Btrfs: fix fallocate deadlock on inode extent lock
[linux-3.10.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/smp_lock.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mpage.h>
32 #include <linux/swap.h>
33 #include <linux/writeback.h>
34 #include <linux/statfs.h>
35 #include <linux/compat.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/xattr.h>
38 #include <linux/posix_acl.h>
39 #include <linux/falloc.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 "ref-cache.h"
52 #include "compression.h"
53 #include "locking.h"
54
55 struct btrfs_iget_args {
56         u64 ino;
57         struct btrfs_root *root;
58 };
59
60 static struct inode_operations btrfs_dir_inode_operations;
61 static struct inode_operations btrfs_symlink_inode_operations;
62 static struct inode_operations btrfs_dir_ro_inode_operations;
63 static struct inode_operations btrfs_special_inode_operations;
64 static struct inode_operations btrfs_file_inode_operations;
65 static struct address_space_operations btrfs_aops;
66 static struct address_space_operations btrfs_symlink_aops;
67 static struct file_operations btrfs_dir_file_operations;
68 static struct extent_io_ops btrfs_extent_io_ops;
69
70 static struct kmem_cache *btrfs_inode_cachep;
71 struct kmem_cache *btrfs_trans_handle_cachep;
72 struct kmem_cache *btrfs_transaction_cachep;
73 struct kmem_cache *btrfs_bit_radix_cachep;
74 struct kmem_cache *btrfs_path_cachep;
75
76 #define S_SHIFT 12
77 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
78         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
79         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
80         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
81         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
82         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
83         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
84         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
85 };
86
87 static void btrfs_truncate(struct inode *inode);
88 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
89 static noinline int cow_file_range(struct inode *inode,
90                                    struct page *locked_page,
91                                    u64 start, u64 end, int *page_started,
92                                    unsigned long *nr_written, int unlock);
93
94 static int btrfs_init_inode_security(struct inode *inode,  struct inode *dir)
95 {
96         int err;
97
98         err = btrfs_init_acl(inode, dir);
99         if (!err)
100                 err = btrfs_xattr_security_init(inode, dir);
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 use_compress = 0;
127
128         if (compressed_size && compressed_pages) {
129                 use_compress = 1;
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 (use_compress) {
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                                                   BTRFS_COMPRESS_ZLIB);
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         BTRFS_I(inode)->disk_i_size = inode->i_size;
195         btrfs_update_inode(trans, root, inode);
196         return 0;
197 fail:
198         btrfs_free_path(path);
199         return err;
200 }
201
202
203 /*
204  * conditionally insert an inline extent into the file.  This
205  * does the checks required to make sure the data is small enough
206  * to fit as an inline extent.
207  */
208 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
209                                  struct btrfs_root *root,
210                                  struct inode *inode, u64 start, u64 end,
211                                  size_t compressed_size,
212                                  struct page **compressed_pages)
213 {
214         u64 isize = i_size_read(inode);
215         u64 actual_end = min(end + 1, isize);
216         u64 inline_len = actual_end - start;
217         u64 aligned_end = (end + root->sectorsize - 1) &
218                         ~((u64)root->sectorsize - 1);
219         u64 hint_byte;
220         u64 data_len = inline_len;
221         int ret;
222
223         if (compressed_size)
224                 data_len = compressed_size;
225
226         if (start > 0 ||
227             actual_end >= PAGE_CACHE_SIZE ||
228             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
229             (!compressed_size &&
230             (actual_end & (root->sectorsize - 1)) == 0) ||
231             end + 1 < isize ||
232             data_len > root->fs_info->max_inline) {
233                 return 1;
234         }
235
236         ret = btrfs_drop_extents(trans, root, inode, start,
237                                  aligned_end, aligned_end, start, &hint_byte);
238         BUG_ON(ret);
239
240         if (isize > actual_end)
241                 inline_len = min_t(u64, isize, actual_end);
242         ret = insert_inline_extent(trans, root, inode, start,
243                                    inline_len, compressed_size,
244                                    compressed_pages);
245         BUG_ON(ret);
246         btrfs_drop_extent_cache(inode, start, aligned_end, 0);
247         return 0;
248 }
249
250 struct async_extent {
251         u64 start;
252         u64 ram_size;
253         u64 compressed_size;
254         struct page **pages;
255         unsigned long nr_pages;
256         struct list_head list;
257 };
258
259 struct async_cow {
260         struct inode *inode;
261         struct btrfs_root *root;
262         struct page *locked_page;
263         u64 start;
264         u64 end;
265         struct list_head extents;
266         struct btrfs_work work;
267 };
268
269 static noinline int add_async_extent(struct async_cow *cow,
270                                      u64 start, u64 ram_size,
271                                      u64 compressed_size,
272                                      struct page **pages,
273                                      unsigned long nr_pages)
274 {
275         struct async_extent *async_extent;
276
277         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
278         async_extent->start = start;
279         async_extent->ram_size = ram_size;
280         async_extent->compressed_size = compressed_size;
281         async_extent->pages = pages;
282         async_extent->nr_pages = nr_pages;
283         list_add_tail(&async_extent->list, &cow->extents);
284         return 0;
285 }
286
287 /*
288  * we create compressed extents in two phases.  The first
289  * phase compresses a range of pages that have already been
290  * locked (both pages and state bits are locked).
291  *
292  * This is done inside an ordered work queue, and the compression
293  * is spread across many cpus.  The actual IO submission is step
294  * two, and the ordered work queue takes care of making sure that
295  * happens in the same order things were put onto the queue by
296  * writepages and friends.
297  *
298  * If this code finds it can't get good compression, it puts an
299  * entry onto the work queue to write the uncompressed bytes.  This
300  * makes sure that both compressed inodes and uncompressed inodes
301  * are written in the same order that pdflush sent them down.
302  */
303 static noinline int compress_file_range(struct inode *inode,
304                                         struct page *locked_page,
305                                         u64 start, u64 end,
306                                         struct async_cow *async_cow,
307                                         int *num_added)
308 {
309         struct btrfs_root *root = BTRFS_I(inode)->root;
310         struct btrfs_trans_handle *trans;
311         u64 num_bytes;
312         u64 orig_start;
313         u64 disk_num_bytes;
314         u64 blocksize = root->sectorsize;
315         u64 actual_end;
316         u64 isize = i_size_read(inode);
317         int ret = 0;
318         struct page **pages = NULL;
319         unsigned long nr_pages;
320         unsigned long nr_pages_ret = 0;
321         unsigned long total_compressed = 0;
322         unsigned long total_in = 0;
323         unsigned long max_compressed = 128 * 1024;
324         unsigned long max_uncompressed = 128 * 1024;
325         int i;
326         int will_compress;
327
328         orig_start = start;
329
330         actual_end = min_t(u64, isize, end + 1);
331 again:
332         will_compress = 0;
333         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
334         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
335
336         /*
337          * we don't want to send crud past the end of i_size through
338          * compression, that's just a waste of CPU time.  So, if the
339          * end of the file is before the start of our current
340          * requested range of bytes, we bail out to the uncompressed
341          * cleanup code that can deal with all of this.
342          *
343          * It isn't really the fastest way to fix things, but this is a
344          * very uncommon corner.
345          */
346         if (actual_end <= start)
347                 goto cleanup_and_bail_uncompressed;
348
349         total_compressed = actual_end - start;
350
351         /* we want to make sure that amount of ram required to uncompress
352          * an extent is reasonable, so we limit the total size in ram
353          * of a compressed extent to 128k.  This is a crucial number
354          * because it also controls how easily we can spread reads across
355          * cpus for decompression.
356          *
357          * We also want to make sure the amount of IO required to do
358          * a random read is reasonably small, so we limit the size of
359          * a compressed extent to 128k.
360          */
361         total_compressed = min(total_compressed, max_uncompressed);
362         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
363         num_bytes = max(blocksize,  num_bytes);
364         disk_num_bytes = num_bytes;
365         total_in = 0;
366         ret = 0;
367
368         /*
369          * we do compression for mount -o compress and when the
370          * inode has not been flagged as nocompress.  This flag can
371          * change at any time if we discover bad compression ratios.
372          */
373         if (!btrfs_test_flag(inode, NOCOMPRESS) &&
374             btrfs_test_opt(root, COMPRESS)) {
375                 WARN_ON(pages);
376                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
377
378                 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
379                                                 total_compressed, pages,
380                                                 nr_pages, &nr_pages_ret,
381                                                 &total_in,
382                                                 &total_compressed,
383                                                 max_compressed);
384
385                 if (!ret) {
386                         unsigned long offset = total_compressed &
387                                 (PAGE_CACHE_SIZE - 1);
388                         struct page *page = pages[nr_pages_ret - 1];
389                         char *kaddr;
390
391                         /* zero the tail end of the last page, we might be
392                          * sending it down to disk
393                          */
394                         if (offset) {
395                                 kaddr = kmap_atomic(page, KM_USER0);
396                                 memset(kaddr + offset, 0,
397                                        PAGE_CACHE_SIZE - offset);
398                                 kunmap_atomic(kaddr, KM_USER0);
399                         }
400                         will_compress = 1;
401                 }
402         }
403         if (start == 0) {
404                 trans = btrfs_join_transaction(root, 1);
405                 BUG_ON(!trans);
406                 btrfs_set_trans_block_group(trans, inode);
407
408                 /* lets try to make an inline extent */
409                 if (ret || total_in < (actual_end - start)) {
410                         /* we didn't compress the entire range, try
411                          * to make an uncompressed inline extent.
412                          */
413                         ret = cow_file_range_inline(trans, root, inode,
414                                                     start, end, 0, NULL);
415                 } else {
416                         /* try making a compressed inline extent */
417                         ret = cow_file_range_inline(trans, root, inode,
418                                                     start, end,
419                                                     total_compressed, pages);
420                 }
421                 btrfs_end_transaction(trans, root);
422                 if (ret == 0) {
423                         /*
424                          * inline extent creation worked, we don't need
425                          * to create any more async work items.  Unlock
426                          * and free up our temp pages.
427                          */
428                         extent_clear_unlock_delalloc(inode,
429                                                      &BTRFS_I(inode)->io_tree,
430                                                      start, end, NULL, 1, 0,
431                                                      0, 1, 1, 1);
432                         ret = 0;
433                         goto free_pages_out;
434                 }
435         }
436
437         if (will_compress) {
438                 /*
439                  * we aren't doing an inline extent round the compressed size
440                  * up to a block size boundary so the allocator does sane
441                  * things
442                  */
443                 total_compressed = (total_compressed + blocksize - 1) &
444                         ~(blocksize - 1);
445
446                 /*
447                  * one last check to make sure the compression is really a
448                  * win, compare the page count read with the blocks on disk
449                  */
450                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
451                         ~(PAGE_CACHE_SIZE - 1);
452                 if (total_compressed >= total_in) {
453                         will_compress = 0;
454                 } else {
455                         disk_num_bytes = total_compressed;
456                         num_bytes = total_in;
457                 }
458         }
459         if (!will_compress && pages) {
460                 /*
461                  * the compression code ran but failed to make things smaller,
462                  * free any pages it allocated and our page pointer array
463                  */
464                 for (i = 0; i < nr_pages_ret; i++) {
465                         WARN_ON(pages[i]->mapping);
466                         page_cache_release(pages[i]);
467                 }
468                 kfree(pages);
469                 pages = NULL;
470                 total_compressed = 0;
471                 nr_pages_ret = 0;
472
473                 /* flag the file so we don't compress in the future */
474                 btrfs_set_flag(inode, NOCOMPRESS);
475         }
476         if (will_compress) {
477                 *num_added += 1;
478
479                 /* the async work queues will take care of doing actual
480                  * allocation on disk for these compressed pages,
481                  * and will submit them to the elevator.
482                  */
483                 add_async_extent(async_cow, start, num_bytes,
484                                  total_compressed, pages, nr_pages_ret);
485
486                 if (start + num_bytes < end && start + num_bytes < actual_end) {
487                         start += num_bytes;
488                         pages = NULL;
489                         cond_resched();
490                         goto again;
491                 }
492         } else {
493 cleanup_and_bail_uncompressed:
494                 /*
495                  * No compression, but we still need to write the pages in
496                  * the file we've been given so far.  redirty the locked
497                  * page if it corresponds to our extent and set things up
498                  * for the async work queue to run cow_file_range to do
499                  * the normal delalloc dance
500                  */
501                 if (page_offset(locked_page) >= start &&
502                     page_offset(locked_page) <= end) {
503                         __set_page_dirty_nobuffers(locked_page);
504                         /* unlocked later on in the async handlers */
505                 }
506                 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
507                 *num_added += 1;
508         }
509
510 out:
511         return 0;
512
513 free_pages_out:
514         for (i = 0; i < nr_pages_ret; i++) {
515                 WARN_ON(pages[i]->mapping);
516                 page_cache_release(pages[i]);
517         }
518         kfree(pages);
519
520         goto out;
521 }
522
523 /*
524  * phase two of compressed writeback.  This is the ordered portion
525  * of the code, which only gets called in the order the work was
526  * queued.  We walk all the async extents created by compress_file_range
527  * and send them down to the disk.
528  */
529 static noinline int submit_compressed_extents(struct inode *inode,
530                                               struct async_cow *async_cow)
531 {
532         struct async_extent *async_extent;
533         u64 alloc_hint = 0;
534         struct btrfs_trans_handle *trans;
535         struct btrfs_key ins;
536         struct extent_map *em;
537         struct btrfs_root *root = BTRFS_I(inode)->root;
538         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
539         struct extent_io_tree *io_tree;
540         int ret;
541
542         if (list_empty(&async_cow->extents))
543                 return 0;
544
545         trans = btrfs_join_transaction(root, 1);
546
547         while (!list_empty(&async_cow->extents)) {
548                 async_extent = list_entry(async_cow->extents.next,
549                                           struct async_extent, list);
550                 list_del(&async_extent->list);
551
552                 io_tree = &BTRFS_I(inode)->io_tree;
553
554                 /* did the compression code fall back to uncompressed IO? */
555                 if (!async_extent->pages) {
556                         int page_started = 0;
557                         unsigned long nr_written = 0;
558
559                         lock_extent(io_tree, async_extent->start,
560                                     async_extent->start +
561                                     async_extent->ram_size - 1, GFP_NOFS);
562
563                         /* allocate blocks */
564                         cow_file_range(inode, async_cow->locked_page,
565                                        async_extent->start,
566                                        async_extent->start +
567                                        async_extent->ram_size - 1,
568                                        &page_started, &nr_written, 0);
569
570                         /*
571                          * if page_started, cow_file_range inserted an
572                          * inline extent and took care of all the unlocking
573                          * and IO for us.  Otherwise, we need to submit
574                          * all those pages down to the drive.
575                          */
576                         if (!page_started)
577                                 extent_write_locked_range(io_tree,
578                                                   inode, async_extent->start,
579                                                   async_extent->start +
580                                                   async_extent->ram_size - 1,
581                                                   btrfs_get_extent,
582                                                   WB_SYNC_ALL);
583                         kfree(async_extent);
584                         cond_resched();
585                         continue;
586                 }
587
588                 lock_extent(io_tree, async_extent->start,
589                             async_extent->start + async_extent->ram_size - 1,
590                             GFP_NOFS);
591                 /*
592                  * here we're doing allocation and writeback of the
593                  * compressed pages
594                  */
595                 btrfs_drop_extent_cache(inode, async_extent->start,
596                                         async_extent->start +
597                                         async_extent->ram_size - 1, 0);
598
599                 ret = btrfs_reserve_extent(trans, root,
600                                            async_extent->compressed_size,
601                                            async_extent->compressed_size,
602                                            0, alloc_hint,
603                                            (u64)-1, &ins, 1);
604                 BUG_ON(ret);
605                 em = alloc_extent_map(GFP_NOFS);
606                 em->start = async_extent->start;
607                 em->len = async_extent->ram_size;
608                 em->orig_start = em->start;
609
610                 em->block_start = ins.objectid;
611                 em->block_len = ins.offset;
612                 em->bdev = root->fs_info->fs_devices->latest_bdev;
613                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
614                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
615
616                 while (1) {
617                         spin_lock(&em_tree->lock);
618                         ret = add_extent_mapping(em_tree, em);
619                         spin_unlock(&em_tree->lock);
620                         if (ret != -EEXIST) {
621                                 free_extent_map(em);
622                                 break;
623                         }
624                         btrfs_drop_extent_cache(inode, async_extent->start,
625                                                 async_extent->start +
626                                                 async_extent->ram_size - 1, 0);
627                 }
628
629                 ret = btrfs_add_ordered_extent(inode, async_extent->start,
630                                                ins.objectid,
631                                                async_extent->ram_size,
632                                                ins.offset,
633                                                BTRFS_ORDERED_COMPRESSED);
634                 BUG_ON(ret);
635
636                 btrfs_end_transaction(trans, root);
637
638                 /*
639                  * clear dirty, set writeback and unlock the pages.
640                  */
641                 extent_clear_unlock_delalloc(inode,
642                                              &BTRFS_I(inode)->io_tree,
643                                              async_extent->start,
644                                              async_extent->start +
645                                              async_extent->ram_size - 1,
646                                              NULL, 1, 1, 0, 1, 1, 0);
647
648                 ret = btrfs_submit_compressed_write(inode,
649                                     async_extent->start,
650                                     async_extent->ram_size,
651                                     ins.objectid,
652                                     ins.offset, async_extent->pages,
653                                     async_extent->nr_pages);
654
655                 BUG_ON(ret);
656                 trans = btrfs_join_transaction(root, 1);
657                 alloc_hint = ins.objectid + ins.offset;
658                 kfree(async_extent);
659                 cond_resched();
660         }
661
662         btrfs_end_transaction(trans, root);
663         return 0;
664 }
665
666 /*
667  * when extent_io.c finds a delayed allocation range in the file,
668  * the call backs end up in this code.  The basic idea is to
669  * allocate extents on disk for the range, and create ordered data structs
670  * in ram to track those extents.
671  *
672  * locked_page is the page that writepage had locked already.  We use
673  * it to make sure we don't do extra locks or unlocks.
674  *
675  * *page_started is set to one if we unlock locked_page and do everything
676  * required to start IO on it.  It may be clean and already done with
677  * IO when we return.
678  */
679 static noinline int cow_file_range(struct inode *inode,
680                                    struct page *locked_page,
681                                    u64 start, u64 end, int *page_started,
682                                    unsigned long *nr_written,
683                                    int unlock)
684 {
685         struct btrfs_root *root = BTRFS_I(inode)->root;
686         struct btrfs_trans_handle *trans;
687         u64 alloc_hint = 0;
688         u64 num_bytes;
689         unsigned long ram_size;
690         u64 disk_num_bytes;
691         u64 cur_alloc_size;
692         u64 blocksize = root->sectorsize;
693         u64 actual_end;
694         u64 isize = i_size_read(inode);
695         struct btrfs_key ins;
696         struct extent_map *em;
697         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
698         int ret = 0;
699
700         trans = btrfs_join_transaction(root, 1);
701         BUG_ON(!trans);
702         btrfs_set_trans_block_group(trans, inode);
703
704         actual_end = min_t(u64, isize, end + 1);
705
706         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
707         num_bytes = max(blocksize,  num_bytes);
708         disk_num_bytes = num_bytes;
709         ret = 0;
710
711         if (start == 0) {
712                 /* lets try to make an inline extent */
713                 ret = cow_file_range_inline(trans, root, inode,
714                                             start, end, 0, NULL);
715                 if (ret == 0) {
716                         extent_clear_unlock_delalloc(inode,
717                                                      &BTRFS_I(inode)->io_tree,
718                                                      start, end, NULL, 1, 1,
719                                                      1, 1, 1, 1);
720                         *nr_written = *nr_written +
721                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
722                         *page_started = 1;
723                         ret = 0;
724                         goto out;
725                 }
726         }
727
728         BUG_ON(disk_num_bytes >
729                btrfs_super_total_bytes(&root->fs_info->super_copy));
730
731         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
732
733         while (disk_num_bytes > 0) {
734                 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
735                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
736                                            root->sectorsize, 0, alloc_hint,
737                                            (u64)-1, &ins, 1);
738                 BUG_ON(ret);
739
740                 em = alloc_extent_map(GFP_NOFS);
741                 em->start = start;
742                 em->orig_start = em->start;
743
744                 ram_size = ins.offset;
745                 em->len = ins.offset;
746
747                 em->block_start = ins.objectid;
748                 em->block_len = ins.offset;
749                 em->bdev = root->fs_info->fs_devices->latest_bdev;
750                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
751
752                 while (1) {
753                         spin_lock(&em_tree->lock);
754                         ret = add_extent_mapping(em_tree, em);
755                         spin_unlock(&em_tree->lock);
756                         if (ret != -EEXIST) {
757                                 free_extent_map(em);
758                                 break;
759                         }
760                         btrfs_drop_extent_cache(inode, start,
761                                                 start + ram_size - 1, 0);
762                 }
763
764                 cur_alloc_size = ins.offset;
765                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
766                                                ram_size, cur_alloc_size, 0);
767                 BUG_ON(ret);
768
769                 if (root->root_key.objectid ==
770                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
771                         ret = btrfs_reloc_clone_csums(inode, start,
772                                                       cur_alloc_size);
773                         BUG_ON(ret);
774                 }
775
776                 if (disk_num_bytes < cur_alloc_size)
777                         break;
778
779                 /* we're not doing compressed IO, don't unlock the first
780                  * page (which the caller expects to stay locked), don't
781                  * clear any dirty bits and don't set any writeback bits
782                  */
783                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
784                                              start, start + ram_size - 1,
785                                              locked_page, unlock, 1,
786                                              1, 0, 0, 0);
787                 disk_num_bytes -= cur_alloc_size;
788                 num_bytes -= cur_alloc_size;
789                 alloc_hint = ins.objectid + ins.offset;
790                 start += cur_alloc_size;
791         }
792 out:
793         ret = 0;
794         btrfs_end_transaction(trans, root);
795
796         return ret;
797 }
798
799 /*
800  * work queue call back to started compression on a file and pages
801  */
802 static noinline void async_cow_start(struct btrfs_work *work)
803 {
804         struct async_cow *async_cow;
805         int num_added = 0;
806         async_cow = container_of(work, struct async_cow, work);
807
808         compress_file_range(async_cow->inode, async_cow->locked_page,
809                             async_cow->start, async_cow->end, async_cow,
810                             &num_added);
811         if (num_added == 0)
812                 async_cow->inode = NULL;
813 }
814
815 /*
816  * work queue call back to submit previously compressed pages
817  */
818 static noinline void async_cow_submit(struct btrfs_work *work)
819 {
820         struct async_cow *async_cow;
821         struct btrfs_root *root;
822         unsigned long nr_pages;
823
824         async_cow = container_of(work, struct async_cow, work);
825
826         root = async_cow->root;
827         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
828                 PAGE_CACHE_SHIFT;
829
830         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
831
832         if (atomic_read(&root->fs_info->async_delalloc_pages) <
833             5 * 1042 * 1024 &&
834             waitqueue_active(&root->fs_info->async_submit_wait))
835                 wake_up(&root->fs_info->async_submit_wait);
836
837         if (async_cow->inode)
838                 submit_compressed_extents(async_cow->inode, async_cow);
839 }
840
841 static noinline void async_cow_free(struct btrfs_work *work)
842 {
843         struct async_cow *async_cow;
844         async_cow = container_of(work, struct async_cow, work);
845         kfree(async_cow);
846 }
847
848 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
849                                 u64 start, u64 end, int *page_started,
850                                 unsigned long *nr_written)
851 {
852         struct async_cow *async_cow;
853         struct btrfs_root *root = BTRFS_I(inode)->root;
854         unsigned long nr_pages;
855         u64 cur_end;
856         int limit = 10 * 1024 * 1042;
857
858         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
859                          EXTENT_DELALLOC, 1, 0, GFP_NOFS);
860         while (start < end) {
861                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
862                 async_cow->inode = inode;
863                 async_cow->root = root;
864                 async_cow->locked_page = locked_page;
865                 async_cow->start = start;
866
867                 if (btrfs_test_flag(inode, NOCOMPRESS))
868                         cur_end = end;
869                 else
870                         cur_end = min(end, start + 512 * 1024 - 1);
871
872                 async_cow->end = cur_end;
873                 INIT_LIST_HEAD(&async_cow->extents);
874
875                 async_cow->work.func = async_cow_start;
876                 async_cow->work.ordered_func = async_cow_submit;
877                 async_cow->work.ordered_free = async_cow_free;
878                 async_cow->work.flags = 0;
879
880                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
881                         PAGE_CACHE_SHIFT;
882                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
883
884                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
885                                    &async_cow->work);
886
887                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
888                         wait_event(root->fs_info->async_submit_wait,
889                            (atomic_read(&root->fs_info->async_delalloc_pages) <
890                             limit));
891                 }
892
893                 while (atomic_read(&root->fs_info->async_submit_draining) &&
894                       atomic_read(&root->fs_info->async_delalloc_pages)) {
895                         wait_event(root->fs_info->async_submit_wait,
896                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
897                            0));
898                 }
899
900                 *nr_written += nr_pages;
901                 start = cur_end + 1;
902         }
903         *page_started = 1;
904         return 0;
905 }
906
907 static noinline int csum_exist_in_range(struct btrfs_root *root,
908                                         u64 bytenr, u64 num_bytes)
909 {
910         int ret;
911         struct btrfs_ordered_sum *sums;
912         LIST_HEAD(list);
913
914         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
915                                        bytenr + num_bytes - 1, &list);
916         if (ret == 0 && list_empty(&list))
917                 return 0;
918
919         while (!list_empty(&list)) {
920                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
921                 list_del(&sums->list);
922                 kfree(sums);
923         }
924         return 1;
925 }
926
927 /*
928  * when nowcow writeback call back.  This checks for snapshots or COW copies
929  * of the extents that exist in the file, and COWs the file as required.
930  *
931  * If no cow copies or snapshots exist, we write directly to the existing
932  * blocks on disk
933  */
934 static noinline int run_delalloc_nocow(struct inode *inode,
935                                        struct page *locked_page,
936                               u64 start, u64 end, int *page_started, int force,
937                               unsigned long *nr_written)
938 {
939         struct btrfs_root *root = BTRFS_I(inode)->root;
940         struct btrfs_trans_handle *trans;
941         struct extent_buffer *leaf;
942         struct btrfs_path *path;
943         struct btrfs_file_extent_item *fi;
944         struct btrfs_key found_key;
945         u64 cow_start;
946         u64 cur_offset;
947         u64 extent_end;
948         u64 disk_bytenr;
949         u64 num_bytes;
950         int extent_type;
951         int ret;
952         int type;
953         int nocow;
954         int check_prev = 1;
955
956         path = btrfs_alloc_path();
957         BUG_ON(!path);
958         trans = btrfs_join_transaction(root, 1);
959         BUG_ON(!trans);
960
961         cow_start = (u64)-1;
962         cur_offset = start;
963         while (1) {
964                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
965                                                cur_offset, 0);
966                 BUG_ON(ret < 0);
967                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
968                         leaf = path->nodes[0];
969                         btrfs_item_key_to_cpu(leaf, &found_key,
970                                               path->slots[0] - 1);
971                         if (found_key.objectid == inode->i_ino &&
972                             found_key.type == BTRFS_EXTENT_DATA_KEY)
973                                 path->slots[0]--;
974                 }
975                 check_prev = 0;
976 next_slot:
977                 leaf = path->nodes[0];
978                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
979                         ret = btrfs_next_leaf(root, path);
980                         if (ret < 0)
981                                 BUG_ON(1);
982                         if (ret > 0)
983                                 break;
984                         leaf = path->nodes[0];
985                 }
986
987                 nocow = 0;
988                 disk_bytenr = 0;
989                 num_bytes = 0;
990                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
991
992                 if (found_key.objectid > inode->i_ino ||
993                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
994                     found_key.offset > end)
995                         break;
996
997                 if (found_key.offset > cur_offset) {
998                         extent_end = found_key.offset;
999                         goto out_check;
1000                 }
1001
1002                 fi = btrfs_item_ptr(leaf, path->slots[0],
1003                                     struct btrfs_file_extent_item);
1004                 extent_type = btrfs_file_extent_type(leaf, fi);
1005
1006                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1007                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1008                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1009                         extent_end = found_key.offset +
1010                                 btrfs_file_extent_num_bytes(leaf, fi);
1011                         if (extent_end <= start) {
1012                                 path->slots[0]++;
1013                                 goto next_slot;
1014                         }
1015                         if (disk_bytenr == 0)
1016                                 goto out_check;
1017                         if (btrfs_file_extent_compression(leaf, fi) ||
1018                             btrfs_file_extent_encryption(leaf, fi) ||
1019                             btrfs_file_extent_other_encoding(leaf, fi))
1020                                 goto out_check;
1021                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1022                                 goto out_check;
1023                         if (btrfs_extent_readonly(root, disk_bytenr))
1024                                 goto out_check;
1025                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1026                                                   disk_bytenr))
1027                                 goto out_check;
1028                         disk_bytenr += btrfs_file_extent_offset(leaf, fi);
1029                         disk_bytenr += cur_offset - found_key.offset;
1030                         num_bytes = min(end + 1, extent_end) - cur_offset;
1031                         /*
1032                          * force cow if csum exists in the range.
1033                          * this ensure that csum for a given extent are
1034                          * either valid or do not exist.
1035                          */
1036                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1037                                 goto out_check;
1038                         nocow = 1;
1039                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1040                         extent_end = found_key.offset +
1041                                 btrfs_file_extent_inline_len(leaf, fi);
1042                         extent_end = ALIGN(extent_end, root->sectorsize);
1043                 } else {
1044                         BUG_ON(1);
1045                 }
1046 out_check:
1047                 if (extent_end <= start) {
1048                         path->slots[0]++;
1049                         goto next_slot;
1050                 }
1051                 if (!nocow) {
1052                         if (cow_start == (u64)-1)
1053                                 cow_start = cur_offset;
1054                         cur_offset = extent_end;
1055                         if (cur_offset > end)
1056                                 break;
1057                         path->slots[0]++;
1058                         goto next_slot;
1059                 }
1060
1061                 btrfs_release_path(root, path);
1062                 if (cow_start != (u64)-1) {
1063                         ret = cow_file_range(inode, locked_page, cow_start,
1064                                         found_key.offset - 1, page_started,
1065                                         nr_written, 1);
1066                         BUG_ON(ret);
1067                         cow_start = (u64)-1;
1068                 }
1069
1070                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1071                         struct extent_map *em;
1072                         struct extent_map_tree *em_tree;
1073                         em_tree = &BTRFS_I(inode)->extent_tree;
1074                         em = alloc_extent_map(GFP_NOFS);
1075                         em->start = cur_offset;
1076                         em->orig_start = em->start;
1077                         em->len = num_bytes;
1078                         em->block_len = num_bytes;
1079                         em->block_start = disk_bytenr;
1080                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1081                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1082                         while (1) {
1083                                 spin_lock(&em_tree->lock);
1084                                 ret = add_extent_mapping(em_tree, em);
1085                                 spin_unlock(&em_tree->lock);
1086                                 if (ret != -EEXIST) {
1087                                         free_extent_map(em);
1088                                         break;
1089                                 }
1090                                 btrfs_drop_extent_cache(inode, em->start,
1091                                                 em->start + em->len - 1, 0);
1092                         }
1093                         type = BTRFS_ORDERED_PREALLOC;
1094                 } else {
1095                         type = BTRFS_ORDERED_NOCOW;
1096                 }
1097
1098                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1099                                                num_bytes, num_bytes, type);
1100                 BUG_ON(ret);
1101
1102                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1103                                         cur_offset, cur_offset + num_bytes - 1,
1104                                         locked_page, 1, 1, 1, 0, 0, 0);
1105                 cur_offset = extent_end;
1106                 if (cur_offset > end)
1107                         break;
1108         }
1109         btrfs_release_path(root, path);
1110
1111         if (cur_offset <= end && cow_start == (u64)-1)
1112                 cow_start = cur_offset;
1113         if (cow_start != (u64)-1) {
1114                 ret = cow_file_range(inode, locked_page, cow_start, end,
1115                                      page_started, nr_written, 1);
1116                 BUG_ON(ret);
1117         }
1118
1119         ret = btrfs_end_transaction(trans, root);
1120         BUG_ON(ret);
1121         btrfs_free_path(path);
1122         return 0;
1123 }
1124
1125 /*
1126  * extent_io.c call back to do delayed allocation processing
1127  */
1128 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1129                               u64 start, u64 end, int *page_started,
1130                               unsigned long *nr_written)
1131 {
1132         int ret;
1133         struct btrfs_root *root = BTRFS_I(inode)->root;
1134
1135         if (btrfs_test_flag(inode, NODATACOW))
1136                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1137                                          page_started, 1, nr_written);
1138         else if (btrfs_test_flag(inode, PREALLOC))
1139                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1140                                          page_started, 0, nr_written);
1141         else if (!btrfs_test_opt(root, COMPRESS))
1142                 ret = cow_file_range(inode, locked_page, start, end,
1143                                       page_started, nr_written, 1);
1144         else
1145                 ret = cow_file_range_async(inode, locked_page, start, end,
1146                                            page_started, nr_written);
1147         return ret;
1148 }
1149
1150 /*
1151  * extent_io.c set_bit_hook, used to track delayed allocation
1152  * bytes in this file, and to maintain the list of inodes that
1153  * have pending delalloc work to be done.
1154  */
1155 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1156                        unsigned long old, unsigned long bits)
1157 {
1158         /*
1159          * set_bit and clear bit hooks normally require _irqsave/restore
1160          * but in this case, we are only testeing for the DELALLOC
1161          * bit, which is only set or cleared with irqs on
1162          */
1163         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1164                 struct btrfs_root *root = BTRFS_I(inode)->root;
1165                 btrfs_delalloc_reserve_space(root, inode, end - start + 1);
1166                 spin_lock(&root->fs_info->delalloc_lock);
1167                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1168                 root->fs_info->delalloc_bytes += end - start + 1;
1169                 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1170                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1171                                       &root->fs_info->delalloc_inodes);
1172                 }
1173                 spin_unlock(&root->fs_info->delalloc_lock);
1174         }
1175         return 0;
1176 }
1177
1178 /*
1179  * extent_io.c clear_bit_hook, see set_bit_hook for why
1180  */
1181 static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
1182                          unsigned long old, unsigned long bits)
1183 {
1184         /*
1185          * set_bit and clear bit hooks normally require _irqsave/restore
1186          * but in this case, we are only testeing for the DELALLOC
1187          * bit, which is only set or cleared with irqs on
1188          */
1189         if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1190                 struct btrfs_root *root = BTRFS_I(inode)->root;
1191
1192                 spin_lock(&root->fs_info->delalloc_lock);
1193                 if (end - start + 1 > root->fs_info->delalloc_bytes) {
1194                         printk(KERN_INFO "btrfs warning: delalloc account "
1195                                "%llu %llu\n",
1196                                (unsigned long long)end - start + 1,
1197                                (unsigned long long)
1198                                root->fs_info->delalloc_bytes);
1199                         btrfs_delalloc_free_space(root, inode, (u64)-1);
1200                         root->fs_info->delalloc_bytes = 0;
1201                         BTRFS_I(inode)->delalloc_bytes = 0;
1202                 } else {
1203                         btrfs_delalloc_free_space(root, inode,
1204                                                   end - start + 1);
1205                         root->fs_info->delalloc_bytes -= end - start + 1;
1206                         BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
1207                 }
1208                 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1209                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1210                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1211                 }
1212                 spin_unlock(&root->fs_info->delalloc_lock);
1213         }
1214         return 0;
1215 }
1216
1217 /*
1218  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1219  * we don't create bios that span stripes or chunks
1220  */
1221 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1222                          size_t size, struct bio *bio,
1223                          unsigned long bio_flags)
1224 {
1225         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1226         struct btrfs_mapping_tree *map_tree;
1227         u64 logical = (u64)bio->bi_sector << 9;
1228         u64 length = 0;
1229         u64 map_length;
1230         int ret;
1231
1232         if (bio_flags & EXTENT_BIO_COMPRESSED)
1233                 return 0;
1234
1235         length = bio->bi_size;
1236         map_tree = &root->fs_info->mapping_tree;
1237         map_length = length;
1238         ret = btrfs_map_block(map_tree, READ, logical,
1239                               &map_length, NULL, 0);
1240
1241         if (map_length < length + size)
1242                 return 1;
1243         return 0;
1244 }
1245
1246 /*
1247  * in order to insert checksums into the metadata in large chunks,
1248  * we wait until bio submission time.   All the pages in the bio are
1249  * checksummed and sums are attached onto the ordered extent record.
1250  *
1251  * At IO completion time the cums attached on the ordered extent record
1252  * are inserted into the btree
1253  */
1254 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1255                                     struct bio *bio, int mirror_num,
1256                                     unsigned long bio_flags)
1257 {
1258         struct btrfs_root *root = BTRFS_I(inode)->root;
1259         int ret = 0;
1260
1261         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1262         BUG_ON(ret);
1263         return 0;
1264 }
1265
1266 /*
1267  * in order to insert checksums into the metadata in large chunks,
1268  * we wait until bio submission time.   All the pages in the bio are
1269  * checksummed and sums are attached onto the ordered extent record.
1270  *
1271  * At IO completion time the cums attached on the ordered extent record
1272  * are inserted into the btree
1273  */
1274 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1275                           int mirror_num, unsigned long bio_flags)
1276 {
1277         struct btrfs_root *root = BTRFS_I(inode)->root;
1278         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1279 }
1280
1281 /*
1282  * extent_io.c submission hook. This does the right thing for csum calculation
1283  * on write, or reading the csums from the tree before a read
1284  */
1285 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1286                           int mirror_num, unsigned long bio_flags)
1287 {
1288         struct btrfs_root *root = BTRFS_I(inode)->root;
1289         int ret = 0;
1290         int skip_sum;
1291
1292         skip_sum = btrfs_test_flag(inode, NODATASUM);
1293
1294         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1295         BUG_ON(ret);
1296
1297         if (!(rw & (1 << BIO_RW))) {
1298                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1299                         return btrfs_submit_compressed_read(inode, bio,
1300                                                     mirror_num, bio_flags);
1301                 } else if (!skip_sum)
1302                         btrfs_lookup_bio_sums(root, inode, bio, NULL);
1303                 goto mapit;
1304         } else if (!skip_sum) {
1305                 /* csum items have already been cloned */
1306                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1307                         goto mapit;
1308                 /* we're doing a write, do the async checksumming */
1309                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1310                                    inode, rw, bio, mirror_num,
1311                                    bio_flags, __btrfs_submit_bio_start,
1312                                    __btrfs_submit_bio_done);
1313         }
1314
1315 mapit:
1316         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1317 }
1318
1319 /*
1320  * given a list of ordered sums record them in the inode.  This happens
1321  * at IO completion time based on sums calculated at bio submission time.
1322  */
1323 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1324                              struct inode *inode, u64 file_offset,
1325                              struct list_head *list)
1326 {
1327         struct btrfs_ordered_sum *sum;
1328
1329         btrfs_set_trans_block_group(trans, inode);
1330
1331         list_for_each_entry(sum, list, list) {
1332                 btrfs_csum_file_blocks(trans,
1333                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1334         }
1335         return 0;
1336 }
1337
1338 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1339 {
1340         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1341                 WARN_ON(1);
1342         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1343                                    GFP_NOFS);
1344 }
1345
1346 /* see btrfs_writepage_start_hook for details on why this is required */
1347 struct btrfs_writepage_fixup {
1348         struct page *page;
1349         struct btrfs_work work;
1350 };
1351
1352 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1353 {
1354         struct btrfs_writepage_fixup *fixup;
1355         struct btrfs_ordered_extent *ordered;
1356         struct page *page;
1357         struct inode *inode;
1358         u64 page_start;
1359         u64 page_end;
1360
1361         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1362         page = fixup->page;
1363 again:
1364         lock_page(page);
1365         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1366                 ClearPageChecked(page);
1367                 goto out_page;
1368         }
1369
1370         inode = page->mapping->host;
1371         page_start = page_offset(page);
1372         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1373
1374         lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1375
1376         /* already ordered? We're done */
1377         if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
1378                              EXTENT_ORDERED, 0)) {
1379                 goto out;
1380         }
1381
1382         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1383         if (ordered) {
1384                 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1385                               page_end, GFP_NOFS);
1386                 unlock_page(page);
1387                 btrfs_start_ordered_extent(inode, ordered, 1);
1388                 goto again;
1389         }
1390
1391         btrfs_set_extent_delalloc(inode, page_start, page_end);
1392         ClearPageChecked(page);
1393 out:
1394         unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1395 out_page:
1396         unlock_page(page);
1397         page_cache_release(page);
1398 }
1399
1400 /*
1401  * There are a few paths in the higher layers of the kernel that directly
1402  * set the page dirty bit without asking the filesystem if it is a
1403  * good idea.  This causes problems because we want to make sure COW
1404  * properly happens and the data=ordered rules are followed.
1405  *
1406  * In our case any range that doesn't have the ORDERED bit set
1407  * hasn't been properly setup for IO.  We kick off an async process
1408  * to fix it up.  The async helper will wait for ordered extents, set
1409  * the delalloc bit and make it safe to write the page.
1410  */
1411 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1412 {
1413         struct inode *inode = page->mapping->host;
1414         struct btrfs_writepage_fixup *fixup;
1415         struct btrfs_root *root = BTRFS_I(inode)->root;
1416         int ret;
1417
1418         ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1419                              EXTENT_ORDERED, 0);
1420         if (ret)
1421                 return 0;
1422
1423         if (PageChecked(page))
1424                 return -EAGAIN;
1425
1426         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1427         if (!fixup)
1428                 return -EAGAIN;
1429
1430         SetPageChecked(page);
1431         page_cache_get(page);
1432         fixup->work.func = btrfs_writepage_fixup_worker;
1433         fixup->page = page;
1434         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1435         return -EAGAIN;
1436 }
1437
1438 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1439                                        struct inode *inode, u64 file_pos,
1440                                        u64 disk_bytenr, u64 disk_num_bytes,
1441                                        u64 num_bytes, u64 ram_bytes,
1442                                        u64 locked_end,
1443                                        u8 compression, u8 encryption,
1444                                        u16 other_encoding, int extent_type)
1445 {
1446         struct btrfs_root *root = BTRFS_I(inode)->root;
1447         struct btrfs_file_extent_item *fi;
1448         struct btrfs_path *path;
1449         struct extent_buffer *leaf;
1450         struct btrfs_key ins;
1451         u64 hint;
1452         int ret;
1453
1454         path = btrfs_alloc_path();
1455         BUG_ON(!path);
1456
1457         path->leave_spinning = 1;
1458         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1459                                  file_pos + num_bytes, locked_end,
1460                                  file_pos, &hint);
1461         BUG_ON(ret);
1462
1463         ins.objectid = inode->i_ino;
1464         ins.offset = file_pos;
1465         ins.type = BTRFS_EXTENT_DATA_KEY;
1466         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1467         BUG_ON(ret);
1468         leaf = path->nodes[0];
1469         fi = btrfs_item_ptr(leaf, path->slots[0],
1470                             struct btrfs_file_extent_item);
1471         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1472         btrfs_set_file_extent_type(leaf, fi, extent_type);
1473         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1474         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1475         btrfs_set_file_extent_offset(leaf, fi, 0);
1476         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1477         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1478         btrfs_set_file_extent_compression(leaf, fi, compression);
1479         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1480         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1481
1482         btrfs_unlock_up_safe(path, 1);
1483         btrfs_set_lock_blocking(leaf);
1484
1485         btrfs_mark_buffer_dirty(leaf);
1486
1487         inode_add_bytes(inode, num_bytes);
1488         btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1489
1490         ins.objectid = disk_bytenr;
1491         ins.offset = disk_num_bytes;
1492         ins.type = BTRFS_EXTENT_ITEM_KEY;
1493         ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
1494                                           root->root_key.objectid,
1495                                           trans->transid, inode->i_ino, &ins);
1496         BUG_ON(ret);
1497         btrfs_free_path(path);
1498
1499         return 0;
1500 }
1501
1502 /*
1503  * helper function for btrfs_finish_ordered_io, this
1504  * just reads in some of the csum leaves to prime them into ram
1505  * before we start the transaction.  It limits the amount of btree
1506  * reads required while inside the transaction.
1507  */
1508 static noinline void reada_csum(struct btrfs_root *root,
1509                                 struct btrfs_path *path,
1510                                 struct btrfs_ordered_extent *ordered_extent)
1511 {
1512         struct btrfs_ordered_sum *sum;
1513         u64 bytenr;
1514
1515         sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum,
1516                          list);
1517         bytenr = sum->sums[0].bytenr;
1518
1519         /*
1520          * we don't care about the results, the point of this search is
1521          * just to get the btree leaves into ram
1522          */
1523         btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0);
1524 }
1525
1526 /* as ordered data IO finishes, this gets called so we can finish
1527  * an ordered extent if the range of bytes in the file it covers are
1528  * fully written.
1529  */
1530 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1531 {
1532         struct btrfs_root *root = BTRFS_I(inode)->root;
1533         struct btrfs_trans_handle *trans;
1534         struct btrfs_ordered_extent *ordered_extent = NULL;
1535         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1536         struct btrfs_path *path;
1537         int compressed = 0;
1538         int ret;
1539
1540         ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1541         if (!ret)
1542                 return 0;
1543
1544         /*
1545          * before we join the transaction, try to do some of our IO.
1546          * This will limit the amount of IO that we have to do with
1547          * the transaction running.  We're unlikely to need to do any
1548          * IO if the file extents are new, the disk_i_size checks
1549          * covers the most common case.
1550          */
1551         if (start < BTRFS_I(inode)->disk_i_size) {
1552                 path = btrfs_alloc_path();
1553                 if (path) {
1554                         ret = btrfs_lookup_file_extent(NULL, root, path,
1555                                                        inode->i_ino,
1556                                                        start, 0);
1557                         ordered_extent = btrfs_lookup_ordered_extent(inode,
1558                                                                      start);
1559                         if (!list_empty(&ordered_extent->list)) {
1560                                 btrfs_release_path(root, path);
1561                                 reada_csum(root, path, ordered_extent);
1562                         }
1563                         btrfs_free_path(path);
1564                 }
1565         }
1566
1567         trans = btrfs_join_transaction(root, 1);
1568
1569         if (!ordered_extent)
1570                 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1571         BUG_ON(!ordered_extent);
1572         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1573                 goto nocow;
1574
1575         lock_extent(io_tree, ordered_extent->file_offset,
1576                     ordered_extent->file_offset + ordered_extent->len - 1,
1577                     GFP_NOFS);
1578
1579         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1580                 compressed = 1;
1581         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1582                 BUG_ON(compressed);
1583                 ret = btrfs_mark_extent_written(trans, root, inode,
1584                                                 ordered_extent->file_offset,
1585                                                 ordered_extent->file_offset +
1586                                                 ordered_extent->len);
1587                 BUG_ON(ret);
1588         } else {
1589                 ret = insert_reserved_file_extent(trans, inode,
1590                                                 ordered_extent->file_offset,
1591                                                 ordered_extent->start,
1592                                                 ordered_extent->disk_len,
1593                                                 ordered_extent->len,
1594                                                 ordered_extent->len,
1595                                                 ordered_extent->file_offset +
1596                                                 ordered_extent->len,
1597                                                 compressed, 0, 0,
1598                                                 BTRFS_FILE_EXTENT_REG);
1599                 BUG_ON(ret);
1600         }
1601         unlock_extent(io_tree, ordered_extent->file_offset,
1602                     ordered_extent->file_offset + ordered_extent->len - 1,
1603                     GFP_NOFS);
1604 nocow:
1605         add_pending_csums(trans, inode, ordered_extent->file_offset,
1606                           &ordered_extent->list);
1607
1608         mutex_lock(&BTRFS_I(inode)->extent_mutex);
1609         btrfs_ordered_update_i_size(inode, ordered_extent);
1610         btrfs_update_inode(trans, root, inode);
1611         btrfs_remove_ordered_extent(inode, ordered_extent);
1612         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1613
1614         /* once for us */
1615         btrfs_put_ordered_extent(ordered_extent);
1616         /* once for the tree */
1617         btrfs_put_ordered_extent(ordered_extent);
1618
1619         btrfs_end_transaction(trans, root);
1620         return 0;
1621 }
1622
1623 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1624                                 struct extent_state *state, int uptodate)
1625 {
1626         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1627 }
1628
1629 /*
1630  * When IO fails, either with EIO or csum verification fails, we
1631  * try other mirrors that might have a good copy of the data.  This
1632  * io_failure_record is used to record state as we go through all the
1633  * mirrors.  If another mirror has good data, the page is set up to date
1634  * and things continue.  If a good mirror can't be found, the original
1635  * bio end_io callback is called to indicate things have failed.
1636  */
1637 struct io_failure_record {
1638         struct page *page;
1639         u64 start;
1640         u64 len;
1641         u64 logical;
1642         unsigned long bio_flags;
1643         int last_mirror;
1644 };
1645
1646 static int btrfs_io_failed_hook(struct bio *failed_bio,
1647                          struct page *page, u64 start, u64 end,
1648                          struct extent_state *state)
1649 {
1650         struct io_failure_record *failrec = NULL;
1651         u64 private;
1652         struct extent_map *em;
1653         struct inode *inode = page->mapping->host;
1654         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1655         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1656         struct bio *bio;
1657         int num_copies;
1658         int ret;
1659         int rw;
1660         u64 logical;
1661
1662         ret = get_state_private(failure_tree, start, &private);
1663         if (ret) {
1664                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1665                 if (!failrec)
1666                         return -ENOMEM;
1667                 failrec->start = start;
1668                 failrec->len = end - start + 1;
1669                 failrec->last_mirror = 0;
1670                 failrec->bio_flags = 0;
1671
1672                 spin_lock(&em_tree->lock);
1673                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1674                 if (em->start > start || em->start + em->len < start) {
1675                         free_extent_map(em);
1676                         em = NULL;
1677                 }
1678                 spin_unlock(&em_tree->lock);
1679
1680                 if (!em || IS_ERR(em)) {
1681                         kfree(failrec);
1682                         return -EIO;
1683                 }
1684                 logical = start - em->start;
1685                 logical = em->block_start + logical;
1686                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1687                         logical = em->block_start;
1688                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1689                 }
1690                 failrec->logical = logical;
1691                 free_extent_map(em);
1692                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1693                                 EXTENT_DIRTY, GFP_NOFS);
1694                 set_state_private(failure_tree, start,
1695                                  (u64)(unsigned long)failrec);
1696         } else {
1697                 failrec = (struct io_failure_record *)(unsigned long)private;
1698         }
1699         num_copies = btrfs_num_copies(
1700                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1701                               failrec->logical, failrec->len);
1702         failrec->last_mirror++;
1703         if (!state) {
1704                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1705                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1706                                                     failrec->start,
1707                                                     EXTENT_LOCKED);
1708                 if (state && state->start != failrec->start)
1709                         state = NULL;
1710                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1711         }
1712         if (!state || failrec->last_mirror > num_copies) {
1713                 set_state_private(failure_tree, failrec->start, 0);
1714                 clear_extent_bits(failure_tree, failrec->start,
1715                                   failrec->start + failrec->len - 1,
1716                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1717                 kfree(failrec);
1718                 return -EIO;
1719         }
1720         bio = bio_alloc(GFP_NOFS, 1);
1721         bio->bi_private = state;
1722         bio->bi_end_io = failed_bio->bi_end_io;
1723         bio->bi_sector = failrec->logical >> 9;
1724         bio->bi_bdev = failed_bio->bi_bdev;
1725         bio->bi_size = 0;
1726
1727         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1728         if (failed_bio->bi_rw & (1 << BIO_RW))
1729                 rw = WRITE;
1730         else
1731                 rw = READ;
1732
1733         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1734                                                       failrec->last_mirror,
1735                                                       failrec->bio_flags);
1736         return 0;
1737 }
1738
1739 /*
1740  * each time an IO finishes, we do a fast check in the IO failure tree
1741  * to see if we need to process or clean up an io_failure_record
1742  */
1743 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1744 {
1745         u64 private;
1746         u64 private_failure;
1747         struct io_failure_record *failure;
1748         int ret;
1749
1750         private = 0;
1751         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1752                              (u64)-1, 1, EXTENT_DIRTY)) {
1753                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1754                                         start, &private_failure);
1755                 if (ret == 0) {
1756                         failure = (struct io_failure_record *)(unsigned long)
1757                                    private_failure;
1758                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1759                                           failure->start, 0);
1760                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1761                                           failure->start,
1762                                           failure->start + failure->len - 1,
1763                                           EXTENT_DIRTY | EXTENT_LOCKED,
1764                                           GFP_NOFS);
1765                         kfree(failure);
1766                 }
1767         }
1768         return 0;
1769 }
1770
1771 /*
1772  * when reads are done, we need to check csums to verify the data is correct
1773  * if there's a match, we allow the bio to finish.  If not, we go through
1774  * the io_failure_record routines to find good copies
1775  */
1776 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1777                                struct extent_state *state)
1778 {
1779         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1780         struct inode *inode = page->mapping->host;
1781         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1782         char *kaddr;
1783         u64 private = ~(u32)0;
1784         int ret;
1785         struct btrfs_root *root = BTRFS_I(inode)->root;
1786         u32 csum = ~(u32)0;
1787
1788         if (PageChecked(page)) {
1789                 ClearPageChecked(page);
1790                 goto good;
1791         }
1792         if (btrfs_test_flag(inode, NODATASUM))
1793                 return 0;
1794
1795         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1796             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1)) {
1797                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1798                                   GFP_NOFS);
1799                 return 0;
1800         }
1801
1802         if (state && state->start == start) {
1803                 private = state->private;
1804                 ret = 0;
1805         } else {
1806                 ret = get_state_private(io_tree, start, &private);
1807         }
1808         kaddr = kmap_atomic(page, KM_USER0);
1809         if (ret)
1810                 goto zeroit;
1811
1812         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1813         btrfs_csum_final(csum, (char *)&csum);
1814         if (csum != private)
1815                 goto zeroit;
1816
1817         kunmap_atomic(kaddr, KM_USER0);
1818 good:
1819         /* if the io failure tree for this inode is non-empty,
1820          * check to see if we've recovered from a failed IO
1821          */
1822         btrfs_clean_io_failures(inode, start);
1823         return 0;
1824
1825 zeroit:
1826         printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1827                "private %llu\n", page->mapping->host->i_ino,
1828                (unsigned long long)start, csum,
1829                (unsigned long long)private);
1830         memset(kaddr + offset, 1, end - start + 1);
1831         flush_dcache_page(page);
1832         kunmap_atomic(kaddr, KM_USER0);
1833         if (private == 0)
1834                 return 0;
1835         return -EIO;
1836 }
1837
1838 /*
1839  * This creates an orphan entry for the given inode in case something goes
1840  * wrong in the middle of an unlink/truncate.
1841  */
1842 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1843 {
1844         struct btrfs_root *root = BTRFS_I(inode)->root;
1845         int ret = 0;
1846
1847         spin_lock(&root->list_lock);
1848
1849         /* already on the orphan list, we're good */
1850         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
1851                 spin_unlock(&root->list_lock);
1852                 return 0;
1853         }
1854
1855         list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1856
1857         spin_unlock(&root->list_lock);
1858
1859         /*
1860          * insert an orphan item to track this unlinked/truncated file
1861          */
1862         ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1863
1864         return ret;
1865 }
1866
1867 /*
1868  * We have done the truncate/delete so we can go ahead and remove the orphan
1869  * item for this particular inode.
1870  */
1871 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1872 {
1873         struct btrfs_root *root = BTRFS_I(inode)->root;
1874         int ret = 0;
1875
1876         spin_lock(&root->list_lock);
1877
1878         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
1879                 spin_unlock(&root->list_lock);
1880                 return 0;
1881         }
1882
1883         list_del_init(&BTRFS_I(inode)->i_orphan);
1884         if (!trans) {
1885                 spin_unlock(&root->list_lock);
1886                 return 0;
1887         }
1888
1889         spin_unlock(&root->list_lock);
1890
1891         ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1892
1893         return ret;
1894 }
1895
1896 /*
1897  * this cleans up any orphans that may be left on the list from the last use
1898  * of this root.
1899  */
1900 void btrfs_orphan_cleanup(struct btrfs_root *root)
1901 {
1902         struct btrfs_path *path;
1903         struct extent_buffer *leaf;
1904         struct btrfs_item *item;
1905         struct btrfs_key key, found_key;
1906         struct btrfs_trans_handle *trans;
1907         struct inode *inode;
1908         int ret = 0, nr_unlink = 0, nr_truncate = 0;
1909
1910         path = btrfs_alloc_path();
1911         if (!path)
1912                 return;
1913         path->reada = -1;
1914
1915         key.objectid = BTRFS_ORPHAN_OBJECTID;
1916         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1917         key.offset = (u64)-1;
1918
1919
1920         while (1) {
1921                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1922                 if (ret < 0) {
1923                         printk(KERN_ERR "Error searching slot for orphan: %d"
1924                                "\n", ret);
1925                         break;
1926                 }
1927
1928                 /*
1929                  * if ret == 0 means we found what we were searching for, which
1930                  * is weird, but possible, so only screw with path if we didnt
1931                  * find the key and see if we have stuff that matches
1932                  */
1933                 if (ret > 0) {
1934                         if (path->slots[0] == 0)
1935                                 break;
1936                         path->slots[0]--;
1937                 }
1938
1939                 /* pull out the item */
1940                 leaf = path->nodes[0];
1941                 item = btrfs_item_nr(leaf, path->slots[0]);
1942                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1943
1944                 /* make sure the item matches what we want */
1945                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1946                         break;
1947                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1948                         break;
1949
1950                 /* release the path since we're done with it */
1951                 btrfs_release_path(root, path);
1952
1953                 /*
1954                  * this is where we are basically btrfs_lookup, without the
1955                  * crossing root thing.  we store the inode number in the
1956                  * offset of the orphan item.
1957                  */
1958                 inode = btrfs_iget_locked(root->fs_info->sb,
1959                                           found_key.offset, root);
1960                 if (!inode)
1961                         break;
1962
1963                 if (inode->i_state & I_NEW) {
1964                         BTRFS_I(inode)->root = root;
1965
1966                         /* have to set the location manually */
1967                         BTRFS_I(inode)->location.objectid = inode->i_ino;
1968                         BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1969                         BTRFS_I(inode)->location.offset = 0;
1970
1971                         btrfs_read_locked_inode(inode);
1972                         unlock_new_inode(inode);
1973                 }
1974
1975                 /*
1976                  * add this inode to the orphan list so btrfs_orphan_del does
1977                  * the proper thing when we hit it
1978                  */
1979                 spin_lock(&root->list_lock);
1980                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1981                 spin_unlock(&root->list_lock);
1982
1983                 /*
1984                  * if this is a bad inode, means we actually succeeded in
1985                  * removing the inode, but not the orphan record, which means
1986                  * we need to manually delete the orphan since iput will just
1987                  * do a destroy_inode
1988                  */
1989                 if (is_bad_inode(inode)) {
1990                         trans = btrfs_start_transaction(root, 1);
1991                         btrfs_orphan_del(trans, inode);
1992                         btrfs_end_transaction(trans, root);
1993                         iput(inode);
1994                         continue;
1995                 }
1996
1997                 /* if we have links, this was a truncate, lets do that */
1998                 if (inode->i_nlink) {
1999                         nr_truncate++;
2000                         btrfs_truncate(inode);
2001                 } else {
2002                         nr_unlink++;
2003                 }
2004
2005                 /* this will do delete_inode and everything for us */
2006                 iput(inode);
2007         }
2008
2009         if (nr_unlink)
2010                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2011         if (nr_truncate)
2012                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2013
2014         btrfs_free_path(path);
2015 }
2016
2017 /*
2018  * read an inode from the btree into the in-memory inode
2019  */
2020 void btrfs_read_locked_inode(struct inode *inode)
2021 {
2022         struct btrfs_path *path;
2023         struct extent_buffer *leaf;
2024         struct btrfs_inode_item *inode_item;
2025         struct btrfs_timespec *tspec;
2026         struct btrfs_root *root = BTRFS_I(inode)->root;
2027         struct btrfs_key location;
2028         u64 alloc_group_block;
2029         u32 rdev;
2030         int ret;
2031
2032         path = btrfs_alloc_path();
2033         BUG_ON(!path);
2034         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2035
2036         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2037         if (ret)
2038                 goto make_bad;
2039
2040         leaf = path->nodes[0];
2041         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2042                                     struct btrfs_inode_item);
2043
2044         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2045         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2046         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2047         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2048         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2049
2050         tspec = btrfs_inode_atime(inode_item);
2051         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2052         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2053
2054         tspec = btrfs_inode_mtime(inode_item);
2055         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2056         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2057
2058         tspec = btrfs_inode_ctime(inode_item);
2059         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2060         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2061
2062         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2063         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2064         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2065         inode->i_generation = BTRFS_I(inode)->generation;
2066         inode->i_rdev = 0;
2067         rdev = btrfs_inode_rdev(leaf, inode_item);
2068
2069         BTRFS_I(inode)->index_cnt = (u64)-1;
2070         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2071
2072         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2073
2074         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2075                                                 alloc_group_block, 0);
2076         btrfs_free_path(path);
2077         inode_item = NULL;
2078
2079         switch (inode->i_mode & S_IFMT) {
2080         case S_IFREG:
2081                 inode->i_mapping->a_ops = &btrfs_aops;
2082                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2083                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2084                 inode->i_fop = &btrfs_file_operations;
2085                 inode->i_op = &btrfs_file_inode_operations;
2086                 break;
2087         case S_IFDIR:
2088                 inode->i_fop = &btrfs_dir_file_operations;
2089                 if (root == root->fs_info->tree_root)
2090                         inode->i_op = &btrfs_dir_ro_inode_operations;
2091                 else
2092                         inode->i_op = &btrfs_dir_inode_operations;
2093                 break;
2094         case S_IFLNK:
2095                 inode->i_op = &btrfs_symlink_inode_operations;
2096                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2097                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2098                 break;
2099         default:
2100                 inode->i_op = &btrfs_special_inode_operations;
2101                 init_special_inode(inode, inode->i_mode, rdev);
2102                 break;
2103         }
2104         return;
2105
2106 make_bad:
2107         btrfs_free_path(path);
2108         make_bad_inode(inode);
2109 }
2110
2111 /*
2112  * given a leaf and an inode, copy the inode fields into the leaf
2113  */
2114 static void fill_inode_item(struct btrfs_trans_handle *trans,
2115                             struct extent_buffer *leaf,
2116                             struct btrfs_inode_item *item,
2117                             struct inode *inode)
2118 {
2119         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2120         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2121         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2122         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2123         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2124
2125         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2126                                inode->i_atime.tv_sec);
2127         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2128                                 inode->i_atime.tv_nsec);
2129
2130         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2131                                inode->i_mtime.tv_sec);
2132         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2133                                 inode->i_mtime.tv_nsec);
2134
2135         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2136                                inode->i_ctime.tv_sec);
2137         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2138                                 inode->i_ctime.tv_nsec);
2139
2140         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2141         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2142         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2143         btrfs_set_inode_transid(leaf, item, trans->transid);
2144         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2145         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2146         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2147 }
2148
2149 /*
2150  * copy everything in the in-memory inode into the btree.
2151  */
2152 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2153                                 struct btrfs_root *root, struct inode *inode)
2154 {
2155         struct btrfs_inode_item *inode_item;
2156         struct btrfs_path *path;
2157         struct extent_buffer *leaf;
2158         int ret;
2159
2160         path = btrfs_alloc_path();
2161         BUG_ON(!path);
2162         path->leave_spinning = 1;
2163         ret = btrfs_lookup_inode(trans, root, path,
2164                                  &BTRFS_I(inode)->location, 1);
2165         if (ret) {
2166                 if (ret > 0)
2167                         ret = -ENOENT;
2168                 goto failed;
2169         }
2170
2171         btrfs_unlock_up_safe(path, 1);
2172         leaf = path->nodes[0];
2173         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2174                                   struct btrfs_inode_item);
2175
2176         fill_inode_item(trans, leaf, inode_item, inode);
2177         btrfs_mark_buffer_dirty(leaf);
2178         btrfs_set_inode_last_trans(trans, inode);
2179         ret = 0;
2180 failed:
2181         btrfs_free_path(path);
2182         return ret;
2183 }
2184
2185
2186 /*
2187  * unlink helper that gets used here in inode.c and in the tree logging
2188  * recovery code.  It remove a link in a directory with a given name, and
2189  * also drops the back refs in the inode to the directory
2190  */
2191 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2192                        struct btrfs_root *root,
2193                        struct inode *dir, struct inode *inode,
2194                        const char *name, int name_len)
2195 {
2196         struct btrfs_path *path;
2197         int ret = 0;
2198         struct extent_buffer *leaf;
2199         struct btrfs_dir_item *di;
2200         struct btrfs_key key;
2201         u64 index;
2202
2203         path = btrfs_alloc_path();
2204         if (!path) {
2205                 ret = -ENOMEM;
2206                 goto err;
2207         }
2208
2209         path->leave_spinning = 1;
2210         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2211                                     name, name_len, -1);
2212         if (IS_ERR(di)) {
2213                 ret = PTR_ERR(di);
2214                 goto err;
2215         }
2216         if (!di) {
2217                 ret = -ENOENT;
2218                 goto err;
2219         }
2220         leaf = path->nodes[0];
2221         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2222         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2223         if (ret)
2224                 goto err;
2225         btrfs_release_path(root, path);
2226
2227         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2228                                   inode->i_ino,
2229                                   dir->i_ino, &index);
2230         if (ret) {
2231                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2232                        "inode %lu parent %lu\n", name_len, name,
2233                        inode->i_ino, dir->i_ino);
2234                 goto err;
2235         }
2236
2237         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2238                                          index, name, name_len, -1);
2239         if (IS_ERR(di)) {
2240                 ret = PTR_ERR(di);
2241                 goto err;
2242         }
2243         if (!di) {
2244                 ret = -ENOENT;
2245                 goto err;
2246         }
2247         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2248         btrfs_release_path(root, path);
2249
2250         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2251                                          inode, dir->i_ino);
2252         BUG_ON(ret != 0 && ret != -ENOENT);
2253
2254         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2255                                            dir, index);
2256         BUG_ON(ret);
2257 err:
2258         btrfs_free_path(path);
2259         if (ret)
2260                 goto out;
2261
2262         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2263         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2264         btrfs_update_inode(trans, root, dir);
2265         btrfs_drop_nlink(inode);
2266         ret = btrfs_update_inode(trans, root, inode);
2267         dir->i_sb->s_dirt = 1;
2268 out:
2269         return ret;
2270 }
2271
2272 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2273 {
2274         struct btrfs_root *root;
2275         struct btrfs_trans_handle *trans;
2276         struct inode *inode = dentry->d_inode;
2277         int ret;
2278         unsigned long nr = 0;
2279
2280         root = BTRFS_I(dir)->root;
2281
2282         trans = btrfs_start_transaction(root, 1);
2283
2284         btrfs_set_trans_block_group(trans, dir);
2285
2286         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2287
2288         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2289                                  dentry->d_name.name, dentry->d_name.len);
2290
2291         if (inode->i_nlink == 0)
2292                 ret = btrfs_orphan_add(trans, inode);
2293
2294         nr = trans->blocks_used;
2295
2296         btrfs_end_transaction_throttle(trans, root);
2297         btrfs_btree_balance_dirty(root, nr);
2298         return ret;
2299 }
2300
2301 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2302 {
2303         struct inode *inode = dentry->d_inode;
2304         int err = 0;
2305         int ret;
2306         struct btrfs_root *root = BTRFS_I(dir)->root;
2307         struct btrfs_trans_handle *trans;
2308         unsigned long nr = 0;
2309
2310         /*
2311          * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2312          * the root of a subvolume or snapshot
2313          */
2314         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2315             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
2316                 return -ENOTEMPTY;
2317         }
2318
2319         trans = btrfs_start_transaction(root, 1);
2320         btrfs_set_trans_block_group(trans, dir);
2321
2322         err = btrfs_orphan_add(trans, inode);
2323         if (err)
2324                 goto fail_trans;
2325
2326         /* now the directory is empty */
2327         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2328                                  dentry->d_name.name, dentry->d_name.len);
2329         if (!err)
2330                 btrfs_i_size_write(inode, 0);
2331
2332 fail_trans:
2333         nr = trans->blocks_used;
2334         ret = btrfs_end_transaction_throttle(trans, root);
2335         btrfs_btree_balance_dirty(root, nr);
2336
2337         if (ret && !err)
2338                 err = ret;
2339         return err;
2340 }
2341
2342 #if 0
2343 /*
2344  * when truncating bytes in a file, it is possible to avoid reading
2345  * the leaves that contain only checksum items.  This can be the
2346  * majority of the IO required to delete a large file, but it must
2347  * be done carefully.
2348  *
2349  * The keys in the level just above the leaves are checked to make sure
2350  * the lowest key in a given leaf is a csum key, and starts at an offset
2351  * after the new  size.
2352  *
2353  * Then the key for the next leaf is checked to make sure it also has
2354  * a checksum item for the same file.  If it does, we know our target leaf
2355  * contains only checksum items, and it can be safely freed without reading
2356  * it.
2357  *
2358  * This is just an optimization targeted at large files.  It may do
2359  * nothing.  It will return 0 unless things went badly.
2360  */
2361 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2362                                      struct btrfs_root *root,
2363                                      struct btrfs_path *path,
2364                                      struct inode *inode, u64 new_size)
2365 {
2366         struct btrfs_key key;
2367         int ret;
2368         int nritems;
2369         struct btrfs_key found_key;
2370         struct btrfs_key other_key;
2371         struct btrfs_leaf_ref *ref;
2372         u64 leaf_gen;
2373         u64 leaf_start;
2374
2375         path->lowest_level = 1;
2376         key.objectid = inode->i_ino;
2377         key.type = BTRFS_CSUM_ITEM_KEY;
2378         key.offset = new_size;
2379 again:
2380         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2381         if (ret < 0)
2382                 goto out;
2383
2384         if (path->nodes[1] == NULL) {
2385                 ret = 0;
2386                 goto out;
2387         }
2388         ret = 0;
2389         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2390         nritems = btrfs_header_nritems(path->nodes[1]);
2391
2392         if (!nritems)
2393                 goto out;
2394
2395         if (path->slots[1] >= nritems)
2396                 goto next_node;
2397
2398         /* did we find a key greater than anything we want to delete? */
2399         if (found_key.objectid > inode->i_ino ||
2400            (found_key.objectid == inode->i_ino && found_key.type > key.type))
2401                 goto out;
2402
2403         /* we check the next key in the node to make sure the leave contains
2404          * only checksum items.  This comparison doesn't work if our
2405          * leaf is the last one in the node
2406          */
2407         if (path->slots[1] + 1 >= nritems) {
2408 next_node:
2409                 /* search forward from the last key in the node, this
2410                  * will bring us into the next node in the tree
2411                  */
2412                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2413
2414                 /* unlikely, but we inc below, so check to be safe */
2415                 if (found_key.offset == (u64)-1)
2416                         goto out;
2417
2418                 /* search_forward needs a path with locks held, do the
2419                  * search again for the original key.  It is possible
2420                  * this will race with a balance and return a path that
2421                  * we could modify, but this drop is just an optimization
2422                  * and is allowed to miss some leaves.
2423                  */
2424                 btrfs_release_path(root, path);
2425                 found_key.offset++;
2426
2427                 /* setup a max key for search_forward */
2428                 other_key.offset = (u64)-1;
2429                 other_key.type = key.type;
2430                 other_key.objectid = key.objectid;
2431
2432                 path->keep_locks = 1;
2433                 ret = btrfs_search_forward(root, &found_key, &other_key,
2434                                            path, 0, 0);
2435                 path->keep_locks = 0;
2436                 if (ret || found_key.objectid != key.objectid ||
2437                     found_key.type != key.type) {
2438                         ret = 0;
2439                         goto out;
2440                 }
2441
2442                 key.offset = found_key.offset;
2443                 btrfs_release_path(root, path);
2444                 cond_resched();
2445                 goto again;
2446         }
2447
2448         /* we know there's one more slot after us in the tree,
2449          * read that key so we can verify it is also a checksum item
2450          */
2451         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2452
2453         if (found_key.objectid < inode->i_ino)
2454                 goto next_key;
2455
2456         if (found_key.type != key.type || found_key.offset < new_size)
2457                 goto next_key;
2458
2459         /*
2460          * if the key for the next leaf isn't a csum key from this objectid,
2461          * we can't be sure there aren't good items inside this leaf.
2462          * Bail out
2463          */
2464         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2465                 goto out;
2466
2467         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2468         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2469         /*
2470          * it is safe to delete this leaf, it contains only
2471          * csum items from this inode at an offset >= new_size
2472          */
2473         ret = btrfs_del_leaf(trans, root, path, leaf_start);
2474         BUG_ON(ret);
2475
2476         if (root->ref_cows && leaf_gen < trans->transid) {
2477                 ref = btrfs_alloc_leaf_ref(root, 0);
2478                 if (ref) {
2479                         ref->root_gen = root->root_key.offset;
2480                         ref->bytenr = leaf_start;
2481                         ref->owner = 0;
2482                         ref->generation = leaf_gen;
2483                         ref->nritems = 0;
2484
2485                         btrfs_sort_leaf_ref(ref);
2486
2487                         ret = btrfs_add_leaf_ref(root, ref, 0);
2488                         WARN_ON(ret);
2489                         btrfs_free_leaf_ref(root, ref);
2490                 } else {
2491                         WARN_ON(1);
2492                 }
2493         }
2494 next_key:
2495         btrfs_release_path(root, path);
2496
2497         if (other_key.objectid == inode->i_ino &&
2498             other_key.type == key.type && other_key.offset > key.offset) {
2499                 key.offset = other_key.offset;
2500                 cond_resched();
2501                 goto again;
2502         }
2503         ret = 0;
2504 out:
2505         /* fixup any changes we've made to the path */
2506         path->lowest_level = 0;
2507         path->keep_locks = 0;
2508         btrfs_release_path(root, path);
2509         return ret;
2510 }
2511
2512 #endif
2513
2514 /*
2515  * this can truncate away extent items, csum items and directory items.
2516  * It starts at a high offset and removes keys until it can't find
2517  * any higher than new_size
2518  *
2519  * csum items that cross the new i_size are truncated to the new size
2520  * as well.
2521  *
2522  * min_type is the minimum key type to truncate down to.  If set to 0, this
2523  * will kill all the items on this inode, including the INODE_ITEM_KEY.
2524  */
2525 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2526                                         struct btrfs_root *root,
2527                                         struct inode *inode,
2528                                         u64 new_size, u32 min_type)
2529 {
2530         int ret;
2531         struct btrfs_path *path;
2532         struct btrfs_key key;
2533         struct btrfs_key found_key;
2534         u32 found_type = (u8)-1;
2535         struct extent_buffer *leaf;
2536         struct btrfs_file_extent_item *fi;
2537         u64 extent_start = 0;
2538         u64 extent_num_bytes = 0;
2539         u64 item_end = 0;
2540         u64 root_gen = 0;
2541         u64 root_owner = 0;
2542         int found_extent;
2543         int del_item;
2544         int pending_del_nr = 0;
2545         int pending_del_slot = 0;
2546         int extent_type = -1;
2547         int encoding;
2548         u64 mask = root->sectorsize - 1;
2549
2550         if (root->ref_cows)
2551                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2552         path = btrfs_alloc_path();
2553         path->reada = -1;
2554         BUG_ON(!path);
2555
2556         /* FIXME, add redo link to tree so we don't leak on crash */
2557         key.objectid = inode->i_ino;
2558         key.offset = (u64)-1;
2559         key.type = (u8)-1;
2560
2561 search_again:
2562         path->leave_spinning = 1;
2563         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2564         if (ret < 0)
2565                 goto error;
2566
2567         if (ret > 0) {
2568                 /* there are no items in the tree for us to truncate, we're
2569                  * done
2570                  */
2571                 if (path->slots[0] == 0) {
2572                         ret = 0;
2573                         goto error;
2574                 }
2575                 path->slots[0]--;
2576         }
2577
2578         while (1) {
2579                 fi = NULL;
2580                 leaf = path->nodes[0];
2581                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2582                 found_type = btrfs_key_type(&found_key);
2583                 encoding = 0;
2584
2585                 if (found_key.objectid != inode->i_ino)
2586                         break;
2587
2588                 if (found_type < min_type)
2589                         break;
2590
2591                 item_end = found_key.offset;
2592                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2593                         fi = btrfs_item_ptr(leaf, path->slots[0],
2594                                             struct btrfs_file_extent_item);
2595                         extent_type = btrfs_file_extent_type(leaf, fi);
2596                         encoding = btrfs_file_extent_compression(leaf, fi);
2597                         encoding |= btrfs_file_extent_encryption(leaf, fi);
2598                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2599
2600                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2601                                 item_end +=
2602                                     btrfs_file_extent_num_bytes(leaf, fi);
2603                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2604                                 item_end += btrfs_file_extent_inline_len(leaf,
2605                                                                          fi);
2606                         }
2607                         item_end--;
2608                 }
2609                 if (item_end < new_size) {
2610                         if (found_type == BTRFS_DIR_ITEM_KEY)
2611                                 found_type = BTRFS_INODE_ITEM_KEY;
2612                         else if (found_type == BTRFS_EXTENT_ITEM_KEY)
2613                                 found_type = BTRFS_EXTENT_DATA_KEY;
2614                         else if (found_type == BTRFS_EXTENT_DATA_KEY)
2615                                 found_type = BTRFS_XATTR_ITEM_KEY;
2616                         else if (found_type == BTRFS_XATTR_ITEM_KEY)
2617                                 found_type = BTRFS_INODE_REF_KEY;
2618                         else if (found_type)
2619                                 found_type--;
2620                         else
2621                                 break;
2622                         btrfs_set_key_type(&key, found_type);
2623                         goto next;
2624                 }
2625                 if (found_key.offset >= new_size)
2626                         del_item = 1;
2627                 else
2628                         del_item = 0;
2629                 found_extent = 0;
2630
2631                 /* FIXME, shrink the extent if the ref count is only 1 */
2632                 if (found_type != BTRFS_EXTENT_DATA_KEY)
2633                         goto delete;
2634
2635                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2636                         u64 num_dec;
2637                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2638                         if (!del_item && !encoding) {
2639                                 u64 orig_num_bytes =
2640                                         btrfs_file_extent_num_bytes(leaf, fi);
2641                                 extent_num_bytes = new_size -
2642                                         found_key.offset + root->sectorsize - 1;
2643                                 extent_num_bytes = extent_num_bytes &
2644                                         ~((u64)root->sectorsize - 1);
2645                                 btrfs_set_file_extent_num_bytes(leaf, fi,
2646                                                          extent_num_bytes);
2647                                 num_dec = (orig_num_bytes -
2648                                            extent_num_bytes);
2649                                 if (root->ref_cows && extent_start != 0)
2650                                         inode_sub_bytes(inode, num_dec);
2651                                 btrfs_mark_buffer_dirty(leaf);
2652                         } else {
2653                                 extent_num_bytes =
2654                                         btrfs_file_extent_disk_num_bytes(leaf,
2655                                                                          fi);
2656                                 /* FIXME blocksize != 4096 */
2657                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2658                                 if (extent_start != 0) {
2659                                         found_extent = 1;
2660                                         if (root->ref_cows)
2661                                                 inode_sub_bytes(inode, num_dec);
2662                                 }
2663                                 root_gen = btrfs_header_generation(leaf);
2664                                 root_owner = btrfs_header_owner(leaf);
2665                         }
2666                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2667                         /*
2668                          * we can't truncate inline items that have had
2669                          * special encodings
2670                          */
2671                         if (!del_item &&
2672                             btrfs_file_extent_compression(leaf, fi) == 0 &&
2673                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
2674                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2675                                 u32 size = new_size - found_key.offset;
2676
2677                                 if (root->ref_cows) {
2678                                         inode_sub_bytes(inode, item_end + 1 -
2679                                                         new_size);
2680                                 }
2681                                 size =
2682                                     btrfs_file_extent_calc_inline_size(size);
2683                                 ret = btrfs_truncate_item(trans, root, path,
2684                                                           size, 1);
2685                                 BUG_ON(ret);
2686                         } else if (root->ref_cows) {
2687                                 inode_sub_bytes(inode, item_end + 1 -
2688                                                 found_key.offset);
2689                         }
2690                 }
2691 delete:
2692                 if (del_item) {
2693                         if (!pending_del_nr) {
2694                                 /* no pending yet, add ourselves */
2695                                 pending_del_slot = path->slots[0];
2696                                 pending_del_nr = 1;
2697                         } else if (pending_del_nr &&
2698                                    path->slots[0] + 1 == pending_del_slot) {
2699                                 /* hop on the pending chunk */
2700                                 pending_del_nr++;
2701                                 pending_del_slot = path->slots[0];
2702                         } else {
2703                                 BUG();
2704                         }
2705                 } else {
2706                         break;
2707                 }
2708                 if (found_extent) {
2709                         btrfs_set_path_blocking(path);
2710                         ret = btrfs_free_extent(trans, root, extent_start,
2711                                                 extent_num_bytes,
2712                                                 leaf->start, root_owner,
2713                                                 root_gen, inode->i_ino, 0);
2714                         BUG_ON(ret);
2715                 }
2716 next:
2717                 if (path->slots[0] == 0) {
2718                         if (pending_del_nr)
2719                                 goto del_pending;
2720                         btrfs_release_path(root, path);
2721                         if (found_type == BTRFS_INODE_ITEM_KEY)
2722                                 break;
2723                         goto search_again;
2724                 }
2725
2726                 path->slots[0]--;
2727                 if (pending_del_nr &&
2728                     path->slots[0] + 1 != pending_del_slot) {
2729                         struct btrfs_key debug;
2730 del_pending:
2731                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
2732                                               pending_del_slot);
2733                         ret = btrfs_del_items(trans, root, path,
2734                                               pending_del_slot,
2735                                               pending_del_nr);
2736                         BUG_ON(ret);
2737                         pending_del_nr = 0;
2738                         btrfs_release_path(root, path);
2739                         if (found_type == BTRFS_INODE_ITEM_KEY)
2740                                 break;
2741                         goto search_again;
2742                 }
2743         }
2744         ret = 0;
2745 error:
2746         if (pending_del_nr) {
2747                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2748                                       pending_del_nr);
2749         }
2750         btrfs_free_path(path);
2751         inode->i_sb->s_dirt = 1;
2752         return ret;
2753 }
2754
2755 /*
2756  * taken from block_truncate_page, but does cow as it zeros out
2757  * any bytes left in the last page in the file.
2758  */
2759 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2760 {
2761         struct inode *inode = mapping->host;
2762         struct btrfs_root *root = BTRFS_I(inode)->root;
2763         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2764         struct btrfs_ordered_extent *ordered;
2765         char *kaddr;
2766         u32 blocksize = root->sectorsize;
2767         pgoff_t index = from >> PAGE_CACHE_SHIFT;
2768         unsigned offset = from & (PAGE_CACHE_SIZE-1);
2769         struct page *page;
2770         int ret = 0;
2771         u64 page_start;
2772         u64 page_end;
2773
2774         if ((offset & (blocksize - 1)) == 0)
2775                 goto out;
2776
2777         ret = -ENOMEM;
2778 again:
2779         page = grab_cache_page(mapping, index);
2780         if (!page)
2781                 goto out;
2782
2783         page_start = page_offset(page);
2784         page_end = page_start + PAGE_CACHE_SIZE - 1;
2785
2786         if (!PageUptodate(page)) {
2787                 ret = btrfs_readpage(NULL, page);
2788                 lock_page(page);
2789                 if (page->mapping != mapping) {
2790                         unlock_page(page);
2791                         page_cache_release(page);
2792                         goto again;
2793                 }
2794                 if (!PageUptodate(page)) {
2795                         ret = -EIO;
2796                         goto out_unlock;
2797                 }
2798         }
2799         wait_on_page_writeback(page);
2800
2801         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2802         set_page_extent_mapped(page);
2803
2804         ordered = btrfs_lookup_ordered_extent(inode, page_start);
2805         if (ordered) {
2806                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2807                 unlock_page(page);
2808                 page_cache_release(page);
2809                 btrfs_start_ordered_extent(inode, ordered, 1);
2810                 btrfs_put_ordered_extent(ordered);
2811                 goto again;
2812         }
2813
2814         btrfs_set_extent_delalloc(inode, page_start, page_end);
2815         ret = 0;
2816         if (offset != PAGE_CACHE_SIZE) {
2817                 kaddr = kmap(page);
2818                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2819                 flush_dcache_page(page);
2820                 kunmap(page);
2821         }
2822         ClearPageChecked(page);
2823         set_page_dirty(page);
2824         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2825
2826 out_unlock:
2827         unlock_page(page);
2828         page_cache_release(page);
2829 out:
2830         return ret;
2831 }
2832
2833 int btrfs_cont_expand(struct inode *inode, loff_t size)
2834 {
2835         struct btrfs_trans_handle *trans;
2836         struct btrfs_root *root = BTRFS_I(inode)->root;
2837         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2838         struct extent_map *em;
2839         u64 mask = root->sectorsize - 1;
2840         u64 hole_start = (inode->i_size + mask) & ~mask;
2841         u64 block_end = (size + mask) & ~mask;
2842         u64 last_byte;
2843         u64 cur_offset;
2844         u64 hole_size;
2845         int err;
2846
2847         if (size <= hole_start)
2848                 return 0;
2849
2850         err = btrfs_check_metadata_free_space(root);
2851         if (err)
2852                 return err;
2853
2854         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2855
2856         while (1) {
2857                 struct btrfs_ordered_extent *ordered;
2858                 btrfs_wait_ordered_range(inode, hole_start,
2859                                          block_end - hole_start);
2860                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2861                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2862                 if (!ordered)
2863                         break;
2864                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2865                 btrfs_put_ordered_extent(ordered);
2866         }
2867
2868         trans = btrfs_start_transaction(root, 1);
2869         btrfs_set_trans_block_group(trans, inode);
2870
2871         cur_offset = hole_start;
2872         while (1) {
2873                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2874                                 block_end - cur_offset, 0);
2875                 BUG_ON(IS_ERR(em) || !em);
2876                 last_byte = min(extent_map_end(em), block_end);
2877                 last_byte = (last_byte + mask) & ~mask;
2878                 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2879                         u64 hint_byte = 0;
2880                         hole_size = last_byte - cur_offset;
2881                         err = btrfs_drop_extents(trans, root, inode,
2882                                                  cur_offset,
2883                                                  cur_offset + hole_size,
2884                                                  block_end,
2885                                                  cur_offset, &hint_byte);
2886                         if (err)
2887                                 break;
2888                         err = btrfs_insert_file_extent(trans, root,
2889                                         inode->i_ino, cur_offset, 0,
2890                                         0, hole_size, 0, hole_size,
2891                                         0, 0, 0);
2892                         btrfs_drop_extent_cache(inode, hole_start,
2893                                         last_byte - 1, 0);
2894                 }
2895                 free_extent_map(em);
2896                 cur_offset = last_byte;
2897                 if (err || cur_offset >= block_end)
2898                         break;
2899         }
2900
2901         btrfs_end_transaction(trans, root);
2902         unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2903         return err;
2904 }
2905
2906 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2907 {
2908         struct inode *inode = dentry->d_inode;
2909         int err;
2910
2911         err = inode_change_ok(inode, attr);
2912         if (err)
2913                 return err;
2914
2915         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
2916                 if (attr->ia_size > inode->i_size) {
2917                         err = btrfs_cont_expand(inode, attr->ia_size);
2918                         if (err)
2919                                 return err;
2920                 } else if (inode->i_size > 0 &&
2921                            attr->ia_size == 0) {
2922
2923                         /* we're truncating a file that used to have good
2924                          * data down to zero.  Make sure it gets into
2925                          * the ordered flush list so that any new writes
2926                          * get down to disk quickly.
2927                          */
2928                         BTRFS_I(inode)->ordered_data_close = 1;
2929                 }
2930         }
2931
2932         err = inode_setattr(inode, attr);
2933
2934         if (!err && ((attr->ia_valid & ATTR_MODE)))
2935                 err = btrfs_acl_chmod(inode);
2936         return err;
2937 }
2938
2939 void btrfs_delete_inode(struct inode *inode)
2940 {
2941         struct btrfs_trans_handle *trans;
2942         struct btrfs_root *root = BTRFS_I(inode)->root;
2943         unsigned long nr;
2944         int ret;
2945
2946         truncate_inode_pages(&inode->i_data, 0);
2947         if (is_bad_inode(inode)) {
2948                 btrfs_orphan_del(NULL, inode);
2949                 goto no_delete;
2950         }
2951         btrfs_wait_ordered_range(inode, 0, (u64)-1);
2952
2953         btrfs_i_size_write(inode, 0);
2954         trans = btrfs_join_transaction(root, 1);
2955
2956         btrfs_set_trans_block_group(trans, inode);
2957         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
2958         if (ret) {
2959                 btrfs_orphan_del(NULL, inode);
2960                 goto no_delete_lock;
2961         }
2962
2963         btrfs_orphan_del(trans, inode);
2964
2965         nr = trans->blocks_used;
2966         clear_inode(inode);
2967
2968         btrfs_end_transaction(trans, root);
2969         btrfs_btree_balance_dirty(root, nr);
2970         return;
2971
2972 no_delete_lock:
2973         nr = trans->blocks_used;
2974         btrfs_end_transaction(trans, root);
2975         btrfs_btree_balance_dirty(root, nr);
2976 no_delete:
2977         clear_inode(inode);
2978 }
2979
2980 /*
2981  * this returns the key found in the dir entry in the location pointer.
2982  * If no dir entries were found, location->objectid is 0.
2983  */
2984 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
2985                                struct btrfs_key *location)
2986 {
2987         const char *name = dentry->d_name.name;
2988         int namelen = dentry->d_name.len;
2989         struct btrfs_dir_item *di;
2990         struct btrfs_path *path;
2991         struct btrfs_root *root = BTRFS_I(dir)->root;
2992         int ret = 0;
2993
2994         path = btrfs_alloc_path();
2995         BUG_ON(!path);
2996
2997         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
2998                                     namelen, 0);
2999         if (IS_ERR(di))
3000                 ret = PTR_ERR(di);
3001
3002         if (!di || IS_ERR(di))
3003                 goto out_err;
3004
3005         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3006 out:
3007         btrfs_free_path(path);
3008         return ret;
3009 out_err:
3010         location->objectid = 0;
3011         goto out;
3012 }
3013
3014 /*
3015  * when we hit a tree root in a directory, the btrfs part of the inode
3016  * needs to be changed to reflect the root directory of the tree root.  This
3017  * is kind of like crossing a mount point.
3018  */
3019 static int fixup_tree_root_location(struct btrfs_root *root,
3020                              struct btrfs_key *location,
3021                              struct btrfs_root **sub_root,
3022                              struct dentry *dentry)
3023 {
3024         struct btrfs_root_item *ri;
3025
3026         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
3027                 return 0;
3028         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
3029                 return 0;
3030
3031         *sub_root = btrfs_read_fs_root(root->fs_info, location,
3032                                         dentry->d_name.name,
3033                                         dentry->d_name.len);
3034         if (IS_ERR(*sub_root))
3035                 return PTR_ERR(*sub_root);
3036
3037         ri = &(*sub_root)->root_item;
3038         location->objectid = btrfs_root_dirid(ri);
3039         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3040         location->offset = 0;
3041
3042         return 0;
3043 }
3044
3045 static noinline void init_btrfs_i(struct inode *inode)
3046 {
3047         struct btrfs_inode *bi = BTRFS_I(inode);
3048
3049         bi->i_acl = NULL;
3050         bi->i_default_acl = NULL;
3051
3052         bi->generation = 0;
3053         bi->sequence = 0;
3054         bi->last_trans = 0;
3055         bi->logged_trans = 0;
3056         bi->delalloc_bytes = 0;
3057         bi->reserved_bytes = 0;
3058         bi->disk_i_size = 0;
3059         bi->flags = 0;
3060         bi->index_cnt = (u64)-1;
3061         bi->last_unlink_trans = 0;
3062         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3063         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3064                              inode->i_mapping, GFP_NOFS);
3065         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3066                              inode->i_mapping, GFP_NOFS);
3067         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3068         INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations);
3069         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3070         mutex_init(&BTRFS_I(inode)->extent_mutex);
3071         mutex_init(&BTRFS_I(inode)->log_mutex);
3072 }
3073
3074 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3075 {
3076         struct btrfs_iget_args *args = p;
3077         inode->i_ino = args->ino;
3078         init_btrfs_i(inode);
3079         BTRFS_I(inode)->root = args->root;
3080         btrfs_set_inode_space_info(args->root, inode);
3081         return 0;
3082 }
3083
3084 static int btrfs_find_actor(struct inode *inode, void *opaque)
3085 {
3086         struct btrfs_iget_args *args = opaque;
3087         return args->ino == inode->i_ino &&
3088                 args->root == BTRFS_I(inode)->root;
3089 }
3090
3091 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
3092                             struct btrfs_root *root, int wait)
3093 {
3094         struct inode *inode;
3095         struct btrfs_iget_args args;
3096         args.ino = objectid;
3097         args.root = root;
3098
3099         if (wait) {
3100                 inode = ilookup5(s, objectid, btrfs_find_actor,
3101                                  (void *)&args);
3102         } else {
3103                 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
3104                                         (void *)&args);
3105         }
3106         return inode;
3107 }
3108
3109 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
3110                                 struct btrfs_root *root)
3111 {
3112         struct inode *inode;
3113         struct btrfs_iget_args args;
3114         args.ino = objectid;
3115         args.root = root;
3116
3117         inode = iget5_locked(s, objectid, btrfs_find_actor,
3118                              btrfs_init_locked_inode,
3119                              (void *)&args);
3120         return inode;
3121 }
3122
3123 /* Get an inode object given its location and corresponding root.
3124  * Returns in *is_new if the inode was read from disk
3125  */
3126 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3127                          struct btrfs_root *root, int *is_new)
3128 {
3129         struct inode *inode;
3130
3131         inode = btrfs_iget_locked(s, location->objectid, root);
3132         if (!inode)
3133                 return ERR_PTR(-EACCES);
3134
3135         if (inode->i_state & I_NEW) {
3136                 BTRFS_I(inode)->root = root;
3137                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3138                 btrfs_read_locked_inode(inode);
3139                 unlock_new_inode(inode);
3140                 if (is_new)
3141                         *is_new = 1;
3142         } else {
3143                 if (is_new)
3144                         *is_new = 0;
3145         }
3146
3147         return inode;
3148 }
3149
3150 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3151 {
3152         struct inode *inode;
3153         struct btrfs_inode *bi = BTRFS_I(dir);
3154         struct btrfs_root *root = bi->root;
3155         struct btrfs_root *sub_root = root;
3156         struct btrfs_key location;
3157         int ret, new;
3158
3159         if (dentry->d_name.len > BTRFS_NAME_LEN)
3160                 return ERR_PTR(-ENAMETOOLONG);
3161
3162         ret = btrfs_inode_by_name(dir, dentry, &location);
3163
3164         if (ret < 0)
3165                 return ERR_PTR(ret);
3166
3167         inode = NULL;
3168         if (location.objectid) {
3169                 ret = fixup_tree_root_location(root, &location, &sub_root,
3170                                                 dentry);
3171                 if (ret < 0)
3172                         return ERR_PTR(ret);
3173                 if (ret > 0)
3174                         return ERR_PTR(-ENOENT);
3175                 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
3176                 if (IS_ERR(inode))
3177                         return ERR_CAST(inode);
3178         }
3179         return inode;
3180 }
3181
3182 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3183                                    struct nameidata *nd)
3184 {
3185         struct inode *inode;
3186
3187         if (dentry->d_name.len > BTRFS_NAME_LEN)
3188                 return ERR_PTR(-ENAMETOOLONG);
3189
3190         inode = btrfs_lookup_dentry(dir, dentry);
3191         if (IS_ERR(inode))
3192                 return ERR_CAST(inode);
3193
3194         return d_splice_alias(inode, dentry);
3195 }
3196
3197 static unsigned char btrfs_filetype_table[] = {
3198         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3199 };
3200
3201 static int btrfs_real_readdir(struct file *filp, void *dirent,
3202                               filldir_t filldir)
3203 {
3204         struct inode *inode = filp->f_dentry->d_inode;
3205         struct btrfs_root *root = BTRFS_I(inode)->root;
3206         struct btrfs_item *item;
3207         struct btrfs_dir_item *di;
3208         struct btrfs_key key;
3209         struct btrfs_key found_key;
3210         struct btrfs_path *path;
3211         int ret;
3212         u32 nritems;
3213         struct extent_buffer *leaf;
3214         int slot;
3215         int advance;
3216         unsigned char d_type;
3217         int over = 0;
3218         u32 di_cur;
3219         u32 di_total;
3220         u32 di_len;
3221         int key_type = BTRFS_DIR_INDEX_KEY;
3222         char tmp_name[32];
3223         char *name_ptr;
3224         int name_len;
3225
3226         /* FIXME, use a real flag for deciding about the key type */
3227         if (root->fs_info->tree_root == root)
3228                 key_type = BTRFS_DIR_ITEM_KEY;
3229
3230         /* special case for "." */
3231         if (filp->f_pos == 0) {
3232                 over = filldir(dirent, ".", 1,
3233                                1, inode->i_ino,
3234                                DT_DIR);
3235                 if (over)
3236                         return 0;
3237                 filp->f_pos = 1;
3238         }
3239         /* special case for .., just use the back ref */
3240         if (filp->f_pos == 1) {
3241                 u64 pino = parent_ino(filp->f_path.dentry);
3242                 over = filldir(dirent, "..", 2,
3243                                2, pino, DT_DIR);
3244                 if (over)
3245                         return 0;
3246                 filp->f_pos = 2;
3247         }
3248         path = btrfs_alloc_path();
3249         path->reada = 2;
3250
3251         btrfs_set_key_type(&key, key_type);
3252         key.offset = filp->f_pos;
3253         key.objectid = inode->i_ino;
3254
3255         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3256         if (ret < 0)
3257                 goto err;
3258         advance = 0;
3259
3260         while (1) {
3261                 leaf = path->nodes[0];
3262                 nritems = btrfs_header_nritems(leaf);
3263                 slot = path->slots[0];
3264                 if (advance || slot >= nritems) {
3265                         if (slot >= nritems - 1) {
3266                                 ret = btrfs_next_leaf(root, path);
3267                                 if (ret)
3268                                         break;
3269                                 leaf = path->nodes[0];
3270                                 nritems = btrfs_header_nritems(leaf);
3271                                 slot = path->slots[0];
3272                         } else {
3273                                 slot++;
3274                                 path->slots[0]++;
3275                         }
3276                 }
3277
3278                 advance = 1;
3279                 item = btrfs_item_nr(leaf, slot);
3280                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3281
3282                 if (found_key.objectid != key.objectid)
3283                         break;
3284                 if (btrfs_key_type(&found_key) != key_type)
3285                         break;
3286                 if (found_key.offset < filp->f_pos)
3287                         continue;
3288
3289                 filp->f_pos = found_key.offset;
3290
3291                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3292                 di_cur = 0;
3293                 di_total = btrfs_item_size(leaf, item);
3294
3295                 while (di_cur < di_total) {
3296                         struct btrfs_key location;
3297
3298                         name_len = btrfs_dir_name_len(leaf, di);
3299                         if (name_len <= sizeof(tmp_name)) {
3300                                 name_ptr = tmp_name;
3301                         } else {
3302                                 name_ptr = kmalloc(name_len, GFP_NOFS);
3303                                 if (!name_ptr) {
3304                                         ret = -ENOMEM;
3305                                         goto err;
3306                                 }
3307                         }
3308                         read_extent_buffer(leaf, name_ptr,
3309                                            (unsigned long)(di + 1), name_len);
3310
3311                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3312                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
3313
3314                         /* is this a reference to our own snapshot? If so
3315                          * skip it
3316                          */
3317                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
3318                             location.objectid == root->root_key.objectid) {
3319                                 over = 0;
3320                                 goto skip;
3321                         }
3322                         over = filldir(dirent, name_ptr, name_len,
3323                                        found_key.offset, location.objectid,
3324                                        d_type);
3325
3326 skip:
3327                         if (name_ptr != tmp_name)
3328                                 kfree(name_ptr);
3329
3330                         if (over)
3331                                 goto nopos;
3332                         di_len = btrfs_dir_name_len(leaf, di) +
3333                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
3334                         di_cur += di_len;
3335                         di = (struct btrfs_dir_item *)((char *)di + di_len);
3336                 }
3337         }
3338
3339         /* Reached end of directory/root. Bump pos past the last item. */
3340         if (key_type == BTRFS_DIR_INDEX_KEY)
3341                 filp->f_pos = INT_LIMIT(off_t);
3342         else
3343                 filp->f_pos++;
3344 nopos:
3345         ret = 0;
3346 err:
3347         btrfs_free_path(path);
3348         return ret;
3349 }
3350
3351 int btrfs_write_inode(struct inode *inode, int wait)
3352 {
3353         struct btrfs_root *root = BTRFS_I(inode)->root;
3354         struct btrfs_trans_handle *trans;
3355         int ret = 0;
3356
3357         if (root->fs_info->btree_inode == inode)
3358                 return 0;
3359
3360         if (wait) {
3361                 trans = btrfs_join_transaction(root, 1);
3362                 btrfs_set_trans_block_group(trans, inode);
3363                 ret = btrfs_commit_transaction(trans, root);
3364         }
3365         return ret;
3366 }
3367
3368 /*
3369  * This is somewhat expensive, updating the tree every time the
3370  * inode changes.  But, it is most likely to find the inode in cache.
3371  * FIXME, needs more benchmarking...there are no reasons other than performance
3372  * to keep or drop this code.
3373  */
3374 void btrfs_dirty_inode(struct inode *inode)
3375 {
3376         struct btrfs_root *root = BTRFS_I(inode)->root;
3377         struct btrfs_trans_handle *trans;
3378
3379         trans = btrfs_join_transaction(root, 1);
3380         btrfs_set_trans_block_group(trans, inode);
3381         btrfs_update_inode(trans, root, inode);
3382         btrfs_end_transaction(trans, root);
3383 }
3384
3385 /*
3386  * find the highest existing sequence number in a directory
3387  * and then set the in-memory index_cnt variable to reflect
3388  * free sequence numbers
3389  */
3390 static int btrfs_set_inode_index_count(struct inode *inode)
3391 {
3392         struct btrfs_root *root = BTRFS_I(inode)->root;
3393         struct btrfs_key key, found_key;
3394         struct btrfs_path *path;
3395         struct extent_buffer *leaf;
3396         int ret;
3397
3398         key.objectid = inode->i_ino;
3399         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3400         key.offset = (u64)-1;
3401
3402         path = btrfs_alloc_path();
3403         if (!path)
3404                 return -ENOMEM;
3405
3406         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3407         if (ret < 0)
3408                 goto out;
3409         /* FIXME: we should be able to handle this */
3410         if (ret == 0)
3411                 goto out;
3412         ret = 0;
3413
3414         /*
3415          * MAGIC NUMBER EXPLANATION:
3416          * since we search a directory based on f_pos we have to start at 2
3417          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3418          * else has to start at 2
3419          */
3420         if (path->slots[0] == 0) {
3421                 BTRFS_I(inode)->index_cnt = 2;
3422                 goto out;
3423         }
3424
3425         path->slots[0]--;
3426
3427         leaf = path->nodes[0];
3428         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3429
3430         if (found_key.objectid != inode->i_ino ||
3431             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3432                 BTRFS_I(inode)->index_cnt = 2;
3433                 goto out;
3434         }
3435
3436         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3437 out:
3438         btrfs_free_path(path);
3439         return ret;
3440 }
3441
3442 /*
3443  * helper to find a free sequence number in a given directory.  This current
3444  * code is very simple, later versions will do smarter things in the btree
3445  */
3446 int btrfs_set_inode_index(struct inode *dir, u64 *index)
3447 {
3448         int ret = 0;
3449
3450         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3451                 ret = btrfs_set_inode_index_count(dir);
3452                 if (ret)
3453                         return ret;
3454         }
3455
3456         *index = BTRFS_I(dir)->index_cnt;
3457         BTRFS_I(dir)->index_cnt++;
3458
3459         return ret;
3460 }
3461
3462 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3463                                      struct btrfs_root *root,
3464                                      struct inode *dir,
3465                                      const char *name, int name_len,
3466                                      u64 ref_objectid, u64 objectid,
3467                                      u64 alloc_hint, int mode, u64 *index)
3468 {
3469         struct inode *inode;
3470         struct btrfs_inode_item *inode_item;
3471         struct btrfs_key *location;
3472         struct btrfs_path *path;
3473         struct btrfs_inode_ref *ref;
3474         struct btrfs_key key[2];
3475         u32 sizes[2];
3476         unsigned long ptr;
3477         int ret;
3478         int owner;
3479
3480         path = btrfs_alloc_path();
3481         BUG_ON(!path);
3482
3483         inode = new_inode(root->fs_info->sb);
3484         if (!inode)
3485                 return ERR_PTR(-ENOMEM);
3486
3487         if (dir) {
3488                 ret = btrfs_set_inode_index(dir, index);
3489                 if (ret) {
3490                         iput(inode);
3491                         return ERR_PTR(ret);
3492                 }
3493         }
3494         /*
3495          * index_cnt is ignored for everything but a dir,
3496          * btrfs_get_inode_index_count has an explanation for the magic
3497          * number
3498          */
3499         init_btrfs_i(inode);
3500         BTRFS_I(inode)->index_cnt = 2;
3501         BTRFS_I(inode)->root = root;
3502         BTRFS_I(inode)->generation = trans->transid;
3503         btrfs_set_inode_space_info(root, inode);
3504
3505         if (mode & S_IFDIR)
3506                 owner = 0;
3507         else
3508                 owner = 1;
3509         BTRFS_I(inode)->block_group =
3510                         btrfs_find_block_group(root, 0, alloc_hint, owner);
3511         if ((mode & S_IFREG)) {
3512                 if (btrfs_test_opt(root, NODATASUM))
3513                         btrfs_set_flag(inode, NODATASUM);
3514                 if (btrfs_test_opt(root, NODATACOW))
3515                         btrfs_set_flag(inode, NODATACOW);
3516         }
3517
3518         key[0].objectid = objectid;
3519         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3520         key[0].offset = 0;
3521
3522         key[1].objectid = objectid;
3523         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3524         key[1].offset = ref_objectid;
3525
3526         sizes[0] = sizeof(struct btrfs_inode_item);
3527         sizes[1] = name_len + sizeof(*ref);
3528
3529         path->leave_spinning = 1;
3530         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3531         if (ret != 0)
3532                 goto fail;
3533
3534         if (objectid > root->highest_inode)
3535                 root->highest_inode = objectid;
3536
3537         inode->i_uid = current_fsuid();
3538
3539         if (dir && (dir->i_mode & S_ISGID)) {
3540                 inode->i_gid = dir->i_gid;
3541                 if (S_ISDIR(mode))
3542                         mode |= S_ISGID;
3543         } else
3544                 inode->i_gid = current_fsgid();
3545
3546         inode->i_mode = mode;
3547         inode->i_ino = objectid;
3548         inode_set_bytes(inode, 0);
3549         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3550         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3551                                   struct btrfs_inode_item);
3552         fill_inode_item(trans, path->nodes[0], inode_item, inode);
3553
3554         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3555                              struct btrfs_inode_ref);
3556         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
3557         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
3558         ptr = (unsigned long)(ref + 1);
3559         write_extent_buffer(path->nodes[0], name, ptr, name_len);
3560
3561         btrfs_mark_buffer_dirty(path->nodes[0]);
3562         btrfs_free_path(path);
3563
3564         location = &BTRFS_I(inode)->location;
3565         location->objectid = objectid;
3566         location->offset = 0;
3567         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3568
3569         insert_inode_hash(inode);
3570         return inode;
3571 fail:
3572         if (dir)
3573                 BTRFS_I(dir)->index_cnt--;
3574         btrfs_free_path(path);
3575         iput(inode);
3576         return ERR_PTR(ret);
3577 }
3578
3579 static inline u8 btrfs_inode_type(struct inode *inode)
3580 {
3581         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3582 }
3583
3584 /*
3585  * utility function to add 'inode' into 'parent_inode' with
3586  * a give name and a given sequence number.
3587  * if 'add_backref' is true, also insert a backref from the
3588  * inode to the parent directory.
3589  */
3590 int btrfs_add_link(struct btrfs_trans_handle *trans,
3591                    struct inode *parent_inode, struct inode *inode,
3592                    const char *name, int name_len, int add_backref, u64 index)
3593 {
3594         int ret;
3595         struct btrfs_key key;
3596         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
3597
3598         key.objectid = inode->i_ino;
3599         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3600         key.offset = 0;
3601
3602         ret = btrfs_insert_dir_item(trans, root, name, name_len,
3603                                     parent_inode->i_ino,
3604                                     &key, btrfs_inode_type(inode),
3605                                     index);
3606         if (ret == 0) {
3607                 if (add_backref) {
3608                         ret = btrfs_insert_inode_ref(trans, root,
3609                                                      name, name_len,
3610                                                      inode->i_ino,
3611                                                      parent_inode->i_ino,
3612                                                      index);
3613                 }
3614                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
3615                                    name_len * 2);
3616                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
3617                 ret = btrfs_update_inode(trans, root, parent_inode);
3618         }
3619         return ret;
3620 }
3621
3622 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
3623                             struct dentry *dentry, struct inode *inode,
3624                             int backref, u64 index)
3625 {
3626         int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3627                                  inode, dentry->d_name.name,
3628                                  dentry->d_name.len, backref, index);
3629         if (!err) {
3630                 d_instantiate(dentry, inode);
3631                 return 0;
3632         }
3633         if (err > 0)
3634                 err = -EEXIST;
3635         return err;
3636 }
3637
3638 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3639                         int mode, dev_t rdev)
3640 {
3641         struct btrfs_trans_handle *trans;
3642         struct btrfs_root *root = BTRFS_I(dir)->root;
3643         struct inode *inode = NULL;
3644         int err;
3645         int drop_inode = 0;
3646         u64 objectid;
3647         unsigned long nr = 0;
3648         u64 index = 0;
3649
3650         if (!new_valid_dev(rdev))
3651                 return -EINVAL;
3652
3653         err = btrfs_check_metadata_free_space(root);
3654         if (err)
3655                 goto fail;
3656
3657         trans = btrfs_start_transaction(root, 1);
3658         btrfs_set_trans_block_group(trans, dir);
3659
3660         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3661         if (err) {
3662                 err = -ENOSPC;
3663                 goto out_unlock;
3664         }
3665
3666         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3667                                 dentry->d_name.len,
3668                                 dentry->d_parent->d_inode->i_ino, objectid,
3669                                 BTRFS_I(dir)->block_group, mode, &index);
3670         err = PTR_ERR(inode);
3671         if (IS_ERR(inode))
3672                 goto out_unlock;
3673
3674         err = btrfs_init_inode_security(inode, dir);
3675         if (err) {
3676                 drop_inode = 1;
3677                 goto out_unlock;
3678         }
3679
3680         btrfs_set_trans_block_group(trans, inode);
3681         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3682         if (err)
3683                 drop_inode = 1;
3684         else {
3685                 inode->i_op = &btrfs_special_inode_operations;
3686                 init_special_inode(inode, inode->i_mode, rdev);
3687                 btrfs_update_inode(trans, root, inode);
3688         }
3689         dir->i_sb->s_dirt = 1;
3690         btrfs_update_inode_block_group(trans, inode);
3691         btrfs_update_inode_block_group(trans, dir);
3692 out_unlock:
3693         nr = trans->blocks_used;
3694         btrfs_end_transaction_throttle(trans, root);
3695 fail:
3696         if (drop_inode) {
3697                 inode_dec_link_count(inode);
3698                 iput(inode);
3699         }
3700         btrfs_btree_balance_dirty(root, nr);
3701         return err;
3702 }
3703
3704 static int btrfs_create(struct inode *dir, struct dentry *dentry,
3705                         int mode, struct nameidata *nd)
3706 {
3707         struct btrfs_trans_handle *trans;
3708         struct btrfs_root *root = BTRFS_I(dir)->root;
3709         struct inode *inode = NULL;
3710         int err;
3711         int drop_inode = 0;
3712         unsigned long nr = 0;
3713         u64 objectid;
3714         u64 index = 0;
3715
3716         err = btrfs_check_metadata_free_space(root);
3717         if (err)
3718                 goto fail;
3719         trans = btrfs_start_transaction(root, 1);
3720         btrfs_set_trans_block_group(trans, dir);
3721
3722         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3723         if (err) {
3724                 err = -ENOSPC;
3725                 goto out_unlock;
3726         }
3727
3728         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3729                                 dentry->d_name.len,
3730                                 dentry->d_parent->d_inode->i_ino,
3731                                 objectid, BTRFS_I(dir)->block_group, mode,
3732                                 &index);
3733         err = PTR_ERR(inode);
3734         if (IS_ERR(inode))
3735                 goto out_unlock;
3736
3737         err = btrfs_init_inode_security(inode, dir);
3738         if (err) {
3739                 drop_inode = 1;
3740                 goto out_unlock;
3741         }
3742
3743         btrfs_set_trans_block_group(trans, inode);
3744         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3745         if (err)
3746                 drop_inode = 1;
3747         else {
3748                 inode->i_mapping->a_ops = &btrfs_aops;
3749                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3750                 inode->i_fop = &btrfs_file_operations;
3751                 inode->i_op = &btrfs_file_inode_operations;
3752                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3753         }
3754         dir->i_sb->s_dirt = 1;
3755         btrfs_update_inode_block_group(trans, inode);
3756         btrfs_update_inode_block_group(trans, dir);
3757 out_unlock:
3758         nr = trans->blocks_used;
3759         btrfs_end_transaction_throttle(trans, root);
3760 fail:
3761         if (drop_inode) {
3762                 inode_dec_link_count(inode);
3763                 iput(inode);
3764         }
3765         btrfs_btree_balance_dirty(root, nr);
3766         return err;
3767 }
3768
3769 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3770                       struct dentry *dentry)
3771 {
3772         struct btrfs_trans_handle *trans;
3773         struct btrfs_root *root = BTRFS_I(dir)->root;
3774         struct inode *inode = old_dentry->d_inode;
3775         u64 index;
3776         unsigned long nr = 0;
3777         int err;
3778         int drop_inode = 0;
3779
3780         if (inode->i_nlink == 0)
3781                 return -ENOENT;
3782
3783         btrfs_inc_nlink(inode);
3784         err = btrfs_check_metadata_free_space(root);
3785         if (err)
3786                 goto fail;
3787         err = btrfs_set_inode_index(dir, &index);
3788         if (err)
3789                 goto fail;
3790
3791         trans = btrfs_start_transaction(root, 1);
3792
3793         btrfs_set_trans_block_group(trans, dir);
3794         atomic_inc(&inode->i_count);
3795
3796         err = btrfs_add_nondir(trans, dentry, inode, 1, index);
3797
3798         if (err)
3799                 drop_inode = 1;
3800
3801         dir->i_sb->s_dirt = 1;
3802         btrfs_update_inode_block_group(trans, dir);
3803         err = btrfs_update_inode(trans, root, inode);
3804
3805         if (err)
3806                 drop_inode = 1;
3807
3808         nr = trans->blocks_used;
3809
3810         btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
3811         btrfs_end_transaction_throttle(trans, root);
3812 fail:
3813         if (drop_inode) {
3814                 inode_dec_link_count(inode);
3815                 iput(inode);
3816         }
3817         btrfs_btree_balance_dirty(root, nr);
3818         return err;
3819 }
3820
3821 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3822 {
3823         struct inode *inode = NULL;
3824         struct btrfs_trans_handle *trans;
3825         struct btrfs_root *root = BTRFS_I(dir)->root;
3826         int err = 0;
3827         int drop_on_err = 0;
3828         u64 objectid = 0;
3829         u64 index = 0;
3830         unsigned long nr = 1;
3831
3832         err = btrfs_check_metadata_free_space(root);
3833         if (err)
3834                 goto out_unlock;
3835
3836         trans = btrfs_start_transaction(root, 1);
3837         btrfs_set_trans_block_group(trans, dir);
3838
3839         if (IS_ERR(trans)) {
3840                 err = PTR_ERR(trans);
3841                 goto out_unlock;
3842         }
3843
3844         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3845         if (err) {
3846                 err = -ENOSPC;
3847                 goto out_unlock;
3848         }
3849
3850         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3851                                 dentry->d_name.len,
3852                                 dentry->d_parent->d_inode->i_ino, objectid,
3853                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3854                                 &index);
3855         if (IS_ERR(inode)) {
3856                 err = PTR_ERR(inode);
3857                 goto out_fail;
3858         }
3859
3860         drop_on_err = 1;
3861
3862         err = btrfs_init_inode_security(inode, dir);
3863         if (err)
3864                 goto out_fail;
3865
3866         inode->i_op = &btrfs_dir_inode_operations;
3867         inode->i_fop = &btrfs_dir_file_operations;
3868         btrfs_set_trans_block_group(trans, inode);
3869
3870         btrfs_i_size_write(inode, 0);
3871         err = btrfs_update_inode(trans, root, inode);
3872         if (err)
3873                 goto out_fail;
3874
3875         err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3876                                  inode, dentry->d_name.name,
3877                                  dentry->d_name.len, 0, index);
3878         if (err)
3879                 goto out_fail;
3880
3881         d_instantiate(dentry, inode);
3882         drop_on_err = 0;
3883         dir->i_sb->s_dirt = 1;
3884         btrfs_update_inode_block_group(trans, inode);
3885         btrfs_update_inode_block_group(trans, dir);
3886
3887 out_fail:
3888         nr = trans->blocks_used;
3889         btrfs_end_transaction_throttle(trans, root);
3890
3891 out_unlock:
3892         if (drop_on_err)
3893                 iput(inode);
3894         btrfs_btree_balance_dirty(root, nr);
3895         return err;
3896 }
3897
3898 /* helper for btfs_get_extent.  Given an existing extent in the tree,
3899  * and an extent that you want to insert, deal with overlap and insert
3900  * the new extent into the tree.
3901  */
3902 static int merge_extent_mapping(struct extent_map_tree *em_tree,
3903                                 struct extent_map *existing,
3904                                 struct extent_map *em,
3905                                 u64 map_start, u64 map_len)
3906 {
3907         u64 start_diff;
3908
3909         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3910         start_diff = map_start - em->start;
3911         em->start = map_start;
3912         em->len = map_len;
3913         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3914             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3915                 em->block_start += start_diff;
3916                 em->block_len -= start_diff;
3917         }
3918         return add_extent_mapping(em_tree, em);
3919 }
3920
3921 static noinline int uncompress_inline(struct btrfs_path *path,
3922                                       struct inode *inode, struct page *page,
3923                                       size_t pg_offset, u64 extent_offset,
3924                                       struct btrfs_file_extent_item *item)
3925 {
3926         int ret;
3927         struct extent_buffer *leaf = path->nodes[0];
3928         char *tmp;
3929         size_t max_size;
3930         unsigned long inline_size;
3931         unsigned long ptr;
3932
3933         WARN_ON(pg_offset != 0);
3934         max_size = btrfs_file_extent_ram_bytes(leaf, item);
3935         inline_size = btrfs_file_extent_inline_item_len(leaf,
3936                                         btrfs_item_nr(leaf, path->slots[0]));
3937         tmp = kmalloc(inline_size, GFP_NOFS);
3938         ptr = btrfs_file_extent_inline_start(item);
3939
3940         read_extent_buffer(leaf, tmp, ptr, inline_size);
3941
3942         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
3943         ret = btrfs_zlib_decompress(tmp, page, extent_offset,
3944                                     inline_size, max_size);
3945         if (ret) {
3946                 char *kaddr = kmap_atomic(page, KM_USER0);
3947                 unsigned long copy_size = min_t(u64,
3948                                   PAGE_CACHE_SIZE - pg_offset,
3949                                   max_size - extent_offset);
3950                 memset(kaddr + pg_offset, 0, copy_size);
3951                 kunmap_atomic(kaddr, KM_USER0);
3952         }
3953         kfree(tmp);
3954         return 0;
3955 }
3956
3957 /*
3958  * a bit scary, this does extent mapping from logical file offset to the disk.
3959  * the ugly parts come from merging extents from the disk with the in-ram
3960  * representation.  This gets more complex because of the data=ordered code,
3961  * where the in-ram extents might be locked pending data=ordered completion.
3962  *
3963  * This also copies inline extents directly into the page.
3964  */
3965
3966 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
3967                                     size_t pg_offset, u64 start, u64 len,
3968                                     int create)
3969 {
3970         int ret;
3971         int err = 0;
3972         u64 bytenr;
3973         u64 extent_start = 0;
3974         u64 extent_end = 0;
3975         u64 objectid = inode->i_ino;
3976         u32 found_type;
3977         struct btrfs_path *path = NULL;
3978         struct btrfs_root *root = BTRFS_I(inode)->root;
3979         struct btrfs_file_extent_item *item;
3980         struct extent_buffer *leaf;
3981         struct btrfs_key found_key;
3982         struct extent_map *em = NULL;
3983         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3984         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3985         struct btrfs_trans_handle *trans = NULL;
3986         int compressed;
3987
3988 again:
3989         spin_lock(&em_tree->lock);
3990         em = lookup_extent_mapping(em_tree, start, len);
3991         if (em)
3992                 em->bdev = root->fs_info->fs_devices->latest_bdev;
3993         spin_unlock(&em_tree->lock);
3994
3995         if (em) {
3996                 if (em->start > start || em->start + em->len <= start)
3997                         free_extent_map(em);
3998                 else if (em->block_start == EXTENT_MAP_INLINE && page)
3999                         free_extent_map(em);
4000                 else
4001                         goto out;
4002         }
4003         em = alloc_extent_map(GFP_NOFS);
4004         if (!em) {
4005                 err = -ENOMEM;
4006                 goto out;
4007         }
4008         em->bdev = root->fs_info->fs_devices->latest_bdev;
4009         em->start = EXTENT_MAP_HOLE;
4010         em->orig_start = EXTENT_MAP_HOLE;
4011         em->len = (u64)-1;
4012         em->block_len = (u64)-1;
4013
4014         if (!path) {
4015                 path = btrfs_alloc_path();
4016                 BUG_ON(!path);
4017         }
4018
4019         ret = btrfs_lookup_file_extent(trans, root, path,
4020                                        objectid, start, trans != NULL);
4021         if (ret < 0) {
4022                 err = ret;
4023                 goto out;
4024         }
4025
4026         if (ret != 0) {
4027                 if (path->slots[0] == 0)
4028                         goto not_found;
4029                 path->slots[0]--;
4030         }
4031
4032         leaf = path->nodes[0];
4033         item = btrfs_item_ptr(leaf, path->slots[0],
4034                               struct btrfs_file_extent_item);
4035         /* are we inside the extent that was found? */
4036         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4037         found_type = btrfs_key_type(&found_key);
4038         if (found_key.objectid != objectid ||
4039             found_type != BTRFS_EXTENT_DATA_KEY) {
4040                 goto not_found;
4041         }
4042
4043         found_type = btrfs_file_extent_type(leaf, item);
4044         extent_start = found_key.offset;
4045         compressed = btrfs_file_extent_compression(leaf, item);
4046         if (found_type == BTRFS_FILE_EXTENT_REG ||
4047             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4048                 extent_end = extent_start +
4049                        btrfs_file_extent_num_bytes(leaf, item);
4050         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4051                 size_t size;
4052                 size = btrfs_file_extent_inline_len(leaf, item);
4053                 extent_end = (extent_start + size + root->sectorsize - 1) &
4054                         ~((u64)root->sectorsize - 1);
4055         }
4056
4057         if (start >= extent_end) {
4058                 path->slots[0]++;
4059                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4060                         ret = btrfs_next_leaf(root, path);
4061                         if (ret < 0) {
4062                                 err = ret;
4063                                 goto out;
4064                         }
4065                         if (ret > 0)
4066                                 goto not_found;
4067                         leaf = path->nodes[0];
4068                 }
4069                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4070                 if (found_key.objectid != objectid ||
4071                     found_key.type != BTRFS_EXTENT_DATA_KEY)
4072                         goto not_found;
4073                 if (start + len <= found_key.offset)
4074                         goto not_found;
4075                 em->start = start;
4076                 em->len = found_key.offset - start;
4077                 goto not_found_em;
4078         }
4079
4080         if (found_type == BTRFS_FILE_EXTENT_REG ||
4081             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4082                 em->start = extent_start;
4083                 em->len = extent_end - extent_start;
4084                 em->orig_start = extent_start -
4085                                  btrfs_file_extent_offset(leaf, item);
4086                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4087                 if (bytenr == 0) {
4088                         em->block_start = EXTENT_MAP_HOLE;
4089                         goto insert;
4090                 }
4091                 if (compressed) {
4092                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4093                         em->block_start = bytenr;
4094                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4095                                                                          item);
4096                 } else {
4097                         bytenr += btrfs_file_extent_offset(leaf, item);
4098                         em->block_start = bytenr;
4099                         em->block_len = em->len;
4100                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4101                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4102                 }
4103                 goto insert;
4104         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4105                 unsigned long ptr;
4106                 char *map;
4107                 size_t size;
4108                 size_t extent_offset;
4109                 size_t copy_size;
4110
4111                 em->block_start = EXTENT_MAP_INLINE;
4112                 if (!page || create) {
4113                         em->start = extent_start;
4114                         em->len = extent_end - extent_start;
4115                         goto out;
4116                 }
4117
4118                 size = btrfs_file_extent_inline_len(leaf, item);
4119                 extent_offset = page_offset(page) + pg_offset - extent_start;
4120                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4121                                 size - extent_offset);
4122                 em->start = extent_start + extent_offset;
4123                 em->len = (copy_size + root->sectorsize - 1) &
4124                         ~((u64)root->sectorsize - 1);
4125                 em->orig_start = EXTENT_MAP_INLINE;
4126                 if (compressed)
4127                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4128                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4129                 if (create == 0 && !PageUptodate(page)) {
4130                         if (btrfs_file_extent_compression(leaf, item) ==
4131                             BTRFS_COMPRESS_ZLIB) {
4132                                 ret = uncompress_inline(path, inode, page,
4133                                                         pg_offset,
4134                                                         extent_offset, item);
4135                                 BUG_ON(ret);
4136                         } else {
4137                                 map = kmap(page);
4138                                 read_extent_buffer(leaf, map + pg_offset, ptr,
4139                                                    copy_size);
4140                                 kunmap(page);
4141                         }
4142                         flush_dcache_page(page);
4143                 } else if (create && PageUptodate(page)) {
4144                         if (!trans) {
4145                                 kunmap(page);
4146                                 free_extent_map(em);
4147                                 em = NULL;
4148                                 btrfs_release_path(root, path);
4149                                 trans = btrfs_join_transaction(root, 1);
4150                                 goto again;
4151                         }
4152                         map = kmap(page);
4153                         write_extent_buffer(leaf, map + pg_offset, ptr,
4154                                             copy_size);
4155                         kunmap(page);
4156                         btrfs_mark_buffer_dirty(leaf);
4157                 }
4158                 set_extent_uptodate(io_tree, em->start,
4159                                     extent_map_end(em) - 1, GFP_NOFS);
4160                 goto insert;
4161         } else {
4162                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
4163                 WARN_ON(1);
4164         }
4165 not_found:
4166         em->start = start;
4167         em->len = len;
4168 not_found_em:
4169         em->block_start = EXTENT_MAP_HOLE;
4170         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4171 insert:
4172         btrfs_release_path(root, path);
4173         if (em->start > start || extent_map_end(em) <= start) {
4174                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4175                        "[%llu %llu]\n", (unsigned long long)em->start,
4176                        (unsigned long long)em->len,
4177                        (unsigned long long)start,
4178                        (unsigned long long)len);
4179                 err = -EIO;
4180                 goto out;
4181         }
4182
4183         err = 0;
4184         spin_lock(&em_tree->lock);
4185         ret = add_extent_mapping(em_tree, em);
4186         /* it is possible that someone inserted the extent into the tree
4187          * while we had the lock dropped.  It is also possible that
4188          * an overlapping map exists in the tree
4189          */
4190         if (ret == -EEXIST) {
4191                 struct extent_map *existing;
4192
4193                 ret = 0;
4194
4195                 existing = lookup_extent_mapping(em_tree, start, len);
4196                 if (existing && (existing->start > start ||
4197                     existing->start + existing->len <= start)) {
4198                         free_extent_map(existing);
4199                         existing = NULL;
4200                 }
4201                 if (!existing) {
4202                         existing = lookup_extent_mapping(em_tree, em->start,
4203                                                          em->len);
4204                         if (existing) {
4205                                 err = merge_extent_mapping(em_tree, existing,
4206                                                            em, start,
4207                                                            root->sectorsize);
4208                                 free_extent_map(existing);
4209                                 if (err) {
4210                                         free_extent_map(em);
4211                                         em = NULL;
4212                                 }
4213                         } else {
4214                                 err = -EIO;
4215                                 free_extent_map(em);
4216                                 em = NULL;
4217                         }
4218                 } else {
4219                         free_extent_map(em);
4220                         em = existing;
4221                         err = 0;
4222                 }
4223         }
4224         spin_unlock(&em_tree->lock);
4225 out:
4226         if (path)
4227                 btrfs_free_path(path);
4228         if (trans) {
4229                 ret = btrfs_end_transaction(trans, root);
4230                 if (!err)
4231                         err = ret;
4232         }
4233         if (err) {
4234                 free_extent_map(em);
4235                 WARN_ON(1);
4236                 return ERR_PTR(err);
4237         }
4238         return em;
4239 }
4240
4241 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4242                         const struct iovec *iov, loff_t offset,
4243                         unsigned long nr_segs)
4244 {
4245         return -EINVAL;
4246 }
4247
4248 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4249                 __u64 start, __u64 len)
4250 {
4251         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
4252 }
4253
4254 int btrfs_readpage(struct file *file, struct page *page)
4255 {
4256         struct extent_io_tree *tree;
4257         tree = &BTRFS_I(page->mapping->host)->io_tree;
4258         return extent_read_full_page(tree, page, btrfs_get_extent);
4259 }
4260
4261 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4262 {
4263         struct extent_io_tree *tree;
4264
4265
4266         if (current->flags & PF_MEMALLOC) {
4267                 redirty_page_for_writepage(wbc, page);
4268                 unlock_page(page);
4269                 return 0;
4270         }
4271         tree = &BTRFS_I(page->mapping->host)->io_tree;
4272         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4273 }
4274
4275 int btrfs_writepages(struct address_space *mapping,
4276                      struct writeback_control *wbc)
4277 {
4278         struct extent_io_tree *tree;
4279
4280         tree = &BTRFS_I(mapping->host)->io_tree;
4281         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4282 }
4283
4284 static int
4285 btrfs_readpages(struct file *file, struct address_space *mapping,
4286                 struct list_head *pages, unsigned nr_pages)
4287 {
4288         struct extent_io_tree *tree;
4289         tree = &BTRFS_I(mapping->host)->io_tree;
4290         return extent_readpages(tree, mapping, pages, nr_pages,
4291                                 btrfs_get_extent);
4292 }
4293 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4294 {
4295         struct extent_io_tree *tree;
4296         struct extent_map_tree *map;
4297         int ret;
4298
4299         tree = &BTRFS_I(page->mapping->host)->io_tree;
4300         map = &BTRFS_I(page->mapping->host)->extent_tree;
4301         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4302         if (ret == 1) {
4303                 ClearPagePrivate(page);
4304                 set_page_private(page, 0);
4305                 page_cache_release(page);
4306         }
4307         return ret;
4308 }
4309
4310 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4311 {
4312         if (PageWriteback(page) || PageDirty(page))
4313                 return 0;
4314         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
4315 }
4316
4317 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4318 {
4319         struct extent_io_tree *tree;
4320         struct btrfs_ordered_extent *ordered;
4321         u64 page_start = page_offset(page);
4322         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4323
4324         wait_on_page_writeback(page);
4325         tree = &BTRFS_I(page->mapping->host)->io_tree;
4326         if (offset) {
4327                 btrfs_releasepage(page, GFP_NOFS);
4328                 return;
4329         }
4330
4331         lock_extent(tree, page_start, page_end, GFP_NOFS);
4332         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4333                                            page_offset(page));
4334         if (ordered) {
4335                 /*
4336                  * IO on this page will never be started, so we need
4337                  * to account for any ordered extents now
4338                  */
4339                 clear_extent_bit(tree, page_start, page_end,
4340                                  EXTENT_DIRTY | EXTENT_DELALLOC |
4341                                  EXTENT_LOCKED, 1, 0, GFP_NOFS);
4342                 btrfs_finish_ordered_io(page->mapping->host,
4343                                         page_start, page_end);
4344                 btrfs_put_ordered_extent(ordered);
4345                 lock_extent(tree, page_start, page_end, GFP_NOFS);
4346         }
4347         clear_extent_bit(tree, page_start, page_end,
4348                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4349                  EXTENT_ORDERED,
4350                  1, 1, GFP_NOFS);
4351         __btrfs_releasepage(page, GFP_NOFS);
4352
4353         ClearPageChecked(page);
4354         if (PagePrivate(page)) {
4355                 ClearPagePrivate(page);
4356                 set_page_private(page, 0);
4357                 page_cache_release(page);
4358         }
4359 }
4360
4361 /*
4362  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4363  * called from a page fault handler when a page is first dirtied. Hence we must
4364  * be careful to check for EOF conditions here. We set the page up correctly
4365  * for a written page which means we get ENOSPC checking when writing into
4366  * holes and correct delalloc and unwritten extent mapping on filesystems that
4367  * support these features.
4368  *
4369  * We are not allowed to take the i_mutex here so we have to play games to
4370  * protect against truncate races as the page could now be beyond EOF.  Because
4371  * vmtruncate() writes the inode size before removing pages, once we have the
4372  * page lock we can determine safely if the page is beyond EOF. If it is not
4373  * beyond EOF, then the page is guaranteed safe against truncation until we
4374  * unlock the page.
4375  */
4376 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4377 {
4378         struct page *page = vmf->page;
4379         struct inode *inode = fdentry(vma->vm_file)->d_inode;
4380         struct btrfs_root *root = BTRFS_I(inode)->root;
4381         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4382         struct btrfs_ordered_extent *ordered;
4383         char *kaddr;
4384         unsigned long zero_start;
4385         loff_t size;
4386         int ret;
4387         u64 page_start;
4388         u64 page_end;
4389
4390         ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
4391         if (ret) {
4392                 if (ret == -ENOMEM)
4393                         ret = VM_FAULT_OOM;
4394                 else /* -ENOSPC, -EIO, etc */
4395                         ret = VM_FAULT_SIGBUS;
4396                 goto out;
4397         }
4398
4399         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
4400 again:
4401         lock_page(page);
4402         size = i_size_read(inode);
4403         page_start = page_offset(page);
4404         page_end = page_start + PAGE_CACHE_SIZE - 1;
4405
4406         if ((page->mapping != inode->i_mapping) ||
4407             (page_start >= size)) {
4408                 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
4409                 /* page got truncated out from underneath us */
4410                 goto out_unlock;
4411         }
4412         wait_on_page_writeback(page);
4413
4414         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4415         set_page_extent_mapped(page);
4416
4417         /*
4418          * we can't set the delalloc bits if there are pending ordered
4419          * extents.  Drop our locks and wait for them to finish
4420          */
4421         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4422         if (ordered) {
4423                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4424                 unlock_page(page);
4425                 btrfs_start_ordered_extent(inode, ordered, 1);
4426                 btrfs_put_ordered_extent(ordered);
4427                 goto again;
4428         }
4429
4430         btrfs_set_extent_delalloc(inode, page_start, page_end);
4431         ret = 0;
4432
4433         /* page is wholly or partially inside EOF */
4434         if (page_start + PAGE_CACHE_SIZE > size)
4435                 zero_start = size & ~PAGE_CACHE_MASK;
4436         else
4437                 zero_start = PAGE_CACHE_SIZE;
4438
4439         if (zero_start != PAGE_CACHE_SIZE) {
4440                 kaddr = kmap(page);
4441                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4442                 flush_dcache_page(page);
4443                 kunmap(page);
4444         }
4445         ClearPageChecked(page);
4446         set_page_dirty(page);
4447
4448         BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
4449         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4450
4451 out_unlock:
4452         unlock_page(page);
4453 out:
4454         return ret;
4455 }
4456
4457 static void btrfs_truncate(struct inode *inode)
4458 {
4459         struct btrfs_root *root = BTRFS_I(inode)->root;
4460         int ret;
4461         struct btrfs_trans_handle *trans;
4462         unsigned long nr;
4463         u64 mask = root->sectorsize - 1;
4464
4465         if (!S_ISREG(inode->i_mode))
4466                 return;
4467         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4468                 return;
4469
4470         btrfs_truncate_page(inode->i_mapping, inode->i_size);
4471         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
4472
4473         trans = btrfs_start_transaction(root, 1);
4474
4475         /*
4476          * setattr is responsible for setting the ordered_data_close flag,
4477          * but that is only tested during the last file release.  That
4478          * could happen well after the next commit, leaving a great big
4479          * window where new writes may get lost if someone chooses to write
4480          * to this file after truncating to zero
4481          *
4482          * The inode doesn't have any dirty data here, and so if we commit
4483          * this is a noop.  If someone immediately starts writing to the inode
4484          * it is very likely we'll catch some of their writes in this
4485          * transaction, and the commit will find this file on the ordered
4486          * data list with good things to send down.
4487          *
4488          * This is a best effort solution, there is still a window where
4489          * using truncate to replace the contents of the file will
4490          * end up with a zero length file after a crash.
4491          */
4492         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
4493                 btrfs_add_ordered_operation(trans, root, inode);
4494
4495         btrfs_set_trans_block_group(trans, inode);
4496         btrfs_i_size_write(inode, inode->i_size);
4497
4498         ret = btrfs_orphan_add(trans, inode);
4499         if (ret)
4500                 goto out;
4501         /* FIXME, add redo link to tree so we don't leak on crash */
4502         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
4503                                       BTRFS_EXTENT_DATA_KEY);
4504         btrfs_update_inode(trans, root, inode);
4505
4506         ret = btrfs_orphan_del(trans, inode);
4507         BUG_ON(ret);
4508
4509 out:
4510         nr = trans->blocks_used;
4511         ret = btrfs_end_transaction_throttle(trans, root);
4512         BUG_ON(ret);
4513         btrfs_btree_balance_dirty(root, nr);
4514 }
4515
4516 /*
4517  * create a new subvolume directory/inode (helper for the ioctl).
4518  */
4519 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
4520                              struct btrfs_root *new_root, struct dentry *dentry,
4521                              u64 new_dirid, u64 alloc_hint)
4522 {
4523         struct inode *inode;
4524         int error;
4525         u64 index = 0;
4526
4527         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
4528                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
4529         if (IS_ERR(inode))
4530                 return PTR_ERR(inode);
4531         inode->i_op = &btrfs_dir_inode_operations;
4532         inode->i_fop = &btrfs_dir_file_operations;
4533
4534         inode->i_nlink = 1;
4535         btrfs_i_size_write(inode, 0);
4536
4537         error = btrfs_update_inode(trans, new_root, inode);
4538         if (error)
4539                 return error;
4540
4541         d_instantiate(dentry, inode);
4542         return 0;
4543 }
4544
4545 /* helper function for file defrag and space balancing.  This
4546  * forces readahead on a given range of bytes in an inode
4547  */
4548 unsigned long btrfs_force_ra(struct address_space *mapping,
4549                               struct file_ra_state *ra, struct file *file,
4550                               pgoff_t offset, pgoff_t last_index)
4551 {
4552         pgoff_t req_size = last_index - offset + 1;
4553
4554         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4555         return offset + req_size;
4556 }
4557
4558 struct inode *btrfs_alloc_inode(struct super_block *sb)
4559 {
4560         struct btrfs_inode *ei;
4561
4562         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4563         if (!ei)
4564                 return NULL;
4565         ei->last_trans = 0;
4566         ei->logged_trans = 0;
4567         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
4568         ei->i_acl = BTRFS_ACL_NOT_CACHED;
4569         ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
4570         INIT_LIST_HEAD(&ei->i_orphan);
4571         INIT_LIST_HEAD(&ei->ordered_operations);
4572         return &ei->vfs_inode;
4573 }
4574
4575 void btrfs_destroy_inode(struct inode *inode)
4576 {
4577         struct btrfs_ordered_extent *ordered;
4578         struct btrfs_root *root = BTRFS_I(inode)->root;
4579
4580         WARN_ON(!list_empty(&inode->i_dentry));
4581         WARN_ON(inode->i_data.nrpages);
4582
4583         if (BTRFS_I(inode)->i_acl &&
4584             BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4585                 posix_acl_release(BTRFS_I(inode)->i_acl);
4586         if (BTRFS_I(inode)->i_default_acl &&
4587             BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4588                 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4589
4590         /*
4591          * Make sure we're properly removed from the ordered operation
4592          * lists.
4593          */
4594         smp_mb();
4595         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
4596                 spin_lock(&root->fs_info->ordered_extent_lock);
4597                 list_del_init(&BTRFS_I(inode)->ordered_operations);
4598                 spin_unlock(&root->fs_info->ordered_extent_lock);
4599         }
4600
4601         spin_lock(&root->list_lock);
4602         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4603                 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4604                        " list\n", inode->i_ino);
4605                 dump_stack();
4606         }
4607         spin_unlock(&root->list_lock);
4608
4609         while (1) {
4610                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4611                 if (!ordered)
4612                         break;
4613                 else {
4614                         printk(KERN_ERR "btrfs found ordered "
4615                                "extent %llu %llu on inode cleanup\n",
4616                                (unsigned long long)ordered->file_offset,
4617                                (unsigned long long)ordered->len);
4618                         btrfs_remove_ordered_extent(inode, ordered);
4619                         btrfs_put_ordered_extent(ordered);
4620                         btrfs_put_ordered_extent(ordered);
4621                 }
4622         }
4623         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
4624         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4625 }
4626
4627 static void init_once(void *foo)
4628 {
4629         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4630
4631         inode_init_once(&ei->vfs_inode);
4632 }
4633
4634 void btrfs_destroy_cachep(void)
4635 {
4636         if (btrfs_inode_cachep)
4637                 kmem_cache_destroy(btrfs_inode_cachep);
4638         if (btrfs_trans_handle_cachep)
4639                 kmem_cache_destroy(btrfs_trans_handle_cachep);
4640         if (btrfs_transaction_cachep)
4641                 kmem_cache_destroy(btrfs_transaction_cachep);
4642         if (btrfs_bit_radix_cachep)
4643                 kmem_cache_destroy(btrfs_bit_radix_cachep);
4644         if (btrfs_path_cachep)
4645                 kmem_cache_destroy(btrfs_path_cachep);
4646 }
4647
4648 int btrfs_init_cachep(void)
4649 {
4650         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
4651                         sizeof(struct btrfs_inode), 0,
4652                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
4653         if (!btrfs_inode_cachep)
4654                 goto fail;
4655
4656         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
4657                         sizeof(struct btrfs_trans_handle), 0,
4658                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
4659         if (!btrfs_trans_handle_cachep)
4660                 goto fail;
4661
4662         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
4663                         sizeof(struct btrfs_transaction), 0,
4664                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
4665         if (!btrfs_transaction_cachep)
4666                 goto fail;
4667
4668         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
4669                         sizeof(struct btrfs_path), 0,
4670                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
4671         if (!btrfs_path_cachep)
4672                 goto fail;
4673
4674         btrfs_bit_radix_cachep = kmem_cache_create("btrfs_radix", 256, 0,
4675                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD |
4676                         SLAB_DESTROY_BY_RCU, NULL);
4677         if (!btrfs_bit_radix_cachep)
4678                 goto fail;
4679         return 0;
4680 fail:
4681         btrfs_destroy_cachep();
4682         return -ENOMEM;
4683 }
4684
4685 static int btrfs_getattr(struct vfsmount *mnt,
4686                          struct dentry *dentry, struct kstat *stat)
4687 {
4688         struct inode *inode = dentry->d_inode;
4689         generic_fillattr(inode, stat);
4690         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
4691         stat->blksize = PAGE_CACHE_SIZE;
4692         stat->blocks = (inode_get_bytes(inode) +
4693                         BTRFS_I(inode)->delalloc_bytes) >> 9;
4694         return 0;
4695 }
4696
4697 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4698                            struct inode *new_dir, struct dentry *new_dentry)
4699 {
4700         struct btrfs_trans_handle *trans;
4701         struct btrfs_root *root = BTRFS_I(old_dir)->root;
4702         struct inode *new_inode = new_dentry->d_inode;
4703         struct inode *old_inode = old_dentry->d_inode;
4704         struct timespec ctime = CURRENT_TIME;
4705         u64 index = 0;
4706         int ret;
4707
4708         /* we're not allowed to rename between subvolumes */
4709         if (BTRFS_I(old_inode)->root->root_key.objectid !=
4710             BTRFS_I(new_dir)->root->root_key.objectid)
4711                 return -EXDEV;
4712
4713         if (S_ISDIR(old_inode->i_mode) && new_inode &&
4714             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4715                 return -ENOTEMPTY;
4716         }
4717
4718         /* to rename a snapshot or subvolume, we need to juggle the
4719          * backrefs.  This isn't coded yet
4720          */
4721         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
4722                 return -EXDEV;
4723
4724         ret = btrfs_check_metadata_free_space(root);
4725         if (ret)
4726                 goto out_unlock;
4727
4728         /*
4729          * we're using rename to replace one file with another.
4730          * and the replacement file is large.  Start IO on it now so
4731          * we don't add too much work to the end of the transaction
4732          */
4733         if (new_inode && old_inode && S_ISREG(old_inode->i_mode) &&
4734             new_inode->i_size &&
4735             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
4736                 filemap_flush(old_inode->i_mapping);
4737
4738         trans = btrfs_start_transaction(root, 1);
4739
4740         /*
4741          * make sure the inode gets flushed if it is replacing
4742          * something.
4743          */
4744         if (new_inode && new_inode->i_size &&
4745             old_inode && S_ISREG(old_inode->i_mode)) {
4746                 btrfs_add_ordered_operation(trans, root, old_inode);
4747         }
4748
4749         /*
4750          * this is an ugly little race, but the rename is required to make
4751          * sure that if we crash, the inode is either at the old name
4752          * or the new one.  pinning the log transaction lets us make sure
4753          * we don't allow a log commit to come in after we unlink the
4754          * name but before we add the new name back in.
4755          */
4756         btrfs_pin_log_trans(root);
4757
4758         btrfs_set_trans_block_group(trans, new_dir);
4759
4760         btrfs_inc_nlink(old_dentry->d_inode);
4761         old_dir->i_ctime = old_dir->i_mtime = ctime;
4762         new_dir->i_ctime = new_dir->i_mtime = ctime;
4763         old_inode->i_ctime = ctime;
4764
4765         if (old_dentry->d_parent != new_dentry->d_parent)
4766                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
4767
4768         ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4769                                  old_dentry->d_name.name,
4770                                  old_dentry->d_name.len);
4771         if (ret)
4772                 goto out_fail;
4773
4774         if (new_inode) {
4775                 new_inode->i_ctime = CURRENT_TIME;
4776                 ret = btrfs_unlink_inode(trans, root, new_dir,
4777                                          new_dentry->d_inode,
4778                                          new_dentry->d_name.name,
4779                                          new_dentry->d_name.len);
4780                 if (ret)
4781                         goto out_fail;
4782                 if (new_inode->i_nlink == 0) {
4783                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4784                         if (ret)
4785                                 goto out_fail;
4786                 }
4787
4788         }
4789         ret = btrfs_set_inode_index(new_dir, &index);
4790         if (ret)
4791                 goto out_fail;
4792
4793         ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4794                              old_inode, new_dentry->d_name.name,
4795                              new_dentry->d_name.len, 1, index);
4796         if (ret)
4797                 goto out_fail;
4798
4799         btrfs_log_new_name(trans, old_inode, old_dir,
4800                                        new_dentry->d_parent);
4801 out_fail:
4802
4803         /* this btrfs_end_log_trans just allows the current
4804          * log-sub transaction to complete
4805          */
4806         btrfs_end_log_trans(root);
4807         btrfs_end_transaction_throttle(trans, root);
4808 out_unlock:
4809         return ret;
4810 }
4811
4812 /*
4813  * some fairly slow code that needs optimization. This walks the list
4814  * of all the inodes with pending delalloc and forces them to disk.
4815  */
4816 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4817 {
4818         struct list_head *head = &root->fs_info->delalloc_inodes;
4819         struct btrfs_inode *binode;
4820         struct inode *inode;
4821
4822         if (root->fs_info->sb->s_flags & MS_RDONLY)
4823                 return -EROFS;
4824
4825         spin_lock(&root->fs_info->delalloc_lock);
4826         while (!list_empty(head)) {
4827                 binode = list_entry(head->next, struct btrfs_inode,
4828                                     delalloc_inodes);
4829                 inode = igrab(&binode->vfs_inode);
4830                 if (!inode)
4831                         list_del_init(&binode->delalloc_inodes);
4832                 spin_unlock(&root->fs_info->delalloc_lock);
4833                 if (inode) {
4834                         filemap_flush(inode->i_mapping);
4835                         iput(inode);
4836                 }
4837                 cond_resched();
4838                 spin_lock(&root->fs_info->delalloc_lock);
4839         }
4840         spin_unlock(&root->fs_info->delalloc_lock);
4841
4842         /* the filemap_flush will queue IO into the worker threads, but
4843          * we have to make sure the IO is actually started and that
4844          * ordered extents get created before we return
4845          */
4846         atomic_inc(&root->fs_info->async_submit_draining);
4847         while (atomic_read(&root->fs_info->nr_async_submits) ||
4848               atomic_read(&root->fs_info->async_delalloc_pages)) {
4849                 wait_event(root->fs_info->async_submit_wait,
4850                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4851                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
4852         }
4853         atomic_dec(&root->fs_info->async_submit_draining);
4854         return 0;
4855 }
4856
4857 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4858                          const char *symname)
4859 {
4860         struct btrfs_trans_handle *trans;
4861         struct btrfs_root *root = BTRFS_I(dir)->root;
4862         struct btrfs_path *path;
4863         struct btrfs_key key;
4864         struct inode *inode = NULL;
4865         int err;
4866         int drop_inode = 0;
4867         u64 objectid;
4868         u64 index = 0 ;
4869         int name_len;
4870         int datasize;
4871         unsigned long ptr;
4872         struct btrfs_file_extent_item *ei;
4873         struct extent_buffer *leaf;
4874         unsigned long nr = 0;
4875
4876         name_len = strlen(symname) + 1;
4877         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4878                 return -ENAMETOOLONG;
4879
4880         err = btrfs_check_metadata_free_space(root);
4881         if (err)
4882                 goto out_fail;
4883
4884         trans = btrfs_start_transaction(root, 1);
4885         btrfs_set_trans_block_group(trans, dir);
4886
4887         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4888         if (err) {
4889                 err = -ENOSPC;
4890                 goto out_unlock;
4891         }
4892
4893         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4894                                 dentry->d_name.len,
4895                                 dentry->d_parent->d_inode->i_ino, objectid,
4896                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4897                                 &index);
4898         err = PTR_ERR(inode);
4899         if (IS_ERR(inode))
4900                 goto out_unlock;
4901
4902         err = btrfs_init_inode_security(inode, dir);
4903         if (err) {
4904                 drop_inode = 1;
4905                 goto out_unlock;
4906         }
4907
4908         btrfs_set_trans_block_group(trans, inode);
4909         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4910         if (err)
4911                 drop_inode = 1;
4912         else {
4913                 inode->i_mapping->a_ops = &btrfs_aops;
4914                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4915                 inode->i_fop = &btrfs_file_operations;
4916                 inode->i_op = &btrfs_file_inode_operations;
4917                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4918         }
4919         dir->i_sb->s_dirt = 1;
4920         btrfs_update_inode_block_group(trans, inode);
4921         btrfs_update_inode_block_group(trans, dir);
4922         if (drop_inode)
4923                 goto out_unlock;
4924
4925         path = btrfs_alloc_path();
4926         BUG_ON(!path);
4927         key.objectid = inode->i_ino;
4928         key.offset = 0;
4929         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4930         datasize = btrfs_file_extent_calc_inline_size(name_len);
4931         err = btrfs_insert_empty_item(trans, root, path, &key,
4932                                       datasize);
4933         if (err) {
4934                 drop_inode = 1;
4935                 goto out_unlock;
4936         }
4937         leaf = path->nodes[0];
4938         ei = btrfs_item_ptr(leaf, path->slots[0],
4939                             struct btrfs_file_extent_item);
4940         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4941         btrfs_set_file_extent_type(leaf, ei,
4942                                    BTRFS_FILE_EXTENT_INLINE);
4943         btrfs_set_file_extent_encryption(leaf, ei, 0);
4944         btrfs_set_file_extent_compression(leaf, ei, 0);
4945         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
4946         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
4947
4948         ptr = btrfs_file_extent_inline_start(ei);
4949         write_extent_buffer(leaf, symname, ptr, name_len);
4950         btrfs_mark_buffer_dirty(leaf);
4951         btrfs_free_path(path);
4952
4953         inode->i_op = &btrfs_symlink_inode_operations;
4954         inode->i_mapping->a_ops = &btrfs_symlink_aops;
4955         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4956         inode_set_bytes(inode, name_len);
4957         btrfs_i_size_write(inode, name_len - 1);
4958         err = btrfs_update_inode(trans, root, inode);
4959         if (err)
4960                 drop_inode = 1;
4961
4962 out_unlock:
4963         nr = trans->blocks_used;
4964         btrfs_end_transaction_throttle(trans, root);
4965 out_fail:
4966         if (drop_inode) {
4967                 inode_dec_link_count(inode);
4968                 iput(inode);
4969         }
4970         btrfs_btree_balance_dirty(root, nr);
4971         return err;
4972 }
4973
4974 static int prealloc_file_range(struct btrfs_trans_handle *trans,
4975                                struct inode *inode, u64 start, u64 end,
4976                                u64 locked_end, u64 alloc_hint, int mode)
4977 {
4978         struct btrfs_root *root = BTRFS_I(inode)->root;
4979         struct btrfs_key ins;
4980         u64 alloc_size;
4981         u64 cur_offset = start;
4982         u64 num_bytes = end - start;
4983         int ret = 0;
4984
4985         while (num_bytes > 0) {
4986                 alloc_size = min(num_bytes, root->fs_info->max_extent);
4987                 ret = btrfs_reserve_extent(trans, root, alloc_size,
4988                                            root->sectorsize, 0, alloc_hint,
4989                                            (u64)-1, &ins, 1);
4990                 if (ret) {
4991                         WARN_ON(1);
4992                         goto out;
4993                 }
4994                 ret = insert_reserved_file_extent(trans, inode,
4995                                                   cur_offset, ins.objectid,
4996                                                   ins.offset, ins.offset,
4997                                                   ins.offset, locked_end,
4998                                                   0, 0, 0,
4999                                                   BTRFS_FILE_EXTENT_PREALLOC);
5000                 BUG_ON(ret);
5001                 num_bytes -= ins.offset;
5002                 cur_offset += ins.offset;
5003                 alloc_hint = ins.objectid + ins.offset;
5004         }
5005 out:
5006         if (cur_offset > start) {
5007                 inode->i_ctime = CURRENT_TIME;
5008                 btrfs_set_flag(inode, PREALLOC);
5009                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
5010                     cur_offset > i_size_read(inode))
5011                         btrfs_i_size_write(inode, cur_offset);
5012                 ret = btrfs_update_inode(trans, root, inode);
5013                 BUG_ON(ret);
5014         }
5015
5016         return ret;
5017 }
5018
5019 static long btrfs_fallocate(struct inode *inode, int mode,
5020                             loff_t offset, loff_t len)
5021 {
5022         u64 cur_offset;
5023         u64 last_byte;
5024         u64 alloc_start;
5025         u64 alloc_end;
5026         u64 alloc_hint = 0;
5027         u64 locked_end;
5028         u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
5029         struct extent_map *em;
5030         struct btrfs_trans_handle *trans;
5031         int ret;
5032
5033         alloc_start = offset & ~mask;
5034         alloc_end =  (offset + len + mask) & ~mask;
5035
5036         /*
5037          * wait for ordered IO before we have any locks.  We'll loop again
5038          * below with the locks held.
5039          */
5040         btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
5041
5042         mutex_lock(&inode->i_mutex);
5043         if (alloc_start > inode->i_size) {
5044                 ret = btrfs_cont_expand(inode, alloc_start);
5045                 if (ret)
5046                         goto out;
5047         }
5048
5049         locked_end = alloc_end - 1;
5050         while (1) {
5051                 struct btrfs_ordered_extent *ordered;
5052
5053                 trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
5054                 if (!trans) {
5055                         ret = -EIO;
5056                         goto out;
5057                 }
5058
5059                 /* the extent lock is ordered inside the running
5060                  * transaction
5061                  */
5062                 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5063                             GFP_NOFS);
5064                 ordered = btrfs_lookup_first_ordered_extent(inode,
5065                                                             alloc_end - 1);
5066                 if (ordered &&
5067                     ordered->file_offset + ordered->len > alloc_start &&
5068                     ordered->file_offset < alloc_end) {
5069                         btrfs_put_ordered_extent(ordered);
5070                         unlock_extent(&BTRFS_I(inode)->io_tree,
5071                                       alloc_start, locked_end, GFP_NOFS);
5072                         btrfs_end_transaction(trans, BTRFS_I(inode)->root);
5073
5074                         /*
5075                          * we can't wait on the range with the transaction
5076                          * running or with the extent lock held
5077                          */
5078                         btrfs_wait_ordered_range(inode, alloc_start,
5079                                                  alloc_end - alloc_start);
5080                 } else {
5081                         if (ordered)
5082                                 btrfs_put_ordered_extent(ordered);
5083                         break;
5084                 }
5085         }
5086
5087         cur_offset = alloc_start;
5088         while (1) {
5089                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5090                                       alloc_end - cur_offset, 0);
5091                 BUG_ON(IS_ERR(em) || !em);
5092                 last_byte = min(extent_map_end(em), alloc_end);
5093                 last_byte = (last_byte + mask) & ~mask;
5094                 if (em->block_start == EXTENT_MAP_HOLE) {
5095                         ret = prealloc_file_range(trans, inode, cur_offset,
5096                                         last_byte, locked_end + 1,
5097                                         alloc_hint, mode);
5098                         if (ret < 0) {
5099                                 free_extent_map(em);
5100                                 break;
5101                         }
5102                 }
5103                 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
5104                         alloc_hint = em->block_start;
5105                 free_extent_map(em);
5106
5107                 cur_offset = last_byte;
5108                 if (cur_offset >= alloc_end) {
5109                         ret = 0;
5110                         break;
5111                 }
5112         }
5113         unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5114                       GFP_NOFS);
5115
5116         btrfs_end_transaction(trans, BTRFS_I(inode)->root);
5117 out:
5118         mutex_unlock(&inode->i_mutex);
5119         return ret;
5120 }
5121
5122 static int btrfs_set_page_dirty(struct page *page)
5123 {
5124         return __set_page_dirty_nobuffers(page);
5125 }
5126
5127 static int btrfs_permission(struct inode *inode, int mask)
5128 {
5129         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
5130                 return -EACCES;
5131         return generic_permission(inode, mask, btrfs_check_acl);
5132 }
5133
5134 static struct inode_operations btrfs_dir_inode_operations = {
5135         .getattr        = btrfs_getattr,
5136         .lookup         = btrfs_lookup,
5137         .create         = btrfs_create,
5138         .unlink         = btrfs_unlink,
5139         .link           = btrfs_link,
5140         .mkdir          = btrfs_mkdir,
5141         .rmdir          = btrfs_rmdir,
5142         .rename         = btrfs_rename,
5143         .symlink        = btrfs_symlink,
5144         .setattr        = btrfs_setattr,
5145         .mknod          = btrfs_mknod,
5146         .setxattr       = btrfs_setxattr,
5147         .getxattr       = btrfs_getxattr,
5148         .listxattr      = btrfs_listxattr,
5149         .removexattr    = btrfs_removexattr,
5150         .permission     = btrfs_permission,
5151 };
5152 static struct inode_operations btrfs_dir_ro_inode_operations = {
5153         .lookup         = btrfs_lookup,
5154         .permission     = btrfs_permission,
5155 };
5156 static struct file_operations btrfs_dir_file_operations = {
5157         .llseek         = generic_file_llseek,
5158         .read           = generic_read_dir,
5159         .readdir        = btrfs_real_readdir,
5160         .unlocked_ioctl = btrfs_ioctl,
5161 #ifdef CONFIG_COMPAT
5162         .compat_ioctl   = btrfs_ioctl,
5163 #endif
5164         .release        = btrfs_release_file,
5165         .fsync          = btrfs_sync_file,
5166 };
5167
5168 static struct extent_io_ops btrfs_extent_io_ops = {
5169         .fill_delalloc = run_delalloc_range,
5170         .submit_bio_hook = btrfs_submit_bio_hook,
5171         .merge_bio_hook = btrfs_merge_bio_hook,
5172         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
5173         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
5174         .writepage_start_hook = btrfs_writepage_start_hook,
5175         .readpage_io_failed_hook = btrfs_io_failed_hook,
5176         .set_bit_hook = btrfs_set_bit_hook,
5177         .clear_bit_hook = btrfs_clear_bit_hook,
5178 };
5179
5180 /*
5181  * btrfs doesn't support the bmap operation because swapfiles
5182  * use bmap to make a mapping of extents in the file.  They assume
5183  * these extents won't change over the life of the file and they
5184  * use the bmap result to do IO directly to the drive.
5185  *
5186  * the btrfs bmap call would return logical addresses that aren't
5187  * suitable for IO and they also will change frequently as COW
5188  * operations happen.  So, swapfile + btrfs == corruption.
5189  *
5190  * For now we're avoiding this by dropping bmap.
5191  */
5192 static struct address_space_operations btrfs_aops = {
5193         .readpage       = btrfs_readpage,
5194         .writepage      = btrfs_writepage,
5195         .writepages     = btrfs_writepages,
5196         .readpages      = btrfs_readpages,
5197         .sync_page      = block_sync_page,
5198         .direct_IO      = btrfs_direct_IO,
5199         .invalidatepage = btrfs_invalidatepage,
5200         .releasepage    = btrfs_releasepage,
5201         .set_page_dirty = btrfs_set_page_dirty,
5202 };
5203
5204 static struct address_space_operations btrfs_symlink_aops = {
5205         .readpage       = btrfs_readpage,
5206         .writepage      = btrfs_writepage,
5207         .invalidatepage = btrfs_invalidatepage,
5208         .releasepage    = btrfs_releasepage,
5209 };
5210
5211 static struct inode_operations btrfs_file_inode_operations = {
5212         .truncate       = btrfs_truncate,
5213         .getattr        = btrfs_getattr,
5214         .setattr        = btrfs_setattr,
5215         .setxattr       = btrfs_setxattr,
5216         .getxattr       = btrfs_getxattr,
5217         .listxattr      = btrfs_listxattr,
5218         .removexattr    = btrfs_removexattr,
5219         .permission     = btrfs_permission,
5220         .fallocate      = btrfs_fallocate,
5221         .fiemap         = btrfs_fiemap,
5222 };
5223 static struct inode_operations btrfs_special_inode_operations = {
5224         .getattr        = btrfs_getattr,
5225         .setattr        = btrfs_setattr,
5226         .permission     = btrfs_permission,
5227         .setxattr       = btrfs_setxattr,
5228         .getxattr       = btrfs_getxattr,
5229         .listxattr      = btrfs_listxattr,
5230         .removexattr    = btrfs_removexattr,
5231 };
5232 static struct inode_operations btrfs_symlink_inode_operations = {
5233         .readlink       = generic_readlink,
5234         .follow_link    = page_follow_link_light,
5235         .put_link       = page_put_link,
5236         .permission     = btrfs_permission,
5237         .setxattr       = btrfs_setxattr,
5238         .getxattr       = btrfs_getxattr,
5239         .listxattr      = btrfs_listxattr,
5240         .removexattr    = btrfs_removexattr,
5241 };