119520bdb9a540d8dc0f9bcf94d66b01df0af068
[pandora-kernel.git] / fs / btrfs / inode.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include "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
54 struct btrfs_iget_args {
55         u64 ino;
56         struct btrfs_root *root;
57 };
58
59 static const struct inode_operations btrfs_dir_inode_operations;
60 static const struct inode_operations btrfs_symlink_inode_operations;
61 static const struct inode_operations btrfs_dir_ro_inode_operations;
62 static const struct inode_operations btrfs_special_inode_operations;
63 static const struct inode_operations btrfs_file_inode_operations;
64 static const struct address_space_operations btrfs_aops;
65 static const struct address_space_operations btrfs_symlink_aops;
66 static const struct file_operations btrfs_dir_file_operations;
67 static struct extent_io_ops btrfs_extent_io_ops;
68
69 static struct kmem_cache *btrfs_inode_cachep;
70 struct kmem_cache *btrfs_trans_handle_cachep;
71 struct kmem_cache *btrfs_transaction_cachep;
72 struct kmem_cache *btrfs_path_cachep;
73
74 #define S_SHIFT 12
75 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
76         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
77         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
78         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
79         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
80         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
81         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
82         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
83 };
84
85 static void btrfs_truncate(struct inode *inode);
86 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
87 static noinline int cow_file_range(struct inode *inode,
88                                    struct page *locked_page,
89                                    u64 start, u64 end, int *page_started,
90                                    unsigned long *nr_written, int unlock);
91
92 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
93                                      struct inode *inode,  struct inode *dir,
94                                      const struct qstr *qstr)
95 {
96         int err;
97
98         err = btrfs_init_acl(trans, inode, dir);
99         if (!err)
100                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
101         return err;
102 }
103
104 /*
105  * this does all the hard work for inserting an inline extent into
106  * the btree.  The caller should have done a btrfs_drop_extents so that
107  * no overlapping inline items exist in the btree
108  */
109 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
110                                 struct btrfs_root *root, struct inode *inode,
111                                 u64 start, size_t size, size_t compressed_size,
112                                 struct page **compressed_pages)
113 {
114         struct btrfs_key key;
115         struct btrfs_path *path;
116         struct extent_buffer *leaf;
117         struct page *page = NULL;
118         char *kaddr;
119         unsigned long ptr;
120         struct btrfs_file_extent_item *ei;
121         int err = 0;
122         int ret;
123         size_t cur_size = size;
124         size_t datasize;
125         unsigned long offset;
126         int compress_type = BTRFS_COMPRESS_NONE;
127
128         if (compressed_size && compressed_pages) {
129                 compress_type = root->fs_info->compress_type;
130                 cur_size = compressed_size;
131         }
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,
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                                    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         async_extent->start = start;
292         async_extent->ram_size = ram_size;
293         async_extent->compressed_size = compressed_size;
294         async_extent->pages = pages;
295         async_extent->nr_pages = nr_pages;
296         async_extent->compress_type = compress_type;
297         list_add_tail(&async_extent->list, &cow->extents);
298         return 0;
299 }
300
301 /*
302  * we create compressed extents in two phases.  The first
303  * phase compresses a range of pages that have already been
304  * locked (both pages and state bits are locked).
305  *
306  * This is done inside an ordered work queue, and the compression
307  * is spread across many cpus.  The actual IO submission is step
308  * two, and the ordered work queue takes care of making sure that
309  * happens in the same order things were put onto the queue by
310  * writepages and friends.
311  *
312  * If this code finds it can't get good compression, it puts an
313  * entry onto the work queue to write the uncompressed bytes.  This
314  * makes sure that both compressed inodes and uncompressed inodes
315  * are written in the same order that pdflush sent them down.
316  */
317 static noinline int compress_file_range(struct inode *inode,
318                                         struct page *locked_page,
319                                         u64 start, u64 end,
320                                         struct async_cow *async_cow,
321                                         int *num_added)
322 {
323         struct btrfs_root *root = BTRFS_I(inode)->root;
324         struct btrfs_trans_handle *trans;
325         u64 num_bytes;
326         u64 blocksize = root->sectorsize;
327         u64 actual_end;
328         u64 isize = i_size_read(inode);
329         int ret = 0;
330         struct page **pages = NULL;
331         unsigned long nr_pages;
332         unsigned long nr_pages_ret = 0;
333         unsigned long total_compressed = 0;
334         unsigned long total_in = 0;
335         unsigned long max_compressed = 128 * 1024;
336         unsigned long max_uncompressed = 128 * 1024;
337         int i;
338         int will_compress;
339         int compress_type = root->fs_info->compress_type;
340
341         actual_end = min_t(u64, isize, end + 1);
342 again:
343         will_compress = 0;
344         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
345         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
346
347         /*
348          * we don't want to send crud past the end of i_size through
349          * compression, that's just a waste of CPU time.  So, if the
350          * end of the file is before the start of our current
351          * requested range of bytes, we bail out to the uncompressed
352          * cleanup code that can deal with all of this.
353          *
354          * It isn't really the fastest way to fix things, but this is a
355          * very uncommon corner.
356          */
357         if (actual_end <= start)
358                 goto cleanup_and_bail_uncompressed;
359
360         total_compressed = actual_end - start;
361
362         /* we want to make sure that amount of ram required to uncompress
363          * an extent is reasonable, so we limit the total size in ram
364          * of a compressed extent to 128k.  This is a crucial number
365          * because it also controls how easily we can spread reads across
366          * cpus for decompression.
367          *
368          * We also want to make sure the amount of IO required to do
369          * a random read is reasonably small, so we limit the size of
370          * a compressed extent to 128k.
371          */
372         total_compressed = min(total_compressed, max_uncompressed);
373         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
374         num_bytes = max(blocksize,  num_bytes);
375         total_in = 0;
376         ret = 0;
377
378         /*
379          * we do compression for mount -o compress and when the
380          * inode has not been flagged as nocompress.  This flag can
381          * change at any time if we discover bad compression ratios.
382          */
383         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
384             (btrfs_test_opt(root, COMPRESS) ||
385              (BTRFS_I(inode)->force_compress))) {
386                 WARN_ON(pages);
387                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
388
389                 if (BTRFS_I(inode)->force_compress)
390                         compress_type = BTRFS_I(inode)->force_compress;
391
392                 ret = btrfs_compress_pages(compress_type,
393                                            inode->i_mapping, start,
394                                            total_compressed, pages,
395                                            nr_pages, &nr_pages_ret,
396                                            &total_in,
397                                            &total_compressed,
398                                            max_compressed);
399
400                 if (!ret) {
401                         unsigned long offset = total_compressed &
402                                 (PAGE_CACHE_SIZE - 1);
403                         struct page *page = pages[nr_pages_ret - 1];
404                         char *kaddr;
405
406                         /* zero the tail end of the last page, we might be
407                          * sending it down to disk
408                          */
409                         if (offset) {
410                                 kaddr = kmap_atomic(page, KM_USER0);
411                                 memset(kaddr + offset, 0,
412                                        PAGE_CACHE_SIZE - offset);
413                                 kunmap_atomic(kaddr, KM_USER0);
414                         }
415                         will_compress = 1;
416                 }
417         }
418         if (start == 0) {
419                 trans = btrfs_join_transaction(root, 1);
420                 BUG_ON(IS_ERR(trans));
421                 btrfs_set_trans_block_group(trans, inode);
422                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
423
424                 /* lets try to make an inline extent */
425                 if (ret || total_in < (actual_end - start)) {
426                         /* we didn't compress the entire range, try
427                          * to make an uncompressed inline extent.
428                          */
429                         ret = cow_file_range_inline(trans, root, inode,
430                                                     start, end, 0, NULL);
431                 } else {
432                         /* try making a compressed inline extent */
433                         ret = cow_file_range_inline(trans, root, inode,
434                                                     start, end,
435                                                     total_compressed, pages);
436                 }
437                 if (ret == 0) {
438                         /*
439                          * inline extent creation worked, we don't need
440                          * to create any more async work items.  Unlock
441                          * and free up our temp pages.
442                          */
443                         extent_clear_unlock_delalloc(inode,
444                              &BTRFS_I(inode)->io_tree,
445                              start, end, NULL,
446                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
447                              EXTENT_CLEAR_DELALLOC |
448                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
449
450                         btrfs_end_transaction(trans, root);
451                         goto free_pages_out;
452                 }
453                 btrfs_end_transaction(trans, root);
454         }
455
456         if (will_compress) {
457                 /*
458                  * we aren't doing an inline extent round the compressed size
459                  * up to a block size boundary so the allocator does sane
460                  * things
461                  */
462                 total_compressed = (total_compressed + blocksize - 1) &
463                         ~(blocksize - 1);
464
465                 /*
466                  * one last check to make sure the compression is really a
467                  * win, compare the page count read with the blocks on disk
468                  */
469                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
470                         ~(PAGE_CACHE_SIZE - 1);
471                 if (total_compressed >= total_in) {
472                         will_compress = 0;
473                 } else {
474                         num_bytes = total_in;
475                 }
476         }
477         if (!will_compress && pages) {
478                 /*
479                  * the compression code ran but failed to make things smaller,
480                  * free any pages it allocated and our page pointer array
481                  */
482                 for (i = 0; i < nr_pages_ret; i++) {
483                         WARN_ON(pages[i]->mapping);
484                         page_cache_release(pages[i]);
485                 }
486                 kfree(pages);
487                 pages = NULL;
488                 total_compressed = 0;
489                 nr_pages_ret = 0;
490
491                 /* flag the file so we don't compress in the future */
492                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
493                     !(BTRFS_I(inode)->force_compress)) {
494                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
495                 }
496         }
497         if (will_compress) {
498                 *num_added += 1;
499
500                 /* the async work queues will take care of doing actual
501                  * allocation on disk for these compressed pages,
502                  * and will submit them to the elevator.
503                  */
504                 add_async_extent(async_cow, start, num_bytes,
505                                  total_compressed, pages, nr_pages_ret,
506                                  compress_type);
507
508                 if (start + num_bytes < end) {
509                         start += num_bytes;
510                         pages = NULL;
511                         cond_resched();
512                         goto again;
513                 }
514         } else {
515 cleanup_and_bail_uncompressed:
516                 /*
517                  * No compression, but we still need to write the pages in
518                  * the file we've been given so far.  redirty the locked
519                  * page if it corresponds to our extent and set things up
520                  * for the async work queue to run cow_file_range to do
521                  * the normal delalloc dance
522                  */
523                 if (page_offset(locked_page) >= start &&
524                     page_offset(locked_page) <= end) {
525                         __set_page_dirty_nobuffers(locked_page);
526                         /* unlocked later on in the async handlers */
527                 }
528                 add_async_extent(async_cow, start, end - start + 1,
529                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
530                 *num_added += 1;
531         }
532
533 out:
534         return 0;
535
536 free_pages_out:
537         for (i = 0; i < nr_pages_ret; i++) {
538                 WARN_ON(pages[i]->mapping);
539                 page_cache_release(pages[i]);
540         }
541         kfree(pages);
542
543         goto out;
544 }
545
546 /*
547  * phase two of compressed writeback.  This is the ordered portion
548  * of the code, which only gets called in the order the work was
549  * queued.  We walk all the async extents created by compress_file_range
550  * and send them down to the disk.
551  */
552 static noinline int submit_compressed_extents(struct inode *inode,
553                                               struct async_cow *async_cow)
554 {
555         struct async_extent *async_extent;
556         u64 alloc_hint = 0;
557         struct btrfs_trans_handle *trans;
558         struct btrfs_key ins;
559         struct extent_map *em;
560         struct btrfs_root *root = BTRFS_I(inode)->root;
561         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
562         struct extent_io_tree *io_tree;
563         int ret = 0;
564
565         if (list_empty(&async_cow->extents))
566                 return 0;
567
568
569         while (!list_empty(&async_cow->extents)) {
570                 async_extent = list_entry(async_cow->extents.next,
571                                           struct async_extent, list);
572                 list_del(&async_extent->list);
573
574                 io_tree = &BTRFS_I(inode)->io_tree;
575
576 retry:
577                 /* did the compression code fall back to uncompressed IO? */
578                 if (!async_extent->pages) {
579                         int page_started = 0;
580                         unsigned long nr_written = 0;
581
582                         lock_extent(io_tree, async_extent->start,
583                                          async_extent->start +
584                                          async_extent->ram_size - 1, GFP_NOFS);
585
586                         /* allocate blocks */
587                         ret = cow_file_range(inode, async_cow->locked_page,
588                                              async_extent->start,
589                                              async_extent->start +
590                                              async_extent->ram_size - 1,
591                                              &page_started, &nr_written, 0);
592
593                         /*
594                          * if page_started, cow_file_range inserted an
595                          * inline extent and took care of all the unlocking
596                          * and IO for us.  Otherwise, we need to submit
597                          * all those pages down to the drive.
598                          */
599                         if (!page_started && !ret)
600                                 extent_write_locked_range(io_tree,
601                                                   inode, async_extent->start,
602                                                   async_extent->start +
603                                                   async_extent->ram_size - 1,
604                                                   btrfs_get_extent,
605                                                   WB_SYNC_ALL);
606                         kfree(async_extent);
607                         cond_resched();
608                         continue;
609                 }
610
611                 lock_extent(io_tree, async_extent->start,
612                             async_extent->start + async_extent->ram_size - 1,
613                             GFP_NOFS);
614
615                 trans = btrfs_join_transaction(root, 1);
616                 BUG_ON(IS_ERR(trans));
617                 ret = btrfs_reserve_extent(trans, root,
618                                            async_extent->compressed_size,
619                                            async_extent->compressed_size,
620                                            0, alloc_hint,
621                                            (u64)-1, &ins, 1);
622                 btrfs_end_transaction(trans, root);
623
624                 if (ret) {
625                         int i;
626                         for (i = 0; i < async_extent->nr_pages; i++) {
627                                 WARN_ON(async_extent->pages[i]->mapping);
628                                 page_cache_release(async_extent->pages[i]);
629                         }
630                         kfree(async_extent->pages);
631                         async_extent->nr_pages = 0;
632                         async_extent->pages = NULL;
633                         unlock_extent(io_tree, async_extent->start,
634                                       async_extent->start +
635                                       async_extent->ram_size - 1, GFP_NOFS);
636                         goto retry;
637                 }
638
639                 /*
640                  * here we're doing allocation and writeback of the
641                  * compressed pages
642                  */
643                 btrfs_drop_extent_cache(inode, async_extent->start,
644                                         async_extent->start +
645                                         async_extent->ram_size - 1, 0);
646
647                 em = alloc_extent_map(GFP_NOFS);
648                 BUG_ON(!em);
649                 em->start = async_extent->start;
650                 em->len = async_extent->ram_size;
651                 em->orig_start = em->start;
652
653                 em->block_start = ins.objectid;
654                 em->block_len = ins.offset;
655                 em->bdev = root->fs_info->fs_devices->latest_bdev;
656                 em->compress_type = async_extent->compress_type;
657                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
658                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
659
660                 while (1) {
661                         write_lock(&em_tree->lock);
662                         ret = add_extent_mapping(em_tree, em);
663                         write_unlock(&em_tree->lock);
664                         if (ret != -EEXIST) {
665                                 free_extent_map(em);
666                                 break;
667                         }
668                         btrfs_drop_extent_cache(inode, async_extent->start,
669                                                 async_extent->start +
670                                                 async_extent->ram_size - 1, 0);
671                 }
672
673                 ret = btrfs_add_ordered_extent_compress(inode,
674                                                 async_extent->start,
675                                                 ins.objectid,
676                                                 async_extent->ram_size,
677                                                 ins.offset,
678                                                 BTRFS_ORDERED_COMPRESSED,
679                                                 async_extent->compress_type);
680                 BUG_ON(ret);
681
682                 /*
683                  * clear dirty, set writeback and unlock the pages.
684                  */
685                 extent_clear_unlock_delalloc(inode,
686                                 &BTRFS_I(inode)->io_tree,
687                                 async_extent->start,
688                                 async_extent->start +
689                                 async_extent->ram_size - 1,
690                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
691                                 EXTENT_CLEAR_UNLOCK |
692                                 EXTENT_CLEAR_DELALLOC |
693                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
694
695                 ret = btrfs_submit_compressed_write(inode,
696                                     async_extent->start,
697                                     async_extent->ram_size,
698                                     ins.objectid,
699                                     ins.offset, async_extent->pages,
700                                     async_extent->nr_pages);
701
702                 BUG_ON(ret);
703                 alloc_hint = ins.objectid + ins.offset;
704                 kfree(async_extent);
705                 cond_resched();
706         }
707
708         return 0;
709 }
710
711 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
712                                       u64 num_bytes)
713 {
714         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
715         struct extent_map *em;
716         u64 alloc_hint = 0;
717
718         read_lock(&em_tree->lock);
719         em = search_extent_mapping(em_tree, start, num_bytes);
720         if (em) {
721                 /*
722                  * if block start isn't an actual block number then find the
723                  * first block in this inode and use that as a hint.  If that
724                  * block is also bogus then just don't worry about it.
725                  */
726                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
727                         free_extent_map(em);
728                         em = search_extent_mapping(em_tree, 0, 0);
729                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
730                                 alloc_hint = em->block_start;
731                         if (em)
732                                 free_extent_map(em);
733                 } else {
734                         alloc_hint = em->block_start;
735                         free_extent_map(em);
736                 }
737         }
738         read_unlock(&em_tree->lock);
739
740         return alloc_hint;
741 }
742
743 /*
744  * when extent_io.c finds a delayed allocation range in the file,
745  * the call backs end up in this code.  The basic idea is to
746  * allocate extents on disk for the range, and create ordered data structs
747  * in ram to track those extents.
748  *
749  * locked_page is the page that writepage had locked already.  We use
750  * it to make sure we don't do extra locks or unlocks.
751  *
752  * *page_started is set to one if we unlock locked_page and do everything
753  * required to start IO on it.  It may be clean and already done with
754  * IO when we return.
755  */
756 static noinline int cow_file_range(struct inode *inode,
757                                    struct page *locked_page,
758                                    u64 start, u64 end, int *page_started,
759                                    unsigned long *nr_written,
760                                    int unlock)
761 {
762         struct btrfs_root *root = BTRFS_I(inode)->root;
763         struct btrfs_trans_handle *trans;
764         u64 alloc_hint = 0;
765         u64 num_bytes;
766         unsigned long ram_size;
767         u64 disk_num_bytes;
768         u64 cur_alloc_size;
769         u64 blocksize = root->sectorsize;
770         struct btrfs_key ins;
771         struct extent_map *em;
772         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
773         int ret = 0;
774
775         BUG_ON(root == root->fs_info->tree_root);
776         trans = btrfs_join_transaction(root, 1);
777         BUG_ON(IS_ERR(trans));
778         btrfs_set_trans_block_group(trans, inode);
779         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
780
781         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
782         num_bytes = max(blocksize,  num_bytes);
783         disk_num_bytes = num_bytes;
784         ret = 0;
785
786         if (start == 0) {
787                 /* lets try to make an inline extent */
788                 ret = cow_file_range_inline(trans, root, inode,
789                                             start, end, 0, NULL);
790                 if (ret == 0) {
791                         extent_clear_unlock_delalloc(inode,
792                                      &BTRFS_I(inode)->io_tree,
793                                      start, end, NULL,
794                                      EXTENT_CLEAR_UNLOCK_PAGE |
795                                      EXTENT_CLEAR_UNLOCK |
796                                      EXTENT_CLEAR_DELALLOC |
797                                      EXTENT_CLEAR_DIRTY |
798                                      EXTENT_SET_WRITEBACK |
799                                      EXTENT_END_WRITEBACK);
800
801                         *nr_written = *nr_written +
802                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
803                         *page_started = 1;
804                         ret = 0;
805                         goto out;
806                 }
807         }
808
809         BUG_ON(disk_num_bytes >
810                btrfs_super_total_bytes(&root->fs_info->super_copy));
811
812         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
813         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
814
815         while (disk_num_bytes > 0) {
816                 unsigned long op;
817
818                 cur_alloc_size = disk_num_bytes;
819                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
820                                            root->sectorsize, 0, alloc_hint,
821                                            (u64)-1, &ins, 1);
822                 BUG_ON(ret);
823
824                 em = alloc_extent_map(GFP_NOFS);
825                 BUG_ON(!em);
826                 em->start = start;
827                 em->orig_start = em->start;
828                 ram_size = ins.offset;
829                 em->len = ins.offset;
830
831                 em->block_start = ins.objectid;
832                 em->block_len = ins.offset;
833                 em->bdev = root->fs_info->fs_devices->latest_bdev;
834                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
835
836                 while (1) {
837                         write_lock(&em_tree->lock);
838                         ret = add_extent_mapping(em_tree, em);
839                         write_unlock(&em_tree->lock);
840                         if (ret != -EEXIST) {
841                                 free_extent_map(em);
842                                 break;
843                         }
844                         btrfs_drop_extent_cache(inode, start,
845                                                 start + ram_size - 1, 0);
846                 }
847
848                 cur_alloc_size = ins.offset;
849                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
850                                                ram_size, cur_alloc_size, 0);
851                 BUG_ON(ret);
852
853                 if (root->root_key.objectid ==
854                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
855                         ret = btrfs_reloc_clone_csums(inode, start,
856                                                       cur_alloc_size);
857                         BUG_ON(ret);
858                 }
859
860                 if (disk_num_bytes < cur_alloc_size)
861                         break;
862
863                 /* we're not doing compressed IO, don't unlock the first
864                  * page (which the caller expects to stay locked), don't
865                  * clear any dirty bits and don't set any writeback bits
866                  *
867                  * Do set the Private2 bit so we know this page was properly
868                  * setup for writepage
869                  */
870                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
871                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
872                         EXTENT_SET_PRIVATE2;
873
874                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
875                                              start, start + ram_size - 1,
876                                              locked_page, op);
877                 disk_num_bytes -= cur_alloc_size;
878                 num_bytes -= cur_alloc_size;
879                 alloc_hint = ins.objectid + ins.offset;
880                 start += cur_alloc_size;
881         }
882 out:
883         ret = 0;
884         btrfs_end_transaction(trans, root);
885
886         return ret;
887 }
888
889 /*
890  * work queue call back to started compression on a file and pages
891  */
892 static noinline void async_cow_start(struct btrfs_work *work)
893 {
894         struct async_cow *async_cow;
895         int num_added = 0;
896         async_cow = container_of(work, struct async_cow, work);
897
898         compress_file_range(async_cow->inode, async_cow->locked_page,
899                             async_cow->start, async_cow->end, async_cow,
900                             &num_added);
901         if (num_added == 0)
902                 async_cow->inode = NULL;
903 }
904
905 /*
906  * work queue call back to submit previously compressed pages
907  */
908 static noinline void async_cow_submit(struct btrfs_work *work)
909 {
910         struct async_cow *async_cow;
911         struct btrfs_root *root;
912         unsigned long nr_pages;
913
914         async_cow = container_of(work, struct async_cow, work);
915
916         root = async_cow->root;
917         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
918                 PAGE_CACHE_SHIFT;
919
920         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
921
922         if (atomic_read(&root->fs_info->async_delalloc_pages) <
923             5 * 1042 * 1024 &&
924             waitqueue_active(&root->fs_info->async_submit_wait))
925                 wake_up(&root->fs_info->async_submit_wait);
926
927         if (async_cow->inode)
928                 submit_compressed_extents(async_cow->inode, async_cow);
929 }
930
931 static noinline void async_cow_free(struct btrfs_work *work)
932 {
933         struct async_cow *async_cow;
934         async_cow = container_of(work, struct async_cow, work);
935         kfree(async_cow);
936 }
937
938 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
939                                 u64 start, u64 end, int *page_started,
940                                 unsigned long *nr_written)
941 {
942         struct async_cow *async_cow;
943         struct btrfs_root *root = BTRFS_I(inode)->root;
944         unsigned long nr_pages;
945         u64 cur_end;
946         int limit = 10 * 1024 * 1042;
947
948         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
949                          1, 0, NULL, GFP_NOFS);
950         while (start < end) {
951                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
952                 async_cow->inode = inode;
953                 async_cow->root = root;
954                 async_cow->locked_page = locked_page;
955                 async_cow->start = start;
956
957                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
958                         cur_end = end;
959                 else
960                         cur_end = min(end, start + 512 * 1024 - 1);
961
962                 async_cow->end = cur_end;
963                 INIT_LIST_HEAD(&async_cow->extents);
964
965                 async_cow->work.func = async_cow_start;
966                 async_cow->work.ordered_func = async_cow_submit;
967                 async_cow->work.ordered_free = async_cow_free;
968                 async_cow->work.flags = 0;
969
970                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
971                         PAGE_CACHE_SHIFT;
972                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
973
974                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
975                                    &async_cow->work);
976
977                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
978                         wait_event(root->fs_info->async_submit_wait,
979                            (atomic_read(&root->fs_info->async_delalloc_pages) <
980                             limit));
981                 }
982
983                 while (atomic_read(&root->fs_info->async_submit_draining) &&
984                       atomic_read(&root->fs_info->async_delalloc_pages)) {
985                         wait_event(root->fs_info->async_submit_wait,
986                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
987                            0));
988                 }
989
990                 *nr_written += nr_pages;
991                 start = cur_end + 1;
992         }
993         *page_started = 1;
994         return 0;
995 }
996
997 static noinline int csum_exist_in_range(struct btrfs_root *root,
998                                         u64 bytenr, u64 num_bytes)
999 {
1000         int ret;
1001         struct btrfs_ordered_sum *sums;
1002         LIST_HEAD(list);
1003
1004         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1005                                        bytenr + num_bytes - 1, &list);
1006         if (ret == 0 && list_empty(&list))
1007                 return 0;
1008
1009         while (!list_empty(&list)) {
1010                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1011                 list_del(&sums->list);
1012                 kfree(sums);
1013         }
1014         return 1;
1015 }
1016
1017 /*
1018  * when nowcow writeback call back.  This checks for snapshots or COW copies
1019  * of the extents that exist in the file, and COWs the file as required.
1020  *
1021  * If no cow copies or snapshots exist, we write directly to the existing
1022  * blocks on disk
1023  */
1024 static noinline int run_delalloc_nocow(struct inode *inode,
1025                                        struct page *locked_page,
1026                               u64 start, u64 end, int *page_started, int force,
1027                               unsigned long *nr_written)
1028 {
1029         struct btrfs_root *root = BTRFS_I(inode)->root;
1030         struct btrfs_trans_handle *trans;
1031         struct extent_buffer *leaf;
1032         struct btrfs_path *path;
1033         struct btrfs_file_extent_item *fi;
1034         struct btrfs_key found_key;
1035         u64 cow_start;
1036         u64 cur_offset;
1037         u64 extent_end;
1038         u64 extent_offset;
1039         u64 disk_bytenr;
1040         u64 num_bytes;
1041         int extent_type;
1042         int ret;
1043         int type;
1044         int nocow;
1045         int check_prev = 1;
1046         bool nolock = false;
1047
1048         path = btrfs_alloc_path();
1049         BUG_ON(!path);
1050         if (root == root->fs_info->tree_root) {
1051                 nolock = true;
1052                 trans = btrfs_join_transaction_nolock(root, 1);
1053         } else {
1054                 trans = btrfs_join_transaction(root, 1);
1055         }
1056         BUG_ON(IS_ERR(trans));
1057
1058         cow_start = (u64)-1;
1059         cur_offset = start;
1060         while (1) {
1061                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1062                                                cur_offset, 0);
1063                 BUG_ON(ret < 0);
1064                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1065                         leaf = path->nodes[0];
1066                         btrfs_item_key_to_cpu(leaf, &found_key,
1067                                               path->slots[0] - 1);
1068                         if (found_key.objectid == inode->i_ino &&
1069                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1070                                 path->slots[0]--;
1071                 }
1072                 check_prev = 0;
1073 next_slot:
1074                 leaf = path->nodes[0];
1075                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1076                         ret = btrfs_next_leaf(root, path);
1077                         if (ret < 0)
1078                                 BUG_ON(1);
1079                         if (ret > 0)
1080                                 break;
1081                         leaf = path->nodes[0];
1082                 }
1083
1084                 nocow = 0;
1085                 disk_bytenr = 0;
1086                 num_bytes = 0;
1087                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1088
1089                 if (found_key.objectid > inode->i_ino ||
1090                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1091                     found_key.offset > end)
1092                         break;
1093
1094                 if (found_key.offset > cur_offset) {
1095                         extent_end = found_key.offset;
1096                         extent_type = 0;
1097                         goto out_check;
1098                 }
1099
1100                 fi = btrfs_item_ptr(leaf, path->slots[0],
1101                                     struct btrfs_file_extent_item);
1102                 extent_type = btrfs_file_extent_type(leaf, fi);
1103
1104                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1105                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1106                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1107                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1108                         extent_end = found_key.offset +
1109                                 btrfs_file_extent_num_bytes(leaf, fi);
1110                         if (extent_end <= start) {
1111                                 path->slots[0]++;
1112                                 goto next_slot;
1113                         }
1114                         if (disk_bytenr == 0)
1115                                 goto out_check;
1116                         if (btrfs_file_extent_compression(leaf, fi) ||
1117                             btrfs_file_extent_encryption(leaf, fi) ||
1118                             btrfs_file_extent_other_encoding(leaf, fi))
1119                                 goto out_check;
1120                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1121                                 goto out_check;
1122                         if (btrfs_extent_readonly(root, disk_bytenr))
1123                                 goto out_check;
1124                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1125                                                   found_key.offset -
1126                                                   extent_offset, disk_bytenr))
1127                                 goto out_check;
1128                         disk_bytenr += extent_offset;
1129                         disk_bytenr += cur_offset - found_key.offset;
1130                         num_bytes = min(end + 1, extent_end) - cur_offset;
1131                         /*
1132                          * force cow if csum exists in the range.
1133                          * this ensure that csum for a given extent are
1134                          * either valid or do not exist.
1135                          */
1136                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1137                                 goto out_check;
1138                         nocow = 1;
1139                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1140                         extent_end = found_key.offset +
1141                                 btrfs_file_extent_inline_len(leaf, fi);
1142                         extent_end = ALIGN(extent_end, root->sectorsize);
1143                 } else {
1144                         BUG_ON(1);
1145                 }
1146 out_check:
1147                 if (extent_end <= start) {
1148                         path->slots[0]++;
1149                         goto next_slot;
1150                 }
1151                 if (!nocow) {
1152                         if (cow_start == (u64)-1)
1153                                 cow_start = cur_offset;
1154                         cur_offset = extent_end;
1155                         if (cur_offset > end)
1156                                 break;
1157                         path->slots[0]++;
1158                         goto next_slot;
1159                 }
1160
1161                 btrfs_release_path(root, path);
1162                 if (cow_start != (u64)-1) {
1163                         ret = cow_file_range(inode, locked_page, cow_start,
1164                                         found_key.offset - 1, page_started,
1165                                         nr_written, 1);
1166                         BUG_ON(ret);
1167                         cow_start = (u64)-1;
1168                 }
1169
1170                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1171                         struct extent_map *em;
1172                         struct extent_map_tree *em_tree;
1173                         em_tree = &BTRFS_I(inode)->extent_tree;
1174                         em = alloc_extent_map(GFP_NOFS);
1175                         BUG_ON(!em);
1176                         em->start = cur_offset;
1177                         em->orig_start = em->start;
1178                         em->len = num_bytes;
1179                         em->block_len = num_bytes;
1180                         em->block_start = disk_bytenr;
1181                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1182                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1183                         while (1) {
1184                                 write_lock(&em_tree->lock);
1185                                 ret = add_extent_mapping(em_tree, em);
1186                                 write_unlock(&em_tree->lock);
1187                                 if (ret != -EEXIST) {
1188                                         free_extent_map(em);
1189                                         break;
1190                                 }
1191                                 btrfs_drop_extent_cache(inode, em->start,
1192                                                 em->start + em->len - 1, 0);
1193                         }
1194                         type = BTRFS_ORDERED_PREALLOC;
1195                 } else {
1196                         type = BTRFS_ORDERED_NOCOW;
1197                 }
1198
1199                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1200                                                num_bytes, num_bytes, type);
1201                 BUG_ON(ret);
1202
1203                 if (root->root_key.objectid ==
1204                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1205                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1206                                                       num_bytes);
1207                         BUG_ON(ret);
1208                 }
1209
1210                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1211                                 cur_offset, cur_offset + num_bytes - 1,
1212                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1213                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1214                                 EXTENT_SET_PRIVATE2);
1215                 cur_offset = extent_end;
1216                 if (cur_offset > end)
1217                         break;
1218         }
1219         btrfs_release_path(root, path);
1220
1221         if (cur_offset <= end && cow_start == (u64)-1)
1222                 cow_start = cur_offset;
1223         if (cow_start != (u64)-1) {
1224                 ret = cow_file_range(inode, locked_page, cow_start, end,
1225                                      page_started, nr_written, 1);
1226                 BUG_ON(ret);
1227         }
1228
1229         if (nolock) {
1230                 ret = btrfs_end_transaction_nolock(trans, root);
1231                 BUG_ON(ret);
1232         } else {
1233                 ret = btrfs_end_transaction(trans, root);
1234                 BUG_ON(ret);
1235         }
1236         btrfs_free_path(path);
1237         return 0;
1238 }
1239
1240 /*
1241  * extent_io.c call back to do delayed allocation processing
1242  */
1243 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1244                               u64 start, u64 end, int *page_started,
1245                               unsigned long *nr_written)
1246 {
1247         int ret;
1248         struct btrfs_root *root = BTRFS_I(inode)->root;
1249
1250         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1251                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1252                                          page_started, 1, nr_written);
1253         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1254                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1255                                          page_started, 0, nr_written);
1256         else if (!btrfs_test_opt(root, COMPRESS) &&
1257                  !(BTRFS_I(inode)->force_compress))
1258                 ret = cow_file_range(inode, locked_page, start, end,
1259                                       page_started, nr_written, 1);
1260         else
1261                 ret = cow_file_range_async(inode, locked_page, start, end,
1262                                            page_started, nr_written);
1263         return ret;
1264 }
1265
1266 static int btrfs_split_extent_hook(struct inode *inode,
1267                                    struct extent_state *orig, u64 split)
1268 {
1269         /* not delalloc, ignore it */
1270         if (!(orig->state & EXTENT_DELALLOC))
1271                 return 0;
1272
1273         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1274         return 0;
1275 }
1276
1277 /*
1278  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1279  * extents so we can keep track of new extents that are just merged onto old
1280  * extents, such as when we are doing sequential writes, so we can properly
1281  * account for the metadata space we'll need.
1282  */
1283 static int btrfs_merge_extent_hook(struct inode *inode,
1284                                    struct extent_state *new,
1285                                    struct extent_state *other)
1286 {
1287         /* not delalloc, ignore it */
1288         if (!(other->state & EXTENT_DELALLOC))
1289                 return 0;
1290
1291         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1292         return 0;
1293 }
1294
1295 /*
1296  * extent_io.c set_bit_hook, used to track delayed allocation
1297  * bytes in this file, and to maintain the list of inodes that
1298  * have pending delalloc work to be done.
1299  */
1300 static int btrfs_set_bit_hook(struct inode *inode,
1301                               struct extent_state *state, int *bits)
1302 {
1303
1304         /*
1305          * set_bit and clear bit hooks normally require _irqsave/restore
1306          * but in this case, we are only testeing for the DELALLOC
1307          * bit, which is only set or cleared with irqs on
1308          */
1309         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1310                 struct btrfs_root *root = BTRFS_I(inode)->root;
1311                 u64 len = state->end + 1 - state->start;
1312                 int do_list = (root->root_key.objectid !=
1313                                BTRFS_ROOT_TREE_OBJECTID);
1314
1315                 if (*bits & EXTENT_FIRST_DELALLOC)
1316                         *bits &= ~EXTENT_FIRST_DELALLOC;
1317                 else
1318                         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1319
1320                 spin_lock(&root->fs_info->delalloc_lock);
1321                 BTRFS_I(inode)->delalloc_bytes += len;
1322                 root->fs_info->delalloc_bytes += len;
1323                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1324                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1325                                       &root->fs_info->delalloc_inodes);
1326                 }
1327                 spin_unlock(&root->fs_info->delalloc_lock);
1328         }
1329         return 0;
1330 }
1331
1332 /*
1333  * extent_io.c clear_bit_hook, see set_bit_hook for why
1334  */
1335 static int btrfs_clear_bit_hook(struct inode *inode,
1336                                 struct extent_state *state, int *bits)
1337 {
1338         /*
1339          * set_bit and clear bit hooks normally require _irqsave/restore
1340          * but in this case, we are only testeing for the DELALLOC
1341          * bit, which is only set or cleared with irqs on
1342          */
1343         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1344                 struct btrfs_root *root = BTRFS_I(inode)->root;
1345                 u64 len = state->end + 1 - state->start;
1346                 int do_list = (root->root_key.objectid !=
1347                                BTRFS_ROOT_TREE_OBJECTID);
1348
1349                 if (*bits & EXTENT_FIRST_DELALLOC)
1350                         *bits &= ~EXTENT_FIRST_DELALLOC;
1351                 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1352                         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1353
1354                 if (*bits & EXTENT_DO_ACCOUNTING)
1355                         btrfs_delalloc_release_metadata(inode, len);
1356
1357                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1358                     && do_list)
1359                         btrfs_free_reserved_data_space(inode, len);
1360
1361                 spin_lock(&root->fs_info->delalloc_lock);
1362                 root->fs_info->delalloc_bytes -= len;
1363                 BTRFS_I(inode)->delalloc_bytes -= len;
1364
1365                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1366                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1367                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1368                 }
1369                 spin_unlock(&root->fs_info->delalloc_lock);
1370         }
1371         return 0;
1372 }
1373
1374 /*
1375  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1376  * we don't create bios that span stripes or chunks
1377  */
1378 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1379                          size_t size, struct bio *bio,
1380                          unsigned long bio_flags)
1381 {
1382         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1383         struct btrfs_mapping_tree *map_tree;
1384         u64 logical = (u64)bio->bi_sector << 9;
1385         u64 length = 0;
1386         u64 map_length;
1387         int ret;
1388
1389         if (bio_flags & EXTENT_BIO_COMPRESSED)
1390                 return 0;
1391
1392         length = bio->bi_size;
1393         map_tree = &root->fs_info->mapping_tree;
1394         map_length = length;
1395         ret = btrfs_map_block(map_tree, READ, logical,
1396                               &map_length, NULL, 0);
1397
1398         if (map_length < length + size)
1399                 return 1;
1400         return ret;
1401 }
1402
1403 /*
1404  * in order to insert checksums into the metadata in large chunks,
1405  * we wait until bio submission time.   All the pages in the bio are
1406  * checksummed and sums are attached onto the ordered extent record.
1407  *
1408  * At IO completion time the cums attached on the ordered extent record
1409  * are inserted into the btree
1410  */
1411 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1412                                     struct bio *bio, int mirror_num,
1413                                     unsigned long bio_flags,
1414                                     u64 bio_offset)
1415 {
1416         struct btrfs_root *root = BTRFS_I(inode)->root;
1417         int ret = 0;
1418
1419         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1420         BUG_ON(ret);
1421         return 0;
1422 }
1423
1424 /*
1425  * in order to insert checksums into the metadata in large chunks,
1426  * we wait until bio submission time.   All the pages in the bio are
1427  * checksummed and sums are attached onto the ordered extent record.
1428  *
1429  * At IO completion time the cums attached on the ordered extent record
1430  * are inserted into the btree
1431  */
1432 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1433                           int mirror_num, unsigned long bio_flags,
1434                           u64 bio_offset)
1435 {
1436         struct btrfs_root *root = BTRFS_I(inode)->root;
1437         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1438 }
1439
1440 /*
1441  * extent_io.c submission hook. This does the right thing for csum calculation
1442  * on write, or reading the csums from the tree before a read
1443  */
1444 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1445                           int mirror_num, unsigned long bio_flags,
1446                           u64 bio_offset)
1447 {
1448         struct btrfs_root *root = BTRFS_I(inode)->root;
1449         int ret = 0;
1450         int skip_sum;
1451
1452         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1453
1454         if (root == root->fs_info->tree_root)
1455                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1456         else
1457                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1458         BUG_ON(ret);
1459
1460         if (!(rw & REQ_WRITE)) {
1461                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1462                         return btrfs_submit_compressed_read(inode, bio,
1463                                                     mirror_num, bio_flags);
1464                 } else if (!skip_sum)
1465                         btrfs_lookup_bio_sums(root, inode, bio, NULL);
1466                 goto mapit;
1467         } else if (!skip_sum) {
1468                 /* csum items have already been cloned */
1469                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1470                         goto mapit;
1471                 /* we're doing a write, do the async checksumming */
1472                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1473                                    inode, rw, bio, mirror_num,
1474                                    bio_flags, bio_offset,
1475                                    __btrfs_submit_bio_start,
1476                                    __btrfs_submit_bio_done);
1477         }
1478
1479 mapit:
1480         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1481 }
1482
1483 /*
1484  * given a list of ordered sums record them in the inode.  This happens
1485  * at IO completion time based on sums calculated at bio submission time.
1486  */
1487 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1488                              struct inode *inode, u64 file_offset,
1489                              struct list_head *list)
1490 {
1491         struct btrfs_ordered_sum *sum;
1492
1493         btrfs_set_trans_block_group(trans, inode);
1494
1495         list_for_each_entry(sum, list, list) {
1496                 btrfs_csum_file_blocks(trans,
1497                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1498         }
1499         return 0;
1500 }
1501
1502 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1503                               struct extent_state **cached_state)
1504 {
1505         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1506                 WARN_ON(1);
1507         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1508                                    cached_state, GFP_NOFS);
1509 }
1510
1511 /* see btrfs_writepage_start_hook for details on why this is required */
1512 struct btrfs_writepage_fixup {
1513         struct page *page;
1514         struct btrfs_work work;
1515 };
1516
1517 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1518 {
1519         struct btrfs_writepage_fixup *fixup;
1520         struct btrfs_ordered_extent *ordered;
1521         struct extent_state *cached_state = NULL;
1522         struct page *page;
1523         struct inode *inode;
1524         u64 page_start;
1525         u64 page_end;
1526
1527         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1528         page = fixup->page;
1529 again:
1530         lock_page(page);
1531         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1532                 ClearPageChecked(page);
1533                 goto out_page;
1534         }
1535
1536         inode = page->mapping->host;
1537         page_start = page_offset(page);
1538         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1539
1540         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1541                          &cached_state, GFP_NOFS);
1542
1543         /* already ordered? We're done */
1544         if (PagePrivate2(page))
1545                 goto out;
1546
1547         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1548         if (ordered) {
1549                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1550                                      page_end, &cached_state, GFP_NOFS);
1551                 unlock_page(page);
1552                 btrfs_start_ordered_extent(inode, ordered, 1);
1553                 goto again;
1554         }
1555
1556         BUG();
1557         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1558         ClearPageChecked(page);
1559 out:
1560         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1561                              &cached_state, GFP_NOFS);
1562 out_page:
1563         unlock_page(page);
1564         page_cache_release(page);
1565         kfree(fixup);
1566 }
1567
1568 /*
1569  * There are a few paths in the higher layers of the kernel that directly
1570  * set the page dirty bit without asking the filesystem if it is a
1571  * good idea.  This causes problems because we want to make sure COW
1572  * properly happens and the data=ordered rules are followed.
1573  *
1574  * In our case any range that doesn't have the ORDERED bit set
1575  * hasn't been properly setup for IO.  We kick off an async process
1576  * to fix it up.  The async helper will wait for ordered extents, set
1577  * the delalloc bit and make it safe to write the page.
1578  */
1579 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1580 {
1581         struct inode *inode = page->mapping->host;
1582         struct btrfs_writepage_fixup *fixup;
1583         struct btrfs_root *root = BTRFS_I(inode)->root;
1584
1585         /* this page is properly in the ordered list */
1586         if (TestClearPagePrivate2(page))
1587                 return 0;
1588
1589         if (PageChecked(page))
1590                 return -EAGAIN;
1591
1592         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1593         if (!fixup)
1594                 return -EAGAIN;
1595
1596         SetPageChecked(page);
1597         page_cache_get(page);
1598         fixup->work.func = btrfs_writepage_fixup_worker;
1599         fixup->page = page;
1600         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1601         return -EAGAIN;
1602 }
1603
1604 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1605                                        struct inode *inode, u64 file_pos,
1606                                        u64 disk_bytenr, u64 disk_num_bytes,
1607                                        u64 num_bytes, u64 ram_bytes,
1608                                        u8 compression, u8 encryption,
1609                                        u16 other_encoding, int extent_type)
1610 {
1611         struct btrfs_root *root = BTRFS_I(inode)->root;
1612         struct btrfs_file_extent_item *fi;
1613         struct btrfs_path *path;
1614         struct extent_buffer *leaf;
1615         struct btrfs_key ins;
1616         u64 hint;
1617         int ret;
1618
1619         path = btrfs_alloc_path();
1620         BUG_ON(!path);
1621
1622         path->leave_spinning = 1;
1623
1624         /*
1625          * we may be replacing one extent in the tree with another.
1626          * The new extent is pinned in the extent map, and we don't want
1627          * to drop it from the cache until it is completely in the btree.
1628          *
1629          * So, tell btrfs_drop_extents to leave this extent in the cache.
1630          * the caller is expected to unpin it and allow it to be merged
1631          * with the others.
1632          */
1633         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1634                                  &hint, 0);
1635         BUG_ON(ret);
1636
1637         ins.objectid = inode->i_ino;
1638         ins.offset = file_pos;
1639         ins.type = BTRFS_EXTENT_DATA_KEY;
1640         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1641         BUG_ON(ret);
1642         leaf = path->nodes[0];
1643         fi = btrfs_item_ptr(leaf, path->slots[0],
1644                             struct btrfs_file_extent_item);
1645         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1646         btrfs_set_file_extent_type(leaf, fi, extent_type);
1647         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1648         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1649         btrfs_set_file_extent_offset(leaf, fi, 0);
1650         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1651         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1652         btrfs_set_file_extent_compression(leaf, fi, compression);
1653         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1654         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1655
1656         btrfs_unlock_up_safe(path, 1);
1657         btrfs_set_lock_blocking(leaf);
1658
1659         btrfs_mark_buffer_dirty(leaf);
1660
1661         inode_add_bytes(inode, num_bytes);
1662
1663         ins.objectid = disk_bytenr;
1664         ins.offset = disk_num_bytes;
1665         ins.type = BTRFS_EXTENT_ITEM_KEY;
1666         ret = btrfs_alloc_reserved_file_extent(trans, root,
1667                                         root->root_key.objectid,
1668                                         inode->i_ino, file_pos, &ins);
1669         BUG_ON(ret);
1670         btrfs_free_path(path);
1671
1672         return 0;
1673 }
1674
1675 /*
1676  * helper function for btrfs_finish_ordered_io, this
1677  * just reads in some of the csum leaves to prime them into ram
1678  * before we start the transaction.  It limits the amount of btree
1679  * reads required while inside the transaction.
1680  */
1681 /* as ordered data IO finishes, this gets called so we can finish
1682  * an ordered extent if the range of bytes in the file it covers are
1683  * fully written.
1684  */
1685 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1686 {
1687         struct btrfs_root *root = BTRFS_I(inode)->root;
1688         struct btrfs_trans_handle *trans = NULL;
1689         struct btrfs_ordered_extent *ordered_extent = NULL;
1690         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1691         struct extent_state *cached_state = NULL;
1692         int compress_type = 0;
1693         int ret;
1694         bool nolock = false;
1695
1696         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1697                                              end - start + 1);
1698         if (!ret)
1699                 return 0;
1700         BUG_ON(!ordered_extent);
1701
1702         nolock = (root == root->fs_info->tree_root);
1703
1704         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1705                 BUG_ON(!list_empty(&ordered_extent->list));
1706                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1707                 if (!ret) {
1708                         if (nolock)
1709                                 trans = btrfs_join_transaction_nolock(root, 1);
1710                         else
1711                                 trans = btrfs_join_transaction(root, 1);
1712                         BUG_ON(IS_ERR(trans));
1713                         btrfs_set_trans_block_group(trans, inode);
1714                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1715                         ret = btrfs_update_inode(trans, root, inode);
1716                         BUG_ON(ret);
1717                 }
1718                 goto out;
1719         }
1720
1721         lock_extent_bits(io_tree, ordered_extent->file_offset,
1722                          ordered_extent->file_offset + ordered_extent->len - 1,
1723                          0, &cached_state, GFP_NOFS);
1724
1725         if (nolock)
1726                 trans = btrfs_join_transaction_nolock(root, 1);
1727         else
1728                 trans = btrfs_join_transaction(root, 1);
1729         BUG_ON(IS_ERR(trans));
1730         btrfs_set_trans_block_group(trans, inode);
1731         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1732
1733         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1734                 compress_type = ordered_extent->compress_type;
1735         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1736                 BUG_ON(compress_type);
1737                 ret = btrfs_mark_extent_written(trans, inode,
1738                                                 ordered_extent->file_offset,
1739                                                 ordered_extent->file_offset +
1740                                                 ordered_extent->len);
1741                 BUG_ON(ret);
1742         } else {
1743                 BUG_ON(root == root->fs_info->tree_root);
1744                 ret = insert_reserved_file_extent(trans, inode,
1745                                                 ordered_extent->file_offset,
1746                                                 ordered_extent->start,
1747                                                 ordered_extent->disk_len,
1748                                                 ordered_extent->len,
1749                                                 ordered_extent->len,
1750                                                 compress_type, 0, 0,
1751                                                 BTRFS_FILE_EXTENT_REG);
1752                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1753                                    ordered_extent->file_offset,
1754                                    ordered_extent->len);
1755                 BUG_ON(ret);
1756         }
1757         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1758                              ordered_extent->file_offset +
1759                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1760
1761         add_pending_csums(trans, inode, ordered_extent->file_offset,
1762                           &ordered_extent->list);
1763
1764         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1765         ret = btrfs_update_inode(trans, root, inode);
1766         BUG_ON(ret);
1767 out:
1768         if (nolock) {
1769                 if (trans)
1770                         btrfs_end_transaction_nolock(trans, root);
1771         } else {
1772                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1773                 if (trans)
1774                         btrfs_end_transaction(trans, root);
1775         }
1776
1777         /* once for us */
1778         btrfs_put_ordered_extent(ordered_extent);
1779         /* once for the tree */
1780         btrfs_put_ordered_extent(ordered_extent);
1781
1782         return 0;
1783 }
1784
1785 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1786                                 struct extent_state *state, int uptodate)
1787 {
1788         ClearPagePrivate2(page);
1789         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1790 }
1791
1792 /*
1793  * When IO fails, either with EIO or csum verification fails, we
1794  * try other mirrors that might have a good copy of the data.  This
1795  * io_failure_record is used to record state as we go through all the
1796  * mirrors.  If another mirror has good data, the page is set up to date
1797  * and things continue.  If a good mirror can't be found, the original
1798  * bio end_io callback is called to indicate things have failed.
1799  */
1800 struct io_failure_record {
1801         struct page *page;
1802         u64 start;
1803         u64 len;
1804         u64 logical;
1805         unsigned long bio_flags;
1806         int last_mirror;
1807 };
1808
1809 static int btrfs_io_failed_hook(struct bio *failed_bio,
1810                          struct page *page, u64 start, u64 end,
1811                          struct extent_state *state)
1812 {
1813         struct io_failure_record *failrec = NULL;
1814         u64 private;
1815         struct extent_map *em;
1816         struct inode *inode = page->mapping->host;
1817         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1818         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1819         struct bio *bio;
1820         int num_copies;
1821         int ret;
1822         int rw;
1823         u64 logical;
1824
1825         ret = get_state_private(failure_tree, start, &private);
1826         if (ret) {
1827                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1828                 if (!failrec)
1829                         return -ENOMEM;
1830                 failrec->start = start;
1831                 failrec->len = end - start + 1;
1832                 failrec->last_mirror = 0;
1833                 failrec->bio_flags = 0;
1834
1835                 read_lock(&em_tree->lock);
1836                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1837                 if (em->start > start || em->start + em->len < start) {
1838                         free_extent_map(em);
1839                         em = NULL;
1840                 }
1841                 read_unlock(&em_tree->lock);
1842
1843                 if (!em || IS_ERR(em)) {
1844                         kfree(failrec);
1845                         return -EIO;
1846                 }
1847                 logical = start - em->start;
1848                 logical = em->block_start + logical;
1849                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1850                         logical = em->block_start;
1851                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1852                         extent_set_compress_type(&failrec->bio_flags,
1853                                                  em->compress_type);
1854                 }
1855                 failrec->logical = logical;
1856                 free_extent_map(em);
1857                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1858                                 EXTENT_DIRTY, GFP_NOFS);
1859                 set_state_private(failure_tree, start,
1860                                  (u64)(unsigned long)failrec);
1861         } else {
1862                 failrec = (struct io_failure_record *)(unsigned long)private;
1863         }
1864         num_copies = btrfs_num_copies(
1865                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1866                               failrec->logical, failrec->len);
1867         failrec->last_mirror++;
1868         if (!state) {
1869                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1870                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1871                                                     failrec->start,
1872                                                     EXTENT_LOCKED);
1873                 if (state && state->start != failrec->start)
1874                         state = NULL;
1875                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1876         }
1877         if (!state || failrec->last_mirror > num_copies) {
1878                 set_state_private(failure_tree, failrec->start, 0);
1879                 clear_extent_bits(failure_tree, failrec->start,
1880                                   failrec->start + failrec->len - 1,
1881                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1882                 kfree(failrec);
1883                 return -EIO;
1884         }
1885         bio = bio_alloc(GFP_NOFS, 1);
1886         bio->bi_private = state;
1887         bio->bi_end_io = failed_bio->bi_end_io;
1888         bio->bi_sector = failrec->logical >> 9;
1889         bio->bi_bdev = failed_bio->bi_bdev;
1890         bio->bi_size = 0;
1891
1892         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1893         if (failed_bio->bi_rw & REQ_WRITE)
1894                 rw = WRITE;
1895         else
1896                 rw = READ;
1897
1898         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1899                                                       failrec->last_mirror,
1900                                                       failrec->bio_flags, 0);
1901         return 0;
1902 }
1903
1904 /*
1905  * each time an IO finishes, we do a fast check in the IO failure tree
1906  * to see if we need to process or clean up an io_failure_record
1907  */
1908 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1909 {
1910         u64 private;
1911         u64 private_failure;
1912         struct io_failure_record *failure;
1913         int ret;
1914
1915         private = 0;
1916         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1917                              (u64)-1, 1, EXTENT_DIRTY, 0)) {
1918                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1919                                         start, &private_failure);
1920                 if (ret == 0) {
1921                         failure = (struct io_failure_record *)(unsigned long)
1922                                    private_failure;
1923                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1924                                           failure->start, 0);
1925                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1926                                           failure->start,
1927                                           failure->start + failure->len - 1,
1928                                           EXTENT_DIRTY | EXTENT_LOCKED,
1929                                           GFP_NOFS);
1930                         kfree(failure);
1931                 }
1932         }
1933         return 0;
1934 }
1935
1936 /*
1937  * when reads are done, we need to check csums to verify the data is correct
1938  * if there's a match, we allow the bio to finish.  If not, we go through
1939  * the io_failure_record routines to find good copies
1940  */
1941 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1942                                struct extent_state *state)
1943 {
1944         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1945         struct inode *inode = page->mapping->host;
1946         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1947         char *kaddr;
1948         u64 private = ~(u32)0;
1949         int ret;
1950         struct btrfs_root *root = BTRFS_I(inode)->root;
1951         u32 csum = ~(u32)0;
1952
1953         if (PageChecked(page)) {
1954                 ClearPageChecked(page);
1955                 goto good;
1956         }
1957
1958         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1959                 return 0;
1960
1961         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1962             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1963                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1964                                   GFP_NOFS);
1965                 return 0;
1966         }
1967
1968         if (state && state->start == start) {
1969                 private = state->private;
1970                 ret = 0;
1971         } else {
1972                 ret = get_state_private(io_tree, start, &private);
1973         }
1974         kaddr = kmap_atomic(page, KM_USER0);
1975         if (ret)
1976                 goto zeroit;
1977
1978         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1979         btrfs_csum_final(csum, (char *)&csum);
1980         if (csum != private)
1981                 goto zeroit;
1982
1983         kunmap_atomic(kaddr, KM_USER0);
1984 good:
1985         /* if the io failure tree for this inode is non-empty,
1986          * check to see if we've recovered from a failed IO
1987          */
1988         btrfs_clean_io_failures(inode, start);
1989         return 0;
1990
1991 zeroit:
1992         if (printk_ratelimit()) {
1993                 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1994                        "private %llu\n", page->mapping->host->i_ino,
1995                        (unsigned long long)start, csum,
1996                        (unsigned long long)private);
1997         }
1998         memset(kaddr + offset, 1, end - start + 1);
1999         flush_dcache_page(page);
2000         kunmap_atomic(kaddr, KM_USER0);
2001         if (private == 0)
2002                 return 0;
2003         return -EIO;
2004 }
2005
2006 struct delayed_iput {
2007         struct list_head list;
2008         struct inode *inode;
2009 };
2010
2011 void btrfs_add_delayed_iput(struct inode *inode)
2012 {
2013         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2014         struct delayed_iput *delayed;
2015
2016         if (atomic_add_unless(&inode->i_count, -1, 1))
2017                 return;
2018
2019         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2020         delayed->inode = inode;
2021
2022         spin_lock(&fs_info->delayed_iput_lock);
2023         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2024         spin_unlock(&fs_info->delayed_iput_lock);
2025 }
2026
2027 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2028 {
2029         LIST_HEAD(list);
2030         struct btrfs_fs_info *fs_info = root->fs_info;
2031         struct delayed_iput *delayed;
2032         int empty;
2033
2034         spin_lock(&fs_info->delayed_iput_lock);
2035         empty = list_empty(&fs_info->delayed_iputs);
2036         spin_unlock(&fs_info->delayed_iput_lock);
2037         if (empty)
2038                 return;
2039
2040         down_read(&root->fs_info->cleanup_work_sem);
2041         spin_lock(&fs_info->delayed_iput_lock);
2042         list_splice_init(&fs_info->delayed_iputs, &list);
2043         spin_unlock(&fs_info->delayed_iput_lock);
2044
2045         while (!list_empty(&list)) {
2046                 delayed = list_entry(list.next, struct delayed_iput, list);
2047                 list_del(&delayed->list);
2048                 iput(delayed->inode);
2049                 kfree(delayed);
2050         }
2051         up_read(&root->fs_info->cleanup_work_sem);
2052 }
2053
2054 /*
2055  * calculate extra metadata reservation when snapshotting a subvolume
2056  * contains orphan files.
2057  */
2058 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2059                                 struct btrfs_pending_snapshot *pending,
2060                                 u64 *bytes_to_reserve)
2061 {
2062         struct btrfs_root *root;
2063         struct btrfs_block_rsv *block_rsv;
2064         u64 num_bytes;
2065         int index;
2066
2067         root = pending->root;
2068         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2069                 return;
2070
2071         block_rsv = root->orphan_block_rsv;
2072
2073         /* orphan block reservation for the snapshot */
2074         num_bytes = block_rsv->size;
2075
2076         /*
2077          * after the snapshot is created, COWing tree blocks may use more
2078          * space than it frees. So we should make sure there is enough
2079          * reserved space.
2080          */
2081         index = trans->transid & 0x1;
2082         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2083                 num_bytes += block_rsv->size -
2084                              (block_rsv->reserved + block_rsv->freed[index]);
2085         }
2086
2087         *bytes_to_reserve += num_bytes;
2088 }
2089
2090 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2091                                 struct btrfs_pending_snapshot *pending)
2092 {
2093         struct btrfs_root *root = pending->root;
2094         struct btrfs_root *snap = pending->snap;
2095         struct btrfs_block_rsv *block_rsv;
2096         u64 num_bytes;
2097         int index;
2098         int ret;
2099
2100         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2101                 return;
2102
2103         /* refill source subvolume's orphan block reservation */
2104         block_rsv = root->orphan_block_rsv;
2105         index = trans->transid & 0x1;
2106         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2107                 num_bytes = block_rsv->size -
2108                             (block_rsv->reserved + block_rsv->freed[index]);
2109                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2110                                               root->orphan_block_rsv,
2111                                               num_bytes);
2112                 BUG_ON(ret);
2113         }
2114
2115         /* setup orphan block reservation for the snapshot */
2116         block_rsv = btrfs_alloc_block_rsv(snap);
2117         BUG_ON(!block_rsv);
2118
2119         btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2120         snap->orphan_block_rsv = block_rsv;
2121
2122         num_bytes = root->orphan_block_rsv->size;
2123         ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2124                                       block_rsv, num_bytes);
2125         BUG_ON(ret);
2126
2127 #if 0
2128         /* insert orphan item for the snapshot */
2129         WARN_ON(!root->orphan_item_inserted);
2130         ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2131                                        snap->root_key.objectid);
2132         BUG_ON(ret);
2133         snap->orphan_item_inserted = 1;
2134 #endif
2135 }
2136
2137 enum btrfs_orphan_cleanup_state {
2138         ORPHAN_CLEANUP_STARTED  = 1,
2139         ORPHAN_CLEANUP_DONE     = 2,
2140 };
2141
2142 /*
2143  * This is called in transaction commmit time. If there are no orphan
2144  * files in the subvolume, it removes orphan item and frees block_rsv
2145  * structure.
2146  */
2147 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2148                               struct btrfs_root *root)
2149 {
2150         int ret;
2151
2152         if (!list_empty(&root->orphan_list) ||
2153             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2154                 return;
2155
2156         if (root->orphan_item_inserted &&
2157             btrfs_root_refs(&root->root_item) > 0) {
2158                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2159                                             root->root_key.objectid);
2160                 BUG_ON(ret);
2161                 root->orphan_item_inserted = 0;
2162         }
2163
2164         if (root->orphan_block_rsv) {
2165                 WARN_ON(root->orphan_block_rsv->size > 0);
2166                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2167                 root->orphan_block_rsv = NULL;
2168         }
2169 }
2170
2171 /*
2172  * This creates an orphan entry for the given inode in case something goes
2173  * wrong in the middle of an unlink/truncate.
2174  *
2175  * NOTE: caller of this function should reserve 5 units of metadata for
2176  *       this function.
2177  */
2178 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2179 {
2180         struct btrfs_root *root = BTRFS_I(inode)->root;
2181         struct btrfs_block_rsv *block_rsv = NULL;
2182         int reserve = 0;
2183         int insert = 0;
2184         int ret;
2185
2186         if (!root->orphan_block_rsv) {
2187                 block_rsv = btrfs_alloc_block_rsv(root);
2188                 BUG_ON(!block_rsv);
2189         }
2190
2191         spin_lock(&root->orphan_lock);
2192         if (!root->orphan_block_rsv) {
2193                 root->orphan_block_rsv = block_rsv;
2194         } else if (block_rsv) {
2195                 btrfs_free_block_rsv(root, block_rsv);
2196                 block_rsv = NULL;
2197         }
2198
2199         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2200                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2201 #if 0
2202                 /*
2203                  * For proper ENOSPC handling, we should do orphan
2204                  * cleanup when mounting. But this introduces backward
2205                  * compatibility issue.
2206                  */
2207                 if (!xchg(&root->orphan_item_inserted, 1))
2208                         insert = 2;
2209                 else
2210                         insert = 1;
2211 #endif
2212                 insert = 1;
2213         } else {
2214                 WARN_ON(!BTRFS_I(inode)->orphan_meta_reserved);
2215         }
2216
2217         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2218                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2219                 reserve = 1;
2220         }
2221         spin_unlock(&root->orphan_lock);
2222
2223         if (block_rsv)
2224                 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2225
2226         /* grab metadata reservation from transaction handle */
2227         if (reserve) {
2228                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2229                 BUG_ON(ret);
2230         }
2231
2232         /* insert an orphan item to track this unlinked/truncated file */
2233         if (insert >= 1) {
2234                 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2235                 BUG_ON(ret);
2236         }
2237
2238         /* insert an orphan item to track subvolume contains orphan files */
2239         if (insert >= 2) {
2240                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2241                                                root->root_key.objectid);
2242                 BUG_ON(ret);
2243         }
2244         return 0;
2245 }
2246
2247 /*
2248  * We have done the truncate/delete so we can go ahead and remove the orphan
2249  * item for this particular inode.
2250  */
2251 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2252 {
2253         struct btrfs_root *root = BTRFS_I(inode)->root;
2254         int delete_item = 0;
2255         int release_rsv = 0;
2256         int ret = 0;
2257
2258         spin_lock(&root->orphan_lock);
2259         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2260                 list_del_init(&BTRFS_I(inode)->i_orphan);
2261                 delete_item = 1;
2262         }
2263
2264         if (BTRFS_I(inode)->orphan_meta_reserved) {
2265                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2266                 release_rsv = 1;
2267         }
2268         spin_unlock(&root->orphan_lock);
2269
2270         if (trans && delete_item) {
2271                 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2272                 BUG_ON(ret);
2273         }
2274
2275         if (release_rsv)
2276                 btrfs_orphan_release_metadata(inode);
2277
2278         return 0;
2279 }
2280
2281 /*
2282  * this cleans up any orphans that may be left on the list from the last use
2283  * of this root.
2284  */
2285 void btrfs_orphan_cleanup(struct btrfs_root *root)
2286 {
2287         struct btrfs_path *path;
2288         struct extent_buffer *leaf;
2289         struct btrfs_key key, found_key;
2290         struct btrfs_trans_handle *trans;
2291         struct inode *inode;
2292         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2293
2294         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2295                 return;
2296
2297         path = btrfs_alloc_path();
2298         BUG_ON(!path);
2299         path->reada = -1;
2300
2301         key.objectid = BTRFS_ORPHAN_OBJECTID;
2302         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2303         key.offset = (u64)-1;
2304
2305         while (1) {
2306                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2307                 if (ret < 0) {
2308                         printk(KERN_ERR "Error searching slot for orphan: %d"
2309                                "\n", ret);
2310                         break;
2311                 }
2312
2313                 /*
2314                  * if ret == 0 means we found what we were searching for, which
2315                  * is weird, but possible, so only screw with path if we didnt
2316                  * find the key and see if we have stuff that matches
2317                  */
2318                 if (ret > 0) {
2319                         if (path->slots[0] == 0)
2320                                 break;
2321                         path->slots[0]--;
2322                 }
2323
2324                 /* pull out the item */
2325                 leaf = path->nodes[0];
2326                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2327
2328                 /* make sure the item matches what we want */
2329                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2330                         break;
2331                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2332                         break;
2333
2334                 /* release the path since we're done with it */
2335                 btrfs_release_path(root, path);
2336
2337                 /*
2338                  * this is where we are basically btrfs_lookup, without the
2339                  * crossing root thing.  we store the inode number in the
2340                  * offset of the orphan item.
2341                  */
2342                 found_key.objectid = found_key.offset;
2343                 found_key.type = BTRFS_INODE_ITEM_KEY;
2344                 found_key.offset = 0;
2345                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2346                 BUG_ON(IS_ERR(inode));
2347
2348                 /*
2349                  * add this inode to the orphan list so btrfs_orphan_del does
2350                  * the proper thing when we hit it
2351                  */
2352                 spin_lock(&root->orphan_lock);
2353                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2354                 spin_unlock(&root->orphan_lock);
2355
2356                 /*
2357                  * if this is a bad inode, means we actually succeeded in
2358                  * removing the inode, but not the orphan record, which means
2359                  * we need to manually delete the orphan since iput will just
2360                  * do a destroy_inode
2361                  */
2362                 if (is_bad_inode(inode)) {
2363                         trans = btrfs_start_transaction(root, 0);
2364                         BUG_ON(IS_ERR(trans));
2365                         btrfs_orphan_del(trans, inode);
2366                         btrfs_end_transaction(trans, root);
2367                         iput(inode);
2368                         continue;
2369                 }
2370
2371                 /* if we have links, this was a truncate, lets do that */
2372                 if (inode->i_nlink) {
2373                         nr_truncate++;
2374                         btrfs_truncate(inode);
2375                 } else {
2376                         nr_unlink++;
2377                 }
2378
2379                 /* this will do delete_inode and everything for us */
2380                 iput(inode);
2381         }
2382         btrfs_free_path(path);
2383
2384         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2385
2386         if (root->orphan_block_rsv)
2387                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2388                                         (u64)-1);
2389
2390         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2391                 trans = btrfs_join_transaction(root, 1);
2392                 BUG_ON(IS_ERR(trans));
2393                 btrfs_end_transaction(trans, root);
2394         }
2395
2396         if (nr_unlink)
2397                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2398         if (nr_truncate)
2399                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2400 }
2401
2402 /*
2403  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2404  * don't find any xattrs, we know there can't be any acls.
2405  *
2406  * slot is the slot the inode is in, objectid is the objectid of the inode
2407  */
2408 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2409                                           int slot, u64 objectid)
2410 {
2411         u32 nritems = btrfs_header_nritems(leaf);
2412         struct btrfs_key found_key;
2413         int scanned = 0;
2414
2415         slot++;
2416         while (slot < nritems) {
2417                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2418
2419                 /* we found a different objectid, there must not be acls */
2420                 if (found_key.objectid != objectid)
2421                         return 0;
2422
2423                 /* we found an xattr, assume we've got an acl */
2424                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2425                         return 1;
2426
2427                 /*
2428                  * we found a key greater than an xattr key, there can't
2429                  * be any acls later on
2430                  */
2431                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2432                         return 0;
2433
2434                 slot++;
2435                 scanned++;
2436
2437                 /*
2438                  * it goes inode, inode backrefs, xattrs, extents,
2439                  * so if there are a ton of hard links to an inode there can
2440                  * be a lot of backrefs.  Don't waste time searching too hard,
2441                  * this is just an optimization
2442                  */
2443                 if (scanned >= 8)
2444                         break;
2445         }
2446         /* we hit the end of the leaf before we found an xattr or
2447          * something larger than an xattr.  We have to assume the inode
2448          * has acls
2449          */
2450         return 1;
2451 }
2452
2453 /*
2454  * read an inode from the btree into the in-memory inode
2455  */
2456 static void btrfs_read_locked_inode(struct inode *inode)
2457 {
2458         struct btrfs_path *path;
2459         struct extent_buffer *leaf;
2460         struct btrfs_inode_item *inode_item;
2461         struct btrfs_timespec *tspec;
2462         struct btrfs_root *root = BTRFS_I(inode)->root;
2463         struct btrfs_key location;
2464         int maybe_acls;
2465         u64 alloc_group_block;
2466         u32 rdev;
2467         int ret;
2468
2469         path = btrfs_alloc_path();
2470         BUG_ON(!path);
2471         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2472
2473         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2474         if (ret)
2475                 goto make_bad;
2476
2477         leaf = path->nodes[0];
2478         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2479                                     struct btrfs_inode_item);
2480
2481         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2482         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2483         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2484         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2485         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2486
2487         tspec = btrfs_inode_atime(inode_item);
2488         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2489         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2490
2491         tspec = btrfs_inode_mtime(inode_item);
2492         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2493         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2494
2495         tspec = btrfs_inode_ctime(inode_item);
2496         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2497         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2498
2499         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2500         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2501         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2502         inode->i_generation = BTRFS_I(inode)->generation;
2503         inode->i_rdev = 0;
2504         rdev = btrfs_inode_rdev(leaf, inode_item);
2505
2506         BTRFS_I(inode)->index_cnt = (u64)-1;
2507         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2508
2509         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2510
2511         /*
2512          * try to precache a NULL acl entry for files that don't have
2513          * any xattrs or acls
2514          */
2515         maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2516         if (!maybe_acls)
2517                 cache_no_acl(inode);
2518
2519         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2520                                                 alloc_group_block, 0);
2521         btrfs_free_path(path);
2522         inode_item = NULL;
2523
2524         switch (inode->i_mode & S_IFMT) {
2525         case S_IFREG:
2526                 inode->i_mapping->a_ops = &btrfs_aops;
2527                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2528                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2529                 inode->i_fop = &btrfs_file_operations;
2530                 inode->i_op = &btrfs_file_inode_operations;
2531                 break;
2532         case S_IFDIR:
2533                 inode->i_fop = &btrfs_dir_file_operations;
2534                 if (root == root->fs_info->tree_root)
2535                         inode->i_op = &btrfs_dir_ro_inode_operations;
2536                 else
2537                         inode->i_op = &btrfs_dir_inode_operations;
2538                 break;
2539         case S_IFLNK:
2540                 inode->i_op = &btrfs_symlink_inode_operations;
2541                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2542                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2543                 break;
2544         default:
2545                 inode->i_op = &btrfs_special_inode_operations;
2546                 init_special_inode(inode, inode->i_mode, rdev);
2547                 break;
2548         }
2549
2550         btrfs_update_iflags(inode);
2551         return;
2552
2553 make_bad:
2554         btrfs_free_path(path);
2555         make_bad_inode(inode);
2556 }
2557
2558 /*
2559  * given a leaf and an inode, copy the inode fields into the leaf
2560  */
2561 static void fill_inode_item(struct btrfs_trans_handle *trans,
2562                             struct extent_buffer *leaf,
2563                             struct btrfs_inode_item *item,
2564                             struct inode *inode)
2565 {
2566         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2567         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2568         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2569         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2570         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2571
2572         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2573                                inode->i_atime.tv_sec);
2574         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2575                                 inode->i_atime.tv_nsec);
2576
2577         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2578                                inode->i_mtime.tv_sec);
2579         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2580                                 inode->i_mtime.tv_nsec);
2581
2582         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2583                                inode->i_ctime.tv_sec);
2584         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2585                                 inode->i_ctime.tv_nsec);
2586
2587         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2588         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2589         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2590         btrfs_set_inode_transid(leaf, item, trans->transid);
2591         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2592         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2593         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2594 }
2595
2596 /*
2597  * copy everything in the in-memory inode into the btree.
2598  */
2599 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2600                                 struct btrfs_root *root, struct inode *inode)
2601 {
2602         struct btrfs_inode_item *inode_item;
2603         struct btrfs_path *path;
2604         struct extent_buffer *leaf;
2605         int ret;
2606
2607         path = btrfs_alloc_path();
2608         BUG_ON(!path);
2609         path->leave_spinning = 1;
2610         ret = btrfs_lookup_inode(trans, root, path,
2611                                  &BTRFS_I(inode)->location, 1);
2612         if (ret) {
2613                 if (ret > 0)
2614                         ret = -ENOENT;
2615                 goto failed;
2616         }
2617
2618         btrfs_unlock_up_safe(path, 1);
2619         leaf = path->nodes[0];
2620         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2621                                   struct btrfs_inode_item);
2622
2623         fill_inode_item(trans, leaf, inode_item, inode);
2624         btrfs_mark_buffer_dirty(leaf);
2625         btrfs_set_inode_last_trans(trans, inode);
2626         ret = 0;
2627 failed:
2628         btrfs_free_path(path);
2629         return ret;
2630 }
2631
2632
2633 /*
2634  * unlink helper that gets used here in inode.c and in the tree logging
2635  * recovery code.  It remove a link in a directory with a given name, and
2636  * also drops the back refs in the inode to the directory
2637  */
2638 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2639                        struct btrfs_root *root,
2640                        struct inode *dir, struct inode *inode,
2641                        const char *name, int name_len)
2642 {
2643         struct btrfs_path *path;
2644         int ret = 0;
2645         struct extent_buffer *leaf;
2646         struct btrfs_dir_item *di;
2647         struct btrfs_key key;
2648         u64 index;
2649
2650         path = btrfs_alloc_path();
2651         if (!path) {
2652                 ret = -ENOMEM;
2653                 goto out;
2654         }
2655
2656         path->leave_spinning = 1;
2657         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2658                                     name, name_len, -1);
2659         if (IS_ERR(di)) {
2660                 ret = PTR_ERR(di);
2661                 goto err;
2662         }
2663         if (!di) {
2664                 ret = -ENOENT;
2665                 goto err;
2666         }
2667         leaf = path->nodes[0];
2668         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2669         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2670         if (ret)
2671                 goto err;
2672         btrfs_release_path(root, path);
2673
2674         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2675                                   inode->i_ino,
2676                                   dir->i_ino, &index);
2677         if (ret) {
2678                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2679                        "inode %lu parent %lu\n", name_len, name,
2680                        inode->i_ino, dir->i_ino);
2681                 goto err;
2682         }
2683
2684         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2685                                          index, name, name_len, -1);
2686         if (IS_ERR(di)) {
2687                 ret = PTR_ERR(di);
2688                 goto err;
2689         }
2690         if (!di) {
2691                 ret = -ENOENT;
2692                 goto err;
2693         }
2694         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2695         btrfs_release_path(root, path);
2696
2697         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2698                                          inode, dir->i_ino);
2699         BUG_ON(ret != 0 && ret != -ENOENT);
2700
2701         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2702                                            dir, index);
2703         if (ret == -ENOENT)
2704                 ret = 0;
2705 err:
2706         btrfs_free_path(path);
2707         if (ret)
2708                 goto out;
2709
2710         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2711         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2712         btrfs_update_inode(trans, root, dir);
2713         btrfs_drop_nlink(inode);
2714         ret = btrfs_update_inode(trans, root, inode);
2715 out:
2716         return ret;
2717 }
2718
2719 /* helper to check if there is any shared block in the path */
2720 static int check_path_shared(struct btrfs_root *root,
2721                              struct btrfs_path *path)
2722 {
2723         struct extent_buffer *eb;
2724         int level;
2725         u64 refs = 1;
2726
2727         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2728                 int ret;
2729
2730                 if (!path->nodes[level])
2731                         break;
2732                 eb = path->nodes[level];
2733                 if (!btrfs_block_can_be_shared(root, eb))
2734                         continue;
2735                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2736                                                &refs, NULL);
2737                 if (refs > 1)
2738                         return 1;
2739         }
2740         return 0;
2741 }
2742
2743 /*
2744  * helper to start transaction for unlink and rmdir.
2745  *
2746  * unlink and rmdir are special in btrfs, they do not always free space.
2747  * so in enospc case, we should make sure they will free space before
2748  * allowing them to use the global metadata reservation.
2749  */
2750 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2751                                                        struct dentry *dentry)
2752 {
2753         struct btrfs_trans_handle *trans;
2754         struct btrfs_root *root = BTRFS_I(dir)->root;
2755         struct btrfs_path *path;
2756         struct btrfs_inode_ref *ref;
2757         struct btrfs_dir_item *di;
2758         struct inode *inode = dentry->d_inode;
2759         u64 index;
2760         int check_link = 1;
2761         int err = -ENOSPC;
2762         int ret;
2763
2764         trans = btrfs_start_transaction(root, 10);
2765         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2766                 return trans;
2767
2768         if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2769                 return ERR_PTR(-ENOSPC);
2770
2771         /* check if there is someone else holds reference */
2772         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2773                 return ERR_PTR(-ENOSPC);
2774
2775         if (atomic_read(&inode->i_count) > 2)
2776                 return ERR_PTR(-ENOSPC);
2777
2778         if (xchg(&root->fs_info->enospc_unlink, 1))
2779                 return ERR_PTR(-ENOSPC);
2780
2781         path = btrfs_alloc_path();
2782         if (!path) {
2783                 root->fs_info->enospc_unlink = 0;
2784                 return ERR_PTR(-ENOMEM);
2785         }
2786
2787         trans = btrfs_start_transaction(root, 0);
2788         if (IS_ERR(trans)) {
2789                 btrfs_free_path(path);
2790                 root->fs_info->enospc_unlink = 0;
2791                 return trans;
2792         }
2793
2794         path->skip_locking = 1;
2795         path->search_commit_root = 1;
2796
2797         ret = btrfs_lookup_inode(trans, root, path,
2798                                 &BTRFS_I(dir)->location, 0);
2799         if (ret < 0) {
2800                 err = ret;
2801                 goto out;
2802         }
2803         if (ret == 0) {
2804                 if (check_path_shared(root, path))
2805                         goto out;
2806         } else {
2807                 check_link = 0;
2808         }
2809         btrfs_release_path(root, path);
2810
2811         ret = btrfs_lookup_inode(trans, root, path,
2812                                 &BTRFS_I(inode)->location, 0);
2813         if (ret < 0) {
2814                 err = ret;
2815                 goto out;
2816         }
2817         if (ret == 0) {
2818                 if (check_path_shared(root, path))
2819                         goto out;
2820         } else {
2821                 check_link = 0;
2822         }
2823         btrfs_release_path(root, path);
2824
2825         if (ret == 0 && S_ISREG(inode->i_mode)) {
2826                 ret = btrfs_lookup_file_extent(trans, root, path,
2827                                                inode->i_ino, (u64)-1, 0);
2828                 if (ret < 0) {
2829                         err = ret;
2830                         goto out;
2831                 }
2832                 BUG_ON(ret == 0);
2833                 if (check_path_shared(root, path))
2834                         goto out;
2835                 btrfs_release_path(root, path);
2836         }
2837
2838         if (!check_link) {
2839                 err = 0;
2840                 goto out;
2841         }
2842
2843         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2844                                 dentry->d_name.name, dentry->d_name.len, 0);
2845         if (IS_ERR(di)) {
2846                 err = PTR_ERR(di);
2847                 goto out;
2848         }
2849         if (di) {
2850                 if (check_path_shared(root, path))
2851                         goto out;
2852         } else {
2853                 err = 0;
2854                 goto out;
2855         }
2856         btrfs_release_path(root, path);
2857
2858         ref = btrfs_lookup_inode_ref(trans, root, path,
2859                                 dentry->d_name.name, dentry->d_name.len,
2860                                 inode->i_ino, dir->i_ino, 0);
2861         if (IS_ERR(ref)) {
2862                 err = PTR_ERR(ref);
2863                 goto out;
2864         }
2865         BUG_ON(!ref);
2866         if (check_path_shared(root, path))
2867                 goto out;
2868         index = btrfs_inode_ref_index(path->nodes[0], ref);
2869         btrfs_release_path(root, path);
2870
2871         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2872                                 dentry->d_name.name, dentry->d_name.len, 0);
2873         if (IS_ERR(di)) {
2874                 err = PTR_ERR(di);
2875                 goto out;
2876         }
2877         BUG_ON(ret == -ENOENT);
2878         if (check_path_shared(root, path))
2879                 goto out;
2880
2881         err = 0;
2882 out:
2883         btrfs_free_path(path);
2884         if (err) {
2885                 btrfs_end_transaction(trans, root);
2886                 root->fs_info->enospc_unlink = 0;
2887                 return ERR_PTR(err);
2888         }
2889
2890         trans->block_rsv = &root->fs_info->global_block_rsv;
2891         return trans;
2892 }
2893
2894 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2895                                struct btrfs_root *root)
2896 {
2897         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2898                 BUG_ON(!root->fs_info->enospc_unlink);
2899                 root->fs_info->enospc_unlink = 0;
2900         }
2901         btrfs_end_transaction_throttle(trans, root);
2902 }
2903
2904 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2905 {
2906         struct btrfs_root *root = BTRFS_I(dir)->root;
2907         struct btrfs_trans_handle *trans;
2908         struct inode *inode = dentry->d_inode;
2909         int ret;
2910         unsigned long nr = 0;
2911
2912         trans = __unlink_start_trans(dir, dentry);
2913         if (IS_ERR(trans))
2914                 return PTR_ERR(trans);
2915
2916         btrfs_set_trans_block_group(trans, dir);
2917
2918         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2919
2920         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2921                                  dentry->d_name.name, dentry->d_name.len);
2922         BUG_ON(ret);
2923
2924         if (inode->i_nlink == 0) {
2925                 ret = btrfs_orphan_add(trans, inode);
2926                 BUG_ON(ret);
2927         }
2928
2929         nr = trans->blocks_used;
2930         __unlink_end_trans(trans, root);
2931         btrfs_btree_balance_dirty(root, nr);
2932         return ret;
2933 }
2934
2935 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2936                         struct btrfs_root *root,
2937                         struct inode *dir, u64 objectid,
2938                         const char *name, int name_len)
2939 {
2940         struct btrfs_path *path;
2941         struct extent_buffer *leaf;
2942         struct btrfs_dir_item *di;
2943         struct btrfs_key key;
2944         u64 index;
2945         int ret;
2946
2947         path = btrfs_alloc_path();
2948         if (!path)
2949                 return -ENOMEM;
2950
2951         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2952                                    name, name_len, -1);
2953         BUG_ON(!di || IS_ERR(di));
2954
2955         leaf = path->nodes[0];
2956         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2957         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2958         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2959         BUG_ON(ret);
2960         btrfs_release_path(root, path);
2961
2962         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2963                                  objectid, root->root_key.objectid,
2964                                  dir->i_ino, &index, name, name_len);
2965         if (ret < 0) {
2966                 BUG_ON(ret != -ENOENT);
2967                 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2968                                                  name, name_len);
2969                 BUG_ON(!di || IS_ERR(di));
2970
2971                 leaf = path->nodes[0];
2972                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2973                 btrfs_release_path(root, path);
2974                 index = key.offset;
2975         }
2976
2977         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2978                                          index, name, name_len, -1);
2979         BUG_ON(!di || IS_ERR(di));
2980
2981         leaf = path->nodes[0];
2982         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2983         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2984         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2985         BUG_ON(ret);
2986         btrfs_release_path(root, path);
2987
2988         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2989         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2990         ret = btrfs_update_inode(trans, root, dir);
2991         BUG_ON(ret);
2992
2993         btrfs_free_path(path);
2994         return 0;
2995 }
2996
2997 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2998 {
2999         struct inode *inode = dentry->d_inode;
3000         int err = 0;
3001         struct btrfs_root *root = BTRFS_I(dir)->root;
3002         struct btrfs_trans_handle *trans;
3003         unsigned long nr = 0;
3004
3005         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3006             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3007                 return -ENOTEMPTY;
3008
3009         trans = __unlink_start_trans(dir, dentry);
3010         if (IS_ERR(trans))
3011                 return PTR_ERR(trans);
3012
3013         btrfs_set_trans_block_group(trans, dir);
3014
3015         if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3016                 err = btrfs_unlink_subvol(trans, root, dir,
3017                                           BTRFS_I(inode)->location.objectid,
3018                                           dentry->d_name.name,
3019                                           dentry->d_name.len);
3020                 goto out;
3021         }
3022
3023         err = btrfs_orphan_add(trans, inode);
3024         if (err)
3025                 goto out;
3026
3027         /* now the directory is empty */
3028         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3029                                  dentry->d_name.name, dentry->d_name.len);
3030         if (!err)
3031                 btrfs_i_size_write(inode, 0);
3032 out:
3033         nr = trans->blocks_used;
3034         __unlink_end_trans(trans, root);
3035         btrfs_btree_balance_dirty(root, nr);
3036
3037         return err;
3038 }
3039
3040 #if 0
3041 /*
3042  * when truncating bytes in a file, it is possible to avoid reading
3043  * the leaves that contain only checksum items.  This can be the
3044  * majority of the IO required to delete a large file, but it must
3045  * be done carefully.
3046  *
3047  * The keys in the level just above the leaves are checked to make sure
3048  * the lowest key in a given leaf is a csum key, and starts at an offset
3049  * after the new  size.
3050  *
3051  * Then the key for the next leaf is checked to make sure it also has
3052  * a checksum item for the same file.  If it does, we know our target leaf
3053  * contains only checksum items, and it can be safely freed without reading
3054  * it.
3055  *
3056  * This is just an optimization targeted at large files.  It may do
3057  * nothing.  It will return 0 unless things went badly.
3058  */
3059 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3060                                      struct btrfs_root *root,
3061                                      struct btrfs_path *path,
3062                                      struct inode *inode, u64 new_size)
3063 {
3064         struct btrfs_key key;
3065         int ret;
3066         int nritems;
3067         struct btrfs_key found_key;
3068         struct btrfs_key other_key;
3069         struct btrfs_leaf_ref *ref;
3070         u64 leaf_gen;
3071         u64 leaf_start;
3072
3073         path->lowest_level = 1;
3074         key.objectid = inode->i_ino;
3075         key.type = BTRFS_CSUM_ITEM_KEY;
3076         key.offset = new_size;
3077 again:
3078         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3079         if (ret < 0)
3080                 goto out;
3081
3082         if (path->nodes[1] == NULL) {
3083                 ret = 0;
3084                 goto out;
3085         }
3086         ret = 0;
3087         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3088         nritems = btrfs_header_nritems(path->nodes[1]);
3089
3090         if (!nritems)
3091                 goto out;
3092
3093         if (path->slots[1] >= nritems)
3094                 goto next_node;
3095
3096         /* did we find a key greater than anything we want to delete? */
3097         if (found_key.objectid > inode->i_ino ||
3098            (found_key.objectid == inode->i_ino && found_key.type > key.type))
3099                 goto out;
3100
3101         /* we check the next key in the node to make sure the leave contains
3102          * only checksum items.  This comparison doesn't work if our
3103          * leaf is the last one in the node
3104          */
3105         if (path->slots[1] + 1 >= nritems) {
3106 next_node:
3107                 /* search forward from the last key in the node, this
3108                  * will bring us into the next node in the tree
3109                  */
3110                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3111
3112                 /* unlikely, but we inc below, so check to be safe */
3113                 if (found_key.offset == (u64)-1)
3114                         goto out;
3115
3116                 /* search_forward needs a path with locks held, do the
3117                  * search again for the original key.  It is possible
3118                  * this will race with a balance and return a path that
3119                  * we could modify, but this drop is just an optimization
3120                  * and is allowed to miss some leaves.
3121                  */
3122                 btrfs_release_path(root, path);
3123                 found_key.offset++;
3124
3125                 /* setup a max key for search_forward */
3126                 other_key.offset = (u64)-1;
3127                 other_key.type = key.type;
3128                 other_key.objectid = key.objectid;
3129
3130                 path->keep_locks = 1;
3131                 ret = btrfs_search_forward(root, &found_key, &other_key,
3132                                            path, 0, 0);
3133                 path->keep_locks = 0;
3134                 if (ret || found_key.objectid != key.objectid ||
3135                     found_key.type != key.type) {
3136                         ret = 0;
3137                         goto out;
3138                 }
3139
3140                 key.offset = found_key.offset;
3141                 btrfs_release_path(root, path);
3142                 cond_resched();
3143                 goto again;
3144         }
3145
3146         /* we know there's one more slot after us in the tree,
3147          * read that key so we can verify it is also a checksum item
3148          */
3149         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3150
3151         if (found_key.objectid < inode->i_ino)
3152                 goto next_key;
3153
3154         if (found_key.type != key.type || found_key.offset < new_size)
3155                 goto next_key;
3156
3157         /*
3158          * if the key for the next leaf isn't a csum key from this objectid,
3159          * we can't be sure there aren't good items inside this leaf.
3160          * Bail out
3161          */
3162         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3163                 goto out;
3164
3165         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3166         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3167         /*
3168          * it is safe to delete this leaf, it contains only
3169          * csum items from this inode at an offset >= new_size
3170          */
3171         ret = btrfs_del_leaf(trans, root, path, leaf_start);
3172         BUG_ON(ret);
3173
3174         if (root->ref_cows && leaf_gen < trans->transid) {
3175                 ref = btrfs_alloc_leaf_ref(root, 0);
3176                 if (ref) {
3177                         ref->root_gen = root->root_key.offset;
3178                         ref->bytenr = leaf_start;
3179                         ref->owner = 0;
3180                         ref->generation = leaf_gen;
3181                         ref->nritems = 0;
3182
3183                         btrfs_sort_leaf_ref(ref);
3184
3185                         ret = btrfs_add_leaf_ref(root, ref, 0);
3186                         WARN_ON(ret);
3187                         btrfs_free_leaf_ref(root, ref);
3188                 } else {
3189                         WARN_ON(1);
3190                 }
3191         }
3192 next_key:
3193         btrfs_release_path(root, path);
3194
3195         if (other_key.objectid == inode->i_ino &&
3196             other_key.type == key.type && other_key.offset > key.offset) {
3197                 key.offset = other_key.offset;
3198                 cond_resched();
3199                 goto again;
3200         }
3201         ret = 0;
3202 out:
3203         /* fixup any changes we've made to the path */
3204         path->lowest_level = 0;
3205         path->keep_locks = 0;
3206         btrfs_release_path(root, path);
3207         return ret;
3208 }
3209
3210 #endif
3211
3212 /*
3213  * this can truncate away extent items, csum items and directory items.
3214  * It starts at a high offset and removes keys until it can't find
3215  * any higher than new_size
3216  *
3217  * csum items that cross the new i_size are truncated to the new size
3218  * as well.
3219  *
3220  * min_type is the minimum key type to truncate down to.  If set to 0, this
3221  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3222  */
3223 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3224                                struct btrfs_root *root,
3225                                struct inode *inode,
3226                                u64 new_size, u32 min_type)
3227 {
3228         struct btrfs_path *path;
3229         struct extent_buffer *leaf;
3230         struct btrfs_file_extent_item *fi;
3231         struct btrfs_key key;
3232         struct btrfs_key found_key;
3233         u64 extent_start = 0;
3234         u64 extent_num_bytes = 0;
3235         u64 extent_offset = 0;
3236         u64 item_end = 0;
3237         u64 mask = root->sectorsize - 1;
3238         u32 found_type = (u8)-1;
3239         int found_extent;
3240         int del_item;
3241         int pending_del_nr = 0;
3242         int pending_del_slot = 0;
3243         int extent_type = -1;
3244         int encoding;
3245         int ret;
3246         int err = 0;
3247
3248         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3249
3250         if (root->ref_cows || root == root->fs_info->tree_root)
3251                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3252
3253         path = btrfs_alloc_path();
3254         BUG_ON(!path);
3255         path->reada = -1;
3256
3257         key.objectid = inode->i_ino;
3258         key.offset = (u64)-1;
3259         key.type = (u8)-1;
3260
3261 search_again:
3262         path->leave_spinning = 1;
3263         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3264         if (ret < 0) {
3265                 err = ret;
3266                 goto out;
3267         }
3268
3269         if (ret > 0) {
3270                 /* there are no items in the tree for us to truncate, we're
3271                  * done
3272                  */
3273                 if (path->slots[0] == 0)
3274                         goto out;
3275                 path->slots[0]--;
3276         }
3277
3278         while (1) {
3279                 fi = NULL;
3280                 leaf = path->nodes[0];
3281                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3282                 found_type = btrfs_key_type(&found_key);
3283                 encoding = 0;
3284
3285                 if (found_key.objectid != inode->i_ino)
3286                         break;
3287
3288                 if (found_type < min_type)
3289                         break;
3290
3291                 item_end = found_key.offset;
3292                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3293                         fi = btrfs_item_ptr(leaf, path->slots[0],
3294                                             struct btrfs_file_extent_item);
3295                         extent_type = btrfs_file_extent_type(leaf, fi);
3296                         encoding = btrfs_file_extent_compression(leaf, fi);
3297                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3298                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3299
3300                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3301                                 item_end +=
3302                                     btrfs_file_extent_num_bytes(leaf, fi);
3303                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3304                                 item_end += btrfs_file_extent_inline_len(leaf,
3305                                                                          fi);
3306                         }
3307                         item_end--;
3308                 }
3309                 if (found_type > min_type) {
3310                         del_item = 1;
3311                 } else {
3312                         if (item_end < new_size)
3313                                 break;
3314                         if (found_key.offset >= new_size)
3315                                 del_item = 1;
3316                         else
3317                                 del_item = 0;
3318                 }
3319                 found_extent = 0;
3320                 /* FIXME, shrink the extent if the ref count is only 1 */
3321                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3322                         goto delete;
3323
3324                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3325                         u64 num_dec;
3326                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3327                         if (!del_item && !encoding) {
3328                                 u64 orig_num_bytes =
3329                                         btrfs_file_extent_num_bytes(leaf, fi);
3330                                 extent_num_bytes = new_size -
3331                                         found_key.offset + root->sectorsize - 1;
3332                                 extent_num_bytes = extent_num_bytes &
3333                                         ~((u64)root->sectorsize - 1);
3334                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3335                                                          extent_num_bytes);
3336                                 num_dec = (orig_num_bytes -
3337                                            extent_num_bytes);
3338                                 if (root->ref_cows && extent_start != 0)
3339                                         inode_sub_bytes(inode, num_dec);
3340                                 btrfs_mark_buffer_dirty(leaf);
3341                         } else {
3342                                 extent_num_bytes =
3343                                         btrfs_file_extent_disk_num_bytes(leaf,
3344                                                                          fi);
3345                                 extent_offset = found_key.offset -
3346                                         btrfs_file_extent_offset(leaf, fi);
3347
3348                                 /* FIXME blocksize != 4096 */
3349                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3350                                 if (extent_start != 0) {
3351                                         found_extent = 1;
3352                                         if (root->ref_cows)
3353                                                 inode_sub_bytes(inode, num_dec);
3354                                 }
3355                         }
3356                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3357                         /*
3358                          * we can't truncate inline items that have had
3359                          * special encodings
3360                          */
3361                         if (!del_item &&
3362                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3363                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3364                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3365                                 u32 size = new_size - found_key.offset;
3366
3367                                 if (root->ref_cows) {
3368                                         inode_sub_bytes(inode, item_end + 1 -
3369                                                         new_size);
3370                                 }
3371                                 size =
3372                                     btrfs_file_extent_calc_inline_size(size);
3373                                 ret = btrfs_truncate_item(trans, root, path,
3374                                                           size, 1);
3375                                 BUG_ON(ret);
3376                         } else if (root->ref_cows) {
3377                                 inode_sub_bytes(inode, item_end + 1 -
3378                                                 found_key.offset);
3379                         }
3380                 }
3381 delete:
3382                 if (del_item) {
3383                         if (!pending_del_nr) {
3384                                 /* no pending yet, add ourselves */
3385                                 pending_del_slot = path->slots[0];
3386                                 pending_del_nr = 1;
3387                         } else if (pending_del_nr &&
3388                                    path->slots[0] + 1 == pending_del_slot) {
3389                                 /* hop on the pending chunk */
3390                                 pending_del_nr++;
3391                                 pending_del_slot = path->slots[0];
3392                         } else {
3393                                 BUG();
3394                         }
3395                 } else {
3396                         break;
3397                 }
3398                 if (found_extent && (root->ref_cows ||
3399                                      root == root->fs_info->tree_root)) {
3400                         btrfs_set_path_blocking(path);
3401                         ret = btrfs_free_extent(trans, root, extent_start,
3402                                                 extent_num_bytes, 0,
3403                                                 btrfs_header_owner(leaf),
3404                                                 inode->i_ino, extent_offset);
3405                         BUG_ON(ret);
3406                 }
3407
3408                 if (found_type == BTRFS_INODE_ITEM_KEY)
3409                         break;
3410
3411                 if (path->slots[0] == 0 ||
3412                     path->slots[0] != pending_del_slot) {
3413                         if (root->ref_cows) {
3414                                 err = -EAGAIN;
3415                                 goto out;
3416                         }
3417                         if (pending_del_nr) {
3418                                 ret = btrfs_del_items(trans, root, path,
3419                                                 pending_del_slot,
3420                                                 pending_del_nr);
3421                                 BUG_ON(ret);
3422                                 pending_del_nr = 0;
3423                         }
3424                         btrfs_release_path(root, path);
3425                         goto search_again;
3426                 } else {
3427                         path->slots[0]--;
3428                 }
3429         }
3430 out:
3431         if (pending_del_nr) {
3432                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3433                                       pending_del_nr);
3434                 BUG_ON(ret);
3435         }
3436         btrfs_free_path(path);
3437         return err;
3438 }
3439
3440 /*
3441  * taken from block_truncate_page, but does cow as it zeros out
3442  * any bytes left in the last page in the file.
3443  */
3444 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3445 {
3446         struct inode *inode = mapping->host;
3447         struct btrfs_root *root = BTRFS_I(inode)->root;
3448         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3449         struct btrfs_ordered_extent *ordered;
3450         struct extent_state *cached_state = NULL;
3451         char *kaddr;
3452         u32 blocksize = root->sectorsize;
3453         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3454         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3455         struct page *page;
3456         int ret = 0;
3457         u64 page_start;
3458         u64 page_end;
3459
3460         if ((offset & (blocksize - 1)) == 0)
3461                 goto out;
3462         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3463         if (ret)
3464                 goto out;
3465
3466         ret = -ENOMEM;
3467 again:
3468         page = grab_cache_page(mapping, index);
3469         if (!page) {
3470                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3471                 goto out;
3472         }
3473
3474         page_start = page_offset(page);
3475         page_end = page_start + PAGE_CACHE_SIZE - 1;
3476
3477         if (!PageUptodate(page)) {
3478                 ret = btrfs_readpage(NULL, page);
3479                 lock_page(page);
3480                 if (page->mapping != mapping) {
3481                         unlock_page(page);
3482                         page_cache_release(page);
3483                         goto again;
3484                 }
3485                 if (!PageUptodate(page)) {
3486                         ret = -EIO;
3487                         goto out_unlock;
3488                 }
3489         }
3490         wait_on_page_writeback(page);
3491
3492         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3493                          GFP_NOFS);
3494         set_page_extent_mapped(page);
3495
3496         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3497         if (ordered) {
3498                 unlock_extent_cached(io_tree, page_start, page_end,
3499                                      &cached_state, GFP_NOFS);
3500                 unlock_page(page);
3501                 page_cache_release(page);
3502                 btrfs_start_ordered_extent(inode, ordered, 1);
3503                 btrfs_put_ordered_extent(ordered);
3504                 goto again;
3505         }
3506
3507         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3508                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3509                           0, 0, &cached_state, GFP_NOFS);
3510
3511         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3512                                         &cached_state);
3513         if (ret) {
3514                 unlock_extent_cached(io_tree, page_start, page_end,
3515                                      &cached_state, GFP_NOFS);
3516                 goto out_unlock;
3517         }
3518
3519         ret = 0;
3520         if (offset != PAGE_CACHE_SIZE) {
3521                 kaddr = kmap(page);
3522                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3523                 flush_dcache_page(page);
3524                 kunmap(page);
3525         }
3526         ClearPageChecked(page);
3527         set_page_dirty(page);
3528         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3529                              GFP_NOFS);
3530
3531 out_unlock:
3532         if (ret)
3533                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3534         unlock_page(page);
3535         page_cache_release(page);
3536 out:
3537         return ret;
3538 }
3539
3540 int btrfs_cont_expand(struct inode *inode, loff_t size)
3541 {
3542         struct btrfs_trans_handle *trans;
3543         struct btrfs_root *root = BTRFS_I(inode)->root;
3544         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3545         struct extent_map *em = NULL;
3546         struct extent_state *cached_state = NULL;
3547         u64 mask = root->sectorsize - 1;
3548         u64 hole_start = (inode->i_size + mask) & ~mask;
3549         u64 block_end = (size + mask) & ~mask;
3550         u64 last_byte;
3551         u64 cur_offset;
3552         u64 hole_size;
3553         int err = 0;
3554
3555         if (size <= hole_start)
3556                 return 0;
3557
3558         while (1) {
3559                 struct btrfs_ordered_extent *ordered;
3560                 btrfs_wait_ordered_range(inode, hole_start,
3561                                          block_end - hole_start);
3562                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3563                                  &cached_state, GFP_NOFS);
3564                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3565                 if (!ordered)
3566                         break;
3567                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3568                                      &cached_state, GFP_NOFS);
3569                 btrfs_put_ordered_extent(ordered);
3570         }
3571
3572         cur_offset = hole_start;
3573         while (1) {
3574                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3575                                 block_end - cur_offset, 0);
3576                 BUG_ON(IS_ERR(em) || !em);
3577                 last_byte = min(extent_map_end(em), block_end);
3578                 last_byte = (last_byte + mask) & ~mask;
3579                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3580                         u64 hint_byte = 0;
3581                         hole_size = last_byte - cur_offset;
3582
3583                         trans = btrfs_start_transaction(root, 2);
3584                         if (IS_ERR(trans)) {
3585                                 err = PTR_ERR(trans);
3586                                 break;
3587                         }
3588                         btrfs_set_trans_block_group(trans, inode);
3589
3590                         err = btrfs_drop_extents(trans, inode, cur_offset,
3591                                                  cur_offset + hole_size,
3592                                                  &hint_byte, 1);
3593                         BUG_ON(err);
3594
3595                         err = btrfs_insert_file_extent(trans, root,
3596                                         inode->i_ino, cur_offset, 0,
3597                                         0, hole_size, 0, hole_size,
3598                                         0, 0, 0);
3599                         BUG_ON(err);
3600
3601                         btrfs_drop_extent_cache(inode, hole_start,
3602                                         last_byte - 1, 0);
3603
3604                         btrfs_end_transaction(trans, root);
3605                 }
3606                 free_extent_map(em);
3607                 em = NULL;
3608                 cur_offset = last_byte;
3609                 if (cur_offset >= block_end)
3610                         break;
3611         }
3612
3613         free_extent_map(em);
3614         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3615                              GFP_NOFS);
3616         return err;
3617 }
3618
3619 static int btrfs_setattr_size(struct inode *inode, struct iattr *attr)
3620 {
3621         struct btrfs_root *root = BTRFS_I(inode)->root;
3622         struct btrfs_trans_handle *trans;
3623         unsigned long nr;
3624         int ret;
3625
3626         if (attr->ia_size == inode->i_size)
3627                 return 0;
3628
3629         if (attr->ia_size > inode->i_size) {
3630                 unsigned long limit;
3631                 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
3632                 if (attr->ia_size > inode->i_sb->s_maxbytes)
3633                         return -EFBIG;
3634                 if (limit != RLIM_INFINITY && attr->ia_size > limit) {
3635                         send_sig(SIGXFSZ, current, 0);
3636                         return -EFBIG;
3637                 }
3638         }
3639
3640         trans = btrfs_start_transaction(root, 5);
3641         if (IS_ERR(trans))
3642                 return PTR_ERR(trans);
3643
3644         btrfs_set_trans_block_group(trans, inode);
3645
3646         ret = btrfs_orphan_add(trans, inode);
3647         BUG_ON(ret);
3648
3649         nr = trans->blocks_used;
3650         btrfs_end_transaction(trans, root);
3651         btrfs_btree_balance_dirty(root, nr);
3652
3653         if (attr->ia_size > inode->i_size) {
3654                 ret = btrfs_cont_expand(inode, attr->ia_size);
3655                 if (ret) {
3656                         btrfs_truncate(inode);
3657                         return ret;
3658                 }
3659
3660                 i_size_write(inode, attr->ia_size);
3661                 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
3662
3663                 trans = btrfs_start_transaction(root, 0);
3664                 BUG_ON(IS_ERR(trans));
3665                 btrfs_set_trans_block_group(trans, inode);
3666                 trans->block_rsv = root->orphan_block_rsv;
3667                 BUG_ON(!trans->block_rsv);
3668
3669                 ret = btrfs_update_inode(trans, root, inode);
3670                 BUG_ON(ret);
3671                 if (inode->i_nlink > 0) {
3672                         ret = btrfs_orphan_del(trans, inode);
3673                         BUG_ON(ret);
3674                 }
3675                 nr = trans->blocks_used;
3676                 btrfs_end_transaction(trans, root);
3677                 btrfs_btree_balance_dirty(root, nr);
3678                 return 0;
3679         }
3680
3681         /*
3682          * We're truncating a file that used to have good data down to
3683          * zero. Make sure it gets into the ordered flush list so that
3684          * any new writes get down to disk quickly.
3685          */
3686         if (attr->ia_size == 0)
3687                 BTRFS_I(inode)->ordered_data_close = 1;
3688
3689         /* we don't support swapfiles, so vmtruncate shouldn't fail */
3690         ret = vmtruncate(inode, attr->ia_size);
3691         BUG_ON(ret);
3692
3693         return 0;
3694 }
3695
3696 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3697 {
3698         struct inode *inode = dentry->d_inode;
3699         struct btrfs_root *root = BTRFS_I(inode)->root;
3700         int err;
3701
3702         if (btrfs_root_readonly(root))
3703                 return -EROFS;
3704
3705         err = inode_change_ok(inode, attr);
3706         if (err)
3707                 return err;
3708
3709         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3710                 err = btrfs_setattr_size(inode, attr);
3711                 if (err)
3712                         return err;
3713         }
3714
3715         if (attr->ia_valid) {
3716                 setattr_copy(inode, attr);
3717                 mark_inode_dirty(inode);
3718
3719                 if (attr->ia_valid & ATTR_MODE)
3720                         err = btrfs_acl_chmod(inode);
3721         }
3722
3723         return err;
3724 }
3725
3726 void btrfs_evict_inode(struct inode *inode)
3727 {
3728         struct btrfs_trans_handle *trans;
3729         struct btrfs_root *root = BTRFS_I(inode)->root;
3730         unsigned long nr;
3731         int ret;
3732
3733         truncate_inode_pages(&inode->i_data, 0);
3734         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3735                                root == root->fs_info->tree_root))
3736                 goto no_delete;
3737
3738         if (is_bad_inode(inode)) {
3739                 btrfs_orphan_del(NULL, inode);
3740                 goto no_delete;
3741         }
3742         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3743         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3744
3745         if (root->fs_info->log_root_recovering) {
3746                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3747                 goto no_delete;
3748         }
3749
3750         if (inode->i_nlink > 0) {
3751                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3752                 goto no_delete;
3753         }
3754
3755         btrfs_i_size_write(inode, 0);
3756
3757         while (1) {
3758                 trans = btrfs_start_transaction(root, 0);
3759                 BUG_ON(IS_ERR(trans));
3760                 btrfs_set_trans_block_group(trans, inode);
3761                 trans->block_rsv = root->orphan_block_rsv;
3762
3763                 ret = btrfs_block_rsv_check(trans, root,
3764                                             root->orphan_block_rsv, 0, 5);
3765                 if (ret) {
3766                         BUG_ON(ret != -EAGAIN);
3767                         ret = btrfs_commit_transaction(trans, root);
3768                         BUG_ON(ret);
3769                         continue;
3770                 }
3771
3772                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3773                 if (ret != -EAGAIN)
3774                         break;
3775
3776                 nr = trans->blocks_used;
3777                 btrfs_end_transaction(trans, root);
3778                 trans = NULL;
3779                 btrfs_btree_balance_dirty(root, nr);
3780
3781         }
3782
3783         if (ret == 0) {
3784                 ret = btrfs_orphan_del(trans, inode);
3785                 BUG_ON(ret);
3786         }
3787
3788         nr = trans->blocks_used;
3789         btrfs_end_transaction(trans, root);
3790         btrfs_btree_balance_dirty(root, nr);
3791 no_delete:
3792         end_writeback(inode);
3793         return;
3794 }
3795
3796 /*
3797  * this returns the key found in the dir entry in the location pointer.
3798  * If no dir entries were found, location->objectid is 0.
3799  */
3800 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3801                                struct btrfs_key *location)
3802 {
3803         const char *name = dentry->d_name.name;
3804         int namelen = dentry->d_name.len;
3805         struct btrfs_dir_item *di;
3806         struct btrfs_path *path;
3807         struct btrfs_root *root = BTRFS_I(dir)->root;
3808         int ret = 0;
3809
3810         path = btrfs_alloc_path();
3811         BUG_ON(!path);
3812
3813         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3814                                     namelen, 0);
3815         if (IS_ERR(di))
3816                 ret = PTR_ERR(di);
3817
3818         if (!di || IS_ERR(di))
3819                 goto out_err;
3820
3821         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3822 out:
3823         btrfs_free_path(path);
3824         return ret;
3825 out_err:
3826         location->objectid = 0;
3827         goto out;
3828 }
3829
3830 /*
3831  * when we hit a tree root in a directory, the btrfs part of the inode
3832  * needs to be changed to reflect the root directory of the tree root.  This
3833  * is kind of like crossing a mount point.
3834  */
3835 static int fixup_tree_root_location(struct btrfs_root *root,
3836                                     struct inode *dir,
3837                                     struct dentry *dentry,
3838                                     struct btrfs_key *location,
3839                                     struct btrfs_root **sub_root)
3840 {
3841         struct btrfs_path *path;
3842         struct btrfs_root *new_root;
3843         struct btrfs_root_ref *ref;
3844         struct extent_buffer *leaf;
3845         int ret;
3846         int err = 0;
3847
3848         path = btrfs_alloc_path();
3849         if (!path) {
3850                 err = -ENOMEM;
3851                 goto out;
3852         }
3853
3854         err = -ENOENT;
3855         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3856                                   BTRFS_I(dir)->root->root_key.objectid,
3857                                   location->objectid);
3858         if (ret) {
3859                 if (ret < 0)
3860                         err = ret;
3861                 goto out;
3862         }
3863
3864         leaf = path->nodes[0];
3865         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3866         if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3867             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3868                 goto out;
3869
3870         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3871                                    (unsigned long)(ref + 1),
3872                                    dentry->d_name.len);
3873         if (ret)
3874                 goto out;
3875
3876         btrfs_release_path(root->fs_info->tree_root, path);
3877
3878         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3879         if (IS_ERR(new_root)) {
3880                 err = PTR_ERR(new_root);
3881                 goto out;
3882         }
3883
3884         if (btrfs_root_refs(&new_root->root_item) == 0) {
3885                 err = -ENOENT;
3886                 goto out;
3887         }
3888
3889         *sub_root = new_root;
3890         location->objectid = btrfs_root_dirid(&new_root->root_item);
3891         location->type = BTRFS_INODE_ITEM_KEY;
3892         location->offset = 0;
3893         err = 0;
3894 out:
3895         btrfs_free_path(path);
3896         return err;
3897 }
3898
3899 static void inode_tree_add(struct inode *inode)
3900 {
3901         struct btrfs_root *root = BTRFS_I(inode)->root;
3902         struct btrfs_inode *entry;
3903         struct rb_node **p;
3904         struct rb_node *parent;
3905 again:
3906         p = &root->inode_tree.rb_node;
3907         parent = NULL;
3908
3909         if (inode_unhashed(inode))
3910                 return;
3911
3912         spin_lock(&root->inode_lock);
3913         while (*p) {
3914                 parent = *p;
3915                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3916
3917                 if (inode->i_ino < entry->vfs_inode.i_ino)
3918                         p = &parent->rb_left;
3919                 else if (inode->i_ino > entry->vfs_inode.i_ino)
3920                         p = &parent->rb_right;
3921                 else {
3922                         WARN_ON(!(entry->vfs_inode.i_state &
3923                                   (I_WILL_FREE | I_FREEING)));
3924                         rb_erase(parent, &root->inode_tree);
3925                         RB_CLEAR_NODE(parent);
3926                         spin_unlock(&root->inode_lock);
3927                         goto again;
3928                 }
3929         }
3930         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3931         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3932         spin_unlock(&root->inode_lock);
3933 }
3934
3935 static void inode_tree_del(struct inode *inode)
3936 {
3937         struct btrfs_root *root = BTRFS_I(inode)->root;
3938         int empty = 0;
3939
3940         spin_lock(&root->inode_lock);
3941         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3942                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3943                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3944                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3945         }
3946         spin_unlock(&root->inode_lock);
3947
3948         /*
3949          * Free space cache has inodes in the tree root, but the tree root has a
3950          * root_refs of 0, so this could end up dropping the tree root as a
3951          * snapshot, so we need the extra !root->fs_info->tree_root check to
3952          * make sure we don't drop it.
3953          */
3954         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3955             root != root->fs_info->tree_root) {
3956                 synchronize_srcu(&root->fs_info->subvol_srcu);
3957                 spin_lock(&root->inode_lock);
3958                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3959                 spin_unlock(&root->inode_lock);
3960                 if (empty)
3961                         btrfs_add_dead_root(root);
3962         }
3963 }
3964
3965 int btrfs_invalidate_inodes(struct btrfs_root *root)
3966 {
3967         struct rb_node *node;
3968         struct rb_node *prev;
3969         struct btrfs_inode *entry;
3970         struct inode *inode;
3971         u64 objectid = 0;
3972
3973         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3974
3975         spin_lock(&root->inode_lock);
3976 again:
3977         node = root->inode_tree.rb_node;
3978         prev = NULL;
3979         while (node) {
3980                 prev = node;
3981                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3982
3983                 if (objectid < entry->vfs_inode.i_ino)
3984                         node = node->rb_left;
3985                 else if (objectid > entry->vfs_inode.i_ino)
3986                         node = node->rb_right;
3987                 else
3988                         break;
3989         }
3990         if (!node) {
3991                 while (prev) {
3992                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3993                         if (objectid <= entry->vfs_inode.i_ino) {
3994                                 node = prev;
3995                                 break;
3996                         }
3997                         prev = rb_next(prev);
3998                 }
3999         }
4000         while (node) {
4001                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4002                 objectid = entry->vfs_inode.i_ino + 1;
4003                 inode = igrab(&entry->vfs_inode);
4004                 if (inode) {
4005                         spin_unlock(&root->inode_lock);
4006                         if (atomic_read(&inode->i_count) > 1)
4007                                 d_prune_aliases(inode);
4008                         /*
4009                          * btrfs_drop_inode will have it removed from
4010                          * the inode cache when its usage count
4011                          * hits zero.
4012                          */
4013                         iput(inode);
4014                         cond_resched();
4015                         spin_lock(&root->inode_lock);
4016                         goto again;
4017                 }
4018
4019                 if (cond_resched_lock(&root->inode_lock))
4020                         goto again;
4021
4022                 node = rb_next(node);
4023         }
4024         spin_unlock(&root->inode_lock);
4025         return 0;
4026 }
4027
4028 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4029 {
4030         struct btrfs_iget_args *args = p;
4031         inode->i_ino = args->ino;
4032         BTRFS_I(inode)->root = args->root;
4033         btrfs_set_inode_space_info(args->root, inode);
4034         return 0;
4035 }
4036
4037 static int btrfs_find_actor(struct inode *inode, void *opaque)
4038 {
4039         struct btrfs_iget_args *args = opaque;
4040         return args->ino == inode->i_ino &&
4041                 args->root == BTRFS_I(inode)->root;
4042 }
4043
4044 static struct inode *btrfs_iget_locked(struct super_block *s,
4045                                        u64 objectid,
4046                                        struct btrfs_root *root)
4047 {
4048         struct inode *inode;
4049         struct btrfs_iget_args args;
4050         args.ino = objectid;
4051         args.root = root;
4052
4053         inode = iget5_locked(s, objectid, btrfs_find_actor,
4054                              btrfs_init_locked_inode,
4055                              (void *)&args);
4056         return inode;
4057 }
4058
4059 /* Get an inode object given its location and corresponding root.
4060  * Returns in *is_new if the inode was read from disk
4061  */
4062 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4063                          struct btrfs_root *root, int *new)
4064 {
4065         struct inode *inode;
4066
4067         inode = btrfs_iget_locked(s, location->objectid, root);
4068         if (!inode)
4069                 return ERR_PTR(-ENOMEM);
4070
4071         if (inode->i_state & I_NEW) {
4072                 BTRFS_I(inode)->root = root;
4073                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4074                 btrfs_read_locked_inode(inode);
4075
4076                 inode_tree_add(inode);
4077                 unlock_new_inode(inode);
4078                 if (new)
4079                         *new = 1;
4080         }
4081
4082         return inode;
4083 }
4084
4085 static struct inode *new_simple_dir(struct super_block *s,
4086                                     struct btrfs_key *key,
4087                                     struct btrfs_root *root)
4088 {
4089         struct inode *inode = new_inode(s);
4090
4091         if (!inode)
4092                 return ERR_PTR(-ENOMEM);
4093
4094         BTRFS_I(inode)->root = root;
4095         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4096         BTRFS_I(inode)->dummy_inode = 1;
4097
4098         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4099         inode->i_op = &simple_dir_inode_operations;
4100         inode->i_fop = &simple_dir_operations;
4101         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4102         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4103
4104         return inode;
4105 }
4106
4107 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4108 {
4109         struct inode *inode;
4110         struct btrfs_root *root = BTRFS_I(dir)->root;
4111         struct btrfs_root *sub_root = root;
4112         struct btrfs_key location;
4113         int index;
4114         int ret;
4115
4116         if (dentry->d_name.len > BTRFS_NAME_LEN)
4117                 return ERR_PTR(-ENAMETOOLONG);
4118
4119         ret = btrfs_inode_by_name(dir, dentry, &location);
4120
4121         if (ret < 0)
4122                 return ERR_PTR(ret);
4123
4124         if (location.objectid == 0)
4125                 return NULL;
4126
4127         if (location.type == BTRFS_INODE_ITEM_KEY) {
4128                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4129                 return inode;
4130         }
4131
4132         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4133
4134         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4135         ret = fixup_tree_root_location(root, dir, dentry,
4136                                        &location, &sub_root);
4137         if (ret < 0) {
4138                 if (ret != -ENOENT)
4139                         inode = ERR_PTR(ret);
4140                 else
4141                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4142         } else {
4143                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4144         }
4145         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4146
4147         if (!IS_ERR(inode) && root != sub_root) {
4148                 down_read(&root->fs_info->cleanup_work_sem);
4149                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4150                         btrfs_orphan_cleanup(sub_root);
4151                 up_read(&root->fs_info->cleanup_work_sem);
4152         }
4153
4154         return inode;
4155 }
4156
4157 static int btrfs_dentry_delete(const struct dentry *dentry)
4158 {
4159         struct btrfs_root *root;
4160
4161         if (!dentry->d_inode && !IS_ROOT(dentry))
4162                 dentry = dentry->d_parent;
4163
4164         if (dentry->d_inode) {
4165                 root = BTRFS_I(dentry->d_inode)->root;
4166                 if (btrfs_root_refs(&root->root_item) == 0)
4167                         return 1;
4168         }
4169         return 0;
4170 }
4171
4172 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4173                                    struct nameidata *nd)
4174 {
4175         struct inode *inode;
4176
4177         inode = btrfs_lookup_dentry(dir, dentry);
4178         if (IS_ERR(inode))
4179                 return ERR_CAST(inode);
4180
4181         return d_splice_alias(inode, dentry);
4182 }
4183
4184 static unsigned char btrfs_filetype_table[] = {
4185         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4186 };
4187
4188 static int btrfs_real_readdir(struct file *filp, void *dirent,
4189                               filldir_t filldir)
4190 {
4191         struct inode *inode = filp->f_dentry->d_inode;
4192         struct btrfs_root *root = BTRFS_I(inode)->root;
4193         struct btrfs_item *item;
4194         struct btrfs_dir_item *di;
4195         struct btrfs_key key;
4196         struct btrfs_key found_key;
4197         struct btrfs_path *path;
4198         int ret;
4199         u32 nritems;
4200         struct extent_buffer *leaf;
4201         int slot;
4202         int advance;
4203         unsigned char d_type;
4204         int over = 0;
4205         u32 di_cur;
4206         u32 di_total;
4207         u32 di_len;
4208         int key_type = BTRFS_DIR_INDEX_KEY;
4209         char tmp_name[32];
4210         char *name_ptr;
4211         int name_len;
4212
4213         /* FIXME, use a real flag for deciding about the key type */
4214         if (root->fs_info->tree_root == root)
4215                 key_type = BTRFS_DIR_ITEM_KEY;
4216
4217         /* special case for "." */
4218         if (filp->f_pos == 0) {
4219                 over = filldir(dirent, ".", 1,
4220                                1, inode->i_ino,
4221                                DT_DIR);
4222                 if (over)
4223                         return 0;
4224                 filp->f_pos = 1;
4225         }
4226         /* special case for .., just use the back ref */
4227         if (filp->f_pos == 1) {
4228                 u64 pino = parent_ino(filp->f_path.dentry);
4229                 over = filldir(dirent, "..", 2,
4230                                2, pino, DT_DIR);
4231                 if (over)
4232                         return 0;
4233                 filp->f_pos = 2;
4234         }
4235         path = btrfs_alloc_path();
4236         path->reada = 2;
4237
4238         btrfs_set_key_type(&key, key_type);
4239         key.offset = filp->f_pos;
4240         key.objectid = inode->i_ino;
4241
4242         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4243         if (ret < 0)
4244                 goto err;
4245         advance = 0;
4246
4247         while (1) {
4248                 leaf = path->nodes[0];
4249                 nritems = btrfs_header_nritems(leaf);
4250                 slot = path->slots[0];
4251                 if (advance || slot >= nritems) {
4252                         if (slot >= nritems - 1) {
4253                                 ret = btrfs_next_leaf(root, path);
4254                                 if (ret)
4255                                         break;
4256                                 leaf = path->nodes[0];
4257                                 nritems = btrfs_header_nritems(leaf);
4258                                 slot = path->slots[0];
4259                         } else {
4260                                 slot++;
4261                                 path->slots[0]++;
4262                         }
4263                 }
4264
4265                 advance = 1;
4266                 item = btrfs_item_nr(leaf, slot);
4267                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4268
4269                 if (found_key.objectid != key.objectid)
4270                         break;
4271                 if (btrfs_key_type(&found_key) != key_type)
4272                         break;
4273                 if (found_key.offset < filp->f_pos)
4274                         continue;
4275
4276                 filp->f_pos = found_key.offset;
4277
4278                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4279                 di_cur = 0;
4280                 di_total = btrfs_item_size(leaf, item);
4281
4282                 while (di_cur < di_total) {
4283                         struct btrfs_key location;
4284
4285                         name_len = btrfs_dir_name_len(leaf, di);
4286                         if (name_len <= sizeof(tmp_name)) {
4287                                 name_ptr = tmp_name;
4288                         } else {
4289                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4290                                 if (!name_ptr) {
4291                                         ret = -ENOMEM;
4292                                         goto err;
4293                                 }
4294                         }
4295                         read_extent_buffer(leaf, name_ptr,
4296                                            (unsigned long)(di + 1), name_len);
4297
4298                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4299                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4300
4301                         /* is this a reference to our own snapshot? If so
4302                          * skip it
4303                          */
4304                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4305                             location.objectid == root->root_key.objectid) {
4306                                 over = 0;
4307                                 goto skip;
4308                         }
4309                         over = filldir(dirent, name_ptr, name_len,
4310                                        found_key.offset, location.objectid,
4311                                        d_type);
4312
4313 skip:
4314                         if (name_ptr != tmp_name)
4315                                 kfree(name_ptr);
4316
4317                         if (over)
4318                                 goto nopos;
4319                         di_len = btrfs_dir_name_len(leaf, di) +
4320                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4321                         di_cur += di_len;
4322                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4323                 }
4324         }
4325
4326         /* Reached end of directory/root. Bump pos past the last item. */
4327         if (key_type == BTRFS_DIR_INDEX_KEY)
4328                 /*
4329                  * 32-bit glibc will use getdents64, but then strtol -
4330                  * so the last number we can serve is this.
4331                  */
4332                 filp->f_pos = 0x7fffffff;
4333         else
4334                 filp->f_pos++;
4335 nopos:
4336         ret = 0;
4337 err:
4338         btrfs_free_path(path);
4339         return ret;
4340 }
4341
4342 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4343 {
4344         struct btrfs_root *root = BTRFS_I(inode)->root;
4345         struct btrfs_trans_handle *trans;
4346         int ret = 0;
4347         bool nolock = false;
4348
4349         if (BTRFS_I(inode)->dummy_inode)
4350                 return 0;
4351
4352         smp_mb();
4353         nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4354
4355         if (wbc->sync_mode == WB_SYNC_ALL) {
4356                 if (nolock)
4357                         trans = btrfs_join_transaction_nolock(root, 1);
4358                 else
4359                         trans = btrfs_join_transaction(root, 1);
4360                 if (IS_ERR(trans))
4361                         return PTR_ERR(trans);
4362                 btrfs_set_trans_block_group(trans, inode);
4363                 if (nolock)
4364                         ret = btrfs_end_transaction_nolock(trans, root);
4365                 else
4366                         ret = btrfs_commit_transaction(trans, root);
4367         }
4368         return ret;
4369 }
4370
4371 /*
4372  * This is somewhat expensive, updating the tree every time the
4373  * inode changes.  But, it is most likely to find the inode in cache.
4374  * FIXME, needs more benchmarking...there are no reasons other than performance
4375  * to keep or drop this code.
4376  */
4377 void btrfs_dirty_inode(struct inode *inode)
4378 {
4379         struct btrfs_root *root = BTRFS_I(inode)->root;
4380         struct btrfs_trans_handle *trans;
4381         int ret;
4382
4383         if (BTRFS_I(inode)->dummy_inode)
4384                 return;
4385
4386         trans = btrfs_join_transaction(root, 1);
4387         BUG_ON(IS_ERR(trans));
4388         btrfs_set_trans_block_group(trans, inode);
4389
4390         ret = btrfs_update_inode(trans, root, inode);
4391         if (ret && ret == -ENOSPC) {
4392                 /* whoops, lets try again with the full transaction */
4393                 btrfs_end_transaction(trans, root);
4394                 trans = btrfs_start_transaction(root, 1);
4395                 if (IS_ERR(trans)) {
4396                         if (printk_ratelimit()) {
4397                                 printk(KERN_ERR "btrfs: fail to "
4398                                        "dirty  inode %lu error %ld\n",
4399                                        inode->i_ino, PTR_ERR(trans));
4400                         }
4401                         return;
4402                 }
4403                 btrfs_set_trans_block_group(trans, inode);
4404
4405                 ret = btrfs_update_inode(trans, root, inode);
4406                 if (ret) {
4407                         if (printk_ratelimit()) {
4408                                 printk(KERN_ERR "btrfs: fail to "
4409                                        "dirty  inode %lu error %d\n",
4410                                        inode->i_ino, ret);
4411                         }
4412                 }
4413         }
4414         btrfs_end_transaction(trans, root);
4415 }
4416
4417 /*
4418  * find the highest existing sequence number in a directory
4419  * and then set the in-memory index_cnt variable to reflect
4420  * free sequence numbers
4421  */
4422 static int btrfs_set_inode_index_count(struct inode *inode)
4423 {
4424         struct btrfs_root *root = BTRFS_I(inode)->root;
4425         struct btrfs_key key, found_key;
4426         struct btrfs_path *path;
4427         struct extent_buffer *leaf;
4428         int ret;
4429
4430         key.objectid = inode->i_ino;
4431         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4432         key.offset = (u64)-1;
4433
4434         path = btrfs_alloc_path();
4435         if (!path)
4436                 return -ENOMEM;
4437
4438         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4439         if (ret < 0)
4440                 goto out;
4441         /* FIXME: we should be able to handle this */
4442         if (ret == 0)
4443                 goto out;
4444         ret = 0;
4445
4446         /*
4447          * MAGIC NUMBER EXPLANATION:
4448          * since we search a directory based on f_pos we have to start at 2
4449          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4450          * else has to start at 2
4451          */
4452         if (path->slots[0] == 0) {
4453                 BTRFS_I(inode)->index_cnt = 2;
4454                 goto out;
4455         }
4456
4457         path->slots[0]--;
4458
4459         leaf = path->nodes[0];
4460         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4461
4462         if (found_key.objectid != inode->i_ino ||
4463             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4464                 BTRFS_I(inode)->index_cnt = 2;
4465                 goto out;
4466         }
4467
4468         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4469 out:
4470         btrfs_free_path(path);
4471         return ret;
4472 }
4473
4474 /*
4475  * helper to find a free sequence number in a given directory.  This current
4476  * code is very simple, later versions will do smarter things in the btree
4477  */
4478 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4479 {
4480         int ret = 0;
4481
4482         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4483                 ret = btrfs_set_inode_index_count(dir);
4484                 if (ret)
4485                         return ret;
4486         }
4487
4488         *index = BTRFS_I(dir)->index_cnt;
4489         BTRFS_I(dir)->index_cnt++;
4490
4491         return ret;
4492 }
4493
4494 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4495                                      struct btrfs_root *root,
4496                                      struct inode *dir,
4497                                      const char *name, int name_len,
4498                                      u64 ref_objectid, u64 objectid,
4499                                      u64 alloc_hint, int mode, u64 *index)
4500 {
4501         struct inode *inode;
4502         struct btrfs_inode_item *inode_item;
4503         struct btrfs_key *location;
4504         struct btrfs_path *path;
4505         struct btrfs_inode_ref *ref;
4506         struct btrfs_key key[2];
4507         u32 sizes[2];
4508         unsigned long ptr;
4509         int ret;
4510         int owner;
4511
4512         path = btrfs_alloc_path();
4513         BUG_ON(!path);
4514
4515         inode = new_inode(root->fs_info->sb);
4516         if (!inode)
4517                 return ERR_PTR(-ENOMEM);
4518
4519         if (dir) {
4520                 ret = btrfs_set_inode_index(dir, index);
4521                 if (ret) {
4522                         iput(inode);
4523                         return ERR_PTR(ret);
4524                 }
4525         }
4526         /*
4527          * index_cnt is ignored for everything but a dir,
4528          * btrfs_get_inode_index_count has an explanation for the magic
4529          * number
4530          */
4531         BTRFS_I(inode)->index_cnt = 2;
4532         BTRFS_I(inode)->root = root;
4533         BTRFS_I(inode)->generation = trans->transid;
4534         inode->i_generation = BTRFS_I(inode)->generation;
4535         btrfs_set_inode_space_info(root, inode);
4536
4537         if (mode & S_IFDIR)
4538                 owner = 0;
4539         else
4540                 owner = 1;
4541         BTRFS_I(inode)->block_group =
4542                         btrfs_find_block_group(root, 0, alloc_hint, owner);
4543
4544         key[0].objectid = objectid;
4545         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4546         key[0].offset = 0;
4547
4548         key[1].objectid = objectid;
4549         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4550         key[1].offset = ref_objectid;
4551
4552         sizes[0] = sizeof(struct btrfs_inode_item);
4553         sizes[1] = name_len + sizeof(*ref);
4554
4555         path->leave_spinning = 1;
4556         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4557         if (ret != 0)
4558                 goto fail;
4559
4560         inode_init_owner(inode, dir, mode);
4561         inode->i_ino = objectid;
4562         inode_set_bytes(inode, 0);
4563         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4564         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4565                                   struct btrfs_inode_item);
4566         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4567
4568         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4569                              struct btrfs_inode_ref);
4570         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4571         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4572         ptr = (unsigned long)(ref + 1);
4573         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4574
4575         btrfs_mark_buffer_dirty(path->nodes[0]);
4576         btrfs_free_path(path);
4577
4578         location = &BTRFS_I(inode)->location;
4579         location->objectid = objectid;
4580         location->offset = 0;
4581         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4582
4583         btrfs_inherit_iflags(inode, dir);
4584
4585         if ((mode & S_IFREG)) {
4586                 if (btrfs_test_opt(root, NODATASUM))
4587                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4588                 if (btrfs_test_opt(root, NODATACOW))
4589                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4590         }
4591
4592         insert_inode_hash(inode);
4593         inode_tree_add(inode);
4594         return inode;
4595 fail:
4596         if (dir)
4597                 BTRFS_I(dir)->index_cnt--;
4598         btrfs_free_path(path);
4599         iput(inode);
4600         return ERR_PTR(ret);
4601 }
4602
4603 static inline u8 btrfs_inode_type(struct inode *inode)
4604 {
4605         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4606 }
4607
4608 /*
4609  * utility function to add 'inode' into 'parent_inode' with
4610  * a give name and a given sequence number.
4611  * if 'add_backref' is true, also insert a backref from the
4612  * inode to the parent directory.
4613  */
4614 int btrfs_add_link(struct btrfs_trans_handle *trans,
4615                    struct inode *parent_inode, struct inode *inode,
4616                    const char *name, int name_len, int add_backref, u64 index)
4617 {
4618         int ret = 0;
4619         struct btrfs_key key;
4620         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4621
4622         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4623                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4624         } else {
4625                 key.objectid = inode->i_ino;
4626                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4627                 key.offset = 0;
4628         }
4629
4630         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4631                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4632                                          key.objectid, root->root_key.objectid,
4633                                          parent_inode->i_ino,
4634                                          index, name, name_len);
4635         } else if (add_backref) {
4636                 ret = btrfs_insert_inode_ref(trans, root,
4637                                              name, name_len, inode->i_ino,
4638                                              parent_inode->i_ino, index);
4639         }
4640
4641         if (ret == 0) {
4642                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4643                                             parent_inode->i_ino, &key,
4644                                             btrfs_inode_type(inode), index);
4645                 BUG_ON(ret);
4646
4647                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4648                                    name_len * 2);
4649                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4650                 ret = btrfs_update_inode(trans, root, parent_inode);
4651         }
4652         return ret;
4653 }
4654
4655 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4656                             struct inode *dir, struct dentry *dentry,
4657                             struct inode *inode, int backref, u64 index)
4658 {
4659         int err = btrfs_add_link(trans, dir, inode,
4660                                  dentry->d_name.name, dentry->d_name.len,
4661                                  backref, index);
4662         if (!err) {
4663                 d_instantiate(dentry, inode);
4664                 return 0;
4665         }
4666         if (err > 0)
4667                 err = -EEXIST;
4668         return err;
4669 }
4670
4671 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4672                         int mode, dev_t rdev)
4673 {
4674         struct btrfs_trans_handle *trans;
4675         struct btrfs_root *root = BTRFS_I(dir)->root;
4676         struct inode *inode = NULL;
4677         int err;
4678         int drop_inode = 0;
4679         u64 objectid;
4680         unsigned long nr = 0;
4681         u64 index = 0;
4682
4683         if (!new_valid_dev(rdev))
4684                 return -EINVAL;
4685
4686         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4687         if (err)
4688                 return err;
4689
4690         /*
4691          * 2 for inode item and ref
4692          * 2 for dir items
4693          * 1 for xattr if selinux is on
4694          */
4695         trans = btrfs_start_transaction(root, 5);
4696         if (IS_ERR(trans))
4697                 return PTR_ERR(trans);
4698
4699         btrfs_set_trans_block_group(trans, dir);
4700
4701         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4702                                 dentry->d_name.len, dir->i_ino, objectid,
4703                                 BTRFS_I(dir)->block_group, mode, &index);
4704         err = PTR_ERR(inode);
4705         if (IS_ERR(inode))
4706                 goto out_unlock;
4707
4708         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4709         if (err) {
4710                 drop_inode = 1;
4711                 goto out_unlock;
4712         }
4713
4714         btrfs_set_trans_block_group(trans, inode);
4715         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4716         if (err)
4717                 drop_inode = 1;
4718         else {
4719                 inode->i_op = &btrfs_special_inode_operations;
4720                 init_special_inode(inode, inode->i_mode, rdev);
4721                 btrfs_update_inode(trans, root, inode);
4722         }
4723         btrfs_update_inode_block_group(trans, inode);
4724         btrfs_update_inode_block_group(trans, dir);
4725 out_unlock:
4726         nr = trans->blocks_used;
4727         btrfs_end_transaction_throttle(trans, root);
4728         btrfs_btree_balance_dirty(root, nr);
4729         if (drop_inode) {
4730                 inode_dec_link_count(inode);
4731                 iput(inode);
4732         }
4733         return err;
4734 }
4735
4736 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4737                         int mode, struct nameidata *nd)
4738 {
4739         struct btrfs_trans_handle *trans;
4740         struct btrfs_root *root = BTRFS_I(dir)->root;
4741         struct inode *inode = NULL;
4742         int drop_inode = 0;
4743         int err;
4744         unsigned long nr = 0;
4745         u64 objectid;
4746         u64 index = 0;
4747
4748         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4749         if (err)
4750                 return err;
4751         /*
4752          * 2 for inode item and ref
4753          * 2 for dir items
4754          * 1 for xattr if selinux is on
4755          */
4756         trans = btrfs_start_transaction(root, 5);
4757         if (IS_ERR(trans))
4758                 return PTR_ERR(trans);
4759
4760         btrfs_set_trans_block_group(trans, dir);
4761
4762         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4763                                 dentry->d_name.len, dir->i_ino, objectid,
4764                                 BTRFS_I(dir)->block_group, mode, &index);
4765         err = PTR_ERR(inode);
4766         if (IS_ERR(inode))
4767                 goto out_unlock;
4768
4769         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4770         if (err) {
4771                 drop_inode = 1;
4772                 goto out_unlock;
4773         }
4774
4775         btrfs_set_trans_block_group(trans, inode);
4776         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4777         if (err)
4778                 drop_inode = 1;
4779         else {
4780                 inode->i_mapping->a_ops = &btrfs_aops;
4781                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4782                 inode->i_fop = &btrfs_file_operations;
4783                 inode->i_op = &btrfs_file_inode_operations;
4784                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4785         }
4786         btrfs_update_inode_block_group(trans, inode);
4787         btrfs_update_inode_block_group(trans, dir);
4788 out_unlock:
4789         nr = trans->blocks_used;
4790         btrfs_end_transaction_throttle(trans, root);
4791         if (drop_inode) {
4792                 inode_dec_link_count(inode);
4793                 iput(inode);
4794         }
4795         btrfs_btree_balance_dirty(root, nr);
4796         return err;
4797 }
4798
4799 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4800                       struct dentry *dentry)
4801 {
4802         struct btrfs_trans_handle *trans;
4803         struct btrfs_root *root = BTRFS_I(dir)->root;
4804         struct inode *inode = old_dentry->d_inode;
4805         u64 index;
4806         unsigned long nr = 0;
4807         int err;
4808         int drop_inode = 0;
4809
4810         /* do not allow sys_link's with other subvols of the same device */
4811         if (root->objectid != BTRFS_I(inode)->root->objectid)
4812                 return -EPERM;
4813
4814         btrfs_inc_nlink(inode);
4815         inode->i_ctime = CURRENT_TIME;
4816
4817         err = btrfs_set_inode_index(dir, &index);
4818         if (err)
4819                 goto fail;
4820
4821         /*
4822          * 2 items for inode and inode ref
4823          * 2 items for dir items
4824          * 1 item for parent inode
4825          */
4826         trans = btrfs_start_transaction(root, 5);
4827         if (IS_ERR(trans)) {
4828                 err = PTR_ERR(trans);
4829                 goto fail;
4830         }
4831
4832         btrfs_set_trans_block_group(trans, dir);
4833         ihold(inode);
4834
4835         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4836
4837         if (err) {
4838                 drop_inode = 1;
4839         } else {
4840                 struct dentry *parent = dget_parent(dentry);
4841                 btrfs_update_inode_block_group(trans, dir);
4842                 err = btrfs_update_inode(trans, root, inode);
4843                 BUG_ON(err);
4844                 btrfs_log_new_name(trans, inode, NULL, parent);
4845                 dput(parent);
4846         }
4847
4848         nr = trans->blocks_used;
4849         btrfs_end_transaction_throttle(trans, root);
4850 fail:
4851         if (drop_inode) {
4852                 inode_dec_link_count(inode);
4853                 iput(inode);
4854         }
4855         btrfs_btree_balance_dirty(root, nr);
4856         return err;
4857 }
4858
4859 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4860 {
4861         struct inode *inode = NULL;
4862         struct btrfs_trans_handle *trans;
4863         struct btrfs_root *root = BTRFS_I(dir)->root;
4864         int err = 0;
4865         int drop_on_err = 0;
4866         u64 objectid = 0;
4867         u64 index = 0;
4868         unsigned long nr = 1;
4869
4870         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4871         if (err)
4872                 return err;
4873
4874         /*
4875          * 2 items for inode and ref
4876          * 2 items for dir items
4877          * 1 for xattr if selinux is on
4878          */
4879         trans = btrfs_start_transaction(root, 5);
4880         if (IS_ERR(trans))
4881                 return PTR_ERR(trans);
4882         btrfs_set_trans_block_group(trans, dir);
4883
4884         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4885                                 dentry->d_name.len, dir->i_ino, objectid,
4886                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4887                                 &index);
4888         if (IS_ERR(inode)) {
4889                 err = PTR_ERR(inode);
4890                 goto out_fail;
4891         }
4892
4893         drop_on_err = 1;
4894
4895         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4896         if (err)
4897                 goto out_fail;
4898
4899         inode->i_op = &btrfs_dir_inode_operations;
4900         inode->i_fop = &btrfs_dir_file_operations;
4901         btrfs_set_trans_block_group(trans, inode);
4902
4903         btrfs_i_size_write(inode, 0);
4904         err = btrfs_update_inode(trans, root, inode);
4905         if (err)
4906                 goto out_fail;
4907
4908         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4909                              dentry->d_name.len, 0, index);
4910         if (err)
4911                 goto out_fail;
4912
4913         d_instantiate(dentry, inode);
4914         drop_on_err = 0;
4915         btrfs_update_inode_block_group(trans, inode);
4916         btrfs_update_inode_block_group(trans, dir);
4917
4918 out_fail:
4919         nr = trans->blocks_used;
4920         btrfs_end_transaction_throttle(trans, root);
4921         if (drop_on_err)
4922                 iput(inode);
4923         btrfs_btree_balance_dirty(root, nr);
4924         return err;
4925 }
4926
4927 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4928  * and an extent that you want to insert, deal with overlap and insert
4929  * the new extent into the tree.
4930  */
4931 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4932                                 struct extent_map *existing,
4933                                 struct extent_map *em,
4934                                 u64 map_start, u64 map_len)
4935 {
4936         u64 start_diff;
4937
4938         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4939         start_diff = map_start - em->start;
4940         em->start = map_start;
4941         em->len = map_len;
4942         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4943             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4944                 em->block_start += start_diff;
4945                 em->block_len -= start_diff;
4946         }
4947         return add_extent_mapping(em_tree, em);
4948 }
4949
4950 static noinline int uncompress_inline(struct btrfs_path *path,
4951                                       struct inode *inode, struct page *page,
4952                                       size_t pg_offset, u64 extent_offset,
4953                                       struct btrfs_file_extent_item *item)
4954 {
4955         int ret;
4956         struct extent_buffer *leaf = path->nodes[0];
4957         char *tmp;
4958         size_t max_size;
4959         unsigned long inline_size;
4960         unsigned long ptr;
4961         int compress_type;
4962
4963         WARN_ON(pg_offset != 0);
4964         compress_type = btrfs_file_extent_compression(leaf, item);
4965         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4966         inline_size = btrfs_file_extent_inline_item_len(leaf,
4967                                         btrfs_item_nr(leaf, path->slots[0]));
4968         tmp = kmalloc(inline_size, GFP_NOFS);
4969         ptr = btrfs_file_extent_inline_start(item);
4970
4971         read_extent_buffer(leaf, tmp, ptr, inline_size);
4972
4973         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4974         ret = btrfs_decompress(compress_type, tmp, page,
4975                                extent_offset, inline_size, max_size);
4976         if (ret) {
4977                 char *kaddr = kmap_atomic(page, KM_USER0);
4978                 unsigned long copy_size = min_t(u64,
4979                                   PAGE_CACHE_SIZE - pg_offset,
4980                                   max_size - extent_offset);
4981                 memset(kaddr + pg_offset, 0, copy_size);
4982                 kunmap_atomic(kaddr, KM_USER0);
4983         }
4984         kfree(tmp);
4985         return 0;
4986 }
4987
4988 /*
4989  * a bit scary, this does extent mapping from logical file offset to the disk.
4990  * the ugly parts come from merging extents from the disk with the in-ram
4991  * representation.  This gets more complex because of the data=ordered code,
4992  * where the in-ram extents might be locked pending data=ordered completion.
4993  *
4994  * This also copies inline extents directly into the page.
4995  */
4996
4997 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4998                                     size_t pg_offset, u64 start, u64 len,
4999                                     int create)
5000 {
5001         int ret;
5002         int err = 0;
5003         u64 bytenr;
5004         u64 extent_start = 0;
5005         u64 extent_end = 0;
5006         u64 objectid = inode->i_ino;
5007         u32 found_type;
5008         struct btrfs_path *path = NULL;
5009         struct btrfs_root *root = BTRFS_I(inode)->root;
5010         struct btrfs_file_extent_item *item;
5011         struct extent_buffer *leaf;
5012         struct btrfs_key found_key;
5013         struct extent_map *em = NULL;
5014         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5015         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5016         struct btrfs_trans_handle *trans = NULL;
5017         int compress_type;
5018
5019 again:
5020         read_lock(&em_tree->lock);
5021         em = lookup_extent_mapping(em_tree, start, len);
5022         if (em)
5023                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5024         read_unlock(&em_tree->lock);
5025
5026         if (em) {
5027                 if (em->start > start || em->start + em->len <= start)
5028                         free_extent_map(em);
5029                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5030                         free_extent_map(em);
5031                 else
5032                         goto out;
5033         }
5034         em = alloc_extent_map(GFP_NOFS);
5035         if (!em) {
5036                 err = -ENOMEM;
5037                 goto out;
5038         }
5039         em->bdev = root->fs_info->fs_devices->latest_bdev;
5040         em->start = EXTENT_MAP_HOLE;
5041         em->orig_start = EXTENT_MAP_HOLE;
5042         em->len = (u64)-1;
5043         em->block_len = (u64)-1;
5044
5045         if (!path) {
5046                 path = btrfs_alloc_path();
5047                 BUG_ON(!path);
5048         }
5049
5050         ret = btrfs_lookup_file_extent(trans, root, path,
5051                                        objectid, start, trans != NULL);
5052         if (ret < 0) {
5053                 err = ret;
5054                 goto out;
5055         }
5056
5057         if (ret != 0) {
5058                 if (path->slots[0] == 0)
5059                         goto not_found;
5060                 path->slots[0]--;
5061         }
5062
5063         leaf = path->nodes[0];
5064         item = btrfs_item_ptr(leaf, path->slots[0],
5065                               struct btrfs_file_extent_item);
5066         /* are we inside the extent that was found? */
5067         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5068         found_type = btrfs_key_type(&found_key);
5069         if (found_key.objectid != objectid ||
5070             found_type != BTRFS_EXTENT_DATA_KEY) {
5071                 goto not_found;
5072         }
5073
5074         found_type = btrfs_file_extent_type(leaf, item);
5075         extent_start = found_key.offset;
5076         compress_type = btrfs_file_extent_compression(leaf, item);
5077         if (found_type == BTRFS_FILE_EXTENT_REG ||
5078             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5079                 extent_end = extent_start +
5080                        btrfs_file_extent_num_bytes(leaf, item);
5081         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5082                 size_t size;
5083                 size = btrfs_file_extent_inline_len(leaf, item);
5084                 extent_end = (extent_start + size + root->sectorsize - 1) &
5085                         ~((u64)root->sectorsize - 1);
5086         }
5087
5088         if (start >= extent_end) {
5089                 path->slots[0]++;
5090                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5091                         ret = btrfs_next_leaf(root, path);
5092                         if (ret < 0) {
5093                                 err = ret;
5094                                 goto out;
5095                         }
5096                         if (ret > 0)
5097                                 goto not_found;
5098                         leaf = path->nodes[0];
5099                 }
5100                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5101                 if (found_key.objectid != objectid ||
5102                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5103                         goto not_found;
5104                 if (start + len <= found_key.offset)
5105                         goto not_found;
5106                 em->start = start;
5107                 em->len = found_key.offset - start;
5108                 goto not_found_em;
5109         }
5110
5111         if (found_type == BTRFS_FILE_EXTENT_REG ||
5112             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5113                 em->start = extent_start;
5114                 em->len = extent_end - extent_start;
5115                 em->orig_start = extent_start -
5116                                  btrfs_file_extent_offset(leaf, item);
5117                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5118                 if (bytenr == 0) {
5119                         em->block_start = EXTENT_MAP_HOLE;
5120                         goto insert;
5121                 }
5122                 if (compress_type != BTRFS_COMPRESS_NONE) {
5123                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5124                         em->compress_type = compress_type;
5125                         em->block_start = bytenr;
5126                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5127                                                                          item);
5128                 } else {
5129                         bytenr += btrfs_file_extent_offset(leaf, item);
5130                         em->block_start = bytenr;
5131                         em->block_len = em->len;
5132                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5133                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5134                 }
5135                 goto insert;
5136         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5137                 unsigned long ptr;
5138                 char *map;
5139                 size_t size;
5140                 size_t extent_offset;
5141                 size_t copy_size;
5142
5143                 em->block_start = EXTENT_MAP_INLINE;
5144                 if (!page || create) {
5145                         em->start = extent_start;
5146                         em->len = extent_end - extent_start;
5147                         goto out;
5148                 }
5149
5150                 size = btrfs_file_extent_inline_len(leaf, item);
5151                 extent_offset = page_offset(page) + pg_offset - extent_start;
5152                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5153                                 size - extent_offset);
5154                 em->start = extent_start + extent_offset;
5155                 em->len = (copy_size + root->sectorsize - 1) &
5156                         ~((u64)root->sectorsize - 1);
5157                 em->orig_start = EXTENT_MAP_INLINE;
5158                 if (compress_type) {
5159                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5160                         em->compress_type = compress_type;
5161                 }
5162                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5163                 if (create == 0 && !PageUptodate(page)) {
5164                         if (btrfs_file_extent_compression(leaf, item) !=
5165                             BTRFS_COMPRESS_NONE) {
5166                                 ret = uncompress_inline(path, inode, page,
5167                                                         pg_offset,
5168                                                         extent_offset, item);
5169                                 BUG_ON(ret);
5170                         } else {
5171                                 map = kmap(page);
5172                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5173                                                    copy_size);
5174                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5175                                         memset(map + pg_offset + copy_size, 0,
5176                                                PAGE_CACHE_SIZE - pg_offset -
5177                                                copy_size);
5178                                 }
5179                                 kunmap(page);
5180                         }
5181                         flush_dcache_page(page);
5182                 } else if (create && PageUptodate(page)) {
5183                         WARN_ON(1);
5184                         if (!trans) {
5185                                 kunmap(page);
5186                                 free_extent_map(em);
5187                                 em = NULL;
5188                                 btrfs_release_path(root, path);
5189                                 trans = btrfs_join_transaction(root, 1);
5190                                 if (IS_ERR(trans))
5191                                         return ERR_CAST(trans);
5192                                 goto again;
5193                         }
5194                         map = kmap(page);
5195                         write_extent_buffer(leaf, map + pg_offset, ptr,
5196                                             copy_size);
5197                         kunmap(page);
5198                         btrfs_mark_buffer_dirty(leaf);
5199                 }
5200                 set_extent_uptodate(io_tree, em->start,
5201                                     extent_map_end(em) - 1, GFP_NOFS);
5202                 goto insert;
5203         } else {
5204                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5205                 WARN_ON(1);
5206         }
5207 not_found:
5208         em->start = start;
5209         em->len = len;
5210 not_found_em:
5211         em->block_start = EXTENT_MAP_HOLE;
5212         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5213 insert:
5214         btrfs_release_path(root, path);
5215         if (em->start > start || extent_map_end(em) <= start) {
5216                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5217                        "[%llu %llu]\n", (unsigned long long)em->start,
5218                        (unsigned long long)em->len,
5219                        (unsigned long long)start,
5220                        (unsigned long long)len);
5221                 err = -EIO;
5222                 goto out;
5223         }
5224
5225         err = 0;
5226         write_lock(&em_tree->lock);
5227         ret = add_extent_mapping(em_tree, em);
5228         /* it is possible that someone inserted the extent into the tree
5229          * while we had the lock dropped.  It is also possible that
5230          * an overlapping map exists in the tree
5231          */
5232         if (ret == -EEXIST) {
5233                 struct extent_map *existing;
5234
5235                 ret = 0;
5236
5237                 existing = lookup_extent_mapping(em_tree, start, len);
5238                 if (existing && (existing->start > start ||
5239                     existing->start + existing->len <= start)) {
5240                         free_extent_map(existing);
5241                         existing = NULL;
5242                 }
5243                 if (!existing) {
5244                         existing = lookup_extent_mapping(em_tree, em->start,
5245                                                          em->len);
5246                         if (existing) {
5247                                 err = merge_extent_mapping(em_tree, existing,
5248                                                            em, start,
5249                                                            root->sectorsize);
5250                                 free_extent_map(existing);
5251                                 if (err) {
5252                                         free_extent_map(em);
5253                                         em = NULL;
5254                                 }
5255                         } else {
5256                                 err = -EIO;
5257                                 free_extent_map(em);
5258                                 em = NULL;
5259                         }
5260                 } else {
5261                         free_extent_map(em);
5262                         em = existing;
5263                         err = 0;
5264                 }
5265         }
5266         write_unlock(&em_tree->lock);
5267 out:
5268         if (path)
5269                 btrfs_free_path(path);
5270         if (trans) {
5271                 ret = btrfs_end_transaction(trans, root);
5272                 if (!err)
5273                         err = ret;
5274         }
5275         if (err) {
5276                 free_extent_map(em);
5277                 return ERR_PTR(err);
5278         }
5279         return em;
5280 }
5281
5282 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5283                                            size_t pg_offset, u64 start, u64 len,
5284                                            int create)
5285 {
5286         struct extent_map *em;
5287         struct extent_map *hole_em = NULL;
5288         u64 range_start = start;
5289         u64 end;
5290         u64 found;
5291         u64 found_end;
5292         int err = 0;
5293
5294         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5295         if (IS_ERR(em))
5296                 return em;
5297         if (em) {
5298                 /*
5299                  * if our em maps to a hole, there might
5300                  * actually be delalloc bytes behind it
5301                  */
5302                 if (em->block_start != EXTENT_MAP_HOLE)
5303                         return em;
5304                 else
5305                         hole_em = em;
5306         }
5307
5308         /* check to see if we've wrapped (len == -1 or similar) */
5309         end = start + len;
5310         if (end < start)
5311                 end = (u64)-1;
5312         else
5313                 end -= 1;
5314
5315         em = NULL;
5316
5317         /* ok, we didn't find anything, lets look for delalloc */
5318         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5319                                  end, len, EXTENT_DELALLOC, 1);
5320         found_end = range_start + found;
5321         if (found_end < range_start)
5322                 found_end = (u64)-1;
5323
5324         /*
5325          * we didn't find anything useful, return
5326          * the original results from get_extent()
5327          */
5328         if (range_start > end || found_end <= start) {
5329                 em = hole_em;
5330                 hole_em = NULL;
5331                 goto out;
5332         }
5333
5334         /* adjust the range_start to make sure it doesn't
5335          * go backwards from the start they passed in
5336          */
5337         range_start = max(start,range_start);
5338         found = found_end - range_start;
5339
5340         if (found > 0) {
5341                 u64 hole_start = start;
5342                 u64 hole_len = len;
5343
5344                 em = alloc_extent_map(GFP_NOFS);
5345                 if (!em) {
5346                         err = -ENOMEM;
5347                         goto out;
5348                 }
5349                 /*
5350                  * when btrfs_get_extent can't find anything it
5351                  * returns one huge hole
5352                  *
5353                  * make sure what it found really fits our range, and
5354                  * adjust to make sure it is based on the start from
5355                  * the caller
5356                  */
5357                 if (hole_em) {
5358                         u64 calc_end = extent_map_end(hole_em);
5359
5360                         if (calc_end <= start || (hole_em->start > end)) {
5361                                 free_extent_map(hole_em);
5362                                 hole_em = NULL;
5363                         } else {
5364                                 hole_start = max(hole_em->start, start);
5365                                 hole_len = calc_end - hole_start;
5366                         }
5367                 }
5368                 em->bdev = NULL;
5369                 if (hole_em && range_start > hole_start) {
5370                         /* our hole starts before our delalloc, so we
5371                          * have to return just the parts of the hole
5372                          * that go until  the delalloc starts
5373                          */
5374                         em->len = min(hole_len,
5375                                       range_start - hole_start);
5376                         em->start = hole_start;
5377                         em->orig_start = hole_start;
5378                         /*
5379                          * don't adjust block start at all,
5380                          * it is fixed at EXTENT_MAP_HOLE
5381                          */
5382                         em->block_start = hole_em->block_start;
5383                         em->block_len = hole_len;
5384                 } else {
5385                         em->start = range_start;
5386                         em->len = found;
5387                         em->orig_start = range_start;
5388                         em->block_start = EXTENT_MAP_DELALLOC;
5389                         em->block_len = found;
5390                 }
5391         } else if (hole_em) {
5392                 return hole_em;
5393         }
5394 out:
5395
5396         free_extent_map(hole_em);
5397         if (err) {
5398                 free_extent_map(em);
5399                 return ERR_PTR(err);
5400         }
5401         return em;
5402 }
5403
5404 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5405                                                   u64 start, u64 len)
5406 {
5407         struct btrfs_root *root = BTRFS_I(inode)->root;
5408         struct btrfs_trans_handle *trans;
5409         struct extent_map *em;
5410         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5411         struct btrfs_key ins;
5412         u64 alloc_hint;
5413         int ret;
5414
5415         btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5416
5417         trans = btrfs_join_transaction(root, 0);
5418         if (IS_ERR(trans))
5419                 return ERR_CAST(trans);
5420
5421         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5422
5423         alloc_hint = get_extent_allocation_hint(inode, start, len);
5424         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5425                                    alloc_hint, (u64)-1, &ins, 1);
5426         if (ret) {
5427                 em = ERR_PTR(ret);
5428                 goto out;
5429         }
5430
5431         em = alloc_extent_map(GFP_NOFS);
5432         if (!em) {
5433                 em = ERR_PTR(-ENOMEM);
5434                 goto out;
5435         }
5436
5437         em->start = start;
5438         em->orig_start = em->start;
5439         em->len = ins.offset;
5440
5441         em->block_start = ins.objectid;
5442         em->block_len = ins.offset;
5443         em->bdev = root->fs_info->fs_devices->latest_bdev;
5444         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5445
5446         while (1) {
5447                 write_lock(&em_tree->lock);
5448                 ret = add_extent_mapping(em_tree, em);
5449                 write_unlock(&em_tree->lock);
5450                 if (ret != -EEXIST)
5451                         break;
5452                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5453         }
5454
5455         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5456                                            ins.offset, ins.offset, 0);
5457         if (ret) {
5458                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5459                 em = ERR_PTR(ret);
5460         }
5461 out:
5462         btrfs_end_transaction(trans, root);
5463         return em;
5464 }
5465
5466 /*
5467  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5468  * block must be cow'd
5469  */
5470 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5471                                       struct inode *inode, u64 offset, u64 len)
5472 {
5473         struct btrfs_path *path;
5474         int ret;
5475         struct extent_buffer *leaf;
5476         struct btrfs_root *root = BTRFS_I(inode)->root;
5477         struct btrfs_file_extent_item *fi;
5478         struct btrfs_key key;
5479         u64 disk_bytenr;
5480         u64 backref_offset;
5481         u64 extent_end;
5482         u64 num_bytes;
5483         int slot;
5484         int found_type;
5485
5486         path = btrfs_alloc_path();
5487         if (!path)
5488                 return -ENOMEM;
5489
5490         ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5491                                        offset, 0);
5492         if (ret < 0)
5493                 goto out;
5494
5495         slot = path->slots[0];
5496         if (ret == 1) {
5497                 if (slot == 0) {
5498                         /* can't find the item, must cow */
5499                         ret = 0;
5500                         goto out;
5501                 }
5502                 slot--;
5503         }
5504         ret = 0;
5505         leaf = path->nodes[0];
5506         btrfs_item_key_to_cpu(leaf, &key, slot);
5507         if (key.objectid != inode->i_ino ||
5508             key.type != BTRFS_EXTENT_DATA_KEY) {
5509                 /* not our file or wrong item type, must cow */
5510                 goto out;
5511         }
5512
5513         if (key.offset > offset) {
5514                 /* Wrong offset, must cow */
5515                 goto out;
5516         }
5517
5518         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5519         found_type = btrfs_file_extent_type(leaf, fi);
5520         if (found_type != BTRFS_FILE_EXTENT_REG &&
5521             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5522                 /* not a regular extent, must cow */
5523                 goto out;
5524         }
5525         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5526         backref_offset = btrfs_file_extent_offset(leaf, fi);
5527
5528         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5529         if (extent_end < offset + len) {
5530                 /* extent doesn't include our full range, must cow */
5531                 goto out;
5532         }
5533
5534         if (btrfs_extent_readonly(root, disk_bytenr))
5535                 goto out;
5536
5537         /*
5538          * look for other files referencing this extent, if we
5539          * find any we must cow
5540          */
5541         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5542                                   key.offset - backref_offset, disk_bytenr))
5543                 goto out;
5544
5545         /*
5546          * adjust disk_bytenr and num_bytes to cover just the bytes
5547          * in this extent we are about to write.  If there
5548          * are any csums in that range we have to cow in order
5549          * to keep the csums correct
5550          */
5551         disk_bytenr += backref_offset;
5552         disk_bytenr += offset - key.offset;
5553         num_bytes = min(offset + len, extent_end) - offset;
5554         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5555                                 goto out;
5556         /*
5557          * all of the above have passed, it is safe to overwrite this extent
5558          * without cow
5559          */
5560         ret = 1;
5561 out:
5562         btrfs_free_path(path);
5563         return ret;
5564 }
5565
5566 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5567                                    struct buffer_head *bh_result, int create)
5568 {
5569         struct extent_map *em;
5570         struct btrfs_root *root = BTRFS_I(inode)->root;
5571         u64 start = iblock << inode->i_blkbits;
5572         u64 len = bh_result->b_size;
5573         struct btrfs_trans_handle *trans;
5574
5575         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5576         if (IS_ERR(em))
5577                 return PTR_ERR(em);
5578
5579         /*
5580          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5581          * io.  INLINE is special, and we could probably kludge it in here, but
5582          * it's still buffered so for safety lets just fall back to the generic
5583          * buffered path.
5584          *
5585          * For COMPRESSED we _have_ to read the entire extent in so we can
5586          * decompress it, so there will be buffering required no matter what we
5587          * do, so go ahead and fallback to buffered.
5588          *
5589          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5590          * to buffered IO.  Don't blame me, this is the price we pay for using
5591          * the generic code.
5592          */
5593         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5594             em->block_start == EXTENT_MAP_INLINE) {
5595                 free_extent_map(em);
5596                 return -ENOTBLK;
5597         }
5598
5599         /* Just a good old fashioned hole, return */
5600         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5601                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5602                 free_extent_map(em);
5603                 /* DIO will do one hole at a time, so just unlock a sector */
5604                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5605                               start + root->sectorsize - 1, GFP_NOFS);
5606                 return 0;
5607         }
5608
5609         /*
5610          * We don't allocate a new extent in the following cases
5611          *
5612          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5613          * existing extent.
5614          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5615          * just use the extent.
5616          *
5617          */
5618         if (!create) {
5619                 len = em->len - (start - em->start);
5620                 goto map;
5621         }
5622
5623         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5624             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5625              em->block_start != EXTENT_MAP_HOLE)) {
5626                 int type;
5627                 int ret;
5628                 u64 block_start;
5629
5630                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5631                         type = BTRFS_ORDERED_PREALLOC;
5632                 else
5633                         type = BTRFS_ORDERED_NOCOW;
5634                 len = min(len, em->len - (start - em->start));
5635                 block_start = em->block_start + (start - em->start);
5636
5637                 /*
5638                  * we're not going to log anything, but we do need
5639                  * to make sure the current transaction stays open
5640                  * while we look for nocow cross refs
5641                  */
5642                 trans = btrfs_join_transaction(root, 0);
5643                 if (IS_ERR(trans))
5644                         goto must_cow;
5645
5646                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5647                         ret = btrfs_add_ordered_extent_dio(inode, start,
5648                                            block_start, len, len, type);
5649                         btrfs_end_transaction(trans, root);
5650                         if (ret) {
5651                                 free_extent_map(em);
5652                                 return ret;
5653                         }
5654                         goto unlock;
5655                 }
5656                 btrfs_end_transaction(trans, root);
5657         }
5658 must_cow:
5659         /*
5660          * this will cow the extent, reset the len in case we changed
5661          * it above
5662          */
5663         len = bh_result->b_size;
5664         free_extent_map(em);
5665         em = btrfs_new_extent_direct(inode, start, len);
5666         if (IS_ERR(em))
5667                 return PTR_ERR(em);
5668         len = min(len, em->len - (start - em->start));
5669 unlock:
5670         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5671                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5672                           0, NULL, GFP_NOFS);
5673 map:
5674         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5675                 inode->i_blkbits;
5676         bh_result->b_size = len;
5677         bh_result->b_bdev = em->bdev;
5678         set_buffer_mapped(bh_result);
5679         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5680                 set_buffer_new(bh_result);
5681
5682         free_extent_map(em);
5683
5684         return 0;
5685 }
5686
5687 struct btrfs_dio_private {
5688         struct inode *inode;
5689         u64 logical_offset;
5690         u64 disk_bytenr;
5691         u64 bytes;
5692         u32 *csums;
5693         void *private;
5694
5695         /* number of bios pending for this dio */
5696         atomic_t pending_bios;
5697
5698         /* IO errors */
5699         int errors;
5700
5701         struct bio *orig_bio;
5702 };
5703
5704 static void btrfs_endio_direct_read(struct bio *bio, int err)
5705 {
5706         struct btrfs_dio_private *dip = bio->bi_private;
5707         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5708         struct bio_vec *bvec = bio->bi_io_vec;
5709         struct inode *inode = dip->inode;
5710         struct btrfs_root *root = BTRFS_I(inode)->root;
5711         u64 start;
5712         u32 *private = dip->csums;
5713
5714         start = dip->logical_offset;
5715         do {
5716                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5717                         struct page *page = bvec->bv_page;
5718                         char *kaddr;
5719                         u32 csum = ~(u32)0;
5720                         unsigned long flags;
5721
5722                         local_irq_save(flags);
5723                         kaddr = kmap_atomic(page, KM_IRQ0);
5724                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5725                                                csum, bvec->bv_len);
5726                         btrfs_csum_final(csum, (char *)&csum);
5727                         kunmap_atomic(kaddr, KM_IRQ0);
5728                         local_irq_restore(flags);
5729
5730                         flush_dcache_page(bvec->bv_page);
5731                         if (csum != *private) {
5732                                 printk(KERN_ERR "btrfs csum failed ino %lu off"
5733                                       " %llu csum %u private %u\n",
5734                                       inode->i_ino, (unsigned long long)start,
5735                                       csum, *private);
5736                                 err = -EIO;
5737                         }
5738                 }
5739
5740                 start += bvec->bv_len;
5741                 private++;
5742                 bvec++;
5743         } while (bvec <= bvec_end);
5744
5745         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5746                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5747         bio->bi_private = dip->private;
5748
5749         kfree(dip->csums);
5750         kfree(dip);
5751         dio_end_io(bio, err);
5752 }
5753
5754 static void btrfs_endio_direct_write(struct bio *bio, int err)
5755 {
5756         struct btrfs_dio_private *dip = bio->bi_private;
5757         struct inode *inode = dip->inode;
5758         struct btrfs_root *root = BTRFS_I(inode)->root;
5759         struct btrfs_trans_handle *trans;
5760         struct btrfs_ordered_extent *ordered = NULL;
5761         struct extent_state *cached_state = NULL;
5762         u64 ordered_offset = dip->logical_offset;
5763         u64 ordered_bytes = dip->bytes;
5764         int ret;
5765
5766         if (err)
5767                 goto out_done;
5768 again:
5769         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5770                                                    &ordered_offset,
5771                                                    ordered_bytes);
5772         if (!ret)
5773                 goto out_test;
5774
5775         BUG_ON(!ordered);
5776
5777         trans = btrfs_join_transaction(root, 1);
5778         if (IS_ERR(trans)) {
5779                 err = -ENOMEM;
5780                 goto out;
5781         }
5782         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5783
5784         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5785                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5786                 if (!ret)
5787                         ret = btrfs_update_inode(trans, root, inode);
5788                 err = ret;
5789                 goto out;
5790         }
5791
5792         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5793                          ordered->file_offset + ordered->len - 1, 0,
5794                          &cached_state, GFP_NOFS);
5795
5796         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5797                 ret = btrfs_mark_extent_written(trans, inode,
5798                                                 ordered->file_offset,
5799                                                 ordered->file_offset +
5800                                                 ordered->len);
5801                 if (ret) {
5802                         err = ret;
5803                         goto out_unlock;
5804                 }
5805         } else {
5806                 ret = insert_reserved_file_extent(trans, inode,
5807                                                   ordered->file_offset,
5808                                                   ordered->start,
5809                                                   ordered->disk_len,
5810                                                   ordered->len,
5811                                                   ordered->len,
5812                                                   0, 0, 0,
5813                                                   BTRFS_FILE_EXTENT_REG);
5814                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5815                                    ordered->file_offset, ordered->len);
5816                 if (ret) {
5817                         err = ret;
5818                         WARN_ON(1);
5819                         goto out_unlock;
5820                 }
5821         }
5822
5823         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5824         btrfs_ordered_update_i_size(inode, 0, ordered);
5825         btrfs_update_inode(trans, root, inode);
5826 out_unlock:
5827         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5828                              ordered->file_offset + ordered->len - 1,
5829                              &cached_state, GFP_NOFS);
5830 out:
5831         btrfs_delalloc_release_metadata(inode, ordered->len);
5832         btrfs_end_transaction(trans, root);
5833         ordered_offset = ordered->file_offset + ordered->len;
5834         btrfs_put_ordered_extent(ordered);
5835         btrfs_put_ordered_extent(ordered);
5836
5837 out_test:
5838         /*
5839          * our bio might span multiple ordered extents.  If we haven't
5840          * completed the accounting for the whole dio, go back and try again
5841          */
5842         if (ordered_offset < dip->logical_offset + dip->bytes) {
5843                 ordered_bytes = dip->logical_offset + dip->bytes -
5844                         ordered_offset;
5845                 goto again;
5846         }
5847 out_done:
5848         bio->bi_private = dip->private;
5849
5850         kfree(dip->csums);
5851         kfree(dip);
5852         dio_end_io(bio, err);
5853 }
5854
5855 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5856                                     struct bio *bio, int mirror_num,
5857                                     unsigned long bio_flags, u64 offset)
5858 {
5859         int ret;
5860         struct btrfs_root *root = BTRFS_I(inode)->root;
5861         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5862         BUG_ON(ret);
5863         return 0;
5864 }
5865
5866 static void btrfs_end_dio_bio(struct bio *bio, int err)
5867 {
5868         struct btrfs_dio_private *dip = bio->bi_private;
5869
5870         if (err) {
5871                 printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu "
5872                       "sector %#Lx len %u err no %d\n",
5873                       dip->inode->i_ino, bio->bi_rw,
5874                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5875                 dip->errors = 1;
5876
5877                 /*
5878                  * before atomic variable goto zero, we must make sure
5879                  * dip->errors is perceived to be set.
5880                  */
5881                 smp_mb__before_atomic_dec();
5882         }
5883
5884         /* if there are more bios still pending for this dio, just exit */
5885         if (!atomic_dec_and_test(&dip->pending_bios))
5886                 goto out;
5887
5888         if (dip->errors)
5889                 bio_io_error(dip->orig_bio);
5890         else {
5891                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5892                 bio_endio(dip->orig_bio, 0);
5893         }
5894 out:
5895         bio_put(bio);
5896 }
5897
5898 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5899                                        u64 first_sector, gfp_t gfp_flags)
5900 {
5901         int nr_vecs = bio_get_nr_vecs(bdev);
5902         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5903 }
5904
5905 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5906                                          int rw, u64 file_offset, int skip_sum,
5907                                          u32 *csums)
5908 {
5909         int write = rw & REQ_WRITE;
5910         struct btrfs_root *root = BTRFS_I(inode)->root;
5911         int ret;
5912
5913         bio_get(bio);
5914         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5915         if (ret)
5916                 goto err;
5917
5918         if (write && !skip_sum) {
5919                 ret = btrfs_wq_submit_bio(root->fs_info,
5920                                    inode, rw, bio, 0, 0,
5921                                    file_offset,
5922                                    __btrfs_submit_bio_start_direct_io,
5923                                    __btrfs_submit_bio_done);
5924                 goto err;
5925         } else if (!skip_sum)
5926                 btrfs_lookup_bio_sums_dio(root, inode, bio,
5927                                           file_offset, csums);
5928
5929         ret = btrfs_map_bio(root, rw, bio, 0, 1);
5930 err:
5931         bio_put(bio);
5932         return ret;
5933 }
5934
5935 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5936                                     int skip_sum)
5937 {
5938         struct inode *inode = dip->inode;
5939         struct btrfs_root *root = BTRFS_I(inode)->root;
5940         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5941         struct bio *bio;
5942         struct bio *orig_bio = dip->orig_bio;
5943         struct bio_vec *bvec = orig_bio->bi_io_vec;
5944         u64 start_sector = orig_bio->bi_sector;
5945         u64 file_offset = dip->logical_offset;
5946         u64 submit_len = 0;
5947         u64 map_length;
5948         int nr_pages = 0;
5949         u32 *csums = dip->csums;
5950         int ret = 0;
5951
5952         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5953         if (!bio)
5954                 return -ENOMEM;
5955         bio->bi_private = dip;
5956         bio->bi_end_io = btrfs_end_dio_bio;
5957         atomic_inc(&dip->pending_bios);
5958
5959         map_length = orig_bio->bi_size;
5960         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5961                               &map_length, NULL, 0);
5962         if (ret) {
5963                 bio_put(bio);
5964                 return -EIO;
5965         }
5966
5967         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5968                 if (unlikely(map_length < submit_len + bvec->bv_len ||
5969                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5970                                  bvec->bv_offset) < bvec->bv_len)) {
5971                         /*
5972                          * inc the count before we submit the bio so
5973                          * we know the end IO handler won't happen before
5974                          * we inc the count. Otherwise, the dip might get freed
5975                          * before we're done setting it up
5976                          */
5977                         atomic_inc(&dip->pending_bios);
5978                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
5979                                                      file_offset, skip_sum,
5980                                                      csums);
5981                         if (ret) {
5982                                 bio_put(bio);
5983                                 atomic_dec(&dip->pending_bios);
5984                                 goto out_err;
5985                         }
5986
5987                         if (!skip_sum)
5988                                 csums = csums + nr_pages;
5989                         start_sector += submit_len >> 9;
5990                         file_offset += submit_len;
5991
5992                         submit_len = 0;
5993                         nr_pages = 0;
5994
5995                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
5996                                                   start_sector, GFP_NOFS);
5997                         if (!bio)
5998                                 goto out_err;
5999                         bio->bi_private = dip;
6000                         bio->bi_end_io = btrfs_end_dio_bio;
6001
6002                         map_length = orig_bio->bi_size;
6003                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6004                                               &map_length, NULL, 0);
6005                         if (ret) {
6006                                 bio_put(bio);
6007                                 goto out_err;
6008                         }
6009                 } else {
6010                         submit_len += bvec->bv_len;
6011                         nr_pages ++;
6012                         bvec++;
6013                 }
6014         }
6015
6016         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6017                                      csums);
6018         if (!ret)
6019                 return 0;
6020
6021         bio_put(bio);
6022 out_err:
6023         dip->errors = 1;
6024         /*
6025          * before atomic variable goto zero, we must
6026          * make sure dip->errors is perceived to be set.
6027          */
6028         smp_mb__before_atomic_dec();
6029         if (atomic_dec_and_test(&dip->pending_bios))
6030                 bio_io_error(dip->orig_bio);
6031
6032         /* bio_end_io() will handle error, so we needn't return it */
6033         return 0;
6034 }
6035
6036 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6037                                 loff_t file_offset)
6038 {
6039         struct btrfs_root *root = BTRFS_I(inode)->root;
6040         struct btrfs_dio_private *dip;
6041         struct bio_vec *bvec = bio->bi_io_vec;
6042         int skip_sum;
6043         int write = rw & REQ_WRITE;
6044         int ret = 0;
6045
6046         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6047
6048         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6049         if (!dip) {
6050                 ret = -ENOMEM;
6051                 goto free_ordered;
6052         }
6053         dip->csums = NULL;
6054
6055         if (!skip_sum) {
6056                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6057                 if (!dip->csums) {
6058                         kfree(dip);
6059                         ret = -ENOMEM;
6060                         goto free_ordered;
6061                 }
6062         }
6063
6064         dip->private = bio->bi_private;
6065         dip->inode = inode;
6066         dip->logical_offset = file_offset;
6067
6068         dip->bytes = 0;
6069         do {
6070                 dip->bytes += bvec->bv_len;
6071                 bvec++;
6072         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6073
6074         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6075         bio->bi_private = dip;
6076         dip->errors = 0;
6077         dip->orig_bio = bio;
6078         atomic_set(&dip->pending_bios, 0);
6079
6080         if (write)
6081                 bio->bi_end_io = btrfs_endio_direct_write;
6082         else
6083                 bio->bi_end_io = btrfs_endio_direct_read;
6084
6085         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6086         if (!ret)
6087                 return;
6088 free_ordered:
6089         /*
6090          * If this is a write, we need to clean up the reserved space and kill
6091          * the ordered extent.
6092          */
6093         if (write) {
6094                 struct btrfs_ordered_extent *ordered;
6095                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6096                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6097                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6098                         btrfs_free_reserved_extent(root, ordered->start,
6099                                                    ordered->disk_len);
6100                 btrfs_put_ordered_extent(ordered);
6101                 btrfs_put_ordered_extent(ordered);
6102         }
6103         bio_endio(bio, ret);
6104 }
6105
6106 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6107                         const struct iovec *iov, loff_t offset,
6108                         unsigned long nr_segs)
6109 {
6110         int seg;
6111         size_t size;
6112         unsigned long addr;
6113         unsigned blocksize_mask = root->sectorsize - 1;
6114         ssize_t retval = -EINVAL;
6115         loff_t end = offset;
6116
6117         if (offset & blocksize_mask)
6118                 goto out;
6119
6120         /* Check the memory alignment.  Blocks cannot straddle pages */
6121         for (seg = 0; seg < nr_segs; seg++) {
6122                 addr = (unsigned long)iov[seg].iov_base;
6123                 size = iov[seg].iov_len;
6124                 end += size;
6125                 if ((addr & blocksize_mask) || (size & blocksize_mask)) 
6126                         goto out;
6127         }
6128         retval = 0;
6129 out:
6130         return retval;
6131 }
6132 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6133                         const struct iovec *iov, loff_t offset,
6134                         unsigned long nr_segs)
6135 {
6136         struct file *file = iocb->ki_filp;
6137         struct inode *inode = file->f_mapping->host;
6138         struct btrfs_ordered_extent *ordered;
6139         struct extent_state *cached_state = NULL;
6140         u64 lockstart, lockend;
6141         ssize_t ret;
6142         int writing = rw & WRITE;
6143         int write_bits = 0;
6144         size_t count = iov_length(iov, nr_segs);
6145
6146         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6147                             offset, nr_segs)) {
6148                 return 0;
6149         }
6150
6151         lockstart = offset;
6152         lockend = offset + count - 1;
6153
6154         if (writing) {
6155                 ret = btrfs_delalloc_reserve_space(inode, count);
6156                 if (ret)
6157                         goto out;
6158         }
6159
6160         while (1) {
6161                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6162                                  0, &cached_state, GFP_NOFS);
6163                 /*
6164                  * We're concerned with the entire range that we're going to be
6165                  * doing DIO to, so we need to make sure theres no ordered
6166                  * extents in this range.
6167                  */
6168                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6169                                                      lockend - lockstart + 1);
6170                 if (!ordered)
6171                         break;
6172                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6173                                      &cached_state, GFP_NOFS);
6174                 btrfs_start_ordered_extent(inode, ordered, 1);
6175                 btrfs_put_ordered_extent(ordered);
6176                 cond_resched();
6177         }
6178
6179         /*
6180          * we don't use btrfs_set_extent_delalloc because we don't want
6181          * the dirty or uptodate bits
6182          */
6183         if (writing) {
6184                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6185                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6186                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6187                                      GFP_NOFS);
6188                 if (ret) {
6189                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6190                                          lockend, EXTENT_LOCKED | write_bits,
6191                                          1, 0, &cached_state, GFP_NOFS);
6192                         goto out;
6193                 }
6194         }
6195
6196         free_extent_state(cached_state);
6197         cached_state = NULL;
6198
6199         ret = __blockdev_direct_IO(rw, iocb, inode,
6200                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6201                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6202                    btrfs_submit_direct, 0);
6203
6204         if (ret < 0 && ret != -EIOCBQUEUED) {
6205                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6206                               offset + iov_length(iov, nr_segs) - 1,
6207                               EXTENT_LOCKED | write_bits, 1, 0,
6208                               &cached_state, GFP_NOFS);
6209         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6210                 /*
6211                  * We're falling back to buffered, unlock the section we didn't
6212                  * do IO on.
6213                  */
6214                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6215                               offset + iov_length(iov, nr_segs) - 1,
6216                               EXTENT_LOCKED | write_bits, 1, 0,
6217                               &cached_state, GFP_NOFS);
6218         }
6219 out:
6220         free_extent_state(cached_state);
6221         return ret;
6222 }
6223
6224 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6225                 __u64 start, __u64 len)
6226 {
6227         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6228 }
6229
6230 int btrfs_readpage(struct file *file, struct page *page)
6231 {
6232         struct extent_io_tree *tree;
6233         tree = &BTRFS_I(page->mapping->host)->io_tree;
6234         return extent_read_full_page(tree, page, btrfs_get_extent);
6235 }
6236
6237 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6238 {
6239         struct extent_io_tree *tree;
6240
6241
6242         if (current->flags & PF_MEMALLOC) {
6243                 redirty_page_for_writepage(wbc, page);
6244                 unlock_page(page);
6245                 return 0;
6246         }
6247         tree = &BTRFS_I(page->mapping->host)->io_tree;
6248         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6249 }
6250
6251 int btrfs_writepages(struct address_space *mapping,
6252                      struct writeback_control *wbc)
6253 {
6254         struct extent_io_tree *tree;
6255
6256         tree = &BTRFS_I(mapping->host)->io_tree;
6257         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6258 }
6259
6260 static int
6261 btrfs_readpages(struct file *file, struct address_space *mapping,
6262                 struct list_head *pages, unsigned nr_pages)
6263 {
6264         struct extent_io_tree *tree;
6265         tree = &BTRFS_I(mapping->host)->io_tree;
6266         return extent_readpages(tree, mapping, pages, nr_pages,
6267                                 btrfs_get_extent);
6268 }
6269 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6270 {
6271         struct extent_io_tree *tree;
6272         struct extent_map_tree *map;
6273         int ret;
6274
6275         tree = &BTRFS_I(page->mapping->host)->io_tree;
6276         map = &BTRFS_I(page->mapping->host)->extent_tree;
6277         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6278         if (ret == 1) {
6279                 ClearPagePrivate(page);
6280                 set_page_private(page, 0);
6281                 page_cache_release(page);
6282         }
6283         return ret;
6284 }
6285
6286 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6287 {
6288         if (PageWriteback(page) || PageDirty(page))
6289                 return 0;
6290         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6291 }
6292
6293 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6294 {
6295         struct extent_io_tree *tree;
6296         struct btrfs_ordered_extent *ordered;
6297         struct extent_state *cached_state = NULL;
6298         u64 page_start = page_offset(page);
6299         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6300
6301
6302         /*
6303          * we have the page locked, so new writeback can't start,
6304          * and the dirty bit won't be cleared while we are here.
6305          *
6306          * Wait for IO on this page so that we can safely clear
6307          * the PagePrivate2 bit and do ordered accounting
6308          */
6309         wait_on_page_writeback(page);
6310
6311         tree = &BTRFS_I(page->mapping->host)->io_tree;
6312         if (offset) {
6313                 btrfs_releasepage(page, GFP_NOFS);
6314                 return;
6315         }
6316         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6317                          GFP_NOFS);
6318         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6319                                            page_offset(page));
6320         if (ordered) {
6321                 /*
6322                  * IO on this page will never be started, so we need
6323                  * to account for any ordered extents now
6324                  */
6325                 clear_extent_bit(tree, page_start, page_end,
6326                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6327                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6328                                  &cached_state, GFP_NOFS);
6329                 /*
6330                  * whoever cleared the private bit is responsible
6331                  * for the finish_ordered_io
6332                  */
6333                 if (TestClearPagePrivate2(page)) {
6334                         btrfs_finish_ordered_io(page->mapping->host,
6335                                                 page_start, page_end);
6336                 }
6337                 btrfs_put_ordered_extent(ordered);
6338                 cached_state = NULL;
6339                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6340                                  GFP_NOFS);
6341         }
6342         clear_extent_bit(tree, page_start, page_end,
6343                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6344                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6345         __btrfs_releasepage(page, GFP_NOFS);
6346
6347         ClearPageChecked(page);
6348         if (PagePrivate(page)) {
6349                 ClearPagePrivate(page);
6350                 set_page_private(page, 0);
6351                 page_cache_release(page);
6352         }
6353 }
6354
6355 /*
6356  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6357  * called from a page fault handler when a page is first dirtied. Hence we must
6358  * be careful to check for EOF conditions here. We set the page up correctly
6359  * for a written page which means we get ENOSPC checking when writing into
6360  * holes and correct delalloc and unwritten extent mapping on filesystems that
6361  * support these features.
6362  *
6363  * We are not allowed to take the i_mutex here so we have to play games to
6364  * protect against truncate races as the page could now be beyond EOF.  Because
6365  * vmtruncate() writes the inode size before removing pages, once we have the
6366  * page lock we can determine safely if the page is beyond EOF. If it is not
6367  * beyond EOF, then the page is guaranteed safe against truncation until we
6368  * unlock the page.
6369  */
6370 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6371 {
6372         struct page *page = vmf->page;
6373         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6374         struct btrfs_root *root = BTRFS_I(inode)->root;
6375         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6376         struct btrfs_ordered_extent *ordered;
6377         struct extent_state *cached_state = NULL;
6378         char *kaddr;
6379         unsigned long zero_start;
6380         loff_t size;
6381         int ret;
6382         u64 page_start;
6383         u64 page_end;
6384
6385         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6386         if (ret) {
6387                 if (ret == -ENOMEM)
6388                         ret = VM_FAULT_OOM;
6389                 else /* -ENOSPC, -EIO, etc */
6390                         ret = VM_FAULT_SIGBUS;
6391                 goto out;
6392         }
6393
6394         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6395 again:
6396         lock_page(page);
6397         size = i_size_read(inode);
6398         page_start = page_offset(page);
6399         page_end = page_start + PAGE_CACHE_SIZE - 1;
6400
6401         if ((page->mapping != inode->i_mapping) ||
6402             (page_start >= size)) {
6403                 /* page got truncated out from underneath us */
6404                 goto out_unlock;
6405         }
6406         wait_on_page_writeback(page);
6407
6408         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6409                          GFP_NOFS);
6410         set_page_extent_mapped(page);
6411
6412         /*
6413          * we can't set the delalloc bits if there are pending ordered
6414          * extents.  Drop our locks and wait for them to finish
6415          */
6416         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6417         if (ordered) {
6418                 unlock_extent_cached(io_tree, page_start, page_end,
6419                                      &cached_state, GFP_NOFS);
6420                 unlock_page(page);
6421                 btrfs_start_ordered_extent(inode, ordered, 1);
6422                 btrfs_put_ordered_extent(ordered);
6423                 goto again;
6424         }
6425
6426         /*
6427          * XXX - page_mkwrite gets called every time the page is dirtied, even
6428          * if it was already dirty, so for space accounting reasons we need to
6429          * clear any delalloc bits for the range we are fixing to save.  There
6430          * is probably a better way to do this, but for now keep consistent with
6431          * prepare_pages in the normal write path.
6432          */
6433         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6434                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6435                           0, 0, &cached_state, GFP_NOFS);
6436
6437         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6438                                         &cached_state);
6439         if (ret) {
6440                 unlock_extent_cached(io_tree, page_start, page_end,
6441                                      &cached_state, GFP_NOFS);
6442                 ret = VM_FAULT_SIGBUS;
6443                 goto out_unlock;
6444         }
6445         ret = 0;
6446
6447         /* page is wholly or partially inside EOF */
6448         if (page_start + PAGE_CACHE_SIZE > size)
6449                 zero_start = size & ~PAGE_CACHE_MASK;
6450         else
6451                 zero_start = PAGE_CACHE_SIZE;
6452
6453         if (zero_start != PAGE_CACHE_SIZE) {
6454                 kaddr = kmap(page);
6455                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6456                 flush_dcache_page(page);
6457                 kunmap(page);
6458         }
6459         ClearPageChecked(page);
6460         set_page_dirty(page);
6461         SetPageUptodate(page);
6462
6463         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6464         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6465
6466         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6467
6468 out_unlock:
6469         if (!ret)
6470                 return VM_FAULT_LOCKED;
6471         unlock_page(page);
6472         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6473 out:
6474         return ret;
6475 }
6476
6477 static void btrfs_truncate(struct inode *inode)
6478 {
6479         struct btrfs_root *root = BTRFS_I(inode)->root;
6480         int ret;
6481         struct btrfs_trans_handle *trans;
6482         unsigned long nr;
6483         u64 mask = root->sectorsize - 1;
6484
6485         if (!S_ISREG(inode->i_mode)) {
6486                 WARN_ON(1);
6487                 return;
6488         }
6489
6490         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6491         if (ret)
6492                 return;
6493
6494         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6495         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6496
6497         trans = btrfs_start_transaction(root, 0);
6498         BUG_ON(IS_ERR(trans));
6499         btrfs_set_trans_block_group(trans, inode);
6500         trans->block_rsv = root->orphan_block_rsv;
6501
6502         /*
6503          * setattr is responsible for setting the ordered_data_close flag,
6504          * but that is only tested during the last file release.  That
6505          * could happen well after the next commit, leaving a great big
6506          * window where new writes may get lost if someone chooses to write
6507          * to this file after truncating to zero
6508          *
6509          * The inode doesn't have any dirty data here, and so if we commit
6510          * this is a noop.  If someone immediately starts writing to the inode
6511          * it is very likely we'll catch some of their writes in this
6512          * transaction, and the commit will find this file on the ordered
6513          * data list with good things to send down.
6514          *
6515          * This is a best effort solution, there is still a window where
6516          * using truncate to replace the contents of the file will
6517          * end up with a zero length file after a crash.
6518          */
6519         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6520                 btrfs_add_ordered_operation(trans, root, inode);
6521
6522         while (1) {
6523                 if (!trans) {
6524                         trans = btrfs_start_transaction(root, 0);
6525                         BUG_ON(IS_ERR(trans));
6526                         btrfs_set_trans_block_group(trans, inode);
6527                         trans->block_rsv = root->orphan_block_rsv;
6528                 }
6529
6530                 ret = btrfs_block_rsv_check(trans, root,
6531                                             root->orphan_block_rsv, 0, 5);
6532                 if (ret) {
6533                         BUG_ON(ret != -EAGAIN);
6534                         ret = btrfs_commit_transaction(trans, root);
6535                         BUG_ON(ret);
6536                         trans = NULL;
6537                         continue;
6538                 }
6539
6540                 ret = btrfs_truncate_inode_items(trans, root, inode,
6541                                                  inode->i_size,
6542                                                  BTRFS_EXTENT_DATA_KEY);
6543                 if (ret != -EAGAIN)
6544                         break;
6545
6546                 ret = btrfs_update_inode(trans, root, inode);
6547                 BUG_ON(ret);
6548
6549                 nr = trans->blocks_used;
6550                 btrfs_end_transaction(trans, root);
6551                 trans = NULL;
6552                 btrfs_btree_balance_dirty(root, nr);
6553         }
6554
6555         if (ret == 0 && inode->i_nlink > 0) {
6556                 ret = btrfs_orphan_del(trans, inode);
6557                 BUG_ON(ret);
6558         }
6559
6560         ret = btrfs_update_inode(trans, root, inode);
6561         BUG_ON(ret);
6562
6563         nr = trans->blocks_used;
6564         ret = btrfs_end_transaction_throttle(trans, root);
6565         BUG_ON(ret);
6566         btrfs_btree_balance_dirty(root, nr);
6567 }
6568
6569 /*
6570  * create a new subvolume directory/inode (helper for the ioctl).
6571  */
6572 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6573                              struct btrfs_root *new_root,
6574                              u64 new_dirid, u64 alloc_hint)
6575 {
6576         struct inode *inode;
6577         int err;
6578         u64 index = 0;
6579
6580         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6581                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6582         if (IS_ERR(inode))
6583                 return PTR_ERR(inode);
6584         inode->i_op = &btrfs_dir_inode_operations;
6585         inode->i_fop = &btrfs_dir_file_operations;
6586
6587         inode->i_nlink = 1;
6588         btrfs_i_size_write(inode, 0);
6589
6590         err = btrfs_update_inode(trans, new_root, inode);
6591         BUG_ON(err);
6592
6593         iput(inode);
6594         return 0;
6595 }
6596
6597 /* helper function for file defrag and space balancing.  This
6598  * forces readahead on a given range of bytes in an inode
6599  */
6600 unsigned long btrfs_force_ra(struct address_space *mapping,
6601                               struct file_ra_state *ra, struct file *file,
6602                               pgoff_t offset, pgoff_t last_index)
6603 {
6604         pgoff_t req_size = last_index - offset + 1;
6605
6606         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6607         return offset + req_size;
6608 }
6609
6610 struct inode *btrfs_alloc_inode(struct super_block *sb)
6611 {
6612         struct btrfs_inode *ei;
6613         struct inode *inode;
6614
6615         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6616         if (!ei)
6617                 return NULL;
6618
6619         ei->root = NULL;
6620         ei->space_info = NULL;
6621         ei->generation = 0;
6622         ei->sequence = 0;
6623         ei->last_trans = 0;
6624         ei->last_sub_trans = 0;
6625         ei->logged_trans = 0;
6626         ei->delalloc_bytes = 0;
6627         ei->reserved_bytes = 0;
6628         ei->disk_i_size = 0;
6629         ei->flags = 0;
6630         ei->index_cnt = (u64)-1;
6631         ei->last_unlink_trans = 0;
6632
6633         spin_lock_init(&ei->accounting_lock);
6634         atomic_set(&ei->outstanding_extents, 0);
6635         ei->reserved_extents = 0;
6636
6637         ei->ordered_data_close = 0;
6638         ei->orphan_meta_reserved = 0;
6639         ei->dummy_inode = 0;
6640         ei->force_compress = BTRFS_COMPRESS_NONE;
6641
6642         inode = &ei->vfs_inode;
6643         extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6644         extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6645         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6646         mutex_init(&ei->log_mutex);
6647         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6648         INIT_LIST_HEAD(&ei->i_orphan);
6649         INIT_LIST_HEAD(&ei->delalloc_inodes);
6650         INIT_LIST_HEAD(&ei->ordered_operations);
6651         RB_CLEAR_NODE(&ei->rb_node);
6652
6653         return inode;
6654 }
6655
6656 static void btrfs_i_callback(struct rcu_head *head)
6657 {
6658         struct inode *inode = container_of(head, struct inode, i_rcu);
6659         INIT_LIST_HEAD(&inode->i_dentry);
6660         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6661 }
6662
6663 void btrfs_destroy_inode(struct inode *inode)
6664 {
6665         struct btrfs_ordered_extent *ordered;
6666         struct btrfs_root *root = BTRFS_I(inode)->root;
6667
6668         WARN_ON(!list_empty(&inode->i_dentry));
6669         WARN_ON(inode->i_data.nrpages);
6670         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6671         WARN_ON(BTRFS_I(inode)->reserved_extents);
6672
6673         /*
6674          * This can happen where we create an inode, but somebody else also
6675          * created the same inode and we need to destroy the one we already
6676          * created.
6677          */
6678         if (!root)
6679                 goto free;
6680
6681         /*
6682          * Make sure we're properly removed from the ordered operation
6683          * lists.
6684          */
6685         smp_mb();
6686         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6687                 spin_lock(&root->fs_info->ordered_extent_lock);
6688                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6689                 spin_unlock(&root->fs_info->ordered_extent_lock);
6690         }
6691
6692         if (root == root->fs_info->tree_root) {
6693                 struct btrfs_block_group_cache *block_group;
6694
6695                 block_group = btrfs_lookup_block_group(root->fs_info,
6696                                                 BTRFS_I(inode)->block_group);
6697                 if (block_group && block_group->inode == inode) {
6698                         spin_lock(&block_group->lock);
6699                         block_group->inode = NULL;
6700                         spin_unlock(&block_group->lock);
6701                         btrfs_put_block_group(block_group);
6702                 } else if (block_group) {
6703                         btrfs_put_block_group(block_group);
6704                 }
6705         }
6706
6707         spin_lock(&root->orphan_lock);
6708         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6709                 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6710                        inode->i_ino);
6711                 list_del_init(&BTRFS_I(inode)->i_orphan);
6712         }
6713         spin_unlock(&root->orphan_lock);
6714
6715         while (1) {
6716                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6717                 if (!ordered)
6718                         break;
6719                 else {
6720                         printk(KERN_ERR "btrfs found ordered "
6721                                "extent %llu %llu on inode cleanup\n",
6722                                (unsigned long long)ordered->file_offset,
6723                                (unsigned long long)ordered->len);
6724                         btrfs_remove_ordered_extent(inode, ordered);
6725                         btrfs_put_ordered_extent(ordered);
6726                         btrfs_put_ordered_extent(ordered);
6727                 }
6728         }
6729         inode_tree_del(inode);
6730         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6731 free:
6732         call_rcu(&inode->i_rcu, btrfs_i_callback);
6733 }
6734
6735 int btrfs_drop_inode(struct inode *inode)
6736 {
6737         struct btrfs_root *root = BTRFS_I(inode)->root;
6738
6739         if (btrfs_root_refs(&root->root_item) == 0 &&
6740             root != root->fs_info->tree_root)
6741                 return 1;
6742         else
6743                 return generic_drop_inode(inode);
6744 }
6745
6746 static void init_once(void *foo)
6747 {
6748         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6749
6750         inode_init_once(&ei->vfs_inode);
6751 }
6752
6753 void btrfs_destroy_cachep(void)
6754 {
6755         if (btrfs_inode_cachep)
6756                 kmem_cache_destroy(btrfs_inode_cachep);
6757         if (btrfs_trans_handle_cachep)
6758                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6759         if (btrfs_transaction_cachep)
6760                 kmem_cache_destroy(btrfs_transaction_cachep);
6761         if (btrfs_path_cachep)
6762                 kmem_cache_destroy(btrfs_path_cachep);
6763 }
6764
6765 int btrfs_init_cachep(void)
6766 {
6767         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6768                         sizeof(struct btrfs_inode), 0,
6769                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6770         if (!btrfs_inode_cachep)
6771                 goto fail;
6772
6773         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6774                         sizeof(struct btrfs_trans_handle), 0,
6775                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6776         if (!btrfs_trans_handle_cachep)
6777                 goto fail;
6778
6779         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6780                         sizeof(struct btrfs_transaction), 0,
6781                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6782         if (!btrfs_transaction_cachep)
6783                 goto fail;
6784
6785         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6786                         sizeof(struct btrfs_path), 0,
6787                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6788         if (!btrfs_path_cachep)
6789                 goto fail;
6790
6791         return 0;
6792 fail:
6793         btrfs_destroy_cachep();
6794         return -ENOMEM;
6795 }
6796
6797 static int btrfs_getattr(struct vfsmount *mnt,
6798                          struct dentry *dentry, struct kstat *stat)
6799 {
6800         struct inode *inode = dentry->d_inode;
6801         generic_fillattr(inode, stat);
6802         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6803         stat->blksize = PAGE_CACHE_SIZE;
6804         stat->blocks = (inode_get_bytes(inode) +
6805                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6806         return 0;
6807 }
6808
6809 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6810                            struct inode *new_dir, struct dentry *new_dentry)
6811 {
6812         struct btrfs_trans_handle *trans;
6813         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6814         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6815         struct inode *new_inode = new_dentry->d_inode;
6816         struct inode *old_inode = old_dentry->d_inode;
6817         struct timespec ctime = CURRENT_TIME;
6818         u64 index = 0;
6819         u64 root_objectid;
6820         int ret;
6821
6822         if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6823                 return -EPERM;
6824
6825         /* we only allow rename subvolume link between subvolumes */
6826         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6827                 return -EXDEV;
6828
6829         if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6830             (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
6831                 return -ENOTEMPTY;
6832
6833         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6834             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6835                 return -ENOTEMPTY;
6836         /*
6837          * we're using rename to replace one file with another.
6838          * and the replacement file is large.  Start IO on it now so
6839          * we don't add too much work to the end of the transaction
6840          */
6841         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6842             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6843                 filemap_flush(old_inode->i_mapping);
6844
6845         /* close the racy window with snapshot create/destroy ioctl */
6846         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6847                 down_read(&root->fs_info->subvol_sem);
6848         /*
6849          * We want to reserve the absolute worst case amount of items.  So if
6850          * both inodes are subvols and we need to unlink them then that would
6851          * require 4 item modifications, but if they are both normal inodes it
6852          * would require 5 item modifications, so we'll assume their normal
6853          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6854          * should cover the worst case number of items we'll modify.
6855          */
6856         trans = btrfs_start_transaction(root, 20);
6857         if (IS_ERR(trans))
6858                 return PTR_ERR(trans);
6859
6860         btrfs_set_trans_block_group(trans, new_dir);
6861
6862         if (dest != root)
6863                 btrfs_record_root_in_trans(trans, dest);
6864
6865         ret = btrfs_set_inode_index(new_dir, &index);
6866         if (ret)
6867                 goto out_fail;
6868
6869         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6870                 /* force full log commit if subvolume involved. */
6871                 root->fs_info->last_trans_log_full_commit = trans->transid;
6872         } else {
6873                 ret = btrfs_insert_inode_ref(trans, dest,
6874                                              new_dentry->d_name.name,
6875                                              new_dentry->d_name.len,
6876                                              old_inode->i_ino,
6877                                              new_dir->i_ino, index);
6878                 if (ret)
6879                         goto out_fail;
6880                 /*
6881                  * this is an ugly little race, but the rename is required
6882                  * to make sure that if we crash, the inode is either at the
6883                  * old name or the new one.  pinning the log transaction lets
6884                  * us make sure we don't allow a log commit to come in after
6885                  * we unlink the name but before we add the new name back in.
6886                  */
6887                 btrfs_pin_log_trans(root);
6888         }
6889         /*
6890          * make sure the inode gets flushed if it is replacing
6891          * something.
6892          */
6893         if (new_inode && new_inode->i_size &&
6894             old_inode && S_ISREG(old_inode->i_mode)) {
6895                 btrfs_add_ordered_operation(trans, root, old_inode);
6896         }
6897
6898         old_dir->i_ctime = old_dir->i_mtime = ctime;
6899         new_dir->i_ctime = new_dir->i_mtime = ctime;
6900         old_inode->i_ctime = ctime;
6901
6902         if (old_dentry->d_parent != new_dentry->d_parent)
6903                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
6904
6905         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6906                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
6907                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
6908                                         old_dentry->d_name.name,
6909                                         old_dentry->d_name.len);
6910         } else {
6911                 btrfs_inc_nlink(old_dentry->d_inode);
6912                 ret = btrfs_unlink_inode(trans, root, old_dir,
6913                                          old_dentry->d_inode,
6914                                          old_dentry->d_name.name,
6915                                          old_dentry->d_name.len);
6916         }
6917         BUG_ON(ret);
6918
6919         if (new_inode) {
6920                 new_inode->i_ctime = CURRENT_TIME;
6921                 if (unlikely(new_inode->i_ino ==
6922                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
6923                         root_objectid = BTRFS_I(new_inode)->location.objectid;
6924                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
6925                                                 root_objectid,
6926                                                 new_dentry->d_name.name,
6927                                                 new_dentry->d_name.len);
6928                         BUG_ON(new_inode->i_nlink == 0);
6929                 } else {
6930                         ret = btrfs_unlink_inode(trans, dest, new_dir,
6931                                                  new_dentry->d_inode,
6932                                                  new_dentry->d_name.name,
6933                                                  new_dentry->d_name.len);
6934                 }
6935                 BUG_ON(ret);
6936                 if (new_inode->i_nlink == 0) {
6937                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
6938                         BUG_ON(ret);
6939                 }
6940         }
6941
6942         ret = btrfs_add_link(trans, new_dir, old_inode,
6943                              new_dentry->d_name.name,
6944                              new_dentry->d_name.len, 0, index);
6945         BUG_ON(ret);
6946
6947         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
6948                 struct dentry *parent = dget_parent(new_dentry);
6949                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
6950                 dput(parent);
6951                 btrfs_end_log_trans(root);
6952         }
6953 out_fail:
6954         btrfs_end_transaction_throttle(trans, root);
6955
6956         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6957                 up_read(&root->fs_info->subvol_sem);
6958
6959         return ret;
6960 }
6961
6962 /*
6963  * some fairly slow code that needs optimization. This walks the list
6964  * of all the inodes with pending delalloc and forces them to disk.
6965  */
6966 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
6967 {
6968         struct list_head *head = &root->fs_info->delalloc_inodes;
6969         struct btrfs_inode *binode;
6970         struct inode *inode;
6971
6972         if (root->fs_info->sb->s_flags & MS_RDONLY)
6973                 return -EROFS;
6974
6975         spin_lock(&root->fs_info->delalloc_lock);
6976         while (!list_empty(head)) {
6977                 binode = list_entry(head->next, struct btrfs_inode,
6978                                     delalloc_inodes);
6979                 inode = igrab(&binode->vfs_inode);
6980                 if (!inode)
6981                         list_del_init(&binode->delalloc_inodes);
6982                 spin_unlock(&root->fs_info->delalloc_lock);
6983                 if (inode) {
6984                         filemap_flush(inode->i_mapping);
6985                         if (delay_iput)
6986                                 btrfs_add_delayed_iput(inode);
6987                         else
6988                                 iput(inode);
6989                 }
6990                 cond_resched();
6991                 spin_lock(&root->fs_info->delalloc_lock);
6992         }
6993         spin_unlock(&root->fs_info->delalloc_lock);
6994
6995         /* the filemap_flush will queue IO into the worker threads, but
6996          * we have to make sure the IO is actually started and that
6997          * ordered extents get created before we return
6998          */
6999         atomic_inc(&root->fs_info->async_submit_draining);
7000         while (atomic_read(&root->fs_info->nr_async_submits) ||
7001               atomic_read(&root->fs_info->async_delalloc_pages)) {
7002                 wait_event(root->fs_info->async_submit_wait,
7003                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7004                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7005         }
7006         atomic_dec(&root->fs_info->async_submit_draining);
7007         return 0;
7008 }
7009
7010 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput,
7011                                    int sync)
7012 {
7013         struct btrfs_inode *binode;
7014         struct inode *inode = NULL;
7015
7016         spin_lock(&root->fs_info->delalloc_lock);
7017         while (!list_empty(&root->fs_info->delalloc_inodes)) {
7018                 binode = list_entry(root->fs_info->delalloc_inodes.next,
7019                                     struct btrfs_inode, delalloc_inodes);
7020                 inode = igrab(&binode->vfs_inode);
7021                 if (inode) {
7022                         list_move_tail(&binode->delalloc_inodes,
7023                                        &root->fs_info->delalloc_inodes);
7024                         break;
7025                 }
7026
7027                 list_del_init(&binode->delalloc_inodes);
7028                 cond_resched_lock(&root->fs_info->delalloc_lock);
7029         }
7030         spin_unlock(&root->fs_info->delalloc_lock);
7031
7032         if (inode) {
7033                 if (sync) {
7034                         filemap_write_and_wait(inode->i_mapping);
7035                         /*
7036                          * We have to do this because compression doesn't
7037                          * actually set PG_writeback until it submits the pages
7038                          * for IO, which happens in an async thread, so we could
7039                          * race and not actually wait for any writeback pages
7040                          * because they've not been submitted yet.  Technically
7041                          * this could still be the case for the ordered stuff
7042                          * since the async thread may not have started to do its
7043                          * work yet.  If this becomes the case then we need to
7044                          * figure out a way to make sure that in writepage we
7045                          * wait for any async pages to be submitted before
7046                          * returning so that fdatawait does what its supposed to
7047                          * do.
7048                          */
7049                         btrfs_wait_ordered_range(inode, 0, (u64)-1);
7050                 } else {
7051                         filemap_flush(inode->i_mapping);
7052                 }
7053                 if (delay_iput)
7054                         btrfs_add_delayed_iput(inode);
7055                 else
7056                         iput(inode);
7057                 return 1;
7058         }
7059         return 0;
7060 }
7061
7062 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7063                          const char *symname)
7064 {
7065         struct btrfs_trans_handle *trans;
7066         struct btrfs_root *root = BTRFS_I(dir)->root;
7067         struct btrfs_path *path;
7068         struct btrfs_key key;
7069         struct inode *inode = NULL;
7070         int err;
7071         int drop_inode = 0;
7072         u64 objectid;
7073         u64 index = 0 ;
7074         int name_len;
7075         int datasize;
7076         unsigned long ptr;
7077         struct btrfs_file_extent_item *ei;
7078         struct extent_buffer *leaf;
7079         unsigned long nr = 0;
7080
7081         name_len = strlen(symname) + 1;
7082         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7083                 return -ENAMETOOLONG;
7084
7085         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
7086         if (err)
7087                 return err;
7088         /*
7089          * 2 items for inode item and ref
7090          * 2 items for dir items
7091          * 1 item for xattr if selinux is on
7092          */
7093         trans = btrfs_start_transaction(root, 5);
7094         if (IS_ERR(trans))
7095                 return PTR_ERR(trans);
7096
7097         btrfs_set_trans_block_group(trans, dir);
7098
7099         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7100                                 dentry->d_name.len, dir->i_ino, objectid,
7101                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
7102                                 &index);
7103         err = PTR_ERR(inode);
7104         if (IS_ERR(inode))
7105                 goto out_unlock;
7106
7107         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7108         if (err) {
7109                 drop_inode = 1;
7110                 goto out_unlock;
7111         }
7112
7113         btrfs_set_trans_block_group(trans, inode);
7114         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7115         if (err)
7116                 drop_inode = 1;
7117         else {
7118                 inode->i_mapping->a_ops = &btrfs_aops;
7119                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7120                 inode->i_fop = &btrfs_file_operations;
7121                 inode->i_op = &btrfs_file_inode_operations;
7122                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7123         }
7124         btrfs_update_inode_block_group(trans, inode);
7125         btrfs_update_inode_block_group(trans, dir);
7126         if (drop_inode)
7127                 goto out_unlock;
7128
7129         path = btrfs_alloc_path();
7130         BUG_ON(!path);
7131         key.objectid = inode->i_ino;
7132         key.offset = 0;
7133         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7134         datasize = btrfs_file_extent_calc_inline_size(name_len);
7135         err = btrfs_insert_empty_item(trans, root, path, &key,
7136                                       datasize);
7137         if (err) {
7138                 drop_inode = 1;
7139                 goto out_unlock;
7140         }
7141         leaf = path->nodes[0];
7142         ei = btrfs_item_ptr(leaf, path->slots[0],
7143                             struct btrfs_file_extent_item);
7144         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7145         btrfs_set_file_extent_type(leaf, ei,
7146                                    BTRFS_FILE_EXTENT_INLINE);
7147         btrfs_set_file_extent_encryption(leaf, ei, 0);
7148         btrfs_set_file_extent_compression(leaf, ei, 0);
7149         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7150         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7151
7152         ptr = btrfs_file_extent_inline_start(ei);
7153         write_extent_buffer(leaf, symname, ptr, name_len);
7154         btrfs_mark_buffer_dirty(leaf);
7155         btrfs_free_path(path);
7156
7157         inode->i_op = &btrfs_symlink_inode_operations;
7158         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7159         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7160         inode_set_bytes(inode, name_len);
7161         btrfs_i_size_write(inode, name_len - 1);
7162         err = btrfs_update_inode(trans, root, inode);
7163         if (err)
7164                 drop_inode = 1;
7165
7166 out_unlock:
7167         nr = trans->blocks_used;
7168         btrfs_end_transaction_throttle(trans, root);
7169         if (drop_inode) {
7170                 inode_dec_link_count(inode);
7171                 iput(inode);
7172         }
7173         btrfs_btree_balance_dirty(root, nr);
7174         return err;
7175 }
7176
7177 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7178                                        u64 start, u64 num_bytes, u64 min_size,
7179                                        loff_t actual_len, u64 *alloc_hint,
7180                                        struct btrfs_trans_handle *trans)
7181 {
7182         struct btrfs_root *root = BTRFS_I(inode)->root;
7183         struct btrfs_key ins;
7184         u64 cur_offset = start;
7185         u64 i_size;
7186         int ret = 0;
7187         bool own_trans = true;
7188
7189         if (trans)
7190                 own_trans = false;
7191         while (num_bytes > 0) {
7192                 if (own_trans) {
7193                         trans = btrfs_start_transaction(root, 3);
7194                         if (IS_ERR(trans)) {
7195                                 ret = PTR_ERR(trans);
7196                                 break;
7197                         }
7198                 }
7199
7200                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7201                                            0, *alloc_hint, (u64)-1, &ins, 1);
7202                 if (ret) {
7203                         if (own_trans)
7204                                 btrfs_end_transaction(trans, root);
7205                         break;
7206                 }
7207
7208                 ret = insert_reserved_file_extent(trans, inode,
7209                                                   cur_offset, ins.objectid,
7210                                                   ins.offset, ins.offset,
7211                                                   ins.offset, 0, 0, 0,
7212                                                   BTRFS_FILE_EXTENT_PREALLOC);
7213                 BUG_ON(ret);
7214                 btrfs_drop_extent_cache(inode, cur_offset,
7215                                         cur_offset + ins.offset -1, 0);
7216
7217                 num_bytes -= ins.offset;
7218                 cur_offset += ins.offset;
7219                 *alloc_hint = ins.objectid + ins.offset;
7220
7221                 inode->i_ctime = CURRENT_TIME;
7222                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7223                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7224                     (actual_len > inode->i_size) &&
7225                     (cur_offset > inode->i_size)) {
7226                         if (cur_offset > actual_len)
7227                                 i_size = actual_len;
7228                         else
7229                                 i_size = cur_offset;
7230                         i_size_write(inode, i_size);
7231                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7232                 }
7233
7234                 ret = btrfs_update_inode(trans, root, inode);
7235                 BUG_ON(ret);
7236
7237                 if (own_trans)
7238                         btrfs_end_transaction(trans, root);
7239         }
7240         return ret;
7241 }
7242
7243 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7244                               u64 start, u64 num_bytes, u64 min_size,
7245                               loff_t actual_len, u64 *alloc_hint)
7246 {
7247         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7248                                            min_size, actual_len, alloc_hint,
7249                                            NULL);
7250 }
7251
7252 int btrfs_prealloc_file_range_trans(struct inode *inode,
7253                                     struct btrfs_trans_handle *trans, int mode,
7254                                     u64 start, u64 num_bytes, u64 min_size,
7255                                     loff_t actual_len, u64 *alloc_hint)
7256 {
7257         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7258                                            min_size, actual_len, alloc_hint, trans);
7259 }
7260
7261 static int btrfs_set_page_dirty(struct page *page)
7262 {
7263         return __set_page_dirty_nobuffers(page);
7264 }
7265
7266 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7267 {
7268         struct btrfs_root *root = BTRFS_I(inode)->root;
7269
7270         if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7271                 return -EROFS;
7272         if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7273                 return -EACCES;
7274         return generic_permission(inode, mask, flags, btrfs_check_acl);
7275 }
7276
7277 static const struct inode_operations btrfs_dir_inode_operations = {
7278         .getattr        = btrfs_getattr,
7279         .lookup         = btrfs_lookup,
7280         .create         = btrfs_create,
7281         .unlink         = btrfs_unlink,
7282         .link           = btrfs_link,
7283         .mkdir          = btrfs_mkdir,
7284         .rmdir          = btrfs_rmdir,
7285         .rename         = btrfs_rename,
7286         .symlink        = btrfs_symlink,
7287         .setattr        = btrfs_setattr,
7288         .mknod          = btrfs_mknod,
7289         .setxattr       = btrfs_setxattr,
7290         .getxattr       = btrfs_getxattr,
7291         .listxattr      = btrfs_listxattr,
7292         .removexattr    = btrfs_removexattr,
7293         .permission     = btrfs_permission,
7294 };
7295 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7296         .lookup         = btrfs_lookup,
7297         .permission     = btrfs_permission,
7298 };
7299
7300 static const struct file_operations btrfs_dir_file_operations = {
7301         .llseek         = generic_file_llseek,
7302         .read           = generic_read_dir,
7303         .readdir        = btrfs_real_readdir,
7304         .unlocked_ioctl = btrfs_ioctl,
7305 #ifdef CONFIG_COMPAT
7306         .compat_ioctl   = btrfs_ioctl,
7307 #endif
7308         .release        = btrfs_release_file,
7309         .fsync          = btrfs_sync_file,
7310 };
7311
7312 static struct extent_io_ops btrfs_extent_io_ops = {
7313         .fill_delalloc = run_delalloc_range,
7314         .submit_bio_hook = btrfs_submit_bio_hook,
7315         .merge_bio_hook = btrfs_merge_bio_hook,
7316         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7317         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7318         .writepage_start_hook = btrfs_writepage_start_hook,
7319         .readpage_io_failed_hook = btrfs_io_failed_hook,
7320         .set_bit_hook = btrfs_set_bit_hook,
7321         .clear_bit_hook = btrfs_clear_bit_hook,
7322         .merge_extent_hook = btrfs_merge_extent_hook,
7323         .split_extent_hook = btrfs_split_extent_hook,
7324 };
7325
7326 /*
7327  * btrfs doesn't support the bmap operation because swapfiles
7328  * use bmap to make a mapping of extents in the file.  They assume
7329  * these extents won't change over the life of the file and they
7330  * use the bmap result to do IO directly to the drive.
7331  *
7332  * the btrfs bmap call would return logical addresses that aren't
7333  * suitable for IO and they also will change frequently as COW
7334  * operations happen.  So, swapfile + btrfs == corruption.
7335  *
7336  * For now we're avoiding this by dropping bmap.
7337  */
7338 static const struct address_space_operations btrfs_aops = {
7339         .readpage       = btrfs_readpage,
7340         .writepage      = btrfs_writepage,
7341         .writepages     = btrfs_writepages,
7342         .readpages      = btrfs_readpages,
7343         .direct_IO      = btrfs_direct_IO,
7344         .invalidatepage = btrfs_invalidatepage,
7345         .releasepage    = btrfs_releasepage,
7346         .set_page_dirty = btrfs_set_page_dirty,
7347         .error_remove_page = generic_error_remove_page,
7348 };
7349
7350 static const struct address_space_operations btrfs_symlink_aops = {
7351         .readpage       = btrfs_readpage,
7352         .writepage      = btrfs_writepage,
7353         .invalidatepage = btrfs_invalidatepage,
7354         .releasepage    = btrfs_releasepage,
7355 };
7356
7357 static const struct inode_operations btrfs_file_inode_operations = {
7358         .truncate       = btrfs_truncate,
7359         .getattr        = btrfs_getattr,
7360         .setattr        = btrfs_setattr,
7361         .setxattr       = btrfs_setxattr,
7362         .getxattr       = btrfs_getxattr,
7363         .listxattr      = btrfs_listxattr,
7364         .removexattr    = btrfs_removexattr,
7365         .permission     = btrfs_permission,
7366         .fiemap         = btrfs_fiemap,
7367 };
7368 static const struct inode_operations btrfs_special_inode_operations = {
7369         .getattr        = btrfs_getattr,
7370         .setattr        = btrfs_setattr,
7371         .permission     = btrfs_permission,
7372         .setxattr       = btrfs_setxattr,
7373         .getxattr       = btrfs_getxattr,
7374         .listxattr      = btrfs_listxattr,
7375         .removexattr    = btrfs_removexattr,
7376 };
7377 static const struct inode_operations btrfs_symlink_inode_operations = {
7378         .readlink       = generic_readlink,
7379         .follow_link    = page_follow_link_light,
7380         .put_link       = page_put_link,
7381         .getattr        = btrfs_getattr,
7382         .permission     = btrfs_permission,
7383         .setxattr       = btrfs_setxattr,
7384         .getxattr       = btrfs_getxattr,
7385         .listxattr      = btrfs_listxattr,
7386         .removexattr    = btrfs_removexattr,
7387 };
7388
7389 const struct dentry_operations btrfs_dentry_operations = {
7390         .d_delete       = btrfs_dentry_delete,
7391 };