Btrfs: cleanup error handling in inode.c
[linux-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 {
98         int err;
99
100         err = btrfs_init_acl(trans, inode, dir);
101         if (!err)
102                 err = btrfs_xattr_security_init(trans, inode, dir);
103         return err;
104 }
105
106 /*
107  * this does all the hard work for inserting an inline extent into
108  * the btree.  The caller should have done a btrfs_drop_extents so that
109  * no overlapping inline items exist in the btree
110  */
111 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
112                                 struct btrfs_root *root, struct inode *inode,
113                                 u64 start, size_t size, size_t compressed_size,
114                                 int compress_type,
115                                 struct page **compressed_pages)
116 {
117         struct btrfs_key key;
118         struct btrfs_path *path;
119         struct extent_buffer *leaf;
120         struct page *page = NULL;
121         char *kaddr;
122         unsigned long ptr;
123         struct btrfs_file_extent_item *ei;
124         int err = 0;
125         int ret;
126         size_t cur_size = size;
127         size_t datasize;
128         unsigned long offset;
129
130         if (compressed_size && compressed_pages)
131                 cur_size = compressed_size;
132
133         path = btrfs_alloc_path();
134         if (!path)
135                 return -ENOMEM;
136
137         path->leave_spinning = 1;
138         btrfs_set_trans_block_group(trans, inode);
139
140         key.objectid = inode->i_ino;
141         key.offset = start;
142         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
143         datasize = btrfs_file_extent_calc_inline_size(cur_size);
144
145         inode_add_bytes(inode, size);
146         ret = btrfs_insert_empty_item(trans, root, path, &key,
147                                       datasize);
148         BUG_ON(ret);
149         if (ret) {
150                 err = ret;
151                 goto fail;
152         }
153         leaf = path->nodes[0];
154         ei = btrfs_item_ptr(leaf, path->slots[0],
155                             struct btrfs_file_extent_item);
156         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
157         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
158         btrfs_set_file_extent_encryption(leaf, ei, 0);
159         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
160         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
161         ptr = btrfs_file_extent_inline_start(ei);
162
163         if (compress_type != BTRFS_COMPRESS_NONE) {
164                 struct page *cpage;
165                 int i = 0;
166                 while (compressed_size > 0) {
167                         cpage = compressed_pages[i];
168                         cur_size = min_t(unsigned long, compressed_size,
169                                        PAGE_CACHE_SIZE);
170
171                         kaddr = kmap_atomic(cpage, KM_USER0);
172                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
173                         kunmap_atomic(kaddr, KM_USER0);
174
175                         i++;
176                         ptr += cur_size;
177                         compressed_size -= cur_size;
178                 }
179                 btrfs_set_file_extent_compression(leaf, ei,
180                                                   compress_type);
181         } else {
182                 page = find_get_page(inode->i_mapping,
183                                      start >> PAGE_CACHE_SHIFT);
184                 btrfs_set_file_extent_compression(leaf, ei, 0);
185                 kaddr = kmap_atomic(page, KM_USER0);
186                 offset = start & (PAGE_CACHE_SIZE - 1);
187                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
188                 kunmap_atomic(kaddr, KM_USER0);
189                 page_cache_release(page);
190         }
191         btrfs_mark_buffer_dirty(leaf);
192         btrfs_free_path(path);
193
194         /*
195          * we're an inline extent, so nobody can
196          * extend the file past i_size without locking
197          * a page we already have locked.
198          *
199          * We must do any isize and inode updates
200          * before we unlock the pages.  Otherwise we
201          * could end up racing with unlink.
202          */
203         BTRFS_I(inode)->disk_i_size = inode->i_size;
204         btrfs_update_inode(trans, root, inode);
205
206         return 0;
207 fail:
208         btrfs_free_path(path);
209         return err;
210 }
211
212
213 /*
214  * conditionally insert an inline extent into the file.  This
215  * does the checks required to make sure the data is small enough
216  * to fit as an inline extent.
217  */
218 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
219                                  struct btrfs_root *root,
220                                  struct inode *inode, u64 start, u64 end,
221                                  size_t compressed_size, int compress_type,
222                                  struct page **compressed_pages)
223 {
224         u64 isize = i_size_read(inode);
225         u64 actual_end = min(end + 1, isize);
226         u64 inline_len = actual_end - start;
227         u64 aligned_end = (end + root->sectorsize - 1) &
228                         ~((u64)root->sectorsize - 1);
229         u64 hint_byte;
230         u64 data_len = inline_len;
231         int ret;
232
233         if (compressed_size)
234                 data_len = compressed_size;
235
236         if (start > 0 ||
237             actual_end >= PAGE_CACHE_SIZE ||
238             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
239             (!compressed_size &&
240             (actual_end & (root->sectorsize - 1)) == 0) ||
241             end + 1 < isize ||
242             data_len > root->fs_info->max_inline) {
243                 return 1;
244         }
245
246         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
247                                  &hint_byte, 1);
248         BUG_ON(ret);
249
250         if (isize > actual_end)
251                 inline_len = min_t(u64, isize, actual_end);
252         ret = insert_inline_extent(trans, root, inode, start,
253                                    inline_len, compressed_size,
254                                    compress_type, compressed_pages);
255         BUG_ON(ret);
256         btrfs_delalloc_release_metadata(inode, end + 1 - start);
257         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
258         return 0;
259 }
260
261 struct async_extent {
262         u64 start;
263         u64 ram_size;
264         u64 compressed_size;
265         struct page **pages;
266         unsigned long nr_pages;
267         int compress_type;
268         struct list_head list;
269 };
270
271 struct async_cow {
272         struct inode *inode;
273         struct btrfs_root *root;
274         struct page *locked_page;
275         u64 start;
276         u64 end;
277         struct list_head extents;
278         struct btrfs_work work;
279 };
280
281 static noinline int add_async_extent(struct async_cow *cow,
282                                      u64 start, u64 ram_size,
283                                      u64 compressed_size,
284                                      struct page **pages,
285                                      unsigned long nr_pages,
286                                      int compress_type)
287 {
288         struct async_extent *async_extent;
289
290         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
291         BUG_ON(!async_extent);
292         async_extent->start = start;
293         async_extent->ram_size = ram_size;
294         async_extent->compressed_size = compressed_size;
295         async_extent->pages = pages;
296         async_extent->nr_pages = nr_pages;
297         async_extent->compress_type = compress_type;
298         list_add_tail(&async_extent->list, &cow->extents);
299         return 0;
300 }
301
302 /*
303  * we create compressed extents in two phases.  The first
304  * phase compresses a range of pages that have already been
305  * locked (both pages and state bits are locked).
306  *
307  * This is done inside an ordered work queue, and the compression
308  * is spread across many cpus.  The actual IO submission is step
309  * two, and the ordered work queue takes care of making sure that
310  * happens in the same order things were put onto the queue by
311  * writepages and friends.
312  *
313  * If this code finds it can't get good compression, it puts an
314  * entry onto the work queue to write the uncompressed bytes.  This
315  * makes sure that both compressed inodes and uncompressed inodes
316  * are written in the same order that pdflush sent them down.
317  */
318 static noinline int compress_file_range(struct inode *inode,
319                                         struct page *locked_page,
320                                         u64 start, u64 end,
321                                         struct async_cow *async_cow,
322                                         int *num_added)
323 {
324         struct btrfs_root *root = BTRFS_I(inode)->root;
325         struct btrfs_trans_handle *trans;
326         u64 num_bytes;
327         u64 blocksize = root->sectorsize;
328         u64 actual_end;
329         u64 isize = i_size_read(inode);
330         int ret = 0;
331         struct page **pages = NULL;
332         unsigned long nr_pages;
333         unsigned long nr_pages_ret = 0;
334         unsigned long total_compressed = 0;
335         unsigned long total_in = 0;
336         unsigned long max_compressed = 128 * 1024;
337         unsigned long max_uncompressed = 128 * 1024;
338         int i;
339         int will_compress;
340         int compress_type = root->fs_info->compress_type;
341
342         actual_end = min_t(u64, isize, end + 1);
343 again:
344         will_compress = 0;
345         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
346         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
347
348         /*
349          * we don't want to send crud past the end of i_size through
350          * compression, that's just a waste of CPU time.  So, if the
351          * end of the file is before the start of our current
352          * requested range of bytes, we bail out to the uncompressed
353          * cleanup code that can deal with all of this.
354          *
355          * It isn't really the fastest way to fix things, but this is a
356          * very uncommon corner.
357          */
358         if (actual_end <= start)
359                 goto cleanup_and_bail_uncompressed;
360
361         total_compressed = actual_end - start;
362
363         /* we want to make sure that amount of ram required to uncompress
364          * an extent is reasonable, so we limit the total size in ram
365          * of a compressed extent to 128k.  This is a crucial number
366          * because it also controls how easily we can spread reads across
367          * cpus for decompression.
368          *
369          * We also want to make sure the amount of IO required to do
370          * a random read is reasonably small, so we limit the size of
371          * a compressed extent to 128k.
372          */
373         total_compressed = min(total_compressed, max_uncompressed);
374         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
375         num_bytes = max(blocksize,  num_bytes);
376         total_in = 0;
377         ret = 0;
378
379         /*
380          * we do compression for mount -o compress and when the
381          * inode has not been flagged as nocompress.  This flag can
382          * change at any time if we discover bad compression ratios.
383          */
384         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
385             (btrfs_test_opt(root, COMPRESS) ||
386              (BTRFS_I(inode)->force_compress) ||
387              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
388                 WARN_ON(pages);
389                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
390                 BUG_ON(!pages);
391
392                 if (BTRFS_I(inode)->force_compress)
393                         compress_type = BTRFS_I(inode)->force_compress;
394
395                 ret = btrfs_compress_pages(compress_type,
396                                            inode->i_mapping, start,
397                                            total_compressed, pages,
398                                            nr_pages, &nr_pages_ret,
399                                            &total_in,
400                                            &total_compressed,
401                                            max_compressed);
402
403                 if (!ret) {
404                         unsigned long offset = total_compressed &
405                                 (PAGE_CACHE_SIZE - 1);
406                         struct page *page = pages[nr_pages_ret - 1];
407                         char *kaddr;
408
409                         /* zero the tail end of the last page, we might be
410                          * sending it down to disk
411                          */
412                         if (offset) {
413                                 kaddr = kmap_atomic(page, KM_USER0);
414                                 memset(kaddr + offset, 0,
415                                        PAGE_CACHE_SIZE - offset);
416                                 kunmap_atomic(kaddr, KM_USER0);
417                         }
418                         will_compress = 1;
419                 }
420         }
421         if (start == 0) {
422                 trans = btrfs_join_transaction(root, 1);
423                 BUG_ON(IS_ERR(trans));
424                 btrfs_set_trans_block_group(trans, inode);
425                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
426
427                 /* lets try to make an inline extent */
428                 if (ret || total_in < (actual_end - start)) {
429                         /* we didn't compress the entire range, try
430                          * to make an uncompressed inline extent.
431                          */
432                         ret = cow_file_range_inline(trans, root, inode,
433                                                     start, end, 0, 0, NULL);
434                 } else {
435                         /* try making a compressed inline extent */
436                         ret = cow_file_range_inline(trans, root, inode,
437                                                     start, end,
438                                                     total_compressed,
439                                                     compress_type, pages);
440                 }
441                 if (ret == 0) {
442                         /*
443                          * inline extent creation worked, we don't need
444                          * to create any more async work items.  Unlock
445                          * and free up our temp pages.
446                          */
447                         extent_clear_unlock_delalloc(inode,
448                              &BTRFS_I(inode)->io_tree,
449                              start, end, NULL,
450                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
451                              EXTENT_CLEAR_DELALLOC |
452                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
453
454                         btrfs_end_transaction(trans, root);
455                         goto free_pages_out;
456                 }
457                 btrfs_end_transaction(trans, root);
458         }
459
460         if (will_compress) {
461                 /*
462                  * we aren't doing an inline extent round the compressed size
463                  * up to a block size boundary so the allocator does sane
464                  * things
465                  */
466                 total_compressed = (total_compressed + blocksize - 1) &
467                         ~(blocksize - 1);
468
469                 /*
470                  * one last check to make sure the compression is really a
471                  * win, compare the page count read with the blocks on disk
472                  */
473                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
474                         ~(PAGE_CACHE_SIZE - 1);
475                 if (total_compressed >= total_in) {
476                         will_compress = 0;
477                 } else {
478                         num_bytes = total_in;
479                 }
480         }
481         if (!will_compress && pages) {
482                 /*
483                  * the compression code ran but failed to make things smaller,
484                  * free any pages it allocated and our page pointer array
485                  */
486                 for (i = 0; i < nr_pages_ret; i++) {
487                         WARN_ON(pages[i]->mapping);
488                         page_cache_release(pages[i]);
489                 }
490                 kfree(pages);
491                 pages = NULL;
492                 total_compressed = 0;
493                 nr_pages_ret = 0;
494
495                 /* flag the file so we don't compress in the future */
496                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
497                     !(BTRFS_I(inode)->force_compress)) {
498                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
499                 }
500         }
501         if (will_compress) {
502                 *num_added += 1;
503
504                 /* the async work queues will take care of doing actual
505                  * allocation on disk for these compressed pages,
506                  * and will submit them to the elevator.
507                  */
508                 add_async_extent(async_cow, start, num_bytes,
509                                  total_compressed, pages, nr_pages_ret,
510                                  compress_type);
511
512                 if (start + num_bytes < end) {
513                         start += num_bytes;
514                         pages = NULL;
515                         cond_resched();
516                         goto again;
517                 }
518         } else {
519 cleanup_and_bail_uncompressed:
520                 /*
521                  * No compression, but we still need to write the pages in
522                  * the file we've been given so far.  redirty the locked
523                  * page if it corresponds to our extent and set things up
524                  * for the async work queue to run cow_file_range to do
525                  * the normal delalloc dance
526                  */
527                 if (page_offset(locked_page) >= start &&
528                     page_offset(locked_page) <= end) {
529                         __set_page_dirty_nobuffers(locked_page);
530                         /* unlocked later on in the async handlers */
531                 }
532                 add_async_extent(async_cow, start, end - start + 1,
533                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
534                 *num_added += 1;
535         }
536
537 out:
538         return 0;
539
540 free_pages_out:
541         for (i = 0; i < nr_pages_ret; i++) {
542                 WARN_ON(pages[i]->mapping);
543                 page_cache_release(pages[i]);
544         }
545         kfree(pages);
546
547         goto out;
548 }
549
550 /*
551  * phase two of compressed writeback.  This is the ordered portion
552  * of the code, which only gets called in the order the work was
553  * queued.  We walk all the async extents created by compress_file_range
554  * and send them down to the disk.
555  */
556 static noinline int submit_compressed_extents(struct inode *inode,
557                                               struct async_cow *async_cow)
558 {
559         struct async_extent *async_extent;
560         u64 alloc_hint = 0;
561         struct btrfs_trans_handle *trans;
562         struct btrfs_key ins;
563         struct extent_map *em;
564         struct btrfs_root *root = BTRFS_I(inode)->root;
565         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
566         struct extent_io_tree *io_tree;
567         int ret = 0;
568
569         if (list_empty(&async_cow->extents))
570                 return 0;
571
572
573         while (!list_empty(&async_cow->extents)) {
574                 async_extent = list_entry(async_cow->extents.next,
575                                           struct async_extent, list);
576                 list_del(&async_extent->list);
577
578                 io_tree = &BTRFS_I(inode)->io_tree;
579
580 retry:
581                 /* did the compression code fall back to uncompressed IO? */
582                 if (!async_extent->pages) {
583                         int page_started = 0;
584                         unsigned long nr_written = 0;
585
586                         lock_extent(io_tree, async_extent->start,
587                                          async_extent->start +
588                                          async_extent->ram_size - 1, GFP_NOFS);
589
590                         /* allocate blocks */
591                         ret = cow_file_range(inode, async_cow->locked_page,
592                                              async_extent->start,
593                                              async_extent->start +
594                                              async_extent->ram_size - 1,
595                                              &page_started, &nr_written, 0);
596
597                         /*
598                          * if page_started, cow_file_range inserted an
599                          * inline extent and took care of all the unlocking
600                          * and IO for us.  Otherwise, we need to submit
601                          * all those pages down to the drive.
602                          */
603                         if (!page_started && !ret)
604                                 extent_write_locked_range(io_tree,
605                                                   inode, async_extent->start,
606                                                   async_extent->start +
607                                                   async_extent->ram_size - 1,
608                                                   btrfs_get_extent,
609                                                   WB_SYNC_ALL);
610                         kfree(async_extent);
611                         cond_resched();
612                         continue;
613                 }
614
615                 lock_extent(io_tree, async_extent->start,
616                             async_extent->start + async_extent->ram_size - 1,
617                             GFP_NOFS);
618
619                 trans = btrfs_join_transaction(root, 1);
620                 BUG_ON(IS_ERR(trans));
621                 ret = btrfs_reserve_extent(trans, root,
622                                            async_extent->compressed_size,
623                                            async_extent->compressed_size,
624                                            0, alloc_hint,
625                                            (u64)-1, &ins, 1);
626                 btrfs_end_transaction(trans, root);
627
628                 if (ret) {
629                         int i;
630                         for (i = 0; i < async_extent->nr_pages; i++) {
631                                 WARN_ON(async_extent->pages[i]->mapping);
632                                 page_cache_release(async_extent->pages[i]);
633                         }
634                         kfree(async_extent->pages);
635                         async_extent->nr_pages = 0;
636                         async_extent->pages = NULL;
637                         unlock_extent(io_tree, async_extent->start,
638                                       async_extent->start +
639                                       async_extent->ram_size - 1, GFP_NOFS);
640                         goto retry;
641                 }
642
643                 /*
644                  * here we're doing allocation and writeback of the
645                  * compressed pages
646                  */
647                 btrfs_drop_extent_cache(inode, async_extent->start,
648                                         async_extent->start +
649                                         async_extent->ram_size - 1, 0);
650
651                 em = alloc_extent_map(GFP_NOFS);
652                 BUG_ON(!em);
653                 em->start = async_extent->start;
654                 em->len = async_extent->ram_size;
655                 em->orig_start = em->start;
656
657                 em->block_start = ins.objectid;
658                 em->block_len = ins.offset;
659                 em->bdev = root->fs_info->fs_devices->latest_bdev;
660                 em->compress_type = async_extent->compress_type;
661                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
662                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
663
664                 while (1) {
665                         write_lock(&em_tree->lock);
666                         ret = add_extent_mapping(em_tree, em);
667                         write_unlock(&em_tree->lock);
668                         if (ret != -EEXIST) {
669                                 free_extent_map(em);
670                                 break;
671                         }
672                         btrfs_drop_extent_cache(inode, async_extent->start,
673                                                 async_extent->start +
674                                                 async_extent->ram_size - 1, 0);
675                 }
676
677                 ret = btrfs_add_ordered_extent_compress(inode,
678                                                 async_extent->start,
679                                                 ins.objectid,
680                                                 async_extent->ram_size,
681                                                 ins.offset,
682                                                 BTRFS_ORDERED_COMPRESSED,
683                                                 async_extent->compress_type);
684                 BUG_ON(ret);
685
686                 /*
687                  * clear dirty, set writeback and unlock the pages.
688                  */
689                 extent_clear_unlock_delalloc(inode,
690                                 &BTRFS_I(inode)->io_tree,
691                                 async_extent->start,
692                                 async_extent->start +
693                                 async_extent->ram_size - 1,
694                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
695                                 EXTENT_CLEAR_UNLOCK |
696                                 EXTENT_CLEAR_DELALLOC |
697                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
698
699                 ret = btrfs_submit_compressed_write(inode,
700                                     async_extent->start,
701                                     async_extent->ram_size,
702                                     ins.objectid,
703                                     ins.offset, async_extent->pages,
704                                     async_extent->nr_pages);
705
706                 BUG_ON(ret);
707                 alloc_hint = ins.objectid + ins.offset;
708                 kfree(async_extent);
709                 cond_resched();
710         }
711
712         return 0;
713 }
714
715 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
716                                       u64 num_bytes)
717 {
718         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
719         struct extent_map *em;
720         u64 alloc_hint = 0;
721
722         read_lock(&em_tree->lock);
723         em = search_extent_mapping(em_tree, start, num_bytes);
724         if (em) {
725                 /*
726                  * if block start isn't an actual block number then find the
727                  * first block in this inode and use that as a hint.  If that
728                  * block is also bogus then just don't worry about it.
729                  */
730                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
731                         free_extent_map(em);
732                         em = search_extent_mapping(em_tree, 0, 0);
733                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
734                                 alloc_hint = em->block_start;
735                         if (em)
736                                 free_extent_map(em);
737                 } else {
738                         alloc_hint = em->block_start;
739                         free_extent_map(em);
740                 }
741         }
742         read_unlock(&em_tree->lock);
743
744         return alloc_hint;
745 }
746
747 /*
748  * when extent_io.c finds a delayed allocation range in the file,
749  * the call backs end up in this code.  The basic idea is to
750  * allocate extents on disk for the range, and create ordered data structs
751  * in ram to track those extents.
752  *
753  * locked_page is the page that writepage had locked already.  We use
754  * it to make sure we don't do extra locks or unlocks.
755  *
756  * *page_started is set to one if we unlock locked_page and do everything
757  * required to start IO on it.  It may be clean and already done with
758  * IO when we return.
759  */
760 static noinline int cow_file_range(struct inode *inode,
761                                    struct page *locked_page,
762                                    u64 start, u64 end, int *page_started,
763                                    unsigned long *nr_written,
764                                    int unlock)
765 {
766         struct btrfs_root *root = BTRFS_I(inode)->root;
767         struct btrfs_trans_handle *trans;
768         u64 alloc_hint = 0;
769         u64 num_bytes;
770         unsigned long ram_size;
771         u64 disk_num_bytes;
772         u64 cur_alloc_size;
773         u64 blocksize = root->sectorsize;
774         struct btrfs_key ins;
775         struct extent_map *em;
776         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
777         int ret = 0;
778
779         BUG_ON(root == root->fs_info->tree_root);
780         trans = btrfs_join_transaction(root, 1);
781         BUG_ON(IS_ERR(trans));
782         btrfs_set_trans_block_group(trans, inode);
783         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
784
785         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
786         num_bytes = max(blocksize,  num_bytes);
787         disk_num_bytes = num_bytes;
788         ret = 0;
789
790         if (start == 0) {
791                 /* lets try to make an inline extent */
792                 ret = cow_file_range_inline(trans, root, inode,
793                                             start, end, 0, 0, NULL);
794                 if (ret == 0) {
795                         extent_clear_unlock_delalloc(inode,
796                                      &BTRFS_I(inode)->io_tree,
797                                      start, end, NULL,
798                                      EXTENT_CLEAR_UNLOCK_PAGE |
799                                      EXTENT_CLEAR_UNLOCK |
800                                      EXTENT_CLEAR_DELALLOC |
801                                      EXTENT_CLEAR_DIRTY |
802                                      EXTENT_SET_WRITEBACK |
803                                      EXTENT_END_WRITEBACK);
804
805                         *nr_written = *nr_written +
806                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
807                         *page_started = 1;
808                         ret = 0;
809                         goto out;
810                 }
811         }
812
813         BUG_ON(disk_num_bytes >
814                btrfs_super_total_bytes(&root->fs_info->super_copy));
815
816         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
817         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
818
819         while (disk_num_bytes > 0) {
820                 unsigned long op;
821
822                 cur_alloc_size = disk_num_bytes;
823                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
824                                            root->sectorsize, 0, alloc_hint,
825                                            (u64)-1, &ins, 1);
826                 BUG_ON(ret);
827
828                 em = alloc_extent_map(GFP_NOFS);
829                 BUG_ON(!em);
830                 em->start = start;
831                 em->orig_start = em->start;
832                 ram_size = ins.offset;
833                 em->len = ins.offset;
834
835                 em->block_start = ins.objectid;
836                 em->block_len = ins.offset;
837                 em->bdev = root->fs_info->fs_devices->latest_bdev;
838                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
839
840                 while (1) {
841                         write_lock(&em_tree->lock);
842                         ret = add_extent_mapping(em_tree, em);
843                         write_unlock(&em_tree->lock);
844                         if (ret != -EEXIST) {
845                                 free_extent_map(em);
846                                 break;
847                         }
848                         btrfs_drop_extent_cache(inode, start,
849                                                 start + ram_size - 1, 0);
850                 }
851
852                 cur_alloc_size = ins.offset;
853                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
854                                                ram_size, cur_alloc_size, 0);
855                 BUG_ON(ret);
856
857                 if (root->root_key.objectid ==
858                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
859                         ret = btrfs_reloc_clone_csums(inode, start,
860                                                       cur_alloc_size);
861                         BUG_ON(ret);
862                 }
863
864                 if (disk_num_bytes < cur_alloc_size)
865                         break;
866
867                 /* we're not doing compressed IO, don't unlock the first
868                  * page (which the caller expects to stay locked), don't
869                  * clear any dirty bits and don't set any writeback bits
870                  *
871                  * Do set the Private2 bit so we know this page was properly
872                  * setup for writepage
873                  */
874                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
875                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
876                         EXTENT_SET_PRIVATE2;
877
878                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
879                                              start, start + ram_size - 1,
880                                              locked_page, op);
881                 disk_num_bytes -= cur_alloc_size;
882                 num_bytes -= cur_alloc_size;
883                 alloc_hint = ins.objectid + ins.offset;
884                 start += cur_alloc_size;
885         }
886 out:
887         ret = 0;
888         btrfs_end_transaction(trans, root);
889
890         return ret;
891 }
892
893 /*
894  * work queue call back to started compression on a file and pages
895  */
896 static noinline void async_cow_start(struct btrfs_work *work)
897 {
898         struct async_cow *async_cow;
899         int num_added = 0;
900         async_cow = container_of(work, struct async_cow, work);
901
902         compress_file_range(async_cow->inode, async_cow->locked_page,
903                             async_cow->start, async_cow->end, async_cow,
904                             &num_added);
905         if (num_added == 0)
906                 async_cow->inode = NULL;
907 }
908
909 /*
910  * work queue call back to submit previously compressed pages
911  */
912 static noinline void async_cow_submit(struct btrfs_work *work)
913 {
914         struct async_cow *async_cow;
915         struct btrfs_root *root;
916         unsigned long nr_pages;
917
918         async_cow = container_of(work, struct async_cow, work);
919
920         root = async_cow->root;
921         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
922                 PAGE_CACHE_SHIFT;
923
924         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
925
926         if (atomic_read(&root->fs_info->async_delalloc_pages) <
927             5 * 1042 * 1024 &&
928             waitqueue_active(&root->fs_info->async_submit_wait))
929                 wake_up(&root->fs_info->async_submit_wait);
930
931         if (async_cow->inode)
932                 submit_compressed_extents(async_cow->inode, async_cow);
933 }
934
935 static noinline void async_cow_free(struct btrfs_work *work)
936 {
937         struct async_cow *async_cow;
938         async_cow = container_of(work, struct async_cow, work);
939         kfree(async_cow);
940 }
941
942 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
943                                 u64 start, u64 end, int *page_started,
944                                 unsigned long *nr_written)
945 {
946         struct async_cow *async_cow;
947         struct btrfs_root *root = BTRFS_I(inode)->root;
948         unsigned long nr_pages;
949         u64 cur_end;
950         int limit = 10 * 1024 * 1042;
951
952         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
953                          1, 0, NULL, GFP_NOFS);
954         while (start < end) {
955                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
956                 BUG_ON(!async_cow);
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         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1774         if (!ret) {
1775                 ret = btrfs_update_inode(trans, root, inode);
1776                 BUG_ON(ret);
1777         }
1778         ret = 0;
1779 out:
1780         if (nolock) {
1781                 if (trans)
1782                         btrfs_end_transaction_nolock(trans, root);
1783         } else {
1784                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1785                 if (trans)
1786                         btrfs_end_transaction(trans, root);
1787         }
1788
1789         /* once for us */
1790         btrfs_put_ordered_extent(ordered_extent);
1791         /* once for the tree */
1792         btrfs_put_ordered_extent(ordered_extent);
1793
1794         return 0;
1795 }
1796
1797 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1798                                 struct extent_state *state, int uptodate)
1799 {
1800         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1801
1802         ClearPagePrivate2(page);
1803         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1804 }
1805
1806 /*
1807  * When IO fails, either with EIO or csum verification fails, we
1808  * try other mirrors that might have a good copy of the data.  This
1809  * io_failure_record is used to record state as we go through all the
1810  * mirrors.  If another mirror has good data, the page is set up to date
1811  * and things continue.  If a good mirror can't be found, the original
1812  * bio end_io callback is called to indicate things have failed.
1813  */
1814 struct io_failure_record {
1815         struct page *page;
1816         u64 start;
1817         u64 len;
1818         u64 logical;
1819         unsigned long bio_flags;
1820         int last_mirror;
1821 };
1822
1823 static int btrfs_io_failed_hook(struct bio *failed_bio,
1824                          struct page *page, u64 start, u64 end,
1825                          struct extent_state *state)
1826 {
1827         struct io_failure_record *failrec = NULL;
1828         u64 private;
1829         struct extent_map *em;
1830         struct inode *inode = page->mapping->host;
1831         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1832         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1833         struct bio *bio;
1834         int num_copies;
1835         int ret;
1836         int rw;
1837         u64 logical;
1838
1839         ret = get_state_private(failure_tree, start, &private);
1840         if (ret) {
1841                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1842                 if (!failrec)
1843                         return -ENOMEM;
1844                 failrec->start = start;
1845                 failrec->len = end - start + 1;
1846                 failrec->last_mirror = 0;
1847                 failrec->bio_flags = 0;
1848
1849                 read_lock(&em_tree->lock);
1850                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1851                 if (em->start > start || em->start + em->len < start) {
1852                         free_extent_map(em);
1853                         em = NULL;
1854                 }
1855                 read_unlock(&em_tree->lock);
1856
1857                 if (!em || IS_ERR(em)) {
1858                         kfree(failrec);
1859                         return -EIO;
1860                 }
1861                 logical = start - em->start;
1862                 logical = em->block_start + logical;
1863                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1864                         logical = em->block_start;
1865                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1866                         extent_set_compress_type(&failrec->bio_flags,
1867                                                  em->compress_type);
1868                 }
1869                 failrec->logical = logical;
1870                 free_extent_map(em);
1871                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1872                                 EXTENT_DIRTY, GFP_NOFS);
1873                 set_state_private(failure_tree, start,
1874                                  (u64)(unsigned long)failrec);
1875         } else {
1876                 failrec = (struct io_failure_record *)(unsigned long)private;
1877         }
1878         num_copies = btrfs_num_copies(
1879                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1880                               failrec->logical, failrec->len);
1881         failrec->last_mirror++;
1882         if (!state) {
1883                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1884                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1885                                                     failrec->start,
1886                                                     EXTENT_LOCKED);
1887                 if (state && state->start != failrec->start)
1888                         state = NULL;
1889                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1890         }
1891         if (!state || failrec->last_mirror > num_copies) {
1892                 set_state_private(failure_tree, failrec->start, 0);
1893                 clear_extent_bits(failure_tree, failrec->start,
1894                                   failrec->start + failrec->len - 1,
1895                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1896                 kfree(failrec);
1897                 return -EIO;
1898         }
1899         bio = bio_alloc(GFP_NOFS, 1);
1900         bio->bi_private = state;
1901         bio->bi_end_io = failed_bio->bi_end_io;
1902         bio->bi_sector = failrec->logical >> 9;
1903         bio->bi_bdev = failed_bio->bi_bdev;
1904         bio->bi_size = 0;
1905
1906         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1907         if (failed_bio->bi_rw & REQ_WRITE)
1908                 rw = WRITE;
1909         else
1910                 rw = READ;
1911
1912         ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1913                                                       failrec->last_mirror,
1914                                                       failrec->bio_flags, 0);
1915         return ret;
1916 }
1917
1918 /*
1919  * each time an IO finishes, we do a fast check in the IO failure tree
1920  * to see if we need to process or clean up an io_failure_record
1921  */
1922 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1923 {
1924         u64 private;
1925         u64 private_failure;
1926         struct io_failure_record *failure;
1927         int ret;
1928
1929         private = 0;
1930         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1931                              (u64)-1, 1, EXTENT_DIRTY, 0)) {
1932                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1933                                         start, &private_failure);
1934                 if (ret == 0) {
1935                         failure = (struct io_failure_record *)(unsigned long)
1936                                    private_failure;
1937                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1938                                           failure->start, 0);
1939                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1940                                           failure->start,
1941                                           failure->start + failure->len - 1,
1942                                           EXTENT_DIRTY | EXTENT_LOCKED,
1943                                           GFP_NOFS);
1944                         kfree(failure);
1945                 }
1946         }
1947         return 0;
1948 }
1949
1950 /*
1951  * when reads are done, we need to check csums to verify the data is correct
1952  * if there's a match, we allow the bio to finish.  If not, we go through
1953  * the io_failure_record routines to find good copies
1954  */
1955 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1956                                struct extent_state *state)
1957 {
1958         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1959         struct inode *inode = page->mapping->host;
1960         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1961         char *kaddr;
1962         u64 private = ~(u32)0;
1963         int ret;
1964         struct btrfs_root *root = BTRFS_I(inode)->root;
1965         u32 csum = ~(u32)0;
1966
1967         if (PageChecked(page)) {
1968                 ClearPageChecked(page);
1969                 goto good;
1970         }
1971
1972         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1973                 return 0;
1974
1975         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1976             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1977                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1978                                   GFP_NOFS);
1979                 return 0;
1980         }
1981
1982         if (state && state->start == start) {
1983                 private = state->private;
1984                 ret = 0;
1985         } else {
1986                 ret = get_state_private(io_tree, start, &private);
1987         }
1988         kaddr = kmap_atomic(page, KM_USER0);
1989         if (ret)
1990                 goto zeroit;
1991
1992         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1993         btrfs_csum_final(csum, (char *)&csum);
1994         if (csum != private)
1995                 goto zeroit;
1996
1997         kunmap_atomic(kaddr, KM_USER0);
1998 good:
1999         /* if the io failure tree for this inode is non-empty,
2000          * check to see if we've recovered from a failed IO
2001          */
2002         btrfs_clean_io_failures(inode, start);
2003         return 0;
2004
2005 zeroit:
2006         if (printk_ratelimit()) {
2007                 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
2008                        "private %llu\n", page->mapping->host->i_ino,
2009                        (unsigned long long)start, csum,
2010                        (unsigned long long)private);
2011         }
2012         memset(kaddr + offset, 1, end - start + 1);
2013         flush_dcache_page(page);
2014         kunmap_atomic(kaddr, KM_USER0);
2015         if (private == 0)
2016                 return 0;
2017         return -EIO;
2018 }
2019
2020 struct delayed_iput {
2021         struct list_head list;
2022         struct inode *inode;
2023 };
2024
2025 void btrfs_add_delayed_iput(struct inode *inode)
2026 {
2027         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2028         struct delayed_iput *delayed;
2029
2030         if (atomic_add_unless(&inode->i_count, -1, 1))
2031                 return;
2032
2033         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2034         delayed->inode = inode;
2035
2036         spin_lock(&fs_info->delayed_iput_lock);
2037         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2038         spin_unlock(&fs_info->delayed_iput_lock);
2039 }
2040
2041 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2042 {
2043         LIST_HEAD(list);
2044         struct btrfs_fs_info *fs_info = root->fs_info;
2045         struct delayed_iput *delayed;
2046         int empty;
2047
2048         spin_lock(&fs_info->delayed_iput_lock);
2049         empty = list_empty(&fs_info->delayed_iputs);
2050         spin_unlock(&fs_info->delayed_iput_lock);
2051         if (empty)
2052                 return;
2053
2054         down_read(&root->fs_info->cleanup_work_sem);
2055         spin_lock(&fs_info->delayed_iput_lock);
2056         list_splice_init(&fs_info->delayed_iputs, &list);
2057         spin_unlock(&fs_info->delayed_iput_lock);
2058
2059         while (!list_empty(&list)) {
2060                 delayed = list_entry(list.next, struct delayed_iput, list);
2061                 list_del(&delayed->list);
2062                 iput(delayed->inode);
2063                 kfree(delayed);
2064         }
2065         up_read(&root->fs_info->cleanup_work_sem);
2066 }
2067
2068 /*
2069  * calculate extra metadata reservation when snapshotting a subvolume
2070  * contains orphan files.
2071  */
2072 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2073                                 struct btrfs_pending_snapshot *pending,
2074                                 u64 *bytes_to_reserve)
2075 {
2076         struct btrfs_root *root;
2077         struct btrfs_block_rsv *block_rsv;
2078         u64 num_bytes;
2079         int index;
2080
2081         root = pending->root;
2082         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2083                 return;
2084
2085         block_rsv = root->orphan_block_rsv;
2086
2087         /* orphan block reservation for the snapshot */
2088         num_bytes = block_rsv->size;
2089
2090         /*
2091          * after the snapshot is created, COWing tree blocks may use more
2092          * space than it frees. So we should make sure there is enough
2093          * reserved space.
2094          */
2095         index = trans->transid & 0x1;
2096         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2097                 num_bytes += block_rsv->size -
2098                              (block_rsv->reserved + block_rsv->freed[index]);
2099         }
2100
2101         *bytes_to_reserve += num_bytes;
2102 }
2103
2104 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2105                                 struct btrfs_pending_snapshot *pending)
2106 {
2107         struct btrfs_root *root = pending->root;
2108         struct btrfs_root *snap = pending->snap;
2109         struct btrfs_block_rsv *block_rsv;
2110         u64 num_bytes;
2111         int index;
2112         int ret;
2113
2114         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2115                 return;
2116
2117         /* refill source subvolume's orphan block reservation */
2118         block_rsv = root->orphan_block_rsv;
2119         index = trans->transid & 0x1;
2120         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2121                 num_bytes = block_rsv->size -
2122                             (block_rsv->reserved + block_rsv->freed[index]);
2123                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2124                                               root->orphan_block_rsv,
2125                                               num_bytes);
2126                 BUG_ON(ret);
2127         }
2128
2129         /* setup orphan block reservation for the snapshot */
2130         block_rsv = btrfs_alloc_block_rsv(snap);
2131         BUG_ON(!block_rsv);
2132
2133         btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2134         snap->orphan_block_rsv = block_rsv;
2135
2136         num_bytes = root->orphan_block_rsv->size;
2137         ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2138                                       block_rsv, num_bytes);
2139         BUG_ON(ret);
2140
2141 #if 0
2142         /* insert orphan item for the snapshot */
2143         WARN_ON(!root->orphan_item_inserted);
2144         ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2145                                        snap->root_key.objectid);
2146         BUG_ON(ret);
2147         snap->orphan_item_inserted = 1;
2148 #endif
2149 }
2150
2151 enum btrfs_orphan_cleanup_state {
2152         ORPHAN_CLEANUP_STARTED  = 1,
2153         ORPHAN_CLEANUP_DONE     = 2,
2154 };
2155
2156 /*
2157  * This is called in transaction commmit time. If there are no orphan
2158  * files in the subvolume, it removes orphan item and frees block_rsv
2159  * structure.
2160  */
2161 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2162                               struct btrfs_root *root)
2163 {
2164         int ret;
2165
2166         if (!list_empty(&root->orphan_list) ||
2167             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2168                 return;
2169
2170         if (root->orphan_item_inserted &&
2171             btrfs_root_refs(&root->root_item) > 0) {
2172                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2173                                             root->root_key.objectid);
2174                 BUG_ON(ret);
2175                 root->orphan_item_inserted = 0;
2176         }
2177
2178         if (root->orphan_block_rsv) {
2179                 WARN_ON(root->orphan_block_rsv->size > 0);
2180                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2181                 root->orphan_block_rsv = NULL;
2182         }
2183 }
2184
2185 /*
2186  * This creates an orphan entry for the given inode in case something goes
2187  * wrong in the middle of an unlink/truncate.
2188  *
2189  * NOTE: caller of this function should reserve 5 units of metadata for
2190  *       this function.
2191  */
2192 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2193 {
2194         struct btrfs_root *root = BTRFS_I(inode)->root;
2195         struct btrfs_block_rsv *block_rsv = NULL;
2196         int reserve = 0;
2197         int insert = 0;
2198         int ret;
2199
2200         if (!root->orphan_block_rsv) {
2201                 block_rsv = btrfs_alloc_block_rsv(root);
2202                 BUG_ON(!block_rsv);
2203         }
2204
2205         spin_lock(&root->orphan_lock);
2206         if (!root->orphan_block_rsv) {
2207                 root->orphan_block_rsv = block_rsv;
2208         } else if (block_rsv) {
2209                 btrfs_free_block_rsv(root, block_rsv);
2210                 block_rsv = NULL;
2211         }
2212
2213         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2214                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2215 #if 0
2216                 /*
2217                  * For proper ENOSPC handling, we should do orphan
2218                  * cleanup when mounting. But this introduces backward
2219                  * compatibility issue.
2220                  */
2221                 if (!xchg(&root->orphan_item_inserted, 1))
2222                         insert = 2;
2223                 else
2224                         insert = 1;
2225 #endif
2226                 insert = 1;
2227         }
2228
2229         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2230                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2231                 reserve = 1;
2232         }
2233         spin_unlock(&root->orphan_lock);
2234
2235         if (block_rsv)
2236                 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2237
2238         /* grab metadata reservation from transaction handle */
2239         if (reserve) {
2240                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2241                 BUG_ON(ret);
2242         }
2243
2244         /* insert an orphan item to track this unlinked/truncated file */
2245         if (insert >= 1) {
2246                 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2247                 BUG_ON(ret);
2248         }
2249
2250         /* insert an orphan item to track subvolume contains orphan files */
2251         if (insert >= 2) {
2252                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2253                                                root->root_key.objectid);
2254                 BUG_ON(ret);
2255         }
2256         return 0;
2257 }
2258
2259 /*
2260  * We have done the truncate/delete so we can go ahead and remove the orphan
2261  * item for this particular inode.
2262  */
2263 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2264 {
2265         struct btrfs_root *root = BTRFS_I(inode)->root;
2266         int delete_item = 0;
2267         int release_rsv = 0;
2268         int ret = 0;
2269
2270         spin_lock(&root->orphan_lock);
2271         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2272                 list_del_init(&BTRFS_I(inode)->i_orphan);
2273                 delete_item = 1;
2274         }
2275
2276         if (BTRFS_I(inode)->orphan_meta_reserved) {
2277                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2278                 release_rsv = 1;
2279         }
2280         spin_unlock(&root->orphan_lock);
2281
2282         if (trans && delete_item) {
2283                 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2284                 BUG_ON(ret);
2285         }
2286
2287         if (release_rsv)
2288                 btrfs_orphan_release_metadata(inode);
2289
2290         return 0;
2291 }
2292
2293 /*
2294  * this cleans up any orphans that may be left on the list from the last use
2295  * of this root.
2296  */
2297 int btrfs_orphan_cleanup(struct btrfs_root *root)
2298 {
2299         struct btrfs_path *path;
2300         struct extent_buffer *leaf;
2301         struct btrfs_key key, found_key;
2302         struct btrfs_trans_handle *trans;
2303         struct inode *inode;
2304         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2305
2306         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2307                 return 0;
2308
2309         path = btrfs_alloc_path();
2310         if (!path) {
2311                 ret = -ENOMEM;
2312                 goto out;
2313         }
2314         path->reada = -1;
2315
2316         key.objectid = BTRFS_ORPHAN_OBJECTID;
2317         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2318         key.offset = (u64)-1;
2319
2320         while (1) {
2321                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2322                 if (ret < 0)
2323                         goto out;
2324
2325                 /*
2326                  * if ret == 0 means we found what we were searching for, which
2327                  * is weird, but possible, so only screw with path if we didnt
2328                  * find the key and see if we have stuff that matches
2329                  */
2330                 if (ret > 0) {
2331                         ret = 0;
2332                         if (path->slots[0] == 0)
2333                                 break;
2334                         path->slots[0]--;
2335                 }
2336
2337                 /* pull out the item */
2338                 leaf = path->nodes[0];
2339                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2340
2341                 /* make sure the item matches what we want */
2342                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2343                         break;
2344                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2345                         break;
2346
2347                 /* release the path since we're done with it */
2348                 btrfs_release_path(root, path);
2349
2350                 /*
2351                  * this is where we are basically btrfs_lookup, without the
2352                  * crossing root thing.  we store the inode number in the
2353                  * offset of the orphan item.
2354                  */
2355                 found_key.objectid = found_key.offset;
2356                 found_key.type = BTRFS_INODE_ITEM_KEY;
2357                 found_key.offset = 0;
2358                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2359                 if (IS_ERR(inode)) {
2360                         ret = PTR_ERR(inode);
2361                         goto out;
2362                 }
2363
2364                 /*
2365                  * add this inode to the orphan list so btrfs_orphan_del does
2366                  * the proper thing when we hit it
2367                  */
2368                 spin_lock(&root->orphan_lock);
2369                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2370                 spin_unlock(&root->orphan_lock);
2371
2372                 /*
2373                  * if this is a bad inode, means we actually succeeded in
2374                  * removing the inode, but not the orphan record, which means
2375                  * we need to manually delete the orphan since iput will just
2376                  * do a destroy_inode
2377                  */
2378                 if (is_bad_inode(inode)) {
2379                         trans = btrfs_start_transaction(root, 0);
2380                         if (IS_ERR(trans)) {
2381                                 ret = PTR_ERR(trans);
2382                                 goto out;
2383                         }
2384                         btrfs_orphan_del(trans, inode);
2385                         btrfs_end_transaction(trans, root);
2386                         iput(inode);
2387                         continue;
2388                 }
2389
2390                 /* if we have links, this was a truncate, lets do that */
2391                 if (inode->i_nlink) {
2392                         if (!S_ISREG(inode->i_mode)) {
2393                                 WARN_ON(1);
2394                                 iput(inode);
2395                                 continue;
2396                         }
2397                         nr_truncate++;
2398                         ret = btrfs_truncate(inode);
2399                 } else {
2400                         nr_unlink++;
2401                 }
2402
2403                 /* this will do delete_inode and everything for us */
2404                 iput(inode);
2405                 if (ret)
2406                         goto out;
2407         }
2408         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2409
2410         if (root->orphan_block_rsv)
2411                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2412                                         (u64)-1);
2413
2414         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2415                 trans = btrfs_join_transaction(root, 1);
2416                 if (!IS_ERR(trans))
2417                         btrfs_end_transaction(trans, root);
2418         }
2419
2420         if (nr_unlink)
2421                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2422         if (nr_truncate)
2423                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2424
2425 out:
2426         if (ret)
2427                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2428         btrfs_free_path(path);
2429         return ret;
2430 }
2431
2432 /*
2433  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2434  * don't find any xattrs, we know there can't be any acls.
2435  *
2436  * slot is the slot the inode is in, objectid is the objectid of the inode
2437  */
2438 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2439                                           int slot, u64 objectid)
2440 {
2441         u32 nritems = btrfs_header_nritems(leaf);
2442         struct btrfs_key found_key;
2443         int scanned = 0;
2444
2445         slot++;
2446         while (slot < nritems) {
2447                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2448
2449                 /* we found a different objectid, there must not be acls */
2450                 if (found_key.objectid != objectid)
2451                         return 0;
2452
2453                 /* we found an xattr, assume we've got an acl */
2454                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2455                         return 1;
2456
2457                 /*
2458                  * we found a key greater than an xattr key, there can't
2459                  * be any acls later on
2460                  */
2461                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2462                         return 0;
2463
2464                 slot++;
2465                 scanned++;
2466
2467                 /*
2468                  * it goes inode, inode backrefs, xattrs, extents,
2469                  * so if there are a ton of hard links to an inode there can
2470                  * be a lot of backrefs.  Don't waste time searching too hard,
2471                  * this is just an optimization
2472                  */
2473                 if (scanned >= 8)
2474                         break;
2475         }
2476         /* we hit the end of the leaf before we found an xattr or
2477          * something larger than an xattr.  We have to assume the inode
2478          * has acls
2479          */
2480         return 1;
2481 }
2482
2483 /*
2484  * read an inode from the btree into the in-memory inode
2485  */
2486 static void btrfs_read_locked_inode(struct inode *inode)
2487 {
2488         struct btrfs_path *path;
2489         struct extent_buffer *leaf;
2490         struct btrfs_inode_item *inode_item;
2491         struct btrfs_timespec *tspec;
2492         struct btrfs_root *root = BTRFS_I(inode)->root;
2493         struct btrfs_key location;
2494         int maybe_acls;
2495         u64 alloc_group_block;
2496         u32 rdev;
2497         int ret;
2498
2499         path = btrfs_alloc_path();
2500         BUG_ON(!path);
2501         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2502
2503         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2504         if (ret)
2505                 goto make_bad;
2506
2507         leaf = path->nodes[0];
2508         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2509                                     struct btrfs_inode_item);
2510
2511         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2512         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2513         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2514         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2515         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2516
2517         tspec = btrfs_inode_atime(inode_item);
2518         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2519         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2520
2521         tspec = btrfs_inode_mtime(inode_item);
2522         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2523         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2524
2525         tspec = btrfs_inode_ctime(inode_item);
2526         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2527         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2528
2529         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2530         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2531         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2532         inode->i_generation = BTRFS_I(inode)->generation;
2533         inode->i_rdev = 0;
2534         rdev = btrfs_inode_rdev(leaf, inode_item);
2535
2536         BTRFS_I(inode)->index_cnt = (u64)-1;
2537         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2538
2539         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2540
2541         /*
2542          * try to precache a NULL acl entry for files that don't have
2543          * any xattrs or acls
2544          */
2545         maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2546         if (!maybe_acls)
2547                 cache_no_acl(inode);
2548
2549         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2550                                                 alloc_group_block, 0);
2551         btrfs_free_path(path);
2552         inode_item = NULL;
2553
2554         switch (inode->i_mode & S_IFMT) {
2555         case S_IFREG:
2556                 inode->i_mapping->a_ops = &btrfs_aops;
2557                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2558                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2559                 inode->i_fop = &btrfs_file_operations;
2560                 inode->i_op = &btrfs_file_inode_operations;
2561                 break;
2562         case S_IFDIR:
2563                 inode->i_fop = &btrfs_dir_file_operations;
2564                 if (root == root->fs_info->tree_root)
2565                         inode->i_op = &btrfs_dir_ro_inode_operations;
2566                 else
2567                         inode->i_op = &btrfs_dir_inode_operations;
2568                 break;
2569         case S_IFLNK:
2570                 inode->i_op = &btrfs_symlink_inode_operations;
2571                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2572                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2573                 break;
2574         default:
2575                 inode->i_op = &btrfs_special_inode_operations;
2576                 init_special_inode(inode, inode->i_mode, rdev);
2577                 break;
2578         }
2579
2580         btrfs_update_iflags(inode);
2581         return;
2582
2583 make_bad:
2584         btrfs_free_path(path);
2585         make_bad_inode(inode);
2586 }
2587
2588 /*
2589  * given a leaf and an inode, copy the inode fields into the leaf
2590  */
2591 static void fill_inode_item(struct btrfs_trans_handle *trans,
2592                             struct extent_buffer *leaf,
2593                             struct btrfs_inode_item *item,
2594                             struct inode *inode)
2595 {
2596         if (!leaf->map_token)
2597                 map_private_extent_buffer(leaf, (unsigned long)item,
2598                                           sizeof(struct btrfs_inode_item),
2599                                           &leaf->map_token, &leaf->kaddr,
2600                                           &leaf->map_start, &leaf->map_len,
2601                                           KM_USER1);
2602
2603         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2604         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2605         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2606         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2607         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2608
2609         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2610                                inode->i_atime.tv_sec);
2611         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2612                                 inode->i_atime.tv_nsec);
2613
2614         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2615                                inode->i_mtime.tv_sec);
2616         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2617                                 inode->i_mtime.tv_nsec);
2618
2619         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2620                                inode->i_ctime.tv_sec);
2621         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2622                                 inode->i_ctime.tv_nsec);
2623
2624         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2625         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2626         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2627         btrfs_set_inode_transid(leaf, item, trans->transid);
2628         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2629         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2630         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2631
2632         if (leaf->map_token) {
2633                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2634                 leaf->map_token = NULL;
2635         }
2636 }
2637
2638 /*
2639  * copy everything in the in-memory inode into the btree.
2640  */
2641 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2642                                 struct btrfs_root *root, struct inode *inode)
2643 {
2644         struct btrfs_inode_item *inode_item;
2645         struct btrfs_path *path;
2646         struct extent_buffer *leaf;
2647         int ret;
2648
2649         path = btrfs_alloc_path();
2650         BUG_ON(!path);
2651         path->leave_spinning = 1;
2652         ret = btrfs_lookup_inode(trans, root, path,
2653                                  &BTRFS_I(inode)->location, 1);
2654         if (ret) {
2655                 if (ret > 0)
2656                         ret = -ENOENT;
2657                 goto failed;
2658         }
2659
2660         btrfs_unlock_up_safe(path, 1);
2661         leaf = path->nodes[0];
2662         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2663                                   struct btrfs_inode_item);
2664
2665         fill_inode_item(trans, leaf, inode_item, inode);
2666         btrfs_mark_buffer_dirty(leaf);
2667         btrfs_set_inode_last_trans(trans, inode);
2668         ret = 0;
2669 failed:
2670         btrfs_free_path(path);
2671         return ret;
2672 }
2673
2674
2675 /*
2676  * unlink helper that gets used here in inode.c and in the tree logging
2677  * recovery code.  It remove a link in a directory with a given name, and
2678  * also drops the back refs in the inode to the directory
2679  */
2680 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2681                                 struct btrfs_root *root,
2682                                 struct inode *dir, struct inode *inode,
2683                                 const char *name, int name_len)
2684 {
2685         struct btrfs_path *path;
2686         int ret = 0;
2687         struct extent_buffer *leaf;
2688         struct btrfs_dir_item *di;
2689         struct btrfs_key key;
2690         u64 index;
2691
2692         path = btrfs_alloc_path();
2693         if (!path) {
2694                 ret = -ENOMEM;
2695                 goto out;
2696         }
2697
2698         path->leave_spinning = 1;
2699         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2700                                     name, name_len, -1);
2701         if (IS_ERR(di)) {
2702                 ret = PTR_ERR(di);
2703                 goto err;
2704         }
2705         if (!di) {
2706                 ret = -ENOENT;
2707                 goto err;
2708         }
2709         leaf = path->nodes[0];
2710         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2711         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2712         if (ret)
2713                 goto err;
2714         btrfs_release_path(root, path);
2715
2716         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2717                                   inode->i_ino,
2718                                   dir->i_ino, &index);
2719         if (ret) {
2720                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2721                        "inode %lu parent %lu\n", name_len, name,
2722                        inode->i_ino, dir->i_ino);
2723                 goto err;
2724         }
2725
2726         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2727                                          index, name, name_len, -1);
2728         if (IS_ERR(di)) {
2729                 ret = PTR_ERR(di);
2730                 goto err;
2731         }
2732         if (!di) {
2733                 ret = -ENOENT;
2734                 goto err;
2735         }
2736         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2737         btrfs_release_path(root, path);
2738
2739         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2740                                          inode, dir->i_ino);
2741         BUG_ON(ret != 0 && ret != -ENOENT);
2742
2743         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2744                                            dir, index);
2745         if (ret == -ENOENT)
2746                 ret = 0;
2747 err:
2748         btrfs_free_path(path);
2749         if (ret)
2750                 goto out;
2751
2752         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2753         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2754         btrfs_update_inode(trans, root, dir);
2755 out:
2756         return ret;
2757 }
2758
2759 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2760                        struct btrfs_root *root,
2761                        struct inode *dir, struct inode *inode,
2762                        const char *name, int name_len)
2763 {
2764         int ret;
2765         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2766         if (!ret) {
2767                 btrfs_drop_nlink(inode);
2768                 ret = btrfs_update_inode(trans, root, inode);
2769         }
2770         return ret;
2771 }
2772                 
2773
2774 /* helper to check if there is any shared block in the path */
2775 static int check_path_shared(struct btrfs_root *root,
2776                              struct btrfs_path *path)
2777 {
2778         struct extent_buffer *eb;
2779         int level;
2780         u64 refs = 1;
2781
2782         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2783                 int ret;
2784
2785                 if (!path->nodes[level])
2786                         break;
2787                 eb = path->nodes[level];
2788                 if (!btrfs_block_can_be_shared(root, eb))
2789                         continue;
2790                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2791                                                &refs, NULL);
2792                 if (refs > 1)
2793                         return 1;
2794         }
2795         return 0;
2796 }
2797
2798 /*
2799  * helper to start transaction for unlink and rmdir.
2800  *
2801  * unlink and rmdir are special in btrfs, they do not always free space.
2802  * so in enospc case, we should make sure they will free space before
2803  * allowing them to use the global metadata reservation.
2804  */
2805 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2806                                                        struct dentry *dentry)
2807 {
2808         struct btrfs_trans_handle *trans;
2809         struct btrfs_root *root = BTRFS_I(dir)->root;
2810         struct btrfs_path *path;
2811         struct btrfs_inode_ref *ref;
2812         struct btrfs_dir_item *di;
2813         struct inode *inode = dentry->d_inode;
2814         u64 index;
2815         int check_link = 1;
2816         int err = -ENOSPC;
2817         int ret;
2818
2819         trans = btrfs_start_transaction(root, 10);
2820         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2821                 return trans;
2822
2823         if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2824                 return ERR_PTR(-ENOSPC);
2825
2826         /* check if there is someone else holds reference */
2827         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2828                 return ERR_PTR(-ENOSPC);
2829
2830         if (atomic_read(&inode->i_count) > 2)
2831                 return ERR_PTR(-ENOSPC);
2832
2833         if (xchg(&root->fs_info->enospc_unlink, 1))
2834                 return ERR_PTR(-ENOSPC);
2835
2836         path = btrfs_alloc_path();
2837         if (!path) {
2838                 root->fs_info->enospc_unlink = 0;
2839                 return ERR_PTR(-ENOMEM);
2840         }
2841
2842         trans = btrfs_start_transaction(root, 0);
2843         if (IS_ERR(trans)) {
2844                 btrfs_free_path(path);
2845                 root->fs_info->enospc_unlink = 0;
2846                 return trans;
2847         }
2848
2849         path->skip_locking = 1;
2850         path->search_commit_root = 1;
2851
2852         ret = btrfs_lookup_inode(trans, root, path,
2853                                 &BTRFS_I(dir)->location, 0);
2854         if (ret < 0) {
2855                 err = ret;
2856                 goto out;
2857         }
2858         if (ret == 0) {
2859                 if (check_path_shared(root, path))
2860                         goto out;
2861         } else {
2862                 check_link = 0;
2863         }
2864         btrfs_release_path(root, path);
2865
2866         ret = btrfs_lookup_inode(trans, root, path,
2867                                 &BTRFS_I(inode)->location, 0);
2868         if (ret < 0) {
2869                 err = ret;
2870                 goto out;
2871         }
2872         if (ret == 0) {
2873                 if (check_path_shared(root, path))
2874                         goto out;
2875         } else {
2876                 check_link = 0;
2877         }
2878         btrfs_release_path(root, path);
2879
2880         if (ret == 0 && S_ISREG(inode->i_mode)) {
2881                 ret = btrfs_lookup_file_extent(trans, root, path,
2882                                                inode->i_ino, (u64)-1, 0);
2883                 if (ret < 0) {
2884                         err = ret;
2885                         goto out;
2886                 }
2887                 BUG_ON(ret == 0);
2888                 if (check_path_shared(root, path))
2889                         goto out;
2890                 btrfs_release_path(root, path);
2891         }
2892
2893         if (!check_link) {
2894                 err = 0;
2895                 goto out;
2896         }
2897
2898         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2899                                 dentry->d_name.name, dentry->d_name.len, 0);
2900         if (IS_ERR(di)) {
2901                 err = PTR_ERR(di);
2902                 goto out;
2903         }
2904         if (di) {
2905                 if (check_path_shared(root, path))
2906                         goto out;
2907         } else {
2908                 err = 0;
2909                 goto out;
2910         }
2911         btrfs_release_path(root, path);
2912
2913         ref = btrfs_lookup_inode_ref(trans, root, path,
2914                                 dentry->d_name.name, dentry->d_name.len,
2915                                 inode->i_ino, dir->i_ino, 0);
2916         if (IS_ERR(ref)) {
2917                 err = PTR_ERR(ref);
2918                 goto out;
2919         }
2920         BUG_ON(!ref);
2921         if (check_path_shared(root, path))
2922                 goto out;
2923         index = btrfs_inode_ref_index(path->nodes[0], ref);
2924         btrfs_release_path(root, path);
2925
2926         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2927                                 dentry->d_name.name, dentry->d_name.len, 0);
2928         if (IS_ERR(di)) {
2929                 err = PTR_ERR(di);
2930                 goto out;
2931         }
2932         BUG_ON(ret == -ENOENT);
2933         if (check_path_shared(root, path))
2934                 goto out;
2935
2936         err = 0;
2937 out:
2938         btrfs_free_path(path);
2939         if (err) {
2940                 btrfs_end_transaction(trans, root);
2941                 root->fs_info->enospc_unlink = 0;
2942                 return ERR_PTR(err);
2943         }
2944
2945         trans->block_rsv = &root->fs_info->global_block_rsv;
2946         return trans;
2947 }
2948
2949 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2950                                struct btrfs_root *root)
2951 {
2952         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2953                 BUG_ON(!root->fs_info->enospc_unlink);
2954                 root->fs_info->enospc_unlink = 0;
2955         }
2956         btrfs_end_transaction_throttle(trans, root);
2957 }
2958
2959 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2960 {
2961         struct btrfs_root *root = BTRFS_I(dir)->root;
2962         struct btrfs_trans_handle *trans;
2963         struct inode *inode = dentry->d_inode;
2964         int ret;
2965         unsigned long nr = 0;
2966
2967         trans = __unlink_start_trans(dir, dentry);
2968         if (IS_ERR(trans))
2969                 return PTR_ERR(trans);
2970
2971         btrfs_set_trans_block_group(trans, dir);
2972
2973         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2974
2975         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2976                                  dentry->d_name.name, dentry->d_name.len);
2977         BUG_ON(ret);
2978
2979         if (inode->i_nlink == 0) {
2980                 ret = btrfs_orphan_add(trans, inode);
2981                 BUG_ON(ret);
2982         }
2983
2984         nr = trans->blocks_used;
2985         __unlink_end_trans(trans, root);
2986         btrfs_btree_balance_dirty(root, nr);
2987         return ret;
2988 }
2989
2990 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2991                         struct btrfs_root *root,
2992                         struct inode *dir, u64 objectid,
2993                         const char *name, int name_len)
2994 {
2995         struct btrfs_path *path;
2996         struct extent_buffer *leaf;
2997         struct btrfs_dir_item *di;
2998         struct btrfs_key key;
2999         u64 index;
3000         int ret;
3001
3002         path = btrfs_alloc_path();
3003         if (!path)
3004                 return -ENOMEM;
3005
3006         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
3007                                    name, name_len, -1);
3008         BUG_ON(!di || IS_ERR(di));
3009
3010         leaf = path->nodes[0];
3011         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3012         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3013         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3014         BUG_ON(ret);
3015         btrfs_release_path(root, path);
3016
3017         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3018                                  objectid, root->root_key.objectid,
3019                                  dir->i_ino, &index, name, name_len);
3020         if (ret < 0) {
3021                 BUG_ON(ret != -ENOENT);
3022                 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
3023                                                  name, name_len);
3024                 BUG_ON(!di || IS_ERR(di));
3025
3026                 leaf = path->nodes[0];
3027                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3028                 btrfs_release_path(root, path);
3029                 index = key.offset;
3030         }
3031
3032         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
3033                                          index, name, name_len, -1);
3034         BUG_ON(!di || IS_ERR(di));
3035
3036         leaf = path->nodes[0];
3037         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3038         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3039         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3040         BUG_ON(ret);
3041         btrfs_release_path(root, path);
3042
3043         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3044         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3045         ret = btrfs_update_inode(trans, root, dir);
3046         BUG_ON(ret);
3047
3048         btrfs_free_path(path);
3049         return 0;
3050 }
3051
3052 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3053 {
3054         struct inode *inode = dentry->d_inode;
3055         int err = 0;
3056         struct btrfs_root *root = BTRFS_I(dir)->root;
3057         struct btrfs_trans_handle *trans;
3058         unsigned long nr = 0;
3059
3060         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3061             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3062                 return -ENOTEMPTY;
3063
3064         trans = __unlink_start_trans(dir, dentry);
3065         if (IS_ERR(trans))
3066                 return PTR_ERR(trans);
3067
3068         btrfs_set_trans_block_group(trans, dir);
3069
3070         if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3071                 err = btrfs_unlink_subvol(trans, root, dir,
3072                                           BTRFS_I(inode)->location.objectid,
3073                                           dentry->d_name.name,
3074                                           dentry->d_name.len);
3075                 goto out;
3076         }
3077
3078         err = btrfs_orphan_add(trans, inode);
3079         if (err)
3080                 goto out;
3081
3082         /* now the directory is empty */
3083         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3084                                  dentry->d_name.name, dentry->d_name.len);
3085         if (!err)
3086                 btrfs_i_size_write(inode, 0);
3087 out:
3088         nr = trans->blocks_used;
3089         __unlink_end_trans(trans, root);
3090         btrfs_btree_balance_dirty(root, nr);
3091
3092         return err;
3093 }
3094
3095 #if 0
3096 /*
3097  * when truncating bytes in a file, it is possible to avoid reading
3098  * the leaves that contain only checksum items.  This can be the
3099  * majority of the IO required to delete a large file, but it must
3100  * be done carefully.
3101  *
3102  * The keys in the level just above the leaves are checked to make sure
3103  * the lowest key in a given leaf is a csum key, and starts at an offset
3104  * after the new  size.
3105  *
3106  * Then the key for the next leaf is checked to make sure it also has
3107  * a checksum item for the same file.  If it does, we know our target leaf
3108  * contains only checksum items, and it can be safely freed without reading
3109  * it.
3110  *
3111  * This is just an optimization targeted at large files.  It may do
3112  * nothing.  It will return 0 unless things went badly.
3113  */
3114 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3115                                      struct btrfs_root *root,
3116                                      struct btrfs_path *path,
3117                                      struct inode *inode, u64 new_size)
3118 {
3119         struct btrfs_key key;
3120         int ret;
3121         int nritems;
3122         struct btrfs_key found_key;
3123         struct btrfs_key other_key;
3124         struct btrfs_leaf_ref *ref;
3125         u64 leaf_gen;
3126         u64 leaf_start;
3127
3128         path->lowest_level = 1;
3129         key.objectid = inode->i_ino;
3130         key.type = BTRFS_CSUM_ITEM_KEY;
3131         key.offset = new_size;
3132 again:
3133         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3134         if (ret < 0)
3135                 goto out;
3136
3137         if (path->nodes[1] == NULL) {
3138                 ret = 0;
3139                 goto out;
3140         }
3141         ret = 0;
3142         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3143         nritems = btrfs_header_nritems(path->nodes[1]);
3144
3145         if (!nritems)
3146                 goto out;
3147
3148         if (path->slots[1] >= nritems)
3149                 goto next_node;
3150
3151         /* did we find a key greater than anything we want to delete? */
3152         if (found_key.objectid > inode->i_ino ||
3153            (found_key.objectid == inode->i_ino && found_key.type > key.type))
3154                 goto out;
3155
3156         /* we check the next key in the node to make sure the leave contains
3157          * only checksum items.  This comparison doesn't work if our
3158          * leaf is the last one in the node
3159          */
3160         if (path->slots[1] + 1 >= nritems) {
3161 next_node:
3162                 /* search forward from the last key in the node, this
3163                  * will bring us into the next node in the tree
3164                  */
3165                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3166
3167                 /* unlikely, but we inc below, so check to be safe */
3168                 if (found_key.offset == (u64)-1)
3169                         goto out;
3170
3171                 /* search_forward needs a path with locks held, do the
3172                  * search again for the original key.  It is possible
3173                  * this will race with a balance and return a path that
3174                  * we could modify, but this drop is just an optimization
3175                  * and is allowed to miss some leaves.
3176                  */
3177                 btrfs_release_path(root, path);
3178                 found_key.offset++;
3179
3180                 /* setup a max key for search_forward */
3181                 other_key.offset = (u64)-1;
3182                 other_key.type = key.type;
3183                 other_key.objectid = key.objectid;
3184
3185                 path->keep_locks = 1;
3186                 ret = btrfs_search_forward(root, &found_key, &other_key,
3187                                            path, 0, 0);
3188                 path->keep_locks = 0;
3189                 if (ret || found_key.objectid != key.objectid ||
3190                     found_key.type != key.type) {
3191                         ret = 0;
3192                         goto out;
3193                 }
3194
3195                 key.offset = found_key.offset;
3196                 btrfs_release_path(root, path);
3197                 cond_resched();
3198                 goto again;
3199         }
3200
3201         /* we know there's one more slot after us in the tree,
3202          * read that key so we can verify it is also a checksum item
3203          */
3204         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3205
3206         if (found_key.objectid < inode->i_ino)
3207                 goto next_key;
3208
3209         if (found_key.type != key.type || found_key.offset < new_size)
3210                 goto next_key;
3211
3212         /*
3213          * if the key for the next leaf isn't a csum key from this objectid,
3214          * we can't be sure there aren't good items inside this leaf.
3215          * Bail out
3216          */
3217         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3218                 goto out;
3219
3220         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3221         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3222         /*
3223          * it is safe to delete this leaf, it contains only
3224          * csum items from this inode at an offset >= new_size
3225          */
3226         ret = btrfs_del_leaf(trans, root, path, leaf_start);
3227         BUG_ON(ret);
3228
3229         if (root->ref_cows && leaf_gen < trans->transid) {
3230                 ref = btrfs_alloc_leaf_ref(root, 0);
3231                 if (ref) {
3232                         ref->root_gen = root->root_key.offset;
3233                         ref->bytenr = leaf_start;
3234                         ref->owner = 0;
3235                         ref->generation = leaf_gen;
3236                         ref->nritems = 0;
3237
3238                         btrfs_sort_leaf_ref(ref);
3239
3240                         ret = btrfs_add_leaf_ref(root, ref, 0);
3241                         WARN_ON(ret);
3242                         btrfs_free_leaf_ref(root, ref);
3243                 } else {
3244                         WARN_ON(1);
3245                 }
3246         }
3247 next_key:
3248         btrfs_release_path(root, path);
3249
3250         if (other_key.objectid == inode->i_ino &&
3251             other_key.type == key.type && other_key.offset > key.offset) {
3252                 key.offset = other_key.offset;
3253                 cond_resched();
3254                 goto again;
3255         }
3256         ret = 0;
3257 out:
3258         /* fixup any changes we've made to the path */
3259         path->lowest_level = 0;
3260         path->keep_locks = 0;
3261         btrfs_release_path(root, path);
3262         return ret;
3263 }
3264
3265 #endif
3266
3267 /*
3268  * this can truncate away extent items, csum items and directory items.
3269  * It starts at a high offset and removes keys until it can't find
3270  * any higher than new_size
3271  *
3272  * csum items that cross the new i_size are truncated to the new size
3273  * as well.
3274  *
3275  * min_type is the minimum key type to truncate down to.  If set to 0, this
3276  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3277  */
3278 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3279                                struct btrfs_root *root,
3280                                struct inode *inode,
3281                                u64 new_size, u32 min_type)
3282 {
3283         struct btrfs_path *path;
3284         struct extent_buffer *leaf;
3285         struct btrfs_file_extent_item *fi;
3286         struct btrfs_key key;
3287         struct btrfs_key found_key;
3288         u64 extent_start = 0;
3289         u64 extent_num_bytes = 0;
3290         u64 extent_offset = 0;
3291         u64 item_end = 0;
3292         u64 mask = root->sectorsize - 1;
3293         u32 found_type = (u8)-1;
3294         int found_extent;
3295         int del_item;
3296         int pending_del_nr = 0;
3297         int pending_del_slot = 0;
3298         int extent_type = -1;
3299         int encoding;
3300         int ret;
3301         int err = 0;
3302
3303         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3304
3305         if (root->ref_cows || root == root->fs_info->tree_root)
3306                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3307
3308         path = btrfs_alloc_path();
3309         BUG_ON(!path);
3310         path->reada = -1;
3311
3312         key.objectid = inode->i_ino;
3313         key.offset = (u64)-1;
3314         key.type = (u8)-1;
3315
3316 search_again:
3317         path->leave_spinning = 1;
3318         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3319         if (ret < 0) {
3320                 err = ret;
3321                 goto out;
3322         }
3323
3324         if (ret > 0) {
3325                 /* there are no items in the tree for us to truncate, we're
3326                  * done
3327                  */
3328                 if (path->slots[0] == 0)
3329                         goto out;
3330                 path->slots[0]--;
3331         }
3332
3333         while (1) {
3334                 fi = NULL;
3335                 leaf = path->nodes[0];
3336                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3337                 found_type = btrfs_key_type(&found_key);
3338                 encoding = 0;
3339
3340                 if (found_key.objectid != inode->i_ino)
3341                         break;
3342
3343                 if (found_type < min_type)
3344                         break;
3345
3346                 item_end = found_key.offset;
3347                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3348                         fi = btrfs_item_ptr(leaf, path->slots[0],
3349                                             struct btrfs_file_extent_item);
3350                         extent_type = btrfs_file_extent_type(leaf, fi);
3351                         encoding = btrfs_file_extent_compression(leaf, fi);
3352                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3353                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3354
3355                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3356                                 item_end +=
3357                                     btrfs_file_extent_num_bytes(leaf, fi);
3358                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3359                                 item_end += btrfs_file_extent_inline_len(leaf,
3360                                                                          fi);
3361                         }
3362                         item_end--;
3363                 }
3364                 if (found_type > min_type) {
3365                         del_item = 1;
3366                 } else {
3367                         if (item_end < new_size)
3368                                 break;
3369                         if (found_key.offset >= new_size)
3370                                 del_item = 1;
3371                         else
3372                                 del_item = 0;
3373                 }
3374                 found_extent = 0;
3375                 /* FIXME, shrink the extent if the ref count is only 1 */
3376                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3377                         goto delete;
3378
3379                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3380                         u64 num_dec;
3381                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3382                         if (!del_item && !encoding) {
3383                                 u64 orig_num_bytes =
3384                                         btrfs_file_extent_num_bytes(leaf, fi);
3385                                 extent_num_bytes = new_size -
3386                                         found_key.offset + root->sectorsize - 1;
3387                                 extent_num_bytes = extent_num_bytes &
3388                                         ~((u64)root->sectorsize - 1);
3389                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3390                                                          extent_num_bytes);
3391                                 num_dec = (orig_num_bytes -
3392                                            extent_num_bytes);
3393                                 if (root->ref_cows && extent_start != 0)
3394                                         inode_sub_bytes(inode, num_dec);
3395                                 btrfs_mark_buffer_dirty(leaf);
3396                         } else {
3397                                 extent_num_bytes =
3398                                         btrfs_file_extent_disk_num_bytes(leaf,
3399                                                                          fi);
3400                                 extent_offset = found_key.offset -
3401                                         btrfs_file_extent_offset(leaf, fi);
3402
3403                                 /* FIXME blocksize != 4096 */
3404                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3405                                 if (extent_start != 0) {
3406                                         found_extent = 1;
3407                                         if (root->ref_cows)
3408                                                 inode_sub_bytes(inode, num_dec);
3409                                 }
3410                         }
3411                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3412                         /*
3413                          * we can't truncate inline items that have had
3414                          * special encodings
3415                          */
3416                         if (!del_item &&
3417                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3418                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3419                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3420                                 u32 size = new_size - found_key.offset;
3421
3422                                 if (root->ref_cows) {
3423                                         inode_sub_bytes(inode, item_end + 1 -
3424                                                         new_size);
3425                                 }
3426                                 size =
3427                                     btrfs_file_extent_calc_inline_size(size);
3428                                 ret = btrfs_truncate_item(trans, root, path,
3429                                                           size, 1);
3430                                 BUG_ON(ret);
3431                         } else if (root->ref_cows) {
3432                                 inode_sub_bytes(inode, item_end + 1 -
3433                                                 found_key.offset);
3434                         }
3435                 }
3436 delete:
3437                 if (del_item) {
3438                         if (!pending_del_nr) {
3439                                 /* no pending yet, add ourselves */
3440                                 pending_del_slot = path->slots[0];
3441                                 pending_del_nr = 1;
3442                         } else if (pending_del_nr &&
3443                                    path->slots[0] + 1 == pending_del_slot) {
3444                                 /* hop on the pending chunk */
3445                                 pending_del_nr++;
3446                                 pending_del_slot = path->slots[0];
3447                         } else {
3448                                 BUG();
3449                         }
3450                 } else {
3451                         break;
3452                 }
3453                 if (found_extent && (root->ref_cows ||
3454                                      root == root->fs_info->tree_root)) {
3455                         btrfs_set_path_blocking(path);
3456                         ret = btrfs_free_extent(trans, root, extent_start,
3457                                                 extent_num_bytes, 0,
3458                                                 btrfs_header_owner(leaf),
3459                                                 inode->i_ino, extent_offset);
3460                         BUG_ON(ret);
3461                 }
3462
3463                 if (found_type == BTRFS_INODE_ITEM_KEY)
3464                         break;
3465
3466                 if (path->slots[0] == 0 ||
3467                     path->slots[0] != pending_del_slot) {
3468                         if (root->ref_cows) {
3469                                 err = -EAGAIN;
3470                                 goto out;
3471                         }
3472                         if (pending_del_nr) {
3473                                 ret = btrfs_del_items(trans, root, path,
3474                                                 pending_del_slot,
3475                                                 pending_del_nr);
3476                                 BUG_ON(ret);
3477                                 pending_del_nr = 0;
3478                         }
3479                         btrfs_release_path(root, path);
3480                         goto search_again;
3481                 } else {
3482                         path->slots[0]--;
3483                 }
3484         }
3485 out:
3486         if (pending_del_nr) {
3487                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3488                                       pending_del_nr);
3489                 BUG_ON(ret);
3490         }
3491         btrfs_free_path(path);
3492         return err;
3493 }
3494
3495 /*
3496  * taken from block_truncate_page, but does cow as it zeros out
3497  * any bytes left in the last page in the file.
3498  */
3499 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3500 {
3501         struct inode *inode = mapping->host;
3502         struct btrfs_root *root = BTRFS_I(inode)->root;
3503         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3504         struct btrfs_ordered_extent *ordered;
3505         struct extent_state *cached_state = NULL;
3506         char *kaddr;
3507         u32 blocksize = root->sectorsize;
3508         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3509         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3510         struct page *page;
3511         int ret = 0;
3512         u64 page_start;
3513         u64 page_end;
3514
3515         if ((offset & (blocksize - 1)) == 0)
3516                 goto out;
3517         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3518         if (ret)
3519                 goto out;
3520
3521         ret = -ENOMEM;
3522 again:
3523         page = grab_cache_page(mapping, index);
3524         if (!page) {
3525                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3526                 goto out;
3527         }
3528
3529         page_start = page_offset(page);
3530         page_end = page_start + PAGE_CACHE_SIZE - 1;
3531
3532         if (!PageUptodate(page)) {
3533                 ret = btrfs_readpage(NULL, page);
3534                 lock_page(page);
3535                 if (page->mapping != mapping) {
3536                         unlock_page(page);
3537                         page_cache_release(page);
3538                         goto again;
3539                 }
3540                 if (!PageUptodate(page)) {
3541                         ret = -EIO;
3542                         goto out_unlock;
3543                 }
3544         }
3545         wait_on_page_writeback(page);
3546
3547         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3548                          GFP_NOFS);
3549         set_page_extent_mapped(page);
3550
3551         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3552         if (ordered) {
3553                 unlock_extent_cached(io_tree, page_start, page_end,
3554                                      &cached_state, GFP_NOFS);
3555                 unlock_page(page);
3556                 page_cache_release(page);
3557                 btrfs_start_ordered_extent(inode, ordered, 1);
3558                 btrfs_put_ordered_extent(ordered);
3559                 goto again;
3560         }
3561
3562         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3563                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3564                           0, 0, &cached_state, GFP_NOFS);
3565
3566         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3567                                         &cached_state);
3568         if (ret) {
3569                 unlock_extent_cached(io_tree, page_start, page_end,
3570                                      &cached_state, GFP_NOFS);
3571                 goto out_unlock;
3572         }
3573
3574         ret = 0;
3575         if (offset != PAGE_CACHE_SIZE) {
3576                 kaddr = kmap(page);
3577                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3578                 flush_dcache_page(page);
3579                 kunmap(page);
3580         }
3581         ClearPageChecked(page);
3582         set_page_dirty(page);
3583         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3584                              GFP_NOFS);
3585
3586 out_unlock:
3587         if (ret)
3588                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3589         unlock_page(page);
3590         page_cache_release(page);
3591 out:
3592         return ret;
3593 }
3594
3595 /*
3596  * This function puts in dummy file extents for the area we're creating a hole
3597  * for.  So if we are truncating this file to a larger size we need to insert
3598  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3599  * the range between oldsize and size
3600  */
3601 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3602 {
3603         struct btrfs_trans_handle *trans;
3604         struct btrfs_root *root = BTRFS_I(inode)->root;
3605         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3606         struct extent_map *em = NULL;
3607         struct extent_state *cached_state = NULL;
3608         u64 mask = root->sectorsize - 1;
3609         u64 hole_start = (oldsize + mask) & ~mask;
3610         u64 block_end = (size + mask) & ~mask;
3611         u64 last_byte;
3612         u64 cur_offset;
3613         u64 hole_size;
3614         int err = 0;
3615
3616         if (size <= hole_start)
3617                 return 0;
3618
3619         while (1) {
3620                 struct btrfs_ordered_extent *ordered;
3621                 btrfs_wait_ordered_range(inode, hole_start,
3622                                          block_end - hole_start);
3623                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3624                                  &cached_state, GFP_NOFS);
3625                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3626                 if (!ordered)
3627                         break;
3628                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3629                                      &cached_state, GFP_NOFS);
3630                 btrfs_put_ordered_extent(ordered);
3631         }
3632
3633         cur_offset = hole_start;
3634         while (1) {
3635                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3636                                 block_end - cur_offset, 0);
3637                 BUG_ON(IS_ERR(em) || !em);
3638                 last_byte = min(extent_map_end(em), block_end);
3639                 last_byte = (last_byte + mask) & ~mask;
3640                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3641                         u64 hint_byte = 0;
3642                         hole_size = last_byte - cur_offset;
3643
3644                         trans = btrfs_start_transaction(root, 2);
3645                         if (IS_ERR(trans)) {
3646                                 err = PTR_ERR(trans);
3647                                 break;
3648                         }
3649                         btrfs_set_trans_block_group(trans, inode);
3650
3651                         err = btrfs_drop_extents(trans, inode, cur_offset,
3652                                                  cur_offset + hole_size,
3653                                                  &hint_byte, 1);
3654                         if (err)
3655                                 break;
3656
3657                         err = btrfs_insert_file_extent(trans, root,
3658                                         inode->i_ino, cur_offset, 0,
3659                                         0, hole_size, 0, hole_size,
3660                                         0, 0, 0);
3661                         if (err)
3662                                 break;
3663
3664                         btrfs_drop_extent_cache(inode, hole_start,
3665                                         last_byte - 1, 0);
3666
3667                         btrfs_end_transaction(trans, root);
3668                 }
3669                 free_extent_map(em);
3670                 em = NULL;
3671                 cur_offset = last_byte;
3672                 if (cur_offset >= block_end)
3673                         break;
3674         }
3675
3676         free_extent_map(em);
3677         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3678                              GFP_NOFS);
3679         return err;
3680 }
3681
3682 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3683 {
3684         loff_t oldsize = i_size_read(inode);
3685         int ret;
3686
3687         if (newsize == oldsize)
3688                 return 0;
3689
3690         if (newsize > oldsize) {
3691                 i_size_write(inode, newsize);
3692                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3693                 truncate_pagecache(inode, oldsize, newsize);
3694                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3695                 if (ret) {
3696                         btrfs_setsize(inode, oldsize);
3697                         return ret;
3698                 }
3699
3700                 mark_inode_dirty(inode);
3701         } else {
3702
3703                 /*
3704                  * We're truncating a file that used to have good data down to
3705                  * zero. Make sure it gets into the ordered flush list so that
3706                  * any new writes get down to disk quickly.
3707                  */
3708                 if (newsize == 0)
3709                         BTRFS_I(inode)->ordered_data_close = 1;
3710
3711                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3712                 truncate_setsize(inode, newsize);
3713                 ret = btrfs_truncate(inode);
3714         }
3715
3716         return ret;
3717 }
3718
3719 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3720 {
3721         struct inode *inode = dentry->d_inode;
3722         struct btrfs_root *root = BTRFS_I(inode)->root;
3723         int err;
3724
3725         if (btrfs_root_readonly(root))
3726                 return -EROFS;
3727
3728         err = inode_change_ok(inode, attr);
3729         if (err)
3730                 return err;
3731
3732         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3733                 err = btrfs_setsize(inode, attr->ia_size);
3734                 if (err)
3735                         return err;
3736         }
3737
3738         if (attr->ia_valid) {
3739                 setattr_copy(inode, attr);
3740                 mark_inode_dirty(inode);
3741
3742                 if (attr->ia_valid & ATTR_MODE)
3743                         err = btrfs_acl_chmod(inode);
3744         }
3745
3746         return err;
3747 }
3748
3749 void btrfs_evict_inode(struct inode *inode)
3750 {
3751         struct btrfs_trans_handle *trans;
3752         struct btrfs_root *root = BTRFS_I(inode)->root;
3753         unsigned long nr;
3754         int ret;
3755
3756         trace_btrfs_inode_evict(inode);
3757
3758         truncate_inode_pages(&inode->i_data, 0);
3759         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3760                                root == root->fs_info->tree_root))
3761                 goto no_delete;
3762
3763         if (is_bad_inode(inode)) {
3764                 btrfs_orphan_del(NULL, inode);
3765                 goto no_delete;
3766         }
3767         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3768         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3769
3770         if (root->fs_info->log_root_recovering) {
3771                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3772                 goto no_delete;
3773         }
3774
3775         if (inode->i_nlink > 0) {
3776                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3777                 goto no_delete;
3778         }
3779
3780         btrfs_i_size_write(inode, 0);
3781
3782         while (1) {
3783                 trans = btrfs_start_transaction(root, 0);
3784                 BUG_ON(IS_ERR(trans));
3785                 btrfs_set_trans_block_group(trans, inode);
3786                 trans->block_rsv = root->orphan_block_rsv;
3787
3788                 ret = btrfs_block_rsv_check(trans, root,
3789                                             root->orphan_block_rsv, 0, 5);
3790                 if (ret) {
3791                         BUG_ON(ret != -EAGAIN);
3792                         ret = btrfs_commit_transaction(trans, root);
3793                         BUG_ON(ret);
3794                         continue;
3795                 }
3796
3797                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3798                 if (ret != -EAGAIN)
3799                         break;
3800
3801                 nr = trans->blocks_used;
3802                 btrfs_end_transaction(trans, root);
3803                 trans = NULL;
3804                 btrfs_btree_balance_dirty(root, nr);
3805
3806         }
3807
3808         if (ret == 0) {
3809                 ret = btrfs_orphan_del(trans, inode);
3810                 BUG_ON(ret);
3811         }
3812
3813         nr = trans->blocks_used;
3814         btrfs_end_transaction(trans, root);
3815         btrfs_btree_balance_dirty(root, nr);
3816 no_delete:
3817         end_writeback(inode);
3818         return;
3819 }
3820
3821 /*
3822  * this returns the key found in the dir entry in the location pointer.
3823  * If no dir entries were found, location->objectid is 0.
3824  */
3825 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3826                                struct btrfs_key *location)
3827 {
3828         const char *name = dentry->d_name.name;
3829         int namelen = dentry->d_name.len;
3830         struct btrfs_dir_item *di;
3831         struct btrfs_path *path;
3832         struct btrfs_root *root = BTRFS_I(dir)->root;
3833         int ret = 0;
3834
3835         path = btrfs_alloc_path();
3836         BUG_ON(!path);
3837
3838         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3839                                     namelen, 0);
3840         if (IS_ERR(di))
3841                 ret = PTR_ERR(di);
3842
3843         if (!di || IS_ERR(di))
3844                 goto out_err;
3845
3846         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3847 out:
3848         btrfs_free_path(path);
3849         return ret;
3850 out_err:
3851         location->objectid = 0;
3852         goto out;
3853 }
3854
3855 /*
3856  * when we hit a tree root in a directory, the btrfs part of the inode
3857  * needs to be changed to reflect the root directory of the tree root.  This
3858  * is kind of like crossing a mount point.
3859  */
3860 static int fixup_tree_root_location(struct btrfs_root *root,
3861                                     struct inode *dir,
3862                                     struct dentry *dentry,
3863                                     struct btrfs_key *location,
3864                                     struct btrfs_root **sub_root)
3865 {
3866         struct btrfs_path *path;
3867         struct btrfs_root *new_root;
3868         struct btrfs_root_ref *ref;
3869         struct extent_buffer *leaf;
3870         int ret;
3871         int err = 0;
3872
3873         path = btrfs_alloc_path();
3874         if (!path) {
3875                 err = -ENOMEM;
3876                 goto out;
3877         }
3878
3879         err = -ENOENT;
3880         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3881                                   BTRFS_I(dir)->root->root_key.objectid,
3882                                   location->objectid);
3883         if (ret) {
3884                 if (ret < 0)
3885                         err = ret;
3886                 goto out;
3887         }
3888
3889         leaf = path->nodes[0];
3890         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3891         if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3892             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3893                 goto out;
3894
3895         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3896                                    (unsigned long)(ref + 1),
3897                                    dentry->d_name.len);
3898         if (ret)
3899                 goto out;
3900
3901         btrfs_release_path(root->fs_info->tree_root, path);
3902
3903         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3904         if (IS_ERR(new_root)) {
3905                 err = PTR_ERR(new_root);
3906                 goto out;
3907         }
3908
3909         if (btrfs_root_refs(&new_root->root_item) == 0) {
3910                 err = -ENOENT;
3911                 goto out;
3912         }
3913
3914         *sub_root = new_root;
3915         location->objectid = btrfs_root_dirid(&new_root->root_item);
3916         location->type = BTRFS_INODE_ITEM_KEY;
3917         location->offset = 0;
3918         err = 0;
3919 out:
3920         btrfs_free_path(path);
3921         return err;
3922 }
3923
3924 static void inode_tree_add(struct inode *inode)
3925 {
3926         struct btrfs_root *root = BTRFS_I(inode)->root;
3927         struct btrfs_inode *entry;
3928         struct rb_node **p;
3929         struct rb_node *parent;
3930 again:
3931         p = &root->inode_tree.rb_node;
3932         parent = NULL;
3933
3934         if (inode_unhashed(inode))
3935                 return;
3936
3937         spin_lock(&root->inode_lock);
3938         while (*p) {
3939                 parent = *p;
3940                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3941
3942                 if (inode->i_ino < entry->vfs_inode.i_ino)
3943                         p = &parent->rb_left;
3944                 else if (inode->i_ino > entry->vfs_inode.i_ino)
3945                         p = &parent->rb_right;
3946                 else {
3947                         WARN_ON(!(entry->vfs_inode.i_state &
3948                                   (I_WILL_FREE | I_FREEING)));
3949                         rb_erase(parent, &root->inode_tree);
3950                         RB_CLEAR_NODE(parent);
3951                         spin_unlock(&root->inode_lock);
3952                         goto again;
3953                 }
3954         }
3955         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3956         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3957         spin_unlock(&root->inode_lock);
3958 }
3959
3960 static void inode_tree_del(struct inode *inode)
3961 {
3962         struct btrfs_root *root = BTRFS_I(inode)->root;
3963         int empty = 0;
3964
3965         spin_lock(&root->inode_lock);
3966         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3967                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3968                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3969                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3970         }
3971         spin_unlock(&root->inode_lock);
3972
3973         /*
3974          * Free space cache has inodes in the tree root, but the tree root has a
3975          * root_refs of 0, so this could end up dropping the tree root as a
3976          * snapshot, so we need the extra !root->fs_info->tree_root check to
3977          * make sure we don't drop it.
3978          */
3979         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3980             root != root->fs_info->tree_root) {
3981                 synchronize_srcu(&root->fs_info->subvol_srcu);
3982                 spin_lock(&root->inode_lock);
3983                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3984                 spin_unlock(&root->inode_lock);
3985                 if (empty)
3986                         btrfs_add_dead_root(root);
3987         }
3988 }
3989
3990 int btrfs_invalidate_inodes(struct btrfs_root *root)
3991 {
3992         struct rb_node *node;
3993         struct rb_node *prev;
3994         struct btrfs_inode *entry;
3995         struct inode *inode;
3996         u64 objectid = 0;
3997
3998         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3999
4000         spin_lock(&root->inode_lock);
4001 again:
4002         node = root->inode_tree.rb_node;
4003         prev = NULL;
4004         while (node) {
4005                 prev = node;
4006                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4007
4008                 if (objectid < entry->vfs_inode.i_ino)
4009                         node = node->rb_left;
4010                 else if (objectid > entry->vfs_inode.i_ino)
4011                         node = node->rb_right;
4012                 else
4013                         break;
4014         }
4015         if (!node) {
4016                 while (prev) {
4017                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4018                         if (objectid <= entry->vfs_inode.i_ino) {
4019                                 node = prev;
4020                                 break;
4021                         }
4022                         prev = rb_next(prev);
4023                 }
4024         }
4025         while (node) {
4026                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4027                 objectid = entry->vfs_inode.i_ino + 1;
4028                 inode = igrab(&entry->vfs_inode);
4029                 if (inode) {
4030                         spin_unlock(&root->inode_lock);
4031                         if (atomic_read(&inode->i_count) > 1)
4032                                 d_prune_aliases(inode);
4033                         /*
4034                          * btrfs_drop_inode will have it removed from
4035                          * the inode cache when its usage count
4036                          * hits zero.
4037                          */
4038                         iput(inode);
4039                         cond_resched();
4040                         spin_lock(&root->inode_lock);
4041                         goto again;
4042                 }
4043
4044                 if (cond_resched_lock(&root->inode_lock))
4045                         goto again;
4046
4047                 node = rb_next(node);
4048         }
4049         spin_unlock(&root->inode_lock);
4050         return 0;
4051 }
4052
4053 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4054 {
4055         struct btrfs_iget_args *args = p;
4056         inode->i_ino = args->ino;
4057         BTRFS_I(inode)->root = args->root;
4058         btrfs_set_inode_space_info(args->root, inode);
4059         return 0;
4060 }
4061
4062 static int btrfs_find_actor(struct inode *inode, void *opaque)
4063 {
4064         struct btrfs_iget_args *args = opaque;
4065         return args->ino == inode->i_ino &&
4066                 args->root == BTRFS_I(inode)->root;
4067 }
4068
4069 static struct inode *btrfs_iget_locked(struct super_block *s,
4070                                        u64 objectid,
4071                                        struct btrfs_root *root)
4072 {
4073         struct inode *inode;
4074         struct btrfs_iget_args args;
4075         args.ino = objectid;
4076         args.root = root;
4077
4078         inode = iget5_locked(s, objectid, btrfs_find_actor,
4079                              btrfs_init_locked_inode,
4080                              (void *)&args);
4081         return inode;
4082 }
4083
4084 /* Get an inode object given its location and corresponding root.
4085  * Returns in *is_new if the inode was read from disk
4086  */
4087 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4088                          struct btrfs_root *root, int *new)
4089 {
4090         struct inode *inode;
4091
4092         inode = btrfs_iget_locked(s, location->objectid, root);
4093         if (!inode)
4094                 return ERR_PTR(-ENOMEM);
4095
4096         if (inode->i_state & I_NEW) {
4097                 BTRFS_I(inode)->root = root;
4098                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4099                 btrfs_read_locked_inode(inode);
4100                 inode_tree_add(inode);
4101                 unlock_new_inode(inode);
4102                 if (new)
4103                         *new = 1;
4104         }
4105
4106         return inode;
4107 }
4108
4109 static struct inode *new_simple_dir(struct super_block *s,
4110                                     struct btrfs_key *key,
4111                                     struct btrfs_root *root)
4112 {
4113         struct inode *inode = new_inode(s);
4114
4115         if (!inode)
4116                 return ERR_PTR(-ENOMEM);
4117
4118         BTRFS_I(inode)->root = root;
4119         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4120         BTRFS_I(inode)->dummy_inode = 1;
4121
4122         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4123         inode->i_op = &simple_dir_inode_operations;
4124         inode->i_fop = &simple_dir_operations;
4125         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4126         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4127
4128         return inode;
4129 }
4130
4131 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4132 {
4133         struct inode *inode;
4134         struct btrfs_root *root = BTRFS_I(dir)->root;
4135         struct btrfs_root *sub_root = root;
4136         struct btrfs_key location;
4137         int index;
4138         int ret;
4139
4140         if (dentry->d_name.len > BTRFS_NAME_LEN)
4141                 return ERR_PTR(-ENAMETOOLONG);
4142
4143         ret = btrfs_inode_by_name(dir, dentry, &location);
4144
4145         if (ret < 0)
4146                 return ERR_PTR(ret);
4147
4148         if (location.objectid == 0)
4149                 return NULL;
4150
4151         if (location.type == BTRFS_INODE_ITEM_KEY) {
4152                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4153                 return inode;
4154         }
4155
4156         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4157
4158         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4159         ret = fixup_tree_root_location(root, dir, dentry,
4160                                        &location, &sub_root);
4161         if (ret < 0) {
4162                 if (ret != -ENOENT)
4163                         inode = ERR_PTR(ret);
4164                 else
4165                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4166         } else {
4167                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4168         }
4169         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4170
4171         if (!IS_ERR(inode) && root != sub_root) {
4172                 down_read(&root->fs_info->cleanup_work_sem);
4173                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4174                         ret = btrfs_orphan_cleanup(sub_root);
4175                 up_read(&root->fs_info->cleanup_work_sem);
4176                 if (ret)
4177                         inode = ERR_PTR(ret);
4178         }
4179
4180         return inode;
4181 }
4182
4183 static int btrfs_dentry_delete(const struct dentry *dentry)
4184 {
4185         struct btrfs_root *root;
4186
4187         if (!dentry->d_inode && !IS_ROOT(dentry))
4188                 dentry = dentry->d_parent;
4189
4190         if (dentry->d_inode) {
4191                 root = BTRFS_I(dentry->d_inode)->root;
4192                 if (btrfs_root_refs(&root->root_item) == 0)
4193                         return 1;
4194         }
4195         return 0;
4196 }
4197
4198 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4199                                    struct nameidata *nd)
4200 {
4201         struct inode *inode;
4202
4203         inode = btrfs_lookup_dentry(dir, dentry);
4204         if (IS_ERR(inode))
4205                 return ERR_CAST(inode);
4206
4207         return d_splice_alias(inode, dentry);
4208 }
4209
4210 static unsigned char btrfs_filetype_table[] = {
4211         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4212 };
4213
4214 static int btrfs_real_readdir(struct file *filp, void *dirent,
4215                               filldir_t filldir)
4216 {
4217         struct inode *inode = filp->f_dentry->d_inode;
4218         struct btrfs_root *root = BTRFS_I(inode)->root;
4219         struct btrfs_item *item;
4220         struct btrfs_dir_item *di;
4221         struct btrfs_key key;
4222         struct btrfs_key found_key;
4223         struct btrfs_path *path;
4224         int ret;
4225         struct extent_buffer *leaf;
4226         int slot;
4227         unsigned char d_type;
4228         int over = 0;
4229         u32 di_cur;
4230         u32 di_total;
4231         u32 di_len;
4232         int key_type = BTRFS_DIR_INDEX_KEY;
4233         char tmp_name[32];
4234         char *name_ptr;
4235         int name_len;
4236
4237         /* FIXME, use a real flag for deciding about the key type */
4238         if (root->fs_info->tree_root == root)
4239                 key_type = BTRFS_DIR_ITEM_KEY;
4240
4241         /* special case for "." */
4242         if (filp->f_pos == 0) {
4243                 over = filldir(dirent, ".", 1,
4244                                1, inode->i_ino,
4245                                DT_DIR);
4246                 if (over)
4247                         return 0;
4248                 filp->f_pos = 1;
4249         }
4250         /* special case for .., just use the back ref */
4251         if (filp->f_pos == 1) {
4252                 u64 pino = parent_ino(filp->f_path.dentry);
4253                 over = filldir(dirent, "..", 2,
4254                                2, pino, DT_DIR);
4255                 if (over)
4256                         return 0;
4257                 filp->f_pos = 2;
4258         }
4259         path = btrfs_alloc_path();
4260         path->reada = 2;
4261
4262         btrfs_set_key_type(&key, key_type);
4263         key.offset = filp->f_pos;
4264         key.objectid = inode->i_ino;
4265
4266         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4267         if (ret < 0)
4268                 goto err;
4269
4270         while (1) {
4271                 leaf = path->nodes[0];
4272                 slot = path->slots[0];
4273                 if (slot >= btrfs_header_nritems(leaf)) {
4274                         ret = btrfs_next_leaf(root, path);
4275                         if (ret < 0)
4276                                 goto err;
4277                         else if (ret > 0)
4278                                 break;
4279                         continue;
4280                 }
4281
4282                 item = btrfs_item_nr(leaf, slot);
4283                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4284
4285                 if (found_key.objectid != key.objectid)
4286                         break;
4287                 if (btrfs_key_type(&found_key) != key_type)
4288                         break;
4289                 if (found_key.offset < filp->f_pos)
4290                         goto next;
4291
4292                 filp->f_pos = found_key.offset;
4293
4294                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4295                 di_cur = 0;
4296                 di_total = btrfs_item_size(leaf, item);
4297
4298                 while (di_cur < di_total) {
4299                         struct btrfs_key location;
4300
4301                         if (verify_dir_item(root, leaf, di))
4302                                 break;
4303
4304                         name_len = btrfs_dir_name_len(leaf, di);
4305                         if (name_len <= sizeof(tmp_name)) {
4306                                 name_ptr = tmp_name;
4307                         } else {
4308                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4309                                 if (!name_ptr) {
4310                                         ret = -ENOMEM;
4311                                         goto err;
4312                                 }
4313                         }
4314                         read_extent_buffer(leaf, name_ptr,
4315                                            (unsigned long)(di + 1), name_len);
4316
4317                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4318                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4319
4320                         /* is this a reference to our own snapshot? If so
4321                          * skip it
4322                          */
4323                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4324                             location.objectid == root->root_key.objectid) {
4325                                 over = 0;
4326                                 goto skip;
4327                         }
4328                         over = filldir(dirent, name_ptr, name_len,
4329                                        found_key.offset, location.objectid,
4330                                        d_type);
4331
4332 skip:
4333                         if (name_ptr != tmp_name)
4334                                 kfree(name_ptr);
4335
4336                         if (over)
4337                                 goto nopos;
4338                         di_len = btrfs_dir_name_len(leaf, di) +
4339                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4340                         di_cur += di_len;
4341                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4342                 }
4343 next:
4344                 path->slots[0]++;
4345         }
4346
4347         /* Reached end of directory/root. Bump pos past the last item. */
4348         if (key_type == BTRFS_DIR_INDEX_KEY)
4349                 /*
4350                  * 32-bit glibc will use getdents64, but then strtol -
4351                  * so the last number we can serve is this.
4352                  */
4353                 filp->f_pos = 0x7fffffff;
4354         else
4355                 filp->f_pos++;
4356 nopos:
4357         ret = 0;
4358 err:
4359         btrfs_free_path(path);
4360         return ret;
4361 }
4362
4363 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4364 {
4365         struct btrfs_root *root = BTRFS_I(inode)->root;
4366         struct btrfs_trans_handle *trans;
4367         int ret = 0;
4368         bool nolock = false;
4369
4370         if (BTRFS_I(inode)->dummy_inode)
4371                 return 0;
4372
4373         smp_mb();
4374         nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4375
4376         if (wbc->sync_mode == WB_SYNC_ALL) {
4377                 if (nolock)
4378                         trans = btrfs_join_transaction_nolock(root, 1);
4379                 else
4380                         trans = btrfs_join_transaction(root, 1);
4381                 if (IS_ERR(trans))
4382                         return PTR_ERR(trans);
4383                 btrfs_set_trans_block_group(trans, inode);
4384                 if (nolock)
4385                         ret = btrfs_end_transaction_nolock(trans, root);
4386                 else
4387                         ret = btrfs_commit_transaction(trans, root);
4388         }
4389         return ret;
4390 }
4391
4392 /*
4393  * This is somewhat expensive, updating the tree every time the
4394  * inode changes.  But, it is most likely to find the inode in cache.
4395  * FIXME, needs more benchmarking...there are no reasons other than performance
4396  * to keep or drop this code.
4397  */
4398 void btrfs_dirty_inode(struct inode *inode)
4399 {
4400         struct btrfs_root *root = BTRFS_I(inode)->root;
4401         struct btrfs_trans_handle *trans;
4402         int ret;
4403
4404         if (BTRFS_I(inode)->dummy_inode)
4405                 return;
4406
4407         trans = btrfs_join_transaction(root, 1);
4408         BUG_ON(IS_ERR(trans));
4409         btrfs_set_trans_block_group(trans, inode);
4410
4411         ret = btrfs_update_inode(trans, root, inode);
4412         if (ret && ret == -ENOSPC) {
4413                 /* whoops, lets try again with the full transaction */
4414                 btrfs_end_transaction(trans, root);
4415                 trans = btrfs_start_transaction(root, 1);
4416                 if (IS_ERR(trans)) {
4417                         if (printk_ratelimit()) {
4418                                 printk(KERN_ERR "btrfs: fail to "
4419                                        "dirty  inode %lu error %ld\n",
4420                                        inode->i_ino, PTR_ERR(trans));
4421                         }
4422                         return;
4423                 }
4424                 btrfs_set_trans_block_group(trans, inode);
4425
4426                 ret = btrfs_update_inode(trans, root, inode);
4427                 if (ret) {
4428                         if (printk_ratelimit()) {
4429                                 printk(KERN_ERR "btrfs: fail to "
4430                                        "dirty  inode %lu error %d\n",
4431                                        inode->i_ino, ret);
4432                         }
4433                 }
4434         }
4435         btrfs_end_transaction(trans, root);
4436 }
4437
4438 /*
4439  * find the highest existing sequence number in a directory
4440  * and then set the in-memory index_cnt variable to reflect
4441  * free sequence numbers
4442  */
4443 static int btrfs_set_inode_index_count(struct inode *inode)
4444 {
4445         struct btrfs_root *root = BTRFS_I(inode)->root;
4446         struct btrfs_key key, found_key;
4447         struct btrfs_path *path;
4448         struct extent_buffer *leaf;
4449         int ret;
4450
4451         key.objectid = inode->i_ino;
4452         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4453         key.offset = (u64)-1;
4454
4455         path = btrfs_alloc_path();
4456         if (!path)
4457                 return -ENOMEM;
4458
4459         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4460         if (ret < 0)
4461                 goto out;
4462         /* FIXME: we should be able to handle this */
4463         if (ret == 0)
4464                 goto out;
4465         ret = 0;
4466
4467         /*
4468          * MAGIC NUMBER EXPLANATION:
4469          * since we search a directory based on f_pos we have to start at 2
4470          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4471          * else has to start at 2
4472          */
4473         if (path->slots[0] == 0) {
4474                 BTRFS_I(inode)->index_cnt = 2;
4475                 goto out;
4476         }
4477
4478         path->slots[0]--;
4479
4480         leaf = path->nodes[0];
4481         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4482
4483         if (found_key.objectid != inode->i_ino ||
4484             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4485                 BTRFS_I(inode)->index_cnt = 2;
4486                 goto out;
4487         }
4488
4489         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4490 out:
4491         btrfs_free_path(path);
4492         return ret;
4493 }
4494
4495 /*
4496  * helper to find a free sequence number in a given directory.  This current
4497  * code is very simple, later versions will do smarter things in the btree
4498  */
4499 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4500 {
4501         int ret = 0;
4502
4503         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4504                 ret = btrfs_set_inode_index_count(dir);
4505                 if (ret)
4506                         return ret;
4507         }
4508
4509         *index = BTRFS_I(dir)->index_cnt;
4510         BTRFS_I(dir)->index_cnt++;
4511
4512         return ret;
4513 }
4514
4515 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4516                                      struct btrfs_root *root,
4517                                      struct inode *dir,
4518                                      const char *name, int name_len,
4519                                      u64 ref_objectid, u64 objectid,
4520                                      u64 alloc_hint, int mode, u64 *index)
4521 {
4522         struct inode *inode;
4523         struct btrfs_inode_item *inode_item;
4524         struct btrfs_key *location;
4525         struct btrfs_path *path;
4526         struct btrfs_inode_ref *ref;
4527         struct btrfs_key key[2];
4528         u32 sizes[2];
4529         unsigned long ptr;
4530         int ret;
4531         int owner;
4532
4533         path = btrfs_alloc_path();
4534         BUG_ON(!path);
4535
4536         inode = new_inode(root->fs_info->sb);
4537         if (!inode) {
4538                 btrfs_free_path(path);
4539                 return ERR_PTR(-ENOMEM);
4540         }
4541
4542         if (dir) {
4543                 trace_btrfs_inode_request(dir);
4544
4545                 ret = btrfs_set_inode_index(dir, index);
4546                 if (ret) {
4547                         btrfs_free_path(path);
4548                         iput(inode);
4549                         return ERR_PTR(ret);
4550                 }
4551         }
4552         /*
4553          * index_cnt is ignored for everything but a dir,
4554          * btrfs_get_inode_index_count has an explanation for the magic
4555          * number
4556          */
4557         BTRFS_I(inode)->index_cnt = 2;
4558         BTRFS_I(inode)->root = root;
4559         BTRFS_I(inode)->generation = trans->transid;
4560         inode->i_generation = BTRFS_I(inode)->generation;
4561         btrfs_set_inode_space_info(root, inode);
4562
4563         if (mode & S_IFDIR)
4564                 owner = 0;
4565         else
4566                 owner = 1;
4567         BTRFS_I(inode)->block_group =
4568                         btrfs_find_block_group(root, 0, alloc_hint, owner);
4569
4570         key[0].objectid = objectid;
4571         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4572         key[0].offset = 0;
4573
4574         key[1].objectid = objectid;
4575         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4576         key[1].offset = ref_objectid;
4577
4578         sizes[0] = sizeof(struct btrfs_inode_item);
4579         sizes[1] = name_len + sizeof(*ref);
4580
4581         path->leave_spinning = 1;
4582         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4583         if (ret != 0)
4584                 goto fail;
4585
4586         inode_init_owner(inode, dir, mode);
4587         inode->i_ino = objectid;
4588         inode_set_bytes(inode, 0);
4589         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4590         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4591                                   struct btrfs_inode_item);
4592         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4593
4594         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4595                              struct btrfs_inode_ref);
4596         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4597         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4598         ptr = (unsigned long)(ref + 1);
4599         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4600
4601         btrfs_mark_buffer_dirty(path->nodes[0]);
4602         btrfs_free_path(path);
4603
4604         location = &BTRFS_I(inode)->location;
4605         location->objectid = objectid;
4606         location->offset = 0;
4607         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4608
4609         btrfs_inherit_iflags(inode, dir);
4610
4611         if ((mode & S_IFREG)) {
4612                 if (btrfs_test_opt(root, NODATASUM))
4613                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4614                 if (btrfs_test_opt(root, NODATACOW) ||
4615                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4616                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4617         }
4618
4619         insert_inode_hash(inode);
4620         inode_tree_add(inode);
4621
4622         trace_btrfs_inode_new(inode);
4623
4624         return inode;
4625 fail:
4626         if (dir)
4627                 BTRFS_I(dir)->index_cnt--;
4628         btrfs_free_path(path);
4629         iput(inode);
4630         return ERR_PTR(ret);
4631 }
4632
4633 static inline u8 btrfs_inode_type(struct inode *inode)
4634 {
4635         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4636 }
4637
4638 /*
4639  * utility function to add 'inode' into 'parent_inode' with
4640  * a give name and a given sequence number.
4641  * if 'add_backref' is true, also insert a backref from the
4642  * inode to the parent directory.
4643  */
4644 int btrfs_add_link(struct btrfs_trans_handle *trans,
4645                    struct inode *parent_inode, struct inode *inode,
4646                    const char *name, int name_len, int add_backref, u64 index)
4647 {
4648         int ret = 0;
4649         struct btrfs_key key;
4650         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4651
4652         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4653                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4654         } else {
4655                 key.objectid = inode->i_ino;
4656                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4657                 key.offset = 0;
4658         }
4659
4660         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4661                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4662                                          key.objectid, root->root_key.objectid,
4663                                          parent_inode->i_ino,
4664                                          index, name, name_len);
4665         } else if (add_backref) {
4666                 ret = btrfs_insert_inode_ref(trans, root,
4667                                              name, name_len, inode->i_ino,
4668                                              parent_inode->i_ino, index);
4669         }
4670
4671         if (ret == 0) {
4672                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4673                                             parent_inode->i_ino, &key,
4674                                             btrfs_inode_type(inode), index);
4675                 BUG_ON(ret);
4676
4677                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4678                                    name_len * 2);
4679                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4680                 ret = btrfs_update_inode(trans, root, parent_inode);
4681         }
4682         return ret;
4683 }
4684
4685 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4686                             struct inode *dir, struct dentry *dentry,
4687                             struct inode *inode, int backref, u64 index)
4688 {
4689         int err = btrfs_add_link(trans, dir, inode,
4690                                  dentry->d_name.name, dentry->d_name.len,
4691                                  backref, index);
4692         if (!err) {
4693                 d_instantiate(dentry, inode);
4694                 return 0;
4695         }
4696         if (err > 0)
4697                 err = -EEXIST;
4698         return err;
4699 }
4700
4701 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4702                         int mode, dev_t rdev)
4703 {
4704         struct btrfs_trans_handle *trans;
4705         struct btrfs_root *root = BTRFS_I(dir)->root;
4706         struct inode *inode = NULL;
4707         int err;
4708         int drop_inode = 0;
4709         u64 objectid;
4710         unsigned long nr = 0;
4711         u64 index = 0;
4712
4713         if (!new_valid_dev(rdev))
4714                 return -EINVAL;
4715
4716         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4717         if (err)
4718                 return err;
4719
4720         /*
4721          * 2 for inode item and ref
4722          * 2 for dir items
4723          * 1 for xattr if selinux is on
4724          */
4725         trans = btrfs_start_transaction(root, 5);
4726         if (IS_ERR(trans))
4727                 return PTR_ERR(trans);
4728
4729         btrfs_set_trans_block_group(trans, dir);
4730
4731         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4732                                 dentry->d_name.len, dir->i_ino, objectid,
4733                                 BTRFS_I(dir)->block_group, mode, &index);
4734         if (IS_ERR(inode)) {
4735                 err = PTR_ERR(inode);
4736                 goto out_unlock;
4737         }
4738
4739         err = btrfs_init_inode_security(trans, inode, dir);
4740         if (err) {
4741                 drop_inode = 1;
4742                 goto out_unlock;
4743         }
4744
4745         btrfs_set_trans_block_group(trans, inode);
4746         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4747         if (err)
4748                 drop_inode = 1;
4749         else {
4750                 inode->i_op = &btrfs_special_inode_operations;
4751                 init_special_inode(inode, inode->i_mode, rdev);
4752                 btrfs_update_inode(trans, root, inode);
4753         }
4754         btrfs_update_inode_block_group(trans, inode);
4755         btrfs_update_inode_block_group(trans, dir);
4756 out_unlock:
4757         nr = trans->blocks_used;
4758         btrfs_end_transaction_throttle(trans, root);
4759         btrfs_btree_balance_dirty(root, nr);
4760         if (drop_inode) {
4761                 inode_dec_link_count(inode);
4762                 iput(inode);
4763         }
4764         return err;
4765 }
4766
4767 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4768                         int mode, struct nameidata *nd)
4769 {
4770         struct btrfs_trans_handle *trans;
4771         struct btrfs_root *root = BTRFS_I(dir)->root;
4772         struct inode *inode = NULL;
4773         int drop_inode = 0;
4774         int err;
4775         unsigned long nr = 0;
4776         u64 objectid;
4777         u64 index = 0;
4778
4779         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4780         if (err)
4781                 return err;
4782         /*
4783          * 2 for inode item and ref
4784          * 2 for dir items
4785          * 1 for xattr if selinux is on
4786          */
4787         trans = btrfs_start_transaction(root, 5);
4788         if (IS_ERR(trans))
4789                 return PTR_ERR(trans);
4790
4791         btrfs_set_trans_block_group(trans, dir);
4792
4793         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4794                                 dentry->d_name.len, dir->i_ino, objectid,
4795                                 BTRFS_I(dir)->block_group, mode, &index);
4796         if (IS_ERR(inode)) {
4797                 err = PTR_ERR(inode);
4798                 goto out_unlock;
4799         }
4800
4801         err = btrfs_init_inode_security(trans, inode, dir);
4802         if (err) {
4803                 drop_inode = 1;
4804                 goto out_unlock;
4805         }
4806
4807         btrfs_set_trans_block_group(trans, inode);
4808         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4809         if (err)
4810                 drop_inode = 1;
4811         else {
4812                 inode->i_mapping->a_ops = &btrfs_aops;
4813                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4814                 inode->i_fop = &btrfs_file_operations;
4815                 inode->i_op = &btrfs_file_inode_operations;
4816                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4817         }
4818         btrfs_update_inode_block_group(trans, inode);
4819         btrfs_update_inode_block_group(trans, dir);
4820 out_unlock:
4821         nr = trans->blocks_used;
4822         btrfs_end_transaction_throttle(trans, root);
4823         if (drop_inode) {
4824                 inode_dec_link_count(inode);
4825                 iput(inode);
4826         }
4827         btrfs_btree_balance_dirty(root, nr);
4828         return err;
4829 }
4830
4831 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4832                       struct dentry *dentry)
4833 {
4834         struct btrfs_trans_handle *trans;
4835         struct btrfs_root *root = BTRFS_I(dir)->root;
4836         struct inode *inode = old_dentry->d_inode;
4837         u64 index;
4838         unsigned long nr = 0;
4839         int err;
4840         int drop_inode = 0;
4841
4842         if (inode->i_nlink == 0)
4843                 return -ENOENT;
4844
4845         /* do not allow sys_link's with other subvols of the same device */
4846         if (root->objectid != BTRFS_I(inode)->root->objectid)
4847                 return -EXDEV;
4848
4849         if (inode->i_nlink == ~0U)
4850                 return -EMLINK;
4851
4852         err = btrfs_set_inode_index(dir, &index);
4853         if (err)
4854                 goto fail;
4855
4856         /*
4857          * 2 items for inode and inode ref
4858          * 2 items for dir items
4859          * 1 item for parent inode
4860          */
4861         trans = btrfs_start_transaction(root, 5);
4862         if (IS_ERR(trans)) {
4863                 err = PTR_ERR(trans);
4864                 goto fail;
4865         }
4866
4867         btrfs_inc_nlink(inode);
4868         inode->i_ctime = CURRENT_TIME;
4869
4870         btrfs_set_trans_block_group(trans, dir);
4871         ihold(inode);
4872
4873         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4874
4875         if (err) {
4876                 drop_inode = 1;
4877         } else {
4878                 struct dentry *parent = dget_parent(dentry);
4879                 btrfs_update_inode_block_group(trans, dir);
4880                 err = btrfs_update_inode(trans, root, inode);
4881                 BUG_ON(err);
4882                 btrfs_log_new_name(trans, inode, NULL, parent);
4883                 dput(parent);
4884         }
4885
4886         nr = trans->blocks_used;
4887         btrfs_end_transaction_throttle(trans, root);
4888 fail:
4889         if (drop_inode) {
4890                 inode_dec_link_count(inode);
4891                 iput(inode);
4892         }
4893         btrfs_btree_balance_dirty(root, nr);
4894         return err;
4895 }
4896
4897 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4898 {
4899         struct inode *inode = NULL;
4900         struct btrfs_trans_handle *trans;
4901         struct btrfs_root *root = BTRFS_I(dir)->root;
4902         int err = 0;
4903         int drop_on_err = 0;
4904         u64 objectid = 0;
4905         u64 index = 0;
4906         unsigned long nr = 1;
4907
4908         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4909         if (err)
4910                 return err;
4911
4912         /*
4913          * 2 items for inode and ref
4914          * 2 items for dir items
4915          * 1 for xattr if selinux is on
4916          */
4917         trans = btrfs_start_transaction(root, 5);
4918         if (IS_ERR(trans))
4919                 return PTR_ERR(trans);
4920         btrfs_set_trans_block_group(trans, dir);
4921
4922         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4923                                 dentry->d_name.len, dir->i_ino, objectid,
4924                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4925                                 &index);
4926         if (IS_ERR(inode)) {
4927                 err = PTR_ERR(inode);
4928                 goto out_fail;
4929         }
4930
4931         drop_on_err = 1;
4932
4933         err = btrfs_init_inode_security(trans, inode, dir);
4934         if (err)
4935                 goto out_fail;
4936
4937         inode->i_op = &btrfs_dir_inode_operations;
4938         inode->i_fop = &btrfs_dir_file_operations;
4939         btrfs_set_trans_block_group(trans, inode);
4940
4941         btrfs_i_size_write(inode, 0);
4942         err = btrfs_update_inode(trans, root, inode);
4943         if (err)
4944                 goto out_fail;
4945
4946         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4947                              dentry->d_name.len, 0, index);
4948         if (err)
4949                 goto out_fail;
4950
4951         d_instantiate(dentry, inode);
4952         drop_on_err = 0;
4953         btrfs_update_inode_block_group(trans, inode);
4954         btrfs_update_inode_block_group(trans, dir);
4955
4956 out_fail:
4957         nr = trans->blocks_used;
4958         btrfs_end_transaction_throttle(trans, root);
4959         if (drop_on_err)
4960                 iput(inode);
4961         btrfs_btree_balance_dirty(root, nr);
4962         return err;
4963 }
4964
4965 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4966  * and an extent that you want to insert, deal with overlap and insert
4967  * the new extent into the tree.
4968  */
4969 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4970                                 struct extent_map *existing,
4971                                 struct extent_map *em,
4972                                 u64 map_start, u64 map_len)
4973 {
4974         u64 start_diff;
4975
4976         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4977         start_diff = map_start - em->start;
4978         em->start = map_start;
4979         em->len = map_len;
4980         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4981             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4982                 em->block_start += start_diff;
4983                 em->block_len -= start_diff;
4984         }
4985         return add_extent_mapping(em_tree, em);
4986 }
4987
4988 static noinline int uncompress_inline(struct btrfs_path *path,
4989                                       struct inode *inode, struct page *page,
4990                                       size_t pg_offset, u64 extent_offset,
4991                                       struct btrfs_file_extent_item *item)
4992 {
4993         int ret;
4994         struct extent_buffer *leaf = path->nodes[0];
4995         char *tmp;
4996         size_t max_size;
4997         unsigned long inline_size;
4998         unsigned long ptr;
4999         int compress_type;
5000
5001         WARN_ON(pg_offset != 0);
5002         compress_type = btrfs_file_extent_compression(leaf, item);
5003         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5004         inline_size = btrfs_file_extent_inline_item_len(leaf,
5005                                         btrfs_item_nr(leaf, path->slots[0]));
5006         tmp = kmalloc(inline_size, GFP_NOFS);
5007         if (!tmp)
5008                 return -ENOMEM;
5009         ptr = btrfs_file_extent_inline_start(item);
5010
5011         read_extent_buffer(leaf, tmp, ptr, inline_size);
5012
5013         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5014         ret = btrfs_decompress(compress_type, tmp, page,
5015                                extent_offset, inline_size, max_size);
5016         if (ret) {
5017                 char *kaddr = kmap_atomic(page, KM_USER0);
5018                 unsigned long copy_size = min_t(u64,
5019                                   PAGE_CACHE_SIZE - pg_offset,
5020                                   max_size - extent_offset);
5021                 memset(kaddr + pg_offset, 0, copy_size);
5022                 kunmap_atomic(kaddr, KM_USER0);
5023         }
5024         kfree(tmp);
5025         return 0;
5026 }
5027
5028 /*
5029  * a bit scary, this does extent mapping from logical file offset to the disk.
5030  * the ugly parts come from merging extents from the disk with the in-ram
5031  * representation.  This gets more complex because of the data=ordered code,
5032  * where the in-ram extents might be locked pending data=ordered completion.
5033  *
5034  * This also copies inline extents directly into the page.
5035  */
5036
5037 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5038                                     size_t pg_offset, u64 start, u64 len,
5039                                     int create)
5040 {
5041         int ret;
5042         int err = 0;
5043         u64 bytenr;
5044         u64 extent_start = 0;
5045         u64 extent_end = 0;
5046         u64 objectid = inode->i_ino;
5047         u32 found_type;
5048         struct btrfs_path *path = NULL;
5049         struct btrfs_root *root = BTRFS_I(inode)->root;
5050         struct btrfs_file_extent_item *item;
5051         struct extent_buffer *leaf;
5052         struct btrfs_key found_key;
5053         struct extent_map *em = NULL;
5054         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5055         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5056         struct btrfs_trans_handle *trans = NULL;
5057         int compress_type;
5058
5059 again:
5060         read_lock(&em_tree->lock);
5061         em = lookup_extent_mapping(em_tree, start, len);
5062         if (em)
5063                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5064         read_unlock(&em_tree->lock);
5065
5066         if (em) {
5067                 if (em->start > start || em->start + em->len <= start)
5068                         free_extent_map(em);
5069                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5070                         free_extent_map(em);
5071                 else
5072                         goto out;
5073         }
5074         em = alloc_extent_map(GFP_NOFS);
5075         if (!em) {
5076                 err = -ENOMEM;
5077                 goto out;
5078         }
5079         em->bdev = root->fs_info->fs_devices->latest_bdev;
5080         em->start = EXTENT_MAP_HOLE;
5081         em->orig_start = EXTENT_MAP_HOLE;
5082         em->len = (u64)-1;
5083         em->block_len = (u64)-1;
5084
5085         if (!path) {
5086                 path = btrfs_alloc_path();
5087                 BUG_ON(!path);
5088         }
5089
5090         ret = btrfs_lookup_file_extent(trans, root, path,
5091                                        objectid, start, trans != NULL);
5092         if (ret < 0) {
5093                 err = ret;
5094                 goto out;
5095         }
5096
5097         if (ret != 0) {
5098                 if (path->slots[0] == 0)
5099                         goto not_found;
5100                 path->slots[0]--;
5101         }
5102
5103         leaf = path->nodes[0];
5104         item = btrfs_item_ptr(leaf, path->slots[0],
5105                               struct btrfs_file_extent_item);
5106         /* are we inside the extent that was found? */
5107         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5108         found_type = btrfs_key_type(&found_key);
5109         if (found_key.objectid != objectid ||
5110             found_type != BTRFS_EXTENT_DATA_KEY) {
5111                 goto not_found;
5112         }
5113
5114         found_type = btrfs_file_extent_type(leaf, item);
5115         extent_start = found_key.offset;
5116         compress_type = btrfs_file_extent_compression(leaf, item);
5117         if (found_type == BTRFS_FILE_EXTENT_REG ||
5118             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5119                 extent_end = extent_start +
5120                        btrfs_file_extent_num_bytes(leaf, item);
5121         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5122                 size_t size;
5123                 size = btrfs_file_extent_inline_len(leaf, item);
5124                 extent_end = (extent_start + size + root->sectorsize - 1) &
5125                         ~((u64)root->sectorsize - 1);
5126         }
5127
5128         if (start >= extent_end) {
5129                 path->slots[0]++;
5130                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5131                         ret = btrfs_next_leaf(root, path);
5132                         if (ret < 0) {
5133                                 err = ret;
5134                                 goto out;
5135                         }
5136                         if (ret > 0)
5137                                 goto not_found;
5138                         leaf = path->nodes[0];
5139                 }
5140                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5141                 if (found_key.objectid != objectid ||
5142                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5143                         goto not_found;
5144                 if (start + len <= found_key.offset)
5145                         goto not_found;
5146                 em->start = start;
5147                 em->len = found_key.offset - start;
5148                 goto not_found_em;
5149         }
5150
5151         if (found_type == BTRFS_FILE_EXTENT_REG ||
5152             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5153                 em->start = extent_start;
5154                 em->len = extent_end - extent_start;
5155                 em->orig_start = extent_start -
5156                                  btrfs_file_extent_offset(leaf, item);
5157                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5158                 if (bytenr == 0) {
5159                         em->block_start = EXTENT_MAP_HOLE;
5160                         goto insert;
5161                 }
5162                 if (compress_type != BTRFS_COMPRESS_NONE) {
5163                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5164                         em->compress_type = compress_type;
5165                         em->block_start = bytenr;
5166                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5167                                                                          item);
5168                 } else {
5169                         bytenr += btrfs_file_extent_offset(leaf, item);
5170                         em->block_start = bytenr;
5171                         em->block_len = em->len;
5172                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5173                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5174                 }
5175                 goto insert;
5176         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5177                 unsigned long ptr;
5178                 char *map;
5179                 size_t size;
5180                 size_t extent_offset;
5181                 size_t copy_size;
5182
5183                 em->block_start = EXTENT_MAP_INLINE;
5184                 if (!page || create) {
5185                         em->start = extent_start;
5186                         em->len = extent_end - extent_start;
5187                         goto out;
5188                 }
5189
5190                 size = btrfs_file_extent_inline_len(leaf, item);
5191                 extent_offset = page_offset(page) + pg_offset - extent_start;
5192                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5193                                 size - extent_offset);
5194                 em->start = extent_start + extent_offset;
5195                 em->len = (copy_size + root->sectorsize - 1) &
5196                         ~((u64)root->sectorsize - 1);
5197                 em->orig_start = EXTENT_MAP_INLINE;
5198                 if (compress_type) {
5199                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5200                         em->compress_type = compress_type;
5201                 }
5202                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5203                 if (create == 0 && !PageUptodate(page)) {
5204                         if (btrfs_file_extent_compression(leaf, item) !=
5205                             BTRFS_COMPRESS_NONE) {
5206                                 ret = uncompress_inline(path, inode, page,
5207                                                         pg_offset,
5208                                                         extent_offset, item);
5209                                 BUG_ON(ret);
5210                         } else {
5211                                 map = kmap(page);
5212                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5213                                                    copy_size);
5214                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5215                                         memset(map + pg_offset + copy_size, 0,
5216                                                PAGE_CACHE_SIZE - pg_offset -
5217                                                copy_size);
5218                                 }
5219                                 kunmap(page);
5220                         }
5221                         flush_dcache_page(page);
5222                 } else if (create && PageUptodate(page)) {
5223                         WARN_ON(1);
5224                         if (!trans) {
5225                                 kunmap(page);
5226                                 free_extent_map(em);
5227                                 em = NULL;
5228                                 btrfs_release_path(root, path);
5229                                 trans = btrfs_join_transaction(root, 1);
5230                                 if (IS_ERR(trans))
5231                                         return ERR_CAST(trans);
5232                                 goto again;
5233                         }
5234                         map = kmap(page);
5235                         write_extent_buffer(leaf, map + pg_offset, ptr,
5236                                             copy_size);
5237                         kunmap(page);
5238                         btrfs_mark_buffer_dirty(leaf);
5239                 }
5240                 set_extent_uptodate(io_tree, em->start,
5241                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5242                 goto insert;
5243         } else {
5244                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5245                 WARN_ON(1);
5246         }
5247 not_found:
5248         em->start = start;
5249         em->len = len;
5250 not_found_em:
5251         em->block_start = EXTENT_MAP_HOLE;
5252         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5253 insert:
5254         btrfs_release_path(root, path);
5255         if (em->start > start || extent_map_end(em) <= start) {
5256                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5257                        "[%llu %llu]\n", (unsigned long long)em->start,
5258                        (unsigned long long)em->len,
5259                        (unsigned long long)start,
5260                        (unsigned long long)len);
5261                 err = -EIO;
5262                 goto out;
5263         }
5264
5265         err = 0;
5266         write_lock(&em_tree->lock);
5267         ret = add_extent_mapping(em_tree, em);
5268         /* it is possible that someone inserted the extent into the tree
5269          * while we had the lock dropped.  It is also possible that
5270          * an overlapping map exists in the tree
5271          */
5272         if (ret == -EEXIST) {
5273                 struct extent_map *existing;
5274
5275                 ret = 0;
5276
5277                 existing = lookup_extent_mapping(em_tree, start, len);
5278                 if (existing && (existing->start > start ||
5279                     existing->start + existing->len <= start)) {
5280                         free_extent_map(existing);
5281                         existing = NULL;
5282                 }
5283                 if (!existing) {
5284                         existing = lookup_extent_mapping(em_tree, em->start,
5285                                                          em->len);
5286                         if (existing) {
5287                                 err = merge_extent_mapping(em_tree, existing,
5288                                                            em, start,
5289                                                            root->sectorsize);
5290                                 free_extent_map(existing);
5291                                 if (err) {
5292                                         free_extent_map(em);
5293                                         em = NULL;
5294                                 }
5295                         } else {
5296                                 err = -EIO;
5297                                 free_extent_map(em);
5298                                 em = NULL;
5299                         }
5300                 } else {
5301                         free_extent_map(em);
5302                         em = existing;
5303                         err = 0;
5304                 }
5305         }
5306         write_unlock(&em_tree->lock);
5307 out:
5308
5309         trace_btrfs_get_extent(root, em);
5310
5311         if (path)
5312                 btrfs_free_path(path);
5313         if (trans) {
5314                 ret = btrfs_end_transaction(trans, root);
5315                 if (!err)
5316                         err = ret;
5317         }
5318         if (err) {
5319                 free_extent_map(em);
5320                 return ERR_PTR(err);
5321         }
5322         return em;
5323 }
5324
5325 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5326                                            size_t pg_offset, u64 start, u64 len,
5327                                            int create)
5328 {
5329         struct extent_map *em;
5330         struct extent_map *hole_em = NULL;
5331         u64 range_start = start;
5332         u64 end;
5333         u64 found;
5334         u64 found_end;
5335         int err = 0;
5336
5337         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5338         if (IS_ERR(em))
5339                 return em;
5340         if (em) {
5341                 /*
5342                  * if our em maps to a hole, there might
5343                  * actually be delalloc bytes behind it
5344                  */
5345                 if (em->block_start != EXTENT_MAP_HOLE)
5346                         return em;
5347                 else
5348                         hole_em = em;
5349         }
5350
5351         /* check to see if we've wrapped (len == -1 or similar) */
5352         end = start + len;
5353         if (end < start)
5354                 end = (u64)-1;
5355         else
5356                 end -= 1;
5357
5358         em = NULL;
5359
5360         /* ok, we didn't find anything, lets look for delalloc */
5361         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5362                                  end, len, EXTENT_DELALLOC, 1);
5363         found_end = range_start + found;
5364         if (found_end < range_start)
5365                 found_end = (u64)-1;
5366
5367         /*
5368          * we didn't find anything useful, return
5369          * the original results from get_extent()
5370          */
5371         if (range_start > end || found_end <= start) {
5372                 em = hole_em;
5373                 hole_em = NULL;
5374                 goto out;
5375         }
5376
5377         /* adjust the range_start to make sure it doesn't
5378          * go backwards from the start they passed in
5379          */
5380         range_start = max(start,range_start);
5381         found = found_end - range_start;
5382
5383         if (found > 0) {
5384                 u64 hole_start = start;
5385                 u64 hole_len = len;
5386
5387                 em = alloc_extent_map(GFP_NOFS);
5388                 if (!em) {
5389                         err = -ENOMEM;
5390                         goto out;
5391                 }
5392                 /*
5393                  * when btrfs_get_extent can't find anything it
5394                  * returns one huge hole
5395                  *
5396                  * make sure what it found really fits our range, and
5397                  * adjust to make sure it is based on the start from
5398                  * the caller
5399                  */
5400                 if (hole_em) {
5401                         u64 calc_end = extent_map_end(hole_em);
5402
5403                         if (calc_end <= start || (hole_em->start > end)) {
5404                                 free_extent_map(hole_em);
5405                                 hole_em = NULL;
5406                         } else {
5407                                 hole_start = max(hole_em->start, start);
5408                                 hole_len = calc_end - hole_start;
5409                         }
5410                 }
5411                 em->bdev = NULL;
5412                 if (hole_em && range_start > hole_start) {
5413                         /* our hole starts before our delalloc, so we
5414                          * have to return just the parts of the hole
5415                          * that go until  the delalloc starts
5416                          */
5417                         em->len = min(hole_len,
5418                                       range_start - hole_start);
5419                         em->start = hole_start;
5420                         em->orig_start = hole_start;
5421                         /*
5422                          * don't adjust block start at all,
5423                          * it is fixed at EXTENT_MAP_HOLE
5424                          */
5425                         em->block_start = hole_em->block_start;
5426                         em->block_len = hole_len;
5427                 } else {
5428                         em->start = range_start;
5429                         em->len = found;
5430                         em->orig_start = range_start;
5431                         em->block_start = EXTENT_MAP_DELALLOC;
5432                         em->block_len = found;
5433                 }
5434         } else if (hole_em) {
5435                 return hole_em;
5436         }
5437 out:
5438
5439         free_extent_map(hole_em);
5440         if (err) {
5441                 free_extent_map(em);
5442                 return ERR_PTR(err);
5443         }
5444         return em;
5445 }
5446
5447 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5448                                                   struct extent_map *em,
5449                                                   u64 start, u64 len)
5450 {
5451         struct btrfs_root *root = BTRFS_I(inode)->root;
5452         struct btrfs_trans_handle *trans;
5453         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5454         struct btrfs_key ins;
5455         u64 alloc_hint;
5456         int ret;
5457         bool insert = false;
5458
5459         /*
5460          * Ok if the extent map we looked up is a hole and is for the exact
5461          * range we want, there is no reason to allocate a new one, however if
5462          * it is not right then we need to free this one and drop the cache for
5463          * our range.
5464          */
5465         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5466             em->len != len) {
5467                 free_extent_map(em);
5468                 em = NULL;
5469                 insert = true;
5470                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5471         }
5472
5473         trans = btrfs_join_transaction(root, 0);
5474         if (IS_ERR(trans))
5475                 return ERR_CAST(trans);
5476
5477         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5478
5479         alloc_hint = get_extent_allocation_hint(inode, start, len);
5480         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5481                                    alloc_hint, (u64)-1, &ins, 1);
5482         if (ret) {
5483                 em = ERR_PTR(ret);
5484                 goto out;
5485         }
5486
5487         if (!em) {
5488                 em = alloc_extent_map(GFP_NOFS);
5489                 if (!em) {
5490                         em = ERR_PTR(-ENOMEM);
5491                         goto out;
5492                 }
5493         }
5494
5495         em->start = start;
5496         em->orig_start = em->start;
5497         em->len = ins.offset;
5498
5499         em->block_start = ins.objectid;
5500         em->block_len = ins.offset;
5501         em->bdev = root->fs_info->fs_devices->latest_bdev;
5502
5503         /*
5504          * We need to do this because if we're using the original em we searched
5505          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5506          */
5507         em->flags = 0;
5508         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5509
5510         while (insert) {
5511                 write_lock(&em_tree->lock);
5512                 ret = add_extent_mapping(em_tree, em);
5513                 write_unlock(&em_tree->lock);
5514                 if (ret != -EEXIST)
5515                         break;
5516                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5517         }
5518
5519         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5520                                            ins.offset, ins.offset, 0);
5521         if (ret) {
5522                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5523                 em = ERR_PTR(ret);
5524         }
5525 out:
5526         btrfs_end_transaction(trans, root);
5527         return em;
5528 }
5529
5530 /*
5531  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5532  * block must be cow'd
5533  */
5534 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5535                                       struct inode *inode, u64 offset, u64 len)
5536 {
5537         struct btrfs_path *path;
5538         int ret;
5539         struct extent_buffer *leaf;
5540         struct btrfs_root *root = BTRFS_I(inode)->root;
5541         struct btrfs_file_extent_item *fi;
5542         struct btrfs_key key;
5543         u64 disk_bytenr;
5544         u64 backref_offset;
5545         u64 extent_end;
5546         u64 num_bytes;
5547         int slot;
5548         int found_type;
5549
5550         path = btrfs_alloc_path();
5551         if (!path)
5552                 return -ENOMEM;
5553
5554         ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5555                                        offset, 0);
5556         if (ret < 0)
5557                 goto out;
5558
5559         slot = path->slots[0];
5560         if (ret == 1) {
5561                 if (slot == 0) {
5562                         /* can't find the item, must cow */
5563                         ret = 0;
5564                         goto out;
5565                 }
5566                 slot--;
5567         }
5568         ret = 0;
5569         leaf = path->nodes[0];
5570         btrfs_item_key_to_cpu(leaf, &key, slot);
5571         if (key.objectid != inode->i_ino ||
5572             key.type != BTRFS_EXTENT_DATA_KEY) {
5573                 /* not our file or wrong item type, must cow */
5574                 goto out;
5575         }
5576
5577         if (key.offset > offset) {
5578                 /* Wrong offset, must cow */
5579                 goto out;
5580         }
5581
5582         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5583         found_type = btrfs_file_extent_type(leaf, fi);
5584         if (found_type != BTRFS_FILE_EXTENT_REG &&
5585             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5586                 /* not a regular extent, must cow */
5587                 goto out;
5588         }
5589         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5590         backref_offset = btrfs_file_extent_offset(leaf, fi);
5591
5592         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5593         if (extent_end < offset + len) {
5594                 /* extent doesn't include our full range, must cow */
5595                 goto out;
5596         }
5597
5598         if (btrfs_extent_readonly(root, disk_bytenr))
5599                 goto out;
5600
5601         /*
5602          * look for other files referencing this extent, if we
5603          * find any we must cow
5604          */
5605         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5606                                   key.offset - backref_offset, disk_bytenr))
5607                 goto out;
5608
5609         /*
5610          * adjust disk_bytenr and num_bytes to cover just the bytes
5611          * in this extent we are about to write.  If there
5612          * are any csums in that range we have to cow in order
5613          * to keep the csums correct
5614          */
5615         disk_bytenr += backref_offset;
5616         disk_bytenr += offset - key.offset;
5617         num_bytes = min(offset + len, extent_end) - offset;
5618         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5619                                 goto out;
5620         /*
5621          * all of the above have passed, it is safe to overwrite this extent
5622          * without cow
5623          */
5624         ret = 1;
5625 out:
5626         btrfs_free_path(path);
5627         return ret;
5628 }
5629
5630 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5631                                    struct buffer_head *bh_result, int create)
5632 {
5633         struct extent_map *em;
5634         struct btrfs_root *root = BTRFS_I(inode)->root;
5635         u64 start = iblock << inode->i_blkbits;
5636         u64 len = bh_result->b_size;
5637         struct btrfs_trans_handle *trans;
5638
5639         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5640         if (IS_ERR(em))
5641                 return PTR_ERR(em);
5642
5643         /*
5644          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5645          * io.  INLINE is special, and we could probably kludge it in here, but
5646          * it's still buffered so for safety lets just fall back to the generic
5647          * buffered path.
5648          *
5649          * For COMPRESSED we _have_ to read the entire extent in&n