115bc05e42b06fee05ef7551cef1ad4174557634
[~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 struct kmem_cache *btrfs_trans_handle_cachep;
75 struct kmem_cache *btrfs_transaction_cachep;
76 struct kmem_cache *btrfs_path_cachep;
77 struct kmem_cache *btrfs_free_space_cachep;
78
79 #define S_SHIFT 12
80 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
81         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
82         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
83         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
84         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
85         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
86         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
87         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
88 };
89
90 static int btrfs_setsize(struct inode *inode, loff_t newsize);
91 static int btrfs_truncate(struct inode *inode);
92 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
93 static noinline int cow_file_range(struct inode *inode,
94                                    struct page *locked_page,
95                                    u64 start, u64 end, int *page_started,
96                                    unsigned long *nr_written, int unlock);
97 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
98                                 struct btrfs_root *root, struct inode *inode);
99
100 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
101                                      struct inode *inode,  struct inode *dir,
102                                      const struct qstr *qstr)
103 {
104         int err;
105
106         err = btrfs_init_acl(trans, inode, dir);
107         if (!err)
108                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
109         return err;
110 }
111
112 /*
113  * this does all the hard work for inserting an inline extent into
114  * the btree.  The caller should have done a btrfs_drop_extents so that
115  * no overlapping inline items exist in the btree
116  */
117 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
118                                 struct btrfs_root *root, struct inode *inode,
119                                 u64 start, size_t size, size_t compressed_size,
120                                 int compress_type,
121                                 struct page **compressed_pages)
122 {
123         struct btrfs_key key;
124         struct btrfs_path *path;
125         struct extent_buffer *leaf;
126         struct page *page = NULL;
127         char *kaddr;
128         unsigned long ptr;
129         struct btrfs_file_extent_item *ei;
130         int err = 0;
131         int ret;
132         size_t cur_size = size;
133         size_t datasize;
134         unsigned long offset;
135
136         if (compressed_size && compressed_pages)
137                 cur_size = compressed_size;
138
139         path = btrfs_alloc_path();
140         if (!path)
141                 return -ENOMEM;
142
143         path->leave_spinning = 1;
144
145         key.objectid = btrfs_ino(inode);
146         key.offset = start;
147         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
148         datasize = btrfs_file_extent_calc_inline_size(cur_size);
149
150         inode_add_bytes(inode, size);
151         ret = btrfs_insert_empty_item(trans, root, path, &key,
152                                       datasize);
153         if (ret) {
154                 err = ret;
155                 goto fail;
156         }
157         leaf = path->nodes[0];
158         ei = btrfs_item_ptr(leaf, path->slots[0],
159                             struct btrfs_file_extent_item);
160         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
161         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
162         btrfs_set_file_extent_encryption(leaf, ei, 0);
163         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
164         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
165         ptr = btrfs_file_extent_inline_start(ei);
166
167         if (compress_type != BTRFS_COMPRESS_NONE) {
168                 struct page *cpage;
169                 int i = 0;
170                 while (compressed_size > 0) {
171                         cpage = compressed_pages[i];
172                         cur_size = min_t(unsigned long, compressed_size,
173                                        PAGE_CACHE_SIZE);
174
175                         kaddr = kmap_atomic(cpage);
176                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
177                         kunmap_atomic(kaddr);
178
179                         i++;
180                         ptr += cur_size;
181                         compressed_size -= cur_size;
182                 }
183                 btrfs_set_file_extent_compression(leaf, ei,
184                                                   compress_type);
185         } else {
186                 page = find_get_page(inode->i_mapping,
187                                      start >> PAGE_CACHE_SHIFT);
188                 btrfs_set_file_extent_compression(leaf, ei, 0);
189                 kaddr = kmap_atomic(page);
190                 offset = start & (PAGE_CACHE_SIZE - 1);
191                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
192                 kunmap_atomic(kaddr);
193                 page_cache_release(page);
194         }
195         btrfs_mark_buffer_dirty(leaf);
196         btrfs_free_path(path);
197
198         /*
199          * we're an inline extent, so nobody can
200          * extend the file past i_size without locking
201          * a page we already have locked.
202          *
203          * We must do any isize and inode updates
204          * before we unlock the pages.  Otherwise we
205          * could end up racing with unlink.
206          */
207         BTRFS_I(inode)->disk_i_size = inode->i_size;
208         ret = btrfs_update_inode(trans, root, inode);
209
210         return ret;
211 fail:
212         btrfs_free_path(path);
213         return err;
214 }
215
216
217 /*
218  * conditionally insert an inline extent into the file.  This
219  * does the checks required to make sure the data is small enough
220  * to fit as an inline extent.
221  */
222 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
223                                  struct btrfs_root *root,
224                                  struct inode *inode, u64 start, u64 end,
225                                  size_t compressed_size, int compress_type,
226                                  struct page **compressed_pages)
227 {
228         u64 isize = i_size_read(inode);
229         u64 actual_end = min(end + 1, isize);
230         u64 inline_len = actual_end - start;
231         u64 aligned_end = (end + root->sectorsize - 1) &
232                         ~((u64)root->sectorsize - 1);
233         u64 hint_byte;
234         u64 data_len = inline_len;
235         int ret;
236
237         if (compressed_size)
238                 data_len = compressed_size;
239
240         if (start > 0 ||
241             actual_end >= PAGE_CACHE_SIZE ||
242             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
243             (!compressed_size &&
244             (actual_end & (root->sectorsize - 1)) == 0) ||
245             end + 1 < isize ||
246             data_len > root->fs_info->max_inline) {
247                 return 1;
248         }
249
250         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
251                                  &hint_byte, 1);
252         if (ret)
253                 return ret;
254
255         if (isize > actual_end)
256                 inline_len = min_t(u64, isize, actual_end);
257         ret = insert_inline_extent(trans, root, inode, start,
258                                    inline_len, compressed_size,
259                                    compress_type, compressed_pages);
260         if (ret) {
261                 btrfs_abort_transaction(trans, root, ret);
262                 return ret;
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 pdflush sent them down.
325  */
326 static noinline int compress_file_range(struct inode *inode,
327                                         struct page *locked_page,
328                                         u64 start, u64 end,
329                                         struct async_cow *async_cow,
330                                         int *num_added)
331 {
332         struct btrfs_root *root = BTRFS_I(inode)->root;
333         struct btrfs_trans_handle *trans;
334         u64 num_bytes;
335         u64 blocksize = root->sectorsize;
336         u64 actual_end;
337         u64 isize = i_size_read(inode);
338         int ret = 0;
339         struct page **pages = NULL;
340         unsigned long nr_pages;
341         unsigned long nr_pages_ret = 0;
342         unsigned long total_compressed = 0;
343         unsigned long total_in = 0;
344         unsigned long max_compressed = 128 * 1024;
345         unsigned long max_uncompressed = 128 * 1024;
346         int i;
347         int will_compress;
348         int compress_type = root->fs_info->compress_type;
349
350         /* if this is a small write inside eof, kick off a defrag */
351         if ((end - start + 1) < 16 * 1024 &&
352             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
353                 btrfs_add_inode_defrag(NULL, inode);
354
355         actual_end = min_t(u64, isize, end + 1);
356 again:
357         will_compress = 0;
358         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
359         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
360
361         /*
362          * we don't want to send crud past the end of i_size through
363          * compression, that's just a waste of CPU time.  So, if the
364          * end of the file is before the start of our current
365          * requested range of bytes, we bail out to the uncompressed
366          * cleanup code that can deal with all of this.
367          *
368          * It isn't really the fastest way to fix things, but this is a
369          * very uncommon corner.
370          */
371         if (actual_end <= start)
372                 goto cleanup_and_bail_uncompressed;
373
374         total_compressed = actual_end - start;
375
376         /* we want to make sure that amount of ram required to uncompress
377          * an extent is reasonable, so we limit the total size in ram
378          * of a compressed extent to 128k.  This is a crucial number
379          * because it also controls how easily we can spread reads across
380          * cpus for decompression.
381          *
382          * We also want to make sure the amount of IO required to do
383          * a random read is reasonably small, so we limit the size of
384          * a compressed extent to 128k.
385          */
386         total_compressed = min(total_compressed, max_uncompressed);
387         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
388         num_bytes = max(blocksize,  num_bytes);
389         total_in = 0;
390         ret = 0;
391
392         /*
393          * we do compression for mount -o compress and when the
394          * inode has not been flagged as nocompress.  This flag can
395          * change at any time if we discover bad compression ratios.
396          */
397         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
398             (btrfs_test_opt(root, COMPRESS) ||
399              (BTRFS_I(inode)->force_compress) ||
400              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
401                 WARN_ON(pages);
402                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
403                 if (!pages) {
404                         /* just bail out to the uncompressed code */
405                         goto cont;
406                 }
407
408                 if (BTRFS_I(inode)->force_compress)
409                         compress_type = BTRFS_I(inode)->force_compress;
410
411                 ret = btrfs_compress_pages(compress_type,
412                                            inode->i_mapping, start,
413                                            total_compressed, pages,
414                                            nr_pages, &nr_pages_ret,
415                                            &total_in,
416                                            &total_compressed,
417                                            max_compressed);
418
419                 if (!ret) {
420                         unsigned long offset = total_compressed &
421                                 (PAGE_CACHE_SIZE - 1);
422                         struct page *page = pages[nr_pages_ret - 1];
423                         char *kaddr;
424
425                         /* zero the tail end of the last page, we might be
426                          * sending it down to disk
427                          */
428                         if (offset) {
429                                 kaddr = kmap_atomic(page);
430                                 memset(kaddr + offset, 0,
431                                        PAGE_CACHE_SIZE - offset);
432                                 kunmap_atomic(kaddr);
433                         }
434                         will_compress = 1;
435                 }
436         }
437 cont:
438         if (start == 0) {
439                 trans = btrfs_join_transaction(root);
440                 if (IS_ERR(trans)) {
441                         ret = PTR_ERR(trans);
442                         trans = NULL;
443                         goto cleanup_and_out;
444                 }
445                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
446
447                 /* lets try to make an inline extent */
448                 if (ret || total_in < (actual_end - start)) {
449                         /* we didn't compress the entire range, try
450                          * to make an uncompressed inline extent.
451                          */
452                         ret = cow_file_range_inline(trans, root, inode,
453                                                     start, end, 0, 0, NULL);
454                 } else {
455                         /* try making a compressed inline extent */
456                         ret = cow_file_range_inline(trans, root, inode,
457                                                     start, end,
458                                                     total_compressed,
459                                                     compress_type, pages);
460                 }
461                 if (ret <= 0) {
462                         /*
463                          * inline extent creation worked or returned error,
464                          * we don't need to create any more async work items.
465                          * Unlock and free up our temp pages.
466                          */
467                         extent_clear_unlock_delalloc(inode,
468                              &BTRFS_I(inode)->io_tree,
469                              start, end, NULL,
470                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
471                              EXTENT_CLEAR_DELALLOC |
472                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
473
474                         btrfs_end_transaction(trans, root);
475                         goto free_pages_out;
476                 }
477                 btrfs_end_transaction(trans, root);
478         }
479
480         if (will_compress) {
481                 /*
482                  * we aren't doing an inline extent round the compressed size
483                  * up to a block size boundary so the allocator does sane
484                  * things
485                  */
486                 total_compressed = (total_compressed + blocksize - 1) &
487                         ~(blocksize - 1);
488
489                 /*
490                  * one last check to make sure the compression is really a
491                  * win, compare the page count read with the blocks on disk
492                  */
493                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
494                         ~(PAGE_CACHE_SIZE - 1);
495                 if (total_compressed >= total_in) {
496                         will_compress = 0;
497                 } else {
498                         num_bytes = total_in;
499                 }
500         }
501         if (!will_compress && pages) {
502                 /*
503                  * the compression code ran but failed to make things smaller,
504                  * free any pages it allocated and our page pointer array
505                  */
506                 for (i = 0; i < nr_pages_ret; i++) {
507                         WARN_ON(pages[i]->mapping);
508                         page_cache_release(pages[i]);
509                 }
510                 kfree(pages);
511                 pages = NULL;
512                 total_compressed = 0;
513                 nr_pages_ret = 0;
514
515                 /* flag the file so we don't compress in the future */
516                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
517                     !(BTRFS_I(inode)->force_compress)) {
518                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
519                 }
520         }
521         if (will_compress) {
522                 *num_added += 1;
523
524                 /* the async work queues will take care of doing actual
525                  * allocation on disk for these compressed pages,
526                  * and will submit them to the elevator.
527                  */
528                 add_async_extent(async_cow, start, num_bytes,
529                                  total_compressed, pages, nr_pages_ret,
530                                  compress_type);
531
532                 if (start + num_bytes < end) {
533                         start += num_bytes;
534                         pages = NULL;
535                         cond_resched();
536                         goto again;
537                 }
538         } else {
539 cleanup_and_bail_uncompressed:
540                 /*
541                  * No compression, but we still need to write the pages in
542                  * the file we've been given so far.  redirty the locked
543                  * page if it corresponds to our extent and set things up
544                  * for the async work queue to run cow_file_range to do
545                  * the normal delalloc dance
546                  */
547                 if (page_offset(locked_page) >= start &&
548                     page_offset(locked_page) <= end) {
549                         __set_page_dirty_nobuffers(locked_page);
550                         /* unlocked later on in the async handlers */
551                 }
552                 add_async_extent(async_cow, start, end - start + 1,
553                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
554                 *num_added += 1;
555         }
556
557 out:
558         return ret;
559
560 free_pages_out:
561         for (i = 0; i < nr_pages_ret; i++) {
562                 WARN_ON(pages[i]->mapping);
563                 page_cache_release(pages[i]);
564         }
565         kfree(pages);
566
567         goto out;
568
569 cleanup_and_out:
570         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
571                                      start, end, NULL,
572                                      EXTENT_CLEAR_UNLOCK_PAGE |
573                                      EXTENT_CLEAR_DIRTY |
574                                      EXTENT_CLEAR_DELALLOC |
575                                      EXTENT_SET_WRITEBACK |
576                                      EXTENT_END_WRITEBACK);
577         if (!trans || IS_ERR(trans))
578                 btrfs_error(root->fs_info, ret, "Failed to join transaction");
579         else
580                 btrfs_abort_transaction(trans, root, ret);
581         goto free_pages_out;
582 }
583
584 /*
585  * phase two of compressed writeback.  This is the ordered portion
586  * of the code, which only gets called in the order the work was
587  * queued.  We walk all the async extents created by compress_file_range
588  * and send them down to the disk.
589  */
590 static noinline int submit_compressed_extents(struct inode *inode,
591                                               struct async_cow *async_cow)
592 {
593         struct async_extent *async_extent;
594         u64 alloc_hint = 0;
595         struct btrfs_trans_handle *trans;
596         struct btrfs_key ins;
597         struct extent_map *em;
598         struct btrfs_root *root = BTRFS_I(inode)->root;
599         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
600         struct extent_io_tree *io_tree;
601         int ret = 0;
602
603         if (list_empty(&async_cow->extents))
604                 return 0;
605
606
607         while (!list_empty(&async_cow->extents)) {
608                 async_extent = list_entry(async_cow->extents.next,
609                                           struct async_extent, list);
610                 list_del(&async_extent->list);
611
612                 io_tree = &BTRFS_I(inode)->io_tree;
613
614 retry:
615                 /* did the compression code fall back to uncompressed IO? */
616                 if (!async_extent->pages) {
617                         int page_started = 0;
618                         unsigned long nr_written = 0;
619
620                         lock_extent(io_tree, async_extent->start,
621                                          async_extent->start +
622                                          async_extent->ram_size - 1);
623
624                         /* allocate blocks */
625                         ret = cow_file_range(inode, async_cow->locked_page,
626                                              async_extent->start,
627                                              async_extent->start +
628                                              async_extent->ram_size - 1,
629                                              &page_started, &nr_written, 0);
630
631                         /* JDM XXX */
632
633                         /*
634                          * if page_started, cow_file_range inserted an
635                          * inline extent and took care of all the unlocking
636                          * and IO for us.  Otherwise, we need to submit
637                          * all those pages down to the drive.
638                          */
639                         if (!page_started && !ret)
640                                 extent_write_locked_range(io_tree,
641                                                   inode, async_extent->start,
642                                                   async_extent->start +
643                                                   async_extent->ram_size - 1,
644                                                   btrfs_get_extent,
645                                                   WB_SYNC_ALL);
646                         kfree(async_extent);
647                         cond_resched();
648                         continue;
649                 }
650
651                 lock_extent(io_tree, async_extent->start,
652                             async_extent->start + async_extent->ram_size - 1);
653
654                 trans = btrfs_join_transaction(root);
655                 if (IS_ERR(trans)) {
656                         ret = PTR_ERR(trans);
657                 } else {
658                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
659                         ret = btrfs_reserve_extent(trans, root,
660                                            async_extent->compressed_size,
661                                            async_extent->compressed_size,
662                                            0, alloc_hint, &ins, 1);
663                         if (ret)
664                                 btrfs_abort_transaction(trans, root, ret);
665                         btrfs_end_transaction(trans, root);
666                 }
667
668                 if (ret) {
669                         int i;
670                         for (i = 0; i < async_extent->nr_pages; i++) {
671                                 WARN_ON(async_extent->pages[i]->mapping);
672                                 page_cache_release(async_extent->pages[i]);
673                         }
674                         kfree(async_extent->pages);
675                         async_extent->nr_pages = 0;
676                         async_extent->pages = NULL;
677                         unlock_extent(io_tree, async_extent->start,
678                                       async_extent->start +
679                                       async_extent->ram_size - 1);
680                         if (ret == -ENOSPC)
681                                 goto retry;
682                         goto out_free; /* JDM: Requeue? */
683                 }
684
685                 /*
686                  * here we're doing allocation and writeback of the
687                  * compressed pages
688                  */
689                 btrfs_drop_extent_cache(inode, async_extent->start,
690                                         async_extent->start +
691                                         async_extent->ram_size - 1, 0);
692
693                 em = alloc_extent_map();
694                 BUG_ON(!em); /* -ENOMEM */
695                 em->start = async_extent->start;
696                 em->len = async_extent->ram_size;
697                 em->orig_start = em->start;
698
699                 em->block_start = ins.objectid;
700                 em->block_len = ins.offset;
701                 em->bdev = root->fs_info->fs_devices->latest_bdev;
702                 em->compress_type = async_extent->compress_type;
703                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
704                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
705
706                 while (1) {
707                         write_lock(&em_tree->lock);
708                         ret = add_extent_mapping(em_tree, em);
709                         write_unlock(&em_tree->lock);
710                         if (ret != -EEXIST) {
711                                 free_extent_map(em);
712                                 break;
713                         }
714                         btrfs_drop_extent_cache(inode, async_extent->start,
715                                                 async_extent->start +
716                                                 async_extent->ram_size - 1, 0);
717                 }
718
719                 ret = btrfs_add_ordered_extent_compress(inode,
720                                                 async_extent->start,
721                                                 ins.objectid,
722                                                 async_extent->ram_size,
723                                                 ins.offset,
724                                                 BTRFS_ORDERED_COMPRESSED,
725                                                 async_extent->compress_type);
726                 BUG_ON(ret); /* -ENOMEM */
727
728                 /*
729                  * clear dirty, set writeback and unlock the pages.
730                  */
731                 extent_clear_unlock_delalloc(inode,
732                                 &BTRFS_I(inode)->io_tree,
733                                 async_extent->start,
734                                 async_extent->start +
735                                 async_extent->ram_size - 1,
736                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
737                                 EXTENT_CLEAR_UNLOCK |
738                                 EXTENT_CLEAR_DELALLOC |
739                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
740
741                 ret = btrfs_submit_compressed_write(inode,
742                                     async_extent->start,
743                                     async_extent->ram_size,
744                                     ins.objectid,
745                                     ins.offset, async_extent->pages,
746                                     async_extent->nr_pages);
747
748                 BUG_ON(ret); /* -ENOMEM */
749                 alloc_hint = ins.objectid + ins.offset;
750                 kfree(async_extent);
751                 cond_resched();
752         }
753         ret = 0;
754 out:
755         return ret;
756 out_free:
757         kfree(async_extent);
758         goto out;
759 }
760
761 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
762                                       u64 num_bytes)
763 {
764         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
765         struct extent_map *em;
766         u64 alloc_hint = 0;
767
768         read_lock(&em_tree->lock);
769         em = search_extent_mapping(em_tree, start, num_bytes);
770         if (em) {
771                 /*
772                  * if block start isn't an actual block number then find the
773                  * first block in this inode and use that as a hint.  If that
774                  * block is also bogus then just don't worry about it.
775                  */
776                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
777                         free_extent_map(em);
778                         em = search_extent_mapping(em_tree, 0, 0);
779                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
780                                 alloc_hint = em->block_start;
781                         if (em)
782                                 free_extent_map(em);
783                 } else {
784                         alloc_hint = em->block_start;
785                         free_extent_map(em);
786                 }
787         }
788         read_unlock(&em_tree->lock);
789
790         return alloc_hint;
791 }
792
793 /*
794  * when extent_io.c finds a delayed allocation range in the file,
795  * the call backs end up in this code.  The basic idea is to
796  * allocate extents on disk for the range, and create ordered data structs
797  * in ram to track those extents.
798  *
799  * locked_page is the page that writepage had locked already.  We use
800  * it to make sure we don't do extra locks or unlocks.
801  *
802  * *page_started is set to one if we unlock locked_page and do everything
803  * required to start IO on it.  It may be clean and already done with
804  * IO when we return.
805  */
806 static noinline int cow_file_range(struct inode *inode,
807                                    struct page *locked_page,
808                                    u64 start, u64 end, int *page_started,
809                                    unsigned long *nr_written,
810                                    int unlock)
811 {
812         struct btrfs_root *root = BTRFS_I(inode)->root;
813         struct btrfs_trans_handle *trans;
814         u64 alloc_hint = 0;
815         u64 num_bytes;
816         unsigned long ram_size;
817         u64 disk_num_bytes;
818         u64 cur_alloc_size;
819         u64 blocksize = root->sectorsize;
820         struct btrfs_key ins;
821         struct extent_map *em;
822         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
823         int ret = 0;
824
825         BUG_ON(btrfs_is_free_space_inode(root, inode));
826         trans = btrfs_join_transaction(root);
827         if (IS_ERR(trans)) {
828                 extent_clear_unlock_delalloc(inode,
829                              &BTRFS_I(inode)->io_tree,
830                              start, end, NULL,
831                              EXTENT_CLEAR_UNLOCK_PAGE |
832                              EXTENT_CLEAR_UNLOCK |
833                              EXTENT_CLEAR_DELALLOC |
834                              EXTENT_CLEAR_DIRTY |
835                              EXTENT_SET_WRITEBACK |
836                              EXTENT_END_WRITEBACK);
837                 return PTR_ERR(trans);
838         }
839         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
840
841         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
842         num_bytes = max(blocksize,  num_bytes);
843         disk_num_bytes = num_bytes;
844         ret = 0;
845
846         /* if this is a small write inside eof, kick off defrag */
847         if (num_bytes < 64 * 1024 &&
848             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
849                 btrfs_add_inode_defrag(trans, inode);
850
851         if (start == 0) {
852                 /* lets try to make an inline extent */
853                 ret = cow_file_range_inline(trans, root, inode,
854                                             start, end, 0, 0, NULL);
855                 if (ret == 0) {
856                         extent_clear_unlock_delalloc(inode,
857                                      &BTRFS_I(inode)->io_tree,
858                                      start, end, NULL,
859                                      EXTENT_CLEAR_UNLOCK_PAGE |
860                                      EXTENT_CLEAR_UNLOCK |
861                                      EXTENT_CLEAR_DELALLOC |
862                                      EXTENT_CLEAR_DIRTY |
863                                      EXTENT_SET_WRITEBACK |
864                                      EXTENT_END_WRITEBACK);
865
866                         *nr_written = *nr_written +
867                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
868                         *page_started = 1;
869                         goto out;
870                 } else if (ret < 0) {
871                         btrfs_abort_transaction(trans, root, ret);
872                         goto out_unlock;
873                 }
874         }
875
876         BUG_ON(disk_num_bytes >
877                btrfs_super_total_bytes(root->fs_info->super_copy));
878
879         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
880         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
881
882         while (disk_num_bytes > 0) {
883                 unsigned long op;
884
885                 cur_alloc_size = disk_num_bytes;
886                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
887                                            root->sectorsize, 0, alloc_hint,
888                                            &ins, 1);
889                 if (ret < 0) {
890                         btrfs_abort_transaction(trans, root, ret);
891                         goto out_unlock;
892                 }
893
894                 em = alloc_extent_map();
895                 BUG_ON(!em); /* -ENOMEM */
896                 em->start = start;
897                 em->orig_start = em->start;
898                 ram_size = ins.offset;
899                 em->len = ins.offset;
900
901                 em->block_start = ins.objectid;
902                 em->block_len = ins.offset;
903                 em->bdev = root->fs_info->fs_devices->latest_bdev;
904                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
905
906                 while (1) {
907                         write_lock(&em_tree->lock);
908                         ret = add_extent_mapping(em_tree, em);
909                         write_unlock(&em_tree->lock);
910                         if (ret != -EEXIST) {
911                                 free_extent_map(em);
912                                 break;
913                         }
914                         btrfs_drop_extent_cache(inode, start,
915                                                 start + ram_size - 1, 0);
916                 }
917
918                 cur_alloc_size = ins.offset;
919                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
920                                                ram_size, cur_alloc_size, 0);
921                 BUG_ON(ret); /* -ENOMEM */
922
923                 if (root->root_key.objectid ==
924                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
925                         ret = btrfs_reloc_clone_csums(inode, start,
926                                                       cur_alloc_size);
927                         if (ret) {
928                                 btrfs_abort_transaction(trans, root, ret);
929                                 goto out_unlock;
930                         }
931                 }
932
933                 if (disk_num_bytes < cur_alloc_size)
934                         break;
935
936                 /* we're not doing compressed IO, don't unlock the first
937                  * page (which the caller expects to stay locked), don't
938                  * clear any dirty bits and don't set any writeback bits
939                  *
940                  * Do set the Private2 bit so we know this page was properly
941                  * setup for writepage
942                  */
943                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
944                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
945                         EXTENT_SET_PRIVATE2;
946
947                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
948                                              start, start + ram_size - 1,
949                                              locked_page, op);
950                 disk_num_bytes -= cur_alloc_size;
951                 num_bytes -= cur_alloc_size;
952                 alloc_hint = ins.objectid + ins.offset;
953                 start += cur_alloc_size;
954         }
955         ret = 0;
956 out:
957         btrfs_end_transaction(trans, root);
958
959         return ret;
960 out_unlock:
961         extent_clear_unlock_delalloc(inode,
962                      &BTRFS_I(inode)->io_tree,
963                      start, end, NULL,
964                      EXTENT_CLEAR_UNLOCK_PAGE |
965                      EXTENT_CLEAR_UNLOCK |
966                      EXTENT_CLEAR_DELALLOC |
967                      EXTENT_CLEAR_DIRTY |
968                      EXTENT_SET_WRITEBACK |
969                      EXTENT_END_WRITEBACK);
970
971         goto out;
972 }
973
974 /*
975  * work queue call back to started compression on a file and pages
976  */
977 static noinline void async_cow_start(struct btrfs_work *work)
978 {
979         struct async_cow *async_cow;
980         int num_added = 0;
981         async_cow = container_of(work, struct async_cow, work);
982
983         compress_file_range(async_cow->inode, async_cow->locked_page,
984                             async_cow->start, async_cow->end, async_cow,
985                             &num_added);
986         if (num_added == 0)
987                 async_cow->inode = NULL;
988 }
989
990 /*
991  * work queue call back to submit previously compressed pages
992  */
993 static noinline void async_cow_submit(struct btrfs_work *work)
994 {
995         struct async_cow *async_cow;
996         struct btrfs_root *root;
997         unsigned long nr_pages;
998
999         async_cow = container_of(work, struct async_cow, work);
1000
1001         root = async_cow->root;
1002         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1003                 PAGE_CACHE_SHIFT;
1004
1005         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
1006
1007         if (atomic_read(&root->fs_info->async_delalloc_pages) <
1008             5 * 1042 * 1024 &&
1009             waitqueue_active(&root->fs_info->async_submit_wait))
1010                 wake_up(&root->fs_info->async_submit_wait);
1011
1012         if (async_cow->inode)
1013                 submit_compressed_extents(async_cow->inode, async_cow);
1014 }
1015
1016 static noinline void async_cow_free(struct btrfs_work *work)
1017 {
1018         struct async_cow *async_cow;
1019         async_cow = container_of(work, struct async_cow, work);
1020         kfree(async_cow);
1021 }
1022
1023 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1024                                 u64 start, u64 end, int *page_started,
1025                                 unsigned long *nr_written)
1026 {
1027         struct async_cow *async_cow;
1028         struct btrfs_root *root = BTRFS_I(inode)->root;
1029         unsigned long nr_pages;
1030         u64 cur_end;
1031         int limit = 10 * 1024 * 1042;
1032
1033         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1034                          1, 0, NULL, GFP_NOFS);
1035         while (start < end) {
1036                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1037                 BUG_ON(!async_cow); /* -ENOMEM */
1038                 async_cow->inode = inode;
1039                 async_cow->root = root;
1040                 async_cow->locked_page = locked_page;
1041                 async_cow->start = start;
1042
1043                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1044                         cur_end = end;
1045                 else
1046                         cur_end = min(end, start + 512 * 1024 - 1);
1047
1048                 async_cow->end = cur_end;
1049                 INIT_LIST_HEAD(&async_cow->extents);
1050
1051                 async_cow->work.func = async_cow_start;
1052                 async_cow->work.ordered_func = async_cow_submit;
1053                 async_cow->work.ordered_free = async_cow_free;
1054                 async_cow->work.flags = 0;
1055
1056                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1057                         PAGE_CACHE_SHIFT;
1058                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1059
1060                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1061                                    &async_cow->work);
1062
1063                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1064                         wait_event(root->fs_info->async_submit_wait,
1065                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1066                             limit));
1067                 }
1068
1069                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1070                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1071                         wait_event(root->fs_info->async_submit_wait,
1072                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1073                            0));
1074                 }
1075
1076                 *nr_written += nr_pages;
1077                 start = cur_end + 1;
1078         }
1079         *page_started = 1;
1080         return 0;
1081 }
1082
1083 static noinline int csum_exist_in_range(struct btrfs_root *root,
1084                                         u64 bytenr, u64 num_bytes)
1085 {
1086         int ret;
1087         struct btrfs_ordered_sum *sums;
1088         LIST_HEAD(list);
1089
1090         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1091                                        bytenr + num_bytes - 1, &list, 0);
1092         if (ret == 0 && list_empty(&list))
1093                 return 0;
1094
1095         while (!list_empty(&list)) {
1096                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1097                 list_del(&sums->list);
1098                 kfree(sums);
1099         }
1100         return 1;
1101 }
1102
1103 /*
1104  * when nowcow writeback call back.  This checks for snapshots or COW copies
1105  * of the extents that exist in the file, and COWs the file as required.
1106  *
1107  * If no cow copies or snapshots exist, we write directly to the existing
1108  * blocks on disk
1109  */
1110 static noinline int run_delalloc_nocow(struct inode *inode,
1111                                        struct page *locked_page,
1112                               u64 start, u64 end, int *page_started, int force,
1113                               unsigned long *nr_written)
1114 {
1115         struct btrfs_root *root = BTRFS_I(inode)->root;
1116         struct btrfs_trans_handle *trans;
1117         struct extent_buffer *leaf;
1118         struct btrfs_path *path;
1119         struct btrfs_file_extent_item *fi;
1120         struct btrfs_key found_key;
1121         u64 cow_start;
1122         u64 cur_offset;
1123         u64 extent_end;
1124         u64 extent_offset;
1125         u64 disk_bytenr;
1126         u64 num_bytes;
1127         int extent_type;
1128         int ret, err;
1129         int type;
1130         int nocow;
1131         int check_prev = 1;
1132         bool nolock;
1133         u64 ino = btrfs_ino(inode);
1134
1135         path = btrfs_alloc_path();
1136         if (!path)
1137                 return -ENOMEM;
1138
1139         nolock = btrfs_is_free_space_inode(root, inode);
1140
1141         if (nolock)
1142                 trans = btrfs_join_transaction_nolock(root);
1143         else
1144                 trans = btrfs_join_transaction(root);
1145
1146         if (IS_ERR(trans)) {
1147                 btrfs_free_path(path);
1148                 return PTR_ERR(trans);
1149         }
1150
1151         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1152
1153         cow_start = (u64)-1;
1154         cur_offset = start;
1155         while (1) {
1156                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1157                                                cur_offset, 0);
1158                 if (ret < 0) {
1159                         btrfs_abort_transaction(trans, root, ret);
1160                         goto error;
1161                 }
1162                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1163                         leaf = path->nodes[0];
1164                         btrfs_item_key_to_cpu(leaf, &found_key,
1165                                               path->slots[0] - 1);
1166                         if (found_key.objectid == ino &&
1167                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1168                                 path->slots[0]--;
1169                 }
1170                 check_prev = 0;
1171 next_slot:
1172                 leaf = path->nodes[0];
1173                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1174                         ret = btrfs_next_leaf(root, path);
1175                         if (ret < 0) {
1176                                 btrfs_abort_transaction(trans, root, ret);
1177                                 goto error;
1178                         }
1179                         if (ret > 0)
1180                                 break;
1181                         leaf = path->nodes[0];
1182                 }
1183
1184                 nocow = 0;
1185                 disk_bytenr = 0;
1186                 num_bytes = 0;
1187                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1188
1189                 if (found_key.objectid > ino ||
1190                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1191                     found_key.offset > end)
1192                         break;
1193
1194                 if (found_key.offset > cur_offset) {
1195                         extent_end = found_key.offset;
1196                         extent_type = 0;
1197                         goto out_check;
1198                 }
1199
1200                 fi = btrfs_item_ptr(leaf, path->slots[0],
1201                                     struct btrfs_file_extent_item);
1202                 extent_type = btrfs_file_extent_type(leaf, fi);
1203
1204                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1205                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1206                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1207                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1208                         extent_end = found_key.offset +
1209                                 btrfs_file_extent_num_bytes(leaf, fi);
1210                         if (extent_end <= start) {
1211                                 path->slots[0]++;
1212                                 goto next_slot;
1213                         }
1214                         if (disk_bytenr == 0)
1215                                 goto out_check;
1216                         if (btrfs_file_extent_compression(leaf, fi) ||
1217                             btrfs_file_extent_encryption(leaf, fi) ||
1218                             btrfs_file_extent_other_encoding(leaf, fi))
1219                                 goto out_check;
1220                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1221                                 goto out_check;
1222                         if (btrfs_extent_readonly(root, disk_bytenr))
1223                                 goto out_check;
1224                         if (btrfs_cross_ref_exist(trans, root, ino,
1225                                                   found_key.offset -
1226                                                   extent_offset, disk_bytenr))
1227                                 goto out_check;
1228                         disk_bytenr += extent_offset;
1229                         disk_bytenr += cur_offset - found_key.offset;
1230                         num_bytes = min(end + 1, extent_end) - cur_offset;
1231                         /*
1232                          * force cow if csum exists in the range.
1233                          * this ensure that csum for a given extent are
1234                          * either valid or do not exist.
1235                          */
1236                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1237                                 goto out_check;
1238                         nocow = 1;
1239                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1240                         extent_end = found_key.offset +
1241                                 btrfs_file_extent_inline_len(leaf, fi);
1242                         extent_end = ALIGN(extent_end, root->sectorsize);
1243                 } else {
1244                         BUG_ON(1);
1245                 }
1246 out_check:
1247                 if (extent_end <= start) {
1248                         path->slots[0]++;
1249                         goto next_slot;
1250                 }
1251                 if (!nocow) {
1252                         if (cow_start == (u64)-1)
1253                                 cow_start = cur_offset;
1254                         cur_offset = extent_end;
1255                         if (cur_offset > end)
1256                                 break;
1257                         path->slots[0]++;
1258                         goto next_slot;
1259                 }
1260
1261                 btrfs_release_path(path);
1262                 if (cow_start != (u64)-1) {
1263                         ret = cow_file_range(inode, locked_page, cow_start,
1264                                         found_key.offset - 1, page_started,
1265                                         nr_written, 1);
1266                         if (ret) {
1267                                 btrfs_abort_transaction(trans, root, ret);
1268                                 goto error;
1269                         }
1270                         cow_start = (u64)-1;
1271                 }
1272
1273                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1274                         struct extent_map *em;
1275                         struct extent_map_tree *em_tree;
1276                         em_tree = &BTRFS_I(inode)->extent_tree;
1277                         em = alloc_extent_map();
1278                         BUG_ON(!em); /* -ENOMEM */
1279                         em->start = cur_offset;
1280                         em->orig_start = em->start;
1281                         em->len = num_bytes;
1282                         em->block_len = num_bytes;
1283                         em->block_start = disk_bytenr;
1284                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1285                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1286                         while (1) {
1287                                 write_lock(&em_tree->lock);
1288                                 ret = add_extent_mapping(em_tree, em);
1289                                 write_unlock(&em_tree->lock);
1290                                 if (ret != -EEXIST) {
1291                                         free_extent_map(em);
1292                                         break;
1293                                 }
1294                                 btrfs_drop_extent_cache(inode, em->start,
1295                                                 em->start + em->len - 1, 0);
1296                         }
1297                         type = BTRFS_ORDERED_PREALLOC;
1298                 } else {
1299                         type = BTRFS_ORDERED_NOCOW;
1300                 }
1301
1302                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1303                                                num_bytes, num_bytes, type);
1304                 BUG_ON(ret); /* -ENOMEM */
1305
1306                 if (root->root_key.objectid ==
1307                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1308                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1309                                                       num_bytes);
1310                         if (ret) {
1311                                 btrfs_abort_transaction(trans, root, ret);
1312                                 goto error;
1313                         }
1314                 }
1315
1316                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1317                                 cur_offset, cur_offset + num_bytes - 1,
1318                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1319                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1320                                 EXTENT_SET_PRIVATE2);
1321                 cur_offset = extent_end;
1322                 if (cur_offset > end)
1323                         break;
1324         }
1325         btrfs_release_path(path);
1326
1327         if (cur_offset <= end && cow_start == (u64)-1)
1328                 cow_start = cur_offset;
1329         if (cow_start != (u64)-1) {
1330                 ret = cow_file_range(inode, locked_page, cow_start, end,
1331                                      page_started, nr_written, 1);
1332                 if (ret) {
1333                         btrfs_abort_transaction(trans, root, ret);
1334                         goto error;
1335                 }
1336         }
1337
1338 error:
1339         if (nolock) {
1340                 err = btrfs_end_transaction_nolock(trans, root);
1341         } else {
1342                 err = btrfs_end_transaction(trans, root);
1343         }
1344         if (!ret)
1345                 ret = err;
1346
1347         btrfs_free_path(path);
1348         return ret;
1349 }
1350
1351 /*
1352  * extent_io.c call back to do delayed allocation processing
1353  */
1354 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1355                               u64 start, u64 end, int *page_started,
1356                               unsigned long *nr_written)
1357 {
1358         int ret;
1359         struct btrfs_root *root = BTRFS_I(inode)->root;
1360
1361         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1362                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1363                                          page_started, 1, nr_written);
1364         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1365                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1366                                          page_started, 0, nr_written);
1367         else if (!btrfs_test_opt(root, COMPRESS) &&
1368                  !(BTRFS_I(inode)->force_compress) &&
1369                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1370                 ret = cow_file_range(inode, locked_page, start, end,
1371                                       page_started, nr_written, 1);
1372         else
1373                 ret = cow_file_range_async(inode, locked_page, start, end,
1374                                            page_started, nr_written);
1375         return ret;
1376 }
1377
1378 static void btrfs_split_extent_hook(struct inode *inode,
1379                                     struct extent_state *orig, u64 split)
1380 {
1381         /* not delalloc, ignore it */
1382         if (!(orig->state & EXTENT_DELALLOC))
1383                 return;
1384
1385         spin_lock(&BTRFS_I(inode)->lock);
1386         BTRFS_I(inode)->outstanding_extents++;
1387         spin_unlock(&BTRFS_I(inode)->lock);
1388 }
1389
1390 /*
1391  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1392  * extents so we can keep track of new extents that are just merged onto old
1393  * extents, such as when we are doing sequential writes, so we can properly
1394  * account for the metadata space we'll need.
1395  */
1396 static void btrfs_merge_extent_hook(struct inode *inode,
1397                                     struct extent_state *new,
1398                                     struct extent_state *other)
1399 {
1400         /* not delalloc, ignore it */
1401         if (!(other->state & EXTENT_DELALLOC))
1402                 return;
1403
1404         spin_lock(&BTRFS_I(inode)->lock);
1405         BTRFS_I(inode)->outstanding_extents--;
1406         spin_unlock(&BTRFS_I(inode)->lock);
1407 }
1408
1409 /*
1410  * extent_io.c set_bit_hook, used to track delayed allocation
1411  * bytes in this file, and to maintain the list of inodes that
1412  * have pending delalloc work to be done.
1413  */
1414 static void btrfs_set_bit_hook(struct inode *inode,
1415                                struct extent_state *state, int *bits)
1416 {
1417
1418         /*
1419          * set_bit and clear bit hooks normally require _irqsave/restore
1420          * but in this case, we are only testing for the DELALLOC
1421          * bit, which is only set or cleared with irqs on
1422          */
1423         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1424                 struct btrfs_root *root = BTRFS_I(inode)->root;
1425                 u64 len = state->end + 1 - state->start;
1426                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1427
1428                 if (*bits & EXTENT_FIRST_DELALLOC) {
1429                         *bits &= ~EXTENT_FIRST_DELALLOC;
1430                 } else {
1431                         spin_lock(&BTRFS_I(inode)->lock);
1432                         BTRFS_I(inode)->outstanding_extents++;
1433                         spin_unlock(&BTRFS_I(inode)->lock);
1434                 }
1435
1436                 spin_lock(&root->fs_info->delalloc_lock);
1437                 BTRFS_I(inode)->delalloc_bytes += len;
1438                 root->fs_info->delalloc_bytes += len;
1439                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1440                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1441                                       &root->fs_info->delalloc_inodes);
1442                 }
1443                 spin_unlock(&root->fs_info->delalloc_lock);
1444         }
1445 }
1446
1447 /*
1448  * extent_io.c clear_bit_hook, see set_bit_hook for why
1449  */
1450 static void btrfs_clear_bit_hook(struct inode *inode,
1451                                  struct extent_state *state, int *bits)
1452 {
1453         /*
1454          * set_bit and clear bit hooks normally require _irqsave/restore
1455          * but in this case, we are only testing for the DELALLOC
1456          * bit, which is only set or cleared with irqs on
1457          */
1458         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1459                 struct btrfs_root *root = BTRFS_I(inode)->root;
1460                 u64 len = state->end + 1 - state->start;
1461                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1462
1463                 if (*bits & EXTENT_FIRST_DELALLOC) {
1464                         *bits &= ~EXTENT_FIRST_DELALLOC;
1465                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1466                         spin_lock(&BTRFS_I(inode)->lock);
1467                         BTRFS_I(inode)->outstanding_extents--;
1468                         spin_unlock(&BTRFS_I(inode)->lock);
1469                 }
1470
1471                 if (*bits & EXTENT_DO_ACCOUNTING)
1472                         btrfs_delalloc_release_metadata(inode, len);
1473
1474                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1475                     && do_list)
1476                         btrfs_free_reserved_data_space(inode, len);
1477
1478                 spin_lock(&root->fs_info->delalloc_lock);
1479                 root->fs_info->delalloc_bytes -= len;
1480                 BTRFS_I(inode)->delalloc_bytes -= len;
1481
1482                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1483                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1484                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1485                 }
1486                 spin_unlock(&root->fs_info->delalloc_lock);
1487         }
1488 }
1489
1490 /*
1491  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1492  * we don't create bios that span stripes or chunks
1493  */
1494 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1495                          size_t size, struct bio *bio,
1496                          unsigned long bio_flags)
1497 {
1498         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1499         struct btrfs_mapping_tree *map_tree;
1500         u64 logical = (u64)bio->bi_sector << 9;
1501         u64 length = 0;
1502         u64 map_length;
1503         int ret;
1504
1505         if (bio_flags & EXTENT_BIO_COMPRESSED)
1506                 return 0;
1507
1508         length = bio->bi_size;
1509         map_tree = &root->fs_info->mapping_tree;
1510         map_length = length;
1511         ret = btrfs_map_block(map_tree, READ, logical,
1512                               &map_length, NULL, 0);
1513         /* Will always return 0 or 1 with map_multi == NULL */
1514         BUG_ON(ret < 0);
1515         if (map_length < length + size)
1516                 return 1;
1517         return 0;
1518 }
1519
1520 /*
1521  * in order to insert checksums into the metadata in large chunks,
1522  * we wait until bio submission time.   All the pages in the bio are
1523  * checksummed and sums are attached onto the ordered extent record.
1524  *
1525  * At IO completion time the cums attached on the ordered extent record
1526  * are inserted into the btree
1527  */
1528 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1529                                     struct bio *bio, int mirror_num,
1530                                     unsigned long bio_flags,
1531                                     u64 bio_offset)
1532 {
1533         struct btrfs_root *root = BTRFS_I(inode)->root;
1534         int ret = 0;
1535
1536         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1537         BUG_ON(ret); /* -ENOMEM */
1538         return 0;
1539 }
1540
1541 /*
1542  * in order to insert checksums into the metadata in large chunks,
1543  * we wait until bio submission time.   All the pages in the bio are
1544  * checksummed and sums are attached onto the ordered extent record.
1545  *
1546  * At IO completion time the cums attached on the ordered extent record
1547  * are inserted into the btree
1548  */
1549 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1550                           int mirror_num, unsigned long bio_flags,
1551                           u64 bio_offset)
1552 {
1553         struct btrfs_root *root = BTRFS_I(inode)->root;
1554         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1555 }
1556
1557 /*
1558  * extent_io.c submission hook. This does the right thing for csum calculation
1559  * on write, or reading the csums from the tree before a read
1560  */
1561 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1562                           int mirror_num, unsigned long bio_flags,
1563                           u64 bio_offset)
1564 {
1565         struct btrfs_root *root = BTRFS_I(inode)->root;
1566         int ret = 0;
1567         int skip_sum;
1568         int metadata = 0;
1569
1570         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1571
1572         if (btrfs_is_free_space_inode(root, inode))
1573                 metadata = 2;
1574
1575         ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1576         if (ret)
1577                 return ret;
1578
1579         if (!(rw & REQ_WRITE)) {
1580                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1581                         return btrfs_submit_compressed_read(inode, bio,
1582                                                     mirror_num, bio_flags);
1583                 } else if (!skip_sum) {
1584                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1585                         if (ret)
1586                                 return ret;
1587                 }
1588                 goto mapit;
1589         } else if (!skip_sum) {
1590                 /* csum items have already been cloned */
1591                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1592                         goto mapit;
1593                 /* we're doing a write, do the async checksumming */
1594                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1595                                    inode, rw, bio, mirror_num,
1596                                    bio_flags, bio_offset,
1597                                    __btrfs_submit_bio_start,
1598                                    __btrfs_submit_bio_done);
1599         }
1600
1601 mapit:
1602         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1603 }
1604
1605 /*
1606  * given a list of ordered sums record them in the inode.  This happens
1607  * at IO completion time based on sums calculated at bio submission time.
1608  */
1609 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1610                              struct inode *inode, u64 file_offset,
1611                              struct list_head *list)
1612 {
1613         struct btrfs_ordered_sum *sum;
1614
1615         list_for_each_entry(sum, list, list) {
1616                 btrfs_csum_file_blocks(trans,
1617                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1618         }
1619         return 0;
1620 }
1621
1622 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1623                               struct extent_state **cached_state)
1624 {
1625         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1626                 WARN_ON(1);
1627         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1628                                    cached_state, GFP_NOFS);
1629 }
1630
1631 /* see btrfs_writepage_start_hook for details on why this is required */
1632 struct btrfs_writepage_fixup {
1633         struct page *page;
1634         struct btrfs_work work;
1635 };
1636
1637 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1638 {
1639         struct btrfs_writepage_fixup *fixup;
1640         struct btrfs_ordered_extent *ordered;
1641         struct extent_state *cached_state = NULL;
1642         struct page *page;
1643         struct inode *inode;
1644         u64 page_start;
1645         u64 page_end;
1646         int ret;
1647
1648         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1649         page = fixup->page;
1650 again:
1651         lock_page(page);
1652         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1653                 ClearPageChecked(page);
1654                 goto out_page;
1655         }
1656
1657         inode = page->mapping->host;
1658         page_start = page_offset(page);
1659         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1660
1661         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1662                          &cached_state);
1663
1664         /* already ordered? We're done */
1665         if (PagePrivate2(page))
1666                 goto out;
1667
1668         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1669         if (ordered) {
1670                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1671                                      page_end, &cached_state, GFP_NOFS);
1672                 unlock_page(page);
1673                 btrfs_start_ordered_extent(inode, ordered, 1);
1674                 btrfs_put_ordered_extent(ordered);
1675                 goto again;
1676         }
1677
1678         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1679         if (ret) {
1680                 mapping_set_error(page->mapping, ret);
1681                 end_extent_writepage(page, ret, page_start, page_end);
1682                 ClearPageChecked(page);
1683                 goto out;
1684          }
1685
1686         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1687         ClearPageChecked(page);
1688         set_page_dirty(page);
1689 out:
1690         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1691                              &cached_state, GFP_NOFS);
1692 out_page:
1693         unlock_page(page);
1694         page_cache_release(page);
1695         kfree(fixup);
1696 }
1697
1698 /*
1699  * There are a few paths in the higher layers of the kernel that directly
1700  * set the page dirty bit without asking the filesystem if it is a
1701  * good idea.  This causes problems because we want to make sure COW
1702  * properly happens and the data=ordered rules are followed.
1703  *
1704  * In our case any range that doesn't have the ORDERED bit set
1705  * hasn't been properly setup for IO.  We kick off an async process
1706  * to fix it up.  The async helper will wait for ordered extents, set
1707  * the delalloc bit and make it safe to write the page.
1708  */
1709 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1710 {
1711         struct inode *inode = page->mapping->host;
1712         struct btrfs_writepage_fixup *fixup;
1713         struct btrfs_root *root = BTRFS_I(inode)->root;
1714
1715         /* this page is properly in the ordered list */
1716         if (TestClearPagePrivate2(page))
1717                 return 0;
1718
1719         if (PageChecked(page))
1720                 return -EAGAIN;
1721
1722         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1723         if (!fixup)
1724                 return -EAGAIN;
1725
1726         SetPageChecked(page);
1727         page_cache_get(page);
1728         fixup->work.func = btrfs_writepage_fixup_worker;
1729         fixup->page = page;
1730         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1731         return -EBUSY;
1732 }
1733
1734 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1735                                        struct inode *inode, u64 file_pos,
1736                                        u64 disk_bytenr, u64 disk_num_bytes,
1737                                        u64 num_bytes, u64 ram_bytes,
1738                                        u8 compression, u8 encryption,
1739                                        u16 other_encoding, int extent_type)
1740 {
1741         struct btrfs_root *root = BTRFS_I(inode)->root;
1742         struct btrfs_file_extent_item *fi;
1743         struct btrfs_path *path;
1744         struct extent_buffer *leaf;
1745         struct btrfs_key ins;
1746         u64 hint;
1747         int ret;
1748
1749         path = btrfs_alloc_path();
1750         if (!path)
1751                 return -ENOMEM;
1752
1753         path->leave_spinning = 1;
1754
1755         /*
1756          * we may be replacing one extent in the tree with another.
1757          * The new extent is pinned in the extent map, and we don't want
1758          * to drop it from the cache until it is completely in the btree.
1759          *
1760          * So, tell btrfs_drop_extents to leave this extent in the cache.
1761          * the caller is expected to unpin it and allow it to be merged
1762          * with the others.
1763          */
1764         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1765                                  &hint, 0);
1766         if (ret)
1767                 goto out;
1768
1769         ins.objectid = btrfs_ino(inode);
1770         ins.offset = file_pos;
1771         ins.type = BTRFS_EXTENT_DATA_KEY;
1772         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1773         if (ret)
1774                 goto out;
1775         leaf = path->nodes[0];
1776         fi = btrfs_item_ptr(leaf, path->slots[0],
1777                             struct btrfs_file_extent_item);
1778         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1779         btrfs_set_file_extent_type(leaf, fi, extent_type);
1780         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1781         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1782         btrfs_set_file_extent_offset(leaf, fi, 0);
1783         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1784         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1785         btrfs_set_file_extent_compression(leaf, fi, compression);
1786         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1787         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1788
1789         btrfs_unlock_up_safe(path, 1);
1790         btrfs_set_lock_blocking(leaf);
1791
1792         btrfs_mark_buffer_dirty(leaf);
1793
1794         inode_add_bytes(inode, num_bytes);
1795
1796         ins.objectid = disk_bytenr;
1797         ins.offset = disk_num_bytes;
1798         ins.type = BTRFS_EXTENT_ITEM_KEY;
1799         ret = btrfs_alloc_reserved_file_extent(trans, root,
1800                                         root->root_key.objectid,
1801                                         btrfs_ino(inode), file_pos, &ins);
1802 out:
1803         btrfs_free_path(path);
1804
1805         return ret;
1806 }
1807
1808 /*
1809  * helper function for btrfs_finish_ordered_io, this
1810  * just reads in some of the csum leaves to prime them into ram
1811  * before we start the transaction.  It limits the amount of btree
1812  * reads required while inside the transaction.
1813  */
1814 /* as ordered data IO finishes, this gets called so we can finish
1815  * an ordered extent if the range of bytes in the file it covers are
1816  * fully written.
1817  */
1818 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1819 {
1820         struct btrfs_root *root = BTRFS_I(inode)->root;
1821         struct btrfs_trans_handle *trans = NULL;
1822         struct btrfs_ordered_extent *ordered_extent = NULL;
1823         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1824         struct extent_state *cached_state = NULL;
1825         int compress_type = 0;
1826         int ret;
1827         bool nolock;
1828
1829         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1830                                              end - start + 1);
1831         if (!ret)
1832                 return 0;
1833         BUG_ON(!ordered_extent); /* Logic error */
1834
1835         nolock = btrfs_is_free_space_inode(root, inode);
1836
1837         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1838                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1839                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1840                 if (!ret) {
1841                         if (nolock)
1842                                 trans = btrfs_join_transaction_nolock(root);
1843                         else
1844                                 trans = btrfs_join_transaction(root);
1845                         if (IS_ERR(trans))
1846                                 return PTR_ERR(trans);
1847                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1848                         ret = btrfs_update_inode_fallback(trans, root, inode);
1849                         if (ret) /* -ENOMEM or corruption */
1850                                 btrfs_abort_transaction(trans, root, ret);
1851                 }
1852                 goto out;
1853         }
1854
1855         lock_extent_bits(io_tree, ordered_extent->file_offset,
1856                          ordered_extent->file_offset + ordered_extent->len - 1,
1857                          0, &cached_state);
1858
1859         if (nolock)
1860                 trans = btrfs_join_transaction_nolock(root);
1861         else
1862                 trans = btrfs_join_transaction(root);
1863         if (IS_ERR(trans)) {
1864                 ret = PTR_ERR(trans);
1865                 trans = NULL;
1866                 goto out_unlock;
1867         }
1868         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1869
1870         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1871                 compress_type = ordered_extent->compress_type;
1872         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1873                 BUG_ON(compress_type);
1874                 ret = btrfs_mark_extent_written(trans, inode,
1875                                                 ordered_extent->file_offset,
1876                                                 ordered_extent->file_offset +
1877                                                 ordered_extent->len);
1878         } else {
1879                 BUG_ON(root == root->fs_info->tree_root);
1880                 ret = insert_reserved_file_extent(trans, inode,
1881                                                 ordered_extent->file_offset,
1882                                                 ordered_extent->start,
1883                                                 ordered_extent->disk_len,
1884                                                 ordered_extent->len,
1885                                                 ordered_extent->len,
1886                                                 compress_type, 0, 0,
1887                                                 BTRFS_FILE_EXTENT_REG);
1888                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1889                                    ordered_extent->file_offset,
1890                                    ordered_extent->len);
1891         }
1892         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1893                              ordered_extent->file_offset +
1894                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1895         if (ret < 0) {
1896                 btrfs_abort_transaction(trans, root, ret);
1897                 goto out;
1898         }
1899
1900         add_pending_csums(trans, inode, ordered_extent->file_offset,
1901                           &ordered_extent->list);
1902
1903         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1904         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1905                 ret = btrfs_update_inode_fallback(trans, root, inode);
1906                 if (ret) { /* -ENOMEM or corruption */
1907                         btrfs_abort_transaction(trans, root, ret);
1908                         goto out;
1909                 }
1910         }
1911         ret = 0;
1912 out:
1913         if (root != root->fs_info->tree_root)
1914                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1915         if (trans) {
1916                 if (nolock)
1917                         btrfs_end_transaction_nolock(trans, root);
1918                 else
1919                         btrfs_end_transaction(trans, root);
1920         }
1921
1922         /* once for us */
1923         btrfs_put_ordered_extent(ordered_extent);
1924         /* once for the tree */
1925         btrfs_put_ordered_extent(ordered_extent);
1926
1927         return 0;
1928 out_unlock:
1929         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1930                              ordered_extent->file_offset +
1931                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1932         goto out;
1933 }
1934
1935 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1936                                 struct extent_state *state, int uptodate)
1937 {
1938         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1939
1940         ClearPagePrivate2(page);
1941         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1942 }
1943
1944 /*
1945  * when reads are done, we need to check csums to verify the data is correct
1946  * if there's a match, we allow the bio to finish.  If not, the code in
1947  * extent_io.c will try to find good copies for us.
1948  */
1949 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1950                                struct extent_state *state)
1951 {
1952         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1953         struct inode *inode = page->mapping->host;
1954         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1955         char *kaddr;
1956         u64 private = ~(u32)0;
1957         int ret;
1958         struct btrfs_root *root = BTRFS_I(inode)->root;
1959         u32 csum = ~(u32)0;
1960
1961         if (PageChecked(page)) {
1962                 ClearPageChecked(page);
1963                 goto good;
1964         }
1965
1966         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1967                 goto good;
1968
1969         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1970             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1971                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1972                                   GFP_NOFS);
1973                 return 0;
1974         }
1975
1976         if (state && state->start == start) {
1977                 private = state->private;
1978                 ret = 0;
1979         } else {
1980                 ret = get_state_private(io_tree, start, &private);
1981         }
1982         kaddr = kmap_atomic(page);
1983         if (ret)
1984                 goto zeroit;
1985
1986         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1987         btrfs_csum_final(csum, (char *)&csum);
1988         if (csum != private)
1989                 goto zeroit;
1990
1991         kunmap_atomic(kaddr);
1992 good:
1993         return 0;
1994
1995 zeroit:
1996         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
1997                        "private %llu\n",
1998                        (unsigned long long)btrfs_ino(page->mapping->host),
1999                        (unsigned long long)start, csum,
2000                        (unsigned long long)private);
2001         memset(kaddr + offset, 1, end - start + 1);
2002         flush_dcache_page(page);
2003         kunmap_atomic(kaddr);
2004         if (private == 0)
2005                 return 0;
2006         return -EIO;
2007 }
2008
2009 struct delayed_iput {
2010         struct list_head list;
2011         struct inode *inode;
2012 };
2013
2014 /* JDM: If this is fs-wide, why can't we add a pointer to
2015  * btrfs_inode instead and avoid the allocation? */
2016 void btrfs_add_delayed_iput(struct inode *inode)
2017 {
2018         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2019         struct delayed_iput *delayed;
2020
2021         if (atomic_add_unless(&inode->i_count, -1, 1))
2022                 return;
2023
2024         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2025         delayed->inode = inode;
2026
2027         spin_lock(&fs_info->delayed_iput_lock);
2028         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2029         spin_unlock(&fs_info->delayed_iput_lock);
2030 }
2031
2032 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2033 {
2034         LIST_HEAD(list);
2035         struct btrfs_fs_info *fs_info = root->fs_info;
2036         struct delayed_iput *delayed;
2037         int empty;
2038
2039         spin_lock(&fs_info->delayed_iput_lock);
2040         empty = list_empty(&fs_info->delayed_iputs);
2041         spin_unlock(&fs_info->delayed_iput_lock);
2042         if (empty)
2043                 return;
2044
2045         down_read(&root->fs_info->cleanup_work_sem);
2046         spin_lock(&fs_info->delayed_iput_lock);
2047         list_splice_init(&fs_info->delayed_iputs, &list);
2048         spin_unlock(&fs_info->delayed_iput_lock);
2049
2050         while (!list_empty(&list)) {
2051                 delayed = list_entry(list.next, struct delayed_iput, list);
2052                 list_del(&delayed->list);
2053                 iput(delayed->inode);
2054                 kfree(delayed);
2055         }
2056         up_read(&root->fs_info->cleanup_work_sem);
2057 }
2058
2059 enum btrfs_orphan_cleanup_state {
2060         ORPHAN_CLEANUP_STARTED  = 1,
2061         ORPHAN_CLEANUP_DONE     = 2,
2062 };
2063
2064 /*
2065  * This is called in transaction commit time. If there are no orphan
2066  * files in the subvolume, it removes orphan item and frees block_rsv
2067  * structure.
2068  */
2069 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2070                               struct btrfs_root *root)
2071 {
2072         struct btrfs_block_rsv *block_rsv;
2073         int ret;
2074
2075         if (!list_empty(&root->orphan_list) ||
2076             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2077                 return;
2078
2079         spin_lock(&root->orphan_lock);
2080         if (!list_empty(&root->orphan_list)) {
2081                 spin_unlock(&root->orphan_lock);
2082                 return;
2083         }
2084
2085         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2086                 spin_unlock(&root->orphan_lock);
2087                 return;
2088         }
2089
2090         block_rsv = root->orphan_block_rsv;
2091         root->orphan_block_rsv = NULL;
2092         spin_unlock(&root->orphan_lock);
2093
2094         if (root->orphan_item_inserted &&
2095             btrfs_root_refs(&root->root_item) > 0) {
2096                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2097                                             root->root_key.objectid);
2098                 BUG_ON(ret);
2099                 root->orphan_item_inserted = 0;
2100         }
2101
2102         if (block_rsv) {
2103                 WARN_ON(block_rsv->size > 0);
2104                 btrfs_free_block_rsv(root, block_rsv);
2105         }
2106 }
2107
2108 /*
2109  * This creates an orphan entry for the given inode in case something goes
2110  * wrong in the middle of an unlink/truncate.
2111  *
2112  * NOTE: caller of this function should reserve 5 units of metadata for
2113  *       this function.
2114  */
2115 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2116 {
2117         struct btrfs_root *root = BTRFS_I(inode)->root;
2118         struct btrfs_block_rsv *block_rsv = NULL;
2119         int reserve = 0;
2120         int insert = 0;
2121         int ret;
2122
2123         if (!root->orphan_block_rsv) {
2124                 block_rsv = btrfs_alloc_block_rsv(root);
2125                 if (!block_rsv)
2126                         return -ENOMEM;
2127         }
2128
2129         spin_lock(&root->orphan_lock);
2130         if (!root->orphan_block_rsv) {
2131                 root->orphan_block_rsv = block_rsv;
2132         } else if (block_rsv) {
2133                 btrfs_free_block_rsv(root, block_rsv);
2134                 block_rsv = NULL;
2135         }
2136
2137         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2138                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2139 #if 0
2140                 /*
2141                  * For proper ENOSPC handling, we should do orphan
2142                  * cleanup when mounting. But this introduces backward
2143                  * compatibility issue.
2144                  */
2145                 if (!xchg(&root->orphan_item_inserted, 1))
2146                         insert = 2;
2147                 else
2148                         insert = 1;
2149 #endif
2150                 insert = 1;
2151         }
2152
2153         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2154                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2155                 reserve = 1;
2156         }
2157         spin_unlock(&root->orphan_lock);
2158
2159         /* grab metadata reservation from transaction handle */
2160         if (reserve) {
2161                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2162                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2163         }
2164
2165         /* insert an orphan item to track this unlinked/truncated file */
2166         if (insert >= 1) {
2167                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2168                 if (ret && ret != -EEXIST) {
2169                         btrfs_abort_transaction(trans, root, ret);
2170                         return ret;
2171                 }
2172                 ret = 0;
2173         }
2174
2175         /* insert an orphan item to track subvolume contains orphan files */
2176         if (insert >= 2) {
2177                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2178                                                root->root_key.objectid);
2179                 if (ret && ret != -EEXIST) {
2180                         btrfs_abort_transaction(trans, root, ret);
2181                         return ret;
2182                 }
2183         }
2184         return 0;
2185 }
2186
2187 /*
2188  * We have done the truncate/delete so we can go ahead and remove the orphan
2189  * item for this particular inode.
2190  */
2191 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2192 {
2193         struct btrfs_root *root = BTRFS_I(inode)->root;
2194         int delete_item = 0;
2195         int release_rsv = 0;
2196         int ret = 0;
2197
2198         spin_lock(&root->orphan_lock);
2199         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2200                 list_del_init(&BTRFS_I(inode)->i_orphan);
2201                 delete_item = 1;
2202         }
2203
2204         if (BTRFS_I(inode)->orphan_meta_reserved) {
2205                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2206                 release_rsv = 1;
2207         }
2208         spin_unlock(&root->orphan_lock);
2209
2210         if (trans && delete_item) {
2211                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2212                 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2213         }
2214
2215         if (release_rsv)
2216                 btrfs_orphan_release_metadata(inode);
2217
2218         return 0;
2219 }
2220
2221 /*
2222  * this cleans up any orphans that may be left on the list from the last use
2223  * of this root.
2224  */
2225 int btrfs_orphan_cleanup(struct btrfs_root *root)
2226 {
2227         struct btrfs_path *path;
2228         struct extent_buffer *leaf;
2229         struct btrfs_key key, found_key;
2230         struct btrfs_trans_handle *trans;
2231         struct inode *inode;
2232         u64 last_objectid = 0;
2233         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2234
2235         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2236                 return 0;
2237
2238         path = btrfs_alloc_path();
2239         if (!path) {
2240                 ret = -ENOMEM;
2241                 goto out;
2242         }
2243         path->reada = -1;
2244
2245         key.objectid = BTRFS_ORPHAN_OBJECTID;
2246         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2247         key.offset = (u64)-1;
2248
2249         while (1) {
2250                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2251                 if (ret < 0)
2252                         goto out;
2253
2254                 /*
2255                  * if ret == 0 means we found what we were searching for, which
2256                  * is weird, but possible, so only screw with path if we didn't
2257                  * find the key and see if we have stuff that matches
2258                  */
2259                 if (ret > 0) {
2260                         ret = 0;
2261                         if (path->slots[0] == 0)
2262                                 break;
2263                         path->slots[0]--;
2264                 }
2265
2266                 /* pull out the item */
2267                 leaf = path->nodes[0];
2268                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2269
2270                 /* make sure the item matches what we want */
2271                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2272                         break;
2273                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2274                         break;
2275
2276                 /* release the path since we're done with it */
2277                 btrfs_release_path(path);
2278
2279                 /*
2280                  * this is where we are basically btrfs_lookup, without the
2281                  * crossing root thing.  we store the inode number in the
2282                  * offset of the orphan item.
2283                  */
2284
2285                 if (found_key.offset == last_objectid) {
2286                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2287                                "stopping orphan cleanup\n");
2288                         ret = -EINVAL;
2289                         goto out;
2290                 }
2291
2292                 last_objectid = found_key.offset;
2293
2294                 found_key.objectid = found_key.offset;
2295                 found_key.type = BTRFS_INODE_ITEM_KEY;
2296                 found_key.offset = 0;
2297                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2298                 ret = PTR_RET(inode);
2299                 if (ret && ret != -ESTALE)
2300                         goto out;
2301
2302                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2303                         struct btrfs_root *dead_root;
2304                         struct btrfs_fs_info *fs_info = root->fs_info;
2305                         int is_dead_root = 0;
2306
2307                         /*
2308                          * this is an orphan in the tree root. Currently these
2309                          * could come from 2 sources:
2310                          *  a) a snapshot deletion in progress
2311                          *  b) a free space cache inode
2312                          * We need to distinguish those two, as the snapshot
2313                          * orphan must not get deleted.
2314                          * find_dead_roots already ran before us, so if this
2315                          * is a snapshot deletion, we should find the root
2316                          * in the dead_roots list
2317                          */
2318                         spin_lock(&fs_info->trans_lock);
2319                         list_for_each_entry(dead_root, &fs_info->dead_roots,
2320                                             root_list) {
2321                                 if (dead_root->root_key.objectid ==
2322                                     found_key.objectid) {
2323                                         is_dead_root = 1;
2324                                         break;
2325                                 }
2326                         }
2327                         spin_unlock(&fs_info->trans_lock);
2328                         if (is_dead_root) {
2329                                 /* prevent this orphan from being found again */
2330                                 key.offset = found_key.objectid - 1;
2331                                 continue;
2332                         }
2333                 }
2334                 /*
2335                  * Inode is already gone but the orphan item is still there,
2336                  * kill the orphan item.
2337                  */
2338                 if (ret == -ESTALE) {
2339                         trans = btrfs_start_transaction(root, 1);
2340                         if (IS_ERR(trans)) {
2341                                 ret = PTR_ERR(trans);
2342                                 goto out;
2343                         }
2344                         ret = btrfs_del_orphan_item(trans, root,
2345                                                     found_key.objectid);
2346                         BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2347                         btrfs_end_transaction(trans, root);
2348                         continue;
2349                 }
2350
2351                 /*
2352                  * add this inode to the orphan list so btrfs_orphan_del does
2353                  * the proper thing when we hit it
2354                  */
2355                 spin_lock(&root->orphan_lock);
2356                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2357                 spin_unlock(&root->orphan_lock);
2358
2359                 /* if we have links, this was a truncate, lets do that */
2360                 if (inode->i_nlink) {
2361                         if (!S_ISREG(inode->i_mode)) {
2362                                 WARN_ON(1);
2363                                 iput(inode);
2364                                 continue;
2365                         }
2366                         nr_truncate++;
2367                         ret = btrfs_truncate(inode);
2368                 } else {
2369                         nr_unlink++;
2370                 }
2371
2372                 /* this will do delete_inode and everything for us */
2373                 iput(inode);
2374                 if (ret)
2375                         goto out;
2376         }
2377         /* release the path since we're done with it */
2378         btrfs_release_path(path);
2379
2380         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2381
2382         if (root->orphan_block_rsv)
2383                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2384                                         (u64)-1);
2385
2386         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2387                 trans = btrfs_join_transaction(root);
2388                 if (!IS_ERR(trans))
2389                         btrfs_end_transaction(trans, root);
2390         }
2391
2392         if (nr_unlink)
2393                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2394         if (nr_truncate)
2395                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2396
2397 out:
2398         if (ret)
2399                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2400         btrfs_free_path(path);
2401         return ret;
2402 }
2403
2404 /*
2405  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2406  * don't find any xattrs, we know there can't be any acls.
2407  *
2408  * slot is the slot the inode is in, objectid is the objectid of the inode
2409  */
2410 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2411                                           int slot, u64 objectid)
2412 {
2413         u32 nritems = btrfs_header_nritems(leaf);
2414         struct btrfs_key found_key;
2415         int scanned = 0;
2416
2417         slot++;
2418         while (slot < nritems) {
2419                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2420
2421                 /* we found a different objectid, there must not be acls */
2422                 if (found_key.objectid != objectid)
2423                         return 0;
2424
2425                 /* we found an xattr, assume we've got an acl */
2426                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2427                         return 1;
2428
2429                 /*
2430                  * we found a key greater than an xattr key, there can't
2431                  * be any acls later on
2432                  */
2433                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2434                         return 0;
2435
2436                 slot++;
2437                 scanned++;
2438
2439                 /*
2440                  * it goes inode, inode backrefs, xattrs, extents,
2441                  * so if there are a ton of hard links to an inode there can
2442                  * be a lot of backrefs.  Don't waste time searching too hard,
2443                  * this is just an optimization
2444                  */
2445                 if (scanned >= 8)
2446                         break;
2447         }
2448         /* we hit the end of the leaf before we found an xattr or
2449          * something larger than an xattr.  We have to assume the inode
2450          * has acls
2451          */
2452         return 1;
2453 }
2454
2455 /*
2456  * read an inode from the btree into the in-memory inode
2457  */
2458 static void btrfs_read_locked_inode(struct inode *inode)
2459 {
2460         struct btrfs_path *path;
2461         struct extent_buffer *leaf;
2462         struct btrfs_inode_item *inode_item;
2463         struct btrfs_timespec *tspec;
2464         struct btrfs_root *root = BTRFS_I(inode)->root;
2465         struct btrfs_key location;
2466         int maybe_acls;
2467         u32 rdev;
2468         int ret;
2469         bool filled = false;
2470
2471         ret = btrfs_fill_inode(inode, &rdev);
2472         if (!ret)
2473                 filled = true;
2474
2475         path = btrfs_alloc_path();
2476         if (!path)
2477                 goto make_bad;
2478
2479         path->leave_spinning = 1;
2480         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2481
2482         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2483         if (ret)
2484                 goto make_bad;
2485
2486         leaf = path->nodes[0];
2487
2488         if (filled)
2489                 goto cache_acl;
2490
2491         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2492                                     struct btrfs_inode_item);
2493         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2494         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2495         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2496         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2497         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2498
2499         tspec = btrfs_inode_atime(inode_item);
2500         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2501         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2502
2503         tspec = btrfs_inode_mtime(inode_item);
2504         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2505         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2506
2507         tspec = btrfs_inode_ctime(inode_item);
2508         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2509         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2510
2511         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2512         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2513         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2514         inode->i_generation = BTRFS_I(inode)->generation;
2515         inode->i_rdev = 0;
2516         rdev = btrfs_inode_rdev(leaf, inode_item);
2517
2518         BTRFS_I(inode)->index_cnt = (u64)-1;
2519         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2520 cache_acl:
2521         /*
2522          * try to precache a NULL acl entry for files that don't have
2523          * any xattrs or acls
2524          */
2525         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2526                                            btrfs_ino(inode));
2527         if (!maybe_acls)
2528                 cache_no_acl(inode);
2529
2530         btrfs_free_path(path);
2531
2532         switch (inode->i_mode & S_IFMT) {
2533         case S_IFREG:
2534                 inode->i_mapping->a_ops = &btrfs_aops;
2535                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2536                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2537                 inode->i_fop = &btrfs_file_operations;
2538                 inode->i_op = &btrfs_file_inode_operations;
2539                 break;
2540         case S_IFDIR:
2541                 inode->i_fop = &btrfs_dir_file_operations;
2542                 if (root == root->fs_info->tree_root)
2543                         inode->i_op = &btrfs_dir_ro_inode_operations;
2544                 else
2545                         inode->i_op = &btrfs_dir_inode_operations;
2546                 break;
2547         case S_IFLNK:
2548                 inode->i_op = &btrfs_symlink_inode_operations;
2549                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2550                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2551                 break;
2552         default:
2553                 inode->i_op = &btrfs_special_inode_operations;
2554                 init_special_inode(inode, inode->i_mode, rdev);
2555                 break;
2556         }
2557
2558         btrfs_update_iflags(inode);
2559         return;
2560
2561 make_bad:
2562         btrfs_free_path(path);
2563         make_bad_inode(inode);
2564 }
2565
2566 /*
2567  * given a leaf and an inode, copy the inode fields into the leaf
2568  */
2569 static void fill_inode_item(struct btrfs_trans_handle *trans,
2570                             struct extent_buffer *leaf,
2571                             struct btrfs_inode_item *item,
2572                             struct inode *inode)
2573 {
2574         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2575         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2576         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2577         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2578         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2579
2580         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2581                                inode->i_atime.tv_sec);
2582         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2583                                 inode->i_atime.tv_nsec);
2584
2585         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2586                                inode->i_mtime.tv_sec);
2587         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2588                                 inode->i_mtime.tv_nsec);
2589
2590         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2591                                inode->i_ctime.tv_sec);
2592         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2593                                 inode->i_ctime.tv_nsec);
2594
2595         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2596         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2597         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2598         btrfs_set_inode_transid(leaf, item, trans->transid);
2599         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2600         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2601         btrfs_set_inode_block_group(leaf, item, 0);
2602 }
2603
2604 /*
2605  * copy everything in the in-memory inode into the btree.
2606  */
2607 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2608                                 struct btrfs_root *root, struct inode *inode)
2609 {
2610         struct btrfs_inode_item *inode_item;
2611         struct btrfs_path *path;
2612         struct extent_buffer *leaf;
2613         int ret;
2614
2615         path = btrfs_alloc_path();
2616         if (!path)
2617                 return -ENOMEM;
2618
2619         path->leave_spinning = 1;
2620         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2621                                  1);
2622         if (ret) {
2623                 if (ret > 0)
2624                         ret = -ENOENT;
2625                 goto failed;
2626         }
2627
2628         btrfs_unlock_up_safe(path, 1);
2629         leaf = path->nodes[0];
2630         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2631                                     struct btrfs_inode_item);
2632
2633         fill_inode_item(trans, leaf, inode_item, inode);
2634         btrfs_mark_buffer_dirty(leaf);
2635         btrfs_set_inode_last_trans(trans, inode);
2636         ret = 0;
2637 failed:
2638         btrfs_free_path(path);
2639         return ret;
2640 }
2641
2642 /*
2643  * copy everything in the in-memory inode into the btree.
2644  */
2645 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2646                                 struct btrfs_root *root, struct inode *inode)
2647 {
2648         int ret;
2649
2650         /*
2651          * If the inode is a free space inode, we can deadlock during commit
2652          * if we put it into the delayed code.
2653          *
2654          * The data relocation inode should also be directly updated
2655          * without delay
2656          */
2657         if (!btrfs_is_free_space_inode(root, inode)
2658             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2659                 ret = btrfs_delayed_update_inode(trans, root, inode);
2660                 if (!ret)
2661                         btrfs_set_inode_last_trans(trans, inode);
2662                 return ret;
2663         }
2664
2665         return btrfs_update_inode_item(trans, root, inode);
2666 }
2667
2668 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2669                                 struct btrfs_root *root, struct inode *inode)
2670 {
2671         int ret;
2672
2673         ret = btrfs_update_inode(trans, root, inode);
2674         if (ret == -ENOSPC)
2675                 return btrfs_update_inode_item(trans, root, inode);
2676         return ret;
2677 }
2678
2679 /*
2680  * unlink helper that gets used here in inode.c and in the tree logging
2681  * recovery code.  It remove a link in a directory with a given name, and
2682  * also drops the back refs in the inode to the directory
2683  */
2684 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2685                                 struct btrfs_root *root,
2686                                 struct inode *dir, struct inode *inode,
2687                                 const char *name, int name_len)
2688 {
2689         struct btrfs_path *path;
2690         int ret = 0;
2691         struct extent_buffer *leaf;
2692         struct btrfs_dir_item *di;
2693         struct btrfs_key key;
2694         u64 index;
2695         u64 ino = btrfs_ino(inode);
2696         u64 dir_ino = btrfs_ino(dir);
2697
2698         path = btrfs_alloc_path();
2699         if (!path) {
2700                 ret = -ENOMEM;
2701                 goto out;
2702         }
2703
2704         path->leave_spinning = 1;
2705         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2706                                     name, name_len, -1);
2707         if (IS_ERR(di)) {
2708                 ret = PTR_ERR(di);
2709                 goto err;
2710         }
2711         if (!di) {
2712                 ret = -ENOENT;
2713                 goto err;
2714         }
2715         leaf = path->nodes[0];
2716         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2717         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2718         if (ret)
2719                 goto err;
2720         btrfs_release_path(path);
2721
2722         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2723                                   dir_ino, &index);
2724         if (ret) {
2725                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2726                        "inode %llu parent %llu\n", name_len, name,
2727                        (unsigned long long)ino, (unsigned long long)dir_ino);
2728                 btrfs_abort_transaction(trans, root, ret);
2729                 goto err;
2730         }
2731
2732         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2733         if (ret) {
2734                 btrfs_abort_transaction(trans, root, ret);
2735                 goto err;
2736         }
2737
2738         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2739                                          inode, dir_ino);
2740         if (ret != 0 && ret != -ENOENT) {
2741                 btrfs_abort_transaction(trans, root, ret);
2742                 goto err;
2743         }
2744
2745         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2746                                            dir, index);
2747         if (ret == -ENOENT)
2748                 ret = 0;
2749 err:
2750         btrfs_free_path(path);
2751         if (ret)
2752                 goto out;
2753
2754         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2755         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2756         btrfs_update_inode(trans, root, dir);
2757 out:
2758         return ret;
2759 }
2760
2761 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2762                        struct btrfs_root *root,
2763                        struct inode *dir, struct inode *inode,
2764                        const char *name, int name_len)
2765 {
2766         int ret;
2767         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2768         if (!ret) {
2769                 btrfs_drop_nlink(inode);
2770                 ret = btrfs_update_inode(trans, root, inode);
2771         }
2772         return ret;
2773 }
2774                 
2775
2776 /* helper to check if there is any shared block in the path */
2777 static int check_path_shared(struct btrfs_root *root,
2778                              struct btrfs_path *path)
2779 {
2780         struct extent_buffer *eb;
2781         int level;
2782         u64 refs = 1;
2783
2784         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2785                 int ret;
2786
2787                 if (!path->nodes[level])
2788                         break;
2789                 eb = path->nodes[level];
2790                 if (!btrfs_block_can_be_shared(root, eb))
2791                         continue;
2792                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2793                                                &refs, NULL);
2794                 if (refs > 1)
2795                         return 1;
2796         }
2797         return 0;
2798 }
2799
2800 /*
2801  * helper to start transaction for unlink and rmdir.
2802  *
2803  * unlink and rmdir are special in btrfs, they do not always free space.
2804  * so in enospc case, we should make sure they will free space before
2805  * allowing them to use the global metadata reservation.
2806  */
2807 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2808                                                        struct dentry *dentry)
2809 {
2810         struct btrfs_trans_handle *trans;
2811         struct btrfs_root *root = BTRFS_I(dir)->root;
2812         struct btrfs_path *path;
2813         struct btrfs_inode_ref *ref;
2814         struct btrfs_dir_item *di;
2815         struct inode *inode = dentry->d_inode;
2816         u64 index;
2817         int check_link = 1;
2818         int err = -ENOSPC;
2819         int ret;
2820         u64 ino = btrfs_ino(inode);
2821         u64 dir_ino = btrfs_ino(dir);
2822
2823         /*
2824          * 1 for the possible orphan item
2825          * 1 for the dir item
2826          * 1 for the dir index
2827          * 1 for the inode ref
2828          * 1 for the inode ref in the tree log
2829          * 2 for the dir entries in the log
2830          * 1 for the inode
2831          */
2832         trans = btrfs_start_transaction(root, 8);
2833         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2834                 return trans;
2835
2836         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2837                 return ERR_PTR(-ENOSPC);
2838
2839         /* check if there is someone else holds reference */
2840         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2841                 return ERR_PTR(-ENOSPC);
2842
2843         if (atomic_read(&inode->i_count) > 2)
2844                 return ERR_PTR(-ENOSPC);
2845
2846         if (xchg(&root->fs_info->enospc_unlink, 1))
2847                 return ERR_PTR(-ENOSPC);
2848
2849         path = btrfs_alloc_path();
2850         if (!path) {
2851                 root->fs_info->enospc_unlink = 0;
2852                 return ERR_PTR(-ENOMEM);
2853         }
2854
2855         /* 1 for the orphan item */
2856         trans = btrfs_start_transaction(root, 1);
2857         if (IS_ERR(trans)) {
2858                 btrfs_free_path(path);
2859                 root->fs_info->enospc_unlink = 0;
2860                 return trans;
2861         }
2862
2863         path->skip_locking = 1;
2864         path->search_commit_root = 1;
2865
2866         ret = btrfs_lookup_inode(trans, root, path,
2867                                 &BTRFS_I(dir)->location, 0);
2868         if (ret < 0) {
2869                 err = ret;
2870                 goto out;
2871         }
2872         if (ret == 0) {
2873                 if (check_path_shared(root, path))
2874                         goto out;
2875         } else {
2876                 check_link = 0;
2877         }
2878         btrfs_release_path(path);
2879
2880         ret = btrfs_lookup_inode(trans, root, path,
2881                                 &BTRFS_I(inode)->location, 0);
2882         if (ret < 0) {
2883                 err = ret;
2884                 goto out;
2885         }
2886         if (ret == 0) {
2887                 if (check_path_shared(root, path))
2888                         goto out;
2889         } else {
2890                 check_link = 0;
2891         }
2892         btrfs_release_path(path);
2893
2894         if (ret == 0 && S_ISREG(inode->i_mode)) {
2895                 ret = btrfs_lookup_file_extent(trans, root, path,
2896                                                ino, (u64)-1, 0);
2897                 if (ret < 0) {
2898                         err = ret;
2899                         goto out;
2900                 }
2901                 BUG_ON(ret == 0); /* Corruption */
2902                 if (check_path_shared(root, path))
2903                         goto out;
2904                 btrfs_release_path(path);
2905         }
2906
2907         if (!check_link) {
2908                 err = 0;
2909                 goto out;
2910         }
2911
2912         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2913                                 dentry->d_name.name, dentry->d_name.len, 0);
2914         if (IS_ERR(di)) {
2915                 err = PTR_ERR(di);
2916                 goto out;
2917         }
2918         if (di) {
2919                 if (check_path_shared(root, path))
2920                         goto out;
2921         } else {
2922                 err = 0;
2923                 goto out;
2924         }
2925         btrfs_release_path(path);
2926
2927         ref = btrfs_lookup_inode_ref(trans, root, path,
2928                                 dentry->d_name.name, dentry->d_name.len,
2929                                 ino, dir_ino, 0);
2930         if (IS_ERR(ref)) {
2931                 err = PTR_ERR(ref);
2932                 goto out;
2933         }
2934         BUG_ON(!ref); /* Logic error */
2935         if (check_path_shared(root, path))
2936                 goto out;
2937         index = btrfs_inode_ref_index(path->nodes[0], ref);
2938         btrfs_release_path(path);
2939
2940         /*
2941          * This is a commit root search, if we can lookup inode item and other
2942          * relative items in the commit root, it means the transaction of
2943          * dir/file creation has been committed, and the dir index item that we
2944          * delay to insert has also been inserted into the commit root. So
2945          * we needn't worry about the delayed insertion of the dir index item
2946          * here.
2947          */
2948         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2949                                 dentry->d_name.name, dentry->d_name.len, 0);
2950         if (IS_ERR(di)) {
2951                 err = PTR_ERR(di);
2952                 goto out;
2953         }
2954         BUG_ON(ret == -ENOENT);
2955         if (check_path_shared(root, path))
2956                 goto out;
2957
2958         err = 0;
2959 out:
2960         btrfs_free_path(path);
2961         /* Migrate the orphan reservation over */
2962         if (!err)
2963                 err = btrfs_block_rsv_migrate(trans->block_rsv,
2964                                 &root->fs_info->global_block_rsv,
2965                                 trans->bytes_reserved);
2966
2967         if (err) {
2968                 btrfs_end_transaction(trans, root);
2969                 root->fs_info->enospc_unlink = 0;
2970                 return ERR_PTR(err);
2971         }
2972
2973         trans->block_rsv = &root->fs_info->global_block_rsv;
2974         return trans;
2975 }
2976
2977 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2978                                struct btrfs_root *root)
2979 {
2980         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2981                 btrfs_block_rsv_release(root, trans->block_rsv,
2982                                         trans->bytes_reserved);
2983                 trans->block_rsv = &root->fs_info->trans_block_rsv;
2984                 BUG_ON(!root->fs_info->enospc_unlink);
2985                 root->fs_info->enospc_unlink = 0;
2986         }
2987         btrfs_end_transaction(trans, root);
2988 }
2989
2990 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2991 {
2992         struct btrfs_root *root = BTRFS_I(dir)->root;
2993         struct btrfs_trans_handle *trans;
2994         struct inode *inode = dentry->d_inode;
2995         int ret;
2996         unsigned long nr = 0;
2997
2998         trans = __unlink_start_trans(dir, dentry);
2999         if (IS_ERR(trans))
3000                 return PTR_ERR(trans);
3001
3002         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3003
3004         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3005                                  dentry->d_name.name, dentry->d_name.len);
3006         if (ret)
3007                 goto out;
3008
3009         if (inode->i_nlink == 0) {
3010                 ret = btrfs_orphan_add(trans, inode);
3011                 if (ret)
3012                         goto out;
3013         }
3014
3015 out:
3016         nr = trans->blocks_used;
3017         __unlink_end_trans(trans, root);
3018         btrfs_btree_balance_dirty(root, nr);
3019         return ret;
3020 }
3021
3022 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3023                         struct btrfs_root *root,
3024                         struct inode *dir, u64 objectid,
3025                         const char *name, int name_len)
3026 {
3027         struct btrfs_path *path;
3028         struct extent_buffer *leaf;
3029         struct btrfs_dir_item *di;
3030         struct btrfs_key key;
3031         u64 index;
3032         int ret;
3033         u64 dir_ino = btrfs_ino(dir);
3034
3035         path = btrfs_alloc_path();
3036         if (!path)
3037                 return -ENOMEM;
3038
3039         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3040                                    name, name_len, -1);
3041         if (IS_ERR_OR_NULL(di)) {
3042                 if (!di)
3043                         ret = -ENOENT;
3044                 else
3045                         ret = PTR_ERR(di);
3046                 goto out;
3047         }
3048
3049         leaf = path->nodes[0];
3050         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3051         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3052         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3053         if (ret) {
3054                 btrfs_abort_transaction(trans, root, ret);
3055                 goto out;
3056         }
3057         btrfs_release_path(path);
3058
3059         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3060                                  objectid, root->root_key.objectid,
3061                                  dir_ino, &index, name, name_len);
3062         if (ret < 0) {
3063                 if (ret != -ENOENT) {
3064                         btrfs_abort_transaction(trans, root, ret);
3065                         goto out;
3066                 }
3067                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3068                                                  name, name_len);
3069                 if (IS_ERR_OR_NULL(di)) {
3070                         if (!di)
3071                                 ret = -ENOENT;
3072                         else
3073                                 ret = PTR_ERR(di);
3074                         btrfs_abort_transaction(trans, root, ret);
3075                         goto out;
3076                 }
3077
3078                 leaf = path->nodes[0];
3079                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3080                 btrfs_release_path(path);
3081                 index = key.offset;
3082         }
3083         btrfs_release_path(path);
3084
3085         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3086         if (ret) {
3087                 btrfs_abort_transaction(trans, root, ret);
3088                 goto out;
3089         }
3090
3091         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3092         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3093         ret = btrfs_update_inode(trans, root, dir);
3094         if (ret)
3095                 btrfs_abort_transaction(trans, root, ret);
3096 out:
3097         btrfs_free_path(path);
3098         return ret;
3099 }
3100
3101 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3102 {
3103         struct inode *inode = dentry->d_inode;
3104         int err = 0;
3105         struct btrfs_root *root = BTRFS_I(dir)->root;
3106         struct btrfs_trans_handle *trans;
3107         unsigned long nr = 0;
3108
3109         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3110             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3111                 return -ENOTEMPTY;
3112
3113         trans = __unlink_start_trans(dir, dentry);
3114         if (IS_ERR(trans))
3115                 return PTR_ERR(trans);
3116
3117         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3118                 err = btrfs_unlink_subvol(trans, root, dir,
3119                                           BTRFS_I(inode)->location.objectid,
3120                                           dentry->d_name.name,
3121                                           dentry->d_name.len);
3122                 goto out;
3123         }
3124
3125         err = btrfs_orphan_add(trans, inode);
3126         if (err)
3127                 goto out;
3128
3129         /* now the directory is empty */
3130         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3131                                  dentry->d_name.name, dentry->d_name.len);
3132         if (!err)
3133                 btrfs_i_size_write(inode, 0);
3134 out:
3135         nr = trans->blocks_used;
3136         __unlink_end_trans(trans, root);
3137         btrfs_btree_balance_dirty(root, nr);
3138
3139         return err;
3140 }
3141
3142 /*
3143  * this can truncate away extent items, csum items and directory items.
3144  * It starts at a high offset and removes keys until it can't find
3145  * any higher than new_size
3146  *
3147  * csum items that cross the new i_size are truncated to the new size
3148  * as well.
3149  *
3150  * min_type is the minimum key type to truncate down to.  If set to 0, this
3151  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3152  */
3153 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3154                                struct btrfs_root *root,
3155                                struct inode *inode,
3156                                u64 new_size, u32 min_type)
3157 {
3158         struct btrfs_path *path;
3159         struct extent_buffer *leaf;
3160         struct btrfs_file_extent_item *fi;
3161         struct btrfs_key key;
3162         struct btrfs_key found_key;
3163         u64 extent_start = 0;
3164         u64 extent_num_bytes = 0;
3165         u64 extent_offset = 0;
3166         u64 item_end = 0;
3167         u64 mask = root->sectorsize - 1;
3168         u32 found_type = (u8)-1;
3169         int found_extent;
3170         int del_item;
3171         int pending_del_nr = 0;
3172         int pending_del_slot = 0;
3173         int extent_type = -1;
3174         int ret;
3175         int err = 0;
3176         u64 ino = btrfs_ino(inode);
3177
3178         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3179
3180         path = btrfs_alloc_path();
3181         if (!path)
3182                 return -ENOMEM;
3183         path->reada = -1;
3184
3185         if (root->ref_cows || root == root->fs_info->tree_root)
3186                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3187
3188         /*
3189          * This function is also used to drop the items in the log tree before
3190          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3191          * it is used to drop the loged items. So we shouldn't kill the delayed
3192          * items.
3193          */
3194         if (min_type == 0 && root == BTRFS_I(inode)->root)
3195                 btrfs_kill_delayed_inode_items(inode);
3196
3197         key.objectid = ino;
3198         key.offset = (u64)-1;
3199         key.type = (u8)-1;
3200
3201 search_again:
3202         path->leave_spinning = 1;
3203         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3204         if (ret < 0) {
3205                 err = ret;
3206                 goto out;
3207         }
3208
3209         if (ret > 0) {
3210                 /* there are no items in the tree for us to truncate, we're
3211                  * done
3212                  */
3213                 if (path->slots[0] == 0)
3214                         goto out;
3215                 path->slots[0]--;
3216         }
3217
3218         while (1) {
3219                 fi = NULL;
3220                 leaf = path->nodes[0];
3221                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3222                 found_type = btrfs_key_type(&found_key);
3223
3224                 if (found_key.objectid != ino)
3225                         break;
3226
3227                 if (found_type < min_type)
3228                         break;
3229
3230                 item_end = found_key.offset;
3231                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3232                         fi = btrfs_item_ptr(leaf, path->slots[0],
3233