]> git.openfabrics.org - ~shefty/rdma-dev.git/blob - fs/btrfs/inode.c
Btrfs: fix permissions of empty files not affected by umask
[~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, loff_t newsize);
92 static int btrfs_truncate(struct inode *inode);
93 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
94 static noinline int cow_file_range(struct inode *inode,
95                                    struct page *locked_page,
96                                    u64 start, u64 end, int *page_started,
97                                    unsigned long *nr_written, int unlock);
98
99 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
100                                      struct inode *inode,  struct inode *dir,
101                                      const struct qstr *qstr)
102 {
103         int err;
104
105         err = btrfs_init_acl(trans, inode, dir);
106         if (!err)
107                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
108         return err;
109 }
110
111 /*
112  * this does all the hard work for inserting an inline extent into
113  * the btree.  The caller should have done a btrfs_drop_extents so that
114  * no overlapping inline items exist in the btree
115  */
116 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
117                                 struct btrfs_root *root, struct inode *inode,
118                                 u64 start, size_t size, size_t compressed_size,
119                                 int compress_type,
120                                 struct page **compressed_pages)
121 {
122         struct btrfs_key key;
123         struct btrfs_path *path;
124         struct extent_buffer *leaf;
125         struct page *page = NULL;
126         char *kaddr;
127         unsigned long ptr;
128         struct btrfs_file_extent_item *ei;
129         int err = 0;
130         int ret;
131         size_t cur_size = size;
132         size_t datasize;
133         unsigned long offset;
134
135         if (compressed_size && compressed_pages)
136                 cur_size = compressed_size;
137
138         path = btrfs_alloc_path();
139         if (!path)
140                 return -ENOMEM;
141
142         path->leave_spinning = 1;
143
144         key.objectid = btrfs_ino(inode);
145         key.offset = start;
146         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
147         datasize = btrfs_file_extent_calc_inline_size(cur_size);
148
149         inode_add_bytes(inode, size);
150         ret = btrfs_insert_empty_item(trans, root, path, &key,
151                                       datasize);
152         if (ret) {
153                 err = ret;
154                 goto fail;
155         }
156         leaf = path->nodes[0];
157         ei = btrfs_item_ptr(leaf, path->slots[0],
158                             struct btrfs_file_extent_item);
159         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
160         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
161         btrfs_set_file_extent_encryption(leaf, ei, 0);
162         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
163         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
164         ptr = btrfs_file_extent_inline_start(ei);
165
166         if (compress_type != BTRFS_COMPRESS_NONE) {
167                 struct page *cpage;
168                 int i = 0;
169                 while (compressed_size > 0) {
170                         cpage = compressed_pages[i];
171                         cur_size = min_t(unsigned long, compressed_size,
172                                        PAGE_CACHE_SIZE);
173
174                         kaddr = kmap_atomic(cpage);
175                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
176                         kunmap_atomic(kaddr);
177
178                         i++;
179                         ptr += cur_size;
180                         compressed_size -= cur_size;
181                 }
182                 btrfs_set_file_extent_compression(leaf, ei,
183                                                   compress_type);
184         } else {
185                 page = find_get_page(inode->i_mapping,
186                                      start >> PAGE_CACHE_SHIFT);
187                 btrfs_set_file_extent_compression(leaf, ei, 0);
188                 kaddr = kmap_atomic(page);
189                 offset = start & (PAGE_CACHE_SIZE - 1);
190                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
191                 kunmap_atomic(kaddr);
192                 page_cache_release(page);
193         }
194         btrfs_mark_buffer_dirty(leaf);
195         btrfs_free_path(path);
196
197         /*
198          * we're an inline extent, so nobody can
199          * extend the file past i_size without locking
200          * a page we already have locked.
201          *
202          * We must do any isize and inode updates
203          * before we unlock the pages.  Otherwise we
204          * could end up racing with unlink.
205          */
206         BTRFS_I(inode)->disk_i_size = inode->i_size;
207         ret = btrfs_update_inode(trans, root, inode);
208
209         return ret;
210 fail:
211         btrfs_free_path(path);
212         return err;
213 }
214
215
216 /*
217  * conditionally insert an inline extent into the file.  This
218  * does the checks required to make sure the data is small enough
219  * to fit as an inline extent.
220  */
221 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
222                                  struct btrfs_root *root,
223                                  struct inode *inode, u64 start, u64 end,
224                                  size_t compressed_size, int compress_type,
225                                  struct page **compressed_pages)
226 {
227         u64 isize = i_size_read(inode);
228         u64 actual_end = min(end + 1, isize);
229         u64 inline_len = actual_end - start;
230         u64 aligned_end = (end + root->sectorsize - 1) &
231                         ~((u64)root->sectorsize - 1);
232         u64 data_len = inline_len;
233         int ret;
234
235         if (compressed_size)
236                 data_len = compressed_size;
237
238         if (start > 0 ||
239             actual_end >= PAGE_CACHE_SIZE ||
240             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
241             (!compressed_size &&
242             (actual_end & (root->sectorsize - 1)) == 0) ||
243             end + 1 < isize ||
244             data_len > root->fs_info->max_inline) {
245                 return 1;
246         }
247
248         ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
249         if (ret)
250                 return ret;
251
252         if (isize > actual_end)
253                 inline_len = min_t(u64, isize, actual_end);
254         ret = insert_inline_extent(trans, root, inode, start,
255                                    inline_len, compressed_size,
256                                    compress_type, compressed_pages);
257         if (ret && ret != -ENOSPC) {
258                 btrfs_abort_transaction(trans, root, ret);
259                 return ret;
260         } else if (ret == -ENOSPC) {
261                 return 1;
262         }
263
264         btrfs_delalloc_release_metadata(inode, end + 1 - start);
265         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
266         return 0;
267 }
268
269 struct async_extent {
270         u64 start;
271         u64 ram_size;
272         u64 compressed_size;
273         struct page **pages;
274         unsigned long nr_pages;
275         int compress_type;
276         struct list_head list;
277 };
278
279 struct async_cow {
280         struct inode *inode;
281         struct btrfs_root *root;
282         struct page *locked_page;
283         u64 start;
284         u64 end;
285         struct list_head extents;
286         struct btrfs_work work;
287 };
288
289 static noinline int add_async_extent(struct async_cow *cow,
290                                      u64 start, u64 ram_size,
291                                      u64 compressed_size,
292                                      struct page **pages,
293                                      unsigned long nr_pages,
294                                      int compress_type)
295 {
296         struct async_extent *async_extent;
297
298         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
299         BUG_ON(!async_extent); /* -ENOMEM */
300         async_extent->start = start;
301         async_extent->ram_size = ram_size;
302         async_extent->compressed_size = compressed_size;
303         async_extent->pages = pages;
304         async_extent->nr_pages = nr_pages;
305         async_extent->compress_type = compress_type;
306         list_add_tail(&async_extent->list, &cow->extents);
307         return 0;
308 }
309
310 /*
311  * we create compressed extents in two phases.  The first
312  * phase compresses a range of pages that have already been
313  * locked (both pages and state bits are locked).
314  *
315  * This is done inside an ordered work queue, and the compression
316  * is spread across many cpus.  The actual IO submission is step
317  * two, and the ordered work queue takes care of making sure that
318  * happens in the same order things were put onto the queue by
319  * writepages and friends.
320  *
321  * If this code finds it can't get good compression, it puts an
322  * entry onto the work queue to write the uncompressed bytes.  This
323  * makes sure that both compressed inodes and uncompressed inodes
324  * are written in the same order that the flusher thread sent them
325  * down.
326  */
327 static noinline int compress_file_range(struct inode *inode,
328                                         struct page *locked_page,
329                                         u64 start, u64 end,
330                                         struct async_cow *async_cow,
331                                         int *num_added)
332 {
333         struct btrfs_root *root = BTRFS_I(inode)->root;
334         struct btrfs_trans_handle *trans;
335         u64 num_bytes;
336         u64 blocksize = root->sectorsize;
337         u64 actual_end;
338         u64 isize = i_size_read(inode);
339         int ret = 0;
340         struct page **pages = NULL;
341         unsigned long nr_pages;
342         unsigned long nr_pages_ret = 0;
343         unsigned long total_compressed = 0;
344         unsigned long total_in = 0;
345         unsigned long max_compressed = 128 * 1024;
346         unsigned long max_uncompressed = 128 * 1024;
347         int i;
348         int will_compress;
349         int compress_type = root->fs_info->compress_type;
350
351         /* if this is a small write inside eof, kick off a defrag */
352         if ((end - start + 1) < 16 * 1024 &&
353             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
354                 btrfs_add_inode_defrag(NULL, inode);
355
356         actual_end = min_t(u64, isize, end + 1);
357 again:
358         will_compress = 0;
359         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
360         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
361
362         /*
363          * we don't want to send crud past the end of i_size through
364          * compression, that's just a waste of CPU time.  So, if the
365          * end of the file is before the start of our current
366          * requested range of bytes, we bail out to the uncompressed
367          * cleanup code that can deal with all of this.
368          *
369          * It isn't really the fastest way to fix things, but this is a
370          * very uncommon corner.
371          */
372         if (actual_end <= start)
373                 goto cleanup_and_bail_uncompressed;
374
375         total_compressed = actual_end - start;
376
377         /* we want to make sure that amount of ram required to uncompress
378          * an extent is reasonable, so we limit the total size in ram
379          * of a compressed extent to 128k.  This is a crucial number
380          * because it also controls how easily we can spread reads across
381          * cpus for decompression.
382          *
383          * We also want to make sure the amount of IO required to do
384          * a random read is reasonably small, so we limit the size of
385          * a compressed extent to 128k.
386          */
387         total_compressed = min(total_compressed, max_uncompressed);
388         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
389         num_bytes = max(blocksize,  num_bytes);
390         total_in = 0;
391         ret = 0;
392
393         /*
394          * we do compression for mount -o compress and when the
395          * inode has not been flagged as nocompress.  This flag can
396          * change at any time if we discover bad compression ratios.
397          */
398         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
399             (btrfs_test_opt(root, COMPRESS) ||
400              (BTRFS_I(inode)->force_compress) ||
401              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
402                 WARN_ON(pages);
403                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
404                 if (!pages) {
405                         /* just bail out to the uncompressed code */
406                         goto cont;
407                 }
408
409                 if (BTRFS_I(inode)->force_compress)
410                         compress_type = BTRFS_I(inode)->force_compress;
411
412                 ret = btrfs_compress_pages(compress_type,
413                                            inode->i_mapping, start,
414                                            total_compressed, pages,
415                                            nr_pages, &nr_pages_ret,
416                                            &total_in,
417                                            &total_compressed,
418                                            max_compressed);
419
420                 if (!ret) {
421                         unsigned long offset = total_compressed &
422                                 (PAGE_CACHE_SIZE - 1);
423                         struct page *page = pages[nr_pages_ret - 1];
424                         char *kaddr;
425
426                         /* zero the tail end of the last page, we might be
427                          * sending it down to disk
428                          */
429                         if (offset) {
430                                 kaddr = kmap_atomic(page);
431                                 memset(kaddr + offset, 0,
432                                        PAGE_CACHE_SIZE - offset);
433                                 kunmap_atomic(kaddr);
434                         }
435                         will_compress = 1;
436                 }
437         }
438 cont:
439         if (start == 0) {
440                 trans = btrfs_join_transaction(root);
441                 if (IS_ERR(trans)) {
442                         ret = PTR_ERR(trans);
443                         trans = NULL;
444                         goto cleanup_and_out;
445                 }
446                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
447
448                 /* lets try to make an inline extent */
449                 if (ret || total_in < (actual_end - start)) {
450                         /* we didn't compress the entire range, try
451                          * to make an uncompressed inline extent.
452                          */
453                         ret = cow_file_range_inline(trans, root, inode,
454                                                     start, end, 0, 0, NULL);
455                 } else {
456                         /* try making a compressed inline extent */
457                         ret = cow_file_range_inline(trans, root, inode,
458                                                     start, end,
459                                                     total_compressed,
460                                                     compress_type, pages);
461                 }
462                 if (ret <= 0) {
463                         /*
464                          * inline extent creation worked or returned error,
465                          * we don't need to create any more async work items.
466                          * Unlock and free up our temp pages.
467                          */
468                         extent_clear_unlock_delalloc(inode,
469                              &BTRFS_I(inode)->io_tree,
470                              start, end, NULL,
471                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
472                              EXTENT_CLEAR_DELALLOC |
473                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
474
475                         btrfs_end_transaction(trans, root);
476                         goto free_pages_out;
477                 }
478                 btrfs_end_transaction(trans, root);
479         }
480
481         if (will_compress) {
482                 /*
483                  * we aren't doing an inline extent round the compressed size
484                  * up to a block size boundary so the allocator does sane
485                  * things
486                  */
487                 total_compressed = (total_compressed + blocksize - 1) &
488                         ~(blocksize - 1);
489
490                 /*
491                  * one last check to make sure the compression is really a
492                  * win, compare the page count read with the blocks on disk
493                  */
494                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
495                         ~(PAGE_CACHE_SIZE - 1);
496                 if (total_compressed >= total_in) {
497                         will_compress = 0;
498                 } else {
499                         num_bytes = total_in;
500                 }
501         }
502         if (!will_compress && pages) {
503                 /*
504                  * the compression code ran but failed to make things smaller,
505                  * free any pages it allocated and our page pointer array
506                  */
507                 for (i = 0; i < nr_pages_ret; i++) {
508                         WARN_ON(pages[i]->mapping);
509                         page_cache_release(pages[i]);
510                 }
511                 kfree(pages);
512                 pages = NULL;
513                 total_compressed = 0;
514                 nr_pages_ret = 0;
515
516                 /* flag the file so we don't compress in the future */
517                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
518                     !(BTRFS_I(inode)->force_compress)) {
519                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
520                 }
521         }
522         if (will_compress) {
523                 *num_added += 1;
524
525                 /* the async work queues will take care of doing actual
526                  * allocation on disk for these compressed pages,
527                  * and will submit them to the elevator.
528                  */
529                 add_async_extent(async_cow, start, num_bytes,
530                                  total_compressed, pages, nr_pages_ret,
531                                  compress_type);
532
533                 if (start + num_bytes < end) {
534                         start += num_bytes;
535                         pages = NULL;
536                         cond_resched();
537                         goto again;
538                 }
539         } else {
540 cleanup_and_bail_uncompressed:
541                 /*
542                  * No compression, but we still need to write the pages in
543                  * the file we've been given so far.  redirty the locked
544                  * page if it corresponds to our extent and set things up
545                  * for the async work queue to run cow_file_range to do
546                  * the normal delalloc dance
547                  */
548                 if (page_offset(locked_page) >= start &&
549                     page_offset(locked_page) <= end) {
550                         __set_page_dirty_nobuffers(locked_page);
551                         /* unlocked later on in the async handlers */
552                 }
553                 add_async_extent(async_cow, start, end - start + 1,
554                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
555                 *num_added += 1;
556         }
557
558 out:
559         return ret;
560
561 free_pages_out:
562         for (i = 0; i < nr_pages_ret; i++) {
563                 WARN_ON(pages[i]->mapping);
564                 page_cache_release(pages[i]);
565         }
566         kfree(pages);
567
568         goto out;
569
570 cleanup_and_out:
571         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
572                                      start, end, NULL,
573                                      EXTENT_CLEAR_UNLOCK_PAGE |
574                                      EXTENT_CLEAR_DIRTY |
575                                      EXTENT_CLEAR_DELALLOC |
576                                      EXTENT_SET_WRITEBACK |
577                                      EXTENT_END_WRITEBACK);
578         if (!trans || IS_ERR(trans))
579                 btrfs_error(root->fs_info, ret, "Failed to join transaction");
580         else
581                 btrfs_abort_transaction(trans, root, ret);
582         goto free_pages_out;
583 }
584
585 /*
586  * phase two of compressed writeback.  This is the ordered portion
587  * of the code, which only gets called in the order the work was
588  * queued.  We walk all the async extents created by compress_file_range
589  * and send them down to the disk.
590  */
591 static noinline int submit_compressed_extents(struct inode *inode,
592                                               struct async_cow *async_cow)
593 {
594         struct async_extent *async_extent;
595         u64 alloc_hint = 0;
596         struct btrfs_trans_handle *trans;
597         struct btrfs_key ins;
598         struct extent_map *em;
599         struct btrfs_root *root = BTRFS_I(inode)->root;
600         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
601         struct extent_io_tree *io_tree;
602         int ret = 0;
603
604         if (list_empty(&async_cow->extents))
605                 return 0;
606
607
608         while (!list_empty(&async_cow->extents)) {
609                 async_extent = list_entry(async_cow->extents.next,
610                                           struct async_extent, list);
611                 list_del(&async_extent->list);
612
613                 io_tree = &BTRFS_I(inode)->io_tree;
614
615 retry:
616                 /* did the compression code fall back to uncompressed IO? */
617                 if (!async_extent->pages) {
618                         int page_started = 0;
619                         unsigned long nr_written = 0;
620
621                         lock_extent(io_tree, async_extent->start,
622                                          async_extent->start +
623                                          async_extent->ram_size - 1);
624
625                         /* allocate blocks */
626                         ret = cow_file_range(inode, async_cow->locked_page,
627                                              async_extent->start,
628                                              async_extent->start +
629                                              async_extent->ram_size - 1,
630                                              &page_started, &nr_written, 0);
631
632                         /* JDM XXX */
633
634                         /*
635                          * if page_started, cow_file_range inserted an
636                          * inline extent and took care of all the unlocking
637                          * and IO for us.  Otherwise, we need to submit
638                          * all those pages down to the drive.
639                          */
640                         if (!page_started && !ret)
641                                 extent_write_locked_range(io_tree,
642                                                   inode, async_extent->start,
643                                                   async_extent->start +
644                                                   async_extent->ram_size - 1,
645                                                   btrfs_get_extent,
646                                                   WB_SYNC_ALL);
647                         kfree(async_extent);
648                         cond_resched();
649                         continue;
650                 }
651
652                 lock_extent(io_tree, async_extent->start,
653                             async_extent->start + async_extent->ram_size - 1);
654
655                 trans = btrfs_join_transaction(root);
656                 if (IS_ERR(trans)) {
657                         ret = PTR_ERR(trans);
658                 } else {
659                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
660                         ret = btrfs_reserve_extent(trans, root,
661                                            async_extent->compressed_size,
662                                            async_extent->compressed_size,
663                                            0, alloc_hint, &ins, 1);
664                         if (ret && ret != -ENOSPC)
665                                 btrfs_abort_transaction(trans, root, ret);
666                         btrfs_end_transaction(trans, root);
667                 }
668
669                 if (ret) {
670                         int i;
671                         for (i = 0; i < async_extent->nr_pages; i++) {
672                                 WARN_ON(async_extent->pages[i]->mapping);
673                                 page_cache_release(async_extent->pages[i]);
674                         }
675                         kfree(async_extent->pages);
676                         async_extent->nr_pages = 0;
677                         async_extent->pages = NULL;
678                         unlock_extent(io_tree, async_extent->start,
679                                       async_extent->start +
680                                       async_extent->ram_size - 1);
681                         if (ret == -ENOSPC)
682                                 goto retry;
683                         goto out_free; /* JDM: Requeue? */
684                 }
685
686                 /*
687                  * here we're doing allocation and writeback of the
688                  * compressed pages
689                  */
690                 btrfs_drop_extent_cache(inode, async_extent->start,
691                                         async_extent->start +
692                                         async_extent->ram_size - 1, 0);
693
694                 em = alloc_extent_map();
695                 BUG_ON(!em); /* -ENOMEM */
696                 em->start = async_extent->start;
697                 em->len = async_extent->ram_size;
698                 em->orig_start = em->start;
699
700                 em->block_start = ins.objectid;
701                 em->block_len = ins.offset;
702                 em->bdev = root->fs_info->fs_devices->latest_bdev;
703                 em->compress_type = async_extent->compress_type;
704                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
705                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
706
707                 while (1) {
708                         write_lock(&em_tree->lock);
709                         ret = add_extent_mapping(em_tree, em);
710                         write_unlock(&em_tree->lock);
711                         if (ret != -EEXIST) {
712                                 free_extent_map(em);
713                                 break;
714                         }
715                         btrfs_drop_extent_cache(inode, async_extent->start,
716                                                 async_extent->start +
717                                                 async_extent->ram_size - 1, 0);
718                 }
719
720                 ret = btrfs_add_ordered_extent_compress(inode,
721                                                 async_extent->start,
722                                                 ins.objectid,
723                                                 async_extent->ram_size,
724                                                 ins.offset,
725                                                 BTRFS_ORDERED_COMPRESSED,
726                                                 async_extent->compress_type);
727                 BUG_ON(ret); /* -ENOMEM */
728
729                 /*
730                  * clear dirty, set writeback and unlock the pages.
731                  */
732                 extent_clear_unlock_delalloc(inode,
733                                 &BTRFS_I(inode)->io_tree,
734                                 async_extent->start,
735                                 async_extent->start +
736                                 async_extent->ram_size - 1,
737                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
738                                 EXTENT_CLEAR_UNLOCK |
739                                 EXTENT_CLEAR_DELALLOC |
740                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
741
742                 ret = btrfs_submit_compressed_write(inode,
743                                     async_extent->start,
744                                     async_extent->ram_size,
745                                     ins.objectid,
746                                     ins.offset, async_extent->pages,
747                                     async_extent->nr_pages);
748
749                 BUG_ON(ret); /* -ENOMEM */
750                 alloc_hint = ins.objectid + ins.offset;
751                 kfree(async_extent);
752                 cond_resched();
753         }
754         ret = 0;
755 out:
756         return ret;
757 out_free:
758         kfree(async_extent);
759         goto out;
760 }
761
762 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
763                                       u64 num_bytes)
764 {
765         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
766         struct extent_map *em;
767         u64 alloc_hint = 0;
768
769         read_lock(&em_tree->lock);
770         em = search_extent_mapping(em_tree, start, num_bytes);
771         if (em) {
772                 /*
773                  * if block start isn't an actual block number then find the
774                  * first block in this inode and use that as a hint.  If that
775                  * block is also bogus then just don't worry about it.
776                  */
777                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
778                         free_extent_map(em);
779                         em = search_extent_mapping(em_tree, 0, 0);
780                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
781                                 alloc_hint = em->block_start;
782                         if (em)
783                                 free_extent_map(em);
784                 } else {
785                         alloc_hint = em->block_start;
786                         free_extent_map(em);
787                 }
788         }
789         read_unlock(&em_tree->lock);
790
791         return alloc_hint;
792 }
793
794 /*
795  * when extent_io.c finds a delayed allocation range in the file,
796  * the call backs end up in this code.  The basic idea is to
797  * allocate extents on disk for the range, and create ordered data structs
798  * in ram to track those extents.
799  *
800  * locked_page is the page that writepage had locked already.  We use
801  * it to make sure we don't do extra locks or unlocks.
802  *
803  * *page_started is set to one if we unlock locked_page and do everything
804  * required to start IO on it.  It may be clean and already done with
805  * IO when we return.
806  */
807 static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
808                                      struct inode *inode,
809                                      struct btrfs_root *root,
810                                      struct page *locked_page,
811                                      u64 start, u64 end, int *page_started,
812                                      unsigned long *nr_written,
813                                      int unlock)
814 {
815         u64 alloc_hint = 0;
816         u64 num_bytes;
817         unsigned long ram_size;
818         u64 disk_num_bytes;
819         u64 cur_alloc_size;
820         u64 blocksize = root->sectorsize;
821         struct btrfs_key ins;
822         struct extent_map *em;
823         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
824         int ret = 0;
825
826         BUG_ON(btrfs_is_free_space_inode(inode));
827
828         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
829         num_bytes = max(blocksize,  num_bytes);
830         disk_num_bytes = num_bytes;
831
832         /* if this is a small write inside eof, kick off defrag */
833         if (num_bytes < 64 * 1024 &&
834             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
835                 btrfs_add_inode_defrag(trans, inode);
836
837         if (start == 0) {
838                 /* lets try to make an inline extent */
839                 ret = cow_file_range_inline(trans, root, inode,
840                                             start, end, 0, 0, NULL);
841                 if (ret == 0) {
842                         extent_clear_unlock_delalloc(inode,
843                                      &BTRFS_I(inode)->io_tree,
844                                      start, end, NULL,
845                                      EXTENT_CLEAR_UNLOCK_PAGE |
846                                      EXTENT_CLEAR_UNLOCK |
847                                      EXTENT_CLEAR_DELALLOC |
848                                      EXTENT_CLEAR_DIRTY |
849                                      EXTENT_SET_WRITEBACK |
850                                      EXTENT_END_WRITEBACK);
851
852                         *nr_written = *nr_written +
853                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
854                         *page_started = 1;
855                         goto out;
856                 } else if (ret < 0) {
857                         btrfs_abort_transaction(trans, root, ret);
858                         goto out_unlock;
859                 }
860         }
861
862         BUG_ON(disk_num_bytes >
863                btrfs_super_total_bytes(root->fs_info->super_copy));
864
865         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
866         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
867
868         while (disk_num_bytes > 0) {
869                 unsigned long op;
870
871                 cur_alloc_size = disk_num_bytes;
872                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
873                                            root->sectorsize, 0, alloc_hint,
874                                            &ins, 1);
875                 if (ret < 0) {
876                         btrfs_abort_transaction(trans, root, ret);
877                         goto out_unlock;
878                 }
879
880                 em = alloc_extent_map();
881                 BUG_ON(!em); /* -ENOMEM */
882                 em->start = start;
883                 em->orig_start = em->start;
884                 ram_size = ins.offset;
885                 em->len = ins.offset;
886
887                 em->block_start = ins.objectid;
888                 em->block_len = ins.offset;
889                 em->bdev = root->fs_info->fs_devices->latest_bdev;
890                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
891
892                 while (1) {
893                         write_lock(&em_tree->lock);
894                         ret = add_extent_mapping(em_tree, em);
895                         write_unlock(&em_tree->lock);
896                         if (ret != -EEXIST) {
897                                 free_extent_map(em);
898                                 break;
899                         }
900                         btrfs_drop_extent_cache(inode, start,
901                                                 start + ram_size - 1, 0);
902                 }
903
904                 cur_alloc_size = ins.offset;
905                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
906                                                ram_size, cur_alloc_size, 0);
907                 BUG_ON(ret); /* -ENOMEM */
908
909                 if (root->root_key.objectid ==
910                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
911                         ret = btrfs_reloc_clone_csums(inode, start,
912                                                       cur_alloc_size);
913                         if (ret) {
914                                 btrfs_abort_transaction(trans, root, ret);
915                                 goto out_unlock;
916                         }
917                 }
918
919                 if (disk_num_bytes < cur_alloc_size)
920                         break;
921
922                 /* we're not doing compressed IO, don't unlock the first
923                  * page (which the caller expects to stay locked), don't
924                  * clear any dirty bits and don't set any writeback bits
925                  *
926                  * Do set the Private2 bit so we know this page was properly
927                  * setup for writepage
928                  */
929                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
930                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
931                         EXTENT_SET_PRIVATE2;
932
933                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
934                                              start, start + ram_size - 1,
935                                              locked_page, op);
936                 disk_num_bytes -= cur_alloc_size;
937                 num_bytes -= cur_alloc_size;
938                 alloc_hint = ins.objectid + ins.offset;
939                 start += cur_alloc_size;
940         }
941 out:
942         return ret;
943
944 out_unlock:
945         extent_clear_unlock_delalloc(inode,
946                      &BTRFS_I(inode)->io_tree,
947                      start, end, locked_page,
948                      EXTENT_CLEAR_UNLOCK_PAGE |
949                      EXTENT_CLEAR_UNLOCK |
950                      EXTENT_CLEAR_DELALLOC |
951                      EXTENT_CLEAR_DIRTY |
952                      EXTENT_SET_WRITEBACK |
953                      EXTENT_END_WRITEBACK);
954
955         goto out;
956 }
957
958 static noinline int cow_file_range(struct inode *inode,
959                                    struct page *locked_page,
960                                    u64 start, u64 end, int *page_started,
961                                    unsigned long *nr_written,
962                                    int unlock)
963 {
964         struct btrfs_trans_handle *trans;
965         struct btrfs_root *root = BTRFS_I(inode)->root;
966         int ret;
967
968         trans = btrfs_join_transaction(root);
969         if (IS_ERR(trans)) {
970                 extent_clear_unlock_delalloc(inode,
971                              &BTRFS_I(inode)->io_tree,
972                              start, end, locked_page,
973                              EXTENT_CLEAR_UNLOCK_PAGE |
974                              EXTENT_CLEAR_UNLOCK |
975                              EXTENT_CLEAR_DELALLOC |
976                              EXTENT_CLEAR_DIRTY |
977                              EXTENT_SET_WRITEBACK |
978                              EXTENT_END_WRITEBACK);
979                 return PTR_ERR(trans);
980         }
981         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
982
983         ret = __cow_file_range(trans, inode, root, locked_page, start, end,
984                                page_started, nr_written, unlock);
985
986         btrfs_end_transaction(trans, root);
987
988         return ret;
989 }
990
991 /*
992  * work queue call back to started compression on a file and pages
993  */
994 static noinline void async_cow_start(struct btrfs_work *work)
995 {
996         struct async_cow *async_cow;
997         int num_added = 0;
998         async_cow = container_of(work, struct async_cow, work);
999
1000         compress_file_range(async_cow->inode, async_cow->locked_page,
1001                             async_cow->start, async_cow->end, async_cow,
1002                             &num_added);
1003         if (num_added == 0) {
1004                 btrfs_add_delayed_iput(async_cow->inode);
1005                 async_cow->inode = NULL;
1006         }
1007 }
1008
1009 /*
1010  * work queue call back to submit previously compressed pages
1011  */
1012 static noinline void async_cow_submit(struct btrfs_work *work)
1013 {
1014         struct async_cow *async_cow;
1015         struct btrfs_root *root;
1016         unsigned long nr_pages;
1017
1018         async_cow = container_of(work, struct async_cow, work);
1019
1020         root = async_cow->root;
1021         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1022                 PAGE_CACHE_SHIFT;
1023
1024         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1025             5 * 1024 * 1024 &&
1026             waitqueue_active(&root->fs_info->async_submit_wait))
1027                 wake_up(&root->fs_info->async_submit_wait);
1028
1029         if (async_cow->inode)
1030                 submit_compressed_extents(async_cow->inode, async_cow);
1031 }
1032
1033 static noinline void async_cow_free(struct btrfs_work *work)
1034 {
1035         struct async_cow *async_cow;
1036         async_cow = container_of(work, struct async_cow, work);
1037         if (async_cow->inode)
1038                 btrfs_add_delayed_iput(async_cow->inode);
1039         kfree(async_cow);
1040 }
1041
1042 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1043                                 u64 start, u64 end, int *page_started,
1044                                 unsigned long *nr_written)
1045 {
1046         struct async_cow *async_cow;
1047         struct btrfs_root *root = BTRFS_I(inode)->root;
1048         unsigned long nr_pages;
1049         u64 cur_end;
1050         int limit = 10 * 1024 * 1024;
1051
1052         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1053                          1, 0, NULL, GFP_NOFS);
1054         while (start < end) {
1055                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1056                 BUG_ON(!async_cow); /* -ENOMEM */
1057                 async_cow->inode = igrab(inode);
1058                 async_cow->root = root;
1059                 async_cow->locked_page = locked_page;
1060                 async_cow->start = start;
1061
1062                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1063                         cur_end = end;
1064                 else
1065                         cur_end = min(end, start + 512 * 1024 - 1);
1066
1067                 async_cow->end = cur_end;
1068                 INIT_LIST_HEAD(&async_cow->extents);
1069
1070                 async_cow->work.func = async_cow_start;
1071                 async_cow->work.ordered_func = async_cow_submit;
1072                 async_cow->work.ordered_free = async_cow_free;
1073                 async_cow->work.flags = 0;
1074
1075                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1076                         PAGE_CACHE_SHIFT;
1077                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1078
1079                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1080                                    &async_cow->work);
1081
1082                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1083                         wait_event(root->fs_info->async_submit_wait,
1084                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1085                             limit));
1086                 }
1087
1088                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1089                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1090                         wait_event(root->fs_info->async_submit_wait,
1091                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1092                            0));
1093                 }
1094
1095                 *nr_written += nr_pages;
1096                 start = cur_end + 1;
1097         }
1098         *page_started = 1;
1099         return 0;
1100 }
1101
1102 static noinline int csum_exist_in_range(struct btrfs_root *root,
1103                                         u64 bytenr, u64 num_bytes)
1104 {
1105         int ret;
1106         struct btrfs_ordered_sum *sums;
1107         LIST_HEAD(list);
1108
1109         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1110                                        bytenr + num_bytes - 1, &list, 0);
1111         if (ret == 0 && list_empty(&list))
1112                 return 0;
1113
1114         while (!list_empty(&list)) {
1115                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1116                 list_del(&sums->list);
1117                 kfree(sums);
1118         }
1119         return 1;
1120 }
1121
1122 /*
1123  * when nowcow writeback call back.  This checks for snapshots or COW copies
1124  * of the extents that exist in the file, and COWs the file as required.
1125  *
1126  * If no cow copies or snapshots exist, we write directly to the existing
1127  * blocks on disk
1128  */
1129 static noinline int run_delalloc_nocow(struct inode *inode,
1130                                        struct page *locked_page,
1131                               u64 start, u64 end, int *page_started, int force,
1132                               unsigned long *nr_written)
1133 {
1134         struct btrfs_root *root = BTRFS_I(inode)->root;
1135         struct btrfs_trans_handle *trans;
1136         struct extent_buffer *leaf;
1137         struct btrfs_path *path;
1138         struct btrfs_file_extent_item *fi;
1139         struct btrfs_key found_key;
1140         u64 cow_start;
1141         u64 cur_offset;
1142         u64 extent_end;
1143         u64 extent_offset;
1144         u64 disk_bytenr;
1145         u64 num_bytes;
1146         int extent_type;
1147         int ret, err;
1148         int type;
1149         int nocow;
1150         int check_prev = 1;
1151         bool nolock;
1152         u64 ino = btrfs_ino(inode);
1153
1154         path = btrfs_alloc_path();
1155         if (!path) {
1156                 extent_clear_unlock_delalloc(inode,
1157                              &BTRFS_I(inode)->io_tree,
1158                              start, end, locked_page,
1159                              EXTENT_CLEAR_UNLOCK_PAGE |
1160                              EXTENT_CLEAR_UNLOCK |
1161                              EXTENT_CLEAR_DELALLOC |
1162                              EXTENT_CLEAR_DIRTY |
1163                              EXTENT_SET_WRITEBACK |
1164                              EXTENT_END_WRITEBACK);
1165                 return -ENOMEM;
1166         }
1167
1168         nolock = btrfs_is_free_space_inode(inode);
1169
1170         if (nolock)
1171                 trans = btrfs_join_transaction_nolock(root);
1172         else
1173                 trans = btrfs_join_transaction(root);
1174
1175         if (IS_ERR(trans)) {
1176                 extent_clear_unlock_delalloc(inode,
1177                              &BTRFS_I(inode)->io_tree,
1178                              start, end, locked_page,
1179                              EXTENT_CLEAR_UNLOCK_PAGE |
1180                              EXTENT_CLEAR_UNLOCK |
1181                              EXTENT_CLEAR_DELALLOC |
1182                              EXTENT_CLEAR_DIRTY |
1183                              EXTENT_SET_WRITEBACK |
1184                              EXTENT_END_WRITEBACK);
1185                 btrfs_free_path(path);
1186                 return PTR_ERR(trans);
1187         }
1188
1189         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1190
1191         cow_start = (u64)-1;
1192         cur_offset = start;
1193         while (1) {
1194                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1195                                                cur_offset, 0);
1196                 if (ret < 0) {
1197                         btrfs_abort_transaction(trans, root, ret);
1198                         goto error;
1199                 }
1200                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1201                         leaf = path->nodes[0];
1202                         btrfs_item_key_to_cpu(leaf, &found_key,
1203                                               path->slots[0] - 1);
1204                         if (found_key.objectid == ino &&
1205                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1206                                 path->slots[0]--;
1207                 }
1208                 check_prev = 0;
1209 next_slot:
1210                 leaf = path->nodes[0];
1211                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1212                         ret = btrfs_next_leaf(root, path);
1213                         if (ret < 0) {
1214                                 btrfs_abort_transaction(trans, root, ret);
1215                                 goto error;
1216                         }
1217                         if (ret > 0)
1218                                 break;
1219                         leaf = path->nodes[0];
1220                 }
1221
1222                 nocow = 0;
1223                 disk_bytenr = 0;
1224                 num_bytes = 0;
1225                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1226
1227                 if (found_key.objectid > ino ||
1228                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1229                     found_key.offset > end)
1230                         break;
1231
1232                 if (found_key.offset > cur_offset) {
1233                         extent_end = found_key.offset;
1234                         extent_type = 0;
1235                         goto out_check;
1236                 }
1237
1238                 fi = btrfs_item_ptr(leaf, path->slots[0],
1239                                     struct btrfs_file_extent_item);
1240                 extent_type = btrfs_file_extent_type(leaf, fi);
1241
1242                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1243                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1244                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1245                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1246                         extent_end = found_key.offset +
1247                                 btrfs_file_extent_num_bytes(leaf, fi);
1248                         if (extent_end <= start) {
1249                                 path->slots[0]++;
1250                                 goto next_slot;
1251                         }
1252                         if (disk_bytenr == 0)
1253                                 goto out_check;
1254                         if (btrfs_file_extent_compression(leaf, fi) ||
1255                             btrfs_file_extent_encryption(leaf, fi) ||
1256                             btrfs_file_extent_other_encoding(leaf, fi))
1257                                 goto out_check;
1258                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1259                                 goto out_check;
1260                         if (btrfs_extent_readonly(root, disk_bytenr))
1261                                 goto out_check;
1262                         if (btrfs_cross_ref_exist(trans, root, ino,
1263                                                   found_key.offset -
1264                                                   extent_offset, disk_bytenr))
1265                                 goto out_check;
1266                         disk_bytenr += extent_offset;
1267                         disk_bytenr += cur_offset - found_key.offset;
1268                         num_bytes = min(end + 1, extent_end) - cur_offset;
1269                         /*
1270                          * force cow if csum exists in the range.
1271                          * this ensure that csum for a given extent are
1272                          * either valid or do not exist.
1273                          */
1274                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1275                                 goto out_check;
1276                         nocow = 1;
1277                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1278                         extent_end = found_key.offset +
1279                                 btrfs_file_extent_inline_len(leaf, fi);
1280                         extent_end = ALIGN(extent_end, root->sectorsize);
1281                 } else {
1282                         BUG_ON(1);
1283                 }
1284 out_check:
1285                 if (extent_end <= start) {
1286                         path->slots[0]++;
1287                         goto next_slot;
1288                 }
1289                 if (!nocow) {
1290                         if (cow_start == (u64)-1)
1291                                 cow_start = cur_offset;
1292                         cur_offset = extent_end;
1293                         if (cur_offset > end)
1294                                 break;
1295                         path->slots[0]++;
1296                         goto next_slot;
1297                 }
1298
1299                 btrfs_release_path(path);
1300                 if (cow_start != (u64)-1) {
1301                         ret = __cow_file_range(trans, inode, root, locked_page,
1302                                                cow_start, found_key.offset - 1,
1303                                                page_started, nr_written, 1);
1304                         if (ret) {
1305                                 btrfs_abort_transaction(trans, root, ret);
1306                                 goto error;
1307                         }
1308                         cow_start = (u64)-1;
1309                 }
1310
1311                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1312                         struct extent_map *em;
1313                         struct extent_map_tree *em_tree;
1314                         em_tree = &BTRFS_I(inode)->extent_tree;
1315                         em = alloc_extent_map();
1316                         BUG_ON(!em); /* -ENOMEM */
1317                         em->start = cur_offset;
1318                         em->orig_start = em->start;
1319                         em->len = num_bytes;
1320                         em->block_len = num_bytes;
1321                         em->block_start = disk_bytenr;
1322                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1323                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1324                         set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
1325                         while (1) {
1326                                 write_lock(&em_tree->lock);
1327                                 ret = add_extent_mapping(em_tree, em);
1328                                 write_unlock(&em_tree->lock);
1329                                 if (ret != -EEXIST) {
1330                                         free_extent_map(em);
1331                                         break;
1332                                 }
1333                                 btrfs_drop_extent_cache(inode, em->start,
1334                                                 em->start + em->len - 1, 0);
1335                         }
1336                         type = BTRFS_ORDERED_PREALLOC;
1337                 } else {
1338                         type = BTRFS_ORDERED_NOCOW;
1339                 }
1340
1341                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1342                                                num_bytes, num_bytes, type);
1343                 BUG_ON(ret); /* -ENOMEM */
1344
1345                 if (root->root_key.objectid ==
1346                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1347                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1348                                                       num_bytes);
1349                         if (ret) {
1350                                 btrfs_abort_transaction(trans, root, ret);
1351                                 goto error;
1352                         }
1353                 }
1354
1355                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1356                                 cur_offset, cur_offset + num_bytes - 1,
1357                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1358                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1359                                 EXTENT_SET_PRIVATE2);
1360                 cur_offset = extent_end;
1361                 if (cur_offset > end)
1362                         break;
1363         }
1364         btrfs_release_path(path);
1365
1366         if (cur_offset <= end && cow_start == (u64)-1) {
1367                 cow_start = cur_offset;
1368                 cur_offset = end;
1369         }
1370
1371         if (cow_start != (u64)-1) {
1372                 ret = __cow_file_range(trans, inode, root, locked_page,
1373                                        cow_start, end,
1374                                        page_started, nr_written, 1);
1375                 if (ret) {
1376                         btrfs_abort_transaction(trans, root, ret);
1377                         goto error;
1378                 }
1379         }
1380
1381 error:
1382         err = btrfs_end_transaction(trans, root);
1383         if (!ret)
1384                 ret = err;
1385
1386         if (ret && cur_offset < end)
1387                 extent_clear_unlock_delalloc(inode,
1388                              &BTRFS_I(inode)->io_tree,
1389                              cur_offset, end, locked_page,
1390                              EXTENT_CLEAR_UNLOCK_PAGE |
1391                              EXTENT_CLEAR_UNLOCK |
1392                              EXTENT_CLEAR_DELALLOC |
1393                              EXTENT_CLEAR_DIRTY |
1394                              EXTENT_SET_WRITEBACK |
1395                              EXTENT_END_WRITEBACK);
1396
1397         btrfs_free_path(path);
1398         return ret;
1399 }
1400
1401 /*
1402  * extent_io.c call back to do delayed allocation processing
1403  */
1404 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1405                               u64 start, u64 end, int *page_started,
1406                               unsigned long *nr_written)
1407 {
1408         int ret;
1409         struct btrfs_root *root = BTRFS_I(inode)->root;
1410
1411         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1412                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1413                                          page_started, 1, nr_written);
1414         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1415                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1416                                          page_started, 0, nr_written);
1417         } else if (!btrfs_test_opt(root, COMPRESS) &&
1418                    !(BTRFS_I(inode)->force_compress) &&
1419                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1420                 ret = cow_file_range(inode, locked_page, start, end,
1421                                       page_started, nr_written, 1);
1422         } else {
1423                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1424                         &BTRFS_I(inode)->runtime_flags);
1425                 ret = cow_file_range_async(inode, locked_page, start, end,
1426                                            page_started, nr_written);
1427         }
1428         return ret;
1429 }
1430
1431 static void btrfs_split_extent_hook(struct inode *inode,
1432                                     struct extent_state *orig, u64 split)
1433 {
1434         /* not delalloc, ignore it */
1435         if (!(orig->state & EXTENT_DELALLOC))
1436                 return;
1437
1438         spin_lock(&BTRFS_I(inode)->lock);
1439         BTRFS_I(inode)->outstanding_extents++;
1440         spin_unlock(&BTRFS_I(inode)->lock);
1441 }
1442
1443 /*
1444  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1445  * extents so we can keep track of new extents that are just merged onto old
1446  * extents, such as when we are doing sequential writes, so we can properly
1447  * account for the metadata space we'll need.
1448  */
1449 static void btrfs_merge_extent_hook(struct inode *inode,
1450                                     struct extent_state *new,
1451                                     struct extent_state *other)
1452 {
1453         /* not delalloc, ignore it */
1454         if (!(other->state & EXTENT_DELALLOC))
1455                 return;
1456
1457         spin_lock(&BTRFS_I(inode)->lock);
1458         BTRFS_I(inode)->outstanding_extents--;
1459         spin_unlock(&BTRFS_I(inode)->lock);
1460 }
1461
1462 /*
1463  * extent_io.c set_bit_hook, used to track delayed allocation
1464  * bytes in this file, and to maintain the list of inodes that
1465  * have pending delalloc work to be done.
1466  */
1467 static void btrfs_set_bit_hook(struct inode *inode,
1468                                struct extent_state *state, int *bits)
1469 {
1470
1471         /*
1472          * set_bit and clear bit hooks normally require _irqsave/restore
1473          * but in this case, we are only testing for the DELALLOC
1474          * bit, which is only set or cleared with irqs on
1475          */
1476         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1477                 struct btrfs_root *root = BTRFS_I(inode)->root;
1478                 u64 len = state->end + 1 - state->start;
1479                 bool do_list = !btrfs_is_free_space_inode(inode);
1480
1481                 if (*bits & EXTENT_FIRST_DELALLOC) {
1482                         *bits &= ~EXTENT_FIRST_DELALLOC;
1483                 } else {
1484                         spin_lock(&BTRFS_I(inode)->lock);
1485                         BTRFS_I(inode)->outstanding_extents++;
1486                         spin_unlock(&BTRFS_I(inode)->lock);
1487                 }
1488
1489                 spin_lock(&root->fs_info->delalloc_lock);
1490                 BTRFS_I(inode)->delalloc_bytes += len;
1491                 root->fs_info->delalloc_bytes += len;
1492                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1493                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1494                                       &root->fs_info->delalloc_inodes);
1495                 }
1496                 spin_unlock(&root->fs_info->delalloc_lock);
1497         }
1498 }
1499
1500 /*
1501  * extent_io.c clear_bit_hook, see set_bit_hook for why
1502  */
1503 static void btrfs_clear_bit_hook(struct inode *inode,
1504                                  struct extent_state *state, int *bits)
1505 {
1506         /*
1507          * set_bit and clear bit hooks normally require _irqsave/restore
1508          * but in this case, we are only testing for the DELALLOC
1509          * bit, which is only set or cleared with irqs on
1510          */
1511         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1512                 struct btrfs_root *root = BTRFS_I(inode)->root;
1513                 u64 len = state->end + 1 - state->start;
1514                 bool do_list = !btrfs_is_free_space_inode(inode);
1515
1516                 if (*bits & EXTENT_FIRST_DELALLOC) {
1517                         *bits &= ~EXTENT_FIRST_DELALLOC;
1518                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1519                         spin_lock(&BTRFS_I(inode)->lock);
1520                         BTRFS_I(inode)->outstanding_extents--;
1521                         spin_unlock(&BTRFS_I(inode)->lock);
1522                 }
1523
1524                 if (*bits & EXTENT_DO_ACCOUNTING)
1525                         btrfs_delalloc_release_metadata(inode, len);
1526
1527                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1528                     && do_list)
1529                         btrfs_free_reserved_data_space(inode, len);
1530
1531                 spin_lock(&root->fs_info->delalloc_lock);
1532                 root->fs_info->delalloc_bytes -= len;
1533                 BTRFS_I(inode)->delalloc_bytes -= len;
1534
1535                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1536                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1537                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1538                 }
1539                 spin_unlock(&root->fs_info->delalloc_lock);
1540         }
1541 }
1542
1543 /*
1544  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1545  * we don't create bios that span stripes or chunks
1546  */
1547 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1548                          size_t size, struct bio *bio,
1549                          unsigned long bio_flags)
1550 {
1551         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1552         u64 logical = (u64)bio->bi_sector << 9;
1553         u64 length = 0;
1554         u64 map_length;
1555         int ret;
1556
1557         if (bio_flags & EXTENT_BIO_COMPRESSED)
1558                 return 0;
1559
1560         length = bio->bi_size;
1561         map_length = length;
1562         ret = btrfs_map_block(root->fs_info, READ, logical,
1563                               &map_length, NULL, 0);
1564         /* Will always return 0 with map_multi == NULL */
1565         BUG_ON(ret < 0);
1566         if (map_length < length + size)
1567                 return 1;
1568         return 0;
1569 }
1570
1571 /*
1572  * in order to insert checksums into the metadata in large chunks,
1573  * we wait until bio submission time.   All the pages in the bio are
1574  * checksummed and sums are attached onto the ordered extent record.
1575  *
1576  * At IO completion time the cums attached on the ordered extent record
1577  * are inserted into the btree
1578  */
1579 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1580                                     struct bio *bio, int mirror_num,
1581                                     unsigned long bio_flags,
1582                                     u64 bio_offset)
1583 {
1584         struct btrfs_root *root = BTRFS_I(inode)->root;
1585         int ret = 0;
1586
1587         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1588         BUG_ON(ret); /* -ENOMEM */
1589         return 0;
1590 }
1591
1592 /*
1593  * in order to insert checksums into the metadata in large chunks,
1594  * we wait until bio submission time.   All the pages in the bio are
1595  * checksummed and sums are attached onto the ordered extent record.
1596  *
1597  * At IO completion time the cums attached on the ordered extent record
1598  * are inserted into the btree
1599  */
1600 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1601                           int mirror_num, unsigned long bio_flags,
1602                           u64 bio_offset)
1603 {
1604         struct btrfs_root *root = BTRFS_I(inode)->root;
1605         int ret;
1606
1607         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1608         if (ret)
1609                 bio_endio(bio, ret);
1610         return ret;
1611 }
1612
1613 /*
1614  * extent_io.c submission hook. This does the right thing for csum calculation
1615  * on write, or reading the csums from the tree before a read
1616  */
1617 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1618                           int mirror_num, unsigned long bio_flags,
1619                           u64 bio_offset)
1620 {
1621         struct btrfs_root *root = BTRFS_I(inode)->root;
1622         int ret = 0;
1623         int skip_sum;
1624         int metadata = 0;
1625
1626         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1627
1628         if (btrfs_is_free_space_inode(inode))
1629                 metadata = 2;
1630
1631         if (!(rw & REQ_WRITE)) {
1632                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1633                 if (ret)
1634                         goto out;
1635
1636                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1637                         ret = btrfs_submit_compressed_read(inode, bio,
1638                                                            mirror_num,
1639                                                            bio_flags);
1640                         goto out;
1641                 } else if (!skip_sum) {
1642                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1643                         if (ret)
1644                                 goto out;
1645                 }
1646                 goto mapit;
1647         } else if (!skip_sum) {
1648                 /* csum items have already been cloned */
1649                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1650                         goto mapit;
1651                 /* we're doing a write, do the async checksumming */
1652                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1653                                    inode, rw, bio, mirror_num,
1654                                    bio_flags, bio_offset,
1655                                    __btrfs_submit_bio_start,
1656                                    __btrfs_submit_bio_done);
1657                 goto out;
1658         }
1659
1660 mapit:
1661         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1662
1663 out:
1664         if (ret < 0)
1665                 bio_endio(bio, ret);
1666         return ret;
1667 }
1668
1669 /*
1670  * given a list of ordered sums record them in the inode.  This happens
1671  * at IO completion time based on sums calculated at bio submission time.
1672  */
1673 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1674                              struct inode *inode, u64 file_offset,
1675                              struct list_head *list)
1676 {
1677         struct btrfs_ordered_sum *sum;
1678
1679         list_for_each_entry(sum, list, list) {
1680                 btrfs_csum_file_blocks(trans,
1681                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1682         }
1683         return 0;
1684 }
1685
1686 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1687                               struct extent_state **cached_state)
1688 {
1689         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1690         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1691                                    cached_state, GFP_NOFS);
1692 }
1693
1694 /* see btrfs_writepage_start_hook for details on why this is required */
1695 struct btrfs_writepage_fixup {
1696         struct page *page;
1697         struct btrfs_work work;
1698 };
1699
1700 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1701 {
1702         struct btrfs_writepage_fixup *fixup;
1703         struct btrfs_ordered_extent *ordered;
1704         struct extent_state *cached_state = NULL;
1705         struct page *page;
1706         struct inode *inode;
1707         u64 page_start;
1708         u64 page_end;
1709         int ret;
1710
1711         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1712         page = fixup->page;
1713 again:
1714         lock_page(page);
1715         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1716                 ClearPageChecked(page);
1717                 goto out_page;
1718         }
1719
1720         inode = page->mapping->host;
1721         page_start = page_offset(page);
1722         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1723
1724         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1725                          &cached_state);
1726
1727         /* already ordered? We're done */
1728         if (PagePrivate2(page))
1729                 goto out;
1730
1731         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1732         if (ordered) {
1733                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1734                                      page_end, &cached_state, GFP_NOFS);
1735                 unlock_page(page);
1736                 btrfs_start_ordered_extent(inode, ordered, 1);
1737                 btrfs_put_ordered_extent(ordered);
1738                 goto again;
1739         }
1740
1741         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1742         if (ret) {
1743                 mapping_set_error(page->mapping, ret);
1744                 end_extent_writepage(page, ret, page_start, page_end);
1745                 ClearPageChecked(page);
1746                 goto out;
1747          }
1748
1749         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1750         ClearPageChecked(page);
1751         set_page_dirty(page);
1752 out:
1753         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1754                              &cached_state, GFP_NOFS);
1755 out_page:
1756         unlock_page(page);
1757         page_cache_release(page);
1758         kfree(fixup);
1759 }
1760
1761 /*
1762  * There are a few paths in the higher layers of the kernel that directly
1763  * set the page dirty bit without asking the filesystem if it is a
1764  * good idea.  This causes problems because we want to make sure COW
1765  * properly happens and the data=ordered rules are followed.
1766  *
1767  * In our case any range that doesn't have the ORDERED bit set
1768  * hasn't been properly setup for IO.  We kick off an async process
1769  * to fix it up.  The async helper will wait for ordered extents, set
1770  * the delalloc bit and make it safe to write the page.
1771  */
1772 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1773 {
1774         struct inode *inode = page->mapping->host;
1775         struct btrfs_writepage_fixup *fixup;
1776         struct btrfs_root *root = BTRFS_I(inode)->root;
1777
1778         /* this page is properly in the ordered list */
1779         if (TestClearPagePrivate2(page))
1780                 return 0;
1781
1782         if (PageChecked(page))
1783                 return -EAGAIN;
1784
1785         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1786         if (!fixup)
1787                 return -EAGAIN;
1788
1789         SetPageChecked(page);
1790         page_cache_get(page);
1791         fixup->work.func = btrfs_writepage_fixup_worker;
1792         fixup->page = page;
1793         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1794         return -EBUSY;
1795 }
1796
1797 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1798                                        struct inode *inode, u64 file_pos,
1799                                        u64 disk_bytenr, u64 disk_num_bytes,
1800                                        u64 num_bytes, u64 ram_bytes,
1801                                        u8 compression, u8 encryption,
1802                                        u16 other_encoding, int extent_type)
1803 {
1804         struct btrfs_root *root = BTRFS_I(inode)->root;
1805         struct btrfs_file_extent_item *fi;
1806         struct btrfs_path *path;
1807         struct extent_buffer *leaf;
1808         struct btrfs_key ins;
1809         int ret;
1810
1811         path = btrfs_alloc_path();
1812         if (!path)
1813                 return -ENOMEM;
1814
1815         path->leave_spinning = 1;
1816
1817         /*
1818          * we may be replacing one extent in the tree with another.
1819          * The new extent is pinned in the extent map, and we don't want
1820          * to drop it from the cache until it is completely in the btree.
1821          *
1822          * So, tell btrfs_drop_extents to leave this extent in the cache.
1823          * the caller is expected to unpin it and allow it to be merged
1824          * with the others.
1825          */
1826         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1827                                  file_pos + num_bytes, 0);
1828         if (ret)
1829                 goto out;
1830
1831         ins.objectid = btrfs_ino(inode);
1832         ins.offset = file_pos;
1833         ins.type = BTRFS_EXTENT_DATA_KEY;
1834         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1835         if (ret)
1836                 goto out;
1837         leaf = path->nodes[0];
1838         fi = btrfs_item_ptr(leaf, path->slots[0],
1839                             struct btrfs_file_extent_item);
1840         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1841         btrfs_set_file_extent_type(leaf, fi, extent_type);
1842         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1843         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1844         btrfs_set_file_extent_offset(leaf, fi, 0);
1845         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1846         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1847         btrfs_set_file_extent_compression(leaf, fi, compression);
1848         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1849         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1850
1851         btrfs_mark_buffer_dirty(leaf);
1852         btrfs_release_path(path);
1853
1854         inode_add_bytes(inode, num_bytes);
1855
1856         ins.objectid = disk_bytenr;
1857         ins.offset = disk_num_bytes;
1858         ins.type = BTRFS_EXTENT_ITEM_KEY;
1859         ret = btrfs_alloc_reserved_file_extent(trans, root,
1860                                         root->root_key.objectid,
1861                                         btrfs_ino(inode), file_pos, &ins);
1862 out:
1863         btrfs_free_path(path);
1864
1865         return ret;
1866 }
1867
1868 /*
1869  * helper function for btrfs_finish_ordered_io, this
1870  * just reads in some of the csum leaves to prime them into ram
1871  * before we start the transaction.  It limits the amount of btree
1872  * reads required while inside the transaction.
1873  */
1874 /* as ordered data IO finishes, this gets called so we can finish
1875  * an ordered extent if the range of bytes in the file it covers are
1876  * fully written.
1877  */
1878 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
1879 {
1880         struct inode *inode = ordered_extent->inode;
1881         struct btrfs_root *root = BTRFS_I(inode)->root;
1882         struct btrfs_trans_handle *trans = NULL;
1883         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1884         struct extent_state *cached_state = NULL;
1885         int compress_type = 0;
1886         int ret;
1887         bool nolock;
1888
1889         nolock = btrfs_is_free_space_inode(inode);
1890
1891         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1892                 ret = -EIO;
1893                 goto out;
1894         }
1895
1896         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1897                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1898                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1899                 if (!ret) {
1900                         if (nolock)
1901                                 trans = btrfs_join_transaction_nolock(root);
1902                         else
1903                                 trans = btrfs_join_transaction(root);
1904                         if (IS_ERR(trans)) {
1905                                 ret = PTR_ERR(trans);
1906                                 trans = NULL;
1907                                 goto out;
1908                         }
1909                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1910                         ret = btrfs_update_inode_fallback(trans, root, inode);
1911                         if (ret) /* -ENOMEM or corruption */
1912                                 btrfs_abort_transaction(trans, root, ret);
1913                 }
1914                 goto out;
1915         }
1916
1917         lock_extent_bits(io_tree, ordered_extent->file_offset,
1918                          ordered_extent->file_offset + ordered_extent->len - 1,
1919                          0, &cached_state);
1920
1921         if (nolock)
1922                 trans = btrfs_join_transaction_nolock(root);
1923         else
1924                 trans = btrfs_join_transaction(root);
1925         if (IS_ERR(trans)) {
1926                 ret = PTR_ERR(trans);
1927                 trans = NULL;
1928                 goto out_unlock;
1929         }
1930         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1931
1932         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1933                 compress_type = ordered_extent->compress_type;
1934         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1935                 BUG_ON(compress_type);
1936                 ret = btrfs_mark_extent_written(trans, inode,
1937                                                 ordered_extent->file_offset,
1938                                                 ordered_extent->file_offset +
1939                                                 ordered_extent->len);
1940         } else {
1941                 BUG_ON(root == root->fs_info->tree_root);
1942                 ret = insert_reserved_file_extent(trans, inode,
1943                                                 ordered_extent->file_offset,
1944                                                 ordered_extent->start,
1945                                                 ordered_extent->disk_len,
1946                                                 ordered_extent->len,
1947                                                 ordered_extent->len,
1948                                                 compress_type, 0, 0,
1949                                                 BTRFS_FILE_EXTENT_REG);
1950         }
1951         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1952                            ordered_extent->file_offset, ordered_extent->len,
1953                            trans->transid);
1954         if (ret < 0) {
1955                 btrfs_abort_transaction(trans, root, ret);
1956                 goto out_unlock;
1957         }
1958
1959         add_pending_csums(trans, inode, ordered_extent->file_offset,
1960                           &ordered_extent->list);
1961
1962         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1963         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1964                 ret = btrfs_update_inode_fallback(trans, root, inode);
1965                 if (ret) { /* -ENOMEM or corruption */
1966                         btrfs_abort_transaction(trans, root, ret);
1967                         goto out_unlock;
1968                 }
1969         } else {
1970                 btrfs_set_inode_last_trans(trans, inode);
1971         }
1972         ret = 0;
1973 out_unlock:
1974         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1975                              ordered_extent->file_offset +
1976                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1977 out:
1978         if (root != root->fs_info->tree_root)
1979                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1980         if (trans)
1981                 btrfs_end_transaction(trans, root);
1982
1983         if (ret)
1984                 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
1985                                       ordered_extent->file_offset +
1986                                       ordered_extent->len - 1, NULL, GFP_NOFS);
1987
1988         /*
1989          * This needs to be done to make sure anybody waiting knows we are done
1990          * updating everything for this ordered extent.
1991          */
1992         btrfs_remove_ordered_extent(inode, ordered_extent);
1993
1994         /* once for us */
1995         btrfs_put_ordered_extent(ordered_extent);
1996         /* once for the tree */
1997         btrfs_put_ordered_extent(ordered_extent);
1998
1999         return ret;
2000 }
2001
2002 static void finish_ordered_fn(struct btrfs_work *work)
2003 {
2004         struct btrfs_ordered_extent *ordered_extent;
2005         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2006         btrfs_finish_ordered_io(ordered_extent);
2007 }
2008
2009 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2010                                 struct extent_state *state, int uptodate)
2011 {
2012         struct inode *inode = page->mapping->host;
2013         struct btrfs_root *root = BTRFS_I(inode)->root;
2014         struct btrfs_ordered_extent *ordered_extent = NULL;
2015         struct btrfs_workers *workers;
2016
2017         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2018
2019         ClearPagePrivate2(page);
2020         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2021                                             end - start + 1, uptodate))
2022                 return 0;
2023
2024         ordered_extent->work.func = finish_ordered_fn;
2025         ordered_extent->work.flags = 0;
2026
2027         if (btrfs_is_free_space_inode(inode))
2028                 workers = &root->fs_info->endio_freespace_worker;
2029         else
2030                 workers = &root->fs_info->endio_write_workers;
2031         btrfs_queue_worker(workers, &ordered_extent->work);
2032
2033         return 0;
2034 }
2035
2036 /*
2037  * when reads are done, we need to check csums to verify the data is correct
2038  * if there's a match, we allow the bio to finish.  If not, the code in
2039  * extent_io.c will try to find good copies for us.
2040  */
2041 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2042                                struct extent_state *state, int mirror)
2043 {
2044         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
2045         struct inode *inode = page->mapping->host;
2046         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2047         char *kaddr;
2048         u64 private = ~(u32)0;
2049         int ret;
2050         struct btrfs_root *root = BTRFS_I(inode)->root;
2051         u32 csum = ~(u32)0;
2052
2053         if (PageChecked(page)) {
2054                 ClearPageChecked(page);
2055                 goto good;
2056         }
2057
2058         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2059                 goto good;
2060
2061         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2062             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2063                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2064                                   GFP_NOFS);
2065                 return 0;
2066         }
2067
2068         if (state && state->start == start) {
2069                 private = state->private;
2070                 ret = 0;
2071         } else {
2072                 ret = get_state_private(io_tree, start, &private);
2073         }
2074         kaddr = kmap_atomic(page);
2075         if (ret)
2076                 goto zeroit;
2077
2078         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2079         btrfs_csum_final(csum, (char *)&csum);
2080         if (csum != private)
2081                 goto zeroit;
2082
2083         kunmap_atomic(kaddr);
2084 good:
2085         return 0;
2086
2087 zeroit:
2088         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2089                        "private %llu\n",
2090                        (unsigned long long)btrfs_ino(page->mapping->host),
2091                        (unsigned long long)start, csum,
2092                        (unsigned long long)private);
2093         memset(kaddr + offset, 1, end - start + 1);
2094         flush_dcache_page(page);
2095         kunmap_atomic(kaddr);
2096         if (private == 0)
2097                 return 0;
2098         return -EIO;
2099 }
2100
2101 struct delayed_iput {
2102         struct list_head list;
2103         struct inode *inode;
2104 };
2105
2106 /* JDM: If this is fs-wide, why can't we add a pointer to
2107  * btrfs_inode instead and avoid the allocation? */
2108 void btrfs_add_delayed_iput(struct inode *inode)
2109 {
2110         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2111         struct delayed_iput *delayed;
2112
2113         if (atomic_add_unless(&inode->i_count, -1, 1))
2114                 return;
2115
2116         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2117         delayed->inode = inode;
2118
2119         spin_lock(&fs_info->delayed_iput_lock);
2120         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2121         spin_unlock(&fs_info->delayed_iput_lock);
2122 }
2123
2124 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2125 {
2126         LIST_HEAD(list);
2127         struct btrfs_fs_info *fs_info = root->fs_info;
2128         struct delayed_iput *delayed;
2129         int empty;
2130
2131         spin_lock(&fs_info->delayed_iput_lock);
2132         empty = list_empty(&fs_info->delayed_iputs);
2133         spin_unlock(&fs_info->delayed_iput_lock);
2134         if (empty)
2135                 return;
2136
2137         spin_lock(&fs_info->delayed_iput_lock);
2138         list_splice_init(&fs_info->delayed_iputs, &list);
2139         spin_unlock(&fs_info->delayed_iput_lock);
2140
2141         while (!list_empty(&list)) {
2142                 delayed = list_entry(list.next, struct delayed_iput, list);
2143                 list_del(&delayed->list);
2144                 iput(delayed->inode);
2145                 kfree(delayed);
2146         }
2147 }
2148
2149 enum btrfs_orphan_cleanup_state {
2150         ORPHAN_CLEANUP_STARTED  = 1,
2151         ORPHAN_CLEANUP_DONE     = 2,
2152 };
2153
2154 /*
2155  * This is called in transaction commit time. If there are no orphan
2156  * files in the subvolume, it removes orphan item and frees block_rsv
2157  * structure.
2158  */
2159 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2160                               struct btrfs_root *root)
2161 {
2162         struct btrfs_block_rsv *block_rsv;
2163         int ret;
2164
2165         if (atomic_read(&root->orphan_inodes) ||
2166             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2167                 return;
2168
2169         spin_lock(&root->orphan_lock);
2170         if (atomic_read(&root->orphan_inodes)) {
2171                 spin_unlock(&root->orphan_lock);
2172                 return;
2173         }
2174
2175         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2176                 spin_unlock(&root->orphan_lock);
2177                 return;
2178         }
2179
2180         block_rsv = root->orphan_block_rsv;
2181         root->orphan_block_rsv = NULL;
2182         spin_unlock(&root->orphan_lock);
2183
2184         if (root->orphan_item_inserted &&
2185             btrfs_root_refs(&root->root_item) > 0) {
2186                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2187                                             root->root_key.objectid);
2188                 BUG_ON(ret);
2189                 root->orphan_item_inserted = 0;
2190         }
2191
2192         if (block_rsv) {
2193                 WARN_ON(block_rsv->size > 0);
2194                 btrfs_free_block_rsv(root, block_rsv);
2195         }
2196 }
2197
2198 /*
2199  * This creates an orphan entry for the given inode in case something goes
2200  * wrong in the middle of an unlink/truncate.
2201  *
2202  * NOTE: caller of this function should reserve 5 units of metadata for
2203  *       this function.
2204  */
2205 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2206 {
2207         struct btrfs_root *root = BTRFS_I(inode)->root;
2208         struct btrfs_block_rsv *block_rsv = NULL;
2209         int reserve = 0;
2210         int insert = 0;
2211         int ret;
2212
2213         if (!root->orphan_block_rsv) {
2214                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2215                 if (!block_rsv)
2216                         return -ENOMEM;
2217         }
2218
2219         spin_lock(&root->orphan_lock);
2220         if (!root->orphan_block_rsv) {
2221                 root->orphan_block_rsv = block_rsv;
2222         } else if (block_rsv) {
2223                 btrfs_free_block_rsv(root, block_rsv);
2224                 block_rsv = NULL;
2225         }
2226
2227         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2228                               &BTRFS_I(inode)->runtime_flags)) {
2229 #if 0
2230                 /*
2231                  * For proper ENOSPC handling, we should do orphan
2232                  * cleanup when mounting. But this introduces backward
2233                  * compatibility issue.
2234                  */
2235                 if (!xchg(&root->orphan_item_inserted, 1))
2236                         insert = 2;
2237                 else
2238                         insert = 1;
2239 #endif
2240                 insert = 1;
2241                 atomic_inc(&root->orphan_inodes);
2242         }
2243
2244         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2245                               &BTRFS_I(inode)->runtime_flags))
2246                 reserve = 1;
2247         spin_unlock(&root->orphan_lock);
2248
2249         /* grab metadata reservation from transaction handle */
2250         if (reserve) {
2251                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2252                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2253         }
2254
2255         /* insert an orphan item to track this unlinked/truncated file */
2256         if (insert >= 1) {
2257                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2258                 if (ret && ret != -EEXIST) {
2259                         clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2260                                   &BTRFS_I(inode)->runtime_flags);
2261                         btrfs_abort_transaction(trans, root, ret);
2262                         return ret;
2263                 }
2264                 ret = 0;
2265         }
2266
2267         /* insert an orphan item to track subvolume contains orphan files */
2268         if (insert >= 2) {
2269                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2270                                                root->root_key.objectid);
2271                 if (ret && ret != -EEXIST) {
2272                         btrfs_abort_transaction(trans, root, ret);
2273                         return ret;
2274                 }
2275         }
2276         return 0;
2277 }
2278
2279 /*
2280  * We have done the truncate/delete so we can go ahead and remove the orphan
2281  * item for this particular inode.
2282  */
2283 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2284 {
2285         struct btrfs_root *root = BTRFS_I(inode)->root;
2286         int delete_item = 0;
2287         int release_rsv = 0;
2288         int ret = 0;
2289
2290         spin_lock(&root->orphan_lock);
2291         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2292                                &BTRFS_I(inode)->runtime_flags))
2293                 delete_item = 1;
2294
2295         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2296                                &BTRFS_I(inode)->runtime_flags))
2297                 release_rsv = 1;
2298         spin_unlock(&root->orphan_lock);
2299
2300         if (trans && delete_item) {
2301                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2302                 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2303         }
2304
2305         if (release_rsv) {
2306                 btrfs_orphan_release_metadata(inode);
2307                 atomic_dec(&root->orphan_inodes);
2308         }
2309
2310         return 0;
2311 }
2312
2313 /*
2314  * this cleans up any orphans that may be left on the list from the last use
2315  * of this root.
2316  */
2317 int btrfs_orphan_cleanup(struct btrfs_root *root)
2318 {
2319         struct btrfs_path *path;
2320         struct extent_buffer *leaf;
2321         struct btrfs_key key, found_key;
2322         struct btrfs_trans_handle *trans;
2323         struct inode *inode;
2324         u64 last_objectid = 0;
2325         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2326
2327         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2328                 return 0;
2329
2330         path = btrfs_alloc_path();
2331         if (!path) {
2332                 ret = -ENOMEM;
2333                 goto out;
2334         }
2335         path->reada = -1;
2336
2337         key.objectid = BTRFS_ORPHAN_OBJECTID;
2338         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2339         key.offset = (u64)-1;
2340
2341         while (1) {
2342                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2343                 if (ret < 0)
2344                         goto out;
2345
2346                 /*
2347                  * if ret == 0 means we found what we were searching for, which
2348                  * is weird, but possible, so only screw with path if we didn't
2349                  * find the key and see if we have stuff that matches
2350                  */
2351                 if (ret > 0) {
2352                         ret = 0;
2353                         if (path->slots[0] == 0)
2354                                 break;
2355                         path->slots[0]--;
2356                 }
2357
2358                 /* pull out the item */
2359                 leaf = path->nodes[0];
2360                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2361
2362                 /* make sure the item matches what we want */
2363                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2364                         break;
2365                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2366                         break;
2367
2368                 /* release the path since we're done with it */
2369                 btrfs_release_path(path);
2370
2371                 /*
2372                  * this is where we are basically btrfs_lookup, without the
2373                  * crossing root thing.  we store the inode number in the
2374                  * offset of the orphan item.
2375                  */
2376
2377                 if (found_key.offset == last_objectid) {
2378                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2379                                "stopping orphan cleanup\n");
2380                         ret = -EINVAL;
2381                         goto out;
2382                 }
2383
2384                 last_objectid = found_key.offset;
2385
2386                 found_key.objectid = found_key.offset;
2387                 found_key.type = BTRFS_INODE_ITEM_KEY;
2388                 found_key.offset = 0;
2389                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2390                 ret = PTR_RET(inode);
2391                 if (ret && ret != -ESTALE)
2392                         goto out;
2393
2394                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2395                         struct btrfs_root *dead_root;
2396                         struct btrfs_fs_info *fs_info = root->fs_info;
2397                         int is_dead_root = 0;
2398
2399                         /*
2400                          * this is an orphan in the tree root. Currently these
2401                          * could come from 2 sources:
2402                          *  a) a snapshot deletion in progress
2403                          *  b) a free space cache inode
2404                          * We need to distinguish those two, as the snapshot
2405                          * orphan must not get deleted.
2406                          * find_dead_roots already ran before us, so if this
2407                          * is a snapshot deletion, we should find the root
2408                          * in the dead_roots list
2409                          */
2410                         spin_lock(&fs_info->trans_lock);
2411                         list_for_each_entry(dead_root, &fs_info->dead_roots,
2412                                             root_list) {
2413                                 if (dead_root->root_key.objectid ==
2414                                     found_key.objectid) {
2415                                         is_dead_root = 1;
2416                                         break;
2417                                 }
2418                         }
2419                         spin_unlock(&fs_info->trans_lock);
2420                         if (is_dead_root) {
2421                                 /* prevent this orphan from being found again */
2422                                 key.offset = found_key.objectid - 1;
2423                                 continue;
2424                         }
2425                 }
2426                 /*
2427                  * Inode is already gone but the orphan item is still there,
2428                  * kill the orphan item.
2429                  */
2430                 if (ret == -ESTALE) {
2431                         trans = btrfs_start_transaction(root, 1);
2432                         if (IS_ERR(trans)) {
2433                                 ret = PTR_ERR(trans);
2434                                 goto out;
2435                         }
2436                         printk(KERN_ERR "auto deleting %Lu\n",
2437                                found_key.objectid);
2438                         ret = btrfs_del_orphan_item(trans, root,
2439                                                     found_key.objectid);
2440                         BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2441                         btrfs_end_transaction(trans, root);
2442                         continue;
2443                 }
2444
2445                 /*
2446                  * add this inode to the orphan list so btrfs_orphan_del does
2447                  * the proper thing when we hit it
2448                  */
2449                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2450                         &BTRFS_I(inode)->runtime_flags);
2451
2452                 /* if we have links, this was a truncate, lets do that */
2453                 if (inode->i_nlink) {
2454                         if (!S_ISREG(inode->i_mode)) {
2455                                 WARN_ON(1);
2456                                 iput(inode);
2457                                 continue;
2458                         }
2459                         nr_truncate++;
2460                         ret = btrfs_truncate(inode);
2461                 } else {
2462                         nr_unlink++;
2463                 }
2464
2465                 /* this will do delete_inode and everything for us */
2466                 iput(inode);
2467                 if (ret)
2468                         goto out;
2469         }
2470         /* release the path since we're done with it */
2471         btrfs_release_path(path);
2472
2473         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2474
2475         if (root->orphan_block_rsv)
2476                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2477                                         (u64)-1);
2478
2479         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2480                 trans = btrfs_join_transaction(root);
2481                 if (!IS_ERR(trans))
2482                         btrfs_end_transaction(trans, root);
2483         }
2484
2485         if (nr_unlink)
2486                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2487         if (nr_truncate)
2488                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2489
2490 out:
2491         if (ret)
2492                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2493         btrfs_free_path(path);
2494         return ret;
2495 }
2496
2497 /*
2498  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2499  * don't find any xattrs, we know there can't be any acls.
2500  *
2501  * slot is the slot the inode is in, objectid is the objectid of the inode
2502  */
2503 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2504                                           int slot, u64 objectid)
2505 {
2506         u32 nritems = btrfs_header_nritems(leaf);
2507         struct btrfs_key found_key;
2508         int scanned = 0;
2509
2510         slot++;
2511         while (slot < nritems) {
2512                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2513
2514                 /* we found a different objectid, there must not be acls */
2515                 if (found_key.objectid != objectid)
2516                         return 0;
2517
2518                 /* we found an xattr, assume we've got an acl */
2519                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2520                         return 1;
2521
2522                 /*
2523                  * we found a key greater than an xattr key, there can't
2524                  * be any acls later on
2525                  */
2526                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2527                         return 0;
2528
2529                 slot++;
2530                 scanned++;
2531
2532                 /*
2533                  * it goes inode, inode backrefs, xattrs, extents,
2534                  * so if there are a ton of hard links to an inode there can
2535                  * be a lot of backrefs.  Don't waste time searching too hard,
2536                  * this is just an optimization
2537                  */
2538                 if (scanned >= 8)
2539                         break;
2540         }
2541         /* we hit the end of the leaf before we found an xattr or
2542          * something larger than an xattr.  We have to assume the inode
2543          * has acls
2544          */
2545         return 1;
2546 }
2547
2548 /*
2549  * read an inode from the btree into the in-memory inode
2550  */
2551 static void btrfs_read_locked_inode(struct inode *inode)
2552 {
2553         struct btrfs_path *path;
2554         struct extent_buffer *leaf;
2555         struct btrfs_inode_item *inode_item;
2556         struct btrfs_timespec *tspec;
2557         struct btrfs_root *root = BTRFS_I(inode)->root;
2558         struct btrfs_key location;
2559         int maybe_acls;
2560         u32 rdev;
2561         int ret;
2562         bool filled = false;
2563
2564         ret = btrfs_fill_inode(inode, &rdev);
2565         if (!ret)
2566                 filled = true;
2567
2568         path = btrfs_alloc_path();
2569         if (!path)
2570                 goto make_bad;
2571
2572         path->leave_spinning = 1;
2573         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2574
2575         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2576         if (ret)
2577                 goto make_bad;
2578
2579         leaf = path->nodes[0];
2580
2581         if (filled)
2582                 goto cache_acl;
2583
2584         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2585                                     struct btrfs_inode_item);
2586         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2587         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2588         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
2589         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
2590         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2591
2592         tspec = btrfs_inode_atime(inode_item);
2593         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2594         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2595
2596         tspec = btrfs_inode_mtime(inode_item);
2597         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2598         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2599
2600         tspec = btrfs_inode_ctime(inode_item);
2601         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2602         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2603
2604         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2605         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2606         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
2607
2608         /*
2609          * If we were modified in the current generation and evicted from memory
2610          * and then re-read we need to do a full sync since we don't have any
2611          * idea about which extents were modified before we were evicted from
2612          * cache.
2613          */
2614         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
2615                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2616                         &BTRFS_I(inode)->runtime_flags);
2617
2618         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2619         inode->i_generation = BTRFS_I(inode)->generation;
2620         inode->i_rdev = 0;
2621         rdev = btrfs_inode_rdev(leaf, inode_item);
2622
2623         BTRFS_I(inode)->index_cnt = (u64)-1;
2624         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2625 cache_acl:
2626         /*
2627          * try to precache a NULL acl entry for files that don't have
2628          * any xattrs or acls
2629          */
2630         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2631                                            btrfs_ino(inode));
2632         if (!maybe_acls)
2633                 cache_no_acl(inode);
2634
2635         btrfs_free_path(path);
2636
2637         switch (inode->i_mode & S_IFMT) {
2638         case S_IFREG:
2639                 inode->i_mapping->a_ops = &btrfs_aops;
2640                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2641                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2642                 inode->i_fop = &btrfs_file_operations;
2643                 inode->i_op = &btrfs_file_inode_operations;
2644                 break;
2645         case S_IFDIR:
2646                 inode->i_fop = &btrfs_dir_file_operations;
2647                 if (root == root->fs_info->tree_root)
2648                         inode->i_op = &btrfs_dir_ro_inode_operations;
2649                 else
2650                         inode->i_op = &btrfs_dir_inode_operations;
2651                 break;
2652         case S_IFLNK:
2653                 inode->i_op = &btrfs_symlink_inode_operations;
2654                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2655                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2656                 break;
2657         default:
2658                 inode->i_op = &btrfs_special_inode_operations;
2659                 init_special_inode(inode, inode->i_mode, rdev);
2660                 break;
2661         }
2662
2663         btrfs_update_iflags(inode);
2664         return;
2665
2666 make_bad:
2667         btrfs_free_path(path);
2668         make_bad_inode(inode);
2669 }
2670
2671 /*
2672  * given a leaf and an inode, copy the inode fields into the leaf
2673  */
2674 static void fill_inode_item(struct btrfs_trans_handle *trans,
2675                             struct extent_buffer *leaf,
2676                             struct btrfs_inode_item *item,
2677                             struct inode *inode)
2678 {
2679         btrfs_set_inode_uid(leaf, item, i_uid_read(inode));
2680         btrfs_set_inode_gid(leaf, item, i_gid_read(inode));
2681         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2682         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2683         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2684
2685         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2686                                inode->i_atime.tv_sec);
2687         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2688                                 inode->i_atime.tv_nsec);
2689
2690         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2691                                inode->i_mtime.tv_sec);
2692         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2693                                 inode->i_mtime.tv_nsec);
2694
2695         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2696                                inode->i_ctime.tv_sec);
2697         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2698                                 inode->i_ctime.tv_nsec);
2699
2700         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2701         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2702         btrfs_set_inode_sequence(leaf, item, inode->i_version);
2703         btrfs_set_inode_transid(leaf, item, trans->transid);
2704         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2705         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2706         btrfs_set_inode_block_group(leaf, item, 0);
2707 }
2708
2709 /*
2710  * copy everything in the in-memory inode into the btree.
2711  */
2712 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2713                                 struct btrfs_root *root, struct inode *inode)
2714 {
2715         struct btrfs_inode_item *inode_item;
2716         struct btrfs_path *path;
2717         struct extent_buffer *leaf;
2718         int ret;
2719
2720         path = btrfs_alloc_path();
2721         if (!path)
2722                 return -ENOMEM;
2723
2724         path->leave_spinning = 1;
2725         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2726                                  1);
2727         if (ret) {
2728                 if (ret > 0)
2729                         ret = -ENOENT;
2730                 goto failed;
2731         }
2732
2733         btrfs_unlock_up_safe(path, 1);
2734         leaf = path->nodes[0];
2735         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2736                                     struct btrfs_inode_item);
2737
2738         fill_inode_item(trans, leaf, inode_item, inode);
2739         btrfs_mark_buffer_dirty(leaf);
2740         btrfs_set_inode_last_trans(trans, inode);
2741         ret = 0;
2742 failed:
2743         btrfs_free_path(path);
2744         return ret;
2745 }
2746
2747 /*
2748  * copy everything in the in-memory inode into the btree.
2749  */
2750 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2751                                 struct btrfs_root *root, struct inode *inode)
2752 {
2753         int ret;
2754
2755         /*
2756          * If the inode is a free space inode, we can deadlock during commit
2757          * if we put it into the delayed code.
2758          *
2759          * The data relocation inode should also be directly updated
2760          * without delay
2761          */
2762         if (!btrfs_is_free_space_inode(inode)
2763             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2764                 btrfs_update_root_times(trans, root);
2765
2766                 ret = btrfs_delayed_update_inode(trans, root, inode);
2767                 if (!ret)
2768                         btrfs_set_inode_last_trans(trans, inode);
2769                 return ret;
2770         }
2771
2772         return btrfs_update_inode_item(trans, root, inode);
2773 }
2774
2775 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2776                                          struct btrfs_root *root,
2777                                          struct inode *inode)
2778 {
2779         int ret;
2780
2781         ret = btrfs_update_inode(trans, root, inode);
2782         if (ret == -ENOSPC)
2783                 return btrfs_update_inode_item(trans, root, inode);
2784         return ret;
2785 }
2786
2787 /*
2788  * unlink helper that gets used here in inode.c and in the tree logging
2789  * recovery code.  It remove a link in a directory with a given name, and
2790  * also drops the back refs in the inode to the directory
2791  */
2792 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2793                                 struct btrfs_root *root,
2794                                 struct inode *dir, struct inode *inode,
2795                                 const char *name, int name_len)
2796 {
2797         struct btrfs_path *path;
2798         int ret = 0;
2799         struct extent_buffer *leaf;
2800         struct btrfs_dir_item *di;
2801         struct btrfs_key key;
2802         u64 index;
2803         u64 ino = btrfs_ino(inode);
2804         u64 dir_ino = btrfs_ino(dir);
2805
2806         path = btrfs_alloc_path();
2807         if (!path) {
2808                 ret = -ENOMEM;
2809                 goto out;
2810         }
2811
2812         path->leave_spinning = 1;
2813         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2814                                     name, name_len, -1);
2815         if (IS_ERR(di)) {
2816                 ret = PTR_ERR(di);
2817                 goto err;
2818         }
2819         if (!di) {
2820                 ret = -ENOENT;
2821                 goto err;
2822         }
2823         leaf = path->nodes[0];
2824         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2825         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2826         if (ret)
2827                 goto err;
2828         btrfs_release_path(path);
2829
2830         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2831                                   dir_ino, &index);
2832         if (ret) {
2833                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2834                        "inode %llu parent %llu\n", name_len, name,
2835                        (unsigned long long)ino, (unsigned long long)dir_ino);
2836                 btrfs_abort_transaction(trans, root, ret);
2837                 goto err;
2838         }
2839
2840         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2841         if (ret) {
2842                 btrfs_abort_transaction(trans, root, ret);
2843                 goto err;
2844         }
2845
2846         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2847                                          inode, dir_ino);
2848         if (ret != 0 && ret != -ENOENT) {
2849                 btrfs_abort_transaction(trans, root, ret);
2850                 goto err;
2851         }
2852
2853         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2854                                            dir, index);
2855         if (ret == -ENOENT)
2856                 ret = 0;
2857 err:
2858         btrfs_free_path(path);
2859         if (ret)
2860                 goto out;
2861
2862         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2863         inode_inc_iversion(inode);
2864         inode_inc_iversion(dir);
2865         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2866         ret = btrfs_update_inode(trans, root, dir);
2867 out:
2868         return ret;
2869 }
2870
2871 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2872                        struct btrfs_root *root,
2873                        struct inode *dir, struct inode *inode,
2874                        const char *name, int name_len)
2875 {
2876         int ret;
2877         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2878         if (!ret) {
2879                 btrfs_drop_nlink(inode);
2880                 ret = btrfs_update_inode(trans, root, inode);
2881         }
2882         return ret;
2883 }
2884                 
2885
2886 /* helper to check if there is any shared block in the path */
2887 static int check_path_shared(struct btrfs_root *root,
2888                              struct btrfs_path *path)
2889 {
2890         struct extent_buffer *eb;
2891         int level;
2892         u64 refs = 1;
2893
2894         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2895                 int ret;
2896
2897                 if (!path->nodes[level])
2898                         break;
2899                 eb = path->nodes[level];
2900                 if (!btrfs_block_can_be_shared(root, eb))
2901                         continue;
2902                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2903                                                &refs, NULL);
2904                 if (refs > 1)
2905                         return 1;
2906         }
2907         return 0;
2908 }
2909
2910 /*
2911  * helper to start transaction for unlink and rmdir.
2912  *
2913  * unlink and rmdir are special in btrfs, they do not always free space.
2914  * so in enospc case, we should make sure they will free space before
2915  * allowing them to use the global metadata reservation.
2916  */
2917 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2918                                                        struct dentry *dentry)
2919 {
2920         struct btrfs_trans_handle *trans;
2921         struct btrfs_root *root = BTRFS_I(dir)->root;
2922         struct btrfs_path *path;
2923         struct btrfs_dir_item *di;
2924         struct inode *inode = dentry->d_inode;
2925         u64 index;
2926         int check_link = 1;
2927         int err = -ENOSPC;
2928         int ret;
2929         u64 ino = btrfs_ino(inode);
2930         u64 dir_ino = btrfs_ino(dir);
2931
2932         /*
2933          * 1 for the possible orphan item
2934          * 1 for the dir item
2935          * 1 for the dir index
2936          * 1 for the inode ref
2937          * 1 for the inode ref in the tree log
2938          * 2 for the dir entries in the log
2939          * 1 for the inode
2940          */
2941         trans = btrfs_start_transaction(root, 8);
2942         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2943                 return trans;
2944
2945         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2946                 return ERR_PTR(-ENOSPC);
2947
2948         /* check if there is someone else holds reference */
2949         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2950                 return ERR_PTR(-ENOSPC);
2951
2952         if (atomic_read(&inode->i_count) > 2)
2953                 return ERR_PTR(-ENOSPC);
2954
2955         if (xchg(&root->fs_info->enospc_unlink, 1))
2956                 return ERR_PTR(-ENOSPC);
2957
2958         path = btrfs_alloc_path();
2959         if (!path) {
2960                 root->fs_info->enospc_unlink = 0;
2961                 return ERR_PTR(-ENOMEM);
2962         }
2963
2964         /* 1 for the orphan item */
2965         trans = btrfs_start_transaction(root, 1);
2966         if (IS_ERR(trans)) {
2967                 btrfs_free_path(path);
2968                 root->fs_info->enospc_unlink = 0;
2969                 return trans;
2970         }
2971
2972         path->skip_locking = 1;
2973         path->search_commit_root = 1;
2974
2975         ret = btrfs_lookup_inode(trans, root, path,
2976                                 &BTRFS_I(dir)->location, 0);
2977         if (ret < 0) {
2978                 err = ret;
2979                 goto out;
2980         }
2981         if (ret == 0) {
2982                 if (check_path_shared(root, path))
2983                         goto out;
2984         } else {
2985                 check_link = 0;
2986         }
2987         btrfs_release_path(path);
2988
2989         ret = btrfs_lookup_inode(trans, root, path,
2990                                 &BTRFS_I(inode)->location, 0);
2991         if (ret < 0) {
2992                 err = ret;
2993                 goto out;
2994         }
2995         if (ret == 0) {
2996                 if (check_path_shared(root, path))
2997                         goto out;
2998         } else {
2999                 check_link = 0;
3000         }
3001         btrfs_release_path(path);
3002
3003         if (ret == 0 && S_ISREG(inode->i_mode)) {
3004                 ret = btrfs_lookup_file_extent(trans, root, path,
3005                                                ino, (u64)-1, 0);
3006                 if (ret < 0) {
3007                         err = ret;
3008                         goto out;
3009                 }
3010                 BUG_ON(ret == 0); /* Corruption */
3011                 if (check_path_shared(root, path))
3012                         goto out;
3013                 btrfs_release_path(path);
3014         }
3015
3016         if (!check_link) {
3017                 err = 0;
3018                 goto out;
3019         }
3020
3021         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3022                                 dentry->d_name.name, dentry->d_name.len, 0);
3023         if (IS_ERR(di)) {
3024                 err = PTR_ERR(di);
3025                 goto out;
3026         }
3027         if (di) {
3028                 if (check_path_shared(root, path))
3029                         goto out;
3030         } else {
3031                 err = 0;
3032                 goto out;
3033         }
3034         btrfs_release_path(path);
3035
3036         ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3037                                         dentry->d_name.len, ino, dir_ino, 0,
3038                                         &index);
3039         if (ret) {
3040                 err = ret;
3041                 goto out;
3042         }
3043
3044         if (check_path_shared(root, path))
3045                 goto out;
3046
3047         btrfs_release_path(path);
3048
3049         /*
3050          * This is a commit root search, if we can lookup inode item and other
3051          * relative items in the commit root, it means the transaction of
3052          * dir/file creation has been committed, and the dir index item that we
3053          * delay to insert has also been inserted into the commit root. So
3054          * we needn't worry about the delayed insertion of the dir index item
3055          * here.
3056          */
3057         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3058                                 dentry->d_name.name, dentry->d_name.len, 0);
3059         if (IS_ERR(di)) {
3060                 err = PTR_ERR(di);
3061                 goto out;
3062         }
3063         BUG_ON(ret == -ENOENT);
3064         if (check_path_shared(root, path))
3065                 goto out;
3066
3067         err = 0;
3068 out:
3069         btrfs_free_path(path);
3070         /* Migrate the orphan reservation over */
3071         if (!err)
3072                 err = btrfs_block_rsv_migrate(trans->block_rsv,
3073                                 &root->fs_info->global_block_rsv,
3074                                 trans->bytes_reserved);
3075
3076         if (err) {
3077                 btrfs_end_transaction(trans, root);
3078                 root->fs_info->enospc_unlink = 0;
3079                 return ERR_PTR(err);
3080         }
3081
3082         trans->block_rsv = &root->fs_info->global_block_rsv;
3083         return trans;
3084 }
3085
3086 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3087                                struct btrfs_root *root)
3088 {
3089         if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3090                 btrfs_block_rsv_release(root, trans->block_rsv,
3091                                         trans->bytes_reserved);
3092                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3093                 BUG_ON(!root->fs_info->enospc_unlink);
3094                 root->fs_info->enospc_unlink = 0;
3095         }
3096         btrfs_end_transaction(trans, root);
3097 }
3098
3099 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3100 {
3101         struct btrfs_root *root = BTRFS_I(dir)->root;
3102         struct btrfs_trans_handle *trans;
3103         struct inode *inode = dentry->d_inode;
3104         int ret;
3105
3106         trans = __unlink_start_trans(dir, dentry);
3107         if (IS_ERR(trans))
3108                 return PTR_ERR(trans);
3109
3110         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3111
3112         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3113                                  dentry->d_name.name, dentry->d_name.len);
3114         if (ret)
3115                 goto out;
3116
3117         if (inode->i_nlink == 0) {
3118                 ret = btrfs_orphan_add(trans, inode);
3119                 if (ret)
3120                         goto out;
3121         }
3122
3123 out:
3124         __unlink_end_trans(trans, root);
3125         btrfs_btree_balance_dirty(root);
3126         return ret;
3127 }
3128
3129 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3130                         struct btrfs_root *root,
3131                         struct inode *dir, u64 objectid,
3132                         const char *name, int name_len)
3133 {
3134         struct btrfs_path *path;
3135         struct extent_buffer *leaf;
3136         struct btrfs_dir_item *di;
3137         struct btrfs_key key;
3138         u64 index;
3139         int ret;
3140         u64 dir_ino = btrfs_ino(dir);
3141
3142         path = btrfs_alloc_path();
3143         if (!path)
3144                 return -ENOMEM;
3145
3146         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3147                                    name, name_len, -1);
3148         if (IS_ERR_OR_NULL(di)) {
3149                 if (!di)
3150                         ret = -ENOENT;
3151                 else
3152                         ret = PTR_ERR(di);
3153                 goto out;
3154         }
3155
3156         leaf = path->nodes[0];
3157         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3158         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3159         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3160         if (ret) {
3161                 btrfs_abort_transaction(trans, root, ret);
3162                 goto out;
3163         }
3164         btrfs_release_path(path);
3165
3166         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3167                                  objectid, root->root_key.objectid,
3168                                  dir_ino, &index, name, name_len);
3169         if (ret < 0) {
3170                 if (ret != -ENOENT) {
3171                         btrfs_abort_transaction(trans, root, ret);
3172                         goto out;
3173                 }
3174                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3175                                                  name, name_len);
3176                 if (IS_ERR_OR_NULL(di)) {
3177                         if (!di)
3178                                 ret = -ENOENT;
3179                         else
3180                                 ret = PTR_ERR(di);
3181                         btrfs_abort_transaction(trans, root, ret);
3182                         goto out;
3183                 }
3184
3185                 leaf = path->nodes[0];
3186                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3187                 btrfs_release_path(path);
3188                 index = key.offset;
3189         }
3190         btrfs_release_path(path);
3191
3192         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3193         if (ret) {
3194                 btrfs_abort_transaction(trans, root, ret);
3195                 goto out;
3196         }
3197
3198         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3199         inode_inc_iversion(dir);
3200         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3201         ret = btrfs_update_inode_fallback(trans, root, dir);
3202         if (ret)
3203                 btrfs_abort_transaction(trans, root, ret);
3204 out:
3205         btrfs_free_path(path);
3206         return ret;
3207 }
3208
3209 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3210 {
3211         struct inode *inode = dentry->d_inode;
3212         int err = 0;
3213         struct btrfs_root *root = BTRFS_I(dir)->root;
3214         struct btrfs_trans_handle *trans;
3215
3216         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)