Btrfs: fix repeated delalloc work allocation
[~shefty/rdma-dev.git] / fs / btrfs / inode.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
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.
7  *
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.
12  *
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.
17  */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.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>
42 #include "compat.h"
43 #include "ctree.h"
44 #include "disk-io.h"
45 #include "transaction.h"
46 #include "btrfs_inode.h"
47 #include "ioctl.h"
48 #include "print-tree.h"
49 #include "ordered-data.h"
50 #include "xattr.h"
51 #include "tree-log.h"
52 #include "volumes.h"
53 #include "compression.h"
54 #include "locking.h"
55 #include "free-space-cache.h"
56 #include "inode-map.h"
57
58 struct btrfs_iget_args {
59         u64 ino;
60         struct btrfs_root *root;
61 };
62
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;
72
73 static struct kmem_cache *btrfs_inode_cachep;
74 static struct kmem_cache *btrfs_delalloc_work_cachep;
75 struct kmem_cache *btrfs_trans_handle_cachep;
76 struct kmem_cache *btrfs_transaction_cachep;
77 struct kmem_cache *btrfs_path_cachep;
78 struct kmem_cache *btrfs_free_space_cachep;
79
80 #define S_SHIFT 12
81 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
82         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
83         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
84         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
85         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
86         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
87         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
88         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
89 };
90
91 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
92 static int btrfs_truncate(struct inode *inode);
93 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
94 static noinline int cow_file_range(struct inode *inode,
95                                    struct page *locked_page,
96                                    u64 start, u64 end, int *page_started,
97                                    unsigned long *nr_written, int unlock);
98 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
99                                            u64 len, u64 orig_start,
100                                            u64 block_start, u64 block_len,
101                                            u64 orig_block_len, int type);
102
103 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
104                                      struct inode *inode,  struct inode *dir,
105                                      const struct qstr *qstr)
106 {
107         int err;
108
109         err = btrfs_init_acl(trans, inode, dir);
110         if (!err)
111                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
112         return err;
113 }
114
115 /*
116  * this does all the hard work for inserting an inline extent into
117  * the btree.  The caller should have done a btrfs_drop_extents so that
118  * no overlapping inline items exist in the btree
119  */
120 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
121                                 struct btrfs_root *root, struct inode *inode,
122                                 u64 start, size_t size, size_t compressed_size,
123                                 int compress_type,
124                                 struct page **compressed_pages)
125 {
126         struct btrfs_key key;
127         struct btrfs_path *path;
128         struct extent_buffer *leaf;
129         struct page *page = NULL;
130         char *kaddr;
131         unsigned long ptr;
132         struct btrfs_file_extent_item *ei;
133         int err = 0;
134         int ret;
135         size_t cur_size = size;
136         size_t datasize;
137         unsigned long offset;
138
139         if (compressed_size && compressed_pages)
140                 cur_size = compressed_size;
141
142         path = btrfs_alloc_path();
143         if (!path)
144                 return -ENOMEM;
145
146         path->leave_spinning = 1;
147
148         key.objectid = btrfs_ino(inode);
149         key.offset = start;
150         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
151         datasize = btrfs_file_extent_calc_inline_size(cur_size);
152
153         inode_add_bytes(inode, size);
154         ret = btrfs_insert_empty_item(trans, root, path, &key,
155                                       datasize);
156         if (ret) {
157                 err = ret;
158                 goto fail;
159         }
160         leaf = path->nodes[0];
161         ei = btrfs_item_ptr(leaf, path->slots[0],
162                             struct btrfs_file_extent_item);
163         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
164         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
165         btrfs_set_file_extent_encryption(leaf, ei, 0);
166         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
167         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
168         ptr = btrfs_file_extent_inline_start(ei);
169
170         if (compress_type != BTRFS_COMPRESS_NONE) {
171                 struct page *cpage;
172                 int i = 0;
173                 while (compressed_size > 0) {
174                         cpage = compressed_pages[i];
175                         cur_size = min_t(unsigned long, compressed_size,
176                                        PAGE_CACHE_SIZE);
177
178                         kaddr = kmap_atomic(cpage);
179                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
180                         kunmap_atomic(kaddr);
181
182                         i++;
183                         ptr += cur_size;
184                         compressed_size -= cur_size;
185                 }
186                 btrfs_set_file_extent_compression(leaf, ei,
187                                                   compress_type);
188         } else {
189                 page = find_get_page(inode->i_mapping,
190                                      start >> PAGE_CACHE_SHIFT);
191                 btrfs_set_file_extent_compression(leaf, ei, 0);
192                 kaddr = kmap_atomic(page);
193                 offset = start & (PAGE_CACHE_SIZE - 1);
194                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
195                 kunmap_atomic(kaddr);
196                 page_cache_release(page);
197         }
198         btrfs_mark_buffer_dirty(leaf);
199         btrfs_free_path(path);
200
201         /*
202          * we're an inline extent, so nobody can
203          * extend the file past i_size without locking
204          * a page we already have locked.
205          *
206          * We must do any isize and inode updates
207          * before we unlock the pages.  Otherwise we
208          * could end up racing with unlink.
209          */
210         BTRFS_I(inode)->disk_i_size = inode->i_size;
211         ret = btrfs_update_inode(trans, root, inode);
212
213         return ret;
214 fail:
215         btrfs_free_path(path);
216         return err;
217 }
218
219
220 /*
221  * conditionally insert an inline extent into the file.  This
222  * does the checks required to make sure the data is small enough
223  * to fit as an inline extent.
224  */
225 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
226                                  struct btrfs_root *root,
227                                  struct inode *inode, u64 start, u64 end,
228                                  size_t compressed_size, int compress_type,
229                                  struct page **compressed_pages)
230 {
231         u64 isize = i_size_read(inode);
232         u64 actual_end = min(end + 1, isize);
233         u64 inline_len = actual_end - start;
234         u64 aligned_end = (end + root->sectorsize - 1) &
235                         ~((u64)root->sectorsize - 1);
236         u64 data_len = inline_len;
237         int ret;
238
239         if (compressed_size)
240                 data_len = compressed_size;
241
242         if (start > 0 ||
243             actual_end >= PAGE_CACHE_SIZE ||
244             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
245             (!compressed_size &&
246             (actual_end & (root->sectorsize - 1)) == 0) ||
247             end + 1 < isize ||
248             data_len > root->fs_info->max_inline) {
249                 return 1;
250         }
251
252         ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
253         if (ret)
254                 return ret;
255
256         if (isize > actual_end)
257                 inline_len = min_t(u64, isize, actual_end);
258         ret = insert_inline_extent(trans, root, inode, start,
259                                    inline_len, compressed_size,
260                                    compress_type, compressed_pages);
261         if (ret && ret != -ENOSPC) {
262                 btrfs_abort_transaction(trans, root, ret);
263                 return ret;
264         } else if (ret == -ENOSPC) {
265                 return 1;
266         }
267
268         btrfs_delalloc_release_metadata(inode, end + 1 - start);
269         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
270         return 0;
271 }
272
273 struct async_extent {
274         u64 start;
275         u64 ram_size;
276         u64 compressed_size;
277         struct page **pages;
278         unsigned long nr_pages;
279         int compress_type;
280         struct list_head list;
281 };
282
283 struct async_cow {
284         struct inode *inode;
285         struct btrfs_root *root;
286         struct page *locked_page;
287         u64 start;
288         u64 end;
289         struct list_head extents;
290         struct btrfs_work work;
291 };
292
293 static noinline int add_async_extent(struct async_cow *cow,
294                                      u64 start, u64 ram_size,
295                                      u64 compressed_size,
296                                      struct page **pages,
297                                      unsigned long nr_pages,
298                                      int compress_type)
299 {
300         struct async_extent *async_extent;
301
302         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
303         BUG_ON(!async_extent); /* -ENOMEM */
304         async_extent->start = start;
305         async_extent->ram_size = ram_size;
306         async_extent->compressed_size = compressed_size;
307         async_extent->pages = pages;
308         async_extent->nr_pages = nr_pages;
309         async_extent->compress_type = compress_type;
310         list_add_tail(&async_extent->list, &cow->extents);
311         return 0;
312 }
313
314 /*
315  * we create compressed extents in two phases.  The first
316  * phase compresses a range of pages that have already been
317  * locked (both pages and state bits are locked).
318  *
319  * This is done inside an ordered work queue, and the compression
320  * is spread across many cpus.  The actual IO submission is step
321  * two, and the ordered work queue takes care of making sure that
322  * happens in the same order things were put onto the queue by
323  * writepages and friends.
324  *
325  * If this code finds it can't get good compression, it puts an
326  * entry onto the work queue to write the uncompressed bytes.  This
327  * makes sure that both compressed inodes and uncompressed inodes
328  * are written in the same order that the flusher thread sent them
329  * down.
330  */
331 static noinline int compress_file_range(struct inode *inode,
332                                         struct page *locked_page,
333                                         u64 start, u64 end,
334                                         struct async_cow *async_cow,
335                                         int *num_added)
336 {
337         struct btrfs_root *root = BTRFS_I(inode)->root;
338         struct btrfs_trans_handle *trans;
339         u64 num_bytes;
340         u64 blocksize = root->sectorsize;
341         u64 actual_end;
342         u64 isize = i_size_read(inode);
343         int ret = 0;
344         struct page **pages = NULL;
345         unsigned long nr_pages;
346         unsigned long nr_pages_ret = 0;
347         unsigned long total_compressed = 0;
348         unsigned long total_in = 0;
349         unsigned long max_compressed = 128 * 1024;
350         unsigned long max_uncompressed = 128 * 1024;
351         int i;
352         int will_compress;
353         int compress_type = root->fs_info->compress_type;
354
355         /* if this is a small write inside eof, kick off a defrag */
356         if ((end - start + 1) < 16 * 1024 &&
357             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
358                 btrfs_add_inode_defrag(NULL, inode);
359
360         actual_end = min_t(u64, isize, end + 1);
361 again:
362         will_compress = 0;
363         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
364         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
365
366         /*
367          * we don't want to send crud past the end of i_size through
368          * compression, that's just a waste of CPU time.  So, if the
369          * end of the file is before the start of our current
370          * requested range of bytes, we bail out to the uncompressed
371          * cleanup code that can deal with all of this.
372          *
373          * It isn't really the fastest way to fix things, but this is a
374          * very uncommon corner.
375          */
376         if (actual_end <= start)
377                 goto cleanup_and_bail_uncompressed;
378
379         total_compressed = actual_end - start;
380
381         /* we want to make sure that amount of ram required to uncompress
382          * an extent is reasonable, so we limit the total size in ram
383          * of a compressed extent to 128k.  This is a crucial number
384          * because it also controls how easily we can spread reads across
385          * cpus for decompression.
386          *
387          * We also want to make sure the amount of IO required to do
388          * a random read is reasonably small, so we limit the size of
389          * a compressed extent to 128k.
390          */
391         total_compressed = min(total_compressed, max_uncompressed);
392         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
393         num_bytes = max(blocksize,  num_bytes);
394         total_in = 0;
395         ret = 0;
396
397         /*
398          * we do compression for mount -o compress and when the
399          * inode has not been flagged as nocompress.  This flag can
400          * change at any time if we discover bad compression ratios.
401          */
402         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
403             (btrfs_test_opt(root, COMPRESS) ||
404              (BTRFS_I(inode)->force_compress) ||
405              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
406                 WARN_ON(pages);
407                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
408                 if (!pages) {
409                         /* just bail out to the uncompressed code */
410                         goto cont;
411                 }
412
413                 if (BTRFS_I(inode)->force_compress)
414                         compress_type = BTRFS_I(inode)->force_compress;
415
416                 ret = btrfs_compress_pages(compress_type,
417                                            inode->i_mapping, start,
418                                            total_compressed, pages,
419                                            nr_pages, &nr_pages_ret,
420                                            &total_in,
421                                            &total_compressed,
422                                            max_compressed);
423
424                 if (!ret) {
425                         unsigned long offset = total_compressed &
426                                 (PAGE_CACHE_SIZE - 1);
427                         struct page *page = pages[nr_pages_ret - 1];
428                         char *kaddr;
429
430                         /* zero the tail end of the last page, we might be
431                          * sending it down to disk
432                          */
433                         if (offset) {
434                                 kaddr = kmap_atomic(page);
435                                 memset(kaddr + offset, 0,
436                                        PAGE_CACHE_SIZE - offset);
437                                 kunmap_atomic(kaddr);
438                         }
439                         will_compress = 1;
440                 }
441         }
442 cont:
443         if (start == 0) {
444                 trans = btrfs_join_transaction(root);
445                 if (IS_ERR(trans)) {
446                         ret = PTR_ERR(trans);
447                         trans = NULL;
448                         goto cleanup_and_out;
449                 }
450                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
451
452                 /* lets try to make an inline extent */
453                 if (ret || total_in < (actual_end - start)) {
454                         /* we didn't compress the entire range, try
455                          * to make an uncompressed inline extent.
456                          */
457                         ret = cow_file_range_inline(trans, root, inode,
458                                                     start, end, 0, 0, NULL);
459                 } else {
460                         /* try making a compressed inline extent */
461                         ret = cow_file_range_inline(trans, root, inode,
462                                                     start, end,
463                                                     total_compressed,
464                                                     compress_type, pages);
465                 }
466                 if (ret <= 0) {
467                         /*
468                          * inline extent creation worked or returned error,
469                          * we don't need to create any more async work items.
470                          * Unlock and free up our temp pages.
471                          */
472                         extent_clear_unlock_delalloc(inode,
473                              &BTRFS_I(inode)->io_tree,
474                              start, end, NULL,
475                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
476                              EXTENT_CLEAR_DELALLOC |
477                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
478
479                         btrfs_end_transaction(trans, root);
480                         goto free_pages_out;
481                 }
482                 btrfs_end_transaction(trans, root);
483         }
484
485         if (will_compress) {
486                 /*
487                  * we aren't doing an inline extent round the compressed size
488                  * up to a block size boundary so the allocator does sane
489                  * things
490                  */
491                 total_compressed = (total_compressed + blocksize - 1) &
492                         ~(blocksize - 1);
493
494                 /*
495                  * one last check to make sure the compression is really a
496                  * win, compare the page count read with the blocks on disk
497                  */
498                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
499                         ~(PAGE_CACHE_SIZE - 1);
500                 if (total_compressed >= total_in) {
501                         will_compress = 0;
502                 } else {
503                         num_bytes = total_in;
504                 }
505         }
506         if (!will_compress && pages) {
507                 /*
508                  * the compression code ran but failed to make things smaller,
509                  * free any pages it allocated and our page pointer array
510                  */
511                 for (i = 0; i < nr_pages_ret; i++) {
512                         WARN_ON(pages[i]->mapping);
513                         page_cache_release(pages[i]);
514                 }
515                 kfree(pages);
516                 pages = NULL;
517                 total_compressed = 0;
518                 nr_pages_ret = 0;
519
520                 /* flag the file so we don't compress in the future */
521                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
522                     !(BTRFS_I(inode)->force_compress)) {
523                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
524                 }
525         }
526         if (will_compress) {
527                 *num_added += 1;
528
529                 /* the async work queues will take care of doing actual
530                  * allocation on disk for these compressed pages,
531                  * and will submit them to the elevator.
532                  */
533                 add_async_extent(async_cow, start, num_bytes,
534                                  total_compressed, pages, nr_pages_ret,
535                                  compress_type);
536
537                 if (start + num_bytes < end) {
538                         start += num_bytes;
539                         pages = NULL;
540                         cond_resched();
541                         goto again;
542                 }
543         } else {
544 cleanup_and_bail_uncompressed:
545                 /*
546                  * No compression, but we still need to write the pages in
547                  * the file we've been given so far.  redirty the locked
548                  * page if it corresponds to our extent and set things up
549                  * for the async work queue to run cow_file_range to do
550                  * the normal delalloc dance
551                  */
552                 if (page_offset(locked_page) >= start &&
553                     page_offset(locked_page) <= end) {
554                         __set_page_dirty_nobuffers(locked_page);
555                         /* unlocked later on in the async handlers */
556                 }
557                 add_async_extent(async_cow, start, end - start + 1,
558                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
559                 *num_added += 1;
560         }
561
562 out:
563         return ret;
564
565 free_pages_out:
566         for (i = 0; i < nr_pages_ret; i++) {
567                 WARN_ON(pages[i]->mapping);
568                 page_cache_release(pages[i]);
569         }
570         kfree(pages);
571
572         goto out;
573
574 cleanup_and_out:
575         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
576                                      start, end, NULL,
577                                      EXTENT_CLEAR_UNLOCK_PAGE |
578                                      EXTENT_CLEAR_DIRTY |
579                                      EXTENT_CLEAR_DELALLOC |
580                                      EXTENT_SET_WRITEBACK |
581                                      EXTENT_END_WRITEBACK);
582         if (!trans || IS_ERR(trans))
583                 btrfs_error(root->fs_info, ret, "Failed to join transaction");
584         else
585                 btrfs_abort_transaction(trans, root, ret);
586         goto free_pages_out;
587 }
588
589 /*
590  * phase two of compressed writeback.  This is the ordered portion
591  * of the code, which only gets called in the order the work was
592  * queued.  We walk all the async extents created by compress_file_range
593  * and send them down to the disk.
594  */
595 static noinline int submit_compressed_extents(struct inode *inode,
596                                               struct async_cow *async_cow)
597 {
598         struct async_extent *async_extent;
599         u64 alloc_hint = 0;
600         struct btrfs_trans_handle *trans;
601         struct btrfs_key ins;
602         struct extent_map *em;
603         struct btrfs_root *root = BTRFS_I(inode)->root;
604         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
605         struct extent_io_tree *io_tree;
606         int ret = 0;
607
608         if (list_empty(&async_cow->extents))
609                 return 0;
610
611
612         while (!list_empty(&async_cow->extents)) {
613                 async_extent = list_entry(async_cow->extents.next,
614                                           struct async_extent, list);
615                 list_del(&async_extent->list);
616
617                 io_tree = &BTRFS_I(inode)->io_tree;
618
619 retry:
620                 /* did the compression code fall back to uncompressed IO? */
621                 if (!async_extent->pages) {
622                         int page_started = 0;
623                         unsigned long nr_written = 0;
624
625                         lock_extent(io_tree, async_extent->start,
626                                          async_extent->start +
627                                          async_extent->ram_size - 1);
628
629                         /* allocate blocks */
630                         ret = cow_file_range(inode, async_cow->locked_page,
631                                              async_extent->start,
632                                              async_extent->start +
633                                              async_extent->ram_size - 1,
634                                              &page_started, &nr_written, 0);
635
636                         /* JDM XXX */
637
638                         /*
639                          * if page_started, cow_file_range inserted an
640                          * inline extent and took care of all the unlocking
641                          * and IO for us.  Otherwise, we need to submit
642                          * all those pages down to the drive.
643                          */
644                         if (!page_started && !ret)
645                                 extent_write_locked_range(io_tree,
646                                                   inode, async_extent->start,
647                                                   async_extent->start +
648                                                   async_extent->ram_size - 1,
649                                                   btrfs_get_extent,
650                                                   WB_SYNC_ALL);
651                         kfree(async_extent);
652                         cond_resched();
653                         continue;
654                 }
655
656                 lock_extent(io_tree, async_extent->start,
657                             async_extent->start + async_extent->ram_size - 1);
658
659                 trans = btrfs_join_transaction(root);
660                 if (IS_ERR(trans)) {
661                         ret = PTR_ERR(trans);
662                 } else {
663                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
664                         ret = btrfs_reserve_extent(trans, root,
665                                            async_extent->compressed_size,
666                                            async_extent->compressed_size,
667                                            0, alloc_hint, &ins, 1);
668                         if (ret && ret != -ENOSPC)
669                                 btrfs_abort_transaction(trans, root, ret);
670                         btrfs_end_transaction(trans, root);
671                 }
672
673                 if (ret) {
674                         int i;
675                         for (i = 0; i < async_extent->nr_pages; i++) {
676                                 WARN_ON(async_extent->pages[i]->mapping);
677                                 page_cache_release(async_extent->pages[i]);
678                         }
679                         kfree(async_extent->pages);
680                         async_extent->nr_pages = 0;
681                         async_extent->pages = NULL;
682                         unlock_extent(io_tree, async_extent->start,
683                                       async_extent->start +
684                                       async_extent->ram_size - 1);
685                         if (ret == -ENOSPC)
686                                 goto retry;
687                         goto out_free; /* JDM: Requeue? */
688                 }
689
690                 /*
691                  * here we're doing allocation and writeback of the
692                  * compressed pages
693                  */
694                 btrfs_drop_extent_cache(inode, async_extent->start,
695                                         async_extent->start +
696                                         async_extent->ram_size - 1, 0);
697
698                 em = alloc_extent_map();
699                 BUG_ON(!em); /* -ENOMEM */
700                 em->start = async_extent->start;
701                 em->len = async_extent->ram_size;
702                 em->orig_start = em->start;
703
704                 em->block_start = ins.objectid;
705                 em->block_len = ins.offset;
706                 em->orig_block_len = ins.offset;
707                 em->bdev = root->fs_info->fs_devices->latest_bdev;
708                 em->compress_type = async_extent->compress_type;
709                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
710                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
711                 em->generation = -1;
712
713                 while (1) {
714                         write_lock(&em_tree->lock);
715                         ret = add_extent_mapping(em_tree, em);
716                         if (!ret)
717                                 list_move(&em->list,
718                                           &em_tree->modified_extents);
719                         write_unlock(&em_tree->lock);
720                         if (ret != -EEXIST) {
721                                 free_extent_map(em);
722                                 break;
723                         }
724                         btrfs_drop_extent_cache(inode, async_extent->start,
725                                                 async_extent->start +
726                                                 async_extent->ram_size - 1, 0);
727                 }
728
729                 ret = btrfs_add_ordered_extent_compress(inode,
730                                                 async_extent->start,
731                                                 ins.objectid,
732                                                 async_extent->ram_size,
733                                                 ins.offset,
734                                                 BTRFS_ORDERED_COMPRESSED,
735                                                 async_extent->compress_type);
736                 BUG_ON(ret); /* -ENOMEM */
737
738                 /*
739                  * clear dirty, set writeback and unlock the pages.
740                  */
741                 extent_clear_unlock_delalloc(inode,
742                                 &BTRFS_I(inode)->io_tree,
743                                 async_extent->start,
744                                 async_extent->start +
745                                 async_extent->ram_size - 1,
746                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
747                                 EXTENT_CLEAR_UNLOCK |
748                                 EXTENT_CLEAR_DELALLOC |
749                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
750
751                 ret = btrfs_submit_compressed_write(inode,
752                                     async_extent->start,
753                                     async_extent->ram_size,
754                                     ins.objectid,
755                                     ins.offset, async_extent->pages,
756                                     async_extent->nr_pages);
757
758                 BUG_ON(ret); /* -ENOMEM */
759                 alloc_hint = ins.objectid + ins.offset;
760                 kfree(async_extent);
761                 cond_resched();
762         }
763         ret = 0;
764 out:
765         return ret;
766 out_free:
767         kfree(async_extent);
768         goto out;
769 }
770
771 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
772                                       u64 num_bytes)
773 {
774         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
775         struct extent_map *em;
776         u64 alloc_hint = 0;
777
778         read_lock(&em_tree->lock);
779         em = search_extent_mapping(em_tree, start, num_bytes);
780         if (em) {
781                 /*
782                  * if block start isn't an actual block number then find the
783                  * first block in this inode and use that as a hint.  If that
784                  * block is also bogus then just don't worry about it.
785                  */
786                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
787                         free_extent_map(em);
788                         em = search_extent_mapping(em_tree, 0, 0);
789                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
790                                 alloc_hint = em->block_start;
791                         if (em)
792                                 free_extent_map(em);
793                 } else {
794                         alloc_hint = em->block_start;
795                         free_extent_map(em);
796                 }
797         }
798         read_unlock(&em_tree->lock);
799
800         return alloc_hint;
801 }
802
803 /*
804  * when extent_io.c finds a delayed allocation range in the file,
805  * the call backs end up in this code.  The basic idea is to
806  * allocate extents on disk for the range, and create ordered data structs
807  * in ram to track those extents.
808  *
809  * locked_page is the page that writepage had locked already.  We use
810  * it to make sure we don't do extra locks or unlocks.
811  *
812  * *page_started is set to one if we unlock locked_page and do everything
813  * required to start IO on it.  It may be clean and already done with
814  * IO when we return.
815  */
816 static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
817                                      struct inode *inode,
818                                      struct btrfs_root *root,
819                                      struct page *locked_page,
820                                      u64 start, u64 end, int *page_started,
821                                      unsigned long *nr_written,
822                                      int unlock)
823 {
824         u64 alloc_hint = 0;
825         u64 num_bytes;
826         unsigned long ram_size;
827         u64 disk_num_bytes;
828         u64 cur_alloc_size;
829         u64 blocksize = root->sectorsize;
830         struct btrfs_key ins;
831         struct extent_map *em;
832         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
833         int ret = 0;
834
835         BUG_ON(btrfs_is_free_space_inode(inode));
836
837         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
838         num_bytes = max(blocksize,  num_bytes);
839         disk_num_bytes = num_bytes;
840
841         /* if this is a small write inside eof, kick off defrag */
842         if (num_bytes < 64 * 1024 &&
843             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
844                 btrfs_add_inode_defrag(trans, inode);
845
846         if (start == 0) {
847                 /* lets try to make an inline extent */
848                 ret = cow_file_range_inline(trans, root, inode,
849                                             start, end, 0, 0, NULL);
850                 if (ret == 0) {
851                         extent_clear_unlock_delalloc(inode,
852                                      &BTRFS_I(inode)->io_tree,
853                                      start, end, NULL,
854                                      EXTENT_CLEAR_UNLOCK_PAGE |
855                                      EXTENT_CLEAR_UNLOCK |
856                                      EXTENT_CLEAR_DELALLOC |
857                                      EXTENT_CLEAR_DIRTY |
858                                      EXTENT_SET_WRITEBACK |
859                                      EXTENT_END_WRITEBACK);
860
861                         *nr_written = *nr_written +
862                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
863                         *page_started = 1;
864                         goto out;
865                 } else if (ret < 0) {
866                         btrfs_abort_transaction(trans, root, ret);
867                         goto out_unlock;
868                 }
869         }
870
871         BUG_ON(disk_num_bytes >
872                btrfs_super_total_bytes(root->fs_info->super_copy));
873
874         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
875         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
876
877         while (disk_num_bytes > 0) {
878                 unsigned long op;
879
880                 cur_alloc_size = disk_num_bytes;
881                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
882                                            root->sectorsize, 0, alloc_hint,
883                                            &ins, 1);
884                 if (ret < 0) {
885                         btrfs_abort_transaction(trans, root, ret);
886                         goto out_unlock;
887                 }
888
889                 em = alloc_extent_map();
890                 BUG_ON(!em); /* -ENOMEM */
891                 em->start = start;
892                 em->orig_start = em->start;
893                 ram_size = ins.offset;
894                 em->len = ins.offset;
895
896                 em->block_start = ins.objectid;
897                 em->block_len = ins.offset;
898                 em->orig_block_len = ins.offset;
899                 em->bdev = root->fs_info->fs_devices->latest_bdev;
900                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
901                 em->generation = -1;
902
903                 while (1) {
904                         write_lock(&em_tree->lock);
905                         ret = add_extent_mapping(em_tree, em);
906                         if (!ret)
907                                 list_move(&em->list,
908                                           &em_tree->modified_extents);
909                         write_unlock(&em_tree->lock);
910                         if (ret != -EEXIST) {
911                                 free_extent_map(em);
912                                 break;
913                         }
914                         btrfs_drop_extent_cache(inode, start,
915                                                 start + ram_size - 1, 0);
916                 }
917
918                 cur_alloc_size = ins.offset;
919                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
920                                                ram_size, cur_alloc_size, 0);
921                 BUG_ON(ret); /* -ENOMEM */
922
923                 if (root->root_key.objectid ==
924                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
925                         ret = btrfs_reloc_clone_csums(inode, start,
926                                                       cur_alloc_size);
927                         if (ret) {
928                                 btrfs_abort_transaction(trans, root, ret);
929                                 goto out_unlock;
930                         }
931                 }
932
933                 if (disk_num_bytes < cur_alloc_size)
934                         break;
935
936                 /* we're not doing compressed IO, don't unlock the first
937                  * page (which the caller expects to stay locked), don't
938                  * clear any dirty bits and don't set any writeback bits
939                  *
940                  * Do set the Private2 bit so we know this page was properly
941                  * setup for writepage
942                  */
943                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
944                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
945                         EXTENT_SET_PRIVATE2;
946
947                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
948                                              start, start + ram_size - 1,
949                                              locked_page, op);
950                 disk_num_bytes -= cur_alloc_size;
951                 num_bytes -= cur_alloc_size;
952                 alloc_hint = ins.objectid + ins.offset;
953                 start += cur_alloc_size;
954         }
955 out:
956         return ret;
957
958 out_unlock:
959         extent_clear_unlock_delalloc(inode,
960                      &BTRFS_I(inode)->io_tree,
961                      start, end, locked_page,
962                      EXTENT_CLEAR_UNLOCK_PAGE |
963                      EXTENT_CLEAR_UNLOCK |
964                      EXTENT_CLEAR_DELALLOC |
965                      EXTENT_CLEAR_DIRTY |
966                      EXTENT_SET_WRITEBACK |
967                      EXTENT_END_WRITEBACK);
968
969         goto out;
970 }
971
972 static noinline int cow_file_range(struct inode *inode,
973                                    struct page *locked_page,
974                                    u64 start, u64 end, int *page_started,
975                                    unsigned long *nr_written,
976                                    int unlock)
977 {
978         struct btrfs_trans_handle *trans;
979         struct btrfs_root *root = BTRFS_I(inode)->root;
980         int ret;
981
982         trans = btrfs_join_transaction(root);
983         if (IS_ERR(trans)) {
984                 extent_clear_unlock_delalloc(inode,
985                              &BTRFS_I(inode)->io_tree,
986                              start, end, locked_page,
987                              EXTENT_CLEAR_UNLOCK_PAGE |
988                              EXTENT_CLEAR_UNLOCK |
989                              EXTENT_CLEAR_DELALLOC |
990                              EXTENT_CLEAR_DIRTY |
991                              EXTENT_SET_WRITEBACK |
992                              EXTENT_END_WRITEBACK);
993                 return PTR_ERR(trans);
994         }
995         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
996
997         ret = __cow_file_range(trans, inode, root, locked_page, start, end,
998                                page_started, nr_written, unlock);
999
1000         btrfs_end_transaction(trans, root);
1001
1002         return ret;
1003 }
1004
1005 /*
1006  * work queue call back to started compression on a file and pages
1007  */
1008 static noinline void async_cow_start(struct btrfs_work *work)
1009 {
1010         struct async_cow *async_cow;
1011         int num_added = 0;
1012         async_cow = container_of(work, struct async_cow, work);
1013
1014         compress_file_range(async_cow->inode, async_cow->locked_page,
1015                             async_cow->start, async_cow->end, async_cow,
1016                             &num_added);
1017         if (num_added == 0) {
1018                 btrfs_add_delayed_iput(async_cow->inode);
1019                 async_cow->inode = NULL;
1020         }
1021 }
1022
1023 /*
1024  * work queue call back to submit previously compressed pages
1025  */
1026 static noinline void async_cow_submit(struct btrfs_work *work)
1027 {
1028         struct async_cow *async_cow;
1029         struct btrfs_root *root;
1030         unsigned long nr_pages;
1031
1032         async_cow = container_of(work, struct async_cow, work);
1033
1034         root = async_cow->root;
1035         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1036                 PAGE_CACHE_SHIFT;
1037
1038         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1039             5 * 1024 * 1024 &&
1040             waitqueue_active(&root->fs_info->async_submit_wait))
1041                 wake_up(&root->fs_info->async_submit_wait);
1042
1043         if (async_cow->inode)
1044                 submit_compressed_extents(async_cow->inode, async_cow);
1045 }
1046
1047 static noinline void async_cow_free(struct btrfs_work *work)
1048 {
1049         struct async_cow *async_cow;
1050         async_cow = container_of(work, struct async_cow, work);
1051         if (async_cow->inode)
1052                 btrfs_add_delayed_iput(async_cow->inode);
1053         kfree(async_cow);
1054 }
1055
1056 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1057                                 u64 start, u64 end, int *page_started,
1058                                 unsigned long *nr_written)
1059 {
1060         struct async_cow *async_cow;
1061         struct btrfs_root *root = BTRFS_I(inode)->root;
1062         unsigned long nr_pages;
1063         u64 cur_end;
1064         int limit = 10 * 1024 * 1024;
1065
1066         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1067                          1, 0, NULL, GFP_NOFS);
1068         while (start < end) {
1069                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1070                 BUG_ON(!async_cow); /* -ENOMEM */
1071                 async_cow->inode = igrab(inode);
1072                 async_cow->root = root;
1073                 async_cow->locked_page = locked_page;
1074                 async_cow->start = start;
1075
1076                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1077                         cur_end = end;
1078                 else
1079                         cur_end = min(end, start + 512 * 1024 - 1);
1080
1081                 async_cow->end = cur_end;
1082                 INIT_LIST_HEAD(&async_cow->extents);
1083
1084                 async_cow->work.func = async_cow_start;
1085                 async_cow->work.ordered_func = async_cow_submit;
1086                 async_cow->work.ordered_free = async_cow_free;
1087                 async_cow->work.flags = 0;
1088
1089                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1090                         PAGE_CACHE_SHIFT;
1091                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1092
1093                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1094                                    &async_cow->work);
1095
1096                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1097                         wait_event(root->fs_info->async_submit_wait,
1098                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1099                             limit));
1100                 }
1101
1102                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1103                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1104                         wait_event(root->fs_info->async_submit_wait,
1105                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1106                            0));
1107                 }
1108
1109                 *nr_written += nr_pages;
1110                 start = cur_end + 1;
1111         }
1112         *page_started = 1;
1113         return 0;
1114 }
1115
1116 static noinline int csum_exist_in_range(struct btrfs_root *root,
1117                                         u64 bytenr, u64 num_bytes)
1118 {
1119         int ret;
1120         struct btrfs_ordered_sum *sums;
1121         LIST_HEAD(list);
1122
1123         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1124                                        bytenr + num_bytes - 1, &list, 0);
1125         if (ret == 0 && list_empty(&list))
1126                 return 0;
1127
1128         while (!list_empty(&list)) {
1129                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1130                 list_del(&sums->list);
1131                 kfree(sums);
1132         }
1133         return 1;
1134 }
1135
1136 /*
1137  * when nowcow writeback call back.  This checks for snapshots or COW copies
1138  * of the extents that exist in the file, and COWs the file as required.
1139  *
1140  * If no cow copies or snapshots exist, we write directly to the existing
1141  * blocks on disk
1142  */
1143 static noinline int run_delalloc_nocow(struct inode *inode,
1144                                        struct page *locked_page,
1145                               u64 start, u64 end, int *page_started, int force,
1146                               unsigned long *nr_written)
1147 {
1148         struct btrfs_root *root = BTRFS_I(inode)->root;
1149         struct btrfs_trans_handle *trans;
1150         struct extent_buffer *leaf;
1151         struct btrfs_path *path;
1152         struct btrfs_file_extent_item *fi;
1153         struct btrfs_key found_key;
1154         u64 cow_start;
1155         u64 cur_offset;
1156         u64 extent_end;
1157         u64 extent_offset;
1158         u64 disk_bytenr;
1159         u64 num_bytes;
1160         u64 disk_num_bytes;
1161         int extent_type;
1162         int ret, err;
1163         int type;
1164         int nocow;
1165         int check_prev = 1;
1166         bool nolock;
1167         u64 ino = btrfs_ino(inode);
1168
1169         path = btrfs_alloc_path();
1170         if (!path) {
1171                 extent_clear_unlock_delalloc(inode,
1172                              &BTRFS_I(inode)->io_tree,
1173                              start, end, locked_page,
1174                              EXTENT_CLEAR_UNLOCK_PAGE |
1175                              EXTENT_CLEAR_UNLOCK |
1176                              EXTENT_CLEAR_DELALLOC |
1177                              EXTENT_CLEAR_DIRTY |
1178                              EXTENT_SET_WRITEBACK |
1179                              EXTENT_END_WRITEBACK);
1180                 return -ENOMEM;
1181         }
1182
1183         nolock = btrfs_is_free_space_inode(inode);
1184
1185         if (nolock)
1186                 trans = btrfs_join_transaction_nolock(root);
1187         else
1188                 trans = btrfs_join_transaction(root);
1189
1190         if (IS_ERR(trans)) {
1191                 extent_clear_unlock_delalloc(inode,
1192                              &BTRFS_I(inode)->io_tree,
1193                              start, end, locked_page,
1194                              EXTENT_CLEAR_UNLOCK_PAGE |
1195                              EXTENT_CLEAR_UNLOCK |
1196                              EXTENT_CLEAR_DELALLOC |
1197                              EXTENT_CLEAR_DIRTY |
1198                              EXTENT_SET_WRITEBACK |
1199                              EXTENT_END_WRITEBACK);
1200                 btrfs_free_path(path);
1201                 return PTR_ERR(trans);
1202         }
1203
1204         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1205
1206         cow_start = (u64)-1;
1207         cur_offset = start;
1208         while (1) {
1209                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1210                                                cur_offset, 0);
1211                 if (ret < 0) {
1212                         btrfs_abort_transaction(trans, root, ret);
1213                         goto error;
1214                 }
1215                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1216                         leaf = path->nodes[0];
1217                         btrfs_item_key_to_cpu(leaf, &found_key,
1218                                               path->slots[0] - 1);
1219                         if (found_key.objectid == ino &&
1220                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1221                                 path->slots[0]--;
1222                 }
1223                 check_prev = 0;
1224 next_slot:
1225                 leaf = path->nodes[0];
1226                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1227                         ret = btrfs_next_leaf(root, path);
1228                         if (ret < 0) {
1229                                 btrfs_abort_transaction(trans, root, ret);
1230                                 goto error;
1231                         }
1232                         if (ret > 0)
1233                                 break;
1234                         leaf = path->nodes[0];
1235                 }
1236
1237                 nocow = 0;
1238                 disk_bytenr = 0;
1239                 num_bytes = 0;
1240                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1241
1242                 if (found_key.objectid > ino ||
1243                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1244                     found_key.offset > end)
1245                         break;
1246
1247                 if (found_key.offset > cur_offset) {
1248                         extent_end = found_key.offset;
1249                         extent_type = 0;
1250                         goto out_check;
1251                 }
1252
1253                 fi = btrfs_item_ptr(leaf, path->slots[0],
1254                                     struct btrfs_file_extent_item);
1255                 extent_type = btrfs_file_extent_type(leaf, fi);
1256
1257                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1258                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1259                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1260                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1261                         extent_end = found_key.offset +
1262                                 btrfs_file_extent_num_bytes(leaf, fi);
1263                         disk_num_bytes =
1264                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1265                         if (extent_end <= start) {
1266                                 path->slots[0]++;
1267                                 goto next_slot;
1268                         }
1269                         if (disk_bytenr == 0)
1270                                 goto out_check;
1271                         if (btrfs_file_extent_compression(leaf, fi) ||
1272                             btrfs_file_extent_encryption(leaf, fi) ||
1273                             btrfs_file_extent_other_encoding(leaf, fi))
1274                                 goto out_check;
1275                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1276                                 goto out_check;
1277                         if (btrfs_extent_readonly(root, disk_bytenr))
1278                                 goto out_check;
1279                         if (btrfs_cross_ref_exist(trans, root, ino,
1280                                                   found_key.offset -
1281                                                   extent_offset, disk_bytenr))
1282                                 goto out_check;
1283                         disk_bytenr += extent_offset;
1284                         disk_bytenr += cur_offset - found_key.offset;
1285                         num_bytes = min(end + 1, extent_end) - cur_offset;
1286                         /*
1287                          * force cow if csum exists in the range.
1288                          * this ensure that csum for a given extent are
1289                          * either valid or do not exist.
1290                          */
1291                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1292                                 goto out_check;
1293                         nocow = 1;
1294                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1295                         extent_end = found_key.offset +
1296                                 btrfs_file_extent_inline_len(leaf, fi);
1297                         extent_end = ALIGN(extent_end, root->sectorsize);
1298                 } else {
1299                         BUG_ON(1);
1300                 }
1301 out_check:
1302                 if (extent_end <= start) {
1303                         path->slots[0]++;
1304                         goto next_slot;
1305                 }
1306                 if (!nocow) {
1307                         if (cow_start == (u64)-1)
1308                                 cow_start = cur_offset;
1309                         cur_offset = extent_end;
1310                         if (cur_offset > end)
1311                                 break;
1312                         path->slots[0]++;
1313                         goto next_slot;
1314                 }
1315
1316                 btrfs_release_path(path);
1317                 if (cow_start != (u64)-1) {
1318                         ret = __cow_file_range(trans, inode, root, locked_page,
1319                                                cow_start, found_key.offset - 1,
1320                                                page_started, nr_written, 1);
1321                         if (ret) {
1322                                 btrfs_abort_transaction(trans, root, ret);
1323                                 goto error;
1324                         }
1325                         cow_start = (u64)-1;
1326                 }
1327
1328                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1329                         struct extent_map *em;
1330                         struct extent_map_tree *em_tree;
1331                         em_tree = &BTRFS_I(inode)->extent_tree;
1332                         em = alloc_extent_map();
1333                         BUG_ON(!em); /* -ENOMEM */
1334                         em->start = cur_offset;
1335                         em->orig_start = found_key.offset - extent_offset;
1336                         em->len = num_bytes;
1337                         em->block_len = num_bytes;
1338                         em->block_start = disk_bytenr;
1339                         em->orig_block_len = disk_num_bytes;
1340                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1341                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1342                         set_bit(EXTENT_FLAG_FILLING, &em->flags);
1343                         em->generation = -1;
1344                         while (1) {
1345                                 write_lock(&em_tree->lock);
1346                                 ret = add_extent_mapping(em_tree, em);
1347                                 if (!ret)
1348                                         list_move(&em->list,
1349                                                   &em_tree->modified_extents);
1350                                 write_unlock(&em_tree->lock);
1351                                 if (ret != -EEXIST) {
1352                                         free_extent_map(em);
1353                                         break;
1354                                 }
1355                                 btrfs_drop_extent_cache(inode, em->start,
1356                                                 em->start + em->len - 1, 0);
1357                         }
1358                         type = BTRFS_ORDERED_PREALLOC;
1359                 } else {
1360                         type = BTRFS_ORDERED_NOCOW;
1361                 }
1362
1363                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1364                                                num_bytes, num_bytes, type);
1365                 BUG_ON(ret); /* -ENOMEM */
1366
1367                 if (root->root_key.objectid ==
1368                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1369                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1370                                                       num_bytes);
1371                         if (ret) {
1372                                 btrfs_abort_transaction(trans, root, ret);
1373                                 goto error;
1374                         }
1375                 }
1376
1377                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1378                                 cur_offset, cur_offset + num_bytes - 1,
1379                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1380                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1381                                 EXTENT_SET_PRIVATE2);
1382                 cur_offset = extent_end;
1383                 if (cur_offset > end)
1384                         break;
1385         }
1386         btrfs_release_path(path);
1387
1388         if (cur_offset <= end && cow_start == (u64)-1) {
1389                 cow_start = cur_offset;
1390                 cur_offset = end;
1391         }
1392
1393         if (cow_start != (u64)-1) {
1394                 ret = __cow_file_range(trans, inode, root, locked_page,
1395                                        cow_start, end,
1396                                        page_started, nr_written, 1);
1397                 if (ret) {
1398                         btrfs_abort_transaction(trans, root, ret);
1399                         goto error;
1400                 }
1401         }
1402
1403 error:
1404         err = btrfs_end_transaction(trans, root);
1405         if (!ret)
1406                 ret = err;
1407
1408         if (ret && cur_offset < end)
1409                 extent_clear_unlock_delalloc(inode,
1410                              &BTRFS_I(inode)->io_tree,
1411                              cur_offset, end, locked_page,
1412                              EXTENT_CLEAR_UNLOCK_PAGE |
1413                              EXTENT_CLEAR_UNLOCK |
1414                              EXTENT_CLEAR_DELALLOC |
1415                              EXTENT_CLEAR_DIRTY |
1416                              EXTENT_SET_WRITEBACK |
1417                              EXTENT_END_WRITEBACK);
1418
1419         btrfs_free_path(path);
1420         return ret;
1421 }
1422
1423 /*
1424  * extent_io.c call back to do delayed allocation processing
1425  */
1426 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1427                               u64 start, u64 end, int *page_started,
1428                               unsigned long *nr_written)
1429 {
1430         int ret;
1431         struct btrfs_root *root = BTRFS_I(inode)->root;
1432
1433         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1434                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1435                                          page_started, 1, nr_written);
1436         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1437                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1438                                          page_started, 0, nr_written);
1439         } else if (!btrfs_test_opt(root, COMPRESS) &&
1440                    !(BTRFS_I(inode)->force_compress) &&
1441                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1442                 ret = cow_file_range(inode, locked_page, start, end,
1443                                       page_started, nr_written, 1);
1444         } else {
1445                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1446                         &BTRFS_I(inode)->runtime_flags);
1447                 ret = cow_file_range_async(inode, locked_page, start, end,
1448                                            page_started, nr_written);
1449         }
1450         return ret;
1451 }
1452
1453 static void btrfs_split_extent_hook(struct inode *inode,
1454                                     struct extent_state *orig, u64 split)
1455 {
1456         /* not delalloc, ignore it */
1457         if (!(orig->state & EXTENT_DELALLOC))
1458                 return;
1459
1460         spin_lock(&BTRFS_I(inode)->lock);
1461         BTRFS_I(inode)->outstanding_extents++;
1462         spin_unlock(&BTRFS_I(inode)->lock);
1463 }
1464
1465 /*
1466  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1467  * extents so we can keep track of new extents that are just merged onto old
1468  * extents, such as when we are doing sequential writes, so we can properly
1469  * account for the metadata space we'll need.
1470  */
1471 static void btrfs_merge_extent_hook(struct inode *inode,
1472                                     struct extent_state *new,
1473                                     struct extent_state *other)
1474 {
1475         /* not delalloc, ignore it */
1476         if (!(other->state & EXTENT_DELALLOC))
1477                 return;
1478
1479         spin_lock(&BTRFS_I(inode)->lock);
1480         BTRFS_I(inode)->outstanding_extents--;
1481         spin_unlock(&BTRFS_I(inode)->lock);
1482 }
1483
1484 /*
1485  * extent_io.c set_bit_hook, used to track delayed allocation
1486  * bytes in this file, and to maintain the list of inodes that
1487  * have pending delalloc work to be done.
1488  */
1489 static void btrfs_set_bit_hook(struct inode *inode,
1490                                struct extent_state *state, int *bits)
1491 {
1492
1493         /*
1494          * set_bit and clear bit hooks normally require _irqsave/restore
1495          * but in this case, we are only testing for the DELALLOC
1496          * bit, which is only set or cleared with irqs on
1497          */
1498         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1499                 struct btrfs_root *root = BTRFS_I(inode)->root;
1500                 u64 len = state->end + 1 - state->start;
1501                 bool do_list = !btrfs_is_free_space_inode(inode);
1502
1503                 if (*bits & EXTENT_FIRST_DELALLOC) {
1504                         *bits &= ~EXTENT_FIRST_DELALLOC;
1505                 } else {
1506                         spin_lock(&BTRFS_I(inode)->lock);
1507                         BTRFS_I(inode)->outstanding_extents++;
1508                         spin_unlock(&BTRFS_I(inode)->lock);
1509                 }
1510
1511                 spin_lock(&root->fs_info->delalloc_lock);
1512                 BTRFS_I(inode)->delalloc_bytes += len;
1513                 root->fs_info->delalloc_bytes += len;
1514                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1515                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1516                                       &root->fs_info->delalloc_inodes);
1517                 }
1518                 spin_unlock(&root->fs_info->delalloc_lock);
1519         }
1520 }
1521
1522 /*
1523  * extent_io.c clear_bit_hook, see set_bit_hook for why
1524  */
1525 static void btrfs_clear_bit_hook(struct inode *inode,
1526                                  struct extent_state *state, int *bits)
1527 {
1528         /*
1529          * set_bit and clear bit hooks normally require _irqsave/restore
1530          * but in this case, we are only testing for the DELALLOC
1531          * bit, which is only set or cleared with irqs on
1532          */
1533         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1534                 struct btrfs_root *root = BTRFS_I(inode)->root;
1535                 u64 len = state->end + 1 - state->start;
1536                 bool do_list = !btrfs_is_free_space_inode(inode);
1537
1538                 if (*bits & EXTENT_FIRST_DELALLOC) {
1539                         *bits &= ~EXTENT_FIRST_DELALLOC;
1540                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1541                         spin_lock(&BTRFS_I(inode)->lock);
1542                         BTRFS_I(inode)->outstanding_extents--;
1543                         spin_unlock(&BTRFS_I(inode)->lock);
1544                 }
1545
1546                 if (*bits & EXTENT_DO_ACCOUNTING)
1547                         btrfs_delalloc_release_metadata(inode, len);
1548
1549                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1550                     && do_list)
1551                         btrfs_free_reserved_data_space(inode, len);
1552
1553                 spin_lock(&root->fs_info->delalloc_lock);
1554                 root->fs_info->delalloc_bytes -= len;
1555                 BTRFS_I(inode)->delalloc_bytes -= len;
1556
1557                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1558                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1559                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1560                 }
1561                 spin_unlock(&root->fs_info->delalloc_lock);
1562         }
1563 }
1564
1565 /*
1566  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1567  * we don't create bios that span stripes or chunks
1568  */
1569 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1570                          size_t size, struct bio *bio,
1571                          unsigned long bio_flags)
1572 {
1573         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1574         u64 logical = (u64)bio->bi_sector << 9;
1575         u64 length = 0;
1576         u64 map_length;
1577         int ret;
1578
1579         if (bio_flags & EXTENT_BIO_COMPRESSED)
1580                 return 0;
1581
1582         length = bio->bi_size;
1583         map_length = length;
1584         ret = btrfs_map_block(root->fs_info, READ, logical,
1585                               &map_length, NULL, 0);
1586         /* Will always return 0 with map_multi == NULL */
1587         BUG_ON(ret < 0);
1588         if (map_length < length + size)
1589                 return 1;
1590         return 0;
1591 }
1592
1593 /*
1594  * in order to insert checksums into the metadata in large chunks,
1595  * we wait until bio submission time.   All the pages in the bio are
1596  * checksummed and sums are attached onto the ordered extent record.
1597  *
1598  * At IO completion time the cums attached on the ordered extent record
1599  * are inserted into the btree
1600  */
1601 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1602                                     struct bio *bio, int mirror_num,
1603                                     unsigned long bio_flags,
1604                                     u64 bio_offset)
1605 {
1606         struct btrfs_root *root = BTRFS_I(inode)->root;
1607         int ret = 0;
1608
1609         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1610         BUG_ON(ret); /* -ENOMEM */
1611         return 0;
1612 }
1613
1614 /*
1615  * in order to insert checksums into the metadata in large chunks,
1616  * we wait until bio submission time.   All the pages in the bio are
1617  * checksummed and sums are attached onto the ordered extent record.
1618  *
1619  * At IO completion time the cums attached on the ordered extent record
1620  * are inserted into the btree
1621  */
1622 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1623                           int mirror_num, unsigned long bio_flags,
1624                           u64 bio_offset)
1625 {
1626         struct btrfs_root *root = BTRFS_I(inode)->root;
1627         int ret;
1628
1629         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1630         if (ret)
1631                 bio_endio(bio, ret);
1632         return ret;
1633 }
1634
1635 /*
1636  * extent_io.c submission hook. This does the right thing for csum calculation
1637  * on write, or reading the csums from the tree before a read
1638  */
1639 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1640                           int mirror_num, unsigned long bio_flags,
1641                           u64 bio_offset)
1642 {
1643         struct btrfs_root *root = BTRFS_I(inode)->root;
1644         int ret = 0;
1645         int skip_sum;
1646         int metadata = 0;
1647         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1648
1649         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1650
1651         if (btrfs_is_free_space_inode(inode))
1652                 metadata = 2;
1653
1654         if (!(rw & REQ_WRITE)) {
1655                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1656                 if (ret)
1657                         goto out;
1658
1659                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1660                         ret = btrfs_submit_compressed_read(inode, bio,
1661                                                            mirror_num,
1662                                                            bio_flags);
1663                         goto out;
1664                 } else if (!skip_sum) {
1665                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1666                         if (ret)
1667                                 goto out;
1668                 }
1669                 goto mapit;
1670         } else if (async && !skip_sum) {
1671                 /* csum items have already been cloned */
1672                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1673                         goto mapit;
1674                 /* we're doing a write, do the async checksumming */
1675                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1676                                    inode, rw, bio, mirror_num,
1677                                    bio_flags, bio_offset,
1678                                    __btrfs_submit_bio_start,
1679                                    __btrfs_submit_bio_done);
1680                 goto out;
1681         } else if (!skip_sum) {
1682                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1683                 if (ret)
1684                         goto out;
1685         }
1686
1687 mapit:
1688         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1689
1690 out:
1691         if (ret < 0)
1692                 bio_endio(bio, ret);
1693         return ret;
1694 }
1695
1696 /*
1697  * given a list of ordered sums record them in the inode.  This happens
1698  * at IO completion time based on sums calculated at bio submission time.
1699  */
1700 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1701                              struct inode *inode, u64 file_offset,
1702                              struct list_head *list)
1703 {
1704         struct btrfs_ordered_sum *sum;
1705
1706         list_for_each_entry(sum, list, list) {
1707                 btrfs_csum_file_blocks(trans,
1708                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1709         }
1710         return 0;
1711 }
1712
1713 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1714                               struct extent_state **cached_state)
1715 {
1716         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1717         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1718                                    cached_state, GFP_NOFS);
1719 }
1720
1721 /* see btrfs_writepage_start_hook for details on why this is required */
1722 struct btrfs_writepage_fixup {
1723         struct page *page;
1724         struct btrfs_work work;
1725 };
1726
1727 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1728 {
1729         struct btrfs_writepage_fixup *fixup;
1730         struct btrfs_ordered_extent *ordered;
1731         struct extent_state *cached_state = NULL;
1732         struct page *page;
1733         struct inode *inode;
1734         u64 page_start;
1735         u64 page_end;
1736         int ret;
1737
1738         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1739         page = fixup->page;
1740 again:
1741         lock_page(page);
1742         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1743                 ClearPageChecked(page);
1744                 goto out_page;
1745         }
1746
1747         inode = page->mapping->host;
1748         page_start = page_offset(page);
1749         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1750
1751         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1752                          &cached_state);
1753
1754         /* already ordered? We're done */
1755         if (PagePrivate2(page))
1756                 goto out;
1757
1758         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1759         if (ordered) {
1760                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1761                                      page_end, &cached_state, GFP_NOFS);
1762                 unlock_page(page);
1763                 btrfs_start_ordered_extent(inode, ordered, 1);
1764                 btrfs_put_ordered_extent(ordered);
1765                 goto again;
1766         }
1767
1768         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1769         if (ret) {
1770                 mapping_set_error(page->mapping, ret);
1771                 end_extent_writepage(page, ret, page_start, page_end);
1772                 ClearPageChecked(page);
1773                 goto out;
1774          }
1775
1776         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1777         ClearPageChecked(page);
1778         set_page_dirty(page);
1779 out:
1780         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1781                              &cached_state, GFP_NOFS);
1782 out_page:
1783         unlock_page(page);
1784         page_cache_release(page);
1785         kfree(fixup);
1786 }
1787
1788 /*
1789  * There are a few paths in the higher layers of the kernel that directly
1790  * set the page dirty bit without asking the filesystem if it is a
1791  * good idea.  This causes problems because we want to make sure COW
1792  * properly happens and the data=ordered rules are followed.
1793  *
1794  * In our case any range that doesn't have the ORDERED bit set
1795  * hasn't been properly setup for IO.  We kick off an async process
1796  * to fix it up.  The async helper will wait for ordered extents, set
1797  * the delalloc bit and make it safe to write the page.
1798  */
1799 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1800 {
1801         struct inode *inode = page->mapping->host;
1802         struct btrfs_writepage_fixup *fixup;
1803         struct btrfs_root *root = BTRFS_I(inode)->root;
1804
1805         /* this page is properly in the ordered list */
1806         if (TestClearPagePrivate2(page))
1807                 return 0;
1808
1809         if (PageChecked(page))
1810                 return -EAGAIN;
1811
1812         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1813         if (!fixup)
1814                 return -EAGAIN;
1815
1816         SetPageChecked(page);
1817         page_cache_get(page);
1818         fixup->work.func = btrfs_writepage_fixup_worker;
1819         fixup->page = page;
1820         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1821         return -EBUSY;
1822 }
1823
1824 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1825                                        struct inode *inode, u64 file_pos,
1826                                        u64 disk_bytenr, u64 disk_num_bytes,
1827                                        u64 num_bytes, u64 ram_bytes,
1828                                        u8 compression, u8 encryption,
1829                                        u16 other_encoding, int extent_type)
1830 {
1831         struct btrfs_root *root = BTRFS_I(inode)->root;
1832         struct btrfs_file_extent_item *fi;
1833         struct btrfs_path *path;
1834         struct extent_buffer *leaf;
1835         struct btrfs_key ins;
1836         int ret;
1837
1838         path = btrfs_alloc_path();
1839         if (!path)
1840                 return -ENOMEM;
1841
1842         path->leave_spinning = 1;
1843
1844         /*
1845          * we may be replacing one extent in the tree with another.
1846          * The new extent is pinned in the extent map, and we don't want
1847          * to drop it from the cache until it is completely in the btree.
1848          *
1849          * So, tell btrfs_drop_extents to leave this extent in the cache.
1850          * the caller is expected to unpin it and allow it to be merged
1851          * with the others.
1852          */
1853         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1854                                  file_pos + num_bytes, 0);
1855         if (ret)
1856                 goto out;
1857
1858         ins.objectid = btrfs_ino(inode);
1859         ins.offset = file_pos;
1860         ins.type = BTRFS_EXTENT_DATA_KEY;
1861         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1862         if (ret)
1863                 goto out;
1864         leaf = path->nodes[0];
1865         fi = btrfs_item_ptr(leaf, path->slots[0],
1866                             struct btrfs_file_extent_item);
1867         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1868         btrfs_set_file_extent_type(leaf, fi, extent_type);
1869         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1870         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1871         btrfs_set_file_extent_offset(leaf, fi, 0);
1872         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1873         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1874         btrfs_set_file_extent_compression(leaf, fi, compression);
1875         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1876         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1877
1878         btrfs_mark_buffer_dirty(leaf);
1879         btrfs_release_path(path);
1880
1881         inode_add_bytes(inode, num_bytes);
1882
1883         ins.objectid = disk_bytenr;
1884         ins.offset = disk_num_bytes;
1885         ins.type = BTRFS_EXTENT_ITEM_KEY;
1886         ret = btrfs_alloc_reserved_file_extent(trans, root,
1887                                         root->root_key.objectid,
1888                                         btrfs_ino(inode), file_pos, &ins);
1889 out:
1890         btrfs_free_path(path);
1891
1892         return ret;
1893 }
1894
1895 /*
1896  * helper function for btrfs_finish_ordered_io, this
1897  * just reads in some of the csum leaves to prime them into ram
1898  * before we start the transaction.  It limits the amount of btree
1899  * reads required while inside the transaction.
1900  */
1901 /* as ordered data IO finishes, this gets called so we can finish
1902  * an ordered extent if the range of bytes in the file it covers are
1903  * fully written.
1904  */
1905 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
1906 {
1907         struct inode *inode = ordered_extent->inode;
1908         struct btrfs_root *root = BTRFS_I(inode)->root;
1909         struct btrfs_trans_handle *trans = NULL;
1910         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1911         struct extent_state *cached_state = NULL;
1912         int compress_type = 0;
1913         int ret;
1914         bool nolock;
1915
1916         nolock = btrfs_is_free_space_inode(inode);
1917
1918         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1919                 ret = -EIO;
1920                 goto out;
1921         }
1922
1923         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1924                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1925                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1926                 if (nolock)
1927                         trans = btrfs_join_transaction_nolock(root);
1928                 else
1929                         trans = btrfs_join_transaction(root);
1930                 if (IS_ERR(trans)) {
1931                         ret = PTR_ERR(trans);
1932                         trans = NULL;
1933                         goto out;
1934                 }
1935                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1936                 ret = btrfs_update_inode_fallback(trans, root, inode);
1937                 if (ret) /* -ENOMEM or corruption */
1938                         btrfs_abort_transaction(trans, root, ret);
1939                 goto out;
1940         }
1941
1942         lock_extent_bits(io_tree, ordered_extent->file_offset,
1943                          ordered_extent->file_offset + ordered_extent->len - 1,
1944                          0, &cached_state);
1945
1946         if (nolock)
1947                 trans = btrfs_join_transaction_nolock(root);
1948         else
1949                 trans = btrfs_join_transaction(root);
1950         if (IS_ERR(trans)) {
1951                 ret = PTR_ERR(trans);
1952                 trans = NULL;
1953                 goto out_unlock;
1954         }
1955         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1956
1957         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1958                 compress_type = ordered_extent->compress_type;
1959         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1960                 BUG_ON(compress_type);
1961                 ret = btrfs_mark_extent_written(trans, inode,
1962                                                 ordered_extent->file_offset,
1963                                                 ordered_extent->file_offset +
1964                                                 ordered_extent->len);
1965         } else {
1966                 BUG_ON(root == root->fs_info->tree_root);
1967                 ret = insert_reserved_file_extent(trans, inode,
1968                                                 ordered_extent->file_offset,
1969                                                 ordered_extent->start,
1970                                                 ordered_extent->disk_len,
1971                                                 ordered_extent->len,
1972                                                 ordered_extent->len,
1973                                                 compress_type, 0, 0,
1974                                                 BTRFS_FILE_EXTENT_REG);
1975         }
1976         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1977                            ordered_extent->file_offset, ordered_extent->len,
1978                            trans->transid);
1979         if (ret < 0) {
1980                 btrfs_abort_transaction(trans, root, ret);
1981                 goto out_unlock;
1982         }
1983
1984         add_pending_csums(trans, inode, ordered_extent->file_offset,
1985                           &ordered_extent->list);
1986
1987         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1988         ret = btrfs_update_inode_fallback(trans, root, inode);
1989         if (ret) { /* -ENOMEM or corruption */
1990                 btrfs_abort_transaction(trans, root, ret);
1991                 goto out_unlock;
1992         }
1993         ret = 0;
1994 out_unlock:
1995         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1996                              ordered_extent->file_offset +
1997                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1998 out:
1999         if (root != root->fs_info->tree_root)
2000                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2001         if (trans)
2002                 btrfs_end_transaction(trans, root);
2003
2004         if (ret)
2005                 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
2006                                       ordered_extent->file_offset +
2007                                       ordered_extent->len - 1, NULL, GFP_NOFS);
2008
2009         /*
2010          * This needs to be done to make sure anybody waiting knows we are done
2011          * updating everything for this ordered extent.
2012          */
2013         btrfs_remove_ordered_extent(inode, ordered_extent);
2014
2015         /* once for us */
2016         btrfs_put_ordered_extent(ordered_extent);
2017         /* once for the tree */
2018         btrfs_put_ordered_extent(ordered_extent);
2019
2020         return ret;
2021 }
2022
2023 static void finish_ordered_fn(struct btrfs_work *work)
2024 {
2025         struct btrfs_ordered_extent *ordered_extent;
2026         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2027         btrfs_finish_ordered_io(ordered_extent);
2028 }
2029
2030 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2031                                 struct extent_state *state, int uptodate)
2032 {
2033         struct inode *inode = page->mapping->host;
2034         struct btrfs_root *root = BTRFS_I(inode)->root;
2035         struct btrfs_ordered_extent *ordered_extent = NULL;
2036         struct btrfs_workers *workers;
2037
2038         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2039
2040         ClearPagePrivate2(page);
2041         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2042                                             end - start + 1, uptodate))
2043                 return 0;
2044
2045         ordered_extent->work.func = finish_ordered_fn;
2046         ordered_extent->work.flags = 0;
2047
2048         if (btrfs_is_free_space_inode(inode))
2049                 workers = &root->fs_info->endio_freespace_worker;
2050         else
2051                 workers = &root->fs_info->endio_write_workers;
2052         btrfs_queue_worker(workers, &ordered_extent->work);
2053
2054         return 0;
2055 }
2056
2057 /*
2058  * when reads are done, we need to check csums to verify the data is correct
2059  * if there's a match, we allow the bio to finish.  If not, the code in
2060  * extent_io.c will try to find good copies for us.
2061  */
2062 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2063                                struct extent_state *state, int mirror)
2064 {
2065         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
2066         struct inode *inode = page->mapping->host;
2067         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2068         char *kaddr;
2069         u64 private = ~(u32)0;
2070         int ret;
2071         struct btrfs_root *root = BTRFS_I(inode)->root;
2072         u32 csum = ~(u32)0;
2073
2074         if (PageChecked(page)) {
2075                 ClearPageChecked(page);
2076                 goto good;
2077         }
2078
2079         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2080                 goto good;
2081
2082         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2083             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2084                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2085                                   GFP_NOFS);
2086                 return 0;
2087         }
2088
2089         if (state && state->start == start) {
2090                 private = state->private;
2091                 ret = 0;
2092         } else {
2093                 ret = get_state_private(io_tree, start, &private);
2094         }
2095         kaddr = kmap_atomic(page);
2096         if (ret)
2097                 goto zeroit;
2098
2099         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2100         btrfs_csum_final(csum, (char *)&csum);
2101         if (csum != private)
2102                 goto zeroit;
2103
2104         kunmap_atomic(kaddr);
2105 good:
2106         return 0;
2107
2108 zeroit:
2109         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2110                        "private %llu\n",
2111                        (unsigned long long)btrfs_ino(page->mapping->host),
2112                        (unsigned long long)start, csum,
2113                        (unsigned long long)private);
2114         memset(kaddr + offset, 1, end - start + 1);
2115         flush_dcache_page(page);
2116         kunmap_atomic(kaddr);
2117         if (private == 0)
2118                 return 0;
2119         return -EIO;
2120 }
2121
2122 struct delayed_iput {
2123         struct list_head list;
2124         struct inode *inode;
2125 };
2126
2127 /* JDM: If this is fs-wide, why can't we add a pointer to
2128  * btrfs_inode instead and avoid the allocation? */
2129 void btrfs_add_delayed_iput(struct inode *inode)
2130 {
2131         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2132         struct delayed_iput *delayed;
2133
2134         if (atomic_add_unless(&inode->i_count, -1, 1))
2135                 return;
2136
2137         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2138         delayed->inode = inode;
2139
2140         spin_lock(&fs_info->delayed_iput_lock);
2141         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2142         spin_unlock(&fs_info->delayed_iput_lock);
2143 }
2144
2145 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2146 {
2147         LIST_HEAD(list);
2148         struct btrfs_fs_info *fs_info = root->fs_info;
2149         struct delayed_iput *delayed;
2150         int empty;
2151
2152         spin_lock(&fs_info->delayed_iput_lock);
2153         empty = list_empty(&fs_info->delayed_iputs);
2154         spin_unlock(&fs_info->delayed_iput_lock);
2155         if (empty)
2156                 return;
2157
2158         spin_lock(&fs_info->delayed_iput_lock);
2159         list_splice_init(&fs_info->delayed_iputs, &list);
2160         spin_unlock(&fs_info->delayed_iput_lock);
2161
2162         while (!list_empty(&list)) {
2163                 delayed = list_entry(list.next, struct delayed_iput, list);
2164                 list_del(&delayed->list);
2165                 iput(delayed->inode);
2166                 kfree(delayed);
2167         }
2168 }
2169
2170 enum btrfs_orphan_cleanup_state {
2171         ORPHAN_CLEANUP_STARTED  = 1,
2172         ORPHAN_CLEANUP_DONE     = 2,
2173 };
2174
2175 /*
2176  * This is called in transaction commit time. If there are no orphan
2177  * files in the subvolume, it removes orphan item and frees block_rsv
2178  * structure.
2179  */
2180 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2181                               struct btrfs_root *root)
2182 {
2183         struct btrfs_block_rsv *block_rsv;
2184         int ret;
2185
2186         if (atomic_read(&root->orphan_inodes) ||
2187             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2188                 return;
2189
2190         spin_lock(&root->orphan_lock);
2191         if (atomic_read(&root->orphan_inodes)) {
2192                 spin_unlock(&root->orphan_lock);
2193                 return;
2194         }
2195
2196         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2197                 spin_unlock(&root->orphan_lock);
2198                 return;
2199         }
2200
2201         block_rsv = root->orphan_block_rsv;
2202         root->orphan_block_rsv = NULL;
2203         spin_unlock(&root->orphan_lock);
2204
2205         if (root->orphan_item_inserted &&
2206             btrfs_root_refs(&root->root_item) > 0) {
2207                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2208                                             root->root_key.objectid);
2209                 BUG_ON(ret);
2210                 root->orphan_item_inserted = 0;
2211         }
2212
2213         if (block_rsv) {
2214                 WARN_ON(block_rsv->size > 0);
2215                 btrfs_free_block_rsv(root, block_rsv);
2216         }
2217 }
2218
2219 /*
2220  * This creates an orphan entry for the given inode in case something goes
2221  * wrong in the middle of an unlink/truncate.
2222  *
2223  * NOTE: caller of this function should reserve 5 units of metadata for
2224  *       this function.
2225  */
2226 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2227 {
2228         struct btrfs_root *root = BTRFS_I(inode)->root;
2229         struct btrfs_block_rsv *block_rsv = NULL;
2230         int reserve = 0;
2231         int insert = 0;
2232         int ret;
2233
2234         if (!root->orphan_block_rsv) {
2235                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2236                 if (!block_rsv)
2237                         return -ENOMEM;
2238         }
2239
2240         spin_lock(&root->orphan_lock);
2241         if (!root->orphan_block_rsv) {
2242                 root->orphan_block_rsv = block_rsv;
2243         } else if (block_rsv) {
2244                 btrfs_free_block_rsv(root, block_rsv);
2245                 block_rsv = NULL;
2246         }
2247
2248         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2249                               &BTRFS_I(inode)->runtime_flags)) {
2250 #if 0
2251                 /*
2252                  * For proper ENOSPC handling, we should do orphan
2253                  * cleanup when mounting. But this introduces backward
2254                  * compatibility issue.
2255                  */
2256                 if (!xchg(&root->orphan_item_inserted, 1))
2257                         insert = 2;
2258                 else
2259                         insert = 1;
2260 #endif
2261                 insert = 1;
2262                 atomic_inc(&root->orphan_inodes);
2263         }
2264
2265         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2266                               &BTRFS_I(inode)->runtime_flags))
2267                 reserve = 1;
2268         spin_unlock(&root->orphan_lock);
2269
2270         /* grab metadata reservation from transaction handle */
2271         if (reserve) {
2272                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2273                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2274         }
2275
2276         /* insert an orphan item to track this unlinked/truncated file */
2277         if (insert >= 1) {
2278                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2279                 if (ret && ret != -EEXIST) {
2280                         clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2281                                   &BTRFS_I(inode)->runtime_flags);
2282                         btrfs_abort_transaction(trans, root, ret);
2283                         return ret;
2284                 }
2285                 ret = 0;
2286         }
2287
2288         /* insert an orphan item to track subvolume contains orphan files */
2289         if (insert >= 2) {
2290                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2291                                                root->root_key.objectid);
2292                 if (ret && ret != -EEXIST) {
2293                         btrfs_abort_transaction(trans, root, ret);
2294                         return ret;
2295                 }
2296         }
2297         return 0;
2298 }
2299
2300 /*
2301  * We have done the truncate/delete so we can go ahead and remove the orphan
2302  * item for this particular inode.
2303  */
2304 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2305 {
2306         struct btrfs_root *root = BTRFS_I(inode)->root;
2307         int delete_item = 0;
2308         int release_rsv = 0;
2309         int ret = 0;
2310
2311         spin_lock(&root->orphan_lock);
2312         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2313                                &BTRFS_I(inode)->runtime_flags))
2314                 delete_item = 1;
2315
2316         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2317                                &BTRFS_I(inode)->runtime_flags))
2318                 release_rsv = 1;
2319         spin_unlock(&root->orphan_lock);
2320
2321         if (trans && delete_item) {
2322                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2323                 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2324         }
2325
2326         if (release_rsv) {
2327                 btrfs_orphan_release_metadata(inode);
2328                 atomic_dec(&root->orphan_inodes);
2329         }
2330
2331         return 0;
2332 }
2333
2334 /*
2335  * this cleans up any orphans that may be left on the list from the last use
2336  * of this root.
2337  */
2338 int btrfs_orphan_cleanup(struct btrfs_root *root)
2339 {
2340         struct btrfs_path *path;
2341         struct extent_buffer *leaf;
2342         struct btrfs_key key, found_key;
2343         struct btrfs_trans_handle *trans;
2344         struct inode *inode;
2345         u64 last_objectid = 0;
2346         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2347
2348         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2349                 return 0;
2350
2351         path = btrfs_alloc_path();
2352         if (!path) {
2353                 ret = -ENOMEM;
2354                 goto out;
2355         }
2356         path->reada = -1;
2357
2358         key.objectid = BTRFS_ORPHAN_OBJECTID;
2359         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2360         key.offset = (u64)-1;
2361
2362         while (1) {
2363                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2364                 if (ret < 0)
2365                         goto out;
2366
2367                 /*
2368                  * if ret == 0 means we found what we were searching for, which
2369                  * is weird, but possible, so only screw with path if we didn't
2370                  * find the key and see if we have stuff that matches
2371                  */
2372                 if (ret > 0) {
2373                         ret = 0;
2374                         if (path->slots[0] == 0)
2375                                 break;
2376                         path->slots[0]--;
2377                 }
2378
2379                 /* pull out the item */
2380                 leaf = path->nodes[0];
2381                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2382
2383                 /* make sure the item matches what we want */
2384                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2385                         break;
2386                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2387                         break;
2388
2389                 /* release the path since we're done with it */
2390                 btrfs_release_path(path);
2391
2392                 /*
2393                  * this is where we are basically btrfs_lookup, without the
2394                  * crossing root thing.  we store the inode number in the
2395                  * offset of the orphan item.
2396                  */
2397
2398                 if (found_key.offset == last_objectid) {
2399                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2400                                "stopping orphan cleanup\n");
2401                         ret = -EINVAL;
2402                         goto out;
2403                 }
2404
2405                 last_objectid = found_key.offset;
2406
2407                 found_key.objectid = found_key.offset;
2408                 found_key.type = BTRFS_INODE_ITEM_KEY;
2409                 found_key.offset = 0;
2410                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2411                 ret = PTR_RET(inode);
2412                 if (ret && ret != -ESTALE)
2413                         goto out;
2414
2415                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2416                         struct btrfs_root *dead_root;
2417                         struct btrfs_fs_info *fs_info = root->fs_info;
2418                         int is_dead_root = 0;
2419
2420                         /*
2421                          * this is an orphan in the tree root. Currently these
2422                          * could come from 2 sources:
2423                          *  a) a snapshot deletion in progress
2424                          *  b) a free space cache inode
2425                          * We need to distinguish those two, as the snapshot
2426                          * orphan must not get deleted.
2427                          * find_dead_roots already ran before us, so if this
2428                          * is a snapshot deletion, we should find the root
2429                          * in the dead_roots list
2430                          */
2431                         spin_lock(&fs_info->trans_lock);
2432                         list_for_each_entry(dead_root, &fs_info->dead_roots,
2433                                             root_list) {
2434                                 if (dead_root->root_key.objectid ==
2435                                     found_key.objectid) {
2436                                         is_dead_root = 1;
2437                                         break;
2438                                 }
2439                         }
2440                         spin_unlock(&fs_info->trans_lock);
2441                         if (is_dead_root) {
2442                                 /* prevent this orphan from being found again */
2443                                 key.offset = found_key.objectid - 1;
2444                                 continue;
2445                         }
2446                 }
2447                 /*
2448                  * Inode is already gone but the orphan item is still there,
2449                  * kill the orphan item.
2450                  */
2451                 if (ret == -ESTALE) {
2452                         trans = btrfs_start_transaction(root, 1);
2453                         if (IS_ERR(trans)) {
2454                                 ret = PTR_ERR(trans);
2455                                 goto out;
2456                         }
2457                         printk(KERN_ERR "auto deleting %Lu\n",
2458                                found_key.objectid);
2459                         ret = btrfs_del_orphan_item(trans, root,
2460                                                     found_key.objectid);
2461                         BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2462                         btrfs_end_transaction(trans, root);
2463                         continue;
2464                 }
2465
2466                 /*
2467                  * add this inode to the orphan list so btrfs_orphan_del does
2468                  * the proper thing when we hit it
2469                  */
2470                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2471                         &BTRFS_I(inode)->runtime_flags);
2472
2473                 /* if we have links, this was a truncate, lets do that */
2474                 if (inode->i_nlink) {
2475                         if (!S_ISREG(inode->i_mode)) {
2476                                 WARN_ON(1);
2477                                 iput(inode);
2478                                 continue;
2479                         }
2480                         nr_truncate++;
2481
2482                         /* 1 for the orphan item deletion. */
2483                         trans = btrfs_start_transaction(root, 1);
2484                         if (IS_ERR(trans)) {
2485                                 ret = PTR_ERR(trans);
2486                                 goto out;
2487                         }
2488                         ret = btrfs_orphan_add(trans, inode);
2489                         btrfs_end_transaction(trans, root);
2490                         if (ret)
2491                                 goto out;
2492
2493                         ret = btrfs_truncate(inode);
2494                 } else {
2495                         nr_unlink++;
2496                 }
2497
2498                 /* this will do delete_inode and everything for us */
2499                 iput(inode);
2500                 if (ret)
2501                         goto out;
2502         }
2503         /* release the path since we're done with it */
2504         btrfs_release_path(path);
2505
2506         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2507
2508         if (root->orphan_block_rsv)
2509                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2510                                         (u64)-1);
2511
2512         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2513                 trans = btrfs_join_transaction(root);
2514                 if (!IS_ERR(trans))
2515                         btrfs_end_transaction(trans, root);
2516         }
2517
2518         if (nr_unlink)
2519                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2520         if (nr_truncate)
2521                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2522
2523 out:
2524         if (ret)
2525                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2526         btrfs_free_path(path);
2527         return ret;
2528 }
2529
2530 /*
2531  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2532  * don't find any xattrs, we know there can't be any acls.
2533  *
2534  * slot is the slot the inode is in, objectid is the objectid of the inode
2535  */
2536 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2537                                           int slot, u64 objectid)
2538 {
2539         u32 nritems = btrfs_header_nritems(leaf);
2540         struct btrfs_key found_key;
2541         int scanned = 0;
2542
2543         slot++;
2544         while (slot < nritems) {
2545                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2546
2547                 /* we found a different objectid, there must not be acls */
2548                 if (found_key.objectid != objectid)
2549                         return 0;
2550
2551                 /* we found an xattr, assume we've got an acl */
2552                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2553                         return 1;
2554
2555                 /*
2556                  * we found a key greater than an xattr key, there can't
2557                  * be any acls later on
2558                  */
2559                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2560                         return 0;
2561
2562                 slot++;
2563                 scanned++;
2564
2565                 /*
2566                  * it goes inode, inode backrefs, xattrs, extents,
2567                  * so if there are a ton of hard links to an inode there can
2568                  * be a lot of backrefs.  Don't waste time searching too hard,
2569                  * this is just an optimization
2570                  */
2571                 if (scanned >= 8)
2572                         break;
2573         }
2574         /* we hit the end of the leaf before we found an xattr or
2575          * something larger than an xattr.  We have to assume the inode
2576          * has acls
2577          */
2578         return 1;
2579 }
2580
2581 /*
2582  * read an inode from the btree into the in-memory inode
2583  */
2584 static void btrfs_read_locked_inode(struct inode *inode)
2585 {
2586         struct btrfs_path *path;
2587         struct extent_buffer *leaf;
2588         struct btrfs_inode_item *inode_item;
2589         struct btrfs_timespec *tspec;
2590         struct btrfs_root *root = BTRFS_I(inode)->root;
2591         struct btrfs_key location;
2592         int maybe_acls;
2593         u32 rdev;
2594         int ret;
2595         bool filled = false;
2596
2597         ret = btrfs_fill_inode(inode, &rdev);
2598         if (!ret)
2599                 filled = true;
2600
2601         path = btrfs_alloc_path();
2602         if (!path)
2603                 goto make_bad;
2604
2605         path->leave_spinning = 1;
2606         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2607
2608         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2609         if (ret)
2610                 goto make_bad;
2611
2612         leaf = path->nodes[0];
2613
2614         if (filled)
2615                 goto cache_acl;
2616
2617         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2618                                     struct btrfs_inode_item);
2619         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2620         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2621         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
2622         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
2623         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2624
2625         tspec = btrfs_inode_atime(inode_item);
2626         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2627         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2628
2629         tspec = btrfs_inode_mtime(inode_item);
2630         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2631         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2632
2633         tspec = btrfs_inode_ctime(inode_item);
2634         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2635         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2636
2637         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2638         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2639         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
2640
2641         /*
2642          * If we were modified in the current generation and evicted from memory
2643          * and then re-read we need to do a full sync since we don't have any
2644          * idea about which extents were modified before we were evicted from
2645          * cache.
2646          */
2647         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
2648                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2649                         &BTRFS_I(inode)->runtime_flags);
2650
2651         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2652         inode->i_generation = BTRFS_I(inode)->generation;
2653         inode->i_rdev = 0;
2654         rdev = btrfs_inode_rdev(leaf, inode_item);
2655
2656         BTRFS_I(inode)->index_cnt = (u64)-1;
2657         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2658 cache_acl:
2659         /*
2660          * try to precache a NULL acl entry for files that don't have
2661          * any xattrs or acls
2662          */
2663         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2664                                            btrfs_ino(inode));
2665         if (!maybe_acls)
2666                 cache_no_acl(inode);
2667
2668         btrfs_free_path(path);
2669
2670         switch (inode->i_mode & S_IFMT) {
2671         case S_IFREG:
2672                 inode->i_mapping->a_ops = &btrfs_aops;
2673                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2674                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2675                 inode->i_fop = &btrfs_file_operations;
2676                 inode->i_op = &btrfs_file_inode_operations;
2677                 break;
2678         case S_IFDIR:
2679                 inode->i_fop = &btrfs_dir_file_operations;
2680                 if (root == root->fs_info->tree_root)
2681                         inode->i_op = &btrfs_dir_ro_inode_operations;
2682                 else
2683                         inode->i_op = &btrfs_dir_inode_operations;
2684                 break;
2685         case S_IFLNK:
2686                 inode->i_op = &btrfs_symlink_inode_operations;
2687                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2688                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2689                 break;
2690         default:
2691                 inode->i_op = &btrfs_special_inode_operations;
2692                 init_special_inode(inode, inode->i_mode, rdev);
2693                 break;
2694         }
2695
2696         btrfs_update_iflags(inode);
2697         return;
2698
2699 make_bad:
2700         btrfs_free_path(path);
2701         make_bad_inode(inode);
2702 }
2703
2704 /*
2705  * given a leaf and an inode, copy the inode fields into the leaf
2706  */
2707 static void fill_inode_item(struct btrfs_trans_handle *trans,
2708                             struct extent_buffer *leaf,
2709                             struct btrfs_inode_item *item,
2710                             struct inode *inode)
2711 {
2712         btrfs_set_inode_uid(leaf, item, i_uid_read(inode));
2713         btrfs_set_inode_gid(leaf, item, i_gid_read(inode));
2714         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2715         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2716         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2717
2718         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2719                                inode->i_atime.tv_sec);
2720         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2721                                 inode->i_atime.tv_nsec);
2722
2723         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2724                                inode->i_mtime.tv_sec);
2725         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2726                                 inode->i_mtime.tv_nsec);
2727
2728         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2729                                inode->i_ctime.tv_sec);
2730         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2731                                 inode->i_ctime.tv_nsec);
2732
2733         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2734         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2735         btrfs_set_inode_sequence(leaf, item, inode->i_version);
2736         btrfs_set_inode_transid(leaf, item, trans->transid);
2737         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2738         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2739         btrfs_set_inode_block_group(leaf, item, 0);
2740 }
2741
2742 /*
2743  * copy everything in the in-memory inode into the btree.
2744  */
2745 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2746                                 struct btrfs_root *root, struct inode *inode)
2747 {
2748         struct btrfs_inode_item *inode_item;
2749         struct btrfs_path *path;
2750         struct extent_buffer *leaf;
2751         int ret;
2752
2753         path = btrfs_alloc_path();
2754         if (!path)
2755                 return -ENOMEM;
2756
2757         path->leave_spinning = 1;
2758         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2759                                  1);
2760         if (ret) {
2761                 if (ret > 0)
2762                         ret = -ENOENT;
2763                 goto failed;
2764         }
2765
2766         btrfs_unlock_up_safe(path, 1);
2767         leaf = path->nodes[0];
2768         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2769                                     struct btrfs_inode_item);
2770
2771         fill_inode_item(trans, leaf, inode_item, inode);
2772         btrfs_mark_buffer_dirty(leaf);
2773         btrfs_set_inode_last_trans(trans, inode);
2774         ret = 0;
2775 failed:
2776         btrfs_free_path(path);
2777         return ret;
2778 }
2779
2780 /*
2781  * copy everything in the in-memory inode into the btree.
2782  */
2783 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2784                                 struct btrfs_root *root, struct inode *inode)
2785 {
2786         int ret;
2787
2788         /*
2789          * If the inode is a free space inode, we can deadlock during commit
2790          * if we put it into the delayed code.
2791          *
2792          * The data relocation inode should also be directly updated
2793          * without delay
2794          */
2795         if (!btrfs_is_free_space_inode(inode)
2796             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2797                 btrfs_update_root_times(trans, root);
2798
2799                 ret = btrfs_delayed_update_inode(trans, root, inode);
2800                 if (!ret)
2801                         btrfs_set_inode_last_trans(trans, inode);
2802                 return ret;
2803         }
2804
2805         return btrfs_update_inode_item(trans, root, inode);
2806 }
2807
2808 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2809                                          struct btrfs_root *root,
2810                                          struct inode *inode)
2811 {
2812         int ret;
2813
2814         ret = btrfs_update_inode(trans, root, inode);
2815         if (ret == -ENOSPC)
2816                 return btrfs_update_inode_item(trans, root, inode);
2817         return ret;
2818 }
2819
2820 /*
2821  * unlink helper that gets used here in inode.c and in the tree logging
2822  * recovery code.  It remove a link in a directory with a given name, and
2823  * also drops the back refs in the inode to the directory
2824  */
2825 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2826                                 struct btrfs_root *root,
2827                                 struct inode *dir, struct inode *inode,
2828                                 const char *name, int name_len)
2829 {
2830         struct btrfs_path *path;
2831         int ret = 0;
2832         struct extent_buffer *leaf;
2833         struct btrfs_dir_item *di;
2834         struct btrfs_key key;
2835         u64 index;
2836         u64 ino = btrfs_ino(inode);
2837         u64 dir_ino = btrfs_ino(dir);
2838
2839         path = btrfs_alloc_path();
2840         if (!path) {
2841                 ret = -ENOMEM;
2842                 goto out;
2843         }
2844
2845         path->leave_spinning = 1;
2846         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2847                                     name, name_len, -1);
2848         if (IS_ERR(di)) {
2849                 ret = PTR_ERR(di);
2850                 goto err;
2851         }
2852         if (!di) {
2853                 ret = -ENOENT;
2854                 goto err;
2855         }
2856         leaf = path->nodes[0];
2857         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2858         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2859         if (ret)
2860                 goto err;
2861         btrfs_release_path(path);
2862
2863         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2864                                   dir_ino, &index);
2865         if (ret) {
2866                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2867                        "inode %llu parent %llu\n", name_len, name,
2868                        (unsigned long long)ino, (unsigned long long)dir_ino);
2869                 btrfs_abort_transaction(trans, root, ret);
2870                 goto err;
2871         }
2872
2873         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2874         if (ret) {
2875                 btrfs_abort_transaction(trans, root, ret);
2876                 goto err;
2877         }
2878
2879         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2880                                          inode, dir_ino);
2881         if (ret != 0 && ret != -ENOENT) {
2882                 btrfs_abort_transaction(trans, root, ret);
2883                 goto err;
2884         }
2885
2886         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2887                                            dir, index);
2888         if (ret == -ENOENT)
2889                 ret = 0;
2890 err:
2891         btrfs_free_path(path);
2892         if (ret)
2893                 goto out;
2894
2895         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2896         inode_inc_iversion(inode);
2897         inode_inc_iversion(dir);
2898         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2899         ret = btrfs_update_inode(trans, root, dir);
2900 out:
2901         return ret;
2902 }
2903
2904 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2905                        struct btrfs_root *root,
2906                        struct inode *dir, struct inode *inode,
2907                        const char *name, int name_len)
2908 {
2909         int ret;
2910         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2911         if (!ret) {
2912                 btrfs_drop_nlink(inode);
2913                 ret = btrfs_update_inode(trans, root, inode);
2914         }
2915         return ret;
2916 }
2917                 
2918
2919 /* helper to check if there is any shared block in the path */
2920 static int check_path_shared(struct btrfs_root *root,
2921                              struct btrfs_path *path)
2922 {
2923         struct extent_buffer *eb;
2924         int level;
2925         u64 refs = 1;
2926
2927         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2928                 int ret;
2929
2930                 if (!path->nodes[level])
2931                         break;
2932                 eb = path->nodes[level];
2933                 if (!btrfs_block_can_be_shared(root, eb))
2934                         continue;
2935                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2936                                                &refs, NULL);
2937                 if (refs > 1)
2938                         return 1;
2939         }
2940         return 0;
2941 }
2942
2943 /*
2944  * helper to start transaction for unlink and rmdir.
2945  *
2946  * unlink and rmdir are special in btrfs, they do not always free space.
2947  * so in enospc case, we should make sure they will free space before
2948  * allowing them to use the global metadata reservation.
2949  */
2950 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2951                                                        struct dentry *dentry)
2952 {
2953         struct btrfs_trans_handle *trans;
2954         struct btrfs_root *root = BTRFS_I(dir)->root;
2955         struct btrfs_path *path;
2956         struct btrfs_dir_item *di;
2957         struct inode *inode = dentry->d_inode;
2958         u64 index;
2959         int check_link = 1;
2960         int err = -ENOSPC;
2961         int ret;
2962         u64 ino = btrfs_ino(inode);
2963         u64 dir_ino = btrfs_ino(dir);
2964
2965         /*
2966          * 1 for the possible orphan item
2967          * 1 for the dir item
2968          * 1 for the dir index
2969          * 1 for the inode ref
2970          * 1 for the inode ref in the tree log
2971          * 2 for the dir entries in the log
2972          * 1 for the inode
2973          */
2974         trans = btrfs_start_transaction(root, 8);
2975         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2976                 return trans;
2977
2978         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2979                 return ERR_PTR(-ENOSPC);
2980
2981         /* check if there is someone else holds reference */
2982         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2983                 return ERR_PTR(-ENOSPC);
2984
2985         if (atomic_read(&inode->i_count) > 2)
2986                 return ERR_PTR(-ENOSPC);
2987
2988         if (xchg(&root->fs_info->enospc_unlink, 1))
2989                 return ERR_PTR(-ENOSPC);
2990
2991         path = btrfs_alloc_path();
2992         if (!path) {
2993                 root->fs_info->enospc_unlink = 0;
2994                 return ERR_PTR(-ENOMEM);
2995         }
2996
2997         /* 1 for the orphan item */
2998         trans = btrfs_start_transaction(root, 1);
2999         if (IS_ERR(trans)) {
3000                 btrfs_free_path(path);
3001                 root->fs_info->enospc_unlink = 0;
3002                 return trans;
3003         }
3004
3005         path->skip_locking = 1;
3006         path->search_commit_root = 1;
3007
3008         ret = btrfs_lookup_inode(trans, root, path,
3009                                 &BTRFS_I(dir)->location, 0);
3010         if (ret < 0) {
3011                 err = ret;
3012                 goto out;
3013         }
3014         if (ret == 0) {
3015                 if (check_path_shared(root, path))
3016                         goto out;
3017         } else {
3018                 check_link = 0;
3019         }
3020         btrfs_release_path(path);
3021
3022         ret = btrfs_lookup_inode(trans, root, path,
3023                                 &BTRFS_I(inode)->location, 0);
3024         if (ret < 0) {
3025                 err = ret;
3026                 goto out;
3027         }
3028         if (ret == 0) {
3029                 if (check_path_shared(root, path))
3030                         goto out;
3031         } else {
3032                 check_link = 0;
3033         }
3034         btrfs_release_path(path);
3035
3036         if (ret == 0 && S_ISREG(inode->i_mode)) {
3037                 ret = btrfs_lookup_file_extent(trans, root, path,
3038                                                ino, (u64)-1, 0);
3039                 if (ret < 0) {
3040                         err = ret;
3041                         goto out;
3042                 }
3043                 BUG_ON(ret == 0); /* Corruption */
3044                 if (check_path_shared(root, path))
3045                         goto out;
3046                 btrfs_release_path(path);
3047         }
3048
3049         if (!check_link) {
3050                 err = 0;
3051                 goto out;
3052         }
3053
3054         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3055                                 dentry->d_name.name, dentry->d_name.len, 0);
3056         if (IS_ERR(di)) {
3057                 err = PTR_ERR(di);
3058                 goto out;
3059         }
3060         if (di) {
3061                 if (check_path_shared(root, path))
3062                         goto out;
3063         } else {
3064                 err = 0;
3065                 goto out;
3066         }
3067         btrfs_release_path(path);
3068
3069         ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3070                                         dentry->d_name.len, ino, dir_ino, 0,
3071                                         &index);
3072         if (ret) {
3073                 err = ret;
3074                 goto out;
3075         }
3076
3077         if (check_path_shared(root, path))
3078                 goto out;
3079
3080         btrfs_release_path(path);
3081
3082         /*
3083          * This is a commit root search, if we can lookup inode item and other
3084          * relative items in the commit root, it means the transaction of
3085          * dir/file creation has been committed, and the dir index item that we
3086          * delay to insert has also been inserted into the commit root. So
3087          * we needn't worry about the delayed insertion of the dir index item
3088          * here.
3089          */
3090         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3091                                 dentry->d_name.name, dentry->d_name.len, 0);
3092         if (IS_ERR(di)) {
3093                 err = PTR_ERR(di);
3094                 goto out;
3095         }
3096         BUG_ON(ret == -ENOENT);
3097         if (check_path_shared(root, path))
3098                 goto out;
3099
3100         err = 0;
3101 out:
3102         btrfs_free_path(path);
3103         /* Migrate the orphan reservation over */
3104         if (!err)
3105                 err = btrfs_block_rsv_migrate(trans->block_rsv,
3106                                 &root->fs_info->global_block_rsv,
3107                                 trans->bytes_reserved);
3108
3109         if (err) {
3110                 btrfs_end_transaction(trans, root);
3111                 root->fs_info->enospc_unlink = 0;
3112                 return ERR_PTR(err);
3113         }
3114
3115         trans->block_rsv = &root->fs_info->global_block_rsv;
3116         return trans;
3117 }
3118
3119 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3120                                struct btrfs_root *root)
3121 {
3122         if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3123                 btrfs_block_rsv_release(root, trans->block_rsv,
3124                                         trans->bytes_reserved);
3125                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3126                 BUG_ON(!root->fs_info->enospc_unlink);
3127                 root->fs_info->enospc_unlink = 0;
3128         }
3129         btrfs_end_transaction(trans, root);
3130 }
3131
3132 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3133 {
3134         struct btrfs_root *root = BTRFS_I(dir)->root;
3135         struct btrfs_trans_handle *trans;
3136         struct inode *inode = dentry->d_inode;
3137         int ret;
3138
3139         trans = __unlink_start_trans(dir, dentry);
3140         if (IS_ERR(trans))
3141                 return PTR_ERR(trans);
3142
3143         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3144
3145         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3146                                  dentry->d_name.name, dentry->d_name.len);
3147         if (ret)
3148                 goto out;
3149
3150         if (inode->i_nlink == 0) {
3151                 ret = btrfs_orphan_add(trans, inode);
3152                 if (ret)
3153                         goto out;
3154         }
3155
3156 out:
3157         __unlink_end_trans(trans, root);
3158         btrfs_btree_balance_dirty(root);
3159         return ret;
3160 }
3161
3162 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3163                         struct btrfs_root *root,
3164                         struct inode *dir, u64 objectid,
3165                         const char *name, int name_len)
3166 {
3167         struct btrfs_path *path;
3168         struct extent_buffer *leaf;
3169         struct btrfs_dir_item *di;
3170         struct btrfs_key key;
3171         u64 index;
3172         int ret;
3173         u64 dir_ino = btrfs_ino(dir);
3174
3175         path = btrfs_alloc_path();
3176         if (!path)
3177                 return -ENOMEM;
3178
3179         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3180                                    name, name_len, -1);
3181         if (IS_ERR_OR_NULL(di)) {
3182                 if (!di)
3183                         ret = -ENOENT;
3184                 else
3185                         ret = PTR_ERR(di);
3186                 goto out;
3187         }
3188
3189         leaf = path->nodes[0];
3190         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3191         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3192         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3193         if (ret) {
3194                 btrfs_abort_transaction(trans, root, ret);
3195                 goto out;
3196         }
3197         btrfs_release_path(path);
3198
3199         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3200                                  objectid, root->root_key.objectid,
3201                                  dir_ino, &index, name, name_len);
3202         if (ret < 0) {
3203                 if (ret != -ENOENT) {
3204                         btrfs_abort_transaction(trans, root, ret);
3205                         goto out;
3206                 }
3207                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3208                                              &nb