Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs...
[sfrench/cifs-2.6.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 "compat.h"
41 #include "ctree.h"
42 #include "disk-io.h"
43 #include "transaction.h"
44 #include "btrfs_inode.h"
45 #include "ioctl.h"
46 #include "print-tree.h"
47 #include "volumes.h"
48 #include "ordered-data.h"
49 #include "xattr.h"
50 #include "tree-log.h"
51 #include "compression.h"
52 #include "locking.h"
53 #include "free-space-cache.h"
54
55 struct btrfs_iget_args {
56         u64 ino;
57         struct btrfs_root *root;
58 };
59
60 static const struct inode_operations btrfs_dir_inode_operations;
61 static const struct inode_operations btrfs_symlink_inode_operations;
62 static const struct inode_operations btrfs_dir_ro_inode_operations;
63 static const struct inode_operations btrfs_special_inode_operations;
64 static const struct inode_operations btrfs_file_inode_operations;
65 static const struct address_space_operations btrfs_aops;
66 static const struct address_space_operations btrfs_symlink_aops;
67 static const struct file_operations btrfs_dir_file_operations;
68 static struct extent_io_ops btrfs_extent_io_ops;
69
70 static struct kmem_cache *btrfs_inode_cachep;
71 struct kmem_cache *btrfs_trans_handle_cachep;
72 struct kmem_cache *btrfs_transaction_cachep;
73 struct kmem_cache *btrfs_path_cachep;
74 struct kmem_cache *btrfs_free_space_cachep;
75
76 #define S_SHIFT 12
77 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
78         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
79         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
80         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
81         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
82         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
83         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
84         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
85 };
86
87 static int btrfs_setsize(struct inode *inode, loff_t newsize);
88 static int btrfs_truncate(struct inode *inode);
89 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
90 static noinline int cow_file_range(struct inode *inode,
91                                    struct page *locked_page,
92                                    u64 start, u64 end, int *page_started,
93                                    unsigned long *nr_written, int unlock);
94
95 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
96                                      struct inode *inode,  struct inode *dir,
97                                      const struct qstr *qstr)
98 {
99         int err;
100
101         err = btrfs_init_acl(trans, inode, dir);
102         if (!err)
103                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
104         return err;
105 }
106
107 /*
108  * this does all the hard work for inserting an inline extent into
109  * the btree.  The caller should have done a btrfs_drop_extents so that
110  * no overlapping inline items exist in the btree
111  */
112 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
113                                 struct btrfs_root *root, struct inode *inode,
114                                 u64 start, size_t size, size_t compressed_size,
115                                 int compress_type,
116                                 struct page **compressed_pages)
117 {
118         struct btrfs_key key;
119         struct btrfs_path *path;
120         struct extent_buffer *leaf;
121         struct page *page = NULL;
122         char *kaddr;
123         unsigned long ptr;
124         struct btrfs_file_extent_item *ei;
125         int err = 0;
126         int ret;
127         size_t cur_size = size;
128         size_t datasize;
129         unsigned long offset;
130
131         if (compressed_size && compressed_pages)
132                 cur_size = compressed_size;
133
134         path = btrfs_alloc_path();
135         if (!path)
136                 return -ENOMEM;
137
138         path->leave_spinning = 1;
139         btrfs_set_trans_block_group(trans, inode);
140
141         key.objectid = inode->i_ino;
142         key.offset = start;
143         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
144         datasize = btrfs_file_extent_calc_inline_size(cur_size);
145
146         inode_add_bytes(inode, size);
147         ret = btrfs_insert_empty_item(trans, root, path, &key,
148                                       datasize);
149         BUG_ON(ret);
150         if (ret) {
151                 err = ret;
152                 goto fail;
153         }
154         leaf = path->nodes[0];
155         ei = btrfs_item_ptr(leaf, path->slots[0],
156                             struct btrfs_file_extent_item);
157         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
158         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
159         btrfs_set_file_extent_encryption(leaf, ei, 0);
160         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
161         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
162         ptr = btrfs_file_extent_inline_start(ei);
163
164         if (compress_type != BTRFS_COMPRESS_NONE) {
165                 struct page *cpage;
166                 int i = 0;
167                 while (compressed_size > 0) {
168                         cpage = compressed_pages[i];
169                         cur_size = min_t(unsigned long, compressed_size,
170                                        PAGE_CACHE_SIZE);
171
172                         kaddr = kmap_atomic(cpage, KM_USER0);
173                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
174                         kunmap_atomic(kaddr, KM_USER0);
175
176                         i++;
177                         ptr += cur_size;
178                         compressed_size -= cur_size;
179                 }
180                 btrfs_set_file_extent_compression(leaf, ei,
181                                                   compress_type);
182         } else {
183                 page = find_get_page(inode->i_mapping,
184                                      start >> PAGE_CACHE_SHIFT);
185                 btrfs_set_file_extent_compression(leaf, ei, 0);
186                 kaddr = kmap_atomic(page, KM_USER0);
187                 offset = start & (PAGE_CACHE_SIZE - 1);
188                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
189                 kunmap_atomic(kaddr, KM_USER0);
190                 page_cache_release(page);
191         }
192         btrfs_mark_buffer_dirty(leaf);
193         btrfs_free_path(path);
194
195         /*
196          * we're an inline extent, so nobody can
197          * extend the file past i_size without locking
198          * a page we already have locked.
199          *
200          * We must do any isize and inode updates
201          * before we unlock the pages.  Otherwise we
202          * could end up racing with unlink.
203          */
204         BTRFS_I(inode)->disk_i_size = inode->i_size;
205         btrfs_update_inode(trans, root, inode);
206
207         return 0;
208 fail:
209         btrfs_free_path(path);
210         return err;
211 }
212
213
214 /*
215  * conditionally insert an inline extent into the file.  This
216  * does the checks required to make sure the data is small enough
217  * to fit as an inline extent.
218  */
219 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
220                                  struct btrfs_root *root,
221                                  struct inode *inode, u64 start, u64 end,
222                                  size_t compressed_size, int compress_type,
223                                  struct page **compressed_pages)
224 {
225         u64 isize = i_size_read(inode);
226         u64 actual_end = min(end + 1, isize);
227         u64 inline_len = actual_end - start;
228         u64 aligned_end = (end + root->sectorsize - 1) &
229                         ~((u64)root->sectorsize - 1);
230         u64 hint_byte;
231         u64 data_len = inline_len;
232         int ret;
233
234         if (compressed_size)
235                 data_len = compressed_size;
236
237         if (start > 0 ||
238             actual_end >= PAGE_CACHE_SIZE ||
239             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
240             (!compressed_size &&
241             (actual_end & (root->sectorsize - 1)) == 0) ||
242             end + 1 < isize ||
243             data_len > root->fs_info->max_inline) {
244                 return 1;
245         }
246
247         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
248                                  &hint_byte, 1);
249         BUG_ON(ret);
250
251         if (isize > actual_end)
252                 inline_len = min_t(u64, isize, actual_end);
253         ret = insert_inline_extent(trans, root, inode, start,
254                                    inline_len, compressed_size,
255                                    compress_type, compressed_pages);
256         BUG_ON(ret);
257         btrfs_delalloc_release_metadata(inode, end + 1 - start);
258         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
259         return 0;
260 }
261
262 struct async_extent {
263         u64 start;
264         u64 ram_size;
265         u64 compressed_size;
266         struct page **pages;
267         unsigned long nr_pages;
268         int compress_type;
269         struct list_head list;
270 };
271
272 struct async_cow {
273         struct inode *inode;
274         struct btrfs_root *root;
275         struct page *locked_page;
276         u64 start;
277         u64 end;
278         struct list_head extents;
279         struct btrfs_work work;
280 };
281
282 static noinline int add_async_extent(struct async_cow *cow,
283                                      u64 start, u64 ram_size,
284                                      u64 compressed_size,
285                                      struct page **pages,
286                                      unsigned long nr_pages,
287                                      int compress_type)
288 {
289         struct async_extent *async_extent;
290
291         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
292         BUG_ON(!async_extent);
293         async_extent->start = start;
294         async_extent->ram_size = ram_size;
295         async_extent->compressed_size = compressed_size;
296         async_extent->pages = pages;
297         async_extent->nr_pages = nr_pages;
298         async_extent->compress_type = compress_type;
299         list_add_tail(&async_extent->list, &cow->extents);
300         return 0;
301 }
302
303 /*
304  * we create compressed extents in two phases.  The first
305  * phase compresses a range of pages that have already been
306  * locked (both pages and state bits are locked).
307  *
308  * This is done inside an ordered work queue, and the compression
309  * is spread across many cpus.  The actual IO submission is step
310  * two, and the ordered work queue takes care of making sure that
311  * happens in the same order things were put onto the queue by
312  * writepages and friends.
313  *
314  * If this code finds it can't get good compression, it puts an
315  * entry onto the work queue to write the uncompressed bytes.  This
316  * makes sure that both compressed inodes and uncompressed inodes
317  * are written in the same order that pdflush sent them down.
318  */
319 static noinline int compress_file_range(struct inode *inode,
320                                         struct page *locked_page,
321                                         u64 start, u64 end,
322                                         struct async_cow *async_cow,
323                                         int *num_added)
324 {
325         struct btrfs_root *root = BTRFS_I(inode)->root;
326         struct btrfs_trans_handle *trans;
327         u64 num_bytes;
328         u64 blocksize = root->sectorsize;
329         u64 actual_end;
330         u64 isize = i_size_read(inode);
331         int ret = 0;
332         struct page **pages = NULL;
333         unsigned long nr_pages;
334         unsigned long nr_pages_ret = 0;
335         unsigned long total_compressed = 0;
336         unsigned long total_in = 0;
337         unsigned long max_compressed = 128 * 1024;
338         unsigned long max_uncompressed = 128 * 1024;
339         int i;
340         int will_compress;
341         int compress_type = root->fs_info->compress_type;
342
343         actual_end = min_t(u64, isize, end + 1);
344 again:
345         will_compress = 0;
346         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
347         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
348
349         /*
350          * we don't want to send crud past the end of i_size through
351          * compression, that's just a waste of CPU time.  So, if the
352          * end of the file is before the start of our current
353          * requested range of bytes, we bail out to the uncompressed
354          * cleanup code that can deal with all of this.
355          *
356          * It isn't really the fastest way to fix things, but this is a
357          * very uncommon corner.
358          */
359         if (actual_end <= start)
360                 goto cleanup_and_bail_uncompressed;
361
362         total_compressed = actual_end - start;
363
364         /* we want to make sure that amount of ram required to uncompress
365          * an extent is reasonable, so we limit the total size in ram
366          * of a compressed extent to 128k.  This is a crucial number
367          * because it also controls how easily we can spread reads across
368          * cpus for decompression.
369          *
370          * We also want to make sure the amount of IO required to do
371          * a random read is reasonably small, so we limit the size of
372          * a compressed extent to 128k.
373          */
374         total_compressed = min(total_compressed, max_uncompressed);
375         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
376         num_bytes = max(blocksize,  num_bytes);
377         total_in = 0;
378         ret = 0;
379
380         /*
381          * we do compression for mount -o compress and when the
382          * inode has not been flagged as nocompress.  This flag can
383          * change at any time if we discover bad compression ratios.
384          */
385         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
386             (btrfs_test_opt(root, COMPRESS) ||
387              (BTRFS_I(inode)->force_compress) ||
388              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
389                 WARN_ON(pages);
390                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
391                 BUG_ON(!pages);
392
393                 if (BTRFS_I(inode)->force_compress)
394                         compress_type = BTRFS_I(inode)->force_compress;
395
396                 ret = btrfs_compress_pages(compress_type,
397                                            inode->i_mapping, start,
398                                            total_compressed, pages,
399                                            nr_pages, &nr_pages_ret,
400                                            &total_in,
401                                            &total_compressed,
402                                            max_compressed);
403
404                 if (!ret) {
405                         unsigned long offset = total_compressed &
406                                 (PAGE_CACHE_SIZE - 1);
407                         struct page *page = pages[nr_pages_ret - 1];
408                         char *kaddr;
409
410                         /* zero the tail end of the last page, we might be
411                          * sending it down to disk
412                          */
413                         if (offset) {
414                                 kaddr = kmap_atomic(page, KM_USER0);
415                                 memset(kaddr + offset, 0,
416                                        PAGE_CACHE_SIZE - offset);
417                                 kunmap_atomic(kaddr, KM_USER0);
418                         }
419                         will_compress = 1;
420                 }
421         }
422         if (start == 0) {
423                 trans = btrfs_join_transaction(root, 1);
424                 BUG_ON(IS_ERR(trans));
425                 btrfs_set_trans_block_group(trans, inode);
426                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
427
428                 /* lets try to make an inline extent */
429                 if (ret || total_in < (actual_end - start)) {
430                         /* we didn't compress the entire range, try
431                          * to make an uncompressed inline extent.
432                          */
433                         ret = cow_file_range_inline(trans, root, inode,
434                                                     start, end, 0, 0, NULL);
435                 } else {
436                         /* try making a compressed inline extent */
437                         ret = cow_file_range_inline(trans, root, inode,
438                                                     start, end,
439                                                     total_compressed,
440                                                     compress_type, pages);
441                 }
442                 if (ret == 0) {
443                         /*
444                          * inline extent creation worked, we don't need
445                          * to create any more async work items.  Unlock
446                          * and free up our temp pages.
447                          */
448                         extent_clear_unlock_delalloc(inode,
449                              &BTRFS_I(inode)->io_tree,
450                              start, end, NULL,
451                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
452                              EXTENT_CLEAR_DELALLOC |
453                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
454
455                         btrfs_end_transaction(trans, root);
456                         goto free_pages_out;
457                 }
458                 btrfs_end_transaction(trans, root);
459         }
460
461         if (will_compress) {
462                 /*
463                  * we aren't doing an inline extent round the compressed size
464                  * up to a block size boundary so the allocator does sane
465                  * things
466                  */
467                 total_compressed = (total_compressed + blocksize - 1) &
468                         ~(blocksize - 1);
469
470                 /*
471                  * one last check to make sure the compression is really a
472                  * win, compare the page count read with the blocks on disk
473                  */
474                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
475                         ~(PAGE_CACHE_SIZE - 1);
476                 if (total_compressed >= total_in) {
477                         will_compress = 0;
478                 } else {
479                         num_bytes = total_in;
480                 }
481         }
482         if (!will_compress && pages) {
483                 /*
484                  * the compression code ran but failed to make things smaller,
485                  * free any pages it allocated and our page pointer array
486                  */
487                 for (i = 0; i < nr_pages_ret; i++) {
488                         WARN_ON(pages[i]->mapping);
489                         page_cache_release(pages[i]);
490                 }
491                 kfree(pages);
492                 pages = NULL;
493                 total_compressed = 0;
494                 nr_pages_ret = 0;
495
496                 /* flag the file so we don't compress in the future */
497                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
498                     !(BTRFS_I(inode)->force_compress)) {
499                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
500                 }
501         }
502         if (will_compress) {
503                 *num_added += 1;
504
505                 /* the async work queues will take care of doing actual
506                  * allocation on disk for these compressed pages,
507                  * and will submit them to the elevator.
508                  */
509                 add_async_extent(async_cow, start, num_bytes,
510                                  total_compressed, pages, nr_pages_ret,
511                                  compress_type);
512
513                 if (start + num_bytes < end) {
514                         start += num_bytes;
515                         pages = NULL;
516                         cond_resched();
517                         goto again;
518                 }
519         } else {
520 cleanup_and_bail_uncompressed:
521                 /*
522                  * No compression, but we still need to write the pages in
523                  * the file we've been given so far.  redirty the locked
524                  * page if it corresponds to our extent and set things up
525                  * for the async work queue to run cow_file_range to do
526                  * the normal delalloc dance
527                  */
528                 if (page_offset(locked_page) >= start &&
529                     page_offset(locked_page) <= end) {
530                         __set_page_dirty_nobuffers(locked_page);
531                         /* unlocked later on in the async handlers */
532                 }
533                 add_async_extent(async_cow, start, end - start + 1,
534                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
535                 *num_added += 1;
536         }
537
538 out:
539         return 0;
540
541 free_pages_out:
542         for (i = 0; i < nr_pages_ret; i++) {
543                 WARN_ON(pages[i]->mapping);
544                 page_cache_release(pages[i]);
545         }
546         kfree(pages);
547
548         goto out;
549 }
550
551 /*
552  * phase two of compressed writeback.  This is the ordered portion
553  * of the code, which only gets called in the order the work was
554  * queued.  We walk all the async extents created by compress_file_range
555  * and send them down to the disk.
556  */
557 static noinline int submit_compressed_extents(struct inode *inode,
558                                               struct async_cow *async_cow)
559 {
560         struct async_extent *async_extent;
561         u64 alloc_hint = 0;
562         struct btrfs_trans_handle *trans;
563         struct btrfs_key ins;
564         struct extent_map *em;
565         struct btrfs_root *root = BTRFS_I(inode)->root;
566         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
567         struct extent_io_tree *io_tree;
568         int ret = 0;
569
570         if (list_empty(&async_cow->extents))
571                 return 0;
572
573
574         while (!list_empty(&async_cow->extents)) {
575                 async_extent = list_entry(async_cow->extents.next,
576                                           struct async_extent, list);
577                 list_del(&async_extent->list);
578
579                 io_tree = &BTRFS_I(inode)->io_tree;
580
581 retry:
582                 /* did the compression code fall back to uncompressed IO? */
583                 if (!async_extent->pages) {
584                         int page_started = 0;
585                         unsigned long nr_written = 0;
586
587                         lock_extent(io_tree, async_extent->start,
588                                          async_extent->start +
589                                          async_extent->ram_size - 1, GFP_NOFS);
590
591                         /* allocate blocks */
592                         ret = cow_file_range(inode, async_cow->locked_page,
593                                              async_extent->start,
594                                              async_extent->start +
595                                              async_extent->ram_size - 1,
596                                              &page_started, &nr_written, 0);
597
598                         /*
599                          * if page_started, cow_file_range inserted an
600                          * inline extent and took care of all the unlocking
601                          * and IO for us.  Otherwise, we need to submit
602                          * all those pages down to the drive.
603                          */
604                         if (!page_started && !ret)
605                                 extent_write_locked_range(io_tree,
606                                                   inode, async_extent->start,
607                                                   async_extent->start +
608                                                   async_extent->ram_size - 1,
609                                                   btrfs_get_extent,
610                                                   WB_SYNC_ALL);
611                         kfree(async_extent);
612                         cond_resched();
613                         continue;
614                 }
615
616                 lock_extent(io_tree, async_extent->start,
617                             async_extent->start + async_extent->ram_size - 1,
618                             GFP_NOFS);
619
620                 trans = btrfs_join_transaction(root, 1);
621                 BUG_ON(IS_ERR(trans));
622                 ret = btrfs_reserve_extent(trans, root,
623                                            async_extent->compressed_size,
624                                            async_extent->compressed_size,
625                                            0, alloc_hint,
626                                            (u64)-1, &ins, 1);
627                 btrfs_end_transaction(trans, root);
628
629                 if (ret) {
630                         int i;
631                         for (i = 0; i < async_extent->nr_pages; i++) {
632                                 WARN_ON(async_extent->pages[i]->mapping);
633                                 page_cache_release(async_extent->pages[i]);
634                         }
635                         kfree(async_extent->pages);
636                         async_extent->nr_pages = 0;
637                         async_extent->pages = NULL;
638                         unlock_extent(io_tree, async_extent->start,
639                                       async_extent->start +
640                                       async_extent->ram_size - 1, GFP_NOFS);
641                         goto retry;
642                 }
643
644                 /*
645                  * here we're doing allocation and writeback of the
646                  * compressed pages
647                  */
648                 btrfs_drop_extent_cache(inode, async_extent->start,
649                                         async_extent->start +
650                                         async_extent->ram_size - 1, 0);
651
652                 em = alloc_extent_map(GFP_NOFS);
653                 BUG_ON(!em);
654                 em->start = async_extent->start;
655                 em->len = async_extent->ram_size;
656                 em->orig_start = em->start;
657
658                 em->block_start = ins.objectid;
659                 em->block_len = ins.offset;
660                 em->bdev = root->fs_info->fs_devices->latest_bdev;
661                 em->compress_type = async_extent->compress_type;
662                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
663                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
664
665                 while (1) {
666                         write_lock(&em_tree->lock);
667                         ret = add_extent_mapping(em_tree, em);
668                         write_unlock(&em_tree->lock);
669                         if (ret != -EEXIST) {
670                                 free_extent_map(em);
671                                 break;
672                         }
673                         btrfs_drop_extent_cache(inode, async_extent->start,
674                                                 async_extent->start +
675                                                 async_extent->ram_size - 1, 0);
676                 }
677
678                 ret = btrfs_add_ordered_extent_compress(inode,
679                                                 async_extent->start,
680                                                 ins.objectid,
681                                                 async_extent->ram_size,
682                                                 ins.offset,
683                                                 BTRFS_ORDERED_COMPRESSED,
684                                                 async_extent->compress_type);
685                 BUG_ON(ret);
686
687                 /*
688                  * clear dirty, set writeback and unlock the pages.
689                  */
690                 extent_clear_unlock_delalloc(inode,
691                                 &BTRFS_I(inode)->io_tree,
692                                 async_extent->start,
693                                 async_extent->start +
694                                 async_extent->ram_size - 1,
695                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
696                                 EXTENT_CLEAR_UNLOCK |
697                                 EXTENT_CLEAR_DELALLOC |
698                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
699
700                 ret = btrfs_submit_compressed_write(inode,
701                                     async_extent->start,
702                                     async_extent->ram_size,
703                                     ins.objectid,
704                                     ins.offset, async_extent->pages,
705                                     async_extent->nr_pages);
706
707                 BUG_ON(ret);
708                 alloc_hint = ins.objectid + ins.offset;
709                 kfree(async_extent);
710                 cond_resched();
711         }
712
713         return 0;
714 }
715
716 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
717                                       u64 num_bytes)
718 {
719         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
720         struct extent_map *em;
721         u64 alloc_hint = 0;
722
723         read_lock(&em_tree->lock);
724         em = search_extent_mapping(em_tree, start, num_bytes);
725         if (em) {
726                 /*
727                  * if block start isn't an actual block number then find the
728                  * first block in this inode and use that as a hint.  If that
729                  * block is also bogus then just don't worry about it.
730                  */
731                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
732                         free_extent_map(em);
733                         em = search_extent_mapping(em_tree, 0, 0);
734                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
735                                 alloc_hint = em->block_start;
736                         if (em)
737                                 free_extent_map(em);
738                 } else {
739                         alloc_hint = em->block_start;
740                         free_extent_map(em);
741                 }
742         }
743         read_unlock(&em_tree->lock);
744
745         return alloc_hint;
746 }
747
748 /*
749  * when extent_io.c finds a delayed allocation range in the file,
750  * the call backs end up in this code.  The basic idea is to
751  * allocate extents on disk for the range, and create ordered data structs
752  * in ram to track those extents.
753  *
754  * locked_page is the page that writepage had locked already.  We use
755  * it to make sure we don't do extra locks or unlocks.
756  *
757  * *page_started is set to one if we unlock locked_page and do everything
758  * required to start IO on it.  It may be clean and already done with
759  * IO when we return.
760  */
761 static noinline int cow_file_range(struct inode *inode,
762                                    struct page *locked_page,
763                                    u64 start, u64 end, int *page_started,
764                                    unsigned long *nr_written,
765                                    int unlock)
766 {
767         struct btrfs_root *root = BTRFS_I(inode)->root;
768         struct btrfs_trans_handle *trans;
769         u64 alloc_hint = 0;
770         u64 num_bytes;
771         unsigned long ram_size;
772         u64 disk_num_bytes;
773         u64 cur_alloc_size;
774         u64 blocksize = root->sectorsize;
775         struct btrfs_key ins;
776         struct extent_map *em;
777         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
778         int ret = 0;
779
780         BUG_ON(root == root->fs_info->tree_root);
781         trans = btrfs_join_transaction(root, 1);
782         BUG_ON(IS_ERR(trans));
783         btrfs_set_trans_block_group(trans, inode);
784         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
785
786         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
787         num_bytes = max(blocksize,  num_bytes);
788         disk_num_bytes = num_bytes;
789         ret = 0;
790
791         if (start == 0) {
792                 /* lets try to make an inline extent */
793                 ret = cow_file_range_inline(trans, root, inode,
794                                             start, end, 0, 0, NULL);
795                 if (ret == 0) {
796                         extent_clear_unlock_delalloc(inode,
797                                      &BTRFS_I(inode)->io_tree,
798                                      start, end, NULL,
799                                      EXTENT_CLEAR_UNLOCK_PAGE |
800                                      EXTENT_CLEAR_UNLOCK |
801                                      EXTENT_CLEAR_DELALLOC |
802                                      EXTENT_CLEAR_DIRTY |
803                                      EXTENT_SET_WRITEBACK |
804                                      EXTENT_END_WRITEBACK);
805
806                         *nr_written = *nr_written +
807                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
808                         *page_started = 1;
809                         ret = 0;
810                         goto out;
811                 }
812         }
813
814         BUG_ON(disk_num_bytes >
815                btrfs_super_total_bytes(&root->fs_info->super_copy));
816
817         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
818         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
819
820         while (disk_num_bytes > 0) {
821                 unsigned long op;
822
823                 cur_alloc_size = disk_num_bytes;
824                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
825                                            root->sectorsize, 0, alloc_hint,
826                                            (u64)-1, &ins, 1);
827                 BUG_ON(ret);
828
829                 em = alloc_extent_map(GFP_NOFS);
830                 BUG_ON(!em);
831                 em->start = start;
832                 em->orig_start = em->start;
833                 ram_size = ins.offset;
834                 em->len = ins.offset;
835
836                 em->block_start = ins.objectid;
837                 em->block_len = ins.offset;
838                 em->bdev = root->fs_info->fs_devices->latest_bdev;
839                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
840
841                 while (1) {
842                         write_lock(&em_tree->lock);
843                         ret = add_extent_mapping(em_tree, em);
844                         write_unlock(&em_tree->lock);
845                         if (ret != -EEXIST) {
846                                 free_extent_map(em);
847                                 break;
848                         }
849                         btrfs_drop_extent_cache(inode, start,
850                                                 start + ram_size - 1, 0);
851                 }
852
853                 cur_alloc_size = ins.offset;
854                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
855                                                ram_size, cur_alloc_size, 0);
856                 BUG_ON(ret);
857
858                 if (root->root_key.objectid ==
859                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
860                         ret = btrfs_reloc_clone_csums(inode, start,
861                                                       cur_alloc_size);
862                         BUG_ON(ret);
863                 }
864
865                 if (disk_num_bytes < cur_alloc_size)
866                         break;
867
868                 /* we're not doing compressed IO, don't unlock the first
869                  * page (which the caller expects to stay locked), don't
870                  * clear any dirty bits and don't set any writeback bits
871                  *
872                  * Do set the Private2 bit so we know this page was properly
873                  * setup for writepage
874                  */
875                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
876                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
877                         EXTENT_SET_PRIVATE2;
878
879                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
880                                              start, start + ram_size - 1,
881                                              locked_page, op);
882                 disk_num_bytes -= cur_alloc_size;
883                 num_bytes -= cur_alloc_size;
884                 alloc_hint = ins.objectid + ins.offset;
885                 start += cur_alloc_size;
886         }
887 out:
888         ret = 0;
889         btrfs_end_transaction(trans, root);
890
891         return ret;
892 }
893
894 /*
895  * work queue call back to started compression on a file and pages
896  */
897 static noinline void async_cow_start(struct btrfs_work *work)
898 {
899         struct async_cow *async_cow;
900         int num_added = 0;
901         async_cow = container_of(work, struct async_cow, work);
902
903         compress_file_range(async_cow->inode, async_cow->locked_page,
904                             async_cow->start, async_cow->end, async_cow,
905                             &num_added);
906         if (num_added == 0)
907                 async_cow->inode = NULL;
908 }
909
910 /*
911  * work queue call back to submit previously compressed pages
912  */
913 static noinline void async_cow_submit(struct btrfs_work *work)
914 {
915         struct async_cow *async_cow;
916         struct btrfs_root *root;
917         unsigned long nr_pages;
918
919         async_cow = container_of(work, struct async_cow, work);
920
921         root = async_cow->root;
922         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
923                 PAGE_CACHE_SHIFT;
924
925         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
926
927         if (atomic_read(&root->fs_info->async_delalloc_pages) <
928             5 * 1042 * 1024 &&
929             waitqueue_active(&root->fs_info->async_submit_wait))
930                 wake_up(&root->fs_info->async_submit_wait);
931
932         if (async_cow->inode)
933                 submit_compressed_extents(async_cow->inode, async_cow);
934 }
935
936 static noinline void async_cow_free(struct btrfs_work *work)
937 {
938         struct async_cow *async_cow;
939         async_cow = container_of(work, struct async_cow, work);
940         kfree(async_cow);
941 }
942
943 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
944                                 u64 start, u64 end, int *page_started,
945                                 unsigned long *nr_written)
946 {
947         struct async_cow *async_cow;
948         struct btrfs_root *root = BTRFS_I(inode)->root;
949         unsigned long nr_pages;
950         u64 cur_end;
951         int limit = 10 * 1024 * 1042;
952
953         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
954                          1, 0, NULL, GFP_NOFS);
955         while (start < end) {
956                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
957                 async_cow->inode = inode;
958                 async_cow->root = root;
959                 async_cow->locked_page = locked_page;
960                 async_cow->start = start;
961
962                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
963                         cur_end = end;
964                 else
965                         cur_end = min(end, start + 512 * 1024 - 1);
966
967                 async_cow->end = cur_end;
968                 INIT_LIST_HEAD(&async_cow->extents);
969
970                 async_cow->work.func = async_cow_start;
971                 async_cow->work.ordered_func = async_cow_submit;
972                 async_cow->work.ordered_free = async_cow_free;
973                 async_cow->work.flags = 0;
974
975                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
976                         PAGE_CACHE_SHIFT;
977                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
978
979                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
980                                    &async_cow->work);
981
982                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
983                         wait_event(root->fs_info->async_submit_wait,
984                            (atomic_read(&root->fs_info->async_delalloc_pages) <
985                             limit));
986                 }
987
988                 while (atomic_read(&root->fs_info->async_submit_draining) &&
989                       atomic_read(&root->fs_info->async_delalloc_pages)) {
990                         wait_event(root->fs_info->async_submit_wait,
991                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
992                            0));
993                 }
994
995                 *nr_written += nr_pages;
996                 start = cur_end + 1;
997         }
998         *page_started = 1;
999         return 0;
1000 }
1001
1002 static noinline int csum_exist_in_range(struct btrfs_root *root,
1003                                         u64 bytenr, u64 num_bytes)
1004 {
1005         int ret;
1006         struct btrfs_ordered_sum *sums;
1007         LIST_HEAD(list);
1008
1009         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1010                                        bytenr + num_bytes - 1, &list);
1011         if (ret == 0 && list_empty(&list))
1012                 return 0;
1013
1014         while (!list_empty(&list)) {
1015                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1016                 list_del(&sums->list);
1017                 kfree(sums);
1018         }
1019         return 1;
1020 }
1021
1022 /*
1023  * when nowcow writeback call back.  This checks for snapshots or COW copies
1024  * of the extents that exist in the file, and COWs the file as required.
1025  *
1026  * If no cow copies or snapshots exist, we write directly to the existing
1027  * blocks on disk
1028  */
1029 static noinline int run_delalloc_nocow(struct inode *inode,
1030                                        struct page *locked_page,
1031                               u64 start, u64 end, int *page_started, int force,
1032                               unsigned long *nr_written)
1033 {
1034         struct btrfs_root *root = BTRFS_I(inode)->root;
1035         struct btrfs_trans_handle *trans;
1036         struct extent_buffer *leaf;
1037         struct btrfs_path *path;
1038         struct btrfs_file_extent_item *fi;
1039         struct btrfs_key found_key;
1040         u64 cow_start;
1041         u64 cur_offset;
1042         u64 extent_end;
1043         u64 extent_offset;
1044         u64 disk_bytenr;
1045         u64 num_bytes;
1046         int extent_type;
1047         int ret;
1048         int type;
1049         int nocow;
1050         int check_prev = 1;
1051         bool nolock = false;
1052
1053         path = btrfs_alloc_path();
1054         BUG_ON(!path);
1055         if (root == root->fs_info->tree_root) {
1056                 nolock = true;
1057                 trans = btrfs_join_transaction_nolock(root, 1);
1058         } else {
1059                 trans = btrfs_join_transaction(root, 1);
1060         }
1061         BUG_ON(IS_ERR(trans));
1062
1063         cow_start = (u64)-1;
1064         cur_offset = start;
1065         while (1) {
1066                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1067                                                cur_offset, 0);
1068                 BUG_ON(ret < 0);
1069                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1070                         leaf = path->nodes[0];
1071                         btrfs_item_key_to_cpu(leaf, &found_key,
1072                                               path->slots[0] - 1);
1073                         if (found_key.objectid == inode->i_ino &&
1074                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1075                                 path->slots[0]--;
1076                 }
1077                 check_prev = 0;
1078 next_slot:
1079                 leaf = path->nodes[0];
1080                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1081                         ret = btrfs_next_leaf(root, path);
1082                         if (ret < 0)
1083                                 BUG_ON(1);
1084                         if (ret > 0)
1085                                 break;
1086                         leaf = path->nodes[0];
1087                 }
1088
1089                 nocow = 0;
1090                 disk_bytenr = 0;
1091                 num_bytes = 0;
1092                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1093
1094                 if (found_key.objectid > inode->i_ino ||
1095                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1096                     found_key.offset > end)
1097                         break;
1098
1099                 if (found_key.offset > cur_offset) {
1100                         extent_end = found_key.offset;
1101                         extent_type = 0;
1102                         goto out_check;
1103                 }
1104
1105                 fi = btrfs_item_ptr(leaf, path->slots[0],
1106                                     struct btrfs_file_extent_item);
1107                 extent_type = btrfs_file_extent_type(leaf, fi);
1108
1109                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1110                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1111                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1112                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1113                         extent_end = found_key.offset +
1114                                 btrfs_file_extent_num_bytes(leaf, fi);
1115                         if (extent_end <= start) {
1116                                 path->slots[0]++;
1117                                 goto next_slot;
1118                         }
1119                         if (disk_bytenr == 0)
1120                                 goto out_check;
1121                         if (btrfs_file_extent_compression(leaf, fi) ||
1122                             btrfs_file_extent_encryption(leaf, fi) ||
1123                             btrfs_file_extent_other_encoding(leaf, fi))
1124                                 goto out_check;
1125                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1126                                 goto out_check;
1127                         if (btrfs_extent_readonly(root, disk_bytenr))
1128                                 goto out_check;
1129                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1130                                                   found_key.offset -
1131                                                   extent_offset, disk_bytenr))
1132                                 goto out_check;
1133                         disk_bytenr += extent_offset;
1134                         disk_bytenr += cur_offset - found_key.offset;
1135                         num_bytes = min(end + 1, extent_end) - cur_offset;
1136                         /*
1137                          * force cow if csum exists in the range.
1138                          * this ensure that csum for a given extent are
1139                          * either valid or do not exist.
1140                          */
1141                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1142                                 goto out_check;
1143                         nocow = 1;
1144                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1145                         extent_end = found_key.offset +
1146                                 btrfs_file_extent_inline_len(leaf, fi);
1147                         extent_end = ALIGN(extent_end, root->sectorsize);
1148                 } else {
1149                         BUG_ON(1);
1150                 }
1151 out_check:
1152                 if (extent_end <= start) {
1153                         path->slots[0]++;
1154                         goto next_slot;
1155                 }
1156                 if (!nocow) {
1157                         if (cow_start == (u64)-1)
1158                                 cow_start = cur_offset;
1159                         cur_offset = extent_end;
1160                         if (cur_offset > end)
1161                                 break;
1162                         path->slots[0]++;
1163                         goto next_slot;
1164                 }
1165
1166                 btrfs_release_path(root, path);
1167                 if (cow_start != (u64)-1) {
1168                         ret = cow_file_range(inode, locked_page, cow_start,
1169                                         found_key.offset - 1, page_started,
1170                                         nr_written, 1);
1171                         BUG_ON(ret);
1172                         cow_start = (u64)-1;
1173                 }
1174
1175                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1176                         struct extent_map *em;
1177                         struct extent_map_tree *em_tree;
1178                         em_tree = &BTRFS_I(inode)->extent_tree;
1179                         em = alloc_extent_map(GFP_NOFS);
1180                         BUG_ON(!em);
1181                         em->start = cur_offset;
1182                         em->orig_start = em->start;
1183                         em->len = num_bytes;
1184                         em->block_len = num_bytes;
1185                         em->block_start = disk_bytenr;
1186                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1187                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1188                         while (1) {
1189                                 write_lock(&em_tree->lock);
1190                                 ret = add_extent_mapping(em_tree, em);
1191                                 write_unlock(&em_tree->lock);
1192                                 if (ret != -EEXIST) {
1193                                         free_extent_map(em);
1194                                         break;
1195                                 }
1196                                 btrfs_drop_extent_cache(inode, em->start,
1197                                                 em->start + em->len - 1, 0);
1198                         }
1199                         type = BTRFS_ORDERED_PREALLOC;
1200                 } else {
1201                         type = BTRFS_ORDERED_NOCOW;
1202                 }
1203
1204                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1205                                                num_bytes, num_bytes, type);
1206                 BUG_ON(ret);
1207
1208                 if (root->root_key.objectid ==
1209                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1210                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1211                                                       num_bytes);
1212                         BUG_ON(ret);
1213                 }
1214
1215                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1216                                 cur_offset, cur_offset + num_bytes - 1,
1217                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1218                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1219                                 EXTENT_SET_PRIVATE2);
1220                 cur_offset = extent_end;
1221                 if (cur_offset > end)
1222                         break;
1223         }
1224         btrfs_release_path(root, path);
1225
1226         if (cur_offset <= end && cow_start == (u64)-1)
1227                 cow_start = cur_offset;
1228         if (cow_start != (u64)-1) {
1229                 ret = cow_file_range(inode, locked_page, cow_start, end,
1230                                      page_started, nr_written, 1);
1231                 BUG_ON(ret);
1232         }
1233
1234         if (nolock) {
1235                 ret = btrfs_end_transaction_nolock(trans, root);
1236                 BUG_ON(ret);
1237         } else {
1238                 ret = btrfs_end_transaction(trans, root);
1239                 BUG_ON(ret);
1240         }
1241         btrfs_free_path(path);
1242         return 0;
1243 }
1244
1245 /*
1246  * extent_io.c call back to do delayed allocation processing
1247  */
1248 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1249                               u64 start, u64 end, int *page_started,
1250                               unsigned long *nr_written)
1251 {
1252         int ret;
1253         struct btrfs_root *root = BTRFS_I(inode)->root;
1254
1255         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1256                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1257                                          page_started, 1, nr_written);
1258         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1259                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1260                                          page_started, 0, nr_written);
1261         else if (!btrfs_test_opt(root, COMPRESS) &&
1262                  !(BTRFS_I(inode)->force_compress) &&
1263                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1264                 ret = cow_file_range(inode, locked_page, start, end,
1265                                       page_started, nr_written, 1);
1266         else
1267                 ret = cow_file_range_async(inode, locked_page, start, end,
1268                                            page_started, nr_written);
1269         return ret;
1270 }
1271
1272 static int btrfs_split_extent_hook(struct inode *inode,
1273                                    struct extent_state *orig, u64 split)
1274 {
1275         /* not delalloc, ignore it */
1276         if (!(orig->state & EXTENT_DELALLOC))
1277                 return 0;
1278
1279         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1280         return 0;
1281 }
1282
1283 /*
1284  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1285  * extents so we can keep track of new extents that are just merged onto old
1286  * extents, such as when we are doing sequential writes, so we can properly
1287  * account for the metadata space we'll need.
1288  */
1289 static int btrfs_merge_extent_hook(struct inode *inode,
1290                                    struct extent_state *new,
1291                                    struct extent_state *other)
1292 {
1293         /* not delalloc, ignore it */
1294         if (!(other->state & EXTENT_DELALLOC))
1295                 return 0;
1296
1297         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1298         return 0;
1299 }
1300
1301 /*
1302  * extent_io.c set_bit_hook, used to track delayed allocation
1303  * bytes in this file, and to maintain the list of inodes that
1304  * have pending delalloc work to be done.
1305  */
1306 static int btrfs_set_bit_hook(struct inode *inode,
1307                               struct extent_state *state, int *bits)
1308 {
1309
1310         /*
1311          * set_bit and clear bit hooks normally require _irqsave/restore
1312          * but in this case, we are only testeing for the DELALLOC
1313          * bit, which is only set or cleared with irqs on
1314          */
1315         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1316                 struct btrfs_root *root = BTRFS_I(inode)->root;
1317                 u64 len = state->end + 1 - state->start;
1318                 int do_list = (root->root_key.objectid !=
1319                                BTRFS_ROOT_TREE_OBJECTID);
1320
1321                 if (*bits & EXTENT_FIRST_DELALLOC)
1322                         *bits &= ~EXTENT_FIRST_DELALLOC;
1323                 else
1324                         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1325
1326                 spin_lock(&root->fs_info->delalloc_lock);
1327                 BTRFS_I(inode)->delalloc_bytes += len;
1328                 root->fs_info->delalloc_bytes += len;
1329                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1330                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1331                                       &root->fs_info->delalloc_inodes);
1332                 }
1333                 spin_unlock(&root->fs_info->delalloc_lock);
1334         }
1335         return 0;
1336 }
1337
1338 /*
1339  * extent_io.c clear_bit_hook, see set_bit_hook for why
1340  */
1341 static int btrfs_clear_bit_hook(struct inode *inode,
1342                                 struct extent_state *state, int *bits)
1343 {
1344         /*
1345          * set_bit and clear bit hooks normally require _irqsave/restore
1346          * but in this case, we are only testeing for the DELALLOC
1347          * bit, which is only set or cleared with irqs on
1348          */
1349         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1350                 struct btrfs_root *root = BTRFS_I(inode)->root;
1351                 u64 len = state->end + 1 - state->start;
1352                 int do_list = (root->root_key.objectid !=
1353                                BTRFS_ROOT_TREE_OBJECTID);
1354
1355                 if (*bits & EXTENT_FIRST_DELALLOC)
1356                         *bits &= ~EXTENT_FIRST_DELALLOC;
1357                 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1358                         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1359
1360                 if (*bits & EXTENT_DO_ACCOUNTING)
1361                         btrfs_delalloc_release_metadata(inode, len);
1362
1363                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1364                     && do_list)
1365                         btrfs_free_reserved_data_space(inode, len);
1366
1367                 spin_lock(&root->fs_info->delalloc_lock);
1368                 root->fs_info->delalloc_bytes -= len;
1369                 BTRFS_I(inode)->delalloc_bytes -= len;
1370
1371                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1372                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1373                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1374                 }
1375                 spin_unlock(&root->fs_info->delalloc_lock);
1376         }
1377         return 0;
1378 }
1379
1380 /*
1381  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1382  * we don't create bios that span stripes or chunks
1383  */
1384 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1385                          size_t size, struct bio *bio,
1386                          unsigned long bio_flags)
1387 {
1388         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1389         struct btrfs_mapping_tree *map_tree;
1390         u64 logical = (u64)bio->bi_sector << 9;
1391         u64 length = 0;
1392         u64 map_length;
1393         int ret;
1394
1395         if (bio_flags & EXTENT_BIO_COMPRESSED)
1396                 return 0;
1397
1398         length = bio->bi_size;
1399         map_tree = &root->fs_info->mapping_tree;
1400         map_length = length;
1401         ret = btrfs_map_block(map_tree, READ, logical,
1402                               &map_length, NULL, 0);
1403
1404         if (map_length < length + size)
1405                 return 1;
1406         return ret;
1407 }
1408
1409 /*
1410  * in order to insert checksums into the metadata in large chunks,
1411  * we wait until bio submission time.   All the pages in the bio are
1412  * checksummed and sums are attached onto the ordered extent record.
1413  *
1414  * At IO completion time the cums attached on the ordered extent record
1415  * are inserted into the btree
1416  */
1417 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1418                                     struct bio *bio, int mirror_num,
1419                                     unsigned long bio_flags,
1420                                     u64 bio_offset)
1421 {
1422         struct btrfs_root *root = BTRFS_I(inode)->root;
1423         int ret = 0;
1424
1425         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1426         BUG_ON(ret);
1427         return 0;
1428 }
1429
1430 /*
1431  * in order to insert checksums into the metadata in large chunks,
1432  * we wait until bio submission time.   All the pages in the bio are
1433  * checksummed and sums are attached onto the ordered extent record.
1434  *
1435  * At IO completion time the cums attached on the ordered extent record
1436  * are inserted into the btree
1437  */
1438 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1439                           int mirror_num, unsigned long bio_flags,
1440                           u64 bio_offset)
1441 {
1442         struct btrfs_root *root = BTRFS_I(inode)->root;
1443         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1444 }
1445
1446 /*
1447  * extent_io.c submission hook. This does the right thing for csum calculation
1448  * on write, or reading the csums from the tree before a read
1449  */
1450 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1451                           int mirror_num, unsigned long bio_flags,
1452                           u64 bio_offset)
1453 {
1454         struct btrfs_root *root = BTRFS_I(inode)->root;
1455         int ret = 0;
1456         int skip_sum;
1457
1458         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1459
1460         if (root == root->fs_info->tree_root)
1461                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1462         else
1463                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1464         BUG_ON(ret);
1465
1466         if (!(rw & REQ_WRITE)) {
1467                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1468                         return btrfs_submit_compressed_read(inode, bio,
1469                                                     mirror_num, bio_flags);
1470                 } else if (!skip_sum) {
1471                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1472                         if (ret)
1473                                 return ret;
1474                 }
1475                 goto mapit;
1476         } else if (!skip_sum) {
1477                 /* csum items have already been cloned */
1478                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1479                         goto mapit;
1480                 /* we're doing a write, do the async checksumming */
1481                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1482                                    inode, rw, bio, mirror_num,
1483                                    bio_flags, bio_offset,
1484                                    __btrfs_submit_bio_start,
1485                                    __btrfs_submit_bio_done);
1486         }
1487
1488 mapit:
1489         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1490 }
1491
1492 /*
1493  * given a list of ordered sums record them in the inode.  This happens
1494  * at IO completion time based on sums calculated at bio submission time.
1495  */
1496 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1497                              struct inode *inode, u64 file_offset,
1498                              struct list_head *list)
1499 {
1500         struct btrfs_ordered_sum *sum;
1501
1502         btrfs_set_trans_block_group(trans, inode);
1503
1504         list_for_each_entry(sum, list, list) {
1505                 btrfs_csum_file_blocks(trans,
1506                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1507         }
1508         return 0;
1509 }
1510
1511 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1512                               struct extent_state **cached_state)
1513 {
1514         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1515                 WARN_ON(1);
1516         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1517                                    cached_state, GFP_NOFS);
1518 }
1519
1520 /* see btrfs_writepage_start_hook for details on why this is required */
1521 struct btrfs_writepage_fixup {
1522         struct page *page;
1523         struct btrfs_work work;
1524 };
1525
1526 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1527 {
1528         struct btrfs_writepage_fixup *fixup;
1529         struct btrfs_ordered_extent *ordered;
1530         struct extent_state *cached_state = NULL;
1531         struct page *page;
1532         struct inode *inode;
1533         u64 page_start;
1534         u64 page_end;
1535
1536         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1537         page = fixup->page;
1538 again:
1539         lock_page(page);
1540         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1541                 ClearPageChecked(page);
1542                 goto out_page;
1543         }
1544
1545         inode = page->mapping->host;
1546         page_start = page_offset(page);
1547         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1548
1549         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1550                          &cached_state, GFP_NOFS);
1551
1552         /* already ordered? We're done */
1553         if (PagePrivate2(page))
1554                 goto out;
1555
1556         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1557         if (ordered) {
1558                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1559                                      page_end, &cached_state, GFP_NOFS);
1560                 unlock_page(page);
1561                 btrfs_start_ordered_extent(inode, ordered, 1);
1562                 goto again;
1563         }
1564
1565         BUG();
1566         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1567         ClearPageChecked(page);
1568 out:
1569         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1570                              &cached_state, GFP_NOFS);
1571 out_page:
1572         unlock_page(page);
1573         page_cache_release(page);
1574         kfree(fixup);
1575 }
1576
1577 /*
1578  * There are a few paths in the higher layers of the kernel that directly
1579  * set the page dirty bit without asking the filesystem if it is a
1580  * good idea.  This causes problems because we want to make sure COW
1581  * properly happens and the data=ordered rules are followed.
1582  *
1583  * In our case any range that doesn't have the ORDERED bit set
1584  * hasn't been properly setup for IO.  We kick off an async process
1585  * to fix it up.  The async helper will wait for ordered extents, set
1586  * the delalloc bit and make it safe to write the page.
1587  */
1588 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1589 {
1590         struct inode *inode = page->mapping->host;
1591         struct btrfs_writepage_fixup *fixup;
1592         struct btrfs_root *root = BTRFS_I(inode)->root;
1593
1594         /* this page is properly in the ordered list */
1595         if (TestClearPagePrivate2(page))
1596                 return 0;
1597
1598         if (PageChecked(page))
1599                 return -EAGAIN;
1600
1601         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1602         if (!fixup)
1603                 return -EAGAIN;
1604
1605         SetPageChecked(page);
1606         page_cache_get(page);
1607         fixup->work.func = btrfs_writepage_fixup_worker;
1608         fixup->page = page;
1609         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1610         return -EAGAIN;
1611 }
1612
1613 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1614                                        struct inode *inode, u64 file_pos,
1615                                        u64 disk_bytenr, u64 disk_num_bytes,
1616                                        u64 num_bytes, u64 ram_bytes,
1617                                        u8 compression, u8 encryption,
1618                                        u16 other_encoding, int extent_type)
1619 {
1620         struct btrfs_root *root = BTRFS_I(inode)->root;
1621         struct btrfs_file_extent_item *fi;
1622         struct btrfs_path *path;
1623         struct extent_buffer *leaf;
1624         struct btrfs_key ins;
1625         u64 hint;
1626         int ret;
1627
1628         path = btrfs_alloc_path();
1629         BUG_ON(!path);
1630
1631         path->leave_spinning = 1;
1632
1633         /*
1634          * we may be replacing one extent in the tree with another.
1635          * The new extent is pinned in the extent map, and we don't want
1636          * to drop it from the cache until it is completely in the btree.
1637          *
1638          * So, tell btrfs_drop_extents to leave this extent in the cache.
1639          * the caller is expected to unpin it and allow it to be merged
1640          * with the others.
1641          */
1642         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1643                                  &hint, 0);
1644         BUG_ON(ret);
1645
1646         ins.objectid = inode->i_ino;
1647         ins.offset = file_pos;
1648         ins.type = BTRFS_EXTENT_DATA_KEY;
1649         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1650         BUG_ON(ret);
1651         leaf = path->nodes[0];
1652         fi = btrfs_item_ptr(leaf, path->slots[0],
1653                             struct btrfs_file_extent_item);
1654         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1655         btrfs_set_file_extent_type(leaf, fi, extent_type);
1656         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1657         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1658         btrfs_set_file_extent_offset(leaf, fi, 0);
1659         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1660         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1661         btrfs_set_file_extent_compression(leaf, fi, compression);
1662         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1663         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1664
1665         btrfs_unlock_up_safe(path, 1);
1666         btrfs_set_lock_blocking(leaf);
1667
1668         btrfs_mark_buffer_dirty(leaf);
1669
1670         inode_add_bytes(inode, num_bytes);
1671
1672         ins.objectid = disk_bytenr;
1673         ins.offset = disk_num_bytes;
1674         ins.type = BTRFS_EXTENT_ITEM_KEY;
1675         ret = btrfs_alloc_reserved_file_extent(trans, root,
1676                                         root->root_key.objectid,
1677                                         inode->i_ino, file_pos, &ins);
1678         BUG_ON(ret);
1679         btrfs_free_path(path);
1680
1681         return 0;
1682 }
1683
1684 /*
1685  * helper function for btrfs_finish_ordered_io, this
1686  * just reads in some of the csum leaves to prime them into ram
1687  * before we start the transaction.  It limits the amount of btree
1688  * reads required while inside the transaction.
1689  */
1690 /* as ordered data IO finishes, this gets called so we can finish
1691  * an ordered extent if the range of bytes in the file it covers are
1692  * fully written.
1693  */
1694 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1695 {
1696         struct btrfs_root *root = BTRFS_I(inode)->root;
1697         struct btrfs_trans_handle *trans = NULL;
1698         struct btrfs_ordered_extent *ordered_extent = NULL;
1699         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1700         struct extent_state *cached_state = NULL;
1701         int compress_type = 0;
1702         int ret;
1703         bool nolock = false;
1704
1705         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1706                                              end - start + 1);
1707         if (!ret)
1708                 return 0;
1709         BUG_ON(!ordered_extent);
1710
1711         nolock = (root == root->fs_info->tree_root);
1712
1713         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1714                 BUG_ON(!list_empty(&ordered_extent->list));
1715                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1716                 if (!ret) {
1717                         if (nolock)
1718                                 trans = btrfs_join_transaction_nolock(root, 1);
1719                         else
1720                                 trans = btrfs_join_transaction(root, 1);
1721                         BUG_ON(IS_ERR(trans));
1722                         btrfs_set_trans_block_group(trans, inode);
1723                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1724                         ret = btrfs_update_inode(trans, root, inode);
1725                         BUG_ON(ret);
1726                 }
1727                 goto out;
1728         }
1729
1730         lock_extent_bits(io_tree, ordered_extent->file_offset,
1731                          ordered_extent->file_offset + ordered_extent->len - 1,
1732                          0, &cached_state, GFP_NOFS);
1733
1734         if (nolock)
1735                 trans = btrfs_join_transaction_nolock(root, 1);
1736         else
1737                 trans = btrfs_join_transaction(root, 1);
1738         BUG_ON(IS_ERR(trans));
1739         btrfs_set_trans_block_group(trans, inode);
1740         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1741
1742         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1743                 compress_type = ordered_extent->compress_type;
1744         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1745                 BUG_ON(compress_type);
1746                 ret = btrfs_mark_extent_written(trans, inode,
1747                                                 ordered_extent->file_offset,
1748                                                 ordered_extent->file_offset +
1749                                                 ordered_extent->len);
1750                 BUG_ON(ret);
1751         } else {
1752                 BUG_ON(root == root->fs_info->tree_root);
1753                 ret = insert_reserved_file_extent(trans, inode,
1754                                                 ordered_extent->file_offset,
1755                                                 ordered_extent->start,
1756                                                 ordered_extent->disk_len,
1757                                                 ordered_extent->len,
1758                                                 ordered_extent->len,
1759                                                 compress_type, 0, 0,
1760                                                 BTRFS_FILE_EXTENT_REG);
1761                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1762                                    ordered_extent->file_offset,
1763                                    ordered_extent->len);
1764                 BUG_ON(ret);
1765         }
1766         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1767                              ordered_extent->file_offset +
1768                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1769
1770         add_pending_csums(trans, inode, ordered_extent->file_offset,
1771                           &ordered_extent->list);
1772
1773         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1774         ret = btrfs_update_inode(trans, root, inode);
1775         BUG_ON(ret);
1776 out:
1777         if (nolock) {
1778                 if (trans)
1779                         btrfs_end_transaction_nolock(trans, root);
1780         } else {
1781                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1782                 if (trans)
1783                         btrfs_end_transaction(trans, root);
1784         }
1785
1786         /* once for us */
1787         btrfs_put_ordered_extent(ordered_extent);
1788         /* once for the tree */
1789         btrfs_put_ordered_extent(ordered_extent);
1790
1791         return 0;
1792 }
1793
1794 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1795                                 struct extent_state *state, int uptodate)
1796 {
1797         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1798
1799         ClearPagePrivate2(page);
1800         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1801 }
1802
1803 /*
1804  * When IO fails, either with EIO or csum verification fails, we
1805  * try other mirrors that might have a good copy of the data.  This
1806  * io_failure_record is used to record state as we go through all the
1807  * mirrors.  If another mirror has good data, the page is set up to date
1808  * and things continue.  If a good mirror can't be found, the original
1809  * bio end_io callback is called to indicate things have failed.
1810  */
1811 struct io_failure_record {
1812         struct page *page;
1813         u64 start;
1814         u64 len;
1815         u64 logical;
1816         unsigned long bio_flags;
1817         int last_mirror;
1818 };
1819
1820 static int btrfs_io_failed_hook(struct bio *failed_bio,
1821                          struct page *page, u64 start, u64 end,
1822                          struct extent_state *state)
1823 {
1824         struct io_failure_record *failrec = NULL;
1825         u64 private;
1826         struct extent_map *em;
1827         struct inode *inode = page->mapping->host;
1828         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1829         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1830         struct bio *bio;
1831         int num_copies;
1832         int ret;
1833         int rw;
1834         u64 logical;
1835
1836         ret = get_state_private(failure_tree, start, &private);
1837         if (ret) {
1838                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1839                 if (!failrec)
1840                         return -ENOMEM;
1841                 failrec->start = start;
1842                 failrec->len = end - start + 1;
1843                 failrec->last_mirror = 0;
1844                 failrec->bio_flags = 0;
1845
1846                 read_lock(&em_tree->lock);
1847                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1848                 if (em->start > start || em->start + em->len < start) {
1849                         free_extent_map(em);
1850                         em = NULL;
1851                 }
1852                 read_unlock(&em_tree->lock);
1853
1854                 if (!em || IS_ERR(em)) {
1855                         kfree(failrec);
1856                         return -EIO;
1857                 }
1858                 logical = start - em->start;
1859                 logical = em->block_start + logical;
1860                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1861                         logical = em->block_start;
1862                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1863                         extent_set_compress_type(&failrec->bio_flags,
1864                                                  em->compress_type);
1865                 }
1866                 failrec->logical = logical;
1867                 free_extent_map(em);
1868                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1869                                 EXTENT_DIRTY, GFP_NOFS);
1870                 set_state_private(failure_tree, start,
1871                                  (u64)(unsigned long)failrec);
1872         } else {
1873                 failrec = (struct io_failure_record *)(unsigned long)private;
1874         }
1875         num_copies = btrfs_num_copies(
1876                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1877                               failrec->logical, failrec->len);
1878         failrec->last_mirror++;
1879         if (!state) {
1880                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1881                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1882                                                     failrec->start,
1883                                                     EXTENT_LOCKED);
1884                 if (state && state->start != failrec->start)
1885                         state = NULL;
1886                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1887         }
1888         if (!state || failrec->last_mirror > num_copies) {
1889                 set_state_private(failure_tree, failrec->start, 0);
1890                 clear_extent_bits(failure_tree, failrec->start,
1891                                   failrec->start + failrec->len - 1,
1892                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1893                 kfree(failrec);
1894                 return -EIO;
1895         }
1896         bio = bio_alloc(GFP_NOFS, 1);
1897         bio->bi_private = state;
1898         bio->bi_end_io = failed_bio->bi_end_io;
1899         bio->bi_sector = failrec->logical >> 9;
1900         bio->bi_bdev = failed_bio->bi_bdev;
1901         bio->bi_size = 0;
1902
1903         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1904         if (failed_bio->bi_rw & REQ_WRITE)
1905                 rw = WRITE;
1906         else
1907                 rw = READ;
1908
1909         ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1910                                                       failrec->last_mirror,
1911                                                       failrec->bio_flags, 0);
1912         return ret;
1913 }
1914
1915 /*
1916  * each time an IO finishes, we do a fast check in the IO failure tree
1917  * to see if we need to process or clean up an io_failure_record
1918  */
1919 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1920 {
1921         u64 private;
1922         u64 private_failure;
1923         struct io_failure_record *failure;
1924         int ret;
1925
1926         private = 0;
1927         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1928                              (u64)-1, 1, EXTENT_DIRTY, 0)) {
1929                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1930                                         start, &private_failure);
1931                 if (ret == 0) {
1932                         failure = (struct io_failure_record *)(unsigned long)
1933                                    private_failure;
1934                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1935                                           failure->start, 0);
1936                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1937                                           failure->start,
1938                                           failure->start + failure->len - 1,
1939                                           EXTENT_DIRTY | EXTENT_LOCKED,
1940                                           GFP_NOFS);
1941                         kfree(failure);
1942                 }
1943         }
1944         return 0;
1945 }
1946
1947 /*
1948  * when reads are done, we need to check csums to verify the data is correct
1949  * if there's a match, we allow the bio to finish.  If not, we go through
1950  * the io_failure_record routines to find good copies
1951  */
1952 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1953                                struct extent_state *state)
1954 {
1955         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1956         struct inode *inode = page->mapping->host;
1957         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1958         char *kaddr;
1959         u64 private = ~(u32)0;
1960         int ret;
1961         struct btrfs_root *root = BTRFS_I(inode)->root;
1962         u32 csum = ~(u32)0;
1963
1964         if (PageChecked(page)) {
1965                 ClearPageChecked(page);
1966                 goto good;
1967         }
1968
1969         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1970                 return 0;
1971
1972         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1973             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1974                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1975                                   GFP_NOFS);
1976                 return 0;
1977         }
1978
1979         if (state && state->start == start) {
1980                 private = state->private;
1981                 ret = 0;
1982         } else {
1983                 ret = get_state_private(io_tree, start, &private);
1984         }
1985         kaddr = kmap_atomic(page, KM_USER0);
1986         if (ret)
1987                 goto zeroit;
1988
1989         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1990         btrfs_csum_final(csum, (char *)&csum);
1991         if (csum != private)
1992                 goto zeroit;
1993
1994         kunmap_atomic(kaddr, KM_USER0);
1995 good:
1996         /* if the io failure tree for this inode is non-empty,
1997          * check to see if we've recovered from a failed IO
1998          */
1999         btrfs_clean_io_failures(inode, start);
2000         return 0;
2001
2002 zeroit:
2003         if (printk_ratelimit()) {
2004                 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
2005                        "private %llu\n", page->mapping->host->i_ino,
2006                        (unsigned long long)start, csum,
2007                        (unsigned long long)private);
2008         }
2009         memset(kaddr + offset, 1, end - start + 1);
2010         flush_dcache_page(page);
2011         kunmap_atomic(kaddr, KM_USER0);
2012         if (private == 0)
2013                 return 0;
2014         return -EIO;
2015 }
2016
2017 struct delayed_iput {
2018         struct list_head list;
2019         struct inode *inode;
2020 };
2021
2022 void btrfs_add_delayed_iput(struct inode *inode)
2023 {
2024         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2025         struct delayed_iput *delayed;
2026
2027         if (atomic_add_unless(&inode->i_count, -1, 1))
2028                 return;
2029
2030         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2031         delayed->inode = inode;
2032
2033         spin_lock(&fs_info->delayed_iput_lock);
2034         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2035         spin_unlock(&fs_info->delayed_iput_lock);
2036 }
2037
2038 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2039 {
2040         LIST_HEAD(list);
2041         struct btrfs_fs_info *fs_info = root->fs_info;
2042         struct delayed_iput *delayed;
2043         int empty;
2044
2045         spin_lock(&fs_info->delayed_iput_lock);
2046         empty = list_empty(&fs_info->delayed_iputs);
2047         spin_unlock(&fs_info->delayed_iput_lock);
2048         if (empty)
2049                 return;
2050
2051         down_read(&root->fs_info->cleanup_work_sem);
2052         spin_lock(&fs_info->delayed_iput_lock);
2053         list_splice_init(&fs_info->delayed_iputs, &list);
2054         spin_unlock(&fs_info->delayed_iput_lock);
2055
2056         while (!list_empty(&list)) {
2057                 delayed = list_entry(list.next, struct delayed_iput, list);
2058                 list_del(&delayed->list);
2059                 iput(delayed->inode);
2060                 kfree(delayed);
2061         }
2062         up_read(&root->fs_info->cleanup_work_sem);
2063 }
2064
2065 /*
2066  * calculate extra metadata reservation when snapshotting a subvolume
2067  * contains orphan files.
2068  */
2069 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2070                                 struct btrfs_pending_snapshot *pending,
2071                                 u64 *bytes_to_reserve)
2072 {
2073         struct btrfs_root *root;
2074         struct btrfs_block_rsv *block_rsv;
2075         u64 num_bytes;
2076         int index;
2077
2078         root = pending->root;
2079         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2080                 return;
2081
2082         block_rsv = root->orphan_block_rsv;
2083
2084         /* orphan block reservation for the snapshot */
2085         num_bytes = block_rsv->size;
2086
2087         /*
2088          * after the snapshot is created, COWing tree blocks may use more
2089          * space than it frees. So we should make sure there is enough
2090          * reserved space.
2091          */
2092         index = trans->transid & 0x1;
2093         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2094                 num_bytes += block_rsv->size -
2095                              (block_rsv->reserved + block_rsv->freed[index]);
2096         }
2097
2098         *bytes_to_reserve += num_bytes;
2099 }
2100
2101 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2102                                 struct btrfs_pending_snapshot *pending)
2103 {
2104         struct btrfs_root *root = pending->root;
2105         struct btrfs_root *snap = pending->snap;
2106         struct btrfs_block_rsv *block_rsv;
2107         u64 num_bytes;
2108         int index;
2109         int ret;
2110
2111         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2112                 return;
2113
2114         /* refill source subvolume's orphan block reservation */
2115         block_rsv = root->orphan_block_rsv;
2116         index = trans->transid & 0x1;
2117         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2118                 num_bytes = block_rsv->size -
2119                             (block_rsv->reserved + block_rsv->freed[index]);
2120                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2121                                               root->orphan_block_rsv,
2122                                               num_bytes);
2123                 BUG_ON(ret);
2124         }
2125
2126         /* setup orphan block reservation for the snapshot */
2127         block_rsv = btrfs_alloc_block_rsv(snap);
2128         BUG_ON(!block_rsv);
2129
2130         btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2131         snap->orphan_block_rsv = block_rsv;
2132
2133         num_bytes = root->orphan_block_rsv->size;
2134         ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2135                                       block_rsv, num_bytes);
2136         BUG_ON(ret);
2137
2138 #if 0
2139         /* insert orphan item for the snapshot */
2140         WARN_ON(!root->orphan_item_inserted);
2141         ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2142                                        snap->root_key.objectid);
2143         BUG_ON(ret);
2144         snap->orphan_item_inserted = 1;
2145 #endif
2146 }
2147
2148 enum btrfs_orphan_cleanup_state {
2149         ORPHAN_CLEANUP_STARTED  = 1,
2150         ORPHAN_CLEANUP_DONE     = 2,
2151 };
2152
2153 /*
2154  * This is called in transaction commmit time. If there are no orphan
2155  * files in the subvolume, it removes orphan item and frees block_rsv
2156  * structure.
2157  */
2158 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2159                               struct btrfs_root *root)
2160 {
2161         int ret;
2162
2163         if (!list_empty(&root->orphan_list) ||
2164             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2165                 return;
2166
2167         if (root->orphan_item_inserted &&
2168             btrfs_root_refs(&root->root_item) > 0) {
2169                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2170                                             root->root_key.objectid);
2171                 BUG_ON(ret);
2172                 root->orphan_item_inserted = 0;
2173         }
2174
2175         if (root->orphan_block_rsv) {
2176                 WARN_ON(root->orphan_block_rsv->size > 0);
2177                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2178                 root->orphan_block_rsv = NULL;
2179         }
2180 }
2181
2182 /*
2183  * This creates an orphan entry for the given inode in case something goes
2184  * wrong in the middle of an unlink/truncate.
2185  *
2186  * NOTE: caller of this function should reserve 5 units of metadata for
2187  *       this function.
2188  */
2189 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2190 {
2191         struct btrfs_root *root = BTRFS_I(inode)->root;
2192         struct btrfs_block_rsv *block_rsv = NULL;
2193         int reserve = 0;
2194         int insert = 0;
2195         int ret;
2196
2197         if (!root->orphan_block_rsv) {
2198                 block_rsv = btrfs_alloc_block_rsv(root);
2199                 BUG_ON(!block_rsv);
2200         }
2201
2202         spin_lock(&root->orphan_lock);
2203         if (!root->orphan_block_rsv) {
2204                 root->orphan_block_rsv = block_rsv;
2205         } else if (block_rsv) {
2206                 btrfs_free_block_rsv(root, block_rsv);
2207                 block_rsv = NULL;
2208         }
2209
2210         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2211                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2212 #if 0
2213                 /*
2214                  * For proper ENOSPC handling, we should do orphan
2215                  * cleanup when mounting. But this introduces backward
2216                  * compatibility issue.
2217                  */
2218                 if (!xchg(&root->orphan_item_inserted, 1))
2219                         insert = 2;
2220                 else
2221                         insert = 1;
2222 #endif
2223                 insert = 1;
2224         }
2225
2226         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2227                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2228                 reserve = 1;
2229         }
2230         spin_unlock(&root->orphan_lock);
2231
2232         if (block_rsv)
2233                 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2234
2235         /* grab metadata reservation from transaction handle */
2236         if (reserve) {
2237                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2238                 BUG_ON(ret);
2239         }
2240
2241         /* insert an orphan item to track this unlinked/truncated file */
2242         if (insert >= 1) {
2243                 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2244                 BUG_ON(ret);
2245         }
2246
2247         /* insert an orphan item to track subvolume contains orphan files */
2248         if (insert >= 2) {
2249                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2250                                                root->root_key.objectid);
2251                 BUG_ON(ret);
2252         }
2253         return 0;
2254 }
2255
2256 /*
2257  * We have done the truncate/delete so we can go ahead and remove the orphan
2258  * item for this particular inode.
2259  */
2260 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2261 {
2262         struct btrfs_root *root = BTRFS_I(inode)->root;
2263         int delete_item = 0;
2264         int release_rsv = 0;
2265         int ret = 0;
2266
2267         spin_lock(&root->orphan_lock);
2268         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2269                 list_del_init(&BTRFS_I(inode)->i_orphan);
2270                 delete_item = 1;
2271         }
2272
2273         if (BTRFS_I(inode)->orphan_meta_reserved) {
2274                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2275                 release_rsv = 1;
2276         }
2277         spin_unlock(&root->orphan_lock);
2278
2279         if (trans && delete_item) {
2280                 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2281                 BUG_ON(ret);
2282         }
2283
2284         if (release_rsv)
2285                 btrfs_orphan_release_metadata(inode);
2286
2287         return 0;
2288 }
2289
2290 /*
2291  * this cleans up any orphans that may be left on the list from the last use
2292  * of this root.
2293  */
2294 int btrfs_orphan_cleanup(struct btrfs_root *root)
2295 {
2296         struct btrfs_path *path;
2297         struct extent_buffer *leaf;
2298         struct btrfs_key key, found_key;
2299         struct btrfs_trans_handle *trans;
2300         struct inode *inode;
2301         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2302
2303         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2304                 return 0;
2305
2306         path = btrfs_alloc_path();
2307         if (!path) {
2308                 ret = -ENOMEM;
2309                 goto out;
2310         }
2311         path->reada = -1;
2312
2313         key.objectid = BTRFS_ORPHAN_OBJECTID;
2314         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2315         key.offset = (u64)-1;
2316
2317         while (1) {
2318                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2319                 if (ret < 0)
2320                         goto out;
2321
2322                 /*
2323                  * if ret == 0 means we found what we were searching for, which
2324                  * is weird, but possible, so only screw with path if we didnt
2325                  * find the key and see if we have stuff that matches
2326                  */
2327                 if (ret > 0) {
2328                         ret = 0;
2329                         if (path->slots[0] == 0)
2330                                 break;
2331                         path->slots[0]--;
2332                 }
2333
2334                 /* pull out the item */
2335                 leaf = path->nodes[0];
2336                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2337
2338                 /* make sure the item matches what we want */
2339                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2340                         break;
2341                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2342                         break;
2343
2344                 /* release the path since we're done with it */
2345                 btrfs_release_path(root, path);
2346
2347                 /*
2348                  * this is where we are basically btrfs_lookup, without the
2349                  * crossing root thing.  we store the inode number in the
2350                  * offset of the orphan item.
2351                  */
2352                 found_key.objectid = found_key.offset;
2353                 found_key.type = BTRFS_INODE_ITEM_KEY;
2354                 found_key.offset = 0;
2355                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2356                 if (IS_ERR(inode)) {
2357                         ret = PTR_ERR(inode);
2358                         goto out;
2359                 }
2360
2361                 /*
2362                  * add this inode to the orphan list so btrfs_orphan_del does
2363                  * the proper thing when we hit it
2364                  */
2365                 spin_lock(&root->orphan_lock);
2366                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2367                 spin_unlock(&root->orphan_lock);
2368
2369                 /*
2370                  * if this is a bad inode, means we actually succeeded in
2371                  * removing the inode, but not the orphan record, which means
2372                  * we need to manually delete the orphan since iput will just
2373                  * do a destroy_inode
2374                  */
2375                 if (is_bad_inode(inode)) {
2376                         trans = btrfs_start_transaction(root, 0);
2377                         if (IS_ERR(trans)) {
2378                                 ret = PTR_ERR(trans);
2379                                 goto out;
2380                         }
2381                         btrfs_orphan_del(trans, inode);
2382                         btrfs_end_transaction(trans, root);
2383                         iput(inode);
2384                         continue;
2385                 }
2386
2387                 /* if we have links, this was a truncate, lets do that */
2388                 if (inode->i_nlink) {
2389                         if (!S_ISREG(inode->i_mode)) {
2390                                 WARN_ON(1);
2391                                 iput(inode);
2392                                 continue;
2393                         }
2394                         nr_truncate++;
2395                         ret = btrfs_truncate(inode);
2396                 } else {
2397                         nr_unlink++;
2398                 }
2399
2400                 /* this will do delete_inode and everything for us */
2401                 iput(inode);
2402                 if (ret)
2403                         goto out;
2404         }
2405         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2406
2407         if (root->orphan_block_rsv)
2408                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2409                                         (u64)-1);
2410
2411         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2412                 trans = btrfs_join_transaction(root, 1);
2413                 if (!IS_ERR(trans))
2414                         btrfs_end_transaction(trans, root);
2415         }
2416
2417         if (nr_unlink)
2418                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2419         if (nr_truncate)
2420                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2421
2422 out:
2423         if (ret)
2424                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2425         btrfs_free_path(path);
2426         return ret;
2427 }
2428
2429 /*
2430  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2431  * don't find any xattrs, we know there can't be any acls.
2432  *
2433  * slot is the slot the inode is in, objectid is the objectid of the inode
2434  */
2435 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2436                                           int slot, u64 objectid)
2437 {
2438         u32 nritems = btrfs_header_nritems(leaf);
2439         struct btrfs_key found_key;
2440         int scanned = 0;
2441
2442         slot++;
2443         while (slot < nritems) {
2444                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2445
2446                 /* we found a different objectid, there must not be acls */
2447                 if (found_key.objectid != objectid)
2448                         return 0;
2449
2450                 /* we found an xattr, assume we've got an acl */
2451                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2452                         return 1;
2453
2454                 /*
2455                  * we found a key greater than an xattr key, there can't
2456                  * be any acls later on
2457                  */
2458                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2459                         return 0;
2460
2461                 slot++;
2462                 scanned++;
2463
2464                 /*
2465                  * it goes inode, inode backrefs, xattrs, extents,
2466                  * so if there are a ton of hard links to an inode there can
2467                  * be a lot of backrefs.  Don't waste time searching too hard,
2468                  * this is just an optimization
2469                  */
2470                 if (scanned >= 8)
2471                         break;
2472         }
2473         /* we hit the end of the leaf before we found an xattr or
2474          * something larger than an xattr.  We have to assume the inode
2475          * has acls
2476          */
2477         return 1;
2478 }
2479
2480 /*
2481  * read an inode from the btree into the in-memory inode
2482  */
2483 static void btrfs_read_locked_inode(struct inode *inode)
2484 {
2485         struct btrfs_path *path;
2486         struct extent_buffer *leaf;
2487         struct btrfs_inode_item *inode_item;
2488         struct btrfs_timespec *tspec;
2489         struct btrfs_root *root = BTRFS_I(inode)->root;
2490         struct btrfs_key location;
2491         int maybe_acls;
2492         u64 alloc_group_block;
2493         u32 rdev;
2494         int ret;
2495
2496         path = btrfs_alloc_path();
2497         BUG_ON(!path);
2498         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2499
2500         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2501         if (ret)
2502                 goto make_bad;
2503
2504         leaf = path->nodes[0];
2505         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2506                                     struct btrfs_inode_item);
2507
2508         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2509         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2510         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2511         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2512         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2513
2514         tspec = btrfs_inode_atime(inode_item);
2515         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2516         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2517
2518         tspec = btrfs_inode_mtime(inode_item);
2519         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2520         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2521
2522         tspec = btrfs_inode_ctime(inode_item);
2523         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2524         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2525
2526         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2527         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2528         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2529         inode->i_generation = BTRFS_I(inode)->generation;
2530         inode->i_rdev = 0;
2531         rdev = btrfs_inode_rdev(leaf, inode_item);
2532
2533         BTRFS_I(inode)->index_cnt = (u64)-1;
2534         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2535
2536         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2537
2538         /*
2539          * try to precache a NULL acl entry for files that don't have
2540          * any xattrs or acls
2541          */
2542         maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2543         if (!maybe_acls)
2544                 cache_no_acl(inode);
2545
2546         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2547                                                 alloc_group_block, 0);
2548         btrfs_free_path(path);
2549         inode_item = NULL;
2550
2551         switch (inode->i_mode & S_IFMT) {
2552         case S_IFREG:
2553                 inode->i_mapping->a_ops = &btrfs_aops;
2554                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2555                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2556                 inode->i_fop = &btrfs_file_operations;
2557                 inode->i_op = &btrfs_file_inode_operations;
2558                 break;
2559         case S_IFDIR:
2560                 inode->i_fop = &btrfs_dir_file_operations;
2561                 if (root == root->fs_info->tree_root)
2562                         inode->i_op = &btrfs_dir_ro_inode_operations;
2563                 else
2564                         inode->i_op = &btrfs_dir_inode_operations;
2565                 break;
2566         case S_IFLNK:
2567                 inode->i_op = &btrfs_symlink_inode_operations;
2568                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2569                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2570                 break;
2571         default:
2572                 inode->i_op = &btrfs_special_inode_operations;
2573                 init_special_inode(inode, inode->i_mode, rdev);
2574                 break;
2575         }
2576
2577         btrfs_update_iflags(inode);
2578         return;
2579
2580 make_bad:
2581         btrfs_free_path(path);
2582         make_bad_inode(inode);
2583 }
2584
2585 /*
2586  * given a leaf and an inode, copy the inode fields into the leaf
2587  */
2588 static void fill_inode_item(struct btrfs_trans_handle *trans,
2589                             struct extent_buffer *leaf,
2590                             struct btrfs_inode_item *item,
2591                             struct inode *inode)
2592 {
2593         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2594         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2595         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2596         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2597         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2598
2599         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2600                                inode->i_atime.tv_sec);
2601         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2602                                 inode->i_atime.tv_nsec);
2603
2604         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2605                                inode->i_mtime.tv_sec);
2606         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2607                                 inode->i_mtime.tv_nsec);
2608
2609         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2610                                inode->i_ctime.tv_sec);
2611         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2612                                 inode->i_ctime.tv_nsec);
2613
2614         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2615         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2616         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2617         btrfs_set_inode_transid(leaf, item, trans->transid);
2618         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2619         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2620         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2621 }
2622
2623 /*
2624  * copy everything in the in-memory inode into the btree.
2625  */
2626 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2627                                 struct btrfs_root *root, struct inode *inode)
2628 {
2629         struct btrfs_inode_item *inode_item;
2630         struct btrfs_path *path;
2631         struct extent_buffer *leaf;
2632         int ret;
2633
2634         path = btrfs_alloc_path();
2635         BUG_ON(!path);
2636         path->leave_spinning = 1;
2637         ret = btrfs_lookup_inode(trans, root, path,
2638                                  &BTRFS_I(inode)->location, 1);
2639         if (ret) {
2640                 if (ret > 0)
2641                         ret = -ENOENT;
2642                 goto failed;
2643         }
2644
2645         btrfs_unlock_up_safe(path, 1);
2646         leaf = path->nodes[0];
2647         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2648                                   struct btrfs_inode_item);
2649
2650         fill_inode_item(trans, leaf, inode_item, inode);
2651         btrfs_mark_buffer_dirty(leaf);
2652         btrfs_set_inode_last_trans(trans, inode);
2653         ret = 0;
2654 failed:
2655         btrfs_free_path(path);
2656         return ret;
2657 }
2658
2659
2660 /*
2661  * unlink helper that gets used here in inode.c and in the tree logging
2662  * recovery code.  It remove a link in a directory with a given name, and
2663  * also drops the back refs in the inode to the directory
2664  */
2665 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2666                                 struct btrfs_root *root,
2667                                 struct inode *dir, struct inode *inode,
2668                                 const char *name, int name_len)
2669 {
2670         struct btrfs_path *path;
2671         int ret = 0;
2672         struct extent_buffer *leaf;
2673         struct btrfs_dir_item *di;
2674         struct btrfs_key key;
2675         u64 index;
2676
2677         path = btrfs_alloc_path();
2678         if (!path) {
2679                 ret = -ENOMEM;
2680                 goto out;
2681         }
2682
2683         path->leave_spinning = 1;
2684         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2685                                     name, name_len, -1);
2686         if (IS_ERR(di)) {
2687                 ret = PTR_ERR(di);
2688                 goto err;
2689         }
2690         if (!di) {
2691                 ret = -ENOENT;
2692                 goto err;
2693         }
2694         leaf = path->nodes[0];
2695         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2696         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2697         if (ret)
2698                 goto err;
2699         btrfs_release_path(root, path);
2700
2701         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2702                                   inode->i_ino,
2703                                   dir->i_ino, &index);
2704         if (ret) {
2705                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2706                        "inode %lu parent %lu\n", name_len, name,
2707                        inode->i_ino, dir->i_ino);
2708                 goto err;
2709         }
2710
2711         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2712                                          index, name, name_len, -1);
2713         if (IS_ERR(di)) {
2714                 ret = PTR_ERR(di);
2715                 goto err;
2716         }
2717         if (!di) {
2718                 ret = -ENOENT;
2719                 goto err;
2720         }
2721         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2722         btrfs_release_path(root, path);
2723
2724         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2725                                          inode, dir->i_ino);
2726         BUG_ON(ret != 0 && ret != -ENOENT);
2727
2728         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2729                                            dir, index);
2730         if (ret == -ENOENT)
2731                 ret = 0;
2732 err:
2733         btrfs_free_path(path);
2734         if (ret)
2735                 goto out;
2736
2737         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2738         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2739         btrfs_update_inode(trans, root, dir);
2740 out:
2741         return ret;
2742 }
2743
2744 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2745                        struct btrfs_root *root,
2746                        struct inode *dir, struct inode *inode,
2747                        const char *name, int name_len)
2748 {
2749         int ret;
2750         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2751         if (!ret) {
2752                 btrfs_drop_nlink(inode);
2753                 ret = btrfs_update_inode(trans, root, inode);
2754         }
2755         return ret;
2756 }
2757                 
2758
2759 /* helper to check if there is any shared block in the path */
2760 static int check_path_shared(struct btrfs_root *root,
2761                              struct btrfs_path *path)
2762 {
2763         struct extent_buffer *eb;
2764         int level;
2765         u64 refs = 1;
2766
2767         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2768                 int ret;
2769
2770                 if (!path->nodes[level])
2771                         break;
2772                 eb = path->nodes[level];
2773                 if (!btrfs_block_can_be_shared(root, eb))
2774                         continue;
2775                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2776                                                &refs, NULL);
2777                 if (refs > 1)
2778                         return 1;
2779         }
2780         return 0;
2781 }
2782
2783 /*
2784  * helper to start transaction for unlink and rmdir.
2785  *
2786  * unlink and rmdir are special in btrfs, they do not always free space.
2787  * so in enospc case, we should make sure they will free space before
2788  * allowing them to use the global metadata reservation.
2789  */
2790 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2791                                                        struct dentry *dentry)
2792 {
2793         struct btrfs_trans_handle *trans;
2794         struct btrfs_root *root = BTRFS_I(dir)->root;
2795         struct btrfs_path *path;
2796         struct btrfs_inode_ref *ref;
2797         struct btrfs_dir_item *di;
2798         struct inode *inode = dentry->d_inode;
2799         u64 index;
2800         int check_link = 1;
2801         int err = -ENOSPC;
2802         int ret;
2803
2804         trans = btrfs_start_transaction(root, 10);
2805         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2806                 return trans;
2807
2808         if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2809                 return ERR_PTR(-ENOSPC);
2810
2811         /* check if there is someone else holds reference */
2812         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2813                 return ERR_PTR(-ENOSPC);
2814
2815         if (atomic_read(&inode->i_count) > 2)
2816                 return ERR_PTR(-ENOSPC);
2817
2818         if (xchg(&root->fs_info->enospc_unlink, 1))
2819                 return ERR_PTR(-ENOSPC);
2820
2821         path = btrfs_alloc_path();
2822         if (!path) {
2823                 root->fs_info->enospc_unlink = 0;
2824                 return ERR_PTR(-ENOMEM);
2825         }
2826
2827         trans = btrfs_start_transaction(root, 0);
2828         if (IS_ERR(trans)) {
2829                 btrfs_free_path(path);
2830                 root->fs_info->enospc_unlink = 0;
2831                 return trans;
2832         }
2833
2834         path->skip_locking = 1;
2835         path->search_commit_root = 1;
2836
2837         ret = btrfs_lookup_inode(trans, root, path,
2838                                 &BTRFS_I(dir)->location, 0);
2839         if (ret < 0) {
2840                 err = ret;
2841                 goto out;
2842         }
2843         if (ret == 0) {
2844                 if (check_path_shared(root, path))
2845                         goto out;
2846         } else {
2847                 check_link = 0;
2848         }
2849         btrfs_release_path(root, path);
2850
2851         ret = btrfs_lookup_inode(trans, root, path,
2852                                 &BTRFS_I(inode)->location, 0);
2853         if (ret < 0) {
2854                 err = ret;
2855                 goto out;
2856         }
2857         if (ret == 0) {
2858                 if (check_path_shared(root, path))
2859                         goto out;
2860         } else {
2861                 check_link = 0;
2862         }
2863         btrfs_release_path(root, path);
2864
2865         if (ret == 0 && S_ISREG(inode->i_mode)) {
2866                 ret = btrfs_lookup_file_extent(trans, root, path,
2867                                                inode->i_ino, (u64)-1, 0);
2868                 if (ret < 0) {
2869                         err = ret;
2870                         goto out;
2871                 }
2872                 BUG_ON(ret == 0);
2873                 if (check_path_shared(root, path))
2874                         goto out;
2875                 btrfs_release_path(root, path);
2876         }
2877
2878         if (!check_link) {
2879                 err = 0;
2880                 goto out;
2881         }
2882
2883         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2884                                 dentry->d_name.name, dentry->d_name.len, 0);
2885         if (IS_ERR(di)) {
2886                 err = PTR_ERR(di);
2887                 goto out;
2888         }
2889         if (di) {
2890                 if (check_path_shared(root, path))
2891                         goto out;
2892         } else {
2893                 err = 0;
2894                 goto out;
2895         }
2896         btrfs_release_path(root, path);
2897
2898         ref = btrfs_lookup_inode_ref(trans, root, path,
2899                                 dentry->d_name.name, dentry->d_name.len,
2900                                 inode->i_ino, dir->i_ino, 0);
2901         if (IS_ERR(ref)) {
2902                 err = PTR_ERR(ref);
2903                 goto out;
2904         }
2905         BUG_ON(!ref);
2906         if (check_path_shared(root, path))
2907                 goto out;
2908         index = btrfs_inode_ref_index(path->nodes[0], ref);
2909         btrfs_release_path(root, path);
2910
2911         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2912                                 dentry->d_name.name, dentry->d_name.len, 0);
2913         if (IS_ERR(di)) {
2914                 err = PTR_ERR(di);
2915                 goto out;
2916         }
2917         BUG_ON(ret == -ENOENT);
2918         if (check_path_shared(root, path))
2919                 goto out;
2920
2921         err = 0;
2922 out:
2923         btrfs_free_path(path);
2924         if (err) {
2925                 btrfs_end_transaction(trans, root);
2926                 root->fs_info->enospc_unlink = 0;
2927                 return ERR_PTR(err);
2928         }
2929
2930         trans->block_rsv = &root->fs_info->global_block_rsv;
2931         return trans;
2932 }
2933
2934 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2935                                struct btrfs_root *root)
2936 {
2937         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2938                 BUG_ON(!root->fs_info->enospc_unlink);
2939                 root->fs_info->enospc_unlink = 0;
2940         }
2941         btrfs_end_transaction_throttle(trans, root);
2942 }
2943
2944 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2945 {
2946         struct btrfs_root *root = BTRFS_I(dir)->root;
2947         struct btrfs_trans_handle *trans;
2948         struct inode *inode = dentry->d_inode;
2949         int ret;
2950         unsigned long nr = 0;
2951
2952         trans = __unlink_start_trans(dir, dentry);
2953         if (IS_ERR(trans))
2954                 return PTR_ERR(trans);
2955
2956         btrfs_set_trans_block_group(trans, dir);
2957
2958         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2959
2960         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2961                                  dentry->d_name.name, dentry->d_name.len);
2962         BUG_ON(ret);
2963
2964         if (inode->i_nlink == 0) {
2965                 ret = btrfs_orphan_add(trans, inode);
2966                 BUG_ON(ret);
2967         }
2968
2969         nr = trans->blocks_used;
2970         __unlink_end_trans(trans, root);
2971         btrfs_btree_balance_dirty(root, nr);
2972         return ret;
2973 }
2974
2975 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2976                         struct btrfs_root *root,
2977                         struct inode *dir, u64 objectid,
2978                         const char *name, int name_len)
2979 {
2980         struct btrfs_path *path;
2981         struct extent_buffer *leaf;
2982         struct btrfs_dir_item *di;
2983         struct btrfs_key key;
2984         u64 index;
2985         int ret;
2986
2987         path = btrfs_alloc_path();
2988         if (!path)
2989                 return -ENOMEM;
2990
2991         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2992                                    name, name_len, -1);
2993         BUG_ON(!di || IS_ERR(di));
2994
2995         leaf = path->nodes[0];
2996         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2997         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2998         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2999         BUG_ON(ret);
3000         btrfs_release_path(root, path);
3001
3002         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3003                                  objectid, root->root_key.objectid,
3004                                  dir->i_ino, &index, name, name_len);
3005         if (ret < 0) {
3006                 BUG_ON(ret != -ENOENT);
3007                 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
3008                                                  name, name_len);
3009                 BUG_ON(!di || IS_ERR(di));
3010
3011                 leaf = path->nodes[0];
3012                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3013                 btrfs_release_path(root, path);
3014                 index = key.offset;
3015         }
3016
3017         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
3018                                          index, name, name_len, -1);
3019         BUG_ON(!di || IS_ERR(di));
3020
3021         leaf = path->nodes[0];
3022         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3023         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3024         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3025         BUG_ON(ret);
3026         btrfs_release_path(root, path);
3027
3028         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3029         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3030         ret = btrfs_update_inode(trans, root, dir);
3031         BUG_ON(ret);
3032
3033         btrfs_free_path(path);
3034         return 0;
3035 }
3036
3037 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3038 {
3039         struct inode *inode = dentry->d_inode;
3040         int err = 0;
3041         struct btrfs_root *root = BTRFS_I(dir)->root;
3042         struct btrfs_trans_handle *trans;
3043         unsigned long nr = 0;
3044
3045         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3046             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3047                 return -ENOTEMPTY;
3048
3049         trans = __unlink_start_trans(dir, dentry);
3050         if (IS_ERR(trans))
3051                 return PTR_ERR(trans);
3052
3053         btrfs_set_trans_block_group(trans, dir);
3054
3055         if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3056                 err = btrfs_unlink_subvol(trans, root, dir,
3057                                           BTRFS_I(inode)->location.objectid,
3058                                           dentry->d_name.name,
3059                                           dentry->d_name.len);
3060                 goto out;
3061         }
3062
3063         err = btrfs_orphan_add(trans, inode);
3064         if (err)
3065                 goto out;
3066
3067         /* now the directory is empty */
3068         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3069                                  dentry->d_name.name, dentry->d_name.len);
3070         if (!err)
3071                 btrfs_i_size_write(inode, 0);
3072 out:
3073         nr = trans->blocks_used;
3074         __unlink_end_trans(trans, root);
3075         btrfs_btree_balance_dirty(root, nr);
3076
3077         return err;
3078 }
3079
3080 #if 0
3081 /*
3082  * when truncating bytes in a file, it is possible to avoid reading
3083  * the leaves that contain only checksum items.  This can be the
3084  * majority of the IO required to delete a large file, but it must
3085  * be done carefully.
3086  *
3087  * The keys in the level just above the leaves are checked to make sure
3088  * the lowest key in a given leaf is a csum key, and starts at an offset
3089  * after the new  size.
3090  *
3091  * Then the key for the next leaf is checked to make sure it also has
3092  * a checksum item for the same file.  If it does, we know our target leaf
3093  * contains only checksum items, and it can be safely freed without reading
3094  * it.
3095  *
3096  * This is just an optimization targeted at large files.  It may do
3097  * nothing.  It will return 0 unless things went badly.
3098  */
3099 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3100                                      struct btrfs_root *root,
3101                                      struct btrfs_path *path,
3102                                      struct inode *inode, u64 new_size)
3103 {
3104         struct btrfs_key key;
3105         int ret;
3106         int nritems;
3107         struct btrfs_key found_key;
3108         struct btrfs_key other_key;
3109         struct btrfs_leaf_ref *ref;
3110         u64 leaf_gen;
3111         u64 leaf_start;
3112
3113         path->lowest_level = 1;
3114         key.objectid = inode->i_ino;
3115         key.type = BTRFS_CSUM_ITEM_KEY;
3116         key.offset = new_size;
3117 again:
3118         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3119         if (ret < 0)
3120                 goto out;
3121
3122         if (path->nodes[1] == NULL) {
3123                 ret = 0;
3124                 goto out;
3125         }
3126         ret = 0;
3127         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3128         nritems = btrfs_header_nritems(path->nodes[1]);
3129
3130         if (!nritems)
3131                 goto out;
3132
3133         if (path->slots[1] >= nritems)
3134                 goto next_node;
3135
3136         /* did we find a key greater than anything we want to delete? */
3137         if (found_key.objectid > inode->i_ino ||
3138            (found_key.objectid == inode->i_ino && found_key.type > key.type))
3139                 goto out;
3140
3141         /* we check the next key in the node to make sure the leave contains
3142          * only checksum items.  This comparison doesn't work if our
3143          * leaf is the last one in the node
3144          */
3145         if (path->slots[1] + 1 >= nritems) {
3146 next_node:
3147                 /* search forward from the last key in