Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland...
[pandora-kernel.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 "compat.h"
42 #include "ctree.h"
43 #include "disk-io.h"
44 #include "transaction.h"
45 #include "btrfs_inode.h"
46 #include "ioctl.h"
47 #include "print-tree.h"
48 #include "ordered-data.h"
49 #include "xattr.h"
50 #include "tree-log.h"
51 #include "volumes.h"
52 #include "compression.h"
53 #include "locking.h"
54 #include "free-space-cache.h"
55 #include "inode-map.h"
56
57 struct btrfs_iget_args {
58         u64 ino;
59         struct btrfs_root *root;
60 };
61
62 static const struct inode_operations btrfs_dir_inode_operations;
63 static const struct inode_operations btrfs_symlink_inode_operations;
64 static const struct inode_operations btrfs_dir_ro_inode_operations;
65 static const struct inode_operations btrfs_special_inode_operations;
66 static const struct inode_operations btrfs_file_inode_operations;
67 static const struct address_space_operations btrfs_aops;
68 static const struct address_space_operations btrfs_symlink_aops;
69 static const struct file_operations btrfs_dir_file_operations;
70 static struct extent_io_ops btrfs_extent_io_ops;
71
72 static struct kmem_cache *btrfs_inode_cachep;
73 struct kmem_cache *btrfs_trans_handle_cachep;
74 struct kmem_cache *btrfs_transaction_cachep;
75 struct kmem_cache *btrfs_path_cachep;
76 struct kmem_cache *btrfs_free_space_cachep;
77
78 #define S_SHIFT 12
79 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
80         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
81         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
82         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
83         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
84         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
85         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
86         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
87 };
88
89 static int btrfs_setsize(struct inode *inode, loff_t newsize);
90 static int btrfs_truncate(struct inode *inode);
91 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
92 static noinline int cow_file_range(struct inode *inode,
93                                    struct page *locked_page,
94                                    u64 start, u64 end, int *page_started,
95                                    unsigned long *nr_written, int unlock);
96 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
97                                 struct btrfs_root *root, struct inode *inode);
98
99 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
100                                      struct inode *inode,  struct inode *dir,
101                                      const struct qstr *qstr)
102 {
103         int err;
104
105         err = btrfs_init_acl(trans, inode, dir);
106         if (!err)
107                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
108         return err;
109 }
110
111 /*
112  * this does all the hard work for inserting an inline extent into
113  * the btree.  The caller should have done a btrfs_drop_extents so that
114  * no overlapping inline items exist in the btree
115  */
116 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
117                                 struct btrfs_root *root, struct inode *inode,
118                                 u64 start, size_t size, size_t compressed_size,
119                                 int compress_type,
120                                 struct page **compressed_pages)
121 {
122         struct btrfs_key key;
123         struct btrfs_path *path;
124         struct extent_buffer *leaf;
125         struct page *page = NULL;
126         char *kaddr;
127         unsigned long ptr;
128         struct btrfs_file_extent_item *ei;
129         int err = 0;
130         int ret;
131         size_t cur_size = size;
132         size_t datasize;
133         unsigned long offset;
134
135         if (compressed_size && compressed_pages)
136                 cur_size = compressed_size;
137
138         path = btrfs_alloc_path();
139         if (!path)
140                 return -ENOMEM;
141
142         path->leave_spinning = 1;
143
144         key.objectid = btrfs_ino(inode);
145         key.offset = start;
146         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
147         datasize = btrfs_file_extent_calc_inline_size(cur_size);
148
149         inode_add_bytes(inode, size);
150         ret = btrfs_insert_empty_item(trans, root, path, &key,
151                                       datasize);
152         BUG_ON(ret);
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, KM_USER0);
176                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
177                         kunmap_atomic(kaddr, KM_USER0);
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, KM_USER0);
190                 offset = start & (PAGE_CACHE_SIZE - 1);
191                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
192                 kunmap_atomic(kaddr, KM_USER0);
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         btrfs_update_inode(trans, root, inode);
209
210         return 0;
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         BUG_ON(ret);
253
254         if (isize > actual_end)
255                 inline_len = min_t(u64, isize, actual_end);
256         ret = insert_inline_extent(trans, root, inode, start,
257                                    inline_len, compressed_size,
258                                    compress_type, compressed_pages);
259         BUG_ON(ret);
260         btrfs_delalloc_release_metadata(inode, end + 1 - start);
261         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
262         return 0;
263 }
264
265 struct async_extent {
266         u64 start;
267         u64 ram_size;
268         u64 compressed_size;
269         struct page **pages;
270         unsigned long nr_pages;
271         int compress_type;
272         struct list_head list;
273 };
274
275 struct async_cow {
276         struct inode *inode;
277         struct btrfs_root *root;
278         struct page *locked_page;
279         u64 start;
280         u64 end;
281         struct list_head extents;
282         struct btrfs_work work;
283 };
284
285 static noinline int add_async_extent(struct async_cow *cow,
286                                      u64 start, u64 ram_size,
287                                      u64 compressed_size,
288                                      struct page **pages,
289                                      unsigned long nr_pages,
290                                      int compress_type)
291 {
292         struct async_extent *async_extent;
293
294         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
295         BUG_ON(!async_extent);
296         async_extent->start = start;
297         async_extent->ram_size = ram_size;
298         async_extent->compressed_size = compressed_size;
299         async_extent->pages = pages;
300         async_extent->nr_pages = nr_pages;
301         async_extent->compress_type = compress_type;
302         list_add_tail(&async_extent->list, &cow->extents);
303         return 0;
304 }
305
306 /*
307  * we create compressed extents in two phases.  The first
308  * phase compresses a range of pages that have already been
309  * locked (both pages and state bits are locked).
310  *
311  * This is done inside an ordered work queue, and the compression
312  * is spread across many cpus.  The actual IO submission is step
313  * two, and the ordered work queue takes care of making sure that
314  * happens in the same order things were put onto the queue by
315  * writepages and friends.
316  *
317  * If this code finds it can't get good compression, it puts an
318  * entry onto the work queue to write the uncompressed bytes.  This
319  * makes sure that both compressed inodes and uncompressed inodes
320  * are written in the same order that pdflush sent them down.
321  */
322 static noinline int compress_file_range(struct inode *inode,
323                                         struct page *locked_page,
324                                         u64 start, u64 end,
325                                         struct async_cow *async_cow,
326                                         int *num_added)
327 {
328         struct btrfs_root *root = BTRFS_I(inode)->root;
329         struct btrfs_trans_handle *trans;
330         u64 num_bytes;
331         u64 blocksize = root->sectorsize;
332         u64 actual_end;
333         u64 isize = i_size_read(inode);
334         int ret = 0;
335         struct page **pages = NULL;
336         unsigned long nr_pages;
337         unsigned long nr_pages_ret = 0;
338         unsigned long total_compressed = 0;
339         unsigned long total_in = 0;
340         unsigned long max_compressed = 128 * 1024;
341         unsigned long max_uncompressed = 128 * 1024;
342         int i;
343         int will_compress;
344         int compress_type = root->fs_info->compress_type;
345
346         /* if this is a small write inside eof, kick off a defragbot */
347         if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024)
348                 btrfs_add_inode_defrag(NULL, inode);
349
350         actual_end = min_t(u64, isize, end + 1);
351 again:
352         will_compress = 0;
353         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
354         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
355
356         /*
357          * we don't want to send crud past the end of i_size through
358          * compression, that's just a waste of CPU time.  So, if the
359          * end of the file is before the start of our current
360          * requested range of bytes, we bail out to the uncompressed
361          * cleanup code that can deal with all of this.
362          *
363          * It isn't really the fastest way to fix things, but this is a
364          * very uncommon corner.
365          */
366         if (actual_end <= start)
367                 goto cleanup_and_bail_uncompressed;
368
369         total_compressed = actual_end - start;
370
371         /* we want to make sure that amount of ram required to uncompress
372          * an extent is reasonable, so we limit the total size in ram
373          * of a compressed extent to 128k.  This is a crucial number
374          * because it also controls how easily we can spread reads across
375          * cpus for decompression.
376          *
377          * We also want to make sure the amount of IO required to do
378          * a random read is reasonably small, so we limit the size of
379          * a compressed extent to 128k.
380          */
381         total_compressed = min(total_compressed, max_uncompressed);
382         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
383         num_bytes = max(blocksize,  num_bytes);
384         total_in = 0;
385         ret = 0;
386
387         /*
388          * we do compression for mount -o compress and when the
389          * inode has not been flagged as nocompress.  This flag can
390          * change at any time if we discover bad compression ratios.
391          */
392         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
393             (btrfs_test_opt(root, COMPRESS) ||
394              (BTRFS_I(inode)->force_compress) ||
395              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
396                 WARN_ON(pages);
397                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
398                 if (!pages) {
399                         /* just bail out to the uncompressed code */
400                         goto cont;
401                 }
402
403                 if (BTRFS_I(inode)->force_compress)
404                         compress_type = BTRFS_I(inode)->force_compress;
405
406                 ret = btrfs_compress_pages(compress_type,
407                                            inode->i_mapping, start,
408                                            total_compressed, pages,
409                                            nr_pages, &nr_pages_ret,
410                                            &total_in,
411                                            &total_compressed,
412                                            max_compressed);
413
414                 if (!ret) {
415                         unsigned long offset = total_compressed &
416                                 (PAGE_CACHE_SIZE - 1);
417                         struct page *page = pages[nr_pages_ret - 1];
418                         char *kaddr;
419
420                         /* zero the tail end of the last page, we might be
421                          * sending it down to disk
422                          */
423                         if (offset) {
424                                 kaddr = kmap_atomic(page, KM_USER0);
425                                 memset(kaddr + offset, 0,
426                                        PAGE_CACHE_SIZE - offset);
427                                 kunmap_atomic(kaddr, KM_USER0);
428                         }
429                         will_compress = 1;
430                 }
431         }
432 cont:
433         if (start == 0) {
434                 trans = btrfs_join_transaction(root);
435                 BUG_ON(IS_ERR(trans));
436                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
437
438                 /* lets try to make an inline extent */
439                 if (ret || total_in < (actual_end - start)) {
440                         /* we didn't compress the entire range, try
441                          * to make an uncompressed inline extent.
442                          */
443                         ret = cow_file_range_inline(trans, root, inode,
444                                                     start, end, 0, 0, NULL);
445                 } else {
446                         /* try making a compressed inline extent */
447                         ret = cow_file_range_inline(trans, root, inode,
448                                                     start, end,
449                                                     total_compressed,
450                                                     compress_type, pages);
451                 }
452                 if (ret == 0) {
453                         /*
454                          * inline extent creation worked, we don't need
455                          * to create any more async work items.  Unlock
456                          * and free up our temp pages.
457                          */
458                         extent_clear_unlock_delalloc(inode,
459                              &BTRFS_I(inode)->io_tree,
460                              start, end, NULL,
461                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
462                              EXTENT_CLEAR_DELALLOC |
463                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
464
465                         btrfs_end_transaction(trans, root);
466                         goto free_pages_out;
467                 }
468                 btrfs_end_transaction(trans, root);
469         }
470
471         if (will_compress) {
472                 /*
473                  * we aren't doing an inline extent round the compressed size
474                  * up to a block size boundary so the allocator does sane
475                  * things
476                  */
477                 total_compressed = (total_compressed + blocksize - 1) &
478                         ~(blocksize - 1);
479
480                 /*
481                  * one last check to make sure the compression is really a
482                  * win, compare the page count read with the blocks on disk
483                  */
484                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
485                         ~(PAGE_CACHE_SIZE - 1);
486                 if (total_compressed >= total_in) {
487                         will_compress = 0;
488                 } else {
489                         num_bytes = total_in;
490                 }
491         }
492         if (!will_compress && pages) {
493                 /*
494                  * the compression code ran but failed to make things smaller,
495                  * free any pages it allocated and our page pointer array
496                  */
497                 for (i = 0; i < nr_pages_ret; i++) {
498                         WARN_ON(pages[i]->mapping);
499                         page_cache_release(pages[i]);
500                 }
501                 kfree(pages);
502                 pages = NULL;
503                 total_compressed = 0;
504                 nr_pages_ret = 0;
505
506                 /* flag the file so we don't compress in the future */
507                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
508                     !(BTRFS_I(inode)->force_compress)) {
509                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
510                 }
511         }
512         if (will_compress) {
513                 *num_added += 1;
514
515                 /* the async work queues will take care of doing actual
516                  * allocation on disk for these compressed pages,
517                  * and will submit them to the elevator.
518                  */
519                 add_async_extent(async_cow, start, num_bytes,
520                                  total_compressed, pages, nr_pages_ret,
521                                  compress_type);
522
523                 if (start + num_bytes < end) {
524                         start += num_bytes;
525                         pages = NULL;
526                         cond_resched();
527                         goto again;
528                 }
529         } else {
530 cleanup_and_bail_uncompressed:
531                 /*
532                  * No compression, but we still need to write the pages in
533                  * the file we've been given so far.  redirty the locked
534                  * page if it corresponds to our extent and set things up
535                  * for the async work queue to run cow_file_range to do
536                  * the normal delalloc dance
537                  */
538                 if (page_offset(locked_page) >= start &&
539                     page_offset(locked_page) <= end) {
540                         __set_page_dirty_nobuffers(locked_page);
541                         /* unlocked later on in the async handlers */
542                 }
543                 add_async_extent(async_cow, start, end - start + 1,
544                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
545                 *num_added += 1;
546         }
547
548 out:
549         return 0;
550
551 free_pages_out:
552         for (i = 0; i < nr_pages_ret; i++) {
553                 WARN_ON(pages[i]->mapping);
554                 page_cache_release(pages[i]);
555         }
556         kfree(pages);
557
558         goto out;
559 }
560
561 /*
562  * phase two of compressed writeback.  This is the ordered portion
563  * of the code, which only gets called in the order the work was
564  * queued.  We walk all the async extents created by compress_file_range
565  * and send them down to the disk.
566  */
567 static noinline int submit_compressed_extents(struct inode *inode,
568                                               struct async_cow *async_cow)
569 {
570         struct async_extent *async_extent;
571         u64 alloc_hint = 0;
572         struct btrfs_trans_handle *trans;
573         struct btrfs_key ins;
574         struct extent_map *em;
575         struct btrfs_root *root = BTRFS_I(inode)->root;
576         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
577         struct extent_io_tree *io_tree;
578         int ret = 0;
579
580         if (list_empty(&async_cow->extents))
581                 return 0;
582
583
584         while (!list_empty(&async_cow->extents)) {
585                 async_extent = list_entry(async_cow->extents.next,
586                                           struct async_extent, list);
587                 list_del(&async_extent->list);
588
589                 io_tree = &BTRFS_I(inode)->io_tree;
590
591 retry:
592                 /* did the compression code fall back to uncompressed IO? */
593                 if (!async_extent->pages) {
594                         int page_started = 0;
595                         unsigned long nr_written = 0;
596
597                         lock_extent(io_tree, async_extent->start,
598                                          async_extent->start +
599                                          async_extent->ram_size - 1, GFP_NOFS);
600
601                         /* allocate blocks */
602                         ret = cow_file_range(inode, async_cow->locked_page,
603                                              async_extent->start,
604                                              async_extent->start +
605                                              async_extent->ram_size - 1,
606                                              &page_started, &nr_written, 0);
607
608                         /*
609                          * if page_started, cow_file_range inserted an
610                          * inline extent and took care of all the unlocking
611                          * and IO for us.  Otherwise, we need to submit
612                          * all those pages down to the drive.
613                          */
614                         if (!page_started && !ret)
615                                 extent_write_locked_range(io_tree,
616                                                   inode, async_extent->start,
617                                                   async_extent->start +
618                                                   async_extent->ram_size - 1,
619                                                   btrfs_get_extent,
620                                                   WB_SYNC_ALL);
621                         kfree(async_extent);
622                         cond_resched();
623                         continue;
624                 }
625
626                 lock_extent(io_tree, async_extent->start,
627                             async_extent->start + async_extent->ram_size - 1,
628                             GFP_NOFS);
629
630                 trans = btrfs_join_transaction(root);
631                 BUG_ON(IS_ERR(trans));
632                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
633                 ret = btrfs_reserve_extent(trans, root,
634                                            async_extent->compressed_size,
635                                            async_extent->compressed_size,
636                                            0, alloc_hint,
637                                            (u64)-1, &ins, 1);
638                 btrfs_end_transaction(trans, root);
639
640                 if (ret) {
641                         int i;
642                         for (i = 0; i < async_extent->nr_pages; i++) {
643                                 WARN_ON(async_extent->pages[i]->mapping);
644                                 page_cache_release(async_extent->pages[i]);
645                         }
646                         kfree(async_extent->pages);
647                         async_extent->nr_pages = 0;
648                         async_extent->pages = NULL;
649                         unlock_extent(io_tree, async_extent->start,
650                                       async_extent->start +
651                                       async_extent->ram_size - 1, GFP_NOFS);
652                         goto retry;
653                 }
654
655                 /*
656                  * here we're doing allocation and writeback of the
657                  * compressed pages
658                  */
659                 btrfs_drop_extent_cache(inode, async_extent->start,
660                                         async_extent->start +
661                                         async_extent->ram_size - 1, 0);
662
663                 em = alloc_extent_map();
664                 BUG_ON(!em);
665                 em->start = async_extent->start;
666                 em->len = async_extent->ram_size;
667                 em->orig_start = em->start;
668
669                 em->block_start = ins.objectid;
670                 em->block_len = ins.offset;
671                 em->bdev = root->fs_info->fs_devices->latest_bdev;
672                 em->compress_type = async_extent->compress_type;
673                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
674                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
675
676                 while (1) {
677                         write_lock(&em_tree->lock);
678                         ret = add_extent_mapping(em_tree, em);
679                         write_unlock(&em_tree->lock);
680                         if (ret != -EEXIST) {
681                                 free_extent_map(em);
682                                 break;
683                         }
684                         btrfs_drop_extent_cache(inode, async_extent->start,
685                                                 async_extent->start +
686                                                 async_extent->ram_size - 1, 0);
687                 }
688
689                 ret = btrfs_add_ordered_extent_compress(inode,
690                                                 async_extent->start,
691                                                 ins.objectid,
692                                                 async_extent->ram_size,
693                                                 ins.offset,
694                                                 BTRFS_ORDERED_COMPRESSED,
695                                                 async_extent->compress_type);
696                 BUG_ON(ret);
697
698                 /*
699                  * clear dirty, set writeback and unlock the pages.
700                  */
701                 extent_clear_unlock_delalloc(inode,
702                                 &BTRFS_I(inode)->io_tree,
703                                 async_extent->start,
704                                 async_extent->start +
705                                 async_extent->ram_size - 1,
706                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
707                                 EXTENT_CLEAR_UNLOCK |
708                                 EXTENT_CLEAR_DELALLOC |
709                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
710
711                 ret = btrfs_submit_compressed_write(inode,
712                                     async_extent->start,
713                                     async_extent->ram_size,
714                                     ins.objectid,
715                                     ins.offset, async_extent->pages,
716                                     async_extent->nr_pages);
717
718                 BUG_ON(ret);
719                 alloc_hint = ins.objectid + ins.offset;
720                 kfree(async_extent);
721                 cond_resched();
722         }
723
724         return 0;
725 }
726
727 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
728                                       u64 num_bytes)
729 {
730         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
731         struct extent_map *em;
732         u64 alloc_hint = 0;
733
734         read_lock(&em_tree->lock);
735         em = search_extent_mapping(em_tree, start, num_bytes);
736         if (em) {
737                 /*
738                  * if block start isn't an actual block number then find the
739                  * first block in this inode and use that as a hint.  If that
740                  * block is also bogus then just don't worry about it.
741                  */
742                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
743                         free_extent_map(em);
744                         em = search_extent_mapping(em_tree, 0, 0);
745                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
746                                 alloc_hint = em->block_start;
747                         if (em)
748                                 free_extent_map(em);
749                 } else {
750                         alloc_hint = em->block_start;
751                         free_extent_map(em);
752                 }
753         }
754         read_unlock(&em_tree->lock);
755
756         return alloc_hint;
757 }
758
759 /*
760  * when extent_io.c finds a delayed allocation range in the file,
761  * the call backs end up in this code.  The basic idea is to
762  * allocate extents on disk for the range, and create ordered data structs
763  * in ram to track those extents.
764  *
765  * locked_page is the page that writepage had locked already.  We use
766  * it to make sure we don't do extra locks or unlocks.
767  *
768  * *page_started is set to one if we unlock locked_page and do everything
769  * required to start IO on it.  It may be clean and already done with
770  * IO when we return.
771  */
772 static noinline int cow_file_range(struct inode *inode,
773                                    struct page *locked_page,
774                                    u64 start, u64 end, int *page_started,
775                                    unsigned long *nr_written,
776                                    int unlock)
777 {
778         struct btrfs_root *root = BTRFS_I(inode)->root;
779         struct btrfs_trans_handle *trans;
780         u64 alloc_hint = 0;
781         u64 num_bytes;
782         unsigned long ram_size;
783         u64 disk_num_bytes;
784         u64 cur_alloc_size;
785         u64 blocksize = root->sectorsize;
786         struct btrfs_key ins;
787         struct extent_map *em;
788         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
789         int ret = 0;
790
791         BUG_ON(btrfs_is_free_space_inode(root, inode));
792         trans = btrfs_join_transaction(root);
793         BUG_ON(IS_ERR(trans));
794         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
795
796         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
797         num_bytes = max(blocksize,  num_bytes);
798         disk_num_bytes = num_bytes;
799         ret = 0;
800
801         /* if this is a small write inside eof, kick off defrag */
802         if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
803                 btrfs_add_inode_defrag(trans, inode);
804
805         if (start == 0) {
806                 /* lets try to make an inline extent */
807                 ret = cow_file_range_inline(trans, root, inode,
808                                             start, end, 0, 0, NULL);
809                 if (ret == 0) {
810                         extent_clear_unlock_delalloc(inode,
811                                      &BTRFS_I(inode)->io_tree,
812                                      start, end, NULL,
813                                      EXTENT_CLEAR_UNLOCK_PAGE |
814                                      EXTENT_CLEAR_UNLOCK |
815                                      EXTENT_CLEAR_DELALLOC |
816                                      EXTENT_CLEAR_DIRTY |
817                                      EXTENT_SET_WRITEBACK |
818                                      EXTENT_END_WRITEBACK);
819
820                         *nr_written = *nr_written +
821                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
822                         *page_started = 1;
823                         ret = 0;
824                         goto out;
825                 }
826         }
827
828         BUG_ON(disk_num_bytes >
829                btrfs_super_total_bytes(root->fs_info->super_copy));
830
831         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
832         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
833
834         while (disk_num_bytes > 0) {
835                 unsigned long op;
836
837                 cur_alloc_size = disk_num_bytes;
838                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
839                                            root->sectorsize, 0, alloc_hint,
840                                            (u64)-1, &ins, 1);
841                 BUG_ON(ret);
842
843                 em = alloc_extent_map();
844                 BUG_ON(!em);
845                 em->start = start;
846                 em->orig_start = em->start;
847                 ram_size = ins.offset;
848                 em->len = ins.offset;
849
850                 em->block_start = ins.objectid;
851                 em->block_len = ins.offset;
852                 em->bdev = root->fs_info->fs_devices->latest_bdev;
853                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
854
855                 while (1) {
856                         write_lock(&em_tree->lock);
857                         ret = add_extent_mapping(em_tree, em);
858                         write_unlock(&em_tree->lock);
859                         if (ret != -EEXIST) {
860                                 free_extent_map(em);
861                                 break;
862                         }
863                         btrfs_drop_extent_cache(inode, start,
864                                                 start + ram_size - 1, 0);
865                 }
866
867                 cur_alloc_size = ins.offset;
868                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
869                                                ram_size, cur_alloc_size, 0);
870                 BUG_ON(ret);
871
872                 if (root->root_key.objectid ==
873                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
874                         ret = btrfs_reloc_clone_csums(inode, start,
875                                                       cur_alloc_size);
876                         BUG_ON(ret);
877                 }
878
879                 if (disk_num_bytes < cur_alloc_size)
880                         break;
881
882                 /* we're not doing compressed IO, don't unlock the first
883                  * page (which the caller expects to stay locked), don't
884                  * clear any dirty bits and don't set any writeback bits
885                  *
886                  * Do set the Private2 bit so we know this page was properly
887                  * setup for writepage
888                  */
889                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
890                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
891                         EXTENT_SET_PRIVATE2;
892
893                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
894                                              start, start + ram_size - 1,
895                                              locked_page, op);
896                 disk_num_bytes -= cur_alloc_size;
897                 num_bytes -= cur_alloc_size;
898                 alloc_hint = ins.objectid + ins.offset;
899                 start += cur_alloc_size;
900         }
901 out:
902         ret = 0;
903         btrfs_end_transaction(trans, root);
904
905         return ret;
906 }
907
908 /*
909  * work queue call back to started compression on a file and pages
910  */
911 static noinline void async_cow_start(struct btrfs_work *work)
912 {
913         struct async_cow *async_cow;
914         int num_added = 0;
915         async_cow = container_of(work, struct async_cow, work);
916
917         compress_file_range(async_cow->inode, async_cow->locked_page,
918                             async_cow->start, async_cow->end, async_cow,
919                             &num_added);
920         if (num_added == 0)
921                 async_cow->inode = NULL;
922 }
923
924 /*
925  * work queue call back to submit previously compressed pages
926  */
927 static noinline void async_cow_submit(struct btrfs_work *work)
928 {
929         struct async_cow *async_cow;
930         struct btrfs_root *root;
931         unsigned long nr_pages;
932
933         async_cow = container_of(work, struct async_cow, work);
934
935         root = async_cow->root;
936         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
937                 PAGE_CACHE_SHIFT;
938
939         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
940
941         if (atomic_read(&root->fs_info->async_delalloc_pages) <
942             5 * 1042 * 1024 &&
943             waitqueue_active(&root->fs_info->async_submit_wait))
944                 wake_up(&root->fs_info->async_submit_wait);
945
946         if (async_cow->inode)
947                 submit_compressed_extents(async_cow->inode, async_cow);
948 }
949
950 static noinline void async_cow_free(struct btrfs_work *work)
951 {
952         struct async_cow *async_cow;
953         async_cow = container_of(work, struct async_cow, work);
954         kfree(async_cow);
955 }
956
957 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
958                                 u64 start, u64 end, int *page_started,
959                                 unsigned long *nr_written)
960 {
961         struct async_cow *async_cow;
962         struct btrfs_root *root = BTRFS_I(inode)->root;
963         unsigned long nr_pages;
964         u64 cur_end;
965         int limit = 10 * 1024 * 1042;
966
967         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
968                          1, 0, NULL, GFP_NOFS);
969         while (start < end) {
970                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
971                 BUG_ON(!async_cow);
972                 async_cow->inode = inode;
973                 async_cow->root = root;
974                 async_cow->locked_page = locked_page;
975                 async_cow->start = start;
976
977                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
978                         cur_end = end;
979                 else
980                         cur_end = min(end, start + 512 * 1024 - 1);
981
982                 async_cow->end = cur_end;
983                 INIT_LIST_HEAD(&async_cow->extents);
984
985                 async_cow->work.func = async_cow_start;
986                 async_cow->work.ordered_func = async_cow_submit;
987                 async_cow->work.ordered_free = async_cow_free;
988                 async_cow->work.flags = 0;
989
990                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
991                         PAGE_CACHE_SHIFT;
992                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
993
994                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
995                                    &async_cow->work);
996
997                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
998                         wait_event(root->fs_info->async_submit_wait,
999                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1000                             limit));
1001                 }
1002
1003                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1004                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1005                         wait_event(root->fs_info->async_submit_wait,
1006                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1007                            0));
1008                 }
1009
1010                 *nr_written += nr_pages;
1011                 start = cur_end + 1;
1012         }
1013         *page_started = 1;
1014         return 0;
1015 }
1016
1017 static noinline int csum_exist_in_range(struct btrfs_root *root,
1018                                         u64 bytenr, u64 num_bytes)
1019 {
1020         int ret;
1021         struct btrfs_ordered_sum *sums;
1022         LIST_HEAD(list);
1023
1024         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1025                                        bytenr + num_bytes - 1, &list, 0);
1026         if (ret == 0 && list_empty(&list))
1027                 return 0;
1028
1029         while (!list_empty(&list)) {
1030                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1031                 list_del(&sums->list);
1032                 kfree(sums);
1033         }
1034         return 1;
1035 }
1036
1037 /*
1038  * when nowcow writeback call back.  This checks for snapshots or COW copies
1039  * of the extents that exist in the file, and COWs the file as required.
1040  *
1041  * If no cow copies or snapshots exist, we write directly to the existing
1042  * blocks on disk
1043  */
1044 static noinline int run_delalloc_nocow(struct inode *inode,
1045                                        struct page *locked_page,
1046                               u64 start, u64 end, int *page_started, int force,
1047                               unsigned long *nr_written)
1048 {
1049         struct btrfs_root *root = BTRFS_I(inode)->root;
1050         struct btrfs_trans_handle *trans;
1051         struct extent_buffer *leaf;
1052         struct btrfs_path *path;
1053         struct btrfs_file_extent_item *fi;
1054         struct btrfs_key found_key;
1055         u64 cow_start;
1056         u64 cur_offset;
1057         u64 extent_end;
1058         u64 extent_offset;
1059         u64 disk_bytenr;
1060         u64 num_bytes;
1061         int extent_type;
1062         int ret;
1063         int type;
1064         int nocow;
1065         int check_prev = 1;
1066         bool nolock;
1067         u64 ino = btrfs_ino(inode);
1068
1069         path = btrfs_alloc_path();
1070         if (!path)
1071                 return -ENOMEM;
1072
1073         nolock = btrfs_is_free_space_inode(root, inode);
1074
1075         if (nolock)
1076                 trans = btrfs_join_transaction_nolock(root);
1077         else
1078                 trans = btrfs_join_transaction(root);
1079
1080         BUG_ON(IS_ERR(trans));
1081         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1082
1083         cow_start = (u64)-1;
1084         cur_offset = start;
1085         while (1) {
1086                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1087                                                cur_offset, 0);
1088                 BUG_ON(ret < 0);
1089                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1090                         leaf = path->nodes[0];
1091                         btrfs_item_key_to_cpu(leaf, &found_key,
1092                                               path->slots[0] - 1);
1093                         if (found_key.objectid == ino &&
1094                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1095                                 path->slots[0]--;
1096                 }
1097                 check_prev = 0;
1098 next_slot:
1099                 leaf = path->nodes[0];
1100                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1101                         ret = btrfs_next_leaf(root, path);
1102                         if (ret < 0)
1103                                 BUG_ON(1);
1104                         if (ret > 0)
1105                                 break;
1106                         leaf = path->nodes[0];
1107                 }
1108
1109                 nocow = 0;
1110                 disk_bytenr = 0;
1111                 num_bytes = 0;
1112                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1113
1114                 if (found_key.objectid > ino ||
1115                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1116                     found_key.offset > end)
1117                         break;
1118
1119                 if (found_key.offset > cur_offset) {
1120                         extent_end = found_key.offset;
1121                         extent_type = 0;
1122                         goto out_check;
1123                 }
1124
1125                 fi = btrfs_item_ptr(leaf, path->slots[0],
1126                                     struct btrfs_file_extent_item);
1127                 extent_type = btrfs_file_extent_type(leaf, fi);
1128
1129                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1130                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1131                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1132                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1133                         extent_end = found_key.offset +
1134                                 btrfs_file_extent_num_bytes(leaf, fi);
1135                         if (extent_end <= start) {
1136                                 path->slots[0]++;
1137                                 goto next_slot;
1138                         }
1139                         if (disk_bytenr == 0)
1140                                 goto out_check;
1141                         if (btrfs_file_extent_compression(leaf, fi) ||
1142                             btrfs_file_extent_encryption(leaf, fi) ||
1143                             btrfs_file_extent_other_encoding(leaf, fi))
1144                                 goto out_check;
1145                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1146                                 goto out_check;
1147                         if (btrfs_extent_readonly(root, disk_bytenr))
1148                                 goto out_check;
1149                         if (btrfs_cross_ref_exist(trans, root, ino,
1150                                                   found_key.offset -
1151                                                   extent_offset, disk_bytenr))
1152                                 goto out_check;
1153                         disk_bytenr += extent_offset;
1154                         disk_bytenr += cur_offset - found_key.offset;
1155                         num_bytes = min(end + 1, extent_end) - cur_offset;
1156                         /*
1157                          * force cow if csum exists in the range.
1158                          * this ensure that csum for a given extent are
1159                          * either valid or do not exist.
1160                          */
1161                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1162                                 goto out_check;
1163                         nocow = 1;
1164                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1165                         extent_end = found_key.offset +
1166                                 btrfs_file_extent_inline_len(leaf, fi);
1167                         extent_end = ALIGN(extent_end, root->sectorsize);
1168                 } else {
1169                         BUG_ON(1);
1170                 }
1171 out_check:
1172                 if (extent_end <= start) {
1173                         path->slots[0]++;
1174                         goto next_slot;
1175                 }
1176                 if (!nocow) {
1177                         if (cow_start == (u64)-1)
1178                                 cow_start = cur_offset;
1179                         cur_offset = extent_end;
1180                         if (cur_offset > end)
1181                                 break;
1182                         path->slots[0]++;
1183                         goto next_slot;
1184                 }
1185
1186                 btrfs_release_path(path);
1187                 if (cow_start != (u64)-1) {
1188                         ret = cow_file_range(inode, locked_page, cow_start,
1189                                         found_key.offset - 1, page_started,
1190                                         nr_written, 1);
1191                         BUG_ON(ret);
1192                         cow_start = (u64)-1;
1193                 }
1194
1195                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1196                         struct extent_map *em;
1197                         struct extent_map_tree *em_tree;
1198                         em_tree = &BTRFS_I(inode)->extent_tree;
1199                         em = alloc_extent_map();
1200                         BUG_ON(!em);
1201                         em->start = cur_offset;
1202                         em->orig_start = em->start;
1203                         em->len = num_bytes;
1204                         em->block_len = num_bytes;
1205                         em->block_start = disk_bytenr;
1206                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1207                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1208                         while (1) {
1209                                 write_lock(&em_tree->lock);
1210                                 ret = add_extent_mapping(em_tree, em);
1211                                 write_unlock(&em_tree->lock);
1212                                 if (ret != -EEXIST) {
1213                                         free_extent_map(em);
1214                                         break;
1215                                 }
1216                                 btrfs_drop_extent_cache(inode, em->start,
1217                                                 em->start + em->len - 1, 0);
1218                         }
1219                         type = BTRFS_ORDERED_PREALLOC;
1220                 } else {
1221                         type = BTRFS_ORDERED_NOCOW;
1222                 }
1223
1224                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1225                                                num_bytes, num_bytes, type);
1226                 BUG_ON(ret);
1227
1228                 if (root->root_key.objectid ==
1229                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1230                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1231                                                       num_bytes);
1232                         BUG_ON(ret);
1233                 }
1234
1235                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1236                                 cur_offset, cur_offset + num_bytes - 1,
1237                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1238                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1239                                 EXTENT_SET_PRIVATE2);
1240                 cur_offset = extent_end;
1241                 if (cur_offset > end)
1242                         break;
1243         }
1244         btrfs_release_path(path);
1245
1246         if (cur_offset <= end && cow_start == (u64)-1)
1247                 cow_start = cur_offset;
1248         if (cow_start != (u64)-1) {
1249                 ret = cow_file_range(inode, locked_page, cow_start, end,
1250                                      page_started, nr_written, 1);
1251                 BUG_ON(ret);
1252         }
1253
1254         if (nolock) {
1255                 ret = btrfs_end_transaction_nolock(trans, root);
1256                 BUG_ON(ret);
1257         } else {
1258                 ret = btrfs_end_transaction(trans, root);
1259                 BUG_ON(ret);
1260         }
1261         btrfs_free_path(path);
1262         return 0;
1263 }
1264
1265 /*
1266  * extent_io.c call back to do delayed allocation processing
1267  */
1268 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1269                               u64 start, u64 end, int *page_started,
1270                               unsigned long *nr_written)
1271 {
1272         int ret;
1273         struct btrfs_root *root = BTRFS_I(inode)->root;
1274
1275         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1276                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1277                                          page_started, 1, nr_written);
1278         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1279                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1280                                          page_started, 0, nr_written);
1281         else if (!btrfs_test_opt(root, COMPRESS) &&
1282                  !(BTRFS_I(inode)->force_compress) &&
1283                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1284                 ret = cow_file_range(inode, locked_page, start, end,
1285                                       page_started, nr_written, 1);
1286         else
1287                 ret = cow_file_range_async(inode, locked_page, start, end,
1288                                            page_started, nr_written);
1289         return ret;
1290 }
1291
1292 static void btrfs_split_extent_hook(struct inode *inode,
1293                                     struct extent_state *orig, u64 split)
1294 {
1295         /* not delalloc, ignore it */
1296         if (!(orig->state & EXTENT_DELALLOC))
1297                 return;
1298
1299         spin_lock(&BTRFS_I(inode)->lock);
1300         BTRFS_I(inode)->outstanding_extents++;
1301         spin_unlock(&BTRFS_I(inode)->lock);
1302 }
1303
1304 /*
1305  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1306  * extents so we can keep track of new extents that are just merged onto old
1307  * extents, such as when we are doing sequential writes, so we can properly
1308  * account for the metadata space we'll need.
1309  */
1310 static void btrfs_merge_extent_hook(struct inode *inode,
1311                                     struct extent_state *new,
1312                                     struct extent_state *other)
1313 {
1314         /* not delalloc, ignore it */
1315         if (!(other->state & EXTENT_DELALLOC))
1316                 return;
1317
1318         spin_lock(&BTRFS_I(inode)->lock);
1319         BTRFS_I(inode)->outstanding_extents--;
1320         spin_unlock(&BTRFS_I(inode)->lock);
1321 }
1322
1323 /*
1324  * extent_io.c set_bit_hook, used to track delayed allocation
1325  * bytes in this file, and to maintain the list of inodes that
1326  * have pending delalloc work to be done.
1327  */
1328 static void btrfs_set_bit_hook(struct inode *inode,
1329                                struct extent_state *state, int *bits)
1330 {
1331
1332         /*
1333          * set_bit and clear bit hooks normally require _irqsave/restore
1334          * but in this case, we are only testing for the DELALLOC
1335          * bit, which is only set or cleared with irqs on
1336          */
1337         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1338                 struct btrfs_root *root = BTRFS_I(inode)->root;
1339                 u64 len = state->end + 1 - state->start;
1340                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1341
1342                 if (*bits & EXTENT_FIRST_DELALLOC) {
1343                         *bits &= ~EXTENT_FIRST_DELALLOC;
1344                 } else {
1345                         spin_lock(&BTRFS_I(inode)->lock);
1346                         BTRFS_I(inode)->outstanding_extents++;
1347                         spin_unlock(&BTRFS_I(inode)->lock);
1348                 }
1349
1350                 spin_lock(&root->fs_info->delalloc_lock);
1351                 BTRFS_I(inode)->delalloc_bytes += len;
1352                 root->fs_info->delalloc_bytes += len;
1353                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1354                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1355                                       &root->fs_info->delalloc_inodes);
1356                 }
1357                 spin_unlock(&root->fs_info->delalloc_lock);
1358         }
1359 }
1360
1361 /*
1362  * extent_io.c clear_bit_hook, see set_bit_hook for why
1363  */
1364 static void btrfs_clear_bit_hook(struct inode *inode,
1365                                  struct extent_state *state, int *bits)
1366 {
1367         /*
1368          * set_bit and clear bit hooks normally require _irqsave/restore
1369          * but in this case, we are only testing for the DELALLOC
1370          * bit, which is only set or cleared with irqs on
1371          */
1372         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1373                 struct btrfs_root *root = BTRFS_I(inode)->root;
1374                 u64 len = state->end + 1 - state->start;
1375                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1376
1377                 if (*bits & EXTENT_FIRST_DELALLOC) {
1378                         *bits &= ~EXTENT_FIRST_DELALLOC;
1379                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1380                         spin_lock(&BTRFS_I(inode)->lock);
1381                         BTRFS_I(inode)->outstanding_extents--;
1382                         spin_unlock(&BTRFS_I(inode)->lock);
1383                 }
1384
1385                 if (*bits & EXTENT_DO_ACCOUNTING)
1386                         btrfs_delalloc_release_metadata(inode, len);
1387
1388                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1389                     && do_list)
1390                         btrfs_free_reserved_data_space(inode, len);
1391
1392                 spin_lock(&root->fs_info->delalloc_lock);
1393                 root->fs_info->delalloc_bytes -= len;
1394                 BTRFS_I(inode)->delalloc_bytes -= len;
1395
1396                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1397                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1398                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1399                 }
1400                 spin_unlock(&root->fs_info->delalloc_lock);
1401         }
1402 }
1403
1404 /*
1405  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1406  * we don't create bios that span stripes or chunks
1407  */
1408 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1409                          size_t size, struct bio *bio,
1410                          unsigned long bio_flags)
1411 {
1412         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1413         struct btrfs_mapping_tree *map_tree;
1414         u64 logical = (u64)bio->bi_sector << 9;
1415         u64 length = 0;
1416         u64 map_length;
1417         int ret;
1418
1419         if (bio_flags & EXTENT_BIO_COMPRESSED)
1420                 return 0;
1421
1422         length = bio->bi_size;
1423         map_tree = &root->fs_info->mapping_tree;
1424         map_length = length;
1425         ret = btrfs_map_block(map_tree, READ, logical,
1426                               &map_length, NULL, 0);
1427
1428         if (map_length < length + size)
1429                 return 1;
1430         return ret;
1431 }
1432
1433 /*
1434  * in order to insert checksums into the metadata in large chunks,
1435  * we wait until bio submission time.   All the pages in the bio are
1436  * checksummed and sums are attached onto the ordered extent record.
1437  *
1438  * At IO completion time the cums attached on the ordered extent record
1439  * are inserted into the btree
1440  */
1441 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1442                                     struct bio *bio, int mirror_num,
1443                                     unsigned long bio_flags,
1444                                     u64 bio_offset)
1445 {
1446         struct btrfs_root *root = BTRFS_I(inode)->root;
1447         int ret = 0;
1448
1449         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1450         BUG_ON(ret);
1451         return 0;
1452 }
1453
1454 /*
1455  * in order to insert checksums into the metadata in large chunks,
1456  * we wait until bio submission time.   All the pages in the bio are
1457  * checksummed and sums are attached onto the ordered extent record.
1458  *
1459  * At IO completion time the cums attached on the ordered extent record
1460  * are inserted into the btree
1461  */
1462 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1463                           int mirror_num, unsigned long bio_flags,
1464                           u64 bio_offset)
1465 {
1466         struct btrfs_root *root = BTRFS_I(inode)->root;
1467         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1468 }
1469
1470 /*
1471  * extent_io.c submission hook. This does the right thing for csum calculation
1472  * on write, or reading the csums from the tree before a read
1473  */
1474 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1475                           int mirror_num, unsigned long bio_flags,
1476                           u64 bio_offset)
1477 {
1478         struct btrfs_root *root = BTRFS_I(inode)->root;
1479         int ret = 0;
1480         int skip_sum;
1481
1482         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1483
1484         if (btrfs_is_free_space_inode(root, inode))
1485                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1486         else
1487                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1488         BUG_ON(ret);
1489
1490         if (!(rw & REQ_WRITE)) {
1491                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1492                         return btrfs_submit_compressed_read(inode, bio,
1493                                                     mirror_num, bio_flags);
1494                 } else if (!skip_sum) {
1495                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1496                         if (ret)
1497                                 return ret;
1498                 }
1499                 goto mapit;
1500         } else if (!skip_sum) {
1501                 /* csum items have already been cloned */
1502                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1503                         goto mapit;
1504                 /* we're doing a write, do the async checksumming */
1505                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1506                                    inode, rw, bio, mirror_num,
1507                                    bio_flags, bio_offset,
1508                                    __btrfs_submit_bio_start,
1509                                    __btrfs_submit_bio_done);
1510         }
1511
1512 mapit:
1513         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1514 }
1515
1516 /*
1517  * given a list of ordered sums record them in the inode.  This happens
1518  * at IO completion time based on sums calculated at bio submission time.
1519  */
1520 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1521                              struct inode *inode, u64 file_offset,
1522                              struct list_head *list)
1523 {
1524         struct btrfs_ordered_sum *sum;
1525
1526         list_for_each_entry(sum, list, list) {
1527                 btrfs_csum_file_blocks(trans,
1528                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1529         }
1530         return 0;
1531 }
1532
1533 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1534                               struct extent_state **cached_state)
1535 {
1536         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1537                 WARN_ON(1);
1538         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1539                                    cached_state, GFP_NOFS);
1540 }
1541
1542 /* see btrfs_writepage_start_hook for details on why this is required */
1543 struct btrfs_writepage_fixup {
1544         struct page *page;
1545         struct btrfs_work work;
1546 };
1547
1548 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1549 {
1550         struct btrfs_writepage_fixup *fixup;
1551         struct btrfs_ordered_extent *ordered;
1552         struct extent_state *cached_state = NULL;
1553         struct page *page;
1554         struct inode *inode;
1555         u64 page_start;
1556         u64 page_end;
1557
1558         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1559         page = fixup->page;
1560 again:
1561         lock_page(page);
1562         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1563                 ClearPageChecked(page);
1564                 goto out_page;
1565         }
1566
1567         inode = page->mapping->host;
1568         page_start = page_offset(page);
1569         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1570
1571         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1572                          &cached_state, GFP_NOFS);
1573
1574         /* already ordered? We're done */
1575         if (PagePrivate2(page))
1576                 goto out;
1577
1578         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1579         if (ordered) {
1580                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1581                                      page_end, &cached_state, GFP_NOFS);
1582                 unlock_page(page);
1583                 btrfs_start_ordered_extent(inode, ordered, 1);
1584                 goto again;
1585         }
1586
1587         BUG();
1588         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1589         ClearPageChecked(page);
1590 out:
1591         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1592                              &cached_state, GFP_NOFS);
1593 out_page:
1594         unlock_page(page);
1595         page_cache_release(page);
1596         kfree(fixup);
1597 }
1598
1599 /*
1600  * There are a few paths in the higher layers of the kernel that directly
1601  * set the page dirty bit without asking the filesystem if it is a
1602  * good idea.  This causes problems because we want to make sure COW
1603  * properly happens and the data=ordered rules are followed.
1604  *
1605  * In our case any range that doesn't have the ORDERED bit set
1606  * hasn't been properly setup for IO.  We kick off an async process
1607  * to fix it up.  The async helper will wait for ordered extents, set
1608  * the delalloc bit and make it safe to write the page.
1609  */
1610 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1611 {
1612         struct inode *inode = page->mapping->host;
1613         struct btrfs_writepage_fixup *fixup;
1614         struct btrfs_root *root = BTRFS_I(inode)->root;
1615
1616         /* this page is properly in the ordered list */
1617         if (TestClearPagePrivate2(page))
1618                 return 0;
1619
1620         if (PageChecked(page))
1621                 return -EAGAIN;
1622
1623         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1624         if (!fixup)
1625                 return -EAGAIN;
1626
1627         SetPageChecked(page);
1628         page_cache_get(page);
1629         fixup->work.func = btrfs_writepage_fixup_worker;
1630         fixup->page = page;
1631         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1632         return -EAGAIN;
1633 }
1634
1635 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1636                                        struct inode *inode, u64 file_pos,
1637                                        u64 disk_bytenr, u64 disk_num_bytes,
1638                                        u64 num_bytes, u64 ram_bytes,
1639                                        u8 compression, u8 encryption,
1640                                        u16 other_encoding, int extent_type)
1641 {
1642         struct btrfs_root *root = BTRFS_I(inode)->root;
1643         struct btrfs_file_extent_item *fi;
1644         struct btrfs_path *path;
1645         struct extent_buffer *leaf;
1646         struct btrfs_key ins;
1647         u64 hint;
1648         int ret;
1649
1650         path = btrfs_alloc_path();
1651         if (!path)
1652                 return -ENOMEM;
1653
1654         path->leave_spinning = 1;
1655
1656         /*
1657          * we may be replacing one extent in the tree with another.
1658          * The new extent is pinned in the extent map, and we don't want
1659          * to drop it from the cache until it is completely in the btree.
1660          *
1661          * So, tell btrfs_drop_extents to leave this extent in the cache.
1662          * the caller is expected to unpin it and allow it to be merged
1663          * with the others.
1664          */
1665         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1666                                  &hint, 0);
1667         BUG_ON(ret);
1668
1669         ins.objectid = btrfs_ino(inode);
1670         ins.offset = file_pos;
1671         ins.type = BTRFS_EXTENT_DATA_KEY;
1672         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1673         BUG_ON(ret);
1674         leaf = path->nodes[0];
1675         fi = btrfs_item_ptr(leaf, path->slots[0],
1676                             struct btrfs_file_extent_item);
1677         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1678         btrfs_set_file_extent_type(leaf, fi, extent_type);
1679         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1680         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1681         btrfs_set_file_extent_offset(leaf, fi, 0);
1682         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1683         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1684         btrfs_set_file_extent_compression(leaf, fi, compression);
1685         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1686         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1687
1688         btrfs_unlock_up_safe(path, 1);
1689         btrfs_set_lock_blocking(leaf);
1690
1691         btrfs_mark_buffer_dirty(leaf);
1692
1693         inode_add_bytes(inode, num_bytes);
1694
1695         ins.objectid = disk_bytenr;
1696         ins.offset = disk_num_bytes;
1697         ins.type = BTRFS_EXTENT_ITEM_KEY;
1698         ret = btrfs_alloc_reserved_file_extent(trans, root,
1699                                         root->root_key.objectid,
1700                                         btrfs_ino(inode), file_pos, &ins);
1701         BUG_ON(ret);
1702         btrfs_free_path(path);
1703
1704         return 0;
1705 }
1706
1707 /*
1708  * helper function for btrfs_finish_ordered_io, this
1709  * just reads in some of the csum leaves to prime them into ram
1710  * before we start the transaction.  It limits the amount of btree
1711  * reads required while inside the transaction.
1712  */
1713 /* as ordered data IO finishes, this gets called so we can finish
1714  * an ordered extent if the range of bytes in the file it covers are
1715  * fully written.
1716  */
1717 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1718 {
1719         struct btrfs_root *root = BTRFS_I(inode)->root;
1720         struct btrfs_trans_handle *trans = NULL;
1721         struct btrfs_ordered_extent *ordered_extent = NULL;
1722         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1723         struct extent_state *cached_state = NULL;
1724         int compress_type = 0;
1725         int ret;
1726         bool nolock;
1727
1728         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1729                                              end - start + 1);
1730         if (!ret)
1731                 return 0;
1732         BUG_ON(!ordered_extent);
1733
1734         nolock = btrfs_is_free_space_inode(root, inode);
1735
1736         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1737                 BUG_ON(!list_empty(&ordered_extent->list));
1738                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1739                 if (!ret) {
1740                         if (nolock)
1741                                 trans = btrfs_join_transaction_nolock(root);
1742                         else
1743                                 trans = btrfs_join_transaction(root);
1744                         BUG_ON(IS_ERR(trans));
1745                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1746                         ret = btrfs_update_inode_fallback(trans, root, inode);
1747                         BUG_ON(ret);
1748                 }
1749                 goto out;
1750         }
1751
1752         lock_extent_bits(io_tree, ordered_extent->file_offset,
1753                          ordered_extent->file_offset + ordered_extent->len - 1,
1754                          0, &cached_state, GFP_NOFS);
1755
1756         if (nolock)
1757                 trans = btrfs_join_transaction_nolock(root);
1758         else
1759                 trans = btrfs_join_transaction(root);
1760         BUG_ON(IS_ERR(trans));
1761         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1762
1763         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1764                 compress_type = ordered_extent->compress_type;
1765         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1766                 BUG_ON(compress_type);
1767                 ret = btrfs_mark_extent_written(trans, inode,
1768                                                 ordered_extent->file_offset,
1769                                                 ordered_extent->file_offset +
1770                                                 ordered_extent->len);
1771                 BUG_ON(ret);
1772         } else {
1773                 BUG_ON(root == root->fs_info->tree_root);
1774                 ret = insert_reserved_file_extent(trans, inode,
1775                                                 ordered_extent->file_offset,
1776                                                 ordered_extent->start,
1777                                                 ordered_extent->disk_len,
1778                                                 ordered_extent->len,
1779                                                 ordered_extent->len,
1780                                                 compress_type, 0, 0,
1781                                                 BTRFS_FILE_EXTENT_REG);
1782                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1783                                    ordered_extent->file_offset,
1784                                    ordered_extent->len);
1785                 BUG_ON(ret);
1786         }
1787         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1788                              ordered_extent->file_offset +
1789                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1790
1791         add_pending_csums(trans, inode, ordered_extent->file_offset,
1792                           &ordered_extent->list);
1793
1794         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1795         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1796                 ret = btrfs_update_inode_fallback(trans, root, inode);
1797                 BUG_ON(ret);
1798         }
1799         ret = 0;
1800 out:
1801         if (root != root->fs_info->tree_root)
1802                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1803         if (trans) {
1804                 if (nolock)
1805                         btrfs_end_transaction_nolock(trans, root);
1806                 else
1807                         btrfs_end_transaction(trans, root);
1808         }
1809
1810         /* once for us */
1811         btrfs_put_ordered_extent(ordered_extent);
1812         /* once for the tree */
1813         btrfs_put_ordered_extent(ordered_extent);
1814
1815         return 0;
1816 }
1817
1818 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1819                                 struct extent_state *state, int uptodate)
1820 {
1821         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1822
1823         ClearPagePrivate2(page);
1824         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1825 }
1826
1827 /*
1828  * when reads are done, we need to check csums to verify the data is correct
1829  * if there's a match, we allow the bio to finish.  If not, the code in
1830  * extent_io.c will try to find good copies for us.
1831  */
1832 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1833                                struct extent_state *state)
1834 {
1835         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1836         struct inode *inode = page->mapping->host;
1837         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1838         char *kaddr;
1839         u64 private = ~(u32)0;
1840         int ret;
1841         struct btrfs_root *root = BTRFS_I(inode)->root;
1842         u32 csum = ~(u32)0;
1843
1844         if (PageChecked(page)) {
1845                 ClearPageChecked(page);
1846                 goto good;
1847         }
1848
1849         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1850                 goto good;
1851
1852         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1853             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1854                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1855                                   GFP_NOFS);
1856                 return 0;
1857         }
1858
1859         if (state && state->start == start) {
1860                 private = state->private;
1861                 ret = 0;
1862         } else {
1863                 ret = get_state_private(io_tree, start, &private);
1864         }
1865         kaddr = kmap_atomic(page, KM_USER0);
1866         if (ret)
1867                 goto zeroit;
1868
1869         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1870         btrfs_csum_final(csum, (char *)&csum);
1871         if (csum != private)
1872                 goto zeroit;
1873
1874         kunmap_atomic(kaddr, KM_USER0);
1875 good:
1876         return 0;
1877
1878 zeroit:
1879         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
1880                        "private %llu\n",
1881                        (unsigned long long)btrfs_ino(page->mapping->host),
1882                        (unsigned long long)start, csum,
1883                        (unsigned long long)private);
1884         memset(kaddr + offset, 1, end - start + 1);
1885         flush_dcache_page(page);
1886         kunmap_atomic(kaddr, KM_USER0);
1887         if (private == 0)
1888                 return 0;
1889         return -EIO;
1890 }
1891
1892 struct delayed_iput {
1893         struct list_head list;
1894         struct inode *inode;
1895 };
1896
1897 void btrfs_add_delayed_iput(struct inode *inode)
1898 {
1899         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1900         struct delayed_iput *delayed;
1901
1902         if (atomic_add_unless(&inode->i_count, -1, 1))
1903                 return;
1904
1905         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
1906         delayed->inode = inode;
1907
1908         spin_lock(&fs_info->delayed_iput_lock);
1909         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
1910         spin_unlock(&fs_info->delayed_iput_lock);
1911 }
1912
1913 void btrfs_run_delayed_iputs(struct btrfs_root *root)
1914 {
1915         LIST_HEAD(list);
1916         struct btrfs_fs_info *fs_info = root->fs_info;
1917         struct delayed_iput *delayed;
1918         int empty;
1919
1920         spin_lock(&fs_info->delayed_iput_lock);
1921         empty = list_empty(&fs_info->delayed_iputs);
1922         spin_unlock(&fs_info->delayed_iput_lock);
1923         if (empty)
1924                 return;
1925
1926         down_read(&root->fs_info->cleanup_work_sem);
1927         spin_lock(&fs_info->delayed_iput_lock);
1928         list_splice_init(&fs_info->delayed_iputs, &list);
1929         spin_unlock(&fs_info->delayed_iput_lock);
1930
1931         while (!list_empty(&list)) {
1932                 delayed = list_entry(list.next, struct delayed_iput, list);
1933                 list_del(&delayed->list);
1934                 iput(delayed->inode);
1935                 kfree(delayed);
1936         }
1937         up_read(&root->fs_info->cleanup_work_sem);
1938 }
1939
1940 enum btrfs_orphan_cleanup_state {
1941         ORPHAN_CLEANUP_STARTED  = 1,
1942         ORPHAN_CLEANUP_DONE     = 2,
1943 };
1944
1945 /*
1946  * This is called in transaction commmit time. If there are no orphan
1947  * files in the subvolume, it removes orphan item and frees block_rsv
1948  * structure.
1949  */
1950 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
1951                               struct btrfs_root *root)
1952 {
1953         int ret;
1954
1955         if (!list_empty(&root->orphan_list) ||
1956             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
1957                 return;
1958
1959         if (root->orphan_item_inserted &&
1960             btrfs_root_refs(&root->root_item) > 0) {
1961                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
1962                                             root->root_key.objectid);
1963                 BUG_ON(ret);
1964                 root->orphan_item_inserted = 0;
1965         }
1966
1967         if (root->orphan_block_rsv) {
1968                 WARN_ON(root->orphan_block_rsv->size > 0);
1969                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
1970                 root->orphan_block_rsv = NULL;
1971         }
1972 }
1973
1974 /*
1975  * This creates an orphan entry for the given inode in case something goes
1976  * wrong in the middle of an unlink/truncate.
1977  *
1978  * NOTE: caller of this function should reserve 5 units of metadata for
1979  *       this function.
1980  */
1981 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1982 {
1983         struct btrfs_root *root = BTRFS_I(inode)->root;
1984         struct btrfs_block_rsv *block_rsv = NULL;
1985         int reserve = 0;
1986         int insert = 0;
1987         int ret;
1988
1989         if (!root->orphan_block_rsv) {
1990                 block_rsv = btrfs_alloc_block_rsv(root);
1991                 if (!block_rsv)
1992                         return -ENOMEM;
1993         }
1994
1995         spin_lock(&root->orphan_lock);
1996         if (!root->orphan_block_rsv) {
1997                 root->orphan_block_rsv = block_rsv;
1998         } else if (block_rsv) {
1999                 btrfs_free_block_rsv(root, block_rsv);
2000                 block_rsv = NULL;
2001         }
2002
2003         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2004                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2005 #if 0
2006                 /*
2007                  * For proper ENOSPC handling, we should do orphan
2008                  * cleanup when mounting. But this introduces backward
2009                  * compatibility issue.
2010                  */
2011                 if (!xchg(&root->orphan_item_inserted, 1))
2012                         insert = 2;
2013                 else
2014                         insert = 1;
2015 #endif
2016                 insert = 1;
2017         }
2018
2019         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2020                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2021                 reserve = 1;
2022         }
2023         spin_unlock(&root->orphan_lock);
2024
2025         /* grab metadata reservation from transaction handle */
2026         if (reserve) {
2027                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2028                 BUG_ON(ret);
2029         }
2030
2031         /* insert an orphan item to track this unlinked/truncated file */
2032         if (insert >= 1) {
2033                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2034                 BUG_ON(ret);
2035         }
2036
2037         /* insert an orphan item to track subvolume contains orphan files */
2038         if (insert >= 2) {
2039                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2040                                                root->root_key.objectid);
2041                 BUG_ON(ret);
2042         }
2043         return 0;
2044 }
2045
2046 /*
2047  * We have done the truncate/delete so we can go ahead and remove the orphan
2048  * item for this particular inode.
2049  */
2050 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2051 {
2052         struct btrfs_root *root = BTRFS_I(inode)->root;
2053         int delete_item = 0;
2054         int release_rsv = 0;
2055         int ret = 0;
2056
2057         spin_lock(&root->orphan_lock);
2058         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2059                 list_del_init(&BTRFS_I(inode)->i_orphan);
2060                 delete_item = 1;
2061         }
2062
2063         if (BTRFS_I(inode)->orphan_meta_reserved) {
2064                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2065                 release_rsv = 1;
2066         }
2067         spin_unlock(&root->orphan_lock);
2068
2069         if (trans && delete_item) {
2070                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2071                 BUG_ON(ret);
2072         }
2073
2074         if (release_rsv)
2075                 btrfs_orphan_release_metadata(inode);
2076
2077         return 0;
2078 }
2079
2080 /*
2081  * this cleans up any orphans that may be left on the list from the last use
2082  * of this root.
2083  */
2084 int btrfs_orphan_cleanup(struct btrfs_root *root)
2085 {
2086         struct btrfs_path *path;
2087         struct extent_buffer *leaf;
2088         struct btrfs_key key, found_key;
2089         struct btrfs_trans_handle *trans;
2090         struct inode *inode;
2091         u64 last_objectid = 0;
2092         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2093
2094         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2095                 return 0;
2096
2097         path = btrfs_alloc_path();
2098         if (!path) {
2099                 ret = -ENOMEM;
2100                 goto out;
2101         }
2102         path->reada = -1;
2103
2104         key.objectid = BTRFS_ORPHAN_OBJECTID;
2105         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2106         key.offset = (u64)-1;
2107
2108         while (1) {
2109                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2110                 if (ret < 0)
2111                         goto out;
2112
2113                 /*
2114                  * if ret == 0 means we found what we were searching for, which
2115                  * is weird, but possible, so only screw with path if we didn't
2116                  * find the key and see if we have stuff that matches
2117                  */
2118                 if (ret > 0) {
2119                         ret = 0;
2120                         if (path->slots[0] == 0)
2121                                 break;
2122                         path->slots[0]--;
2123                 }
2124
2125                 /* pull out the item */
2126                 leaf = path->nodes[0];
2127                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2128
2129                 /* make sure the item matches what we want */
2130                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2131                         break;
2132                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2133                         break;
2134
2135                 /* release the path since we're done with it */
2136                 btrfs_release_path(path);
2137
2138                 /*
2139                  * this is where we are basically btrfs_lookup, without the
2140                  * crossing root thing.  we store the inode number in the
2141                  * offset of the orphan item.
2142                  */
2143
2144                 if (found_key.offset == last_objectid) {
2145                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2146                                "stopping orphan cleanup\n");
2147                         ret = -EINVAL;
2148                         goto out;
2149                 }
2150
2151                 last_objectid = found_key.offset;
2152
2153                 found_key.objectid = found_key.offset;
2154                 found_key.type = BTRFS_INODE_ITEM_KEY;
2155                 found_key.offset = 0;
2156                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2157                 ret = PTR_RET(inode);
2158                 if (ret && ret != -ESTALE)
2159                         goto out;
2160
2161                 /*
2162                  * Inode is already gone but the orphan item is still there,
2163                  * kill the orphan item.
2164                  */
2165                 if (ret == -ESTALE) {
2166                         trans = btrfs_start_transaction(root, 1);
2167                         if (IS_ERR(trans)) {
2168                                 ret = PTR_ERR(trans);
2169                                 goto out;
2170                         }
2171                         ret = btrfs_del_orphan_item(trans, root,
2172                                                     found_key.objectid);
2173                         BUG_ON(ret);
2174                         btrfs_end_transaction(trans, root);
2175                         continue;
2176                 }
2177
2178                 /*
2179                  * add this inode to the orphan list so btrfs_orphan_del does
2180                  * the proper thing when we hit it
2181                  */
2182                 spin_lock(&root->orphan_lock);
2183                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2184                 spin_unlock(&root->orphan_lock);
2185
2186                 /* if we have links, this was a truncate, lets do that */
2187                 if (inode->i_nlink) {
2188                         if (!S_ISREG(inode->i_mode)) {
2189                                 WARN_ON(1);
2190                                 iput(inode);
2191                                 continue;
2192                         }
2193                         nr_truncate++;
2194                         ret = btrfs_truncate(inode);
2195                 } else {
2196                         nr_unlink++;
2197                 }
2198
2199                 /* this will do delete_inode and everything for us */
2200                 iput(inode);
2201                 if (ret)
2202                         goto out;
2203         }
2204         /* release the path since we're done with it */
2205         btrfs_release_path(path);
2206
2207         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2208
2209         if (root->orphan_block_rsv)
2210                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2211                                         (u64)-1);
2212
2213         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2214                 trans = btrfs_join_transaction(root);
2215                 if (!IS_ERR(trans))
2216                         btrfs_end_transaction(trans, root);
2217         }
2218
2219         if (nr_unlink)
2220                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2221         if (nr_truncate)
2222                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2223
2224 out:
2225         if (ret)
2226                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2227         btrfs_free_path(path);
2228         return ret;
2229 }
2230
2231 /*
2232  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2233  * don't find any xattrs, we know there can't be any acls.
2234  *
2235  * slot is the slot the inode is in, objectid is the objectid of the inode
2236  */
2237 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2238                                           int slot, u64 objectid)
2239 {
2240         u32 nritems = btrfs_header_nritems(leaf);
2241         struct btrfs_key found_key;
2242         int scanned = 0;
2243
2244         slot++;
2245         while (slot < nritems) {
2246                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2247
2248                 /* we found a different objectid, there must not be acls */
2249                 if (found_key.objectid != objectid)
2250                         return 0;
2251
2252                 /* we found an xattr, assume we've got an acl */
2253                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2254                         return 1;
2255
2256                 /*
2257                  * we found a key greater than an xattr key, there can't
2258                  * be any acls later on
2259                  */
2260                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2261                         return 0;
2262
2263                 slot++;
2264                 scanned++;
2265
2266                 /*
2267                  * it goes inode, inode backrefs, xattrs, extents,
2268                  * so if there are a ton of hard links to an inode there can
2269                  * be a lot of backrefs.  Don't waste time searching too hard,
2270                  * this is just an optimization
2271                  */
2272                 if (scanned >= 8)
2273                         break;
2274         }
2275         /* we hit the end of the leaf before we found an xattr or
2276          * something larger than an xattr.  We have to assume the inode
2277          * has acls
2278          */
2279         return 1;
2280 }
2281
2282 /*
2283  * read an inode from the btree into the in-memory inode
2284  */
2285 static void btrfs_read_locked_inode(struct inode *inode)
2286 {
2287         struct btrfs_path *path;
2288         struct extent_buffer *leaf;
2289         struct btrfs_inode_item *inode_item;
2290         struct btrfs_timespec *tspec;
2291         struct btrfs_root *root = BTRFS_I(inode)->root;
2292         struct btrfs_key location;
2293         int maybe_acls;
2294         u32 rdev;
2295         int ret;
2296         bool filled = false;
2297
2298         ret = btrfs_fill_inode(inode, &rdev);
2299         if (!ret)
2300                 filled = true;
2301
2302         path = btrfs_alloc_path();
2303         if (!path)
2304                 goto make_bad;
2305
2306         path->leave_spinning = 1;
2307         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2308
2309         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2310         if (ret)
2311                 goto make_bad;
2312
2313         leaf = path->nodes[0];
2314
2315         if (filled)
2316                 goto cache_acl;
2317
2318         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2319                                     struct btrfs_inode_item);
2320         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2321         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2322         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2323         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2324         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2325
2326         tspec = btrfs_inode_atime(inode_item);
2327         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2328         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2329
2330         tspec = btrfs_inode_mtime(inode_item);
2331         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2332         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2333
2334         tspec = btrfs_inode_ctime(inode_item);
2335         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2336         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2337
2338         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2339         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2340         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2341         inode->i_generation = BTRFS_I(inode)->generation;
2342         inode->i_rdev = 0;
2343         rdev = btrfs_inode_rdev(leaf, inode_item);
2344
2345         BTRFS_I(inode)->index_cnt = (u64)-1;
2346         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2347 cache_acl:
2348         /*
2349          * try to precache a NULL acl entry for files that don't have
2350          * any xattrs or acls
2351          */
2352         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2353                                            btrfs_ino(inode));
2354         if (!maybe_acls)
2355                 cache_no_acl(inode);
2356
2357         btrfs_free_path(path);
2358
2359         switch (inode->i_mode & S_IFMT) {
2360         case S_IFREG:
2361                 inode->i_mapping->a_ops = &btrfs_aops;
2362                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2363                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2364                 inode->i_fop = &btrfs_file_operations;
2365                 inode->i_op = &btrfs_file_inode_operations;
2366                 break;
2367         case S_IFDIR:
2368                 inode->i_fop = &btrfs_dir_file_operations;
2369                 if (root == root->fs_info->tree_root)
2370                         inode->i_op = &btrfs_dir_ro_inode_operations;
2371                 else
2372                         inode->i_op = &btrfs_dir_inode_operations;
2373                 break;
2374         case S_IFLNK:
2375                 inode->i_op = &btrfs_symlink_inode_operations;
2376                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2377                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2378                 break;
2379         default:
2380                 inode->i_op = &btrfs_special_inode_operations;
2381                 init_special_inode(inode, inode->i_mode, rdev);
2382                 break;
2383         }
2384
2385         btrfs_update_iflags(inode);
2386         return;
2387
2388 make_bad:
2389         btrfs_free_path(path);
2390         make_bad_inode(inode);
2391 }
2392
2393 /*
2394  * given a leaf and an inode, copy the inode fields into the leaf
2395  */
2396 static void fill_inode_item(struct btrfs_trans_handle *trans,
2397                             struct extent_buffer *leaf,
2398                             struct btrfs_inode_item *item,
2399                             struct inode *inode)
2400 {
2401         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2402         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2403         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2404         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2405         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2406
2407         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2408                                inode->i_atime.tv_sec);
2409         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2410                                 inode->i_atime.tv_nsec);
2411
2412         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2413                                inode->i_mtime.tv_sec);
2414         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2415                                 inode->i_mtime.tv_nsec);
2416
2417         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2418                                inode->i_ctime.tv_sec);
2419         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2420                                 inode->i_ctime.tv_nsec);
2421
2422         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2423         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2424         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2425         btrfs_set_inode_transid(leaf, item, trans->transid);
2426         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2427         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2428         btrfs_set_inode_block_group(leaf, item, 0);
2429 }
2430
2431 /*
2432  * copy everything in the in-memory inode into the btree.
2433  */
2434 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2435                                 struct btrfs_root *root, struct inode *inode)
2436 {
2437         struct btrfs_inode_item *inode_item;
2438         struct btrfs_path *path;
2439         struct extent_buffer *leaf;
2440         int ret;
2441
2442         path = btrfs_alloc_path();
2443         if (!path)
2444                 return -ENOMEM;
2445
2446         path->leave_spinning = 1;
2447         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2448                                  1);
2449         if (ret) {
2450                 if (ret > 0)
2451                         ret = -ENOENT;
2452                 goto failed;
2453         }
2454
2455         btrfs_unlock_up_safe(path, 1);
2456         leaf = path->nodes[0];
2457         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2458                                     struct btrfs_inode_item);
2459
2460         fill_inode_item(trans, leaf, inode_item, inode);
2461         btrfs_mark_buffer_dirty(leaf);
2462         btrfs_set_inode_last_trans(trans, inode);
2463         ret = 0;
2464 failed:
2465         btrfs_free_path(path);
2466         return ret;
2467 }
2468
2469 /*
2470  * copy everything in the in-memory inode into the btree.
2471  */
2472 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2473                                 struct btrfs_root *root, struct inode *inode)
2474 {
2475         int ret;
2476
2477         /*
2478          * If the inode is a free space inode, we can deadlock during commit
2479          * if we put it into the delayed code.
2480          *
2481          * The data relocation inode should also be directly updated
2482          * without delay
2483          */
2484         if (!btrfs_is_free_space_inode(root, inode)
2485             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2486                 ret = btrfs_delayed_update_inode(trans, root, inode);
2487                 if (!ret)
2488                         btrfs_set_inode_last_trans(trans, inode);
2489                 return ret;
2490         }
2491
2492         return btrfs_update_inode_item(trans, root, inode);
2493 }
2494
2495 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2496                                 struct btrfs_root *root, struct inode *inode)
2497 {
2498         int ret;
2499
2500         ret = btrfs_update_inode(trans, root, inode);
2501         if (ret == -ENOSPC)
2502                 return btrfs_update_inode_item(trans, root, inode);
2503         return ret;
2504 }
2505
2506 /*
2507  * unlink helper that gets used here in inode.c and in the tree logging
2508  * recovery code.  It remove a link in a directory with a given name, and
2509  * also drops the back refs in the inode to the directory
2510  */
2511 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2512                                 struct btrfs_root *root,
2513                                 struct inode *dir, struct inode *inode,
2514                                 const char *name, int name_len)
2515 {
2516         struct btrfs_path *path;
2517         int ret = 0;
2518         struct extent_buffer *leaf;
2519         struct btrfs_dir_item *di;
2520         struct btrfs_key key;
2521         u64 index;
2522         u64 ino = btrfs_ino(inode);
2523         u64 dir_ino = btrfs_ino(dir);
2524
2525         path = btrfs_alloc_path();
2526         if (!path) {
2527                 ret = -ENOMEM;
2528                 goto out;
2529         }
2530
2531         path->leave_spinning = 1;
2532         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2533                                     name, name_len, -1);
2534         if (IS_ERR(di)) {
2535                 ret = PTR_ERR(di);
2536                 goto err;
2537         }
2538         if (!di) {
2539                 ret = -ENOENT;
2540                 goto err;
2541         }
2542         leaf = path->nodes[0];
2543         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2544         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2545         if (ret)
2546                 goto err;
2547         btrfs_release_path(path);
2548
2549         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2550                                   dir_ino, &index);
2551         if (ret) {
2552                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2553                        "inode %llu parent %llu\n", name_len, name,
2554                        (unsigned long long)ino, (unsigned long long)dir_ino);
2555                 goto err;
2556         }
2557
2558         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2559         if (ret)
2560                 goto err;
2561
2562         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2563                                          inode, dir_ino);
2564         BUG_ON(ret != 0 && ret != -ENOENT);
2565
2566         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2567                                            dir, index);
2568         if (ret == -ENOENT)
2569                 ret = 0;
2570 err:
2571         btrfs_free_path(path);
2572         if (ret)
2573                 goto out;
2574
2575         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2576         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2577         btrfs_update_inode(trans, root, dir);
2578 out:
2579         return ret;
2580 }
2581
2582 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2583                        struct btrfs_root *root,
2584                        struct inode *dir, struct inode *inode,
2585                        const char *name, int name_len)
2586 {
2587         int ret;
2588         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2589         if (!ret) {
2590                 btrfs_drop_nlink(inode);
2591                 ret = btrfs_update_inode(trans, root, inode);
2592         }
2593         return ret;
2594 }
2595                 
2596
2597 /* helper to check if there is any shared block in the path */
2598 static int check_path_shared(struct btrfs_root *root,
2599                              struct btrfs_path *path)
2600 {
2601         struct extent_buffer *eb;
2602         int level;
2603         u64 refs = 1;
2604
2605         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2606                 int ret;
2607
2608                 if (!path->nodes[level])
2609                         break;
2610                 eb = path->nodes[level];
2611                 if (!btrfs_block_can_be_shared(root, eb))
2612                         continue;
2613                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2614                                                &refs, NULL);
2615                 if (refs > 1)
2616                         return 1;
2617         }
2618         return 0;
2619 }
2620
2621 /*
2622  * helper to start transaction for unlink and rmdir.
2623  *
2624  * unlink and rmdir are special in btrfs, they do not always free space.
2625  * so in enospc case, we should make sure they will free space before
2626  * allowing them to use the global metadata reservation.
2627  */
2628 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2629                                                        struct dentry *dentry)
2630 {
2631         struct btrfs_trans_handle *trans;
2632         struct btrfs_root *root = BTRFS_I(dir)->root;
2633         struct btrfs_path *path;
2634         struct btrfs_inode_ref *ref;
2635         struct btrfs_dir_item *di;
2636         struct inode *inode = dentry->d_inode;
2637         u64 index;
2638         int check_link = 1;
2639         int err = -ENOSPC;
2640         int ret;
2641         u64 ino = btrfs_ino(inode);
2642         u64 dir_ino = btrfs_ino(dir);
2643
2644         /*
2645          * 1 for the possible orphan item
2646          * 1 for the dir item
2647          * 1 for the dir index
2648          * 1 for the inode ref
2649          * 1 for the inode ref in the tree log
2650          * 2 for the dir entries in the log
2651          * 1 for the inode
2652          */
2653         trans = btrfs_start_transaction(root, 8);
2654         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2655                 return trans;
2656
2657         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2658                 return ERR_PTR(-ENOSPC);
2659
2660         /* check if there is someone else holds reference */
2661         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2662                 return ERR_PTR(-ENOSPC);
2663
2664         if (atomic_read(&inode->i_count) > 2)
2665                 return ERR_PTR(-ENOSPC);
2666
2667         if (xchg(&root->fs_info->enospc_unlink, 1))
2668                 return ERR_PTR(-ENOSPC);
2669
2670         path = btrfs_alloc_path();
2671         if (!path) {
2672                 root->fs_info->enospc_unlink = 0;
2673                 return ERR_PTR(-ENOMEM);
2674         }
2675
2676         /* 1 for the orphan item */
2677         trans = btrfs_start_transaction(root, 1);
2678         if (IS_ERR(trans)) {
2679                 btrfs_free_path(path);
2680                 root->fs_info->enospc_unlink = 0;
2681                 return trans;
2682         }
2683
2684         path->skip_locking = 1;
2685         path->search_commit_root = 1;
2686
2687         ret = btrfs_lookup_inode(trans, root, path,
2688                                 &BTRFS_I(dir)->location, 0);
2689         if (ret < 0) {
2690                 err = ret;
2691                 goto out;
2692         }
2693         if (ret == 0) {
2694                 if (check_path_shared(root, path))
2695                         goto out;
2696         } else {
2697                 check_link = 0;
2698         }
2699         btrfs_release_path(path);
2700
2701         ret = btrfs_lookup_inode(trans, root, path,
2702                                 &BTRFS_I(inode)->location, 0);
2703         if (ret < 0) {
2704                 err = ret;
2705                 goto out;
2706         }
2707         if (ret == 0) {
2708                 if (check_path_shared(root, path))
2709                         goto out;
2710         } else {
2711                 check_link = 0;
2712         }
2713         btrfs_release_path(path);
2714
2715         if (ret == 0 && S_ISREG(inode->i_mode)) {
2716                 ret = btrfs_lookup_file_extent(trans, root, path,
2717                                                ino, (u64)-1, 0);
2718                 if (ret < 0) {
2719                         err = ret;
2720                         goto out;
2721                 }
2722                 BUG_ON(ret == 0);
2723                 if (check_path_shared(root, path))
2724                         goto out;
2725                 btrfs_release_path(path);
2726         }
2727
2728         if (!check_link) {
2729                 err = 0;
2730                 goto out;
2731         }
2732
2733         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2734                                 dentry->d_name.name, dentry->d_name.len, 0);
2735         if (IS_ERR(di)) {
2736                 err = PTR_ERR(di);
2737                 goto out;
2738         }
2739         if (di) {
2740                 if (check_path_shared(root, path))
2741                         goto out;
2742         } else {
2743                 err = 0;
2744                 goto out;
2745         }
2746         btrfs_release_path(path);
2747
2748         ref = btrfs_lookup_inode_ref(trans, root, path,
2749                                 dentry->d_name.name, dentry->d_name.len,
2750                                 ino, dir_ino, 0);
2751         if (IS_ERR(ref)) {
2752                 err = PTR_ERR(ref);
2753                 goto out;
2754         }
2755         BUG_ON(!ref);
2756         if (check_path_shared(root, path))
2757                 goto out;
2758         index = btrfs_inode_ref_index(path->nodes[0], ref);
2759         btrfs_release_path(path);
2760
2761         /*
2762          * This is a commit root search, if we can lookup inode item and other
2763          * relative items in the commit root, it means the transaction of
2764          * dir/file creation has been committed, and the dir index item that we
2765          * delay to insert has also been inserted into the commit root. So
2766          * we needn't worry about the delayed insertion of the dir index item
2767          * here.
2768          */
2769         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2770                                 dentry->d_name.name, dentry->d_name.len, 0);
2771         if (IS_ERR(di)) {
2772                 err = PTR_ERR(di);
2773                 goto out;
2774         }
2775         BUG_ON(ret == -ENOENT);
2776         if (check_path_shared(root, path))
2777                 goto out;
2778
2779         err = 0;
2780 out:
2781         btrfs_free_path(path);
2782         /* Migrate the orphan reservation over */
2783         if (!err)
2784                 err = btrfs_block_rsv_migrate(trans->block_rsv,
2785                                 &root->fs_info->global_block_rsv,
2786                                 trans->bytes_reserved);
2787
2788         if (err) {
2789                 btrfs_end_transaction(trans, root);
2790                 root->fs_info->enospc_unlink = 0;
2791                 return ERR_PTR(err);
2792         }
2793
2794         trans->block_rsv = &root->fs_info->global_block_rsv;
2795         return trans;
2796 }
2797
2798 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2799                                struct btrfs_root *root)
2800 {
2801         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2802                 btrfs_block_rsv_release(root, trans->block_rsv,
2803                                         trans->bytes_reserved);
2804                 trans->block_rsv = &root->fs_info->trans_block_rsv;
2805                 BUG_ON(!root->fs_info->enospc_unlink);
2806                 root->fs_info->enospc_unlink = 0;
2807         }
2808         btrfs_end_transaction_throttle(trans, root);
2809 }
2810
2811 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2812 {
2813         struct btrfs_root *root = BTRFS_I(dir)->root;
2814         struct btrfs_trans_handle *trans;
2815         struct inode *inode = dentry->d_inode;
2816         int ret;
2817         unsigned long nr = 0;
2818
2819         trans = __unlink_start_trans(dir, dentry);
2820         if (IS_ERR(trans))
2821                 return PTR_ERR(trans);
2822
2823         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2824
2825         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2826                                  dentry->d_name.name, dentry->d_name.len);
2827         if (ret)
2828                 goto out;
2829
2830         if (inode->i_nlink == 0) {
2831                 ret = btrfs_orphan_add(trans, inode);
2832                 if (ret)
2833                         goto out;
2834         }
2835
2836 out:
2837         nr = trans->blocks_used;
2838         __unlink_end_trans(trans, root);
2839         btrfs_btree_balance_dirty(root, nr);
2840         return ret;
2841 }
2842
2843 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2844                         struct btrfs_root *root,
2845                         struct inode *dir, u64 objectid,
2846                         const char *name, int name_len)
2847 {
2848         struct btrfs_path *path;
2849         struct extent_buffer *leaf;
2850         struct btrfs_dir_item *di;
2851         struct btrfs_key key;
2852         u64 index;
2853         int ret;
2854         u64 dir_ino = btrfs_ino(dir);
2855
2856         path = btrfs_alloc_path();
2857         if (!path)
2858                 return -ENOMEM;
2859
2860         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2861                                    name, name_len, -1);
2862         BUG_ON(IS_ERR_OR_NULL(di));
2863
2864         leaf = path->nodes[0];
2865         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2866         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2867         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2868         BUG_ON(ret);
2869         btrfs_release_path(path);
2870
2871         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2872                                  objectid, root->root_key.objectid,
2873                                  dir_ino, &index, name, name_len);
2874         if (ret < 0) {
2875                 BUG_ON(ret != -ENOENT);
2876                 di = btrfs_search_dir_index_item(root, path, dir_ino,
2877                                                  name, name_len);
2878                 BUG_ON(IS_ERR_OR_NULL(di));
2879
2880                 leaf = path->nodes[0];
2881                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2882                 btrfs_release_path(path);
2883                 index = key.offset;
2884         }
2885         btrfs_release_path(path);
2886
2887         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2888         BUG_ON(ret);
2889
2890         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2891         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2892         ret = btrfs_update_inode(trans, root, dir);
2893         BUG_ON(ret);
2894
2895         btrfs_free_path(path);
2896         return 0;
2897 }
2898
2899 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2900 {
2901         struct inode *inode = dentry->d_inode;
2902         int err = 0;
2903         struct btrfs_root *root = BTRFS_I(dir)->root;
2904         struct btrfs_trans_handle *trans;
2905         unsigned long nr = 0;
2906
2907         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2908             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
2909                 return -ENOTEMPTY;
2910
2911         trans = __unlink_start_trans(dir, dentry);
2912         if (IS_ERR(trans))
2913                 return PTR_ERR(trans);
2914
2915         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2916                 err = btrfs_unlink_subvol(trans, root, dir,
2917                                           BTRFS_I(inode)->location.objectid,
2918                                           dentry->d_name.name,
2919                                           dentry->d_name.len);
2920                 goto out;
2921         }
2922
2923         err = btrfs_orphan_add(trans, inode);
2924         if (err)
2925                 goto out;
2926
2927         /* now the directory is empty */
2928         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2929                                  dentry->d_name.name, dentry->d_name.len);
2930         if (!err)
2931                 btrfs_i_size_write(inode, 0);
2932 out:
2933         nr = trans->blocks_used;
2934         __unlink_end_trans(trans, root);
2935         btrfs_btree_balance_dirty(root, nr);
2936
2937         return err;
2938 }
2939
2940 /*
2941  * this can truncate away extent items, csum items and directory items.
2942  * It starts at a high offset and removes keys until it can't find
2943  * any higher than new_size
2944  *
2945  * csum items that cross the new i_size are truncated to the new size
2946  * as well.
2947  *
2948  * min_type is the minimum key type to truncate down to.  If set to 0, this
2949  * will kill all the items on this inode, including the INODE_ITEM_KEY.
2950  */
2951 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2952                                struct btrfs_root *root,
2953                                struct inode *inode,
2954                                u64 new_size, u32 min_type)
2955 {
2956         struct btrfs_path *path;
2957         struct extent_buffer *leaf;
2958         struct btrfs_file_extent_item *fi;
2959         struct btrfs_key key;
2960         struct btrfs_key found_key;
2961         u64 extent_start = 0;
2962         u64 extent_num_bytes = 0;
2963         u64 extent_offset = 0;
2964         u64 item_end = 0;
2965         u64 mask = root->sectorsize - 1;
2966         u32 found_type = (u8)-1;
2967         int found_extent;
2968         int del_item;
2969         int pending_del_nr = 0;
2970         int pending_del_slot = 0;
2971         int extent_type = -1;
2972         int encoding;
2973         int ret;
2974         int err = 0;
2975         u64 ino = btrfs_ino(inode);
2976
2977         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
2978
2979         path = btrfs_alloc_path();
2980         if (!path)
2981                 return -ENOMEM;
2982         path->reada = -1;
2983
2984         if (root->ref_cows || root == root->fs_info->tree_root)
2985                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2986
2987         /*
2988          * This function is also used to drop the items in the log tree before
2989          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
2990          * it is used to drop the loged items. So we shouldn't kill the delayed
2991          * items.
2992          */
2993         if (min_type == 0 && root == BTRFS_I(inode)->root)
2994                 btrfs_kill_delayed_inode_items(inode);
2995
2996         key.objectid = ino;
2997         key.offset = (u64)-1;
2998         key.type = (u8)-1;
2999
3000 search_again:
3001         path->leave_spinning = 1;
3002         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3003         if (ret < 0) {
3004                 err = ret;
3005                 goto out;
3006         }
3007
3008         if (ret > 0) {
3009                 /* there are no items in the tree for us to truncate, we're
3010                  * done
3011                  */
3012                 if (path->slots[0] == 0)
3013                         goto out;
3014                 path->slots[0]--;
3015         }
3016
3017         while (1) {
3018                 fi = NULL;
3019                 leaf = path->nodes[0];
3020                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3021                 found_type = btrfs_key_type(&found_key);
3022                 encoding = 0;
3023
3024                 if (found_key.objectid != ino)
3025                         break;
3026
3027                 if (found_type < min_type)
3028                         break;
3029
3030                 item_end = found_key.offset;
3031                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3032                         fi = btrfs_item_ptr(leaf, path->slots[0],
3033                                             struct btrfs_file_extent_item);
3034                         extent_type = btrfs_file_extent_type(leaf, fi);
3035                         encoding = btrfs_file_extent_compression(leaf, fi);
3036                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3037                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3038
3039                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3040                                 item_end +=
3041                                     btrfs_file_extent_num_bytes(leaf, fi);
3042                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3043                                 item_end += btrfs_file_extent_inline_len(leaf,
3044                                                                          fi);
3045                         }
3046                         item_end--;
3047                 }
3048                 if (found_type > min_type) {
3049                         del_item = 1;
3050                 } else {
3051                         if (item_end < new_size)
3052                                 break;
3053                         if (found_key.offset >= new_size)
3054                                 del_item = 1;
3055                         else
3056                                 del_item = 0;
3057                 }
3058                 found_extent = 0;
3059                 /* FIXME, shrink the extent if the ref count is only 1 */
3060                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3061                         goto delete;
3062
3063                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3064                         u64 num_dec;
3065                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3066                         if (!del_item && !encoding) {
3067                                 u64 orig_num_bytes =
3068                                         btrfs_file_extent_num_bytes(leaf, fi);
3069                                 extent_num_bytes = new_size -
3070                                         found_key.offset + root->sectorsize - 1;
3071                                 extent_num_bytes = extent_num_bytes &
3072                                         ~((u64)root->sectorsize - 1);
3073                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3074                                                          extent_num_bytes);
3075                                 num_dec = (orig_num_bytes -
3076                                            extent_num_bytes);
3077                                 if (root->ref_cows && extent_start != 0)
3078                                         inode_sub_bytes(inode, num_dec);
3079                                 btrfs_mark_buffer_dirty(leaf);
3080                         } else {
3081                                 extent_num_bytes =
3082                                         btrfs_file_extent_disk_num_bytes(leaf,
3083