Get "no acls for this inode" right, fix shmem breakage
[linux-2.6.git] / fs / btrfs / inode.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
23 #include <linux/fs.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/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 "compression.h"
52 #include "locking.h"
53
54 struct btrfs_iget_args {
55         u64 ino;
56         struct btrfs_root *root;
57 };
58
59 static struct inode_operations btrfs_dir_inode_operations;
60 static struct inode_operations btrfs_symlink_inode_operations;
61 static struct inode_operations btrfs_dir_ro_inode_operations;
62 static struct inode_operations btrfs_special_inode_operations;
63 static struct inode_operations btrfs_file_inode_operations;
64 static struct address_space_operations btrfs_aops;
65 static struct address_space_operations btrfs_symlink_aops;
66 static struct file_operations btrfs_dir_file_operations;
67 static struct extent_io_ops btrfs_extent_io_ops;
68
69 static struct kmem_cache *btrfs_inode_cachep;
70 struct kmem_cache *btrfs_trans_handle_cachep;
71 struct kmem_cache *btrfs_transaction_cachep;
72 struct kmem_cache *btrfs_path_cachep;
73
74 #define S_SHIFT 12
75 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
76         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
77         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
78         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
79         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
80         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
81         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
82         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
83 };
84
85 static void btrfs_truncate(struct inode *inode);
86 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
87 static noinline int cow_file_range(struct inode *inode,
88                                    struct page *locked_page,
89                                    u64 start, u64 end, int *page_started,
90                                    unsigned long *nr_written, int unlock);
91
92 static int btrfs_init_inode_security(struct inode *inode,  struct inode *dir)
93 {
94         int err;
95
96         err = btrfs_init_acl(inode, dir);
97         if (!err)
98                 err = btrfs_xattr_security_init(inode, dir);
99         return err;
100 }
101
102 /*
103  * this does all the hard work for inserting an inline extent into
104  * the btree.  The caller should have done a btrfs_drop_extents so that
105  * no overlapping inline items exist in the btree
106  */
107 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
108                                 struct btrfs_root *root, struct inode *inode,
109                                 u64 start, size_t size, size_t compressed_size,
110                                 struct page **compressed_pages)
111 {
112         struct btrfs_key key;
113         struct btrfs_path *path;
114         struct extent_buffer *leaf;
115         struct page *page = NULL;
116         char *kaddr;
117         unsigned long ptr;
118         struct btrfs_file_extent_item *ei;
119         int err = 0;
120         int ret;
121         size_t cur_size = size;
122         size_t datasize;
123         unsigned long offset;
124         int use_compress = 0;
125
126         if (compressed_size && compressed_pages) {
127                 use_compress = 1;
128                 cur_size = compressed_size;
129         }
130
131         path = btrfs_alloc_path();
132         if (!path)
133                 return -ENOMEM;
134
135         path->leave_spinning = 1;
136         btrfs_set_trans_block_group(trans, inode);
137
138         key.objectid = inode->i_ino;
139         key.offset = start;
140         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
141         datasize = btrfs_file_extent_calc_inline_size(cur_size);
142
143         inode_add_bytes(inode, size);
144         ret = btrfs_insert_empty_item(trans, root, path, &key,
145                                       datasize);
146         BUG_ON(ret);
147         if (ret) {
148                 err = ret;
149                 goto fail;
150         }
151         leaf = path->nodes[0];
152         ei = btrfs_item_ptr(leaf, path->slots[0],
153                             struct btrfs_file_extent_item);
154         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
155         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
156         btrfs_set_file_extent_encryption(leaf, ei, 0);
157         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
158         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
159         ptr = btrfs_file_extent_inline_start(ei);
160
161         if (use_compress) {
162                 struct page *cpage;
163                 int i = 0;
164                 while (compressed_size > 0) {
165                         cpage = compressed_pages[i];
166                         cur_size = min_t(unsigned long, compressed_size,
167                                        PAGE_CACHE_SIZE);
168
169                         kaddr = kmap_atomic(cpage, KM_USER0);
170                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
171                         kunmap_atomic(kaddr, KM_USER0);
172
173                         i++;
174                         ptr += cur_size;
175                         compressed_size -= cur_size;
176                 }
177                 btrfs_set_file_extent_compression(leaf, ei,
178                                                   BTRFS_COMPRESS_ZLIB);
179         } else {
180                 page = find_get_page(inode->i_mapping,
181                                      start >> PAGE_CACHE_SHIFT);
182                 btrfs_set_file_extent_compression(leaf, ei, 0);
183                 kaddr = kmap_atomic(page, KM_USER0);
184                 offset = start & (PAGE_CACHE_SIZE - 1);
185                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
186                 kunmap_atomic(kaddr, KM_USER0);
187                 page_cache_release(page);
188         }
189         btrfs_mark_buffer_dirty(leaf);
190         btrfs_free_path(path);
191
192         BTRFS_I(inode)->disk_i_size = inode->i_size;
193         btrfs_update_inode(trans, root, inode);
194         return 0;
195 fail:
196         btrfs_free_path(path);
197         return err;
198 }
199
200
201 /*
202  * conditionally insert an inline extent into the file.  This
203  * does the checks required to make sure the data is small enough
204  * to fit as an inline extent.
205  */
206 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
207                                  struct btrfs_root *root,
208                                  struct inode *inode, u64 start, u64 end,
209                                  size_t compressed_size,
210                                  struct page **compressed_pages)
211 {
212         u64 isize = i_size_read(inode);
213         u64 actual_end = min(end + 1, isize);
214         u64 inline_len = actual_end - start;
215         u64 aligned_end = (end + root->sectorsize - 1) &
216                         ~((u64)root->sectorsize - 1);
217         u64 hint_byte;
218         u64 data_len = inline_len;
219         int ret;
220
221         if (compressed_size)
222                 data_len = compressed_size;
223
224         if (start > 0 ||
225             actual_end >= PAGE_CACHE_SIZE ||
226             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
227             (!compressed_size &&
228             (actual_end & (root->sectorsize - 1)) == 0) ||
229             end + 1 < isize ||
230             data_len > root->fs_info->max_inline) {
231                 return 1;
232         }
233
234         ret = btrfs_drop_extents(trans, root, inode, start,
235                                  aligned_end, aligned_end, start, &hint_byte);
236         BUG_ON(ret);
237
238         if (isize > actual_end)
239                 inline_len = min_t(u64, isize, actual_end);
240         ret = insert_inline_extent(trans, root, inode, start,
241                                    inline_len, compressed_size,
242                                    compressed_pages);
243         BUG_ON(ret);
244         btrfs_drop_extent_cache(inode, start, aligned_end, 0);
245         return 0;
246 }
247
248 struct async_extent {
249         u64 start;
250         u64 ram_size;
251         u64 compressed_size;
252         struct page **pages;
253         unsigned long nr_pages;
254         struct list_head list;
255 };
256
257 struct async_cow {
258         struct inode *inode;
259         struct btrfs_root *root;
260         struct page *locked_page;
261         u64 start;
262         u64 end;
263         struct list_head extents;
264         struct btrfs_work work;
265 };
266
267 static noinline int add_async_extent(struct async_cow *cow,
268                                      u64 start, u64 ram_size,
269                                      u64 compressed_size,
270                                      struct page **pages,
271                                      unsigned long nr_pages)
272 {
273         struct async_extent *async_extent;
274
275         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
276         async_extent->start = start;
277         async_extent->ram_size = ram_size;
278         async_extent->compressed_size = compressed_size;
279         async_extent->pages = pages;
280         async_extent->nr_pages = nr_pages;
281         list_add_tail(&async_extent->list, &cow->extents);
282         return 0;
283 }
284
285 /*
286  * we create compressed extents in two phases.  The first
287  * phase compresses a range of pages that have already been
288  * locked (both pages and state bits are locked).
289  *
290  * This is done inside an ordered work queue, and the compression
291  * is spread across many cpus.  The actual IO submission is step
292  * two, and the ordered work queue takes care of making sure that
293  * happens in the same order things were put onto the queue by
294  * writepages and friends.
295  *
296  * If this code finds it can't get good compression, it puts an
297  * entry onto the work queue to write the uncompressed bytes.  This
298  * makes sure that both compressed inodes and uncompressed inodes
299  * are written in the same order that pdflush sent them down.
300  */
301 static noinline int compress_file_range(struct inode *inode,
302                                         struct page *locked_page,
303                                         u64 start, u64 end,
304                                         struct async_cow *async_cow,
305                                         int *num_added)
306 {
307         struct btrfs_root *root = BTRFS_I(inode)->root;
308         struct btrfs_trans_handle *trans;
309         u64 num_bytes;
310         u64 orig_start;
311         u64 disk_num_bytes;
312         u64 blocksize = root->sectorsize;
313         u64 actual_end;
314         u64 isize = i_size_read(inode);
315         int ret = 0;
316         struct page **pages = NULL;
317         unsigned long nr_pages;
318         unsigned long nr_pages_ret = 0;
319         unsigned long total_compressed = 0;
320         unsigned long total_in = 0;
321         unsigned long max_compressed = 128 * 1024;
322         unsigned long max_uncompressed = 128 * 1024;
323         int i;
324         int will_compress;
325
326         orig_start = start;
327
328         actual_end = min_t(u64, isize, end + 1);
329 again:
330         will_compress = 0;
331         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
332         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
333
334         /*
335          * we don't want to send crud past the end of i_size through
336          * compression, that's just a waste of CPU time.  So, if the
337          * end of the file is before the start of our current
338          * requested range of bytes, we bail out to the uncompressed
339          * cleanup code that can deal with all of this.
340          *
341          * It isn't really the fastest way to fix things, but this is a
342          * very uncommon corner.
343          */
344         if (actual_end <= start)
345                 goto cleanup_and_bail_uncompressed;
346
347         total_compressed = actual_end - start;
348
349         /* we want to make sure that amount of ram required to uncompress
350          * an extent is reasonable, so we limit the total size in ram
351          * of a compressed extent to 128k.  This is a crucial number
352          * because it also controls how easily we can spread reads across
353          * cpus for decompression.
354          *
355          * We also want to make sure the amount of IO required to do
356          * a random read is reasonably small, so we limit the size of
357          * a compressed extent to 128k.
358          */
359         total_compressed = min(total_compressed, max_uncompressed);
360         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
361         num_bytes = max(blocksize,  num_bytes);
362         disk_num_bytes = num_bytes;
363         total_in = 0;
364         ret = 0;
365
366         /*
367          * we do compression for mount -o compress and when the
368          * inode has not been flagged as nocompress.  This flag can
369          * change at any time if we discover bad compression ratios.
370          */
371         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
372             btrfs_test_opt(root, COMPRESS)) {
373                 WARN_ON(pages);
374                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
375
376                 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
377                                                 total_compressed, pages,
378                                                 nr_pages, &nr_pages_ret,
379                                                 &total_in,
380                                                 &total_compressed,
381                                                 max_compressed);
382
383                 if (!ret) {
384                         unsigned long offset = total_compressed &
385                                 (PAGE_CACHE_SIZE - 1);
386                         struct page *page = pages[nr_pages_ret - 1];
387                         char *kaddr;
388
389                         /* zero the tail end of the last page, we might be
390                          * sending it down to disk
391                          */
392                         if (offset) {
393                                 kaddr = kmap_atomic(page, KM_USER0);
394                                 memset(kaddr + offset, 0,
395                                        PAGE_CACHE_SIZE - offset);
396                                 kunmap_atomic(kaddr, KM_USER0);
397                         }
398                         will_compress = 1;
399                 }
400         }
401         if (start == 0) {
402                 trans = btrfs_join_transaction(root, 1);
403                 BUG_ON(!trans);
404                 btrfs_set_trans_block_group(trans, inode);
405
406                 /* lets try to make an inline extent */
407                 if (ret || total_in < (actual_end - start)) {
408                         /* we didn't compress the entire range, try
409                          * to make an uncompressed inline extent.
410                          */
411                         ret = cow_file_range_inline(trans, root, inode,
412                                                     start, end, 0, NULL);
413                 } else {
414                         /* try making a compressed inline extent */
415                         ret = cow_file_range_inline(trans, root, inode,
416                                                     start, end,
417                                                     total_compressed, pages);
418                 }
419                 btrfs_end_transaction(trans, root);
420                 if (ret == 0) {
421                         /*
422                          * inline extent creation worked, we don't need
423                          * to create any more async work items.  Unlock
424                          * and free up our temp pages.
425                          */
426                         extent_clear_unlock_delalloc(inode,
427                                                      &BTRFS_I(inode)->io_tree,
428                                                      start, end, NULL, 1, 0,
429                                                      0, 1, 1, 1);
430                         ret = 0;
431                         goto free_pages_out;
432                 }
433         }
434
435         if (will_compress) {
436                 /*
437                  * we aren't doing an inline extent round the compressed size
438                  * up to a block size boundary so the allocator does sane
439                  * things
440                  */
441                 total_compressed = (total_compressed + blocksize - 1) &
442                         ~(blocksize - 1);
443
444                 /*
445                  * one last check to make sure the compression is really a
446                  * win, compare the page count read with the blocks on disk
447                  */
448                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
449                         ~(PAGE_CACHE_SIZE - 1);
450                 if (total_compressed >= total_in) {
451                         will_compress = 0;
452                 } else {
453                         disk_num_bytes = total_compressed;
454                         num_bytes = total_in;
455                 }
456         }
457         if (!will_compress && pages) {
458                 /*
459                  * the compression code ran but failed to make things smaller,
460                  * free any pages it allocated and our page pointer array
461                  */
462                 for (i = 0; i < nr_pages_ret; i++) {
463                         WARN_ON(pages[i]->mapping);
464                         page_cache_release(pages[i]);
465                 }
466                 kfree(pages);
467                 pages = NULL;
468                 total_compressed = 0;
469                 nr_pages_ret = 0;
470
471                 /* flag the file so we don't compress in the future */
472                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
473         }
474         if (will_compress) {
475                 *num_added += 1;
476
477                 /* the async work queues will take care of doing actual
478                  * allocation on disk for these compressed pages,
479                  * and will submit them to the elevator.
480                  */
481                 add_async_extent(async_cow, start, num_bytes,
482                                  total_compressed, pages, nr_pages_ret);
483
484                 if (start + num_bytes < end && start + num_bytes < actual_end) {
485                         start += num_bytes;
486                         pages = NULL;
487                         cond_resched();
488                         goto again;
489                 }
490         } else {
491 cleanup_and_bail_uncompressed:
492                 /*
493                  * No compression, but we still need to write the pages in
494                  * the file we've been given so far.  redirty the locked
495                  * page if it corresponds to our extent and set things up
496                  * for the async work queue to run cow_file_range to do
497                  * the normal delalloc dance
498                  */
499                 if (page_offset(locked_page) >= start &&
500                     page_offset(locked_page) <= end) {
501                         __set_page_dirty_nobuffers(locked_page);
502                         /* unlocked later on in the async handlers */
503                 }
504                 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
505                 *num_added += 1;
506         }
507
508 out:
509         return 0;
510
511 free_pages_out:
512         for (i = 0; i < nr_pages_ret; i++) {
513                 WARN_ON(pages[i]->mapping);
514                 page_cache_release(pages[i]);
515         }
516         kfree(pages);
517
518         goto out;
519 }
520
521 /*
522  * phase two of compressed writeback.  This is the ordered portion
523  * of the code, which only gets called in the order the work was
524  * queued.  We walk all the async extents created by compress_file_range
525  * and send them down to the disk.
526  */
527 static noinline int submit_compressed_extents(struct inode *inode,
528                                               struct async_cow *async_cow)
529 {
530         struct async_extent *async_extent;
531         u64 alloc_hint = 0;
532         struct btrfs_trans_handle *trans;
533         struct btrfs_key ins;
534         struct extent_map *em;
535         struct btrfs_root *root = BTRFS_I(inode)->root;
536         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
537         struct extent_io_tree *io_tree;
538         int ret;
539
540         if (list_empty(&async_cow->extents))
541                 return 0;
542
543         trans = btrfs_join_transaction(root, 1);
544
545         while (!list_empty(&async_cow->extents)) {
546                 async_extent = list_entry(async_cow->extents.next,
547                                           struct async_extent, list);
548                 list_del(&async_extent->list);
549
550                 io_tree = &BTRFS_I(inode)->io_tree;
551
552                 /* did the compression code fall back to uncompressed IO? */
553                 if (!async_extent->pages) {
554                         int page_started = 0;
555                         unsigned long nr_written = 0;
556
557                         lock_extent(io_tree, async_extent->start,
558                                     async_extent->start +
559                                     async_extent->ram_size - 1, GFP_NOFS);
560
561                         /* allocate blocks */
562                         cow_file_range(inode, async_cow->locked_page,
563                                        async_extent->start,
564                                        async_extent->start +
565                                        async_extent->ram_size - 1,
566                                        &page_started, &nr_written, 0);
567
568                         /*
569                          * if page_started, cow_file_range inserted an
570                          * inline extent and took care of all the unlocking
571                          * and IO for us.  Otherwise, we need to submit
572                          * all those pages down to the drive.
573                          */
574                         if (!page_started)
575                                 extent_write_locked_range(io_tree,
576                                                   inode, async_extent->start,
577                                                   async_extent->start +
578                                                   async_extent->ram_size - 1,
579                                                   btrfs_get_extent,
580                                                   WB_SYNC_ALL);
581                         kfree(async_extent);
582                         cond_resched();
583                         continue;
584                 }
585
586                 lock_extent(io_tree, async_extent->start,
587                             async_extent->start + async_extent->ram_size - 1,
588                             GFP_NOFS);
589                 /*
590                  * here we're doing allocation and writeback of the
591                  * compressed pages
592                  */
593                 btrfs_drop_extent_cache(inode, async_extent->start,
594                                         async_extent->start +
595                                         async_extent->ram_size - 1, 0);
596
597                 ret = btrfs_reserve_extent(trans, root,
598                                            async_extent->compressed_size,
599                                            async_extent->compressed_size,
600                                            0, alloc_hint,
601                                            (u64)-1, &ins, 1);
602                 BUG_ON(ret);
603                 em = alloc_extent_map(GFP_NOFS);
604                 em->start = async_extent->start;
605                 em->len = async_extent->ram_size;
606                 em->orig_start = em->start;
607
608                 em->block_start = ins.objectid;
609                 em->block_len = ins.offset;
610                 em->bdev = root->fs_info->fs_devices->latest_bdev;
611                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
612                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
613
614                 while (1) {
615                         spin_lock(&em_tree->lock);
616                         ret = add_extent_mapping(em_tree, em);
617                         spin_unlock(&em_tree->lock);
618                         if (ret != -EEXIST) {
619                                 free_extent_map(em);
620                                 break;
621                         }
622                         btrfs_drop_extent_cache(inode, async_extent->start,
623                                                 async_extent->start +
624                                                 async_extent->ram_size - 1, 0);
625                 }
626
627                 ret = btrfs_add_ordered_extent(inode, async_extent->start,
628                                                ins.objectid,
629                                                async_extent->ram_size,
630                                                ins.offset,
631                                                BTRFS_ORDERED_COMPRESSED);
632                 BUG_ON(ret);
633
634                 btrfs_end_transaction(trans, root);
635
636                 /*
637                  * clear dirty, set writeback and unlock the pages.
638                  */
639                 extent_clear_unlock_delalloc(inode,
640                                              &BTRFS_I(inode)->io_tree,
641                                              async_extent->start,
642                                              async_extent->start +
643                                              async_extent->ram_size - 1,
644                                              NULL, 1, 1, 0, 1, 1, 0);
645
646                 ret = btrfs_submit_compressed_write(inode,
647                                     async_extent->start,
648                                     async_extent->ram_size,
649                                     ins.objectid,
650                                     ins.offset, async_extent->pages,
651                                     async_extent->nr_pages);
652
653                 BUG_ON(ret);
654                 trans = btrfs_join_transaction(root, 1);
655                 alloc_hint = ins.objectid + ins.offset;
656                 kfree(async_extent);
657                 cond_resched();
658         }
659
660         btrfs_end_transaction(trans, root);
661         return 0;
662 }
663
664 /*
665  * when extent_io.c finds a delayed allocation range in the file,
666  * the call backs end up in this code.  The basic idea is to
667  * allocate extents on disk for the range, and create ordered data structs
668  * in ram to track those extents.
669  *
670  * locked_page is the page that writepage had locked already.  We use
671  * it to make sure we don't do extra locks or unlocks.
672  *
673  * *page_started is set to one if we unlock locked_page and do everything
674  * required to start IO on it.  It may be clean and already done with
675  * IO when we return.
676  */
677 static noinline int cow_file_range(struct inode *inode,
678                                    struct page *locked_page,
679                                    u64 start, u64 end, int *page_started,
680                                    unsigned long *nr_written,
681                                    int unlock)
682 {
683         struct btrfs_root *root = BTRFS_I(inode)->root;
684         struct btrfs_trans_handle *trans;
685         u64 alloc_hint = 0;
686         u64 num_bytes;
687         unsigned long ram_size;
688         u64 disk_num_bytes;
689         u64 cur_alloc_size;
690         u64 blocksize = root->sectorsize;
691         u64 actual_end;
692         u64 isize = i_size_read(inode);
693         struct btrfs_key ins;
694         struct extent_map *em;
695         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
696         int ret = 0;
697
698         trans = btrfs_join_transaction(root, 1);
699         BUG_ON(!trans);
700         btrfs_set_trans_block_group(trans, inode);
701
702         actual_end = min_t(u64, isize, end + 1);
703
704         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
705         num_bytes = max(blocksize,  num_bytes);
706         disk_num_bytes = num_bytes;
707         ret = 0;
708
709         if (start == 0) {
710                 /* lets try to make an inline extent */
711                 ret = cow_file_range_inline(trans, root, inode,
712                                             start, end, 0, NULL);
713                 if (ret == 0) {
714                         extent_clear_unlock_delalloc(inode,
715                                                      &BTRFS_I(inode)->io_tree,
716                                                      start, end, NULL, 1, 1,
717                                                      1, 1, 1, 1);
718                         *nr_written = *nr_written +
719                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
720                         *page_started = 1;
721                         ret = 0;
722                         goto out;
723                 }
724         }
725
726         BUG_ON(disk_num_bytes >
727                btrfs_super_total_bytes(&root->fs_info->super_copy));
728
729         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
730
731         while (disk_num_bytes > 0) {
732                 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
733                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
734                                            root->sectorsize, 0, alloc_hint,
735                                            (u64)-1, &ins, 1);
736                 BUG_ON(ret);
737
738                 em = alloc_extent_map(GFP_NOFS);
739                 em->start = start;
740                 em->orig_start = em->start;
741
742                 ram_size = ins.offset;
743                 em->len = ins.offset;
744
745                 em->block_start = ins.objectid;
746                 em->block_len = ins.offset;
747                 em->bdev = root->fs_info->fs_devices->latest_bdev;
748                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
749
750                 while (1) {
751                         spin_lock(&em_tree->lock);
752                         ret = add_extent_mapping(em_tree, em);
753                         spin_unlock(&em_tree->lock);
754                         if (ret != -EEXIST) {
755                                 free_extent_map(em);
756                                 break;
757                         }
758                         btrfs_drop_extent_cache(inode, start,
759                                                 start + ram_size - 1, 0);
760                 }
761
762                 cur_alloc_size = ins.offset;
763                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
764                                                ram_size, cur_alloc_size, 0);
765                 BUG_ON(ret);
766
767                 if (root->root_key.objectid ==
768                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
769                         ret = btrfs_reloc_clone_csums(inode, start,
770                                                       cur_alloc_size);
771                         BUG_ON(ret);
772                 }
773
774                 if (disk_num_bytes < cur_alloc_size)
775                         break;
776
777                 /* we're not doing compressed IO, don't unlock the first
778                  * page (which the caller expects to stay locked), don't
779                  * clear any dirty bits and don't set any writeback bits
780                  */
781                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
782                                              start, start + ram_size - 1,
783                                              locked_page, unlock, 1,
784                                              1, 0, 0, 0);
785                 disk_num_bytes -= cur_alloc_size;
786                 num_bytes -= cur_alloc_size;
787                 alloc_hint = ins.objectid + ins.offset;
788                 start += cur_alloc_size;
789         }
790 out:
791         ret = 0;
792         btrfs_end_transaction(trans, root);
793
794         return ret;
795 }
796
797 /*
798  * work queue call back to started compression on a file and pages
799  */
800 static noinline void async_cow_start(struct btrfs_work *work)
801 {
802         struct async_cow *async_cow;
803         int num_added = 0;
804         async_cow = container_of(work, struct async_cow, work);
805
806         compress_file_range(async_cow->inode, async_cow->locked_page,
807                             async_cow->start, async_cow->end, async_cow,
808                             &num_added);
809         if (num_added == 0)
810                 async_cow->inode = NULL;
811 }
812
813 /*
814  * work queue call back to submit previously compressed pages
815  */
816 static noinline void async_cow_submit(struct btrfs_work *work)
817 {
818         struct async_cow *async_cow;
819         struct btrfs_root *root;
820         unsigned long nr_pages;
821
822         async_cow = container_of(work, struct async_cow, work);
823
824         root = async_cow->root;
825         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
826                 PAGE_CACHE_SHIFT;
827
828         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
829
830         if (atomic_read(&root->fs_info->async_delalloc_pages) <
831             5 * 1042 * 1024 &&
832             waitqueue_active(&root->fs_info->async_submit_wait))
833                 wake_up(&root->fs_info->async_submit_wait);
834
835         if (async_cow->inode)
836                 submit_compressed_extents(async_cow->inode, async_cow);
837 }
838
839 static noinline void async_cow_free(struct btrfs_work *work)
840 {
841         struct async_cow *async_cow;
842         async_cow = container_of(work, struct async_cow, work);
843         kfree(async_cow);
844 }
845
846 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
847                                 u64 start, u64 end, int *page_started,
848                                 unsigned long *nr_written)
849 {
850         struct async_cow *async_cow;
851         struct btrfs_root *root = BTRFS_I(inode)->root;
852         unsigned long nr_pages;
853         u64 cur_end;
854         int limit = 10 * 1024 * 1042;
855
856         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
857                          EXTENT_DELALLOC, 1, 0, GFP_NOFS);
858         while (start < end) {
859                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
860                 async_cow->inode = inode;
861                 async_cow->root = root;
862                 async_cow->locked_page = locked_page;
863                 async_cow->start = start;
864
865                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
866                         cur_end = end;
867                 else
868                         cur_end = min(end, start + 512 * 1024 - 1);
869
870                 async_cow->end = cur_end;
871                 INIT_LIST_HEAD(&async_cow->extents);
872
873                 async_cow->work.func = async_cow_start;
874                 async_cow->work.ordered_func = async_cow_submit;
875                 async_cow->work.ordered_free = async_cow_free;
876                 async_cow->work.flags = 0;
877
878                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
879                         PAGE_CACHE_SHIFT;
880                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
881
882                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
883                                    &async_cow->work);
884
885                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
886                         wait_event(root->fs_info->async_submit_wait,
887                            (atomic_read(&root->fs_info->async_delalloc_pages) <
888                             limit));
889                 }
890
891                 while (atomic_read(&root->fs_info->async_submit_draining) &&
892                       atomic_read(&root->fs_info->async_delalloc_pages)) {
893                         wait_event(root->fs_info->async_submit_wait,
894                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
895                            0));
896                 }
897
898                 *nr_written += nr_pages;
899                 start = cur_end + 1;
900         }
901         *page_started = 1;
902         return 0;
903 }
904
905 static noinline int csum_exist_in_range(struct btrfs_root *root,
906                                         u64 bytenr, u64 num_bytes)
907 {
908         int ret;
909         struct btrfs_ordered_sum *sums;
910         LIST_HEAD(list);
911
912         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
913                                        bytenr + num_bytes - 1, &list);
914         if (ret == 0 && list_empty(&list))
915                 return 0;
916
917         while (!list_empty(&list)) {
918                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
919                 list_del(&sums->list);
920                 kfree(sums);
921         }
922         return 1;
923 }
924
925 /*
926  * when nowcow writeback call back.  This checks for snapshots or COW copies
927  * of the extents that exist in the file, and COWs the file as required.
928  *
929  * If no cow copies or snapshots exist, we write directly to the existing
930  * blocks on disk
931  */
932 static noinline int run_delalloc_nocow(struct inode *inode,
933                                        struct page *locked_page,
934                               u64 start, u64 end, int *page_started, int force,
935                               unsigned long *nr_written)
936 {
937         struct btrfs_root *root = BTRFS_I(inode)->root;
938         struct btrfs_trans_handle *trans;
939         struct extent_buffer *leaf;
940         struct btrfs_path *path;
941         struct btrfs_file_extent_item *fi;
942         struct btrfs_key found_key;
943         u64 cow_start;
944         u64 cur_offset;
945         u64 extent_end;
946         u64 extent_offset;
947         u64 disk_bytenr;
948         u64 num_bytes;
949         int extent_type;
950         int ret;
951         int type;
952         int nocow;
953         int check_prev = 1;
954
955         path = btrfs_alloc_path();
956         BUG_ON(!path);
957         trans = btrfs_join_transaction(root, 1);
958         BUG_ON(!trans);
959
960         cow_start = (u64)-1;
961         cur_offset = start;
962         while (1) {
963                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
964                                                cur_offset, 0);
965                 BUG_ON(ret < 0);
966                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
967                         leaf = path->nodes[0];
968                         btrfs_item_key_to_cpu(leaf, &found_key,
969                                               path->slots[0] - 1);
970                         if (found_key.objectid == inode->i_ino &&
971                             found_key.type == BTRFS_EXTENT_DATA_KEY)
972                                 path->slots[0]--;
973                 }
974                 check_prev = 0;
975 next_slot:
976                 leaf = path->nodes[0];
977                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
978                         ret = btrfs_next_leaf(root, path);
979                         if (ret < 0)
980                                 BUG_ON(1);
981                         if (ret > 0)
982                                 break;
983                         leaf = path->nodes[0];
984                 }
985
986                 nocow = 0;
987                 disk_bytenr = 0;
988                 num_bytes = 0;
989                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
990
991                 if (found_key.objectid > inode->i_ino ||
992                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
993                     found_key.offset > end)
994                         break;
995
996                 if (found_key.offset > cur_offset) {
997                         extent_end = found_key.offset;
998                         goto out_check;
999                 }
1000
1001                 fi = btrfs_item_ptr(leaf, path->slots[0],
1002                                     struct btrfs_file_extent_item);
1003                 extent_type = btrfs_file_extent_type(leaf, fi);
1004
1005                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1006                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1007                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1008                         extent_offset = btrfs_file_extent_offset(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                                                   found_key.offset -
1027                                                   extent_offset, disk_bytenr))
1028                                 goto out_check;
1029                         disk_bytenr += extent_offset;
1030                         disk_bytenr += cur_offset - found_key.offset;
1031                         num_bytes = min(end + 1, extent_end) - cur_offset;
1032                         /*
1033                          * force cow if csum exists in the range.
1034                          * this ensure that csum for a given extent are
1035                          * either valid or do not exist.
1036                          */
1037                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1038                                 goto out_check;
1039                         nocow = 1;
1040                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1041                         extent_end = found_key.offset +
1042                                 btrfs_file_extent_inline_len(leaf, fi);
1043                         extent_end = ALIGN(extent_end, root->sectorsize);
1044                 } else {
1045                         BUG_ON(1);
1046                 }
1047 out_check:
1048                 if (extent_end <= start) {
1049                         path->slots[0]++;
1050                         goto next_slot;
1051                 }
1052                 if (!nocow) {
1053                         if (cow_start == (u64)-1)
1054                                 cow_start = cur_offset;
1055                         cur_offset = extent_end;
1056                         if (cur_offset > end)
1057                                 break;
1058                         path->slots[0]++;
1059                         goto next_slot;
1060                 }
1061
1062                 btrfs_release_path(root, path);
1063                 if (cow_start != (u64)-1) {
1064                         ret = cow_file_range(inode, locked_page, cow_start,
1065                                         found_key.offset - 1, page_started,
1066                                         nr_written, 1);
1067                         BUG_ON(ret);
1068                         cow_start = (u64)-1;
1069                 }
1070
1071                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1072                         struct extent_map *em;
1073                         struct extent_map_tree *em_tree;
1074                         em_tree = &BTRFS_I(inode)->extent_tree;
1075                         em = alloc_extent_map(GFP_NOFS);
1076                         em->start = cur_offset;
1077                         em->orig_start = em->start;
1078                         em->len = num_bytes;
1079                         em->block_len = num_bytes;
1080                         em->block_start = disk_bytenr;
1081                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1082                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1083                         while (1) {
1084                                 spin_lock(&em_tree->lock);
1085                                 ret = add_extent_mapping(em_tree, em);
1086                                 spin_unlock(&em_tree->lock);
1087                                 if (ret != -EEXIST) {
1088                                         free_extent_map(em);
1089                                         break;
1090                                 }
1091                                 btrfs_drop_extent_cache(inode, em->start,
1092                                                 em->start + em->len - 1, 0);
1093                         }
1094                         type = BTRFS_ORDERED_PREALLOC;
1095                 } else {
1096                         type = BTRFS_ORDERED_NOCOW;
1097                 }
1098
1099                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1100                                                num_bytes, num_bytes, type);
1101                 BUG_ON(ret);
1102
1103                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1104                                         cur_offset, cur_offset + num_bytes - 1,
1105                                         locked_page, 1, 1, 1, 0, 0, 0);
1106                 cur_offset = extent_end;
1107                 if (cur_offset > end)
1108                         break;
1109         }
1110         btrfs_release_path(root, path);
1111
1112         if (cur_offset <= end && cow_start == (u64)-1)
1113                 cow_start = cur_offset;
1114         if (cow_start != (u64)-1) {
1115                 ret = cow_file_range(inode, locked_page, cow_start, end,
1116                                      page_started, nr_written, 1);
1117                 BUG_ON(ret);
1118         }
1119
1120         ret = btrfs_end_transaction(trans, root);
1121         BUG_ON(ret);
1122         btrfs_free_path(path);
1123         return 0;
1124 }
1125
1126 /*
1127  * extent_io.c call back to do delayed allocation processing
1128  */
1129 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1130                               u64 start, u64 end, int *page_started,
1131                               unsigned long *nr_written)
1132 {
1133         int ret;
1134         struct btrfs_root *root = BTRFS_I(inode)->root;
1135
1136         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1137                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1138                                          page_started, 1, nr_written);
1139         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1140                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1141                                          page_started, 0, nr_written);
1142         else if (!btrfs_test_opt(root, COMPRESS))
1143                 ret = cow_file_range(inode, locked_page, start, end,
1144                                       page_started, nr_written, 1);
1145         else
1146                 ret = cow_file_range_async(inode, locked_page, start, end,
1147                                            page_started, nr_written);
1148         return ret;
1149 }
1150
1151 /*
1152  * extent_io.c set_bit_hook, used to track delayed allocation
1153  * bytes in this file, and to maintain the list of inodes that
1154  * have pending delalloc work to be done.
1155  */
1156 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1157                        unsigned long old, unsigned long bits)
1158 {
1159         /*
1160          * set_bit and clear bit hooks normally require _irqsave/restore
1161          * but in this case, we are only testeing for the DELALLOC
1162          * bit, which is only set or cleared with irqs on
1163          */
1164         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1165                 struct btrfs_root *root = BTRFS_I(inode)->root;
1166                 btrfs_delalloc_reserve_space(root, inode, end - start + 1);
1167                 spin_lock(&root->fs_info->delalloc_lock);
1168                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1169                 root->fs_info->delalloc_bytes += end - start + 1;
1170                 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1171                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1172                                       &root->fs_info->delalloc_inodes);
1173                 }
1174                 spin_unlock(&root->fs_info->delalloc_lock);
1175         }
1176         return 0;
1177 }
1178
1179 /*
1180  * extent_io.c clear_bit_hook, see set_bit_hook for why
1181  */
1182 static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
1183                          unsigned long old, unsigned long bits)
1184 {
1185         /*
1186          * set_bit and clear bit hooks normally require _irqsave/restore
1187          * but in this case, we are only testeing for the DELALLOC
1188          * bit, which is only set or cleared with irqs on
1189          */
1190         if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1191                 struct btrfs_root *root = BTRFS_I(inode)->root;
1192
1193                 spin_lock(&root->fs_info->delalloc_lock);
1194                 if (end - start + 1 > root->fs_info->delalloc_bytes) {
1195                         printk(KERN_INFO "btrfs warning: delalloc account "
1196                                "%llu %llu\n",
1197                                (unsigned long long)end - start + 1,
1198                                (unsigned long long)
1199                                root->fs_info->delalloc_bytes);
1200                         btrfs_delalloc_free_space(root, inode, (u64)-1);
1201                         root->fs_info->delalloc_bytes = 0;
1202                         BTRFS_I(inode)->delalloc_bytes = 0;
1203                 } else {
1204                         btrfs_delalloc_free_space(root, inode,
1205                                                   end - start + 1);
1206                         root->fs_info->delalloc_bytes -= end - start + 1;
1207                         BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
1208                 }
1209                 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1210                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1211                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1212                 }
1213                 spin_unlock(&root->fs_info->delalloc_lock);
1214         }
1215         return 0;
1216 }
1217
1218 /*
1219  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1220  * we don't create bios that span stripes or chunks
1221  */
1222 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1223                          size_t size, struct bio *bio,
1224                          unsigned long bio_flags)
1225 {
1226         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1227         struct btrfs_mapping_tree *map_tree;
1228         u64 logical = (u64)bio->bi_sector << 9;
1229         u64 length = 0;
1230         u64 map_length;
1231         int ret;
1232
1233         if (bio_flags & EXTENT_BIO_COMPRESSED)
1234                 return 0;
1235
1236         length = bio->bi_size;
1237         map_tree = &root->fs_info->mapping_tree;
1238         map_length = length;
1239         ret = btrfs_map_block(map_tree, READ, logical,
1240                               &map_length, NULL, 0);
1241
1242         if (map_length < length + size)
1243                 return 1;
1244         return 0;
1245 }
1246
1247 /*
1248  * in order to insert checksums into the metadata in large chunks,
1249  * we wait until bio submission time.   All the pages in the bio are
1250  * checksummed and sums are attached onto the ordered extent record.
1251  *
1252  * At IO completion time the cums attached on the ordered extent record
1253  * are inserted into the btree
1254  */
1255 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1256                                     struct bio *bio, int mirror_num,
1257                                     unsigned long bio_flags)
1258 {
1259         struct btrfs_root *root = BTRFS_I(inode)->root;
1260         int ret = 0;
1261
1262         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1263         BUG_ON(ret);
1264         return 0;
1265 }
1266
1267 /*
1268  * in order to insert checksums into the metadata in large chunks,
1269  * we wait until bio submission time.   All the pages in the bio are
1270  * checksummed and sums are attached onto the ordered extent record.
1271  *
1272  * At IO completion time the cums attached on the ordered extent record
1273  * are inserted into the btree
1274  */
1275 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1276                           int mirror_num, unsigned long bio_flags)
1277 {
1278         struct btrfs_root *root = BTRFS_I(inode)->root;
1279         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1280 }
1281
1282 /*
1283  * extent_io.c submission hook. This does the right thing for csum calculation
1284  * on write, or reading the csums from the tree before a read
1285  */
1286 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1287                           int mirror_num, unsigned long bio_flags)
1288 {
1289         struct btrfs_root *root = BTRFS_I(inode)->root;
1290         int ret = 0;
1291         int skip_sum;
1292
1293         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1294
1295         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1296         BUG_ON(ret);
1297
1298         if (!(rw & (1 << BIO_RW))) {
1299                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1300                         return btrfs_submit_compressed_read(inode, bio,
1301                                                     mirror_num, bio_flags);
1302                 } else if (!skip_sum)
1303                         btrfs_lookup_bio_sums(root, inode, bio, NULL);
1304                 goto mapit;
1305         } else if (!skip_sum) {
1306                 /* csum items have already been cloned */
1307                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1308                         goto mapit;
1309                 /* we're doing a write, do the async checksumming */
1310                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1311                                    inode, rw, bio, mirror_num,
1312                                    bio_flags, __btrfs_submit_bio_start,
1313                                    __btrfs_submit_bio_done);
1314         }
1315
1316 mapit:
1317         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1318 }
1319
1320 /*
1321  * given a list of ordered sums record them in the inode.  This happens
1322  * at IO completion time based on sums calculated at bio submission time.
1323  */
1324 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1325                              struct inode *inode, u64 file_offset,
1326                              struct list_head *list)
1327 {
1328         struct btrfs_ordered_sum *sum;
1329
1330         btrfs_set_trans_block_group(trans, inode);
1331
1332         list_for_each_entry(sum, list, list) {
1333                 btrfs_csum_file_blocks(trans,
1334                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1335         }
1336         return 0;
1337 }
1338
1339 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1340 {
1341         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1342                 WARN_ON(1);
1343         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1344                                    GFP_NOFS);
1345 }
1346
1347 /* see btrfs_writepage_start_hook for details on why this is required */
1348 struct btrfs_writepage_fixup {
1349         struct page *page;
1350         struct btrfs_work work;
1351 };
1352
1353 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1354 {
1355         struct btrfs_writepage_fixup *fixup;
1356         struct btrfs_ordered_extent *ordered;
1357         struct page *page;
1358         struct inode *inode;
1359         u64 page_start;
1360         u64 page_end;
1361
1362         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1363         page = fixup->page;
1364 again:
1365         lock_page(page);
1366         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1367                 ClearPageChecked(page);
1368                 goto out_page;
1369         }
1370
1371         inode = page->mapping->host;
1372         page_start = page_offset(page);
1373         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1374
1375         lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1376
1377         /* already ordered? We're done */
1378         if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
1379                              EXTENT_ORDERED, 0)) {
1380                 goto out;
1381         }
1382
1383         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1384         if (ordered) {
1385                 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1386                               page_end, GFP_NOFS);
1387                 unlock_page(page);
1388                 btrfs_start_ordered_extent(inode, ordered, 1);
1389                 goto again;
1390         }
1391
1392         btrfs_set_extent_delalloc(inode, page_start, page_end);
1393         ClearPageChecked(page);
1394 out:
1395         unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1396 out_page:
1397         unlock_page(page);
1398         page_cache_release(page);
1399 }
1400
1401 /*
1402  * There are a few paths in the higher layers of the kernel that directly
1403  * set the page dirty bit without asking the filesystem if it is a
1404  * good idea.  This causes problems because we want to make sure COW
1405  * properly happens and the data=ordered rules are followed.
1406  *
1407  * In our case any range that doesn't have the ORDERED bit set
1408  * hasn't been properly setup for IO.  We kick off an async process
1409  * to fix it up.  The async helper will wait for ordered extents, set
1410  * the delalloc bit and make it safe to write the page.
1411  */
1412 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1413 {
1414         struct inode *inode = page->mapping->host;
1415         struct btrfs_writepage_fixup *fixup;
1416         struct btrfs_root *root = BTRFS_I(inode)->root;
1417         int ret;
1418
1419         ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1420                              EXTENT_ORDERED, 0);
1421         if (ret)
1422                 return 0;
1423
1424         if (PageChecked(page))
1425                 return -EAGAIN;
1426
1427         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1428         if (!fixup)
1429                 return -EAGAIN;
1430
1431         SetPageChecked(page);
1432         page_cache_get(page);
1433         fixup->work.func = btrfs_writepage_fixup_worker;
1434         fixup->page = page;
1435         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1436         return -EAGAIN;
1437 }
1438
1439 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1440                                        struct inode *inode, u64 file_pos,
1441                                        u64 disk_bytenr, u64 disk_num_bytes,
1442                                        u64 num_bytes, u64 ram_bytes,
1443                                        u64 locked_end,
1444                                        u8 compression, u8 encryption,
1445                                        u16 other_encoding, int extent_type)
1446 {
1447         struct btrfs_root *root = BTRFS_I(inode)->root;
1448         struct btrfs_file_extent_item *fi;
1449         struct btrfs_path *path;
1450         struct extent_buffer *leaf;
1451         struct btrfs_key ins;
1452         u64 hint;
1453         int ret;
1454
1455         path = btrfs_alloc_path();
1456         BUG_ON(!path);
1457
1458         path->leave_spinning = 1;
1459         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1460                                  file_pos + num_bytes, locked_end,
1461                                  file_pos, &hint);
1462         BUG_ON(ret);
1463
1464         ins.objectid = inode->i_ino;
1465         ins.offset = file_pos;
1466         ins.type = BTRFS_EXTENT_DATA_KEY;
1467         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1468         BUG_ON(ret);
1469         leaf = path->nodes[0];
1470         fi = btrfs_item_ptr(leaf, path->slots[0],
1471                             struct btrfs_file_extent_item);
1472         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1473         btrfs_set_file_extent_type(leaf, fi, extent_type);
1474         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1475         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1476         btrfs_set_file_extent_offset(leaf, fi, 0);
1477         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1478         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1479         btrfs_set_file_extent_compression(leaf, fi, compression);
1480         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1481         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1482
1483         btrfs_unlock_up_safe(path, 1);
1484         btrfs_set_lock_blocking(leaf);
1485
1486         btrfs_mark_buffer_dirty(leaf);
1487
1488         inode_add_bytes(inode, num_bytes);
1489         btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1490
1491         ins.objectid = disk_bytenr;
1492         ins.offset = disk_num_bytes;
1493         ins.type = BTRFS_EXTENT_ITEM_KEY;
1494         ret = btrfs_alloc_reserved_file_extent(trans, root,
1495                                         root->root_key.objectid,
1496                                         inode->i_ino, file_pos, &ins);
1497         BUG_ON(ret);
1498         btrfs_free_path(path);
1499
1500         return 0;
1501 }
1502
1503 /*
1504  * helper function for btrfs_finish_ordered_io, this
1505  * just reads in some of the csum leaves to prime them into ram
1506  * before we start the transaction.  It limits the amount of btree
1507  * reads required while inside the transaction.
1508  */
1509 static noinline void reada_csum(struct btrfs_root *root,
1510                                 struct btrfs_path *path,
1511                                 struct btrfs_ordered_extent *ordered_extent)
1512 {
1513         struct btrfs_ordered_sum *sum;
1514         u64 bytenr;
1515
1516         sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum,
1517                          list);
1518         bytenr = sum->sums[0].bytenr;
1519
1520         /*
1521          * we don't care about the results, the point of this search is
1522          * just to get the btree leaves into ram
1523          */
1524         btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0);
1525 }
1526
1527 /* as ordered data IO finishes, this gets called so we can finish
1528  * an ordered extent if the range of bytes in the file it covers are
1529  * fully written.
1530  */
1531 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1532 {
1533         struct btrfs_root *root = BTRFS_I(inode)->root;
1534         struct btrfs_trans_handle *trans;
1535         struct btrfs_ordered_extent *ordered_extent = NULL;
1536         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1537         struct btrfs_path *path;
1538         int compressed = 0;
1539         int ret;
1540
1541         ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1542         if (!ret)
1543                 return 0;
1544
1545         /*
1546          * before we join the transaction, try to do some of our IO.
1547          * This will limit the amount of IO that we have to do with
1548          * the transaction running.  We're unlikely to need to do any
1549          * IO if the file extents are new, the disk_i_size checks
1550          * covers the most common case.
1551          */
1552         if (start < BTRFS_I(inode)->disk_i_size) {
1553                 path = btrfs_alloc_path();
1554                 if (path) {
1555                         ret = btrfs_lookup_file_extent(NULL, root, path,
1556                                                        inode->i_ino,
1557                                                        start, 0);
1558                         ordered_extent = btrfs_lookup_ordered_extent(inode,
1559                                                                      start);
1560                         if (!list_empty(&ordered_extent->list)) {
1561                                 btrfs_release_path(root, path);
1562                                 reada_csum(root, path, ordered_extent);
1563                         }
1564                         btrfs_free_path(path);
1565                 }
1566         }
1567
1568         trans = btrfs_join_transaction(root, 1);
1569
1570         if (!ordered_extent)
1571                 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1572         BUG_ON(!ordered_extent);
1573         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1574                 goto nocow;
1575
1576         lock_extent(io_tree, ordered_extent->file_offset,
1577                     ordered_extent->file_offset + ordered_extent->len - 1,
1578                     GFP_NOFS);
1579
1580         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1581                 compressed = 1;
1582         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1583                 BUG_ON(compressed);
1584                 ret = btrfs_mark_extent_written(trans, root, inode,
1585                                                 ordered_extent->file_offset,
1586                                                 ordered_extent->file_offset +
1587                                                 ordered_extent->len);
1588                 BUG_ON(ret);
1589         } else {
1590                 ret = insert_reserved_file_extent(trans, inode,
1591                                                 ordered_extent->file_offset,
1592                                                 ordered_extent->start,
1593                                                 ordered_extent->disk_len,
1594                                                 ordered_extent->len,
1595                                                 ordered_extent->len,
1596                                                 ordered_extent->file_offset +
1597                                                 ordered_extent->len,
1598                                                 compressed, 0, 0,
1599                                                 BTRFS_FILE_EXTENT_REG);
1600                 BUG_ON(ret);
1601         }
1602         unlock_extent(io_tree, ordered_extent->file_offset,
1603                     ordered_extent->file_offset + ordered_extent->len - 1,
1604                     GFP_NOFS);
1605 nocow:
1606         add_pending_csums(trans, inode, ordered_extent->file_offset,
1607                           &ordered_extent->list);
1608
1609         mutex_lock(&BTRFS_I(inode)->extent_mutex);
1610         btrfs_ordered_update_i_size(inode, ordered_extent);
1611         btrfs_update_inode(trans, root, inode);
1612         btrfs_remove_ordered_extent(inode, ordered_extent);
1613         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1614
1615         /* once for us */
1616         btrfs_put_ordered_extent(ordered_extent);
1617         /* once for the tree */
1618         btrfs_put_ordered_extent(ordered_extent);
1619
1620         btrfs_end_transaction(trans, root);
1621         return 0;
1622 }
1623
1624 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1625                                 struct extent_state *state, int uptodate)
1626 {
1627         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1628 }
1629
1630 /*
1631  * When IO fails, either with EIO or csum verification fails, we
1632  * try other mirrors that might have a good copy of the data.  This
1633  * io_failure_record is used to record state as we go through all the
1634  * mirrors.  If another mirror has good data, the page is set up to date
1635  * and things continue.  If a good mirror can't be found, the original
1636  * bio end_io callback is called to indicate things have failed.
1637  */
1638 struct io_failure_record {
1639         struct page *page;
1640         u64 start;
1641         u64 len;
1642         u64 logical;
1643         unsigned long bio_flags;
1644         int last_mirror;
1645 };
1646
1647 static int btrfs_io_failed_hook(struct bio *failed_bio,
1648                          struct page *page, u64 start, u64 end,
1649                          struct extent_state *state)
1650 {
1651         struct io_failure_record *failrec = NULL;
1652         u64 private;
1653         struct extent_map *em;
1654         struct inode *inode = page->mapping->host;
1655         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1656         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1657         struct bio *bio;
1658         int num_copies;
1659         int ret;
1660         int rw;
1661         u64 logical;
1662
1663         ret = get_state_private(failure_tree, start, &private);
1664         if (ret) {
1665                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1666                 if (!failrec)
1667                         return -ENOMEM;
1668                 failrec->start = start;
1669                 failrec->len = end - start + 1;
1670                 failrec->last_mirror = 0;
1671                 failrec->bio_flags = 0;
1672
1673                 spin_lock(&em_tree->lock);
1674                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1675                 if (em->start > start || em->start + em->len < start) {
1676                         free_extent_map(em);
1677                         em = NULL;
1678                 }
1679                 spin_unlock(&em_tree->lock);
1680
1681                 if (!em || IS_ERR(em)) {
1682                         kfree(failrec);
1683                         return -EIO;
1684                 }
1685                 logical = start - em->start;
1686                 logical = em->block_start + logical;
1687                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1688                         logical = em->block_start;
1689                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1690                 }
1691                 failrec->logical = logical;
1692                 free_extent_map(em);
1693                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1694                                 EXTENT_DIRTY, GFP_NOFS);
1695                 set_state_private(failure_tree, start,
1696                                  (u64)(unsigned long)failrec);
1697         } else {
1698                 failrec = (struct io_failure_record *)(unsigned long)private;
1699         }
1700         num_copies = btrfs_num_copies(
1701                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1702                               failrec->logical, failrec->len);
1703         failrec->last_mirror++;
1704         if (!state) {
1705                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1706                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1707                                                     failrec->start,
1708                                                     EXTENT_LOCKED);
1709                 if (state && state->start != failrec->start)
1710                         state = NULL;
1711                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1712         }
1713         if (!state || failrec->last_mirror > num_copies) {
1714                 set_state_private(failure_tree, failrec->start, 0);
1715                 clear_extent_bits(failure_tree, failrec->start,
1716                                   failrec->start + failrec->len - 1,
1717                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1718                 kfree(failrec);
1719                 return -EIO;
1720         }
1721         bio = bio_alloc(GFP_NOFS, 1);
1722         bio->bi_private = state;
1723         bio->bi_end_io = failed_bio->bi_end_io;
1724         bio->bi_sector = failrec->logical >> 9;
1725         bio->bi_bdev = failed_bio->bi_bdev;
1726         bio->bi_size = 0;
1727
1728         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1729         if (failed_bio->bi_rw & (1 << BIO_RW))
1730                 rw = WRITE;
1731         else
1732                 rw = READ;
1733
1734         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1735                                                       failrec->last_mirror,
1736                                                       failrec->bio_flags);
1737         return 0;
1738 }
1739
1740 /*
1741  * each time an IO finishes, we do a fast check in the IO failure tree
1742  * to see if we need to process or clean up an io_failure_record
1743  */
1744 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1745 {
1746         u64 private;
1747         u64 private_failure;
1748         struct io_failure_record *failure;
1749         int ret;
1750
1751         private = 0;
1752         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1753                              (u64)-1, 1, EXTENT_DIRTY)) {
1754                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1755                                         start, &private_failure);
1756                 if (ret == 0) {
1757                         failure = (struct io_failure_record *)(unsigned long)
1758                                    private_failure;
1759                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1760                                           failure->start, 0);
1761                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1762                                           failure->start,
1763                                           failure->start + failure->len - 1,
1764                                           EXTENT_DIRTY | EXTENT_LOCKED,
1765                                           GFP_NOFS);
1766                         kfree(failure);
1767                 }
1768         }
1769         return 0;
1770 }
1771
1772 /*
1773  * when reads are done, we need to check csums to verify the data is correct
1774  * if there's a match, we allow the bio to finish.  If not, we go through
1775  * the io_failure_record routines to find good copies
1776  */
1777 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1778                                struct extent_state *state)
1779 {
1780         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1781         struct inode *inode = page->mapping->host;
1782         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1783         char *kaddr;
1784         u64 private = ~(u32)0;
1785         int ret;
1786         struct btrfs_root *root = BTRFS_I(inode)->root;
1787         u32 csum = ~(u32)0;
1788
1789         if (PageChecked(page)) {
1790                 ClearPageChecked(page);
1791                 goto good;
1792         }
1793
1794         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1795                 return 0;
1796
1797         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1798             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1)) {
1799                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1800                                   GFP_NOFS);
1801                 return 0;
1802         }
1803
1804         if (state && state->start == start) {
1805                 private = state->private;
1806                 ret = 0;
1807         } else {
1808                 ret = get_state_private(io_tree, start, &private);
1809         }
1810         kaddr = kmap_atomic(page, KM_USER0);
1811         if (ret)
1812                 goto zeroit;
1813
1814         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1815         btrfs_csum_final(csum, (char *)&csum);
1816         if (csum != private)
1817                 goto zeroit;
1818
1819         kunmap_atomic(kaddr, KM_USER0);
1820 good:
1821         /* if the io failure tree for this inode is non-empty,
1822          * check to see if we've recovered from a failed IO
1823          */
1824         btrfs_clean_io_failures(inode, start);
1825         return 0;
1826
1827 zeroit:
1828         if (printk_ratelimit()) {
1829                 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1830                        "private %llu\n", page->mapping->host->i_ino,
1831                        (unsigned long long)start, csum,
1832                        (unsigned long long)private);
1833         }
1834         memset(kaddr + offset, 1, end - start + 1);
1835         flush_dcache_page(page);
1836         kunmap_atomic(kaddr, KM_USER0);
1837         if (private == 0)
1838                 return 0;
1839         return -EIO;
1840 }
1841
1842 /*
1843  * This creates an orphan entry for the given inode in case something goes
1844  * wrong in the middle of an unlink/truncate.
1845  */
1846 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1847 {
1848         struct btrfs_root *root = BTRFS_I(inode)->root;
1849         int ret = 0;
1850
1851         spin_lock(&root->list_lock);
1852
1853         /* already on the orphan list, we're good */
1854         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
1855                 spin_unlock(&root->list_lock);
1856                 return 0;
1857         }
1858
1859         list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1860
1861         spin_unlock(&root->list_lock);
1862
1863         /*
1864          * insert an orphan item to track this unlinked/truncated file
1865          */
1866         ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1867
1868         return ret;
1869 }
1870
1871 /*
1872  * We have done the truncate/delete so we can go ahead and remove the orphan
1873  * item for this particular inode.
1874  */
1875 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1876 {
1877         struct btrfs_root *root = BTRFS_I(inode)->root;
1878         int ret = 0;
1879
1880         spin_lock(&root->list_lock);
1881
1882         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
1883                 spin_unlock(&root->list_lock);
1884                 return 0;
1885         }
1886
1887         list_del_init(&BTRFS_I(inode)->i_orphan);
1888         if (!trans) {
1889                 spin_unlock(&root->list_lock);
1890                 return 0;
1891         }
1892
1893         spin_unlock(&root->list_lock);
1894
1895         ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1896
1897         return ret;
1898 }
1899
1900 /*
1901  * this cleans up any orphans that may be left on the list from the last use
1902  * of this root.
1903  */
1904 void btrfs_orphan_cleanup(struct btrfs_root *root)
1905 {
1906         struct btrfs_path *path;
1907         struct extent_buffer *leaf;
1908         struct btrfs_item *item;
1909         struct btrfs_key key, found_key;
1910         struct btrfs_trans_handle *trans;
1911         struct inode *inode;
1912         int ret = 0, nr_unlink = 0, nr_truncate = 0;
1913
1914         path = btrfs_alloc_path();
1915         if (!path)
1916                 return;
1917         path->reada = -1;
1918
1919         key.objectid = BTRFS_ORPHAN_OBJECTID;
1920         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1921         key.offset = (u64)-1;
1922
1923
1924         while (1) {
1925                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1926                 if (ret < 0) {
1927                         printk(KERN_ERR "Error searching slot for orphan: %d"
1928                                "\n", ret);
1929                         break;
1930                 }
1931
1932                 /*
1933                  * if ret == 0 means we found what we were searching for, which
1934                  * is weird, but possible, so only screw with path if we didnt
1935                  * find the key and see if we have stuff that matches
1936                  */
1937                 if (ret > 0) {
1938                         if (path->slots[0] == 0)
1939                                 break;
1940                         path->slots[0]--;
1941                 }
1942
1943                 /* pull out the item */
1944                 leaf = path->nodes[0];
1945                 item = btrfs_item_nr(leaf, path->slots[0]);
1946                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1947
1948                 /* make sure the item matches what we want */
1949                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1950                         break;
1951                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1952                         break;
1953
1954                 /* release the path since we're done with it */
1955                 btrfs_release_path(root, path);
1956
1957                 /*
1958                  * this is where we are basically btrfs_lookup, without the
1959                  * crossing root thing.  we store the inode number in the
1960                  * offset of the orphan item.
1961                  */
1962                 found_key.objectid = found_key.offset;
1963                 found_key.type = BTRFS_INODE_ITEM_KEY;
1964                 found_key.offset = 0;
1965                 inode = btrfs_iget(root->fs_info->sb, &found_key, root);
1966                 if (IS_ERR(inode))
1967                         break;
1968
1969                 /*
1970                  * add this inode to the orphan list so btrfs_orphan_del does
1971                  * the proper thing when we hit it
1972                  */
1973                 spin_lock(&root->list_lock);
1974                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1975                 spin_unlock(&root->list_lock);
1976
1977                 /*
1978                  * if this is a bad inode, means we actually succeeded in
1979                  * removing the inode, but not the orphan record, which means
1980                  * we need to manually delete the orphan since iput will just
1981                  * do a destroy_inode
1982                  */
1983                 if (is_bad_inode(inode)) {
1984                         trans = btrfs_start_transaction(root, 1);
1985                         btrfs_orphan_del(trans, inode);
1986                         btrfs_end_transaction(trans, root);
1987                         iput(inode);
1988                         continue;
1989                 }
1990
1991                 /* if we have links, this was a truncate, lets do that */
1992                 if (inode->i_nlink) {
1993                         nr_truncate++;
1994                         btrfs_truncate(inode);
1995                 } else {
1996                         nr_unlink++;
1997                 }
1998
1999                 /* this will do delete_inode and everything for us */
2000                 iput(inode);
2001         }
2002
2003         if (nr_unlink)
2004                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2005         if (nr_truncate)
2006                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2007
2008         btrfs_free_path(path);
2009 }
2010
2011 /*
2012  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2013  * don't find any xattrs, we know there can't be any acls.
2014  *
2015  * slot is the slot the inode is in, objectid is the objectid of the inode
2016  */
2017 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2018                                           int slot, u64 objectid)
2019 {
2020         u32 nritems = btrfs_header_nritems(leaf);
2021         struct btrfs_key found_key;
2022         int scanned = 0;
2023
2024         slot++;
2025         while (slot < nritems) {
2026                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2027
2028                 /* we found a different objectid, there must not be acls */
2029                 if (found_key.objectid != objectid)
2030                         return 0;
2031
2032                 /* we found an xattr, assume we've got an acl */
2033                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2034                         return 1;
2035
2036                 /*
2037                  * we found a key greater than an xattr key, there can't
2038                  * be any acls later on
2039                  */
2040                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2041                         return 0;
2042
2043                 slot++;
2044                 scanned++;
2045
2046                 /*
2047                  * it goes inode, inode backrefs, xattrs, extents,
2048                  * so if there are a ton of hard links to an inode there can
2049                  * be a lot of backrefs.  Don't waste time searching too hard,
2050                  * this is just an optimization
2051                  */
2052                 if (scanned >= 8)
2053                         break;
2054         }
2055         /* we hit the end of the leaf before we found an xattr or
2056          * something larger than an xattr.  We have to assume the inode
2057          * has acls
2058          */
2059         return 1;
2060 }
2061
2062 /*
2063  * read an inode from the btree into the in-memory inode
2064  */
2065 static void btrfs_read_locked_inode(struct inode *inode)
2066 {
2067         struct btrfs_path *path;
2068         struct extent_buffer *leaf;
2069         struct btrfs_inode_item *inode_item;
2070         struct btrfs_timespec *tspec;
2071         struct btrfs_root *root = BTRFS_I(inode)->root;
2072         struct btrfs_key location;
2073         int maybe_acls;
2074         u64 alloc_group_block;
2075         u32 rdev;
2076         int ret;
2077
2078         path = btrfs_alloc_path();
2079         BUG_ON(!path);
2080         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2081
2082         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2083         if (ret)
2084                 goto make_bad;
2085
2086         leaf = path->nodes[0];
2087         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2088                                     struct btrfs_inode_item);
2089
2090         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2091         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2092         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2093         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2094         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2095
2096         tspec = btrfs_inode_atime(inode_item);
2097         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2098         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2099
2100         tspec = btrfs_inode_mtime(inode_item);
2101         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2102         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2103
2104         tspec = btrfs_inode_ctime(inode_item);
2105         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2106         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2107
2108         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2109         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2110         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2111         inode->i_generation = BTRFS_I(inode)->generation;
2112         inode->i_rdev = 0;
2113         rdev = btrfs_inode_rdev(leaf, inode_item);
2114
2115         BTRFS_I(inode)->index_cnt = (u64)-1;
2116         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2117
2118         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2119
2120         /*
2121          * try to precache a NULL acl entry for files that don't have
2122          * any xattrs or acls
2123          */
2124         maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2125         if (!maybe_acls)
2126                 cache_no_acl(inode);
2127
2128         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2129                                                 alloc_group_block, 0);
2130         btrfs_free_path(path);
2131         inode_item = NULL;
2132
2133         switch (inode->i_mode & S_IFMT) {
2134         case S_IFREG:
2135                 inode->i_mapping->a_ops = &btrfs_aops;
2136                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2137                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2138                 inode->i_fop = &btrfs_file_operations;
2139                 inode->i_op = &btrfs_file_inode_operations;
2140                 break;
2141         case S_IFDIR:
2142                 inode->i_fop = &btrfs_dir_file_operations;
2143                 if (root == root->fs_info->tree_root)
2144                         inode->i_op = &btrfs_dir_ro_inode_operations;
2145                 else
2146                         inode->i_op = &btrfs_dir_inode_operations;
2147                 break;
2148         case S_IFLNK:
2149                 inode->i_op = &btrfs_symlink_inode_operations;
2150                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2151                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2152                 break;
2153         default:
2154                 inode->i_op = &btrfs_special_inode_operations;
2155                 init_special_inode(inode, inode->i_mode, rdev);
2156                 break;
2157         }
2158
2159         btrfs_update_iflags(inode);
2160         return;
2161
2162 make_bad:
2163         btrfs_free_path(path);
2164         make_bad_inode(inode);
2165 }
2166
2167 /*
2168  * given a leaf and an inode, copy the inode fields into the leaf
2169  */
2170 static void fill_inode_item(struct btrfs_trans_handle *trans,
2171                             struct extent_buffer *leaf,
2172                             struct btrfs_inode_item *item,
2173                             struct inode *inode)
2174 {
2175         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2176         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2177         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2178         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2179         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2180
2181         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2182                                inode->i_atime.tv_sec);
2183         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2184                                 inode->i_atime.tv_nsec);
2185
2186         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2187                                inode->i_mtime.tv_sec);
2188         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2189                                 inode->i_mtime.tv_nsec);
2190
2191         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2192                                inode->i_ctime.tv_sec);
2193         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2194                                 inode->i_ctime.tv_nsec);
2195
2196         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2197         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2198         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2199         btrfs_set_inode_transid(leaf, item, trans->transid);
2200         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2201         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2202         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2203 }
2204
2205 /*
2206  * copy everything in the in-memory inode into the btree.
2207  */
2208 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2209                                 struct btrfs_root *root, struct inode *inode)
2210 {
2211         struct btrfs_inode_item *inode_item;
2212         struct btrfs_path *path;
2213         struct extent_buffer *leaf;
2214         int ret;
2215
2216         path = btrfs_alloc_path();
2217         BUG_ON(!path);
2218         path->leave_spinning = 1;
2219         ret = btrfs_lookup_inode(trans, root, path,
2220                                  &BTRFS_I(inode)->location, 1);
2221         if (ret) {
2222                 if (ret > 0)
2223                         ret = -ENOENT;
2224                 goto failed;
2225         }
2226
2227         btrfs_unlock_up_safe(path, 1);
2228         leaf = path->nodes[0];
2229         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2230                                   struct btrfs_inode_item);
2231
2232         fill_inode_item(trans, leaf, inode_item, inode);
2233         btrfs_mark_buffer_dirty(leaf);
2234         btrfs_set_inode_last_trans(trans, inode);
2235         ret = 0;
2236 failed:
2237         btrfs_free_path(path);
2238         return ret;
2239 }
2240
2241
2242 /*
2243  * unlink helper that gets used here in inode.c and in the tree logging
2244  * recovery code.  It remove a link in a directory with a given name, and
2245  * also drops the back refs in the inode to the directory
2246  */
2247 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2248                        struct btrfs_root *root,
2249                        struct inode *dir, struct inode *inode,
2250                        const char *name, int name_len)
2251 {
2252         struct btrfs_path *path;
2253         int ret = 0;
2254         struct extent_buffer *leaf;
2255         struct btrfs_dir_item *di;
2256         struct btrfs_key key;
2257         u64 index;
2258
2259         path = btrfs_alloc_path();
2260         if (!path) {
2261                 ret = -ENOMEM;
2262                 goto err;
2263         }
2264
2265         path->leave_spinning = 1;
2266         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2267                                     name, name_len, -1);
2268         if (IS_ERR(di)) {
2269                 ret = PTR_ERR(di);
2270                 goto err;
2271         }
2272         if (!di) {
2273                 ret = -ENOENT;
2274                 goto err;
2275         }
2276         leaf = path->nodes[0];
2277         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2278         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2279         if (ret)
2280                 goto err;
2281         btrfs_release_path(root, path);
2282
2283         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2284                                   inode->i_ino,
2285                                   dir->i_ino, &index);
2286         if (ret) {
2287                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2288                        "inode %lu parent %lu\n", name_len, name,
2289                        inode->i_ino, dir->i_ino);
2290                 goto err;
2291         }
2292
2293         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2294                                          index, name, name_len, -1);
2295         if (IS_ERR(di)) {
2296                 ret = PTR_ERR(di);
2297                 goto err;
2298         }
2299         if (!di) {
2300                 ret = -ENOENT;
2301                 goto err;
2302         }
2303         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2304         btrfs_release_path(root, path);
2305
2306         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2307                                          inode, dir->i_ino);
2308         BUG_ON(ret != 0 && ret != -ENOENT);
2309
2310         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2311                                            dir, index);
2312         BUG_ON(ret);
2313 err:
2314         btrfs_free_path(path);
2315         if (ret)
2316                 goto out;
2317
2318         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2319         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2320         btrfs_update_inode(trans, root, dir);
2321         btrfs_drop_nlink(inode);
2322         ret = btrfs_update_inode(trans, root, inode);
2323 out:
2324         return ret;
2325 }
2326
2327 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2328 {
2329         struct btrfs_root *root;
2330         struct btrfs_trans_handle *trans;
2331         struct inode *inode = dentry->d_inode;
2332         int ret;
2333         unsigned long nr = 0;
2334
2335         root = BTRFS_I(dir)->root;
2336
2337         trans = btrfs_start_transaction(root, 1);
2338
2339         btrfs_set_trans_block_group(trans, dir);
2340
2341         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2342
2343         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2344                                  dentry->d_name.name, dentry->d_name.len);
2345
2346         if (inode->i_nlink == 0)
2347                 ret = btrfs_orphan_add(trans, inode);
2348
2349         nr = trans->blocks_used;
2350
2351         btrfs_end_transaction_throttle(trans, root);
2352         btrfs_btree_balance_dirty(root, nr);
2353         return ret;
2354 }
2355
2356 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2357 {
2358         struct inode *inode = dentry->d_inode;
2359         int err = 0;
2360         int ret;
2361         struct btrfs_root *root = BTRFS_I(dir)->root;
2362         struct btrfs_trans_handle *trans;
2363         unsigned long nr = 0;
2364
2365         /*
2366          * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2367          * the root of a subvolume or snapshot
2368          */
2369         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2370             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
2371                 return -ENOTEMPTY;
2372         }
2373
2374         trans = btrfs_start_transaction(root, 1);
2375         btrfs_set_trans_block_group(trans, dir);
2376
2377         err = btrfs_orphan_add(trans, inode);
2378         if (err)
2379                 goto fail_trans;
2380
2381         /* now the directory is empty */
2382         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2383                                  dentry->d_name.name, dentry->d_name.len);
2384         if (!err)
2385                 btrfs_i_size_write(inode, 0);
2386
2387 fail_trans:
2388         nr = trans->blocks_used;
2389         ret = btrfs_end_transaction_throttle(trans, root);
2390         btrfs_btree_balance_dirty(root, nr);
2391
2392         if (ret && !err)
2393                 err = ret;
2394         return err;
2395 }
2396
2397 #if 0
2398 /*
2399  * when truncating bytes in a file, it is possible to avoid reading
2400  * the leaves that contain only checksum items.  This can be the
2401  * majority of the IO required to delete a large file, but it must
2402  * be done carefully.
2403  *
2404  * The keys in the level just above the leaves are checked to make sure
2405  * the lowest key in a given leaf is a csum key, and starts at an offset
2406  * after the new  size.
2407  *
2408  * Then the key for the next leaf is checked to make sure it also has
2409  * a checksum item for the same file.  If it does, we know our target leaf
2410  * contains only checksum items, and it can be safely freed without reading
2411  * it.
2412  *
2413  * This is just an optimization targeted at large files.  It may do
2414  * nothing.  It will return 0 unless things went badly.
2415  */
2416 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2417                                      struct btrfs_root *root,
2418                                      struct btrfs_path *path,
2419                                      struct inode *inode, u64 new_size)
2420 {
2421         struct btrfs_key key;
2422         int ret;
2423         int nritems;
2424         struct btrfs_key found_key;
2425         struct btrfs_key other_key;
2426         struct btrfs_leaf_ref *ref;
2427         u64 leaf_gen;
2428         u64 leaf_start;
2429
2430         path->lowest_level = 1;
2431         key.objectid = inode->i_ino;
2432         key.type = BTRFS_CSUM_ITEM_KEY;
2433         key.offset = new_size;
2434 again:
2435         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2436         if (ret < 0)
2437                 goto out;
2438
2439         if (path->nodes[1] == NULL) {
2440                 ret = 0;
2441                 goto out;
2442         }
2443         ret = 0;
2444         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2445         nritems = btrfs_header_nritems(path->nodes[1]);
2446
2447         if (!nritems)
2448                 goto out;
2449
2450         if (path->slots[1] >= nritems)
2451                 goto next_node;
2452
2453         /* did we find a key greater than anything we want to delete? */
2454         if (found_key.objectid > inode->i_ino ||
2455            (found_key.objectid == inode->i_ino && found_key.type > key.type))
2456                 goto out;
2457
2458         /* we check the next key in the node to make sure the leave contains
2459          * only checksum items.  This comparison doesn't work if our
2460          * leaf is the last one in the node
2461          */
2462         if (path->slots[1] + 1 >= nritems) {
2463 next_node:
2464                 /* search forward from the last key in the node, this
2465                  * will bring us into the next node in the tree
2466                  */
2467                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2468
2469                 /* unlikely, but we inc below, so check to be safe */
2470                 if (found_key.offset == (u64)-1)
2471                         goto out;
2472
2473                 /* search_forward needs a path with locks held, do the
2474                  * search again for the original key.  It is possible
2475                  * this will race with a balance and return a path that
2476                  * we could modify, but this drop is just an optimization
2477                  * and is allowed to miss some leaves.
2478                  */
2479                 btrfs_release_path(root, path);
2480                 found_key.offset++;
2481
2482                 /* setup a max key for search_forward */
2483                 other_key.offset = (u64)-1;
2484                 other_key.type = key.type;
2485                 other_key.objectid = key.objectid;
2486
2487                 path->keep_locks = 1;
2488                 ret = btrfs_search_forward(root, &found_key, &other_key,
2489                                            path, 0, 0);
2490                 path->keep_locks = 0;
2491                 if (ret || found_key.objectid != key.objectid ||
2492                     found_key.type != key.type) {
2493                         ret = 0;
2494                         goto out;
2495                 }
2496
2497                 key.offset = found_key.offset;
2498                 btrfs_release_path(root, path);
2499                 cond_resched();
2500                 goto again;
2501         }
2502
2503         /* we know there's one more slot after us in the tree,
2504          * read that key so we can verify it is also a checksum item
2505          */
2506         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2507
2508         if (found_key.objectid < inode->i_ino)
2509                 goto next_key;
2510
2511         if (found_key.type != key.type || found_key.offset < new_size)
2512                 goto next_key;
2513
2514         /*
2515          * if the key for the next leaf isn't a csum key from this objectid,
2516          * we can't be sure there aren't good items inside this leaf.
2517          * Bail out
2518          */
2519         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2520                 goto out;
2521
2522         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2523         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2524         /*
2525          * it is safe to delete this leaf, it contains only
2526          * csum items from this inode at an offset >= new_size
2527          */
2528         ret = btrfs_del_leaf(trans, root, path, leaf_start);
2529         BUG_ON(ret);
2530
2531         if (root->ref_cows && leaf_gen < trans->transid) {
2532                 ref = btrfs_alloc_leaf_ref(root, 0);
2533                 if (ref) {
2534                         ref->root_gen = root->root_key.offset;
2535                         ref->bytenr = leaf_start;
2536                         ref->owner = 0;
2537                         ref->generation = leaf_gen;
2538                         ref->nritems = 0;
2539
2540                         btrfs_sort_leaf_ref(ref);
2541
2542                         ret = btrfs_add_leaf_ref(root, ref, 0);
2543                         WARN_ON(ret);
2544                         btrfs_free_leaf_ref(root, ref);
2545                 } else {
2546                         WARN_ON(1);
2547                 }
2548         }
2549 next_key:
2550         btrfs_release_path(root, path);
2551
2552         if (other_key.objectid == inode->i_ino &&
2553             other_key.type == key.type && other_key.offset > key.offset) {
2554                 key.offset = other_key.offset;
2555                 cond_resched();
2556                 goto again;
2557         }
2558         ret = 0;
2559 out:
2560         /* fixup any changes we've made to the path */
2561         path->lowest_level = 0;
2562         path->keep_locks = 0;
2563         btrfs_release_path(root, path);
2564         return ret;
2565 }
2566
2567 #endif
2568
2569 /*
2570  * this can truncate away extent items, csum items and directory items.
2571  * It starts at a high offset and removes keys until it can't find
2572  * any higher than new_size
2573  *
2574  * csum items that cross the new i_size are truncated to the new size
2575  * as well.
2576  *
2577  * min_type is the minimum key type to truncate down to.  If set to 0, this
2578  * will kill all the items on this inode, including the INODE_ITEM_KEY.
2579  */
2580 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2581                                         struct btrfs_root *root,
2582                                         struct inode *inode,
2583                                         u64 new_size, u32 min_type)
2584 {
2585         int ret;
2586         struct btrfs_path *path;
2587         struct btrfs_key key;
2588         struct btrfs_key found_key;
2589         u32 found_type = (u8)-1;
2590         struct extent_buffer *leaf;
2591         struct btrfs_file_extent_item *fi;
2592         u64 extent_start = 0;
2593         u64 extent_num_bytes = 0;
2594         u64 extent_offset = 0;
2595         u64 item_end = 0;
2596         int found_extent;
2597         int del_item;
2598         int pending_del_nr = 0;
2599         int pending_del_slot = 0;
2600         int extent_type = -1;
2601         int encoding;
2602         u64 mask = root->sectorsize - 1;
2603
2604         if (root->ref_cows)
2605                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2606         path = btrfs_alloc_path();
2607         path->reada = -1;
2608         BUG_ON(!path);
2609
2610         /* FIXME, add redo link to tree so we don't leak on crash */
2611         key.objectid = inode->i_ino;
2612         key.offset = (u64)-1;
2613         key.type = (u8)-1;
2614
2615 search_again:
2616         path->leave_spinning = 1;
2617         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2618         if (ret < 0)
2619                 goto error;
2620
2621         if (ret > 0) {
2622                 /* there are no items in the tree for us to truncate, we're
2623                  * done
2624                  */
2625                 if (path->slots[0] == 0) {
2626                         ret = 0;
2627                         goto error;
2628                 }
2629                 path->slots[0]--;
2630         }
2631
2632         while (1) {
2633                 fi = NULL;
2634                 leaf = path->nodes[0];
2635                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2636                 found_type = btrfs_key_type(&found_key);
2637                 encoding = 0;
2638
2639                 if (found_key.objectid != inode->i_ino)
2640                         break;
2641
2642                 if (found_type < min_type)
2643                         break;
2644
2645                 item_end = found_key.offset;
2646                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2647                         fi = btrfs_item_ptr(leaf, path->slots[0],
2648                                             struct btrfs_file_extent_item);
2649                         extent_type = btrfs_file_extent_type(leaf, fi);
2650                         encoding = btrfs_file_extent_compression(leaf, fi);
2651                         encoding |= btrfs_file_extent_encryption(leaf, fi);
2652                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2653
2654                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2655                                 item_end +=
2656                                     btrfs_file_extent_num_bytes(leaf, fi);
2657                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2658                                 item_end += btrfs_file_extent_inline_len(leaf,
2659                                                                          fi);
2660                         }
2661                         item_end--;
2662                 }
2663                 if (item_end < new_size) {
2664                         if (found_type == BTRFS_DIR_ITEM_KEY)
2665                                 found_type = BTRFS_INODE_ITEM_KEY;
2666                         else if (found_type == BTRFS_EXTENT_ITEM_KEY)
2667                                 found_type = BTRFS_EXTENT_DATA_KEY;
2668                         else if (found_type == BTRFS_EXTENT_DATA_KEY)
2669                                 found_type = BTRFS_XATTR_ITEM_KEY;
2670                         else if (found_type == BTRFS_XATTR_ITEM_KEY)
2671                                 found_type = BTRFS_INODE_REF_KEY;
2672                         else if (found_type)
2673                                 found_type--;
2674                         else
2675                                 break;
2676                         btrfs_set_key_type(&key, found_type);
2677                         goto next;
2678                 }
2679                 if (found_key.offset >= new_size)
2680                         del_item = 1;
2681                 else
2682                         del_item = 0;
2683                 found_extent = 0;
2684
2685                 /* FIXME, shrink the extent if the ref count is only 1 */
2686                 if (found_type != BTRFS_EXTENT_DATA_KEY)
2687                         goto delete;
2688
2689                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2690                         u64 num_dec;
2691                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2692                         if (!del_item && !encoding) {
2693                                 u64 orig_num_bytes =
2694                                         btrfs_file_extent_num_bytes(leaf, fi);
2695                                 extent_num_bytes = new_size -
2696                                         found_key.offset + root->sectorsize - 1;
2697                                 extent_num_bytes = extent_num_bytes &
2698                                         ~((u64)root->sectorsize - 1);
2699                                 btrfs_set_file_extent_num_bytes(leaf, fi,
2700                                                          extent_num_bytes);
2701                                 num_dec = (orig_num_bytes -
2702                                            extent_num_bytes);
2703                                 if (root->ref_cows && extent_start != 0)
2704                                         inode_sub_bytes(inode, num_dec);
2705                                 btrfs_mark_buffer_dirty(leaf);
2706                         } else {
2707                                 extent_num_bytes =
2708                                         btrfs_file_extent_disk_num_bytes(leaf,
2709                                                                          fi);
2710                                 extent_offset = found_key.offset -
2711                                         btrfs_file_extent_offset(leaf, fi);
2712
2713                                 /* FIXME blocksize != 4096 */
2714                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2715                                 if (extent_start != 0) {
2716                                         found_extent = 1;
2717                                         if (root->ref_cows)
2718                                                 inode_sub_bytes(inode, num_dec);
2719                                 }
2720                         }
2721                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2722                         /*
2723                          * we can't truncate inline items that have had
2724                          * special encodings
2725                          */
2726                         if (!del_item &&
2727                             btrfs_file_extent_compression(leaf, fi) == 0 &&
2728                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
2729                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2730                                 u32 size = new_size - found_key.offset;
2731
2732                                 if (root->ref_cows) {
2733                                         inode_sub_bytes(inode, item_end + 1 -
2734                                                         new_size);
2735                                 }
2736                                 size =
2737                                     btrfs_file_extent_calc_inline_size(size);
2738                                 ret = btrfs_truncate_item(trans, root, path,
2739                                                           size, 1);
2740                                 BUG_ON(ret);
2741                         } else if (root->ref_cows) {
2742                                 inode_sub_bytes(inode, item_end + 1 -
2743                                                 found_key.offset);
2744                         }
2745                 }
2746 delete:
2747                 if (del_item) {
2748                         if (!pending_del_nr) {
2749                                 /* no pending yet, add ourselves */
2750                                 pending_del_slot = path->slots[0];
2751                                 pending_del_nr = 1;
2752                         } else if (pending_del_nr &&
2753                                    path->slots[0] + 1 == pending_del_slot) {
2754                                 /* hop on the pending chunk */
2755                                 pending_del_nr++;
2756                                 pending_del_slot = path->slots[0];
2757                         } else {
2758                                 BUG();
2759                         }
2760                 } else {
2761                         break;
2762                 }
2763                 if (found_extent && root->ref_cows) {
2764                         btrfs_set_path_blocking(path);
2765                         ret = btrfs_free_extent(trans, root, extent_start,
2766                                                 extent_num_bytes, 0,
2767                                                 btrfs_header_owner(leaf),
2768                                                 inode->i_ino, extent_offset);
2769                         BUG_ON(ret);
2770                 }
2771 next:
2772                 if (path->slots[0] == 0) {
2773                         if (pending_del_nr)
2774                                 goto del_pending;
2775                         btrfs_release_path(root, path);
2776                         if (found_type == BTRFS_INODE_ITEM_KEY)
2777                                 break;
2778                         goto search_again;
2779                 }
2780
2781                 path->slots[0]--;
2782                 if (pending_del_nr &&
2783                     path->slots[0] + 1 != pending_del_slot) {
2784                         struct btrfs_key debug;
2785 del_pending:
2786                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
2787                                               pending_del_slot);
2788                         ret = btrfs_del_items(trans, root, path,
2789                                               pending_del_slot,
2790                                               pending_del_nr);
2791                         BUG_ON(ret);
2792                         pending_del_nr = 0;
2793                         btrfs_release_path(root, path);
2794                         if (found_type == BTRFS_INODE_ITEM_KEY)
2795                                 break;
2796                         goto search_again;
2797                 }
2798         }
2799         ret = 0;
2800 error:
2801         if (pending_del_nr) {
2802                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2803                                       pending_del_nr);
2804         }
2805         btrfs_free_path(path);
2806         return ret;
2807 }
2808
2809 /*
2810  * taken from block_truncate_page, but does cow as it zeros out
2811  * any bytes left in the last page in the file.
2812  */
2813 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2814 {
2815         struct inode *inode = mapping->host;
2816         struct btrfs_root *root = BTRFS_I(inode)->root;
2817         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2818         struct btrfs_ordered_extent *ordered;
2819         char *kaddr;
2820         u32 blocksize = root->sectorsize;
2821         pgoff_t index = from >> PAGE_CACHE_SHIFT;
2822         unsigned offset = from & (PAGE_CACHE_SIZE-1);
2823         struct page *page;
2824         int ret = 0;
2825         u64 page_start;
2826         u64 page_end;
2827
2828         if ((offset & (blocksize - 1)) == 0)
2829                 goto out;
2830
2831         ret = -ENOMEM;
2832 again:
2833         page = grab_cache_page(mapping, index);
2834         if (!page)
2835                 goto out;
2836
2837         page_start = page_offset(page);
2838         page_end = page_start + PAGE_CACHE_SIZE - 1;
2839
2840         if (!PageUptodate(page)) {
2841                 ret = btrfs_readpage(NULL, page);
2842                 lock_page(page);
2843                 if (page->mapping != mapping) {
2844                         unlock_page(page);
2845                         page_cache_release(page);
2846                         goto again;
2847                 }
2848                 if (!PageUptodate(page)) {
2849                         ret = -EIO;
2850                         goto out_unlock;
2851                 }
2852         }
2853         wait_on_page_writeback(page);
2854
2855         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2856         set_page_extent_mapped(page);
2857
2858         ordered = btrfs_lookup_ordered_extent(inode, page_start);
2859         if (ordered) {
2860                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2861                 unlock_page(page);
2862                 page_cache_release(page);
2863                 btrfs_start_ordered_extent(inode, ordered, 1);
2864                 btrfs_put_ordered_extent(ordered);
2865                 goto again;
2866         }
2867
2868         btrfs_set_extent_delalloc(inode, page_start, page_end);
2869         ret = 0;
2870         if (offset != PAGE_CACHE_SIZE) {
2871                 kaddr = kmap(page);
2872                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2873                 flush_dcache_page(page);
2874                 kunmap(page);
2875         }
2876         ClearPageChecked(page);
2877         set_page_dirty(page);
2878         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2879
2880 out_unlock:
2881         unlock_page(page);
2882         page_cache_release(page);
2883 out:
2884         return ret;
2885 }
2886
2887 int btrfs_cont_expand(struct inode *inode, loff_t size)
2888 {
2889         struct btrfs_trans_handle *trans;
2890         struct btrfs_root *root = BTRFS_I(inode)->root;
2891         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2892         struct extent_map *em;
2893         u64 mask = root->sectorsize - 1;
2894         u64 hole_start = (inode->i_size + mask) & ~mask;
2895         u64 block_end = (size + mask) & ~mask;
2896         u64 last_byte;
2897         u64 cur_offset;
2898         u64 hole_size;
2899         int err;
2900
2901         if (size <= hole_start)
2902                 return 0;
2903
2904         err = btrfs_check_metadata_free_space(root);
2905         if (err)
2906                 return err;
2907
2908         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2909
2910         while (1) {
2911                 struct btrfs_ordered_extent *ordered;
2912                 btrfs_wait_ordered_range(inode, hole_start,
2913                                          block_end - hole_start);
2914                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2915                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2916                 if (!ordered)
2917                         break;
2918                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2919                 btrfs_put_ordered_extent(ordered);
2920         }
2921
2922         trans = btrfs_start_transaction(root, 1);
2923         btrfs_set_trans_block_group(trans, inode);
2924
2925         cur_offset = hole_start;
2926         while (1) {
2927                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2928                                 block_end - cur_offset, 0);
2929                 BUG_ON(IS_ERR(em) || !em);
2930                 last_byte = min(extent_map_end(em), block_end);
2931                 last_byte = (last_byte + mask) & ~mask;
2932                 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2933                         u64 hint_byte = 0;
2934                         hole_size = last_byte - cur_offset;
2935                         err = btrfs_drop_extents(trans, root, inode,
2936                                                  cur_offset,
2937                                                  cur_offset + hole_size,
2938                                                  block_end,
2939                                                  cur_offset, &hint_byte);
2940                         if (err)
2941                                 break;
2942                         err = btrfs_insert_file_extent(trans, root,
2943                                         inode->i_ino, cur_offset, 0,
2944                                         0, hole_size, 0, hole_size,
2945                                         0, 0, 0);
2946                         btrfs_drop_extent_cache(inode, hole_start,
2947                                         last_byte - 1, 0);
2948                 }
2949                 free_extent_map(em);
2950                 cur_offset = last_byte;
2951                 if (err || cur_offset >= block_end)
2952                         break;
2953         }
2954
2955         btrfs_end_transaction(trans, root);
2956         unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2957         return err;
2958 }
2959
2960 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2961 {
2962         struct inode *inode = dentry->d_inode;
2963         int err;
2964
2965         err = inode_change_ok(inode, attr);
2966         if (err)
2967                 return err;
2968
2969         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
2970                 if (attr->ia_size > inode->i_size) {
2971                         err = btrfs_cont_expand(inode, attr->ia_size);
2972                         if (err)
2973                                 return err;
2974                 } else if (inode->i_size > 0 &&
2975                            attr->ia_size == 0) {
2976
2977                         /* we're truncating a file that used to have good
2978                          * data down to zero.  Make sure it gets into
2979                          * the ordered flush list so that any new writes
2980                          * get down to disk quickly.
2981                          */
2982                         BTRFS_I(inode)->ordered_data_close = 1;
2983                 }
2984         }
2985
2986         err = inode_setattr(inode, attr);
2987
2988         if (!err && ((attr->ia_valid & ATTR_MODE)))
2989                 err = btrfs_acl_chmod(inode);
2990         return err;
2991 }
2992
2993 void btrfs_delete_inode(struct inode *inode)
2994 {
2995         struct btrfs_trans_handle *trans;
2996         struct btrfs_root *root = BTRFS_I(inode)->root;
2997         unsigned long nr;
2998         int ret;
2999
3000         truncate_inode_pages(&inode->i_data, 0);
3001         if (is_bad_inode(inode)) {
3002                 btrfs_orphan_del(NULL, inode);
3003                 goto no_delete;
3004         }
3005         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3006
3007         btrfs_i_size_write(inode, 0);
3008         trans = btrfs_join_transaction(root, 1);
3009
3010         btrfs_set_trans_block_group(trans, inode);
3011         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
3012         if (ret) {
3013                 btrfs_orphan_del(NULL, inode);
3014                 goto no_delete_lock;
3015         }
3016
3017         btrfs_orphan_del(trans, inode);
3018
3019         nr = trans->blocks_used;
3020         clear_inode(inode);
3021
3022         btrfs_end_transaction(trans, root);
3023         btrfs_btree_balance_dirty(root, nr);
3024         return;
3025
3026 no_delete_lock:
3027         nr = trans->blocks_used;
3028         btrfs_end_transaction(trans, root);
3029         btrfs_btree_balance_dirty(root, nr);
3030 no_delete:
3031         clear_inode(inode);
3032 }
3033
3034 /*
3035  * this returns the key found in the dir entry in the location pointer.
3036  * If no dir entries were found, location->objectid is 0.
3037  */
3038 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3039                                struct btrfs_key *location)
3040 {
3041         const char *name = dentry->d_name.name;
3042         int namelen = dentry->d_name.len;
3043         struct btrfs_dir_item *di;
3044         struct btrfs_path *path;
3045         struct btrfs_root *root = BTRFS_I(dir)->root;
3046         int ret = 0;
3047
3048         path = btrfs_alloc_path();
3049         BUG_ON(!path);
3050
3051         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3052                                     namelen, 0);
3053         if (IS_ERR(di))
3054                 ret = PTR_ERR(di);
3055
3056         if (!di || IS_ERR(di))
3057                 goto out_err;
3058
3059         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3060 out:
3061         btrfs_free_path(path);
3062         return ret;
3063 out_err:
3064         location->objectid = 0;
3065         goto out;
3066 }
3067
3068 /*
3069  * when we hit a tree root in a directory, the btrfs part of the inode
3070  * needs to be changed to reflect the root directory of the tree root.  This
3071  * is kind of like crossing a mount point.
3072  */
3073 static int fixup_tree_root_location(struct btrfs_root *root,
3074                              struct btrfs_key *location,
3075                              struct btrfs_root **sub_root,
3076                              struct dentry *dentry)
3077 {
3078         struct btrfs_root_item *ri;
3079
3080         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
3081                 return 0;
3082         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
3083                 return 0;
3084
3085         *sub_root = btrfs_read_fs_root(root->fs_info, location,
3086                                         dentry->d_name.name,
3087                                         dentry->d_name.len);
3088         if (IS_ERR(*sub_root))
3089                 return PTR_ERR(*sub_root);
3090
3091         ri = &(*sub_root)->root_item;
3092         location->objectid = btrfs_root_dirid(ri);
3093         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3094         location->offset = 0;
3095
3096         return 0;
3097 }
3098
3099 static void inode_tree_add(struct inode *inode)
3100 {
3101         struct btrfs_root *root = BTRFS_I(inode)->root;
3102         struct btrfs_inode *entry;
3103         struct rb_node **p = &root->inode_tree.rb_node;
3104         struct rb_node *parent = NULL;
3105
3106         spin_lock(&root->inode_lock);
3107         while (*p) {
3108                 parent = *p;
3109                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3110
3111                 if (inode->i_ino < entry->vfs_inode.i_ino)
3112                         p = &(*p)->rb_left;
3113                 else if (inode->i_ino > entry->vfs_inode.i_ino)
3114                         p = &(*p)->rb_right;
3115                 else {
3116                         WARN_ON(!(entry->vfs_inode.i_state &
3117                                   (I_WILL_FREE | I_FREEING | I_CLEAR)));
3118                         break;
3119                 }
3120         }
3121         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3122         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3123         spin_unlock(&root->inode_lock);
3124 }
3125
3126 static void inode_tree_del(struct inode *inode)
3127 {
3128         struct btrfs_root *root = BTRFS_I(inode)->root;
3129
3130         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3131                 spin_lock(&root->inode_lock);
3132                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3133                 spin_unlock(&root->inode_lock);
3134                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3135         }
3136 }
3137
3138 static noinline void init_btrfs_i(struct inode *inode)
3139 {
3140         struct btrfs_inode *bi = BTRFS_I(inode);
3141
3142         bi->generation = 0;
3143         bi->sequence = 0;
3144         bi->last_trans = 0;
3145         bi->logged_trans = 0;
3146         bi->delalloc_bytes = 0;
3147         bi->reserved_bytes = 0;
3148         bi->disk_i_size = 0;
3149         bi->flags = 0;
3150         bi->index_cnt = (u64)-1;
3151         bi->last_unlink_trans = 0;
3152         bi->ordered_data_close = 0;
3153         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3154         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3155                              inode->i_mapping, GFP_NOFS);
3156         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3157                              inode->i_mapping, GFP_NOFS);
3158         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3159         INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations);
3160         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3161         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3162         mutex_init(&BTRFS_I(inode)->extent_mutex);
3163         mutex_init(&BTRFS_I(inode)->log_mutex);
3164 }
3165
3166 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3167 {
3168         struct btrfs_iget_args *args = p;
3169         inode->i_ino = args->ino;
3170         init_btrfs_i(inode);
3171         BTRFS_I(inode)->root = args->root;
3172         btrfs_set_inode_space_info(args->root, inode);
3173         return 0;
3174 }
3175
3176 static int btrfs_find_actor(struct inode *inode, void *opaque)
3177 {
3178         struct btrfs_iget_args *args = opaque;
3179         return args->ino == inode->i_ino &&
3180                 args->root == BTRFS_I(inode)->root;
3181 }
3182
3183 static struct inode *btrfs_iget_locked(struct super_block *s,
3184                                        u64 objectid,
3185                                        struct btrfs_root *root)
3186 {
3187         struct inode *inode;
3188         struct btrfs_iget_args args;
3189         args.ino = objectid;
3190         args.root = root;
3191
3192         inode = iget5_locked(s, objectid, btrfs_find_actor,
3193                              btrfs_init_locked_inode,
3194                              (void *)&args);
3195         return inode;
3196 }
3197
3198 /* Get an inode object given its location and corresponding root.
3199  * Returns in *is_new if the inode was read from disk
3200  */
3201 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3202                          struct btrfs_root *root)
3203 {
3204         struct inode *inode;
3205
3206         inode = btrfs_iget_locked(s, location->objectid, root);
3207         if (!inode)
3208                 return ERR_PTR(-ENOMEM);
3209
3210         if (inode->i_state & I_NEW) {
3211                 BTRFS_I(inode)->root = root;
3212                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3213                 btrfs_read_locked_inode(inode);
3214
3215                 inode_tree_add(inode);
3216                 unlock_new_inode(inode);
3217         }
3218
3219         return inode;
3220 }
3221
3222 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3223 {
3224         struct inode *inode;
3225         struct btrfs_inode *bi = BTRFS_I(dir);
3226         struct btrfs_root *root = bi->root;
3227         struct btrfs_root *sub_root = root;
3228         struct btrfs_key location;
3229         int ret;
3230
3231         if (dentry->d_name.len > BTRFS_NAME_LEN)
3232                 return ERR_PTR(-ENAMETOOLONG);
3233
3234         ret = btrfs_inode_by_name(dir, dentry, &location);
3235
3236         if (ret < 0)
3237                 return ERR_PTR(ret);
3238
3239         inode = NULL;
3240         if (location.objectid) {
3241                 ret = fixup_tree_root_location(root, &location, &sub_root,
3242                                                 dentry);
3243                 if (ret < 0)
3244                         return ERR_PTR(ret);
3245                 if (ret > 0)
3246                         return ERR_PTR(-ENOENT);
3247                 inode = btrfs_iget(dir->i_sb, &location, sub_root);
3248                 if (IS_ERR(inode))
3249                         return ERR_CAST(inode);
3250         }
3251         return inode;
3252 }
3253
3254 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3255                                    struct nameidata *nd)
3256 {
3257         struct inode *inode;
3258
3259         if (dentry->d_name.len > BTRFS_NAME_LEN)
3260                 return ERR_PTR(-ENAMETOOLONG);
3261
3262         inode = btrfs_lookup_dentry(dir, dentry);
3263         if (IS_ERR(inode))
3264                 return ERR_CAST(inode);
3265
3266         return d_splice_alias(inode, dentry);
3267 }
3268
3269 static unsigned char btrfs_filetype_table[] = {
3270         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3271 };
3272
3273 static int btrfs_real_readdir(struct file *filp, void *dirent,
3274                               filldir_t filldir)
3275 {
3276         struct inode *inode = filp->f_dentry->d_inode;
3277         struct btrfs_root *root = BTRFS_I(inode)->root;
3278         struct btrfs_item *item;
3279         struct btrfs_dir_item *di;
3280         struct btrfs_key key;
3281         struct btrfs_key found_key;
3282         struct btrfs_path *path;
3283         int ret;
3284         u32 nritems;
3285         struct extent_buffer *leaf;
3286         int slot;
3287         int advance;
3288         unsigned char d_type;
3289         int over = 0;
3290         u32 di_cur;
3291         u32 di_total;
3292         u32 di_len;
3293         int key_type = BTRFS_DIR_INDEX_KEY;
3294         char tmp_name[32];
3295         char *name_ptr;
3296         int name_len;
3297
3298         /* FIXME, use a real flag for deciding about the key type */
3299         if (root->fs_info->tree_root == root)
3300                 key_type = BTRFS_DIR_ITEM_KEY;
3301
3302         /* special case for "." */
3303         if (filp->f_pos == 0) {
3304                 over = filldir(dirent, ".", 1,
3305                                1, inode->i_ino,
3306                                DT_DIR);
3307                 if (over)
3308                         return 0;
3309                 filp->f_pos = 1;
3310         }
3311         /* special case for .., just use the back ref */
3312         if (filp->f_pos == 1) {
3313                 u64 pino = parent_ino(filp->f_path.dentry);
3314                 over = filldir(dirent, "..", 2,
3315                                2, pino, DT_DIR);
3316                 if (over)
3317                         return 0;
3318                 filp->f_pos = 2;
3319         }
3320         path = btrfs_alloc_path();
3321         path->reada = 2;
3322
3323         btrfs_set_key_type(&key, key_type);
3324         key.offset = filp->f_pos;
3325         key.objectid = inode->i_ino;
3326
3327         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3328         if (ret < 0)
3329                 goto err;
3330         advance = 0;
3331
3332         while (1) {
3333                 leaf = path->nodes[0];
3334                 nritems = btrfs_header_nritems(leaf);
3335                 slot = path->slots[0];
3336                 if (advance || slot >= nritems) {
3337                         if (slot >= nritems - 1) {
3338                                 ret = btrfs_next_leaf(root, path);
3339                                 if (ret)
3340                                         break;
3341                                 leaf = path->nodes[0];
3342                                 nritems = btrfs_header_nritems(leaf);
3343                                 slot = path->slots[0];
3344                         } else {
3345                                 slot++;
3346                                 path->slots[0]++;
3347                         }
3348                 }
3349
3350                 advance = 1;
3351                 item = btrfs_item_nr(leaf, slot);
3352                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3353
3354                 if (found_key.objectid != key.objectid)
3355                         break;
3356                 if (btrfs_key_type(&found_key) != key_type)
3357                         break;
3358                 if (found_key.offset < filp->f_pos)
3359                         continue;
3360
3361                 filp->f_pos = found_key.offset;
3362
3363                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3364                 di_cur = 0;
3365                 di_total = btrfs_item_size(leaf, item);
3366
3367                 while (di_cur < di_total) {
3368                         struct btrfs_key location;
3369
3370                         name_len = btrfs_dir_name_len(leaf, di);
3371                         if (name_len <= sizeof(tmp_name)) {
3372                                 name_ptr = tmp_name;
3373                         } else {
3374                                 name_ptr = kmalloc(name_len, GFP_NOFS);
3375                                 if (!name_ptr) {
3376                                         ret = -ENOMEM;
3377                                         goto err;
3378                                 }
3379                         }
3380                         read_extent_buffer(leaf, name_ptr,
3381                                            (unsigned long)(di + 1), name_len);
3382
3383                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3384                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
3385
3386                         /* is this a reference to our own snapshot? If so
3387                          * skip it
3388                          */
3389                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
3390                             location.objectid == root->root_key.objectid) {
3391                                 over = 0;
3392                                 goto skip;
3393                         }
3394                         over = filldir(dirent, name_ptr, name_len,
3395                                        found_key.offset, location.objectid,
3396                                        d_type);
3397
3398 skip:
3399                         if (name_ptr != tmp_name)
3400                                 kfree(name_ptr);
3401
3402                         if (over)
3403                                 goto nopos;
3404                         di_len = btrfs_dir_name_len(leaf, di) +
3405                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
3406                         di_cur += di_len;
3407                         di = (struct btrfs_dir_item *)((char *)di + di_len);
3408                 }
3409         }
3410
3411         /* Reached end of directory/root. Bump pos past the last item. */
3412         if (key_type == BTRFS_DIR_INDEX_KEY)
3413                 filp->f_pos = INT_LIMIT(off_t);
3414         else
3415                 filp->f_pos++;
3416 nopos:
3417         ret = 0;
3418 err:
3419         btrfs_free_path(path);
3420         return ret;
3421 }
3422
3423 int btrfs_write_inode(struct inode *inode, int wait)
3424 {
3425         struct btrfs_root *root = BTRFS_I(inode)->root;
3426         struct btrfs_trans_handle *trans;
3427         int ret = 0;
3428
3429         if (root->fs_info->btree_inode == inode)
3430                 return 0;
3431
3432         if (wait) {
3433                 trans = btrfs_join_transaction(root, 1);
3434                 btrfs_set_trans_block_group(trans, inode);
3435                 ret = btrfs_commit_transaction(trans, root);
3436         }
3437         return ret;
3438 }
3439
3440 /*
3441  * This is somewhat expensive, updating the tree every time the
3442  * inode changes.  But, it is most likely to find the inode in cache.
3443  * FIXME, needs more benchmarking...there are no reasons other than performance
3444  * to keep or drop this code.
3445  */
3446 void btrfs_dirty_inode(struct inode *inode)
3447 {
3448         struct btrfs_root *root = BTRFS_I(inode)->root;
3449         struct btrfs_trans_handle *trans;
3450
3451         trans = btrfs_join_transaction(root, 1);
3452         btrfs_set_trans_block_group(trans, inode);
3453         btrfs_update_inode(trans, root, inode);
3454         btrfs_end_transaction(trans, root);
3455 }
3456
3457 /*
3458  * find the highest existing sequence number in a directory
3459  * and then set the in-memory index_cnt variable to reflect
3460  * free sequence numbers
3461  */
3462 static int btrfs_set_inode_index_count(struct inode *inode)
3463 {
3464         struct btrfs_root *root = BTRFS_I(inode)->root;
3465         struct btrfs_key key, found_key;
3466         struct btrfs_path *path;
3467         struct extent_buffer *leaf;
3468         int ret;
3469
3470         key.objectid = inode->i_ino;
3471         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3472         key.offset = (u64)-1;
3473
3474         path = btrfs_alloc_path();
3475         if (!path)
3476                 return -ENOMEM;
3477
3478         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3479         if (ret < 0)
3480                 goto out;
3481         /* FIXME: we should be able to handle this */
3482         if (ret == 0)
3483                 goto out;
3484         ret = 0;
3485
3486         /*
3487          * MAGIC NUMBER EXPLANATION:
3488          * since we search a directory based on f_pos we have to start at 2
3489          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3490          * else has to start at 2
3491          */
3492         if (path->slots[0] == 0) {
3493                 BTRFS_I(inode)->index_cnt = 2;
3494                 goto out;
3495         }
3496
3497         path->slots[0]--;
3498
3499         leaf = path->nodes[0];
3500         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3501
3502         if (found_key.objectid != inode->i_ino ||
3503             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3504                 BTRFS_I(inode)->index_cnt = 2;
3505                 goto out;
3506         }
3507
3508         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3509 out:
3510         btrfs_free_path(path);
3511         return ret;
3512 }
3513
3514 /*
3515  * helper to find a free sequence number in a given directory.  This current
3516  * code is very simple, later versions will do smarter things in the btree
3517  */
3518 int btrfs_set_inode_index(struct inode *dir, u64 *index)
3519 {
3520         int ret = 0;
3521
3522         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3523                 ret = btrfs_set_inode_index_count(dir);
3524                 if (ret)
3525                         return ret;
3526         }
3527
3528         *index = BTRFS_I(dir)->index_cnt;
3529         BTRFS_I(dir)->index_cnt++;
3530
3531         return ret;
3532 }
3533
3534 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3535                                      struct btrfs_root *root,
3536                                      struct inode *dir,
3537                                      const char *name, int name_len,
3538                                      u64 ref_objectid, u64 objectid,
3539                                      u64 alloc_hint, int mode, u64 *index)
3540 {
3541         struct inode *inode;
3542         struct btrfs_inode_item *inode_item;
3543         struct btrfs_key *location;
3544         struct btrfs_path *path;
3545         struct btrfs_inode_ref *ref;
3546         struct btrfs_key key[2];
3547         u32 sizes[2];
3548         unsigned long ptr;
3549         int ret;
3550         int owner;
3551
3552         path = btrfs_alloc_path();
3553         BUG_ON(!path);
3554
3555         inode = new_inode(root->fs_info->sb);
3556         if (!inode)
3557                 return ERR_PTR(-ENOMEM);
3558
3559         if (dir) {
3560                 ret = btrfs_set_inode_index(dir, index);
3561                 if (ret) {
3562                         iput(inode);
3563                         return ERR_PTR(ret);
3564                 }
3565         }
3566         /*
3567          * index_cnt is ignored for everything but a dir,
3568          * btrfs_get_inode_index_count has an explanation for the magic
3569          * number
3570          */
3571         init_btrfs_i(inode);
3572         BTRFS_I(inode)->index_cnt = 2;
3573         BTRFS_I(inode)->root = root;
3574         BTRFS_I(inode)->generation = trans->transid;
3575         btrfs_set_inode_space_info(root, inode);
3576
3577         if (mode & S_IFDIR)
3578                 owner = 0;
3579         else
3580                 owner = 1;
3581         BTRFS_I(inode)->block_group =
3582                         btrfs_find_block_group(root, 0, alloc_hint, owner);
3583         if ((mode & S_IFREG)) {
3584                 if (btrfs_test_opt(root, NODATASUM))
3585                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
3586                 if (btrfs_test_opt(root, NODATACOW))
3587                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
3588         }
3589
3590         key[0].objectid = objectid;
3591         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3592         key[0].offset = 0;
3593
3594         key[1].objectid = objectid;
3595         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3596         key[1].offset = ref_objectid;
3597
3598         sizes[0] = sizeof(struct btrfs_inode_item);
3599         sizes[1] = name_len + sizeof(*ref);
3600
3601         path->leave_spinning = 1;
3602         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3603         if (ret != 0)
3604                 goto fail;
3605
3606         if (objectid > root->highest_inode)
3607                 root->highest_inode = objectid;
3608
3609         inode->i_uid = current_fsuid();
3610
3611         if (dir && (dir->i_mode & S_ISGID)) {
3612                 inode->i_gid = dir->i_gid;
3613                 if (S_ISDIR(mode))
3614                         mode |= S_ISGID;
3615         } else
3616                 inode->i_gid = current_fsgid();
3617
3618         inode->i_mode = mode;
3619         inode->i_ino = objectid;
3620         inode_set_bytes(inode, 0);
3621         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3622         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3623                                   struct btrfs_inode_item);
3624         fill_inode_item(trans, path->nodes[0], inode_item, inode);
3625
3626         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3627                              struct btrfs_inode_ref);
3628         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
3629         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
3630         ptr = (unsigned long)(ref + 1);
3631         write_extent_buffer(path->nodes[0], name, ptr, name_len);
3632
3633         btrfs_mark_buffer_dirty(path->nodes[0]);
3634         btrfs_free_path(path);
3635
3636         location = &BTRFS_I(inode)->location;
3637         location->objectid = objectid;
3638         location->offset = 0;
3639         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3640
3641         btrfs_inherit_iflags(inode, dir);
3642
3643         insert_inode_hash(inode);
3644         inode_tree_add(inode);
3645         return inode;
3646 fail:
3647         if (dir)
3648                 BTRFS_I(dir)->index_cnt--;
3649         btrfs_free_path(path);
3650         iput(inode);
3651         return ERR_PTR(ret);
3652 }
3653
3654 static inline u8 btrfs_inode_type(struct inode *inode)
3655 {
3656         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3657 }
3658
3659 /*
3660  * utility function to add 'inode' into 'parent_inode' with
3661  * a give name and a given sequence number.
3662  * if 'add_backref' is true, also insert a backref from the
3663  * inode to the parent directory.
3664  */
3665 int btrfs_add_link(struct btrfs_trans_handle *trans,
3666                    struct inode *parent_inode, struct inode *inode,
3667                    const char *name, int name_len, int add_backref, u64 index)
3668 {
3669         int ret;
3670         struct btrfs_key key;
3671         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
3672
3673         key.objectid = inode->i_ino;
3674         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3675         key.offset = 0;
3676
3677         ret = btrfs_insert_dir_item(trans, root, name, name_len,
3678                                     parent_inode->i_ino,
3679                                     &key, btrfs_inode_type(inode),
3680                                     index);
3681         if (ret == 0) {
3682                 if (add_backref) {
3683                         ret = btrfs_insert_inode_ref(trans, root,
3684                                                      name, name_len,
3685                                                      inode->i_ino,
3686                                                      parent_inode->i_ino,
3687                                                      index);
3688                 }
3689                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
3690                                    name_len * 2);
3691                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
3692                 ret = btrfs_update_inode(trans, root, parent_inode);
3693         }
3694         return ret;
3695 }
3696
3697 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
3698                             struct dentry *dentry, struct inode *inode,
3699                             int backref, u64 index)
3700 {
3701         int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3702                                  inode, dentry->d_name.name,
3703                                  dentry->d_name.len, backref, index);
3704         if (!err) {
3705                 d_instantiate(dentry, inode);
3706                 return 0;
3707         }
3708         if (err > 0)
3709                 err = -EEXIST;
3710         return err;
3711 }
3712
3713 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3714                         int mode, dev_t rdev)
3715 {
3716         struct btrfs_trans_handle *trans;
3717         struct btrfs_root *root = BTRFS_I(dir)->root;
3718         struct inode *inode = NULL;
3719         int err;
3720         int drop_inode = 0;
3721         u64 objectid;
3722         unsigned long nr = 0;
3723         u64 index = 0;
3724
3725         if (!new_valid_dev(rdev))
3726                 return -EINVAL;
3727
3728         err = btrfs_check_metadata_free_space(root);
3729         if (err)
3730                 goto fail;
3731
3732         trans = btrfs_start_transaction(root, 1);
3733         btrfs_set_trans_block_group(trans, dir);
3734
3735         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3736         if (err) {
3737                 err = -ENOSPC;
3738                 goto out_unlock;
3739         }
3740
3741         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3742                                 dentry->d_name.len,
3743                                 dentry->d_parent->d_inode->i_ino, objectid,
3744                                 BTRFS_I(dir)->block_group, mode, &index);
3745         err = PTR_ERR(inode);
3746         if (IS_ERR(inode))
3747                 goto out_unlock;
3748
3749         err = btrfs_init_inode_security(inode, dir);
3750         if (err) {
3751                 drop_inode = 1;
3752                 goto out_unlock;
3753         }
3754
3755         btrfs_set_trans_block_group(trans, inode);
3756         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3757         if (err)
3758                 drop_inode = 1;
3759         else {
3760                 inode->i_op = &btrfs_special_inode_operations;
3761                 init_special_inode(inode, inode->i_mode, rdev);
3762                 btrfs_update_inode(trans, root, inode);
3763         }
3764         btrfs_update_inode_block_group(trans, inode);
3765         btrfs_update_inode_block_group(trans, dir);
3766 out_unlock:
3767         nr = trans->blocks_used;
3768         btrfs_end_transaction_throttle(trans, root);
3769 fail:
3770         if (drop_inode) {
3771                 inode_dec_link_count(inode);
3772                 iput(inode);
3773         }
3774         btrfs_btree_balance_dirty(root, nr);
3775         return err;
3776 }
3777
3778 static int btrfs_create(struct inode *dir, struct dentry *dentry,
3779                         int mode, struct nameidata *nd)
3780 {
3781         struct btrfs_trans_handle *trans;
3782         struct btrfs_root *root = BTRFS_I(dir)->root;
3783         struct inode *inode = NULL;
3784         int err;
3785         int drop_inode = 0;
3786         unsigned long nr = 0;
3787         u64 objectid;
3788         u64 index = 0;
3789
3790         err = btrfs_check_metadata_free_space(root);
3791         if (err)
3792                 goto fail;
3793         trans = btrfs_start_transaction(root, 1);
3794         btrfs_set_trans_block_group(trans, dir);
3795
3796         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3797         if (err) {
3798                 err = -ENOSPC;
3799                 goto out_unlock;
3800         }
3801
3802         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3803                                 dentry->d_name.len,
3804                                 dentry->d_parent->d_inode->i_ino,
3805                                 objectid, BTRFS_I(dir)->block_group, mode,
3806                                 &index);
3807         err = PTR_ERR(inode);
3808         if (IS_ERR(inode))
3809                 goto out_unlock;
3810
3811         err = btrfs_init_inode_security(inode, dir);
3812         if (err) {
3813                 drop_inode = 1;
3814                 goto out_unlock;
3815         }
3816
3817         btrfs_set_trans_block_group(trans, inode);
3818         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3819         if (err)
3820                 drop_inode = 1;
3821         else {
3822                 inode->i_mapping->a_ops = &btrfs_aops;
3823                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3824                 inode->i_fop = &btrfs_file_operations;
3825                 inode->i_op = &btrfs_file_inode_operations;
3826                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3827         }
3828         btrfs_update_inode_block_group(trans, inode);
3829         btrfs_update_inode_block_group(trans, dir);
3830 out_unlock:
3831         nr = trans->blocks_used;
3832         btrfs_end_transaction_throttle(trans, root);
3833 fail:
3834         if (drop_inode) {
3835                 inode_dec_link_count(inode);
3836                 iput(inode);
3837         }
3838         btrfs_btree_balance_dirty(root, nr);
3839         return err;
3840 }
3841
3842 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3843                       struct dentry *dentry)
3844 {
3845         struct btrfs_trans_handle *trans;
3846         struct btrfs_root *root = BTRFS_I(dir)->root;
3847         struct inode *inode = old_dentry->d_inode;
3848         u64 index;
3849         unsigned long nr = 0;
3850         int err;
3851         int drop_inode = 0;
3852
3853         if (inode->i_nlink == 0)
3854                 return -ENOENT;
3855
3856         btrfs_inc_nlink(inode);
3857         err = btrfs_check_metadata_free_space(root);
3858         if (err)
3859                 goto fail;
3860         err = btrfs_set_inode_index(dir, &index);
3861         if (err)
3862                 goto fail;
3863
3864         trans = btrfs_start_transaction(root, 1);
3865
3866         btrfs_set_trans_block_group(trans, dir);
3867         atomic_inc(&inode->i_count);
3868
3869         err = btrfs_add_nondir(trans, dentry, inode, 1, index);
3870
3871         if (err)
3872                 drop_inode = 1;
3873
3874         btrfs_update_inode_block_group(trans, dir);
3875         err = btrfs_update_inode(trans, root, inode);
3876
3877         if (err)
3878                 drop_inode = 1;
3879
3880         nr = trans->blocks_used;
3881
3882         btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
3883         btrfs_end_transaction_throttle(trans, root);
3884 fail:
3885         if (drop_inode) {
3886                 inode_dec_link_count(inode);
3887                 iput(inode);
3888         }
3889         btrfs_btree_balance_dirty(root, nr);
3890         return err;
3891 }
3892
3893 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3894 {
3895         struct inode *inode = NULL;
3896         struct btrfs_trans_handle *trans;
3897         struct btrfs_root *root = BTRFS_I(dir)->root;
3898         int err = 0;
3899         int drop_on_err = 0;
3900         u64 objectid = 0;
3901         u64 index = 0;
3902         unsigned long nr = 1;
3903
3904         err = btrfs_check_metadata_free_space(root);
3905         if (err)
3906                 goto out_unlock;
3907
3908         trans = btrfs_start_transaction(root, 1);
3909         btrfs_set_trans_block_group(trans, dir);
3910
3911         if (IS_ERR(trans)) {
3912                 err = PTR_ERR(trans);
3913                 goto out_unlock;
3914         }
3915
3916         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3917         if (err) {
3918                 err = -ENOSPC;
3919                 goto out_unlock;
3920         }
3921
3922         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3923                                 dentry->d_name.len,
3924                                 dentry->d_parent->d_inode->i_ino, objectid,
3925                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3926                                 &index);
3927         if (IS_ERR(inode)) {
3928                 err = PTR_ERR(inode);
3929                 goto out_fail;
3930         }
3931
3932         drop_on_err = 1;
3933
3934         err = btrfs_init_inode_security(inode, dir);
3935         if (err)
3936                 goto out_fail;
3937
3938         inode->i_op = &btrfs_dir_inode_operations;
3939         inode->i_fop = &btrfs_dir_file_operations;
3940         btrfs_set_trans_block_group(trans, inode);
3941
3942         btrfs_i_size_write(inode, 0);
3943         err = btrfs_update_inode(trans, root, inode);
3944         if (err)
3945                 goto out_fail;
3946
3947         err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3948                                  inode, dentry->d_name.name,
3949                                  dentry->d_name.len, 0, index);
3950         if (err)
3951                 goto out_fail;
3952
3953         d_instantiate(dentry, inode);
3954         drop_on_err = 0;
3955         btrfs_update_inode_block_group(trans, inode);
3956         btrfs_update_inode_block_group(trans, dir);
3957
3958 out_fail:
3959         nr = trans->blocks_used;
3960         btrfs_end_transaction_throttle(trans, root);
3961
3962 out_unlock:
3963         if (drop_on_err)
3964                 iput(inode);
3965         btrfs_btree_balance_dirty(root, nr);
3966         return err;
3967 }
3968
3969 /* helper for btfs_get_extent.  Given an existing extent in the tree,
3970  * and an extent that you want to insert, deal with overlap and insert
3971  * the new extent into the tree.
3972  */
3973 static int merge_extent_mapping(struct extent_map_tree *em_tree,
3974                                 struct extent_map *existing,
3975                                 struct extent_map *em,
3976                                 u64 map_start, u64 map_len)
3977 {
3978         u64 start_diff;
3979
3980         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3981         start_diff = map_start - em->start;
3982         em->start = map_start;
3983         em->len = map_len;
3984         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3985             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3986                 em->block_start += start_diff;
3987                 em->block_len -= start_diff;
3988         }
3989         return add_extent_mapping(em_tree, em);
3990 }
3991
3992 static noinline int uncompress_inline(struct btrfs_path *path,
3993                                       struct inode *inode, struct page *page,
3994                                       size_t pg_offset, u64 extent_offset,
3995                                       struct btrfs_file_extent_item *item)
3996 {
3997         int ret;
3998         struct extent_buffer *leaf = path->nodes[0];
3999         char *tmp;
4000         size_t max_size;
4001         unsigned long inline_size;
4002         unsigned long ptr;
4003
4004         WARN_ON(pg_offset != 0);
4005         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4006         inline_size = btrfs_file_extent_inline_item_len(leaf,
4007                                         btrfs_item_nr(leaf, path->slots[0]));
4008         tmp = kmalloc(inline_size, GFP_NOFS);
4009         ptr = btrfs_file_extent_inline_start(item);
4010
4011         read_extent_buffer(leaf, tmp, ptr, inline_size);
4012
4013         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4014         ret = btrfs_zlib_decompress(tmp, page, extent_offset,
4015                                     inline_size, max_size);
4016         if (ret) {
4017                 char *kaddr = kmap_atomic(page, KM_USER0);
4018                 unsigned long copy_size = min_t(u64,
4019                                   PAGE_CACHE_SIZE - pg_offset,
4020                                   max_size - extent_offset);
4021                 memset(kaddr + pg_offset, 0, copy_size);
4022                 kunmap_atomic(kaddr, KM_USER0);
4023         }
4024         kfree(tmp);
4025         return 0;
4026 }
4027
4028 /*
4029  * a bit scary, this does extent mapping from logical file offset to the disk.
4030  * the ugly parts come from merging extents from the disk with the in-ram
4031  * representation.  This gets more complex because of the data=ordered code,
4032  * where the in-ram extents might be locked pending data=ordered completion.
4033  *
4034  * This also copies inline extents directly into the page.
4035  */
4036
4037 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4038                                     size_t pg_offset, u64 start, u64 len,
4039                                     int create)
4040 {
4041         int ret;
4042         int err = 0;
4043         u64 bytenr;
4044         u64 extent_start = 0;
4045         u64 extent_end = 0;
4046         u64 objectid = inode->i_ino;
4047         u32 found_type;
4048         struct btrfs_path *path = NULL;
4049         struct btrfs_root *root = BTRFS_I(inode)->root;
4050         struct btrfs_file_extent_item *item;
4051         struct extent_buffer *leaf;
4052         struct btrfs_key found_key;
4053         struct extent_map *em = NULL;
4054         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4055         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4056         struct btrfs_trans_handle *trans = NULL;
4057         int compressed;
4058
4059 again:
4060         spin_lock(&em_tree->lock);
4061         em = lookup_extent_mapping(em_tree, start, len);
4062         if (em)
4063                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4064         spin_unlock(&em_tree->lock);
4065
4066         if (em) {
4067                 if (em->start > start || em->start + em->len <= start)
4068                         free_extent_map(em);
4069                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4070                         free_extent_map(em);
4071                 else
4072                         goto out;
4073         }
4074         em = alloc_extent_map(GFP_NOFS);
4075         if (!em) {
4076                 err = -ENOMEM;
4077                 goto out;
4078         }
4079         em->bdev = root->fs_info->fs_devices->latest_bdev;
4080         em->start = EXTENT_MAP_HOLE;
4081         em->orig_start = EXTENT_MAP_HOLE;
4082         em->len = (u64)-1;
4083         em->block_len = (u64)-1;
4084
4085         if (!path) {
4086                 path = btrfs_alloc_path();
4087                 BUG_ON(!path);
4088         }
4089
4090         ret = btrfs_lookup_file_extent(trans, root, path,
4091                                        objectid, start, trans != NULL);
4092         if (ret < 0) {
4093                 err = ret;
4094                 goto out;
4095         }
4096
4097         if (ret != 0) {
4098                 if (path->slots[0] == 0)
4099                         goto not_found;
4100                 path->slots[0]--;
4101         }
4102
4103         leaf = path->nodes[0];
4104         item = btrfs_item_ptr(leaf, path->slots[0],
4105                               struct btrfs_file_extent_item);
4106         /* are we inside the extent that was found? */
4107         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4108         found_type = btrfs_key_type(&found_key);
4109         if (found_key.objectid != objectid ||
4110             found_type != BTRFS_EXTENT_DATA_KEY) {
4111                 goto not_found;
4112         }
4113
4114         found_type = btrfs_file_extent_type(leaf, item);
4115         extent_start = found_key.offset;
4116         compressed = btrfs_file_extent_compression(leaf, item);
4117         if (found_type == BTRFS_FILE_EXTENT_REG ||
4118             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4119                 extent_end = extent_start +
4120                        btrfs_file_extent_num_bytes(leaf, item);
4121         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4122                 size_t size;
4123                 size = btrfs_file_extent_inline_len(leaf, item);
4124                 extent_end = (extent_start + size + root->sectorsize - 1) &
4125                         ~((u64)root->sectorsize - 1);
4126         }
4127
4128         if (start >= extent_end) {
4129                 path->slots[0]++;
4130                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4131                         ret = btrfs_next_leaf(root, path);
4132                         if (ret < 0) {
4133                                 err = ret;
4134                                 goto out;
4135                         }
4136                         if (ret > 0)
4137                                 goto not_found;
4138                         leaf = path->nodes[0];
4139                 }
4140                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4141                 if (found_key.objectid != objectid ||
4142                     found_key.type != BTRFS_EXTENT_DATA_KEY)
4143                         goto not_found;
4144                 if (start + len <= found_key.offset)
4145                         goto not_found;
4146                 em->start = start;
4147                 em->len = found_key.offset - start;
4148                 goto not_found_em;
4149         }
4150
4151         if (found_type == BTRFS_FILE_EXTENT_REG ||
4152             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4153                 em->start = extent_start;
4154                 em->len = extent_end - extent_start;
4155                 em->orig_start = extent_start -
4156                                  btrfs_file_extent_offset(leaf, item);
4157                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4158                 if (bytenr == 0) {
4159                         em->block_start = EXTENT_MAP_HOLE;
4160                         goto insert;
4161                 }
4162                 if (compressed) {
4163                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4164                         em->block_start = bytenr;
4165                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4166                                                                          item);
4167                 } else {
4168                         bytenr += btrfs_file_extent_offset(leaf, item);
4169                         em->block_start = bytenr;
4170                         em->block_len = em->len;
4171                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4172                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4173                 }
4174                 goto insert;
4175         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4176                 unsigned long ptr;
4177                 char *map;
4178                 size_t size;
4179                 size_t extent_offset;
4180                 size_t copy_size;
4181
4182                 em->block_start = EXTENT_MAP_INLINE;
4183                 if (!page || create) {
4184                         em->start = extent_start;
4185                         em->len = extent_end - extent_start;
4186                         goto out;
4187                 }
4188
4189                 size = btrfs_file_extent_inline_len(leaf, item);
4190                 extent_offset = page_offset(page) + pg_offset - extent_start;
4191                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4192                                 size - extent_offset);
4193                 em->start = extent_start + extent_offset;
4194                 em->len = (copy_size + root->sectorsize - 1) &
4195                         ~((u64)root->sectorsize - 1);
4196                 em->orig_start = EXTENT_MAP_INLINE;
4197                 if (compressed)
4198                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4199                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4200                 if (create == 0 && !PageUptodate(page)) {
4201                         if (btrfs_file_extent_compression(leaf, item) ==
4202                             BTRFS_COMPRESS_ZLIB) {
4203                                 ret = uncompress_inline(path, inode, page,
4204                                                         pg_offset,
4205                                                         extent_offset, item);
4206                                 BUG_ON(ret);
4207                         } else {
4208                                 map = kmap(page);
4209                                 read_extent_buffer(leaf, map + pg_offset, ptr,
4210                                                    copy_size);
4211                                 kunmap(page);
4212                         }
4213                         flush_dcache_page(page);
4214                 } else if (create && PageUptodate(page)) {
4215                         if (!trans) {
4216                                 kunmap(page);
4217                                 free_extent_map(em);
4218                                 em = NULL;
4219                                 btrfs_release_path(root, path);
4220                                 trans = btrfs_join_transaction(root, 1);
4221                                 goto again;
4222                         }
4223                         map = kmap(page);
4224                         write_extent_buffer(leaf, map + pg_offset, ptr,
4225                                             copy_size);
4226                         kunmap(page);
4227                         btrfs_mark_buffer_dirty(leaf);
4228                 }
4229                 set_extent_uptodate(io_tree, em->start,
4230                                     extent_map_end(em) - 1, GFP_NOFS);
4231                 goto insert;
4232         } else {
4233                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
4234                 WARN_ON(1);
4235         }
4236 not_found:
4237         em->start = start;
4238         em->len = len;
4239 not_found_em:
4240         em->block_start = EXTENT_MAP_HOLE;
4241         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4242 insert:
4243         btrfs_release_path(root, path);
4244         if (em->start > start || extent_map_end(em) <= start) {
4245                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4246                        "[%llu %llu]\n", (unsigned long long)em->start,
4247                        (unsigned long long)em->len,
4248                        (unsigned long long)start,
4249                        (unsigned long long)len);
4250                 err = -EIO;
4251                 goto out;
4252         }
4253
4254         err = 0;
4255         spin_lock(&em_tree->lock);
4256         ret = add_extent_mapping(em_tree, em);
4257         /* it is possible that someone inserted the extent into the tree
4258          * while we had the lock dropped.  It is also possible that
4259          * an overlapping map exists in the tree
4260          */
4261         if (ret == -EEXIST) {
4262                 struct extent_map *existing;
4263
4264                 ret = 0;
4265
4266                 existing = lookup_extent_mapping(em_tree, start, len);
4267                 if (existing && (existing->start > start ||
4268                     existing->start + existing->len <= start)) {
4269                         free_extent_map(existing);
4270                         existing = NULL;
4271                 }
4272                 if (!existing) {
4273                         existing = lookup_extent_mapping(em_tree, em->start,
4274                                                          em->len);
4275                         if (existing) {
4276                                 err = merge_extent_mapping(em_tree, existing,
4277                                                            em, start,
4278                                                            root->sectorsize);
4279                                 free_extent_map(existing);
4280                                 if (err) {
4281                                         free_extent_map(em);
4282                                         em = NULL;
4283                                 }
4284                         } else {
4285                                 err = -EIO;
4286                                 free_extent_map(em);
4287                                 em = NULL;
4288                         }
4289                 } else {
4290                         free_extent_map(em);
4291                         em = existing;
4292                         err = 0;
4293                 }
4294         }
4295         spin_unlock(&em_tree->lock);
4296 out:
4297         if (path)
4298                 btrfs_free_path(path);
4299         if (trans) {
4300                 ret = btrfs_end_transaction(trans, root);
4301                 if (!err)
4302                         err = ret;
4303         }
4304         if (err) {
4305                 free_extent_map(em);
4306                 return ERR_PTR(err);
4307         }
4308         return em;
4309 }
4310
4311 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4312                         const struct iovec *iov, loff_t offset,
4313                         unsigned long nr_segs)
4314 {
4315         return -EINVAL;
4316 }
4317
4318 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4319                 __u64 start, __u64 len)
4320 {
4321         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
4322 }
4323
4324 int btrfs_readpage(struct file *file, struct page *page)
4325 {
4326         struct extent_io_tree *tree;
4327         tree = &BTRFS_I(page->mapping->host)->io_tree;
4328         return extent_read_full_page(tree, page, btrfs_get_extent);
4329 }
4330
4331 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4332 {
4333         struct extent_io_tree *tree;
4334
4335
4336         if (current->flags & PF_MEMALLOC) {
4337                 redirty_page_for_writepage(wbc, page);
4338                 unlock_page(page);
4339                 return 0;
4340         }
4341         tree = &BTRFS_I(page->mapping->host)->io_tree;
4342         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4343 }
4344
4345 int btrfs_writepages(struct address_space *mapping,
4346                      struct writeback_control *wbc)
4347 {
4348         struct extent_io_tree *tree;
4349
4350         tree = &BTRFS_I(mapping->host)->io_tree;
4351         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4352 }
4353
4354 static int
4355 btrfs_readpages(struct file *file, struct address_space *mapping,
4356                 struct list_head *pages, unsigned nr_pages)
4357 {
4358         struct extent_io_tree *tree;
4359         tree = &BTRFS_I(mapping->host)->io_tree;
4360         return extent_readpages(tree, mapping, pages, nr_pages,
4361                                 btrfs_get_extent);
4362 }
4363 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4364 {
4365         struct extent_io_tree *tree;
4366         struct extent_map_tree *map;
4367         int ret;
4368
4369         tree = &BTRFS_I(page->mapping->host)->io_tree;
4370         map = &BTRFS_I(page->mapping->host)->extent_tree;
4371         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4372         if (ret == 1) {
4373                 ClearPagePrivate(page);
4374                 set_page_private(page, 0);
4375                 page_cache_release(page);
4376         }
4377         return ret;
4378 }
4379
4380 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4381 {
4382         if (PageWriteback(page) || PageDirty(page))
4383                 return 0;
4384         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
4385 }
4386
4387 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4388 {
4389         struct extent_io_tree *tree;
4390         struct btrfs_ordered_extent *ordered;
4391         u64 page_start = page_offset(page);
4392         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4393
4394         wait_on_page_writeback(page);
4395         tree = &BTRFS_I(page->mapping->host)->io_tree;
4396         if (offset) {
4397                 btrfs_releasepage(page, GFP_NOFS);
4398                 return;
4399         }
4400
4401         lock_extent(tree, page_start, page_end, GFP_NOFS);
4402         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4403                                            page_offset(page));
4404         if (ordered) {
4405                 /*
4406                  * IO on this page will never be started, so we need
4407                  * to account for any ordered extents now
4408                  */
4409                 clear_extent_bit(tree, page_start, page_end,
4410                                  EXTENT_DIRTY | EXTENT_DELALLOC |
4411                                  EXTENT_LOCKED, 1, 0, GFP_NOFS);
4412                 btrfs_finish_ordered_io(page->mapping->host,
4413                                         page_start, page_end);
4414                 btrfs_put_ordered_extent(ordered);
4415                 lock_extent(tree, page_start, page_end, GFP_NOFS);
4416         }
4417         clear_extent_bit(tree, page_start, page_end,
4418                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4419                  EXTENT_ORDERED,
4420                  1, 1, GFP_NOFS);
4421         __btrfs_releasepage(page, GFP_NOFS);
4422
4423         ClearPageChecked(page);
4424         if (PagePrivate(page)) {
4425                 ClearPagePrivate(page);
4426                 set_page_private(page, 0);
4427                 page_cache_release(page);
4428         }
4429 }
4430
4431 /*
4432  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4433  * called from a page fault handler when a page is first dirtied. Hence we must
4434  * be careful to check for EOF conditions here. We set the page up correctly
4435  * for a written page which means we get ENOSPC checking when writing into
4436  * holes and correct delalloc and unwritten extent mapping on filesystems that
4437  * support these features.
4438  *
4439  * We are not allowed to take the i_mutex here so we have to play games to
4440  * protect against truncate races as the page could now be beyond EOF.  Because
4441  * vmtruncate() writes the inode size before removing pages, once we have the
4442  * page lock we can determine safely if the page is beyond EOF. If it is not
4443  * beyond EOF, then the page is guaranteed safe against truncation until we
4444  * unlock the page.
4445  */
4446 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
4447 {
4448         struct page *page = vmf->page;
4449         struct inode *inode = fdentry(vma->vm_file)->d_inode;
4450         struct btrfs_root *root = BTRFS_I(inode)->root;
4451         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4452         struct btrfs_ordered_extent *ordered;
4453         char *kaddr;
4454         unsigned long zero_start;
4455         loff_t size;
4456         int ret;
4457         u64 page_start;
4458         u64 page_end;
4459
4460         ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
4461         if (ret) {
4462                 if (ret == -ENOMEM)
4463                         ret = VM_FAULT_OOM;
4464                 else /* -ENOSPC, -EIO, etc */
4465                         ret = VM_FAULT_SIGBUS;
4466                 goto out;
4467         }
4468
4469         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
4470 again:
4471         lock_page(page);
4472         size = i_size_read(inode);
4473         page_start = page_offset(page);
4474         page_end = page_start + PAGE_CACHE_SIZE - 1;
4475
4476         if ((page->mapping != inode->i_mapping) ||
4477             (page_start >= size)) {
4478                 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
4479                 /* page got truncated out from underneath us */
4480                 goto out_unlock;
4481         }
4482         wait_on_page_writeback(page);
4483
4484         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4485         set_page_extent_mapped(page);
4486
4487         /*
4488          * we can't set the delalloc bits if there are pending ordered
4489          * extents.  Drop our locks and wait for them to finish
4490          */
4491         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4492         if (ordered) {
4493                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4494                 unlock_page(page);
4495                 btrfs_start_ordered_extent(inode, ordered, 1);
4496                 btrfs_put_ordered_extent(ordered);
4497                 goto again;
4498         }
4499
4500         btrfs_set_extent_delalloc(inode, page_start, page_end);
4501         ret = 0;
4502
4503         /* page is wholly or partially inside EOF */
4504         if (page_start + PAGE_CACHE_SIZE > size)
4505                 zero_start = size & ~PAGE_CACHE_MASK;
4506         else
4507                 zero_start = PAGE_CACHE_SIZE;
4508
4509         if (zero_start != PAGE_CACHE_SIZE) {
4510                 kaddr = kmap(page);
4511                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4512                 flush_dcache_page(page);
4513                 kunmap(page);
4514         }
4515         ClearPageChecked(page);
4516         set_page_dirty(page);
4517
4518         BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
4519         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4520
4521 out_unlock:
4522         unlock_page(page);
4523 out:
4524         return ret;
4525 }
4526
4527 static void btrfs_truncate(struct inode *inode)
4528 {
4529         struct btrfs_root *root = BTRFS_I(inode)->root;
4530         int ret;
4531         struct btrfs_trans_handle *trans;
4532         unsigned long nr;
4533         u64 mask = root->sectorsize - 1;
4534
4535         if (!S_ISREG(inode->i_mode))
4536                 return;
4537         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4538                 return;
4539
4540         btrfs_truncate_page(inode->i_mapping, inode->i_size);
4541         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
4542
4543         trans = btrfs_start_transaction(root, 1);
4544
4545         /*
4546          * setattr is responsible for setting the ordered_data_close flag,
4547          * but that is only tested during the last file release.  That
4548          * could happen well after the next commit, leaving a great big
4549          * window where new writes may get lost if someone chooses to write
4550          * to this file after truncating to zero
4551          *
4552          * The inode doesn't have any dirty data here, and so if we commit
4553          * this is a noop.  If someone immediately starts writing to the inode
4554          * it is very likely we'll catch some of their writes in this
4555          * transaction, and the commit will find this file on the ordered
4556          * data list with good things to send down.
4557          *
4558          * This is a best effort solution, there is still a window where
4559          * using truncate to replace the contents of the file will
4560          * end up with a zero length file after a crash.
4561          */
4562         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
4563                 btrfs_add_ordered_operation(trans, root, inode);
4564
4565         btrfs_set_trans_block_group(trans, inode);
4566         btrfs_i_size_write(inode, inode->i_size);
4567
4568         ret = btrfs_orphan_add(trans, inode);
4569         if (ret)
4570                 goto out;
4571         /* FIXME, add redo link to tree so we don't leak on crash */
4572         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
4573                                       BTRFS_EXTENT_DATA_KEY);
4574         btrfs_update_inode(trans, root, inode);
4575
4576         ret = btrfs_orphan_del(trans, inode);
4577         BUG_ON(ret);
4578
4579 out:
4580         nr = trans->blocks_used;
4581         ret = btrfs_end_transaction_throttle(trans, root);
4582         BUG_ON(ret);
4583         btrfs_btree_balance_dirty(root, nr);
4584 }
4585
4586 /*
4587  * create a new subvolume directory/inode (helper for the ioctl).
4588  */
4589 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
4590                              struct btrfs_root *new_root, struct dentry *dentry,
4591                              u64 new_dirid, u64 alloc_hint)
4592 {
4593         struct inode *inode;
4594         int error;
4595         u64 index = 0;
4596
4597         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
4598                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
4599         if (IS_ERR(inode))
4600                 return PTR_ERR(inode);
4601         inode->i_op = &btrfs_dir_inode_operations;
4602         inode->i_fop = &btrfs_dir_file_operations;
4603
4604         inode->i_nlink = 1;
4605         btrfs_i_size_write(inode, 0);
4606
4607         error = btrfs_update_inode(trans, new_root, inode);
4608         if (error)
4609                 return error;
4610
4611         d_instantiate(dentry, inode);
4612         return 0;
4613 }
4614
4615 /* helper function for file defrag and space balancing.  This
4616  * forces readahead on a given range of bytes in an inode
4617  */
4618 unsigned long btrfs_force_ra(struct address_space *mapping,
4619                               struct file_ra_state *ra, struct file *file,
4620                               pgoff_t offset, pgoff_t last_index)
4621 {
4622         pgoff_t req_size = last_index - offset + 1;
4623
4624         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4625         return offset + req_size;
4626 }
4627
4628 struct inode *btrfs_alloc_inode(struct super_block *sb)
4629 {
4630         struct btrfs_inode *ei;
4631
4632         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4633         if (!ei)
4634                 return NULL;
4635         ei->last_trans = 0;
4636         ei->logged_trans = 0;
4637         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
4638         INIT_LIST_HEAD(&ei->i_orphan);
4639         INIT_LIST_HEAD(&ei->ordered_operations);
4640         return &ei->vfs_inode;
4641 }
4642
4643 void btrfs_destroy_inode(struct inode *inode)
4644 {
4645         struct btrfs_ordered_extent *ordered;
4646         struct btrfs_root *root = BTRFS_I(inode)->root;
4647
4648         WARN_ON(!list_empty(&inode->i_dentry));
4649         WARN_ON(inode->i_data.nrpages);
4650
4651         /*
4652          * Make sure we're properly removed from the ordered operation
4653          * lists.
4654          */
4655         smp_mb();
4656         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
4657                 spin_lock(&root->fs_info->ordered_extent_lock);
4658                 list_del_init(&BTRFS_I(inode)->ordered_operations);
4659                 spin_unlock(&root->fs_info->ordered_extent_lock);
4660         }
4661
4662         spin_lock(&root->list_lock);
4663         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4664                 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4665                        " list\n", inode->i_ino);
4666                 dump_stack();
4667         }
4668         spin_unlock(&root->list_lock);
4669
4670         while (1) {
4671                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4672                 if (!ordered)
4673                         break;
4674                 else {
4675                         printk(KERN_ERR "btrfs found ordered "
4676                                "extent %llu %llu on inode cleanup\n",
4677                                (unsigned long long)ordered->file_offset,
4678                                (unsigned long long)ordered->len);
4679                         btrfs_remove_ordered_extent(inode, ordered);
4680                         btrfs_put_ordered_extent(ordered);
4681                         btrfs_put_ordered_extent(ordered);
4682                 }
4683         }
4684         inode_tree_del(inode);
4685         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
4686         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4687 }
4688
4689 static void init_once(void *foo)
4690 {
4691         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4692
4693         inode_init_once(&ei->vfs_inode);
4694 }
4695
4696 void btrfs_destroy_cachep(void)
4697 {
4698         if (btrfs_inode_cachep)
4699                 kmem_cache_destroy(btrfs_inode_cachep);
4700         if (btrfs_trans_handle_cachep)
4701                 kmem_cache_destroy(btrfs_trans_handle_cachep);
4702         if (btrfs_transaction_cachep)
4703                 kmem_cache_destroy(btrfs_transaction_cachep);
4704         if (btrfs_path_cachep)
4705                 kmem_cache_destroy(btrfs_path_cachep);
4706 }
4707
4708 int btrfs_init_cachep(void)
4709 {
4710         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
4711                         sizeof(struct btrfs_inode), 0,
4712                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
4713         if (!btrfs_inode_cachep)
4714                 goto fail;
4715
4716         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
4717                         sizeof(struct btrfs_trans_handle), 0,
4718                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
4719         if (!btrfs_trans_handle_cachep)
4720                 goto fail;
4721
4722         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
4723                         sizeof(struct btrfs_transaction), 0,
4724                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
4725         if (!btrfs_transaction_cachep)
4726                 goto fail;
4727
4728         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
4729                         sizeof(struct btrfs_path), 0,
4730                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
4731         if (!btrfs_path_cachep)
4732                 goto fail;
4733
4734         return 0;
4735 fail:
4736         btrfs_destroy_cachep();
4737         return -ENOMEM;
4738 }
4739
4740 static int btrfs_getattr(struct vfsmount *mnt,
4741                          struct dentry *dentry, struct kstat *stat)
4742 {
4743         struct inode *inode = dentry->d_inode;
4744         generic_fillattr(inode, stat);
4745         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
4746         stat->blksize = PAGE_CACHE_SIZE;
4747         stat->blocks = (inode_get_bytes(inode) +
4748                         BTRFS_I(inode)->delalloc_bytes) >> 9;
4749         return 0;
4750 }
4751
4752 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4753                            struct inode *new_dir, struct dentry *new_dentry)
4754 {
4755         struct btrfs_trans_handle *trans;
4756         struct btrfs_root *root = BTRFS_I(old_dir)->root;
4757         struct inode *new_inode = new_dentry->d_inode;
4758         struct inode *old_inode = old_dentry->d_inode;
4759         struct timespec ctime = CURRENT_TIME;
4760         u64 index = 0;
4761         int ret;
4762
4763         /* we're not allowed to rename between subvolumes */
4764         if (BTRFS_I(old_inode)->root->root_key.objectid !=
4765             BTRFS_I(new_dir)->root->root_key.objectid)
4766                 return -EXDEV;
4767
4768         if (S_ISDIR(old_inode->i_mode) && new_inode &&
4769             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4770                 return -ENOTEMPTY;
4771         }
4772
4773         /* to rename a snapshot or subvolume, we need to juggle the
4774          * backrefs.  This isn't coded yet
4775          */
4776         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
4777                 return -EXDEV;
4778
4779         ret = btrfs_check_metadata_free_space(root);
4780         if (ret)
4781                 goto out_unlock;
4782
4783         /*
4784          * we're using rename to replace one file with another.
4785          * and the replacement file is large.  Start IO on it now so
4786          * we don't add too much work to the end of the transaction
4787          */
4788         if (new_inode && old_inode && S_ISREG(old_inode->i_mode) &&
4789             new_inode->i_size &&
4790             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
4791                 filemap_flush(old_inode->i_mapping);
4792
4793         trans = btrfs_start_transaction(root, 1);
4794
4795         /*
4796          * make sure the inode gets flushed if it is replacing
4797          * something.
4798          */
4799         if (new_inode && new_inode->i_size &&
4800             old_inode && S_ISREG(old_inode->i_mode)) {
4801                 btrfs_add_ordered_operation(trans, root, old_inode);
4802         }
4803
4804         /*
4805          * this is an ugly little race, but the rename is required to make
4806          * sure that if we crash, the inode is either at the old name
4807          * or the new one.  pinning the log transaction lets us make sure
4808          * we don't allow a log commit to come in after we unlink the
4809          * name but before we add the new name back in.
4810          */
4811         btrfs_pin_log_trans(root);
4812
4813         btrfs_set_trans_block_group(trans, new_dir);
4814
4815         btrfs_inc_nlink(old_dentry->d_inode);
4816         old_dir->i_ctime = old_dir->i_mtime = ctime;
4817         new_dir->i_ctime = new_dir->i_mtime = ctime;
4818         old_inode->i_ctime = ctime;
4819
4820         if (old_dentry->d_parent != new_dentry->d_parent)
4821                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
4822
4823         ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4824                                  old_dentry->d_name.name,
4825                                  old_dentry->d_name.len);
4826         if (ret)
4827                 goto out_fail;
4828
4829         if (new_inode) {
4830                 new_inode->i_ctime = CURRENT_TIME;
4831                 ret = btrfs_unlink_inode(trans, root, new_dir,
4832                                          new_dentry->d_inode,
4833                                          new_dentry->d_name.name,
4834                                          new_dentry->d_name.len);
4835                 if (ret)
4836                         goto out_fail;
4837                 if (new_inode->i_nlink == 0) {
4838                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4839                         if (ret)
4840                                 goto out_fail;
4841                 }
4842
4843         }
4844         ret = btrfs_set_inode_index(new_dir, &index);
4845         if (ret)
4846                 goto out_fail;
4847
4848         ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4849                              old_inode, new_dentry->d_name.name,
4850                              new_dentry->d_name.len, 1, index);
4851         if (ret)
4852                 goto out_fail;
4853
4854         btrfs_log_new_name(trans, old_inode, old_dir,
4855                                        new_dentry->d_parent);
4856 out_fail:
4857
4858         /* this btrfs_end_log_trans just allows the current
4859          * log-sub transaction to complete
4860          */
4861         btrfs_end_log_trans(root);
4862         btrfs_end_transaction_throttle(trans, root);
4863 out_unlock:
4864         return ret;
4865 }
4866
4867 /*
4868  * some fairly slow code that needs optimization. This walks the list
4869  * of all the inodes with pending delalloc and forces them to disk.
4870  */
4871 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4872 {
4873         struct list_head *head = &root->fs_info->delalloc_inodes;
4874         struct btrfs_inode *binode;
4875         struct inode *inode;
4876
4877         if (root->fs_info->sb->s_flags & MS_RDONLY)
4878                 return -EROFS;
4879
4880         spin_lock(&root->fs_info->delalloc_lock);
4881         while (!list_empty(head)) {
4882                 binode = list_entry(head->next, struct btrfs_inode,
4883                                     delalloc_inodes);
4884                 inode = igrab(&binode->vfs_inode);
4885                 if (!inode)
4886                         list_del_init(&binode->delalloc_inodes);
4887                 spin_unlock(&root->fs_info->delalloc_lock);
4888                 if (inode) {
4889                         filemap_flush(inode->i_mapping);
4890                         iput(inode);
4891                 }
4892                 cond_resched();
4893                 spin_lock(&root->fs_info->delalloc_lock);
4894         }
4895         spin_unlock(&root->fs_info->delalloc_lock);
4896
4897         /* the filemap_flush will queue IO into the worker threads, but
4898          * we have to make sure the IO is actually started and that
4899          * ordered extents get created before we return
4900          */
4901         atomic_inc(&root->fs_info->async_submit_draining);
4902         while (atomic_read(&root->fs_info->nr_async_submits) ||
4903               atomic_read(&root->fs_info->async_delalloc_pages)) {
4904                 wait_event(root->fs_info->async_submit_wait,
4905                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4906                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
4907         }
4908         atomic_dec(&root->fs_info->async_submit_draining);
4909         return 0;
4910 }
4911
4912 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4913                          const char *symname)
4914 {
4915         struct btrfs_trans_handle *trans;
4916         struct btrfs_root *root = BTRFS_I(dir)->root;
4917         struct btrfs_path *path;
4918         struct btrfs_key key;
4919         struct inode *inode = NULL;
4920         int err;
4921         int drop_inode = 0;
4922         u64 objectid;
4923         u64 index = 0 ;
4924         int name_len;
4925         int datasize;
4926         unsigned long ptr;
4927         struct btrfs_file_extent_item *ei;
4928         struct extent_buffer *leaf;
4929         unsigned long nr = 0;
4930
4931         name_len = strlen(symname) + 1;
4932         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4933                 return -ENAMETOOLONG;
4934
4935         err = btrfs_check_metadata_free_space(root);
4936         if (err)
4937                 goto out_fail;
4938
4939         trans = btrfs_start_transaction(root, 1);
4940         btrfs_set_trans_block_group(trans, dir);
4941
4942         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4943         if (err) {
4944                 err = -ENOSPC;
4945                 goto out_unlock;
4946         }
4947
4948         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4949                                 dentry->d_name.len,
4950                                 dentry->d_parent->d_inode->i_ino, objectid,
4951                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4952                                 &index);
4953         err = PTR_ERR(inode);
4954         if (IS_ERR(inode))
4955                 goto out_unlock;
4956
4957         err = btrfs_init_inode_security(inode, dir);
4958         if (err) {
4959                 drop_inode = 1;
4960                 goto out_unlock;
4961         }
4962
4963         btrfs_set_trans_block_group(trans, inode);
4964         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4965         if (err)
4966                 drop_inode = 1;
4967         else {
4968                 inode->i_mapping->a_ops = &btrfs_aops;
4969                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4970                 inode->i_fop = &btrfs_file_operations;
4971                 inode->i_op = &btrfs_file_inode_operations;
4972                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4973         }
4974         btrfs_update_inode_block_group(trans, inode);
4975         btrfs_update_inode_block_group(trans, dir);
4976         if (drop_inode)
4977                 goto out_unlock;
4978
4979         path = btrfs_alloc_path();
4980         BUG_ON(!path);
4981         key.objectid = inode->i_ino;
4982         key.offset = 0;
4983         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4984         datasize = btrfs_file_extent_calc_inline_size(name_len);
4985         err = btrfs_insert_empty_item(trans, root, path, &key,
4986                                       datasize);
4987         if (err) {
4988                 drop_inode = 1;
4989                 goto out_unlock;
4990         }
4991         leaf = path->nodes[0];
4992         ei = btrfs_item_ptr(leaf, path->slots[0],
4993                             struct btrfs_file_extent_item);
4994         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4995         btrfs_set_file_extent_type(leaf, ei,
4996                                    BTRFS_FILE_EXTENT_INLINE);
4997         btrfs_set_file_extent_encryption(leaf, ei, 0);
4998         btrfs_set_file_extent_compression(leaf, ei, 0);
4999         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
5000         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
5001
5002         ptr = btrfs_file_extent_inline_start(ei);
5003         write_extent_buffer(leaf, symname, ptr, name_len);
5004         btrfs_mark_buffer_dirty(leaf);
5005         btrfs_free_path(path);
5006
5007         inode->i_op = &btrfs_symlink_inode_operations;
5008         inode->i_mapping->a_ops = &btrfs_symlink_aops;
5009         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5010         inode_set_bytes(inode, name_len);
5011         btrfs_i_size_write(inode, name_len - 1);
5012         err = btrfs_update_inode(trans, root, inode);
5013         if (err)
5014                 drop_inode = 1;
5015
5016 out_unlock:
5017         nr = trans->blocks_used;
5018         btrfs_end_transaction_throttle(trans, root);
5019 out_fail:
5020         if (drop_inode) {
5021                 inode_dec_link_count(inode);
5022                 iput(inode);
5023         }
5024         btrfs_btree_balance_dirty(root, nr);
5025         return err;
5026 }
5027
5028 static int prealloc_file_range(struct btrfs_trans_handle *trans,
5029                                struct inode *inode, u64 start, u64 end,
5030                                u64 locked_end, u64 alloc_hint, int mode)
5031 {
5032         struct btrfs_root *root = BTRFS_I(inode)->root;
5033         struct btrfs_key ins;
5034         u64 alloc_size;
5035         u64 cur_offset = start;
5036         u64 num_bytes = end - start;
5037         int ret = 0;
5038
5039         while (num_bytes > 0) {
5040                 alloc_size = min(num_bytes, root->fs_info->max_extent);
5041                 ret = btrfs_reserve_extent(trans, root, alloc_size,
5042                                            root->sectorsize, 0, alloc_hint,
5043                                            (u64)-1, &ins, 1);
5044                 if (ret) {
5045                         WARN_ON(1);
5046                         goto out;
5047                 }
5048                 ret = insert_reserved_file_extent(trans, inode,
5049                                                   cur_offset, ins.objectid,
5050                                                   ins.offset, ins.offset,
5051                                                   ins.offset, locked_end,
5052                                                   0, 0, 0,
5053                                                   BTRFS_FILE_EXTENT_PREALLOC);
5054                 BUG_ON(ret);
5055                 num_bytes -= ins.offset;
5056                 cur_offset += ins.offset;
5057                 alloc_hint = ins.objectid + ins.offset;
5058         }
5059 out:
5060         if (cur_offset > start) {
5061                 inode->i_ctime = CURRENT_TIME;
5062                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
5063                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
5064                     cur_offset > i_size_read(inode))
5065                         btrfs_i_size_write(inode, cur_offset);
5066                 ret = btrfs_update_inode(trans, root, inode);
5067                 BUG_ON(ret);
5068         }
5069
5070         return ret;
5071 }
5072
5073 static long btrfs_fallocate(struct inode *inode, int mode,
5074                             loff_t offset, loff_t len)
5075 {
5076         u64 cur_offset;
5077         u64 last_byte;
5078         u64 alloc_start;
5079         u64 alloc_end;
5080         u64 alloc_hint = 0;
5081         u64 locked_end;
5082         u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
5083         struct extent_map *em;
5084         struct btrfs_trans_handle *trans;
5085         int ret;
5086
5087         alloc_start = offset & ~mask;
5088         alloc_end =  (offset + len + mask) & ~mask;
5089
5090         /*
5091          * wait for ordered IO before we have any locks.  We'll loop again
5092          * below with the locks held.
5093          */
5094         btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
5095
5096         mutex_lock(&inode->i_mutex);
5097         if (alloc_start > inode->i_size) {
5098                 ret = btrfs_cont_expand(inode, alloc_start);
5099                 if (ret)
5100                         goto out;
5101         }
5102
5103         locked_end = alloc_end - 1;
5104         while (1) {
5105                 struct btrfs_ordered_extent *ordered;
5106
5107                 trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
5108                 if (!trans) {
5109                         ret = -EIO;
5110                         goto out;
5111                 }
5112
5113                 /* the extent lock is ordered inside the running
5114                  * transaction
5115                  */
5116                 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5117                             GFP_NOFS);
5118                 ordered = btrfs_lookup_first_ordered_extent(inode,
5119                                                             alloc_end - 1);
5120                 if (ordered &&
5121                     ordered->file_offset + ordered->len > alloc_start &&
5122                     ordered->file_offset < alloc_end) {
5123                         btrfs_put_ordered_extent(ordered);
5124                         unlock_extent(&BTRFS_I(inode)->io_tree,
5125                                       alloc_start, locked_end, GFP_NOFS);
5126                         btrfs_end_transaction(trans, BTRFS_I(inode)->root);
5127
5128                         /*
5129                          * we can't wait on the range with the transaction
5130                          * running or with the extent lock held
5131                          */
5132                         btrfs_wait_ordered_range(inode, alloc_start,
5133                                                  alloc_end - alloc_start);
5134                 } else {
5135                         if (ordered)
5136                                 btrfs_put_ordered_extent(ordered);
5137                         break;
5138                 }
5139         }
5140
5141         cur_offset = alloc_start;
5142         while (1) {
5143                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5144                                       alloc_end - cur_offset, 0);
5145                 BUG_ON(IS_ERR(em) || !em);
5146                 last_byte = min(extent_map_end(em), alloc_end);
5147                 last_byte = (last_byte + mask) & ~mask;
5148                 if (em->block_start == EXTENT_MAP_HOLE) {
5149                         ret = prealloc_file_range(trans, inode, cur_offset,
5150                                         last_byte, locked_end + 1,
5151                                         alloc_hint, mode);
5152                         if (ret < 0) {
5153                                 free_extent_map(em);
5154                                 break;
5155                         }
5156                 }
5157                 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
5158                         alloc_hint = em->block_start;
5159                 free_extent_map(em);
5160
5161                 cur_offset = last_byte;
5162                 if (cur_offset >= alloc_end) {
5163                         ret = 0;
5164                         break;
5165                 }
5166         }
5167         unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5168                       GFP_NOFS);
5169
5170         btrfs_end_transaction(trans, BTRFS_I(inode)->root);
5171 out:
5172         mutex_unlock(&inode->i_mutex);
5173         return ret;
5174 }
5175
5176 static int btrfs_set_page_dirty(struct page *page)
5177 {
5178         return __set_page_dirty_nobuffers(page);
5179 }
5180
5181 static int btrfs_permission(struct inode *inode, int mask)
5182 {
5183         if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
5184                 return -EACCES;
5185         return generic_permission(inode, mask, btrfs_check_acl);
5186 }
5187
5188 static struct inode_operations btrfs_dir_inode_operations = {
5189         .getattr        = btrfs_getattr,
5190         .lookup         = btrfs_lookup,
5191         .create         = btrfs_create,
5192         .unlink         = btrfs_unlink,
5193         .link           = btrfs_link,
5194         .mkdir          = btrfs_mkdir,
5195         .rmdir          = btrfs_rmdir,
5196         .rename         = btrfs_rename,
5197         .symlink        = btrfs_symlink,
5198         .setattr        = btrfs_setattr,
5199         .mknod          = btrfs_mknod,
5200         .setxattr       = btrfs_setxattr,
5201         .getxattr       = btrfs_getxattr,
5202         .listxattr      = btrfs_listxattr,
5203         .removexattr    = btrfs_removexattr,
5204         .permission     = btrfs_permission,
5205 };
5206 static struct inode_operations btrfs_dir_ro_inode_operations = {
5207         .lookup         = btrfs_lookup,
5208         .permission     = btrfs_permission,
5209 };
5210 static struct file_operations btrfs_dir_file_operations = {
5211         .llseek         = generic_file_llseek,
5212         .read           = generic_read_dir,
5213         .readdir        = btrfs_real_readdir,
5214         .unlocked_ioctl = btrfs_ioctl,
5215 #ifdef CONFIG_COMPAT
5216         .compat_ioctl   = btrfs_ioctl,
5217 #endif
5218         .release        = btrfs_release_file,
5219         .fsync          = btrfs_sync_file,
5220 };
5221
5222 static struct extent_io_ops btrfs_extent_io_ops = {
5223         .fill_delalloc = run_delalloc_range,
5224         .submit_bio_hook = btrfs_submit_bio_hook,
5225         .merge_bio_hook = btrfs_merge_bio_hook,
5226         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
5227         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
5228         .writepage_start_hook = btrfs_writepage_start_hook,
5229         .readpage_io_failed_hook = btrfs_io_failed_hook,
5230         .set_bit_hook = btrfs_set_bit_hook,
5231         .clear_bit_hook = btrfs_clear_bit_hook,
5232 };
5233
5234 /*
5235  * btrfs doesn't support the bmap operation because swapfiles
5236  * use bmap to make a mapping of extents in the file.  They assume
5237  * these extents won't change over the life of the file and they
5238  * use the bmap result to do IO directly to the drive.
5239  *
5240  * the btrfs bmap call would return logical addresses that aren't
5241  * suitable for IO and they also will change frequently as COW
5242  * operations happen.  So, swapfile + btrfs == corruption.
5243  *
5244  * For now we're avoiding this by dropping bmap.
5245  */
5246 static struct address_space_operations btrfs_aops = {
5247         .readpage       = btrfs_readpage,
5248         .writepage      = btrfs_writepage,
5249         .writepages     = btrfs_writepages,
5250         .readpages      = btrfs_readpages,
5251         .sync_page      = block_sync_page,
5252         .direct_IO      = btrfs_direct_IO,
5253         .invalidatepage = btrfs_invalidatepage,
5254         .releasepage    = btrfs_releasepage,
5255         .set_page_dirty = btrfs_set_page_dirty,
5256 };
5257
5258 static struct address_space_operations btrfs_symlink_aops = {
5259         .readpage       = btrfs_readpage,
5260         .writepage      = btrfs_writepage,
5261         .invalidatepage = btrfs_invalidatepage,
5262         .releasepage    = btrfs_releasepage,
5263 };
5264
5265 static struct inode_operations btrfs_file_inode_operations = {
5266         .truncate       = btrfs_truncate,
5267         .getattr        = btrfs_getattr,
5268         .setattr        = btrfs_setattr,
5269         .setxattr       = btrfs_setxattr,
5270         .getxattr       = btrfs_getxattr,
5271         .listxattr      = btrfs_listxattr,
5272         .removexattr    = btrfs_removexattr,
5273         .permission     = btrfs_permission,
5274         .fallocate      = btrfs_fallocate,
5275         .fiemap         = btrfs_fiemap,
5276 };
5277 static struct inode_operations btrfs_special_inode_operations = {
5278         .getattr        = btrfs_getattr,
5279         .setattr        = btrfs_setattr,
5280         .permission     = btrfs_permission,
5281         .setxattr       = btrfs_setxattr,
5282         .getxattr       = btrfs_getxattr,
5283         .listxattr      = btrfs_listxattr,
5284         .removexattr    = btrfs_removexattr,
5285 };
5286 static struct inode_operations btrfs_symlink_inode_operations = {
5287         .readlink       = generic_readlink,
5288         .follow_link    = page_follow_link_light,
5289         .put_link       = page_put_link,
5290         .permission     = btrfs_permission,
5291         .setxattr       = btrfs_setxattr,
5292         .getxattr       = btrfs_getxattr,
5293         .listxattr      = btrfs_listxattr,
5294         .removexattr    = btrfs_removexattr,
5295 };