cows || root == root->fs_info->tree_root)
3041 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3044 * This function is also used to drop the items in the log tree before
3045 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3046 * it is used to drop the loged items. So we shouldn't kill the delayed
3049 if (min_type == 0 && root == BTRFS_I(inode)->root)
3050 btrfs_kill_delayed_inode_items(inode);
3053 key.offset = (u64)-1;
3057 path->leave_spinning = 1;
3058 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3065 /* there are no items in the tree for us to truncate, we're
3068 if (path->slots[0] == 0)
3075 leaf = path->nodes[0];
3076 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3077 found_type = btrfs_key_type(&found_key);
3079 if (found_key.objectid != ino)
3082 if (found_type < min_type)
3085 item_end = found_key.offset;
3086 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3087 fi = btrfs_item_ptr(leaf, path->slots[0],
3088 struct btrfs_file_extent_item);
3089 extent_type = btrfs_file_extent_type(leaf, fi);
3090 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3092 btrfs_file_extent_num_bytes(leaf, fi);
3093 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3094 item_end += btrfs_file_extent_inline_len(leaf,
3099 if (found_type > min_type) {
3102 if (item_end < new_size)
3104 if (found_key.offset >= new_size)
3110 /* FIXME, shrink the extent if the ref count is only 1 */
3111 if (found_type != BTRFS_EXTENT_DATA_KEY)
3114 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3116 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3118 u64 orig_num_bytes =
3119 btrfs_file_extent_num_bytes(leaf, fi);
3120 extent_num_bytes = new_size -
3121 found_key.offset + root->sectorsize - 1;
3122 extent_num_bytes = extent_num_bytes &
3123 ~((u64)root->sectorsize - 1);
3124 btrfs_set_file_extent_num_bytes(leaf, fi,
3126 num_dec = (orig_num_bytes -
3128 if (root->ref_cows && extent_start != 0)
3129 inode_sub_bytes(inode, num_dec);
3130 btrfs_mark_buffer_dirty(leaf);
3133 btrfs_file_extent_disk_num_bytes(leaf,
3135 extent_offset = found_key.offset -
3136 btrfs_file_extent_offset(leaf, fi);
3138 /* FIXME blocksize != 4096 */
3139 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3140 if (extent_start != 0) {
3143 inode_sub_bytes(inode, num_dec);
3146 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3148 * we can't truncate inline items that have had
3152 btrfs_file_extent_compression(leaf, fi) == 0 &&
3153 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3154 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3155 u32 size = new_size - found_key.offset;
3157 if (root->ref_cows) {
3158 inode_sub_bytes(inode, item_end + 1 -
3162 btrfs_file_extent_calc_inline_size(size);
3163 ret = btrfs_truncate_item(trans, root, path,
3165 } else if (root->ref_cows) {
3166 inode_sub_bytes(inode, item_end + 1 -
3172 if (!pending_del_nr) {
3173 /* no pending yet, add ourselves */
3174 pending_del_slot = path->slots[0];
3176 } else if (pending_del_nr &&
3177 path->slots[0] + 1 == pending_del_slot) {
3178 /* hop on the pending chunk */
3180 pending_del_slot = path->slots[0];
3187 if (found_extent && (root->ref_cows ||
3188 root == root->fs_info->tree_root)) {
3189 btrfs_set_path_blocking(path);
3190 ret = btrfs_free_extent(trans, root, extent_start,
3191 extent_num_bytes, 0,
3192 btrfs_header_owner(leaf),
3193 ino, extent_offset, 0);
3197 if (found_type == BTRFS_INODE_ITEM_KEY)
3200 if (path->slots[0] == 0 ||
3201 path->slots[0] != pending_del_slot) {
3202 if (root->ref_cows &&
3203 BTRFS_I(inode)->location.objectid !=
3204 BTRFS_FREE_INO_OBJECTID) {
3208 if (pending_del_nr) {
3209 ret = btrfs_del_items(trans, root, path,
3215 btrfs_release_path(path);
3222 if (pending_del_nr) {
3223 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3227 btrfs_free_path(path);
3232 * taken from block_truncate_page, but does cow as it zeros out
3233 * any bytes left in the last page in the file.
3235 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3237 struct inode *inode = mapping->host;
3238 struct btrfs_root *root = BTRFS_I(inode)->root;
3239 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3240 struct btrfs_ordered_extent *ordered;
3241 struct extent_state *cached_state = NULL;
3243 u32 blocksize = root->sectorsize;
3244 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3245 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3247 gfp_t mask = btrfs_alloc_write_mask(mapping);
3252 if ((offset & (blocksize - 1)) == 0)
3254 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3260 page = find_or_create_page(mapping, index, mask);
3262 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3266 page_start = page_offset(page);
3267 page_end = page_start + PAGE_CACHE_SIZE - 1;
3269 if (!PageUptodate(page)) {
3270 ret = btrfs_readpage(NULL, page);
3272 if (page->mapping != mapping) {
3274 page_cache_release(page);
3277 if (!PageUptodate(page)) {
3282 wait_on_page_writeback(page);
3284 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3286 set_page_extent_mapped(page);
3288 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3290 unlock_extent_cached(io_tree, page_start, page_end,
3291 &cached_state, GFP_NOFS);
3293 page_cache_release(page);
3294 btrfs_start_ordered_extent(inode, ordered, 1);
3295 btrfs_put_ordered_extent(ordered);
3299 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3300 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3301 0, 0, &cached_state, GFP_NOFS);
3303 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3306 unlock_extent_cached(io_tree, page_start, page_end,
3307 &cached_state, GFP_NOFS);
3312 if (offset != PAGE_CACHE_SIZE) {
3314 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3315 flush_dcache_page(page);
3318 ClearPageChecked(page);
3319 set_page_dirty(page);
3320 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3325 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3327 page_cache_release(page);
3333 * This function puts in dummy file extents for the area we're creating a hole
3334 * for. So if we are truncating this file to a larger size we need to insert
3335 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3336 * the range between oldsize and size
3338 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3340 struct btrfs_trans_handle *trans;
3341 struct btrfs_root *root = BTRFS_I(inode)->root;
3342 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3343 struct extent_map *em = NULL;
3344 struct extent_state *cached_state = NULL;
3345 u64 mask = root->sectorsize - 1;
3346 u64 hole_start = (oldsize + mask) & ~mask;
3347 u64 block_end = (size + mask) & ~mask;
3353 if (size <= hole_start)
3357 struct btrfs_ordered_extent *ordered;
3358 btrfs_wait_ordered_range(inode, hole_start,
3359 block_end - hole_start);
3360 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3361 &cached_state, GFP_NOFS);
3362 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3365 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3366 &cached_state, GFP_NOFS);
3367 btrfs_put_ordered_extent(ordered);
3370 cur_offset = hole_start;
3372 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3373 block_end - cur_offset, 0);
3374 BUG_ON(IS_ERR_OR_NULL(em));
3375 last_byte = min(extent_map_end(em), block_end);
3376 last_byte = (last_byte + mask) & ~mask;
3377 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3379 hole_size = last_byte - cur_offset;
3381 trans = btrfs_start_transaction(root, 3);
3382 if (IS_ERR(trans)) {
3383 err = PTR_ERR(trans);
3387 err = btrfs_drop_extents(trans, inode, cur_offset,
3388 cur_offset + hole_size,
3391 btrfs_update_inode(trans, root, inode);
3392 btrfs_end_transaction(trans, root);
3396 err = btrfs_insert_file_extent(trans, root,
3397 btrfs_ino(inode), cur_offset, 0,
3398 0, hole_size, 0, hole_size,
3401 btrfs_update_inode(trans, root, inode);
3402 btrfs_end_transaction(trans, root);
3406 btrfs_drop_extent_cache(inode, hole_start,
3409 btrfs_update_inode(trans, root, inode);
3410 btrfs_end_transaction(trans, root);
3412 free_extent_map(em);
3414 cur_offset = last_byte;
3415 if (cur_offset >= block_end)
3419 free_extent_map(em);
3420 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3425 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3427 struct btrfs_root *root = BTRFS_I(inode)->root;
3428 struct btrfs_trans_handle *trans;
3429 loff_t oldsize = i_size_read(inode);
3432 if (newsize == oldsize)
3435 if (newsize > oldsize) {
3436 truncate_pagecache(inode, oldsize, newsize);
3437 ret = btrfs_cont_expand(inode, oldsize, newsize);
3441 trans = btrfs_start_transaction(root, 1);
3443 return PTR_ERR(trans);
3445 i_size_write(inode, newsize);
3446 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3447 ret = btrfs_update_inode(trans, root, inode);
3448 btrfs_end_transaction(trans, root);
3452 * We're truncating a file that used to have good data down to
3453 * zero. Make sure it gets into the ordered flush list so that
3454 * any new writes get down to disk quickly.
3457 BTRFS_I(inode)->ordered_data_close = 1;
3459 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3460 truncate_setsize(inode, newsize);
3461 ret = btrfs_truncate(inode);
3467 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3469 struct inode *inode = dentry->d_inode;
3470 struct btrfs_root *root = BTRFS_I(inode)->root;
3473 if (btrfs_root_readonly(root))
3476 err = inode_change_ok(inode, attr);
3480 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3481 err = btrfs_setsize(inode, attr->ia_size);
3486 if (attr->ia_valid) {
3487 setattr_copy(inode, attr);
3488 err = btrfs_dirty_inode(inode);
3490 if (!err && attr->ia_valid & ATTR_MODE)
3491 err = btrfs_acl_chmod(inode);
3497 void btrfs_evict_inode(struct inode *inode)
3499 struct btrfs_trans_handle *trans;
3500 struct btrfs_root *root = BTRFS_I(inode)->root;
3501 struct btrfs_block_rsv *rsv, *global_rsv;
3502 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3506 trace_btrfs_inode_evict(inode);
3508 truncate_inode_pages(&inode->i_data, 0);
3509 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3510 btrfs_is_free_space_inode(root, inode)))
3513 if (is_bad_inode(inode)) {
3514 btrfs_orphan_del(NULL, inode);
3517 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3518 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3520 if (root->fs_info->log_root_recovering) {
3521 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3525 if (inode->i_nlink > 0) {
3526 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3530 rsv = btrfs_alloc_block_rsv(root);
3532 btrfs_orphan_del(NULL, inode);
3535 rsv->size = min_size;
3536 global_rsv = &root->fs_info->global_block_rsv;
3538 btrfs_i_size_write(inode, 0);
3541 * This is a bit simpler than btrfs_truncate since
3543 * 1) We've already reserved our space for our orphan item in the
3545 * 2) We're going to delete the inode item, so we don't need to update
3548 * So we just need to reserve some slack space in case we add bytes when
3549 * doing the truncate.
3552 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
3555 * Try and steal from the global reserve since we will
3556 * likely not use this space anyway, we want to try as
3557 * hard as possible to get this to work.
3560 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3563 printk(KERN_WARNING "Could not get space for a "
3564 "delete, will truncate on mount %d\n", ret);
3565 btrfs_orphan_del(NULL, inode);
3566 btrfs_free_block_rsv(root, rsv);
3570 trans = btrfs_start_transaction(root, 0);
3571 if (IS_ERR(trans)) {
3572 btrfs_orphan_del(NULL, inode);
3573 btrfs_free_block_rsv(root, rsv);
3577 trans->block_rsv = rsv;
3579 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3583 nr = trans->blocks_used;
3584 btrfs_end_transaction(trans, root);
3586 btrfs_btree_balance_dirty(root, nr);
3589 btrfs_free_block_rsv(root, rsv);
3592 trans->block_rsv = root->orphan_block_rsv;
3593 ret = btrfs_orphan_del(trans, inode);
3597 trans->block_rsv = &root->fs_info->trans_block_rsv;
3598 if (!(root == root->fs_info->tree_root ||
3599 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3600 btrfs_return_ino(root, btrfs_ino(inode));
3602 nr = trans->blocks_used;
3603 btrfs_end_transaction(trans, root);
3604 btrfs_btree_balance_dirty(root, nr);
3606 end_writeback(inode);
3611 * this returns the key found in the dir entry in the location pointer.
3612 * If no dir entries were found, location->objectid is 0.
3614 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3615 struct btrfs_key *location)
3617 const char *name = dentry->d_name.name;
3618 int namelen = dentry->d_name.len;
3619 struct btrfs_dir_item *di;
3620 struct btrfs_path *path;
3621 struct btrfs_root *root = BTRFS_I(dir)->root;
3624 path = btrfs_alloc_path();
3628 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3633 if (IS_ERR_OR_NULL(di))
3636 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3638 btrfs_free_path(path);
3641 location->objectid = 0;
3646 * when we hit a tree root in a directory, the btrfs part of the inode
3647 * needs to be changed to reflect the root directory of the tree root. This
3648 * is kind of like crossing a mount point.
3650 static int fixup_tree_root_location(struct btrfs_root *root,
3652 struct dentry *dentry,
3653 struct btrfs_key *location,
3654 struct btrfs_root **sub_root)
3656 struct btrfs_path *path;
3657 struct btrfs_root *new_root;
3658 struct btrfs_root_ref *ref;
3659 struct extent_buffer *leaf;
3663 path = btrfs_alloc_path();
3670 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3671 BTRFS_I(dir)->root->root_key.objectid,
3672 location->objectid);
3679 leaf = path->nodes[0];
3680 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3681 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3682 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3685 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3686 (unsigned long)(ref + 1),
3687 dentry->d_name.len);
3691 btrfs_release_path(path);
3693 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3694 if (IS_ERR(new_root)) {
3695 err = PTR_ERR(new_root);
3699 if (btrfs_root_refs(&new_root->root_item) == 0) {
3704 *sub_root = new_root;
3705 location->objectid = btrfs_root_dirid(&new_root->root_item);
3706 location->type = BTRFS_INODE_ITEM_KEY;
3707 location->offset = 0;
3710 btrfs_free_path(path);
3714 static void inode_tree_add(struct inode *inode)
3716 struct btrfs_root *root = BTRFS_I(inode)->root;
3717 struct btrfs_inode *entry;
3719 struct rb_node *parent;
3720 u64 ino = btrfs_ino(inode);
3722 p = &root->inode_tree.rb_node;
3725 if (inode_unhashed(inode))
3728 spin_lock(&root->inode_lock);
3731 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3733 if (ino < btrfs_ino(&entry->vfs_inode))
3734 p = &parent->rb_left;
3735 else if (ino > btrfs_ino(&entry->vfs_inode))
3736 p = &parent->rb_right;
3738 WARN_ON(!(entry->vfs_inode.i_state &
3739 (I_WILL_FREE | I_FREEING)));
3740 rb_erase(parent, &root->inode_tree);
3741 RB_CLEAR_NODE(parent);
3742 spin_unlock(&root->inode_lock);
3746 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3747 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3748 spin_unlock(&root->inode_lock);
3751 static void inode_tree_del(struct inode *inode)
3753 struct btrfs_root *root = BTRFS_I(inode)->root;
3756 spin_lock(&root->inode_lock);
3757 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3758 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3759 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3760 empty = RB_EMPTY_ROOT(&root->inode_tree);
3762 spin_unlock(&root->inode_lock);
3765 * Free space cache has inodes in the tree root, but the tree root has a
3766 * root_refs of 0, so this could end up dropping the tree root as a
3767 * snapshot, so we need the extra !root->fs_info->tree_root check to
3768 * make sure we don't drop it.
3770 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3771 root != root->fs_info->tree_root) {
3772 synchronize_srcu(&root->fs_info->subvol_srcu);
3773 spin_lock(&root->inode_lock);
3774 empty = RB_EMPTY_ROOT(&root->inode_tree);
3775 spin_unlock(&root->inode_lock);
3777 btrfs_add_dead_root(root);
3781 int btrfs_invalidate_inodes(struct btrfs_root *root)
3783 struct rb_node *node;
3784 struct rb_node *prev;
3785 struct btrfs_inode *entry;
3786 struct inode *inode;
3789 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3791 spin_lock(&root->inode_lock);
3793 node = root->inode_tree.rb_node;
3797 entry = rb_entry(node, struct btrfs_inode, rb_node);
3799 if (objectid < btrfs_ino(&entry->vfs_inode))
3800 node = node->rb_left;
3801 else if (objectid > btrfs_ino(&entry->vfs_inode))
3802 node = node->rb_right;
3808 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3809 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3813 prev = rb_next(prev);
3817 entry = rb_entry(node, struct btrfs_inode, rb_node);
3818 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3819 inode = igrab(&entry->vfs_inode);
3821 spin_unlock(&root->inode_lock);
3822 if (atomic_read(&inode->i_count) > 1)
3823 d_prune_aliases(inode);
3825 * btrfs_drop_inode will have it removed from
3826 * the inode cache when its usage count
3831 spin_lock(&root->inode_lock);
3835 if (cond_resched_lock(&root->inode_lock))
3838 node = rb_next(node);
3840 spin_unlock(&root->inode_lock);
3844 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3846 struct btrfs_iget_args *args = p;
3847 inode->i_ino = args->ino;
3848 BTRFS_I(inode)->root = args->root;
3849 btrfs_set_inode_space_info(args->root, inode);
3853 static int btrfs_find_actor(struct inode *inode, void *opaque)
3855 struct btrfs_iget_args *args = opaque;
3856 return args->ino == btrfs_ino(inode) &&
3857 args->root == BTRFS_I(inode)->root;
3860 static struct inode *btrfs_iget_locked(struct super_block *s,
3862 struct btrfs_root *root)
3864 struct inode *inode;
3865 struct btrfs_iget_args args;
3866 args.ino = objectid;
3869 inode = iget5_locked(s, objectid, btrfs_find_actor,
3870 btrfs_init_locked_inode,
3875 /* Get an inode object given its location and corresponding root.
3876 * Returns in *is_new if the inode was read from disk
3878 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3879 struct btrfs_root *root, int *new)
3881 struct inode *inode;
3883 inode = btrfs_iget_locked(s, location->objectid, root);
3885 return ERR_PTR(-ENOMEM);
3887 if (inode->i_state & I_NEW) {
3888 BTRFS_I(inode)->root = root;
3889 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3890 btrfs_read_locked_inode(inode);
3891 if (!is_bad_inode(inode)) {
3892 inode_tree_add(inode);
3893 unlock_new_inode(inode);
3897 unlock_new_inode(inode);
3899 inode = ERR_PTR(-ESTALE);
3906 static struct inode *new_simple_dir(struct super_block *s,
3907 struct btrfs_key *key,
3908 struct btrfs_root *root)
3910 struct inode *inode = new_inode(s);
3913 return ERR_PTR(-ENOMEM);
3915 BTRFS_I(inode)->root = root;
3916 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3917 BTRFS_I(inode)->dummy_inode = 1;
3919 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3920 inode->i_op = &simple_dir_inode_operations;
3921 inode->i_fop = &simple_dir_operations;
3922 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3923 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3928 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3930 struct inode *inode;
3931 struct btrfs_root *root = BTRFS_I(dir)->root;
3932 struct btrfs_root *sub_root = root;
3933 struct btrfs_key location;
3937 if (dentry->d_name.len > BTRFS_NAME_LEN)
3938 return ERR_PTR(-ENAMETOOLONG);
3940 if (unlikely(d_need_lookup(dentry))) {
3941 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
3942 kfree(dentry->d_fsdata);
3943 dentry->d_fsdata = NULL;
3944 /* This thing is hashed, drop it for now */
3947 ret = btrfs_inode_by_name(dir, dentry, &location);
3951 return ERR_PTR(ret);
3953 if (location.objectid == 0)
3956 if (location.type == BTRFS_INODE_ITEM_KEY) {
3957 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
3961 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3963 index = srcu_read_lock(&root->fs_info->subvol_srcu);
3964 ret = fixup_tree_root_location(root, dir, dentry,
3965 &location, &sub_root);
3968 inode = ERR_PTR(ret);
3970 inode = new_simple_dir(dir->i_sb, &location, sub_root);
3972 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
3974 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3976 if (!IS_ERR(inode) && root != sub_root) {
3977 down_read(&root->fs_info->cleanup_work_sem);
3978 if (!(inode->i_sb->s_flags & MS_RDONLY))
3979 ret = btrfs_orphan_cleanup(sub_root);
3980 up_read(&root->fs_info->cleanup_work_sem);
3982 inode = ERR_PTR(ret);
3988 static int btrfs_dentry_delete(const struct dentry *dentry)
3990 struct btrfs_root *root;
3992 if (!dentry->d_inode && !IS_ROOT(dentry))
3993 dentry = dentry->d_parent;
3995 if (dentry->d_inode) {
3996 root = BTRFS_I(dentry->d_inode)->root;
3997 if (btrfs_root_refs(&root->root_item) == 0)
4003 static void btrfs_dentry_release(struct dentry *dentry)
4005 if (dentry->d_fsdata)
4006 kfree(dentry->d_fsdata);
4009 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4010 struct nameidata *nd)
4014 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4015 if (unlikely(d_need_lookup(dentry))) {
4016 spin_lock(&dentry->d_lock);
4017 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4018 spin_unlock(&dentry->d_lock);
4023 unsigned char btrfs_filetype_table[] = {
4024 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4027 static int btrfs_real_readdir(struct file *filp, void *dirent,
4030 struct inode *inode = filp->f_dentry->d_inode;
4031 struct btrfs_root *root = BTRFS_I(inode)->root;
4032 struct btrfs_item *item;
4033 struct btrfs_dir_item *di;
4034 struct btrfs_key key;
4035 struct btrfs_key found_key;
4036 struct btrfs_path *path;
4037 struct list_head ins_list;
4038 struct list_head del_list;
4041 struct extent_buffer *leaf;
4043 unsigned char d_type;
4048 int key_type = BTRFS_DIR_INDEX_KEY;
4052 int is_curr = 0; /* filp->f_pos points to the current index? */
4054 /* FIXME, use a real flag for deciding about the key type */
4055 if (root->fs_info->tree_root == root)
4056 key_type = BTRFS_DIR_ITEM_KEY;
4058 /* special case for "." */
4059 if (filp->f_pos == 0) {
4060 over = filldir(dirent, ".", 1,
4061 filp->f_pos, btrfs_ino(inode), DT_DIR);
4066 /* special case for .., just use the back ref */
4067 if (filp->f_pos == 1) {
4068 u64 pino = parent_ino(filp->f_path.dentry);
4069 over = filldir(dirent, "..", 2,
4070 filp->f_pos, pino, DT_DIR);
4075 path = btrfs_alloc_path();
4081 if (key_type == BTRFS_DIR_INDEX_KEY) {
4082 INIT_LIST_HEAD(&ins_list);
4083 INIT_LIST_HEAD(&del_list);
4084 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4087 btrfs_set_key_type(&key, key_type);
4088 key.offset = filp->f_pos;
4089 key.objectid = btrfs_ino(inode);
4091 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4096 leaf = path->nodes[0];
4097 slot = path->slots[0];
4098 if (slot >= btrfs_header_nritems(leaf)) {
4099 ret = btrfs_next_leaf(root, path);
4107 item = btrfs_item_nr(leaf, slot);
4108 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4110 if (found_key.objectid != key.objectid)
4112 if (btrfs_key_type(&found_key) != key_type)
4114 if (found_key.offset < filp->f_pos)
4116 if (key_type == BTRFS_DIR_INDEX_KEY &&
4117 btrfs_should_delete_dir_index(&del_list,
4121 filp->f_pos = found_key.offset;
4124 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4126 di_total = btrfs_item_size(leaf, item);
4128 while (di_cur < di_total) {
4129 struct btrfs_key location;
4132 if (verify_dir_item(root, leaf, di))
4135 name_len = btrfs_dir_name_len(leaf, di);
4136 if (name_len <= sizeof(tmp_name)) {
4137 name_ptr = tmp_name;
4139 name_ptr = kmalloc(name_len, GFP_NOFS);
4145 read_extent_buffer(leaf, name_ptr,
4146 (unsigned long)(di + 1), name_len);
4148 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4149 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4153 q.hash = full_name_hash(q.name, q.len);
4154 tmp = d_lookup(filp->f_dentry, &q);
4156 struct btrfs_key *newkey;
4158 newkey = kzalloc(sizeof(struct btrfs_key),
4162 tmp = d_alloc(filp->f_dentry, &q);
4168 memcpy(newkey, &location,
4169 sizeof(struct btrfs_key));
4170 tmp->d_fsdata = newkey;
4171 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4178 /* is this a reference to our own snapshot? If so
4181 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4182 location.objectid == root->root_key.objectid) {
4186 over = filldir(dirent, name_ptr, name_len,
4187 found_key.offset, location.objectid,
4191 if (name_ptr != tmp_name)
4196 di_len = btrfs_dir_name_len(leaf, di) +
4197 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4199 di = (struct btrfs_dir_item *)((char *)di + di_len);
4205 if (key_type == BTRFS_DIR_INDEX_KEY) {
4208 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4214 /* Reached end of directory/root. Bump pos past the last item. */
4215 if (key_type == BTRFS_DIR_INDEX_KEY)
4217 * 32-bit glibc will use getdents64, but then strtol -
4218 * so the last number we can serve is this.
4220 filp->f_pos = 0x7fffffff;
4226 if (key_type == BTRFS_DIR_INDEX_KEY)
4227 btrfs_put_delayed_items(&ins_list, &del_list);
4228 btrfs_free_path(path);
4232 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4234 struct btrfs_root *root = BTRFS_I(inode)->root;
4235 struct btrfs_trans_handle *trans;
4237 bool nolock = false;
4239 if (BTRFS_I(inode)->dummy_inode)
4242 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4245 if (wbc->sync_mode == WB_SYNC_ALL) {
4247 trans = btrfs_join_transaction_nolock(root);
4249 trans = btrfs_join_transaction(root);
4251 return PTR_ERR(trans);
4253 ret = btrfs_end_transaction_nolock(trans, root);
4255 ret = btrfs_commit_transaction(trans, root);
4261 * This is somewhat expensive, updating the tree every time the
4262 * inode changes. But, it is most likely to find the inode in cache.
4263 * FIXME, needs more benchmarking...there are no reasons other than performance
4264 * to keep or drop this code.
4266 int btrfs_dirty_inode(struct inode *inode)
4268 struct btrfs_root *root = BTRFS_I(inode)->root;
4269 struct btrfs_trans_handle *trans;
4272 if (BTRFS_I(inode)->dummy_inode)
4275 trans = btrfs_join_transaction(root);
4277 return PTR_ERR(trans);
4279 ret = btrfs_update_inode(trans, root, inode);
4280 if (ret && ret == -ENOSPC) {
4281 /* whoops, lets try again with the full transaction */
4282 btrfs_end_transaction(trans, root);
4283 trans = btrfs_start_transaction(root, 1);
4285 return PTR_ERR(trans);
4287 ret = btrfs_update_inode(trans, root, inode);
4289 btrfs_end_transaction(trans, root);
4290 if (BTRFS_I(inode)->delayed_node)
4291 btrfs_balance_delayed_items(root);
4297 * This is a copy of file_update_time. We need this so we can return error on
4298 * ENOSPC for updating the inode in the case of file write and mmap writes.
4300 int btrfs_update_time(struct file *file)
4302 struct inode *inode = file->f_path.dentry->d_inode;
4303 struct timespec now;
4305 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
4307 /* First try to exhaust all avenues to not sync */
4308 if (IS_NOCMTIME(inode))
4311 now = current_fs_time(inode->i_sb);
4312 if (!timespec_equal(&inode->i_mtime, &now))
4315 if (!timespec_equal(&inode->i_ctime, &now))
4318 if (IS_I_VERSION(inode))
4319 sync_it |= S_VERSION;
4324 /* Finally allowed to write? Takes lock. */
4325 if (mnt_want_write_file(file))
4328 /* Only change inode inside the lock region */
4329 if (sync_it & S_VERSION)
4330 inode_inc_iversion(inode);
4331 if (sync_it & S_CTIME)
4332 inode->i_ctime = now;
4333 if (sync_it & S_MTIME)
4334 inode->i_mtime = now;
4335 ret = btrfs_dirty_inode(inode);
4337 mark_inode_dirty_sync(inode);
4338 mnt_drop_write(file->f_path.mnt);
4343 * find the highest existing sequence number in a directory
4344 * and then set the in-memory index_cnt variable to reflect
4345 * free sequence numbers
4347 static int btrfs_set_inode_index_count(struct inode *inode)
4349 struct btrfs_root *root = BTRFS_I(inode)->root;
4350 struct btrfs_key key, found_key;
4351 struct btrfs_path *path;
4352 struct extent_buffer *leaf;
4355 key.objectid = btrfs_ino(inode);
4356 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4357 key.offset = (u64)-1;
4359 path = btrfs_alloc_path();
4363 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4366 /* FIXME: we should be able to handle this */
4372 * MAGIC NUMBER EXPLANATION:
4373 * since we search a directory based on f_pos we have to start at 2
4374 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4375 * else has to start at 2
4377 if (path->slots[0] == 0) {
4378 BTRFS_I(inode)->index_cnt = 2;
4384 leaf = path->nodes[0];
4385 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4387 if (found_key.objectid != btrfs_ino(inode) ||
4388 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4389 BTRFS_I(inode)->index_cnt = 2;
4393 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4395 btrfs_free_path(path);
4400 * helper to find a free sequence number in a given directory. This current
4401 * code is very simple, later versions will do smarter things in the btree
4403 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4407 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4408 ret = btrfs_inode_delayed_dir_index_count(dir);
4410 ret = btrfs_set_inode_index_count(dir);
4416 *index = BTRFS_I(dir)->index_cnt;
4417 BTRFS_I(dir)->index_cnt++;
4422 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4423 struct btrfs_root *root,
4425 const char *name, int name_len,
4426 u64 ref_objectid, u64 objectid,
4427 umode_t mode, u64 *index)
4429 struct inode *inode;
4430 struct btrfs_inode_item *inode_item;
4431 struct btrfs_key *location;
4432 struct btrfs_path *path;
4433 struct btrfs_inode_ref *ref;
4434 struct btrfs_key key[2];
4440 path = btrfs_alloc_path();
4442 return ERR_PTR(-ENOMEM);
4444 inode = new_inode(root->fs_info->sb);
4446 btrfs_free_path(path);
4447 return ERR_PTR(-ENOMEM);
4451 * we have to initialize this early, so we can reclaim the inode
4452 * number if we fail afterwards in this function.
4454 inode->i_ino = objectid;
4457 trace_btrfs_inode_request(dir);
4459 ret = btrfs_set_inode_index(dir, index);
4461 btrfs_free_path(path);
4463 return ERR_PTR(ret);
4467 * index_cnt is ignored for everything but a dir,
4468 * btrfs_get_inode_index_count has an explanation for the magic
4471 BTRFS_I(inode)->index_cnt = 2;
4472 BTRFS_I(inode)->root = root;
4473 BTRFS_I(inode)->generation = trans->transid;
4474 inode->i_generation = BTRFS_I(inode)->generation;
4475 btrfs_set_inode_space_info(root, inode);
4482 key[0].objectid = objectid;
4483 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4486 key[1].objectid = objectid;
4487 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4488 key[1].offset = ref_objectid;
4490 sizes[0] = sizeof(struct btrfs_inode_item);
4491 sizes[1] = name_len + sizeof(*ref);
4493 path->leave_spinning = 1;
4494 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4498 inode_init_owner(inode, dir, mode);
4499 inode_set_bytes(inode, 0);
4500 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4501 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4502 struct btrfs_inode_item);
4503 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4505 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4506 struct btrfs_inode_ref);
4507 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4508 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4509 ptr = (unsigned long)(ref + 1);
4510 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4512 btrfs_mark_buffer_dirty(path->nodes[0]);
4513 btrfs_free_path(path);
4515 location = &BTRFS_I(inode)->location;
4516 location->objectid = objectid;
4517 location->offset = 0;
4518 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4520 btrfs_inherit_iflags(inode, dir);
4522 if (S_ISREG(mode)) {
4523 if (btrfs_test_opt(root, NODATASUM))
4524 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4525 if (btrfs_test_opt(root, NODATACOW) ||
4526 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4527 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4530 insert_inode_hash(inode);
4531 inode_tree_add(inode);
4533 trace_btrfs_inode_new(inode);
4534 btrfs_set_inode_last_trans(trans, inode);
4539 BTRFS_I(dir)->index_cnt--;
4540 btrfs_free_path(path);
4542 return ERR_PTR(ret);
4545 static inline u8 btrfs_inode_type(struct inode *inode)
4547 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4551 * utility function to add 'inode' into 'parent_inode' with
4552 * a give name and a given sequence number.
4553 * if 'add_backref' is true, also insert a backref from the
4554 * inode to the parent directory.
4556 int btrfs_add_link(struct btrfs_trans_handle *trans,
4557 struct inode *parent_inode, struct inode *inode,
4558 const char *name, int name_len, int add_backref, u64 index)
4561 struct btrfs_key key;
4562 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4563 u64 ino = btrfs_ino(inode);
4564 u64 parent_ino = btrfs_ino(parent_inode);
4566 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4567 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4570 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4574 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4575 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4576 key.objectid, root->root_key.objectid,
4577 parent_ino, index, name, name_len);
4578 } else if (add_backref) {
4579 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4584 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4586 btrfs_inode_type(inode), index);
4590 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4592 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4593 ret = btrfs_update_inode(trans, root, parent_inode);
4598 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4601 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4602 key.objectid, root->root_key.objectid,
4603 parent_ino, &local_index, name, name_len);
4605 } else if (add_backref) {
4609 err = btrfs_del_inode_ref(trans, root, name, name_len,
4610 ino, parent_ino, &local_index);
4615 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4616 struct inode *dir, struct dentry *dentry,
4617 struct inode *inode, int backref, u64 index)
4619 int err = btrfs_add_link(trans, dir, inode,
4620 dentry->d_name.name, dentry->d_name.len,
4627 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4628 umode_t mode, dev_t rdev)
4630 struct btrfs_trans_handle *trans;
4631 struct btrfs_root *root = BTRFS_I(dir)->root;
4632 struct inode *inode = NULL;
4636 unsigned long nr = 0;
4639 if (!new_valid_dev(rdev))
4643 * 2 for inode item and ref
4645 * 1 for xattr if selinux is on
4647 trans = btrfs_start_transaction(root, 5);
4649 return PTR_ERR(trans);
4651 err = btrfs_find_free_ino(root, &objectid);
4655 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4656 dentry->d_name.len, btrfs_ino(dir), objectid,
4658 if (IS_ERR(inode)) {
4659 err = PTR_ERR(inode);
4663 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4670 * If the active LSM wants to access the inode during
4671 * d_instantiate it needs these. Smack checks to see
4672 * if the filesystem supports xattrs by looking at the
4676 inode->i_op = &btrfs_special_inode_operations;
4677 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4681 init_special_inode(inode, inode->i_mode, rdev);
4682 btrfs_update_inode(trans, root, inode);
4683 d_instantiate(dentry, inode);
4686 nr = trans->blocks_used;
4687 btrfs_end_transaction(trans, root);
4688 btrfs_btree_balance_dirty(root, nr);
4690 inode_dec_link_count(inode);
4696 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4697 umode_t mode, struct nameidata *nd)
4699 struct btrfs_trans_handle *trans;
4700 struct btrfs_root *root = BTRFS_I(dir)->root;
4701 struct inode *inode = NULL;
4704 unsigned long nr = 0;
4709 * 2 for inode item and ref
4711 * 1 for xattr if selinux is on
4713 trans = btrfs_start_transaction(root, 5);
4715 return PTR_ERR(trans);
4717 err = btrfs_find_free_ino(root, &objectid);
4721 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4722 dentry->d_name.len, btrfs_ino(dir), objectid,
4724 if (IS_ERR(inode)) {
4725 err = PTR_ERR(inode);
4729 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4736 * If the active LSM wants to access the inode during
4737 * d_instantiate it needs these. Smack checks to see
4738 * if the filesystem supports xattrs by looking at the
4741 inode->i_fop = &btrfs_file_operations;
4742 inode->i_op = &btrfs_file_inode_operations;
4744 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4748 inode->i_mapping->a_ops = &btrfs_aops;
4749 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4750 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4751 d_instantiate(dentry, inode);
4754 nr = trans->blocks_used;
4755 btrfs_end_transaction(trans, root);
4757 inode_dec_link_count(inode);
4760 btrfs_btree_balance_dirty(root, nr);
4764 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4765 struct dentry *dentry)
4767 struct btrfs_trans_handle *trans;
4768 struct btrfs_root *root = BTRFS_I(dir)->root;
4769 struct inode *inode = old_dentry->d_inode;
4771 unsigned long nr = 0;
4775 /* do not allow sys_link's with other subvols of the same device */
4776 if (root->objectid != BTRFS_I(inode)->root->objectid)
4779 if (inode->i_nlink == ~0U)
4782 err = btrfs_set_inode_index(dir, &index);
4787 * 2 items for inode and inode ref
4788 * 2 items for dir items
4789 * 1 item for parent inode
4791 trans = btrfs_start_transaction(root, 5);
4792 if (IS_ERR(trans)) {
4793 err = PTR_ERR(trans);
4797 btrfs_inc_nlink(inode);
4798 inode->i_ctime = CURRENT_TIME;
4801 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4806 struct dentry *parent = dentry->d_parent;
4807 err = btrfs_update_inode(trans, root, inode);
4809 d_instantiate(dentry, inode);
4810 btrfs_log_new_name(trans, inode, NULL, parent);
4813 nr = trans->blocks_used;
4814 btrfs_end_transaction(trans, root);
4817 inode_dec_link_count(inode);
4820 btrfs_btree_balance_dirty(root, nr);
4824 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
4826 struct inode *inode = NULL;
4827 struct btrfs_trans_handle *trans;
4828 struct btrfs_root *root = BTRFS_I(dir)->root;
4830 int drop_on_err = 0;
4833 unsigned long nr = 1;
4836 * 2 items for inode and ref
4837 * 2 items for dir items
4838 * 1 for xattr if selinux is on
4840 trans = btrfs_start_transaction(root, 5);
4842 return PTR_ERR(trans);
4844 err = btrfs_find_free_ino(root, &objectid);
4848 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4849 dentry->d_name.len, btrfs_ino(dir), objectid,
4850 S_IFDIR | mode, &index);
4851 if (IS_ERR(inode)) {
4852 err = PTR_ERR(inode);
4858 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4862 inode->i_op = &btrfs_dir_inode_operations;
4863 inode->i_fop = &btrfs_dir_file_operations;
4865 btrfs_i_size_write(inode, 0);
4866 err = btrfs_update_inode(trans, root, inode);
4870 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4871 dentry->d_name.len, 0, index);
4875 d_instantiate(dentry, inode);
4879 nr = trans->blocks_used;
4880 btrfs_end_transaction(trans, root);
4883 btrfs_btree_balance_dirty(root, nr);
4887 /* helper for btfs_get_extent. Given an existing extent in the tree,
4888 * and an extent that you want to insert, deal with overlap and insert
4889 * the new extent into the tree.
4891 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4892 struct extent_map *existing,
4893 struct extent_map *em,
4894 u64 map_start, u64 map_len)
4898 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4899 start_diff = map_start - em->start;
4900 em->start = map_start;
4902 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4903 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4904 em->block_start += start_diff;
4905 em->block_len -= start_diff;
4907 return add_extent_mapping(em_tree, em);
4910 static noinline int uncompress_inline(struct btrfs_path *path,
4911 struct inode *inode, struct page *page,
4912 size_t pg_offset, u64 extent_offset,
4913 struct btrfs_file_extent_item *item)
4916 struct extent_buffer *leaf = path->nodes[0];
4919 unsigned long inline_size;
4923 WARN_ON(pg_offset != 0);
4924 compress_type = btrfs_file_extent_compression(leaf, item);
4925 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4926 inline_size = btrfs_file_extent_inline_item_len(leaf,
4927 btrfs_item_nr(leaf, path->slots[0]));
4928 tmp = kmalloc(inline_size, GFP_NOFS);
4931 ptr = btrfs_file_extent_inline_start(item);
4933 read_extent_buffer(leaf, tmp, ptr, inline_size);
4935 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4936 ret = btrfs_decompress(compress_type, tmp, page,
4937 extent_offset, inline_size, max_size);
4939 char *kaddr = kmap_atomic(page, KM_USER0);
4940 unsigned long copy_size = min_t(u64,
4941 PAGE_CACHE_SIZE - pg_offset,
4942 max_size - extent_offset);
4943 memset(kaddr + pg_offset, 0, copy_size);
4944 kunmap_atomic(kaddr, KM_USER0);
4951 * a bit scary, this does extent mapping from logical file offset to the disk.
4952 * the ugly parts come from merging extents from the disk with the in-ram
4953 * representation. This gets more complex because of the data=ordered code,
4954 * where the in-ram extents might be locked pending data=ordered completion.
4956 * This also copies inline extents directly into the page.
4959 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4960 size_t pg_offset, u64 start, u64 len,
4966 u64 extent_start = 0;
4968 u64 objectid = btrfs_ino(inode);
4970 struct btrfs_path *path = NULL;
4971 struct btrfs_root *root = BTRFS_I(inode)->root;
4972 struct btrfs_file_extent_item *item;
4973 struct extent_buffer *leaf;
4974 struct btrfs_key found_key;
4975 struct extent_map *em = NULL;
4976 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4977 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4978 struct btrfs_trans_handle *trans = NULL;
4982 read_lock(&em_tree->lock);
4983 em = lookup_extent_mapping(em_tree, start, len);
4985 em->bdev = root->fs_info->fs_devices->latest_bdev;
4986 read_unlock(&em_tree->lock);
4989 if (em->start > start || em->start + em->len <= start)
4990 free_extent_map(em);
4991 else if (em->block_start == EXTENT_MAP_INLINE && page)
4992 free_extent_map(em);
4996 em = alloc_extent_map();
5001 em->bdev = root->fs_info->fs_devices->latest_bdev;
5002 em->start = EXTENT_MAP_HOLE;
5003 em->orig_start = EXTENT_MAP_HOLE;
5005 em->block_len = (u64)-1;
5008 path = btrfs_alloc_path();
5014 * Chances are we'll be called again, so go ahead and do
5020 ret = btrfs_lookup_file_extent(trans, root, path,
5021 objectid, start, trans != NULL);
5028 if (path->slots[0] == 0)
5033 leaf = path->nodes[0];
5034 item = btrfs_item_ptr(leaf, path->slots[0],
5035 struct btrfs_file_extent_item);
5036 /* are we inside the extent that was found? */
5037 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5038 found_type = btrfs_key_type(&found_key);
5039 if (found_key.objectid != objectid ||
5040 found_type != BTRFS_EXTENT_DATA_KEY) {
5044 found_type = btrfs_file_extent_type(leaf, item);
5045 extent_start = found_key.offset;
5046 compress_type = btrfs_file_extent_compression(leaf, item);
5047 if (found_type == BTRFS_FILE_EXTENT_REG ||
5048 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5049 extent_end = extent_start +
5050 btrfs_file_extent_num_bytes(leaf, item);
5051 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5053 size = btrfs_file_extent_inline_len(leaf, item);
5054 extent_end = (extent_start + size + root->sectorsize - 1) &
5055 ~((u64)root->sectorsize - 1);
5058 if (start >= extent_end) {
5060 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5061 ret = btrfs_next_leaf(root, path);
5068 leaf = path->nodes[0];
5070 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5071 if (found_key.objectid != objectid ||
5072 found_key.type != BTRFS_EXTENT_DATA_KEY)
5074 if (start + len <= found_key.offset)
5077 em->len = found_key.offset - start;
5081 if (found_type == BTRFS_FILE_EXTENT_REG ||
5082 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5083 em->start = extent_start;
5084 em->len = extent_end - extent_start;
5085 em->orig_start = extent_start -
5086 btrfs_file_extent_offset(leaf, item);
5087 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5089 em->block_start = EXTENT_MAP_HOLE;
5092 if (compress_type != BTRFS_COMPRESS_NONE) {
5093 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5094 em->compress_type = compress_type;
5095 em->block_start = bytenr;
5096 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5099 bytenr += btrfs_file_extent_offset(leaf, item);
5100 em->block_start = bytenr;
5101 em->block_len = em->len;
5102 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5103 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5106 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5110 size_t extent_offset;
5113 em->block_start = EXTENT_MAP_INLINE;
5114 if (!page || create) {
5115 em->start = extent_start;
5116 em->len = extent_end - extent_start;
5120 size = btrfs_file_extent_inline_len(leaf, item);
5121 extent_offset = page_offset(page) + pg_offset - extent_start;
5122 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5123 size - extent_offset);
5124 em->start = extent_start + extent_offset;
5125 em->len = (copy_size + root->sectorsize - 1) &
5126 ~((u64)root->sectorsize - 1);
5127 em->orig_start = EXTENT_MAP_INLINE;
5128 if (compress_type) {
5129 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5130 em->compress_type = compress_type;
5132 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5133 if (create == 0 && !PageUptodate(page)) {
5134 if (btrfs_file_extent_compression(leaf, item) !=
5135 BTRFS_COMPRESS_NONE) {
5136 ret = uncompress_inline(path, inode, page,
5138 extent_offset, item);
5142 read_extent_buffer(leaf, map + pg_offset, ptr,
5144 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5145 memset(map + pg_offset + copy_size, 0,
5146 PAGE_CACHE_SIZE - pg_offset -
5151 flush_dcache_page(page);
5152 } else if (create && PageUptodate(page)) {
5156 free_extent_map(em);
5159 btrfs_release_path(path);
5160 trans = btrfs_join_transaction(root);
5163 return ERR_CAST(trans);
5167 write_extent_buffer(leaf, map + pg_offset, ptr,
5170 btrfs_mark_buffer_dirty(leaf);
5172 set_extent_uptodate(io_tree, em->start,
5173 extent_map_end(em) - 1, NULL, GFP_NOFS);
5176 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5183 em->block_start = EXTENT_MAP_HOLE;
5184 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5186 btrfs_release_path(path);
5187 if (em->start > start || extent_map_end(em) <= start) {
5188 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5189 "[%llu %llu]\n", (unsigned long long)em->start,
5190 (unsigned long long)em->len,
5191 (unsigned long long)start,
5192 (unsigned long long)len);
5198 write_lock(&em_tree->lock);
5199 ret = add_extent_mapping(em_tree, em);
5200 /* it is possible that someone inserted the extent into the tree
5201 * while we had the lock dropped. It is also possible that
5202 * an overlapping map exists in the tree
5204 if (ret == -EEXIST) {
5205 struct extent_map *existing;
5209 existing = lookup_extent_mapping(em_tree, start, len);
5210 if (existing && (existing->start > start ||
5211 existing->start + existing->len <= start)) {
5212 free_extent_map(existing);
5216 existing = lookup_extent_mapping(em_tree, em->start,
5219 err = merge_extent_mapping(em_tree, existing,
5222 free_extent_map(existing);
5224 free_extent_map(em);
5229 free_extent_map(em);
5233 free_extent_map(em);
5238 write_unlock(&em_tree->lock);
5241 trace_btrfs_get_extent(root, em);
5244 btrfs_free_path(path);
5246 ret = btrfs_end_transaction(trans, root);
5251 free_extent_map(em);
5252 return ERR_PTR(err);
5257 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5258 size_t pg_offset, u64 start, u64 len,
5261 struct extent_map *em;
5262 struct extent_map *hole_em = NULL;
5263 u64 range_start = start;
5269 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5274 * if our em maps to a hole, there might
5275 * actually be delalloc bytes behind it
5277 if (em->block_start != EXTENT_MAP_HOLE)
5283 /* check to see if we've wrapped (len == -1 or similar) */
5292 /* ok, we didn't find anything, lets look for delalloc */
5293 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5294 end, len, EXTENT_DELALLOC, 1);
5295 found_end = range_start + found;
5296 if (found_end < range_start)
5297 found_end = (u64)-1;
5300 * we didn't find anything useful, return
5301 * the original results from get_extent()
5303 if (range_start > end || found_end <= start) {
5309 /* adjust the range_start to make sure it doesn't
5310 * go backwards from the start they passed in
5312 range_start = max(start,range_start);
5313 found = found_end - range_start;
5316 u64 hole_start = start;
5319 em = alloc_extent_map();
5325 * when btrfs_get_extent can't find anything it
5326 * returns one huge hole
5328 * make sure what it found really fits our range, and
5329 * adjust to make sure it is based on the start from
5333 u64 calc_end = extent_map_end(hole_em);
5335 if (calc_end <= start || (hole_em->start > end)) {
5336 free_extent_map(hole_em);
5339 hole_start = max(hole_em->start, start);
5340 hole_len = calc_end - hole_start;
5344 if (hole_em && range_start > hole_start) {
5345 /* our hole starts before our delalloc, so we
5346 * have to return just the parts of the hole
5347 * that go until the delalloc starts
5349 em->len = min(hole_len,
5350 range_start - hole_start);
5351 em->start = hole_start;
5352 em->orig_start = hole_start;
5354 * don't adjust block start at all,
5355 * it is fixed at EXTENT_MAP_HOLE
5357 em->block_start = hole_em->block_start;
5358 em->block_len = hole_len;
5360 em->start = range_start;
5362 em->orig_start = range_start;
5363 em->block_start = EXTENT_MAP_DELALLOC;
5364 em->block_len = found;
5366 } else if (hole_em) {
5371 free_extent_map(hole_em);
5373 free_extent_map(em);
5374 return ERR_PTR(err);
5379 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5380 struct extent_map *em,
5383 struct btrfs_root *root = BTRFS_I(inode)->root;
5384 struct btrfs_trans_handle *trans;
5385 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5386 struct btrfs_key ins;
5389 bool insert = false;
5392 * Ok if the extent map we looked up is a hole and is for the exact
5393 * range we want, there is no reason to allocate a new one, however if
5394 * it is not right then we need to free this one and drop the cache for
5397 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5399 free_extent_map(em);
5402 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5405 trans = btrfs_join_transaction(root);
5407 return ERR_CAST(trans);
5409 if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5410 btrfs_add_inode_defrag(trans, inode);
5412 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5414 alloc_hint = get_extent_allocation_hint(inode, start, len);
5415 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5416 alloc_hint, &ins, 1);
5423 em = alloc_extent_map();
5425 em = ERR_PTR(-ENOMEM);
5431 em->orig_start = em->start;
5432 em->len = ins.offset;
5434 em->block_start = ins.objectid;
5435 em->block_len = ins.offset;
5436 em->bdev = root->fs_info->fs_devices->latest_bdev;
5439 * We need to do this because if we're using the original em we searched
5440 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5443 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5446 write_lock(&em_tree->lock);
5447 ret = add_extent_mapping(em_tree, em);
5448 write_unlock(&em_tree->lock);
5451 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5454 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5455 ins.offset, ins.offset, 0);
5457 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5461 btrfs_end_transaction(trans, root);
5466 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5467 * block must be cow'd
5469 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5470 struct inode *inode, u64 offset, u64 len)
5472 struct btrfs_path *path;
5474 struct extent_buffer *leaf;
5475 struct btrfs_root *root = BTRFS_I(inode)->root;
5476 struct btrfs_file_extent_item *fi;
5477 struct btrfs_key key;
5485 path = btrfs_alloc_path();
5489 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5494 slot = path->slots[0];
5497 /* can't find the item, must cow */
5504 leaf = path->nodes[0];
5505 btrfs_item_key_to_cpu(leaf, &key, slot);
5506 if (key.objectid != btrfs_ino(inode) ||
5507 key.type != BTRFS_EXTENT_DATA_KEY) {
5508 /* not our file or wrong item type, must cow */
5512 if (key.offset > offset) {
5513 /* Wrong offset, must cow */
5517 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5518 found_type = btrfs_file_extent_type(leaf, fi);
5519 if (found_type != BTRFS_FILE_EXTENT_REG &&
5520 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5521 /* not a regular extent, must cow */
5524 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5525 backref_offset = btrfs_file_extent_offset(leaf, fi);
5527 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5528 if (extent_end < offset + len) {
5529 /* extent doesn't include our full range, must cow */
5533 if (btrfs_extent_readonly(root, disk_bytenr))
5537 * look for other files referencing this extent, if we
5538 * find any we must cow
5540 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5541 key.offset - backref_offset, disk_bytenr))
5545 * adjust disk_bytenr and num_bytes to cover just the bytes
5546 * in this extent we are about to write. If there
5547 * are any csums in that range we have to cow in order
5548 * to keep the csums correct
5550 disk_bytenr += backref_offset;
5551 disk_bytenr += offset - key.offset;
5552 num_bytes = min(offset + len, extent_end) - offset;
5553 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5556 * all of the above have passed, it is safe to overwrite this extent
5561 btrfs_free_path(path);
5565 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5566 struct buffer_head *bh_result, int create)
5568 struct extent_map *em;
5569 struct btrfs_root *root = BTRFS_I(inode)->root;
5570 u64 start = iblock << inode->i_blkbits;
5571 u64 len = bh_result->b_size;
5572 struct btrfs_trans_handle *trans;
5574 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5579 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5580 * io. INLINE is special, and we could probably kludge it in here, but
5581 * it's still buffered so for safety lets just fall back to the generic
5584 * For COMPRESSED we _have_ to read the entire extent in so we can
5585 * decompress it, so there will be buffering required no matter what we
5586 * do, so go ahead and fallback to buffered.
5588 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5589 * to buffered IO. Don't blame me, this is the price we pay for using
5592 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5593 em->block_start == EXTENT_MAP_INLINE) {
5594 free_extent_map(em);
5598 /* Just a good old fashioned hole, return */
5599 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5600 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5601 free_extent_map(em);
5602 /* DIO will do one hole at a time, so just unlock a sector */
5603 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5604 start + root->sectorsize - 1, GFP_NOFS);
5609 * We don't allocate a new extent in the following cases
5611 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5613 * 2) The extent is marked as PREALLOC. We're good to go here and can
5614 * just use the extent.
5618 len = em->len - (start - em->start);
5622 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5623 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5624 em->block_start != EXTENT_MAP_HOLE)) {
5629 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5630 type = BTRFS_ORDERED_PREALLOC;
5632 type = BTRFS_ORDERED_NOCOW;
5633 len = min(len, em->len - (start - em->start));
5634 block_start = em->block_start + (start - em->start);
5637 * we're not going to log anything, but we do need
5638 * to make sure the current transaction stays open
5639 * while we look for nocow cross refs
5641 trans = btrfs_join_transaction(root);
5645 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5646 ret = btrfs_add_ordered_extent_dio(inode, start,
5647 block_start, len, len, type);
5648 btrfs_end_transaction(trans, root);
5650 free_extent_map(em);
5655 btrfs_end_transaction(trans, root);
5659 * this will cow the extent, reset the len in case we changed
5662 len = bh_result->b_size;
5663 em = btrfs_new_extent_direct(inode, em, start, len);
5666 len = min(len, em->len - (start - em->start));
5668 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5669 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5672 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5674 bh_result->b_size = len;
5675 bh_result->b_bdev = em->bdev;
5676 set_buffer_mapped(bh_result);
5677 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5678 set_buffer_new(bh_result);
5680 free_extent_map(em);
5685 struct btrfs_dio_private {
5686 struct inode *inode;
5693 /* number of bios pending for this dio */
5694 atomic_t pending_bios;
5699 struct bio *orig_bio;
5702 static void btrfs_endio_direct_read(struct bio *bio, int err)
5704 struct btrfs_dio_private *dip = bio->bi_private;
5705 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5706 struct bio_vec *bvec = bio->bi_io_vec;
5707 struct inode *inode = dip->inode;
5708 struct btrfs_root *root = BTRFS_I(inode)->root;
5710 u32 *private = dip->csums;
5712 start = dip->logical_offset;
5714 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5715 struct page *page = bvec->bv_page;
5718 unsigned long flags;
5720 local_irq_save(flags);
5721 kaddr = kmap_atomic(page, KM_IRQ0);
5722 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5723 csum, bvec->bv_len);
5724 btrfs_csum_final(csum, (char *)&csum);
5725 kunmap_atomic(kaddr, KM_IRQ0);
5726 local_irq_restore(flags);
5728 flush_dcache_page(bvec->bv_page);
5729 if (csum != *private) {
5730 printk(KERN_ERR "btrfs csum failed ino %llu off"
5731 " %llu csum %u private %u\n",
5732 (unsigned long long)btrfs_ino(inode),
5733 (unsigned long long)start,
5739 start += bvec->bv_len;
5742 } while (bvec <= bvec_end);
5744 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5745 dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5746 bio->bi_private = dip->private;
5751 /* If we had a csum failure make sure to clear the uptodate flag */
5753 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5754 dio_end_io(bio, err);
5757 static void btrfs_endio_direct_write(struct bio *bio, int err)
5759 struct btrfs_dio_private *dip = bio->bi_private;
5760 struct inode *inode = dip->inode;
5761 struct btrfs_root *root = BTRFS_I(inode)->root;
5762 struct btrfs_trans_handle *trans;
5763 struct btrfs_ordered_extent *ordered = NULL;
5764 struct extent_state *cached_state = NULL;
5765 u64 ordered_offset = dip->logical_offset;
5766 u64 ordered_bytes = dip->bytes;
5772 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5780 trans = btrfs_join_transaction(root);
5781 if (IS_ERR(trans)) {
5785 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5787 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5788 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5790 err = btrfs_update_inode_fallback(trans, root, inode);
5794 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5795 ordered->file_offset + ordered->len - 1, 0,
5796 &cached_state, GFP_NOFS);
5798 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5799 ret = btrfs_mark_extent_written(trans, inode,
5800 ordered->file_offset,
5801 ordered->file_offset +
5808 ret = insert_reserved_file_extent(trans, inode,
5809 ordered->file_offset,
5815 BTRFS_FILE_EXTENT_REG);
5816 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5817 ordered->file_offset, ordered->len);
5825 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5826 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5827 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
5828 btrfs_update_inode_fallback(trans, root, inode);
5831 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5832 ordered->file_offset + ordered->len - 1,
5833 &cached_state, GFP_NOFS);
5835 btrfs_delalloc_release_metadata(inode, ordered->len);
5836 btrfs_end_transaction(trans, root);
5837 ordered_offset = ordered->file_offset + ordered->len;
5838 btrfs_put_ordered_extent(ordered);
5839 btrfs_put_ordered_extent(ordered);
5843 * our bio might span multiple ordered extents. If we haven't
5844 * completed the accounting for the whole dio, go back and try again
5846 if (ordered_offset < dip->logical_offset + dip->bytes) {
5847 ordered_bytes = dip->logical_offset + dip->bytes -
5852 bio->bi_private = dip->private;
5857 /* If we had an error make sure to clear the uptodate flag */
5859 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5860 dio_end_io(bio, err);
5863 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5864 struct bio *bio, int mirror_num,
5865 unsigned long bio_flags, u64 offset)
5868 struct btrfs_root *root = BTRFS_I(inode)->root;
5869 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5874 static void btrfs_end_dio_bio(struct bio *bio, int err)
5876 struct btrfs_dio_private *dip = bio->bi_private;
5879 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5880 "sector %#Lx len %u err no %d\n",
5881 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5882 (unsigned long long)bio->bi_sector, bio->bi_size, err);
5886 * before atomic variable goto zero, we must make sure
5887 * dip->errors is perceived to be set.
5889 smp_mb__before_atomic_dec();
5892 /* if there are more bios still pending for this dio, just exit */
5893 if (!atomic_dec_and_test(&dip->pending_bios))
5897 bio_io_error(dip->orig_bio);
5899 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5900 bio_endio(dip->orig_bio, 0);
5906 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5907 u64 first_sector, gfp_t gfp_flags)
5909 int nr_vecs = bio_get_nr_vecs(bdev);
5910 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5913 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5914 int rw, u64 file_offset, int skip_sum,
5915 u32 *csums, int async_submit)
5917 int write = rw & REQ_WRITE;
5918 struct btrfs_root *root = BTRFS_I(inode)->root;
5922 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5929 if (write && async_submit) {
5930 ret = btrfs_wq_submit_bio(root->fs_info,
5931 inode, rw, bio, 0, 0,
5933 __btrfs_submit_bio_start_direct_io,
5934 __btrfs_submit_bio_done);
5938 * If we aren't doing async submit, calculate the csum of the
5941 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5944 } else if (!skip_sum) {
5945 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5946 file_offset, csums);
5952 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5958 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5961 struct inode *inode = dip->inode;
5962 struct btrfs_root *root = BTRFS_I(inode)->root;
5963 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5965 struct bio *orig_bio = dip->orig_bio;
5966 struct bio_vec *bvec = orig_bio->bi_io_vec;
5967 u64 start_sector = orig_bio->bi_sector;
5968 u64 file_offset = dip->logical_offset;
5972 u32 *csums = dip->csums;
5974 int async_submit = 0;
5975 int write = rw & REQ_WRITE;
5977 map_length = orig_bio->bi_size;
5978 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5979 &map_length, NULL, 0);
5985 if (map_length >= orig_bio->bi_size) {
5991 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5994 bio->bi_private = dip;
5995 bio->bi_end_io = btrfs_end_dio_bio;
5996 atomic_inc(&dip->pending_bios);
5998 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5999 if (unlikely(map_length < submit_len + bvec->bv_len ||
6000 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6001 bvec->bv_offset) < bvec->bv_len)) {
6003 * inc the count before we submit the bio so
6004 * we know the end IO handler won't happen before
6005 * we inc the count. Otherwise, the dip might get freed
6006 * before we're done setting it up
6008 atomic_inc(&dip->pending_bios);
6009 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6010 file_offset, skip_sum,
6011 csums, async_submit);
6014 atomic_dec(&dip->pending_bios);
6018 /* Write's use the ordered csums */
6019 if (!write && !skip_sum)
6020 csums = csums + nr_pages;
6021 start_sector += submit_len >> 9;
6022 file_offset += submit_len;
6027 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6028 start_sector, GFP_NOFS);
6031 bio->bi_private = dip;
6032 bio->bi_end_io = btrfs_end_dio_bio;
6034 map_length = orig_bio->bi_size;
6035 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6036 &map_length, NULL, 0);
6042 submit_len += bvec->bv_len;
6049 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6050 csums, async_submit);
6058 * before atomic variable goto zero, we must
6059 * make sure dip->errors is perceived to be set.
6061 smp_mb__before_atomic_dec();
6062 if (atomic_dec_and_test(&dip->pending_bios))
6063 bio_io_error(dip->orig_bio);
6065 /* bio_end_io() will handle error, so we needn't return it */
6069 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6072 struct btrfs_root *root = BTRFS_I(inode)->root;
6073 struct btrfs_dio_private *dip;
6074 struct bio_vec *bvec = bio->bi_io_vec;
6076 int write = rw & REQ_WRITE;
6079 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6081 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6088 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6089 if (!write && !skip_sum) {
6090 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6098 dip->private = bio->bi_private;
6100 dip->logical_offset = file_offset;
6104 dip->bytes += bvec->bv_len;
6106 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6108 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6109 bio->bi_private = dip;
6111 dip->orig_bio = bio;
6112 atomic_set(&dip->pending_bios, 0);
6115 bio->bi_end_io = btrfs_endio_direct_write;
6117 bio->bi_end_io = btrfs_endio_direct_read;
6119 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6124 * If this is a write, we need to clean up the reserved space and kill
6125 * the ordered extent.
6128 struct btrfs_ordered_extent *ordered;
6129 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6130 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6131 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6132 btrfs_free_reserved_extent(root, ordered->start,
6134 btrfs_put_ordered_extent(ordered);
6135 btrfs_put_ordered_extent(ordered);
6137 bio_endio(bio, ret);
6140 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6141 const struct iovec *iov, loff_t offset,
6142 unsigned long nr_segs)
6148 unsigned blocksize_mask = root->sectorsize - 1;
6149 ssize_t retval = -EINVAL;
6150 loff_t end = offset;
6152 if (offset & blocksize_mask)
6155 /* Check the memory alignment. Blocks cannot straddle pages */
6156 for (seg = 0; seg < nr_segs; seg++) {
6157 addr = (unsigned long)iov[seg].iov_base;
6158 size = iov[seg].iov_len;
6160 if ((addr & blocksize_mask) || (size & blocksize_mask))
6163 /* If this is a write we don't need to check anymore */
6168 * Check to make sure we don't have duplicate iov_base's in this
6169 * iovec, if so return EINVAL, otherwise we'll get csum errors
6170 * when reading back.
6172 for (i = seg + 1; i < nr_segs; i++) {
6173 if (iov[seg].iov_base == iov[i].iov_base)
6181 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6182 const struct iovec *iov, loff_t offset,
6183 unsigned long nr_segs)
6185 struct file *file = iocb->ki_filp;
6186 struct inode *inode = file->f_mapping->host;
6187 struct btrfs_ordered_extent *ordered;
6188 struct extent_state *cached_state = NULL;
6189 u64 lockstart, lockend;
6191 int writing = rw & WRITE;
6193 size_t count = iov_length(iov, nr_segs);
6195 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6201 lockend = offset + count - 1;
6204 ret = btrfs_delalloc_reserve_space(inode, count);
6210 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6211 0, &cached_state, GFP_NOFS);
6213 * We're concerned with the entire range that we're going to be
6214 * doing DIO to, so we need to make sure theres no ordered
6215 * extents in this range.
6217 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6218 lockend - lockstart + 1);
6221 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6222 &cached_state, GFP_NOFS);
6223 btrfs_start_ordered_extent(inode, ordered, 1);
6224 btrfs_put_ordered_extent(ordered);
6229 * we don't use btrfs_set_extent_delalloc because we don't want
6230 * the dirty or uptodate bits
6233 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6234 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6235 EXTENT_DELALLOC, 0, NULL, &cached_state,
6238 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6239 lockend, EXTENT_LOCKED | write_bits,
6240 1, 0, &cached_state, GFP_NOFS);
6245 free_extent_state(cached_state);
6246 cached_state = NULL;
6248 ret = __blockdev_direct_IO(rw, iocb, inode,
6249 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6250 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6251 btrfs_submit_direct, 0);
6253 if (ret < 0 && ret != -EIOCBQUEUED) {
6254 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6255 offset + iov_length(iov, nr_segs) - 1,
6256 EXTENT_LOCKED | write_bits, 1, 0,
6257 &cached_state, GFP_NOFS);
6258 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6260 * We're falling back to buffered, unlock the section we didn't
6263 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6264 offset + iov_length(iov, nr_segs) - 1,
6265 EXTENT_LOCKED | write_bits, 1, 0,
6266 &cached_state, GFP_NOFS);
6269 free_extent_state(cached_state);
6273 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6274 __u64 start, __u64 len)
6276 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6279 int btrfs_readpage(struct file *file, struct page *page)
6281 struct extent_io_tree *tree;
6282 tree = &BTRFS_I(page->mapping->host)->io_tree;
6283 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6286 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6288 struct extent_io_tree *tree;
6291 if (current->flags & PF_MEMALLOC) {
6292 redirty_page_for_writepage(wbc, page);
6296 tree = &BTRFS_I(page->mapping->host)->io_tree;
6297 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6300 int btrfs_writepages(struct address_space *mapping,
6301 struct writeback_control *wbc)
6303 struct extent_io_tree *tree;
6305 tree = &BTRFS_I(mapping->host)->io_tree;
6306 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6310 btrfs_readpages(struct file *file, struct address_space *mapping,
6311 struct list_head *pages, unsigned nr_pages)
6313 struct extent_io_tree *tree;
6314 tree = &BTRFS_I(mapping->host)->io_tree;
6315 return extent_readpages(tree, mapping, pages, nr_pages,
6318 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6320 struct extent_io_tree *tree;
6321 struct extent_map_tree *map;
6324 tree = &BTRFS_I(page->mapping->host)->io_tree;
6325 map = &BTRFS_I(page->mapping->host)->extent_tree;
6326 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6328 ClearPagePrivate(page);
6329 set_page_private(page, 0);
6330 page_cache_release(page);
6335 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6337 if (PageWriteback(page) || PageDirty(page))
6339 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6342 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6344 struct extent_io_tree *tree;
6345 struct btrfs_ordered_extent *ordered;
6346 struct extent_state *cached_state = NULL;
6347 u64 page_start = page_offset(page);
6348 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6352 * we have the page locked, so new writeback can't start,
6353 * and the dirty bit won't be cleared while we are here.
6355 * Wait for IO on this page so that we can safely clear
6356 * the PagePrivate2 bit and do ordered accounting
6358 wait_on_page_writeback(page);
6360 tree = &BTRFS_I(page->mapping->host)->io_tree;
6362 btrfs_releasepage(page, GFP_NOFS);
6365 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6367 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6371 * IO on this page will never be started, so we need
6372 * to account for any ordered extents now
6374 clear_extent_bit(tree, page_start, page_end,
6375 EXTENT_DIRTY | EXTENT_DELALLOC |
6376 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6377 &cached_state, GFP_NOFS);
6379 * whoever cleared the private bit is responsible
6380 * for the finish_ordered_io
6382 if (TestClearPagePrivate2(page)) {
6383 btrfs_finish_ordered_io(page->mapping->host,
6384 page_start, page_end);
6386 btrfs_put_ordered_extent(ordered);
6387 cached_state = NULL;
6388 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6391 clear_extent_bit(tree, page_start, page_end,
6392 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6393 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6394 __btrfs_releasepage(page, GFP_NOFS);
6396 ClearPageChecked(page);
6397 if (PagePrivate(page)) {
6398 ClearPagePrivate(page);
6399 set_page_private(page, 0);
6400 page_cache_release(page);
6405 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6406 * called from a page fault handler when a page is first dirtied. Hence we must
6407 * be careful to check for EOF conditions here. We set the page up correctly
6408 * for a written page which means we get ENOSPC checking when writing into
6409 * holes and correct delalloc and unwritten extent mapping on filesystems that
6410 * support these features.
6412 * We are not allowed to take the i_mutex here so we have to play games to
6413 * protect against truncate races as the page could now be beyond EOF. Because
6414 * vmtruncate() writes the inode size before removing pages, once we have the
6415 * page lock we can determine safely if the page is beyond EOF. If it is not
6416 * beyond EOF, then the page is guaranteed safe against truncation until we
6419 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6421 struct page *page = vmf->page;
6422 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6423 struct btrfs_root *root = BTRFS_I(inode)->root;
6424 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6425 struct btrfs_ordered_extent *ordered;
6426 struct extent_state *cached_state = NULL;
6428 unsigned long zero_start;
6435 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6437 ret = btrfs_update_time(vma->vm_file);
6443 else /* -ENOSPC, -EIO, etc */
6444 ret = VM_FAULT_SIGBUS;
6450 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6453 size = i_size_read(inode);
6454 page_start = page_offset(page);
6455 page_end = page_start + PAGE_CACHE_SIZE - 1;
6457 if ((page->mapping != inode->i_mapping) ||
6458 (page_start >= size)) {
6459 /* page got truncated out from underneath us */
6462 wait_on_page_writeback(page);
6464 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6466 set_page_extent_mapped(page);
6469 * we can't set the delalloc bits if there are pending ordered
6470 * extents. Drop our locks and wait for them to finish
6472 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6474 unlock_extent_cached(io_tree, page_start, page_end,
6475 &cached_state, GFP_NOFS);
6477 btrfs_start_ordered_extent(inode, ordered, 1);
6478 btrfs_put_ordered_extent(ordered);
6483 * XXX - page_mkwrite gets called every time the page is dirtied, even
6484 * if it was already dirty, so for space accounting reasons we need to
6485 * clear any delalloc bits for the range we are fixing to save. There
6486 * is probably a better way to do this, but for now keep consistent with
6487 * prepare_pages in the normal write path.
6489 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6490 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6491 0, 0, &cached_state, GFP_NOFS);
6493 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6496 unlock_extent_cached(io_tree, page_start, page_end,
6497 &cached_state, GFP_NOFS);
6498 ret = VM_FAULT_SIGBUS;
6503 /* page is wholly or partially inside EOF */
6504 if (page_start + PAGE_CACHE_SIZE > size)
6505 zero_start = size & ~PAGE_CACHE_MASK;
6507 zero_start = PAGE_CACHE_SIZE;
6509 if (zero_start != PAGE_CACHE_SIZE) {
6511 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6512 flush_dcache_page(page);
6515 ClearPageChecked(page);
6516 set_page_dirty(page);
6517 SetPageUptodate(page);
6519 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6520 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6522 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6526 return VM_FAULT_LOCKED;
6529 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6534 static int btrfs_truncate(struct inode *inode)
6536 struct btrfs_root *root = BTRFS_I(inode)->root;
6537 struct btrfs_block_rsv *rsv;
6540 struct btrfs_trans_handle *trans;
6542 u64 mask = root->sectorsize - 1;
6543 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6545 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6549 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6550 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6553 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6554 * 3 things going on here
6556 * 1) We need to reserve space for our orphan item and the space to
6557 * delete our orphan item. Lord knows we don't want to have a dangling
6558 * orphan item because we didn't reserve space to remove it.
6560 * 2) We need to reserve space to update our inode.
6562 * 3) We need to have something to cache all the space that is going to
6563 * be free'd up by the truncate operation, but also have some slack
6564 * space reserved in case it uses space during the truncate (thank you
6565 * very much snapshotting).
6567 * And we need these to all be seperate. The fact is we can use alot of
6568 * space doing the truncate, and we have no earthly idea how much space
6569 * we will use, so we need the truncate reservation to be seperate so it
6570 * doesn't end up using space reserved for updating the inode or
6571 * removing the orphan item. We also need to be able to stop the
6572 * transaction and start a new one, which means we need to be able to
6573 * update the inode several times, and we have no idea of knowing how
6574 * many times that will be, so we can't just reserve 1 item for the
6575 * entirety of the opration, so that has to be done seperately as well.
6576 * Then there is the orphan item, which does indeed need to be held on
6577 * to for the whole operation, and we need nobody to touch this reserved
6578 * space except the orphan code.
6580 * So that leaves us with
6582 * 1) root->orphan_block_rsv - for the orphan deletion.
6583 * 2) rsv - for the truncate reservation, which we will steal from the
6584 * transaction reservation.
6585 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6586 * updating the inode.
6588 rsv = btrfs_alloc_block_rsv(root);
6591 rsv->size = min_size;
6594 * 1 for the truncate slack space
6595 * 1 for the orphan item we're going to add
6596 * 1 for the orphan item deletion
6597 * 1 for updating the inode.
6599 trans = btrfs_start_transaction(root, 4);
6600 if (IS_ERR(trans)) {
6601 err = PTR_ERR(trans);
6605 /* Migrate the slack space for the truncate to our reserve */
6606 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6610 ret = btrfs_orphan_add(trans, inode);
6612 btrfs_end_transaction(trans, root);
6617 * setattr is responsible for setting the ordered_data_close flag,
6618 * but that is only tested during the last file release. That
6619 * could happen well after the next commit, leaving a great big
6620 * window where new writes may get lost if someone chooses to write
6621 * to this file after truncating to zero
6623 * The inode doesn't have any dirty data here, and so if we commit
6624 * this is a noop. If someone immediately starts writing to the inode
6625 * it is very likely we'll catch some of their writes in this
6626 * transaction, and the commit will find this file on the ordered
6627 * data list with good things to send down.
6629 * This is a best effort solution, there is still a window where
6630 * using truncate to replace the contents of the file will
6631 * end up with a zero length file after a crash.
6633 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6634 btrfs_add_ordered_operation(trans, root, inode);
6637 ret = btrfs_block_rsv_refill(root, rsv, min_size);
6640 * This can only happen with the original transaction we
6641 * started above, every other time we shouldn't have a
6642 * transaction started yet.
6651 /* Just need the 1 for updating the inode */
6652 trans = btrfs_start_transaction(root, 1);
6653 if (IS_ERR(trans)) {
6654 ret = err = PTR_ERR(trans);
6660 trans->block_rsv = rsv;
6662 ret = btrfs_truncate_inode_items(trans, root, inode,
6664 BTRFS_EXTENT_DATA_KEY);
6665 if (ret != -EAGAIN) {
6670 trans->block_rsv = &root->fs_info->trans_block_rsv;
6671 ret = btrfs_update_inode(trans, root, inode);
6677 nr = trans->blocks_used;
6678 btrfs_end_transaction(trans, root);
6680 btrfs_btree_balance_dirty(root, nr);
6683 if (ret == 0 && inode->i_nlink > 0) {
6684 trans->block_rsv = root->orphan_block_rsv;
6685 ret = btrfs_orphan_del(trans, inode);
6688 } else if (ret && inode->i_nlink > 0) {
6690 * Failed to do the truncate, remove us from the in memory
6693 ret = btrfs_orphan_del(NULL, inode);
6697 trans->block_rsv = &root->fs_info->trans_block_rsv;
6698 ret = btrfs_update_inode(trans, root, inode);
6702 nr = trans->blocks_used;
6703 ret = btrfs_end_transaction(trans, root);
6704 btrfs_btree_balance_dirty(root, nr);
6708 btrfs_free_block_rsv(root, rsv);
6717 * create a new subvolume directory/inode (helper for the ioctl).
6719 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6720 struct btrfs_root *new_root, u64 new_dirid)
6722 struct inode *inode;
6726 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
6727 new_dirid, new_dirid,
6728 S_IFDIR | (~current_umask() & S_IRWXUGO),
6731 return PTR_ERR(inode);
6732 inode->i_op = &btrfs_dir_inode_operations;
6733 inode->i_fop = &btrfs_dir_file_operations;
6735 set_nlink(inode, 1);
6736 btrfs_i_size_write(inode, 0);
6738 err = btrfs_update_inode(trans, new_root, inode);
6745 struct inode *btrfs_alloc_inode(struct super_block *sb)
6747 struct btrfs_inode *ei;
6748 struct inode *inode;
6750 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6755 ei->space_info = NULL;
6759 ei->last_sub_trans = 0;
6760 ei->logged_trans = 0;
6761 ei->delalloc_bytes = 0;
6762 ei->disk_i_size = 0;
6765 ei->index_cnt = (u64)-1;
6766 ei->last_unlink_trans = 0;
6768 spin_lock_init(&ei->lock);
6769 ei->outstanding_extents = 0;
6770 ei->reserved_extents = 0;
6772 ei->ordered_data_close = 0;
6773 ei->orphan_meta_reserved = 0;
6774 ei->dummy_inode = 0;
6776 ei->delalloc_meta_reserved = 0;
6777 ei->force_compress = BTRFS_COMPRESS_NONE;
6779 ei->delayed_node = NULL;
6781 inode = &ei->vfs_inode;
6782 extent_map_tree_init(&ei->extent_tree);
6783 extent_io_tree_init(&ei->io_tree, &inode->i_data);
6784 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6785 mutex_init(&ei->log_mutex);
6786 mutex_init(&ei->delalloc_mutex);
6787 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6788 INIT_LIST_HEAD(&ei->i_orphan);
6789 INIT_LIST_HEAD(&ei->delalloc_inodes);
6790 INIT_LIST_HEAD(&ei->ordered_operations);
6791 RB_CLEAR_NODE(&ei->rb_node);
6796 static void btrfs_i_callback(struct rcu_head *head)
6798 struct inode *inode = container_of(head, struct inode, i_rcu);
6799 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6802 void btrfs_destroy_inode(struct inode *inode)
6804 struct btrfs_ordered_extent *ordered;
6805 struct btrfs_root *root = BTRFS_I(inode)->root;
6807 WARN_ON(!list_empty(&inode->i_dentry));
6808 WARN_ON(inode->i_data.nrpages);
6809 WARN_ON(BTRFS_I(inode)->outstanding_extents);
6810 WARN_ON(BTRFS_I(inode)->reserved_extents);
6811 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6812 WARN_ON(BTRFS_I(inode)->csum_bytes);
6815 * This can happen where we create an inode, but somebody else also
6816 * created the same inode and we need to destroy the one we already
6823 * Make sure we're properly removed from the ordered operation
6827 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6828 spin_lock(&root->fs_info->ordered_extent_lock);
6829 list_del_init(&BTRFS_I(inode)->ordered_operations);
6830 spin_unlock(&root->fs_info->ordered_extent_lock);
6833 spin_lock(&root->orphan_lock);
6834 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6835 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6836 (unsigned long long)btrfs_ino(inode));
6837 list_del_init(&BTRFS_I(inode)->i_orphan);
6839 spin_unlock(&root->orphan_lock);
6842 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6846 printk(KERN_ERR "btrfs found ordered "
6847 "extent %llu %llu on inode cleanup\n",
6848 (unsigned long long)ordered->file_offset,
6849 (unsigned long long)ordered->len);
6850 btrfs_remove_ordered_extent(inode, ordered);
6851 btrfs_put_ordered_extent(ordered);
6852 btrfs_put_ordered_extent(ordered);
6855 inode_tree_del(inode);
6856 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6858 btrfs_remove_delayed_node(inode);
6859 call_rcu(&inode->i_rcu, btrfs_i_callback);
6862 int btrfs_drop_inode(struct inode *inode)
6864 struct btrfs_root *root = BTRFS_I(inode)->root;
6866 if (btrfs_root_refs(&root->root_item) == 0 &&
6867 !btrfs_is_free_space_inode(root, inode))
6870 return generic_drop_inode(inode);
6873 static void init_once(void *foo)
6875 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6877 inode_init_once(&ei->vfs_inode);
6880 void btrfs_destroy_cachep(void)
6882 if (btrfs_inode_cachep)
6883 kmem_cache_destroy(btrfs_inode_cachep);
6884 if (btrfs_trans_handle_cachep)
6885 kmem_cache_destroy(btrfs_trans_handle_cachep);
6886 if (btrfs_transaction_cachep)
6887 kmem_cache_destroy(btrfs_transaction_cachep);
6888 if (btrfs_path_cachep)
6889 kmem_cache_destroy(btrfs_path_cachep);
6890 if (btrfs_free_space_cachep)
6891 kmem_cache_destroy(btrfs_free_space_cachep);
6894 int btrfs_init_cachep(void)
6896 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6897 sizeof(struct btrfs_inode), 0,
6898 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6899 if (!btrfs_inode_cachep)
6902 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6903 sizeof(struct btrfs_trans_handle), 0,
6904 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6905 if (!btrfs_trans_handle_cachep)
6908 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6909 sizeof(struct btrfs_transaction), 0,
6910 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6911 if (!btrfs_transaction_cachep)
6914 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6915 sizeof(struct btrfs_path), 0,
6916 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6917 if (!btrfs_path_cachep)
6920 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6921 sizeof(struct btrfs_free_space), 0,
6922 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6923 if (!btrfs_free_space_cachep)
6928 btrfs_destroy_cachep();
6932 static int btrfs_getattr(struct vfsmount *mnt,
6933 struct dentry *dentry, struct kstat *stat)
6935 struct inode *inode = dentry->d_inode;
6936 u32 blocksize = inode->i_sb->s_blocksize;
6938 generic_fillattr(inode, stat);
6939 stat->dev = BTRFS_I(inode)->root->anon_dev;
6940 stat->blksize = PAGE_CACHE_SIZE;
6941 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
6942 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
6947 * If a file is moved, it will inherit the cow and compression flags of the new
6950 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6952 struct btrfs_inode *b_dir = BTRFS_I(dir);
6953 struct btrfs_inode *b_inode = BTRFS_I(inode);
6955 if (b_dir->flags & BTRFS_INODE_NODATACOW)
6956 b_inode->flags |= BTRFS_INODE_NODATACOW;
6958 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6960 if (b_dir->flags & BTRFS_INODE_COMPRESS)
6961 b_inode->flags |= BTRFS_INODE_COMPRESS;
6963 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6966 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6967 struct inode *new_dir, struct dentry *new_dentry)
6969 struct btrfs_trans_handle *trans;
6970 struct btrfs_root *root = BTRFS_I(old_dir)->root;
6971 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6972 struct inode *new_inode = new_dentry->d_inode;
6973 struct inode *old_inode = old_dentry->d_inode;
6974 struct timespec ctime = CURRENT_TIME;
6978 u64 old_ino = btrfs_ino(old_inode);
6980 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6983 /* we only allow rename subvolume link between subvolumes */
6984 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6987 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6988 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6991 if (S_ISDIR(old_inode->i_mode) && new_inode &&
6992 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6995 * we're using rename to replace one file with another.
6996 * and the replacement file is large. Start IO on it now so
6997 * we don't add too much work to the end of the transaction
6999 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7000 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7001 filemap_flush(old_inode->i_mapping);
7003 /* close the racy window with snapshot create/destroy ioctl */
7004 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7005 down_read(&root->fs_info->subvol_sem);
7007 * We want to reserve the absolute worst case amount of items. So if
7008 * both inodes are subvols and we need to unlink them then that would
7009 * require 4 item modifications, but if they are both normal inodes it
7010 * would require 5 item modifications, so we'll assume their normal
7011 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7012 * should cover the worst case number of items we'll modify.
7014 trans = btrfs_start_transaction(root, 20);
7015 if (IS_ERR(trans)) {
7016 ret = PTR_ERR(trans);
7021 btrfs_record_root_in_trans(trans, dest);
7023 ret = btrfs_set_inode_index(new_dir, &index);
7027 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7028 /* force full log commit if subvolume involved. */
7029 root->fs_info->last_trans_log_full_commit = trans->transid;
7031 ret = btrfs_insert_inode_ref(trans, dest,
7032 new_dentry->d_name.name,
7033 new_dentry->d_name.len,
7035 btrfs_ino(new_dir), index);
7039 * this is an ugly little race, but the rename is required
7040 * to make sure that if we crash, the inode is either at the
7041 * old name or the new one. pinning the log transaction lets
7042 * us make sure we don't allow a log commit to come in after
7043 * we unlink the name but before we add the new name back in.
7045 btrfs_pin_log_trans(root);
7048 * make sure the inode gets flushed if it is replacing
7051 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7052 btrfs_add_ordered_operation(trans, root, old_inode);
7054 old_dir->i_ctime = old_dir->i_mtime = ctime;
7055 new_dir->i_ctime = new_dir->i_mtime = ctime;
7056 old_inode->i_ctime = ctime;
7058 if (old_dentry->d_parent != new_dentry->d_parent)
7059 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7061 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7062 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7063 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7064 old_dentry->d_name.name,
7065 old_dentry->d_name.len);
7067 ret = __btrfs_unlink_inode(trans, root, old_dir,
7068 old_dentry->d_inode,
7069 old_dentry->d_name.name,
7070 old_dentry->d_name.len);
7072 ret = btrfs_update_inode(trans, root, old_inode);
7077 new_inode->i_ctime = CURRENT_TIME;
7078 if (unlikely(btrfs_ino(new_inode) ==
7079 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7080 root_objectid = BTRFS_I(new_inode)->location.objectid;
7081 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7083 new_dentry->d_name.name,
7084 new_dentry->d_name.len);
7085 BUG_ON(new_inode->i_nlink == 0);
7087 ret = btrfs_unlink_inode(trans, dest, new_dir,
7088 new_dentry->d_inode,
7089 new_dentry->d_name.name,
7090 new_dentry->d_name.len);
7093 if (new_inode->i_nlink == 0) {
7094 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7099 fixup_inode_flags(new_dir, old_inode);
7101 ret = btrfs_add_link(trans, new_dir, old_inode,
7102 new_dentry->d_name.name,
7103 new_dentry->d_name.len, 0, index);
7106 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7107 struct dentry *parent = new_dentry->d_parent;
7108 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7109 btrfs_end_log_trans(root);
7112 btrfs_end_transaction(trans, root);
7114 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7115 up_read(&root->fs_info->subvol_sem);
7121 * some fairly slow code that needs optimization. This walks the list
7122 * of all the inodes with pending delalloc and forces them to disk.
7124 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7126 struct list_head *head = &root->fs_info->delalloc_inodes;
7127 struct btrfs_inode *binode;
7128 struct inode *inode;
7130 if (root->fs_info->sb->s_flags & MS_RDONLY)
7133 spin_lock(&root->fs_info->delalloc_lock);
7134 while (!list_empty(head)) {
7135 binode = list_entry(head->next, struct btrfs_inode,
7137 inode = igrab(&binode->vfs_inode);
7139 list_del_init(&binode->delalloc_inodes);
7140 spin_unlock(&root->fs_info->delalloc_lock);
7142 filemap_flush(inode->i_mapping);
7144 btrfs_add_delayed_iput(inode);
7149 spin_lock(&root->fs_info->delalloc_lock);
7151 spin_unlock(&root->fs_info->delalloc_lock);
7153 /* the filemap_flush will queue IO into the worker threads, but
7154 * we have to make sure the IO is actually started and that
7155 * ordered extents get created before we return
7157 atomic_inc(&root->fs_info->async_submit_draining);
7158 while (atomic_read(&root->fs_info->nr_async_submits) ||
7159 atomic_read(&root->fs_info->async_delalloc_pages)) {
7160 wait_event(root->fs_info->async_submit_wait,
7161 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7162 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7164 atomic_dec(&root->fs_info->async_submit_draining);
7168 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7169 const char *symname)
7171 struct btrfs_trans_handle *trans;
7172 struct btrfs_root *root = BTRFS_I(dir)->root;
7173 struct btrfs_path *path;
7174 struct btrfs_key key;
7175 struct inode *inode = NULL;
7183 struct btrfs_file_extent_item *ei;
7184 struct extent_buffer *leaf;
7185 unsigned long nr = 0;
7187 name_len = strlen(symname) + 1;
7188 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7189 return -ENAMETOOLONG;
7192 * 2 items for inode item and ref
7193 * 2 items for dir items
7194 * 1 item for xattr if selinux is on
7196 trans = btrfs_start_transaction(root, 5);
7198 return PTR_ERR(trans);
7200 err = btrfs_find_free_ino(root, &objectid);
7204 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7205 dentry->d_name.len, btrfs_ino(dir), objectid,
7206 S_IFLNK|S_IRWXUGO, &index);
7207 if (IS_ERR(inode)) {
7208 err = PTR_ERR(inode);
7212 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7219 * If the active LSM wants to access the inode during
7220 * d_instantiate it needs these. Smack checks to see
7221 * if the filesystem supports xattrs by looking at the
7224 inode->i_fop = &btrfs_file_operations;
7225 inode->i_op = &btrfs_file_inode_operations;
7227 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7231 inode->i_mapping->a_ops = &btrfs_aops;
7232 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7233 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7238 path = btrfs_alloc_path();
7244 key.objectid = btrfs_ino(inode);
7246 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7247 datasize = btrfs_file_extent_calc_inline_size(name_len);
7248 err = btrfs_insert_empty_item(trans, root, path, &key,
7252 btrfs_free_path(path);
7255 leaf = path->nodes[0];
7256 ei = btrfs_item_ptr(leaf, path->slots[0],
7257 struct btrfs_file_extent_item);
7258 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7259 btrfs_set_file_extent_type(leaf, ei,
7260 BTRFS_FILE_EXTENT_INLINE);
7261 btrfs_set_file_extent_encryption(leaf, ei, 0);
7262 btrfs_set_file_extent_compression(leaf, ei, 0);
7263 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7264 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7266 ptr = btrfs_file_extent_inline_start(ei);
7267 write_extent_buffer(leaf, symname, ptr, name_len);
7268 btrfs_mark_buffer_dirty(leaf);
7269 btrfs_free_path(path);
7271 inode->i_op = &btrfs_symlink_inode_operations;
7272 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7273 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7274 inode_set_bytes(inode, name_len);
7275 btrfs_i_size_write(inode, name_len - 1);
7276 err = btrfs_update_inode(trans, root, inode);
7282 d_instantiate(dentry, inode);
7283 nr = trans->blocks_used;
7284 btrfs_end_transaction(trans, root);
7286 inode_dec_link_count(inode);
7289 btrfs_btree_balance_dirty(root, nr);
7293 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7294 u64 start, u64 num_bytes, u64 min_size,
7295 loff_t actual_len, u64 *alloc_hint,
7296 struct btrfs_trans_handle *trans)
7298 struct btrfs_root *root = BTRFS_I(inode)->root;
7299 struct btrfs_key ins;
7300 u64 cur_offset = start;
7303 bool own_trans = true;
7307 while (num_bytes > 0) {
7309 trans = btrfs_start_transaction(root, 3);
7310 if (IS_ERR(trans)) {
7311 ret = PTR_ERR(trans);
7316 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7317 0, *alloc_hint, &ins, 1);
7320 btrfs_end_transaction(trans, root);
7324 ret = insert_reserved_file_extent(trans, inode,
7325 cur_offset, ins.objectid,
7326 ins.offset, ins.offset,
7327 ins.offset, 0, 0, 0,
7328 BTRFS_FILE_EXTENT_PREALLOC);
7330 btrfs_drop_extent_cache(inode, cur_offset,
7331 cur_offset + ins.offset -1, 0);
7333 num_bytes -= ins.offset;
7334 cur_offset += ins.offset;
7335 *alloc_hint = ins.objectid + ins.offset;
7337 inode->i_ctime = CURRENT_TIME;
7338 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7339 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7340 (actual_len > inode->i_size) &&
7341 (cur_offset > inode->i_size)) {
7342 if (cur_offset > actual_len)
7343 i_size = actual_len;
7345 i_size = cur_offset;
7346 i_size_write(inode, i_size);
7347 btrfs_ordered_update_i_size(inode, i_size, NULL);
7350 ret = btrfs_update_inode(trans, root, inode);
7354 btrfs_end_transaction(trans, root);
7359 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7360 u64 start, u64 num_bytes, u64 min_size,
7361 loff_t actual_len, u64 *alloc_hint)
7363 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7364 min_size, actual_len, alloc_hint,
7368 int btrfs_prealloc_file_range_trans(struct inode *inode,
7369 struct btrfs_trans_handle *trans, int mode,
7370 u64 start, u64 num_bytes, u64 min_size,
7371 loff_t actual_len, u64 *alloc_hint)
7373 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7374 min_size, actual_len, alloc_hint, trans);
7377 static int btrfs_set_page_dirty(struct page *page)
7379 return __set_page_dirty_nobuffers(page);
7382 static int btrfs_permission(struct inode *inode, int mask)
7384 struct btrfs_root *root = BTRFS_I(inode)->root;
7385 umode_t mode = inode->i_mode;
7387 if (mask & MAY_WRITE &&
7388 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7389 if (btrfs_root_readonly(root))
7391 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7394 return generic_permission(inode, mask);
7397 static const struct inode_operations btrfs_dir_inode_operations = {
7398 .getattr = btrfs_getattr,
7399 .lookup = btrfs_lookup,
7400 .create = btrfs_create,
7401 .unlink = btrfs_unlink,
7403 .mkdir = btrfs_mkdir,
7404 .rmdir = btrfs_rmdir,
7405 .rename = btrfs_rename,
7406 .symlink = btrfs_symlink,
7407 .setattr = btrfs_setattr,
7408 .mknod = btrfs_mknod,
7409 .setxattr = btrfs_setxattr,
7410 .getxattr = btrfs_getxattr,
7411 .listxattr = btrfs_listxattr,
7412 .removexattr = btrfs_removexattr,
7413 .permission = btrfs_permission,
7414 .get_acl = btrfs_get_acl,
7416 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7417 .lookup = btrfs_lookup,
7418 .permission = btrfs_permission,
7419 .get_acl = btrfs_get_acl,
7422 static const struct file_operations btrfs_dir_file_operations = {
7423 .llseek = generic_file_llseek,
7424 .read = generic_read_dir,
7425 .readdir = btrfs_real_readdir,
7426 .unlocked_ioctl = btrfs_ioctl,
7427 #ifdef CONFIG_COMPAT
7428 .compat_ioctl = btrfs_ioctl,
7430 .release = btrfs_release_file,
7431 .fsync = btrfs_sync_file,
7434 static struct extent_io_ops btrfs_extent_io_ops = {
7435 .fill_delalloc = run_delalloc_range,
7436 .submit_bio_hook = btrfs_submit_bio_hook,
7437 .merge_bio_hook = btrfs_merge_bio_hook,
7438 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7439 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7440 .writepage_start_hook = btrfs_writepage_start_hook,
7441 .set_bit_hook = btrfs_set_bit_hook,
7442 .clear_bit_hook = btrfs_clear_bit_hook,
7443 .merge_extent_hook = btrfs_merge_extent_hook,
7444 .split_extent_hook = btrfs_split_extent_hook,
7448 * btrfs doesn't support the bmap operation because swapfiles
7449 * use bmap to make a mapping of extents in the file. They assume
7450 * these extents won't change over the life of the file and they
7451 * use the bmap result to do IO directly to the drive.
7453 * the btrfs bmap call would return logical addresses that aren't
7454 * suitable for IO and they also will change frequently as COW
7455 * operations happen. So, swapfile + btrfs == corruption.
7457 * For now we're avoiding this by dropping bmap.
7459 static const struct address_space_operations btrfs_aops = {
7460 .readpage = btrfs_readpage,
7461 .writepage = btrfs_writepage,
7462 .writepages = btrfs_writepages,
7463 .readpages = btrfs_readpages,
7464 .direct_IO = btrfs_direct_IO,
7465 .invalidatepage = btrfs_invalidatepage,
7466 .releasepage = btrfs_releasepage,
7467 .set_page_dirty = btrfs_set_page_dirty,
7468 .error_remove_page = generic_error_remove_page,
7471 static const struct address_space_operations btrfs_symlink_aops = {
7472 .readpage = btrfs_readpage,
7473 .writepage = btrfs_writepage,
7474 .invalidatepage = btrfs_invalidatepage,
7475 .releasepage = btrfs_releasepage,
7478 static const struct inode_operations btrfs_file_inode_operations = {
7479 .getattr = btrfs_getattr,
7480 .setattr = btrfs_setattr,
7481 .setxattr = btrfs_setxattr,
7482 .getxattr = btrfs_getxattr,
7483 .listxattr = btrfs_listxattr,
7484 .removexattr = btrfs_removexattr,
7485 .permission = btrfs_permission,
7486 .fiemap = btrfs_fiemap,
7487 .get_acl = btrfs_get_acl,
7489 static const struct inode_operations btrfs_special_inode_operations = {
7490 .getattr = btrfs_getattr,
7491 .setattr = btrfs_setattr,
7492 .permission = btrfs_permission,
7493 .setxattr = btrfs_setxattr,
7494 .getxattr = btrfs_getxattr,
7495 .listxattr = btrfs_listxattr,
7496 .removexattr = btrfs_removexattr,
7497 .get_acl = btrfs_get_acl,
7499 static const struct inode_operations btrfs_symlink_inode_operations = {
7500 .readlink = generic_readlink,
7501 .follow_link = page_follow_link_light,
7502 .put_link = page_put_link,
7503 .getattr = btrfs_getattr,
7504 .setattr = btrfs_setattr,
7505 .permission = btrfs_permission,
7506 .setxattr = btrfs_setxattr,
7507 .getxattr = btrfs_getxattr,
7508 .listxattr = btrfs_listxattr,
7509 .removexattr = btrfs_removexattr,
7510 .get_acl = btrfs_get_acl,
7513 const struct dentry_operations btrfs_dentry_operations = {
7514 .d_delete = btrfs_dentry_delete,
7515 .d_release = btrfs_dentry_release,